Loading...
1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18#include <linux/sched.h>
19#include <linux/pagemap.h>
20#include <linux/writeback.h>
21#include <linux/blkdev.h>
22#include <linux/sort.h>
23#include <linux/rcupdate.h>
24#include <linux/kthread.h>
25#include <linux/slab.h>
26#include <linux/ratelimit.h>
27#include <linux/percpu_counter.h>
28#include "hash.h"
29#include "tree-log.h"
30#include "disk-io.h"
31#include "print-tree.h"
32#include "volumes.h"
33#include "raid56.h"
34#include "locking.h"
35#include "free-space-cache.h"
36#include "free-space-tree.h"
37#include "math.h"
38#include "sysfs.h"
39#include "qgroup.h"
40
41#undef SCRAMBLE_DELAYED_REFS
42
43/*
44 * control flags for do_chunk_alloc's force field
45 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
46 * if we really need one.
47 *
48 * CHUNK_ALLOC_LIMITED means to only try and allocate one
49 * if we have very few chunks already allocated. This is
50 * used as part of the clustering code to help make sure
51 * we have a good pool of storage to cluster in, without
52 * filling the FS with empty chunks
53 *
54 * CHUNK_ALLOC_FORCE means it must try to allocate one
55 *
56 */
57enum {
58 CHUNK_ALLOC_NO_FORCE = 0,
59 CHUNK_ALLOC_LIMITED = 1,
60 CHUNK_ALLOC_FORCE = 2,
61};
62
63/*
64 * Control how reservations are dealt with.
65 *
66 * RESERVE_FREE - freeing a reservation.
67 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * ENOSPC accounting
69 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
70 * bytes_may_use as the ENOSPC accounting is done elsewhere
71 */
72enum {
73 RESERVE_FREE = 0,
74 RESERVE_ALLOC = 1,
75 RESERVE_ALLOC_NO_ACCOUNT = 2,
76};
77
78static int update_block_group(struct btrfs_trans_handle *trans,
79 struct btrfs_root *root, u64 bytenr,
80 u64 num_bytes, int alloc);
81static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
82 struct btrfs_root *root,
83 struct btrfs_delayed_ref_node *node, u64 parent,
84 u64 root_objectid, u64 owner_objectid,
85 u64 owner_offset, int refs_to_drop,
86 struct btrfs_delayed_extent_op *extra_op);
87static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
88 struct extent_buffer *leaf,
89 struct btrfs_extent_item *ei);
90static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
91 struct btrfs_root *root,
92 u64 parent, u64 root_objectid,
93 u64 flags, u64 owner, u64 offset,
94 struct btrfs_key *ins, int ref_mod);
95static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
96 struct btrfs_root *root,
97 u64 parent, u64 root_objectid,
98 u64 flags, struct btrfs_disk_key *key,
99 int level, struct btrfs_key *ins);
100static int do_chunk_alloc(struct btrfs_trans_handle *trans,
101 struct btrfs_root *extent_root, u64 flags,
102 int force);
103static int find_next_key(struct btrfs_path *path, int level,
104 struct btrfs_key *key);
105static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
106 int dump_block_groups);
107static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
108 u64 num_bytes, int reserve,
109 int delalloc);
110static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
111 u64 num_bytes);
112int btrfs_pin_extent(struct btrfs_root *root,
113 u64 bytenr, u64 num_bytes, int reserved);
114
115static noinline int
116block_group_cache_done(struct btrfs_block_group_cache *cache)
117{
118 smp_mb();
119 return cache->cached == BTRFS_CACHE_FINISHED ||
120 cache->cached == BTRFS_CACHE_ERROR;
121}
122
123static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
124{
125 return (cache->flags & bits) == bits;
126}
127
128void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
129{
130 atomic_inc(&cache->count);
131}
132
133void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
134{
135 if (atomic_dec_and_test(&cache->count)) {
136 WARN_ON(cache->pinned > 0);
137 WARN_ON(cache->reserved > 0);
138 kfree(cache->free_space_ctl);
139 kfree(cache);
140 }
141}
142
143/*
144 * this adds the block group to the fs_info rb tree for the block group
145 * cache
146 */
147static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
148 struct btrfs_block_group_cache *block_group)
149{
150 struct rb_node **p;
151 struct rb_node *parent = NULL;
152 struct btrfs_block_group_cache *cache;
153
154 spin_lock(&info->block_group_cache_lock);
155 p = &info->block_group_cache_tree.rb_node;
156
157 while (*p) {
158 parent = *p;
159 cache = rb_entry(parent, struct btrfs_block_group_cache,
160 cache_node);
161 if (block_group->key.objectid < cache->key.objectid) {
162 p = &(*p)->rb_left;
163 } else if (block_group->key.objectid > cache->key.objectid) {
164 p = &(*p)->rb_right;
165 } else {
166 spin_unlock(&info->block_group_cache_lock);
167 return -EEXIST;
168 }
169 }
170
171 rb_link_node(&block_group->cache_node, parent, p);
172 rb_insert_color(&block_group->cache_node,
173 &info->block_group_cache_tree);
174
175 if (info->first_logical_byte > block_group->key.objectid)
176 info->first_logical_byte = block_group->key.objectid;
177
178 spin_unlock(&info->block_group_cache_lock);
179
180 return 0;
181}
182
183/*
184 * This will return the block group at or after bytenr if contains is 0, else
185 * it will return the block group that contains the bytenr
186 */
187static struct btrfs_block_group_cache *
188block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
189 int contains)
190{
191 struct btrfs_block_group_cache *cache, *ret = NULL;
192 struct rb_node *n;
193 u64 end, start;
194
195 spin_lock(&info->block_group_cache_lock);
196 n = info->block_group_cache_tree.rb_node;
197
198 while (n) {
199 cache = rb_entry(n, struct btrfs_block_group_cache,
200 cache_node);
201 end = cache->key.objectid + cache->key.offset - 1;
202 start = cache->key.objectid;
203
204 if (bytenr < start) {
205 if (!contains && (!ret || start < ret->key.objectid))
206 ret = cache;
207 n = n->rb_left;
208 } else if (bytenr > start) {
209 if (contains && bytenr <= end) {
210 ret = cache;
211 break;
212 }
213 n = n->rb_right;
214 } else {
215 ret = cache;
216 break;
217 }
218 }
219 if (ret) {
220 btrfs_get_block_group(ret);
221 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
222 info->first_logical_byte = ret->key.objectid;
223 }
224 spin_unlock(&info->block_group_cache_lock);
225
226 return ret;
227}
228
229static int add_excluded_extent(struct btrfs_root *root,
230 u64 start, u64 num_bytes)
231{
232 u64 end = start + num_bytes - 1;
233 set_extent_bits(&root->fs_info->freed_extents[0],
234 start, end, EXTENT_UPTODATE, GFP_NOFS);
235 set_extent_bits(&root->fs_info->freed_extents[1],
236 start, end, EXTENT_UPTODATE, GFP_NOFS);
237 return 0;
238}
239
240static void free_excluded_extents(struct btrfs_root *root,
241 struct btrfs_block_group_cache *cache)
242{
243 u64 start, end;
244
245 start = cache->key.objectid;
246 end = start + cache->key.offset - 1;
247
248 clear_extent_bits(&root->fs_info->freed_extents[0],
249 start, end, EXTENT_UPTODATE, GFP_NOFS);
250 clear_extent_bits(&root->fs_info->freed_extents[1],
251 start, end, EXTENT_UPTODATE, GFP_NOFS);
252}
253
254static int exclude_super_stripes(struct btrfs_root *root,
255 struct btrfs_block_group_cache *cache)
256{
257 u64 bytenr;
258 u64 *logical;
259 int stripe_len;
260 int i, nr, ret;
261
262 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
263 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
264 cache->bytes_super += stripe_len;
265 ret = add_excluded_extent(root, cache->key.objectid,
266 stripe_len);
267 if (ret)
268 return ret;
269 }
270
271 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
272 bytenr = btrfs_sb_offset(i);
273 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
274 cache->key.objectid, bytenr,
275 0, &logical, &nr, &stripe_len);
276 if (ret)
277 return ret;
278
279 while (nr--) {
280 u64 start, len;
281
282 if (logical[nr] > cache->key.objectid +
283 cache->key.offset)
284 continue;
285
286 if (logical[nr] + stripe_len <= cache->key.objectid)
287 continue;
288
289 start = logical[nr];
290 if (start < cache->key.objectid) {
291 start = cache->key.objectid;
292 len = (logical[nr] + stripe_len) - start;
293 } else {
294 len = min_t(u64, stripe_len,
295 cache->key.objectid +
296 cache->key.offset - start);
297 }
298
299 cache->bytes_super += len;
300 ret = add_excluded_extent(root, start, len);
301 if (ret) {
302 kfree(logical);
303 return ret;
304 }
305 }
306
307 kfree(logical);
308 }
309 return 0;
310}
311
312static struct btrfs_caching_control *
313get_caching_control(struct btrfs_block_group_cache *cache)
314{
315 struct btrfs_caching_control *ctl;
316
317 spin_lock(&cache->lock);
318 if (!cache->caching_ctl) {
319 spin_unlock(&cache->lock);
320 return NULL;
321 }
322
323 ctl = cache->caching_ctl;
324 atomic_inc(&ctl->count);
325 spin_unlock(&cache->lock);
326 return ctl;
327}
328
329static void put_caching_control(struct btrfs_caching_control *ctl)
330{
331 if (atomic_dec_and_test(&ctl->count))
332 kfree(ctl);
333}
334
335#ifdef CONFIG_BTRFS_DEBUG
336static void fragment_free_space(struct btrfs_root *root,
337 struct btrfs_block_group_cache *block_group)
338{
339 u64 start = block_group->key.objectid;
340 u64 len = block_group->key.offset;
341 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
342 root->nodesize : root->sectorsize;
343 u64 step = chunk << 1;
344
345 while (len > chunk) {
346 btrfs_remove_free_space(block_group, start, chunk);
347 start += step;
348 if (len < step)
349 len = 0;
350 else
351 len -= step;
352 }
353}
354#endif
355
356/*
357 * this is only called by cache_block_group, since we could have freed extents
358 * we need to check the pinned_extents for any extents that can't be used yet
359 * since their free space will be released as soon as the transaction commits.
360 */
361u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
362 struct btrfs_fs_info *info, u64 start, u64 end)
363{
364 u64 extent_start, extent_end, size, total_added = 0;
365 int ret;
366
367 while (start < end) {
368 ret = find_first_extent_bit(info->pinned_extents, start,
369 &extent_start, &extent_end,
370 EXTENT_DIRTY | EXTENT_UPTODATE,
371 NULL);
372 if (ret)
373 break;
374
375 if (extent_start <= start) {
376 start = extent_end + 1;
377 } else if (extent_start > start && extent_start < end) {
378 size = extent_start - start;
379 total_added += size;
380 ret = btrfs_add_free_space(block_group, start,
381 size);
382 BUG_ON(ret); /* -ENOMEM or logic error */
383 start = extent_end + 1;
384 } else {
385 break;
386 }
387 }
388
389 if (start < end) {
390 size = end - start;
391 total_added += size;
392 ret = btrfs_add_free_space(block_group, start, size);
393 BUG_ON(ret); /* -ENOMEM or logic error */
394 }
395
396 return total_added;
397}
398
399static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
400{
401 struct btrfs_block_group_cache *block_group;
402 struct btrfs_fs_info *fs_info;
403 struct btrfs_root *extent_root;
404 struct btrfs_path *path;
405 struct extent_buffer *leaf;
406 struct btrfs_key key;
407 u64 total_found = 0;
408 u64 last = 0;
409 u32 nritems;
410 int ret;
411 bool wakeup = true;
412
413 block_group = caching_ctl->block_group;
414 fs_info = block_group->fs_info;
415 extent_root = fs_info->extent_root;
416
417 path = btrfs_alloc_path();
418 if (!path)
419 return -ENOMEM;
420
421 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
422
423#ifdef CONFIG_BTRFS_DEBUG
424 /*
425 * If we're fragmenting we don't want to make anybody think we can
426 * allocate from this block group until we've had a chance to fragment
427 * the free space.
428 */
429 if (btrfs_should_fragment_free_space(extent_root, block_group))
430 wakeup = false;
431#endif
432 /*
433 * We don't want to deadlock with somebody trying to allocate a new
434 * extent for the extent root while also trying to search the extent
435 * root to add free space. So we skip locking and search the commit
436 * root, since its read-only
437 */
438 path->skip_locking = 1;
439 path->search_commit_root = 1;
440 path->reada = READA_FORWARD;
441
442 key.objectid = last;
443 key.offset = 0;
444 key.type = BTRFS_EXTENT_ITEM_KEY;
445
446next:
447 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
448 if (ret < 0)
449 goto out;
450
451 leaf = path->nodes[0];
452 nritems = btrfs_header_nritems(leaf);
453
454 while (1) {
455 if (btrfs_fs_closing(fs_info) > 1) {
456 last = (u64)-1;
457 break;
458 }
459
460 if (path->slots[0] < nritems) {
461 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
462 } else {
463 ret = find_next_key(path, 0, &key);
464 if (ret)
465 break;
466
467 if (need_resched() ||
468 rwsem_is_contended(&fs_info->commit_root_sem)) {
469 if (wakeup)
470 caching_ctl->progress = last;
471 btrfs_release_path(path);
472 up_read(&fs_info->commit_root_sem);
473 mutex_unlock(&caching_ctl->mutex);
474 cond_resched();
475 mutex_lock(&caching_ctl->mutex);
476 down_read(&fs_info->commit_root_sem);
477 goto next;
478 }
479
480 ret = btrfs_next_leaf(extent_root, path);
481 if (ret < 0)
482 goto out;
483 if (ret)
484 break;
485 leaf = path->nodes[0];
486 nritems = btrfs_header_nritems(leaf);
487 continue;
488 }
489
490 if (key.objectid < last) {
491 key.objectid = last;
492 key.offset = 0;
493 key.type = BTRFS_EXTENT_ITEM_KEY;
494
495 if (wakeup)
496 caching_ctl->progress = last;
497 btrfs_release_path(path);
498 goto next;
499 }
500
501 if (key.objectid < block_group->key.objectid) {
502 path->slots[0]++;
503 continue;
504 }
505
506 if (key.objectid >= block_group->key.objectid +
507 block_group->key.offset)
508 break;
509
510 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
511 key.type == BTRFS_METADATA_ITEM_KEY) {
512 total_found += add_new_free_space(block_group,
513 fs_info, last,
514 key.objectid);
515 if (key.type == BTRFS_METADATA_ITEM_KEY)
516 last = key.objectid +
517 fs_info->tree_root->nodesize;
518 else
519 last = key.objectid + key.offset;
520
521 if (total_found > CACHING_CTL_WAKE_UP) {
522 total_found = 0;
523 if (wakeup)
524 wake_up(&caching_ctl->wait);
525 }
526 }
527 path->slots[0]++;
528 }
529 ret = 0;
530
531 total_found += add_new_free_space(block_group, fs_info, last,
532 block_group->key.objectid +
533 block_group->key.offset);
534 caching_ctl->progress = (u64)-1;
535
536out:
537 btrfs_free_path(path);
538 return ret;
539}
540
541static noinline void caching_thread(struct btrfs_work *work)
542{
543 struct btrfs_block_group_cache *block_group;
544 struct btrfs_fs_info *fs_info;
545 struct btrfs_caching_control *caching_ctl;
546 struct btrfs_root *extent_root;
547 int ret;
548
549 caching_ctl = container_of(work, struct btrfs_caching_control, work);
550 block_group = caching_ctl->block_group;
551 fs_info = block_group->fs_info;
552 extent_root = fs_info->extent_root;
553
554 mutex_lock(&caching_ctl->mutex);
555 down_read(&fs_info->commit_root_sem);
556
557 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
558 ret = load_free_space_tree(caching_ctl);
559 else
560 ret = load_extent_tree_free(caching_ctl);
561
562 spin_lock(&block_group->lock);
563 block_group->caching_ctl = NULL;
564 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
565 spin_unlock(&block_group->lock);
566
567#ifdef CONFIG_BTRFS_DEBUG
568 if (btrfs_should_fragment_free_space(extent_root, block_group)) {
569 u64 bytes_used;
570
571 spin_lock(&block_group->space_info->lock);
572 spin_lock(&block_group->lock);
573 bytes_used = block_group->key.offset -
574 btrfs_block_group_used(&block_group->item);
575 block_group->space_info->bytes_used += bytes_used >> 1;
576 spin_unlock(&block_group->lock);
577 spin_unlock(&block_group->space_info->lock);
578 fragment_free_space(extent_root, block_group);
579 }
580#endif
581
582 caching_ctl->progress = (u64)-1;
583
584 up_read(&fs_info->commit_root_sem);
585 free_excluded_extents(fs_info->extent_root, block_group);
586 mutex_unlock(&caching_ctl->mutex);
587
588 wake_up(&caching_ctl->wait);
589
590 put_caching_control(caching_ctl);
591 btrfs_put_block_group(block_group);
592}
593
594static int cache_block_group(struct btrfs_block_group_cache *cache,
595 int load_cache_only)
596{
597 DEFINE_WAIT(wait);
598 struct btrfs_fs_info *fs_info = cache->fs_info;
599 struct btrfs_caching_control *caching_ctl;
600 int ret = 0;
601
602 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
603 if (!caching_ctl)
604 return -ENOMEM;
605
606 INIT_LIST_HEAD(&caching_ctl->list);
607 mutex_init(&caching_ctl->mutex);
608 init_waitqueue_head(&caching_ctl->wait);
609 caching_ctl->block_group = cache;
610 caching_ctl->progress = cache->key.objectid;
611 atomic_set(&caching_ctl->count, 1);
612 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
613 caching_thread, NULL, NULL);
614
615 spin_lock(&cache->lock);
616 /*
617 * This should be a rare occasion, but this could happen I think in the
618 * case where one thread starts to load the space cache info, and then
619 * some other thread starts a transaction commit which tries to do an
620 * allocation while the other thread is still loading the space cache
621 * info. The previous loop should have kept us from choosing this block
622 * group, but if we've moved to the state where we will wait on caching
623 * block groups we need to first check if we're doing a fast load here,
624 * so we can wait for it to finish, otherwise we could end up allocating
625 * from a block group who's cache gets evicted for one reason or
626 * another.
627 */
628 while (cache->cached == BTRFS_CACHE_FAST) {
629 struct btrfs_caching_control *ctl;
630
631 ctl = cache->caching_ctl;
632 atomic_inc(&ctl->count);
633 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
634 spin_unlock(&cache->lock);
635
636 schedule();
637
638 finish_wait(&ctl->wait, &wait);
639 put_caching_control(ctl);
640 spin_lock(&cache->lock);
641 }
642
643 if (cache->cached != BTRFS_CACHE_NO) {
644 spin_unlock(&cache->lock);
645 kfree(caching_ctl);
646 return 0;
647 }
648 WARN_ON(cache->caching_ctl);
649 cache->caching_ctl = caching_ctl;
650 cache->cached = BTRFS_CACHE_FAST;
651 spin_unlock(&cache->lock);
652
653 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
654 mutex_lock(&caching_ctl->mutex);
655 ret = load_free_space_cache(fs_info, cache);
656
657 spin_lock(&cache->lock);
658 if (ret == 1) {
659 cache->caching_ctl = NULL;
660 cache->cached = BTRFS_CACHE_FINISHED;
661 cache->last_byte_to_unpin = (u64)-1;
662 caching_ctl->progress = (u64)-1;
663 } else {
664 if (load_cache_only) {
665 cache->caching_ctl = NULL;
666 cache->cached = BTRFS_CACHE_NO;
667 } else {
668 cache->cached = BTRFS_CACHE_STARTED;
669 cache->has_caching_ctl = 1;
670 }
671 }
672 spin_unlock(&cache->lock);
673#ifdef CONFIG_BTRFS_DEBUG
674 if (ret == 1 &&
675 btrfs_should_fragment_free_space(fs_info->extent_root,
676 cache)) {
677 u64 bytes_used;
678
679 spin_lock(&cache->space_info->lock);
680 spin_lock(&cache->lock);
681 bytes_used = cache->key.offset -
682 btrfs_block_group_used(&cache->item);
683 cache->space_info->bytes_used += bytes_used >> 1;
684 spin_unlock(&cache->lock);
685 spin_unlock(&cache->space_info->lock);
686 fragment_free_space(fs_info->extent_root, cache);
687 }
688#endif
689 mutex_unlock(&caching_ctl->mutex);
690
691 wake_up(&caching_ctl->wait);
692 if (ret == 1) {
693 put_caching_control(caching_ctl);
694 free_excluded_extents(fs_info->extent_root, cache);
695 return 0;
696 }
697 } else {
698 /*
699 * We're either using the free space tree or no caching at all.
700 * Set cached to the appropriate value and wakeup any waiters.
701 */
702 spin_lock(&cache->lock);
703 if (load_cache_only) {
704 cache->caching_ctl = NULL;
705 cache->cached = BTRFS_CACHE_NO;
706 } else {
707 cache->cached = BTRFS_CACHE_STARTED;
708 cache->has_caching_ctl = 1;
709 }
710 spin_unlock(&cache->lock);
711 wake_up(&caching_ctl->wait);
712 }
713
714 if (load_cache_only) {
715 put_caching_control(caching_ctl);
716 return 0;
717 }
718
719 down_write(&fs_info->commit_root_sem);
720 atomic_inc(&caching_ctl->count);
721 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
722 up_write(&fs_info->commit_root_sem);
723
724 btrfs_get_block_group(cache);
725
726 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
727
728 return ret;
729}
730
731/*
732 * return the block group that starts at or after bytenr
733 */
734static struct btrfs_block_group_cache *
735btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
736{
737 struct btrfs_block_group_cache *cache;
738
739 cache = block_group_cache_tree_search(info, bytenr, 0);
740
741 return cache;
742}
743
744/*
745 * return the block group that contains the given bytenr
746 */
747struct btrfs_block_group_cache *btrfs_lookup_block_group(
748 struct btrfs_fs_info *info,
749 u64 bytenr)
750{
751 struct btrfs_block_group_cache *cache;
752
753 cache = block_group_cache_tree_search(info, bytenr, 1);
754
755 return cache;
756}
757
758static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
759 u64 flags)
760{
761 struct list_head *head = &info->space_info;
762 struct btrfs_space_info *found;
763
764 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
765
766 rcu_read_lock();
767 list_for_each_entry_rcu(found, head, list) {
768 if (found->flags & flags) {
769 rcu_read_unlock();
770 return found;
771 }
772 }
773 rcu_read_unlock();
774 return NULL;
775}
776
777/*
778 * after adding space to the filesystem, we need to clear the full flags
779 * on all the space infos.
780 */
781void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
782{
783 struct list_head *head = &info->space_info;
784 struct btrfs_space_info *found;
785
786 rcu_read_lock();
787 list_for_each_entry_rcu(found, head, list)
788 found->full = 0;
789 rcu_read_unlock();
790}
791
792/* simple helper to search for an existing data extent at a given offset */
793int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
794{
795 int ret;
796 struct btrfs_key key;
797 struct btrfs_path *path;
798
799 path = btrfs_alloc_path();
800 if (!path)
801 return -ENOMEM;
802
803 key.objectid = start;
804 key.offset = len;
805 key.type = BTRFS_EXTENT_ITEM_KEY;
806 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
807 0, 0);
808 btrfs_free_path(path);
809 return ret;
810}
811
812/*
813 * helper function to lookup reference count and flags of a tree block.
814 *
815 * the head node for delayed ref is used to store the sum of all the
816 * reference count modifications queued up in the rbtree. the head
817 * node may also store the extent flags to set. This way you can check
818 * to see what the reference count and extent flags would be if all of
819 * the delayed refs are not processed.
820 */
821int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
822 struct btrfs_root *root, u64 bytenr,
823 u64 offset, int metadata, u64 *refs, u64 *flags)
824{
825 struct btrfs_delayed_ref_head *head;
826 struct btrfs_delayed_ref_root *delayed_refs;
827 struct btrfs_path *path;
828 struct btrfs_extent_item *ei;
829 struct extent_buffer *leaf;
830 struct btrfs_key key;
831 u32 item_size;
832 u64 num_refs;
833 u64 extent_flags;
834 int ret;
835
836 /*
837 * If we don't have skinny metadata, don't bother doing anything
838 * different
839 */
840 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
841 offset = root->nodesize;
842 metadata = 0;
843 }
844
845 path = btrfs_alloc_path();
846 if (!path)
847 return -ENOMEM;
848
849 if (!trans) {
850 path->skip_locking = 1;
851 path->search_commit_root = 1;
852 }
853
854search_again:
855 key.objectid = bytenr;
856 key.offset = offset;
857 if (metadata)
858 key.type = BTRFS_METADATA_ITEM_KEY;
859 else
860 key.type = BTRFS_EXTENT_ITEM_KEY;
861
862 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
863 &key, path, 0, 0);
864 if (ret < 0)
865 goto out_free;
866
867 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
868 if (path->slots[0]) {
869 path->slots[0]--;
870 btrfs_item_key_to_cpu(path->nodes[0], &key,
871 path->slots[0]);
872 if (key.objectid == bytenr &&
873 key.type == BTRFS_EXTENT_ITEM_KEY &&
874 key.offset == root->nodesize)
875 ret = 0;
876 }
877 }
878
879 if (ret == 0) {
880 leaf = path->nodes[0];
881 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
882 if (item_size >= sizeof(*ei)) {
883 ei = btrfs_item_ptr(leaf, path->slots[0],
884 struct btrfs_extent_item);
885 num_refs = btrfs_extent_refs(leaf, ei);
886 extent_flags = btrfs_extent_flags(leaf, ei);
887 } else {
888#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
889 struct btrfs_extent_item_v0 *ei0;
890 BUG_ON(item_size != sizeof(*ei0));
891 ei0 = btrfs_item_ptr(leaf, path->slots[0],
892 struct btrfs_extent_item_v0);
893 num_refs = btrfs_extent_refs_v0(leaf, ei0);
894 /* FIXME: this isn't correct for data */
895 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
896#else
897 BUG();
898#endif
899 }
900 BUG_ON(num_refs == 0);
901 } else {
902 num_refs = 0;
903 extent_flags = 0;
904 ret = 0;
905 }
906
907 if (!trans)
908 goto out;
909
910 delayed_refs = &trans->transaction->delayed_refs;
911 spin_lock(&delayed_refs->lock);
912 head = btrfs_find_delayed_ref_head(trans, bytenr);
913 if (head) {
914 if (!mutex_trylock(&head->mutex)) {
915 atomic_inc(&head->node.refs);
916 spin_unlock(&delayed_refs->lock);
917
918 btrfs_release_path(path);
919
920 /*
921 * Mutex was contended, block until it's released and try
922 * again
923 */
924 mutex_lock(&head->mutex);
925 mutex_unlock(&head->mutex);
926 btrfs_put_delayed_ref(&head->node);
927 goto search_again;
928 }
929 spin_lock(&head->lock);
930 if (head->extent_op && head->extent_op->update_flags)
931 extent_flags |= head->extent_op->flags_to_set;
932 else
933 BUG_ON(num_refs == 0);
934
935 num_refs += head->node.ref_mod;
936 spin_unlock(&head->lock);
937 mutex_unlock(&head->mutex);
938 }
939 spin_unlock(&delayed_refs->lock);
940out:
941 WARN_ON(num_refs == 0);
942 if (refs)
943 *refs = num_refs;
944 if (flags)
945 *flags = extent_flags;
946out_free:
947 btrfs_free_path(path);
948 return ret;
949}
950
951/*
952 * Back reference rules. Back refs have three main goals:
953 *
954 * 1) differentiate between all holders of references to an extent so that
955 * when a reference is dropped we can make sure it was a valid reference
956 * before freeing the extent.
957 *
958 * 2) Provide enough information to quickly find the holders of an extent
959 * if we notice a given block is corrupted or bad.
960 *
961 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
962 * maintenance. This is actually the same as #2, but with a slightly
963 * different use case.
964 *
965 * There are two kinds of back refs. The implicit back refs is optimized
966 * for pointers in non-shared tree blocks. For a given pointer in a block,
967 * back refs of this kind provide information about the block's owner tree
968 * and the pointer's key. These information allow us to find the block by
969 * b-tree searching. The full back refs is for pointers in tree blocks not
970 * referenced by their owner trees. The location of tree block is recorded
971 * in the back refs. Actually the full back refs is generic, and can be
972 * used in all cases the implicit back refs is used. The major shortcoming
973 * of the full back refs is its overhead. Every time a tree block gets
974 * COWed, we have to update back refs entry for all pointers in it.
975 *
976 * For a newly allocated tree block, we use implicit back refs for
977 * pointers in it. This means most tree related operations only involve
978 * implicit back refs. For a tree block created in old transaction, the
979 * only way to drop a reference to it is COW it. So we can detect the
980 * event that tree block loses its owner tree's reference and do the
981 * back refs conversion.
982 *
983 * When a tree block is COW'd through a tree, there are four cases:
984 *
985 * The reference count of the block is one and the tree is the block's
986 * owner tree. Nothing to do in this case.
987 *
988 * The reference count of the block is one and the tree is not the
989 * block's owner tree. In this case, full back refs is used for pointers
990 * in the block. Remove these full back refs, add implicit back refs for
991 * every pointers in the new block.
992 *
993 * The reference count of the block is greater than one and the tree is
994 * the block's owner tree. In this case, implicit back refs is used for
995 * pointers in the block. Add full back refs for every pointers in the
996 * block, increase lower level extents' reference counts. The original
997 * implicit back refs are entailed to the new block.
998 *
999 * The reference count of the block is greater than one and the tree is
1000 * not the block's owner tree. Add implicit back refs for every pointer in
1001 * the new block, increase lower level extents' reference count.
1002 *
1003 * Back Reference Key composing:
1004 *
1005 * The key objectid corresponds to the first byte in the extent,
1006 * The key type is used to differentiate between types of back refs.
1007 * There are different meanings of the key offset for different types
1008 * of back refs.
1009 *
1010 * File extents can be referenced by:
1011 *
1012 * - multiple snapshots, subvolumes, or different generations in one subvol
1013 * - different files inside a single subvolume
1014 * - different offsets inside a file (bookend extents in file.c)
1015 *
1016 * The extent ref structure for the implicit back refs has fields for:
1017 *
1018 * - Objectid of the subvolume root
1019 * - objectid of the file holding the reference
1020 * - original offset in the file
1021 * - how many bookend extents
1022 *
1023 * The key offset for the implicit back refs is hash of the first
1024 * three fields.
1025 *
1026 * The extent ref structure for the full back refs has field for:
1027 *
1028 * - number of pointers in the tree leaf
1029 *
1030 * The key offset for the implicit back refs is the first byte of
1031 * the tree leaf
1032 *
1033 * When a file extent is allocated, The implicit back refs is used.
1034 * the fields are filled in:
1035 *
1036 * (root_key.objectid, inode objectid, offset in file, 1)
1037 *
1038 * When a file extent is removed file truncation, we find the
1039 * corresponding implicit back refs and check the following fields:
1040 *
1041 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1042 *
1043 * Btree extents can be referenced by:
1044 *
1045 * - Different subvolumes
1046 *
1047 * Both the implicit back refs and the full back refs for tree blocks
1048 * only consist of key. The key offset for the implicit back refs is
1049 * objectid of block's owner tree. The key offset for the full back refs
1050 * is the first byte of parent block.
1051 *
1052 * When implicit back refs is used, information about the lowest key and
1053 * level of the tree block are required. These information are stored in
1054 * tree block info structure.
1055 */
1056
1057#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1058static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1059 struct btrfs_root *root,
1060 struct btrfs_path *path,
1061 u64 owner, u32 extra_size)
1062{
1063 struct btrfs_extent_item *item;
1064 struct btrfs_extent_item_v0 *ei0;
1065 struct btrfs_extent_ref_v0 *ref0;
1066 struct btrfs_tree_block_info *bi;
1067 struct extent_buffer *leaf;
1068 struct btrfs_key key;
1069 struct btrfs_key found_key;
1070 u32 new_size = sizeof(*item);
1071 u64 refs;
1072 int ret;
1073
1074 leaf = path->nodes[0];
1075 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1076
1077 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1078 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1079 struct btrfs_extent_item_v0);
1080 refs = btrfs_extent_refs_v0(leaf, ei0);
1081
1082 if (owner == (u64)-1) {
1083 while (1) {
1084 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1085 ret = btrfs_next_leaf(root, path);
1086 if (ret < 0)
1087 return ret;
1088 BUG_ON(ret > 0); /* Corruption */
1089 leaf = path->nodes[0];
1090 }
1091 btrfs_item_key_to_cpu(leaf, &found_key,
1092 path->slots[0]);
1093 BUG_ON(key.objectid != found_key.objectid);
1094 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1095 path->slots[0]++;
1096 continue;
1097 }
1098 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1099 struct btrfs_extent_ref_v0);
1100 owner = btrfs_ref_objectid_v0(leaf, ref0);
1101 break;
1102 }
1103 }
1104 btrfs_release_path(path);
1105
1106 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1107 new_size += sizeof(*bi);
1108
1109 new_size -= sizeof(*ei0);
1110 ret = btrfs_search_slot(trans, root, &key, path,
1111 new_size + extra_size, 1);
1112 if (ret < 0)
1113 return ret;
1114 BUG_ON(ret); /* Corruption */
1115
1116 btrfs_extend_item(root, path, new_size);
1117
1118 leaf = path->nodes[0];
1119 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1120 btrfs_set_extent_refs(leaf, item, refs);
1121 /* FIXME: get real generation */
1122 btrfs_set_extent_generation(leaf, item, 0);
1123 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1124 btrfs_set_extent_flags(leaf, item,
1125 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1126 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1127 bi = (struct btrfs_tree_block_info *)(item + 1);
1128 /* FIXME: get first key of the block */
1129 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1130 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1131 } else {
1132 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1133 }
1134 btrfs_mark_buffer_dirty(leaf);
1135 return 0;
1136}
1137#endif
1138
1139static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1140{
1141 u32 high_crc = ~(u32)0;
1142 u32 low_crc = ~(u32)0;
1143 __le64 lenum;
1144
1145 lenum = cpu_to_le64(root_objectid);
1146 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1147 lenum = cpu_to_le64(owner);
1148 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1149 lenum = cpu_to_le64(offset);
1150 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1151
1152 return ((u64)high_crc << 31) ^ (u64)low_crc;
1153}
1154
1155static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1156 struct btrfs_extent_data_ref *ref)
1157{
1158 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1159 btrfs_extent_data_ref_objectid(leaf, ref),
1160 btrfs_extent_data_ref_offset(leaf, ref));
1161}
1162
1163static int match_extent_data_ref(struct extent_buffer *leaf,
1164 struct btrfs_extent_data_ref *ref,
1165 u64 root_objectid, u64 owner, u64 offset)
1166{
1167 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1168 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1169 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1170 return 0;
1171 return 1;
1172}
1173
1174static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1175 struct btrfs_root *root,
1176 struct btrfs_path *path,
1177 u64 bytenr, u64 parent,
1178 u64 root_objectid,
1179 u64 owner, u64 offset)
1180{
1181 struct btrfs_key key;
1182 struct btrfs_extent_data_ref *ref;
1183 struct extent_buffer *leaf;
1184 u32 nritems;
1185 int ret;
1186 int recow;
1187 int err = -ENOENT;
1188
1189 key.objectid = bytenr;
1190 if (parent) {
1191 key.type = BTRFS_SHARED_DATA_REF_KEY;
1192 key.offset = parent;
1193 } else {
1194 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1195 key.offset = hash_extent_data_ref(root_objectid,
1196 owner, offset);
1197 }
1198again:
1199 recow = 0;
1200 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1201 if (ret < 0) {
1202 err = ret;
1203 goto fail;
1204 }
1205
1206 if (parent) {
1207 if (!ret)
1208 return 0;
1209#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1210 key.type = BTRFS_EXTENT_REF_V0_KEY;
1211 btrfs_release_path(path);
1212 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1213 if (ret < 0) {
1214 err = ret;
1215 goto fail;
1216 }
1217 if (!ret)
1218 return 0;
1219#endif
1220 goto fail;
1221 }
1222
1223 leaf = path->nodes[0];
1224 nritems = btrfs_header_nritems(leaf);
1225 while (1) {
1226 if (path->slots[0] >= nritems) {
1227 ret = btrfs_next_leaf(root, path);
1228 if (ret < 0)
1229 err = ret;
1230 if (ret)
1231 goto fail;
1232
1233 leaf = path->nodes[0];
1234 nritems = btrfs_header_nritems(leaf);
1235 recow = 1;
1236 }
1237
1238 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1239 if (key.objectid != bytenr ||
1240 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1241 goto fail;
1242
1243 ref = btrfs_item_ptr(leaf, path->slots[0],
1244 struct btrfs_extent_data_ref);
1245
1246 if (match_extent_data_ref(leaf, ref, root_objectid,
1247 owner, offset)) {
1248 if (recow) {
1249 btrfs_release_path(path);
1250 goto again;
1251 }
1252 err = 0;
1253 break;
1254 }
1255 path->slots[0]++;
1256 }
1257fail:
1258 return err;
1259}
1260
1261static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1262 struct btrfs_root *root,
1263 struct btrfs_path *path,
1264 u64 bytenr, u64 parent,
1265 u64 root_objectid, u64 owner,
1266 u64 offset, int refs_to_add)
1267{
1268 struct btrfs_key key;
1269 struct extent_buffer *leaf;
1270 u32 size;
1271 u32 num_refs;
1272 int ret;
1273
1274 key.objectid = bytenr;
1275 if (parent) {
1276 key.type = BTRFS_SHARED_DATA_REF_KEY;
1277 key.offset = parent;
1278 size = sizeof(struct btrfs_shared_data_ref);
1279 } else {
1280 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1281 key.offset = hash_extent_data_ref(root_objectid,
1282 owner, offset);
1283 size = sizeof(struct btrfs_extent_data_ref);
1284 }
1285
1286 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1287 if (ret && ret != -EEXIST)
1288 goto fail;
1289
1290 leaf = path->nodes[0];
1291 if (parent) {
1292 struct btrfs_shared_data_ref *ref;
1293 ref = btrfs_item_ptr(leaf, path->slots[0],
1294 struct btrfs_shared_data_ref);
1295 if (ret == 0) {
1296 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1297 } else {
1298 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1299 num_refs += refs_to_add;
1300 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1301 }
1302 } else {
1303 struct btrfs_extent_data_ref *ref;
1304 while (ret == -EEXIST) {
1305 ref = btrfs_item_ptr(leaf, path->slots[0],
1306 struct btrfs_extent_data_ref);
1307 if (match_extent_data_ref(leaf, ref, root_objectid,
1308 owner, offset))
1309 break;
1310 btrfs_release_path(path);
1311 key.offset++;
1312 ret = btrfs_insert_empty_item(trans, root, path, &key,
1313 size);
1314 if (ret && ret != -EEXIST)
1315 goto fail;
1316
1317 leaf = path->nodes[0];
1318 }
1319 ref = btrfs_item_ptr(leaf, path->slots[0],
1320 struct btrfs_extent_data_ref);
1321 if (ret == 0) {
1322 btrfs_set_extent_data_ref_root(leaf, ref,
1323 root_objectid);
1324 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1325 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1326 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1327 } else {
1328 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1329 num_refs += refs_to_add;
1330 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1331 }
1332 }
1333 btrfs_mark_buffer_dirty(leaf);
1334 ret = 0;
1335fail:
1336 btrfs_release_path(path);
1337 return ret;
1338}
1339
1340static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1341 struct btrfs_root *root,
1342 struct btrfs_path *path,
1343 int refs_to_drop, int *last_ref)
1344{
1345 struct btrfs_key key;
1346 struct btrfs_extent_data_ref *ref1 = NULL;
1347 struct btrfs_shared_data_ref *ref2 = NULL;
1348 struct extent_buffer *leaf;
1349 u32 num_refs = 0;
1350 int ret = 0;
1351
1352 leaf = path->nodes[0];
1353 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1354
1355 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1356 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1357 struct btrfs_extent_data_ref);
1358 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1359 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1360 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1361 struct btrfs_shared_data_ref);
1362 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1363#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1364 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1365 struct btrfs_extent_ref_v0 *ref0;
1366 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1367 struct btrfs_extent_ref_v0);
1368 num_refs = btrfs_ref_count_v0(leaf, ref0);
1369#endif
1370 } else {
1371 BUG();
1372 }
1373
1374 BUG_ON(num_refs < refs_to_drop);
1375 num_refs -= refs_to_drop;
1376
1377 if (num_refs == 0) {
1378 ret = btrfs_del_item(trans, root, path);
1379 *last_ref = 1;
1380 } else {
1381 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1382 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1383 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1384 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1385#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1386 else {
1387 struct btrfs_extent_ref_v0 *ref0;
1388 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1389 struct btrfs_extent_ref_v0);
1390 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1391 }
1392#endif
1393 btrfs_mark_buffer_dirty(leaf);
1394 }
1395 return ret;
1396}
1397
1398static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1399 struct btrfs_extent_inline_ref *iref)
1400{
1401 struct btrfs_key key;
1402 struct extent_buffer *leaf;
1403 struct btrfs_extent_data_ref *ref1;
1404 struct btrfs_shared_data_ref *ref2;
1405 u32 num_refs = 0;
1406
1407 leaf = path->nodes[0];
1408 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1409 if (iref) {
1410 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1411 BTRFS_EXTENT_DATA_REF_KEY) {
1412 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1413 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1414 } else {
1415 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1416 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1417 }
1418 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1419 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1420 struct btrfs_extent_data_ref);
1421 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1422 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1423 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1424 struct btrfs_shared_data_ref);
1425 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1426#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1427 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1428 struct btrfs_extent_ref_v0 *ref0;
1429 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1430 struct btrfs_extent_ref_v0);
1431 num_refs = btrfs_ref_count_v0(leaf, ref0);
1432#endif
1433 } else {
1434 WARN_ON(1);
1435 }
1436 return num_refs;
1437}
1438
1439static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1440 struct btrfs_root *root,
1441 struct btrfs_path *path,
1442 u64 bytenr, u64 parent,
1443 u64 root_objectid)
1444{
1445 struct btrfs_key key;
1446 int ret;
1447
1448 key.objectid = bytenr;
1449 if (parent) {
1450 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1451 key.offset = parent;
1452 } else {
1453 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1454 key.offset = root_objectid;
1455 }
1456
1457 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1458 if (ret > 0)
1459 ret = -ENOENT;
1460#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1461 if (ret == -ENOENT && parent) {
1462 btrfs_release_path(path);
1463 key.type = BTRFS_EXTENT_REF_V0_KEY;
1464 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1465 if (ret > 0)
1466 ret = -ENOENT;
1467 }
1468#endif
1469 return ret;
1470}
1471
1472static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1473 struct btrfs_root *root,
1474 struct btrfs_path *path,
1475 u64 bytenr, u64 parent,
1476 u64 root_objectid)
1477{
1478 struct btrfs_key key;
1479 int ret;
1480
1481 key.objectid = bytenr;
1482 if (parent) {
1483 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1484 key.offset = parent;
1485 } else {
1486 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1487 key.offset = root_objectid;
1488 }
1489
1490 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1491 btrfs_release_path(path);
1492 return ret;
1493}
1494
1495static inline int extent_ref_type(u64 parent, u64 owner)
1496{
1497 int type;
1498 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1499 if (parent > 0)
1500 type = BTRFS_SHARED_BLOCK_REF_KEY;
1501 else
1502 type = BTRFS_TREE_BLOCK_REF_KEY;
1503 } else {
1504 if (parent > 0)
1505 type = BTRFS_SHARED_DATA_REF_KEY;
1506 else
1507 type = BTRFS_EXTENT_DATA_REF_KEY;
1508 }
1509 return type;
1510}
1511
1512static int find_next_key(struct btrfs_path *path, int level,
1513 struct btrfs_key *key)
1514
1515{
1516 for (; level < BTRFS_MAX_LEVEL; level++) {
1517 if (!path->nodes[level])
1518 break;
1519 if (path->slots[level] + 1 >=
1520 btrfs_header_nritems(path->nodes[level]))
1521 continue;
1522 if (level == 0)
1523 btrfs_item_key_to_cpu(path->nodes[level], key,
1524 path->slots[level] + 1);
1525 else
1526 btrfs_node_key_to_cpu(path->nodes[level], key,
1527 path->slots[level] + 1);
1528 return 0;
1529 }
1530 return 1;
1531}
1532
1533/*
1534 * look for inline back ref. if back ref is found, *ref_ret is set
1535 * to the address of inline back ref, and 0 is returned.
1536 *
1537 * if back ref isn't found, *ref_ret is set to the address where it
1538 * should be inserted, and -ENOENT is returned.
1539 *
1540 * if insert is true and there are too many inline back refs, the path
1541 * points to the extent item, and -EAGAIN is returned.
1542 *
1543 * NOTE: inline back refs are ordered in the same way that back ref
1544 * items in the tree are ordered.
1545 */
1546static noinline_for_stack
1547int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1548 struct btrfs_root *root,
1549 struct btrfs_path *path,
1550 struct btrfs_extent_inline_ref **ref_ret,
1551 u64 bytenr, u64 num_bytes,
1552 u64 parent, u64 root_objectid,
1553 u64 owner, u64 offset, int insert)
1554{
1555 struct btrfs_key key;
1556 struct extent_buffer *leaf;
1557 struct btrfs_extent_item *ei;
1558 struct btrfs_extent_inline_ref *iref;
1559 u64 flags;
1560 u64 item_size;
1561 unsigned long ptr;
1562 unsigned long end;
1563 int extra_size;
1564 int type;
1565 int want;
1566 int ret;
1567 int err = 0;
1568 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1569 SKINNY_METADATA);
1570
1571 key.objectid = bytenr;
1572 key.type = BTRFS_EXTENT_ITEM_KEY;
1573 key.offset = num_bytes;
1574
1575 want = extent_ref_type(parent, owner);
1576 if (insert) {
1577 extra_size = btrfs_extent_inline_ref_size(want);
1578 path->keep_locks = 1;
1579 } else
1580 extra_size = -1;
1581
1582 /*
1583 * Owner is our parent level, so we can just add one to get the level
1584 * for the block we are interested in.
1585 */
1586 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1587 key.type = BTRFS_METADATA_ITEM_KEY;
1588 key.offset = owner;
1589 }
1590
1591again:
1592 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1593 if (ret < 0) {
1594 err = ret;
1595 goto out;
1596 }
1597
1598 /*
1599 * We may be a newly converted file system which still has the old fat
1600 * extent entries for metadata, so try and see if we have one of those.
1601 */
1602 if (ret > 0 && skinny_metadata) {
1603 skinny_metadata = false;
1604 if (path->slots[0]) {
1605 path->slots[0]--;
1606 btrfs_item_key_to_cpu(path->nodes[0], &key,
1607 path->slots[0]);
1608 if (key.objectid == bytenr &&
1609 key.type == BTRFS_EXTENT_ITEM_KEY &&
1610 key.offset == num_bytes)
1611 ret = 0;
1612 }
1613 if (ret) {
1614 key.objectid = bytenr;
1615 key.type = BTRFS_EXTENT_ITEM_KEY;
1616 key.offset = num_bytes;
1617 btrfs_release_path(path);
1618 goto again;
1619 }
1620 }
1621
1622 if (ret && !insert) {
1623 err = -ENOENT;
1624 goto out;
1625 } else if (WARN_ON(ret)) {
1626 err = -EIO;
1627 goto out;
1628 }
1629
1630 leaf = path->nodes[0];
1631 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1632#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1633 if (item_size < sizeof(*ei)) {
1634 if (!insert) {
1635 err = -ENOENT;
1636 goto out;
1637 }
1638 ret = convert_extent_item_v0(trans, root, path, owner,
1639 extra_size);
1640 if (ret < 0) {
1641 err = ret;
1642 goto out;
1643 }
1644 leaf = path->nodes[0];
1645 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1646 }
1647#endif
1648 BUG_ON(item_size < sizeof(*ei));
1649
1650 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1651 flags = btrfs_extent_flags(leaf, ei);
1652
1653 ptr = (unsigned long)(ei + 1);
1654 end = (unsigned long)ei + item_size;
1655
1656 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1657 ptr += sizeof(struct btrfs_tree_block_info);
1658 BUG_ON(ptr > end);
1659 }
1660
1661 err = -ENOENT;
1662 while (1) {
1663 if (ptr >= end) {
1664 WARN_ON(ptr > end);
1665 break;
1666 }
1667 iref = (struct btrfs_extent_inline_ref *)ptr;
1668 type = btrfs_extent_inline_ref_type(leaf, iref);
1669 if (want < type)
1670 break;
1671 if (want > type) {
1672 ptr += btrfs_extent_inline_ref_size(type);
1673 continue;
1674 }
1675
1676 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1677 struct btrfs_extent_data_ref *dref;
1678 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1679 if (match_extent_data_ref(leaf, dref, root_objectid,
1680 owner, offset)) {
1681 err = 0;
1682 break;
1683 }
1684 if (hash_extent_data_ref_item(leaf, dref) <
1685 hash_extent_data_ref(root_objectid, owner, offset))
1686 break;
1687 } else {
1688 u64 ref_offset;
1689 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1690 if (parent > 0) {
1691 if (parent == ref_offset) {
1692 err = 0;
1693 break;
1694 }
1695 if (ref_offset < parent)
1696 break;
1697 } else {
1698 if (root_objectid == ref_offset) {
1699 err = 0;
1700 break;
1701 }
1702 if (ref_offset < root_objectid)
1703 break;
1704 }
1705 }
1706 ptr += btrfs_extent_inline_ref_size(type);
1707 }
1708 if (err == -ENOENT && insert) {
1709 if (item_size + extra_size >=
1710 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1711 err = -EAGAIN;
1712 goto out;
1713 }
1714 /*
1715 * To add new inline back ref, we have to make sure
1716 * there is no corresponding back ref item.
1717 * For simplicity, we just do not add new inline back
1718 * ref if there is any kind of item for this block
1719 */
1720 if (find_next_key(path, 0, &key) == 0 &&
1721 key.objectid == bytenr &&
1722 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1723 err = -EAGAIN;
1724 goto out;
1725 }
1726 }
1727 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1728out:
1729 if (insert) {
1730 path->keep_locks = 0;
1731 btrfs_unlock_up_safe(path, 1);
1732 }
1733 return err;
1734}
1735
1736/*
1737 * helper to add new inline back ref
1738 */
1739static noinline_for_stack
1740void setup_inline_extent_backref(struct btrfs_root *root,
1741 struct btrfs_path *path,
1742 struct btrfs_extent_inline_ref *iref,
1743 u64 parent, u64 root_objectid,
1744 u64 owner, u64 offset, int refs_to_add,
1745 struct btrfs_delayed_extent_op *extent_op)
1746{
1747 struct extent_buffer *leaf;
1748 struct btrfs_extent_item *ei;
1749 unsigned long ptr;
1750 unsigned long end;
1751 unsigned long item_offset;
1752 u64 refs;
1753 int size;
1754 int type;
1755
1756 leaf = path->nodes[0];
1757 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1758 item_offset = (unsigned long)iref - (unsigned long)ei;
1759
1760 type = extent_ref_type(parent, owner);
1761 size = btrfs_extent_inline_ref_size(type);
1762
1763 btrfs_extend_item(root, path, size);
1764
1765 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1766 refs = btrfs_extent_refs(leaf, ei);
1767 refs += refs_to_add;
1768 btrfs_set_extent_refs(leaf, ei, refs);
1769 if (extent_op)
1770 __run_delayed_extent_op(extent_op, leaf, ei);
1771
1772 ptr = (unsigned long)ei + item_offset;
1773 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1774 if (ptr < end - size)
1775 memmove_extent_buffer(leaf, ptr + size, ptr,
1776 end - size - ptr);
1777
1778 iref = (struct btrfs_extent_inline_ref *)ptr;
1779 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1780 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1781 struct btrfs_extent_data_ref *dref;
1782 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1783 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1784 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1785 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1786 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1787 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1788 struct btrfs_shared_data_ref *sref;
1789 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1790 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1791 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1792 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1793 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1794 } else {
1795 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1796 }
1797 btrfs_mark_buffer_dirty(leaf);
1798}
1799
1800static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1801 struct btrfs_root *root,
1802 struct btrfs_path *path,
1803 struct btrfs_extent_inline_ref **ref_ret,
1804 u64 bytenr, u64 num_bytes, u64 parent,
1805 u64 root_objectid, u64 owner, u64 offset)
1806{
1807 int ret;
1808
1809 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1810 bytenr, num_bytes, parent,
1811 root_objectid, owner, offset, 0);
1812 if (ret != -ENOENT)
1813 return ret;
1814
1815 btrfs_release_path(path);
1816 *ref_ret = NULL;
1817
1818 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1819 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1820 root_objectid);
1821 } else {
1822 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1823 root_objectid, owner, offset);
1824 }
1825 return ret;
1826}
1827
1828/*
1829 * helper to update/remove inline back ref
1830 */
1831static noinline_for_stack
1832void update_inline_extent_backref(struct btrfs_root *root,
1833 struct btrfs_path *path,
1834 struct btrfs_extent_inline_ref *iref,
1835 int refs_to_mod,
1836 struct btrfs_delayed_extent_op *extent_op,
1837 int *last_ref)
1838{
1839 struct extent_buffer *leaf;
1840 struct btrfs_extent_item *ei;
1841 struct btrfs_extent_data_ref *dref = NULL;
1842 struct btrfs_shared_data_ref *sref = NULL;
1843 unsigned long ptr;
1844 unsigned long end;
1845 u32 item_size;
1846 int size;
1847 int type;
1848 u64 refs;
1849
1850 leaf = path->nodes[0];
1851 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1852 refs = btrfs_extent_refs(leaf, ei);
1853 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1854 refs += refs_to_mod;
1855 btrfs_set_extent_refs(leaf, ei, refs);
1856 if (extent_op)
1857 __run_delayed_extent_op(extent_op, leaf, ei);
1858
1859 type = btrfs_extent_inline_ref_type(leaf, iref);
1860
1861 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1862 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1863 refs = btrfs_extent_data_ref_count(leaf, dref);
1864 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1865 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1866 refs = btrfs_shared_data_ref_count(leaf, sref);
1867 } else {
1868 refs = 1;
1869 BUG_ON(refs_to_mod != -1);
1870 }
1871
1872 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1873 refs += refs_to_mod;
1874
1875 if (refs > 0) {
1876 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1877 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1878 else
1879 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1880 } else {
1881 *last_ref = 1;
1882 size = btrfs_extent_inline_ref_size(type);
1883 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1884 ptr = (unsigned long)iref;
1885 end = (unsigned long)ei + item_size;
1886 if (ptr + size < end)
1887 memmove_extent_buffer(leaf, ptr, ptr + size,
1888 end - ptr - size);
1889 item_size -= size;
1890 btrfs_truncate_item(root, path, item_size, 1);
1891 }
1892 btrfs_mark_buffer_dirty(leaf);
1893}
1894
1895static noinline_for_stack
1896int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1897 struct btrfs_root *root,
1898 struct btrfs_path *path,
1899 u64 bytenr, u64 num_bytes, u64 parent,
1900 u64 root_objectid, u64 owner,
1901 u64 offset, int refs_to_add,
1902 struct btrfs_delayed_extent_op *extent_op)
1903{
1904 struct btrfs_extent_inline_ref *iref;
1905 int ret;
1906
1907 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1908 bytenr, num_bytes, parent,
1909 root_objectid, owner, offset, 1);
1910 if (ret == 0) {
1911 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1912 update_inline_extent_backref(root, path, iref,
1913 refs_to_add, extent_op, NULL);
1914 } else if (ret == -ENOENT) {
1915 setup_inline_extent_backref(root, path, iref, parent,
1916 root_objectid, owner, offset,
1917 refs_to_add, extent_op);
1918 ret = 0;
1919 }
1920 return ret;
1921}
1922
1923static int insert_extent_backref(struct btrfs_trans_handle *trans,
1924 struct btrfs_root *root,
1925 struct btrfs_path *path,
1926 u64 bytenr, u64 parent, u64 root_objectid,
1927 u64 owner, u64 offset, int refs_to_add)
1928{
1929 int ret;
1930 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1931 BUG_ON(refs_to_add != 1);
1932 ret = insert_tree_block_ref(trans, root, path, bytenr,
1933 parent, root_objectid);
1934 } else {
1935 ret = insert_extent_data_ref(trans, root, path, bytenr,
1936 parent, root_objectid,
1937 owner, offset, refs_to_add);
1938 }
1939 return ret;
1940}
1941
1942static int remove_extent_backref(struct btrfs_trans_handle *trans,
1943 struct btrfs_root *root,
1944 struct btrfs_path *path,
1945 struct btrfs_extent_inline_ref *iref,
1946 int refs_to_drop, int is_data, int *last_ref)
1947{
1948 int ret = 0;
1949
1950 BUG_ON(!is_data && refs_to_drop != 1);
1951 if (iref) {
1952 update_inline_extent_backref(root, path, iref,
1953 -refs_to_drop, NULL, last_ref);
1954 } else if (is_data) {
1955 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1956 last_ref);
1957 } else {
1958 *last_ref = 1;
1959 ret = btrfs_del_item(trans, root, path);
1960 }
1961 return ret;
1962}
1963
1964#define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1965static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1966 u64 *discarded_bytes)
1967{
1968 int j, ret = 0;
1969 u64 bytes_left, end;
1970 u64 aligned_start = ALIGN(start, 1 << 9);
1971
1972 if (WARN_ON(start != aligned_start)) {
1973 len -= aligned_start - start;
1974 len = round_down(len, 1 << 9);
1975 start = aligned_start;
1976 }
1977
1978 *discarded_bytes = 0;
1979
1980 if (!len)
1981 return 0;
1982
1983 end = start + len;
1984 bytes_left = len;
1985
1986 /* Skip any superblocks on this device. */
1987 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1988 u64 sb_start = btrfs_sb_offset(j);
1989 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1990 u64 size = sb_start - start;
1991
1992 if (!in_range(sb_start, start, bytes_left) &&
1993 !in_range(sb_end, start, bytes_left) &&
1994 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1995 continue;
1996
1997 /*
1998 * Superblock spans beginning of range. Adjust start and
1999 * try again.
2000 */
2001 if (sb_start <= start) {
2002 start += sb_end - start;
2003 if (start > end) {
2004 bytes_left = 0;
2005 break;
2006 }
2007 bytes_left = end - start;
2008 continue;
2009 }
2010
2011 if (size) {
2012 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2013 GFP_NOFS, 0);
2014 if (!ret)
2015 *discarded_bytes += size;
2016 else if (ret != -EOPNOTSUPP)
2017 return ret;
2018 }
2019
2020 start = sb_end;
2021 if (start > end) {
2022 bytes_left = 0;
2023 break;
2024 }
2025 bytes_left = end - start;
2026 }
2027
2028 if (bytes_left) {
2029 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2030 GFP_NOFS, 0);
2031 if (!ret)
2032 *discarded_bytes += bytes_left;
2033 }
2034 return ret;
2035}
2036
2037int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
2038 u64 num_bytes, u64 *actual_bytes)
2039{
2040 int ret;
2041 u64 discarded_bytes = 0;
2042 struct btrfs_bio *bbio = NULL;
2043
2044
2045 /* Tell the block device(s) that the sectors can be discarded */
2046 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
2047 bytenr, &num_bytes, &bbio, 0);
2048 /* Error condition is -ENOMEM */
2049 if (!ret) {
2050 struct btrfs_bio_stripe *stripe = bbio->stripes;
2051 int i;
2052
2053
2054 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2055 u64 bytes;
2056 if (!stripe->dev->can_discard)
2057 continue;
2058
2059 ret = btrfs_issue_discard(stripe->dev->bdev,
2060 stripe->physical,
2061 stripe->length,
2062 &bytes);
2063 if (!ret)
2064 discarded_bytes += bytes;
2065 else if (ret != -EOPNOTSUPP)
2066 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2067
2068 /*
2069 * Just in case we get back EOPNOTSUPP for some reason,
2070 * just ignore the return value so we don't screw up
2071 * people calling discard_extent.
2072 */
2073 ret = 0;
2074 }
2075 btrfs_put_bbio(bbio);
2076 }
2077
2078 if (actual_bytes)
2079 *actual_bytes = discarded_bytes;
2080
2081
2082 if (ret == -EOPNOTSUPP)
2083 ret = 0;
2084 return ret;
2085}
2086
2087/* Can return -ENOMEM */
2088int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2089 struct btrfs_root *root,
2090 u64 bytenr, u64 num_bytes, u64 parent,
2091 u64 root_objectid, u64 owner, u64 offset)
2092{
2093 int ret;
2094 struct btrfs_fs_info *fs_info = root->fs_info;
2095
2096 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2097 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2098
2099 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2100 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2101 num_bytes,
2102 parent, root_objectid, (int)owner,
2103 BTRFS_ADD_DELAYED_REF, NULL);
2104 } else {
2105 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2106 num_bytes, parent, root_objectid,
2107 owner, offset, 0,
2108 BTRFS_ADD_DELAYED_REF, NULL);
2109 }
2110 return ret;
2111}
2112
2113static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2114 struct btrfs_root *root,
2115 struct btrfs_delayed_ref_node *node,
2116 u64 parent, u64 root_objectid,
2117 u64 owner, u64 offset, int refs_to_add,
2118 struct btrfs_delayed_extent_op *extent_op)
2119{
2120 struct btrfs_fs_info *fs_info = root->fs_info;
2121 struct btrfs_path *path;
2122 struct extent_buffer *leaf;
2123 struct btrfs_extent_item *item;
2124 struct btrfs_key key;
2125 u64 bytenr = node->bytenr;
2126 u64 num_bytes = node->num_bytes;
2127 u64 refs;
2128 int ret;
2129
2130 path = btrfs_alloc_path();
2131 if (!path)
2132 return -ENOMEM;
2133
2134 path->reada = READA_FORWARD;
2135 path->leave_spinning = 1;
2136 /* this will setup the path even if it fails to insert the back ref */
2137 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
2138 bytenr, num_bytes, parent,
2139 root_objectid, owner, offset,
2140 refs_to_add, extent_op);
2141 if ((ret < 0 && ret != -EAGAIN) || !ret)
2142 goto out;
2143
2144 /*
2145 * Ok we had -EAGAIN which means we didn't have space to insert and
2146 * inline extent ref, so just update the reference count and add a
2147 * normal backref.
2148 */
2149 leaf = path->nodes[0];
2150 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2151 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2152 refs = btrfs_extent_refs(leaf, item);
2153 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2154 if (extent_op)
2155 __run_delayed_extent_op(extent_op, leaf, item);
2156
2157 btrfs_mark_buffer_dirty(leaf);
2158 btrfs_release_path(path);
2159
2160 path->reada = READA_FORWARD;
2161 path->leave_spinning = 1;
2162 /* now insert the actual backref */
2163 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2164 path, bytenr, parent, root_objectid,
2165 owner, offset, refs_to_add);
2166 if (ret)
2167 btrfs_abort_transaction(trans, root, ret);
2168out:
2169 btrfs_free_path(path);
2170 return ret;
2171}
2172
2173static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2174 struct btrfs_root *root,
2175 struct btrfs_delayed_ref_node *node,
2176 struct btrfs_delayed_extent_op *extent_op,
2177 int insert_reserved)
2178{
2179 int ret = 0;
2180 struct btrfs_delayed_data_ref *ref;
2181 struct btrfs_key ins;
2182 u64 parent = 0;
2183 u64 ref_root = 0;
2184 u64 flags = 0;
2185
2186 ins.objectid = node->bytenr;
2187 ins.offset = node->num_bytes;
2188 ins.type = BTRFS_EXTENT_ITEM_KEY;
2189
2190 ref = btrfs_delayed_node_to_data_ref(node);
2191 trace_run_delayed_data_ref(node, ref, node->action);
2192
2193 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2194 parent = ref->parent;
2195 ref_root = ref->root;
2196
2197 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2198 if (extent_op)
2199 flags |= extent_op->flags_to_set;
2200 ret = alloc_reserved_file_extent(trans, root,
2201 parent, ref_root, flags,
2202 ref->objectid, ref->offset,
2203 &ins, node->ref_mod);
2204 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2205 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2206 ref_root, ref->objectid,
2207 ref->offset, node->ref_mod,
2208 extent_op);
2209 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2210 ret = __btrfs_free_extent(trans, root, node, parent,
2211 ref_root, ref->objectid,
2212 ref->offset, node->ref_mod,
2213 extent_op);
2214 } else {
2215 BUG();
2216 }
2217 return ret;
2218}
2219
2220static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2221 struct extent_buffer *leaf,
2222 struct btrfs_extent_item *ei)
2223{
2224 u64 flags = btrfs_extent_flags(leaf, ei);
2225 if (extent_op->update_flags) {
2226 flags |= extent_op->flags_to_set;
2227 btrfs_set_extent_flags(leaf, ei, flags);
2228 }
2229
2230 if (extent_op->update_key) {
2231 struct btrfs_tree_block_info *bi;
2232 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2233 bi = (struct btrfs_tree_block_info *)(ei + 1);
2234 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2235 }
2236}
2237
2238static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2239 struct btrfs_root *root,
2240 struct btrfs_delayed_ref_node *node,
2241 struct btrfs_delayed_extent_op *extent_op)
2242{
2243 struct btrfs_key key;
2244 struct btrfs_path *path;
2245 struct btrfs_extent_item *ei;
2246 struct extent_buffer *leaf;
2247 u32 item_size;
2248 int ret;
2249 int err = 0;
2250 int metadata = !extent_op->is_data;
2251
2252 if (trans->aborted)
2253 return 0;
2254
2255 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2256 metadata = 0;
2257
2258 path = btrfs_alloc_path();
2259 if (!path)
2260 return -ENOMEM;
2261
2262 key.objectid = node->bytenr;
2263
2264 if (metadata) {
2265 key.type = BTRFS_METADATA_ITEM_KEY;
2266 key.offset = extent_op->level;
2267 } else {
2268 key.type = BTRFS_EXTENT_ITEM_KEY;
2269 key.offset = node->num_bytes;
2270 }
2271
2272again:
2273 path->reada = READA_FORWARD;
2274 path->leave_spinning = 1;
2275 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2276 path, 0, 1);
2277 if (ret < 0) {
2278 err = ret;
2279 goto out;
2280 }
2281 if (ret > 0) {
2282 if (metadata) {
2283 if (path->slots[0] > 0) {
2284 path->slots[0]--;
2285 btrfs_item_key_to_cpu(path->nodes[0], &key,
2286 path->slots[0]);
2287 if (key.objectid == node->bytenr &&
2288 key.type == BTRFS_EXTENT_ITEM_KEY &&
2289 key.offset == node->num_bytes)
2290 ret = 0;
2291 }
2292 if (ret > 0) {
2293 btrfs_release_path(path);
2294 metadata = 0;
2295
2296 key.objectid = node->bytenr;
2297 key.offset = node->num_bytes;
2298 key.type = BTRFS_EXTENT_ITEM_KEY;
2299 goto again;
2300 }
2301 } else {
2302 err = -EIO;
2303 goto out;
2304 }
2305 }
2306
2307 leaf = path->nodes[0];
2308 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2309#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2310 if (item_size < sizeof(*ei)) {
2311 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2312 path, (u64)-1, 0);
2313 if (ret < 0) {
2314 err = ret;
2315 goto out;
2316 }
2317 leaf = path->nodes[0];
2318 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2319 }
2320#endif
2321 BUG_ON(item_size < sizeof(*ei));
2322 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2323 __run_delayed_extent_op(extent_op, leaf, ei);
2324
2325 btrfs_mark_buffer_dirty(leaf);
2326out:
2327 btrfs_free_path(path);
2328 return err;
2329}
2330
2331static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2332 struct btrfs_root *root,
2333 struct btrfs_delayed_ref_node *node,
2334 struct btrfs_delayed_extent_op *extent_op,
2335 int insert_reserved)
2336{
2337 int ret = 0;
2338 struct btrfs_delayed_tree_ref *ref;
2339 struct btrfs_key ins;
2340 u64 parent = 0;
2341 u64 ref_root = 0;
2342 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2343 SKINNY_METADATA);
2344
2345 ref = btrfs_delayed_node_to_tree_ref(node);
2346 trace_run_delayed_tree_ref(node, ref, node->action);
2347
2348 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2349 parent = ref->parent;
2350 ref_root = ref->root;
2351
2352 ins.objectid = node->bytenr;
2353 if (skinny_metadata) {
2354 ins.offset = ref->level;
2355 ins.type = BTRFS_METADATA_ITEM_KEY;
2356 } else {
2357 ins.offset = node->num_bytes;
2358 ins.type = BTRFS_EXTENT_ITEM_KEY;
2359 }
2360
2361 BUG_ON(node->ref_mod != 1);
2362 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2363 BUG_ON(!extent_op || !extent_op->update_flags);
2364 ret = alloc_reserved_tree_block(trans, root,
2365 parent, ref_root,
2366 extent_op->flags_to_set,
2367 &extent_op->key,
2368 ref->level, &ins);
2369 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2370 ret = __btrfs_inc_extent_ref(trans, root, node,
2371 parent, ref_root,
2372 ref->level, 0, 1,
2373 extent_op);
2374 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2375 ret = __btrfs_free_extent(trans, root, node,
2376 parent, ref_root,
2377 ref->level, 0, 1, extent_op);
2378 } else {
2379 BUG();
2380 }
2381 return ret;
2382}
2383
2384/* helper function to actually process a single delayed ref entry */
2385static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2386 struct btrfs_root *root,
2387 struct btrfs_delayed_ref_node *node,
2388 struct btrfs_delayed_extent_op *extent_op,
2389 int insert_reserved)
2390{
2391 int ret = 0;
2392
2393 if (trans->aborted) {
2394 if (insert_reserved)
2395 btrfs_pin_extent(root, node->bytenr,
2396 node->num_bytes, 1);
2397 return 0;
2398 }
2399
2400 if (btrfs_delayed_ref_is_head(node)) {
2401 struct btrfs_delayed_ref_head *head;
2402 /*
2403 * we've hit the end of the chain and we were supposed
2404 * to insert this extent into the tree. But, it got
2405 * deleted before we ever needed to insert it, so all
2406 * we have to do is clean up the accounting
2407 */
2408 BUG_ON(extent_op);
2409 head = btrfs_delayed_node_to_head(node);
2410 trace_run_delayed_ref_head(node, head, node->action);
2411
2412 if (insert_reserved) {
2413 btrfs_pin_extent(root, node->bytenr,
2414 node->num_bytes, 1);
2415 if (head->is_data) {
2416 ret = btrfs_del_csums(trans, root,
2417 node->bytenr,
2418 node->num_bytes);
2419 }
2420 }
2421
2422 /* Also free its reserved qgroup space */
2423 btrfs_qgroup_free_delayed_ref(root->fs_info,
2424 head->qgroup_ref_root,
2425 head->qgroup_reserved);
2426 return ret;
2427 }
2428
2429 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2430 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2431 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2432 insert_reserved);
2433 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2434 node->type == BTRFS_SHARED_DATA_REF_KEY)
2435 ret = run_delayed_data_ref(trans, root, node, extent_op,
2436 insert_reserved);
2437 else
2438 BUG();
2439 return ret;
2440}
2441
2442static inline struct btrfs_delayed_ref_node *
2443select_delayed_ref(struct btrfs_delayed_ref_head *head)
2444{
2445 struct btrfs_delayed_ref_node *ref;
2446
2447 if (list_empty(&head->ref_list))
2448 return NULL;
2449
2450 /*
2451 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2452 * This is to prevent a ref count from going down to zero, which deletes
2453 * the extent item from the extent tree, when there still are references
2454 * to add, which would fail because they would not find the extent item.
2455 */
2456 list_for_each_entry(ref, &head->ref_list, list) {
2457 if (ref->action == BTRFS_ADD_DELAYED_REF)
2458 return ref;
2459 }
2460
2461 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2462 list);
2463}
2464
2465/*
2466 * Returns 0 on success or if called with an already aborted transaction.
2467 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2468 */
2469static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2470 struct btrfs_root *root,
2471 unsigned long nr)
2472{
2473 struct btrfs_delayed_ref_root *delayed_refs;
2474 struct btrfs_delayed_ref_node *ref;
2475 struct btrfs_delayed_ref_head *locked_ref = NULL;
2476 struct btrfs_delayed_extent_op *extent_op;
2477 struct btrfs_fs_info *fs_info = root->fs_info;
2478 ktime_t start = ktime_get();
2479 int ret;
2480 unsigned long count = 0;
2481 unsigned long actual_count = 0;
2482 int must_insert_reserved = 0;
2483
2484 delayed_refs = &trans->transaction->delayed_refs;
2485 while (1) {
2486 if (!locked_ref) {
2487 if (count >= nr)
2488 break;
2489
2490 spin_lock(&delayed_refs->lock);
2491 locked_ref = btrfs_select_ref_head(trans);
2492 if (!locked_ref) {
2493 spin_unlock(&delayed_refs->lock);
2494 break;
2495 }
2496
2497 /* grab the lock that says we are going to process
2498 * all the refs for this head */
2499 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2500 spin_unlock(&delayed_refs->lock);
2501 /*
2502 * we may have dropped the spin lock to get the head
2503 * mutex lock, and that might have given someone else
2504 * time to free the head. If that's true, it has been
2505 * removed from our list and we can move on.
2506 */
2507 if (ret == -EAGAIN) {
2508 locked_ref = NULL;
2509 count++;
2510 continue;
2511 }
2512 }
2513
2514 /*
2515 * We need to try and merge add/drops of the same ref since we
2516 * can run into issues with relocate dropping the implicit ref
2517 * and then it being added back again before the drop can
2518 * finish. If we merged anything we need to re-loop so we can
2519 * get a good ref.
2520 * Or we can get node references of the same type that weren't
2521 * merged when created due to bumps in the tree mod seq, and
2522 * we need to merge them to prevent adding an inline extent
2523 * backref before dropping it (triggering a BUG_ON at
2524 * insert_inline_extent_backref()).
2525 */
2526 spin_lock(&locked_ref->lock);
2527 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2528 locked_ref);
2529
2530 /*
2531 * locked_ref is the head node, so we have to go one
2532 * node back for any delayed ref updates
2533 */
2534 ref = select_delayed_ref(locked_ref);
2535
2536 if (ref && ref->seq &&
2537 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2538 spin_unlock(&locked_ref->lock);
2539 btrfs_delayed_ref_unlock(locked_ref);
2540 spin_lock(&delayed_refs->lock);
2541 locked_ref->processing = 0;
2542 delayed_refs->num_heads_ready++;
2543 spin_unlock(&delayed_refs->lock);
2544 locked_ref = NULL;
2545 cond_resched();
2546 count++;
2547 continue;
2548 }
2549
2550 /*
2551 * record the must insert reserved flag before we
2552 * drop the spin lock.
2553 */
2554 must_insert_reserved = locked_ref->must_insert_reserved;
2555 locked_ref->must_insert_reserved = 0;
2556
2557 extent_op = locked_ref->extent_op;
2558 locked_ref->extent_op = NULL;
2559
2560 if (!ref) {
2561
2562
2563 /* All delayed refs have been processed, Go ahead
2564 * and send the head node to run_one_delayed_ref,
2565 * so that any accounting fixes can happen
2566 */
2567 ref = &locked_ref->node;
2568
2569 if (extent_op && must_insert_reserved) {
2570 btrfs_free_delayed_extent_op(extent_op);
2571 extent_op = NULL;
2572 }
2573
2574 if (extent_op) {
2575 spin_unlock(&locked_ref->lock);
2576 ret = run_delayed_extent_op(trans, root,
2577 ref, extent_op);
2578 btrfs_free_delayed_extent_op(extent_op);
2579
2580 if (ret) {
2581 /*
2582 * Need to reset must_insert_reserved if
2583 * there was an error so the abort stuff
2584 * can cleanup the reserved space
2585 * properly.
2586 */
2587 if (must_insert_reserved)
2588 locked_ref->must_insert_reserved = 1;
2589 locked_ref->processing = 0;
2590 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2591 btrfs_delayed_ref_unlock(locked_ref);
2592 return ret;
2593 }
2594 continue;
2595 }
2596
2597 /*
2598 * Need to drop our head ref lock and re-aqcuire the
2599 * delayed ref lock and then re-check to make sure
2600 * nobody got added.
2601 */
2602 spin_unlock(&locked_ref->lock);
2603 spin_lock(&delayed_refs->lock);
2604 spin_lock(&locked_ref->lock);
2605 if (!list_empty(&locked_ref->ref_list) ||
2606 locked_ref->extent_op) {
2607 spin_unlock(&locked_ref->lock);
2608 spin_unlock(&delayed_refs->lock);
2609 continue;
2610 }
2611 ref->in_tree = 0;
2612 delayed_refs->num_heads--;
2613 rb_erase(&locked_ref->href_node,
2614 &delayed_refs->href_root);
2615 spin_unlock(&delayed_refs->lock);
2616 } else {
2617 actual_count++;
2618 ref->in_tree = 0;
2619 list_del(&ref->list);
2620 }
2621 atomic_dec(&delayed_refs->num_entries);
2622
2623 if (!btrfs_delayed_ref_is_head(ref)) {
2624 /*
2625 * when we play the delayed ref, also correct the
2626 * ref_mod on head
2627 */
2628 switch (ref->action) {
2629 case BTRFS_ADD_DELAYED_REF:
2630 case BTRFS_ADD_DELAYED_EXTENT:
2631 locked_ref->node.ref_mod -= ref->ref_mod;
2632 break;
2633 case BTRFS_DROP_DELAYED_REF:
2634 locked_ref->node.ref_mod += ref->ref_mod;
2635 break;
2636 default:
2637 WARN_ON(1);
2638 }
2639 }
2640 spin_unlock(&locked_ref->lock);
2641
2642 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2643 must_insert_reserved);
2644
2645 btrfs_free_delayed_extent_op(extent_op);
2646 if (ret) {
2647 locked_ref->processing = 0;
2648 btrfs_delayed_ref_unlock(locked_ref);
2649 btrfs_put_delayed_ref(ref);
2650 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2651 return ret;
2652 }
2653
2654 /*
2655 * If this node is a head, that means all the refs in this head
2656 * have been dealt with, and we will pick the next head to deal
2657 * with, so we must unlock the head and drop it from the cluster
2658 * list before we release it.
2659 */
2660 if (btrfs_delayed_ref_is_head(ref)) {
2661 if (locked_ref->is_data &&
2662 locked_ref->total_ref_mod < 0) {
2663 spin_lock(&delayed_refs->lock);
2664 delayed_refs->pending_csums -= ref->num_bytes;
2665 spin_unlock(&delayed_refs->lock);
2666 }
2667 btrfs_delayed_ref_unlock(locked_ref);
2668 locked_ref = NULL;
2669 }
2670 btrfs_put_delayed_ref(ref);
2671 count++;
2672 cond_resched();
2673 }
2674
2675 /*
2676 * We don't want to include ref heads since we can have empty ref heads
2677 * and those will drastically skew our runtime down since we just do
2678 * accounting, no actual extent tree updates.
2679 */
2680 if (actual_count > 0) {
2681 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2682 u64 avg;
2683
2684 /*
2685 * We weigh the current average higher than our current runtime
2686 * to avoid large swings in the average.
2687 */
2688 spin_lock(&delayed_refs->lock);
2689 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2690 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2691 spin_unlock(&delayed_refs->lock);
2692 }
2693 return 0;
2694}
2695
2696#ifdef SCRAMBLE_DELAYED_REFS
2697/*
2698 * Normally delayed refs get processed in ascending bytenr order. This
2699 * correlates in most cases to the order added. To expose dependencies on this
2700 * order, we start to process the tree in the middle instead of the beginning
2701 */
2702static u64 find_middle(struct rb_root *root)
2703{
2704 struct rb_node *n = root->rb_node;
2705 struct btrfs_delayed_ref_node *entry;
2706 int alt = 1;
2707 u64 middle;
2708 u64 first = 0, last = 0;
2709
2710 n = rb_first(root);
2711 if (n) {
2712 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2713 first = entry->bytenr;
2714 }
2715 n = rb_last(root);
2716 if (n) {
2717 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2718 last = entry->bytenr;
2719 }
2720 n = root->rb_node;
2721
2722 while (n) {
2723 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2724 WARN_ON(!entry->in_tree);
2725
2726 middle = entry->bytenr;
2727
2728 if (alt)
2729 n = n->rb_left;
2730 else
2731 n = n->rb_right;
2732
2733 alt = 1 - alt;
2734 }
2735 return middle;
2736}
2737#endif
2738
2739static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2740{
2741 u64 num_bytes;
2742
2743 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2744 sizeof(struct btrfs_extent_inline_ref));
2745 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2746 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2747
2748 /*
2749 * We don't ever fill up leaves all the way so multiply by 2 just to be
2750 * closer to what we're really going to want to ouse.
2751 */
2752 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2753}
2754
2755/*
2756 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2757 * would require to store the csums for that many bytes.
2758 */
2759u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2760{
2761 u64 csum_size;
2762 u64 num_csums_per_leaf;
2763 u64 num_csums;
2764
2765 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2766 num_csums_per_leaf = div64_u64(csum_size,
2767 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2768 num_csums = div64_u64(csum_bytes, root->sectorsize);
2769 num_csums += num_csums_per_leaf - 1;
2770 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2771 return num_csums;
2772}
2773
2774int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2775 struct btrfs_root *root)
2776{
2777 struct btrfs_block_rsv *global_rsv;
2778 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2779 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2780 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2781 u64 num_bytes, num_dirty_bgs_bytes;
2782 int ret = 0;
2783
2784 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2785 num_heads = heads_to_leaves(root, num_heads);
2786 if (num_heads > 1)
2787 num_bytes += (num_heads - 1) * root->nodesize;
2788 num_bytes <<= 1;
2789 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2790 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2791 num_dirty_bgs);
2792 global_rsv = &root->fs_info->global_block_rsv;
2793
2794 /*
2795 * If we can't allocate any more chunks lets make sure we have _lots_ of
2796 * wiggle room since running delayed refs can create more delayed refs.
2797 */
2798 if (global_rsv->space_info->full) {
2799 num_dirty_bgs_bytes <<= 1;
2800 num_bytes <<= 1;
2801 }
2802
2803 spin_lock(&global_rsv->lock);
2804 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2805 ret = 1;
2806 spin_unlock(&global_rsv->lock);
2807 return ret;
2808}
2809
2810int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2811 struct btrfs_root *root)
2812{
2813 struct btrfs_fs_info *fs_info = root->fs_info;
2814 u64 num_entries =
2815 atomic_read(&trans->transaction->delayed_refs.num_entries);
2816 u64 avg_runtime;
2817 u64 val;
2818
2819 smp_mb();
2820 avg_runtime = fs_info->avg_delayed_ref_runtime;
2821 val = num_entries * avg_runtime;
2822 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2823 return 1;
2824 if (val >= NSEC_PER_SEC / 2)
2825 return 2;
2826
2827 return btrfs_check_space_for_delayed_refs(trans, root);
2828}
2829
2830struct async_delayed_refs {
2831 struct btrfs_root *root;
2832 int count;
2833 int error;
2834 int sync;
2835 struct completion wait;
2836 struct btrfs_work work;
2837};
2838
2839static void delayed_ref_async_start(struct btrfs_work *work)
2840{
2841 struct async_delayed_refs *async;
2842 struct btrfs_trans_handle *trans;
2843 int ret;
2844
2845 async = container_of(work, struct async_delayed_refs, work);
2846
2847 trans = btrfs_join_transaction(async->root);
2848 if (IS_ERR(trans)) {
2849 async->error = PTR_ERR(trans);
2850 goto done;
2851 }
2852
2853 /*
2854 * trans->sync means that when we call end_transaciton, we won't
2855 * wait on delayed refs
2856 */
2857 trans->sync = true;
2858 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2859 if (ret)
2860 async->error = ret;
2861
2862 ret = btrfs_end_transaction(trans, async->root);
2863 if (ret && !async->error)
2864 async->error = ret;
2865done:
2866 if (async->sync)
2867 complete(&async->wait);
2868 else
2869 kfree(async);
2870}
2871
2872int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2873 unsigned long count, int wait)
2874{
2875 struct async_delayed_refs *async;
2876 int ret;
2877
2878 async = kmalloc(sizeof(*async), GFP_NOFS);
2879 if (!async)
2880 return -ENOMEM;
2881
2882 async->root = root->fs_info->tree_root;
2883 async->count = count;
2884 async->error = 0;
2885 if (wait)
2886 async->sync = 1;
2887 else
2888 async->sync = 0;
2889 init_completion(&async->wait);
2890
2891 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2892 delayed_ref_async_start, NULL, NULL);
2893
2894 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2895
2896 if (wait) {
2897 wait_for_completion(&async->wait);
2898 ret = async->error;
2899 kfree(async);
2900 return ret;
2901 }
2902 return 0;
2903}
2904
2905/*
2906 * this starts processing the delayed reference count updates and
2907 * extent insertions we have queued up so far. count can be
2908 * 0, which means to process everything in the tree at the start
2909 * of the run (but not newly added entries), or it can be some target
2910 * number you'd like to process.
2911 *
2912 * Returns 0 on success or if called with an aborted transaction
2913 * Returns <0 on error and aborts the transaction
2914 */
2915int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2916 struct btrfs_root *root, unsigned long count)
2917{
2918 struct rb_node *node;
2919 struct btrfs_delayed_ref_root *delayed_refs;
2920 struct btrfs_delayed_ref_head *head;
2921 int ret;
2922 int run_all = count == (unsigned long)-1;
2923 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2924
2925 /* We'll clean this up in btrfs_cleanup_transaction */
2926 if (trans->aborted)
2927 return 0;
2928
2929 if (root->fs_info->creating_free_space_tree)
2930 return 0;
2931
2932 if (root == root->fs_info->extent_root)
2933 root = root->fs_info->tree_root;
2934
2935 delayed_refs = &trans->transaction->delayed_refs;
2936 if (count == 0)
2937 count = atomic_read(&delayed_refs->num_entries) * 2;
2938
2939again:
2940#ifdef SCRAMBLE_DELAYED_REFS
2941 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2942#endif
2943 trans->can_flush_pending_bgs = false;
2944 ret = __btrfs_run_delayed_refs(trans, root, count);
2945 if (ret < 0) {
2946 btrfs_abort_transaction(trans, root, ret);
2947 return ret;
2948 }
2949
2950 if (run_all) {
2951 if (!list_empty(&trans->new_bgs))
2952 btrfs_create_pending_block_groups(trans, root);
2953
2954 spin_lock(&delayed_refs->lock);
2955 node = rb_first(&delayed_refs->href_root);
2956 if (!node) {
2957 spin_unlock(&delayed_refs->lock);
2958 goto out;
2959 }
2960 count = (unsigned long)-1;
2961
2962 while (node) {
2963 head = rb_entry(node, struct btrfs_delayed_ref_head,
2964 href_node);
2965 if (btrfs_delayed_ref_is_head(&head->node)) {
2966 struct btrfs_delayed_ref_node *ref;
2967
2968 ref = &head->node;
2969 atomic_inc(&ref->refs);
2970
2971 spin_unlock(&delayed_refs->lock);
2972 /*
2973 * Mutex was contended, block until it's
2974 * released and try again
2975 */
2976 mutex_lock(&head->mutex);
2977 mutex_unlock(&head->mutex);
2978
2979 btrfs_put_delayed_ref(ref);
2980 cond_resched();
2981 goto again;
2982 } else {
2983 WARN_ON(1);
2984 }
2985 node = rb_next(node);
2986 }
2987 spin_unlock(&delayed_refs->lock);
2988 cond_resched();
2989 goto again;
2990 }
2991out:
2992 assert_qgroups_uptodate(trans);
2993 trans->can_flush_pending_bgs = can_flush_pending_bgs;
2994 return 0;
2995}
2996
2997int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2998 struct btrfs_root *root,
2999 u64 bytenr, u64 num_bytes, u64 flags,
3000 int level, int is_data)
3001{
3002 struct btrfs_delayed_extent_op *extent_op;
3003 int ret;
3004
3005 extent_op = btrfs_alloc_delayed_extent_op();
3006 if (!extent_op)
3007 return -ENOMEM;
3008
3009 extent_op->flags_to_set = flags;
3010 extent_op->update_flags = true;
3011 extent_op->update_key = false;
3012 extent_op->is_data = is_data ? true : false;
3013 extent_op->level = level;
3014
3015 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
3016 num_bytes, extent_op);
3017 if (ret)
3018 btrfs_free_delayed_extent_op(extent_op);
3019 return ret;
3020}
3021
3022static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
3023 struct btrfs_root *root,
3024 struct btrfs_path *path,
3025 u64 objectid, u64 offset, u64 bytenr)
3026{
3027 struct btrfs_delayed_ref_head *head;
3028 struct btrfs_delayed_ref_node *ref;
3029 struct btrfs_delayed_data_ref *data_ref;
3030 struct btrfs_delayed_ref_root *delayed_refs;
3031 int ret = 0;
3032
3033 delayed_refs = &trans->transaction->delayed_refs;
3034 spin_lock(&delayed_refs->lock);
3035 head = btrfs_find_delayed_ref_head(trans, bytenr);
3036 if (!head) {
3037 spin_unlock(&delayed_refs->lock);
3038 return 0;
3039 }
3040
3041 if (!mutex_trylock(&head->mutex)) {
3042 atomic_inc(&head->node.refs);
3043 spin_unlock(&delayed_refs->lock);
3044
3045 btrfs_release_path(path);
3046
3047 /*
3048 * Mutex was contended, block until it's released and let
3049 * caller try again
3050 */
3051 mutex_lock(&head->mutex);
3052 mutex_unlock(&head->mutex);
3053 btrfs_put_delayed_ref(&head->node);
3054 return -EAGAIN;
3055 }
3056 spin_unlock(&delayed_refs->lock);
3057
3058 spin_lock(&head->lock);
3059 list_for_each_entry(ref, &head->ref_list, list) {
3060 /* If it's a shared ref we know a cross reference exists */
3061 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3062 ret = 1;
3063 break;
3064 }
3065
3066 data_ref = btrfs_delayed_node_to_data_ref(ref);
3067
3068 /*
3069 * If our ref doesn't match the one we're currently looking at
3070 * then we have a cross reference.
3071 */
3072 if (data_ref->root != root->root_key.objectid ||
3073 data_ref->objectid != objectid ||
3074 data_ref->offset != offset) {
3075 ret = 1;
3076 break;
3077 }
3078 }
3079 spin_unlock(&head->lock);
3080 mutex_unlock(&head->mutex);
3081 return ret;
3082}
3083
3084static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
3085 struct btrfs_root *root,
3086 struct btrfs_path *path,
3087 u64 objectid, u64 offset, u64 bytenr)
3088{
3089 struct btrfs_root *extent_root = root->fs_info->extent_root;
3090 struct extent_buffer *leaf;
3091 struct btrfs_extent_data_ref *ref;
3092 struct btrfs_extent_inline_ref *iref;
3093 struct btrfs_extent_item *ei;
3094 struct btrfs_key key;
3095 u32 item_size;
3096 int ret;
3097
3098 key.objectid = bytenr;
3099 key.offset = (u64)-1;
3100 key.type = BTRFS_EXTENT_ITEM_KEY;
3101
3102 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3103 if (ret < 0)
3104 goto out;
3105 BUG_ON(ret == 0); /* Corruption */
3106
3107 ret = -ENOENT;
3108 if (path->slots[0] == 0)
3109 goto out;
3110
3111 path->slots[0]--;
3112 leaf = path->nodes[0];
3113 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3114
3115 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3116 goto out;
3117
3118 ret = 1;
3119 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3120#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3121 if (item_size < sizeof(*ei)) {
3122 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3123 goto out;
3124 }
3125#endif
3126 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3127
3128 if (item_size != sizeof(*ei) +
3129 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3130 goto out;
3131
3132 if (btrfs_extent_generation(leaf, ei) <=
3133 btrfs_root_last_snapshot(&root->root_item))
3134 goto out;
3135
3136 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3137 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3138 BTRFS_EXTENT_DATA_REF_KEY)
3139 goto out;
3140
3141 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3142 if (btrfs_extent_refs(leaf, ei) !=
3143 btrfs_extent_data_ref_count(leaf, ref) ||
3144 btrfs_extent_data_ref_root(leaf, ref) !=
3145 root->root_key.objectid ||
3146 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3147 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3148 goto out;
3149
3150 ret = 0;
3151out:
3152 return ret;
3153}
3154
3155int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3156 struct btrfs_root *root,
3157 u64 objectid, u64 offset, u64 bytenr)
3158{
3159 struct btrfs_path *path;
3160 int ret;
3161 int ret2;
3162
3163 path = btrfs_alloc_path();
3164 if (!path)
3165 return -ENOENT;
3166
3167 do {
3168 ret = check_committed_ref(trans, root, path, objectid,
3169 offset, bytenr);
3170 if (ret && ret != -ENOENT)
3171 goto out;
3172
3173 ret2 = check_delayed_ref(trans, root, path, objectid,
3174 offset, bytenr);
3175 } while (ret2 == -EAGAIN);
3176
3177 if (ret2 && ret2 != -ENOENT) {
3178 ret = ret2;
3179 goto out;
3180 }
3181
3182 if (ret != -ENOENT || ret2 != -ENOENT)
3183 ret = 0;
3184out:
3185 btrfs_free_path(path);
3186 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3187 WARN_ON(ret > 0);
3188 return ret;
3189}
3190
3191static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3192 struct btrfs_root *root,
3193 struct extent_buffer *buf,
3194 int full_backref, int inc)
3195{
3196 u64 bytenr;
3197 u64 num_bytes;
3198 u64 parent;
3199 u64 ref_root;
3200 u32 nritems;
3201 struct btrfs_key key;
3202 struct btrfs_file_extent_item *fi;
3203 int i;
3204 int level;
3205 int ret = 0;
3206 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3207 u64, u64, u64, u64, u64, u64);
3208
3209
3210 if (btrfs_test_is_dummy_root(root))
3211 return 0;
3212
3213 ref_root = btrfs_header_owner(buf);
3214 nritems = btrfs_header_nritems(buf);
3215 level = btrfs_header_level(buf);
3216
3217 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3218 return 0;
3219
3220 if (inc)
3221 process_func = btrfs_inc_extent_ref;
3222 else
3223 process_func = btrfs_free_extent;
3224
3225 if (full_backref)
3226 parent = buf->start;
3227 else
3228 parent = 0;
3229
3230 for (i = 0; i < nritems; i++) {
3231 if (level == 0) {
3232 btrfs_item_key_to_cpu(buf, &key, i);
3233 if (key.type != BTRFS_EXTENT_DATA_KEY)
3234 continue;
3235 fi = btrfs_item_ptr(buf, i,
3236 struct btrfs_file_extent_item);
3237 if (btrfs_file_extent_type(buf, fi) ==
3238 BTRFS_FILE_EXTENT_INLINE)
3239 continue;
3240 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3241 if (bytenr == 0)
3242 continue;
3243
3244 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3245 key.offset -= btrfs_file_extent_offset(buf, fi);
3246 ret = process_func(trans, root, bytenr, num_bytes,
3247 parent, ref_root, key.objectid,
3248 key.offset);
3249 if (ret)
3250 goto fail;
3251 } else {
3252 bytenr = btrfs_node_blockptr(buf, i);
3253 num_bytes = root->nodesize;
3254 ret = process_func(trans, root, bytenr, num_bytes,
3255 parent, ref_root, level - 1, 0);
3256 if (ret)
3257 goto fail;
3258 }
3259 }
3260 return 0;
3261fail:
3262 return ret;
3263}
3264
3265int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3266 struct extent_buffer *buf, int full_backref)
3267{
3268 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3269}
3270
3271int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3272 struct extent_buffer *buf, int full_backref)
3273{
3274 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3275}
3276
3277static int write_one_cache_group(struct btrfs_trans_handle *trans,
3278 struct btrfs_root *root,
3279 struct btrfs_path *path,
3280 struct btrfs_block_group_cache *cache)
3281{
3282 int ret;
3283 struct btrfs_root *extent_root = root->fs_info->extent_root;
3284 unsigned long bi;
3285 struct extent_buffer *leaf;
3286
3287 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3288 if (ret) {
3289 if (ret > 0)
3290 ret = -ENOENT;
3291 goto fail;
3292 }
3293
3294 leaf = path->nodes[0];
3295 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3296 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3297 btrfs_mark_buffer_dirty(leaf);
3298fail:
3299 btrfs_release_path(path);
3300 return ret;
3301
3302}
3303
3304static struct btrfs_block_group_cache *
3305next_block_group(struct btrfs_root *root,
3306 struct btrfs_block_group_cache *cache)
3307{
3308 struct rb_node *node;
3309
3310 spin_lock(&root->fs_info->block_group_cache_lock);
3311
3312 /* If our block group was removed, we need a full search. */
3313 if (RB_EMPTY_NODE(&cache->cache_node)) {
3314 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3315
3316 spin_unlock(&root->fs_info->block_group_cache_lock);
3317 btrfs_put_block_group(cache);
3318 cache = btrfs_lookup_first_block_group(root->fs_info,
3319 next_bytenr);
3320 return cache;
3321 }
3322 node = rb_next(&cache->cache_node);
3323 btrfs_put_block_group(cache);
3324 if (node) {
3325 cache = rb_entry(node, struct btrfs_block_group_cache,
3326 cache_node);
3327 btrfs_get_block_group(cache);
3328 } else
3329 cache = NULL;
3330 spin_unlock(&root->fs_info->block_group_cache_lock);
3331 return cache;
3332}
3333
3334static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3335 struct btrfs_trans_handle *trans,
3336 struct btrfs_path *path)
3337{
3338 struct btrfs_root *root = block_group->fs_info->tree_root;
3339 struct inode *inode = NULL;
3340 u64 alloc_hint = 0;
3341 int dcs = BTRFS_DC_ERROR;
3342 u64 num_pages = 0;
3343 int retries = 0;
3344 int ret = 0;
3345
3346 /*
3347 * If this block group is smaller than 100 megs don't bother caching the
3348 * block group.
3349 */
3350 if (block_group->key.offset < (100 * SZ_1M)) {
3351 spin_lock(&block_group->lock);
3352 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3353 spin_unlock(&block_group->lock);
3354 return 0;
3355 }
3356
3357 if (trans->aborted)
3358 return 0;
3359again:
3360 inode = lookup_free_space_inode(root, block_group, path);
3361 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3362 ret = PTR_ERR(inode);
3363 btrfs_release_path(path);
3364 goto out;
3365 }
3366
3367 if (IS_ERR(inode)) {
3368 BUG_ON(retries);
3369 retries++;
3370
3371 if (block_group->ro)
3372 goto out_free;
3373
3374 ret = create_free_space_inode(root, trans, block_group, path);
3375 if (ret)
3376 goto out_free;
3377 goto again;
3378 }
3379
3380 /* We've already setup this transaction, go ahead and exit */
3381 if (block_group->cache_generation == trans->transid &&
3382 i_size_read(inode)) {
3383 dcs = BTRFS_DC_SETUP;
3384 goto out_put;
3385 }
3386
3387 /*
3388 * We want to set the generation to 0, that way if anything goes wrong
3389 * from here on out we know not to trust this cache when we load up next
3390 * time.
3391 */
3392 BTRFS_I(inode)->generation = 0;
3393 ret = btrfs_update_inode(trans, root, inode);
3394 if (ret) {
3395 /*
3396 * So theoretically we could recover from this, simply set the
3397 * super cache generation to 0 so we know to invalidate the
3398 * cache, but then we'd have to keep track of the block groups
3399 * that fail this way so we know we _have_ to reset this cache
3400 * before the next commit or risk reading stale cache. So to
3401 * limit our exposure to horrible edge cases lets just abort the
3402 * transaction, this only happens in really bad situations
3403 * anyway.
3404 */
3405 btrfs_abort_transaction(trans, root, ret);
3406 goto out_put;
3407 }
3408 WARN_ON(ret);
3409
3410 if (i_size_read(inode) > 0) {
3411 ret = btrfs_check_trunc_cache_free_space(root,
3412 &root->fs_info->global_block_rsv);
3413 if (ret)
3414 goto out_put;
3415
3416 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3417 if (ret)
3418 goto out_put;
3419 }
3420
3421 spin_lock(&block_group->lock);
3422 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3423 !btrfs_test_opt(root, SPACE_CACHE)) {
3424 /*
3425 * don't bother trying to write stuff out _if_
3426 * a) we're not cached,
3427 * b) we're with nospace_cache mount option.
3428 */
3429 dcs = BTRFS_DC_WRITTEN;
3430 spin_unlock(&block_group->lock);
3431 goto out_put;
3432 }
3433 spin_unlock(&block_group->lock);
3434
3435 /*
3436 * We hit an ENOSPC when setting up the cache in this transaction, just
3437 * skip doing the setup, we've already cleared the cache so we're safe.
3438 */
3439 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3440 ret = -ENOSPC;
3441 goto out_put;
3442 }
3443
3444 /*
3445 * Try to preallocate enough space based on how big the block group is.
3446 * Keep in mind this has to include any pinned space which could end up
3447 * taking up quite a bit since it's not folded into the other space
3448 * cache.
3449 */
3450 num_pages = div_u64(block_group->key.offset, SZ_256M);
3451 if (!num_pages)
3452 num_pages = 1;
3453
3454 num_pages *= 16;
3455 num_pages *= PAGE_SIZE;
3456
3457 ret = btrfs_check_data_free_space(inode, 0, num_pages);
3458 if (ret)
3459 goto out_put;
3460
3461 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3462 num_pages, num_pages,
3463 &alloc_hint);
3464 /*
3465 * Our cache requires contiguous chunks so that we don't modify a bunch
3466 * of metadata or split extents when writing the cache out, which means
3467 * we can enospc if we are heavily fragmented in addition to just normal
3468 * out of space conditions. So if we hit this just skip setting up any
3469 * other block groups for this transaction, maybe we'll unpin enough
3470 * space the next time around.
3471 */
3472 if (!ret)
3473 dcs = BTRFS_DC_SETUP;
3474 else if (ret == -ENOSPC)
3475 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3476 btrfs_free_reserved_data_space(inode, 0, num_pages);
3477
3478out_put:
3479 iput(inode);
3480out_free:
3481 btrfs_release_path(path);
3482out:
3483 spin_lock(&block_group->lock);
3484 if (!ret && dcs == BTRFS_DC_SETUP)
3485 block_group->cache_generation = trans->transid;
3486 block_group->disk_cache_state = dcs;
3487 spin_unlock(&block_group->lock);
3488
3489 return ret;
3490}
3491
3492int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3493 struct btrfs_root *root)
3494{
3495 struct btrfs_block_group_cache *cache, *tmp;
3496 struct btrfs_transaction *cur_trans = trans->transaction;
3497 struct btrfs_path *path;
3498
3499 if (list_empty(&cur_trans->dirty_bgs) ||
3500 !btrfs_test_opt(root, SPACE_CACHE))
3501 return 0;
3502
3503 path = btrfs_alloc_path();
3504 if (!path)
3505 return -ENOMEM;
3506
3507 /* Could add new block groups, use _safe just in case */
3508 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3509 dirty_list) {
3510 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3511 cache_save_setup(cache, trans, path);
3512 }
3513
3514 btrfs_free_path(path);
3515 return 0;
3516}
3517
3518/*
3519 * transaction commit does final block group cache writeback during a
3520 * critical section where nothing is allowed to change the FS. This is
3521 * required in order for the cache to actually match the block group,
3522 * but can introduce a lot of latency into the commit.
3523 *
3524 * So, btrfs_start_dirty_block_groups is here to kick off block group
3525 * cache IO. There's a chance we'll have to redo some of it if the
3526 * block group changes again during the commit, but it greatly reduces
3527 * the commit latency by getting rid of the easy block groups while
3528 * we're still allowing others to join the commit.
3529 */
3530int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3531 struct btrfs_root *root)
3532{
3533 struct btrfs_block_group_cache *cache;
3534 struct btrfs_transaction *cur_trans = trans->transaction;
3535 int ret = 0;
3536 int should_put;
3537 struct btrfs_path *path = NULL;
3538 LIST_HEAD(dirty);
3539 struct list_head *io = &cur_trans->io_bgs;
3540 int num_started = 0;
3541 int loops = 0;
3542
3543 spin_lock(&cur_trans->dirty_bgs_lock);
3544 if (list_empty(&cur_trans->dirty_bgs)) {
3545 spin_unlock(&cur_trans->dirty_bgs_lock);
3546 return 0;
3547 }
3548 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3549 spin_unlock(&cur_trans->dirty_bgs_lock);
3550
3551again:
3552 /*
3553 * make sure all the block groups on our dirty list actually
3554 * exist
3555 */
3556 btrfs_create_pending_block_groups(trans, root);
3557
3558 if (!path) {
3559 path = btrfs_alloc_path();
3560 if (!path)
3561 return -ENOMEM;
3562 }
3563
3564 /*
3565 * cache_write_mutex is here only to save us from balance or automatic
3566 * removal of empty block groups deleting this block group while we are
3567 * writing out the cache
3568 */
3569 mutex_lock(&trans->transaction->cache_write_mutex);
3570 while (!list_empty(&dirty)) {
3571 cache = list_first_entry(&dirty,
3572 struct btrfs_block_group_cache,
3573 dirty_list);
3574 /*
3575 * this can happen if something re-dirties a block
3576 * group that is already under IO. Just wait for it to
3577 * finish and then do it all again
3578 */
3579 if (!list_empty(&cache->io_list)) {
3580 list_del_init(&cache->io_list);
3581 btrfs_wait_cache_io(root, trans, cache,
3582 &cache->io_ctl, path,
3583 cache->key.objectid);
3584 btrfs_put_block_group(cache);
3585 }
3586
3587
3588 /*
3589 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3590 * if it should update the cache_state. Don't delete
3591 * until after we wait.
3592 *
3593 * Since we're not running in the commit critical section
3594 * we need the dirty_bgs_lock to protect from update_block_group
3595 */
3596 spin_lock(&cur_trans->dirty_bgs_lock);
3597 list_del_init(&cache->dirty_list);
3598 spin_unlock(&cur_trans->dirty_bgs_lock);
3599
3600 should_put = 1;
3601
3602 cache_save_setup(cache, trans, path);
3603
3604 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3605 cache->io_ctl.inode = NULL;
3606 ret = btrfs_write_out_cache(root, trans, cache, path);
3607 if (ret == 0 && cache->io_ctl.inode) {
3608 num_started++;
3609 should_put = 0;
3610
3611 /*
3612 * the cache_write_mutex is protecting
3613 * the io_list
3614 */
3615 list_add_tail(&cache->io_list, io);
3616 } else {
3617 /*
3618 * if we failed to write the cache, the
3619 * generation will be bad and life goes on
3620 */
3621 ret = 0;
3622 }
3623 }
3624 if (!ret) {
3625 ret = write_one_cache_group(trans, root, path, cache);
3626 /*
3627 * Our block group might still be attached to the list
3628 * of new block groups in the transaction handle of some
3629 * other task (struct btrfs_trans_handle->new_bgs). This
3630 * means its block group item isn't yet in the extent
3631 * tree. If this happens ignore the error, as we will
3632 * try again later in the critical section of the
3633 * transaction commit.
3634 */
3635 if (ret == -ENOENT) {
3636 ret = 0;
3637 spin_lock(&cur_trans->dirty_bgs_lock);
3638 if (list_empty(&cache->dirty_list)) {
3639 list_add_tail(&cache->dirty_list,
3640 &cur_trans->dirty_bgs);
3641 btrfs_get_block_group(cache);
3642 }
3643 spin_unlock(&cur_trans->dirty_bgs_lock);
3644 } else if (ret) {
3645 btrfs_abort_transaction(trans, root, ret);
3646 }
3647 }
3648
3649 /* if its not on the io list, we need to put the block group */
3650 if (should_put)
3651 btrfs_put_block_group(cache);
3652
3653 if (ret)
3654 break;
3655
3656 /*
3657 * Avoid blocking other tasks for too long. It might even save
3658 * us from writing caches for block groups that are going to be
3659 * removed.
3660 */
3661 mutex_unlock(&trans->transaction->cache_write_mutex);
3662 mutex_lock(&trans->transaction->cache_write_mutex);
3663 }
3664 mutex_unlock(&trans->transaction->cache_write_mutex);
3665
3666 /*
3667 * go through delayed refs for all the stuff we've just kicked off
3668 * and then loop back (just once)
3669 */
3670 ret = btrfs_run_delayed_refs(trans, root, 0);
3671 if (!ret && loops == 0) {
3672 loops++;
3673 spin_lock(&cur_trans->dirty_bgs_lock);
3674 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3675 /*
3676 * dirty_bgs_lock protects us from concurrent block group
3677 * deletes too (not just cache_write_mutex).
3678 */
3679 if (!list_empty(&dirty)) {
3680 spin_unlock(&cur_trans->dirty_bgs_lock);
3681 goto again;
3682 }
3683 spin_unlock(&cur_trans->dirty_bgs_lock);
3684 }
3685
3686 btrfs_free_path(path);
3687 return ret;
3688}
3689
3690int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3691 struct btrfs_root *root)
3692{
3693 struct btrfs_block_group_cache *cache;
3694 struct btrfs_transaction *cur_trans = trans->transaction;
3695 int ret = 0;
3696 int should_put;
3697 struct btrfs_path *path;
3698 struct list_head *io = &cur_trans->io_bgs;
3699 int num_started = 0;
3700
3701 path = btrfs_alloc_path();
3702 if (!path)
3703 return -ENOMEM;
3704
3705 /*
3706 * Even though we are in the critical section of the transaction commit,
3707 * we can still have concurrent tasks adding elements to this
3708 * transaction's list of dirty block groups. These tasks correspond to
3709 * endio free space workers started when writeback finishes for a
3710 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3711 * allocate new block groups as a result of COWing nodes of the root
3712 * tree when updating the free space inode. The writeback for the space
3713 * caches is triggered by an earlier call to
3714 * btrfs_start_dirty_block_groups() and iterations of the following
3715 * loop.
3716 * Also we want to do the cache_save_setup first and then run the
3717 * delayed refs to make sure we have the best chance at doing this all
3718 * in one shot.
3719 */
3720 spin_lock(&cur_trans->dirty_bgs_lock);
3721 while (!list_empty(&cur_trans->dirty_bgs)) {
3722 cache = list_first_entry(&cur_trans->dirty_bgs,
3723 struct btrfs_block_group_cache,
3724 dirty_list);
3725
3726 /*
3727 * this can happen if cache_save_setup re-dirties a block
3728 * group that is already under IO. Just wait for it to
3729 * finish and then do it all again
3730 */
3731 if (!list_empty(&cache->io_list)) {
3732 spin_unlock(&cur_trans->dirty_bgs_lock);
3733 list_del_init(&cache->io_list);
3734 btrfs_wait_cache_io(root, trans, cache,
3735 &cache->io_ctl, path,
3736 cache->key.objectid);
3737 btrfs_put_block_group(cache);
3738 spin_lock(&cur_trans->dirty_bgs_lock);
3739 }
3740
3741 /*
3742 * don't remove from the dirty list until after we've waited
3743 * on any pending IO
3744 */
3745 list_del_init(&cache->dirty_list);
3746 spin_unlock(&cur_trans->dirty_bgs_lock);
3747 should_put = 1;
3748
3749 cache_save_setup(cache, trans, path);
3750
3751 if (!ret)
3752 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3753
3754 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3755 cache->io_ctl.inode = NULL;
3756 ret = btrfs_write_out_cache(root, trans, cache, path);
3757 if (ret == 0 && cache->io_ctl.inode) {
3758 num_started++;
3759 should_put = 0;
3760 list_add_tail(&cache->io_list, io);
3761 } else {
3762 /*
3763 * if we failed to write the cache, the
3764 * generation will be bad and life goes on
3765 */
3766 ret = 0;
3767 }
3768 }
3769 if (!ret) {
3770 ret = write_one_cache_group(trans, root, path, cache);
3771 /*
3772 * One of the free space endio workers might have
3773 * created a new block group while updating a free space
3774 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3775 * and hasn't released its transaction handle yet, in
3776 * which case the new block group is still attached to
3777 * its transaction handle and its creation has not
3778 * finished yet (no block group item in the extent tree
3779 * yet, etc). If this is the case, wait for all free
3780 * space endio workers to finish and retry. This is a
3781 * a very rare case so no need for a more efficient and
3782 * complex approach.
3783 */
3784 if (ret == -ENOENT) {
3785 wait_event(cur_trans->writer_wait,
3786 atomic_read(&cur_trans->num_writers) == 1);
3787 ret = write_one_cache_group(trans, root, path,
3788 cache);
3789 }
3790 if (ret)
3791 btrfs_abort_transaction(trans, root, ret);
3792 }
3793
3794 /* if its not on the io list, we need to put the block group */
3795 if (should_put)
3796 btrfs_put_block_group(cache);
3797 spin_lock(&cur_trans->dirty_bgs_lock);
3798 }
3799 spin_unlock(&cur_trans->dirty_bgs_lock);
3800
3801 while (!list_empty(io)) {
3802 cache = list_first_entry(io, struct btrfs_block_group_cache,
3803 io_list);
3804 list_del_init(&cache->io_list);
3805 btrfs_wait_cache_io(root, trans, cache,
3806 &cache->io_ctl, path, cache->key.objectid);
3807 btrfs_put_block_group(cache);
3808 }
3809
3810 btrfs_free_path(path);
3811 return ret;
3812}
3813
3814int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3815{
3816 struct btrfs_block_group_cache *block_group;
3817 int readonly = 0;
3818
3819 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3820 if (!block_group || block_group->ro)
3821 readonly = 1;
3822 if (block_group)
3823 btrfs_put_block_group(block_group);
3824 return readonly;
3825}
3826
3827static const char *alloc_name(u64 flags)
3828{
3829 switch (flags) {
3830 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3831 return "mixed";
3832 case BTRFS_BLOCK_GROUP_METADATA:
3833 return "metadata";
3834 case BTRFS_BLOCK_GROUP_DATA:
3835 return "data";
3836 case BTRFS_BLOCK_GROUP_SYSTEM:
3837 return "system";
3838 default:
3839 WARN_ON(1);
3840 return "invalid-combination";
3841 };
3842}
3843
3844static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3845 u64 total_bytes, u64 bytes_used,
3846 struct btrfs_space_info **space_info)
3847{
3848 struct btrfs_space_info *found;
3849 int i;
3850 int factor;
3851 int ret;
3852
3853 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3854 BTRFS_BLOCK_GROUP_RAID10))
3855 factor = 2;
3856 else
3857 factor = 1;
3858
3859 found = __find_space_info(info, flags);
3860 if (found) {
3861 spin_lock(&found->lock);
3862 found->total_bytes += total_bytes;
3863 found->disk_total += total_bytes * factor;
3864 found->bytes_used += bytes_used;
3865 found->disk_used += bytes_used * factor;
3866 if (total_bytes > 0)
3867 found->full = 0;
3868 spin_unlock(&found->lock);
3869 *space_info = found;
3870 return 0;
3871 }
3872 found = kzalloc(sizeof(*found), GFP_NOFS);
3873 if (!found)
3874 return -ENOMEM;
3875
3876 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3877 if (ret) {
3878 kfree(found);
3879 return ret;
3880 }
3881
3882 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3883 INIT_LIST_HEAD(&found->block_groups[i]);
3884 init_rwsem(&found->groups_sem);
3885 spin_lock_init(&found->lock);
3886 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3887 found->total_bytes = total_bytes;
3888 found->disk_total = total_bytes * factor;
3889 found->bytes_used = bytes_used;
3890 found->disk_used = bytes_used * factor;
3891 found->bytes_pinned = 0;
3892 found->bytes_reserved = 0;
3893 found->bytes_readonly = 0;
3894 found->bytes_may_use = 0;
3895 found->full = 0;
3896 found->max_extent_size = 0;
3897 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3898 found->chunk_alloc = 0;
3899 found->flush = 0;
3900 init_waitqueue_head(&found->wait);
3901 INIT_LIST_HEAD(&found->ro_bgs);
3902
3903 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3904 info->space_info_kobj, "%s",
3905 alloc_name(found->flags));
3906 if (ret) {
3907 kfree(found);
3908 return ret;
3909 }
3910
3911 *space_info = found;
3912 list_add_rcu(&found->list, &info->space_info);
3913 if (flags & BTRFS_BLOCK_GROUP_DATA)
3914 info->data_sinfo = found;
3915
3916 return ret;
3917}
3918
3919static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3920{
3921 u64 extra_flags = chunk_to_extended(flags) &
3922 BTRFS_EXTENDED_PROFILE_MASK;
3923
3924 write_seqlock(&fs_info->profiles_lock);
3925 if (flags & BTRFS_BLOCK_GROUP_DATA)
3926 fs_info->avail_data_alloc_bits |= extra_flags;
3927 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3928 fs_info->avail_metadata_alloc_bits |= extra_flags;
3929 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3930 fs_info->avail_system_alloc_bits |= extra_flags;
3931 write_sequnlock(&fs_info->profiles_lock);
3932}
3933
3934/*
3935 * returns target flags in extended format or 0 if restripe for this
3936 * chunk_type is not in progress
3937 *
3938 * should be called with either volume_mutex or balance_lock held
3939 */
3940static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3941{
3942 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3943 u64 target = 0;
3944
3945 if (!bctl)
3946 return 0;
3947
3948 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3949 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3950 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3951 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3952 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3953 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3954 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3955 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3956 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3957 }
3958
3959 return target;
3960}
3961
3962/*
3963 * @flags: available profiles in extended format (see ctree.h)
3964 *
3965 * Returns reduced profile in chunk format. If profile changing is in
3966 * progress (either running or paused) picks the target profile (if it's
3967 * already available), otherwise falls back to plain reducing.
3968 */
3969static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3970{
3971 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3972 u64 target;
3973 u64 raid_type;
3974 u64 allowed = 0;
3975
3976 /*
3977 * see if restripe for this chunk_type is in progress, if so
3978 * try to reduce to the target profile
3979 */
3980 spin_lock(&root->fs_info->balance_lock);
3981 target = get_restripe_target(root->fs_info, flags);
3982 if (target) {
3983 /* pick target profile only if it's already available */
3984 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3985 spin_unlock(&root->fs_info->balance_lock);
3986 return extended_to_chunk(target);
3987 }
3988 }
3989 spin_unlock(&root->fs_info->balance_lock);
3990
3991 /* First, mask out the RAID levels which aren't possible */
3992 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3993 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
3994 allowed |= btrfs_raid_group[raid_type];
3995 }
3996 allowed &= flags;
3997
3998 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
3999 allowed = BTRFS_BLOCK_GROUP_RAID6;
4000 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4001 allowed = BTRFS_BLOCK_GROUP_RAID5;
4002 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4003 allowed = BTRFS_BLOCK_GROUP_RAID10;
4004 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4005 allowed = BTRFS_BLOCK_GROUP_RAID1;
4006 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4007 allowed = BTRFS_BLOCK_GROUP_RAID0;
4008
4009 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4010
4011 return extended_to_chunk(flags | allowed);
4012}
4013
4014static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
4015{
4016 unsigned seq;
4017 u64 flags;
4018
4019 do {
4020 flags = orig_flags;
4021 seq = read_seqbegin(&root->fs_info->profiles_lock);
4022
4023 if (flags & BTRFS_BLOCK_GROUP_DATA)
4024 flags |= root->fs_info->avail_data_alloc_bits;
4025 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4026 flags |= root->fs_info->avail_system_alloc_bits;
4027 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4028 flags |= root->fs_info->avail_metadata_alloc_bits;
4029 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
4030
4031 return btrfs_reduce_alloc_profile(root, flags);
4032}
4033
4034u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
4035{
4036 u64 flags;
4037 u64 ret;
4038
4039 if (data)
4040 flags = BTRFS_BLOCK_GROUP_DATA;
4041 else if (root == root->fs_info->chunk_root)
4042 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4043 else
4044 flags = BTRFS_BLOCK_GROUP_METADATA;
4045
4046 ret = get_alloc_profile(root, flags);
4047 return ret;
4048}
4049
4050int btrfs_alloc_data_chunk_ondemand(struct inode *inode, u64 bytes)
4051{
4052 struct btrfs_space_info *data_sinfo;
4053 struct btrfs_root *root = BTRFS_I(inode)->root;
4054 struct btrfs_fs_info *fs_info = root->fs_info;
4055 u64 used;
4056 int ret = 0;
4057 int need_commit = 2;
4058 int have_pinned_space;
4059
4060 /* make sure bytes are sectorsize aligned */
4061 bytes = ALIGN(bytes, root->sectorsize);
4062
4063 if (btrfs_is_free_space_inode(inode)) {
4064 need_commit = 0;
4065 ASSERT(current->journal_info);
4066 }
4067
4068 data_sinfo = fs_info->data_sinfo;
4069 if (!data_sinfo)
4070 goto alloc;
4071
4072again:
4073 /* make sure we have enough space to handle the data first */
4074 spin_lock(&data_sinfo->lock);
4075 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
4076 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
4077 data_sinfo->bytes_may_use;
4078
4079 if (used + bytes > data_sinfo->total_bytes) {
4080 struct btrfs_trans_handle *trans;
4081
4082 /*
4083 * if we don't have enough free bytes in this space then we need
4084 * to alloc a new chunk.
4085 */
4086 if (!data_sinfo->full) {
4087 u64 alloc_target;
4088
4089 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4090 spin_unlock(&data_sinfo->lock);
4091alloc:
4092 alloc_target = btrfs_get_alloc_profile(root, 1);
4093 /*
4094 * It is ugly that we don't call nolock join
4095 * transaction for the free space inode case here.
4096 * But it is safe because we only do the data space
4097 * reservation for the free space cache in the
4098 * transaction context, the common join transaction
4099 * just increase the counter of the current transaction
4100 * handler, doesn't try to acquire the trans_lock of
4101 * the fs.
4102 */
4103 trans = btrfs_join_transaction(root);
4104 if (IS_ERR(trans))
4105 return PTR_ERR(trans);
4106
4107 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4108 alloc_target,
4109 CHUNK_ALLOC_NO_FORCE);
4110 btrfs_end_transaction(trans, root);
4111 if (ret < 0) {
4112 if (ret != -ENOSPC)
4113 return ret;
4114 else {
4115 have_pinned_space = 1;
4116 goto commit_trans;
4117 }
4118 }
4119
4120 if (!data_sinfo)
4121 data_sinfo = fs_info->data_sinfo;
4122
4123 goto again;
4124 }
4125
4126 /*
4127 * If we don't have enough pinned space to deal with this
4128 * allocation, and no removed chunk in current transaction,
4129 * don't bother committing the transaction.
4130 */
4131 have_pinned_space = percpu_counter_compare(
4132 &data_sinfo->total_bytes_pinned,
4133 used + bytes - data_sinfo->total_bytes);
4134 spin_unlock(&data_sinfo->lock);
4135
4136 /* commit the current transaction and try again */
4137commit_trans:
4138 if (need_commit &&
4139 !atomic_read(&root->fs_info->open_ioctl_trans)) {
4140 need_commit--;
4141
4142 if (need_commit > 0) {
4143 btrfs_start_delalloc_roots(fs_info, 0, -1);
4144 btrfs_wait_ordered_roots(fs_info, -1);
4145 }
4146
4147 trans = btrfs_join_transaction(root);
4148 if (IS_ERR(trans))
4149 return PTR_ERR(trans);
4150 if (have_pinned_space >= 0 ||
4151 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4152 &trans->transaction->flags) ||
4153 need_commit > 0) {
4154 ret = btrfs_commit_transaction(trans, root);
4155 if (ret)
4156 return ret;
4157 /*
4158 * The cleaner kthread might still be doing iput
4159 * operations. Wait for it to finish so that
4160 * more space is released.
4161 */
4162 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
4163 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
4164 goto again;
4165 } else {
4166 btrfs_end_transaction(trans, root);
4167 }
4168 }
4169
4170 trace_btrfs_space_reservation(root->fs_info,
4171 "space_info:enospc",
4172 data_sinfo->flags, bytes, 1);
4173 return -ENOSPC;
4174 }
4175 data_sinfo->bytes_may_use += bytes;
4176 trace_btrfs_space_reservation(root->fs_info, "space_info",
4177 data_sinfo->flags, bytes, 1);
4178 spin_unlock(&data_sinfo->lock);
4179
4180 return ret;
4181}
4182
4183/*
4184 * New check_data_free_space() with ability for precious data reservation
4185 * Will replace old btrfs_check_data_free_space(), but for patch split,
4186 * add a new function first and then replace it.
4187 */
4188int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len)
4189{
4190 struct btrfs_root *root = BTRFS_I(inode)->root;
4191 int ret;
4192
4193 /* align the range */
4194 len = round_up(start + len, root->sectorsize) -
4195 round_down(start, root->sectorsize);
4196 start = round_down(start, root->sectorsize);
4197
4198 ret = btrfs_alloc_data_chunk_ondemand(inode, len);
4199 if (ret < 0)
4200 return ret;
4201
4202 /*
4203 * Use new btrfs_qgroup_reserve_data to reserve precious data space
4204 *
4205 * TODO: Find a good method to avoid reserve data space for NOCOW
4206 * range, but don't impact performance on quota disable case.
4207 */
4208 ret = btrfs_qgroup_reserve_data(inode, start, len);
4209 return ret;
4210}
4211
4212/*
4213 * Called if we need to clear a data reservation for this inode
4214 * Normally in a error case.
4215 *
4216 * This one will *NOT* use accurate qgroup reserved space API, just for case
4217 * which we can't sleep and is sure it won't affect qgroup reserved space.
4218 * Like clear_bit_hook().
4219 */
4220void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4221 u64 len)
4222{
4223 struct btrfs_root *root = BTRFS_I(inode)->root;
4224 struct btrfs_space_info *data_sinfo;
4225
4226 /* Make sure the range is aligned to sectorsize */
4227 len = round_up(start + len, root->sectorsize) -
4228 round_down(start, root->sectorsize);
4229 start = round_down(start, root->sectorsize);
4230
4231 data_sinfo = root->fs_info->data_sinfo;
4232 spin_lock(&data_sinfo->lock);
4233 if (WARN_ON(data_sinfo->bytes_may_use < len))
4234 data_sinfo->bytes_may_use = 0;
4235 else
4236 data_sinfo->bytes_may_use -= len;
4237 trace_btrfs_space_reservation(root->fs_info, "space_info",
4238 data_sinfo->flags, len, 0);
4239 spin_unlock(&data_sinfo->lock);
4240}
4241
4242/*
4243 * Called if we need to clear a data reservation for this inode
4244 * Normally in a error case.
4245 *
4246 * This one will handle the per-indoe data rsv map for accurate reserved
4247 * space framework.
4248 */
4249void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len)
4250{
4251 btrfs_free_reserved_data_space_noquota(inode, start, len);
4252 btrfs_qgroup_free_data(inode, start, len);
4253}
4254
4255static void force_metadata_allocation(struct btrfs_fs_info *info)
4256{
4257 struct list_head *head = &info->space_info;
4258 struct btrfs_space_info *found;
4259
4260 rcu_read_lock();
4261 list_for_each_entry_rcu(found, head, list) {
4262 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4263 found->force_alloc = CHUNK_ALLOC_FORCE;
4264 }
4265 rcu_read_unlock();
4266}
4267
4268static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4269{
4270 return (global->size << 1);
4271}
4272
4273static int should_alloc_chunk(struct btrfs_root *root,
4274 struct btrfs_space_info *sinfo, int force)
4275{
4276 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4277 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4278 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4279 u64 thresh;
4280
4281 if (force == CHUNK_ALLOC_FORCE)
4282 return 1;
4283
4284 /*
4285 * We need to take into account the global rsv because for all intents
4286 * and purposes it's used space. Don't worry about locking the
4287 * global_rsv, it doesn't change except when the transaction commits.
4288 */
4289 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4290 num_allocated += calc_global_rsv_need_space(global_rsv);
4291
4292 /*
4293 * in limited mode, we want to have some free space up to
4294 * about 1% of the FS size.
4295 */
4296 if (force == CHUNK_ALLOC_LIMITED) {
4297 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4298 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4299
4300 if (num_bytes - num_allocated < thresh)
4301 return 1;
4302 }
4303
4304 if (num_allocated + SZ_2M < div_factor(num_bytes, 8))
4305 return 0;
4306 return 1;
4307}
4308
4309static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4310{
4311 u64 num_dev;
4312
4313 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4314 BTRFS_BLOCK_GROUP_RAID0 |
4315 BTRFS_BLOCK_GROUP_RAID5 |
4316 BTRFS_BLOCK_GROUP_RAID6))
4317 num_dev = root->fs_info->fs_devices->rw_devices;
4318 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4319 num_dev = 2;
4320 else
4321 num_dev = 1; /* DUP or single */
4322
4323 return num_dev;
4324}
4325
4326/*
4327 * If @is_allocation is true, reserve space in the system space info necessary
4328 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4329 * removing a chunk.
4330 */
4331void check_system_chunk(struct btrfs_trans_handle *trans,
4332 struct btrfs_root *root,
4333 u64 type)
4334{
4335 struct btrfs_space_info *info;
4336 u64 left;
4337 u64 thresh;
4338 int ret = 0;
4339 u64 num_devs;
4340
4341 /*
4342 * Needed because we can end up allocating a system chunk and for an
4343 * atomic and race free space reservation in the chunk block reserve.
4344 */
4345 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4346
4347 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4348 spin_lock(&info->lock);
4349 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4350 info->bytes_reserved - info->bytes_readonly -
4351 info->bytes_may_use;
4352 spin_unlock(&info->lock);
4353
4354 num_devs = get_profile_num_devs(root, type);
4355
4356 /* num_devs device items to update and 1 chunk item to add or remove */
4357 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4358 btrfs_calc_trans_metadata_size(root, 1);
4359
4360 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4361 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4362 left, thresh, type);
4363 dump_space_info(info, 0, 0);
4364 }
4365
4366 if (left < thresh) {
4367 u64 flags;
4368
4369 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4370 /*
4371 * Ignore failure to create system chunk. We might end up not
4372 * needing it, as we might not need to COW all nodes/leafs from
4373 * the paths we visit in the chunk tree (they were already COWed
4374 * or created in the current transaction for example).
4375 */
4376 ret = btrfs_alloc_chunk(trans, root, flags);
4377 }
4378
4379 if (!ret) {
4380 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4381 &root->fs_info->chunk_block_rsv,
4382 thresh, BTRFS_RESERVE_NO_FLUSH);
4383 if (!ret)
4384 trans->chunk_bytes_reserved += thresh;
4385 }
4386}
4387
4388static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4389 struct btrfs_root *extent_root, u64 flags, int force)
4390{
4391 struct btrfs_space_info *space_info;
4392 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4393 int wait_for_alloc = 0;
4394 int ret = 0;
4395
4396 /* Don't re-enter if we're already allocating a chunk */
4397 if (trans->allocating_chunk)
4398 return -ENOSPC;
4399
4400 space_info = __find_space_info(extent_root->fs_info, flags);
4401 if (!space_info) {
4402 ret = update_space_info(extent_root->fs_info, flags,
4403 0, 0, &space_info);
4404 BUG_ON(ret); /* -ENOMEM */
4405 }
4406 BUG_ON(!space_info); /* Logic error */
4407
4408again:
4409 spin_lock(&space_info->lock);
4410 if (force < space_info->force_alloc)
4411 force = space_info->force_alloc;
4412 if (space_info->full) {
4413 if (should_alloc_chunk(extent_root, space_info, force))
4414 ret = -ENOSPC;
4415 else
4416 ret = 0;
4417 spin_unlock(&space_info->lock);
4418 return ret;
4419 }
4420
4421 if (!should_alloc_chunk(extent_root, space_info, force)) {
4422 spin_unlock(&space_info->lock);
4423 return 0;
4424 } else if (space_info->chunk_alloc) {
4425 wait_for_alloc = 1;
4426 } else {
4427 space_info->chunk_alloc = 1;
4428 }
4429
4430 spin_unlock(&space_info->lock);
4431
4432 mutex_lock(&fs_info->chunk_mutex);
4433
4434 /*
4435 * The chunk_mutex is held throughout the entirety of a chunk
4436 * allocation, so once we've acquired the chunk_mutex we know that the
4437 * other guy is done and we need to recheck and see if we should
4438 * allocate.
4439 */
4440 if (wait_for_alloc) {
4441 mutex_unlock(&fs_info->chunk_mutex);
4442 wait_for_alloc = 0;
4443 goto again;
4444 }
4445
4446 trans->allocating_chunk = true;
4447
4448 /*
4449 * If we have mixed data/metadata chunks we want to make sure we keep
4450 * allocating mixed chunks instead of individual chunks.
4451 */
4452 if (btrfs_mixed_space_info(space_info))
4453 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4454
4455 /*
4456 * if we're doing a data chunk, go ahead and make sure that
4457 * we keep a reasonable number of metadata chunks allocated in the
4458 * FS as well.
4459 */
4460 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4461 fs_info->data_chunk_allocations++;
4462 if (!(fs_info->data_chunk_allocations %
4463 fs_info->metadata_ratio))
4464 force_metadata_allocation(fs_info);
4465 }
4466
4467 /*
4468 * Check if we have enough space in SYSTEM chunk because we may need
4469 * to update devices.
4470 */
4471 check_system_chunk(trans, extent_root, flags);
4472
4473 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4474 trans->allocating_chunk = false;
4475
4476 spin_lock(&space_info->lock);
4477 if (ret < 0 && ret != -ENOSPC)
4478 goto out;
4479 if (ret)
4480 space_info->full = 1;
4481 else
4482 ret = 1;
4483
4484 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4485out:
4486 space_info->chunk_alloc = 0;
4487 spin_unlock(&space_info->lock);
4488 mutex_unlock(&fs_info->chunk_mutex);
4489 /*
4490 * When we allocate a new chunk we reserve space in the chunk block
4491 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4492 * add new nodes/leafs to it if we end up needing to do it when
4493 * inserting the chunk item and updating device items as part of the
4494 * second phase of chunk allocation, performed by
4495 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4496 * large number of new block groups to create in our transaction
4497 * handle's new_bgs list to avoid exhausting the chunk block reserve
4498 * in extreme cases - like having a single transaction create many new
4499 * block groups when starting to write out the free space caches of all
4500 * the block groups that were made dirty during the lifetime of the
4501 * transaction.
4502 */
4503 if (trans->can_flush_pending_bgs &&
4504 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4505 btrfs_create_pending_block_groups(trans, trans->root);
4506 btrfs_trans_release_chunk_metadata(trans);
4507 }
4508 return ret;
4509}
4510
4511static int can_overcommit(struct btrfs_root *root,
4512 struct btrfs_space_info *space_info, u64 bytes,
4513 enum btrfs_reserve_flush_enum flush)
4514{
4515 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4516 u64 profile = btrfs_get_alloc_profile(root, 0);
4517 u64 space_size;
4518 u64 avail;
4519 u64 used;
4520
4521 used = space_info->bytes_used + space_info->bytes_reserved +
4522 space_info->bytes_pinned + space_info->bytes_readonly;
4523
4524 /*
4525 * We only want to allow over committing if we have lots of actual space
4526 * free, but if we don't have enough space to handle the global reserve
4527 * space then we could end up having a real enospc problem when trying
4528 * to allocate a chunk or some other such important allocation.
4529 */
4530 spin_lock(&global_rsv->lock);
4531 space_size = calc_global_rsv_need_space(global_rsv);
4532 spin_unlock(&global_rsv->lock);
4533 if (used + space_size >= space_info->total_bytes)
4534 return 0;
4535
4536 used += space_info->bytes_may_use;
4537
4538 spin_lock(&root->fs_info->free_chunk_lock);
4539 avail = root->fs_info->free_chunk_space;
4540 spin_unlock(&root->fs_info->free_chunk_lock);
4541
4542 /*
4543 * If we have dup, raid1 or raid10 then only half of the free
4544 * space is actually useable. For raid56, the space info used
4545 * doesn't include the parity drive, so we don't have to
4546 * change the math
4547 */
4548 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4549 BTRFS_BLOCK_GROUP_RAID1 |
4550 BTRFS_BLOCK_GROUP_RAID10))
4551 avail >>= 1;
4552
4553 /*
4554 * If we aren't flushing all things, let us overcommit up to
4555 * 1/2th of the space. If we can flush, don't let us overcommit
4556 * too much, let it overcommit up to 1/8 of the space.
4557 */
4558 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4559 avail >>= 3;
4560 else
4561 avail >>= 1;
4562
4563 if (used + bytes < space_info->total_bytes + avail)
4564 return 1;
4565 return 0;
4566}
4567
4568static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4569 unsigned long nr_pages, int nr_items)
4570{
4571 struct super_block *sb = root->fs_info->sb;
4572
4573 if (down_read_trylock(&sb->s_umount)) {
4574 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4575 up_read(&sb->s_umount);
4576 } else {
4577 /*
4578 * We needn't worry the filesystem going from r/w to r/o though
4579 * we don't acquire ->s_umount mutex, because the filesystem
4580 * should guarantee the delalloc inodes list be empty after
4581 * the filesystem is readonly(all dirty pages are written to
4582 * the disk).
4583 */
4584 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4585 if (!current->journal_info)
4586 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4587 }
4588}
4589
4590static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4591{
4592 u64 bytes;
4593 int nr;
4594
4595 bytes = btrfs_calc_trans_metadata_size(root, 1);
4596 nr = (int)div64_u64(to_reclaim, bytes);
4597 if (!nr)
4598 nr = 1;
4599 return nr;
4600}
4601
4602#define EXTENT_SIZE_PER_ITEM SZ_256K
4603
4604/*
4605 * shrink metadata reservation for delalloc
4606 */
4607static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4608 bool wait_ordered)
4609{
4610 struct btrfs_block_rsv *block_rsv;
4611 struct btrfs_space_info *space_info;
4612 struct btrfs_trans_handle *trans;
4613 u64 delalloc_bytes;
4614 u64 max_reclaim;
4615 long time_left;
4616 unsigned long nr_pages;
4617 int loops;
4618 int items;
4619 enum btrfs_reserve_flush_enum flush;
4620
4621 /* Calc the number of the pages we need flush for space reservation */
4622 items = calc_reclaim_items_nr(root, to_reclaim);
4623 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4624
4625 trans = (struct btrfs_trans_handle *)current->journal_info;
4626 block_rsv = &root->fs_info->delalloc_block_rsv;
4627 space_info = block_rsv->space_info;
4628
4629 delalloc_bytes = percpu_counter_sum_positive(
4630 &root->fs_info->delalloc_bytes);
4631 if (delalloc_bytes == 0) {
4632 if (trans)
4633 return;
4634 if (wait_ordered)
4635 btrfs_wait_ordered_roots(root->fs_info, items);
4636 return;
4637 }
4638
4639 loops = 0;
4640 while (delalloc_bytes && loops < 3) {
4641 max_reclaim = min(delalloc_bytes, to_reclaim);
4642 nr_pages = max_reclaim >> PAGE_SHIFT;
4643 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4644 /*
4645 * We need to wait for the async pages to actually start before
4646 * we do anything.
4647 */
4648 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4649 if (!max_reclaim)
4650 goto skip_async;
4651
4652 if (max_reclaim <= nr_pages)
4653 max_reclaim = 0;
4654 else
4655 max_reclaim -= nr_pages;
4656
4657 wait_event(root->fs_info->async_submit_wait,
4658 atomic_read(&root->fs_info->async_delalloc_pages) <=
4659 (int)max_reclaim);
4660skip_async:
4661 if (!trans)
4662 flush = BTRFS_RESERVE_FLUSH_ALL;
4663 else
4664 flush = BTRFS_RESERVE_NO_FLUSH;
4665 spin_lock(&space_info->lock);
4666 if (can_overcommit(root, space_info, orig, flush)) {
4667 spin_unlock(&space_info->lock);
4668 break;
4669 }
4670 spin_unlock(&space_info->lock);
4671
4672 loops++;
4673 if (wait_ordered && !trans) {
4674 btrfs_wait_ordered_roots(root->fs_info, items);
4675 } else {
4676 time_left = schedule_timeout_killable(1);
4677 if (time_left)
4678 break;
4679 }
4680 delalloc_bytes = percpu_counter_sum_positive(
4681 &root->fs_info->delalloc_bytes);
4682 }
4683}
4684
4685/**
4686 * maybe_commit_transaction - possibly commit the transaction if its ok to
4687 * @root - the root we're allocating for
4688 * @bytes - the number of bytes we want to reserve
4689 * @force - force the commit
4690 *
4691 * This will check to make sure that committing the transaction will actually
4692 * get us somewhere and then commit the transaction if it does. Otherwise it
4693 * will return -ENOSPC.
4694 */
4695static int may_commit_transaction(struct btrfs_root *root,
4696 struct btrfs_space_info *space_info,
4697 u64 bytes, int force)
4698{
4699 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4700 struct btrfs_trans_handle *trans;
4701
4702 trans = (struct btrfs_trans_handle *)current->journal_info;
4703 if (trans)
4704 return -EAGAIN;
4705
4706 if (force)
4707 goto commit;
4708
4709 /* See if there is enough pinned space to make this reservation */
4710 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4711 bytes) >= 0)
4712 goto commit;
4713
4714 /*
4715 * See if there is some space in the delayed insertion reservation for
4716 * this reservation.
4717 */
4718 if (space_info != delayed_rsv->space_info)
4719 return -ENOSPC;
4720
4721 spin_lock(&delayed_rsv->lock);
4722 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4723 bytes - delayed_rsv->size) >= 0) {
4724 spin_unlock(&delayed_rsv->lock);
4725 return -ENOSPC;
4726 }
4727 spin_unlock(&delayed_rsv->lock);
4728
4729commit:
4730 trans = btrfs_join_transaction(root);
4731 if (IS_ERR(trans))
4732 return -ENOSPC;
4733
4734 return btrfs_commit_transaction(trans, root);
4735}
4736
4737enum flush_state {
4738 FLUSH_DELAYED_ITEMS_NR = 1,
4739 FLUSH_DELAYED_ITEMS = 2,
4740 FLUSH_DELALLOC = 3,
4741 FLUSH_DELALLOC_WAIT = 4,
4742 ALLOC_CHUNK = 5,
4743 COMMIT_TRANS = 6,
4744};
4745
4746static int flush_space(struct btrfs_root *root,
4747 struct btrfs_space_info *space_info, u64 num_bytes,
4748 u64 orig_bytes, int state)
4749{
4750 struct btrfs_trans_handle *trans;
4751 int nr;
4752 int ret = 0;
4753
4754 switch (state) {
4755 case FLUSH_DELAYED_ITEMS_NR:
4756 case FLUSH_DELAYED_ITEMS:
4757 if (state == FLUSH_DELAYED_ITEMS_NR)
4758 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4759 else
4760 nr = -1;
4761
4762 trans = btrfs_join_transaction(root);
4763 if (IS_ERR(trans)) {
4764 ret = PTR_ERR(trans);
4765 break;
4766 }
4767 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4768 btrfs_end_transaction(trans, root);
4769 break;
4770 case FLUSH_DELALLOC:
4771 case FLUSH_DELALLOC_WAIT:
4772 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4773 state == FLUSH_DELALLOC_WAIT);
4774 break;
4775 case ALLOC_CHUNK:
4776 trans = btrfs_join_transaction(root);
4777 if (IS_ERR(trans)) {
4778 ret = PTR_ERR(trans);
4779 break;
4780 }
4781 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4782 btrfs_get_alloc_profile(root, 0),
4783 CHUNK_ALLOC_NO_FORCE);
4784 btrfs_end_transaction(trans, root);
4785 if (ret == -ENOSPC)
4786 ret = 0;
4787 break;
4788 case COMMIT_TRANS:
4789 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4790 break;
4791 default:
4792 ret = -ENOSPC;
4793 break;
4794 }
4795
4796 return ret;
4797}
4798
4799static inline u64
4800btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4801 struct btrfs_space_info *space_info)
4802{
4803 u64 used;
4804 u64 expected;
4805 u64 to_reclaim;
4806
4807 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4808 spin_lock(&space_info->lock);
4809 if (can_overcommit(root, space_info, to_reclaim,
4810 BTRFS_RESERVE_FLUSH_ALL)) {
4811 to_reclaim = 0;
4812 goto out;
4813 }
4814
4815 used = space_info->bytes_used + space_info->bytes_reserved +
4816 space_info->bytes_pinned + space_info->bytes_readonly +
4817 space_info->bytes_may_use;
4818 if (can_overcommit(root, space_info, SZ_1M, BTRFS_RESERVE_FLUSH_ALL))
4819 expected = div_factor_fine(space_info->total_bytes, 95);
4820 else
4821 expected = div_factor_fine(space_info->total_bytes, 90);
4822
4823 if (used > expected)
4824 to_reclaim = used - expected;
4825 else
4826 to_reclaim = 0;
4827 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4828 space_info->bytes_reserved);
4829out:
4830 spin_unlock(&space_info->lock);
4831
4832 return to_reclaim;
4833}
4834
4835static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4836 struct btrfs_fs_info *fs_info, u64 used)
4837{
4838 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4839
4840 /* If we're just plain full then async reclaim just slows us down. */
4841 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
4842 return 0;
4843
4844 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4845 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4846}
4847
4848static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4849 struct btrfs_fs_info *fs_info,
4850 int flush_state)
4851{
4852 u64 used;
4853
4854 spin_lock(&space_info->lock);
4855 /*
4856 * We run out of space and have not got any free space via flush_space,
4857 * so don't bother doing async reclaim.
4858 */
4859 if (flush_state > COMMIT_TRANS && space_info->full) {
4860 spin_unlock(&space_info->lock);
4861 return 0;
4862 }
4863
4864 used = space_info->bytes_used + space_info->bytes_reserved +
4865 space_info->bytes_pinned + space_info->bytes_readonly +
4866 space_info->bytes_may_use;
4867 if (need_do_async_reclaim(space_info, fs_info, used)) {
4868 spin_unlock(&space_info->lock);
4869 return 1;
4870 }
4871 spin_unlock(&space_info->lock);
4872
4873 return 0;
4874}
4875
4876static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4877{
4878 struct btrfs_fs_info *fs_info;
4879 struct btrfs_space_info *space_info;
4880 u64 to_reclaim;
4881 int flush_state;
4882
4883 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4884 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4885
4886 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4887 space_info);
4888 if (!to_reclaim)
4889 return;
4890
4891 flush_state = FLUSH_DELAYED_ITEMS_NR;
4892 do {
4893 flush_space(fs_info->fs_root, space_info, to_reclaim,
4894 to_reclaim, flush_state);
4895 flush_state++;
4896 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4897 flush_state))
4898 return;
4899 } while (flush_state < COMMIT_TRANS);
4900}
4901
4902void btrfs_init_async_reclaim_work(struct work_struct *work)
4903{
4904 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4905}
4906
4907/**
4908 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4909 * @root - the root we're allocating for
4910 * @block_rsv - the block_rsv we're allocating for
4911 * @orig_bytes - the number of bytes we want
4912 * @flush - whether or not we can flush to make our reservation
4913 *
4914 * This will reserve orgi_bytes number of bytes from the space info associated
4915 * with the block_rsv. If there is not enough space it will make an attempt to
4916 * flush out space to make room. It will do this by flushing delalloc if
4917 * possible or committing the transaction. If flush is 0 then no attempts to
4918 * regain reservations will be made and this will fail if there is not enough
4919 * space already.
4920 */
4921static int reserve_metadata_bytes(struct btrfs_root *root,
4922 struct btrfs_block_rsv *block_rsv,
4923 u64 orig_bytes,
4924 enum btrfs_reserve_flush_enum flush)
4925{
4926 struct btrfs_space_info *space_info = block_rsv->space_info;
4927 u64 used;
4928 u64 num_bytes = orig_bytes;
4929 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4930 int ret = 0;
4931 bool flushing = false;
4932
4933again:
4934 ret = 0;
4935 spin_lock(&space_info->lock);
4936 /*
4937 * We only want to wait if somebody other than us is flushing and we
4938 * are actually allowed to flush all things.
4939 */
4940 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4941 space_info->flush) {
4942 spin_unlock(&space_info->lock);
4943 /*
4944 * If we have a trans handle we can't wait because the flusher
4945 * may have to commit the transaction, which would mean we would
4946 * deadlock since we are waiting for the flusher to finish, but
4947 * hold the current transaction open.
4948 */
4949 if (current->journal_info)
4950 return -EAGAIN;
4951 ret = wait_event_killable(space_info->wait, !space_info->flush);
4952 /* Must have been killed, return */
4953 if (ret)
4954 return -EINTR;
4955
4956 spin_lock(&space_info->lock);
4957 }
4958
4959 ret = -ENOSPC;
4960 used = space_info->bytes_used + space_info->bytes_reserved +
4961 space_info->bytes_pinned + space_info->bytes_readonly +
4962 space_info->bytes_may_use;
4963
4964 /*
4965 * The idea here is that we've not already over-reserved the block group
4966 * then we can go ahead and save our reservation first and then start
4967 * flushing if we need to. Otherwise if we've already overcommitted
4968 * lets start flushing stuff first and then come back and try to make
4969 * our reservation.
4970 */
4971 if (used <= space_info->total_bytes) {
4972 if (used + orig_bytes <= space_info->total_bytes) {
4973 space_info->bytes_may_use += orig_bytes;
4974 trace_btrfs_space_reservation(root->fs_info,
4975 "space_info", space_info->flags, orig_bytes, 1);
4976 ret = 0;
4977 } else {
4978 /*
4979 * Ok set num_bytes to orig_bytes since we aren't
4980 * overocmmitted, this way we only try and reclaim what
4981 * we need.
4982 */
4983 num_bytes = orig_bytes;
4984 }
4985 } else {
4986 /*
4987 * Ok we're over committed, set num_bytes to the overcommitted
4988 * amount plus the amount of bytes that we need for this
4989 * reservation.
4990 */
4991 num_bytes = used - space_info->total_bytes +
4992 (orig_bytes * 2);
4993 }
4994
4995 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4996 space_info->bytes_may_use += orig_bytes;
4997 trace_btrfs_space_reservation(root->fs_info, "space_info",
4998 space_info->flags, orig_bytes,
4999 1);
5000 ret = 0;
5001 }
5002
5003 /*
5004 * Couldn't make our reservation, save our place so while we're trying
5005 * to reclaim space we can actually use it instead of somebody else
5006 * stealing it from us.
5007 *
5008 * We make the other tasks wait for the flush only when we can flush
5009 * all things.
5010 */
5011 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5012 flushing = true;
5013 space_info->flush = 1;
5014 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5015 used += orig_bytes;
5016 /*
5017 * We will do the space reservation dance during log replay,
5018 * which means we won't have fs_info->fs_root set, so don't do
5019 * the async reclaim as we will panic.
5020 */
5021 if (!root->fs_info->log_root_recovering &&
5022 need_do_async_reclaim(space_info, root->fs_info, used) &&
5023 !work_busy(&root->fs_info->async_reclaim_work))
5024 queue_work(system_unbound_wq,
5025 &root->fs_info->async_reclaim_work);
5026 }
5027 spin_unlock(&space_info->lock);
5028
5029 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5030 goto out;
5031
5032 ret = flush_space(root, space_info, num_bytes, orig_bytes,
5033 flush_state);
5034 flush_state++;
5035
5036 /*
5037 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
5038 * would happen. So skip delalloc flush.
5039 */
5040 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
5041 (flush_state == FLUSH_DELALLOC ||
5042 flush_state == FLUSH_DELALLOC_WAIT))
5043 flush_state = ALLOC_CHUNK;
5044
5045 if (!ret)
5046 goto again;
5047 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
5048 flush_state < COMMIT_TRANS)
5049 goto again;
5050 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
5051 flush_state <= COMMIT_TRANS)
5052 goto again;
5053
5054out:
5055 if (ret == -ENOSPC &&
5056 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5057 struct btrfs_block_rsv *global_rsv =
5058 &root->fs_info->global_block_rsv;
5059
5060 if (block_rsv != global_rsv &&
5061 !block_rsv_use_bytes(global_rsv, orig_bytes))
5062 ret = 0;
5063 }
5064 if (ret == -ENOSPC)
5065 trace_btrfs_space_reservation(root->fs_info,
5066 "space_info:enospc",
5067 space_info->flags, orig_bytes, 1);
5068 if (flushing) {
5069 spin_lock(&space_info->lock);
5070 space_info->flush = 0;
5071 wake_up_all(&space_info->wait);
5072 spin_unlock(&space_info->lock);
5073 }
5074 return ret;
5075}
5076
5077static struct btrfs_block_rsv *get_block_rsv(
5078 const struct btrfs_trans_handle *trans,
5079 const struct btrfs_root *root)
5080{
5081 struct btrfs_block_rsv *block_rsv = NULL;
5082
5083 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5084 (root == root->fs_info->csum_root && trans->adding_csums) ||
5085 (root == root->fs_info->uuid_root))
5086 block_rsv = trans->block_rsv;
5087
5088 if (!block_rsv)
5089 block_rsv = root->block_rsv;
5090
5091 if (!block_rsv)
5092 block_rsv = &root->fs_info->empty_block_rsv;
5093
5094 return block_rsv;
5095}
5096
5097static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5098 u64 num_bytes)
5099{
5100 int ret = -ENOSPC;
5101 spin_lock(&block_rsv->lock);
5102 if (block_rsv->reserved >= num_bytes) {
5103 block_rsv->reserved -= num_bytes;
5104 if (block_rsv->reserved < block_rsv->size)
5105 block_rsv->full = 0;
5106 ret = 0;
5107 }
5108 spin_unlock(&block_rsv->lock);
5109 return ret;
5110}
5111
5112static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5113 u64 num_bytes, int update_size)
5114{
5115 spin_lock(&block_rsv->lock);
5116 block_rsv->reserved += num_bytes;
5117 if (update_size)
5118 block_rsv->size += num_bytes;
5119 else if (block_rsv->reserved >= block_rsv->size)
5120 block_rsv->full = 1;
5121 spin_unlock(&block_rsv->lock);
5122}
5123
5124int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5125 struct btrfs_block_rsv *dest, u64 num_bytes,
5126 int min_factor)
5127{
5128 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5129 u64 min_bytes;
5130
5131 if (global_rsv->space_info != dest->space_info)
5132 return -ENOSPC;
5133
5134 spin_lock(&global_rsv->lock);
5135 min_bytes = div_factor(global_rsv->size, min_factor);
5136 if (global_rsv->reserved < min_bytes + num_bytes) {
5137 spin_unlock(&global_rsv->lock);
5138 return -ENOSPC;
5139 }
5140 global_rsv->reserved -= num_bytes;
5141 if (global_rsv->reserved < global_rsv->size)
5142 global_rsv->full = 0;
5143 spin_unlock(&global_rsv->lock);
5144
5145 block_rsv_add_bytes(dest, num_bytes, 1);
5146 return 0;
5147}
5148
5149static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5150 struct btrfs_block_rsv *block_rsv,
5151 struct btrfs_block_rsv *dest, u64 num_bytes)
5152{
5153 struct btrfs_space_info *space_info = block_rsv->space_info;
5154
5155 spin_lock(&block_rsv->lock);
5156 if (num_bytes == (u64)-1)
5157 num_bytes = block_rsv->size;
5158 block_rsv->size -= num_bytes;
5159 if (block_rsv->reserved >= block_rsv->size) {
5160 num_bytes = block_rsv->reserved - block_rsv->size;
5161 block_rsv->reserved = block_rsv->size;
5162 block_rsv->full = 1;
5163 } else {
5164 num_bytes = 0;
5165 }
5166 spin_unlock(&block_rsv->lock);
5167
5168 if (num_bytes > 0) {
5169 if (dest) {
5170 spin_lock(&dest->lock);
5171 if (!dest->full) {
5172 u64 bytes_to_add;
5173
5174 bytes_to_add = dest->size - dest->reserved;
5175 bytes_to_add = min(num_bytes, bytes_to_add);
5176 dest->reserved += bytes_to_add;
5177 if (dest->reserved >= dest->size)
5178 dest->full = 1;
5179 num_bytes -= bytes_to_add;
5180 }
5181 spin_unlock(&dest->lock);
5182 }
5183 if (num_bytes) {
5184 spin_lock(&space_info->lock);
5185 space_info->bytes_may_use -= num_bytes;
5186 trace_btrfs_space_reservation(fs_info, "space_info",
5187 space_info->flags, num_bytes, 0);
5188 spin_unlock(&space_info->lock);
5189 }
5190 }
5191}
5192
5193static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
5194 struct btrfs_block_rsv *dst, u64 num_bytes)
5195{
5196 int ret;
5197
5198 ret = block_rsv_use_bytes(src, num_bytes);
5199 if (ret)
5200 return ret;
5201
5202 block_rsv_add_bytes(dst, num_bytes, 1);
5203 return 0;
5204}
5205
5206void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5207{
5208 memset(rsv, 0, sizeof(*rsv));
5209 spin_lock_init(&rsv->lock);
5210 rsv->type = type;
5211}
5212
5213struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5214 unsigned short type)
5215{
5216 struct btrfs_block_rsv *block_rsv;
5217 struct btrfs_fs_info *fs_info = root->fs_info;
5218
5219 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5220 if (!block_rsv)
5221 return NULL;
5222
5223 btrfs_init_block_rsv(block_rsv, type);
5224 block_rsv->space_info = __find_space_info(fs_info,
5225 BTRFS_BLOCK_GROUP_METADATA);
5226 return block_rsv;
5227}
5228
5229void btrfs_free_block_rsv(struct btrfs_root *root,
5230 struct btrfs_block_rsv *rsv)
5231{
5232 if (!rsv)
5233 return;
5234 btrfs_block_rsv_release(root, rsv, (u64)-1);
5235 kfree(rsv);
5236}
5237
5238void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5239{
5240 kfree(rsv);
5241}
5242
5243int btrfs_block_rsv_add(struct btrfs_root *root,
5244 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5245 enum btrfs_reserve_flush_enum flush)
5246{
5247 int ret;
5248
5249 if (num_bytes == 0)
5250 return 0;
5251
5252 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5253 if (!ret) {
5254 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5255 return 0;
5256 }
5257
5258 return ret;
5259}
5260
5261int btrfs_block_rsv_check(struct btrfs_root *root,
5262 struct btrfs_block_rsv *block_rsv, int min_factor)
5263{
5264 u64 num_bytes = 0;
5265 int ret = -ENOSPC;
5266
5267 if (!block_rsv)
5268 return 0;
5269
5270 spin_lock(&block_rsv->lock);
5271 num_bytes = div_factor(block_rsv->size, min_factor);
5272 if (block_rsv->reserved >= num_bytes)
5273 ret = 0;
5274 spin_unlock(&block_rsv->lock);
5275
5276 return ret;
5277}
5278
5279int btrfs_block_rsv_refill(struct btrfs_root *root,
5280 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5281 enum btrfs_reserve_flush_enum flush)
5282{
5283 u64 num_bytes = 0;
5284 int ret = -ENOSPC;
5285
5286 if (!block_rsv)
5287 return 0;
5288
5289 spin_lock(&block_rsv->lock);
5290 num_bytes = min_reserved;
5291 if (block_rsv->reserved >= num_bytes)
5292 ret = 0;
5293 else
5294 num_bytes -= block_rsv->reserved;
5295 spin_unlock(&block_rsv->lock);
5296
5297 if (!ret)
5298 return 0;
5299
5300 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5301 if (!ret) {
5302 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5303 return 0;
5304 }
5305
5306 return ret;
5307}
5308
5309int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5310 struct btrfs_block_rsv *dst_rsv,
5311 u64 num_bytes)
5312{
5313 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5314}
5315
5316void btrfs_block_rsv_release(struct btrfs_root *root,
5317 struct btrfs_block_rsv *block_rsv,
5318 u64 num_bytes)
5319{
5320 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5321 if (global_rsv == block_rsv ||
5322 block_rsv->space_info != global_rsv->space_info)
5323 global_rsv = NULL;
5324 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5325 num_bytes);
5326}
5327
5328/*
5329 * helper to calculate size of global block reservation.
5330 * the desired value is sum of space used by extent tree,
5331 * checksum tree and root tree
5332 */
5333static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5334{
5335 struct btrfs_space_info *sinfo;
5336 u64 num_bytes;
5337 u64 meta_used;
5338 u64 data_used;
5339 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5340
5341 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5342 spin_lock(&sinfo->lock);
5343 data_used = sinfo->bytes_used;
5344 spin_unlock(&sinfo->lock);
5345
5346 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5347 spin_lock(&sinfo->lock);
5348 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5349 data_used = 0;
5350 meta_used = sinfo->bytes_used;
5351 spin_unlock(&sinfo->lock);
5352
5353 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5354 csum_size * 2;
5355 num_bytes += div_u64(data_used + meta_used, 50);
5356
5357 if (num_bytes * 3 > meta_used)
5358 num_bytes = div_u64(meta_used, 3);
5359
5360 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5361}
5362
5363static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5364{
5365 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5366 struct btrfs_space_info *sinfo = block_rsv->space_info;
5367 u64 num_bytes;
5368
5369 num_bytes = calc_global_metadata_size(fs_info);
5370
5371 spin_lock(&sinfo->lock);
5372 spin_lock(&block_rsv->lock);
5373
5374 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5375
5376 if (block_rsv->reserved < block_rsv->size) {
5377 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5378 sinfo->bytes_reserved + sinfo->bytes_readonly +
5379 sinfo->bytes_may_use;
5380 if (sinfo->total_bytes > num_bytes) {
5381 num_bytes = sinfo->total_bytes - num_bytes;
5382 num_bytes = min(num_bytes,
5383 block_rsv->size - block_rsv->reserved);
5384 block_rsv->reserved += num_bytes;
5385 sinfo->bytes_may_use += num_bytes;
5386 trace_btrfs_space_reservation(fs_info, "space_info",
5387 sinfo->flags, num_bytes,
5388 1);
5389 }
5390 } else if (block_rsv->reserved > block_rsv->size) {
5391 num_bytes = block_rsv->reserved - block_rsv->size;
5392 sinfo->bytes_may_use -= num_bytes;
5393 trace_btrfs_space_reservation(fs_info, "space_info",
5394 sinfo->flags, num_bytes, 0);
5395 block_rsv->reserved = block_rsv->size;
5396 }
5397
5398 if (block_rsv->reserved == block_rsv->size)
5399 block_rsv->full = 1;
5400 else
5401 block_rsv->full = 0;
5402
5403 spin_unlock(&block_rsv->lock);
5404 spin_unlock(&sinfo->lock);
5405}
5406
5407static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5408{
5409 struct btrfs_space_info *space_info;
5410
5411 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5412 fs_info->chunk_block_rsv.space_info = space_info;
5413
5414 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5415 fs_info->global_block_rsv.space_info = space_info;
5416 fs_info->delalloc_block_rsv.space_info = space_info;
5417 fs_info->trans_block_rsv.space_info = space_info;
5418 fs_info->empty_block_rsv.space_info = space_info;
5419 fs_info->delayed_block_rsv.space_info = space_info;
5420
5421 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5422 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5423 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5424 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5425 if (fs_info->quota_root)
5426 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5427 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5428
5429 update_global_block_rsv(fs_info);
5430}
5431
5432static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5433{
5434 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5435 (u64)-1);
5436 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5437 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5438 WARN_ON(fs_info->trans_block_rsv.size > 0);
5439 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5440 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5441 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5442 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5443 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5444}
5445
5446void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5447 struct btrfs_root *root)
5448{
5449 if (!trans->block_rsv)
5450 return;
5451
5452 if (!trans->bytes_reserved)
5453 return;
5454
5455 trace_btrfs_space_reservation(root->fs_info, "transaction",
5456 trans->transid, trans->bytes_reserved, 0);
5457 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5458 trans->bytes_reserved = 0;
5459}
5460
5461/*
5462 * To be called after all the new block groups attached to the transaction
5463 * handle have been created (btrfs_create_pending_block_groups()).
5464 */
5465void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5466{
5467 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5468
5469 if (!trans->chunk_bytes_reserved)
5470 return;
5471
5472 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5473
5474 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5475 trans->chunk_bytes_reserved);
5476 trans->chunk_bytes_reserved = 0;
5477}
5478
5479/* Can only return 0 or -ENOSPC */
5480int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5481 struct inode *inode)
5482{
5483 struct btrfs_root *root = BTRFS_I(inode)->root;
5484 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5485 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5486
5487 /*
5488 * We need to hold space in order to delete our orphan item once we've
5489 * added it, so this takes the reservation so we can release it later
5490 * when we are truly done with the orphan item.
5491 */
5492 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5493 trace_btrfs_space_reservation(root->fs_info, "orphan",
5494 btrfs_ino(inode), num_bytes, 1);
5495 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5496}
5497
5498void btrfs_orphan_release_metadata(struct inode *inode)
5499{
5500 struct btrfs_root *root = BTRFS_I(inode)->root;
5501 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5502 trace_btrfs_space_reservation(root->fs_info, "orphan",
5503 btrfs_ino(inode), num_bytes, 0);
5504 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5505}
5506
5507/*
5508 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5509 * root: the root of the parent directory
5510 * rsv: block reservation
5511 * items: the number of items that we need do reservation
5512 * qgroup_reserved: used to return the reserved size in qgroup
5513 *
5514 * This function is used to reserve the space for snapshot/subvolume
5515 * creation and deletion. Those operations are different with the
5516 * common file/directory operations, they change two fs/file trees
5517 * and root tree, the number of items that the qgroup reserves is
5518 * different with the free space reservation. So we can not use
5519 * the space reseravtion mechanism in start_transaction().
5520 */
5521int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5522 struct btrfs_block_rsv *rsv,
5523 int items,
5524 u64 *qgroup_reserved,
5525 bool use_global_rsv)
5526{
5527 u64 num_bytes;
5528 int ret;
5529 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5530
5531 if (root->fs_info->quota_enabled) {
5532 /* One for parent inode, two for dir entries */
5533 num_bytes = 3 * root->nodesize;
5534 ret = btrfs_qgroup_reserve_meta(root, num_bytes);
5535 if (ret)
5536 return ret;
5537 } else {
5538 num_bytes = 0;
5539 }
5540
5541 *qgroup_reserved = num_bytes;
5542
5543 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5544 rsv->space_info = __find_space_info(root->fs_info,
5545 BTRFS_BLOCK_GROUP_METADATA);
5546 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5547 BTRFS_RESERVE_FLUSH_ALL);
5548
5549 if (ret == -ENOSPC && use_global_rsv)
5550 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5551
5552 if (ret && *qgroup_reserved)
5553 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5554
5555 return ret;
5556}
5557
5558void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5559 struct btrfs_block_rsv *rsv,
5560 u64 qgroup_reserved)
5561{
5562 btrfs_block_rsv_release(root, rsv, (u64)-1);
5563}
5564
5565/**
5566 * drop_outstanding_extent - drop an outstanding extent
5567 * @inode: the inode we're dropping the extent for
5568 * @num_bytes: the number of bytes we're relaseing.
5569 *
5570 * This is called when we are freeing up an outstanding extent, either called
5571 * after an error or after an extent is written. This will return the number of
5572 * reserved extents that need to be freed. This must be called with
5573 * BTRFS_I(inode)->lock held.
5574 */
5575static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5576{
5577 unsigned drop_inode_space = 0;
5578 unsigned dropped_extents = 0;
5579 unsigned num_extents = 0;
5580
5581 num_extents = (unsigned)div64_u64(num_bytes +
5582 BTRFS_MAX_EXTENT_SIZE - 1,
5583 BTRFS_MAX_EXTENT_SIZE);
5584 ASSERT(num_extents);
5585 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5586 BTRFS_I(inode)->outstanding_extents -= num_extents;
5587
5588 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5589 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5590 &BTRFS_I(inode)->runtime_flags))
5591 drop_inode_space = 1;
5592
5593 /*
5594 * If we have more or the same amount of outsanding extents than we have
5595 * reserved then we need to leave the reserved extents count alone.
5596 */
5597 if (BTRFS_I(inode)->outstanding_extents >=
5598 BTRFS_I(inode)->reserved_extents)
5599 return drop_inode_space;
5600
5601 dropped_extents = BTRFS_I(inode)->reserved_extents -
5602 BTRFS_I(inode)->outstanding_extents;
5603 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5604 return dropped_extents + drop_inode_space;
5605}
5606
5607/**
5608 * calc_csum_metadata_size - return the amount of metada space that must be
5609 * reserved/free'd for the given bytes.
5610 * @inode: the inode we're manipulating
5611 * @num_bytes: the number of bytes in question
5612 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5613 *
5614 * This adjusts the number of csum_bytes in the inode and then returns the
5615 * correct amount of metadata that must either be reserved or freed. We
5616 * calculate how many checksums we can fit into one leaf and then divide the
5617 * number of bytes that will need to be checksumed by this value to figure out
5618 * how many checksums will be required. If we are adding bytes then the number
5619 * may go up and we will return the number of additional bytes that must be
5620 * reserved. If it is going down we will return the number of bytes that must
5621 * be freed.
5622 *
5623 * This must be called with BTRFS_I(inode)->lock held.
5624 */
5625static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5626 int reserve)
5627{
5628 struct btrfs_root *root = BTRFS_I(inode)->root;
5629 u64 old_csums, num_csums;
5630
5631 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5632 BTRFS_I(inode)->csum_bytes == 0)
5633 return 0;
5634
5635 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5636 if (reserve)
5637 BTRFS_I(inode)->csum_bytes += num_bytes;
5638 else
5639 BTRFS_I(inode)->csum_bytes -= num_bytes;
5640 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5641
5642 /* No change, no need to reserve more */
5643 if (old_csums == num_csums)
5644 return 0;
5645
5646 if (reserve)
5647 return btrfs_calc_trans_metadata_size(root,
5648 num_csums - old_csums);
5649
5650 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5651}
5652
5653int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5654{
5655 struct btrfs_root *root = BTRFS_I(inode)->root;
5656 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5657 u64 to_reserve = 0;
5658 u64 csum_bytes;
5659 unsigned nr_extents = 0;
5660 int extra_reserve = 0;
5661 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5662 int ret = 0;
5663 bool delalloc_lock = true;
5664 u64 to_free = 0;
5665 unsigned dropped;
5666
5667 /* If we are a free space inode we need to not flush since we will be in
5668 * the middle of a transaction commit. We also don't need the delalloc
5669 * mutex since we won't race with anybody. We need this mostly to make
5670 * lockdep shut its filthy mouth.
5671 */
5672 if (btrfs_is_free_space_inode(inode)) {
5673 flush = BTRFS_RESERVE_NO_FLUSH;
5674 delalloc_lock = false;
5675 }
5676
5677 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5678 btrfs_transaction_in_commit(root->fs_info))
5679 schedule_timeout(1);
5680
5681 if (delalloc_lock)
5682 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5683
5684 num_bytes = ALIGN(num_bytes, root->sectorsize);
5685
5686 spin_lock(&BTRFS_I(inode)->lock);
5687 nr_extents = (unsigned)div64_u64(num_bytes +
5688 BTRFS_MAX_EXTENT_SIZE - 1,
5689 BTRFS_MAX_EXTENT_SIZE);
5690 BTRFS_I(inode)->outstanding_extents += nr_extents;
5691 nr_extents = 0;
5692
5693 if (BTRFS_I(inode)->outstanding_extents >
5694 BTRFS_I(inode)->reserved_extents)
5695 nr_extents = BTRFS_I(inode)->outstanding_extents -
5696 BTRFS_I(inode)->reserved_extents;
5697
5698 /*
5699 * Add an item to reserve for updating the inode when we complete the
5700 * delalloc io.
5701 */
5702 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5703 &BTRFS_I(inode)->runtime_flags)) {
5704 nr_extents++;
5705 extra_reserve = 1;
5706 }
5707
5708 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5709 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5710 csum_bytes = BTRFS_I(inode)->csum_bytes;
5711 spin_unlock(&BTRFS_I(inode)->lock);
5712
5713 if (root->fs_info->quota_enabled) {
5714 ret = btrfs_qgroup_reserve_meta(root,
5715 nr_extents * root->nodesize);
5716 if (ret)
5717 goto out_fail;
5718 }
5719
5720 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5721 if (unlikely(ret)) {
5722 btrfs_qgroup_free_meta(root, nr_extents * root->nodesize);
5723 goto out_fail;
5724 }
5725
5726 spin_lock(&BTRFS_I(inode)->lock);
5727 if (extra_reserve) {
5728 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5729 &BTRFS_I(inode)->runtime_flags);
5730 nr_extents--;
5731 }
5732 BTRFS_I(inode)->reserved_extents += nr_extents;
5733 spin_unlock(&BTRFS_I(inode)->lock);
5734
5735 if (delalloc_lock)
5736 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5737
5738 if (to_reserve)
5739 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5740 btrfs_ino(inode), to_reserve, 1);
5741 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5742
5743 return 0;
5744
5745out_fail:
5746 spin_lock(&BTRFS_I(inode)->lock);
5747 dropped = drop_outstanding_extent(inode, num_bytes);
5748 /*
5749 * If the inodes csum_bytes is the same as the original
5750 * csum_bytes then we know we haven't raced with any free()ers
5751 * so we can just reduce our inodes csum bytes and carry on.
5752 */
5753 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5754 calc_csum_metadata_size(inode, num_bytes, 0);
5755 } else {
5756 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5757 u64 bytes;
5758
5759 /*
5760 * This is tricky, but first we need to figure out how much we
5761 * free'd from any free-ers that occurred during this
5762 * reservation, so we reset ->csum_bytes to the csum_bytes
5763 * before we dropped our lock, and then call the free for the
5764 * number of bytes that were freed while we were trying our
5765 * reservation.
5766 */
5767 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5768 BTRFS_I(inode)->csum_bytes = csum_bytes;
5769 to_free = calc_csum_metadata_size(inode, bytes, 0);
5770
5771
5772 /*
5773 * Now we need to see how much we would have freed had we not
5774 * been making this reservation and our ->csum_bytes were not
5775 * artificially inflated.
5776 */
5777 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5778 bytes = csum_bytes - orig_csum_bytes;
5779 bytes = calc_csum_metadata_size(inode, bytes, 0);
5780
5781 /*
5782 * Now reset ->csum_bytes to what it should be. If bytes is
5783 * more than to_free then we would have free'd more space had we
5784 * not had an artificially high ->csum_bytes, so we need to free
5785 * the remainder. If bytes is the same or less then we don't
5786 * need to do anything, the other free-ers did the correct
5787 * thing.
5788 */
5789 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5790 if (bytes > to_free)
5791 to_free = bytes - to_free;
5792 else
5793 to_free = 0;
5794 }
5795 spin_unlock(&BTRFS_I(inode)->lock);
5796 if (dropped)
5797 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5798
5799 if (to_free) {
5800 btrfs_block_rsv_release(root, block_rsv, to_free);
5801 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5802 btrfs_ino(inode), to_free, 0);
5803 }
5804 if (delalloc_lock)
5805 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5806 return ret;
5807}
5808
5809/**
5810 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5811 * @inode: the inode to release the reservation for
5812 * @num_bytes: the number of bytes we're releasing
5813 *
5814 * This will release the metadata reservation for an inode. This can be called
5815 * once we complete IO for a given set of bytes to release their metadata
5816 * reservations.
5817 */
5818void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5819{
5820 struct btrfs_root *root = BTRFS_I(inode)->root;
5821 u64 to_free = 0;
5822 unsigned dropped;
5823
5824 num_bytes = ALIGN(num_bytes, root->sectorsize);
5825 spin_lock(&BTRFS_I(inode)->lock);
5826 dropped = drop_outstanding_extent(inode, num_bytes);
5827
5828 if (num_bytes)
5829 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5830 spin_unlock(&BTRFS_I(inode)->lock);
5831 if (dropped > 0)
5832 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5833
5834 if (btrfs_test_is_dummy_root(root))
5835 return;
5836
5837 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5838 btrfs_ino(inode), to_free, 0);
5839
5840 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5841 to_free);
5842}
5843
5844/**
5845 * btrfs_delalloc_reserve_space - reserve data and metadata space for
5846 * delalloc
5847 * @inode: inode we're writing to
5848 * @start: start range we are writing to
5849 * @len: how long the range we are writing to
5850 *
5851 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
5852 *
5853 * This will do the following things
5854 *
5855 * o reserve space in data space info for num bytes
5856 * and reserve precious corresponding qgroup space
5857 * (Done in check_data_free_space)
5858 *
5859 * o reserve space for metadata space, based on the number of outstanding
5860 * extents and how much csums will be needed
5861 * also reserve metadata space in a per root over-reserve method.
5862 * o add to the inodes->delalloc_bytes
5863 * o add it to the fs_info's delalloc inodes list.
5864 * (Above 3 all done in delalloc_reserve_metadata)
5865 *
5866 * Return 0 for success
5867 * Return <0 for error(-ENOSPC or -EQUOT)
5868 */
5869int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len)
5870{
5871 int ret;
5872
5873 ret = btrfs_check_data_free_space(inode, start, len);
5874 if (ret < 0)
5875 return ret;
5876 ret = btrfs_delalloc_reserve_metadata(inode, len);
5877 if (ret < 0)
5878 btrfs_free_reserved_data_space(inode, start, len);
5879 return ret;
5880}
5881
5882/**
5883 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5884 * @inode: inode we're releasing space for
5885 * @start: start position of the space already reserved
5886 * @len: the len of the space already reserved
5887 *
5888 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5889 * called in the case that we don't need the metadata AND data reservations
5890 * anymore. So if there is an error or we insert an inline extent.
5891 *
5892 * This function will release the metadata space that was not used and will
5893 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5894 * list if there are no delalloc bytes left.
5895 * Also it will handle the qgroup reserved space.
5896 */
5897void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len)
5898{
5899 btrfs_delalloc_release_metadata(inode, len);
5900 btrfs_free_reserved_data_space(inode, start, len);
5901}
5902
5903static int update_block_group(struct btrfs_trans_handle *trans,
5904 struct btrfs_root *root, u64 bytenr,
5905 u64 num_bytes, int alloc)
5906{
5907 struct btrfs_block_group_cache *cache = NULL;
5908 struct btrfs_fs_info *info = root->fs_info;
5909 u64 total = num_bytes;
5910 u64 old_val;
5911 u64 byte_in_group;
5912 int factor;
5913
5914 /* block accounting for super block */
5915 spin_lock(&info->delalloc_root_lock);
5916 old_val = btrfs_super_bytes_used(info->super_copy);
5917 if (alloc)
5918 old_val += num_bytes;
5919 else
5920 old_val -= num_bytes;
5921 btrfs_set_super_bytes_used(info->super_copy, old_val);
5922 spin_unlock(&info->delalloc_root_lock);
5923
5924 while (total) {
5925 cache = btrfs_lookup_block_group(info, bytenr);
5926 if (!cache)
5927 return -ENOENT;
5928 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5929 BTRFS_BLOCK_GROUP_RAID1 |
5930 BTRFS_BLOCK_GROUP_RAID10))
5931 factor = 2;
5932 else
5933 factor = 1;
5934 /*
5935 * If this block group has free space cache written out, we
5936 * need to make sure to load it if we are removing space. This
5937 * is because we need the unpinning stage to actually add the
5938 * space back to the block group, otherwise we will leak space.
5939 */
5940 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5941 cache_block_group(cache, 1);
5942
5943 byte_in_group = bytenr - cache->key.objectid;
5944 WARN_ON(byte_in_group > cache->key.offset);
5945
5946 spin_lock(&cache->space_info->lock);
5947 spin_lock(&cache->lock);
5948
5949 if (btrfs_test_opt(root, SPACE_CACHE) &&
5950 cache->disk_cache_state < BTRFS_DC_CLEAR)
5951 cache->disk_cache_state = BTRFS_DC_CLEAR;
5952
5953 old_val = btrfs_block_group_used(&cache->item);
5954 num_bytes = min(total, cache->key.offset - byte_in_group);
5955 if (alloc) {
5956 old_val += num_bytes;
5957 btrfs_set_block_group_used(&cache->item, old_val);
5958 cache->reserved -= num_bytes;
5959 cache->space_info->bytes_reserved -= num_bytes;
5960 cache->space_info->bytes_used += num_bytes;
5961 cache->space_info->disk_used += num_bytes * factor;
5962 spin_unlock(&cache->lock);
5963 spin_unlock(&cache->space_info->lock);
5964 } else {
5965 old_val -= num_bytes;
5966 btrfs_set_block_group_used(&cache->item, old_val);
5967 cache->pinned += num_bytes;
5968 cache->space_info->bytes_pinned += num_bytes;
5969 cache->space_info->bytes_used -= num_bytes;
5970 cache->space_info->disk_used -= num_bytes * factor;
5971 spin_unlock(&cache->lock);
5972 spin_unlock(&cache->space_info->lock);
5973
5974 set_extent_dirty(info->pinned_extents,
5975 bytenr, bytenr + num_bytes - 1,
5976 GFP_NOFS | __GFP_NOFAIL);
5977 }
5978
5979 spin_lock(&trans->transaction->dirty_bgs_lock);
5980 if (list_empty(&cache->dirty_list)) {
5981 list_add_tail(&cache->dirty_list,
5982 &trans->transaction->dirty_bgs);
5983 trans->transaction->num_dirty_bgs++;
5984 btrfs_get_block_group(cache);
5985 }
5986 spin_unlock(&trans->transaction->dirty_bgs_lock);
5987
5988 /*
5989 * No longer have used bytes in this block group, queue it for
5990 * deletion. We do this after adding the block group to the
5991 * dirty list to avoid races between cleaner kthread and space
5992 * cache writeout.
5993 */
5994 if (!alloc && old_val == 0) {
5995 spin_lock(&info->unused_bgs_lock);
5996 if (list_empty(&cache->bg_list)) {
5997 btrfs_get_block_group(cache);
5998 list_add_tail(&cache->bg_list,
5999 &info->unused_bgs);
6000 }
6001 spin_unlock(&info->unused_bgs_lock);
6002 }
6003
6004 btrfs_put_block_group(cache);
6005 total -= num_bytes;
6006 bytenr += num_bytes;
6007 }
6008 return 0;
6009}
6010
6011static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
6012{
6013 struct btrfs_block_group_cache *cache;
6014 u64 bytenr;
6015
6016 spin_lock(&root->fs_info->block_group_cache_lock);
6017 bytenr = root->fs_info->first_logical_byte;
6018 spin_unlock(&root->fs_info->block_group_cache_lock);
6019
6020 if (bytenr < (u64)-1)
6021 return bytenr;
6022
6023 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
6024 if (!cache)
6025 return 0;
6026
6027 bytenr = cache->key.objectid;
6028 btrfs_put_block_group(cache);
6029
6030 return bytenr;
6031}
6032
6033static int pin_down_extent(struct btrfs_root *root,
6034 struct btrfs_block_group_cache *cache,
6035 u64 bytenr, u64 num_bytes, int reserved)
6036{
6037 spin_lock(&cache->space_info->lock);
6038 spin_lock(&cache->lock);
6039 cache->pinned += num_bytes;
6040 cache->space_info->bytes_pinned += num_bytes;
6041 if (reserved) {
6042 cache->reserved -= num_bytes;
6043 cache->space_info->bytes_reserved -= num_bytes;
6044 }
6045 spin_unlock(&cache->lock);
6046 spin_unlock(&cache->space_info->lock);
6047
6048 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
6049 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6050 if (reserved)
6051 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
6052 return 0;
6053}
6054
6055/*
6056 * this function must be called within transaction
6057 */
6058int btrfs_pin_extent(struct btrfs_root *root,
6059 u64 bytenr, u64 num_bytes, int reserved)
6060{
6061 struct btrfs_block_group_cache *cache;
6062
6063 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6064 BUG_ON(!cache); /* Logic error */
6065
6066 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
6067
6068 btrfs_put_block_group(cache);
6069 return 0;
6070}
6071
6072/*
6073 * this function must be called within transaction
6074 */
6075int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
6076 u64 bytenr, u64 num_bytes)
6077{
6078 struct btrfs_block_group_cache *cache;
6079 int ret;
6080
6081 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6082 if (!cache)
6083 return -EINVAL;
6084
6085 /*
6086 * pull in the free space cache (if any) so that our pin
6087 * removes the free space from the cache. We have load_only set
6088 * to one because the slow code to read in the free extents does check
6089 * the pinned extents.
6090 */
6091 cache_block_group(cache, 1);
6092
6093 pin_down_extent(root, cache, bytenr, num_bytes, 0);
6094
6095 /* remove us from the free space cache (if we're there at all) */
6096 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6097 btrfs_put_block_group(cache);
6098 return ret;
6099}
6100
6101static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
6102{
6103 int ret;
6104 struct btrfs_block_group_cache *block_group;
6105 struct btrfs_caching_control *caching_ctl;
6106
6107 block_group = btrfs_lookup_block_group(root->fs_info, start);
6108 if (!block_group)
6109 return -EINVAL;
6110
6111 cache_block_group(block_group, 0);
6112 caching_ctl = get_caching_control(block_group);
6113
6114 if (!caching_ctl) {
6115 /* Logic error */
6116 BUG_ON(!block_group_cache_done(block_group));
6117 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6118 } else {
6119 mutex_lock(&caching_ctl->mutex);
6120
6121 if (start >= caching_ctl->progress) {
6122 ret = add_excluded_extent(root, start, num_bytes);
6123 } else if (start + num_bytes <= caching_ctl->progress) {
6124 ret = btrfs_remove_free_space(block_group,
6125 start, num_bytes);
6126 } else {
6127 num_bytes = caching_ctl->progress - start;
6128 ret = btrfs_remove_free_space(block_group,
6129 start, num_bytes);
6130 if (ret)
6131 goto out_lock;
6132
6133 num_bytes = (start + num_bytes) -
6134 caching_ctl->progress;
6135 start = caching_ctl->progress;
6136 ret = add_excluded_extent(root, start, num_bytes);
6137 }
6138out_lock:
6139 mutex_unlock(&caching_ctl->mutex);
6140 put_caching_control(caching_ctl);
6141 }
6142 btrfs_put_block_group(block_group);
6143 return ret;
6144}
6145
6146int btrfs_exclude_logged_extents(struct btrfs_root *log,
6147 struct extent_buffer *eb)
6148{
6149 struct btrfs_file_extent_item *item;
6150 struct btrfs_key key;
6151 int found_type;
6152 int i;
6153
6154 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
6155 return 0;
6156
6157 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6158 btrfs_item_key_to_cpu(eb, &key, i);
6159 if (key.type != BTRFS_EXTENT_DATA_KEY)
6160 continue;
6161 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6162 found_type = btrfs_file_extent_type(eb, item);
6163 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6164 continue;
6165 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6166 continue;
6167 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6168 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6169 __exclude_logged_extent(log, key.objectid, key.offset);
6170 }
6171
6172 return 0;
6173}
6174
6175/**
6176 * btrfs_update_reserved_bytes - update the block_group and space info counters
6177 * @cache: The cache we are manipulating
6178 * @num_bytes: The number of bytes in question
6179 * @reserve: One of the reservation enums
6180 * @delalloc: The blocks are allocated for the delalloc write
6181 *
6182 * This is called by the allocator when it reserves space, or by somebody who is
6183 * freeing space that was never actually used on disk. For example if you
6184 * reserve some space for a new leaf in transaction A and before transaction A
6185 * commits you free that leaf, you call this with reserve set to 0 in order to
6186 * clear the reservation.
6187 *
6188 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
6189 * ENOSPC accounting. For data we handle the reservation through clearing the
6190 * delalloc bits in the io_tree. We have to do this since we could end up
6191 * allocating less disk space for the amount of data we have reserved in the
6192 * case of compression.
6193 *
6194 * If this is a reservation and the block group has become read only we cannot
6195 * make the reservation and return -EAGAIN, otherwise this function always
6196 * succeeds.
6197 */
6198static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
6199 u64 num_bytes, int reserve, int delalloc)
6200{
6201 struct btrfs_space_info *space_info = cache->space_info;
6202 int ret = 0;
6203
6204 spin_lock(&space_info->lock);
6205 spin_lock(&cache->lock);
6206 if (reserve != RESERVE_FREE) {
6207 if (cache->ro) {
6208 ret = -EAGAIN;
6209 } else {
6210 cache->reserved += num_bytes;
6211 space_info->bytes_reserved += num_bytes;
6212 if (reserve == RESERVE_ALLOC) {
6213 trace_btrfs_space_reservation(cache->fs_info,
6214 "space_info", space_info->flags,
6215 num_bytes, 0);
6216 space_info->bytes_may_use -= num_bytes;
6217 }
6218
6219 if (delalloc)
6220 cache->delalloc_bytes += num_bytes;
6221 }
6222 } else {
6223 if (cache->ro)
6224 space_info->bytes_readonly += num_bytes;
6225 cache->reserved -= num_bytes;
6226 space_info->bytes_reserved -= num_bytes;
6227
6228 if (delalloc)
6229 cache->delalloc_bytes -= num_bytes;
6230 }
6231 spin_unlock(&cache->lock);
6232 spin_unlock(&space_info->lock);
6233 return ret;
6234}
6235
6236void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6237 struct btrfs_root *root)
6238{
6239 struct btrfs_fs_info *fs_info = root->fs_info;
6240 struct btrfs_caching_control *next;
6241 struct btrfs_caching_control *caching_ctl;
6242 struct btrfs_block_group_cache *cache;
6243
6244 down_write(&fs_info->commit_root_sem);
6245
6246 list_for_each_entry_safe(caching_ctl, next,
6247 &fs_info->caching_block_groups, list) {
6248 cache = caching_ctl->block_group;
6249 if (block_group_cache_done(cache)) {
6250 cache->last_byte_to_unpin = (u64)-1;
6251 list_del_init(&caching_ctl->list);
6252 put_caching_control(caching_ctl);
6253 } else {
6254 cache->last_byte_to_unpin = caching_ctl->progress;
6255 }
6256 }
6257
6258 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6259 fs_info->pinned_extents = &fs_info->freed_extents[1];
6260 else
6261 fs_info->pinned_extents = &fs_info->freed_extents[0];
6262
6263 up_write(&fs_info->commit_root_sem);
6264
6265 update_global_block_rsv(fs_info);
6266}
6267
6268/*
6269 * Returns the free cluster for the given space info and sets empty_cluster to
6270 * what it should be based on the mount options.
6271 */
6272static struct btrfs_free_cluster *
6273fetch_cluster_info(struct btrfs_root *root, struct btrfs_space_info *space_info,
6274 u64 *empty_cluster)
6275{
6276 struct btrfs_free_cluster *ret = NULL;
6277 bool ssd = btrfs_test_opt(root, SSD);
6278
6279 *empty_cluster = 0;
6280 if (btrfs_mixed_space_info(space_info))
6281 return ret;
6282
6283 if (ssd)
6284 *empty_cluster = SZ_2M;
6285 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6286 ret = &root->fs_info->meta_alloc_cluster;
6287 if (!ssd)
6288 *empty_cluster = SZ_64K;
6289 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) {
6290 ret = &root->fs_info->data_alloc_cluster;
6291 }
6292
6293 return ret;
6294}
6295
6296static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6297 const bool return_free_space)
6298{
6299 struct btrfs_fs_info *fs_info = root->fs_info;
6300 struct btrfs_block_group_cache *cache = NULL;
6301 struct btrfs_space_info *space_info;
6302 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6303 struct btrfs_free_cluster *cluster = NULL;
6304 u64 len;
6305 u64 total_unpinned = 0;
6306 u64 empty_cluster = 0;
6307 bool readonly;
6308
6309 while (start <= end) {
6310 readonly = false;
6311 if (!cache ||
6312 start >= cache->key.objectid + cache->key.offset) {
6313 if (cache)
6314 btrfs_put_block_group(cache);
6315 total_unpinned = 0;
6316 cache = btrfs_lookup_block_group(fs_info, start);
6317 BUG_ON(!cache); /* Logic error */
6318
6319 cluster = fetch_cluster_info(root,
6320 cache->space_info,
6321 &empty_cluster);
6322 empty_cluster <<= 1;
6323 }
6324
6325 len = cache->key.objectid + cache->key.offset - start;
6326 len = min(len, end + 1 - start);
6327
6328 if (start < cache->last_byte_to_unpin) {
6329 len = min(len, cache->last_byte_to_unpin - start);
6330 if (return_free_space)
6331 btrfs_add_free_space(cache, start, len);
6332 }
6333
6334 start += len;
6335 total_unpinned += len;
6336 space_info = cache->space_info;
6337
6338 /*
6339 * If this space cluster has been marked as fragmented and we've
6340 * unpinned enough in this block group to potentially allow a
6341 * cluster to be created inside of it go ahead and clear the
6342 * fragmented check.
6343 */
6344 if (cluster && cluster->fragmented &&
6345 total_unpinned > empty_cluster) {
6346 spin_lock(&cluster->lock);
6347 cluster->fragmented = 0;
6348 spin_unlock(&cluster->lock);
6349 }
6350
6351 spin_lock(&space_info->lock);
6352 spin_lock(&cache->lock);
6353 cache->pinned -= len;
6354 space_info->bytes_pinned -= len;
6355 space_info->max_extent_size = 0;
6356 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6357 if (cache->ro) {
6358 space_info->bytes_readonly += len;
6359 readonly = true;
6360 }
6361 spin_unlock(&cache->lock);
6362 if (!readonly && global_rsv->space_info == space_info) {
6363 spin_lock(&global_rsv->lock);
6364 if (!global_rsv->full) {
6365 len = min(len, global_rsv->size -
6366 global_rsv->reserved);
6367 global_rsv->reserved += len;
6368 space_info->bytes_may_use += len;
6369 if (global_rsv->reserved >= global_rsv->size)
6370 global_rsv->full = 1;
6371 }
6372 spin_unlock(&global_rsv->lock);
6373 }
6374 spin_unlock(&space_info->lock);
6375 }
6376
6377 if (cache)
6378 btrfs_put_block_group(cache);
6379 return 0;
6380}
6381
6382int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6383 struct btrfs_root *root)
6384{
6385 struct btrfs_fs_info *fs_info = root->fs_info;
6386 struct btrfs_block_group_cache *block_group, *tmp;
6387 struct list_head *deleted_bgs;
6388 struct extent_io_tree *unpin;
6389 u64 start;
6390 u64 end;
6391 int ret;
6392
6393 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6394 unpin = &fs_info->freed_extents[1];
6395 else
6396 unpin = &fs_info->freed_extents[0];
6397
6398 while (!trans->aborted) {
6399 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6400 ret = find_first_extent_bit(unpin, 0, &start, &end,
6401 EXTENT_DIRTY, NULL);
6402 if (ret) {
6403 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6404 break;
6405 }
6406
6407 if (btrfs_test_opt(root, DISCARD))
6408 ret = btrfs_discard_extent(root, start,
6409 end + 1 - start, NULL);
6410
6411 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6412 unpin_extent_range(root, start, end, true);
6413 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6414 cond_resched();
6415 }
6416
6417 /*
6418 * Transaction is finished. We don't need the lock anymore. We
6419 * do need to clean up the block groups in case of a transaction
6420 * abort.
6421 */
6422 deleted_bgs = &trans->transaction->deleted_bgs;
6423 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6424 u64 trimmed = 0;
6425
6426 ret = -EROFS;
6427 if (!trans->aborted)
6428 ret = btrfs_discard_extent(root,
6429 block_group->key.objectid,
6430 block_group->key.offset,
6431 &trimmed);
6432
6433 list_del_init(&block_group->bg_list);
6434 btrfs_put_block_group_trimming(block_group);
6435 btrfs_put_block_group(block_group);
6436
6437 if (ret) {
6438 const char *errstr = btrfs_decode_error(ret);
6439 btrfs_warn(fs_info,
6440 "Discard failed while removing blockgroup: errno=%d %s\n",
6441 ret, errstr);
6442 }
6443 }
6444
6445 return 0;
6446}
6447
6448static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6449 u64 owner, u64 root_objectid)
6450{
6451 struct btrfs_space_info *space_info;
6452 u64 flags;
6453
6454 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6455 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6456 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6457 else
6458 flags = BTRFS_BLOCK_GROUP_METADATA;
6459 } else {
6460 flags = BTRFS_BLOCK_GROUP_DATA;
6461 }
6462
6463 space_info = __find_space_info(fs_info, flags);
6464 BUG_ON(!space_info); /* Logic bug */
6465 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6466}
6467
6468
6469static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6470 struct btrfs_root *root,
6471 struct btrfs_delayed_ref_node *node, u64 parent,
6472 u64 root_objectid, u64 owner_objectid,
6473 u64 owner_offset, int refs_to_drop,
6474 struct btrfs_delayed_extent_op *extent_op)
6475{
6476 struct btrfs_key key;
6477 struct btrfs_path *path;
6478 struct btrfs_fs_info *info = root->fs_info;
6479 struct btrfs_root *extent_root = info->extent_root;
6480 struct extent_buffer *leaf;
6481 struct btrfs_extent_item *ei;
6482 struct btrfs_extent_inline_ref *iref;
6483 int ret;
6484 int is_data;
6485 int extent_slot = 0;
6486 int found_extent = 0;
6487 int num_to_del = 1;
6488 u32 item_size;
6489 u64 refs;
6490 u64 bytenr = node->bytenr;
6491 u64 num_bytes = node->num_bytes;
6492 int last_ref = 0;
6493 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6494 SKINNY_METADATA);
6495
6496 path = btrfs_alloc_path();
6497 if (!path)
6498 return -ENOMEM;
6499
6500 path->reada = READA_FORWARD;
6501 path->leave_spinning = 1;
6502
6503 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6504 BUG_ON(!is_data && refs_to_drop != 1);
6505
6506 if (is_data)
6507 skinny_metadata = 0;
6508
6509 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6510 bytenr, num_bytes, parent,
6511 root_objectid, owner_objectid,
6512 owner_offset);
6513 if (ret == 0) {
6514 extent_slot = path->slots[0];
6515 while (extent_slot >= 0) {
6516 btrfs_item_key_to_cpu(path->nodes[0], &key,
6517 extent_slot);
6518 if (key.objectid != bytenr)
6519 break;
6520 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6521 key.offset == num_bytes) {
6522 found_extent = 1;
6523 break;
6524 }
6525 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6526 key.offset == owner_objectid) {
6527 found_extent = 1;
6528 break;
6529 }
6530 if (path->slots[0] - extent_slot > 5)
6531 break;
6532 extent_slot--;
6533 }
6534#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6535 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6536 if (found_extent && item_size < sizeof(*ei))
6537 found_extent = 0;
6538#endif
6539 if (!found_extent) {
6540 BUG_ON(iref);
6541 ret = remove_extent_backref(trans, extent_root, path,
6542 NULL, refs_to_drop,
6543 is_data, &last_ref);
6544 if (ret) {
6545 btrfs_abort_transaction(trans, extent_root, ret);
6546 goto out;
6547 }
6548 btrfs_release_path(path);
6549 path->leave_spinning = 1;
6550
6551 key.objectid = bytenr;
6552 key.type = BTRFS_EXTENT_ITEM_KEY;
6553 key.offset = num_bytes;
6554
6555 if (!is_data && skinny_metadata) {
6556 key.type = BTRFS_METADATA_ITEM_KEY;
6557 key.offset = owner_objectid;
6558 }
6559
6560 ret = btrfs_search_slot(trans, extent_root,
6561 &key, path, -1, 1);
6562 if (ret > 0 && skinny_metadata && path->slots[0]) {
6563 /*
6564 * Couldn't find our skinny metadata item,
6565 * see if we have ye olde extent item.
6566 */
6567 path->slots[0]--;
6568 btrfs_item_key_to_cpu(path->nodes[0], &key,
6569 path->slots[0]);
6570 if (key.objectid == bytenr &&
6571 key.type == BTRFS_EXTENT_ITEM_KEY &&
6572 key.offset == num_bytes)
6573 ret = 0;
6574 }
6575
6576 if (ret > 0 && skinny_metadata) {
6577 skinny_metadata = false;
6578 key.objectid = bytenr;
6579 key.type = BTRFS_EXTENT_ITEM_KEY;
6580 key.offset = num_bytes;
6581 btrfs_release_path(path);
6582 ret = btrfs_search_slot(trans, extent_root,
6583 &key, path, -1, 1);
6584 }
6585
6586 if (ret) {
6587 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6588 ret, bytenr);
6589 if (ret > 0)
6590 btrfs_print_leaf(extent_root,
6591 path->nodes[0]);
6592 }
6593 if (ret < 0) {
6594 btrfs_abort_transaction(trans, extent_root, ret);
6595 goto out;
6596 }
6597 extent_slot = path->slots[0];
6598 }
6599 } else if (WARN_ON(ret == -ENOENT)) {
6600 btrfs_print_leaf(extent_root, path->nodes[0]);
6601 btrfs_err(info,
6602 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6603 bytenr, parent, root_objectid, owner_objectid,
6604 owner_offset);
6605 btrfs_abort_transaction(trans, extent_root, ret);
6606 goto out;
6607 } else {
6608 btrfs_abort_transaction(trans, extent_root, ret);
6609 goto out;
6610 }
6611
6612 leaf = path->nodes[0];
6613 item_size = btrfs_item_size_nr(leaf, extent_slot);
6614#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6615 if (item_size < sizeof(*ei)) {
6616 BUG_ON(found_extent || extent_slot != path->slots[0]);
6617 ret = convert_extent_item_v0(trans, extent_root, path,
6618 owner_objectid, 0);
6619 if (ret < 0) {
6620 btrfs_abort_transaction(trans, extent_root, ret);
6621 goto out;
6622 }
6623
6624 btrfs_release_path(path);
6625 path->leave_spinning = 1;
6626
6627 key.objectid = bytenr;
6628 key.type = BTRFS_EXTENT_ITEM_KEY;
6629 key.offset = num_bytes;
6630
6631 ret = btrfs_search_slot(trans, extent_root, &key, path,
6632 -1, 1);
6633 if (ret) {
6634 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6635 ret, bytenr);
6636 btrfs_print_leaf(extent_root, path->nodes[0]);
6637 }
6638 if (ret < 0) {
6639 btrfs_abort_transaction(trans, extent_root, ret);
6640 goto out;
6641 }
6642
6643 extent_slot = path->slots[0];
6644 leaf = path->nodes[0];
6645 item_size = btrfs_item_size_nr(leaf, extent_slot);
6646 }
6647#endif
6648 BUG_ON(item_size < sizeof(*ei));
6649 ei = btrfs_item_ptr(leaf, extent_slot,
6650 struct btrfs_extent_item);
6651 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6652 key.type == BTRFS_EXTENT_ITEM_KEY) {
6653 struct btrfs_tree_block_info *bi;
6654 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6655 bi = (struct btrfs_tree_block_info *)(ei + 1);
6656 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6657 }
6658
6659 refs = btrfs_extent_refs(leaf, ei);
6660 if (refs < refs_to_drop) {
6661 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6662 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6663 ret = -EINVAL;
6664 btrfs_abort_transaction(trans, extent_root, ret);
6665 goto out;
6666 }
6667 refs -= refs_to_drop;
6668
6669 if (refs > 0) {
6670 if (extent_op)
6671 __run_delayed_extent_op(extent_op, leaf, ei);
6672 /*
6673 * In the case of inline back ref, reference count will
6674 * be updated by remove_extent_backref
6675 */
6676 if (iref) {
6677 BUG_ON(!found_extent);
6678 } else {
6679 btrfs_set_extent_refs(leaf, ei, refs);
6680 btrfs_mark_buffer_dirty(leaf);
6681 }
6682 if (found_extent) {
6683 ret = remove_extent_backref(trans, extent_root, path,
6684 iref, refs_to_drop,
6685 is_data, &last_ref);
6686 if (ret) {
6687 btrfs_abort_transaction(trans, extent_root, ret);
6688 goto out;
6689 }
6690 }
6691 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6692 root_objectid);
6693 } else {
6694 if (found_extent) {
6695 BUG_ON(is_data && refs_to_drop !=
6696 extent_data_ref_count(path, iref));
6697 if (iref) {
6698 BUG_ON(path->slots[0] != extent_slot);
6699 } else {
6700 BUG_ON(path->slots[0] != extent_slot + 1);
6701 path->slots[0] = extent_slot;
6702 num_to_del = 2;
6703 }
6704 }
6705
6706 last_ref = 1;
6707 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6708 num_to_del);
6709 if (ret) {
6710 btrfs_abort_transaction(trans, extent_root, ret);
6711 goto out;
6712 }
6713 btrfs_release_path(path);
6714
6715 if (is_data) {
6716 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6717 if (ret) {
6718 btrfs_abort_transaction(trans, extent_root, ret);
6719 goto out;
6720 }
6721 }
6722
6723 ret = add_to_free_space_tree(trans, root->fs_info, bytenr,
6724 num_bytes);
6725 if (ret) {
6726 btrfs_abort_transaction(trans, extent_root, ret);
6727 goto out;
6728 }
6729
6730 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6731 if (ret) {
6732 btrfs_abort_transaction(trans, extent_root, ret);
6733 goto out;
6734 }
6735 }
6736 btrfs_release_path(path);
6737
6738out:
6739 btrfs_free_path(path);
6740 return ret;
6741}
6742
6743/*
6744 * when we free an block, it is possible (and likely) that we free the last
6745 * delayed ref for that extent as well. This searches the delayed ref tree for
6746 * a given extent, and if there are no other delayed refs to be processed, it
6747 * removes it from the tree.
6748 */
6749static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6750 struct btrfs_root *root, u64 bytenr)
6751{
6752 struct btrfs_delayed_ref_head *head;
6753 struct btrfs_delayed_ref_root *delayed_refs;
6754 int ret = 0;
6755
6756 delayed_refs = &trans->transaction->delayed_refs;
6757 spin_lock(&delayed_refs->lock);
6758 head = btrfs_find_delayed_ref_head(trans, bytenr);
6759 if (!head)
6760 goto out_delayed_unlock;
6761
6762 spin_lock(&head->lock);
6763 if (!list_empty(&head->ref_list))
6764 goto out;
6765
6766 if (head->extent_op) {
6767 if (!head->must_insert_reserved)
6768 goto out;
6769 btrfs_free_delayed_extent_op(head->extent_op);
6770 head->extent_op = NULL;
6771 }
6772
6773 /*
6774 * waiting for the lock here would deadlock. If someone else has it
6775 * locked they are already in the process of dropping it anyway
6776 */
6777 if (!mutex_trylock(&head->mutex))
6778 goto out;
6779
6780 /*
6781 * at this point we have a head with no other entries. Go
6782 * ahead and process it.
6783 */
6784 head->node.in_tree = 0;
6785 rb_erase(&head->href_node, &delayed_refs->href_root);
6786
6787 atomic_dec(&delayed_refs->num_entries);
6788
6789 /*
6790 * we don't take a ref on the node because we're removing it from the
6791 * tree, so we just steal the ref the tree was holding.
6792 */
6793 delayed_refs->num_heads--;
6794 if (head->processing == 0)
6795 delayed_refs->num_heads_ready--;
6796 head->processing = 0;
6797 spin_unlock(&head->lock);
6798 spin_unlock(&delayed_refs->lock);
6799
6800 BUG_ON(head->extent_op);
6801 if (head->must_insert_reserved)
6802 ret = 1;
6803
6804 mutex_unlock(&head->mutex);
6805 btrfs_put_delayed_ref(&head->node);
6806 return ret;
6807out:
6808 spin_unlock(&head->lock);
6809
6810out_delayed_unlock:
6811 spin_unlock(&delayed_refs->lock);
6812 return 0;
6813}
6814
6815void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6816 struct btrfs_root *root,
6817 struct extent_buffer *buf,
6818 u64 parent, int last_ref)
6819{
6820 int pin = 1;
6821 int ret;
6822
6823 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6824 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6825 buf->start, buf->len,
6826 parent, root->root_key.objectid,
6827 btrfs_header_level(buf),
6828 BTRFS_DROP_DELAYED_REF, NULL);
6829 BUG_ON(ret); /* -ENOMEM */
6830 }
6831
6832 if (!last_ref)
6833 return;
6834
6835 if (btrfs_header_generation(buf) == trans->transid) {
6836 struct btrfs_block_group_cache *cache;
6837
6838 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6839 ret = check_ref_cleanup(trans, root, buf->start);
6840 if (!ret)
6841 goto out;
6842 }
6843
6844 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6845
6846 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6847 pin_down_extent(root, cache, buf->start, buf->len, 1);
6848 btrfs_put_block_group(cache);
6849 goto out;
6850 }
6851
6852 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6853
6854 btrfs_add_free_space(cache, buf->start, buf->len);
6855 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6856 btrfs_put_block_group(cache);
6857 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6858 pin = 0;
6859 }
6860out:
6861 if (pin)
6862 add_pinned_bytes(root->fs_info, buf->len,
6863 btrfs_header_level(buf),
6864 root->root_key.objectid);
6865
6866 /*
6867 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6868 * anymore.
6869 */
6870 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6871}
6872
6873/* Can return -ENOMEM */
6874int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6875 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6876 u64 owner, u64 offset)
6877{
6878 int ret;
6879 struct btrfs_fs_info *fs_info = root->fs_info;
6880
6881 if (btrfs_test_is_dummy_root(root))
6882 return 0;
6883
6884 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6885
6886 /*
6887 * tree log blocks never actually go into the extent allocation
6888 * tree, just update pinning info and exit early.
6889 */
6890 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6891 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6892 /* unlocks the pinned mutex */
6893 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6894 ret = 0;
6895 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6896 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6897 num_bytes,
6898 parent, root_objectid, (int)owner,
6899 BTRFS_DROP_DELAYED_REF, NULL);
6900 } else {
6901 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6902 num_bytes,
6903 parent, root_objectid, owner,
6904 offset, 0,
6905 BTRFS_DROP_DELAYED_REF, NULL);
6906 }
6907 return ret;
6908}
6909
6910/*
6911 * when we wait for progress in the block group caching, its because
6912 * our allocation attempt failed at least once. So, we must sleep
6913 * and let some progress happen before we try again.
6914 *
6915 * This function will sleep at least once waiting for new free space to
6916 * show up, and then it will check the block group free space numbers
6917 * for our min num_bytes. Another option is to have it go ahead
6918 * and look in the rbtree for a free extent of a given size, but this
6919 * is a good start.
6920 *
6921 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6922 * any of the information in this block group.
6923 */
6924static noinline void
6925wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6926 u64 num_bytes)
6927{
6928 struct btrfs_caching_control *caching_ctl;
6929
6930 caching_ctl = get_caching_control(cache);
6931 if (!caching_ctl)
6932 return;
6933
6934 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6935 (cache->free_space_ctl->free_space >= num_bytes));
6936
6937 put_caching_control(caching_ctl);
6938}
6939
6940static noinline int
6941wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6942{
6943 struct btrfs_caching_control *caching_ctl;
6944 int ret = 0;
6945
6946 caching_ctl = get_caching_control(cache);
6947 if (!caching_ctl)
6948 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6949
6950 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6951 if (cache->cached == BTRFS_CACHE_ERROR)
6952 ret = -EIO;
6953 put_caching_control(caching_ctl);
6954 return ret;
6955}
6956
6957int __get_raid_index(u64 flags)
6958{
6959 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6960 return BTRFS_RAID_RAID10;
6961 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6962 return BTRFS_RAID_RAID1;
6963 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6964 return BTRFS_RAID_DUP;
6965 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6966 return BTRFS_RAID_RAID0;
6967 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6968 return BTRFS_RAID_RAID5;
6969 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6970 return BTRFS_RAID_RAID6;
6971
6972 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6973}
6974
6975int get_block_group_index(struct btrfs_block_group_cache *cache)
6976{
6977 return __get_raid_index(cache->flags);
6978}
6979
6980static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6981 [BTRFS_RAID_RAID10] = "raid10",
6982 [BTRFS_RAID_RAID1] = "raid1",
6983 [BTRFS_RAID_DUP] = "dup",
6984 [BTRFS_RAID_RAID0] = "raid0",
6985 [BTRFS_RAID_SINGLE] = "single",
6986 [BTRFS_RAID_RAID5] = "raid5",
6987 [BTRFS_RAID_RAID6] = "raid6",
6988};
6989
6990static const char *get_raid_name(enum btrfs_raid_types type)
6991{
6992 if (type >= BTRFS_NR_RAID_TYPES)
6993 return NULL;
6994
6995 return btrfs_raid_type_names[type];
6996}
6997
6998enum btrfs_loop_type {
6999 LOOP_CACHING_NOWAIT = 0,
7000 LOOP_CACHING_WAIT = 1,
7001 LOOP_ALLOC_CHUNK = 2,
7002 LOOP_NO_EMPTY_SIZE = 3,
7003};
7004
7005static inline void
7006btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7007 int delalloc)
7008{
7009 if (delalloc)
7010 down_read(&cache->data_rwsem);
7011}
7012
7013static inline void
7014btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7015 int delalloc)
7016{
7017 btrfs_get_block_group(cache);
7018 if (delalloc)
7019 down_read(&cache->data_rwsem);
7020}
7021
7022static struct btrfs_block_group_cache *
7023btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7024 struct btrfs_free_cluster *cluster,
7025 int delalloc)
7026{
7027 struct btrfs_block_group_cache *used_bg = NULL;
7028 bool locked = false;
7029again:
7030 spin_lock(&cluster->refill_lock);
7031 if (locked) {
7032 if (used_bg == cluster->block_group)
7033 return used_bg;
7034
7035 up_read(&used_bg->data_rwsem);
7036 btrfs_put_block_group(used_bg);
7037 }
7038
7039 used_bg = cluster->block_group;
7040 if (!used_bg)
7041 return NULL;
7042
7043 if (used_bg == block_group)
7044 return used_bg;
7045
7046 btrfs_get_block_group(used_bg);
7047
7048 if (!delalloc)
7049 return used_bg;
7050
7051 if (down_read_trylock(&used_bg->data_rwsem))
7052 return used_bg;
7053
7054 spin_unlock(&cluster->refill_lock);
7055 down_read(&used_bg->data_rwsem);
7056 locked = true;
7057 goto again;
7058}
7059
7060static inline void
7061btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7062 int delalloc)
7063{
7064 if (delalloc)
7065 up_read(&cache->data_rwsem);
7066 btrfs_put_block_group(cache);
7067}
7068
7069/*
7070 * walks the btree of allocated extents and find a hole of a given size.
7071 * The key ins is changed to record the hole:
7072 * ins->objectid == start position
7073 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7074 * ins->offset == the size of the hole.
7075 * Any available blocks before search_start are skipped.
7076 *
7077 * If there is no suitable free space, we will record the max size of
7078 * the free space extent currently.
7079 */
7080static noinline int find_free_extent(struct btrfs_root *orig_root,
7081 u64 num_bytes, u64 empty_size,
7082 u64 hint_byte, struct btrfs_key *ins,
7083 u64 flags, int delalloc)
7084{
7085 int ret = 0;
7086 struct btrfs_root *root = orig_root->fs_info->extent_root;
7087 struct btrfs_free_cluster *last_ptr = NULL;
7088 struct btrfs_block_group_cache *block_group = NULL;
7089 u64 search_start = 0;
7090 u64 max_extent_size = 0;
7091 u64 empty_cluster = 0;
7092 struct btrfs_space_info *space_info;
7093 int loop = 0;
7094 int index = __get_raid_index(flags);
7095 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
7096 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
7097 bool failed_cluster_refill = false;
7098 bool failed_alloc = false;
7099 bool use_cluster = true;
7100 bool have_caching_bg = false;
7101 bool orig_have_caching_bg = false;
7102 bool full_search = false;
7103
7104 WARN_ON(num_bytes < root->sectorsize);
7105 ins->type = BTRFS_EXTENT_ITEM_KEY;
7106 ins->objectid = 0;
7107 ins->offset = 0;
7108
7109 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
7110
7111 space_info = __find_space_info(root->fs_info, flags);
7112 if (!space_info) {
7113 btrfs_err(root->fs_info, "No space info for %llu", flags);
7114 return -ENOSPC;
7115 }
7116
7117 /*
7118 * If our free space is heavily fragmented we may not be able to make
7119 * big contiguous allocations, so instead of doing the expensive search
7120 * for free space, simply return ENOSPC with our max_extent_size so we
7121 * can go ahead and search for a more manageable chunk.
7122 *
7123 * If our max_extent_size is large enough for our allocation simply
7124 * disable clustering since we will likely not be able to find enough
7125 * space to create a cluster and induce latency trying.
7126 */
7127 if (unlikely(space_info->max_extent_size)) {
7128 spin_lock(&space_info->lock);
7129 if (space_info->max_extent_size &&
7130 num_bytes > space_info->max_extent_size) {
7131 ins->offset = space_info->max_extent_size;
7132 spin_unlock(&space_info->lock);
7133 return -ENOSPC;
7134 } else if (space_info->max_extent_size) {
7135 use_cluster = false;
7136 }
7137 spin_unlock(&space_info->lock);
7138 }
7139
7140 last_ptr = fetch_cluster_info(orig_root, space_info, &empty_cluster);
7141 if (last_ptr) {
7142 spin_lock(&last_ptr->lock);
7143 if (last_ptr->block_group)
7144 hint_byte = last_ptr->window_start;
7145 if (last_ptr->fragmented) {
7146 /*
7147 * We still set window_start so we can keep track of the
7148 * last place we found an allocation to try and save
7149 * some time.
7150 */
7151 hint_byte = last_ptr->window_start;
7152 use_cluster = false;
7153 }
7154 spin_unlock(&last_ptr->lock);
7155 }
7156
7157 search_start = max(search_start, first_logical_byte(root, 0));
7158 search_start = max(search_start, hint_byte);
7159 if (search_start == hint_byte) {
7160 block_group = btrfs_lookup_block_group(root->fs_info,
7161 search_start);
7162 /*
7163 * we don't want to use the block group if it doesn't match our
7164 * allocation bits, or if its not cached.
7165 *
7166 * However if we are re-searching with an ideal block group
7167 * picked out then we don't care that the block group is cached.
7168 */
7169 if (block_group && block_group_bits(block_group, flags) &&
7170 block_group->cached != BTRFS_CACHE_NO) {
7171 down_read(&space_info->groups_sem);
7172 if (list_empty(&block_group->list) ||
7173 block_group->ro) {
7174 /*
7175 * someone is removing this block group,
7176 * we can't jump into the have_block_group
7177 * target because our list pointers are not
7178 * valid
7179 */
7180 btrfs_put_block_group(block_group);
7181 up_read(&space_info->groups_sem);
7182 } else {
7183 index = get_block_group_index(block_group);
7184 btrfs_lock_block_group(block_group, delalloc);
7185 goto have_block_group;
7186 }
7187 } else if (block_group) {
7188 btrfs_put_block_group(block_group);
7189 }
7190 }
7191search:
7192 have_caching_bg = false;
7193 if (index == 0 || index == __get_raid_index(flags))
7194 full_search = true;
7195 down_read(&space_info->groups_sem);
7196 list_for_each_entry(block_group, &space_info->block_groups[index],
7197 list) {
7198 u64 offset;
7199 int cached;
7200
7201 btrfs_grab_block_group(block_group, delalloc);
7202 search_start = block_group->key.objectid;
7203
7204 /*
7205 * this can happen if we end up cycling through all the
7206 * raid types, but we want to make sure we only allocate
7207 * for the proper type.
7208 */
7209 if (!block_group_bits(block_group, flags)) {
7210 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7211 BTRFS_BLOCK_GROUP_RAID1 |
7212 BTRFS_BLOCK_GROUP_RAID5 |
7213 BTRFS_BLOCK_GROUP_RAID6 |
7214 BTRFS_BLOCK_GROUP_RAID10;
7215
7216 /*
7217 * if they asked for extra copies and this block group
7218 * doesn't provide them, bail. This does allow us to
7219 * fill raid0 from raid1.
7220 */
7221 if ((flags & extra) && !(block_group->flags & extra))
7222 goto loop;
7223 }
7224
7225have_block_group:
7226 cached = block_group_cache_done(block_group);
7227 if (unlikely(!cached)) {
7228 have_caching_bg = true;
7229 ret = cache_block_group(block_group, 0);
7230 BUG_ON(ret < 0);
7231 ret = 0;
7232 }
7233
7234 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7235 goto loop;
7236 if (unlikely(block_group->ro))
7237 goto loop;
7238
7239 /*
7240 * Ok we want to try and use the cluster allocator, so
7241 * lets look there
7242 */
7243 if (last_ptr && use_cluster) {
7244 struct btrfs_block_group_cache *used_block_group;
7245 unsigned long aligned_cluster;
7246 /*
7247 * the refill lock keeps out other
7248 * people trying to start a new cluster
7249 */
7250 used_block_group = btrfs_lock_cluster(block_group,
7251 last_ptr,
7252 delalloc);
7253 if (!used_block_group)
7254 goto refill_cluster;
7255
7256 if (used_block_group != block_group &&
7257 (used_block_group->ro ||
7258 !block_group_bits(used_block_group, flags)))
7259 goto release_cluster;
7260
7261 offset = btrfs_alloc_from_cluster(used_block_group,
7262 last_ptr,
7263 num_bytes,
7264 used_block_group->key.objectid,
7265 &max_extent_size);
7266 if (offset) {
7267 /* we have a block, we're done */
7268 spin_unlock(&last_ptr->refill_lock);
7269 trace_btrfs_reserve_extent_cluster(root,
7270 used_block_group,
7271 search_start, num_bytes);
7272 if (used_block_group != block_group) {
7273 btrfs_release_block_group(block_group,
7274 delalloc);
7275 block_group = used_block_group;
7276 }
7277 goto checks;
7278 }
7279
7280 WARN_ON(last_ptr->block_group != used_block_group);
7281release_cluster:
7282 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7283 * set up a new clusters, so lets just skip it
7284 * and let the allocator find whatever block
7285 * it can find. If we reach this point, we
7286 * will have tried the cluster allocator
7287 * plenty of times and not have found
7288 * anything, so we are likely way too
7289 * fragmented for the clustering stuff to find
7290 * anything.
7291 *
7292 * However, if the cluster is taken from the
7293 * current block group, release the cluster
7294 * first, so that we stand a better chance of
7295 * succeeding in the unclustered
7296 * allocation. */
7297 if (loop >= LOOP_NO_EMPTY_SIZE &&
7298 used_block_group != block_group) {
7299 spin_unlock(&last_ptr->refill_lock);
7300 btrfs_release_block_group(used_block_group,
7301 delalloc);
7302 goto unclustered_alloc;
7303 }
7304
7305 /*
7306 * this cluster didn't work out, free it and
7307 * start over
7308 */
7309 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7310
7311 if (used_block_group != block_group)
7312 btrfs_release_block_group(used_block_group,
7313 delalloc);
7314refill_cluster:
7315 if (loop >= LOOP_NO_EMPTY_SIZE) {
7316 spin_unlock(&last_ptr->refill_lock);
7317 goto unclustered_alloc;
7318 }
7319
7320 aligned_cluster = max_t(unsigned long,
7321 empty_cluster + empty_size,
7322 block_group->full_stripe_len);
7323
7324 /* allocate a cluster in this block group */
7325 ret = btrfs_find_space_cluster(root, block_group,
7326 last_ptr, search_start,
7327 num_bytes,
7328 aligned_cluster);
7329 if (ret == 0) {
7330 /*
7331 * now pull our allocation out of this
7332 * cluster
7333 */
7334 offset = btrfs_alloc_from_cluster(block_group,
7335 last_ptr,
7336 num_bytes,
7337 search_start,
7338 &max_extent_size);
7339 if (offset) {
7340 /* we found one, proceed */
7341 spin_unlock(&last_ptr->refill_lock);
7342 trace_btrfs_reserve_extent_cluster(root,
7343 block_group, search_start,
7344 num_bytes);
7345 goto checks;
7346 }
7347 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7348 && !failed_cluster_refill) {
7349 spin_unlock(&last_ptr->refill_lock);
7350
7351 failed_cluster_refill = true;
7352 wait_block_group_cache_progress(block_group,
7353 num_bytes + empty_cluster + empty_size);
7354 goto have_block_group;
7355 }
7356
7357 /*
7358 * at this point we either didn't find a cluster
7359 * or we weren't able to allocate a block from our
7360 * cluster. Free the cluster we've been trying
7361 * to use, and go to the next block group
7362 */
7363 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7364 spin_unlock(&last_ptr->refill_lock);
7365 goto loop;
7366 }
7367
7368unclustered_alloc:
7369 /*
7370 * We are doing an unclustered alloc, set the fragmented flag so
7371 * we don't bother trying to setup a cluster again until we get
7372 * more space.
7373 */
7374 if (unlikely(last_ptr)) {
7375 spin_lock(&last_ptr->lock);
7376 last_ptr->fragmented = 1;
7377 spin_unlock(&last_ptr->lock);
7378 }
7379 spin_lock(&block_group->free_space_ctl->tree_lock);
7380 if (cached &&
7381 block_group->free_space_ctl->free_space <
7382 num_bytes + empty_cluster + empty_size) {
7383 if (block_group->free_space_ctl->free_space >
7384 max_extent_size)
7385 max_extent_size =
7386 block_group->free_space_ctl->free_space;
7387 spin_unlock(&block_group->free_space_ctl->tree_lock);
7388 goto loop;
7389 }
7390 spin_unlock(&block_group->free_space_ctl->tree_lock);
7391
7392 offset = btrfs_find_space_for_alloc(block_group, search_start,
7393 num_bytes, empty_size,
7394 &max_extent_size);
7395 /*
7396 * If we didn't find a chunk, and we haven't failed on this
7397 * block group before, and this block group is in the middle of
7398 * caching and we are ok with waiting, then go ahead and wait
7399 * for progress to be made, and set failed_alloc to true.
7400 *
7401 * If failed_alloc is true then we've already waited on this
7402 * block group once and should move on to the next block group.
7403 */
7404 if (!offset && !failed_alloc && !cached &&
7405 loop > LOOP_CACHING_NOWAIT) {
7406 wait_block_group_cache_progress(block_group,
7407 num_bytes + empty_size);
7408 failed_alloc = true;
7409 goto have_block_group;
7410 } else if (!offset) {
7411 goto loop;
7412 }
7413checks:
7414 search_start = ALIGN(offset, root->stripesize);
7415
7416 /* move on to the next group */
7417 if (search_start + num_bytes >
7418 block_group->key.objectid + block_group->key.offset) {
7419 btrfs_add_free_space(block_group, offset, num_bytes);
7420 goto loop;
7421 }
7422
7423 if (offset < search_start)
7424 btrfs_add_free_space(block_group, offset,
7425 search_start - offset);
7426 BUG_ON(offset > search_start);
7427
7428 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7429 alloc_type, delalloc);
7430 if (ret == -EAGAIN) {
7431 btrfs_add_free_space(block_group, offset, num_bytes);
7432 goto loop;
7433 }
7434
7435 /* we are all good, lets return */
7436 ins->objectid = search_start;
7437 ins->offset = num_bytes;
7438
7439 trace_btrfs_reserve_extent(orig_root, block_group,
7440 search_start, num_bytes);
7441 btrfs_release_block_group(block_group, delalloc);
7442 break;
7443loop:
7444 failed_cluster_refill = false;
7445 failed_alloc = false;
7446 BUG_ON(index != get_block_group_index(block_group));
7447 btrfs_release_block_group(block_group, delalloc);
7448 }
7449 up_read(&space_info->groups_sem);
7450
7451 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7452 && !orig_have_caching_bg)
7453 orig_have_caching_bg = true;
7454
7455 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7456 goto search;
7457
7458 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7459 goto search;
7460
7461 /*
7462 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7463 * caching kthreads as we move along
7464 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7465 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7466 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7467 * again
7468 */
7469 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7470 index = 0;
7471 if (loop == LOOP_CACHING_NOWAIT) {
7472 /*
7473 * We want to skip the LOOP_CACHING_WAIT step if we
7474 * don't have any unached bgs and we've alrelady done a
7475 * full search through.
7476 */
7477 if (orig_have_caching_bg || !full_search)
7478 loop = LOOP_CACHING_WAIT;
7479 else
7480 loop = LOOP_ALLOC_CHUNK;
7481 } else {
7482 loop++;
7483 }
7484
7485 if (loop == LOOP_ALLOC_CHUNK) {
7486 struct btrfs_trans_handle *trans;
7487 int exist = 0;
7488
7489 trans = current->journal_info;
7490 if (trans)
7491 exist = 1;
7492 else
7493 trans = btrfs_join_transaction(root);
7494
7495 if (IS_ERR(trans)) {
7496 ret = PTR_ERR(trans);
7497 goto out;
7498 }
7499
7500 ret = do_chunk_alloc(trans, root, flags,
7501 CHUNK_ALLOC_FORCE);
7502
7503 /*
7504 * If we can't allocate a new chunk we've already looped
7505 * through at least once, move on to the NO_EMPTY_SIZE
7506 * case.
7507 */
7508 if (ret == -ENOSPC)
7509 loop = LOOP_NO_EMPTY_SIZE;
7510
7511 /*
7512 * Do not bail out on ENOSPC since we
7513 * can do more things.
7514 */
7515 if (ret < 0 && ret != -ENOSPC)
7516 btrfs_abort_transaction(trans,
7517 root, ret);
7518 else
7519 ret = 0;
7520 if (!exist)
7521 btrfs_end_transaction(trans, root);
7522 if (ret)
7523 goto out;
7524 }
7525
7526 if (loop == LOOP_NO_EMPTY_SIZE) {
7527 /*
7528 * Don't loop again if we already have no empty_size and
7529 * no empty_cluster.
7530 */
7531 if (empty_size == 0 &&
7532 empty_cluster == 0) {
7533 ret = -ENOSPC;
7534 goto out;
7535 }
7536 empty_size = 0;
7537 empty_cluster = 0;
7538 }
7539
7540 goto search;
7541 } else if (!ins->objectid) {
7542 ret = -ENOSPC;
7543 } else if (ins->objectid) {
7544 if (!use_cluster && last_ptr) {
7545 spin_lock(&last_ptr->lock);
7546 last_ptr->window_start = ins->objectid;
7547 spin_unlock(&last_ptr->lock);
7548 }
7549 ret = 0;
7550 }
7551out:
7552 if (ret == -ENOSPC) {
7553 spin_lock(&space_info->lock);
7554 space_info->max_extent_size = max_extent_size;
7555 spin_unlock(&space_info->lock);
7556 ins->offset = max_extent_size;
7557 }
7558 return ret;
7559}
7560
7561static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7562 int dump_block_groups)
7563{
7564 struct btrfs_block_group_cache *cache;
7565 int index = 0;
7566
7567 spin_lock(&info->lock);
7568 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7569 info->flags,
7570 info->total_bytes - info->bytes_used - info->bytes_pinned -
7571 info->bytes_reserved - info->bytes_readonly,
7572 (info->full) ? "" : "not ");
7573 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7574 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7575 info->total_bytes, info->bytes_used, info->bytes_pinned,
7576 info->bytes_reserved, info->bytes_may_use,
7577 info->bytes_readonly);
7578 spin_unlock(&info->lock);
7579
7580 if (!dump_block_groups)
7581 return;
7582
7583 down_read(&info->groups_sem);
7584again:
7585 list_for_each_entry(cache, &info->block_groups[index], list) {
7586 spin_lock(&cache->lock);
7587 printk(KERN_INFO "BTRFS: "
7588 "block group %llu has %llu bytes, "
7589 "%llu used %llu pinned %llu reserved %s\n",
7590 cache->key.objectid, cache->key.offset,
7591 btrfs_block_group_used(&cache->item), cache->pinned,
7592 cache->reserved, cache->ro ? "[readonly]" : "");
7593 btrfs_dump_free_space(cache, bytes);
7594 spin_unlock(&cache->lock);
7595 }
7596 if (++index < BTRFS_NR_RAID_TYPES)
7597 goto again;
7598 up_read(&info->groups_sem);
7599}
7600
7601int btrfs_reserve_extent(struct btrfs_root *root,
7602 u64 num_bytes, u64 min_alloc_size,
7603 u64 empty_size, u64 hint_byte,
7604 struct btrfs_key *ins, int is_data, int delalloc)
7605{
7606 bool final_tried = num_bytes == min_alloc_size;
7607 u64 flags;
7608 int ret;
7609
7610 flags = btrfs_get_alloc_profile(root, is_data);
7611again:
7612 WARN_ON(num_bytes < root->sectorsize);
7613 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7614 flags, delalloc);
7615
7616 if (ret == -ENOSPC) {
7617 if (!final_tried && ins->offset) {
7618 num_bytes = min(num_bytes >> 1, ins->offset);
7619 num_bytes = round_down(num_bytes, root->sectorsize);
7620 num_bytes = max(num_bytes, min_alloc_size);
7621 if (num_bytes == min_alloc_size)
7622 final_tried = true;
7623 goto again;
7624 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7625 struct btrfs_space_info *sinfo;
7626
7627 sinfo = __find_space_info(root->fs_info, flags);
7628 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7629 flags, num_bytes);
7630 if (sinfo)
7631 dump_space_info(sinfo, num_bytes, 1);
7632 }
7633 }
7634
7635 return ret;
7636}
7637
7638static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7639 u64 start, u64 len,
7640 int pin, int delalloc)
7641{
7642 struct btrfs_block_group_cache *cache;
7643 int ret = 0;
7644
7645 cache = btrfs_lookup_block_group(root->fs_info, start);
7646 if (!cache) {
7647 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7648 start);
7649 return -ENOSPC;
7650 }
7651
7652 if (pin)
7653 pin_down_extent(root, cache, start, len, 1);
7654 else {
7655 if (btrfs_test_opt(root, DISCARD))
7656 ret = btrfs_discard_extent(root, start, len, NULL);
7657 btrfs_add_free_space(cache, start, len);
7658 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7659 }
7660
7661 btrfs_put_block_group(cache);
7662
7663 trace_btrfs_reserved_extent_free(root, start, len);
7664
7665 return ret;
7666}
7667
7668int btrfs_free_reserved_extent(struct btrfs_root *root,
7669 u64 start, u64 len, int delalloc)
7670{
7671 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7672}
7673
7674int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7675 u64 start, u64 len)
7676{
7677 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7678}
7679
7680static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7681 struct btrfs_root *root,
7682 u64 parent, u64 root_objectid,
7683 u64 flags, u64 owner, u64 offset,
7684 struct btrfs_key *ins, int ref_mod)
7685{
7686 int ret;
7687 struct btrfs_fs_info *fs_info = root->fs_info;
7688 struct btrfs_extent_item *extent_item;
7689 struct btrfs_extent_inline_ref *iref;
7690 struct btrfs_path *path;
7691 struct extent_buffer *leaf;
7692 int type;
7693 u32 size;
7694
7695 if (parent > 0)
7696 type = BTRFS_SHARED_DATA_REF_KEY;
7697 else
7698 type = BTRFS_EXTENT_DATA_REF_KEY;
7699
7700 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7701
7702 path = btrfs_alloc_path();
7703 if (!path)
7704 return -ENOMEM;
7705
7706 path->leave_spinning = 1;
7707 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7708 ins, size);
7709 if (ret) {
7710 btrfs_free_path(path);
7711 return ret;
7712 }
7713
7714 leaf = path->nodes[0];
7715 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7716 struct btrfs_extent_item);
7717 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7718 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7719 btrfs_set_extent_flags(leaf, extent_item,
7720 flags | BTRFS_EXTENT_FLAG_DATA);
7721
7722 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7723 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7724 if (parent > 0) {
7725 struct btrfs_shared_data_ref *ref;
7726 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7727 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7728 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7729 } else {
7730 struct btrfs_extent_data_ref *ref;
7731 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7732 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7733 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7734 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7735 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7736 }
7737
7738 btrfs_mark_buffer_dirty(path->nodes[0]);
7739 btrfs_free_path(path);
7740
7741 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
7742 ins->offset);
7743 if (ret)
7744 return ret;
7745
7746 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7747 if (ret) { /* -ENOENT, logic error */
7748 btrfs_err(fs_info, "update block group failed for %llu %llu",
7749 ins->objectid, ins->offset);
7750 BUG();
7751 }
7752 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7753 return ret;
7754}
7755
7756static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7757 struct btrfs_root *root,
7758 u64 parent, u64 root_objectid,
7759 u64 flags, struct btrfs_disk_key *key,
7760 int level, struct btrfs_key *ins)
7761{
7762 int ret;
7763 struct btrfs_fs_info *fs_info = root->fs_info;
7764 struct btrfs_extent_item *extent_item;
7765 struct btrfs_tree_block_info *block_info;
7766 struct btrfs_extent_inline_ref *iref;
7767 struct btrfs_path *path;
7768 struct extent_buffer *leaf;
7769 u32 size = sizeof(*extent_item) + sizeof(*iref);
7770 u64 num_bytes = ins->offset;
7771 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7772 SKINNY_METADATA);
7773
7774 if (!skinny_metadata)
7775 size += sizeof(*block_info);
7776
7777 path = btrfs_alloc_path();
7778 if (!path) {
7779 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7780 root->nodesize);
7781 return -ENOMEM;
7782 }
7783
7784 path->leave_spinning = 1;
7785 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7786 ins, size);
7787 if (ret) {
7788 btrfs_free_path(path);
7789 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7790 root->nodesize);
7791 return ret;
7792 }
7793
7794 leaf = path->nodes[0];
7795 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7796 struct btrfs_extent_item);
7797 btrfs_set_extent_refs(leaf, extent_item, 1);
7798 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7799 btrfs_set_extent_flags(leaf, extent_item,
7800 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7801
7802 if (skinny_metadata) {
7803 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7804 num_bytes = root->nodesize;
7805 } else {
7806 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7807 btrfs_set_tree_block_key(leaf, block_info, key);
7808 btrfs_set_tree_block_level(leaf, block_info, level);
7809 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7810 }
7811
7812 if (parent > 0) {
7813 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7814 btrfs_set_extent_inline_ref_type(leaf, iref,
7815 BTRFS_SHARED_BLOCK_REF_KEY);
7816 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7817 } else {
7818 btrfs_set_extent_inline_ref_type(leaf, iref,
7819 BTRFS_TREE_BLOCK_REF_KEY);
7820 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7821 }
7822
7823 btrfs_mark_buffer_dirty(leaf);
7824 btrfs_free_path(path);
7825
7826 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
7827 num_bytes);
7828 if (ret)
7829 return ret;
7830
7831 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7832 1);
7833 if (ret) { /* -ENOENT, logic error */
7834 btrfs_err(fs_info, "update block group failed for %llu %llu",
7835 ins->objectid, ins->offset);
7836 BUG();
7837 }
7838
7839 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7840 return ret;
7841}
7842
7843int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7844 struct btrfs_root *root,
7845 u64 root_objectid, u64 owner,
7846 u64 offset, u64 ram_bytes,
7847 struct btrfs_key *ins)
7848{
7849 int ret;
7850
7851 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7852
7853 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7854 ins->offset, 0,
7855 root_objectid, owner, offset,
7856 ram_bytes, BTRFS_ADD_DELAYED_EXTENT,
7857 NULL);
7858 return ret;
7859}
7860
7861/*
7862 * this is used by the tree logging recovery code. It records that
7863 * an extent has been allocated and makes sure to clear the free
7864 * space cache bits as well
7865 */
7866int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7867 struct btrfs_root *root,
7868 u64 root_objectid, u64 owner, u64 offset,
7869 struct btrfs_key *ins)
7870{
7871 int ret;
7872 struct btrfs_block_group_cache *block_group;
7873
7874 /*
7875 * Mixed block groups will exclude before processing the log so we only
7876 * need to do the exlude dance if this fs isn't mixed.
7877 */
7878 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7879 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7880 if (ret)
7881 return ret;
7882 }
7883
7884 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7885 if (!block_group)
7886 return -EINVAL;
7887
7888 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7889 RESERVE_ALLOC_NO_ACCOUNT, 0);
7890 BUG_ON(ret); /* logic error */
7891 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7892 0, owner, offset, ins, 1);
7893 btrfs_put_block_group(block_group);
7894 return ret;
7895}
7896
7897static struct extent_buffer *
7898btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7899 u64 bytenr, int level)
7900{
7901 struct extent_buffer *buf;
7902
7903 buf = btrfs_find_create_tree_block(root, bytenr);
7904 if (!buf)
7905 return ERR_PTR(-ENOMEM);
7906 btrfs_set_header_generation(buf, trans->transid);
7907 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7908 btrfs_tree_lock(buf);
7909 clean_tree_block(trans, root->fs_info, buf);
7910 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7911
7912 btrfs_set_lock_blocking(buf);
7913 set_extent_buffer_uptodate(buf);
7914
7915 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7916 buf->log_index = root->log_transid % 2;
7917 /*
7918 * we allow two log transactions at a time, use different
7919 * EXENT bit to differentiate dirty pages.
7920 */
7921 if (buf->log_index == 0)
7922 set_extent_dirty(&root->dirty_log_pages, buf->start,
7923 buf->start + buf->len - 1, GFP_NOFS);
7924 else
7925 set_extent_new(&root->dirty_log_pages, buf->start,
7926 buf->start + buf->len - 1, GFP_NOFS);
7927 } else {
7928 buf->log_index = -1;
7929 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7930 buf->start + buf->len - 1, GFP_NOFS);
7931 }
7932 trans->blocks_used++;
7933 /* this returns a buffer locked for blocking */
7934 return buf;
7935}
7936
7937static struct btrfs_block_rsv *
7938use_block_rsv(struct btrfs_trans_handle *trans,
7939 struct btrfs_root *root, u32 blocksize)
7940{
7941 struct btrfs_block_rsv *block_rsv;
7942 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7943 int ret;
7944 bool global_updated = false;
7945
7946 block_rsv = get_block_rsv(trans, root);
7947
7948 if (unlikely(block_rsv->size == 0))
7949 goto try_reserve;
7950again:
7951 ret = block_rsv_use_bytes(block_rsv, blocksize);
7952 if (!ret)
7953 return block_rsv;
7954
7955 if (block_rsv->failfast)
7956 return ERR_PTR(ret);
7957
7958 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7959 global_updated = true;
7960 update_global_block_rsv(root->fs_info);
7961 goto again;
7962 }
7963
7964 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7965 static DEFINE_RATELIMIT_STATE(_rs,
7966 DEFAULT_RATELIMIT_INTERVAL * 10,
7967 /*DEFAULT_RATELIMIT_BURST*/ 1);
7968 if (__ratelimit(&_rs))
7969 WARN(1, KERN_DEBUG
7970 "BTRFS: block rsv returned %d\n", ret);
7971 }
7972try_reserve:
7973 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7974 BTRFS_RESERVE_NO_FLUSH);
7975 if (!ret)
7976 return block_rsv;
7977 /*
7978 * If we couldn't reserve metadata bytes try and use some from
7979 * the global reserve if its space type is the same as the global
7980 * reservation.
7981 */
7982 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7983 block_rsv->space_info == global_rsv->space_info) {
7984 ret = block_rsv_use_bytes(global_rsv, blocksize);
7985 if (!ret)
7986 return global_rsv;
7987 }
7988 return ERR_PTR(ret);
7989}
7990
7991static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7992 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7993{
7994 block_rsv_add_bytes(block_rsv, blocksize, 0);
7995 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7996}
7997
7998/*
7999 * finds a free extent and does all the dirty work required for allocation
8000 * returns the tree buffer or an ERR_PTR on error.
8001 */
8002struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8003 struct btrfs_root *root,
8004 u64 parent, u64 root_objectid,
8005 struct btrfs_disk_key *key, int level,
8006 u64 hint, u64 empty_size)
8007{
8008 struct btrfs_key ins;
8009 struct btrfs_block_rsv *block_rsv;
8010 struct extent_buffer *buf;
8011 struct btrfs_delayed_extent_op *extent_op;
8012 u64 flags = 0;
8013 int ret;
8014 u32 blocksize = root->nodesize;
8015 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
8016 SKINNY_METADATA);
8017
8018 if (btrfs_test_is_dummy_root(root)) {
8019 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8020 level);
8021 if (!IS_ERR(buf))
8022 root->alloc_bytenr += blocksize;
8023 return buf;
8024 }
8025
8026 block_rsv = use_block_rsv(trans, root, blocksize);
8027 if (IS_ERR(block_rsv))
8028 return ERR_CAST(block_rsv);
8029
8030 ret = btrfs_reserve_extent(root, blocksize, blocksize,
8031 empty_size, hint, &ins, 0, 0);
8032 if (ret)
8033 goto out_unuse;
8034
8035 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8036 if (IS_ERR(buf)) {
8037 ret = PTR_ERR(buf);
8038 goto out_free_reserved;
8039 }
8040
8041 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8042 if (parent == 0)
8043 parent = ins.objectid;
8044 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8045 } else
8046 BUG_ON(parent > 0);
8047
8048 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8049 extent_op = btrfs_alloc_delayed_extent_op();
8050 if (!extent_op) {
8051 ret = -ENOMEM;
8052 goto out_free_buf;
8053 }
8054 if (key)
8055 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8056 else
8057 memset(&extent_op->key, 0, sizeof(extent_op->key));
8058 extent_op->flags_to_set = flags;
8059 extent_op->update_key = skinny_metadata ? false : true;
8060 extent_op->update_flags = true;
8061 extent_op->is_data = false;
8062 extent_op->level = level;
8063
8064 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
8065 ins.objectid, ins.offset,
8066 parent, root_objectid, level,
8067 BTRFS_ADD_DELAYED_EXTENT,
8068 extent_op);
8069 if (ret)
8070 goto out_free_delayed;
8071 }
8072 return buf;
8073
8074out_free_delayed:
8075 btrfs_free_delayed_extent_op(extent_op);
8076out_free_buf:
8077 free_extent_buffer(buf);
8078out_free_reserved:
8079 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
8080out_unuse:
8081 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
8082 return ERR_PTR(ret);
8083}
8084
8085struct walk_control {
8086 u64 refs[BTRFS_MAX_LEVEL];
8087 u64 flags[BTRFS_MAX_LEVEL];
8088 struct btrfs_key update_progress;
8089 int stage;
8090 int level;
8091 int shared_level;
8092 int update_ref;
8093 int keep_locks;
8094 int reada_slot;
8095 int reada_count;
8096 int for_reloc;
8097};
8098
8099#define DROP_REFERENCE 1
8100#define UPDATE_BACKREF 2
8101
8102static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8103 struct btrfs_root *root,
8104 struct walk_control *wc,
8105 struct btrfs_path *path)
8106{
8107 u64 bytenr;
8108 u64 generation;
8109 u64 refs;
8110 u64 flags;
8111 u32 nritems;
8112 u32 blocksize;
8113 struct btrfs_key key;
8114 struct extent_buffer *eb;
8115 int ret;
8116 int slot;
8117 int nread = 0;
8118
8119 if (path->slots[wc->level] < wc->reada_slot) {
8120 wc->reada_count = wc->reada_count * 2 / 3;
8121 wc->reada_count = max(wc->reada_count, 2);
8122 } else {
8123 wc->reada_count = wc->reada_count * 3 / 2;
8124 wc->reada_count = min_t(int, wc->reada_count,
8125 BTRFS_NODEPTRS_PER_BLOCK(root));
8126 }
8127
8128 eb = path->nodes[wc->level];
8129 nritems = btrfs_header_nritems(eb);
8130 blocksize = root->nodesize;
8131
8132 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8133 if (nread >= wc->reada_count)
8134 break;
8135
8136 cond_resched();
8137 bytenr = btrfs_node_blockptr(eb, slot);
8138 generation = btrfs_node_ptr_generation(eb, slot);
8139
8140 if (slot == path->slots[wc->level])
8141 goto reada;
8142
8143 if (wc->stage == UPDATE_BACKREF &&
8144 generation <= root->root_key.offset)
8145 continue;
8146
8147 /* We don't lock the tree block, it's OK to be racy here */
8148 ret = btrfs_lookup_extent_info(trans, root, bytenr,
8149 wc->level - 1, 1, &refs,
8150 &flags);
8151 /* We don't care about errors in readahead. */
8152 if (ret < 0)
8153 continue;
8154 BUG_ON(refs == 0);
8155
8156 if (wc->stage == DROP_REFERENCE) {
8157 if (refs == 1)
8158 goto reada;
8159
8160 if (wc->level == 1 &&
8161 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8162 continue;
8163 if (!wc->update_ref ||
8164 generation <= root->root_key.offset)
8165 continue;
8166 btrfs_node_key_to_cpu(eb, &key, slot);
8167 ret = btrfs_comp_cpu_keys(&key,
8168 &wc->update_progress);
8169 if (ret < 0)
8170 continue;
8171 } else {
8172 if (wc->level == 1 &&
8173 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8174 continue;
8175 }
8176reada:
8177 readahead_tree_block(root, bytenr);
8178 nread++;
8179 }
8180 wc->reada_slot = slot;
8181}
8182
8183/*
8184 * These may not be seen by the usual inc/dec ref code so we have to
8185 * add them here.
8186 */
8187static int record_one_subtree_extent(struct btrfs_trans_handle *trans,
8188 struct btrfs_root *root, u64 bytenr,
8189 u64 num_bytes)
8190{
8191 struct btrfs_qgroup_extent_record *qrecord;
8192 struct btrfs_delayed_ref_root *delayed_refs;
8193
8194 qrecord = kmalloc(sizeof(*qrecord), GFP_NOFS);
8195 if (!qrecord)
8196 return -ENOMEM;
8197
8198 qrecord->bytenr = bytenr;
8199 qrecord->num_bytes = num_bytes;
8200 qrecord->old_roots = NULL;
8201
8202 delayed_refs = &trans->transaction->delayed_refs;
8203 spin_lock(&delayed_refs->lock);
8204 if (btrfs_qgroup_insert_dirty_extent(delayed_refs, qrecord))
8205 kfree(qrecord);
8206 spin_unlock(&delayed_refs->lock);
8207
8208 return 0;
8209}
8210
8211static int account_leaf_items(struct btrfs_trans_handle *trans,
8212 struct btrfs_root *root,
8213 struct extent_buffer *eb)
8214{
8215 int nr = btrfs_header_nritems(eb);
8216 int i, extent_type, ret;
8217 struct btrfs_key key;
8218 struct btrfs_file_extent_item *fi;
8219 u64 bytenr, num_bytes;
8220
8221 /* We can be called directly from walk_up_proc() */
8222 if (!root->fs_info->quota_enabled)
8223 return 0;
8224
8225 for (i = 0; i < nr; i++) {
8226 btrfs_item_key_to_cpu(eb, &key, i);
8227
8228 if (key.type != BTRFS_EXTENT_DATA_KEY)
8229 continue;
8230
8231 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
8232 /* filter out non qgroup-accountable extents */
8233 extent_type = btrfs_file_extent_type(eb, fi);
8234
8235 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
8236 continue;
8237
8238 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
8239 if (!bytenr)
8240 continue;
8241
8242 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
8243
8244 ret = record_one_subtree_extent(trans, root, bytenr, num_bytes);
8245 if (ret)
8246 return ret;
8247 }
8248 return 0;
8249}
8250
8251/*
8252 * Walk up the tree from the bottom, freeing leaves and any interior
8253 * nodes which have had all slots visited. If a node (leaf or
8254 * interior) is freed, the node above it will have it's slot
8255 * incremented. The root node will never be freed.
8256 *
8257 * At the end of this function, we should have a path which has all
8258 * slots incremented to the next position for a search. If we need to
8259 * read a new node it will be NULL and the node above it will have the
8260 * correct slot selected for a later read.
8261 *
8262 * If we increment the root nodes slot counter past the number of
8263 * elements, 1 is returned to signal completion of the search.
8264 */
8265static int adjust_slots_upwards(struct btrfs_root *root,
8266 struct btrfs_path *path, int root_level)
8267{
8268 int level = 0;
8269 int nr, slot;
8270 struct extent_buffer *eb;
8271
8272 if (root_level == 0)
8273 return 1;
8274
8275 while (level <= root_level) {
8276 eb = path->nodes[level];
8277 nr = btrfs_header_nritems(eb);
8278 path->slots[level]++;
8279 slot = path->slots[level];
8280 if (slot >= nr || level == 0) {
8281 /*
8282 * Don't free the root - we will detect this
8283 * condition after our loop and return a
8284 * positive value for caller to stop walking the tree.
8285 */
8286 if (level != root_level) {
8287 btrfs_tree_unlock_rw(eb, path->locks[level]);
8288 path->locks[level] = 0;
8289
8290 free_extent_buffer(eb);
8291 path->nodes[level] = NULL;
8292 path->slots[level] = 0;
8293 }
8294 } else {
8295 /*
8296 * We have a valid slot to walk back down
8297 * from. Stop here so caller can process these
8298 * new nodes.
8299 */
8300 break;
8301 }
8302
8303 level++;
8304 }
8305
8306 eb = path->nodes[root_level];
8307 if (path->slots[root_level] >= btrfs_header_nritems(eb))
8308 return 1;
8309
8310 return 0;
8311}
8312
8313/*
8314 * root_eb is the subtree root and is locked before this function is called.
8315 */
8316static int account_shared_subtree(struct btrfs_trans_handle *trans,
8317 struct btrfs_root *root,
8318 struct extent_buffer *root_eb,
8319 u64 root_gen,
8320 int root_level)
8321{
8322 int ret = 0;
8323 int level;
8324 struct extent_buffer *eb = root_eb;
8325 struct btrfs_path *path = NULL;
8326
8327 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
8328 BUG_ON(root_eb == NULL);
8329
8330 if (!root->fs_info->quota_enabled)
8331 return 0;
8332
8333 if (!extent_buffer_uptodate(root_eb)) {
8334 ret = btrfs_read_buffer(root_eb, root_gen);
8335 if (ret)
8336 goto out;
8337 }
8338
8339 if (root_level == 0) {
8340 ret = account_leaf_items(trans, root, root_eb);
8341 goto out;
8342 }
8343
8344 path = btrfs_alloc_path();
8345 if (!path)
8346 return -ENOMEM;
8347
8348 /*
8349 * Walk down the tree. Missing extent blocks are filled in as
8350 * we go. Metadata is accounted every time we read a new
8351 * extent block.
8352 *
8353 * When we reach a leaf, we account for file extent items in it,
8354 * walk back up the tree (adjusting slot pointers as we go)
8355 * and restart the search process.
8356 */
8357 extent_buffer_get(root_eb); /* For path */
8358 path->nodes[root_level] = root_eb;
8359 path->slots[root_level] = 0;
8360 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
8361walk_down:
8362 level = root_level;
8363 while (level >= 0) {
8364 if (path->nodes[level] == NULL) {
8365 int parent_slot;
8366 u64 child_gen;
8367 u64 child_bytenr;
8368
8369 /* We need to get child blockptr/gen from
8370 * parent before we can read it. */
8371 eb = path->nodes[level + 1];
8372 parent_slot = path->slots[level + 1];
8373 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
8374 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
8375
8376 eb = read_tree_block(root, child_bytenr, child_gen);
8377 if (IS_ERR(eb)) {
8378 ret = PTR_ERR(eb);
8379 goto out;
8380 } else if (!extent_buffer_uptodate(eb)) {
8381 free_extent_buffer(eb);
8382 ret = -EIO;
8383 goto out;
8384 }
8385
8386 path->nodes[level] = eb;
8387 path->slots[level] = 0;
8388
8389 btrfs_tree_read_lock(eb);
8390 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8391 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8392
8393 ret = record_one_subtree_extent(trans, root, child_bytenr,
8394 root->nodesize);
8395 if (ret)
8396 goto out;
8397 }
8398
8399 if (level == 0) {
8400 ret = account_leaf_items(trans, root, path->nodes[level]);
8401 if (ret)
8402 goto out;
8403
8404 /* Nonzero return here means we completed our search */
8405 ret = adjust_slots_upwards(root, path, root_level);
8406 if (ret)
8407 break;
8408
8409 /* Restart search with new slots */
8410 goto walk_down;
8411 }
8412
8413 level--;
8414 }
8415
8416 ret = 0;
8417out:
8418 btrfs_free_path(path);
8419
8420 return ret;
8421}
8422
8423/*
8424 * helper to process tree block while walking down the tree.
8425 *
8426 * when wc->stage == UPDATE_BACKREF, this function updates
8427 * back refs for pointers in the block.
8428 *
8429 * NOTE: return value 1 means we should stop walking down.
8430 */
8431static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8432 struct btrfs_root *root,
8433 struct btrfs_path *path,
8434 struct walk_control *wc, int lookup_info)
8435{
8436 int level = wc->level;
8437 struct extent_buffer *eb = path->nodes[level];
8438 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8439 int ret;
8440
8441 if (wc->stage == UPDATE_BACKREF &&
8442 btrfs_header_owner(eb) != root->root_key.objectid)
8443 return 1;
8444
8445 /*
8446 * when reference count of tree block is 1, it won't increase
8447 * again. once full backref flag is set, we never clear it.
8448 */
8449 if (lookup_info &&
8450 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8451 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8452 BUG_ON(!path->locks[level]);
8453 ret = btrfs_lookup_extent_info(trans, root,
8454 eb->start, level, 1,
8455 &wc->refs[level],
8456 &wc->flags[level]);
8457 BUG_ON(ret == -ENOMEM);
8458 if (ret)
8459 return ret;
8460 BUG_ON(wc->refs[level] == 0);
8461 }
8462
8463 if (wc->stage == DROP_REFERENCE) {
8464 if (wc->refs[level] > 1)
8465 return 1;
8466
8467 if (path->locks[level] && !wc->keep_locks) {
8468 btrfs_tree_unlock_rw(eb, path->locks[level]);
8469 path->locks[level] = 0;
8470 }
8471 return 0;
8472 }
8473
8474 /* wc->stage == UPDATE_BACKREF */
8475 if (!(wc->flags[level] & flag)) {
8476 BUG_ON(!path->locks[level]);
8477 ret = btrfs_inc_ref(trans, root, eb, 1);
8478 BUG_ON(ret); /* -ENOMEM */
8479 ret = btrfs_dec_ref(trans, root, eb, 0);
8480 BUG_ON(ret); /* -ENOMEM */
8481 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8482 eb->len, flag,
8483 btrfs_header_level(eb), 0);
8484 BUG_ON(ret); /* -ENOMEM */
8485 wc->flags[level] |= flag;
8486 }
8487
8488 /*
8489 * the block is shared by multiple trees, so it's not good to
8490 * keep the tree lock
8491 */
8492 if (path->locks[level] && level > 0) {
8493 btrfs_tree_unlock_rw(eb, path->locks[level]);
8494 path->locks[level] = 0;
8495 }
8496 return 0;
8497}
8498
8499/*
8500 * helper to process tree block pointer.
8501 *
8502 * when wc->stage == DROP_REFERENCE, this function checks
8503 * reference count of the block pointed to. if the block
8504 * is shared and we need update back refs for the subtree
8505 * rooted at the block, this function changes wc->stage to
8506 * UPDATE_BACKREF. if the block is shared and there is no
8507 * need to update back, this function drops the reference
8508 * to the block.
8509 *
8510 * NOTE: return value 1 means we should stop walking down.
8511 */
8512static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8513 struct btrfs_root *root,
8514 struct btrfs_path *path,
8515 struct walk_control *wc, int *lookup_info)
8516{
8517 u64 bytenr;
8518 u64 generation;
8519 u64 parent;
8520 u32 blocksize;
8521 struct btrfs_key key;
8522 struct extent_buffer *next;
8523 int level = wc->level;
8524 int reada = 0;
8525 int ret = 0;
8526 bool need_account = false;
8527
8528 generation = btrfs_node_ptr_generation(path->nodes[level],
8529 path->slots[level]);
8530 /*
8531 * if the lower level block was created before the snapshot
8532 * was created, we know there is no need to update back refs
8533 * for the subtree
8534 */
8535 if (wc->stage == UPDATE_BACKREF &&
8536 generation <= root->root_key.offset) {
8537 *lookup_info = 1;
8538 return 1;
8539 }
8540
8541 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8542 blocksize = root->nodesize;
8543
8544 next = btrfs_find_tree_block(root->fs_info, bytenr);
8545 if (!next) {
8546 next = btrfs_find_create_tree_block(root, bytenr);
8547 if (!next)
8548 return -ENOMEM;
8549 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8550 level - 1);
8551 reada = 1;
8552 }
8553 btrfs_tree_lock(next);
8554 btrfs_set_lock_blocking(next);
8555
8556 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8557 &wc->refs[level - 1],
8558 &wc->flags[level - 1]);
8559 if (ret < 0) {
8560 btrfs_tree_unlock(next);
8561 return ret;
8562 }
8563
8564 if (unlikely(wc->refs[level - 1] == 0)) {
8565 btrfs_err(root->fs_info, "Missing references.");
8566 BUG();
8567 }
8568 *lookup_info = 0;
8569
8570 if (wc->stage == DROP_REFERENCE) {
8571 if (wc->refs[level - 1] > 1) {
8572 need_account = true;
8573 if (level == 1 &&
8574 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8575 goto skip;
8576
8577 if (!wc->update_ref ||
8578 generation <= root->root_key.offset)
8579 goto skip;
8580
8581 btrfs_node_key_to_cpu(path->nodes[level], &key,
8582 path->slots[level]);
8583 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8584 if (ret < 0)
8585 goto skip;
8586
8587 wc->stage = UPDATE_BACKREF;
8588 wc->shared_level = level - 1;
8589 }
8590 } else {
8591 if (level == 1 &&
8592 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8593 goto skip;
8594 }
8595
8596 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8597 btrfs_tree_unlock(next);
8598 free_extent_buffer(next);
8599 next = NULL;
8600 *lookup_info = 1;
8601 }
8602
8603 if (!next) {
8604 if (reada && level == 1)
8605 reada_walk_down(trans, root, wc, path);
8606 next = read_tree_block(root, bytenr, generation);
8607 if (IS_ERR(next)) {
8608 return PTR_ERR(next);
8609 } else if (!extent_buffer_uptodate(next)) {
8610 free_extent_buffer(next);
8611 return -EIO;
8612 }
8613 btrfs_tree_lock(next);
8614 btrfs_set_lock_blocking(next);
8615 }
8616
8617 level--;
8618 BUG_ON(level != btrfs_header_level(next));
8619 path->nodes[level] = next;
8620 path->slots[level] = 0;
8621 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8622 wc->level = level;
8623 if (wc->level == 1)
8624 wc->reada_slot = 0;
8625 return 0;
8626skip:
8627 wc->refs[level - 1] = 0;
8628 wc->flags[level - 1] = 0;
8629 if (wc->stage == DROP_REFERENCE) {
8630 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8631 parent = path->nodes[level]->start;
8632 } else {
8633 BUG_ON(root->root_key.objectid !=
8634 btrfs_header_owner(path->nodes[level]));
8635 parent = 0;
8636 }
8637
8638 if (need_account) {
8639 ret = account_shared_subtree(trans, root, next,
8640 generation, level - 1);
8641 if (ret) {
8642 btrfs_err_rl(root->fs_info,
8643 "Error "
8644 "%d accounting shared subtree. Quota "
8645 "is out of sync, rescan required.",
8646 ret);
8647 }
8648 }
8649 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8650 root->root_key.objectid, level - 1, 0);
8651 BUG_ON(ret); /* -ENOMEM */
8652 }
8653 btrfs_tree_unlock(next);
8654 free_extent_buffer(next);
8655 *lookup_info = 1;
8656 return 1;
8657}
8658
8659/*
8660 * helper to process tree block while walking up the tree.
8661 *
8662 * when wc->stage == DROP_REFERENCE, this function drops
8663 * reference count on the block.
8664 *
8665 * when wc->stage == UPDATE_BACKREF, this function changes
8666 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8667 * to UPDATE_BACKREF previously while processing the block.
8668 *
8669 * NOTE: return value 1 means we should stop walking up.
8670 */
8671static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8672 struct btrfs_root *root,
8673 struct btrfs_path *path,
8674 struct walk_control *wc)
8675{
8676 int ret;
8677 int level = wc->level;
8678 struct extent_buffer *eb = path->nodes[level];
8679 u64 parent = 0;
8680
8681 if (wc->stage == UPDATE_BACKREF) {
8682 BUG_ON(wc->shared_level < level);
8683 if (level < wc->shared_level)
8684 goto out;
8685
8686 ret = find_next_key(path, level + 1, &wc->update_progress);
8687 if (ret > 0)
8688 wc->update_ref = 0;
8689
8690 wc->stage = DROP_REFERENCE;
8691 wc->shared_level = -1;
8692 path->slots[level] = 0;
8693
8694 /*
8695 * check reference count again if the block isn't locked.
8696 * we should start walking down the tree again if reference
8697 * count is one.
8698 */
8699 if (!path->locks[level]) {
8700 BUG_ON(level == 0);
8701 btrfs_tree_lock(eb);
8702 btrfs_set_lock_blocking(eb);
8703 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8704
8705 ret = btrfs_lookup_extent_info(trans, root,
8706 eb->start, level, 1,
8707 &wc->refs[level],
8708 &wc->flags[level]);
8709 if (ret < 0) {
8710 btrfs_tree_unlock_rw(eb, path->locks[level]);
8711 path->locks[level] = 0;
8712 return ret;
8713 }
8714 BUG_ON(wc->refs[level] == 0);
8715 if (wc->refs[level] == 1) {
8716 btrfs_tree_unlock_rw(eb, path->locks[level]);
8717 path->locks[level] = 0;
8718 return 1;
8719 }
8720 }
8721 }
8722
8723 /* wc->stage == DROP_REFERENCE */
8724 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8725
8726 if (wc->refs[level] == 1) {
8727 if (level == 0) {
8728 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8729 ret = btrfs_dec_ref(trans, root, eb, 1);
8730 else
8731 ret = btrfs_dec_ref(trans, root, eb, 0);
8732 BUG_ON(ret); /* -ENOMEM */
8733 ret = account_leaf_items(trans, root, eb);
8734 if (ret) {
8735 btrfs_err_rl(root->fs_info,
8736 "error "
8737 "%d accounting leaf items. Quota "
8738 "is out of sync, rescan required.",
8739 ret);
8740 }
8741 }
8742 /* make block locked assertion in clean_tree_block happy */
8743 if (!path->locks[level] &&
8744 btrfs_header_generation(eb) == trans->transid) {
8745 btrfs_tree_lock(eb);
8746 btrfs_set_lock_blocking(eb);
8747 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8748 }
8749 clean_tree_block(trans, root->fs_info, eb);
8750 }
8751
8752 if (eb == root->node) {
8753 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8754 parent = eb->start;
8755 else
8756 BUG_ON(root->root_key.objectid !=
8757 btrfs_header_owner(eb));
8758 } else {
8759 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8760 parent = path->nodes[level + 1]->start;
8761 else
8762 BUG_ON(root->root_key.objectid !=
8763 btrfs_header_owner(path->nodes[level + 1]));
8764 }
8765
8766 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8767out:
8768 wc->refs[level] = 0;
8769 wc->flags[level] = 0;
8770 return 0;
8771}
8772
8773static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8774 struct btrfs_root *root,
8775 struct btrfs_path *path,
8776 struct walk_control *wc)
8777{
8778 int level = wc->level;
8779 int lookup_info = 1;
8780 int ret;
8781
8782 while (level >= 0) {
8783 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8784 if (ret > 0)
8785 break;
8786
8787 if (level == 0)
8788 break;
8789
8790 if (path->slots[level] >=
8791 btrfs_header_nritems(path->nodes[level]))
8792 break;
8793
8794 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8795 if (ret > 0) {
8796 path->slots[level]++;
8797 continue;
8798 } else if (ret < 0)
8799 return ret;
8800 level = wc->level;
8801 }
8802 return 0;
8803}
8804
8805static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8806 struct btrfs_root *root,
8807 struct btrfs_path *path,
8808 struct walk_control *wc, int max_level)
8809{
8810 int level = wc->level;
8811 int ret;
8812
8813 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8814 while (level < max_level && path->nodes[level]) {
8815 wc->level = level;
8816 if (path->slots[level] + 1 <
8817 btrfs_header_nritems(path->nodes[level])) {
8818 path->slots[level]++;
8819 return 0;
8820 } else {
8821 ret = walk_up_proc(trans, root, path, wc);
8822 if (ret > 0)
8823 return 0;
8824
8825 if (path->locks[level]) {
8826 btrfs_tree_unlock_rw(path->nodes[level],
8827 path->locks[level]);
8828 path->locks[level] = 0;
8829 }
8830 free_extent_buffer(path->nodes[level]);
8831 path->nodes[level] = NULL;
8832 level++;
8833 }
8834 }
8835 return 1;
8836}
8837
8838/*
8839 * drop a subvolume tree.
8840 *
8841 * this function traverses the tree freeing any blocks that only
8842 * referenced by the tree.
8843 *
8844 * when a shared tree block is found. this function decreases its
8845 * reference count by one. if update_ref is true, this function
8846 * also make sure backrefs for the shared block and all lower level
8847 * blocks are properly updated.
8848 *
8849 * If called with for_reloc == 0, may exit early with -EAGAIN
8850 */
8851int btrfs_drop_snapshot(struct btrfs_root *root,
8852 struct btrfs_block_rsv *block_rsv, int update_ref,
8853 int for_reloc)
8854{
8855 struct btrfs_path *path;
8856 struct btrfs_trans_handle *trans;
8857 struct btrfs_root *tree_root = root->fs_info->tree_root;
8858 struct btrfs_root_item *root_item = &root->root_item;
8859 struct walk_control *wc;
8860 struct btrfs_key key;
8861 int err = 0;
8862 int ret;
8863 int level;
8864 bool root_dropped = false;
8865
8866 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8867
8868 path = btrfs_alloc_path();
8869 if (!path) {
8870 err = -ENOMEM;
8871 goto out;
8872 }
8873
8874 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8875 if (!wc) {
8876 btrfs_free_path(path);
8877 err = -ENOMEM;
8878 goto out;
8879 }
8880
8881 trans = btrfs_start_transaction(tree_root, 0);
8882 if (IS_ERR(trans)) {
8883 err = PTR_ERR(trans);
8884 goto out_free;
8885 }
8886
8887 if (block_rsv)
8888 trans->block_rsv = block_rsv;
8889
8890 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8891 level = btrfs_header_level(root->node);
8892 path->nodes[level] = btrfs_lock_root_node(root);
8893 btrfs_set_lock_blocking(path->nodes[level]);
8894 path->slots[level] = 0;
8895 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8896 memset(&wc->update_progress, 0,
8897 sizeof(wc->update_progress));
8898 } else {
8899 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8900 memcpy(&wc->update_progress, &key,
8901 sizeof(wc->update_progress));
8902
8903 level = root_item->drop_level;
8904 BUG_ON(level == 0);
8905 path->lowest_level = level;
8906 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8907 path->lowest_level = 0;
8908 if (ret < 0) {
8909 err = ret;
8910 goto out_end_trans;
8911 }
8912 WARN_ON(ret > 0);
8913
8914 /*
8915 * unlock our path, this is safe because only this
8916 * function is allowed to delete this snapshot
8917 */
8918 btrfs_unlock_up_safe(path, 0);
8919
8920 level = btrfs_header_level(root->node);
8921 while (1) {
8922 btrfs_tree_lock(path->nodes[level]);
8923 btrfs_set_lock_blocking(path->nodes[level]);
8924 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8925
8926 ret = btrfs_lookup_extent_info(trans, root,
8927 path->nodes[level]->start,
8928 level, 1, &wc->refs[level],
8929 &wc->flags[level]);
8930 if (ret < 0) {
8931 err = ret;
8932 goto out_end_trans;
8933 }
8934 BUG_ON(wc->refs[level] == 0);
8935
8936 if (level == root_item->drop_level)
8937 break;
8938
8939 btrfs_tree_unlock(path->nodes[level]);
8940 path->locks[level] = 0;
8941 WARN_ON(wc->refs[level] != 1);
8942 level--;
8943 }
8944 }
8945
8946 wc->level = level;
8947 wc->shared_level = -1;
8948 wc->stage = DROP_REFERENCE;
8949 wc->update_ref = update_ref;
8950 wc->keep_locks = 0;
8951 wc->for_reloc = for_reloc;
8952 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8953
8954 while (1) {
8955
8956 ret = walk_down_tree(trans, root, path, wc);
8957 if (ret < 0) {
8958 err = ret;
8959 break;
8960 }
8961
8962 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8963 if (ret < 0) {
8964 err = ret;
8965 break;
8966 }
8967
8968 if (ret > 0) {
8969 BUG_ON(wc->stage != DROP_REFERENCE);
8970 break;
8971 }
8972
8973 if (wc->stage == DROP_REFERENCE) {
8974 level = wc->level;
8975 btrfs_node_key(path->nodes[level],
8976 &root_item->drop_progress,
8977 path->slots[level]);
8978 root_item->drop_level = level;
8979 }
8980
8981 BUG_ON(wc->level == 0);
8982 if (btrfs_should_end_transaction(trans, tree_root) ||
8983 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8984 ret = btrfs_update_root(trans, tree_root,
8985 &root->root_key,
8986 root_item);
8987 if (ret) {
8988 btrfs_abort_transaction(trans, tree_root, ret);
8989 err = ret;
8990 goto out_end_trans;
8991 }
8992
8993 btrfs_end_transaction_throttle(trans, tree_root);
8994 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8995 pr_debug("BTRFS: drop snapshot early exit\n");
8996 err = -EAGAIN;
8997 goto out_free;
8998 }
8999
9000 trans = btrfs_start_transaction(tree_root, 0);
9001 if (IS_ERR(trans)) {
9002 err = PTR_ERR(trans);
9003 goto out_free;
9004 }
9005 if (block_rsv)
9006 trans->block_rsv = block_rsv;
9007 }
9008 }
9009 btrfs_release_path(path);
9010 if (err)
9011 goto out_end_trans;
9012
9013 ret = btrfs_del_root(trans, tree_root, &root->root_key);
9014 if (ret) {
9015 btrfs_abort_transaction(trans, tree_root, ret);
9016 goto out_end_trans;
9017 }
9018
9019 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9020 ret = btrfs_find_root(tree_root, &root->root_key, path,
9021 NULL, NULL);
9022 if (ret < 0) {
9023 btrfs_abort_transaction(trans, tree_root, ret);
9024 err = ret;
9025 goto out_end_trans;
9026 } else if (ret > 0) {
9027 /* if we fail to delete the orphan item this time
9028 * around, it'll get picked up the next time.
9029 *
9030 * The most common failure here is just -ENOENT.
9031 */
9032 btrfs_del_orphan_item(trans, tree_root,
9033 root->root_key.objectid);
9034 }
9035 }
9036
9037 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9038 btrfs_add_dropped_root(trans, root);
9039 } else {
9040 free_extent_buffer(root->node);
9041 free_extent_buffer(root->commit_root);
9042 btrfs_put_fs_root(root);
9043 }
9044 root_dropped = true;
9045out_end_trans:
9046 btrfs_end_transaction_throttle(trans, tree_root);
9047out_free:
9048 kfree(wc);
9049 btrfs_free_path(path);
9050out:
9051 /*
9052 * So if we need to stop dropping the snapshot for whatever reason we
9053 * need to make sure to add it back to the dead root list so that we
9054 * keep trying to do the work later. This also cleans up roots if we
9055 * don't have it in the radix (like when we recover after a power fail
9056 * or unmount) so we don't leak memory.
9057 */
9058 if (!for_reloc && root_dropped == false)
9059 btrfs_add_dead_root(root);
9060 if (err && err != -EAGAIN)
9061 btrfs_std_error(root->fs_info, err, NULL);
9062 return err;
9063}
9064
9065/*
9066 * drop subtree rooted at tree block 'node'.
9067 *
9068 * NOTE: this function will unlock and release tree block 'node'
9069 * only used by relocation code
9070 */
9071int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9072 struct btrfs_root *root,
9073 struct extent_buffer *node,
9074 struct extent_buffer *parent)
9075{
9076 struct btrfs_path *path;
9077 struct walk_control *wc;
9078 int level;
9079 int parent_level;
9080 int ret = 0;
9081 int wret;
9082
9083 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9084
9085 path = btrfs_alloc_path();
9086 if (!path)
9087 return -ENOMEM;
9088
9089 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9090 if (!wc) {
9091 btrfs_free_path(path);
9092 return -ENOMEM;
9093 }
9094
9095 btrfs_assert_tree_locked(parent);
9096 parent_level = btrfs_header_level(parent);
9097 extent_buffer_get(parent);
9098 path->nodes[parent_level] = parent;
9099 path->slots[parent_level] = btrfs_header_nritems(parent);
9100
9101 btrfs_assert_tree_locked(node);
9102 level = btrfs_header_level(node);
9103 path->nodes[level] = node;
9104 path->slots[level] = 0;
9105 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9106
9107 wc->refs[parent_level] = 1;
9108 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9109 wc->level = level;
9110 wc->shared_level = -1;
9111 wc->stage = DROP_REFERENCE;
9112 wc->update_ref = 0;
9113 wc->keep_locks = 1;
9114 wc->for_reloc = 1;
9115 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9116
9117 while (1) {
9118 wret = walk_down_tree(trans, root, path, wc);
9119 if (wret < 0) {
9120 ret = wret;
9121 break;
9122 }
9123
9124 wret = walk_up_tree(trans, root, path, wc, parent_level);
9125 if (wret < 0)
9126 ret = wret;
9127 if (wret != 0)
9128 break;
9129 }
9130
9131 kfree(wc);
9132 btrfs_free_path(path);
9133 return ret;
9134}
9135
9136static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
9137{
9138 u64 num_devices;
9139 u64 stripped;
9140
9141 /*
9142 * if restripe for this chunk_type is on pick target profile and
9143 * return, otherwise do the usual balance
9144 */
9145 stripped = get_restripe_target(root->fs_info, flags);
9146 if (stripped)
9147 return extended_to_chunk(stripped);
9148
9149 num_devices = root->fs_info->fs_devices->rw_devices;
9150
9151 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9152 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9153 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9154
9155 if (num_devices == 1) {
9156 stripped |= BTRFS_BLOCK_GROUP_DUP;
9157 stripped = flags & ~stripped;
9158
9159 /* turn raid0 into single device chunks */
9160 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9161 return stripped;
9162
9163 /* turn mirroring into duplication */
9164 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9165 BTRFS_BLOCK_GROUP_RAID10))
9166 return stripped | BTRFS_BLOCK_GROUP_DUP;
9167 } else {
9168 /* they already had raid on here, just return */
9169 if (flags & stripped)
9170 return flags;
9171
9172 stripped |= BTRFS_BLOCK_GROUP_DUP;
9173 stripped = flags & ~stripped;
9174
9175 /* switch duplicated blocks with raid1 */
9176 if (flags & BTRFS_BLOCK_GROUP_DUP)
9177 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9178
9179 /* this is drive concat, leave it alone */
9180 }
9181
9182 return flags;
9183}
9184
9185static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9186{
9187 struct btrfs_space_info *sinfo = cache->space_info;
9188 u64 num_bytes;
9189 u64 min_allocable_bytes;
9190 int ret = -ENOSPC;
9191
9192 /*
9193 * We need some metadata space and system metadata space for
9194 * allocating chunks in some corner cases until we force to set
9195 * it to be readonly.
9196 */
9197 if ((sinfo->flags &
9198 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9199 !force)
9200 min_allocable_bytes = SZ_1M;
9201 else
9202 min_allocable_bytes = 0;
9203
9204 spin_lock(&sinfo->lock);
9205 spin_lock(&cache->lock);
9206
9207 if (cache->ro) {
9208 cache->ro++;
9209 ret = 0;
9210 goto out;
9211 }
9212
9213 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9214 cache->bytes_super - btrfs_block_group_used(&cache->item);
9215
9216 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
9217 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
9218 min_allocable_bytes <= sinfo->total_bytes) {
9219 sinfo->bytes_readonly += num_bytes;
9220 cache->ro++;
9221 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9222 ret = 0;
9223 }
9224out:
9225 spin_unlock(&cache->lock);
9226 spin_unlock(&sinfo->lock);
9227 return ret;
9228}
9229
9230int btrfs_inc_block_group_ro(struct btrfs_root *root,
9231 struct btrfs_block_group_cache *cache)
9232
9233{
9234 struct btrfs_trans_handle *trans;
9235 u64 alloc_flags;
9236 int ret;
9237
9238again:
9239 trans = btrfs_join_transaction(root);
9240 if (IS_ERR(trans))
9241 return PTR_ERR(trans);
9242
9243 /*
9244 * we're not allowed to set block groups readonly after the dirty
9245 * block groups cache has started writing. If it already started,
9246 * back off and let this transaction commit
9247 */
9248 mutex_lock(&root->fs_info->ro_block_group_mutex);
9249 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9250 u64 transid = trans->transid;
9251
9252 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9253 btrfs_end_transaction(trans, root);
9254
9255 ret = btrfs_wait_for_commit(root, transid);
9256 if (ret)
9257 return ret;
9258 goto again;
9259 }
9260
9261 /*
9262 * if we are changing raid levels, try to allocate a corresponding
9263 * block group with the new raid level.
9264 */
9265 alloc_flags = update_block_group_flags(root, cache->flags);
9266 if (alloc_flags != cache->flags) {
9267 ret = do_chunk_alloc(trans, root, alloc_flags,
9268 CHUNK_ALLOC_FORCE);
9269 /*
9270 * ENOSPC is allowed here, we may have enough space
9271 * already allocated at the new raid level to
9272 * carry on
9273 */
9274 if (ret == -ENOSPC)
9275 ret = 0;
9276 if (ret < 0)
9277 goto out;
9278 }
9279
9280 ret = inc_block_group_ro(cache, 0);
9281 if (!ret)
9282 goto out;
9283 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
9284 ret = do_chunk_alloc(trans, root, alloc_flags,
9285 CHUNK_ALLOC_FORCE);
9286 if (ret < 0)
9287 goto out;
9288 ret = inc_block_group_ro(cache, 0);
9289out:
9290 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9291 alloc_flags = update_block_group_flags(root, cache->flags);
9292 lock_chunks(root->fs_info->chunk_root);
9293 check_system_chunk(trans, root, alloc_flags);
9294 unlock_chunks(root->fs_info->chunk_root);
9295 }
9296 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9297
9298 btrfs_end_transaction(trans, root);
9299 return ret;
9300}
9301
9302int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9303 struct btrfs_root *root, u64 type)
9304{
9305 u64 alloc_flags = get_alloc_profile(root, type);
9306 return do_chunk_alloc(trans, root, alloc_flags,
9307 CHUNK_ALLOC_FORCE);
9308}
9309
9310/*
9311 * helper to account the unused space of all the readonly block group in the
9312 * space_info. takes mirrors into account.
9313 */
9314u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9315{
9316 struct btrfs_block_group_cache *block_group;
9317 u64 free_bytes = 0;
9318 int factor;
9319
9320 /* It's df, we don't care if it's racey */
9321 if (list_empty(&sinfo->ro_bgs))
9322 return 0;
9323
9324 spin_lock(&sinfo->lock);
9325 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9326 spin_lock(&block_group->lock);
9327
9328 if (!block_group->ro) {
9329 spin_unlock(&block_group->lock);
9330 continue;
9331 }
9332
9333 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9334 BTRFS_BLOCK_GROUP_RAID10 |
9335 BTRFS_BLOCK_GROUP_DUP))
9336 factor = 2;
9337 else
9338 factor = 1;
9339
9340 free_bytes += (block_group->key.offset -
9341 btrfs_block_group_used(&block_group->item)) *
9342 factor;
9343
9344 spin_unlock(&block_group->lock);
9345 }
9346 spin_unlock(&sinfo->lock);
9347
9348 return free_bytes;
9349}
9350
9351void btrfs_dec_block_group_ro(struct btrfs_root *root,
9352 struct btrfs_block_group_cache *cache)
9353{
9354 struct btrfs_space_info *sinfo = cache->space_info;
9355 u64 num_bytes;
9356
9357 BUG_ON(!cache->ro);
9358
9359 spin_lock(&sinfo->lock);
9360 spin_lock(&cache->lock);
9361 if (!--cache->ro) {
9362 num_bytes = cache->key.offset - cache->reserved -
9363 cache->pinned - cache->bytes_super -
9364 btrfs_block_group_used(&cache->item);
9365 sinfo->bytes_readonly -= num_bytes;
9366 list_del_init(&cache->ro_list);
9367 }
9368 spin_unlock(&cache->lock);
9369 spin_unlock(&sinfo->lock);
9370}
9371
9372/*
9373 * checks to see if its even possible to relocate this block group.
9374 *
9375 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9376 * ok to go ahead and try.
9377 */
9378int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
9379{
9380 struct btrfs_block_group_cache *block_group;
9381 struct btrfs_space_info *space_info;
9382 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
9383 struct btrfs_device *device;
9384 struct btrfs_trans_handle *trans;
9385 u64 min_free;
9386 u64 dev_min = 1;
9387 u64 dev_nr = 0;
9388 u64 target;
9389 int debug;
9390 int index;
9391 int full = 0;
9392 int ret = 0;
9393
9394 debug = btrfs_test_opt(root, ENOSPC_DEBUG);
9395
9396 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9397
9398 /* odd, couldn't find the block group, leave it alone */
9399 if (!block_group) {
9400 if (debug)
9401 btrfs_warn(root->fs_info,
9402 "can't find block group for bytenr %llu",
9403 bytenr);
9404 return -1;
9405 }
9406
9407 min_free = btrfs_block_group_used(&block_group->item);
9408
9409 /* no bytes used, we're good */
9410 if (!min_free)
9411 goto out;
9412
9413 space_info = block_group->space_info;
9414 spin_lock(&space_info->lock);
9415
9416 full = space_info->full;
9417
9418 /*
9419 * if this is the last block group we have in this space, we can't
9420 * relocate it unless we're able to allocate a new chunk below.
9421 *
9422 * Otherwise, we need to make sure we have room in the space to handle
9423 * all of the extents from this block group. If we can, we're good
9424 */
9425 if ((space_info->total_bytes != block_group->key.offset) &&
9426 (space_info->bytes_used + space_info->bytes_reserved +
9427 space_info->bytes_pinned + space_info->bytes_readonly +
9428 min_free < space_info->total_bytes)) {
9429 spin_unlock(&space_info->lock);
9430 goto out;
9431 }
9432 spin_unlock(&space_info->lock);
9433
9434 /*
9435 * ok we don't have enough space, but maybe we have free space on our
9436 * devices to allocate new chunks for relocation, so loop through our
9437 * alloc devices and guess if we have enough space. if this block
9438 * group is going to be restriped, run checks against the target
9439 * profile instead of the current one.
9440 */
9441 ret = -1;
9442
9443 /*
9444 * index:
9445 * 0: raid10
9446 * 1: raid1
9447 * 2: dup
9448 * 3: raid0
9449 * 4: single
9450 */
9451 target = get_restripe_target(root->fs_info, block_group->flags);
9452 if (target) {
9453 index = __get_raid_index(extended_to_chunk(target));
9454 } else {
9455 /*
9456 * this is just a balance, so if we were marked as full
9457 * we know there is no space for a new chunk
9458 */
9459 if (full) {
9460 if (debug)
9461 btrfs_warn(root->fs_info,
9462 "no space to alloc new chunk for block group %llu",
9463 block_group->key.objectid);
9464 goto out;
9465 }
9466
9467 index = get_block_group_index(block_group);
9468 }
9469
9470 if (index == BTRFS_RAID_RAID10) {
9471 dev_min = 4;
9472 /* Divide by 2 */
9473 min_free >>= 1;
9474 } else if (index == BTRFS_RAID_RAID1) {
9475 dev_min = 2;
9476 } else if (index == BTRFS_RAID_DUP) {
9477 /* Multiply by 2 */
9478 min_free <<= 1;
9479 } else if (index == BTRFS_RAID_RAID0) {
9480 dev_min = fs_devices->rw_devices;
9481 min_free = div64_u64(min_free, dev_min);
9482 }
9483
9484 /* We need to do this so that we can look at pending chunks */
9485 trans = btrfs_join_transaction(root);
9486 if (IS_ERR(trans)) {
9487 ret = PTR_ERR(trans);
9488 goto out;
9489 }
9490
9491 mutex_lock(&root->fs_info->chunk_mutex);
9492 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9493 u64 dev_offset;
9494
9495 /*
9496 * check to make sure we can actually find a chunk with enough
9497 * space to fit our block group in.
9498 */
9499 if (device->total_bytes > device->bytes_used + min_free &&
9500 !device->is_tgtdev_for_dev_replace) {
9501 ret = find_free_dev_extent(trans, device, min_free,
9502 &dev_offset, NULL);
9503 if (!ret)
9504 dev_nr++;
9505
9506 if (dev_nr >= dev_min)
9507 break;
9508
9509 ret = -1;
9510 }
9511 }
9512 if (debug && ret == -1)
9513 btrfs_warn(root->fs_info,
9514 "no space to allocate a new chunk for block group %llu",
9515 block_group->key.objectid);
9516 mutex_unlock(&root->fs_info->chunk_mutex);
9517 btrfs_end_transaction(trans, root);
9518out:
9519 btrfs_put_block_group(block_group);
9520 return ret;
9521}
9522
9523static int find_first_block_group(struct btrfs_root *root,
9524 struct btrfs_path *path, struct btrfs_key *key)
9525{
9526 int ret = 0;
9527 struct btrfs_key found_key;
9528 struct extent_buffer *leaf;
9529 int slot;
9530
9531 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9532 if (ret < 0)
9533 goto out;
9534
9535 while (1) {
9536 slot = path->slots[0];
9537 leaf = path->nodes[0];
9538 if (slot >= btrfs_header_nritems(leaf)) {
9539 ret = btrfs_next_leaf(root, path);
9540 if (ret == 0)
9541 continue;
9542 if (ret < 0)
9543 goto out;
9544 break;
9545 }
9546 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9547
9548 if (found_key.objectid >= key->objectid &&
9549 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9550 ret = 0;
9551 goto out;
9552 }
9553 path->slots[0]++;
9554 }
9555out:
9556 return ret;
9557}
9558
9559void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9560{
9561 struct btrfs_block_group_cache *block_group;
9562 u64 last = 0;
9563
9564 while (1) {
9565 struct inode *inode;
9566
9567 block_group = btrfs_lookup_first_block_group(info, last);
9568 while (block_group) {
9569 spin_lock(&block_group->lock);
9570 if (block_group->iref)
9571 break;
9572 spin_unlock(&block_group->lock);
9573 block_group = next_block_group(info->tree_root,
9574 block_group);
9575 }
9576 if (!block_group) {
9577 if (last == 0)
9578 break;
9579 last = 0;
9580 continue;
9581 }
9582
9583 inode = block_group->inode;
9584 block_group->iref = 0;
9585 block_group->inode = NULL;
9586 spin_unlock(&block_group->lock);
9587 iput(inode);
9588 last = block_group->key.objectid + block_group->key.offset;
9589 btrfs_put_block_group(block_group);
9590 }
9591}
9592
9593int btrfs_free_block_groups(struct btrfs_fs_info *info)
9594{
9595 struct btrfs_block_group_cache *block_group;
9596 struct btrfs_space_info *space_info;
9597 struct btrfs_caching_control *caching_ctl;
9598 struct rb_node *n;
9599
9600 down_write(&info->commit_root_sem);
9601 while (!list_empty(&info->caching_block_groups)) {
9602 caching_ctl = list_entry(info->caching_block_groups.next,
9603 struct btrfs_caching_control, list);
9604 list_del(&caching_ctl->list);
9605 put_caching_control(caching_ctl);
9606 }
9607 up_write(&info->commit_root_sem);
9608
9609 spin_lock(&info->unused_bgs_lock);
9610 while (!list_empty(&info->unused_bgs)) {
9611 block_group = list_first_entry(&info->unused_bgs,
9612 struct btrfs_block_group_cache,
9613 bg_list);
9614 list_del_init(&block_group->bg_list);
9615 btrfs_put_block_group(block_group);
9616 }
9617 spin_unlock(&info->unused_bgs_lock);
9618
9619 spin_lock(&info->block_group_cache_lock);
9620 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9621 block_group = rb_entry(n, struct btrfs_block_group_cache,
9622 cache_node);
9623 rb_erase(&block_group->cache_node,
9624 &info->block_group_cache_tree);
9625 RB_CLEAR_NODE(&block_group->cache_node);
9626 spin_unlock(&info->block_group_cache_lock);
9627
9628 down_write(&block_group->space_info->groups_sem);
9629 list_del(&block_group->list);
9630 up_write(&block_group->space_info->groups_sem);
9631
9632 if (block_group->cached == BTRFS_CACHE_STARTED)
9633 wait_block_group_cache_done(block_group);
9634
9635 /*
9636 * We haven't cached this block group, which means we could
9637 * possibly have excluded extents on this block group.
9638 */
9639 if (block_group->cached == BTRFS_CACHE_NO ||
9640 block_group->cached == BTRFS_CACHE_ERROR)
9641 free_excluded_extents(info->extent_root, block_group);
9642
9643 btrfs_remove_free_space_cache(block_group);
9644 btrfs_put_block_group(block_group);
9645
9646 spin_lock(&info->block_group_cache_lock);
9647 }
9648 spin_unlock(&info->block_group_cache_lock);
9649
9650 /* now that all the block groups are freed, go through and
9651 * free all the space_info structs. This is only called during
9652 * the final stages of unmount, and so we know nobody is
9653 * using them. We call synchronize_rcu() once before we start,
9654 * just to be on the safe side.
9655 */
9656 synchronize_rcu();
9657
9658 release_global_block_rsv(info);
9659
9660 while (!list_empty(&info->space_info)) {
9661 int i;
9662
9663 space_info = list_entry(info->space_info.next,
9664 struct btrfs_space_info,
9665 list);
9666 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9667 if (WARN_ON(space_info->bytes_pinned > 0 ||
9668 space_info->bytes_reserved > 0 ||
9669 space_info->bytes_may_use > 0)) {
9670 dump_space_info(space_info, 0, 0);
9671 }
9672 }
9673 list_del(&space_info->list);
9674 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9675 struct kobject *kobj;
9676 kobj = space_info->block_group_kobjs[i];
9677 space_info->block_group_kobjs[i] = NULL;
9678 if (kobj) {
9679 kobject_del(kobj);
9680 kobject_put(kobj);
9681 }
9682 }
9683 kobject_del(&space_info->kobj);
9684 kobject_put(&space_info->kobj);
9685 }
9686 return 0;
9687}
9688
9689static void __link_block_group(struct btrfs_space_info *space_info,
9690 struct btrfs_block_group_cache *cache)
9691{
9692 int index = get_block_group_index(cache);
9693 bool first = false;
9694
9695 down_write(&space_info->groups_sem);
9696 if (list_empty(&space_info->block_groups[index]))
9697 first = true;
9698 list_add_tail(&cache->list, &space_info->block_groups[index]);
9699 up_write(&space_info->groups_sem);
9700
9701 if (first) {
9702 struct raid_kobject *rkobj;
9703 int ret;
9704
9705 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9706 if (!rkobj)
9707 goto out_err;
9708 rkobj->raid_type = index;
9709 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9710 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9711 "%s", get_raid_name(index));
9712 if (ret) {
9713 kobject_put(&rkobj->kobj);
9714 goto out_err;
9715 }
9716 space_info->block_group_kobjs[index] = &rkobj->kobj;
9717 }
9718
9719 return;
9720out_err:
9721 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9722}
9723
9724static struct btrfs_block_group_cache *
9725btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9726{
9727 struct btrfs_block_group_cache *cache;
9728
9729 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9730 if (!cache)
9731 return NULL;
9732
9733 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9734 GFP_NOFS);
9735 if (!cache->free_space_ctl) {
9736 kfree(cache);
9737 return NULL;
9738 }
9739
9740 cache->key.objectid = start;
9741 cache->key.offset = size;
9742 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9743
9744 cache->sectorsize = root->sectorsize;
9745 cache->fs_info = root->fs_info;
9746 cache->full_stripe_len = btrfs_full_stripe_len(root,
9747 &root->fs_info->mapping_tree,
9748 start);
9749 set_free_space_tree_thresholds(cache);
9750
9751 atomic_set(&cache->count, 1);
9752 spin_lock_init(&cache->lock);
9753 init_rwsem(&cache->data_rwsem);
9754 INIT_LIST_HEAD(&cache->list);
9755 INIT_LIST_HEAD(&cache->cluster_list);
9756 INIT_LIST_HEAD(&cache->bg_list);
9757 INIT_LIST_HEAD(&cache->ro_list);
9758 INIT_LIST_HEAD(&cache->dirty_list);
9759 INIT_LIST_HEAD(&cache->io_list);
9760 btrfs_init_free_space_ctl(cache);
9761 atomic_set(&cache->trimming, 0);
9762 mutex_init(&cache->free_space_lock);
9763
9764 return cache;
9765}
9766
9767int btrfs_read_block_groups(struct btrfs_root *root)
9768{
9769 struct btrfs_path *path;
9770 int ret;
9771 struct btrfs_block_group_cache *cache;
9772 struct btrfs_fs_info *info = root->fs_info;
9773 struct btrfs_space_info *space_info;
9774 struct btrfs_key key;
9775 struct btrfs_key found_key;
9776 struct extent_buffer *leaf;
9777 int need_clear = 0;
9778 u64 cache_gen;
9779
9780 root = info->extent_root;
9781 key.objectid = 0;
9782 key.offset = 0;
9783 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9784 path = btrfs_alloc_path();
9785 if (!path)
9786 return -ENOMEM;
9787 path->reada = READA_FORWARD;
9788
9789 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9790 if (btrfs_test_opt(root, SPACE_CACHE) &&
9791 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9792 need_clear = 1;
9793 if (btrfs_test_opt(root, CLEAR_CACHE))
9794 need_clear = 1;
9795
9796 while (1) {
9797 ret = find_first_block_group(root, path, &key);
9798 if (ret > 0)
9799 break;
9800 if (ret != 0)
9801 goto error;
9802
9803 leaf = path->nodes[0];
9804 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9805
9806 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9807 found_key.offset);
9808 if (!cache) {
9809 ret = -ENOMEM;
9810 goto error;
9811 }
9812
9813 if (need_clear) {
9814 /*
9815 * When we mount with old space cache, we need to
9816 * set BTRFS_DC_CLEAR and set dirty flag.
9817 *
9818 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9819 * truncate the old free space cache inode and
9820 * setup a new one.
9821 * b) Setting 'dirty flag' makes sure that we flush
9822 * the new space cache info onto disk.
9823 */
9824 if (btrfs_test_opt(root, SPACE_CACHE))
9825 cache->disk_cache_state = BTRFS_DC_CLEAR;
9826 }
9827
9828 read_extent_buffer(leaf, &cache->item,
9829 btrfs_item_ptr_offset(leaf, path->slots[0]),
9830 sizeof(cache->item));
9831 cache->flags = btrfs_block_group_flags(&cache->item);
9832
9833 key.objectid = found_key.objectid + found_key.offset;
9834 btrfs_release_path(path);
9835
9836 /*
9837 * We need to exclude the super stripes now so that the space
9838 * info has super bytes accounted for, otherwise we'll think
9839 * we have more space than we actually do.
9840 */
9841 ret = exclude_super_stripes(root, cache);
9842 if (ret) {
9843 /*
9844 * We may have excluded something, so call this just in
9845 * case.
9846 */
9847 free_excluded_extents(root, cache);
9848 btrfs_put_block_group(cache);
9849 goto error;
9850 }
9851
9852 /*
9853 * check for two cases, either we are full, and therefore
9854 * don't need to bother with the caching work since we won't
9855 * find any space, or we are empty, and we can just add all
9856 * the space in and be done with it. This saves us _alot_ of
9857 * time, particularly in the full case.
9858 */
9859 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9860 cache->last_byte_to_unpin = (u64)-1;
9861 cache->cached = BTRFS_CACHE_FINISHED;
9862 free_excluded_extents(root, cache);
9863 } else if (btrfs_block_group_used(&cache->item) == 0) {
9864 cache->last_byte_to_unpin = (u64)-1;
9865 cache->cached = BTRFS_CACHE_FINISHED;
9866 add_new_free_space(cache, root->fs_info,
9867 found_key.objectid,
9868 found_key.objectid +
9869 found_key.offset);
9870 free_excluded_extents(root, cache);
9871 }
9872
9873 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9874 if (ret) {
9875 btrfs_remove_free_space_cache(cache);
9876 btrfs_put_block_group(cache);
9877 goto error;
9878 }
9879
9880 ret = update_space_info(info, cache->flags, found_key.offset,
9881 btrfs_block_group_used(&cache->item),
9882 &space_info);
9883 if (ret) {
9884 btrfs_remove_free_space_cache(cache);
9885 spin_lock(&info->block_group_cache_lock);
9886 rb_erase(&cache->cache_node,
9887 &info->block_group_cache_tree);
9888 RB_CLEAR_NODE(&cache->cache_node);
9889 spin_unlock(&info->block_group_cache_lock);
9890 btrfs_put_block_group(cache);
9891 goto error;
9892 }
9893
9894 cache->space_info = space_info;
9895 spin_lock(&cache->space_info->lock);
9896 cache->space_info->bytes_readonly += cache->bytes_super;
9897 spin_unlock(&cache->space_info->lock);
9898
9899 __link_block_group(space_info, cache);
9900
9901 set_avail_alloc_bits(root->fs_info, cache->flags);
9902 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9903 inc_block_group_ro(cache, 1);
9904 } else if (btrfs_block_group_used(&cache->item) == 0) {
9905 spin_lock(&info->unused_bgs_lock);
9906 /* Should always be true but just in case. */
9907 if (list_empty(&cache->bg_list)) {
9908 btrfs_get_block_group(cache);
9909 list_add_tail(&cache->bg_list,
9910 &info->unused_bgs);
9911 }
9912 spin_unlock(&info->unused_bgs_lock);
9913 }
9914 }
9915
9916 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9917 if (!(get_alloc_profile(root, space_info->flags) &
9918 (BTRFS_BLOCK_GROUP_RAID10 |
9919 BTRFS_BLOCK_GROUP_RAID1 |
9920 BTRFS_BLOCK_GROUP_RAID5 |
9921 BTRFS_BLOCK_GROUP_RAID6 |
9922 BTRFS_BLOCK_GROUP_DUP)))
9923 continue;
9924 /*
9925 * avoid allocating from un-mirrored block group if there are
9926 * mirrored block groups.
9927 */
9928 list_for_each_entry(cache,
9929 &space_info->block_groups[BTRFS_RAID_RAID0],
9930 list)
9931 inc_block_group_ro(cache, 1);
9932 list_for_each_entry(cache,
9933 &space_info->block_groups[BTRFS_RAID_SINGLE],
9934 list)
9935 inc_block_group_ro(cache, 1);
9936 }
9937
9938 init_global_block_rsv(info);
9939 ret = 0;
9940error:
9941 btrfs_free_path(path);
9942 return ret;
9943}
9944
9945void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9946 struct btrfs_root *root)
9947{
9948 struct btrfs_block_group_cache *block_group, *tmp;
9949 struct btrfs_root *extent_root = root->fs_info->extent_root;
9950 struct btrfs_block_group_item item;
9951 struct btrfs_key key;
9952 int ret = 0;
9953 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
9954
9955 trans->can_flush_pending_bgs = false;
9956 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9957 if (ret)
9958 goto next;
9959
9960 spin_lock(&block_group->lock);
9961 memcpy(&item, &block_group->item, sizeof(item));
9962 memcpy(&key, &block_group->key, sizeof(key));
9963 spin_unlock(&block_group->lock);
9964
9965 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9966 sizeof(item));
9967 if (ret)
9968 btrfs_abort_transaction(trans, extent_root, ret);
9969 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9970 key.objectid, key.offset);
9971 if (ret)
9972 btrfs_abort_transaction(trans, extent_root, ret);
9973 add_block_group_free_space(trans, root->fs_info, block_group);
9974 /* already aborted the transaction if it failed. */
9975next:
9976 list_del_init(&block_group->bg_list);
9977 }
9978 trans->can_flush_pending_bgs = can_flush_pending_bgs;
9979}
9980
9981int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9982 struct btrfs_root *root, u64 bytes_used,
9983 u64 type, u64 chunk_objectid, u64 chunk_offset,
9984 u64 size)
9985{
9986 int ret;
9987 struct btrfs_root *extent_root;
9988 struct btrfs_block_group_cache *cache;
9989
9990 extent_root = root->fs_info->extent_root;
9991
9992 btrfs_set_log_full_commit(root->fs_info, trans);
9993
9994 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9995 if (!cache)
9996 return -ENOMEM;
9997
9998 btrfs_set_block_group_used(&cache->item, bytes_used);
9999 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
10000 btrfs_set_block_group_flags(&cache->item, type);
10001
10002 cache->flags = type;
10003 cache->last_byte_to_unpin = (u64)-1;
10004 cache->cached = BTRFS_CACHE_FINISHED;
10005 cache->needs_free_space = 1;
10006 ret = exclude_super_stripes(root, cache);
10007 if (ret) {
10008 /*
10009 * We may have excluded something, so call this just in
10010 * case.
10011 */
10012 free_excluded_extents(root, cache);
10013 btrfs_put_block_group(cache);
10014 return ret;
10015 }
10016
10017 add_new_free_space(cache, root->fs_info, chunk_offset,
10018 chunk_offset + size);
10019
10020 free_excluded_extents(root, cache);
10021
10022#ifdef CONFIG_BTRFS_DEBUG
10023 if (btrfs_should_fragment_free_space(root, cache)) {
10024 u64 new_bytes_used = size - bytes_used;
10025
10026 bytes_used += new_bytes_used >> 1;
10027 fragment_free_space(root, cache);
10028 }
10029#endif
10030 /*
10031 * Call to ensure the corresponding space_info object is created and
10032 * assigned to our block group, but don't update its counters just yet.
10033 * We want our bg to be added to the rbtree with its ->space_info set.
10034 */
10035 ret = update_space_info(root->fs_info, cache->flags, 0, 0,
10036 &cache->space_info);
10037 if (ret) {
10038 btrfs_remove_free_space_cache(cache);
10039 btrfs_put_block_group(cache);
10040 return ret;
10041 }
10042
10043 ret = btrfs_add_block_group_cache(root->fs_info, cache);
10044 if (ret) {
10045 btrfs_remove_free_space_cache(cache);
10046 btrfs_put_block_group(cache);
10047 return ret;
10048 }
10049
10050 /*
10051 * Now that our block group has its ->space_info set and is inserted in
10052 * the rbtree, update the space info's counters.
10053 */
10054 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
10055 &cache->space_info);
10056 if (ret) {
10057 btrfs_remove_free_space_cache(cache);
10058 spin_lock(&root->fs_info->block_group_cache_lock);
10059 rb_erase(&cache->cache_node,
10060 &root->fs_info->block_group_cache_tree);
10061 RB_CLEAR_NODE(&cache->cache_node);
10062 spin_unlock(&root->fs_info->block_group_cache_lock);
10063 btrfs_put_block_group(cache);
10064 return ret;
10065 }
10066 update_global_block_rsv(root->fs_info);
10067
10068 spin_lock(&cache->space_info->lock);
10069 cache->space_info->bytes_readonly += cache->bytes_super;
10070 spin_unlock(&cache->space_info->lock);
10071
10072 __link_block_group(cache->space_info, cache);
10073
10074 list_add_tail(&cache->bg_list, &trans->new_bgs);
10075
10076 set_avail_alloc_bits(extent_root->fs_info, type);
10077
10078 return 0;
10079}
10080
10081static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10082{
10083 u64 extra_flags = chunk_to_extended(flags) &
10084 BTRFS_EXTENDED_PROFILE_MASK;
10085
10086 write_seqlock(&fs_info->profiles_lock);
10087 if (flags & BTRFS_BLOCK_GROUP_DATA)
10088 fs_info->avail_data_alloc_bits &= ~extra_flags;
10089 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10090 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10091 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10092 fs_info->avail_system_alloc_bits &= ~extra_flags;
10093 write_sequnlock(&fs_info->profiles_lock);
10094}
10095
10096int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10097 struct btrfs_root *root, u64 group_start,
10098 struct extent_map *em)
10099{
10100 struct btrfs_path *path;
10101 struct btrfs_block_group_cache *block_group;
10102 struct btrfs_free_cluster *cluster;
10103 struct btrfs_root *tree_root = root->fs_info->tree_root;
10104 struct btrfs_key key;
10105 struct inode *inode;
10106 struct kobject *kobj = NULL;
10107 int ret;
10108 int index;
10109 int factor;
10110 struct btrfs_caching_control *caching_ctl = NULL;
10111 bool remove_em;
10112
10113 root = root->fs_info->extent_root;
10114
10115 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
10116 BUG_ON(!block_group);
10117 BUG_ON(!block_group->ro);
10118
10119 /*
10120 * Free the reserved super bytes from this block group before
10121 * remove it.
10122 */
10123 free_excluded_extents(root, block_group);
10124
10125 memcpy(&key, &block_group->key, sizeof(key));
10126 index = get_block_group_index(block_group);
10127 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10128 BTRFS_BLOCK_GROUP_RAID1 |
10129 BTRFS_BLOCK_GROUP_RAID10))
10130 factor = 2;
10131 else
10132 factor = 1;
10133
10134 /* make sure this block group isn't part of an allocation cluster */
10135 cluster = &root->fs_info->data_alloc_cluster;
10136 spin_lock(&cluster->refill_lock);
10137 btrfs_return_cluster_to_free_space(block_group, cluster);
10138 spin_unlock(&cluster->refill_lock);
10139
10140 /*
10141 * make sure this block group isn't part of a metadata
10142 * allocation cluster
10143 */
10144 cluster = &root->fs_info->meta_alloc_cluster;
10145 spin_lock(&cluster->refill_lock);
10146 btrfs_return_cluster_to_free_space(block_group, cluster);
10147 spin_unlock(&cluster->refill_lock);
10148
10149 path = btrfs_alloc_path();
10150 if (!path) {
10151 ret = -ENOMEM;
10152 goto out;
10153 }
10154
10155 /*
10156 * get the inode first so any iput calls done for the io_list
10157 * aren't the final iput (no unlinks allowed now)
10158 */
10159 inode = lookup_free_space_inode(tree_root, block_group, path);
10160
10161 mutex_lock(&trans->transaction->cache_write_mutex);
10162 /*
10163 * make sure our free spache cache IO is done before remove the
10164 * free space inode
10165 */
10166 spin_lock(&trans->transaction->dirty_bgs_lock);
10167 if (!list_empty(&block_group->io_list)) {
10168 list_del_init(&block_group->io_list);
10169
10170 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10171
10172 spin_unlock(&trans->transaction->dirty_bgs_lock);
10173 btrfs_wait_cache_io(root, trans, block_group,
10174 &block_group->io_ctl, path,
10175 block_group->key.objectid);
10176 btrfs_put_block_group(block_group);
10177 spin_lock(&trans->transaction->dirty_bgs_lock);
10178 }
10179
10180 if (!list_empty(&block_group->dirty_list)) {
10181 list_del_init(&block_group->dirty_list);
10182 btrfs_put_block_group(block_group);
10183 }
10184 spin_unlock(&trans->transaction->dirty_bgs_lock);
10185 mutex_unlock(&trans->transaction->cache_write_mutex);
10186
10187 if (!IS_ERR(inode)) {
10188 ret = btrfs_orphan_add(trans, inode);
10189 if (ret) {
10190 btrfs_add_delayed_iput(inode);
10191 goto out;
10192 }
10193 clear_nlink(inode);
10194 /* One for the block groups ref */
10195 spin_lock(&block_group->lock);
10196 if (block_group->iref) {
10197 block_group->iref = 0;
10198 block_group->inode = NULL;
10199 spin_unlock(&block_group->lock);
10200 iput(inode);
10201 } else {
10202 spin_unlock(&block_group->lock);
10203 }
10204 /* One for our lookup ref */
10205 btrfs_add_delayed_iput(inode);
10206 }
10207
10208 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10209 key.offset = block_group->key.objectid;
10210 key.type = 0;
10211
10212 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10213 if (ret < 0)
10214 goto out;
10215 if (ret > 0)
10216 btrfs_release_path(path);
10217 if (ret == 0) {
10218 ret = btrfs_del_item(trans, tree_root, path);
10219 if (ret)
10220 goto out;
10221 btrfs_release_path(path);
10222 }
10223
10224 spin_lock(&root->fs_info->block_group_cache_lock);
10225 rb_erase(&block_group->cache_node,
10226 &root->fs_info->block_group_cache_tree);
10227 RB_CLEAR_NODE(&block_group->cache_node);
10228
10229 if (root->fs_info->first_logical_byte == block_group->key.objectid)
10230 root->fs_info->first_logical_byte = (u64)-1;
10231 spin_unlock(&root->fs_info->block_group_cache_lock);
10232
10233 down_write(&block_group->space_info->groups_sem);
10234 /*
10235 * we must use list_del_init so people can check to see if they
10236 * are still on the list after taking the semaphore
10237 */
10238 list_del_init(&block_group->list);
10239 if (list_empty(&block_group->space_info->block_groups[index])) {
10240 kobj = block_group->space_info->block_group_kobjs[index];
10241 block_group->space_info->block_group_kobjs[index] = NULL;
10242 clear_avail_alloc_bits(root->fs_info, block_group->flags);
10243 }
10244 up_write(&block_group->space_info->groups_sem);
10245 if (kobj) {
10246 kobject_del(kobj);
10247 kobject_put(kobj);
10248 }
10249
10250 if (block_group->has_caching_ctl)
10251 caching_ctl = get_caching_control(block_group);
10252 if (block_group->cached == BTRFS_CACHE_STARTED)
10253 wait_block_group_cache_done(block_group);
10254 if (block_group->has_caching_ctl) {
10255 down_write(&root->fs_info->commit_root_sem);
10256 if (!caching_ctl) {
10257 struct btrfs_caching_control *ctl;
10258
10259 list_for_each_entry(ctl,
10260 &root->fs_info->caching_block_groups, list)
10261 if (ctl->block_group == block_group) {
10262 caching_ctl = ctl;
10263 atomic_inc(&caching_ctl->count);
10264 break;
10265 }
10266 }
10267 if (caching_ctl)
10268 list_del_init(&caching_ctl->list);
10269 up_write(&root->fs_info->commit_root_sem);
10270 if (caching_ctl) {
10271 /* Once for the caching bgs list and once for us. */
10272 put_caching_control(caching_ctl);
10273 put_caching_control(caching_ctl);
10274 }
10275 }
10276
10277 spin_lock(&trans->transaction->dirty_bgs_lock);
10278 if (!list_empty(&block_group->dirty_list)) {
10279 WARN_ON(1);
10280 }
10281 if (!list_empty(&block_group->io_list)) {
10282 WARN_ON(1);
10283 }
10284 spin_unlock(&trans->transaction->dirty_bgs_lock);
10285 btrfs_remove_free_space_cache(block_group);
10286
10287 spin_lock(&block_group->space_info->lock);
10288 list_del_init(&block_group->ro_list);
10289
10290 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
10291 WARN_ON(block_group->space_info->total_bytes
10292 < block_group->key.offset);
10293 WARN_ON(block_group->space_info->bytes_readonly
10294 < block_group->key.offset);
10295 WARN_ON(block_group->space_info->disk_total
10296 < block_group->key.offset * factor);
10297 }
10298 block_group->space_info->total_bytes -= block_group->key.offset;
10299 block_group->space_info->bytes_readonly -= block_group->key.offset;
10300 block_group->space_info->disk_total -= block_group->key.offset * factor;
10301
10302 spin_unlock(&block_group->space_info->lock);
10303
10304 memcpy(&key, &block_group->key, sizeof(key));
10305
10306 lock_chunks(root);
10307 if (!list_empty(&em->list)) {
10308 /* We're in the transaction->pending_chunks list. */
10309 free_extent_map(em);
10310 }
10311 spin_lock(&block_group->lock);
10312 block_group->removed = 1;
10313 /*
10314 * At this point trimming can't start on this block group, because we
10315 * removed the block group from the tree fs_info->block_group_cache_tree
10316 * so no one can't find it anymore and even if someone already got this
10317 * block group before we removed it from the rbtree, they have already
10318 * incremented block_group->trimming - if they didn't, they won't find
10319 * any free space entries because we already removed them all when we
10320 * called btrfs_remove_free_space_cache().
10321 *
10322 * And we must not remove the extent map from the fs_info->mapping_tree
10323 * to prevent the same logical address range and physical device space
10324 * ranges from being reused for a new block group. This is because our
10325 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10326 * completely transactionless, so while it is trimming a range the
10327 * currently running transaction might finish and a new one start,
10328 * allowing for new block groups to be created that can reuse the same
10329 * physical device locations unless we take this special care.
10330 *
10331 * There may also be an implicit trim operation if the file system
10332 * is mounted with -odiscard. The same protections must remain
10333 * in place until the extents have been discarded completely when
10334 * the transaction commit has completed.
10335 */
10336 remove_em = (atomic_read(&block_group->trimming) == 0);
10337 /*
10338 * Make sure a trimmer task always sees the em in the pinned_chunks list
10339 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10340 * before checking block_group->removed).
10341 */
10342 if (!remove_em) {
10343 /*
10344 * Our em might be in trans->transaction->pending_chunks which
10345 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10346 * and so is the fs_info->pinned_chunks list.
10347 *
10348 * So at this point we must be holding the chunk_mutex to avoid
10349 * any races with chunk allocation (more specifically at
10350 * volumes.c:contains_pending_extent()), to ensure it always
10351 * sees the em, either in the pending_chunks list or in the
10352 * pinned_chunks list.
10353 */
10354 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
10355 }
10356 spin_unlock(&block_group->lock);
10357
10358 if (remove_em) {
10359 struct extent_map_tree *em_tree;
10360
10361 em_tree = &root->fs_info->mapping_tree.map_tree;
10362 write_lock(&em_tree->lock);
10363 /*
10364 * The em might be in the pending_chunks list, so make sure the
10365 * chunk mutex is locked, since remove_extent_mapping() will
10366 * delete us from that list.
10367 */
10368 remove_extent_mapping(em_tree, em);
10369 write_unlock(&em_tree->lock);
10370 /* once for the tree */
10371 free_extent_map(em);
10372 }
10373
10374 unlock_chunks(root);
10375
10376 ret = remove_block_group_free_space(trans, root->fs_info, block_group);
10377 if (ret)
10378 goto out;
10379
10380 btrfs_put_block_group(block_group);
10381 btrfs_put_block_group(block_group);
10382
10383 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10384 if (ret > 0)
10385 ret = -EIO;
10386 if (ret < 0)
10387 goto out;
10388
10389 ret = btrfs_del_item(trans, root, path);
10390out:
10391 btrfs_free_path(path);
10392 return ret;
10393}
10394
10395struct btrfs_trans_handle *
10396btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10397 const u64 chunk_offset)
10398{
10399 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10400 struct extent_map *em;
10401 struct map_lookup *map;
10402 unsigned int num_items;
10403
10404 read_lock(&em_tree->lock);
10405 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10406 read_unlock(&em_tree->lock);
10407 ASSERT(em && em->start == chunk_offset);
10408
10409 /*
10410 * We need to reserve 3 + N units from the metadata space info in order
10411 * to remove a block group (done at btrfs_remove_chunk() and at
10412 * btrfs_remove_block_group()), which are used for:
10413 *
10414 * 1 unit for adding the free space inode's orphan (located in the tree
10415 * of tree roots).
10416 * 1 unit for deleting the block group item (located in the extent
10417 * tree).
10418 * 1 unit for deleting the free space item (located in tree of tree
10419 * roots).
10420 * N units for deleting N device extent items corresponding to each
10421 * stripe (located in the device tree).
10422 *
10423 * In order to remove a block group we also need to reserve units in the
10424 * system space info in order to update the chunk tree (update one or
10425 * more device items and remove one chunk item), but this is done at
10426 * btrfs_remove_chunk() through a call to check_system_chunk().
10427 */
10428 map = em->map_lookup;
10429 num_items = 3 + map->num_stripes;
10430 free_extent_map(em);
10431
10432 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10433 num_items, 1);
10434}
10435
10436/*
10437 * Process the unused_bgs list and remove any that don't have any allocated
10438 * space inside of them.
10439 */
10440void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10441{
10442 struct btrfs_block_group_cache *block_group;
10443 struct btrfs_space_info *space_info;
10444 struct btrfs_root *root = fs_info->extent_root;
10445 struct btrfs_trans_handle *trans;
10446 int ret = 0;
10447
10448 if (!fs_info->open)
10449 return;
10450
10451 spin_lock(&fs_info->unused_bgs_lock);
10452 while (!list_empty(&fs_info->unused_bgs)) {
10453 u64 start, end;
10454 int trimming;
10455
10456 block_group = list_first_entry(&fs_info->unused_bgs,
10457 struct btrfs_block_group_cache,
10458 bg_list);
10459 list_del_init(&block_group->bg_list);
10460
10461 space_info = block_group->space_info;
10462
10463 if (ret || btrfs_mixed_space_info(space_info)) {
10464 btrfs_put_block_group(block_group);
10465 continue;
10466 }
10467 spin_unlock(&fs_info->unused_bgs_lock);
10468
10469 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10470
10471 /* Don't want to race with allocators so take the groups_sem */
10472 down_write(&space_info->groups_sem);
10473 spin_lock(&block_group->lock);
10474 if (block_group->reserved ||
10475 btrfs_block_group_used(&block_group->item) ||
10476 block_group->ro ||
10477 list_is_singular(&block_group->list)) {
10478 /*
10479 * We want to bail if we made new allocations or have
10480 * outstanding allocations in this block group. We do
10481 * the ro check in case balance is currently acting on
10482 * this block group.
10483 */
10484 spin_unlock(&block_group->lock);
10485 up_write(&space_info->groups_sem);
10486 goto next;
10487 }
10488 spin_unlock(&block_group->lock);
10489
10490 /* We don't want to force the issue, only flip if it's ok. */
10491 ret = inc_block_group_ro(block_group, 0);
10492 up_write(&space_info->groups_sem);
10493 if (ret < 0) {
10494 ret = 0;
10495 goto next;
10496 }
10497
10498 /*
10499 * Want to do this before we do anything else so we can recover
10500 * properly if we fail to join the transaction.
10501 */
10502 trans = btrfs_start_trans_remove_block_group(fs_info,
10503 block_group->key.objectid);
10504 if (IS_ERR(trans)) {
10505 btrfs_dec_block_group_ro(root, block_group);
10506 ret = PTR_ERR(trans);
10507 goto next;
10508 }
10509
10510 /*
10511 * We could have pending pinned extents for this block group,
10512 * just delete them, we don't care about them anymore.
10513 */
10514 start = block_group->key.objectid;
10515 end = start + block_group->key.offset - 1;
10516 /*
10517 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10518 * btrfs_finish_extent_commit(). If we are at transaction N,
10519 * another task might be running finish_extent_commit() for the
10520 * previous transaction N - 1, and have seen a range belonging
10521 * to the block group in freed_extents[] before we were able to
10522 * clear the whole block group range from freed_extents[]. This
10523 * means that task can lookup for the block group after we
10524 * unpinned it from freed_extents[] and removed it, leading to
10525 * a BUG_ON() at btrfs_unpin_extent_range().
10526 */
10527 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10528 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10529 EXTENT_DIRTY, GFP_NOFS);
10530 if (ret) {
10531 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10532 btrfs_dec_block_group_ro(root, block_group);
10533 goto end_trans;
10534 }
10535 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10536 EXTENT_DIRTY, GFP_NOFS);
10537 if (ret) {
10538 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10539 btrfs_dec_block_group_ro(root, block_group);
10540 goto end_trans;
10541 }
10542 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10543
10544 /* Reset pinned so btrfs_put_block_group doesn't complain */
10545 spin_lock(&space_info->lock);
10546 spin_lock(&block_group->lock);
10547
10548 space_info->bytes_pinned -= block_group->pinned;
10549 space_info->bytes_readonly += block_group->pinned;
10550 percpu_counter_add(&space_info->total_bytes_pinned,
10551 -block_group->pinned);
10552 block_group->pinned = 0;
10553
10554 spin_unlock(&block_group->lock);
10555 spin_unlock(&space_info->lock);
10556
10557 /* DISCARD can flip during remount */
10558 trimming = btrfs_test_opt(root, DISCARD);
10559
10560 /* Implicit trim during transaction commit. */
10561 if (trimming)
10562 btrfs_get_block_group_trimming(block_group);
10563
10564 /*
10565 * Btrfs_remove_chunk will abort the transaction if things go
10566 * horribly wrong.
10567 */
10568 ret = btrfs_remove_chunk(trans, root,
10569 block_group->key.objectid);
10570
10571 if (ret) {
10572 if (trimming)
10573 btrfs_put_block_group_trimming(block_group);
10574 goto end_trans;
10575 }
10576
10577 /*
10578 * If we're not mounted with -odiscard, we can just forget
10579 * about this block group. Otherwise we'll need to wait
10580 * until transaction commit to do the actual discard.
10581 */
10582 if (trimming) {
10583 spin_lock(&fs_info->unused_bgs_lock);
10584 /*
10585 * A concurrent scrub might have added us to the list
10586 * fs_info->unused_bgs, so use a list_move operation
10587 * to add the block group to the deleted_bgs list.
10588 */
10589 list_move(&block_group->bg_list,
10590 &trans->transaction->deleted_bgs);
10591 spin_unlock(&fs_info->unused_bgs_lock);
10592 btrfs_get_block_group(block_group);
10593 }
10594end_trans:
10595 btrfs_end_transaction(trans, root);
10596next:
10597 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10598 btrfs_put_block_group(block_group);
10599 spin_lock(&fs_info->unused_bgs_lock);
10600 }
10601 spin_unlock(&fs_info->unused_bgs_lock);
10602}
10603
10604int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10605{
10606 struct btrfs_space_info *space_info;
10607 struct btrfs_super_block *disk_super;
10608 u64 features;
10609 u64 flags;
10610 int mixed = 0;
10611 int ret;
10612
10613 disk_super = fs_info->super_copy;
10614 if (!btrfs_super_root(disk_super))
10615 return -EINVAL;
10616
10617 features = btrfs_super_incompat_flags(disk_super);
10618 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10619 mixed = 1;
10620
10621 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10622 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10623 if (ret)
10624 goto out;
10625
10626 if (mixed) {
10627 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10628 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10629 } else {
10630 flags = BTRFS_BLOCK_GROUP_METADATA;
10631 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10632 if (ret)
10633 goto out;
10634
10635 flags = BTRFS_BLOCK_GROUP_DATA;
10636 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10637 }
10638out:
10639 return ret;
10640}
10641
10642int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10643{
10644 return unpin_extent_range(root, start, end, false);
10645}
10646
10647/*
10648 * It used to be that old block groups would be left around forever.
10649 * Iterating over them would be enough to trim unused space. Since we
10650 * now automatically remove them, we also need to iterate over unallocated
10651 * space.
10652 *
10653 * We don't want a transaction for this since the discard may take a
10654 * substantial amount of time. We don't require that a transaction be
10655 * running, but we do need to take a running transaction into account
10656 * to ensure that we're not discarding chunks that were released in
10657 * the current transaction.
10658 *
10659 * Holding the chunks lock will prevent other threads from allocating
10660 * or releasing chunks, but it won't prevent a running transaction
10661 * from committing and releasing the memory that the pending chunks
10662 * list head uses. For that, we need to take a reference to the
10663 * transaction.
10664 */
10665static int btrfs_trim_free_extents(struct btrfs_device *device,
10666 u64 minlen, u64 *trimmed)
10667{
10668 u64 start = 0, len = 0;
10669 int ret;
10670
10671 *trimmed = 0;
10672
10673 /* Not writeable = nothing to do. */
10674 if (!device->writeable)
10675 return 0;
10676
10677 /* No free space = nothing to do. */
10678 if (device->total_bytes <= device->bytes_used)
10679 return 0;
10680
10681 ret = 0;
10682
10683 while (1) {
10684 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
10685 struct btrfs_transaction *trans;
10686 u64 bytes;
10687
10688 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10689 if (ret)
10690 return ret;
10691
10692 down_read(&fs_info->commit_root_sem);
10693
10694 spin_lock(&fs_info->trans_lock);
10695 trans = fs_info->running_transaction;
10696 if (trans)
10697 atomic_inc(&trans->use_count);
10698 spin_unlock(&fs_info->trans_lock);
10699
10700 ret = find_free_dev_extent_start(trans, device, minlen, start,
10701 &start, &len);
10702 if (trans)
10703 btrfs_put_transaction(trans);
10704
10705 if (ret) {
10706 up_read(&fs_info->commit_root_sem);
10707 mutex_unlock(&fs_info->chunk_mutex);
10708 if (ret == -ENOSPC)
10709 ret = 0;
10710 break;
10711 }
10712
10713 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10714 up_read(&fs_info->commit_root_sem);
10715 mutex_unlock(&fs_info->chunk_mutex);
10716
10717 if (ret)
10718 break;
10719
10720 start += len;
10721 *trimmed += bytes;
10722
10723 if (fatal_signal_pending(current)) {
10724 ret = -ERESTARTSYS;
10725 break;
10726 }
10727
10728 cond_resched();
10729 }
10730
10731 return ret;
10732}
10733
10734int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10735{
10736 struct btrfs_fs_info *fs_info = root->fs_info;
10737 struct btrfs_block_group_cache *cache = NULL;
10738 struct btrfs_device *device;
10739 struct list_head *devices;
10740 u64 group_trimmed;
10741 u64 start;
10742 u64 end;
10743 u64 trimmed = 0;
10744 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10745 int ret = 0;
10746
10747 /*
10748 * try to trim all FS space, our block group may start from non-zero.
10749 */
10750 if (range->len == total_bytes)
10751 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10752 else
10753 cache = btrfs_lookup_block_group(fs_info, range->start);
10754
10755 while (cache) {
10756 if (cache->key.objectid >= (range->start + range->len)) {
10757 btrfs_put_block_group(cache);
10758 break;
10759 }
10760
10761 start = max(range->start, cache->key.objectid);
10762 end = min(range->start + range->len,
10763 cache->key.objectid + cache->key.offset);
10764
10765 if (end - start >= range->minlen) {
10766 if (!block_group_cache_done(cache)) {
10767 ret = cache_block_group(cache, 0);
10768 if (ret) {
10769 btrfs_put_block_group(cache);
10770 break;
10771 }
10772 ret = wait_block_group_cache_done(cache);
10773 if (ret) {
10774 btrfs_put_block_group(cache);
10775 break;
10776 }
10777 }
10778 ret = btrfs_trim_block_group(cache,
10779 &group_trimmed,
10780 start,
10781 end,
10782 range->minlen);
10783
10784 trimmed += group_trimmed;
10785 if (ret) {
10786 btrfs_put_block_group(cache);
10787 break;
10788 }
10789 }
10790
10791 cache = next_block_group(fs_info->tree_root, cache);
10792 }
10793
10794 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
10795 devices = &root->fs_info->fs_devices->alloc_list;
10796 list_for_each_entry(device, devices, dev_alloc_list) {
10797 ret = btrfs_trim_free_extents(device, range->minlen,
10798 &group_trimmed);
10799 if (ret)
10800 break;
10801
10802 trimmed += group_trimmed;
10803 }
10804 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
10805
10806 range->len = trimmed;
10807 return ret;
10808}
10809
10810/*
10811 * btrfs_{start,end}_write_no_snapshoting() are similar to
10812 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10813 * data into the page cache through nocow before the subvolume is snapshoted,
10814 * but flush the data into disk after the snapshot creation, or to prevent
10815 * operations while snapshoting is ongoing and that cause the snapshot to be
10816 * inconsistent (writes followed by expanding truncates for example).
10817 */
10818void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10819{
10820 percpu_counter_dec(&root->subv_writers->counter);
10821 /*
10822 * Make sure counter is updated before we wake up waiters.
10823 */
10824 smp_mb();
10825 if (waitqueue_active(&root->subv_writers->wait))
10826 wake_up(&root->subv_writers->wait);
10827}
10828
10829int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10830{
10831 if (atomic_read(&root->will_be_snapshoted))
10832 return 0;
10833
10834 percpu_counter_inc(&root->subv_writers->counter);
10835 /*
10836 * Make sure counter is updated before we check for snapshot creation.
10837 */
10838 smp_mb();
10839 if (atomic_read(&root->will_be_snapshoted)) {
10840 btrfs_end_write_no_snapshoting(root);
10841 return 0;
10842 }
10843 return 1;
10844}
10845
10846static int wait_snapshoting_atomic_t(atomic_t *a)
10847{
10848 schedule();
10849 return 0;
10850}
10851
10852void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
10853{
10854 while (true) {
10855 int ret;
10856
10857 ret = btrfs_start_write_no_snapshoting(root);
10858 if (ret)
10859 break;
10860 wait_on_atomic_t(&root->will_be_snapshoted,
10861 wait_snapshoting_atomic_t,
10862 TASK_UNINTERRUPTIBLE);
10863 }
10864}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6#include <linux/sched.h>
7#include <linux/sched/signal.h>
8#include <linux/pagemap.h>
9#include <linux/writeback.h>
10#include <linux/blkdev.h>
11#include <linux/sort.h>
12#include <linux/rcupdate.h>
13#include <linux/kthread.h>
14#include <linux/slab.h>
15#include <linux/ratelimit.h>
16#include <linux/percpu_counter.h>
17#include <linux/lockdep.h>
18#include <linux/crc32c.h>
19#include "ctree.h"
20#include "extent-tree.h"
21#include "transaction.h"
22#include "disk-io.h"
23#include "print-tree.h"
24#include "volumes.h"
25#include "raid56.h"
26#include "locking.h"
27#include "free-space-cache.h"
28#include "free-space-tree.h"
29#include "qgroup.h"
30#include "ref-verify.h"
31#include "space-info.h"
32#include "block-rsv.h"
33#include "discard.h"
34#include "zoned.h"
35#include "dev-replace.h"
36#include "fs.h"
37#include "accessors.h"
38#include "root-tree.h"
39#include "file-item.h"
40#include "orphan.h"
41#include "tree-checker.h"
42#include "raid-stripe-tree.h"
43
44#undef SCRAMBLE_DELAYED_REFS
45
46
47static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
48 struct btrfs_delayed_ref_head *href,
49 struct btrfs_delayed_ref_node *node,
50 struct btrfs_delayed_extent_op *extra_op);
51static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
52 struct extent_buffer *leaf,
53 struct btrfs_extent_item *ei);
54static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
55 u64 parent, u64 root_objectid,
56 u64 flags, u64 owner, u64 offset,
57 struct btrfs_key *ins, int ref_mod, u64 oref_root);
58static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
59 struct btrfs_delayed_ref_node *node,
60 struct btrfs_delayed_extent_op *extent_op);
61static int find_next_key(struct btrfs_path *path, int level,
62 struct btrfs_key *key);
63
64static int block_group_bits(struct btrfs_block_group *cache, u64 bits)
65{
66 return (cache->flags & bits) == bits;
67}
68
69/* simple helper to search for an existing data extent at a given offset */
70int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
71{
72 struct btrfs_root *root = btrfs_extent_root(fs_info, start);
73 int ret;
74 struct btrfs_key key;
75 struct btrfs_path *path;
76
77 path = btrfs_alloc_path();
78 if (!path)
79 return -ENOMEM;
80
81 key.objectid = start;
82 key.offset = len;
83 key.type = BTRFS_EXTENT_ITEM_KEY;
84 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
85 btrfs_free_path(path);
86 return ret;
87}
88
89/*
90 * helper function to lookup reference count and flags of a tree block.
91 *
92 * the head node for delayed ref is used to store the sum of all the
93 * reference count modifications queued up in the rbtree. the head
94 * node may also store the extent flags to set. This way you can check
95 * to see what the reference count and extent flags would be if all of
96 * the delayed refs are not processed.
97 */
98int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
99 struct btrfs_fs_info *fs_info, u64 bytenr,
100 u64 offset, int metadata, u64 *refs, u64 *flags,
101 u64 *owning_root)
102{
103 struct btrfs_root *extent_root;
104 struct btrfs_delayed_ref_head *head;
105 struct btrfs_delayed_ref_root *delayed_refs;
106 struct btrfs_path *path;
107 struct btrfs_key key;
108 u64 num_refs;
109 u64 extent_flags;
110 u64 owner = 0;
111 int ret;
112
113 /*
114 * If we don't have skinny metadata, don't bother doing anything
115 * different
116 */
117 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
118 offset = fs_info->nodesize;
119 metadata = 0;
120 }
121
122 path = btrfs_alloc_path();
123 if (!path)
124 return -ENOMEM;
125
126search_again:
127 key.objectid = bytenr;
128 key.offset = offset;
129 if (metadata)
130 key.type = BTRFS_METADATA_ITEM_KEY;
131 else
132 key.type = BTRFS_EXTENT_ITEM_KEY;
133
134 extent_root = btrfs_extent_root(fs_info, bytenr);
135 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
136 if (ret < 0)
137 goto out_free;
138
139 if (ret > 0 && key.type == BTRFS_METADATA_ITEM_KEY) {
140 if (path->slots[0]) {
141 path->slots[0]--;
142 btrfs_item_key_to_cpu(path->nodes[0], &key,
143 path->slots[0]);
144 if (key.objectid == bytenr &&
145 key.type == BTRFS_EXTENT_ITEM_KEY &&
146 key.offset == fs_info->nodesize)
147 ret = 0;
148 }
149 }
150
151 if (ret == 0) {
152 struct extent_buffer *leaf = path->nodes[0];
153 struct btrfs_extent_item *ei;
154 const u32 item_size = btrfs_item_size(leaf, path->slots[0]);
155
156 if (unlikely(item_size < sizeof(*ei))) {
157 ret = -EUCLEAN;
158 btrfs_err(fs_info,
159 "unexpected extent item size, has %u expect >= %zu",
160 item_size, sizeof(*ei));
161 btrfs_abort_transaction(trans, ret);
162 goto out_free;
163 }
164
165 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
166 num_refs = btrfs_extent_refs(leaf, ei);
167 if (unlikely(num_refs == 0)) {
168 ret = -EUCLEAN;
169 btrfs_err(fs_info,
170 "unexpected zero reference count for extent item (%llu %u %llu)",
171 key.objectid, key.type, key.offset);
172 btrfs_abort_transaction(trans, ret);
173 goto out_free;
174 }
175 extent_flags = btrfs_extent_flags(leaf, ei);
176 owner = btrfs_get_extent_owner_root(fs_info, leaf, path->slots[0]);
177 } else {
178 num_refs = 0;
179 extent_flags = 0;
180 ret = 0;
181 }
182
183 delayed_refs = &trans->transaction->delayed_refs;
184 spin_lock(&delayed_refs->lock);
185 head = btrfs_find_delayed_ref_head(fs_info, delayed_refs, bytenr);
186 if (head) {
187 if (!mutex_trylock(&head->mutex)) {
188 refcount_inc(&head->refs);
189 spin_unlock(&delayed_refs->lock);
190
191 btrfs_release_path(path);
192
193 /*
194 * Mutex was contended, block until it's released and try
195 * again
196 */
197 mutex_lock(&head->mutex);
198 mutex_unlock(&head->mutex);
199 btrfs_put_delayed_ref_head(head);
200 goto search_again;
201 }
202 spin_lock(&head->lock);
203 if (head->extent_op && head->extent_op->update_flags)
204 extent_flags |= head->extent_op->flags_to_set;
205
206 num_refs += head->ref_mod;
207 spin_unlock(&head->lock);
208 mutex_unlock(&head->mutex);
209 }
210 spin_unlock(&delayed_refs->lock);
211
212 WARN_ON(num_refs == 0);
213 if (refs)
214 *refs = num_refs;
215 if (flags)
216 *flags = extent_flags;
217 if (owning_root)
218 *owning_root = owner;
219out_free:
220 btrfs_free_path(path);
221 return ret;
222}
223
224/*
225 * Back reference rules. Back refs have three main goals:
226 *
227 * 1) differentiate between all holders of references to an extent so that
228 * when a reference is dropped we can make sure it was a valid reference
229 * before freeing the extent.
230 *
231 * 2) Provide enough information to quickly find the holders of an extent
232 * if we notice a given block is corrupted or bad.
233 *
234 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
235 * maintenance. This is actually the same as #2, but with a slightly
236 * different use case.
237 *
238 * There are two kinds of back refs. The implicit back refs is optimized
239 * for pointers in non-shared tree blocks. For a given pointer in a block,
240 * back refs of this kind provide information about the block's owner tree
241 * and the pointer's key. These information allow us to find the block by
242 * b-tree searching. The full back refs is for pointers in tree blocks not
243 * referenced by their owner trees. The location of tree block is recorded
244 * in the back refs. Actually the full back refs is generic, and can be
245 * used in all cases the implicit back refs is used. The major shortcoming
246 * of the full back refs is its overhead. Every time a tree block gets
247 * COWed, we have to update back refs entry for all pointers in it.
248 *
249 * For a newly allocated tree block, we use implicit back refs for
250 * pointers in it. This means most tree related operations only involve
251 * implicit back refs. For a tree block created in old transaction, the
252 * only way to drop a reference to it is COW it. So we can detect the
253 * event that tree block loses its owner tree's reference and do the
254 * back refs conversion.
255 *
256 * When a tree block is COWed through a tree, there are four cases:
257 *
258 * The reference count of the block is one and the tree is the block's
259 * owner tree. Nothing to do in this case.
260 *
261 * The reference count of the block is one and the tree is not the
262 * block's owner tree. In this case, full back refs is used for pointers
263 * in the block. Remove these full back refs, add implicit back refs for
264 * every pointers in the new block.
265 *
266 * The reference count of the block is greater than one and the tree is
267 * the block's owner tree. In this case, implicit back refs is used for
268 * pointers in the block. Add full back refs for every pointers in the
269 * block, increase lower level extents' reference counts. The original
270 * implicit back refs are entailed to the new block.
271 *
272 * The reference count of the block is greater than one and the tree is
273 * not the block's owner tree. Add implicit back refs for every pointer in
274 * the new block, increase lower level extents' reference count.
275 *
276 * Back Reference Key composing:
277 *
278 * The key objectid corresponds to the first byte in the extent,
279 * The key type is used to differentiate between types of back refs.
280 * There are different meanings of the key offset for different types
281 * of back refs.
282 *
283 * File extents can be referenced by:
284 *
285 * - multiple snapshots, subvolumes, or different generations in one subvol
286 * - different files inside a single subvolume
287 * - different offsets inside a file (bookend extents in file.c)
288 *
289 * The extent ref structure for the implicit back refs has fields for:
290 *
291 * - Objectid of the subvolume root
292 * - objectid of the file holding the reference
293 * - original offset in the file
294 * - how many bookend extents
295 *
296 * The key offset for the implicit back refs is hash of the first
297 * three fields.
298 *
299 * The extent ref structure for the full back refs has field for:
300 *
301 * - number of pointers in the tree leaf
302 *
303 * The key offset for the implicit back refs is the first byte of
304 * the tree leaf
305 *
306 * When a file extent is allocated, The implicit back refs is used.
307 * the fields are filled in:
308 *
309 * (root_key.objectid, inode objectid, offset in file, 1)
310 *
311 * When a file extent is removed file truncation, we find the
312 * corresponding implicit back refs and check the following fields:
313 *
314 * (btrfs_header_owner(leaf), inode objectid, offset in file)
315 *
316 * Btree extents can be referenced by:
317 *
318 * - Different subvolumes
319 *
320 * Both the implicit back refs and the full back refs for tree blocks
321 * only consist of key. The key offset for the implicit back refs is
322 * objectid of block's owner tree. The key offset for the full back refs
323 * is the first byte of parent block.
324 *
325 * When implicit back refs is used, information about the lowest key and
326 * level of the tree block are required. These information are stored in
327 * tree block info structure.
328 */
329
330/*
331 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
332 * is_data == BTRFS_REF_TYPE_DATA, data type is requiried,
333 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
334 */
335int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
336 struct btrfs_extent_inline_ref *iref,
337 enum btrfs_inline_ref_type is_data)
338{
339 struct btrfs_fs_info *fs_info = eb->fs_info;
340 int type = btrfs_extent_inline_ref_type(eb, iref);
341 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
342
343 if (type == BTRFS_EXTENT_OWNER_REF_KEY) {
344 ASSERT(btrfs_fs_incompat(fs_info, SIMPLE_QUOTA));
345 return type;
346 }
347
348 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
349 type == BTRFS_SHARED_BLOCK_REF_KEY ||
350 type == BTRFS_SHARED_DATA_REF_KEY ||
351 type == BTRFS_EXTENT_DATA_REF_KEY) {
352 if (is_data == BTRFS_REF_TYPE_BLOCK) {
353 if (type == BTRFS_TREE_BLOCK_REF_KEY)
354 return type;
355 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
356 ASSERT(fs_info);
357 /*
358 * Every shared one has parent tree block,
359 * which must be aligned to sector size.
360 */
361 if (offset && IS_ALIGNED(offset, fs_info->sectorsize))
362 return type;
363 }
364 } else if (is_data == BTRFS_REF_TYPE_DATA) {
365 if (type == BTRFS_EXTENT_DATA_REF_KEY)
366 return type;
367 if (type == BTRFS_SHARED_DATA_REF_KEY) {
368 ASSERT(fs_info);
369 /*
370 * Every shared one has parent tree block,
371 * which must be aligned to sector size.
372 */
373 if (offset &&
374 IS_ALIGNED(offset, fs_info->sectorsize))
375 return type;
376 }
377 } else {
378 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
379 return type;
380 }
381 }
382
383 WARN_ON(1);
384 btrfs_print_leaf(eb);
385 btrfs_err(fs_info,
386 "eb %llu iref 0x%lx invalid extent inline ref type %d",
387 eb->start, (unsigned long)iref, type);
388
389 return BTRFS_REF_TYPE_INVALID;
390}
391
392u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
393{
394 u32 high_crc = ~(u32)0;
395 u32 low_crc = ~(u32)0;
396 __le64 lenum;
397
398 lenum = cpu_to_le64(root_objectid);
399 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
400 lenum = cpu_to_le64(owner);
401 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
402 lenum = cpu_to_le64(offset);
403 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
404
405 return ((u64)high_crc << 31) ^ (u64)low_crc;
406}
407
408static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
409 struct btrfs_extent_data_ref *ref)
410{
411 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
412 btrfs_extent_data_ref_objectid(leaf, ref),
413 btrfs_extent_data_ref_offset(leaf, ref));
414}
415
416static int match_extent_data_ref(struct extent_buffer *leaf,
417 struct btrfs_extent_data_ref *ref,
418 u64 root_objectid, u64 owner, u64 offset)
419{
420 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
421 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
422 btrfs_extent_data_ref_offset(leaf, ref) != offset)
423 return 0;
424 return 1;
425}
426
427static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
428 struct btrfs_path *path,
429 u64 bytenr, u64 parent,
430 u64 root_objectid,
431 u64 owner, u64 offset)
432{
433 struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
434 struct btrfs_key key;
435 struct btrfs_extent_data_ref *ref;
436 struct extent_buffer *leaf;
437 u32 nritems;
438 int recow;
439 int ret;
440
441 key.objectid = bytenr;
442 if (parent) {
443 key.type = BTRFS_SHARED_DATA_REF_KEY;
444 key.offset = parent;
445 } else {
446 key.type = BTRFS_EXTENT_DATA_REF_KEY;
447 key.offset = hash_extent_data_ref(root_objectid,
448 owner, offset);
449 }
450again:
451 recow = 0;
452 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
453 if (ret < 0)
454 return ret;
455
456 if (parent) {
457 if (ret)
458 return -ENOENT;
459 return 0;
460 }
461
462 ret = -ENOENT;
463 leaf = path->nodes[0];
464 nritems = btrfs_header_nritems(leaf);
465 while (1) {
466 if (path->slots[0] >= nritems) {
467 ret = btrfs_next_leaf(root, path);
468 if (ret) {
469 if (ret > 0)
470 return -ENOENT;
471 return ret;
472 }
473
474 leaf = path->nodes[0];
475 nritems = btrfs_header_nritems(leaf);
476 recow = 1;
477 }
478
479 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
480 if (key.objectid != bytenr ||
481 key.type != BTRFS_EXTENT_DATA_REF_KEY)
482 goto fail;
483
484 ref = btrfs_item_ptr(leaf, path->slots[0],
485 struct btrfs_extent_data_ref);
486
487 if (match_extent_data_ref(leaf, ref, root_objectid,
488 owner, offset)) {
489 if (recow) {
490 btrfs_release_path(path);
491 goto again;
492 }
493 ret = 0;
494 break;
495 }
496 path->slots[0]++;
497 }
498fail:
499 return ret;
500}
501
502static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
503 struct btrfs_path *path,
504 struct btrfs_delayed_ref_node *node,
505 u64 bytenr)
506{
507 struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
508 struct btrfs_key key;
509 struct extent_buffer *leaf;
510 u64 owner = btrfs_delayed_ref_owner(node);
511 u64 offset = btrfs_delayed_ref_offset(node);
512 u32 size;
513 u32 num_refs;
514 int ret;
515
516 key.objectid = bytenr;
517 if (node->parent) {
518 key.type = BTRFS_SHARED_DATA_REF_KEY;
519 key.offset = node->parent;
520 size = sizeof(struct btrfs_shared_data_ref);
521 } else {
522 key.type = BTRFS_EXTENT_DATA_REF_KEY;
523 key.offset = hash_extent_data_ref(node->ref_root, owner, offset);
524 size = sizeof(struct btrfs_extent_data_ref);
525 }
526
527 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
528 if (ret && ret != -EEXIST)
529 goto fail;
530
531 leaf = path->nodes[0];
532 if (node->parent) {
533 struct btrfs_shared_data_ref *ref;
534 ref = btrfs_item_ptr(leaf, path->slots[0],
535 struct btrfs_shared_data_ref);
536 if (ret == 0) {
537 btrfs_set_shared_data_ref_count(leaf, ref, node->ref_mod);
538 } else {
539 num_refs = btrfs_shared_data_ref_count(leaf, ref);
540 num_refs += node->ref_mod;
541 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
542 }
543 } else {
544 struct btrfs_extent_data_ref *ref;
545 while (ret == -EEXIST) {
546 ref = btrfs_item_ptr(leaf, path->slots[0],
547 struct btrfs_extent_data_ref);
548 if (match_extent_data_ref(leaf, ref, node->ref_root,
549 owner, offset))
550 break;
551 btrfs_release_path(path);
552 key.offset++;
553 ret = btrfs_insert_empty_item(trans, root, path, &key,
554 size);
555 if (ret && ret != -EEXIST)
556 goto fail;
557
558 leaf = path->nodes[0];
559 }
560 ref = btrfs_item_ptr(leaf, path->slots[0],
561 struct btrfs_extent_data_ref);
562 if (ret == 0) {
563 btrfs_set_extent_data_ref_root(leaf, ref, node->ref_root);
564 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
565 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
566 btrfs_set_extent_data_ref_count(leaf, ref, node->ref_mod);
567 } else {
568 num_refs = btrfs_extent_data_ref_count(leaf, ref);
569 num_refs += node->ref_mod;
570 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
571 }
572 }
573 btrfs_mark_buffer_dirty(trans, leaf);
574 ret = 0;
575fail:
576 btrfs_release_path(path);
577 return ret;
578}
579
580static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
581 struct btrfs_root *root,
582 struct btrfs_path *path,
583 int refs_to_drop)
584{
585 struct btrfs_key key;
586 struct btrfs_extent_data_ref *ref1 = NULL;
587 struct btrfs_shared_data_ref *ref2 = NULL;
588 struct extent_buffer *leaf;
589 u32 num_refs = 0;
590 int ret = 0;
591
592 leaf = path->nodes[0];
593 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
594
595 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
596 ref1 = btrfs_item_ptr(leaf, path->slots[0],
597 struct btrfs_extent_data_ref);
598 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
599 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
600 ref2 = btrfs_item_ptr(leaf, path->slots[0],
601 struct btrfs_shared_data_ref);
602 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
603 } else {
604 btrfs_err(trans->fs_info,
605 "unrecognized backref key (%llu %u %llu)",
606 key.objectid, key.type, key.offset);
607 btrfs_abort_transaction(trans, -EUCLEAN);
608 return -EUCLEAN;
609 }
610
611 BUG_ON(num_refs < refs_to_drop);
612 num_refs -= refs_to_drop;
613
614 if (num_refs == 0) {
615 ret = btrfs_del_item(trans, root, path);
616 } else {
617 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
618 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
619 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
620 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
621 btrfs_mark_buffer_dirty(trans, leaf);
622 }
623 return ret;
624}
625
626static noinline u32 extent_data_ref_count(struct btrfs_path *path,
627 struct btrfs_extent_inline_ref *iref)
628{
629 struct btrfs_key key;
630 struct extent_buffer *leaf;
631 struct btrfs_extent_data_ref *ref1;
632 struct btrfs_shared_data_ref *ref2;
633 u32 num_refs = 0;
634 int type;
635
636 leaf = path->nodes[0];
637 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
638
639 if (iref) {
640 /*
641 * If type is invalid, we should have bailed out earlier than
642 * this call.
643 */
644 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
645 ASSERT(type != BTRFS_REF_TYPE_INVALID);
646 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
647 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
648 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
649 } else {
650 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
651 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
652 }
653 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
654 ref1 = btrfs_item_ptr(leaf, path->slots[0],
655 struct btrfs_extent_data_ref);
656 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
657 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
658 ref2 = btrfs_item_ptr(leaf, path->slots[0],
659 struct btrfs_shared_data_ref);
660 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
661 } else {
662 WARN_ON(1);
663 }
664 return num_refs;
665}
666
667static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
668 struct btrfs_path *path,
669 u64 bytenr, u64 parent,
670 u64 root_objectid)
671{
672 struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
673 struct btrfs_key key;
674 int ret;
675
676 key.objectid = bytenr;
677 if (parent) {
678 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
679 key.offset = parent;
680 } else {
681 key.type = BTRFS_TREE_BLOCK_REF_KEY;
682 key.offset = root_objectid;
683 }
684
685 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
686 if (ret > 0)
687 ret = -ENOENT;
688 return ret;
689}
690
691static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
692 struct btrfs_path *path,
693 struct btrfs_delayed_ref_node *node,
694 u64 bytenr)
695{
696 struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
697 struct btrfs_key key;
698 int ret;
699
700 key.objectid = bytenr;
701 if (node->parent) {
702 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
703 key.offset = node->parent;
704 } else {
705 key.type = BTRFS_TREE_BLOCK_REF_KEY;
706 key.offset = node->ref_root;
707 }
708
709 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
710 btrfs_release_path(path);
711 return ret;
712}
713
714static inline int extent_ref_type(u64 parent, u64 owner)
715{
716 int type;
717 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
718 if (parent > 0)
719 type = BTRFS_SHARED_BLOCK_REF_KEY;
720 else
721 type = BTRFS_TREE_BLOCK_REF_KEY;
722 } else {
723 if (parent > 0)
724 type = BTRFS_SHARED_DATA_REF_KEY;
725 else
726 type = BTRFS_EXTENT_DATA_REF_KEY;
727 }
728 return type;
729}
730
731static int find_next_key(struct btrfs_path *path, int level,
732 struct btrfs_key *key)
733
734{
735 for (; level < BTRFS_MAX_LEVEL; level++) {
736 if (!path->nodes[level])
737 break;
738 if (path->slots[level] + 1 >=
739 btrfs_header_nritems(path->nodes[level]))
740 continue;
741 if (level == 0)
742 btrfs_item_key_to_cpu(path->nodes[level], key,
743 path->slots[level] + 1);
744 else
745 btrfs_node_key_to_cpu(path->nodes[level], key,
746 path->slots[level] + 1);
747 return 0;
748 }
749 return 1;
750}
751
752/*
753 * look for inline back ref. if back ref is found, *ref_ret is set
754 * to the address of inline back ref, and 0 is returned.
755 *
756 * if back ref isn't found, *ref_ret is set to the address where it
757 * should be inserted, and -ENOENT is returned.
758 *
759 * if insert is true and there are too many inline back refs, the path
760 * points to the extent item, and -EAGAIN is returned.
761 *
762 * NOTE: inline back refs are ordered in the same way that back ref
763 * items in the tree are ordered.
764 */
765static noinline_for_stack
766int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
767 struct btrfs_path *path,
768 struct btrfs_extent_inline_ref **ref_ret,
769 u64 bytenr, u64 num_bytes,
770 u64 parent, u64 root_objectid,
771 u64 owner, u64 offset, int insert)
772{
773 struct btrfs_fs_info *fs_info = trans->fs_info;
774 struct btrfs_root *root = btrfs_extent_root(fs_info, bytenr);
775 struct btrfs_key key;
776 struct extent_buffer *leaf;
777 struct btrfs_extent_item *ei;
778 struct btrfs_extent_inline_ref *iref;
779 u64 flags;
780 u64 item_size;
781 unsigned long ptr;
782 unsigned long end;
783 int extra_size;
784 int type;
785 int want;
786 int ret;
787 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
788 int needed;
789
790 key.objectid = bytenr;
791 key.type = BTRFS_EXTENT_ITEM_KEY;
792 key.offset = num_bytes;
793
794 want = extent_ref_type(parent, owner);
795 if (insert) {
796 extra_size = btrfs_extent_inline_ref_size(want);
797 path->search_for_extension = 1;
798 } else
799 extra_size = -1;
800
801 /*
802 * Owner is our level, so we can just add one to get the level for the
803 * block we are interested in.
804 */
805 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
806 key.type = BTRFS_METADATA_ITEM_KEY;
807 key.offset = owner;
808 }
809
810again:
811 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
812 if (ret < 0)
813 goto out;
814
815 /*
816 * We may be a newly converted file system which still has the old fat
817 * extent entries for metadata, so try and see if we have one of those.
818 */
819 if (ret > 0 && skinny_metadata) {
820 skinny_metadata = false;
821 if (path->slots[0]) {
822 path->slots[0]--;
823 btrfs_item_key_to_cpu(path->nodes[0], &key,
824 path->slots[0]);
825 if (key.objectid == bytenr &&
826 key.type == BTRFS_EXTENT_ITEM_KEY &&
827 key.offset == num_bytes)
828 ret = 0;
829 }
830 if (ret) {
831 key.objectid = bytenr;
832 key.type = BTRFS_EXTENT_ITEM_KEY;
833 key.offset = num_bytes;
834 btrfs_release_path(path);
835 goto again;
836 }
837 }
838
839 if (ret && !insert) {
840 ret = -ENOENT;
841 goto out;
842 } else if (WARN_ON(ret)) {
843 btrfs_print_leaf(path->nodes[0]);
844 btrfs_err(fs_info,
845"extent item not found for insert, bytenr %llu num_bytes %llu parent %llu root_objectid %llu owner %llu offset %llu",
846 bytenr, num_bytes, parent, root_objectid, owner,
847 offset);
848 ret = -EUCLEAN;
849 goto out;
850 }
851
852 leaf = path->nodes[0];
853 item_size = btrfs_item_size(leaf, path->slots[0]);
854 if (unlikely(item_size < sizeof(*ei))) {
855 ret = -EUCLEAN;
856 btrfs_err(fs_info,
857 "unexpected extent item size, has %llu expect >= %zu",
858 item_size, sizeof(*ei));
859 btrfs_abort_transaction(trans, ret);
860 goto out;
861 }
862
863 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
864 flags = btrfs_extent_flags(leaf, ei);
865
866 ptr = (unsigned long)(ei + 1);
867 end = (unsigned long)ei + item_size;
868
869 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
870 ptr += sizeof(struct btrfs_tree_block_info);
871 BUG_ON(ptr > end);
872 }
873
874 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
875 needed = BTRFS_REF_TYPE_DATA;
876 else
877 needed = BTRFS_REF_TYPE_BLOCK;
878
879 ret = -ENOENT;
880 while (ptr < end) {
881 iref = (struct btrfs_extent_inline_ref *)ptr;
882 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
883 if (type == BTRFS_EXTENT_OWNER_REF_KEY) {
884 ASSERT(btrfs_fs_incompat(fs_info, SIMPLE_QUOTA));
885 ptr += btrfs_extent_inline_ref_size(type);
886 continue;
887 }
888 if (type == BTRFS_REF_TYPE_INVALID) {
889 ret = -EUCLEAN;
890 goto out;
891 }
892
893 if (want < type)
894 break;
895 if (want > type) {
896 ptr += btrfs_extent_inline_ref_size(type);
897 continue;
898 }
899
900 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
901 struct btrfs_extent_data_ref *dref;
902 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
903 if (match_extent_data_ref(leaf, dref, root_objectid,
904 owner, offset)) {
905 ret = 0;
906 break;
907 }
908 if (hash_extent_data_ref_item(leaf, dref) <
909 hash_extent_data_ref(root_objectid, owner, offset))
910 break;
911 } else {
912 u64 ref_offset;
913 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
914 if (parent > 0) {
915 if (parent == ref_offset) {
916 ret = 0;
917 break;
918 }
919 if (ref_offset < parent)
920 break;
921 } else {
922 if (root_objectid == ref_offset) {
923 ret = 0;
924 break;
925 }
926 if (ref_offset < root_objectid)
927 break;
928 }
929 }
930 ptr += btrfs_extent_inline_ref_size(type);
931 }
932
933 if (unlikely(ptr > end)) {
934 ret = -EUCLEAN;
935 btrfs_print_leaf(path->nodes[0]);
936 btrfs_crit(fs_info,
937"overrun extent record at slot %d while looking for inline extent for root %llu owner %llu offset %llu parent %llu",
938 path->slots[0], root_objectid, owner, offset, parent);
939 goto out;
940 }
941
942 if (ret == -ENOENT && insert) {
943 if (item_size + extra_size >=
944 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
945 ret = -EAGAIN;
946 goto out;
947 }
948
949 if (path->slots[0] + 1 < btrfs_header_nritems(path->nodes[0])) {
950 struct btrfs_key tmp_key;
951
952 btrfs_item_key_to_cpu(path->nodes[0], &tmp_key, path->slots[0] + 1);
953 if (tmp_key.objectid == bytenr &&
954 tmp_key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
955 ret = -EAGAIN;
956 goto out;
957 }
958 goto out_no_entry;
959 }
960
961 if (!path->keep_locks) {
962 btrfs_release_path(path);
963 path->keep_locks = 1;
964 goto again;
965 }
966
967 /*
968 * To add new inline back ref, we have to make sure
969 * there is no corresponding back ref item.
970 * For simplicity, we just do not add new inline back
971 * ref if there is any kind of item for this block
972 */
973 if (find_next_key(path, 0, &key) == 0 &&
974 key.objectid == bytenr &&
975 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
976 ret = -EAGAIN;
977 goto out;
978 }
979 }
980out_no_entry:
981 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
982out:
983 if (path->keep_locks) {
984 path->keep_locks = 0;
985 btrfs_unlock_up_safe(path, 1);
986 }
987 if (insert)
988 path->search_for_extension = 0;
989 return ret;
990}
991
992/*
993 * helper to add new inline back ref
994 */
995static noinline_for_stack
996void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
997 struct btrfs_path *path,
998 struct btrfs_extent_inline_ref *iref,
999 u64 parent, u64 root_objectid,
1000 u64 owner, u64 offset, int refs_to_add,
1001 struct btrfs_delayed_extent_op *extent_op)
1002{
1003 struct extent_buffer *leaf;
1004 struct btrfs_extent_item *ei;
1005 unsigned long ptr;
1006 unsigned long end;
1007 unsigned long item_offset;
1008 u64 refs;
1009 int size;
1010 int type;
1011
1012 leaf = path->nodes[0];
1013 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1014 item_offset = (unsigned long)iref - (unsigned long)ei;
1015
1016 type = extent_ref_type(parent, owner);
1017 size = btrfs_extent_inline_ref_size(type);
1018
1019 btrfs_extend_item(trans, path, size);
1020
1021 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1022 refs = btrfs_extent_refs(leaf, ei);
1023 refs += refs_to_add;
1024 btrfs_set_extent_refs(leaf, ei, refs);
1025 if (extent_op)
1026 __run_delayed_extent_op(extent_op, leaf, ei);
1027
1028 ptr = (unsigned long)ei + item_offset;
1029 end = (unsigned long)ei + btrfs_item_size(leaf, path->slots[0]);
1030 if (ptr < end - size)
1031 memmove_extent_buffer(leaf, ptr + size, ptr,
1032 end - size - ptr);
1033
1034 iref = (struct btrfs_extent_inline_ref *)ptr;
1035 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1036 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1037 struct btrfs_extent_data_ref *dref;
1038 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1039 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1040 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1041 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1042 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1043 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1044 struct btrfs_shared_data_ref *sref;
1045 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1046 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1047 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1048 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1049 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1050 } else {
1051 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1052 }
1053 btrfs_mark_buffer_dirty(trans, leaf);
1054}
1055
1056static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1057 struct btrfs_path *path,
1058 struct btrfs_extent_inline_ref **ref_ret,
1059 u64 bytenr, u64 num_bytes, u64 parent,
1060 u64 root_objectid, u64 owner, u64 offset)
1061{
1062 int ret;
1063
1064 ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr,
1065 num_bytes, parent, root_objectid,
1066 owner, offset, 0);
1067 if (ret != -ENOENT)
1068 return ret;
1069
1070 btrfs_release_path(path);
1071 *ref_ret = NULL;
1072
1073 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1074 ret = lookup_tree_block_ref(trans, path, bytenr, parent,
1075 root_objectid);
1076 } else {
1077 ret = lookup_extent_data_ref(trans, path, bytenr, parent,
1078 root_objectid, owner, offset);
1079 }
1080 return ret;
1081}
1082
1083/*
1084 * helper to update/remove inline back ref
1085 */
1086static noinline_for_stack int update_inline_extent_backref(
1087 struct btrfs_trans_handle *trans,
1088 struct btrfs_path *path,
1089 struct btrfs_extent_inline_ref *iref,
1090 int refs_to_mod,
1091 struct btrfs_delayed_extent_op *extent_op)
1092{
1093 struct extent_buffer *leaf = path->nodes[0];
1094 struct btrfs_fs_info *fs_info = leaf->fs_info;
1095 struct btrfs_extent_item *ei;
1096 struct btrfs_extent_data_ref *dref = NULL;
1097 struct btrfs_shared_data_ref *sref = NULL;
1098 unsigned long ptr;
1099 unsigned long end;
1100 u32 item_size;
1101 int size;
1102 int type;
1103 u64 refs;
1104
1105 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1106 refs = btrfs_extent_refs(leaf, ei);
1107 if (unlikely(refs_to_mod < 0 && refs + refs_to_mod <= 0)) {
1108 struct btrfs_key key;
1109 u32 extent_size;
1110
1111 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1112 if (key.type == BTRFS_METADATA_ITEM_KEY)
1113 extent_size = fs_info->nodesize;
1114 else
1115 extent_size = key.offset;
1116 btrfs_print_leaf(leaf);
1117 btrfs_err(fs_info,
1118 "invalid refs_to_mod for extent %llu num_bytes %u, has %d expect >= -%llu",
1119 key.objectid, extent_size, refs_to_mod, refs);
1120 return -EUCLEAN;
1121 }
1122 refs += refs_to_mod;
1123 btrfs_set_extent_refs(leaf, ei, refs);
1124 if (extent_op)
1125 __run_delayed_extent_op(extent_op, leaf, ei);
1126
1127 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1128 /*
1129 * Function btrfs_get_extent_inline_ref_type() has already printed
1130 * error messages.
1131 */
1132 if (unlikely(type == BTRFS_REF_TYPE_INVALID))
1133 return -EUCLEAN;
1134
1135 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1136 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1137 refs = btrfs_extent_data_ref_count(leaf, dref);
1138 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1139 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1140 refs = btrfs_shared_data_ref_count(leaf, sref);
1141 } else {
1142 refs = 1;
1143 /*
1144 * For tree blocks we can only drop one ref for it, and tree
1145 * blocks should not have refs > 1.
1146 *
1147 * Furthermore if we're inserting a new inline backref, we
1148 * won't reach this path either. That would be
1149 * setup_inline_extent_backref().
1150 */
1151 if (unlikely(refs_to_mod != -1)) {
1152 struct btrfs_key key;
1153
1154 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1155
1156 btrfs_print_leaf(leaf);
1157 btrfs_err(fs_info,
1158 "invalid refs_to_mod for tree block %llu, has %d expect -1",
1159 key.objectid, refs_to_mod);
1160 return -EUCLEAN;
1161 }
1162 }
1163
1164 if (unlikely(refs_to_mod < 0 && refs < -refs_to_mod)) {
1165 struct btrfs_key key;
1166 u32 extent_size;
1167
1168 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1169 if (key.type == BTRFS_METADATA_ITEM_KEY)
1170 extent_size = fs_info->nodesize;
1171 else
1172 extent_size = key.offset;
1173 btrfs_print_leaf(leaf);
1174 btrfs_err(fs_info,
1175"invalid refs_to_mod for backref entry, iref %lu extent %llu num_bytes %u, has %d expect >= -%llu",
1176 (unsigned long)iref, key.objectid, extent_size,
1177 refs_to_mod, refs);
1178 return -EUCLEAN;
1179 }
1180 refs += refs_to_mod;
1181
1182 if (refs > 0) {
1183 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1184 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1185 else
1186 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1187 } else {
1188 size = btrfs_extent_inline_ref_size(type);
1189 item_size = btrfs_item_size(leaf, path->slots[0]);
1190 ptr = (unsigned long)iref;
1191 end = (unsigned long)ei + item_size;
1192 if (ptr + size < end)
1193 memmove_extent_buffer(leaf, ptr, ptr + size,
1194 end - ptr - size);
1195 item_size -= size;
1196 btrfs_truncate_item(trans, path, item_size, 1);
1197 }
1198 btrfs_mark_buffer_dirty(trans, leaf);
1199 return 0;
1200}
1201
1202static noinline_for_stack
1203int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1204 struct btrfs_path *path,
1205 u64 bytenr, u64 num_bytes, u64 parent,
1206 u64 root_objectid, u64 owner,
1207 u64 offset, int refs_to_add,
1208 struct btrfs_delayed_extent_op *extent_op)
1209{
1210 struct btrfs_extent_inline_ref *iref;
1211 int ret;
1212
1213 ret = lookup_inline_extent_backref(trans, path, &iref, bytenr,
1214 num_bytes, parent, root_objectid,
1215 owner, offset, 1);
1216 if (ret == 0) {
1217 /*
1218 * We're adding refs to a tree block we already own, this
1219 * should not happen at all.
1220 */
1221 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1222 btrfs_print_leaf(path->nodes[0]);
1223 btrfs_crit(trans->fs_info,
1224"adding refs to an existing tree ref, bytenr %llu num_bytes %llu root_objectid %llu slot %u",
1225 bytenr, num_bytes, root_objectid, path->slots[0]);
1226 return -EUCLEAN;
1227 }
1228 ret = update_inline_extent_backref(trans, path, iref,
1229 refs_to_add, extent_op);
1230 } else if (ret == -ENOENT) {
1231 setup_inline_extent_backref(trans, path, iref, parent,
1232 root_objectid, owner, offset,
1233 refs_to_add, extent_op);
1234 ret = 0;
1235 }
1236 return ret;
1237}
1238
1239static int remove_extent_backref(struct btrfs_trans_handle *trans,
1240 struct btrfs_root *root,
1241 struct btrfs_path *path,
1242 struct btrfs_extent_inline_ref *iref,
1243 int refs_to_drop, int is_data)
1244{
1245 int ret = 0;
1246
1247 BUG_ON(!is_data && refs_to_drop != 1);
1248 if (iref)
1249 ret = update_inline_extent_backref(trans, path, iref,
1250 -refs_to_drop, NULL);
1251 else if (is_data)
1252 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1253 else
1254 ret = btrfs_del_item(trans, root, path);
1255 return ret;
1256}
1257
1258static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1259 u64 *discarded_bytes)
1260{
1261 int j, ret = 0;
1262 u64 bytes_left, end;
1263 u64 aligned_start = ALIGN(start, 1 << SECTOR_SHIFT);
1264
1265 /* Adjust the range to be aligned to 512B sectors if necessary. */
1266 if (start != aligned_start) {
1267 len -= aligned_start - start;
1268 len = round_down(len, 1 << SECTOR_SHIFT);
1269 start = aligned_start;
1270 }
1271
1272 *discarded_bytes = 0;
1273
1274 if (!len)
1275 return 0;
1276
1277 end = start + len;
1278 bytes_left = len;
1279
1280 /* Skip any superblocks on this device. */
1281 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1282 u64 sb_start = btrfs_sb_offset(j);
1283 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1284 u64 size = sb_start - start;
1285
1286 if (!in_range(sb_start, start, bytes_left) &&
1287 !in_range(sb_end, start, bytes_left) &&
1288 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1289 continue;
1290
1291 /*
1292 * Superblock spans beginning of range. Adjust start and
1293 * try again.
1294 */
1295 if (sb_start <= start) {
1296 start += sb_end - start;
1297 if (start > end) {
1298 bytes_left = 0;
1299 break;
1300 }
1301 bytes_left = end - start;
1302 continue;
1303 }
1304
1305 if (size) {
1306 ret = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT,
1307 size >> SECTOR_SHIFT,
1308 GFP_NOFS);
1309 if (!ret)
1310 *discarded_bytes += size;
1311 else if (ret != -EOPNOTSUPP)
1312 return ret;
1313 }
1314
1315 start = sb_end;
1316 if (start > end) {
1317 bytes_left = 0;
1318 break;
1319 }
1320 bytes_left = end - start;
1321 }
1322
1323 while (bytes_left) {
1324 u64 bytes_to_discard = min(BTRFS_MAX_DISCARD_CHUNK_SIZE, bytes_left);
1325
1326 ret = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT,
1327 bytes_to_discard >> SECTOR_SHIFT,
1328 GFP_NOFS);
1329
1330 if (ret) {
1331 if (ret != -EOPNOTSUPP)
1332 break;
1333 continue;
1334 }
1335
1336 start += bytes_to_discard;
1337 bytes_left -= bytes_to_discard;
1338 *discarded_bytes += bytes_to_discard;
1339
1340 if (btrfs_trim_interrupted()) {
1341 ret = -ERESTARTSYS;
1342 break;
1343 }
1344 }
1345
1346 return ret;
1347}
1348
1349static int do_discard_extent(struct btrfs_discard_stripe *stripe, u64 *bytes)
1350{
1351 struct btrfs_device *dev = stripe->dev;
1352 struct btrfs_fs_info *fs_info = dev->fs_info;
1353 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1354 u64 phys = stripe->physical;
1355 u64 len = stripe->length;
1356 u64 discarded = 0;
1357 int ret = 0;
1358
1359 /* Zone reset on a zoned filesystem */
1360 if (btrfs_can_zone_reset(dev, phys, len)) {
1361 u64 src_disc;
1362
1363 ret = btrfs_reset_device_zone(dev, phys, len, &discarded);
1364 if (ret)
1365 goto out;
1366
1367 if (!btrfs_dev_replace_is_ongoing(dev_replace) ||
1368 dev != dev_replace->srcdev)
1369 goto out;
1370
1371 src_disc = discarded;
1372
1373 /* Send to replace target as well */
1374 ret = btrfs_reset_device_zone(dev_replace->tgtdev, phys, len,
1375 &discarded);
1376 discarded += src_disc;
1377 } else if (bdev_max_discard_sectors(stripe->dev->bdev)) {
1378 ret = btrfs_issue_discard(dev->bdev, phys, len, &discarded);
1379 } else {
1380 ret = 0;
1381 *bytes = 0;
1382 }
1383
1384out:
1385 *bytes = discarded;
1386 return ret;
1387}
1388
1389int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
1390 u64 num_bytes, u64 *actual_bytes)
1391{
1392 int ret = 0;
1393 u64 discarded_bytes = 0;
1394 u64 end = bytenr + num_bytes;
1395 u64 cur = bytenr;
1396
1397 /*
1398 * Avoid races with device replace and make sure the devices in the
1399 * stripes don't go away while we are discarding.
1400 */
1401 btrfs_bio_counter_inc_blocked(fs_info);
1402 while (cur < end) {
1403 struct btrfs_discard_stripe *stripes;
1404 unsigned int num_stripes;
1405 int i;
1406
1407 num_bytes = end - cur;
1408 stripes = btrfs_map_discard(fs_info, cur, &num_bytes, &num_stripes);
1409 if (IS_ERR(stripes)) {
1410 ret = PTR_ERR(stripes);
1411 if (ret == -EOPNOTSUPP)
1412 ret = 0;
1413 break;
1414 }
1415
1416 for (i = 0; i < num_stripes; i++) {
1417 struct btrfs_discard_stripe *stripe = stripes + i;
1418 u64 bytes;
1419
1420 if (!stripe->dev->bdev) {
1421 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
1422 continue;
1423 }
1424
1425 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
1426 &stripe->dev->dev_state))
1427 continue;
1428
1429 ret = do_discard_extent(stripe, &bytes);
1430 if (ret) {
1431 /*
1432 * Keep going if discard is not supported by the
1433 * device.
1434 */
1435 if (ret != -EOPNOTSUPP)
1436 break;
1437 ret = 0;
1438 } else {
1439 discarded_bytes += bytes;
1440 }
1441 }
1442 kfree(stripes);
1443 if (ret)
1444 break;
1445 cur += num_bytes;
1446 }
1447 btrfs_bio_counter_dec(fs_info);
1448 if (actual_bytes)
1449 *actual_bytes = discarded_bytes;
1450 return ret;
1451}
1452
1453/* Can return -ENOMEM */
1454int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1455 struct btrfs_ref *generic_ref)
1456{
1457 struct btrfs_fs_info *fs_info = trans->fs_info;
1458 int ret;
1459
1460 ASSERT(generic_ref->type != BTRFS_REF_NOT_SET &&
1461 generic_ref->action);
1462 BUG_ON(generic_ref->type == BTRFS_REF_METADATA &&
1463 generic_ref->ref_root == BTRFS_TREE_LOG_OBJECTID);
1464
1465 if (generic_ref->type == BTRFS_REF_METADATA)
1466 ret = btrfs_add_delayed_tree_ref(trans, generic_ref, NULL);
1467 else
1468 ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0);
1469
1470 btrfs_ref_tree_mod(fs_info, generic_ref);
1471
1472 return ret;
1473}
1474
1475/*
1476 * Insert backreference for a given extent.
1477 *
1478 * The counterpart is in __btrfs_free_extent(), with examples and more details
1479 * how it works.
1480 *
1481 * @trans: Handle of transaction
1482 *
1483 * @node: The delayed ref node used to get the bytenr/length for
1484 * extent whose references are incremented.
1485 *
1486 * @extent_op Pointer to a structure, holding information necessary when
1487 * updating a tree block's flags
1488 *
1489 */
1490static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1491 struct btrfs_delayed_ref_node *node,
1492 struct btrfs_delayed_extent_op *extent_op)
1493{
1494 struct btrfs_path *path;
1495 struct extent_buffer *leaf;
1496 struct btrfs_extent_item *item;
1497 struct btrfs_key key;
1498 u64 bytenr = node->bytenr;
1499 u64 num_bytes = node->num_bytes;
1500 u64 owner = btrfs_delayed_ref_owner(node);
1501 u64 offset = btrfs_delayed_ref_offset(node);
1502 u64 refs;
1503 int refs_to_add = node->ref_mod;
1504 int ret;
1505
1506 path = btrfs_alloc_path();
1507 if (!path)
1508 return -ENOMEM;
1509
1510 /* this will setup the path even if it fails to insert the back ref */
1511 ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
1512 node->parent, node->ref_root, owner,
1513 offset, refs_to_add, extent_op);
1514 if ((ret < 0 && ret != -EAGAIN) || !ret)
1515 goto out;
1516
1517 /*
1518 * Ok we had -EAGAIN which means we didn't have space to insert and
1519 * inline extent ref, so just update the reference count and add a
1520 * normal backref.
1521 */
1522 leaf = path->nodes[0];
1523 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1524 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1525 refs = btrfs_extent_refs(leaf, item);
1526 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1527 if (extent_op)
1528 __run_delayed_extent_op(extent_op, leaf, item);
1529
1530 btrfs_mark_buffer_dirty(trans, leaf);
1531 btrfs_release_path(path);
1532
1533 /* now insert the actual backref */
1534 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1535 ret = insert_tree_block_ref(trans, path, node, bytenr);
1536 else
1537 ret = insert_extent_data_ref(trans, path, node, bytenr);
1538
1539 if (ret)
1540 btrfs_abort_transaction(trans, ret);
1541out:
1542 btrfs_free_path(path);
1543 return ret;
1544}
1545
1546static void free_head_ref_squota_rsv(struct btrfs_fs_info *fs_info,
1547 struct btrfs_delayed_ref_head *href)
1548{
1549 u64 root = href->owning_root;
1550
1551 /*
1552 * Don't check must_insert_reserved, as this is called from contexts
1553 * where it has already been unset.
1554 */
1555 if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_SIMPLE ||
1556 !href->is_data || !is_fstree(root))
1557 return;
1558
1559 btrfs_qgroup_free_refroot(fs_info, root, href->reserved_bytes,
1560 BTRFS_QGROUP_RSV_DATA);
1561}
1562
1563static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1564 struct btrfs_delayed_ref_head *href,
1565 struct btrfs_delayed_ref_node *node,
1566 struct btrfs_delayed_extent_op *extent_op,
1567 bool insert_reserved)
1568{
1569 int ret = 0;
1570 u64 parent = 0;
1571 u64 flags = 0;
1572
1573 trace_run_delayed_data_ref(trans->fs_info, node);
1574
1575 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1576 parent = node->parent;
1577
1578 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1579 struct btrfs_key key;
1580 struct btrfs_squota_delta delta = {
1581 .root = href->owning_root,
1582 .num_bytes = node->num_bytes,
1583 .is_data = true,
1584 .is_inc = true,
1585 .generation = trans->transid,
1586 };
1587 u64 owner = btrfs_delayed_ref_owner(node);
1588 u64 offset = btrfs_delayed_ref_offset(node);
1589
1590 if (extent_op)
1591 flags |= extent_op->flags_to_set;
1592
1593 key.objectid = node->bytenr;
1594 key.type = BTRFS_EXTENT_ITEM_KEY;
1595 key.offset = node->num_bytes;
1596
1597 ret = alloc_reserved_file_extent(trans, parent, node->ref_root,
1598 flags, owner, offset, &key,
1599 node->ref_mod,
1600 href->owning_root);
1601 free_head_ref_squota_rsv(trans->fs_info, href);
1602 if (!ret)
1603 ret = btrfs_record_squota_delta(trans->fs_info, &delta);
1604 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1605 ret = __btrfs_inc_extent_ref(trans, node, extent_op);
1606 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1607 ret = __btrfs_free_extent(trans, href, node, extent_op);
1608 } else {
1609 BUG();
1610 }
1611 return ret;
1612}
1613
1614static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1615 struct extent_buffer *leaf,
1616 struct btrfs_extent_item *ei)
1617{
1618 u64 flags = btrfs_extent_flags(leaf, ei);
1619 if (extent_op->update_flags) {
1620 flags |= extent_op->flags_to_set;
1621 btrfs_set_extent_flags(leaf, ei, flags);
1622 }
1623
1624 if (extent_op->update_key) {
1625 struct btrfs_tree_block_info *bi;
1626 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1627 bi = (struct btrfs_tree_block_info *)(ei + 1);
1628 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1629 }
1630}
1631
1632static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1633 struct btrfs_delayed_ref_head *head,
1634 struct btrfs_delayed_extent_op *extent_op)
1635{
1636 struct btrfs_fs_info *fs_info = trans->fs_info;
1637 struct btrfs_root *root;
1638 struct btrfs_key key;
1639 struct btrfs_path *path;
1640 struct btrfs_extent_item *ei;
1641 struct extent_buffer *leaf;
1642 u32 item_size;
1643 int ret;
1644 int metadata = 1;
1645
1646 if (TRANS_ABORTED(trans))
1647 return 0;
1648
1649 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1650 metadata = 0;
1651
1652 path = btrfs_alloc_path();
1653 if (!path)
1654 return -ENOMEM;
1655
1656 key.objectid = head->bytenr;
1657
1658 if (metadata) {
1659 key.type = BTRFS_METADATA_ITEM_KEY;
1660 key.offset = head->level;
1661 } else {
1662 key.type = BTRFS_EXTENT_ITEM_KEY;
1663 key.offset = head->num_bytes;
1664 }
1665
1666 root = btrfs_extent_root(fs_info, key.objectid);
1667again:
1668 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1669 if (ret < 0) {
1670 goto out;
1671 } else if (ret > 0) {
1672 if (metadata) {
1673 if (path->slots[0] > 0) {
1674 path->slots[0]--;
1675 btrfs_item_key_to_cpu(path->nodes[0], &key,
1676 path->slots[0]);
1677 if (key.objectid == head->bytenr &&
1678 key.type == BTRFS_EXTENT_ITEM_KEY &&
1679 key.offset == head->num_bytes)
1680 ret = 0;
1681 }
1682 if (ret > 0) {
1683 btrfs_release_path(path);
1684 metadata = 0;
1685
1686 key.objectid = head->bytenr;
1687 key.offset = head->num_bytes;
1688 key.type = BTRFS_EXTENT_ITEM_KEY;
1689 goto again;
1690 }
1691 } else {
1692 ret = -EUCLEAN;
1693 btrfs_err(fs_info,
1694 "missing extent item for extent %llu num_bytes %llu level %d",
1695 head->bytenr, head->num_bytes, head->level);
1696 goto out;
1697 }
1698 }
1699
1700 leaf = path->nodes[0];
1701 item_size = btrfs_item_size(leaf, path->slots[0]);
1702
1703 if (unlikely(item_size < sizeof(*ei))) {
1704 ret = -EUCLEAN;
1705 btrfs_err(fs_info,
1706 "unexpected extent item size, has %u expect >= %zu",
1707 item_size, sizeof(*ei));
1708 btrfs_abort_transaction(trans, ret);
1709 goto out;
1710 }
1711
1712 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1713 __run_delayed_extent_op(extent_op, leaf, ei);
1714
1715 btrfs_mark_buffer_dirty(trans, leaf);
1716out:
1717 btrfs_free_path(path);
1718 return ret;
1719}
1720
1721static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1722 struct btrfs_delayed_ref_head *href,
1723 struct btrfs_delayed_ref_node *node,
1724 struct btrfs_delayed_extent_op *extent_op,
1725 bool insert_reserved)
1726{
1727 int ret = 0;
1728 struct btrfs_fs_info *fs_info = trans->fs_info;
1729 u64 parent = 0;
1730 u64 ref_root = 0;
1731
1732 trace_run_delayed_tree_ref(trans->fs_info, node);
1733
1734 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1735 parent = node->parent;
1736 ref_root = node->ref_root;
1737
1738 if (unlikely(node->ref_mod != 1)) {
1739 btrfs_err(trans->fs_info,
1740 "btree block %llu has %d references rather than 1: action %d ref_root %llu parent %llu",
1741 node->bytenr, node->ref_mod, node->action, ref_root,
1742 parent);
1743 return -EUCLEAN;
1744 }
1745 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1746 struct btrfs_squota_delta delta = {
1747 .root = href->owning_root,
1748 .num_bytes = fs_info->nodesize,
1749 .is_data = false,
1750 .is_inc = true,
1751 .generation = trans->transid,
1752 };
1753
1754 ret = alloc_reserved_tree_block(trans, node, extent_op);
1755 if (!ret)
1756 btrfs_record_squota_delta(fs_info, &delta);
1757 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1758 ret = __btrfs_inc_extent_ref(trans, node, extent_op);
1759 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1760 ret = __btrfs_free_extent(trans, href, node, extent_op);
1761 } else {
1762 BUG();
1763 }
1764 return ret;
1765}
1766
1767/* helper function to actually process a single delayed ref entry */
1768static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1769 struct btrfs_delayed_ref_head *href,
1770 struct btrfs_delayed_ref_node *node,
1771 struct btrfs_delayed_extent_op *extent_op,
1772 bool insert_reserved)
1773{
1774 int ret = 0;
1775
1776 if (TRANS_ABORTED(trans)) {
1777 if (insert_reserved) {
1778 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1779 free_head_ref_squota_rsv(trans->fs_info, href);
1780 }
1781 return 0;
1782 }
1783
1784 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1785 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1786 ret = run_delayed_tree_ref(trans, href, node, extent_op,
1787 insert_reserved);
1788 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1789 node->type == BTRFS_SHARED_DATA_REF_KEY)
1790 ret = run_delayed_data_ref(trans, href, node, extent_op,
1791 insert_reserved);
1792 else if (node->type == BTRFS_EXTENT_OWNER_REF_KEY)
1793 ret = 0;
1794 else
1795 BUG();
1796 if (ret && insert_reserved)
1797 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1798 if (ret < 0)
1799 btrfs_err(trans->fs_info,
1800"failed to run delayed ref for logical %llu num_bytes %llu type %u action %u ref_mod %d: %d",
1801 node->bytenr, node->num_bytes, node->type,
1802 node->action, node->ref_mod, ret);
1803 return ret;
1804}
1805
1806static inline struct btrfs_delayed_ref_node *
1807select_delayed_ref(struct btrfs_delayed_ref_head *head)
1808{
1809 struct btrfs_delayed_ref_node *ref;
1810
1811 if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
1812 return NULL;
1813
1814 /*
1815 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
1816 * This is to prevent a ref count from going down to zero, which deletes
1817 * the extent item from the extent tree, when there still are references
1818 * to add, which would fail because they would not find the extent item.
1819 */
1820 if (!list_empty(&head->ref_add_list))
1821 return list_first_entry(&head->ref_add_list,
1822 struct btrfs_delayed_ref_node, add_list);
1823
1824 ref = rb_entry(rb_first_cached(&head->ref_tree),
1825 struct btrfs_delayed_ref_node, ref_node);
1826 ASSERT(list_empty(&ref->add_list));
1827 return ref;
1828}
1829
1830static struct btrfs_delayed_extent_op *cleanup_extent_op(
1831 struct btrfs_delayed_ref_head *head)
1832{
1833 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
1834
1835 if (!extent_op)
1836 return NULL;
1837
1838 if (head->must_insert_reserved) {
1839 head->extent_op = NULL;
1840 btrfs_free_delayed_extent_op(extent_op);
1841 return NULL;
1842 }
1843 return extent_op;
1844}
1845
1846static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
1847 struct btrfs_delayed_ref_head *head)
1848{
1849 struct btrfs_delayed_extent_op *extent_op;
1850 int ret;
1851
1852 extent_op = cleanup_extent_op(head);
1853 if (!extent_op)
1854 return 0;
1855 head->extent_op = NULL;
1856 spin_unlock(&head->lock);
1857 ret = run_delayed_extent_op(trans, head, extent_op);
1858 btrfs_free_delayed_extent_op(extent_op);
1859 return ret ? ret : 1;
1860}
1861
1862u64 btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
1863 struct btrfs_delayed_ref_root *delayed_refs,
1864 struct btrfs_delayed_ref_head *head)
1865{
1866 u64 ret = 0;
1867
1868 /*
1869 * We had csum deletions accounted for in our delayed refs rsv, we need
1870 * to drop the csum leaves for this update from our delayed_refs_rsv.
1871 */
1872 if (head->total_ref_mod < 0 && head->is_data) {
1873 int nr_csums;
1874
1875 spin_lock(&delayed_refs->lock);
1876 delayed_refs->pending_csums -= head->num_bytes;
1877 spin_unlock(&delayed_refs->lock);
1878 nr_csums = btrfs_csum_bytes_to_leaves(fs_info, head->num_bytes);
1879
1880 btrfs_delayed_refs_rsv_release(fs_info, 0, nr_csums);
1881
1882 ret = btrfs_calc_delayed_ref_csum_bytes(fs_info, nr_csums);
1883 }
1884 /* must_insert_reserved can be set only if we didn't run the head ref. */
1885 if (head->must_insert_reserved)
1886 free_head_ref_squota_rsv(fs_info, head);
1887
1888 return ret;
1889}
1890
1891static int cleanup_ref_head(struct btrfs_trans_handle *trans,
1892 struct btrfs_delayed_ref_head *head,
1893 u64 *bytes_released)
1894{
1895
1896 struct btrfs_fs_info *fs_info = trans->fs_info;
1897 struct btrfs_delayed_ref_root *delayed_refs;
1898 int ret;
1899
1900 delayed_refs = &trans->transaction->delayed_refs;
1901
1902 ret = run_and_cleanup_extent_op(trans, head);
1903 if (ret < 0) {
1904 btrfs_unselect_ref_head(delayed_refs, head);
1905 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
1906 return ret;
1907 } else if (ret) {
1908 return ret;
1909 }
1910
1911 /*
1912 * Need to drop our head ref lock and re-acquire the delayed ref lock
1913 * and then re-check to make sure nobody got added.
1914 */
1915 spin_unlock(&head->lock);
1916 spin_lock(&delayed_refs->lock);
1917 spin_lock(&head->lock);
1918 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
1919 spin_unlock(&head->lock);
1920 spin_unlock(&delayed_refs->lock);
1921 return 1;
1922 }
1923 btrfs_delete_ref_head(fs_info, delayed_refs, head);
1924 spin_unlock(&head->lock);
1925 spin_unlock(&delayed_refs->lock);
1926
1927 if (head->must_insert_reserved) {
1928 btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1);
1929 if (head->is_data) {
1930 struct btrfs_root *csum_root;
1931
1932 csum_root = btrfs_csum_root(fs_info, head->bytenr);
1933 ret = btrfs_del_csums(trans, csum_root, head->bytenr,
1934 head->num_bytes);
1935 }
1936 }
1937
1938 *bytes_released += btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
1939
1940 trace_run_delayed_ref_head(fs_info, head, 0);
1941 btrfs_delayed_ref_unlock(head);
1942 btrfs_put_delayed_ref_head(head);
1943 return ret;
1944}
1945
1946static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
1947 struct btrfs_delayed_ref_head *locked_ref,
1948 u64 *bytes_released)
1949{
1950 struct btrfs_fs_info *fs_info = trans->fs_info;
1951 struct btrfs_delayed_ref_root *delayed_refs;
1952 struct btrfs_delayed_extent_op *extent_op;
1953 struct btrfs_delayed_ref_node *ref;
1954 bool must_insert_reserved;
1955 int ret;
1956
1957 delayed_refs = &trans->transaction->delayed_refs;
1958
1959 lockdep_assert_held(&locked_ref->mutex);
1960 lockdep_assert_held(&locked_ref->lock);
1961
1962 while ((ref = select_delayed_ref(locked_ref))) {
1963 if (ref->seq &&
1964 btrfs_check_delayed_seq(fs_info, ref->seq)) {
1965 spin_unlock(&locked_ref->lock);
1966 btrfs_unselect_ref_head(delayed_refs, locked_ref);
1967 return -EAGAIN;
1968 }
1969
1970 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
1971 RB_CLEAR_NODE(&ref->ref_node);
1972 if (!list_empty(&ref->add_list))
1973 list_del(&ref->add_list);
1974 /*
1975 * When we play the delayed ref, also correct the ref_mod on
1976 * head
1977 */
1978 switch (ref->action) {
1979 case BTRFS_ADD_DELAYED_REF:
1980 case BTRFS_ADD_DELAYED_EXTENT:
1981 locked_ref->ref_mod -= ref->ref_mod;
1982 break;
1983 case BTRFS_DROP_DELAYED_REF:
1984 locked_ref->ref_mod += ref->ref_mod;
1985 break;
1986 default:
1987 WARN_ON(1);
1988 }
1989
1990 /*
1991 * Record the must_insert_reserved flag before we drop the
1992 * spin lock.
1993 */
1994 must_insert_reserved = locked_ref->must_insert_reserved;
1995 /*
1996 * Unsetting this on the head ref relinquishes ownership of
1997 * the rsv_bytes, so it is critical that every possible code
1998 * path from here forward frees all reserves including qgroup
1999 * reserve.
2000 */
2001 locked_ref->must_insert_reserved = false;
2002
2003 extent_op = locked_ref->extent_op;
2004 locked_ref->extent_op = NULL;
2005 spin_unlock(&locked_ref->lock);
2006
2007 ret = run_one_delayed_ref(trans, locked_ref, ref, extent_op,
2008 must_insert_reserved);
2009 btrfs_delayed_refs_rsv_release(fs_info, 1, 0);
2010 *bytes_released += btrfs_calc_delayed_ref_bytes(fs_info, 1);
2011
2012 btrfs_free_delayed_extent_op(extent_op);
2013 if (ret) {
2014 btrfs_unselect_ref_head(delayed_refs, locked_ref);
2015 btrfs_put_delayed_ref(ref);
2016 return ret;
2017 }
2018
2019 btrfs_put_delayed_ref(ref);
2020 cond_resched();
2021
2022 spin_lock(&locked_ref->lock);
2023 btrfs_merge_delayed_refs(fs_info, delayed_refs, locked_ref);
2024 }
2025
2026 return 0;
2027}
2028
2029/*
2030 * Returns 0 on success or if called with an already aborted transaction.
2031 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2032 */
2033static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2034 u64 min_bytes)
2035{
2036 struct btrfs_fs_info *fs_info = trans->fs_info;
2037 struct btrfs_delayed_ref_root *delayed_refs;
2038 struct btrfs_delayed_ref_head *locked_ref = NULL;
2039 int ret;
2040 unsigned long count = 0;
2041 unsigned long max_count = 0;
2042 u64 bytes_processed = 0;
2043
2044 delayed_refs = &trans->transaction->delayed_refs;
2045 if (min_bytes == 0) {
2046 max_count = delayed_refs->num_heads_ready;
2047 min_bytes = U64_MAX;
2048 }
2049
2050 do {
2051 if (!locked_ref) {
2052 locked_ref = btrfs_select_ref_head(fs_info, delayed_refs);
2053 if (IS_ERR_OR_NULL(locked_ref)) {
2054 if (PTR_ERR(locked_ref) == -EAGAIN) {
2055 continue;
2056 } else {
2057 break;
2058 }
2059 }
2060 count++;
2061 }
2062 /*
2063 * We need to try and merge add/drops of the same ref since we
2064 * can run into issues with relocate dropping the implicit ref
2065 * and then it being added back again before the drop can
2066 * finish. If we merged anything we need to re-loop so we can
2067 * get a good ref.
2068 * Or we can get node references of the same type that weren't
2069 * merged when created due to bumps in the tree mod seq, and
2070 * we need to merge them to prevent adding an inline extent
2071 * backref before dropping it (triggering a BUG_ON at
2072 * insert_inline_extent_backref()).
2073 */
2074 spin_lock(&locked_ref->lock);
2075 btrfs_merge_delayed_refs(fs_info, delayed_refs, locked_ref);
2076
2077 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref, &bytes_processed);
2078 if (ret < 0 && ret != -EAGAIN) {
2079 /*
2080 * Error, btrfs_run_delayed_refs_for_head already
2081 * unlocked everything so just bail out
2082 */
2083 return ret;
2084 } else if (!ret) {
2085 /*
2086 * Success, perform the usual cleanup of a processed
2087 * head
2088 */
2089 ret = cleanup_ref_head(trans, locked_ref, &bytes_processed);
2090 if (ret > 0 ) {
2091 /* We dropped our lock, we need to loop. */
2092 ret = 0;
2093 continue;
2094 } else if (ret) {
2095 return ret;
2096 }
2097 }
2098
2099 /*
2100 * Either success case or btrfs_run_delayed_refs_for_head
2101 * returned -EAGAIN, meaning we need to select another head
2102 */
2103
2104 locked_ref = NULL;
2105 cond_resched();
2106 } while ((min_bytes != U64_MAX && bytes_processed < min_bytes) ||
2107 (max_count > 0 && count < max_count) ||
2108 locked_ref);
2109
2110 return 0;
2111}
2112
2113#ifdef SCRAMBLE_DELAYED_REFS
2114/*
2115 * Normally delayed refs get processed in ascending bytenr order. This
2116 * correlates in most cases to the order added. To expose dependencies on this
2117 * order, we start to process the tree in the middle instead of the beginning
2118 */
2119static u64 find_middle(struct rb_root *root)
2120{
2121 struct rb_node *n = root->rb_node;
2122 struct btrfs_delayed_ref_node *entry;
2123 int alt = 1;
2124 u64 middle;
2125 u64 first = 0, last = 0;
2126
2127 n = rb_first(root);
2128 if (n) {
2129 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2130 first = entry->bytenr;
2131 }
2132 n = rb_last(root);
2133 if (n) {
2134 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2135 last = entry->bytenr;
2136 }
2137 n = root->rb_node;
2138
2139 while (n) {
2140 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2141 WARN_ON(!entry->in_tree);
2142
2143 middle = entry->bytenr;
2144
2145 if (alt)
2146 n = n->rb_left;
2147 else
2148 n = n->rb_right;
2149
2150 alt = 1 - alt;
2151 }
2152 return middle;
2153}
2154#endif
2155
2156/*
2157 * Start processing the delayed reference count updates and extent insertions
2158 * we have queued up so far.
2159 *
2160 * @trans: Transaction handle.
2161 * @min_bytes: How many bytes of delayed references to process. After this
2162 * many bytes we stop processing delayed references if there are
2163 * any more. If 0 it means to run all existing delayed references,
2164 * but not new ones added after running all existing ones.
2165 * Use (u64)-1 (U64_MAX) to run all existing delayed references
2166 * plus any new ones that are added.
2167 *
2168 * Returns 0 on success or if called with an aborted transaction
2169 * Returns <0 on error and aborts the transaction
2170 */
2171int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, u64 min_bytes)
2172{
2173 struct btrfs_fs_info *fs_info = trans->fs_info;
2174 struct btrfs_delayed_ref_root *delayed_refs;
2175 int ret;
2176
2177 /* We'll clean this up in btrfs_cleanup_transaction */
2178 if (TRANS_ABORTED(trans))
2179 return 0;
2180
2181 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2182 return 0;
2183
2184 delayed_refs = &trans->transaction->delayed_refs;
2185again:
2186#ifdef SCRAMBLE_DELAYED_REFS
2187 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2188#endif
2189 ret = __btrfs_run_delayed_refs(trans, min_bytes);
2190 if (ret < 0) {
2191 btrfs_abort_transaction(trans, ret);
2192 return ret;
2193 }
2194
2195 if (min_bytes == U64_MAX) {
2196 btrfs_create_pending_block_groups(trans);
2197
2198 spin_lock(&delayed_refs->lock);
2199 if (xa_empty(&delayed_refs->head_refs)) {
2200 spin_unlock(&delayed_refs->lock);
2201 return 0;
2202 }
2203 spin_unlock(&delayed_refs->lock);
2204
2205 cond_resched();
2206 goto again;
2207 }
2208
2209 return 0;
2210}
2211
2212int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2213 struct extent_buffer *eb, u64 flags)
2214{
2215 struct btrfs_delayed_extent_op *extent_op;
2216 int ret;
2217
2218 extent_op = btrfs_alloc_delayed_extent_op();
2219 if (!extent_op)
2220 return -ENOMEM;
2221
2222 extent_op->flags_to_set = flags;
2223 extent_op->update_flags = true;
2224 extent_op->update_key = false;
2225
2226 ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len,
2227 btrfs_header_level(eb), extent_op);
2228 if (ret)
2229 btrfs_free_delayed_extent_op(extent_op);
2230 return ret;
2231}
2232
2233static noinline int check_delayed_ref(struct btrfs_root *root,
2234 struct btrfs_path *path,
2235 u64 objectid, u64 offset, u64 bytenr)
2236{
2237 struct btrfs_delayed_ref_head *head;
2238 struct btrfs_delayed_ref_node *ref;
2239 struct btrfs_delayed_ref_root *delayed_refs;
2240 struct btrfs_transaction *cur_trans;
2241 struct rb_node *node;
2242 int ret = 0;
2243
2244 spin_lock(&root->fs_info->trans_lock);
2245 cur_trans = root->fs_info->running_transaction;
2246 if (cur_trans)
2247 refcount_inc(&cur_trans->use_count);
2248 spin_unlock(&root->fs_info->trans_lock);
2249 if (!cur_trans)
2250 return 0;
2251
2252 delayed_refs = &cur_trans->delayed_refs;
2253 spin_lock(&delayed_refs->lock);
2254 head = btrfs_find_delayed_ref_head(root->fs_info, delayed_refs, bytenr);
2255 if (!head) {
2256 spin_unlock(&delayed_refs->lock);
2257 btrfs_put_transaction(cur_trans);
2258 return 0;
2259 }
2260
2261 if (!mutex_trylock(&head->mutex)) {
2262 if (path->nowait) {
2263 spin_unlock(&delayed_refs->lock);
2264 btrfs_put_transaction(cur_trans);
2265 return -EAGAIN;
2266 }
2267
2268 refcount_inc(&head->refs);
2269 spin_unlock(&delayed_refs->lock);
2270
2271 btrfs_release_path(path);
2272
2273 /*
2274 * Mutex was contended, block until it's released and let
2275 * caller try again
2276 */
2277 mutex_lock(&head->mutex);
2278 mutex_unlock(&head->mutex);
2279 btrfs_put_delayed_ref_head(head);
2280 btrfs_put_transaction(cur_trans);
2281 return -EAGAIN;
2282 }
2283 spin_unlock(&delayed_refs->lock);
2284
2285 spin_lock(&head->lock);
2286 /*
2287 * XXX: We should replace this with a proper search function in the
2288 * future.
2289 */
2290 for (node = rb_first_cached(&head->ref_tree); node;
2291 node = rb_next(node)) {
2292 u64 ref_owner;
2293 u64 ref_offset;
2294
2295 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
2296 /* If it's a shared ref we know a cross reference exists */
2297 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2298 ret = 1;
2299 break;
2300 }
2301
2302 ref_owner = btrfs_delayed_ref_owner(ref);
2303 ref_offset = btrfs_delayed_ref_offset(ref);
2304
2305 /*
2306 * If our ref doesn't match the one we're currently looking at
2307 * then we have a cross reference.
2308 */
2309 if (ref->ref_root != btrfs_root_id(root) ||
2310 ref_owner != objectid || ref_offset != offset) {
2311 ret = 1;
2312 break;
2313 }
2314 }
2315 spin_unlock(&head->lock);
2316 mutex_unlock(&head->mutex);
2317 btrfs_put_transaction(cur_trans);
2318 return ret;
2319}
2320
2321static noinline int check_committed_ref(struct btrfs_root *root,
2322 struct btrfs_path *path,
2323 u64 objectid, u64 offset, u64 bytenr,
2324 bool strict)
2325{
2326 struct btrfs_fs_info *fs_info = root->fs_info;
2327 struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr);
2328 struct extent_buffer *leaf;
2329 struct btrfs_extent_data_ref *ref;
2330 struct btrfs_extent_inline_ref *iref;
2331 struct btrfs_extent_item *ei;
2332 struct btrfs_key key;
2333 u32 item_size;
2334 u32 expected_size;
2335 int type;
2336 int ret;
2337
2338 key.objectid = bytenr;
2339 key.offset = (u64)-1;
2340 key.type = BTRFS_EXTENT_ITEM_KEY;
2341
2342 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2343 if (ret < 0)
2344 goto out;
2345 if (ret == 0) {
2346 /*
2347 * Key with offset -1 found, there would have to exist an extent
2348 * item with such offset, but this is out of the valid range.
2349 */
2350 ret = -EUCLEAN;
2351 goto out;
2352 }
2353
2354 ret = -ENOENT;
2355 if (path->slots[0] == 0)
2356 goto out;
2357
2358 path->slots[0]--;
2359 leaf = path->nodes[0];
2360 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2361
2362 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2363 goto out;
2364
2365 ret = 1;
2366 item_size = btrfs_item_size(leaf, path->slots[0]);
2367 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2368 expected_size = sizeof(*ei) + btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY);
2369
2370 /* No inline refs; we need to bail before checking for owner ref. */
2371 if (item_size == sizeof(*ei))
2372 goto out;
2373
2374 /* Check for an owner ref; skip over it to the real inline refs. */
2375 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2376 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
2377 if (btrfs_fs_incompat(fs_info, SIMPLE_QUOTA) && type == BTRFS_EXTENT_OWNER_REF_KEY) {
2378 expected_size += btrfs_extent_inline_ref_size(BTRFS_EXTENT_OWNER_REF_KEY);
2379 iref = (struct btrfs_extent_inline_ref *)(iref + 1);
2380 }
2381
2382 /* If extent item has more than 1 inline ref then it's shared */
2383 if (item_size != expected_size)
2384 goto out;
2385
2386 /*
2387 * If extent created before last snapshot => it's shared unless the
2388 * snapshot has been deleted. Use the heuristic if strict is false.
2389 */
2390 if (!strict &&
2391 (btrfs_extent_generation(leaf, ei) <=
2392 btrfs_root_last_snapshot(&root->root_item)))
2393 goto out;
2394
2395 /* If this extent has SHARED_DATA_REF then it's shared */
2396 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
2397 if (type != BTRFS_EXTENT_DATA_REF_KEY)
2398 goto out;
2399
2400 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2401 if (btrfs_extent_refs(leaf, ei) !=
2402 btrfs_extent_data_ref_count(leaf, ref) ||
2403 btrfs_extent_data_ref_root(leaf, ref) != btrfs_root_id(root) ||
2404 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2405 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2406 goto out;
2407
2408 ret = 0;
2409out:
2410 return ret;
2411}
2412
2413int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
2414 u64 bytenr, bool strict, struct btrfs_path *path)
2415{
2416 int ret;
2417
2418 do {
2419 ret = check_committed_ref(root, path, objectid,
2420 offset, bytenr, strict);
2421 if (ret && ret != -ENOENT)
2422 goto out;
2423
2424 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
2425 } while (ret == -EAGAIN && !path->nowait);
2426
2427out:
2428 btrfs_release_path(path);
2429 if (btrfs_is_data_reloc_root(root))
2430 WARN_ON(ret > 0);
2431 return ret;
2432}
2433
2434static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2435 struct btrfs_root *root,
2436 struct extent_buffer *buf,
2437 int full_backref, int inc)
2438{
2439 struct btrfs_fs_info *fs_info = root->fs_info;
2440 u64 parent;
2441 u64 ref_root;
2442 u32 nritems;
2443 struct btrfs_key key;
2444 struct btrfs_file_extent_item *fi;
2445 bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC);
2446 int i;
2447 int action;
2448 int level;
2449 int ret = 0;
2450
2451 if (btrfs_is_testing(fs_info))
2452 return 0;
2453
2454 ref_root = btrfs_header_owner(buf);
2455 nritems = btrfs_header_nritems(buf);
2456 level = btrfs_header_level(buf);
2457
2458 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0)
2459 return 0;
2460
2461 if (full_backref)
2462 parent = buf->start;
2463 else
2464 parent = 0;
2465 if (inc)
2466 action = BTRFS_ADD_DELAYED_REF;
2467 else
2468 action = BTRFS_DROP_DELAYED_REF;
2469
2470 for (i = 0; i < nritems; i++) {
2471 struct btrfs_ref ref = {
2472 .action = action,
2473 .parent = parent,
2474 .ref_root = ref_root,
2475 };
2476
2477 if (level == 0) {
2478 btrfs_item_key_to_cpu(buf, &key, i);
2479 if (key.type != BTRFS_EXTENT_DATA_KEY)
2480 continue;
2481 fi = btrfs_item_ptr(buf, i,
2482 struct btrfs_file_extent_item);
2483 if (btrfs_file_extent_type(buf, fi) ==
2484 BTRFS_FILE_EXTENT_INLINE)
2485 continue;
2486 ref.bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2487 if (ref.bytenr == 0)
2488 continue;
2489
2490 ref.num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2491 ref.owning_root = ref_root;
2492
2493 key.offset -= btrfs_file_extent_offset(buf, fi);
2494 btrfs_init_data_ref(&ref, key.objectid, key.offset,
2495 btrfs_root_id(root), for_reloc);
2496 if (inc)
2497 ret = btrfs_inc_extent_ref(trans, &ref);
2498 else
2499 ret = btrfs_free_extent(trans, &ref);
2500 if (ret)
2501 goto fail;
2502 } else {
2503 /* We don't know the owning_root, leave as 0. */
2504 ref.bytenr = btrfs_node_blockptr(buf, i);
2505 ref.num_bytes = fs_info->nodesize;
2506
2507 btrfs_init_tree_ref(&ref, level - 1,
2508 btrfs_root_id(root), for_reloc);
2509 if (inc)
2510 ret = btrfs_inc_extent_ref(trans, &ref);
2511 else
2512 ret = btrfs_free_extent(trans, &ref);
2513 if (ret)
2514 goto fail;
2515 }
2516 }
2517 return 0;
2518fail:
2519 return ret;
2520}
2521
2522int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2523 struct extent_buffer *buf, int full_backref)
2524{
2525 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2526}
2527
2528int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2529 struct extent_buffer *buf, int full_backref)
2530{
2531 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2532}
2533
2534static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
2535{
2536 struct btrfs_fs_info *fs_info = root->fs_info;
2537 u64 flags;
2538 u64 ret;
2539
2540 if (data)
2541 flags = BTRFS_BLOCK_GROUP_DATA;
2542 else if (root == fs_info->chunk_root)
2543 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2544 else
2545 flags = BTRFS_BLOCK_GROUP_METADATA;
2546
2547 ret = btrfs_get_alloc_profile(fs_info, flags);
2548 return ret;
2549}
2550
2551static u64 first_logical_byte(struct btrfs_fs_info *fs_info)
2552{
2553 struct rb_node *leftmost;
2554 u64 bytenr = 0;
2555
2556 read_lock(&fs_info->block_group_cache_lock);
2557 /* Get the block group with the lowest logical start address. */
2558 leftmost = rb_first_cached(&fs_info->block_group_cache_tree);
2559 if (leftmost) {
2560 struct btrfs_block_group *bg;
2561
2562 bg = rb_entry(leftmost, struct btrfs_block_group, cache_node);
2563 bytenr = bg->start;
2564 }
2565 read_unlock(&fs_info->block_group_cache_lock);
2566
2567 return bytenr;
2568}
2569
2570static int pin_down_extent(struct btrfs_trans_handle *trans,
2571 struct btrfs_block_group *cache,
2572 u64 bytenr, u64 num_bytes, int reserved)
2573{
2574 struct btrfs_fs_info *fs_info = cache->fs_info;
2575
2576 spin_lock(&cache->space_info->lock);
2577 spin_lock(&cache->lock);
2578 cache->pinned += num_bytes;
2579 btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info,
2580 num_bytes);
2581 if (reserved) {
2582 cache->reserved -= num_bytes;
2583 cache->space_info->bytes_reserved -= num_bytes;
2584 }
2585 spin_unlock(&cache->lock);
2586 spin_unlock(&cache->space_info->lock);
2587
2588 set_extent_bit(&trans->transaction->pinned_extents, bytenr,
2589 bytenr + num_bytes - 1, EXTENT_DIRTY, NULL);
2590 return 0;
2591}
2592
2593int btrfs_pin_extent(struct btrfs_trans_handle *trans,
2594 u64 bytenr, u64 num_bytes, int reserved)
2595{
2596 struct btrfs_block_group *cache;
2597
2598 cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2599 BUG_ON(!cache); /* Logic error */
2600
2601 pin_down_extent(trans, cache, bytenr, num_bytes, reserved);
2602
2603 btrfs_put_block_group(cache);
2604 return 0;
2605}
2606
2607int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
2608 const struct extent_buffer *eb)
2609{
2610 struct btrfs_block_group *cache;
2611 int ret;
2612
2613 cache = btrfs_lookup_block_group(trans->fs_info, eb->start);
2614 if (!cache)
2615 return -EINVAL;
2616
2617 /*
2618 * Fully cache the free space first so that our pin removes the free space
2619 * from the cache.
2620 */
2621 ret = btrfs_cache_block_group(cache, true);
2622 if (ret)
2623 goto out;
2624
2625 pin_down_extent(trans, cache, eb->start, eb->len, 0);
2626
2627 /* remove us from the free space cache (if we're there at all) */
2628 ret = btrfs_remove_free_space(cache, eb->start, eb->len);
2629out:
2630 btrfs_put_block_group(cache);
2631 return ret;
2632}
2633
2634static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
2635 u64 start, u64 num_bytes)
2636{
2637 int ret;
2638 struct btrfs_block_group *block_group;
2639
2640 block_group = btrfs_lookup_block_group(fs_info, start);
2641 if (!block_group)
2642 return -EINVAL;
2643
2644 ret = btrfs_cache_block_group(block_group, true);
2645 if (ret)
2646 goto out;
2647
2648 ret = btrfs_remove_free_space(block_group, start, num_bytes);
2649out:
2650 btrfs_put_block_group(block_group);
2651 return ret;
2652}
2653
2654int btrfs_exclude_logged_extents(struct extent_buffer *eb)
2655{
2656 struct btrfs_fs_info *fs_info = eb->fs_info;
2657 struct btrfs_file_extent_item *item;
2658 struct btrfs_key key;
2659 int found_type;
2660 int i;
2661 int ret = 0;
2662
2663 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
2664 return 0;
2665
2666 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2667 btrfs_item_key_to_cpu(eb, &key, i);
2668 if (key.type != BTRFS_EXTENT_DATA_KEY)
2669 continue;
2670 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
2671 found_type = btrfs_file_extent_type(eb, item);
2672 if (found_type == BTRFS_FILE_EXTENT_INLINE)
2673 continue;
2674 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
2675 continue;
2676 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
2677 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
2678 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
2679 if (ret)
2680 break;
2681 }
2682
2683 return ret;
2684}
2685
2686static void
2687btrfs_inc_block_group_reservations(struct btrfs_block_group *bg)
2688{
2689 atomic_inc(&bg->reservations);
2690}
2691
2692/*
2693 * Returns the free cluster for the given space info and sets empty_cluster to
2694 * what it should be based on the mount options.
2695 */
2696static struct btrfs_free_cluster *
2697fetch_cluster_info(struct btrfs_fs_info *fs_info,
2698 struct btrfs_space_info *space_info, u64 *empty_cluster)
2699{
2700 struct btrfs_free_cluster *ret = NULL;
2701
2702 *empty_cluster = 0;
2703 if (btrfs_mixed_space_info(space_info))
2704 return ret;
2705
2706 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
2707 ret = &fs_info->meta_alloc_cluster;
2708 if (btrfs_test_opt(fs_info, SSD))
2709 *empty_cluster = SZ_2M;
2710 else
2711 *empty_cluster = SZ_64K;
2712 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
2713 btrfs_test_opt(fs_info, SSD_SPREAD)) {
2714 *empty_cluster = SZ_2M;
2715 ret = &fs_info->data_alloc_cluster;
2716 }
2717
2718 return ret;
2719}
2720
2721static int unpin_extent_range(struct btrfs_fs_info *fs_info,
2722 u64 start, u64 end,
2723 const bool return_free_space)
2724{
2725 struct btrfs_block_group *cache = NULL;
2726 struct btrfs_space_info *space_info;
2727 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
2728 struct btrfs_free_cluster *cluster = NULL;
2729 u64 len;
2730 u64 total_unpinned = 0;
2731 u64 empty_cluster = 0;
2732 bool readonly;
2733 int ret = 0;
2734
2735 while (start <= end) {
2736 readonly = false;
2737 if (!cache ||
2738 start >= cache->start + cache->length) {
2739 if (cache)
2740 btrfs_put_block_group(cache);
2741 total_unpinned = 0;
2742 cache = btrfs_lookup_block_group(fs_info, start);
2743 if (cache == NULL) {
2744 /* Logic error, something removed the block group. */
2745 ret = -EUCLEAN;
2746 goto out;
2747 }
2748
2749 cluster = fetch_cluster_info(fs_info,
2750 cache->space_info,
2751 &empty_cluster);
2752 empty_cluster <<= 1;
2753 }
2754
2755 len = cache->start + cache->length - start;
2756 len = min(len, end + 1 - start);
2757
2758 if (return_free_space)
2759 btrfs_add_free_space(cache, start, len);
2760
2761 start += len;
2762 total_unpinned += len;
2763 space_info = cache->space_info;
2764
2765 /*
2766 * If this space cluster has been marked as fragmented and we've
2767 * unpinned enough in this block group to potentially allow a
2768 * cluster to be created inside of it go ahead and clear the
2769 * fragmented check.
2770 */
2771 if (cluster && cluster->fragmented &&
2772 total_unpinned > empty_cluster) {
2773 spin_lock(&cluster->lock);
2774 cluster->fragmented = 0;
2775 spin_unlock(&cluster->lock);
2776 }
2777
2778 spin_lock(&space_info->lock);
2779 spin_lock(&cache->lock);
2780 cache->pinned -= len;
2781 btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len);
2782 space_info->max_extent_size = 0;
2783 if (cache->ro) {
2784 space_info->bytes_readonly += len;
2785 readonly = true;
2786 } else if (btrfs_is_zoned(fs_info)) {
2787 /* Need reset before reusing in a zoned block group */
2788 btrfs_space_info_update_bytes_zone_unusable(fs_info, space_info,
2789 len);
2790 readonly = true;
2791 }
2792 spin_unlock(&cache->lock);
2793 if (!readonly && return_free_space &&
2794 global_rsv->space_info == space_info) {
2795 spin_lock(&global_rsv->lock);
2796 if (!global_rsv->full) {
2797 u64 to_add = min(len, global_rsv->size -
2798 global_rsv->reserved);
2799
2800 global_rsv->reserved += to_add;
2801 btrfs_space_info_update_bytes_may_use(fs_info,
2802 space_info, to_add);
2803 if (global_rsv->reserved >= global_rsv->size)
2804 global_rsv->full = 1;
2805 len -= to_add;
2806 }
2807 spin_unlock(&global_rsv->lock);
2808 }
2809 /* Add to any tickets we may have */
2810 if (!readonly && return_free_space && len)
2811 btrfs_try_granting_tickets(fs_info, space_info);
2812 spin_unlock(&space_info->lock);
2813 }
2814
2815 if (cache)
2816 btrfs_put_block_group(cache);
2817out:
2818 return ret;
2819}
2820
2821int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
2822{
2823 struct btrfs_fs_info *fs_info = trans->fs_info;
2824 struct btrfs_block_group *block_group, *tmp;
2825 struct list_head *deleted_bgs;
2826 struct extent_io_tree *unpin;
2827 u64 start;
2828 u64 end;
2829 int ret;
2830
2831 unpin = &trans->transaction->pinned_extents;
2832
2833 while (!TRANS_ABORTED(trans)) {
2834 struct extent_state *cached_state = NULL;
2835
2836 mutex_lock(&fs_info->unused_bg_unpin_mutex);
2837 if (!find_first_extent_bit(unpin, 0, &start, &end,
2838 EXTENT_DIRTY, &cached_state)) {
2839 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2840 break;
2841 }
2842
2843 if (btrfs_test_opt(fs_info, DISCARD_SYNC))
2844 ret = btrfs_discard_extent(fs_info, start,
2845 end + 1 - start, NULL);
2846
2847 clear_extent_dirty(unpin, start, end, &cached_state);
2848 ret = unpin_extent_range(fs_info, start, end, true);
2849 BUG_ON(ret);
2850 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2851 free_extent_state(cached_state);
2852 cond_resched();
2853 }
2854
2855 if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) {
2856 btrfs_discard_calc_delay(&fs_info->discard_ctl);
2857 btrfs_discard_schedule_work(&fs_info->discard_ctl, true);
2858 }
2859
2860 /*
2861 * Transaction is finished. We don't need the lock anymore. We
2862 * do need to clean up the block groups in case of a transaction
2863 * abort.
2864 */
2865 deleted_bgs = &trans->transaction->deleted_bgs;
2866 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
2867 u64 trimmed = 0;
2868
2869 ret = -EROFS;
2870 if (!TRANS_ABORTED(trans))
2871 ret = btrfs_discard_extent(fs_info,
2872 block_group->start,
2873 block_group->length,
2874 &trimmed);
2875
2876 list_del_init(&block_group->bg_list);
2877 btrfs_unfreeze_block_group(block_group);
2878 btrfs_put_block_group(block_group);
2879
2880 if (ret) {
2881 const char *errstr = btrfs_decode_error(ret);
2882 btrfs_warn(fs_info,
2883 "discard failed while removing blockgroup: errno=%d %s",
2884 ret, errstr);
2885 }
2886 }
2887
2888 return 0;
2889}
2890
2891/*
2892 * Parse an extent item's inline extents looking for a simple quotas owner ref.
2893 *
2894 * @fs_info: the btrfs_fs_info for this mount
2895 * @leaf: a leaf in the extent tree containing the extent item
2896 * @slot: the slot in the leaf where the extent item is found
2897 *
2898 * Returns the objectid of the root that originally allocated the extent item
2899 * if the inline owner ref is expected and present, otherwise 0.
2900 *
2901 * If an extent item has an owner ref item, it will be the first inline ref
2902 * item. Therefore the logic is to check whether there are any inline ref
2903 * items, then check the type of the first one.
2904 */
2905u64 btrfs_get_extent_owner_root(struct btrfs_fs_info *fs_info,
2906 struct extent_buffer *leaf, int slot)
2907{
2908 struct btrfs_extent_item *ei;
2909 struct btrfs_extent_inline_ref *iref;
2910 struct btrfs_extent_owner_ref *oref;
2911 unsigned long ptr;
2912 unsigned long end;
2913 int type;
2914
2915 if (!btrfs_fs_incompat(fs_info, SIMPLE_QUOTA))
2916 return 0;
2917
2918 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
2919 ptr = (unsigned long)(ei + 1);
2920 end = (unsigned long)ei + btrfs_item_size(leaf, slot);
2921
2922 /* No inline ref items of any kind, can't check type. */
2923 if (ptr == end)
2924 return 0;
2925
2926 iref = (struct btrfs_extent_inline_ref *)ptr;
2927 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
2928
2929 /* We found an owner ref, get the root out of it. */
2930 if (type == BTRFS_EXTENT_OWNER_REF_KEY) {
2931 oref = (struct btrfs_extent_owner_ref *)(&iref->offset);
2932 return btrfs_extent_owner_ref_root_id(leaf, oref);
2933 }
2934
2935 /* We have inline refs, but not an owner ref. */
2936 return 0;
2937}
2938
2939static int do_free_extent_accounting(struct btrfs_trans_handle *trans,
2940 u64 bytenr, struct btrfs_squota_delta *delta)
2941{
2942 int ret;
2943 u64 num_bytes = delta->num_bytes;
2944
2945 if (delta->is_data) {
2946 struct btrfs_root *csum_root;
2947
2948 csum_root = btrfs_csum_root(trans->fs_info, bytenr);
2949 ret = btrfs_del_csums(trans, csum_root, bytenr, num_bytes);
2950 if (ret) {
2951 btrfs_abort_transaction(trans, ret);
2952 return ret;
2953 }
2954
2955 ret = btrfs_delete_raid_extent(trans, bytenr, num_bytes);
2956 if (ret) {
2957 btrfs_abort_transaction(trans, ret);
2958 return ret;
2959 }
2960 }
2961
2962 ret = btrfs_record_squota_delta(trans->fs_info, delta);
2963 if (ret) {
2964 btrfs_abort_transaction(trans, ret);
2965 return ret;
2966 }
2967
2968 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
2969 if (ret) {
2970 btrfs_abort_transaction(trans, ret);
2971 return ret;
2972 }
2973
2974 ret = btrfs_update_block_group(trans, bytenr, num_bytes, false);
2975 if (ret)
2976 btrfs_abort_transaction(trans, ret);
2977
2978 return ret;
2979}
2980
2981#define abort_and_dump(trans, path, fmt, args...) \
2982({ \
2983 btrfs_abort_transaction(trans, -EUCLEAN); \
2984 btrfs_print_leaf(path->nodes[0]); \
2985 btrfs_crit(trans->fs_info, fmt, ##args); \
2986})
2987
2988/*
2989 * Drop one or more refs of @node.
2990 *
2991 * 1. Locate the extent refs.
2992 * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item.
2993 * Locate it, then reduce the refs number or remove the ref line completely.
2994 *
2995 * 2. Update the refs count in EXTENT/METADATA_ITEM
2996 *
2997 * Inline backref case:
2998 *
2999 * in extent tree we have:
3000 *
3001 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
3002 * refs 2 gen 6 flags DATA
3003 * extent data backref root FS_TREE objectid 258 offset 0 count 1
3004 * extent data backref root FS_TREE objectid 257 offset 0 count 1
3005 *
3006 * This function gets called with:
3007 *
3008 * node->bytenr = 13631488
3009 * node->num_bytes = 1048576
3010 * root_objectid = FS_TREE
3011 * owner_objectid = 257
3012 * owner_offset = 0
3013 * refs_to_drop = 1
3014 *
3015 * Then we should get some like:
3016 *
3017 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
3018 * refs 1 gen 6 flags DATA
3019 * extent data backref root FS_TREE objectid 258 offset 0 count 1
3020 *
3021 * Keyed backref case:
3022 *
3023 * in extent tree we have:
3024 *
3025 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
3026 * refs 754 gen 6 flags DATA
3027 * [...]
3028 * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28
3029 * extent data backref root FS_TREE objectid 866 offset 0 count 1
3030 *
3031 * This function get called with:
3032 *
3033 * node->bytenr = 13631488
3034 * node->num_bytes = 1048576
3035 * root_objectid = FS_TREE
3036 * owner_objectid = 866
3037 * owner_offset = 0
3038 * refs_to_drop = 1
3039 *
3040 * Then we should get some like:
3041 *
3042 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
3043 * refs 753 gen 6 flags DATA
3044 *
3045 * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed.
3046 */
3047static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3048 struct btrfs_delayed_ref_head *href,
3049 struct btrfs_delayed_ref_node *node,
3050 struct btrfs_delayed_extent_op *extent_op)
3051{
3052 struct btrfs_fs_info *info = trans->fs_info;
3053 struct btrfs_key key;
3054 struct btrfs_path *path;
3055 struct btrfs_root *extent_root;
3056 struct extent_buffer *leaf;
3057 struct btrfs_extent_item *ei;
3058 struct btrfs_extent_inline_ref *iref;
3059 int ret;
3060 int is_data;
3061 int extent_slot = 0;
3062 int found_extent = 0;
3063 int num_to_del = 1;
3064 int refs_to_drop = node->ref_mod;
3065 u32 item_size;
3066 u64 refs;
3067 u64 bytenr = node->bytenr;
3068 u64 num_bytes = node->num_bytes;
3069 u64 owner_objectid = btrfs_delayed_ref_owner(node);
3070 u64 owner_offset = btrfs_delayed_ref_offset(node);
3071 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
3072 u64 delayed_ref_root = href->owning_root;
3073
3074 extent_root = btrfs_extent_root(info, bytenr);
3075 ASSERT(extent_root);
3076
3077 path = btrfs_alloc_path();
3078 if (!path)
3079 return -ENOMEM;
3080
3081 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3082
3083 if (!is_data && refs_to_drop != 1) {
3084 btrfs_crit(info,
3085"invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u",
3086 node->bytenr, refs_to_drop);
3087 ret = -EINVAL;
3088 btrfs_abort_transaction(trans, ret);
3089 goto out;
3090 }
3091
3092 if (is_data)
3093 skinny_metadata = false;
3094
3095 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
3096 node->parent, node->ref_root, owner_objectid,
3097 owner_offset);
3098 if (ret == 0) {
3099 /*
3100 * Either the inline backref or the SHARED_DATA_REF/
3101 * SHARED_BLOCK_REF is found
3102 *
3103 * Here is a quick path to locate EXTENT/METADATA_ITEM.
3104 * It's possible the EXTENT/METADATA_ITEM is near current slot.
3105 */
3106 extent_slot = path->slots[0];
3107 while (extent_slot >= 0) {
3108 btrfs_item_key_to_cpu(path->nodes[0], &key,
3109 extent_slot);
3110 if (key.objectid != bytenr)
3111 break;
3112 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3113 key.offset == num_bytes) {
3114 found_extent = 1;
3115 break;
3116 }
3117 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3118 key.offset == owner_objectid) {
3119 found_extent = 1;
3120 break;
3121 }
3122
3123 /* Quick path didn't find the EXTENT/METADATA_ITEM */
3124 if (path->slots[0] - extent_slot > 5)
3125 break;
3126 extent_slot--;
3127 }
3128
3129 if (!found_extent) {
3130 if (iref) {
3131 abort_and_dump(trans, path,
3132"invalid iref slot %u, no EXTENT/METADATA_ITEM found but has inline extent ref",
3133 path->slots[0]);
3134 ret = -EUCLEAN;
3135 goto out;
3136 }
3137 /* Must be SHARED_* item, remove the backref first */
3138 ret = remove_extent_backref(trans, extent_root, path,
3139 NULL, refs_to_drop, is_data);
3140 if (ret) {
3141 btrfs_abort_transaction(trans, ret);
3142 goto out;
3143 }
3144 btrfs_release_path(path);
3145
3146 /* Slow path to locate EXTENT/METADATA_ITEM */
3147 key.objectid = bytenr;
3148 key.type = BTRFS_EXTENT_ITEM_KEY;
3149 key.offset = num_bytes;
3150
3151 if (!is_data && skinny_metadata) {
3152 key.type = BTRFS_METADATA_ITEM_KEY;
3153 key.offset = owner_objectid;
3154 }
3155
3156 ret = btrfs_search_slot(trans, extent_root,
3157 &key, path, -1, 1);
3158 if (ret > 0 && skinny_metadata && path->slots[0]) {
3159 /*
3160 * Couldn't find our skinny metadata item,
3161 * see if we have ye olde extent item.
3162 */
3163 path->slots[0]--;
3164 btrfs_item_key_to_cpu(path->nodes[0], &key,
3165 path->slots[0]);
3166 if (key.objectid == bytenr &&
3167 key.type == BTRFS_EXTENT_ITEM_KEY &&
3168 key.offset == num_bytes)
3169 ret = 0;
3170 }
3171
3172 if (ret > 0 && skinny_metadata) {
3173 skinny_metadata = false;
3174 key.objectid = bytenr;
3175 key.type = BTRFS_EXTENT_ITEM_KEY;
3176 key.offset = num_bytes;
3177 btrfs_release_path(path);
3178 ret = btrfs_search_slot(trans, extent_root,
3179 &key, path, -1, 1);
3180 }
3181
3182 if (ret) {
3183 if (ret > 0)
3184 btrfs_print_leaf(path->nodes[0]);
3185 btrfs_err(info,
3186 "umm, got %d back from search, was looking for %llu, slot %d",
3187 ret, bytenr, path->slots[0]);
3188 }
3189 if (ret < 0) {
3190 btrfs_abort_transaction(trans, ret);
3191 goto out;
3192 }
3193 extent_slot = path->slots[0];
3194 }
3195 } else if (WARN_ON(ret == -ENOENT)) {
3196 abort_and_dump(trans, path,
3197"unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu slot %d",
3198 bytenr, node->parent, node->ref_root, owner_objectid,
3199 owner_offset, path->slots[0]);
3200 goto out;
3201 } else {
3202 btrfs_abort_transaction(trans, ret);
3203 goto out;
3204 }
3205
3206 leaf = path->nodes[0];
3207 item_size = btrfs_item_size(leaf, extent_slot);
3208 if (unlikely(item_size < sizeof(*ei))) {
3209 ret = -EUCLEAN;
3210 btrfs_err(trans->fs_info,
3211 "unexpected extent item size, has %u expect >= %zu",
3212 item_size, sizeof(*ei));
3213 btrfs_abort_transaction(trans, ret);
3214 goto out;
3215 }
3216 ei = btrfs_item_ptr(leaf, extent_slot,
3217 struct btrfs_extent_item);
3218 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
3219 key.type == BTRFS_EXTENT_ITEM_KEY) {
3220 struct btrfs_tree_block_info *bi;
3221
3222 if (item_size < sizeof(*ei) + sizeof(*bi)) {
3223 abort_and_dump(trans, path,
3224"invalid extent item size for key (%llu, %u, %llu) slot %u owner %llu, has %u expect >= %zu",
3225 key.objectid, key.type, key.offset,
3226 path->slots[0], owner_objectid, item_size,
3227 sizeof(*ei) + sizeof(*bi));
3228 ret = -EUCLEAN;
3229 goto out;
3230 }
3231 bi = (struct btrfs_tree_block_info *)(ei + 1);
3232 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3233 }
3234
3235 refs = btrfs_extent_refs(leaf, ei);
3236 if (refs < refs_to_drop) {
3237 abort_and_dump(trans, path,
3238 "trying to drop %d refs but we only have %llu for bytenr %llu slot %u",
3239 refs_to_drop, refs, bytenr, path->slots[0]);
3240 ret = -EUCLEAN;
3241 goto out;
3242 }
3243 refs -= refs_to_drop;
3244
3245 if (refs > 0) {
3246 if (extent_op)
3247 __run_delayed_extent_op(extent_op, leaf, ei);
3248 /*
3249 * In the case of inline back ref, reference count will
3250 * be updated by remove_extent_backref
3251 */
3252 if (iref) {
3253 if (!found_extent) {
3254 abort_and_dump(trans, path,
3255"invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found, slot %u",
3256 path->slots[0]);
3257 ret = -EUCLEAN;
3258 goto out;
3259 }
3260 } else {
3261 btrfs_set_extent_refs(leaf, ei, refs);
3262 btrfs_mark_buffer_dirty(trans, leaf);
3263 }
3264 if (found_extent) {
3265 ret = remove_extent_backref(trans, extent_root, path,
3266 iref, refs_to_drop, is_data);
3267 if (ret) {
3268 btrfs_abort_transaction(trans, ret);
3269 goto out;
3270 }
3271 }
3272 } else {
3273 struct btrfs_squota_delta delta = {
3274 .root = delayed_ref_root,
3275 .num_bytes = num_bytes,
3276 .is_data = is_data,
3277 .is_inc = false,
3278 .generation = btrfs_extent_generation(leaf, ei),
3279 };
3280
3281 /* In this branch refs == 1 */
3282 if (found_extent) {
3283 if (is_data && refs_to_drop !=
3284 extent_data_ref_count(path, iref)) {
3285 abort_and_dump(trans, path,
3286 "invalid refs_to_drop, current refs %u refs_to_drop %u slot %u",
3287 extent_data_ref_count(path, iref),
3288 refs_to_drop, path->slots[0]);
3289 ret = -EUCLEAN;
3290 goto out;
3291 }
3292 if (iref) {
3293 if (path->slots[0] != extent_slot) {
3294 abort_and_dump(trans, path,
3295"invalid iref, extent item key (%llu %u %llu) slot %u doesn't have wanted iref",
3296 key.objectid, key.type,
3297 key.offset, path->slots[0]);
3298 ret = -EUCLEAN;
3299 goto out;
3300 }
3301 } else {
3302 /*
3303 * No inline ref, we must be at SHARED_* item,
3304 * And it's single ref, it must be:
3305 * | extent_slot ||extent_slot + 1|
3306 * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ]
3307 */
3308 if (path->slots[0] != extent_slot + 1) {
3309 abort_and_dump(trans, path,
3310 "invalid SHARED_* item slot %u, previous item is not EXTENT/METADATA_ITEM",
3311 path->slots[0]);
3312 ret = -EUCLEAN;
3313 goto out;
3314 }
3315 path->slots[0] = extent_slot;
3316 num_to_del = 2;
3317 }
3318 }
3319 /*
3320 * We can't infer the data owner from the delayed ref, so we need
3321 * to try to get it from the owning ref item.
3322 *
3323 * If it is not present, then that extent was not written under
3324 * simple quotas mode, so we don't need to account for its deletion.
3325 */
3326 if (is_data)
3327 delta.root = btrfs_get_extent_owner_root(trans->fs_info,
3328 leaf, extent_slot);
3329
3330 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3331 num_to_del);
3332 if (ret) {
3333 btrfs_abort_transaction(trans, ret);
3334 goto out;
3335 }
3336 btrfs_release_path(path);
3337
3338 ret = do_free_extent_accounting(trans, bytenr, &delta);
3339 }
3340 btrfs_release_path(path);
3341
3342out:
3343 btrfs_free_path(path);
3344 return ret;
3345}
3346
3347/*
3348 * when we free an block, it is possible (and likely) that we free the last
3349 * delayed ref for that extent as well. This searches the delayed ref tree for
3350 * a given extent, and if there are no other delayed refs to be processed, it
3351 * removes it from the tree.
3352 */
3353static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3354 u64 bytenr)
3355{
3356 struct btrfs_fs_info *fs_info = trans->fs_info;
3357 struct btrfs_delayed_ref_head *head;
3358 struct btrfs_delayed_ref_root *delayed_refs;
3359 int ret = 0;
3360
3361 delayed_refs = &trans->transaction->delayed_refs;
3362 spin_lock(&delayed_refs->lock);
3363 head = btrfs_find_delayed_ref_head(fs_info, delayed_refs, bytenr);
3364 if (!head)
3365 goto out_delayed_unlock;
3366
3367 spin_lock(&head->lock);
3368 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
3369 goto out;
3370
3371 if (cleanup_extent_op(head) != NULL)
3372 goto out;
3373
3374 /*
3375 * waiting for the lock here would deadlock. If someone else has it
3376 * locked they are already in the process of dropping it anyway
3377 */
3378 if (!mutex_trylock(&head->mutex))
3379 goto out;
3380
3381 btrfs_delete_ref_head(fs_info, delayed_refs, head);
3382 head->processing = false;
3383
3384 spin_unlock(&head->lock);
3385 spin_unlock(&delayed_refs->lock);
3386
3387 BUG_ON(head->extent_op);
3388 if (head->must_insert_reserved)
3389 ret = 1;
3390
3391 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
3392 mutex_unlock(&head->mutex);
3393 btrfs_put_delayed_ref_head(head);
3394 return ret;
3395out:
3396 spin_unlock(&head->lock);
3397
3398out_delayed_unlock:
3399 spin_unlock(&delayed_refs->lock);
3400 return 0;
3401}
3402
3403int btrfs_free_tree_block(struct btrfs_trans_handle *trans,
3404 u64 root_id,
3405 struct extent_buffer *buf,
3406 u64 parent, int last_ref)
3407{
3408 struct btrfs_fs_info *fs_info = trans->fs_info;
3409 struct btrfs_block_group *bg;
3410 int ret;
3411
3412 if (root_id != BTRFS_TREE_LOG_OBJECTID) {
3413 struct btrfs_ref generic_ref = {
3414 .action = BTRFS_DROP_DELAYED_REF,
3415 .bytenr = buf->start,
3416 .num_bytes = buf->len,
3417 .parent = parent,
3418 .owning_root = btrfs_header_owner(buf),
3419 .ref_root = root_id,
3420 };
3421
3422 /*
3423 * Assert that the extent buffer is not cleared due to
3424 * EXTENT_BUFFER_ZONED_ZEROOUT. Please refer
3425 * btrfs_clear_buffer_dirty() and btree_csum_one_bio() for
3426 * detail.
3427 */
3428 ASSERT(btrfs_header_bytenr(buf) != 0);
3429
3430 btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf), 0, false);
3431 btrfs_ref_tree_mod(fs_info, &generic_ref);
3432 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL);
3433 if (ret < 0)
3434 return ret;
3435 }
3436
3437 if (!last_ref)
3438 return 0;
3439
3440 if (btrfs_header_generation(buf) != trans->transid)
3441 goto out;
3442
3443 if (root_id != BTRFS_TREE_LOG_OBJECTID) {
3444 ret = check_ref_cleanup(trans, buf->start);
3445 if (!ret)
3446 goto out;
3447 }
3448
3449 bg = btrfs_lookup_block_group(fs_info, buf->start);
3450
3451 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3452 pin_down_extent(trans, bg, buf->start, buf->len, 1);
3453 btrfs_put_block_group(bg);
3454 goto out;
3455 }
3456
3457 /*
3458 * If there are tree mod log users we may have recorded mod log
3459 * operations for this node. If we re-allocate this node we
3460 * could replay operations on this node that happened when it
3461 * existed in a completely different root. For example if it
3462 * was part of root A, then was reallocated to root B, and we
3463 * are doing a btrfs_old_search_slot(root b), we could replay
3464 * operations that happened when the block was part of root A,
3465 * giving us an inconsistent view of the btree.
3466 *
3467 * We are safe from races here because at this point no other
3468 * node or root points to this extent buffer, so if after this
3469 * check a new tree mod log user joins we will not have an
3470 * existing log of operations on this node that we have to
3471 * contend with.
3472 */
3473
3474 if (test_bit(BTRFS_FS_TREE_MOD_LOG_USERS, &fs_info->flags)
3475 || btrfs_is_zoned(fs_info)) {
3476 pin_down_extent(trans, bg, buf->start, buf->len, 1);
3477 btrfs_put_block_group(bg);
3478 goto out;
3479 }
3480
3481 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
3482
3483 btrfs_add_free_space(bg, buf->start, buf->len);
3484 btrfs_free_reserved_bytes(bg, buf->len, 0);
3485 btrfs_put_block_group(bg);
3486 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
3487
3488out:
3489
3490 /*
3491 * Deleting the buffer, clear the corrupt flag since it doesn't
3492 * matter anymore.
3493 */
3494 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
3495 return 0;
3496}
3497
3498/* Can return -ENOMEM */
3499int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref)
3500{
3501 struct btrfs_fs_info *fs_info = trans->fs_info;
3502 int ret;
3503
3504 if (btrfs_is_testing(fs_info))
3505 return 0;
3506
3507 /*
3508 * tree log blocks never actually go into the extent allocation
3509 * tree, just update pinning info and exit early.
3510 */
3511 if (ref->ref_root == BTRFS_TREE_LOG_OBJECTID) {
3512 btrfs_pin_extent(trans, ref->bytenr, ref->num_bytes, 1);
3513 ret = 0;
3514 } else if (ref->type == BTRFS_REF_METADATA) {
3515 ret = btrfs_add_delayed_tree_ref(trans, ref, NULL);
3516 } else {
3517 ret = btrfs_add_delayed_data_ref(trans, ref, 0);
3518 }
3519
3520 if (ref->ref_root != BTRFS_TREE_LOG_OBJECTID)
3521 btrfs_ref_tree_mod(fs_info, ref);
3522
3523 return ret;
3524}
3525
3526enum btrfs_loop_type {
3527 /*
3528 * Start caching block groups but do not wait for progress or for them
3529 * to be done.
3530 */
3531 LOOP_CACHING_NOWAIT,
3532
3533 /*
3534 * Wait for the block group free_space >= the space we're waiting for if
3535 * the block group isn't cached.
3536 */
3537 LOOP_CACHING_WAIT,
3538
3539 /*
3540 * Allow allocations to happen from block groups that do not yet have a
3541 * size classification.
3542 */
3543 LOOP_UNSET_SIZE_CLASS,
3544
3545 /*
3546 * Allocate a chunk and then retry the allocation.
3547 */
3548 LOOP_ALLOC_CHUNK,
3549
3550 /*
3551 * Ignore the size class restrictions for this allocation.
3552 */
3553 LOOP_WRONG_SIZE_CLASS,
3554
3555 /*
3556 * Ignore the empty size, only try to allocate the number of bytes
3557 * needed for this allocation.
3558 */
3559 LOOP_NO_EMPTY_SIZE,
3560};
3561
3562static inline void
3563btrfs_lock_block_group(struct btrfs_block_group *cache,
3564 int delalloc)
3565{
3566 if (delalloc)
3567 down_read(&cache->data_rwsem);
3568}
3569
3570static inline void btrfs_grab_block_group(struct btrfs_block_group *cache,
3571 int delalloc)
3572{
3573 btrfs_get_block_group(cache);
3574 if (delalloc)
3575 down_read(&cache->data_rwsem);
3576}
3577
3578static struct btrfs_block_group *btrfs_lock_cluster(
3579 struct btrfs_block_group *block_group,
3580 struct btrfs_free_cluster *cluster,
3581 int delalloc)
3582 __acquires(&cluster->refill_lock)
3583{
3584 struct btrfs_block_group *used_bg = NULL;
3585
3586 spin_lock(&cluster->refill_lock);
3587 while (1) {
3588 used_bg = cluster->block_group;
3589 if (!used_bg)
3590 return NULL;
3591
3592 if (used_bg == block_group)
3593 return used_bg;
3594
3595 btrfs_get_block_group(used_bg);
3596
3597 if (!delalloc)
3598 return used_bg;
3599
3600 if (down_read_trylock(&used_bg->data_rwsem))
3601 return used_bg;
3602
3603 spin_unlock(&cluster->refill_lock);
3604
3605 /* We should only have one-level nested. */
3606 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
3607
3608 spin_lock(&cluster->refill_lock);
3609 if (used_bg == cluster->block_group)
3610 return used_bg;
3611
3612 up_read(&used_bg->data_rwsem);
3613 btrfs_put_block_group(used_bg);
3614 }
3615}
3616
3617static inline void
3618btrfs_release_block_group(struct btrfs_block_group *cache,
3619 int delalloc)
3620{
3621 if (delalloc)
3622 up_read(&cache->data_rwsem);
3623 btrfs_put_block_group(cache);
3624}
3625
3626/*
3627 * Helper function for find_free_extent().
3628 *
3629 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
3630 * Return >0 to inform caller that we find nothing
3631 * Return 0 means we have found a location and set ffe_ctl->found_offset.
3632 */
3633static int find_free_extent_clustered(struct btrfs_block_group *bg,
3634 struct find_free_extent_ctl *ffe_ctl,
3635 struct btrfs_block_group **cluster_bg_ret)
3636{
3637 struct btrfs_block_group *cluster_bg;
3638 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3639 u64 aligned_cluster;
3640 u64 offset;
3641 int ret;
3642
3643 cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
3644 if (!cluster_bg)
3645 goto refill_cluster;
3646 if (cluster_bg != bg && (cluster_bg->ro ||
3647 !block_group_bits(cluster_bg, ffe_ctl->flags)))
3648 goto release_cluster;
3649
3650 offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
3651 ffe_ctl->num_bytes, cluster_bg->start,
3652 &ffe_ctl->max_extent_size);
3653 if (offset) {
3654 /* We have a block, we're done */
3655 spin_unlock(&last_ptr->refill_lock);
3656 trace_btrfs_reserve_extent_cluster(cluster_bg, ffe_ctl);
3657 *cluster_bg_ret = cluster_bg;
3658 ffe_ctl->found_offset = offset;
3659 return 0;
3660 }
3661 WARN_ON(last_ptr->block_group != cluster_bg);
3662
3663release_cluster:
3664 /*
3665 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
3666 * lets just skip it and let the allocator find whatever block it can
3667 * find. If we reach this point, we will have tried the cluster
3668 * allocator plenty of times and not have found anything, so we are
3669 * likely way too fragmented for the clustering stuff to find anything.
3670 *
3671 * However, if the cluster is taken from the current block group,
3672 * release the cluster first, so that we stand a better chance of
3673 * succeeding in the unclustered allocation.
3674 */
3675 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
3676 spin_unlock(&last_ptr->refill_lock);
3677 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3678 return -ENOENT;
3679 }
3680
3681 /* This cluster didn't work out, free it and start over */
3682 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3683
3684 if (cluster_bg != bg)
3685 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3686
3687refill_cluster:
3688 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
3689 spin_unlock(&last_ptr->refill_lock);
3690 return -ENOENT;
3691 }
3692
3693 aligned_cluster = max_t(u64,
3694 ffe_ctl->empty_cluster + ffe_ctl->empty_size,
3695 bg->full_stripe_len);
3696 ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start,
3697 ffe_ctl->num_bytes, aligned_cluster);
3698 if (ret == 0) {
3699 /* Now pull our allocation out of this cluster */
3700 offset = btrfs_alloc_from_cluster(bg, last_ptr,
3701 ffe_ctl->num_bytes, ffe_ctl->search_start,
3702 &ffe_ctl->max_extent_size);
3703 if (offset) {
3704 /* We found one, proceed */
3705 spin_unlock(&last_ptr->refill_lock);
3706 ffe_ctl->found_offset = offset;
3707 trace_btrfs_reserve_extent_cluster(bg, ffe_ctl);
3708 return 0;
3709 }
3710 }
3711 /*
3712 * At this point we either didn't find a cluster or we weren't able to
3713 * allocate a block from our cluster. Free the cluster we've been
3714 * trying to use, and go to the next block group.
3715 */
3716 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3717 spin_unlock(&last_ptr->refill_lock);
3718 return 1;
3719}
3720
3721/*
3722 * Return >0 to inform caller that we find nothing
3723 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
3724 */
3725static int find_free_extent_unclustered(struct btrfs_block_group *bg,
3726 struct find_free_extent_ctl *ffe_ctl)
3727{
3728 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3729 u64 offset;
3730
3731 /*
3732 * We are doing an unclustered allocation, set the fragmented flag so
3733 * we don't bother trying to setup a cluster again until we get more
3734 * space.
3735 */
3736 if (unlikely(last_ptr)) {
3737 spin_lock(&last_ptr->lock);
3738 last_ptr->fragmented = 1;
3739 spin_unlock(&last_ptr->lock);
3740 }
3741 if (ffe_ctl->cached) {
3742 struct btrfs_free_space_ctl *free_space_ctl;
3743
3744 free_space_ctl = bg->free_space_ctl;
3745 spin_lock(&free_space_ctl->tree_lock);
3746 if (free_space_ctl->free_space <
3747 ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
3748 ffe_ctl->empty_size) {
3749 ffe_ctl->total_free_space = max_t(u64,
3750 ffe_ctl->total_free_space,
3751 free_space_ctl->free_space);
3752 spin_unlock(&free_space_ctl->tree_lock);
3753 return 1;
3754 }
3755 spin_unlock(&free_space_ctl->tree_lock);
3756 }
3757
3758 offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
3759 ffe_ctl->num_bytes, ffe_ctl->empty_size,
3760 &ffe_ctl->max_extent_size);
3761 if (!offset)
3762 return 1;
3763 ffe_ctl->found_offset = offset;
3764 return 0;
3765}
3766
3767static int do_allocation_clustered(struct btrfs_block_group *block_group,
3768 struct find_free_extent_ctl *ffe_ctl,
3769 struct btrfs_block_group **bg_ret)
3770{
3771 int ret;
3772
3773 /* We want to try and use the cluster allocator, so lets look there */
3774 if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) {
3775 ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret);
3776 if (ret >= 0)
3777 return ret;
3778 /* ret == -ENOENT case falls through */
3779 }
3780
3781 return find_free_extent_unclustered(block_group, ffe_ctl);
3782}
3783
3784/*
3785 * Tree-log block group locking
3786 * ============================
3787 *
3788 * fs_info::treelog_bg_lock protects the fs_info::treelog_bg which
3789 * indicates the starting address of a block group, which is reserved only
3790 * for tree-log metadata.
3791 *
3792 * Lock nesting
3793 * ============
3794 *
3795 * space_info::lock
3796 * block_group::lock
3797 * fs_info::treelog_bg_lock
3798 */
3799
3800/*
3801 * Simple allocator for sequential-only block group. It only allows sequential
3802 * allocation. No need to play with trees. This function also reserves the
3803 * bytes as in btrfs_add_reserved_bytes.
3804 */
3805static int do_allocation_zoned(struct btrfs_block_group *block_group,
3806 struct find_free_extent_ctl *ffe_ctl,
3807 struct btrfs_block_group **bg_ret)
3808{
3809 struct btrfs_fs_info *fs_info = block_group->fs_info;
3810 struct btrfs_space_info *space_info = block_group->space_info;
3811 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3812 u64 start = block_group->start;
3813 u64 num_bytes = ffe_ctl->num_bytes;
3814 u64 avail;
3815 u64 bytenr = block_group->start;
3816 u64 log_bytenr;
3817 u64 data_reloc_bytenr;
3818 int ret = 0;
3819 bool skip = false;
3820
3821 ASSERT(btrfs_is_zoned(block_group->fs_info));
3822
3823 /*
3824 * Do not allow non-tree-log blocks in the dedicated tree-log block
3825 * group, and vice versa.
3826 */
3827 spin_lock(&fs_info->treelog_bg_lock);
3828 log_bytenr = fs_info->treelog_bg;
3829 if (log_bytenr && ((ffe_ctl->for_treelog && bytenr != log_bytenr) ||
3830 (!ffe_ctl->for_treelog && bytenr == log_bytenr)))
3831 skip = true;
3832 spin_unlock(&fs_info->treelog_bg_lock);
3833 if (skip)
3834 return 1;
3835
3836 /*
3837 * Do not allow non-relocation blocks in the dedicated relocation block
3838 * group, and vice versa.
3839 */
3840 spin_lock(&fs_info->relocation_bg_lock);
3841 data_reloc_bytenr = fs_info->data_reloc_bg;
3842 if (data_reloc_bytenr &&
3843 ((ffe_ctl->for_data_reloc && bytenr != data_reloc_bytenr) ||
3844 (!ffe_ctl->for_data_reloc && bytenr == data_reloc_bytenr)))
3845 skip = true;
3846 spin_unlock(&fs_info->relocation_bg_lock);
3847 if (skip)
3848 return 1;
3849
3850 /* Check RO and no space case before trying to activate it */
3851 spin_lock(&block_group->lock);
3852 if (block_group->ro || btrfs_zoned_bg_is_full(block_group)) {
3853 ret = 1;
3854 /*
3855 * May need to clear fs_info->{treelog,data_reloc}_bg.
3856 * Return the error after taking the locks.
3857 */
3858 }
3859 spin_unlock(&block_group->lock);
3860
3861 /* Metadata block group is activated at write time. */
3862 if (!ret && (block_group->flags & BTRFS_BLOCK_GROUP_DATA) &&
3863 !btrfs_zone_activate(block_group)) {
3864 ret = 1;
3865 /*
3866 * May need to clear fs_info->{treelog,data_reloc}_bg.
3867 * Return the error after taking the locks.
3868 */
3869 }
3870
3871 spin_lock(&space_info->lock);
3872 spin_lock(&block_group->lock);
3873 spin_lock(&fs_info->treelog_bg_lock);
3874 spin_lock(&fs_info->relocation_bg_lock);
3875
3876 if (ret)
3877 goto out;
3878
3879 ASSERT(!ffe_ctl->for_treelog ||
3880 block_group->start == fs_info->treelog_bg ||
3881 fs_info->treelog_bg == 0);
3882 ASSERT(!ffe_ctl->for_data_reloc ||
3883 block_group->start == fs_info->data_reloc_bg ||
3884 fs_info->data_reloc_bg == 0);
3885
3886 if (block_group->ro ||
3887 (!ffe_ctl->for_data_reloc &&
3888 test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags))) {
3889 ret = 1;
3890 goto out;
3891 }
3892
3893 /*
3894 * Do not allow currently using block group to be tree-log dedicated
3895 * block group.
3896 */
3897 if (ffe_ctl->for_treelog && !fs_info->treelog_bg &&
3898 (block_group->used || block_group->reserved)) {
3899 ret = 1;
3900 goto out;
3901 }
3902
3903 /*
3904 * Do not allow currently used block group to be the data relocation
3905 * dedicated block group.
3906 */
3907 if (ffe_ctl->for_data_reloc && !fs_info->data_reloc_bg &&
3908 (block_group->used || block_group->reserved)) {
3909 ret = 1;
3910 goto out;
3911 }
3912
3913 WARN_ON_ONCE(block_group->alloc_offset > block_group->zone_capacity);
3914 avail = block_group->zone_capacity - block_group->alloc_offset;
3915 if (avail < num_bytes) {
3916 if (ffe_ctl->max_extent_size < avail) {
3917 /*
3918 * With sequential allocator, free space is always
3919 * contiguous
3920 */
3921 ffe_ctl->max_extent_size = avail;
3922 ffe_ctl->total_free_space = avail;
3923 }
3924 ret = 1;
3925 goto out;
3926 }
3927
3928 if (ffe_ctl->for_treelog && !fs_info->treelog_bg)
3929 fs_info->treelog_bg = block_group->start;
3930
3931 if (ffe_ctl->for_data_reloc) {
3932 if (!fs_info->data_reloc_bg)
3933 fs_info->data_reloc_bg = block_group->start;
3934 /*
3935 * Do not allow allocations from this block group, unless it is
3936 * for data relocation. Compared to increasing the ->ro, setting
3937 * the ->zoned_data_reloc_ongoing flag still allows nocow
3938 * writers to come in. See btrfs_inc_nocow_writers().
3939 *
3940 * We need to disable an allocation to avoid an allocation of
3941 * regular (non-relocation data) extent. With mix of relocation
3942 * extents and regular extents, we can dispatch WRITE commands
3943 * (for relocation extents) and ZONE APPEND commands (for
3944 * regular extents) at the same time to the same zone, which
3945 * easily break the write pointer.
3946 *
3947 * Also, this flag avoids this block group to be zone finished.
3948 */
3949 set_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags);
3950 }
3951
3952 ffe_ctl->found_offset = start + block_group->alloc_offset;
3953 block_group->alloc_offset += num_bytes;
3954 spin_lock(&ctl->tree_lock);
3955 ctl->free_space -= num_bytes;
3956 spin_unlock(&ctl->tree_lock);
3957
3958 /*
3959 * We do not check if found_offset is aligned to stripesize. The
3960 * address is anyway rewritten when using zone append writing.
3961 */
3962
3963 ffe_ctl->search_start = ffe_ctl->found_offset;
3964
3965out:
3966 if (ret && ffe_ctl->for_treelog)
3967 fs_info->treelog_bg = 0;
3968 if (ret && ffe_ctl->for_data_reloc)
3969 fs_info->data_reloc_bg = 0;
3970 spin_unlock(&fs_info->relocation_bg_lock);
3971 spin_unlock(&fs_info->treelog_bg_lock);
3972 spin_unlock(&block_group->lock);
3973 spin_unlock(&space_info->lock);
3974 return ret;
3975}
3976
3977static int do_allocation(struct btrfs_block_group *block_group,
3978 struct find_free_extent_ctl *ffe_ctl,
3979 struct btrfs_block_group **bg_ret)
3980{
3981 switch (ffe_ctl->policy) {
3982 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3983 return do_allocation_clustered(block_group, ffe_ctl, bg_ret);
3984 case BTRFS_EXTENT_ALLOC_ZONED:
3985 return do_allocation_zoned(block_group, ffe_ctl, bg_ret);
3986 default:
3987 BUG();
3988 }
3989}
3990
3991static void release_block_group(struct btrfs_block_group *block_group,
3992 struct find_free_extent_ctl *ffe_ctl,
3993 int delalloc)
3994{
3995 switch (ffe_ctl->policy) {
3996 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3997 ffe_ctl->retry_uncached = false;
3998 break;
3999 case BTRFS_EXTENT_ALLOC_ZONED:
4000 /* Nothing to do */
4001 break;
4002 default:
4003 BUG();
4004 }
4005
4006 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
4007 ffe_ctl->index);
4008 btrfs_release_block_group(block_group, delalloc);
4009}
4010
4011static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl,
4012 struct btrfs_key *ins)
4013{
4014 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
4015
4016 if (!ffe_ctl->use_cluster && last_ptr) {
4017 spin_lock(&last_ptr->lock);
4018 last_ptr->window_start = ins->objectid;
4019 spin_unlock(&last_ptr->lock);
4020 }
4021}
4022
4023static void found_extent(struct find_free_extent_ctl *ffe_ctl,
4024 struct btrfs_key *ins)
4025{
4026 switch (ffe_ctl->policy) {
4027 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4028 found_extent_clustered(ffe_ctl, ins);
4029 break;
4030 case BTRFS_EXTENT_ALLOC_ZONED:
4031 /* Nothing to do */
4032 break;
4033 default:
4034 BUG();
4035 }
4036}
4037
4038static int can_allocate_chunk_zoned(struct btrfs_fs_info *fs_info,
4039 struct find_free_extent_ctl *ffe_ctl)
4040{
4041 /* Block group's activeness is not a requirement for METADATA block groups. */
4042 if (!(ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA))
4043 return 0;
4044
4045 /* If we can activate new zone, just allocate a chunk and use it */
4046 if (btrfs_can_activate_zone(fs_info->fs_devices, ffe_ctl->flags))
4047 return 0;
4048
4049 /*
4050 * We already reached the max active zones. Try to finish one block
4051 * group to make a room for a new block group. This is only possible
4052 * for a data block group because btrfs_zone_finish() may need to wait
4053 * for a running transaction which can cause a deadlock for metadata
4054 * allocation.
4055 */
4056 if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA) {
4057 int ret = btrfs_zone_finish_one_bg(fs_info);
4058
4059 if (ret == 1)
4060 return 0;
4061 else if (ret < 0)
4062 return ret;
4063 }
4064
4065 /*
4066 * If we have enough free space left in an already active block group
4067 * and we can't activate any other zone now, do not allow allocating a
4068 * new chunk and let find_free_extent() retry with a smaller size.
4069 */
4070 if (ffe_ctl->max_extent_size >= ffe_ctl->min_alloc_size)
4071 return -ENOSPC;
4072
4073 /*
4074 * Even min_alloc_size is not left in any block groups. Since we cannot
4075 * activate a new block group, allocating it may not help. Let's tell a
4076 * caller to try again and hope it progress something by writing some
4077 * parts of the region. That is only possible for data block groups,
4078 * where a part of the region can be written.
4079 */
4080 if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA)
4081 return -EAGAIN;
4082
4083 /*
4084 * We cannot activate a new block group and no enough space left in any
4085 * block groups. So, allocating a new block group may not help. But,
4086 * there is nothing to do anyway, so let's go with it.
4087 */
4088 return 0;
4089}
4090
4091static int can_allocate_chunk(struct btrfs_fs_info *fs_info,
4092 struct find_free_extent_ctl *ffe_ctl)
4093{
4094 switch (ffe_ctl->policy) {
4095 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4096 return 0;
4097 case BTRFS_EXTENT_ALLOC_ZONED:
4098 return can_allocate_chunk_zoned(fs_info, ffe_ctl);
4099 default:
4100 BUG();
4101 }
4102}
4103
4104/*
4105 * Return >0 means caller needs to re-search for free extent
4106 * Return 0 means we have the needed free extent.
4107 * Return <0 means we failed to locate any free extent.
4108 */
4109static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
4110 struct btrfs_key *ins,
4111 struct find_free_extent_ctl *ffe_ctl,
4112 bool full_search)
4113{
4114 struct btrfs_root *root = fs_info->chunk_root;
4115 int ret;
4116
4117 if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
4118 ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
4119 ffe_ctl->orig_have_caching_bg = true;
4120
4121 if (ins->objectid) {
4122 found_extent(ffe_ctl, ins);
4123 return 0;
4124 }
4125
4126 if (ffe_ctl->loop >= LOOP_CACHING_WAIT && ffe_ctl->have_caching_bg)
4127 return 1;
4128
4129 ffe_ctl->index++;
4130 if (ffe_ctl->index < BTRFS_NR_RAID_TYPES)
4131 return 1;
4132
4133 /* See the comments for btrfs_loop_type for an explanation of the phases. */
4134 if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
4135 ffe_ctl->index = 0;
4136 /*
4137 * We want to skip the LOOP_CACHING_WAIT step if we don't have
4138 * any uncached bgs and we've already done a full search
4139 * through.
4140 */
4141 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT &&
4142 (!ffe_ctl->orig_have_caching_bg && full_search))
4143 ffe_ctl->loop++;
4144 ffe_ctl->loop++;
4145
4146 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
4147 struct btrfs_trans_handle *trans;
4148 int exist = 0;
4149
4150 /* Check if allocation policy allows to create a new chunk */
4151 ret = can_allocate_chunk(fs_info, ffe_ctl);
4152 if (ret)
4153 return ret;
4154
4155 trans = current->journal_info;
4156 if (trans)
4157 exist = 1;
4158 else
4159 trans = btrfs_join_transaction(root);
4160
4161 if (IS_ERR(trans)) {
4162 ret = PTR_ERR(trans);
4163 return ret;
4164 }
4165
4166 ret = btrfs_chunk_alloc(trans, ffe_ctl->flags,
4167 CHUNK_ALLOC_FORCE_FOR_EXTENT);
4168
4169 /* Do not bail out on ENOSPC since we can do more. */
4170 if (ret == -ENOSPC) {
4171 ret = 0;
4172 ffe_ctl->loop++;
4173 }
4174 else if (ret < 0)
4175 btrfs_abort_transaction(trans, ret);
4176 else
4177 ret = 0;
4178 if (!exist)
4179 btrfs_end_transaction(trans);
4180 if (ret)
4181 return ret;
4182 }
4183
4184 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
4185 if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED)
4186 return -ENOSPC;
4187
4188 /*
4189 * Don't loop again if we already have no empty_size and
4190 * no empty_cluster.
4191 */
4192 if (ffe_ctl->empty_size == 0 &&
4193 ffe_ctl->empty_cluster == 0)
4194 return -ENOSPC;
4195 ffe_ctl->empty_size = 0;
4196 ffe_ctl->empty_cluster = 0;
4197 }
4198 return 1;
4199 }
4200 return -ENOSPC;
4201}
4202
4203static bool find_free_extent_check_size_class(struct find_free_extent_ctl *ffe_ctl,
4204 struct btrfs_block_group *bg)
4205{
4206 if (ffe_ctl->policy == BTRFS_EXTENT_ALLOC_ZONED)
4207 return true;
4208 if (!btrfs_block_group_should_use_size_class(bg))
4209 return true;
4210 if (ffe_ctl->loop >= LOOP_WRONG_SIZE_CLASS)
4211 return true;
4212 if (ffe_ctl->loop >= LOOP_UNSET_SIZE_CLASS &&
4213 bg->size_class == BTRFS_BG_SZ_NONE)
4214 return true;
4215 return ffe_ctl->size_class == bg->size_class;
4216}
4217
4218static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info,
4219 struct find_free_extent_ctl *ffe_ctl,
4220 struct btrfs_space_info *space_info,
4221 struct btrfs_key *ins)
4222{
4223 /*
4224 * If our free space is heavily fragmented we may not be able to make
4225 * big contiguous allocations, so instead of doing the expensive search
4226 * for free space, simply return ENOSPC with our max_extent_size so we
4227 * can go ahead and search for a more manageable chunk.
4228 *
4229 * If our max_extent_size is large enough for our allocation simply
4230 * disable clustering since we will likely not be able to find enough
4231 * space to create a cluster and induce latency trying.
4232 */
4233 if (space_info->max_extent_size) {
4234 spin_lock(&space_info->lock);
4235 if (space_info->max_extent_size &&
4236 ffe_ctl->num_bytes > space_info->max_extent_size) {
4237 ins->offset = space_info->max_extent_size;
4238 spin_unlock(&space_info->lock);
4239 return -ENOSPC;
4240 } else if (space_info->max_extent_size) {
4241 ffe_ctl->use_cluster = false;
4242 }
4243 spin_unlock(&space_info->lock);
4244 }
4245
4246 ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info,
4247 &ffe_ctl->empty_cluster);
4248 if (ffe_ctl->last_ptr) {
4249 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
4250
4251 spin_lock(&last_ptr->lock);
4252 if (last_ptr->block_group)
4253 ffe_ctl->hint_byte = last_ptr->window_start;
4254 if (last_ptr->fragmented) {
4255 /*
4256 * We still set window_start so we can keep track of the
4257 * last place we found an allocation to try and save
4258 * some time.
4259 */
4260 ffe_ctl->hint_byte = last_ptr->window_start;
4261 ffe_ctl->use_cluster = false;
4262 }
4263 spin_unlock(&last_ptr->lock);
4264 }
4265
4266 return 0;
4267}
4268
4269static int prepare_allocation_zoned(struct btrfs_fs_info *fs_info,
4270 struct find_free_extent_ctl *ffe_ctl)
4271{
4272 if (ffe_ctl->for_treelog) {
4273 spin_lock(&fs_info->treelog_bg_lock);
4274 if (fs_info->treelog_bg)
4275 ffe_ctl->hint_byte = fs_info->treelog_bg;
4276 spin_unlock(&fs_info->treelog_bg_lock);
4277 } else if (ffe_ctl->for_data_reloc) {
4278 spin_lock(&fs_info->relocation_bg_lock);
4279 if (fs_info->data_reloc_bg)
4280 ffe_ctl->hint_byte = fs_info->data_reloc_bg;
4281 spin_unlock(&fs_info->relocation_bg_lock);
4282 } else if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA) {
4283 struct btrfs_block_group *block_group;
4284
4285 spin_lock(&fs_info->zone_active_bgs_lock);
4286 list_for_each_entry(block_group, &fs_info->zone_active_bgs, active_bg_list) {
4287 /*
4288 * No lock is OK here because avail is monotinically
4289 * decreasing, and this is just a hint.
4290 */
4291 u64 avail = block_group->zone_capacity - block_group->alloc_offset;
4292
4293 if (block_group_bits(block_group, ffe_ctl->flags) &&
4294 avail >= ffe_ctl->num_bytes) {
4295 ffe_ctl->hint_byte = block_group->start;
4296 break;
4297 }
4298 }
4299 spin_unlock(&fs_info->zone_active_bgs_lock);
4300 }
4301
4302 return 0;
4303}
4304
4305static int prepare_allocation(struct btrfs_fs_info *fs_info,
4306 struct find_free_extent_ctl *ffe_ctl,
4307 struct btrfs_space_info *space_info,
4308 struct btrfs_key *ins)
4309{
4310 switch (ffe_ctl->policy) {
4311 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4312 return prepare_allocation_clustered(fs_info, ffe_ctl,
4313 space_info, ins);
4314 case BTRFS_EXTENT_ALLOC_ZONED:
4315 return prepare_allocation_zoned(fs_info, ffe_ctl);
4316 default:
4317 BUG();
4318 }
4319}
4320
4321/*
4322 * walks the btree of allocated extents and find a hole of a given size.
4323 * The key ins is changed to record the hole:
4324 * ins->objectid == start position
4325 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4326 * ins->offset == the size of the hole.
4327 * Any available blocks before search_start are skipped.
4328 *
4329 * If there is no suitable free space, we will record the max size of
4330 * the free space extent currently.
4331 *
4332 * The overall logic and call chain:
4333 *
4334 * find_free_extent()
4335 * |- Iterate through all block groups
4336 * | |- Get a valid block group
4337 * | |- Try to do clustered allocation in that block group
4338 * | |- Try to do unclustered allocation in that block group
4339 * | |- Check if the result is valid
4340 * | | |- If valid, then exit
4341 * | |- Jump to next block group
4342 * |
4343 * |- Push harder to find free extents
4344 * |- If not found, re-iterate all block groups
4345 */
4346static noinline int find_free_extent(struct btrfs_root *root,
4347 struct btrfs_key *ins,
4348 struct find_free_extent_ctl *ffe_ctl)
4349{
4350 struct btrfs_fs_info *fs_info = root->fs_info;
4351 int ret = 0;
4352 int cache_block_group_error = 0;
4353 struct btrfs_block_group *block_group = NULL;
4354 struct btrfs_space_info *space_info;
4355 bool full_search = false;
4356
4357 WARN_ON(ffe_ctl->num_bytes < fs_info->sectorsize);
4358
4359 ffe_ctl->search_start = 0;
4360 /* For clustered allocation */
4361 ffe_ctl->empty_cluster = 0;
4362 ffe_ctl->last_ptr = NULL;
4363 ffe_ctl->use_cluster = true;
4364 ffe_ctl->have_caching_bg = false;
4365 ffe_ctl->orig_have_caching_bg = false;
4366 ffe_ctl->index = btrfs_bg_flags_to_raid_index(ffe_ctl->flags);
4367 ffe_ctl->loop = 0;
4368 ffe_ctl->retry_uncached = false;
4369 ffe_ctl->cached = 0;
4370 ffe_ctl->max_extent_size = 0;
4371 ffe_ctl->total_free_space = 0;
4372 ffe_ctl->found_offset = 0;
4373 ffe_ctl->policy = BTRFS_EXTENT_ALLOC_CLUSTERED;
4374 ffe_ctl->size_class = btrfs_calc_block_group_size_class(ffe_ctl->num_bytes);
4375
4376 if (btrfs_is_zoned(fs_info))
4377 ffe_ctl->policy = BTRFS_EXTENT_ALLOC_ZONED;
4378
4379 ins->type = BTRFS_EXTENT_ITEM_KEY;
4380 ins->objectid = 0;
4381 ins->offset = 0;
4382
4383 trace_find_free_extent(root, ffe_ctl);
4384
4385 space_info = btrfs_find_space_info(fs_info, ffe_ctl->flags);
4386 if (!space_info) {
4387 btrfs_err(fs_info, "No space info for %llu", ffe_ctl->flags);
4388 return -ENOSPC;
4389 }
4390
4391 ret = prepare_allocation(fs_info, ffe_ctl, space_info, ins);
4392 if (ret < 0)
4393 return ret;
4394
4395 ffe_ctl->search_start = max(ffe_ctl->search_start,
4396 first_logical_byte(fs_info));
4397 ffe_ctl->search_start = max(ffe_ctl->search_start, ffe_ctl->hint_byte);
4398 if (ffe_ctl->search_start == ffe_ctl->hint_byte) {
4399 block_group = btrfs_lookup_block_group(fs_info,
4400 ffe_ctl->search_start);
4401 /*
4402 * we don't want to use the block group if it doesn't match our
4403 * allocation bits, or if its not cached.
4404 *
4405 * However if we are re-searching with an ideal block group
4406 * picked out then we don't care that the block group is cached.
4407 */
4408 if (block_group && block_group_bits(block_group, ffe_ctl->flags) &&
4409 block_group->cached != BTRFS_CACHE_NO) {
4410 down_read(&space_info->groups_sem);
4411 if (list_empty(&block_group->list) ||
4412 block_group->ro) {
4413 /*
4414 * someone is removing this block group,
4415 * we can't jump into the have_block_group
4416 * target because our list pointers are not
4417 * valid
4418 */
4419 btrfs_put_block_group(block_group);
4420 up_read(&space_info->groups_sem);
4421 } else {
4422 ffe_ctl->index = btrfs_bg_flags_to_raid_index(
4423 block_group->flags);
4424 btrfs_lock_block_group(block_group,
4425 ffe_ctl->delalloc);
4426 ffe_ctl->hinted = true;
4427 goto have_block_group;
4428 }
4429 } else if (block_group) {
4430 btrfs_put_block_group(block_group);
4431 }
4432 }
4433search:
4434 trace_find_free_extent_search_loop(root, ffe_ctl);
4435 ffe_ctl->have_caching_bg = false;
4436 if (ffe_ctl->index == btrfs_bg_flags_to_raid_index(ffe_ctl->flags) ||
4437 ffe_ctl->index == 0)
4438 full_search = true;
4439 down_read(&space_info->groups_sem);
4440 list_for_each_entry(block_group,
4441 &space_info->block_groups[ffe_ctl->index], list) {
4442 struct btrfs_block_group *bg_ret;
4443
4444 ffe_ctl->hinted = false;
4445 /* If the block group is read-only, we can skip it entirely. */
4446 if (unlikely(block_group->ro)) {
4447 if (ffe_ctl->for_treelog)
4448 btrfs_clear_treelog_bg(block_group);
4449 if (ffe_ctl->for_data_reloc)
4450 btrfs_clear_data_reloc_bg(block_group);
4451 continue;
4452 }
4453
4454 btrfs_grab_block_group(block_group, ffe_ctl->delalloc);
4455 ffe_ctl->search_start = block_group->start;
4456
4457 /*
4458 * this can happen if we end up cycling through all the
4459 * raid types, but we want to make sure we only allocate
4460 * for the proper type.
4461 */
4462 if (!block_group_bits(block_group, ffe_ctl->flags)) {
4463 u64 extra = BTRFS_BLOCK_GROUP_DUP |
4464 BTRFS_BLOCK_GROUP_RAID1_MASK |
4465 BTRFS_BLOCK_GROUP_RAID56_MASK |
4466 BTRFS_BLOCK_GROUP_RAID10;
4467
4468 /*
4469 * if they asked for extra copies and this block group
4470 * doesn't provide them, bail. This does allow us to
4471 * fill raid0 from raid1.
4472 */
4473 if ((ffe_ctl->flags & extra) && !(block_group->flags & extra))
4474 goto loop;
4475
4476 /*
4477 * This block group has different flags than we want.
4478 * It's possible that we have MIXED_GROUP flag but no
4479 * block group is mixed. Just skip such block group.
4480 */
4481 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4482 continue;
4483 }
4484
4485have_block_group:
4486 trace_find_free_extent_have_block_group(root, ffe_ctl, block_group);
4487 ffe_ctl->cached = btrfs_block_group_done(block_group);
4488 if (unlikely(!ffe_ctl->cached)) {
4489 ffe_ctl->have_caching_bg = true;
4490 ret = btrfs_cache_block_group(block_group, false);
4491
4492 /*
4493 * If we get ENOMEM here or something else we want to
4494 * try other block groups, because it may not be fatal.
4495 * However if we can't find anything else we need to
4496 * save our return here so that we return the actual
4497 * error that caused problems, not ENOSPC.
4498 */
4499 if (ret < 0) {
4500 if (!cache_block_group_error)
4501 cache_block_group_error = ret;
4502 ret = 0;
4503 goto loop;
4504 }
4505 ret = 0;
4506 }
4507
4508 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) {
4509 if (!cache_block_group_error)
4510 cache_block_group_error = -EIO;
4511 goto loop;
4512 }
4513
4514 if (!find_free_extent_check_size_class(ffe_ctl, block_group))
4515 goto loop;
4516
4517 bg_ret = NULL;
4518 ret = do_allocation(block_group, ffe_ctl, &bg_ret);
4519 if (ret > 0)
4520 goto loop;
4521
4522 if (bg_ret && bg_ret != block_group) {
4523 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4524 block_group = bg_ret;
4525 }
4526
4527 /* Checks */
4528 ffe_ctl->search_start = round_up(ffe_ctl->found_offset,
4529 fs_info->stripesize);
4530
4531 /* move on to the next group */
4532 if (ffe_ctl->search_start + ffe_ctl->num_bytes >
4533 block_group->start + block_group->length) {
4534 btrfs_add_free_space_unused(block_group,
4535 ffe_ctl->found_offset,
4536 ffe_ctl->num_bytes);
4537 goto loop;
4538 }
4539
4540 if (ffe_ctl->found_offset < ffe_ctl->search_start)
4541 btrfs_add_free_space_unused(block_group,
4542 ffe_ctl->found_offset,
4543 ffe_ctl->search_start - ffe_ctl->found_offset);
4544
4545 ret = btrfs_add_reserved_bytes(block_group, ffe_ctl->ram_bytes,
4546 ffe_ctl->num_bytes,
4547 ffe_ctl->delalloc,
4548 ffe_ctl->loop >= LOOP_WRONG_SIZE_CLASS);
4549 if (ret == -EAGAIN) {
4550 btrfs_add_free_space_unused(block_group,
4551 ffe_ctl->found_offset,
4552 ffe_ctl->num_bytes);
4553 goto loop;
4554 }
4555 btrfs_inc_block_group_reservations(block_group);
4556
4557 /* we are all good, lets return */
4558 ins->objectid = ffe_ctl->search_start;
4559 ins->offset = ffe_ctl->num_bytes;
4560
4561 trace_btrfs_reserve_extent(block_group, ffe_ctl);
4562 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4563 break;
4564loop:
4565 if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
4566 !ffe_ctl->retry_uncached) {
4567 ffe_ctl->retry_uncached = true;
4568 btrfs_wait_block_group_cache_progress(block_group,
4569 ffe_ctl->num_bytes +
4570 ffe_ctl->empty_cluster +
4571 ffe_ctl->empty_size);
4572 goto have_block_group;
4573 }
4574 release_block_group(block_group, ffe_ctl, ffe_ctl->delalloc);
4575 cond_resched();
4576 }
4577 up_read(&space_info->groups_sem);
4578
4579 ret = find_free_extent_update_loop(fs_info, ins, ffe_ctl, full_search);
4580 if (ret > 0)
4581 goto search;
4582
4583 if (ret == -ENOSPC && !cache_block_group_error) {
4584 /*
4585 * Use ffe_ctl->total_free_space as fallback if we can't find
4586 * any contiguous hole.
4587 */
4588 if (!ffe_ctl->max_extent_size)
4589 ffe_ctl->max_extent_size = ffe_ctl->total_free_space;
4590 spin_lock(&space_info->lock);
4591 space_info->max_extent_size = ffe_ctl->max_extent_size;
4592 spin_unlock(&space_info->lock);
4593 ins->offset = ffe_ctl->max_extent_size;
4594 } else if (ret == -ENOSPC) {
4595 ret = cache_block_group_error;
4596 }
4597 return ret;
4598}
4599
4600/*
4601 * Entry point to the extent allocator. Tries to find a hole that is at least
4602 * as big as @num_bytes.
4603 *
4604 * @root - The root that will contain this extent
4605 *
4606 * @ram_bytes - The amount of space in ram that @num_bytes take. This
4607 * is used for accounting purposes. This value differs
4608 * from @num_bytes only in the case of compressed extents.
4609 *
4610 * @num_bytes - Number of bytes to allocate on-disk.
4611 *
4612 * @min_alloc_size - Indicates the minimum amount of space that the
4613 * allocator should try to satisfy. In some cases
4614 * @num_bytes may be larger than what is required and if
4615 * the filesystem is fragmented then allocation fails.
4616 * However, the presence of @min_alloc_size gives a
4617 * chance to try and satisfy the smaller allocation.
4618 *
4619 * @empty_size - A hint that you plan on doing more COW. This is the
4620 * size in bytes the allocator should try to find free
4621 * next to the block it returns. This is just a hint and
4622 * may be ignored by the allocator.
4623 *
4624 * @hint_byte - Hint to the allocator to start searching above the byte
4625 * address passed. It might be ignored.
4626 *
4627 * @ins - This key is modified to record the found hole. It will
4628 * have the following values:
4629 * ins->objectid == start position
4630 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4631 * ins->offset == the size of the hole.
4632 *
4633 * @is_data - Boolean flag indicating whether an extent is
4634 * allocated for data (true) or metadata (false)
4635 *
4636 * @delalloc - Boolean flag indicating whether this allocation is for
4637 * delalloc or not. If 'true' data_rwsem of block groups
4638 * is going to be acquired.
4639 *
4640 *
4641 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
4642 * case -ENOSPC is returned then @ins->offset will contain the size of the
4643 * largest available hole the allocator managed to find.
4644 */
4645int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
4646 u64 num_bytes, u64 min_alloc_size,
4647 u64 empty_size, u64 hint_byte,
4648 struct btrfs_key *ins, int is_data, int delalloc)
4649{
4650 struct btrfs_fs_info *fs_info = root->fs_info;
4651 struct find_free_extent_ctl ffe_ctl = {};
4652 bool final_tried = num_bytes == min_alloc_size;
4653 u64 flags;
4654 int ret;
4655 bool for_treelog = (btrfs_root_id(root) == BTRFS_TREE_LOG_OBJECTID);
4656 bool for_data_reloc = (btrfs_is_data_reloc_root(root) && is_data);
4657
4658 flags = get_alloc_profile_by_root(root, is_data);
4659again:
4660 WARN_ON(num_bytes < fs_info->sectorsize);
4661
4662 ffe_ctl.ram_bytes = ram_bytes;
4663 ffe_ctl.num_bytes = num_bytes;
4664 ffe_ctl.min_alloc_size = min_alloc_size;
4665 ffe_ctl.empty_size = empty_size;
4666 ffe_ctl.flags = flags;
4667 ffe_ctl.delalloc = delalloc;
4668 ffe_ctl.hint_byte = hint_byte;
4669 ffe_ctl.for_treelog = for_treelog;
4670 ffe_ctl.for_data_reloc = for_data_reloc;
4671
4672 ret = find_free_extent(root, ins, &ffe_ctl);
4673 if (!ret && !is_data) {
4674 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
4675 } else if (ret == -ENOSPC) {
4676 if (!final_tried && ins->offset) {
4677 num_bytes = min(num_bytes >> 1, ins->offset);
4678 num_bytes = round_down(num_bytes,
4679 fs_info->sectorsize);
4680 num_bytes = max(num_bytes, min_alloc_size);
4681 ram_bytes = num_bytes;
4682 if (num_bytes == min_alloc_size)
4683 final_tried = true;
4684 goto again;
4685 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4686 struct btrfs_space_info *sinfo;
4687
4688 sinfo = btrfs_find_space_info(fs_info, flags);
4689 btrfs_err(fs_info,
4690 "allocation failed flags %llu, wanted %llu tree-log %d, relocation: %d",
4691 flags, num_bytes, for_treelog, for_data_reloc);
4692 if (sinfo)
4693 btrfs_dump_space_info(fs_info, sinfo,
4694 num_bytes, 1);
4695 }
4696 }
4697
4698 return ret;
4699}
4700
4701int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
4702 u64 start, u64 len, int delalloc)
4703{
4704 struct btrfs_block_group *cache;
4705
4706 cache = btrfs_lookup_block_group(fs_info, start);
4707 if (!cache) {
4708 btrfs_err(fs_info, "Unable to find block group for %llu",
4709 start);
4710 return -ENOSPC;
4711 }
4712
4713 btrfs_add_free_space(cache, start, len);
4714 btrfs_free_reserved_bytes(cache, len, delalloc);
4715 trace_btrfs_reserved_extent_free(fs_info, start, len);
4716
4717 btrfs_put_block_group(cache);
4718 return 0;
4719}
4720
4721int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans,
4722 const struct extent_buffer *eb)
4723{
4724 struct btrfs_block_group *cache;
4725 int ret = 0;
4726
4727 cache = btrfs_lookup_block_group(trans->fs_info, eb->start);
4728 if (!cache) {
4729 btrfs_err(trans->fs_info, "unable to find block group for %llu",
4730 eb->start);
4731 return -ENOSPC;
4732 }
4733
4734 ret = pin_down_extent(trans, cache, eb->start, eb->len, 1);
4735 btrfs_put_block_group(cache);
4736 return ret;
4737}
4738
4739static int alloc_reserved_extent(struct btrfs_trans_handle *trans, u64 bytenr,
4740 u64 num_bytes)
4741{
4742 struct btrfs_fs_info *fs_info = trans->fs_info;
4743 int ret;
4744
4745 ret = remove_from_free_space_tree(trans, bytenr, num_bytes);
4746 if (ret)
4747 return ret;
4748
4749 ret = btrfs_update_block_group(trans, bytenr, num_bytes, true);
4750 if (ret) {
4751 ASSERT(!ret);
4752 btrfs_err(fs_info, "update block group failed for %llu %llu",
4753 bytenr, num_bytes);
4754 return ret;
4755 }
4756
4757 trace_btrfs_reserved_extent_alloc(fs_info, bytenr, num_bytes);
4758 return 0;
4759}
4760
4761static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4762 u64 parent, u64 root_objectid,
4763 u64 flags, u64 owner, u64 offset,
4764 struct btrfs_key *ins, int ref_mod, u64 oref_root)
4765{
4766 struct btrfs_fs_info *fs_info = trans->fs_info;
4767 struct btrfs_root *extent_root;
4768 int ret;
4769 struct btrfs_extent_item *extent_item;
4770 struct btrfs_extent_owner_ref *oref;
4771 struct btrfs_extent_inline_ref *iref;
4772 struct btrfs_path *path;
4773 struct extent_buffer *leaf;
4774 int type;
4775 u32 size;
4776 const bool simple_quota = (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE);
4777
4778 if (parent > 0)
4779 type = BTRFS_SHARED_DATA_REF_KEY;
4780 else
4781 type = BTRFS_EXTENT_DATA_REF_KEY;
4782
4783 size = sizeof(*extent_item);
4784 if (simple_quota)
4785 size += btrfs_extent_inline_ref_size(BTRFS_EXTENT_OWNER_REF_KEY);
4786 size += btrfs_extent_inline_ref_size(type);
4787
4788 path = btrfs_alloc_path();
4789 if (!path)
4790 return -ENOMEM;
4791
4792 extent_root = btrfs_extent_root(fs_info, ins->objectid);
4793 ret = btrfs_insert_empty_item(trans, extent_root, path, ins, size);
4794 if (ret) {
4795 btrfs_free_path(path);
4796 return ret;
4797 }
4798
4799 leaf = path->nodes[0];
4800 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4801 struct btrfs_extent_item);
4802 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4803 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4804 btrfs_set_extent_flags(leaf, extent_item,
4805 flags | BTRFS_EXTENT_FLAG_DATA);
4806
4807 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4808 if (simple_quota) {
4809 btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_EXTENT_OWNER_REF_KEY);
4810 oref = (struct btrfs_extent_owner_ref *)(&iref->offset);
4811 btrfs_set_extent_owner_ref_root_id(leaf, oref, oref_root);
4812 iref = (struct btrfs_extent_inline_ref *)(oref + 1);
4813 }
4814 btrfs_set_extent_inline_ref_type(leaf, iref, type);
4815
4816 if (parent > 0) {
4817 struct btrfs_shared_data_ref *ref;
4818 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4819 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4820 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4821 } else {
4822 struct btrfs_extent_data_ref *ref;
4823 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4824 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4825 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4826 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4827 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4828 }
4829
4830 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
4831 btrfs_free_path(path);
4832
4833 return alloc_reserved_extent(trans, ins->objectid, ins->offset);
4834}
4835
4836static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4837 struct btrfs_delayed_ref_node *node,
4838 struct btrfs_delayed_extent_op *extent_op)
4839{
4840 struct btrfs_fs_info *fs_info = trans->fs_info;
4841 struct btrfs_root *extent_root;
4842 int ret;
4843 struct btrfs_extent_item *extent_item;
4844 struct btrfs_key extent_key;
4845 struct btrfs_tree_block_info *block_info;
4846 struct btrfs_extent_inline_ref *iref;
4847 struct btrfs_path *path;
4848 struct extent_buffer *leaf;
4849 u32 size = sizeof(*extent_item) + sizeof(*iref);
4850 const u64 flags = (extent_op ? extent_op->flags_to_set : 0);
4851 /* The owner of a tree block is the level. */
4852 int level = btrfs_delayed_ref_owner(node);
4853 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4854
4855 extent_key.objectid = node->bytenr;
4856 if (skinny_metadata) {
4857 /* The owner of a tree block is the level. */
4858 extent_key.offset = level;
4859 extent_key.type = BTRFS_METADATA_ITEM_KEY;
4860 } else {
4861 extent_key.offset = node->num_bytes;
4862 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4863 size += sizeof(*block_info);
4864 }
4865
4866 path = btrfs_alloc_path();
4867 if (!path)
4868 return -ENOMEM;
4869
4870 extent_root = btrfs_extent_root(fs_info, extent_key.objectid);
4871 ret = btrfs_insert_empty_item(trans, extent_root, path, &extent_key,
4872 size);
4873 if (ret) {
4874 btrfs_free_path(path);
4875 return ret;
4876 }
4877
4878 leaf = path->nodes[0];
4879 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4880 struct btrfs_extent_item);
4881 btrfs_set_extent_refs(leaf, extent_item, 1);
4882 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4883 btrfs_set_extent_flags(leaf, extent_item,
4884 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4885
4886 if (skinny_metadata) {
4887 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4888 } else {
4889 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4890 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
4891 btrfs_set_tree_block_level(leaf, block_info, level);
4892 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4893 }
4894
4895 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
4896 btrfs_set_extent_inline_ref_type(leaf, iref,
4897 BTRFS_SHARED_BLOCK_REF_KEY);
4898 btrfs_set_extent_inline_ref_offset(leaf, iref, node->parent);
4899 } else {
4900 btrfs_set_extent_inline_ref_type(leaf, iref,
4901 BTRFS_TREE_BLOCK_REF_KEY);
4902 btrfs_set_extent_inline_ref_offset(leaf, iref, node->ref_root);
4903 }
4904
4905 btrfs_mark_buffer_dirty(trans, leaf);
4906 btrfs_free_path(path);
4907
4908 return alloc_reserved_extent(trans, node->bytenr, fs_info->nodesize);
4909}
4910
4911int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4912 struct btrfs_root *root, u64 owner,
4913 u64 offset, u64 ram_bytes,
4914 struct btrfs_key *ins)
4915{
4916 struct btrfs_ref generic_ref = {
4917 .action = BTRFS_ADD_DELAYED_EXTENT,
4918 .bytenr = ins->objectid,
4919 .num_bytes = ins->offset,
4920 .owning_root = btrfs_root_id(root),
4921 .ref_root = btrfs_root_id(root),
4922 };
4923
4924 ASSERT(generic_ref.ref_root != BTRFS_TREE_LOG_OBJECTID);
4925
4926 if (btrfs_is_data_reloc_root(root) && is_fstree(root->relocation_src_root))
4927 generic_ref.owning_root = root->relocation_src_root;
4928
4929 btrfs_init_data_ref(&generic_ref, owner, offset, 0, false);
4930 btrfs_ref_tree_mod(root->fs_info, &generic_ref);
4931
4932 return btrfs_add_delayed_data_ref(trans, &generic_ref, ram_bytes);
4933}
4934
4935/*
4936 * this is used by the tree logging recovery code. It records that
4937 * an extent has been allocated and makes sure to clear the free
4938 * space cache bits as well
4939 */
4940int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4941 u64 root_objectid, u64 owner, u64 offset,
4942 struct btrfs_key *ins)
4943{
4944 struct btrfs_fs_info *fs_info = trans->fs_info;
4945 int ret;
4946 struct btrfs_block_group *block_group;
4947 struct btrfs_space_info *space_info;
4948 struct btrfs_squota_delta delta = {
4949 .root = root_objectid,
4950 .num_bytes = ins->offset,
4951 .generation = trans->transid,
4952 .is_data = true,
4953 .is_inc = true,
4954 };
4955
4956 /*
4957 * Mixed block groups will exclude before processing the log so we only
4958 * need to do the exclude dance if this fs isn't mixed.
4959 */
4960 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
4961 ret = __exclude_logged_extent(fs_info, ins->objectid,
4962 ins->offset);
4963 if (ret)
4964 return ret;
4965 }
4966
4967 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
4968 if (!block_group)
4969 return -EINVAL;
4970
4971 space_info = block_group->space_info;
4972 spin_lock(&space_info->lock);
4973 spin_lock(&block_group->lock);
4974 space_info->bytes_reserved += ins->offset;
4975 block_group->reserved += ins->offset;
4976 spin_unlock(&block_group->lock);
4977 spin_unlock(&space_info->lock);
4978
4979 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
4980 offset, ins, 1, root_objectid);
4981 if (ret)
4982 btrfs_pin_extent(trans, ins->objectid, ins->offset, 1);
4983 ret = btrfs_record_squota_delta(fs_info, &delta);
4984 btrfs_put_block_group(block_group);
4985 return ret;
4986}
4987
4988#ifdef CONFIG_BTRFS_DEBUG
4989/*
4990 * Extra safety check in case the extent tree is corrupted and extent allocator
4991 * chooses to use a tree block which is already used and locked.
4992 */
4993static bool check_eb_lock_owner(const struct extent_buffer *eb)
4994{
4995 if (eb->lock_owner == current->pid) {
4996 btrfs_err_rl(eb->fs_info,
4997"tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
4998 eb->start, btrfs_header_owner(eb), current->pid);
4999 return true;
5000 }
5001 return false;
5002}
5003#else
5004static bool check_eb_lock_owner(struct extent_buffer *eb)
5005{
5006 return false;
5007}
5008#endif
5009
5010static struct extent_buffer *
5011btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5012 u64 bytenr, int level, u64 owner,
5013 enum btrfs_lock_nesting nest)
5014{
5015 struct btrfs_fs_info *fs_info = root->fs_info;
5016 struct extent_buffer *buf;
5017 u64 lockdep_owner = owner;
5018
5019 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner, level);
5020 if (IS_ERR(buf))
5021 return buf;
5022
5023 if (check_eb_lock_owner(buf)) {
5024 free_extent_buffer(buf);
5025 return ERR_PTR(-EUCLEAN);
5026 }
5027
5028 /*
5029 * The reloc trees are just snapshots, so we need them to appear to be
5030 * just like any other fs tree WRT lockdep.
5031 *
5032 * The exception however is in replace_path() in relocation, where we
5033 * hold the lock on the original fs root and then search for the reloc
5034 * root. At that point we need to make sure any reloc root buffers are
5035 * set to the BTRFS_TREE_RELOC_OBJECTID lockdep class in order to make
5036 * lockdep happy.
5037 */
5038 if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID &&
5039 !test_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &root->state))
5040 lockdep_owner = BTRFS_FS_TREE_OBJECTID;
5041
5042 /* btrfs_clear_buffer_dirty() accesses generation field. */
5043 btrfs_set_header_generation(buf, trans->transid);
5044
5045 /*
5046 * This needs to stay, because we could allocate a freed block from an
5047 * old tree into a new tree, so we need to make sure this new block is
5048 * set to the appropriate level and owner.
5049 */
5050 btrfs_set_buffer_lockdep_class(lockdep_owner, buf, level);
5051
5052 btrfs_tree_lock_nested(buf, nest);
5053 btrfs_clear_buffer_dirty(trans, buf);
5054 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
5055 clear_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &buf->bflags);
5056
5057 set_extent_buffer_uptodate(buf);
5058
5059 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
5060 btrfs_set_header_level(buf, level);
5061 btrfs_set_header_bytenr(buf, buf->start);
5062 btrfs_set_header_generation(buf, trans->transid);
5063 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
5064 btrfs_set_header_owner(buf, owner);
5065 write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
5066 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
5067 if (btrfs_root_id(root) == BTRFS_TREE_LOG_OBJECTID) {
5068 buf->log_index = root->log_transid % 2;
5069 /*
5070 * we allow two log transactions at a time, use different
5071 * EXTENT bit to differentiate dirty pages.
5072 */
5073 if (buf->log_index == 0)
5074 set_extent_bit(&root->dirty_log_pages, buf->start,
5075 buf->start + buf->len - 1,
5076 EXTENT_DIRTY, NULL);
5077 else
5078 set_extent_bit(&root->dirty_log_pages, buf->start,
5079 buf->start + buf->len - 1,
5080 EXTENT_NEW, NULL);
5081 } else {
5082 buf->log_index = -1;
5083 set_extent_bit(&trans->transaction->dirty_pages, buf->start,
5084 buf->start + buf->len - 1, EXTENT_DIRTY, NULL);
5085 }
5086 /* this returns a buffer locked for blocking */
5087 return buf;
5088}
5089
5090/*
5091 * finds a free extent and does all the dirty work required for allocation
5092 * returns the tree buffer or an ERR_PTR on error.
5093 */
5094struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
5095 struct btrfs_root *root,
5096 u64 parent, u64 root_objectid,
5097 const struct btrfs_disk_key *key,
5098 int level, u64 hint,
5099 u64 empty_size,
5100 u64 reloc_src_root,
5101 enum btrfs_lock_nesting nest)
5102{
5103 struct btrfs_fs_info *fs_info = root->fs_info;
5104 struct btrfs_key ins;
5105 struct btrfs_block_rsv *block_rsv;
5106 struct extent_buffer *buf;
5107 u64 flags = 0;
5108 int ret;
5109 u32 blocksize = fs_info->nodesize;
5110 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
5111 u64 owning_root;
5112
5113#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5114 if (btrfs_is_testing(fs_info)) {
5115 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
5116 level, root_objectid, nest);
5117 if (!IS_ERR(buf))
5118 root->alloc_bytenr += blocksize;
5119 return buf;
5120 }
5121#endif
5122
5123 block_rsv = btrfs_use_block_rsv(trans, root, blocksize);
5124 if (IS_ERR(block_rsv))
5125 return ERR_CAST(block_rsv);
5126
5127 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
5128 empty_size, hint, &ins, 0, 0);
5129 if (ret)
5130 goto out_unuse;
5131
5132 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
5133 root_objectid, nest);
5134 if (IS_ERR(buf)) {
5135 ret = PTR_ERR(buf);
5136 goto out_free_reserved;
5137 }
5138 owning_root = btrfs_header_owner(buf);
5139
5140 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5141 if (parent == 0)
5142 parent = ins.objectid;
5143 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5144 owning_root = reloc_src_root;
5145 } else
5146 BUG_ON(parent > 0);
5147
5148 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5149 struct btrfs_delayed_extent_op *extent_op;
5150 struct btrfs_ref generic_ref = {
5151 .action = BTRFS_ADD_DELAYED_EXTENT,
5152 .bytenr = ins.objectid,
5153 .num_bytes = ins.offset,
5154 .parent = parent,
5155 .owning_root = owning_root,
5156 .ref_root = root_objectid,
5157 };
5158
5159 if (!skinny_metadata || flags != 0) {
5160 extent_op = btrfs_alloc_delayed_extent_op();
5161 if (!extent_op) {
5162 ret = -ENOMEM;
5163 goto out_free_buf;
5164 }
5165 if (key)
5166 memcpy(&extent_op->key, key, sizeof(extent_op->key));
5167 else
5168 memset(&extent_op->key, 0, sizeof(extent_op->key));
5169 extent_op->flags_to_set = flags;
5170 extent_op->update_key = (skinny_metadata ? false : true);
5171 extent_op->update_flags = (flags != 0);
5172 } else {
5173 extent_op = NULL;
5174 }
5175
5176 btrfs_init_tree_ref(&generic_ref, level, btrfs_root_id(root), false);
5177 btrfs_ref_tree_mod(fs_info, &generic_ref);
5178 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, extent_op);
5179 if (ret) {
5180 btrfs_free_delayed_extent_op(extent_op);
5181 goto out_free_buf;
5182 }
5183 }
5184 return buf;
5185
5186out_free_buf:
5187 btrfs_tree_unlock(buf);
5188 free_extent_buffer(buf);
5189out_free_reserved:
5190 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
5191out_unuse:
5192 btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize);
5193 return ERR_PTR(ret);
5194}
5195
5196struct walk_control {
5197 u64 refs[BTRFS_MAX_LEVEL];
5198 u64 flags[BTRFS_MAX_LEVEL];
5199 struct btrfs_key update_progress;
5200 struct btrfs_key drop_progress;
5201 int drop_level;
5202 int stage;
5203 int level;
5204 int shared_level;
5205 int update_ref;
5206 int keep_locks;
5207 int reada_slot;
5208 int reada_count;
5209 int restarted;
5210 /* Indicate that extent info needs to be looked up when walking the tree. */
5211 int lookup_info;
5212};
5213
5214/*
5215 * This is our normal stage. We are traversing blocks the current snapshot owns
5216 * and we are dropping any of our references to any children we are able to, and
5217 * then freeing the block once we've processed all of the children.
5218 */
5219#define DROP_REFERENCE 1
5220
5221/*
5222 * We enter this stage when we have to walk into a child block (meaning we can't
5223 * simply drop our reference to it from our current parent node) and there are
5224 * more than one reference on it. If we are the owner of any of the children
5225 * blocks from the current parent node then we have to do the FULL_BACKREF dance
5226 * on them in order to drop our normal ref and add the shared ref.
5227 */
5228#define UPDATE_BACKREF 2
5229
5230/*
5231 * Decide if we need to walk down into this node to adjust the references.
5232 *
5233 * @root: the root we are currently deleting
5234 * @wc: the walk control for this deletion
5235 * @eb: the parent eb that we're currently visiting
5236 * @refs: the number of refs for wc->level - 1
5237 * @flags: the flags for wc->level - 1
5238 * @slot: the slot in the eb that we're currently checking
5239 *
5240 * This is meant to be called when we're evaluating if a node we point to at
5241 * wc->level should be read and walked into, or if we can simply delete our
5242 * reference to it. We return true if we should walk into the node, false if we
5243 * can skip it.
5244 *
5245 * We have assertions in here to make sure this is called correctly. We assume
5246 * that sanity checking on the blocks read to this point has been done, so any
5247 * corrupted file systems must have been caught before calling this function.
5248 */
5249static bool visit_node_for_delete(struct btrfs_root *root, struct walk_control *wc,
5250 struct extent_buffer *eb, u64 flags, int slot)
5251{
5252 struct btrfs_key key;
5253 u64 generation;
5254 int level = wc->level;
5255
5256 ASSERT(level > 0);
5257 ASSERT(wc->refs[level - 1] > 0);
5258
5259 /*
5260 * The update backref stage we only want to skip if we already have
5261 * FULL_BACKREF set, otherwise we need to read.
5262 */
5263 if (wc->stage == UPDATE_BACKREF) {
5264 if (level == 1 && flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5265 return false;
5266 return true;
5267 }
5268
5269 /*
5270 * We're the last ref on this block, we must walk into it and process
5271 * any refs it's pointing at.
5272 */
5273 if (wc->refs[level - 1] == 1)
5274 return true;
5275
5276 /*
5277 * If we're already FULL_BACKREF then we know we can just drop our
5278 * current reference.
5279 */
5280 if (level == 1 && flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5281 return false;
5282
5283 /*
5284 * This block is older than our creation generation, we can drop our
5285 * reference to it.
5286 */
5287 generation = btrfs_node_ptr_generation(eb, slot);
5288 if (!wc->update_ref || generation <= btrfs_root_origin_generation(root))
5289 return false;
5290
5291 /*
5292 * This block was processed from a previous snapshot deletion run, we
5293 * can skip it.
5294 */
5295 btrfs_node_key_to_cpu(eb, &key, slot);
5296 if (btrfs_comp_cpu_keys(&key, &wc->update_progress) < 0)
5297 return false;
5298
5299 /* All other cases we need to wander into the node. */
5300 return true;
5301}
5302
5303static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5304 struct btrfs_root *root,
5305 struct walk_control *wc,
5306 struct btrfs_path *path)
5307{
5308 struct btrfs_fs_info *fs_info = root->fs_info;
5309 u64 bytenr;
5310 u64 generation;
5311 u64 refs;
5312 u64 flags;
5313 u32 nritems;
5314 struct extent_buffer *eb;
5315 int ret;
5316 int slot;
5317 int nread = 0;
5318
5319 if (path->slots[wc->level] < wc->reada_slot) {
5320 wc->reada_count = wc->reada_count * 2 / 3;
5321 wc->reada_count = max(wc->reada_count, 2);
5322 } else {
5323 wc->reada_count = wc->reada_count * 3 / 2;
5324 wc->reada_count = min_t(int, wc->reada_count,
5325 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
5326 }
5327
5328 eb = path->nodes[wc->level];
5329 nritems = btrfs_header_nritems(eb);
5330
5331 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5332 if (nread >= wc->reada_count)
5333 break;
5334
5335 cond_resched();
5336 bytenr = btrfs_node_blockptr(eb, slot);
5337 generation = btrfs_node_ptr_generation(eb, slot);
5338
5339 if (slot == path->slots[wc->level])
5340 goto reada;
5341
5342 if (wc->stage == UPDATE_BACKREF &&
5343 generation <= btrfs_root_origin_generation(root))
5344 continue;
5345
5346 /* We don't lock the tree block, it's OK to be racy here */
5347 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
5348 wc->level - 1, 1, &refs,
5349 &flags, NULL);
5350 /* We don't care about errors in readahead. */
5351 if (ret < 0)
5352 continue;
5353
5354 /*
5355 * This could be racey, it's conceivable that we raced and end
5356 * up with a bogus refs count, if that's the case just skip, if
5357 * we are actually corrupt we will notice when we look up
5358 * everything again with our locks.
5359 */
5360 if (refs == 0)
5361 continue;
5362
5363 /* If we don't need to visit this node don't reada. */
5364 if (!visit_node_for_delete(root, wc, eb, flags, slot))
5365 continue;
5366reada:
5367 btrfs_readahead_node_child(eb, slot);
5368 nread++;
5369 }
5370 wc->reada_slot = slot;
5371}
5372
5373/*
5374 * helper to process tree block while walking down the tree.
5375 *
5376 * when wc->stage == UPDATE_BACKREF, this function updates
5377 * back refs for pointers in the block.
5378 *
5379 * NOTE: return value 1 means we should stop walking down.
5380 */
5381static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5382 struct btrfs_root *root,
5383 struct btrfs_path *path,
5384 struct walk_control *wc)
5385{
5386 struct btrfs_fs_info *fs_info = root->fs_info;
5387 int level = wc->level;
5388 struct extent_buffer *eb = path->nodes[level];
5389 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5390 int ret;
5391
5392 if (wc->stage == UPDATE_BACKREF && btrfs_header_owner(eb) != btrfs_root_id(root))
5393 return 1;
5394
5395 /*
5396 * when reference count of tree block is 1, it won't increase
5397 * again. once full backref flag is set, we never clear it.
5398 */
5399 if (wc->lookup_info &&
5400 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5401 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5402 ASSERT(path->locks[level]);
5403 ret = btrfs_lookup_extent_info(trans, fs_info,
5404 eb->start, level, 1,
5405 &wc->refs[level],
5406 &wc->flags[level],
5407 NULL);
5408 if (ret)
5409 return ret;
5410 if (unlikely(wc->refs[level] == 0)) {
5411 btrfs_err(fs_info, "bytenr %llu has 0 references, expect > 0",
5412 eb->start);
5413 return -EUCLEAN;
5414 }
5415 }
5416
5417 if (wc->stage == DROP_REFERENCE) {
5418 if (wc->refs[level] > 1)
5419 return 1;
5420
5421 if (path->locks[level] && !wc->keep_locks) {
5422 btrfs_tree_unlock_rw(eb, path->locks[level]);
5423 path->locks[level] = 0;
5424 }
5425 return 0;
5426 }
5427
5428 /* wc->stage == UPDATE_BACKREF */
5429 if (!(wc->flags[level] & flag)) {
5430 ASSERT(path->locks[level]);
5431 ret = btrfs_inc_ref(trans, root, eb, 1);
5432 if (ret) {
5433 btrfs_abort_transaction(trans, ret);
5434 return ret;
5435 }
5436 ret = btrfs_dec_ref(trans, root, eb, 0);
5437 if (ret) {
5438 btrfs_abort_transaction(trans, ret);
5439 return ret;
5440 }
5441 ret = btrfs_set_disk_extent_flags(trans, eb, flag);
5442 if (ret) {
5443 btrfs_abort_transaction(trans, ret);
5444 return ret;
5445 }
5446 wc->flags[level] |= flag;
5447 }
5448
5449 /*
5450 * the block is shared by multiple trees, so it's not good to
5451 * keep the tree lock
5452 */
5453 if (path->locks[level] && level > 0) {
5454 btrfs_tree_unlock_rw(eb, path->locks[level]);
5455 path->locks[level] = 0;
5456 }
5457 return 0;
5458}
5459
5460/*
5461 * This is used to verify a ref exists for this root to deal with a bug where we
5462 * would have a drop_progress key that hadn't been updated properly.
5463 */
5464static int check_ref_exists(struct btrfs_trans_handle *trans,
5465 struct btrfs_root *root, u64 bytenr, u64 parent,
5466 int level)
5467{
5468 struct btrfs_delayed_ref_root *delayed_refs;
5469 struct btrfs_delayed_ref_head *head;
5470 struct btrfs_path *path;
5471 struct btrfs_extent_inline_ref *iref;
5472 int ret;
5473 bool exists = false;
5474
5475 path = btrfs_alloc_path();
5476 if (!path)
5477 return -ENOMEM;
5478again:
5479 ret = lookup_extent_backref(trans, path, &iref, bytenr,
5480 root->fs_info->nodesize, parent,
5481 btrfs_root_id(root), level, 0);
5482 if (ret != -ENOENT) {
5483 /*
5484 * If we get 0 then we found our reference, return 1, else
5485 * return the error if it's not -ENOENT;
5486 */
5487 btrfs_free_path(path);
5488 return (ret < 0 ) ? ret : 1;
5489 }
5490
5491 /*
5492 * We could have a delayed ref with this reference, so look it up while
5493 * we're holding the path open to make sure we don't race with the
5494 * delayed ref running.
5495 */
5496 delayed_refs = &trans->transaction->delayed_refs;
5497 spin_lock(&delayed_refs->lock);
5498 head = btrfs_find_delayed_ref_head(root->fs_info, delayed_refs, bytenr);
5499 if (!head)
5500 goto out;
5501 if (!mutex_trylock(&head->mutex)) {
5502 /*
5503 * We're contended, means that the delayed ref is running, get a
5504 * reference and wait for the ref head to be complete and then
5505 * try again.
5506 */
5507 refcount_inc(&head->refs);
5508 spin_unlock(&delayed_refs->lock);
5509
5510 btrfs_release_path(path);
5511
5512 mutex_lock(&head->mutex);
5513 mutex_unlock(&head->mutex);
5514 btrfs_put_delayed_ref_head(head);
5515 goto again;
5516 }
5517
5518 exists = btrfs_find_delayed_tree_ref(head, root->root_key.objectid, parent);
5519 mutex_unlock(&head->mutex);
5520out:
5521 spin_unlock(&delayed_refs->lock);
5522 btrfs_free_path(path);
5523 return exists ? 1 : 0;
5524}
5525
5526/*
5527 * We may not have an uptodate block, so if we are going to walk down into this
5528 * block we need to drop the lock, read it off of the disk, re-lock it and
5529 * return to continue dropping the snapshot.
5530 */
5531static int check_next_block_uptodate(struct btrfs_trans_handle *trans,
5532 struct btrfs_root *root,
5533 struct btrfs_path *path,
5534 struct walk_control *wc,
5535 struct extent_buffer *next)
5536{
5537 struct btrfs_tree_parent_check check = { 0 };
5538 u64 generation;
5539 int level = wc->level;
5540 int ret;
5541
5542 btrfs_assert_tree_write_locked(next);
5543
5544 generation = btrfs_node_ptr_generation(path->nodes[level], path->slots[level]);
5545
5546 if (btrfs_buffer_uptodate(next, generation, 0))
5547 return 0;
5548
5549 check.level = level - 1;
5550 check.transid = generation;
5551 check.owner_root = btrfs_root_id(root);
5552 check.has_first_key = true;
5553 btrfs_node_key_to_cpu(path->nodes[level], &check.first_key, path->slots[level]);
5554
5555 btrfs_tree_unlock(next);
5556 if (level == 1)
5557 reada_walk_down(trans, root, wc, path);
5558 ret = btrfs_read_extent_buffer(next, &check);
5559 if (ret) {
5560 free_extent_buffer(next);
5561 return ret;
5562 }
5563 btrfs_tree_lock(next);
5564 wc->lookup_info = 1;
5565 return 0;
5566}
5567
5568/*
5569 * If we determine that we don't have to visit wc->level - 1 then we need to
5570 * determine if we can drop our reference.
5571 *
5572 * If we are UPDATE_BACKREF then we will not, we need to update our backrefs.
5573 *
5574 * If we are DROP_REFERENCE this will figure out if we need to drop our current
5575 * reference, skipping it if we dropped it from a previous incompleted drop, or
5576 * dropping it if we still have a reference to it.
5577 */
5578static int maybe_drop_reference(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5579 struct btrfs_path *path, struct walk_control *wc,
5580 struct extent_buffer *next, u64 owner_root)
5581{
5582 struct btrfs_ref ref = {
5583 .action = BTRFS_DROP_DELAYED_REF,
5584 .bytenr = next->start,
5585 .num_bytes = root->fs_info->nodesize,
5586 .owning_root = owner_root,
5587 .ref_root = btrfs_root_id(root),
5588 };
5589 int level = wc->level;
5590 int ret;
5591
5592 /* We are UPDATE_BACKREF, we're not dropping anything. */
5593 if (wc->stage == UPDATE_BACKREF)
5594 return 0;
5595
5596 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5597 ref.parent = path->nodes[level]->start;
5598 } else {
5599 ASSERT(btrfs_root_id(root) == btrfs_header_owner(path->nodes[level]));
5600 if (btrfs_root_id(root) != btrfs_header_owner(path->nodes[level])) {
5601 btrfs_err(root->fs_info, "mismatched block owner");
5602 return -EIO;
5603 }
5604 }
5605
5606 /*
5607 * If we had a drop_progress we need to verify the refs are set as
5608 * expected. If we find our ref then we know that from here on out
5609 * everything should be correct, and we can clear the
5610 * ->restarted flag.
5611 */
5612 if (wc->restarted) {
5613 ret = check_ref_exists(trans, root, next->start, ref.parent,
5614 level - 1);
5615 if (ret <= 0)
5616 return ret;
5617 ret = 0;
5618 wc->restarted = 0;
5619 }
5620
5621 /*
5622 * Reloc tree doesn't contribute to qgroup numbers, and we have already
5623 * accounted them at merge time (replace_path), thus we could skip
5624 * expensive subtree trace here.
5625 */
5626 if (btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID &&
5627 wc->refs[level - 1] > 1) {
5628 u64 generation = btrfs_node_ptr_generation(path->nodes[level],
5629 path->slots[level]);
5630
5631 ret = btrfs_qgroup_trace_subtree(trans, next, generation, level - 1);
5632 if (ret) {
5633 btrfs_err_rl(root->fs_info,
5634"error %d accounting shared subtree, quota is out of sync, rescan required",
5635 ret);
5636 }
5637 }
5638
5639 /*
5640 * We need to update the next key in our walk control so we can update
5641 * the drop_progress key accordingly. We don't care if find_next_key
5642 * doesn't find a key because that means we're at the end and are going
5643 * to clean up now.
5644 */
5645 wc->drop_level = level;
5646 find_next_key(path, level, &wc->drop_progress);
5647
5648 btrfs_init_tree_ref(&ref, level - 1, 0, false);
5649 return btrfs_free_extent(trans, &ref);
5650}
5651
5652/*
5653 * helper to process tree block pointer.
5654 *
5655 * when wc->stage == DROP_REFERENCE, this function checks
5656 * reference count of the block pointed to. if the block
5657 * is shared and we need update back refs for the subtree
5658 * rooted at the block, this function changes wc->stage to
5659 * UPDATE_BACKREF. if the block is shared and there is no
5660 * need to update back, this function drops the reference
5661 * to the block.
5662 *
5663 * NOTE: return value 1 means we should stop walking down.
5664 */
5665static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5666 struct btrfs_root *root,
5667 struct btrfs_path *path,
5668 struct walk_control *wc)
5669{
5670 struct btrfs_fs_info *fs_info = root->fs_info;
5671 u64 bytenr;
5672 u64 generation;
5673 u64 owner_root = 0;
5674 struct extent_buffer *next;
5675 int level = wc->level;
5676 int ret = 0;
5677
5678 generation = btrfs_node_ptr_generation(path->nodes[level],
5679 path->slots[level]);
5680 /*
5681 * if the lower level block was created before the snapshot
5682 * was created, we know there is no need to update back refs
5683 * for the subtree
5684 */
5685 if (wc->stage == UPDATE_BACKREF &&
5686 generation <= btrfs_root_origin_generation(root)) {
5687 wc->lookup_info = 1;
5688 return 1;
5689 }
5690
5691 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5692
5693 next = btrfs_find_create_tree_block(fs_info, bytenr, btrfs_root_id(root),
5694 level - 1);
5695 if (IS_ERR(next))
5696 return PTR_ERR(next);
5697
5698 btrfs_tree_lock(next);
5699
5700 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
5701 &wc->refs[level - 1],
5702 &wc->flags[level - 1],
5703 &owner_root);
5704 if (ret < 0)
5705 goto out_unlock;
5706
5707 if (unlikely(wc->refs[level - 1] == 0)) {
5708 btrfs_err(fs_info, "bytenr %llu has 0 references, expect > 0",
5709 bytenr);
5710 ret = -EUCLEAN;
5711 goto out_unlock;
5712 }
5713 wc->lookup_info = 0;
5714
5715 /* If we don't have to walk into this node skip it. */
5716 if (!visit_node_for_delete(root, wc, path->nodes[level],
5717 wc->flags[level - 1], path->slots[level]))
5718 goto skip;
5719
5720 /*
5721 * We have to walk down into this node, and if we're currently at the
5722 * DROP_REFERNCE stage and this block is shared then we need to switch
5723 * to the UPDATE_BACKREF stage in order to convert to FULL_BACKREF.
5724 */
5725 if (wc->stage == DROP_REFERENCE && wc->refs[level - 1] > 1) {
5726 wc->stage = UPDATE_BACKREF;
5727 wc->shared_level = level - 1;
5728 }
5729
5730 ret = check_next_block_uptodate(trans, root, path, wc, next);
5731 if (ret)
5732 return ret;
5733
5734 level--;
5735 ASSERT(level == btrfs_header_level(next));
5736 if (level != btrfs_header_level(next)) {
5737 btrfs_err(root->fs_info, "mismatched level");
5738 ret = -EIO;
5739 goto out_unlock;
5740 }
5741 path->nodes[level] = next;
5742 path->slots[level] = 0;
5743 path->locks[level] = BTRFS_WRITE_LOCK;
5744 wc->level = level;
5745 if (wc->level == 1)
5746 wc->reada_slot = 0;
5747 return 0;
5748skip:
5749 ret = maybe_drop_reference(trans, root, path, wc, next, owner_root);
5750 if (ret)
5751 goto out_unlock;
5752 wc->refs[level - 1] = 0;
5753 wc->flags[level - 1] = 0;
5754 wc->lookup_info = 1;
5755 ret = 1;
5756
5757out_unlock:
5758 btrfs_tree_unlock(next);
5759 free_extent_buffer(next);
5760
5761 return ret;
5762}
5763
5764/*
5765 * helper to process tree block while walking up the tree.
5766 *
5767 * when wc->stage == DROP_REFERENCE, this function drops
5768 * reference count on the block.
5769 *
5770 * when wc->stage == UPDATE_BACKREF, this function changes
5771 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5772 * to UPDATE_BACKREF previously while processing the block.
5773 *
5774 * NOTE: return value 1 means we should stop walking up.
5775 */
5776static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5777 struct btrfs_root *root,
5778 struct btrfs_path *path,
5779 struct walk_control *wc)
5780{
5781 struct btrfs_fs_info *fs_info = root->fs_info;
5782 int ret = 0;
5783 int level = wc->level;
5784 struct extent_buffer *eb = path->nodes[level];
5785 u64 parent = 0;
5786
5787 if (wc->stage == UPDATE_BACKREF) {
5788 ASSERT(wc->shared_level >= level);
5789 if (level < wc->shared_level)
5790 goto out;
5791
5792 ret = find_next_key(path, level + 1, &wc->update_progress);
5793 if (ret > 0)
5794 wc->update_ref = 0;
5795
5796 wc->stage = DROP_REFERENCE;
5797 wc->shared_level = -1;
5798 path->slots[level] = 0;
5799
5800 /*
5801 * check reference count again if the block isn't locked.
5802 * we should start walking down the tree again if reference
5803 * count is one.
5804 */
5805 if (!path->locks[level]) {
5806 ASSERT(level > 0);
5807 btrfs_tree_lock(eb);
5808 path->locks[level] = BTRFS_WRITE_LOCK;
5809
5810 ret = btrfs_lookup_extent_info(trans, fs_info,
5811 eb->start, level, 1,
5812 &wc->refs[level],
5813 &wc->flags[level],
5814 NULL);
5815 if (ret < 0) {
5816 btrfs_tree_unlock_rw(eb, path->locks[level]);
5817 path->locks[level] = 0;
5818 return ret;
5819 }
5820 if (unlikely(wc->refs[level] == 0)) {
5821 btrfs_tree_unlock_rw(eb, path->locks[level]);
5822 btrfs_err(fs_info, "bytenr %llu has 0 references, expect > 0",
5823 eb->start);
5824 return -EUCLEAN;
5825 }
5826 if (wc->refs[level] == 1) {
5827 btrfs_tree_unlock_rw(eb, path->locks[level]);
5828 path->locks[level] = 0;
5829 return 1;
5830 }
5831 }
5832 }
5833
5834 /* wc->stage == DROP_REFERENCE */
5835 ASSERT(path->locks[level] || wc->refs[level] == 1);
5836
5837 if (wc->refs[level] == 1) {
5838 if (level == 0) {
5839 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5840 ret = btrfs_dec_ref(trans, root, eb, 1);
5841 else
5842 ret = btrfs_dec_ref(trans, root, eb, 0);
5843 if (ret) {
5844 btrfs_abort_transaction(trans, ret);
5845 return ret;
5846 }
5847 if (is_fstree(btrfs_root_id(root))) {
5848 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
5849 if (ret) {
5850 btrfs_err_rl(fs_info,
5851 "error %d accounting leaf items, quota is out of sync, rescan required",
5852 ret);
5853 }
5854 }
5855 }
5856 /* Make block locked assertion in btrfs_clear_buffer_dirty happy. */
5857 if (!path->locks[level]) {
5858 btrfs_tree_lock(eb);
5859 path->locks[level] = BTRFS_WRITE_LOCK;
5860 }
5861 btrfs_clear_buffer_dirty(trans, eb);
5862 }
5863
5864 if (eb == root->node) {
5865 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5866 parent = eb->start;
5867 else if (btrfs_root_id(root) != btrfs_header_owner(eb))
5868 goto owner_mismatch;
5869 } else {
5870 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5871 parent = path->nodes[level + 1]->start;
5872 else if (btrfs_root_id(root) !=
5873 btrfs_header_owner(path->nodes[level + 1]))
5874 goto owner_mismatch;
5875 }
5876
5877 ret = btrfs_free_tree_block(trans, btrfs_root_id(root), eb, parent,
5878 wc->refs[level] == 1);
5879 if (ret < 0)
5880 btrfs_abort_transaction(trans, ret);
5881out:
5882 wc->refs[level] = 0;
5883 wc->flags[level] = 0;
5884 return ret;
5885
5886owner_mismatch:
5887 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
5888 btrfs_header_owner(eb), btrfs_root_id(root));
5889 return -EUCLEAN;
5890}
5891
5892/*
5893 * walk_down_tree consists of two steps.
5894 *
5895 * walk_down_proc(). Look up the reference count and reference of our current
5896 * wc->level. At this point path->nodes[wc->level] should be populated and
5897 * uptodate, and in most cases should already be locked. If we are in
5898 * DROP_REFERENCE and our refcount is > 1 then we've entered a shared node and
5899 * we can walk back up the tree. If we are UPDATE_BACKREF we have to set
5900 * FULL_BACKREF on this node if it's not already set, and then do the
5901 * FULL_BACKREF conversion dance, which is to drop the root reference and add
5902 * the shared reference to all of this nodes children.
5903 *
5904 * do_walk_down(). This is where we actually start iterating on the children of
5905 * our current path->nodes[wc->level]. For DROP_REFERENCE that means dropping
5906 * our reference to the children that return false from visit_node_for_delete(),
5907 * which has various conditions where we know we can just drop our reference
5908 * without visiting the node. For UPDATE_BACKREF we will skip any children that
5909 * visit_node_for_delete() returns false for, only walking down when necessary.
5910 * The bulk of the work for UPDATE_BACKREF occurs in the walk_up_tree() part of
5911 * snapshot deletion.
5912 */
5913static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5914 struct btrfs_root *root,
5915 struct btrfs_path *path,
5916 struct walk_control *wc)
5917{
5918 int level = wc->level;
5919 int ret = 0;
5920
5921 wc->lookup_info = 1;
5922 while (level >= 0) {
5923 ret = walk_down_proc(trans, root, path, wc);
5924 if (ret)
5925 break;
5926
5927 if (level == 0)
5928 break;
5929
5930 if (path->slots[level] >=
5931 btrfs_header_nritems(path->nodes[level]))
5932 break;
5933
5934 ret = do_walk_down(trans, root, path, wc);
5935 if (ret > 0) {
5936 path->slots[level]++;
5937 continue;
5938 } else if (ret < 0)
5939 break;
5940 level = wc->level;
5941 }
5942 return (ret == 1) ? 0 : ret;
5943}
5944
5945/*
5946 * walk_up_tree() is responsible for making sure we visit every slot on our
5947 * current node, and if we're at the end of that node then we call
5948 * walk_up_proc() on our current node which will do one of a few things based on
5949 * our stage.
5950 *
5951 * UPDATE_BACKREF. If we wc->level is currently less than our wc->shared_level
5952 * then we need to walk back up the tree, and then going back down into the
5953 * other slots via walk_down_tree to update any other children from our original
5954 * wc->shared_level. Once we're at or above our wc->shared_level we can switch
5955 * back to DROP_REFERENCE, lookup the current nodes refs and flags, and carry on.
5956 *
5957 * DROP_REFERENCE. If our refs == 1 then we're going to free this tree block.
5958 * If we're level 0 then we need to btrfs_dec_ref() on all of the data extents
5959 * in our current leaf. After that we call btrfs_free_tree_block() on the
5960 * current node and walk up to the next node to walk down the next slot.
5961 */
5962static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5963 struct btrfs_root *root,
5964 struct btrfs_path *path,
5965 struct walk_control *wc, int max_level)
5966{
5967 int level = wc->level;
5968 int ret;
5969
5970 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5971 while (level < max_level && path->nodes[level]) {
5972 wc->level = level;
5973 if (path->slots[level] + 1 <
5974 btrfs_header_nritems(path->nodes[level])) {
5975 path->slots[level]++;
5976 return 0;
5977 } else {
5978 ret = walk_up_proc(trans, root, path, wc);
5979 if (ret > 0)
5980 return 0;
5981 if (ret < 0)
5982 return ret;
5983
5984 if (path->locks[level]) {
5985 btrfs_tree_unlock_rw(path->nodes[level],
5986 path->locks[level]);
5987 path->locks[level] = 0;
5988 }
5989 free_extent_buffer(path->nodes[level]);
5990 path->nodes[level] = NULL;
5991 level++;
5992 }
5993 }
5994 return 1;
5995}
5996
5997/*
5998 * drop a subvolume tree.
5999 *
6000 * this function traverses the tree freeing any blocks that only
6001 * referenced by the tree.
6002 *
6003 * when a shared tree block is found. this function decreases its
6004 * reference count by one. if update_ref is true, this function
6005 * also make sure backrefs for the shared block and all lower level
6006 * blocks are properly updated.
6007 *
6008 * If called with for_reloc == 0, may exit early with -EAGAIN
6009 */
6010int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc)
6011{
6012 const bool is_reloc_root = (btrfs_root_id(root) == BTRFS_TREE_RELOC_OBJECTID);
6013 struct btrfs_fs_info *fs_info = root->fs_info;
6014 struct btrfs_path *path;
6015 struct btrfs_trans_handle *trans;
6016 struct btrfs_root *tree_root = fs_info->tree_root;
6017 struct btrfs_root_item *root_item = &root->root_item;
6018 struct walk_control *wc;
6019 struct btrfs_key key;
6020 const u64 rootid = btrfs_root_id(root);
6021 int ret = 0;
6022 int level;
6023 bool root_dropped = false;
6024 bool unfinished_drop = false;
6025
6026 btrfs_debug(fs_info, "Drop subvolume %llu", btrfs_root_id(root));
6027
6028 path = btrfs_alloc_path();
6029 if (!path) {
6030 ret = -ENOMEM;
6031 goto out;
6032 }
6033
6034 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6035 if (!wc) {
6036 btrfs_free_path(path);
6037 ret = -ENOMEM;
6038 goto out;
6039 }
6040
6041 /*
6042 * Use join to avoid potential EINTR from transaction start. See
6043 * wait_reserve_ticket and the whole reservation callchain.
6044 */
6045 if (for_reloc)
6046 trans = btrfs_join_transaction(tree_root);
6047 else
6048 trans = btrfs_start_transaction(tree_root, 0);
6049 if (IS_ERR(trans)) {
6050 ret = PTR_ERR(trans);
6051 goto out_free;
6052 }
6053
6054 ret = btrfs_run_delayed_items(trans);
6055 if (ret)
6056 goto out_end_trans;
6057
6058 /*
6059 * This will help us catch people modifying the fs tree while we're
6060 * dropping it. It is unsafe to mess with the fs tree while it's being
6061 * dropped as we unlock the root node and parent nodes as we walk down
6062 * the tree, assuming nothing will change. If something does change
6063 * then we'll have stale information and drop references to blocks we've
6064 * already dropped.
6065 */
6066 set_bit(BTRFS_ROOT_DELETING, &root->state);
6067 unfinished_drop = test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state);
6068
6069 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6070 level = btrfs_header_level(root->node);
6071 path->nodes[level] = btrfs_lock_root_node(root);
6072 path->slots[level] = 0;
6073 path->locks[level] = BTRFS_WRITE_LOCK;
6074 memset(&wc->update_progress, 0,
6075 sizeof(wc->update_progress));
6076 } else {
6077 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6078 memcpy(&wc->update_progress, &key,
6079 sizeof(wc->update_progress));
6080
6081 level = btrfs_root_drop_level(root_item);
6082 BUG_ON(level == 0);
6083 path->lowest_level = level;
6084 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6085 path->lowest_level = 0;
6086 if (ret < 0)
6087 goto out_end_trans;
6088
6089 WARN_ON(ret > 0);
6090 ret = 0;
6091
6092 /*
6093 * unlock our path, this is safe because only this
6094 * function is allowed to delete this snapshot
6095 */
6096 btrfs_unlock_up_safe(path, 0);
6097
6098 level = btrfs_header_level(root->node);
6099 while (1) {
6100 btrfs_tree_lock(path->nodes[level]);
6101 path->locks[level] = BTRFS_WRITE_LOCK;
6102
6103 /*
6104 * btrfs_lookup_extent_info() returns 0 for success,
6105 * or < 0 for error.
6106 */
6107 ret = btrfs_lookup_extent_info(trans, fs_info,
6108 path->nodes[level]->start,
6109 level, 1, &wc->refs[level],
6110 &wc->flags[level], NULL);
6111 if (ret < 0)
6112 goto out_end_trans;
6113
6114 BUG_ON(wc->refs[level] == 0);
6115
6116 if (level == btrfs_root_drop_level(root_item))
6117 break;
6118
6119 btrfs_tree_unlock(path->nodes[level]);
6120 path->locks[level] = 0;
6121 WARN_ON(wc->refs[level] != 1);
6122 level--;
6123 }
6124 }
6125
6126 wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
6127 wc->level = level;
6128 wc->shared_level = -1;
6129 wc->stage = DROP_REFERENCE;
6130 wc->update_ref = update_ref;
6131 wc->keep_locks = 0;
6132 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
6133
6134 while (1) {
6135
6136 ret = walk_down_tree(trans, root, path, wc);
6137 if (ret < 0) {
6138 btrfs_abort_transaction(trans, ret);
6139 break;
6140 }
6141
6142 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6143 if (ret < 0) {
6144 btrfs_abort_transaction(trans, ret);
6145 break;
6146 }
6147
6148 if (ret > 0) {
6149 BUG_ON(wc->stage != DROP_REFERENCE);
6150 ret = 0;
6151 break;
6152 }
6153
6154 if (wc->stage == DROP_REFERENCE) {
6155 wc->drop_level = wc->level;
6156 btrfs_node_key_to_cpu(path->nodes[wc->drop_level],
6157 &wc->drop_progress,
6158 path->slots[wc->drop_level]);
6159 }
6160 btrfs_cpu_key_to_disk(&root_item->drop_progress,
6161 &wc->drop_progress);
6162 btrfs_set_root_drop_level(root_item, wc->drop_level);
6163
6164 BUG_ON(wc->level == 0);
6165 if (btrfs_should_end_transaction(trans) ||
6166 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
6167 ret = btrfs_update_root(trans, tree_root,
6168 &root->root_key,
6169 root_item);
6170 if (ret) {
6171 btrfs_abort_transaction(trans, ret);
6172 goto out_end_trans;
6173 }
6174
6175 if (!is_reloc_root)
6176 btrfs_set_last_root_drop_gen(fs_info, trans->transid);
6177
6178 btrfs_end_transaction_throttle(trans);
6179 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
6180 btrfs_debug(fs_info,
6181 "drop snapshot early exit");
6182 ret = -EAGAIN;
6183 goto out_free;
6184 }
6185
6186 /*
6187 * Use join to avoid potential EINTR from transaction
6188 * start. See wait_reserve_ticket and the whole
6189 * reservation callchain.
6190 */
6191 if (for_reloc)
6192 trans = btrfs_join_transaction(tree_root);
6193 else
6194 trans = btrfs_start_transaction(tree_root, 0);
6195 if (IS_ERR(trans)) {
6196 ret = PTR_ERR(trans);
6197 goto out_free;
6198 }
6199 }
6200 }
6201 btrfs_release_path(path);
6202 if (ret)
6203 goto out_end_trans;
6204
6205 ret = btrfs_del_root(trans, &root->root_key);
6206 if (ret) {
6207 btrfs_abort_transaction(trans, ret);
6208 goto out_end_trans;
6209 }
6210
6211 if (!is_reloc_root) {
6212 ret = btrfs_find_root(tree_root, &root->root_key, path,
6213 NULL, NULL);
6214 if (ret < 0) {
6215 btrfs_abort_transaction(trans, ret);
6216 goto out_end_trans;
6217 } else if (ret > 0) {
6218 ret = 0;
6219 /*
6220 * If we fail to delete the orphan item this time
6221 * around, it'll get picked up the next time.
6222 *
6223 * The most common failure here is just -ENOENT.
6224 */
6225 btrfs_del_orphan_item(trans, tree_root, btrfs_root_id(root));
6226 }
6227 }
6228
6229 /*
6230 * This subvolume is going to be completely dropped, and won't be
6231 * recorded as dirty roots, thus pertrans meta rsv will not be freed at
6232 * commit transaction time. So free it here manually.
6233 */
6234 btrfs_qgroup_convert_reserved_meta(root, INT_MAX);
6235 btrfs_qgroup_free_meta_all_pertrans(root);
6236
6237 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state))
6238 btrfs_add_dropped_root(trans, root);
6239 else
6240 btrfs_put_root(root);
6241 root_dropped = true;
6242out_end_trans:
6243 if (!is_reloc_root)
6244 btrfs_set_last_root_drop_gen(fs_info, trans->transid);
6245
6246 btrfs_end_transaction_throttle(trans);
6247out_free:
6248 kfree(wc);
6249 btrfs_free_path(path);
6250out:
6251 if (!ret && root_dropped) {
6252 ret = btrfs_qgroup_cleanup_dropped_subvolume(fs_info, rootid);
6253 if (ret < 0)
6254 btrfs_warn_rl(fs_info,
6255 "failed to cleanup qgroup 0/%llu: %d",
6256 rootid, ret);
6257 ret = 0;
6258 }
6259 /*
6260 * We were an unfinished drop root, check to see if there are any
6261 * pending, and if not clear and wake up any waiters.
6262 */
6263 if (!ret && unfinished_drop)
6264 btrfs_maybe_wake_unfinished_drop(fs_info);
6265
6266 /*
6267 * So if we need to stop dropping the snapshot for whatever reason we
6268 * need to make sure to add it back to the dead root list so that we
6269 * keep trying to do the work later. This also cleans up roots if we
6270 * don't have it in the radix (like when we recover after a power fail
6271 * or unmount) so we don't leak memory.
6272 */
6273 if (!for_reloc && !root_dropped)
6274 btrfs_add_dead_root(root);
6275 return ret;
6276}
6277
6278/*
6279 * drop subtree rooted at tree block 'node'.
6280 *
6281 * NOTE: this function will unlock and release tree block 'node'
6282 * only used by relocation code
6283 */
6284int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6285 struct btrfs_root *root,
6286 struct extent_buffer *node,
6287 struct extent_buffer *parent)
6288{
6289 struct btrfs_fs_info *fs_info = root->fs_info;
6290 struct btrfs_path *path;
6291 struct walk_control *wc;
6292 int level;
6293 int parent_level;
6294 int ret = 0;
6295
6296 BUG_ON(btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID);
6297
6298 path = btrfs_alloc_path();
6299 if (!path)
6300 return -ENOMEM;
6301
6302 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6303 if (!wc) {
6304 btrfs_free_path(path);
6305 return -ENOMEM;
6306 }
6307
6308 btrfs_assert_tree_write_locked(parent);
6309 parent_level = btrfs_header_level(parent);
6310 atomic_inc(&parent->refs);
6311 path->nodes[parent_level] = parent;
6312 path->slots[parent_level] = btrfs_header_nritems(parent);
6313
6314 btrfs_assert_tree_write_locked(node);
6315 level = btrfs_header_level(node);
6316 path->nodes[level] = node;
6317 path->slots[level] = 0;
6318 path->locks[level] = BTRFS_WRITE_LOCK;
6319
6320 wc->refs[parent_level] = 1;
6321 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6322 wc->level = level;
6323 wc->shared_level = -1;
6324 wc->stage = DROP_REFERENCE;
6325 wc->update_ref = 0;
6326 wc->keep_locks = 1;
6327 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
6328
6329 while (1) {
6330 ret = walk_down_tree(trans, root, path, wc);
6331 if (ret < 0)
6332 break;
6333
6334 ret = walk_up_tree(trans, root, path, wc, parent_level);
6335 if (ret) {
6336 if (ret > 0)
6337 ret = 0;
6338 break;
6339 }
6340 }
6341
6342 kfree(wc);
6343 btrfs_free_path(path);
6344 return ret;
6345}
6346
6347/*
6348 * Unpin the extent range in an error context and don't add the space back.
6349 * Errors are not propagated further.
6350 */
6351void btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, u64 start, u64 end)
6352{
6353 unpin_extent_range(fs_info, start, end, false);
6354}
6355
6356/*
6357 * It used to be that old block groups would be left around forever.
6358 * Iterating over them would be enough to trim unused space. Since we
6359 * now automatically remove them, we also need to iterate over unallocated
6360 * space.
6361 *
6362 * We don't want a transaction for this since the discard may take a
6363 * substantial amount of time. We don't require that a transaction be
6364 * running, but we do need to take a running transaction into account
6365 * to ensure that we're not discarding chunks that were released or
6366 * allocated in the current transaction.
6367 *
6368 * Holding the chunks lock will prevent other threads from allocating
6369 * or releasing chunks, but it won't prevent a running transaction
6370 * from committing and releasing the memory that the pending chunks
6371 * list head uses. For that, we need to take a reference to the
6372 * transaction and hold the commit root sem. We only need to hold
6373 * it while performing the free space search since we have already
6374 * held back allocations.
6375 */
6376static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed)
6377{
6378 u64 start = BTRFS_DEVICE_RANGE_RESERVED, len = 0, end = 0;
6379 int ret;
6380
6381 *trimmed = 0;
6382
6383 /* Discard not supported = nothing to do. */
6384 if (!bdev_max_discard_sectors(device->bdev))
6385 return 0;
6386
6387 /* Not writable = nothing to do. */
6388 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
6389 return 0;
6390
6391 /* No free space = nothing to do. */
6392 if (device->total_bytes <= device->bytes_used)
6393 return 0;
6394
6395 ret = 0;
6396
6397 while (1) {
6398 struct btrfs_fs_info *fs_info = device->fs_info;
6399 u64 bytes;
6400
6401 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
6402 if (ret)
6403 break;
6404
6405 find_first_clear_extent_bit(&device->alloc_state, start,
6406 &start, &end,
6407 CHUNK_TRIMMED | CHUNK_ALLOCATED);
6408
6409 /* Check if there are any CHUNK_* bits left */
6410 if (start > device->total_bytes) {
6411 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
6412 btrfs_warn_in_rcu(fs_info,
6413"ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu",
6414 start, end - start + 1,
6415 btrfs_dev_name(device),
6416 device->total_bytes);
6417 mutex_unlock(&fs_info->chunk_mutex);
6418 ret = 0;
6419 break;
6420 }
6421
6422 /* Ensure we skip the reserved space on each device. */
6423 start = max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED);
6424
6425 /*
6426 * If find_first_clear_extent_bit find a range that spans the
6427 * end of the device it will set end to -1, in this case it's up
6428 * to the caller to trim the value to the size of the device.
6429 */
6430 end = min(end, device->total_bytes - 1);
6431
6432 len = end - start + 1;
6433
6434 /* We didn't find any extents */
6435 if (!len) {
6436 mutex_unlock(&fs_info->chunk_mutex);
6437 ret = 0;
6438 break;
6439 }
6440
6441 ret = btrfs_issue_discard(device->bdev, start, len,
6442 &bytes);
6443 if (!ret)
6444 set_extent_bit(&device->alloc_state, start,
6445 start + bytes - 1, CHUNK_TRIMMED, NULL);
6446 mutex_unlock(&fs_info->chunk_mutex);
6447
6448 if (ret)
6449 break;
6450
6451 start += len;
6452 *trimmed += bytes;
6453
6454 if (btrfs_trim_interrupted()) {
6455 ret = -ERESTARTSYS;
6456 break;
6457 }
6458
6459 cond_resched();
6460 }
6461
6462 return ret;
6463}
6464
6465/*
6466 * Trim the whole filesystem by:
6467 * 1) trimming the free space in each block group
6468 * 2) trimming the unallocated space on each device
6469 *
6470 * This will also continue trimming even if a block group or device encounters
6471 * an error. The return value will be the last error, or 0 if nothing bad
6472 * happens.
6473 */
6474int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
6475{
6476 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6477 struct btrfs_block_group *cache = NULL;
6478 struct btrfs_device *device;
6479 u64 group_trimmed;
6480 u64 range_end = U64_MAX;
6481 u64 start;
6482 u64 end;
6483 u64 trimmed = 0;
6484 u64 bg_failed = 0;
6485 u64 dev_failed = 0;
6486 int bg_ret = 0;
6487 int dev_ret = 0;
6488 int ret = 0;
6489
6490 if (range->start == U64_MAX)
6491 return -EINVAL;
6492
6493 /*
6494 * Check range overflow if range->len is set.
6495 * The default range->len is U64_MAX.
6496 */
6497 if (range->len != U64_MAX &&
6498 check_add_overflow(range->start, range->len, &range_end))
6499 return -EINVAL;
6500
6501 cache = btrfs_lookup_first_block_group(fs_info, range->start);
6502 for (; cache; cache = btrfs_next_block_group(cache)) {
6503 if (cache->start >= range_end) {
6504 btrfs_put_block_group(cache);
6505 break;
6506 }
6507
6508 start = max(range->start, cache->start);
6509 end = min(range_end, cache->start + cache->length);
6510
6511 if (end - start >= range->minlen) {
6512 if (!btrfs_block_group_done(cache)) {
6513 ret = btrfs_cache_block_group(cache, true);
6514 if (ret) {
6515 bg_failed++;
6516 bg_ret = ret;
6517 continue;
6518 }
6519 }
6520 ret = btrfs_trim_block_group(cache,
6521 &group_trimmed,
6522 start,
6523 end,
6524 range->minlen);
6525
6526 trimmed += group_trimmed;
6527 if (ret) {
6528 bg_failed++;
6529 bg_ret = ret;
6530 continue;
6531 }
6532 }
6533 }
6534
6535 if (bg_failed)
6536 btrfs_warn(fs_info,
6537 "failed to trim %llu block group(s), last error %d",
6538 bg_failed, bg_ret);
6539
6540 mutex_lock(&fs_devices->device_list_mutex);
6541 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6542 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
6543 continue;
6544
6545 ret = btrfs_trim_free_extents(device, &group_trimmed);
6546
6547 trimmed += group_trimmed;
6548 if (ret) {
6549 dev_failed++;
6550 dev_ret = ret;
6551 break;
6552 }
6553 }
6554 mutex_unlock(&fs_devices->device_list_mutex);
6555
6556 if (dev_failed)
6557 btrfs_warn(fs_info,
6558 "failed to trim %llu device(s), last error %d",
6559 dev_failed, dev_ret);
6560 range->len = trimmed;
6561 if (bg_ret)
6562 return bg_ret;
6563 return dev_ret;
6564}