Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6#include <linux/slab.h>
7#include <linux/blkdev.h>
8#include <linux/writeback.h>
9#include <linux/sched/mm.h>
10#include "messages.h"
11#include "misc.h"
12#include "ctree.h"
13#include "transaction.h"
14#include "btrfs_inode.h"
15#include "extent_io.h"
16#include "disk-io.h"
17#include "compression.h"
18#include "delalloc-space.h"
19#include "qgroup.h"
20#include "subpage.h"
21#include "file.h"
22#include "super.h"
23
24static struct kmem_cache *btrfs_ordered_extent_cache;
25
26static u64 entry_end(struct btrfs_ordered_extent *entry)
27{
28 if (entry->file_offset + entry->num_bytes < entry->file_offset)
29 return (u64)-1;
30 return entry->file_offset + entry->num_bytes;
31}
32
33/* returns NULL if the insertion worked, or it returns the node it did find
34 * in the tree
35 */
36static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
37 struct rb_node *node)
38{
39 struct rb_node **p = &root->rb_node;
40 struct rb_node *parent = NULL;
41 struct btrfs_ordered_extent *entry;
42
43 while (*p) {
44 parent = *p;
45 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
46
47 if (file_offset < entry->file_offset)
48 p = &(*p)->rb_left;
49 else if (file_offset >= entry_end(entry))
50 p = &(*p)->rb_right;
51 else
52 return parent;
53 }
54
55 rb_link_node(node, parent, p);
56 rb_insert_color(node, root);
57 return NULL;
58}
59
60/*
61 * look for a given offset in the tree, and if it can't be found return the
62 * first lesser offset
63 */
64static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
65 struct rb_node **prev_ret)
66{
67 struct rb_node *n = root->rb_node;
68 struct rb_node *prev = NULL;
69 struct rb_node *test;
70 struct btrfs_ordered_extent *entry;
71 struct btrfs_ordered_extent *prev_entry = NULL;
72
73 while (n) {
74 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
75 prev = n;
76 prev_entry = entry;
77
78 if (file_offset < entry->file_offset)
79 n = n->rb_left;
80 else if (file_offset >= entry_end(entry))
81 n = n->rb_right;
82 else
83 return n;
84 }
85 if (!prev_ret)
86 return NULL;
87
88 while (prev && file_offset >= entry_end(prev_entry)) {
89 test = rb_next(prev);
90 if (!test)
91 break;
92 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
93 rb_node);
94 if (file_offset < entry_end(prev_entry))
95 break;
96
97 prev = test;
98 }
99 if (prev)
100 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
101 rb_node);
102 while (prev && file_offset < entry_end(prev_entry)) {
103 test = rb_prev(prev);
104 if (!test)
105 break;
106 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
107 rb_node);
108 prev = test;
109 }
110 *prev_ret = prev;
111 return NULL;
112}
113
114static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
115 u64 len)
116{
117 if (file_offset + len <= entry->file_offset ||
118 entry->file_offset + entry->num_bytes <= file_offset)
119 return 0;
120 return 1;
121}
122
123/*
124 * look find the first ordered struct that has this offset, otherwise
125 * the first one less than this offset
126 */
127static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
128 u64 file_offset)
129{
130 struct rb_root *root = &tree->tree;
131 struct rb_node *prev = NULL;
132 struct rb_node *ret;
133 struct btrfs_ordered_extent *entry;
134
135 if (tree->last) {
136 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
137 rb_node);
138 if (in_range(file_offset, entry->file_offset, entry->num_bytes))
139 return tree->last;
140 }
141 ret = __tree_search(root, file_offset, &prev);
142 if (!ret)
143 ret = prev;
144 if (ret)
145 tree->last = ret;
146 return ret;
147}
148
149/*
150 * Add an ordered extent to the per-inode tree.
151 *
152 * @inode: Inode that this extent is for.
153 * @file_offset: Logical offset in file where the extent starts.
154 * @num_bytes: Logical length of extent in file.
155 * @ram_bytes: Full length of unencoded data.
156 * @disk_bytenr: Offset of extent on disk.
157 * @disk_num_bytes: Size of extent on disk.
158 * @offset: Offset into unencoded data where file data starts.
159 * @flags: Flags specifying type of extent (1 << BTRFS_ORDERED_*).
160 * @compress_type: Compression algorithm used for data.
161 *
162 * Most of these parameters correspond to &struct btrfs_file_extent_item. The
163 * tree is given a single reference on the ordered extent that was inserted.
164 *
165 * Return: 0 or -ENOMEM.
166 */
167int btrfs_add_ordered_extent(struct btrfs_inode *inode, u64 file_offset,
168 u64 num_bytes, u64 ram_bytes, u64 disk_bytenr,
169 u64 disk_num_bytes, u64 offset, unsigned flags,
170 int compress_type)
171{
172 struct btrfs_root *root = inode->root;
173 struct btrfs_fs_info *fs_info = root->fs_info;
174 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
175 struct rb_node *node;
176 struct btrfs_ordered_extent *entry;
177 int ret;
178
179 if (flags &
180 ((1 << BTRFS_ORDERED_NOCOW) | (1 << BTRFS_ORDERED_PREALLOC))) {
181 /* For nocow write, we can release the qgroup rsv right now */
182 ret = btrfs_qgroup_free_data(inode, NULL, file_offset, num_bytes);
183 if (ret < 0)
184 return ret;
185 ret = 0;
186 } else {
187 /*
188 * The ordered extent has reserved qgroup space, release now
189 * and pass the reserved number for qgroup_record to free.
190 */
191 ret = btrfs_qgroup_release_data(inode, file_offset, num_bytes);
192 if (ret < 0)
193 return ret;
194 }
195 entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
196 if (!entry)
197 return -ENOMEM;
198
199 entry->file_offset = file_offset;
200 entry->num_bytes = num_bytes;
201 entry->ram_bytes = ram_bytes;
202 entry->disk_bytenr = disk_bytenr;
203 entry->disk_num_bytes = disk_num_bytes;
204 entry->offset = offset;
205 entry->bytes_left = num_bytes;
206 entry->inode = igrab(&inode->vfs_inode);
207 entry->compress_type = compress_type;
208 entry->truncated_len = (u64)-1;
209 entry->qgroup_rsv = ret;
210 entry->physical = (u64)-1;
211
212 ASSERT((flags & ~BTRFS_ORDERED_TYPE_FLAGS) == 0);
213 entry->flags = flags;
214
215 percpu_counter_add_batch(&fs_info->ordered_bytes, num_bytes,
216 fs_info->delalloc_batch);
217
218 /* one ref for the tree */
219 refcount_set(&entry->refs, 1);
220 init_waitqueue_head(&entry->wait);
221 INIT_LIST_HEAD(&entry->list);
222 INIT_LIST_HEAD(&entry->log_list);
223 INIT_LIST_HEAD(&entry->root_extent_list);
224 INIT_LIST_HEAD(&entry->work_list);
225 init_completion(&entry->completion);
226
227 trace_btrfs_ordered_extent_add(inode, entry);
228
229 spin_lock_irq(&tree->lock);
230 node = tree_insert(&tree->tree, file_offset,
231 &entry->rb_node);
232 if (node)
233 btrfs_panic(fs_info, -EEXIST,
234 "inconsistency in ordered tree at offset %llu",
235 file_offset);
236 spin_unlock_irq(&tree->lock);
237
238 spin_lock(&root->ordered_extent_lock);
239 list_add_tail(&entry->root_extent_list,
240 &root->ordered_extents);
241 root->nr_ordered_extents++;
242 if (root->nr_ordered_extents == 1) {
243 spin_lock(&fs_info->ordered_root_lock);
244 BUG_ON(!list_empty(&root->ordered_root));
245 list_add_tail(&root->ordered_root, &fs_info->ordered_roots);
246 spin_unlock(&fs_info->ordered_root_lock);
247 }
248 spin_unlock(&root->ordered_extent_lock);
249
250 /*
251 * We don't need the count_max_extents here, we can assume that all of
252 * that work has been done at higher layers, so this is truly the
253 * smallest the extent is going to get.
254 */
255 spin_lock(&inode->lock);
256 btrfs_mod_outstanding_extents(inode, 1);
257 spin_unlock(&inode->lock);
258
259 return 0;
260}
261
262/*
263 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
264 * when an ordered extent is finished. If the list covers more than one
265 * ordered extent, it is split across multiples.
266 */
267void btrfs_add_ordered_sum(struct btrfs_ordered_extent *entry,
268 struct btrfs_ordered_sum *sum)
269{
270 struct btrfs_ordered_inode_tree *tree;
271
272 tree = &BTRFS_I(entry->inode)->ordered_tree;
273 spin_lock_irq(&tree->lock);
274 list_add_tail(&sum->list, &entry->list);
275 spin_unlock_irq(&tree->lock);
276}
277
278static void finish_ordered_fn(struct btrfs_work *work)
279{
280 struct btrfs_ordered_extent *ordered_extent;
281
282 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
283 btrfs_finish_ordered_io(ordered_extent);
284}
285
286/*
287 * Mark all ordered extents io inside the specified range finished.
288 *
289 * @page: The involved page for the operation.
290 * For uncompressed buffered IO, the page status also needs to be
291 * updated to indicate whether the pending ordered io is finished.
292 * Can be NULL for direct IO and compressed write.
293 * For these cases, callers are ensured they won't execute the
294 * endio function twice.
295 *
296 * This function is called for endio, thus the range must have ordered
297 * extent(s) covering it.
298 */
299void btrfs_mark_ordered_io_finished(struct btrfs_inode *inode,
300 struct page *page, u64 file_offset,
301 u64 num_bytes, bool uptodate)
302{
303 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
304 struct btrfs_fs_info *fs_info = inode->root->fs_info;
305 struct btrfs_workqueue *wq;
306 struct rb_node *node;
307 struct btrfs_ordered_extent *entry = NULL;
308 unsigned long flags;
309 u64 cur = file_offset;
310
311 if (btrfs_is_free_space_inode(inode))
312 wq = fs_info->endio_freespace_worker;
313 else
314 wq = fs_info->endio_write_workers;
315
316 if (page)
317 ASSERT(page->mapping && page_offset(page) <= file_offset &&
318 file_offset + num_bytes <= page_offset(page) + PAGE_SIZE);
319
320 spin_lock_irqsave(&tree->lock, flags);
321 while (cur < file_offset + num_bytes) {
322 u64 entry_end;
323 u64 end;
324 u32 len;
325
326 node = tree_search(tree, cur);
327 /* No ordered extents at all */
328 if (!node)
329 break;
330
331 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
332 entry_end = entry->file_offset + entry->num_bytes;
333 /*
334 * |<-- OE --->| |
335 * cur
336 * Go to next OE.
337 */
338 if (cur >= entry_end) {
339 node = rb_next(node);
340 /* No more ordered extents, exit */
341 if (!node)
342 break;
343 entry = rb_entry(node, struct btrfs_ordered_extent,
344 rb_node);
345
346 /* Go to next ordered extent and continue */
347 cur = entry->file_offset;
348 continue;
349 }
350 /*
351 * | |<--- OE --->|
352 * cur
353 * Go to the start of OE.
354 */
355 if (cur < entry->file_offset) {
356 cur = entry->file_offset;
357 continue;
358 }
359
360 /*
361 * Now we are definitely inside one ordered extent.
362 *
363 * |<--- OE --->|
364 * |
365 * cur
366 */
367 end = min(entry->file_offset + entry->num_bytes,
368 file_offset + num_bytes) - 1;
369 ASSERT(end + 1 - cur < U32_MAX);
370 len = end + 1 - cur;
371
372 if (page) {
373 /*
374 * Ordered (Private2) bit indicates whether we still
375 * have pending io unfinished for the ordered extent.
376 *
377 * If there's no such bit, we need to skip to next range.
378 */
379 if (!btrfs_page_test_ordered(fs_info, page, cur, len)) {
380 cur += len;
381 continue;
382 }
383 btrfs_page_clear_ordered(fs_info, page, cur, len);
384 }
385
386 /* Now we're fine to update the accounting */
387 if (unlikely(len > entry->bytes_left)) {
388 WARN_ON(1);
389 btrfs_crit(fs_info,
390"bad ordered extent accounting, root=%llu ino=%llu OE offset=%llu OE len=%llu to_dec=%u left=%llu",
391 inode->root->root_key.objectid,
392 btrfs_ino(inode),
393 entry->file_offset,
394 entry->num_bytes,
395 len, entry->bytes_left);
396 entry->bytes_left = 0;
397 } else {
398 entry->bytes_left -= len;
399 }
400
401 if (!uptodate)
402 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
403
404 /*
405 * All the IO of the ordered extent is finished, we need to queue
406 * the finish_func to be executed.
407 */
408 if (entry->bytes_left == 0) {
409 set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
410 cond_wake_up(&entry->wait);
411 refcount_inc(&entry->refs);
412 trace_btrfs_ordered_extent_mark_finished(inode, entry);
413 spin_unlock_irqrestore(&tree->lock, flags);
414 btrfs_init_work(&entry->work, finish_ordered_fn, NULL, NULL);
415 btrfs_queue_work(wq, &entry->work);
416 spin_lock_irqsave(&tree->lock, flags);
417 }
418 cur += len;
419 }
420 spin_unlock_irqrestore(&tree->lock, flags);
421}
422
423/*
424 * Finish IO for one ordered extent across a given range. The range can only
425 * contain one ordered extent.
426 *
427 * @cached: The cached ordered extent. If not NULL, we can skip the tree
428 * search and use the ordered extent directly.
429 * Will be also used to store the finished ordered extent.
430 * @file_offset: File offset for the finished IO
431 * @io_size: Length of the finish IO range
432 *
433 * Return true if the ordered extent is finished in the range, and update
434 * @cached.
435 * Return false otherwise.
436 *
437 * NOTE: The range can NOT cross multiple ordered extents.
438 * Thus caller should ensure the range doesn't cross ordered extents.
439 */
440bool btrfs_dec_test_ordered_pending(struct btrfs_inode *inode,
441 struct btrfs_ordered_extent **cached,
442 u64 file_offset, u64 io_size)
443{
444 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
445 struct rb_node *node;
446 struct btrfs_ordered_extent *entry = NULL;
447 unsigned long flags;
448 bool finished = false;
449
450 spin_lock_irqsave(&tree->lock, flags);
451 if (cached && *cached) {
452 entry = *cached;
453 goto have_entry;
454 }
455
456 node = tree_search(tree, file_offset);
457 if (!node)
458 goto out;
459
460 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
461have_entry:
462 if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
463 goto out;
464
465 if (io_size > entry->bytes_left)
466 btrfs_crit(inode->root->fs_info,
467 "bad ordered accounting left %llu size %llu",
468 entry->bytes_left, io_size);
469
470 entry->bytes_left -= io_size;
471
472 if (entry->bytes_left == 0) {
473 /*
474 * Ensure only one caller can set the flag and finished_ret
475 * accordingly
476 */
477 finished = !test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
478 /* test_and_set_bit implies a barrier */
479 cond_wake_up_nomb(&entry->wait);
480 }
481out:
482 if (finished && cached && entry) {
483 *cached = entry;
484 refcount_inc(&entry->refs);
485 trace_btrfs_ordered_extent_dec_test_pending(inode, entry);
486 }
487 spin_unlock_irqrestore(&tree->lock, flags);
488 return finished;
489}
490
491/*
492 * used to drop a reference on an ordered extent. This will free
493 * the extent if the last reference is dropped
494 */
495void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
496{
497 struct list_head *cur;
498 struct btrfs_ordered_sum *sum;
499
500 trace_btrfs_ordered_extent_put(BTRFS_I(entry->inode), entry);
501
502 if (refcount_dec_and_test(&entry->refs)) {
503 ASSERT(list_empty(&entry->root_extent_list));
504 ASSERT(list_empty(&entry->log_list));
505 ASSERT(RB_EMPTY_NODE(&entry->rb_node));
506 if (entry->inode)
507 btrfs_add_delayed_iput(BTRFS_I(entry->inode));
508 while (!list_empty(&entry->list)) {
509 cur = entry->list.next;
510 sum = list_entry(cur, struct btrfs_ordered_sum, list);
511 list_del(&sum->list);
512 kvfree(sum);
513 }
514 kmem_cache_free(btrfs_ordered_extent_cache, entry);
515 }
516}
517
518/*
519 * remove an ordered extent from the tree. No references are dropped
520 * and waiters are woken up.
521 */
522void btrfs_remove_ordered_extent(struct btrfs_inode *btrfs_inode,
523 struct btrfs_ordered_extent *entry)
524{
525 struct btrfs_ordered_inode_tree *tree;
526 struct btrfs_root *root = btrfs_inode->root;
527 struct btrfs_fs_info *fs_info = root->fs_info;
528 struct rb_node *node;
529 bool pending;
530 bool freespace_inode;
531
532 /*
533 * If this is a free space inode the thread has not acquired the ordered
534 * extents lockdep map.
535 */
536 freespace_inode = btrfs_is_free_space_inode(btrfs_inode);
537
538 btrfs_lockdep_acquire(fs_info, btrfs_trans_pending_ordered);
539 /* This is paired with btrfs_add_ordered_extent. */
540 spin_lock(&btrfs_inode->lock);
541 btrfs_mod_outstanding_extents(btrfs_inode, -1);
542 spin_unlock(&btrfs_inode->lock);
543 if (root != fs_info->tree_root) {
544 u64 release;
545
546 if (test_bit(BTRFS_ORDERED_ENCODED, &entry->flags))
547 release = entry->disk_num_bytes;
548 else
549 release = entry->num_bytes;
550 btrfs_delalloc_release_metadata(btrfs_inode, release, false);
551 }
552
553 percpu_counter_add_batch(&fs_info->ordered_bytes, -entry->num_bytes,
554 fs_info->delalloc_batch);
555
556 tree = &btrfs_inode->ordered_tree;
557 spin_lock_irq(&tree->lock);
558 node = &entry->rb_node;
559 rb_erase(node, &tree->tree);
560 RB_CLEAR_NODE(node);
561 if (tree->last == node)
562 tree->last = NULL;
563 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
564 pending = test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags);
565 spin_unlock_irq(&tree->lock);
566
567 /*
568 * The current running transaction is waiting on us, we need to let it
569 * know that we're complete and wake it up.
570 */
571 if (pending) {
572 struct btrfs_transaction *trans;
573
574 /*
575 * The checks for trans are just a formality, it should be set,
576 * but if it isn't we don't want to deref/assert under the spin
577 * lock, so be nice and check if trans is set, but ASSERT() so
578 * if it isn't set a developer will notice.
579 */
580 spin_lock(&fs_info->trans_lock);
581 trans = fs_info->running_transaction;
582 if (trans)
583 refcount_inc(&trans->use_count);
584 spin_unlock(&fs_info->trans_lock);
585
586 ASSERT(trans);
587 if (trans) {
588 if (atomic_dec_and_test(&trans->pending_ordered))
589 wake_up(&trans->pending_wait);
590 btrfs_put_transaction(trans);
591 }
592 }
593
594 btrfs_lockdep_release(fs_info, btrfs_trans_pending_ordered);
595
596 spin_lock(&root->ordered_extent_lock);
597 list_del_init(&entry->root_extent_list);
598 root->nr_ordered_extents--;
599
600 trace_btrfs_ordered_extent_remove(btrfs_inode, entry);
601
602 if (!root->nr_ordered_extents) {
603 spin_lock(&fs_info->ordered_root_lock);
604 BUG_ON(list_empty(&root->ordered_root));
605 list_del_init(&root->ordered_root);
606 spin_unlock(&fs_info->ordered_root_lock);
607 }
608 spin_unlock(&root->ordered_extent_lock);
609 wake_up(&entry->wait);
610 if (!freespace_inode)
611 btrfs_lockdep_release(fs_info, btrfs_ordered_extent);
612}
613
614static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
615{
616 struct btrfs_ordered_extent *ordered;
617
618 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
619 btrfs_start_ordered_extent(ordered, 1);
620 complete(&ordered->completion);
621}
622
623/*
624 * wait for all the ordered extents in a root. This is done when balancing
625 * space between drives.
626 */
627u64 btrfs_wait_ordered_extents(struct btrfs_root *root, u64 nr,
628 const u64 range_start, const u64 range_len)
629{
630 struct btrfs_fs_info *fs_info = root->fs_info;
631 LIST_HEAD(splice);
632 LIST_HEAD(skipped);
633 LIST_HEAD(works);
634 struct btrfs_ordered_extent *ordered, *next;
635 u64 count = 0;
636 const u64 range_end = range_start + range_len;
637
638 mutex_lock(&root->ordered_extent_mutex);
639 spin_lock(&root->ordered_extent_lock);
640 list_splice_init(&root->ordered_extents, &splice);
641 while (!list_empty(&splice) && nr) {
642 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
643 root_extent_list);
644
645 if (range_end <= ordered->disk_bytenr ||
646 ordered->disk_bytenr + ordered->disk_num_bytes <= range_start) {
647 list_move_tail(&ordered->root_extent_list, &skipped);
648 cond_resched_lock(&root->ordered_extent_lock);
649 continue;
650 }
651
652 list_move_tail(&ordered->root_extent_list,
653 &root->ordered_extents);
654 refcount_inc(&ordered->refs);
655 spin_unlock(&root->ordered_extent_lock);
656
657 btrfs_init_work(&ordered->flush_work,
658 btrfs_run_ordered_extent_work, NULL, NULL);
659 list_add_tail(&ordered->work_list, &works);
660 btrfs_queue_work(fs_info->flush_workers, &ordered->flush_work);
661
662 cond_resched();
663 spin_lock(&root->ordered_extent_lock);
664 if (nr != U64_MAX)
665 nr--;
666 count++;
667 }
668 list_splice_tail(&skipped, &root->ordered_extents);
669 list_splice_tail(&splice, &root->ordered_extents);
670 spin_unlock(&root->ordered_extent_lock);
671
672 list_for_each_entry_safe(ordered, next, &works, work_list) {
673 list_del_init(&ordered->work_list);
674 wait_for_completion(&ordered->completion);
675 btrfs_put_ordered_extent(ordered);
676 cond_resched();
677 }
678 mutex_unlock(&root->ordered_extent_mutex);
679
680 return count;
681}
682
683void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, u64 nr,
684 const u64 range_start, const u64 range_len)
685{
686 struct btrfs_root *root;
687 struct list_head splice;
688 u64 done;
689
690 INIT_LIST_HEAD(&splice);
691
692 mutex_lock(&fs_info->ordered_operations_mutex);
693 spin_lock(&fs_info->ordered_root_lock);
694 list_splice_init(&fs_info->ordered_roots, &splice);
695 while (!list_empty(&splice) && nr) {
696 root = list_first_entry(&splice, struct btrfs_root,
697 ordered_root);
698 root = btrfs_grab_root(root);
699 BUG_ON(!root);
700 list_move_tail(&root->ordered_root,
701 &fs_info->ordered_roots);
702 spin_unlock(&fs_info->ordered_root_lock);
703
704 done = btrfs_wait_ordered_extents(root, nr,
705 range_start, range_len);
706 btrfs_put_root(root);
707
708 spin_lock(&fs_info->ordered_root_lock);
709 if (nr != U64_MAX) {
710 nr -= done;
711 }
712 }
713 list_splice_tail(&splice, &fs_info->ordered_roots);
714 spin_unlock(&fs_info->ordered_root_lock);
715 mutex_unlock(&fs_info->ordered_operations_mutex);
716}
717
718/*
719 * Used to start IO or wait for a given ordered extent to finish.
720 *
721 * If wait is one, this effectively waits on page writeback for all the pages
722 * in the extent, and it waits on the io completion code to insert
723 * metadata into the btree corresponding to the extent
724 */
725void btrfs_start_ordered_extent(struct btrfs_ordered_extent *entry, int wait)
726{
727 u64 start = entry->file_offset;
728 u64 end = start + entry->num_bytes - 1;
729 struct btrfs_inode *inode = BTRFS_I(entry->inode);
730 bool freespace_inode;
731
732 trace_btrfs_ordered_extent_start(inode, entry);
733
734 /*
735 * If this is a free space inode do not take the ordered extents lockdep
736 * map.
737 */
738 freespace_inode = btrfs_is_free_space_inode(inode);
739
740 /*
741 * pages in the range can be dirty, clean or writeback. We
742 * start IO on any dirty ones so the wait doesn't stall waiting
743 * for the flusher thread to find them
744 */
745 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
746 filemap_fdatawrite_range(inode->vfs_inode.i_mapping, start, end);
747 if (wait) {
748 if (!freespace_inode)
749 btrfs_might_wait_for_event(inode->root->fs_info, btrfs_ordered_extent);
750 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
751 &entry->flags));
752 }
753}
754
755/*
756 * Used to wait on ordered extents across a large range of bytes.
757 */
758int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
759{
760 int ret = 0;
761 int ret_wb = 0;
762 u64 end;
763 u64 orig_end;
764 struct btrfs_ordered_extent *ordered;
765
766 if (start + len < start) {
767 orig_end = OFFSET_MAX;
768 } else {
769 orig_end = start + len - 1;
770 if (orig_end > OFFSET_MAX)
771 orig_end = OFFSET_MAX;
772 }
773
774 /* start IO across the range first to instantiate any delalloc
775 * extents
776 */
777 ret = btrfs_fdatawrite_range(inode, start, orig_end);
778 if (ret)
779 return ret;
780
781 /*
782 * If we have a writeback error don't return immediately. Wait first
783 * for any ordered extents that haven't completed yet. This is to make
784 * sure no one can dirty the same page ranges and call writepages()
785 * before the ordered extents complete - to avoid failures (-EEXIST)
786 * when adding the new ordered extents to the ordered tree.
787 */
788 ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
789
790 end = orig_end;
791 while (1) {
792 ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode), end);
793 if (!ordered)
794 break;
795 if (ordered->file_offset > orig_end) {
796 btrfs_put_ordered_extent(ordered);
797 break;
798 }
799 if (ordered->file_offset + ordered->num_bytes <= start) {
800 btrfs_put_ordered_extent(ordered);
801 break;
802 }
803 btrfs_start_ordered_extent(ordered, 1);
804 end = ordered->file_offset;
805 /*
806 * If the ordered extent had an error save the error but don't
807 * exit without waiting first for all other ordered extents in
808 * the range to complete.
809 */
810 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
811 ret = -EIO;
812 btrfs_put_ordered_extent(ordered);
813 if (end == 0 || end == start)
814 break;
815 end--;
816 }
817 return ret_wb ? ret_wb : ret;
818}
819
820/*
821 * find an ordered extent corresponding to file_offset. return NULL if
822 * nothing is found, otherwise take a reference on the extent and return it
823 */
824struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct btrfs_inode *inode,
825 u64 file_offset)
826{
827 struct btrfs_ordered_inode_tree *tree;
828 struct rb_node *node;
829 struct btrfs_ordered_extent *entry = NULL;
830 unsigned long flags;
831
832 tree = &inode->ordered_tree;
833 spin_lock_irqsave(&tree->lock, flags);
834 node = tree_search(tree, file_offset);
835 if (!node)
836 goto out;
837
838 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
839 if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
840 entry = NULL;
841 if (entry) {
842 refcount_inc(&entry->refs);
843 trace_btrfs_ordered_extent_lookup(inode, entry);
844 }
845out:
846 spin_unlock_irqrestore(&tree->lock, flags);
847 return entry;
848}
849
850/* Since the DIO code tries to lock a wide area we need to look for any ordered
851 * extents that exist in the range, rather than just the start of the range.
852 */
853struct btrfs_ordered_extent *btrfs_lookup_ordered_range(
854 struct btrfs_inode *inode, u64 file_offset, u64 len)
855{
856 struct btrfs_ordered_inode_tree *tree;
857 struct rb_node *node;
858 struct btrfs_ordered_extent *entry = NULL;
859
860 tree = &inode->ordered_tree;
861 spin_lock_irq(&tree->lock);
862 node = tree_search(tree, file_offset);
863 if (!node) {
864 node = tree_search(tree, file_offset + len);
865 if (!node)
866 goto out;
867 }
868
869 while (1) {
870 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
871 if (range_overlaps(entry, file_offset, len))
872 break;
873
874 if (entry->file_offset >= file_offset + len) {
875 entry = NULL;
876 break;
877 }
878 entry = NULL;
879 node = rb_next(node);
880 if (!node)
881 break;
882 }
883out:
884 if (entry) {
885 refcount_inc(&entry->refs);
886 trace_btrfs_ordered_extent_lookup_range(inode, entry);
887 }
888 spin_unlock_irq(&tree->lock);
889 return entry;
890}
891
892/*
893 * Adds all ordered extents to the given list. The list ends up sorted by the
894 * file_offset of the ordered extents.
895 */
896void btrfs_get_ordered_extents_for_logging(struct btrfs_inode *inode,
897 struct list_head *list)
898{
899 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
900 struct rb_node *n;
901
902 ASSERT(inode_is_locked(&inode->vfs_inode));
903
904 spin_lock_irq(&tree->lock);
905 for (n = rb_first(&tree->tree); n; n = rb_next(n)) {
906 struct btrfs_ordered_extent *ordered;
907
908 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
909
910 if (test_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
911 continue;
912
913 ASSERT(list_empty(&ordered->log_list));
914 list_add_tail(&ordered->log_list, list);
915 refcount_inc(&ordered->refs);
916 trace_btrfs_ordered_extent_lookup_for_logging(inode, ordered);
917 }
918 spin_unlock_irq(&tree->lock);
919}
920
921/*
922 * lookup and return any extent before 'file_offset'. NULL is returned
923 * if none is found
924 */
925struct btrfs_ordered_extent *
926btrfs_lookup_first_ordered_extent(struct btrfs_inode *inode, u64 file_offset)
927{
928 struct btrfs_ordered_inode_tree *tree;
929 struct rb_node *node;
930 struct btrfs_ordered_extent *entry = NULL;
931
932 tree = &inode->ordered_tree;
933 spin_lock_irq(&tree->lock);
934 node = tree_search(tree, file_offset);
935 if (!node)
936 goto out;
937
938 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
939 refcount_inc(&entry->refs);
940 trace_btrfs_ordered_extent_lookup_first(inode, entry);
941out:
942 spin_unlock_irq(&tree->lock);
943 return entry;
944}
945
946/*
947 * Lookup the first ordered extent that overlaps the range
948 * [@file_offset, @file_offset + @len).
949 *
950 * The difference between this and btrfs_lookup_first_ordered_extent() is
951 * that this one won't return any ordered extent that does not overlap the range.
952 * And the difference against btrfs_lookup_ordered_extent() is, this function
953 * ensures the first ordered extent gets returned.
954 */
955struct btrfs_ordered_extent *btrfs_lookup_first_ordered_range(
956 struct btrfs_inode *inode, u64 file_offset, u64 len)
957{
958 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
959 struct rb_node *node;
960 struct rb_node *cur;
961 struct rb_node *prev;
962 struct rb_node *next;
963 struct btrfs_ordered_extent *entry = NULL;
964
965 spin_lock_irq(&tree->lock);
966 node = tree->tree.rb_node;
967 /*
968 * Here we don't want to use tree_search() which will use tree->last
969 * and screw up the search order.
970 * And __tree_search() can't return the adjacent ordered extents
971 * either, thus here we do our own search.
972 */
973 while (node) {
974 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
975
976 if (file_offset < entry->file_offset) {
977 node = node->rb_left;
978 } else if (file_offset >= entry_end(entry)) {
979 node = node->rb_right;
980 } else {
981 /*
982 * Direct hit, got an ordered extent that starts at
983 * @file_offset
984 */
985 goto out;
986 }
987 }
988 if (!entry) {
989 /* Empty tree */
990 goto out;
991 }
992
993 cur = &entry->rb_node;
994 /* We got an entry around @file_offset, check adjacent entries */
995 if (entry->file_offset < file_offset) {
996 prev = cur;
997 next = rb_next(cur);
998 } else {
999 prev = rb_prev(cur);
1000 next = cur;
1001 }
1002 if (prev) {
1003 entry = rb_entry(prev, struct btrfs_ordered_extent, rb_node);
1004 if (range_overlaps(entry, file_offset, len))
1005 goto out;
1006 }
1007 if (next) {
1008 entry = rb_entry(next, struct btrfs_ordered_extent, rb_node);
1009 if (range_overlaps(entry, file_offset, len))
1010 goto out;
1011 }
1012 /* No ordered extent in the range */
1013 entry = NULL;
1014out:
1015 if (entry) {
1016 refcount_inc(&entry->refs);
1017 trace_btrfs_ordered_extent_lookup_first_range(inode, entry);
1018 }
1019
1020 spin_unlock_irq(&tree->lock);
1021 return entry;
1022}
1023
1024/*
1025 * Lock the passed range and ensures all pending ordered extents in it are run
1026 * to completion.
1027 *
1028 * @inode: Inode whose ordered tree is to be searched
1029 * @start: Beginning of range to flush
1030 * @end: Last byte of range to lock
1031 * @cached_state: If passed, will return the extent state responsible for the
1032 * locked range. It's the caller's responsibility to free the
1033 * cached state.
1034 *
1035 * Always return with the given range locked, ensuring after it's called no
1036 * order extent can be pending.
1037 */
1038void btrfs_lock_and_flush_ordered_range(struct btrfs_inode *inode, u64 start,
1039 u64 end,
1040 struct extent_state **cached_state)
1041{
1042 struct btrfs_ordered_extent *ordered;
1043 struct extent_state *cache = NULL;
1044 struct extent_state **cachedp = &cache;
1045
1046 if (cached_state)
1047 cachedp = cached_state;
1048
1049 while (1) {
1050 lock_extent(&inode->io_tree, start, end, cachedp);
1051 ordered = btrfs_lookup_ordered_range(inode, start,
1052 end - start + 1);
1053 if (!ordered) {
1054 /*
1055 * If no external cached_state has been passed then
1056 * decrement the extra ref taken for cachedp since we
1057 * aren't exposing it outside of this function
1058 */
1059 if (!cached_state)
1060 refcount_dec(&cache->refs);
1061 break;
1062 }
1063 unlock_extent(&inode->io_tree, start, end, cachedp);
1064 btrfs_start_ordered_extent(ordered, 1);
1065 btrfs_put_ordered_extent(ordered);
1066 }
1067}
1068
1069/*
1070 * Lock the passed range and ensure all pending ordered extents in it are run
1071 * to completion in nowait mode.
1072 *
1073 * Return true if btrfs_lock_ordered_range does not return any extents,
1074 * otherwise false.
1075 */
1076bool btrfs_try_lock_ordered_range(struct btrfs_inode *inode, u64 start, u64 end,
1077 struct extent_state **cached_state)
1078{
1079 struct btrfs_ordered_extent *ordered;
1080
1081 if (!try_lock_extent(&inode->io_tree, start, end, cached_state))
1082 return false;
1083
1084 ordered = btrfs_lookup_ordered_range(inode, start, end - start + 1);
1085 if (!ordered)
1086 return true;
1087
1088 btrfs_put_ordered_extent(ordered);
1089 unlock_extent(&inode->io_tree, start, end, cached_state);
1090
1091 return false;
1092}
1093
1094
1095static int clone_ordered_extent(struct btrfs_ordered_extent *ordered, u64 pos,
1096 u64 len)
1097{
1098 struct inode *inode = ordered->inode;
1099 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1100 u64 file_offset = ordered->file_offset + pos;
1101 u64 disk_bytenr = ordered->disk_bytenr + pos;
1102 unsigned long flags = ordered->flags & BTRFS_ORDERED_TYPE_FLAGS;
1103
1104 /*
1105 * The splitting extent is already counted and will be added again in
1106 * btrfs_add_ordered_extent_*(). Subtract len to avoid double counting.
1107 */
1108 percpu_counter_add_batch(&fs_info->ordered_bytes, -len,
1109 fs_info->delalloc_batch);
1110 WARN_ON_ONCE(flags & (1 << BTRFS_ORDERED_COMPRESSED));
1111 return btrfs_add_ordered_extent(BTRFS_I(inode), file_offset, len, len,
1112 disk_bytenr, len, 0, flags,
1113 ordered->compress_type);
1114}
1115
1116int btrfs_split_ordered_extent(struct btrfs_ordered_extent *ordered, u64 pre,
1117 u64 post)
1118{
1119 struct inode *inode = ordered->inode;
1120 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
1121 struct rb_node *node;
1122 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1123 int ret = 0;
1124
1125 trace_btrfs_ordered_extent_split(BTRFS_I(inode), ordered);
1126
1127 spin_lock_irq(&tree->lock);
1128 /* Remove from tree once */
1129 node = &ordered->rb_node;
1130 rb_erase(node, &tree->tree);
1131 RB_CLEAR_NODE(node);
1132 if (tree->last == node)
1133 tree->last = NULL;
1134
1135 ordered->file_offset += pre;
1136 ordered->disk_bytenr += pre;
1137 ordered->num_bytes -= (pre + post);
1138 ordered->disk_num_bytes -= (pre + post);
1139 ordered->bytes_left -= (pre + post);
1140
1141 /* Re-insert the node */
1142 node = tree_insert(&tree->tree, ordered->file_offset, &ordered->rb_node);
1143 if (node)
1144 btrfs_panic(fs_info, -EEXIST,
1145 "zoned: inconsistency in ordered tree at offset %llu",
1146 ordered->file_offset);
1147
1148 spin_unlock_irq(&tree->lock);
1149
1150 if (pre)
1151 ret = clone_ordered_extent(ordered, 0, pre);
1152 if (ret == 0 && post)
1153 ret = clone_ordered_extent(ordered, pre + ordered->disk_num_bytes,
1154 post);
1155
1156 return ret;
1157}
1158
1159int __init ordered_data_init(void)
1160{
1161 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1162 sizeof(struct btrfs_ordered_extent), 0,
1163 SLAB_MEM_SPREAD,
1164 NULL);
1165 if (!btrfs_ordered_extent_cache)
1166 return -ENOMEM;
1167
1168 return 0;
1169}
1170
1171void __cold ordered_data_exit(void)
1172{
1173 kmem_cache_destroy(btrfs_ordered_extent_cache);
1174}
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
19#include <linux/slab.h>
20#include <linux/blkdev.h>
21#include <linux/writeback.h>
22#include <linux/pagevec.h>
23#include "ctree.h"
24#include "transaction.h"
25#include "btrfs_inode.h"
26#include "extent_io.h"
27#include "disk-io.h"
28
29static struct kmem_cache *btrfs_ordered_extent_cache;
30
31static u64 entry_end(struct btrfs_ordered_extent *entry)
32{
33 if (entry->file_offset + entry->len < entry->file_offset)
34 return (u64)-1;
35 return entry->file_offset + entry->len;
36}
37
38/* returns NULL if the insertion worked, or it returns the node it did find
39 * in the tree
40 */
41static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
42 struct rb_node *node)
43{
44 struct rb_node **p = &root->rb_node;
45 struct rb_node *parent = NULL;
46 struct btrfs_ordered_extent *entry;
47
48 while (*p) {
49 parent = *p;
50 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
51
52 if (file_offset < entry->file_offset)
53 p = &(*p)->rb_left;
54 else if (file_offset >= entry_end(entry))
55 p = &(*p)->rb_right;
56 else
57 return parent;
58 }
59
60 rb_link_node(node, parent, p);
61 rb_insert_color(node, root);
62 return NULL;
63}
64
65static void ordered_data_tree_panic(struct inode *inode, int errno,
66 u64 offset)
67{
68 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
69 btrfs_panic(fs_info, errno, "Inconsistency in ordered tree at offset "
70 "%llu\n", offset);
71}
72
73/*
74 * look for a given offset in the tree, and if it can't be found return the
75 * first lesser offset
76 */
77static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
78 struct rb_node **prev_ret)
79{
80 struct rb_node *n = root->rb_node;
81 struct rb_node *prev = NULL;
82 struct rb_node *test;
83 struct btrfs_ordered_extent *entry;
84 struct btrfs_ordered_extent *prev_entry = NULL;
85
86 while (n) {
87 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
88 prev = n;
89 prev_entry = entry;
90
91 if (file_offset < entry->file_offset)
92 n = n->rb_left;
93 else if (file_offset >= entry_end(entry))
94 n = n->rb_right;
95 else
96 return n;
97 }
98 if (!prev_ret)
99 return NULL;
100
101 while (prev && file_offset >= entry_end(prev_entry)) {
102 test = rb_next(prev);
103 if (!test)
104 break;
105 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
106 rb_node);
107 if (file_offset < entry_end(prev_entry))
108 break;
109
110 prev = test;
111 }
112 if (prev)
113 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
114 rb_node);
115 while (prev && file_offset < entry_end(prev_entry)) {
116 test = rb_prev(prev);
117 if (!test)
118 break;
119 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
120 rb_node);
121 prev = test;
122 }
123 *prev_ret = prev;
124 return NULL;
125}
126
127/*
128 * helper to check if a given offset is inside a given entry
129 */
130static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
131{
132 if (file_offset < entry->file_offset ||
133 entry->file_offset + entry->len <= file_offset)
134 return 0;
135 return 1;
136}
137
138static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
139 u64 len)
140{
141 if (file_offset + len <= entry->file_offset ||
142 entry->file_offset + entry->len <= file_offset)
143 return 0;
144 return 1;
145}
146
147/*
148 * look find the first ordered struct that has this offset, otherwise
149 * the first one less than this offset
150 */
151static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
152 u64 file_offset)
153{
154 struct rb_root *root = &tree->tree;
155 struct rb_node *prev = NULL;
156 struct rb_node *ret;
157 struct btrfs_ordered_extent *entry;
158
159 if (tree->last) {
160 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
161 rb_node);
162 if (offset_in_entry(entry, file_offset))
163 return tree->last;
164 }
165 ret = __tree_search(root, file_offset, &prev);
166 if (!ret)
167 ret = prev;
168 if (ret)
169 tree->last = ret;
170 return ret;
171}
172
173/* allocate and add a new ordered_extent into the per-inode tree.
174 * file_offset is the logical offset in the file
175 *
176 * start is the disk block number of an extent already reserved in the
177 * extent allocation tree
178 *
179 * len is the length of the extent
180 *
181 * The tree is given a single reference on the ordered extent that was
182 * inserted.
183 */
184static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
185 u64 start, u64 len, u64 disk_len,
186 int type, int dio, int compress_type)
187{
188 struct btrfs_root *root = BTRFS_I(inode)->root;
189 struct btrfs_ordered_inode_tree *tree;
190 struct rb_node *node;
191 struct btrfs_ordered_extent *entry;
192
193 tree = &BTRFS_I(inode)->ordered_tree;
194 entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
195 if (!entry)
196 return -ENOMEM;
197
198 entry->file_offset = file_offset;
199 entry->start = start;
200 entry->len = len;
201 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) &&
202 !(type == BTRFS_ORDERED_NOCOW))
203 entry->csum_bytes_left = disk_len;
204 entry->disk_len = disk_len;
205 entry->bytes_left = len;
206 entry->inode = igrab(inode);
207 entry->compress_type = compress_type;
208 entry->truncated_len = (u64)-1;
209 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
210 set_bit(type, &entry->flags);
211
212 if (dio)
213 set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
214
215 /* one ref for the tree */
216 atomic_set(&entry->refs, 1);
217 init_waitqueue_head(&entry->wait);
218 INIT_LIST_HEAD(&entry->list);
219 INIT_LIST_HEAD(&entry->root_extent_list);
220 INIT_LIST_HEAD(&entry->work_list);
221 init_completion(&entry->completion);
222 INIT_LIST_HEAD(&entry->log_list);
223
224 trace_btrfs_ordered_extent_add(inode, entry);
225
226 spin_lock_irq(&tree->lock);
227 node = tree_insert(&tree->tree, file_offset,
228 &entry->rb_node);
229 if (node)
230 ordered_data_tree_panic(inode, -EEXIST, file_offset);
231 spin_unlock_irq(&tree->lock);
232
233 spin_lock(&root->ordered_extent_lock);
234 list_add_tail(&entry->root_extent_list,
235 &root->ordered_extents);
236 root->nr_ordered_extents++;
237 if (root->nr_ordered_extents == 1) {
238 spin_lock(&root->fs_info->ordered_root_lock);
239 BUG_ON(!list_empty(&root->ordered_root));
240 list_add_tail(&root->ordered_root,
241 &root->fs_info->ordered_roots);
242 spin_unlock(&root->fs_info->ordered_root_lock);
243 }
244 spin_unlock(&root->ordered_extent_lock);
245
246 return 0;
247}
248
249int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
250 u64 start, u64 len, u64 disk_len, int type)
251{
252 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
253 disk_len, type, 0,
254 BTRFS_COMPRESS_NONE);
255}
256
257int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
258 u64 start, u64 len, u64 disk_len, int type)
259{
260 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
261 disk_len, type, 1,
262 BTRFS_COMPRESS_NONE);
263}
264
265int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset,
266 u64 start, u64 len, u64 disk_len,
267 int type, int compress_type)
268{
269 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
270 disk_len, type, 0,
271 compress_type);
272}
273
274/*
275 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
276 * when an ordered extent is finished. If the list covers more than one
277 * ordered extent, it is split across multiples.
278 */
279void btrfs_add_ordered_sum(struct inode *inode,
280 struct btrfs_ordered_extent *entry,
281 struct btrfs_ordered_sum *sum)
282{
283 struct btrfs_ordered_inode_tree *tree;
284
285 tree = &BTRFS_I(inode)->ordered_tree;
286 spin_lock_irq(&tree->lock);
287 list_add_tail(&sum->list, &entry->list);
288 WARN_ON(entry->csum_bytes_left < sum->len);
289 entry->csum_bytes_left -= sum->len;
290 if (entry->csum_bytes_left == 0)
291 wake_up(&entry->wait);
292 spin_unlock_irq(&tree->lock);
293}
294
295/*
296 * this is used to account for finished IO across a given range
297 * of the file. The IO may span ordered extents. If
298 * a given ordered_extent is completely done, 1 is returned, otherwise
299 * 0.
300 *
301 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
302 * to make sure this function only returns 1 once for a given ordered extent.
303 *
304 * file_offset is updated to one byte past the range that is recorded as
305 * complete. This allows you to walk forward in the file.
306 */
307int btrfs_dec_test_first_ordered_pending(struct inode *inode,
308 struct btrfs_ordered_extent **cached,
309 u64 *file_offset, u64 io_size, int uptodate)
310{
311 struct btrfs_ordered_inode_tree *tree;
312 struct rb_node *node;
313 struct btrfs_ordered_extent *entry = NULL;
314 int ret;
315 unsigned long flags;
316 u64 dec_end;
317 u64 dec_start;
318 u64 to_dec;
319
320 tree = &BTRFS_I(inode)->ordered_tree;
321 spin_lock_irqsave(&tree->lock, flags);
322 node = tree_search(tree, *file_offset);
323 if (!node) {
324 ret = 1;
325 goto out;
326 }
327
328 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
329 if (!offset_in_entry(entry, *file_offset)) {
330 ret = 1;
331 goto out;
332 }
333
334 dec_start = max(*file_offset, entry->file_offset);
335 dec_end = min(*file_offset + io_size, entry->file_offset +
336 entry->len);
337 *file_offset = dec_end;
338 if (dec_start > dec_end) {
339 btrfs_crit(BTRFS_I(inode)->root->fs_info,
340 "bad ordering dec_start %llu end %llu", dec_start, dec_end);
341 }
342 to_dec = dec_end - dec_start;
343 if (to_dec > entry->bytes_left) {
344 btrfs_crit(BTRFS_I(inode)->root->fs_info,
345 "bad ordered accounting left %llu size %llu",
346 entry->bytes_left, to_dec);
347 }
348 entry->bytes_left -= to_dec;
349 if (!uptodate)
350 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
351
352 if (entry->bytes_left == 0) {
353 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
354 if (waitqueue_active(&entry->wait))
355 wake_up(&entry->wait);
356 } else {
357 ret = 1;
358 }
359out:
360 if (!ret && cached && entry) {
361 *cached = entry;
362 atomic_inc(&entry->refs);
363 }
364 spin_unlock_irqrestore(&tree->lock, flags);
365 return ret == 0;
366}
367
368/*
369 * this is used to account for finished IO across a given range
370 * of the file. The IO should not span ordered extents. If
371 * a given ordered_extent is completely done, 1 is returned, otherwise
372 * 0.
373 *
374 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
375 * to make sure this function only returns 1 once for a given ordered extent.
376 */
377int btrfs_dec_test_ordered_pending(struct inode *inode,
378 struct btrfs_ordered_extent **cached,
379 u64 file_offset, u64 io_size, int uptodate)
380{
381 struct btrfs_ordered_inode_tree *tree;
382 struct rb_node *node;
383 struct btrfs_ordered_extent *entry = NULL;
384 unsigned long flags;
385 int ret;
386
387 tree = &BTRFS_I(inode)->ordered_tree;
388 spin_lock_irqsave(&tree->lock, flags);
389 if (cached && *cached) {
390 entry = *cached;
391 goto have_entry;
392 }
393
394 node = tree_search(tree, file_offset);
395 if (!node) {
396 ret = 1;
397 goto out;
398 }
399
400 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
401have_entry:
402 if (!offset_in_entry(entry, file_offset)) {
403 ret = 1;
404 goto out;
405 }
406
407 if (io_size > entry->bytes_left) {
408 btrfs_crit(BTRFS_I(inode)->root->fs_info,
409 "bad ordered accounting left %llu size %llu",
410 entry->bytes_left, io_size);
411 }
412 entry->bytes_left -= io_size;
413 if (!uptodate)
414 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
415
416 if (entry->bytes_left == 0) {
417 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
418 if (waitqueue_active(&entry->wait))
419 wake_up(&entry->wait);
420 } else {
421 ret = 1;
422 }
423out:
424 if (!ret && cached && entry) {
425 *cached = entry;
426 atomic_inc(&entry->refs);
427 }
428 spin_unlock_irqrestore(&tree->lock, flags);
429 return ret == 0;
430}
431
432/* Needs to either be called under a log transaction or the log_mutex */
433void btrfs_get_logged_extents(struct inode *inode,
434 struct list_head *logged_list)
435{
436 struct btrfs_ordered_inode_tree *tree;
437 struct btrfs_ordered_extent *ordered;
438 struct rb_node *n;
439
440 tree = &BTRFS_I(inode)->ordered_tree;
441 spin_lock_irq(&tree->lock);
442 for (n = rb_first(&tree->tree); n; n = rb_next(n)) {
443 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
444 if (!list_empty(&ordered->log_list))
445 continue;
446 list_add_tail(&ordered->log_list, logged_list);
447 atomic_inc(&ordered->refs);
448 }
449 spin_unlock_irq(&tree->lock);
450}
451
452void btrfs_put_logged_extents(struct list_head *logged_list)
453{
454 struct btrfs_ordered_extent *ordered;
455
456 while (!list_empty(logged_list)) {
457 ordered = list_first_entry(logged_list,
458 struct btrfs_ordered_extent,
459 log_list);
460 list_del_init(&ordered->log_list);
461 btrfs_put_ordered_extent(ordered);
462 }
463}
464
465void btrfs_submit_logged_extents(struct list_head *logged_list,
466 struct btrfs_root *log)
467{
468 int index = log->log_transid % 2;
469
470 spin_lock_irq(&log->log_extents_lock[index]);
471 list_splice_tail(logged_list, &log->logged_list[index]);
472 spin_unlock_irq(&log->log_extents_lock[index]);
473}
474
475void btrfs_wait_logged_extents(struct btrfs_root *log, u64 transid)
476{
477 struct btrfs_ordered_extent *ordered;
478 int index = transid % 2;
479
480 spin_lock_irq(&log->log_extents_lock[index]);
481 while (!list_empty(&log->logged_list[index])) {
482 ordered = list_first_entry(&log->logged_list[index],
483 struct btrfs_ordered_extent,
484 log_list);
485 list_del_init(&ordered->log_list);
486 spin_unlock_irq(&log->log_extents_lock[index]);
487 wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE,
488 &ordered->flags));
489 btrfs_put_ordered_extent(ordered);
490 spin_lock_irq(&log->log_extents_lock[index]);
491 }
492 spin_unlock_irq(&log->log_extents_lock[index]);
493}
494
495void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid)
496{
497 struct btrfs_ordered_extent *ordered;
498 int index = transid % 2;
499
500 spin_lock_irq(&log->log_extents_lock[index]);
501 while (!list_empty(&log->logged_list[index])) {
502 ordered = list_first_entry(&log->logged_list[index],
503 struct btrfs_ordered_extent,
504 log_list);
505 list_del_init(&ordered->log_list);
506 spin_unlock_irq(&log->log_extents_lock[index]);
507 btrfs_put_ordered_extent(ordered);
508 spin_lock_irq(&log->log_extents_lock[index]);
509 }
510 spin_unlock_irq(&log->log_extents_lock[index]);
511}
512
513/*
514 * used to drop a reference on an ordered extent. This will free
515 * the extent if the last reference is dropped
516 */
517void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
518{
519 struct list_head *cur;
520 struct btrfs_ordered_sum *sum;
521
522 trace_btrfs_ordered_extent_put(entry->inode, entry);
523
524 if (atomic_dec_and_test(&entry->refs)) {
525 if (entry->inode)
526 btrfs_add_delayed_iput(entry->inode);
527 while (!list_empty(&entry->list)) {
528 cur = entry->list.next;
529 sum = list_entry(cur, struct btrfs_ordered_sum, list);
530 list_del(&sum->list);
531 kfree(sum);
532 }
533 kmem_cache_free(btrfs_ordered_extent_cache, entry);
534 }
535}
536
537/*
538 * remove an ordered extent from the tree. No references are dropped
539 * and waiters are woken up.
540 */
541void btrfs_remove_ordered_extent(struct inode *inode,
542 struct btrfs_ordered_extent *entry)
543{
544 struct btrfs_ordered_inode_tree *tree;
545 struct btrfs_root *root = BTRFS_I(inode)->root;
546 struct rb_node *node;
547
548 tree = &BTRFS_I(inode)->ordered_tree;
549 spin_lock_irq(&tree->lock);
550 node = &entry->rb_node;
551 rb_erase(node, &tree->tree);
552 if (tree->last == node)
553 tree->last = NULL;
554 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
555 spin_unlock_irq(&tree->lock);
556
557 spin_lock(&root->ordered_extent_lock);
558 list_del_init(&entry->root_extent_list);
559 root->nr_ordered_extents--;
560
561 trace_btrfs_ordered_extent_remove(inode, entry);
562
563 /*
564 * we have no more ordered extents for this inode and
565 * no dirty pages. We can safely remove it from the
566 * list of ordered extents
567 */
568 if (RB_EMPTY_ROOT(&tree->tree) &&
569 !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
570 spin_lock(&root->fs_info->ordered_root_lock);
571 list_del_init(&BTRFS_I(inode)->ordered_operations);
572 spin_unlock(&root->fs_info->ordered_root_lock);
573 }
574
575 if (!root->nr_ordered_extents) {
576 spin_lock(&root->fs_info->ordered_root_lock);
577 BUG_ON(list_empty(&root->ordered_root));
578 list_del_init(&root->ordered_root);
579 spin_unlock(&root->fs_info->ordered_root_lock);
580 }
581 spin_unlock(&root->ordered_extent_lock);
582 wake_up(&entry->wait);
583}
584
585static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
586{
587 struct btrfs_ordered_extent *ordered;
588
589 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
590 btrfs_start_ordered_extent(ordered->inode, ordered, 1);
591 complete(&ordered->completion);
592}
593
594/*
595 * wait for all the ordered extents in a root. This is done when balancing
596 * space between drives.
597 */
598int btrfs_wait_ordered_extents(struct btrfs_root *root, int nr)
599{
600 struct list_head splice, works;
601 struct btrfs_ordered_extent *ordered, *next;
602 int count = 0;
603
604 INIT_LIST_HEAD(&splice);
605 INIT_LIST_HEAD(&works);
606
607 mutex_lock(&root->ordered_extent_mutex);
608 spin_lock(&root->ordered_extent_lock);
609 list_splice_init(&root->ordered_extents, &splice);
610 while (!list_empty(&splice) && nr) {
611 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
612 root_extent_list);
613 list_move_tail(&ordered->root_extent_list,
614 &root->ordered_extents);
615 atomic_inc(&ordered->refs);
616 spin_unlock(&root->ordered_extent_lock);
617
618 btrfs_init_work(&ordered->flush_work,
619 btrfs_run_ordered_extent_work, NULL, NULL);
620 list_add_tail(&ordered->work_list, &works);
621 btrfs_queue_work(root->fs_info->flush_workers,
622 &ordered->flush_work);
623
624 cond_resched();
625 spin_lock(&root->ordered_extent_lock);
626 if (nr != -1)
627 nr--;
628 count++;
629 }
630 list_splice_tail(&splice, &root->ordered_extents);
631 spin_unlock(&root->ordered_extent_lock);
632
633 list_for_each_entry_safe(ordered, next, &works, work_list) {
634 list_del_init(&ordered->work_list);
635 wait_for_completion(&ordered->completion);
636 btrfs_put_ordered_extent(ordered);
637 cond_resched();
638 }
639 mutex_unlock(&root->ordered_extent_mutex);
640
641 return count;
642}
643
644void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int nr)
645{
646 struct btrfs_root *root;
647 struct list_head splice;
648 int done;
649
650 INIT_LIST_HEAD(&splice);
651
652 mutex_lock(&fs_info->ordered_operations_mutex);
653 spin_lock(&fs_info->ordered_root_lock);
654 list_splice_init(&fs_info->ordered_roots, &splice);
655 while (!list_empty(&splice) && nr) {
656 root = list_first_entry(&splice, struct btrfs_root,
657 ordered_root);
658 root = btrfs_grab_fs_root(root);
659 BUG_ON(!root);
660 list_move_tail(&root->ordered_root,
661 &fs_info->ordered_roots);
662 spin_unlock(&fs_info->ordered_root_lock);
663
664 done = btrfs_wait_ordered_extents(root, nr);
665 btrfs_put_fs_root(root);
666
667 spin_lock(&fs_info->ordered_root_lock);
668 if (nr != -1) {
669 nr -= done;
670 WARN_ON(nr < 0);
671 }
672 }
673 list_splice_tail(&splice, &fs_info->ordered_roots);
674 spin_unlock(&fs_info->ordered_root_lock);
675 mutex_unlock(&fs_info->ordered_operations_mutex);
676}
677
678/*
679 * this is used during transaction commit to write all the inodes
680 * added to the ordered operation list. These files must be fully on
681 * disk before the transaction commits.
682 *
683 * we have two modes here, one is to just start the IO via filemap_flush
684 * and the other is to wait for all the io. When we wait, we have an
685 * extra check to make sure the ordered operation list really is empty
686 * before we return
687 */
688int btrfs_run_ordered_operations(struct btrfs_trans_handle *trans,
689 struct btrfs_root *root, int wait)
690{
691 struct btrfs_inode *btrfs_inode;
692 struct inode *inode;
693 struct btrfs_transaction *cur_trans = trans->transaction;
694 struct list_head splice;
695 struct list_head works;
696 struct btrfs_delalloc_work *work, *next;
697 int ret = 0;
698
699 INIT_LIST_HEAD(&splice);
700 INIT_LIST_HEAD(&works);
701
702 mutex_lock(&root->fs_info->ordered_extent_flush_mutex);
703 spin_lock(&root->fs_info->ordered_root_lock);
704 list_splice_init(&cur_trans->ordered_operations, &splice);
705 while (!list_empty(&splice)) {
706 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
707 ordered_operations);
708 inode = &btrfs_inode->vfs_inode;
709
710 list_del_init(&btrfs_inode->ordered_operations);
711
712 /*
713 * the inode may be getting freed (in sys_unlink path).
714 */
715 inode = igrab(inode);
716 if (!inode)
717 continue;
718
719 if (!wait)
720 list_add_tail(&BTRFS_I(inode)->ordered_operations,
721 &cur_trans->ordered_operations);
722 spin_unlock(&root->fs_info->ordered_root_lock);
723
724 work = btrfs_alloc_delalloc_work(inode, wait, 1);
725 if (!work) {
726 spin_lock(&root->fs_info->ordered_root_lock);
727 if (list_empty(&BTRFS_I(inode)->ordered_operations))
728 list_add_tail(&btrfs_inode->ordered_operations,
729 &splice);
730 list_splice_tail(&splice,
731 &cur_trans->ordered_operations);
732 spin_unlock(&root->fs_info->ordered_root_lock);
733 ret = -ENOMEM;
734 goto out;
735 }
736 list_add_tail(&work->list, &works);
737 btrfs_queue_work(root->fs_info->flush_workers,
738 &work->work);
739
740 cond_resched();
741 spin_lock(&root->fs_info->ordered_root_lock);
742 }
743 spin_unlock(&root->fs_info->ordered_root_lock);
744out:
745 list_for_each_entry_safe(work, next, &works, list) {
746 list_del_init(&work->list);
747 btrfs_wait_and_free_delalloc_work(work);
748 }
749 mutex_unlock(&root->fs_info->ordered_extent_flush_mutex);
750 return ret;
751}
752
753/*
754 * Used to start IO or wait for a given ordered extent to finish.
755 *
756 * If wait is one, this effectively waits on page writeback for all the pages
757 * in the extent, and it waits on the io completion code to insert
758 * metadata into the btree corresponding to the extent
759 */
760void btrfs_start_ordered_extent(struct inode *inode,
761 struct btrfs_ordered_extent *entry,
762 int wait)
763{
764 u64 start = entry->file_offset;
765 u64 end = start + entry->len - 1;
766
767 trace_btrfs_ordered_extent_start(inode, entry);
768
769 /*
770 * pages in the range can be dirty, clean or writeback. We
771 * start IO on any dirty ones so the wait doesn't stall waiting
772 * for the flusher thread to find them
773 */
774 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
775 filemap_fdatawrite_range(inode->i_mapping, start, end);
776 if (wait) {
777 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
778 &entry->flags));
779 }
780}
781
782/*
783 * Used to wait on ordered extents across a large range of bytes.
784 */
785int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
786{
787 int ret = 0;
788 u64 end;
789 u64 orig_end;
790 struct btrfs_ordered_extent *ordered;
791
792 if (start + len < start) {
793 orig_end = INT_LIMIT(loff_t);
794 } else {
795 orig_end = start + len - 1;
796 if (orig_end > INT_LIMIT(loff_t))
797 orig_end = INT_LIMIT(loff_t);
798 }
799
800 /* start IO across the range first to instantiate any delalloc
801 * extents
802 */
803 ret = filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
804 if (ret)
805 return ret;
806 /*
807 * So with compression we will find and lock a dirty page and clear the
808 * first one as dirty, setup an async extent, and immediately return
809 * with the entire range locked but with nobody actually marked with
810 * writeback. So we can't just filemap_write_and_wait_range() and
811 * expect it to work since it will just kick off a thread to do the
812 * actual work. So we need to call filemap_fdatawrite_range _again_
813 * since it will wait on the page lock, which won't be unlocked until
814 * after the pages have been marked as writeback and so we're good to go
815 * from there. We have to do this otherwise we'll miss the ordered
816 * extents and that results in badness. Please Josef, do not think you
817 * know better and pull this out at some point in the future, it is
818 * right and you are wrong.
819 */
820 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
821 &BTRFS_I(inode)->runtime_flags)) {
822 ret = filemap_fdatawrite_range(inode->i_mapping, start,
823 orig_end);
824 if (ret)
825 return ret;
826 }
827 ret = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
828 if (ret)
829 return ret;
830
831 end = orig_end;
832 while (1) {
833 ordered = btrfs_lookup_first_ordered_extent(inode, end);
834 if (!ordered)
835 break;
836 if (ordered->file_offset > orig_end) {
837 btrfs_put_ordered_extent(ordered);
838 break;
839 }
840 if (ordered->file_offset + ordered->len <= start) {
841 btrfs_put_ordered_extent(ordered);
842 break;
843 }
844 btrfs_start_ordered_extent(inode, ordered, 1);
845 end = ordered->file_offset;
846 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
847 ret = -EIO;
848 btrfs_put_ordered_extent(ordered);
849 if (ret || end == 0 || end == start)
850 break;
851 end--;
852 }
853 return ret;
854}
855
856/*
857 * find an ordered extent corresponding to file_offset. return NULL if
858 * nothing is found, otherwise take a reference on the extent and return it
859 */
860struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
861 u64 file_offset)
862{
863 struct btrfs_ordered_inode_tree *tree;
864 struct rb_node *node;
865 struct btrfs_ordered_extent *entry = NULL;
866
867 tree = &BTRFS_I(inode)->ordered_tree;
868 spin_lock_irq(&tree->lock);
869 node = tree_search(tree, file_offset);
870 if (!node)
871 goto out;
872
873 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
874 if (!offset_in_entry(entry, file_offset))
875 entry = NULL;
876 if (entry)
877 atomic_inc(&entry->refs);
878out:
879 spin_unlock_irq(&tree->lock);
880 return entry;
881}
882
883/* Since the DIO code tries to lock a wide area we need to look for any ordered
884 * extents that exist in the range, rather than just the start of the range.
885 */
886struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode,
887 u64 file_offset,
888 u64 len)
889{
890 struct btrfs_ordered_inode_tree *tree;
891 struct rb_node *node;
892 struct btrfs_ordered_extent *entry = NULL;
893
894 tree = &BTRFS_I(inode)->ordered_tree;
895 spin_lock_irq(&tree->lock);
896 node = tree_search(tree, file_offset);
897 if (!node) {
898 node = tree_search(tree, file_offset + len);
899 if (!node)
900 goto out;
901 }
902
903 while (1) {
904 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
905 if (range_overlaps(entry, file_offset, len))
906 break;
907
908 if (entry->file_offset >= file_offset + len) {
909 entry = NULL;
910 break;
911 }
912 entry = NULL;
913 node = rb_next(node);
914 if (!node)
915 break;
916 }
917out:
918 if (entry)
919 atomic_inc(&entry->refs);
920 spin_unlock_irq(&tree->lock);
921 return entry;
922}
923
924/*
925 * lookup and return any extent before 'file_offset'. NULL is returned
926 * if none is found
927 */
928struct btrfs_ordered_extent *
929btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
930{
931 struct btrfs_ordered_inode_tree *tree;
932 struct rb_node *node;
933 struct btrfs_ordered_extent *entry = NULL;
934
935 tree = &BTRFS_I(inode)->ordered_tree;
936 spin_lock_irq(&tree->lock);
937 node = tree_search(tree, file_offset);
938 if (!node)
939 goto out;
940
941 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
942 atomic_inc(&entry->refs);
943out:
944 spin_unlock_irq(&tree->lock);
945 return entry;
946}
947
948/*
949 * After an extent is done, call this to conditionally update the on disk
950 * i_size. i_size is updated to cover any fully written part of the file.
951 */
952int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
953 struct btrfs_ordered_extent *ordered)
954{
955 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
956 u64 disk_i_size;
957 u64 new_i_size;
958 u64 i_size = i_size_read(inode);
959 struct rb_node *node;
960 struct rb_node *prev = NULL;
961 struct btrfs_ordered_extent *test;
962 int ret = 1;
963
964 spin_lock_irq(&tree->lock);
965 if (ordered) {
966 offset = entry_end(ordered);
967 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags))
968 offset = min(offset,
969 ordered->file_offset +
970 ordered->truncated_len);
971 } else {
972 offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
973 }
974 disk_i_size = BTRFS_I(inode)->disk_i_size;
975
976 /* truncate file */
977 if (disk_i_size > i_size) {
978 BTRFS_I(inode)->disk_i_size = i_size;
979 ret = 0;
980 goto out;
981 }
982
983 /*
984 * if the disk i_size is already at the inode->i_size, or
985 * this ordered extent is inside the disk i_size, we're done
986 */
987 if (disk_i_size == i_size)
988 goto out;
989
990 /*
991 * We still need to update disk_i_size if outstanding_isize is greater
992 * than disk_i_size.
993 */
994 if (offset <= disk_i_size &&
995 (!ordered || ordered->outstanding_isize <= disk_i_size))
996 goto out;
997
998 /*
999 * walk backward from this ordered extent to disk_i_size.
1000 * if we find an ordered extent then we can't update disk i_size
1001 * yet
1002 */
1003 if (ordered) {
1004 node = rb_prev(&ordered->rb_node);
1005 } else {
1006 prev = tree_search(tree, offset);
1007 /*
1008 * we insert file extents without involving ordered struct,
1009 * so there should be no ordered struct cover this offset
1010 */
1011 if (prev) {
1012 test = rb_entry(prev, struct btrfs_ordered_extent,
1013 rb_node);
1014 BUG_ON(offset_in_entry(test, offset));
1015 }
1016 node = prev;
1017 }
1018 for (; node; node = rb_prev(node)) {
1019 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
1020
1021 /* We treat this entry as if it doesnt exist */
1022 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags))
1023 continue;
1024 if (test->file_offset + test->len <= disk_i_size)
1025 break;
1026 if (test->file_offset >= i_size)
1027 break;
1028 if (entry_end(test) > disk_i_size) {
1029 /*
1030 * we don't update disk_i_size now, so record this
1031 * undealt i_size. Or we will not know the real
1032 * i_size.
1033 */
1034 if (test->outstanding_isize < offset)
1035 test->outstanding_isize = offset;
1036 if (ordered &&
1037 ordered->outstanding_isize >
1038 test->outstanding_isize)
1039 test->outstanding_isize =
1040 ordered->outstanding_isize;
1041 goto out;
1042 }
1043 }
1044 new_i_size = min_t(u64, offset, i_size);
1045
1046 /*
1047 * Some ordered extents may completed before the current one, and
1048 * we hold the real i_size in ->outstanding_isize.
1049 */
1050 if (ordered && ordered->outstanding_isize > new_i_size)
1051 new_i_size = min_t(u64, ordered->outstanding_isize, i_size);
1052 BTRFS_I(inode)->disk_i_size = new_i_size;
1053 ret = 0;
1054out:
1055 /*
1056 * We need to do this because we can't remove ordered extents until
1057 * after the i_disk_size has been updated and then the inode has been
1058 * updated to reflect the change, so we need to tell anybody who finds
1059 * this ordered extent that we've already done all the real work, we
1060 * just haven't completed all the other work.
1061 */
1062 if (ordered)
1063 set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags);
1064 spin_unlock_irq(&tree->lock);
1065 return ret;
1066}
1067
1068/*
1069 * search the ordered extents for one corresponding to 'offset' and
1070 * try to find a checksum. This is used because we allow pages to
1071 * be reclaimed before their checksum is actually put into the btree
1072 */
1073int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
1074 u32 *sum, int len)
1075{
1076 struct btrfs_ordered_sum *ordered_sum;
1077 struct btrfs_ordered_extent *ordered;
1078 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
1079 unsigned long num_sectors;
1080 unsigned long i;
1081 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
1082 int index = 0;
1083
1084 ordered = btrfs_lookup_ordered_extent(inode, offset);
1085 if (!ordered)
1086 return 0;
1087
1088 spin_lock_irq(&tree->lock);
1089 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
1090 if (disk_bytenr >= ordered_sum->bytenr &&
1091 disk_bytenr < ordered_sum->bytenr + ordered_sum->len) {
1092 i = (disk_bytenr - ordered_sum->bytenr) >>
1093 inode->i_sb->s_blocksize_bits;
1094 num_sectors = ordered_sum->len >>
1095 inode->i_sb->s_blocksize_bits;
1096 num_sectors = min_t(int, len - index, num_sectors - i);
1097 memcpy(sum + index, ordered_sum->sums + i,
1098 num_sectors);
1099
1100 index += (int)num_sectors;
1101 if (index == len)
1102 goto out;
1103 disk_bytenr += num_sectors * sectorsize;
1104 }
1105 }
1106out:
1107 spin_unlock_irq(&tree->lock);
1108 btrfs_put_ordered_extent(ordered);
1109 return index;
1110}
1111
1112
1113/*
1114 * add a given inode to the list of inodes that must be fully on
1115 * disk before a transaction commit finishes.
1116 *
1117 * This basically gives us the ext3 style data=ordered mode, and it is mostly
1118 * used to make sure renamed files are fully on disk.
1119 *
1120 * It is a noop if the inode is already fully on disk.
1121 *
1122 * If trans is not null, we'll do a friendly check for a transaction that
1123 * is already flushing things and force the IO down ourselves.
1124 */
1125void btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
1126 struct btrfs_root *root, struct inode *inode)
1127{
1128 struct btrfs_transaction *cur_trans = trans->transaction;
1129 u64 last_mod;
1130
1131 last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
1132
1133 /*
1134 * if this file hasn't been changed since the last transaction
1135 * commit, we can safely return without doing anything
1136 */
1137 if (last_mod <= root->fs_info->last_trans_committed)
1138 return;
1139
1140 spin_lock(&root->fs_info->ordered_root_lock);
1141 if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
1142 list_add_tail(&BTRFS_I(inode)->ordered_operations,
1143 &cur_trans->ordered_operations);
1144 }
1145 spin_unlock(&root->fs_info->ordered_root_lock);
1146}
1147
1148int __init ordered_data_init(void)
1149{
1150 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1151 sizeof(struct btrfs_ordered_extent), 0,
1152 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
1153 NULL);
1154 if (!btrfs_ordered_extent_cache)
1155 return -ENOMEM;
1156
1157 return 0;
1158}
1159
1160void ordered_data_exit(void)
1161{
1162 if (btrfs_ordered_extent_cache)
1163 kmem_cache_destroy(btrfs_ordered_extent_cache);
1164}