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
19#include <linux/fs.h>
20#include <linux/blkdev.h>
21#include <linux/scatterlist.h>
22#include <linux/swap.h>
23#include <linux/radix-tree.h>
24#include <linux/writeback.h>
25#include <linux/buffer_head.h>
26#include <linux/workqueue.h>
27#include <linux/kthread.h>
28#include <linux/freezer.h>
29#include <linux/slab.h>
30#include <linux/migrate.h>
31#include <linux/ratelimit.h>
32#include <linux/uuid.h>
33#include <linux/semaphore.h>
34#include <asm/unaligned.h>
35#include "ctree.h"
36#include "disk-io.h"
37#include "hash.h"
38#include "transaction.h"
39#include "btrfs_inode.h"
40#include "volumes.h"
41#include "print-tree.h"
42#include "async-thread.h"
43#include "locking.h"
44#include "tree-log.h"
45#include "free-space-cache.h"
46#include "inode-map.h"
47#include "check-integrity.h"
48#include "rcu-string.h"
49#include "dev-replace.h"
50#include "raid56.h"
51#include "sysfs.h"
52
53#ifdef CONFIG_X86
54#include <asm/cpufeature.h>
55#endif
56
57static struct extent_io_ops btree_extent_io_ops;
58static void end_workqueue_fn(struct btrfs_work *work);
59static void free_fs_root(struct btrfs_root *root);
60static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
61 int read_only);
62static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
63 struct btrfs_root *root);
64static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
65static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
66 struct btrfs_root *root);
67static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
68static int btrfs_destroy_marked_extents(struct btrfs_root *root,
69 struct extent_io_tree *dirty_pages,
70 int mark);
71static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
72 struct extent_io_tree *pinned_extents);
73static int btrfs_cleanup_transaction(struct btrfs_root *root);
74static void btrfs_error_commit_super(struct btrfs_root *root);
75
76/*
77 * end_io_wq structs are used to do processing in task context when an IO is
78 * complete. This is used during reads to verify checksums, and it is used
79 * by writes to insert metadata for new file extents after IO is complete.
80 */
81struct end_io_wq {
82 struct bio *bio;
83 bio_end_io_t *end_io;
84 void *private;
85 struct btrfs_fs_info *info;
86 int error;
87 int metadata;
88 struct list_head list;
89 struct btrfs_work work;
90};
91
92/*
93 * async submit bios are used to offload expensive checksumming
94 * onto the worker threads. They checksum file and metadata bios
95 * just before they are sent down the IO stack.
96 */
97struct async_submit_bio {
98 struct inode *inode;
99 struct bio *bio;
100 struct list_head list;
101 extent_submit_bio_hook_t *submit_bio_start;
102 extent_submit_bio_hook_t *submit_bio_done;
103 int rw;
104 int mirror_num;
105 unsigned long bio_flags;
106 /*
107 * bio_offset is optional, can be used if the pages in the bio
108 * can't tell us where in the file the bio should go
109 */
110 u64 bio_offset;
111 struct btrfs_work work;
112 int error;
113};
114
115/*
116 * Lockdep class keys for extent_buffer->lock's in this root. For a given
117 * eb, the lockdep key is determined by the btrfs_root it belongs to and
118 * the level the eb occupies in the tree.
119 *
120 * Different roots are used for different purposes and may nest inside each
121 * other and they require separate keysets. As lockdep keys should be
122 * static, assign keysets according to the purpose of the root as indicated
123 * by btrfs_root->objectid. This ensures that all special purpose roots
124 * have separate keysets.
125 *
126 * Lock-nesting across peer nodes is always done with the immediate parent
127 * node locked thus preventing deadlock. As lockdep doesn't know this, use
128 * subclass to avoid triggering lockdep warning in such cases.
129 *
130 * The key is set by the readpage_end_io_hook after the buffer has passed
131 * csum validation but before the pages are unlocked. It is also set by
132 * btrfs_init_new_buffer on freshly allocated blocks.
133 *
134 * We also add a check to make sure the highest level of the tree is the
135 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
136 * needs update as well.
137 */
138#ifdef CONFIG_DEBUG_LOCK_ALLOC
139# if BTRFS_MAX_LEVEL != 8
140# error
141# endif
142
143static struct btrfs_lockdep_keyset {
144 u64 id; /* root objectid */
145 const char *name_stem; /* lock name stem */
146 char names[BTRFS_MAX_LEVEL + 1][20];
147 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
148} btrfs_lockdep_keysets[] = {
149 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
150 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
151 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
152 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
153 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
154 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
155 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
156 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
157 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
158 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
159 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
160 { .id = 0, .name_stem = "tree" },
161};
162
163void __init btrfs_init_lockdep(void)
164{
165 int i, j;
166
167 /* initialize lockdep class names */
168 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
169 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
170
171 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
172 snprintf(ks->names[j], sizeof(ks->names[j]),
173 "btrfs-%s-%02d", ks->name_stem, j);
174 }
175}
176
177void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
178 int level)
179{
180 struct btrfs_lockdep_keyset *ks;
181
182 BUG_ON(level >= ARRAY_SIZE(ks->keys));
183
184 /* find the matching keyset, id 0 is the default entry */
185 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
186 if (ks->id == objectid)
187 break;
188
189 lockdep_set_class_and_name(&eb->lock,
190 &ks->keys[level], ks->names[level]);
191}
192
193#endif
194
195/*
196 * extents on the btree inode are pretty simple, there's one extent
197 * that covers the entire device
198 */
199static struct extent_map *btree_get_extent(struct inode *inode,
200 struct page *page, size_t pg_offset, u64 start, u64 len,
201 int create)
202{
203 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
204 struct extent_map *em;
205 int ret;
206
207 read_lock(&em_tree->lock);
208 em = lookup_extent_mapping(em_tree, start, len);
209 if (em) {
210 em->bdev =
211 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
212 read_unlock(&em_tree->lock);
213 goto out;
214 }
215 read_unlock(&em_tree->lock);
216
217 em = alloc_extent_map();
218 if (!em) {
219 em = ERR_PTR(-ENOMEM);
220 goto out;
221 }
222 em->start = 0;
223 em->len = (u64)-1;
224 em->block_len = (u64)-1;
225 em->block_start = 0;
226 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
227
228 write_lock(&em_tree->lock);
229 ret = add_extent_mapping(em_tree, em, 0);
230 if (ret == -EEXIST) {
231 free_extent_map(em);
232 em = lookup_extent_mapping(em_tree, start, len);
233 if (!em)
234 em = ERR_PTR(-EIO);
235 } else if (ret) {
236 free_extent_map(em);
237 em = ERR_PTR(ret);
238 }
239 write_unlock(&em_tree->lock);
240
241out:
242 return em;
243}
244
245u32 btrfs_csum_data(char *data, u32 seed, size_t len)
246{
247 return btrfs_crc32c(seed, data, len);
248}
249
250void btrfs_csum_final(u32 crc, char *result)
251{
252 put_unaligned_le32(~crc, result);
253}
254
255/*
256 * compute the csum for a btree block, and either verify it or write it
257 * into the csum field of the block.
258 */
259static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
260 int verify)
261{
262 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
263 char *result = NULL;
264 unsigned long len;
265 unsigned long cur_len;
266 unsigned long offset = BTRFS_CSUM_SIZE;
267 char *kaddr;
268 unsigned long map_start;
269 unsigned long map_len;
270 int err;
271 u32 crc = ~(u32)0;
272 unsigned long inline_result;
273
274 len = buf->len - offset;
275 while (len > 0) {
276 err = map_private_extent_buffer(buf, offset, 32,
277 &kaddr, &map_start, &map_len);
278 if (err)
279 return 1;
280 cur_len = min(len, map_len - (offset - map_start));
281 crc = btrfs_csum_data(kaddr + offset - map_start,
282 crc, cur_len);
283 len -= cur_len;
284 offset += cur_len;
285 }
286 if (csum_size > sizeof(inline_result)) {
287 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
288 if (!result)
289 return 1;
290 } else {
291 result = (char *)&inline_result;
292 }
293
294 btrfs_csum_final(crc, result);
295
296 if (verify) {
297 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
298 u32 val;
299 u32 found = 0;
300 memcpy(&found, result, csum_size);
301
302 read_extent_buffer(buf, &val, 0, csum_size);
303 printk_ratelimited(KERN_INFO
304 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
305 "level %d\n",
306 root->fs_info->sb->s_id, buf->start,
307 val, found, btrfs_header_level(buf));
308 if (result != (char *)&inline_result)
309 kfree(result);
310 return 1;
311 }
312 } else {
313 write_extent_buffer(buf, result, 0, csum_size);
314 }
315 if (result != (char *)&inline_result)
316 kfree(result);
317 return 0;
318}
319
320/*
321 * we can't consider a given block up to date unless the transid of the
322 * block matches the transid in the parent node's pointer. This is how we
323 * detect blocks that either didn't get written at all or got written
324 * in the wrong place.
325 */
326static int verify_parent_transid(struct extent_io_tree *io_tree,
327 struct extent_buffer *eb, u64 parent_transid,
328 int atomic)
329{
330 struct extent_state *cached_state = NULL;
331 int ret;
332 bool need_lock = (current->journal_info ==
333 (void *)BTRFS_SEND_TRANS_STUB);
334
335 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
336 return 0;
337
338 if (atomic)
339 return -EAGAIN;
340
341 if (need_lock) {
342 btrfs_tree_read_lock(eb);
343 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
344 }
345
346 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
347 0, &cached_state);
348 if (extent_buffer_uptodate(eb) &&
349 btrfs_header_generation(eb) == parent_transid) {
350 ret = 0;
351 goto out;
352 }
353 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
354 "found %llu\n",
355 eb->start, parent_transid, btrfs_header_generation(eb));
356 ret = 1;
357
358 /*
359 * Things reading via commit roots that don't have normal protection,
360 * like send, can have a really old block in cache that may point at a
361 * block that has been free'd and re-allocated. So don't clear uptodate
362 * if we find an eb that is under IO (dirty/writeback) because we could
363 * end up reading in the stale data and then writing it back out and
364 * making everybody very sad.
365 */
366 if (!extent_buffer_under_io(eb))
367 clear_extent_buffer_uptodate(eb);
368out:
369 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
370 &cached_state, GFP_NOFS);
371 btrfs_tree_read_unlock_blocking(eb);
372 return ret;
373}
374
375/*
376 * Return 0 if the superblock checksum type matches the checksum value of that
377 * algorithm. Pass the raw disk superblock data.
378 */
379static int btrfs_check_super_csum(char *raw_disk_sb)
380{
381 struct btrfs_super_block *disk_sb =
382 (struct btrfs_super_block *)raw_disk_sb;
383 u16 csum_type = btrfs_super_csum_type(disk_sb);
384 int ret = 0;
385
386 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
387 u32 crc = ~(u32)0;
388 const int csum_size = sizeof(crc);
389 char result[csum_size];
390
391 /*
392 * The super_block structure does not span the whole
393 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
394 * is filled with zeros and is included in the checkum.
395 */
396 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
397 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
398 btrfs_csum_final(crc, result);
399
400 if (memcmp(raw_disk_sb, result, csum_size))
401 ret = 1;
402
403 if (ret && btrfs_super_generation(disk_sb) < 10) {
404 printk(KERN_WARNING
405 "BTRFS: super block crcs don't match, older mkfs detected\n");
406 ret = 0;
407 }
408 }
409
410 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
411 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
412 csum_type);
413 ret = 1;
414 }
415
416 return ret;
417}
418
419/*
420 * helper to read a given tree block, doing retries as required when
421 * the checksums don't match and we have alternate mirrors to try.
422 */
423static int btree_read_extent_buffer_pages(struct btrfs_root *root,
424 struct extent_buffer *eb,
425 u64 start, u64 parent_transid)
426{
427 struct extent_io_tree *io_tree;
428 int failed = 0;
429 int ret;
430 int num_copies = 0;
431 int mirror_num = 0;
432 int failed_mirror = 0;
433
434 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
435 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
436 while (1) {
437 ret = read_extent_buffer_pages(io_tree, eb, start,
438 WAIT_COMPLETE,
439 btree_get_extent, mirror_num);
440 if (!ret) {
441 if (!verify_parent_transid(io_tree, eb,
442 parent_transid, 0))
443 break;
444 else
445 ret = -EIO;
446 }
447
448 /*
449 * This buffer's crc is fine, but its contents are corrupted, so
450 * there is no reason to read the other copies, they won't be
451 * any less wrong.
452 */
453 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
454 break;
455
456 num_copies = btrfs_num_copies(root->fs_info,
457 eb->start, eb->len);
458 if (num_copies == 1)
459 break;
460
461 if (!failed_mirror) {
462 failed = 1;
463 failed_mirror = eb->read_mirror;
464 }
465
466 mirror_num++;
467 if (mirror_num == failed_mirror)
468 mirror_num++;
469
470 if (mirror_num > num_copies)
471 break;
472 }
473
474 if (failed && !ret && failed_mirror)
475 repair_eb_io_failure(root, eb, failed_mirror);
476
477 return ret;
478}
479
480/*
481 * checksum a dirty tree block before IO. This has extra checks to make sure
482 * we only fill in the checksum field in the first page of a multi-page block
483 */
484
485static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
486{
487 u64 start = page_offset(page);
488 u64 found_start;
489 struct extent_buffer *eb;
490
491 eb = (struct extent_buffer *)page->private;
492 if (page != eb->pages[0])
493 return 0;
494 found_start = btrfs_header_bytenr(eb);
495 if (WARN_ON(found_start != start || !PageUptodate(page)))
496 return 0;
497 csum_tree_block(root, eb, 0);
498 return 0;
499}
500
501static int check_tree_block_fsid(struct btrfs_root *root,
502 struct extent_buffer *eb)
503{
504 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
505 u8 fsid[BTRFS_UUID_SIZE];
506 int ret = 1;
507
508 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
509 while (fs_devices) {
510 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
511 ret = 0;
512 break;
513 }
514 fs_devices = fs_devices->seed;
515 }
516 return ret;
517}
518
519#define CORRUPT(reason, eb, root, slot) \
520 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
521 "root=%llu, slot=%d", reason, \
522 btrfs_header_bytenr(eb), root->objectid, slot)
523
524static noinline int check_leaf(struct btrfs_root *root,
525 struct extent_buffer *leaf)
526{
527 struct btrfs_key key;
528 struct btrfs_key leaf_key;
529 u32 nritems = btrfs_header_nritems(leaf);
530 int slot;
531
532 if (nritems == 0)
533 return 0;
534
535 /* Check the 0 item */
536 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
537 BTRFS_LEAF_DATA_SIZE(root)) {
538 CORRUPT("invalid item offset size pair", leaf, root, 0);
539 return -EIO;
540 }
541
542 /*
543 * Check to make sure each items keys are in the correct order and their
544 * offsets make sense. We only have to loop through nritems-1 because
545 * we check the current slot against the next slot, which verifies the
546 * next slot's offset+size makes sense and that the current's slot
547 * offset is correct.
548 */
549 for (slot = 0; slot < nritems - 1; slot++) {
550 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
551 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
552
553 /* Make sure the keys are in the right order */
554 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
555 CORRUPT("bad key order", leaf, root, slot);
556 return -EIO;
557 }
558
559 /*
560 * Make sure the offset and ends are right, remember that the
561 * item data starts at the end of the leaf and grows towards the
562 * front.
563 */
564 if (btrfs_item_offset_nr(leaf, slot) !=
565 btrfs_item_end_nr(leaf, slot + 1)) {
566 CORRUPT("slot offset bad", leaf, root, slot);
567 return -EIO;
568 }
569
570 /*
571 * Check to make sure that we don't point outside of the leaf,
572 * just incase all the items are consistent to eachother, but
573 * all point outside of the leaf.
574 */
575 if (btrfs_item_end_nr(leaf, slot) >
576 BTRFS_LEAF_DATA_SIZE(root)) {
577 CORRUPT("slot end outside of leaf", leaf, root, slot);
578 return -EIO;
579 }
580 }
581
582 return 0;
583}
584
585static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
586 u64 phy_offset, struct page *page,
587 u64 start, u64 end, int mirror)
588{
589 u64 found_start;
590 int found_level;
591 struct extent_buffer *eb;
592 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
593 int ret = 0;
594 int reads_done;
595
596 if (!page->private)
597 goto out;
598
599 eb = (struct extent_buffer *)page->private;
600
601 /* the pending IO might have been the only thing that kept this buffer
602 * in memory. Make sure we have a ref for all this other checks
603 */
604 extent_buffer_get(eb);
605
606 reads_done = atomic_dec_and_test(&eb->io_pages);
607 if (!reads_done)
608 goto err;
609
610 eb->read_mirror = mirror;
611 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
612 ret = -EIO;
613 goto err;
614 }
615
616 found_start = btrfs_header_bytenr(eb);
617 if (found_start != eb->start) {
618 printk_ratelimited(KERN_INFO "BTRFS: bad tree block start "
619 "%llu %llu\n",
620 found_start, eb->start);
621 ret = -EIO;
622 goto err;
623 }
624 if (check_tree_block_fsid(root, eb)) {
625 printk_ratelimited(KERN_INFO "BTRFS: bad fsid on block %llu\n",
626 eb->start);
627 ret = -EIO;
628 goto err;
629 }
630 found_level = btrfs_header_level(eb);
631 if (found_level >= BTRFS_MAX_LEVEL) {
632 btrfs_info(root->fs_info, "bad tree block level %d",
633 (int)btrfs_header_level(eb));
634 ret = -EIO;
635 goto err;
636 }
637
638 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
639 eb, found_level);
640
641 ret = csum_tree_block(root, eb, 1);
642 if (ret) {
643 ret = -EIO;
644 goto err;
645 }
646
647 /*
648 * If this is a leaf block and it is corrupt, set the corrupt bit so
649 * that we don't try and read the other copies of this block, just
650 * return -EIO.
651 */
652 if (found_level == 0 && check_leaf(root, eb)) {
653 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
654 ret = -EIO;
655 }
656
657 if (!ret)
658 set_extent_buffer_uptodate(eb);
659err:
660 if (reads_done &&
661 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
662 btree_readahead_hook(root, eb, eb->start, ret);
663
664 if (ret) {
665 /*
666 * our io error hook is going to dec the io pages
667 * again, we have to make sure it has something
668 * to decrement
669 */
670 atomic_inc(&eb->io_pages);
671 clear_extent_buffer_uptodate(eb);
672 }
673 free_extent_buffer(eb);
674out:
675 return ret;
676}
677
678static int btree_io_failed_hook(struct page *page, int failed_mirror)
679{
680 struct extent_buffer *eb;
681 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
682
683 eb = (struct extent_buffer *)page->private;
684 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
685 eb->read_mirror = failed_mirror;
686 atomic_dec(&eb->io_pages);
687 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
688 btree_readahead_hook(root, eb, eb->start, -EIO);
689 return -EIO; /* we fixed nothing */
690}
691
692static void end_workqueue_bio(struct bio *bio, int err)
693{
694 struct end_io_wq *end_io_wq = bio->bi_private;
695 struct btrfs_fs_info *fs_info;
696
697 fs_info = end_io_wq->info;
698 end_io_wq->error = err;
699 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
700
701 if (bio->bi_rw & REQ_WRITE) {
702 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
703 btrfs_queue_work(fs_info->endio_meta_write_workers,
704 &end_io_wq->work);
705 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
706 btrfs_queue_work(fs_info->endio_freespace_worker,
707 &end_io_wq->work);
708 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
709 btrfs_queue_work(fs_info->endio_raid56_workers,
710 &end_io_wq->work);
711 else
712 btrfs_queue_work(fs_info->endio_write_workers,
713 &end_io_wq->work);
714 } else {
715 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
716 btrfs_queue_work(fs_info->endio_raid56_workers,
717 &end_io_wq->work);
718 else if (end_io_wq->metadata)
719 btrfs_queue_work(fs_info->endio_meta_workers,
720 &end_io_wq->work);
721 else
722 btrfs_queue_work(fs_info->endio_workers,
723 &end_io_wq->work);
724 }
725}
726
727/*
728 * For the metadata arg you want
729 *
730 * 0 - if data
731 * 1 - if normal metadta
732 * 2 - if writing to the free space cache area
733 * 3 - raid parity work
734 */
735int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
736 int metadata)
737{
738 struct end_io_wq *end_io_wq;
739 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
740 if (!end_io_wq)
741 return -ENOMEM;
742
743 end_io_wq->private = bio->bi_private;
744 end_io_wq->end_io = bio->bi_end_io;
745 end_io_wq->info = info;
746 end_io_wq->error = 0;
747 end_io_wq->bio = bio;
748 end_io_wq->metadata = metadata;
749
750 bio->bi_private = end_io_wq;
751 bio->bi_end_io = end_workqueue_bio;
752 return 0;
753}
754
755unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
756{
757 unsigned long limit = min_t(unsigned long,
758 info->thread_pool_size,
759 info->fs_devices->open_devices);
760 return 256 * limit;
761}
762
763static void run_one_async_start(struct btrfs_work *work)
764{
765 struct async_submit_bio *async;
766 int ret;
767
768 async = container_of(work, struct async_submit_bio, work);
769 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
770 async->mirror_num, async->bio_flags,
771 async->bio_offset);
772 if (ret)
773 async->error = ret;
774}
775
776static void run_one_async_done(struct btrfs_work *work)
777{
778 struct btrfs_fs_info *fs_info;
779 struct async_submit_bio *async;
780 int limit;
781
782 async = container_of(work, struct async_submit_bio, work);
783 fs_info = BTRFS_I(async->inode)->root->fs_info;
784
785 limit = btrfs_async_submit_limit(fs_info);
786 limit = limit * 2 / 3;
787
788 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
789 waitqueue_active(&fs_info->async_submit_wait))
790 wake_up(&fs_info->async_submit_wait);
791
792 /* If an error occured we just want to clean up the bio and move on */
793 if (async->error) {
794 bio_endio(async->bio, async->error);
795 return;
796 }
797
798 async->submit_bio_done(async->inode, async->rw, async->bio,
799 async->mirror_num, async->bio_flags,
800 async->bio_offset);
801}
802
803static void run_one_async_free(struct btrfs_work *work)
804{
805 struct async_submit_bio *async;
806
807 async = container_of(work, struct async_submit_bio, work);
808 kfree(async);
809}
810
811int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
812 int rw, struct bio *bio, int mirror_num,
813 unsigned long bio_flags,
814 u64 bio_offset,
815 extent_submit_bio_hook_t *submit_bio_start,
816 extent_submit_bio_hook_t *submit_bio_done)
817{
818 struct async_submit_bio *async;
819
820 async = kmalloc(sizeof(*async), GFP_NOFS);
821 if (!async)
822 return -ENOMEM;
823
824 async->inode = inode;
825 async->rw = rw;
826 async->bio = bio;
827 async->mirror_num = mirror_num;
828 async->submit_bio_start = submit_bio_start;
829 async->submit_bio_done = submit_bio_done;
830
831 btrfs_init_work(&async->work, run_one_async_start,
832 run_one_async_done, run_one_async_free);
833
834 async->bio_flags = bio_flags;
835 async->bio_offset = bio_offset;
836
837 async->error = 0;
838
839 atomic_inc(&fs_info->nr_async_submits);
840
841 if (rw & REQ_SYNC)
842 btrfs_set_work_high_priority(&async->work);
843
844 btrfs_queue_work(fs_info->workers, &async->work);
845
846 while (atomic_read(&fs_info->async_submit_draining) &&
847 atomic_read(&fs_info->nr_async_submits)) {
848 wait_event(fs_info->async_submit_wait,
849 (atomic_read(&fs_info->nr_async_submits) == 0));
850 }
851
852 return 0;
853}
854
855static int btree_csum_one_bio(struct bio *bio)
856{
857 struct bio_vec *bvec;
858 struct btrfs_root *root;
859 int i, ret = 0;
860
861 bio_for_each_segment_all(bvec, bio, i) {
862 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
863 ret = csum_dirty_buffer(root, bvec->bv_page);
864 if (ret)
865 break;
866 }
867
868 return ret;
869}
870
871static int __btree_submit_bio_start(struct inode *inode, int rw,
872 struct bio *bio, int mirror_num,
873 unsigned long bio_flags,
874 u64 bio_offset)
875{
876 /*
877 * when we're called for a write, we're already in the async
878 * submission context. Just jump into btrfs_map_bio
879 */
880 return btree_csum_one_bio(bio);
881}
882
883static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
884 int mirror_num, unsigned long bio_flags,
885 u64 bio_offset)
886{
887 int ret;
888
889 /*
890 * when we're called for a write, we're already in the async
891 * submission context. Just jump into btrfs_map_bio
892 */
893 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
894 if (ret)
895 bio_endio(bio, ret);
896 return ret;
897}
898
899static int check_async_write(struct inode *inode, unsigned long bio_flags)
900{
901 if (bio_flags & EXTENT_BIO_TREE_LOG)
902 return 0;
903#ifdef CONFIG_X86
904 if (cpu_has_xmm4_2)
905 return 0;
906#endif
907 return 1;
908}
909
910static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
911 int mirror_num, unsigned long bio_flags,
912 u64 bio_offset)
913{
914 int async = check_async_write(inode, bio_flags);
915 int ret;
916
917 if (!(rw & REQ_WRITE)) {
918 /*
919 * called for a read, do the setup so that checksum validation
920 * can happen in the async kernel threads
921 */
922 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
923 bio, 1);
924 if (ret)
925 goto out_w_error;
926 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
927 mirror_num, 0);
928 } else if (!async) {
929 ret = btree_csum_one_bio(bio);
930 if (ret)
931 goto out_w_error;
932 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
933 mirror_num, 0);
934 } else {
935 /*
936 * kthread helpers are used to submit writes so that
937 * checksumming can happen in parallel across all CPUs
938 */
939 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
940 inode, rw, bio, mirror_num, 0,
941 bio_offset,
942 __btree_submit_bio_start,
943 __btree_submit_bio_done);
944 }
945
946 if (ret) {
947out_w_error:
948 bio_endio(bio, ret);
949 }
950 return ret;
951}
952
953#ifdef CONFIG_MIGRATION
954static int btree_migratepage(struct address_space *mapping,
955 struct page *newpage, struct page *page,
956 enum migrate_mode mode)
957{
958 /*
959 * we can't safely write a btree page from here,
960 * we haven't done the locking hook
961 */
962 if (PageDirty(page))
963 return -EAGAIN;
964 /*
965 * Buffers may be managed in a filesystem specific way.
966 * We must have no buffers or drop them.
967 */
968 if (page_has_private(page) &&
969 !try_to_release_page(page, GFP_KERNEL))
970 return -EAGAIN;
971 return migrate_page(mapping, newpage, page, mode);
972}
973#endif
974
975
976static int btree_writepages(struct address_space *mapping,
977 struct writeback_control *wbc)
978{
979 struct btrfs_fs_info *fs_info;
980 int ret;
981
982 if (wbc->sync_mode == WB_SYNC_NONE) {
983
984 if (wbc->for_kupdate)
985 return 0;
986
987 fs_info = BTRFS_I(mapping->host)->root->fs_info;
988 /* this is a bit racy, but that's ok */
989 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
990 BTRFS_DIRTY_METADATA_THRESH);
991 if (ret < 0)
992 return 0;
993 }
994 return btree_write_cache_pages(mapping, wbc);
995}
996
997static int btree_readpage(struct file *file, struct page *page)
998{
999 struct extent_io_tree *tree;
1000 tree = &BTRFS_I(page->mapping->host)->io_tree;
1001 return extent_read_full_page(tree, page, btree_get_extent, 0);
1002}
1003
1004static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1005{
1006 if (PageWriteback(page) || PageDirty(page))
1007 return 0;
1008
1009 return try_release_extent_buffer(page);
1010}
1011
1012static void btree_invalidatepage(struct page *page, unsigned int offset,
1013 unsigned int length)
1014{
1015 struct extent_io_tree *tree;
1016 tree = &BTRFS_I(page->mapping->host)->io_tree;
1017 extent_invalidatepage(tree, page, offset);
1018 btree_releasepage(page, GFP_NOFS);
1019 if (PagePrivate(page)) {
1020 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1021 "page private not zero on page %llu",
1022 (unsigned long long)page_offset(page));
1023 ClearPagePrivate(page);
1024 set_page_private(page, 0);
1025 page_cache_release(page);
1026 }
1027}
1028
1029static int btree_set_page_dirty(struct page *page)
1030{
1031#ifdef DEBUG
1032 struct extent_buffer *eb;
1033
1034 BUG_ON(!PagePrivate(page));
1035 eb = (struct extent_buffer *)page->private;
1036 BUG_ON(!eb);
1037 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1038 BUG_ON(!atomic_read(&eb->refs));
1039 btrfs_assert_tree_locked(eb);
1040#endif
1041 return __set_page_dirty_nobuffers(page);
1042}
1043
1044static const struct address_space_operations btree_aops = {
1045 .readpage = btree_readpage,
1046 .writepages = btree_writepages,
1047 .releasepage = btree_releasepage,
1048 .invalidatepage = btree_invalidatepage,
1049#ifdef CONFIG_MIGRATION
1050 .migratepage = btree_migratepage,
1051#endif
1052 .set_page_dirty = btree_set_page_dirty,
1053};
1054
1055int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1056 u64 parent_transid)
1057{
1058 struct extent_buffer *buf = NULL;
1059 struct inode *btree_inode = root->fs_info->btree_inode;
1060 int ret = 0;
1061
1062 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1063 if (!buf)
1064 return 0;
1065 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1066 buf, 0, WAIT_NONE, btree_get_extent, 0);
1067 free_extent_buffer(buf);
1068 return ret;
1069}
1070
1071int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1072 int mirror_num, struct extent_buffer **eb)
1073{
1074 struct extent_buffer *buf = NULL;
1075 struct inode *btree_inode = root->fs_info->btree_inode;
1076 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1077 int ret;
1078
1079 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1080 if (!buf)
1081 return 0;
1082
1083 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1084
1085 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1086 btree_get_extent, mirror_num);
1087 if (ret) {
1088 free_extent_buffer(buf);
1089 return ret;
1090 }
1091
1092 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1093 free_extent_buffer(buf);
1094 return -EIO;
1095 } else if (extent_buffer_uptodate(buf)) {
1096 *eb = buf;
1097 } else {
1098 free_extent_buffer(buf);
1099 }
1100 return 0;
1101}
1102
1103struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1104 u64 bytenr, u32 blocksize)
1105{
1106 return find_extent_buffer(root->fs_info, bytenr);
1107}
1108
1109struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1110 u64 bytenr, u32 blocksize)
1111{
1112 return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
1113}
1114
1115
1116int btrfs_write_tree_block(struct extent_buffer *buf)
1117{
1118 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1119 buf->start + buf->len - 1);
1120}
1121
1122int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1123{
1124 return filemap_fdatawait_range(buf->pages[0]->mapping,
1125 buf->start, buf->start + buf->len - 1);
1126}
1127
1128struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1129 u32 blocksize, u64 parent_transid)
1130{
1131 struct extent_buffer *buf = NULL;
1132 int ret;
1133
1134 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1135 if (!buf)
1136 return NULL;
1137
1138 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1139 if (ret) {
1140 free_extent_buffer(buf);
1141 return NULL;
1142 }
1143 return buf;
1144
1145}
1146
1147void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1148 struct extent_buffer *buf)
1149{
1150 struct btrfs_fs_info *fs_info = root->fs_info;
1151
1152 if (btrfs_header_generation(buf) ==
1153 fs_info->running_transaction->transid) {
1154 btrfs_assert_tree_locked(buf);
1155
1156 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1157 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1158 -buf->len,
1159 fs_info->dirty_metadata_batch);
1160 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1161 btrfs_set_lock_blocking(buf);
1162 clear_extent_buffer_dirty(buf);
1163 }
1164 }
1165}
1166
1167static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1168{
1169 struct btrfs_subvolume_writers *writers;
1170 int ret;
1171
1172 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1173 if (!writers)
1174 return ERR_PTR(-ENOMEM);
1175
1176 ret = percpu_counter_init(&writers->counter, 0);
1177 if (ret < 0) {
1178 kfree(writers);
1179 return ERR_PTR(ret);
1180 }
1181
1182 init_waitqueue_head(&writers->wait);
1183 return writers;
1184}
1185
1186static void
1187btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1188{
1189 percpu_counter_destroy(&writers->counter);
1190 kfree(writers);
1191}
1192
1193static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1194 u32 stripesize, struct btrfs_root *root,
1195 struct btrfs_fs_info *fs_info,
1196 u64 objectid)
1197{
1198 root->node = NULL;
1199 root->commit_root = NULL;
1200 root->sectorsize = sectorsize;
1201 root->nodesize = nodesize;
1202 root->leafsize = leafsize;
1203 root->stripesize = stripesize;
1204 root->ref_cows = 0;
1205 root->track_dirty = 0;
1206 root->in_radix = 0;
1207 root->orphan_item_inserted = 0;
1208 root->orphan_cleanup_state = 0;
1209
1210 root->objectid = objectid;
1211 root->last_trans = 0;
1212 root->highest_objectid = 0;
1213 root->nr_delalloc_inodes = 0;
1214 root->nr_ordered_extents = 0;
1215 root->name = NULL;
1216 root->inode_tree = RB_ROOT;
1217 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1218 root->block_rsv = NULL;
1219 root->orphan_block_rsv = NULL;
1220
1221 INIT_LIST_HEAD(&root->dirty_list);
1222 INIT_LIST_HEAD(&root->root_list);
1223 INIT_LIST_HEAD(&root->delalloc_inodes);
1224 INIT_LIST_HEAD(&root->delalloc_root);
1225 INIT_LIST_HEAD(&root->ordered_extents);
1226 INIT_LIST_HEAD(&root->ordered_root);
1227 INIT_LIST_HEAD(&root->logged_list[0]);
1228 INIT_LIST_HEAD(&root->logged_list[1]);
1229 spin_lock_init(&root->orphan_lock);
1230 spin_lock_init(&root->inode_lock);
1231 spin_lock_init(&root->delalloc_lock);
1232 spin_lock_init(&root->ordered_extent_lock);
1233 spin_lock_init(&root->accounting_lock);
1234 spin_lock_init(&root->log_extents_lock[0]);
1235 spin_lock_init(&root->log_extents_lock[1]);
1236 mutex_init(&root->objectid_mutex);
1237 mutex_init(&root->log_mutex);
1238 mutex_init(&root->ordered_extent_mutex);
1239 mutex_init(&root->delalloc_mutex);
1240 init_waitqueue_head(&root->log_writer_wait);
1241 init_waitqueue_head(&root->log_commit_wait[0]);
1242 init_waitqueue_head(&root->log_commit_wait[1]);
1243 INIT_LIST_HEAD(&root->log_ctxs[0]);
1244 INIT_LIST_HEAD(&root->log_ctxs[1]);
1245 atomic_set(&root->log_commit[0], 0);
1246 atomic_set(&root->log_commit[1], 0);
1247 atomic_set(&root->log_writers, 0);
1248 atomic_set(&root->log_batch, 0);
1249 atomic_set(&root->orphan_inodes, 0);
1250 atomic_set(&root->refs, 1);
1251 atomic_set(&root->will_be_snapshoted, 0);
1252 root->log_transid = 0;
1253 root->log_transid_committed = -1;
1254 root->last_log_commit = 0;
1255 if (fs_info)
1256 extent_io_tree_init(&root->dirty_log_pages,
1257 fs_info->btree_inode->i_mapping);
1258
1259 memset(&root->root_key, 0, sizeof(root->root_key));
1260 memset(&root->root_item, 0, sizeof(root->root_item));
1261 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1262 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1263 if (fs_info)
1264 root->defrag_trans_start = fs_info->generation;
1265 else
1266 root->defrag_trans_start = 0;
1267 init_completion(&root->kobj_unregister);
1268 root->defrag_running = 0;
1269 root->root_key.objectid = objectid;
1270 root->anon_dev = 0;
1271
1272 spin_lock_init(&root->root_item_lock);
1273}
1274
1275static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1276{
1277 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1278 if (root)
1279 root->fs_info = fs_info;
1280 return root;
1281}
1282
1283#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1284/* Should only be used by the testing infrastructure */
1285struct btrfs_root *btrfs_alloc_dummy_root(void)
1286{
1287 struct btrfs_root *root;
1288
1289 root = btrfs_alloc_root(NULL);
1290 if (!root)
1291 return ERR_PTR(-ENOMEM);
1292 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1293 root->dummy_root = 1;
1294
1295 return root;
1296}
1297#endif
1298
1299struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1300 struct btrfs_fs_info *fs_info,
1301 u64 objectid)
1302{
1303 struct extent_buffer *leaf;
1304 struct btrfs_root *tree_root = fs_info->tree_root;
1305 struct btrfs_root *root;
1306 struct btrfs_key key;
1307 int ret = 0;
1308 uuid_le uuid;
1309
1310 root = btrfs_alloc_root(fs_info);
1311 if (!root)
1312 return ERR_PTR(-ENOMEM);
1313
1314 __setup_root(tree_root->nodesize, tree_root->leafsize,
1315 tree_root->sectorsize, tree_root->stripesize,
1316 root, fs_info, objectid);
1317 root->root_key.objectid = objectid;
1318 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1319 root->root_key.offset = 0;
1320
1321 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1322 0, objectid, NULL, 0, 0, 0);
1323 if (IS_ERR(leaf)) {
1324 ret = PTR_ERR(leaf);
1325 leaf = NULL;
1326 goto fail;
1327 }
1328
1329 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1330 btrfs_set_header_bytenr(leaf, leaf->start);
1331 btrfs_set_header_generation(leaf, trans->transid);
1332 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1333 btrfs_set_header_owner(leaf, objectid);
1334 root->node = leaf;
1335
1336 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1337 BTRFS_FSID_SIZE);
1338 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1339 btrfs_header_chunk_tree_uuid(leaf),
1340 BTRFS_UUID_SIZE);
1341 btrfs_mark_buffer_dirty(leaf);
1342
1343 root->commit_root = btrfs_root_node(root);
1344 root->track_dirty = 1;
1345
1346
1347 root->root_item.flags = 0;
1348 root->root_item.byte_limit = 0;
1349 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1350 btrfs_set_root_generation(&root->root_item, trans->transid);
1351 btrfs_set_root_level(&root->root_item, 0);
1352 btrfs_set_root_refs(&root->root_item, 1);
1353 btrfs_set_root_used(&root->root_item, leaf->len);
1354 btrfs_set_root_last_snapshot(&root->root_item, 0);
1355 btrfs_set_root_dirid(&root->root_item, 0);
1356 uuid_le_gen(&uuid);
1357 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1358 root->root_item.drop_level = 0;
1359
1360 key.objectid = objectid;
1361 key.type = BTRFS_ROOT_ITEM_KEY;
1362 key.offset = 0;
1363 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1364 if (ret)
1365 goto fail;
1366
1367 btrfs_tree_unlock(leaf);
1368
1369 return root;
1370
1371fail:
1372 if (leaf) {
1373 btrfs_tree_unlock(leaf);
1374 free_extent_buffer(leaf);
1375 }
1376 kfree(root);
1377
1378 return ERR_PTR(ret);
1379}
1380
1381static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1382 struct btrfs_fs_info *fs_info)
1383{
1384 struct btrfs_root *root;
1385 struct btrfs_root *tree_root = fs_info->tree_root;
1386 struct extent_buffer *leaf;
1387
1388 root = btrfs_alloc_root(fs_info);
1389 if (!root)
1390 return ERR_PTR(-ENOMEM);
1391
1392 __setup_root(tree_root->nodesize, tree_root->leafsize,
1393 tree_root->sectorsize, tree_root->stripesize,
1394 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1395
1396 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1397 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1398 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1399 /*
1400 * log trees do not get reference counted because they go away
1401 * before a real commit is actually done. They do store pointers
1402 * to file data extents, and those reference counts still get
1403 * updated (along with back refs to the log tree).
1404 */
1405 root->ref_cows = 0;
1406
1407 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1408 BTRFS_TREE_LOG_OBJECTID, NULL,
1409 0, 0, 0);
1410 if (IS_ERR(leaf)) {
1411 kfree(root);
1412 return ERR_CAST(leaf);
1413 }
1414
1415 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1416 btrfs_set_header_bytenr(leaf, leaf->start);
1417 btrfs_set_header_generation(leaf, trans->transid);
1418 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1419 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1420 root->node = leaf;
1421
1422 write_extent_buffer(root->node, root->fs_info->fsid,
1423 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1424 btrfs_mark_buffer_dirty(root->node);
1425 btrfs_tree_unlock(root->node);
1426 return root;
1427}
1428
1429int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1430 struct btrfs_fs_info *fs_info)
1431{
1432 struct btrfs_root *log_root;
1433
1434 log_root = alloc_log_tree(trans, fs_info);
1435 if (IS_ERR(log_root))
1436 return PTR_ERR(log_root);
1437 WARN_ON(fs_info->log_root_tree);
1438 fs_info->log_root_tree = log_root;
1439 return 0;
1440}
1441
1442int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1443 struct btrfs_root *root)
1444{
1445 struct btrfs_root *log_root;
1446 struct btrfs_inode_item *inode_item;
1447
1448 log_root = alloc_log_tree(trans, root->fs_info);
1449 if (IS_ERR(log_root))
1450 return PTR_ERR(log_root);
1451
1452 log_root->last_trans = trans->transid;
1453 log_root->root_key.offset = root->root_key.objectid;
1454
1455 inode_item = &log_root->root_item.inode;
1456 btrfs_set_stack_inode_generation(inode_item, 1);
1457 btrfs_set_stack_inode_size(inode_item, 3);
1458 btrfs_set_stack_inode_nlink(inode_item, 1);
1459 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1460 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1461
1462 btrfs_set_root_node(&log_root->root_item, log_root->node);
1463
1464 WARN_ON(root->log_root);
1465 root->log_root = log_root;
1466 root->log_transid = 0;
1467 root->log_transid_committed = -1;
1468 root->last_log_commit = 0;
1469 return 0;
1470}
1471
1472static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1473 struct btrfs_key *key)
1474{
1475 struct btrfs_root *root;
1476 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1477 struct btrfs_path *path;
1478 u64 generation;
1479 u32 blocksize;
1480 int ret;
1481
1482 path = btrfs_alloc_path();
1483 if (!path)
1484 return ERR_PTR(-ENOMEM);
1485
1486 root = btrfs_alloc_root(fs_info);
1487 if (!root) {
1488 ret = -ENOMEM;
1489 goto alloc_fail;
1490 }
1491
1492 __setup_root(tree_root->nodesize, tree_root->leafsize,
1493 tree_root->sectorsize, tree_root->stripesize,
1494 root, fs_info, key->objectid);
1495
1496 ret = btrfs_find_root(tree_root, key, path,
1497 &root->root_item, &root->root_key);
1498 if (ret) {
1499 if (ret > 0)
1500 ret = -ENOENT;
1501 goto find_fail;
1502 }
1503
1504 generation = btrfs_root_generation(&root->root_item);
1505 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1506 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1507 blocksize, generation);
1508 if (!root->node) {
1509 ret = -ENOMEM;
1510 goto find_fail;
1511 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1512 ret = -EIO;
1513 goto read_fail;
1514 }
1515 root->commit_root = btrfs_root_node(root);
1516out:
1517 btrfs_free_path(path);
1518 return root;
1519
1520read_fail:
1521 free_extent_buffer(root->node);
1522find_fail:
1523 kfree(root);
1524alloc_fail:
1525 root = ERR_PTR(ret);
1526 goto out;
1527}
1528
1529struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1530 struct btrfs_key *location)
1531{
1532 struct btrfs_root *root;
1533
1534 root = btrfs_read_tree_root(tree_root, location);
1535 if (IS_ERR(root))
1536 return root;
1537
1538 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1539 root->ref_cows = 1;
1540 btrfs_check_and_init_root_item(&root->root_item);
1541 }
1542
1543 return root;
1544}
1545
1546int btrfs_init_fs_root(struct btrfs_root *root)
1547{
1548 int ret;
1549 struct btrfs_subvolume_writers *writers;
1550
1551 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1552 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1553 GFP_NOFS);
1554 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1555 ret = -ENOMEM;
1556 goto fail;
1557 }
1558
1559 writers = btrfs_alloc_subvolume_writers();
1560 if (IS_ERR(writers)) {
1561 ret = PTR_ERR(writers);
1562 goto fail;
1563 }
1564 root->subv_writers = writers;
1565
1566 btrfs_init_free_ino_ctl(root);
1567 spin_lock_init(&root->cache_lock);
1568 init_waitqueue_head(&root->cache_wait);
1569
1570 ret = get_anon_bdev(&root->anon_dev);
1571 if (ret)
1572 goto free_writers;
1573 return 0;
1574
1575free_writers:
1576 btrfs_free_subvolume_writers(root->subv_writers);
1577fail:
1578 kfree(root->free_ino_ctl);
1579 kfree(root->free_ino_pinned);
1580 return ret;
1581}
1582
1583static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1584 u64 root_id)
1585{
1586 struct btrfs_root *root;
1587
1588 spin_lock(&fs_info->fs_roots_radix_lock);
1589 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1590 (unsigned long)root_id);
1591 spin_unlock(&fs_info->fs_roots_radix_lock);
1592 return root;
1593}
1594
1595int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1596 struct btrfs_root *root)
1597{
1598 int ret;
1599
1600 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1601 if (ret)
1602 return ret;
1603
1604 spin_lock(&fs_info->fs_roots_radix_lock);
1605 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1606 (unsigned long)root->root_key.objectid,
1607 root);
1608 if (ret == 0)
1609 root->in_radix = 1;
1610 spin_unlock(&fs_info->fs_roots_radix_lock);
1611 radix_tree_preload_end();
1612
1613 return ret;
1614}
1615
1616struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1617 struct btrfs_key *location,
1618 bool check_ref)
1619{
1620 struct btrfs_root *root;
1621 int ret;
1622
1623 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1624 return fs_info->tree_root;
1625 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1626 return fs_info->extent_root;
1627 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1628 return fs_info->chunk_root;
1629 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1630 return fs_info->dev_root;
1631 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1632 return fs_info->csum_root;
1633 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1634 return fs_info->quota_root ? fs_info->quota_root :
1635 ERR_PTR(-ENOENT);
1636 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1637 return fs_info->uuid_root ? fs_info->uuid_root :
1638 ERR_PTR(-ENOENT);
1639again:
1640 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1641 if (root) {
1642 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1643 return ERR_PTR(-ENOENT);
1644 return root;
1645 }
1646
1647 root = btrfs_read_fs_root(fs_info->tree_root, location);
1648 if (IS_ERR(root))
1649 return root;
1650
1651 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1652 ret = -ENOENT;
1653 goto fail;
1654 }
1655
1656 ret = btrfs_init_fs_root(root);
1657 if (ret)
1658 goto fail;
1659
1660 ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID,
1661 location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
1662 if (ret < 0)
1663 goto fail;
1664 if (ret == 0)
1665 root->orphan_item_inserted = 1;
1666
1667 ret = btrfs_insert_fs_root(fs_info, root);
1668 if (ret) {
1669 if (ret == -EEXIST) {
1670 free_fs_root(root);
1671 goto again;
1672 }
1673 goto fail;
1674 }
1675 return root;
1676fail:
1677 free_fs_root(root);
1678 return ERR_PTR(ret);
1679}
1680
1681static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1682{
1683 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1684 int ret = 0;
1685 struct btrfs_device *device;
1686 struct backing_dev_info *bdi;
1687
1688 rcu_read_lock();
1689 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1690 if (!device->bdev)
1691 continue;
1692 bdi = blk_get_backing_dev_info(device->bdev);
1693 if (bdi && bdi_congested(bdi, bdi_bits)) {
1694 ret = 1;
1695 break;
1696 }
1697 }
1698 rcu_read_unlock();
1699 return ret;
1700}
1701
1702/*
1703 * If this fails, caller must call bdi_destroy() to get rid of the
1704 * bdi again.
1705 */
1706static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1707{
1708 int err;
1709
1710 bdi->capabilities = BDI_CAP_MAP_COPY;
1711 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1712 if (err)
1713 return err;
1714
1715 bdi->ra_pages = default_backing_dev_info.ra_pages;
1716 bdi->congested_fn = btrfs_congested_fn;
1717 bdi->congested_data = info;
1718 return 0;
1719}
1720
1721/*
1722 * called by the kthread helper functions to finally call the bio end_io
1723 * functions. This is where read checksum verification actually happens
1724 */
1725static void end_workqueue_fn(struct btrfs_work *work)
1726{
1727 struct bio *bio;
1728 struct end_io_wq *end_io_wq;
1729 int error;
1730
1731 end_io_wq = container_of(work, struct end_io_wq, work);
1732 bio = end_io_wq->bio;
1733
1734 error = end_io_wq->error;
1735 bio->bi_private = end_io_wq->private;
1736 bio->bi_end_io = end_io_wq->end_io;
1737 kfree(end_io_wq);
1738 bio_endio_nodec(bio, error);
1739}
1740
1741static int cleaner_kthread(void *arg)
1742{
1743 struct btrfs_root *root = arg;
1744 int again;
1745
1746 do {
1747 again = 0;
1748
1749 /* Make the cleaner go to sleep early. */
1750 if (btrfs_need_cleaner_sleep(root))
1751 goto sleep;
1752
1753 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1754 goto sleep;
1755
1756 /*
1757 * Avoid the problem that we change the status of the fs
1758 * during the above check and trylock.
1759 */
1760 if (btrfs_need_cleaner_sleep(root)) {
1761 mutex_unlock(&root->fs_info->cleaner_mutex);
1762 goto sleep;
1763 }
1764
1765 btrfs_run_delayed_iputs(root);
1766 again = btrfs_clean_one_deleted_snapshot(root);
1767 mutex_unlock(&root->fs_info->cleaner_mutex);
1768
1769 /*
1770 * The defragger has dealt with the R/O remount and umount,
1771 * needn't do anything special here.
1772 */
1773 btrfs_run_defrag_inodes(root->fs_info);
1774sleep:
1775 if (!try_to_freeze() && !again) {
1776 set_current_state(TASK_INTERRUPTIBLE);
1777 if (!kthread_should_stop())
1778 schedule();
1779 __set_current_state(TASK_RUNNING);
1780 }
1781 } while (!kthread_should_stop());
1782 return 0;
1783}
1784
1785static int transaction_kthread(void *arg)
1786{
1787 struct btrfs_root *root = arg;
1788 struct btrfs_trans_handle *trans;
1789 struct btrfs_transaction *cur;
1790 u64 transid;
1791 unsigned long now;
1792 unsigned long delay;
1793 bool cannot_commit;
1794
1795 do {
1796 cannot_commit = false;
1797 delay = HZ * root->fs_info->commit_interval;
1798 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1799
1800 spin_lock(&root->fs_info->trans_lock);
1801 cur = root->fs_info->running_transaction;
1802 if (!cur) {
1803 spin_unlock(&root->fs_info->trans_lock);
1804 goto sleep;
1805 }
1806
1807 now = get_seconds();
1808 if (cur->state < TRANS_STATE_BLOCKED &&
1809 (now < cur->start_time ||
1810 now - cur->start_time < root->fs_info->commit_interval)) {
1811 spin_unlock(&root->fs_info->trans_lock);
1812 delay = HZ * 5;
1813 goto sleep;
1814 }
1815 transid = cur->transid;
1816 spin_unlock(&root->fs_info->trans_lock);
1817
1818 /* If the file system is aborted, this will always fail. */
1819 trans = btrfs_attach_transaction(root);
1820 if (IS_ERR(trans)) {
1821 if (PTR_ERR(trans) != -ENOENT)
1822 cannot_commit = true;
1823 goto sleep;
1824 }
1825 if (transid == trans->transid) {
1826 btrfs_commit_transaction(trans, root);
1827 } else {
1828 btrfs_end_transaction(trans, root);
1829 }
1830sleep:
1831 wake_up_process(root->fs_info->cleaner_kthread);
1832 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1833
1834 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1835 &root->fs_info->fs_state)))
1836 btrfs_cleanup_transaction(root);
1837 if (!try_to_freeze()) {
1838 set_current_state(TASK_INTERRUPTIBLE);
1839 if (!kthread_should_stop() &&
1840 (!btrfs_transaction_blocked(root->fs_info) ||
1841 cannot_commit))
1842 schedule_timeout(delay);
1843 __set_current_state(TASK_RUNNING);
1844 }
1845 } while (!kthread_should_stop());
1846 return 0;
1847}
1848
1849/*
1850 * this will find the highest generation in the array of
1851 * root backups. The index of the highest array is returned,
1852 * or -1 if we can't find anything.
1853 *
1854 * We check to make sure the array is valid by comparing the
1855 * generation of the latest root in the array with the generation
1856 * in the super block. If they don't match we pitch it.
1857 */
1858static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1859{
1860 u64 cur;
1861 int newest_index = -1;
1862 struct btrfs_root_backup *root_backup;
1863 int i;
1864
1865 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1866 root_backup = info->super_copy->super_roots + i;
1867 cur = btrfs_backup_tree_root_gen(root_backup);
1868 if (cur == newest_gen)
1869 newest_index = i;
1870 }
1871
1872 /* check to see if we actually wrapped around */
1873 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1874 root_backup = info->super_copy->super_roots;
1875 cur = btrfs_backup_tree_root_gen(root_backup);
1876 if (cur == newest_gen)
1877 newest_index = 0;
1878 }
1879 return newest_index;
1880}
1881
1882
1883/*
1884 * find the oldest backup so we know where to store new entries
1885 * in the backup array. This will set the backup_root_index
1886 * field in the fs_info struct
1887 */
1888static void find_oldest_super_backup(struct btrfs_fs_info *info,
1889 u64 newest_gen)
1890{
1891 int newest_index = -1;
1892
1893 newest_index = find_newest_super_backup(info, newest_gen);
1894 /* if there was garbage in there, just move along */
1895 if (newest_index == -1) {
1896 info->backup_root_index = 0;
1897 } else {
1898 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1899 }
1900}
1901
1902/*
1903 * copy all the root pointers into the super backup array.
1904 * this will bump the backup pointer by one when it is
1905 * done
1906 */
1907static void backup_super_roots(struct btrfs_fs_info *info)
1908{
1909 int next_backup;
1910 struct btrfs_root_backup *root_backup;
1911 int last_backup;
1912
1913 next_backup = info->backup_root_index;
1914 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1915 BTRFS_NUM_BACKUP_ROOTS;
1916
1917 /*
1918 * just overwrite the last backup if we're at the same generation
1919 * this happens only at umount
1920 */
1921 root_backup = info->super_for_commit->super_roots + last_backup;
1922 if (btrfs_backup_tree_root_gen(root_backup) ==
1923 btrfs_header_generation(info->tree_root->node))
1924 next_backup = last_backup;
1925
1926 root_backup = info->super_for_commit->super_roots + next_backup;
1927
1928 /*
1929 * make sure all of our padding and empty slots get zero filled
1930 * regardless of which ones we use today
1931 */
1932 memset(root_backup, 0, sizeof(*root_backup));
1933
1934 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1935
1936 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1937 btrfs_set_backup_tree_root_gen(root_backup,
1938 btrfs_header_generation(info->tree_root->node));
1939
1940 btrfs_set_backup_tree_root_level(root_backup,
1941 btrfs_header_level(info->tree_root->node));
1942
1943 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1944 btrfs_set_backup_chunk_root_gen(root_backup,
1945 btrfs_header_generation(info->chunk_root->node));
1946 btrfs_set_backup_chunk_root_level(root_backup,
1947 btrfs_header_level(info->chunk_root->node));
1948
1949 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1950 btrfs_set_backup_extent_root_gen(root_backup,
1951 btrfs_header_generation(info->extent_root->node));
1952 btrfs_set_backup_extent_root_level(root_backup,
1953 btrfs_header_level(info->extent_root->node));
1954
1955 /*
1956 * we might commit during log recovery, which happens before we set
1957 * the fs_root. Make sure it is valid before we fill it in.
1958 */
1959 if (info->fs_root && info->fs_root->node) {
1960 btrfs_set_backup_fs_root(root_backup,
1961 info->fs_root->node->start);
1962 btrfs_set_backup_fs_root_gen(root_backup,
1963 btrfs_header_generation(info->fs_root->node));
1964 btrfs_set_backup_fs_root_level(root_backup,
1965 btrfs_header_level(info->fs_root->node));
1966 }
1967
1968 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1969 btrfs_set_backup_dev_root_gen(root_backup,
1970 btrfs_header_generation(info->dev_root->node));
1971 btrfs_set_backup_dev_root_level(root_backup,
1972 btrfs_header_level(info->dev_root->node));
1973
1974 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1975 btrfs_set_backup_csum_root_gen(root_backup,
1976 btrfs_header_generation(info->csum_root->node));
1977 btrfs_set_backup_csum_root_level(root_backup,
1978 btrfs_header_level(info->csum_root->node));
1979
1980 btrfs_set_backup_total_bytes(root_backup,
1981 btrfs_super_total_bytes(info->super_copy));
1982 btrfs_set_backup_bytes_used(root_backup,
1983 btrfs_super_bytes_used(info->super_copy));
1984 btrfs_set_backup_num_devices(root_backup,
1985 btrfs_super_num_devices(info->super_copy));
1986
1987 /*
1988 * if we don't copy this out to the super_copy, it won't get remembered
1989 * for the next commit
1990 */
1991 memcpy(&info->super_copy->super_roots,
1992 &info->super_for_commit->super_roots,
1993 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1994}
1995
1996/*
1997 * this copies info out of the root backup array and back into
1998 * the in-memory super block. It is meant to help iterate through
1999 * the array, so you send it the number of backups you've already
2000 * tried and the last backup index you used.
2001 *
2002 * this returns -1 when it has tried all the backups
2003 */
2004static noinline int next_root_backup(struct btrfs_fs_info *info,
2005 struct btrfs_super_block *super,
2006 int *num_backups_tried, int *backup_index)
2007{
2008 struct btrfs_root_backup *root_backup;
2009 int newest = *backup_index;
2010
2011 if (*num_backups_tried == 0) {
2012 u64 gen = btrfs_super_generation(super);
2013
2014 newest = find_newest_super_backup(info, gen);
2015 if (newest == -1)
2016 return -1;
2017
2018 *backup_index = newest;
2019 *num_backups_tried = 1;
2020 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2021 /* we've tried all the backups, all done */
2022 return -1;
2023 } else {
2024 /* jump to the next oldest backup */
2025 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2026 BTRFS_NUM_BACKUP_ROOTS;
2027 *backup_index = newest;
2028 *num_backups_tried += 1;
2029 }
2030 root_backup = super->super_roots + newest;
2031
2032 btrfs_set_super_generation(super,
2033 btrfs_backup_tree_root_gen(root_backup));
2034 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2035 btrfs_set_super_root_level(super,
2036 btrfs_backup_tree_root_level(root_backup));
2037 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2038
2039 /*
2040 * fixme: the total bytes and num_devices need to match or we should
2041 * need a fsck
2042 */
2043 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2044 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2045 return 0;
2046}
2047
2048/* helper to cleanup workers */
2049static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2050{
2051 btrfs_destroy_workqueue(fs_info->fixup_workers);
2052 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2053 btrfs_destroy_workqueue(fs_info->workers);
2054 btrfs_destroy_workqueue(fs_info->endio_workers);
2055 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2056 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2057 btrfs_destroy_workqueue(fs_info->rmw_workers);
2058 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2059 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2060 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2061 btrfs_destroy_workqueue(fs_info->submit_workers);
2062 btrfs_destroy_workqueue(fs_info->delayed_workers);
2063 btrfs_destroy_workqueue(fs_info->caching_workers);
2064 btrfs_destroy_workqueue(fs_info->readahead_workers);
2065 btrfs_destroy_workqueue(fs_info->flush_workers);
2066 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2067}
2068
2069static void free_root_extent_buffers(struct btrfs_root *root)
2070{
2071 if (root) {
2072 free_extent_buffer(root->node);
2073 free_extent_buffer(root->commit_root);
2074 root->node = NULL;
2075 root->commit_root = NULL;
2076 }
2077}
2078
2079/* helper to cleanup tree roots */
2080static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2081{
2082 free_root_extent_buffers(info->tree_root);
2083
2084 free_root_extent_buffers(info->dev_root);
2085 free_root_extent_buffers(info->extent_root);
2086 free_root_extent_buffers(info->csum_root);
2087 free_root_extent_buffers(info->quota_root);
2088 free_root_extent_buffers(info->uuid_root);
2089 if (chunk_root)
2090 free_root_extent_buffers(info->chunk_root);
2091}
2092
2093static void del_fs_roots(struct btrfs_fs_info *fs_info)
2094{
2095 int ret;
2096 struct btrfs_root *gang[8];
2097 int i;
2098
2099 while (!list_empty(&fs_info->dead_roots)) {
2100 gang[0] = list_entry(fs_info->dead_roots.next,
2101 struct btrfs_root, root_list);
2102 list_del(&gang[0]->root_list);
2103
2104 if (gang[0]->in_radix) {
2105 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2106 } else {
2107 free_extent_buffer(gang[0]->node);
2108 free_extent_buffer(gang[0]->commit_root);
2109 btrfs_put_fs_root(gang[0]);
2110 }
2111 }
2112
2113 while (1) {
2114 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2115 (void **)gang, 0,
2116 ARRAY_SIZE(gang));
2117 if (!ret)
2118 break;
2119 for (i = 0; i < ret; i++)
2120 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2121 }
2122
2123 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2124 btrfs_free_log_root_tree(NULL, fs_info);
2125 btrfs_destroy_pinned_extent(fs_info->tree_root,
2126 fs_info->pinned_extents);
2127 }
2128}
2129
2130int open_ctree(struct super_block *sb,
2131 struct btrfs_fs_devices *fs_devices,
2132 char *options)
2133{
2134 u32 sectorsize;
2135 u32 nodesize;
2136 u32 leafsize;
2137 u32 blocksize;
2138 u32 stripesize;
2139 u64 generation;
2140 u64 features;
2141 struct btrfs_key location;
2142 struct buffer_head *bh;
2143 struct btrfs_super_block *disk_super;
2144 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2145 struct btrfs_root *tree_root;
2146 struct btrfs_root *extent_root;
2147 struct btrfs_root *csum_root;
2148 struct btrfs_root *chunk_root;
2149 struct btrfs_root *dev_root;
2150 struct btrfs_root *quota_root;
2151 struct btrfs_root *uuid_root;
2152 struct btrfs_root *log_tree_root;
2153 int ret;
2154 int err = -EINVAL;
2155 int num_backups_tried = 0;
2156 int backup_index = 0;
2157 int max_active;
2158 int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2159 bool create_uuid_tree;
2160 bool check_uuid_tree;
2161
2162 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2163 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2164 if (!tree_root || !chunk_root) {
2165 err = -ENOMEM;
2166 goto fail;
2167 }
2168
2169 ret = init_srcu_struct(&fs_info->subvol_srcu);
2170 if (ret) {
2171 err = ret;
2172 goto fail;
2173 }
2174
2175 ret = setup_bdi(fs_info, &fs_info->bdi);
2176 if (ret) {
2177 err = ret;
2178 goto fail_srcu;
2179 }
2180
2181 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2182 if (ret) {
2183 err = ret;
2184 goto fail_bdi;
2185 }
2186 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2187 (1 + ilog2(nr_cpu_ids));
2188
2189 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2190 if (ret) {
2191 err = ret;
2192 goto fail_dirty_metadata_bytes;
2193 }
2194
2195 ret = percpu_counter_init(&fs_info->bio_counter, 0);
2196 if (ret) {
2197 err = ret;
2198 goto fail_delalloc_bytes;
2199 }
2200
2201 fs_info->btree_inode = new_inode(sb);
2202 if (!fs_info->btree_inode) {
2203 err = -ENOMEM;
2204 goto fail_bio_counter;
2205 }
2206
2207 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2208
2209 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2210 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2211 INIT_LIST_HEAD(&fs_info->trans_list);
2212 INIT_LIST_HEAD(&fs_info->dead_roots);
2213 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2214 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2215 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2216 spin_lock_init(&fs_info->delalloc_root_lock);
2217 spin_lock_init(&fs_info->trans_lock);
2218 spin_lock_init(&fs_info->fs_roots_radix_lock);
2219 spin_lock_init(&fs_info->delayed_iput_lock);
2220 spin_lock_init(&fs_info->defrag_inodes_lock);
2221 spin_lock_init(&fs_info->free_chunk_lock);
2222 spin_lock_init(&fs_info->tree_mod_seq_lock);
2223 spin_lock_init(&fs_info->super_lock);
2224 spin_lock_init(&fs_info->buffer_lock);
2225 rwlock_init(&fs_info->tree_mod_log_lock);
2226 mutex_init(&fs_info->reloc_mutex);
2227 mutex_init(&fs_info->delalloc_root_mutex);
2228 seqlock_init(&fs_info->profiles_lock);
2229
2230 init_completion(&fs_info->kobj_unregister);
2231 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2232 INIT_LIST_HEAD(&fs_info->space_info);
2233 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2234 btrfs_mapping_init(&fs_info->mapping_tree);
2235 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2236 BTRFS_BLOCK_RSV_GLOBAL);
2237 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2238 BTRFS_BLOCK_RSV_DELALLOC);
2239 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2240 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2241 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2242 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2243 BTRFS_BLOCK_RSV_DELOPS);
2244 atomic_set(&fs_info->nr_async_submits, 0);
2245 atomic_set(&fs_info->async_delalloc_pages, 0);
2246 atomic_set(&fs_info->async_submit_draining, 0);
2247 atomic_set(&fs_info->nr_async_bios, 0);
2248 atomic_set(&fs_info->defrag_running, 0);
2249 atomic64_set(&fs_info->tree_mod_seq, 0);
2250 fs_info->sb = sb;
2251 fs_info->max_inline = 8192 * 1024;
2252 fs_info->metadata_ratio = 0;
2253 fs_info->defrag_inodes = RB_ROOT;
2254 fs_info->free_chunk_space = 0;
2255 fs_info->tree_mod_log = RB_ROOT;
2256 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2257 fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2258 /* readahead state */
2259 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2260 spin_lock_init(&fs_info->reada_lock);
2261
2262 fs_info->thread_pool_size = min_t(unsigned long,
2263 num_online_cpus() + 2, 8);
2264
2265 INIT_LIST_HEAD(&fs_info->ordered_roots);
2266 spin_lock_init(&fs_info->ordered_root_lock);
2267 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2268 GFP_NOFS);
2269 if (!fs_info->delayed_root) {
2270 err = -ENOMEM;
2271 goto fail_iput;
2272 }
2273 btrfs_init_delayed_root(fs_info->delayed_root);
2274
2275 mutex_init(&fs_info->scrub_lock);
2276 atomic_set(&fs_info->scrubs_running, 0);
2277 atomic_set(&fs_info->scrub_pause_req, 0);
2278 atomic_set(&fs_info->scrubs_paused, 0);
2279 atomic_set(&fs_info->scrub_cancel_req, 0);
2280 init_waitqueue_head(&fs_info->replace_wait);
2281 init_waitqueue_head(&fs_info->scrub_pause_wait);
2282 fs_info->scrub_workers_refcnt = 0;
2283#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2284 fs_info->check_integrity_print_mask = 0;
2285#endif
2286
2287 spin_lock_init(&fs_info->balance_lock);
2288 mutex_init(&fs_info->balance_mutex);
2289 atomic_set(&fs_info->balance_running, 0);
2290 atomic_set(&fs_info->balance_pause_req, 0);
2291 atomic_set(&fs_info->balance_cancel_req, 0);
2292 fs_info->balance_ctl = NULL;
2293 init_waitqueue_head(&fs_info->balance_wait_q);
2294
2295 sb->s_blocksize = 4096;
2296 sb->s_blocksize_bits = blksize_bits(4096);
2297 sb->s_bdi = &fs_info->bdi;
2298
2299 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2300 set_nlink(fs_info->btree_inode, 1);
2301 /*
2302 * we set the i_size on the btree inode to the max possible int.
2303 * the real end of the address space is determined by all of
2304 * the devices in the system
2305 */
2306 fs_info->btree_inode->i_size = OFFSET_MAX;
2307 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2308 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2309
2310 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2311 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2312 fs_info->btree_inode->i_mapping);
2313 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2314 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2315
2316 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2317
2318 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2319 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2320 sizeof(struct btrfs_key));
2321 set_bit(BTRFS_INODE_DUMMY,
2322 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2323 btrfs_insert_inode_hash(fs_info->btree_inode);
2324
2325 spin_lock_init(&fs_info->block_group_cache_lock);
2326 fs_info->block_group_cache_tree = RB_ROOT;
2327 fs_info->first_logical_byte = (u64)-1;
2328
2329 extent_io_tree_init(&fs_info->freed_extents[0],
2330 fs_info->btree_inode->i_mapping);
2331 extent_io_tree_init(&fs_info->freed_extents[1],
2332 fs_info->btree_inode->i_mapping);
2333 fs_info->pinned_extents = &fs_info->freed_extents[0];
2334 fs_info->do_barriers = 1;
2335
2336
2337 mutex_init(&fs_info->ordered_operations_mutex);
2338 mutex_init(&fs_info->ordered_extent_flush_mutex);
2339 mutex_init(&fs_info->tree_log_mutex);
2340 mutex_init(&fs_info->chunk_mutex);
2341 mutex_init(&fs_info->transaction_kthread_mutex);
2342 mutex_init(&fs_info->cleaner_mutex);
2343 mutex_init(&fs_info->volume_mutex);
2344 init_rwsem(&fs_info->commit_root_sem);
2345 init_rwsem(&fs_info->cleanup_work_sem);
2346 init_rwsem(&fs_info->subvol_sem);
2347 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2348 fs_info->dev_replace.lock_owner = 0;
2349 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2350 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2351 mutex_init(&fs_info->dev_replace.lock_management_lock);
2352 mutex_init(&fs_info->dev_replace.lock);
2353
2354 spin_lock_init(&fs_info->qgroup_lock);
2355 mutex_init(&fs_info->qgroup_ioctl_lock);
2356 fs_info->qgroup_tree = RB_ROOT;
2357 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2358 fs_info->qgroup_seq = 1;
2359 fs_info->quota_enabled = 0;
2360 fs_info->pending_quota_state = 0;
2361 fs_info->qgroup_ulist = NULL;
2362 mutex_init(&fs_info->qgroup_rescan_lock);
2363
2364 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2365 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2366
2367 init_waitqueue_head(&fs_info->transaction_throttle);
2368 init_waitqueue_head(&fs_info->transaction_wait);
2369 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2370 init_waitqueue_head(&fs_info->async_submit_wait);
2371
2372 ret = btrfs_alloc_stripe_hash_table(fs_info);
2373 if (ret) {
2374 err = ret;
2375 goto fail_alloc;
2376 }
2377
2378 __setup_root(4096, 4096, 4096, 4096, tree_root,
2379 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2380
2381 invalidate_bdev(fs_devices->latest_bdev);
2382
2383 /*
2384 * Read super block and check the signature bytes only
2385 */
2386 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2387 if (!bh) {
2388 err = -EINVAL;
2389 goto fail_alloc;
2390 }
2391
2392 /*
2393 * We want to check superblock checksum, the type is stored inside.
2394 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2395 */
2396 if (btrfs_check_super_csum(bh->b_data)) {
2397 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2398 err = -EINVAL;
2399 goto fail_alloc;
2400 }
2401
2402 /*
2403 * super_copy is zeroed at allocation time and we never touch the
2404 * following bytes up to INFO_SIZE, the checksum is calculated from
2405 * the whole block of INFO_SIZE
2406 */
2407 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2408 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2409 sizeof(*fs_info->super_for_commit));
2410 brelse(bh);
2411
2412 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2413
2414 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2415 if (ret) {
2416 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2417 err = -EINVAL;
2418 goto fail_alloc;
2419 }
2420
2421 disk_super = fs_info->super_copy;
2422 if (!btrfs_super_root(disk_super))
2423 goto fail_alloc;
2424
2425 /* check FS state, whether FS is broken. */
2426 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2427 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2428
2429 /*
2430 * run through our array of backup supers and setup
2431 * our ring pointer to the oldest one
2432 */
2433 generation = btrfs_super_generation(disk_super);
2434 find_oldest_super_backup(fs_info, generation);
2435
2436 /*
2437 * In the long term, we'll store the compression type in the super
2438 * block, and it'll be used for per file compression control.
2439 */
2440 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2441
2442 ret = btrfs_parse_options(tree_root, options);
2443 if (ret) {
2444 err = ret;
2445 goto fail_alloc;
2446 }
2447
2448 features = btrfs_super_incompat_flags(disk_super) &
2449 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2450 if (features) {
2451 printk(KERN_ERR "BTRFS: couldn't mount because of "
2452 "unsupported optional features (%Lx).\n",
2453 features);
2454 err = -EINVAL;
2455 goto fail_alloc;
2456 }
2457
2458 if (btrfs_super_leafsize(disk_super) !=
2459 btrfs_super_nodesize(disk_super)) {
2460 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2461 "blocksizes don't match. node %d leaf %d\n",
2462 btrfs_super_nodesize(disk_super),
2463 btrfs_super_leafsize(disk_super));
2464 err = -EINVAL;
2465 goto fail_alloc;
2466 }
2467 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2468 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2469 "blocksize (%d) was too large\n",
2470 btrfs_super_leafsize(disk_super));
2471 err = -EINVAL;
2472 goto fail_alloc;
2473 }
2474
2475 features = btrfs_super_incompat_flags(disk_super);
2476 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2477 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2478 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2479
2480 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2481 printk(KERN_ERR "BTRFS: has skinny extents\n");
2482
2483 /*
2484 * flag our filesystem as having big metadata blocks if
2485 * they are bigger than the page size
2486 */
2487 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2488 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2489 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2490 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2491 }
2492
2493 nodesize = btrfs_super_nodesize(disk_super);
2494 leafsize = btrfs_super_leafsize(disk_super);
2495 sectorsize = btrfs_super_sectorsize(disk_super);
2496 stripesize = btrfs_super_stripesize(disk_super);
2497 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2498 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2499
2500 /*
2501 * mixed block groups end up with duplicate but slightly offset
2502 * extent buffers for the same range. It leads to corruptions
2503 */
2504 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2505 (sectorsize != leafsize)) {
2506 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
2507 "are not allowed for mixed block groups on %s\n",
2508 sb->s_id);
2509 goto fail_alloc;
2510 }
2511
2512 /*
2513 * Needn't use the lock because there is no other task which will
2514 * update the flag.
2515 */
2516 btrfs_set_super_incompat_flags(disk_super, features);
2517
2518 features = btrfs_super_compat_ro_flags(disk_super) &
2519 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2520 if (!(sb->s_flags & MS_RDONLY) && features) {
2521 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2522 "unsupported option features (%Lx).\n",
2523 features);
2524 err = -EINVAL;
2525 goto fail_alloc;
2526 }
2527
2528 max_active = fs_info->thread_pool_size;
2529
2530 fs_info->workers =
2531 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2532 max_active, 16);
2533
2534 fs_info->delalloc_workers =
2535 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2536
2537 fs_info->flush_workers =
2538 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2539
2540 fs_info->caching_workers =
2541 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2542
2543 /*
2544 * a higher idle thresh on the submit workers makes it much more
2545 * likely that bios will be send down in a sane order to the
2546 * devices
2547 */
2548 fs_info->submit_workers =
2549 btrfs_alloc_workqueue("submit", flags,
2550 min_t(u64, fs_devices->num_devices,
2551 max_active), 64);
2552
2553 fs_info->fixup_workers =
2554 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2555
2556 /*
2557 * endios are largely parallel and should have a very
2558 * low idle thresh
2559 */
2560 fs_info->endio_workers =
2561 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2562 fs_info->endio_meta_workers =
2563 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2564 fs_info->endio_meta_write_workers =
2565 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2566 fs_info->endio_raid56_workers =
2567 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2568 fs_info->rmw_workers =
2569 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2570 fs_info->endio_write_workers =
2571 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2572 fs_info->endio_freespace_worker =
2573 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2574 fs_info->delayed_workers =
2575 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2576 fs_info->readahead_workers =
2577 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2578 fs_info->qgroup_rescan_workers =
2579 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2580
2581 if (!(fs_info->workers && fs_info->delalloc_workers &&
2582 fs_info->submit_workers && fs_info->flush_workers &&
2583 fs_info->endio_workers && fs_info->endio_meta_workers &&
2584 fs_info->endio_meta_write_workers &&
2585 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2586 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2587 fs_info->caching_workers && fs_info->readahead_workers &&
2588 fs_info->fixup_workers && fs_info->delayed_workers &&
2589 fs_info->qgroup_rescan_workers)) {
2590 err = -ENOMEM;
2591 goto fail_sb_buffer;
2592 }
2593
2594 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2595 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2596 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2597
2598 tree_root->nodesize = nodesize;
2599 tree_root->leafsize = leafsize;
2600 tree_root->sectorsize = sectorsize;
2601 tree_root->stripesize = stripesize;
2602
2603 sb->s_blocksize = sectorsize;
2604 sb->s_blocksize_bits = blksize_bits(sectorsize);
2605
2606 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2607 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
2608 goto fail_sb_buffer;
2609 }
2610
2611 if (sectorsize != PAGE_SIZE) {
2612 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
2613 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2614 goto fail_sb_buffer;
2615 }
2616
2617 mutex_lock(&fs_info->chunk_mutex);
2618 ret = btrfs_read_sys_array(tree_root);
2619 mutex_unlock(&fs_info->chunk_mutex);
2620 if (ret) {
2621 printk(KERN_WARNING "BTRFS: failed to read the system "
2622 "array on %s\n", sb->s_id);
2623 goto fail_sb_buffer;
2624 }
2625
2626 blocksize = btrfs_level_size(tree_root,
2627 btrfs_super_chunk_root_level(disk_super));
2628 generation = btrfs_super_chunk_root_generation(disk_super);
2629
2630 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2631 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2632
2633 chunk_root->node = read_tree_block(chunk_root,
2634 btrfs_super_chunk_root(disk_super),
2635 blocksize, generation);
2636 if (!chunk_root->node ||
2637 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2638 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
2639 sb->s_id);
2640 goto fail_tree_roots;
2641 }
2642 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2643 chunk_root->commit_root = btrfs_root_node(chunk_root);
2644
2645 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2646 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2647
2648 ret = btrfs_read_chunk_tree(chunk_root);
2649 if (ret) {
2650 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
2651 sb->s_id);
2652 goto fail_tree_roots;
2653 }
2654
2655 /*
2656 * keep the device that is marked to be the target device for the
2657 * dev_replace procedure
2658 */
2659 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2660
2661 if (!fs_devices->latest_bdev) {
2662 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
2663 sb->s_id);
2664 goto fail_tree_roots;
2665 }
2666
2667retry_root_backup:
2668 blocksize = btrfs_level_size(tree_root,
2669 btrfs_super_root_level(disk_super));
2670 generation = btrfs_super_generation(disk_super);
2671
2672 tree_root->node = read_tree_block(tree_root,
2673 btrfs_super_root(disk_super),
2674 blocksize, generation);
2675 if (!tree_root->node ||
2676 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2677 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2678 sb->s_id);
2679
2680 goto recovery_tree_root;
2681 }
2682
2683 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2684 tree_root->commit_root = btrfs_root_node(tree_root);
2685 btrfs_set_root_refs(&tree_root->root_item, 1);
2686
2687 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2688 location.type = BTRFS_ROOT_ITEM_KEY;
2689 location.offset = 0;
2690
2691 extent_root = btrfs_read_tree_root(tree_root, &location);
2692 if (IS_ERR(extent_root)) {
2693 ret = PTR_ERR(extent_root);
2694 goto recovery_tree_root;
2695 }
2696 extent_root->track_dirty = 1;
2697 fs_info->extent_root = extent_root;
2698
2699 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2700 dev_root = btrfs_read_tree_root(tree_root, &location);
2701 if (IS_ERR(dev_root)) {
2702 ret = PTR_ERR(dev_root);
2703 goto recovery_tree_root;
2704 }
2705 dev_root->track_dirty = 1;
2706 fs_info->dev_root = dev_root;
2707 btrfs_init_devices_late(fs_info);
2708
2709 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2710 csum_root = btrfs_read_tree_root(tree_root, &location);
2711 if (IS_ERR(csum_root)) {
2712 ret = PTR_ERR(csum_root);
2713 goto recovery_tree_root;
2714 }
2715 csum_root->track_dirty = 1;
2716 fs_info->csum_root = csum_root;
2717
2718 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2719 quota_root = btrfs_read_tree_root(tree_root, &location);
2720 if (!IS_ERR(quota_root)) {
2721 quota_root->track_dirty = 1;
2722 fs_info->quota_enabled = 1;
2723 fs_info->pending_quota_state = 1;
2724 fs_info->quota_root = quota_root;
2725 }
2726
2727 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2728 uuid_root = btrfs_read_tree_root(tree_root, &location);
2729 if (IS_ERR(uuid_root)) {
2730 ret = PTR_ERR(uuid_root);
2731 if (ret != -ENOENT)
2732 goto recovery_tree_root;
2733 create_uuid_tree = true;
2734 check_uuid_tree = false;
2735 } else {
2736 uuid_root->track_dirty = 1;
2737 fs_info->uuid_root = uuid_root;
2738 create_uuid_tree = false;
2739 check_uuid_tree =
2740 generation != btrfs_super_uuid_tree_generation(disk_super);
2741 }
2742
2743 fs_info->generation = generation;
2744 fs_info->last_trans_committed = generation;
2745
2746 ret = btrfs_recover_balance(fs_info);
2747 if (ret) {
2748 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
2749 goto fail_block_groups;
2750 }
2751
2752 ret = btrfs_init_dev_stats(fs_info);
2753 if (ret) {
2754 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2755 ret);
2756 goto fail_block_groups;
2757 }
2758
2759 ret = btrfs_init_dev_replace(fs_info);
2760 if (ret) {
2761 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2762 goto fail_block_groups;
2763 }
2764
2765 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2766
2767 ret = btrfs_sysfs_add_one(fs_info);
2768 if (ret) {
2769 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2770 goto fail_block_groups;
2771 }
2772
2773 ret = btrfs_init_space_info(fs_info);
2774 if (ret) {
2775 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2776 goto fail_sysfs;
2777 }
2778
2779 ret = btrfs_read_block_groups(extent_root);
2780 if (ret) {
2781 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2782 goto fail_sysfs;
2783 }
2784 fs_info->num_tolerated_disk_barrier_failures =
2785 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2786 if (fs_info->fs_devices->missing_devices >
2787 fs_info->num_tolerated_disk_barrier_failures &&
2788 !(sb->s_flags & MS_RDONLY)) {
2789 printk(KERN_WARNING "BTRFS: "
2790 "too many missing devices, writeable mount is not allowed\n");
2791 goto fail_sysfs;
2792 }
2793
2794 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2795 "btrfs-cleaner");
2796 if (IS_ERR(fs_info->cleaner_kthread))
2797 goto fail_sysfs;
2798
2799 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2800 tree_root,
2801 "btrfs-transaction");
2802 if (IS_ERR(fs_info->transaction_kthread))
2803 goto fail_cleaner;
2804
2805 if (!btrfs_test_opt(tree_root, SSD) &&
2806 !btrfs_test_opt(tree_root, NOSSD) &&
2807 !fs_info->fs_devices->rotating) {
2808 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2809 "mode\n");
2810 btrfs_set_opt(fs_info->mount_opt, SSD);
2811 }
2812
2813 /* Set the real inode map cache flag */
2814 if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE))
2815 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE);
2816
2817#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2818 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2819 ret = btrfsic_mount(tree_root, fs_devices,
2820 btrfs_test_opt(tree_root,
2821 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2822 1 : 0,
2823 fs_info->check_integrity_print_mask);
2824 if (ret)
2825 printk(KERN_WARNING "BTRFS: failed to initialize"
2826 " integrity check module %s\n", sb->s_id);
2827 }
2828#endif
2829 ret = btrfs_read_qgroup_config(fs_info);
2830 if (ret)
2831 goto fail_trans_kthread;
2832
2833 /* do not make disk changes in broken FS */
2834 if (btrfs_super_log_root(disk_super) != 0) {
2835 u64 bytenr = btrfs_super_log_root(disk_super);
2836
2837 if (fs_devices->rw_devices == 0) {
2838 printk(KERN_WARNING "BTRFS: log replay required "
2839 "on RO media\n");
2840 err = -EIO;
2841 goto fail_qgroup;
2842 }
2843 blocksize =
2844 btrfs_level_size(tree_root,
2845 btrfs_super_log_root_level(disk_super));
2846
2847 log_tree_root = btrfs_alloc_root(fs_info);
2848 if (!log_tree_root) {
2849 err = -ENOMEM;
2850 goto fail_qgroup;
2851 }
2852
2853 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2854 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2855
2856 log_tree_root->node = read_tree_block(tree_root, bytenr,
2857 blocksize,
2858 generation + 1);
2859 if (!log_tree_root->node ||
2860 !extent_buffer_uptodate(log_tree_root->node)) {
2861 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2862 free_extent_buffer(log_tree_root->node);
2863 kfree(log_tree_root);
2864 goto fail_qgroup;
2865 }
2866 /* returns with log_tree_root freed on success */
2867 ret = btrfs_recover_log_trees(log_tree_root);
2868 if (ret) {
2869 btrfs_error(tree_root->fs_info, ret,
2870 "Failed to recover log tree");
2871 free_extent_buffer(log_tree_root->node);
2872 kfree(log_tree_root);
2873 goto fail_qgroup;
2874 }
2875
2876 if (sb->s_flags & MS_RDONLY) {
2877 ret = btrfs_commit_super(tree_root);
2878 if (ret)
2879 goto fail_qgroup;
2880 }
2881 }
2882
2883 ret = btrfs_find_orphan_roots(tree_root);
2884 if (ret)
2885 goto fail_qgroup;
2886
2887 if (!(sb->s_flags & MS_RDONLY)) {
2888 ret = btrfs_cleanup_fs_roots(fs_info);
2889 if (ret)
2890 goto fail_qgroup;
2891
2892 ret = btrfs_recover_relocation(tree_root);
2893 if (ret < 0) {
2894 printk(KERN_WARNING
2895 "BTRFS: failed to recover relocation\n");
2896 err = -EINVAL;
2897 goto fail_qgroup;
2898 }
2899 }
2900
2901 location.objectid = BTRFS_FS_TREE_OBJECTID;
2902 location.type = BTRFS_ROOT_ITEM_KEY;
2903 location.offset = 0;
2904
2905 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2906 if (IS_ERR(fs_info->fs_root)) {
2907 err = PTR_ERR(fs_info->fs_root);
2908 goto fail_qgroup;
2909 }
2910
2911 if (sb->s_flags & MS_RDONLY)
2912 return 0;
2913
2914 down_read(&fs_info->cleanup_work_sem);
2915 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2916 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2917 up_read(&fs_info->cleanup_work_sem);
2918 close_ctree(tree_root);
2919 return ret;
2920 }
2921 up_read(&fs_info->cleanup_work_sem);
2922
2923 ret = btrfs_resume_balance_async(fs_info);
2924 if (ret) {
2925 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2926 close_ctree(tree_root);
2927 return ret;
2928 }
2929
2930 ret = btrfs_resume_dev_replace_async(fs_info);
2931 if (ret) {
2932 pr_warn("BTRFS: failed to resume dev_replace\n");
2933 close_ctree(tree_root);
2934 return ret;
2935 }
2936
2937 btrfs_qgroup_rescan_resume(fs_info);
2938
2939 if (create_uuid_tree) {
2940 pr_info("BTRFS: creating UUID tree\n");
2941 ret = btrfs_create_uuid_tree(fs_info);
2942 if (ret) {
2943 pr_warn("BTRFS: failed to create the UUID tree %d\n",
2944 ret);
2945 close_ctree(tree_root);
2946 return ret;
2947 }
2948 } else if (check_uuid_tree ||
2949 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2950 pr_info("BTRFS: checking UUID tree\n");
2951 ret = btrfs_check_uuid_tree(fs_info);
2952 if (ret) {
2953 pr_warn("BTRFS: failed to check the UUID tree %d\n",
2954 ret);
2955 close_ctree(tree_root);
2956 return ret;
2957 }
2958 } else {
2959 fs_info->update_uuid_tree_gen = 1;
2960 }
2961
2962 return 0;
2963
2964fail_qgroup:
2965 btrfs_free_qgroup_config(fs_info);
2966fail_trans_kthread:
2967 kthread_stop(fs_info->transaction_kthread);
2968 btrfs_cleanup_transaction(fs_info->tree_root);
2969 del_fs_roots(fs_info);
2970fail_cleaner:
2971 kthread_stop(fs_info->cleaner_kthread);
2972
2973 /*
2974 * make sure we're done with the btree inode before we stop our
2975 * kthreads
2976 */
2977 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2978
2979fail_sysfs:
2980 btrfs_sysfs_remove_one(fs_info);
2981
2982fail_block_groups:
2983 btrfs_put_block_group_cache(fs_info);
2984 btrfs_free_block_groups(fs_info);
2985
2986fail_tree_roots:
2987 free_root_pointers(fs_info, 1);
2988 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2989
2990fail_sb_buffer:
2991 btrfs_stop_all_workers(fs_info);
2992fail_alloc:
2993fail_iput:
2994 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2995
2996 iput(fs_info->btree_inode);
2997fail_bio_counter:
2998 percpu_counter_destroy(&fs_info->bio_counter);
2999fail_delalloc_bytes:
3000 percpu_counter_destroy(&fs_info->delalloc_bytes);
3001fail_dirty_metadata_bytes:
3002 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3003fail_bdi:
3004 bdi_destroy(&fs_info->bdi);
3005fail_srcu:
3006 cleanup_srcu_struct(&fs_info->subvol_srcu);
3007fail:
3008 btrfs_free_stripe_hash_table(fs_info);
3009 btrfs_close_devices(fs_info->fs_devices);
3010 return err;
3011
3012recovery_tree_root:
3013 if (!btrfs_test_opt(tree_root, RECOVERY))
3014 goto fail_tree_roots;
3015
3016 free_root_pointers(fs_info, 0);
3017
3018 /* don't use the log in recovery mode, it won't be valid */
3019 btrfs_set_super_log_root(disk_super, 0);
3020
3021 /* we can't trust the free space cache either */
3022 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3023
3024 ret = next_root_backup(fs_info, fs_info->super_copy,
3025 &num_backups_tried, &backup_index);
3026 if (ret == -1)
3027 goto fail_block_groups;
3028 goto retry_root_backup;
3029}
3030
3031static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3032{
3033 if (uptodate) {
3034 set_buffer_uptodate(bh);
3035 } else {
3036 struct btrfs_device *device = (struct btrfs_device *)
3037 bh->b_private;
3038
3039 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3040 "I/O error on %s\n",
3041 rcu_str_deref(device->name));
3042 /* note, we dont' set_buffer_write_io_error because we have
3043 * our own ways of dealing with the IO errors
3044 */
3045 clear_buffer_uptodate(bh);
3046 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3047 }
3048 unlock_buffer(bh);
3049 put_bh(bh);
3050}
3051
3052struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3053{
3054 struct buffer_head *bh;
3055 struct buffer_head *latest = NULL;
3056 struct btrfs_super_block *super;
3057 int i;
3058 u64 transid = 0;
3059 u64 bytenr;
3060
3061 /* we would like to check all the supers, but that would make
3062 * a btrfs mount succeed after a mkfs from a different FS.
3063 * So, we need to add a special mount option to scan for
3064 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3065 */
3066 for (i = 0; i < 1; i++) {
3067 bytenr = btrfs_sb_offset(i);
3068 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3069 i_size_read(bdev->bd_inode))
3070 break;
3071 bh = __bread(bdev, bytenr / 4096,
3072 BTRFS_SUPER_INFO_SIZE);
3073 if (!bh)
3074 continue;
3075
3076 super = (struct btrfs_super_block *)bh->b_data;
3077 if (btrfs_super_bytenr(super) != bytenr ||
3078 btrfs_super_magic(super) != BTRFS_MAGIC) {
3079 brelse(bh);
3080 continue;
3081 }
3082
3083 if (!latest || btrfs_super_generation(super) > transid) {
3084 brelse(latest);
3085 latest = bh;
3086 transid = btrfs_super_generation(super);
3087 } else {
3088 brelse(bh);
3089 }
3090 }
3091 return latest;
3092}
3093
3094/*
3095 * this should be called twice, once with wait == 0 and
3096 * once with wait == 1. When wait == 0 is done, all the buffer heads
3097 * we write are pinned.
3098 *
3099 * They are released when wait == 1 is done.
3100 * max_mirrors must be the same for both runs, and it indicates how
3101 * many supers on this one device should be written.
3102 *
3103 * max_mirrors == 0 means to write them all.
3104 */
3105static int write_dev_supers(struct btrfs_device *device,
3106 struct btrfs_super_block *sb,
3107 int do_barriers, int wait, int max_mirrors)
3108{
3109 struct buffer_head *bh;
3110 int i;
3111 int ret;
3112 int errors = 0;
3113 u32 crc;
3114 u64 bytenr;
3115
3116 if (max_mirrors == 0)
3117 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3118
3119 for (i = 0; i < max_mirrors; i++) {
3120 bytenr = btrfs_sb_offset(i);
3121 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3122 break;
3123
3124 if (wait) {
3125 bh = __find_get_block(device->bdev, bytenr / 4096,
3126 BTRFS_SUPER_INFO_SIZE);
3127 if (!bh) {
3128 errors++;
3129 continue;
3130 }
3131 wait_on_buffer(bh);
3132 if (!buffer_uptodate(bh))
3133 errors++;
3134
3135 /* drop our reference */
3136 brelse(bh);
3137
3138 /* drop the reference from the wait == 0 run */
3139 brelse(bh);
3140 continue;
3141 } else {
3142 btrfs_set_super_bytenr(sb, bytenr);
3143
3144 crc = ~(u32)0;
3145 crc = btrfs_csum_data((char *)sb +
3146 BTRFS_CSUM_SIZE, crc,
3147 BTRFS_SUPER_INFO_SIZE -
3148 BTRFS_CSUM_SIZE);
3149 btrfs_csum_final(crc, sb->csum);
3150
3151 /*
3152 * one reference for us, and we leave it for the
3153 * caller
3154 */
3155 bh = __getblk(device->bdev, bytenr / 4096,
3156 BTRFS_SUPER_INFO_SIZE);
3157 if (!bh) {
3158 printk(KERN_ERR "BTRFS: couldn't get super "
3159 "buffer head for bytenr %Lu\n", bytenr);
3160 errors++;
3161 continue;
3162 }
3163
3164 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3165
3166 /* one reference for submit_bh */
3167 get_bh(bh);
3168
3169 set_buffer_uptodate(bh);
3170 lock_buffer(bh);
3171 bh->b_end_io = btrfs_end_buffer_write_sync;
3172 bh->b_private = device;
3173 }
3174
3175 /*
3176 * we fua the first super. The others we allow
3177 * to go down lazy.
3178 */
3179 if (i == 0)
3180 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3181 else
3182 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3183 if (ret)
3184 errors++;
3185 }
3186 return errors < i ? 0 : -1;
3187}
3188
3189/*
3190 * endio for the write_dev_flush, this will wake anyone waiting
3191 * for the barrier when it is done
3192 */
3193static void btrfs_end_empty_barrier(struct bio *bio, int err)
3194{
3195 if (err) {
3196 if (err == -EOPNOTSUPP)
3197 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3198 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3199 }
3200 if (bio->bi_private)
3201 complete(bio->bi_private);
3202 bio_put(bio);
3203}
3204
3205/*
3206 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3207 * sent down. With wait == 1, it waits for the previous flush.
3208 *
3209 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3210 * capable
3211 */
3212static int write_dev_flush(struct btrfs_device *device, int wait)
3213{
3214 struct bio *bio;
3215 int ret = 0;
3216
3217 if (device->nobarriers)
3218 return 0;
3219
3220 if (wait) {
3221 bio = device->flush_bio;
3222 if (!bio)
3223 return 0;
3224
3225 wait_for_completion(&device->flush_wait);
3226
3227 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3228 printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3229 rcu_str_deref(device->name));
3230 device->nobarriers = 1;
3231 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3232 ret = -EIO;
3233 btrfs_dev_stat_inc_and_print(device,
3234 BTRFS_DEV_STAT_FLUSH_ERRS);
3235 }
3236
3237 /* drop the reference from the wait == 0 run */
3238 bio_put(bio);
3239 device->flush_bio = NULL;
3240
3241 return ret;
3242 }
3243
3244 /*
3245 * one reference for us, and we leave it for the
3246 * caller
3247 */
3248 device->flush_bio = NULL;
3249 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3250 if (!bio)
3251 return -ENOMEM;
3252
3253 bio->bi_end_io = btrfs_end_empty_barrier;
3254 bio->bi_bdev = device->bdev;
3255 init_completion(&device->flush_wait);
3256 bio->bi_private = &device->flush_wait;
3257 device->flush_bio = bio;
3258
3259 bio_get(bio);
3260 btrfsic_submit_bio(WRITE_FLUSH, bio);
3261
3262 return 0;
3263}
3264
3265/*
3266 * send an empty flush down to each device in parallel,
3267 * then wait for them
3268 */
3269static int barrier_all_devices(struct btrfs_fs_info *info)
3270{
3271 struct list_head *head;
3272 struct btrfs_device *dev;
3273 int errors_send = 0;
3274 int errors_wait = 0;
3275 int ret;
3276
3277 /* send down all the barriers */
3278 head = &info->fs_devices->devices;
3279 list_for_each_entry_rcu(dev, head, dev_list) {
3280 if (dev->missing)
3281 continue;
3282 if (!dev->bdev) {
3283 errors_send++;
3284 continue;
3285 }
3286 if (!dev->in_fs_metadata || !dev->writeable)
3287 continue;
3288
3289 ret = write_dev_flush(dev, 0);
3290 if (ret)
3291 errors_send++;
3292 }
3293
3294 /* wait for all the barriers */
3295 list_for_each_entry_rcu(dev, head, dev_list) {
3296 if (dev->missing)
3297 continue;
3298 if (!dev->bdev) {
3299 errors_wait++;
3300 continue;
3301 }
3302 if (!dev->in_fs_metadata || !dev->writeable)
3303 continue;
3304
3305 ret = write_dev_flush(dev, 1);
3306 if (ret)
3307 errors_wait++;
3308 }
3309 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3310 errors_wait > info->num_tolerated_disk_barrier_failures)
3311 return -EIO;
3312 return 0;
3313}
3314
3315int btrfs_calc_num_tolerated_disk_barrier_failures(
3316 struct btrfs_fs_info *fs_info)
3317{
3318 struct btrfs_ioctl_space_info space;
3319 struct btrfs_space_info *sinfo;
3320 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3321 BTRFS_BLOCK_GROUP_SYSTEM,
3322 BTRFS_BLOCK_GROUP_METADATA,
3323 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3324 int num_types = 4;
3325 int i;
3326 int c;
3327 int num_tolerated_disk_barrier_failures =
3328 (int)fs_info->fs_devices->num_devices;
3329
3330 for (i = 0; i < num_types; i++) {
3331 struct btrfs_space_info *tmp;
3332
3333 sinfo = NULL;
3334 rcu_read_lock();
3335 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3336 if (tmp->flags == types[i]) {
3337 sinfo = tmp;
3338 break;
3339 }
3340 }
3341 rcu_read_unlock();
3342
3343 if (!sinfo)
3344 continue;
3345
3346 down_read(&sinfo->groups_sem);
3347 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3348 if (!list_empty(&sinfo->block_groups[c])) {
3349 u64 flags;
3350
3351 btrfs_get_block_group_info(
3352 &sinfo->block_groups[c], &space);
3353 if (space.total_bytes == 0 ||
3354 space.used_bytes == 0)
3355 continue;
3356 flags = space.flags;
3357 /*
3358 * return
3359 * 0: if dup, single or RAID0 is configured for
3360 * any of metadata, system or data, else
3361 * 1: if RAID5 is configured, or if RAID1 or
3362 * RAID10 is configured and only two mirrors
3363 * are used, else
3364 * 2: if RAID6 is configured, else
3365 * num_mirrors - 1: if RAID1 or RAID10 is
3366 * configured and more than
3367 * 2 mirrors are used.
3368 */
3369 if (num_tolerated_disk_barrier_failures > 0 &&
3370 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3371 BTRFS_BLOCK_GROUP_RAID0)) ||
3372 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3373 == 0)))
3374 num_tolerated_disk_barrier_failures = 0;
3375 else if (num_tolerated_disk_barrier_failures > 1) {
3376 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3377 BTRFS_BLOCK_GROUP_RAID5 |
3378 BTRFS_BLOCK_GROUP_RAID10)) {
3379 num_tolerated_disk_barrier_failures = 1;
3380 } else if (flags &
3381 BTRFS_BLOCK_GROUP_RAID6) {
3382 num_tolerated_disk_barrier_failures = 2;
3383 }
3384 }
3385 }
3386 }
3387 up_read(&sinfo->groups_sem);
3388 }
3389
3390 return num_tolerated_disk_barrier_failures;
3391}
3392
3393static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3394{
3395 struct list_head *head;
3396 struct btrfs_device *dev;
3397 struct btrfs_super_block *sb;
3398 struct btrfs_dev_item *dev_item;
3399 int ret;
3400 int do_barriers;
3401 int max_errors;
3402 int total_errors = 0;
3403 u64 flags;
3404
3405 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3406 backup_super_roots(root->fs_info);
3407
3408 sb = root->fs_info->super_for_commit;
3409 dev_item = &sb->dev_item;
3410
3411 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3412 head = &root->fs_info->fs_devices->devices;
3413 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3414
3415 if (do_barriers) {
3416 ret = barrier_all_devices(root->fs_info);
3417 if (ret) {
3418 mutex_unlock(
3419 &root->fs_info->fs_devices->device_list_mutex);
3420 btrfs_error(root->fs_info, ret,
3421 "errors while submitting device barriers.");
3422 return ret;
3423 }
3424 }
3425
3426 list_for_each_entry_rcu(dev, head, dev_list) {
3427 if (!dev->bdev) {
3428 total_errors++;
3429 continue;
3430 }
3431 if (!dev->in_fs_metadata || !dev->writeable)
3432 continue;
3433
3434 btrfs_set_stack_device_generation(dev_item, 0);
3435 btrfs_set_stack_device_type(dev_item, dev->type);
3436 btrfs_set_stack_device_id(dev_item, dev->devid);
3437 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3438 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3439 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3440 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3441 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3442 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3443 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3444
3445 flags = btrfs_super_flags(sb);
3446 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3447
3448 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3449 if (ret)
3450 total_errors++;
3451 }
3452 if (total_errors > max_errors) {
3453 btrfs_err(root->fs_info, "%d errors while writing supers",
3454 total_errors);
3455 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3456
3457 /* FUA is masked off if unsupported and can't be the reason */
3458 btrfs_error(root->fs_info, -EIO,
3459 "%d errors while writing supers", total_errors);
3460 return -EIO;
3461 }
3462
3463 total_errors = 0;
3464 list_for_each_entry_rcu(dev, head, dev_list) {
3465 if (!dev->bdev)
3466 continue;
3467 if (!dev->in_fs_metadata || !dev->writeable)
3468 continue;
3469
3470 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3471 if (ret)
3472 total_errors++;
3473 }
3474 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3475 if (total_errors > max_errors) {
3476 btrfs_error(root->fs_info, -EIO,
3477 "%d errors while writing supers", total_errors);
3478 return -EIO;
3479 }
3480 return 0;
3481}
3482
3483int write_ctree_super(struct btrfs_trans_handle *trans,
3484 struct btrfs_root *root, int max_mirrors)
3485{
3486 return write_all_supers(root, max_mirrors);
3487}
3488
3489/* Drop a fs root from the radix tree and free it. */
3490void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3491 struct btrfs_root *root)
3492{
3493 spin_lock(&fs_info->fs_roots_radix_lock);
3494 radix_tree_delete(&fs_info->fs_roots_radix,
3495 (unsigned long)root->root_key.objectid);
3496 spin_unlock(&fs_info->fs_roots_radix_lock);
3497
3498 if (btrfs_root_refs(&root->root_item) == 0)
3499 synchronize_srcu(&fs_info->subvol_srcu);
3500
3501 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3502 btrfs_free_log(NULL, root);
3503
3504 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3505 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3506 free_fs_root(root);
3507}
3508
3509static void free_fs_root(struct btrfs_root *root)
3510{
3511 iput(root->cache_inode);
3512 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3513 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3514 root->orphan_block_rsv = NULL;
3515 if (root->anon_dev)
3516 free_anon_bdev(root->anon_dev);
3517 if (root->subv_writers)
3518 btrfs_free_subvolume_writers(root->subv_writers);
3519 free_extent_buffer(root->node);
3520 free_extent_buffer(root->commit_root);
3521 kfree(root->free_ino_ctl);
3522 kfree(root->free_ino_pinned);
3523 kfree(root->name);
3524 btrfs_put_fs_root(root);
3525}
3526
3527void btrfs_free_fs_root(struct btrfs_root *root)
3528{
3529 free_fs_root(root);
3530}
3531
3532int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3533{
3534 u64 root_objectid = 0;
3535 struct btrfs_root *gang[8];
3536 int i;
3537 int ret;
3538
3539 while (1) {
3540 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3541 (void **)gang, root_objectid,
3542 ARRAY_SIZE(gang));
3543 if (!ret)
3544 break;
3545
3546 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3547 for (i = 0; i < ret; i++) {
3548 int err;
3549
3550 root_objectid = gang[i]->root_key.objectid;
3551 err = btrfs_orphan_cleanup(gang[i]);
3552 if (err)
3553 return err;
3554 }
3555 root_objectid++;
3556 }
3557 return 0;
3558}
3559
3560int btrfs_commit_super(struct btrfs_root *root)
3561{
3562 struct btrfs_trans_handle *trans;
3563
3564 mutex_lock(&root->fs_info->cleaner_mutex);
3565 btrfs_run_delayed_iputs(root);
3566 mutex_unlock(&root->fs_info->cleaner_mutex);
3567 wake_up_process(root->fs_info->cleaner_kthread);
3568
3569 /* wait until ongoing cleanup work done */
3570 down_write(&root->fs_info->cleanup_work_sem);
3571 up_write(&root->fs_info->cleanup_work_sem);
3572
3573 trans = btrfs_join_transaction(root);
3574 if (IS_ERR(trans))
3575 return PTR_ERR(trans);
3576 return btrfs_commit_transaction(trans, root);
3577}
3578
3579int close_ctree(struct btrfs_root *root)
3580{
3581 struct btrfs_fs_info *fs_info = root->fs_info;
3582 int ret;
3583
3584 fs_info->closing = 1;
3585 smp_mb();
3586
3587 /* wait for the uuid_scan task to finish */
3588 down(&fs_info->uuid_tree_rescan_sem);
3589 /* avoid complains from lockdep et al., set sem back to initial state */
3590 up(&fs_info->uuid_tree_rescan_sem);
3591
3592 /* pause restriper - we want to resume on mount */
3593 btrfs_pause_balance(fs_info);
3594
3595 btrfs_dev_replace_suspend_for_unmount(fs_info);
3596
3597 btrfs_scrub_cancel(fs_info);
3598
3599 /* wait for any defraggers to finish */
3600 wait_event(fs_info->transaction_wait,
3601 (atomic_read(&fs_info->defrag_running) == 0));
3602
3603 /* clear out the rbtree of defraggable inodes */
3604 btrfs_cleanup_defrag_inodes(fs_info);
3605
3606 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3607 ret = btrfs_commit_super(root);
3608 if (ret)
3609 btrfs_err(root->fs_info, "commit super ret %d", ret);
3610 }
3611
3612 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3613 btrfs_error_commit_super(root);
3614
3615 kthread_stop(fs_info->transaction_kthread);
3616 kthread_stop(fs_info->cleaner_kthread);
3617
3618 fs_info->closing = 2;
3619 smp_mb();
3620
3621 btrfs_free_qgroup_config(root->fs_info);
3622
3623 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3624 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3625 percpu_counter_sum(&fs_info->delalloc_bytes));
3626 }
3627
3628 btrfs_sysfs_remove_one(fs_info);
3629
3630 del_fs_roots(fs_info);
3631
3632 btrfs_put_block_group_cache(fs_info);
3633
3634 btrfs_free_block_groups(fs_info);
3635
3636 btrfs_stop_all_workers(fs_info);
3637
3638 free_root_pointers(fs_info, 1);
3639
3640 iput(fs_info->btree_inode);
3641
3642#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3643 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3644 btrfsic_unmount(root, fs_info->fs_devices);
3645#endif
3646
3647 btrfs_close_devices(fs_info->fs_devices);
3648 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3649
3650 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3651 percpu_counter_destroy(&fs_info->delalloc_bytes);
3652 percpu_counter_destroy(&fs_info->bio_counter);
3653 bdi_destroy(&fs_info->bdi);
3654 cleanup_srcu_struct(&fs_info->subvol_srcu);
3655
3656 btrfs_free_stripe_hash_table(fs_info);
3657
3658 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3659 root->orphan_block_rsv = NULL;
3660
3661 return 0;
3662}
3663
3664int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3665 int atomic)
3666{
3667 int ret;
3668 struct inode *btree_inode = buf->pages[0]->mapping->host;
3669
3670 ret = extent_buffer_uptodate(buf);
3671 if (!ret)
3672 return ret;
3673
3674 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3675 parent_transid, atomic);
3676 if (ret == -EAGAIN)
3677 return ret;
3678 return !ret;
3679}
3680
3681int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3682{
3683 return set_extent_buffer_uptodate(buf);
3684}
3685
3686void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3687{
3688 struct btrfs_root *root;
3689 u64 transid = btrfs_header_generation(buf);
3690 int was_dirty;
3691
3692#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3693 /*
3694 * This is a fast path so only do this check if we have sanity tests
3695 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3696 * outside of the sanity tests.
3697 */
3698 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3699 return;
3700#endif
3701 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3702 btrfs_assert_tree_locked(buf);
3703 if (transid != root->fs_info->generation)
3704 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3705 "found %llu running %llu\n",
3706 buf->start, transid, root->fs_info->generation);
3707 was_dirty = set_extent_buffer_dirty(buf);
3708 if (!was_dirty)
3709 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3710 buf->len,
3711 root->fs_info->dirty_metadata_batch);
3712}
3713
3714static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3715 int flush_delayed)
3716{
3717 /*
3718 * looks as though older kernels can get into trouble with
3719 * this code, they end up stuck in balance_dirty_pages forever
3720 */
3721 int ret;
3722
3723 if (current->flags & PF_MEMALLOC)
3724 return;
3725
3726 if (flush_delayed)
3727 btrfs_balance_delayed_items(root);
3728
3729 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3730 BTRFS_DIRTY_METADATA_THRESH);
3731 if (ret > 0) {
3732 balance_dirty_pages_ratelimited(
3733 root->fs_info->btree_inode->i_mapping);
3734 }
3735 return;
3736}
3737
3738void btrfs_btree_balance_dirty(struct btrfs_root *root)
3739{
3740 __btrfs_btree_balance_dirty(root, 1);
3741}
3742
3743void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3744{
3745 __btrfs_btree_balance_dirty(root, 0);
3746}
3747
3748int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3749{
3750 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3751 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3752}
3753
3754static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3755 int read_only)
3756{
3757 /*
3758 * Placeholder for checks
3759 */
3760 return 0;
3761}
3762
3763static void btrfs_error_commit_super(struct btrfs_root *root)
3764{
3765 mutex_lock(&root->fs_info->cleaner_mutex);
3766 btrfs_run_delayed_iputs(root);
3767 mutex_unlock(&root->fs_info->cleaner_mutex);
3768
3769 down_write(&root->fs_info->cleanup_work_sem);
3770 up_write(&root->fs_info->cleanup_work_sem);
3771
3772 /* cleanup FS via transaction */
3773 btrfs_cleanup_transaction(root);
3774}
3775
3776static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3777 struct btrfs_root *root)
3778{
3779 struct btrfs_inode *btrfs_inode;
3780 struct list_head splice;
3781
3782 INIT_LIST_HEAD(&splice);
3783
3784 mutex_lock(&root->fs_info->ordered_operations_mutex);
3785 spin_lock(&root->fs_info->ordered_root_lock);
3786
3787 list_splice_init(&t->ordered_operations, &splice);
3788 while (!list_empty(&splice)) {
3789 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3790 ordered_operations);
3791
3792 list_del_init(&btrfs_inode->ordered_operations);
3793 spin_unlock(&root->fs_info->ordered_root_lock);
3794
3795 btrfs_invalidate_inodes(btrfs_inode->root);
3796
3797 spin_lock(&root->fs_info->ordered_root_lock);
3798 }
3799
3800 spin_unlock(&root->fs_info->ordered_root_lock);
3801 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3802}
3803
3804static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3805{
3806 struct btrfs_ordered_extent *ordered;
3807
3808 spin_lock(&root->ordered_extent_lock);
3809 /*
3810 * This will just short circuit the ordered completion stuff which will
3811 * make sure the ordered extent gets properly cleaned up.
3812 */
3813 list_for_each_entry(ordered, &root->ordered_extents,
3814 root_extent_list)
3815 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3816 spin_unlock(&root->ordered_extent_lock);
3817}
3818
3819static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3820{
3821 struct btrfs_root *root;
3822 struct list_head splice;
3823
3824 INIT_LIST_HEAD(&splice);
3825
3826 spin_lock(&fs_info->ordered_root_lock);
3827 list_splice_init(&fs_info->ordered_roots, &splice);
3828 while (!list_empty(&splice)) {
3829 root = list_first_entry(&splice, struct btrfs_root,
3830 ordered_root);
3831 list_move_tail(&root->ordered_root,
3832 &fs_info->ordered_roots);
3833
3834 spin_unlock(&fs_info->ordered_root_lock);
3835 btrfs_destroy_ordered_extents(root);
3836
3837 cond_resched();
3838 spin_lock(&fs_info->ordered_root_lock);
3839 }
3840 spin_unlock(&fs_info->ordered_root_lock);
3841}
3842
3843static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3844 struct btrfs_root *root)
3845{
3846 struct rb_node *node;
3847 struct btrfs_delayed_ref_root *delayed_refs;
3848 struct btrfs_delayed_ref_node *ref;
3849 int ret = 0;
3850
3851 delayed_refs = &trans->delayed_refs;
3852
3853 spin_lock(&delayed_refs->lock);
3854 if (atomic_read(&delayed_refs->num_entries) == 0) {
3855 spin_unlock(&delayed_refs->lock);
3856 btrfs_info(root->fs_info, "delayed_refs has NO entry");
3857 return ret;
3858 }
3859
3860 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3861 struct btrfs_delayed_ref_head *head;
3862 bool pin_bytes = false;
3863
3864 head = rb_entry(node, struct btrfs_delayed_ref_head,
3865 href_node);
3866 if (!mutex_trylock(&head->mutex)) {
3867 atomic_inc(&head->node.refs);
3868 spin_unlock(&delayed_refs->lock);
3869
3870 mutex_lock(&head->mutex);
3871 mutex_unlock(&head->mutex);
3872 btrfs_put_delayed_ref(&head->node);
3873 spin_lock(&delayed_refs->lock);
3874 continue;
3875 }
3876 spin_lock(&head->lock);
3877 while ((node = rb_first(&head->ref_root)) != NULL) {
3878 ref = rb_entry(node, struct btrfs_delayed_ref_node,
3879 rb_node);
3880 ref->in_tree = 0;
3881 rb_erase(&ref->rb_node, &head->ref_root);
3882 atomic_dec(&delayed_refs->num_entries);
3883 btrfs_put_delayed_ref(ref);
3884 }
3885 if (head->must_insert_reserved)
3886 pin_bytes = true;
3887 btrfs_free_delayed_extent_op(head->extent_op);
3888 delayed_refs->num_heads--;
3889 if (head->processing == 0)
3890 delayed_refs->num_heads_ready--;
3891 atomic_dec(&delayed_refs->num_entries);
3892 head->node.in_tree = 0;
3893 rb_erase(&head->href_node, &delayed_refs->href_root);
3894 spin_unlock(&head->lock);
3895 spin_unlock(&delayed_refs->lock);
3896 mutex_unlock(&head->mutex);
3897
3898 if (pin_bytes)
3899 btrfs_pin_extent(root, head->node.bytenr,
3900 head->node.num_bytes, 1);
3901 btrfs_put_delayed_ref(&head->node);
3902 cond_resched();
3903 spin_lock(&delayed_refs->lock);
3904 }
3905
3906 spin_unlock(&delayed_refs->lock);
3907
3908 return ret;
3909}
3910
3911static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3912{
3913 struct btrfs_inode *btrfs_inode;
3914 struct list_head splice;
3915
3916 INIT_LIST_HEAD(&splice);
3917
3918 spin_lock(&root->delalloc_lock);
3919 list_splice_init(&root->delalloc_inodes, &splice);
3920
3921 while (!list_empty(&splice)) {
3922 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3923 delalloc_inodes);
3924
3925 list_del_init(&btrfs_inode->delalloc_inodes);
3926 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3927 &btrfs_inode->runtime_flags);
3928 spin_unlock(&root->delalloc_lock);
3929
3930 btrfs_invalidate_inodes(btrfs_inode->root);
3931
3932 spin_lock(&root->delalloc_lock);
3933 }
3934
3935 spin_unlock(&root->delalloc_lock);
3936}
3937
3938static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3939{
3940 struct btrfs_root *root;
3941 struct list_head splice;
3942
3943 INIT_LIST_HEAD(&splice);
3944
3945 spin_lock(&fs_info->delalloc_root_lock);
3946 list_splice_init(&fs_info->delalloc_roots, &splice);
3947 while (!list_empty(&splice)) {
3948 root = list_first_entry(&splice, struct btrfs_root,
3949 delalloc_root);
3950 list_del_init(&root->delalloc_root);
3951 root = btrfs_grab_fs_root(root);
3952 BUG_ON(!root);
3953 spin_unlock(&fs_info->delalloc_root_lock);
3954
3955 btrfs_destroy_delalloc_inodes(root);
3956 btrfs_put_fs_root(root);
3957
3958 spin_lock(&fs_info->delalloc_root_lock);
3959 }
3960 spin_unlock(&fs_info->delalloc_root_lock);
3961}
3962
3963static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3964 struct extent_io_tree *dirty_pages,
3965 int mark)
3966{
3967 int ret;
3968 struct extent_buffer *eb;
3969 u64 start = 0;
3970 u64 end;
3971
3972 while (1) {
3973 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3974 mark, NULL);
3975 if (ret)
3976 break;
3977
3978 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3979 while (start <= end) {
3980 eb = btrfs_find_tree_block(root, start,
3981 root->leafsize);
3982 start += root->leafsize;
3983 if (!eb)
3984 continue;
3985 wait_on_extent_buffer_writeback(eb);
3986
3987 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3988 &eb->bflags))
3989 clear_extent_buffer_dirty(eb);
3990 free_extent_buffer_stale(eb);
3991 }
3992 }
3993
3994 return ret;
3995}
3996
3997static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3998 struct extent_io_tree *pinned_extents)
3999{
4000 struct extent_io_tree *unpin;
4001 u64 start;
4002 u64 end;
4003 int ret;
4004 bool loop = true;
4005
4006 unpin = pinned_extents;
4007again:
4008 while (1) {
4009 ret = find_first_extent_bit(unpin, 0, &start, &end,
4010 EXTENT_DIRTY, NULL);
4011 if (ret)
4012 break;
4013
4014 /* opt_discard */
4015 if (btrfs_test_opt(root, DISCARD))
4016 ret = btrfs_error_discard_extent(root, start,
4017 end + 1 - start,
4018 NULL);
4019
4020 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4021 btrfs_error_unpin_extent_range(root, start, end);
4022 cond_resched();
4023 }
4024
4025 if (loop) {
4026 if (unpin == &root->fs_info->freed_extents[0])
4027 unpin = &root->fs_info->freed_extents[1];
4028 else
4029 unpin = &root->fs_info->freed_extents[0];
4030 loop = false;
4031 goto again;
4032 }
4033
4034 return 0;
4035}
4036
4037void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4038 struct btrfs_root *root)
4039{
4040 btrfs_destroy_ordered_operations(cur_trans, root);
4041
4042 btrfs_destroy_delayed_refs(cur_trans, root);
4043
4044 cur_trans->state = TRANS_STATE_COMMIT_START;
4045 wake_up(&root->fs_info->transaction_blocked_wait);
4046
4047 cur_trans->state = TRANS_STATE_UNBLOCKED;
4048 wake_up(&root->fs_info->transaction_wait);
4049
4050 btrfs_destroy_delayed_inodes(root);
4051 btrfs_assert_delayed_root_empty(root);
4052
4053 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4054 EXTENT_DIRTY);
4055 btrfs_destroy_pinned_extent(root,
4056 root->fs_info->pinned_extents);
4057
4058 cur_trans->state =TRANS_STATE_COMPLETED;
4059 wake_up(&cur_trans->commit_wait);
4060
4061 /*
4062 memset(cur_trans, 0, sizeof(*cur_trans));
4063 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4064 */
4065}
4066
4067static int btrfs_cleanup_transaction(struct btrfs_root *root)
4068{
4069 struct btrfs_transaction *t;
4070
4071 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4072
4073 spin_lock(&root->fs_info->trans_lock);
4074 while (!list_empty(&root->fs_info->trans_list)) {
4075 t = list_first_entry(&root->fs_info->trans_list,
4076 struct btrfs_transaction, list);
4077 if (t->state >= TRANS_STATE_COMMIT_START) {
4078 atomic_inc(&t->use_count);
4079 spin_unlock(&root->fs_info->trans_lock);
4080 btrfs_wait_for_commit(root, t->transid);
4081 btrfs_put_transaction(t);
4082 spin_lock(&root->fs_info->trans_lock);
4083 continue;
4084 }
4085 if (t == root->fs_info->running_transaction) {
4086 t->state = TRANS_STATE_COMMIT_DOING;
4087 spin_unlock(&root->fs_info->trans_lock);
4088 /*
4089 * We wait for 0 num_writers since we don't hold a trans
4090 * handle open currently for this transaction.
4091 */
4092 wait_event(t->writer_wait,
4093 atomic_read(&t->num_writers) == 0);
4094 } else {
4095 spin_unlock(&root->fs_info->trans_lock);
4096 }
4097 btrfs_cleanup_one_transaction(t, root);
4098
4099 spin_lock(&root->fs_info->trans_lock);
4100 if (t == root->fs_info->running_transaction)
4101 root->fs_info->running_transaction = NULL;
4102 list_del_init(&t->list);
4103 spin_unlock(&root->fs_info->trans_lock);
4104
4105 btrfs_put_transaction(t);
4106 trace_btrfs_transaction_commit(root);
4107 spin_lock(&root->fs_info->trans_lock);
4108 }
4109 spin_unlock(&root->fs_info->trans_lock);
4110 btrfs_destroy_all_ordered_extents(root->fs_info);
4111 btrfs_destroy_delayed_inodes(root);
4112 btrfs_assert_delayed_root_empty(root);
4113 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4114 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4115 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4116
4117 return 0;
4118}
4119
4120static struct extent_io_ops btree_extent_io_ops = {
4121 .readpage_end_io_hook = btree_readpage_end_io_hook,
4122 .readpage_io_failed_hook = btree_io_failed_hook,
4123 .submit_bio_hook = btree_submit_bio_hook,
4124 /* note we're sharing with inode.c for the merge bio hook */
4125 .merge_bio_hook = btrfs_merge_bio_hook,
4126};
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6#include <linux/fs.h>
7#include <linux/blkdev.h>
8#include <linux/radix-tree.h>
9#include <linux/writeback.h>
10#include <linux/workqueue.h>
11#include <linux/kthread.h>
12#include <linux/slab.h>
13#include <linux/migrate.h>
14#include <linux/ratelimit.h>
15#include <linux/uuid.h>
16#include <linux/semaphore.h>
17#include <linux/error-injection.h>
18#include <linux/crc32c.h>
19#include <linux/sched/mm.h>
20#include <asm/unaligned.h>
21#include <crypto/hash.h>
22#include "ctree.h"
23#include "disk-io.h"
24#include "transaction.h"
25#include "btrfs_inode.h"
26#include "bio.h"
27#include "print-tree.h"
28#include "locking.h"
29#include "tree-log.h"
30#include "free-space-cache.h"
31#include "free-space-tree.h"
32#include "dev-replace.h"
33#include "raid56.h"
34#include "sysfs.h"
35#include "qgroup.h"
36#include "compression.h"
37#include "tree-checker.h"
38#include "ref-verify.h"
39#include "block-group.h"
40#include "discard.h"
41#include "space-info.h"
42#include "zoned.h"
43#include "subpage.h"
44#include "fs.h"
45#include "accessors.h"
46#include "extent-tree.h"
47#include "root-tree.h"
48#include "defrag.h"
49#include "uuid-tree.h"
50#include "relocation.h"
51#include "scrub.h"
52#include "super.h"
53
54#define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
55 BTRFS_HEADER_FLAG_RELOC |\
56 BTRFS_SUPER_FLAG_ERROR |\
57 BTRFS_SUPER_FLAG_SEEDING |\
58 BTRFS_SUPER_FLAG_METADUMP |\
59 BTRFS_SUPER_FLAG_METADUMP_V2)
60
61static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
62static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
63
64static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
65{
66 if (fs_info->csum_shash)
67 crypto_free_shash(fs_info->csum_shash);
68}
69
70/*
71 * Compute the csum of a btree block and store the result to provided buffer.
72 */
73static void csum_tree_block(struct extent_buffer *buf, u8 *result)
74{
75 struct btrfs_fs_info *fs_info = buf->fs_info;
76 int num_pages;
77 u32 first_page_part;
78 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
79 char *kaddr;
80 int i;
81
82 shash->tfm = fs_info->csum_shash;
83 crypto_shash_init(shash);
84
85 if (buf->addr) {
86 /* Pages are contiguous, handle them as a big one. */
87 kaddr = buf->addr;
88 first_page_part = fs_info->nodesize;
89 num_pages = 1;
90 } else {
91 kaddr = folio_address(buf->folios[0]);
92 first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
93 num_pages = num_extent_pages(buf);
94 }
95
96 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
97 first_page_part - BTRFS_CSUM_SIZE);
98
99 /*
100 * Multiple single-page folios case would reach here.
101 *
102 * nodesize <= PAGE_SIZE and large folio all handled by above
103 * crypto_shash_update() already.
104 */
105 for (i = 1; i < num_pages && INLINE_EXTENT_BUFFER_PAGES > 1; i++) {
106 kaddr = folio_address(buf->folios[i]);
107 crypto_shash_update(shash, kaddr, PAGE_SIZE);
108 }
109 memset(result, 0, BTRFS_CSUM_SIZE);
110 crypto_shash_final(shash, result);
111}
112
113/*
114 * we can't consider a given block up to date unless the transid of the
115 * block matches the transid in the parent node's pointer. This is how we
116 * detect blocks that either didn't get written at all or got written
117 * in the wrong place.
118 */
119int btrfs_buffer_uptodate(struct extent_buffer *eb, u64 parent_transid, int atomic)
120{
121 if (!extent_buffer_uptodate(eb))
122 return 0;
123
124 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
125 return 1;
126
127 if (atomic)
128 return -EAGAIN;
129
130 if (!extent_buffer_uptodate(eb) ||
131 btrfs_header_generation(eb) != parent_transid) {
132 btrfs_err_rl(eb->fs_info,
133"parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
134 eb->start, eb->read_mirror,
135 parent_transid, btrfs_header_generation(eb));
136 clear_extent_buffer_uptodate(eb);
137 return 0;
138 }
139 return 1;
140}
141
142static bool btrfs_supported_super_csum(u16 csum_type)
143{
144 switch (csum_type) {
145 case BTRFS_CSUM_TYPE_CRC32:
146 case BTRFS_CSUM_TYPE_XXHASH:
147 case BTRFS_CSUM_TYPE_SHA256:
148 case BTRFS_CSUM_TYPE_BLAKE2:
149 return true;
150 default:
151 return false;
152 }
153}
154
155/*
156 * Return 0 if the superblock checksum type matches the checksum value of that
157 * algorithm. Pass the raw disk superblock data.
158 */
159int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
160 const struct btrfs_super_block *disk_sb)
161{
162 char result[BTRFS_CSUM_SIZE];
163 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
164
165 shash->tfm = fs_info->csum_shash;
166
167 /*
168 * The super_block structure does not span the whole
169 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
170 * filled with zeros and is included in the checksum.
171 */
172 crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
173 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
174
175 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
176 return 1;
177
178 return 0;
179}
180
181static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
182 int mirror_num)
183{
184 struct btrfs_fs_info *fs_info = eb->fs_info;
185 int num_folios = num_extent_folios(eb);
186 int ret = 0;
187
188 if (sb_rdonly(fs_info->sb))
189 return -EROFS;
190
191 for (int i = 0; i < num_folios; i++) {
192 struct folio *folio = eb->folios[i];
193 u64 start = max_t(u64, eb->start, folio_pos(folio));
194 u64 end = min_t(u64, eb->start + eb->len,
195 folio_pos(folio) + eb->folio_size);
196 u32 len = end - start;
197
198 ret = btrfs_repair_io_failure(fs_info, 0, start, len,
199 start, folio, offset_in_folio(folio, start),
200 mirror_num);
201 if (ret)
202 break;
203 }
204
205 return ret;
206}
207
208/*
209 * helper to read a given tree block, doing retries as required when
210 * the checksums don't match and we have alternate mirrors to try.
211 *
212 * @check: expected tree parentness check, see the comments of the
213 * structure for details.
214 */
215int btrfs_read_extent_buffer(struct extent_buffer *eb,
216 struct btrfs_tree_parent_check *check)
217{
218 struct btrfs_fs_info *fs_info = eb->fs_info;
219 int failed = 0;
220 int ret;
221 int num_copies = 0;
222 int mirror_num = 0;
223 int failed_mirror = 0;
224
225 ASSERT(check);
226
227 while (1) {
228 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
229 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check);
230 if (!ret)
231 break;
232
233 num_copies = btrfs_num_copies(fs_info,
234 eb->start, eb->len);
235 if (num_copies == 1)
236 break;
237
238 if (!failed_mirror) {
239 failed = 1;
240 failed_mirror = eb->read_mirror;
241 }
242
243 mirror_num++;
244 if (mirror_num == failed_mirror)
245 mirror_num++;
246
247 if (mirror_num > num_copies)
248 break;
249 }
250
251 if (failed && !ret && failed_mirror)
252 btrfs_repair_eb_io_failure(eb, failed_mirror);
253
254 return ret;
255}
256
257/*
258 * Checksum a dirty tree block before IO.
259 */
260blk_status_t btree_csum_one_bio(struct btrfs_bio *bbio)
261{
262 struct extent_buffer *eb = bbio->private;
263 struct btrfs_fs_info *fs_info = eb->fs_info;
264 u64 found_start = btrfs_header_bytenr(eb);
265 u64 last_trans;
266 u8 result[BTRFS_CSUM_SIZE];
267 int ret;
268
269 /* Btree blocks are always contiguous on disk. */
270 if (WARN_ON_ONCE(bbio->file_offset != eb->start))
271 return BLK_STS_IOERR;
272 if (WARN_ON_ONCE(bbio->bio.bi_iter.bi_size != eb->len))
273 return BLK_STS_IOERR;
274
275 /*
276 * If an extent_buffer is marked as EXTENT_BUFFER_ZONED_ZEROOUT, don't
277 * checksum it but zero-out its content. This is done to preserve
278 * ordering of I/O without unnecessarily writing out data.
279 */
280 if (test_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags)) {
281 memzero_extent_buffer(eb, 0, eb->len);
282 return BLK_STS_OK;
283 }
284
285 if (WARN_ON_ONCE(found_start != eb->start))
286 return BLK_STS_IOERR;
287 if (WARN_ON(!btrfs_folio_test_uptodate(fs_info, eb->folios[0],
288 eb->start, eb->len)))
289 return BLK_STS_IOERR;
290
291 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
292 offsetof(struct btrfs_header, fsid),
293 BTRFS_FSID_SIZE) == 0);
294 csum_tree_block(eb, result);
295
296 if (btrfs_header_level(eb))
297 ret = btrfs_check_node(eb);
298 else
299 ret = btrfs_check_leaf(eb);
300
301 if (ret < 0)
302 goto error;
303
304 /*
305 * Also check the generation, the eb reached here must be newer than
306 * last committed. Or something seriously wrong happened.
307 */
308 last_trans = btrfs_get_last_trans_committed(fs_info);
309 if (unlikely(btrfs_header_generation(eb) <= last_trans)) {
310 ret = -EUCLEAN;
311 btrfs_err(fs_info,
312 "block=%llu bad generation, have %llu expect > %llu",
313 eb->start, btrfs_header_generation(eb), last_trans);
314 goto error;
315 }
316 write_extent_buffer(eb, result, 0, fs_info->csum_size);
317 return BLK_STS_OK;
318
319error:
320 btrfs_print_tree(eb, 0);
321 btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
322 eb->start);
323 /*
324 * Be noisy if this is an extent buffer from a log tree. We don't abort
325 * a transaction in case there's a bad log tree extent buffer, we just
326 * fallback to a transaction commit. Still we want to know when there is
327 * a bad log tree extent buffer, as that may signal a bug somewhere.
328 */
329 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
330 btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
331 return errno_to_blk_status(ret);
332}
333
334static bool check_tree_block_fsid(struct extent_buffer *eb)
335{
336 struct btrfs_fs_info *fs_info = eb->fs_info;
337 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
338 u8 fsid[BTRFS_FSID_SIZE];
339
340 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
341 BTRFS_FSID_SIZE);
342
343 /*
344 * alloc_fsid_devices() copies the fsid into fs_devices::metadata_uuid.
345 * This is then overwritten by metadata_uuid if it is present in the
346 * device_list_add(). The same true for a seed device as well. So use of
347 * fs_devices::metadata_uuid is appropriate here.
348 */
349 if (memcmp(fsid, fs_info->fs_devices->metadata_uuid, BTRFS_FSID_SIZE) == 0)
350 return false;
351
352 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
353 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
354 return false;
355
356 return true;
357}
358
359/* Do basic extent buffer checks at read time */
360int btrfs_validate_extent_buffer(struct extent_buffer *eb,
361 struct btrfs_tree_parent_check *check)
362{
363 struct btrfs_fs_info *fs_info = eb->fs_info;
364 u64 found_start;
365 const u32 csum_size = fs_info->csum_size;
366 u8 found_level;
367 u8 result[BTRFS_CSUM_SIZE];
368 const u8 *header_csum;
369 int ret = 0;
370
371 ASSERT(check);
372
373 found_start = btrfs_header_bytenr(eb);
374 if (found_start != eb->start) {
375 btrfs_err_rl(fs_info,
376 "bad tree block start, mirror %u want %llu have %llu",
377 eb->read_mirror, eb->start, found_start);
378 ret = -EIO;
379 goto out;
380 }
381 if (check_tree_block_fsid(eb)) {
382 btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
383 eb->start, eb->read_mirror);
384 ret = -EIO;
385 goto out;
386 }
387 found_level = btrfs_header_level(eb);
388 if (found_level >= BTRFS_MAX_LEVEL) {
389 btrfs_err(fs_info,
390 "bad tree block level, mirror %u level %d on logical %llu",
391 eb->read_mirror, btrfs_header_level(eb), eb->start);
392 ret = -EIO;
393 goto out;
394 }
395
396 csum_tree_block(eb, result);
397 header_csum = folio_address(eb->folios[0]) +
398 get_eb_offset_in_folio(eb, offsetof(struct btrfs_header, csum));
399
400 if (memcmp(result, header_csum, csum_size) != 0) {
401 btrfs_warn_rl(fs_info,
402"checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d",
403 eb->start, eb->read_mirror,
404 CSUM_FMT_VALUE(csum_size, header_csum),
405 CSUM_FMT_VALUE(csum_size, result),
406 btrfs_header_level(eb));
407 ret = -EUCLEAN;
408 goto out;
409 }
410
411 if (found_level != check->level) {
412 btrfs_err(fs_info,
413 "level verify failed on logical %llu mirror %u wanted %u found %u",
414 eb->start, eb->read_mirror, check->level, found_level);
415 ret = -EIO;
416 goto out;
417 }
418 if (unlikely(check->transid &&
419 btrfs_header_generation(eb) != check->transid)) {
420 btrfs_err_rl(eb->fs_info,
421"parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
422 eb->start, eb->read_mirror, check->transid,
423 btrfs_header_generation(eb));
424 ret = -EIO;
425 goto out;
426 }
427 if (check->has_first_key) {
428 struct btrfs_key *expect_key = &check->first_key;
429 struct btrfs_key found_key;
430
431 if (found_level)
432 btrfs_node_key_to_cpu(eb, &found_key, 0);
433 else
434 btrfs_item_key_to_cpu(eb, &found_key, 0);
435 if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
436 btrfs_err(fs_info,
437"tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
438 eb->start, check->transid,
439 expect_key->objectid,
440 expect_key->type, expect_key->offset,
441 found_key.objectid, found_key.type,
442 found_key.offset);
443 ret = -EUCLEAN;
444 goto out;
445 }
446 }
447 if (check->owner_root) {
448 ret = btrfs_check_eb_owner(eb, check->owner_root);
449 if (ret < 0)
450 goto out;
451 }
452
453 /*
454 * If this is a leaf block and it is corrupt, set the corrupt bit so
455 * that we don't try and read the other copies of this block, just
456 * return -EIO.
457 */
458 if (found_level == 0 && btrfs_check_leaf(eb)) {
459 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
460 ret = -EIO;
461 }
462
463 if (found_level > 0 && btrfs_check_node(eb))
464 ret = -EIO;
465
466 if (ret)
467 btrfs_err(fs_info,
468 "read time tree block corruption detected on logical %llu mirror %u",
469 eb->start, eb->read_mirror);
470out:
471 return ret;
472}
473
474#ifdef CONFIG_MIGRATION
475static int btree_migrate_folio(struct address_space *mapping,
476 struct folio *dst, struct folio *src, enum migrate_mode mode)
477{
478 /*
479 * we can't safely write a btree page from here,
480 * we haven't done the locking hook
481 */
482 if (folio_test_dirty(src))
483 return -EAGAIN;
484 /*
485 * Buffers may be managed in a filesystem specific way.
486 * We must have no buffers or drop them.
487 */
488 if (folio_get_private(src) &&
489 !filemap_release_folio(src, GFP_KERNEL))
490 return -EAGAIN;
491 return migrate_folio(mapping, dst, src, mode);
492}
493#else
494#define btree_migrate_folio NULL
495#endif
496
497static int btree_writepages(struct address_space *mapping,
498 struct writeback_control *wbc)
499{
500 int ret;
501
502 if (wbc->sync_mode == WB_SYNC_NONE) {
503 struct btrfs_fs_info *fs_info;
504
505 if (wbc->for_kupdate)
506 return 0;
507
508 fs_info = inode_to_fs_info(mapping->host);
509 /* this is a bit racy, but that's ok */
510 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
511 BTRFS_DIRTY_METADATA_THRESH,
512 fs_info->dirty_metadata_batch);
513 if (ret < 0)
514 return 0;
515 }
516 return btree_write_cache_pages(mapping, wbc);
517}
518
519static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
520{
521 if (folio_test_writeback(folio) || folio_test_dirty(folio))
522 return false;
523
524 return try_release_extent_buffer(&folio->page);
525}
526
527static void btree_invalidate_folio(struct folio *folio, size_t offset,
528 size_t length)
529{
530 struct extent_io_tree *tree;
531
532 tree = &folio_to_inode(folio)->io_tree;
533 extent_invalidate_folio(tree, folio, offset);
534 btree_release_folio(folio, GFP_NOFS);
535 if (folio_get_private(folio)) {
536 btrfs_warn(folio_to_fs_info(folio),
537 "folio private not zero on folio %llu",
538 (unsigned long long)folio_pos(folio));
539 folio_detach_private(folio);
540 }
541}
542
543#ifdef DEBUG
544static bool btree_dirty_folio(struct address_space *mapping,
545 struct folio *folio)
546{
547 struct btrfs_fs_info *fs_info = inode_to_fs_info(mapping->host);
548 struct btrfs_subpage_info *spi = fs_info->subpage_info;
549 struct btrfs_subpage *subpage;
550 struct extent_buffer *eb;
551 int cur_bit = 0;
552 u64 page_start = folio_pos(folio);
553
554 if (fs_info->sectorsize == PAGE_SIZE) {
555 eb = folio_get_private(folio);
556 BUG_ON(!eb);
557 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
558 BUG_ON(!atomic_read(&eb->refs));
559 btrfs_assert_tree_write_locked(eb);
560 return filemap_dirty_folio(mapping, folio);
561 }
562
563 ASSERT(spi);
564 subpage = folio_get_private(folio);
565
566 for (cur_bit = spi->dirty_offset;
567 cur_bit < spi->dirty_offset + spi->bitmap_nr_bits;
568 cur_bit++) {
569 unsigned long flags;
570 u64 cur;
571
572 spin_lock_irqsave(&subpage->lock, flags);
573 if (!test_bit(cur_bit, subpage->bitmaps)) {
574 spin_unlock_irqrestore(&subpage->lock, flags);
575 continue;
576 }
577 spin_unlock_irqrestore(&subpage->lock, flags);
578 cur = page_start + cur_bit * fs_info->sectorsize;
579
580 eb = find_extent_buffer(fs_info, cur);
581 ASSERT(eb);
582 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
583 ASSERT(atomic_read(&eb->refs));
584 btrfs_assert_tree_write_locked(eb);
585 free_extent_buffer(eb);
586
587 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits) - 1;
588 }
589 return filemap_dirty_folio(mapping, folio);
590}
591#else
592#define btree_dirty_folio filemap_dirty_folio
593#endif
594
595static const struct address_space_operations btree_aops = {
596 .writepages = btree_writepages,
597 .release_folio = btree_release_folio,
598 .invalidate_folio = btree_invalidate_folio,
599 .migrate_folio = btree_migrate_folio,
600 .dirty_folio = btree_dirty_folio,
601};
602
603struct extent_buffer *btrfs_find_create_tree_block(
604 struct btrfs_fs_info *fs_info,
605 u64 bytenr, u64 owner_root,
606 int level)
607{
608 if (btrfs_is_testing(fs_info))
609 return alloc_test_extent_buffer(fs_info, bytenr);
610 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
611}
612
613/*
614 * Read tree block at logical address @bytenr and do variant basic but critical
615 * verification.
616 *
617 * @check: expected tree parentness check, see comments of the
618 * structure for details.
619 */
620struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
621 struct btrfs_tree_parent_check *check)
622{
623 struct extent_buffer *buf = NULL;
624 int ret;
625
626 ASSERT(check);
627
628 buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
629 check->level);
630 if (IS_ERR(buf))
631 return buf;
632
633 ret = btrfs_read_extent_buffer(buf, check);
634 if (ret) {
635 free_extent_buffer_stale(buf);
636 return ERR_PTR(ret);
637 }
638 if (btrfs_check_eb_owner(buf, check->owner_root)) {
639 free_extent_buffer_stale(buf);
640 return ERR_PTR(-EUCLEAN);
641 }
642 return buf;
643
644}
645
646static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
647 u64 objectid)
648{
649 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
650
651 memset(&root->root_key, 0, sizeof(root->root_key));
652 memset(&root->root_item, 0, sizeof(root->root_item));
653 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
654 root->fs_info = fs_info;
655 root->root_key.objectid = objectid;
656 root->node = NULL;
657 root->commit_root = NULL;
658 root->state = 0;
659 RB_CLEAR_NODE(&root->rb_node);
660
661 root->last_trans = 0;
662 root->free_objectid = 0;
663 root->nr_delalloc_inodes = 0;
664 root->nr_ordered_extents = 0;
665 root->inode_tree = RB_ROOT;
666 /* GFP flags are compatible with XA_FLAGS_*. */
667 xa_init_flags(&root->delayed_nodes, GFP_ATOMIC);
668
669 btrfs_init_root_block_rsv(root);
670
671 INIT_LIST_HEAD(&root->dirty_list);
672 INIT_LIST_HEAD(&root->root_list);
673 INIT_LIST_HEAD(&root->delalloc_inodes);
674 INIT_LIST_HEAD(&root->delalloc_root);
675 INIT_LIST_HEAD(&root->ordered_extents);
676 INIT_LIST_HEAD(&root->ordered_root);
677 INIT_LIST_HEAD(&root->reloc_dirty_list);
678 spin_lock_init(&root->inode_lock);
679 spin_lock_init(&root->delalloc_lock);
680 spin_lock_init(&root->ordered_extent_lock);
681 spin_lock_init(&root->accounting_lock);
682 spin_lock_init(&root->qgroup_meta_rsv_lock);
683 mutex_init(&root->objectid_mutex);
684 mutex_init(&root->log_mutex);
685 mutex_init(&root->ordered_extent_mutex);
686 mutex_init(&root->delalloc_mutex);
687 init_waitqueue_head(&root->qgroup_flush_wait);
688 init_waitqueue_head(&root->log_writer_wait);
689 init_waitqueue_head(&root->log_commit_wait[0]);
690 init_waitqueue_head(&root->log_commit_wait[1]);
691 INIT_LIST_HEAD(&root->log_ctxs[0]);
692 INIT_LIST_HEAD(&root->log_ctxs[1]);
693 atomic_set(&root->log_commit[0], 0);
694 atomic_set(&root->log_commit[1], 0);
695 atomic_set(&root->log_writers, 0);
696 atomic_set(&root->log_batch, 0);
697 refcount_set(&root->refs, 1);
698 atomic_set(&root->snapshot_force_cow, 0);
699 atomic_set(&root->nr_swapfiles, 0);
700 btrfs_set_root_log_transid(root, 0);
701 root->log_transid_committed = -1;
702 btrfs_set_root_last_log_commit(root, 0);
703 root->anon_dev = 0;
704 if (!dummy) {
705 extent_io_tree_init(fs_info, &root->dirty_log_pages,
706 IO_TREE_ROOT_DIRTY_LOG_PAGES);
707 extent_io_tree_init(fs_info, &root->log_csum_range,
708 IO_TREE_LOG_CSUM_RANGE);
709 }
710
711 spin_lock_init(&root->root_item_lock);
712 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
713#ifdef CONFIG_BTRFS_DEBUG
714 INIT_LIST_HEAD(&root->leak_list);
715 spin_lock(&fs_info->fs_roots_radix_lock);
716 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
717 spin_unlock(&fs_info->fs_roots_radix_lock);
718#endif
719}
720
721static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
722 u64 objectid, gfp_t flags)
723{
724 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
725 if (root)
726 __setup_root(root, fs_info, objectid);
727 return root;
728}
729
730#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
731/* Should only be used by the testing infrastructure */
732struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
733{
734 struct btrfs_root *root;
735
736 if (!fs_info)
737 return ERR_PTR(-EINVAL);
738
739 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
740 if (!root)
741 return ERR_PTR(-ENOMEM);
742
743 /* We don't use the stripesize in selftest, set it as sectorsize */
744 root->alloc_bytenr = 0;
745
746 return root;
747}
748#endif
749
750static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
751{
752 const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
753 const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
754
755 return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
756}
757
758static int global_root_key_cmp(const void *k, const struct rb_node *node)
759{
760 const struct btrfs_key *key = k;
761 const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
762
763 return btrfs_comp_cpu_keys(key, &root->root_key);
764}
765
766int btrfs_global_root_insert(struct btrfs_root *root)
767{
768 struct btrfs_fs_info *fs_info = root->fs_info;
769 struct rb_node *tmp;
770 int ret = 0;
771
772 write_lock(&fs_info->global_root_lock);
773 tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
774 write_unlock(&fs_info->global_root_lock);
775
776 if (tmp) {
777 ret = -EEXIST;
778 btrfs_warn(fs_info, "global root %llu %llu already exists",
779 root->root_key.objectid, root->root_key.offset);
780 }
781 return ret;
782}
783
784void btrfs_global_root_delete(struct btrfs_root *root)
785{
786 struct btrfs_fs_info *fs_info = root->fs_info;
787
788 write_lock(&fs_info->global_root_lock);
789 rb_erase(&root->rb_node, &fs_info->global_root_tree);
790 write_unlock(&fs_info->global_root_lock);
791}
792
793struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
794 struct btrfs_key *key)
795{
796 struct rb_node *node;
797 struct btrfs_root *root = NULL;
798
799 read_lock(&fs_info->global_root_lock);
800 node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
801 if (node)
802 root = container_of(node, struct btrfs_root, rb_node);
803 read_unlock(&fs_info->global_root_lock);
804
805 return root;
806}
807
808static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
809{
810 struct btrfs_block_group *block_group;
811 u64 ret;
812
813 if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
814 return 0;
815
816 if (bytenr)
817 block_group = btrfs_lookup_block_group(fs_info, bytenr);
818 else
819 block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
820 ASSERT(block_group);
821 if (!block_group)
822 return 0;
823 ret = block_group->global_root_id;
824 btrfs_put_block_group(block_group);
825
826 return ret;
827}
828
829struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
830{
831 struct btrfs_key key = {
832 .objectid = BTRFS_CSUM_TREE_OBJECTID,
833 .type = BTRFS_ROOT_ITEM_KEY,
834 .offset = btrfs_global_root_id(fs_info, bytenr),
835 };
836
837 return btrfs_global_root(fs_info, &key);
838}
839
840struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
841{
842 struct btrfs_key key = {
843 .objectid = BTRFS_EXTENT_TREE_OBJECTID,
844 .type = BTRFS_ROOT_ITEM_KEY,
845 .offset = btrfs_global_root_id(fs_info, bytenr),
846 };
847
848 return btrfs_global_root(fs_info, &key);
849}
850
851struct btrfs_root *btrfs_block_group_root(struct btrfs_fs_info *fs_info)
852{
853 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE))
854 return fs_info->block_group_root;
855 return btrfs_extent_root(fs_info, 0);
856}
857
858struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
859 u64 objectid)
860{
861 struct btrfs_fs_info *fs_info = trans->fs_info;
862 struct extent_buffer *leaf;
863 struct btrfs_root *tree_root = fs_info->tree_root;
864 struct btrfs_root *root;
865 struct btrfs_key key;
866 unsigned int nofs_flag;
867 int ret = 0;
868
869 /*
870 * We're holding a transaction handle, so use a NOFS memory allocation
871 * context to avoid deadlock if reclaim happens.
872 */
873 nofs_flag = memalloc_nofs_save();
874 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
875 memalloc_nofs_restore(nofs_flag);
876 if (!root)
877 return ERR_PTR(-ENOMEM);
878
879 root->root_key.objectid = objectid;
880 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
881 root->root_key.offset = 0;
882
883 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
884 0, BTRFS_NESTING_NORMAL);
885 if (IS_ERR(leaf)) {
886 ret = PTR_ERR(leaf);
887 leaf = NULL;
888 goto fail;
889 }
890
891 root->node = leaf;
892 btrfs_mark_buffer_dirty(trans, leaf);
893
894 root->commit_root = btrfs_root_node(root);
895 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
896
897 btrfs_set_root_flags(&root->root_item, 0);
898 btrfs_set_root_limit(&root->root_item, 0);
899 btrfs_set_root_bytenr(&root->root_item, leaf->start);
900 btrfs_set_root_generation(&root->root_item, trans->transid);
901 btrfs_set_root_level(&root->root_item, 0);
902 btrfs_set_root_refs(&root->root_item, 1);
903 btrfs_set_root_used(&root->root_item, leaf->len);
904 btrfs_set_root_last_snapshot(&root->root_item, 0);
905 btrfs_set_root_dirid(&root->root_item, 0);
906 if (is_fstree(objectid))
907 generate_random_guid(root->root_item.uuid);
908 else
909 export_guid(root->root_item.uuid, &guid_null);
910 btrfs_set_root_drop_level(&root->root_item, 0);
911
912 btrfs_tree_unlock(leaf);
913
914 key.objectid = objectid;
915 key.type = BTRFS_ROOT_ITEM_KEY;
916 key.offset = 0;
917 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
918 if (ret)
919 goto fail;
920
921 return root;
922
923fail:
924 btrfs_put_root(root);
925
926 return ERR_PTR(ret);
927}
928
929static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
930 struct btrfs_fs_info *fs_info)
931{
932 struct btrfs_root *root;
933
934 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
935 if (!root)
936 return ERR_PTR(-ENOMEM);
937
938 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
939 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
940 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
941
942 return root;
943}
944
945int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
946 struct btrfs_root *root)
947{
948 struct extent_buffer *leaf;
949
950 /*
951 * DON'T set SHAREABLE bit for log trees.
952 *
953 * Log trees are not exposed to user space thus can't be snapshotted,
954 * and they go away before a real commit is actually done.
955 *
956 * They do store pointers to file data extents, and those reference
957 * counts still get updated (along with back refs to the log tree).
958 */
959
960 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
961 NULL, 0, 0, 0, 0, BTRFS_NESTING_NORMAL);
962 if (IS_ERR(leaf))
963 return PTR_ERR(leaf);
964
965 root->node = leaf;
966
967 btrfs_mark_buffer_dirty(trans, root->node);
968 btrfs_tree_unlock(root->node);
969
970 return 0;
971}
972
973int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
974 struct btrfs_fs_info *fs_info)
975{
976 struct btrfs_root *log_root;
977
978 log_root = alloc_log_tree(trans, fs_info);
979 if (IS_ERR(log_root))
980 return PTR_ERR(log_root);
981
982 if (!btrfs_is_zoned(fs_info)) {
983 int ret = btrfs_alloc_log_tree_node(trans, log_root);
984
985 if (ret) {
986 btrfs_put_root(log_root);
987 return ret;
988 }
989 }
990
991 WARN_ON(fs_info->log_root_tree);
992 fs_info->log_root_tree = log_root;
993 return 0;
994}
995
996int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
997 struct btrfs_root *root)
998{
999 struct btrfs_fs_info *fs_info = root->fs_info;
1000 struct btrfs_root *log_root;
1001 struct btrfs_inode_item *inode_item;
1002 int ret;
1003
1004 log_root = alloc_log_tree(trans, fs_info);
1005 if (IS_ERR(log_root))
1006 return PTR_ERR(log_root);
1007
1008 ret = btrfs_alloc_log_tree_node(trans, log_root);
1009 if (ret) {
1010 btrfs_put_root(log_root);
1011 return ret;
1012 }
1013
1014 log_root->last_trans = trans->transid;
1015 log_root->root_key.offset = root->root_key.objectid;
1016
1017 inode_item = &log_root->root_item.inode;
1018 btrfs_set_stack_inode_generation(inode_item, 1);
1019 btrfs_set_stack_inode_size(inode_item, 3);
1020 btrfs_set_stack_inode_nlink(inode_item, 1);
1021 btrfs_set_stack_inode_nbytes(inode_item,
1022 fs_info->nodesize);
1023 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1024
1025 btrfs_set_root_node(&log_root->root_item, log_root->node);
1026
1027 WARN_ON(root->log_root);
1028 root->log_root = log_root;
1029 btrfs_set_root_log_transid(root, 0);
1030 root->log_transid_committed = -1;
1031 btrfs_set_root_last_log_commit(root, 0);
1032 return 0;
1033}
1034
1035static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1036 struct btrfs_path *path,
1037 struct btrfs_key *key)
1038{
1039 struct btrfs_root *root;
1040 struct btrfs_tree_parent_check check = { 0 };
1041 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1042 u64 generation;
1043 int ret;
1044 int level;
1045
1046 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1047 if (!root)
1048 return ERR_PTR(-ENOMEM);
1049
1050 ret = btrfs_find_root(tree_root, key, path,
1051 &root->root_item, &root->root_key);
1052 if (ret) {
1053 if (ret > 0)
1054 ret = -ENOENT;
1055 goto fail;
1056 }
1057
1058 generation = btrfs_root_generation(&root->root_item);
1059 level = btrfs_root_level(&root->root_item);
1060 check.level = level;
1061 check.transid = generation;
1062 check.owner_root = key->objectid;
1063 root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item),
1064 &check);
1065 if (IS_ERR(root->node)) {
1066 ret = PTR_ERR(root->node);
1067 root->node = NULL;
1068 goto fail;
1069 }
1070 if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1071 ret = -EIO;
1072 goto fail;
1073 }
1074
1075 /*
1076 * For real fs, and not log/reloc trees, root owner must
1077 * match its root node owner
1078 */
1079 if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) &&
1080 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1081 root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1082 root->root_key.objectid != btrfs_header_owner(root->node)) {
1083 btrfs_crit(fs_info,
1084"root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1085 root->root_key.objectid, root->node->start,
1086 btrfs_header_owner(root->node),
1087 root->root_key.objectid);
1088 ret = -EUCLEAN;
1089 goto fail;
1090 }
1091 root->commit_root = btrfs_root_node(root);
1092 return root;
1093fail:
1094 btrfs_put_root(root);
1095 return ERR_PTR(ret);
1096}
1097
1098struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1099 struct btrfs_key *key)
1100{
1101 struct btrfs_root *root;
1102 struct btrfs_path *path;
1103
1104 path = btrfs_alloc_path();
1105 if (!path)
1106 return ERR_PTR(-ENOMEM);
1107 root = read_tree_root_path(tree_root, path, key);
1108 btrfs_free_path(path);
1109
1110 return root;
1111}
1112
1113/*
1114 * Initialize subvolume root in-memory structure
1115 *
1116 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1117 */
1118static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1119{
1120 int ret;
1121
1122 btrfs_drew_lock_init(&root->snapshot_lock);
1123
1124 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1125 !btrfs_is_data_reloc_root(root) &&
1126 is_fstree(root->root_key.objectid)) {
1127 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1128 btrfs_check_and_init_root_item(&root->root_item);
1129 }
1130
1131 /*
1132 * Don't assign anonymous block device to roots that are not exposed to
1133 * userspace, the id pool is limited to 1M
1134 */
1135 if (is_fstree(root->root_key.objectid) &&
1136 btrfs_root_refs(&root->root_item) > 0) {
1137 if (!anon_dev) {
1138 ret = get_anon_bdev(&root->anon_dev);
1139 if (ret)
1140 goto fail;
1141 } else {
1142 root->anon_dev = anon_dev;
1143 }
1144 }
1145
1146 mutex_lock(&root->objectid_mutex);
1147 ret = btrfs_init_root_free_objectid(root);
1148 if (ret) {
1149 mutex_unlock(&root->objectid_mutex);
1150 goto fail;
1151 }
1152
1153 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1154
1155 mutex_unlock(&root->objectid_mutex);
1156
1157 return 0;
1158fail:
1159 /* The caller is responsible to call btrfs_free_fs_root */
1160 return ret;
1161}
1162
1163static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1164 u64 root_id)
1165{
1166 struct btrfs_root *root;
1167
1168 spin_lock(&fs_info->fs_roots_radix_lock);
1169 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1170 (unsigned long)root_id);
1171 root = btrfs_grab_root(root);
1172 spin_unlock(&fs_info->fs_roots_radix_lock);
1173 return root;
1174}
1175
1176static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1177 u64 objectid)
1178{
1179 struct btrfs_key key = {
1180 .objectid = objectid,
1181 .type = BTRFS_ROOT_ITEM_KEY,
1182 .offset = 0,
1183 };
1184
1185 switch (objectid) {
1186 case BTRFS_ROOT_TREE_OBJECTID:
1187 return btrfs_grab_root(fs_info->tree_root);
1188 case BTRFS_EXTENT_TREE_OBJECTID:
1189 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1190 case BTRFS_CHUNK_TREE_OBJECTID:
1191 return btrfs_grab_root(fs_info->chunk_root);
1192 case BTRFS_DEV_TREE_OBJECTID:
1193 return btrfs_grab_root(fs_info->dev_root);
1194 case BTRFS_CSUM_TREE_OBJECTID:
1195 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1196 case BTRFS_QUOTA_TREE_OBJECTID:
1197 return btrfs_grab_root(fs_info->quota_root);
1198 case BTRFS_UUID_TREE_OBJECTID:
1199 return btrfs_grab_root(fs_info->uuid_root);
1200 case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
1201 return btrfs_grab_root(fs_info->block_group_root);
1202 case BTRFS_FREE_SPACE_TREE_OBJECTID:
1203 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1204 case BTRFS_RAID_STRIPE_TREE_OBJECTID:
1205 return btrfs_grab_root(fs_info->stripe_root);
1206 default:
1207 return NULL;
1208 }
1209}
1210
1211int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1212 struct btrfs_root *root)
1213{
1214 int ret;
1215
1216 ret = radix_tree_preload(GFP_NOFS);
1217 if (ret)
1218 return ret;
1219
1220 spin_lock(&fs_info->fs_roots_radix_lock);
1221 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1222 (unsigned long)root->root_key.objectid,
1223 root);
1224 if (ret == 0) {
1225 btrfs_grab_root(root);
1226 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1227 }
1228 spin_unlock(&fs_info->fs_roots_radix_lock);
1229 radix_tree_preload_end();
1230
1231 return ret;
1232}
1233
1234void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1235{
1236#ifdef CONFIG_BTRFS_DEBUG
1237 struct btrfs_root *root;
1238
1239 while (!list_empty(&fs_info->allocated_roots)) {
1240 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1241
1242 root = list_first_entry(&fs_info->allocated_roots,
1243 struct btrfs_root, leak_list);
1244 btrfs_err(fs_info, "leaked root %s refcount %d",
1245 btrfs_root_name(&root->root_key, buf),
1246 refcount_read(&root->refs));
1247 WARN_ON_ONCE(1);
1248 while (refcount_read(&root->refs) > 1)
1249 btrfs_put_root(root);
1250 btrfs_put_root(root);
1251 }
1252#endif
1253}
1254
1255static void free_global_roots(struct btrfs_fs_info *fs_info)
1256{
1257 struct btrfs_root *root;
1258 struct rb_node *node;
1259
1260 while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1261 root = rb_entry(node, struct btrfs_root, rb_node);
1262 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1263 btrfs_put_root(root);
1264 }
1265}
1266
1267void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1268{
1269 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1270 percpu_counter_destroy(&fs_info->delalloc_bytes);
1271 percpu_counter_destroy(&fs_info->ordered_bytes);
1272 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1273 btrfs_free_csum_hash(fs_info);
1274 btrfs_free_stripe_hash_table(fs_info);
1275 btrfs_free_ref_cache(fs_info);
1276 kfree(fs_info->balance_ctl);
1277 kfree(fs_info->delayed_root);
1278 free_global_roots(fs_info);
1279 btrfs_put_root(fs_info->tree_root);
1280 btrfs_put_root(fs_info->chunk_root);
1281 btrfs_put_root(fs_info->dev_root);
1282 btrfs_put_root(fs_info->quota_root);
1283 btrfs_put_root(fs_info->uuid_root);
1284 btrfs_put_root(fs_info->fs_root);
1285 btrfs_put_root(fs_info->data_reloc_root);
1286 btrfs_put_root(fs_info->block_group_root);
1287 btrfs_put_root(fs_info->stripe_root);
1288 btrfs_check_leaked_roots(fs_info);
1289 btrfs_extent_buffer_leak_debug_check(fs_info);
1290 kfree(fs_info->super_copy);
1291 kfree(fs_info->super_for_commit);
1292 kfree(fs_info->subpage_info);
1293 kvfree(fs_info);
1294}
1295
1296
1297/*
1298 * Get an in-memory reference of a root structure.
1299 *
1300 * For essential trees like root/extent tree, we grab it from fs_info directly.
1301 * For subvolume trees, we check the cached filesystem roots first. If not
1302 * found, then read it from disk and add it to cached fs roots.
1303 *
1304 * Caller should release the root by calling btrfs_put_root() after the usage.
1305 *
1306 * NOTE: Reloc and log trees can't be read by this function as they share the
1307 * same root objectid.
1308 *
1309 * @objectid: root id
1310 * @anon_dev: preallocated anonymous block device number for new roots,
1311 * pass NULL for a new allocation.
1312 * @check_ref: whether to check root item references, If true, return -ENOENT
1313 * for orphan roots
1314 */
1315static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1316 u64 objectid, dev_t *anon_dev,
1317 bool check_ref)
1318{
1319 struct btrfs_root *root;
1320 struct btrfs_path *path;
1321 struct btrfs_key key;
1322 int ret;
1323
1324 root = btrfs_get_global_root(fs_info, objectid);
1325 if (root)
1326 return root;
1327
1328 /*
1329 * If we're called for non-subvolume trees, and above function didn't
1330 * find one, do not try to read it from disk.
1331 *
1332 * This is namely for free-space-tree and quota tree, which can change
1333 * at runtime and should only be grabbed from fs_info.
1334 */
1335 if (!is_fstree(objectid) && objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1336 return ERR_PTR(-ENOENT);
1337again:
1338 root = btrfs_lookup_fs_root(fs_info, objectid);
1339 if (root) {
1340 /*
1341 * Some other caller may have read out the newly inserted
1342 * subvolume already (for things like backref walk etc). Not
1343 * that common but still possible. In that case, we just need
1344 * to free the anon_dev.
1345 */
1346 if (unlikely(anon_dev && *anon_dev)) {
1347 free_anon_bdev(*anon_dev);
1348 *anon_dev = 0;
1349 }
1350
1351 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1352 btrfs_put_root(root);
1353 return ERR_PTR(-ENOENT);
1354 }
1355 return root;
1356 }
1357
1358 key.objectid = objectid;
1359 key.type = BTRFS_ROOT_ITEM_KEY;
1360 key.offset = (u64)-1;
1361 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1362 if (IS_ERR(root))
1363 return root;
1364
1365 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1366 ret = -ENOENT;
1367 goto fail;
1368 }
1369
1370 ret = btrfs_init_fs_root(root, anon_dev ? *anon_dev : 0);
1371 if (ret)
1372 goto fail;
1373
1374 path = btrfs_alloc_path();
1375 if (!path) {
1376 ret = -ENOMEM;
1377 goto fail;
1378 }
1379 key.objectid = BTRFS_ORPHAN_OBJECTID;
1380 key.type = BTRFS_ORPHAN_ITEM_KEY;
1381 key.offset = objectid;
1382
1383 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1384 btrfs_free_path(path);
1385 if (ret < 0)
1386 goto fail;
1387 if (ret == 0)
1388 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1389
1390 ret = btrfs_insert_fs_root(fs_info, root);
1391 if (ret) {
1392 if (ret == -EEXIST) {
1393 btrfs_put_root(root);
1394 goto again;
1395 }
1396 goto fail;
1397 }
1398 return root;
1399fail:
1400 /*
1401 * If our caller provided us an anonymous device, then it's his
1402 * responsibility to free it in case we fail. So we have to set our
1403 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1404 * and once again by our caller.
1405 */
1406 if (anon_dev && *anon_dev)
1407 root->anon_dev = 0;
1408 btrfs_put_root(root);
1409 return ERR_PTR(ret);
1410}
1411
1412/*
1413 * Get in-memory reference of a root structure
1414 *
1415 * @objectid: tree objectid
1416 * @check_ref: if set, verify that the tree exists and the item has at least
1417 * one reference
1418 */
1419struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1420 u64 objectid, bool check_ref)
1421{
1422 return btrfs_get_root_ref(fs_info, objectid, NULL, check_ref);
1423}
1424
1425/*
1426 * Get in-memory reference of a root structure, created as new, optionally pass
1427 * the anonymous block device id
1428 *
1429 * @objectid: tree objectid
1430 * @anon_dev: if NULL, allocate a new anonymous block device or use the
1431 * parameter value if not NULL
1432 */
1433struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1434 u64 objectid, dev_t *anon_dev)
1435{
1436 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1437}
1438
1439/*
1440 * Return a root for the given objectid.
1441 *
1442 * @fs_info: the fs_info
1443 * @objectid: the objectid we need to lookup
1444 *
1445 * This is exclusively used for backref walking, and exists specifically because
1446 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1447 * creation time, which means we may have to read the tree_root in order to look
1448 * up a fs root that is not in memory. If the root is not in memory we will
1449 * read the tree root commit root and look up the fs root from there. This is a
1450 * temporary root, it will not be inserted into the radix tree as it doesn't
1451 * have the most uptodate information, it'll simply be discarded once the
1452 * backref code is finished using the root.
1453 */
1454struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1455 struct btrfs_path *path,
1456 u64 objectid)
1457{
1458 struct btrfs_root *root;
1459 struct btrfs_key key;
1460
1461 ASSERT(path->search_commit_root && path->skip_locking);
1462
1463 /*
1464 * This can return -ENOENT if we ask for a root that doesn't exist, but
1465 * since this is called via the backref walking code we won't be looking
1466 * up a root that doesn't exist, unless there's corruption. So if root
1467 * != NULL just return it.
1468 */
1469 root = btrfs_get_global_root(fs_info, objectid);
1470 if (root)
1471 return root;
1472
1473 root = btrfs_lookup_fs_root(fs_info, objectid);
1474 if (root)
1475 return root;
1476
1477 key.objectid = objectid;
1478 key.type = BTRFS_ROOT_ITEM_KEY;
1479 key.offset = (u64)-1;
1480 root = read_tree_root_path(fs_info->tree_root, path, &key);
1481 btrfs_release_path(path);
1482
1483 return root;
1484}
1485
1486static int cleaner_kthread(void *arg)
1487{
1488 struct btrfs_fs_info *fs_info = arg;
1489 int again;
1490
1491 while (1) {
1492 again = 0;
1493
1494 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1495
1496 /* Make the cleaner go to sleep early. */
1497 if (btrfs_need_cleaner_sleep(fs_info))
1498 goto sleep;
1499
1500 /*
1501 * Do not do anything if we might cause open_ctree() to block
1502 * before we have finished mounting the filesystem.
1503 */
1504 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1505 goto sleep;
1506
1507 if (!mutex_trylock(&fs_info->cleaner_mutex))
1508 goto sleep;
1509
1510 /*
1511 * Avoid the problem that we change the status of the fs
1512 * during the above check and trylock.
1513 */
1514 if (btrfs_need_cleaner_sleep(fs_info)) {
1515 mutex_unlock(&fs_info->cleaner_mutex);
1516 goto sleep;
1517 }
1518
1519 if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags))
1520 btrfs_sysfs_feature_update(fs_info);
1521
1522 btrfs_run_delayed_iputs(fs_info);
1523
1524 again = btrfs_clean_one_deleted_snapshot(fs_info);
1525 mutex_unlock(&fs_info->cleaner_mutex);
1526
1527 /*
1528 * The defragger has dealt with the R/O remount and umount,
1529 * needn't do anything special here.
1530 */
1531 btrfs_run_defrag_inodes(fs_info);
1532
1533 /*
1534 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1535 * with relocation (btrfs_relocate_chunk) and relocation
1536 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1537 * after acquiring fs_info->reclaim_bgs_lock. So we
1538 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1539 * unused block groups.
1540 */
1541 btrfs_delete_unused_bgs(fs_info);
1542
1543 /*
1544 * Reclaim block groups in the reclaim_bgs list after we deleted
1545 * all unused block_groups. This possibly gives us some more free
1546 * space.
1547 */
1548 btrfs_reclaim_bgs(fs_info);
1549sleep:
1550 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1551 if (kthread_should_park())
1552 kthread_parkme();
1553 if (kthread_should_stop())
1554 return 0;
1555 if (!again) {
1556 set_current_state(TASK_INTERRUPTIBLE);
1557 schedule();
1558 __set_current_state(TASK_RUNNING);
1559 }
1560 }
1561}
1562
1563static int transaction_kthread(void *arg)
1564{
1565 struct btrfs_root *root = arg;
1566 struct btrfs_fs_info *fs_info = root->fs_info;
1567 struct btrfs_trans_handle *trans;
1568 struct btrfs_transaction *cur;
1569 u64 transid;
1570 time64_t delta;
1571 unsigned long delay;
1572 bool cannot_commit;
1573
1574 do {
1575 cannot_commit = false;
1576 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1577 mutex_lock(&fs_info->transaction_kthread_mutex);
1578
1579 spin_lock(&fs_info->trans_lock);
1580 cur = fs_info->running_transaction;
1581 if (!cur) {
1582 spin_unlock(&fs_info->trans_lock);
1583 goto sleep;
1584 }
1585
1586 delta = ktime_get_seconds() - cur->start_time;
1587 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1588 cur->state < TRANS_STATE_COMMIT_PREP &&
1589 delta < fs_info->commit_interval) {
1590 spin_unlock(&fs_info->trans_lock);
1591 delay -= msecs_to_jiffies((delta - 1) * 1000);
1592 delay = min(delay,
1593 msecs_to_jiffies(fs_info->commit_interval * 1000));
1594 goto sleep;
1595 }
1596 transid = cur->transid;
1597 spin_unlock(&fs_info->trans_lock);
1598
1599 /* If the file system is aborted, this will always fail. */
1600 trans = btrfs_attach_transaction(root);
1601 if (IS_ERR(trans)) {
1602 if (PTR_ERR(trans) != -ENOENT)
1603 cannot_commit = true;
1604 goto sleep;
1605 }
1606 if (transid == trans->transid) {
1607 btrfs_commit_transaction(trans);
1608 } else {
1609 btrfs_end_transaction(trans);
1610 }
1611sleep:
1612 wake_up_process(fs_info->cleaner_kthread);
1613 mutex_unlock(&fs_info->transaction_kthread_mutex);
1614
1615 if (BTRFS_FS_ERROR(fs_info))
1616 btrfs_cleanup_transaction(fs_info);
1617 if (!kthread_should_stop() &&
1618 (!btrfs_transaction_blocked(fs_info) ||
1619 cannot_commit))
1620 schedule_timeout_interruptible(delay);
1621 } while (!kthread_should_stop());
1622 return 0;
1623}
1624
1625/*
1626 * This will find the highest generation in the array of root backups. The
1627 * index of the highest array is returned, or -EINVAL if we can't find
1628 * anything.
1629 *
1630 * We check to make sure the array is valid by comparing the
1631 * generation of the latest root in the array with the generation
1632 * in the super block. If they don't match we pitch it.
1633 */
1634static int find_newest_super_backup(struct btrfs_fs_info *info)
1635{
1636 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1637 u64 cur;
1638 struct btrfs_root_backup *root_backup;
1639 int i;
1640
1641 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1642 root_backup = info->super_copy->super_roots + i;
1643 cur = btrfs_backup_tree_root_gen(root_backup);
1644 if (cur == newest_gen)
1645 return i;
1646 }
1647
1648 return -EINVAL;
1649}
1650
1651/*
1652 * copy all the root pointers into the super backup array.
1653 * this will bump the backup pointer by one when it is
1654 * done
1655 */
1656static void backup_super_roots(struct btrfs_fs_info *info)
1657{
1658 const int next_backup = info->backup_root_index;
1659 struct btrfs_root_backup *root_backup;
1660
1661 root_backup = info->super_for_commit->super_roots + next_backup;
1662
1663 /*
1664 * make sure all of our padding and empty slots get zero filled
1665 * regardless of which ones we use today
1666 */
1667 memset(root_backup, 0, sizeof(*root_backup));
1668
1669 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1670
1671 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1672 btrfs_set_backup_tree_root_gen(root_backup,
1673 btrfs_header_generation(info->tree_root->node));
1674
1675 btrfs_set_backup_tree_root_level(root_backup,
1676 btrfs_header_level(info->tree_root->node));
1677
1678 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1679 btrfs_set_backup_chunk_root_gen(root_backup,
1680 btrfs_header_generation(info->chunk_root->node));
1681 btrfs_set_backup_chunk_root_level(root_backup,
1682 btrfs_header_level(info->chunk_root->node));
1683
1684 if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
1685 struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
1686 struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
1687
1688 btrfs_set_backup_extent_root(root_backup,
1689 extent_root->node->start);
1690 btrfs_set_backup_extent_root_gen(root_backup,
1691 btrfs_header_generation(extent_root->node));
1692 btrfs_set_backup_extent_root_level(root_backup,
1693 btrfs_header_level(extent_root->node));
1694
1695 btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
1696 btrfs_set_backup_csum_root_gen(root_backup,
1697 btrfs_header_generation(csum_root->node));
1698 btrfs_set_backup_csum_root_level(root_backup,
1699 btrfs_header_level(csum_root->node));
1700 }
1701
1702 /*
1703 * we might commit during log recovery, which happens before we set
1704 * the fs_root. Make sure it is valid before we fill it in.
1705 */
1706 if (info->fs_root && info->fs_root->node) {
1707 btrfs_set_backup_fs_root(root_backup,
1708 info->fs_root->node->start);
1709 btrfs_set_backup_fs_root_gen(root_backup,
1710 btrfs_header_generation(info->fs_root->node));
1711 btrfs_set_backup_fs_root_level(root_backup,
1712 btrfs_header_level(info->fs_root->node));
1713 }
1714
1715 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1716 btrfs_set_backup_dev_root_gen(root_backup,
1717 btrfs_header_generation(info->dev_root->node));
1718 btrfs_set_backup_dev_root_level(root_backup,
1719 btrfs_header_level(info->dev_root->node));
1720
1721 btrfs_set_backup_total_bytes(root_backup,
1722 btrfs_super_total_bytes(info->super_copy));
1723 btrfs_set_backup_bytes_used(root_backup,
1724 btrfs_super_bytes_used(info->super_copy));
1725 btrfs_set_backup_num_devices(root_backup,
1726 btrfs_super_num_devices(info->super_copy));
1727
1728 /*
1729 * if we don't copy this out to the super_copy, it won't get remembered
1730 * for the next commit
1731 */
1732 memcpy(&info->super_copy->super_roots,
1733 &info->super_for_commit->super_roots,
1734 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1735}
1736
1737/*
1738 * Reads a backup root based on the passed priority. Prio 0 is the newest, prio
1739 * 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1740 *
1741 * @fs_info: filesystem whose backup roots need to be read
1742 * @priority: priority of backup root required
1743 *
1744 * Returns backup root index on success and -EINVAL otherwise.
1745 */
1746static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1747{
1748 int backup_index = find_newest_super_backup(fs_info);
1749 struct btrfs_super_block *super = fs_info->super_copy;
1750 struct btrfs_root_backup *root_backup;
1751
1752 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1753 if (priority == 0)
1754 return backup_index;
1755
1756 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1757 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1758 } else {
1759 return -EINVAL;
1760 }
1761
1762 root_backup = super->super_roots + backup_index;
1763
1764 btrfs_set_super_generation(super,
1765 btrfs_backup_tree_root_gen(root_backup));
1766 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1767 btrfs_set_super_root_level(super,
1768 btrfs_backup_tree_root_level(root_backup));
1769 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1770
1771 /*
1772 * Fixme: the total bytes and num_devices need to match or we should
1773 * need a fsck
1774 */
1775 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1776 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1777
1778 return backup_index;
1779}
1780
1781/* helper to cleanup workers */
1782static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1783{
1784 btrfs_destroy_workqueue(fs_info->fixup_workers);
1785 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1786 btrfs_destroy_workqueue(fs_info->workers);
1787 if (fs_info->endio_workers)
1788 destroy_workqueue(fs_info->endio_workers);
1789 if (fs_info->rmw_workers)
1790 destroy_workqueue(fs_info->rmw_workers);
1791 if (fs_info->compressed_write_workers)
1792 destroy_workqueue(fs_info->compressed_write_workers);
1793 btrfs_destroy_workqueue(fs_info->endio_write_workers);
1794 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1795 btrfs_destroy_workqueue(fs_info->delayed_workers);
1796 btrfs_destroy_workqueue(fs_info->caching_workers);
1797 btrfs_destroy_workqueue(fs_info->flush_workers);
1798 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1799 if (fs_info->discard_ctl.discard_workers)
1800 destroy_workqueue(fs_info->discard_ctl.discard_workers);
1801 /*
1802 * Now that all other work queues are destroyed, we can safely destroy
1803 * the queues used for metadata I/O, since tasks from those other work
1804 * queues can do metadata I/O operations.
1805 */
1806 if (fs_info->endio_meta_workers)
1807 destroy_workqueue(fs_info->endio_meta_workers);
1808}
1809
1810static void free_root_extent_buffers(struct btrfs_root *root)
1811{
1812 if (root) {
1813 free_extent_buffer(root->node);
1814 free_extent_buffer(root->commit_root);
1815 root->node = NULL;
1816 root->commit_root = NULL;
1817 }
1818}
1819
1820static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
1821{
1822 struct btrfs_root *root, *tmp;
1823
1824 rbtree_postorder_for_each_entry_safe(root, tmp,
1825 &fs_info->global_root_tree,
1826 rb_node)
1827 free_root_extent_buffers(root);
1828}
1829
1830/* helper to cleanup tree roots */
1831static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1832{
1833 free_root_extent_buffers(info->tree_root);
1834
1835 free_global_root_pointers(info);
1836 free_root_extent_buffers(info->dev_root);
1837 free_root_extent_buffers(info->quota_root);
1838 free_root_extent_buffers(info->uuid_root);
1839 free_root_extent_buffers(info->fs_root);
1840 free_root_extent_buffers(info->data_reloc_root);
1841 free_root_extent_buffers(info->block_group_root);
1842 free_root_extent_buffers(info->stripe_root);
1843 if (free_chunk_root)
1844 free_root_extent_buffers(info->chunk_root);
1845}
1846
1847void btrfs_put_root(struct btrfs_root *root)
1848{
1849 if (!root)
1850 return;
1851
1852 if (refcount_dec_and_test(&root->refs)) {
1853 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1854 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
1855 if (root->anon_dev)
1856 free_anon_bdev(root->anon_dev);
1857 free_root_extent_buffers(root);
1858#ifdef CONFIG_BTRFS_DEBUG
1859 spin_lock(&root->fs_info->fs_roots_radix_lock);
1860 list_del_init(&root->leak_list);
1861 spin_unlock(&root->fs_info->fs_roots_radix_lock);
1862#endif
1863 kfree(root);
1864 }
1865}
1866
1867void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
1868{
1869 int ret;
1870 struct btrfs_root *gang[8];
1871 int i;
1872
1873 while (!list_empty(&fs_info->dead_roots)) {
1874 gang[0] = list_entry(fs_info->dead_roots.next,
1875 struct btrfs_root, root_list);
1876 list_del(&gang[0]->root_list);
1877
1878 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
1879 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
1880 btrfs_put_root(gang[0]);
1881 }
1882
1883 while (1) {
1884 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1885 (void **)gang, 0,
1886 ARRAY_SIZE(gang));
1887 if (!ret)
1888 break;
1889 for (i = 0; i < ret; i++)
1890 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
1891 }
1892}
1893
1894static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
1895{
1896 mutex_init(&fs_info->scrub_lock);
1897 atomic_set(&fs_info->scrubs_running, 0);
1898 atomic_set(&fs_info->scrub_pause_req, 0);
1899 atomic_set(&fs_info->scrubs_paused, 0);
1900 atomic_set(&fs_info->scrub_cancel_req, 0);
1901 init_waitqueue_head(&fs_info->scrub_pause_wait);
1902 refcount_set(&fs_info->scrub_workers_refcnt, 0);
1903}
1904
1905static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
1906{
1907 spin_lock_init(&fs_info->balance_lock);
1908 mutex_init(&fs_info->balance_mutex);
1909 atomic_set(&fs_info->balance_pause_req, 0);
1910 atomic_set(&fs_info->balance_cancel_req, 0);
1911 fs_info->balance_ctl = NULL;
1912 init_waitqueue_head(&fs_info->balance_wait_q);
1913 atomic_set(&fs_info->reloc_cancel_req, 0);
1914}
1915
1916static int btrfs_init_btree_inode(struct super_block *sb)
1917{
1918 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1919 unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
1920 fs_info->tree_root);
1921 struct inode *inode;
1922
1923 inode = new_inode(sb);
1924 if (!inode)
1925 return -ENOMEM;
1926
1927 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1928 set_nlink(inode, 1);
1929 /*
1930 * we set the i_size on the btree inode to the max possible int.
1931 * the real end of the address space is determined by all of
1932 * the devices in the system
1933 */
1934 inode->i_size = OFFSET_MAX;
1935 inode->i_mapping->a_ops = &btree_aops;
1936 mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
1937
1938 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
1939 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
1940 IO_TREE_BTREE_INODE_IO);
1941 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
1942
1943 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
1944 BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID;
1945 BTRFS_I(inode)->location.type = 0;
1946 BTRFS_I(inode)->location.offset = 0;
1947 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
1948 __insert_inode_hash(inode, hash);
1949 fs_info->btree_inode = inode;
1950
1951 return 0;
1952}
1953
1954static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
1955{
1956 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
1957 init_rwsem(&fs_info->dev_replace.rwsem);
1958 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
1959}
1960
1961static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
1962{
1963 spin_lock_init(&fs_info->qgroup_lock);
1964 mutex_init(&fs_info->qgroup_ioctl_lock);
1965 fs_info->qgroup_tree = RB_ROOT;
1966 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
1967 fs_info->qgroup_seq = 1;
1968 fs_info->qgroup_ulist = NULL;
1969 fs_info->qgroup_rescan_running = false;
1970 fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
1971 mutex_init(&fs_info->qgroup_rescan_lock);
1972}
1973
1974static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
1975{
1976 u32 max_active = fs_info->thread_pool_size;
1977 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
1978 unsigned int ordered_flags = WQ_MEM_RECLAIM | WQ_FREEZABLE;
1979
1980 fs_info->workers =
1981 btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
1982
1983 fs_info->delalloc_workers =
1984 btrfs_alloc_workqueue(fs_info, "delalloc",
1985 flags, max_active, 2);
1986
1987 fs_info->flush_workers =
1988 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
1989 flags, max_active, 0);
1990
1991 fs_info->caching_workers =
1992 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
1993
1994 fs_info->fixup_workers =
1995 btrfs_alloc_ordered_workqueue(fs_info, "fixup", ordered_flags);
1996
1997 fs_info->endio_workers =
1998 alloc_workqueue("btrfs-endio", flags, max_active);
1999 fs_info->endio_meta_workers =
2000 alloc_workqueue("btrfs-endio-meta", flags, max_active);
2001 fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
2002 fs_info->endio_write_workers =
2003 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2004 max_active, 2);
2005 fs_info->compressed_write_workers =
2006 alloc_workqueue("btrfs-compressed-write", flags, max_active);
2007 fs_info->endio_freespace_worker =
2008 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2009 max_active, 0);
2010 fs_info->delayed_workers =
2011 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2012 max_active, 0);
2013 fs_info->qgroup_rescan_workers =
2014 btrfs_alloc_ordered_workqueue(fs_info, "qgroup-rescan",
2015 ordered_flags);
2016 fs_info->discard_ctl.discard_workers =
2017 alloc_ordered_workqueue("btrfs_discard", WQ_FREEZABLE);
2018
2019 if (!(fs_info->workers &&
2020 fs_info->delalloc_workers && fs_info->flush_workers &&
2021 fs_info->endio_workers && fs_info->endio_meta_workers &&
2022 fs_info->compressed_write_workers &&
2023 fs_info->endio_write_workers &&
2024 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2025 fs_info->caching_workers && fs_info->fixup_workers &&
2026 fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2027 fs_info->discard_ctl.discard_workers)) {
2028 return -ENOMEM;
2029 }
2030
2031 return 0;
2032}
2033
2034static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2035{
2036 struct crypto_shash *csum_shash;
2037 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2038
2039 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2040
2041 if (IS_ERR(csum_shash)) {
2042 btrfs_err(fs_info, "error allocating %s hash for checksum",
2043 csum_driver);
2044 return PTR_ERR(csum_shash);
2045 }
2046
2047 fs_info->csum_shash = csum_shash;
2048
2049 /*
2050 * Check if the checksum implementation is a fast accelerated one.
2051 * As-is this is a bit of a hack and should be replaced once the csum
2052 * implementations provide that information themselves.
2053 */
2054 switch (csum_type) {
2055 case BTRFS_CSUM_TYPE_CRC32:
2056 if (!strstr(crypto_shash_driver_name(csum_shash), "generic"))
2057 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2058 break;
2059 case BTRFS_CSUM_TYPE_XXHASH:
2060 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2061 break;
2062 default:
2063 break;
2064 }
2065
2066 btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2067 btrfs_super_csum_name(csum_type),
2068 crypto_shash_driver_name(csum_shash));
2069 return 0;
2070}
2071
2072static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2073 struct btrfs_fs_devices *fs_devices)
2074{
2075 int ret;
2076 struct btrfs_tree_parent_check check = { 0 };
2077 struct btrfs_root *log_tree_root;
2078 struct btrfs_super_block *disk_super = fs_info->super_copy;
2079 u64 bytenr = btrfs_super_log_root(disk_super);
2080 int level = btrfs_super_log_root_level(disk_super);
2081
2082 if (fs_devices->rw_devices == 0) {
2083 btrfs_warn(fs_info, "log replay required on RO media");
2084 return -EIO;
2085 }
2086
2087 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2088 GFP_KERNEL);
2089 if (!log_tree_root)
2090 return -ENOMEM;
2091
2092 check.level = level;
2093 check.transid = fs_info->generation + 1;
2094 check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2095 log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
2096 if (IS_ERR(log_tree_root->node)) {
2097 btrfs_warn(fs_info, "failed to read log tree");
2098 ret = PTR_ERR(log_tree_root->node);
2099 log_tree_root->node = NULL;
2100 btrfs_put_root(log_tree_root);
2101 return ret;
2102 }
2103 if (!extent_buffer_uptodate(log_tree_root->node)) {
2104 btrfs_err(fs_info, "failed to read log tree");
2105 btrfs_put_root(log_tree_root);
2106 return -EIO;
2107 }
2108
2109 /* returns with log_tree_root freed on success */
2110 ret = btrfs_recover_log_trees(log_tree_root);
2111 if (ret) {
2112 btrfs_handle_fs_error(fs_info, ret,
2113 "Failed to recover log tree");
2114 btrfs_put_root(log_tree_root);
2115 return ret;
2116 }
2117
2118 if (sb_rdonly(fs_info->sb)) {
2119 ret = btrfs_commit_super(fs_info);
2120 if (ret)
2121 return ret;
2122 }
2123
2124 return 0;
2125}
2126
2127static int load_global_roots_objectid(struct btrfs_root *tree_root,
2128 struct btrfs_path *path, u64 objectid,
2129 const char *name)
2130{
2131 struct btrfs_fs_info *fs_info = tree_root->fs_info;
2132 struct btrfs_root *root;
2133 u64 max_global_id = 0;
2134 int ret;
2135 struct btrfs_key key = {
2136 .objectid = objectid,
2137 .type = BTRFS_ROOT_ITEM_KEY,
2138 .offset = 0,
2139 };
2140 bool found = false;
2141
2142 /* If we have IGNOREDATACSUMS skip loading these roots. */
2143 if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2144 btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2145 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2146 return 0;
2147 }
2148
2149 while (1) {
2150 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2151 if (ret < 0)
2152 break;
2153
2154 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2155 ret = btrfs_next_leaf(tree_root, path);
2156 if (ret) {
2157 if (ret > 0)
2158 ret = 0;
2159 break;
2160 }
2161 }
2162 ret = 0;
2163
2164 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2165 if (key.objectid != objectid)
2166 break;
2167 btrfs_release_path(path);
2168
2169 /*
2170 * Just worry about this for extent tree, it'll be the same for
2171 * everybody.
2172 */
2173 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2174 max_global_id = max(max_global_id, key.offset);
2175
2176 found = true;
2177 root = read_tree_root_path(tree_root, path, &key);
2178 if (IS_ERR(root)) {
2179 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2180 ret = PTR_ERR(root);
2181 break;
2182 }
2183 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2184 ret = btrfs_global_root_insert(root);
2185 if (ret) {
2186 btrfs_put_root(root);
2187 break;
2188 }
2189 key.offset++;
2190 }
2191 btrfs_release_path(path);
2192
2193 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2194 fs_info->nr_global_roots = max_global_id + 1;
2195
2196 if (!found || ret) {
2197 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2198 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2199
2200 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2201 ret = ret ? ret : -ENOENT;
2202 else
2203 ret = 0;
2204 btrfs_err(fs_info, "failed to load root %s", name);
2205 }
2206 return ret;
2207}
2208
2209static int load_global_roots(struct btrfs_root *tree_root)
2210{
2211 struct btrfs_path *path;
2212 int ret = 0;
2213
2214 path = btrfs_alloc_path();
2215 if (!path)
2216 return -ENOMEM;
2217
2218 ret = load_global_roots_objectid(tree_root, path,
2219 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2220 if (ret)
2221 goto out;
2222 ret = load_global_roots_objectid(tree_root, path,
2223 BTRFS_CSUM_TREE_OBJECTID, "csum");
2224 if (ret)
2225 goto out;
2226 if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2227 goto out;
2228 ret = load_global_roots_objectid(tree_root, path,
2229 BTRFS_FREE_SPACE_TREE_OBJECTID,
2230 "free space");
2231out:
2232 btrfs_free_path(path);
2233 return ret;
2234}
2235
2236static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2237{
2238 struct btrfs_root *tree_root = fs_info->tree_root;
2239 struct btrfs_root *root;
2240 struct btrfs_key location;
2241 int ret;
2242
2243 ASSERT(fs_info->tree_root);
2244
2245 ret = load_global_roots(tree_root);
2246 if (ret)
2247 return ret;
2248
2249 location.type = BTRFS_ROOT_ITEM_KEY;
2250 location.offset = 0;
2251
2252 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2253 location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2254 root = btrfs_read_tree_root(tree_root, &location);
2255 if (IS_ERR(root)) {
2256 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2257 ret = PTR_ERR(root);
2258 goto out;
2259 }
2260 } else {
2261 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2262 fs_info->block_group_root = root;
2263 }
2264 }
2265
2266 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2267 root = btrfs_read_tree_root(tree_root, &location);
2268 if (IS_ERR(root)) {
2269 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2270 ret = PTR_ERR(root);
2271 goto out;
2272 }
2273 } else {
2274 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2275 fs_info->dev_root = root;
2276 }
2277 /* Initialize fs_info for all devices in any case */
2278 ret = btrfs_init_devices_late(fs_info);
2279 if (ret)
2280 goto out;
2281
2282 /*
2283 * This tree can share blocks with some other fs tree during relocation
2284 * and we need a proper setup by btrfs_get_fs_root
2285 */
2286 root = btrfs_get_fs_root(tree_root->fs_info,
2287 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2288 if (IS_ERR(root)) {
2289 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2290 ret = PTR_ERR(root);
2291 goto out;
2292 }
2293 } else {
2294 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2295 fs_info->data_reloc_root = root;
2296 }
2297
2298 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2299 root = btrfs_read_tree_root(tree_root, &location);
2300 if (!IS_ERR(root)) {
2301 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2302 fs_info->quota_root = root;
2303 }
2304
2305 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2306 root = btrfs_read_tree_root(tree_root, &location);
2307 if (IS_ERR(root)) {
2308 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2309 ret = PTR_ERR(root);
2310 if (ret != -ENOENT)
2311 goto out;
2312 }
2313 } else {
2314 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2315 fs_info->uuid_root = root;
2316 }
2317
2318 if (btrfs_fs_incompat(fs_info, RAID_STRIPE_TREE)) {
2319 location.objectid = BTRFS_RAID_STRIPE_TREE_OBJECTID;
2320 root = btrfs_read_tree_root(tree_root, &location);
2321 if (IS_ERR(root)) {
2322 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2323 ret = PTR_ERR(root);
2324 goto out;
2325 }
2326 } else {
2327 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2328 fs_info->stripe_root = root;
2329 }
2330 }
2331
2332 return 0;
2333out:
2334 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2335 location.objectid, ret);
2336 return ret;
2337}
2338
2339/*
2340 * Real super block validation
2341 * NOTE: super csum type and incompat features will not be checked here.
2342 *
2343 * @sb: super block to check
2344 * @mirror_num: the super block number to check its bytenr:
2345 * 0 the primary (1st) sb
2346 * 1, 2 2nd and 3rd backup copy
2347 * -1 skip bytenr check
2348 */
2349int btrfs_validate_super(struct btrfs_fs_info *fs_info,
2350 struct btrfs_super_block *sb, int mirror_num)
2351{
2352 u64 nodesize = btrfs_super_nodesize(sb);
2353 u64 sectorsize = btrfs_super_sectorsize(sb);
2354 int ret = 0;
2355
2356 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2357 btrfs_err(fs_info, "no valid FS found");
2358 ret = -EINVAL;
2359 }
2360 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2361 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2362 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2363 ret = -EINVAL;
2364 }
2365 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2366 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2367 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2368 ret = -EINVAL;
2369 }
2370 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2371 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2372 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2373 ret = -EINVAL;
2374 }
2375 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2376 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2377 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2378 ret = -EINVAL;
2379 }
2380
2381 /*
2382 * Check sectorsize and nodesize first, other check will need it.
2383 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2384 */
2385 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2386 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2387 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2388 ret = -EINVAL;
2389 }
2390
2391 /*
2392 * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2393 *
2394 * We can support 16K sectorsize with 64K page size without problem,
2395 * but such sectorsize/pagesize combination doesn't make much sense.
2396 * 4K will be our future standard, PAGE_SIZE is supported from the very
2397 * beginning.
2398 */
2399 if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2400 btrfs_err(fs_info,
2401 "sectorsize %llu not yet supported for page size %lu",
2402 sectorsize, PAGE_SIZE);
2403 ret = -EINVAL;
2404 }
2405
2406 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2407 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2408 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2409 ret = -EINVAL;
2410 }
2411 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2412 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2413 le32_to_cpu(sb->__unused_leafsize), nodesize);
2414 ret = -EINVAL;
2415 }
2416
2417 /* Root alignment check */
2418 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2419 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2420 btrfs_super_root(sb));
2421 ret = -EINVAL;
2422 }
2423 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2424 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2425 btrfs_super_chunk_root(sb));
2426 ret = -EINVAL;
2427 }
2428 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2429 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2430 btrfs_super_log_root(sb));
2431 ret = -EINVAL;
2432 }
2433
2434 if (!fs_info->fs_devices->temp_fsid &&
2435 memcmp(fs_info->fs_devices->fsid, sb->fsid, BTRFS_FSID_SIZE) != 0) {
2436 btrfs_err(fs_info,
2437 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2438 sb->fsid, fs_info->fs_devices->fsid);
2439 ret = -EINVAL;
2440 }
2441
2442 if (memcmp(fs_info->fs_devices->metadata_uuid, btrfs_sb_fsid_ptr(sb),
2443 BTRFS_FSID_SIZE) != 0) {
2444 btrfs_err(fs_info,
2445"superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2446 btrfs_sb_fsid_ptr(sb), fs_info->fs_devices->metadata_uuid);
2447 ret = -EINVAL;
2448 }
2449
2450 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2451 BTRFS_FSID_SIZE) != 0) {
2452 btrfs_err(fs_info,
2453 "dev_item UUID does not match metadata fsid: %pU != %pU",
2454 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2455 ret = -EINVAL;
2456 }
2457
2458 /*
2459 * Artificial requirement for block-group-tree to force newer features
2460 * (free-space-tree, no-holes) so the test matrix is smaller.
2461 */
2462 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2463 (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2464 !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2465 btrfs_err(fs_info,
2466 "block-group-tree feature requires fres-space-tree and no-holes");
2467 ret = -EINVAL;
2468 }
2469
2470 /*
2471 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2472 * done later
2473 */
2474 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2475 btrfs_err(fs_info, "bytes_used is too small %llu",
2476 btrfs_super_bytes_used(sb));
2477 ret = -EINVAL;
2478 }
2479 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2480 btrfs_err(fs_info, "invalid stripesize %u",
2481 btrfs_super_stripesize(sb));
2482 ret = -EINVAL;
2483 }
2484 if (btrfs_super_num_devices(sb) > (1UL << 31))
2485 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2486 btrfs_super_num_devices(sb));
2487 if (btrfs_super_num_devices(sb) == 0) {
2488 btrfs_err(fs_info, "number of devices is 0");
2489 ret = -EINVAL;
2490 }
2491
2492 if (mirror_num >= 0 &&
2493 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2494 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2495 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2496 ret = -EINVAL;
2497 }
2498
2499 /*
2500 * Obvious sys_chunk_array corruptions, it must hold at least one key
2501 * and one chunk
2502 */
2503 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2504 btrfs_err(fs_info, "system chunk array too big %u > %u",
2505 btrfs_super_sys_array_size(sb),
2506 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2507 ret = -EINVAL;
2508 }
2509 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2510 + sizeof(struct btrfs_chunk)) {
2511 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2512 btrfs_super_sys_array_size(sb),
2513 sizeof(struct btrfs_disk_key)
2514 + sizeof(struct btrfs_chunk));
2515 ret = -EINVAL;
2516 }
2517
2518 /*
2519 * The generation is a global counter, we'll trust it more than the others
2520 * but it's still possible that it's the one that's wrong.
2521 */
2522 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2523 btrfs_warn(fs_info,
2524 "suspicious: generation < chunk_root_generation: %llu < %llu",
2525 btrfs_super_generation(sb),
2526 btrfs_super_chunk_root_generation(sb));
2527 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2528 && btrfs_super_cache_generation(sb) != (u64)-1)
2529 btrfs_warn(fs_info,
2530 "suspicious: generation < cache_generation: %llu < %llu",
2531 btrfs_super_generation(sb),
2532 btrfs_super_cache_generation(sb));
2533
2534 return ret;
2535}
2536
2537/*
2538 * Validation of super block at mount time.
2539 * Some checks already done early at mount time, like csum type and incompat
2540 * flags will be skipped.
2541 */
2542static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2543{
2544 return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2545}
2546
2547/*
2548 * Validation of super block at write time.
2549 * Some checks like bytenr check will be skipped as their values will be
2550 * overwritten soon.
2551 * Extra checks like csum type and incompat flags will be done here.
2552 */
2553static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2554 struct btrfs_super_block *sb)
2555{
2556 int ret;
2557
2558 ret = btrfs_validate_super(fs_info, sb, -1);
2559 if (ret < 0)
2560 goto out;
2561 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2562 ret = -EUCLEAN;
2563 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2564 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2565 goto out;
2566 }
2567 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2568 ret = -EUCLEAN;
2569 btrfs_err(fs_info,
2570 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2571 btrfs_super_incompat_flags(sb),
2572 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2573 goto out;
2574 }
2575out:
2576 if (ret < 0)
2577 btrfs_err(fs_info,
2578 "super block corruption detected before writing it to disk");
2579 return ret;
2580}
2581
2582static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2583{
2584 struct btrfs_tree_parent_check check = {
2585 .level = level,
2586 .transid = gen,
2587 .owner_root = root->root_key.objectid
2588 };
2589 int ret = 0;
2590
2591 root->node = read_tree_block(root->fs_info, bytenr, &check);
2592 if (IS_ERR(root->node)) {
2593 ret = PTR_ERR(root->node);
2594 root->node = NULL;
2595 return ret;
2596 }
2597 if (!extent_buffer_uptodate(root->node)) {
2598 free_extent_buffer(root->node);
2599 root->node = NULL;
2600 return -EIO;
2601 }
2602
2603 btrfs_set_root_node(&root->root_item, root->node);
2604 root->commit_root = btrfs_root_node(root);
2605 btrfs_set_root_refs(&root->root_item, 1);
2606 return ret;
2607}
2608
2609static int load_important_roots(struct btrfs_fs_info *fs_info)
2610{
2611 struct btrfs_super_block *sb = fs_info->super_copy;
2612 u64 gen, bytenr;
2613 int level, ret;
2614
2615 bytenr = btrfs_super_root(sb);
2616 gen = btrfs_super_generation(sb);
2617 level = btrfs_super_root_level(sb);
2618 ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2619 if (ret) {
2620 btrfs_warn(fs_info, "couldn't read tree root");
2621 return ret;
2622 }
2623 return 0;
2624}
2625
2626static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2627{
2628 int backup_index = find_newest_super_backup(fs_info);
2629 struct btrfs_super_block *sb = fs_info->super_copy;
2630 struct btrfs_root *tree_root = fs_info->tree_root;
2631 bool handle_error = false;
2632 int ret = 0;
2633 int i;
2634
2635 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2636 if (handle_error) {
2637 if (!IS_ERR(tree_root->node))
2638 free_extent_buffer(tree_root->node);
2639 tree_root->node = NULL;
2640
2641 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2642 break;
2643
2644 free_root_pointers(fs_info, 0);
2645
2646 /*
2647 * Don't use the log in recovery mode, it won't be
2648 * valid
2649 */
2650 btrfs_set_super_log_root(sb, 0);
2651
2652 btrfs_warn(fs_info, "try to load backup roots slot %d", i);
2653 ret = read_backup_root(fs_info, i);
2654 backup_index = ret;
2655 if (ret < 0)
2656 return ret;
2657 }
2658
2659 ret = load_important_roots(fs_info);
2660 if (ret) {
2661 handle_error = true;
2662 continue;
2663 }
2664
2665 /*
2666 * No need to hold btrfs_root::objectid_mutex since the fs
2667 * hasn't been fully initialised and we are the only user
2668 */
2669 ret = btrfs_init_root_free_objectid(tree_root);
2670 if (ret < 0) {
2671 handle_error = true;
2672 continue;
2673 }
2674
2675 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2676
2677 ret = btrfs_read_roots(fs_info);
2678 if (ret < 0) {
2679 handle_error = true;
2680 continue;
2681 }
2682
2683 /* All successful */
2684 fs_info->generation = btrfs_header_generation(tree_root->node);
2685 btrfs_set_last_trans_committed(fs_info, fs_info->generation);
2686 fs_info->last_reloc_trans = 0;
2687
2688 /* Always begin writing backup roots after the one being used */
2689 if (backup_index < 0) {
2690 fs_info->backup_root_index = 0;
2691 } else {
2692 fs_info->backup_root_index = backup_index + 1;
2693 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2694 }
2695 break;
2696 }
2697
2698 return ret;
2699}
2700
2701void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2702{
2703 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2704 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2705 INIT_LIST_HEAD(&fs_info->trans_list);
2706 INIT_LIST_HEAD(&fs_info->dead_roots);
2707 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2708 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2709 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2710 spin_lock_init(&fs_info->delalloc_root_lock);
2711 spin_lock_init(&fs_info->trans_lock);
2712 spin_lock_init(&fs_info->fs_roots_radix_lock);
2713 spin_lock_init(&fs_info->delayed_iput_lock);
2714 spin_lock_init(&fs_info->defrag_inodes_lock);
2715 spin_lock_init(&fs_info->super_lock);
2716 spin_lock_init(&fs_info->buffer_lock);
2717 spin_lock_init(&fs_info->unused_bgs_lock);
2718 spin_lock_init(&fs_info->treelog_bg_lock);
2719 spin_lock_init(&fs_info->zone_active_bgs_lock);
2720 spin_lock_init(&fs_info->relocation_bg_lock);
2721 rwlock_init(&fs_info->tree_mod_log_lock);
2722 rwlock_init(&fs_info->global_root_lock);
2723 mutex_init(&fs_info->unused_bg_unpin_mutex);
2724 mutex_init(&fs_info->reclaim_bgs_lock);
2725 mutex_init(&fs_info->reloc_mutex);
2726 mutex_init(&fs_info->delalloc_root_mutex);
2727 mutex_init(&fs_info->zoned_meta_io_lock);
2728 mutex_init(&fs_info->zoned_data_reloc_io_lock);
2729 seqlock_init(&fs_info->profiles_lock);
2730
2731 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
2732 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
2733 btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
2734 btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
2735 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_prep,
2736 BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2737 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
2738 BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2739 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
2740 BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2741 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
2742 BTRFS_LOCKDEP_TRANS_COMPLETED);
2743
2744 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2745 INIT_LIST_HEAD(&fs_info->space_info);
2746 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2747 INIT_LIST_HEAD(&fs_info->unused_bgs);
2748 INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2749 INIT_LIST_HEAD(&fs_info->zone_active_bgs);
2750#ifdef CONFIG_BTRFS_DEBUG
2751 INIT_LIST_HEAD(&fs_info->allocated_roots);
2752 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2753 spin_lock_init(&fs_info->eb_leak_lock);
2754#endif
2755 fs_info->mapping_tree = RB_ROOT_CACHED;
2756 rwlock_init(&fs_info->mapping_tree_lock);
2757 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2758 BTRFS_BLOCK_RSV_GLOBAL);
2759 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2760 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2761 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2762 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2763 BTRFS_BLOCK_RSV_DELOPS);
2764 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2765 BTRFS_BLOCK_RSV_DELREFS);
2766
2767 atomic_set(&fs_info->async_delalloc_pages, 0);
2768 atomic_set(&fs_info->defrag_running, 0);
2769 atomic_set(&fs_info->nr_delayed_iputs, 0);
2770 atomic64_set(&fs_info->tree_mod_seq, 0);
2771 fs_info->global_root_tree = RB_ROOT;
2772 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2773 fs_info->metadata_ratio = 0;
2774 fs_info->defrag_inodes = RB_ROOT;
2775 atomic64_set(&fs_info->free_chunk_space, 0);
2776 fs_info->tree_mod_log = RB_ROOT;
2777 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2778 btrfs_init_ref_verify(fs_info);
2779
2780 fs_info->thread_pool_size = min_t(unsigned long,
2781 num_online_cpus() + 2, 8);
2782
2783 INIT_LIST_HEAD(&fs_info->ordered_roots);
2784 spin_lock_init(&fs_info->ordered_root_lock);
2785
2786 btrfs_init_scrub(fs_info);
2787 btrfs_init_balance(fs_info);
2788 btrfs_init_async_reclaim_work(fs_info);
2789
2790 rwlock_init(&fs_info->block_group_cache_lock);
2791 fs_info->block_group_cache_tree = RB_ROOT_CACHED;
2792
2793 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2794 IO_TREE_FS_EXCLUDED_EXTENTS);
2795
2796 mutex_init(&fs_info->ordered_operations_mutex);
2797 mutex_init(&fs_info->tree_log_mutex);
2798 mutex_init(&fs_info->chunk_mutex);
2799 mutex_init(&fs_info->transaction_kthread_mutex);
2800 mutex_init(&fs_info->cleaner_mutex);
2801 mutex_init(&fs_info->ro_block_group_mutex);
2802 init_rwsem(&fs_info->commit_root_sem);
2803 init_rwsem(&fs_info->cleanup_work_sem);
2804 init_rwsem(&fs_info->subvol_sem);
2805 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2806
2807 btrfs_init_dev_replace_locks(fs_info);
2808 btrfs_init_qgroup(fs_info);
2809 btrfs_discard_init(fs_info);
2810
2811 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2812 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2813
2814 init_waitqueue_head(&fs_info->transaction_throttle);
2815 init_waitqueue_head(&fs_info->transaction_wait);
2816 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2817 init_waitqueue_head(&fs_info->async_submit_wait);
2818 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2819
2820 /* Usable values until the real ones are cached from the superblock */
2821 fs_info->nodesize = 4096;
2822 fs_info->sectorsize = 4096;
2823 fs_info->sectorsize_bits = ilog2(4096);
2824 fs_info->stripesize = 4096;
2825
2826 /* Default compress algorithm when user does -o compress */
2827 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2828
2829 fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
2830
2831 spin_lock_init(&fs_info->swapfile_pins_lock);
2832 fs_info->swapfile_pins = RB_ROOT;
2833
2834 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2835 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2836}
2837
2838static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2839{
2840 int ret;
2841
2842 fs_info->sb = sb;
2843 /* Temporary fixed values for block size until we read the superblock. */
2844 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2845 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2846
2847 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2848 if (ret)
2849 return ret;
2850
2851 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2852 if (ret)
2853 return ret;
2854
2855 fs_info->dirty_metadata_batch = PAGE_SIZE *
2856 (1 + ilog2(nr_cpu_ids));
2857
2858 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2859 if (ret)
2860 return ret;
2861
2862 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2863 GFP_KERNEL);
2864 if (ret)
2865 return ret;
2866
2867 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2868 GFP_KERNEL);
2869 if (!fs_info->delayed_root)
2870 return -ENOMEM;
2871 btrfs_init_delayed_root(fs_info->delayed_root);
2872
2873 if (sb_rdonly(sb))
2874 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2875
2876 return btrfs_alloc_stripe_hash_table(fs_info);
2877}
2878
2879static int btrfs_uuid_rescan_kthread(void *data)
2880{
2881 struct btrfs_fs_info *fs_info = data;
2882 int ret;
2883
2884 /*
2885 * 1st step is to iterate through the existing UUID tree and
2886 * to delete all entries that contain outdated data.
2887 * 2nd step is to add all missing entries to the UUID tree.
2888 */
2889 ret = btrfs_uuid_tree_iterate(fs_info);
2890 if (ret < 0) {
2891 if (ret != -EINTR)
2892 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2893 ret);
2894 up(&fs_info->uuid_tree_rescan_sem);
2895 return ret;
2896 }
2897 return btrfs_uuid_scan_kthread(data);
2898}
2899
2900static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2901{
2902 struct task_struct *task;
2903
2904 down(&fs_info->uuid_tree_rescan_sem);
2905 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2906 if (IS_ERR(task)) {
2907 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2908 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2909 up(&fs_info->uuid_tree_rescan_sem);
2910 return PTR_ERR(task);
2911 }
2912
2913 return 0;
2914}
2915
2916static int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2917{
2918 u64 root_objectid = 0;
2919 struct btrfs_root *gang[8];
2920 int i = 0;
2921 int err = 0;
2922 unsigned int ret = 0;
2923
2924 while (1) {
2925 spin_lock(&fs_info->fs_roots_radix_lock);
2926 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2927 (void **)gang, root_objectid,
2928 ARRAY_SIZE(gang));
2929 if (!ret) {
2930 spin_unlock(&fs_info->fs_roots_radix_lock);
2931 break;
2932 }
2933 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2934
2935 for (i = 0; i < ret; i++) {
2936 /* Avoid to grab roots in dead_roots. */
2937 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
2938 gang[i] = NULL;
2939 continue;
2940 }
2941 /* Grab all the search result for later use. */
2942 gang[i] = btrfs_grab_root(gang[i]);
2943 }
2944 spin_unlock(&fs_info->fs_roots_radix_lock);
2945
2946 for (i = 0; i < ret; i++) {
2947 if (!gang[i])
2948 continue;
2949 root_objectid = gang[i]->root_key.objectid;
2950 err = btrfs_orphan_cleanup(gang[i]);
2951 if (err)
2952 goto out;
2953 btrfs_put_root(gang[i]);
2954 }
2955 root_objectid++;
2956 }
2957out:
2958 /* Release the uncleaned roots due to error. */
2959 for (; i < ret; i++) {
2960 if (gang[i])
2961 btrfs_put_root(gang[i]);
2962 }
2963 return err;
2964}
2965
2966/*
2967 * Mounting logic specific to read-write file systems. Shared by open_ctree
2968 * and btrfs_remount when remounting from read-only to read-write.
2969 */
2970int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
2971{
2972 int ret;
2973 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
2974 bool rebuild_free_space_tree = false;
2975
2976 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2977 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2978 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
2979 btrfs_warn(fs_info,
2980 "'clear_cache' option is ignored with extent tree v2");
2981 else
2982 rebuild_free_space_tree = true;
2983 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2984 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2985 btrfs_warn(fs_info, "free space tree is invalid");
2986 rebuild_free_space_tree = true;
2987 }
2988
2989 if (rebuild_free_space_tree) {
2990 btrfs_info(fs_info, "rebuilding free space tree");
2991 ret = btrfs_rebuild_free_space_tree(fs_info);
2992 if (ret) {
2993 btrfs_warn(fs_info,
2994 "failed to rebuild free space tree: %d", ret);
2995 goto out;
2996 }
2997 }
2998
2999 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3000 !btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
3001 btrfs_info(fs_info, "disabling free space tree");
3002 ret = btrfs_delete_free_space_tree(fs_info);
3003 if (ret) {
3004 btrfs_warn(fs_info,
3005 "failed to disable free space tree: %d", ret);
3006 goto out;
3007 }
3008 }
3009
3010 /*
3011 * btrfs_find_orphan_roots() is responsible for finding all the dead
3012 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3013 * them into the fs_info->fs_roots_radix tree. This must be done before
3014 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3015 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3016 * item before the root's tree is deleted - this means that if we unmount
3017 * or crash before the deletion completes, on the next mount we will not
3018 * delete what remains of the tree because the orphan item does not
3019 * exists anymore, which is what tells us we have a pending deletion.
3020 */
3021 ret = btrfs_find_orphan_roots(fs_info);
3022 if (ret)
3023 goto out;
3024
3025 ret = btrfs_cleanup_fs_roots(fs_info);
3026 if (ret)
3027 goto out;
3028
3029 down_read(&fs_info->cleanup_work_sem);
3030 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3031 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3032 up_read(&fs_info->cleanup_work_sem);
3033 goto out;
3034 }
3035 up_read(&fs_info->cleanup_work_sem);
3036
3037 mutex_lock(&fs_info->cleaner_mutex);
3038 ret = btrfs_recover_relocation(fs_info);
3039 mutex_unlock(&fs_info->cleaner_mutex);
3040 if (ret < 0) {
3041 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3042 goto out;
3043 }
3044
3045 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3046 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3047 btrfs_info(fs_info, "creating free space tree");
3048 ret = btrfs_create_free_space_tree(fs_info);
3049 if (ret) {
3050 btrfs_warn(fs_info,
3051 "failed to create free space tree: %d", ret);
3052 goto out;
3053 }
3054 }
3055
3056 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3057 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3058 if (ret)
3059 goto out;
3060 }
3061
3062 ret = btrfs_resume_balance_async(fs_info);
3063 if (ret)
3064 goto out;
3065
3066 ret = btrfs_resume_dev_replace_async(fs_info);
3067 if (ret) {
3068 btrfs_warn(fs_info, "failed to resume dev_replace");
3069 goto out;
3070 }
3071
3072 btrfs_qgroup_rescan_resume(fs_info);
3073
3074 if (!fs_info->uuid_root) {
3075 btrfs_info(fs_info, "creating UUID tree");
3076 ret = btrfs_create_uuid_tree(fs_info);
3077 if (ret) {
3078 btrfs_warn(fs_info,
3079 "failed to create the UUID tree %d", ret);
3080 goto out;
3081 }
3082 }
3083
3084out:
3085 return ret;
3086}
3087
3088/*
3089 * Do various sanity and dependency checks of different features.
3090 *
3091 * @is_rw_mount: If the mount is read-write.
3092 *
3093 * This is the place for less strict checks (like for subpage or artificial
3094 * feature dependencies).
3095 *
3096 * For strict checks or possible corruption detection, see
3097 * btrfs_validate_super().
3098 *
3099 * This should be called after btrfs_parse_options(), as some mount options
3100 * (space cache related) can modify on-disk format like free space tree and
3101 * screw up certain feature dependencies.
3102 */
3103int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3104{
3105 struct btrfs_super_block *disk_super = fs_info->super_copy;
3106 u64 incompat = btrfs_super_incompat_flags(disk_super);
3107 const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3108 const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3109
3110 if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3111 btrfs_err(fs_info,
3112 "cannot mount because of unknown incompat features (0x%llx)",
3113 incompat);
3114 return -EINVAL;
3115 }
3116
3117 /* Runtime limitation for mixed block groups. */
3118 if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3119 (fs_info->sectorsize != fs_info->nodesize)) {
3120 btrfs_err(fs_info,
3121"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3122 fs_info->nodesize, fs_info->sectorsize);
3123 return -EINVAL;
3124 }
3125
3126 /* Mixed backref is an always-enabled feature. */
3127 incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3128
3129 /* Set compression related flags just in case. */
3130 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3131 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3132 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3133 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3134
3135 /*
3136 * An ancient flag, which should really be marked deprecated.
3137 * Such runtime limitation doesn't really need a incompat flag.
3138 */
3139 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3140 incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3141
3142 if (compat_ro_unsupp && is_rw_mount) {
3143 btrfs_err(fs_info,
3144 "cannot mount read-write because of unknown compat_ro features (0x%llx)",
3145 compat_ro);
3146 return -EINVAL;
3147 }
3148
3149 /*
3150 * We have unsupported RO compat features, although RO mounted, we
3151 * should not cause any metadata writes, including log replay.
3152 * Or we could screw up whatever the new feature requires.
3153 */
3154 if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3155 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3156 btrfs_err(fs_info,
3157"cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3158 compat_ro);
3159 return -EINVAL;
3160 }
3161
3162 /*
3163 * Artificial limitations for block group tree, to force
3164 * block-group-tree to rely on no-holes and free-space-tree.
3165 */
3166 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3167 (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3168 !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3169 btrfs_err(fs_info,
3170"block-group-tree feature requires no-holes and free-space-tree features");
3171 return -EINVAL;
3172 }
3173
3174 /*
3175 * Subpage runtime limitation on v1 cache.
3176 *
3177 * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3178 * we're already defaulting to v2 cache, no need to bother v1 as it's
3179 * going to be deprecated anyway.
3180 */
3181 if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3182 btrfs_warn(fs_info,
3183 "v1 space cache is not supported for page size %lu with sectorsize %u",
3184 PAGE_SIZE, fs_info->sectorsize);
3185 return -EINVAL;
3186 }
3187
3188 /* This can be called by remount, we need to protect the super block. */
3189 spin_lock(&fs_info->super_lock);
3190 btrfs_set_super_incompat_flags(disk_super, incompat);
3191 spin_unlock(&fs_info->super_lock);
3192
3193 return 0;
3194}
3195
3196int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3197 char *options)
3198{
3199 u32 sectorsize;
3200 u32 nodesize;
3201 u32 stripesize;
3202 u64 generation;
3203 u16 csum_type;
3204 struct btrfs_super_block *disk_super;
3205 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3206 struct btrfs_root *tree_root;
3207 struct btrfs_root *chunk_root;
3208 int ret;
3209 int level;
3210
3211 ret = init_mount_fs_info(fs_info, sb);
3212 if (ret)
3213 goto fail;
3214
3215 /* These need to be init'ed before we start creating inodes and such. */
3216 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3217 GFP_KERNEL);
3218 fs_info->tree_root = tree_root;
3219 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3220 GFP_KERNEL);
3221 fs_info->chunk_root = chunk_root;
3222 if (!tree_root || !chunk_root) {
3223 ret = -ENOMEM;
3224 goto fail;
3225 }
3226
3227 ret = btrfs_init_btree_inode(sb);
3228 if (ret)
3229 goto fail;
3230
3231 invalidate_bdev(fs_devices->latest_dev->bdev);
3232
3233 /*
3234 * Read super block and check the signature bytes only
3235 */
3236 disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3237 if (IS_ERR(disk_super)) {
3238 ret = PTR_ERR(disk_super);
3239 goto fail_alloc;
3240 }
3241
3242 btrfs_info(fs_info, "first mount of filesystem %pU", disk_super->fsid);
3243 /*
3244 * Verify the type first, if that or the checksum value are
3245 * corrupted, we'll find out
3246 */
3247 csum_type = btrfs_super_csum_type(disk_super);
3248 if (!btrfs_supported_super_csum(csum_type)) {
3249 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3250 csum_type);
3251 ret = -EINVAL;
3252 btrfs_release_disk_super(disk_super);
3253 goto fail_alloc;
3254 }
3255
3256 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3257
3258 ret = btrfs_init_csum_hash(fs_info, csum_type);
3259 if (ret) {
3260 btrfs_release_disk_super(disk_super);
3261 goto fail_alloc;
3262 }
3263
3264 /*
3265 * We want to check superblock checksum, the type is stored inside.
3266 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3267 */
3268 if (btrfs_check_super_csum(fs_info, disk_super)) {
3269 btrfs_err(fs_info, "superblock checksum mismatch");
3270 ret = -EINVAL;
3271 btrfs_release_disk_super(disk_super);
3272 goto fail_alloc;
3273 }
3274
3275 /*
3276 * super_copy is zeroed at allocation time and we never touch the
3277 * following bytes up to INFO_SIZE, the checksum is calculated from
3278 * the whole block of INFO_SIZE
3279 */
3280 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3281 btrfs_release_disk_super(disk_super);
3282
3283 disk_super = fs_info->super_copy;
3284
3285 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3286 sizeof(*fs_info->super_for_commit));
3287
3288 ret = btrfs_validate_mount_super(fs_info);
3289 if (ret) {
3290 btrfs_err(fs_info, "superblock contains fatal errors");
3291 ret = -EINVAL;
3292 goto fail_alloc;
3293 }
3294
3295 if (!btrfs_super_root(disk_super)) {
3296 btrfs_err(fs_info, "invalid superblock tree root bytenr");
3297 ret = -EINVAL;
3298 goto fail_alloc;
3299 }
3300
3301 /* check FS state, whether FS is broken. */
3302 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3303 WRITE_ONCE(fs_info->fs_error, -EUCLEAN);
3304
3305 /* Set up fs_info before parsing mount options */
3306 nodesize = btrfs_super_nodesize(disk_super);
3307 sectorsize = btrfs_super_sectorsize(disk_super);
3308 stripesize = sectorsize;
3309 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3310 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3311
3312 fs_info->nodesize = nodesize;
3313 fs_info->sectorsize = sectorsize;
3314 fs_info->sectorsize_bits = ilog2(sectorsize);
3315 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3316 fs_info->stripesize = stripesize;
3317
3318 /*
3319 * Handle the space caching options appropriately now that we have the
3320 * super block loaded and validated.
3321 */
3322 btrfs_set_free_space_cache_settings(fs_info);
3323
3324 if (!btrfs_check_options(fs_info, &fs_info->mount_opt, sb->s_flags)) {
3325 ret = -EINVAL;
3326 goto fail_alloc;
3327 }
3328
3329 ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3330 if (ret < 0)
3331 goto fail_alloc;
3332
3333 /*
3334 * At this point our mount options are validated, if we set ->max_inline
3335 * to something non-standard make sure we truncate it to sectorsize.
3336 */
3337 fs_info->max_inline = min_t(u64, fs_info->max_inline, fs_info->sectorsize);
3338
3339 if (sectorsize < PAGE_SIZE) {
3340 struct btrfs_subpage_info *subpage_info;
3341
3342 btrfs_warn(fs_info,
3343 "read-write for sector size %u with page size %lu is experimental",
3344 sectorsize, PAGE_SIZE);
3345 subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3346 if (!subpage_info) {
3347 ret = -ENOMEM;
3348 goto fail_alloc;
3349 }
3350 btrfs_init_subpage_info(subpage_info, sectorsize);
3351 fs_info->subpage_info = subpage_info;
3352 }
3353
3354 ret = btrfs_init_workqueues(fs_info);
3355 if (ret)
3356 goto fail_sb_buffer;
3357
3358 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3359 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3360
3361 /* Update the values for the current filesystem. */
3362 sb->s_blocksize = sectorsize;
3363 sb->s_blocksize_bits = blksize_bits(sectorsize);
3364 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3365
3366 mutex_lock(&fs_info->chunk_mutex);
3367 ret = btrfs_read_sys_array(fs_info);
3368 mutex_unlock(&fs_info->chunk_mutex);
3369 if (ret) {
3370 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3371 goto fail_sb_buffer;
3372 }
3373
3374 generation = btrfs_super_chunk_root_generation(disk_super);
3375 level = btrfs_super_chunk_root_level(disk_super);
3376 ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3377 generation, level);
3378 if (ret) {
3379 btrfs_err(fs_info, "failed to read chunk root");
3380 goto fail_tree_roots;
3381 }
3382
3383 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3384 offsetof(struct btrfs_header, chunk_tree_uuid),
3385 BTRFS_UUID_SIZE);
3386
3387 ret = btrfs_read_chunk_tree(fs_info);
3388 if (ret) {
3389 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3390 goto fail_tree_roots;
3391 }
3392
3393 /*
3394 * At this point we know all the devices that make this filesystem,
3395 * including the seed devices but we don't know yet if the replace
3396 * target is required. So free devices that are not part of this
3397 * filesystem but skip the replace target device which is checked
3398 * below in btrfs_init_dev_replace().
3399 */
3400 btrfs_free_extra_devids(fs_devices);
3401 if (!fs_devices->latest_dev->bdev) {
3402 btrfs_err(fs_info, "failed to read devices");
3403 ret = -EIO;
3404 goto fail_tree_roots;
3405 }
3406
3407 ret = init_tree_roots(fs_info);
3408 if (ret)
3409 goto fail_tree_roots;
3410
3411 /*
3412 * Get zone type information of zoned block devices. This will also
3413 * handle emulation of a zoned filesystem if a regular device has the
3414 * zoned incompat feature flag set.
3415 */
3416 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3417 if (ret) {
3418 btrfs_err(fs_info,
3419 "zoned: failed to read device zone info: %d", ret);
3420 goto fail_block_groups;
3421 }
3422
3423 /*
3424 * If we have a uuid root and we're not being told to rescan we need to
3425 * check the generation here so we can set the
3426 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3427 * transaction during a balance or the log replay without updating the
3428 * uuid generation, and then if we crash we would rescan the uuid tree,
3429 * even though it was perfectly fine.
3430 */
3431 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3432 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3433 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3434
3435 ret = btrfs_verify_dev_extents(fs_info);
3436 if (ret) {
3437 btrfs_err(fs_info,
3438 "failed to verify dev extents against chunks: %d",
3439 ret);
3440 goto fail_block_groups;
3441 }
3442 ret = btrfs_recover_balance(fs_info);
3443 if (ret) {
3444 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3445 goto fail_block_groups;
3446 }
3447
3448 ret = btrfs_init_dev_stats(fs_info);
3449 if (ret) {
3450 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3451 goto fail_block_groups;
3452 }
3453
3454 ret = btrfs_init_dev_replace(fs_info);
3455 if (ret) {
3456 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3457 goto fail_block_groups;
3458 }
3459
3460 ret = btrfs_check_zoned_mode(fs_info);
3461 if (ret) {
3462 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3463 ret);
3464 goto fail_block_groups;
3465 }
3466
3467 ret = btrfs_sysfs_add_fsid(fs_devices);
3468 if (ret) {
3469 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3470 ret);
3471 goto fail_block_groups;
3472 }
3473
3474 ret = btrfs_sysfs_add_mounted(fs_info);
3475 if (ret) {
3476 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3477 goto fail_fsdev_sysfs;
3478 }
3479
3480 ret = btrfs_init_space_info(fs_info);
3481 if (ret) {
3482 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3483 goto fail_sysfs;
3484 }
3485
3486 ret = btrfs_read_block_groups(fs_info);
3487 if (ret) {
3488 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3489 goto fail_sysfs;
3490 }
3491
3492 btrfs_free_zone_cache(fs_info);
3493
3494 btrfs_check_active_zone_reservation(fs_info);
3495
3496 if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3497 !btrfs_check_rw_degradable(fs_info, NULL)) {
3498 btrfs_warn(fs_info,
3499 "writable mount is not allowed due to too many missing devices");
3500 ret = -EINVAL;
3501 goto fail_sysfs;
3502 }
3503
3504 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3505 "btrfs-cleaner");
3506 if (IS_ERR(fs_info->cleaner_kthread)) {
3507 ret = PTR_ERR(fs_info->cleaner_kthread);
3508 goto fail_sysfs;
3509 }
3510
3511 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3512 tree_root,
3513 "btrfs-transaction");
3514 if (IS_ERR(fs_info->transaction_kthread)) {
3515 ret = PTR_ERR(fs_info->transaction_kthread);
3516 goto fail_cleaner;
3517 }
3518
3519 ret = btrfs_read_qgroup_config(fs_info);
3520 if (ret)
3521 goto fail_trans_kthread;
3522
3523 if (btrfs_build_ref_tree(fs_info))
3524 btrfs_err(fs_info, "couldn't build ref tree");
3525
3526 /* do not make disk changes in broken FS or nologreplay is given */
3527 if (btrfs_super_log_root(disk_super) != 0 &&
3528 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3529 btrfs_info(fs_info, "start tree-log replay");
3530 ret = btrfs_replay_log(fs_info, fs_devices);
3531 if (ret)
3532 goto fail_qgroup;
3533 }
3534
3535 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3536 if (IS_ERR(fs_info->fs_root)) {
3537 ret = PTR_ERR(fs_info->fs_root);
3538 btrfs_warn(fs_info, "failed to read fs tree: %d", ret);
3539 fs_info->fs_root = NULL;
3540 goto fail_qgroup;
3541 }
3542
3543 if (sb_rdonly(sb))
3544 return 0;
3545
3546 ret = btrfs_start_pre_rw_mount(fs_info);
3547 if (ret) {
3548 close_ctree(fs_info);
3549 return ret;
3550 }
3551 btrfs_discard_resume(fs_info);
3552
3553 if (fs_info->uuid_root &&
3554 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3555 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3556 btrfs_info(fs_info, "checking UUID tree");
3557 ret = btrfs_check_uuid_tree(fs_info);
3558 if (ret) {
3559 btrfs_warn(fs_info,
3560 "failed to check the UUID tree: %d", ret);
3561 close_ctree(fs_info);
3562 return ret;
3563 }
3564 }
3565
3566 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3567
3568 /* Kick the cleaner thread so it'll start deleting snapshots. */
3569 if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3570 wake_up_process(fs_info->cleaner_kthread);
3571
3572 return 0;
3573
3574fail_qgroup:
3575 btrfs_free_qgroup_config(fs_info);
3576fail_trans_kthread:
3577 kthread_stop(fs_info->transaction_kthread);
3578 btrfs_cleanup_transaction(fs_info);
3579 btrfs_free_fs_roots(fs_info);
3580fail_cleaner:
3581 kthread_stop(fs_info->cleaner_kthread);
3582
3583 /*
3584 * make sure we're done with the btree inode before we stop our
3585 * kthreads
3586 */
3587 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3588
3589fail_sysfs:
3590 btrfs_sysfs_remove_mounted(fs_info);
3591
3592fail_fsdev_sysfs:
3593 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3594
3595fail_block_groups:
3596 btrfs_put_block_group_cache(fs_info);
3597
3598fail_tree_roots:
3599 if (fs_info->data_reloc_root)
3600 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3601 free_root_pointers(fs_info, true);
3602 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3603
3604fail_sb_buffer:
3605 btrfs_stop_all_workers(fs_info);
3606 btrfs_free_block_groups(fs_info);
3607fail_alloc:
3608 btrfs_mapping_tree_free(fs_info);
3609
3610 iput(fs_info->btree_inode);
3611fail:
3612 btrfs_close_devices(fs_info->fs_devices);
3613 ASSERT(ret < 0);
3614 return ret;
3615}
3616ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3617
3618static void btrfs_end_super_write(struct bio *bio)
3619{
3620 struct btrfs_device *device = bio->bi_private;
3621 struct bio_vec *bvec;
3622 struct bvec_iter_all iter_all;
3623 struct page *page;
3624
3625 bio_for_each_segment_all(bvec, bio, iter_all) {
3626 page = bvec->bv_page;
3627
3628 if (bio->bi_status) {
3629 btrfs_warn_rl_in_rcu(device->fs_info,
3630 "lost page write due to IO error on %s (%d)",
3631 btrfs_dev_name(device),
3632 blk_status_to_errno(bio->bi_status));
3633 ClearPageUptodate(page);
3634 SetPageError(page);
3635 btrfs_dev_stat_inc_and_print(device,
3636 BTRFS_DEV_STAT_WRITE_ERRS);
3637 } else {
3638 SetPageUptodate(page);
3639 }
3640
3641 put_page(page);
3642 unlock_page(page);
3643 }
3644
3645 bio_put(bio);
3646}
3647
3648struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3649 int copy_num, bool drop_cache)
3650{
3651 struct btrfs_super_block *super;
3652 struct page *page;
3653 u64 bytenr, bytenr_orig;
3654 struct address_space *mapping = bdev->bd_inode->i_mapping;
3655 int ret;
3656
3657 bytenr_orig = btrfs_sb_offset(copy_num);
3658 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3659 if (ret == -ENOENT)
3660 return ERR_PTR(-EINVAL);
3661 else if (ret)
3662 return ERR_PTR(ret);
3663
3664 if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3665 return ERR_PTR(-EINVAL);
3666
3667 if (drop_cache) {
3668 /* This should only be called with the primary sb. */
3669 ASSERT(copy_num == 0);
3670
3671 /*
3672 * Drop the page of the primary superblock, so later read will
3673 * always read from the device.
3674 */
3675 invalidate_inode_pages2_range(mapping,
3676 bytenr >> PAGE_SHIFT,
3677 (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3678 }
3679
3680 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3681 if (IS_ERR(page))
3682 return ERR_CAST(page);
3683
3684 super = page_address(page);
3685 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3686 btrfs_release_disk_super(super);
3687 return ERR_PTR(-ENODATA);
3688 }
3689
3690 if (btrfs_super_bytenr(super) != bytenr_orig) {
3691 btrfs_release_disk_super(super);
3692 return ERR_PTR(-EINVAL);
3693 }
3694
3695 return super;
3696}
3697
3698
3699struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3700{
3701 struct btrfs_super_block *super, *latest = NULL;
3702 int i;
3703 u64 transid = 0;
3704
3705 /* we would like to check all the supers, but that would make
3706 * a btrfs mount succeed after a mkfs from a different FS.
3707 * So, we need to add a special mount option to scan for
3708 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3709 */
3710 for (i = 0; i < 1; i++) {
3711 super = btrfs_read_dev_one_super(bdev, i, false);
3712 if (IS_ERR(super))
3713 continue;
3714
3715 if (!latest || btrfs_super_generation(super) > transid) {
3716 if (latest)
3717 btrfs_release_disk_super(super);
3718
3719 latest = super;
3720 transid = btrfs_super_generation(super);
3721 }
3722 }
3723
3724 return super;
3725}
3726
3727/*
3728 * Write superblock @sb to the @device. Do not wait for completion, all the
3729 * pages we use for writing are locked.
3730 *
3731 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3732 * the expected device size at commit time. Note that max_mirrors must be
3733 * same for write and wait phases.
3734 *
3735 * Return number of errors when page is not found or submission fails.
3736 */
3737static int write_dev_supers(struct btrfs_device *device,
3738 struct btrfs_super_block *sb, int max_mirrors)
3739{
3740 struct btrfs_fs_info *fs_info = device->fs_info;
3741 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3742 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3743 int i;
3744 int errors = 0;
3745 int ret;
3746 u64 bytenr, bytenr_orig;
3747
3748 if (max_mirrors == 0)
3749 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3750
3751 shash->tfm = fs_info->csum_shash;
3752
3753 for (i = 0; i < max_mirrors; i++) {
3754 struct page *page;
3755 struct bio *bio;
3756 struct btrfs_super_block *disk_super;
3757
3758 bytenr_orig = btrfs_sb_offset(i);
3759 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3760 if (ret == -ENOENT) {
3761 continue;
3762 } else if (ret < 0) {
3763 btrfs_err(device->fs_info,
3764 "couldn't get super block location for mirror %d",
3765 i);
3766 errors++;
3767 continue;
3768 }
3769 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3770 device->commit_total_bytes)
3771 break;
3772
3773 btrfs_set_super_bytenr(sb, bytenr_orig);
3774
3775 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3776 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3777 sb->csum);
3778
3779 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3780 GFP_NOFS);
3781 if (!page) {
3782 btrfs_err(device->fs_info,
3783 "couldn't get super block page for bytenr %llu",
3784 bytenr);
3785 errors++;
3786 continue;
3787 }
3788
3789 /* Bump the refcount for wait_dev_supers() */
3790 get_page(page);
3791
3792 disk_super = page_address(page);
3793 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3794
3795 /*
3796 * Directly use bios here instead of relying on the page cache
3797 * to do I/O, so we don't lose the ability to do integrity
3798 * checking.
3799 */
3800 bio = bio_alloc(device->bdev, 1,
3801 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
3802 GFP_NOFS);
3803 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3804 bio->bi_private = device;
3805 bio->bi_end_io = btrfs_end_super_write;
3806 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3807 offset_in_page(bytenr));
3808
3809 /*
3810 * We FUA only the first super block. The others we allow to
3811 * go down lazy and there's a short window where the on-disk
3812 * copies might still contain the older version.
3813 */
3814 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3815 bio->bi_opf |= REQ_FUA;
3816 submit_bio(bio);
3817
3818 if (btrfs_advance_sb_log(device, i))
3819 errors++;
3820 }
3821 return errors < i ? 0 : -1;
3822}
3823
3824/*
3825 * Wait for write completion of superblocks done by write_dev_supers,
3826 * @max_mirrors same for write and wait phases.
3827 *
3828 * Return number of errors when page is not found or not marked up to
3829 * date.
3830 */
3831static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3832{
3833 int i;
3834 int errors = 0;
3835 bool primary_failed = false;
3836 int ret;
3837 u64 bytenr;
3838
3839 if (max_mirrors == 0)
3840 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3841
3842 for (i = 0; i < max_mirrors; i++) {
3843 struct page *page;
3844
3845 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3846 if (ret == -ENOENT) {
3847 break;
3848 } else if (ret < 0) {
3849 errors++;
3850 if (i == 0)
3851 primary_failed = true;
3852 continue;
3853 }
3854 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3855 device->commit_total_bytes)
3856 break;
3857
3858 page = find_get_page(device->bdev->bd_inode->i_mapping,
3859 bytenr >> PAGE_SHIFT);
3860 if (!page) {
3861 errors++;
3862 if (i == 0)
3863 primary_failed = true;
3864 continue;
3865 }
3866 /* Page is submitted locked and unlocked once the IO completes */
3867 wait_on_page_locked(page);
3868 if (PageError(page)) {
3869 errors++;
3870 if (i == 0)
3871 primary_failed = true;
3872 }
3873
3874 /* Drop our reference */
3875 put_page(page);
3876
3877 /* Drop the reference from the writing run */
3878 put_page(page);
3879 }
3880
3881 /* log error, force error return */
3882 if (primary_failed) {
3883 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3884 device->devid);
3885 return -1;
3886 }
3887
3888 return errors < i ? 0 : -1;
3889}
3890
3891/*
3892 * endio for the write_dev_flush, this will wake anyone waiting
3893 * for the barrier when it is done
3894 */
3895static void btrfs_end_empty_barrier(struct bio *bio)
3896{
3897 bio_uninit(bio);
3898 complete(bio->bi_private);
3899}
3900
3901/*
3902 * Submit a flush request to the device if it supports it. Error handling is
3903 * done in the waiting counterpart.
3904 */
3905static void write_dev_flush(struct btrfs_device *device)
3906{
3907 struct bio *bio = &device->flush_bio;
3908
3909 device->last_flush_error = BLK_STS_OK;
3910
3911 bio_init(bio, device->bdev, NULL, 0,
3912 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
3913 bio->bi_end_io = btrfs_end_empty_barrier;
3914 init_completion(&device->flush_wait);
3915 bio->bi_private = &device->flush_wait;
3916 submit_bio(bio);
3917 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3918}
3919
3920/*
3921 * If the flush bio has been submitted by write_dev_flush, wait for it.
3922 * Return true for any error, and false otherwise.
3923 */
3924static bool wait_dev_flush(struct btrfs_device *device)
3925{
3926 struct bio *bio = &device->flush_bio;
3927
3928 if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3929 return false;
3930
3931 wait_for_completion_io(&device->flush_wait);
3932
3933 if (bio->bi_status) {
3934 device->last_flush_error = bio->bi_status;
3935 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_FLUSH_ERRS);
3936 return true;
3937 }
3938
3939 return false;
3940}
3941
3942/*
3943 * send an empty flush down to each device in parallel,
3944 * then wait for them
3945 */
3946static int barrier_all_devices(struct btrfs_fs_info *info)
3947{
3948 struct list_head *head;
3949 struct btrfs_device *dev;
3950 int errors_wait = 0;
3951
3952 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3953 /* send down all the barriers */
3954 head = &info->fs_devices->devices;
3955 list_for_each_entry(dev, head, dev_list) {
3956 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3957 continue;
3958 if (!dev->bdev)
3959 continue;
3960 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3961 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3962 continue;
3963
3964 write_dev_flush(dev);
3965 }
3966
3967 /* wait for all the barriers */
3968 list_for_each_entry(dev, head, dev_list) {
3969 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3970 continue;
3971 if (!dev->bdev) {
3972 errors_wait++;
3973 continue;
3974 }
3975 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3976 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3977 continue;
3978
3979 if (wait_dev_flush(dev))
3980 errors_wait++;
3981 }
3982
3983 /*
3984 * Checks last_flush_error of disks in order to determine the device
3985 * state.
3986 */
3987 if (errors_wait && !btrfs_check_rw_degradable(info, NULL))
3988 return -EIO;
3989
3990 return 0;
3991}
3992
3993int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3994{
3995 int raid_type;
3996 int min_tolerated = INT_MAX;
3997
3998 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3999 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4000 min_tolerated = min_t(int, min_tolerated,
4001 btrfs_raid_array[BTRFS_RAID_SINGLE].
4002 tolerated_failures);
4003
4004 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4005 if (raid_type == BTRFS_RAID_SINGLE)
4006 continue;
4007 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4008 continue;
4009 min_tolerated = min_t(int, min_tolerated,
4010 btrfs_raid_array[raid_type].
4011 tolerated_failures);
4012 }
4013
4014 if (min_tolerated == INT_MAX) {
4015 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4016 min_tolerated = 0;
4017 }
4018
4019 return min_tolerated;
4020}
4021
4022int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4023{
4024 struct list_head *head;
4025 struct btrfs_device *dev;
4026 struct btrfs_super_block *sb;
4027 struct btrfs_dev_item *dev_item;
4028 int ret;
4029 int do_barriers;
4030 int max_errors;
4031 int total_errors = 0;
4032 u64 flags;
4033
4034 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4035
4036 /*
4037 * max_mirrors == 0 indicates we're from commit_transaction,
4038 * not from fsync where the tree roots in fs_info have not
4039 * been consistent on disk.
4040 */
4041 if (max_mirrors == 0)
4042 backup_super_roots(fs_info);
4043
4044 sb = fs_info->super_for_commit;
4045 dev_item = &sb->dev_item;
4046
4047 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4048 head = &fs_info->fs_devices->devices;
4049 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4050
4051 if (do_barriers) {
4052 ret = barrier_all_devices(fs_info);
4053 if (ret) {
4054 mutex_unlock(
4055 &fs_info->fs_devices->device_list_mutex);
4056 btrfs_handle_fs_error(fs_info, ret,
4057 "errors while submitting device barriers.");
4058 return ret;
4059 }
4060 }
4061
4062 list_for_each_entry(dev, head, dev_list) {
4063 if (!dev->bdev) {
4064 total_errors++;
4065 continue;
4066 }
4067 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4068 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4069 continue;
4070
4071 btrfs_set_stack_device_generation(dev_item, 0);
4072 btrfs_set_stack_device_type(dev_item, dev->type);
4073 btrfs_set_stack_device_id(dev_item, dev->devid);
4074 btrfs_set_stack_device_total_bytes(dev_item,
4075 dev->commit_total_bytes);
4076 btrfs_set_stack_device_bytes_used(dev_item,
4077 dev->commit_bytes_used);
4078 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4079 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4080 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4081 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4082 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4083 BTRFS_FSID_SIZE);
4084
4085 flags = btrfs_super_flags(sb);
4086 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4087
4088 ret = btrfs_validate_write_super(fs_info, sb);
4089 if (ret < 0) {
4090 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4091 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4092 "unexpected superblock corruption detected");
4093 return -EUCLEAN;
4094 }
4095
4096 ret = write_dev_supers(dev, sb, max_mirrors);
4097 if (ret)
4098 total_errors++;
4099 }
4100 if (total_errors > max_errors) {
4101 btrfs_err(fs_info, "%d errors while writing supers",
4102 total_errors);
4103 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4104
4105 /* FUA is masked off if unsupported and can't be the reason */
4106 btrfs_handle_fs_error(fs_info, -EIO,
4107 "%d errors while writing supers",
4108 total_errors);
4109 return -EIO;
4110 }
4111
4112 total_errors = 0;
4113 list_for_each_entry(dev, head, dev_list) {
4114 if (!dev->bdev)
4115 continue;
4116 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4117 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4118 continue;
4119
4120 ret = wait_dev_supers(dev, max_mirrors);
4121 if (ret)
4122 total_errors++;
4123 }
4124 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4125 if (total_errors > max_errors) {
4126 btrfs_handle_fs_error(fs_info, -EIO,
4127 "%d errors while writing supers",
4128 total_errors);
4129 return -EIO;
4130 }
4131 return 0;
4132}
4133
4134/* Drop a fs root from the radix tree and free it. */
4135void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4136 struct btrfs_root *root)
4137{
4138 bool drop_ref = false;
4139
4140 spin_lock(&fs_info->fs_roots_radix_lock);
4141 radix_tree_delete(&fs_info->fs_roots_radix,
4142 (unsigned long)root->root_key.objectid);
4143 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4144 drop_ref = true;
4145 spin_unlock(&fs_info->fs_roots_radix_lock);
4146
4147 if (BTRFS_FS_ERROR(fs_info)) {
4148 ASSERT(root->log_root == NULL);
4149 if (root->reloc_root) {
4150 btrfs_put_root(root->reloc_root);
4151 root->reloc_root = NULL;
4152 }
4153 }
4154
4155 if (drop_ref)
4156 btrfs_put_root(root);
4157}
4158
4159int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4160{
4161 struct btrfs_root *root = fs_info->tree_root;
4162 struct btrfs_trans_handle *trans;
4163
4164 mutex_lock(&fs_info->cleaner_mutex);
4165 btrfs_run_delayed_iputs(fs_info);
4166 mutex_unlock(&fs_info->cleaner_mutex);
4167 wake_up_process(fs_info->cleaner_kthread);
4168
4169 /* wait until ongoing cleanup work done */
4170 down_write(&fs_info->cleanup_work_sem);
4171 up_write(&fs_info->cleanup_work_sem);
4172
4173 trans = btrfs_join_transaction(root);
4174 if (IS_ERR(trans))
4175 return PTR_ERR(trans);
4176 return btrfs_commit_transaction(trans);
4177}
4178
4179static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4180{
4181 struct btrfs_transaction *trans;
4182 struct btrfs_transaction *tmp;
4183 bool found = false;
4184
4185 if (list_empty(&fs_info->trans_list))
4186 return;
4187
4188 /*
4189 * This function is only called at the very end of close_ctree(),
4190 * thus no other running transaction, no need to take trans_lock.
4191 */
4192 ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4193 list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4194 struct extent_state *cached = NULL;
4195 u64 dirty_bytes = 0;
4196 u64 cur = 0;
4197 u64 found_start;
4198 u64 found_end;
4199
4200 found = true;
4201 while (find_first_extent_bit(&trans->dirty_pages, cur,
4202 &found_start, &found_end, EXTENT_DIRTY, &cached)) {
4203 dirty_bytes += found_end + 1 - found_start;
4204 cur = found_end + 1;
4205 }
4206 btrfs_warn(fs_info,
4207 "transaction %llu (with %llu dirty metadata bytes) is not committed",
4208 trans->transid, dirty_bytes);
4209 btrfs_cleanup_one_transaction(trans, fs_info);
4210
4211 if (trans == fs_info->running_transaction)
4212 fs_info->running_transaction = NULL;
4213 list_del_init(&trans->list);
4214
4215 btrfs_put_transaction(trans);
4216 trace_btrfs_transaction_commit(fs_info);
4217 }
4218 ASSERT(!found);
4219}
4220
4221void __cold close_ctree(struct btrfs_fs_info *fs_info)
4222{
4223 int ret;
4224
4225 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4226
4227 /*
4228 * If we had UNFINISHED_DROPS we could still be processing them, so
4229 * clear that bit and wake up relocation so it can stop.
4230 * We must do this before stopping the block group reclaim task, because
4231 * at btrfs_relocate_block_group() we wait for this bit, and after the
4232 * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4233 * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4234 * return 1.
4235 */
4236 btrfs_wake_unfinished_drop(fs_info);
4237
4238 /*
4239 * We may have the reclaim task running and relocating a data block group,
4240 * in which case it may create delayed iputs. So stop it before we park
4241 * the cleaner kthread otherwise we can get new delayed iputs after
4242 * parking the cleaner, and that can make the async reclaim task to hang
4243 * if it's waiting for delayed iputs to complete, since the cleaner is
4244 * parked and can not run delayed iputs - this will make us hang when
4245 * trying to stop the async reclaim task.
4246 */
4247 cancel_work_sync(&fs_info->reclaim_bgs_work);
4248 /*
4249 * We don't want the cleaner to start new transactions, add more delayed
4250 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4251 * because that frees the task_struct, and the transaction kthread might
4252 * still try to wake up the cleaner.
4253 */
4254 kthread_park(fs_info->cleaner_kthread);
4255
4256 /* wait for the qgroup rescan worker to stop */
4257 btrfs_qgroup_wait_for_completion(fs_info, false);
4258
4259 /* wait for the uuid_scan task to finish */
4260 down(&fs_info->uuid_tree_rescan_sem);
4261 /* avoid complains from lockdep et al., set sem back to initial state */
4262 up(&fs_info->uuid_tree_rescan_sem);
4263
4264 /* pause restriper - we want to resume on mount */
4265 btrfs_pause_balance(fs_info);
4266
4267 btrfs_dev_replace_suspend_for_unmount(fs_info);
4268
4269 btrfs_scrub_cancel(fs_info);
4270
4271 /* wait for any defraggers to finish */
4272 wait_event(fs_info->transaction_wait,
4273 (atomic_read(&fs_info->defrag_running) == 0));
4274
4275 /* clear out the rbtree of defraggable inodes */
4276 btrfs_cleanup_defrag_inodes(fs_info);
4277
4278 /*
4279 * After we parked the cleaner kthread, ordered extents may have
4280 * completed and created new delayed iputs. If one of the async reclaim
4281 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4282 * can hang forever trying to stop it, because if a delayed iput is
4283 * added after it ran btrfs_run_delayed_iputs() and before it called
4284 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4285 * no one else to run iputs.
4286 *
4287 * So wait for all ongoing ordered extents to complete and then run
4288 * delayed iputs. This works because once we reach this point no one
4289 * can either create new ordered extents nor create delayed iputs
4290 * through some other means.
4291 *
4292 * Also note that btrfs_wait_ordered_roots() is not safe here, because
4293 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4294 * but the delayed iput for the respective inode is made only when doing
4295 * the final btrfs_put_ordered_extent() (which must happen at
4296 * btrfs_finish_ordered_io() when we are unmounting).
4297 */
4298 btrfs_flush_workqueue(fs_info->endio_write_workers);
4299 /* Ordered extents for free space inodes. */
4300 btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4301 btrfs_run_delayed_iputs(fs_info);
4302
4303 cancel_work_sync(&fs_info->async_reclaim_work);
4304 cancel_work_sync(&fs_info->async_data_reclaim_work);
4305 cancel_work_sync(&fs_info->preempt_reclaim_work);
4306
4307 /* Cancel or finish ongoing discard work */
4308 btrfs_discard_cleanup(fs_info);
4309
4310 if (!sb_rdonly(fs_info->sb)) {
4311 /*
4312 * The cleaner kthread is stopped, so do one final pass over
4313 * unused block groups.
4314 */
4315 btrfs_delete_unused_bgs(fs_info);
4316
4317 /*
4318 * There might be existing delayed inode workers still running
4319 * and holding an empty delayed inode item. We must wait for
4320 * them to complete first because they can create a transaction.
4321 * This happens when someone calls btrfs_balance_delayed_items()
4322 * and then a transaction commit runs the same delayed nodes
4323 * before any delayed worker has done something with the nodes.
4324 * We must wait for any worker here and not at transaction
4325 * commit time since that could cause a deadlock.
4326 * This is a very rare case.
4327 */
4328 btrfs_flush_workqueue(fs_info->delayed_workers);
4329
4330 ret = btrfs_commit_super(fs_info);
4331 if (ret)
4332 btrfs_err(fs_info, "commit super ret %d", ret);
4333 }
4334
4335 if (BTRFS_FS_ERROR(fs_info))
4336 btrfs_error_commit_super(fs_info);
4337
4338 kthread_stop(fs_info->transaction_kthread);
4339 kthread_stop(fs_info->cleaner_kthread);
4340
4341 ASSERT(list_empty(&fs_info->delayed_iputs));
4342 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4343
4344 if (btrfs_check_quota_leak(fs_info)) {
4345 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4346 btrfs_err(fs_info, "qgroup reserved space leaked");
4347 }
4348
4349 btrfs_free_qgroup_config(fs_info);
4350 ASSERT(list_empty(&fs_info->delalloc_roots));
4351
4352 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4353 btrfs_info(fs_info, "at unmount delalloc count %lld",
4354 percpu_counter_sum(&fs_info->delalloc_bytes));
4355 }
4356
4357 if (percpu_counter_sum(&fs_info->ordered_bytes))
4358 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4359 percpu_counter_sum(&fs_info->ordered_bytes));
4360
4361 btrfs_sysfs_remove_mounted(fs_info);
4362 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4363
4364 btrfs_put_block_group_cache(fs_info);
4365
4366 /*
4367 * we must make sure there is not any read request to
4368 * submit after we stopping all workers.
4369 */
4370 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4371 btrfs_stop_all_workers(fs_info);
4372
4373 /* We shouldn't have any transaction open at this point */
4374 warn_about_uncommitted_trans(fs_info);
4375
4376 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4377 free_root_pointers(fs_info, true);
4378 btrfs_free_fs_roots(fs_info);
4379
4380 /*
4381 * We must free the block groups after dropping the fs_roots as we could
4382 * have had an IO error and have left over tree log blocks that aren't
4383 * cleaned up until the fs roots are freed. This makes the block group
4384 * accounting appear to be wrong because there's pending reserved bytes,
4385 * so make sure we do the block group cleanup afterwards.
4386 */
4387 btrfs_free_block_groups(fs_info);
4388
4389 iput(fs_info->btree_inode);
4390
4391 btrfs_mapping_tree_free(fs_info);
4392 btrfs_close_devices(fs_info->fs_devices);
4393}
4394
4395void btrfs_mark_buffer_dirty(struct btrfs_trans_handle *trans,
4396 struct extent_buffer *buf)
4397{
4398 struct btrfs_fs_info *fs_info = buf->fs_info;
4399 u64 transid = btrfs_header_generation(buf);
4400
4401#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4402 /*
4403 * This is a fast path so only do this check if we have sanity tests
4404 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4405 * outside of the sanity tests.
4406 */
4407 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4408 return;
4409#endif
4410 /* This is an active transaction (its state < TRANS_STATE_UNBLOCKED). */
4411 ASSERT(trans->transid == fs_info->generation);
4412 btrfs_assert_tree_write_locked(buf);
4413 if (unlikely(transid != fs_info->generation)) {
4414 btrfs_abort_transaction(trans, -EUCLEAN);
4415 btrfs_crit(fs_info,
4416"dirty buffer transid mismatch, logical %llu found transid %llu running transid %llu",
4417 buf->start, transid, fs_info->generation);
4418 }
4419 set_extent_buffer_dirty(buf);
4420}
4421
4422static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4423 int flush_delayed)
4424{
4425 /*
4426 * looks as though older kernels can get into trouble with
4427 * this code, they end up stuck in balance_dirty_pages forever
4428 */
4429 int ret;
4430
4431 if (current->flags & PF_MEMALLOC)
4432 return;
4433
4434 if (flush_delayed)
4435 btrfs_balance_delayed_items(fs_info);
4436
4437 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4438 BTRFS_DIRTY_METADATA_THRESH,
4439 fs_info->dirty_metadata_batch);
4440 if (ret > 0) {
4441 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4442 }
4443}
4444
4445void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4446{
4447 __btrfs_btree_balance_dirty(fs_info, 1);
4448}
4449
4450void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4451{
4452 __btrfs_btree_balance_dirty(fs_info, 0);
4453}
4454
4455static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4456{
4457 /* cleanup FS via transaction */
4458 btrfs_cleanup_transaction(fs_info);
4459
4460 mutex_lock(&fs_info->cleaner_mutex);
4461 btrfs_run_delayed_iputs(fs_info);
4462 mutex_unlock(&fs_info->cleaner_mutex);
4463
4464 down_write(&fs_info->cleanup_work_sem);
4465 up_write(&fs_info->cleanup_work_sem);
4466}
4467
4468static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4469{
4470 struct btrfs_root *gang[8];
4471 u64 root_objectid = 0;
4472 int ret;
4473
4474 spin_lock(&fs_info->fs_roots_radix_lock);
4475 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4476 (void **)gang, root_objectid,
4477 ARRAY_SIZE(gang))) != 0) {
4478 int i;
4479
4480 for (i = 0; i < ret; i++)
4481 gang[i] = btrfs_grab_root(gang[i]);
4482 spin_unlock(&fs_info->fs_roots_radix_lock);
4483
4484 for (i = 0; i < ret; i++) {
4485 if (!gang[i])
4486 continue;
4487 root_objectid = gang[i]->root_key.objectid;
4488 btrfs_free_log(NULL, gang[i]);
4489 btrfs_put_root(gang[i]);
4490 }
4491 root_objectid++;
4492 spin_lock(&fs_info->fs_roots_radix_lock);
4493 }
4494 spin_unlock(&fs_info->fs_roots_radix_lock);
4495 btrfs_free_log_root_tree(NULL, fs_info);
4496}
4497
4498static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4499{
4500 struct btrfs_ordered_extent *ordered;
4501
4502 spin_lock(&root->ordered_extent_lock);
4503 /*
4504 * This will just short circuit the ordered completion stuff which will
4505 * make sure the ordered extent gets properly cleaned up.
4506 */
4507 list_for_each_entry(ordered, &root->ordered_extents,
4508 root_extent_list)
4509 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4510 spin_unlock(&root->ordered_extent_lock);
4511}
4512
4513static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4514{
4515 struct btrfs_root *root;
4516 LIST_HEAD(splice);
4517
4518 spin_lock(&fs_info->ordered_root_lock);
4519 list_splice_init(&fs_info->ordered_roots, &splice);
4520 while (!list_empty(&splice)) {
4521 root = list_first_entry(&splice, struct btrfs_root,
4522 ordered_root);
4523 list_move_tail(&root->ordered_root,
4524 &fs_info->ordered_roots);
4525
4526 spin_unlock(&fs_info->ordered_root_lock);
4527 btrfs_destroy_ordered_extents(root);
4528
4529 cond_resched();
4530 spin_lock(&fs_info->ordered_root_lock);
4531 }
4532 spin_unlock(&fs_info->ordered_root_lock);
4533
4534 /*
4535 * We need this here because if we've been flipped read-only we won't
4536 * get sync() from the umount, so we need to make sure any ordered
4537 * extents that haven't had their dirty pages IO start writeout yet
4538 * actually get run and error out properly.
4539 */
4540 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4541}
4542
4543static void btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4544 struct btrfs_fs_info *fs_info)
4545{
4546 struct rb_node *node;
4547 struct btrfs_delayed_ref_root *delayed_refs;
4548 struct btrfs_delayed_ref_node *ref;
4549
4550 delayed_refs = &trans->delayed_refs;
4551
4552 spin_lock(&delayed_refs->lock);
4553 if (atomic_read(&delayed_refs->num_entries) == 0) {
4554 spin_unlock(&delayed_refs->lock);
4555 btrfs_debug(fs_info, "delayed_refs has NO entry");
4556 return;
4557 }
4558
4559 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4560 struct btrfs_delayed_ref_head *head;
4561 struct rb_node *n;
4562 bool pin_bytes = false;
4563
4564 head = rb_entry(node, struct btrfs_delayed_ref_head,
4565 href_node);
4566 if (btrfs_delayed_ref_lock(delayed_refs, head))
4567 continue;
4568
4569 spin_lock(&head->lock);
4570 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4571 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4572 ref_node);
4573 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4574 RB_CLEAR_NODE(&ref->ref_node);
4575 if (!list_empty(&ref->add_list))
4576 list_del(&ref->add_list);
4577 atomic_dec(&delayed_refs->num_entries);
4578 btrfs_put_delayed_ref(ref);
4579 btrfs_delayed_refs_rsv_release(fs_info, 1, 0);
4580 }
4581 if (head->must_insert_reserved)
4582 pin_bytes = true;
4583 btrfs_free_delayed_extent_op(head->extent_op);
4584 btrfs_delete_ref_head(delayed_refs, head);
4585 spin_unlock(&head->lock);
4586 spin_unlock(&delayed_refs->lock);
4587 mutex_unlock(&head->mutex);
4588
4589 if (pin_bytes) {
4590 struct btrfs_block_group *cache;
4591
4592 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4593 BUG_ON(!cache);
4594
4595 spin_lock(&cache->space_info->lock);
4596 spin_lock(&cache->lock);
4597 cache->pinned += head->num_bytes;
4598 btrfs_space_info_update_bytes_pinned(fs_info,
4599 cache->space_info, head->num_bytes);
4600 cache->reserved -= head->num_bytes;
4601 cache->space_info->bytes_reserved -= head->num_bytes;
4602 spin_unlock(&cache->lock);
4603 spin_unlock(&cache->space_info->lock);
4604
4605 btrfs_put_block_group(cache);
4606
4607 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4608 head->bytenr + head->num_bytes - 1);
4609 }
4610 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4611 btrfs_put_delayed_ref_head(head);
4612 cond_resched();
4613 spin_lock(&delayed_refs->lock);
4614 }
4615 btrfs_qgroup_destroy_extent_records(trans);
4616
4617 spin_unlock(&delayed_refs->lock);
4618}
4619
4620static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4621{
4622 struct btrfs_inode *btrfs_inode;
4623 LIST_HEAD(splice);
4624
4625 spin_lock(&root->delalloc_lock);
4626 list_splice_init(&root->delalloc_inodes, &splice);
4627
4628 while (!list_empty(&splice)) {
4629 struct inode *inode = NULL;
4630 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4631 delalloc_inodes);
4632 btrfs_del_delalloc_inode(btrfs_inode);
4633 spin_unlock(&root->delalloc_lock);
4634
4635 /*
4636 * Make sure we get a live inode and that it'll not disappear
4637 * meanwhile.
4638 */
4639 inode = igrab(&btrfs_inode->vfs_inode);
4640 if (inode) {
4641 unsigned int nofs_flag;
4642
4643 nofs_flag = memalloc_nofs_save();
4644 invalidate_inode_pages2(inode->i_mapping);
4645 memalloc_nofs_restore(nofs_flag);
4646 iput(inode);
4647 }
4648 spin_lock(&root->delalloc_lock);
4649 }
4650 spin_unlock(&root->delalloc_lock);
4651}
4652
4653static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4654{
4655 struct btrfs_root *root;
4656 LIST_HEAD(splice);
4657
4658 spin_lock(&fs_info->delalloc_root_lock);
4659 list_splice_init(&fs_info->delalloc_roots, &splice);
4660 while (!list_empty(&splice)) {
4661 root = list_first_entry(&splice, struct btrfs_root,
4662 delalloc_root);
4663 root = btrfs_grab_root(root);
4664 BUG_ON(!root);
4665 spin_unlock(&fs_info->delalloc_root_lock);
4666
4667 btrfs_destroy_delalloc_inodes(root);
4668 btrfs_put_root(root);
4669
4670 spin_lock(&fs_info->delalloc_root_lock);
4671 }
4672 spin_unlock(&fs_info->delalloc_root_lock);
4673}
4674
4675static void btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4676 struct extent_io_tree *dirty_pages,
4677 int mark)
4678{
4679 struct extent_buffer *eb;
4680 u64 start = 0;
4681 u64 end;
4682
4683 while (find_first_extent_bit(dirty_pages, start, &start, &end,
4684 mark, NULL)) {
4685 clear_extent_bits(dirty_pages, start, end, mark);
4686 while (start <= end) {
4687 eb = find_extent_buffer(fs_info, start);
4688 start += fs_info->nodesize;
4689 if (!eb)
4690 continue;
4691
4692 btrfs_tree_lock(eb);
4693 wait_on_extent_buffer_writeback(eb);
4694 btrfs_clear_buffer_dirty(NULL, eb);
4695 btrfs_tree_unlock(eb);
4696
4697 free_extent_buffer_stale(eb);
4698 }
4699 }
4700}
4701
4702static void btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4703 struct extent_io_tree *unpin)
4704{
4705 u64 start;
4706 u64 end;
4707
4708 while (1) {
4709 struct extent_state *cached_state = NULL;
4710
4711 /*
4712 * The btrfs_finish_extent_commit() may get the same range as
4713 * ours between find_first_extent_bit and clear_extent_dirty.
4714 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4715 * the same extent range.
4716 */
4717 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4718 if (!find_first_extent_bit(unpin, 0, &start, &end,
4719 EXTENT_DIRTY, &cached_state)) {
4720 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4721 break;
4722 }
4723
4724 clear_extent_dirty(unpin, start, end, &cached_state);
4725 free_extent_state(cached_state);
4726 btrfs_error_unpin_extent_range(fs_info, start, end);
4727 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4728 cond_resched();
4729 }
4730}
4731
4732static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4733{
4734 struct inode *inode;
4735
4736 inode = cache->io_ctl.inode;
4737 if (inode) {
4738 unsigned int nofs_flag;
4739
4740 nofs_flag = memalloc_nofs_save();
4741 invalidate_inode_pages2(inode->i_mapping);
4742 memalloc_nofs_restore(nofs_flag);
4743
4744 BTRFS_I(inode)->generation = 0;
4745 cache->io_ctl.inode = NULL;
4746 iput(inode);
4747 }
4748 ASSERT(cache->io_ctl.pages == NULL);
4749 btrfs_put_block_group(cache);
4750}
4751
4752void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4753 struct btrfs_fs_info *fs_info)
4754{
4755 struct btrfs_block_group *cache;
4756
4757 spin_lock(&cur_trans->dirty_bgs_lock);
4758 while (!list_empty(&cur_trans->dirty_bgs)) {
4759 cache = list_first_entry(&cur_trans->dirty_bgs,
4760 struct btrfs_block_group,
4761 dirty_list);
4762
4763 if (!list_empty(&cache->io_list)) {
4764 spin_unlock(&cur_trans->dirty_bgs_lock);
4765 list_del_init(&cache->io_list);
4766 btrfs_cleanup_bg_io(cache);
4767 spin_lock(&cur_trans->dirty_bgs_lock);
4768 }
4769
4770 list_del_init(&cache->dirty_list);
4771 spin_lock(&cache->lock);
4772 cache->disk_cache_state = BTRFS_DC_ERROR;
4773 spin_unlock(&cache->lock);
4774
4775 spin_unlock(&cur_trans->dirty_bgs_lock);
4776 btrfs_put_block_group(cache);
4777 btrfs_dec_delayed_refs_rsv_bg_updates(fs_info);
4778 spin_lock(&cur_trans->dirty_bgs_lock);
4779 }
4780 spin_unlock(&cur_trans->dirty_bgs_lock);
4781
4782 /*
4783 * Refer to the definition of io_bgs member for details why it's safe
4784 * to use it without any locking
4785 */
4786 while (!list_empty(&cur_trans->io_bgs)) {
4787 cache = list_first_entry(&cur_trans->io_bgs,
4788 struct btrfs_block_group,
4789 io_list);
4790
4791 list_del_init(&cache->io_list);
4792 spin_lock(&cache->lock);
4793 cache->disk_cache_state = BTRFS_DC_ERROR;
4794 spin_unlock(&cache->lock);
4795 btrfs_cleanup_bg_io(cache);
4796 }
4797}
4798
4799static void btrfs_free_all_qgroup_pertrans(struct btrfs_fs_info *fs_info)
4800{
4801 struct btrfs_root *gang[8];
4802 int i;
4803 int ret;
4804
4805 spin_lock(&fs_info->fs_roots_radix_lock);
4806 while (1) {
4807 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
4808 (void **)gang, 0,
4809 ARRAY_SIZE(gang),
4810 BTRFS_ROOT_TRANS_TAG);
4811 if (ret == 0)
4812 break;
4813 for (i = 0; i < ret; i++) {
4814 struct btrfs_root *root = gang[i];
4815
4816 btrfs_qgroup_free_meta_all_pertrans(root);
4817 radix_tree_tag_clear(&fs_info->fs_roots_radix,
4818 (unsigned long)root->root_key.objectid,
4819 BTRFS_ROOT_TRANS_TAG);
4820 }
4821 }
4822 spin_unlock(&fs_info->fs_roots_radix_lock);
4823}
4824
4825void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4826 struct btrfs_fs_info *fs_info)
4827{
4828 struct btrfs_device *dev, *tmp;
4829
4830 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4831 ASSERT(list_empty(&cur_trans->dirty_bgs));
4832 ASSERT(list_empty(&cur_trans->io_bgs));
4833
4834 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4835 post_commit_list) {
4836 list_del_init(&dev->post_commit_list);
4837 }
4838
4839 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4840
4841 cur_trans->state = TRANS_STATE_COMMIT_START;
4842 wake_up(&fs_info->transaction_blocked_wait);
4843
4844 cur_trans->state = TRANS_STATE_UNBLOCKED;
4845 wake_up(&fs_info->transaction_wait);
4846
4847 btrfs_destroy_delayed_inodes(fs_info);
4848
4849 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4850 EXTENT_DIRTY);
4851 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4852
4853 btrfs_free_all_qgroup_pertrans(fs_info);
4854
4855 cur_trans->state =TRANS_STATE_COMPLETED;
4856 wake_up(&cur_trans->commit_wait);
4857}
4858
4859static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4860{
4861 struct btrfs_transaction *t;
4862
4863 mutex_lock(&fs_info->transaction_kthread_mutex);
4864
4865 spin_lock(&fs_info->trans_lock);
4866 while (!list_empty(&fs_info->trans_list)) {
4867 t = list_first_entry(&fs_info->trans_list,
4868 struct btrfs_transaction, list);
4869 if (t->state >= TRANS_STATE_COMMIT_PREP) {
4870 refcount_inc(&t->use_count);
4871 spin_unlock(&fs_info->trans_lock);
4872 btrfs_wait_for_commit(fs_info, t->transid);
4873 btrfs_put_transaction(t);
4874 spin_lock(&fs_info->trans_lock);
4875 continue;
4876 }
4877 if (t == fs_info->running_transaction) {
4878 t->state = TRANS_STATE_COMMIT_DOING;
4879 spin_unlock(&fs_info->trans_lock);
4880 /*
4881 * We wait for 0 num_writers since we don't hold a trans
4882 * handle open currently for this transaction.
4883 */
4884 wait_event(t->writer_wait,
4885 atomic_read(&t->num_writers) == 0);
4886 } else {
4887 spin_unlock(&fs_info->trans_lock);
4888 }
4889 btrfs_cleanup_one_transaction(t, fs_info);
4890
4891 spin_lock(&fs_info->trans_lock);
4892 if (t == fs_info->running_transaction)
4893 fs_info->running_transaction = NULL;
4894 list_del_init(&t->list);
4895 spin_unlock(&fs_info->trans_lock);
4896
4897 btrfs_put_transaction(t);
4898 trace_btrfs_transaction_commit(fs_info);
4899 spin_lock(&fs_info->trans_lock);
4900 }
4901 spin_unlock(&fs_info->trans_lock);
4902 btrfs_destroy_all_ordered_extents(fs_info);
4903 btrfs_destroy_delayed_inodes(fs_info);
4904 btrfs_assert_delayed_root_empty(fs_info);
4905 btrfs_destroy_all_delalloc_inodes(fs_info);
4906 btrfs_drop_all_logs(fs_info);
4907 mutex_unlock(&fs_info->transaction_kthread_mutex);
4908
4909 return 0;
4910}
4911
4912int btrfs_init_root_free_objectid(struct btrfs_root *root)
4913{
4914 struct btrfs_path *path;
4915 int ret;
4916 struct extent_buffer *l;
4917 struct btrfs_key search_key;
4918 struct btrfs_key found_key;
4919 int slot;
4920
4921 path = btrfs_alloc_path();
4922 if (!path)
4923 return -ENOMEM;
4924
4925 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4926 search_key.type = -1;
4927 search_key.offset = (u64)-1;
4928 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4929 if (ret < 0)
4930 goto error;
4931 if (ret == 0) {
4932 /*
4933 * Key with offset -1 found, there would have to exist a root
4934 * with such id, but this is out of valid range.
4935 */
4936 ret = -EUCLEAN;
4937 goto error;
4938 }
4939 if (path->slots[0] > 0) {
4940 slot = path->slots[0] - 1;
4941 l = path->nodes[0];
4942 btrfs_item_key_to_cpu(l, &found_key, slot);
4943 root->free_objectid = max_t(u64, found_key.objectid + 1,
4944 BTRFS_FIRST_FREE_OBJECTID);
4945 } else {
4946 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4947 }
4948 ret = 0;
4949error:
4950 btrfs_free_path(path);
4951 return ret;
4952}
4953
4954int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4955{
4956 int ret;
4957 mutex_lock(&root->objectid_mutex);
4958
4959 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4960 btrfs_warn(root->fs_info,
4961 "the objectid of root %llu reaches its highest value",
4962 root->root_key.objectid);
4963 ret = -ENOSPC;
4964 goto out;
4965 }
4966
4967 *objectid = root->free_objectid++;
4968 ret = 0;
4969out:
4970 mutex_unlock(&root->objectid_mutex);
4971 return ret;
4972}