Loading...
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 <linux/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 const 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 const 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 const bool ignore_csum = btrfs_test_opt(fs_info, IGNOREMETACSUMS);
371
372 ASSERT(check);
373
374 found_start = btrfs_header_bytenr(eb);
375 if (found_start != eb->start) {
376 btrfs_err_rl(fs_info,
377 "bad tree block start, mirror %u want %llu have %llu",
378 eb->read_mirror, eb->start, found_start);
379 ret = -EIO;
380 goto out;
381 }
382 if (check_tree_block_fsid(eb)) {
383 btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
384 eb->start, eb->read_mirror);
385 ret = -EIO;
386 goto out;
387 }
388 found_level = btrfs_header_level(eb);
389 if (found_level >= BTRFS_MAX_LEVEL) {
390 btrfs_err(fs_info,
391 "bad tree block level, mirror %u level %d on logical %llu",
392 eb->read_mirror, btrfs_header_level(eb), eb->start);
393 ret = -EIO;
394 goto out;
395 }
396
397 csum_tree_block(eb, result);
398 header_csum = folio_address(eb->folios[0]) +
399 get_eb_offset_in_folio(eb, offsetof(struct btrfs_header, csum));
400
401 if (memcmp(result, header_csum, csum_size) != 0) {
402 btrfs_warn_rl(fs_info,
403"checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d%s",
404 eb->start, eb->read_mirror,
405 CSUM_FMT_VALUE(csum_size, header_csum),
406 CSUM_FMT_VALUE(csum_size, result),
407 btrfs_header_level(eb),
408 ignore_csum ? ", ignored" : "");
409 if (!ignore_csum) {
410 ret = -EUCLEAN;
411 goto out;
412 }
413 }
414
415 if (found_level != check->level) {
416 btrfs_err(fs_info,
417 "level verify failed on logical %llu mirror %u wanted %u found %u",
418 eb->start, eb->read_mirror, check->level, found_level);
419 ret = -EIO;
420 goto out;
421 }
422 if (unlikely(check->transid &&
423 btrfs_header_generation(eb) != check->transid)) {
424 btrfs_err_rl(eb->fs_info,
425"parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
426 eb->start, eb->read_mirror, check->transid,
427 btrfs_header_generation(eb));
428 ret = -EIO;
429 goto out;
430 }
431 if (check->has_first_key) {
432 const struct btrfs_key *expect_key = &check->first_key;
433 struct btrfs_key found_key;
434
435 if (found_level)
436 btrfs_node_key_to_cpu(eb, &found_key, 0);
437 else
438 btrfs_item_key_to_cpu(eb, &found_key, 0);
439 if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
440 btrfs_err(fs_info,
441"tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
442 eb->start, check->transid,
443 expect_key->objectid,
444 expect_key->type, expect_key->offset,
445 found_key.objectid, found_key.type,
446 found_key.offset);
447 ret = -EUCLEAN;
448 goto out;
449 }
450 }
451 if (check->owner_root) {
452 ret = btrfs_check_eb_owner(eb, check->owner_root);
453 if (ret < 0)
454 goto out;
455 }
456
457 /*
458 * If this is a leaf block and it is corrupt, set the corrupt bit so
459 * that we don't try and read the other copies of this block, just
460 * return -EIO.
461 */
462 if (found_level == 0 && btrfs_check_leaf(eb)) {
463 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
464 ret = -EIO;
465 }
466
467 if (found_level > 0 && btrfs_check_node(eb))
468 ret = -EIO;
469
470 if (ret)
471 btrfs_err(fs_info,
472 "read time tree block corruption detected on logical %llu mirror %u",
473 eb->start, eb->read_mirror);
474out:
475 return ret;
476}
477
478#ifdef CONFIG_MIGRATION
479static int btree_migrate_folio(struct address_space *mapping,
480 struct folio *dst, struct folio *src, enum migrate_mode mode)
481{
482 /*
483 * we can't safely write a btree page from here,
484 * we haven't done the locking hook
485 */
486 if (folio_test_dirty(src))
487 return -EAGAIN;
488 /*
489 * Buffers may be managed in a filesystem specific way.
490 * We must have no buffers or drop them.
491 */
492 if (folio_get_private(src) &&
493 !filemap_release_folio(src, GFP_KERNEL))
494 return -EAGAIN;
495 return migrate_folio(mapping, dst, src, mode);
496}
497#else
498#define btree_migrate_folio NULL
499#endif
500
501static int btree_writepages(struct address_space *mapping,
502 struct writeback_control *wbc)
503{
504 int ret;
505
506 if (wbc->sync_mode == WB_SYNC_NONE) {
507 struct btrfs_fs_info *fs_info;
508
509 if (wbc->for_kupdate)
510 return 0;
511
512 fs_info = inode_to_fs_info(mapping->host);
513 /* this is a bit racy, but that's ok */
514 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
515 BTRFS_DIRTY_METADATA_THRESH,
516 fs_info->dirty_metadata_batch);
517 if (ret < 0)
518 return 0;
519 }
520 return btree_write_cache_pages(mapping, wbc);
521}
522
523static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
524{
525 if (folio_test_writeback(folio) || folio_test_dirty(folio))
526 return false;
527
528 return try_release_extent_buffer(folio);
529}
530
531static void btree_invalidate_folio(struct folio *folio, size_t offset,
532 size_t length)
533{
534 struct extent_io_tree *tree;
535
536 tree = &folio_to_inode(folio)->io_tree;
537 extent_invalidate_folio(tree, folio, offset);
538 btree_release_folio(folio, GFP_NOFS);
539 if (folio_get_private(folio)) {
540 btrfs_warn(folio_to_fs_info(folio),
541 "folio private not zero on folio %llu",
542 (unsigned long long)folio_pos(folio));
543 folio_detach_private(folio);
544 }
545}
546
547#ifdef DEBUG
548static bool btree_dirty_folio(struct address_space *mapping,
549 struct folio *folio)
550{
551 struct btrfs_fs_info *fs_info = inode_to_fs_info(mapping->host);
552 struct btrfs_subpage_info *spi = fs_info->subpage_info;
553 struct btrfs_subpage *subpage;
554 struct extent_buffer *eb;
555 int cur_bit = 0;
556 u64 page_start = folio_pos(folio);
557
558 if (fs_info->sectorsize == PAGE_SIZE) {
559 eb = folio_get_private(folio);
560 BUG_ON(!eb);
561 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
562 BUG_ON(!atomic_read(&eb->refs));
563 btrfs_assert_tree_write_locked(eb);
564 return filemap_dirty_folio(mapping, folio);
565 }
566
567 ASSERT(spi);
568 subpage = folio_get_private(folio);
569
570 for (cur_bit = spi->dirty_offset;
571 cur_bit < spi->dirty_offset + spi->bitmap_nr_bits;
572 cur_bit++) {
573 unsigned long flags;
574 u64 cur;
575
576 spin_lock_irqsave(&subpage->lock, flags);
577 if (!test_bit(cur_bit, subpage->bitmaps)) {
578 spin_unlock_irqrestore(&subpage->lock, flags);
579 continue;
580 }
581 spin_unlock_irqrestore(&subpage->lock, flags);
582 cur = page_start + cur_bit * fs_info->sectorsize;
583
584 eb = find_extent_buffer(fs_info, cur);
585 ASSERT(eb);
586 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
587 ASSERT(atomic_read(&eb->refs));
588 btrfs_assert_tree_write_locked(eb);
589 free_extent_buffer(eb);
590
591 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits) - 1;
592 }
593 return filemap_dirty_folio(mapping, folio);
594}
595#else
596#define btree_dirty_folio filemap_dirty_folio
597#endif
598
599static const struct address_space_operations btree_aops = {
600 .writepages = btree_writepages,
601 .release_folio = btree_release_folio,
602 .invalidate_folio = btree_invalidate_folio,
603 .migrate_folio = btree_migrate_folio,
604 .dirty_folio = btree_dirty_folio,
605};
606
607struct extent_buffer *btrfs_find_create_tree_block(
608 struct btrfs_fs_info *fs_info,
609 u64 bytenr, u64 owner_root,
610 int level)
611{
612 if (btrfs_is_testing(fs_info))
613 return alloc_test_extent_buffer(fs_info, bytenr);
614 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
615}
616
617/*
618 * Read tree block at logical address @bytenr and do variant basic but critical
619 * verification.
620 *
621 * @check: expected tree parentness check, see comments of the
622 * structure for details.
623 */
624struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
625 struct btrfs_tree_parent_check *check)
626{
627 struct extent_buffer *buf = NULL;
628 int ret;
629
630 ASSERT(check);
631
632 buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
633 check->level);
634 if (IS_ERR(buf))
635 return buf;
636
637 ret = btrfs_read_extent_buffer(buf, check);
638 if (ret) {
639 free_extent_buffer_stale(buf);
640 return ERR_PTR(ret);
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 = btrfs_is_testing(fs_info);
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 btrfs_set_root_last_trans(root, 0);
662 root->free_objectid = 0;
663 root->nr_delalloc_inodes = 0;
664 root->nr_ordered_extents = 0;
665 xa_init(&root->inodes);
666 xa_init(&root->delayed_nodes);
667
668 btrfs_init_root_block_rsv(root);
669
670 INIT_LIST_HEAD(&root->dirty_list);
671 INIT_LIST_HEAD(&root->root_list);
672 INIT_LIST_HEAD(&root->delalloc_inodes);
673 INIT_LIST_HEAD(&root->delalloc_root);
674 INIT_LIST_HEAD(&root->ordered_extents);
675 INIT_LIST_HEAD(&root->ordered_root);
676 INIT_LIST_HEAD(&root->reloc_dirty_list);
677 spin_lock_init(&root->delalloc_lock);
678 spin_lock_init(&root->ordered_extent_lock);
679 spin_lock_init(&root->accounting_lock);
680 spin_lock_init(&root->qgroup_meta_rsv_lock);
681 mutex_init(&root->objectid_mutex);
682 mutex_init(&root->log_mutex);
683 mutex_init(&root->ordered_extent_mutex);
684 mutex_init(&root->delalloc_mutex);
685 init_waitqueue_head(&root->qgroup_flush_wait);
686 init_waitqueue_head(&root->log_writer_wait);
687 init_waitqueue_head(&root->log_commit_wait[0]);
688 init_waitqueue_head(&root->log_commit_wait[1]);
689 INIT_LIST_HEAD(&root->log_ctxs[0]);
690 INIT_LIST_HEAD(&root->log_ctxs[1]);
691 atomic_set(&root->log_commit[0], 0);
692 atomic_set(&root->log_commit[1], 0);
693 atomic_set(&root->log_writers, 0);
694 atomic_set(&root->log_batch, 0);
695 refcount_set(&root->refs, 1);
696 atomic_set(&root->snapshot_force_cow, 0);
697 atomic_set(&root->nr_swapfiles, 0);
698 btrfs_set_root_log_transid(root, 0);
699 root->log_transid_committed = -1;
700 btrfs_set_root_last_log_commit(root, 0);
701 root->anon_dev = 0;
702 if (!dummy) {
703 extent_io_tree_init(fs_info, &root->dirty_log_pages,
704 IO_TREE_ROOT_DIRTY_LOG_PAGES);
705 extent_io_tree_init(fs_info, &root->log_csum_range,
706 IO_TREE_LOG_CSUM_RANGE);
707 }
708
709 spin_lock_init(&root->root_item_lock);
710 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
711#ifdef CONFIG_BTRFS_DEBUG
712 INIT_LIST_HEAD(&root->leak_list);
713 spin_lock(&fs_info->fs_roots_radix_lock);
714 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
715 spin_unlock(&fs_info->fs_roots_radix_lock);
716#endif
717}
718
719static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
720 u64 objectid, gfp_t flags)
721{
722 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
723 if (root)
724 __setup_root(root, fs_info, objectid);
725 return root;
726}
727
728#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
729/* Should only be used by the testing infrastructure */
730struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
731{
732 struct btrfs_root *root;
733
734 if (!fs_info)
735 return ERR_PTR(-EINVAL);
736
737 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
738 if (!root)
739 return ERR_PTR(-ENOMEM);
740
741 /* We don't use the stripesize in selftest, set it as sectorsize */
742 root->alloc_bytenr = 0;
743
744 return root;
745}
746#endif
747
748static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
749{
750 const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
751 const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
752
753 return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
754}
755
756static int global_root_key_cmp(const void *k, const struct rb_node *node)
757{
758 const struct btrfs_key *key = k;
759 const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
760
761 return btrfs_comp_cpu_keys(key, &root->root_key);
762}
763
764int btrfs_global_root_insert(struct btrfs_root *root)
765{
766 struct btrfs_fs_info *fs_info = root->fs_info;
767 struct rb_node *tmp;
768 int ret = 0;
769
770 write_lock(&fs_info->global_root_lock);
771 tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
772 write_unlock(&fs_info->global_root_lock);
773
774 if (tmp) {
775 ret = -EEXIST;
776 btrfs_warn(fs_info, "global root %llu %llu already exists",
777 btrfs_root_id(root), root->root_key.offset);
778 }
779 return ret;
780}
781
782void btrfs_global_root_delete(struct btrfs_root *root)
783{
784 struct btrfs_fs_info *fs_info = root->fs_info;
785
786 write_lock(&fs_info->global_root_lock);
787 rb_erase(&root->rb_node, &fs_info->global_root_tree);
788 write_unlock(&fs_info->global_root_lock);
789}
790
791struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
792 struct btrfs_key *key)
793{
794 struct rb_node *node;
795 struct btrfs_root *root = NULL;
796
797 read_lock(&fs_info->global_root_lock);
798 node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
799 if (node)
800 root = container_of(node, struct btrfs_root, rb_node);
801 read_unlock(&fs_info->global_root_lock);
802
803 return root;
804}
805
806static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
807{
808 struct btrfs_block_group *block_group;
809 u64 ret;
810
811 if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
812 return 0;
813
814 if (bytenr)
815 block_group = btrfs_lookup_block_group(fs_info, bytenr);
816 else
817 block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
818 ASSERT(block_group);
819 if (!block_group)
820 return 0;
821 ret = block_group->global_root_id;
822 btrfs_put_block_group(block_group);
823
824 return ret;
825}
826
827struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
828{
829 struct btrfs_key key = {
830 .objectid = BTRFS_CSUM_TREE_OBJECTID,
831 .type = BTRFS_ROOT_ITEM_KEY,
832 .offset = btrfs_global_root_id(fs_info, bytenr),
833 };
834
835 return btrfs_global_root(fs_info, &key);
836}
837
838struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
839{
840 struct btrfs_key key = {
841 .objectid = BTRFS_EXTENT_TREE_OBJECTID,
842 .type = BTRFS_ROOT_ITEM_KEY,
843 .offset = btrfs_global_root_id(fs_info, bytenr),
844 };
845
846 return btrfs_global_root(fs_info, &key);
847}
848
849struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
850 u64 objectid)
851{
852 struct btrfs_fs_info *fs_info = trans->fs_info;
853 struct extent_buffer *leaf;
854 struct btrfs_root *tree_root = fs_info->tree_root;
855 struct btrfs_root *root;
856 struct btrfs_key key;
857 unsigned int nofs_flag;
858 int ret = 0;
859
860 /*
861 * We're holding a transaction handle, so use a NOFS memory allocation
862 * context to avoid deadlock if reclaim happens.
863 */
864 nofs_flag = memalloc_nofs_save();
865 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
866 memalloc_nofs_restore(nofs_flag);
867 if (!root)
868 return ERR_PTR(-ENOMEM);
869
870 root->root_key.objectid = objectid;
871 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
872 root->root_key.offset = 0;
873
874 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
875 0, BTRFS_NESTING_NORMAL);
876 if (IS_ERR(leaf)) {
877 ret = PTR_ERR(leaf);
878 leaf = NULL;
879 goto fail;
880 }
881
882 root->node = leaf;
883 btrfs_mark_buffer_dirty(trans, leaf);
884
885 root->commit_root = btrfs_root_node(root);
886 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
887
888 btrfs_set_root_flags(&root->root_item, 0);
889 btrfs_set_root_limit(&root->root_item, 0);
890 btrfs_set_root_bytenr(&root->root_item, leaf->start);
891 btrfs_set_root_generation(&root->root_item, trans->transid);
892 btrfs_set_root_level(&root->root_item, 0);
893 btrfs_set_root_refs(&root->root_item, 1);
894 btrfs_set_root_used(&root->root_item, leaf->len);
895 btrfs_set_root_last_snapshot(&root->root_item, 0);
896 btrfs_set_root_dirid(&root->root_item, 0);
897 if (is_fstree(objectid))
898 generate_random_guid(root->root_item.uuid);
899 else
900 export_guid(root->root_item.uuid, &guid_null);
901 btrfs_set_root_drop_level(&root->root_item, 0);
902
903 btrfs_tree_unlock(leaf);
904
905 key.objectid = objectid;
906 key.type = BTRFS_ROOT_ITEM_KEY;
907 key.offset = 0;
908 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
909 if (ret)
910 goto fail;
911
912 return root;
913
914fail:
915 btrfs_put_root(root);
916
917 return ERR_PTR(ret);
918}
919
920static struct btrfs_root *alloc_log_tree(struct btrfs_fs_info *fs_info)
921{
922 struct btrfs_root *root;
923
924 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
925 if (!root)
926 return ERR_PTR(-ENOMEM);
927
928 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
929 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
930 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
931
932 return root;
933}
934
935int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
936 struct btrfs_root *root)
937{
938 struct extent_buffer *leaf;
939
940 /*
941 * DON'T set SHAREABLE bit for log trees.
942 *
943 * Log trees are not exposed to user space thus can't be snapshotted,
944 * and they go away before a real commit is actually done.
945 *
946 * They do store pointers to file data extents, and those reference
947 * counts still get updated (along with back refs to the log tree).
948 */
949
950 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
951 NULL, 0, 0, 0, 0, BTRFS_NESTING_NORMAL);
952 if (IS_ERR(leaf))
953 return PTR_ERR(leaf);
954
955 root->node = leaf;
956
957 btrfs_mark_buffer_dirty(trans, root->node);
958 btrfs_tree_unlock(root->node);
959
960 return 0;
961}
962
963int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
964 struct btrfs_fs_info *fs_info)
965{
966 struct btrfs_root *log_root;
967
968 log_root = alloc_log_tree(fs_info);
969 if (IS_ERR(log_root))
970 return PTR_ERR(log_root);
971
972 if (!btrfs_is_zoned(fs_info)) {
973 int ret = btrfs_alloc_log_tree_node(trans, log_root);
974
975 if (ret) {
976 btrfs_put_root(log_root);
977 return ret;
978 }
979 }
980
981 WARN_ON(fs_info->log_root_tree);
982 fs_info->log_root_tree = log_root;
983 return 0;
984}
985
986int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
987 struct btrfs_root *root)
988{
989 struct btrfs_fs_info *fs_info = root->fs_info;
990 struct btrfs_root *log_root;
991 struct btrfs_inode_item *inode_item;
992 int ret;
993
994 log_root = alloc_log_tree(fs_info);
995 if (IS_ERR(log_root))
996 return PTR_ERR(log_root);
997
998 ret = btrfs_alloc_log_tree_node(trans, log_root);
999 if (ret) {
1000 btrfs_put_root(log_root);
1001 return ret;
1002 }
1003
1004 btrfs_set_root_last_trans(log_root, trans->transid);
1005 log_root->root_key.offset = btrfs_root_id(root);
1006
1007 inode_item = &log_root->root_item.inode;
1008 btrfs_set_stack_inode_generation(inode_item, 1);
1009 btrfs_set_stack_inode_size(inode_item, 3);
1010 btrfs_set_stack_inode_nlink(inode_item, 1);
1011 btrfs_set_stack_inode_nbytes(inode_item,
1012 fs_info->nodesize);
1013 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1014
1015 btrfs_set_root_node(&log_root->root_item, log_root->node);
1016
1017 WARN_ON(root->log_root);
1018 root->log_root = log_root;
1019 btrfs_set_root_log_transid(root, 0);
1020 root->log_transid_committed = -1;
1021 btrfs_set_root_last_log_commit(root, 0);
1022 return 0;
1023}
1024
1025static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1026 struct btrfs_path *path,
1027 const struct btrfs_key *key)
1028{
1029 struct btrfs_root *root;
1030 struct btrfs_tree_parent_check check = { 0 };
1031 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1032 u64 generation;
1033 int ret;
1034 int level;
1035
1036 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1037 if (!root)
1038 return ERR_PTR(-ENOMEM);
1039
1040 ret = btrfs_find_root(tree_root, key, path,
1041 &root->root_item, &root->root_key);
1042 if (ret) {
1043 if (ret > 0)
1044 ret = -ENOENT;
1045 goto fail;
1046 }
1047
1048 generation = btrfs_root_generation(&root->root_item);
1049 level = btrfs_root_level(&root->root_item);
1050 check.level = level;
1051 check.transid = generation;
1052 check.owner_root = key->objectid;
1053 root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item),
1054 &check);
1055 if (IS_ERR(root->node)) {
1056 ret = PTR_ERR(root->node);
1057 root->node = NULL;
1058 goto fail;
1059 }
1060 if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1061 ret = -EIO;
1062 goto fail;
1063 }
1064
1065 /*
1066 * For real fs, and not log/reloc trees, root owner must
1067 * match its root node owner
1068 */
1069 if (!btrfs_is_testing(fs_info) &&
1070 btrfs_root_id(root) != BTRFS_TREE_LOG_OBJECTID &&
1071 btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID &&
1072 btrfs_root_id(root) != btrfs_header_owner(root->node)) {
1073 btrfs_crit(fs_info,
1074"root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1075 btrfs_root_id(root), root->node->start,
1076 btrfs_header_owner(root->node),
1077 btrfs_root_id(root));
1078 ret = -EUCLEAN;
1079 goto fail;
1080 }
1081 root->commit_root = btrfs_root_node(root);
1082 return root;
1083fail:
1084 btrfs_put_root(root);
1085 return ERR_PTR(ret);
1086}
1087
1088struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1089 const struct btrfs_key *key)
1090{
1091 struct btrfs_root *root;
1092 struct btrfs_path *path;
1093
1094 path = btrfs_alloc_path();
1095 if (!path)
1096 return ERR_PTR(-ENOMEM);
1097 root = read_tree_root_path(tree_root, path, key);
1098 btrfs_free_path(path);
1099
1100 return root;
1101}
1102
1103/*
1104 * Initialize subvolume root in-memory structure
1105 *
1106 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1107 */
1108static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1109{
1110 int ret;
1111
1112 btrfs_drew_lock_init(&root->snapshot_lock);
1113
1114 if (btrfs_root_id(root) != BTRFS_TREE_LOG_OBJECTID &&
1115 !btrfs_is_data_reloc_root(root) &&
1116 is_fstree(btrfs_root_id(root))) {
1117 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1118 btrfs_check_and_init_root_item(&root->root_item);
1119 }
1120
1121 /*
1122 * Don't assign anonymous block device to roots that are not exposed to
1123 * userspace, the id pool is limited to 1M
1124 */
1125 if (is_fstree(btrfs_root_id(root)) &&
1126 btrfs_root_refs(&root->root_item) > 0) {
1127 if (!anon_dev) {
1128 ret = get_anon_bdev(&root->anon_dev);
1129 if (ret)
1130 goto fail;
1131 } else {
1132 root->anon_dev = anon_dev;
1133 }
1134 }
1135
1136 mutex_lock(&root->objectid_mutex);
1137 ret = btrfs_init_root_free_objectid(root);
1138 if (ret) {
1139 mutex_unlock(&root->objectid_mutex);
1140 goto fail;
1141 }
1142
1143 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1144
1145 mutex_unlock(&root->objectid_mutex);
1146
1147 return 0;
1148fail:
1149 /* The caller is responsible to call btrfs_free_fs_root */
1150 return ret;
1151}
1152
1153static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1154 u64 root_id)
1155{
1156 struct btrfs_root *root;
1157
1158 spin_lock(&fs_info->fs_roots_radix_lock);
1159 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1160 (unsigned long)root_id);
1161 root = btrfs_grab_root(root);
1162 spin_unlock(&fs_info->fs_roots_radix_lock);
1163 return root;
1164}
1165
1166static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1167 u64 objectid)
1168{
1169 struct btrfs_key key = {
1170 .objectid = objectid,
1171 .type = BTRFS_ROOT_ITEM_KEY,
1172 .offset = 0,
1173 };
1174
1175 switch (objectid) {
1176 case BTRFS_ROOT_TREE_OBJECTID:
1177 return btrfs_grab_root(fs_info->tree_root);
1178 case BTRFS_EXTENT_TREE_OBJECTID:
1179 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1180 case BTRFS_CHUNK_TREE_OBJECTID:
1181 return btrfs_grab_root(fs_info->chunk_root);
1182 case BTRFS_DEV_TREE_OBJECTID:
1183 return btrfs_grab_root(fs_info->dev_root);
1184 case BTRFS_CSUM_TREE_OBJECTID:
1185 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1186 case BTRFS_QUOTA_TREE_OBJECTID:
1187 return btrfs_grab_root(fs_info->quota_root);
1188 case BTRFS_UUID_TREE_OBJECTID:
1189 return btrfs_grab_root(fs_info->uuid_root);
1190 case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
1191 return btrfs_grab_root(fs_info->block_group_root);
1192 case BTRFS_FREE_SPACE_TREE_OBJECTID:
1193 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1194 case BTRFS_RAID_STRIPE_TREE_OBJECTID:
1195 return btrfs_grab_root(fs_info->stripe_root);
1196 default:
1197 return NULL;
1198 }
1199}
1200
1201int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1202 struct btrfs_root *root)
1203{
1204 int ret;
1205
1206 ret = radix_tree_preload(GFP_NOFS);
1207 if (ret)
1208 return ret;
1209
1210 spin_lock(&fs_info->fs_roots_radix_lock);
1211 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1212 (unsigned long)btrfs_root_id(root),
1213 root);
1214 if (ret == 0) {
1215 btrfs_grab_root(root);
1216 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1217 }
1218 spin_unlock(&fs_info->fs_roots_radix_lock);
1219 radix_tree_preload_end();
1220
1221 return ret;
1222}
1223
1224void btrfs_check_leaked_roots(const struct btrfs_fs_info *fs_info)
1225{
1226#ifdef CONFIG_BTRFS_DEBUG
1227 struct btrfs_root *root;
1228
1229 while (!list_empty(&fs_info->allocated_roots)) {
1230 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1231
1232 root = list_first_entry(&fs_info->allocated_roots,
1233 struct btrfs_root, leak_list);
1234 btrfs_err(fs_info, "leaked root %s refcount %d",
1235 btrfs_root_name(&root->root_key, buf),
1236 refcount_read(&root->refs));
1237 WARN_ON_ONCE(1);
1238 while (refcount_read(&root->refs) > 1)
1239 btrfs_put_root(root);
1240 btrfs_put_root(root);
1241 }
1242#endif
1243}
1244
1245static void free_global_roots(struct btrfs_fs_info *fs_info)
1246{
1247 struct btrfs_root *root;
1248 struct rb_node *node;
1249
1250 while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1251 root = rb_entry(node, struct btrfs_root, rb_node);
1252 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1253 btrfs_put_root(root);
1254 }
1255}
1256
1257void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1258{
1259 struct percpu_counter *em_counter = &fs_info->evictable_extent_maps;
1260
1261 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1262 percpu_counter_destroy(&fs_info->delalloc_bytes);
1263 percpu_counter_destroy(&fs_info->ordered_bytes);
1264 if (percpu_counter_initialized(em_counter))
1265 ASSERT(percpu_counter_sum_positive(em_counter) == 0);
1266 percpu_counter_destroy(em_counter);
1267 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1268 btrfs_free_csum_hash(fs_info);
1269 btrfs_free_stripe_hash_table(fs_info);
1270 btrfs_free_ref_cache(fs_info);
1271 kfree(fs_info->balance_ctl);
1272 kfree(fs_info->delayed_root);
1273 free_global_roots(fs_info);
1274 btrfs_put_root(fs_info->tree_root);
1275 btrfs_put_root(fs_info->chunk_root);
1276 btrfs_put_root(fs_info->dev_root);
1277 btrfs_put_root(fs_info->quota_root);
1278 btrfs_put_root(fs_info->uuid_root);
1279 btrfs_put_root(fs_info->fs_root);
1280 btrfs_put_root(fs_info->data_reloc_root);
1281 btrfs_put_root(fs_info->block_group_root);
1282 btrfs_put_root(fs_info->stripe_root);
1283 btrfs_check_leaked_roots(fs_info);
1284 btrfs_extent_buffer_leak_debug_check(fs_info);
1285 kfree(fs_info->super_copy);
1286 kfree(fs_info->super_for_commit);
1287 kvfree(fs_info);
1288}
1289
1290
1291/*
1292 * Get an in-memory reference of a root structure.
1293 *
1294 * For essential trees like root/extent tree, we grab it from fs_info directly.
1295 * For subvolume trees, we check the cached filesystem roots first. If not
1296 * found, then read it from disk and add it to cached fs roots.
1297 *
1298 * Caller should release the root by calling btrfs_put_root() after the usage.
1299 *
1300 * NOTE: Reloc and log trees can't be read by this function as they share the
1301 * same root objectid.
1302 *
1303 * @objectid: root id
1304 * @anon_dev: preallocated anonymous block device number for new roots,
1305 * pass NULL for a new allocation.
1306 * @check_ref: whether to check root item references, If true, return -ENOENT
1307 * for orphan roots
1308 */
1309static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1310 u64 objectid, dev_t *anon_dev,
1311 bool check_ref)
1312{
1313 struct btrfs_root *root;
1314 struct btrfs_path *path;
1315 struct btrfs_key key;
1316 int ret;
1317
1318 root = btrfs_get_global_root(fs_info, objectid);
1319 if (root)
1320 return root;
1321
1322 /*
1323 * If we're called for non-subvolume trees, and above function didn't
1324 * find one, do not try to read it from disk.
1325 *
1326 * This is namely for free-space-tree and quota tree, which can change
1327 * at runtime and should only be grabbed from fs_info.
1328 */
1329 if (!is_fstree(objectid) && objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1330 return ERR_PTR(-ENOENT);
1331again:
1332 root = btrfs_lookup_fs_root(fs_info, objectid);
1333 if (root) {
1334 /*
1335 * Some other caller may have read out the newly inserted
1336 * subvolume already (for things like backref walk etc). Not
1337 * that common but still possible. In that case, we just need
1338 * to free the anon_dev.
1339 */
1340 if (unlikely(anon_dev && *anon_dev)) {
1341 free_anon_bdev(*anon_dev);
1342 *anon_dev = 0;
1343 }
1344
1345 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1346 btrfs_put_root(root);
1347 return ERR_PTR(-ENOENT);
1348 }
1349 return root;
1350 }
1351
1352 key.objectid = objectid;
1353 key.type = BTRFS_ROOT_ITEM_KEY;
1354 key.offset = (u64)-1;
1355 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1356 if (IS_ERR(root))
1357 return root;
1358
1359 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1360 ret = -ENOENT;
1361 goto fail;
1362 }
1363
1364 ret = btrfs_init_fs_root(root, anon_dev ? *anon_dev : 0);
1365 if (ret)
1366 goto fail;
1367
1368 path = btrfs_alloc_path();
1369 if (!path) {
1370 ret = -ENOMEM;
1371 goto fail;
1372 }
1373 key.objectid = BTRFS_ORPHAN_OBJECTID;
1374 key.type = BTRFS_ORPHAN_ITEM_KEY;
1375 key.offset = objectid;
1376
1377 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1378 btrfs_free_path(path);
1379 if (ret < 0)
1380 goto fail;
1381 if (ret == 0)
1382 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1383
1384 ret = btrfs_insert_fs_root(fs_info, root);
1385 if (ret) {
1386 if (ret == -EEXIST) {
1387 btrfs_put_root(root);
1388 goto again;
1389 }
1390 goto fail;
1391 }
1392 return root;
1393fail:
1394 /*
1395 * If our caller provided us an anonymous device, then it's his
1396 * responsibility to free it in case we fail. So we have to set our
1397 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1398 * and once again by our caller.
1399 */
1400 if (anon_dev && *anon_dev)
1401 root->anon_dev = 0;
1402 btrfs_put_root(root);
1403 return ERR_PTR(ret);
1404}
1405
1406/*
1407 * Get in-memory reference of a root structure
1408 *
1409 * @objectid: tree objectid
1410 * @check_ref: if set, verify that the tree exists and the item has at least
1411 * one reference
1412 */
1413struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1414 u64 objectid, bool check_ref)
1415{
1416 return btrfs_get_root_ref(fs_info, objectid, NULL, check_ref);
1417}
1418
1419/*
1420 * Get in-memory reference of a root structure, created as new, optionally pass
1421 * the anonymous block device id
1422 *
1423 * @objectid: tree objectid
1424 * @anon_dev: if NULL, allocate a new anonymous block device or use the
1425 * parameter value if not NULL
1426 */
1427struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1428 u64 objectid, dev_t *anon_dev)
1429{
1430 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1431}
1432
1433/*
1434 * Return a root for the given objectid.
1435 *
1436 * @fs_info: the fs_info
1437 * @objectid: the objectid we need to lookup
1438 *
1439 * This is exclusively used for backref walking, and exists specifically because
1440 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1441 * creation time, which means we may have to read the tree_root in order to look
1442 * up a fs root that is not in memory. If the root is not in memory we will
1443 * read the tree root commit root and look up the fs root from there. This is a
1444 * temporary root, it will not be inserted into the radix tree as it doesn't
1445 * have the most uptodate information, it'll simply be discarded once the
1446 * backref code is finished using the root.
1447 */
1448struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1449 struct btrfs_path *path,
1450 u64 objectid)
1451{
1452 struct btrfs_root *root;
1453 struct btrfs_key key;
1454
1455 ASSERT(path->search_commit_root && path->skip_locking);
1456
1457 /*
1458 * This can return -ENOENT if we ask for a root that doesn't exist, but
1459 * since this is called via the backref walking code we won't be looking
1460 * up a root that doesn't exist, unless there's corruption. So if root
1461 * != NULL just return it.
1462 */
1463 root = btrfs_get_global_root(fs_info, objectid);
1464 if (root)
1465 return root;
1466
1467 root = btrfs_lookup_fs_root(fs_info, objectid);
1468 if (root)
1469 return root;
1470
1471 key.objectid = objectid;
1472 key.type = BTRFS_ROOT_ITEM_KEY;
1473 key.offset = (u64)-1;
1474 root = read_tree_root_path(fs_info->tree_root, path, &key);
1475 btrfs_release_path(path);
1476
1477 return root;
1478}
1479
1480static int cleaner_kthread(void *arg)
1481{
1482 struct btrfs_fs_info *fs_info = arg;
1483 int again;
1484
1485 while (1) {
1486 again = 0;
1487
1488 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1489
1490 /* Make the cleaner go to sleep early. */
1491 if (btrfs_need_cleaner_sleep(fs_info))
1492 goto sleep;
1493
1494 /*
1495 * Do not do anything if we might cause open_ctree() to block
1496 * before we have finished mounting the filesystem.
1497 */
1498 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1499 goto sleep;
1500
1501 if (!mutex_trylock(&fs_info->cleaner_mutex))
1502 goto sleep;
1503
1504 /*
1505 * Avoid the problem that we change the status of the fs
1506 * during the above check and trylock.
1507 */
1508 if (btrfs_need_cleaner_sleep(fs_info)) {
1509 mutex_unlock(&fs_info->cleaner_mutex);
1510 goto sleep;
1511 }
1512
1513 if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags))
1514 btrfs_sysfs_feature_update(fs_info);
1515
1516 btrfs_run_delayed_iputs(fs_info);
1517
1518 again = btrfs_clean_one_deleted_snapshot(fs_info);
1519 mutex_unlock(&fs_info->cleaner_mutex);
1520
1521 /*
1522 * The defragger has dealt with the R/O remount and umount,
1523 * needn't do anything special here.
1524 */
1525 btrfs_run_defrag_inodes(fs_info);
1526
1527 /*
1528 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1529 * with relocation (btrfs_relocate_chunk) and relocation
1530 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1531 * after acquiring fs_info->reclaim_bgs_lock. So we
1532 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1533 * unused block groups.
1534 */
1535 btrfs_delete_unused_bgs(fs_info);
1536
1537 /*
1538 * Reclaim block groups in the reclaim_bgs list after we deleted
1539 * all unused block_groups. This possibly gives us some more free
1540 * space.
1541 */
1542 btrfs_reclaim_bgs(fs_info);
1543sleep:
1544 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1545 if (kthread_should_park())
1546 kthread_parkme();
1547 if (kthread_should_stop())
1548 return 0;
1549 if (!again) {
1550 set_current_state(TASK_INTERRUPTIBLE);
1551 schedule();
1552 __set_current_state(TASK_RUNNING);
1553 }
1554 }
1555}
1556
1557static int transaction_kthread(void *arg)
1558{
1559 struct btrfs_root *root = arg;
1560 struct btrfs_fs_info *fs_info = root->fs_info;
1561 struct btrfs_trans_handle *trans;
1562 struct btrfs_transaction *cur;
1563 u64 transid;
1564 time64_t delta;
1565 unsigned long delay;
1566 bool cannot_commit;
1567
1568 do {
1569 cannot_commit = false;
1570 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1571 mutex_lock(&fs_info->transaction_kthread_mutex);
1572
1573 spin_lock(&fs_info->trans_lock);
1574 cur = fs_info->running_transaction;
1575 if (!cur) {
1576 spin_unlock(&fs_info->trans_lock);
1577 goto sleep;
1578 }
1579
1580 delta = ktime_get_seconds() - cur->start_time;
1581 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1582 cur->state < TRANS_STATE_COMMIT_PREP &&
1583 delta < fs_info->commit_interval) {
1584 spin_unlock(&fs_info->trans_lock);
1585 delay -= msecs_to_jiffies((delta - 1) * 1000);
1586 delay = min(delay,
1587 msecs_to_jiffies(fs_info->commit_interval * 1000));
1588 goto sleep;
1589 }
1590 transid = cur->transid;
1591 spin_unlock(&fs_info->trans_lock);
1592
1593 /* If the file system is aborted, this will always fail. */
1594 trans = btrfs_attach_transaction(root);
1595 if (IS_ERR(trans)) {
1596 if (PTR_ERR(trans) != -ENOENT)
1597 cannot_commit = true;
1598 goto sleep;
1599 }
1600 if (transid == trans->transid) {
1601 btrfs_commit_transaction(trans);
1602 } else {
1603 btrfs_end_transaction(trans);
1604 }
1605sleep:
1606 wake_up_process(fs_info->cleaner_kthread);
1607 mutex_unlock(&fs_info->transaction_kthread_mutex);
1608
1609 if (BTRFS_FS_ERROR(fs_info))
1610 btrfs_cleanup_transaction(fs_info);
1611 if (!kthread_should_stop() &&
1612 (!btrfs_transaction_blocked(fs_info) ||
1613 cannot_commit))
1614 schedule_timeout_interruptible(delay);
1615 } while (!kthread_should_stop());
1616 return 0;
1617}
1618
1619/*
1620 * This will find the highest generation in the array of root backups. The
1621 * index of the highest array is returned, or -EINVAL if we can't find
1622 * anything.
1623 *
1624 * We check to make sure the array is valid by comparing the
1625 * generation of the latest root in the array with the generation
1626 * in the super block. If they don't match we pitch it.
1627 */
1628static int find_newest_super_backup(struct btrfs_fs_info *info)
1629{
1630 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1631 u64 cur;
1632 struct btrfs_root_backup *root_backup;
1633 int i;
1634
1635 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1636 root_backup = info->super_copy->super_roots + i;
1637 cur = btrfs_backup_tree_root_gen(root_backup);
1638 if (cur == newest_gen)
1639 return i;
1640 }
1641
1642 return -EINVAL;
1643}
1644
1645/*
1646 * copy all the root pointers into the super backup array.
1647 * this will bump the backup pointer by one when it is
1648 * done
1649 */
1650static void backup_super_roots(struct btrfs_fs_info *info)
1651{
1652 const int next_backup = info->backup_root_index;
1653 struct btrfs_root_backup *root_backup;
1654
1655 root_backup = info->super_for_commit->super_roots + next_backup;
1656
1657 /*
1658 * make sure all of our padding and empty slots get zero filled
1659 * regardless of which ones we use today
1660 */
1661 memset(root_backup, 0, sizeof(*root_backup));
1662
1663 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1664
1665 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1666 btrfs_set_backup_tree_root_gen(root_backup,
1667 btrfs_header_generation(info->tree_root->node));
1668
1669 btrfs_set_backup_tree_root_level(root_backup,
1670 btrfs_header_level(info->tree_root->node));
1671
1672 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1673 btrfs_set_backup_chunk_root_gen(root_backup,
1674 btrfs_header_generation(info->chunk_root->node));
1675 btrfs_set_backup_chunk_root_level(root_backup,
1676 btrfs_header_level(info->chunk_root->node));
1677
1678 if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
1679 struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
1680 struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
1681
1682 btrfs_set_backup_extent_root(root_backup,
1683 extent_root->node->start);
1684 btrfs_set_backup_extent_root_gen(root_backup,
1685 btrfs_header_generation(extent_root->node));
1686 btrfs_set_backup_extent_root_level(root_backup,
1687 btrfs_header_level(extent_root->node));
1688
1689 btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
1690 btrfs_set_backup_csum_root_gen(root_backup,
1691 btrfs_header_generation(csum_root->node));
1692 btrfs_set_backup_csum_root_level(root_backup,
1693 btrfs_header_level(csum_root->node));
1694 }
1695
1696 /*
1697 * we might commit during log recovery, which happens before we set
1698 * the fs_root. Make sure it is valid before we fill it in.
1699 */
1700 if (info->fs_root && info->fs_root->node) {
1701 btrfs_set_backup_fs_root(root_backup,
1702 info->fs_root->node->start);
1703 btrfs_set_backup_fs_root_gen(root_backup,
1704 btrfs_header_generation(info->fs_root->node));
1705 btrfs_set_backup_fs_root_level(root_backup,
1706 btrfs_header_level(info->fs_root->node));
1707 }
1708
1709 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1710 btrfs_set_backup_dev_root_gen(root_backup,
1711 btrfs_header_generation(info->dev_root->node));
1712 btrfs_set_backup_dev_root_level(root_backup,
1713 btrfs_header_level(info->dev_root->node));
1714
1715 btrfs_set_backup_total_bytes(root_backup,
1716 btrfs_super_total_bytes(info->super_copy));
1717 btrfs_set_backup_bytes_used(root_backup,
1718 btrfs_super_bytes_used(info->super_copy));
1719 btrfs_set_backup_num_devices(root_backup,
1720 btrfs_super_num_devices(info->super_copy));
1721
1722 /*
1723 * if we don't copy this out to the super_copy, it won't get remembered
1724 * for the next commit
1725 */
1726 memcpy(&info->super_copy->super_roots,
1727 &info->super_for_commit->super_roots,
1728 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1729}
1730
1731/*
1732 * Reads a backup root based on the passed priority. Prio 0 is the newest, prio
1733 * 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1734 *
1735 * @fs_info: filesystem whose backup roots need to be read
1736 * @priority: priority of backup root required
1737 *
1738 * Returns backup root index on success and -EINVAL otherwise.
1739 */
1740static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1741{
1742 int backup_index = find_newest_super_backup(fs_info);
1743 struct btrfs_super_block *super = fs_info->super_copy;
1744 struct btrfs_root_backup *root_backup;
1745
1746 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1747 if (priority == 0)
1748 return backup_index;
1749
1750 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1751 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1752 } else {
1753 return -EINVAL;
1754 }
1755
1756 root_backup = super->super_roots + backup_index;
1757
1758 btrfs_set_super_generation(super,
1759 btrfs_backup_tree_root_gen(root_backup));
1760 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1761 btrfs_set_super_root_level(super,
1762 btrfs_backup_tree_root_level(root_backup));
1763 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1764
1765 /*
1766 * Fixme: the total bytes and num_devices need to match or we should
1767 * need a fsck
1768 */
1769 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1770 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1771
1772 return backup_index;
1773}
1774
1775/* helper to cleanup workers */
1776static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1777{
1778 btrfs_destroy_workqueue(fs_info->fixup_workers);
1779 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1780 btrfs_destroy_workqueue(fs_info->workers);
1781 if (fs_info->endio_workers)
1782 destroy_workqueue(fs_info->endio_workers);
1783 if (fs_info->rmw_workers)
1784 destroy_workqueue(fs_info->rmw_workers);
1785 if (fs_info->compressed_write_workers)
1786 destroy_workqueue(fs_info->compressed_write_workers);
1787 btrfs_destroy_workqueue(fs_info->endio_write_workers);
1788 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1789 btrfs_destroy_workqueue(fs_info->delayed_workers);
1790 btrfs_destroy_workqueue(fs_info->caching_workers);
1791 btrfs_destroy_workqueue(fs_info->flush_workers);
1792 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1793 if (fs_info->discard_ctl.discard_workers)
1794 destroy_workqueue(fs_info->discard_ctl.discard_workers);
1795 /*
1796 * Now that all other work queues are destroyed, we can safely destroy
1797 * the queues used for metadata I/O, since tasks from those other work
1798 * queues can do metadata I/O operations.
1799 */
1800 if (fs_info->endio_meta_workers)
1801 destroy_workqueue(fs_info->endio_meta_workers);
1802}
1803
1804static void free_root_extent_buffers(struct btrfs_root *root)
1805{
1806 if (root) {
1807 free_extent_buffer(root->node);
1808 free_extent_buffer(root->commit_root);
1809 root->node = NULL;
1810 root->commit_root = NULL;
1811 }
1812}
1813
1814static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
1815{
1816 struct btrfs_root *root, *tmp;
1817
1818 rbtree_postorder_for_each_entry_safe(root, tmp,
1819 &fs_info->global_root_tree,
1820 rb_node)
1821 free_root_extent_buffers(root);
1822}
1823
1824/* helper to cleanup tree roots */
1825static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1826{
1827 free_root_extent_buffers(info->tree_root);
1828
1829 free_global_root_pointers(info);
1830 free_root_extent_buffers(info->dev_root);
1831 free_root_extent_buffers(info->quota_root);
1832 free_root_extent_buffers(info->uuid_root);
1833 free_root_extent_buffers(info->fs_root);
1834 free_root_extent_buffers(info->data_reloc_root);
1835 free_root_extent_buffers(info->block_group_root);
1836 free_root_extent_buffers(info->stripe_root);
1837 if (free_chunk_root)
1838 free_root_extent_buffers(info->chunk_root);
1839}
1840
1841void btrfs_put_root(struct btrfs_root *root)
1842{
1843 if (!root)
1844 return;
1845
1846 if (refcount_dec_and_test(&root->refs)) {
1847 if (WARN_ON(!xa_empty(&root->inodes)))
1848 xa_destroy(&root->inodes);
1849 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
1850 if (root->anon_dev)
1851 free_anon_bdev(root->anon_dev);
1852 free_root_extent_buffers(root);
1853#ifdef CONFIG_BTRFS_DEBUG
1854 spin_lock(&root->fs_info->fs_roots_radix_lock);
1855 list_del_init(&root->leak_list);
1856 spin_unlock(&root->fs_info->fs_roots_radix_lock);
1857#endif
1858 kfree(root);
1859 }
1860}
1861
1862void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
1863{
1864 int ret;
1865 struct btrfs_root *gang[8];
1866 int i;
1867
1868 while (!list_empty(&fs_info->dead_roots)) {
1869 gang[0] = list_entry(fs_info->dead_roots.next,
1870 struct btrfs_root, root_list);
1871 list_del(&gang[0]->root_list);
1872
1873 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
1874 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
1875 btrfs_put_root(gang[0]);
1876 }
1877
1878 while (1) {
1879 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1880 (void **)gang, 0,
1881 ARRAY_SIZE(gang));
1882 if (!ret)
1883 break;
1884 for (i = 0; i < ret; i++)
1885 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
1886 }
1887}
1888
1889static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
1890{
1891 mutex_init(&fs_info->scrub_lock);
1892 atomic_set(&fs_info->scrubs_running, 0);
1893 atomic_set(&fs_info->scrub_pause_req, 0);
1894 atomic_set(&fs_info->scrubs_paused, 0);
1895 atomic_set(&fs_info->scrub_cancel_req, 0);
1896 init_waitqueue_head(&fs_info->scrub_pause_wait);
1897 refcount_set(&fs_info->scrub_workers_refcnt, 0);
1898}
1899
1900static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
1901{
1902 spin_lock_init(&fs_info->balance_lock);
1903 mutex_init(&fs_info->balance_mutex);
1904 atomic_set(&fs_info->balance_pause_req, 0);
1905 atomic_set(&fs_info->balance_cancel_req, 0);
1906 fs_info->balance_ctl = NULL;
1907 init_waitqueue_head(&fs_info->balance_wait_q);
1908 atomic_set(&fs_info->reloc_cancel_req, 0);
1909}
1910
1911static int btrfs_init_btree_inode(struct super_block *sb)
1912{
1913 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1914 unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
1915 fs_info->tree_root);
1916 struct inode *inode;
1917
1918 inode = new_inode(sb);
1919 if (!inode)
1920 return -ENOMEM;
1921
1922 btrfs_set_inode_number(BTRFS_I(inode), BTRFS_BTREE_INODE_OBJECTID);
1923 set_nlink(inode, 1);
1924 /*
1925 * we set the i_size on the btree inode to the max possible int.
1926 * the real end of the address space is determined by all of
1927 * the devices in the system
1928 */
1929 inode->i_size = OFFSET_MAX;
1930 inode->i_mapping->a_ops = &btree_aops;
1931 mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
1932
1933 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
1934 IO_TREE_BTREE_INODE_IO);
1935 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
1936
1937 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
1938 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
1939 __insert_inode_hash(inode, hash);
1940 fs_info->btree_inode = inode;
1941
1942 return 0;
1943}
1944
1945static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
1946{
1947 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
1948 init_rwsem(&fs_info->dev_replace.rwsem);
1949 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
1950}
1951
1952static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
1953{
1954 spin_lock_init(&fs_info->qgroup_lock);
1955 mutex_init(&fs_info->qgroup_ioctl_lock);
1956 fs_info->qgroup_tree = RB_ROOT;
1957 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
1958 fs_info->qgroup_seq = 1;
1959 fs_info->qgroup_ulist = NULL;
1960 fs_info->qgroup_rescan_running = false;
1961 fs_info->qgroup_drop_subtree_thres = BTRFS_QGROUP_DROP_SUBTREE_THRES_DEFAULT;
1962 mutex_init(&fs_info->qgroup_rescan_lock);
1963}
1964
1965static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
1966{
1967 u32 max_active = fs_info->thread_pool_size;
1968 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
1969 unsigned int ordered_flags = WQ_MEM_RECLAIM | WQ_FREEZABLE;
1970
1971 fs_info->workers =
1972 btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
1973
1974 fs_info->delalloc_workers =
1975 btrfs_alloc_workqueue(fs_info, "delalloc",
1976 flags, max_active, 2);
1977
1978 fs_info->flush_workers =
1979 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
1980 flags, max_active, 0);
1981
1982 fs_info->caching_workers =
1983 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
1984
1985 fs_info->fixup_workers =
1986 btrfs_alloc_ordered_workqueue(fs_info, "fixup", ordered_flags);
1987
1988 fs_info->endio_workers =
1989 alloc_workqueue("btrfs-endio", flags, max_active);
1990 fs_info->endio_meta_workers =
1991 alloc_workqueue("btrfs-endio-meta", flags, max_active);
1992 fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
1993 fs_info->endio_write_workers =
1994 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
1995 max_active, 2);
1996 fs_info->compressed_write_workers =
1997 alloc_workqueue("btrfs-compressed-write", flags, max_active);
1998 fs_info->endio_freespace_worker =
1999 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2000 max_active, 0);
2001 fs_info->delayed_workers =
2002 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2003 max_active, 0);
2004 fs_info->qgroup_rescan_workers =
2005 btrfs_alloc_ordered_workqueue(fs_info, "qgroup-rescan",
2006 ordered_flags);
2007 fs_info->discard_ctl.discard_workers =
2008 alloc_ordered_workqueue("btrfs_discard", WQ_FREEZABLE);
2009
2010 if (!(fs_info->workers &&
2011 fs_info->delalloc_workers && fs_info->flush_workers &&
2012 fs_info->endio_workers && fs_info->endio_meta_workers &&
2013 fs_info->compressed_write_workers &&
2014 fs_info->endio_write_workers &&
2015 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2016 fs_info->caching_workers && fs_info->fixup_workers &&
2017 fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2018 fs_info->discard_ctl.discard_workers)) {
2019 return -ENOMEM;
2020 }
2021
2022 return 0;
2023}
2024
2025static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2026{
2027 struct crypto_shash *csum_shash;
2028 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2029
2030 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2031
2032 if (IS_ERR(csum_shash)) {
2033 btrfs_err(fs_info, "error allocating %s hash for checksum",
2034 csum_driver);
2035 return PTR_ERR(csum_shash);
2036 }
2037
2038 fs_info->csum_shash = csum_shash;
2039
2040 /*
2041 * Check if the checksum implementation is a fast accelerated one.
2042 * As-is this is a bit of a hack and should be replaced once the csum
2043 * implementations provide that information themselves.
2044 */
2045 switch (csum_type) {
2046 case BTRFS_CSUM_TYPE_CRC32:
2047 if (!strstr(crypto_shash_driver_name(csum_shash), "generic"))
2048 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2049 break;
2050 case BTRFS_CSUM_TYPE_XXHASH:
2051 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2052 break;
2053 default:
2054 break;
2055 }
2056
2057 btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2058 btrfs_super_csum_name(csum_type),
2059 crypto_shash_driver_name(csum_shash));
2060 return 0;
2061}
2062
2063static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2064 struct btrfs_fs_devices *fs_devices)
2065{
2066 int ret;
2067 struct btrfs_tree_parent_check check = { 0 };
2068 struct btrfs_root *log_tree_root;
2069 struct btrfs_super_block *disk_super = fs_info->super_copy;
2070 u64 bytenr = btrfs_super_log_root(disk_super);
2071 int level = btrfs_super_log_root_level(disk_super);
2072
2073 if (fs_devices->rw_devices == 0) {
2074 btrfs_warn(fs_info, "log replay required on RO media");
2075 return -EIO;
2076 }
2077
2078 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2079 GFP_KERNEL);
2080 if (!log_tree_root)
2081 return -ENOMEM;
2082
2083 check.level = level;
2084 check.transid = fs_info->generation + 1;
2085 check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2086 log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
2087 if (IS_ERR(log_tree_root->node)) {
2088 btrfs_warn(fs_info, "failed to read log tree");
2089 ret = PTR_ERR(log_tree_root->node);
2090 log_tree_root->node = NULL;
2091 btrfs_put_root(log_tree_root);
2092 return ret;
2093 }
2094 if (!extent_buffer_uptodate(log_tree_root->node)) {
2095 btrfs_err(fs_info, "failed to read log tree");
2096 btrfs_put_root(log_tree_root);
2097 return -EIO;
2098 }
2099
2100 /* returns with log_tree_root freed on success */
2101 ret = btrfs_recover_log_trees(log_tree_root);
2102 if (ret) {
2103 btrfs_handle_fs_error(fs_info, ret,
2104 "Failed to recover log tree");
2105 btrfs_put_root(log_tree_root);
2106 return ret;
2107 }
2108
2109 if (sb_rdonly(fs_info->sb)) {
2110 ret = btrfs_commit_super(fs_info);
2111 if (ret)
2112 return ret;
2113 }
2114
2115 return 0;
2116}
2117
2118static int load_global_roots_objectid(struct btrfs_root *tree_root,
2119 struct btrfs_path *path, u64 objectid,
2120 const char *name)
2121{
2122 struct btrfs_fs_info *fs_info = tree_root->fs_info;
2123 struct btrfs_root *root;
2124 u64 max_global_id = 0;
2125 int ret;
2126 struct btrfs_key key = {
2127 .objectid = objectid,
2128 .type = BTRFS_ROOT_ITEM_KEY,
2129 .offset = 0,
2130 };
2131 bool found = false;
2132
2133 /* If we have IGNOREDATACSUMS skip loading these roots. */
2134 if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2135 btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2136 set_bit(BTRFS_FS_STATE_NO_DATA_CSUMS, &fs_info->fs_state);
2137 return 0;
2138 }
2139
2140 while (1) {
2141 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2142 if (ret < 0)
2143 break;
2144
2145 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2146 ret = btrfs_next_leaf(tree_root, path);
2147 if (ret) {
2148 if (ret > 0)
2149 ret = 0;
2150 break;
2151 }
2152 }
2153 ret = 0;
2154
2155 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2156 if (key.objectid != objectid)
2157 break;
2158 btrfs_release_path(path);
2159
2160 /*
2161 * Just worry about this for extent tree, it'll be the same for
2162 * everybody.
2163 */
2164 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2165 max_global_id = max(max_global_id, key.offset);
2166
2167 found = true;
2168 root = read_tree_root_path(tree_root, path, &key);
2169 if (IS_ERR(root)) {
2170 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2171 ret = PTR_ERR(root);
2172 break;
2173 }
2174 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2175 ret = btrfs_global_root_insert(root);
2176 if (ret) {
2177 btrfs_put_root(root);
2178 break;
2179 }
2180 key.offset++;
2181 }
2182 btrfs_release_path(path);
2183
2184 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2185 fs_info->nr_global_roots = max_global_id + 1;
2186
2187 if (!found || ret) {
2188 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2189 set_bit(BTRFS_FS_STATE_NO_DATA_CSUMS, &fs_info->fs_state);
2190
2191 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2192 ret = ret ? ret : -ENOENT;
2193 else
2194 ret = 0;
2195 btrfs_err(fs_info, "failed to load root %s", name);
2196 }
2197 return ret;
2198}
2199
2200static int load_global_roots(struct btrfs_root *tree_root)
2201{
2202 struct btrfs_path *path;
2203 int ret = 0;
2204
2205 path = btrfs_alloc_path();
2206 if (!path)
2207 return -ENOMEM;
2208
2209 ret = load_global_roots_objectid(tree_root, path,
2210 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2211 if (ret)
2212 goto out;
2213 ret = load_global_roots_objectid(tree_root, path,
2214 BTRFS_CSUM_TREE_OBJECTID, "csum");
2215 if (ret)
2216 goto out;
2217 if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2218 goto out;
2219 ret = load_global_roots_objectid(tree_root, path,
2220 BTRFS_FREE_SPACE_TREE_OBJECTID,
2221 "free space");
2222out:
2223 btrfs_free_path(path);
2224 return ret;
2225}
2226
2227static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2228{
2229 struct btrfs_root *tree_root = fs_info->tree_root;
2230 struct btrfs_root *root;
2231 struct btrfs_key location;
2232 int ret;
2233
2234 ASSERT(fs_info->tree_root);
2235
2236 ret = load_global_roots(tree_root);
2237 if (ret)
2238 return ret;
2239
2240 location.type = BTRFS_ROOT_ITEM_KEY;
2241 location.offset = 0;
2242
2243 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2244 location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2245 root = btrfs_read_tree_root(tree_root, &location);
2246 if (IS_ERR(root)) {
2247 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2248 ret = PTR_ERR(root);
2249 goto out;
2250 }
2251 } else {
2252 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2253 fs_info->block_group_root = root;
2254 }
2255 }
2256
2257 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2258 root = btrfs_read_tree_root(tree_root, &location);
2259 if (IS_ERR(root)) {
2260 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2261 ret = PTR_ERR(root);
2262 goto out;
2263 }
2264 } else {
2265 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2266 fs_info->dev_root = root;
2267 }
2268 /* Initialize fs_info for all devices in any case */
2269 ret = btrfs_init_devices_late(fs_info);
2270 if (ret)
2271 goto out;
2272
2273 /*
2274 * This tree can share blocks with some other fs tree during relocation
2275 * and we need a proper setup by btrfs_get_fs_root
2276 */
2277 root = btrfs_get_fs_root(tree_root->fs_info,
2278 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2279 if (IS_ERR(root)) {
2280 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2281 ret = PTR_ERR(root);
2282 goto out;
2283 }
2284 } else {
2285 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2286 fs_info->data_reloc_root = root;
2287 }
2288
2289 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2290 root = btrfs_read_tree_root(tree_root, &location);
2291 if (!IS_ERR(root)) {
2292 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2293 fs_info->quota_root = root;
2294 }
2295
2296 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2297 root = btrfs_read_tree_root(tree_root, &location);
2298 if (IS_ERR(root)) {
2299 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2300 ret = PTR_ERR(root);
2301 if (ret != -ENOENT)
2302 goto out;
2303 }
2304 } else {
2305 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2306 fs_info->uuid_root = root;
2307 }
2308
2309 if (btrfs_fs_incompat(fs_info, RAID_STRIPE_TREE)) {
2310 location.objectid = BTRFS_RAID_STRIPE_TREE_OBJECTID;
2311 root = btrfs_read_tree_root(tree_root, &location);
2312 if (IS_ERR(root)) {
2313 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2314 ret = PTR_ERR(root);
2315 goto out;
2316 }
2317 } else {
2318 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2319 fs_info->stripe_root = root;
2320 }
2321 }
2322
2323 return 0;
2324out:
2325 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2326 location.objectid, ret);
2327 return ret;
2328}
2329
2330/*
2331 * Real super block validation
2332 * NOTE: super csum type and incompat features will not be checked here.
2333 *
2334 * @sb: super block to check
2335 * @mirror_num: the super block number to check its bytenr:
2336 * 0 the primary (1st) sb
2337 * 1, 2 2nd and 3rd backup copy
2338 * -1 skip bytenr check
2339 */
2340int btrfs_validate_super(const struct btrfs_fs_info *fs_info,
2341 const struct btrfs_super_block *sb, int mirror_num)
2342{
2343 u64 nodesize = btrfs_super_nodesize(sb);
2344 u64 sectorsize = btrfs_super_sectorsize(sb);
2345 int ret = 0;
2346 const bool ignore_flags = btrfs_test_opt(fs_info, IGNORESUPERFLAGS);
2347
2348 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2349 btrfs_err(fs_info, "no valid FS found");
2350 ret = -EINVAL;
2351 }
2352 if ((btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)) {
2353 if (!ignore_flags) {
2354 btrfs_err(fs_info,
2355 "unrecognized or unsupported super flag 0x%llx",
2356 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2357 ret = -EINVAL;
2358 } else {
2359 btrfs_info(fs_info,
2360 "unrecognized or unsupported super flags: 0x%llx, ignored",
2361 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2362 }
2363 }
2364 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2365 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2366 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2367 ret = -EINVAL;
2368 }
2369 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2370 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2371 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2372 ret = -EINVAL;
2373 }
2374 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2375 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2376 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2377 ret = -EINVAL;
2378 }
2379
2380 /*
2381 * Check sectorsize and nodesize first, other check will need it.
2382 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2383 */
2384 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2385 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2386 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2387 ret = -EINVAL;
2388 }
2389
2390 /*
2391 * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2392 *
2393 * We can support 16K sectorsize with 64K page size without problem,
2394 * but such sectorsize/pagesize combination doesn't make much sense.
2395 * 4K will be our future standard, PAGE_SIZE is supported from the very
2396 * beginning.
2397 */
2398 if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2399 btrfs_err(fs_info,
2400 "sectorsize %llu not yet supported for page size %lu",
2401 sectorsize, PAGE_SIZE);
2402 ret = -EINVAL;
2403 }
2404
2405 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2406 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2407 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2408 ret = -EINVAL;
2409 }
2410 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2411 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2412 le32_to_cpu(sb->__unused_leafsize), nodesize);
2413 ret = -EINVAL;
2414 }
2415
2416 /* Root alignment check */
2417 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2418 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2419 btrfs_super_root(sb));
2420 ret = -EINVAL;
2421 }
2422 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2423 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2424 btrfs_super_chunk_root(sb));
2425 ret = -EINVAL;
2426 }
2427 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2428 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2429 btrfs_super_log_root(sb));
2430 ret = -EINVAL;
2431 }
2432
2433 if (!fs_info->fs_devices->temp_fsid &&
2434 memcmp(fs_info->fs_devices->fsid, sb->fsid, BTRFS_FSID_SIZE) != 0) {
2435 btrfs_err(fs_info,
2436 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2437 sb->fsid, fs_info->fs_devices->fsid);
2438 ret = -EINVAL;
2439 }
2440
2441 if (memcmp(fs_info->fs_devices->metadata_uuid, btrfs_sb_fsid_ptr(sb),
2442 BTRFS_FSID_SIZE) != 0) {
2443 btrfs_err(fs_info,
2444"superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2445 btrfs_sb_fsid_ptr(sb), fs_info->fs_devices->metadata_uuid);
2446 ret = -EINVAL;
2447 }
2448
2449 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2450 BTRFS_FSID_SIZE) != 0) {
2451 btrfs_err(fs_info,
2452 "dev_item UUID does not match metadata fsid: %pU != %pU",
2453 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2454 ret = -EINVAL;
2455 }
2456
2457 /*
2458 * Artificial requirement for block-group-tree to force newer features
2459 * (free-space-tree, no-holes) so the test matrix is smaller.
2460 */
2461 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2462 (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2463 !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2464 btrfs_err(fs_info,
2465 "block-group-tree feature requires free-space-tree and no-holes");
2466 ret = -EINVAL;
2467 }
2468
2469 /*
2470 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2471 * done later
2472 */
2473 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2474 btrfs_err(fs_info, "bytes_used is too small %llu",
2475 btrfs_super_bytes_used(sb));
2476 ret = -EINVAL;
2477 }
2478 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2479 btrfs_err(fs_info, "invalid stripesize %u",
2480 btrfs_super_stripesize(sb));
2481 ret = -EINVAL;
2482 }
2483 if (btrfs_super_num_devices(sb) > (1UL << 31))
2484 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2485 btrfs_super_num_devices(sb));
2486 if (btrfs_super_num_devices(sb) == 0) {
2487 btrfs_err(fs_info, "number of devices is 0");
2488 ret = -EINVAL;
2489 }
2490
2491 if (mirror_num >= 0 &&
2492 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2493 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2494 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2495 ret = -EINVAL;
2496 }
2497
2498 /*
2499 * Obvious sys_chunk_array corruptions, it must hold at least one key
2500 * and one chunk
2501 */
2502 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2503 btrfs_err(fs_info, "system chunk array too big %u > %u",
2504 btrfs_super_sys_array_size(sb),
2505 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2506 ret = -EINVAL;
2507 }
2508 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2509 + sizeof(struct btrfs_chunk)) {
2510 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2511 btrfs_super_sys_array_size(sb),
2512 sizeof(struct btrfs_disk_key)
2513 + sizeof(struct btrfs_chunk));
2514 ret = -EINVAL;
2515 }
2516
2517 /*
2518 * The generation is a global counter, we'll trust it more than the others
2519 * but it's still possible that it's the one that's wrong.
2520 */
2521 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2522 btrfs_warn(fs_info,
2523 "suspicious: generation < chunk_root_generation: %llu < %llu",
2524 btrfs_super_generation(sb),
2525 btrfs_super_chunk_root_generation(sb));
2526 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2527 && btrfs_super_cache_generation(sb) != (u64)-1)
2528 btrfs_warn(fs_info,
2529 "suspicious: generation < cache_generation: %llu < %llu",
2530 btrfs_super_generation(sb),
2531 btrfs_super_cache_generation(sb));
2532
2533 return ret;
2534}
2535
2536/*
2537 * Validation of super block at mount time.
2538 * Some checks already done early at mount time, like csum type and incompat
2539 * flags will be skipped.
2540 */
2541static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2542{
2543 return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2544}
2545
2546/*
2547 * Validation of super block at write time.
2548 * Some checks like bytenr check will be skipped as their values will be
2549 * overwritten soon.
2550 * Extra checks like csum type and incompat flags will be done here.
2551 */
2552static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2553 struct btrfs_super_block *sb)
2554{
2555 int ret;
2556
2557 ret = btrfs_validate_super(fs_info, sb, -1);
2558 if (ret < 0)
2559 goto out;
2560 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2561 ret = -EUCLEAN;
2562 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2563 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2564 goto out;
2565 }
2566 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2567 ret = -EUCLEAN;
2568 btrfs_err(fs_info,
2569 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2570 btrfs_super_incompat_flags(sb),
2571 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2572 goto out;
2573 }
2574out:
2575 if (ret < 0)
2576 btrfs_err(fs_info,
2577 "super block corruption detected before writing it to disk");
2578 return ret;
2579}
2580
2581static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2582{
2583 struct btrfs_tree_parent_check check = {
2584 .level = level,
2585 .transid = gen,
2586 .owner_root = btrfs_root_id(root)
2587 };
2588 int ret = 0;
2589
2590 root->node = read_tree_block(root->fs_info, bytenr, &check);
2591 if (IS_ERR(root->node)) {
2592 ret = PTR_ERR(root->node);
2593 root->node = NULL;
2594 return ret;
2595 }
2596 if (!extent_buffer_uptodate(root->node)) {
2597 free_extent_buffer(root->node);
2598 root->node = NULL;
2599 return -EIO;
2600 }
2601
2602 btrfs_set_root_node(&root->root_item, root->node);
2603 root->commit_root = btrfs_root_node(root);
2604 btrfs_set_root_refs(&root->root_item, 1);
2605 return ret;
2606}
2607
2608static int load_important_roots(struct btrfs_fs_info *fs_info)
2609{
2610 struct btrfs_super_block *sb = fs_info->super_copy;
2611 u64 gen, bytenr;
2612 int level, ret;
2613
2614 bytenr = btrfs_super_root(sb);
2615 gen = btrfs_super_generation(sb);
2616 level = btrfs_super_root_level(sb);
2617 ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2618 if (ret) {
2619 btrfs_warn(fs_info, "couldn't read tree root");
2620 return ret;
2621 }
2622 return 0;
2623}
2624
2625static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2626{
2627 int backup_index = find_newest_super_backup(fs_info);
2628 struct btrfs_super_block *sb = fs_info->super_copy;
2629 struct btrfs_root *tree_root = fs_info->tree_root;
2630 bool handle_error = false;
2631 int ret = 0;
2632 int i;
2633
2634 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2635 if (handle_error) {
2636 if (!IS_ERR(tree_root->node))
2637 free_extent_buffer(tree_root->node);
2638 tree_root->node = NULL;
2639
2640 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2641 break;
2642
2643 free_root_pointers(fs_info, 0);
2644
2645 /*
2646 * Don't use the log in recovery mode, it won't be
2647 * valid
2648 */
2649 btrfs_set_super_log_root(sb, 0);
2650
2651 btrfs_warn(fs_info, "try to load backup roots slot %d", i);
2652 ret = read_backup_root(fs_info, i);
2653 backup_index = ret;
2654 if (ret < 0)
2655 return ret;
2656 }
2657
2658 ret = load_important_roots(fs_info);
2659 if (ret) {
2660 handle_error = true;
2661 continue;
2662 }
2663
2664 /*
2665 * No need to hold btrfs_root::objectid_mutex since the fs
2666 * hasn't been fully initialised and we are the only user
2667 */
2668 ret = btrfs_init_root_free_objectid(tree_root);
2669 if (ret < 0) {
2670 handle_error = true;
2671 continue;
2672 }
2673
2674 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2675
2676 ret = btrfs_read_roots(fs_info);
2677 if (ret < 0) {
2678 handle_error = true;
2679 continue;
2680 }
2681
2682 /* All successful */
2683 fs_info->generation = btrfs_header_generation(tree_root->node);
2684 btrfs_set_last_trans_committed(fs_info, fs_info->generation);
2685 fs_info->last_reloc_trans = 0;
2686
2687 /* Always begin writing backup roots after the one being used */
2688 if (backup_index < 0) {
2689 fs_info->backup_root_index = 0;
2690 } else {
2691 fs_info->backup_root_index = backup_index + 1;
2692 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2693 }
2694 break;
2695 }
2696
2697 return ret;
2698}
2699
2700void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2701{
2702 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2703 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2704 INIT_LIST_HEAD(&fs_info->trans_list);
2705 INIT_LIST_HEAD(&fs_info->dead_roots);
2706 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2707 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2708 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2709 spin_lock_init(&fs_info->delalloc_root_lock);
2710 spin_lock_init(&fs_info->trans_lock);
2711 spin_lock_init(&fs_info->fs_roots_radix_lock);
2712 spin_lock_init(&fs_info->delayed_iput_lock);
2713 spin_lock_init(&fs_info->defrag_inodes_lock);
2714 spin_lock_init(&fs_info->super_lock);
2715 spin_lock_init(&fs_info->buffer_lock);
2716 spin_lock_init(&fs_info->unused_bgs_lock);
2717 spin_lock_init(&fs_info->treelog_bg_lock);
2718 spin_lock_init(&fs_info->zone_active_bgs_lock);
2719 spin_lock_init(&fs_info->relocation_bg_lock);
2720 rwlock_init(&fs_info->tree_mod_log_lock);
2721 rwlock_init(&fs_info->global_root_lock);
2722 mutex_init(&fs_info->unused_bg_unpin_mutex);
2723 mutex_init(&fs_info->reclaim_bgs_lock);
2724 mutex_init(&fs_info->reloc_mutex);
2725 mutex_init(&fs_info->delalloc_root_mutex);
2726 mutex_init(&fs_info->zoned_meta_io_lock);
2727 mutex_init(&fs_info->zoned_data_reloc_io_lock);
2728 seqlock_init(&fs_info->profiles_lock);
2729
2730 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
2731 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
2732 btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
2733 btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
2734 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_prep,
2735 BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2736 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
2737 BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2738 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
2739 BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2740 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
2741 BTRFS_LOCKDEP_TRANS_COMPLETED);
2742
2743 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2744 INIT_LIST_HEAD(&fs_info->space_info);
2745 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2746 INIT_LIST_HEAD(&fs_info->unused_bgs);
2747 INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2748 INIT_LIST_HEAD(&fs_info->zone_active_bgs);
2749#ifdef CONFIG_BTRFS_DEBUG
2750 INIT_LIST_HEAD(&fs_info->allocated_roots);
2751 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2752 spin_lock_init(&fs_info->eb_leak_lock);
2753#endif
2754 fs_info->mapping_tree = RB_ROOT_CACHED;
2755 rwlock_init(&fs_info->mapping_tree_lock);
2756 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2757 BTRFS_BLOCK_RSV_GLOBAL);
2758 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2759 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2760 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2761 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2762 BTRFS_BLOCK_RSV_DELOPS);
2763 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2764 BTRFS_BLOCK_RSV_DELREFS);
2765
2766 atomic_set(&fs_info->async_delalloc_pages, 0);
2767 atomic_set(&fs_info->defrag_running, 0);
2768 atomic_set(&fs_info->nr_delayed_iputs, 0);
2769 atomic64_set(&fs_info->tree_mod_seq, 0);
2770 fs_info->global_root_tree = RB_ROOT;
2771 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2772 fs_info->metadata_ratio = 0;
2773 fs_info->defrag_inodes = RB_ROOT;
2774 atomic64_set(&fs_info->free_chunk_space, 0);
2775 fs_info->tree_mod_log = RB_ROOT;
2776 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2777 btrfs_init_ref_verify(fs_info);
2778
2779 fs_info->thread_pool_size = min_t(unsigned long,
2780 num_online_cpus() + 2, 8);
2781
2782 INIT_LIST_HEAD(&fs_info->ordered_roots);
2783 spin_lock_init(&fs_info->ordered_root_lock);
2784
2785 btrfs_init_scrub(fs_info);
2786 btrfs_init_balance(fs_info);
2787 btrfs_init_async_reclaim_work(fs_info);
2788 btrfs_init_extent_map_shrinker_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->evictable_extent_maps, 0, GFP_KERNEL);
2852 if (ret)
2853 return ret;
2854
2855 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2856 if (ret)
2857 return ret;
2858
2859 fs_info->dirty_metadata_batch = PAGE_SIZE *
2860 (1 + ilog2(nr_cpu_ids));
2861
2862 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2863 if (ret)
2864 return ret;
2865
2866 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2867 GFP_KERNEL);
2868 if (ret)
2869 return ret;
2870
2871 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2872 GFP_KERNEL);
2873 if (!fs_info->delayed_root)
2874 return -ENOMEM;
2875 btrfs_init_delayed_root(fs_info->delayed_root);
2876
2877 if (sb_rdonly(sb))
2878 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2879 if (btrfs_test_opt(fs_info, IGNOREMETACSUMS))
2880 set_bit(BTRFS_FS_STATE_SKIP_META_CSUMS, &fs_info->fs_state);
2881
2882 return btrfs_alloc_stripe_hash_table(fs_info);
2883}
2884
2885static int btrfs_uuid_rescan_kthread(void *data)
2886{
2887 struct btrfs_fs_info *fs_info = data;
2888 int ret;
2889
2890 /*
2891 * 1st step is to iterate through the existing UUID tree and
2892 * to delete all entries that contain outdated data.
2893 * 2nd step is to add all missing entries to the UUID tree.
2894 */
2895 ret = btrfs_uuid_tree_iterate(fs_info);
2896 if (ret < 0) {
2897 if (ret != -EINTR)
2898 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2899 ret);
2900 up(&fs_info->uuid_tree_rescan_sem);
2901 return ret;
2902 }
2903 return btrfs_uuid_scan_kthread(data);
2904}
2905
2906static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2907{
2908 struct task_struct *task;
2909
2910 down(&fs_info->uuid_tree_rescan_sem);
2911 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2912 if (IS_ERR(task)) {
2913 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2914 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2915 up(&fs_info->uuid_tree_rescan_sem);
2916 return PTR_ERR(task);
2917 }
2918
2919 return 0;
2920}
2921
2922static int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2923{
2924 u64 root_objectid = 0;
2925 struct btrfs_root *gang[8];
2926 int ret = 0;
2927
2928 while (1) {
2929 unsigned int found;
2930
2931 spin_lock(&fs_info->fs_roots_radix_lock);
2932 found = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2933 (void **)gang, root_objectid,
2934 ARRAY_SIZE(gang));
2935 if (!found) {
2936 spin_unlock(&fs_info->fs_roots_radix_lock);
2937 break;
2938 }
2939 root_objectid = btrfs_root_id(gang[found - 1]) + 1;
2940
2941 for (int i = 0; i < found; i++) {
2942 /* Avoid to grab roots in dead_roots. */
2943 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
2944 gang[i] = NULL;
2945 continue;
2946 }
2947 /* Grab all the search result for later use. */
2948 gang[i] = btrfs_grab_root(gang[i]);
2949 }
2950 spin_unlock(&fs_info->fs_roots_radix_lock);
2951
2952 for (int i = 0; i < found; i++) {
2953 if (!gang[i])
2954 continue;
2955 root_objectid = btrfs_root_id(gang[i]);
2956 /*
2957 * Continue to release the remaining roots after the first
2958 * error without cleanup and preserve the first error
2959 * for the return.
2960 */
2961 if (!ret)
2962 ret = btrfs_orphan_cleanup(gang[i]);
2963 btrfs_put_root(gang[i]);
2964 }
2965 if (ret)
2966 break;
2967
2968 root_objectid++;
2969 }
2970 return ret;
2971}
2972
2973/*
2974 * Mounting logic specific to read-write file systems. Shared by open_ctree
2975 * and btrfs_remount when remounting from read-only to read-write.
2976 */
2977int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
2978{
2979 int ret;
2980 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
2981 bool rebuild_free_space_tree = false;
2982
2983 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2984 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2985 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
2986 btrfs_warn(fs_info,
2987 "'clear_cache' option is ignored with extent tree v2");
2988 else
2989 rebuild_free_space_tree = true;
2990 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2991 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2992 btrfs_warn(fs_info, "free space tree is invalid");
2993 rebuild_free_space_tree = true;
2994 }
2995
2996 if (rebuild_free_space_tree) {
2997 btrfs_info(fs_info, "rebuilding free space tree");
2998 ret = btrfs_rebuild_free_space_tree(fs_info);
2999 if (ret) {
3000 btrfs_warn(fs_info,
3001 "failed to rebuild free space tree: %d", ret);
3002 goto out;
3003 }
3004 }
3005
3006 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3007 !btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
3008 btrfs_info(fs_info, "disabling free space tree");
3009 ret = btrfs_delete_free_space_tree(fs_info);
3010 if (ret) {
3011 btrfs_warn(fs_info,
3012 "failed to disable free space tree: %d", ret);
3013 goto out;
3014 }
3015 }
3016
3017 /*
3018 * btrfs_find_orphan_roots() is responsible for finding all the dead
3019 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3020 * them into the fs_info->fs_roots_radix tree. This must be done before
3021 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3022 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3023 * item before the root's tree is deleted - this means that if we unmount
3024 * or crash before the deletion completes, on the next mount we will not
3025 * delete what remains of the tree because the orphan item does not
3026 * exists anymore, which is what tells us we have a pending deletion.
3027 */
3028 ret = btrfs_find_orphan_roots(fs_info);
3029 if (ret)
3030 goto out;
3031
3032 ret = btrfs_cleanup_fs_roots(fs_info);
3033 if (ret)
3034 goto out;
3035
3036 down_read(&fs_info->cleanup_work_sem);
3037 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3038 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3039 up_read(&fs_info->cleanup_work_sem);
3040 goto out;
3041 }
3042 up_read(&fs_info->cleanup_work_sem);
3043
3044 mutex_lock(&fs_info->cleaner_mutex);
3045 ret = btrfs_recover_relocation(fs_info);
3046 mutex_unlock(&fs_info->cleaner_mutex);
3047 if (ret < 0) {
3048 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3049 goto out;
3050 }
3051
3052 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3053 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3054 btrfs_info(fs_info, "creating free space tree");
3055 ret = btrfs_create_free_space_tree(fs_info);
3056 if (ret) {
3057 btrfs_warn(fs_info,
3058 "failed to create free space tree: %d", ret);
3059 goto out;
3060 }
3061 }
3062
3063 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3064 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3065 if (ret)
3066 goto out;
3067 }
3068
3069 ret = btrfs_resume_balance_async(fs_info);
3070 if (ret)
3071 goto out;
3072
3073 ret = btrfs_resume_dev_replace_async(fs_info);
3074 if (ret) {
3075 btrfs_warn(fs_info, "failed to resume dev_replace");
3076 goto out;
3077 }
3078
3079 btrfs_qgroup_rescan_resume(fs_info);
3080
3081 if (!fs_info->uuid_root) {
3082 btrfs_info(fs_info, "creating UUID tree");
3083 ret = btrfs_create_uuid_tree(fs_info);
3084 if (ret) {
3085 btrfs_warn(fs_info,
3086 "failed to create the UUID tree %d", ret);
3087 goto out;
3088 }
3089 }
3090
3091out:
3092 return ret;
3093}
3094
3095/*
3096 * Do various sanity and dependency checks of different features.
3097 *
3098 * @is_rw_mount: If the mount is read-write.
3099 *
3100 * This is the place for less strict checks (like for subpage or artificial
3101 * feature dependencies).
3102 *
3103 * For strict checks or possible corruption detection, see
3104 * btrfs_validate_super().
3105 *
3106 * This should be called after btrfs_parse_options(), as some mount options
3107 * (space cache related) can modify on-disk format like free space tree and
3108 * screw up certain feature dependencies.
3109 */
3110int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3111{
3112 struct btrfs_super_block *disk_super = fs_info->super_copy;
3113 u64 incompat = btrfs_super_incompat_flags(disk_super);
3114 const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3115 const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3116
3117 if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3118 btrfs_err(fs_info,
3119 "cannot mount because of unknown incompat features (0x%llx)",
3120 incompat);
3121 return -EINVAL;
3122 }
3123
3124 /* Runtime limitation for mixed block groups. */
3125 if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3126 (fs_info->sectorsize != fs_info->nodesize)) {
3127 btrfs_err(fs_info,
3128"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3129 fs_info->nodesize, fs_info->sectorsize);
3130 return -EINVAL;
3131 }
3132
3133 /* Mixed backref is an always-enabled feature. */
3134 incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3135
3136 /* Set compression related flags just in case. */
3137 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3138 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3139 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3140 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3141
3142 /*
3143 * An ancient flag, which should really be marked deprecated.
3144 * Such runtime limitation doesn't really need a incompat flag.
3145 */
3146 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3147 incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3148
3149 if (compat_ro_unsupp && is_rw_mount) {
3150 btrfs_err(fs_info,
3151 "cannot mount read-write because of unknown compat_ro features (0x%llx)",
3152 compat_ro);
3153 return -EINVAL;
3154 }
3155
3156 /*
3157 * We have unsupported RO compat features, although RO mounted, we
3158 * should not cause any metadata writes, including log replay.
3159 * Or we could screw up whatever the new feature requires.
3160 */
3161 if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3162 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3163 btrfs_err(fs_info,
3164"cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3165 compat_ro);
3166 return -EINVAL;
3167 }
3168
3169 /*
3170 * Artificial limitations for block group tree, to force
3171 * block-group-tree to rely on no-holes and free-space-tree.
3172 */
3173 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3174 (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3175 !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3176 btrfs_err(fs_info,
3177"block-group-tree feature requires no-holes and free-space-tree features");
3178 return -EINVAL;
3179 }
3180
3181 /*
3182 * Subpage runtime limitation on v1 cache.
3183 *
3184 * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3185 * we're already defaulting to v2 cache, no need to bother v1 as it's
3186 * going to be deprecated anyway.
3187 */
3188 if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3189 btrfs_warn(fs_info,
3190 "v1 space cache is not supported for page size %lu with sectorsize %u",
3191 PAGE_SIZE, fs_info->sectorsize);
3192 return -EINVAL;
3193 }
3194
3195 /* This can be called by remount, we need to protect the super block. */
3196 spin_lock(&fs_info->super_lock);
3197 btrfs_set_super_incompat_flags(disk_super, incompat);
3198 spin_unlock(&fs_info->super_lock);
3199
3200 return 0;
3201}
3202
3203int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices)
3204{
3205 u32 sectorsize;
3206 u32 nodesize;
3207 u32 stripesize;
3208 u64 generation;
3209 u16 csum_type;
3210 struct btrfs_super_block *disk_super;
3211 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3212 struct btrfs_root *tree_root;
3213 struct btrfs_root *chunk_root;
3214 int ret;
3215 int level;
3216
3217 ret = init_mount_fs_info(fs_info, sb);
3218 if (ret)
3219 goto fail;
3220
3221 /* These need to be init'ed before we start creating inodes and such. */
3222 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3223 GFP_KERNEL);
3224 fs_info->tree_root = tree_root;
3225 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3226 GFP_KERNEL);
3227 fs_info->chunk_root = chunk_root;
3228 if (!tree_root || !chunk_root) {
3229 ret = -ENOMEM;
3230 goto fail;
3231 }
3232
3233 ret = btrfs_init_btree_inode(sb);
3234 if (ret)
3235 goto fail;
3236
3237 invalidate_bdev(fs_devices->latest_dev->bdev);
3238
3239 /*
3240 * Read super block and check the signature bytes only
3241 */
3242 disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3243 if (IS_ERR(disk_super)) {
3244 ret = PTR_ERR(disk_super);
3245 goto fail_alloc;
3246 }
3247
3248 btrfs_info(fs_info, "first mount of filesystem %pU", disk_super->fsid);
3249 /*
3250 * Verify the type first, if that or the checksum value are
3251 * corrupted, we'll find out
3252 */
3253 csum_type = btrfs_super_csum_type(disk_super);
3254 if (!btrfs_supported_super_csum(csum_type)) {
3255 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3256 csum_type);
3257 ret = -EINVAL;
3258 btrfs_release_disk_super(disk_super);
3259 goto fail_alloc;
3260 }
3261
3262 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3263
3264 ret = btrfs_init_csum_hash(fs_info, csum_type);
3265 if (ret) {
3266 btrfs_release_disk_super(disk_super);
3267 goto fail_alloc;
3268 }
3269
3270 /*
3271 * We want to check superblock checksum, the type is stored inside.
3272 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3273 */
3274 if (btrfs_check_super_csum(fs_info, disk_super)) {
3275 btrfs_err(fs_info, "superblock checksum mismatch");
3276 ret = -EINVAL;
3277 btrfs_release_disk_super(disk_super);
3278 goto fail_alloc;
3279 }
3280
3281 /*
3282 * super_copy is zeroed at allocation time and we never touch the
3283 * following bytes up to INFO_SIZE, the checksum is calculated from
3284 * the whole block of INFO_SIZE
3285 */
3286 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3287 btrfs_release_disk_super(disk_super);
3288
3289 disk_super = fs_info->super_copy;
3290
3291 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3292 sizeof(*fs_info->super_for_commit));
3293
3294 ret = btrfs_validate_mount_super(fs_info);
3295 if (ret) {
3296 btrfs_err(fs_info, "superblock contains fatal errors");
3297 ret = -EINVAL;
3298 goto fail_alloc;
3299 }
3300
3301 if (!btrfs_super_root(disk_super)) {
3302 btrfs_err(fs_info, "invalid superblock tree root bytenr");
3303 ret = -EINVAL;
3304 goto fail_alloc;
3305 }
3306
3307 /* check FS state, whether FS is broken. */
3308 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3309 WRITE_ONCE(fs_info->fs_error, -EUCLEAN);
3310
3311 /* Set up fs_info before parsing mount options */
3312 nodesize = btrfs_super_nodesize(disk_super);
3313 sectorsize = btrfs_super_sectorsize(disk_super);
3314 stripesize = sectorsize;
3315 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3316 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3317
3318 fs_info->nodesize = nodesize;
3319 fs_info->sectorsize = sectorsize;
3320 fs_info->sectorsize_bits = ilog2(sectorsize);
3321 fs_info->sectors_per_page = (PAGE_SIZE >> fs_info->sectorsize_bits);
3322 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3323 fs_info->stripesize = stripesize;
3324
3325 /*
3326 * Handle the space caching options appropriately now that we have the
3327 * super block loaded and validated.
3328 */
3329 btrfs_set_free_space_cache_settings(fs_info);
3330
3331 if (!btrfs_check_options(fs_info, &fs_info->mount_opt, sb->s_flags)) {
3332 ret = -EINVAL;
3333 goto fail_alloc;
3334 }
3335
3336 ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3337 if (ret < 0)
3338 goto fail_alloc;
3339
3340 /*
3341 * At this point our mount options are validated, if we set ->max_inline
3342 * to something non-standard make sure we truncate it to sectorsize.
3343 */
3344 fs_info->max_inline = min_t(u64, fs_info->max_inline, fs_info->sectorsize);
3345
3346 if (sectorsize < PAGE_SIZE)
3347 btrfs_warn(fs_info,
3348 "read-write for sector size %u with page size %lu is experimental",
3349 sectorsize, PAGE_SIZE);
3350
3351 ret = btrfs_init_workqueues(fs_info);
3352 if (ret)
3353 goto fail_sb_buffer;
3354
3355 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3356 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3357
3358 /* Update the values for the current filesystem. */
3359 sb->s_blocksize = sectorsize;
3360 sb->s_blocksize_bits = blksize_bits(sectorsize);
3361 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3362
3363 mutex_lock(&fs_info->chunk_mutex);
3364 ret = btrfs_read_sys_array(fs_info);
3365 mutex_unlock(&fs_info->chunk_mutex);
3366 if (ret) {
3367 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3368 goto fail_sb_buffer;
3369 }
3370
3371 generation = btrfs_super_chunk_root_generation(disk_super);
3372 level = btrfs_super_chunk_root_level(disk_super);
3373 ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3374 generation, level);
3375 if (ret) {
3376 btrfs_err(fs_info, "failed to read chunk root");
3377 goto fail_tree_roots;
3378 }
3379
3380 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3381 offsetof(struct btrfs_header, chunk_tree_uuid),
3382 BTRFS_UUID_SIZE);
3383
3384 ret = btrfs_read_chunk_tree(fs_info);
3385 if (ret) {
3386 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3387 goto fail_tree_roots;
3388 }
3389
3390 /*
3391 * At this point we know all the devices that make this filesystem,
3392 * including the seed devices but we don't know yet if the replace
3393 * target is required. So free devices that are not part of this
3394 * filesystem but skip the replace target device which is checked
3395 * below in btrfs_init_dev_replace().
3396 */
3397 btrfs_free_extra_devids(fs_devices);
3398 if (!fs_devices->latest_dev->bdev) {
3399 btrfs_err(fs_info, "failed to read devices");
3400 ret = -EIO;
3401 goto fail_tree_roots;
3402 }
3403
3404 ret = init_tree_roots(fs_info);
3405 if (ret)
3406 goto fail_tree_roots;
3407
3408 /*
3409 * Get zone type information of zoned block devices. This will also
3410 * handle emulation of a zoned filesystem if a regular device has the
3411 * zoned incompat feature flag set.
3412 */
3413 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3414 if (ret) {
3415 btrfs_err(fs_info,
3416 "zoned: failed to read device zone info: %d", ret);
3417 goto fail_block_groups;
3418 }
3419
3420 /*
3421 * If we have a uuid root and we're not being told to rescan we need to
3422 * check the generation here so we can set the
3423 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3424 * transaction during a balance or the log replay without updating the
3425 * uuid generation, and then if we crash we would rescan the uuid tree,
3426 * even though it was perfectly fine.
3427 */
3428 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3429 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3430 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3431
3432 ret = btrfs_verify_dev_extents(fs_info);
3433 if (ret) {
3434 btrfs_err(fs_info,
3435 "failed to verify dev extents against chunks: %d",
3436 ret);
3437 goto fail_block_groups;
3438 }
3439 ret = btrfs_recover_balance(fs_info);
3440 if (ret) {
3441 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3442 goto fail_block_groups;
3443 }
3444
3445 ret = btrfs_init_dev_stats(fs_info);
3446 if (ret) {
3447 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3448 goto fail_block_groups;
3449 }
3450
3451 ret = btrfs_init_dev_replace(fs_info);
3452 if (ret) {
3453 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3454 goto fail_block_groups;
3455 }
3456
3457 ret = btrfs_check_zoned_mode(fs_info);
3458 if (ret) {
3459 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3460 ret);
3461 goto fail_block_groups;
3462 }
3463
3464 ret = btrfs_sysfs_add_fsid(fs_devices);
3465 if (ret) {
3466 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3467 ret);
3468 goto fail_block_groups;
3469 }
3470
3471 ret = btrfs_sysfs_add_mounted(fs_info);
3472 if (ret) {
3473 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3474 goto fail_fsdev_sysfs;
3475 }
3476
3477 ret = btrfs_init_space_info(fs_info);
3478 if (ret) {
3479 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3480 goto fail_sysfs;
3481 }
3482
3483 ret = btrfs_read_block_groups(fs_info);
3484 if (ret) {
3485 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3486 goto fail_sysfs;
3487 }
3488
3489 btrfs_free_zone_cache(fs_info);
3490
3491 btrfs_check_active_zone_reservation(fs_info);
3492
3493 if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3494 !btrfs_check_rw_degradable(fs_info, NULL)) {
3495 btrfs_warn(fs_info,
3496 "writable mount is not allowed due to too many missing devices");
3497 ret = -EINVAL;
3498 goto fail_sysfs;
3499 }
3500
3501 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3502 "btrfs-cleaner");
3503 if (IS_ERR(fs_info->cleaner_kthread)) {
3504 ret = PTR_ERR(fs_info->cleaner_kthread);
3505 goto fail_sysfs;
3506 }
3507
3508 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3509 tree_root,
3510 "btrfs-transaction");
3511 if (IS_ERR(fs_info->transaction_kthread)) {
3512 ret = PTR_ERR(fs_info->transaction_kthread);
3513 goto fail_cleaner;
3514 }
3515
3516 ret = btrfs_read_qgroup_config(fs_info);
3517 if (ret)
3518 goto fail_trans_kthread;
3519
3520 if (btrfs_build_ref_tree(fs_info))
3521 btrfs_err(fs_info, "couldn't build ref tree");
3522
3523 /* do not make disk changes in broken FS or nologreplay is given */
3524 if (btrfs_super_log_root(disk_super) != 0 &&
3525 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3526 btrfs_info(fs_info, "start tree-log replay");
3527 ret = btrfs_replay_log(fs_info, fs_devices);
3528 if (ret)
3529 goto fail_qgroup;
3530 }
3531
3532 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3533 if (IS_ERR(fs_info->fs_root)) {
3534 ret = PTR_ERR(fs_info->fs_root);
3535 btrfs_warn(fs_info, "failed to read fs tree: %d", ret);
3536 fs_info->fs_root = NULL;
3537 goto fail_qgroup;
3538 }
3539
3540 if (sb_rdonly(sb))
3541 return 0;
3542
3543 ret = btrfs_start_pre_rw_mount(fs_info);
3544 if (ret) {
3545 close_ctree(fs_info);
3546 return ret;
3547 }
3548 btrfs_discard_resume(fs_info);
3549
3550 if (fs_info->uuid_root &&
3551 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3552 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3553 btrfs_info(fs_info, "checking UUID tree");
3554 ret = btrfs_check_uuid_tree(fs_info);
3555 if (ret) {
3556 btrfs_warn(fs_info,
3557 "failed to check the UUID tree: %d", ret);
3558 close_ctree(fs_info);
3559 return ret;
3560 }
3561 }
3562
3563 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3564
3565 /* Kick the cleaner thread so it'll start deleting snapshots. */
3566 if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3567 wake_up_process(fs_info->cleaner_kthread);
3568
3569 return 0;
3570
3571fail_qgroup:
3572 btrfs_free_qgroup_config(fs_info);
3573fail_trans_kthread:
3574 kthread_stop(fs_info->transaction_kthread);
3575 btrfs_cleanup_transaction(fs_info);
3576 btrfs_free_fs_roots(fs_info);
3577fail_cleaner:
3578 kthread_stop(fs_info->cleaner_kthread);
3579
3580 /*
3581 * make sure we're done with the btree inode before we stop our
3582 * kthreads
3583 */
3584 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3585
3586fail_sysfs:
3587 btrfs_sysfs_remove_mounted(fs_info);
3588
3589fail_fsdev_sysfs:
3590 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3591
3592fail_block_groups:
3593 btrfs_put_block_group_cache(fs_info);
3594
3595fail_tree_roots:
3596 if (fs_info->data_reloc_root)
3597 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3598 free_root_pointers(fs_info, true);
3599 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3600
3601fail_sb_buffer:
3602 btrfs_stop_all_workers(fs_info);
3603 btrfs_free_block_groups(fs_info);
3604fail_alloc:
3605 btrfs_mapping_tree_free(fs_info);
3606
3607 iput(fs_info->btree_inode);
3608fail:
3609 btrfs_close_devices(fs_info->fs_devices);
3610 ASSERT(ret < 0);
3611 return ret;
3612}
3613ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3614
3615static void btrfs_end_super_write(struct bio *bio)
3616{
3617 struct btrfs_device *device = bio->bi_private;
3618 struct folio_iter fi;
3619
3620 bio_for_each_folio_all(fi, bio) {
3621 if (bio->bi_status) {
3622 btrfs_warn_rl_in_rcu(device->fs_info,
3623 "lost super block write due to IO error on %s (%d)",
3624 btrfs_dev_name(device),
3625 blk_status_to_errno(bio->bi_status));
3626 btrfs_dev_stat_inc_and_print(device,
3627 BTRFS_DEV_STAT_WRITE_ERRS);
3628 /* Ensure failure if the primary sb fails. */
3629 if (bio->bi_opf & REQ_FUA)
3630 atomic_add(BTRFS_SUPER_PRIMARY_WRITE_ERROR,
3631 &device->sb_write_errors);
3632 else
3633 atomic_inc(&device->sb_write_errors);
3634 }
3635 folio_unlock(fi.folio);
3636 folio_put(fi.folio);
3637 }
3638
3639 bio_put(bio);
3640}
3641
3642struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3643 int copy_num, bool drop_cache)
3644{
3645 struct btrfs_super_block *super;
3646 struct page *page;
3647 u64 bytenr, bytenr_orig;
3648 struct address_space *mapping = bdev->bd_mapping;
3649 int ret;
3650
3651 bytenr_orig = btrfs_sb_offset(copy_num);
3652 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3653 if (ret == -ENOENT)
3654 return ERR_PTR(-EINVAL);
3655 else if (ret)
3656 return ERR_PTR(ret);
3657
3658 if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3659 return ERR_PTR(-EINVAL);
3660
3661 if (drop_cache) {
3662 /* This should only be called with the primary sb. */
3663 ASSERT(copy_num == 0);
3664
3665 /*
3666 * Drop the page of the primary superblock, so later read will
3667 * always read from the device.
3668 */
3669 invalidate_inode_pages2_range(mapping,
3670 bytenr >> PAGE_SHIFT,
3671 (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3672 }
3673
3674 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3675 if (IS_ERR(page))
3676 return ERR_CAST(page);
3677
3678 super = page_address(page);
3679 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3680 btrfs_release_disk_super(super);
3681 return ERR_PTR(-ENODATA);
3682 }
3683
3684 if (btrfs_super_bytenr(super) != bytenr_orig) {
3685 btrfs_release_disk_super(super);
3686 return ERR_PTR(-EINVAL);
3687 }
3688
3689 return super;
3690}
3691
3692
3693struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3694{
3695 struct btrfs_super_block *super, *latest = NULL;
3696 int i;
3697 u64 transid = 0;
3698
3699 /* we would like to check all the supers, but that would make
3700 * a btrfs mount succeed after a mkfs from a different FS.
3701 * So, we need to add a special mount option to scan for
3702 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3703 */
3704 for (i = 0; i < 1; i++) {
3705 super = btrfs_read_dev_one_super(bdev, i, false);
3706 if (IS_ERR(super))
3707 continue;
3708
3709 if (!latest || btrfs_super_generation(super) > transid) {
3710 if (latest)
3711 btrfs_release_disk_super(super);
3712
3713 latest = super;
3714 transid = btrfs_super_generation(super);
3715 }
3716 }
3717
3718 return super;
3719}
3720
3721/*
3722 * Write superblock @sb to the @device. Do not wait for completion, all the
3723 * folios we use for writing are locked.
3724 *
3725 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3726 * the expected device size at commit time. Note that max_mirrors must be
3727 * same for write and wait phases.
3728 *
3729 * Return number of errors when folio is not found or submission fails.
3730 */
3731static int write_dev_supers(struct btrfs_device *device,
3732 struct btrfs_super_block *sb, int max_mirrors)
3733{
3734 struct btrfs_fs_info *fs_info = device->fs_info;
3735 struct address_space *mapping = device->bdev->bd_mapping;
3736 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3737 int i;
3738 int ret;
3739 u64 bytenr, bytenr_orig;
3740
3741 atomic_set(&device->sb_write_errors, 0);
3742
3743 if (max_mirrors == 0)
3744 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3745
3746 shash->tfm = fs_info->csum_shash;
3747
3748 for (i = 0; i < max_mirrors; i++) {
3749 struct folio *folio;
3750 struct bio *bio;
3751 struct btrfs_super_block *disk_super;
3752 size_t offset;
3753
3754 bytenr_orig = btrfs_sb_offset(i);
3755 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3756 if (ret == -ENOENT) {
3757 continue;
3758 } else if (ret < 0) {
3759 btrfs_err(device->fs_info,
3760 "couldn't get super block location for mirror %d",
3761 i);
3762 atomic_inc(&device->sb_write_errors);
3763 continue;
3764 }
3765 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3766 device->commit_total_bytes)
3767 break;
3768
3769 btrfs_set_super_bytenr(sb, bytenr_orig);
3770
3771 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3772 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3773 sb->csum);
3774
3775 folio = __filemap_get_folio(mapping, bytenr >> PAGE_SHIFT,
3776 FGP_LOCK | FGP_ACCESSED | FGP_CREAT,
3777 GFP_NOFS);
3778 if (IS_ERR(folio)) {
3779 btrfs_err(device->fs_info,
3780 "couldn't get super block page for bytenr %llu",
3781 bytenr);
3782 atomic_inc(&device->sb_write_errors);
3783 continue;
3784 }
3785 ASSERT(folio_order(folio) == 0);
3786
3787 offset = offset_in_folio(folio, bytenr);
3788 disk_super = folio_address(folio) + offset;
3789 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3790
3791 /*
3792 * Directly use bios here instead of relying on the page cache
3793 * to do I/O, so we don't lose the ability to do integrity
3794 * checking.
3795 */
3796 bio = bio_alloc(device->bdev, 1,
3797 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
3798 GFP_NOFS);
3799 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3800 bio->bi_private = device;
3801 bio->bi_end_io = btrfs_end_super_write;
3802 bio_add_folio_nofail(bio, folio, BTRFS_SUPER_INFO_SIZE, offset);
3803
3804 /*
3805 * We FUA only the first super block. The others we allow to
3806 * go down lazy and there's a short window where the on-disk
3807 * copies might still contain the older version.
3808 */
3809 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3810 bio->bi_opf |= REQ_FUA;
3811 submit_bio(bio);
3812
3813 if (btrfs_advance_sb_log(device, i))
3814 atomic_inc(&device->sb_write_errors);
3815 }
3816 return atomic_read(&device->sb_write_errors) < i ? 0 : -1;
3817}
3818
3819/*
3820 * Wait for write completion of superblocks done by write_dev_supers,
3821 * @max_mirrors same for write and wait phases.
3822 *
3823 * Return -1 if primary super block write failed or when there were no super block
3824 * copies written. Otherwise 0.
3825 */
3826static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3827{
3828 int i;
3829 int errors = 0;
3830 bool primary_failed = false;
3831 int ret;
3832 u64 bytenr;
3833
3834 if (max_mirrors == 0)
3835 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3836
3837 for (i = 0; i < max_mirrors; i++) {
3838 struct folio *folio;
3839
3840 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3841 if (ret == -ENOENT) {
3842 break;
3843 } else if (ret < 0) {
3844 errors++;
3845 if (i == 0)
3846 primary_failed = true;
3847 continue;
3848 }
3849 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3850 device->commit_total_bytes)
3851 break;
3852
3853 folio = filemap_get_folio(device->bdev->bd_mapping,
3854 bytenr >> PAGE_SHIFT);
3855 /* If the folio has been removed, then we know it completed. */
3856 if (IS_ERR(folio))
3857 continue;
3858 ASSERT(folio_order(folio) == 0);
3859
3860 /* Folio will be unlocked once the write completes. */
3861 folio_wait_locked(folio);
3862 folio_put(folio);
3863 }
3864
3865 errors += atomic_read(&device->sb_write_errors);
3866 if (errors >= BTRFS_SUPER_PRIMARY_WRITE_ERROR)
3867 primary_failed = true;
3868 if (primary_failed) {
3869 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3870 device->devid);
3871 return -1;
3872 }
3873
3874 return errors < i ? 0 : -1;
3875}
3876
3877/*
3878 * endio for the write_dev_flush, this will wake anyone waiting
3879 * for the barrier when it is done
3880 */
3881static void btrfs_end_empty_barrier(struct bio *bio)
3882{
3883 bio_uninit(bio);
3884 complete(bio->bi_private);
3885}
3886
3887/*
3888 * Submit a flush request to the device if it supports it. Error handling is
3889 * done in the waiting counterpart.
3890 */
3891static void write_dev_flush(struct btrfs_device *device)
3892{
3893 struct bio *bio = &device->flush_bio;
3894
3895 device->last_flush_error = BLK_STS_OK;
3896
3897 bio_init(bio, device->bdev, NULL, 0,
3898 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
3899 bio->bi_end_io = btrfs_end_empty_barrier;
3900 init_completion(&device->flush_wait);
3901 bio->bi_private = &device->flush_wait;
3902 submit_bio(bio);
3903 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3904}
3905
3906/*
3907 * If the flush bio has been submitted by write_dev_flush, wait for it.
3908 * Return true for any error, and false otherwise.
3909 */
3910static bool wait_dev_flush(struct btrfs_device *device)
3911{
3912 struct bio *bio = &device->flush_bio;
3913
3914 if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3915 return false;
3916
3917 wait_for_completion_io(&device->flush_wait);
3918
3919 if (bio->bi_status) {
3920 device->last_flush_error = bio->bi_status;
3921 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_FLUSH_ERRS);
3922 return true;
3923 }
3924
3925 return false;
3926}
3927
3928/*
3929 * send an empty flush down to each device in parallel,
3930 * then wait for them
3931 */
3932static int barrier_all_devices(struct btrfs_fs_info *info)
3933{
3934 struct list_head *head;
3935 struct btrfs_device *dev;
3936 int errors_wait = 0;
3937
3938 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3939 /* send down all the barriers */
3940 head = &info->fs_devices->devices;
3941 list_for_each_entry(dev, head, dev_list) {
3942 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3943 continue;
3944 if (!dev->bdev)
3945 continue;
3946 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3947 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3948 continue;
3949
3950 write_dev_flush(dev);
3951 }
3952
3953 /* wait for all the barriers */
3954 list_for_each_entry(dev, head, dev_list) {
3955 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3956 continue;
3957 if (!dev->bdev) {
3958 errors_wait++;
3959 continue;
3960 }
3961 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3962 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3963 continue;
3964
3965 if (wait_dev_flush(dev))
3966 errors_wait++;
3967 }
3968
3969 /*
3970 * Checks last_flush_error of disks in order to determine the device
3971 * state.
3972 */
3973 if (errors_wait && !btrfs_check_rw_degradable(info, NULL))
3974 return -EIO;
3975
3976 return 0;
3977}
3978
3979int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3980{
3981 int raid_type;
3982 int min_tolerated = INT_MAX;
3983
3984 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3985 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3986 min_tolerated = min_t(int, min_tolerated,
3987 btrfs_raid_array[BTRFS_RAID_SINGLE].
3988 tolerated_failures);
3989
3990 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3991 if (raid_type == BTRFS_RAID_SINGLE)
3992 continue;
3993 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3994 continue;
3995 min_tolerated = min_t(int, min_tolerated,
3996 btrfs_raid_array[raid_type].
3997 tolerated_failures);
3998 }
3999
4000 if (min_tolerated == INT_MAX) {
4001 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4002 min_tolerated = 0;
4003 }
4004
4005 return min_tolerated;
4006}
4007
4008int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4009{
4010 struct list_head *head;
4011 struct btrfs_device *dev;
4012 struct btrfs_super_block *sb;
4013 struct btrfs_dev_item *dev_item;
4014 int ret;
4015 int do_barriers;
4016 int max_errors;
4017 int total_errors = 0;
4018 u64 flags;
4019
4020 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4021
4022 /*
4023 * max_mirrors == 0 indicates we're from commit_transaction,
4024 * not from fsync where the tree roots in fs_info have not
4025 * been consistent on disk.
4026 */
4027 if (max_mirrors == 0)
4028 backup_super_roots(fs_info);
4029
4030 sb = fs_info->super_for_commit;
4031 dev_item = &sb->dev_item;
4032
4033 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4034 head = &fs_info->fs_devices->devices;
4035 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4036
4037 if (do_barriers) {
4038 ret = barrier_all_devices(fs_info);
4039 if (ret) {
4040 mutex_unlock(
4041 &fs_info->fs_devices->device_list_mutex);
4042 btrfs_handle_fs_error(fs_info, ret,
4043 "errors while submitting device barriers.");
4044 return ret;
4045 }
4046 }
4047
4048 list_for_each_entry(dev, head, dev_list) {
4049 if (!dev->bdev) {
4050 total_errors++;
4051 continue;
4052 }
4053 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4054 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4055 continue;
4056
4057 btrfs_set_stack_device_generation(dev_item, 0);
4058 btrfs_set_stack_device_type(dev_item, dev->type);
4059 btrfs_set_stack_device_id(dev_item, dev->devid);
4060 btrfs_set_stack_device_total_bytes(dev_item,
4061 dev->commit_total_bytes);
4062 btrfs_set_stack_device_bytes_used(dev_item,
4063 dev->commit_bytes_used);
4064 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4065 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4066 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4067 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4068 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4069 BTRFS_FSID_SIZE);
4070
4071 flags = btrfs_super_flags(sb);
4072 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4073
4074 ret = btrfs_validate_write_super(fs_info, sb);
4075 if (ret < 0) {
4076 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4077 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4078 "unexpected superblock corruption detected");
4079 return -EUCLEAN;
4080 }
4081
4082 ret = write_dev_supers(dev, sb, max_mirrors);
4083 if (ret)
4084 total_errors++;
4085 }
4086 if (total_errors > max_errors) {
4087 btrfs_err(fs_info, "%d errors while writing supers",
4088 total_errors);
4089 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4090
4091 /* FUA is masked off if unsupported and can't be the reason */
4092 btrfs_handle_fs_error(fs_info, -EIO,
4093 "%d errors while writing supers",
4094 total_errors);
4095 return -EIO;
4096 }
4097
4098 total_errors = 0;
4099 list_for_each_entry(dev, head, dev_list) {
4100 if (!dev->bdev)
4101 continue;
4102 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4103 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4104 continue;
4105
4106 ret = wait_dev_supers(dev, max_mirrors);
4107 if (ret)
4108 total_errors++;
4109 }
4110 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4111 if (total_errors > max_errors) {
4112 btrfs_handle_fs_error(fs_info, -EIO,
4113 "%d errors while writing supers",
4114 total_errors);
4115 return -EIO;
4116 }
4117 return 0;
4118}
4119
4120/* Drop a fs root from the radix tree and free it. */
4121void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4122 struct btrfs_root *root)
4123{
4124 bool drop_ref = false;
4125
4126 spin_lock(&fs_info->fs_roots_radix_lock);
4127 radix_tree_delete(&fs_info->fs_roots_radix,
4128 (unsigned long)btrfs_root_id(root));
4129 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4130 drop_ref = true;
4131 spin_unlock(&fs_info->fs_roots_radix_lock);
4132
4133 if (BTRFS_FS_ERROR(fs_info)) {
4134 ASSERT(root->log_root == NULL);
4135 if (root->reloc_root) {
4136 btrfs_put_root(root->reloc_root);
4137 root->reloc_root = NULL;
4138 }
4139 }
4140
4141 if (drop_ref)
4142 btrfs_put_root(root);
4143}
4144
4145int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4146{
4147 mutex_lock(&fs_info->cleaner_mutex);
4148 btrfs_run_delayed_iputs(fs_info);
4149 mutex_unlock(&fs_info->cleaner_mutex);
4150 wake_up_process(fs_info->cleaner_kthread);
4151
4152 /* wait until ongoing cleanup work done */
4153 down_write(&fs_info->cleanup_work_sem);
4154 up_write(&fs_info->cleanup_work_sem);
4155
4156 return btrfs_commit_current_transaction(fs_info->tree_root);
4157}
4158
4159static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4160{
4161 struct btrfs_transaction *trans;
4162 struct btrfs_transaction *tmp;
4163 bool found = false;
4164
4165 /*
4166 * This function is only called at the very end of close_ctree(),
4167 * thus no other running transaction, no need to take trans_lock.
4168 */
4169 ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4170 list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4171 struct extent_state *cached = NULL;
4172 u64 dirty_bytes = 0;
4173 u64 cur = 0;
4174 u64 found_start;
4175 u64 found_end;
4176
4177 found = true;
4178 while (find_first_extent_bit(&trans->dirty_pages, cur,
4179 &found_start, &found_end, EXTENT_DIRTY, &cached)) {
4180 dirty_bytes += found_end + 1 - found_start;
4181 cur = found_end + 1;
4182 }
4183 btrfs_warn(fs_info,
4184 "transaction %llu (with %llu dirty metadata bytes) is not committed",
4185 trans->transid, dirty_bytes);
4186 btrfs_cleanup_one_transaction(trans);
4187
4188 if (trans == fs_info->running_transaction)
4189 fs_info->running_transaction = NULL;
4190 list_del_init(&trans->list);
4191
4192 btrfs_put_transaction(trans);
4193 trace_btrfs_transaction_commit(fs_info);
4194 }
4195 ASSERT(!found);
4196}
4197
4198void __cold close_ctree(struct btrfs_fs_info *fs_info)
4199{
4200 int ret;
4201
4202 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4203
4204 /*
4205 * If we had UNFINISHED_DROPS we could still be processing them, so
4206 * clear that bit and wake up relocation so it can stop.
4207 * We must do this before stopping the block group reclaim task, because
4208 * at btrfs_relocate_block_group() we wait for this bit, and after the
4209 * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4210 * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4211 * return 1.
4212 */
4213 btrfs_wake_unfinished_drop(fs_info);
4214
4215 /*
4216 * We may have the reclaim task running and relocating a data block group,
4217 * in which case it may create delayed iputs. So stop it before we park
4218 * the cleaner kthread otherwise we can get new delayed iputs after
4219 * parking the cleaner, and that can make the async reclaim task to hang
4220 * if it's waiting for delayed iputs to complete, since the cleaner is
4221 * parked and can not run delayed iputs - this will make us hang when
4222 * trying to stop the async reclaim task.
4223 */
4224 cancel_work_sync(&fs_info->reclaim_bgs_work);
4225 /*
4226 * We don't want the cleaner to start new transactions, add more delayed
4227 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4228 * because that frees the task_struct, and the transaction kthread might
4229 * still try to wake up the cleaner.
4230 */
4231 kthread_park(fs_info->cleaner_kthread);
4232
4233 /* wait for the qgroup rescan worker to stop */
4234 btrfs_qgroup_wait_for_completion(fs_info, false);
4235
4236 /* wait for the uuid_scan task to finish */
4237 down(&fs_info->uuid_tree_rescan_sem);
4238 /* avoid complains from lockdep et al., set sem back to initial state */
4239 up(&fs_info->uuid_tree_rescan_sem);
4240
4241 /* pause restriper - we want to resume on mount */
4242 btrfs_pause_balance(fs_info);
4243
4244 btrfs_dev_replace_suspend_for_unmount(fs_info);
4245
4246 btrfs_scrub_cancel(fs_info);
4247
4248 /* wait for any defraggers to finish */
4249 wait_event(fs_info->transaction_wait,
4250 (atomic_read(&fs_info->defrag_running) == 0));
4251
4252 /* clear out the rbtree of defraggable inodes */
4253 btrfs_cleanup_defrag_inodes(fs_info);
4254
4255 /*
4256 * Wait for any fixup workers to complete.
4257 * If we don't wait for them here and they are still running by the time
4258 * we call kthread_stop() against the cleaner kthread further below, we
4259 * get an use-after-free on the cleaner because the fixup worker adds an
4260 * inode to the list of delayed iputs and then attempts to wakeup the
4261 * cleaner kthread, which was already stopped and destroyed. We parked
4262 * already the cleaner, but below we run all pending delayed iputs.
4263 */
4264 btrfs_flush_workqueue(fs_info->fixup_workers);
4265 /*
4266 * Similar case here, we have to wait for delalloc workers before we
4267 * proceed below and stop the cleaner kthread, otherwise we trigger a
4268 * use-after-tree on the cleaner kthread task_struct when a delalloc
4269 * worker running submit_compressed_extents() adds a delayed iput, which
4270 * does a wake up on the cleaner kthread, which was already freed below
4271 * when we call kthread_stop().
4272 */
4273 btrfs_flush_workqueue(fs_info->delalloc_workers);
4274
4275 /*
4276 * After we parked the cleaner kthread, ordered extents may have
4277 * completed and created new delayed iputs. If one of the async reclaim
4278 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4279 * can hang forever trying to stop it, because if a delayed iput is
4280 * added after it ran btrfs_run_delayed_iputs() and before it called
4281 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4282 * no one else to run iputs.
4283 *
4284 * So wait for all ongoing ordered extents to complete and then run
4285 * delayed iputs. This works because once we reach this point no one
4286 * can either create new ordered extents nor create delayed iputs
4287 * through some other means.
4288 *
4289 * Also note that btrfs_wait_ordered_roots() is not safe here, because
4290 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4291 * but the delayed iput for the respective inode is made only when doing
4292 * the final btrfs_put_ordered_extent() (which must happen at
4293 * btrfs_finish_ordered_io() when we are unmounting).
4294 */
4295 btrfs_flush_workqueue(fs_info->endio_write_workers);
4296 /* Ordered extents for free space inodes. */
4297 btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4298 btrfs_run_delayed_iputs(fs_info);
4299
4300 cancel_work_sync(&fs_info->async_reclaim_work);
4301 cancel_work_sync(&fs_info->async_data_reclaim_work);
4302 cancel_work_sync(&fs_info->preempt_reclaim_work);
4303 cancel_work_sync(&fs_info->em_shrinker_work);
4304
4305 /* Cancel or finish ongoing discard work */
4306 btrfs_discard_cleanup(fs_info);
4307
4308 if (!sb_rdonly(fs_info->sb)) {
4309 /*
4310 * The cleaner kthread is stopped, so do one final pass over
4311 * unused block groups.
4312 */
4313 btrfs_delete_unused_bgs(fs_info);
4314
4315 /*
4316 * There might be existing delayed inode workers still running
4317 * and holding an empty delayed inode item. We must wait for
4318 * them to complete first because they can create a transaction.
4319 * This happens when someone calls btrfs_balance_delayed_items()
4320 * and then a transaction commit runs the same delayed nodes
4321 * before any delayed worker has done something with the nodes.
4322 * We must wait for any worker here and not at transaction
4323 * commit time since that could cause a deadlock.
4324 * This is a very rare case.
4325 */
4326 btrfs_flush_workqueue(fs_info->delayed_workers);
4327
4328 ret = btrfs_commit_super(fs_info);
4329 if (ret)
4330 btrfs_err(fs_info, "commit super ret %d", ret);
4331 }
4332
4333 if (BTRFS_FS_ERROR(fs_info))
4334 btrfs_error_commit_super(fs_info);
4335
4336 kthread_stop(fs_info->transaction_kthread);
4337 kthread_stop(fs_info->cleaner_kthread);
4338
4339 ASSERT(list_empty(&fs_info->delayed_iputs));
4340 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4341
4342 if (btrfs_check_quota_leak(fs_info)) {
4343 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4344 btrfs_err(fs_info, "qgroup reserved space leaked");
4345 }
4346
4347 btrfs_free_qgroup_config(fs_info);
4348 ASSERT(list_empty(&fs_info->delalloc_roots));
4349
4350 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4351 btrfs_info(fs_info, "at unmount delalloc count %lld",
4352 percpu_counter_sum(&fs_info->delalloc_bytes));
4353 }
4354
4355 if (percpu_counter_sum(&fs_info->ordered_bytes))
4356 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4357 percpu_counter_sum(&fs_info->ordered_bytes));
4358
4359 btrfs_sysfs_remove_mounted(fs_info);
4360 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4361
4362 btrfs_put_block_group_cache(fs_info);
4363
4364 /*
4365 * we must make sure there is not any read request to
4366 * submit after we stopping all workers.
4367 */
4368 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4369 btrfs_stop_all_workers(fs_info);
4370
4371 /* We shouldn't have any transaction open at this point */
4372 warn_about_uncommitted_trans(fs_info);
4373
4374 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4375 free_root_pointers(fs_info, true);
4376 btrfs_free_fs_roots(fs_info);
4377
4378 /*
4379 * We must free the block groups after dropping the fs_roots as we could
4380 * have had an IO error and have left over tree log blocks that aren't
4381 * cleaned up until the fs roots are freed. This makes the block group
4382 * accounting appear to be wrong because there's pending reserved bytes,
4383 * so make sure we do the block group cleanup afterwards.
4384 */
4385 btrfs_free_block_groups(fs_info);
4386
4387 iput(fs_info->btree_inode);
4388
4389 btrfs_mapping_tree_free(fs_info);
4390 btrfs_close_devices(fs_info->fs_devices);
4391}
4392
4393void btrfs_mark_buffer_dirty(struct btrfs_trans_handle *trans,
4394 struct extent_buffer *buf)
4395{
4396 struct btrfs_fs_info *fs_info = buf->fs_info;
4397 u64 transid = btrfs_header_generation(buf);
4398
4399#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4400 /*
4401 * This is a fast path so only do this check if we have sanity tests
4402 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4403 * outside of the sanity tests.
4404 */
4405 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4406 return;
4407#endif
4408 /* This is an active transaction (its state < TRANS_STATE_UNBLOCKED). */
4409 ASSERT(trans->transid == fs_info->generation);
4410 btrfs_assert_tree_write_locked(buf);
4411 if (unlikely(transid != fs_info->generation)) {
4412 btrfs_abort_transaction(trans, -EUCLEAN);
4413 btrfs_crit(fs_info,
4414"dirty buffer transid mismatch, logical %llu found transid %llu running transid %llu",
4415 buf->start, transid, fs_info->generation);
4416 }
4417 set_extent_buffer_dirty(buf);
4418}
4419
4420static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4421 int flush_delayed)
4422{
4423 /*
4424 * looks as though older kernels can get into trouble with
4425 * this code, they end up stuck in balance_dirty_pages forever
4426 */
4427 int ret;
4428
4429 if (current->flags & PF_MEMALLOC)
4430 return;
4431
4432 if (flush_delayed)
4433 btrfs_balance_delayed_items(fs_info);
4434
4435 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4436 BTRFS_DIRTY_METADATA_THRESH,
4437 fs_info->dirty_metadata_batch);
4438 if (ret > 0) {
4439 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4440 }
4441}
4442
4443void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4444{
4445 __btrfs_btree_balance_dirty(fs_info, 1);
4446}
4447
4448void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4449{
4450 __btrfs_btree_balance_dirty(fs_info, 0);
4451}
4452
4453static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4454{
4455 /* cleanup FS via transaction */
4456 btrfs_cleanup_transaction(fs_info);
4457
4458 mutex_lock(&fs_info->cleaner_mutex);
4459 btrfs_run_delayed_iputs(fs_info);
4460 mutex_unlock(&fs_info->cleaner_mutex);
4461
4462 down_write(&fs_info->cleanup_work_sem);
4463 up_write(&fs_info->cleanup_work_sem);
4464}
4465
4466static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4467{
4468 struct btrfs_root *gang[8];
4469 u64 root_objectid = 0;
4470 int ret;
4471
4472 spin_lock(&fs_info->fs_roots_radix_lock);
4473 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4474 (void **)gang, root_objectid,
4475 ARRAY_SIZE(gang))) != 0) {
4476 int i;
4477
4478 for (i = 0; i < ret; i++)
4479 gang[i] = btrfs_grab_root(gang[i]);
4480 spin_unlock(&fs_info->fs_roots_radix_lock);
4481
4482 for (i = 0; i < ret; i++) {
4483 if (!gang[i])
4484 continue;
4485 root_objectid = btrfs_root_id(gang[i]);
4486 btrfs_free_log(NULL, gang[i]);
4487 btrfs_put_root(gang[i]);
4488 }
4489 root_objectid++;
4490 spin_lock(&fs_info->fs_roots_radix_lock);
4491 }
4492 spin_unlock(&fs_info->fs_roots_radix_lock);
4493 btrfs_free_log_root_tree(NULL, fs_info);
4494}
4495
4496static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4497{
4498 struct btrfs_ordered_extent *ordered;
4499
4500 spin_lock(&root->ordered_extent_lock);
4501 /*
4502 * This will just short circuit the ordered completion stuff which will
4503 * make sure the ordered extent gets properly cleaned up.
4504 */
4505 list_for_each_entry(ordered, &root->ordered_extents,
4506 root_extent_list)
4507 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4508 spin_unlock(&root->ordered_extent_lock);
4509}
4510
4511static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4512{
4513 struct btrfs_root *root;
4514 LIST_HEAD(splice);
4515
4516 spin_lock(&fs_info->ordered_root_lock);
4517 list_splice_init(&fs_info->ordered_roots, &splice);
4518 while (!list_empty(&splice)) {
4519 root = list_first_entry(&splice, struct btrfs_root,
4520 ordered_root);
4521 list_move_tail(&root->ordered_root,
4522 &fs_info->ordered_roots);
4523
4524 spin_unlock(&fs_info->ordered_root_lock);
4525 btrfs_destroy_ordered_extents(root);
4526
4527 cond_resched();
4528 spin_lock(&fs_info->ordered_root_lock);
4529 }
4530 spin_unlock(&fs_info->ordered_root_lock);
4531
4532 /*
4533 * We need this here because if we've been flipped read-only we won't
4534 * get sync() from the umount, so we need to make sure any ordered
4535 * extents that haven't had their dirty pages IO start writeout yet
4536 * actually get run and error out properly.
4537 */
4538 btrfs_wait_ordered_roots(fs_info, U64_MAX, NULL);
4539}
4540
4541static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4542{
4543 struct btrfs_inode *btrfs_inode;
4544 LIST_HEAD(splice);
4545
4546 spin_lock(&root->delalloc_lock);
4547 list_splice_init(&root->delalloc_inodes, &splice);
4548
4549 while (!list_empty(&splice)) {
4550 struct inode *inode = NULL;
4551 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4552 delalloc_inodes);
4553 btrfs_del_delalloc_inode(btrfs_inode);
4554 spin_unlock(&root->delalloc_lock);
4555
4556 /*
4557 * Make sure we get a live inode and that it'll not disappear
4558 * meanwhile.
4559 */
4560 inode = igrab(&btrfs_inode->vfs_inode);
4561 if (inode) {
4562 unsigned int nofs_flag;
4563
4564 nofs_flag = memalloc_nofs_save();
4565 invalidate_inode_pages2(inode->i_mapping);
4566 memalloc_nofs_restore(nofs_flag);
4567 iput(inode);
4568 }
4569 spin_lock(&root->delalloc_lock);
4570 }
4571 spin_unlock(&root->delalloc_lock);
4572}
4573
4574static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4575{
4576 struct btrfs_root *root;
4577 LIST_HEAD(splice);
4578
4579 spin_lock(&fs_info->delalloc_root_lock);
4580 list_splice_init(&fs_info->delalloc_roots, &splice);
4581 while (!list_empty(&splice)) {
4582 root = list_first_entry(&splice, struct btrfs_root,
4583 delalloc_root);
4584 root = btrfs_grab_root(root);
4585 BUG_ON(!root);
4586 spin_unlock(&fs_info->delalloc_root_lock);
4587
4588 btrfs_destroy_delalloc_inodes(root);
4589 btrfs_put_root(root);
4590
4591 spin_lock(&fs_info->delalloc_root_lock);
4592 }
4593 spin_unlock(&fs_info->delalloc_root_lock);
4594}
4595
4596static void btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4597 struct extent_io_tree *dirty_pages,
4598 int mark)
4599{
4600 struct extent_buffer *eb;
4601 u64 start = 0;
4602 u64 end;
4603
4604 while (find_first_extent_bit(dirty_pages, start, &start, &end,
4605 mark, NULL)) {
4606 clear_extent_bits(dirty_pages, start, end, mark);
4607 while (start <= end) {
4608 eb = find_extent_buffer(fs_info, start);
4609 start += fs_info->nodesize;
4610 if (!eb)
4611 continue;
4612
4613 btrfs_tree_lock(eb);
4614 wait_on_extent_buffer_writeback(eb);
4615 btrfs_clear_buffer_dirty(NULL, eb);
4616 btrfs_tree_unlock(eb);
4617
4618 free_extent_buffer_stale(eb);
4619 }
4620 }
4621}
4622
4623static void btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4624 struct extent_io_tree *unpin)
4625{
4626 u64 start;
4627 u64 end;
4628
4629 while (1) {
4630 struct extent_state *cached_state = NULL;
4631
4632 /*
4633 * The btrfs_finish_extent_commit() may get the same range as
4634 * ours between find_first_extent_bit and clear_extent_dirty.
4635 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4636 * the same extent range.
4637 */
4638 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4639 if (!find_first_extent_bit(unpin, 0, &start, &end,
4640 EXTENT_DIRTY, &cached_state)) {
4641 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4642 break;
4643 }
4644
4645 clear_extent_dirty(unpin, start, end, &cached_state);
4646 free_extent_state(cached_state);
4647 btrfs_error_unpin_extent_range(fs_info, start, end);
4648 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4649 cond_resched();
4650 }
4651}
4652
4653static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4654{
4655 struct inode *inode;
4656
4657 inode = cache->io_ctl.inode;
4658 if (inode) {
4659 unsigned int nofs_flag;
4660
4661 nofs_flag = memalloc_nofs_save();
4662 invalidate_inode_pages2(inode->i_mapping);
4663 memalloc_nofs_restore(nofs_flag);
4664
4665 BTRFS_I(inode)->generation = 0;
4666 cache->io_ctl.inode = NULL;
4667 iput(inode);
4668 }
4669 ASSERT(cache->io_ctl.pages == NULL);
4670 btrfs_put_block_group(cache);
4671}
4672
4673void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4674 struct btrfs_fs_info *fs_info)
4675{
4676 struct btrfs_block_group *cache;
4677
4678 spin_lock(&cur_trans->dirty_bgs_lock);
4679 while (!list_empty(&cur_trans->dirty_bgs)) {
4680 cache = list_first_entry(&cur_trans->dirty_bgs,
4681 struct btrfs_block_group,
4682 dirty_list);
4683
4684 if (!list_empty(&cache->io_list)) {
4685 spin_unlock(&cur_trans->dirty_bgs_lock);
4686 list_del_init(&cache->io_list);
4687 btrfs_cleanup_bg_io(cache);
4688 spin_lock(&cur_trans->dirty_bgs_lock);
4689 }
4690
4691 list_del_init(&cache->dirty_list);
4692 spin_lock(&cache->lock);
4693 cache->disk_cache_state = BTRFS_DC_ERROR;
4694 spin_unlock(&cache->lock);
4695
4696 spin_unlock(&cur_trans->dirty_bgs_lock);
4697 btrfs_put_block_group(cache);
4698 btrfs_dec_delayed_refs_rsv_bg_updates(fs_info);
4699 spin_lock(&cur_trans->dirty_bgs_lock);
4700 }
4701 spin_unlock(&cur_trans->dirty_bgs_lock);
4702
4703 /*
4704 * Refer to the definition of io_bgs member for details why it's safe
4705 * to use it without any locking
4706 */
4707 while (!list_empty(&cur_trans->io_bgs)) {
4708 cache = list_first_entry(&cur_trans->io_bgs,
4709 struct btrfs_block_group,
4710 io_list);
4711
4712 list_del_init(&cache->io_list);
4713 spin_lock(&cache->lock);
4714 cache->disk_cache_state = BTRFS_DC_ERROR;
4715 spin_unlock(&cache->lock);
4716 btrfs_cleanup_bg_io(cache);
4717 }
4718}
4719
4720static void btrfs_free_all_qgroup_pertrans(struct btrfs_fs_info *fs_info)
4721{
4722 struct btrfs_root *gang[8];
4723 int i;
4724 int ret;
4725
4726 spin_lock(&fs_info->fs_roots_radix_lock);
4727 while (1) {
4728 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
4729 (void **)gang, 0,
4730 ARRAY_SIZE(gang),
4731 BTRFS_ROOT_TRANS_TAG);
4732 if (ret == 0)
4733 break;
4734 for (i = 0; i < ret; i++) {
4735 struct btrfs_root *root = gang[i];
4736
4737 btrfs_qgroup_free_meta_all_pertrans(root);
4738 radix_tree_tag_clear(&fs_info->fs_roots_radix,
4739 (unsigned long)btrfs_root_id(root),
4740 BTRFS_ROOT_TRANS_TAG);
4741 }
4742 }
4743 spin_unlock(&fs_info->fs_roots_radix_lock);
4744}
4745
4746void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans)
4747{
4748 struct btrfs_fs_info *fs_info = cur_trans->fs_info;
4749 struct btrfs_device *dev, *tmp;
4750
4751 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4752 ASSERT(list_empty(&cur_trans->dirty_bgs));
4753 ASSERT(list_empty(&cur_trans->io_bgs));
4754
4755 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4756 post_commit_list) {
4757 list_del_init(&dev->post_commit_list);
4758 }
4759
4760 btrfs_destroy_delayed_refs(cur_trans);
4761
4762 cur_trans->state = TRANS_STATE_COMMIT_START;
4763 wake_up(&fs_info->transaction_blocked_wait);
4764
4765 cur_trans->state = TRANS_STATE_UNBLOCKED;
4766 wake_up(&fs_info->transaction_wait);
4767
4768 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4769 EXTENT_DIRTY);
4770 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4771
4772 cur_trans->state =TRANS_STATE_COMPLETED;
4773 wake_up(&cur_trans->commit_wait);
4774}
4775
4776static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4777{
4778 struct btrfs_transaction *t;
4779
4780 mutex_lock(&fs_info->transaction_kthread_mutex);
4781
4782 spin_lock(&fs_info->trans_lock);
4783 while (!list_empty(&fs_info->trans_list)) {
4784 t = list_first_entry(&fs_info->trans_list,
4785 struct btrfs_transaction, list);
4786 if (t->state >= TRANS_STATE_COMMIT_PREP) {
4787 refcount_inc(&t->use_count);
4788 spin_unlock(&fs_info->trans_lock);
4789 btrfs_wait_for_commit(fs_info, t->transid);
4790 btrfs_put_transaction(t);
4791 spin_lock(&fs_info->trans_lock);
4792 continue;
4793 }
4794 if (t == fs_info->running_transaction) {
4795 t->state = TRANS_STATE_COMMIT_DOING;
4796 spin_unlock(&fs_info->trans_lock);
4797 /*
4798 * We wait for 0 num_writers since we don't hold a trans
4799 * handle open currently for this transaction.
4800 */
4801 wait_event(t->writer_wait,
4802 atomic_read(&t->num_writers) == 0);
4803 } else {
4804 spin_unlock(&fs_info->trans_lock);
4805 }
4806 btrfs_cleanup_one_transaction(t);
4807
4808 spin_lock(&fs_info->trans_lock);
4809 if (t == fs_info->running_transaction)
4810 fs_info->running_transaction = NULL;
4811 list_del_init(&t->list);
4812 spin_unlock(&fs_info->trans_lock);
4813
4814 btrfs_put_transaction(t);
4815 trace_btrfs_transaction_commit(fs_info);
4816 spin_lock(&fs_info->trans_lock);
4817 }
4818 spin_unlock(&fs_info->trans_lock);
4819 btrfs_destroy_all_ordered_extents(fs_info);
4820 btrfs_destroy_delayed_inodes(fs_info);
4821 btrfs_assert_delayed_root_empty(fs_info);
4822 btrfs_destroy_all_delalloc_inodes(fs_info);
4823 btrfs_drop_all_logs(fs_info);
4824 btrfs_free_all_qgroup_pertrans(fs_info);
4825 mutex_unlock(&fs_info->transaction_kthread_mutex);
4826
4827 return 0;
4828}
4829
4830int btrfs_init_root_free_objectid(struct btrfs_root *root)
4831{
4832 struct btrfs_path *path;
4833 int ret;
4834 struct extent_buffer *l;
4835 struct btrfs_key search_key;
4836 struct btrfs_key found_key;
4837 int slot;
4838
4839 path = btrfs_alloc_path();
4840 if (!path)
4841 return -ENOMEM;
4842
4843 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4844 search_key.type = -1;
4845 search_key.offset = (u64)-1;
4846 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4847 if (ret < 0)
4848 goto error;
4849 if (ret == 0) {
4850 /*
4851 * Key with offset -1 found, there would have to exist a root
4852 * with such id, but this is out of valid range.
4853 */
4854 ret = -EUCLEAN;
4855 goto error;
4856 }
4857 if (path->slots[0] > 0) {
4858 slot = path->slots[0] - 1;
4859 l = path->nodes[0];
4860 btrfs_item_key_to_cpu(l, &found_key, slot);
4861 root->free_objectid = max_t(u64, found_key.objectid + 1,
4862 BTRFS_FIRST_FREE_OBJECTID);
4863 } else {
4864 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4865 }
4866 ret = 0;
4867error:
4868 btrfs_free_path(path);
4869 return ret;
4870}
4871
4872int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4873{
4874 int ret;
4875 mutex_lock(&root->objectid_mutex);
4876
4877 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4878 btrfs_warn(root->fs_info,
4879 "the objectid of root %llu reaches its highest value",
4880 btrfs_root_id(root));
4881 ret = -ENOSPC;
4882 goto out;
4883 }
4884
4885 *objectid = root->free_objectid++;
4886 ret = 0;
4887out:
4888 mutex_unlock(&root->objectid_mutex);
4889 return ret;
4890}
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/crc32c.h>
30#include <linux/slab.h>
31#include <linux/migrate.h>
32#include <linux/ratelimit.h>
33#include <asm/unaligned.h>
34#include "compat.h"
35#include "ctree.h"
36#include "disk-io.h"
37#include "transaction.h"
38#include "btrfs_inode.h"
39#include "volumes.h"
40#include "print-tree.h"
41#include "async-thread.h"
42#include "locking.h"
43#include "tree-log.h"
44#include "free-space-cache.h"
45#include "inode-map.h"
46#include "check-integrity.h"
47#include "rcu-string.h"
48
49static struct extent_io_ops btree_extent_io_ops;
50static void end_workqueue_fn(struct btrfs_work *work);
51static void free_fs_root(struct btrfs_root *root);
52static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
53 int read_only);
54static void btrfs_destroy_ordered_operations(struct btrfs_root *root);
55static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
56static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
57 struct btrfs_root *root);
58static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
59static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
60static int btrfs_destroy_marked_extents(struct btrfs_root *root,
61 struct extent_io_tree *dirty_pages,
62 int mark);
63static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
64 struct extent_io_tree *pinned_extents);
65
66/*
67 * end_io_wq structs are used to do processing in task context when an IO is
68 * complete. This is used during reads to verify checksums, and it is used
69 * by writes to insert metadata for new file extents after IO is complete.
70 */
71struct end_io_wq {
72 struct bio *bio;
73 bio_end_io_t *end_io;
74 void *private;
75 struct btrfs_fs_info *info;
76 int error;
77 int metadata;
78 struct list_head list;
79 struct btrfs_work work;
80};
81
82/*
83 * async submit bios are used to offload expensive checksumming
84 * onto the worker threads. They checksum file and metadata bios
85 * just before they are sent down the IO stack.
86 */
87struct async_submit_bio {
88 struct inode *inode;
89 struct bio *bio;
90 struct list_head list;
91 extent_submit_bio_hook_t *submit_bio_start;
92 extent_submit_bio_hook_t *submit_bio_done;
93 int rw;
94 int mirror_num;
95 unsigned long bio_flags;
96 /*
97 * bio_offset is optional, can be used if the pages in the bio
98 * can't tell us where in the file the bio should go
99 */
100 u64 bio_offset;
101 struct btrfs_work work;
102 int error;
103};
104
105/*
106 * Lockdep class keys for extent_buffer->lock's in this root. For a given
107 * eb, the lockdep key is determined by the btrfs_root it belongs to and
108 * the level the eb occupies in the tree.
109 *
110 * Different roots are used for different purposes and may nest inside each
111 * other and they require separate keysets. As lockdep keys should be
112 * static, assign keysets according to the purpose of the root as indicated
113 * by btrfs_root->objectid. This ensures that all special purpose roots
114 * have separate keysets.
115 *
116 * Lock-nesting across peer nodes is always done with the immediate parent
117 * node locked thus preventing deadlock. As lockdep doesn't know this, use
118 * subclass to avoid triggering lockdep warning in such cases.
119 *
120 * The key is set by the readpage_end_io_hook after the buffer has passed
121 * csum validation but before the pages are unlocked. It is also set by
122 * btrfs_init_new_buffer on freshly allocated blocks.
123 *
124 * We also add a check to make sure the highest level of the tree is the
125 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
126 * needs update as well.
127 */
128#ifdef CONFIG_DEBUG_LOCK_ALLOC
129# if BTRFS_MAX_LEVEL != 8
130# error
131# endif
132
133static struct btrfs_lockdep_keyset {
134 u64 id; /* root objectid */
135 const char *name_stem; /* lock name stem */
136 char names[BTRFS_MAX_LEVEL + 1][20];
137 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
138} btrfs_lockdep_keysets[] = {
139 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
140 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
141 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
142 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
143 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
144 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
145 { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
146 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
147 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
148 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
149 { .id = 0, .name_stem = "tree" },
150};
151
152void __init btrfs_init_lockdep(void)
153{
154 int i, j;
155
156 /* initialize lockdep class names */
157 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
158 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
159
160 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
161 snprintf(ks->names[j], sizeof(ks->names[j]),
162 "btrfs-%s-%02d", ks->name_stem, j);
163 }
164}
165
166void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
167 int level)
168{
169 struct btrfs_lockdep_keyset *ks;
170
171 BUG_ON(level >= ARRAY_SIZE(ks->keys));
172
173 /* find the matching keyset, id 0 is the default entry */
174 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
175 if (ks->id == objectid)
176 break;
177
178 lockdep_set_class_and_name(&eb->lock,
179 &ks->keys[level], ks->names[level]);
180}
181
182#endif
183
184/*
185 * extents on the btree inode are pretty simple, there's one extent
186 * that covers the entire device
187 */
188static struct extent_map *btree_get_extent(struct inode *inode,
189 struct page *page, size_t pg_offset, u64 start, u64 len,
190 int create)
191{
192 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
193 struct extent_map *em;
194 int ret;
195
196 read_lock(&em_tree->lock);
197 em = lookup_extent_mapping(em_tree, start, len);
198 if (em) {
199 em->bdev =
200 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
201 read_unlock(&em_tree->lock);
202 goto out;
203 }
204 read_unlock(&em_tree->lock);
205
206 em = alloc_extent_map();
207 if (!em) {
208 em = ERR_PTR(-ENOMEM);
209 goto out;
210 }
211 em->start = 0;
212 em->len = (u64)-1;
213 em->block_len = (u64)-1;
214 em->block_start = 0;
215 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
216
217 write_lock(&em_tree->lock);
218 ret = add_extent_mapping(em_tree, em);
219 if (ret == -EEXIST) {
220 u64 failed_start = em->start;
221 u64 failed_len = em->len;
222
223 free_extent_map(em);
224 em = lookup_extent_mapping(em_tree, start, len);
225 if (em) {
226 ret = 0;
227 } else {
228 em = lookup_extent_mapping(em_tree, failed_start,
229 failed_len);
230 ret = -EIO;
231 }
232 } else if (ret) {
233 free_extent_map(em);
234 em = NULL;
235 }
236 write_unlock(&em_tree->lock);
237
238 if (ret)
239 em = ERR_PTR(ret);
240out:
241 return em;
242}
243
244u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
245{
246 return crc32c(seed, data, len);
247}
248
249void btrfs_csum_final(u32 crc, char *result)
250{
251 put_unaligned_le32(~crc, result);
252}
253
254/*
255 * compute the csum for a btree block, and either verify it or write it
256 * into the csum field of the block.
257 */
258static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
259 int verify)
260{
261 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
262 char *result = NULL;
263 unsigned long len;
264 unsigned long cur_len;
265 unsigned long offset = BTRFS_CSUM_SIZE;
266 char *kaddr;
267 unsigned long map_start;
268 unsigned long map_len;
269 int err;
270 u32 crc = ~(u32)0;
271 unsigned long inline_result;
272
273 len = buf->len - offset;
274 while (len > 0) {
275 err = map_private_extent_buffer(buf, offset, 32,
276 &kaddr, &map_start, &map_len);
277 if (err)
278 return 1;
279 cur_len = min(len, map_len - (offset - map_start));
280 crc = btrfs_csum_data(root, kaddr + offset - map_start,
281 crc, cur_len);
282 len -= cur_len;
283 offset += cur_len;
284 }
285 if (csum_size > sizeof(inline_result)) {
286 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
287 if (!result)
288 return 1;
289 } else {
290 result = (char *)&inline_result;
291 }
292
293 btrfs_csum_final(crc, result);
294
295 if (verify) {
296 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
297 u32 val;
298 u32 found = 0;
299 memcpy(&found, result, csum_size);
300
301 read_extent_buffer(buf, &val, 0, csum_size);
302 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
303 "failed on %llu wanted %X found %X "
304 "level %d\n",
305 root->fs_info->sb->s_id,
306 (unsigned long long)buf->start, val, found,
307 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
333 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
334 return 0;
335
336 if (atomic)
337 return -EAGAIN;
338
339 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
340 0, &cached_state);
341 if (extent_buffer_uptodate(eb) &&
342 btrfs_header_generation(eb) == parent_transid) {
343 ret = 0;
344 goto out;
345 }
346 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
347 "found %llu\n",
348 (unsigned long long)eb->start,
349 (unsigned long long)parent_transid,
350 (unsigned long long)btrfs_header_generation(eb));
351 ret = 1;
352 clear_extent_buffer_uptodate(eb);
353out:
354 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
355 &cached_state, GFP_NOFS);
356 return ret;
357}
358
359/*
360 * helper to read a given tree block, doing retries as required when
361 * the checksums don't match and we have alternate mirrors to try.
362 */
363static int btree_read_extent_buffer_pages(struct btrfs_root *root,
364 struct extent_buffer *eb,
365 u64 start, u64 parent_transid)
366{
367 struct extent_io_tree *io_tree;
368 int failed = 0;
369 int ret;
370 int num_copies = 0;
371 int mirror_num = 0;
372 int failed_mirror = 0;
373
374 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
375 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
376 while (1) {
377 ret = read_extent_buffer_pages(io_tree, eb, start,
378 WAIT_COMPLETE,
379 btree_get_extent, mirror_num);
380 if (!ret && !verify_parent_transid(io_tree, eb,
381 parent_transid, 0))
382 break;
383
384 /*
385 * This buffer's crc is fine, but its contents are corrupted, so
386 * there is no reason to read the other copies, they won't be
387 * any less wrong.
388 */
389 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
390 break;
391
392 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
393 eb->start, eb->len);
394 if (num_copies == 1)
395 break;
396
397 if (!failed_mirror) {
398 failed = 1;
399 failed_mirror = eb->read_mirror;
400 }
401
402 mirror_num++;
403 if (mirror_num == failed_mirror)
404 mirror_num++;
405
406 if (mirror_num > num_copies)
407 break;
408 }
409
410 if (failed && !ret)
411 repair_eb_io_failure(root, eb, failed_mirror);
412
413 return ret;
414}
415
416/*
417 * checksum a dirty tree block before IO. This has extra checks to make sure
418 * we only fill in the checksum field in the first page of a multi-page block
419 */
420
421static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
422{
423 struct extent_io_tree *tree;
424 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
425 u64 found_start;
426 struct extent_buffer *eb;
427
428 tree = &BTRFS_I(page->mapping->host)->io_tree;
429
430 eb = (struct extent_buffer *)page->private;
431 if (page != eb->pages[0])
432 return 0;
433 found_start = btrfs_header_bytenr(eb);
434 if (found_start != start) {
435 WARN_ON(1);
436 return 0;
437 }
438 if (eb->pages[0] != page) {
439 WARN_ON(1);
440 return 0;
441 }
442 if (!PageUptodate(page)) {
443 WARN_ON(1);
444 return 0;
445 }
446 csum_tree_block(root, eb, 0);
447 return 0;
448}
449
450static int check_tree_block_fsid(struct btrfs_root *root,
451 struct extent_buffer *eb)
452{
453 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
454 u8 fsid[BTRFS_UUID_SIZE];
455 int ret = 1;
456
457 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
458 BTRFS_FSID_SIZE);
459 while (fs_devices) {
460 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
461 ret = 0;
462 break;
463 }
464 fs_devices = fs_devices->seed;
465 }
466 return ret;
467}
468
469#define CORRUPT(reason, eb, root, slot) \
470 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
471 "root=%llu, slot=%d\n", reason, \
472 (unsigned long long)btrfs_header_bytenr(eb), \
473 (unsigned long long)root->objectid, slot)
474
475static noinline int check_leaf(struct btrfs_root *root,
476 struct extent_buffer *leaf)
477{
478 struct btrfs_key key;
479 struct btrfs_key leaf_key;
480 u32 nritems = btrfs_header_nritems(leaf);
481 int slot;
482
483 if (nritems == 0)
484 return 0;
485
486 /* Check the 0 item */
487 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
488 BTRFS_LEAF_DATA_SIZE(root)) {
489 CORRUPT("invalid item offset size pair", leaf, root, 0);
490 return -EIO;
491 }
492
493 /*
494 * Check to make sure each items keys are in the correct order and their
495 * offsets make sense. We only have to loop through nritems-1 because
496 * we check the current slot against the next slot, which verifies the
497 * next slot's offset+size makes sense and that the current's slot
498 * offset is correct.
499 */
500 for (slot = 0; slot < nritems - 1; slot++) {
501 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
502 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
503
504 /* Make sure the keys are in the right order */
505 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
506 CORRUPT("bad key order", leaf, root, slot);
507 return -EIO;
508 }
509
510 /*
511 * Make sure the offset and ends are right, remember that the
512 * item data starts at the end of the leaf and grows towards the
513 * front.
514 */
515 if (btrfs_item_offset_nr(leaf, slot) !=
516 btrfs_item_end_nr(leaf, slot + 1)) {
517 CORRUPT("slot offset bad", leaf, root, slot);
518 return -EIO;
519 }
520
521 /*
522 * Check to make sure that we don't point outside of the leaf,
523 * just incase all the items are consistent to eachother, but
524 * all point outside of the leaf.
525 */
526 if (btrfs_item_end_nr(leaf, slot) >
527 BTRFS_LEAF_DATA_SIZE(root)) {
528 CORRUPT("slot end outside of leaf", leaf, root, slot);
529 return -EIO;
530 }
531 }
532
533 return 0;
534}
535
536struct extent_buffer *find_eb_for_page(struct extent_io_tree *tree,
537 struct page *page, int max_walk)
538{
539 struct extent_buffer *eb;
540 u64 start = page_offset(page);
541 u64 target = start;
542 u64 min_start;
543
544 if (start < max_walk)
545 min_start = 0;
546 else
547 min_start = start - max_walk;
548
549 while (start >= min_start) {
550 eb = find_extent_buffer(tree, start, 0);
551 if (eb) {
552 /*
553 * we found an extent buffer and it contains our page
554 * horray!
555 */
556 if (eb->start <= target &&
557 eb->start + eb->len > target)
558 return eb;
559
560 /* we found an extent buffer that wasn't for us */
561 free_extent_buffer(eb);
562 return NULL;
563 }
564 if (start == 0)
565 break;
566 start -= PAGE_CACHE_SIZE;
567 }
568 return NULL;
569}
570
571static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
572 struct extent_state *state, int mirror)
573{
574 struct extent_io_tree *tree;
575 u64 found_start;
576 int found_level;
577 struct extent_buffer *eb;
578 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
579 int ret = 0;
580 int reads_done;
581
582 if (!page->private)
583 goto out;
584
585 tree = &BTRFS_I(page->mapping->host)->io_tree;
586 eb = (struct extent_buffer *)page->private;
587
588 /* the pending IO might have been the only thing that kept this buffer
589 * in memory. Make sure we have a ref for all this other checks
590 */
591 extent_buffer_get(eb);
592
593 reads_done = atomic_dec_and_test(&eb->io_pages);
594 if (!reads_done)
595 goto err;
596
597 eb->read_mirror = mirror;
598 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
599 ret = -EIO;
600 goto err;
601 }
602
603 found_start = btrfs_header_bytenr(eb);
604 if (found_start != eb->start) {
605 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
606 "%llu %llu\n",
607 (unsigned long long)found_start,
608 (unsigned long long)eb->start);
609 ret = -EIO;
610 goto err;
611 }
612 if (check_tree_block_fsid(root, eb)) {
613 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
614 (unsigned long long)eb->start);
615 ret = -EIO;
616 goto err;
617 }
618 found_level = btrfs_header_level(eb);
619
620 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
621 eb, found_level);
622
623 ret = csum_tree_block(root, eb, 1);
624 if (ret) {
625 ret = -EIO;
626 goto err;
627 }
628
629 /*
630 * If this is a leaf block and it is corrupt, set the corrupt bit so
631 * that we don't try and read the other copies of this block, just
632 * return -EIO.
633 */
634 if (found_level == 0 && check_leaf(root, eb)) {
635 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
636 ret = -EIO;
637 }
638
639 if (!ret)
640 set_extent_buffer_uptodate(eb);
641err:
642 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
643 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
644 btree_readahead_hook(root, eb, eb->start, ret);
645 }
646
647 if (ret)
648 clear_extent_buffer_uptodate(eb);
649 free_extent_buffer(eb);
650out:
651 return ret;
652}
653
654static int btree_io_failed_hook(struct page *page, int failed_mirror)
655{
656 struct extent_buffer *eb;
657 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
658
659 eb = (struct extent_buffer *)page->private;
660 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
661 eb->read_mirror = failed_mirror;
662 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
663 btree_readahead_hook(root, eb, eb->start, -EIO);
664 return -EIO; /* we fixed nothing */
665}
666
667static void end_workqueue_bio(struct bio *bio, int err)
668{
669 struct end_io_wq *end_io_wq = bio->bi_private;
670 struct btrfs_fs_info *fs_info;
671
672 fs_info = end_io_wq->info;
673 end_io_wq->error = err;
674 end_io_wq->work.func = end_workqueue_fn;
675 end_io_wq->work.flags = 0;
676
677 if (bio->bi_rw & REQ_WRITE) {
678 if (end_io_wq->metadata == 1)
679 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
680 &end_io_wq->work);
681 else if (end_io_wq->metadata == 2)
682 btrfs_queue_worker(&fs_info->endio_freespace_worker,
683 &end_io_wq->work);
684 else
685 btrfs_queue_worker(&fs_info->endio_write_workers,
686 &end_io_wq->work);
687 } else {
688 if (end_io_wq->metadata)
689 btrfs_queue_worker(&fs_info->endio_meta_workers,
690 &end_io_wq->work);
691 else
692 btrfs_queue_worker(&fs_info->endio_workers,
693 &end_io_wq->work);
694 }
695}
696
697/*
698 * For the metadata arg you want
699 *
700 * 0 - if data
701 * 1 - if normal metadta
702 * 2 - if writing to the free space cache area
703 */
704int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
705 int metadata)
706{
707 struct end_io_wq *end_io_wq;
708 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
709 if (!end_io_wq)
710 return -ENOMEM;
711
712 end_io_wq->private = bio->bi_private;
713 end_io_wq->end_io = bio->bi_end_io;
714 end_io_wq->info = info;
715 end_io_wq->error = 0;
716 end_io_wq->bio = bio;
717 end_io_wq->metadata = metadata;
718
719 bio->bi_private = end_io_wq;
720 bio->bi_end_io = end_workqueue_bio;
721 return 0;
722}
723
724unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
725{
726 unsigned long limit = min_t(unsigned long,
727 info->workers.max_workers,
728 info->fs_devices->open_devices);
729 return 256 * limit;
730}
731
732static void run_one_async_start(struct btrfs_work *work)
733{
734 struct async_submit_bio *async;
735 int ret;
736
737 async = container_of(work, struct async_submit_bio, work);
738 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
739 async->mirror_num, async->bio_flags,
740 async->bio_offset);
741 if (ret)
742 async->error = ret;
743}
744
745static void run_one_async_done(struct btrfs_work *work)
746{
747 struct btrfs_fs_info *fs_info;
748 struct async_submit_bio *async;
749 int limit;
750
751 async = container_of(work, struct async_submit_bio, work);
752 fs_info = BTRFS_I(async->inode)->root->fs_info;
753
754 limit = btrfs_async_submit_limit(fs_info);
755 limit = limit * 2 / 3;
756
757 atomic_dec(&fs_info->nr_async_submits);
758
759 if (atomic_read(&fs_info->nr_async_submits) < limit &&
760 waitqueue_active(&fs_info->async_submit_wait))
761 wake_up(&fs_info->async_submit_wait);
762
763 /* If an error occured we just want to clean up the bio and move on */
764 if (async->error) {
765 bio_endio(async->bio, async->error);
766 return;
767 }
768
769 async->submit_bio_done(async->inode, async->rw, async->bio,
770 async->mirror_num, async->bio_flags,
771 async->bio_offset);
772}
773
774static void run_one_async_free(struct btrfs_work *work)
775{
776 struct async_submit_bio *async;
777
778 async = container_of(work, struct async_submit_bio, work);
779 kfree(async);
780}
781
782int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
783 int rw, struct bio *bio, int mirror_num,
784 unsigned long bio_flags,
785 u64 bio_offset,
786 extent_submit_bio_hook_t *submit_bio_start,
787 extent_submit_bio_hook_t *submit_bio_done)
788{
789 struct async_submit_bio *async;
790
791 async = kmalloc(sizeof(*async), GFP_NOFS);
792 if (!async)
793 return -ENOMEM;
794
795 async->inode = inode;
796 async->rw = rw;
797 async->bio = bio;
798 async->mirror_num = mirror_num;
799 async->submit_bio_start = submit_bio_start;
800 async->submit_bio_done = submit_bio_done;
801
802 async->work.func = run_one_async_start;
803 async->work.ordered_func = run_one_async_done;
804 async->work.ordered_free = run_one_async_free;
805
806 async->work.flags = 0;
807 async->bio_flags = bio_flags;
808 async->bio_offset = bio_offset;
809
810 async->error = 0;
811
812 atomic_inc(&fs_info->nr_async_submits);
813
814 if (rw & REQ_SYNC)
815 btrfs_set_work_high_prio(&async->work);
816
817 btrfs_queue_worker(&fs_info->workers, &async->work);
818
819 while (atomic_read(&fs_info->async_submit_draining) &&
820 atomic_read(&fs_info->nr_async_submits)) {
821 wait_event(fs_info->async_submit_wait,
822 (atomic_read(&fs_info->nr_async_submits) == 0));
823 }
824
825 return 0;
826}
827
828static int btree_csum_one_bio(struct bio *bio)
829{
830 struct bio_vec *bvec = bio->bi_io_vec;
831 int bio_index = 0;
832 struct btrfs_root *root;
833 int ret = 0;
834
835 WARN_ON(bio->bi_vcnt <= 0);
836 while (bio_index < bio->bi_vcnt) {
837 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
838 ret = csum_dirty_buffer(root, bvec->bv_page);
839 if (ret)
840 break;
841 bio_index++;
842 bvec++;
843 }
844 return ret;
845}
846
847static int __btree_submit_bio_start(struct inode *inode, int rw,
848 struct bio *bio, int mirror_num,
849 unsigned long bio_flags,
850 u64 bio_offset)
851{
852 /*
853 * when we're called for a write, we're already in the async
854 * submission context. Just jump into btrfs_map_bio
855 */
856 return btree_csum_one_bio(bio);
857}
858
859static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
860 int mirror_num, unsigned long bio_flags,
861 u64 bio_offset)
862{
863 /*
864 * when we're called for a write, we're already in the async
865 * submission context. Just jump into btrfs_map_bio
866 */
867 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
868}
869
870static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
871 int mirror_num, unsigned long bio_flags,
872 u64 bio_offset)
873{
874 int ret;
875
876 if (!(rw & REQ_WRITE)) {
877
878 /*
879 * called for a read, do the setup so that checksum validation
880 * can happen in the async kernel threads
881 */
882 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
883 bio, 1);
884 if (ret)
885 return ret;
886 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
887 mirror_num, 0);
888 }
889
890 /*
891 * kthread helpers are used to submit writes so that checksumming
892 * can happen in parallel across all CPUs
893 */
894 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
895 inode, rw, bio, mirror_num, 0,
896 bio_offset,
897 __btree_submit_bio_start,
898 __btree_submit_bio_done);
899}
900
901#ifdef CONFIG_MIGRATION
902static int btree_migratepage(struct address_space *mapping,
903 struct page *newpage, struct page *page,
904 enum migrate_mode mode)
905{
906 /*
907 * we can't safely write a btree page from here,
908 * we haven't done the locking hook
909 */
910 if (PageDirty(page))
911 return -EAGAIN;
912 /*
913 * Buffers may be managed in a filesystem specific way.
914 * We must have no buffers or drop them.
915 */
916 if (page_has_private(page) &&
917 !try_to_release_page(page, GFP_KERNEL))
918 return -EAGAIN;
919 return migrate_page(mapping, newpage, page, mode);
920}
921#endif
922
923
924static int btree_writepages(struct address_space *mapping,
925 struct writeback_control *wbc)
926{
927 struct extent_io_tree *tree;
928 tree = &BTRFS_I(mapping->host)->io_tree;
929 if (wbc->sync_mode == WB_SYNC_NONE) {
930 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
931 u64 num_dirty;
932 unsigned long thresh = 32 * 1024 * 1024;
933
934 if (wbc->for_kupdate)
935 return 0;
936
937 /* this is a bit racy, but that's ok */
938 num_dirty = root->fs_info->dirty_metadata_bytes;
939 if (num_dirty < thresh)
940 return 0;
941 }
942 return btree_write_cache_pages(mapping, wbc);
943}
944
945static int btree_readpage(struct file *file, struct page *page)
946{
947 struct extent_io_tree *tree;
948 tree = &BTRFS_I(page->mapping->host)->io_tree;
949 return extent_read_full_page(tree, page, btree_get_extent, 0);
950}
951
952static int btree_releasepage(struct page *page, gfp_t gfp_flags)
953{
954 if (PageWriteback(page) || PageDirty(page))
955 return 0;
956 /*
957 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
958 * slab allocation from alloc_extent_state down the callchain where
959 * it'd hit a BUG_ON as those flags are not allowed.
960 */
961 gfp_flags &= ~GFP_SLAB_BUG_MASK;
962
963 return try_release_extent_buffer(page, gfp_flags);
964}
965
966static void btree_invalidatepage(struct page *page, unsigned long offset)
967{
968 struct extent_io_tree *tree;
969 tree = &BTRFS_I(page->mapping->host)->io_tree;
970 extent_invalidatepage(tree, page, offset);
971 btree_releasepage(page, GFP_NOFS);
972 if (PagePrivate(page)) {
973 printk(KERN_WARNING "btrfs warning page private not zero "
974 "on page %llu\n", (unsigned long long)page_offset(page));
975 ClearPagePrivate(page);
976 set_page_private(page, 0);
977 page_cache_release(page);
978 }
979}
980
981static int btree_set_page_dirty(struct page *page)
982{
983 struct extent_buffer *eb;
984
985 BUG_ON(!PagePrivate(page));
986 eb = (struct extent_buffer *)page->private;
987 BUG_ON(!eb);
988 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
989 BUG_ON(!atomic_read(&eb->refs));
990 btrfs_assert_tree_locked(eb);
991 return __set_page_dirty_nobuffers(page);
992}
993
994static const struct address_space_operations btree_aops = {
995 .readpage = btree_readpage,
996 .writepages = btree_writepages,
997 .releasepage = btree_releasepage,
998 .invalidatepage = btree_invalidatepage,
999#ifdef CONFIG_MIGRATION
1000 .migratepage = btree_migratepage,
1001#endif
1002 .set_page_dirty = btree_set_page_dirty,
1003};
1004
1005int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1006 u64 parent_transid)
1007{
1008 struct extent_buffer *buf = NULL;
1009 struct inode *btree_inode = root->fs_info->btree_inode;
1010 int ret = 0;
1011
1012 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1013 if (!buf)
1014 return 0;
1015 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1016 buf, 0, WAIT_NONE, btree_get_extent, 0);
1017 free_extent_buffer(buf);
1018 return ret;
1019}
1020
1021int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1022 int mirror_num, struct extent_buffer **eb)
1023{
1024 struct extent_buffer *buf = NULL;
1025 struct inode *btree_inode = root->fs_info->btree_inode;
1026 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1027 int ret;
1028
1029 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1030 if (!buf)
1031 return 0;
1032
1033 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1034
1035 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1036 btree_get_extent, mirror_num);
1037 if (ret) {
1038 free_extent_buffer(buf);
1039 return ret;
1040 }
1041
1042 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1043 free_extent_buffer(buf);
1044 return -EIO;
1045 } else if (extent_buffer_uptodate(buf)) {
1046 *eb = buf;
1047 } else {
1048 free_extent_buffer(buf);
1049 }
1050 return 0;
1051}
1052
1053struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1054 u64 bytenr, u32 blocksize)
1055{
1056 struct inode *btree_inode = root->fs_info->btree_inode;
1057 struct extent_buffer *eb;
1058 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1059 bytenr, blocksize);
1060 return eb;
1061}
1062
1063struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1064 u64 bytenr, u32 blocksize)
1065{
1066 struct inode *btree_inode = root->fs_info->btree_inode;
1067 struct extent_buffer *eb;
1068
1069 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1070 bytenr, blocksize);
1071 return eb;
1072}
1073
1074
1075int btrfs_write_tree_block(struct extent_buffer *buf)
1076{
1077 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1078 buf->start + buf->len - 1);
1079}
1080
1081int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1082{
1083 return filemap_fdatawait_range(buf->pages[0]->mapping,
1084 buf->start, buf->start + buf->len - 1);
1085}
1086
1087struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1088 u32 blocksize, u64 parent_transid)
1089{
1090 struct extent_buffer *buf = NULL;
1091 int ret;
1092
1093 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1094 if (!buf)
1095 return NULL;
1096
1097 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1098 return buf;
1099
1100}
1101
1102void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1103 struct extent_buffer *buf)
1104{
1105 if (btrfs_header_generation(buf) ==
1106 root->fs_info->running_transaction->transid) {
1107 btrfs_assert_tree_locked(buf);
1108
1109 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1110 spin_lock(&root->fs_info->delalloc_lock);
1111 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1112 root->fs_info->dirty_metadata_bytes -= buf->len;
1113 else {
1114 spin_unlock(&root->fs_info->delalloc_lock);
1115 btrfs_panic(root->fs_info, -EOVERFLOW,
1116 "Can't clear %lu bytes from "
1117 " dirty_mdatadata_bytes (%lu)",
1118 buf->len,
1119 root->fs_info->dirty_metadata_bytes);
1120 }
1121 spin_unlock(&root->fs_info->delalloc_lock);
1122 }
1123
1124 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1125 btrfs_set_lock_blocking(buf);
1126 clear_extent_buffer_dirty(buf);
1127 }
1128}
1129
1130static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1131 u32 stripesize, struct btrfs_root *root,
1132 struct btrfs_fs_info *fs_info,
1133 u64 objectid)
1134{
1135 root->node = NULL;
1136 root->commit_root = NULL;
1137 root->sectorsize = sectorsize;
1138 root->nodesize = nodesize;
1139 root->leafsize = leafsize;
1140 root->stripesize = stripesize;
1141 root->ref_cows = 0;
1142 root->track_dirty = 0;
1143 root->in_radix = 0;
1144 root->orphan_item_inserted = 0;
1145 root->orphan_cleanup_state = 0;
1146
1147 root->objectid = objectid;
1148 root->last_trans = 0;
1149 root->highest_objectid = 0;
1150 root->name = NULL;
1151 root->inode_tree = RB_ROOT;
1152 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1153 root->block_rsv = NULL;
1154 root->orphan_block_rsv = NULL;
1155
1156 INIT_LIST_HEAD(&root->dirty_list);
1157 INIT_LIST_HEAD(&root->root_list);
1158 spin_lock_init(&root->orphan_lock);
1159 spin_lock_init(&root->inode_lock);
1160 spin_lock_init(&root->accounting_lock);
1161 mutex_init(&root->objectid_mutex);
1162 mutex_init(&root->log_mutex);
1163 init_waitqueue_head(&root->log_writer_wait);
1164 init_waitqueue_head(&root->log_commit_wait[0]);
1165 init_waitqueue_head(&root->log_commit_wait[1]);
1166 atomic_set(&root->log_commit[0], 0);
1167 atomic_set(&root->log_commit[1], 0);
1168 atomic_set(&root->log_writers, 0);
1169 atomic_set(&root->orphan_inodes, 0);
1170 root->log_batch = 0;
1171 root->log_transid = 0;
1172 root->last_log_commit = 0;
1173 extent_io_tree_init(&root->dirty_log_pages,
1174 fs_info->btree_inode->i_mapping);
1175
1176 memset(&root->root_key, 0, sizeof(root->root_key));
1177 memset(&root->root_item, 0, sizeof(root->root_item));
1178 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1179 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1180 root->defrag_trans_start = fs_info->generation;
1181 init_completion(&root->kobj_unregister);
1182 root->defrag_running = 0;
1183 root->root_key.objectid = objectid;
1184 root->anon_dev = 0;
1185}
1186
1187static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
1188 struct btrfs_fs_info *fs_info,
1189 u64 objectid,
1190 struct btrfs_root *root)
1191{
1192 int ret;
1193 u32 blocksize;
1194 u64 generation;
1195
1196 __setup_root(tree_root->nodesize, tree_root->leafsize,
1197 tree_root->sectorsize, tree_root->stripesize,
1198 root, fs_info, objectid);
1199 ret = btrfs_find_last_root(tree_root, objectid,
1200 &root->root_item, &root->root_key);
1201 if (ret > 0)
1202 return -ENOENT;
1203 else if (ret < 0)
1204 return ret;
1205
1206 generation = btrfs_root_generation(&root->root_item);
1207 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1208 root->commit_root = NULL;
1209 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1210 blocksize, generation);
1211 if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) {
1212 free_extent_buffer(root->node);
1213 root->node = NULL;
1214 return -EIO;
1215 }
1216 root->commit_root = btrfs_root_node(root);
1217 return 0;
1218}
1219
1220static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1221{
1222 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1223 if (root)
1224 root->fs_info = fs_info;
1225 return root;
1226}
1227
1228static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1229 struct btrfs_fs_info *fs_info)
1230{
1231 struct btrfs_root *root;
1232 struct btrfs_root *tree_root = fs_info->tree_root;
1233 struct extent_buffer *leaf;
1234
1235 root = btrfs_alloc_root(fs_info);
1236 if (!root)
1237 return ERR_PTR(-ENOMEM);
1238
1239 __setup_root(tree_root->nodesize, tree_root->leafsize,
1240 tree_root->sectorsize, tree_root->stripesize,
1241 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1242
1243 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1244 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1245 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1246 /*
1247 * log trees do not get reference counted because they go away
1248 * before a real commit is actually done. They do store pointers
1249 * to file data extents, and those reference counts still get
1250 * updated (along with back refs to the log tree).
1251 */
1252 root->ref_cows = 0;
1253
1254 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1255 BTRFS_TREE_LOG_OBJECTID, NULL,
1256 0, 0, 0);
1257 if (IS_ERR(leaf)) {
1258 kfree(root);
1259 return ERR_CAST(leaf);
1260 }
1261
1262 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1263 btrfs_set_header_bytenr(leaf, leaf->start);
1264 btrfs_set_header_generation(leaf, trans->transid);
1265 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1266 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1267 root->node = leaf;
1268
1269 write_extent_buffer(root->node, root->fs_info->fsid,
1270 (unsigned long)btrfs_header_fsid(root->node),
1271 BTRFS_FSID_SIZE);
1272 btrfs_mark_buffer_dirty(root->node);
1273 btrfs_tree_unlock(root->node);
1274 return root;
1275}
1276
1277int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1278 struct btrfs_fs_info *fs_info)
1279{
1280 struct btrfs_root *log_root;
1281
1282 log_root = alloc_log_tree(trans, fs_info);
1283 if (IS_ERR(log_root))
1284 return PTR_ERR(log_root);
1285 WARN_ON(fs_info->log_root_tree);
1286 fs_info->log_root_tree = log_root;
1287 return 0;
1288}
1289
1290int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1291 struct btrfs_root *root)
1292{
1293 struct btrfs_root *log_root;
1294 struct btrfs_inode_item *inode_item;
1295
1296 log_root = alloc_log_tree(trans, root->fs_info);
1297 if (IS_ERR(log_root))
1298 return PTR_ERR(log_root);
1299
1300 log_root->last_trans = trans->transid;
1301 log_root->root_key.offset = root->root_key.objectid;
1302
1303 inode_item = &log_root->root_item.inode;
1304 inode_item->generation = cpu_to_le64(1);
1305 inode_item->size = cpu_to_le64(3);
1306 inode_item->nlink = cpu_to_le32(1);
1307 inode_item->nbytes = cpu_to_le64(root->leafsize);
1308 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1309
1310 btrfs_set_root_node(&log_root->root_item, log_root->node);
1311
1312 WARN_ON(root->log_root);
1313 root->log_root = log_root;
1314 root->log_transid = 0;
1315 root->last_log_commit = 0;
1316 return 0;
1317}
1318
1319struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1320 struct btrfs_key *location)
1321{
1322 struct btrfs_root *root;
1323 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1324 struct btrfs_path *path;
1325 struct extent_buffer *l;
1326 u64 generation;
1327 u32 blocksize;
1328 int ret = 0;
1329
1330 root = btrfs_alloc_root(fs_info);
1331 if (!root)
1332 return ERR_PTR(-ENOMEM);
1333 if (location->offset == (u64)-1) {
1334 ret = find_and_setup_root(tree_root, fs_info,
1335 location->objectid, root);
1336 if (ret) {
1337 kfree(root);
1338 return ERR_PTR(ret);
1339 }
1340 goto out;
1341 }
1342
1343 __setup_root(tree_root->nodesize, tree_root->leafsize,
1344 tree_root->sectorsize, tree_root->stripesize,
1345 root, fs_info, location->objectid);
1346
1347 path = btrfs_alloc_path();
1348 if (!path) {
1349 kfree(root);
1350 return ERR_PTR(-ENOMEM);
1351 }
1352 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1353 if (ret == 0) {
1354 l = path->nodes[0];
1355 read_extent_buffer(l, &root->root_item,
1356 btrfs_item_ptr_offset(l, path->slots[0]),
1357 sizeof(root->root_item));
1358 memcpy(&root->root_key, location, sizeof(*location));
1359 }
1360 btrfs_free_path(path);
1361 if (ret) {
1362 kfree(root);
1363 if (ret > 0)
1364 ret = -ENOENT;
1365 return ERR_PTR(ret);
1366 }
1367
1368 generation = btrfs_root_generation(&root->root_item);
1369 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1370 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1371 blocksize, generation);
1372 root->commit_root = btrfs_root_node(root);
1373 BUG_ON(!root->node); /* -ENOMEM */
1374out:
1375 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1376 root->ref_cows = 1;
1377 btrfs_check_and_init_root_item(&root->root_item);
1378 }
1379
1380 return root;
1381}
1382
1383struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1384 struct btrfs_key *location)
1385{
1386 struct btrfs_root *root;
1387 int ret;
1388
1389 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1390 return fs_info->tree_root;
1391 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1392 return fs_info->extent_root;
1393 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1394 return fs_info->chunk_root;
1395 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1396 return fs_info->dev_root;
1397 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1398 return fs_info->csum_root;
1399again:
1400 spin_lock(&fs_info->fs_roots_radix_lock);
1401 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1402 (unsigned long)location->objectid);
1403 spin_unlock(&fs_info->fs_roots_radix_lock);
1404 if (root)
1405 return root;
1406
1407 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1408 if (IS_ERR(root))
1409 return root;
1410
1411 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1412 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1413 GFP_NOFS);
1414 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1415 ret = -ENOMEM;
1416 goto fail;
1417 }
1418
1419 btrfs_init_free_ino_ctl(root);
1420 mutex_init(&root->fs_commit_mutex);
1421 spin_lock_init(&root->cache_lock);
1422 init_waitqueue_head(&root->cache_wait);
1423
1424 ret = get_anon_bdev(&root->anon_dev);
1425 if (ret)
1426 goto fail;
1427
1428 if (btrfs_root_refs(&root->root_item) == 0) {
1429 ret = -ENOENT;
1430 goto fail;
1431 }
1432
1433 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1434 if (ret < 0)
1435 goto fail;
1436 if (ret == 0)
1437 root->orphan_item_inserted = 1;
1438
1439 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1440 if (ret)
1441 goto fail;
1442
1443 spin_lock(&fs_info->fs_roots_radix_lock);
1444 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1445 (unsigned long)root->root_key.objectid,
1446 root);
1447 if (ret == 0)
1448 root->in_radix = 1;
1449
1450 spin_unlock(&fs_info->fs_roots_radix_lock);
1451 radix_tree_preload_end();
1452 if (ret) {
1453 if (ret == -EEXIST) {
1454 free_fs_root(root);
1455 goto again;
1456 }
1457 goto fail;
1458 }
1459
1460 ret = btrfs_find_dead_roots(fs_info->tree_root,
1461 root->root_key.objectid);
1462 WARN_ON(ret);
1463 return root;
1464fail:
1465 free_fs_root(root);
1466 return ERR_PTR(ret);
1467}
1468
1469static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1470{
1471 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1472 int ret = 0;
1473 struct btrfs_device *device;
1474 struct backing_dev_info *bdi;
1475
1476 rcu_read_lock();
1477 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1478 if (!device->bdev)
1479 continue;
1480 bdi = blk_get_backing_dev_info(device->bdev);
1481 if (bdi && bdi_congested(bdi, bdi_bits)) {
1482 ret = 1;
1483 break;
1484 }
1485 }
1486 rcu_read_unlock();
1487 return ret;
1488}
1489
1490/*
1491 * If this fails, caller must call bdi_destroy() to get rid of the
1492 * bdi again.
1493 */
1494static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1495{
1496 int err;
1497
1498 bdi->capabilities = BDI_CAP_MAP_COPY;
1499 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1500 if (err)
1501 return err;
1502
1503 bdi->ra_pages = default_backing_dev_info.ra_pages;
1504 bdi->congested_fn = btrfs_congested_fn;
1505 bdi->congested_data = info;
1506 return 0;
1507}
1508
1509/*
1510 * called by the kthread helper functions to finally call the bio end_io
1511 * functions. This is where read checksum verification actually happens
1512 */
1513static void end_workqueue_fn(struct btrfs_work *work)
1514{
1515 struct bio *bio;
1516 struct end_io_wq *end_io_wq;
1517 struct btrfs_fs_info *fs_info;
1518 int error;
1519
1520 end_io_wq = container_of(work, struct end_io_wq, work);
1521 bio = end_io_wq->bio;
1522 fs_info = end_io_wq->info;
1523
1524 error = end_io_wq->error;
1525 bio->bi_private = end_io_wq->private;
1526 bio->bi_end_io = end_io_wq->end_io;
1527 kfree(end_io_wq);
1528 bio_endio(bio, error);
1529}
1530
1531static int cleaner_kthread(void *arg)
1532{
1533 struct btrfs_root *root = arg;
1534
1535 do {
1536 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1537
1538 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1539 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1540 btrfs_run_delayed_iputs(root);
1541 btrfs_clean_old_snapshots(root);
1542 mutex_unlock(&root->fs_info->cleaner_mutex);
1543 btrfs_run_defrag_inodes(root->fs_info);
1544 }
1545
1546 if (!try_to_freeze()) {
1547 set_current_state(TASK_INTERRUPTIBLE);
1548 if (!kthread_should_stop())
1549 schedule();
1550 __set_current_state(TASK_RUNNING);
1551 }
1552 } while (!kthread_should_stop());
1553 return 0;
1554}
1555
1556static int transaction_kthread(void *arg)
1557{
1558 struct btrfs_root *root = arg;
1559 struct btrfs_trans_handle *trans;
1560 struct btrfs_transaction *cur;
1561 u64 transid;
1562 unsigned long now;
1563 unsigned long delay;
1564 bool cannot_commit;
1565
1566 do {
1567 cannot_commit = false;
1568 delay = HZ * 30;
1569 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1570 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1571
1572 spin_lock(&root->fs_info->trans_lock);
1573 cur = root->fs_info->running_transaction;
1574 if (!cur) {
1575 spin_unlock(&root->fs_info->trans_lock);
1576 goto sleep;
1577 }
1578
1579 now = get_seconds();
1580 if (!cur->blocked &&
1581 (now < cur->start_time || now - cur->start_time < 30)) {
1582 spin_unlock(&root->fs_info->trans_lock);
1583 delay = HZ * 5;
1584 goto sleep;
1585 }
1586 transid = cur->transid;
1587 spin_unlock(&root->fs_info->trans_lock);
1588
1589 /* If the file system is aborted, this will always fail. */
1590 trans = btrfs_join_transaction(root);
1591 if (IS_ERR(trans)) {
1592 cannot_commit = true;
1593 goto sleep;
1594 }
1595 if (transid == trans->transid) {
1596 btrfs_commit_transaction(trans, root);
1597 } else {
1598 btrfs_end_transaction(trans, root);
1599 }
1600sleep:
1601 wake_up_process(root->fs_info->cleaner_kthread);
1602 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1603
1604 if (!try_to_freeze()) {
1605 set_current_state(TASK_INTERRUPTIBLE);
1606 if (!kthread_should_stop() &&
1607 (!btrfs_transaction_blocked(root->fs_info) ||
1608 cannot_commit))
1609 schedule_timeout(delay);
1610 __set_current_state(TASK_RUNNING);
1611 }
1612 } while (!kthread_should_stop());
1613 return 0;
1614}
1615
1616/*
1617 * this will find the highest generation in the array of
1618 * root backups. The index of the highest array is returned,
1619 * or -1 if we can't find anything.
1620 *
1621 * We check to make sure the array is valid by comparing the
1622 * generation of the latest root in the array with the generation
1623 * in the super block. If they don't match we pitch it.
1624 */
1625static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1626{
1627 u64 cur;
1628 int newest_index = -1;
1629 struct btrfs_root_backup *root_backup;
1630 int i;
1631
1632 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1633 root_backup = info->super_copy->super_roots + i;
1634 cur = btrfs_backup_tree_root_gen(root_backup);
1635 if (cur == newest_gen)
1636 newest_index = i;
1637 }
1638
1639 /* check to see if we actually wrapped around */
1640 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1641 root_backup = info->super_copy->super_roots;
1642 cur = btrfs_backup_tree_root_gen(root_backup);
1643 if (cur == newest_gen)
1644 newest_index = 0;
1645 }
1646 return newest_index;
1647}
1648
1649
1650/*
1651 * find the oldest backup so we know where to store new entries
1652 * in the backup array. This will set the backup_root_index
1653 * field in the fs_info struct
1654 */
1655static void find_oldest_super_backup(struct btrfs_fs_info *info,
1656 u64 newest_gen)
1657{
1658 int newest_index = -1;
1659
1660 newest_index = find_newest_super_backup(info, newest_gen);
1661 /* if there was garbage in there, just move along */
1662 if (newest_index == -1) {
1663 info->backup_root_index = 0;
1664 } else {
1665 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1666 }
1667}
1668
1669/*
1670 * copy all the root pointers into the super backup array.
1671 * this will bump the backup pointer by one when it is
1672 * done
1673 */
1674static void backup_super_roots(struct btrfs_fs_info *info)
1675{
1676 int next_backup;
1677 struct btrfs_root_backup *root_backup;
1678 int last_backup;
1679
1680 next_backup = info->backup_root_index;
1681 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1682 BTRFS_NUM_BACKUP_ROOTS;
1683
1684 /*
1685 * just overwrite the last backup if we're at the same generation
1686 * this happens only at umount
1687 */
1688 root_backup = info->super_for_commit->super_roots + last_backup;
1689 if (btrfs_backup_tree_root_gen(root_backup) ==
1690 btrfs_header_generation(info->tree_root->node))
1691 next_backup = last_backup;
1692
1693 root_backup = info->super_for_commit->super_roots + next_backup;
1694
1695 /*
1696 * make sure all of our padding and empty slots get zero filled
1697 * regardless of which ones we use today
1698 */
1699 memset(root_backup, 0, sizeof(*root_backup));
1700
1701 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1702
1703 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1704 btrfs_set_backup_tree_root_gen(root_backup,
1705 btrfs_header_generation(info->tree_root->node));
1706
1707 btrfs_set_backup_tree_root_level(root_backup,
1708 btrfs_header_level(info->tree_root->node));
1709
1710 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1711 btrfs_set_backup_chunk_root_gen(root_backup,
1712 btrfs_header_generation(info->chunk_root->node));
1713 btrfs_set_backup_chunk_root_level(root_backup,
1714 btrfs_header_level(info->chunk_root->node));
1715
1716 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1717 btrfs_set_backup_extent_root_gen(root_backup,
1718 btrfs_header_generation(info->extent_root->node));
1719 btrfs_set_backup_extent_root_level(root_backup,
1720 btrfs_header_level(info->extent_root->node));
1721
1722 /*
1723 * we might commit during log recovery, which happens before we set
1724 * the fs_root. Make sure it is valid before we fill it in.
1725 */
1726 if (info->fs_root && info->fs_root->node) {
1727 btrfs_set_backup_fs_root(root_backup,
1728 info->fs_root->node->start);
1729 btrfs_set_backup_fs_root_gen(root_backup,
1730 btrfs_header_generation(info->fs_root->node));
1731 btrfs_set_backup_fs_root_level(root_backup,
1732 btrfs_header_level(info->fs_root->node));
1733 }
1734
1735 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1736 btrfs_set_backup_dev_root_gen(root_backup,
1737 btrfs_header_generation(info->dev_root->node));
1738 btrfs_set_backup_dev_root_level(root_backup,
1739 btrfs_header_level(info->dev_root->node));
1740
1741 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1742 btrfs_set_backup_csum_root_gen(root_backup,
1743 btrfs_header_generation(info->csum_root->node));
1744 btrfs_set_backup_csum_root_level(root_backup,
1745 btrfs_header_level(info->csum_root->node));
1746
1747 btrfs_set_backup_total_bytes(root_backup,
1748 btrfs_super_total_bytes(info->super_copy));
1749 btrfs_set_backup_bytes_used(root_backup,
1750 btrfs_super_bytes_used(info->super_copy));
1751 btrfs_set_backup_num_devices(root_backup,
1752 btrfs_super_num_devices(info->super_copy));
1753
1754 /*
1755 * if we don't copy this out to the super_copy, it won't get remembered
1756 * for the next commit
1757 */
1758 memcpy(&info->super_copy->super_roots,
1759 &info->super_for_commit->super_roots,
1760 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1761}
1762
1763/*
1764 * this copies info out of the root backup array and back into
1765 * the in-memory super block. It is meant to help iterate through
1766 * the array, so you send it the number of backups you've already
1767 * tried and the last backup index you used.
1768 *
1769 * this returns -1 when it has tried all the backups
1770 */
1771static noinline int next_root_backup(struct btrfs_fs_info *info,
1772 struct btrfs_super_block *super,
1773 int *num_backups_tried, int *backup_index)
1774{
1775 struct btrfs_root_backup *root_backup;
1776 int newest = *backup_index;
1777
1778 if (*num_backups_tried == 0) {
1779 u64 gen = btrfs_super_generation(super);
1780
1781 newest = find_newest_super_backup(info, gen);
1782 if (newest == -1)
1783 return -1;
1784
1785 *backup_index = newest;
1786 *num_backups_tried = 1;
1787 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1788 /* we've tried all the backups, all done */
1789 return -1;
1790 } else {
1791 /* jump to the next oldest backup */
1792 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1793 BTRFS_NUM_BACKUP_ROOTS;
1794 *backup_index = newest;
1795 *num_backups_tried += 1;
1796 }
1797 root_backup = super->super_roots + newest;
1798
1799 btrfs_set_super_generation(super,
1800 btrfs_backup_tree_root_gen(root_backup));
1801 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1802 btrfs_set_super_root_level(super,
1803 btrfs_backup_tree_root_level(root_backup));
1804 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1805
1806 /*
1807 * fixme: the total bytes and num_devices need to match or we should
1808 * need a fsck
1809 */
1810 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1811 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1812 return 0;
1813}
1814
1815/* helper to cleanup tree roots */
1816static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1817{
1818 free_extent_buffer(info->tree_root->node);
1819 free_extent_buffer(info->tree_root->commit_root);
1820 free_extent_buffer(info->dev_root->node);
1821 free_extent_buffer(info->dev_root->commit_root);
1822 free_extent_buffer(info->extent_root->node);
1823 free_extent_buffer(info->extent_root->commit_root);
1824 free_extent_buffer(info->csum_root->node);
1825 free_extent_buffer(info->csum_root->commit_root);
1826
1827 info->tree_root->node = NULL;
1828 info->tree_root->commit_root = NULL;
1829 info->dev_root->node = NULL;
1830 info->dev_root->commit_root = NULL;
1831 info->extent_root->node = NULL;
1832 info->extent_root->commit_root = NULL;
1833 info->csum_root->node = NULL;
1834 info->csum_root->commit_root = NULL;
1835
1836 if (chunk_root) {
1837 free_extent_buffer(info->chunk_root->node);
1838 free_extent_buffer(info->chunk_root->commit_root);
1839 info->chunk_root->node = NULL;
1840 info->chunk_root->commit_root = NULL;
1841 }
1842}
1843
1844
1845int open_ctree(struct super_block *sb,
1846 struct btrfs_fs_devices *fs_devices,
1847 char *options)
1848{
1849 u32 sectorsize;
1850 u32 nodesize;
1851 u32 leafsize;
1852 u32 blocksize;
1853 u32 stripesize;
1854 u64 generation;
1855 u64 features;
1856 struct btrfs_key location;
1857 struct buffer_head *bh;
1858 struct btrfs_super_block *disk_super;
1859 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1860 struct btrfs_root *tree_root;
1861 struct btrfs_root *extent_root;
1862 struct btrfs_root *csum_root;
1863 struct btrfs_root *chunk_root;
1864 struct btrfs_root *dev_root;
1865 struct btrfs_root *log_tree_root;
1866 int ret;
1867 int err = -EINVAL;
1868 int num_backups_tried = 0;
1869 int backup_index = 0;
1870
1871 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
1872 extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
1873 csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
1874 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
1875 dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
1876
1877 if (!tree_root || !extent_root || !csum_root ||
1878 !chunk_root || !dev_root) {
1879 err = -ENOMEM;
1880 goto fail;
1881 }
1882
1883 ret = init_srcu_struct(&fs_info->subvol_srcu);
1884 if (ret) {
1885 err = ret;
1886 goto fail;
1887 }
1888
1889 ret = setup_bdi(fs_info, &fs_info->bdi);
1890 if (ret) {
1891 err = ret;
1892 goto fail_srcu;
1893 }
1894
1895 fs_info->btree_inode = new_inode(sb);
1896 if (!fs_info->btree_inode) {
1897 err = -ENOMEM;
1898 goto fail_bdi;
1899 }
1900
1901 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1902
1903 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1904 INIT_LIST_HEAD(&fs_info->trans_list);
1905 INIT_LIST_HEAD(&fs_info->dead_roots);
1906 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1907 INIT_LIST_HEAD(&fs_info->hashers);
1908 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1909 INIT_LIST_HEAD(&fs_info->ordered_operations);
1910 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1911 spin_lock_init(&fs_info->delalloc_lock);
1912 spin_lock_init(&fs_info->trans_lock);
1913 spin_lock_init(&fs_info->ref_cache_lock);
1914 spin_lock_init(&fs_info->fs_roots_radix_lock);
1915 spin_lock_init(&fs_info->delayed_iput_lock);
1916 spin_lock_init(&fs_info->defrag_inodes_lock);
1917 spin_lock_init(&fs_info->free_chunk_lock);
1918 spin_lock_init(&fs_info->tree_mod_seq_lock);
1919 rwlock_init(&fs_info->tree_mod_log_lock);
1920 mutex_init(&fs_info->reloc_mutex);
1921
1922 init_completion(&fs_info->kobj_unregister);
1923 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1924 INIT_LIST_HEAD(&fs_info->space_info);
1925 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
1926 btrfs_mapping_init(&fs_info->mapping_tree);
1927 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1928 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1929 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1930 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1931 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1932 btrfs_init_block_rsv(&fs_info->delayed_block_rsv);
1933 atomic_set(&fs_info->nr_async_submits, 0);
1934 atomic_set(&fs_info->async_delalloc_pages, 0);
1935 atomic_set(&fs_info->async_submit_draining, 0);
1936 atomic_set(&fs_info->nr_async_bios, 0);
1937 atomic_set(&fs_info->defrag_running, 0);
1938 atomic_set(&fs_info->tree_mod_seq, 0);
1939 fs_info->sb = sb;
1940 fs_info->max_inline = 8192 * 1024;
1941 fs_info->metadata_ratio = 0;
1942 fs_info->defrag_inodes = RB_ROOT;
1943 fs_info->trans_no_join = 0;
1944 fs_info->free_chunk_space = 0;
1945 fs_info->tree_mod_log = RB_ROOT;
1946
1947 /* readahead state */
1948 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
1949 spin_lock_init(&fs_info->reada_lock);
1950
1951 fs_info->thread_pool_size = min_t(unsigned long,
1952 num_online_cpus() + 2, 8);
1953
1954 INIT_LIST_HEAD(&fs_info->ordered_extents);
1955 spin_lock_init(&fs_info->ordered_extent_lock);
1956 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1957 GFP_NOFS);
1958 if (!fs_info->delayed_root) {
1959 err = -ENOMEM;
1960 goto fail_iput;
1961 }
1962 btrfs_init_delayed_root(fs_info->delayed_root);
1963
1964 mutex_init(&fs_info->scrub_lock);
1965 atomic_set(&fs_info->scrubs_running, 0);
1966 atomic_set(&fs_info->scrub_pause_req, 0);
1967 atomic_set(&fs_info->scrubs_paused, 0);
1968 atomic_set(&fs_info->scrub_cancel_req, 0);
1969 init_waitqueue_head(&fs_info->scrub_pause_wait);
1970 init_rwsem(&fs_info->scrub_super_lock);
1971 fs_info->scrub_workers_refcnt = 0;
1972#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1973 fs_info->check_integrity_print_mask = 0;
1974#endif
1975
1976 spin_lock_init(&fs_info->balance_lock);
1977 mutex_init(&fs_info->balance_mutex);
1978 atomic_set(&fs_info->balance_running, 0);
1979 atomic_set(&fs_info->balance_pause_req, 0);
1980 atomic_set(&fs_info->balance_cancel_req, 0);
1981 fs_info->balance_ctl = NULL;
1982 init_waitqueue_head(&fs_info->balance_wait_q);
1983
1984 sb->s_blocksize = 4096;
1985 sb->s_blocksize_bits = blksize_bits(4096);
1986 sb->s_bdi = &fs_info->bdi;
1987
1988 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1989 set_nlink(fs_info->btree_inode, 1);
1990 /*
1991 * we set the i_size on the btree inode to the max possible int.
1992 * the real end of the address space is determined by all of
1993 * the devices in the system
1994 */
1995 fs_info->btree_inode->i_size = OFFSET_MAX;
1996 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1997 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1998
1999 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2000 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2001 fs_info->btree_inode->i_mapping);
2002 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2003 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2004
2005 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2006
2007 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2008 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2009 sizeof(struct btrfs_key));
2010 set_bit(BTRFS_INODE_DUMMY,
2011 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2012 insert_inode_hash(fs_info->btree_inode);
2013
2014 spin_lock_init(&fs_info->block_group_cache_lock);
2015 fs_info->block_group_cache_tree = RB_ROOT;
2016
2017 extent_io_tree_init(&fs_info->freed_extents[0],
2018 fs_info->btree_inode->i_mapping);
2019 extent_io_tree_init(&fs_info->freed_extents[1],
2020 fs_info->btree_inode->i_mapping);
2021 fs_info->pinned_extents = &fs_info->freed_extents[0];
2022 fs_info->do_barriers = 1;
2023
2024
2025 mutex_init(&fs_info->ordered_operations_mutex);
2026 mutex_init(&fs_info->tree_log_mutex);
2027 mutex_init(&fs_info->chunk_mutex);
2028 mutex_init(&fs_info->transaction_kthread_mutex);
2029 mutex_init(&fs_info->cleaner_mutex);
2030 mutex_init(&fs_info->volume_mutex);
2031 init_rwsem(&fs_info->extent_commit_sem);
2032 init_rwsem(&fs_info->cleanup_work_sem);
2033 init_rwsem(&fs_info->subvol_sem);
2034
2035 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2036 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2037
2038 init_waitqueue_head(&fs_info->transaction_throttle);
2039 init_waitqueue_head(&fs_info->transaction_wait);
2040 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2041 init_waitqueue_head(&fs_info->async_submit_wait);
2042
2043 __setup_root(4096, 4096, 4096, 4096, tree_root,
2044 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2045
2046 invalidate_bdev(fs_devices->latest_bdev);
2047 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2048 if (!bh) {
2049 err = -EINVAL;
2050 goto fail_alloc;
2051 }
2052
2053 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2054 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2055 sizeof(*fs_info->super_for_commit));
2056 brelse(bh);
2057
2058 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2059
2060 disk_super = fs_info->super_copy;
2061 if (!btrfs_super_root(disk_super))
2062 goto fail_alloc;
2063
2064 /* check FS state, whether FS is broken. */
2065 fs_info->fs_state |= btrfs_super_flags(disk_super);
2066
2067 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2068 if (ret) {
2069 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2070 err = ret;
2071 goto fail_alloc;
2072 }
2073
2074 /*
2075 * run through our array of backup supers and setup
2076 * our ring pointer to the oldest one
2077 */
2078 generation = btrfs_super_generation(disk_super);
2079 find_oldest_super_backup(fs_info, generation);
2080
2081 /*
2082 * In the long term, we'll store the compression type in the super
2083 * block, and it'll be used for per file compression control.
2084 */
2085 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2086
2087 ret = btrfs_parse_options(tree_root, options);
2088 if (ret) {
2089 err = ret;
2090 goto fail_alloc;
2091 }
2092
2093 features = btrfs_super_incompat_flags(disk_super) &
2094 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2095 if (features) {
2096 printk(KERN_ERR "BTRFS: couldn't mount because of "
2097 "unsupported optional features (%Lx).\n",
2098 (unsigned long long)features);
2099 err = -EINVAL;
2100 goto fail_alloc;
2101 }
2102
2103 if (btrfs_super_leafsize(disk_super) !=
2104 btrfs_super_nodesize(disk_super)) {
2105 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2106 "blocksizes don't match. node %d leaf %d\n",
2107 btrfs_super_nodesize(disk_super),
2108 btrfs_super_leafsize(disk_super));
2109 err = -EINVAL;
2110 goto fail_alloc;
2111 }
2112 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2113 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2114 "blocksize (%d) was too large\n",
2115 btrfs_super_leafsize(disk_super));
2116 err = -EINVAL;
2117 goto fail_alloc;
2118 }
2119
2120 features = btrfs_super_incompat_flags(disk_super);
2121 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2122 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2123 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2124
2125 /*
2126 * flag our filesystem as having big metadata blocks if
2127 * they are bigger than the page size
2128 */
2129 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2130 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2131 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2132 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2133 }
2134
2135 nodesize = btrfs_super_nodesize(disk_super);
2136 leafsize = btrfs_super_leafsize(disk_super);
2137 sectorsize = btrfs_super_sectorsize(disk_super);
2138 stripesize = btrfs_super_stripesize(disk_super);
2139
2140 /*
2141 * mixed block groups end up with duplicate but slightly offset
2142 * extent buffers for the same range. It leads to corruptions
2143 */
2144 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2145 (sectorsize != leafsize)) {
2146 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2147 "are not allowed for mixed block groups on %s\n",
2148 sb->s_id);
2149 goto fail_alloc;
2150 }
2151
2152 btrfs_set_super_incompat_flags(disk_super, features);
2153
2154 features = btrfs_super_compat_ro_flags(disk_super) &
2155 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2156 if (!(sb->s_flags & MS_RDONLY) && features) {
2157 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2158 "unsupported option features (%Lx).\n",
2159 (unsigned long long)features);
2160 err = -EINVAL;
2161 goto fail_alloc;
2162 }
2163
2164 btrfs_init_workers(&fs_info->generic_worker,
2165 "genwork", 1, NULL);
2166
2167 btrfs_init_workers(&fs_info->workers, "worker",
2168 fs_info->thread_pool_size,
2169 &fs_info->generic_worker);
2170
2171 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2172 fs_info->thread_pool_size,
2173 &fs_info->generic_worker);
2174
2175 btrfs_init_workers(&fs_info->submit_workers, "submit",
2176 min_t(u64, fs_devices->num_devices,
2177 fs_info->thread_pool_size),
2178 &fs_info->generic_worker);
2179
2180 btrfs_init_workers(&fs_info->caching_workers, "cache",
2181 2, &fs_info->generic_worker);
2182
2183 /* a higher idle thresh on the submit workers makes it much more
2184 * likely that bios will be send down in a sane order to the
2185 * devices
2186 */
2187 fs_info->submit_workers.idle_thresh = 64;
2188
2189 fs_info->workers.idle_thresh = 16;
2190 fs_info->workers.ordered = 1;
2191
2192 fs_info->delalloc_workers.idle_thresh = 2;
2193 fs_info->delalloc_workers.ordered = 1;
2194
2195 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2196 &fs_info->generic_worker);
2197 btrfs_init_workers(&fs_info->endio_workers, "endio",
2198 fs_info->thread_pool_size,
2199 &fs_info->generic_worker);
2200 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2201 fs_info->thread_pool_size,
2202 &fs_info->generic_worker);
2203 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2204 "endio-meta-write", fs_info->thread_pool_size,
2205 &fs_info->generic_worker);
2206 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2207 fs_info->thread_pool_size,
2208 &fs_info->generic_worker);
2209 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2210 1, &fs_info->generic_worker);
2211 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2212 fs_info->thread_pool_size,
2213 &fs_info->generic_worker);
2214 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2215 fs_info->thread_pool_size,
2216 &fs_info->generic_worker);
2217
2218 /*
2219 * endios are largely parallel and should have a very
2220 * low idle thresh
2221 */
2222 fs_info->endio_workers.idle_thresh = 4;
2223 fs_info->endio_meta_workers.idle_thresh = 4;
2224
2225 fs_info->endio_write_workers.idle_thresh = 2;
2226 fs_info->endio_meta_write_workers.idle_thresh = 2;
2227 fs_info->readahead_workers.idle_thresh = 2;
2228
2229 /*
2230 * btrfs_start_workers can really only fail because of ENOMEM so just
2231 * return -ENOMEM if any of these fail.
2232 */
2233 ret = btrfs_start_workers(&fs_info->workers);
2234 ret |= btrfs_start_workers(&fs_info->generic_worker);
2235 ret |= btrfs_start_workers(&fs_info->submit_workers);
2236 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2237 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2238 ret |= btrfs_start_workers(&fs_info->endio_workers);
2239 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2240 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2241 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2242 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2243 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2244 ret |= btrfs_start_workers(&fs_info->caching_workers);
2245 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2246 if (ret) {
2247 ret = -ENOMEM;
2248 goto fail_sb_buffer;
2249 }
2250
2251 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2252 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2253 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2254
2255 tree_root->nodesize = nodesize;
2256 tree_root->leafsize = leafsize;
2257 tree_root->sectorsize = sectorsize;
2258 tree_root->stripesize = stripesize;
2259
2260 sb->s_blocksize = sectorsize;
2261 sb->s_blocksize_bits = blksize_bits(sectorsize);
2262
2263 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2264 sizeof(disk_super->magic))) {
2265 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2266 goto fail_sb_buffer;
2267 }
2268
2269 if (sectorsize != PAGE_SIZE) {
2270 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2271 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2272 goto fail_sb_buffer;
2273 }
2274
2275 mutex_lock(&fs_info->chunk_mutex);
2276 ret = btrfs_read_sys_array(tree_root);
2277 mutex_unlock(&fs_info->chunk_mutex);
2278 if (ret) {
2279 printk(KERN_WARNING "btrfs: failed to read the system "
2280 "array on %s\n", sb->s_id);
2281 goto fail_sb_buffer;
2282 }
2283
2284 blocksize = btrfs_level_size(tree_root,
2285 btrfs_super_chunk_root_level(disk_super));
2286 generation = btrfs_super_chunk_root_generation(disk_super);
2287
2288 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2289 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2290
2291 chunk_root->node = read_tree_block(chunk_root,
2292 btrfs_super_chunk_root(disk_super),
2293 blocksize, generation);
2294 BUG_ON(!chunk_root->node); /* -ENOMEM */
2295 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2296 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2297 sb->s_id);
2298 goto fail_tree_roots;
2299 }
2300 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2301 chunk_root->commit_root = btrfs_root_node(chunk_root);
2302
2303 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2304 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2305 BTRFS_UUID_SIZE);
2306
2307 ret = btrfs_read_chunk_tree(chunk_root);
2308 if (ret) {
2309 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2310 sb->s_id);
2311 goto fail_tree_roots;
2312 }
2313
2314 btrfs_close_extra_devices(fs_devices);
2315
2316 if (!fs_devices->latest_bdev) {
2317 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2318 sb->s_id);
2319 goto fail_tree_roots;
2320 }
2321
2322retry_root_backup:
2323 blocksize = btrfs_level_size(tree_root,
2324 btrfs_super_root_level(disk_super));
2325 generation = btrfs_super_generation(disk_super);
2326
2327 tree_root->node = read_tree_block(tree_root,
2328 btrfs_super_root(disk_super),
2329 blocksize, generation);
2330 if (!tree_root->node ||
2331 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2332 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2333 sb->s_id);
2334
2335 goto recovery_tree_root;
2336 }
2337
2338 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2339 tree_root->commit_root = btrfs_root_node(tree_root);
2340
2341 ret = find_and_setup_root(tree_root, fs_info,
2342 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2343 if (ret)
2344 goto recovery_tree_root;
2345 extent_root->track_dirty = 1;
2346
2347 ret = find_and_setup_root(tree_root, fs_info,
2348 BTRFS_DEV_TREE_OBJECTID, dev_root);
2349 if (ret)
2350 goto recovery_tree_root;
2351 dev_root->track_dirty = 1;
2352
2353 ret = find_and_setup_root(tree_root, fs_info,
2354 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2355 if (ret)
2356 goto recovery_tree_root;
2357 csum_root->track_dirty = 1;
2358
2359 fs_info->generation = generation;
2360 fs_info->last_trans_committed = generation;
2361
2362 ret = btrfs_recover_balance(fs_info);
2363 if (ret) {
2364 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2365 goto fail_block_groups;
2366 }
2367
2368 ret = btrfs_init_dev_stats(fs_info);
2369 if (ret) {
2370 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2371 ret);
2372 goto fail_block_groups;
2373 }
2374
2375 ret = btrfs_init_space_info(fs_info);
2376 if (ret) {
2377 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2378 goto fail_block_groups;
2379 }
2380
2381 ret = btrfs_read_block_groups(extent_root);
2382 if (ret) {
2383 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2384 goto fail_block_groups;
2385 }
2386
2387 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2388 "btrfs-cleaner");
2389 if (IS_ERR(fs_info->cleaner_kthread))
2390 goto fail_block_groups;
2391
2392 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2393 tree_root,
2394 "btrfs-transaction");
2395 if (IS_ERR(fs_info->transaction_kthread))
2396 goto fail_cleaner;
2397
2398 if (!btrfs_test_opt(tree_root, SSD) &&
2399 !btrfs_test_opt(tree_root, NOSSD) &&
2400 !fs_info->fs_devices->rotating) {
2401 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2402 "mode\n");
2403 btrfs_set_opt(fs_info->mount_opt, SSD);
2404 }
2405
2406#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2407 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2408 ret = btrfsic_mount(tree_root, fs_devices,
2409 btrfs_test_opt(tree_root,
2410 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2411 1 : 0,
2412 fs_info->check_integrity_print_mask);
2413 if (ret)
2414 printk(KERN_WARNING "btrfs: failed to initialize"
2415 " integrity check module %s\n", sb->s_id);
2416 }
2417#endif
2418
2419 /* do not make disk changes in broken FS */
2420 if (btrfs_super_log_root(disk_super) != 0 &&
2421 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2422 u64 bytenr = btrfs_super_log_root(disk_super);
2423
2424 if (fs_devices->rw_devices == 0) {
2425 printk(KERN_WARNING "Btrfs log replay required "
2426 "on RO media\n");
2427 err = -EIO;
2428 goto fail_trans_kthread;
2429 }
2430 blocksize =
2431 btrfs_level_size(tree_root,
2432 btrfs_super_log_root_level(disk_super));
2433
2434 log_tree_root = btrfs_alloc_root(fs_info);
2435 if (!log_tree_root) {
2436 err = -ENOMEM;
2437 goto fail_trans_kthread;
2438 }
2439
2440 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2441 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2442
2443 log_tree_root->node = read_tree_block(tree_root, bytenr,
2444 blocksize,
2445 generation + 1);
2446 /* returns with log_tree_root freed on success */
2447 ret = btrfs_recover_log_trees(log_tree_root);
2448 if (ret) {
2449 btrfs_error(tree_root->fs_info, ret,
2450 "Failed to recover log tree");
2451 free_extent_buffer(log_tree_root->node);
2452 kfree(log_tree_root);
2453 goto fail_trans_kthread;
2454 }
2455
2456 if (sb->s_flags & MS_RDONLY) {
2457 ret = btrfs_commit_super(tree_root);
2458 if (ret)
2459 goto fail_trans_kthread;
2460 }
2461 }
2462
2463 ret = btrfs_find_orphan_roots(tree_root);
2464 if (ret)
2465 goto fail_trans_kthread;
2466
2467 if (!(sb->s_flags & MS_RDONLY)) {
2468 ret = btrfs_cleanup_fs_roots(fs_info);
2469 if (ret) {
2470 }
2471
2472 ret = btrfs_recover_relocation(tree_root);
2473 if (ret < 0) {
2474 printk(KERN_WARNING
2475 "btrfs: failed to recover relocation\n");
2476 err = -EINVAL;
2477 goto fail_trans_kthread;
2478 }
2479 }
2480
2481 location.objectid = BTRFS_FS_TREE_OBJECTID;
2482 location.type = BTRFS_ROOT_ITEM_KEY;
2483 location.offset = (u64)-1;
2484
2485 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2486 if (!fs_info->fs_root)
2487 goto fail_trans_kthread;
2488 if (IS_ERR(fs_info->fs_root)) {
2489 err = PTR_ERR(fs_info->fs_root);
2490 goto fail_trans_kthread;
2491 }
2492
2493 if (sb->s_flags & MS_RDONLY)
2494 return 0;
2495
2496 down_read(&fs_info->cleanup_work_sem);
2497 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2498 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2499 up_read(&fs_info->cleanup_work_sem);
2500 close_ctree(tree_root);
2501 return ret;
2502 }
2503 up_read(&fs_info->cleanup_work_sem);
2504
2505 ret = btrfs_resume_balance_async(fs_info);
2506 if (ret) {
2507 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2508 close_ctree(tree_root);
2509 return ret;
2510 }
2511
2512 return 0;
2513
2514fail_trans_kthread:
2515 kthread_stop(fs_info->transaction_kthread);
2516fail_cleaner:
2517 kthread_stop(fs_info->cleaner_kthread);
2518
2519 /*
2520 * make sure we're done with the btree inode before we stop our
2521 * kthreads
2522 */
2523 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2524 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2525
2526fail_block_groups:
2527 btrfs_free_block_groups(fs_info);
2528
2529fail_tree_roots:
2530 free_root_pointers(fs_info, 1);
2531
2532fail_sb_buffer:
2533 btrfs_stop_workers(&fs_info->generic_worker);
2534 btrfs_stop_workers(&fs_info->readahead_workers);
2535 btrfs_stop_workers(&fs_info->fixup_workers);
2536 btrfs_stop_workers(&fs_info->delalloc_workers);
2537 btrfs_stop_workers(&fs_info->workers);
2538 btrfs_stop_workers(&fs_info->endio_workers);
2539 btrfs_stop_workers(&fs_info->endio_meta_workers);
2540 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2541 btrfs_stop_workers(&fs_info->endio_write_workers);
2542 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2543 btrfs_stop_workers(&fs_info->submit_workers);
2544 btrfs_stop_workers(&fs_info->delayed_workers);
2545 btrfs_stop_workers(&fs_info->caching_workers);
2546fail_alloc:
2547fail_iput:
2548 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2549
2550 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2551 iput(fs_info->btree_inode);
2552fail_bdi:
2553 bdi_destroy(&fs_info->bdi);
2554fail_srcu:
2555 cleanup_srcu_struct(&fs_info->subvol_srcu);
2556fail:
2557 btrfs_close_devices(fs_info->fs_devices);
2558 return err;
2559
2560recovery_tree_root:
2561 if (!btrfs_test_opt(tree_root, RECOVERY))
2562 goto fail_tree_roots;
2563
2564 free_root_pointers(fs_info, 0);
2565
2566 /* don't use the log in recovery mode, it won't be valid */
2567 btrfs_set_super_log_root(disk_super, 0);
2568
2569 /* we can't trust the free space cache either */
2570 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2571
2572 ret = next_root_backup(fs_info, fs_info->super_copy,
2573 &num_backups_tried, &backup_index);
2574 if (ret == -1)
2575 goto fail_block_groups;
2576 goto retry_root_backup;
2577}
2578
2579static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2580{
2581 if (uptodate) {
2582 set_buffer_uptodate(bh);
2583 } else {
2584 struct btrfs_device *device = (struct btrfs_device *)
2585 bh->b_private;
2586
2587 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2588 "I/O error on %s\n",
2589 rcu_str_deref(device->name));
2590 /* note, we dont' set_buffer_write_io_error because we have
2591 * our own ways of dealing with the IO errors
2592 */
2593 clear_buffer_uptodate(bh);
2594 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
2595 }
2596 unlock_buffer(bh);
2597 put_bh(bh);
2598}
2599
2600struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2601{
2602 struct buffer_head *bh;
2603 struct buffer_head *latest = NULL;
2604 struct btrfs_super_block *super;
2605 int i;
2606 u64 transid = 0;
2607 u64 bytenr;
2608
2609 /* we would like to check all the supers, but that would make
2610 * a btrfs mount succeed after a mkfs from a different FS.
2611 * So, we need to add a special mount option to scan for
2612 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2613 */
2614 for (i = 0; i < 1; i++) {
2615 bytenr = btrfs_sb_offset(i);
2616 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2617 break;
2618 bh = __bread(bdev, bytenr / 4096, 4096);
2619 if (!bh)
2620 continue;
2621
2622 super = (struct btrfs_super_block *)bh->b_data;
2623 if (btrfs_super_bytenr(super) != bytenr ||
2624 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2625 sizeof(super->magic))) {
2626 brelse(bh);
2627 continue;
2628 }
2629
2630 if (!latest || btrfs_super_generation(super) > transid) {
2631 brelse(latest);
2632 latest = bh;
2633 transid = btrfs_super_generation(super);
2634 } else {
2635 brelse(bh);
2636 }
2637 }
2638 return latest;
2639}
2640
2641/*
2642 * this should be called twice, once with wait == 0 and
2643 * once with wait == 1. When wait == 0 is done, all the buffer heads
2644 * we write are pinned.
2645 *
2646 * They are released when wait == 1 is done.
2647 * max_mirrors must be the same for both runs, and it indicates how
2648 * many supers on this one device should be written.
2649 *
2650 * max_mirrors == 0 means to write them all.
2651 */
2652static int write_dev_supers(struct btrfs_device *device,
2653 struct btrfs_super_block *sb,
2654 int do_barriers, int wait, int max_mirrors)
2655{
2656 struct buffer_head *bh;
2657 int i;
2658 int ret;
2659 int errors = 0;
2660 u32 crc;
2661 u64 bytenr;
2662
2663 if (max_mirrors == 0)
2664 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2665
2666 for (i = 0; i < max_mirrors; i++) {
2667 bytenr = btrfs_sb_offset(i);
2668 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2669 break;
2670
2671 if (wait) {
2672 bh = __find_get_block(device->bdev, bytenr / 4096,
2673 BTRFS_SUPER_INFO_SIZE);
2674 BUG_ON(!bh);
2675 wait_on_buffer(bh);
2676 if (!buffer_uptodate(bh))
2677 errors++;
2678
2679 /* drop our reference */
2680 brelse(bh);
2681
2682 /* drop the reference from the wait == 0 run */
2683 brelse(bh);
2684 continue;
2685 } else {
2686 btrfs_set_super_bytenr(sb, bytenr);
2687
2688 crc = ~(u32)0;
2689 crc = btrfs_csum_data(NULL, (char *)sb +
2690 BTRFS_CSUM_SIZE, crc,
2691 BTRFS_SUPER_INFO_SIZE -
2692 BTRFS_CSUM_SIZE);
2693 btrfs_csum_final(crc, sb->csum);
2694
2695 /*
2696 * one reference for us, and we leave it for the
2697 * caller
2698 */
2699 bh = __getblk(device->bdev, bytenr / 4096,
2700 BTRFS_SUPER_INFO_SIZE);
2701 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2702
2703 /* one reference for submit_bh */
2704 get_bh(bh);
2705
2706 set_buffer_uptodate(bh);
2707 lock_buffer(bh);
2708 bh->b_end_io = btrfs_end_buffer_write_sync;
2709 bh->b_private = device;
2710 }
2711
2712 /*
2713 * we fua the first super. The others we allow
2714 * to go down lazy.
2715 */
2716 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2717 if (ret)
2718 errors++;
2719 }
2720 return errors < i ? 0 : -1;
2721}
2722
2723/*
2724 * endio for the write_dev_flush, this will wake anyone waiting
2725 * for the barrier when it is done
2726 */
2727static void btrfs_end_empty_barrier(struct bio *bio, int err)
2728{
2729 if (err) {
2730 if (err == -EOPNOTSUPP)
2731 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2732 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2733 }
2734 if (bio->bi_private)
2735 complete(bio->bi_private);
2736 bio_put(bio);
2737}
2738
2739/*
2740 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2741 * sent down. With wait == 1, it waits for the previous flush.
2742 *
2743 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2744 * capable
2745 */
2746static int write_dev_flush(struct btrfs_device *device, int wait)
2747{
2748 struct bio *bio;
2749 int ret = 0;
2750
2751 if (device->nobarriers)
2752 return 0;
2753
2754 if (wait) {
2755 bio = device->flush_bio;
2756 if (!bio)
2757 return 0;
2758
2759 wait_for_completion(&device->flush_wait);
2760
2761 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2762 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
2763 rcu_str_deref(device->name));
2764 device->nobarriers = 1;
2765 }
2766 if (!bio_flagged(bio, BIO_UPTODATE)) {
2767 ret = -EIO;
2768 if (!bio_flagged(bio, BIO_EOPNOTSUPP))
2769 btrfs_dev_stat_inc_and_print(device,
2770 BTRFS_DEV_STAT_FLUSH_ERRS);
2771 }
2772
2773 /* drop the reference from the wait == 0 run */
2774 bio_put(bio);
2775 device->flush_bio = NULL;
2776
2777 return ret;
2778 }
2779
2780 /*
2781 * one reference for us, and we leave it for the
2782 * caller
2783 */
2784 device->flush_bio = NULL;
2785 bio = bio_alloc(GFP_NOFS, 0);
2786 if (!bio)
2787 return -ENOMEM;
2788
2789 bio->bi_end_io = btrfs_end_empty_barrier;
2790 bio->bi_bdev = device->bdev;
2791 init_completion(&device->flush_wait);
2792 bio->bi_private = &device->flush_wait;
2793 device->flush_bio = bio;
2794
2795 bio_get(bio);
2796 btrfsic_submit_bio(WRITE_FLUSH, bio);
2797
2798 return 0;
2799}
2800
2801/*
2802 * send an empty flush down to each device in parallel,
2803 * then wait for them
2804 */
2805static int barrier_all_devices(struct btrfs_fs_info *info)
2806{
2807 struct list_head *head;
2808 struct btrfs_device *dev;
2809 int errors = 0;
2810 int ret;
2811
2812 /* send down all the barriers */
2813 head = &info->fs_devices->devices;
2814 list_for_each_entry_rcu(dev, head, dev_list) {
2815 if (!dev->bdev) {
2816 errors++;
2817 continue;
2818 }
2819 if (!dev->in_fs_metadata || !dev->writeable)
2820 continue;
2821
2822 ret = write_dev_flush(dev, 0);
2823 if (ret)
2824 errors++;
2825 }
2826
2827 /* wait for all the barriers */
2828 list_for_each_entry_rcu(dev, head, dev_list) {
2829 if (!dev->bdev) {
2830 errors++;
2831 continue;
2832 }
2833 if (!dev->in_fs_metadata || !dev->writeable)
2834 continue;
2835
2836 ret = write_dev_flush(dev, 1);
2837 if (ret)
2838 errors++;
2839 }
2840 if (errors)
2841 return -EIO;
2842 return 0;
2843}
2844
2845int write_all_supers(struct btrfs_root *root, int max_mirrors)
2846{
2847 struct list_head *head;
2848 struct btrfs_device *dev;
2849 struct btrfs_super_block *sb;
2850 struct btrfs_dev_item *dev_item;
2851 int ret;
2852 int do_barriers;
2853 int max_errors;
2854 int total_errors = 0;
2855 u64 flags;
2856
2857 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2858 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2859 backup_super_roots(root->fs_info);
2860
2861 sb = root->fs_info->super_for_commit;
2862 dev_item = &sb->dev_item;
2863
2864 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2865 head = &root->fs_info->fs_devices->devices;
2866
2867 if (do_barriers)
2868 barrier_all_devices(root->fs_info);
2869
2870 list_for_each_entry_rcu(dev, head, dev_list) {
2871 if (!dev->bdev) {
2872 total_errors++;
2873 continue;
2874 }
2875 if (!dev->in_fs_metadata || !dev->writeable)
2876 continue;
2877
2878 btrfs_set_stack_device_generation(dev_item, 0);
2879 btrfs_set_stack_device_type(dev_item, dev->type);
2880 btrfs_set_stack_device_id(dev_item, dev->devid);
2881 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2882 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2883 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2884 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2885 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2886 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2887 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2888
2889 flags = btrfs_super_flags(sb);
2890 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2891
2892 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2893 if (ret)
2894 total_errors++;
2895 }
2896 if (total_errors > max_errors) {
2897 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2898 total_errors);
2899
2900 /* This shouldn't happen. FUA is masked off if unsupported */
2901 BUG();
2902 }
2903
2904 total_errors = 0;
2905 list_for_each_entry_rcu(dev, head, dev_list) {
2906 if (!dev->bdev)
2907 continue;
2908 if (!dev->in_fs_metadata || !dev->writeable)
2909 continue;
2910
2911 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2912 if (ret)
2913 total_errors++;
2914 }
2915 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2916 if (total_errors > max_errors) {
2917 btrfs_error(root->fs_info, -EIO,
2918 "%d errors while writing supers", total_errors);
2919 return -EIO;
2920 }
2921 return 0;
2922}
2923
2924int write_ctree_super(struct btrfs_trans_handle *trans,
2925 struct btrfs_root *root, int max_mirrors)
2926{
2927 int ret;
2928
2929 ret = write_all_supers(root, max_mirrors);
2930 return ret;
2931}
2932
2933void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2934{
2935 spin_lock(&fs_info->fs_roots_radix_lock);
2936 radix_tree_delete(&fs_info->fs_roots_radix,
2937 (unsigned long)root->root_key.objectid);
2938 spin_unlock(&fs_info->fs_roots_radix_lock);
2939
2940 if (btrfs_root_refs(&root->root_item) == 0)
2941 synchronize_srcu(&fs_info->subvol_srcu);
2942
2943 __btrfs_remove_free_space_cache(root->free_ino_pinned);
2944 __btrfs_remove_free_space_cache(root->free_ino_ctl);
2945 free_fs_root(root);
2946}
2947
2948static void free_fs_root(struct btrfs_root *root)
2949{
2950 iput(root->cache_inode);
2951 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2952 if (root->anon_dev)
2953 free_anon_bdev(root->anon_dev);
2954 free_extent_buffer(root->node);
2955 free_extent_buffer(root->commit_root);
2956 kfree(root->free_ino_ctl);
2957 kfree(root->free_ino_pinned);
2958 kfree(root->name);
2959 kfree(root);
2960}
2961
2962static void del_fs_roots(struct btrfs_fs_info *fs_info)
2963{
2964 int ret;
2965 struct btrfs_root *gang[8];
2966 int i;
2967
2968 while (!list_empty(&fs_info->dead_roots)) {
2969 gang[0] = list_entry(fs_info->dead_roots.next,
2970 struct btrfs_root, root_list);
2971 list_del(&gang[0]->root_list);
2972
2973 if (gang[0]->in_radix) {
2974 btrfs_free_fs_root(fs_info, gang[0]);
2975 } else {
2976 free_extent_buffer(gang[0]->node);
2977 free_extent_buffer(gang[0]->commit_root);
2978 kfree(gang[0]);
2979 }
2980 }
2981
2982 while (1) {
2983 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2984 (void **)gang, 0,
2985 ARRAY_SIZE(gang));
2986 if (!ret)
2987 break;
2988 for (i = 0; i < ret; i++)
2989 btrfs_free_fs_root(fs_info, gang[i]);
2990 }
2991}
2992
2993int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2994{
2995 u64 root_objectid = 0;
2996 struct btrfs_root *gang[8];
2997 int i;
2998 int ret;
2999
3000 while (1) {
3001 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3002 (void **)gang, root_objectid,
3003 ARRAY_SIZE(gang));
3004 if (!ret)
3005 break;
3006
3007 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3008 for (i = 0; i < ret; i++) {
3009 int err;
3010
3011 root_objectid = gang[i]->root_key.objectid;
3012 err = btrfs_orphan_cleanup(gang[i]);
3013 if (err)
3014 return err;
3015 }
3016 root_objectid++;
3017 }
3018 return 0;
3019}
3020
3021int btrfs_commit_super(struct btrfs_root *root)
3022{
3023 struct btrfs_trans_handle *trans;
3024 int ret;
3025
3026 mutex_lock(&root->fs_info->cleaner_mutex);
3027 btrfs_run_delayed_iputs(root);
3028 btrfs_clean_old_snapshots(root);
3029 mutex_unlock(&root->fs_info->cleaner_mutex);
3030
3031 /* wait until ongoing cleanup work done */
3032 down_write(&root->fs_info->cleanup_work_sem);
3033 up_write(&root->fs_info->cleanup_work_sem);
3034
3035 trans = btrfs_join_transaction(root);
3036 if (IS_ERR(trans))
3037 return PTR_ERR(trans);
3038 ret = btrfs_commit_transaction(trans, root);
3039 if (ret)
3040 return ret;
3041 /* run commit again to drop the original snapshot */
3042 trans = btrfs_join_transaction(root);
3043 if (IS_ERR(trans))
3044 return PTR_ERR(trans);
3045 ret = btrfs_commit_transaction(trans, root);
3046 if (ret)
3047 return ret;
3048 ret = btrfs_write_and_wait_transaction(NULL, root);
3049 if (ret) {
3050 btrfs_error(root->fs_info, ret,
3051 "Failed to sync btree inode to disk.");
3052 return ret;
3053 }
3054
3055 ret = write_ctree_super(NULL, root, 0);
3056 return ret;
3057}
3058
3059int close_ctree(struct btrfs_root *root)
3060{
3061 struct btrfs_fs_info *fs_info = root->fs_info;
3062 int ret;
3063
3064 fs_info->closing = 1;
3065 smp_mb();
3066
3067 /* pause restriper - we want to resume on mount */
3068 btrfs_pause_balance(root->fs_info);
3069
3070 btrfs_scrub_cancel(root);
3071
3072 /* wait for any defraggers to finish */
3073 wait_event(fs_info->transaction_wait,
3074 (atomic_read(&fs_info->defrag_running) == 0));
3075
3076 /* clear out the rbtree of defraggable inodes */
3077 btrfs_run_defrag_inodes(fs_info);
3078
3079 /*
3080 * Here come 2 situations when btrfs is broken to flip readonly:
3081 *
3082 * 1. when btrfs flips readonly somewhere else before
3083 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
3084 * and btrfs will skip to write sb directly to keep
3085 * ERROR state on disk.
3086 *
3087 * 2. when btrfs flips readonly just in btrfs_commit_super,
3088 * and in such case, btrfs cannot write sb via btrfs_commit_super,
3089 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
3090 * btrfs will cleanup all FS resources first and write sb then.
3091 */
3092 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3093 ret = btrfs_commit_super(root);
3094 if (ret)
3095 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3096 }
3097
3098 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
3099 ret = btrfs_error_commit_super(root);
3100 if (ret)
3101 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3102 }
3103
3104 btrfs_put_block_group_cache(fs_info);
3105
3106 kthread_stop(fs_info->transaction_kthread);
3107 kthread_stop(fs_info->cleaner_kthread);
3108
3109 fs_info->closing = 2;
3110 smp_mb();
3111
3112 if (fs_info->delalloc_bytes) {
3113 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
3114 (unsigned long long)fs_info->delalloc_bytes);
3115 }
3116 if (fs_info->total_ref_cache_size) {
3117 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
3118 (unsigned long long)fs_info->total_ref_cache_size);
3119 }
3120
3121 free_extent_buffer(fs_info->extent_root->node);
3122 free_extent_buffer(fs_info->extent_root->commit_root);
3123 free_extent_buffer(fs_info->tree_root->node);
3124 free_extent_buffer(fs_info->tree_root->commit_root);
3125 free_extent_buffer(fs_info->chunk_root->node);
3126 free_extent_buffer(fs_info->chunk_root->commit_root);
3127 free_extent_buffer(fs_info->dev_root->node);
3128 free_extent_buffer(fs_info->dev_root->commit_root);
3129 free_extent_buffer(fs_info->csum_root->node);
3130 free_extent_buffer(fs_info->csum_root->commit_root);
3131
3132 btrfs_free_block_groups(fs_info);
3133
3134 del_fs_roots(fs_info);
3135
3136 iput(fs_info->btree_inode);
3137
3138 btrfs_stop_workers(&fs_info->generic_worker);
3139 btrfs_stop_workers(&fs_info->fixup_workers);
3140 btrfs_stop_workers(&fs_info->delalloc_workers);
3141 btrfs_stop_workers(&fs_info->workers);
3142 btrfs_stop_workers(&fs_info->endio_workers);
3143 btrfs_stop_workers(&fs_info->endio_meta_workers);
3144 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
3145 btrfs_stop_workers(&fs_info->endio_write_workers);
3146 btrfs_stop_workers(&fs_info->endio_freespace_worker);
3147 btrfs_stop_workers(&fs_info->submit_workers);
3148 btrfs_stop_workers(&fs_info->delayed_workers);
3149 btrfs_stop_workers(&fs_info->caching_workers);
3150 btrfs_stop_workers(&fs_info->readahead_workers);
3151
3152#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3153 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3154 btrfsic_unmount(root, fs_info->fs_devices);
3155#endif
3156
3157 btrfs_close_devices(fs_info->fs_devices);
3158 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3159
3160 bdi_destroy(&fs_info->bdi);
3161 cleanup_srcu_struct(&fs_info->subvol_srcu);
3162
3163 return 0;
3164}
3165
3166int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3167 int atomic)
3168{
3169 int ret;
3170 struct inode *btree_inode = buf->pages[0]->mapping->host;
3171
3172 ret = extent_buffer_uptodate(buf);
3173 if (!ret)
3174 return ret;
3175
3176 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3177 parent_transid, atomic);
3178 if (ret == -EAGAIN)
3179 return ret;
3180 return !ret;
3181}
3182
3183int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3184{
3185 return set_extent_buffer_uptodate(buf);
3186}
3187
3188void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3189{
3190 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3191 u64 transid = btrfs_header_generation(buf);
3192 int was_dirty;
3193
3194 btrfs_assert_tree_locked(buf);
3195 if (transid != root->fs_info->generation) {
3196 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
3197 "found %llu running %llu\n",
3198 (unsigned long long)buf->start,
3199 (unsigned long long)transid,
3200 (unsigned long long)root->fs_info->generation);
3201 WARN_ON(1);
3202 }
3203 was_dirty = set_extent_buffer_dirty(buf);
3204 if (!was_dirty) {
3205 spin_lock(&root->fs_info->delalloc_lock);
3206 root->fs_info->dirty_metadata_bytes += buf->len;
3207 spin_unlock(&root->fs_info->delalloc_lock);
3208 }
3209}
3210
3211void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3212{
3213 /*
3214 * looks as though older kernels can get into trouble with
3215 * this code, they end up stuck in balance_dirty_pages forever
3216 */
3217 u64 num_dirty;
3218 unsigned long thresh = 32 * 1024 * 1024;
3219
3220 if (current->flags & PF_MEMALLOC)
3221 return;
3222
3223 btrfs_balance_delayed_items(root);
3224
3225 num_dirty = root->fs_info->dirty_metadata_bytes;
3226
3227 if (num_dirty > thresh) {
3228 balance_dirty_pages_ratelimited_nr(
3229 root->fs_info->btree_inode->i_mapping, 1);
3230 }
3231 return;
3232}
3233
3234void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3235{
3236 /*
3237 * looks as though older kernels can get into trouble with
3238 * this code, they end up stuck in balance_dirty_pages forever
3239 */
3240 u64 num_dirty;
3241 unsigned long thresh = 32 * 1024 * 1024;
3242
3243 if (current->flags & PF_MEMALLOC)
3244 return;
3245
3246 num_dirty = root->fs_info->dirty_metadata_bytes;
3247
3248 if (num_dirty > thresh) {
3249 balance_dirty_pages_ratelimited_nr(
3250 root->fs_info->btree_inode->i_mapping, 1);
3251 }
3252 return;
3253}
3254
3255int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3256{
3257 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3258 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3259}
3260
3261static int btree_lock_page_hook(struct page *page, void *data,
3262 void (*flush_fn)(void *))
3263{
3264 struct inode *inode = page->mapping->host;
3265 struct btrfs_root *root = BTRFS_I(inode)->root;
3266 struct extent_buffer *eb;
3267
3268 /*
3269 * We culled this eb but the page is still hanging out on the mapping,
3270 * carry on.
3271 */
3272 if (!PagePrivate(page))
3273 goto out;
3274
3275 eb = (struct extent_buffer *)page->private;
3276 if (!eb) {
3277 WARN_ON(1);
3278 goto out;
3279 }
3280 if (page != eb->pages[0])
3281 goto out;
3282
3283 if (!btrfs_try_tree_write_lock(eb)) {
3284 flush_fn(data);
3285 btrfs_tree_lock(eb);
3286 }
3287 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3288
3289 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3290 spin_lock(&root->fs_info->delalloc_lock);
3291 if (root->fs_info->dirty_metadata_bytes >= eb->len)
3292 root->fs_info->dirty_metadata_bytes -= eb->len;
3293 else
3294 WARN_ON(1);
3295 spin_unlock(&root->fs_info->delalloc_lock);
3296 }
3297
3298 btrfs_tree_unlock(eb);
3299out:
3300 if (!trylock_page(page)) {
3301 flush_fn(data);
3302 lock_page(page);
3303 }
3304 return 0;
3305}
3306
3307static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3308 int read_only)
3309{
3310 if (btrfs_super_csum_type(fs_info->super_copy) >= ARRAY_SIZE(btrfs_csum_sizes)) {
3311 printk(KERN_ERR "btrfs: unsupported checksum algorithm\n");
3312 return -EINVAL;
3313 }
3314
3315 if (read_only)
3316 return 0;
3317
3318 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
3319 printk(KERN_WARNING "warning: mount fs with errors, "
3320 "running btrfsck is recommended\n");
3321 }
3322
3323 return 0;
3324}
3325
3326int btrfs_error_commit_super(struct btrfs_root *root)
3327{
3328 int ret;
3329
3330 mutex_lock(&root->fs_info->cleaner_mutex);
3331 btrfs_run_delayed_iputs(root);
3332 mutex_unlock(&root->fs_info->cleaner_mutex);
3333
3334 down_write(&root->fs_info->cleanup_work_sem);
3335 up_write(&root->fs_info->cleanup_work_sem);
3336
3337 /* cleanup FS via transaction */
3338 btrfs_cleanup_transaction(root);
3339
3340 ret = write_ctree_super(NULL, root, 0);
3341
3342 return ret;
3343}
3344
3345static void btrfs_destroy_ordered_operations(struct btrfs_root *root)
3346{
3347 struct btrfs_inode *btrfs_inode;
3348 struct list_head splice;
3349
3350 INIT_LIST_HEAD(&splice);
3351
3352 mutex_lock(&root->fs_info->ordered_operations_mutex);
3353 spin_lock(&root->fs_info->ordered_extent_lock);
3354
3355 list_splice_init(&root->fs_info->ordered_operations, &splice);
3356 while (!list_empty(&splice)) {
3357 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3358 ordered_operations);
3359
3360 list_del_init(&btrfs_inode->ordered_operations);
3361
3362 btrfs_invalidate_inodes(btrfs_inode->root);
3363 }
3364
3365 spin_unlock(&root->fs_info->ordered_extent_lock);
3366 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3367}
3368
3369static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3370{
3371 struct list_head splice;
3372 struct btrfs_ordered_extent *ordered;
3373 struct inode *inode;
3374
3375 INIT_LIST_HEAD(&splice);
3376
3377 spin_lock(&root->fs_info->ordered_extent_lock);
3378
3379 list_splice_init(&root->fs_info->ordered_extents, &splice);
3380 while (!list_empty(&splice)) {
3381 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3382 root_extent_list);
3383
3384 list_del_init(&ordered->root_extent_list);
3385 atomic_inc(&ordered->refs);
3386
3387 /* the inode may be getting freed (in sys_unlink path). */
3388 inode = igrab(ordered->inode);
3389
3390 spin_unlock(&root->fs_info->ordered_extent_lock);
3391 if (inode)
3392 iput(inode);
3393
3394 atomic_set(&ordered->refs, 1);
3395 btrfs_put_ordered_extent(ordered);
3396
3397 spin_lock(&root->fs_info->ordered_extent_lock);
3398 }
3399
3400 spin_unlock(&root->fs_info->ordered_extent_lock);
3401}
3402
3403int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3404 struct btrfs_root *root)
3405{
3406 struct rb_node *node;
3407 struct btrfs_delayed_ref_root *delayed_refs;
3408 struct btrfs_delayed_ref_node *ref;
3409 int ret = 0;
3410
3411 delayed_refs = &trans->delayed_refs;
3412
3413 spin_lock(&delayed_refs->lock);
3414 if (delayed_refs->num_entries == 0) {
3415 spin_unlock(&delayed_refs->lock);
3416 printk(KERN_INFO "delayed_refs has NO entry\n");
3417 return ret;
3418 }
3419
3420 while ((node = rb_first(&delayed_refs->root)) != NULL) {
3421 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3422
3423 atomic_set(&ref->refs, 1);
3424 if (btrfs_delayed_ref_is_head(ref)) {
3425 struct btrfs_delayed_ref_head *head;
3426
3427 head = btrfs_delayed_node_to_head(ref);
3428 if (!mutex_trylock(&head->mutex)) {
3429 atomic_inc(&ref->refs);
3430 spin_unlock(&delayed_refs->lock);
3431
3432 /* Need to wait for the delayed ref to run */
3433 mutex_lock(&head->mutex);
3434 mutex_unlock(&head->mutex);
3435 btrfs_put_delayed_ref(ref);
3436
3437 spin_lock(&delayed_refs->lock);
3438 continue;
3439 }
3440
3441 kfree(head->extent_op);
3442 delayed_refs->num_heads--;
3443 if (list_empty(&head->cluster))
3444 delayed_refs->num_heads_ready--;
3445 list_del_init(&head->cluster);
3446 }
3447 ref->in_tree = 0;
3448 rb_erase(&ref->rb_node, &delayed_refs->root);
3449 delayed_refs->num_entries--;
3450
3451 spin_unlock(&delayed_refs->lock);
3452 btrfs_put_delayed_ref(ref);
3453
3454 cond_resched();
3455 spin_lock(&delayed_refs->lock);
3456 }
3457
3458 spin_unlock(&delayed_refs->lock);
3459
3460 return ret;
3461}
3462
3463static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3464{
3465 struct btrfs_pending_snapshot *snapshot;
3466 struct list_head splice;
3467
3468 INIT_LIST_HEAD(&splice);
3469
3470 list_splice_init(&t->pending_snapshots, &splice);
3471
3472 while (!list_empty(&splice)) {
3473 snapshot = list_entry(splice.next,
3474 struct btrfs_pending_snapshot,
3475 list);
3476
3477 list_del_init(&snapshot->list);
3478
3479 kfree(snapshot);
3480 }
3481}
3482
3483static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3484{
3485 struct btrfs_inode *btrfs_inode;
3486 struct list_head splice;
3487
3488 INIT_LIST_HEAD(&splice);
3489
3490 spin_lock(&root->fs_info->delalloc_lock);
3491 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3492
3493 while (!list_empty(&splice)) {
3494 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3495 delalloc_inodes);
3496
3497 list_del_init(&btrfs_inode->delalloc_inodes);
3498
3499 btrfs_invalidate_inodes(btrfs_inode->root);
3500 }
3501
3502 spin_unlock(&root->fs_info->delalloc_lock);
3503}
3504
3505static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3506 struct extent_io_tree *dirty_pages,
3507 int mark)
3508{
3509 int ret;
3510 struct page *page;
3511 struct inode *btree_inode = root->fs_info->btree_inode;
3512 struct extent_buffer *eb;
3513 u64 start = 0;
3514 u64 end;
3515 u64 offset;
3516 unsigned long index;
3517
3518 while (1) {
3519 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3520 mark);
3521 if (ret)
3522 break;
3523
3524 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3525 while (start <= end) {
3526 index = start >> PAGE_CACHE_SHIFT;
3527 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3528 page = find_get_page(btree_inode->i_mapping, index);
3529 if (!page)
3530 continue;
3531 offset = page_offset(page);
3532
3533 spin_lock(&dirty_pages->buffer_lock);
3534 eb = radix_tree_lookup(
3535 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3536 offset >> PAGE_CACHE_SHIFT);
3537 spin_unlock(&dirty_pages->buffer_lock);
3538 if (eb)
3539 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3540 &eb->bflags);
3541 if (PageWriteback(page))
3542 end_page_writeback(page);
3543
3544 lock_page(page);
3545 if (PageDirty(page)) {
3546 clear_page_dirty_for_io(page);
3547 spin_lock_irq(&page->mapping->tree_lock);
3548 radix_tree_tag_clear(&page->mapping->page_tree,
3549 page_index(page),
3550 PAGECACHE_TAG_DIRTY);
3551 spin_unlock_irq(&page->mapping->tree_lock);
3552 }
3553
3554 unlock_page(page);
3555 page_cache_release(page);
3556 }
3557 }
3558
3559 return ret;
3560}
3561
3562static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3563 struct extent_io_tree *pinned_extents)
3564{
3565 struct extent_io_tree *unpin;
3566 u64 start;
3567 u64 end;
3568 int ret;
3569 bool loop = true;
3570
3571 unpin = pinned_extents;
3572again:
3573 while (1) {
3574 ret = find_first_extent_bit(unpin, 0, &start, &end,
3575 EXTENT_DIRTY);
3576 if (ret)
3577 break;
3578
3579 /* opt_discard */
3580 if (btrfs_test_opt(root, DISCARD))
3581 ret = btrfs_error_discard_extent(root, start,
3582 end + 1 - start,
3583 NULL);
3584
3585 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3586 btrfs_error_unpin_extent_range(root, start, end);
3587 cond_resched();
3588 }
3589
3590 if (loop) {
3591 if (unpin == &root->fs_info->freed_extents[0])
3592 unpin = &root->fs_info->freed_extents[1];
3593 else
3594 unpin = &root->fs_info->freed_extents[0];
3595 loop = false;
3596 goto again;
3597 }
3598
3599 return 0;
3600}
3601
3602void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3603 struct btrfs_root *root)
3604{
3605 btrfs_destroy_delayed_refs(cur_trans, root);
3606 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3607 cur_trans->dirty_pages.dirty_bytes);
3608
3609 /* FIXME: cleanup wait for commit */
3610 cur_trans->in_commit = 1;
3611 cur_trans->blocked = 1;
3612 wake_up(&root->fs_info->transaction_blocked_wait);
3613
3614 cur_trans->blocked = 0;
3615 wake_up(&root->fs_info->transaction_wait);
3616
3617 cur_trans->commit_done = 1;
3618 wake_up(&cur_trans->commit_wait);
3619
3620 btrfs_destroy_delayed_inodes(root);
3621 btrfs_assert_delayed_root_empty(root);
3622
3623 btrfs_destroy_pending_snapshots(cur_trans);
3624
3625 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3626 EXTENT_DIRTY);
3627 btrfs_destroy_pinned_extent(root,
3628 root->fs_info->pinned_extents);
3629
3630 /*
3631 memset(cur_trans, 0, sizeof(*cur_trans));
3632 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3633 */
3634}
3635
3636int btrfs_cleanup_transaction(struct btrfs_root *root)
3637{
3638 struct btrfs_transaction *t;
3639 LIST_HEAD(list);
3640
3641 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3642
3643 spin_lock(&root->fs_info->trans_lock);
3644 list_splice_init(&root->fs_info->trans_list, &list);
3645 root->fs_info->trans_no_join = 1;
3646 spin_unlock(&root->fs_info->trans_lock);
3647
3648 while (!list_empty(&list)) {
3649 t = list_entry(list.next, struct btrfs_transaction, list);
3650 if (!t)
3651 break;
3652
3653 btrfs_destroy_ordered_operations(root);
3654
3655 btrfs_destroy_ordered_extents(root);
3656
3657 btrfs_destroy_delayed_refs(t, root);
3658
3659 btrfs_block_rsv_release(root,
3660 &root->fs_info->trans_block_rsv,
3661 t->dirty_pages.dirty_bytes);
3662
3663 /* FIXME: cleanup wait for commit */
3664 t->in_commit = 1;
3665 t->blocked = 1;
3666 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3667 wake_up(&root->fs_info->transaction_blocked_wait);
3668
3669 t->blocked = 0;
3670 if (waitqueue_active(&root->fs_info->transaction_wait))
3671 wake_up(&root->fs_info->transaction_wait);
3672
3673 t->commit_done = 1;
3674 if (waitqueue_active(&t->commit_wait))
3675 wake_up(&t->commit_wait);
3676
3677 btrfs_destroy_delayed_inodes(root);
3678 btrfs_assert_delayed_root_empty(root);
3679
3680 btrfs_destroy_pending_snapshots(t);
3681
3682 btrfs_destroy_delalloc_inodes(root);
3683
3684 spin_lock(&root->fs_info->trans_lock);
3685 root->fs_info->running_transaction = NULL;
3686 spin_unlock(&root->fs_info->trans_lock);
3687
3688 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3689 EXTENT_DIRTY);
3690
3691 btrfs_destroy_pinned_extent(root,
3692 root->fs_info->pinned_extents);
3693
3694 atomic_set(&t->use_count, 0);
3695 list_del_init(&t->list);
3696 memset(t, 0, sizeof(*t));
3697 kmem_cache_free(btrfs_transaction_cachep, t);
3698 }
3699
3700 spin_lock(&root->fs_info->trans_lock);
3701 root->fs_info->trans_no_join = 0;
3702 spin_unlock(&root->fs_info->trans_lock);
3703 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3704
3705 return 0;
3706}
3707
3708static struct extent_io_ops btree_extent_io_ops = {
3709 .write_cache_pages_lock_hook = btree_lock_page_hook,
3710 .readpage_end_io_hook = btree_readpage_end_io_hook,
3711 .readpage_io_failed_hook = btree_io_failed_hook,
3712 .submit_bio_hook = btree_submit_bio_hook,
3713 /* note we're sharing with inode.c for the merge bio hook */
3714 .merge_bio_hook = btrfs_merge_bio_hook,
3715};