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/buffer_head.h>
11#include <linux/workqueue.h>
12#include <linux/kthread.h>
13#include <linux/slab.h>
14#include <linux/migrate.h>
15#include <linux/ratelimit.h>
16#include <linux/uuid.h>
17#include <linux/semaphore.h>
18#include <linux/error-injection.h>
19#include <linux/crc32c.h>
20#include <linux/sched/mm.h>
21#include <asm/unaligned.h>
22#include <crypto/hash.h>
23#include "ctree.h"
24#include "disk-io.h"
25#include "transaction.h"
26#include "btrfs_inode.h"
27#include "volumes.h"
28#include "print-tree.h"
29#include "locking.h"
30#include "tree-log.h"
31#include "free-space-cache.h"
32#include "free-space-tree.h"
33#include "inode-map.h"
34#include "check-integrity.h"
35#include "rcu-string.h"
36#include "dev-replace.h"
37#include "raid56.h"
38#include "sysfs.h"
39#include "qgroup.h"
40#include "compression.h"
41#include "tree-checker.h"
42#include "ref-verify.h"
43#include "block-group.h"
44
45#define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
46 BTRFS_HEADER_FLAG_RELOC |\
47 BTRFS_SUPER_FLAG_ERROR |\
48 BTRFS_SUPER_FLAG_SEEDING |\
49 BTRFS_SUPER_FLAG_METADUMP |\
50 BTRFS_SUPER_FLAG_METADUMP_V2)
51
52static const struct extent_io_ops btree_extent_io_ops;
53static void end_workqueue_fn(struct btrfs_work *work);
54static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
55static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
56 struct btrfs_fs_info *fs_info);
57static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
58static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
59 struct extent_io_tree *dirty_pages,
60 int mark);
61static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
62 struct extent_io_tree *pinned_extents);
63static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
64static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
65
66/*
67 * btrfs_end_io_wq structs are used to do processing in task context when an IO
68 * is 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 btrfs_end_io_wq {
72 struct bio *bio;
73 bio_end_io_t *end_io;
74 void *private;
75 struct btrfs_fs_info *info;
76 blk_status_t status;
77 enum btrfs_wq_endio_type metadata;
78 struct btrfs_work work;
79};
80
81static struct kmem_cache *btrfs_end_io_wq_cache;
82
83int __init btrfs_end_io_wq_init(void)
84{
85 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
86 sizeof(struct btrfs_end_io_wq),
87 0,
88 SLAB_MEM_SPREAD,
89 NULL);
90 if (!btrfs_end_io_wq_cache)
91 return -ENOMEM;
92 return 0;
93}
94
95void __cold btrfs_end_io_wq_exit(void)
96{
97 kmem_cache_destroy(btrfs_end_io_wq_cache);
98}
99
100/*
101 * async submit bios are used to offload expensive checksumming
102 * onto the worker threads. They checksum file and metadata bios
103 * just before they are sent down the IO stack.
104 */
105struct async_submit_bio {
106 void *private_data;
107 struct bio *bio;
108 extent_submit_bio_start_t *submit_bio_start;
109 int mirror_num;
110 /*
111 * bio_offset is optional, can be used if the pages in the bio
112 * can't tell us where in the file the bio should go
113 */
114 u64 bio_offset;
115 struct btrfs_work work;
116 blk_status_t status;
117};
118
119/*
120 * Lockdep class keys for extent_buffer->lock's in this root. For a given
121 * eb, the lockdep key is determined by the btrfs_root it belongs to and
122 * the level the eb occupies in the tree.
123 *
124 * Different roots are used for different purposes and may nest inside each
125 * other and they require separate keysets. As lockdep keys should be
126 * static, assign keysets according to the purpose of the root as indicated
127 * by btrfs_root->root_key.objectid. This ensures that all special purpose
128 * roots have separate keysets.
129 *
130 * Lock-nesting across peer nodes is always done with the immediate parent
131 * node locked thus preventing deadlock. As lockdep doesn't know this, use
132 * subclass to avoid triggering lockdep warning in such cases.
133 *
134 * The key is set by the readpage_end_io_hook after the buffer has passed
135 * csum validation but before the pages are unlocked. It is also set by
136 * btrfs_init_new_buffer on freshly allocated blocks.
137 *
138 * We also add a check to make sure the highest level of the tree is the
139 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
140 * needs update as well.
141 */
142#ifdef CONFIG_DEBUG_LOCK_ALLOC
143# if BTRFS_MAX_LEVEL != 8
144# error
145# endif
146
147static struct btrfs_lockdep_keyset {
148 u64 id; /* root objectid */
149 const char *name_stem; /* lock name stem */
150 char names[BTRFS_MAX_LEVEL + 1][20];
151 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
152} btrfs_lockdep_keysets[] = {
153 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
154 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
155 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
156 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
157 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
158 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
159 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
160 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
161 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
162 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
163 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
164 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
165 { .id = 0, .name_stem = "tree" },
166};
167
168void __init btrfs_init_lockdep(void)
169{
170 int i, j;
171
172 /* initialize lockdep class names */
173 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
174 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
175
176 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
177 snprintf(ks->names[j], sizeof(ks->names[j]),
178 "btrfs-%s-%02d", ks->name_stem, j);
179 }
180}
181
182void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
183 int level)
184{
185 struct btrfs_lockdep_keyset *ks;
186
187 BUG_ON(level >= ARRAY_SIZE(ks->keys));
188
189 /* find the matching keyset, id 0 is the default entry */
190 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
191 if (ks->id == objectid)
192 break;
193
194 lockdep_set_class_and_name(&eb->lock,
195 &ks->keys[level], ks->names[level]);
196}
197
198#endif
199
200/*
201 * extents on the btree inode are pretty simple, there's one extent
202 * that covers the entire device
203 */
204struct extent_map *btree_get_extent(struct btrfs_inode *inode,
205 struct page *page, size_t pg_offset, u64 start, u64 len,
206 int create)
207{
208 struct btrfs_fs_info *fs_info = inode->root->fs_info;
209 struct extent_map_tree *em_tree = &inode->extent_tree;
210 struct extent_map *em;
211 int ret;
212
213 read_lock(&em_tree->lock);
214 em = lookup_extent_mapping(em_tree, start, len);
215 if (em) {
216 em->bdev = fs_info->fs_devices->latest_bdev;
217 read_unlock(&em_tree->lock);
218 goto out;
219 }
220 read_unlock(&em_tree->lock);
221
222 em = alloc_extent_map();
223 if (!em) {
224 em = ERR_PTR(-ENOMEM);
225 goto out;
226 }
227 em->start = 0;
228 em->len = (u64)-1;
229 em->block_len = (u64)-1;
230 em->block_start = 0;
231 em->bdev = fs_info->fs_devices->latest_bdev;
232
233 write_lock(&em_tree->lock);
234 ret = add_extent_mapping(em_tree, em, 0);
235 if (ret == -EEXIST) {
236 free_extent_map(em);
237 em = lookup_extent_mapping(em_tree, start, len);
238 if (!em)
239 em = ERR_PTR(-EIO);
240 } else if (ret) {
241 free_extent_map(em);
242 em = ERR_PTR(ret);
243 }
244 write_unlock(&em_tree->lock);
245
246out:
247 return em;
248}
249
250/*
251 * Compute the csum of a btree block and store the result to provided buffer.
252 *
253 * Returns error if the extent buffer cannot be mapped.
254 */
255static int csum_tree_block(struct extent_buffer *buf, u8 *result)
256{
257 struct btrfs_fs_info *fs_info = buf->fs_info;
258 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
259 unsigned long len;
260 unsigned long cur_len;
261 unsigned long offset = BTRFS_CSUM_SIZE;
262 char *kaddr;
263 unsigned long map_start;
264 unsigned long map_len;
265 int err;
266
267 shash->tfm = fs_info->csum_shash;
268 crypto_shash_init(shash);
269
270 len = buf->len - offset;
271
272 while (len > 0) {
273 /*
274 * Note: we don't need to check for the err == 1 case here, as
275 * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
276 * and 'min_len = 32' and the currently implemented mapping
277 * algorithm we cannot cross a page boundary.
278 */
279 err = map_private_extent_buffer(buf, offset, 32,
280 &kaddr, &map_start, &map_len);
281 if (WARN_ON(err))
282 return err;
283 cur_len = min(len, map_len - (offset - map_start));
284 crypto_shash_update(shash, kaddr + offset - map_start, cur_len);
285 len -= cur_len;
286 offset += cur_len;
287 }
288 memset(result, 0, BTRFS_CSUM_SIZE);
289
290 crypto_shash_final(shash, result);
291
292 return 0;
293}
294
295/*
296 * we can't consider a given block up to date unless the transid of the
297 * block matches the transid in the parent node's pointer. This is how we
298 * detect blocks that either didn't get written at all or got written
299 * in the wrong place.
300 */
301static int verify_parent_transid(struct extent_io_tree *io_tree,
302 struct extent_buffer *eb, u64 parent_transid,
303 int atomic)
304{
305 struct extent_state *cached_state = NULL;
306 int ret;
307 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
308
309 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
310 return 0;
311
312 if (atomic)
313 return -EAGAIN;
314
315 if (need_lock) {
316 btrfs_tree_read_lock(eb);
317 btrfs_set_lock_blocking_read(eb);
318 }
319
320 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
321 &cached_state);
322 if (extent_buffer_uptodate(eb) &&
323 btrfs_header_generation(eb) == parent_transid) {
324 ret = 0;
325 goto out;
326 }
327 btrfs_err_rl(eb->fs_info,
328 "parent transid verify failed on %llu wanted %llu found %llu",
329 eb->start,
330 parent_transid, btrfs_header_generation(eb));
331 ret = 1;
332
333 /*
334 * Things reading via commit roots that don't have normal protection,
335 * like send, can have a really old block in cache that may point at a
336 * block that has been freed and re-allocated. So don't clear uptodate
337 * if we find an eb that is under IO (dirty/writeback) because we could
338 * end up reading in the stale data and then writing it back out and
339 * making everybody very sad.
340 */
341 if (!extent_buffer_under_io(eb))
342 clear_extent_buffer_uptodate(eb);
343out:
344 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
345 &cached_state);
346 if (need_lock)
347 btrfs_tree_read_unlock_blocking(eb);
348 return ret;
349}
350
351static bool btrfs_supported_super_csum(u16 csum_type)
352{
353 switch (csum_type) {
354 case BTRFS_CSUM_TYPE_CRC32:
355 return true;
356 default:
357 return false;
358 }
359}
360
361/*
362 * Return 0 if the superblock checksum type matches the checksum value of that
363 * algorithm. Pass the raw disk superblock data.
364 */
365static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
366 char *raw_disk_sb)
367{
368 struct btrfs_super_block *disk_sb =
369 (struct btrfs_super_block *)raw_disk_sb;
370 char result[BTRFS_CSUM_SIZE];
371 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
372
373 shash->tfm = fs_info->csum_shash;
374 crypto_shash_init(shash);
375
376 /*
377 * The super_block structure does not span the whole
378 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
379 * filled with zeros and is included in the checksum.
380 */
381 crypto_shash_update(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
382 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
383 crypto_shash_final(shash, result);
384
385 if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb)))
386 return 1;
387
388 return 0;
389}
390
391int btrfs_verify_level_key(struct extent_buffer *eb, int level,
392 struct btrfs_key *first_key, u64 parent_transid)
393{
394 struct btrfs_fs_info *fs_info = eb->fs_info;
395 int found_level;
396 struct btrfs_key found_key;
397 int ret;
398
399 found_level = btrfs_header_level(eb);
400 if (found_level != level) {
401 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
402 KERN_ERR "BTRFS: tree level check failed\n");
403 btrfs_err(fs_info,
404"tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
405 eb->start, level, found_level);
406 return -EIO;
407 }
408
409 if (!first_key)
410 return 0;
411
412 /*
413 * For live tree block (new tree blocks in current transaction),
414 * we need proper lock context to avoid race, which is impossible here.
415 * So we only checks tree blocks which is read from disk, whose
416 * generation <= fs_info->last_trans_committed.
417 */
418 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
419 return 0;
420
421 /* We have @first_key, so this @eb must have at least one item */
422 if (btrfs_header_nritems(eb) == 0) {
423 btrfs_err(fs_info,
424 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
425 eb->start);
426 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
427 return -EUCLEAN;
428 }
429
430 if (found_level)
431 btrfs_node_key_to_cpu(eb, &found_key, 0);
432 else
433 btrfs_item_key_to_cpu(eb, &found_key, 0);
434 ret = btrfs_comp_cpu_keys(first_key, &found_key);
435
436 if (ret) {
437 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
438 KERN_ERR "BTRFS: tree first key check failed\n");
439 btrfs_err(fs_info,
440"tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
441 eb->start, parent_transid, first_key->objectid,
442 first_key->type, first_key->offset,
443 found_key.objectid, found_key.type,
444 found_key.offset);
445 }
446 return ret;
447}
448
449/*
450 * helper to read a given tree block, doing retries as required when
451 * the checksums don't match and we have alternate mirrors to try.
452 *
453 * @parent_transid: expected transid, skip check if 0
454 * @level: expected level, mandatory check
455 * @first_key: expected key of first slot, skip check if NULL
456 */
457static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
458 u64 parent_transid, int level,
459 struct btrfs_key *first_key)
460{
461 struct btrfs_fs_info *fs_info = eb->fs_info;
462 struct extent_io_tree *io_tree;
463 int failed = 0;
464 int ret;
465 int num_copies = 0;
466 int mirror_num = 0;
467 int failed_mirror = 0;
468
469 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
470 while (1) {
471 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
472 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
473 if (!ret) {
474 if (verify_parent_transid(io_tree, eb,
475 parent_transid, 0))
476 ret = -EIO;
477 else if (btrfs_verify_level_key(eb, level,
478 first_key, parent_transid))
479 ret = -EUCLEAN;
480 else
481 break;
482 }
483
484 num_copies = btrfs_num_copies(fs_info,
485 eb->start, eb->len);
486 if (num_copies == 1)
487 break;
488
489 if (!failed_mirror) {
490 failed = 1;
491 failed_mirror = eb->read_mirror;
492 }
493
494 mirror_num++;
495 if (mirror_num == failed_mirror)
496 mirror_num++;
497
498 if (mirror_num > num_copies)
499 break;
500 }
501
502 if (failed && !ret && failed_mirror)
503 btrfs_repair_eb_io_failure(eb, failed_mirror);
504
505 return ret;
506}
507
508/*
509 * checksum a dirty tree block before IO. This has extra checks to make sure
510 * we only fill in the checksum field in the first page of a multi-page block
511 */
512
513static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
514{
515 u64 start = page_offset(page);
516 u64 found_start;
517 u8 result[BTRFS_CSUM_SIZE];
518 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
519 struct extent_buffer *eb;
520 int ret;
521
522 eb = (struct extent_buffer *)page->private;
523 if (page != eb->pages[0])
524 return 0;
525
526 found_start = btrfs_header_bytenr(eb);
527 /*
528 * Please do not consolidate these warnings into a single if.
529 * It is useful to know what went wrong.
530 */
531 if (WARN_ON(found_start != start))
532 return -EUCLEAN;
533 if (WARN_ON(!PageUptodate(page)))
534 return -EUCLEAN;
535
536 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
537 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
538
539 if (csum_tree_block(eb, result))
540 return -EINVAL;
541
542 if (btrfs_header_level(eb))
543 ret = btrfs_check_node(eb);
544 else
545 ret = btrfs_check_leaf_full(eb);
546
547 if (ret < 0) {
548 btrfs_err(fs_info,
549 "block=%llu write time tree block corruption detected",
550 eb->start);
551 return ret;
552 }
553 write_extent_buffer(eb, result, 0, csum_size);
554
555 return 0;
556}
557
558static int check_tree_block_fsid(struct extent_buffer *eb)
559{
560 struct btrfs_fs_info *fs_info = eb->fs_info;
561 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
562 u8 fsid[BTRFS_FSID_SIZE];
563 int ret = 1;
564
565 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
566 while (fs_devices) {
567 u8 *metadata_uuid;
568
569 /*
570 * Checking the incompat flag is only valid for the current
571 * fs. For seed devices it's forbidden to have their uuid
572 * changed so reading ->fsid in this case is fine
573 */
574 if (fs_devices == fs_info->fs_devices &&
575 btrfs_fs_incompat(fs_info, METADATA_UUID))
576 metadata_uuid = fs_devices->metadata_uuid;
577 else
578 metadata_uuid = fs_devices->fsid;
579
580 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
581 ret = 0;
582 break;
583 }
584 fs_devices = fs_devices->seed;
585 }
586 return ret;
587}
588
589static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
590 u64 phy_offset, struct page *page,
591 u64 start, u64 end, int mirror)
592{
593 u64 found_start;
594 int found_level;
595 struct extent_buffer *eb;
596 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
597 struct btrfs_fs_info *fs_info = root->fs_info;
598 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
599 int ret = 0;
600 u8 result[BTRFS_CSUM_SIZE];
601 int reads_done;
602
603 if (!page->private)
604 goto out;
605
606 eb = (struct extent_buffer *)page->private;
607
608 /* the pending IO might have been the only thing that kept this buffer
609 * in memory. Make sure we have a ref for all this other checks
610 */
611 extent_buffer_get(eb);
612
613 reads_done = atomic_dec_and_test(&eb->io_pages);
614 if (!reads_done)
615 goto err;
616
617 eb->read_mirror = mirror;
618 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
619 ret = -EIO;
620 goto err;
621 }
622
623 found_start = btrfs_header_bytenr(eb);
624 if (found_start != eb->start) {
625 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
626 eb->start, found_start);
627 ret = -EIO;
628 goto err;
629 }
630 if (check_tree_block_fsid(eb)) {
631 btrfs_err_rl(fs_info, "bad fsid on block %llu",
632 eb->start);
633 ret = -EIO;
634 goto err;
635 }
636 found_level = btrfs_header_level(eb);
637 if (found_level >= BTRFS_MAX_LEVEL) {
638 btrfs_err(fs_info, "bad tree block level %d on %llu",
639 (int)btrfs_header_level(eb), eb->start);
640 ret = -EIO;
641 goto err;
642 }
643
644 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
645 eb, found_level);
646
647 ret = csum_tree_block(eb, result);
648 if (ret)
649 goto err;
650
651 if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
652 u32 val;
653 u32 found = 0;
654
655 memcpy(&found, result, csum_size);
656
657 read_extent_buffer(eb, &val, 0, csum_size);
658 btrfs_warn_rl(fs_info,
659 "%s checksum verify failed on %llu wanted %x found %x level %d",
660 fs_info->sb->s_id, eb->start,
661 val, found, btrfs_header_level(eb));
662 ret = -EUCLEAN;
663 goto err;
664 }
665
666 /*
667 * If this is a leaf block and it is corrupt, set the corrupt bit so
668 * that we don't try and read the other copies of this block, just
669 * return -EIO.
670 */
671 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
672 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
673 ret = -EIO;
674 }
675
676 if (found_level > 0 && btrfs_check_node(eb))
677 ret = -EIO;
678
679 if (!ret)
680 set_extent_buffer_uptodate(eb);
681 else
682 btrfs_err(fs_info,
683 "block=%llu read time tree block corruption detected",
684 eb->start);
685err:
686 if (reads_done &&
687 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
688 btree_readahead_hook(eb, ret);
689
690 if (ret) {
691 /*
692 * our io error hook is going to dec the io pages
693 * again, we have to make sure it has something
694 * to decrement
695 */
696 atomic_inc(&eb->io_pages);
697 clear_extent_buffer_uptodate(eb);
698 }
699 free_extent_buffer(eb);
700out:
701 return ret;
702}
703
704static void end_workqueue_bio(struct bio *bio)
705{
706 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
707 struct btrfs_fs_info *fs_info;
708 struct btrfs_workqueue *wq;
709 btrfs_work_func_t func;
710
711 fs_info = end_io_wq->info;
712 end_io_wq->status = bio->bi_status;
713
714 if (bio_op(bio) == REQ_OP_WRITE) {
715 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
716 wq = fs_info->endio_meta_write_workers;
717 func = btrfs_endio_meta_write_helper;
718 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
719 wq = fs_info->endio_freespace_worker;
720 func = btrfs_freespace_write_helper;
721 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
722 wq = fs_info->endio_raid56_workers;
723 func = btrfs_endio_raid56_helper;
724 } else {
725 wq = fs_info->endio_write_workers;
726 func = btrfs_endio_write_helper;
727 }
728 } else {
729 if (unlikely(end_io_wq->metadata ==
730 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
731 wq = fs_info->endio_repair_workers;
732 func = btrfs_endio_repair_helper;
733 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
734 wq = fs_info->endio_raid56_workers;
735 func = btrfs_endio_raid56_helper;
736 } else if (end_io_wq->metadata) {
737 wq = fs_info->endio_meta_workers;
738 func = btrfs_endio_meta_helper;
739 } else {
740 wq = fs_info->endio_workers;
741 func = btrfs_endio_helper;
742 }
743 }
744
745 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
746 btrfs_queue_work(wq, &end_io_wq->work);
747}
748
749blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
750 enum btrfs_wq_endio_type metadata)
751{
752 struct btrfs_end_io_wq *end_io_wq;
753
754 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
755 if (!end_io_wq)
756 return BLK_STS_RESOURCE;
757
758 end_io_wq->private = bio->bi_private;
759 end_io_wq->end_io = bio->bi_end_io;
760 end_io_wq->info = info;
761 end_io_wq->status = 0;
762 end_io_wq->bio = bio;
763 end_io_wq->metadata = metadata;
764
765 bio->bi_private = end_io_wq;
766 bio->bi_end_io = end_workqueue_bio;
767 return 0;
768}
769
770static void run_one_async_start(struct btrfs_work *work)
771{
772 struct async_submit_bio *async;
773 blk_status_t ret;
774
775 async = container_of(work, struct async_submit_bio, work);
776 ret = async->submit_bio_start(async->private_data, async->bio,
777 async->bio_offset);
778 if (ret)
779 async->status = ret;
780}
781
782/*
783 * In order to insert checksums into the metadata in large chunks, we wait
784 * until bio submission time. All the pages in the bio are checksummed and
785 * sums are attached onto the ordered extent record.
786 *
787 * At IO completion time the csums attached on the ordered extent record are
788 * inserted into the tree.
789 */
790static void run_one_async_done(struct btrfs_work *work)
791{
792 struct async_submit_bio *async;
793 struct inode *inode;
794 blk_status_t ret;
795
796 async = container_of(work, struct async_submit_bio, work);
797 inode = async->private_data;
798
799 /* If an error occurred we just want to clean up the bio and move on */
800 if (async->status) {
801 async->bio->bi_status = async->status;
802 bio_endio(async->bio);
803 return;
804 }
805
806 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio,
807 async->mirror_num, 1);
808 if (ret) {
809 async->bio->bi_status = ret;
810 bio_endio(async->bio);
811 }
812}
813
814static void run_one_async_free(struct btrfs_work *work)
815{
816 struct async_submit_bio *async;
817
818 async = container_of(work, struct async_submit_bio, work);
819 kfree(async);
820}
821
822blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
823 int mirror_num, unsigned long bio_flags,
824 u64 bio_offset, void *private_data,
825 extent_submit_bio_start_t *submit_bio_start)
826{
827 struct async_submit_bio *async;
828
829 async = kmalloc(sizeof(*async), GFP_NOFS);
830 if (!async)
831 return BLK_STS_RESOURCE;
832
833 async->private_data = private_data;
834 async->bio = bio;
835 async->mirror_num = mirror_num;
836 async->submit_bio_start = submit_bio_start;
837
838 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
839 run_one_async_done, run_one_async_free);
840
841 async->bio_offset = bio_offset;
842
843 async->status = 0;
844
845 if (op_is_sync(bio->bi_opf))
846 btrfs_set_work_high_priority(&async->work);
847
848 btrfs_queue_work(fs_info->workers, &async->work);
849 return 0;
850}
851
852static blk_status_t btree_csum_one_bio(struct bio *bio)
853{
854 struct bio_vec *bvec;
855 struct btrfs_root *root;
856 int ret = 0;
857 struct bvec_iter_all iter_all;
858
859 ASSERT(!bio_flagged(bio, BIO_CLONED));
860 bio_for_each_segment_all(bvec, bio, iter_all) {
861 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
862 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
863 if (ret)
864 break;
865 }
866
867 return errno_to_blk_status(ret);
868}
869
870static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
871 u64 bio_offset)
872{
873 /*
874 * when we're called for a write, we're already in the async
875 * submission context. Just jump into btrfs_map_bio
876 */
877 return btree_csum_one_bio(bio);
878}
879
880static int check_async_write(struct btrfs_fs_info *fs_info,
881 struct btrfs_inode *bi)
882{
883 if (atomic_read(&bi->sync_writers))
884 return 0;
885 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
886 return 0;
887 return 1;
888}
889
890static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
891 int mirror_num,
892 unsigned long bio_flags)
893{
894 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
895 int async = check_async_write(fs_info, BTRFS_I(inode));
896 blk_status_t ret;
897
898 if (bio_op(bio) != REQ_OP_WRITE) {
899 /*
900 * called for a read, do the setup so that checksum validation
901 * can happen in the async kernel threads
902 */
903 ret = btrfs_bio_wq_end_io(fs_info, bio,
904 BTRFS_WQ_ENDIO_METADATA);
905 if (ret)
906 goto out_w_error;
907 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
908 } else if (!async) {
909 ret = btree_csum_one_bio(bio);
910 if (ret)
911 goto out_w_error;
912 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
913 } else {
914 /*
915 * kthread helpers are used to submit writes so that
916 * checksumming can happen in parallel across all CPUs
917 */
918 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
919 0, inode, btree_submit_bio_start);
920 }
921
922 if (ret)
923 goto out_w_error;
924 return 0;
925
926out_w_error:
927 bio->bi_status = ret;
928 bio_endio(bio);
929 return ret;
930}
931
932#ifdef CONFIG_MIGRATION
933static int btree_migratepage(struct address_space *mapping,
934 struct page *newpage, struct page *page,
935 enum migrate_mode mode)
936{
937 /*
938 * we can't safely write a btree page from here,
939 * we haven't done the locking hook
940 */
941 if (PageDirty(page))
942 return -EAGAIN;
943 /*
944 * Buffers may be managed in a filesystem specific way.
945 * We must have no buffers or drop them.
946 */
947 if (page_has_private(page) &&
948 !try_to_release_page(page, GFP_KERNEL))
949 return -EAGAIN;
950 return migrate_page(mapping, newpage, page, mode);
951}
952#endif
953
954
955static int btree_writepages(struct address_space *mapping,
956 struct writeback_control *wbc)
957{
958 struct btrfs_fs_info *fs_info;
959 int ret;
960
961 if (wbc->sync_mode == WB_SYNC_NONE) {
962
963 if (wbc->for_kupdate)
964 return 0;
965
966 fs_info = BTRFS_I(mapping->host)->root->fs_info;
967 /* this is a bit racy, but that's ok */
968 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
969 BTRFS_DIRTY_METADATA_THRESH,
970 fs_info->dirty_metadata_batch);
971 if (ret < 0)
972 return 0;
973 }
974 return btree_write_cache_pages(mapping, wbc);
975}
976
977static int btree_readpage(struct file *file, struct page *page)
978{
979 struct extent_io_tree *tree;
980 tree = &BTRFS_I(page->mapping->host)->io_tree;
981 return extent_read_full_page(tree, page, btree_get_extent, 0);
982}
983
984static int btree_releasepage(struct page *page, gfp_t gfp_flags)
985{
986 if (PageWriteback(page) || PageDirty(page))
987 return 0;
988
989 return try_release_extent_buffer(page);
990}
991
992static void btree_invalidatepage(struct page *page, unsigned int offset,
993 unsigned int length)
994{
995 struct extent_io_tree *tree;
996 tree = &BTRFS_I(page->mapping->host)->io_tree;
997 extent_invalidatepage(tree, page, offset);
998 btree_releasepage(page, GFP_NOFS);
999 if (PagePrivate(page)) {
1000 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1001 "page private not zero on page %llu",
1002 (unsigned long long)page_offset(page));
1003 ClearPagePrivate(page);
1004 set_page_private(page, 0);
1005 put_page(page);
1006 }
1007}
1008
1009static int btree_set_page_dirty(struct page *page)
1010{
1011#ifdef DEBUG
1012 struct extent_buffer *eb;
1013
1014 BUG_ON(!PagePrivate(page));
1015 eb = (struct extent_buffer *)page->private;
1016 BUG_ON(!eb);
1017 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1018 BUG_ON(!atomic_read(&eb->refs));
1019 btrfs_assert_tree_locked(eb);
1020#endif
1021 return __set_page_dirty_nobuffers(page);
1022}
1023
1024static const struct address_space_operations btree_aops = {
1025 .readpage = btree_readpage,
1026 .writepages = btree_writepages,
1027 .releasepage = btree_releasepage,
1028 .invalidatepage = btree_invalidatepage,
1029#ifdef CONFIG_MIGRATION
1030 .migratepage = btree_migratepage,
1031#endif
1032 .set_page_dirty = btree_set_page_dirty,
1033};
1034
1035void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1036{
1037 struct extent_buffer *buf = NULL;
1038 int ret;
1039
1040 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1041 if (IS_ERR(buf))
1042 return;
1043
1044 ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1045 if (ret < 0)
1046 free_extent_buffer_stale(buf);
1047 else
1048 free_extent_buffer(buf);
1049}
1050
1051struct extent_buffer *btrfs_find_create_tree_block(
1052 struct btrfs_fs_info *fs_info,
1053 u64 bytenr)
1054{
1055 if (btrfs_is_testing(fs_info))
1056 return alloc_test_extent_buffer(fs_info, bytenr);
1057 return alloc_extent_buffer(fs_info, bytenr);
1058}
1059
1060/*
1061 * Read tree block at logical address @bytenr and do variant basic but critical
1062 * verification.
1063 *
1064 * @parent_transid: expected transid of this tree block, skip check if 0
1065 * @level: expected level, mandatory check
1066 * @first_key: expected key in slot 0, skip check if NULL
1067 */
1068struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1069 u64 parent_transid, int level,
1070 struct btrfs_key *first_key)
1071{
1072 struct extent_buffer *buf = NULL;
1073 int ret;
1074
1075 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1076 if (IS_ERR(buf))
1077 return buf;
1078
1079 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1080 level, first_key);
1081 if (ret) {
1082 free_extent_buffer_stale(buf);
1083 return ERR_PTR(ret);
1084 }
1085 return buf;
1086
1087}
1088
1089void btrfs_clean_tree_block(struct extent_buffer *buf)
1090{
1091 struct btrfs_fs_info *fs_info = buf->fs_info;
1092 if (btrfs_header_generation(buf) ==
1093 fs_info->running_transaction->transid) {
1094 btrfs_assert_tree_locked(buf);
1095
1096 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1097 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1098 -buf->len,
1099 fs_info->dirty_metadata_batch);
1100 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1101 btrfs_set_lock_blocking_write(buf);
1102 clear_extent_buffer_dirty(buf);
1103 }
1104 }
1105}
1106
1107static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1108{
1109 struct btrfs_subvolume_writers *writers;
1110 int ret;
1111
1112 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1113 if (!writers)
1114 return ERR_PTR(-ENOMEM);
1115
1116 ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1117 if (ret < 0) {
1118 kfree(writers);
1119 return ERR_PTR(ret);
1120 }
1121
1122 init_waitqueue_head(&writers->wait);
1123 return writers;
1124}
1125
1126static void
1127btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1128{
1129 percpu_counter_destroy(&writers->counter);
1130 kfree(writers);
1131}
1132
1133static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1134 u64 objectid)
1135{
1136 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1137 root->node = NULL;
1138 root->commit_root = NULL;
1139 root->state = 0;
1140 root->orphan_cleanup_state = 0;
1141
1142 root->last_trans = 0;
1143 root->highest_objectid = 0;
1144 root->nr_delalloc_inodes = 0;
1145 root->nr_ordered_extents = 0;
1146 root->inode_tree = RB_ROOT;
1147 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1148 root->block_rsv = NULL;
1149
1150 INIT_LIST_HEAD(&root->dirty_list);
1151 INIT_LIST_HEAD(&root->root_list);
1152 INIT_LIST_HEAD(&root->delalloc_inodes);
1153 INIT_LIST_HEAD(&root->delalloc_root);
1154 INIT_LIST_HEAD(&root->ordered_extents);
1155 INIT_LIST_HEAD(&root->ordered_root);
1156 INIT_LIST_HEAD(&root->reloc_dirty_list);
1157 INIT_LIST_HEAD(&root->logged_list[0]);
1158 INIT_LIST_HEAD(&root->logged_list[1]);
1159 spin_lock_init(&root->inode_lock);
1160 spin_lock_init(&root->delalloc_lock);
1161 spin_lock_init(&root->ordered_extent_lock);
1162 spin_lock_init(&root->accounting_lock);
1163 spin_lock_init(&root->log_extents_lock[0]);
1164 spin_lock_init(&root->log_extents_lock[1]);
1165 spin_lock_init(&root->qgroup_meta_rsv_lock);
1166 mutex_init(&root->objectid_mutex);
1167 mutex_init(&root->log_mutex);
1168 mutex_init(&root->ordered_extent_mutex);
1169 mutex_init(&root->delalloc_mutex);
1170 init_waitqueue_head(&root->log_writer_wait);
1171 init_waitqueue_head(&root->log_commit_wait[0]);
1172 init_waitqueue_head(&root->log_commit_wait[1]);
1173 INIT_LIST_HEAD(&root->log_ctxs[0]);
1174 INIT_LIST_HEAD(&root->log_ctxs[1]);
1175 atomic_set(&root->log_commit[0], 0);
1176 atomic_set(&root->log_commit[1], 0);
1177 atomic_set(&root->log_writers, 0);
1178 atomic_set(&root->log_batch, 0);
1179 refcount_set(&root->refs, 1);
1180 atomic_set(&root->will_be_snapshotted, 0);
1181 atomic_set(&root->snapshot_force_cow, 0);
1182 atomic_set(&root->nr_swapfiles, 0);
1183 root->log_transid = 0;
1184 root->log_transid_committed = -1;
1185 root->last_log_commit = 0;
1186 if (!dummy)
1187 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1188 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1189
1190 memset(&root->root_key, 0, sizeof(root->root_key));
1191 memset(&root->root_item, 0, sizeof(root->root_item));
1192 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1193 if (!dummy)
1194 root->defrag_trans_start = fs_info->generation;
1195 else
1196 root->defrag_trans_start = 0;
1197 root->root_key.objectid = objectid;
1198 root->anon_dev = 0;
1199
1200 spin_lock_init(&root->root_item_lock);
1201 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1202}
1203
1204static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1205 gfp_t flags)
1206{
1207 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1208 if (root)
1209 root->fs_info = fs_info;
1210 return root;
1211}
1212
1213#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1214/* Should only be used by the testing infrastructure */
1215struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1216{
1217 struct btrfs_root *root;
1218
1219 if (!fs_info)
1220 return ERR_PTR(-EINVAL);
1221
1222 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1223 if (!root)
1224 return ERR_PTR(-ENOMEM);
1225
1226 /* We don't use the stripesize in selftest, set it as sectorsize */
1227 __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1228 root->alloc_bytenr = 0;
1229
1230 return root;
1231}
1232#endif
1233
1234struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1235 u64 objectid)
1236{
1237 struct btrfs_fs_info *fs_info = trans->fs_info;
1238 struct extent_buffer *leaf;
1239 struct btrfs_root *tree_root = fs_info->tree_root;
1240 struct btrfs_root *root;
1241 struct btrfs_key key;
1242 unsigned int nofs_flag;
1243 int ret = 0;
1244 uuid_le uuid = NULL_UUID_LE;
1245
1246 /*
1247 * We're holding a transaction handle, so use a NOFS memory allocation
1248 * context to avoid deadlock if reclaim happens.
1249 */
1250 nofs_flag = memalloc_nofs_save();
1251 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1252 memalloc_nofs_restore(nofs_flag);
1253 if (!root)
1254 return ERR_PTR(-ENOMEM);
1255
1256 __setup_root(root, fs_info, objectid);
1257 root->root_key.objectid = objectid;
1258 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1259 root->root_key.offset = 0;
1260
1261 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1262 if (IS_ERR(leaf)) {
1263 ret = PTR_ERR(leaf);
1264 leaf = NULL;
1265 goto fail;
1266 }
1267
1268 root->node = leaf;
1269 btrfs_mark_buffer_dirty(leaf);
1270
1271 root->commit_root = btrfs_root_node(root);
1272 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1273
1274 root->root_item.flags = 0;
1275 root->root_item.byte_limit = 0;
1276 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1277 btrfs_set_root_generation(&root->root_item, trans->transid);
1278 btrfs_set_root_level(&root->root_item, 0);
1279 btrfs_set_root_refs(&root->root_item, 1);
1280 btrfs_set_root_used(&root->root_item, leaf->len);
1281 btrfs_set_root_last_snapshot(&root->root_item, 0);
1282 btrfs_set_root_dirid(&root->root_item, 0);
1283 if (is_fstree(objectid))
1284 uuid_le_gen(&uuid);
1285 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1286 root->root_item.drop_level = 0;
1287
1288 key.objectid = objectid;
1289 key.type = BTRFS_ROOT_ITEM_KEY;
1290 key.offset = 0;
1291 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1292 if (ret)
1293 goto fail;
1294
1295 btrfs_tree_unlock(leaf);
1296
1297 return root;
1298
1299fail:
1300 if (leaf) {
1301 btrfs_tree_unlock(leaf);
1302 free_extent_buffer(root->commit_root);
1303 free_extent_buffer(leaf);
1304 }
1305 kfree(root);
1306
1307 return ERR_PTR(ret);
1308}
1309
1310static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1311 struct btrfs_fs_info *fs_info)
1312{
1313 struct btrfs_root *root;
1314 struct extent_buffer *leaf;
1315
1316 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1317 if (!root)
1318 return ERR_PTR(-ENOMEM);
1319
1320 __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1321
1322 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1323 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1324 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1325
1326 /*
1327 * DON'T set REF_COWS for log trees
1328 *
1329 * log trees do not get reference counted because they go away
1330 * before a real commit is actually done. They do store pointers
1331 * to file data extents, and those reference counts still get
1332 * updated (along with back refs to the log tree).
1333 */
1334
1335 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1336 NULL, 0, 0, 0);
1337 if (IS_ERR(leaf)) {
1338 kfree(root);
1339 return ERR_CAST(leaf);
1340 }
1341
1342 root->node = leaf;
1343
1344 btrfs_mark_buffer_dirty(root->node);
1345 btrfs_tree_unlock(root->node);
1346 return root;
1347}
1348
1349int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1350 struct btrfs_fs_info *fs_info)
1351{
1352 struct btrfs_root *log_root;
1353
1354 log_root = alloc_log_tree(trans, fs_info);
1355 if (IS_ERR(log_root))
1356 return PTR_ERR(log_root);
1357 WARN_ON(fs_info->log_root_tree);
1358 fs_info->log_root_tree = log_root;
1359 return 0;
1360}
1361
1362int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1363 struct btrfs_root *root)
1364{
1365 struct btrfs_fs_info *fs_info = root->fs_info;
1366 struct btrfs_root *log_root;
1367 struct btrfs_inode_item *inode_item;
1368
1369 log_root = alloc_log_tree(trans, fs_info);
1370 if (IS_ERR(log_root))
1371 return PTR_ERR(log_root);
1372
1373 log_root->last_trans = trans->transid;
1374 log_root->root_key.offset = root->root_key.objectid;
1375
1376 inode_item = &log_root->root_item.inode;
1377 btrfs_set_stack_inode_generation(inode_item, 1);
1378 btrfs_set_stack_inode_size(inode_item, 3);
1379 btrfs_set_stack_inode_nlink(inode_item, 1);
1380 btrfs_set_stack_inode_nbytes(inode_item,
1381 fs_info->nodesize);
1382 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1383
1384 btrfs_set_root_node(&log_root->root_item, log_root->node);
1385
1386 WARN_ON(root->log_root);
1387 root->log_root = log_root;
1388 root->log_transid = 0;
1389 root->log_transid_committed = -1;
1390 root->last_log_commit = 0;
1391 return 0;
1392}
1393
1394static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1395 struct btrfs_key *key)
1396{
1397 struct btrfs_root *root;
1398 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1399 struct btrfs_path *path;
1400 u64 generation;
1401 int ret;
1402 int level;
1403
1404 path = btrfs_alloc_path();
1405 if (!path)
1406 return ERR_PTR(-ENOMEM);
1407
1408 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1409 if (!root) {
1410 ret = -ENOMEM;
1411 goto alloc_fail;
1412 }
1413
1414 __setup_root(root, fs_info, key->objectid);
1415
1416 ret = btrfs_find_root(tree_root, key, path,
1417 &root->root_item, &root->root_key);
1418 if (ret) {
1419 if (ret > 0)
1420 ret = -ENOENT;
1421 goto find_fail;
1422 }
1423
1424 generation = btrfs_root_generation(&root->root_item);
1425 level = btrfs_root_level(&root->root_item);
1426 root->node = read_tree_block(fs_info,
1427 btrfs_root_bytenr(&root->root_item),
1428 generation, level, NULL);
1429 if (IS_ERR(root->node)) {
1430 ret = PTR_ERR(root->node);
1431 goto find_fail;
1432 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1433 ret = -EIO;
1434 free_extent_buffer(root->node);
1435 goto find_fail;
1436 }
1437 root->commit_root = btrfs_root_node(root);
1438out:
1439 btrfs_free_path(path);
1440 return root;
1441
1442find_fail:
1443 kfree(root);
1444alloc_fail:
1445 root = ERR_PTR(ret);
1446 goto out;
1447}
1448
1449struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1450 struct btrfs_key *location)
1451{
1452 struct btrfs_root *root;
1453
1454 root = btrfs_read_tree_root(tree_root, location);
1455 if (IS_ERR(root))
1456 return root;
1457
1458 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1459 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1460 btrfs_check_and_init_root_item(&root->root_item);
1461 }
1462
1463 return root;
1464}
1465
1466int btrfs_init_fs_root(struct btrfs_root *root)
1467{
1468 int ret;
1469 struct btrfs_subvolume_writers *writers;
1470
1471 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1472 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1473 GFP_NOFS);
1474 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1475 ret = -ENOMEM;
1476 goto fail;
1477 }
1478
1479 writers = btrfs_alloc_subvolume_writers();
1480 if (IS_ERR(writers)) {
1481 ret = PTR_ERR(writers);
1482 goto fail;
1483 }
1484 root->subv_writers = writers;
1485
1486 btrfs_init_free_ino_ctl(root);
1487 spin_lock_init(&root->ino_cache_lock);
1488 init_waitqueue_head(&root->ino_cache_wait);
1489
1490 ret = get_anon_bdev(&root->anon_dev);
1491 if (ret)
1492 goto fail;
1493
1494 mutex_lock(&root->objectid_mutex);
1495 ret = btrfs_find_highest_objectid(root,
1496 &root->highest_objectid);
1497 if (ret) {
1498 mutex_unlock(&root->objectid_mutex);
1499 goto fail;
1500 }
1501
1502 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1503
1504 mutex_unlock(&root->objectid_mutex);
1505
1506 return 0;
1507fail:
1508 /* The caller is responsible to call btrfs_free_fs_root */
1509 return ret;
1510}
1511
1512struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1513 u64 root_id)
1514{
1515 struct btrfs_root *root;
1516
1517 spin_lock(&fs_info->fs_roots_radix_lock);
1518 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1519 (unsigned long)root_id);
1520 spin_unlock(&fs_info->fs_roots_radix_lock);
1521 return root;
1522}
1523
1524int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1525 struct btrfs_root *root)
1526{
1527 int ret;
1528
1529 ret = radix_tree_preload(GFP_NOFS);
1530 if (ret)
1531 return ret;
1532
1533 spin_lock(&fs_info->fs_roots_radix_lock);
1534 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1535 (unsigned long)root->root_key.objectid,
1536 root);
1537 if (ret == 0)
1538 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1539 spin_unlock(&fs_info->fs_roots_radix_lock);
1540 radix_tree_preload_end();
1541
1542 return ret;
1543}
1544
1545struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1546 struct btrfs_key *location,
1547 bool check_ref)
1548{
1549 struct btrfs_root *root;
1550 struct btrfs_path *path;
1551 struct btrfs_key key;
1552 int ret;
1553
1554 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1555 return fs_info->tree_root;
1556 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1557 return fs_info->extent_root;
1558 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1559 return fs_info->chunk_root;
1560 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1561 return fs_info->dev_root;
1562 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1563 return fs_info->csum_root;
1564 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1565 return fs_info->quota_root ? fs_info->quota_root :
1566 ERR_PTR(-ENOENT);
1567 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1568 return fs_info->uuid_root ? fs_info->uuid_root :
1569 ERR_PTR(-ENOENT);
1570 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1571 return fs_info->free_space_root ? fs_info->free_space_root :
1572 ERR_PTR(-ENOENT);
1573again:
1574 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1575 if (root) {
1576 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1577 return ERR_PTR(-ENOENT);
1578 return root;
1579 }
1580
1581 root = btrfs_read_fs_root(fs_info->tree_root, location);
1582 if (IS_ERR(root))
1583 return root;
1584
1585 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1586 ret = -ENOENT;
1587 goto fail;
1588 }
1589
1590 ret = btrfs_init_fs_root(root);
1591 if (ret)
1592 goto fail;
1593
1594 path = btrfs_alloc_path();
1595 if (!path) {
1596 ret = -ENOMEM;
1597 goto fail;
1598 }
1599 key.objectid = BTRFS_ORPHAN_OBJECTID;
1600 key.type = BTRFS_ORPHAN_ITEM_KEY;
1601 key.offset = location->objectid;
1602
1603 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1604 btrfs_free_path(path);
1605 if (ret < 0)
1606 goto fail;
1607 if (ret == 0)
1608 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1609
1610 ret = btrfs_insert_fs_root(fs_info, root);
1611 if (ret) {
1612 if (ret == -EEXIST) {
1613 btrfs_free_fs_root(root);
1614 goto again;
1615 }
1616 goto fail;
1617 }
1618 return root;
1619fail:
1620 btrfs_free_fs_root(root);
1621 return ERR_PTR(ret);
1622}
1623
1624static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1625{
1626 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1627 int ret = 0;
1628 struct btrfs_device *device;
1629 struct backing_dev_info *bdi;
1630
1631 rcu_read_lock();
1632 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1633 if (!device->bdev)
1634 continue;
1635 bdi = device->bdev->bd_bdi;
1636 if (bdi_congested(bdi, bdi_bits)) {
1637 ret = 1;
1638 break;
1639 }
1640 }
1641 rcu_read_unlock();
1642 return ret;
1643}
1644
1645/*
1646 * called by the kthread helper functions to finally call the bio end_io
1647 * functions. This is where read checksum verification actually happens
1648 */
1649static void end_workqueue_fn(struct btrfs_work *work)
1650{
1651 struct bio *bio;
1652 struct btrfs_end_io_wq *end_io_wq;
1653
1654 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1655 bio = end_io_wq->bio;
1656
1657 bio->bi_status = end_io_wq->status;
1658 bio->bi_private = end_io_wq->private;
1659 bio->bi_end_io = end_io_wq->end_io;
1660 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1661 bio_endio(bio);
1662}
1663
1664static int cleaner_kthread(void *arg)
1665{
1666 struct btrfs_root *root = arg;
1667 struct btrfs_fs_info *fs_info = root->fs_info;
1668 int again;
1669
1670 while (1) {
1671 again = 0;
1672
1673 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1674
1675 /* Make the cleaner go to sleep early. */
1676 if (btrfs_need_cleaner_sleep(fs_info))
1677 goto sleep;
1678
1679 /*
1680 * Do not do anything if we might cause open_ctree() to block
1681 * before we have finished mounting the filesystem.
1682 */
1683 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1684 goto sleep;
1685
1686 if (!mutex_trylock(&fs_info->cleaner_mutex))
1687 goto sleep;
1688
1689 /*
1690 * Avoid the problem that we change the status of the fs
1691 * during the above check and trylock.
1692 */
1693 if (btrfs_need_cleaner_sleep(fs_info)) {
1694 mutex_unlock(&fs_info->cleaner_mutex);
1695 goto sleep;
1696 }
1697
1698 btrfs_run_delayed_iputs(fs_info);
1699
1700 again = btrfs_clean_one_deleted_snapshot(root);
1701 mutex_unlock(&fs_info->cleaner_mutex);
1702
1703 /*
1704 * The defragger has dealt with the R/O remount and umount,
1705 * needn't do anything special here.
1706 */
1707 btrfs_run_defrag_inodes(fs_info);
1708
1709 /*
1710 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1711 * with relocation (btrfs_relocate_chunk) and relocation
1712 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1713 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1714 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1715 * unused block groups.
1716 */
1717 btrfs_delete_unused_bgs(fs_info);
1718sleep:
1719 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1720 if (kthread_should_park())
1721 kthread_parkme();
1722 if (kthread_should_stop())
1723 return 0;
1724 if (!again) {
1725 set_current_state(TASK_INTERRUPTIBLE);
1726 schedule();
1727 __set_current_state(TASK_RUNNING);
1728 }
1729 }
1730}
1731
1732static int transaction_kthread(void *arg)
1733{
1734 struct btrfs_root *root = arg;
1735 struct btrfs_fs_info *fs_info = root->fs_info;
1736 struct btrfs_trans_handle *trans;
1737 struct btrfs_transaction *cur;
1738 u64 transid;
1739 time64_t now;
1740 unsigned long delay;
1741 bool cannot_commit;
1742
1743 do {
1744 cannot_commit = false;
1745 delay = HZ * fs_info->commit_interval;
1746 mutex_lock(&fs_info->transaction_kthread_mutex);
1747
1748 spin_lock(&fs_info->trans_lock);
1749 cur = fs_info->running_transaction;
1750 if (!cur) {
1751 spin_unlock(&fs_info->trans_lock);
1752 goto sleep;
1753 }
1754
1755 now = ktime_get_seconds();
1756 if (cur->state < TRANS_STATE_BLOCKED &&
1757 !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1758 (now < cur->start_time ||
1759 now - cur->start_time < fs_info->commit_interval)) {
1760 spin_unlock(&fs_info->trans_lock);
1761 delay = HZ * 5;
1762 goto sleep;
1763 }
1764 transid = cur->transid;
1765 spin_unlock(&fs_info->trans_lock);
1766
1767 /* If the file system is aborted, this will always fail. */
1768 trans = btrfs_attach_transaction(root);
1769 if (IS_ERR(trans)) {
1770 if (PTR_ERR(trans) != -ENOENT)
1771 cannot_commit = true;
1772 goto sleep;
1773 }
1774 if (transid == trans->transid) {
1775 btrfs_commit_transaction(trans);
1776 } else {
1777 btrfs_end_transaction(trans);
1778 }
1779sleep:
1780 wake_up_process(fs_info->cleaner_kthread);
1781 mutex_unlock(&fs_info->transaction_kthread_mutex);
1782
1783 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1784 &fs_info->fs_state)))
1785 btrfs_cleanup_transaction(fs_info);
1786 if (!kthread_should_stop() &&
1787 (!btrfs_transaction_blocked(fs_info) ||
1788 cannot_commit))
1789 schedule_timeout_interruptible(delay);
1790 } while (!kthread_should_stop());
1791 return 0;
1792}
1793
1794/*
1795 * this will find the highest generation in the array of
1796 * root backups. The index of the highest array is returned,
1797 * or -1 if we can't find anything.
1798 *
1799 * We check to make sure the array is valid by comparing the
1800 * generation of the latest root in the array with the generation
1801 * in the super block. If they don't match we pitch it.
1802 */
1803static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1804{
1805 u64 cur;
1806 int newest_index = -1;
1807 struct btrfs_root_backup *root_backup;
1808 int i;
1809
1810 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1811 root_backup = info->super_copy->super_roots + i;
1812 cur = btrfs_backup_tree_root_gen(root_backup);
1813 if (cur == newest_gen)
1814 newest_index = i;
1815 }
1816
1817 /* check to see if we actually wrapped around */
1818 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1819 root_backup = info->super_copy->super_roots;
1820 cur = btrfs_backup_tree_root_gen(root_backup);
1821 if (cur == newest_gen)
1822 newest_index = 0;
1823 }
1824 return newest_index;
1825}
1826
1827
1828/*
1829 * find the oldest backup so we know where to store new entries
1830 * in the backup array. This will set the backup_root_index
1831 * field in the fs_info struct
1832 */
1833static void find_oldest_super_backup(struct btrfs_fs_info *info,
1834 u64 newest_gen)
1835{
1836 int newest_index = -1;
1837
1838 newest_index = find_newest_super_backup(info, newest_gen);
1839 /* if there was garbage in there, just move along */
1840 if (newest_index == -1) {
1841 info->backup_root_index = 0;
1842 } else {
1843 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1844 }
1845}
1846
1847/*
1848 * copy all the root pointers into the super backup array.
1849 * this will bump the backup pointer by one when it is
1850 * done
1851 */
1852static void backup_super_roots(struct btrfs_fs_info *info)
1853{
1854 int next_backup;
1855 struct btrfs_root_backup *root_backup;
1856 int last_backup;
1857
1858 next_backup = info->backup_root_index;
1859 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1860 BTRFS_NUM_BACKUP_ROOTS;
1861
1862 /*
1863 * just overwrite the last backup if we're at the same generation
1864 * this happens only at umount
1865 */
1866 root_backup = info->super_for_commit->super_roots + last_backup;
1867 if (btrfs_backup_tree_root_gen(root_backup) ==
1868 btrfs_header_generation(info->tree_root->node))
1869 next_backup = last_backup;
1870
1871 root_backup = info->super_for_commit->super_roots + next_backup;
1872
1873 /*
1874 * make sure all of our padding and empty slots get zero filled
1875 * regardless of which ones we use today
1876 */
1877 memset(root_backup, 0, sizeof(*root_backup));
1878
1879 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1880
1881 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1882 btrfs_set_backup_tree_root_gen(root_backup,
1883 btrfs_header_generation(info->tree_root->node));
1884
1885 btrfs_set_backup_tree_root_level(root_backup,
1886 btrfs_header_level(info->tree_root->node));
1887
1888 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1889 btrfs_set_backup_chunk_root_gen(root_backup,
1890 btrfs_header_generation(info->chunk_root->node));
1891 btrfs_set_backup_chunk_root_level(root_backup,
1892 btrfs_header_level(info->chunk_root->node));
1893
1894 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1895 btrfs_set_backup_extent_root_gen(root_backup,
1896 btrfs_header_generation(info->extent_root->node));
1897 btrfs_set_backup_extent_root_level(root_backup,
1898 btrfs_header_level(info->extent_root->node));
1899
1900 /*
1901 * we might commit during log recovery, which happens before we set
1902 * the fs_root. Make sure it is valid before we fill it in.
1903 */
1904 if (info->fs_root && info->fs_root->node) {
1905 btrfs_set_backup_fs_root(root_backup,
1906 info->fs_root->node->start);
1907 btrfs_set_backup_fs_root_gen(root_backup,
1908 btrfs_header_generation(info->fs_root->node));
1909 btrfs_set_backup_fs_root_level(root_backup,
1910 btrfs_header_level(info->fs_root->node));
1911 }
1912
1913 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1914 btrfs_set_backup_dev_root_gen(root_backup,
1915 btrfs_header_generation(info->dev_root->node));
1916 btrfs_set_backup_dev_root_level(root_backup,
1917 btrfs_header_level(info->dev_root->node));
1918
1919 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1920 btrfs_set_backup_csum_root_gen(root_backup,
1921 btrfs_header_generation(info->csum_root->node));
1922 btrfs_set_backup_csum_root_level(root_backup,
1923 btrfs_header_level(info->csum_root->node));
1924
1925 btrfs_set_backup_total_bytes(root_backup,
1926 btrfs_super_total_bytes(info->super_copy));
1927 btrfs_set_backup_bytes_used(root_backup,
1928 btrfs_super_bytes_used(info->super_copy));
1929 btrfs_set_backup_num_devices(root_backup,
1930 btrfs_super_num_devices(info->super_copy));
1931
1932 /*
1933 * if we don't copy this out to the super_copy, it won't get remembered
1934 * for the next commit
1935 */
1936 memcpy(&info->super_copy->super_roots,
1937 &info->super_for_commit->super_roots,
1938 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1939}
1940
1941/*
1942 * this copies info out of the root backup array and back into
1943 * the in-memory super block. It is meant to help iterate through
1944 * the array, so you send it the number of backups you've already
1945 * tried and the last backup index you used.
1946 *
1947 * this returns -1 when it has tried all the backups
1948 */
1949static noinline int next_root_backup(struct btrfs_fs_info *info,
1950 struct btrfs_super_block *super,
1951 int *num_backups_tried, int *backup_index)
1952{
1953 struct btrfs_root_backup *root_backup;
1954 int newest = *backup_index;
1955
1956 if (*num_backups_tried == 0) {
1957 u64 gen = btrfs_super_generation(super);
1958
1959 newest = find_newest_super_backup(info, gen);
1960 if (newest == -1)
1961 return -1;
1962
1963 *backup_index = newest;
1964 *num_backups_tried = 1;
1965 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1966 /* we've tried all the backups, all done */
1967 return -1;
1968 } else {
1969 /* jump to the next oldest backup */
1970 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1971 BTRFS_NUM_BACKUP_ROOTS;
1972 *backup_index = newest;
1973 *num_backups_tried += 1;
1974 }
1975 root_backup = super->super_roots + newest;
1976
1977 btrfs_set_super_generation(super,
1978 btrfs_backup_tree_root_gen(root_backup));
1979 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1980 btrfs_set_super_root_level(super,
1981 btrfs_backup_tree_root_level(root_backup));
1982 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1983
1984 /*
1985 * fixme: the total bytes and num_devices need to match or we should
1986 * need a fsck
1987 */
1988 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1989 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1990 return 0;
1991}
1992
1993/* helper to cleanup workers */
1994static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1995{
1996 btrfs_destroy_workqueue(fs_info->fixup_workers);
1997 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1998 btrfs_destroy_workqueue(fs_info->workers);
1999 btrfs_destroy_workqueue(fs_info->endio_workers);
2000 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2001 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2002 btrfs_destroy_workqueue(fs_info->rmw_workers);
2003 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2004 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2005 btrfs_destroy_workqueue(fs_info->submit_workers);
2006 btrfs_destroy_workqueue(fs_info->delayed_workers);
2007 btrfs_destroy_workqueue(fs_info->caching_workers);
2008 btrfs_destroy_workqueue(fs_info->readahead_workers);
2009 btrfs_destroy_workqueue(fs_info->flush_workers);
2010 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2011 /*
2012 * Now that all other work queues are destroyed, we can safely destroy
2013 * the queues used for metadata I/O, since tasks from those other work
2014 * queues can do metadata I/O operations.
2015 */
2016 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2017 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2018}
2019
2020static void free_root_extent_buffers(struct btrfs_root *root)
2021{
2022 if (root) {
2023 free_extent_buffer(root->node);
2024 free_extent_buffer(root->commit_root);
2025 root->node = NULL;
2026 root->commit_root = NULL;
2027 }
2028}
2029
2030/* helper to cleanup tree roots */
2031static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2032{
2033 free_root_extent_buffers(info->tree_root);
2034
2035 free_root_extent_buffers(info->dev_root);
2036 free_root_extent_buffers(info->extent_root);
2037 free_root_extent_buffers(info->csum_root);
2038 free_root_extent_buffers(info->quota_root);
2039 free_root_extent_buffers(info->uuid_root);
2040 if (chunk_root)
2041 free_root_extent_buffers(info->chunk_root);
2042 free_root_extent_buffers(info->free_space_root);
2043}
2044
2045void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2046{
2047 int ret;
2048 struct btrfs_root *gang[8];
2049 int i;
2050
2051 while (!list_empty(&fs_info->dead_roots)) {
2052 gang[0] = list_entry(fs_info->dead_roots.next,
2053 struct btrfs_root, root_list);
2054 list_del(&gang[0]->root_list);
2055
2056 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2057 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2058 } else {
2059 free_extent_buffer(gang[0]->node);
2060 free_extent_buffer(gang[0]->commit_root);
2061 btrfs_put_fs_root(gang[0]);
2062 }
2063 }
2064
2065 while (1) {
2066 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2067 (void **)gang, 0,
2068 ARRAY_SIZE(gang));
2069 if (!ret)
2070 break;
2071 for (i = 0; i < ret; i++)
2072 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2073 }
2074
2075 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2076 btrfs_free_log_root_tree(NULL, fs_info);
2077 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2078 }
2079}
2080
2081static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2082{
2083 mutex_init(&fs_info->scrub_lock);
2084 atomic_set(&fs_info->scrubs_running, 0);
2085 atomic_set(&fs_info->scrub_pause_req, 0);
2086 atomic_set(&fs_info->scrubs_paused, 0);
2087 atomic_set(&fs_info->scrub_cancel_req, 0);
2088 init_waitqueue_head(&fs_info->scrub_pause_wait);
2089 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2090}
2091
2092static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2093{
2094 spin_lock_init(&fs_info->balance_lock);
2095 mutex_init(&fs_info->balance_mutex);
2096 atomic_set(&fs_info->balance_pause_req, 0);
2097 atomic_set(&fs_info->balance_cancel_req, 0);
2098 fs_info->balance_ctl = NULL;
2099 init_waitqueue_head(&fs_info->balance_wait_q);
2100}
2101
2102static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2103{
2104 struct inode *inode = fs_info->btree_inode;
2105
2106 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2107 set_nlink(inode, 1);
2108 /*
2109 * we set the i_size on the btree inode to the max possible int.
2110 * the real end of the address space is determined by all of
2111 * the devices in the system
2112 */
2113 inode->i_size = OFFSET_MAX;
2114 inode->i_mapping->a_ops = &btree_aops;
2115
2116 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2117 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2118 IO_TREE_INODE_IO, inode);
2119 BTRFS_I(inode)->io_tree.track_uptodate = false;
2120 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2121
2122 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2123
2124 BTRFS_I(inode)->root = fs_info->tree_root;
2125 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2126 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2127 btrfs_insert_inode_hash(inode);
2128}
2129
2130static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2131{
2132 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2133 init_rwsem(&fs_info->dev_replace.rwsem);
2134 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2135}
2136
2137static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2138{
2139 spin_lock_init(&fs_info->qgroup_lock);
2140 mutex_init(&fs_info->qgroup_ioctl_lock);
2141 fs_info->qgroup_tree = RB_ROOT;
2142 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2143 fs_info->qgroup_seq = 1;
2144 fs_info->qgroup_ulist = NULL;
2145 fs_info->qgroup_rescan_running = false;
2146 mutex_init(&fs_info->qgroup_rescan_lock);
2147}
2148
2149static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2150 struct btrfs_fs_devices *fs_devices)
2151{
2152 u32 max_active = fs_info->thread_pool_size;
2153 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2154
2155 fs_info->workers =
2156 btrfs_alloc_workqueue(fs_info, "worker",
2157 flags | WQ_HIGHPRI, max_active, 16);
2158
2159 fs_info->delalloc_workers =
2160 btrfs_alloc_workqueue(fs_info, "delalloc",
2161 flags, max_active, 2);
2162
2163 fs_info->flush_workers =
2164 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2165 flags, max_active, 0);
2166
2167 fs_info->caching_workers =
2168 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2169
2170 /*
2171 * a higher idle thresh on the submit workers makes it much more
2172 * likely that bios will be send down in a sane order to the
2173 * devices
2174 */
2175 fs_info->submit_workers =
2176 btrfs_alloc_workqueue(fs_info, "submit", flags,
2177 min_t(u64, fs_devices->num_devices,
2178 max_active), 64);
2179
2180 fs_info->fixup_workers =
2181 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2182
2183 /*
2184 * endios are largely parallel and should have a very
2185 * low idle thresh
2186 */
2187 fs_info->endio_workers =
2188 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2189 fs_info->endio_meta_workers =
2190 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2191 max_active, 4);
2192 fs_info->endio_meta_write_workers =
2193 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2194 max_active, 2);
2195 fs_info->endio_raid56_workers =
2196 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2197 max_active, 4);
2198 fs_info->endio_repair_workers =
2199 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2200 fs_info->rmw_workers =
2201 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2202 fs_info->endio_write_workers =
2203 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2204 max_active, 2);
2205 fs_info->endio_freespace_worker =
2206 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2207 max_active, 0);
2208 fs_info->delayed_workers =
2209 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2210 max_active, 0);
2211 fs_info->readahead_workers =
2212 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2213 max_active, 2);
2214 fs_info->qgroup_rescan_workers =
2215 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2216
2217 if (!(fs_info->workers && fs_info->delalloc_workers &&
2218 fs_info->submit_workers && fs_info->flush_workers &&
2219 fs_info->endio_workers && fs_info->endio_meta_workers &&
2220 fs_info->endio_meta_write_workers &&
2221 fs_info->endio_repair_workers &&
2222 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2223 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2224 fs_info->caching_workers && fs_info->readahead_workers &&
2225 fs_info->fixup_workers && fs_info->delayed_workers &&
2226 fs_info->qgroup_rescan_workers)) {
2227 return -ENOMEM;
2228 }
2229
2230 return 0;
2231}
2232
2233static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2234{
2235 struct crypto_shash *csum_shash;
2236 const char *csum_name = btrfs_super_csum_name(csum_type);
2237
2238 csum_shash = crypto_alloc_shash(csum_name, 0, 0);
2239
2240 if (IS_ERR(csum_shash)) {
2241 btrfs_err(fs_info, "error allocating %s hash for checksum",
2242 csum_name);
2243 return PTR_ERR(csum_shash);
2244 }
2245
2246 fs_info->csum_shash = csum_shash;
2247
2248 return 0;
2249}
2250
2251static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
2252{
2253 crypto_free_shash(fs_info->csum_shash);
2254}
2255
2256static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2257 struct btrfs_fs_devices *fs_devices)
2258{
2259 int ret;
2260 struct btrfs_root *log_tree_root;
2261 struct btrfs_super_block *disk_super = fs_info->super_copy;
2262 u64 bytenr = btrfs_super_log_root(disk_super);
2263 int level = btrfs_super_log_root_level(disk_super);
2264
2265 if (fs_devices->rw_devices == 0) {
2266 btrfs_warn(fs_info, "log replay required on RO media");
2267 return -EIO;
2268 }
2269
2270 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2271 if (!log_tree_root)
2272 return -ENOMEM;
2273
2274 __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2275
2276 log_tree_root->node = read_tree_block(fs_info, bytenr,
2277 fs_info->generation + 1,
2278 level, NULL);
2279 if (IS_ERR(log_tree_root->node)) {
2280 btrfs_warn(fs_info, "failed to read log tree");
2281 ret = PTR_ERR(log_tree_root->node);
2282 kfree(log_tree_root);
2283 return ret;
2284 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2285 btrfs_err(fs_info, "failed to read log tree");
2286 free_extent_buffer(log_tree_root->node);
2287 kfree(log_tree_root);
2288 return -EIO;
2289 }
2290 /* returns with log_tree_root freed on success */
2291 ret = btrfs_recover_log_trees(log_tree_root);
2292 if (ret) {
2293 btrfs_handle_fs_error(fs_info, ret,
2294 "Failed to recover log tree");
2295 free_extent_buffer(log_tree_root->node);
2296 kfree(log_tree_root);
2297 return ret;
2298 }
2299
2300 if (sb_rdonly(fs_info->sb)) {
2301 ret = btrfs_commit_super(fs_info);
2302 if (ret)
2303 return ret;
2304 }
2305
2306 return 0;
2307}
2308
2309static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2310{
2311 struct btrfs_root *tree_root = fs_info->tree_root;
2312 struct btrfs_root *root;
2313 struct btrfs_key location;
2314 int ret;
2315
2316 BUG_ON(!fs_info->tree_root);
2317
2318 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2319 location.type = BTRFS_ROOT_ITEM_KEY;
2320 location.offset = 0;
2321
2322 root = btrfs_read_tree_root(tree_root, &location);
2323 if (IS_ERR(root)) {
2324 ret = PTR_ERR(root);
2325 goto out;
2326 }
2327 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2328 fs_info->extent_root = root;
2329
2330 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2331 root = btrfs_read_tree_root(tree_root, &location);
2332 if (IS_ERR(root)) {
2333 ret = PTR_ERR(root);
2334 goto out;
2335 }
2336 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2337 fs_info->dev_root = root;
2338 btrfs_init_devices_late(fs_info);
2339
2340 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2341 root = btrfs_read_tree_root(tree_root, &location);
2342 if (IS_ERR(root)) {
2343 ret = PTR_ERR(root);
2344 goto out;
2345 }
2346 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2347 fs_info->csum_root = root;
2348
2349 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2350 root = btrfs_read_tree_root(tree_root, &location);
2351 if (!IS_ERR(root)) {
2352 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2353 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2354 fs_info->quota_root = root;
2355 }
2356
2357 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2358 root = btrfs_read_tree_root(tree_root, &location);
2359 if (IS_ERR(root)) {
2360 ret = PTR_ERR(root);
2361 if (ret != -ENOENT)
2362 goto out;
2363 } else {
2364 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2365 fs_info->uuid_root = root;
2366 }
2367
2368 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2369 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2370 root = btrfs_read_tree_root(tree_root, &location);
2371 if (IS_ERR(root)) {
2372 ret = PTR_ERR(root);
2373 goto out;
2374 }
2375 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2376 fs_info->free_space_root = root;
2377 }
2378
2379 return 0;
2380out:
2381 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2382 location.objectid, ret);
2383 return ret;
2384}
2385
2386/*
2387 * Real super block validation
2388 * NOTE: super csum type and incompat features will not be checked here.
2389 *
2390 * @sb: super block to check
2391 * @mirror_num: the super block number to check its bytenr:
2392 * 0 the primary (1st) sb
2393 * 1, 2 2nd and 3rd backup copy
2394 * -1 skip bytenr check
2395 */
2396static int validate_super(struct btrfs_fs_info *fs_info,
2397 struct btrfs_super_block *sb, int mirror_num)
2398{
2399 u64 nodesize = btrfs_super_nodesize(sb);
2400 u64 sectorsize = btrfs_super_sectorsize(sb);
2401 int ret = 0;
2402
2403 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2404 btrfs_err(fs_info, "no valid FS found");
2405 ret = -EINVAL;
2406 }
2407 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2408 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2409 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2410 ret = -EINVAL;
2411 }
2412 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2413 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2414 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2415 ret = -EINVAL;
2416 }
2417 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2418 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2419 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2420 ret = -EINVAL;
2421 }
2422 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2423 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2424 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2425 ret = -EINVAL;
2426 }
2427
2428 /*
2429 * Check sectorsize and nodesize first, other check will need it.
2430 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2431 */
2432 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2433 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2434 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2435 ret = -EINVAL;
2436 }
2437 /* Only PAGE SIZE is supported yet */
2438 if (sectorsize != PAGE_SIZE) {
2439 btrfs_err(fs_info,
2440 "sectorsize %llu not supported yet, only support %lu",
2441 sectorsize, PAGE_SIZE);
2442 ret = -EINVAL;
2443 }
2444 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2445 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2446 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2447 ret = -EINVAL;
2448 }
2449 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2450 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2451 le32_to_cpu(sb->__unused_leafsize), nodesize);
2452 ret = -EINVAL;
2453 }
2454
2455 /* Root alignment check */
2456 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2457 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2458 btrfs_super_root(sb));
2459 ret = -EINVAL;
2460 }
2461 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2462 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2463 btrfs_super_chunk_root(sb));
2464 ret = -EINVAL;
2465 }
2466 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2467 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2468 btrfs_super_log_root(sb));
2469 ret = -EINVAL;
2470 }
2471
2472 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2473 BTRFS_FSID_SIZE) != 0) {
2474 btrfs_err(fs_info,
2475 "dev_item UUID does not match metadata fsid: %pU != %pU",
2476 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2477 ret = -EINVAL;
2478 }
2479
2480 /*
2481 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2482 * done later
2483 */
2484 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2485 btrfs_err(fs_info, "bytes_used is too small %llu",
2486 btrfs_super_bytes_used(sb));
2487 ret = -EINVAL;
2488 }
2489 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2490 btrfs_err(fs_info, "invalid stripesize %u",
2491 btrfs_super_stripesize(sb));
2492 ret = -EINVAL;
2493 }
2494 if (btrfs_super_num_devices(sb) > (1UL << 31))
2495 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2496 btrfs_super_num_devices(sb));
2497 if (btrfs_super_num_devices(sb) == 0) {
2498 btrfs_err(fs_info, "number of devices is 0");
2499 ret = -EINVAL;
2500 }
2501
2502 if (mirror_num >= 0 &&
2503 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2504 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2505 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2506 ret = -EINVAL;
2507 }
2508
2509 /*
2510 * Obvious sys_chunk_array corruptions, it must hold at least one key
2511 * and one chunk
2512 */
2513 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2514 btrfs_err(fs_info, "system chunk array too big %u > %u",
2515 btrfs_super_sys_array_size(sb),
2516 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2517 ret = -EINVAL;
2518 }
2519 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2520 + sizeof(struct btrfs_chunk)) {
2521 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2522 btrfs_super_sys_array_size(sb),
2523 sizeof(struct btrfs_disk_key)
2524 + sizeof(struct btrfs_chunk));
2525 ret = -EINVAL;
2526 }
2527
2528 /*
2529 * The generation is a global counter, we'll trust it more than the others
2530 * but it's still possible that it's the one that's wrong.
2531 */
2532 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2533 btrfs_warn(fs_info,
2534 "suspicious: generation < chunk_root_generation: %llu < %llu",
2535 btrfs_super_generation(sb),
2536 btrfs_super_chunk_root_generation(sb));
2537 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2538 && btrfs_super_cache_generation(sb) != (u64)-1)
2539 btrfs_warn(fs_info,
2540 "suspicious: generation < cache_generation: %llu < %llu",
2541 btrfs_super_generation(sb),
2542 btrfs_super_cache_generation(sb));
2543
2544 return ret;
2545}
2546
2547/*
2548 * Validation of super block at mount time.
2549 * Some checks already done early at mount time, like csum type and incompat
2550 * flags will be skipped.
2551 */
2552static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2553{
2554 return validate_super(fs_info, fs_info->super_copy, 0);
2555}
2556
2557/*
2558 * Validation of super block at write time.
2559 * Some checks like bytenr check will be skipped as their values will be
2560 * overwritten soon.
2561 * Extra checks like csum type and incompat flags will be done here.
2562 */
2563static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2564 struct btrfs_super_block *sb)
2565{
2566 int ret;
2567
2568 ret = validate_super(fs_info, sb, -1);
2569 if (ret < 0)
2570 goto out;
2571 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2572 ret = -EUCLEAN;
2573 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2574 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2575 goto out;
2576 }
2577 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2578 ret = -EUCLEAN;
2579 btrfs_err(fs_info,
2580 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2581 btrfs_super_incompat_flags(sb),
2582 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2583 goto out;
2584 }
2585out:
2586 if (ret < 0)
2587 btrfs_err(fs_info,
2588 "super block corruption detected before writing it to disk");
2589 return ret;
2590}
2591
2592int open_ctree(struct super_block *sb,
2593 struct btrfs_fs_devices *fs_devices,
2594 char *options)
2595{
2596 u32 sectorsize;
2597 u32 nodesize;
2598 u32 stripesize;
2599 u64 generation;
2600 u64 features;
2601 u16 csum_type;
2602 struct btrfs_key location;
2603 struct buffer_head *bh;
2604 struct btrfs_super_block *disk_super;
2605 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2606 struct btrfs_root *tree_root;
2607 struct btrfs_root *chunk_root;
2608 int ret;
2609 int err = -EINVAL;
2610 int num_backups_tried = 0;
2611 int backup_index = 0;
2612 int clear_free_space_tree = 0;
2613 int level;
2614
2615 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2616 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2617 if (!tree_root || !chunk_root) {
2618 err = -ENOMEM;
2619 goto fail;
2620 }
2621
2622 ret = init_srcu_struct(&fs_info->subvol_srcu);
2623 if (ret) {
2624 err = ret;
2625 goto fail;
2626 }
2627
2628 ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2629 if (ret) {
2630 err = ret;
2631 goto fail_srcu;
2632 }
2633
2634 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2635 if (ret) {
2636 err = ret;
2637 goto fail_dio_bytes;
2638 }
2639 fs_info->dirty_metadata_batch = PAGE_SIZE *
2640 (1 + ilog2(nr_cpu_ids));
2641
2642 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2643 if (ret) {
2644 err = ret;
2645 goto fail_dirty_metadata_bytes;
2646 }
2647
2648 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2649 GFP_KERNEL);
2650 if (ret) {
2651 err = ret;
2652 goto fail_delalloc_bytes;
2653 }
2654
2655 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2656 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2657 INIT_LIST_HEAD(&fs_info->trans_list);
2658 INIT_LIST_HEAD(&fs_info->dead_roots);
2659 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2660 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2661 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2662 spin_lock_init(&fs_info->delalloc_root_lock);
2663 spin_lock_init(&fs_info->trans_lock);
2664 spin_lock_init(&fs_info->fs_roots_radix_lock);
2665 spin_lock_init(&fs_info->delayed_iput_lock);
2666 spin_lock_init(&fs_info->defrag_inodes_lock);
2667 spin_lock_init(&fs_info->tree_mod_seq_lock);
2668 spin_lock_init(&fs_info->super_lock);
2669 spin_lock_init(&fs_info->buffer_lock);
2670 spin_lock_init(&fs_info->unused_bgs_lock);
2671 rwlock_init(&fs_info->tree_mod_log_lock);
2672 mutex_init(&fs_info->unused_bg_unpin_mutex);
2673 mutex_init(&fs_info->delete_unused_bgs_mutex);
2674 mutex_init(&fs_info->reloc_mutex);
2675 mutex_init(&fs_info->delalloc_root_mutex);
2676 seqlock_init(&fs_info->profiles_lock);
2677
2678 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2679 INIT_LIST_HEAD(&fs_info->space_info);
2680 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2681 INIT_LIST_HEAD(&fs_info->unused_bgs);
2682 extent_map_tree_init(&fs_info->mapping_tree);
2683 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2684 BTRFS_BLOCK_RSV_GLOBAL);
2685 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2686 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2687 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2688 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2689 BTRFS_BLOCK_RSV_DELOPS);
2690 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2691 BTRFS_BLOCK_RSV_DELREFS);
2692
2693 atomic_set(&fs_info->async_delalloc_pages, 0);
2694 atomic_set(&fs_info->defrag_running, 0);
2695 atomic_set(&fs_info->reada_works_cnt, 0);
2696 atomic_set(&fs_info->nr_delayed_iputs, 0);
2697 atomic64_set(&fs_info->tree_mod_seq, 0);
2698 fs_info->sb = sb;
2699 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2700 fs_info->metadata_ratio = 0;
2701 fs_info->defrag_inodes = RB_ROOT;
2702 atomic64_set(&fs_info->free_chunk_space, 0);
2703 fs_info->tree_mod_log = RB_ROOT;
2704 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2705 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2706 /* readahead state */
2707 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2708 spin_lock_init(&fs_info->reada_lock);
2709 btrfs_init_ref_verify(fs_info);
2710
2711 fs_info->thread_pool_size = min_t(unsigned long,
2712 num_online_cpus() + 2, 8);
2713
2714 INIT_LIST_HEAD(&fs_info->ordered_roots);
2715 spin_lock_init(&fs_info->ordered_root_lock);
2716
2717 fs_info->btree_inode = new_inode(sb);
2718 if (!fs_info->btree_inode) {
2719 err = -ENOMEM;
2720 goto fail_bio_counter;
2721 }
2722 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2723
2724 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2725 GFP_KERNEL);
2726 if (!fs_info->delayed_root) {
2727 err = -ENOMEM;
2728 goto fail_iput;
2729 }
2730 btrfs_init_delayed_root(fs_info->delayed_root);
2731
2732 btrfs_init_scrub(fs_info);
2733#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2734 fs_info->check_integrity_print_mask = 0;
2735#endif
2736 btrfs_init_balance(fs_info);
2737 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2738
2739 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2740 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2741
2742 btrfs_init_btree_inode(fs_info);
2743
2744 spin_lock_init(&fs_info->block_group_cache_lock);
2745 fs_info->block_group_cache_tree = RB_ROOT;
2746 fs_info->first_logical_byte = (u64)-1;
2747
2748 extent_io_tree_init(fs_info, &fs_info->freed_extents[0],
2749 IO_TREE_FS_INFO_FREED_EXTENTS0, NULL);
2750 extent_io_tree_init(fs_info, &fs_info->freed_extents[1],
2751 IO_TREE_FS_INFO_FREED_EXTENTS1, NULL);
2752 fs_info->pinned_extents = &fs_info->freed_extents[0];
2753 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2754
2755 mutex_init(&fs_info->ordered_operations_mutex);
2756 mutex_init(&fs_info->tree_log_mutex);
2757 mutex_init(&fs_info->chunk_mutex);
2758 mutex_init(&fs_info->transaction_kthread_mutex);
2759 mutex_init(&fs_info->cleaner_mutex);
2760 mutex_init(&fs_info->ro_block_group_mutex);
2761 init_rwsem(&fs_info->commit_root_sem);
2762 init_rwsem(&fs_info->cleanup_work_sem);
2763 init_rwsem(&fs_info->subvol_sem);
2764 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2765
2766 btrfs_init_dev_replace_locks(fs_info);
2767 btrfs_init_qgroup(fs_info);
2768
2769 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2770 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2771
2772 init_waitqueue_head(&fs_info->transaction_throttle);
2773 init_waitqueue_head(&fs_info->transaction_wait);
2774 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2775 init_waitqueue_head(&fs_info->async_submit_wait);
2776 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2777
2778 /* Usable values until the real ones are cached from the superblock */
2779 fs_info->nodesize = 4096;
2780 fs_info->sectorsize = 4096;
2781 fs_info->stripesize = 4096;
2782
2783 spin_lock_init(&fs_info->swapfile_pins_lock);
2784 fs_info->swapfile_pins = RB_ROOT;
2785
2786 fs_info->send_in_progress = 0;
2787
2788 ret = btrfs_alloc_stripe_hash_table(fs_info);
2789 if (ret) {
2790 err = ret;
2791 goto fail_alloc;
2792 }
2793
2794 __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2795
2796 invalidate_bdev(fs_devices->latest_bdev);
2797
2798 /*
2799 * Read super block and check the signature bytes only
2800 */
2801 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2802 if (IS_ERR(bh)) {
2803 err = PTR_ERR(bh);
2804 goto fail_alloc;
2805 }
2806
2807 /*
2808 * Verify the type first, if that or the the checksum value are
2809 * corrupted, we'll find out
2810 */
2811 csum_type = btrfs_super_csum_type((struct btrfs_super_block *)bh->b_data);
2812 if (!btrfs_supported_super_csum(csum_type)) {
2813 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2814 csum_type);
2815 err = -EINVAL;
2816 brelse(bh);
2817 goto fail_alloc;
2818 }
2819
2820 ret = btrfs_init_csum_hash(fs_info, csum_type);
2821 if (ret) {
2822 err = ret;
2823 goto fail_alloc;
2824 }
2825
2826 /*
2827 * We want to check superblock checksum, the type is stored inside.
2828 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2829 */
2830 if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2831 btrfs_err(fs_info, "superblock checksum mismatch");
2832 err = -EINVAL;
2833 brelse(bh);
2834 goto fail_csum;
2835 }
2836
2837 /*
2838 * super_copy is zeroed at allocation time and we never touch the
2839 * following bytes up to INFO_SIZE, the checksum is calculated from
2840 * the whole block of INFO_SIZE
2841 */
2842 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2843 brelse(bh);
2844
2845 disk_super = fs_info->super_copy;
2846
2847 ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2848 BTRFS_FSID_SIZE));
2849
2850 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2851 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2852 fs_info->super_copy->metadata_uuid,
2853 BTRFS_FSID_SIZE));
2854 }
2855
2856 features = btrfs_super_flags(disk_super);
2857 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2858 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2859 btrfs_set_super_flags(disk_super, features);
2860 btrfs_info(fs_info,
2861 "found metadata UUID change in progress flag, clearing");
2862 }
2863
2864 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2865 sizeof(*fs_info->super_for_commit));
2866
2867 ret = btrfs_validate_mount_super(fs_info);
2868 if (ret) {
2869 btrfs_err(fs_info, "superblock contains fatal errors");
2870 err = -EINVAL;
2871 goto fail_csum;
2872 }
2873
2874 if (!btrfs_super_root(disk_super))
2875 goto fail_csum;
2876
2877 /* check FS state, whether FS is broken. */
2878 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2879 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2880
2881 /*
2882 * run through our array of backup supers and setup
2883 * our ring pointer to the oldest one
2884 */
2885 generation = btrfs_super_generation(disk_super);
2886 find_oldest_super_backup(fs_info, generation);
2887
2888 /*
2889 * In the long term, we'll store the compression type in the super
2890 * block, and it'll be used for per file compression control.
2891 */
2892 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2893
2894 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2895 if (ret) {
2896 err = ret;
2897 goto fail_csum;
2898 }
2899
2900 features = btrfs_super_incompat_flags(disk_super) &
2901 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2902 if (features) {
2903 btrfs_err(fs_info,
2904 "cannot mount because of unsupported optional features (%llx)",
2905 features);
2906 err = -EINVAL;
2907 goto fail_csum;
2908 }
2909
2910 features = btrfs_super_incompat_flags(disk_super);
2911 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2912 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2913 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2914 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2915 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2916
2917 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2918 btrfs_info(fs_info, "has skinny extents");
2919
2920 /*
2921 * flag our filesystem as having big metadata blocks if
2922 * they are bigger than the page size
2923 */
2924 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2925 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2926 btrfs_info(fs_info,
2927 "flagging fs with big metadata feature");
2928 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2929 }
2930
2931 nodesize = btrfs_super_nodesize(disk_super);
2932 sectorsize = btrfs_super_sectorsize(disk_super);
2933 stripesize = sectorsize;
2934 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2935 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2936
2937 /* Cache block sizes */
2938 fs_info->nodesize = nodesize;
2939 fs_info->sectorsize = sectorsize;
2940 fs_info->stripesize = stripesize;
2941
2942 /*
2943 * mixed block groups end up with duplicate but slightly offset
2944 * extent buffers for the same range. It leads to corruptions
2945 */
2946 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2947 (sectorsize != nodesize)) {
2948 btrfs_err(fs_info,
2949"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2950 nodesize, sectorsize);
2951 goto fail_csum;
2952 }
2953
2954 /*
2955 * Needn't use the lock because there is no other task which will
2956 * update the flag.
2957 */
2958 btrfs_set_super_incompat_flags(disk_super, features);
2959
2960 features = btrfs_super_compat_ro_flags(disk_super) &
2961 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2962 if (!sb_rdonly(sb) && features) {
2963 btrfs_err(fs_info,
2964 "cannot mount read-write because of unsupported optional features (%llx)",
2965 features);
2966 err = -EINVAL;
2967 goto fail_csum;
2968 }
2969
2970 ret = btrfs_init_workqueues(fs_info, fs_devices);
2971 if (ret) {
2972 err = ret;
2973 goto fail_sb_buffer;
2974 }
2975
2976 sb->s_bdi->congested_fn = btrfs_congested_fn;
2977 sb->s_bdi->congested_data = fs_info;
2978 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2979 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
2980 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2981 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2982
2983 sb->s_blocksize = sectorsize;
2984 sb->s_blocksize_bits = blksize_bits(sectorsize);
2985 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
2986
2987 mutex_lock(&fs_info->chunk_mutex);
2988 ret = btrfs_read_sys_array(fs_info);
2989 mutex_unlock(&fs_info->chunk_mutex);
2990 if (ret) {
2991 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2992 goto fail_sb_buffer;
2993 }
2994
2995 generation = btrfs_super_chunk_root_generation(disk_super);
2996 level = btrfs_super_chunk_root_level(disk_super);
2997
2998 __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2999
3000 chunk_root->node = read_tree_block(fs_info,
3001 btrfs_super_chunk_root(disk_super),
3002 generation, level, NULL);
3003 if (IS_ERR(chunk_root->node) ||
3004 !extent_buffer_uptodate(chunk_root->node)) {
3005 btrfs_err(fs_info, "failed to read chunk root");
3006 if (!IS_ERR(chunk_root->node))
3007 free_extent_buffer(chunk_root->node);
3008 chunk_root->node = NULL;
3009 goto fail_tree_roots;
3010 }
3011 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3012 chunk_root->commit_root = btrfs_root_node(chunk_root);
3013
3014 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3015 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
3016
3017 ret = btrfs_read_chunk_tree(fs_info);
3018 if (ret) {
3019 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3020 goto fail_tree_roots;
3021 }
3022
3023 /*
3024 * Keep the devid that is marked to be the target device for the
3025 * device replace procedure
3026 */
3027 btrfs_free_extra_devids(fs_devices, 0);
3028
3029 if (!fs_devices->latest_bdev) {
3030 btrfs_err(fs_info, "failed to read devices");
3031 goto fail_tree_roots;
3032 }
3033
3034retry_root_backup:
3035 generation = btrfs_super_generation(disk_super);
3036 level = btrfs_super_root_level(disk_super);
3037
3038 tree_root->node = read_tree_block(fs_info,
3039 btrfs_super_root(disk_super),
3040 generation, level, NULL);
3041 if (IS_ERR(tree_root->node) ||
3042 !extent_buffer_uptodate(tree_root->node)) {
3043 btrfs_warn(fs_info, "failed to read tree root");
3044 if (!IS_ERR(tree_root->node))
3045 free_extent_buffer(tree_root->node);
3046 tree_root->node = NULL;
3047 goto recovery_tree_root;
3048 }
3049
3050 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3051 tree_root->commit_root = btrfs_root_node(tree_root);
3052 btrfs_set_root_refs(&tree_root->root_item, 1);
3053
3054 mutex_lock(&tree_root->objectid_mutex);
3055 ret = btrfs_find_highest_objectid(tree_root,
3056 &tree_root->highest_objectid);
3057 if (ret) {
3058 mutex_unlock(&tree_root->objectid_mutex);
3059 goto recovery_tree_root;
3060 }
3061
3062 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3063
3064 mutex_unlock(&tree_root->objectid_mutex);
3065
3066 ret = btrfs_read_roots(fs_info);
3067 if (ret)
3068 goto recovery_tree_root;
3069
3070 fs_info->generation = generation;
3071 fs_info->last_trans_committed = generation;
3072
3073 ret = btrfs_verify_dev_extents(fs_info);
3074 if (ret) {
3075 btrfs_err(fs_info,
3076 "failed to verify dev extents against chunks: %d",
3077 ret);
3078 goto fail_block_groups;
3079 }
3080 ret = btrfs_recover_balance(fs_info);
3081 if (ret) {
3082 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3083 goto fail_block_groups;
3084 }
3085
3086 ret = btrfs_init_dev_stats(fs_info);
3087 if (ret) {
3088 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3089 goto fail_block_groups;
3090 }
3091
3092 ret = btrfs_init_dev_replace(fs_info);
3093 if (ret) {
3094 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3095 goto fail_block_groups;
3096 }
3097
3098 btrfs_free_extra_devids(fs_devices, 1);
3099
3100 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3101 if (ret) {
3102 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3103 ret);
3104 goto fail_block_groups;
3105 }
3106
3107 ret = btrfs_sysfs_add_device(fs_devices);
3108 if (ret) {
3109 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3110 ret);
3111 goto fail_fsdev_sysfs;
3112 }
3113
3114 ret = btrfs_sysfs_add_mounted(fs_info);
3115 if (ret) {
3116 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3117 goto fail_fsdev_sysfs;
3118 }
3119
3120 ret = btrfs_init_space_info(fs_info);
3121 if (ret) {
3122 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3123 goto fail_sysfs;
3124 }
3125
3126 ret = btrfs_read_block_groups(fs_info);
3127 if (ret) {
3128 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3129 goto fail_sysfs;
3130 }
3131
3132 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3133 btrfs_warn(fs_info,
3134 "writable mount is not allowed due to too many missing devices");
3135 goto fail_sysfs;
3136 }
3137
3138 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3139 "btrfs-cleaner");
3140 if (IS_ERR(fs_info->cleaner_kthread))
3141 goto fail_sysfs;
3142
3143 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3144 tree_root,
3145 "btrfs-transaction");
3146 if (IS_ERR(fs_info->transaction_kthread))
3147 goto fail_cleaner;
3148
3149 if (!btrfs_test_opt(fs_info, NOSSD) &&
3150 !fs_info->fs_devices->rotating) {
3151 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3152 }
3153
3154 /*
3155 * Mount does not set all options immediately, we can do it now and do
3156 * not have to wait for transaction commit
3157 */
3158 btrfs_apply_pending_changes(fs_info);
3159
3160#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3161 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3162 ret = btrfsic_mount(fs_info, fs_devices,
3163 btrfs_test_opt(fs_info,
3164 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3165 1 : 0,
3166 fs_info->check_integrity_print_mask);
3167 if (ret)
3168 btrfs_warn(fs_info,
3169 "failed to initialize integrity check module: %d",
3170 ret);
3171 }
3172#endif
3173 ret = btrfs_read_qgroup_config(fs_info);
3174 if (ret)
3175 goto fail_trans_kthread;
3176
3177 if (btrfs_build_ref_tree(fs_info))
3178 btrfs_err(fs_info, "couldn't build ref tree");
3179
3180 /* do not make disk changes in broken FS or nologreplay is given */
3181 if (btrfs_super_log_root(disk_super) != 0 &&
3182 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3183 ret = btrfs_replay_log(fs_info, fs_devices);
3184 if (ret) {
3185 err = ret;
3186 goto fail_qgroup;
3187 }
3188 }
3189
3190 ret = btrfs_find_orphan_roots(fs_info);
3191 if (ret)
3192 goto fail_qgroup;
3193
3194 if (!sb_rdonly(sb)) {
3195 ret = btrfs_cleanup_fs_roots(fs_info);
3196 if (ret)
3197 goto fail_qgroup;
3198
3199 mutex_lock(&fs_info->cleaner_mutex);
3200 ret = btrfs_recover_relocation(tree_root);
3201 mutex_unlock(&fs_info->cleaner_mutex);
3202 if (ret < 0) {
3203 btrfs_warn(fs_info, "failed to recover relocation: %d",
3204 ret);
3205 err = -EINVAL;
3206 goto fail_qgroup;
3207 }
3208 }
3209
3210 location.objectid = BTRFS_FS_TREE_OBJECTID;
3211 location.type = BTRFS_ROOT_ITEM_KEY;
3212 location.offset = 0;
3213
3214 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3215 if (IS_ERR(fs_info->fs_root)) {
3216 err = PTR_ERR(fs_info->fs_root);
3217 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3218 goto fail_qgroup;
3219 }
3220
3221 if (sb_rdonly(sb))
3222 return 0;
3223
3224 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3225 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3226 clear_free_space_tree = 1;
3227 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3228 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3229 btrfs_warn(fs_info, "free space tree is invalid");
3230 clear_free_space_tree = 1;
3231 }
3232
3233 if (clear_free_space_tree) {
3234 btrfs_info(fs_info, "clearing free space tree");
3235 ret = btrfs_clear_free_space_tree(fs_info);
3236 if (ret) {
3237 btrfs_warn(fs_info,
3238 "failed to clear free space tree: %d", ret);
3239 close_ctree(fs_info);
3240 return ret;
3241 }
3242 }
3243
3244 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3245 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3246 btrfs_info(fs_info, "creating free space tree");
3247 ret = btrfs_create_free_space_tree(fs_info);
3248 if (ret) {
3249 btrfs_warn(fs_info,
3250 "failed to create free space tree: %d", ret);
3251 close_ctree(fs_info);
3252 return ret;
3253 }
3254 }
3255
3256 down_read(&fs_info->cleanup_work_sem);
3257 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3258 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3259 up_read(&fs_info->cleanup_work_sem);
3260 close_ctree(fs_info);
3261 return ret;
3262 }
3263 up_read(&fs_info->cleanup_work_sem);
3264
3265 ret = btrfs_resume_balance_async(fs_info);
3266 if (ret) {
3267 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3268 close_ctree(fs_info);
3269 return ret;
3270 }
3271
3272 ret = btrfs_resume_dev_replace_async(fs_info);
3273 if (ret) {
3274 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3275 close_ctree(fs_info);
3276 return ret;
3277 }
3278
3279 btrfs_qgroup_rescan_resume(fs_info);
3280
3281 if (!fs_info->uuid_root) {
3282 btrfs_info(fs_info, "creating UUID tree");
3283 ret = btrfs_create_uuid_tree(fs_info);
3284 if (ret) {
3285 btrfs_warn(fs_info,
3286 "failed to create the UUID tree: %d", ret);
3287 close_ctree(fs_info);
3288 return ret;
3289 }
3290 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3291 fs_info->generation !=
3292 btrfs_super_uuid_tree_generation(disk_super)) {
3293 btrfs_info(fs_info, "checking UUID tree");
3294 ret = btrfs_check_uuid_tree(fs_info);
3295 if (ret) {
3296 btrfs_warn(fs_info,
3297 "failed to check the UUID tree: %d", ret);
3298 close_ctree(fs_info);
3299 return ret;
3300 }
3301 } else {
3302 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3303 }
3304 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3305
3306 /*
3307 * backuproot only affect mount behavior, and if open_ctree succeeded,
3308 * no need to keep the flag
3309 */
3310 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3311
3312 return 0;
3313
3314fail_qgroup:
3315 btrfs_free_qgroup_config(fs_info);
3316fail_trans_kthread:
3317 kthread_stop(fs_info->transaction_kthread);
3318 btrfs_cleanup_transaction(fs_info);
3319 btrfs_free_fs_roots(fs_info);
3320fail_cleaner:
3321 kthread_stop(fs_info->cleaner_kthread);
3322
3323 /*
3324 * make sure we're done with the btree inode before we stop our
3325 * kthreads
3326 */
3327 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3328
3329fail_sysfs:
3330 btrfs_sysfs_remove_mounted(fs_info);
3331
3332fail_fsdev_sysfs:
3333 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3334
3335fail_block_groups:
3336 btrfs_put_block_group_cache(fs_info);
3337
3338fail_tree_roots:
3339 free_root_pointers(fs_info, 1);
3340 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3341
3342fail_sb_buffer:
3343 btrfs_stop_all_workers(fs_info);
3344 btrfs_free_block_groups(fs_info);
3345fail_csum:
3346 btrfs_free_csum_hash(fs_info);
3347fail_alloc:
3348fail_iput:
3349 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3350
3351 iput(fs_info->btree_inode);
3352fail_bio_counter:
3353 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
3354fail_delalloc_bytes:
3355 percpu_counter_destroy(&fs_info->delalloc_bytes);
3356fail_dirty_metadata_bytes:
3357 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3358fail_dio_bytes:
3359 percpu_counter_destroy(&fs_info->dio_bytes);
3360fail_srcu:
3361 cleanup_srcu_struct(&fs_info->subvol_srcu);
3362fail:
3363 btrfs_free_stripe_hash_table(fs_info);
3364 btrfs_close_devices(fs_info->fs_devices);
3365 return err;
3366
3367recovery_tree_root:
3368 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3369 goto fail_tree_roots;
3370
3371 free_root_pointers(fs_info, 0);
3372
3373 /* don't use the log in recovery mode, it won't be valid */
3374 btrfs_set_super_log_root(disk_super, 0);
3375
3376 /* we can't trust the free space cache either */
3377 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3378
3379 ret = next_root_backup(fs_info, fs_info->super_copy,
3380 &num_backups_tried, &backup_index);
3381 if (ret == -1)
3382 goto fail_block_groups;
3383 goto retry_root_backup;
3384}
3385ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3386
3387static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3388{
3389 if (uptodate) {
3390 set_buffer_uptodate(bh);
3391 } else {
3392 struct btrfs_device *device = (struct btrfs_device *)
3393 bh->b_private;
3394
3395 btrfs_warn_rl_in_rcu(device->fs_info,
3396 "lost page write due to IO error on %s",
3397 rcu_str_deref(device->name));
3398 /* note, we don't set_buffer_write_io_error because we have
3399 * our own ways of dealing with the IO errors
3400 */
3401 clear_buffer_uptodate(bh);
3402 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3403 }
3404 unlock_buffer(bh);
3405 put_bh(bh);
3406}
3407
3408int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3409 struct buffer_head **bh_ret)
3410{
3411 struct buffer_head *bh;
3412 struct btrfs_super_block *super;
3413 u64 bytenr;
3414
3415 bytenr = btrfs_sb_offset(copy_num);
3416 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3417 return -EINVAL;
3418
3419 bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3420 /*
3421 * If we fail to read from the underlying devices, as of now
3422 * the best option we have is to mark it EIO.
3423 */
3424 if (!bh)
3425 return -EIO;
3426
3427 super = (struct btrfs_super_block *)bh->b_data;
3428 if (btrfs_super_bytenr(super) != bytenr ||
3429 btrfs_super_magic(super) != BTRFS_MAGIC) {
3430 brelse(bh);
3431 return -EINVAL;
3432 }
3433
3434 *bh_ret = bh;
3435 return 0;
3436}
3437
3438
3439struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3440{
3441 struct buffer_head *bh;
3442 struct buffer_head *latest = NULL;
3443 struct btrfs_super_block *super;
3444 int i;
3445 u64 transid = 0;
3446 int ret = -EINVAL;
3447
3448 /* we would like to check all the supers, but that would make
3449 * a btrfs mount succeed after a mkfs from a different FS.
3450 * So, we need to add a special mount option to scan for
3451 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3452 */
3453 for (i = 0; i < 1; i++) {
3454 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3455 if (ret)
3456 continue;
3457
3458 super = (struct btrfs_super_block *)bh->b_data;
3459
3460 if (!latest || btrfs_super_generation(super) > transid) {
3461 brelse(latest);
3462 latest = bh;
3463 transid = btrfs_super_generation(super);
3464 } else {
3465 brelse(bh);
3466 }
3467 }
3468
3469 if (!latest)
3470 return ERR_PTR(ret);
3471
3472 return latest;
3473}
3474
3475/*
3476 * Write superblock @sb to the @device. Do not wait for completion, all the
3477 * buffer heads we write are pinned.
3478 *
3479 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3480 * the expected device size at commit time. Note that max_mirrors must be
3481 * same for write and wait phases.
3482 *
3483 * Return number of errors when buffer head is not found or submission fails.
3484 */
3485static int write_dev_supers(struct btrfs_device *device,
3486 struct btrfs_super_block *sb, int max_mirrors)
3487{
3488 struct btrfs_fs_info *fs_info = device->fs_info;
3489 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3490 struct buffer_head *bh;
3491 int i;
3492 int ret;
3493 int errors = 0;
3494 u64 bytenr;
3495 int op_flags;
3496
3497 if (max_mirrors == 0)
3498 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3499
3500 shash->tfm = fs_info->csum_shash;
3501
3502 for (i = 0; i < max_mirrors; i++) {
3503 bytenr = btrfs_sb_offset(i);
3504 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3505 device->commit_total_bytes)
3506 break;
3507
3508 btrfs_set_super_bytenr(sb, bytenr);
3509
3510 crypto_shash_init(shash);
3511 crypto_shash_update(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3512 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3513 crypto_shash_final(shash, sb->csum);
3514
3515 /* One reference for us, and we leave it for the caller */
3516 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3517 BTRFS_SUPER_INFO_SIZE);
3518 if (!bh) {
3519 btrfs_err(device->fs_info,
3520 "couldn't get super buffer head for bytenr %llu",
3521 bytenr);
3522 errors++;
3523 continue;
3524 }
3525
3526 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3527
3528 /* one reference for submit_bh */
3529 get_bh(bh);
3530
3531 set_buffer_uptodate(bh);
3532 lock_buffer(bh);
3533 bh->b_end_io = btrfs_end_buffer_write_sync;
3534 bh->b_private = device;
3535
3536 /*
3537 * we fua the first super. The others we allow
3538 * to go down lazy.
3539 */
3540 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3541 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3542 op_flags |= REQ_FUA;
3543 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3544 if (ret)
3545 errors++;
3546 }
3547 return errors < i ? 0 : -1;
3548}
3549
3550/*
3551 * Wait for write completion of superblocks done by write_dev_supers,
3552 * @max_mirrors same for write and wait phases.
3553 *
3554 * Return number of errors when buffer head is not found or not marked up to
3555 * date.
3556 */
3557static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3558{
3559 struct buffer_head *bh;
3560 int i;
3561 int errors = 0;
3562 bool primary_failed = false;
3563 u64 bytenr;
3564
3565 if (max_mirrors == 0)
3566 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3567
3568 for (i = 0; i < max_mirrors; i++) {
3569 bytenr = btrfs_sb_offset(i);
3570 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3571 device->commit_total_bytes)
3572 break;
3573
3574 bh = __find_get_block(device->bdev,
3575 bytenr / BTRFS_BDEV_BLOCKSIZE,
3576 BTRFS_SUPER_INFO_SIZE);
3577 if (!bh) {
3578 errors++;
3579 if (i == 0)
3580 primary_failed = true;
3581 continue;
3582 }
3583 wait_on_buffer(bh);
3584 if (!buffer_uptodate(bh)) {
3585 errors++;
3586 if (i == 0)
3587 primary_failed = true;
3588 }
3589
3590 /* drop our reference */
3591 brelse(bh);
3592
3593 /* drop the reference from the writing run */
3594 brelse(bh);
3595 }
3596
3597 /* log error, force error return */
3598 if (primary_failed) {
3599 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3600 device->devid);
3601 return -1;
3602 }
3603
3604 return errors < i ? 0 : -1;
3605}
3606
3607/*
3608 * endio for the write_dev_flush, this will wake anyone waiting
3609 * for the barrier when it is done
3610 */
3611static void btrfs_end_empty_barrier(struct bio *bio)
3612{
3613 complete(bio->bi_private);
3614}
3615
3616/*
3617 * Submit a flush request to the device if it supports it. Error handling is
3618 * done in the waiting counterpart.
3619 */
3620static void write_dev_flush(struct btrfs_device *device)
3621{
3622 struct request_queue *q = bdev_get_queue(device->bdev);
3623 struct bio *bio = device->flush_bio;
3624
3625 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3626 return;
3627
3628 bio_reset(bio);
3629 bio->bi_end_io = btrfs_end_empty_barrier;
3630 bio_set_dev(bio, device->bdev);
3631 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3632 init_completion(&device->flush_wait);
3633 bio->bi_private = &device->flush_wait;
3634
3635 btrfsic_submit_bio(bio);
3636 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3637}
3638
3639/*
3640 * If the flush bio has been submitted by write_dev_flush, wait for it.
3641 */
3642static blk_status_t wait_dev_flush(struct btrfs_device *device)
3643{
3644 struct bio *bio = device->flush_bio;
3645
3646 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3647 return BLK_STS_OK;
3648
3649 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3650 wait_for_completion_io(&device->flush_wait);
3651
3652 return bio->bi_status;
3653}
3654
3655static int check_barrier_error(struct btrfs_fs_info *fs_info)
3656{
3657 if (!btrfs_check_rw_degradable(fs_info, NULL))
3658 return -EIO;
3659 return 0;
3660}
3661
3662/*
3663 * send an empty flush down to each device in parallel,
3664 * then wait for them
3665 */
3666static int barrier_all_devices(struct btrfs_fs_info *info)
3667{
3668 struct list_head *head;
3669 struct btrfs_device *dev;
3670 int errors_wait = 0;
3671 blk_status_t ret;
3672
3673 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3674 /* send down all the barriers */
3675 head = &info->fs_devices->devices;
3676 list_for_each_entry(dev, head, dev_list) {
3677 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3678 continue;
3679 if (!dev->bdev)
3680 continue;
3681 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3682 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3683 continue;
3684
3685 write_dev_flush(dev);
3686 dev->last_flush_error = BLK_STS_OK;
3687 }
3688
3689 /* wait for all the barriers */
3690 list_for_each_entry(dev, head, dev_list) {
3691 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3692 continue;
3693 if (!dev->bdev) {
3694 errors_wait++;
3695 continue;
3696 }
3697 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3698 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3699 continue;
3700
3701 ret = wait_dev_flush(dev);
3702 if (ret) {
3703 dev->last_flush_error = ret;
3704 btrfs_dev_stat_inc_and_print(dev,
3705 BTRFS_DEV_STAT_FLUSH_ERRS);
3706 errors_wait++;
3707 }
3708 }
3709
3710 if (errors_wait) {
3711 /*
3712 * At some point we need the status of all disks
3713 * to arrive at the volume status. So error checking
3714 * is being pushed to a separate loop.
3715 */
3716 return check_barrier_error(info);
3717 }
3718 return 0;
3719}
3720
3721int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3722{
3723 int raid_type;
3724 int min_tolerated = INT_MAX;
3725
3726 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3727 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3728 min_tolerated = min_t(int, min_tolerated,
3729 btrfs_raid_array[BTRFS_RAID_SINGLE].
3730 tolerated_failures);
3731
3732 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3733 if (raid_type == BTRFS_RAID_SINGLE)
3734 continue;
3735 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3736 continue;
3737 min_tolerated = min_t(int, min_tolerated,
3738 btrfs_raid_array[raid_type].
3739 tolerated_failures);
3740 }
3741
3742 if (min_tolerated == INT_MAX) {
3743 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3744 min_tolerated = 0;
3745 }
3746
3747 return min_tolerated;
3748}
3749
3750int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3751{
3752 struct list_head *head;
3753 struct btrfs_device *dev;
3754 struct btrfs_super_block *sb;
3755 struct btrfs_dev_item *dev_item;
3756 int ret;
3757 int do_barriers;
3758 int max_errors;
3759 int total_errors = 0;
3760 u64 flags;
3761
3762 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3763
3764 /*
3765 * max_mirrors == 0 indicates we're from commit_transaction,
3766 * not from fsync where the tree roots in fs_info have not
3767 * been consistent on disk.
3768 */
3769 if (max_mirrors == 0)
3770 backup_super_roots(fs_info);
3771
3772 sb = fs_info->super_for_commit;
3773 dev_item = &sb->dev_item;
3774
3775 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3776 head = &fs_info->fs_devices->devices;
3777 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3778
3779 if (do_barriers) {
3780 ret = barrier_all_devices(fs_info);
3781 if (ret) {
3782 mutex_unlock(
3783 &fs_info->fs_devices->device_list_mutex);
3784 btrfs_handle_fs_error(fs_info, ret,
3785 "errors while submitting device barriers.");
3786 return ret;
3787 }
3788 }
3789
3790 list_for_each_entry(dev, head, dev_list) {
3791 if (!dev->bdev) {
3792 total_errors++;
3793 continue;
3794 }
3795 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3796 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3797 continue;
3798
3799 btrfs_set_stack_device_generation(dev_item, 0);
3800 btrfs_set_stack_device_type(dev_item, dev->type);
3801 btrfs_set_stack_device_id(dev_item, dev->devid);
3802 btrfs_set_stack_device_total_bytes(dev_item,
3803 dev->commit_total_bytes);
3804 btrfs_set_stack_device_bytes_used(dev_item,
3805 dev->commit_bytes_used);
3806 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3807 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3808 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3809 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3810 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3811 BTRFS_FSID_SIZE);
3812
3813 flags = btrfs_super_flags(sb);
3814 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3815
3816 ret = btrfs_validate_write_super(fs_info, sb);
3817 if (ret < 0) {
3818 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3819 btrfs_handle_fs_error(fs_info, -EUCLEAN,
3820 "unexpected superblock corruption detected");
3821 return -EUCLEAN;
3822 }
3823
3824 ret = write_dev_supers(dev, sb, max_mirrors);
3825 if (ret)
3826 total_errors++;
3827 }
3828 if (total_errors > max_errors) {
3829 btrfs_err(fs_info, "%d errors while writing supers",
3830 total_errors);
3831 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3832
3833 /* FUA is masked off if unsupported and can't be the reason */
3834 btrfs_handle_fs_error(fs_info, -EIO,
3835 "%d errors while writing supers",
3836 total_errors);
3837 return -EIO;
3838 }
3839
3840 total_errors = 0;
3841 list_for_each_entry(dev, head, dev_list) {
3842 if (!dev->bdev)
3843 continue;
3844 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3845 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3846 continue;
3847
3848 ret = wait_dev_supers(dev, max_mirrors);
3849 if (ret)
3850 total_errors++;
3851 }
3852 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3853 if (total_errors > max_errors) {
3854 btrfs_handle_fs_error(fs_info, -EIO,
3855 "%d errors while writing supers",
3856 total_errors);
3857 return -EIO;
3858 }
3859 return 0;
3860}
3861
3862/* Drop a fs root from the radix tree and free it. */
3863void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3864 struct btrfs_root *root)
3865{
3866 spin_lock(&fs_info->fs_roots_radix_lock);
3867 radix_tree_delete(&fs_info->fs_roots_radix,
3868 (unsigned long)root->root_key.objectid);
3869 spin_unlock(&fs_info->fs_roots_radix_lock);
3870
3871 if (btrfs_root_refs(&root->root_item) == 0)
3872 synchronize_srcu(&fs_info->subvol_srcu);
3873
3874 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3875 btrfs_free_log(NULL, root);
3876 if (root->reloc_root) {
3877 free_extent_buffer(root->reloc_root->node);
3878 free_extent_buffer(root->reloc_root->commit_root);
3879 btrfs_put_fs_root(root->reloc_root);
3880 root->reloc_root = NULL;
3881 }
3882 }
3883
3884 if (root->free_ino_pinned)
3885 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3886 if (root->free_ino_ctl)
3887 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3888 btrfs_free_fs_root(root);
3889}
3890
3891void btrfs_free_fs_root(struct btrfs_root *root)
3892{
3893 iput(root->ino_cache_inode);
3894 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3895 if (root->anon_dev)
3896 free_anon_bdev(root->anon_dev);
3897 if (root->subv_writers)
3898 btrfs_free_subvolume_writers(root->subv_writers);
3899 free_extent_buffer(root->node);
3900 free_extent_buffer(root->commit_root);
3901 kfree(root->free_ino_ctl);
3902 kfree(root->free_ino_pinned);
3903 btrfs_put_fs_root(root);
3904}
3905
3906int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3907{
3908 u64 root_objectid = 0;
3909 struct btrfs_root *gang[8];
3910 int i = 0;
3911 int err = 0;
3912 unsigned int ret = 0;
3913 int index;
3914
3915 while (1) {
3916 index = srcu_read_lock(&fs_info->subvol_srcu);
3917 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3918 (void **)gang, root_objectid,
3919 ARRAY_SIZE(gang));
3920 if (!ret) {
3921 srcu_read_unlock(&fs_info->subvol_srcu, index);
3922 break;
3923 }
3924 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3925
3926 for (i = 0; i < ret; i++) {
3927 /* Avoid to grab roots in dead_roots */
3928 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3929 gang[i] = NULL;
3930 continue;
3931 }
3932 /* grab all the search result for later use */
3933 gang[i] = btrfs_grab_fs_root(gang[i]);
3934 }
3935 srcu_read_unlock(&fs_info->subvol_srcu, index);
3936
3937 for (i = 0; i < ret; i++) {
3938 if (!gang[i])
3939 continue;
3940 root_objectid = gang[i]->root_key.objectid;
3941 err = btrfs_orphan_cleanup(gang[i]);
3942 if (err)
3943 break;
3944 btrfs_put_fs_root(gang[i]);
3945 }
3946 root_objectid++;
3947 }
3948
3949 /* release the uncleaned roots due to error */
3950 for (; i < ret; i++) {
3951 if (gang[i])
3952 btrfs_put_fs_root(gang[i]);
3953 }
3954 return err;
3955}
3956
3957int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3958{
3959 struct btrfs_root *root = fs_info->tree_root;
3960 struct btrfs_trans_handle *trans;
3961
3962 mutex_lock(&fs_info->cleaner_mutex);
3963 btrfs_run_delayed_iputs(fs_info);
3964 mutex_unlock(&fs_info->cleaner_mutex);
3965 wake_up_process(fs_info->cleaner_kthread);
3966
3967 /* wait until ongoing cleanup work done */
3968 down_write(&fs_info->cleanup_work_sem);
3969 up_write(&fs_info->cleanup_work_sem);
3970
3971 trans = btrfs_join_transaction(root);
3972 if (IS_ERR(trans))
3973 return PTR_ERR(trans);
3974 return btrfs_commit_transaction(trans);
3975}
3976
3977void close_ctree(struct btrfs_fs_info *fs_info)
3978{
3979 int ret;
3980
3981 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3982 /*
3983 * We don't want the cleaner to start new transactions, add more delayed
3984 * iputs, etc. while we're closing. We can't use kthread_stop() yet
3985 * because that frees the task_struct, and the transaction kthread might
3986 * still try to wake up the cleaner.
3987 */
3988 kthread_park(fs_info->cleaner_kthread);
3989
3990 /* wait for the qgroup rescan worker to stop */
3991 btrfs_qgroup_wait_for_completion(fs_info, false);
3992
3993 /* wait for the uuid_scan task to finish */
3994 down(&fs_info->uuid_tree_rescan_sem);
3995 /* avoid complains from lockdep et al., set sem back to initial state */
3996 up(&fs_info->uuid_tree_rescan_sem);
3997
3998 /* pause restriper - we want to resume on mount */
3999 btrfs_pause_balance(fs_info);
4000
4001 btrfs_dev_replace_suspend_for_unmount(fs_info);
4002
4003 btrfs_scrub_cancel(fs_info);
4004
4005 /* wait for any defraggers to finish */
4006 wait_event(fs_info->transaction_wait,
4007 (atomic_read(&fs_info->defrag_running) == 0));
4008
4009 /* clear out the rbtree of defraggable inodes */
4010 btrfs_cleanup_defrag_inodes(fs_info);
4011
4012 cancel_work_sync(&fs_info->async_reclaim_work);
4013
4014 if (!sb_rdonly(fs_info->sb)) {
4015 /*
4016 * The cleaner kthread is stopped, so do one final pass over
4017 * unused block groups.
4018 */
4019 btrfs_delete_unused_bgs(fs_info);
4020
4021 ret = btrfs_commit_super(fs_info);
4022 if (ret)
4023 btrfs_err(fs_info, "commit super ret %d", ret);
4024 }
4025
4026 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4027 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4028 btrfs_error_commit_super(fs_info);
4029
4030 kthread_stop(fs_info->transaction_kthread);
4031 kthread_stop(fs_info->cleaner_kthread);
4032
4033 ASSERT(list_empty(&fs_info->delayed_iputs));
4034 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4035
4036 btrfs_free_qgroup_config(fs_info);
4037 ASSERT(list_empty(&fs_info->delalloc_roots));
4038
4039 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4040 btrfs_info(fs_info, "at unmount delalloc count %lld",
4041 percpu_counter_sum(&fs_info->delalloc_bytes));
4042 }
4043
4044 if (percpu_counter_sum(&fs_info->dio_bytes))
4045 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4046 percpu_counter_sum(&fs_info->dio_bytes));
4047
4048 btrfs_sysfs_remove_mounted(fs_info);
4049 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4050
4051 btrfs_free_fs_roots(fs_info);
4052
4053 btrfs_put_block_group_cache(fs_info);
4054
4055 /*
4056 * we must make sure there is not any read request to
4057 * submit after we stopping all workers.
4058 */
4059 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4060 btrfs_stop_all_workers(fs_info);
4061
4062 btrfs_free_block_groups(fs_info);
4063
4064 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4065 free_root_pointers(fs_info, 1);
4066
4067 iput(fs_info->btree_inode);
4068
4069#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4070 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4071 btrfsic_unmount(fs_info->fs_devices);
4072#endif
4073
4074 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4075 btrfs_close_devices(fs_info->fs_devices);
4076
4077 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4078 percpu_counter_destroy(&fs_info->delalloc_bytes);
4079 percpu_counter_destroy(&fs_info->dio_bytes);
4080 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
4081 cleanup_srcu_struct(&fs_info->subvol_srcu);
4082
4083 btrfs_free_csum_hash(fs_info);
4084 btrfs_free_stripe_hash_table(fs_info);
4085 btrfs_free_ref_cache(fs_info);
4086}
4087
4088int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4089 int atomic)
4090{
4091 int ret;
4092 struct inode *btree_inode = buf->pages[0]->mapping->host;
4093
4094 ret = extent_buffer_uptodate(buf);
4095 if (!ret)
4096 return ret;
4097
4098 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4099 parent_transid, atomic);
4100 if (ret == -EAGAIN)
4101 return ret;
4102 return !ret;
4103}
4104
4105void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4106{
4107 struct btrfs_fs_info *fs_info;
4108 struct btrfs_root *root;
4109 u64 transid = btrfs_header_generation(buf);
4110 int was_dirty;
4111
4112#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4113 /*
4114 * This is a fast path so only do this check if we have sanity tests
4115 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4116 * outside of the sanity tests.
4117 */
4118 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4119 return;
4120#endif
4121 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4122 fs_info = root->fs_info;
4123 btrfs_assert_tree_locked(buf);
4124 if (transid != fs_info->generation)
4125 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4126 buf->start, transid, fs_info->generation);
4127 was_dirty = set_extent_buffer_dirty(buf);
4128 if (!was_dirty)
4129 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4130 buf->len,
4131 fs_info->dirty_metadata_batch);
4132#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4133 /*
4134 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4135 * but item data not updated.
4136 * So here we should only check item pointers, not item data.
4137 */
4138 if (btrfs_header_level(buf) == 0 &&
4139 btrfs_check_leaf_relaxed(buf)) {
4140 btrfs_print_leaf(buf);
4141 ASSERT(0);
4142 }
4143#endif
4144}
4145
4146static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4147 int flush_delayed)
4148{
4149 /*
4150 * looks as though older kernels can get into trouble with
4151 * this code, they end up stuck in balance_dirty_pages forever
4152 */
4153 int ret;
4154
4155 if (current->flags & PF_MEMALLOC)
4156 return;
4157
4158 if (flush_delayed)
4159 btrfs_balance_delayed_items(fs_info);
4160
4161 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4162 BTRFS_DIRTY_METADATA_THRESH,
4163 fs_info->dirty_metadata_batch);
4164 if (ret > 0) {
4165 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4166 }
4167}
4168
4169void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4170{
4171 __btrfs_btree_balance_dirty(fs_info, 1);
4172}
4173
4174void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4175{
4176 __btrfs_btree_balance_dirty(fs_info, 0);
4177}
4178
4179int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4180 struct btrfs_key *first_key)
4181{
4182 return btree_read_extent_buffer_pages(buf, parent_transid,
4183 level, first_key);
4184}
4185
4186static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4187{
4188 /* cleanup FS via transaction */
4189 btrfs_cleanup_transaction(fs_info);
4190
4191 mutex_lock(&fs_info->cleaner_mutex);
4192 btrfs_run_delayed_iputs(fs_info);
4193 mutex_unlock(&fs_info->cleaner_mutex);
4194
4195 down_write(&fs_info->cleanup_work_sem);
4196 up_write(&fs_info->cleanup_work_sem);
4197}
4198
4199static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4200{
4201 struct btrfs_ordered_extent *ordered;
4202
4203 spin_lock(&root->ordered_extent_lock);
4204 /*
4205 * This will just short circuit the ordered completion stuff which will
4206 * make sure the ordered extent gets properly cleaned up.
4207 */
4208 list_for_each_entry(ordered, &root->ordered_extents,
4209 root_extent_list)
4210 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4211 spin_unlock(&root->ordered_extent_lock);
4212}
4213
4214static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4215{
4216 struct btrfs_root *root;
4217 struct list_head splice;
4218
4219 INIT_LIST_HEAD(&splice);
4220
4221 spin_lock(&fs_info->ordered_root_lock);
4222 list_splice_init(&fs_info->ordered_roots, &splice);
4223 while (!list_empty(&splice)) {
4224 root = list_first_entry(&splice, struct btrfs_root,
4225 ordered_root);
4226 list_move_tail(&root->ordered_root,
4227 &fs_info->ordered_roots);
4228
4229 spin_unlock(&fs_info->ordered_root_lock);
4230 btrfs_destroy_ordered_extents(root);
4231
4232 cond_resched();
4233 spin_lock(&fs_info->ordered_root_lock);
4234 }
4235 spin_unlock(&fs_info->ordered_root_lock);
4236
4237 /*
4238 * We need this here because if we've been flipped read-only we won't
4239 * get sync() from the umount, so we need to make sure any ordered
4240 * extents that haven't had their dirty pages IO start writeout yet
4241 * actually get run and error out properly.
4242 */
4243 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4244}
4245
4246static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4247 struct btrfs_fs_info *fs_info)
4248{
4249 struct rb_node *node;
4250 struct btrfs_delayed_ref_root *delayed_refs;
4251 struct btrfs_delayed_ref_node *ref;
4252 int ret = 0;
4253
4254 delayed_refs = &trans->delayed_refs;
4255
4256 spin_lock(&delayed_refs->lock);
4257 if (atomic_read(&delayed_refs->num_entries) == 0) {
4258 spin_unlock(&delayed_refs->lock);
4259 btrfs_info(fs_info, "delayed_refs has NO entry");
4260 return ret;
4261 }
4262
4263 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4264 struct btrfs_delayed_ref_head *head;
4265 struct rb_node *n;
4266 bool pin_bytes = false;
4267
4268 head = rb_entry(node, struct btrfs_delayed_ref_head,
4269 href_node);
4270 if (btrfs_delayed_ref_lock(delayed_refs, head))
4271 continue;
4272
4273 spin_lock(&head->lock);
4274 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4275 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4276 ref_node);
4277 ref->in_tree = 0;
4278 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4279 RB_CLEAR_NODE(&ref->ref_node);
4280 if (!list_empty(&ref->add_list))
4281 list_del(&ref->add_list);
4282 atomic_dec(&delayed_refs->num_entries);
4283 btrfs_put_delayed_ref(ref);
4284 }
4285 if (head->must_insert_reserved)
4286 pin_bytes = true;
4287 btrfs_free_delayed_extent_op(head->extent_op);
4288 btrfs_delete_ref_head(delayed_refs, head);
4289 spin_unlock(&head->lock);
4290 spin_unlock(&delayed_refs->lock);
4291 mutex_unlock(&head->mutex);
4292
4293 if (pin_bytes)
4294 btrfs_pin_extent(fs_info, head->bytenr,
4295 head->num_bytes, 1);
4296 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4297 btrfs_put_delayed_ref_head(head);
4298 cond_resched();
4299 spin_lock(&delayed_refs->lock);
4300 }
4301
4302 spin_unlock(&delayed_refs->lock);
4303
4304 return ret;
4305}
4306
4307static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4308{
4309 struct btrfs_inode *btrfs_inode;
4310 struct list_head splice;
4311
4312 INIT_LIST_HEAD(&splice);
4313
4314 spin_lock(&root->delalloc_lock);
4315 list_splice_init(&root->delalloc_inodes, &splice);
4316
4317 while (!list_empty(&splice)) {
4318 struct inode *inode = NULL;
4319 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4320 delalloc_inodes);
4321 __btrfs_del_delalloc_inode(root, btrfs_inode);
4322 spin_unlock(&root->delalloc_lock);
4323
4324 /*
4325 * Make sure we get a live inode and that it'll not disappear
4326 * meanwhile.
4327 */
4328 inode = igrab(&btrfs_inode->vfs_inode);
4329 if (inode) {
4330 invalidate_inode_pages2(inode->i_mapping);
4331 iput(inode);
4332 }
4333 spin_lock(&root->delalloc_lock);
4334 }
4335 spin_unlock(&root->delalloc_lock);
4336}
4337
4338static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4339{
4340 struct btrfs_root *root;
4341 struct list_head splice;
4342
4343 INIT_LIST_HEAD(&splice);
4344
4345 spin_lock(&fs_info->delalloc_root_lock);
4346 list_splice_init(&fs_info->delalloc_roots, &splice);
4347 while (!list_empty(&splice)) {
4348 root = list_first_entry(&splice, struct btrfs_root,
4349 delalloc_root);
4350 root = btrfs_grab_fs_root(root);
4351 BUG_ON(!root);
4352 spin_unlock(&fs_info->delalloc_root_lock);
4353
4354 btrfs_destroy_delalloc_inodes(root);
4355 btrfs_put_fs_root(root);
4356
4357 spin_lock(&fs_info->delalloc_root_lock);
4358 }
4359 spin_unlock(&fs_info->delalloc_root_lock);
4360}
4361
4362static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4363 struct extent_io_tree *dirty_pages,
4364 int mark)
4365{
4366 int ret;
4367 struct extent_buffer *eb;
4368 u64 start = 0;
4369 u64 end;
4370
4371 while (1) {
4372 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4373 mark, NULL);
4374 if (ret)
4375 break;
4376
4377 clear_extent_bits(dirty_pages, start, end, mark);
4378 while (start <= end) {
4379 eb = find_extent_buffer(fs_info, start);
4380 start += fs_info->nodesize;
4381 if (!eb)
4382 continue;
4383 wait_on_extent_buffer_writeback(eb);
4384
4385 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4386 &eb->bflags))
4387 clear_extent_buffer_dirty(eb);
4388 free_extent_buffer_stale(eb);
4389 }
4390 }
4391
4392 return ret;
4393}
4394
4395static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4396 struct extent_io_tree *pinned_extents)
4397{
4398 struct extent_io_tree *unpin;
4399 u64 start;
4400 u64 end;
4401 int ret;
4402 bool loop = true;
4403
4404 unpin = pinned_extents;
4405again:
4406 while (1) {
4407 struct extent_state *cached_state = NULL;
4408
4409 /*
4410 * The btrfs_finish_extent_commit() may get the same range as
4411 * ours between find_first_extent_bit and clear_extent_dirty.
4412 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4413 * the same extent range.
4414 */
4415 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4416 ret = find_first_extent_bit(unpin, 0, &start, &end,
4417 EXTENT_DIRTY, &cached_state);
4418 if (ret) {
4419 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4420 break;
4421 }
4422
4423 clear_extent_dirty(unpin, start, end, &cached_state);
4424 free_extent_state(cached_state);
4425 btrfs_error_unpin_extent_range(fs_info, start, end);
4426 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4427 cond_resched();
4428 }
4429
4430 if (loop) {
4431 if (unpin == &fs_info->freed_extents[0])
4432 unpin = &fs_info->freed_extents[1];
4433 else
4434 unpin = &fs_info->freed_extents[0];
4435 loop = false;
4436 goto again;
4437 }
4438
4439 return 0;
4440}
4441
4442static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4443{
4444 struct inode *inode;
4445
4446 inode = cache->io_ctl.inode;
4447 if (inode) {
4448 invalidate_inode_pages2(inode->i_mapping);
4449 BTRFS_I(inode)->generation = 0;
4450 cache->io_ctl.inode = NULL;
4451 iput(inode);
4452 }
4453 btrfs_put_block_group(cache);
4454}
4455
4456void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4457 struct btrfs_fs_info *fs_info)
4458{
4459 struct btrfs_block_group_cache *cache;
4460
4461 spin_lock(&cur_trans->dirty_bgs_lock);
4462 while (!list_empty(&cur_trans->dirty_bgs)) {
4463 cache = list_first_entry(&cur_trans->dirty_bgs,
4464 struct btrfs_block_group_cache,
4465 dirty_list);
4466
4467 if (!list_empty(&cache->io_list)) {
4468 spin_unlock(&cur_trans->dirty_bgs_lock);
4469 list_del_init(&cache->io_list);
4470 btrfs_cleanup_bg_io(cache);
4471 spin_lock(&cur_trans->dirty_bgs_lock);
4472 }
4473
4474 list_del_init(&cache->dirty_list);
4475 spin_lock(&cache->lock);
4476 cache->disk_cache_state = BTRFS_DC_ERROR;
4477 spin_unlock(&cache->lock);
4478
4479 spin_unlock(&cur_trans->dirty_bgs_lock);
4480 btrfs_put_block_group(cache);
4481 btrfs_delayed_refs_rsv_release(fs_info, 1);
4482 spin_lock(&cur_trans->dirty_bgs_lock);
4483 }
4484 spin_unlock(&cur_trans->dirty_bgs_lock);
4485
4486 /*
4487 * Refer to the definition of io_bgs member for details why it's safe
4488 * to use it without any locking
4489 */
4490 while (!list_empty(&cur_trans->io_bgs)) {
4491 cache = list_first_entry(&cur_trans->io_bgs,
4492 struct btrfs_block_group_cache,
4493 io_list);
4494
4495 list_del_init(&cache->io_list);
4496 spin_lock(&cache->lock);
4497 cache->disk_cache_state = BTRFS_DC_ERROR;
4498 spin_unlock(&cache->lock);
4499 btrfs_cleanup_bg_io(cache);
4500 }
4501}
4502
4503void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4504 struct btrfs_fs_info *fs_info)
4505{
4506 struct btrfs_device *dev, *tmp;
4507
4508 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4509 ASSERT(list_empty(&cur_trans->dirty_bgs));
4510 ASSERT(list_empty(&cur_trans->io_bgs));
4511
4512 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4513 post_commit_list) {
4514 list_del_init(&dev->post_commit_list);
4515 }
4516
4517 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4518
4519 cur_trans->state = TRANS_STATE_COMMIT_START;
4520 wake_up(&fs_info->transaction_blocked_wait);
4521
4522 cur_trans->state = TRANS_STATE_UNBLOCKED;
4523 wake_up(&fs_info->transaction_wait);
4524
4525 btrfs_destroy_delayed_inodes(fs_info);
4526 btrfs_assert_delayed_root_empty(fs_info);
4527
4528 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4529 EXTENT_DIRTY);
4530 btrfs_destroy_pinned_extent(fs_info,
4531 fs_info->pinned_extents);
4532
4533 cur_trans->state =TRANS_STATE_COMPLETED;
4534 wake_up(&cur_trans->commit_wait);
4535}
4536
4537static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4538{
4539 struct btrfs_transaction *t;
4540
4541 mutex_lock(&fs_info->transaction_kthread_mutex);
4542
4543 spin_lock(&fs_info->trans_lock);
4544 while (!list_empty(&fs_info->trans_list)) {
4545 t = list_first_entry(&fs_info->trans_list,
4546 struct btrfs_transaction, list);
4547 if (t->state >= TRANS_STATE_COMMIT_START) {
4548 refcount_inc(&t->use_count);
4549 spin_unlock(&fs_info->trans_lock);
4550 btrfs_wait_for_commit(fs_info, t->transid);
4551 btrfs_put_transaction(t);
4552 spin_lock(&fs_info->trans_lock);
4553 continue;
4554 }
4555 if (t == fs_info->running_transaction) {
4556 t->state = TRANS_STATE_COMMIT_DOING;
4557 spin_unlock(&fs_info->trans_lock);
4558 /*
4559 * We wait for 0 num_writers since we don't hold a trans
4560 * handle open currently for this transaction.
4561 */
4562 wait_event(t->writer_wait,
4563 atomic_read(&t->num_writers) == 0);
4564 } else {
4565 spin_unlock(&fs_info->trans_lock);
4566 }
4567 btrfs_cleanup_one_transaction(t, fs_info);
4568
4569 spin_lock(&fs_info->trans_lock);
4570 if (t == fs_info->running_transaction)
4571 fs_info->running_transaction = NULL;
4572 list_del_init(&t->list);
4573 spin_unlock(&fs_info->trans_lock);
4574
4575 btrfs_put_transaction(t);
4576 trace_btrfs_transaction_commit(fs_info->tree_root);
4577 spin_lock(&fs_info->trans_lock);
4578 }
4579 spin_unlock(&fs_info->trans_lock);
4580 btrfs_destroy_all_ordered_extents(fs_info);
4581 btrfs_destroy_delayed_inodes(fs_info);
4582 btrfs_assert_delayed_root_empty(fs_info);
4583 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4584 btrfs_destroy_all_delalloc_inodes(fs_info);
4585 mutex_unlock(&fs_info->transaction_kthread_mutex);
4586
4587 return 0;
4588}
4589
4590static const struct extent_io_ops btree_extent_io_ops = {
4591 /* mandatory callbacks */
4592 .submit_bio_hook = btree_submit_bio_hook,
4593 .readpage_end_io_hook = btree_readpage_end_io_hook,
4594};
1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/fs.h>
20#include <linux/blkdev.h>
21#include <linux/scatterlist.h>
22#include <linux/swap.h>
23#include <linux/radix-tree.h>
24#include <linux/writeback.h>
25#include <linux/buffer_head.h>
26#include <linux/workqueue.h>
27#include <linux/kthread.h>
28#include <linux/freezer.h>
29#include <linux/slab.h>
30#include <linux/migrate.h>
31#include <linux/ratelimit.h>
32#include <linux/uuid.h>
33#include <linux/semaphore.h>
34#include <asm/unaligned.h>
35#include "ctree.h"
36#include "disk-io.h"
37#include "hash.h"
38#include "transaction.h"
39#include "btrfs_inode.h"
40#include "volumes.h"
41#include "print-tree.h"
42#include "async-thread.h"
43#include "locking.h"
44#include "tree-log.h"
45#include "free-space-cache.h"
46#include "inode-map.h"
47#include "check-integrity.h"
48#include "rcu-string.h"
49#include "dev-replace.h"
50#include "raid56.h"
51#include "sysfs.h"
52
53#ifdef CONFIG_X86
54#include <asm/cpufeature.h>
55#endif
56
57static struct extent_io_ops btree_extent_io_ops;
58static void end_workqueue_fn(struct btrfs_work *work);
59static void free_fs_root(struct btrfs_root *root);
60static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
61 int read_only);
62static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
63 struct btrfs_root *root);
64static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
65static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
66 struct btrfs_root *root);
67static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
68static int btrfs_destroy_marked_extents(struct btrfs_root *root,
69 struct extent_io_tree *dirty_pages,
70 int mark);
71static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
72 struct extent_io_tree *pinned_extents);
73static int btrfs_cleanup_transaction(struct btrfs_root *root);
74static void btrfs_error_commit_super(struct btrfs_root *root);
75
76/*
77 * end_io_wq structs are used to do processing in task context when an IO is
78 * complete. This is used during reads to verify checksums, and it is used
79 * by writes to insert metadata for new file extents after IO is complete.
80 */
81struct end_io_wq {
82 struct bio *bio;
83 bio_end_io_t *end_io;
84 void *private;
85 struct btrfs_fs_info *info;
86 int error;
87 int metadata;
88 struct list_head list;
89 struct btrfs_work work;
90};
91
92/*
93 * async submit bios are used to offload expensive checksumming
94 * onto the worker threads. They checksum file and metadata bios
95 * just before they are sent down the IO stack.
96 */
97struct async_submit_bio {
98 struct inode *inode;
99 struct bio *bio;
100 struct list_head list;
101 extent_submit_bio_hook_t *submit_bio_start;
102 extent_submit_bio_hook_t *submit_bio_done;
103 int rw;
104 int mirror_num;
105 unsigned long bio_flags;
106 /*
107 * bio_offset is optional, can be used if the pages in the bio
108 * can't tell us where in the file the bio should go
109 */
110 u64 bio_offset;
111 struct btrfs_work work;
112 int error;
113};
114
115/*
116 * Lockdep class keys for extent_buffer->lock's in this root. For a given
117 * eb, the lockdep key is determined by the btrfs_root it belongs to and
118 * the level the eb occupies in the tree.
119 *
120 * Different roots are used for different purposes and may nest inside each
121 * other and they require separate keysets. As lockdep keys should be
122 * static, assign keysets according to the purpose of the root as indicated
123 * by btrfs_root->objectid. This ensures that all special purpose roots
124 * have separate keysets.
125 *
126 * Lock-nesting across peer nodes is always done with the immediate parent
127 * node locked thus preventing deadlock. As lockdep doesn't know this, use
128 * subclass to avoid triggering lockdep warning in such cases.
129 *
130 * The key is set by the readpage_end_io_hook after the buffer has passed
131 * csum validation but before the pages are unlocked. It is also set by
132 * btrfs_init_new_buffer on freshly allocated blocks.
133 *
134 * We also add a check to make sure the highest level of the tree is the
135 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
136 * needs update as well.
137 */
138#ifdef CONFIG_DEBUG_LOCK_ALLOC
139# if BTRFS_MAX_LEVEL != 8
140# error
141# endif
142
143static struct btrfs_lockdep_keyset {
144 u64 id; /* root objectid */
145 const char *name_stem; /* lock name stem */
146 char names[BTRFS_MAX_LEVEL + 1][20];
147 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
148} btrfs_lockdep_keysets[] = {
149 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
150 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
151 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
152 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
153 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
154 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
155 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
156 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
157 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
158 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
159 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
160 { .id = 0, .name_stem = "tree" },
161};
162
163void __init btrfs_init_lockdep(void)
164{
165 int i, j;
166
167 /* initialize lockdep class names */
168 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
169 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
170
171 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
172 snprintf(ks->names[j], sizeof(ks->names[j]),
173 "btrfs-%s-%02d", ks->name_stem, j);
174 }
175}
176
177void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
178 int level)
179{
180 struct btrfs_lockdep_keyset *ks;
181
182 BUG_ON(level >= ARRAY_SIZE(ks->keys));
183
184 /* find the matching keyset, id 0 is the default entry */
185 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
186 if (ks->id == objectid)
187 break;
188
189 lockdep_set_class_and_name(&eb->lock,
190 &ks->keys[level], ks->names[level]);
191}
192
193#endif
194
195/*
196 * extents on the btree inode are pretty simple, there's one extent
197 * that covers the entire device
198 */
199static struct extent_map *btree_get_extent(struct inode *inode,
200 struct page *page, size_t pg_offset, u64 start, u64 len,
201 int create)
202{
203 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
204 struct extent_map *em;
205 int ret;
206
207 read_lock(&em_tree->lock);
208 em = lookup_extent_mapping(em_tree, start, len);
209 if (em) {
210 em->bdev =
211 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
212 read_unlock(&em_tree->lock);
213 goto out;
214 }
215 read_unlock(&em_tree->lock);
216
217 em = alloc_extent_map();
218 if (!em) {
219 em = ERR_PTR(-ENOMEM);
220 goto out;
221 }
222 em->start = 0;
223 em->len = (u64)-1;
224 em->block_len = (u64)-1;
225 em->block_start = 0;
226 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
227
228 write_lock(&em_tree->lock);
229 ret = add_extent_mapping(em_tree, em, 0);
230 if (ret == -EEXIST) {
231 free_extent_map(em);
232 em = lookup_extent_mapping(em_tree, start, len);
233 if (!em)
234 em = ERR_PTR(-EIO);
235 } else if (ret) {
236 free_extent_map(em);
237 em = ERR_PTR(ret);
238 }
239 write_unlock(&em_tree->lock);
240
241out:
242 return em;
243}
244
245u32 btrfs_csum_data(char *data, u32 seed, size_t len)
246{
247 return btrfs_crc32c(seed, data, len);
248}
249
250void btrfs_csum_final(u32 crc, char *result)
251{
252 put_unaligned_le32(~crc, result);
253}
254
255/*
256 * compute the csum for a btree block, and either verify it or write it
257 * into the csum field of the block.
258 */
259static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
260 int verify)
261{
262 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
263 char *result = NULL;
264 unsigned long len;
265 unsigned long cur_len;
266 unsigned long offset = BTRFS_CSUM_SIZE;
267 char *kaddr;
268 unsigned long map_start;
269 unsigned long map_len;
270 int err;
271 u32 crc = ~(u32)0;
272 unsigned long inline_result;
273
274 len = buf->len - offset;
275 while (len > 0) {
276 err = map_private_extent_buffer(buf, offset, 32,
277 &kaddr, &map_start, &map_len);
278 if (err)
279 return 1;
280 cur_len = min(len, map_len - (offset - map_start));
281 crc = btrfs_csum_data(kaddr + offset - map_start,
282 crc, cur_len);
283 len -= cur_len;
284 offset += cur_len;
285 }
286 if (csum_size > sizeof(inline_result)) {
287 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
288 if (!result)
289 return 1;
290 } else {
291 result = (char *)&inline_result;
292 }
293
294 btrfs_csum_final(crc, result);
295
296 if (verify) {
297 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
298 u32 val;
299 u32 found = 0;
300 memcpy(&found, result, csum_size);
301
302 read_extent_buffer(buf, &val, 0, csum_size);
303 printk_ratelimited(KERN_INFO
304 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
305 "level %d\n",
306 root->fs_info->sb->s_id, buf->start,
307 val, found, btrfs_header_level(buf));
308 if (result != (char *)&inline_result)
309 kfree(result);
310 return 1;
311 }
312 } else {
313 write_extent_buffer(buf, result, 0, csum_size);
314 }
315 if (result != (char *)&inline_result)
316 kfree(result);
317 return 0;
318}
319
320/*
321 * we can't consider a given block up to date unless the transid of the
322 * block matches the transid in the parent node's pointer. This is how we
323 * detect blocks that either didn't get written at all or got written
324 * in the wrong place.
325 */
326static int verify_parent_transid(struct extent_io_tree *io_tree,
327 struct extent_buffer *eb, u64 parent_transid,
328 int atomic)
329{
330 struct extent_state *cached_state = NULL;
331 int ret;
332 bool need_lock = (current->journal_info ==
333 (void *)BTRFS_SEND_TRANS_STUB);
334
335 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
336 return 0;
337
338 if (atomic)
339 return -EAGAIN;
340
341 if (need_lock) {
342 btrfs_tree_read_lock(eb);
343 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
344 }
345
346 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
347 0, &cached_state);
348 if (extent_buffer_uptodate(eb) &&
349 btrfs_header_generation(eb) == parent_transid) {
350 ret = 0;
351 goto out;
352 }
353 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
354 "found %llu\n",
355 eb->start, parent_transid, btrfs_header_generation(eb));
356 ret = 1;
357
358 /*
359 * Things reading via commit roots that don't have normal protection,
360 * like send, can have a really old block in cache that may point at a
361 * block that has been free'd and re-allocated. So don't clear uptodate
362 * if we find an eb that is under IO (dirty/writeback) because we could
363 * end up reading in the stale data and then writing it back out and
364 * making everybody very sad.
365 */
366 if (!extent_buffer_under_io(eb))
367 clear_extent_buffer_uptodate(eb);
368out:
369 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
370 &cached_state, GFP_NOFS);
371 btrfs_tree_read_unlock_blocking(eb);
372 return ret;
373}
374
375/*
376 * Return 0 if the superblock checksum type matches the checksum value of that
377 * algorithm. Pass the raw disk superblock data.
378 */
379static int btrfs_check_super_csum(char *raw_disk_sb)
380{
381 struct btrfs_super_block *disk_sb =
382 (struct btrfs_super_block *)raw_disk_sb;
383 u16 csum_type = btrfs_super_csum_type(disk_sb);
384 int ret = 0;
385
386 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
387 u32 crc = ~(u32)0;
388 const int csum_size = sizeof(crc);
389 char result[csum_size];
390
391 /*
392 * The super_block structure does not span the whole
393 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
394 * is filled with zeros and is included in the checkum.
395 */
396 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
397 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
398 btrfs_csum_final(crc, result);
399
400 if (memcmp(raw_disk_sb, result, csum_size))
401 ret = 1;
402
403 if (ret && btrfs_super_generation(disk_sb) < 10) {
404 printk(KERN_WARNING
405 "BTRFS: super block crcs don't match, older mkfs detected\n");
406 ret = 0;
407 }
408 }
409
410 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
411 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
412 csum_type);
413 ret = 1;
414 }
415
416 return ret;
417}
418
419/*
420 * helper to read a given tree block, doing retries as required when
421 * the checksums don't match and we have alternate mirrors to try.
422 */
423static int btree_read_extent_buffer_pages(struct btrfs_root *root,
424 struct extent_buffer *eb,
425 u64 start, u64 parent_transid)
426{
427 struct extent_io_tree *io_tree;
428 int failed = 0;
429 int ret;
430 int num_copies = 0;
431 int mirror_num = 0;
432 int failed_mirror = 0;
433
434 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
435 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
436 while (1) {
437 ret = read_extent_buffer_pages(io_tree, eb, start,
438 WAIT_COMPLETE,
439 btree_get_extent, mirror_num);
440 if (!ret) {
441 if (!verify_parent_transid(io_tree, eb,
442 parent_transid, 0))
443 break;
444 else
445 ret = -EIO;
446 }
447
448 /*
449 * This buffer's crc is fine, but its contents are corrupted, so
450 * there is no reason to read the other copies, they won't be
451 * any less wrong.
452 */
453 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
454 break;
455
456 num_copies = btrfs_num_copies(root->fs_info,
457 eb->start, eb->len);
458 if (num_copies == 1)
459 break;
460
461 if (!failed_mirror) {
462 failed = 1;
463 failed_mirror = eb->read_mirror;
464 }
465
466 mirror_num++;
467 if (mirror_num == failed_mirror)
468 mirror_num++;
469
470 if (mirror_num > num_copies)
471 break;
472 }
473
474 if (failed && !ret && failed_mirror)
475 repair_eb_io_failure(root, eb, failed_mirror);
476
477 return ret;
478}
479
480/*
481 * checksum a dirty tree block before IO. This has extra checks to make sure
482 * we only fill in the checksum field in the first page of a multi-page block
483 */
484
485static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
486{
487 u64 start = page_offset(page);
488 u64 found_start;
489 struct extent_buffer *eb;
490
491 eb = (struct extent_buffer *)page->private;
492 if (page != eb->pages[0])
493 return 0;
494 found_start = btrfs_header_bytenr(eb);
495 if (WARN_ON(found_start != start || !PageUptodate(page)))
496 return 0;
497 csum_tree_block(root, eb, 0);
498 return 0;
499}
500
501static int check_tree_block_fsid(struct btrfs_root *root,
502 struct extent_buffer *eb)
503{
504 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
505 u8 fsid[BTRFS_UUID_SIZE];
506 int ret = 1;
507
508 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
509 while (fs_devices) {
510 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
511 ret = 0;
512 break;
513 }
514 fs_devices = fs_devices->seed;
515 }
516 return ret;
517}
518
519#define CORRUPT(reason, eb, root, slot) \
520 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
521 "root=%llu, slot=%d", reason, \
522 btrfs_header_bytenr(eb), root->objectid, slot)
523
524static noinline int check_leaf(struct btrfs_root *root,
525 struct extent_buffer *leaf)
526{
527 struct btrfs_key key;
528 struct btrfs_key leaf_key;
529 u32 nritems = btrfs_header_nritems(leaf);
530 int slot;
531
532 if (nritems == 0)
533 return 0;
534
535 /* Check the 0 item */
536 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
537 BTRFS_LEAF_DATA_SIZE(root)) {
538 CORRUPT("invalid item offset size pair", leaf, root, 0);
539 return -EIO;
540 }
541
542 /*
543 * Check to make sure each items keys are in the correct order and their
544 * offsets make sense. We only have to loop through nritems-1 because
545 * we check the current slot against the next slot, which verifies the
546 * next slot's offset+size makes sense and that the current's slot
547 * offset is correct.
548 */
549 for (slot = 0; slot < nritems - 1; slot++) {
550 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
551 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
552
553 /* Make sure the keys are in the right order */
554 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
555 CORRUPT("bad key order", leaf, root, slot);
556 return -EIO;
557 }
558
559 /*
560 * Make sure the offset and ends are right, remember that the
561 * item data starts at the end of the leaf and grows towards the
562 * front.
563 */
564 if (btrfs_item_offset_nr(leaf, slot) !=
565 btrfs_item_end_nr(leaf, slot + 1)) {
566 CORRUPT("slot offset bad", leaf, root, slot);
567 return -EIO;
568 }
569
570 /*
571 * Check to make sure that we don't point outside of the leaf,
572 * just incase all the items are consistent to eachother, but
573 * all point outside of the leaf.
574 */
575 if (btrfs_item_end_nr(leaf, slot) >
576 BTRFS_LEAF_DATA_SIZE(root)) {
577 CORRUPT("slot end outside of leaf", leaf, root, slot);
578 return -EIO;
579 }
580 }
581
582 return 0;
583}
584
585static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
586 u64 phy_offset, struct page *page,
587 u64 start, u64 end, int mirror)
588{
589 u64 found_start;
590 int found_level;
591 struct extent_buffer *eb;
592 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
593 int ret = 0;
594 int reads_done;
595
596 if (!page->private)
597 goto out;
598
599 eb = (struct extent_buffer *)page->private;
600
601 /* the pending IO might have been the only thing that kept this buffer
602 * in memory. Make sure we have a ref for all this other checks
603 */
604 extent_buffer_get(eb);
605
606 reads_done = atomic_dec_and_test(&eb->io_pages);
607 if (!reads_done)
608 goto err;
609
610 eb->read_mirror = mirror;
611 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
612 ret = -EIO;
613 goto err;
614 }
615
616 found_start = btrfs_header_bytenr(eb);
617 if (found_start != eb->start) {
618 printk_ratelimited(KERN_INFO "BTRFS: bad tree block start "
619 "%llu %llu\n",
620 found_start, eb->start);
621 ret = -EIO;
622 goto err;
623 }
624 if (check_tree_block_fsid(root, eb)) {
625 printk_ratelimited(KERN_INFO "BTRFS: bad fsid on block %llu\n",
626 eb->start);
627 ret = -EIO;
628 goto err;
629 }
630 found_level = btrfs_header_level(eb);
631 if (found_level >= BTRFS_MAX_LEVEL) {
632 btrfs_info(root->fs_info, "bad tree block level %d",
633 (int)btrfs_header_level(eb));
634 ret = -EIO;
635 goto err;
636 }
637
638 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
639 eb, found_level);
640
641 ret = csum_tree_block(root, eb, 1);
642 if (ret) {
643 ret = -EIO;
644 goto err;
645 }
646
647 /*
648 * If this is a leaf block and it is corrupt, set the corrupt bit so
649 * that we don't try and read the other copies of this block, just
650 * return -EIO.
651 */
652 if (found_level == 0 && check_leaf(root, eb)) {
653 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
654 ret = -EIO;
655 }
656
657 if (!ret)
658 set_extent_buffer_uptodate(eb);
659err:
660 if (reads_done &&
661 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
662 btree_readahead_hook(root, eb, eb->start, ret);
663
664 if (ret) {
665 /*
666 * our io error hook is going to dec the io pages
667 * again, we have to make sure it has something
668 * to decrement
669 */
670 atomic_inc(&eb->io_pages);
671 clear_extent_buffer_uptodate(eb);
672 }
673 free_extent_buffer(eb);
674out:
675 return ret;
676}
677
678static int btree_io_failed_hook(struct page *page, int failed_mirror)
679{
680 struct extent_buffer *eb;
681 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
682
683 eb = (struct extent_buffer *)page->private;
684 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
685 eb->read_mirror = failed_mirror;
686 atomic_dec(&eb->io_pages);
687 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
688 btree_readahead_hook(root, eb, eb->start, -EIO);
689 return -EIO; /* we fixed nothing */
690}
691
692static void end_workqueue_bio(struct bio *bio, int err)
693{
694 struct end_io_wq *end_io_wq = bio->bi_private;
695 struct btrfs_fs_info *fs_info;
696
697 fs_info = end_io_wq->info;
698 end_io_wq->error = err;
699 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
700
701 if (bio->bi_rw & REQ_WRITE) {
702 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
703 btrfs_queue_work(fs_info->endio_meta_write_workers,
704 &end_io_wq->work);
705 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
706 btrfs_queue_work(fs_info->endio_freespace_worker,
707 &end_io_wq->work);
708 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
709 btrfs_queue_work(fs_info->endio_raid56_workers,
710 &end_io_wq->work);
711 else
712 btrfs_queue_work(fs_info->endio_write_workers,
713 &end_io_wq->work);
714 } else {
715 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
716 btrfs_queue_work(fs_info->endio_raid56_workers,
717 &end_io_wq->work);
718 else if (end_io_wq->metadata)
719 btrfs_queue_work(fs_info->endio_meta_workers,
720 &end_io_wq->work);
721 else
722 btrfs_queue_work(fs_info->endio_workers,
723 &end_io_wq->work);
724 }
725}
726
727/*
728 * For the metadata arg you want
729 *
730 * 0 - if data
731 * 1 - if normal metadta
732 * 2 - if writing to the free space cache area
733 * 3 - raid parity work
734 */
735int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
736 int metadata)
737{
738 struct end_io_wq *end_io_wq;
739 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
740 if (!end_io_wq)
741 return -ENOMEM;
742
743 end_io_wq->private = bio->bi_private;
744 end_io_wq->end_io = bio->bi_end_io;
745 end_io_wq->info = info;
746 end_io_wq->error = 0;
747 end_io_wq->bio = bio;
748 end_io_wq->metadata = metadata;
749
750 bio->bi_private = end_io_wq;
751 bio->bi_end_io = end_workqueue_bio;
752 return 0;
753}
754
755unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
756{
757 unsigned long limit = min_t(unsigned long,
758 info->thread_pool_size,
759 info->fs_devices->open_devices);
760 return 256 * limit;
761}
762
763static void run_one_async_start(struct btrfs_work *work)
764{
765 struct async_submit_bio *async;
766 int ret;
767
768 async = container_of(work, struct async_submit_bio, work);
769 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
770 async->mirror_num, async->bio_flags,
771 async->bio_offset);
772 if (ret)
773 async->error = ret;
774}
775
776static void run_one_async_done(struct btrfs_work *work)
777{
778 struct btrfs_fs_info *fs_info;
779 struct async_submit_bio *async;
780 int limit;
781
782 async = container_of(work, struct async_submit_bio, work);
783 fs_info = BTRFS_I(async->inode)->root->fs_info;
784
785 limit = btrfs_async_submit_limit(fs_info);
786 limit = limit * 2 / 3;
787
788 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
789 waitqueue_active(&fs_info->async_submit_wait))
790 wake_up(&fs_info->async_submit_wait);
791
792 /* If an error occured we just want to clean up the bio and move on */
793 if (async->error) {
794 bio_endio(async->bio, async->error);
795 return;
796 }
797
798 async->submit_bio_done(async->inode, async->rw, async->bio,
799 async->mirror_num, async->bio_flags,
800 async->bio_offset);
801}
802
803static void run_one_async_free(struct btrfs_work *work)
804{
805 struct async_submit_bio *async;
806
807 async = container_of(work, struct async_submit_bio, work);
808 kfree(async);
809}
810
811int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
812 int rw, struct bio *bio, int mirror_num,
813 unsigned long bio_flags,
814 u64 bio_offset,
815 extent_submit_bio_hook_t *submit_bio_start,
816 extent_submit_bio_hook_t *submit_bio_done)
817{
818 struct async_submit_bio *async;
819
820 async = kmalloc(sizeof(*async), GFP_NOFS);
821 if (!async)
822 return -ENOMEM;
823
824 async->inode = inode;
825 async->rw = rw;
826 async->bio = bio;
827 async->mirror_num = mirror_num;
828 async->submit_bio_start = submit_bio_start;
829 async->submit_bio_done = submit_bio_done;
830
831 btrfs_init_work(&async->work, run_one_async_start,
832 run_one_async_done, run_one_async_free);
833
834 async->bio_flags = bio_flags;
835 async->bio_offset = bio_offset;
836
837 async->error = 0;
838
839 atomic_inc(&fs_info->nr_async_submits);
840
841 if (rw & REQ_SYNC)
842 btrfs_set_work_high_priority(&async->work);
843
844 btrfs_queue_work(fs_info->workers, &async->work);
845
846 while (atomic_read(&fs_info->async_submit_draining) &&
847 atomic_read(&fs_info->nr_async_submits)) {
848 wait_event(fs_info->async_submit_wait,
849 (atomic_read(&fs_info->nr_async_submits) == 0));
850 }
851
852 return 0;
853}
854
855static int btree_csum_one_bio(struct bio *bio)
856{
857 struct bio_vec *bvec;
858 struct btrfs_root *root;
859 int i, ret = 0;
860
861 bio_for_each_segment_all(bvec, bio, i) {
862 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
863 ret = csum_dirty_buffer(root, bvec->bv_page);
864 if (ret)
865 break;
866 }
867
868 return ret;
869}
870
871static int __btree_submit_bio_start(struct inode *inode, int rw,
872 struct bio *bio, int mirror_num,
873 unsigned long bio_flags,
874 u64 bio_offset)
875{
876 /*
877 * when we're called for a write, we're already in the async
878 * submission context. Just jump into btrfs_map_bio
879 */
880 return btree_csum_one_bio(bio);
881}
882
883static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
884 int mirror_num, unsigned long bio_flags,
885 u64 bio_offset)
886{
887 int ret;
888
889 /*
890 * when we're called for a write, we're already in the async
891 * submission context. Just jump into btrfs_map_bio
892 */
893 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
894 if (ret)
895 bio_endio(bio, ret);
896 return ret;
897}
898
899static int check_async_write(struct inode *inode, unsigned long bio_flags)
900{
901 if (bio_flags & EXTENT_BIO_TREE_LOG)
902 return 0;
903#ifdef CONFIG_X86
904 if (cpu_has_xmm4_2)
905 return 0;
906#endif
907 return 1;
908}
909
910static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
911 int mirror_num, unsigned long bio_flags,
912 u64 bio_offset)
913{
914 int async = check_async_write(inode, bio_flags);
915 int ret;
916
917 if (!(rw & REQ_WRITE)) {
918 /*
919 * called for a read, do the setup so that checksum validation
920 * can happen in the async kernel threads
921 */
922 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
923 bio, 1);
924 if (ret)
925 goto out_w_error;
926 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
927 mirror_num, 0);
928 } else if (!async) {
929 ret = btree_csum_one_bio(bio);
930 if (ret)
931 goto out_w_error;
932 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
933 mirror_num, 0);
934 } else {
935 /*
936 * kthread helpers are used to submit writes so that
937 * checksumming can happen in parallel across all CPUs
938 */
939 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
940 inode, rw, bio, mirror_num, 0,
941 bio_offset,
942 __btree_submit_bio_start,
943 __btree_submit_bio_done);
944 }
945
946 if (ret) {
947out_w_error:
948 bio_endio(bio, ret);
949 }
950 return ret;
951}
952
953#ifdef CONFIG_MIGRATION
954static int btree_migratepage(struct address_space *mapping,
955 struct page *newpage, struct page *page,
956 enum migrate_mode mode)
957{
958 /*
959 * we can't safely write a btree page from here,
960 * we haven't done the locking hook
961 */
962 if (PageDirty(page))
963 return -EAGAIN;
964 /*
965 * Buffers may be managed in a filesystem specific way.
966 * We must have no buffers or drop them.
967 */
968 if (page_has_private(page) &&
969 !try_to_release_page(page, GFP_KERNEL))
970 return -EAGAIN;
971 return migrate_page(mapping, newpage, page, mode);
972}
973#endif
974
975
976static int btree_writepages(struct address_space *mapping,
977 struct writeback_control *wbc)
978{
979 struct btrfs_fs_info *fs_info;
980 int ret;
981
982 if (wbc->sync_mode == WB_SYNC_NONE) {
983
984 if (wbc->for_kupdate)
985 return 0;
986
987 fs_info = BTRFS_I(mapping->host)->root->fs_info;
988 /* this is a bit racy, but that's ok */
989 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
990 BTRFS_DIRTY_METADATA_THRESH);
991 if (ret < 0)
992 return 0;
993 }
994 return btree_write_cache_pages(mapping, wbc);
995}
996
997static int btree_readpage(struct file *file, struct page *page)
998{
999 struct extent_io_tree *tree;
1000 tree = &BTRFS_I(page->mapping->host)->io_tree;
1001 return extent_read_full_page(tree, page, btree_get_extent, 0);
1002}
1003
1004static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1005{
1006 if (PageWriteback(page) || PageDirty(page))
1007 return 0;
1008
1009 return try_release_extent_buffer(page);
1010}
1011
1012static void btree_invalidatepage(struct page *page, unsigned int offset,
1013 unsigned int length)
1014{
1015 struct extent_io_tree *tree;
1016 tree = &BTRFS_I(page->mapping->host)->io_tree;
1017 extent_invalidatepage(tree, page, offset);
1018 btree_releasepage(page, GFP_NOFS);
1019 if (PagePrivate(page)) {
1020 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1021 "page private not zero on page %llu",
1022 (unsigned long long)page_offset(page));
1023 ClearPagePrivate(page);
1024 set_page_private(page, 0);
1025 page_cache_release(page);
1026 }
1027}
1028
1029static int btree_set_page_dirty(struct page *page)
1030{
1031#ifdef DEBUG
1032 struct extent_buffer *eb;
1033
1034 BUG_ON(!PagePrivate(page));
1035 eb = (struct extent_buffer *)page->private;
1036 BUG_ON(!eb);
1037 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1038 BUG_ON(!atomic_read(&eb->refs));
1039 btrfs_assert_tree_locked(eb);
1040#endif
1041 return __set_page_dirty_nobuffers(page);
1042}
1043
1044static const struct address_space_operations btree_aops = {
1045 .readpage = btree_readpage,
1046 .writepages = btree_writepages,
1047 .releasepage = btree_releasepage,
1048 .invalidatepage = btree_invalidatepage,
1049#ifdef CONFIG_MIGRATION
1050 .migratepage = btree_migratepage,
1051#endif
1052 .set_page_dirty = btree_set_page_dirty,
1053};
1054
1055int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1056 u64 parent_transid)
1057{
1058 struct extent_buffer *buf = NULL;
1059 struct inode *btree_inode = root->fs_info->btree_inode;
1060 int ret = 0;
1061
1062 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1063 if (!buf)
1064 return 0;
1065 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1066 buf, 0, WAIT_NONE, btree_get_extent, 0);
1067 free_extent_buffer(buf);
1068 return ret;
1069}
1070
1071int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1072 int mirror_num, struct extent_buffer **eb)
1073{
1074 struct extent_buffer *buf = NULL;
1075 struct inode *btree_inode = root->fs_info->btree_inode;
1076 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1077 int ret;
1078
1079 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1080 if (!buf)
1081 return 0;
1082
1083 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1084
1085 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1086 btree_get_extent, mirror_num);
1087 if (ret) {
1088 free_extent_buffer(buf);
1089 return ret;
1090 }
1091
1092 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1093 free_extent_buffer(buf);
1094 return -EIO;
1095 } else if (extent_buffer_uptodate(buf)) {
1096 *eb = buf;
1097 } else {
1098 free_extent_buffer(buf);
1099 }
1100 return 0;
1101}
1102
1103struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1104 u64 bytenr, u32 blocksize)
1105{
1106 return find_extent_buffer(root->fs_info, bytenr);
1107}
1108
1109struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1110 u64 bytenr, u32 blocksize)
1111{
1112 return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
1113}
1114
1115
1116int btrfs_write_tree_block(struct extent_buffer *buf)
1117{
1118 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1119 buf->start + buf->len - 1);
1120}
1121
1122int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1123{
1124 return filemap_fdatawait_range(buf->pages[0]->mapping,
1125 buf->start, buf->start + buf->len - 1);
1126}
1127
1128struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1129 u32 blocksize, u64 parent_transid)
1130{
1131 struct extent_buffer *buf = NULL;
1132 int ret;
1133
1134 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1135 if (!buf)
1136 return NULL;
1137
1138 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1139 if (ret) {
1140 free_extent_buffer(buf);
1141 return NULL;
1142 }
1143 return buf;
1144
1145}
1146
1147void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1148 struct extent_buffer *buf)
1149{
1150 struct btrfs_fs_info *fs_info = root->fs_info;
1151
1152 if (btrfs_header_generation(buf) ==
1153 fs_info->running_transaction->transid) {
1154 btrfs_assert_tree_locked(buf);
1155
1156 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1157 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1158 -buf->len,
1159 fs_info->dirty_metadata_batch);
1160 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1161 btrfs_set_lock_blocking(buf);
1162 clear_extent_buffer_dirty(buf);
1163 }
1164 }
1165}
1166
1167static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1168{
1169 struct btrfs_subvolume_writers *writers;
1170 int ret;
1171
1172 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1173 if (!writers)
1174 return ERR_PTR(-ENOMEM);
1175
1176 ret = percpu_counter_init(&writers->counter, 0);
1177 if (ret < 0) {
1178 kfree(writers);
1179 return ERR_PTR(ret);
1180 }
1181
1182 init_waitqueue_head(&writers->wait);
1183 return writers;
1184}
1185
1186static void
1187btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1188{
1189 percpu_counter_destroy(&writers->counter);
1190 kfree(writers);
1191}
1192
1193static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1194 u32 stripesize, struct btrfs_root *root,
1195 struct btrfs_fs_info *fs_info,
1196 u64 objectid)
1197{
1198 root->node = NULL;
1199 root->commit_root = NULL;
1200 root->sectorsize = sectorsize;
1201 root->nodesize = nodesize;
1202 root->leafsize = leafsize;
1203 root->stripesize = stripesize;
1204 root->ref_cows = 0;
1205 root->track_dirty = 0;
1206 root->in_radix = 0;
1207 root->orphan_item_inserted = 0;
1208 root->orphan_cleanup_state = 0;
1209
1210 root->objectid = objectid;
1211 root->last_trans = 0;
1212 root->highest_objectid = 0;
1213 root->nr_delalloc_inodes = 0;
1214 root->nr_ordered_extents = 0;
1215 root->name = NULL;
1216 root->inode_tree = RB_ROOT;
1217 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1218 root->block_rsv = NULL;
1219 root->orphan_block_rsv = NULL;
1220
1221 INIT_LIST_HEAD(&root->dirty_list);
1222 INIT_LIST_HEAD(&root->root_list);
1223 INIT_LIST_HEAD(&root->delalloc_inodes);
1224 INIT_LIST_HEAD(&root->delalloc_root);
1225 INIT_LIST_HEAD(&root->ordered_extents);
1226 INIT_LIST_HEAD(&root->ordered_root);
1227 INIT_LIST_HEAD(&root->logged_list[0]);
1228 INIT_LIST_HEAD(&root->logged_list[1]);
1229 spin_lock_init(&root->orphan_lock);
1230 spin_lock_init(&root->inode_lock);
1231 spin_lock_init(&root->delalloc_lock);
1232 spin_lock_init(&root->ordered_extent_lock);
1233 spin_lock_init(&root->accounting_lock);
1234 spin_lock_init(&root->log_extents_lock[0]);
1235 spin_lock_init(&root->log_extents_lock[1]);
1236 mutex_init(&root->objectid_mutex);
1237 mutex_init(&root->log_mutex);
1238 mutex_init(&root->ordered_extent_mutex);
1239 mutex_init(&root->delalloc_mutex);
1240 init_waitqueue_head(&root->log_writer_wait);
1241 init_waitqueue_head(&root->log_commit_wait[0]);
1242 init_waitqueue_head(&root->log_commit_wait[1]);
1243 INIT_LIST_HEAD(&root->log_ctxs[0]);
1244 INIT_LIST_HEAD(&root->log_ctxs[1]);
1245 atomic_set(&root->log_commit[0], 0);
1246 atomic_set(&root->log_commit[1], 0);
1247 atomic_set(&root->log_writers, 0);
1248 atomic_set(&root->log_batch, 0);
1249 atomic_set(&root->orphan_inodes, 0);
1250 atomic_set(&root->refs, 1);
1251 atomic_set(&root->will_be_snapshoted, 0);
1252 root->log_transid = 0;
1253 root->log_transid_committed = -1;
1254 root->last_log_commit = 0;
1255 if (fs_info)
1256 extent_io_tree_init(&root->dirty_log_pages,
1257 fs_info->btree_inode->i_mapping);
1258
1259 memset(&root->root_key, 0, sizeof(root->root_key));
1260 memset(&root->root_item, 0, sizeof(root->root_item));
1261 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1262 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1263 if (fs_info)
1264 root->defrag_trans_start = fs_info->generation;
1265 else
1266 root->defrag_trans_start = 0;
1267 init_completion(&root->kobj_unregister);
1268 root->defrag_running = 0;
1269 root->root_key.objectid = objectid;
1270 root->anon_dev = 0;
1271
1272 spin_lock_init(&root->root_item_lock);
1273}
1274
1275static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1276{
1277 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1278 if (root)
1279 root->fs_info = fs_info;
1280 return root;
1281}
1282
1283#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1284/* Should only be used by the testing infrastructure */
1285struct btrfs_root *btrfs_alloc_dummy_root(void)
1286{
1287 struct btrfs_root *root;
1288
1289 root = btrfs_alloc_root(NULL);
1290 if (!root)
1291 return ERR_PTR(-ENOMEM);
1292 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1293 root->dummy_root = 1;
1294
1295 return root;
1296}
1297#endif
1298
1299struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1300 struct btrfs_fs_info *fs_info,
1301 u64 objectid)
1302{
1303 struct extent_buffer *leaf;
1304 struct btrfs_root *tree_root = fs_info->tree_root;
1305 struct btrfs_root *root;
1306 struct btrfs_key key;
1307 int ret = 0;
1308 uuid_le uuid;
1309
1310 root = btrfs_alloc_root(fs_info);
1311 if (!root)
1312 return ERR_PTR(-ENOMEM);
1313
1314 __setup_root(tree_root->nodesize, tree_root->leafsize,
1315 tree_root->sectorsize, tree_root->stripesize,
1316 root, fs_info, objectid);
1317 root->root_key.objectid = objectid;
1318 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1319 root->root_key.offset = 0;
1320
1321 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1322 0, objectid, NULL, 0, 0, 0);
1323 if (IS_ERR(leaf)) {
1324 ret = PTR_ERR(leaf);
1325 leaf = NULL;
1326 goto fail;
1327 }
1328
1329 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1330 btrfs_set_header_bytenr(leaf, leaf->start);
1331 btrfs_set_header_generation(leaf, trans->transid);
1332 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1333 btrfs_set_header_owner(leaf, objectid);
1334 root->node = leaf;
1335
1336 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1337 BTRFS_FSID_SIZE);
1338 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1339 btrfs_header_chunk_tree_uuid(leaf),
1340 BTRFS_UUID_SIZE);
1341 btrfs_mark_buffer_dirty(leaf);
1342
1343 root->commit_root = btrfs_root_node(root);
1344 root->track_dirty = 1;
1345
1346
1347 root->root_item.flags = 0;
1348 root->root_item.byte_limit = 0;
1349 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1350 btrfs_set_root_generation(&root->root_item, trans->transid);
1351 btrfs_set_root_level(&root->root_item, 0);
1352 btrfs_set_root_refs(&root->root_item, 1);
1353 btrfs_set_root_used(&root->root_item, leaf->len);
1354 btrfs_set_root_last_snapshot(&root->root_item, 0);
1355 btrfs_set_root_dirid(&root->root_item, 0);
1356 uuid_le_gen(&uuid);
1357 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1358 root->root_item.drop_level = 0;
1359
1360 key.objectid = objectid;
1361 key.type = BTRFS_ROOT_ITEM_KEY;
1362 key.offset = 0;
1363 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1364 if (ret)
1365 goto fail;
1366
1367 btrfs_tree_unlock(leaf);
1368
1369 return root;
1370
1371fail:
1372 if (leaf) {
1373 btrfs_tree_unlock(leaf);
1374 free_extent_buffer(leaf);
1375 }
1376 kfree(root);
1377
1378 return ERR_PTR(ret);
1379}
1380
1381static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1382 struct btrfs_fs_info *fs_info)
1383{
1384 struct btrfs_root *root;
1385 struct btrfs_root *tree_root = fs_info->tree_root;
1386 struct extent_buffer *leaf;
1387
1388 root = btrfs_alloc_root(fs_info);
1389 if (!root)
1390 return ERR_PTR(-ENOMEM);
1391
1392 __setup_root(tree_root->nodesize, tree_root->leafsize,
1393 tree_root->sectorsize, tree_root->stripesize,
1394 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1395
1396 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1397 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1398 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1399 /*
1400 * log trees do not get reference counted because they go away
1401 * before a real commit is actually done. They do store pointers
1402 * to file data extents, and those reference counts still get
1403 * updated (along with back refs to the log tree).
1404 */
1405 root->ref_cows = 0;
1406
1407 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1408 BTRFS_TREE_LOG_OBJECTID, NULL,
1409 0, 0, 0);
1410 if (IS_ERR(leaf)) {
1411 kfree(root);
1412 return ERR_CAST(leaf);
1413 }
1414
1415 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1416 btrfs_set_header_bytenr(leaf, leaf->start);
1417 btrfs_set_header_generation(leaf, trans->transid);
1418 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1419 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1420 root->node = leaf;
1421
1422 write_extent_buffer(root->node, root->fs_info->fsid,
1423 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1424 btrfs_mark_buffer_dirty(root->node);
1425 btrfs_tree_unlock(root->node);
1426 return root;
1427}
1428
1429int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1430 struct btrfs_fs_info *fs_info)
1431{
1432 struct btrfs_root *log_root;
1433
1434 log_root = alloc_log_tree(trans, fs_info);
1435 if (IS_ERR(log_root))
1436 return PTR_ERR(log_root);
1437 WARN_ON(fs_info->log_root_tree);
1438 fs_info->log_root_tree = log_root;
1439 return 0;
1440}
1441
1442int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1443 struct btrfs_root *root)
1444{
1445 struct btrfs_root *log_root;
1446 struct btrfs_inode_item *inode_item;
1447
1448 log_root = alloc_log_tree(trans, root->fs_info);
1449 if (IS_ERR(log_root))
1450 return PTR_ERR(log_root);
1451
1452 log_root->last_trans = trans->transid;
1453 log_root->root_key.offset = root->root_key.objectid;
1454
1455 inode_item = &log_root->root_item.inode;
1456 btrfs_set_stack_inode_generation(inode_item, 1);
1457 btrfs_set_stack_inode_size(inode_item, 3);
1458 btrfs_set_stack_inode_nlink(inode_item, 1);
1459 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1460 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1461
1462 btrfs_set_root_node(&log_root->root_item, log_root->node);
1463
1464 WARN_ON(root->log_root);
1465 root->log_root = log_root;
1466 root->log_transid = 0;
1467 root->log_transid_committed = -1;
1468 root->last_log_commit = 0;
1469 return 0;
1470}
1471
1472static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1473 struct btrfs_key *key)
1474{
1475 struct btrfs_root *root;
1476 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1477 struct btrfs_path *path;
1478 u64 generation;
1479 u32 blocksize;
1480 int ret;
1481
1482 path = btrfs_alloc_path();
1483 if (!path)
1484 return ERR_PTR(-ENOMEM);
1485
1486 root = btrfs_alloc_root(fs_info);
1487 if (!root) {
1488 ret = -ENOMEM;
1489 goto alloc_fail;
1490 }
1491
1492 __setup_root(tree_root->nodesize, tree_root->leafsize,
1493 tree_root->sectorsize, tree_root->stripesize,
1494 root, fs_info, key->objectid);
1495
1496 ret = btrfs_find_root(tree_root, key, path,
1497 &root->root_item, &root->root_key);
1498 if (ret) {
1499 if (ret > 0)
1500 ret = -ENOENT;
1501 goto find_fail;
1502 }
1503
1504 generation = btrfs_root_generation(&root->root_item);
1505 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1506 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1507 blocksize, generation);
1508 if (!root->node) {
1509 ret = -ENOMEM;
1510 goto find_fail;
1511 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1512 ret = -EIO;
1513 goto read_fail;
1514 }
1515 root->commit_root = btrfs_root_node(root);
1516out:
1517 btrfs_free_path(path);
1518 return root;
1519
1520read_fail:
1521 free_extent_buffer(root->node);
1522find_fail:
1523 kfree(root);
1524alloc_fail:
1525 root = ERR_PTR(ret);
1526 goto out;
1527}
1528
1529struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1530 struct btrfs_key *location)
1531{
1532 struct btrfs_root *root;
1533
1534 root = btrfs_read_tree_root(tree_root, location);
1535 if (IS_ERR(root))
1536 return root;
1537
1538 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1539 root->ref_cows = 1;
1540 btrfs_check_and_init_root_item(&root->root_item);
1541 }
1542
1543 return root;
1544}
1545
1546int btrfs_init_fs_root(struct btrfs_root *root)
1547{
1548 int ret;
1549 struct btrfs_subvolume_writers *writers;
1550
1551 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1552 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1553 GFP_NOFS);
1554 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1555 ret = -ENOMEM;
1556 goto fail;
1557 }
1558
1559 writers = btrfs_alloc_subvolume_writers();
1560 if (IS_ERR(writers)) {
1561 ret = PTR_ERR(writers);
1562 goto fail;
1563 }
1564 root->subv_writers = writers;
1565
1566 btrfs_init_free_ino_ctl(root);
1567 spin_lock_init(&root->cache_lock);
1568 init_waitqueue_head(&root->cache_wait);
1569
1570 ret = get_anon_bdev(&root->anon_dev);
1571 if (ret)
1572 goto free_writers;
1573 return 0;
1574
1575free_writers:
1576 btrfs_free_subvolume_writers(root->subv_writers);
1577fail:
1578 kfree(root->free_ino_ctl);
1579 kfree(root->free_ino_pinned);
1580 return ret;
1581}
1582
1583static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1584 u64 root_id)
1585{
1586 struct btrfs_root *root;
1587
1588 spin_lock(&fs_info->fs_roots_radix_lock);
1589 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1590 (unsigned long)root_id);
1591 spin_unlock(&fs_info->fs_roots_radix_lock);
1592 return root;
1593}
1594
1595int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1596 struct btrfs_root *root)
1597{
1598 int ret;
1599
1600 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1601 if (ret)
1602 return ret;
1603
1604 spin_lock(&fs_info->fs_roots_radix_lock);
1605 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1606 (unsigned long)root->root_key.objectid,
1607 root);
1608 if (ret == 0)
1609 root->in_radix = 1;
1610 spin_unlock(&fs_info->fs_roots_radix_lock);
1611 radix_tree_preload_end();
1612
1613 return ret;
1614}
1615
1616struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1617 struct btrfs_key *location,
1618 bool check_ref)
1619{
1620 struct btrfs_root *root;
1621 int ret;
1622
1623 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1624 return fs_info->tree_root;
1625 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1626 return fs_info->extent_root;
1627 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1628 return fs_info->chunk_root;
1629 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1630 return fs_info->dev_root;
1631 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1632 return fs_info->csum_root;
1633 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1634 return fs_info->quota_root ? fs_info->quota_root :
1635 ERR_PTR(-ENOENT);
1636 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1637 return fs_info->uuid_root ? fs_info->uuid_root :
1638 ERR_PTR(-ENOENT);
1639again:
1640 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1641 if (root) {
1642 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1643 return ERR_PTR(-ENOENT);
1644 return root;
1645 }
1646
1647 root = btrfs_read_fs_root(fs_info->tree_root, location);
1648 if (IS_ERR(root))
1649 return root;
1650
1651 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1652 ret = -ENOENT;
1653 goto fail;
1654 }
1655
1656 ret = btrfs_init_fs_root(root);
1657 if (ret)
1658 goto fail;
1659
1660 ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID,
1661 location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
1662 if (ret < 0)
1663 goto fail;
1664 if (ret == 0)
1665 root->orphan_item_inserted = 1;
1666
1667 ret = btrfs_insert_fs_root(fs_info, root);
1668 if (ret) {
1669 if (ret == -EEXIST) {
1670 free_fs_root(root);
1671 goto again;
1672 }
1673 goto fail;
1674 }
1675 return root;
1676fail:
1677 free_fs_root(root);
1678 return ERR_PTR(ret);
1679}
1680
1681static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1682{
1683 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1684 int ret = 0;
1685 struct btrfs_device *device;
1686 struct backing_dev_info *bdi;
1687
1688 rcu_read_lock();
1689 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1690 if (!device->bdev)
1691 continue;
1692 bdi = blk_get_backing_dev_info(device->bdev);
1693 if (bdi && bdi_congested(bdi, bdi_bits)) {
1694 ret = 1;
1695 break;
1696 }
1697 }
1698 rcu_read_unlock();
1699 return ret;
1700}
1701
1702/*
1703 * If this fails, caller must call bdi_destroy() to get rid of the
1704 * bdi again.
1705 */
1706static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1707{
1708 int err;
1709
1710 bdi->capabilities = BDI_CAP_MAP_COPY;
1711 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1712 if (err)
1713 return err;
1714
1715 bdi->ra_pages = default_backing_dev_info.ra_pages;
1716 bdi->congested_fn = btrfs_congested_fn;
1717 bdi->congested_data = info;
1718 return 0;
1719}
1720
1721/*
1722 * called by the kthread helper functions to finally call the bio end_io
1723 * functions. This is where read checksum verification actually happens
1724 */
1725static void end_workqueue_fn(struct btrfs_work *work)
1726{
1727 struct bio *bio;
1728 struct end_io_wq *end_io_wq;
1729 int error;
1730
1731 end_io_wq = container_of(work, struct end_io_wq, work);
1732 bio = end_io_wq->bio;
1733
1734 error = end_io_wq->error;
1735 bio->bi_private = end_io_wq->private;
1736 bio->bi_end_io = end_io_wq->end_io;
1737 kfree(end_io_wq);
1738 bio_endio_nodec(bio, error);
1739}
1740
1741static int cleaner_kthread(void *arg)
1742{
1743 struct btrfs_root *root = arg;
1744 int again;
1745
1746 do {
1747 again = 0;
1748
1749 /* Make the cleaner go to sleep early. */
1750 if (btrfs_need_cleaner_sleep(root))
1751 goto sleep;
1752
1753 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1754 goto sleep;
1755
1756 /*
1757 * Avoid the problem that we change the status of the fs
1758 * during the above check and trylock.
1759 */
1760 if (btrfs_need_cleaner_sleep(root)) {
1761 mutex_unlock(&root->fs_info->cleaner_mutex);
1762 goto sleep;
1763 }
1764
1765 btrfs_run_delayed_iputs(root);
1766 again = btrfs_clean_one_deleted_snapshot(root);
1767 mutex_unlock(&root->fs_info->cleaner_mutex);
1768
1769 /*
1770 * The defragger has dealt with the R/O remount and umount,
1771 * needn't do anything special here.
1772 */
1773 btrfs_run_defrag_inodes(root->fs_info);
1774sleep:
1775 if (!try_to_freeze() && !again) {
1776 set_current_state(TASK_INTERRUPTIBLE);
1777 if (!kthread_should_stop())
1778 schedule();
1779 __set_current_state(TASK_RUNNING);
1780 }
1781 } while (!kthread_should_stop());
1782 return 0;
1783}
1784
1785static int transaction_kthread(void *arg)
1786{
1787 struct btrfs_root *root = arg;
1788 struct btrfs_trans_handle *trans;
1789 struct btrfs_transaction *cur;
1790 u64 transid;
1791 unsigned long now;
1792 unsigned long delay;
1793 bool cannot_commit;
1794
1795 do {
1796 cannot_commit = false;
1797 delay = HZ * root->fs_info->commit_interval;
1798 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1799
1800 spin_lock(&root->fs_info->trans_lock);
1801 cur = root->fs_info->running_transaction;
1802 if (!cur) {
1803 spin_unlock(&root->fs_info->trans_lock);
1804 goto sleep;
1805 }
1806
1807 now = get_seconds();
1808 if (cur->state < TRANS_STATE_BLOCKED &&
1809 (now < cur->start_time ||
1810 now - cur->start_time < root->fs_info->commit_interval)) {
1811 spin_unlock(&root->fs_info->trans_lock);
1812 delay = HZ * 5;
1813 goto sleep;
1814 }
1815 transid = cur->transid;
1816 spin_unlock(&root->fs_info->trans_lock);
1817
1818 /* If the file system is aborted, this will always fail. */
1819 trans = btrfs_attach_transaction(root);
1820 if (IS_ERR(trans)) {
1821 if (PTR_ERR(trans) != -ENOENT)
1822 cannot_commit = true;
1823 goto sleep;
1824 }
1825 if (transid == trans->transid) {
1826 btrfs_commit_transaction(trans, root);
1827 } else {
1828 btrfs_end_transaction(trans, root);
1829 }
1830sleep:
1831 wake_up_process(root->fs_info->cleaner_kthread);
1832 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1833
1834 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1835 &root->fs_info->fs_state)))
1836 btrfs_cleanup_transaction(root);
1837 if (!try_to_freeze()) {
1838 set_current_state(TASK_INTERRUPTIBLE);
1839 if (!kthread_should_stop() &&
1840 (!btrfs_transaction_blocked(root->fs_info) ||
1841 cannot_commit))
1842 schedule_timeout(delay);
1843 __set_current_state(TASK_RUNNING);
1844 }
1845 } while (!kthread_should_stop());
1846 return 0;
1847}
1848
1849/*
1850 * this will find the highest generation in the array of
1851 * root backups. The index of the highest array is returned,
1852 * or -1 if we can't find anything.
1853 *
1854 * We check to make sure the array is valid by comparing the
1855 * generation of the latest root in the array with the generation
1856 * in the super block. If they don't match we pitch it.
1857 */
1858static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1859{
1860 u64 cur;
1861 int newest_index = -1;
1862 struct btrfs_root_backup *root_backup;
1863 int i;
1864
1865 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1866 root_backup = info->super_copy->super_roots + i;
1867 cur = btrfs_backup_tree_root_gen(root_backup);
1868 if (cur == newest_gen)
1869 newest_index = i;
1870 }
1871
1872 /* check to see if we actually wrapped around */
1873 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1874 root_backup = info->super_copy->super_roots;
1875 cur = btrfs_backup_tree_root_gen(root_backup);
1876 if (cur == newest_gen)
1877 newest_index = 0;
1878 }
1879 return newest_index;
1880}
1881
1882
1883/*
1884 * find the oldest backup so we know where to store new entries
1885 * in the backup array. This will set the backup_root_index
1886 * field in the fs_info struct
1887 */
1888static void find_oldest_super_backup(struct btrfs_fs_info *info,
1889 u64 newest_gen)
1890{
1891 int newest_index = -1;
1892
1893 newest_index = find_newest_super_backup(info, newest_gen);
1894 /* if there was garbage in there, just move along */
1895 if (newest_index == -1) {
1896 info->backup_root_index = 0;
1897 } else {
1898 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1899 }
1900}
1901
1902/*
1903 * copy all the root pointers into the super backup array.
1904 * this will bump the backup pointer by one when it is
1905 * done
1906 */
1907static void backup_super_roots(struct btrfs_fs_info *info)
1908{
1909 int next_backup;
1910 struct btrfs_root_backup *root_backup;
1911 int last_backup;
1912
1913 next_backup = info->backup_root_index;
1914 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1915 BTRFS_NUM_BACKUP_ROOTS;
1916
1917 /*
1918 * just overwrite the last backup if we're at the same generation
1919 * this happens only at umount
1920 */
1921 root_backup = info->super_for_commit->super_roots + last_backup;
1922 if (btrfs_backup_tree_root_gen(root_backup) ==
1923 btrfs_header_generation(info->tree_root->node))
1924 next_backup = last_backup;
1925
1926 root_backup = info->super_for_commit->super_roots + next_backup;
1927
1928 /*
1929 * make sure all of our padding and empty slots get zero filled
1930 * regardless of which ones we use today
1931 */
1932 memset(root_backup, 0, sizeof(*root_backup));
1933
1934 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1935
1936 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1937 btrfs_set_backup_tree_root_gen(root_backup,
1938 btrfs_header_generation(info->tree_root->node));
1939
1940 btrfs_set_backup_tree_root_level(root_backup,
1941 btrfs_header_level(info->tree_root->node));
1942
1943 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1944 btrfs_set_backup_chunk_root_gen(root_backup,
1945 btrfs_header_generation(info->chunk_root->node));
1946 btrfs_set_backup_chunk_root_level(root_backup,
1947 btrfs_header_level(info->chunk_root->node));
1948
1949 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1950 btrfs_set_backup_extent_root_gen(root_backup,
1951 btrfs_header_generation(info->extent_root->node));
1952 btrfs_set_backup_extent_root_level(root_backup,
1953 btrfs_header_level(info->extent_root->node));
1954
1955 /*
1956 * we might commit during log recovery, which happens before we set
1957 * the fs_root. Make sure it is valid before we fill it in.
1958 */
1959 if (info->fs_root && info->fs_root->node) {
1960 btrfs_set_backup_fs_root(root_backup,
1961 info->fs_root->node->start);
1962 btrfs_set_backup_fs_root_gen(root_backup,
1963 btrfs_header_generation(info->fs_root->node));
1964 btrfs_set_backup_fs_root_level(root_backup,
1965 btrfs_header_level(info->fs_root->node));
1966 }
1967
1968 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1969 btrfs_set_backup_dev_root_gen(root_backup,
1970 btrfs_header_generation(info->dev_root->node));
1971 btrfs_set_backup_dev_root_level(root_backup,
1972 btrfs_header_level(info->dev_root->node));
1973
1974 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1975 btrfs_set_backup_csum_root_gen(root_backup,
1976 btrfs_header_generation(info->csum_root->node));
1977 btrfs_set_backup_csum_root_level(root_backup,
1978 btrfs_header_level(info->csum_root->node));
1979
1980 btrfs_set_backup_total_bytes(root_backup,
1981 btrfs_super_total_bytes(info->super_copy));
1982 btrfs_set_backup_bytes_used(root_backup,
1983 btrfs_super_bytes_used(info->super_copy));
1984 btrfs_set_backup_num_devices(root_backup,
1985 btrfs_super_num_devices(info->super_copy));
1986
1987 /*
1988 * if we don't copy this out to the super_copy, it won't get remembered
1989 * for the next commit
1990 */
1991 memcpy(&info->super_copy->super_roots,
1992 &info->super_for_commit->super_roots,
1993 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1994}
1995
1996/*
1997 * this copies info out of the root backup array and back into
1998 * the in-memory super block. It is meant to help iterate through
1999 * the array, so you send it the number of backups you've already
2000 * tried and the last backup index you used.
2001 *
2002 * this returns -1 when it has tried all the backups
2003 */
2004static noinline int next_root_backup(struct btrfs_fs_info *info,
2005 struct btrfs_super_block *super,
2006 int *num_backups_tried, int *backup_index)
2007{
2008 struct btrfs_root_backup *root_backup;
2009 int newest = *backup_index;
2010
2011 if (*num_backups_tried == 0) {
2012 u64 gen = btrfs_super_generation(super);
2013
2014 newest = find_newest_super_backup(info, gen);
2015 if (newest == -1)
2016 return -1;
2017
2018 *backup_index = newest;
2019 *num_backups_tried = 1;
2020 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2021 /* we've tried all the backups, all done */
2022 return -1;
2023 } else {
2024 /* jump to the next oldest backup */
2025 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2026 BTRFS_NUM_BACKUP_ROOTS;
2027 *backup_index = newest;
2028 *num_backups_tried += 1;
2029 }
2030 root_backup = super->super_roots + newest;
2031
2032 btrfs_set_super_generation(super,
2033 btrfs_backup_tree_root_gen(root_backup));
2034 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2035 btrfs_set_super_root_level(super,
2036 btrfs_backup_tree_root_level(root_backup));
2037 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2038
2039 /*
2040 * fixme: the total bytes and num_devices need to match or we should
2041 * need a fsck
2042 */
2043 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2044 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2045 return 0;
2046}
2047
2048/* helper to cleanup workers */
2049static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2050{
2051 btrfs_destroy_workqueue(fs_info->fixup_workers);
2052 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2053 btrfs_destroy_workqueue(fs_info->workers);
2054 btrfs_destroy_workqueue(fs_info->endio_workers);
2055 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2056 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2057 btrfs_destroy_workqueue(fs_info->rmw_workers);
2058 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2059 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2060 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2061 btrfs_destroy_workqueue(fs_info->submit_workers);
2062 btrfs_destroy_workqueue(fs_info->delayed_workers);
2063 btrfs_destroy_workqueue(fs_info->caching_workers);
2064 btrfs_destroy_workqueue(fs_info->readahead_workers);
2065 btrfs_destroy_workqueue(fs_info->flush_workers);
2066 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2067}
2068
2069static void free_root_extent_buffers(struct btrfs_root *root)
2070{
2071 if (root) {
2072 free_extent_buffer(root->node);
2073 free_extent_buffer(root->commit_root);
2074 root->node = NULL;
2075 root->commit_root = NULL;
2076 }
2077}
2078
2079/* helper to cleanup tree roots */
2080static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2081{
2082 free_root_extent_buffers(info->tree_root);
2083
2084 free_root_extent_buffers(info->dev_root);
2085 free_root_extent_buffers(info->extent_root);
2086 free_root_extent_buffers(info->csum_root);
2087 free_root_extent_buffers(info->quota_root);
2088 free_root_extent_buffers(info->uuid_root);
2089 if (chunk_root)
2090 free_root_extent_buffers(info->chunk_root);
2091}
2092
2093static void del_fs_roots(struct btrfs_fs_info *fs_info)
2094{
2095 int ret;
2096 struct btrfs_root *gang[8];
2097 int i;
2098
2099 while (!list_empty(&fs_info->dead_roots)) {
2100 gang[0] = list_entry(fs_info->dead_roots.next,
2101 struct btrfs_root, root_list);
2102 list_del(&gang[0]->root_list);
2103
2104 if (gang[0]->in_radix) {
2105 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2106 } else {
2107 free_extent_buffer(gang[0]->node);
2108 free_extent_buffer(gang[0]->commit_root);
2109 btrfs_put_fs_root(gang[0]);
2110 }
2111 }
2112
2113 while (1) {
2114 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2115 (void **)gang, 0,
2116 ARRAY_SIZE(gang));
2117 if (!ret)
2118 break;
2119 for (i = 0; i < ret; i++)
2120 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2121 }
2122
2123 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2124 btrfs_free_log_root_tree(NULL, fs_info);
2125 btrfs_destroy_pinned_extent(fs_info->tree_root,
2126 fs_info->pinned_extents);
2127 }
2128}
2129
2130int open_ctree(struct super_block *sb,
2131 struct btrfs_fs_devices *fs_devices,
2132 char *options)
2133{
2134 u32 sectorsize;
2135 u32 nodesize;
2136 u32 leafsize;
2137 u32 blocksize;
2138 u32 stripesize;
2139 u64 generation;
2140 u64 features;
2141 struct btrfs_key location;
2142 struct buffer_head *bh;
2143 struct btrfs_super_block *disk_super;
2144 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2145 struct btrfs_root *tree_root;
2146 struct btrfs_root *extent_root;
2147 struct btrfs_root *csum_root;
2148 struct btrfs_root *chunk_root;
2149 struct btrfs_root *dev_root;
2150 struct btrfs_root *quota_root;
2151 struct btrfs_root *uuid_root;
2152 struct btrfs_root *log_tree_root;
2153 int ret;
2154 int err = -EINVAL;
2155 int num_backups_tried = 0;
2156 int backup_index = 0;
2157 int max_active;
2158 int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2159 bool create_uuid_tree;
2160 bool check_uuid_tree;
2161
2162 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2163 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2164 if (!tree_root || !chunk_root) {
2165 err = -ENOMEM;
2166 goto fail;
2167 }
2168
2169 ret = init_srcu_struct(&fs_info->subvol_srcu);
2170 if (ret) {
2171 err = ret;
2172 goto fail;
2173 }
2174
2175 ret = setup_bdi(fs_info, &fs_info->bdi);
2176 if (ret) {
2177 err = ret;
2178 goto fail_srcu;
2179 }
2180
2181 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2182 if (ret) {
2183 err = ret;
2184 goto fail_bdi;
2185 }
2186 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2187 (1 + ilog2(nr_cpu_ids));
2188
2189 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2190 if (ret) {
2191 err = ret;
2192 goto fail_dirty_metadata_bytes;
2193 }
2194
2195 ret = percpu_counter_init(&fs_info->bio_counter, 0);
2196 if (ret) {
2197 err = ret;
2198 goto fail_delalloc_bytes;
2199 }
2200
2201 fs_info->btree_inode = new_inode(sb);
2202 if (!fs_info->btree_inode) {
2203 err = -ENOMEM;
2204 goto fail_bio_counter;
2205 }
2206
2207 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2208
2209 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2210 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2211 INIT_LIST_HEAD(&fs_info->trans_list);
2212 INIT_LIST_HEAD(&fs_info->dead_roots);
2213 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2214 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2215 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2216 spin_lock_init(&fs_info->delalloc_root_lock);
2217 spin_lock_init(&fs_info->trans_lock);
2218 spin_lock_init(&fs_info->fs_roots_radix_lock);
2219 spin_lock_init(&fs_info->delayed_iput_lock);
2220 spin_lock_init(&fs_info->defrag_inodes_lock);
2221 spin_lock_init(&fs_info->free_chunk_lock);
2222 spin_lock_init(&fs_info->tree_mod_seq_lock);
2223 spin_lock_init(&fs_info->super_lock);
2224 spin_lock_init(&fs_info->buffer_lock);
2225 rwlock_init(&fs_info->tree_mod_log_lock);
2226 mutex_init(&fs_info->reloc_mutex);
2227 mutex_init(&fs_info->delalloc_root_mutex);
2228 seqlock_init(&fs_info->profiles_lock);
2229
2230 init_completion(&fs_info->kobj_unregister);
2231 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2232 INIT_LIST_HEAD(&fs_info->space_info);
2233 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2234 btrfs_mapping_init(&fs_info->mapping_tree);
2235 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2236 BTRFS_BLOCK_RSV_GLOBAL);
2237 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2238 BTRFS_BLOCK_RSV_DELALLOC);
2239 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2240 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2241 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2242 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2243 BTRFS_BLOCK_RSV_DELOPS);
2244 atomic_set(&fs_info->nr_async_submits, 0);
2245 atomic_set(&fs_info->async_delalloc_pages, 0);
2246 atomic_set(&fs_info->async_submit_draining, 0);
2247 atomic_set(&fs_info->nr_async_bios, 0);
2248 atomic_set(&fs_info->defrag_running, 0);
2249 atomic64_set(&fs_info->tree_mod_seq, 0);
2250 fs_info->sb = sb;
2251 fs_info->max_inline = 8192 * 1024;
2252 fs_info->metadata_ratio = 0;
2253 fs_info->defrag_inodes = RB_ROOT;
2254 fs_info->free_chunk_space = 0;
2255 fs_info->tree_mod_log = RB_ROOT;
2256 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2257 fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2258 /* readahead state */
2259 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2260 spin_lock_init(&fs_info->reada_lock);
2261
2262 fs_info->thread_pool_size = min_t(unsigned long,
2263 num_online_cpus() + 2, 8);
2264
2265 INIT_LIST_HEAD(&fs_info->ordered_roots);
2266 spin_lock_init(&fs_info->ordered_root_lock);
2267 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2268 GFP_NOFS);
2269 if (!fs_info->delayed_root) {
2270 err = -ENOMEM;
2271 goto fail_iput;
2272 }
2273 btrfs_init_delayed_root(fs_info->delayed_root);
2274
2275 mutex_init(&fs_info->scrub_lock);
2276 atomic_set(&fs_info->scrubs_running, 0);
2277 atomic_set(&fs_info->scrub_pause_req, 0);
2278 atomic_set(&fs_info->scrubs_paused, 0);
2279 atomic_set(&fs_info->scrub_cancel_req, 0);
2280 init_waitqueue_head(&fs_info->replace_wait);
2281 init_waitqueue_head(&fs_info->scrub_pause_wait);
2282 fs_info->scrub_workers_refcnt = 0;
2283#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2284 fs_info->check_integrity_print_mask = 0;
2285#endif
2286
2287 spin_lock_init(&fs_info->balance_lock);
2288 mutex_init(&fs_info->balance_mutex);
2289 atomic_set(&fs_info->balance_running, 0);
2290 atomic_set(&fs_info->balance_pause_req, 0);
2291 atomic_set(&fs_info->balance_cancel_req, 0);
2292 fs_info->balance_ctl = NULL;
2293 init_waitqueue_head(&fs_info->balance_wait_q);
2294
2295 sb->s_blocksize = 4096;
2296 sb->s_blocksize_bits = blksize_bits(4096);
2297 sb->s_bdi = &fs_info->bdi;
2298
2299 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2300 set_nlink(fs_info->btree_inode, 1);
2301 /*
2302 * we set the i_size on the btree inode to the max possible int.
2303 * the real end of the address space is determined by all of
2304 * the devices in the system
2305 */
2306 fs_info->btree_inode->i_size = OFFSET_MAX;
2307 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2308 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2309
2310 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2311 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2312 fs_info->btree_inode->i_mapping);
2313 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2314 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2315
2316 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2317
2318 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2319 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2320 sizeof(struct btrfs_key));
2321 set_bit(BTRFS_INODE_DUMMY,
2322 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2323 btrfs_insert_inode_hash(fs_info->btree_inode);
2324
2325 spin_lock_init(&fs_info->block_group_cache_lock);
2326 fs_info->block_group_cache_tree = RB_ROOT;
2327 fs_info->first_logical_byte = (u64)-1;
2328
2329 extent_io_tree_init(&fs_info->freed_extents[0],
2330 fs_info->btree_inode->i_mapping);
2331 extent_io_tree_init(&fs_info->freed_extents[1],
2332 fs_info->btree_inode->i_mapping);
2333 fs_info->pinned_extents = &fs_info->freed_extents[0];
2334 fs_info->do_barriers = 1;
2335
2336
2337 mutex_init(&fs_info->ordered_operations_mutex);
2338 mutex_init(&fs_info->ordered_extent_flush_mutex);
2339 mutex_init(&fs_info->tree_log_mutex);
2340 mutex_init(&fs_info->chunk_mutex);
2341 mutex_init(&fs_info->transaction_kthread_mutex);
2342 mutex_init(&fs_info->cleaner_mutex);
2343 mutex_init(&fs_info->volume_mutex);
2344 init_rwsem(&fs_info->commit_root_sem);
2345 init_rwsem(&fs_info->cleanup_work_sem);
2346 init_rwsem(&fs_info->subvol_sem);
2347 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2348 fs_info->dev_replace.lock_owner = 0;
2349 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2350 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2351 mutex_init(&fs_info->dev_replace.lock_management_lock);
2352 mutex_init(&fs_info->dev_replace.lock);
2353
2354 spin_lock_init(&fs_info->qgroup_lock);
2355 mutex_init(&fs_info->qgroup_ioctl_lock);
2356 fs_info->qgroup_tree = RB_ROOT;
2357 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2358 fs_info->qgroup_seq = 1;
2359 fs_info->quota_enabled = 0;
2360 fs_info->pending_quota_state = 0;
2361 fs_info->qgroup_ulist = NULL;
2362 mutex_init(&fs_info->qgroup_rescan_lock);
2363
2364 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2365 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2366
2367 init_waitqueue_head(&fs_info->transaction_throttle);
2368 init_waitqueue_head(&fs_info->transaction_wait);
2369 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2370 init_waitqueue_head(&fs_info->async_submit_wait);
2371
2372 ret = btrfs_alloc_stripe_hash_table(fs_info);
2373 if (ret) {
2374 err = ret;
2375 goto fail_alloc;
2376 }
2377
2378 __setup_root(4096, 4096, 4096, 4096, tree_root,
2379 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2380
2381 invalidate_bdev(fs_devices->latest_bdev);
2382
2383 /*
2384 * Read super block and check the signature bytes only
2385 */
2386 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2387 if (!bh) {
2388 err = -EINVAL;
2389 goto fail_alloc;
2390 }
2391
2392 /*
2393 * We want to check superblock checksum, the type is stored inside.
2394 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2395 */
2396 if (btrfs_check_super_csum(bh->b_data)) {
2397 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2398 err = -EINVAL;
2399 goto fail_alloc;
2400 }
2401
2402 /*
2403 * super_copy is zeroed at allocation time and we never touch the
2404 * following bytes up to INFO_SIZE, the checksum is calculated from
2405 * the whole block of INFO_SIZE
2406 */
2407 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2408 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2409 sizeof(*fs_info->super_for_commit));
2410 brelse(bh);
2411
2412 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2413
2414 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2415 if (ret) {
2416 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2417 err = -EINVAL;
2418 goto fail_alloc;
2419 }
2420
2421 disk_super = fs_info->super_copy;
2422 if (!btrfs_super_root(disk_super))
2423 goto fail_alloc;
2424
2425 /* check FS state, whether FS is broken. */
2426 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2427 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2428
2429 /*
2430 * run through our array of backup supers and setup
2431 * our ring pointer to the oldest one
2432 */
2433 generation = btrfs_super_generation(disk_super);
2434 find_oldest_super_backup(fs_info, generation);
2435
2436 /*
2437 * In the long term, we'll store the compression type in the super
2438 * block, and it'll be used for per file compression control.
2439 */
2440 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2441
2442 ret = btrfs_parse_options(tree_root, options);
2443 if (ret) {
2444 err = ret;
2445 goto fail_alloc;
2446 }
2447
2448 features = btrfs_super_incompat_flags(disk_super) &
2449 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2450 if (features) {
2451 printk(KERN_ERR "BTRFS: couldn't mount because of "
2452 "unsupported optional features (%Lx).\n",
2453 features);
2454 err = -EINVAL;
2455 goto fail_alloc;
2456 }
2457
2458 if (btrfs_super_leafsize(disk_super) !=
2459 btrfs_super_nodesize(disk_super)) {
2460 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2461 "blocksizes don't match. node %d leaf %d\n",
2462 btrfs_super_nodesize(disk_super),
2463 btrfs_super_leafsize(disk_super));
2464 err = -EINVAL;
2465 goto fail_alloc;
2466 }
2467 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2468 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2469 "blocksize (%d) was too large\n",
2470 btrfs_super_leafsize(disk_super));
2471 err = -EINVAL;
2472 goto fail_alloc;
2473 }
2474
2475 features = btrfs_super_incompat_flags(disk_super);
2476 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2477 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2478 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2479
2480 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2481 printk(KERN_ERR "BTRFS: has skinny extents\n");
2482
2483 /*
2484 * flag our filesystem as having big metadata blocks if
2485 * they are bigger than the page size
2486 */
2487 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2488 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2489 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2490 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2491 }
2492
2493 nodesize = btrfs_super_nodesize(disk_super);
2494 leafsize = btrfs_super_leafsize(disk_super);
2495 sectorsize = btrfs_super_sectorsize(disk_super);
2496 stripesize = btrfs_super_stripesize(disk_super);
2497 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2498 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2499
2500 /*
2501 * mixed block groups end up with duplicate but slightly offset
2502 * extent buffers for the same range. It leads to corruptions
2503 */
2504 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2505 (sectorsize != leafsize)) {
2506 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
2507 "are not allowed for mixed block groups on %s\n",
2508 sb->s_id);
2509 goto fail_alloc;
2510 }
2511
2512 /*
2513 * Needn't use the lock because there is no other task which will
2514 * update the flag.
2515 */
2516 btrfs_set_super_incompat_flags(disk_super, features);
2517
2518 features = btrfs_super_compat_ro_flags(disk_super) &
2519 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2520 if (!(sb->s_flags & MS_RDONLY) && features) {
2521 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2522 "unsupported option features (%Lx).\n",
2523 features);
2524 err = -EINVAL;
2525 goto fail_alloc;
2526 }
2527
2528 max_active = fs_info->thread_pool_size;
2529
2530 fs_info->workers =
2531 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2532 max_active, 16);
2533
2534 fs_info->delalloc_workers =
2535 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2536
2537 fs_info->flush_workers =
2538 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2539
2540 fs_info->caching_workers =
2541 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2542
2543 /*
2544 * a higher idle thresh on the submit workers makes it much more
2545 * likely that bios will be send down in a sane order to the
2546 * devices
2547 */
2548 fs_info->submit_workers =
2549 btrfs_alloc_workqueue("submit", flags,
2550 min_t(u64, fs_devices->num_devices,
2551 max_active), 64);
2552
2553 fs_info->fixup_workers =
2554 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2555
2556 /*
2557 * endios are largely parallel and should have a very
2558 * low idle thresh
2559 */
2560 fs_info->endio_workers =
2561 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2562 fs_info->endio_meta_workers =
2563 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2564 fs_info->endio_meta_write_workers =
2565 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2566 fs_info->endio_raid56_workers =
2567 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2568 fs_info->rmw_workers =
2569 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2570 fs_info->endio_write_workers =
2571 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2572 fs_info->endio_freespace_worker =
2573 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2574 fs_info->delayed_workers =
2575 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2576 fs_info->readahead_workers =
2577 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2578 fs_info->qgroup_rescan_workers =
2579 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2580
2581 if (!(fs_info->workers && fs_info->delalloc_workers &&
2582 fs_info->submit_workers && fs_info->flush_workers &&
2583 fs_info->endio_workers && fs_info->endio_meta_workers &&
2584 fs_info->endio_meta_write_workers &&
2585 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2586 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2587 fs_info->caching_workers && fs_info->readahead_workers &&
2588 fs_info->fixup_workers && fs_info->delayed_workers &&
2589 fs_info->qgroup_rescan_workers)) {
2590 err = -ENOMEM;
2591 goto fail_sb_buffer;
2592 }
2593
2594 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2595 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2596 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2597
2598 tree_root->nodesize = nodesize;
2599 tree_root->leafsize = leafsize;
2600 tree_root->sectorsize = sectorsize;
2601 tree_root->stripesize = stripesize;
2602
2603 sb->s_blocksize = sectorsize;
2604 sb->s_blocksize_bits = blksize_bits(sectorsize);
2605
2606 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2607 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
2608 goto fail_sb_buffer;
2609 }
2610
2611 if (sectorsize != PAGE_SIZE) {
2612 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
2613 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2614 goto fail_sb_buffer;
2615 }
2616
2617 mutex_lock(&fs_info->chunk_mutex);
2618 ret = btrfs_read_sys_array(tree_root);
2619 mutex_unlock(&fs_info->chunk_mutex);
2620 if (ret) {
2621 printk(KERN_WARNING "BTRFS: failed to read the system "
2622 "array on %s\n", sb->s_id);
2623 goto fail_sb_buffer;
2624 }
2625
2626 blocksize = btrfs_level_size(tree_root,
2627 btrfs_super_chunk_root_level(disk_super));
2628 generation = btrfs_super_chunk_root_generation(disk_super);
2629
2630 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2631 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2632
2633 chunk_root->node = read_tree_block(chunk_root,
2634 btrfs_super_chunk_root(disk_super),
2635 blocksize, generation);
2636 if (!chunk_root->node ||
2637 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2638 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
2639 sb->s_id);
2640 goto fail_tree_roots;
2641 }
2642 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2643 chunk_root->commit_root = btrfs_root_node(chunk_root);
2644
2645 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2646 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2647
2648 ret = btrfs_read_chunk_tree(chunk_root);
2649 if (ret) {
2650 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
2651 sb->s_id);
2652 goto fail_tree_roots;
2653 }
2654
2655 /*
2656 * keep the device that is marked to be the target device for the
2657 * dev_replace procedure
2658 */
2659 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2660
2661 if (!fs_devices->latest_bdev) {
2662 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
2663 sb->s_id);
2664 goto fail_tree_roots;
2665 }
2666
2667retry_root_backup:
2668 blocksize = btrfs_level_size(tree_root,
2669 btrfs_super_root_level(disk_super));
2670 generation = btrfs_super_generation(disk_super);
2671
2672 tree_root->node = read_tree_block(tree_root,
2673 btrfs_super_root(disk_super),
2674 blocksize, generation);
2675 if (!tree_root->node ||
2676 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2677 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2678 sb->s_id);
2679
2680 goto recovery_tree_root;
2681 }
2682
2683 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2684 tree_root->commit_root = btrfs_root_node(tree_root);
2685 btrfs_set_root_refs(&tree_root->root_item, 1);
2686
2687 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2688 location.type = BTRFS_ROOT_ITEM_KEY;
2689 location.offset = 0;
2690
2691 extent_root = btrfs_read_tree_root(tree_root, &location);
2692 if (IS_ERR(extent_root)) {
2693 ret = PTR_ERR(extent_root);
2694 goto recovery_tree_root;
2695 }
2696 extent_root->track_dirty = 1;
2697 fs_info->extent_root = extent_root;
2698
2699 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2700 dev_root = btrfs_read_tree_root(tree_root, &location);
2701 if (IS_ERR(dev_root)) {
2702 ret = PTR_ERR(dev_root);
2703 goto recovery_tree_root;
2704 }
2705 dev_root->track_dirty = 1;
2706 fs_info->dev_root = dev_root;
2707 btrfs_init_devices_late(fs_info);
2708
2709 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2710 csum_root = btrfs_read_tree_root(tree_root, &location);
2711 if (IS_ERR(csum_root)) {
2712 ret = PTR_ERR(csum_root);
2713 goto recovery_tree_root;
2714 }
2715 csum_root->track_dirty = 1;
2716 fs_info->csum_root = csum_root;
2717
2718 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2719 quota_root = btrfs_read_tree_root(tree_root, &location);
2720 if (!IS_ERR(quota_root)) {
2721 quota_root->track_dirty = 1;
2722 fs_info->quota_enabled = 1;
2723 fs_info->pending_quota_state = 1;
2724 fs_info->quota_root = quota_root;
2725 }
2726
2727 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2728 uuid_root = btrfs_read_tree_root(tree_root, &location);
2729 if (IS_ERR(uuid_root)) {
2730 ret = PTR_ERR(uuid_root);
2731 if (ret != -ENOENT)
2732 goto recovery_tree_root;
2733 create_uuid_tree = true;
2734 check_uuid_tree = false;
2735 } else {
2736 uuid_root->track_dirty = 1;
2737 fs_info->uuid_root = uuid_root;
2738 create_uuid_tree = false;
2739 check_uuid_tree =
2740 generation != btrfs_super_uuid_tree_generation(disk_super);
2741 }
2742
2743 fs_info->generation = generation;
2744 fs_info->last_trans_committed = generation;
2745
2746 ret = btrfs_recover_balance(fs_info);
2747 if (ret) {
2748 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
2749 goto fail_block_groups;
2750 }
2751
2752 ret = btrfs_init_dev_stats(fs_info);
2753 if (ret) {
2754 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2755 ret);
2756 goto fail_block_groups;
2757 }
2758
2759 ret = btrfs_init_dev_replace(fs_info);
2760 if (ret) {
2761 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2762 goto fail_block_groups;
2763 }
2764
2765 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2766
2767 ret = btrfs_sysfs_add_one(fs_info);
2768 if (ret) {
2769 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2770 goto fail_block_groups;
2771 }
2772
2773 ret = btrfs_init_space_info(fs_info);
2774 if (ret) {
2775 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2776 goto fail_sysfs;
2777 }
2778
2779 ret = btrfs_read_block_groups(extent_root);
2780 if (ret) {
2781 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2782 goto fail_sysfs;
2783 }
2784 fs_info->num_tolerated_disk_barrier_failures =
2785 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2786 if (fs_info->fs_devices->missing_devices >
2787 fs_info->num_tolerated_disk_barrier_failures &&
2788 !(sb->s_flags & MS_RDONLY)) {
2789 printk(KERN_WARNING "BTRFS: "
2790 "too many missing devices, writeable mount is not allowed\n");
2791 goto fail_sysfs;
2792 }
2793
2794 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2795 "btrfs-cleaner");
2796 if (IS_ERR(fs_info->cleaner_kthread))
2797 goto fail_sysfs;
2798
2799 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2800 tree_root,
2801 "btrfs-transaction");
2802 if (IS_ERR(fs_info->transaction_kthread))
2803 goto fail_cleaner;
2804
2805 if (!btrfs_test_opt(tree_root, SSD) &&
2806 !btrfs_test_opt(tree_root, NOSSD) &&
2807 !fs_info->fs_devices->rotating) {
2808 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2809 "mode\n");
2810 btrfs_set_opt(fs_info->mount_opt, SSD);
2811 }
2812
2813 /* Set the real inode map cache flag */
2814 if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE))
2815 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE);
2816
2817#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2818 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2819 ret = btrfsic_mount(tree_root, fs_devices,
2820 btrfs_test_opt(tree_root,
2821 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2822 1 : 0,
2823 fs_info->check_integrity_print_mask);
2824 if (ret)
2825 printk(KERN_WARNING "BTRFS: failed to initialize"
2826 " integrity check module %s\n", sb->s_id);
2827 }
2828#endif
2829 ret = btrfs_read_qgroup_config(fs_info);
2830 if (ret)
2831 goto fail_trans_kthread;
2832
2833 /* do not make disk changes in broken FS */
2834 if (btrfs_super_log_root(disk_super) != 0) {
2835 u64 bytenr = btrfs_super_log_root(disk_super);
2836
2837 if (fs_devices->rw_devices == 0) {
2838 printk(KERN_WARNING "BTRFS: log replay required "
2839 "on RO media\n");
2840 err = -EIO;
2841 goto fail_qgroup;
2842 }
2843 blocksize =
2844 btrfs_level_size(tree_root,
2845 btrfs_super_log_root_level(disk_super));
2846
2847 log_tree_root = btrfs_alloc_root(fs_info);
2848 if (!log_tree_root) {
2849 err = -ENOMEM;
2850 goto fail_qgroup;
2851 }
2852
2853 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2854 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2855
2856 log_tree_root->node = read_tree_block(tree_root, bytenr,
2857 blocksize,
2858 generation + 1);
2859 if (!log_tree_root->node ||
2860 !extent_buffer_uptodate(log_tree_root->node)) {
2861 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2862 free_extent_buffer(log_tree_root->node);
2863 kfree(log_tree_root);
2864 goto fail_qgroup;
2865 }
2866 /* returns with log_tree_root freed on success */
2867 ret = btrfs_recover_log_trees(log_tree_root);
2868 if (ret) {
2869 btrfs_error(tree_root->fs_info, ret,
2870 "Failed to recover log tree");
2871 free_extent_buffer(log_tree_root->node);
2872 kfree(log_tree_root);
2873 goto fail_qgroup;
2874 }
2875
2876 if (sb->s_flags & MS_RDONLY) {
2877 ret = btrfs_commit_super(tree_root);
2878 if (ret)
2879 goto fail_qgroup;
2880 }
2881 }
2882
2883 ret = btrfs_find_orphan_roots(tree_root);
2884 if (ret)
2885 goto fail_qgroup;
2886
2887 if (!(sb->s_flags & MS_RDONLY)) {
2888 ret = btrfs_cleanup_fs_roots(fs_info);
2889 if (ret)
2890 goto fail_qgroup;
2891
2892 ret = btrfs_recover_relocation(tree_root);
2893 if (ret < 0) {
2894 printk(KERN_WARNING
2895 "BTRFS: failed to recover relocation\n");
2896 err = -EINVAL;
2897 goto fail_qgroup;
2898 }
2899 }
2900
2901 location.objectid = BTRFS_FS_TREE_OBJECTID;
2902 location.type = BTRFS_ROOT_ITEM_KEY;
2903 location.offset = 0;
2904
2905 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2906 if (IS_ERR(fs_info->fs_root)) {
2907 err = PTR_ERR(fs_info->fs_root);
2908 goto fail_qgroup;
2909 }
2910
2911 if (sb->s_flags & MS_RDONLY)
2912 return 0;
2913
2914 down_read(&fs_info->cleanup_work_sem);
2915 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2916 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2917 up_read(&fs_info->cleanup_work_sem);
2918 close_ctree(tree_root);
2919 return ret;
2920 }
2921 up_read(&fs_info->cleanup_work_sem);
2922
2923 ret = btrfs_resume_balance_async(fs_info);
2924 if (ret) {
2925 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2926 close_ctree(tree_root);
2927 return ret;
2928 }
2929
2930 ret = btrfs_resume_dev_replace_async(fs_info);
2931 if (ret) {
2932 pr_warn("BTRFS: failed to resume dev_replace\n");
2933 close_ctree(tree_root);
2934 return ret;
2935 }
2936
2937 btrfs_qgroup_rescan_resume(fs_info);
2938
2939 if (create_uuid_tree) {
2940 pr_info("BTRFS: creating UUID tree\n");
2941 ret = btrfs_create_uuid_tree(fs_info);
2942 if (ret) {
2943 pr_warn("BTRFS: failed to create the UUID tree %d\n",
2944 ret);
2945 close_ctree(tree_root);
2946 return ret;
2947 }
2948 } else if (check_uuid_tree ||
2949 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2950 pr_info("BTRFS: checking UUID tree\n");
2951 ret = btrfs_check_uuid_tree(fs_info);
2952 if (ret) {
2953 pr_warn("BTRFS: failed to check the UUID tree %d\n",
2954 ret);
2955 close_ctree(tree_root);
2956 return ret;
2957 }
2958 } else {
2959 fs_info->update_uuid_tree_gen = 1;
2960 }
2961
2962 return 0;
2963
2964fail_qgroup:
2965 btrfs_free_qgroup_config(fs_info);
2966fail_trans_kthread:
2967 kthread_stop(fs_info->transaction_kthread);
2968 btrfs_cleanup_transaction(fs_info->tree_root);
2969 del_fs_roots(fs_info);
2970fail_cleaner:
2971 kthread_stop(fs_info->cleaner_kthread);
2972
2973 /*
2974 * make sure we're done with the btree inode before we stop our
2975 * kthreads
2976 */
2977 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2978
2979fail_sysfs:
2980 btrfs_sysfs_remove_one(fs_info);
2981
2982fail_block_groups:
2983 btrfs_put_block_group_cache(fs_info);
2984 btrfs_free_block_groups(fs_info);
2985
2986fail_tree_roots:
2987 free_root_pointers(fs_info, 1);
2988 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2989
2990fail_sb_buffer:
2991 btrfs_stop_all_workers(fs_info);
2992fail_alloc:
2993fail_iput:
2994 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2995
2996 iput(fs_info->btree_inode);
2997fail_bio_counter:
2998 percpu_counter_destroy(&fs_info->bio_counter);
2999fail_delalloc_bytes:
3000 percpu_counter_destroy(&fs_info->delalloc_bytes);
3001fail_dirty_metadata_bytes:
3002 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3003fail_bdi:
3004 bdi_destroy(&fs_info->bdi);
3005fail_srcu:
3006 cleanup_srcu_struct(&fs_info->subvol_srcu);
3007fail:
3008 btrfs_free_stripe_hash_table(fs_info);
3009 btrfs_close_devices(fs_info->fs_devices);
3010 return err;
3011
3012recovery_tree_root:
3013 if (!btrfs_test_opt(tree_root, RECOVERY))
3014 goto fail_tree_roots;
3015
3016 free_root_pointers(fs_info, 0);
3017
3018 /* don't use the log in recovery mode, it won't be valid */
3019 btrfs_set_super_log_root(disk_super, 0);
3020
3021 /* we can't trust the free space cache either */
3022 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3023
3024 ret = next_root_backup(fs_info, fs_info->super_copy,
3025 &num_backups_tried, &backup_index);
3026 if (ret == -1)
3027 goto fail_block_groups;
3028 goto retry_root_backup;
3029}
3030
3031static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3032{
3033 if (uptodate) {
3034 set_buffer_uptodate(bh);
3035 } else {
3036 struct btrfs_device *device = (struct btrfs_device *)
3037 bh->b_private;
3038
3039 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3040 "I/O error on %s\n",
3041 rcu_str_deref(device->name));
3042 /* note, we dont' set_buffer_write_io_error because we have
3043 * our own ways of dealing with the IO errors
3044 */
3045 clear_buffer_uptodate(bh);
3046 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3047 }
3048 unlock_buffer(bh);
3049 put_bh(bh);
3050}
3051
3052struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3053{
3054 struct buffer_head *bh;
3055 struct buffer_head *latest = NULL;
3056 struct btrfs_super_block *super;
3057 int i;
3058 u64 transid = 0;
3059 u64 bytenr;
3060
3061 /* we would like to check all the supers, but that would make
3062 * a btrfs mount succeed after a mkfs from a different FS.
3063 * So, we need to add a special mount option to scan for
3064 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3065 */
3066 for (i = 0; i < 1; i++) {
3067 bytenr = btrfs_sb_offset(i);
3068 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3069 i_size_read(bdev->bd_inode))
3070 break;
3071 bh = __bread(bdev, bytenr / 4096,
3072 BTRFS_SUPER_INFO_SIZE);
3073 if (!bh)
3074 continue;
3075
3076 super = (struct btrfs_super_block *)bh->b_data;
3077 if (btrfs_super_bytenr(super) != bytenr ||
3078 btrfs_super_magic(super) != BTRFS_MAGIC) {
3079 brelse(bh);
3080 continue;
3081 }
3082
3083 if (!latest || btrfs_super_generation(super) > transid) {
3084 brelse(latest);
3085 latest = bh;
3086 transid = btrfs_super_generation(super);
3087 } else {
3088 brelse(bh);
3089 }
3090 }
3091 return latest;
3092}
3093
3094/*
3095 * this should be called twice, once with wait == 0 and
3096 * once with wait == 1. When wait == 0 is done, all the buffer heads
3097 * we write are pinned.
3098 *
3099 * They are released when wait == 1 is done.
3100 * max_mirrors must be the same for both runs, and it indicates how
3101 * many supers on this one device should be written.
3102 *
3103 * max_mirrors == 0 means to write them all.
3104 */
3105static int write_dev_supers(struct btrfs_device *device,
3106 struct btrfs_super_block *sb,
3107 int do_barriers, int wait, int max_mirrors)
3108{
3109 struct buffer_head *bh;
3110 int i;
3111 int ret;
3112 int errors = 0;
3113 u32 crc;
3114 u64 bytenr;
3115
3116 if (max_mirrors == 0)
3117 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3118
3119 for (i = 0; i < max_mirrors; i++) {
3120 bytenr = btrfs_sb_offset(i);
3121 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3122 break;
3123
3124 if (wait) {
3125 bh = __find_get_block(device->bdev, bytenr / 4096,
3126 BTRFS_SUPER_INFO_SIZE);
3127 if (!bh) {
3128 errors++;
3129 continue;
3130 }
3131 wait_on_buffer(bh);
3132 if (!buffer_uptodate(bh))
3133 errors++;
3134
3135 /* drop our reference */
3136 brelse(bh);
3137
3138 /* drop the reference from the wait == 0 run */
3139 brelse(bh);
3140 continue;
3141 } else {
3142 btrfs_set_super_bytenr(sb, bytenr);
3143
3144 crc = ~(u32)0;
3145 crc = btrfs_csum_data((char *)sb +
3146 BTRFS_CSUM_SIZE, crc,
3147 BTRFS_SUPER_INFO_SIZE -
3148 BTRFS_CSUM_SIZE);
3149 btrfs_csum_final(crc, sb->csum);
3150
3151 /*
3152 * one reference for us, and we leave it for the
3153 * caller
3154 */
3155 bh = __getblk(device->bdev, bytenr / 4096,
3156 BTRFS_SUPER_INFO_SIZE);
3157 if (!bh) {
3158 printk(KERN_ERR "BTRFS: couldn't get super "
3159 "buffer head for bytenr %Lu\n", bytenr);
3160 errors++;
3161 continue;
3162 }
3163
3164 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3165
3166 /* one reference for submit_bh */
3167 get_bh(bh);
3168
3169 set_buffer_uptodate(bh);
3170 lock_buffer(bh);
3171 bh->b_end_io = btrfs_end_buffer_write_sync;
3172 bh->b_private = device;
3173 }
3174
3175 /*
3176 * we fua the first super. The others we allow
3177 * to go down lazy.
3178 */
3179 if (i == 0)
3180 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3181 else
3182 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3183 if (ret)
3184 errors++;
3185 }
3186 return errors < i ? 0 : -1;
3187}
3188
3189/*
3190 * endio for the write_dev_flush, this will wake anyone waiting
3191 * for the barrier when it is done
3192 */
3193static void btrfs_end_empty_barrier(struct bio *bio, int err)
3194{
3195 if (err) {
3196 if (err == -EOPNOTSUPP)
3197 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3198 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3199 }
3200 if (bio->bi_private)
3201 complete(bio->bi_private);
3202 bio_put(bio);
3203}
3204
3205/*
3206 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3207 * sent down. With wait == 1, it waits for the previous flush.
3208 *
3209 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3210 * capable
3211 */
3212static int write_dev_flush(struct btrfs_device *device, int wait)
3213{
3214 struct bio *bio;
3215 int ret = 0;
3216
3217 if (device->nobarriers)
3218 return 0;
3219
3220 if (wait) {
3221 bio = device->flush_bio;
3222 if (!bio)
3223 return 0;
3224
3225 wait_for_completion(&device->flush_wait);
3226
3227 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3228 printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3229 rcu_str_deref(device->name));
3230 device->nobarriers = 1;
3231 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3232 ret = -EIO;
3233 btrfs_dev_stat_inc_and_print(device,
3234 BTRFS_DEV_STAT_FLUSH_ERRS);
3235 }
3236
3237 /* drop the reference from the wait == 0 run */
3238 bio_put(bio);
3239 device->flush_bio = NULL;
3240
3241 return ret;
3242 }
3243
3244 /*
3245 * one reference for us, and we leave it for the
3246 * caller
3247 */
3248 device->flush_bio = NULL;
3249 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3250 if (!bio)
3251 return -ENOMEM;
3252
3253 bio->bi_end_io = btrfs_end_empty_barrier;
3254 bio->bi_bdev = device->bdev;
3255 init_completion(&device->flush_wait);
3256 bio->bi_private = &device->flush_wait;
3257 device->flush_bio = bio;
3258
3259 bio_get(bio);
3260 btrfsic_submit_bio(WRITE_FLUSH, bio);
3261
3262 return 0;
3263}
3264
3265/*
3266 * send an empty flush down to each device in parallel,
3267 * then wait for them
3268 */
3269static int barrier_all_devices(struct btrfs_fs_info *info)
3270{
3271 struct list_head *head;
3272 struct btrfs_device *dev;
3273 int errors_send = 0;
3274 int errors_wait = 0;
3275 int ret;
3276
3277 /* send down all the barriers */
3278 head = &info->fs_devices->devices;
3279 list_for_each_entry_rcu(dev, head, dev_list) {
3280 if (dev->missing)
3281 continue;
3282 if (!dev->bdev) {
3283 errors_send++;
3284 continue;
3285 }
3286 if (!dev->in_fs_metadata || !dev->writeable)
3287 continue;
3288
3289 ret = write_dev_flush(dev, 0);
3290 if (ret)
3291 errors_send++;
3292 }
3293
3294 /* wait for all the barriers */
3295 list_for_each_entry_rcu(dev, head, dev_list) {
3296 if (dev->missing)
3297 continue;
3298 if (!dev->bdev) {
3299 errors_wait++;
3300 continue;
3301 }
3302 if (!dev->in_fs_metadata || !dev->writeable)
3303 continue;
3304
3305 ret = write_dev_flush(dev, 1);
3306 if (ret)
3307 errors_wait++;
3308 }
3309 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3310 errors_wait > info->num_tolerated_disk_barrier_failures)
3311 return -EIO;
3312 return 0;
3313}
3314
3315int btrfs_calc_num_tolerated_disk_barrier_failures(
3316 struct btrfs_fs_info *fs_info)
3317{
3318 struct btrfs_ioctl_space_info space;
3319 struct btrfs_space_info *sinfo;
3320 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3321 BTRFS_BLOCK_GROUP_SYSTEM,
3322 BTRFS_BLOCK_GROUP_METADATA,
3323 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3324 int num_types = 4;
3325 int i;
3326 int c;
3327 int num_tolerated_disk_barrier_failures =
3328 (int)fs_info->fs_devices->num_devices;
3329
3330 for (i = 0; i < num_types; i++) {
3331 struct btrfs_space_info *tmp;
3332
3333 sinfo = NULL;
3334 rcu_read_lock();
3335 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3336 if (tmp->flags == types[i]) {
3337 sinfo = tmp;
3338 break;
3339 }
3340 }
3341 rcu_read_unlock();
3342
3343 if (!sinfo)
3344 continue;
3345
3346 down_read(&sinfo->groups_sem);
3347 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3348 if (!list_empty(&sinfo->block_groups[c])) {
3349 u64 flags;
3350
3351 btrfs_get_block_group_info(
3352 &sinfo->block_groups[c], &space);
3353 if (space.total_bytes == 0 ||
3354 space.used_bytes == 0)
3355 continue;
3356 flags = space.flags;
3357 /*
3358 * return
3359 * 0: if dup, single or RAID0 is configured for
3360 * any of metadata, system or data, else
3361 * 1: if RAID5 is configured, or if RAID1 or
3362 * RAID10 is configured and only two mirrors
3363 * are used, else
3364 * 2: if RAID6 is configured, else
3365 * num_mirrors - 1: if RAID1 or RAID10 is
3366 * configured and more than
3367 * 2 mirrors are used.
3368 */
3369 if (num_tolerated_disk_barrier_failures > 0 &&
3370 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3371 BTRFS_BLOCK_GROUP_RAID0)) ||
3372 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3373 == 0)))
3374 num_tolerated_disk_barrier_failures = 0;
3375 else if (num_tolerated_disk_barrier_failures > 1) {
3376 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3377 BTRFS_BLOCK_GROUP_RAID5 |
3378 BTRFS_BLOCK_GROUP_RAID10)) {
3379 num_tolerated_disk_barrier_failures = 1;
3380 } else if (flags &
3381 BTRFS_BLOCK_GROUP_RAID6) {
3382 num_tolerated_disk_barrier_failures = 2;
3383 }
3384 }
3385 }
3386 }
3387 up_read(&sinfo->groups_sem);
3388 }
3389
3390 return num_tolerated_disk_barrier_failures;
3391}
3392
3393static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3394{
3395 struct list_head *head;
3396 struct btrfs_device *dev;
3397 struct btrfs_super_block *sb;
3398 struct btrfs_dev_item *dev_item;
3399 int ret;
3400 int do_barriers;
3401 int max_errors;
3402 int total_errors = 0;
3403 u64 flags;
3404
3405 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3406 backup_super_roots(root->fs_info);
3407
3408 sb = root->fs_info->super_for_commit;
3409 dev_item = &sb->dev_item;
3410
3411 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3412 head = &root->fs_info->fs_devices->devices;
3413 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3414
3415 if (do_barriers) {
3416 ret = barrier_all_devices(root->fs_info);
3417 if (ret) {
3418 mutex_unlock(
3419 &root->fs_info->fs_devices->device_list_mutex);
3420 btrfs_error(root->fs_info, ret,
3421 "errors while submitting device barriers.");
3422 return ret;
3423 }
3424 }
3425
3426 list_for_each_entry_rcu(dev, head, dev_list) {
3427 if (!dev->bdev) {
3428 total_errors++;
3429 continue;
3430 }
3431 if (!dev->in_fs_metadata || !dev->writeable)
3432 continue;
3433
3434 btrfs_set_stack_device_generation(dev_item, 0);
3435 btrfs_set_stack_device_type(dev_item, dev->type);
3436 btrfs_set_stack_device_id(dev_item, dev->devid);
3437 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3438 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3439 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3440 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3441 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3442 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3443 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3444
3445 flags = btrfs_super_flags(sb);
3446 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3447
3448 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3449 if (ret)
3450 total_errors++;
3451 }
3452 if (total_errors > max_errors) {
3453 btrfs_err(root->fs_info, "%d errors while writing supers",
3454 total_errors);
3455 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3456
3457 /* FUA is masked off if unsupported and can't be the reason */
3458 btrfs_error(root->fs_info, -EIO,
3459 "%d errors while writing supers", total_errors);
3460 return -EIO;
3461 }
3462
3463 total_errors = 0;
3464 list_for_each_entry_rcu(dev, head, dev_list) {
3465 if (!dev->bdev)
3466 continue;
3467 if (!dev->in_fs_metadata || !dev->writeable)
3468 continue;
3469
3470 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3471 if (ret)
3472 total_errors++;
3473 }
3474 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3475 if (total_errors > max_errors) {
3476 btrfs_error(root->fs_info, -EIO,
3477 "%d errors while writing supers", total_errors);
3478 return -EIO;
3479 }
3480 return 0;
3481}
3482
3483int write_ctree_super(struct btrfs_trans_handle *trans,
3484 struct btrfs_root *root, int max_mirrors)
3485{
3486 return write_all_supers(root, max_mirrors);
3487}
3488
3489/* Drop a fs root from the radix tree and free it. */
3490void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3491 struct btrfs_root *root)
3492{
3493 spin_lock(&fs_info->fs_roots_radix_lock);
3494 radix_tree_delete(&fs_info->fs_roots_radix,
3495 (unsigned long)root->root_key.objectid);
3496 spin_unlock(&fs_info->fs_roots_radix_lock);
3497
3498 if (btrfs_root_refs(&root->root_item) == 0)
3499 synchronize_srcu(&fs_info->subvol_srcu);
3500
3501 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3502 btrfs_free_log(NULL, root);
3503
3504 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3505 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3506 free_fs_root(root);
3507}
3508
3509static void free_fs_root(struct btrfs_root *root)
3510{
3511 iput(root->cache_inode);
3512 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3513 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3514 root->orphan_block_rsv = NULL;
3515 if (root->anon_dev)
3516 free_anon_bdev(root->anon_dev);
3517 if (root->subv_writers)
3518 btrfs_free_subvolume_writers(root->subv_writers);
3519 free_extent_buffer(root->node);
3520 free_extent_buffer(root->commit_root);
3521 kfree(root->free_ino_ctl);
3522 kfree(root->free_ino_pinned);
3523 kfree(root->name);
3524 btrfs_put_fs_root(root);
3525}
3526
3527void btrfs_free_fs_root(struct btrfs_root *root)
3528{
3529 free_fs_root(root);
3530}
3531
3532int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3533{
3534 u64 root_objectid = 0;
3535 struct btrfs_root *gang[8];
3536 int i;
3537 int ret;
3538
3539 while (1) {
3540 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3541 (void **)gang, root_objectid,
3542 ARRAY_SIZE(gang));
3543 if (!ret)
3544 break;
3545
3546 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3547 for (i = 0; i < ret; i++) {
3548 int err;
3549
3550 root_objectid = gang[i]->root_key.objectid;
3551 err = btrfs_orphan_cleanup(gang[i]);
3552 if (err)
3553 return err;
3554 }
3555 root_objectid++;
3556 }
3557 return 0;
3558}
3559
3560int btrfs_commit_super(struct btrfs_root *root)
3561{
3562 struct btrfs_trans_handle *trans;
3563
3564 mutex_lock(&root->fs_info->cleaner_mutex);
3565 btrfs_run_delayed_iputs(root);
3566 mutex_unlock(&root->fs_info->cleaner_mutex);
3567 wake_up_process(root->fs_info->cleaner_kthread);
3568
3569 /* wait until ongoing cleanup work done */
3570 down_write(&root->fs_info->cleanup_work_sem);
3571 up_write(&root->fs_info->cleanup_work_sem);
3572
3573 trans = btrfs_join_transaction(root);
3574 if (IS_ERR(trans))
3575 return PTR_ERR(trans);
3576 return btrfs_commit_transaction(trans, root);
3577}
3578
3579int close_ctree(struct btrfs_root *root)
3580{
3581 struct btrfs_fs_info *fs_info = root->fs_info;
3582 int ret;
3583
3584 fs_info->closing = 1;
3585 smp_mb();
3586
3587 /* wait for the uuid_scan task to finish */
3588 down(&fs_info->uuid_tree_rescan_sem);
3589 /* avoid complains from lockdep et al., set sem back to initial state */
3590 up(&fs_info->uuid_tree_rescan_sem);
3591
3592 /* pause restriper - we want to resume on mount */
3593 btrfs_pause_balance(fs_info);
3594
3595 btrfs_dev_replace_suspend_for_unmount(fs_info);
3596
3597 btrfs_scrub_cancel(fs_info);
3598
3599 /* wait for any defraggers to finish */
3600 wait_event(fs_info->transaction_wait,
3601 (atomic_read(&fs_info->defrag_running) == 0));
3602
3603 /* clear out the rbtree of defraggable inodes */
3604 btrfs_cleanup_defrag_inodes(fs_info);
3605
3606 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3607 ret = btrfs_commit_super(root);
3608 if (ret)
3609 btrfs_err(root->fs_info, "commit super ret %d", ret);
3610 }
3611
3612 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3613 btrfs_error_commit_super(root);
3614
3615 kthread_stop(fs_info->transaction_kthread);
3616 kthread_stop(fs_info->cleaner_kthread);
3617
3618 fs_info->closing = 2;
3619 smp_mb();
3620
3621 btrfs_free_qgroup_config(root->fs_info);
3622
3623 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3624 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3625 percpu_counter_sum(&fs_info->delalloc_bytes));
3626 }
3627
3628 btrfs_sysfs_remove_one(fs_info);
3629
3630 del_fs_roots(fs_info);
3631
3632 btrfs_put_block_group_cache(fs_info);
3633
3634 btrfs_free_block_groups(fs_info);
3635
3636 btrfs_stop_all_workers(fs_info);
3637
3638 free_root_pointers(fs_info, 1);
3639
3640 iput(fs_info->btree_inode);
3641
3642#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3643 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3644 btrfsic_unmount(root, fs_info->fs_devices);
3645#endif
3646
3647 btrfs_close_devices(fs_info->fs_devices);
3648 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3649
3650 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3651 percpu_counter_destroy(&fs_info->delalloc_bytes);
3652 percpu_counter_destroy(&fs_info->bio_counter);
3653 bdi_destroy(&fs_info->bdi);
3654 cleanup_srcu_struct(&fs_info->subvol_srcu);
3655
3656 btrfs_free_stripe_hash_table(fs_info);
3657
3658 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3659 root->orphan_block_rsv = NULL;
3660
3661 return 0;
3662}
3663
3664int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3665 int atomic)
3666{
3667 int ret;
3668 struct inode *btree_inode = buf->pages[0]->mapping->host;
3669
3670 ret = extent_buffer_uptodate(buf);
3671 if (!ret)
3672 return ret;
3673
3674 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3675 parent_transid, atomic);
3676 if (ret == -EAGAIN)
3677 return ret;
3678 return !ret;
3679}
3680
3681int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3682{
3683 return set_extent_buffer_uptodate(buf);
3684}
3685
3686void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3687{
3688 struct btrfs_root *root;
3689 u64 transid = btrfs_header_generation(buf);
3690 int was_dirty;
3691
3692#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3693 /*
3694 * This is a fast path so only do this check if we have sanity tests
3695 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3696 * outside of the sanity tests.
3697 */
3698 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3699 return;
3700#endif
3701 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3702 btrfs_assert_tree_locked(buf);
3703 if (transid != root->fs_info->generation)
3704 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3705 "found %llu running %llu\n",
3706 buf->start, transid, root->fs_info->generation);
3707 was_dirty = set_extent_buffer_dirty(buf);
3708 if (!was_dirty)
3709 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3710 buf->len,
3711 root->fs_info->dirty_metadata_batch);
3712}
3713
3714static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3715 int flush_delayed)
3716{
3717 /*
3718 * looks as though older kernels can get into trouble with
3719 * this code, they end up stuck in balance_dirty_pages forever
3720 */
3721 int ret;
3722
3723 if (current->flags & PF_MEMALLOC)
3724 return;
3725
3726 if (flush_delayed)
3727 btrfs_balance_delayed_items(root);
3728
3729 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3730 BTRFS_DIRTY_METADATA_THRESH);
3731 if (ret > 0) {
3732 balance_dirty_pages_ratelimited(
3733 root->fs_info->btree_inode->i_mapping);
3734 }
3735 return;
3736}
3737
3738void btrfs_btree_balance_dirty(struct btrfs_root *root)
3739{
3740 __btrfs_btree_balance_dirty(root, 1);
3741}
3742
3743void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3744{
3745 __btrfs_btree_balance_dirty(root, 0);
3746}
3747
3748int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3749{
3750 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3751 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3752}
3753
3754static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3755 int read_only)
3756{
3757 /*
3758 * Placeholder for checks
3759 */
3760 return 0;
3761}
3762
3763static void btrfs_error_commit_super(struct btrfs_root *root)
3764{
3765 mutex_lock(&root->fs_info->cleaner_mutex);
3766 btrfs_run_delayed_iputs(root);
3767 mutex_unlock(&root->fs_info->cleaner_mutex);
3768
3769 down_write(&root->fs_info->cleanup_work_sem);
3770 up_write(&root->fs_info->cleanup_work_sem);
3771
3772 /* cleanup FS via transaction */
3773 btrfs_cleanup_transaction(root);
3774}
3775
3776static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3777 struct btrfs_root *root)
3778{
3779 struct btrfs_inode *btrfs_inode;
3780 struct list_head splice;
3781
3782 INIT_LIST_HEAD(&splice);
3783
3784 mutex_lock(&root->fs_info->ordered_operations_mutex);
3785 spin_lock(&root->fs_info->ordered_root_lock);
3786
3787 list_splice_init(&t->ordered_operations, &splice);
3788 while (!list_empty(&splice)) {
3789 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3790 ordered_operations);
3791
3792 list_del_init(&btrfs_inode->ordered_operations);
3793 spin_unlock(&root->fs_info->ordered_root_lock);
3794
3795 btrfs_invalidate_inodes(btrfs_inode->root);
3796
3797 spin_lock(&root->fs_info->ordered_root_lock);
3798 }
3799
3800 spin_unlock(&root->fs_info->ordered_root_lock);
3801 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3802}
3803
3804static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3805{
3806 struct btrfs_ordered_extent *ordered;
3807
3808 spin_lock(&root->ordered_extent_lock);
3809 /*
3810 * This will just short circuit the ordered completion stuff which will
3811 * make sure the ordered extent gets properly cleaned up.
3812 */
3813 list_for_each_entry(ordered, &root->ordered_extents,
3814 root_extent_list)
3815 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3816 spin_unlock(&root->ordered_extent_lock);
3817}
3818
3819static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3820{
3821 struct btrfs_root *root;
3822 struct list_head splice;
3823
3824 INIT_LIST_HEAD(&splice);
3825
3826 spin_lock(&fs_info->ordered_root_lock);
3827 list_splice_init(&fs_info->ordered_roots, &splice);
3828 while (!list_empty(&splice)) {
3829 root = list_first_entry(&splice, struct btrfs_root,
3830 ordered_root);
3831 list_move_tail(&root->ordered_root,
3832 &fs_info->ordered_roots);
3833
3834 spin_unlock(&fs_info->ordered_root_lock);
3835 btrfs_destroy_ordered_extents(root);
3836
3837 cond_resched();
3838 spin_lock(&fs_info->ordered_root_lock);
3839 }
3840 spin_unlock(&fs_info->ordered_root_lock);
3841}
3842
3843static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3844 struct btrfs_root *root)
3845{
3846 struct rb_node *node;
3847 struct btrfs_delayed_ref_root *delayed_refs;
3848 struct btrfs_delayed_ref_node *ref;
3849 int ret = 0;
3850
3851 delayed_refs = &trans->delayed_refs;
3852
3853 spin_lock(&delayed_refs->lock);
3854 if (atomic_read(&delayed_refs->num_entries) == 0) {
3855 spin_unlock(&delayed_refs->lock);
3856 btrfs_info(root->fs_info, "delayed_refs has NO entry");
3857 return ret;
3858 }
3859
3860 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3861 struct btrfs_delayed_ref_head *head;
3862 bool pin_bytes = false;
3863
3864 head = rb_entry(node, struct btrfs_delayed_ref_head,
3865 href_node);
3866 if (!mutex_trylock(&head->mutex)) {
3867 atomic_inc(&head->node.refs);
3868 spin_unlock(&delayed_refs->lock);
3869
3870 mutex_lock(&head->mutex);
3871 mutex_unlock(&head->mutex);
3872 btrfs_put_delayed_ref(&head->node);
3873 spin_lock(&delayed_refs->lock);
3874 continue;
3875 }
3876 spin_lock(&head->lock);
3877 while ((node = rb_first(&head->ref_root)) != NULL) {
3878 ref = rb_entry(node, struct btrfs_delayed_ref_node,
3879 rb_node);
3880 ref->in_tree = 0;
3881 rb_erase(&ref->rb_node, &head->ref_root);
3882 atomic_dec(&delayed_refs->num_entries);
3883 btrfs_put_delayed_ref(ref);
3884 }
3885 if (head->must_insert_reserved)
3886 pin_bytes = true;
3887 btrfs_free_delayed_extent_op(head->extent_op);
3888 delayed_refs->num_heads--;
3889 if (head->processing == 0)
3890 delayed_refs->num_heads_ready--;
3891 atomic_dec(&delayed_refs->num_entries);
3892 head->node.in_tree = 0;
3893 rb_erase(&head->href_node, &delayed_refs->href_root);
3894 spin_unlock(&head->lock);
3895 spin_unlock(&delayed_refs->lock);
3896 mutex_unlock(&head->mutex);
3897
3898 if (pin_bytes)
3899 btrfs_pin_extent(root, head->node.bytenr,
3900 head->node.num_bytes, 1);
3901 btrfs_put_delayed_ref(&head->node);
3902 cond_resched();
3903 spin_lock(&delayed_refs->lock);
3904 }
3905
3906 spin_unlock(&delayed_refs->lock);
3907
3908 return ret;
3909}
3910
3911static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3912{
3913 struct btrfs_inode *btrfs_inode;
3914 struct list_head splice;
3915
3916 INIT_LIST_HEAD(&splice);
3917
3918 spin_lock(&root->delalloc_lock);
3919 list_splice_init(&root->delalloc_inodes, &splice);
3920
3921 while (!list_empty(&splice)) {
3922 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3923 delalloc_inodes);
3924
3925 list_del_init(&btrfs_inode->delalloc_inodes);
3926 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3927 &btrfs_inode->runtime_flags);
3928 spin_unlock(&root->delalloc_lock);
3929
3930 btrfs_invalidate_inodes(btrfs_inode->root);
3931
3932 spin_lock(&root->delalloc_lock);
3933 }
3934
3935 spin_unlock(&root->delalloc_lock);
3936}
3937
3938static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3939{
3940 struct btrfs_root *root;
3941 struct list_head splice;
3942
3943 INIT_LIST_HEAD(&splice);
3944
3945 spin_lock(&fs_info->delalloc_root_lock);
3946 list_splice_init(&fs_info->delalloc_roots, &splice);
3947 while (!list_empty(&splice)) {
3948 root = list_first_entry(&splice, struct btrfs_root,
3949 delalloc_root);
3950 list_del_init(&root->delalloc_root);
3951 root = btrfs_grab_fs_root(root);
3952 BUG_ON(!root);
3953 spin_unlock(&fs_info->delalloc_root_lock);
3954
3955 btrfs_destroy_delalloc_inodes(root);
3956 btrfs_put_fs_root(root);
3957
3958 spin_lock(&fs_info->delalloc_root_lock);
3959 }
3960 spin_unlock(&fs_info->delalloc_root_lock);
3961}
3962
3963static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3964 struct extent_io_tree *dirty_pages,
3965 int mark)
3966{
3967 int ret;
3968 struct extent_buffer *eb;
3969 u64 start = 0;
3970 u64 end;
3971
3972 while (1) {
3973 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3974 mark, NULL);
3975 if (ret)
3976 break;
3977
3978 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3979 while (start <= end) {
3980 eb = btrfs_find_tree_block(root, start,
3981 root->leafsize);
3982 start += root->leafsize;
3983 if (!eb)
3984 continue;
3985 wait_on_extent_buffer_writeback(eb);
3986
3987 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3988 &eb->bflags))
3989 clear_extent_buffer_dirty(eb);
3990 free_extent_buffer_stale(eb);
3991 }
3992 }
3993
3994 return ret;
3995}
3996
3997static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3998 struct extent_io_tree *pinned_extents)
3999{
4000 struct extent_io_tree *unpin;
4001 u64 start;
4002 u64 end;
4003 int ret;
4004 bool loop = true;
4005
4006 unpin = pinned_extents;
4007again:
4008 while (1) {
4009 ret = find_first_extent_bit(unpin, 0, &start, &end,
4010 EXTENT_DIRTY, NULL);
4011 if (ret)
4012 break;
4013
4014 /* opt_discard */
4015 if (btrfs_test_opt(root, DISCARD))
4016 ret = btrfs_error_discard_extent(root, start,
4017 end + 1 - start,
4018 NULL);
4019
4020 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4021 btrfs_error_unpin_extent_range(root, start, end);
4022 cond_resched();
4023 }
4024
4025 if (loop) {
4026 if (unpin == &root->fs_info->freed_extents[0])
4027 unpin = &root->fs_info->freed_extents[1];
4028 else
4029 unpin = &root->fs_info->freed_extents[0];
4030 loop = false;
4031 goto again;
4032 }
4033
4034 return 0;
4035}
4036
4037void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4038 struct btrfs_root *root)
4039{
4040 btrfs_destroy_ordered_operations(cur_trans, root);
4041
4042 btrfs_destroy_delayed_refs(cur_trans, root);
4043
4044 cur_trans->state = TRANS_STATE_COMMIT_START;
4045 wake_up(&root->fs_info->transaction_blocked_wait);
4046
4047 cur_trans->state = TRANS_STATE_UNBLOCKED;
4048 wake_up(&root->fs_info->transaction_wait);
4049
4050 btrfs_destroy_delayed_inodes(root);
4051 btrfs_assert_delayed_root_empty(root);
4052
4053 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4054 EXTENT_DIRTY);
4055 btrfs_destroy_pinned_extent(root,
4056 root->fs_info->pinned_extents);
4057
4058 cur_trans->state =TRANS_STATE_COMPLETED;
4059 wake_up(&cur_trans->commit_wait);
4060
4061 /*
4062 memset(cur_trans, 0, sizeof(*cur_trans));
4063 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4064 */
4065}
4066
4067static int btrfs_cleanup_transaction(struct btrfs_root *root)
4068{
4069 struct btrfs_transaction *t;
4070
4071 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4072
4073 spin_lock(&root->fs_info->trans_lock);
4074 while (!list_empty(&root->fs_info->trans_list)) {
4075 t = list_first_entry(&root->fs_info->trans_list,
4076 struct btrfs_transaction, list);
4077 if (t->state >= TRANS_STATE_COMMIT_START) {
4078 atomic_inc(&t->use_count);
4079 spin_unlock(&root->fs_info->trans_lock);
4080 btrfs_wait_for_commit(root, t->transid);
4081 btrfs_put_transaction(t);
4082 spin_lock(&root->fs_info->trans_lock);
4083 continue;
4084 }
4085 if (t == root->fs_info->running_transaction) {
4086 t->state = TRANS_STATE_COMMIT_DOING;
4087 spin_unlock(&root->fs_info->trans_lock);
4088 /*
4089 * We wait for 0 num_writers since we don't hold a trans
4090 * handle open currently for this transaction.
4091 */
4092 wait_event(t->writer_wait,
4093 atomic_read(&t->num_writers) == 0);
4094 } else {
4095 spin_unlock(&root->fs_info->trans_lock);
4096 }
4097 btrfs_cleanup_one_transaction(t, root);
4098
4099 spin_lock(&root->fs_info->trans_lock);
4100 if (t == root->fs_info->running_transaction)
4101 root->fs_info->running_transaction = NULL;
4102 list_del_init(&t->list);
4103 spin_unlock(&root->fs_info->trans_lock);
4104
4105 btrfs_put_transaction(t);
4106 trace_btrfs_transaction_commit(root);
4107 spin_lock(&root->fs_info->trans_lock);
4108 }
4109 spin_unlock(&root->fs_info->trans_lock);
4110 btrfs_destroy_all_ordered_extents(root->fs_info);
4111 btrfs_destroy_delayed_inodes(root);
4112 btrfs_assert_delayed_root_empty(root);
4113 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4114 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4115 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4116
4117 return 0;
4118}
4119
4120static struct extent_io_ops btree_extent_io_ops = {
4121 .readpage_end_io_hook = btree_readpage_end_io_hook,
4122 .readpage_io_failed_hook = btree_io_failed_hook,
4123 .submit_bio_hook = btree_submit_bio_hook,
4124 /* note we're sharing with inode.c for the merge bio hook */
4125 .merge_bio_hook = btrfs_merge_bio_hook,
4126};