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/slab.h>
29#include <linux/migrate.h>
30#include <linux/ratelimit.h>
31#include <linux/uuid.h>
32#include <linux/semaphore.h>
33#include <asm/unaligned.h>
34#include "ctree.h"
35#include "disk-io.h"
36#include "hash.h"
37#include "transaction.h"
38#include "btrfs_inode.h"
39#include "volumes.h"
40#include "print-tree.h"
41#include "locking.h"
42#include "tree-log.h"
43#include "free-space-cache.h"
44#include "free-space-tree.h"
45#include "inode-map.h"
46#include "check-integrity.h"
47#include "rcu-string.h"
48#include "dev-replace.h"
49#include "raid56.h"
50#include "sysfs.h"
51#include "qgroup.h"
52#include "compression.h"
53
54#ifdef CONFIG_X86
55#include <asm/cpufeature.h>
56#endif
57
58#define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
59 BTRFS_HEADER_FLAG_RELOC |\
60 BTRFS_SUPER_FLAG_ERROR |\
61 BTRFS_SUPER_FLAG_SEEDING |\
62 BTRFS_SUPER_FLAG_METADUMP)
63
64static const struct extent_io_ops btree_extent_io_ops;
65static void end_workqueue_fn(struct btrfs_work *work);
66static void free_fs_root(struct btrfs_root *root);
67static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
68 int read_only);
69static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
70static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
71 struct btrfs_root *root);
72static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
73static int btrfs_destroy_marked_extents(struct btrfs_root *root,
74 struct extent_io_tree *dirty_pages,
75 int mark);
76static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
77 struct extent_io_tree *pinned_extents);
78static int btrfs_cleanup_transaction(struct btrfs_root *root);
79static void btrfs_error_commit_super(struct btrfs_root *root);
80
81/*
82 * btrfs_end_io_wq structs are used to do processing in task context when an IO
83 * is complete. This is used during reads to verify checksums, and it is used
84 * by writes to insert metadata for new file extents after IO is complete.
85 */
86struct btrfs_end_io_wq {
87 struct bio *bio;
88 bio_end_io_t *end_io;
89 void *private;
90 struct btrfs_fs_info *info;
91 int error;
92 enum btrfs_wq_endio_type metadata;
93 struct list_head list;
94 struct btrfs_work work;
95};
96
97static struct kmem_cache *btrfs_end_io_wq_cache;
98
99int __init btrfs_end_io_wq_init(void)
100{
101 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
102 sizeof(struct btrfs_end_io_wq),
103 0,
104 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
105 NULL);
106 if (!btrfs_end_io_wq_cache)
107 return -ENOMEM;
108 return 0;
109}
110
111void btrfs_end_io_wq_exit(void)
112{
113 kmem_cache_destroy(btrfs_end_io_wq_cache);
114}
115
116/*
117 * async submit bios are used to offload expensive checksumming
118 * onto the worker threads. They checksum file and metadata bios
119 * just before they are sent down the IO stack.
120 */
121struct async_submit_bio {
122 struct inode *inode;
123 struct bio *bio;
124 struct list_head list;
125 extent_submit_bio_hook_t *submit_bio_start;
126 extent_submit_bio_hook_t *submit_bio_done;
127 int rw;
128 int mirror_num;
129 unsigned long bio_flags;
130 /*
131 * bio_offset is optional, can be used if the pages in the bio
132 * can't tell us where in the file the bio should go
133 */
134 u64 bio_offset;
135 struct btrfs_work work;
136 int error;
137};
138
139/*
140 * Lockdep class keys for extent_buffer->lock's in this root. For a given
141 * eb, the lockdep key is determined by the btrfs_root it belongs to and
142 * the level the eb occupies in the tree.
143 *
144 * Different roots are used for different purposes and may nest inside each
145 * other and they require separate keysets. As lockdep keys should be
146 * static, assign keysets according to the purpose of the root as indicated
147 * by btrfs_root->objectid. This ensures that all special purpose roots
148 * have separate keysets.
149 *
150 * Lock-nesting across peer nodes is always done with the immediate parent
151 * node locked thus preventing deadlock. As lockdep doesn't know this, use
152 * subclass to avoid triggering lockdep warning in such cases.
153 *
154 * The key is set by the readpage_end_io_hook after the buffer has passed
155 * csum validation but before the pages are unlocked. It is also set by
156 * btrfs_init_new_buffer on freshly allocated blocks.
157 *
158 * We also add a check to make sure the highest level of the tree is the
159 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
160 * needs update as well.
161 */
162#ifdef CONFIG_DEBUG_LOCK_ALLOC
163# if BTRFS_MAX_LEVEL != 8
164# error
165# endif
166
167static struct btrfs_lockdep_keyset {
168 u64 id; /* root objectid */
169 const char *name_stem; /* lock name stem */
170 char names[BTRFS_MAX_LEVEL + 1][20];
171 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
172} btrfs_lockdep_keysets[] = {
173 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
174 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
175 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
176 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
177 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
178 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
179 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
180 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
181 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
182 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
183 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
184 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
185 { .id = 0, .name_stem = "tree" },
186};
187
188void __init btrfs_init_lockdep(void)
189{
190 int i, j;
191
192 /* initialize lockdep class names */
193 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
194 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
195
196 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
197 snprintf(ks->names[j], sizeof(ks->names[j]),
198 "btrfs-%s-%02d", ks->name_stem, j);
199 }
200}
201
202void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
203 int level)
204{
205 struct btrfs_lockdep_keyset *ks;
206
207 BUG_ON(level >= ARRAY_SIZE(ks->keys));
208
209 /* find the matching keyset, id 0 is the default entry */
210 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
211 if (ks->id == objectid)
212 break;
213
214 lockdep_set_class_and_name(&eb->lock,
215 &ks->keys[level], ks->names[level]);
216}
217
218#endif
219
220/*
221 * extents on the btree inode are pretty simple, there's one extent
222 * that covers the entire device
223 */
224static struct extent_map *btree_get_extent(struct inode *inode,
225 struct page *page, size_t pg_offset, u64 start, u64 len,
226 int create)
227{
228 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
229 struct extent_map *em;
230 int ret;
231
232 read_lock(&em_tree->lock);
233 em = lookup_extent_mapping(em_tree, start, len);
234 if (em) {
235 em->bdev =
236 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
237 read_unlock(&em_tree->lock);
238 goto out;
239 }
240 read_unlock(&em_tree->lock);
241
242 em = alloc_extent_map();
243 if (!em) {
244 em = ERR_PTR(-ENOMEM);
245 goto out;
246 }
247 em->start = 0;
248 em->len = (u64)-1;
249 em->block_len = (u64)-1;
250 em->block_start = 0;
251 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
252
253 write_lock(&em_tree->lock);
254 ret = add_extent_mapping(em_tree, em, 0);
255 if (ret == -EEXIST) {
256 free_extent_map(em);
257 em = lookup_extent_mapping(em_tree, start, len);
258 if (!em)
259 em = ERR_PTR(-EIO);
260 } else if (ret) {
261 free_extent_map(em);
262 em = ERR_PTR(ret);
263 }
264 write_unlock(&em_tree->lock);
265
266out:
267 return em;
268}
269
270u32 btrfs_csum_data(char *data, u32 seed, size_t len)
271{
272 return btrfs_crc32c(seed, data, len);
273}
274
275void btrfs_csum_final(u32 crc, char *result)
276{
277 put_unaligned_le32(~crc, result);
278}
279
280/*
281 * compute the csum for a btree block, and either verify it or write it
282 * into the csum field of the block.
283 */
284static int csum_tree_block(struct btrfs_fs_info *fs_info,
285 struct extent_buffer *buf,
286 int verify)
287{
288 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
289 char *result = NULL;
290 unsigned long len;
291 unsigned long cur_len;
292 unsigned long offset = BTRFS_CSUM_SIZE;
293 char *kaddr;
294 unsigned long map_start;
295 unsigned long map_len;
296 int err;
297 u32 crc = ~(u32)0;
298 unsigned long inline_result;
299
300 len = buf->len - offset;
301 while (len > 0) {
302 err = map_private_extent_buffer(buf, offset, 32,
303 &kaddr, &map_start, &map_len);
304 if (err)
305 return err;
306 cur_len = min(len, map_len - (offset - map_start));
307 crc = btrfs_csum_data(kaddr + offset - map_start,
308 crc, cur_len);
309 len -= cur_len;
310 offset += cur_len;
311 }
312 if (csum_size > sizeof(inline_result)) {
313 result = kzalloc(csum_size, GFP_NOFS);
314 if (!result)
315 return -ENOMEM;
316 } else {
317 result = (char *)&inline_result;
318 }
319
320 btrfs_csum_final(crc, result);
321
322 if (verify) {
323 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
324 u32 val;
325 u32 found = 0;
326 memcpy(&found, result, csum_size);
327
328 read_extent_buffer(buf, &val, 0, csum_size);
329 btrfs_warn_rl(fs_info,
330 "%s checksum verify failed on %llu wanted %X found %X "
331 "level %d",
332 fs_info->sb->s_id, buf->start,
333 val, found, btrfs_header_level(buf));
334 if (result != (char *)&inline_result)
335 kfree(result);
336 return -EUCLEAN;
337 }
338 } else {
339 write_extent_buffer(buf, result, 0, csum_size);
340 }
341 if (result != (char *)&inline_result)
342 kfree(result);
343 return 0;
344}
345
346/*
347 * we can't consider a given block up to date unless the transid of the
348 * block matches the transid in the parent node's pointer. This is how we
349 * detect blocks that either didn't get written at all or got written
350 * in the wrong place.
351 */
352static int verify_parent_transid(struct extent_io_tree *io_tree,
353 struct extent_buffer *eb, u64 parent_transid,
354 int atomic)
355{
356 struct extent_state *cached_state = NULL;
357 int ret;
358 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
359
360 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
361 return 0;
362
363 if (atomic)
364 return -EAGAIN;
365
366 if (need_lock) {
367 btrfs_tree_read_lock(eb);
368 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
369 }
370
371 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
372 &cached_state);
373 if (extent_buffer_uptodate(eb) &&
374 btrfs_header_generation(eb) == parent_transid) {
375 ret = 0;
376 goto out;
377 }
378 btrfs_err_rl(eb->fs_info,
379 "parent transid verify failed on %llu wanted %llu found %llu",
380 eb->start,
381 parent_transid, btrfs_header_generation(eb));
382 ret = 1;
383
384 /*
385 * Things reading via commit roots that don't have normal protection,
386 * like send, can have a really old block in cache that may point at a
387 * block that has been free'd and re-allocated. So don't clear uptodate
388 * if we find an eb that is under IO (dirty/writeback) because we could
389 * end up reading in the stale data and then writing it back out and
390 * making everybody very sad.
391 */
392 if (!extent_buffer_under_io(eb))
393 clear_extent_buffer_uptodate(eb);
394out:
395 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
396 &cached_state, GFP_NOFS);
397 if (need_lock)
398 btrfs_tree_read_unlock_blocking(eb);
399 return ret;
400}
401
402/*
403 * Return 0 if the superblock checksum type matches the checksum value of that
404 * algorithm. Pass the raw disk superblock data.
405 */
406static int btrfs_check_super_csum(char *raw_disk_sb)
407{
408 struct btrfs_super_block *disk_sb =
409 (struct btrfs_super_block *)raw_disk_sb;
410 u16 csum_type = btrfs_super_csum_type(disk_sb);
411 int ret = 0;
412
413 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
414 u32 crc = ~(u32)0;
415 const int csum_size = sizeof(crc);
416 char result[csum_size];
417
418 /*
419 * The super_block structure does not span the whole
420 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
421 * is filled with zeros and is included in the checkum.
422 */
423 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
424 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
425 btrfs_csum_final(crc, result);
426
427 if (memcmp(raw_disk_sb, result, csum_size))
428 ret = 1;
429 }
430
431 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
432 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
433 csum_type);
434 ret = 1;
435 }
436
437 return ret;
438}
439
440/*
441 * helper to read a given tree block, doing retries as required when
442 * the checksums don't match and we have alternate mirrors to try.
443 */
444static int btree_read_extent_buffer_pages(struct btrfs_root *root,
445 struct extent_buffer *eb,
446 u64 start, u64 parent_transid)
447{
448 struct extent_io_tree *io_tree;
449 int failed = 0;
450 int ret;
451 int num_copies = 0;
452 int mirror_num = 0;
453 int failed_mirror = 0;
454
455 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
456 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
457 while (1) {
458 ret = read_extent_buffer_pages(io_tree, eb, start,
459 WAIT_COMPLETE,
460 btree_get_extent, mirror_num);
461 if (!ret) {
462 if (!verify_parent_transid(io_tree, eb,
463 parent_transid, 0))
464 break;
465 else
466 ret = -EIO;
467 }
468
469 /*
470 * This buffer's crc is fine, but its contents are corrupted, so
471 * there is no reason to read the other copies, they won't be
472 * any less wrong.
473 */
474 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
475 break;
476
477 num_copies = btrfs_num_copies(root->fs_info,
478 eb->start, eb->len);
479 if (num_copies == 1)
480 break;
481
482 if (!failed_mirror) {
483 failed = 1;
484 failed_mirror = eb->read_mirror;
485 }
486
487 mirror_num++;
488 if (mirror_num == failed_mirror)
489 mirror_num++;
490
491 if (mirror_num > num_copies)
492 break;
493 }
494
495 if (failed && !ret && failed_mirror)
496 repair_eb_io_failure(root, eb, failed_mirror);
497
498 return ret;
499}
500
501/*
502 * checksum a dirty tree block before IO. This has extra checks to make sure
503 * we only fill in the checksum field in the first page of a multi-page block
504 */
505
506static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
507{
508 u64 start = page_offset(page);
509 u64 found_start;
510 struct extent_buffer *eb;
511
512 eb = (struct extent_buffer *)page->private;
513 if (page != eb->pages[0])
514 return 0;
515
516 found_start = btrfs_header_bytenr(eb);
517 /*
518 * Please do not consolidate these warnings into a single if.
519 * It is useful to know what went wrong.
520 */
521 if (WARN_ON(found_start != start))
522 return -EUCLEAN;
523 if (WARN_ON(!PageUptodate(page)))
524 return -EUCLEAN;
525
526 ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
527 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
528
529 return csum_tree_block(fs_info, eb, 0);
530}
531
532static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
533 struct extent_buffer *eb)
534{
535 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
536 u8 fsid[BTRFS_UUID_SIZE];
537 int ret = 1;
538
539 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
540 while (fs_devices) {
541 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
542 ret = 0;
543 break;
544 }
545 fs_devices = fs_devices->seed;
546 }
547 return ret;
548}
549
550#define CORRUPT(reason, eb, root, slot) \
551 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
552 "root=%llu, slot=%d", reason, \
553 btrfs_header_bytenr(eb), root->objectid, slot)
554
555static noinline int check_leaf(struct btrfs_root *root,
556 struct extent_buffer *leaf)
557{
558 struct btrfs_key key;
559 struct btrfs_key leaf_key;
560 u32 nritems = btrfs_header_nritems(leaf);
561 int slot;
562
563 if (nritems == 0)
564 return 0;
565
566 /* Check the 0 item */
567 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
568 BTRFS_LEAF_DATA_SIZE(root)) {
569 CORRUPT("invalid item offset size pair", leaf, root, 0);
570 return -EIO;
571 }
572
573 /*
574 * Check to make sure each items keys are in the correct order and their
575 * offsets make sense. We only have to loop through nritems-1 because
576 * we check the current slot against the next slot, which verifies the
577 * next slot's offset+size makes sense and that the current's slot
578 * offset is correct.
579 */
580 for (slot = 0; slot < nritems - 1; slot++) {
581 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
582 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
583
584 /* Make sure the keys are in the right order */
585 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
586 CORRUPT("bad key order", leaf, root, slot);
587 return -EIO;
588 }
589
590 /*
591 * Make sure the offset and ends are right, remember that the
592 * item data starts at the end of the leaf and grows towards the
593 * front.
594 */
595 if (btrfs_item_offset_nr(leaf, slot) !=
596 btrfs_item_end_nr(leaf, slot + 1)) {
597 CORRUPT("slot offset bad", leaf, root, slot);
598 return -EIO;
599 }
600
601 /*
602 * Check to make sure that we don't point outside of the leaf,
603 * just incase all the items are consistent to eachother, but
604 * all point outside of the leaf.
605 */
606 if (btrfs_item_end_nr(leaf, slot) >
607 BTRFS_LEAF_DATA_SIZE(root)) {
608 CORRUPT("slot end outside of leaf", leaf, root, slot);
609 return -EIO;
610 }
611 }
612
613 return 0;
614}
615
616static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
617 u64 phy_offset, struct page *page,
618 u64 start, u64 end, int mirror)
619{
620 u64 found_start;
621 int found_level;
622 struct extent_buffer *eb;
623 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
624 struct btrfs_fs_info *fs_info = root->fs_info;
625 int ret = 0;
626 int reads_done;
627
628 if (!page->private)
629 goto out;
630
631 eb = (struct extent_buffer *)page->private;
632
633 /* the pending IO might have been the only thing that kept this buffer
634 * in memory. Make sure we have a ref for all this other checks
635 */
636 extent_buffer_get(eb);
637
638 reads_done = atomic_dec_and_test(&eb->io_pages);
639 if (!reads_done)
640 goto err;
641
642 eb->read_mirror = mirror;
643 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
644 ret = -EIO;
645 goto err;
646 }
647
648 found_start = btrfs_header_bytenr(eb);
649 if (found_start != eb->start) {
650 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
651 found_start, eb->start);
652 ret = -EIO;
653 goto err;
654 }
655 if (check_tree_block_fsid(fs_info, eb)) {
656 btrfs_err_rl(fs_info, "bad fsid on block %llu",
657 eb->start);
658 ret = -EIO;
659 goto err;
660 }
661 found_level = btrfs_header_level(eb);
662 if (found_level >= BTRFS_MAX_LEVEL) {
663 btrfs_err(fs_info, "bad tree block level %d",
664 (int)btrfs_header_level(eb));
665 ret = -EIO;
666 goto err;
667 }
668
669 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
670 eb, found_level);
671
672 ret = csum_tree_block(fs_info, eb, 1);
673 if (ret)
674 goto err;
675
676 /*
677 * If this is a leaf block and it is corrupt, set the corrupt bit so
678 * that we don't try and read the other copies of this block, just
679 * return -EIO.
680 */
681 if (found_level == 0 && check_leaf(root, eb)) {
682 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
683 ret = -EIO;
684 }
685
686 if (!ret)
687 set_extent_buffer_uptodate(eb);
688err:
689 if (reads_done &&
690 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
691 btree_readahead_hook(fs_info, eb, eb->start, ret);
692
693 if (ret) {
694 /*
695 * our io error hook is going to dec the io pages
696 * again, we have to make sure it has something
697 * to decrement
698 */
699 atomic_inc(&eb->io_pages);
700 clear_extent_buffer_uptodate(eb);
701 }
702 free_extent_buffer(eb);
703out:
704 return ret;
705}
706
707static int btree_io_failed_hook(struct page *page, int failed_mirror)
708{
709 struct extent_buffer *eb;
710
711 eb = (struct extent_buffer *)page->private;
712 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
713 eb->read_mirror = failed_mirror;
714 atomic_dec(&eb->io_pages);
715 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
716 btree_readahead_hook(eb->fs_info, eb, eb->start, -EIO);
717 return -EIO; /* we fixed nothing */
718}
719
720static void end_workqueue_bio(struct bio *bio)
721{
722 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
723 struct btrfs_fs_info *fs_info;
724 struct btrfs_workqueue *wq;
725 btrfs_work_func_t func;
726
727 fs_info = end_io_wq->info;
728 end_io_wq->error = bio->bi_error;
729
730 if (bio->bi_rw & REQ_WRITE) {
731 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
732 wq = fs_info->endio_meta_write_workers;
733 func = btrfs_endio_meta_write_helper;
734 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
735 wq = fs_info->endio_freespace_worker;
736 func = btrfs_freespace_write_helper;
737 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
738 wq = fs_info->endio_raid56_workers;
739 func = btrfs_endio_raid56_helper;
740 } else {
741 wq = fs_info->endio_write_workers;
742 func = btrfs_endio_write_helper;
743 }
744 } else {
745 if (unlikely(end_io_wq->metadata ==
746 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
747 wq = fs_info->endio_repair_workers;
748 func = btrfs_endio_repair_helper;
749 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
750 wq = fs_info->endio_raid56_workers;
751 func = btrfs_endio_raid56_helper;
752 } else if (end_io_wq->metadata) {
753 wq = fs_info->endio_meta_workers;
754 func = btrfs_endio_meta_helper;
755 } else {
756 wq = fs_info->endio_workers;
757 func = btrfs_endio_helper;
758 }
759 }
760
761 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
762 btrfs_queue_work(wq, &end_io_wq->work);
763}
764
765int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
766 enum btrfs_wq_endio_type metadata)
767{
768 struct btrfs_end_io_wq *end_io_wq;
769
770 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
771 if (!end_io_wq)
772 return -ENOMEM;
773
774 end_io_wq->private = bio->bi_private;
775 end_io_wq->end_io = bio->bi_end_io;
776 end_io_wq->info = info;
777 end_io_wq->error = 0;
778 end_io_wq->bio = bio;
779 end_io_wq->metadata = metadata;
780
781 bio->bi_private = end_io_wq;
782 bio->bi_end_io = end_workqueue_bio;
783 return 0;
784}
785
786unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
787{
788 unsigned long limit = min_t(unsigned long,
789 info->thread_pool_size,
790 info->fs_devices->open_devices);
791 return 256 * limit;
792}
793
794static void run_one_async_start(struct btrfs_work *work)
795{
796 struct async_submit_bio *async;
797 int ret;
798
799 async = container_of(work, struct async_submit_bio, work);
800 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
801 async->mirror_num, async->bio_flags,
802 async->bio_offset);
803 if (ret)
804 async->error = ret;
805}
806
807static void run_one_async_done(struct btrfs_work *work)
808{
809 struct btrfs_fs_info *fs_info;
810 struct async_submit_bio *async;
811 int limit;
812
813 async = container_of(work, struct async_submit_bio, work);
814 fs_info = BTRFS_I(async->inode)->root->fs_info;
815
816 limit = btrfs_async_submit_limit(fs_info);
817 limit = limit * 2 / 3;
818
819 /*
820 * atomic_dec_return implies a barrier for waitqueue_active
821 */
822 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
823 waitqueue_active(&fs_info->async_submit_wait))
824 wake_up(&fs_info->async_submit_wait);
825
826 /* If an error occurred we just want to clean up the bio and move on */
827 if (async->error) {
828 async->bio->bi_error = async->error;
829 bio_endio(async->bio);
830 return;
831 }
832
833 async->submit_bio_done(async->inode, async->rw, async->bio,
834 async->mirror_num, async->bio_flags,
835 async->bio_offset);
836}
837
838static void run_one_async_free(struct btrfs_work *work)
839{
840 struct async_submit_bio *async;
841
842 async = container_of(work, struct async_submit_bio, work);
843 kfree(async);
844}
845
846int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
847 int rw, struct bio *bio, int mirror_num,
848 unsigned long bio_flags,
849 u64 bio_offset,
850 extent_submit_bio_hook_t *submit_bio_start,
851 extent_submit_bio_hook_t *submit_bio_done)
852{
853 struct async_submit_bio *async;
854
855 async = kmalloc(sizeof(*async), GFP_NOFS);
856 if (!async)
857 return -ENOMEM;
858
859 async->inode = inode;
860 async->rw = rw;
861 async->bio = bio;
862 async->mirror_num = mirror_num;
863 async->submit_bio_start = submit_bio_start;
864 async->submit_bio_done = submit_bio_done;
865
866 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
867 run_one_async_done, run_one_async_free);
868
869 async->bio_flags = bio_flags;
870 async->bio_offset = bio_offset;
871
872 async->error = 0;
873
874 atomic_inc(&fs_info->nr_async_submits);
875
876 if (rw & REQ_SYNC)
877 btrfs_set_work_high_priority(&async->work);
878
879 btrfs_queue_work(fs_info->workers, &async->work);
880
881 while (atomic_read(&fs_info->async_submit_draining) &&
882 atomic_read(&fs_info->nr_async_submits)) {
883 wait_event(fs_info->async_submit_wait,
884 (atomic_read(&fs_info->nr_async_submits) == 0));
885 }
886
887 return 0;
888}
889
890static int btree_csum_one_bio(struct bio *bio)
891{
892 struct bio_vec *bvec;
893 struct btrfs_root *root;
894 int i, ret = 0;
895
896 bio_for_each_segment_all(bvec, bio, i) {
897 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
898 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
899 if (ret)
900 break;
901 }
902
903 return ret;
904}
905
906static int __btree_submit_bio_start(struct inode *inode, int rw,
907 struct bio *bio, int mirror_num,
908 unsigned long bio_flags,
909 u64 bio_offset)
910{
911 /*
912 * when we're called for a write, we're already in the async
913 * submission context. Just jump into btrfs_map_bio
914 */
915 return btree_csum_one_bio(bio);
916}
917
918static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
919 int mirror_num, unsigned long bio_flags,
920 u64 bio_offset)
921{
922 int ret;
923
924 /*
925 * when we're called for a write, we're already in the async
926 * submission context. Just jump into btrfs_map_bio
927 */
928 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
929 if (ret) {
930 bio->bi_error = ret;
931 bio_endio(bio);
932 }
933 return ret;
934}
935
936static int check_async_write(struct inode *inode, unsigned long bio_flags)
937{
938 if (bio_flags & EXTENT_BIO_TREE_LOG)
939 return 0;
940#ifdef CONFIG_X86
941 if (static_cpu_has(X86_FEATURE_XMM4_2))
942 return 0;
943#endif
944 return 1;
945}
946
947static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
948 int mirror_num, unsigned long bio_flags,
949 u64 bio_offset)
950{
951 int async = check_async_write(inode, bio_flags);
952 int ret;
953
954 if (!(rw & REQ_WRITE)) {
955 /*
956 * called for a read, do the setup so that checksum validation
957 * can happen in the async kernel threads
958 */
959 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
960 bio, BTRFS_WQ_ENDIO_METADATA);
961 if (ret)
962 goto out_w_error;
963 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
964 mirror_num, 0);
965 } else if (!async) {
966 ret = btree_csum_one_bio(bio);
967 if (ret)
968 goto out_w_error;
969 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
970 mirror_num, 0);
971 } else {
972 /*
973 * kthread helpers are used to submit writes so that
974 * checksumming can happen in parallel across all CPUs
975 */
976 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
977 inode, rw, bio, mirror_num, 0,
978 bio_offset,
979 __btree_submit_bio_start,
980 __btree_submit_bio_done);
981 }
982
983 if (ret)
984 goto out_w_error;
985 return 0;
986
987out_w_error:
988 bio->bi_error = ret;
989 bio_endio(bio);
990 return ret;
991}
992
993#ifdef CONFIG_MIGRATION
994static int btree_migratepage(struct address_space *mapping,
995 struct page *newpage, struct page *page,
996 enum migrate_mode mode)
997{
998 /*
999 * we can't safely write a btree page from here,
1000 * we haven't done the locking hook
1001 */
1002 if (PageDirty(page))
1003 return -EAGAIN;
1004 /*
1005 * Buffers may be managed in a filesystem specific way.
1006 * We must have no buffers or drop them.
1007 */
1008 if (page_has_private(page) &&
1009 !try_to_release_page(page, GFP_KERNEL))
1010 return -EAGAIN;
1011 return migrate_page(mapping, newpage, page, mode);
1012}
1013#endif
1014
1015
1016static int btree_writepages(struct address_space *mapping,
1017 struct writeback_control *wbc)
1018{
1019 struct btrfs_fs_info *fs_info;
1020 int ret;
1021
1022 if (wbc->sync_mode == WB_SYNC_NONE) {
1023
1024 if (wbc->for_kupdate)
1025 return 0;
1026
1027 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1028 /* this is a bit racy, but that's ok */
1029 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1030 BTRFS_DIRTY_METADATA_THRESH);
1031 if (ret < 0)
1032 return 0;
1033 }
1034 return btree_write_cache_pages(mapping, wbc);
1035}
1036
1037static int btree_readpage(struct file *file, struct page *page)
1038{
1039 struct extent_io_tree *tree;
1040 tree = &BTRFS_I(page->mapping->host)->io_tree;
1041 return extent_read_full_page(tree, page, btree_get_extent, 0);
1042}
1043
1044static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1045{
1046 if (PageWriteback(page) || PageDirty(page))
1047 return 0;
1048
1049 return try_release_extent_buffer(page);
1050}
1051
1052static void btree_invalidatepage(struct page *page, unsigned int offset,
1053 unsigned int length)
1054{
1055 struct extent_io_tree *tree;
1056 tree = &BTRFS_I(page->mapping->host)->io_tree;
1057 extent_invalidatepage(tree, page, offset);
1058 btree_releasepage(page, GFP_NOFS);
1059 if (PagePrivate(page)) {
1060 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1061 "page private not zero on page %llu",
1062 (unsigned long long)page_offset(page));
1063 ClearPagePrivate(page);
1064 set_page_private(page, 0);
1065 put_page(page);
1066 }
1067}
1068
1069static int btree_set_page_dirty(struct page *page)
1070{
1071#ifdef DEBUG
1072 struct extent_buffer *eb;
1073
1074 BUG_ON(!PagePrivate(page));
1075 eb = (struct extent_buffer *)page->private;
1076 BUG_ON(!eb);
1077 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1078 BUG_ON(!atomic_read(&eb->refs));
1079 btrfs_assert_tree_locked(eb);
1080#endif
1081 return __set_page_dirty_nobuffers(page);
1082}
1083
1084static const struct address_space_operations btree_aops = {
1085 .readpage = btree_readpage,
1086 .writepages = btree_writepages,
1087 .releasepage = btree_releasepage,
1088 .invalidatepage = btree_invalidatepage,
1089#ifdef CONFIG_MIGRATION
1090 .migratepage = btree_migratepage,
1091#endif
1092 .set_page_dirty = btree_set_page_dirty,
1093};
1094
1095void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1096{
1097 struct extent_buffer *buf = NULL;
1098 struct inode *btree_inode = root->fs_info->btree_inode;
1099
1100 buf = btrfs_find_create_tree_block(root, bytenr);
1101 if (!buf)
1102 return;
1103 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1104 buf, 0, WAIT_NONE, btree_get_extent, 0);
1105 free_extent_buffer(buf);
1106}
1107
1108int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1109 int mirror_num, struct extent_buffer **eb)
1110{
1111 struct extent_buffer *buf = NULL;
1112 struct inode *btree_inode = root->fs_info->btree_inode;
1113 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1114 int ret;
1115
1116 buf = btrfs_find_create_tree_block(root, bytenr);
1117 if (!buf)
1118 return 0;
1119
1120 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1121
1122 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1123 btree_get_extent, mirror_num);
1124 if (ret) {
1125 free_extent_buffer(buf);
1126 return ret;
1127 }
1128
1129 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1130 free_extent_buffer(buf);
1131 return -EIO;
1132 } else if (extent_buffer_uptodate(buf)) {
1133 *eb = buf;
1134 } else {
1135 free_extent_buffer(buf);
1136 }
1137 return 0;
1138}
1139
1140struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1141 u64 bytenr)
1142{
1143 return find_extent_buffer(fs_info, bytenr);
1144}
1145
1146struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1147 u64 bytenr)
1148{
1149 if (btrfs_test_is_dummy_root(root))
1150 return alloc_test_extent_buffer(root->fs_info, bytenr);
1151 return alloc_extent_buffer(root->fs_info, bytenr);
1152}
1153
1154
1155int btrfs_write_tree_block(struct extent_buffer *buf)
1156{
1157 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1158 buf->start + buf->len - 1);
1159}
1160
1161int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1162{
1163 return filemap_fdatawait_range(buf->pages[0]->mapping,
1164 buf->start, buf->start + buf->len - 1);
1165}
1166
1167struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1168 u64 parent_transid)
1169{
1170 struct extent_buffer *buf = NULL;
1171 int ret;
1172
1173 buf = btrfs_find_create_tree_block(root, bytenr);
1174 if (!buf)
1175 return ERR_PTR(-ENOMEM);
1176
1177 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1178 if (ret) {
1179 free_extent_buffer(buf);
1180 return ERR_PTR(ret);
1181 }
1182 return buf;
1183
1184}
1185
1186void clean_tree_block(struct btrfs_trans_handle *trans,
1187 struct btrfs_fs_info *fs_info,
1188 struct extent_buffer *buf)
1189{
1190 if (btrfs_header_generation(buf) ==
1191 fs_info->running_transaction->transid) {
1192 btrfs_assert_tree_locked(buf);
1193
1194 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1195 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1196 -buf->len,
1197 fs_info->dirty_metadata_batch);
1198 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1199 btrfs_set_lock_blocking(buf);
1200 clear_extent_buffer_dirty(buf);
1201 }
1202 }
1203}
1204
1205static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1206{
1207 struct btrfs_subvolume_writers *writers;
1208 int ret;
1209
1210 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1211 if (!writers)
1212 return ERR_PTR(-ENOMEM);
1213
1214 ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1215 if (ret < 0) {
1216 kfree(writers);
1217 return ERR_PTR(ret);
1218 }
1219
1220 init_waitqueue_head(&writers->wait);
1221 return writers;
1222}
1223
1224static void
1225btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1226{
1227 percpu_counter_destroy(&writers->counter);
1228 kfree(writers);
1229}
1230
1231static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1232 struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1233 u64 objectid)
1234{
1235 root->node = NULL;
1236 root->commit_root = NULL;
1237 root->sectorsize = sectorsize;
1238 root->nodesize = nodesize;
1239 root->stripesize = stripesize;
1240 root->state = 0;
1241 root->orphan_cleanup_state = 0;
1242
1243 root->objectid = objectid;
1244 root->last_trans = 0;
1245 root->highest_objectid = 0;
1246 root->nr_delalloc_inodes = 0;
1247 root->nr_ordered_extents = 0;
1248 root->name = NULL;
1249 root->inode_tree = RB_ROOT;
1250 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1251 root->block_rsv = NULL;
1252 root->orphan_block_rsv = NULL;
1253
1254 INIT_LIST_HEAD(&root->dirty_list);
1255 INIT_LIST_HEAD(&root->root_list);
1256 INIT_LIST_HEAD(&root->delalloc_inodes);
1257 INIT_LIST_HEAD(&root->delalloc_root);
1258 INIT_LIST_HEAD(&root->ordered_extents);
1259 INIT_LIST_HEAD(&root->ordered_root);
1260 INIT_LIST_HEAD(&root->logged_list[0]);
1261 INIT_LIST_HEAD(&root->logged_list[1]);
1262 spin_lock_init(&root->orphan_lock);
1263 spin_lock_init(&root->inode_lock);
1264 spin_lock_init(&root->delalloc_lock);
1265 spin_lock_init(&root->ordered_extent_lock);
1266 spin_lock_init(&root->accounting_lock);
1267 spin_lock_init(&root->log_extents_lock[0]);
1268 spin_lock_init(&root->log_extents_lock[1]);
1269 mutex_init(&root->objectid_mutex);
1270 mutex_init(&root->log_mutex);
1271 mutex_init(&root->ordered_extent_mutex);
1272 mutex_init(&root->delalloc_mutex);
1273 init_waitqueue_head(&root->log_writer_wait);
1274 init_waitqueue_head(&root->log_commit_wait[0]);
1275 init_waitqueue_head(&root->log_commit_wait[1]);
1276 INIT_LIST_HEAD(&root->log_ctxs[0]);
1277 INIT_LIST_HEAD(&root->log_ctxs[1]);
1278 atomic_set(&root->log_commit[0], 0);
1279 atomic_set(&root->log_commit[1], 0);
1280 atomic_set(&root->log_writers, 0);
1281 atomic_set(&root->log_batch, 0);
1282 atomic_set(&root->orphan_inodes, 0);
1283 atomic_set(&root->refs, 1);
1284 atomic_set(&root->will_be_snapshoted, 0);
1285 atomic_set(&root->qgroup_meta_rsv, 0);
1286 root->log_transid = 0;
1287 root->log_transid_committed = -1;
1288 root->last_log_commit = 0;
1289 if (fs_info)
1290 extent_io_tree_init(&root->dirty_log_pages,
1291 fs_info->btree_inode->i_mapping);
1292
1293 memset(&root->root_key, 0, sizeof(root->root_key));
1294 memset(&root->root_item, 0, sizeof(root->root_item));
1295 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1296 if (fs_info)
1297 root->defrag_trans_start = fs_info->generation;
1298 else
1299 root->defrag_trans_start = 0;
1300 root->root_key.objectid = objectid;
1301 root->anon_dev = 0;
1302
1303 spin_lock_init(&root->root_item_lock);
1304}
1305
1306static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1307 gfp_t flags)
1308{
1309 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1310 if (root)
1311 root->fs_info = fs_info;
1312 return root;
1313}
1314
1315#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1316/* Should only be used by the testing infrastructure */
1317struct btrfs_root *btrfs_alloc_dummy_root(void)
1318{
1319 struct btrfs_root *root;
1320
1321 root = btrfs_alloc_root(NULL, GFP_KERNEL);
1322 if (!root)
1323 return ERR_PTR(-ENOMEM);
1324 __setup_root(4096, 4096, 4096, root, NULL, 1);
1325 set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1326 root->alloc_bytenr = 0;
1327
1328 return root;
1329}
1330#endif
1331
1332struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1333 struct btrfs_fs_info *fs_info,
1334 u64 objectid)
1335{
1336 struct extent_buffer *leaf;
1337 struct btrfs_root *tree_root = fs_info->tree_root;
1338 struct btrfs_root *root;
1339 struct btrfs_key key;
1340 int ret = 0;
1341 uuid_le uuid;
1342
1343 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1344 if (!root)
1345 return ERR_PTR(-ENOMEM);
1346
1347 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1348 tree_root->stripesize, root, fs_info, objectid);
1349 root->root_key.objectid = objectid;
1350 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1351 root->root_key.offset = 0;
1352
1353 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1354 if (IS_ERR(leaf)) {
1355 ret = PTR_ERR(leaf);
1356 leaf = NULL;
1357 goto fail;
1358 }
1359
1360 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1361 btrfs_set_header_bytenr(leaf, leaf->start);
1362 btrfs_set_header_generation(leaf, trans->transid);
1363 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1364 btrfs_set_header_owner(leaf, objectid);
1365 root->node = leaf;
1366
1367 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1368 BTRFS_FSID_SIZE);
1369 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1370 btrfs_header_chunk_tree_uuid(leaf),
1371 BTRFS_UUID_SIZE);
1372 btrfs_mark_buffer_dirty(leaf);
1373
1374 root->commit_root = btrfs_root_node(root);
1375 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1376
1377 root->root_item.flags = 0;
1378 root->root_item.byte_limit = 0;
1379 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1380 btrfs_set_root_generation(&root->root_item, trans->transid);
1381 btrfs_set_root_level(&root->root_item, 0);
1382 btrfs_set_root_refs(&root->root_item, 1);
1383 btrfs_set_root_used(&root->root_item, leaf->len);
1384 btrfs_set_root_last_snapshot(&root->root_item, 0);
1385 btrfs_set_root_dirid(&root->root_item, 0);
1386 uuid_le_gen(&uuid);
1387 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1388 root->root_item.drop_level = 0;
1389
1390 key.objectid = objectid;
1391 key.type = BTRFS_ROOT_ITEM_KEY;
1392 key.offset = 0;
1393 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1394 if (ret)
1395 goto fail;
1396
1397 btrfs_tree_unlock(leaf);
1398
1399 return root;
1400
1401fail:
1402 if (leaf) {
1403 btrfs_tree_unlock(leaf);
1404 free_extent_buffer(root->commit_root);
1405 free_extent_buffer(leaf);
1406 }
1407 kfree(root);
1408
1409 return ERR_PTR(ret);
1410}
1411
1412static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1413 struct btrfs_fs_info *fs_info)
1414{
1415 struct btrfs_root *root;
1416 struct btrfs_root *tree_root = fs_info->tree_root;
1417 struct extent_buffer *leaf;
1418
1419 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1420 if (!root)
1421 return ERR_PTR(-ENOMEM);
1422
1423 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1424 tree_root->stripesize, root, fs_info,
1425 BTRFS_TREE_LOG_OBJECTID);
1426
1427 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1428 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1429 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1430
1431 /*
1432 * DON'T set REF_COWS for log trees
1433 *
1434 * log trees do not get reference counted because they go away
1435 * before a real commit is actually done. They do store pointers
1436 * to file data extents, and those reference counts still get
1437 * updated (along with back refs to the log tree).
1438 */
1439
1440 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1441 NULL, 0, 0, 0);
1442 if (IS_ERR(leaf)) {
1443 kfree(root);
1444 return ERR_CAST(leaf);
1445 }
1446
1447 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1448 btrfs_set_header_bytenr(leaf, leaf->start);
1449 btrfs_set_header_generation(leaf, trans->transid);
1450 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1451 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1452 root->node = leaf;
1453
1454 write_extent_buffer(root->node, root->fs_info->fsid,
1455 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1456 btrfs_mark_buffer_dirty(root->node);
1457 btrfs_tree_unlock(root->node);
1458 return root;
1459}
1460
1461int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1462 struct btrfs_fs_info *fs_info)
1463{
1464 struct btrfs_root *log_root;
1465
1466 log_root = alloc_log_tree(trans, fs_info);
1467 if (IS_ERR(log_root))
1468 return PTR_ERR(log_root);
1469 WARN_ON(fs_info->log_root_tree);
1470 fs_info->log_root_tree = log_root;
1471 return 0;
1472}
1473
1474int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1475 struct btrfs_root *root)
1476{
1477 struct btrfs_root *log_root;
1478 struct btrfs_inode_item *inode_item;
1479
1480 log_root = alloc_log_tree(trans, root->fs_info);
1481 if (IS_ERR(log_root))
1482 return PTR_ERR(log_root);
1483
1484 log_root->last_trans = trans->transid;
1485 log_root->root_key.offset = root->root_key.objectid;
1486
1487 inode_item = &log_root->root_item.inode;
1488 btrfs_set_stack_inode_generation(inode_item, 1);
1489 btrfs_set_stack_inode_size(inode_item, 3);
1490 btrfs_set_stack_inode_nlink(inode_item, 1);
1491 btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1492 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1493
1494 btrfs_set_root_node(&log_root->root_item, log_root->node);
1495
1496 WARN_ON(root->log_root);
1497 root->log_root = log_root;
1498 root->log_transid = 0;
1499 root->log_transid_committed = -1;
1500 root->last_log_commit = 0;
1501 return 0;
1502}
1503
1504static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1505 struct btrfs_key *key)
1506{
1507 struct btrfs_root *root;
1508 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1509 struct btrfs_path *path;
1510 u64 generation;
1511 int ret;
1512
1513 path = btrfs_alloc_path();
1514 if (!path)
1515 return ERR_PTR(-ENOMEM);
1516
1517 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1518 if (!root) {
1519 ret = -ENOMEM;
1520 goto alloc_fail;
1521 }
1522
1523 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1524 tree_root->stripesize, root, fs_info, key->objectid);
1525
1526 ret = btrfs_find_root(tree_root, key, path,
1527 &root->root_item, &root->root_key);
1528 if (ret) {
1529 if (ret > 0)
1530 ret = -ENOENT;
1531 goto find_fail;
1532 }
1533
1534 generation = btrfs_root_generation(&root->root_item);
1535 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1536 generation);
1537 if (IS_ERR(root->node)) {
1538 ret = PTR_ERR(root->node);
1539 goto find_fail;
1540 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1541 ret = -EIO;
1542 free_extent_buffer(root->node);
1543 goto find_fail;
1544 }
1545 root->commit_root = btrfs_root_node(root);
1546out:
1547 btrfs_free_path(path);
1548 return root;
1549
1550find_fail:
1551 kfree(root);
1552alloc_fail:
1553 root = ERR_PTR(ret);
1554 goto out;
1555}
1556
1557struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1558 struct btrfs_key *location)
1559{
1560 struct btrfs_root *root;
1561
1562 root = btrfs_read_tree_root(tree_root, location);
1563 if (IS_ERR(root))
1564 return root;
1565
1566 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1567 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1568 btrfs_check_and_init_root_item(&root->root_item);
1569 }
1570
1571 return root;
1572}
1573
1574int btrfs_init_fs_root(struct btrfs_root *root)
1575{
1576 int ret;
1577 struct btrfs_subvolume_writers *writers;
1578
1579 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1580 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1581 GFP_NOFS);
1582 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1583 ret = -ENOMEM;
1584 goto fail;
1585 }
1586
1587 writers = btrfs_alloc_subvolume_writers();
1588 if (IS_ERR(writers)) {
1589 ret = PTR_ERR(writers);
1590 goto fail;
1591 }
1592 root->subv_writers = writers;
1593
1594 btrfs_init_free_ino_ctl(root);
1595 spin_lock_init(&root->ino_cache_lock);
1596 init_waitqueue_head(&root->ino_cache_wait);
1597
1598 ret = get_anon_bdev(&root->anon_dev);
1599 if (ret)
1600 goto free_writers;
1601
1602 mutex_lock(&root->objectid_mutex);
1603 ret = btrfs_find_highest_objectid(root,
1604 &root->highest_objectid);
1605 if (ret) {
1606 mutex_unlock(&root->objectid_mutex);
1607 goto free_root_dev;
1608 }
1609
1610 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1611
1612 mutex_unlock(&root->objectid_mutex);
1613
1614 return 0;
1615
1616free_root_dev:
1617 free_anon_bdev(root->anon_dev);
1618free_writers:
1619 btrfs_free_subvolume_writers(root->subv_writers);
1620fail:
1621 kfree(root->free_ino_ctl);
1622 kfree(root->free_ino_pinned);
1623 return ret;
1624}
1625
1626static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1627 u64 root_id)
1628{
1629 struct btrfs_root *root;
1630
1631 spin_lock(&fs_info->fs_roots_radix_lock);
1632 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1633 (unsigned long)root_id);
1634 spin_unlock(&fs_info->fs_roots_radix_lock);
1635 return root;
1636}
1637
1638int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1639 struct btrfs_root *root)
1640{
1641 int ret;
1642
1643 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1644 if (ret)
1645 return ret;
1646
1647 spin_lock(&fs_info->fs_roots_radix_lock);
1648 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1649 (unsigned long)root->root_key.objectid,
1650 root);
1651 if (ret == 0)
1652 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1653 spin_unlock(&fs_info->fs_roots_radix_lock);
1654 radix_tree_preload_end();
1655
1656 return ret;
1657}
1658
1659struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1660 struct btrfs_key *location,
1661 bool check_ref)
1662{
1663 struct btrfs_root *root;
1664 struct btrfs_path *path;
1665 struct btrfs_key key;
1666 int ret;
1667
1668 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1669 return fs_info->tree_root;
1670 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1671 return fs_info->extent_root;
1672 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1673 return fs_info->chunk_root;
1674 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1675 return fs_info->dev_root;
1676 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1677 return fs_info->csum_root;
1678 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1679 return fs_info->quota_root ? fs_info->quota_root :
1680 ERR_PTR(-ENOENT);
1681 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1682 return fs_info->uuid_root ? fs_info->uuid_root :
1683 ERR_PTR(-ENOENT);
1684 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1685 return fs_info->free_space_root ? fs_info->free_space_root :
1686 ERR_PTR(-ENOENT);
1687again:
1688 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1689 if (root) {
1690 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1691 return ERR_PTR(-ENOENT);
1692 return root;
1693 }
1694
1695 root = btrfs_read_fs_root(fs_info->tree_root, location);
1696 if (IS_ERR(root))
1697 return root;
1698
1699 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1700 ret = -ENOENT;
1701 goto fail;
1702 }
1703
1704 ret = btrfs_init_fs_root(root);
1705 if (ret)
1706 goto fail;
1707
1708 path = btrfs_alloc_path();
1709 if (!path) {
1710 ret = -ENOMEM;
1711 goto fail;
1712 }
1713 key.objectid = BTRFS_ORPHAN_OBJECTID;
1714 key.type = BTRFS_ORPHAN_ITEM_KEY;
1715 key.offset = location->objectid;
1716
1717 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1718 btrfs_free_path(path);
1719 if (ret < 0)
1720 goto fail;
1721 if (ret == 0)
1722 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1723
1724 ret = btrfs_insert_fs_root(fs_info, root);
1725 if (ret) {
1726 if (ret == -EEXIST) {
1727 free_fs_root(root);
1728 goto again;
1729 }
1730 goto fail;
1731 }
1732 return root;
1733fail:
1734 free_fs_root(root);
1735 return ERR_PTR(ret);
1736}
1737
1738static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1739{
1740 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1741 int ret = 0;
1742 struct btrfs_device *device;
1743 struct backing_dev_info *bdi;
1744
1745 rcu_read_lock();
1746 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1747 if (!device->bdev)
1748 continue;
1749 bdi = blk_get_backing_dev_info(device->bdev);
1750 if (bdi_congested(bdi, bdi_bits)) {
1751 ret = 1;
1752 break;
1753 }
1754 }
1755 rcu_read_unlock();
1756 return ret;
1757}
1758
1759static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1760{
1761 int err;
1762
1763 err = bdi_setup_and_register(bdi, "btrfs");
1764 if (err)
1765 return err;
1766
1767 bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
1768 bdi->congested_fn = btrfs_congested_fn;
1769 bdi->congested_data = info;
1770 bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1771 return 0;
1772}
1773
1774/*
1775 * called by the kthread helper functions to finally call the bio end_io
1776 * functions. This is where read checksum verification actually happens
1777 */
1778static void end_workqueue_fn(struct btrfs_work *work)
1779{
1780 struct bio *bio;
1781 struct btrfs_end_io_wq *end_io_wq;
1782
1783 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1784 bio = end_io_wq->bio;
1785
1786 bio->bi_error = end_io_wq->error;
1787 bio->bi_private = end_io_wq->private;
1788 bio->bi_end_io = end_io_wq->end_io;
1789 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1790 bio_endio(bio);
1791}
1792
1793static int cleaner_kthread(void *arg)
1794{
1795 struct btrfs_root *root = arg;
1796 int again;
1797 struct btrfs_trans_handle *trans;
1798
1799 do {
1800 again = 0;
1801
1802 /* Make the cleaner go to sleep early. */
1803 if (btrfs_need_cleaner_sleep(root))
1804 goto sleep;
1805
1806 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1807 goto sleep;
1808
1809 /*
1810 * Avoid the problem that we change the status of the fs
1811 * during the above check and trylock.
1812 */
1813 if (btrfs_need_cleaner_sleep(root)) {
1814 mutex_unlock(&root->fs_info->cleaner_mutex);
1815 goto sleep;
1816 }
1817
1818 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
1819 btrfs_run_delayed_iputs(root);
1820 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
1821
1822 again = btrfs_clean_one_deleted_snapshot(root);
1823 mutex_unlock(&root->fs_info->cleaner_mutex);
1824
1825 /*
1826 * The defragger has dealt with the R/O remount and umount,
1827 * needn't do anything special here.
1828 */
1829 btrfs_run_defrag_inodes(root->fs_info);
1830
1831 /*
1832 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1833 * with relocation (btrfs_relocate_chunk) and relocation
1834 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1835 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1836 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1837 * unused block groups.
1838 */
1839 btrfs_delete_unused_bgs(root->fs_info);
1840sleep:
1841 if (!again) {
1842 set_current_state(TASK_INTERRUPTIBLE);
1843 if (!kthread_should_stop())
1844 schedule();
1845 __set_current_state(TASK_RUNNING);
1846 }
1847 } while (!kthread_should_stop());
1848
1849 /*
1850 * Transaction kthread is stopped before us and wakes us up.
1851 * However we might have started a new transaction and COWed some
1852 * tree blocks when deleting unused block groups for example. So
1853 * make sure we commit the transaction we started to have a clean
1854 * shutdown when evicting the btree inode - if it has dirty pages
1855 * when we do the final iput() on it, eviction will trigger a
1856 * writeback for it which will fail with null pointer dereferences
1857 * since work queues and other resources were already released and
1858 * destroyed by the time the iput/eviction/writeback is made.
1859 */
1860 trans = btrfs_attach_transaction(root);
1861 if (IS_ERR(trans)) {
1862 if (PTR_ERR(trans) != -ENOENT)
1863 btrfs_err(root->fs_info,
1864 "cleaner transaction attach returned %ld",
1865 PTR_ERR(trans));
1866 } else {
1867 int ret;
1868
1869 ret = btrfs_commit_transaction(trans, root);
1870 if (ret)
1871 btrfs_err(root->fs_info,
1872 "cleaner open transaction commit returned %d",
1873 ret);
1874 }
1875
1876 return 0;
1877}
1878
1879static int transaction_kthread(void *arg)
1880{
1881 struct btrfs_root *root = arg;
1882 struct btrfs_trans_handle *trans;
1883 struct btrfs_transaction *cur;
1884 u64 transid;
1885 unsigned long now;
1886 unsigned long delay;
1887 bool cannot_commit;
1888
1889 do {
1890 cannot_commit = false;
1891 delay = HZ * root->fs_info->commit_interval;
1892 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1893
1894 spin_lock(&root->fs_info->trans_lock);
1895 cur = root->fs_info->running_transaction;
1896 if (!cur) {
1897 spin_unlock(&root->fs_info->trans_lock);
1898 goto sleep;
1899 }
1900
1901 now = get_seconds();
1902 if (cur->state < TRANS_STATE_BLOCKED &&
1903 (now < cur->start_time ||
1904 now - cur->start_time < root->fs_info->commit_interval)) {
1905 spin_unlock(&root->fs_info->trans_lock);
1906 delay = HZ * 5;
1907 goto sleep;
1908 }
1909 transid = cur->transid;
1910 spin_unlock(&root->fs_info->trans_lock);
1911
1912 /* If the file system is aborted, this will always fail. */
1913 trans = btrfs_attach_transaction(root);
1914 if (IS_ERR(trans)) {
1915 if (PTR_ERR(trans) != -ENOENT)
1916 cannot_commit = true;
1917 goto sleep;
1918 }
1919 if (transid == trans->transid) {
1920 btrfs_commit_transaction(trans, root);
1921 } else {
1922 btrfs_end_transaction(trans, root);
1923 }
1924sleep:
1925 wake_up_process(root->fs_info->cleaner_kthread);
1926 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1927
1928 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1929 &root->fs_info->fs_state)))
1930 btrfs_cleanup_transaction(root);
1931 set_current_state(TASK_INTERRUPTIBLE);
1932 if (!kthread_should_stop() &&
1933 (!btrfs_transaction_blocked(root->fs_info) ||
1934 cannot_commit))
1935 schedule_timeout(delay);
1936 __set_current_state(TASK_RUNNING);
1937 } while (!kthread_should_stop());
1938 return 0;
1939}
1940
1941/*
1942 * this will find the highest generation in the array of
1943 * root backups. The index of the highest array is returned,
1944 * or -1 if we can't find anything.
1945 *
1946 * We check to make sure the array is valid by comparing the
1947 * generation of the latest root in the array with the generation
1948 * in the super block. If they don't match we pitch it.
1949 */
1950static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1951{
1952 u64 cur;
1953 int newest_index = -1;
1954 struct btrfs_root_backup *root_backup;
1955 int i;
1956
1957 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1958 root_backup = info->super_copy->super_roots + i;
1959 cur = btrfs_backup_tree_root_gen(root_backup);
1960 if (cur == newest_gen)
1961 newest_index = i;
1962 }
1963
1964 /* check to see if we actually wrapped around */
1965 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1966 root_backup = info->super_copy->super_roots;
1967 cur = btrfs_backup_tree_root_gen(root_backup);
1968 if (cur == newest_gen)
1969 newest_index = 0;
1970 }
1971 return newest_index;
1972}
1973
1974
1975/*
1976 * find the oldest backup so we know where to store new entries
1977 * in the backup array. This will set the backup_root_index
1978 * field in the fs_info struct
1979 */
1980static void find_oldest_super_backup(struct btrfs_fs_info *info,
1981 u64 newest_gen)
1982{
1983 int newest_index = -1;
1984
1985 newest_index = find_newest_super_backup(info, newest_gen);
1986 /* if there was garbage in there, just move along */
1987 if (newest_index == -1) {
1988 info->backup_root_index = 0;
1989 } else {
1990 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1991 }
1992}
1993
1994/*
1995 * copy all the root pointers into the super backup array.
1996 * this will bump the backup pointer by one when it is
1997 * done
1998 */
1999static void backup_super_roots(struct btrfs_fs_info *info)
2000{
2001 int next_backup;
2002 struct btrfs_root_backup *root_backup;
2003 int last_backup;
2004
2005 next_backup = info->backup_root_index;
2006 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2007 BTRFS_NUM_BACKUP_ROOTS;
2008
2009 /*
2010 * just overwrite the last backup if we're at the same generation
2011 * this happens only at umount
2012 */
2013 root_backup = info->super_for_commit->super_roots + last_backup;
2014 if (btrfs_backup_tree_root_gen(root_backup) ==
2015 btrfs_header_generation(info->tree_root->node))
2016 next_backup = last_backup;
2017
2018 root_backup = info->super_for_commit->super_roots + next_backup;
2019
2020 /*
2021 * make sure all of our padding and empty slots get zero filled
2022 * regardless of which ones we use today
2023 */
2024 memset(root_backup, 0, sizeof(*root_backup));
2025
2026 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2027
2028 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2029 btrfs_set_backup_tree_root_gen(root_backup,
2030 btrfs_header_generation(info->tree_root->node));
2031
2032 btrfs_set_backup_tree_root_level(root_backup,
2033 btrfs_header_level(info->tree_root->node));
2034
2035 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2036 btrfs_set_backup_chunk_root_gen(root_backup,
2037 btrfs_header_generation(info->chunk_root->node));
2038 btrfs_set_backup_chunk_root_level(root_backup,
2039 btrfs_header_level(info->chunk_root->node));
2040
2041 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2042 btrfs_set_backup_extent_root_gen(root_backup,
2043 btrfs_header_generation(info->extent_root->node));
2044 btrfs_set_backup_extent_root_level(root_backup,
2045 btrfs_header_level(info->extent_root->node));
2046
2047 /*
2048 * we might commit during log recovery, which happens before we set
2049 * the fs_root. Make sure it is valid before we fill it in.
2050 */
2051 if (info->fs_root && info->fs_root->node) {
2052 btrfs_set_backup_fs_root(root_backup,
2053 info->fs_root->node->start);
2054 btrfs_set_backup_fs_root_gen(root_backup,
2055 btrfs_header_generation(info->fs_root->node));
2056 btrfs_set_backup_fs_root_level(root_backup,
2057 btrfs_header_level(info->fs_root->node));
2058 }
2059
2060 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2061 btrfs_set_backup_dev_root_gen(root_backup,
2062 btrfs_header_generation(info->dev_root->node));
2063 btrfs_set_backup_dev_root_level(root_backup,
2064 btrfs_header_level(info->dev_root->node));
2065
2066 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2067 btrfs_set_backup_csum_root_gen(root_backup,
2068 btrfs_header_generation(info->csum_root->node));
2069 btrfs_set_backup_csum_root_level(root_backup,
2070 btrfs_header_level(info->csum_root->node));
2071
2072 btrfs_set_backup_total_bytes(root_backup,
2073 btrfs_super_total_bytes(info->super_copy));
2074 btrfs_set_backup_bytes_used(root_backup,
2075 btrfs_super_bytes_used(info->super_copy));
2076 btrfs_set_backup_num_devices(root_backup,
2077 btrfs_super_num_devices(info->super_copy));
2078
2079 /*
2080 * if we don't copy this out to the super_copy, it won't get remembered
2081 * for the next commit
2082 */
2083 memcpy(&info->super_copy->super_roots,
2084 &info->super_for_commit->super_roots,
2085 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2086}
2087
2088/*
2089 * this copies info out of the root backup array and back into
2090 * the in-memory super block. It is meant to help iterate through
2091 * the array, so you send it the number of backups you've already
2092 * tried and the last backup index you used.
2093 *
2094 * this returns -1 when it has tried all the backups
2095 */
2096static noinline int next_root_backup(struct btrfs_fs_info *info,
2097 struct btrfs_super_block *super,
2098 int *num_backups_tried, int *backup_index)
2099{
2100 struct btrfs_root_backup *root_backup;
2101 int newest = *backup_index;
2102
2103 if (*num_backups_tried == 0) {
2104 u64 gen = btrfs_super_generation(super);
2105
2106 newest = find_newest_super_backup(info, gen);
2107 if (newest == -1)
2108 return -1;
2109
2110 *backup_index = newest;
2111 *num_backups_tried = 1;
2112 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2113 /* we've tried all the backups, all done */
2114 return -1;
2115 } else {
2116 /* jump to the next oldest backup */
2117 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2118 BTRFS_NUM_BACKUP_ROOTS;
2119 *backup_index = newest;
2120 *num_backups_tried += 1;
2121 }
2122 root_backup = super->super_roots + newest;
2123
2124 btrfs_set_super_generation(super,
2125 btrfs_backup_tree_root_gen(root_backup));
2126 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2127 btrfs_set_super_root_level(super,
2128 btrfs_backup_tree_root_level(root_backup));
2129 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2130
2131 /*
2132 * fixme: the total bytes and num_devices need to match or we should
2133 * need a fsck
2134 */
2135 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2136 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2137 return 0;
2138}
2139
2140/* helper to cleanup workers */
2141static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2142{
2143 btrfs_destroy_workqueue(fs_info->fixup_workers);
2144 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2145 btrfs_destroy_workqueue(fs_info->workers);
2146 btrfs_destroy_workqueue(fs_info->endio_workers);
2147 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2148 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2149 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2150 btrfs_destroy_workqueue(fs_info->rmw_workers);
2151 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2152 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2153 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2154 btrfs_destroy_workqueue(fs_info->submit_workers);
2155 btrfs_destroy_workqueue(fs_info->delayed_workers);
2156 btrfs_destroy_workqueue(fs_info->caching_workers);
2157 btrfs_destroy_workqueue(fs_info->readahead_workers);
2158 btrfs_destroy_workqueue(fs_info->flush_workers);
2159 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2160 btrfs_destroy_workqueue(fs_info->extent_workers);
2161}
2162
2163static void free_root_extent_buffers(struct btrfs_root *root)
2164{
2165 if (root) {
2166 free_extent_buffer(root->node);
2167 free_extent_buffer(root->commit_root);
2168 root->node = NULL;
2169 root->commit_root = NULL;
2170 }
2171}
2172
2173/* helper to cleanup tree roots */
2174static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2175{
2176 free_root_extent_buffers(info->tree_root);
2177
2178 free_root_extent_buffers(info->dev_root);
2179 free_root_extent_buffers(info->extent_root);
2180 free_root_extent_buffers(info->csum_root);
2181 free_root_extent_buffers(info->quota_root);
2182 free_root_extent_buffers(info->uuid_root);
2183 if (chunk_root)
2184 free_root_extent_buffers(info->chunk_root);
2185 free_root_extent_buffers(info->free_space_root);
2186}
2187
2188void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2189{
2190 int ret;
2191 struct btrfs_root *gang[8];
2192 int i;
2193
2194 while (!list_empty(&fs_info->dead_roots)) {
2195 gang[0] = list_entry(fs_info->dead_roots.next,
2196 struct btrfs_root, root_list);
2197 list_del(&gang[0]->root_list);
2198
2199 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2200 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2201 } else {
2202 free_extent_buffer(gang[0]->node);
2203 free_extent_buffer(gang[0]->commit_root);
2204 btrfs_put_fs_root(gang[0]);
2205 }
2206 }
2207
2208 while (1) {
2209 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2210 (void **)gang, 0,
2211 ARRAY_SIZE(gang));
2212 if (!ret)
2213 break;
2214 for (i = 0; i < ret; i++)
2215 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2216 }
2217
2218 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2219 btrfs_free_log_root_tree(NULL, fs_info);
2220 btrfs_destroy_pinned_extent(fs_info->tree_root,
2221 fs_info->pinned_extents);
2222 }
2223}
2224
2225static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2226{
2227 mutex_init(&fs_info->scrub_lock);
2228 atomic_set(&fs_info->scrubs_running, 0);
2229 atomic_set(&fs_info->scrub_pause_req, 0);
2230 atomic_set(&fs_info->scrubs_paused, 0);
2231 atomic_set(&fs_info->scrub_cancel_req, 0);
2232 init_waitqueue_head(&fs_info->scrub_pause_wait);
2233 fs_info->scrub_workers_refcnt = 0;
2234}
2235
2236static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2237{
2238 spin_lock_init(&fs_info->balance_lock);
2239 mutex_init(&fs_info->balance_mutex);
2240 atomic_set(&fs_info->balance_running, 0);
2241 atomic_set(&fs_info->balance_pause_req, 0);
2242 atomic_set(&fs_info->balance_cancel_req, 0);
2243 fs_info->balance_ctl = NULL;
2244 init_waitqueue_head(&fs_info->balance_wait_q);
2245}
2246
2247static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2248 struct btrfs_root *tree_root)
2249{
2250 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2251 set_nlink(fs_info->btree_inode, 1);
2252 /*
2253 * we set the i_size on the btree inode to the max possible int.
2254 * the real end of the address space is determined by all of
2255 * the devices in the system
2256 */
2257 fs_info->btree_inode->i_size = OFFSET_MAX;
2258 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2259
2260 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2261 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2262 fs_info->btree_inode->i_mapping);
2263 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2264 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2265
2266 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2267
2268 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2269 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2270 sizeof(struct btrfs_key));
2271 set_bit(BTRFS_INODE_DUMMY,
2272 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2273 btrfs_insert_inode_hash(fs_info->btree_inode);
2274}
2275
2276static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2277{
2278 fs_info->dev_replace.lock_owner = 0;
2279 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2280 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2281 rwlock_init(&fs_info->dev_replace.lock);
2282 atomic_set(&fs_info->dev_replace.read_locks, 0);
2283 atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2284 init_waitqueue_head(&fs_info->replace_wait);
2285 init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2286}
2287
2288static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2289{
2290 spin_lock_init(&fs_info->qgroup_lock);
2291 mutex_init(&fs_info->qgroup_ioctl_lock);
2292 fs_info->qgroup_tree = RB_ROOT;
2293 fs_info->qgroup_op_tree = RB_ROOT;
2294 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2295 fs_info->qgroup_seq = 1;
2296 fs_info->quota_enabled = 0;
2297 fs_info->pending_quota_state = 0;
2298 fs_info->qgroup_ulist = NULL;
2299 mutex_init(&fs_info->qgroup_rescan_lock);
2300}
2301
2302static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2303 struct btrfs_fs_devices *fs_devices)
2304{
2305 int max_active = fs_info->thread_pool_size;
2306 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2307
2308 fs_info->workers =
2309 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2310 max_active, 16);
2311
2312 fs_info->delalloc_workers =
2313 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2314
2315 fs_info->flush_workers =
2316 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2317
2318 fs_info->caching_workers =
2319 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2320
2321 /*
2322 * a higher idle thresh on the submit workers makes it much more
2323 * likely that bios will be send down in a sane order to the
2324 * devices
2325 */
2326 fs_info->submit_workers =
2327 btrfs_alloc_workqueue("submit", flags,
2328 min_t(u64, fs_devices->num_devices,
2329 max_active), 64);
2330
2331 fs_info->fixup_workers =
2332 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2333
2334 /*
2335 * endios are largely parallel and should have a very
2336 * low idle thresh
2337 */
2338 fs_info->endio_workers =
2339 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2340 fs_info->endio_meta_workers =
2341 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2342 fs_info->endio_meta_write_workers =
2343 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2344 fs_info->endio_raid56_workers =
2345 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2346 fs_info->endio_repair_workers =
2347 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2348 fs_info->rmw_workers =
2349 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2350 fs_info->endio_write_workers =
2351 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2352 fs_info->endio_freespace_worker =
2353 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2354 fs_info->delayed_workers =
2355 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2356 fs_info->readahead_workers =
2357 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2358 fs_info->qgroup_rescan_workers =
2359 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2360 fs_info->extent_workers =
2361 btrfs_alloc_workqueue("extent-refs", flags,
2362 min_t(u64, fs_devices->num_devices,
2363 max_active), 8);
2364
2365 if (!(fs_info->workers && fs_info->delalloc_workers &&
2366 fs_info->submit_workers && fs_info->flush_workers &&
2367 fs_info->endio_workers && fs_info->endio_meta_workers &&
2368 fs_info->endio_meta_write_workers &&
2369 fs_info->endio_repair_workers &&
2370 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2371 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2372 fs_info->caching_workers && fs_info->readahead_workers &&
2373 fs_info->fixup_workers && fs_info->delayed_workers &&
2374 fs_info->extent_workers &&
2375 fs_info->qgroup_rescan_workers)) {
2376 return -ENOMEM;
2377 }
2378
2379 return 0;
2380}
2381
2382static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2383 struct btrfs_fs_devices *fs_devices)
2384{
2385 int ret;
2386 struct btrfs_root *tree_root = fs_info->tree_root;
2387 struct btrfs_root *log_tree_root;
2388 struct btrfs_super_block *disk_super = fs_info->super_copy;
2389 u64 bytenr = btrfs_super_log_root(disk_super);
2390
2391 if (fs_devices->rw_devices == 0) {
2392 btrfs_warn(fs_info, "log replay required on RO media");
2393 return -EIO;
2394 }
2395
2396 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2397 if (!log_tree_root)
2398 return -ENOMEM;
2399
2400 __setup_root(tree_root->nodesize, tree_root->sectorsize,
2401 tree_root->stripesize, log_tree_root, fs_info,
2402 BTRFS_TREE_LOG_OBJECTID);
2403
2404 log_tree_root->node = read_tree_block(tree_root, bytenr,
2405 fs_info->generation + 1);
2406 if (IS_ERR(log_tree_root->node)) {
2407 btrfs_warn(fs_info, "failed to read log tree");
2408 ret = PTR_ERR(log_tree_root->node);
2409 kfree(log_tree_root);
2410 return ret;
2411 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2412 btrfs_err(fs_info, "failed to read log tree");
2413 free_extent_buffer(log_tree_root->node);
2414 kfree(log_tree_root);
2415 return -EIO;
2416 }
2417 /* returns with log_tree_root freed on success */
2418 ret = btrfs_recover_log_trees(log_tree_root);
2419 if (ret) {
2420 btrfs_std_error(tree_root->fs_info, ret,
2421 "Failed to recover log tree");
2422 free_extent_buffer(log_tree_root->node);
2423 kfree(log_tree_root);
2424 return ret;
2425 }
2426
2427 if (fs_info->sb->s_flags & MS_RDONLY) {
2428 ret = btrfs_commit_super(tree_root);
2429 if (ret)
2430 return ret;
2431 }
2432
2433 return 0;
2434}
2435
2436static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2437 struct btrfs_root *tree_root)
2438{
2439 struct btrfs_root *root;
2440 struct btrfs_key location;
2441 int ret;
2442
2443 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2444 location.type = BTRFS_ROOT_ITEM_KEY;
2445 location.offset = 0;
2446
2447 root = btrfs_read_tree_root(tree_root, &location);
2448 if (IS_ERR(root))
2449 return PTR_ERR(root);
2450 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2451 fs_info->extent_root = root;
2452
2453 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2454 root = btrfs_read_tree_root(tree_root, &location);
2455 if (IS_ERR(root))
2456 return PTR_ERR(root);
2457 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2458 fs_info->dev_root = root;
2459 btrfs_init_devices_late(fs_info);
2460
2461 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2462 root = btrfs_read_tree_root(tree_root, &location);
2463 if (IS_ERR(root))
2464 return PTR_ERR(root);
2465 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2466 fs_info->csum_root = root;
2467
2468 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2469 root = btrfs_read_tree_root(tree_root, &location);
2470 if (!IS_ERR(root)) {
2471 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2472 fs_info->quota_enabled = 1;
2473 fs_info->pending_quota_state = 1;
2474 fs_info->quota_root = root;
2475 }
2476
2477 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2478 root = btrfs_read_tree_root(tree_root, &location);
2479 if (IS_ERR(root)) {
2480 ret = PTR_ERR(root);
2481 if (ret != -ENOENT)
2482 return ret;
2483 } else {
2484 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2485 fs_info->uuid_root = root;
2486 }
2487
2488 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2489 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2490 root = btrfs_read_tree_root(tree_root, &location);
2491 if (IS_ERR(root))
2492 return PTR_ERR(root);
2493 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2494 fs_info->free_space_root = root;
2495 }
2496
2497 return 0;
2498}
2499
2500int open_ctree(struct super_block *sb,
2501 struct btrfs_fs_devices *fs_devices,
2502 char *options)
2503{
2504 u32 sectorsize;
2505 u32 nodesize;
2506 u32 stripesize;
2507 u64 generation;
2508 u64 features;
2509 struct btrfs_key location;
2510 struct buffer_head *bh;
2511 struct btrfs_super_block *disk_super;
2512 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2513 struct btrfs_root *tree_root;
2514 struct btrfs_root *chunk_root;
2515 int ret;
2516 int err = -EINVAL;
2517 int num_backups_tried = 0;
2518 int backup_index = 0;
2519 int max_active;
2520
2521 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2522 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2523 if (!tree_root || !chunk_root) {
2524 err = -ENOMEM;
2525 goto fail;
2526 }
2527
2528 ret = init_srcu_struct(&fs_info->subvol_srcu);
2529 if (ret) {
2530 err = ret;
2531 goto fail;
2532 }
2533
2534 ret = setup_bdi(fs_info, &fs_info->bdi);
2535 if (ret) {
2536 err = ret;
2537 goto fail_srcu;
2538 }
2539
2540 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2541 if (ret) {
2542 err = ret;
2543 goto fail_bdi;
2544 }
2545 fs_info->dirty_metadata_batch = PAGE_SIZE *
2546 (1 + ilog2(nr_cpu_ids));
2547
2548 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2549 if (ret) {
2550 err = ret;
2551 goto fail_dirty_metadata_bytes;
2552 }
2553
2554 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2555 if (ret) {
2556 err = ret;
2557 goto fail_delalloc_bytes;
2558 }
2559
2560 fs_info->btree_inode = new_inode(sb);
2561 if (!fs_info->btree_inode) {
2562 err = -ENOMEM;
2563 goto fail_bio_counter;
2564 }
2565
2566 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2567
2568 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2569 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2570 INIT_LIST_HEAD(&fs_info->trans_list);
2571 INIT_LIST_HEAD(&fs_info->dead_roots);
2572 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2573 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2574 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2575 spin_lock_init(&fs_info->delalloc_root_lock);
2576 spin_lock_init(&fs_info->trans_lock);
2577 spin_lock_init(&fs_info->fs_roots_radix_lock);
2578 spin_lock_init(&fs_info->delayed_iput_lock);
2579 spin_lock_init(&fs_info->defrag_inodes_lock);
2580 spin_lock_init(&fs_info->free_chunk_lock);
2581 spin_lock_init(&fs_info->tree_mod_seq_lock);
2582 spin_lock_init(&fs_info->super_lock);
2583 spin_lock_init(&fs_info->qgroup_op_lock);
2584 spin_lock_init(&fs_info->buffer_lock);
2585 spin_lock_init(&fs_info->unused_bgs_lock);
2586 rwlock_init(&fs_info->tree_mod_log_lock);
2587 mutex_init(&fs_info->unused_bg_unpin_mutex);
2588 mutex_init(&fs_info->delete_unused_bgs_mutex);
2589 mutex_init(&fs_info->reloc_mutex);
2590 mutex_init(&fs_info->delalloc_root_mutex);
2591 mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2592 seqlock_init(&fs_info->profiles_lock);
2593
2594 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2595 INIT_LIST_HEAD(&fs_info->space_info);
2596 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2597 INIT_LIST_HEAD(&fs_info->unused_bgs);
2598 btrfs_mapping_init(&fs_info->mapping_tree);
2599 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2600 BTRFS_BLOCK_RSV_GLOBAL);
2601 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2602 BTRFS_BLOCK_RSV_DELALLOC);
2603 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2604 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2605 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2606 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2607 BTRFS_BLOCK_RSV_DELOPS);
2608 atomic_set(&fs_info->nr_async_submits, 0);
2609 atomic_set(&fs_info->async_delalloc_pages, 0);
2610 atomic_set(&fs_info->async_submit_draining, 0);
2611 atomic_set(&fs_info->nr_async_bios, 0);
2612 atomic_set(&fs_info->defrag_running, 0);
2613 atomic_set(&fs_info->qgroup_op_seq, 0);
2614 atomic_set(&fs_info->reada_works_cnt, 0);
2615 atomic64_set(&fs_info->tree_mod_seq, 0);
2616 fs_info->sb = sb;
2617 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2618 fs_info->metadata_ratio = 0;
2619 fs_info->defrag_inodes = RB_ROOT;
2620 fs_info->free_chunk_space = 0;
2621 fs_info->tree_mod_log = RB_ROOT;
2622 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2623 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2624 /* readahead state */
2625 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2626 spin_lock_init(&fs_info->reada_lock);
2627
2628 fs_info->thread_pool_size = min_t(unsigned long,
2629 num_online_cpus() + 2, 8);
2630
2631 INIT_LIST_HEAD(&fs_info->ordered_roots);
2632 spin_lock_init(&fs_info->ordered_root_lock);
2633 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2634 GFP_KERNEL);
2635 if (!fs_info->delayed_root) {
2636 err = -ENOMEM;
2637 goto fail_iput;
2638 }
2639 btrfs_init_delayed_root(fs_info->delayed_root);
2640
2641 btrfs_init_scrub(fs_info);
2642#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2643 fs_info->check_integrity_print_mask = 0;
2644#endif
2645 btrfs_init_balance(fs_info);
2646 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2647
2648 sb->s_blocksize = 4096;
2649 sb->s_blocksize_bits = blksize_bits(4096);
2650 sb->s_bdi = &fs_info->bdi;
2651
2652 btrfs_init_btree_inode(fs_info, tree_root);
2653
2654 spin_lock_init(&fs_info->block_group_cache_lock);
2655 fs_info->block_group_cache_tree = RB_ROOT;
2656 fs_info->first_logical_byte = (u64)-1;
2657
2658 extent_io_tree_init(&fs_info->freed_extents[0],
2659 fs_info->btree_inode->i_mapping);
2660 extent_io_tree_init(&fs_info->freed_extents[1],
2661 fs_info->btree_inode->i_mapping);
2662 fs_info->pinned_extents = &fs_info->freed_extents[0];
2663 fs_info->do_barriers = 1;
2664
2665
2666 mutex_init(&fs_info->ordered_operations_mutex);
2667 mutex_init(&fs_info->tree_log_mutex);
2668 mutex_init(&fs_info->chunk_mutex);
2669 mutex_init(&fs_info->transaction_kthread_mutex);
2670 mutex_init(&fs_info->cleaner_mutex);
2671 mutex_init(&fs_info->volume_mutex);
2672 mutex_init(&fs_info->ro_block_group_mutex);
2673 init_rwsem(&fs_info->commit_root_sem);
2674 init_rwsem(&fs_info->cleanup_work_sem);
2675 init_rwsem(&fs_info->subvol_sem);
2676 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2677
2678 btrfs_init_dev_replace_locks(fs_info);
2679 btrfs_init_qgroup(fs_info);
2680
2681 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2682 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2683
2684 init_waitqueue_head(&fs_info->transaction_throttle);
2685 init_waitqueue_head(&fs_info->transaction_wait);
2686 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2687 init_waitqueue_head(&fs_info->async_submit_wait);
2688
2689 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2690
2691 ret = btrfs_alloc_stripe_hash_table(fs_info);
2692 if (ret) {
2693 err = ret;
2694 goto fail_alloc;
2695 }
2696
2697 __setup_root(4096, 4096, 4096, tree_root,
2698 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2699
2700 invalidate_bdev(fs_devices->latest_bdev);
2701
2702 /*
2703 * Read super block and check the signature bytes only
2704 */
2705 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2706 if (IS_ERR(bh)) {
2707 err = PTR_ERR(bh);
2708 goto fail_alloc;
2709 }
2710
2711 /*
2712 * We want to check superblock checksum, the type is stored inside.
2713 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2714 */
2715 if (btrfs_check_super_csum(bh->b_data)) {
2716 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2717 err = -EINVAL;
2718 brelse(bh);
2719 goto fail_alloc;
2720 }
2721
2722 /*
2723 * super_copy is zeroed at allocation time and we never touch the
2724 * following bytes up to INFO_SIZE, the checksum is calculated from
2725 * the whole block of INFO_SIZE
2726 */
2727 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2728 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2729 sizeof(*fs_info->super_for_commit));
2730 brelse(bh);
2731
2732 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2733
2734 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2735 if (ret) {
2736 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2737 err = -EINVAL;
2738 goto fail_alloc;
2739 }
2740
2741 disk_super = fs_info->super_copy;
2742 if (!btrfs_super_root(disk_super))
2743 goto fail_alloc;
2744
2745 /* check FS state, whether FS is broken. */
2746 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2747 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2748
2749 /*
2750 * run through our array of backup supers and setup
2751 * our ring pointer to the oldest one
2752 */
2753 generation = btrfs_super_generation(disk_super);
2754 find_oldest_super_backup(fs_info, generation);
2755
2756 /*
2757 * In the long term, we'll store the compression type in the super
2758 * block, and it'll be used for per file compression control.
2759 */
2760 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2761
2762 ret = btrfs_parse_options(tree_root, options, sb->s_flags);
2763 if (ret) {
2764 err = ret;
2765 goto fail_alloc;
2766 }
2767
2768 features = btrfs_super_incompat_flags(disk_super) &
2769 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2770 if (features) {
2771 printk(KERN_ERR "BTRFS: couldn't mount because of "
2772 "unsupported optional features (%Lx).\n",
2773 features);
2774 err = -EINVAL;
2775 goto fail_alloc;
2776 }
2777
2778 features = btrfs_super_incompat_flags(disk_super);
2779 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2780 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2781 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2782
2783 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2784 printk(KERN_INFO "BTRFS: has skinny extents\n");
2785
2786 /*
2787 * flag our filesystem as having big metadata blocks if
2788 * they are bigger than the page size
2789 */
2790 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2791 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2792 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2793 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2794 }
2795
2796 nodesize = btrfs_super_nodesize(disk_super);
2797 sectorsize = btrfs_super_sectorsize(disk_super);
2798 stripesize = btrfs_super_stripesize(disk_super);
2799 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2800 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2801
2802 /*
2803 * mixed block groups end up with duplicate but slightly offset
2804 * extent buffers for the same range. It leads to corruptions
2805 */
2806 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2807 (sectorsize != nodesize)) {
2808 printk(KERN_ERR "BTRFS: unequal leaf/node/sector sizes "
2809 "are not allowed for mixed block groups on %s\n",
2810 sb->s_id);
2811 goto fail_alloc;
2812 }
2813
2814 /*
2815 * Needn't use the lock because there is no other task which will
2816 * update the flag.
2817 */
2818 btrfs_set_super_incompat_flags(disk_super, features);
2819
2820 features = btrfs_super_compat_ro_flags(disk_super) &
2821 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2822 if (!(sb->s_flags & MS_RDONLY) && features) {
2823 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2824 "unsupported option features (%Lx).\n",
2825 features);
2826 err = -EINVAL;
2827 goto fail_alloc;
2828 }
2829
2830 max_active = fs_info->thread_pool_size;
2831
2832 ret = btrfs_init_workqueues(fs_info, fs_devices);
2833 if (ret) {
2834 err = ret;
2835 goto fail_sb_buffer;
2836 }
2837
2838 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2839 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2840 SZ_4M / PAGE_SIZE);
2841
2842 tree_root->nodesize = nodesize;
2843 tree_root->sectorsize = sectorsize;
2844 tree_root->stripesize = stripesize;
2845
2846 sb->s_blocksize = sectorsize;
2847 sb->s_blocksize_bits = blksize_bits(sectorsize);
2848
2849 mutex_lock(&fs_info->chunk_mutex);
2850 ret = btrfs_read_sys_array(tree_root);
2851 mutex_unlock(&fs_info->chunk_mutex);
2852 if (ret) {
2853 printk(KERN_ERR "BTRFS: failed to read the system "
2854 "array on %s\n", sb->s_id);
2855 goto fail_sb_buffer;
2856 }
2857
2858 generation = btrfs_super_chunk_root_generation(disk_super);
2859
2860 __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2861 fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2862
2863 chunk_root->node = read_tree_block(chunk_root,
2864 btrfs_super_chunk_root(disk_super),
2865 generation);
2866 if (IS_ERR(chunk_root->node) ||
2867 !extent_buffer_uptodate(chunk_root->node)) {
2868 printk(KERN_ERR "BTRFS: failed to read chunk root on %s\n",
2869 sb->s_id);
2870 if (!IS_ERR(chunk_root->node))
2871 free_extent_buffer(chunk_root->node);
2872 chunk_root->node = NULL;
2873 goto fail_tree_roots;
2874 }
2875 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2876 chunk_root->commit_root = btrfs_root_node(chunk_root);
2877
2878 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2879 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2880
2881 ret = btrfs_read_chunk_tree(chunk_root);
2882 if (ret) {
2883 printk(KERN_ERR "BTRFS: failed to read chunk tree on %s\n",
2884 sb->s_id);
2885 goto fail_tree_roots;
2886 }
2887
2888 /*
2889 * keep the device that is marked to be the target device for the
2890 * dev_replace procedure
2891 */
2892 btrfs_close_extra_devices(fs_devices, 0);
2893
2894 if (!fs_devices->latest_bdev) {
2895 printk(KERN_ERR "BTRFS: failed to read devices on %s\n",
2896 sb->s_id);
2897 goto fail_tree_roots;
2898 }
2899
2900retry_root_backup:
2901 generation = btrfs_super_generation(disk_super);
2902
2903 tree_root->node = read_tree_block(tree_root,
2904 btrfs_super_root(disk_super),
2905 generation);
2906 if (IS_ERR(tree_root->node) ||
2907 !extent_buffer_uptodate(tree_root->node)) {
2908 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2909 sb->s_id);
2910 if (!IS_ERR(tree_root->node))
2911 free_extent_buffer(tree_root->node);
2912 tree_root->node = NULL;
2913 goto recovery_tree_root;
2914 }
2915
2916 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2917 tree_root->commit_root = btrfs_root_node(tree_root);
2918 btrfs_set_root_refs(&tree_root->root_item, 1);
2919
2920 mutex_lock(&tree_root->objectid_mutex);
2921 ret = btrfs_find_highest_objectid(tree_root,
2922 &tree_root->highest_objectid);
2923 if (ret) {
2924 mutex_unlock(&tree_root->objectid_mutex);
2925 goto recovery_tree_root;
2926 }
2927
2928 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2929
2930 mutex_unlock(&tree_root->objectid_mutex);
2931
2932 ret = btrfs_read_roots(fs_info, tree_root);
2933 if (ret)
2934 goto recovery_tree_root;
2935
2936 fs_info->generation = generation;
2937 fs_info->last_trans_committed = generation;
2938
2939 ret = btrfs_recover_balance(fs_info);
2940 if (ret) {
2941 printk(KERN_ERR "BTRFS: failed to recover balance\n");
2942 goto fail_block_groups;
2943 }
2944
2945 ret = btrfs_init_dev_stats(fs_info);
2946 if (ret) {
2947 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2948 ret);
2949 goto fail_block_groups;
2950 }
2951
2952 ret = btrfs_init_dev_replace(fs_info);
2953 if (ret) {
2954 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2955 goto fail_block_groups;
2956 }
2957
2958 btrfs_close_extra_devices(fs_devices, 1);
2959
2960 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2961 if (ret) {
2962 pr_err("BTRFS: failed to init sysfs fsid interface: %d\n", ret);
2963 goto fail_block_groups;
2964 }
2965
2966 ret = btrfs_sysfs_add_device(fs_devices);
2967 if (ret) {
2968 pr_err("BTRFS: failed to init sysfs device interface: %d\n", ret);
2969 goto fail_fsdev_sysfs;
2970 }
2971
2972 ret = btrfs_sysfs_add_mounted(fs_info);
2973 if (ret) {
2974 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2975 goto fail_fsdev_sysfs;
2976 }
2977
2978 ret = btrfs_init_space_info(fs_info);
2979 if (ret) {
2980 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2981 goto fail_sysfs;
2982 }
2983
2984 ret = btrfs_read_block_groups(fs_info->extent_root);
2985 if (ret) {
2986 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2987 goto fail_sysfs;
2988 }
2989 fs_info->num_tolerated_disk_barrier_failures =
2990 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2991 if (fs_info->fs_devices->missing_devices >
2992 fs_info->num_tolerated_disk_barrier_failures &&
2993 !(sb->s_flags & MS_RDONLY)) {
2994 pr_warn("BTRFS: missing devices(%llu) exceeds the limit(%d), writeable mount is not allowed\n",
2995 fs_info->fs_devices->missing_devices,
2996 fs_info->num_tolerated_disk_barrier_failures);
2997 goto fail_sysfs;
2998 }
2999
3000 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3001 "btrfs-cleaner");
3002 if (IS_ERR(fs_info->cleaner_kthread))
3003 goto fail_sysfs;
3004
3005 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3006 tree_root,
3007 "btrfs-transaction");
3008 if (IS_ERR(fs_info->transaction_kthread))
3009 goto fail_cleaner;
3010
3011 if (!btrfs_test_opt(tree_root, SSD) &&
3012 !btrfs_test_opt(tree_root, NOSSD) &&
3013 !fs_info->fs_devices->rotating) {
3014 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
3015 "mode\n");
3016 btrfs_set_opt(fs_info->mount_opt, SSD);
3017 }
3018
3019 /*
3020 * Mount does not set all options immediatelly, we can do it now and do
3021 * not have to wait for transaction commit
3022 */
3023 btrfs_apply_pending_changes(fs_info);
3024
3025#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3026 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
3027 ret = btrfsic_mount(tree_root, fs_devices,
3028 btrfs_test_opt(tree_root,
3029 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3030 1 : 0,
3031 fs_info->check_integrity_print_mask);
3032 if (ret)
3033 printk(KERN_WARNING "BTRFS: failed to initialize"
3034 " integrity check module %s\n", sb->s_id);
3035 }
3036#endif
3037 ret = btrfs_read_qgroup_config(fs_info);
3038 if (ret)
3039 goto fail_trans_kthread;
3040
3041 /* do not make disk changes in broken FS or nologreplay is given */
3042 if (btrfs_super_log_root(disk_super) != 0 &&
3043 !btrfs_test_opt(tree_root, NOLOGREPLAY)) {
3044 ret = btrfs_replay_log(fs_info, fs_devices);
3045 if (ret) {
3046 err = ret;
3047 goto fail_qgroup;
3048 }
3049 }
3050
3051 ret = btrfs_find_orphan_roots(tree_root);
3052 if (ret)
3053 goto fail_qgroup;
3054
3055 if (!(sb->s_flags & MS_RDONLY)) {
3056 ret = btrfs_cleanup_fs_roots(fs_info);
3057 if (ret)
3058 goto fail_qgroup;
3059
3060 mutex_lock(&fs_info->cleaner_mutex);
3061 ret = btrfs_recover_relocation(tree_root);
3062 mutex_unlock(&fs_info->cleaner_mutex);
3063 if (ret < 0) {
3064 printk(KERN_WARNING
3065 "BTRFS: failed to recover relocation\n");
3066 err = -EINVAL;
3067 goto fail_qgroup;
3068 }
3069 }
3070
3071 location.objectid = BTRFS_FS_TREE_OBJECTID;
3072 location.type = BTRFS_ROOT_ITEM_KEY;
3073 location.offset = 0;
3074
3075 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3076 if (IS_ERR(fs_info->fs_root)) {
3077 err = PTR_ERR(fs_info->fs_root);
3078 goto fail_qgroup;
3079 }
3080
3081 if (sb->s_flags & MS_RDONLY)
3082 return 0;
3083
3084 if (btrfs_test_opt(tree_root, FREE_SPACE_TREE) &&
3085 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3086 pr_info("BTRFS: creating free space tree\n");
3087 ret = btrfs_create_free_space_tree(fs_info);
3088 if (ret) {
3089 pr_warn("BTRFS: failed to create free space tree %d\n",
3090 ret);
3091 close_ctree(tree_root);
3092 return ret;
3093 }
3094 }
3095
3096 down_read(&fs_info->cleanup_work_sem);
3097 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3098 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3099 up_read(&fs_info->cleanup_work_sem);
3100 close_ctree(tree_root);
3101 return ret;
3102 }
3103 up_read(&fs_info->cleanup_work_sem);
3104
3105 ret = btrfs_resume_balance_async(fs_info);
3106 if (ret) {
3107 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
3108 close_ctree(tree_root);
3109 return ret;
3110 }
3111
3112 ret = btrfs_resume_dev_replace_async(fs_info);
3113 if (ret) {
3114 pr_warn("BTRFS: failed to resume dev_replace\n");
3115 close_ctree(tree_root);
3116 return ret;
3117 }
3118
3119 btrfs_qgroup_rescan_resume(fs_info);
3120
3121 if (btrfs_test_opt(tree_root, CLEAR_CACHE) &&
3122 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3123 pr_info("BTRFS: clearing free space tree\n");
3124 ret = btrfs_clear_free_space_tree(fs_info);
3125 if (ret) {
3126 pr_warn("BTRFS: failed to clear free space tree %d\n",
3127 ret);
3128 close_ctree(tree_root);
3129 return ret;
3130 }
3131 }
3132
3133 if (!fs_info->uuid_root) {
3134 pr_info("BTRFS: creating UUID tree\n");
3135 ret = btrfs_create_uuid_tree(fs_info);
3136 if (ret) {
3137 pr_warn("BTRFS: failed to create the UUID tree %d\n",
3138 ret);
3139 close_ctree(tree_root);
3140 return ret;
3141 }
3142 } else if (btrfs_test_opt(tree_root, RESCAN_UUID_TREE) ||
3143 fs_info->generation !=
3144 btrfs_super_uuid_tree_generation(disk_super)) {
3145 pr_info("BTRFS: checking UUID tree\n");
3146 ret = btrfs_check_uuid_tree(fs_info);
3147 if (ret) {
3148 pr_warn("BTRFS: failed to check the UUID tree %d\n",
3149 ret);
3150 close_ctree(tree_root);
3151 return ret;
3152 }
3153 } else {
3154 fs_info->update_uuid_tree_gen = 1;
3155 }
3156
3157 fs_info->open = 1;
3158
3159 /*
3160 * backuproot only affect mount behavior, and if open_ctree succeeded,
3161 * no need to keep the flag
3162 */
3163 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3164
3165 return 0;
3166
3167fail_qgroup:
3168 btrfs_free_qgroup_config(fs_info);
3169fail_trans_kthread:
3170 kthread_stop(fs_info->transaction_kthread);
3171 btrfs_cleanup_transaction(fs_info->tree_root);
3172 btrfs_free_fs_roots(fs_info);
3173fail_cleaner:
3174 kthread_stop(fs_info->cleaner_kthread);
3175
3176 /*
3177 * make sure we're done with the btree inode before we stop our
3178 * kthreads
3179 */
3180 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3181
3182fail_sysfs:
3183 btrfs_sysfs_remove_mounted(fs_info);
3184
3185fail_fsdev_sysfs:
3186 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3187
3188fail_block_groups:
3189 btrfs_put_block_group_cache(fs_info);
3190 btrfs_free_block_groups(fs_info);
3191
3192fail_tree_roots:
3193 free_root_pointers(fs_info, 1);
3194 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3195
3196fail_sb_buffer:
3197 btrfs_stop_all_workers(fs_info);
3198fail_alloc:
3199fail_iput:
3200 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3201
3202 iput(fs_info->btree_inode);
3203fail_bio_counter:
3204 percpu_counter_destroy(&fs_info->bio_counter);
3205fail_delalloc_bytes:
3206 percpu_counter_destroy(&fs_info->delalloc_bytes);
3207fail_dirty_metadata_bytes:
3208 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3209fail_bdi:
3210 bdi_destroy(&fs_info->bdi);
3211fail_srcu:
3212 cleanup_srcu_struct(&fs_info->subvol_srcu);
3213fail:
3214 btrfs_free_stripe_hash_table(fs_info);
3215 btrfs_close_devices(fs_info->fs_devices);
3216 return err;
3217
3218recovery_tree_root:
3219 if (!btrfs_test_opt(tree_root, USEBACKUPROOT))
3220 goto fail_tree_roots;
3221
3222 free_root_pointers(fs_info, 0);
3223
3224 /* don't use the log in recovery mode, it won't be valid */
3225 btrfs_set_super_log_root(disk_super, 0);
3226
3227 /* we can't trust the free space cache either */
3228 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3229
3230 ret = next_root_backup(fs_info, fs_info->super_copy,
3231 &num_backups_tried, &backup_index);
3232 if (ret == -1)
3233 goto fail_block_groups;
3234 goto retry_root_backup;
3235}
3236
3237static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3238{
3239 if (uptodate) {
3240 set_buffer_uptodate(bh);
3241 } else {
3242 struct btrfs_device *device = (struct btrfs_device *)
3243 bh->b_private;
3244
3245 btrfs_warn_rl_in_rcu(device->dev_root->fs_info,
3246 "lost page write due to IO error on %s",
3247 rcu_str_deref(device->name));
3248 /* note, we dont' set_buffer_write_io_error because we have
3249 * our own ways of dealing with the IO errors
3250 */
3251 clear_buffer_uptodate(bh);
3252 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3253 }
3254 unlock_buffer(bh);
3255 put_bh(bh);
3256}
3257
3258int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3259 struct buffer_head **bh_ret)
3260{
3261 struct buffer_head *bh;
3262 struct btrfs_super_block *super;
3263 u64 bytenr;
3264
3265 bytenr = btrfs_sb_offset(copy_num);
3266 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3267 return -EINVAL;
3268
3269 bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3270 /*
3271 * If we fail to read from the underlying devices, as of now
3272 * the best option we have is to mark it EIO.
3273 */
3274 if (!bh)
3275 return -EIO;
3276
3277 super = (struct btrfs_super_block *)bh->b_data;
3278 if (btrfs_super_bytenr(super) != bytenr ||
3279 btrfs_super_magic(super) != BTRFS_MAGIC) {
3280 brelse(bh);
3281 return -EINVAL;
3282 }
3283
3284 *bh_ret = bh;
3285 return 0;
3286}
3287
3288
3289struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3290{
3291 struct buffer_head *bh;
3292 struct buffer_head *latest = NULL;
3293 struct btrfs_super_block *super;
3294 int i;
3295 u64 transid = 0;
3296 int ret = -EINVAL;
3297
3298 /* we would like to check all the supers, but that would make
3299 * a btrfs mount succeed after a mkfs from a different FS.
3300 * So, we need to add a special mount option to scan for
3301 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3302 */
3303 for (i = 0; i < 1; i++) {
3304 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3305 if (ret)
3306 continue;
3307
3308 super = (struct btrfs_super_block *)bh->b_data;
3309
3310 if (!latest || btrfs_super_generation(super) > transid) {
3311 brelse(latest);
3312 latest = bh;
3313 transid = btrfs_super_generation(super);
3314 } else {
3315 brelse(bh);
3316 }
3317 }
3318
3319 if (!latest)
3320 return ERR_PTR(ret);
3321
3322 return latest;
3323}
3324
3325/*
3326 * this should be called twice, once with wait == 0 and
3327 * once with wait == 1. When wait == 0 is done, all the buffer heads
3328 * we write are pinned.
3329 *
3330 * They are released when wait == 1 is done.
3331 * max_mirrors must be the same for both runs, and it indicates how
3332 * many supers on this one device should be written.
3333 *
3334 * max_mirrors == 0 means to write them all.
3335 */
3336static int write_dev_supers(struct btrfs_device *device,
3337 struct btrfs_super_block *sb,
3338 int do_barriers, int wait, int max_mirrors)
3339{
3340 struct buffer_head *bh;
3341 int i;
3342 int ret;
3343 int errors = 0;
3344 u32 crc;
3345 u64 bytenr;
3346
3347 if (max_mirrors == 0)
3348 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3349
3350 for (i = 0; i < max_mirrors; i++) {
3351 bytenr = btrfs_sb_offset(i);
3352 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3353 device->commit_total_bytes)
3354 break;
3355
3356 if (wait) {
3357 bh = __find_get_block(device->bdev, bytenr / 4096,
3358 BTRFS_SUPER_INFO_SIZE);
3359 if (!bh) {
3360 errors++;
3361 continue;
3362 }
3363 wait_on_buffer(bh);
3364 if (!buffer_uptodate(bh))
3365 errors++;
3366
3367 /* drop our reference */
3368 brelse(bh);
3369
3370 /* drop the reference from the wait == 0 run */
3371 brelse(bh);
3372 continue;
3373 } else {
3374 btrfs_set_super_bytenr(sb, bytenr);
3375
3376 crc = ~(u32)0;
3377 crc = btrfs_csum_data((char *)sb +
3378 BTRFS_CSUM_SIZE, crc,
3379 BTRFS_SUPER_INFO_SIZE -
3380 BTRFS_CSUM_SIZE);
3381 btrfs_csum_final(crc, sb->csum);
3382
3383 /*
3384 * one reference for us, and we leave it for the
3385 * caller
3386 */
3387 bh = __getblk(device->bdev, bytenr / 4096,
3388 BTRFS_SUPER_INFO_SIZE);
3389 if (!bh) {
3390 btrfs_err(device->dev_root->fs_info,
3391 "couldn't get super buffer head for bytenr %llu",
3392 bytenr);
3393 errors++;
3394 continue;
3395 }
3396
3397 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3398
3399 /* one reference for submit_bh */
3400 get_bh(bh);
3401
3402 set_buffer_uptodate(bh);
3403 lock_buffer(bh);
3404 bh->b_end_io = btrfs_end_buffer_write_sync;
3405 bh->b_private = device;
3406 }
3407
3408 /*
3409 * we fua the first super. The others we allow
3410 * to go down lazy.
3411 */
3412 if (i == 0)
3413 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3414 else
3415 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3416 if (ret)
3417 errors++;
3418 }
3419 return errors < i ? 0 : -1;
3420}
3421
3422/*
3423 * endio for the write_dev_flush, this will wake anyone waiting
3424 * for the barrier when it is done
3425 */
3426static void btrfs_end_empty_barrier(struct bio *bio)
3427{
3428 if (bio->bi_private)
3429 complete(bio->bi_private);
3430 bio_put(bio);
3431}
3432
3433/*
3434 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3435 * sent down. With wait == 1, it waits for the previous flush.
3436 *
3437 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3438 * capable
3439 */
3440static int write_dev_flush(struct btrfs_device *device, int wait)
3441{
3442 struct bio *bio;
3443 int ret = 0;
3444
3445 if (device->nobarriers)
3446 return 0;
3447
3448 if (wait) {
3449 bio = device->flush_bio;
3450 if (!bio)
3451 return 0;
3452
3453 wait_for_completion(&device->flush_wait);
3454
3455 if (bio->bi_error) {
3456 ret = bio->bi_error;
3457 btrfs_dev_stat_inc_and_print(device,
3458 BTRFS_DEV_STAT_FLUSH_ERRS);
3459 }
3460
3461 /* drop the reference from the wait == 0 run */
3462 bio_put(bio);
3463 device->flush_bio = NULL;
3464
3465 return ret;
3466 }
3467
3468 /*
3469 * one reference for us, and we leave it for the
3470 * caller
3471 */
3472 device->flush_bio = NULL;
3473 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3474 if (!bio)
3475 return -ENOMEM;
3476
3477 bio->bi_end_io = btrfs_end_empty_barrier;
3478 bio->bi_bdev = device->bdev;
3479 init_completion(&device->flush_wait);
3480 bio->bi_private = &device->flush_wait;
3481 device->flush_bio = bio;
3482
3483 bio_get(bio);
3484 btrfsic_submit_bio(WRITE_FLUSH, bio);
3485
3486 return 0;
3487}
3488
3489/*
3490 * send an empty flush down to each device in parallel,
3491 * then wait for them
3492 */
3493static int barrier_all_devices(struct btrfs_fs_info *info)
3494{
3495 struct list_head *head;
3496 struct btrfs_device *dev;
3497 int errors_send = 0;
3498 int errors_wait = 0;
3499 int ret;
3500
3501 /* send down all the barriers */
3502 head = &info->fs_devices->devices;
3503 list_for_each_entry_rcu(dev, head, dev_list) {
3504 if (dev->missing)
3505 continue;
3506 if (!dev->bdev) {
3507 errors_send++;
3508 continue;
3509 }
3510 if (!dev->in_fs_metadata || !dev->writeable)
3511 continue;
3512
3513 ret = write_dev_flush(dev, 0);
3514 if (ret)
3515 errors_send++;
3516 }
3517
3518 /* wait for all the barriers */
3519 list_for_each_entry_rcu(dev, head, dev_list) {
3520 if (dev->missing)
3521 continue;
3522 if (!dev->bdev) {
3523 errors_wait++;
3524 continue;
3525 }
3526 if (!dev->in_fs_metadata || !dev->writeable)
3527 continue;
3528
3529 ret = write_dev_flush(dev, 1);
3530 if (ret)
3531 errors_wait++;
3532 }
3533 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3534 errors_wait > info->num_tolerated_disk_barrier_failures)
3535 return -EIO;
3536 return 0;
3537}
3538
3539int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3540{
3541 int raid_type;
3542 int min_tolerated = INT_MAX;
3543
3544 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3545 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3546 min_tolerated = min(min_tolerated,
3547 btrfs_raid_array[BTRFS_RAID_SINGLE].
3548 tolerated_failures);
3549
3550 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3551 if (raid_type == BTRFS_RAID_SINGLE)
3552 continue;
3553 if (!(flags & btrfs_raid_group[raid_type]))
3554 continue;
3555 min_tolerated = min(min_tolerated,
3556 btrfs_raid_array[raid_type].
3557 tolerated_failures);
3558 }
3559
3560 if (min_tolerated == INT_MAX) {
3561 pr_warn("BTRFS: unknown raid flag: %llu\n", flags);
3562 min_tolerated = 0;
3563 }
3564
3565 return min_tolerated;
3566}
3567
3568int btrfs_calc_num_tolerated_disk_barrier_failures(
3569 struct btrfs_fs_info *fs_info)
3570{
3571 struct btrfs_ioctl_space_info space;
3572 struct btrfs_space_info *sinfo;
3573 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3574 BTRFS_BLOCK_GROUP_SYSTEM,
3575 BTRFS_BLOCK_GROUP_METADATA,
3576 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3577 int i;
3578 int c;
3579 int num_tolerated_disk_barrier_failures =
3580 (int)fs_info->fs_devices->num_devices;
3581
3582 for (i = 0; i < ARRAY_SIZE(types); i++) {
3583 struct btrfs_space_info *tmp;
3584
3585 sinfo = NULL;
3586 rcu_read_lock();
3587 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3588 if (tmp->flags == types[i]) {
3589 sinfo = tmp;
3590 break;
3591 }
3592 }
3593 rcu_read_unlock();
3594
3595 if (!sinfo)
3596 continue;
3597
3598 down_read(&sinfo->groups_sem);
3599 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3600 u64 flags;
3601
3602 if (list_empty(&sinfo->block_groups[c]))
3603 continue;
3604
3605 btrfs_get_block_group_info(&sinfo->block_groups[c],
3606 &space);
3607 if (space.total_bytes == 0 || space.used_bytes == 0)
3608 continue;
3609 flags = space.flags;
3610
3611 num_tolerated_disk_barrier_failures = min(
3612 num_tolerated_disk_barrier_failures,
3613 btrfs_get_num_tolerated_disk_barrier_failures(
3614 flags));
3615 }
3616 up_read(&sinfo->groups_sem);
3617 }
3618
3619 return num_tolerated_disk_barrier_failures;
3620}
3621
3622static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3623{
3624 struct list_head *head;
3625 struct btrfs_device *dev;
3626 struct btrfs_super_block *sb;
3627 struct btrfs_dev_item *dev_item;
3628 int ret;
3629 int do_barriers;
3630 int max_errors;
3631 int total_errors = 0;
3632 u64 flags;
3633
3634 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3635 backup_super_roots(root->fs_info);
3636
3637 sb = root->fs_info->super_for_commit;
3638 dev_item = &sb->dev_item;
3639
3640 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3641 head = &root->fs_info->fs_devices->devices;
3642 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3643
3644 if (do_barriers) {
3645 ret = barrier_all_devices(root->fs_info);
3646 if (ret) {
3647 mutex_unlock(
3648 &root->fs_info->fs_devices->device_list_mutex);
3649 btrfs_std_error(root->fs_info, ret,
3650 "errors while submitting device barriers.");
3651 return ret;
3652 }
3653 }
3654
3655 list_for_each_entry_rcu(dev, head, dev_list) {
3656 if (!dev->bdev) {
3657 total_errors++;
3658 continue;
3659 }
3660 if (!dev->in_fs_metadata || !dev->writeable)
3661 continue;
3662
3663 btrfs_set_stack_device_generation(dev_item, 0);
3664 btrfs_set_stack_device_type(dev_item, dev->type);
3665 btrfs_set_stack_device_id(dev_item, dev->devid);
3666 btrfs_set_stack_device_total_bytes(dev_item,
3667 dev->commit_total_bytes);
3668 btrfs_set_stack_device_bytes_used(dev_item,
3669 dev->commit_bytes_used);
3670 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3671 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3672 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3673 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3674 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3675
3676 flags = btrfs_super_flags(sb);
3677 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3678
3679 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3680 if (ret)
3681 total_errors++;
3682 }
3683 if (total_errors > max_errors) {
3684 btrfs_err(root->fs_info, "%d errors while writing supers",
3685 total_errors);
3686 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3687
3688 /* FUA is masked off if unsupported and can't be the reason */
3689 btrfs_std_error(root->fs_info, -EIO,
3690 "%d errors while writing supers", total_errors);
3691 return -EIO;
3692 }
3693
3694 total_errors = 0;
3695 list_for_each_entry_rcu(dev, head, dev_list) {
3696 if (!dev->bdev)
3697 continue;
3698 if (!dev->in_fs_metadata || !dev->writeable)
3699 continue;
3700
3701 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3702 if (ret)
3703 total_errors++;
3704 }
3705 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3706 if (total_errors > max_errors) {
3707 btrfs_std_error(root->fs_info, -EIO,
3708 "%d errors while writing supers", total_errors);
3709 return -EIO;
3710 }
3711 return 0;
3712}
3713
3714int write_ctree_super(struct btrfs_trans_handle *trans,
3715 struct btrfs_root *root, int max_mirrors)
3716{
3717 return write_all_supers(root, max_mirrors);
3718}
3719
3720/* Drop a fs root from the radix tree and free it. */
3721void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3722 struct btrfs_root *root)
3723{
3724 spin_lock(&fs_info->fs_roots_radix_lock);
3725 radix_tree_delete(&fs_info->fs_roots_radix,
3726 (unsigned long)root->root_key.objectid);
3727 spin_unlock(&fs_info->fs_roots_radix_lock);
3728
3729 if (btrfs_root_refs(&root->root_item) == 0)
3730 synchronize_srcu(&fs_info->subvol_srcu);
3731
3732 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3733 btrfs_free_log(NULL, root);
3734
3735 if (root->free_ino_pinned)
3736 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3737 if (root->free_ino_ctl)
3738 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3739 free_fs_root(root);
3740}
3741
3742static void free_fs_root(struct btrfs_root *root)
3743{
3744 iput(root->ino_cache_inode);
3745 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3746 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3747 root->orphan_block_rsv = NULL;
3748 if (root->anon_dev)
3749 free_anon_bdev(root->anon_dev);
3750 if (root->subv_writers)
3751 btrfs_free_subvolume_writers(root->subv_writers);
3752 free_extent_buffer(root->node);
3753 free_extent_buffer(root->commit_root);
3754 kfree(root->free_ino_ctl);
3755 kfree(root->free_ino_pinned);
3756 kfree(root->name);
3757 btrfs_put_fs_root(root);
3758}
3759
3760void btrfs_free_fs_root(struct btrfs_root *root)
3761{
3762 free_fs_root(root);
3763}
3764
3765int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3766{
3767 u64 root_objectid = 0;
3768 struct btrfs_root *gang[8];
3769 int i = 0;
3770 int err = 0;
3771 unsigned int ret = 0;
3772 int index;
3773
3774 while (1) {
3775 index = srcu_read_lock(&fs_info->subvol_srcu);
3776 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3777 (void **)gang, root_objectid,
3778 ARRAY_SIZE(gang));
3779 if (!ret) {
3780 srcu_read_unlock(&fs_info->subvol_srcu, index);
3781 break;
3782 }
3783 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3784
3785 for (i = 0; i < ret; i++) {
3786 /* Avoid to grab roots in dead_roots */
3787 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3788 gang[i] = NULL;
3789 continue;
3790 }
3791 /* grab all the search result for later use */
3792 gang[i] = btrfs_grab_fs_root(gang[i]);
3793 }
3794 srcu_read_unlock(&fs_info->subvol_srcu, index);
3795
3796 for (i = 0; i < ret; i++) {
3797 if (!gang[i])
3798 continue;
3799 root_objectid = gang[i]->root_key.objectid;
3800 err = btrfs_orphan_cleanup(gang[i]);
3801 if (err)
3802 break;
3803 btrfs_put_fs_root(gang[i]);
3804 }
3805 root_objectid++;
3806 }
3807
3808 /* release the uncleaned roots due to error */
3809 for (; i < ret; i++) {
3810 if (gang[i])
3811 btrfs_put_fs_root(gang[i]);
3812 }
3813 return err;
3814}
3815
3816int btrfs_commit_super(struct btrfs_root *root)
3817{
3818 struct btrfs_trans_handle *trans;
3819
3820 mutex_lock(&root->fs_info->cleaner_mutex);
3821 btrfs_run_delayed_iputs(root);
3822 mutex_unlock(&root->fs_info->cleaner_mutex);
3823 wake_up_process(root->fs_info->cleaner_kthread);
3824
3825 /* wait until ongoing cleanup work done */
3826 down_write(&root->fs_info->cleanup_work_sem);
3827 up_write(&root->fs_info->cleanup_work_sem);
3828
3829 trans = btrfs_join_transaction(root);
3830 if (IS_ERR(trans))
3831 return PTR_ERR(trans);
3832 return btrfs_commit_transaction(trans, root);
3833}
3834
3835void close_ctree(struct btrfs_root *root)
3836{
3837 struct btrfs_fs_info *fs_info = root->fs_info;
3838 int ret;
3839
3840 fs_info->closing = 1;
3841 smp_mb();
3842
3843 /* wait for the qgroup rescan worker to stop */
3844 btrfs_qgroup_wait_for_completion(fs_info);
3845
3846 /* wait for the uuid_scan task to finish */
3847 down(&fs_info->uuid_tree_rescan_sem);
3848 /* avoid complains from lockdep et al., set sem back to initial state */
3849 up(&fs_info->uuid_tree_rescan_sem);
3850
3851 /* pause restriper - we want to resume on mount */
3852 btrfs_pause_balance(fs_info);
3853
3854 btrfs_dev_replace_suspend_for_unmount(fs_info);
3855
3856 btrfs_scrub_cancel(fs_info);
3857
3858 /* wait for any defraggers to finish */
3859 wait_event(fs_info->transaction_wait,
3860 (atomic_read(&fs_info->defrag_running) == 0));
3861
3862 /* clear out the rbtree of defraggable inodes */
3863 btrfs_cleanup_defrag_inodes(fs_info);
3864
3865 cancel_work_sync(&fs_info->async_reclaim_work);
3866
3867 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3868 /*
3869 * If the cleaner thread is stopped and there are
3870 * block groups queued for removal, the deletion will be
3871 * skipped when we quit the cleaner thread.
3872 */
3873 btrfs_delete_unused_bgs(root->fs_info);
3874
3875 ret = btrfs_commit_super(root);
3876 if (ret)
3877 btrfs_err(fs_info, "commit super ret %d", ret);
3878 }
3879
3880 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3881 btrfs_error_commit_super(root);
3882
3883 kthread_stop(fs_info->transaction_kthread);
3884 kthread_stop(fs_info->cleaner_kthread);
3885
3886 fs_info->closing = 2;
3887 smp_mb();
3888
3889 btrfs_free_qgroup_config(fs_info);
3890
3891 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3892 btrfs_info(fs_info, "at unmount delalloc count %lld",
3893 percpu_counter_sum(&fs_info->delalloc_bytes));
3894 }
3895
3896 btrfs_sysfs_remove_mounted(fs_info);
3897 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3898
3899 btrfs_free_fs_roots(fs_info);
3900
3901 btrfs_put_block_group_cache(fs_info);
3902
3903 btrfs_free_block_groups(fs_info);
3904
3905 /*
3906 * we must make sure there is not any read request to
3907 * submit after we stopping all workers.
3908 */
3909 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3910 btrfs_stop_all_workers(fs_info);
3911
3912 fs_info->open = 0;
3913 free_root_pointers(fs_info, 1);
3914
3915 iput(fs_info->btree_inode);
3916
3917#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3918 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3919 btrfsic_unmount(root, fs_info->fs_devices);
3920#endif
3921
3922 btrfs_close_devices(fs_info->fs_devices);
3923 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3924
3925 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3926 percpu_counter_destroy(&fs_info->delalloc_bytes);
3927 percpu_counter_destroy(&fs_info->bio_counter);
3928 bdi_destroy(&fs_info->bdi);
3929 cleanup_srcu_struct(&fs_info->subvol_srcu);
3930
3931 btrfs_free_stripe_hash_table(fs_info);
3932
3933 __btrfs_free_block_rsv(root->orphan_block_rsv);
3934 root->orphan_block_rsv = NULL;
3935
3936 lock_chunks(root);
3937 while (!list_empty(&fs_info->pinned_chunks)) {
3938 struct extent_map *em;
3939
3940 em = list_first_entry(&fs_info->pinned_chunks,
3941 struct extent_map, list);
3942 list_del_init(&em->list);
3943 free_extent_map(em);
3944 }
3945 unlock_chunks(root);
3946}
3947
3948int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3949 int atomic)
3950{
3951 int ret;
3952 struct inode *btree_inode = buf->pages[0]->mapping->host;
3953
3954 ret = extent_buffer_uptodate(buf);
3955 if (!ret)
3956 return ret;
3957
3958 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3959 parent_transid, atomic);
3960 if (ret == -EAGAIN)
3961 return ret;
3962 return !ret;
3963}
3964
3965void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3966{
3967 struct btrfs_root *root;
3968 u64 transid = btrfs_header_generation(buf);
3969 int was_dirty;
3970
3971#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3972 /*
3973 * This is a fast path so only do this check if we have sanity tests
3974 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3975 * outside of the sanity tests.
3976 */
3977 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3978 return;
3979#endif
3980 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3981 btrfs_assert_tree_locked(buf);
3982 if (transid != root->fs_info->generation)
3983 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3984 "found %llu running %llu\n",
3985 buf->start, transid, root->fs_info->generation);
3986 was_dirty = set_extent_buffer_dirty(buf);
3987 if (!was_dirty)
3988 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3989 buf->len,
3990 root->fs_info->dirty_metadata_batch);
3991#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3992 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3993 btrfs_print_leaf(root, buf);
3994 ASSERT(0);
3995 }
3996#endif
3997}
3998
3999static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
4000 int flush_delayed)
4001{
4002 /*
4003 * looks as though older kernels can get into trouble with
4004 * this code, they end up stuck in balance_dirty_pages forever
4005 */
4006 int ret;
4007
4008 if (current->flags & PF_MEMALLOC)
4009 return;
4010
4011 if (flush_delayed)
4012 btrfs_balance_delayed_items(root);
4013
4014 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
4015 BTRFS_DIRTY_METADATA_THRESH);
4016 if (ret > 0) {
4017 balance_dirty_pages_ratelimited(
4018 root->fs_info->btree_inode->i_mapping);
4019 }
4020}
4021
4022void btrfs_btree_balance_dirty(struct btrfs_root *root)
4023{
4024 __btrfs_btree_balance_dirty(root, 1);
4025}
4026
4027void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
4028{
4029 __btrfs_btree_balance_dirty(root, 0);
4030}
4031
4032int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4033{
4034 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4035 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
4036}
4037
4038static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
4039 int read_only)
4040{
4041 struct btrfs_super_block *sb = fs_info->super_copy;
4042 u64 nodesize = btrfs_super_nodesize(sb);
4043 u64 sectorsize = btrfs_super_sectorsize(sb);
4044 int ret = 0;
4045
4046 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4047 printk(KERN_ERR "BTRFS: no valid FS found\n");
4048 ret = -EINVAL;
4049 }
4050 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4051 printk(KERN_WARNING "BTRFS: unrecognized super flag: %llu\n",
4052 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4053 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4054 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
4055 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4056 ret = -EINVAL;
4057 }
4058 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4059 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
4060 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4061 ret = -EINVAL;
4062 }
4063 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4064 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
4065 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4066 ret = -EINVAL;
4067 }
4068
4069 /*
4070 * Check sectorsize and nodesize first, other check will need it.
4071 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4072 */
4073 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4074 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4075 printk(KERN_ERR "BTRFS: invalid sectorsize %llu\n", sectorsize);
4076 ret = -EINVAL;
4077 }
4078 /* Only PAGE SIZE is supported yet */
4079 if (sectorsize != PAGE_SIZE) {
4080 printk(KERN_ERR "BTRFS: sectorsize %llu not supported yet, only support %lu\n",
4081 sectorsize, PAGE_SIZE);
4082 ret = -EINVAL;
4083 }
4084 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4085 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4086 printk(KERN_ERR "BTRFS: invalid nodesize %llu\n", nodesize);
4087 ret = -EINVAL;
4088 }
4089 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4090 printk(KERN_ERR "BTRFS: invalid leafsize %u, should be %llu\n",
4091 le32_to_cpu(sb->__unused_leafsize),
4092 nodesize);
4093 ret = -EINVAL;
4094 }
4095
4096 /* Root alignment check */
4097 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4098 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
4099 btrfs_super_root(sb));
4100 ret = -EINVAL;
4101 }
4102 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4103 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
4104 btrfs_super_chunk_root(sb));
4105 ret = -EINVAL;
4106 }
4107 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4108 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
4109 btrfs_super_log_root(sb));
4110 ret = -EINVAL;
4111 }
4112
4113 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4114 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4115 fs_info->fsid, sb->dev_item.fsid);
4116 ret = -EINVAL;
4117 }
4118
4119 /*
4120 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4121 * done later
4122 */
4123 if (btrfs_super_num_devices(sb) > (1UL << 31))
4124 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
4125 btrfs_super_num_devices(sb));
4126 if (btrfs_super_num_devices(sb) == 0) {
4127 printk(KERN_ERR "BTRFS: number of devices is 0\n");
4128 ret = -EINVAL;
4129 }
4130
4131 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4132 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
4133 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4134 ret = -EINVAL;
4135 }
4136
4137 /*
4138 * Obvious sys_chunk_array corruptions, it must hold at least one key
4139 * and one chunk
4140 */
4141 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4142 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
4143 btrfs_super_sys_array_size(sb),
4144 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4145 ret = -EINVAL;
4146 }
4147 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4148 + sizeof(struct btrfs_chunk)) {
4149 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
4150 btrfs_super_sys_array_size(sb),
4151 sizeof(struct btrfs_disk_key)
4152 + sizeof(struct btrfs_chunk));
4153 ret = -EINVAL;
4154 }
4155
4156 /*
4157 * The generation is a global counter, we'll trust it more than the others
4158 * but it's still possible that it's the one that's wrong.
4159 */
4160 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4161 printk(KERN_WARNING
4162 "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4163 btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4164 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4165 && btrfs_super_cache_generation(sb) != (u64)-1)
4166 printk(KERN_WARNING
4167 "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4168 btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4169
4170 return ret;
4171}
4172
4173static void btrfs_error_commit_super(struct btrfs_root *root)
4174{
4175 mutex_lock(&root->fs_info->cleaner_mutex);
4176 btrfs_run_delayed_iputs(root);
4177 mutex_unlock(&root->fs_info->cleaner_mutex);
4178
4179 down_write(&root->fs_info->cleanup_work_sem);
4180 up_write(&root->fs_info->cleanup_work_sem);
4181
4182 /* cleanup FS via transaction */
4183 btrfs_cleanup_transaction(root);
4184}
4185
4186static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4187{
4188 struct btrfs_ordered_extent *ordered;
4189
4190 spin_lock(&root->ordered_extent_lock);
4191 /*
4192 * This will just short circuit the ordered completion stuff which will
4193 * make sure the ordered extent gets properly cleaned up.
4194 */
4195 list_for_each_entry(ordered, &root->ordered_extents,
4196 root_extent_list)
4197 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4198 spin_unlock(&root->ordered_extent_lock);
4199}
4200
4201static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4202{
4203 struct btrfs_root *root;
4204 struct list_head splice;
4205
4206 INIT_LIST_HEAD(&splice);
4207
4208 spin_lock(&fs_info->ordered_root_lock);
4209 list_splice_init(&fs_info->ordered_roots, &splice);
4210 while (!list_empty(&splice)) {
4211 root = list_first_entry(&splice, struct btrfs_root,
4212 ordered_root);
4213 list_move_tail(&root->ordered_root,
4214 &fs_info->ordered_roots);
4215
4216 spin_unlock(&fs_info->ordered_root_lock);
4217 btrfs_destroy_ordered_extents(root);
4218
4219 cond_resched();
4220 spin_lock(&fs_info->ordered_root_lock);
4221 }
4222 spin_unlock(&fs_info->ordered_root_lock);
4223}
4224
4225static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4226 struct btrfs_root *root)
4227{
4228 struct rb_node *node;
4229 struct btrfs_delayed_ref_root *delayed_refs;
4230 struct btrfs_delayed_ref_node *ref;
4231 int ret = 0;
4232
4233 delayed_refs = &trans->delayed_refs;
4234
4235 spin_lock(&delayed_refs->lock);
4236 if (atomic_read(&delayed_refs->num_entries) == 0) {
4237 spin_unlock(&delayed_refs->lock);
4238 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4239 return ret;
4240 }
4241
4242 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4243 struct btrfs_delayed_ref_head *head;
4244 struct btrfs_delayed_ref_node *tmp;
4245 bool pin_bytes = false;
4246
4247 head = rb_entry(node, struct btrfs_delayed_ref_head,
4248 href_node);
4249 if (!mutex_trylock(&head->mutex)) {
4250 atomic_inc(&head->node.refs);
4251 spin_unlock(&delayed_refs->lock);
4252
4253 mutex_lock(&head->mutex);
4254 mutex_unlock(&head->mutex);
4255 btrfs_put_delayed_ref(&head->node);
4256 spin_lock(&delayed_refs->lock);
4257 continue;
4258 }
4259 spin_lock(&head->lock);
4260 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4261 list) {
4262 ref->in_tree = 0;
4263 list_del(&ref->list);
4264 atomic_dec(&delayed_refs->num_entries);
4265 btrfs_put_delayed_ref(ref);
4266 }
4267 if (head->must_insert_reserved)
4268 pin_bytes = true;
4269 btrfs_free_delayed_extent_op(head->extent_op);
4270 delayed_refs->num_heads--;
4271 if (head->processing == 0)
4272 delayed_refs->num_heads_ready--;
4273 atomic_dec(&delayed_refs->num_entries);
4274 head->node.in_tree = 0;
4275 rb_erase(&head->href_node, &delayed_refs->href_root);
4276 spin_unlock(&head->lock);
4277 spin_unlock(&delayed_refs->lock);
4278 mutex_unlock(&head->mutex);
4279
4280 if (pin_bytes)
4281 btrfs_pin_extent(root, head->node.bytenr,
4282 head->node.num_bytes, 1);
4283 btrfs_put_delayed_ref(&head->node);
4284 cond_resched();
4285 spin_lock(&delayed_refs->lock);
4286 }
4287
4288 spin_unlock(&delayed_refs->lock);
4289
4290 return ret;
4291}
4292
4293static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4294{
4295 struct btrfs_inode *btrfs_inode;
4296 struct list_head splice;
4297
4298 INIT_LIST_HEAD(&splice);
4299
4300 spin_lock(&root->delalloc_lock);
4301 list_splice_init(&root->delalloc_inodes, &splice);
4302
4303 while (!list_empty(&splice)) {
4304 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4305 delalloc_inodes);
4306
4307 list_del_init(&btrfs_inode->delalloc_inodes);
4308 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4309 &btrfs_inode->runtime_flags);
4310 spin_unlock(&root->delalloc_lock);
4311
4312 btrfs_invalidate_inodes(btrfs_inode->root);
4313
4314 spin_lock(&root->delalloc_lock);
4315 }
4316
4317 spin_unlock(&root->delalloc_lock);
4318}
4319
4320static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4321{
4322 struct btrfs_root *root;
4323 struct list_head splice;
4324
4325 INIT_LIST_HEAD(&splice);
4326
4327 spin_lock(&fs_info->delalloc_root_lock);
4328 list_splice_init(&fs_info->delalloc_roots, &splice);
4329 while (!list_empty(&splice)) {
4330 root = list_first_entry(&splice, struct btrfs_root,
4331 delalloc_root);
4332 list_del_init(&root->delalloc_root);
4333 root = btrfs_grab_fs_root(root);
4334 BUG_ON(!root);
4335 spin_unlock(&fs_info->delalloc_root_lock);
4336
4337 btrfs_destroy_delalloc_inodes(root);
4338 btrfs_put_fs_root(root);
4339
4340 spin_lock(&fs_info->delalloc_root_lock);
4341 }
4342 spin_unlock(&fs_info->delalloc_root_lock);
4343}
4344
4345static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4346 struct extent_io_tree *dirty_pages,
4347 int mark)
4348{
4349 int ret;
4350 struct extent_buffer *eb;
4351 u64 start = 0;
4352 u64 end;
4353
4354 while (1) {
4355 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4356 mark, NULL);
4357 if (ret)
4358 break;
4359
4360 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4361 while (start <= end) {
4362 eb = btrfs_find_tree_block(root->fs_info, start);
4363 start += root->nodesize;
4364 if (!eb)
4365 continue;
4366 wait_on_extent_buffer_writeback(eb);
4367
4368 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4369 &eb->bflags))
4370 clear_extent_buffer_dirty(eb);
4371 free_extent_buffer_stale(eb);
4372 }
4373 }
4374
4375 return ret;
4376}
4377
4378static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4379 struct extent_io_tree *pinned_extents)
4380{
4381 struct extent_io_tree *unpin;
4382 u64 start;
4383 u64 end;
4384 int ret;
4385 bool loop = true;
4386
4387 unpin = pinned_extents;
4388again:
4389 while (1) {
4390 ret = find_first_extent_bit(unpin, 0, &start, &end,
4391 EXTENT_DIRTY, NULL);
4392 if (ret)
4393 break;
4394
4395 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4396 btrfs_error_unpin_extent_range(root, start, end);
4397 cond_resched();
4398 }
4399
4400 if (loop) {
4401 if (unpin == &root->fs_info->freed_extents[0])
4402 unpin = &root->fs_info->freed_extents[1];
4403 else
4404 unpin = &root->fs_info->freed_extents[0];
4405 loop = false;
4406 goto again;
4407 }
4408
4409 return 0;
4410}
4411
4412void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4413 struct btrfs_root *root)
4414{
4415 btrfs_destroy_delayed_refs(cur_trans, root);
4416
4417 cur_trans->state = TRANS_STATE_COMMIT_START;
4418 wake_up(&root->fs_info->transaction_blocked_wait);
4419
4420 cur_trans->state = TRANS_STATE_UNBLOCKED;
4421 wake_up(&root->fs_info->transaction_wait);
4422
4423 btrfs_destroy_delayed_inodes(root);
4424 btrfs_assert_delayed_root_empty(root);
4425
4426 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4427 EXTENT_DIRTY);
4428 btrfs_destroy_pinned_extent(root,
4429 root->fs_info->pinned_extents);
4430
4431 cur_trans->state =TRANS_STATE_COMPLETED;
4432 wake_up(&cur_trans->commit_wait);
4433
4434 /*
4435 memset(cur_trans, 0, sizeof(*cur_trans));
4436 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4437 */
4438}
4439
4440static int btrfs_cleanup_transaction(struct btrfs_root *root)
4441{
4442 struct btrfs_transaction *t;
4443
4444 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4445
4446 spin_lock(&root->fs_info->trans_lock);
4447 while (!list_empty(&root->fs_info->trans_list)) {
4448 t = list_first_entry(&root->fs_info->trans_list,
4449 struct btrfs_transaction, list);
4450 if (t->state >= TRANS_STATE_COMMIT_START) {
4451 atomic_inc(&t->use_count);
4452 spin_unlock(&root->fs_info->trans_lock);
4453 btrfs_wait_for_commit(root, t->transid);
4454 btrfs_put_transaction(t);
4455 spin_lock(&root->fs_info->trans_lock);
4456 continue;
4457 }
4458 if (t == root->fs_info->running_transaction) {
4459 t->state = TRANS_STATE_COMMIT_DOING;
4460 spin_unlock(&root->fs_info->trans_lock);
4461 /*
4462 * We wait for 0 num_writers since we don't hold a trans
4463 * handle open currently for this transaction.
4464 */
4465 wait_event(t->writer_wait,
4466 atomic_read(&t->num_writers) == 0);
4467 } else {
4468 spin_unlock(&root->fs_info->trans_lock);
4469 }
4470 btrfs_cleanup_one_transaction(t, root);
4471
4472 spin_lock(&root->fs_info->trans_lock);
4473 if (t == root->fs_info->running_transaction)
4474 root->fs_info->running_transaction = NULL;
4475 list_del_init(&t->list);
4476 spin_unlock(&root->fs_info->trans_lock);
4477
4478 btrfs_put_transaction(t);
4479 trace_btrfs_transaction_commit(root);
4480 spin_lock(&root->fs_info->trans_lock);
4481 }
4482 spin_unlock(&root->fs_info->trans_lock);
4483 btrfs_destroy_all_ordered_extents(root->fs_info);
4484 btrfs_destroy_delayed_inodes(root);
4485 btrfs_assert_delayed_root_empty(root);
4486 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4487 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4488 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4489
4490 return 0;
4491}
4492
4493static const struct extent_io_ops btree_extent_io_ops = {
4494 .readpage_end_io_hook = btree_readpage_end_io_hook,
4495 .readpage_io_failed_hook = btree_io_failed_hook,
4496 .submit_bio_hook = btree_submit_bio_hook,
4497 /* note we're sharing with inode.c for the merge bio hook */
4498 .merge_bio_hook = btrfs_merge_bio_hook,
4499};