Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6#include <linux/fs.h>
7#include <linux/blkdev.h>
8#include <linux/radix-tree.h>
9#include <linux/writeback.h>
10#include <linux/workqueue.h>
11#include <linux/kthread.h>
12#include <linux/slab.h>
13#include <linux/migrate.h>
14#include <linux/ratelimit.h>
15#include <linux/uuid.h>
16#include <linux/semaphore.h>
17#include <linux/error-injection.h>
18#include <linux/crc32c.h>
19#include <linux/sched/mm.h>
20#include <asm/unaligned.h>
21#include <crypto/hash.h>
22#include "ctree.h"
23#include "disk-io.h"
24#include "transaction.h"
25#include "btrfs_inode.h"
26#include "volumes.h"
27#include "print-tree.h"
28#include "locking.h"
29#include "tree-log.h"
30#include "free-space-cache.h"
31#include "free-space-tree.h"
32#include "inode-map.h"
33#include "check-integrity.h"
34#include "rcu-string.h"
35#include "dev-replace.h"
36#include "raid56.h"
37#include "sysfs.h"
38#include "qgroup.h"
39#include "compression.h"
40#include "tree-checker.h"
41#include "ref-verify.h"
42#include "block-group.h"
43#include "discard.h"
44#include "space-info.h"
45
46#define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
47 BTRFS_HEADER_FLAG_RELOC |\
48 BTRFS_SUPER_FLAG_ERROR |\
49 BTRFS_SUPER_FLAG_SEEDING |\
50 BTRFS_SUPER_FLAG_METADUMP |\
51 BTRFS_SUPER_FLAG_METADUMP_V2)
52
53static const struct extent_io_ops btree_extent_io_ops;
54static void end_workqueue_fn(struct btrfs_work *work);
55static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
56static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
57 struct btrfs_fs_info *fs_info);
58static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
59static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
60 struct extent_io_tree *dirty_pages,
61 int mark);
62static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
63 struct extent_io_tree *pinned_extents);
64static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
65static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
66
67/*
68 * btrfs_end_io_wq structs are used to do processing in task context when an IO
69 * is complete. This is used during reads to verify checksums, and it is used
70 * by writes to insert metadata for new file extents after IO is complete.
71 */
72struct btrfs_end_io_wq {
73 struct bio *bio;
74 bio_end_io_t *end_io;
75 void *private;
76 struct btrfs_fs_info *info;
77 blk_status_t status;
78 enum btrfs_wq_endio_type metadata;
79 struct btrfs_work work;
80};
81
82static struct kmem_cache *btrfs_end_io_wq_cache;
83
84int __init btrfs_end_io_wq_init(void)
85{
86 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
87 sizeof(struct btrfs_end_io_wq),
88 0,
89 SLAB_MEM_SPREAD,
90 NULL);
91 if (!btrfs_end_io_wq_cache)
92 return -ENOMEM;
93 return 0;
94}
95
96void __cold btrfs_end_io_wq_exit(void)
97{
98 kmem_cache_destroy(btrfs_end_io_wq_cache);
99}
100
101static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
102{
103 if (fs_info->csum_shash)
104 crypto_free_shash(fs_info->csum_shash);
105}
106
107/*
108 * async submit bios are used to offload expensive checksumming
109 * onto the worker threads. They checksum file and metadata bios
110 * just before they are sent down the IO stack.
111 */
112struct async_submit_bio {
113 void *private_data;
114 struct bio *bio;
115 extent_submit_bio_start_t *submit_bio_start;
116 int mirror_num;
117 /*
118 * bio_offset is optional, can be used if the pages in the bio
119 * can't tell us where in the file the bio should go
120 */
121 u64 bio_offset;
122 struct btrfs_work work;
123 blk_status_t status;
124};
125
126/*
127 * Lockdep class keys for extent_buffer->lock's in this root. For a given
128 * eb, the lockdep key is determined by the btrfs_root it belongs to and
129 * the level the eb occupies in the tree.
130 *
131 * Different roots are used for different purposes and may nest inside each
132 * other and they require separate keysets. As lockdep keys should be
133 * static, assign keysets according to the purpose of the root as indicated
134 * by btrfs_root->root_key.objectid. This ensures that all special purpose
135 * roots have separate keysets.
136 *
137 * Lock-nesting across peer nodes is always done with the immediate parent
138 * node locked thus preventing deadlock. As lockdep doesn't know this, use
139 * subclass to avoid triggering lockdep warning in such cases.
140 *
141 * The key is set by the readpage_end_io_hook after the buffer has passed
142 * csum validation but before the pages are unlocked. It is also set by
143 * btrfs_init_new_buffer on freshly allocated blocks.
144 *
145 * We also add a check to make sure the highest level of the tree is the
146 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
147 * needs update as well.
148 */
149#ifdef CONFIG_DEBUG_LOCK_ALLOC
150# if BTRFS_MAX_LEVEL != 8
151# error
152# endif
153
154static struct btrfs_lockdep_keyset {
155 u64 id; /* root objectid */
156 const char *name_stem; /* lock name stem */
157 char names[BTRFS_MAX_LEVEL + 1][20];
158 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
159} btrfs_lockdep_keysets[] = {
160 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
161 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
162 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
163 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
164 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
165 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
166 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
167 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
168 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
169 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
170 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
171 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
172 { .id = 0, .name_stem = "tree" },
173};
174
175void __init btrfs_init_lockdep(void)
176{
177 int i, j;
178
179 /* initialize lockdep class names */
180 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
181 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
182
183 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
184 snprintf(ks->names[j], sizeof(ks->names[j]),
185 "btrfs-%s-%02d", ks->name_stem, j);
186 }
187}
188
189void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
190 int level)
191{
192 struct btrfs_lockdep_keyset *ks;
193
194 BUG_ON(level >= ARRAY_SIZE(ks->keys));
195
196 /* find the matching keyset, id 0 is the default entry */
197 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
198 if (ks->id == objectid)
199 break;
200
201 lockdep_set_class_and_name(&eb->lock,
202 &ks->keys[level], ks->names[level]);
203}
204
205#endif
206
207/*
208 * extents on the btree inode are pretty simple, there's one extent
209 * that covers the entire device
210 */
211struct extent_map *btree_get_extent(struct btrfs_inode *inode,
212 struct page *page, size_t pg_offset,
213 u64 start, u64 len)
214{
215 struct extent_map_tree *em_tree = &inode->extent_tree;
216 struct extent_map *em;
217 int ret;
218
219 read_lock(&em_tree->lock);
220 em = lookup_extent_mapping(em_tree, start, len);
221 if (em) {
222 read_unlock(&em_tree->lock);
223 goto out;
224 }
225 read_unlock(&em_tree->lock);
226
227 em = alloc_extent_map();
228 if (!em) {
229 em = ERR_PTR(-ENOMEM);
230 goto out;
231 }
232 em->start = 0;
233 em->len = (u64)-1;
234 em->block_len = (u64)-1;
235 em->block_start = 0;
236
237 write_lock(&em_tree->lock);
238 ret = add_extent_mapping(em_tree, em, 0);
239 if (ret == -EEXIST) {
240 free_extent_map(em);
241 em = lookup_extent_mapping(em_tree, start, len);
242 if (!em)
243 em = ERR_PTR(-EIO);
244 } else if (ret) {
245 free_extent_map(em);
246 em = ERR_PTR(ret);
247 }
248 write_unlock(&em_tree->lock);
249
250out:
251 return em;
252}
253
254/*
255 * Compute the csum of a btree block and store the result to provided buffer.
256 */
257static void csum_tree_block(struct extent_buffer *buf, u8 *result)
258{
259 struct btrfs_fs_info *fs_info = buf->fs_info;
260 const int num_pages = fs_info->nodesize >> PAGE_SHIFT;
261 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
262 char *kaddr;
263 int i;
264
265 shash->tfm = fs_info->csum_shash;
266 crypto_shash_init(shash);
267 kaddr = page_address(buf->pages[0]);
268 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
269 PAGE_SIZE - BTRFS_CSUM_SIZE);
270
271 for (i = 1; i < num_pages; i++) {
272 kaddr = page_address(buf->pages[i]);
273 crypto_shash_update(shash, kaddr, PAGE_SIZE);
274 }
275 memset(result, 0, BTRFS_CSUM_SIZE);
276 crypto_shash_final(shash, result);
277}
278
279/*
280 * we can't consider a given block up to date unless the transid of the
281 * block matches the transid in the parent node's pointer. This is how we
282 * detect blocks that either didn't get written at all or got written
283 * in the wrong place.
284 */
285static int verify_parent_transid(struct extent_io_tree *io_tree,
286 struct extent_buffer *eb, u64 parent_transid,
287 int atomic)
288{
289 struct extent_state *cached_state = NULL;
290 int ret;
291 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
292
293 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
294 return 0;
295
296 if (atomic)
297 return -EAGAIN;
298
299 if (need_lock) {
300 btrfs_tree_read_lock(eb);
301 btrfs_set_lock_blocking_read(eb);
302 }
303
304 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
305 &cached_state);
306 if (extent_buffer_uptodate(eb) &&
307 btrfs_header_generation(eb) == parent_transid) {
308 ret = 0;
309 goto out;
310 }
311 btrfs_err_rl(eb->fs_info,
312 "parent transid verify failed on %llu wanted %llu found %llu",
313 eb->start,
314 parent_transid, btrfs_header_generation(eb));
315 ret = 1;
316
317 /*
318 * Things reading via commit roots that don't have normal protection,
319 * like send, can have a really old block in cache that may point at a
320 * block that has been freed and re-allocated. So don't clear uptodate
321 * if we find an eb that is under IO (dirty/writeback) because we could
322 * end up reading in the stale data and then writing it back out and
323 * making everybody very sad.
324 */
325 if (!extent_buffer_under_io(eb))
326 clear_extent_buffer_uptodate(eb);
327out:
328 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
329 &cached_state);
330 if (need_lock)
331 btrfs_tree_read_unlock_blocking(eb);
332 return ret;
333}
334
335static bool btrfs_supported_super_csum(u16 csum_type)
336{
337 switch (csum_type) {
338 case BTRFS_CSUM_TYPE_CRC32:
339 case BTRFS_CSUM_TYPE_XXHASH:
340 case BTRFS_CSUM_TYPE_SHA256:
341 case BTRFS_CSUM_TYPE_BLAKE2:
342 return true;
343 default:
344 return false;
345 }
346}
347
348/*
349 * Return 0 if the superblock checksum type matches the checksum value of that
350 * algorithm. Pass the raw disk superblock data.
351 */
352static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
353 char *raw_disk_sb)
354{
355 struct btrfs_super_block *disk_sb =
356 (struct btrfs_super_block *)raw_disk_sb;
357 char result[BTRFS_CSUM_SIZE];
358 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
359
360 shash->tfm = fs_info->csum_shash;
361
362 /*
363 * The super_block structure does not span the whole
364 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
365 * filled with zeros and is included in the checksum.
366 */
367 crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
368 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
369
370 if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb)))
371 return 1;
372
373 return 0;
374}
375
376int btrfs_verify_level_key(struct extent_buffer *eb, int level,
377 struct btrfs_key *first_key, u64 parent_transid)
378{
379 struct btrfs_fs_info *fs_info = eb->fs_info;
380 int found_level;
381 struct btrfs_key found_key;
382 int ret;
383
384 found_level = btrfs_header_level(eb);
385 if (found_level != level) {
386 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
387 KERN_ERR "BTRFS: tree level check failed\n");
388 btrfs_err(fs_info,
389"tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
390 eb->start, level, found_level);
391 return -EIO;
392 }
393
394 if (!first_key)
395 return 0;
396
397 /*
398 * For live tree block (new tree blocks in current transaction),
399 * we need proper lock context to avoid race, which is impossible here.
400 * So we only checks tree blocks which is read from disk, whose
401 * generation <= fs_info->last_trans_committed.
402 */
403 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
404 return 0;
405
406 /* We have @first_key, so this @eb must have at least one item */
407 if (btrfs_header_nritems(eb) == 0) {
408 btrfs_err(fs_info,
409 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
410 eb->start);
411 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
412 return -EUCLEAN;
413 }
414
415 if (found_level)
416 btrfs_node_key_to_cpu(eb, &found_key, 0);
417 else
418 btrfs_item_key_to_cpu(eb, &found_key, 0);
419 ret = btrfs_comp_cpu_keys(first_key, &found_key);
420
421 if (ret) {
422 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
423 KERN_ERR "BTRFS: tree first key check failed\n");
424 btrfs_err(fs_info,
425"tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
426 eb->start, parent_transid, first_key->objectid,
427 first_key->type, first_key->offset,
428 found_key.objectid, found_key.type,
429 found_key.offset);
430 }
431 return ret;
432}
433
434/*
435 * helper to read a given tree block, doing retries as required when
436 * the checksums don't match and we have alternate mirrors to try.
437 *
438 * @parent_transid: expected transid, skip check if 0
439 * @level: expected level, mandatory check
440 * @first_key: expected key of first slot, skip check if NULL
441 */
442static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
443 u64 parent_transid, int level,
444 struct btrfs_key *first_key)
445{
446 struct btrfs_fs_info *fs_info = eb->fs_info;
447 struct extent_io_tree *io_tree;
448 int failed = 0;
449 int ret;
450 int num_copies = 0;
451 int mirror_num = 0;
452 int failed_mirror = 0;
453
454 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
455 while (1) {
456 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
457 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
458 if (!ret) {
459 if (verify_parent_transid(io_tree, eb,
460 parent_transid, 0))
461 ret = -EIO;
462 else if (btrfs_verify_level_key(eb, level,
463 first_key, parent_transid))
464 ret = -EUCLEAN;
465 else
466 break;
467 }
468
469 num_copies = btrfs_num_copies(fs_info,
470 eb->start, eb->len);
471 if (num_copies == 1)
472 break;
473
474 if (!failed_mirror) {
475 failed = 1;
476 failed_mirror = eb->read_mirror;
477 }
478
479 mirror_num++;
480 if (mirror_num == failed_mirror)
481 mirror_num++;
482
483 if (mirror_num > num_copies)
484 break;
485 }
486
487 if (failed && !ret && failed_mirror)
488 btrfs_repair_eb_io_failure(eb, failed_mirror);
489
490 return ret;
491}
492
493/*
494 * checksum a dirty tree block before IO. This has extra checks to make sure
495 * we only fill in the checksum field in the first page of a multi-page block
496 */
497
498static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
499{
500 u64 start = page_offset(page);
501 u64 found_start;
502 u8 result[BTRFS_CSUM_SIZE];
503 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
504 struct extent_buffer *eb;
505 int ret;
506
507 eb = (struct extent_buffer *)page->private;
508 if (page != eb->pages[0])
509 return 0;
510
511 found_start = btrfs_header_bytenr(eb);
512 /*
513 * Please do not consolidate these warnings into a single if.
514 * It is useful to know what went wrong.
515 */
516 if (WARN_ON(found_start != start))
517 return -EUCLEAN;
518 if (WARN_ON(!PageUptodate(page)))
519 return -EUCLEAN;
520
521 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
522 offsetof(struct btrfs_header, fsid),
523 BTRFS_FSID_SIZE) == 0);
524
525 csum_tree_block(eb, result);
526
527 if (btrfs_header_level(eb))
528 ret = btrfs_check_node(eb);
529 else
530 ret = btrfs_check_leaf_full(eb);
531
532 if (ret < 0) {
533 btrfs_print_tree(eb, 0);
534 btrfs_err(fs_info,
535 "block=%llu write time tree block corruption detected",
536 eb->start);
537 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
538 return ret;
539 }
540 write_extent_buffer(eb, result, 0, csum_size);
541
542 return 0;
543}
544
545static int check_tree_block_fsid(struct extent_buffer *eb)
546{
547 struct btrfs_fs_info *fs_info = eb->fs_info;
548 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
549 u8 fsid[BTRFS_FSID_SIZE];
550 int ret = 1;
551
552 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
553 BTRFS_FSID_SIZE);
554 while (fs_devices) {
555 u8 *metadata_uuid;
556
557 /*
558 * Checking the incompat flag is only valid for the current
559 * fs. For seed devices it's forbidden to have their uuid
560 * changed so reading ->fsid in this case is fine
561 */
562 if (fs_devices == fs_info->fs_devices &&
563 btrfs_fs_incompat(fs_info, METADATA_UUID))
564 metadata_uuid = fs_devices->metadata_uuid;
565 else
566 metadata_uuid = fs_devices->fsid;
567
568 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
569 ret = 0;
570 break;
571 }
572 fs_devices = fs_devices->seed;
573 }
574 return ret;
575}
576
577static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
578 u64 phy_offset, struct page *page,
579 u64 start, u64 end, int mirror)
580{
581 u64 found_start;
582 int found_level;
583 struct extent_buffer *eb;
584 struct btrfs_fs_info *fs_info;
585 u16 csum_size;
586 int ret = 0;
587 u8 result[BTRFS_CSUM_SIZE];
588 int reads_done;
589
590 if (!page->private)
591 goto out;
592
593 eb = (struct extent_buffer *)page->private;
594 fs_info = eb->fs_info;
595 csum_size = btrfs_super_csum_size(fs_info->super_copy);
596
597 /* the pending IO might have been the only thing that kept this buffer
598 * in memory. Make sure we have a ref for all this other checks
599 */
600 atomic_inc(&eb->refs);
601
602 reads_done = atomic_dec_and_test(&eb->io_pages);
603 if (!reads_done)
604 goto err;
605
606 eb->read_mirror = mirror;
607 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
608 ret = -EIO;
609 goto err;
610 }
611
612 found_start = btrfs_header_bytenr(eb);
613 if (found_start != eb->start) {
614 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
615 eb->start, found_start);
616 ret = -EIO;
617 goto err;
618 }
619 if (check_tree_block_fsid(eb)) {
620 btrfs_err_rl(fs_info, "bad fsid on block %llu",
621 eb->start);
622 ret = -EIO;
623 goto err;
624 }
625 found_level = btrfs_header_level(eb);
626 if (found_level >= BTRFS_MAX_LEVEL) {
627 btrfs_err(fs_info, "bad tree block level %d on %llu",
628 (int)btrfs_header_level(eb), eb->start);
629 ret = -EIO;
630 goto err;
631 }
632
633 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
634 eb, found_level);
635
636 csum_tree_block(eb, result);
637
638 if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
639 u8 val[BTRFS_CSUM_SIZE] = { 0 };
640
641 read_extent_buffer(eb, &val, 0, csum_size);
642 btrfs_warn_rl(fs_info,
643 "%s checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d",
644 fs_info->sb->s_id, eb->start,
645 CSUM_FMT_VALUE(csum_size, val),
646 CSUM_FMT_VALUE(csum_size, result),
647 btrfs_header_level(eb));
648 ret = -EUCLEAN;
649 goto err;
650 }
651
652 /*
653 * If this is a leaf block and it is corrupt, set the corrupt bit so
654 * that we don't try and read the other copies of this block, just
655 * return -EIO.
656 */
657 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
658 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
659 ret = -EIO;
660 }
661
662 if (found_level > 0 && btrfs_check_node(eb))
663 ret = -EIO;
664
665 if (!ret)
666 set_extent_buffer_uptodate(eb);
667 else
668 btrfs_err(fs_info,
669 "block=%llu read time tree block corruption detected",
670 eb->start);
671err:
672 if (reads_done &&
673 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
674 btree_readahead_hook(eb, ret);
675
676 if (ret) {
677 /*
678 * our io error hook is going to dec the io pages
679 * again, we have to make sure it has something
680 * to decrement
681 */
682 atomic_inc(&eb->io_pages);
683 clear_extent_buffer_uptodate(eb);
684 }
685 free_extent_buffer(eb);
686out:
687 return ret;
688}
689
690static void end_workqueue_bio(struct bio *bio)
691{
692 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
693 struct btrfs_fs_info *fs_info;
694 struct btrfs_workqueue *wq;
695
696 fs_info = end_io_wq->info;
697 end_io_wq->status = bio->bi_status;
698
699 if (bio_op(bio) == REQ_OP_WRITE) {
700 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
701 wq = fs_info->endio_meta_write_workers;
702 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
703 wq = fs_info->endio_freespace_worker;
704 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
705 wq = fs_info->endio_raid56_workers;
706 else
707 wq = fs_info->endio_write_workers;
708 } else {
709 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
710 wq = fs_info->endio_raid56_workers;
711 else if (end_io_wq->metadata)
712 wq = fs_info->endio_meta_workers;
713 else
714 wq = fs_info->endio_workers;
715 }
716
717 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
718 btrfs_queue_work(wq, &end_io_wq->work);
719}
720
721blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
722 enum btrfs_wq_endio_type metadata)
723{
724 struct btrfs_end_io_wq *end_io_wq;
725
726 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
727 if (!end_io_wq)
728 return BLK_STS_RESOURCE;
729
730 end_io_wq->private = bio->bi_private;
731 end_io_wq->end_io = bio->bi_end_io;
732 end_io_wq->info = info;
733 end_io_wq->status = 0;
734 end_io_wq->bio = bio;
735 end_io_wq->metadata = metadata;
736
737 bio->bi_private = end_io_wq;
738 bio->bi_end_io = end_workqueue_bio;
739 return 0;
740}
741
742static void run_one_async_start(struct btrfs_work *work)
743{
744 struct async_submit_bio *async;
745 blk_status_t ret;
746
747 async = container_of(work, struct async_submit_bio, work);
748 ret = async->submit_bio_start(async->private_data, async->bio,
749 async->bio_offset);
750 if (ret)
751 async->status = ret;
752}
753
754/*
755 * In order to insert checksums into the metadata in large chunks, we wait
756 * until bio submission time. All the pages in the bio are checksummed and
757 * sums are attached onto the ordered extent record.
758 *
759 * At IO completion time the csums attached on the ordered extent record are
760 * inserted into the tree.
761 */
762static void run_one_async_done(struct btrfs_work *work)
763{
764 struct async_submit_bio *async;
765 struct inode *inode;
766 blk_status_t ret;
767
768 async = container_of(work, struct async_submit_bio, work);
769 inode = async->private_data;
770
771 /* If an error occurred we just want to clean up the bio and move on */
772 if (async->status) {
773 async->bio->bi_status = async->status;
774 bio_endio(async->bio);
775 return;
776 }
777
778 /*
779 * All of the bios that pass through here are from async helpers.
780 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
781 * This changes nothing when cgroups aren't in use.
782 */
783 async->bio->bi_opf |= REQ_CGROUP_PUNT;
784 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
785 if (ret) {
786 async->bio->bi_status = ret;
787 bio_endio(async->bio);
788 }
789}
790
791static void run_one_async_free(struct btrfs_work *work)
792{
793 struct async_submit_bio *async;
794
795 async = container_of(work, struct async_submit_bio, work);
796 kfree(async);
797}
798
799blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
800 int mirror_num, unsigned long bio_flags,
801 u64 bio_offset, void *private_data,
802 extent_submit_bio_start_t *submit_bio_start)
803{
804 struct async_submit_bio *async;
805
806 async = kmalloc(sizeof(*async), GFP_NOFS);
807 if (!async)
808 return BLK_STS_RESOURCE;
809
810 async->private_data = private_data;
811 async->bio = bio;
812 async->mirror_num = mirror_num;
813 async->submit_bio_start = submit_bio_start;
814
815 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
816 run_one_async_free);
817
818 async->bio_offset = bio_offset;
819
820 async->status = 0;
821
822 if (op_is_sync(bio->bi_opf))
823 btrfs_set_work_high_priority(&async->work);
824
825 btrfs_queue_work(fs_info->workers, &async->work);
826 return 0;
827}
828
829static blk_status_t btree_csum_one_bio(struct bio *bio)
830{
831 struct bio_vec *bvec;
832 struct btrfs_root *root;
833 int ret = 0;
834 struct bvec_iter_all iter_all;
835
836 ASSERT(!bio_flagged(bio, BIO_CLONED));
837 bio_for_each_segment_all(bvec, bio, iter_all) {
838 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
839 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
840 if (ret)
841 break;
842 }
843
844 return errno_to_blk_status(ret);
845}
846
847static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
848 u64 bio_offset)
849{
850 /*
851 * when we're called for a write, we're already in the async
852 * submission context. Just jump into btrfs_map_bio
853 */
854 return btree_csum_one_bio(bio);
855}
856
857static int check_async_write(struct btrfs_fs_info *fs_info,
858 struct btrfs_inode *bi)
859{
860 if (atomic_read(&bi->sync_writers))
861 return 0;
862 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
863 return 0;
864 return 1;
865}
866
867static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
868 int mirror_num,
869 unsigned long bio_flags)
870{
871 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
872 int async = check_async_write(fs_info, BTRFS_I(inode));
873 blk_status_t ret;
874
875 if (bio_op(bio) != REQ_OP_WRITE) {
876 /*
877 * called for a read, do the setup so that checksum validation
878 * can happen in the async kernel threads
879 */
880 ret = btrfs_bio_wq_end_io(fs_info, bio,
881 BTRFS_WQ_ENDIO_METADATA);
882 if (ret)
883 goto out_w_error;
884 ret = btrfs_map_bio(fs_info, bio, mirror_num);
885 } else if (!async) {
886 ret = btree_csum_one_bio(bio);
887 if (ret)
888 goto out_w_error;
889 ret = btrfs_map_bio(fs_info, bio, mirror_num);
890 } else {
891 /*
892 * kthread helpers are used to submit writes so that
893 * checksumming can happen in parallel across all CPUs
894 */
895 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
896 0, inode, btree_submit_bio_start);
897 }
898
899 if (ret)
900 goto out_w_error;
901 return 0;
902
903out_w_error:
904 bio->bi_status = ret;
905 bio_endio(bio);
906 return ret;
907}
908
909#ifdef CONFIG_MIGRATION
910static int btree_migratepage(struct address_space *mapping,
911 struct page *newpage, struct page *page,
912 enum migrate_mode mode)
913{
914 /*
915 * we can't safely write a btree page from here,
916 * we haven't done the locking hook
917 */
918 if (PageDirty(page))
919 return -EAGAIN;
920 /*
921 * Buffers may be managed in a filesystem specific way.
922 * We must have no buffers or drop them.
923 */
924 if (page_has_private(page) &&
925 !try_to_release_page(page, GFP_KERNEL))
926 return -EAGAIN;
927 return migrate_page(mapping, newpage, page, mode);
928}
929#endif
930
931
932static int btree_writepages(struct address_space *mapping,
933 struct writeback_control *wbc)
934{
935 struct btrfs_fs_info *fs_info;
936 int ret;
937
938 if (wbc->sync_mode == WB_SYNC_NONE) {
939
940 if (wbc->for_kupdate)
941 return 0;
942
943 fs_info = BTRFS_I(mapping->host)->root->fs_info;
944 /* this is a bit racy, but that's ok */
945 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
946 BTRFS_DIRTY_METADATA_THRESH,
947 fs_info->dirty_metadata_batch);
948 if (ret < 0)
949 return 0;
950 }
951 return btree_write_cache_pages(mapping, wbc);
952}
953
954static int btree_readpage(struct file *file, struct page *page)
955{
956 return extent_read_full_page(page, btree_get_extent, 0);
957}
958
959static int btree_releasepage(struct page *page, gfp_t gfp_flags)
960{
961 if (PageWriteback(page) || PageDirty(page))
962 return 0;
963
964 return try_release_extent_buffer(page);
965}
966
967static void btree_invalidatepage(struct page *page, unsigned int offset,
968 unsigned int length)
969{
970 struct extent_io_tree *tree;
971 tree = &BTRFS_I(page->mapping->host)->io_tree;
972 extent_invalidatepage(tree, page, offset);
973 btree_releasepage(page, GFP_NOFS);
974 if (PagePrivate(page)) {
975 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
976 "page private not zero on page %llu",
977 (unsigned long long)page_offset(page));
978 detach_page_private(page);
979 }
980}
981
982static int btree_set_page_dirty(struct page *page)
983{
984#ifdef DEBUG
985 struct extent_buffer *eb;
986
987 BUG_ON(!PagePrivate(page));
988 eb = (struct extent_buffer *)page->private;
989 BUG_ON(!eb);
990 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
991 BUG_ON(!atomic_read(&eb->refs));
992 btrfs_assert_tree_locked(eb);
993#endif
994 return __set_page_dirty_nobuffers(page);
995}
996
997static const struct address_space_operations btree_aops = {
998 .readpage = btree_readpage,
999 .writepages = btree_writepages,
1000 .releasepage = btree_releasepage,
1001 .invalidatepage = btree_invalidatepage,
1002#ifdef CONFIG_MIGRATION
1003 .migratepage = btree_migratepage,
1004#endif
1005 .set_page_dirty = btree_set_page_dirty,
1006};
1007
1008void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1009{
1010 struct extent_buffer *buf = NULL;
1011 int ret;
1012
1013 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1014 if (IS_ERR(buf))
1015 return;
1016
1017 ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1018 if (ret < 0)
1019 free_extent_buffer_stale(buf);
1020 else
1021 free_extent_buffer(buf);
1022}
1023
1024struct extent_buffer *btrfs_find_create_tree_block(
1025 struct btrfs_fs_info *fs_info,
1026 u64 bytenr)
1027{
1028 if (btrfs_is_testing(fs_info))
1029 return alloc_test_extent_buffer(fs_info, bytenr);
1030 return alloc_extent_buffer(fs_info, bytenr);
1031}
1032
1033/*
1034 * Read tree block at logical address @bytenr and do variant basic but critical
1035 * verification.
1036 *
1037 * @parent_transid: expected transid of this tree block, skip check if 0
1038 * @level: expected level, mandatory check
1039 * @first_key: expected key in slot 0, skip check if NULL
1040 */
1041struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1042 u64 parent_transid, int level,
1043 struct btrfs_key *first_key)
1044{
1045 struct extent_buffer *buf = NULL;
1046 int ret;
1047
1048 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1049 if (IS_ERR(buf))
1050 return buf;
1051
1052 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1053 level, first_key);
1054 if (ret) {
1055 free_extent_buffer_stale(buf);
1056 return ERR_PTR(ret);
1057 }
1058 return buf;
1059
1060}
1061
1062void btrfs_clean_tree_block(struct extent_buffer *buf)
1063{
1064 struct btrfs_fs_info *fs_info = buf->fs_info;
1065 if (btrfs_header_generation(buf) ==
1066 fs_info->running_transaction->transid) {
1067 btrfs_assert_tree_locked(buf);
1068
1069 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1070 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1071 -buf->len,
1072 fs_info->dirty_metadata_batch);
1073 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1074 btrfs_set_lock_blocking_write(buf);
1075 clear_extent_buffer_dirty(buf);
1076 }
1077 }
1078}
1079
1080static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1081 u64 objectid)
1082{
1083 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1084 root->fs_info = fs_info;
1085 root->node = NULL;
1086 root->commit_root = NULL;
1087 root->state = 0;
1088 root->orphan_cleanup_state = 0;
1089
1090 root->last_trans = 0;
1091 root->highest_objectid = 0;
1092 root->nr_delalloc_inodes = 0;
1093 root->nr_ordered_extents = 0;
1094 root->inode_tree = RB_ROOT;
1095 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1096 root->block_rsv = NULL;
1097
1098 INIT_LIST_HEAD(&root->dirty_list);
1099 INIT_LIST_HEAD(&root->root_list);
1100 INIT_LIST_HEAD(&root->delalloc_inodes);
1101 INIT_LIST_HEAD(&root->delalloc_root);
1102 INIT_LIST_HEAD(&root->ordered_extents);
1103 INIT_LIST_HEAD(&root->ordered_root);
1104 INIT_LIST_HEAD(&root->reloc_dirty_list);
1105 INIT_LIST_HEAD(&root->logged_list[0]);
1106 INIT_LIST_HEAD(&root->logged_list[1]);
1107 spin_lock_init(&root->inode_lock);
1108 spin_lock_init(&root->delalloc_lock);
1109 spin_lock_init(&root->ordered_extent_lock);
1110 spin_lock_init(&root->accounting_lock);
1111 spin_lock_init(&root->log_extents_lock[0]);
1112 spin_lock_init(&root->log_extents_lock[1]);
1113 spin_lock_init(&root->qgroup_meta_rsv_lock);
1114 mutex_init(&root->objectid_mutex);
1115 mutex_init(&root->log_mutex);
1116 mutex_init(&root->ordered_extent_mutex);
1117 mutex_init(&root->delalloc_mutex);
1118 init_waitqueue_head(&root->qgroup_flush_wait);
1119 init_waitqueue_head(&root->log_writer_wait);
1120 init_waitqueue_head(&root->log_commit_wait[0]);
1121 init_waitqueue_head(&root->log_commit_wait[1]);
1122 INIT_LIST_HEAD(&root->log_ctxs[0]);
1123 INIT_LIST_HEAD(&root->log_ctxs[1]);
1124 atomic_set(&root->log_commit[0], 0);
1125 atomic_set(&root->log_commit[1], 0);
1126 atomic_set(&root->log_writers, 0);
1127 atomic_set(&root->log_batch, 0);
1128 refcount_set(&root->refs, 1);
1129 atomic_set(&root->snapshot_force_cow, 0);
1130 atomic_set(&root->nr_swapfiles, 0);
1131 root->log_transid = 0;
1132 root->log_transid_committed = -1;
1133 root->last_log_commit = 0;
1134 if (!dummy) {
1135 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1136 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1137 extent_io_tree_init(fs_info, &root->log_csum_range,
1138 IO_TREE_LOG_CSUM_RANGE, NULL);
1139 }
1140
1141 memset(&root->root_key, 0, sizeof(root->root_key));
1142 memset(&root->root_item, 0, sizeof(root->root_item));
1143 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1144 root->root_key.objectid = objectid;
1145 root->anon_dev = 0;
1146
1147 spin_lock_init(&root->root_item_lock);
1148 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1149#ifdef CONFIG_BTRFS_DEBUG
1150 INIT_LIST_HEAD(&root->leak_list);
1151 spin_lock(&fs_info->fs_roots_radix_lock);
1152 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1153 spin_unlock(&fs_info->fs_roots_radix_lock);
1154#endif
1155}
1156
1157static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1158 u64 objectid, gfp_t flags)
1159{
1160 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1161 if (root)
1162 __setup_root(root, fs_info, objectid);
1163 return root;
1164}
1165
1166#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1167/* Should only be used by the testing infrastructure */
1168struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1169{
1170 struct btrfs_root *root;
1171
1172 if (!fs_info)
1173 return ERR_PTR(-EINVAL);
1174
1175 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1176 if (!root)
1177 return ERR_PTR(-ENOMEM);
1178
1179 /* We don't use the stripesize in selftest, set it as sectorsize */
1180 root->alloc_bytenr = 0;
1181
1182 return root;
1183}
1184#endif
1185
1186struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1187 u64 objectid)
1188{
1189 struct btrfs_fs_info *fs_info = trans->fs_info;
1190 struct extent_buffer *leaf;
1191 struct btrfs_root *tree_root = fs_info->tree_root;
1192 struct btrfs_root *root;
1193 struct btrfs_key key;
1194 unsigned int nofs_flag;
1195 int ret = 0;
1196
1197 /*
1198 * We're holding a transaction handle, so use a NOFS memory allocation
1199 * context to avoid deadlock if reclaim happens.
1200 */
1201 nofs_flag = memalloc_nofs_save();
1202 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1203 memalloc_nofs_restore(nofs_flag);
1204 if (!root)
1205 return ERR_PTR(-ENOMEM);
1206
1207 root->root_key.objectid = objectid;
1208 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1209 root->root_key.offset = 0;
1210
1211 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1212 if (IS_ERR(leaf)) {
1213 ret = PTR_ERR(leaf);
1214 leaf = NULL;
1215 goto fail;
1216 }
1217
1218 root->node = leaf;
1219 btrfs_mark_buffer_dirty(leaf);
1220
1221 root->commit_root = btrfs_root_node(root);
1222 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1223
1224 root->root_item.flags = 0;
1225 root->root_item.byte_limit = 0;
1226 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1227 btrfs_set_root_generation(&root->root_item, trans->transid);
1228 btrfs_set_root_level(&root->root_item, 0);
1229 btrfs_set_root_refs(&root->root_item, 1);
1230 btrfs_set_root_used(&root->root_item, leaf->len);
1231 btrfs_set_root_last_snapshot(&root->root_item, 0);
1232 btrfs_set_root_dirid(&root->root_item, 0);
1233 if (is_fstree(objectid))
1234 generate_random_guid(root->root_item.uuid);
1235 else
1236 export_guid(root->root_item.uuid, &guid_null);
1237 root->root_item.drop_level = 0;
1238
1239 key.objectid = objectid;
1240 key.type = BTRFS_ROOT_ITEM_KEY;
1241 key.offset = 0;
1242 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1243 if (ret)
1244 goto fail;
1245
1246 btrfs_tree_unlock(leaf);
1247
1248 return root;
1249
1250fail:
1251 if (leaf)
1252 btrfs_tree_unlock(leaf);
1253 btrfs_put_root(root);
1254
1255 return ERR_PTR(ret);
1256}
1257
1258static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1259 struct btrfs_fs_info *fs_info)
1260{
1261 struct btrfs_root *root;
1262 struct extent_buffer *leaf;
1263
1264 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1265 if (!root)
1266 return ERR_PTR(-ENOMEM);
1267
1268 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1269 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1270 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1271
1272 /*
1273 * DON'T set SHAREABLE bit for log trees.
1274 *
1275 * Log trees are not exposed to user space thus can't be snapshotted,
1276 * and they go away before a real commit is actually done.
1277 *
1278 * They do store pointers to file data extents, and those reference
1279 * counts still get updated (along with back refs to the log tree).
1280 */
1281
1282 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1283 NULL, 0, 0, 0);
1284 if (IS_ERR(leaf)) {
1285 btrfs_put_root(root);
1286 return ERR_CAST(leaf);
1287 }
1288
1289 root->node = leaf;
1290
1291 btrfs_mark_buffer_dirty(root->node);
1292 btrfs_tree_unlock(root->node);
1293 return root;
1294}
1295
1296int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1297 struct btrfs_fs_info *fs_info)
1298{
1299 struct btrfs_root *log_root;
1300
1301 log_root = alloc_log_tree(trans, fs_info);
1302 if (IS_ERR(log_root))
1303 return PTR_ERR(log_root);
1304 WARN_ON(fs_info->log_root_tree);
1305 fs_info->log_root_tree = log_root;
1306 return 0;
1307}
1308
1309int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1310 struct btrfs_root *root)
1311{
1312 struct btrfs_fs_info *fs_info = root->fs_info;
1313 struct btrfs_root *log_root;
1314 struct btrfs_inode_item *inode_item;
1315
1316 log_root = alloc_log_tree(trans, fs_info);
1317 if (IS_ERR(log_root))
1318 return PTR_ERR(log_root);
1319
1320 log_root->last_trans = trans->transid;
1321 log_root->root_key.offset = root->root_key.objectid;
1322
1323 inode_item = &log_root->root_item.inode;
1324 btrfs_set_stack_inode_generation(inode_item, 1);
1325 btrfs_set_stack_inode_size(inode_item, 3);
1326 btrfs_set_stack_inode_nlink(inode_item, 1);
1327 btrfs_set_stack_inode_nbytes(inode_item,
1328 fs_info->nodesize);
1329 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1330
1331 btrfs_set_root_node(&log_root->root_item, log_root->node);
1332
1333 WARN_ON(root->log_root);
1334 root->log_root = log_root;
1335 root->log_transid = 0;
1336 root->log_transid_committed = -1;
1337 root->last_log_commit = 0;
1338 return 0;
1339}
1340
1341struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1342 struct btrfs_key *key)
1343{
1344 struct btrfs_root *root;
1345 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1346 struct btrfs_path *path;
1347 u64 generation;
1348 int ret;
1349 int level;
1350
1351 path = btrfs_alloc_path();
1352 if (!path)
1353 return ERR_PTR(-ENOMEM);
1354
1355 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1356 if (!root) {
1357 ret = -ENOMEM;
1358 goto alloc_fail;
1359 }
1360
1361 ret = btrfs_find_root(tree_root, key, path,
1362 &root->root_item, &root->root_key);
1363 if (ret) {
1364 if (ret > 0)
1365 ret = -ENOENT;
1366 goto find_fail;
1367 }
1368
1369 generation = btrfs_root_generation(&root->root_item);
1370 level = btrfs_root_level(&root->root_item);
1371 root->node = read_tree_block(fs_info,
1372 btrfs_root_bytenr(&root->root_item),
1373 generation, level, NULL);
1374 if (IS_ERR(root->node)) {
1375 ret = PTR_ERR(root->node);
1376 root->node = NULL;
1377 goto find_fail;
1378 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1379 ret = -EIO;
1380 goto find_fail;
1381 }
1382 root->commit_root = btrfs_root_node(root);
1383out:
1384 btrfs_free_path(path);
1385 return root;
1386
1387find_fail:
1388 btrfs_put_root(root);
1389alloc_fail:
1390 root = ERR_PTR(ret);
1391 goto out;
1392}
1393
1394/*
1395 * Initialize subvolume root in-memory structure
1396 *
1397 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1398 */
1399static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1400{
1401 int ret;
1402 unsigned int nofs_flag;
1403
1404 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1405 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1406 GFP_NOFS);
1407 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1408 ret = -ENOMEM;
1409 goto fail;
1410 }
1411
1412 /*
1413 * We might be called under a transaction (e.g. indirect backref
1414 * resolution) which could deadlock if it triggers memory reclaim
1415 */
1416 nofs_flag = memalloc_nofs_save();
1417 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1418 memalloc_nofs_restore(nofs_flag);
1419 if (ret)
1420 goto fail;
1421
1422 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1423 root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
1424 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1425 btrfs_check_and_init_root_item(&root->root_item);
1426 }
1427
1428 btrfs_init_free_ino_ctl(root);
1429 spin_lock_init(&root->ino_cache_lock);
1430 init_waitqueue_head(&root->ino_cache_wait);
1431
1432 /*
1433 * Don't assign anonymous block device to roots that are not exposed to
1434 * userspace, the id pool is limited to 1M
1435 */
1436 if (is_fstree(root->root_key.objectid) &&
1437 btrfs_root_refs(&root->root_item) > 0) {
1438 if (!anon_dev) {
1439 ret = get_anon_bdev(&root->anon_dev);
1440 if (ret)
1441 goto fail;
1442 } else {
1443 root->anon_dev = anon_dev;
1444 }
1445 }
1446
1447 mutex_lock(&root->objectid_mutex);
1448 ret = btrfs_find_highest_objectid(root,
1449 &root->highest_objectid);
1450 if (ret) {
1451 mutex_unlock(&root->objectid_mutex);
1452 goto fail;
1453 }
1454
1455 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1456
1457 mutex_unlock(&root->objectid_mutex);
1458
1459 return 0;
1460fail:
1461 /* The caller is responsible to call btrfs_free_fs_root */
1462 return ret;
1463}
1464
1465static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1466 u64 root_id)
1467{
1468 struct btrfs_root *root;
1469
1470 spin_lock(&fs_info->fs_roots_radix_lock);
1471 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1472 (unsigned long)root_id);
1473 if (root)
1474 root = btrfs_grab_root(root);
1475 spin_unlock(&fs_info->fs_roots_radix_lock);
1476 return root;
1477}
1478
1479int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1480 struct btrfs_root *root)
1481{
1482 int ret;
1483
1484 ret = radix_tree_preload(GFP_NOFS);
1485 if (ret)
1486 return ret;
1487
1488 spin_lock(&fs_info->fs_roots_radix_lock);
1489 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1490 (unsigned long)root->root_key.objectid,
1491 root);
1492 if (ret == 0) {
1493 btrfs_grab_root(root);
1494 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1495 }
1496 spin_unlock(&fs_info->fs_roots_radix_lock);
1497 radix_tree_preload_end();
1498
1499 return ret;
1500}
1501
1502void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1503{
1504#ifdef CONFIG_BTRFS_DEBUG
1505 struct btrfs_root *root;
1506
1507 while (!list_empty(&fs_info->allocated_roots)) {
1508 root = list_first_entry(&fs_info->allocated_roots,
1509 struct btrfs_root, leak_list);
1510 btrfs_err(fs_info, "leaked root %llu-%llu refcount %d",
1511 root->root_key.objectid, root->root_key.offset,
1512 refcount_read(&root->refs));
1513 while (refcount_read(&root->refs) > 1)
1514 btrfs_put_root(root);
1515 btrfs_put_root(root);
1516 }
1517#endif
1518}
1519
1520void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1521{
1522 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1523 percpu_counter_destroy(&fs_info->delalloc_bytes);
1524 percpu_counter_destroy(&fs_info->dio_bytes);
1525 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1526 btrfs_free_csum_hash(fs_info);
1527 btrfs_free_stripe_hash_table(fs_info);
1528 btrfs_free_ref_cache(fs_info);
1529 kfree(fs_info->balance_ctl);
1530 kfree(fs_info->delayed_root);
1531 btrfs_put_root(fs_info->extent_root);
1532 btrfs_put_root(fs_info->tree_root);
1533 btrfs_put_root(fs_info->chunk_root);
1534 btrfs_put_root(fs_info->dev_root);
1535 btrfs_put_root(fs_info->csum_root);
1536 btrfs_put_root(fs_info->quota_root);
1537 btrfs_put_root(fs_info->uuid_root);
1538 btrfs_put_root(fs_info->free_space_root);
1539 btrfs_put_root(fs_info->fs_root);
1540 btrfs_put_root(fs_info->data_reloc_root);
1541 btrfs_check_leaked_roots(fs_info);
1542 btrfs_extent_buffer_leak_debug_check(fs_info);
1543 kfree(fs_info->super_copy);
1544 kfree(fs_info->super_for_commit);
1545 kvfree(fs_info);
1546}
1547
1548
1549/*
1550 * Get an in-memory reference of a root structure.
1551 *
1552 * For essential trees like root/extent tree, we grab it from fs_info directly.
1553 * For subvolume trees, we check the cached filesystem roots first. If not
1554 * found, then read it from disk and add it to cached fs roots.
1555 *
1556 * Caller should release the root by calling btrfs_put_root() after the usage.
1557 *
1558 * NOTE: Reloc and log trees can't be read by this function as they share the
1559 * same root objectid.
1560 *
1561 * @objectid: root id
1562 * @anon_dev: preallocated anonymous block device number for new roots,
1563 * pass 0 for new allocation.
1564 * @check_ref: whether to check root item references, If true, return -ENOENT
1565 * for orphan roots
1566 */
1567static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1568 u64 objectid, dev_t anon_dev,
1569 bool check_ref)
1570{
1571 struct btrfs_root *root;
1572 struct btrfs_path *path;
1573 struct btrfs_key key;
1574 int ret;
1575
1576 if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1577 return btrfs_grab_root(fs_info->tree_root);
1578 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1579 return btrfs_grab_root(fs_info->extent_root);
1580 if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1581 return btrfs_grab_root(fs_info->chunk_root);
1582 if (objectid == BTRFS_DEV_TREE_OBJECTID)
1583 return btrfs_grab_root(fs_info->dev_root);
1584 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1585 return btrfs_grab_root(fs_info->csum_root);
1586 if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1587 return btrfs_grab_root(fs_info->quota_root) ?
1588 fs_info->quota_root : ERR_PTR(-ENOENT);
1589 if (objectid == BTRFS_UUID_TREE_OBJECTID)
1590 return btrfs_grab_root(fs_info->uuid_root) ?
1591 fs_info->uuid_root : ERR_PTR(-ENOENT);
1592 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1593 return btrfs_grab_root(fs_info->free_space_root) ?
1594 fs_info->free_space_root : ERR_PTR(-ENOENT);
1595again:
1596 root = btrfs_lookup_fs_root(fs_info, objectid);
1597 if (root) {
1598 /* Shouldn't get preallocated anon_dev for cached roots */
1599 ASSERT(!anon_dev);
1600 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1601 btrfs_put_root(root);
1602 return ERR_PTR(-ENOENT);
1603 }
1604 return root;
1605 }
1606
1607 key.objectid = objectid;
1608 key.type = BTRFS_ROOT_ITEM_KEY;
1609 key.offset = (u64)-1;
1610 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1611 if (IS_ERR(root))
1612 return root;
1613
1614 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1615 ret = -ENOENT;
1616 goto fail;
1617 }
1618
1619 ret = btrfs_init_fs_root(root, anon_dev);
1620 if (ret)
1621 goto fail;
1622
1623 path = btrfs_alloc_path();
1624 if (!path) {
1625 ret = -ENOMEM;
1626 goto fail;
1627 }
1628 key.objectid = BTRFS_ORPHAN_OBJECTID;
1629 key.type = BTRFS_ORPHAN_ITEM_KEY;
1630 key.offset = objectid;
1631
1632 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1633 btrfs_free_path(path);
1634 if (ret < 0)
1635 goto fail;
1636 if (ret == 0)
1637 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1638
1639 ret = btrfs_insert_fs_root(fs_info, root);
1640 if (ret) {
1641 btrfs_put_root(root);
1642 if (ret == -EEXIST)
1643 goto again;
1644 goto fail;
1645 }
1646 return root;
1647fail:
1648 btrfs_put_root(root);
1649 return ERR_PTR(ret);
1650}
1651
1652/*
1653 * Get in-memory reference of a root structure
1654 *
1655 * @objectid: tree objectid
1656 * @check_ref: if set, verify that the tree exists and the item has at least
1657 * one reference
1658 */
1659struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1660 u64 objectid, bool check_ref)
1661{
1662 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1663}
1664
1665/*
1666 * Get in-memory reference of a root structure, created as new, optionally pass
1667 * the anonymous block device id
1668 *
1669 * @objectid: tree objectid
1670 * @anon_dev: if zero, allocate a new anonymous block device or use the
1671 * parameter value
1672 */
1673struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1674 u64 objectid, dev_t anon_dev)
1675{
1676 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1677}
1678
1679/*
1680 * called by the kthread helper functions to finally call the bio end_io
1681 * functions. This is where read checksum verification actually happens
1682 */
1683static void end_workqueue_fn(struct btrfs_work *work)
1684{
1685 struct bio *bio;
1686 struct btrfs_end_io_wq *end_io_wq;
1687
1688 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1689 bio = end_io_wq->bio;
1690
1691 bio->bi_status = end_io_wq->status;
1692 bio->bi_private = end_io_wq->private;
1693 bio->bi_end_io = end_io_wq->end_io;
1694 bio_endio(bio);
1695 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1696}
1697
1698static int cleaner_kthread(void *arg)
1699{
1700 struct btrfs_root *root = arg;
1701 struct btrfs_fs_info *fs_info = root->fs_info;
1702 int again;
1703
1704 while (1) {
1705 again = 0;
1706
1707 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1708
1709 /* Make the cleaner go to sleep early. */
1710 if (btrfs_need_cleaner_sleep(fs_info))
1711 goto sleep;
1712
1713 /*
1714 * Do not do anything if we might cause open_ctree() to block
1715 * before we have finished mounting the filesystem.
1716 */
1717 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1718 goto sleep;
1719
1720 if (!mutex_trylock(&fs_info->cleaner_mutex))
1721 goto sleep;
1722
1723 /*
1724 * Avoid the problem that we change the status of the fs
1725 * during the above check and trylock.
1726 */
1727 if (btrfs_need_cleaner_sleep(fs_info)) {
1728 mutex_unlock(&fs_info->cleaner_mutex);
1729 goto sleep;
1730 }
1731
1732 btrfs_run_delayed_iputs(fs_info);
1733
1734 again = btrfs_clean_one_deleted_snapshot(root);
1735 mutex_unlock(&fs_info->cleaner_mutex);
1736
1737 /*
1738 * The defragger has dealt with the R/O remount and umount,
1739 * needn't do anything special here.
1740 */
1741 btrfs_run_defrag_inodes(fs_info);
1742
1743 /*
1744 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1745 * with relocation (btrfs_relocate_chunk) and relocation
1746 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1747 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1748 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1749 * unused block groups.
1750 */
1751 btrfs_delete_unused_bgs(fs_info);
1752sleep:
1753 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1754 if (kthread_should_park())
1755 kthread_parkme();
1756 if (kthread_should_stop())
1757 return 0;
1758 if (!again) {
1759 set_current_state(TASK_INTERRUPTIBLE);
1760 schedule();
1761 __set_current_state(TASK_RUNNING);
1762 }
1763 }
1764}
1765
1766static int transaction_kthread(void *arg)
1767{
1768 struct btrfs_root *root = arg;
1769 struct btrfs_fs_info *fs_info = root->fs_info;
1770 struct btrfs_trans_handle *trans;
1771 struct btrfs_transaction *cur;
1772 u64 transid;
1773 time64_t now;
1774 unsigned long delay;
1775 bool cannot_commit;
1776
1777 do {
1778 cannot_commit = false;
1779 delay = HZ * fs_info->commit_interval;
1780 mutex_lock(&fs_info->transaction_kthread_mutex);
1781
1782 spin_lock(&fs_info->trans_lock);
1783 cur = fs_info->running_transaction;
1784 if (!cur) {
1785 spin_unlock(&fs_info->trans_lock);
1786 goto sleep;
1787 }
1788
1789 now = ktime_get_seconds();
1790 if (cur->state < TRANS_STATE_COMMIT_START &&
1791 (now < cur->start_time ||
1792 now - cur->start_time < fs_info->commit_interval)) {
1793 spin_unlock(&fs_info->trans_lock);
1794 delay = HZ * 5;
1795 goto sleep;
1796 }
1797 transid = cur->transid;
1798 spin_unlock(&fs_info->trans_lock);
1799
1800 /* If the file system is aborted, this will always fail. */
1801 trans = btrfs_attach_transaction(root);
1802 if (IS_ERR(trans)) {
1803 if (PTR_ERR(trans) != -ENOENT)
1804 cannot_commit = true;
1805 goto sleep;
1806 }
1807 if (transid == trans->transid) {
1808 btrfs_commit_transaction(trans);
1809 } else {
1810 btrfs_end_transaction(trans);
1811 }
1812sleep:
1813 wake_up_process(fs_info->cleaner_kthread);
1814 mutex_unlock(&fs_info->transaction_kthread_mutex);
1815
1816 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1817 &fs_info->fs_state)))
1818 btrfs_cleanup_transaction(fs_info);
1819 if (!kthread_should_stop() &&
1820 (!btrfs_transaction_blocked(fs_info) ||
1821 cannot_commit))
1822 schedule_timeout_interruptible(delay);
1823 } while (!kthread_should_stop());
1824 return 0;
1825}
1826
1827/*
1828 * This will find the highest generation in the array of root backups. The
1829 * index of the highest array is returned, or -EINVAL if we can't find
1830 * anything.
1831 *
1832 * We check to make sure the array is valid by comparing the
1833 * generation of the latest root in the array with the generation
1834 * in the super block. If they don't match we pitch it.
1835 */
1836static int find_newest_super_backup(struct btrfs_fs_info *info)
1837{
1838 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1839 u64 cur;
1840 struct btrfs_root_backup *root_backup;
1841 int i;
1842
1843 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1844 root_backup = info->super_copy->super_roots + i;
1845 cur = btrfs_backup_tree_root_gen(root_backup);
1846 if (cur == newest_gen)
1847 return i;
1848 }
1849
1850 return -EINVAL;
1851}
1852
1853/*
1854 * copy all the root pointers into the super backup array.
1855 * this will bump the backup pointer by one when it is
1856 * done
1857 */
1858static void backup_super_roots(struct btrfs_fs_info *info)
1859{
1860 const int next_backup = info->backup_root_index;
1861 struct btrfs_root_backup *root_backup;
1862
1863 root_backup = info->super_for_commit->super_roots + next_backup;
1864
1865 /*
1866 * make sure all of our padding and empty slots get zero filled
1867 * regardless of which ones we use today
1868 */
1869 memset(root_backup, 0, sizeof(*root_backup));
1870
1871 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1872
1873 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1874 btrfs_set_backup_tree_root_gen(root_backup,
1875 btrfs_header_generation(info->tree_root->node));
1876
1877 btrfs_set_backup_tree_root_level(root_backup,
1878 btrfs_header_level(info->tree_root->node));
1879
1880 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1881 btrfs_set_backup_chunk_root_gen(root_backup,
1882 btrfs_header_generation(info->chunk_root->node));
1883 btrfs_set_backup_chunk_root_level(root_backup,
1884 btrfs_header_level(info->chunk_root->node));
1885
1886 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1887 btrfs_set_backup_extent_root_gen(root_backup,
1888 btrfs_header_generation(info->extent_root->node));
1889 btrfs_set_backup_extent_root_level(root_backup,
1890 btrfs_header_level(info->extent_root->node));
1891
1892 /*
1893 * we might commit during log recovery, which happens before we set
1894 * the fs_root. Make sure it is valid before we fill it in.
1895 */
1896 if (info->fs_root && info->fs_root->node) {
1897 btrfs_set_backup_fs_root(root_backup,
1898 info->fs_root->node->start);
1899 btrfs_set_backup_fs_root_gen(root_backup,
1900 btrfs_header_generation(info->fs_root->node));
1901 btrfs_set_backup_fs_root_level(root_backup,
1902 btrfs_header_level(info->fs_root->node));
1903 }
1904
1905 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1906 btrfs_set_backup_dev_root_gen(root_backup,
1907 btrfs_header_generation(info->dev_root->node));
1908 btrfs_set_backup_dev_root_level(root_backup,
1909 btrfs_header_level(info->dev_root->node));
1910
1911 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1912 btrfs_set_backup_csum_root_gen(root_backup,
1913 btrfs_header_generation(info->csum_root->node));
1914 btrfs_set_backup_csum_root_level(root_backup,
1915 btrfs_header_level(info->csum_root->node));
1916
1917 btrfs_set_backup_total_bytes(root_backup,
1918 btrfs_super_total_bytes(info->super_copy));
1919 btrfs_set_backup_bytes_used(root_backup,
1920 btrfs_super_bytes_used(info->super_copy));
1921 btrfs_set_backup_num_devices(root_backup,
1922 btrfs_super_num_devices(info->super_copy));
1923
1924 /*
1925 * if we don't copy this out to the super_copy, it won't get remembered
1926 * for the next commit
1927 */
1928 memcpy(&info->super_copy->super_roots,
1929 &info->super_for_commit->super_roots,
1930 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1931}
1932
1933/*
1934 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
1935 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1936 *
1937 * fs_info - filesystem whose backup roots need to be read
1938 * priority - priority of backup root required
1939 *
1940 * Returns backup root index on success and -EINVAL otherwise.
1941 */
1942static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1943{
1944 int backup_index = find_newest_super_backup(fs_info);
1945 struct btrfs_super_block *super = fs_info->super_copy;
1946 struct btrfs_root_backup *root_backup;
1947
1948 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1949 if (priority == 0)
1950 return backup_index;
1951
1952 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1953 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1954 } else {
1955 return -EINVAL;
1956 }
1957
1958 root_backup = super->super_roots + backup_index;
1959
1960 btrfs_set_super_generation(super,
1961 btrfs_backup_tree_root_gen(root_backup));
1962 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1963 btrfs_set_super_root_level(super,
1964 btrfs_backup_tree_root_level(root_backup));
1965 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1966
1967 /*
1968 * Fixme: the total bytes and num_devices need to match or we should
1969 * need a fsck
1970 */
1971 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1972 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1973
1974 return backup_index;
1975}
1976
1977/* helper to cleanup workers */
1978static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1979{
1980 btrfs_destroy_workqueue(fs_info->fixup_workers);
1981 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1982 btrfs_destroy_workqueue(fs_info->workers);
1983 btrfs_destroy_workqueue(fs_info->endio_workers);
1984 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
1985 btrfs_destroy_workqueue(fs_info->rmw_workers);
1986 btrfs_destroy_workqueue(fs_info->endio_write_workers);
1987 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1988 btrfs_destroy_workqueue(fs_info->delayed_workers);
1989 btrfs_destroy_workqueue(fs_info->caching_workers);
1990 btrfs_destroy_workqueue(fs_info->readahead_workers);
1991 btrfs_destroy_workqueue(fs_info->flush_workers);
1992 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1993 if (fs_info->discard_ctl.discard_workers)
1994 destroy_workqueue(fs_info->discard_ctl.discard_workers);
1995 /*
1996 * Now that all other work queues are destroyed, we can safely destroy
1997 * the queues used for metadata I/O, since tasks from those other work
1998 * queues can do metadata I/O operations.
1999 */
2000 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2001 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2002}
2003
2004static void free_root_extent_buffers(struct btrfs_root *root)
2005{
2006 if (root) {
2007 free_extent_buffer(root->node);
2008 free_extent_buffer(root->commit_root);
2009 root->node = NULL;
2010 root->commit_root = NULL;
2011 }
2012}
2013
2014/* helper to cleanup tree roots */
2015static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2016{
2017 free_root_extent_buffers(info->tree_root);
2018
2019 free_root_extent_buffers(info->dev_root);
2020 free_root_extent_buffers(info->extent_root);
2021 free_root_extent_buffers(info->csum_root);
2022 free_root_extent_buffers(info->quota_root);
2023 free_root_extent_buffers(info->uuid_root);
2024 free_root_extent_buffers(info->fs_root);
2025 free_root_extent_buffers(info->data_reloc_root);
2026 if (free_chunk_root)
2027 free_root_extent_buffers(info->chunk_root);
2028 free_root_extent_buffers(info->free_space_root);
2029}
2030
2031void btrfs_put_root(struct btrfs_root *root)
2032{
2033 if (!root)
2034 return;
2035
2036 if (refcount_dec_and_test(&root->refs)) {
2037 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2038 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2039 if (root->anon_dev)
2040 free_anon_bdev(root->anon_dev);
2041 btrfs_drew_lock_destroy(&root->snapshot_lock);
2042 free_root_extent_buffers(root);
2043 kfree(root->free_ino_ctl);
2044 kfree(root->free_ino_pinned);
2045#ifdef CONFIG_BTRFS_DEBUG
2046 spin_lock(&root->fs_info->fs_roots_radix_lock);
2047 list_del_init(&root->leak_list);
2048 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2049#endif
2050 kfree(root);
2051 }
2052}
2053
2054void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2055{
2056 int ret;
2057 struct btrfs_root *gang[8];
2058 int i;
2059
2060 while (!list_empty(&fs_info->dead_roots)) {
2061 gang[0] = list_entry(fs_info->dead_roots.next,
2062 struct btrfs_root, root_list);
2063 list_del(&gang[0]->root_list);
2064
2065 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2066 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2067 btrfs_put_root(gang[0]);
2068 }
2069
2070 while (1) {
2071 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2072 (void **)gang, 0,
2073 ARRAY_SIZE(gang));
2074 if (!ret)
2075 break;
2076 for (i = 0; i < ret; i++)
2077 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
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 = btrfs_grab_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 fs_info->fixup_workers =
2171 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2172
2173 /*
2174 * endios are largely parallel and should have a very
2175 * low idle thresh
2176 */
2177 fs_info->endio_workers =
2178 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2179 fs_info->endio_meta_workers =
2180 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2181 max_active, 4);
2182 fs_info->endio_meta_write_workers =
2183 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2184 max_active, 2);
2185 fs_info->endio_raid56_workers =
2186 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2187 max_active, 4);
2188 fs_info->rmw_workers =
2189 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2190 fs_info->endio_write_workers =
2191 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2192 max_active, 2);
2193 fs_info->endio_freespace_worker =
2194 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2195 max_active, 0);
2196 fs_info->delayed_workers =
2197 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2198 max_active, 0);
2199 fs_info->readahead_workers =
2200 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2201 max_active, 2);
2202 fs_info->qgroup_rescan_workers =
2203 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2204 fs_info->discard_ctl.discard_workers =
2205 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2206
2207 if (!(fs_info->workers && fs_info->delalloc_workers &&
2208 fs_info->flush_workers &&
2209 fs_info->endio_workers && fs_info->endio_meta_workers &&
2210 fs_info->endio_meta_write_workers &&
2211 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2212 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2213 fs_info->caching_workers && fs_info->readahead_workers &&
2214 fs_info->fixup_workers && fs_info->delayed_workers &&
2215 fs_info->qgroup_rescan_workers &&
2216 fs_info->discard_ctl.discard_workers)) {
2217 return -ENOMEM;
2218 }
2219
2220 return 0;
2221}
2222
2223static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2224{
2225 struct crypto_shash *csum_shash;
2226 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2227
2228 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2229
2230 if (IS_ERR(csum_shash)) {
2231 btrfs_err(fs_info, "error allocating %s hash for checksum",
2232 csum_driver);
2233 return PTR_ERR(csum_shash);
2234 }
2235
2236 fs_info->csum_shash = csum_shash;
2237
2238 return 0;
2239}
2240
2241static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2242 struct btrfs_fs_devices *fs_devices)
2243{
2244 int ret;
2245 struct btrfs_root *log_tree_root;
2246 struct btrfs_super_block *disk_super = fs_info->super_copy;
2247 u64 bytenr = btrfs_super_log_root(disk_super);
2248 int level = btrfs_super_log_root_level(disk_super);
2249
2250 if (fs_devices->rw_devices == 0) {
2251 btrfs_warn(fs_info, "log replay required on RO media");
2252 return -EIO;
2253 }
2254
2255 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2256 GFP_KERNEL);
2257 if (!log_tree_root)
2258 return -ENOMEM;
2259
2260 log_tree_root->node = read_tree_block(fs_info, bytenr,
2261 fs_info->generation + 1,
2262 level, NULL);
2263 if (IS_ERR(log_tree_root->node)) {
2264 btrfs_warn(fs_info, "failed to read log tree");
2265 ret = PTR_ERR(log_tree_root->node);
2266 log_tree_root->node = NULL;
2267 btrfs_put_root(log_tree_root);
2268 return ret;
2269 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2270 btrfs_err(fs_info, "failed to read log tree");
2271 btrfs_put_root(log_tree_root);
2272 return -EIO;
2273 }
2274 /* returns with log_tree_root freed on success */
2275 ret = btrfs_recover_log_trees(log_tree_root);
2276 if (ret) {
2277 btrfs_handle_fs_error(fs_info, ret,
2278 "Failed to recover log tree");
2279 btrfs_put_root(log_tree_root);
2280 return ret;
2281 }
2282
2283 if (sb_rdonly(fs_info->sb)) {
2284 ret = btrfs_commit_super(fs_info);
2285 if (ret)
2286 return ret;
2287 }
2288
2289 return 0;
2290}
2291
2292static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2293{
2294 struct btrfs_root *tree_root = fs_info->tree_root;
2295 struct btrfs_root *root;
2296 struct btrfs_key location;
2297 int ret;
2298
2299 BUG_ON(!fs_info->tree_root);
2300
2301 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2302 location.type = BTRFS_ROOT_ITEM_KEY;
2303 location.offset = 0;
2304
2305 root = btrfs_read_tree_root(tree_root, &location);
2306 if (IS_ERR(root)) {
2307 ret = PTR_ERR(root);
2308 goto out;
2309 }
2310 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2311 fs_info->extent_root = root;
2312
2313 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2314 root = btrfs_read_tree_root(tree_root, &location);
2315 if (IS_ERR(root)) {
2316 ret = PTR_ERR(root);
2317 goto out;
2318 }
2319 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2320 fs_info->dev_root = root;
2321 btrfs_init_devices_late(fs_info);
2322
2323 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2324 root = btrfs_read_tree_root(tree_root, &location);
2325 if (IS_ERR(root)) {
2326 ret = PTR_ERR(root);
2327 goto out;
2328 }
2329 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2330 fs_info->csum_root = root;
2331
2332 /*
2333 * This tree can share blocks with some other fs tree during relocation
2334 * and we need a proper setup by btrfs_get_fs_root
2335 */
2336 root = btrfs_get_fs_root(tree_root->fs_info,
2337 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2338 if (IS_ERR(root)) {
2339 ret = PTR_ERR(root);
2340 goto out;
2341 }
2342 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2343 fs_info->data_reloc_root = root;
2344
2345 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2346 root = btrfs_read_tree_root(tree_root, &location);
2347 if (!IS_ERR(root)) {
2348 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2349 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2350 fs_info->quota_root = root;
2351 }
2352
2353 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2354 root = btrfs_read_tree_root(tree_root, &location);
2355 if (IS_ERR(root)) {
2356 ret = PTR_ERR(root);
2357 if (ret != -ENOENT)
2358 goto out;
2359 } else {
2360 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2361 fs_info->uuid_root = root;
2362 }
2363
2364 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2365 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2366 root = btrfs_read_tree_root(tree_root, &location);
2367 if (IS_ERR(root)) {
2368 ret = PTR_ERR(root);
2369 goto out;
2370 }
2371 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2372 fs_info->free_space_root = root;
2373 }
2374
2375 return 0;
2376out:
2377 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2378 location.objectid, ret);
2379 return ret;
2380}
2381
2382/*
2383 * Real super block validation
2384 * NOTE: super csum type and incompat features will not be checked here.
2385 *
2386 * @sb: super block to check
2387 * @mirror_num: the super block number to check its bytenr:
2388 * 0 the primary (1st) sb
2389 * 1, 2 2nd and 3rd backup copy
2390 * -1 skip bytenr check
2391 */
2392static int validate_super(struct btrfs_fs_info *fs_info,
2393 struct btrfs_super_block *sb, int mirror_num)
2394{
2395 u64 nodesize = btrfs_super_nodesize(sb);
2396 u64 sectorsize = btrfs_super_sectorsize(sb);
2397 int ret = 0;
2398
2399 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2400 btrfs_err(fs_info, "no valid FS found");
2401 ret = -EINVAL;
2402 }
2403 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2404 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2405 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2406 ret = -EINVAL;
2407 }
2408 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2409 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2410 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2411 ret = -EINVAL;
2412 }
2413 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2414 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2415 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2416 ret = -EINVAL;
2417 }
2418 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2419 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2420 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2421 ret = -EINVAL;
2422 }
2423
2424 /*
2425 * Check sectorsize and nodesize first, other check will need it.
2426 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2427 */
2428 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2429 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2430 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2431 ret = -EINVAL;
2432 }
2433 /* Only PAGE SIZE is supported yet */
2434 if (sectorsize != PAGE_SIZE) {
2435 btrfs_err(fs_info,
2436 "sectorsize %llu not supported yet, only support %lu",
2437 sectorsize, PAGE_SIZE);
2438 ret = -EINVAL;
2439 }
2440 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2441 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2442 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2443 ret = -EINVAL;
2444 }
2445 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2446 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2447 le32_to_cpu(sb->__unused_leafsize), nodesize);
2448 ret = -EINVAL;
2449 }
2450
2451 /* Root alignment check */
2452 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2453 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2454 btrfs_super_root(sb));
2455 ret = -EINVAL;
2456 }
2457 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2458 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2459 btrfs_super_chunk_root(sb));
2460 ret = -EINVAL;
2461 }
2462 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2463 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2464 btrfs_super_log_root(sb));
2465 ret = -EINVAL;
2466 }
2467
2468 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2469 BTRFS_FSID_SIZE) != 0) {
2470 btrfs_err(fs_info,
2471 "dev_item UUID does not match metadata fsid: %pU != %pU",
2472 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2473 ret = -EINVAL;
2474 }
2475
2476 /*
2477 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2478 * done later
2479 */
2480 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2481 btrfs_err(fs_info, "bytes_used is too small %llu",
2482 btrfs_super_bytes_used(sb));
2483 ret = -EINVAL;
2484 }
2485 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2486 btrfs_err(fs_info, "invalid stripesize %u",
2487 btrfs_super_stripesize(sb));
2488 ret = -EINVAL;
2489 }
2490 if (btrfs_super_num_devices(sb) > (1UL << 31))
2491 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2492 btrfs_super_num_devices(sb));
2493 if (btrfs_super_num_devices(sb) == 0) {
2494 btrfs_err(fs_info, "number of devices is 0");
2495 ret = -EINVAL;
2496 }
2497
2498 if (mirror_num >= 0 &&
2499 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2500 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2501 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2502 ret = -EINVAL;
2503 }
2504
2505 /*
2506 * Obvious sys_chunk_array corruptions, it must hold at least one key
2507 * and one chunk
2508 */
2509 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2510 btrfs_err(fs_info, "system chunk array too big %u > %u",
2511 btrfs_super_sys_array_size(sb),
2512 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2513 ret = -EINVAL;
2514 }
2515 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2516 + sizeof(struct btrfs_chunk)) {
2517 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2518 btrfs_super_sys_array_size(sb),
2519 sizeof(struct btrfs_disk_key)
2520 + sizeof(struct btrfs_chunk));
2521 ret = -EINVAL;
2522 }
2523
2524 /*
2525 * The generation is a global counter, we'll trust it more than the others
2526 * but it's still possible that it's the one that's wrong.
2527 */
2528 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2529 btrfs_warn(fs_info,
2530 "suspicious: generation < chunk_root_generation: %llu < %llu",
2531 btrfs_super_generation(sb),
2532 btrfs_super_chunk_root_generation(sb));
2533 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2534 && btrfs_super_cache_generation(sb) != (u64)-1)
2535 btrfs_warn(fs_info,
2536 "suspicious: generation < cache_generation: %llu < %llu",
2537 btrfs_super_generation(sb),
2538 btrfs_super_cache_generation(sb));
2539
2540 return ret;
2541}
2542
2543/*
2544 * Validation of super block at mount time.
2545 * Some checks already done early at mount time, like csum type and incompat
2546 * flags will be skipped.
2547 */
2548static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2549{
2550 return validate_super(fs_info, fs_info->super_copy, 0);
2551}
2552
2553/*
2554 * Validation of super block at write time.
2555 * Some checks like bytenr check will be skipped as their values will be
2556 * overwritten soon.
2557 * Extra checks like csum type and incompat flags will be done here.
2558 */
2559static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2560 struct btrfs_super_block *sb)
2561{
2562 int ret;
2563
2564 ret = validate_super(fs_info, sb, -1);
2565 if (ret < 0)
2566 goto out;
2567 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2568 ret = -EUCLEAN;
2569 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2570 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2571 goto out;
2572 }
2573 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2574 ret = -EUCLEAN;
2575 btrfs_err(fs_info,
2576 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2577 btrfs_super_incompat_flags(sb),
2578 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2579 goto out;
2580 }
2581out:
2582 if (ret < 0)
2583 btrfs_err(fs_info,
2584 "super block corruption detected before writing it to disk");
2585 return ret;
2586}
2587
2588static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2589{
2590 int backup_index = find_newest_super_backup(fs_info);
2591 struct btrfs_super_block *sb = fs_info->super_copy;
2592 struct btrfs_root *tree_root = fs_info->tree_root;
2593 bool handle_error = false;
2594 int ret = 0;
2595 int i;
2596
2597 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2598 u64 generation;
2599 int level;
2600
2601 if (handle_error) {
2602 if (!IS_ERR(tree_root->node))
2603 free_extent_buffer(tree_root->node);
2604 tree_root->node = NULL;
2605
2606 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2607 break;
2608
2609 free_root_pointers(fs_info, 0);
2610
2611 /*
2612 * Don't use the log in recovery mode, it won't be
2613 * valid
2614 */
2615 btrfs_set_super_log_root(sb, 0);
2616
2617 /* We can't trust the free space cache either */
2618 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2619
2620 ret = read_backup_root(fs_info, i);
2621 backup_index = ret;
2622 if (ret < 0)
2623 return ret;
2624 }
2625 generation = btrfs_super_generation(sb);
2626 level = btrfs_super_root_level(sb);
2627 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2628 generation, level, NULL);
2629 if (IS_ERR(tree_root->node) ||
2630 !extent_buffer_uptodate(tree_root->node)) {
2631 handle_error = true;
2632
2633 if (IS_ERR(tree_root->node)) {
2634 ret = PTR_ERR(tree_root->node);
2635 tree_root->node = NULL;
2636 } else if (!extent_buffer_uptodate(tree_root->node)) {
2637 ret = -EUCLEAN;
2638 }
2639
2640 btrfs_warn(fs_info, "failed to read tree root");
2641 continue;
2642 }
2643
2644 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2645 tree_root->commit_root = btrfs_root_node(tree_root);
2646 btrfs_set_root_refs(&tree_root->root_item, 1);
2647
2648 /*
2649 * No need to hold btrfs_root::objectid_mutex since the fs
2650 * hasn't been fully initialised and we are the only user
2651 */
2652 ret = btrfs_find_highest_objectid(tree_root,
2653 &tree_root->highest_objectid);
2654 if (ret < 0) {
2655 handle_error = true;
2656 continue;
2657 }
2658
2659 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2660
2661 ret = btrfs_read_roots(fs_info);
2662 if (ret < 0) {
2663 handle_error = true;
2664 continue;
2665 }
2666
2667 /* All successful */
2668 fs_info->generation = generation;
2669 fs_info->last_trans_committed = generation;
2670
2671 /* Always begin writing backup roots after the one being used */
2672 if (backup_index < 0) {
2673 fs_info->backup_root_index = 0;
2674 } else {
2675 fs_info->backup_root_index = backup_index + 1;
2676 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2677 }
2678 break;
2679 }
2680
2681 return ret;
2682}
2683
2684void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2685{
2686 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2687 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2688 INIT_LIST_HEAD(&fs_info->trans_list);
2689 INIT_LIST_HEAD(&fs_info->dead_roots);
2690 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2691 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2692 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2693 spin_lock_init(&fs_info->delalloc_root_lock);
2694 spin_lock_init(&fs_info->trans_lock);
2695 spin_lock_init(&fs_info->fs_roots_radix_lock);
2696 spin_lock_init(&fs_info->delayed_iput_lock);
2697 spin_lock_init(&fs_info->defrag_inodes_lock);
2698 spin_lock_init(&fs_info->super_lock);
2699 spin_lock_init(&fs_info->buffer_lock);
2700 spin_lock_init(&fs_info->unused_bgs_lock);
2701 rwlock_init(&fs_info->tree_mod_log_lock);
2702 mutex_init(&fs_info->unused_bg_unpin_mutex);
2703 mutex_init(&fs_info->delete_unused_bgs_mutex);
2704 mutex_init(&fs_info->reloc_mutex);
2705 mutex_init(&fs_info->delalloc_root_mutex);
2706 seqlock_init(&fs_info->profiles_lock);
2707
2708 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2709 INIT_LIST_HEAD(&fs_info->space_info);
2710 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2711 INIT_LIST_HEAD(&fs_info->unused_bgs);
2712#ifdef CONFIG_BTRFS_DEBUG
2713 INIT_LIST_HEAD(&fs_info->allocated_roots);
2714 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2715 spin_lock_init(&fs_info->eb_leak_lock);
2716#endif
2717 extent_map_tree_init(&fs_info->mapping_tree);
2718 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2719 BTRFS_BLOCK_RSV_GLOBAL);
2720 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2721 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2722 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2723 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2724 BTRFS_BLOCK_RSV_DELOPS);
2725 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2726 BTRFS_BLOCK_RSV_DELREFS);
2727
2728 atomic_set(&fs_info->async_delalloc_pages, 0);
2729 atomic_set(&fs_info->defrag_running, 0);
2730 atomic_set(&fs_info->reada_works_cnt, 0);
2731 atomic_set(&fs_info->nr_delayed_iputs, 0);
2732 atomic64_set(&fs_info->tree_mod_seq, 0);
2733 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2734 fs_info->metadata_ratio = 0;
2735 fs_info->defrag_inodes = RB_ROOT;
2736 atomic64_set(&fs_info->free_chunk_space, 0);
2737 fs_info->tree_mod_log = RB_ROOT;
2738 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2739 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2740 /* readahead state */
2741 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2742 spin_lock_init(&fs_info->reada_lock);
2743 btrfs_init_ref_verify(fs_info);
2744
2745 fs_info->thread_pool_size = min_t(unsigned long,
2746 num_online_cpus() + 2, 8);
2747
2748 INIT_LIST_HEAD(&fs_info->ordered_roots);
2749 spin_lock_init(&fs_info->ordered_root_lock);
2750
2751 btrfs_init_scrub(fs_info);
2752#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2753 fs_info->check_integrity_print_mask = 0;
2754#endif
2755 btrfs_init_balance(fs_info);
2756 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2757
2758 spin_lock_init(&fs_info->block_group_cache_lock);
2759 fs_info->block_group_cache_tree = RB_ROOT;
2760 fs_info->first_logical_byte = (u64)-1;
2761
2762 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2763 IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2764 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2765
2766 mutex_init(&fs_info->ordered_operations_mutex);
2767 mutex_init(&fs_info->tree_log_mutex);
2768 mutex_init(&fs_info->chunk_mutex);
2769 mutex_init(&fs_info->transaction_kthread_mutex);
2770 mutex_init(&fs_info->cleaner_mutex);
2771 mutex_init(&fs_info->ro_block_group_mutex);
2772 init_rwsem(&fs_info->commit_root_sem);
2773 init_rwsem(&fs_info->cleanup_work_sem);
2774 init_rwsem(&fs_info->subvol_sem);
2775 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2776
2777 btrfs_init_dev_replace_locks(fs_info);
2778 btrfs_init_qgroup(fs_info);
2779 btrfs_discard_init(fs_info);
2780
2781 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2782 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2783
2784 init_waitqueue_head(&fs_info->transaction_throttle);
2785 init_waitqueue_head(&fs_info->transaction_wait);
2786 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2787 init_waitqueue_head(&fs_info->async_submit_wait);
2788 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2789
2790 /* Usable values until the real ones are cached from the superblock */
2791 fs_info->nodesize = 4096;
2792 fs_info->sectorsize = 4096;
2793 fs_info->stripesize = 4096;
2794
2795 spin_lock_init(&fs_info->swapfile_pins_lock);
2796 fs_info->swapfile_pins = RB_ROOT;
2797
2798 fs_info->send_in_progress = 0;
2799}
2800
2801static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2802{
2803 int ret;
2804
2805 fs_info->sb = sb;
2806 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2807 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2808
2809 ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2810 if (ret)
2811 return ret;
2812
2813 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2814 if (ret)
2815 return ret;
2816
2817 fs_info->dirty_metadata_batch = PAGE_SIZE *
2818 (1 + ilog2(nr_cpu_ids));
2819
2820 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2821 if (ret)
2822 return ret;
2823
2824 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2825 GFP_KERNEL);
2826 if (ret)
2827 return ret;
2828
2829 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2830 GFP_KERNEL);
2831 if (!fs_info->delayed_root)
2832 return -ENOMEM;
2833 btrfs_init_delayed_root(fs_info->delayed_root);
2834
2835 return btrfs_alloc_stripe_hash_table(fs_info);
2836}
2837
2838static int btrfs_uuid_rescan_kthread(void *data)
2839{
2840 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
2841 int ret;
2842
2843 /*
2844 * 1st step is to iterate through the existing UUID tree and
2845 * to delete all entries that contain outdated data.
2846 * 2nd step is to add all missing entries to the UUID tree.
2847 */
2848 ret = btrfs_uuid_tree_iterate(fs_info);
2849 if (ret < 0) {
2850 if (ret != -EINTR)
2851 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2852 ret);
2853 up(&fs_info->uuid_tree_rescan_sem);
2854 return ret;
2855 }
2856 return btrfs_uuid_scan_kthread(data);
2857}
2858
2859static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2860{
2861 struct task_struct *task;
2862
2863 down(&fs_info->uuid_tree_rescan_sem);
2864 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2865 if (IS_ERR(task)) {
2866 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2867 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2868 up(&fs_info->uuid_tree_rescan_sem);
2869 return PTR_ERR(task);
2870 }
2871
2872 return 0;
2873}
2874
2875int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
2876 char *options)
2877{
2878 u32 sectorsize;
2879 u32 nodesize;
2880 u32 stripesize;
2881 u64 generation;
2882 u64 features;
2883 u16 csum_type;
2884 struct btrfs_super_block *disk_super;
2885 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2886 struct btrfs_root *tree_root;
2887 struct btrfs_root *chunk_root;
2888 int ret;
2889 int err = -EINVAL;
2890 int clear_free_space_tree = 0;
2891 int level;
2892
2893 ret = init_mount_fs_info(fs_info, sb);
2894 if (ret) {
2895 err = ret;
2896 goto fail;
2897 }
2898
2899 /* These need to be init'ed before we start creating inodes and such. */
2900 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
2901 GFP_KERNEL);
2902 fs_info->tree_root = tree_root;
2903 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
2904 GFP_KERNEL);
2905 fs_info->chunk_root = chunk_root;
2906 if (!tree_root || !chunk_root) {
2907 err = -ENOMEM;
2908 goto fail;
2909 }
2910
2911 fs_info->btree_inode = new_inode(sb);
2912 if (!fs_info->btree_inode) {
2913 err = -ENOMEM;
2914 goto fail;
2915 }
2916 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2917 btrfs_init_btree_inode(fs_info);
2918
2919 invalidate_bdev(fs_devices->latest_bdev);
2920
2921 /*
2922 * Read super block and check the signature bytes only
2923 */
2924 disk_super = btrfs_read_dev_super(fs_devices->latest_bdev);
2925 if (IS_ERR(disk_super)) {
2926 err = PTR_ERR(disk_super);
2927 goto fail_alloc;
2928 }
2929
2930 /*
2931 * Verify the type first, if that or the the checksum value are
2932 * corrupted, we'll find out
2933 */
2934 csum_type = btrfs_super_csum_type(disk_super);
2935 if (!btrfs_supported_super_csum(csum_type)) {
2936 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2937 csum_type);
2938 err = -EINVAL;
2939 btrfs_release_disk_super(disk_super);
2940 goto fail_alloc;
2941 }
2942
2943 ret = btrfs_init_csum_hash(fs_info, csum_type);
2944 if (ret) {
2945 err = ret;
2946 btrfs_release_disk_super(disk_super);
2947 goto fail_alloc;
2948 }
2949
2950 /*
2951 * We want to check superblock checksum, the type is stored inside.
2952 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2953 */
2954 if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
2955 btrfs_err(fs_info, "superblock checksum mismatch");
2956 err = -EINVAL;
2957 btrfs_release_disk_super(disk_super);
2958 goto fail_alloc;
2959 }
2960
2961 /*
2962 * super_copy is zeroed at allocation time and we never touch the
2963 * following bytes up to INFO_SIZE, the checksum is calculated from
2964 * the whole block of INFO_SIZE
2965 */
2966 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
2967 btrfs_release_disk_super(disk_super);
2968
2969 disk_super = fs_info->super_copy;
2970
2971 ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2972 BTRFS_FSID_SIZE));
2973
2974 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2975 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2976 fs_info->super_copy->metadata_uuid,
2977 BTRFS_FSID_SIZE));
2978 }
2979
2980 features = btrfs_super_flags(disk_super);
2981 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2982 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2983 btrfs_set_super_flags(disk_super, features);
2984 btrfs_info(fs_info,
2985 "found metadata UUID change in progress flag, clearing");
2986 }
2987
2988 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2989 sizeof(*fs_info->super_for_commit));
2990
2991 ret = btrfs_validate_mount_super(fs_info);
2992 if (ret) {
2993 btrfs_err(fs_info, "superblock contains fatal errors");
2994 err = -EINVAL;
2995 goto fail_alloc;
2996 }
2997
2998 if (!btrfs_super_root(disk_super))
2999 goto fail_alloc;
3000
3001 /* check FS state, whether FS is broken. */
3002 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3003 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3004
3005 /*
3006 * In the long term, we'll store the compression type in the super
3007 * block, and it'll be used for per file compression control.
3008 */
3009 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3010
3011 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3012 if (ret) {
3013 err = ret;
3014 goto fail_alloc;
3015 }
3016
3017 features = btrfs_super_incompat_flags(disk_super) &
3018 ~BTRFS_FEATURE_INCOMPAT_SUPP;
3019 if (features) {
3020 btrfs_err(fs_info,
3021 "cannot mount because of unsupported optional features (%llx)",
3022 features);
3023 err = -EINVAL;
3024 goto fail_alloc;
3025 }
3026
3027 features = btrfs_super_incompat_flags(disk_super);
3028 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3029 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3030 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3031 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3032 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3033
3034 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3035 btrfs_info(fs_info, "has skinny extents");
3036
3037 /*
3038 * flag our filesystem as having big metadata blocks if
3039 * they are bigger than the page size
3040 */
3041 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3042 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3043 btrfs_info(fs_info,
3044 "flagging fs with big metadata feature");
3045 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3046 }
3047
3048 nodesize = btrfs_super_nodesize(disk_super);
3049 sectorsize = btrfs_super_sectorsize(disk_super);
3050 stripesize = sectorsize;
3051 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3052 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3053
3054 /* Cache block sizes */
3055 fs_info->nodesize = nodesize;
3056 fs_info->sectorsize = sectorsize;
3057 fs_info->stripesize = stripesize;
3058
3059 /*
3060 * mixed block groups end up with duplicate but slightly offset
3061 * extent buffers for the same range. It leads to corruptions
3062 */
3063 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3064 (sectorsize != nodesize)) {
3065 btrfs_err(fs_info,
3066"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3067 nodesize, sectorsize);
3068 goto fail_alloc;
3069 }
3070
3071 /*
3072 * Needn't use the lock because there is no other task which will
3073 * update the flag.
3074 */
3075 btrfs_set_super_incompat_flags(disk_super, features);
3076
3077 features = btrfs_super_compat_ro_flags(disk_super) &
3078 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3079 if (!sb_rdonly(sb) && features) {
3080 btrfs_err(fs_info,
3081 "cannot mount read-write because of unsupported optional features (%llx)",
3082 features);
3083 err = -EINVAL;
3084 goto fail_alloc;
3085 }
3086
3087 ret = btrfs_init_workqueues(fs_info, fs_devices);
3088 if (ret) {
3089 err = ret;
3090 goto fail_sb_buffer;
3091 }
3092
3093 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
3094 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
3095 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3096 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3097
3098 sb->s_blocksize = sectorsize;
3099 sb->s_blocksize_bits = blksize_bits(sectorsize);
3100 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3101
3102 mutex_lock(&fs_info->chunk_mutex);
3103 ret = btrfs_read_sys_array(fs_info);
3104 mutex_unlock(&fs_info->chunk_mutex);
3105 if (ret) {
3106 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3107 goto fail_sb_buffer;
3108 }
3109
3110 generation = btrfs_super_chunk_root_generation(disk_super);
3111 level = btrfs_super_chunk_root_level(disk_super);
3112
3113 chunk_root->node = read_tree_block(fs_info,
3114 btrfs_super_chunk_root(disk_super),
3115 generation, level, NULL);
3116 if (IS_ERR(chunk_root->node) ||
3117 !extent_buffer_uptodate(chunk_root->node)) {
3118 btrfs_err(fs_info, "failed to read chunk root");
3119 if (!IS_ERR(chunk_root->node))
3120 free_extent_buffer(chunk_root->node);
3121 chunk_root->node = NULL;
3122 goto fail_tree_roots;
3123 }
3124 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3125 chunk_root->commit_root = btrfs_root_node(chunk_root);
3126
3127 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3128 offsetof(struct btrfs_header, chunk_tree_uuid),
3129 BTRFS_UUID_SIZE);
3130
3131 ret = btrfs_read_chunk_tree(fs_info);
3132 if (ret) {
3133 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3134 goto fail_tree_roots;
3135 }
3136
3137 /*
3138 * Keep the devid that is marked to be the target device for the
3139 * device replace procedure
3140 */
3141 btrfs_free_extra_devids(fs_devices, 0);
3142
3143 if (!fs_devices->latest_bdev) {
3144 btrfs_err(fs_info, "failed to read devices");
3145 goto fail_tree_roots;
3146 }
3147
3148 ret = init_tree_roots(fs_info);
3149 if (ret)
3150 goto fail_tree_roots;
3151
3152 /*
3153 * If we have a uuid root and we're not being told to rescan we need to
3154 * check the generation here so we can set the
3155 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3156 * transaction during a balance or the log replay without updating the
3157 * uuid generation, and then if we crash we would rescan the uuid tree,
3158 * even though it was perfectly fine.
3159 */
3160 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3161 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3162 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3163
3164 ret = btrfs_verify_dev_extents(fs_info);
3165 if (ret) {
3166 btrfs_err(fs_info,
3167 "failed to verify dev extents against chunks: %d",
3168 ret);
3169 goto fail_block_groups;
3170 }
3171 ret = btrfs_recover_balance(fs_info);
3172 if (ret) {
3173 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3174 goto fail_block_groups;
3175 }
3176
3177 ret = btrfs_init_dev_stats(fs_info);
3178 if (ret) {
3179 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3180 goto fail_block_groups;
3181 }
3182
3183 ret = btrfs_init_dev_replace(fs_info);
3184 if (ret) {
3185 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3186 goto fail_block_groups;
3187 }
3188
3189 btrfs_free_extra_devids(fs_devices, 1);
3190
3191 ret = btrfs_sysfs_add_fsid(fs_devices);
3192 if (ret) {
3193 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3194 ret);
3195 goto fail_block_groups;
3196 }
3197
3198 ret = btrfs_sysfs_add_mounted(fs_info);
3199 if (ret) {
3200 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3201 goto fail_fsdev_sysfs;
3202 }
3203
3204 ret = btrfs_init_space_info(fs_info);
3205 if (ret) {
3206 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3207 goto fail_sysfs;
3208 }
3209
3210 ret = btrfs_read_block_groups(fs_info);
3211 if (ret) {
3212 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3213 goto fail_sysfs;
3214 }
3215
3216 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3217 btrfs_warn(fs_info,
3218 "writable mount is not allowed due to too many missing devices");
3219 goto fail_sysfs;
3220 }
3221
3222 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3223 "btrfs-cleaner");
3224 if (IS_ERR(fs_info->cleaner_kthread))
3225 goto fail_sysfs;
3226
3227 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3228 tree_root,
3229 "btrfs-transaction");
3230 if (IS_ERR(fs_info->transaction_kthread))
3231 goto fail_cleaner;
3232
3233 if (!btrfs_test_opt(fs_info, NOSSD) &&
3234 !fs_info->fs_devices->rotating) {
3235 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3236 }
3237
3238 /*
3239 * Mount does not set all options immediately, we can do it now and do
3240 * not have to wait for transaction commit
3241 */
3242 btrfs_apply_pending_changes(fs_info);
3243
3244#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3245 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3246 ret = btrfsic_mount(fs_info, fs_devices,
3247 btrfs_test_opt(fs_info,
3248 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3249 1 : 0,
3250 fs_info->check_integrity_print_mask);
3251 if (ret)
3252 btrfs_warn(fs_info,
3253 "failed to initialize integrity check module: %d",
3254 ret);
3255 }
3256#endif
3257 ret = btrfs_read_qgroup_config(fs_info);
3258 if (ret)
3259 goto fail_trans_kthread;
3260
3261 if (btrfs_build_ref_tree(fs_info))
3262 btrfs_err(fs_info, "couldn't build ref tree");
3263
3264 /* do not make disk changes in broken FS or nologreplay is given */
3265 if (btrfs_super_log_root(disk_super) != 0 &&
3266 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3267 btrfs_info(fs_info, "start tree-log replay");
3268 ret = btrfs_replay_log(fs_info, fs_devices);
3269 if (ret) {
3270 err = ret;
3271 goto fail_qgroup;
3272 }
3273 }
3274
3275 ret = btrfs_find_orphan_roots(fs_info);
3276 if (ret)
3277 goto fail_qgroup;
3278
3279 if (!sb_rdonly(sb)) {
3280 ret = btrfs_cleanup_fs_roots(fs_info);
3281 if (ret)
3282 goto fail_qgroup;
3283
3284 mutex_lock(&fs_info->cleaner_mutex);
3285 ret = btrfs_recover_relocation(tree_root);
3286 mutex_unlock(&fs_info->cleaner_mutex);
3287 if (ret < 0) {
3288 btrfs_warn(fs_info, "failed to recover relocation: %d",
3289 ret);
3290 err = -EINVAL;
3291 goto fail_qgroup;
3292 }
3293 }
3294
3295 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3296 if (IS_ERR(fs_info->fs_root)) {
3297 err = PTR_ERR(fs_info->fs_root);
3298 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3299 fs_info->fs_root = NULL;
3300 goto fail_qgroup;
3301 }
3302
3303 if (sb_rdonly(sb))
3304 return 0;
3305
3306 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3307 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3308 clear_free_space_tree = 1;
3309 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3310 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3311 btrfs_warn(fs_info, "free space tree is invalid");
3312 clear_free_space_tree = 1;
3313 }
3314
3315 if (clear_free_space_tree) {
3316 btrfs_info(fs_info, "clearing free space tree");
3317 ret = btrfs_clear_free_space_tree(fs_info);
3318 if (ret) {
3319 btrfs_warn(fs_info,
3320 "failed to clear free space tree: %d", ret);
3321 close_ctree(fs_info);
3322 return ret;
3323 }
3324 }
3325
3326 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3327 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3328 btrfs_info(fs_info, "creating free space tree");
3329 ret = btrfs_create_free_space_tree(fs_info);
3330 if (ret) {
3331 btrfs_warn(fs_info,
3332 "failed to create free space tree: %d", ret);
3333 close_ctree(fs_info);
3334 return ret;
3335 }
3336 }
3337
3338 down_read(&fs_info->cleanup_work_sem);
3339 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3340 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3341 up_read(&fs_info->cleanup_work_sem);
3342 close_ctree(fs_info);
3343 return ret;
3344 }
3345 up_read(&fs_info->cleanup_work_sem);
3346
3347 ret = btrfs_resume_balance_async(fs_info);
3348 if (ret) {
3349 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3350 close_ctree(fs_info);
3351 return ret;
3352 }
3353
3354 ret = btrfs_resume_dev_replace_async(fs_info);
3355 if (ret) {
3356 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3357 close_ctree(fs_info);
3358 return ret;
3359 }
3360
3361 btrfs_qgroup_rescan_resume(fs_info);
3362 btrfs_discard_resume(fs_info);
3363
3364 if (!fs_info->uuid_root) {
3365 btrfs_info(fs_info, "creating UUID tree");
3366 ret = btrfs_create_uuid_tree(fs_info);
3367 if (ret) {
3368 btrfs_warn(fs_info,
3369 "failed to create the UUID tree: %d", ret);
3370 close_ctree(fs_info);
3371 return ret;
3372 }
3373 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3374 fs_info->generation !=
3375 btrfs_super_uuid_tree_generation(disk_super)) {
3376 btrfs_info(fs_info, "checking UUID tree");
3377 ret = btrfs_check_uuid_tree(fs_info);
3378 if (ret) {
3379 btrfs_warn(fs_info,
3380 "failed to check the UUID tree: %d", ret);
3381 close_ctree(fs_info);
3382 return ret;
3383 }
3384 }
3385 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3386
3387 /*
3388 * backuproot only affect mount behavior, and if open_ctree succeeded,
3389 * no need to keep the flag
3390 */
3391 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3392
3393 return 0;
3394
3395fail_qgroup:
3396 btrfs_free_qgroup_config(fs_info);
3397fail_trans_kthread:
3398 kthread_stop(fs_info->transaction_kthread);
3399 btrfs_cleanup_transaction(fs_info);
3400 btrfs_free_fs_roots(fs_info);
3401fail_cleaner:
3402 kthread_stop(fs_info->cleaner_kthread);
3403
3404 /*
3405 * make sure we're done with the btree inode before we stop our
3406 * kthreads
3407 */
3408 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3409
3410fail_sysfs:
3411 btrfs_sysfs_remove_mounted(fs_info);
3412
3413fail_fsdev_sysfs:
3414 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3415
3416fail_block_groups:
3417 btrfs_put_block_group_cache(fs_info);
3418
3419fail_tree_roots:
3420 if (fs_info->data_reloc_root)
3421 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3422 free_root_pointers(fs_info, true);
3423 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3424
3425fail_sb_buffer:
3426 btrfs_stop_all_workers(fs_info);
3427 btrfs_free_block_groups(fs_info);
3428fail_alloc:
3429 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3430
3431 iput(fs_info->btree_inode);
3432fail:
3433 btrfs_close_devices(fs_info->fs_devices);
3434 return err;
3435}
3436ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3437
3438static void btrfs_end_super_write(struct bio *bio)
3439{
3440 struct btrfs_device *device = bio->bi_private;
3441 struct bio_vec *bvec;
3442 struct bvec_iter_all iter_all;
3443 struct page *page;
3444
3445 bio_for_each_segment_all(bvec, bio, iter_all) {
3446 page = bvec->bv_page;
3447
3448 if (bio->bi_status) {
3449 btrfs_warn_rl_in_rcu(device->fs_info,
3450 "lost page write due to IO error on %s (%d)",
3451 rcu_str_deref(device->name),
3452 blk_status_to_errno(bio->bi_status));
3453 ClearPageUptodate(page);
3454 SetPageError(page);
3455 btrfs_dev_stat_inc_and_print(device,
3456 BTRFS_DEV_STAT_WRITE_ERRS);
3457 } else {
3458 SetPageUptodate(page);
3459 }
3460
3461 put_page(page);
3462 unlock_page(page);
3463 }
3464
3465 bio_put(bio);
3466}
3467
3468struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3469 int copy_num)
3470{
3471 struct btrfs_super_block *super;
3472 struct page *page;
3473 u64 bytenr;
3474 struct address_space *mapping = bdev->bd_inode->i_mapping;
3475
3476 bytenr = btrfs_sb_offset(copy_num);
3477 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3478 return ERR_PTR(-EINVAL);
3479
3480 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3481 if (IS_ERR(page))
3482 return ERR_CAST(page);
3483
3484 super = page_address(page);
3485 if (btrfs_super_bytenr(super) != bytenr ||
3486 btrfs_super_magic(super) != BTRFS_MAGIC) {
3487 btrfs_release_disk_super(super);
3488 return ERR_PTR(-EINVAL);
3489 }
3490
3491 return super;
3492}
3493
3494
3495struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3496{
3497 struct btrfs_super_block *super, *latest = NULL;
3498 int i;
3499 u64 transid = 0;
3500
3501 /* we would like to check all the supers, but that would make
3502 * a btrfs mount succeed after a mkfs from a different FS.
3503 * So, we need to add a special mount option to scan for
3504 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3505 */
3506 for (i = 0; i < 1; i++) {
3507 super = btrfs_read_dev_one_super(bdev, i);
3508 if (IS_ERR(super))
3509 continue;
3510
3511 if (!latest || btrfs_super_generation(super) > transid) {
3512 if (latest)
3513 btrfs_release_disk_super(super);
3514
3515 latest = super;
3516 transid = btrfs_super_generation(super);
3517 }
3518 }
3519
3520 return super;
3521}
3522
3523/*
3524 * Write superblock @sb to the @device. Do not wait for completion, all the
3525 * pages we use for writing are locked.
3526 *
3527 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3528 * the expected device size at commit time. Note that max_mirrors must be
3529 * same for write and wait phases.
3530 *
3531 * Return number of errors when page is not found or submission fails.
3532 */
3533static int write_dev_supers(struct btrfs_device *device,
3534 struct btrfs_super_block *sb, int max_mirrors)
3535{
3536 struct btrfs_fs_info *fs_info = device->fs_info;
3537 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3538 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3539 int i;
3540 int errors = 0;
3541 u64 bytenr;
3542
3543 if (max_mirrors == 0)
3544 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3545
3546 shash->tfm = fs_info->csum_shash;
3547
3548 for (i = 0; i < max_mirrors; i++) {
3549 struct page *page;
3550 struct bio *bio;
3551 struct btrfs_super_block *disk_super;
3552
3553 bytenr = btrfs_sb_offset(i);
3554 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3555 device->commit_total_bytes)
3556 break;
3557
3558 btrfs_set_super_bytenr(sb, bytenr);
3559
3560 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3561 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3562 sb->csum);
3563
3564 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3565 GFP_NOFS);
3566 if (!page) {
3567 btrfs_err(device->fs_info,
3568 "couldn't get super block page for bytenr %llu",
3569 bytenr);
3570 errors++;
3571 continue;
3572 }
3573
3574 /* Bump the refcount for wait_dev_supers() */
3575 get_page(page);
3576
3577 disk_super = page_address(page);
3578 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3579
3580 /*
3581 * Directly use bios here instead of relying on the page cache
3582 * to do I/O, so we don't lose the ability to do integrity
3583 * checking.
3584 */
3585 bio = bio_alloc(GFP_NOFS, 1);
3586 bio_set_dev(bio, device->bdev);
3587 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3588 bio->bi_private = device;
3589 bio->bi_end_io = btrfs_end_super_write;
3590 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3591 offset_in_page(bytenr));
3592
3593 /*
3594 * We FUA only the first super block. The others we allow to
3595 * go down lazy and there's a short window where the on-disk
3596 * copies might still contain the older version.
3597 */
3598 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3599 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3600 bio->bi_opf |= REQ_FUA;
3601
3602 btrfsic_submit_bio(bio);
3603 }
3604 return errors < i ? 0 : -1;
3605}
3606
3607/*
3608 * Wait for write completion of superblocks done by write_dev_supers,
3609 * @max_mirrors same for write and wait phases.
3610 *
3611 * Return number of errors when page is not found or not marked up to
3612 * date.
3613 */
3614static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3615{
3616 int i;
3617 int errors = 0;
3618 bool primary_failed = false;
3619 u64 bytenr;
3620
3621 if (max_mirrors == 0)
3622 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3623
3624 for (i = 0; i < max_mirrors; i++) {
3625 struct page *page;
3626
3627 bytenr = btrfs_sb_offset(i);
3628 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3629 device->commit_total_bytes)
3630 break;
3631
3632 page = find_get_page(device->bdev->bd_inode->i_mapping,
3633 bytenr >> PAGE_SHIFT);
3634 if (!page) {
3635 errors++;
3636 if (i == 0)
3637 primary_failed = true;
3638 continue;
3639 }
3640 /* Page is submitted locked and unlocked once the IO completes */
3641 wait_on_page_locked(page);
3642 if (PageError(page)) {
3643 errors++;
3644 if (i == 0)
3645 primary_failed = true;
3646 }
3647
3648 /* Drop our reference */
3649 put_page(page);
3650
3651 /* Drop the reference from the writing run */
3652 put_page(page);
3653 }
3654
3655 /* log error, force error return */
3656 if (primary_failed) {
3657 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3658 device->devid);
3659 return -1;
3660 }
3661
3662 return errors < i ? 0 : -1;
3663}
3664
3665/*
3666 * endio for the write_dev_flush, this will wake anyone waiting
3667 * for the barrier when it is done
3668 */
3669static void btrfs_end_empty_barrier(struct bio *bio)
3670{
3671 complete(bio->bi_private);
3672}
3673
3674/*
3675 * Submit a flush request to the device if it supports it. Error handling is
3676 * done in the waiting counterpart.
3677 */
3678static void write_dev_flush(struct btrfs_device *device)
3679{
3680 struct request_queue *q = bdev_get_queue(device->bdev);
3681 struct bio *bio = device->flush_bio;
3682
3683 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3684 return;
3685
3686 bio_reset(bio);
3687 bio->bi_end_io = btrfs_end_empty_barrier;
3688 bio_set_dev(bio, device->bdev);
3689 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3690 init_completion(&device->flush_wait);
3691 bio->bi_private = &device->flush_wait;
3692
3693 btrfsic_submit_bio(bio);
3694 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3695}
3696
3697/*
3698 * If the flush bio has been submitted by write_dev_flush, wait for it.
3699 */
3700static blk_status_t wait_dev_flush(struct btrfs_device *device)
3701{
3702 struct bio *bio = device->flush_bio;
3703
3704 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3705 return BLK_STS_OK;
3706
3707 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3708 wait_for_completion_io(&device->flush_wait);
3709
3710 return bio->bi_status;
3711}
3712
3713static int check_barrier_error(struct btrfs_fs_info *fs_info)
3714{
3715 if (!btrfs_check_rw_degradable(fs_info, NULL))
3716 return -EIO;
3717 return 0;
3718}
3719
3720/*
3721 * send an empty flush down to each device in parallel,
3722 * then wait for them
3723 */
3724static int barrier_all_devices(struct btrfs_fs_info *info)
3725{
3726 struct list_head *head;
3727 struct btrfs_device *dev;
3728 int errors_wait = 0;
3729 blk_status_t ret;
3730
3731 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3732 /* send down all the barriers */
3733 head = &info->fs_devices->devices;
3734 list_for_each_entry(dev, head, dev_list) {
3735 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3736 continue;
3737 if (!dev->bdev)
3738 continue;
3739 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3740 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3741 continue;
3742
3743 write_dev_flush(dev);
3744 dev->last_flush_error = BLK_STS_OK;
3745 }
3746
3747 /* wait for all the barriers */
3748 list_for_each_entry(dev, head, dev_list) {
3749 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3750 continue;
3751 if (!dev->bdev) {
3752 errors_wait++;
3753 continue;
3754 }
3755 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3756 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3757 continue;
3758
3759 ret = wait_dev_flush(dev);
3760 if (ret) {
3761 dev->last_flush_error = ret;
3762 btrfs_dev_stat_inc_and_print(dev,
3763 BTRFS_DEV_STAT_FLUSH_ERRS);
3764 errors_wait++;
3765 }
3766 }
3767
3768 if (errors_wait) {
3769 /*
3770 * At some point we need the status of all disks
3771 * to arrive at the volume status. So error checking
3772 * is being pushed to a separate loop.
3773 */
3774 return check_barrier_error(info);
3775 }
3776 return 0;
3777}
3778
3779int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3780{
3781 int raid_type;
3782 int min_tolerated = INT_MAX;
3783
3784 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3785 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3786 min_tolerated = min_t(int, min_tolerated,
3787 btrfs_raid_array[BTRFS_RAID_SINGLE].
3788 tolerated_failures);
3789
3790 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3791 if (raid_type == BTRFS_RAID_SINGLE)
3792 continue;
3793 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3794 continue;
3795 min_tolerated = min_t(int, min_tolerated,
3796 btrfs_raid_array[raid_type].
3797 tolerated_failures);
3798 }
3799
3800 if (min_tolerated == INT_MAX) {
3801 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3802 min_tolerated = 0;
3803 }
3804
3805 return min_tolerated;
3806}
3807
3808int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3809{
3810 struct list_head *head;
3811 struct btrfs_device *dev;
3812 struct btrfs_super_block *sb;
3813 struct btrfs_dev_item *dev_item;
3814 int ret;
3815 int do_barriers;
3816 int max_errors;
3817 int total_errors = 0;
3818 u64 flags;
3819
3820 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3821
3822 /*
3823 * max_mirrors == 0 indicates we're from commit_transaction,
3824 * not from fsync where the tree roots in fs_info have not
3825 * been consistent on disk.
3826 */
3827 if (max_mirrors == 0)
3828 backup_super_roots(fs_info);
3829
3830 sb = fs_info->super_for_commit;
3831 dev_item = &sb->dev_item;
3832
3833 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3834 head = &fs_info->fs_devices->devices;
3835 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3836
3837 if (do_barriers) {
3838 ret = barrier_all_devices(fs_info);
3839 if (ret) {
3840 mutex_unlock(
3841 &fs_info->fs_devices->device_list_mutex);
3842 btrfs_handle_fs_error(fs_info, ret,
3843 "errors while submitting device barriers.");
3844 return ret;
3845 }
3846 }
3847
3848 list_for_each_entry(dev, head, dev_list) {
3849 if (!dev->bdev) {
3850 total_errors++;
3851 continue;
3852 }
3853 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3854 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3855 continue;
3856
3857 btrfs_set_stack_device_generation(dev_item, 0);
3858 btrfs_set_stack_device_type(dev_item, dev->type);
3859 btrfs_set_stack_device_id(dev_item, dev->devid);
3860 btrfs_set_stack_device_total_bytes(dev_item,
3861 dev->commit_total_bytes);
3862 btrfs_set_stack_device_bytes_used(dev_item,
3863 dev->commit_bytes_used);
3864 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3865 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3866 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3867 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3868 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3869 BTRFS_FSID_SIZE);
3870
3871 flags = btrfs_super_flags(sb);
3872 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3873
3874 ret = btrfs_validate_write_super(fs_info, sb);
3875 if (ret < 0) {
3876 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3877 btrfs_handle_fs_error(fs_info, -EUCLEAN,
3878 "unexpected superblock corruption detected");
3879 return -EUCLEAN;
3880 }
3881
3882 ret = write_dev_supers(dev, sb, max_mirrors);
3883 if (ret)
3884 total_errors++;
3885 }
3886 if (total_errors > max_errors) {
3887 btrfs_err(fs_info, "%d errors while writing supers",
3888 total_errors);
3889 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3890
3891 /* FUA is masked off if unsupported and can't be the reason */
3892 btrfs_handle_fs_error(fs_info, -EIO,
3893 "%d errors while writing supers",
3894 total_errors);
3895 return -EIO;
3896 }
3897
3898 total_errors = 0;
3899 list_for_each_entry(dev, head, dev_list) {
3900 if (!dev->bdev)
3901 continue;
3902 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3903 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3904 continue;
3905
3906 ret = wait_dev_supers(dev, max_mirrors);
3907 if (ret)
3908 total_errors++;
3909 }
3910 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3911 if (total_errors > max_errors) {
3912 btrfs_handle_fs_error(fs_info, -EIO,
3913 "%d errors while writing supers",
3914 total_errors);
3915 return -EIO;
3916 }
3917 return 0;
3918}
3919
3920/* Drop a fs root from the radix tree and free it. */
3921void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3922 struct btrfs_root *root)
3923{
3924 bool drop_ref = false;
3925
3926 spin_lock(&fs_info->fs_roots_radix_lock);
3927 radix_tree_delete(&fs_info->fs_roots_radix,
3928 (unsigned long)root->root_key.objectid);
3929 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
3930 drop_ref = true;
3931 spin_unlock(&fs_info->fs_roots_radix_lock);
3932
3933 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3934 ASSERT(root->log_root == NULL);
3935 if (root->reloc_root) {
3936 btrfs_put_root(root->reloc_root);
3937 root->reloc_root = NULL;
3938 }
3939 }
3940
3941 if (root->free_ino_pinned)
3942 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3943 if (root->free_ino_ctl)
3944 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3945 if (root->ino_cache_inode) {
3946 iput(root->ino_cache_inode);
3947 root->ino_cache_inode = NULL;
3948 }
3949 if (drop_ref)
3950 btrfs_put_root(root);
3951}
3952
3953int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3954{
3955 u64 root_objectid = 0;
3956 struct btrfs_root *gang[8];
3957 int i = 0;
3958 int err = 0;
3959 unsigned int ret = 0;
3960
3961 while (1) {
3962 spin_lock(&fs_info->fs_roots_radix_lock);
3963 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3964 (void **)gang, root_objectid,
3965 ARRAY_SIZE(gang));
3966 if (!ret) {
3967 spin_unlock(&fs_info->fs_roots_radix_lock);
3968 break;
3969 }
3970 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3971
3972 for (i = 0; i < ret; i++) {
3973 /* Avoid to grab roots in dead_roots */
3974 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3975 gang[i] = NULL;
3976 continue;
3977 }
3978 /* grab all the search result for later use */
3979 gang[i] = btrfs_grab_root(gang[i]);
3980 }
3981 spin_unlock(&fs_info->fs_roots_radix_lock);
3982
3983 for (i = 0; i < ret; i++) {
3984 if (!gang[i])
3985 continue;
3986 root_objectid = gang[i]->root_key.objectid;
3987 err = btrfs_orphan_cleanup(gang[i]);
3988 if (err)
3989 break;
3990 btrfs_put_root(gang[i]);
3991 }
3992 root_objectid++;
3993 }
3994
3995 /* release the uncleaned roots due to error */
3996 for (; i < ret; i++) {
3997 if (gang[i])
3998 btrfs_put_root(gang[i]);
3999 }
4000 return err;
4001}
4002
4003int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4004{
4005 struct btrfs_root *root = fs_info->tree_root;
4006 struct btrfs_trans_handle *trans;
4007
4008 mutex_lock(&fs_info->cleaner_mutex);
4009 btrfs_run_delayed_iputs(fs_info);
4010 mutex_unlock(&fs_info->cleaner_mutex);
4011 wake_up_process(fs_info->cleaner_kthread);
4012
4013 /* wait until ongoing cleanup work done */
4014 down_write(&fs_info->cleanup_work_sem);
4015 up_write(&fs_info->cleanup_work_sem);
4016
4017 trans = btrfs_join_transaction(root);
4018 if (IS_ERR(trans))
4019 return PTR_ERR(trans);
4020 return btrfs_commit_transaction(trans);
4021}
4022
4023void __cold close_ctree(struct btrfs_fs_info *fs_info)
4024{
4025 int ret;
4026
4027 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4028 /*
4029 * We don't want the cleaner to start new transactions, add more delayed
4030 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4031 * because that frees the task_struct, and the transaction kthread might
4032 * still try to wake up the cleaner.
4033 */
4034 kthread_park(fs_info->cleaner_kthread);
4035
4036 /* wait for the qgroup rescan worker to stop */
4037 btrfs_qgroup_wait_for_completion(fs_info, false);
4038
4039 /* wait for the uuid_scan task to finish */
4040 down(&fs_info->uuid_tree_rescan_sem);
4041 /* avoid complains from lockdep et al., set sem back to initial state */
4042 up(&fs_info->uuid_tree_rescan_sem);
4043
4044 /* pause restriper - we want to resume on mount */
4045 btrfs_pause_balance(fs_info);
4046
4047 btrfs_dev_replace_suspend_for_unmount(fs_info);
4048
4049 btrfs_scrub_cancel(fs_info);
4050
4051 /* wait for any defraggers to finish */
4052 wait_event(fs_info->transaction_wait,
4053 (atomic_read(&fs_info->defrag_running) == 0));
4054
4055 /* clear out the rbtree of defraggable inodes */
4056 btrfs_cleanup_defrag_inodes(fs_info);
4057
4058 cancel_work_sync(&fs_info->async_reclaim_work);
4059
4060 /* Cancel or finish ongoing discard work */
4061 btrfs_discard_cleanup(fs_info);
4062
4063 if (!sb_rdonly(fs_info->sb)) {
4064 /*
4065 * The cleaner kthread is stopped, so do one final pass over
4066 * unused block groups.
4067 */
4068 btrfs_delete_unused_bgs(fs_info);
4069
4070 /*
4071 * There might be existing delayed inode workers still running
4072 * and holding an empty delayed inode item. We must wait for
4073 * them to complete first because they can create a transaction.
4074 * This happens when someone calls btrfs_balance_delayed_items()
4075 * and then a transaction commit runs the same delayed nodes
4076 * before any delayed worker has done something with the nodes.
4077 * We must wait for any worker here and not at transaction
4078 * commit time since that could cause a deadlock.
4079 * This is a very rare case.
4080 */
4081 btrfs_flush_workqueue(fs_info->delayed_workers);
4082
4083 ret = btrfs_commit_super(fs_info);
4084 if (ret)
4085 btrfs_err(fs_info, "commit super ret %d", ret);
4086 }
4087
4088 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4089 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4090 btrfs_error_commit_super(fs_info);
4091
4092 kthread_stop(fs_info->transaction_kthread);
4093 kthread_stop(fs_info->cleaner_kthread);
4094
4095 ASSERT(list_empty(&fs_info->delayed_iputs));
4096 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4097
4098 if (btrfs_check_quota_leak(fs_info)) {
4099 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4100 btrfs_err(fs_info, "qgroup reserved space leaked");
4101 }
4102
4103 btrfs_free_qgroup_config(fs_info);
4104 ASSERT(list_empty(&fs_info->delalloc_roots));
4105
4106 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4107 btrfs_info(fs_info, "at unmount delalloc count %lld",
4108 percpu_counter_sum(&fs_info->delalloc_bytes));
4109 }
4110
4111 if (percpu_counter_sum(&fs_info->dio_bytes))
4112 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4113 percpu_counter_sum(&fs_info->dio_bytes));
4114
4115 btrfs_sysfs_remove_mounted(fs_info);
4116 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4117
4118 btrfs_put_block_group_cache(fs_info);
4119
4120 /*
4121 * we must make sure there is not any read request to
4122 * submit after we stopping all workers.
4123 */
4124 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4125 btrfs_stop_all_workers(fs_info);
4126
4127 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4128 free_root_pointers(fs_info, true);
4129 btrfs_free_fs_roots(fs_info);
4130
4131 /*
4132 * We must free the block groups after dropping the fs_roots as we could
4133 * have had an IO error and have left over tree log blocks that aren't
4134 * cleaned up until the fs roots are freed. This makes the block group
4135 * accounting appear to be wrong because there's pending reserved bytes,
4136 * so make sure we do the block group cleanup afterwards.
4137 */
4138 btrfs_free_block_groups(fs_info);
4139
4140 iput(fs_info->btree_inode);
4141
4142#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4143 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4144 btrfsic_unmount(fs_info->fs_devices);
4145#endif
4146
4147 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4148 btrfs_close_devices(fs_info->fs_devices);
4149}
4150
4151int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4152 int atomic)
4153{
4154 int ret;
4155 struct inode *btree_inode = buf->pages[0]->mapping->host;
4156
4157 ret = extent_buffer_uptodate(buf);
4158 if (!ret)
4159 return ret;
4160
4161 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4162 parent_transid, atomic);
4163 if (ret == -EAGAIN)
4164 return ret;
4165 return !ret;
4166}
4167
4168void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4169{
4170 struct btrfs_fs_info *fs_info;
4171 struct btrfs_root *root;
4172 u64 transid = btrfs_header_generation(buf);
4173 int was_dirty;
4174
4175#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4176 /*
4177 * This is a fast path so only do this check if we have sanity tests
4178 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4179 * outside of the sanity tests.
4180 */
4181 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4182 return;
4183#endif
4184 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4185 fs_info = root->fs_info;
4186 btrfs_assert_tree_locked(buf);
4187 if (transid != fs_info->generation)
4188 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4189 buf->start, transid, fs_info->generation);
4190 was_dirty = set_extent_buffer_dirty(buf);
4191 if (!was_dirty)
4192 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4193 buf->len,
4194 fs_info->dirty_metadata_batch);
4195#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4196 /*
4197 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4198 * but item data not updated.
4199 * So here we should only check item pointers, not item data.
4200 */
4201 if (btrfs_header_level(buf) == 0 &&
4202 btrfs_check_leaf_relaxed(buf)) {
4203 btrfs_print_leaf(buf);
4204 ASSERT(0);
4205 }
4206#endif
4207}
4208
4209static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4210 int flush_delayed)
4211{
4212 /*
4213 * looks as though older kernels can get into trouble with
4214 * this code, they end up stuck in balance_dirty_pages forever
4215 */
4216 int ret;
4217
4218 if (current->flags & PF_MEMALLOC)
4219 return;
4220
4221 if (flush_delayed)
4222 btrfs_balance_delayed_items(fs_info);
4223
4224 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4225 BTRFS_DIRTY_METADATA_THRESH,
4226 fs_info->dirty_metadata_batch);
4227 if (ret > 0) {
4228 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4229 }
4230}
4231
4232void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4233{
4234 __btrfs_btree_balance_dirty(fs_info, 1);
4235}
4236
4237void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4238{
4239 __btrfs_btree_balance_dirty(fs_info, 0);
4240}
4241
4242int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4243 struct btrfs_key *first_key)
4244{
4245 return btree_read_extent_buffer_pages(buf, parent_transid,
4246 level, first_key);
4247}
4248
4249static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4250{
4251 /* cleanup FS via transaction */
4252 btrfs_cleanup_transaction(fs_info);
4253
4254 mutex_lock(&fs_info->cleaner_mutex);
4255 btrfs_run_delayed_iputs(fs_info);
4256 mutex_unlock(&fs_info->cleaner_mutex);
4257
4258 down_write(&fs_info->cleanup_work_sem);
4259 up_write(&fs_info->cleanup_work_sem);
4260}
4261
4262static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4263{
4264 struct btrfs_root *gang[8];
4265 u64 root_objectid = 0;
4266 int ret;
4267
4268 spin_lock(&fs_info->fs_roots_radix_lock);
4269 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4270 (void **)gang, root_objectid,
4271 ARRAY_SIZE(gang))) != 0) {
4272 int i;
4273
4274 for (i = 0; i < ret; i++)
4275 gang[i] = btrfs_grab_root(gang[i]);
4276 spin_unlock(&fs_info->fs_roots_radix_lock);
4277
4278 for (i = 0; i < ret; i++) {
4279 if (!gang[i])
4280 continue;
4281 root_objectid = gang[i]->root_key.objectid;
4282 btrfs_free_log(NULL, gang[i]);
4283 btrfs_put_root(gang[i]);
4284 }
4285 root_objectid++;
4286 spin_lock(&fs_info->fs_roots_radix_lock);
4287 }
4288 spin_unlock(&fs_info->fs_roots_radix_lock);
4289 btrfs_free_log_root_tree(NULL, fs_info);
4290}
4291
4292static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4293{
4294 struct btrfs_ordered_extent *ordered;
4295
4296 spin_lock(&root->ordered_extent_lock);
4297 /*
4298 * This will just short circuit the ordered completion stuff which will
4299 * make sure the ordered extent gets properly cleaned up.
4300 */
4301 list_for_each_entry(ordered, &root->ordered_extents,
4302 root_extent_list)
4303 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4304 spin_unlock(&root->ordered_extent_lock);
4305}
4306
4307static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4308{
4309 struct btrfs_root *root;
4310 struct list_head splice;
4311
4312 INIT_LIST_HEAD(&splice);
4313
4314 spin_lock(&fs_info->ordered_root_lock);
4315 list_splice_init(&fs_info->ordered_roots, &splice);
4316 while (!list_empty(&splice)) {
4317 root = list_first_entry(&splice, struct btrfs_root,
4318 ordered_root);
4319 list_move_tail(&root->ordered_root,
4320 &fs_info->ordered_roots);
4321
4322 spin_unlock(&fs_info->ordered_root_lock);
4323 btrfs_destroy_ordered_extents(root);
4324
4325 cond_resched();
4326 spin_lock(&fs_info->ordered_root_lock);
4327 }
4328 spin_unlock(&fs_info->ordered_root_lock);
4329
4330 /*
4331 * We need this here because if we've been flipped read-only we won't
4332 * get sync() from the umount, so we need to make sure any ordered
4333 * extents that haven't had their dirty pages IO start writeout yet
4334 * actually get run and error out properly.
4335 */
4336 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4337}
4338
4339static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4340 struct btrfs_fs_info *fs_info)
4341{
4342 struct rb_node *node;
4343 struct btrfs_delayed_ref_root *delayed_refs;
4344 struct btrfs_delayed_ref_node *ref;
4345 int ret = 0;
4346
4347 delayed_refs = &trans->delayed_refs;
4348
4349 spin_lock(&delayed_refs->lock);
4350 if (atomic_read(&delayed_refs->num_entries) == 0) {
4351 spin_unlock(&delayed_refs->lock);
4352 btrfs_debug(fs_info, "delayed_refs has NO entry");
4353 return ret;
4354 }
4355
4356 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4357 struct btrfs_delayed_ref_head *head;
4358 struct rb_node *n;
4359 bool pin_bytes = false;
4360
4361 head = rb_entry(node, struct btrfs_delayed_ref_head,
4362 href_node);
4363 if (btrfs_delayed_ref_lock(delayed_refs, head))
4364 continue;
4365
4366 spin_lock(&head->lock);
4367 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4368 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4369 ref_node);
4370 ref->in_tree = 0;
4371 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4372 RB_CLEAR_NODE(&ref->ref_node);
4373 if (!list_empty(&ref->add_list))
4374 list_del(&ref->add_list);
4375 atomic_dec(&delayed_refs->num_entries);
4376 btrfs_put_delayed_ref(ref);
4377 }
4378 if (head->must_insert_reserved)
4379 pin_bytes = true;
4380 btrfs_free_delayed_extent_op(head->extent_op);
4381 btrfs_delete_ref_head(delayed_refs, head);
4382 spin_unlock(&head->lock);
4383 spin_unlock(&delayed_refs->lock);
4384 mutex_unlock(&head->mutex);
4385
4386 if (pin_bytes) {
4387 struct btrfs_block_group *cache;
4388
4389 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4390 BUG_ON(!cache);
4391
4392 spin_lock(&cache->space_info->lock);
4393 spin_lock(&cache->lock);
4394 cache->pinned += head->num_bytes;
4395 btrfs_space_info_update_bytes_pinned(fs_info,
4396 cache->space_info, head->num_bytes);
4397 cache->reserved -= head->num_bytes;
4398 cache->space_info->bytes_reserved -= head->num_bytes;
4399 spin_unlock(&cache->lock);
4400 spin_unlock(&cache->space_info->lock);
4401 percpu_counter_add_batch(
4402 &cache->space_info->total_bytes_pinned,
4403 head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
4404
4405 btrfs_put_block_group(cache);
4406
4407 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4408 head->bytenr + head->num_bytes - 1);
4409 }
4410 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4411 btrfs_put_delayed_ref_head(head);
4412 cond_resched();
4413 spin_lock(&delayed_refs->lock);
4414 }
4415 btrfs_qgroup_destroy_extent_records(trans);
4416
4417 spin_unlock(&delayed_refs->lock);
4418
4419 return ret;
4420}
4421
4422static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4423{
4424 struct btrfs_inode *btrfs_inode;
4425 struct list_head splice;
4426
4427 INIT_LIST_HEAD(&splice);
4428
4429 spin_lock(&root->delalloc_lock);
4430 list_splice_init(&root->delalloc_inodes, &splice);
4431
4432 while (!list_empty(&splice)) {
4433 struct inode *inode = NULL;
4434 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4435 delalloc_inodes);
4436 __btrfs_del_delalloc_inode(root, btrfs_inode);
4437 spin_unlock(&root->delalloc_lock);
4438
4439 /*
4440 * Make sure we get a live inode and that it'll not disappear
4441 * meanwhile.
4442 */
4443 inode = igrab(&btrfs_inode->vfs_inode);
4444 if (inode) {
4445 invalidate_inode_pages2(inode->i_mapping);
4446 iput(inode);
4447 }
4448 spin_lock(&root->delalloc_lock);
4449 }
4450 spin_unlock(&root->delalloc_lock);
4451}
4452
4453static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4454{
4455 struct btrfs_root *root;
4456 struct list_head splice;
4457
4458 INIT_LIST_HEAD(&splice);
4459
4460 spin_lock(&fs_info->delalloc_root_lock);
4461 list_splice_init(&fs_info->delalloc_roots, &splice);
4462 while (!list_empty(&splice)) {
4463 root = list_first_entry(&splice, struct btrfs_root,
4464 delalloc_root);
4465 root = btrfs_grab_root(root);
4466 BUG_ON(!root);
4467 spin_unlock(&fs_info->delalloc_root_lock);
4468
4469 btrfs_destroy_delalloc_inodes(root);
4470 btrfs_put_root(root);
4471
4472 spin_lock(&fs_info->delalloc_root_lock);
4473 }
4474 spin_unlock(&fs_info->delalloc_root_lock);
4475}
4476
4477static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4478 struct extent_io_tree *dirty_pages,
4479 int mark)
4480{
4481 int ret;
4482 struct extent_buffer *eb;
4483 u64 start = 0;
4484 u64 end;
4485
4486 while (1) {
4487 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4488 mark, NULL);
4489 if (ret)
4490 break;
4491
4492 clear_extent_bits(dirty_pages, start, end, mark);
4493 while (start <= end) {
4494 eb = find_extent_buffer(fs_info, start);
4495 start += fs_info->nodesize;
4496 if (!eb)
4497 continue;
4498 wait_on_extent_buffer_writeback(eb);
4499
4500 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4501 &eb->bflags))
4502 clear_extent_buffer_dirty(eb);
4503 free_extent_buffer_stale(eb);
4504 }
4505 }
4506
4507 return ret;
4508}
4509
4510static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4511 struct extent_io_tree *unpin)
4512{
4513 u64 start;
4514 u64 end;
4515 int ret;
4516
4517 while (1) {
4518 struct extent_state *cached_state = NULL;
4519
4520 /*
4521 * The btrfs_finish_extent_commit() may get the same range as
4522 * ours between find_first_extent_bit and clear_extent_dirty.
4523 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4524 * the same extent range.
4525 */
4526 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4527 ret = find_first_extent_bit(unpin, 0, &start, &end,
4528 EXTENT_DIRTY, &cached_state);
4529 if (ret) {
4530 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4531 break;
4532 }
4533
4534 clear_extent_dirty(unpin, start, end, &cached_state);
4535 free_extent_state(cached_state);
4536 btrfs_error_unpin_extent_range(fs_info, start, end);
4537 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4538 cond_resched();
4539 }
4540
4541 return 0;
4542}
4543
4544static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4545{
4546 struct inode *inode;
4547
4548 inode = cache->io_ctl.inode;
4549 if (inode) {
4550 invalidate_inode_pages2(inode->i_mapping);
4551 BTRFS_I(inode)->generation = 0;
4552 cache->io_ctl.inode = NULL;
4553 iput(inode);
4554 }
4555 ASSERT(cache->io_ctl.pages == NULL);
4556 btrfs_put_block_group(cache);
4557}
4558
4559void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4560 struct btrfs_fs_info *fs_info)
4561{
4562 struct btrfs_block_group *cache;
4563
4564 spin_lock(&cur_trans->dirty_bgs_lock);
4565 while (!list_empty(&cur_trans->dirty_bgs)) {
4566 cache = list_first_entry(&cur_trans->dirty_bgs,
4567 struct btrfs_block_group,
4568 dirty_list);
4569
4570 if (!list_empty(&cache->io_list)) {
4571 spin_unlock(&cur_trans->dirty_bgs_lock);
4572 list_del_init(&cache->io_list);
4573 btrfs_cleanup_bg_io(cache);
4574 spin_lock(&cur_trans->dirty_bgs_lock);
4575 }
4576
4577 list_del_init(&cache->dirty_list);
4578 spin_lock(&cache->lock);
4579 cache->disk_cache_state = BTRFS_DC_ERROR;
4580 spin_unlock(&cache->lock);
4581
4582 spin_unlock(&cur_trans->dirty_bgs_lock);
4583 btrfs_put_block_group(cache);
4584 btrfs_delayed_refs_rsv_release(fs_info, 1);
4585 spin_lock(&cur_trans->dirty_bgs_lock);
4586 }
4587 spin_unlock(&cur_trans->dirty_bgs_lock);
4588
4589 /*
4590 * Refer to the definition of io_bgs member for details why it's safe
4591 * to use it without any locking
4592 */
4593 while (!list_empty(&cur_trans->io_bgs)) {
4594 cache = list_first_entry(&cur_trans->io_bgs,
4595 struct btrfs_block_group,
4596 io_list);
4597
4598 list_del_init(&cache->io_list);
4599 spin_lock(&cache->lock);
4600 cache->disk_cache_state = BTRFS_DC_ERROR;
4601 spin_unlock(&cache->lock);
4602 btrfs_cleanup_bg_io(cache);
4603 }
4604}
4605
4606void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4607 struct btrfs_fs_info *fs_info)
4608{
4609 struct btrfs_device *dev, *tmp;
4610
4611 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4612 ASSERT(list_empty(&cur_trans->dirty_bgs));
4613 ASSERT(list_empty(&cur_trans->io_bgs));
4614
4615 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4616 post_commit_list) {
4617 list_del_init(&dev->post_commit_list);
4618 }
4619
4620 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4621
4622 cur_trans->state = TRANS_STATE_COMMIT_START;
4623 wake_up(&fs_info->transaction_blocked_wait);
4624
4625 cur_trans->state = TRANS_STATE_UNBLOCKED;
4626 wake_up(&fs_info->transaction_wait);
4627
4628 btrfs_destroy_delayed_inodes(fs_info);
4629
4630 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4631 EXTENT_DIRTY);
4632 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4633
4634 cur_trans->state =TRANS_STATE_COMPLETED;
4635 wake_up(&cur_trans->commit_wait);
4636}
4637
4638static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4639{
4640 struct btrfs_transaction *t;
4641
4642 mutex_lock(&fs_info->transaction_kthread_mutex);
4643
4644 spin_lock(&fs_info->trans_lock);
4645 while (!list_empty(&fs_info->trans_list)) {
4646 t = list_first_entry(&fs_info->trans_list,
4647 struct btrfs_transaction, list);
4648 if (t->state >= TRANS_STATE_COMMIT_START) {
4649 refcount_inc(&t->use_count);
4650 spin_unlock(&fs_info->trans_lock);
4651 btrfs_wait_for_commit(fs_info, t->transid);
4652 btrfs_put_transaction(t);
4653 spin_lock(&fs_info->trans_lock);
4654 continue;
4655 }
4656 if (t == fs_info->running_transaction) {
4657 t->state = TRANS_STATE_COMMIT_DOING;
4658 spin_unlock(&fs_info->trans_lock);
4659 /*
4660 * We wait for 0 num_writers since we don't hold a trans
4661 * handle open currently for this transaction.
4662 */
4663 wait_event(t->writer_wait,
4664 atomic_read(&t->num_writers) == 0);
4665 } else {
4666 spin_unlock(&fs_info->trans_lock);
4667 }
4668 btrfs_cleanup_one_transaction(t, fs_info);
4669
4670 spin_lock(&fs_info->trans_lock);
4671 if (t == fs_info->running_transaction)
4672 fs_info->running_transaction = NULL;
4673 list_del_init(&t->list);
4674 spin_unlock(&fs_info->trans_lock);
4675
4676 btrfs_put_transaction(t);
4677 trace_btrfs_transaction_commit(fs_info->tree_root);
4678 spin_lock(&fs_info->trans_lock);
4679 }
4680 spin_unlock(&fs_info->trans_lock);
4681 btrfs_destroy_all_ordered_extents(fs_info);
4682 btrfs_destroy_delayed_inodes(fs_info);
4683 btrfs_assert_delayed_root_empty(fs_info);
4684 btrfs_destroy_all_delalloc_inodes(fs_info);
4685 btrfs_drop_all_logs(fs_info);
4686 mutex_unlock(&fs_info->transaction_kthread_mutex);
4687
4688 return 0;
4689}
4690
4691static const struct extent_io_ops btree_extent_io_ops = {
4692 /* mandatory callbacks */
4693 .submit_bio_hook = btree_submit_bio_hook,
4694 .readpage_end_io_hook = btree_readpage_end_io_hook,
4695};
1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/fs.h>
20#include <linux/blkdev.h>
21#include <linux/scatterlist.h>
22#include <linux/swap.h>
23#include <linux/radix-tree.h>
24#include <linux/writeback.h>
25#include <linux/buffer_head.h>
26#include <linux/workqueue.h>
27#include <linux/kthread.h>
28#include <linux/freezer.h>
29#include <linux/crc32c.h>
30#include <linux/slab.h>
31#include <linux/migrate.h>
32#include <linux/ratelimit.h>
33#include <asm/unaligned.h>
34#include "compat.h"
35#include "ctree.h"
36#include "disk-io.h"
37#include "transaction.h"
38#include "btrfs_inode.h"
39#include "volumes.h"
40#include "print-tree.h"
41#include "async-thread.h"
42#include "locking.h"
43#include "tree-log.h"
44#include "free-space-cache.h"
45#include "inode-map.h"
46#include "check-integrity.h"
47#include "rcu-string.h"
48
49static struct extent_io_ops btree_extent_io_ops;
50static void end_workqueue_fn(struct btrfs_work *work);
51static void free_fs_root(struct btrfs_root *root);
52static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
53 int read_only);
54static void btrfs_destroy_ordered_operations(struct btrfs_root *root);
55static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
56static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
57 struct btrfs_root *root);
58static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
59static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
60static int btrfs_destroy_marked_extents(struct btrfs_root *root,
61 struct extent_io_tree *dirty_pages,
62 int mark);
63static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
64 struct extent_io_tree *pinned_extents);
65
66/*
67 * end_io_wq structs are used to do processing in task context when an IO is
68 * complete. This is used during reads to verify checksums, and it is used
69 * by writes to insert metadata for new file extents after IO is complete.
70 */
71struct end_io_wq {
72 struct bio *bio;
73 bio_end_io_t *end_io;
74 void *private;
75 struct btrfs_fs_info *info;
76 int error;
77 int metadata;
78 struct list_head list;
79 struct btrfs_work work;
80};
81
82/*
83 * async submit bios are used to offload expensive checksumming
84 * onto the worker threads. They checksum file and metadata bios
85 * just before they are sent down the IO stack.
86 */
87struct async_submit_bio {
88 struct inode *inode;
89 struct bio *bio;
90 struct list_head list;
91 extent_submit_bio_hook_t *submit_bio_start;
92 extent_submit_bio_hook_t *submit_bio_done;
93 int rw;
94 int mirror_num;
95 unsigned long bio_flags;
96 /*
97 * bio_offset is optional, can be used if the pages in the bio
98 * can't tell us where in the file the bio should go
99 */
100 u64 bio_offset;
101 struct btrfs_work work;
102 int error;
103};
104
105/*
106 * Lockdep class keys for extent_buffer->lock's in this root. For a given
107 * eb, the lockdep key is determined by the btrfs_root it belongs to and
108 * the level the eb occupies in the tree.
109 *
110 * Different roots are used for different purposes and may nest inside each
111 * other and they require separate keysets. As lockdep keys should be
112 * static, assign keysets according to the purpose of the root as indicated
113 * by btrfs_root->objectid. This ensures that all special purpose roots
114 * have separate keysets.
115 *
116 * Lock-nesting across peer nodes is always done with the immediate parent
117 * node locked thus preventing deadlock. As lockdep doesn't know this, use
118 * subclass to avoid triggering lockdep warning in such cases.
119 *
120 * The key is set by the readpage_end_io_hook after the buffer has passed
121 * csum validation but before the pages are unlocked. It is also set by
122 * btrfs_init_new_buffer on freshly allocated blocks.
123 *
124 * We also add a check to make sure the highest level of the tree is the
125 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
126 * needs update as well.
127 */
128#ifdef CONFIG_DEBUG_LOCK_ALLOC
129# if BTRFS_MAX_LEVEL != 8
130# error
131# endif
132
133static struct btrfs_lockdep_keyset {
134 u64 id; /* root objectid */
135 const char *name_stem; /* lock name stem */
136 char names[BTRFS_MAX_LEVEL + 1][20];
137 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
138} btrfs_lockdep_keysets[] = {
139 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
140 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
141 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
142 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
143 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
144 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
145 { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
146 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
147 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
148 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
149 { .id = 0, .name_stem = "tree" },
150};
151
152void __init btrfs_init_lockdep(void)
153{
154 int i, j;
155
156 /* initialize lockdep class names */
157 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
158 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
159
160 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
161 snprintf(ks->names[j], sizeof(ks->names[j]),
162 "btrfs-%s-%02d", ks->name_stem, j);
163 }
164}
165
166void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
167 int level)
168{
169 struct btrfs_lockdep_keyset *ks;
170
171 BUG_ON(level >= ARRAY_SIZE(ks->keys));
172
173 /* find the matching keyset, id 0 is the default entry */
174 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
175 if (ks->id == objectid)
176 break;
177
178 lockdep_set_class_and_name(&eb->lock,
179 &ks->keys[level], ks->names[level]);
180}
181
182#endif
183
184/*
185 * extents on the btree inode are pretty simple, there's one extent
186 * that covers the entire device
187 */
188static struct extent_map *btree_get_extent(struct inode *inode,
189 struct page *page, size_t pg_offset, u64 start, u64 len,
190 int create)
191{
192 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
193 struct extent_map *em;
194 int ret;
195
196 read_lock(&em_tree->lock);
197 em = lookup_extent_mapping(em_tree, start, len);
198 if (em) {
199 em->bdev =
200 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
201 read_unlock(&em_tree->lock);
202 goto out;
203 }
204 read_unlock(&em_tree->lock);
205
206 em = alloc_extent_map();
207 if (!em) {
208 em = ERR_PTR(-ENOMEM);
209 goto out;
210 }
211 em->start = 0;
212 em->len = (u64)-1;
213 em->block_len = (u64)-1;
214 em->block_start = 0;
215 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
216
217 write_lock(&em_tree->lock);
218 ret = add_extent_mapping(em_tree, em);
219 if (ret == -EEXIST) {
220 u64 failed_start = em->start;
221 u64 failed_len = em->len;
222
223 free_extent_map(em);
224 em = lookup_extent_mapping(em_tree, start, len);
225 if (em) {
226 ret = 0;
227 } else {
228 em = lookup_extent_mapping(em_tree, failed_start,
229 failed_len);
230 ret = -EIO;
231 }
232 } else if (ret) {
233 free_extent_map(em);
234 em = NULL;
235 }
236 write_unlock(&em_tree->lock);
237
238 if (ret)
239 em = ERR_PTR(ret);
240out:
241 return em;
242}
243
244u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
245{
246 return crc32c(seed, data, len);
247}
248
249void btrfs_csum_final(u32 crc, char *result)
250{
251 put_unaligned_le32(~crc, result);
252}
253
254/*
255 * compute the csum for a btree block, and either verify it or write it
256 * into the csum field of the block.
257 */
258static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
259 int verify)
260{
261 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
262 char *result = NULL;
263 unsigned long len;
264 unsigned long cur_len;
265 unsigned long offset = BTRFS_CSUM_SIZE;
266 char *kaddr;
267 unsigned long map_start;
268 unsigned long map_len;
269 int err;
270 u32 crc = ~(u32)0;
271 unsigned long inline_result;
272
273 len = buf->len - offset;
274 while (len > 0) {
275 err = map_private_extent_buffer(buf, offset, 32,
276 &kaddr, &map_start, &map_len);
277 if (err)
278 return 1;
279 cur_len = min(len, map_len - (offset - map_start));
280 crc = btrfs_csum_data(root, kaddr + offset - map_start,
281 crc, cur_len);
282 len -= cur_len;
283 offset += cur_len;
284 }
285 if (csum_size > sizeof(inline_result)) {
286 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
287 if (!result)
288 return 1;
289 } else {
290 result = (char *)&inline_result;
291 }
292
293 btrfs_csum_final(crc, result);
294
295 if (verify) {
296 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
297 u32 val;
298 u32 found = 0;
299 memcpy(&found, result, csum_size);
300
301 read_extent_buffer(buf, &val, 0, csum_size);
302 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
303 "failed on %llu wanted %X found %X "
304 "level %d\n",
305 root->fs_info->sb->s_id,
306 (unsigned long long)buf->start, val, found,
307 btrfs_header_level(buf));
308 if (result != (char *)&inline_result)
309 kfree(result);
310 return 1;
311 }
312 } else {
313 write_extent_buffer(buf, result, 0, csum_size);
314 }
315 if (result != (char *)&inline_result)
316 kfree(result);
317 return 0;
318}
319
320/*
321 * we can't consider a given block up to date unless the transid of the
322 * block matches the transid in the parent node's pointer. This is how we
323 * detect blocks that either didn't get written at all or got written
324 * in the wrong place.
325 */
326static int verify_parent_transid(struct extent_io_tree *io_tree,
327 struct extent_buffer *eb, u64 parent_transid,
328 int atomic)
329{
330 struct extent_state *cached_state = NULL;
331 int ret;
332
333 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
334 return 0;
335
336 if (atomic)
337 return -EAGAIN;
338
339 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
340 0, &cached_state);
341 if (extent_buffer_uptodate(eb) &&
342 btrfs_header_generation(eb) == parent_transid) {
343 ret = 0;
344 goto out;
345 }
346 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
347 "found %llu\n",
348 (unsigned long long)eb->start,
349 (unsigned long long)parent_transid,
350 (unsigned long long)btrfs_header_generation(eb));
351 ret = 1;
352 clear_extent_buffer_uptodate(eb);
353out:
354 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
355 &cached_state, GFP_NOFS);
356 return ret;
357}
358
359/*
360 * helper to read a given tree block, doing retries as required when
361 * the checksums don't match and we have alternate mirrors to try.
362 */
363static int btree_read_extent_buffer_pages(struct btrfs_root *root,
364 struct extent_buffer *eb,
365 u64 start, u64 parent_transid)
366{
367 struct extent_io_tree *io_tree;
368 int failed = 0;
369 int ret;
370 int num_copies = 0;
371 int mirror_num = 0;
372 int failed_mirror = 0;
373
374 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
375 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
376 while (1) {
377 ret = read_extent_buffer_pages(io_tree, eb, start,
378 WAIT_COMPLETE,
379 btree_get_extent, mirror_num);
380 if (!ret && !verify_parent_transid(io_tree, eb,
381 parent_transid, 0))
382 break;
383
384 /*
385 * This buffer's crc is fine, but its contents are corrupted, so
386 * there is no reason to read the other copies, they won't be
387 * any less wrong.
388 */
389 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
390 break;
391
392 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
393 eb->start, eb->len);
394 if (num_copies == 1)
395 break;
396
397 if (!failed_mirror) {
398 failed = 1;
399 failed_mirror = eb->read_mirror;
400 }
401
402 mirror_num++;
403 if (mirror_num == failed_mirror)
404 mirror_num++;
405
406 if (mirror_num > num_copies)
407 break;
408 }
409
410 if (failed && !ret)
411 repair_eb_io_failure(root, eb, failed_mirror);
412
413 return ret;
414}
415
416/*
417 * checksum a dirty tree block before IO. This has extra checks to make sure
418 * we only fill in the checksum field in the first page of a multi-page block
419 */
420
421static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
422{
423 struct extent_io_tree *tree;
424 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
425 u64 found_start;
426 struct extent_buffer *eb;
427
428 tree = &BTRFS_I(page->mapping->host)->io_tree;
429
430 eb = (struct extent_buffer *)page->private;
431 if (page != eb->pages[0])
432 return 0;
433 found_start = btrfs_header_bytenr(eb);
434 if (found_start != start) {
435 WARN_ON(1);
436 return 0;
437 }
438 if (eb->pages[0] != page) {
439 WARN_ON(1);
440 return 0;
441 }
442 if (!PageUptodate(page)) {
443 WARN_ON(1);
444 return 0;
445 }
446 csum_tree_block(root, eb, 0);
447 return 0;
448}
449
450static int check_tree_block_fsid(struct btrfs_root *root,
451 struct extent_buffer *eb)
452{
453 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
454 u8 fsid[BTRFS_UUID_SIZE];
455 int ret = 1;
456
457 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
458 BTRFS_FSID_SIZE);
459 while (fs_devices) {
460 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
461 ret = 0;
462 break;
463 }
464 fs_devices = fs_devices->seed;
465 }
466 return ret;
467}
468
469#define CORRUPT(reason, eb, root, slot) \
470 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
471 "root=%llu, slot=%d\n", reason, \
472 (unsigned long long)btrfs_header_bytenr(eb), \
473 (unsigned long long)root->objectid, slot)
474
475static noinline int check_leaf(struct btrfs_root *root,
476 struct extent_buffer *leaf)
477{
478 struct btrfs_key key;
479 struct btrfs_key leaf_key;
480 u32 nritems = btrfs_header_nritems(leaf);
481 int slot;
482
483 if (nritems == 0)
484 return 0;
485
486 /* Check the 0 item */
487 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
488 BTRFS_LEAF_DATA_SIZE(root)) {
489 CORRUPT("invalid item offset size pair", leaf, root, 0);
490 return -EIO;
491 }
492
493 /*
494 * Check to make sure each items keys are in the correct order and their
495 * offsets make sense. We only have to loop through nritems-1 because
496 * we check the current slot against the next slot, which verifies the
497 * next slot's offset+size makes sense and that the current's slot
498 * offset is correct.
499 */
500 for (slot = 0; slot < nritems - 1; slot++) {
501 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
502 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
503
504 /* Make sure the keys are in the right order */
505 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
506 CORRUPT("bad key order", leaf, root, slot);
507 return -EIO;
508 }
509
510 /*
511 * Make sure the offset and ends are right, remember that the
512 * item data starts at the end of the leaf and grows towards the
513 * front.
514 */
515 if (btrfs_item_offset_nr(leaf, slot) !=
516 btrfs_item_end_nr(leaf, slot + 1)) {
517 CORRUPT("slot offset bad", leaf, root, slot);
518 return -EIO;
519 }
520
521 /*
522 * Check to make sure that we don't point outside of the leaf,
523 * just incase all the items are consistent to eachother, but
524 * all point outside of the leaf.
525 */
526 if (btrfs_item_end_nr(leaf, slot) >
527 BTRFS_LEAF_DATA_SIZE(root)) {
528 CORRUPT("slot end outside of leaf", leaf, root, slot);
529 return -EIO;
530 }
531 }
532
533 return 0;
534}
535
536struct extent_buffer *find_eb_for_page(struct extent_io_tree *tree,
537 struct page *page, int max_walk)
538{
539 struct extent_buffer *eb;
540 u64 start = page_offset(page);
541 u64 target = start;
542 u64 min_start;
543
544 if (start < max_walk)
545 min_start = 0;
546 else
547 min_start = start - max_walk;
548
549 while (start >= min_start) {
550 eb = find_extent_buffer(tree, start, 0);
551 if (eb) {
552 /*
553 * we found an extent buffer and it contains our page
554 * horray!
555 */
556 if (eb->start <= target &&
557 eb->start + eb->len > target)
558 return eb;
559
560 /* we found an extent buffer that wasn't for us */
561 free_extent_buffer(eb);
562 return NULL;
563 }
564 if (start == 0)
565 break;
566 start -= PAGE_CACHE_SIZE;
567 }
568 return NULL;
569}
570
571static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
572 struct extent_state *state, int mirror)
573{
574 struct extent_io_tree *tree;
575 u64 found_start;
576 int found_level;
577 struct extent_buffer *eb;
578 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
579 int ret = 0;
580 int reads_done;
581
582 if (!page->private)
583 goto out;
584
585 tree = &BTRFS_I(page->mapping->host)->io_tree;
586 eb = (struct extent_buffer *)page->private;
587
588 /* the pending IO might have been the only thing that kept this buffer
589 * in memory. Make sure we have a ref for all this other checks
590 */
591 extent_buffer_get(eb);
592
593 reads_done = atomic_dec_and_test(&eb->io_pages);
594 if (!reads_done)
595 goto err;
596
597 eb->read_mirror = mirror;
598 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
599 ret = -EIO;
600 goto err;
601 }
602
603 found_start = btrfs_header_bytenr(eb);
604 if (found_start != eb->start) {
605 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
606 "%llu %llu\n",
607 (unsigned long long)found_start,
608 (unsigned long long)eb->start);
609 ret = -EIO;
610 goto err;
611 }
612 if (check_tree_block_fsid(root, eb)) {
613 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
614 (unsigned long long)eb->start);
615 ret = -EIO;
616 goto err;
617 }
618 found_level = btrfs_header_level(eb);
619
620 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
621 eb, found_level);
622
623 ret = csum_tree_block(root, eb, 1);
624 if (ret) {
625 ret = -EIO;
626 goto err;
627 }
628
629 /*
630 * If this is a leaf block and it is corrupt, set the corrupt bit so
631 * that we don't try and read the other copies of this block, just
632 * return -EIO.
633 */
634 if (found_level == 0 && check_leaf(root, eb)) {
635 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
636 ret = -EIO;
637 }
638
639 if (!ret)
640 set_extent_buffer_uptodate(eb);
641err:
642 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
643 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
644 btree_readahead_hook(root, eb, eb->start, ret);
645 }
646
647 if (ret)
648 clear_extent_buffer_uptodate(eb);
649 free_extent_buffer(eb);
650out:
651 return ret;
652}
653
654static int btree_io_failed_hook(struct page *page, int failed_mirror)
655{
656 struct extent_buffer *eb;
657 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
658
659 eb = (struct extent_buffer *)page->private;
660 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
661 eb->read_mirror = failed_mirror;
662 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
663 btree_readahead_hook(root, eb, eb->start, -EIO);
664 return -EIO; /* we fixed nothing */
665}
666
667static void end_workqueue_bio(struct bio *bio, int err)
668{
669 struct end_io_wq *end_io_wq = bio->bi_private;
670 struct btrfs_fs_info *fs_info;
671
672 fs_info = end_io_wq->info;
673 end_io_wq->error = err;
674 end_io_wq->work.func = end_workqueue_fn;
675 end_io_wq->work.flags = 0;
676
677 if (bio->bi_rw & REQ_WRITE) {
678 if (end_io_wq->metadata == 1)
679 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
680 &end_io_wq->work);
681 else if (end_io_wq->metadata == 2)
682 btrfs_queue_worker(&fs_info->endio_freespace_worker,
683 &end_io_wq->work);
684 else
685 btrfs_queue_worker(&fs_info->endio_write_workers,
686 &end_io_wq->work);
687 } else {
688 if (end_io_wq->metadata)
689 btrfs_queue_worker(&fs_info->endio_meta_workers,
690 &end_io_wq->work);
691 else
692 btrfs_queue_worker(&fs_info->endio_workers,
693 &end_io_wq->work);
694 }
695}
696
697/*
698 * For the metadata arg you want
699 *
700 * 0 - if data
701 * 1 - if normal metadta
702 * 2 - if writing to the free space cache area
703 */
704int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
705 int metadata)
706{
707 struct end_io_wq *end_io_wq;
708 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
709 if (!end_io_wq)
710 return -ENOMEM;
711
712 end_io_wq->private = bio->bi_private;
713 end_io_wq->end_io = bio->bi_end_io;
714 end_io_wq->info = info;
715 end_io_wq->error = 0;
716 end_io_wq->bio = bio;
717 end_io_wq->metadata = metadata;
718
719 bio->bi_private = end_io_wq;
720 bio->bi_end_io = end_workqueue_bio;
721 return 0;
722}
723
724unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
725{
726 unsigned long limit = min_t(unsigned long,
727 info->workers.max_workers,
728 info->fs_devices->open_devices);
729 return 256 * limit;
730}
731
732static void run_one_async_start(struct btrfs_work *work)
733{
734 struct async_submit_bio *async;
735 int ret;
736
737 async = container_of(work, struct async_submit_bio, work);
738 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
739 async->mirror_num, async->bio_flags,
740 async->bio_offset);
741 if (ret)
742 async->error = ret;
743}
744
745static void run_one_async_done(struct btrfs_work *work)
746{
747 struct btrfs_fs_info *fs_info;
748 struct async_submit_bio *async;
749 int limit;
750
751 async = container_of(work, struct async_submit_bio, work);
752 fs_info = BTRFS_I(async->inode)->root->fs_info;
753
754 limit = btrfs_async_submit_limit(fs_info);
755 limit = limit * 2 / 3;
756
757 atomic_dec(&fs_info->nr_async_submits);
758
759 if (atomic_read(&fs_info->nr_async_submits) < limit &&
760 waitqueue_active(&fs_info->async_submit_wait))
761 wake_up(&fs_info->async_submit_wait);
762
763 /* If an error occured we just want to clean up the bio and move on */
764 if (async->error) {
765 bio_endio(async->bio, async->error);
766 return;
767 }
768
769 async->submit_bio_done(async->inode, async->rw, async->bio,
770 async->mirror_num, async->bio_flags,
771 async->bio_offset);
772}
773
774static void run_one_async_free(struct btrfs_work *work)
775{
776 struct async_submit_bio *async;
777
778 async = container_of(work, struct async_submit_bio, work);
779 kfree(async);
780}
781
782int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
783 int rw, struct bio *bio, int mirror_num,
784 unsigned long bio_flags,
785 u64 bio_offset,
786 extent_submit_bio_hook_t *submit_bio_start,
787 extent_submit_bio_hook_t *submit_bio_done)
788{
789 struct async_submit_bio *async;
790
791 async = kmalloc(sizeof(*async), GFP_NOFS);
792 if (!async)
793 return -ENOMEM;
794
795 async->inode = inode;
796 async->rw = rw;
797 async->bio = bio;
798 async->mirror_num = mirror_num;
799 async->submit_bio_start = submit_bio_start;
800 async->submit_bio_done = submit_bio_done;
801
802 async->work.func = run_one_async_start;
803 async->work.ordered_func = run_one_async_done;
804 async->work.ordered_free = run_one_async_free;
805
806 async->work.flags = 0;
807 async->bio_flags = bio_flags;
808 async->bio_offset = bio_offset;
809
810 async->error = 0;
811
812 atomic_inc(&fs_info->nr_async_submits);
813
814 if (rw & REQ_SYNC)
815 btrfs_set_work_high_prio(&async->work);
816
817 btrfs_queue_worker(&fs_info->workers, &async->work);
818
819 while (atomic_read(&fs_info->async_submit_draining) &&
820 atomic_read(&fs_info->nr_async_submits)) {
821 wait_event(fs_info->async_submit_wait,
822 (atomic_read(&fs_info->nr_async_submits) == 0));
823 }
824
825 return 0;
826}
827
828static int btree_csum_one_bio(struct bio *bio)
829{
830 struct bio_vec *bvec = bio->bi_io_vec;
831 int bio_index = 0;
832 struct btrfs_root *root;
833 int ret = 0;
834
835 WARN_ON(bio->bi_vcnt <= 0);
836 while (bio_index < bio->bi_vcnt) {
837 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
838 ret = csum_dirty_buffer(root, bvec->bv_page);
839 if (ret)
840 break;
841 bio_index++;
842 bvec++;
843 }
844 return ret;
845}
846
847static int __btree_submit_bio_start(struct inode *inode, int rw,
848 struct bio *bio, int mirror_num,
849 unsigned long bio_flags,
850 u64 bio_offset)
851{
852 /*
853 * when we're called for a write, we're already in the async
854 * submission context. Just jump into btrfs_map_bio
855 */
856 return btree_csum_one_bio(bio);
857}
858
859static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
860 int mirror_num, unsigned long bio_flags,
861 u64 bio_offset)
862{
863 /*
864 * when we're called for a write, we're already in the async
865 * submission context. Just jump into btrfs_map_bio
866 */
867 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
868}
869
870static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
871 int mirror_num, unsigned long bio_flags,
872 u64 bio_offset)
873{
874 int ret;
875
876 if (!(rw & REQ_WRITE)) {
877
878 /*
879 * called for a read, do the setup so that checksum validation
880 * can happen in the async kernel threads
881 */
882 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
883 bio, 1);
884 if (ret)
885 return ret;
886 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
887 mirror_num, 0);
888 }
889
890 /*
891 * kthread helpers are used to submit writes so that checksumming
892 * can happen in parallel across all CPUs
893 */
894 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
895 inode, rw, bio, mirror_num, 0,
896 bio_offset,
897 __btree_submit_bio_start,
898 __btree_submit_bio_done);
899}
900
901#ifdef CONFIG_MIGRATION
902static int btree_migratepage(struct address_space *mapping,
903 struct page *newpage, struct page *page,
904 enum migrate_mode mode)
905{
906 /*
907 * we can't safely write a btree page from here,
908 * we haven't done the locking hook
909 */
910 if (PageDirty(page))
911 return -EAGAIN;
912 /*
913 * Buffers may be managed in a filesystem specific way.
914 * We must have no buffers or drop them.
915 */
916 if (page_has_private(page) &&
917 !try_to_release_page(page, GFP_KERNEL))
918 return -EAGAIN;
919 return migrate_page(mapping, newpage, page, mode);
920}
921#endif
922
923
924static int btree_writepages(struct address_space *mapping,
925 struct writeback_control *wbc)
926{
927 struct extent_io_tree *tree;
928 tree = &BTRFS_I(mapping->host)->io_tree;
929 if (wbc->sync_mode == WB_SYNC_NONE) {
930 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
931 u64 num_dirty;
932 unsigned long thresh = 32 * 1024 * 1024;
933
934 if (wbc->for_kupdate)
935 return 0;
936
937 /* this is a bit racy, but that's ok */
938 num_dirty = root->fs_info->dirty_metadata_bytes;
939 if (num_dirty < thresh)
940 return 0;
941 }
942 return btree_write_cache_pages(mapping, wbc);
943}
944
945static int btree_readpage(struct file *file, struct page *page)
946{
947 struct extent_io_tree *tree;
948 tree = &BTRFS_I(page->mapping->host)->io_tree;
949 return extent_read_full_page(tree, page, btree_get_extent, 0);
950}
951
952static int btree_releasepage(struct page *page, gfp_t gfp_flags)
953{
954 if (PageWriteback(page) || PageDirty(page))
955 return 0;
956 /*
957 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
958 * slab allocation from alloc_extent_state down the callchain where
959 * it'd hit a BUG_ON as those flags are not allowed.
960 */
961 gfp_flags &= ~GFP_SLAB_BUG_MASK;
962
963 return try_release_extent_buffer(page, gfp_flags);
964}
965
966static void btree_invalidatepage(struct page *page, unsigned long offset)
967{
968 struct extent_io_tree *tree;
969 tree = &BTRFS_I(page->mapping->host)->io_tree;
970 extent_invalidatepage(tree, page, offset);
971 btree_releasepage(page, GFP_NOFS);
972 if (PagePrivate(page)) {
973 printk(KERN_WARNING "btrfs warning page private not zero "
974 "on page %llu\n", (unsigned long long)page_offset(page));
975 ClearPagePrivate(page);
976 set_page_private(page, 0);
977 page_cache_release(page);
978 }
979}
980
981static int btree_set_page_dirty(struct page *page)
982{
983 struct extent_buffer *eb;
984
985 BUG_ON(!PagePrivate(page));
986 eb = (struct extent_buffer *)page->private;
987 BUG_ON(!eb);
988 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
989 BUG_ON(!atomic_read(&eb->refs));
990 btrfs_assert_tree_locked(eb);
991 return __set_page_dirty_nobuffers(page);
992}
993
994static const struct address_space_operations btree_aops = {
995 .readpage = btree_readpage,
996 .writepages = btree_writepages,
997 .releasepage = btree_releasepage,
998 .invalidatepage = btree_invalidatepage,
999#ifdef CONFIG_MIGRATION
1000 .migratepage = btree_migratepage,
1001#endif
1002 .set_page_dirty = btree_set_page_dirty,
1003};
1004
1005int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1006 u64 parent_transid)
1007{
1008 struct extent_buffer *buf = NULL;
1009 struct inode *btree_inode = root->fs_info->btree_inode;
1010 int ret = 0;
1011
1012 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1013 if (!buf)
1014 return 0;
1015 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1016 buf, 0, WAIT_NONE, btree_get_extent, 0);
1017 free_extent_buffer(buf);
1018 return ret;
1019}
1020
1021int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1022 int mirror_num, struct extent_buffer **eb)
1023{
1024 struct extent_buffer *buf = NULL;
1025 struct inode *btree_inode = root->fs_info->btree_inode;
1026 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1027 int ret;
1028
1029 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1030 if (!buf)
1031 return 0;
1032
1033 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1034
1035 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1036 btree_get_extent, mirror_num);
1037 if (ret) {
1038 free_extent_buffer(buf);
1039 return ret;
1040 }
1041
1042 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1043 free_extent_buffer(buf);
1044 return -EIO;
1045 } else if (extent_buffer_uptodate(buf)) {
1046 *eb = buf;
1047 } else {
1048 free_extent_buffer(buf);
1049 }
1050 return 0;
1051}
1052
1053struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1054 u64 bytenr, u32 blocksize)
1055{
1056 struct inode *btree_inode = root->fs_info->btree_inode;
1057 struct extent_buffer *eb;
1058 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1059 bytenr, blocksize);
1060 return eb;
1061}
1062
1063struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1064 u64 bytenr, u32 blocksize)
1065{
1066 struct inode *btree_inode = root->fs_info->btree_inode;
1067 struct extent_buffer *eb;
1068
1069 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1070 bytenr, blocksize);
1071 return eb;
1072}
1073
1074
1075int btrfs_write_tree_block(struct extent_buffer *buf)
1076{
1077 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1078 buf->start + buf->len - 1);
1079}
1080
1081int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1082{
1083 return filemap_fdatawait_range(buf->pages[0]->mapping,
1084 buf->start, buf->start + buf->len - 1);
1085}
1086
1087struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1088 u32 blocksize, u64 parent_transid)
1089{
1090 struct extent_buffer *buf = NULL;
1091 int ret;
1092
1093 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1094 if (!buf)
1095 return NULL;
1096
1097 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1098 return buf;
1099
1100}
1101
1102void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1103 struct extent_buffer *buf)
1104{
1105 if (btrfs_header_generation(buf) ==
1106 root->fs_info->running_transaction->transid) {
1107 btrfs_assert_tree_locked(buf);
1108
1109 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1110 spin_lock(&root->fs_info->delalloc_lock);
1111 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1112 root->fs_info->dirty_metadata_bytes -= buf->len;
1113 else {
1114 spin_unlock(&root->fs_info->delalloc_lock);
1115 btrfs_panic(root->fs_info, -EOVERFLOW,
1116 "Can't clear %lu bytes from "
1117 " dirty_mdatadata_bytes (%lu)",
1118 buf->len,
1119 root->fs_info->dirty_metadata_bytes);
1120 }
1121 spin_unlock(&root->fs_info->delalloc_lock);
1122 }
1123
1124 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1125 btrfs_set_lock_blocking(buf);
1126 clear_extent_buffer_dirty(buf);
1127 }
1128}
1129
1130static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1131 u32 stripesize, struct btrfs_root *root,
1132 struct btrfs_fs_info *fs_info,
1133 u64 objectid)
1134{
1135 root->node = NULL;
1136 root->commit_root = NULL;
1137 root->sectorsize = sectorsize;
1138 root->nodesize = nodesize;
1139 root->leafsize = leafsize;
1140 root->stripesize = stripesize;
1141 root->ref_cows = 0;
1142 root->track_dirty = 0;
1143 root->in_radix = 0;
1144 root->orphan_item_inserted = 0;
1145 root->orphan_cleanup_state = 0;
1146
1147 root->objectid = objectid;
1148 root->last_trans = 0;
1149 root->highest_objectid = 0;
1150 root->name = NULL;
1151 root->inode_tree = RB_ROOT;
1152 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1153 root->block_rsv = NULL;
1154 root->orphan_block_rsv = NULL;
1155
1156 INIT_LIST_HEAD(&root->dirty_list);
1157 INIT_LIST_HEAD(&root->root_list);
1158 spin_lock_init(&root->orphan_lock);
1159 spin_lock_init(&root->inode_lock);
1160 spin_lock_init(&root->accounting_lock);
1161 mutex_init(&root->objectid_mutex);
1162 mutex_init(&root->log_mutex);
1163 init_waitqueue_head(&root->log_writer_wait);
1164 init_waitqueue_head(&root->log_commit_wait[0]);
1165 init_waitqueue_head(&root->log_commit_wait[1]);
1166 atomic_set(&root->log_commit[0], 0);
1167 atomic_set(&root->log_commit[1], 0);
1168 atomic_set(&root->log_writers, 0);
1169 atomic_set(&root->orphan_inodes, 0);
1170 root->log_batch = 0;
1171 root->log_transid = 0;
1172 root->last_log_commit = 0;
1173 extent_io_tree_init(&root->dirty_log_pages,
1174 fs_info->btree_inode->i_mapping);
1175
1176 memset(&root->root_key, 0, sizeof(root->root_key));
1177 memset(&root->root_item, 0, sizeof(root->root_item));
1178 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1179 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1180 root->defrag_trans_start = fs_info->generation;
1181 init_completion(&root->kobj_unregister);
1182 root->defrag_running = 0;
1183 root->root_key.objectid = objectid;
1184 root->anon_dev = 0;
1185}
1186
1187static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
1188 struct btrfs_fs_info *fs_info,
1189 u64 objectid,
1190 struct btrfs_root *root)
1191{
1192 int ret;
1193 u32 blocksize;
1194 u64 generation;
1195
1196 __setup_root(tree_root->nodesize, tree_root->leafsize,
1197 tree_root->sectorsize, tree_root->stripesize,
1198 root, fs_info, objectid);
1199 ret = btrfs_find_last_root(tree_root, objectid,
1200 &root->root_item, &root->root_key);
1201 if (ret > 0)
1202 return -ENOENT;
1203 else if (ret < 0)
1204 return ret;
1205
1206 generation = btrfs_root_generation(&root->root_item);
1207 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1208 root->commit_root = NULL;
1209 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1210 blocksize, generation);
1211 if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) {
1212 free_extent_buffer(root->node);
1213 root->node = NULL;
1214 return -EIO;
1215 }
1216 root->commit_root = btrfs_root_node(root);
1217 return 0;
1218}
1219
1220static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1221{
1222 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1223 if (root)
1224 root->fs_info = fs_info;
1225 return root;
1226}
1227
1228static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1229 struct btrfs_fs_info *fs_info)
1230{
1231 struct btrfs_root *root;
1232 struct btrfs_root *tree_root = fs_info->tree_root;
1233 struct extent_buffer *leaf;
1234
1235 root = btrfs_alloc_root(fs_info);
1236 if (!root)
1237 return ERR_PTR(-ENOMEM);
1238
1239 __setup_root(tree_root->nodesize, tree_root->leafsize,
1240 tree_root->sectorsize, tree_root->stripesize,
1241 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1242
1243 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1244 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1245 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1246 /*
1247 * log trees do not get reference counted because they go away
1248 * before a real commit is actually done. They do store pointers
1249 * to file data extents, and those reference counts still get
1250 * updated (along with back refs to the log tree).
1251 */
1252 root->ref_cows = 0;
1253
1254 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1255 BTRFS_TREE_LOG_OBJECTID, NULL,
1256 0, 0, 0);
1257 if (IS_ERR(leaf)) {
1258 kfree(root);
1259 return ERR_CAST(leaf);
1260 }
1261
1262 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1263 btrfs_set_header_bytenr(leaf, leaf->start);
1264 btrfs_set_header_generation(leaf, trans->transid);
1265 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1266 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1267 root->node = leaf;
1268
1269 write_extent_buffer(root->node, root->fs_info->fsid,
1270 (unsigned long)btrfs_header_fsid(root->node),
1271 BTRFS_FSID_SIZE);
1272 btrfs_mark_buffer_dirty(root->node);
1273 btrfs_tree_unlock(root->node);
1274 return root;
1275}
1276
1277int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1278 struct btrfs_fs_info *fs_info)
1279{
1280 struct btrfs_root *log_root;
1281
1282 log_root = alloc_log_tree(trans, fs_info);
1283 if (IS_ERR(log_root))
1284 return PTR_ERR(log_root);
1285 WARN_ON(fs_info->log_root_tree);
1286 fs_info->log_root_tree = log_root;
1287 return 0;
1288}
1289
1290int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1291 struct btrfs_root *root)
1292{
1293 struct btrfs_root *log_root;
1294 struct btrfs_inode_item *inode_item;
1295
1296 log_root = alloc_log_tree(trans, root->fs_info);
1297 if (IS_ERR(log_root))
1298 return PTR_ERR(log_root);
1299
1300 log_root->last_trans = trans->transid;
1301 log_root->root_key.offset = root->root_key.objectid;
1302
1303 inode_item = &log_root->root_item.inode;
1304 inode_item->generation = cpu_to_le64(1);
1305 inode_item->size = cpu_to_le64(3);
1306 inode_item->nlink = cpu_to_le32(1);
1307 inode_item->nbytes = cpu_to_le64(root->leafsize);
1308 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1309
1310 btrfs_set_root_node(&log_root->root_item, log_root->node);
1311
1312 WARN_ON(root->log_root);
1313 root->log_root = log_root;
1314 root->log_transid = 0;
1315 root->last_log_commit = 0;
1316 return 0;
1317}
1318
1319struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1320 struct btrfs_key *location)
1321{
1322 struct btrfs_root *root;
1323 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1324 struct btrfs_path *path;
1325 struct extent_buffer *l;
1326 u64 generation;
1327 u32 blocksize;
1328 int ret = 0;
1329
1330 root = btrfs_alloc_root(fs_info);
1331 if (!root)
1332 return ERR_PTR(-ENOMEM);
1333 if (location->offset == (u64)-1) {
1334 ret = find_and_setup_root(tree_root, fs_info,
1335 location->objectid, root);
1336 if (ret) {
1337 kfree(root);
1338 return ERR_PTR(ret);
1339 }
1340 goto out;
1341 }
1342
1343 __setup_root(tree_root->nodesize, tree_root->leafsize,
1344 tree_root->sectorsize, tree_root->stripesize,
1345 root, fs_info, location->objectid);
1346
1347 path = btrfs_alloc_path();
1348 if (!path) {
1349 kfree(root);
1350 return ERR_PTR(-ENOMEM);
1351 }
1352 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1353 if (ret == 0) {
1354 l = path->nodes[0];
1355 read_extent_buffer(l, &root->root_item,
1356 btrfs_item_ptr_offset(l, path->slots[0]),
1357 sizeof(root->root_item));
1358 memcpy(&root->root_key, location, sizeof(*location));
1359 }
1360 btrfs_free_path(path);
1361 if (ret) {
1362 kfree(root);
1363 if (ret > 0)
1364 ret = -ENOENT;
1365 return ERR_PTR(ret);
1366 }
1367
1368 generation = btrfs_root_generation(&root->root_item);
1369 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1370 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1371 blocksize, generation);
1372 root->commit_root = btrfs_root_node(root);
1373 BUG_ON(!root->node); /* -ENOMEM */
1374out:
1375 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1376 root->ref_cows = 1;
1377 btrfs_check_and_init_root_item(&root->root_item);
1378 }
1379
1380 return root;
1381}
1382
1383struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1384 struct btrfs_key *location)
1385{
1386 struct btrfs_root *root;
1387 int ret;
1388
1389 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1390 return fs_info->tree_root;
1391 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1392 return fs_info->extent_root;
1393 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1394 return fs_info->chunk_root;
1395 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1396 return fs_info->dev_root;
1397 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1398 return fs_info->csum_root;
1399again:
1400 spin_lock(&fs_info->fs_roots_radix_lock);
1401 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1402 (unsigned long)location->objectid);
1403 spin_unlock(&fs_info->fs_roots_radix_lock);
1404 if (root)
1405 return root;
1406
1407 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1408 if (IS_ERR(root))
1409 return root;
1410
1411 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1412 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1413 GFP_NOFS);
1414 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1415 ret = -ENOMEM;
1416 goto fail;
1417 }
1418
1419 btrfs_init_free_ino_ctl(root);
1420 mutex_init(&root->fs_commit_mutex);
1421 spin_lock_init(&root->cache_lock);
1422 init_waitqueue_head(&root->cache_wait);
1423
1424 ret = get_anon_bdev(&root->anon_dev);
1425 if (ret)
1426 goto fail;
1427
1428 if (btrfs_root_refs(&root->root_item) == 0) {
1429 ret = -ENOENT;
1430 goto fail;
1431 }
1432
1433 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1434 if (ret < 0)
1435 goto fail;
1436 if (ret == 0)
1437 root->orphan_item_inserted = 1;
1438
1439 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1440 if (ret)
1441 goto fail;
1442
1443 spin_lock(&fs_info->fs_roots_radix_lock);
1444 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1445 (unsigned long)root->root_key.objectid,
1446 root);
1447 if (ret == 0)
1448 root->in_radix = 1;
1449
1450 spin_unlock(&fs_info->fs_roots_radix_lock);
1451 radix_tree_preload_end();
1452 if (ret) {
1453 if (ret == -EEXIST) {
1454 free_fs_root(root);
1455 goto again;
1456 }
1457 goto fail;
1458 }
1459
1460 ret = btrfs_find_dead_roots(fs_info->tree_root,
1461 root->root_key.objectid);
1462 WARN_ON(ret);
1463 return root;
1464fail:
1465 free_fs_root(root);
1466 return ERR_PTR(ret);
1467}
1468
1469static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1470{
1471 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1472 int ret = 0;
1473 struct btrfs_device *device;
1474 struct backing_dev_info *bdi;
1475
1476 rcu_read_lock();
1477 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1478 if (!device->bdev)
1479 continue;
1480 bdi = blk_get_backing_dev_info(device->bdev);
1481 if (bdi && bdi_congested(bdi, bdi_bits)) {
1482 ret = 1;
1483 break;
1484 }
1485 }
1486 rcu_read_unlock();
1487 return ret;
1488}
1489
1490/*
1491 * If this fails, caller must call bdi_destroy() to get rid of the
1492 * bdi again.
1493 */
1494static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1495{
1496 int err;
1497
1498 bdi->capabilities = BDI_CAP_MAP_COPY;
1499 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1500 if (err)
1501 return err;
1502
1503 bdi->ra_pages = default_backing_dev_info.ra_pages;
1504 bdi->congested_fn = btrfs_congested_fn;
1505 bdi->congested_data = info;
1506 return 0;
1507}
1508
1509/*
1510 * called by the kthread helper functions to finally call the bio end_io
1511 * functions. This is where read checksum verification actually happens
1512 */
1513static void end_workqueue_fn(struct btrfs_work *work)
1514{
1515 struct bio *bio;
1516 struct end_io_wq *end_io_wq;
1517 struct btrfs_fs_info *fs_info;
1518 int error;
1519
1520 end_io_wq = container_of(work, struct end_io_wq, work);
1521 bio = end_io_wq->bio;
1522 fs_info = end_io_wq->info;
1523
1524 error = end_io_wq->error;
1525 bio->bi_private = end_io_wq->private;
1526 bio->bi_end_io = end_io_wq->end_io;
1527 kfree(end_io_wq);
1528 bio_endio(bio, error);
1529}
1530
1531static int cleaner_kthread(void *arg)
1532{
1533 struct btrfs_root *root = arg;
1534
1535 do {
1536 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1537
1538 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1539 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1540 btrfs_run_delayed_iputs(root);
1541 btrfs_clean_old_snapshots(root);
1542 mutex_unlock(&root->fs_info->cleaner_mutex);
1543 btrfs_run_defrag_inodes(root->fs_info);
1544 }
1545
1546 if (!try_to_freeze()) {
1547 set_current_state(TASK_INTERRUPTIBLE);
1548 if (!kthread_should_stop())
1549 schedule();
1550 __set_current_state(TASK_RUNNING);
1551 }
1552 } while (!kthread_should_stop());
1553 return 0;
1554}
1555
1556static int transaction_kthread(void *arg)
1557{
1558 struct btrfs_root *root = arg;
1559 struct btrfs_trans_handle *trans;
1560 struct btrfs_transaction *cur;
1561 u64 transid;
1562 unsigned long now;
1563 unsigned long delay;
1564 bool cannot_commit;
1565
1566 do {
1567 cannot_commit = false;
1568 delay = HZ * 30;
1569 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1570 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1571
1572 spin_lock(&root->fs_info->trans_lock);
1573 cur = root->fs_info->running_transaction;
1574 if (!cur) {
1575 spin_unlock(&root->fs_info->trans_lock);
1576 goto sleep;
1577 }
1578
1579 now = get_seconds();
1580 if (!cur->blocked &&
1581 (now < cur->start_time || now - cur->start_time < 30)) {
1582 spin_unlock(&root->fs_info->trans_lock);
1583 delay = HZ * 5;
1584 goto sleep;
1585 }
1586 transid = cur->transid;
1587 spin_unlock(&root->fs_info->trans_lock);
1588
1589 /* If the file system is aborted, this will always fail. */
1590 trans = btrfs_join_transaction(root);
1591 if (IS_ERR(trans)) {
1592 cannot_commit = true;
1593 goto sleep;
1594 }
1595 if (transid == trans->transid) {
1596 btrfs_commit_transaction(trans, root);
1597 } else {
1598 btrfs_end_transaction(trans, root);
1599 }
1600sleep:
1601 wake_up_process(root->fs_info->cleaner_kthread);
1602 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1603
1604 if (!try_to_freeze()) {
1605 set_current_state(TASK_INTERRUPTIBLE);
1606 if (!kthread_should_stop() &&
1607 (!btrfs_transaction_blocked(root->fs_info) ||
1608 cannot_commit))
1609 schedule_timeout(delay);
1610 __set_current_state(TASK_RUNNING);
1611 }
1612 } while (!kthread_should_stop());
1613 return 0;
1614}
1615
1616/*
1617 * this will find the highest generation in the array of
1618 * root backups. The index of the highest array is returned,
1619 * or -1 if we can't find anything.
1620 *
1621 * We check to make sure the array is valid by comparing the
1622 * generation of the latest root in the array with the generation
1623 * in the super block. If they don't match we pitch it.
1624 */
1625static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1626{
1627 u64 cur;
1628 int newest_index = -1;
1629 struct btrfs_root_backup *root_backup;
1630 int i;
1631
1632 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1633 root_backup = info->super_copy->super_roots + i;
1634 cur = btrfs_backup_tree_root_gen(root_backup);
1635 if (cur == newest_gen)
1636 newest_index = i;
1637 }
1638
1639 /* check to see if we actually wrapped around */
1640 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1641 root_backup = info->super_copy->super_roots;
1642 cur = btrfs_backup_tree_root_gen(root_backup);
1643 if (cur == newest_gen)
1644 newest_index = 0;
1645 }
1646 return newest_index;
1647}
1648
1649
1650/*
1651 * find the oldest backup so we know where to store new entries
1652 * in the backup array. This will set the backup_root_index
1653 * field in the fs_info struct
1654 */
1655static void find_oldest_super_backup(struct btrfs_fs_info *info,
1656 u64 newest_gen)
1657{
1658 int newest_index = -1;
1659
1660 newest_index = find_newest_super_backup(info, newest_gen);
1661 /* if there was garbage in there, just move along */
1662 if (newest_index == -1) {
1663 info->backup_root_index = 0;
1664 } else {
1665 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1666 }
1667}
1668
1669/*
1670 * copy all the root pointers into the super backup array.
1671 * this will bump the backup pointer by one when it is
1672 * done
1673 */
1674static void backup_super_roots(struct btrfs_fs_info *info)
1675{
1676 int next_backup;
1677 struct btrfs_root_backup *root_backup;
1678 int last_backup;
1679
1680 next_backup = info->backup_root_index;
1681 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1682 BTRFS_NUM_BACKUP_ROOTS;
1683
1684 /*
1685 * just overwrite the last backup if we're at the same generation
1686 * this happens only at umount
1687 */
1688 root_backup = info->super_for_commit->super_roots + last_backup;
1689 if (btrfs_backup_tree_root_gen(root_backup) ==
1690 btrfs_header_generation(info->tree_root->node))
1691 next_backup = last_backup;
1692
1693 root_backup = info->super_for_commit->super_roots + next_backup;
1694
1695 /*
1696 * make sure all of our padding and empty slots get zero filled
1697 * regardless of which ones we use today
1698 */
1699 memset(root_backup, 0, sizeof(*root_backup));
1700
1701 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1702
1703 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1704 btrfs_set_backup_tree_root_gen(root_backup,
1705 btrfs_header_generation(info->tree_root->node));
1706
1707 btrfs_set_backup_tree_root_level(root_backup,
1708 btrfs_header_level(info->tree_root->node));
1709
1710 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1711 btrfs_set_backup_chunk_root_gen(root_backup,
1712 btrfs_header_generation(info->chunk_root->node));
1713 btrfs_set_backup_chunk_root_level(root_backup,
1714 btrfs_header_level(info->chunk_root->node));
1715
1716 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1717 btrfs_set_backup_extent_root_gen(root_backup,
1718 btrfs_header_generation(info->extent_root->node));
1719 btrfs_set_backup_extent_root_level(root_backup,
1720 btrfs_header_level(info->extent_root->node));
1721
1722 /*
1723 * we might commit during log recovery, which happens before we set
1724 * the fs_root. Make sure it is valid before we fill it in.
1725 */
1726 if (info->fs_root && info->fs_root->node) {
1727 btrfs_set_backup_fs_root(root_backup,
1728 info->fs_root->node->start);
1729 btrfs_set_backup_fs_root_gen(root_backup,
1730 btrfs_header_generation(info->fs_root->node));
1731 btrfs_set_backup_fs_root_level(root_backup,
1732 btrfs_header_level(info->fs_root->node));
1733 }
1734
1735 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1736 btrfs_set_backup_dev_root_gen(root_backup,
1737 btrfs_header_generation(info->dev_root->node));
1738 btrfs_set_backup_dev_root_level(root_backup,
1739 btrfs_header_level(info->dev_root->node));
1740
1741 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1742 btrfs_set_backup_csum_root_gen(root_backup,
1743 btrfs_header_generation(info->csum_root->node));
1744 btrfs_set_backup_csum_root_level(root_backup,
1745 btrfs_header_level(info->csum_root->node));
1746
1747 btrfs_set_backup_total_bytes(root_backup,
1748 btrfs_super_total_bytes(info->super_copy));
1749 btrfs_set_backup_bytes_used(root_backup,
1750 btrfs_super_bytes_used(info->super_copy));
1751 btrfs_set_backup_num_devices(root_backup,
1752 btrfs_super_num_devices(info->super_copy));
1753
1754 /*
1755 * if we don't copy this out to the super_copy, it won't get remembered
1756 * for the next commit
1757 */
1758 memcpy(&info->super_copy->super_roots,
1759 &info->super_for_commit->super_roots,
1760 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1761}
1762
1763/*
1764 * this copies info out of the root backup array and back into
1765 * the in-memory super block. It is meant to help iterate through
1766 * the array, so you send it the number of backups you've already
1767 * tried and the last backup index you used.
1768 *
1769 * this returns -1 when it has tried all the backups
1770 */
1771static noinline int next_root_backup(struct btrfs_fs_info *info,
1772 struct btrfs_super_block *super,
1773 int *num_backups_tried, int *backup_index)
1774{
1775 struct btrfs_root_backup *root_backup;
1776 int newest = *backup_index;
1777
1778 if (*num_backups_tried == 0) {
1779 u64 gen = btrfs_super_generation(super);
1780
1781 newest = find_newest_super_backup(info, gen);
1782 if (newest == -1)
1783 return -1;
1784
1785 *backup_index = newest;
1786 *num_backups_tried = 1;
1787 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1788 /* we've tried all the backups, all done */
1789 return -1;
1790 } else {
1791 /* jump to the next oldest backup */
1792 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1793 BTRFS_NUM_BACKUP_ROOTS;
1794 *backup_index = newest;
1795 *num_backups_tried += 1;
1796 }
1797 root_backup = super->super_roots + newest;
1798
1799 btrfs_set_super_generation(super,
1800 btrfs_backup_tree_root_gen(root_backup));
1801 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1802 btrfs_set_super_root_level(super,
1803 btrfs_backup_tree_root_level(root_backup));
1804 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1805
1806 /*
1807 * fixme: the total bytes and num_devices need to match or we should
1808 * need a fsck
1809 */
1810 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1811 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1812 return 0;
1813}
1814
1815/* helper to cleanup tree roots */
1816static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1817{
1818 free_extent_buffer(info->tree_root->node);
1819 free_extent_buffer(info->tree_root->commit_root);
1820 free_extent_buffer(info->dev_root->node);
1821 free_extent_buffer(info->dev_root->commit_root);
1822 free_extent_buffer(info->extent_root->node);
1823 free_extent_buffer(info->extent_root->commit_root);
1824 free_extent_buffer(info->csum_root->node);
1825 free_extent_buffer(info->csum_root->commit_root);
1826
1827 info->tree_root->node = NULL;
1828 info->tree_root->commit_root = NULL;
1829 info->dev_root->node = NULL;
1830 info->dev_root->commit_root = NULL;
1831 info->extent_root->node = NULL;
1832 info->extent_root->commit_root = NULL;
1833 info->csum_root->node = NULL;
1834 info->csum_root->commit_root = NULL;
1835
1836 if (chunk_root) {
1837 free_extent_buffer(info->chunk_root->node);
1838 free_extent_buffer(info->chunk_root->commit_root);
1839 info->chunk_root->node = NULL;
1840 info->chunk_root->commit_root = NULL;
1841 }
1842}
1843
1844
1845int open_ctree(struct super_block *sb,
1846 struct btrfs_fs_devices *fs_devices,
1847 char *options)
1848{
1849 u32 sectorsize;
1850 u32 nodesize;
1851 u32 leafsize;
1852 u32 blocksize;
1853 u32 stripesize;
1854 u64 generation;
1855 u64 features;
1856 struct btrfs_key location;
1857 struct buffer_head *bh;
1858 struct btrfs_super_block *disk_super;
1859 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1860 struct btrfs_root *tree_root;
1861 struct btrfs_root *extent_root;
1862 struct btrfs_root *csum_root;
1863 struct btrfs_root *chunk_root;
1864 struct btrfs_root *dev_root;
1865 struct btrfs_root *log_tree_root;
1866 int ret;
1867 int err = -EINVAL;
1868 int num_backups_tried = 0;
1869 int backup_index = 0;
1870
1871 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
1872 extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
1873 csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
1874 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
1875 dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
1876
1877 if (!tree_root || !extent_root || !csum_root ||
1878 !chunk_root || !dev_root) {
1879 err = -ENOMEM;
1880 goto fail;
1881 }
1882
1883 ret = init_srcu_struct(&fs_info->subvol_srcu);
1884 if (ret) {
1885 err = ret;
1886 goto fail;
1887 }
1888
1889 ret = setup_bdi(fs_info, &fs_info->bdi);
1890 if (ret) {
1891 err = ret;
1892 goto fail_srcu;
1893 }
1894
1895 fs_info->btree_inode = new_inode(sb);
1896 if (!fs_info->btree_inode) {
1897 err = -ENOMEM;
1898 goto fail_bdi;
1899 }
1900
1901 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1902
1903 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1904 INIT_LIST_HEAD(&fs_info->trans_list);
1905 INIT_LIST_HEAD(&fs_info->dead_roots);
1906 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1907 INIT_LIST_HEAD(&fs_info->hashers);
1908 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1909 INIT_LIST_HEAD(&fs_info->ordered_operations);
1910 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1911 spin_lock_init(&fs_info->delalloc_lock);
1912 spin_lock_init(&fs_info->trans_lock);
1913 spin_lock_init(&fs_info->ref_cache_lock);
1914 spin_lock_init(&fs_info->fs_roots_radix_lock);
1915 spin_lock_init(&fs_info->delayed_iput_lock);
1916 spin_lock_init(&fs_info->defrag_inodes_lock);
1917 spin_lock_init(&fs_info->free_chunk_lock);
1918 spin_lock_init(&fs_info->tree_mod_seq_lock);
1919 rwlock_init(&fs_info->tree_mod_log_lock);
1920 mutex_init(&fs_info->reloc_mutex);
1921
1922 init_completion(&fs_info->kobj_unregister);
1923 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1924 INIT_LIST_HEAD(&fs_info->space_info);
1925 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
1926 btrfs_mapping_init(&fs_info->mapping_tree);
1927 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1928 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1929 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1930 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1931 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1932 btrfs_init_block_rsv(&fs_info->delayed_block_rsv);
1933 atomic_set(&fs_info->nr_async_submits, 0);
1934 atomic_set(&fs_info->async_delalloc_pages, 0);
1935 atomic_set(&fs_info->async_submit_draining, 0);
1936 atomic_set(&fs_info->nr_async_bios, 0);
1937 atomic_set(&fs_info->defrag_running, 0);
1938 atomic_set(&fs_info->tree_mod_seq, 0);
1939 fs_info->sb = sb;
1940 fs_info->max_inline = 8192 * 1024;
1941 fs_info->metadata_ratio = 0;
1942 fs_info->defrag_inodes = RB_ROOT;
1943 fs_info->trans_no_join = 0;
1944 fs_info->free_chunk_space = 0;
1945 fs_info->tree_mod_log = RB_ROOT;
1946
1947 /* readahead state */
1948 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
1949 spin_lock_init(&fs_info->reada_lock);
1950
1951 fs_info->thread_pool_size = min_t(unsigned long,
1952 num_online_cpus() + 2, 8);
1953
1954 INIT_LIST_HEAD(&fs_info->ordered_extents);
1955 spin_lock_init(&fs_info->ordered_extent_lock);
1956 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1957 GFP_NOFS);
1958 if (!fs_info->delayed_root) {
1959 err = -ENOMEM;
1960 goto fail_iput;
1961 }
1962 btrfs_init_delayed_root(fs_info->delayed_root);
1963
1964 mutex_init(&fs_info->scrub_lock);
1965 atomic_set(&fs_info->scrubs_running, 0);
1966 atomic_set(&fs_info->scrub_pause_req, 0);
1967 atomic_set(&fs_info->scrubs_paused, 0);
1968 atomic_set(&fs_info->scrub_cancel_req, 0);
1969 init_waitqueue_head(&fs_info->scrub_pause_wait);
1970 init_rwsem(&fs_info->scrub_super_lock);
1971 fs_info->scrub_workers_refcnt = 0;
1972#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1973 fs_info->check_integrity_print_mask = 0;
1974#endif
1975
1976 spin_lock_init(&fs_info->balance_lock);
1977 mutex_init(&fs_info->balance_mutex);
1978 atomic_set(&fs_info->balance_running, 0);
1979 atomic_set(&fs_info->balance_pause_req, 0);
1980 atomic_set(&fs_info->balance_cancel_req, 0);
1981 fs_info->balance_ctl = NULL;
1982 init_waitqueue_head(&fs_info->balance_wait_q);
1983
1984 sb->s_blocksize = 4096;
1985 sb->s_blocksize_bits = blksize_bits(4096);
1986 sb->s_bdi = &fs_info->bdi;
1987
1988 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1989 set_nlink(fs_info->btree_inode, 1);
1990 /*
1991 * we set the i_size on the btree inode to the max possible int.
1992 * the real end of the address space is determined by all of
1993 * the devices in the system
1994 */
1995 fs_info->btree_inode->i_size = OFFSET_MAX;
1996 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1997 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1998
1999 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2000 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2001 fs_info->btree_inode->i_mapping);
2002 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2003 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2004
2005 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2006
2007 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2008 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2009 sizeof(struct btrfs_key));
2010 set_bit(BTRFS_INODE_DUMMY,
2011 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2012 insert_inode_hash(fs_info->btree_inode);
2013
2014 spin_lock_init(&fs_info->block_group_cache_lock);
2015 fs_info->block_group_cache_tree = RB_ROOT;
2016
2017 extent_io_tree_init(&fs_info->freed_extents[0],
2018 fs_info->btree_inode->i_mapping);
2019 extent_io_tree_init(&fs_info->freed_extents[1],
2020 fs_info->btree_inode->i_mapping);
2021 fs_info->pinned_extents = &fs_info->freed_extents[0];
2022 fs_info->do_barriers = 1;
2023
2024
2025 mutex_init(&fs_info->ordered_operations_mutex);
2026 mutex_init(&fs_info->tree_log_mutex);
2027 mutex_init(&fs_info->chunk_mutex);
2028 mutex_init(&fs_info->transaction_kthread_mutex);
2029 mutex_init(&fs_info->cleaner_mutex);
2030 mutex_init(&fs_info->volume_mutex);
2031 init_rwsem(&fs_info->extent_commit_sem);
2032 init_rwsem(&fs_info->cleanup_work_sem);
2033 init_rwsem(&fs_info->subvol_sem);
2034
2035 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2036 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2037
2038 init_waitqueue_head(&fs_info->transaction_throttle);
2039 init_waitqueue_head(&fs_info->transaction_wait);
2040 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2041 init_waitqueue_head(&fs_info->async_submit_wait);
2042
2043 __setup_root(4096, 4096, 4096, 4096, tree_root,
2044 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2045
2046 invalidate_bdev(fs_devices->latest_bdev);
2047 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2048 if (!bh) {
2049 err = -EINVAL;
2050 goto fail_alloc;
2051 }
2052
2053 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2054 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2055 sizeof(*fs_info->super_for_commit));
2056 brelse(bh);
2057
2058 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2059
2060 disk_super = fs_info->super_copy;
2061 if (!btrfs_super_root(disk_super))
2062 goto fail_alloc;
2063
2064 /* check FS state, whether FS is broken. */
2065 fs_info->fs_state |= btrfs_super_flags(disk_super);
2066
2067 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2068 if (ret) {
2069 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2070 err = ret;
2071 goto fail_alloc;
2072 }
2073
2074 /*
2075 * run through our array of backup supers and setup
2076 * our ring pointer to the oldest one
2077 */
2078 generation = btrfs_super_generation(disk_super);
2079 find_oldest_super_backup(fs_info, generation);
2080
2081 /*
2082 * In the long term, we'll store the compression type in the super
2083 * block, and it'll be used for per file compression control.
2084 */
2085 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2086
2087 ret = btrfs_parse_options(tree_root, options);
2088 if (ret) {
2089 err = ret;
2090 goto fail_alloc;
2091 }
2092
2093 features = btrfs_super_incompat_flags(disk_super) &
2094 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2095 if (features) {
2096 printk(KERN_ERR "BTRFS: couldn't mount because of "
2097 "unsupported optional features (%Lx).\n",
2098 (unsigned long long)features);
2099 err = -EINVAL;
2100 goto fail_alloc;
2101 }
2102
2103 if (btrfs_super_leafsize(disk_super) !=
2104 btrfs_super_nodesize(disk_super)) {
2105 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2106 "blocksizes don't match. node %d leaf %d\n",
2107 btrfs_super_nodesize(disk_super),
2108 btrfs_super_leafsize(disk_super));
2109 err = -EINVAL;
2110 goto fail_alloc;
2111 }
2112 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2113 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2114 "blocksize (%d) was too large\n",
2115 btrfs_super_leafsize(disk_super));
2116 err = -EINVAL;
2117 goto fail_alloc;
2118 }
2119
2120 features = btrfs_super_incompat_flags(disk_super);
2121 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2122 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2123 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2124
2125 /*
2126 * flag our filesystem as having big metadata blocks if
2127 * they are bigger than the page size
2128 */
2129 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2130 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2131 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2132 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2133 }
2134
2135 nodesize = btrfs_super_nodesize(disk_super);
2136 leafsize = btrfs_super_leafsize(disk_super);
2137 sectorsize = btrfs_super_sectorsize(disk_super);
2138 stripesize = btrfs_super_stripesize(disk_super);
2139
2140 /*
2141 * mixed block groups end up with duplicate but slightly offset
2142 * extent buffers for the same range. It leads to corruptions
2143 */
2144 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2145 (sectorsize != leafsize)) {
2146 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2147 "are not allowed for mixed block groups on %s\n",
2148 sb->s_id);
2149 goto fail_alloc;
2150 }
2151
2152 btrfs_set_super_incompat_flags(disk_super, features);
2153
2154 features = btrfs_super_compat_ro_flags(disk_super) &
2155 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2156 if (!(sb->s_flags & MS_RDONLY) && features) {
2157 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2158 "unsupported option features (%Lx).\n",
2159 (unsigned long long)features);
2160 err = -EINVAL;
2161 goto fail_alloc;
2162 }
2163
2164 btrfs_init_workers(&fs_info->generic_worker,
2165 "genwork", 1, NULL);
2166
2167 btrfs_init_workers(&fs_info->workers, "worker",
2168 fs_info->thread_pool_size,
2169 &fs_info->generic_worker);
2170
2171 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2172 fs_info->thread_pool_size,
2173 &fs_info->generic_worker);
2174
2175 btrfs_init_workers(&fs_info->submit_workers, "submit",
2176 min_t(u64, fs_devices->num_devices,
2177 fs_info->thread_pool_size),
2178 &fs_info->generic_worker);
2179
2180 btrfs_init_workers(&fs_info->caching_workers, "cache",
2181 2, &fs_info->generic_worker);
2182
2183 /* a higher idle thresh on the submit workers makes it much more
2184 * likely that bios will be send down in a sane order to the
2185 * devices
2186 */
2187 fs_info->submit_workers.idle_thresh = 64;
2188
2189 fs_info->workers.idle_thresh = 16;
2190 fs_info->workers.ordered = 1;
2191
2192 fs_info->delalloc_workers.idle_thresh = 2;
2193 fs_info->delalloc_workers.ordered = 1;
2194
2195 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2196 &fs_info->generic_worker);
2197 btrfs_init_workers(&fs_info->endio_workers, "endio",
2198 fs_info->thread_pool_size,
2199 &fs_info->generic_worker);
2200 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2201 fs_info->thread_pool_size,
2202 &fs_info->generic_worker);
2203 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2204 "endio-meta-write", fs_info->thread_pool_size,
2205 &fs_info->generic_worker);
2206 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2207 fs_info->thread_pool_size,
2208 &fs_info->generic_worker);
2209 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2210 1, &fs_info->generic_worker);
2211 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2212 fs_info->thread_pool_size,
2213 &fs_info->generic_worker);
2214 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2215 fs_info->thread_pool_size,
2216 &fs_info->generic_worker);
2217
2218 /*
2219 * endios are largely parallel and should have a very
2220 * low idle thresh
2221 */
2222 fs_info->endio_workers.idle_thresh = 4;
2223 fs_info->endio_meta_workers.idle_thresh = 4;
2224
2225 fs_info->endio_write_workers.idle_thresh = 2;
2226 fs_info->endio_meta_write_workers.idle_thresh = 2;
2227 fs_info->readahead_workers.idle_thresh = 2;
2228
2229 /*
2230 * btrfs_start_workers can really only fail because of ENOMEM so just
2231 * return -ENOMEM if any of these fail.
2232 */
2233 ret = btrfs_start_workers(&fs_info->workers);
2234 ret |= btrfs_start_workers(&fs_info->generic_worker);
2235 ret |= btrfs_start_workers(&fs_info->submit_workers);
2236 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2237 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2238 ret |= btrfs_start_workers(&fs_info->endio_workers);
2239 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2240 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2241 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2242 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2243 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2244 ret |= btrfs_start_workers(&fs_info->caching_workers);
2245 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2246 if (ret) {
2247 ret = -ENOMEM;
2248 goto fail_sb_buffer;
2249 }
2250
2251 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2252 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2253 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2254
2255 tree_root->nodesize = nodesize;
2256 tree_root->leafsize = leafsize;
2257 tree_root->sectorsize = sectorsize;
2258 tree_root->stripesize = stripesize;
2259
2260 sb->s_blocksize = sectorsize;
2261 sb->s_blocksize_bits = blksize_bits(sectorsize);
2262
2263 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2264 sizeof(disk_super->magic))) {
2265 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2266 goto fail_sb_buffer;
2267 }
2268
2269 if (sectorsize != PAGE_SIZE) {
2270 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2271 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2272 goto fail_sb_buffer;
2273 }
2274
2275 mutex_lock(&fs_info->chunk_mutex);
2276 ret = btrfs_read_sys_array(tree_root);
2277 mutex_unlock(&fs_info->chunk_mutex);
2278 if (ret) {
2279 printk(KERN_WARNING "btrfs: failed to read the system "
2280 "array on %s\n", sb->s_id);
2281 goto fail_sb_buffer;
2282 }
2283
2284 blocksize = btrfs_level_size(tree_root,
2285 btrfs_super_chunk_root_level(disk_super));
2286 generation = btrfs_super_chunk_root_generation(disk_super);
2287
2288 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2289 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2290
2291 chunk_root->node = read_tree_block(chunk_root,
2292 btrfs_super_chunk_root(disk_super),
2293 blocksize, generation);
2294 BUG_ON(!chunk_root->node); /* -ENOMEM */
2295 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2296 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2297 sb->s_id);
2298 goto fail_tree_roots;
2299 }
2300 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2301 chunk_root->commit_root = btrfs_root_node(chunk_root);
2302
2303 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2304 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2305 BTRFS_UUID_SIZE);
2306
2307 ret = btrfs_read_chunk_tree(chunk_root);
2308 if (ret) {
2309 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2310 sb->s_id);
2311 goto fail_tree_roots;
2312 }
2313
2314 btrfs_close_extra_devices(fs_devices);
2315
2316 if (!fs_devices->latest_bdev) {
2317 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2318 sb->s_id);
2319 goto fail_tree_roots;
2320 }
2321
2322retry_root_backup:
2323 blocksize = btrfs_level_size(tree_root,
2324 btrfs_super_root_level(disk_super));
2325 generation = btrfs_super_generation(disk_super);
2326
2327 tree_root->node = read_tree_block(tree_root,
2328 btrfs_super_root(disk_super),
2329 blocksize, generation);
2330 if (!tree_root->node ||
2331 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2332 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2333 sb->s_id);
2334
2335 goto recovery_tree_root;
2336 }
2337
2338 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2339 tree_root->commit_root = btrfs_root_node(tree_root);
2340
2341 ret = find_and_setup_root(tree_root, fs_info,
2342 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2343 if (ret)
2344 goto recovery_tree_root;
2345 extent_root->track_dirty = 1;
2346
2347 ret = find_and_setup_root(tree_root, fs_info,
2348 BTRFS_DEV_TREE_OBJECTID, dev_root);
2349 if (ret)
2350 goto recovery_tree_root;
2351 dev_root->track_dirty = 1;
2352
2353 ret = find_and_setup_root(tree_root, fs_info,
2354 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2355 if (ret)
2356 goto recovery_tree_root;
2357 csum_root->track_dirty = 1;
2358
2359 fs_info->generation = generation;
2360 fs_info->last_trans_committed = generation;
2361
2362 ret = btrfs_recover_balance(fs_info);
2363 if (ret) {
2364 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2365 goto fail_block_groups;
2366 }
2367
2368 ret = btrfs_init_dev_stats(fs_info);
2369 if (ret) {
2370 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2371 ret);
2372 goto fail_block_groups;
2373 }
2374
2375 ret = btrfs_init_space_info(fs_info);
2376 if (ret) {
2377 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2378 goto fail_block_groups;
2379 }
2380
2381 ret = btrfs_read_block_groups(extent_root);
2382 if (ret) {
2383 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2384 goto fail_block_groups;
2385 }
2386
2387 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2388 "btrfs-cleaner");
2389 if (IS_ERR(fs_info->cleaner_kthread))
2390 goto fail_block_groups;
2391
2392 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2393 tree_root,
2394 "btrfs-transaction");
2395 if (IS_ERR(fs_info->transaction_kthread))
2396 goto fail_cleaner;
2397
2398 if (!btrfs_test_opt(tree_root, SSD) &&
2399 !btrfs_test_opt(tree_root, NOSSD) &&
2400 !fs_info->fs_devices->rotating) {
2401 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2402 "mode\n");
2403 btrfs_set_opt(fs_info->mount_opt, SSD);
2404 }
2405
2406#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2407 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2408 ret = btrfsic_mount(tree_root, fs_devices,
2409 btrfs_test_opt(tree_root,
2410 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2411 1 : 0,
2412 fs_info->check_integrity_print_mask);
2413 if (ret)
2414 printk(KERN_WARNING "btrfs: failed to initialize"
2415 " integrity check module %s\n", sb->s_id);
2416 }
2417#endif
2418
2419 /* do not make disk changes in broken FS */
2420 if (btrfs_super_log_root(disk_super) != 0 &&
2421 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2422 u64 bytenr = btrfs_super_log_root(disk_super);
2423
2424 if (fs_devices->rw_devices == 0) {
2425 printk(KERN_WARNING "Btrfs log replay required "
2426 "on RO media\n");
2427 err = -EIO;
2428 goto fail_trans_kthread;
2429 }
2430 blocksize =
2431 btrfs_level_size(tree_root,
2432 btrfs_super_log_root_level(disk_super));
2433
2434 log_tree_root = btrfs_alloc_root(fs_info);
2435 if (!log_tree_root) {
2436 err = -ENOMEM;
2437 goto fail_trans_kthread;
2438 }
2439
2440 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2441 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2442
2443 log_tree_root->node = read_tree_block(tree_root, bytenr,
2444 blocksize,
2445 generation + 1);
2446 /* returns with log_tree_root freed on success */
2447 ret = btrfs_recover_log_trees(log_tree_root);
2448 if (ret) {
2449 btrfs_error(tree_root->fs_info, ret,
2450 "Failed to recover log tree");
2451 free_extent_buffer(log_tree_root->node);
2452 kfree(log_tree_root);
2453 goto fail_trans_kthread;
2454 }
2455
2456 if (sb->s_flags & MS_RDONLY) {
2457 ret = btrfs_commit_super(tree_root);
2458 if (ret)
2459 goto fail_trans_kthread;
2460 }
2461 }
2462
2463 ret = btrfs_find_orphan_roots(tree_root);
2464 if (ret)
2465 goto fail_trans_kthread;
2466
2467 if (!(sb->s_flags & MS_RDONLY)) {
2468 ret = btrfs_cleanup_fs_roots(fs_info);
2469 if (ret) {
2470 }
2471
2472 ret = btrfs_recover_relocation(tree_root);
2473 if (ret < 0) {
2474 printk(KERN_WARNING
2475 "btrfs: failed to recover relocation\n");
2476 err = -EINVAL;
2477 goto fail_trans_kthread;
2478 }
2479 }
2480
2481 location.objectid = BTRFS_FS_TREE_OBJECTID;
2482 location.type = BTRFS_ROOT_ITEM_KEY;
2483 location.offset = (u64)-1;
2484
2485 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2486 if (!fs_info->fs_root)
2487 goto fail_trans_kthread;
2488 if (IS_ERR(fs_info->fs_root)) {
2489 err = PTR_ERR(fs_info->fs_root);
2490 goto fail_trans_kthread;
2491 }
2492
2493 if (sb->s_flags & MS_RDONLY)
2494 return 0;
2495
2496 down_read(&fs_info->cleanup_work_sem);
2497 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2498 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2499 up_read(&fs_info->cleanup_work_sem);
2500 close_ctree(tree_root);
2501 return ret;
2502 }
2503 up_read(&fs_info->cleanup_work_sem);
2504
2505 ret = btrfs_resume_balance_async(fs_info);
2506 if (ret) {
2507 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2508 close_ctree(tree_root);
2509 return ret;
2510 }
2511
2512 return 0;
2513
2514fail_trans_kthread:
2515 kthread_stop(fs_info->transaction_kthread);
2516fail_cleaner:
2517 kthread_stop(fs_info->cleaner_kthread);
2518
2519 /*
2520 * make sure we're done with the btree inode before we stop our
2521 * kthreads
2522 */
2523 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2524 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2525
2526fail_block_groups:
2527 btrfs_free_block_groups(fs_info);
2528
2529fail_tree_roots:
2530 free_root_pointers(fs_info, 1);
2531
2532fail_sb_buffer:
2533 btrfs_stop_workers(&fs_info->generic_worker);
2534 btrfs_stop_workers(&fs_info->readahead_workers);
2535 btrfs_stop_workers(&fs_info->fixup_workers);
2536 btrfs_stop_workers(&fs_info->delalloc_workers);
2537 btrfs_stop_workers(&fs_info->workers);
2538 btrfs_stop_workers(&fs_info->endio_workers);
2539 btrfs_stop_workers(&fs_info->endio_meta_workers);
2540 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2541 btrfs_stop_workers(&fs_info->endio_write_workers);
2542 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2543 btrfs_stop_workers(&fs_info->submit_workers);
2544 btrfs_stop_workers(&fs_info->delayed_workers);
2545 btrfs_stop_workers(&fs_info->caching_workers);
2546fail_alloc:
2547fail_iput:
2548 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2549
2550 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2551 iput(fs_info->btree_inode);
2552fail_bdi:
2553 bdi_destroy(&fs_info->bdi);
2554fail_srcu:
2555 cleanup_srcu_struct(&fs_info->subvol_srcu);
2556fail:
2557 btrfs_close_devices(fs_info->fs_devices);
2558 return err;
2559
2560recovery_tree_root:
2561 if (!btrfs_test_opt(tree_root, RECOVERY))
2562 goto fail_tree_roots;
2563
2564 free_root_pointers(fs_info, 0);
2565
2566 /* don't use the log in recovery mode, it won't be valid */
2567 btrfs_set_super_log_root(disk_super, 0);
2568
2569 /* we can't trust the free space cache either */
2570 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2571
2572 ret = next_root_backup(fs_info, fs_info->super_copy,
2573 &num_backups_tried, &backup_index);
2574 if (ret == -1)
2575 goto fail_block_groups;
2576 goto retry_root_backup;
2577}
2578
2579static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2580{
2581 if (uptodate) {
2582 set_buffer_uptodate(bh);
2583 } else {
2584 struct btrfs_device *device = (struct btrfs_device *)
2585 bh->b_private;
2586
2587 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2588 "I/O error on %s\n",
2589 rcu_str_deref(device->name));
2590 /* note, we dont' set_buffer_write_io_error because we have
2591 * our own ways of dealing with the IO errors
2592 */
2593 clear_buffer_uptodate(bh);
2594 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
2595 }
2596 unlock_buffer(bh);
2597 put_bh(bh);
2598}
2599
2600struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2601{
2602 struct buffer_head *bh;
2603 struct buffer_head *latest = NULL;
2604 struct btrfs_super_block *super;
2605 int i;
2606 u64 transid = 0;
2607 u64 bytenr;
2608
2609 /* we would like to check all the supers, but that would make
2610 * a btrfs mount succeed after a mkfs from a different FS.
2611 * So, we need to add a special mount option to scan for
2612 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2613 */
2614 for (i = 0; i < 1; i++) {
2615 bytenr = btrfs_sb_offset(i);
2616 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2617 break;
2618 bh = __bread(bdev, bytenr / 4096, 4096);
2619 if (!bh)
2620 continue;
2621
2622 super = (struct btrfs_super_block *)bh->b_data;
2623 if (btrfs_super_bytenr(super) != bytenr ||
2624 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2625 sizeof(super->magic))) {
2626 brelse(bh);
2627 continue;
2628 }
2629
2630 if (!latest || btrfs_super_generation(super) > transid) {
2631 brelse(latest);
2632 latest = bh;
2633 transid = btrfs_super_generation(super);
2634 } else {
2635 brelse(bh);
2636 }
2637 }
2638 return latest;
2639}
2640
2641/*
2642 * this should be called twice, once with wait == 0 and
2643 * once with wait == 1. When wait == 0 is done, all the buffer heads
2644 * we write are pinned.
2645 *
2646 * They are released when wait == 1 is done.
2647 * max_mirrors must be the same for both runs, and it indicates how
2648 * many supers on this one device should be written.
2649 *
2650 * max_mirrors == 0 means to write them all.
2651 */
2652static int write_dev_supers(struct btrfs_device *device,
2653 struct btrfs_super_block *sb,
2654 int do_barriers, int wait, int max_mirrors)
2655{
2656 struct buffer_head *bh;
2657 int i;
2658 int ret;
2659 int errors = 0;
2660 u32 crc;
2661 u64 bytenr;
2662
2663 if (max_mirrors == 0)
2664 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2665
2666 for (i = 0; i < max_mirrors; i++) {
2667 bytenr = btrfs_sb_offset(i);
2668 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2669 break;
2670
2671 if (wait) {
2672 bh = __find_get_block(device->bdev, bytenr / 4096,
2673 BTRFS_SUPER_INFO_SIZE);
2674 BUG_ON(!bh);
2675 wait_on_buffer(bh);
2676 if (!buffer_uptodate(bh))
2677 errors++;
2678
2679 /* drop our reference */
2680 brelse(bh);
2681
2682 /* drop the reference from the wait == 0 run */
2683 brelse(bh);
2684 continue;
2685 } else {
2686 btrfs_set_super_bytenr(sb, bytenr);
2687
2688 crc = ~(u32)0;
2689 crc = btrfs_csum_data(NULL, (char *)sb +
2690 BTRFS_CSUM_SIZE, crc,
2691 BTRFS_SUPER_INFO_SIZE -
2692 BTRFS_CSUM_SIZE);
2693 btrfs_csum_final(crc, sb->csum);
2694
2695 /*
2696 * one reference for us, and we leave it for the
2697 * caller
2698 */
2699 bh = __getblk(device->bdev, bytenr / 4096,
2700 BTRFS_SUPER_INFO_SIZE);
2701 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2702
2703 /* one reference for submit_bh */
2704 get_bh(bh);
2705
2706 set_buffer_uptodate(bh);
2707 lock_buffer(bh);
2708 bh->b_end_io = btrfs_end_buffer_write_sync;
2709 bh->b_private = device;
2710 }
2711
2712 /*
2713 * we fua the first super. The others we allow
2714 * to go down lazy.
2715 */
2716 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2717 if (ret)
2718 errors++;
2719 }
2720 return errors < i ? 0 : -1;
2721}
2722
2723/*
2724 * endio for the write_dev_flush, this will wake anyone waiting
2725 * for the barrier when it is done
2726 */
2727static void btrfs_end_empty_barrier(struct bio *bio, int err)
2728{
2729 if (err) {
2730 if (err == -EOPNOTSUPP)
2731 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2732 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2733 }
2734 if (bio->bi_private)
2735 complete(bio->bi_private);
2736 bio_put(bio);
2737}
2738
2739/*
2740 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2741 * sent down. With wait == 1, it waits for the previous flush.
2742 *
2743 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2744 * capable
2745 */
2746static int write_dev_flush(struct btrfs_device *device, int wait)
2747{
2748 struct bio *bio;
2749 int ret = 0;
2750
2751 if (device->nobarriers)
2752 return 0;
2753
2754 if (wait) {
2755 bio = device->flush_bio;
2756 if (!bio)
2757 return 0;
2758
2759 wait_for_completion(&device->flush_wait);
2760
2761 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2762 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
2763 rcu_str_deref(device->name));
2764 device->nobarriers = 1;
2765 }
2766 if (!bio_flagged(bio, BIO_UPTODATE)) {
2767 ret = -EIO;
2768 if (!bio_flagged(bio, BIO_EOPNOTSUPP))
2769 btrfs_dev_stat_inc_and_print(device,
2770 BTRFS_DEV_STAT_FLUSH_ERRS);
2771 }
2772
2773 /* drop the reference from the wait == 0 run */
2774 bio_put(bio);
2775 device->flush_bio = NULL;
2776
2777 return ret;
2778 }
2779
2780 /*
2781 * one reference for us, and we leave it for the
2782 * caller
2783 */
2784 device->flush_bio = NULL;
2785 bio = bio_alloc(GFP_NOFS, 0);
2786 if (!bio)
2787 return -ENOMEM;
2788
2789 bio->bi_end_io = btrfs_end_empty_barrier;
2790 bio->bi_bdev = device->bdev;
2791 init_completion(&device->flush_wait);
2792 bio->bi_private = &device->flush_wait;
2793 device->flush_bio = bio;
2794
2795 bio_get(bio);
2796 btrfsic_submit_bio(WRITE_FLUSH, bio);
2797
2798 return 0;
2799}
2800
2801/*
2802 * send an empty flush down to each device in parallel,
2803 * then wait for them
2804 */
2805static int barrier_all_devices(struct btrfs_fs_info *info)
2806{
2807 struct list_head *head;
2808 struct btrfs_device *dev;
2809 int errors = 0;
2810 int ret;
2811
2812 /* send down all the barriers */
2813 head = &info->fs_devices->devices;
2814 list_for_each_entry_rcu(dev, head, dev_list) {
2815 if (!dev->bdev) {
2816 errors++;
2817 continue;
2818 }
2819 if (!dev->in_fs_metadata || !dev->writeable)
2820 continue;
2821
2822 ret = write_dev_flush(dev, 0);
2823 if (ret)
2824 errors++;
2825 }
2826
2827 /* wait for all the barriers */
2828 list_for_each_entry_rcu(dev, head, dev_list) {
2829 if (!dev->bdev) {
2830 errors++;
2831 continue;
2832 }
2833 if (!dev->in_fs_metadata || !dev->writeable)
2834 continue;
2835
2836 ret = write_dev_flush(dev, 1);
2837 if (ret)
2838 errors++;
2839 }
2840 if (errors)
2841 return -EIO;
2842 return 0;
2843}
2844
2845int write_all_supers(struct btrfs_root *root, int max_mirrors)
2846{
2847 struct list_head *head;
2848 struct btrfs_device *dev;
2849 struct btrfs_super_block *sb;
2850 struct btrfs_dev_item *dev_item;
2851 int ret;
2852 int do_barriers;
2853 int max_errors;
2854 int total_errors = 0;
2855 u64 flags;
2856
2857 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2858 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2859 backup_super_roots(root->fs_info);
2860
2861 sb = root->fs_info->super_for_commit;
2862 dev_item = &sb->dev_item;
2863
2864 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2865 head = &root->fs_info->fs_devices->devices;
2866
2867 if (do_barriers)
2868 barrier_all_devices(root->fs_info);
2869
2870 list_for_each_entry_rcu(dev, head, dev_list) {
2871 if (!dev->bdev) {
2872 total_errors++;
2873 continue;
2874 }
2875 if (!dev->in_fs_metadata || !dev->writeable)
2876 continue;
2877
2878 btrfs_set_stack_device_generation(dev_item, 0);
2879 btrfs_set_stack_device_type(dev_item, dev->type);
2880 btrfs_set_stack_device_id(dev_item, dev->devid);
2881 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2882 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2883 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2884 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2885 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2886 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2887 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2888
2889 flags = btrfs_super_flags(sb);
2890 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2891
2892 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2893 if (ret)
2894 total_errors++;
2895 }
2896 if (total_errors > max_errors) {
2897 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2898 total_errors);
2899
2900 /* This shouldn't happen. FUA is masked off if unsupported */
2901 BUG();
2902 }
2903
2904 total_errors = 0;
2905 list_for_each_entry_rcu(dev, head, dev_list) {
2906 if (!dev->bdev)
2907 continue;
2908 if (!dev->in_fs_metadata || !dev->writeable)
2909 continue;
2910
2911 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2912 if (ret)
2913 total_errors++;
2914 }
2915 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2916 if (total_errors > max_errors) {
2917 btrfs_error(root->fs_info, -EIO,
2918 "%d errors while writing supers", total_errors);
2919 return -EIO;
2920 }
2921 return 0;
2922}
2923
2924int write_ctree_super(struct btrfs_trans_handle *trans,
2925 struct btrfs_root *root, int max_mirrors)
2926{
2927 int ret;
2928
2929 ret = write_all_supers(root, max_mirrors);
2930 return ret;
2931}
2932
2933void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2934{
2935 spin_lock(&fs_info->fs_roots_radix_lock);
2936 radix_tree_delete(&fs_info->fs_roots_radix,
2937 (unsigned long)root->root_key.objectid);
2938 spin_unlock(&fs_info->fs_roots_radix_lock);
2939
2940 if (btrfs_root_refs(&root->root_item) == 0)
2941 synchronize_srcu(&fs_info->subvol_srcu);
2942
2943 __btrfs_remove_free_space_cache(root->free_ino_pinned);
2944 __btrfs_remove_free_space_cache(root->free_ino_ctl);
2945 free_fs_root(root);
2946}
2947
2948static void free_fs_root(struct btrfs_root *root)
2949{
2950 iput(root->cache_inode);
2951 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2952 if (root->anon_dev)
2953 free_anon_bdev(root->anon_dev);
2954 free_extent_buffer(root->node);
2955 free_extent_buffer(root->commit_root);
2956 kfree(root->free_ino_ctl);
2957 kfree(root->free_ino_pinned);
2958 kfree(root->name);
2959 kfree(root);
2960}
2961
2962static void del_fs_roots(struct btrfs_fs_info *fs_info)
2963{
2964 int ret;
2965 struct btrfs_root *gang[8];
2966 int i;
2967
2968 while (!list_empty(&fs_info->dead_roots)) {
2969 gang[0] = list_entry(fs_info->dead_roots.next,
2970 struct btrfs_root, root_list);
2971 list_del(&gang[0]->root_list);
2972
2973 if (gang[0]->in_radix) {
2974 btrfs_free_fs_root(fs_info, gang[0]);
2975 } else {
2976 free_extent_buffer(gang[0]->node);
2977 free_extent_buffer(gang[0]->commit_root);
2978 kfree(gang[0]);
2979 }
2980 }
2981
2982 while (1) {
2983 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2984 (void **)gang, 0,
2985 ARRAY_SIZE(gang));
2986 if (!ret)
2987 break;
2988 for (i = 0; i < ret; i++)
2989 btrfs_free_fs_root(fs_info, gang[i]);
2990 }
2991}
2992
2993int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2994{
2995 u64 root_objectid = 0;
2996 struct btrfs_root *gang[8];
2997 int i;
2998 int ret;
2999
3000 while (1) {
3001 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3002 (void **)gang, root_objectid,
3003 ARRAY_SIZE(gang));
3004 if (!ret)
3005 break;
3006
3007 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3008 for (i = 0; i < ret; i++) {
3009 int err;
3010
3011 root_objectid = gang[i]->root_key.objectid;
3012 err = btrfs_orphan_cleanup(gang[i]);
3013 if (err)
3014 return err;
3015 }
3016 root_objectid++;
3017 }
3018 return 0;
3019}
3020
3021int btrfs_commit_super(struct btrfs_root *root)
3022{
3023 struct btrfs_trans_handle *trans;
3024 int ret;
3025
3026 mutex_lock(&root->fs_info->cleaner_mutex);
3027 btrfs_run_delayed_iputs(root);
3028 btrfs_clean_old_snapshots(root);
3029 mutex_unlock(&root->fs_info->cleaner_mutex);
3030
3031 /* wait until ongoing cleanup work done */
3032 down_write(&root->fs_info->cleanup_work_sem);
3033 up_write(&root->fs_info->cleanup_work_sem);
3034
3035 trans = btrfs_join_transaction(root);
3036 if (IS_ERR(trans))
3037 return PTR_ERR(trans);
3038 ret = btrfs_commit_transaction(trans, root);
3039 if (ret)
3040 return ret;
3041 /* run commit again to drop the original snapshot */
3042 trans = btrfs_join_transaction(root);
3043 if (IS_ERR(trans))
3044 return PTR_ERR(trans);
3045 ret = btrfs_commit_transaction(trans, root);
3046 if (ret)
3047 return ret;
3048 ret = btrfs_write_and_wait_transaction(NULL, root);
3049 if (ret) {
3050 btrfs_error(root->fs_info, ret,
3051 "Failed to sync btree inode to disk.");
3052 return ret;
3053 }
3054
3055 ret = write_ctree_super(NULL, root, 0);
3056 return ret;
3057}
3058
3059int close_ctree(struct btrfs_root *root)
3060{
3061 struct btrfs_fs_info *fs_info = root->fs_info;
3062 int ret;
3063
3064 fs_info->closing = 1;
3065 smp_mb();
3066
3067 /* pause restriper - we want to resume on mount */
3068 btrfs_pause_balance(root->fs_info);
3069
3070 btrfs_scrub_cancel(root);
3071
3072 /* wait for any defraggers to finish */
3073 wait_event(fs_info->transaction_wait,
3074 (atomic_read(&fs_info->defrag_running) == 0));
3075
3076 /* clear out the rbtree of defraggable inodes */
3077 btrfs_run_defrag_inodes(fs_info);
3078
3079 /*
3080 * Here come 2 situations when btrfs is broken to flip readonly:
3081 *
3082 * 1. when btrfs flips readonly somewhere else before
3083 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
3084 * and btrfs will skip to write sb directly to keep
3085 * ERROR state on disk.
3086 *
3087 * 2. when btrfs flips readonly just in btrfs_commit_super,
3088 * and in such case, btrfs cannot write sb via btrfs_commit_super,
3089 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
3090 * btrfs will cleanup all FS resources first and write sb then.
3091 */
3092 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3093 ret = btrfs_commit_super(root);
3094 if (ret)
3095 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3096 }
3097
3098 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
3099 ret = btrfs_error_commit_super(root);
3100 if (ret)
3101 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3102 }
3103
3104 btrfs_put_block_group_cache(fs_info);
3105
3106 kthread_stop(fs_info->transaction_kthread);
3107 kthread_stop(fs_info->cleaner_kthread);
3108
3109 fs_info->closing = 2;
3110 smp_mb();
3111
3112 if (fs_info->delalloc_bytes) {
3113 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
3114 (unsigned long long)fs_info->delalloc_bytes);
3115 }
3116 if (fs_info->total_ref_cache_size) {
3117 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
3118 (unsigned long long)fs_info->total_ref_cache_size);
3119 }
3120
3121 free_extent_buffer(fs_info->extent_root->node);
3122 free_extent_buffer(fs_info->extent_root->commit_root);
3123 free_extent_buffer(fs_info->tree_root->node);
3124 free_extent_buffer(fs_info->tree_root->commit_root);
3125 free_extent_buffer(fs_info->chunk_root->node);
3126 free_extent_buffer(fs_info->chunk_root->commit_root);
3127 free_extent_buffer(fs_info->dev_root->node);
3128 free_extent_buffer(fs_info->dev_root->commit_root);
3129 free_extent_buffer(fs_info->csum_root->node);
3130 free_extent_buffer(fs_info->csum_root->commit_root);
3131
3132 btrfs_free_block_groups(fs_info);
3133
3134 del_fs_roots(fs_info);
3135
3136 iput(fs_info->btree_inode);
3137
3138 btrfs_stop_workers(&fs_info->generic_worker);
3139 btrfs_stop_workers(&fs_info->fixup_workers);
3140 btrfs_stop_workers(&fs_info->delalloc_workers);
3141 btrfs_stop_workers(&fs_info->workers);
3142 btrfs_stop_workers(&fs_info->endio_workers);
3143 btrfs_stop_workers(&fs_info->endio_meta_workers);
3144 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
3145 btrfs_stop_workers(&fs_info->endio_write_workers);
3146 btrfs_stop_workers(&fs_info->endio_freespace_worker);
3147 btrfs_stop_workers(&fs_info->submit_workers);
3148 btrfs_stop_workers(&fs_info->delayed_workers);
3149 btrfs_stop_workers(&fs_info->caching_workers);
3150 btrfs_stop_workers(&fs_info->readahead_workers);
3151
3152#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3153 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3154 btrfsic_unmount(root, fs_info->fs_devices);
3155#endif
3156
3157 btrfs_close_devices(fs_info->fs_devices);
3158 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3159
3160 bdi_destroy(&fs_info->bdi);
3161 cleanup_srcu_struct(&fs_info->subvol_srcu);
3162
3163 return 0;
3164}
3165
3166int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3167 int atomic)
3168{
3169 int ret;
3170 struct inode *btree_inode = buf->pages[0]->mapping->host;
3171
3172 ret = extent_buffer_uptodate(buf);
3173 if (!ret)
3174 return ret;
3175
3176 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3177 parent_transid, atomic);
3178 if (ret == -EAGAIN)
3179 return ret;
3180 return !ret;
3181}
3182
3183int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3184{
3185 return set_extent_buffer_uptodate(buf);
3186}
3187
3188void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3189{
3190 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3191 u64 transid = btrfs_header_generation(buf);
3192 int was_dirty;
3193
3194 btrfs_assert_tree_locked(buf);
3195 if (transid != root->fs_info->generation) {
3196 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
3197 "found %llu running %llu\n",
3198 (unsigned long long)buf->start,
3199 (unsigned long long)transid,
3200 (unsigned long long)root->fs_info->generation);
3201 WARN_ON(1);
3202 }
3203 was_dirty = set_extent_buffer_dirty(buf);
3204 if (!was_dirty) {
3205 spin_lock(&root->fs_info->delalloc_lock);
3206 root->fs_info->dirty_metadata_bytes += buf->len;
3207 spin_unlock(&root->fs_info->delalloc_lock);
3208 }
3209}
3210
3211void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3212{
3213 /*
3214 * looks as though older kernels can get into trouble with
3215 * this code, they end up stuck in balance_dirty_pages forever
3216 */
3217 u64 num_dirty;
3218 unsigned long thresh = 32 * 1024 * 1024;
3219
3220 if (current->flags & PF_MEMALLOC)
3221 return;
3222
3223 btrfs_balance_delayed_items(root);
3224
3225 num_dirty = root->fs_info->dirty_metadata_bytes;
3226
3227 if (num_dirty > thresh) {
3228 balance_dirty_pages_ratelimited_nr(
3229 root->fs_info->btree_inode->i_mapping, 1);
3230 }
3231 return;
3232}
3233
3234void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3235{
3236 /*
3237 * looks as though older kernels can get into trouble with
3238 * this code, they end up stuck in balance_dirty_pages forever
3239 */
3240 u64 num_dirty;
3241 unsigned long thresh = 32 * 1024 * 1024;
3242
3243 if (current->flags & PF_MEMALLOC)
3244 return;
3245
3246 num_dirty = root->fs_info->dirty_metadata_bytes;
3247
3248 if (num_dirty > thresh) {
3249 balance_dirty_pages_ratelimited_nr(
3250 root->fs_info->btree_inode->i_mapping, 1);
3251 }
3252 return;
3253}
3254
3255int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3256{
3257 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3258 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3259}
3260
3261static int btree_lock_page_hook(struct page *page, void *data,
3262 void (*flush_fn)(void *))
3263{
3264 struct inode *inode = page->mapping->host;
3265 struct btrfs_root *root = BTRFS_I(inode)->root;
3266 struct extent_buffer *eb;
3267
3268 /*
3269 * We culled this eb but the page is still hanging out on the mapping,
3270 * carry on.
3271 */
3272 if (!PagePrivate(page))
3273 goto out;
3274
3275 eb = (struct extent_buffer *)page->private;
3276 if (!eb) {
3277 WARN_ON(1);
3278 goto out;
3279 }
3280 if (page != eb->pages[0])
3281 goto out;
3282
3283 if (!btrfs_try_tree_write_lock(eb)) {
3284 flush_fn(data);
3285 btrfs_tree_lock(eb);
3286 }
3287 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3288
3289 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3290 spin_lock(&root->fs_info->delalloc_lock);
3291 if (root->fs_info->dirty_metadata_bytes >= eb->len)
3292 root->fs_info->dirty_metadata_bytes -= eb->len;
3293 else
3294 WARN_ON(1);
3295 spin_unlock(&root->fs_info->delalloc_lock);
3296 }
3297
3298 btrfs_tree_unlock(eb);
3299out:
3300 if (!trylock_page(page)) {
3301 flush_fn(data);
3302 lock_page(page);
3303 }
3304 return 0;
3305}
3306
3307static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3308 int read_only)
3309{
3310 if (btrfs_super_csum_type(fs_info->super_copy) >= ARRAY_SIZE(btrfs_csum_sizes)) {
3311 printk(KERN_ERR "btrfs: unsupported checksum algorithm\n");
3312 return -EINVAL;
3313 }
3314
3315 if (read_only)
3316 return 0;
3317
3318 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
3319 printk(KERN_WARNING "warning: mount fs with errors, "
3320 "running btrfsck is recommended\n");
3321 }
3322
3323 return 0;
3324}
3325
3326int btrfs_error_commit_super(struct btrfs_root *root)
3327{
3328 int ret;
3329
3330 mutex_lock(&root->fs_info->cleaner_mutex);
3331 btrfs_run_delayed_iputs(root);
3332 mutex_unlock(&root->fs_info->cleaner_mutex);
3333
3334 down_write(&root->fs_info->cleanup_work_sem);
3335 up_write(&root->fs_info->cleanup_work_sem);
3336
3337 /* cleanup FS via transaction */
3338 btrfs_cleanup_transaction(root);
3339
3340 ret = write_ctree_super(NULL, root, 0);
3341
3342 return ret;
3343}
3344
3345static void btrfs_destroy_ordered_operations(struct btrfs_root *root)
3346{
3347 struct btrfs_inode *btrfs_inode;
3348 struct list_head splice;
3349
3350 INIT_LIST_HEAD(&splice);
3351
3352 mutex_lock(&root->fs_info->ordered_operations_mutex);
3353 spin_lock(&root->fs_info->ordered_extent_lock);
3354
3355 list_splice_init(&root->fs_info->ordered_operations, &splice);
3356 while (!list_empty(&splice)) {
3357 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3358 ordered_operations);
3359
3360 list_del_init(&btrfs_inode->ordered_operations);
3361
3362 btrfs_invalidate_inodes(btrfs_inode->root);
3363 }
3364
3365 spin_unlock(&root->fs_info->ordered_extent_lock);
3366 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3367}
3368
3369static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3370{
3371 struct list_head splice;
3372 struct btrfs_ordered_extent *ordered;
3373 struct inode *inode;
3374
3375 INIT_LIST_HEAD(&splice);
3376
3377 spin_lock(&root->fs_info->ordered_extent_lock);
3378
3379 list_splice_init(&root->fs_info->ordered_extents, &splice);
3380 while (!list_empty(&splice)) {
3381 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3382 root_extent_list);
3383
3384 list_del_init(&ordered->root_extent_list);
3385 atomic_inc(&ordered->refs);
3386
3387 /* the inode may be getting freed (in sys_unlink path). */
3388 inode = igrab(ordered->inode);
3389
3390 spin_unlock(&root->fs_info->ordered_extent_lock);
3391 if (inode)
3392 iput(inode);
3393
3394 atomic_set(&ordered->refs, 1);
3395 btrfs_put_ordered_extent(ordered);
3396
3397 spin_lock(&root->fs_info->ordered_extent_lock);
3398 }
3399
3400 spin_unlock(&root->fs_info->ordered_extent_lock);
3401}
3402
3403int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3404 struct btrfs_root *root)
3405{
3406 struct rb_node *node;
3407 struct btrfs_delayed_ref_root *delayed_refs;
3408 struct btrfs_delayed_ref_node *ref;
3409 int ret = 0;
3410
3411 delayed_refs = &trans->delayed_refs;
3412
3413 spin_lock(&delayed_refs->lock);
3414 if (delayed_refs->num_entries == 0) {
3415 spin_unlock(&delayed_refs->lock);
3416 printk(KERN_INFO "delayed_refs has NO entry\n");
3417 return ret;
3418 }
3419
3420 while ((node = rb_first(&delayed_refs->root)) != NULL) {
3421 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3422
3423 atomic_set(&ref->refs, 1);
3424 if (btrfs_delayed_ref_is_head(ref)) {
3425 struct btrfs_delayed_ref_head *head;
3426
3427 head = btrfs_delayed_node_to_head(ref);
3428 if (!mutex_trylock(&head->mutex)) {
3429 atomic_inc(&ref->refs);
3430 spin_unlock(&delayed_refs->lock);
3431
3432 /* Need to wait for the delayed ref to run */
3433 mutex_lock(&head->mutex);
3434 mutex_unlock(&head->mutex);
3435 btrfs_put_delayed_ref(ref);
3436
3437 spin_lock(&delayed_refs->lock);
3438 continue;
3439 }
3440
3441 kfree(head->extent_op);
3442 delayed_refs->num_heads--;
3443 if (list_empty(&head->cluster))
3444 delayed_refs->num_heads_ready--;
3445 list_del_init(&head->cluster);
3446 }
3447 ref->in_tree = 0;
3448 rb_erase(&ref->rb_node, &delayed_refs->root);
3449 delayed_refs->num_entries--;
3450
3451 spin_unlock(&delayed_refs->lock);
3452 btrfs_put_delayed_ref(ref);
3453
3454 cond_resched();
3455 spin_lock(&delayed_refs->lock);
3456 }
3457
3458 spin_unlock(&delayed_refs->lock);
3459
3460 return ret;
3461}
3462
3463static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3464{
3465 struct btrfs_pending_snapshot *snapshot;
3466 struct list_head splice;
3467
3468 INIT_LIST_HEAD(&splice);
3469
3470 list_splice_init(&t->pending_snapshots, &splice);
3471
3472 while (!list_empty(&splice)) {
3473 snapshot = list_entry(splice.next,
3474 struct btrfs_pending_snapshot,
3475 list);
3476
3477 list_del_init(&snapshot->list);
3478
3479 kfree(snapshot);
3480 }
3481}
3482
3483static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3484{
3485 struct btrfs_inode *btrfs_inode;
3486 struct list_head splice;
3487
3488 INIT_LIST_HEAD(&splice);
3489
3490 spin_lock(&root->fs_info->delalloc_lock);
3491 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3492
3493 while (!list_empty(&splice)) {
3494 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3495 delalloc_inodes);
3496
3497 list_del_init(&btrfs_inode->delalloc_inodes);
3498
3499 btrfs_invalidate_inodes(btrfs_inode->root);
3500 }
3501
3502 spin_unlock(&root->fs_info->delalloc_lock);
3503}
3504
3505static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3506 struct extent_io_tree *dirty_pages,
3507 int mark)
3508{
3509 int ret;
3510 struct page *page;
3511 struct inode *btree_inode = root->fs_info->btree_inode;
3512 struct extent_buffer *eb;
3513 u64 start = 0;
3514 u64 end;
3515 u64 offset;
3516 unsigned long index;
3517
3518 while (1) {
3519 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3520 mark);
3521 if (ret)
3522 break;
3523
3524 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3525 while (start <= end) {
3526 index = start >> PAGE_CACHE_SHIFT;
3527 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3528 page = find_get_page(btree_inode->i_mapping, index);
3529 if (!page)
3530 continue;
3531 offset = page_offset(page);
3532
3533 spin_lock(&dirty_pages->buffer_lock);
3534 eb = radix_tree_lookup(
3535 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3536 offset >> PAGE_CACHE_SHIFT);
3537 spin_unlock(&dirty_pages->buffer_lock);
3538 if (eb)
3539 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3540 &eb->bflags);
3541 if (PageWriteback(page))
3542 end_page_writeback(page);
3543
3544 lock_page(page);
3545 if (PageDirty(page)) {
3546 clear_page_dirty_for_io(page);
3547 spin_lock_irq(&page->mapping->tree_lock);
3548 radix_tree_tag_clear(&page->mapping->page_tree,
3549 page_index(page),
3550 PAGECACHE_TAG_DIRTY);
3551 spin_unlock_irq(&page->mapping->tree_lock);
3552 }
3553
3554 unlock_page(page);
3555 page_cache_release(page);
3556 }
3557 }
3558
3559 return ret;
3560}
3561
3562static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3563 struct extent_io_tree *pinned_extents)
3564{
3565 struct extent_io_tree *unpin;
3566 u64 start;
3567 u64 end;
3568 int ret;
3569 bool loop = true;
3570
3571 unpin = pinned_extents;
3572again:
3573 while (1) {
3574 ret = find_first_extent_bit(unpin, 0, &start, &end,
3575 EXTENT_DIRTY);
3576 if (ret)
3577 break;
3578
3579 /* opt_discard */
3580 if (btrfs_test_opt(root, DISCARD))
3581 ret = btrfs_error_discard_extent(root, start,
3582 end + 1 - start,
3583 NULL);
3584
3585 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3586 btrfs_error_unpin_extent_range(root, start, end);
3587 cond_resched();
3588 }
3589
3590 if (loop) {
3591 if (unpin == &root->fs_info->freed_extents[0])
3592 unpin = &root->fs_info->freed_extents[1];
3593 else
3594 unpin = &root->fs_info->freed_extents[0];
3595 loop = false;
3596 goto again;
3597 }
3598
3599 return 0;
3600}
3601
3602void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3603 struct btrfs_root *root)
3604{
3605 btrfs_destroy_delayed_refs(cur_trans, root);
3606 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3607 cur_trans->dirty_pages.dirty_bytes);
3608
3609 /* FIXME: cleanup wait for commit */
3610 cur_trans->in_commit = 1;
3611 cur_trans->blocked = 1;
3612 wake_up(&root->fs_info->transaction_blocked_wait);
3613
3614 cur_trans->blocked = 0;
3615 wake_up(&root->fs_info->transaction_wait);
3616
3617 cur_trans->commit_done = 1;
3618 wake_up(&cur_trans->commit_wait);
3619
3620 btrfs_destroy_delayed_inodes(root);
3621 btrfs_assert_delayed_root_empty(root);
3622
3623 btrfs_destroy_pending_snapshots(cur_trans);
3624
3625 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3626 EXTENT_DIRTY);
3627 btrfs_destroy_pinned_extent(root,
3628 root->fs_info->pinned_extents);
3629
3630 /*
3631 memset(cur_trans, 0, sizeof(*cur_trans));
3632 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3633 */
3634}
3635
3636int btrfs_cleanup_transaction(struct btrfs_root *root)
3637{
3638 struct btrfs_transaction *t;
3639 LIST_HEAD(list);
3640
3641 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3642
3643 spin_lock(&root->fs_info->trans_lock);
3644 list_splice_init(&root->fs_info->trans_list, &list);
3645 root->fs_info->trans_no_join = 1;
3646 spin_unlock(&root->fs_info->trans_lock);
3647
3648 while (!list_empty(&list)) {
3649 t = list_entry(list.next, struct btrfs_transaction, list);
3650 if (!t)
3651 break;
3652
3653 btrfs_destroy_ordered_operations(root);
3654
3655 btrfs_destroy_ordered_extents(root);
3656
3657 btrfs_destroy_delayed_refs(t, root);
3658
3659 btrfs_block_rsv_release(root,
3660 &root->fs_info->trans_block_rsv,
3661 t->dirty_pages.dirty_bytes);
3662
3663 /* FIXME: cleanup wait for commit */
3664 t->in_commit = 1;
3665 t->blocked = 1;
3666 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3667 wake_up(&root->fs_info->transaction_blocked_wait);
3668
3669 t->blocked = 0;
3670 if (waitqueue_active(&root->fs_info->transaction_wait))
3671 wake_up(&root->fs_info->transaction_wait);
3672
3673 t->commit_done = 1;
3674 if (waitqueue_active(&t->commit_wait))
3675 wake_up(&t->commit_wait);
3676
3677 btrfs_destroy_delayed_inodes(root);
3678 btrfs_assert_delayed_root_empty(root);
3679
3680 btrfs_destroy_pending_snapshots(t);
3681
3682 btrfs_destroy_delalloc_inodes(root);
3683
3684 spin_lock(&root->fs_info->trans_lock);
3685 root->fs_info->running_transaction = NULL;
3686 spin_unlock(&root->fs_info->trans_lock);
3687
3688 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3689 EXTENT_DIRTY);
3690
3691 btrfs_destroy_pinned_extent(root,
3692 root->fs_info->pinned_extents);
3693
3694 atomic_set(&t->use_count, 0);
3695 list_del_init(&t->list);
3696 memset(t, 0, sizeof(*t));
3697 kmem_cache_free(btrfs_transaction_cachep, t);
3698 }
3699
3700 spin_lock(&root->fs_info->trans_lock);
3701 root->fs_info->trans_no_join = 0;
3702 spin_unlock(&root->fs_info->trans_lock);
3703 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3704
3705 return 0;
3706}
3707
3708static struct extent_io_ops btree_extent_io_ops = {
3709 .write_cache_pages_lock_hook = btree_lock_page_hook,
3710 .readpage_end_io_hook = btree_readpage_end_io_hook,
3711 .readpage_io_failed_hook = btree_io_failed_hook,
3712 .submit_bio_hook = btree_submit_bio_hook,
3713 /* note we're sharing with inode.c for the merge bio hook */
3714 .merge_bio_hook = btrfs_merge_bio_hook,
3715};