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 "check-integrity.h"
33#include "rcu-string.h"
34#include "dev-replace.h"
35#include "raid56.h"
36#include "sysfs.h"
37#include "qgroup.h"
38#include "compression.h"
39#include "tree-checker.h"
40#include "ref-verify.h"
41#include "block-group.h"
42#include "discard.h"
43#include "space-info.h"
44#include "zoned.h"
45#include "subpage.h"
46
47#define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
48 BTRFS_HEADER_FLAG_RELOC |\
49 BTRFS_SUPER_FLAG_ERROR |\
50 BTRFS_SUPER_FLAG_SEEDING |\
51 BTRFS_SUPER_FLAG_METADUMP |\
52 BTRFS_SUPER_FLAG_METADUMP_V2)
53
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 struct inode *inode;
114 struct bio *bio;
115 extent_submit_bio_start_t *submit_bio_start;
116 int mirror_num;
117
118 /* Optional parameter for submit_bio_start used by direct io */
119 u64 dio_file_offset;
120 struct btrfs_work work;
121 blk_status_t status;
122};
123
124/*
125 * Lockdep class keys for extent_buffer->lock's in this root. For a given
126 * eb, the lockdep key is determined by the btrfs_root it belongs to and
127 * the level the eb occupies in the tree.
128 *
129 * Different roots are used for different purposes and may nest inside each
130 * other and they require separate keysets. As lockdep keys should be
131 * static, assign keysets according to the purpose of the root as indicated
132 * by btrfs_root->root_key.objectid. This ensures that all special purpose
133 * roots have separate keysets.
134 *
135 * Lock-nesting across peer nodes is always done with the immediate parent
136 * node locked thus preventing deadlock. As lockdep doesn't know this, use
137 * subclass to avoid triggering lockdep warning in such cases.
138 *
139 * The key is set by the readpage_end_io_hook after the buffer has passed
140 * csum validation but before the pages are unlocked. It is also set by
141 * btrfs_init_new_buffer on freshly allocated blocks.
142 *
143 * We also add a check to make sure the highest level of the tree is the
144 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
145 * needs update as well.
146 */
147#ifdef CONFIG_DEBUG_LOCK_ALLOC
148# if BTRFS_MAX_LEVEL != 8
149# error
150# endif
151
152#define DEFINE_LEVEL(stem, level) \
153 .names[level] = "btrfs-" stem "-0" #level,
154
155#define DEFINE_NAME(stem) \
156 DEFINE_LEVEL(stem, 0) \
157 DEFINE_LEVEL(stem, 1) \
158 DEFINE_LEVEL(stem, 2) \
159 DEFINE_LEVEL(stem, 3) \
160 DEFINE_LEVEL(stem, 4) \
161 DEFINE_LEVEL(stem, 5) \
162 DEFINE_LEVEL(stem, 6) \
163 DEFINE_LEVEL(stem, 7)
164
165static struct btrfs_lockdep_keyset {
166 u64 id; /* root objectid */
167 /* Longest entry: btrfs-free-space-00 */
168 char names[BTRFS_MAX_LEVEL][20];
169 struct lock_class_key keys[BTRFS_MAX_LEVEL];
170} btrfs_lockdep_keysets[] = {
171 { .id = BTRFS_ROOT_TREE_OBJECTID, DEFINE_NAME("root") },
172 { .id = BTRFS_EXTENT_TREE_OBJECTID, DEFINE_NAME("extent") },
173 { .id = BTRFS_CHUNK_TREE_OBJECTID, DEFINE_NAME("chunk") },
174 { .id = BTRFS_DEV_TREE_OBJECTID, DEFINE_NAME("dev") },
175 { .id = BTRFS_CSUM_TREE_OBJECTID, DEFINE_NAME("csum") },
176 { .id = BTRFS_QUOTA_TREE_OBJECTID, DEFINE_NAME("quota") },
177 { .id = BTRFS_TREE_LOG_OBJECTID, DEFINE_NAME("log") },
178 { .id = BTRFS_TREE_RELOC_OBJECTID, DEFINE_NAME("treloc") },
179 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, DEFINE_NAME("dreloc") },
180 { .id = BTRFS_UUID_TREE_OBJECTID, DEFINE_NAME("uuid") },
181 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, DEFINE_NAME("free-space") },
182 { .id = 0, DEFINE_NAME("tree") },
183};
184
185#undef DEFINE_LEVEL
186#undef DEFINE_NAME
187
188void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
189 int level)
190{
191 struct btrfs_lockdep_keyset *ks;
192
193 BUG_ON(level >= ARRAY_SIZE(ks->keys));
194
195 /* find the matching keyset, id 0 is the default entry */
196 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
197 if (ks->id == objectid)
198 break;
199
200 lockdep_set_class_and_name(&eb->lock,
201 &ks->keys[level], ks->names[level]);
202}
203
204#endif
205
206/*
207 * Compute the csum of a btree block and store the result to provided buffer.
208 */
209static void csum_tree_block(struct extent_buffer *buf, u8 *result)
210{
211 struct btrfs_fs_info *fs_info = buf->fs_info;
212 const int num_pages = num_extent_pages(buf);
213 const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
214 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
215 char *kaddr;
216 int i;
217
218 shash->tfm = fs_info->csum_shash;
219 crypto_shash_init(shash);
220 kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
221 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
222 first_page_part - BTRFS_CSUM_SIZE);
223
224 for (i = 1; i < num_pages; i++) {
225 kaddr = page_address(buf->pages[i]);
226 crypto_shash_update(shash, kaddr, PAGE_SIZE);
227 }
228 memset(result, 0, BTRFS_CSUM_SIZE);
229 crypto_shash_final(shash, result);
230}
231
232/*
233 * we can't consider a given block up to date unless the transid of the
234 * block matches the transid in the parent node's pointer. This is how we
235 * detect blocks that either didn't get written at all or got written
236 * in the wrong place.
237 */
238static int verify_parent_transid(struct extent_io_tree *io_tree,
239 struct extent_buffer *eb, u64 parent_transid,
240 int atomic)
241{
242 struct extent_state *cached_state = NULL;
243 int ret;
244
245 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
246 return 0;
247
248 if (atomic)
249 return -EAGAIN;
250
251 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
252 &cached_state);
253 if (extent_buffer_uptodate(eb) &&
254 btrfs_header_generation(eb) == parent_transid) {
255 ret = 0;
256 goto out;
257 }
258 btrfs_err_rl(eb->fs_info,
259 "parent transid verify failed on %llu wanted %llu found %llu",
260 eb->start,
261 parent_transid, btrfs_header_generation(eb));
262 ret = 1;
263 clear_extent_buffer_uptodate(eb);
264out:
265 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
266 &cached_state);
267 return ret;
268}
269
270static bool btrfs_supported_super_csum(u16 csum_type)
271{
272 switch (csum_type) {
273 case BTRFS_CSUM_TYPE_CRC32:
274 case BTRFS_CSUM_TYPE_XXHASH:
275 case BTRFS_CSUM_TYPE_SHA256:
276 case BTRFS_CSUM_TYPE_BLAKE2:
277 return true;
278 default:
279 return false;
280 }
281}
282
283/*
284 * Return 0 if the superblock checksum type matches the checksum value of that
285 * algorithm. Pass the raw disk superblock data.
286 */
287static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
288 char *raw_disk_sb)
289{
290 struct btrfs_super_block *disk_sb =
291 (struct btrfs_super_block *)raw_disk_sb;
292 char result[BTRFS_CSUM_SIZE];
293 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
294
295 shash->tfm = fs_info->csum_shash;
296
297 /*
298 * The super_block structure does not span the whole
299 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
300 * filled with zeros and is included in the checksum.
301 */
302 crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
303 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
304
305 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
306 return 1;
307
308 return 0;
309}
310
311int btrfs_verify_level_key(struct extent_buffer *eb, int level,
312 struct btrfs_key *first_key, u64 parent_transid)
313{
314 struct btrfs_fs_info *fs_info = eb->fs_info;
315 int found_level;
316 struct btrfs_key found_key;
317 int ret;
318
319 found_level = btrfs_header_level(eb);
320 if (found_level != level) {
321 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
322 KERN_ERR "BTRFS: tree level check failed\n");
323 btrfs_err(fs_info,
324"tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
325 eb->start, level, found_level);
326 return -EIO;
327 }
328
329 if (!first_key)
330 return 0;
331
332 /*
333 * For live tree block (new tree blocks in current transaction),
334 * we need proper lock context to avoid race, which is impossible here.
335 * So we only checks tree blocks which is read from disk, whose
336 * generation <= fs_info->last_trans_committed.
337 */
338 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
339 return 0;
340
341 /* We have @first_key, so this @eb must have at least one item */
342 if (btrfs_header_nritems(eb) == 0) {
343 btrfs_err(fs_info,
344 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
345 eb->start);
346 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
347 return -EUCLEAN;
348 }
349
350 if (found_level)
351 btrfs_node_key_to_cpu(eb, &found_key, 0);
352 else
353 btrfs_item_key_to_cpu(eb, &found_key, 0);
354 ret = btrfs_comp_cpu_keys(first_key, &found_key);
355
356 if (ret) {
357 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
358 KERN_ERR "BTRFS: tree first key check failed\n");
359 btrfs_err(fs_info,
360"tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
361 eb->start, parent_transid, first_key->objectid,
362 first_key->type, first_key->offset,
363 found_key.objectid, found_key.type,
364 found_key.offset);
365 }
366 return ret;
367}
368
369/*
370 * helper to read a given tree block, doing retries as required when
371 * the checksums don't match and we have alternate mirrors to try.
372 *
373 * @parent_transid: expected transid, skip check if 0
374 * @level: expected level, mandatory check
375 * @first_key: expected key of first slot, skip check if NULL
376 */
377static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
378 u64 parent_transid, int level,
379 struct btrfs_key *first_key)
380{
381 struct btrfs_fs_info *fs_info = eb->fs_info;
382 struct extent_io_tree *io_tree;
383 int failed = 0;
384 int ret;
385 int num_copies = 0;
386 int mirror_num = 0;
387 int failed_mirror = 0;
388
389 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
390 while (1) {
391 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
392 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
393 if (!ret) {
394 if (verify_parent_transid(io_tree, eb,
395 parent_transid, 0))
396 ret = -EIO;
397 else if (btrfs_verify_level_key(eb, level,
398 first_key, parent_transid))
399 ret = -EUCLEAN;
400 else
401 break;
402 }
403
404 num_copies = btrfs_num_copies(fs_info,
405 eb->start, eb->len);
406 if (num_copies == 1)
407 break;
408
409 if (!failed_mirror) {
410 failed = 1;
411 failed_mirror = eb->read_mirror;
412 }
413
414 mirror_num++;
415 if (mirror_num == failed_mirror)
416 mirror_num++;
417
418 if (mirror_num > num_copies)
419 break;
420 }
421
422 if (failed && !ret && failed_mirror)
423 btrfs_repair_eb_io_failure(eb, failed_mirror);
424
425 return ret;
426}
427
428static int csum_one_extent_buffer(struct extent_buffer *eb)
429{
430 struct btrfs_fs_info *fs_info = eb->fs_info;
431 u8 result[BTRFS_CSUM_SIZE];
432 int ret;
433
434 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
435 offsetof(struct btrfs_header, fsid),
436 BTRFS_FSID_SIZE) == 0);
437 csum_tree_block(eb, result);
438
439 if (btrfs_header_level(eb))
440 ret = btrfs_check_node(eb);
441 else
442 ret = btrfs_check_leaf_full(eb);
443
444 if (ret < 0) {
445 btrfs_print_tree(eb, 0);
446 btrfs_err(fs_info,
447 "block=%llu write time tree block corruption detected",
448 eb->start);
449 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
450 return ret;
451 }
452 write_extent_buffer(eb, result, 0, fs_info->csum_size);
453
454 return 0;
455}
456
457/* Checksum all dirty extent buffers in one bio_vec */
458static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info,
459 struct bio_vec *bvec)
460{
461 struct page *page = bvec->bv_page;
462 u64 bvec_start = page_offset(page) + bvec->bv_offset;
463 u64 cur;
464 int ret = 0;
465
466 for (cur = bvec_start; cur < bvec_start + bvec->bv_len;
467 cur += fs_info->nodesize) {
468 struct extent_buffer *eb;
469 bool uptodate;
470
471 eb = find_extent_buffer(fs_info, cur);
472 uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur,
473 fs_info->nodesize);
474
475 /* A dirty eb shouldn't disappear from buffer_radix */
476 if (WARN_ON(!eb))
477 return -EUCLEAN;
478
479 if (WARN_ON(cur != btrfs_header_bytenr(eb))) {
480 free_extent_buffer(eb);
481 return -EUCLEAN;
482 }
483 if (WARN_ON(!uptodate)) {
484 free_extent_buffer(eb);
485 return -EUCLEAN;
486 }
487
488 ret = csum_one_extent_buffer(eb);
489 free_extent_buffer(eb);
490 if (ret < 0)
491 return ret;
492 }
493 return ret;
494}
495
496/*
497 * Checksum a dirty tree block before IO. This has extra checks to make sure
498 * we only fill in the checksum field in the first page of a multi-page block.
499 * For subpage extent buffers we need bvec to also read the offset in the page.
500 */
501static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
502{
503 struct page *page = bvec->bv_page;
504 u64 start = page_offset(page);
505 u64 found_start;
506 struct extent_buffer *eb;
507
508 if (fs_info->sectorsize < PAGE_SIZE)
509 return csum_dirty_subpage_buffers(fs_info, bvec);
510
511 eb = (struct extent_buffer *)page->private;
512 if (page != eb->pages[0])
513 return 0;
514
515 found_start = btrfs_header_bytenr(eb);
516
517 if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
518 WARN_ON(found_start != 0);
519 return 0;
520 }
521
522 /*
523 * Please do not consolidate these warnings into a single if.
524 * It is useful to know what went wrong.
525 */
526 if (WARN_ON(found_start != start))
527 return -EUCLEAN;
528 if (WARN_ON(!PageUptodate(page)))
529 return -EUCLEAN;
530
531 return csum_one_extent_buffer(eb);
532}
533
534static int check_tree_block_fsid(struct extent_buffer *eb)
535{
536 struct btrfs_fs_info *fs_info = eb->fs_info;
537 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
538 u8 fsid[BTRFS_FSID_SIZE];
539 u8 *metadata_uuid;
540
541 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
542 BTRFS_FSID_SIZE);
543 /*
544 * Checking the incompat flag is only valid for the current fs. For
545 * seed devices it's forbidden to have their uuid changed so reading
546 * ->fsid in this case is fine
547 */
548 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
549 metadata_uuid = fs_devices->metadata_uuid;
550 else
551 metadata_uuid = fs_devices->fsid;
552
553 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
554 return 0;
555
556 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
557 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
558 return 0;
559
560 return 1;
561}
562
563/* Do basic extent buffer checks at read time */
564static int validate_extent_buffer(struct extent_buffer *eb)
565{
566 struct btrfs_fs_info *fs_info = eb->fs_info;
567 u64 found_start;
568 const u32 csum_size = fs_info->csum_size;
569 u8 found_level;
570 u8 result[BTRFS_CSUM_SIZE];
571 const u8 *header_csum;
572 int ret = 0;
573
574 found_start = btrfs_header_bytenr(eb);
575 if (found_start != eb->start) {
576 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
577 eb->start, found_start);
578 ret = -EIO;
579 goto out;
580 }
581 if (check_tree_block_fsid(eb)) {
582 btrfs_err_rl(fs_info, "bad fsid on block %llu",
583 eb->start);
584 ret = -EIO;
585 goto out;
586 }
587 found_level = btrfs_header_level(eb);
588 if (found_level >= BTRFS_MAX_LEVEL) {
589 btrfs_err(fs_info, "bad tree block level %d on %llu",
590 (int)btrfs_header_level(eb), eb->start);
591 ret = -EIO;
592 goto out;
593 }
594
595 csum_tree_block(eb, result);
596 header_csum = page_address(eb->pages[0]) +
597 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
598
599 if (memcmp(result, header_csum, csum_size) != 0) {
600 btrfs_warn_rl(fs_info,
601 "checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d",
602 eb->start,
603 CSUM_FMT_VALUE(csum_size, header_csum),
604 CSUM_FMT_VALUE(csum_size, result),
605 btrfs_header_level(eb));
606 ret = -EUCLEAN;
607 goto out;
608 }
609
610 /*
611 * If this is a leaf block and it is corrupt, set the corrupt bit so
612 * that we don't try and read the other copies of this block, just
613 * return -EIO.
614 */
615 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
616 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
617 ret = -EIO;
618 }
619
620 if (found_level > 0 && btrfs_check_node(eb))
621 ret = -EIO;
622
623 if (!ret)
624 set_extent_buffer_uptodate(eb);
625 else
626 btrfs_err(fs_info,
627 "block=%llu read time tree block corruption detected",
628 eb->start);
629out:
630 return ret;
631}
632
633static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
634 int mirror)
635{
636 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
637 struct extent_buffer *eb;
638 bool reads_done;
639 int ret = 0;
640
641 /*
642 * We don't allow bio merge for subpage metadata read, so we should
643 * only get one eb for each endio hook.
644 */
645 ASSERT(end == start + fs_info->nodesize - 1);
646 ASSERT(PagePrivate(page));
647
648 eb = find_extent_buffer(fs_info, start);
649 /*
650 * When we are reading one tree block, eb must have been inserted into
651 * the radix tree. If not, something is wrong.
652 */
653 ASSERT(eb);
654
655 reads_done = atomic_dec_and_test(&eb->io_pages);
656 /* Subpage read must finish in page read */
657 ASSERT(reads_done);
658
659 eb->read_mirror = mirror;
660 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
661 ret = -EIO;
662 goto err;
663 }
664 ret = validate_extent_buffer(eb);
665 if (ret < 0)
666 goto err;
667
668 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
669 btree_readahead_hook(eb, ret);
670
671 set_extent_buffer_uptodate(eb);
672
673 free_extent_buffer(eb);
674 return ret;
675err:
676 /*
677 * end_bio_extent_readpage decrements io_pages in case of error,
678 * make sure it has something to decrement.
679 */
680 atomic_inc(&eb->io_pages);
681 clear_extent_buffer_uptodate(eb);
682 free_extent_buffer(eb);
683 return ret;
684}
685
686int btrfs_validate_metadata_buffer(struct btrfs_io_bio *io_bio,
687 struct page *page, u64 start, u64 end,
688 int mirror)
689{
690 struct extent_buffer *eb;
691 int ret = 0;
692 int reads_done;
693
694 ASSERT(page->private);
695
696 if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE)
697 return validate_subpage_buffer(page, start, end, mirror);
698
699 eb = (struct extent_buffer *)page->private;
700
701 /*
702 * The pending IO might have been the only thing that kept this buffer
703 * in memory. Make sure we have a ref for all this other checks
704 */
705 atomic_inc(&eb->refs);
706
707 reads_done = atomic_dec_and_test(&eb->io_pages);
708 if (!reads_done)
709 goto err;
710
711 eb->read_mirror = mirror;
712 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
713 ret = -EIO;
714 goto err;
715 }
716 ret = validate_extent_buffer(eb);
717err:
718 if (reads_done &&
719 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
720 btree_readahead_hook(eb, ret);
721
722 if (ret) {
723 /*
724 * our io error hook is going to dec the io pages
725 * again, we have to make sure it has something
726 * to decrement
727 */
728 atomic_inc(&eb->io_pages);
729 clear_extent_buffer_uptodate(eb);
730 }
731 free_extent_buffer(eb);
732
733 return ret;
734}
735
736static void end_workqueue_bio(struct bio *bio)
737{
738 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
739 struct btrfs_fs_info *fs_info;
740 struct btrfs_workqueue *wq;
741
742 fs_info = end_io_wq->info;
743 end_io_wq->status = bio->bi_status;
744
745 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
746 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
747 wq = fs_info->endio_meta_write_workers;
748 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
749 wq = fs_info->endio_freespace_worker;
750 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
751 wq = fs_info->endio_raid56_workers;
752 else
753 wq = fs_info->endio_write_workers;
754 } else {
755 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
756 wq = fs_info->endio_raid56_workers;
757 else if (end_io_wq->metadata)
758 wq = fs_info->endio_meta_workers;
759 else
760 wq = fs_info->endio_workers;
761 }
762
763 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
764 btrfs_queue_work(wq, &end_io_wq->work);
765}
766
767blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
768 enum btrfs_wq_endio_type metadata)
769{
770 struct btrfs_end_io_wq *end_io_wq;
771
772 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
773 if (!end_io_wq)
774 return BLK_STS_RESOURCE;
775
776 end_io_wq->private = bio->bi_private;
777 end_io_wq->end_io = bio->bi_end_io;
778 end_io_wq->info = info;
779 end_io_wq->status = 0;
780 end_io_wq->bio = bio;
781 end_io_wq->metadata = metadata;
782
783 bio->bi_private = end_io_wq;
784 bio->bi_end_io = end_workqueue_bio;
785 return 0;
786}
787
788static void run_one_async_start(struct btrfs_work *work)
789{
790 struct async_submit_bio *async;
791 blk_status_t ret;
792
793 async = container_of(work, struct async_submit_bio, work);
794 ret = async->submit_bio_start(async->inode, async->bio,
795 async->dio_file_offset);
796 if (ret)
797 async->status = ret;
798}
799
800/*
801 * In order to insert checksums into the metadata in large chunks, we wait
802 * until bio submission time. All the pages in the bio are checksummed and
803 * sums are attached onto the ordered extent record.
804 *
805 * At IO completion time the csums attached on the ordered extent record are
806 * inserted into the tree.
807 */
808static void run_one_async_done(struct btrfs_work *work)
809{
810 struct async_submit_bio *async;
811 struct inode *inode;
812 blk_status_t ret;
813
814 async = container_of(work, struct async_submit_bio, work);
815 inode = async->inode;
816
817 /* If an error occurred we just want to clean up the bio and move on */
818 if (async->status) {
819 async->bio->bi_status = async->status;
820 bio_endio(async->bio);
821 return;
822 }
823
824 /*
825 * All of the bios that pass through here are from async helpers.
826 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
827 * This changes nothing when cgroups aren't in use.
828 */
829 async->bio->bi_opf |= REQ_CGROUP_PUNT;
830 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
831 if (ret) {
832 async->bio->bi_status = ret;
833 bio_endio(async->bio);
834 }
835}
836
837static void run_one_async_free(struct btrfs_work *work)
838{
839 struct async_submit_bio *async;
840
841 async = container_of(work, struct async_submit_bio, work);
842 kfree(async);
843}
844
845blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio,
846 int mirror_num, unsigned long bio_flags,
847 u64 dio_file_offset,
848 extent_submit_bio_start_t *submit_bio_start)
849{
850 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
851 struct async_submit_bio *async;
852
853 async = kmalloc(sizeof(*async), GFP_NOFS);
854 if (!async)
855 return BLK_STS_RESOURCE;
856
857 async->inode = inode;
858 async->bio = bio;
859 async->mirror_num = mirror_num;
860 async->submit_bio_start = submit_bio_start;
861
862 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
863 run_one_async_free);
864
865 async->dio_file_offset = dio_file_offset;
866
867 async->status = 0;
868
869 if (op_is_sync(bio->bi_opf))
870 btrfs_set_work_high_priority(&async->work);
871
872 btrfs_queue_work(fs_info->workers, &async->work);
873 return 0;
874}
875
876static blk_status_t btree_csum_one_bio(struct bio *bio)
877{
878 struct bio_vec *bvec;
879 struct btrfs_root *root;
880 int ret = 0;
881 struct bvec_iter_all iter_all;
882
883 ASSERT(!bio_flagged(bio, BIO_CLONED));
884 bio_for_each_segment_all(bvec, bio, iter_all) {
885 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
886 ret = csum_dirty_buffer(root->fs_info, bvec);
887 if (ret)
888 break;
889 }
890
891 return errno_to_blk_status(ret);
892}
893
894static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio,
895 u64 dio_file_offset)
896{
897 /*
898 * when we're called for a write, we're already in the async
899 * submission context. Just jump into btrfs_map_bio
900 */
901 return btree_csum_one_bio(bio);
902}
903
904static bool should_async_write(struct btrfs_fs_info *fs_info,
905 struct btrfs_inode *bi)
906{
907 if (btrfs_is_zoned(fs_info))
908 return false;
909 if (atomic_read(&bi->sync_writers))
910 return false;
911 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
912 return false;
913 return true;
914}
915
916blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio,
917 int mirror_num, unsigned long bio_flags)
918{
919 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
920 blk_status_t ret;
921
922 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
923 /*
924 * called for a read, do the setup so that checksum validation
925 * can happen in the async kernel threads
926 */
927 ret = btrfs_bio_wq_end_io(fs_info, bio,
928 BTRFS_WQ_ENDIO_METADATA);
929 if (ret)
930 goto out_w_error;
931 ret = btrfs_map_bio(fs_info, bio, mirror_num);
932 } else if (!should_async_write(fs_info, BTRFS_I(inode))) {
933 ret = btree_csum_one_bio(bio);
934 if (ret)
935 goto out_w_error;
936 ret = btrfs_map_bio(fs_info, bio, mirror_num);
937 } else {
938 /*
939 * kthread helpers are used to submit writes so that
940 * checksumming can happen in parallel across all CPUs
941 */
942 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0,
943 0, btree_submit_bio_start);
944 }
945
946 if (ret)
947 goto out_w_error;
948 return 0;
949
950out_w_error:
951 bio->bi_status = ret;
952 bio_endio(bio);
953 return ret;
954}
955
956#ifdef CONFIG_MIGRATION
957static int btree_migratepage(struct address_space *mapping,
958 struct page *newpage, struct page *page,
959 enum migrate_mode mode)
960{
961 /*
962 * we can't safely write a btree page from here,
963 * we haven't done the locking hook
964 */
965 if (PageDirty(page))
966 return -EAGAIN;
967 /*
968 * Buffers may be managed in a filesystem specific way.
969 * We must have no buffers or drop them.
970 */
971 if (page_has_private(page) &&
972 !try_to_release_page(page, GFP_KERNEL))
973 return -EAGAIN;
974 return migrate_page(mapping, newpage, page, mode);
975}
976#endif
977
978
979static int btree_writepages(struct address_space *mapping,
980 struct writeback_control *wbc)
981{
982 struct btrfs_fs_info *fs_info;
983 int ret;
984
985 if (wbc->sync_mode == WB_SYNC_NONE) {
986
987 if (wbc->for_kupdate)
988 return 0;
989
990 fs_info = BTRFS_I(mapping->host)->root->fs_info;
991 /* this is a bit racy, but that's ok */
992 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
993 BTRFS_DIRTY_METADATA_THRESH,
994 fs_info->dirty_metadata_batch);
995 if (ret < 0)
996 return 0;
997 }
998 return btree_write_cache_pages(mapping, wbc);
999}
1000
1001static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1002{
1003 if (PageWriteback(page) || PageDirty(page))
1004 return 0;
1005
1006 return try_release_extent_buffer(page);
1007}
1008
1009static void btree_invalidatepage(struct page *page, unsigned int offset,
1010 unsigned int length)
1011{
1012 struct extent_io_tree *tree;
1013 tree = &BTRFS_I(page->mapping->host)->io_tree;
1014 extent_invalidatepage(tree, page, offset);
1015 btree_releasepage(page, GFP_NOFS);
1016 if (PagePrivate(page)) {
1017 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1018 "page private not zero on page %llu",
1019 (unsigned long long)page_offset(page));
1020 detach_page_private(page);
1021 }
1022}
1023
1024static int btree_set_page_dirty(struct page *page)
1025{
1026#ifdef DEBUG
1027 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
1028 struct btrfs_subpage *subpage;
1029 struct extent_buffer *eb;
1030 int cur_bit = 0;
1031 u64 page_start = page_offset(page);
1032
1033 if (fs_info->sectorsize == PAGE_SIZE) {
1034 BUG_ON(!PagePrivate(page));
1035 eb = (struct extent_buffer *)page->private;
1036 BUG_ON(!eb);
1037 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1038 BUG_ON(!atomic_read(&eb->refs));
1039 btrfs_assert_tree_locked(eb);
1040 return __set_page_dirty_nobuffers(page);
1041 }
1042 ASSERT(PagePrivate(page) && page->private);
1043 subpage = (struct btrfs_subpage *)page->private;
1044
1045 ASSERT(subpage->dirty_bitmap);
1046 while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) {
1047 unsigned long flags;
1048 u64 cur;
1049 u16 tmp = (1 << cur_bit);
1050
1051 spin_lock_irqsave(&subpage->lock, flags);
1052 if (!(tmp & subpage->dirty_bitmap)) {
1053 spin_unlock_irqrestore(&subpage->lock, flags);
1054 cur_bit++;
1055 continue;
1056 }
1057 spin_unlock_irqrestore(&subpage->lock, flags);
1058 cur = page_start + cur_bit * fs_info->sectorsize;
1059
1060 eb = find_extent_buffer(fs_info, cur);
1061 ASSERT(eb);
1062 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1063 ASSERT(atomic_read(&eb->refs));
1064 btrfs_assert_tree_locked(eb);
1065 free_extent_buffer(eb);
1066
1067 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits);
1068 }
1069#endif
1070 return __set_page_dirty_nobuffers(page);
1071}
1072
1073static const struct address_space_operations btree_aops = {
1074 .writepages = btree_writepages,
1075 .releasepage = btree_releasepage,
1076 .invalidatepage = btree_invalidatepage,
1077#ifdef CONFIG_MIGRATION
1078 .migratepage = btree_migratepage,
1079#endif
1080 .set_page_dirty = btree_set_page_dirty,
1081};
1082
1083struct extent_buffer *btrfs_find_create_tree_block(
1084 struct btrfs_fs_info *fs_info,
1085 u64 bytenr, u64 owner_root,
1086 int level)
1087{
1088 if (btrfs_is_testing(fs_info))
1089 return alloc_test_extent_buffer(fs_info, bytenr);
1090 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
1091}
1092
1093/*
1094 * Read tree block at logical address @bytenr and do variant basic but critical
1095 * verification.
1096 *
1097 * @owner_root: the objectid of the root owner for this block.
1098 * @parent_transid: expected transid of this tree block, skip check if 0
1099 * @level: expected level, mandatory check
1100 * @first_key: expected key in slot 0, skip check if NULL
1101 */
1102struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1103 u64 owner_root, u64 parent_transid,
1104 int level, struct btrfs_key *first_key)
1105{
1106 struct extent_buffer *buf = NULL;
1107 int ret;
1108
1109 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
1110 if (IS_ERR(buf))
1111 return buf;
1112
1113 ret = btree_read_extent_buffer_pages(buf, parent_transid,
1114 level, first_key);
1115 if (ret) {
1116 free_extent_buffer_stale(buf);
1117 return ERR_PTR(ret);
1118 }
1119 return buf;
1120
1121}
1122
1123void btrfs_clean_tree_block(struct extent_buffer *buf)
1124{
1125 struct btrfs_fs_info *fs_info = buf->fs_info;
1126 if (btrfs_header_generation(buf) ==
1127 fs_info->running_transaction->transid) {
1128 btrfs_assert_tree_locked(buf);
1129
1130 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1131 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1132 -buf->len,
1133 fs_info->dirty_metadata_batch);
1134 clear_extent_buffer_dirty(buf);
1135 }
1136 }
1137}
1138
1139static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1140 u64 objectid)
1141{
1142 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1143 root->fs_info = fs_info;
1144 root->node = NULL;
1145 root->commit_root = NULL;
1146 root->state = 0;
1147 root->orphan_cleanup_state = 0;
1148
1149 root->last_trans = 0;
1150 root->free_objectid = 0;
1151 root->nr_delalloc_inodes = 0;
1152 root->nr_ordered_extents = 0;
1153 root->inode_tree = RB_ROOT;
1154 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1155 root->block_rsv = NULL;
1156
1157 INIT_LIST_HEAD(&root->dirty_list);
1158 INIT_LIST_HEAD(&root->root_list);
1159 INIT_LIST_HEAD(&root->delalloc_inodes);
1160 INIT_LIST_HEAD(&root->delalloc_root);
1161 INIT_LIST_HEAD(&root->ordered_extents);
1162 INIT_LIST_HEAD(&root->ordered_root);
1163 INIT_LIST_HEAD(&root->reloc_dirty_list);
1164 INIT_LIST_HEAD(&root->logged_list[0]);
1165 INIT_LIST_HEAD(&root->logged_list[1]);
1166 spin_lock_init(&root->inode_lock);
1167 spin_lock_init(&root->delalloc_lock);
1168 spin_lock_init(&root->ordered_extent_lock);
1169 spin_lock_init(&root->accounting_lock);
1170 spin_lock_init(&root->log_extents_lock[0]);
1171 spin_lock_init(&root->log_extents_lock[1]);
1172 spin_lock_init(&root->qgroup_meta_rsv_lock);
1173 mutex_init(&root->objectid_mutex);
1174 mutex_init(&root->log_mutex);
1175 mutex_init(&root->ordered_extent_mutex);
1176 mutex_init(&root->delalloc_mutex);
1177 init_waitqueue_head(&root->qgroup_flush_wait);
1178 init_waitqueue_head(&root->log_writer_wait);
1179 init_waitqueue_head(&root->log_commit_wait[0]);
1180 init_waitqueue_head(&root->log_commit_wait[1]);
1181 INIT_LIST_HEAD(&root->log_ctxs[0]);
1182 INIT_LIST_HEAD(&root->log_ctxs[1]);
1183 atomic_set(&root->log_commit[0], 0);
1184 atomic_set(&root->log_commit[1], 0);
1185 atomic_set(&root->log_writers, 0);
1186 atomic_set(&root->log_batch, 0);
1187 refcount_set(&root->refs, 1);
1188 atomic_set(&root->snapshot_force_cow, 0);
1189 atomic_set(&root->nr_swapfiles, 0);
1190 root->log_transid = 0;
1191 root->log_transid_committed = -1;
1192 root->last_log_commit = 0;
1193 if (!dummy) {
1194 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1195 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1196 extent_io_tree_init(fs_info, &root->log_csum_range,
1197 IO_TREE_LOG_CSUM_RANGE, NULL);
1198 }
1199
1200 memset(&root->root_key, 0, sizeof(root->root_key));
1201 memset(&root->root_item, 0, sizeof(root->root_item));
1202 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1203 root->root_key.objectid = objectid;
1204 root->anon_dev = 0;
1205
1206 spin_lock_init(&root->root_item_lock);
1207 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1208#ifdef CONFIG_BTRFS_DEBUG
1209 INIT_LIST_HEAD(&root->leak_list);
1210 spin_lock(&fs_info->fs_roots_radix_lock);
1211 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1212 spin_unlock(&fs_info->fs_roots_radix_lock);
1213#endif
1214}
1215
1216static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1217 u64 objectid, gfp_t flags)
1218{
1219 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1220 if (root)
1221 __setup_root(root, fs_info, objectid);
1222 return root;
1223}
1224
1225#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1226/* Should only be used by the testing infrastructure */
1227struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1228{
1229 struct btrfs_root *root;
1230
1231 if (!fs_info)
1232 return ERR_PTR(-EINVAL);
1233
1234 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1235 if (!root)
1236 return ERR_PTR(-ENOMEM);
1237
1238 /* We don't use the stripesize in selftest, set it as sectorsize */
1239 root->alloc_bytenr = 0;
1240
1241 return root;
1242}
1243#endif
1244
1245struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1246 u64 objectid)
1247{
1248 struct btrfs_fs_info *fs_info = trans->fs_info;
1249 struct extent_buffer *leaf;
1250 struct btrfs_root *tree_root = fs_info->tree_root;
1251 struct btrfs_root *root;
1252 struct btrfs_key key;
1253 unsigned int nofs_flag;
1254 int ret = 0;
1255
1256 /*
1257 * We're holding a transaction handle, so use a NOFS memory allocation
1258 * context to avoid deadlock if reclaim happens.
1259 */
1260 nofs_flag = memalloc_nofs_save();
1261 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1262 memalloc_nofs_restore(nofs_flag);
1263 if (!root)
1264 return ERR_PTR(-ENOMEM);
1265
1266 root->root_key.objectid = objectid;
1267 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1268 root->root_key.offset = 0;
1269
1270 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1271 BTRFS_NESTING_NORMAL);
1272 if (IS_ERR(leaf)) {
1273 ret = PTR_ERR(leaf);
1274 leaf = NULL;
1275 goto fail_unlock;
1276 }
1277
1278 root->node = leaf;
1279 btrfs_mark_buffer_dirty(leaf);
1280
1281 root->commit_root = btrfs_root_node(root);
1282 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1283
1284 btrfs_set_root_flags(&root->root_item, 0);
1285 btrfs_set_root_limit(&root->root_item, 0);
1286 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1287 btrfs_set_root_generation(&root->root_item, trans->transid);
1288 btrfs_set_root_level(&root->root_item, 0);
1289 btrfs_set_root_refs(&root->root_item, 1);
1290 btrfs_set_root_used(&root->root_item, leaf->len);
1291 btrfs_set_root_last_snapshot(&root->root_item, 0);
1292 btrfs_set_root_dirid(&root->root_item, 0);
1293 if (is_fstree(objectid))
1294 generate_random_guid(root->root_item.uuid);
1295 else
1296 export_guid(root->root_item.uuid, &guid_null);
1297 btrfs_set_root_drop_level(&root->root_item, 0);
1298
1299 btrfs_tree_unlock(leaf);
1300
1301 key.objectid = objectid;
1302 key.type = BTRFS_ROOT_ITEM_KEY;
1303 key.offset = 0;
1304 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1305 if (ret)
1306 goto fail;
1307
1308 return root;
1309
1310fail_unlock:
1311 if (leaf)
1312 btrfs_tree_unlock(leaf);
1313fail:
1314 btrfs_put_root(root);
1315
1316 return ERR_PTR(ret);
1317}
1318
1319static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1320 struct btrfs_fs_info *fs_info)
1321{
1322 struct btrfs_root *root;
1323
1324 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1325 if (!root)
1326 return ERR_PTR(-ENOMEM);
1327
1328 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1329 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1330 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1331
1332 return root;
1333}
1334
1335int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
1336 struct btrfs_root *root)
1337{
1338 struct extent_buffer *leaf;
1339
1340 /*
1341 * DON'T set SHAREABLE bit for log trees.
1342 *
1343 * Log trees are not exposed to user space thus can't be snapshotted,
1344 * and they go away before a real commit is actually done.
1345 *
1346 * They do store pointers to file data extents, and those reference
1347 * counts still get updated (along with back refs to the log tree).
1348 */
1349
1350 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1351 NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1352 if (IS_ERR(leaf))
1353 return PTR_ERR(leaf);
1354
1355 root->node = leaf;
1356
1357 btrfs_mark_buffer_dirty(root->node);
1358 btrfs_tree_unlock(root->node);
1359
1360 return 0;
1361}
1362
1363int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1364 struct btrfs_fs_info *fs_info)
1365{
1366 struct btrfs_root *log_root;
1367
1368 log_root = alloc_log_tree(trans, fs_info);
1369 if (IS_ERR(log_root))
1370 return PTR_ERR(log_root);
1371
1372 if (!btrfs_is_zoned(fs_info)) {
1373 int ret = btrfs_alloc_log_tree_node(trans, log_root);
1374
1375 if (ret) {
1376 btrfs_put_root(log_root);
1377 return ret;
1378 }
1379 }
1380
1381 WARN_ON(fs_info->log_root_tree);
1382 fs_info->log_root_tree = log_root;
1383 return 0;
1384}
1385
1386int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1387 struct btrfs_root *root)
1388{
1389 struct btrfs_fs_info *fs_info = root->fs_info;
1390 struct btrfs_root *log_root;
1391 struct btrfs_inode_item *inode_item;
1392 int ret;
1393
1394 log_root = alloc_log_tree(trans, fs_info);
1395 if (IS_ERR(log_root))
1396 return PTR_ERR(log_root);
1397
1398 ret = btrfs_alloc_log_tree_node(trans, log_root);
1399 if (ret) {
1400 btrfs_put_root(log_root);
1401 return ret;
1402 }
1403
1404 log_root->last_trans = trans->transid;
1405 log_root->root_key.offset = root->root_key.objectid;
1406
1407 inode_item = &log_root->root_item.inode;
1408 btrfs_set_stack_inode_generation(inode_item, 1);
1409 btrfs_set_stack_inode_size(inode_item, 3);
1410 btrfs_set_stack_inode_nlink(inode_item, 1);
1411 btrfs_set_stack_inode_nbytes(inode_item,
1412 fs_info->nodesize);
1413 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1414
1415 btrfs_set_root_node(&log_root->root_item, log_root->node);
1416
1417 WARN_ON(root->log_root);
1418 root->log_root = log_root;
1419 root->log_transid = 0;
1420 root->log_transid_committed = -1;
1421 root->last_log_commit = 0;
1422 return 0;
1423}
1424
1425static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1426 struct btrfs_path *path,
1427 struct btrfs_key *key)
1428{
1429 struct btrfs_root *root;
1430 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1431 u64 generation;
1432 int ret;
1433 int level;
1434
1435 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1436 if (!root)
1437 return ERR_PTR(-ENOMEM);
1438
1439 ret = btrfs_find_root(tree_root, key, path,
1440 &root->root_item, &root->root_key);
1441 if (ret) {
1442 if (ret > 0)
1443 ret = -ENOENT;
1444 goto fail;
1445 }
1446
1447 generation = btrfs_root_generation(&root->root_item);
1448 level = btrfs_root_level(&root->root_item);
1449 root->node = read_tree_block(fs_info,
1450 btrfs_root_bytenr(&root->root_item),
1451 key->objectid, generation, level, NULL);
1452 if (IS_ERR(root->node)) {
1453 ret = PTR_ERR(root->node);
1454 root->node = NULL;
1455 goto fail;
1456 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1457 ret = -EIO;
1458 goto fail;
1459 }
1460 root->commit_root = btrfs_root_node(root);
1461 return root;
1462fail:
1463 btrfs_put_root(root);
1464 return ERR_PTR(ret);
1465}
1466
1467struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1468 struct btrfs_key *key)
1469{
1470 struct btrfs_root *root;
1471 struct btrfs_path *path;
1472
1473 path = btrfs_alloc_path();
1474 if (!path)
1475 return ERR_PTR(-ENOMEM);
1476 root = read_tree_root_path(tree_root, path, key);
1477 btrfs_free_path(path);
1478
1479 return root;
1480}
1481
1482/*
1483 * Initialize subvolume root in-memory structure
1484 *
1485 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1486 */
1487static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1488{
1489 int ret;
1490 unsigned int nofs_flag;
1491
1492 /*
1493 * We might be called under a transaction (e.g. indirect backref
1494 * resolution) which could deadlock if it triggers memory reclaim
1495 */
1496 nofs_flag = memalloc_nofs_save();
1497 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1498 memalloc_nofs_restore(nofs_flag);
1499 if (ret)
1500 goto fail;
1501
1502 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1503 root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
1504 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1505 btrfs_check_and_init_root_item(&root->root_item);
1506 }
1507
1508 /*
1509 * Don't assign anonymous block device to roots that are not exposed to
1510 * userspace, the id pool is limited to 1M
1511 */
1512 if (is_fstree(root->root_key.objectid) &&
1513 btrfs_root_refs(&root->root_item) > 0) {
1514 if (!anon_dev) {
1515 ret = get_anon_bdev(&root->anon_dev);
1516 if (ret)
1517 goto fail;
1518 } else {
1519 root->anon_dev = anon_dev;
1520 }
1521 }
1522
1523 mutex_lock(&root->objectid_mutex);
1524 ret = btrfs_init_root_free_objectid(root);
1525 if (ret) {
1526 mutex_unlock(&root->objectid_mutex);
1527 goto fail;
1528 }
1529
1530 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1531
1532 mutex_unlock(&root->objectid_mutex);
1533
1534 return 0;
1535fail:
1536 /* The caller is responsible to call btrfs_free_fs_root */
1537 return ret;
1538}
1539
1540static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1541 u64 root_id)
1542{
1543 struct btrfs_root *root;
1544
1545 spin_lock(&fs_info->fs_roots_radix_lock);
1546 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1547 (unsigned long)root_id);
1548 if (root)
1549 root = btrfs_grab_root(root);
1550 spin_unlock(&fs_info->fs_roots_radix_lock);
1551 return root;
1552}
1553
1554static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1555 u64 objectid)
1556{
1557 if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1558 return btrfs_grab_root(fs_info->tree_root);
1559 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1560 return btrfs_grab_root(fs_info->extent_root);
1561 if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1562 return btrfs_grab_root(fs_info->chunk_root);
1563 if (objectid == BTRFS_DEV_TREE_OBJECTID)
1564 return btrfs_grab_root(fs_info->dev_root);
1565 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1566 return btrfs_grab_root(fs_info->csum_root);
1567 if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1568 return btrfs_grab_root(fs_info->quota_root) ?
1569 fs_info->quota_root : ERR_PTR(-ENOENT);
1570 if (objectid == BTRFS_UUID_TREE_OBJECTID)
1571 return btrfs_grab_root(fs_info->uuid_root) ?
1572 fs_info->uuid_root : ERR_PTR(-ENOENT);
1573 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1574 return btrfs_grab_root(fs_info->free_space_root) ?
1575 fs_info->free_space_root : ERR_PTR(-ENOENT);
1576 return NULL;
1577}
1578
1579int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1580 struct btrfs_root *root)
1581{
1582 int ret;
1583
1584 ret = radix_tree_preload(GFP_NOFS);
1585 if (ret)
1586 return ret;
1587
1588 spin_lock(&fs_info->fs_roots_radix_lock);
1589 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1590 (unsigned long)root->root_key.objectid,
1591 root);
1592 if (ret == 0) {
1593 btrfs_grab_root(root);
1594 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1595 }
1596 spin_unlock(&fs_info->fs_roots_radix_lock);
1597 radix_tree_preload_end();
1598
1599 return ret;
1600}
1601
1602void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1603{
1604#ifdef CONFIG_BTRFS_DEBUG
1605 struct btrfs_root *root;
1606
1607 while (!list_empty(&fs_info->allocated_roots)) {
1608 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1609
1610 root = list_first_entry(&fs_info->allocated_roots,
1611 struct btrfs_root, leak_list);
1612 btrfs_err(fs_info, "leaked root %s refcount %d",
1613 btrfs_root_name(&root->root_key, buf),
1614 refcount_read(&root->refs));
1615 while (refcount_read(&root->refs) > 1)
1616 btrfs_put_root(root);
1617 btrfs_put_root(root);
1618 }
1619#endif
1620}
1621
1622void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1623{
1624 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1625 percpu_counter_destroy(&fs_info->delalloc_bytes);
1626 percpu_counter_destroy(&fs_info->ordered_bytes);
1627 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1628 btrfs_free_csum_hash(fs_info);
1629 btrfs_free_stripe_hash_table(fs_info);
1630 btrfs_free_ref_cache(fs_info);
1631 kfree(fs_info->balance_ctl);
1632 kfree(fs_info->delayed_root);
1633 btrfs_put_root(fs_info->extent_root);
1634 btrfs_put_root(fs_info->tree_root);
1635 btrfs_put_root(fs_info->chunk_root);
1636 btrfs_put_root(fs_info->dev_root);
1637 btrfs_put_root(fs_info->csum_root);
1638 btrfs_put_root(fs_info->quota_root);
1639 btrfs_put_root(fs_info->uuid_root);
1640 btrfs_put_root(fs_info->free_space_root);
1641 btrfs_put_root(fs_info->fs_root);
1642 btrfs_put_root(fs_info->data_reloc_root);
1643 btrfs_check_leaked_roots(fs_info);
1644 btrfs_extent_buffer_leak_debug_check(fs_info);
1645 kfree(fs_info->super_copy);
1646 kfree(fs_info->super_for_commit);
1647 kvfree(fs_info);
1648}
1649
1650
1651/*
1652 * Get an in-memory reference of a root structure.
1653 *
1654 * For essential trees like root/extent tree, we grab it from fs_info directly.
1655 * For subvolume trees, we check the cached filesystem roots first. If not
1656 * found, then read it from disk and add it to cached fs roots.
1657 *
1658 * Caller should release the root by calling btrfs_put_root() after the usage.
1659 *
1660 * NOTE: Reloc and log trees can't be read by this function as they share the
1661 * same root objectid.
1662 *
1663 * @objectid: root id
1664 * @anon_dev: preallocated anonymous block device number for new roots,
1665 * pass 0 for new allocation.
1666 * @check_ref: whether to check root item references, If true, return -ENOENT
1667 * for orphan roots
1668 */
1669static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1670 u64 objectid, dev_t anon_dev,
1671 bool check_ref)
1672{
1673 struct btrfs_root *root;
1674 struct btrfs_path *path;
1675 struct btrfs_key key;
1676 int ret;
1677
1678 root = btrfs_get_global_root(fs_info, objectid);
1679 if (root)
1680 return root;
1681again:
1682 root = btrfs_lookup_fs_root(fs_info, objectid);
1683 if (root) {
1684 /* Shouldn't get preallocated anon_dev for cached roots */
1685 ASSERT(!anon_dev);
1686 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1687 btrfs_put_root(root);
1688 return ERR_PTR(-ENOENT);
1689 }
1690 return root;
1691 }
1692
1693 key.objectid = objectid;
1694 key.type = BTRFS_ROOT_ITEM_KEY;
1695 key.offset = (u64)-1;
1696 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1697 if (IS_ERR(root))
1698 return root;
1699
1700 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1701 ret = -ENOENT;
1702 goto fail;
1703 }
1704
1705 ret = btrfs_init_fs_root(root, anon_dev);
1706 if (ret)
1707 goto fail;
1708
1709 path = btrfs_alloc_path();
1710 if (!path) {
1711 ret = -ENOMEM;
1712 goto fail;
1713 }
1714 key.objectid = BTRFS_ORPHAN_OBJECTID;
1715 key.type = BTRFS_ORPHAN_ITEM_KEY;
1716 key.offset = objectid;
1717
1718 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1719 btrfs_free_path(path);
1720 if (ret < 0)
1721 goto fail;
1722 if (ret == 0)
1723 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1724
1725 ret = btrfs_insert_fs_root(fs_info, root);
1726 if (ret) {
1727 btrfs_put_root(root);
1728 if (ret == -EEXIST)
1729 goto again;
1730 goto fail;
1731 }
1732 return root;
1733fail:
1734 btrfs_put_root(root);
1735 return ERR_PTR(ret);
1736}
1737
1738/*
1739 * Get in-memory reference of a root structure
1740 *
1741 * @objectid: tree objectid
1742 * @check_ref: if set, verify that the tree exists and the item has at least
1743 * one reference
1744 */
1745struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1746 u64 objectid, bool check_ref)
1747{
1748 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1749}
1750
1751/*
1752 * Get in-memory reference of a root structure, created as new, optionally pass
1753 * the anonymous block device id
1754 *
1755 * @objectid: tree objectid
1756 * @anon_dev: if zero, allocate a new anonymous block device or use the
1757 * parameter value
1758 */
1759struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1760 u64 objectid, dev_t anon_dev)
1761{
1762 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1763}
1764
1765/*
1766 * btrfs_get_fs_root_commit_root - return a root for the given objectid
1767 * @fs_info: the fs_info
1768 * @objectid: the objectid we need to lookup
1769 *
1770 * This is exclusively used for backref walking, and exists specifically because
1771 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1772 * creation time, which means we may have to read the tree_root in order to look
1773 * up a fs root that is not in memory. If the root is not in memory we will
1774 * read the tree root commit root and look up the fs root from there. This is a
1775 * temporary root, it will not be inserted into the radix tree as it doesn't
1776 * have the most uptodate information, it'll simply be discarded once the
1777 * backref code is finished using the root.
1778 */
1779struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1780 struct btrfs_path *path,
1781 u64 objectid)
1782{
1783 struct btrfs_root *root;
1784 struct btrfs_key key;
1785
1786 ASSERT(path->search_commit_root && path->skip_locking);
1787
1788 /*
1789 * This can return -ENOENT if we ask for a root that doesn't exist, but
1790 * since this is called via the backref walking code we won't be looking
1791 * up a root that doesn't exist, unless there's corruption. So if root
1792 * != NULL just return it.
1793 */
1794 root = btrfs_get_global_root(fs_info, objectid);
1795 if (root)
1796 return root;
1797
1798 root = btrfs_lookup_fs_root(fs_info, objectid);
1799 if (root)
1800 return root;
1801
1802 key.objectid = objectid;
1803 key.type = BTRFS_ROOT_ITEM_KEY;
1804 key.offset = (u64)-1;
1805 root = read_tree_root_path(fs_info->tree_root, path, &key);
1806 btrfs_release_path(path);
1807
1808 return root;
1809}
1810
1811/*
1812 * called by the kthread helper functions to finally call the bio end_io
1813 * functions. This is where read checksum verification actually happens
1814 */
1815static void end_workqueue_fn(struct btrfs_work *work)
1816{
1817 struct bio *bio;
1818 struct btrfs_end_io_wq *end_io_wq;
1819
1820 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1821 bio = end_io_wq->bio;
1822
1823 bio->bi_status = end_io_wq->status;
1824 bio->bi_private = end_io_wq->private;
1825 bio->bi_end_io = end_io_wq->end_io;
1826 bio_endio(bio);
1827 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1828}
1829
1830static int cleaner_kthread(void *arg)
1831{
1832 struct btrfs_root *root = arg;
1833 struct btrfs_fs_info *fs_info = root->fs_info;
1834 int again;
1835
1836 while (1) {
1837 again = 0;
1838
1839 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1840
1841 /* Make the cleaner go to sleep early. */
1842 if (btrfs_need_cleaner_sleep(fs_info))
1843 goto sleep;
1844
1845 /*
1846 * Do not do anything if we might cause open_ctree() to block
1847 * before we have finished mounting the filesystem.
1848 */
1849 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1850 goto sleep;
1851
1852 if (!mutex_trylock(&fs_info->cleaner_mutex))
1853 goto sleep;
1854
1855 /*
1856 * Avoid the problem that we change the status of the fs
1857 * during the above check and trylock.
1858 */
1859 if (btrfs_need_cleaner_sleep(fs_info)) {
1860 mutex_unlock(&fs_info->cleaner_mutex);
1861 goto sleep;
1862 }
1863
1864 btrfs_run_delayed_iputs(fs_info);
1865
1866 again = btrfs_clean_one_deleted_snapshot(root);
1867 mutex_unlock(&fs_info->cleaner_mutex);
1868
1869 /*
1870 * The defragger has dealt with the R/O remount and umount,
1871 * needn't do anything special here.
1872 */
1873 btrfs_run_defrag_inodes(fs_info);
1874
1875 /*
1876 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1877 * with relocation (btrfs_relocate_chunk) and relocation
1878 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1879 * after acquiring fs_info->reclaim_bgs_lock. So we
1880 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1881 * unused block groups.
1882 */
1883 btrfs_delete_unused_bgs(fs_info);
1884
1885 /*
1886 * Reclaim block groups in the reclaim_bgs list after we deleted
1887 * all unused block_groups. This possibly gives us some more free
1888 * space.
1889 */
1890 btrfs_reclaim_bgs(fs_info);
1891sleep:
1892 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1893 if (kthread_should_park())
1894 kthread_parkme();
1895 if (kthread_should_stop())
1896 return 0;
1897 if (!again) {
1898 set_current_state(TASK_INTERRUPTIBLE);
1899 schedule();
1900 __set_current_state(TASK_RUNNING);
1901 }
1902 }
1903}
1904
1905static int transaction_kthread(void *arg)
1906{
1907 struct btrfs_root *root = arg;
1908 struct btrfs_fs_info *fs_info = root->fs_info;
1909 struct btrfs_trans_handle *trans;
1910 struct btrfs_transaction *cur;
1911 u64 transid;
1912 time64_t delta;
1913 unsigned long delay;
1914 bool cannot_commit;
1915
1916 do {
1917 cannot_commit = false;
1918 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1919 mutex_lock(&fs_info->transaction_kthread_mutex);
1920
1921 spin_lock(&fs_info->trans_lock);
1922 cur = fs_info->running_transaction;
1923 if (!cur) {
1924 spin_unlock(&fs_info->trans_lock);
1925 goto sleep;
1926 }
1927
1928 delta = ktime_get_seconds() - cur->start_time;
1929 if (cur->state < TRANS_STATE_COMMIT_START &&
1930 delta < fs_info->commit_interval) {
1931 spin_unlock(&fs_info->trans_lock);
1932 delay -= msecs_to_jiffies((delta - 1) * 1000);
1933 delay = min(delay,
1934 msecs_to_jiffies(fs_info->commit_interval * 1000));
1935 goto sleep;
1936 }
1937 transid = cur->transid;
1938 spin_unlock(&fs_info->trans_lock);
1939
1940 /* If the file system is aborted, this will always fail. */
1941 trans = btrfs_attach_transaction(root);
1942 if (IS_ERR(trans)) {
1943 if (PTR_ERR(trans) != -ENOENT)
1944 cannot_commit = true;
1945 goto sleep;
1946 }
1947 if (transid == trans->transid) {
1948 btrfs_commit_transaction(trans);
1949 } else {
1950 btrfs_end_transaction(trans);
1951 }
1952sleep:
1953 wake_up_process(fs_info->cleaner_kthread);
1954 mutex_unlock(&fs_info->transaction_kthread_mutex);
1955
1956 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1957 &fs_info->fs_state)))
1958 btrfs_cleanup_transaction(fs_info);
1959 if (!kthread_should_stop() &&
1960 (!btrfs_transaction_blocked(fs_info) ||
1961 cannot_commit))
1962 schedule_timeout_interruptible(delay);
1963 } while (!kthread_should_stop());
1964 return 0;
1965}
1966
1967/*
1968 * This will find the highest generation in the array of root backups. The
1969 * index of the highest array is returned, or -EINVAL if we can't find
1970 * anything.
1971 *
1972 * We check to make sure the array is valid by comparing the
1973 * generation of the latest root in the array with the generation
1974 * in the super block. If they don't match we pitch it.
1975 */
1976static int find_newest_super_backup(struct btrfs_fs_info *info)
1977{
1978 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1979 u64 cur;
1980 struct btrfs_root_backup *root_backup;
1981 int i;
1982
1983 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1984 root_backup = info->super_copy->super_roots + i;
1985 cur = btrfs_backup_tree_root_gen(root_backup);
1986 if (cur == newest_gen)
1987 return i;
1988 }
1989
1990 return -EINVAL;
1991}
1992
1993/*
1994 * copy all the root pointers into the super backup array.
1995 * this will bump the backup pointer by one when it is
1996 * done
1997 */
1998static void backup_super_roots(struct btrfs_fs_info *info)
1999{
2000 const int next_backup = info->backup_root_index;
2001 struct btrfs_root_backup *root_backup;
2002
2003 root_backup = info->super_for_commit->super_roots + next_backup;
2004
2005 /*
2006 * make sure all of our padding and empty slots get zero filled
2007 * regardless of which ones we use today
2008 */
2009 memset(root_backup, 0, sizeof(*root_backup));
2010
2011 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2012
2013 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2014 btrfs_set_backup_tree_root_gen(root_backup,
2015 btrfs_header_generation(info->tree_root->node));
2016
2017 btrfs_set_backup_tree_root_level(root_backup,
2018 btrfs_header_level(info->tree_root->node));
2019
2020 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2021 btrfs_set_backup_chunk_root_gen(root_backup,
2022 btrfs_header_generation(info->chunk_root->node));
2023 btrfs_set_backup_chunk_root_level(root_backup,
2024 btrfs_header_level(info->chunk_root->node));
2025
2026 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2027 btrfs_set_backup_extent_root_gen(root_backup,
2028 btrfs_header_generation(info->extent_root->node));
2029 btrfs_set_backup_extent_root_level(root_backup,
2030 btrfs_header_level(info->extent_root->node));
2031
2032 /*
2033 * we might commit during log recovery, which happens before we set
2034 * the fs_root. Make sure it is valid before we fill it in.
2035 */
2036 if (info->fs_root && info->fs_root->node) {
2037 btrfs_set_backup_fs_root(root_backup,
2038 info->fs_root->node->start);
2039 btrfs_set_backup_fs_root_gen(root_backup,
2040 btrfs_header_generation(info->fs_root->node));
2041 btrfs_set_backup_fs_root_level(root_backup,
2042 btrfs_header_level(info->fs_root->node));
2043 }
2044
2045 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2046 btrfs_set_backup_dev_root_gen(root_backup,
2047 btrfs_header_generation(info->dev_root->node));
2048 btrfs_set_backup_dev_root_level(root_backup,
2049 btrfs_header_level(info->dev_root->node));
2050
2051 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2052 btrfs_set_backup_csum_root_gen(root_backup,
2053 btrfs_header_generation(info->csum_root->node));
2054 btrfs_set_backup_csum_root_level(root_backup,
2055 btrfs_header_level(info->csum_root->node));
2056
2057 btrfs_set_backup_total_bytes(root_backup,
2058 btrfs_super_total_bytes(info->super_copy));
2059 btrfs_set_backup_bytes_used(root_backup,
2060 btrfs_super_bytes_used(info->super_copy));
2061 btrfs_set_backup_num_devices(root_backup,
2062 btrfs_super_num_devices(info->super_copy));
2063
2064 /*
2065 * if we don't copy this out to the super_copy, it won't get remembered
2066 * for the next commit
2067 */
2068 memcpy(&info->super_copy->super_roots,
2069 &info->super_for_commit->super_roots,
2070 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2071}
2072
2073/*
2074 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
2075 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
2076 *
2077 * fs_info - filesystem whose backup roots need to be read
2078 * priority - priority of backup root required
2079 *
2080 * Returns backup root index on success and -EINVAL otherwise.
2081 */
2082static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
2083{
2084 int backup_index = find_newest_super_backup(fs_info);
2085 struct btrfs_super_block *super = fs_info->super_copy;
2086 struct btrfs_root_backup *root_backup;
2087
2088 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
2089 if (priority == 0)
2090 return backup_index;
2091
2092 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
2093 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
2094 } else {
2095 return -EINVAL;
2096 }
2097
2098 root_backup = super->super_roots + backup_index;
2099
2100 btrfs_set_super_generation(super,
2101 btrfs_backup_tree_root_gen(root_backup));
2102 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2103 btrfs_set_super_root_level(super,
2104 btrfs_backup_tree_root_level(root_backup));
2105 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2106
2107 /*
2108 * Fixme: the total bytes and num_devices need to match or we should
2109 * need a fsck
2110 */
2111 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2112 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2113
2114 return backup_index;
2115}
2116
2117/* helper to cleanup workers */
2118static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2119{
2120 btrfs_destroy_workqueue(fs_info->fixup_workers);
2121 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2122 btrfs_destroy_workqueue(fs_info->workers);
2123 btrfs_destroy_workqueue(fs_info->endio_workers);
2124 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2125 btrfs_destroy_workqueue(fs_info->rmw_workers);
2126 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2127 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2128 btrfs_destroy_workqueue(fs_info->delayed_workers);
2129 btrfs_destroy_workqueue(fs_info->caching_workers);
2130 btrfs_destroy_workqueue(fs_info->readahead_workers);
2131 btrfs_destroy_workqueue(fs_info->flush_workers);
2132 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2133 if (fs_info->discard_ctl.discard_workers)
2134 destroy_workqueue(fs_info->discard_ctl.discard_workers);
2135 /*
2136 * Now that all other work queues are destroyed, we can safely destroy
2137 * the queues used for metadata I/O, since tasks from those other work
2138 * queues can do metadata I/O operations.
2139 */
2140 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2141 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2142}
2143
2144static void free_root_extent_buffers(struct btrfs_root *root)
2145{
2146 if (root) {
2147 free_extent_buffer(root->node);
2148 free_extent_buffer(root->commit_root);
2149 root->node = NULL;
2150 root->commit_root = NULL;
2151 }
2152}
2153
2154/* helper to cleanup tree roots */
2155static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2156{
2157 free_root_extent_buffers(info->tree_root);
2158
2159 free_root_extent_buffers(info->dev_root);
2160 free_root_extent_buffers(info->extent_root);
2161 free_root_extent_buffers(info->csum_root);
2162 free_root_extent_buffers(info->quota_root);
2163 free_root_extent_buffers(info->uuid_root);
2164 free_root_extent_buffers(info->fs_root);
2165 free_root_extent_buffers(info->data_reloc_root);
2166 if (free_chunk_root)
2167 free_root_extent_buffers(info->chunk_root);
2168 free_root_extent_buffers(info->free_space_root);
2169}
2170
2171void btrfs_put_root(struct btrfs_root *root)
2172{
2173 if (!root)
2174 return;
2175
2176 if (refcount_dec_and_test(&root->refs)) {
2177 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2178 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2179 if (root->anon_dev)
2180 free_anon_bdev(root->anon_dev);
2181 btrfs_drew_lock_destroy(&root->snapshot_lock);
2182 free_root_extent_buffers(root);
2183#ifdef CONFIG_BTRFS_DEBUG
2184 spin_lock(&root->fs_info->fs_roots_radix_lock);
2185 list_del_init(&root->leak_list);
2186 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2187#endif
2188 kfree(root);
2189 }
2190}
2191
2192void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2193{
2194 int ret;
2195 struct btrfs_root *gang[8];
2196 int i;
2197
2198 while (!list_empty(&fs_info->dead_roots)) {
2199 gang[0] = list_entry(fs_info->dead_roots.next,
2200 struct btrfs_root, root_list);
2201 list_del(&gang[0]->root_list);
2202
2203 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2204 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2205 btrfs_put_root(gang[0]);
2206 }
2207
2208 while (1) {
2209 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2210 (void **)gang, 0,
2211 ARRAY_SIZE(gang));
2212 if (!ret)
2213 break;
2214 for (i = 0; i < ret; i++)
2215 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2216 }
2217}
2218
2219static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2220{
2221 mutex_init(&fs_info->scrub_lock);
2222 atomic_set(&fs_info->scrubs_running, 0);
2223 atomic_set(&fs_info->scrub_pause_req, 0);
2224 atomic_set(&fs_info->scrubs_paused, 0);
2225 atomic_set(&fs_info->scrub_cancel_req, 0);
2226 init_waitqueue_head(&fs_info->scrub_pause_wait);
2227 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2228}
2229
2230static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2231{
2232 spin_lock_init(&fs_info->balance_lock);
2233 mutex_init(&fs_info->balance_mutex);
2234 atomic_set(&fs_info->balance_pause_req, 0);
2235 atomic_set(&fs_info->balance_cancel_req, 0);
2236 fs_info->balance_ctl = NULL;
2237 init_waitqueue_head(&fs_info->balance_wait_q);
2238 atomic_set(&fs_info->reloc_cancel_req, 0);
2239}
2240
2241static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2242{
2243 struct inode *inode = fs_info->btree_inode;
2244
2245 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2246 set_nlink(inode, 1);
2247 /*
2248 * we set the i_size on the btree inode to the max possible int.
2249 * the real end of the address space is determined by all of
2250 * the devices in the system
2251 */
2252 inode->i_size = OFFSET_MAX;
2253 inode->i_mapping->a_ops = &btree_aops;
2254
2255 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2256 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2257 IO_TREE_BTREE_INODE_IO, inode);
2258 BTRFS_I(inode)->io_tree.track_uptodate = false;
2259 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2260
2261 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2262 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2263 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2264 btrfs_insert_inode_hash(inode);
2265}
2266
2267static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2268{
2269 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2270 init_rwsem(&fs_info->dev_replace.rwsem);
2271 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2272}
2273
2274static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2275{
2276 spin_lock_init(&fs_info->qgroup_lock);
2277 mutex_init(&fs_info->qgroup_ioctl_lock);
2278 fs_info->qgroup_tree = RB_ROOT;
2279 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2280 fs_info->qgroup_seq = 1;
2281 fs_info->qgroup_ulist = NULL;
2282 fs_info->qgroup_rescan_running = false;
2283 mutex_init(&fs_info->qgroup_rescan_lock);
2284}
2285
2286static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2287 struct btrfs_fs_devices *fs_devices)
2288{
2289 u32 max_active = fs_info->thread_pool_size;
2290 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2291
2292 fs_info->workers =
2293 btrfs_alloc_workqueue(fs_info, "worker",
2294 flags | WQ_HIGHPRI, max_active, 16);
2295
2296 fs_info->delalloc_workers =
2297 btrfs_alloc_workqueue(fs_info, "delalloc",
2298 flags, max_active, 2);
2299
2300 fs_info->flush_workers =
2301 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2302 flags, max_active, 0);
2303
2304 fs_info->caching_workers =
2305 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2306
2307 fs_info->fixup_workers =
2308 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2309
2310 /*
2311 * endios are largely parallel and should have a very
2312 * low idle thresh
2313 */
2314 fs_info->endio_workers =
2315 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2316 fs_info->endio_meta_workers =
2317 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2318 max_active, 4);
2319 fs_info->endio_meta_write_workers =
2320 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2321 max_active, 2);
2322 fs_info->endio_raid56_workers =
2323 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2324 max_active, 4);
2325 fs_info->rmw_workers =
2326 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2327 fs_info->endio_write_workers =
2328 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2329 max_active, 2);
2330 fs_info->endio_freespace_worker =
2331 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2332 max_active, 0);
2333 fs_info->delayed_workers =
2334 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2335 max_active, 0);
2336 fs_info->readahead_workers =
2337 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2338 max_active, 2);
2339 fs_info->qgroup_rescan_workers =
2340 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2341 fs_info->discard_ctl.discard_workers =
2342 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2343
2344 if (!(fs_info->workers && fs_info->delalloc_workers &&
2345 fs_info->flush_workers &&
2346 fs_info->endio_workers && fs_info->endio_meta_workers &&
2347 fs_info->endio_meta_write_workers &&
2348 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2349 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2350 fs_info->caching_workers && fs_info->readahead_workers &&
2351 fs_info->fixup_workers && fs_info->delayed_workers &&
2352 fs_info->qgroup_rescan_workers &&
2353 fs_info->discard_ctl.discard_workers)) {
2354 return -ENOMEM;
2355 }
2356
2357 return 0;
2358}
2359
2360static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2361{
2362 struct crypto_shash *csum_shash;
2363 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2364
2365 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2366
2367 if (IS_ERR(csum_shash)) {
2368 btrfs_err(fs_info, "error allocating %s hash for checksum",
2369 csum_driver);
2370 return PTR_ERR(csum_shash);
2371 }
2372
2373 fs_info->csum_shash = csum_shash;
2374
2375 return 0;
2376}
2377
2378static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2379 struct btrfs_fs_devices *fs_devices)
2380{
2381 int ret;
2382 struct btrfs_root *log_tree_root;
2383 struct btrfs_super_block *disk_super = fs_info->super_copy;
2384 u64 bytenr = btrfs_super_log_root(disk_super);
2385 int level = btrfs_super_log_root_level(disk_super);
2386
2387 if (fs_devices->rw_devices == 0) {
2388 btrfs_warn(fs_info, "log replay required on RO media");
2389 return -EIO;
2390 }
2391
2392 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2393 GFP_KERNEL);
2394 if (!log_tree_root)
2395 return -ENOMEM;
2396
2397 log_tree_root->node = read_tree_block(fs_info, bytenr,
2398 BTRFS_TREE_LOG_OBJECTID,
2399 fs_info->generation + 1, level,
2400 NULL);
2401 if (IS_ERR(log_tree_root->node)) {
2402 btrfs_warn(fs_info, "failed to read log tree");
2403 ret = PTR_ERR(log_tree_root->node);
2404 log_tree_root->node = NULL;
2405 btrfs_put_root(log_tree_root);
2406 return ret;
2407 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2408 btrfs_err(fs_info, "failed to read log tree");
2409 btrfs_put_root(log_tree_root);
2410 return -EIO;
2411 }
2412 /* returns with log_tree_root freed on success */
2413 ret = btrfs_recover_log_trees(log_tree_root);
2414 if (ret) {
2415 btrfs_handle_fs_error(fs_info, ret,
2416 "Failed to recover log tree");
2417 btrfs_put_root(log_tree_root);
2418 return ret;
2419 }
2420
2421 if (sb_rdonly(fs_info->sb)) {
2422 ret = btrfs_commit_super(fs_info);
2423 if (ret)
2424 return ret;
2425 }
2426
2427 return 0;
2428}
2429
2430static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2431{
2432 struct btrfs_root *tree_root = fs_info->tree_root;
2433 struct btrfs_root *root;
2434 struct btrfs_key location;
2435 int ret;
2436
2437 BUG_ON(!fs_info->tree_root);
2438
2439 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2440 location.type = BTRFS_ROOT_ITEM_KEY;
2441 location.offset = 0;
2442
2443 root = btrfs_read_tree_root(tree_root, &location);
2444 if (IS_ERR(root)) {
2445 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2446 ret = PTR_ERR(root);
2447 goto out;
2448 }
2449 } else {
2450 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2451 fs_info->extent_root = root;
2452 }
2453
2454 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2455 root = btrfs_read_tree_root(tree_root, &location);
2456 if (IS_ERR(root)) {
2457 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2458 ret = PTR_ERR(root);
2459 goto out;
2460 }
2461 } else {
2462 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2463 fs_info->dev_root = root;
2464 }
2465 /* Initialize fs_info for all devices in any case */
2466 btrfs_init_devices_late(fs_info);
2467
2468 /* If IGNOREDATACSUMS is set don't bother reading the csum root. */
2469 if (!btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2470 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2471 root = btrfs_read_tree_root(tree_root, &location);
2472 if (IS_ERR(root)) {
2473 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2474 ret = PTR_ERR(root);
2475 goto out;
2476 }
2477 } else {
2478 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2479 fs_info->csum_root = root;
2480 }
2481 }
2482
2483 /*
2484 * This tree can share blocks with some other fs tree during relocation
2485 * and we need a proper setup by btrfs_get_fs_root
2486 */
2487 root = btrfs_get_fs_root(tree_root->fs_info,
2488 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2489 if (IS_ERR(root)) {
2490 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2491 ret = PTR_ERR(root);
2492 goto out;
2493 }
2494 } else {
2495 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2496 fs_info->data_reloc_root = root;
2497 }
2498
2499 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2500 root = btrfs_read_tree_root(tree_root, &location);
2501 if (!IS_ERR(root)) {
2502 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2503 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2504 fs_info->quota_root = root;
2505 }
2506
2507 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2508 root = btrfs_read_tree_root(tree_root, &location);
2509 if (IS_ERR(root)) {
2510 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2511 ret = PTR_ERR(root);
2512 if (ret != -ENOENT)
2513 goto out;
2514 }
2515 } else {
2516 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2517 fs_info->uuid_root = root;
2518 }
2519
2520 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2521 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2522 root = btrfs_read_tree_root(tree_root, &location);
2523 if (IS_ERR(root)) {
2524 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2525 ret = PTR_ERR(root);
2526 goto out;
2527 }
2528 } else {
2529 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2530 fs_info->free_space_root = root;
2531 }
2532 }
2533
2534 return 0;
2535out:
2536 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2537 location.objectid, ret);
2538 return ret;
2539}
2540
2541/*
2542 * Real super block validation
2543 * NOTE: super csum type and incompat features will not be checked here.
2544 *
2545 * @sb: super block to check
2546 * @mirror_num: the super block number to check its bytenr:
2547 * 0 the primary (1st) sb
2548 * 1, 2 2nd and 3rd backup copy
2549 * -1 skip bytenr check
2550 */
2551static int validate_super(struct btrfs_fs_info *fs_info,
2552 struct btrfs_super_block *sb, int mirror_num)
2553{
2554 u64 nodesize = btrfs_super_nodesize(sb);
2555 u64 sectorsize = btrfs_super_sectorsize(sb);
2556 int ret = 0;
2557
2558 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2559 btrfs_err(fs_info, "no valid FS found");
2560 ret = -EINVAL;
2561 }
2562 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2563 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2564 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2565 ret = -EINVAL;
2566 }
2567 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2568 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2569 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2570 ret = -EINVAL;
2571 }
2572 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2573 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2574 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2575 ret = -EINVAL;
2576 }
2577 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2578 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2579 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2580 ret = -EINVAL;
2581 }
2582
2583 /*
2584 * Check sectorsize and nodesize first, other check will need it.
2585 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2586 */
2587 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2588 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2589 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2590 ret = -EINVAL;
2591 }
2592
2593 /*
2594 * For 4K page size, we only support 4K sector size.
2595 * For 64K page size, we support read-write for 64K sector size, and
2596 * read-only for 4K sector size.
2597 */
2598 if ((PAGE_SIZE == SZ_4K && sectorsize != PAGE_SIZE) ||
2599 (PAGE_SIZE == SZ_64K && (sectorsize != SZ_4K &&
2600 sectorsize != SZ_64K))) {
2601 btrfs_err(fs_info,
2602 "sectorsize %llu not yet supported for page size %lu",
2603 sectorsize, PAGE_SIZE);
2604 ret = -EINVAL;
2605 }
2606
2607 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2608 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2609 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2610 ret = -EINVAL;
2611 }
2612 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2613 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2614 le32_to_cpu(sb->__unused_leafsize), nodesize);
2615 ret = -EINVAL;
2616 }
2617
2618 /* Root alignment check */
2619 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2620 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2621 btrfs_super_root(sb));
2622 ret = -EINVAL;
2623 }
2624 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2625 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2626 btrfs_super_chunk_root(sb));
2627 ret = -EINVAL;
2628 }
2629 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2630 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2631 btrfs_super_log_root(sb));
2632 ret = -EINVAL;
2633 }
2634
2635 if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2636 BTRFS_FSID_SIZE)) {
2637 btrfs_err(fs_info,
2638 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2639 fs_info->super_copy->fsid, fs_info->fs_devices->fsid);
2640 ret = -EINVAL;
2641 }
2642
2643 if (btrfs_fs_incompat(fs_info, METADATA_UUID) &&
2644 memcmp(fs_info->fs_devices->metadata_uuid,
2645 fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) {
2646 btrfs_err(fs_info,
2647"superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2648 fs_info->super_copy->metadata_uuid,
2649 fs_info->fs_devices->metadata_uuid);
2650 ret = -EINVAL;
2651 }
2652
2653 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2654 BTRFS_FSID_SIZE) != 0) {
2655 btrfs_err(fs_info,
2656 "dev_item UUID does not match metadata fsid: %pU != %pU",
2657 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2658 ret = -EINVAL;
2659 }
2660
2661 /*
2662 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2663 * done later
2664 */
2665 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2666 btrfs_err(fs_info, "bytes_used is too small %llu",
2667 btrfs_super_bytes_used(sb));
2668 ret = -EINVAL;
2669 }
2670 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2671 btrfs_err(fs_info, "invalid stripesize %u",
2672 btrfs_super_stripesize(sb));
2673 ret = -EINVAL;
2674 }
2675 if (btrfs_super_num_devices(sb) > (1UL << 31))
2676 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2677 btrfs_super_num_devices(sb));
2678 if (btrfs_super_num_devices(sb) == 0) {
2679 btrfs_err(fs_info, "number of devices is 0");
2680 ret = -EINVAL;
2681 }
2682
2683 if (mirror_num >= 0 &&
2684 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2685 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2686 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2687 ret = -EINVAL;
2688 }
2689
2690 /*
2691 * Obvious sys_chunk_array corruptions, it must hold at least one key
2692 * and one chunk
2693 */
2694 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2695 btrfs_err(fs_info, "system chunk array too big %u > %u",
2696 btrfs_super_sys_array_size(sb),
2697 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2698 ret = -EINVAL;
2699 }
2700 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2701 + sizeof(struct btrfs_chunk)) {
2702 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2703 btrfs_super_sys_array_size(sb),
2704 sizeof(struct btrfs_disk_key)
2705 + sizeof(struct btrfs_chunk));
2706 ret = -EINVAL;
2707 }
2708
2709 /*
2710 * The generation is a global counter, we'll trust it more than the others
2711 * but it's still possible that it's the one that's wrong.
2712 */
2713 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2714 btrfs_warn(fs_info,
2715 "suspicious: generation < chunk_root_generation: %llu < %llu",
2716 btrfs_super_generation(sb),
2717 btrfs_super_chunk_root_generation(sb));
2718 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2719 && btrfs_super_cache_generation(sb) != (u64)-1)
2720 btrfs_warn(fs_info,
2721 "suspicious: generation < cache_generation: %llu < %llu",
2722 btrfs_super_generation(sb),
2723 btrfs_super_cache_generation(sb));
2724
2725 return ret;
2726}
2727
2728/*
2729 * Validation of super block at mount time.
2730 * Some checks already done early at mount time, like csum type and incompat
2731 * flags will be skipped.
2732 */
2733static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2734{
2735 return validate_super(fs_info, fs_info->super_copy, 0);
2736}
2737
2738/*
2739 * Validation of super block at write time.
2740 * Some checks like bytenr check will be skipped as their values will be
2741 * overwritten soon.
2742 * Extra checks like csum type and incompat flags will be done here.
2743 */
2744static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2745 struct btrfs_super_block *sb)
2746{
2747 int ret;
2748
2749 ret = validate_super(fs_info, sb, -1);
2750 if (ret < 0)
2751 goto out;
2752 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2753 ret = -EUCLEAN;
2754 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2755 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2756 goto out;
2757 }
2758 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2759 ret = -EUCLEAN;
2760 btrfs_err(fs_info,
2761 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2762 btrfs_super_incompat_flags(sb),
2763 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2764 goto out;
2765 }
2766out:
2767 if (ret < 0)
2768 btrfs_err(fs_info,
2769 "super block corruption detected before writing it to disk");
2770 return ret;
2771}
2772
2773static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2774{
2775 int backup_index = find_newest_super_backup(fs_info);
2776 struct btrfs_super_block *sb = fs_info->super_copy;
2777 struct btrfs_root *tree_root = fs_info->tree_root;
2778 bool handle_error = false;
2779 int ret = 0;
2780 int i;
2781
2782 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2783 u64 generation;
2784 int level;
2785
2786 if (handle_error) {
2787 if (!IS_ERR(tree_root->node))
2788 free_extent_buffer(tree_root->node);
2789 tree_root->node = NULL;
2790
2791 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2792 break;
2793
2794 free_root_pointers(fs_info, 0);
2795
2796 /*
2797 * Don't use the log in recovery mode, it won't be
2798 * valid
2799 */
2800 btrfs_set_super_log_root(sb, 0);
2801
2802 /* We can't trust the free space cache either */
2803 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2804
2805 ret = read_backup_root(fs_info, i);
2806 backup_index = ret;
2807 if (ret < 0)
2808 return ret;
2809 }
2810 generation = btrfs_super_generation(sb);
2811 level = btrfs_super_root_level(sb);
2812 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2813 BTRFS_ROOT_TREE_OBJECTID,
2814 generation, level, NULL);
2815 if (IS_ERR(tree_root->node)) {
2816 handle_error = true;
2817 ret = PTR_ERR(tree_root->node);
2818 tree_root->node = NULL;
2819 btrfs_warn(fs_info, "couldn't read tree root");
2820 continue;
2821
2822 } else if (!extent_buffer_uptodate(tree_root->node)) {
2823 handle_error = true;
2824 ret = -EIO;
2825 btrfs_warn(fs_info, "error while reading tree root");
2826 continue;
2827 }
2828
2829 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2830 tree_root->commit_root = btrfs_root_node(tree_root);
2831 btrfs_set_root_refs(&tree_root->root_item, 1);
2832
2833 /*
2834 * No need to hold btrfs_root::objectid_mutex since the fs
2835 * hasn't been fully initialised and we are the only user
2836 */
2837 ret = btrfs_init_root_free_objectid(tree_root);
2838 if (ret < 0) {
2839 handle_error = true;
2840 continue;
2841 }
2842
2843 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2844
2845 ret = btrfs_read_roots(fs_info);
2846 if (ret < 0) {
2847 handle_error = true;
2848 continue;
2849 }
2850
2851 /* All successful */
2852 fs_info->generation = generation;
2853 fs_info->last_trans_committed = generation;
2854
2855 /* Always begin writing backup roots after the one being used */
2856 if (backup_index < 0) {
2857 fs_info->backup_root_index = 0;
2858 } else {
2859 fs_info->backup_root_index = backup_index + 1;
2860 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2861 }
2862 break;
2863 }
2864
2865 return ret;
2866}
2867
2868void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2869{
2870 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2871 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2872 INIT_LIST_HEAD(&fs_info->trans_list);
2873 INIT_LIST_HEAD(&fs_info->dead_roots);
2874 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2875 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2876 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2877 spin_lock_init(&fs_info->delalloc_root_lock);
2878 spin_lock_init(&fs_info->trans_lock);
2879 spin_lock_init(&fs_info->fs_roots_radix_lock);
2880 spin_lock_init(&fs_info->delayed_iput_lock);
2881 spin_lock_init(&fs_info->defrag_inodes_lock);
2882 spin_lock_init(&fs_info->super_lock);
2883 spin_lock_init(&fs_info->buffer_lock);
2884 spin_lock_init(&fs_info->unused_bgs_lock);
2885 spin_lock_init(&fs_info->treelog_bg_lock);
2886 rwlock_init(&fs_info->tree_mod_log_lock);
2887 mutex_init(&fs_info->unused_bg_unpin_mutex);
2888 mutex_init(&fs_info->reclaim_bgs_lock);
2889 mutex_init(&fs_info->reloc_mutex);
2890 mutex_init(&fs_info->delalloc_root_mutex);
2891 mutex_init(&fs_info->zoned_meta_io_lock);
2892 seqlock_init(&fs_info->profiles_lock);
2893
2894 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2895 INIT_LIST_HEAD(&fs_info->space_info);
2896 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2897 INIT_LIST_HEAD(&fs_info->unused_bgs);
2898 INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2899#ifdef CONFIG_BTRFS_DEBUG
2900 INIT_LIST_HEAD(&fs_info->allocated_roots);
2901 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2902 spin_lock_init(&fs_info->eb_leak_lock);
2903#endif
2904 extent_map_tree_init(&fs_info->mapping_tree);
2905 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2906 BTRFS_BLOCK_RSV_GLOBAL);
2907 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2908 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2909 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2910 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2911 BTRFS_BLOCK_RSV_DELOPS);
2912 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2913 BTRFS_BLOCK_RSV_DELREFS);
2914
2915 atomic_set(&fs_info->async_delalloc_pages, 0);
2916 atomic_set(&fs_info->defrag_running, 0);
2917 atomic_set(&fs_info->reada_works_cnt, 0);
2918 atomic_set(&fs_info->nr_delayed_iputs, 0);
2919 atomic64_set(&fs_info->tree_mod_seq, 0);
2920 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2921 fs_info->metadata_ratio = 0;
2922 fs_info->defrag_inodes = RB_ROOT;
2923 atomic64_set(&fs_info->free_chunk_space, 0);
2924 fs_info->tree_mod_log = RB_ROOT;
2925 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2926 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2927 /* readahead state */
2928 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2929 spin_lock_init(&fs_info->reada_lock);
2930 btrfs_init_ref_verify(fs_info);
2931
2932 fs_info->thread_pool_size = min_t(unsigned long,
2933 num_online_cpus() + 2, 8);
2934
2935 INIT_LIST_HEAD(&fs_info->ordered_roots);
2936 spin_lock_init(&fs_info->ordered_root_lock);
2937
2938 btrfs_init_scrub(fs_info);
2939#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2940 fs_info->check_integrity_print_mask = 0;
2941#endif
2942 btrfs_init_balance(fs_info);
2943 btrfs_init_async_reclaim_work(fs_info);
2944
2945 spin_lock_init(&fs_info->block_group_cache_lock);
2946 fs_info->block_group_cache_tree = RB_ROOT;
2947 fs_info->first_logical_byte = (u64)-1;
2948
2949 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2950 IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2951 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2952
2953 mutex_init(&fs_info->ordered_operations_mutex);
2954 mutex_init(&fs_info->tree_log_mutex);
2955 mutex_init(&fs_info->chunk_mutex);
2956 mutex_init(&fs_info->transaction_kthread_mutex);
2957 mutex_init(&fs_info->cleaner_mutex);
2958 mutex_init(&fs_info->ro_block_group_mutex);
2959 init_rwsem(&fs_info->commit_root_sem);
2960 init_rwsem(&fs_info->cleanup_work_sem);
2961 init_rwsem(&fs_info->subvol_sem);
2962 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2963
2964 btrfs_init_dev_replace_locks(fs_info);
2965 btrfs_init_qgroup(fs_info);
2966 btrfs_discard_init(fs_info);
2967
2968 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2969 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2970
2971 init_waitqueue_head(&fs_info->transaction_throttle);
2972 init_waitqueue_head(&fs_info->transaction_wait);
2973 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2974 init_waitqueue_head(&fs_info->async_submit_wait);
2975 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2976
2977 /* Usable values until the real ones are cached from the superblock */
2978 fs_info->nodesize = 4096;
2979 fs_info->sectorsize = 4096;
2980 fs_info->sectorsize_bits = ilog2(4096);
2981 fs_info->stripesize = 4096;
2982
2983 spin_lock_init(&fs_info->swapfile_pins_lock);
2984 fs_info->swapfile_pins = RB_ROOT;
2985
2986 spin_lock_init(&fs_info->send_reloc_lock);
2987 fs_info->send_in_progress = 0;
2988
2989 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2990 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2991}
2992
2993static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2994{
2995 int ret;
2996
2997 fs_info->sb = sb;
2998 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2999 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
3000
3001 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
3002 if (ret)
3003 return ret;
3004
3005 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
3006 if (ret)
3007 return ret;
3008
3009 fs_info->dirty_metadata_batch = PAGE_SIZE *
3010 (1 + ilog2(nr_cpu_ids));
3011
3012 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
3013 if (ret)
3014 return ret;
3015
3016 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
3017 GFP_KERNEL);
3018 if (ret)
3019 return ret;
3020
3021 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
3022 GFP_KERNEL);
3023 if (!fs_info->delayed_root)
3024 return -ENOMEM;
3025 btrfs_init_delayed_root(fs_info->delayed_root);
3026
3027 if (sb_rdonly(sb))
3028 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
3029
3030 return btrfs_alloc_stripe_hash_table(fs_info);
3031}
3032
3033static int btrfs_uuid_rescan_kthread(void *data)
3034{
3035 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3036 int ret;
3037
3038 /*
3039 * 1st step is to iterate through the existing UUID tree and
3040 * to delete all entries that contain outdated data.
3041 * 2nd step is to add all missing entries to the UUID tree.
3042 */
3043 ret = btrfs_uuid_tree_iterate(fs_info);
3044 if (ret < 0) {
3045 if (ret != -EINTR)
3046 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
3047 ret);
3048 up(&fs_info->uuid_tree_rescan_sem);
3049 return ret;
3050 }
3051 return btrfs_uuid_scan_kthread(data);
3052}
3053
3054static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3055{
3056 struct task_struct *task;
3057
3058 down(&fs_info->uuid_tree_rescan_sem);
3059 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3060 if (IS_ERR(task)) {
3061 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3062 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3063 up(&fs_info->uuid_tree_rescan_sem);
3064 return PTR_ERR(task);
3065 }
3066
3067 return 0;
3068}
3069
3070/*
3071 * Some options only have meaning at mount time and shouldn't persist across
3072 * remounts, or be displayed. Clear these at the end of mount and remount
3073 * code paths.
3074 */
3075void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
3076{
3077 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3078 btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
3079}
3080
3081/*
3082 * Mounting logic specific to read-write file systems. Shared by open_ctree
3083 * and btrfs_remount when remounting from read-only to read-write.
3084 */
3085int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3086{
3087 int ret;
3088 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3089 bool clear_free_space_tree = false;
3090
3091 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3092 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3093 clear_free_space_tree = true;
3094 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3095 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3096 btrfs_warn(fs_info, "free space tree is invalid");
3097 clear_free_space_tree = true;
3098 }
3099
3100 if (clear_free_space_tree) {
3101 btrfs_info(fs_info, "clearing free space tree");
3102 ret = btrfs_clear_free_space_tree(fs_info);
3103 if (ret) {
3104 btrfs_warn(fs_info,
3105 "failed to clear free space tree: %d", ret);
3106 goto out;
3107 }
3108 }
3109
3110 /*
3111 * btrfs_find_orphan_roots() is responsible for finding all the dead
3112 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3113 * them into the fs_info->fs_roots_radix tree. This must be done before
3114 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3115 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3116 * item before the root's tree is deleted - this means that if we unmount
3117 * or crash before the deletion completes, on the next mount we will not
3118 * delete what remains of the tree because the orphan item does not
3119 * exists anymore, which is what tells us we have a pending deletion.
3120 */
3121 ret = btrfs_find_orphan_roots(fs_info);
3122 if (ret)
3123 goto out;
3124
3125 ret = btrfs_cleanup_fs_roots(fs_info);
3126 if (ret)
3127 goto out;
3128
3129 down_read(&fs_info->cleanup_work_sem);
3130 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3131 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3132 up_read(&fs_info->cleanup_work_sem);
3133 goto out;
3134 }
3135 up_read(&fs_info->cleanup_work_sem);
3136
3137 mutex_lock(&fs_info->cleaner_mutex);
3138 ret = btrfs_recover_relocation(fs_info->tree_root);
3139 mutex_unlock(&fs_info->cleaner_mutex);
3140 if (ret < 0) {
3141 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3142 goto out;
3143 }
3144
3145 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3146 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3147 btrfs_info(fs_info, "creating free space tree");
3148 ret = btrfs_create_free_space_tree(fs_info);
3149 if (ret) {
3150 btrfs_warn(fs_info,
3151 "failed to create free space tree: %d", ret);
3152 goto out;
3153 }
3154 }
3155
3156 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3157 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3158 if (ret)
3159 goto out;
3160 }
3161
3162 ret = btrfs_resume_balance_async(fs_info);
3163 if (ret)
3164 goto out;
3165
3166 ret = btrfs_resume_dev_replace_async(fs_info);
3167 if (ret) {
3168 btrfs_warn(fs_info, "failed to resume dev_replace");
3169 goto out;
3170 }
3171
3172 btrfs_qgroup_rescan_resume(fs_info);
3173
3174 if (!fs_info->uuid_root) {
3175 btrfs_info(fs_info, "creating UUID tree");
3176 ret = btrfs_create_uuid_tree(fs_info);
3177 if (ret) {
3178 btrfs_warn(fs_info,
3179 "failed to create the UUID tree %d", ret);
3180 goto out;
3181 }
3182 }
3183
3184out:
3185 return ret;
3186}
3187
3188int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3189 char *options)
3190{
3191 u32 sectorsize;
3192 u32 nodesize;
3193 u32 stripesize;
3194 u64 generation;
3195 u64 features;
3196 u16 csum_type;
3197 struct btrfs_super_block *disk_super;
3198 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3199 struct btrfs_root *tree_root;
3200 struct btrfs_root *chunk_root;
3201 int ret;
3202 int err = -EINVAL;
3203 int level;
3204
3205 ret = init_mount_fs_info(fs_info, sb);
3206 if (ret) {
3207 err = ret;
3208 goto fail;
3209 }
3210
3211 /* These need to be init'ed before we start creating inodes and such. */
3212 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3213 GFP_KERNEL);
3214 fs_info->tree_root = tree_root;
3215 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3216 GFP_KERNEL);
3217 fs_info->chunk_root = chunk_root;
3218 if (!tree_root || !chunk_root) {
3219 err = -ENOMEM;
3220 goto fail;
3221 }
3222
3223 fs_info->btree_inode = new_inode(sb);
3224 if (!fs_info->btree_inode) {
3225 err = -ENOMEM;
3226 goto fail;
3227 }
3228 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
3229 btrfs_init_btree_inode(fs_info);
3230
3231 invalidate_bdev(fs_devices->latest_bdev);
3232
3233 /*
3234 * Read super block and check the signature bytes only
3235 */
3236 disk_super = btrfs_read_dev_super(fs_devices->latest_bdev);
3237 if (IS_ERR(disk_super)) {
3238 err = PTR_ERR(disk_super);
3239 goto fail_alloc;
3240 }
3241
3242 /*
3243 * Verify the type first, if that or the checksum value are
3244 * corrupted, we'll find out
3245 */
3246 csum_type = btrfs_super_csum_type(disk_super);
3247 if (!btrfs_supported_super_csum(csum_type)) {
3248 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3249 csum_type);
3250 err = -EINVAL;
3251 btrfs_release_disk_super(disk_super);
3252 goto fail_alloc;
3253 }
3254
3255 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3256
3257 ret = btrfs_init_csum_hash(fs_info, csum_type);
3258 if (ret) {
3259 err = ret;
3260 btrfs_release_disk_super(disk_super);
3261 goto fail_alloc;
3262 }
3263
3264 /*
3265 * We want to check superblock checksum, the type is stored inside.
3266 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3267 */
3268 if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
3269 btrfs_err(fs_info, "superblock checksum mismatch");
3270 err = -EINVAL;
3271 btrfs_release_disk_super(disk_super);
3272 goto fail_alloc;
3273 }
3274
3275 /*
3276 * super_copy is zeroed at allocation time and we never touch the
3277 * following bytes up to INFO_SIZE, the checksum is calculated from
3278 * the whole block of INFO_SIZE
3279 */
3280 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3281 btrfs_release_disk_super(disk_super);
3282
3283 disk_super = fs_info->super_copy;
3284
3285
3286 features = btrfs_super_flags(disk_super);
3287 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3288 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3289 btrfs_set_super_flags(disk_super, features);
3290 btrfs_info(fs_info,
3291 "found metadata UUID change in progress flag, clearing");
3292 }
3293
3294 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3295 sizeof(*fs_info->super_for_commit));
3296
3297 ret = btrfs_validate_mount_super(fs_info);
3298 if (ret) {
3299 btrfs_err(fs_info, "superblock contains fatal errors");
3300 err = -EINVAL;
3301 goto fail_alloc;
3302 }
3303
3304 if (!btrfs_super_root(disk_super))
3305 goto fail_alloc;
3306
3307 /* check FS state, whether FS is broken. */
3308 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3309 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3310
3311 /*
3312 * In the long term, we'll store the compression type in the super
3313 * block, and it'll be used for per file compression control.
3314 */
3315 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3316
3317 /*
3318 * Flag our filesystem as having big metadata blocks if they are bigger
3319 * than the page size.
3320 */
3321 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3322 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3323 btrfs_info(fs_info,
3324 "flagging fs with big metadata feature");
3325 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3326 }
3327
3328 /* Set up fs_info before parsing mount options */
3329 nodesize = btrfs_super_nodesize(disk_super);
3330 sectorsize = btrfs_super_sectorsize(disk_super);
3331 stripesize = sectorsize;
3332 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3333 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3334
3335 fs_info->nodesize = nodesize;
3336 fs_info->sectorsize = sectorsize;
3337 fs_info->sectorsize_bits = ilog2(sectorsize);
3338 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3339 fs_info->stripesize = stripesize;
3340
3341 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3342 if (ret) {
3343 err = ret;
3344 goto fail_alloc;
3345 }
3346
3347 features = btrfs_super_incompat_flags(disk_super) &
3348 ~BTRFS_FEATURE_INCOMPAT_SUPP;
3349 if (features) {
3350 btrfs_err(fs_info,
3351 "cannot mount because of unsupported optional features (%llx)",
3352 features);
3353 err = -EINVAL;
3354 goto fail_alloc;
3355 }
3356
3357 features = btrfs_super_incompat_flags(disk_super);
3358 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3359 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3360 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3361 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3362 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3363
3364 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3365 btrfs_info(fs_info, "has skinny extents");
3366
3367 /*
3368 * mixed block groups end up with duplicate but slightly offset
3369 * extent buffers for the same range. It leads to corruptions
3370 */
3371 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3372 (sectorsize != nodesize)) {
3373 btrfs_err(fs_info,
3374"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3375 nodesize, sectorsize);
3376 goto fail_alloc;
3377 }
3378
3379 /*
3380 * Needn't use the lock because there is no other task which will
3381 * update the flag.
3382 */
3383 btrfs_set_super_incompat_flags(disk_super, features);
3384
3385 features = btrfs_super_compat_ro_flags(disk_super) &
3386 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3387 if (!sb_rdonly(sb) && features) {
3388 btrfs_err(fs_info,
3389 "cannot mount read-write because of unsupported optional features (%llx)",
3390 features);
3391 err = -EINVAL;
3392 goto fail_alloc;
3393 }
3394
3395 /* For 4K sector size support, it's only read-only */
3396 if (PAGE_SIZE == SZ_64K && sectorsize == SZ_4K) {
3397 if (!sb_rdonly(sb) || btrfs_super_log_root(disk_super)) {
3398 btrfs_err(fs_info,
3399 "subpage sectorsize %u only supported read-only for page size %lu",
3400 sectorsize, PAGE_SIZE);
3401 err = -EINVAL;
3402 goto fail_alloc;
3403 }
3404 }
3405
3406 ret = btrfs_init_workqueues(fs_info, fs_devices);
3407 if (ret) {
3408 err = ret;
3409 goto fail_sb_buffer;
3410 }
3411
3412 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3413 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3414
3415 sb->s_blocksize = sectorsize;
3416 sb->s_blocksize_bits = blksize_bits(sectorsize);
3417 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3418
3419 mutex_lock(&fs_info->chunk_mutex);
3420 ret = btrfs_read_sys_array(fs_info);
3421 mutex_unlock(&fs_info->chunk_mutex);
3422 if (ret) {
3423 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3424 goto fail_sb_buffer;
3425 }
3426
3427 generation = btrfs_super_chunk_root_generation(disk_super);
3428 level = btrfs_super_chunk_root_level(disk_super);
3429
3430 chunk_root->node = read_tree_block(fs_info,
3431 btrfs_super_chunk_root(disk_super),
3432 BTRFS_CHUNK_TREE_OBJECTID,
3433 generation, level, NULL);
3434 if (IS_ERR(chunk_root->node) ||
3435 !extent_buffer_uptodate(chunk_root->node)) {
3436 btrfs_err(fs_info, "failed to read chunk root");
3437 if (!IS_ERR(chunk_root->node))
3438 free_extent_buffer(chunk_root->node);
3439 chunk_root->node = NULL;
3440 goto fail_tree_roots;
3441 }
3442 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3443 chunk_root->commit_root = btrfs_root_node(chunk_root);
3444
3445 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3446 offsetof(struct btrfs_header, chunk_tree_uuid),
3447 BTRFS_UUID_SIZE);
3448
3449 ret = btrfs_read_chunk_tree(fs_info);
3450 if (ret) {
3451 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3452 goto fail_tree_roots;
3453 }
3454
3455 /*
3456 * At this point we know all the devices that make this filesystem,
3457 * including the seed devices but we don't know yet if the replace
3458 * target is required. So free devices that are not part of this
3459 * filesystem but skip the replace target device which is checked
3460 * below in btrfs_init_dev_replace().
3461 */
3462 btrfs_free_extra_devids(fs_devices);
3463 if (!fs_devices->latest_bdev) {
3464 btrfs_err(fs_info, "failed to read devices");
3465 goto fail_tree_roots;
3466 }
3467
3468 ret = init_tree_roots(fs_info);
3469 if (ret)
3470 goto fail_tree_roots;
3471
3472 /*
3473 * Get zone type information of zoned block devices. This will also
3474 * handle emulation of a zoned filesystem if a regular device has the
3475 * zoned incompat feature flag set.
3476 */
3477 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3478 if (ret) {
3479 btrfs_err(fs_info,
3480 "zoned: failed to read device zone info: %d",
3481 ret);
3482 goto fail_block_groups;
3483 }
3484
3485 /*
3486 * If we have a uuid root and we're not being told to rescan we need to
3487 * check the generation here so we can set the
3488 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3489 * transaction during a balance or the log replay without updating the
3490 * uuid generation, and then if we crash we would rescan the uuid tree,
3491 * even though it was perfectly fine.
3492 */
3493 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3494 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3495 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3496
3497 ret = btrfs_verify_dev_extents(fs_info);
3498 if (ret) {
3499 btrfs_err(fs_info,
3500 "failed to verify dev extents against chunks: %d",
3501 ret);
3502 goto fail_block_groups;
3503 }
3504 ret = btrfs_recover_balance(fs_info);
3505 if (ret) {
3506 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3507 goto fail_block_groups;
3508 }
3509
3510 ret = btrfs_init_dev_stats(fs_info);
3511 if (ret) {
3512 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3513 goto fail_block_groups;
3514 }
3515
3516 ret = btrfs_init_dev_replace(fs_info);
3517 if (ret) {
3518 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3519 goto fail_block_groups;
3520 }
3521
3522 ret = btrfs_check_zoned_mode(fs_info);
3523 if (ret) {
3524 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3525 ret);
3526 goto fail_block_groups;
3527 }
3528
3529 ret = btrfs_sysfs_add_fsid(fs_devices);
3530 if (ret) {
3531 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3532 ret);
3533 goto fail_block_groups;
3534 }
3535
3536 ret = btrfs_sysfs_add_mounted(fs_info);
3537 if (ret) {
3538 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3539 goto fail_fsdev_sysfs;
3540 }
3541
3542 ret = btrfs_init_space_info(fs_info);
3543 if (ret) {
3544 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3545 goto fail_sysfs;
3546 }
3547
3548 ret = btrfs_read_block_groups(fs_info);
3549 if (ret) {
3550 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3551 goto fail_sysfs;
3552 }
3553
3554 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3555 btrfs_warn(fs_info,
3556 "writable mount is not allowed due to too many missing devices");
3557 goto fail_sysfs;
3558 }
3559
3560 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3561 "btrfs-cleaner");
3562 if (IS_ERR(fs_info->cleaner_kthread))
3563 goto fail_sysfs;
3564
3565 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3566 tree_root,
3567 "btrfs-transaction");
3568 if (IS_ERR(fs_info->transaction_kthread))
3569 goto fail_cleaner;
3570
3571 if (!btrfs_test_opt(fs_info, NOSSD) &&
3572 !fs_info->fs_devices->rotating) {
3573 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3574 }
3575
3576 /*
3577 * Mount does not set all options immediately, we can do it now and do
3578 * not have to wait for transaction commit
3579 */
3580 btrfs_apply_pending_changes(fs_info);
3581
3582#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3583 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3584 ret = btrfsic_mount(fs_info, fs_devices,
3585 btrfs_test_opt(fs_info,
3586 CHECK_INTEGRITY_DATA) ? 1 : 0,
3587 fs_info->check_integrity_print_mask);
3588 if (ret)
3589 btrfs_warn(fs_info,
3590 "failed to initialize integrity check module: %d",
3591 ret);
3592 }
3593#endif
3594 ret = btrfs_read_qgroup_config(fs_info);
3595 if (ret)
3596 goto fail_trans_kthread;
3597
3598 if (btrfs_build_ref_tree(fs_info))
3599 btrfs_err(fs_info, "couldn't build ref tree");
3600
3601 /* do not make disk changes in broken FS or nologreplay is given */
3602 if (btrfs_super_log_root(disk_super) != 0 &&
3603 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3604 btrfs_info(fs_info, "start tree-log replay");
3605 ret = btrfs_replay_log(fs_info, fs_devices);
3606 if (ret) {
3607 err = ret;
3608 goto fail_qgroup;
3609 }
3610 }
3611
3612 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3613 if (IS_ERR(fs_info->fs_root)) {
3614 err = PTR_ERR(fs_info->fs_root);
3615 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3616 fs_info->fs_root = NULL;
3617 goto fail_qgroup;
3618 }
3619
3620 if (sb_rdonly(sb))
3621 goto clear_oneshot;
3622
3623 ret = btrfs_start_pre_rw_mount(fs_info);
3624 if (ret) {
3625 close_ctree(fs_info);
3626 return ret;
3627 }
3628 btrfs_discard_resume(fs_info);
3629
3630 if (fs_info->uuid_root &&
3631 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3632 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3633 btrfs_info(fs_info, "checking UUID tree");
3634 ret = btrfs_check_uuid_tree(fs_info);
3635 if (ret) {
3636 btrfs_warn(fs_info,
3637 "failed to check the UUID tree: %d", ret);
3638 close_ctree(fs_info);
3639 return ret;
3640 }
3641 }
3642
3643 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3644
3645clear_oneshot:
3646 btrfs_clear_oneshot_options(fs_info);
3647 return 0;
3648
3649fail_qgroup:
3650 btrfs_free_qgroup_config(fs_info);
3651fail_trans_kthread:
3652 kthread_stop(fs_info->transaction_kthread);
3653 btrfs_cleanup_transaction(fs_info);
3654 btrfs_free_fs_roots(fs_info);
3655fail_cleaner:
3656 kthread_stop(fs_info->cleaner_kthread);
3657
3658 /*
3659 * make sure we're done with the btree inode before we stop our
3660 * kthreads
3661 */
3662 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3663
3664fail_sysfs:
3665 btrfs_sysfs_remove_mounted(fs_info);
3666
3667fail_fsdev_sysfs:
3668 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3669
3670fail_block_groups:
3671 btrfs_put_block_group_cache(fs_info);
3672
3673fail_tree_roots:
3674 if (fs_info->data_reloc_root)
3675 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3676 free_root_pointers(fs_info, true);
3677 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3678
3679fail_sb_buffer:
3680 btrfs_stop_all_workers(fs_info);
3681 btrfs_free_block_groups(fs_info);
3682fail_alloc:
3683 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3684
3685 iput(fs_info->btree_inode);
3686fail:
3687 btrfs_close_devices(fs_info->fs_devices);
3688 return err;
3689}
3690ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3691
3692static void btrfs_end_super_write(struct bio *bio)
3693{
3694 struct btrfs_device *device = bio->bi_private;
3695 struct bio_vec *bvec;
3696 struct bvec_iter_all iter_all;
3697 struct page *page;
3698
3699 bio_for_each_segment_all(bvec, bio, iter_all) {
3700 page = bvec->bv_page;
3701
3702 if (bio->bi_status) {
3703 btrfs_warn_rl_in_rcu(device->fs_info,
3704 "lost page write due to IO error on %s (%d)",
3705 rcu_str_deref(device->name),
3706 blk_status_to_errno(bio->bi_status));
3707 ClearPageUptodate(page);
3708 SetPageError(page);
3709 btrfs_dev_stat_inc_and_print(device,
3710 BTRFS_DEV_STAT_WRITE_ERRS);
3711 } else {
3712 SetPageUptodate(page);
3713 }
3714
3715 put_page(page);
3716 unlock_page(page);
3717 }
3718
3719 bio_put(bio);
3720}
3721
3722struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3723 int copy_num)
3724{
3725 struct btrfs_super_block *super;
3726 struct page *page;
3727 u64 bytenr, bytenr_orig;
3728 struct address_space *mapping = bdev->bd_inode->i_mapping;
3729 int ret;
3730
3731 bytenr_orig = btrfs_sb_offset(copy_num);
3732 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3733 if (ret == -ENOENT)
3734 return ERR_PTR(-EINVAL);
3735 else if (ret)
3736 return ERR_PTR(ret);
3737
3738 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3739 return ERR_PTR(-EINVAL);
3740
3741 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3742 if (IS_ERR(page))
3743 return ERR_CAST(page);
3744
3745 super = page_address(page);
3746 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3747 btrfs_release_disk_super(super);
3748 return ERR_PTR(-ENODATA);
3749 }
3750
3751 if (btrfs_super_bytenr(super) != bytenr_orig) {
3752 btrfs_release_disk_super(super);
3753 return ERR_PTR(-EINVAL);
3754 }
3755
3756 return super;
3757}
3758
3759
3760struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3761{
3762 struct btrfs_super_block *super, *latest = NULL;
3763 int i;
3764 u64 transid = 0;
3765
3766 /* we would like to check all the supers, but that would make
3767 * a btrfs mount succeed after a mkfs from a different FS.
3768 * So, we need to add a special mount option to scan for
3769 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3770 */
3771 for (i = 0; i < 1; i++) {
3772 super = btrfs_read_dev_one_super(bdev, i);
3773 if (IS_ERR(super))
3774 continue;
3775
3776 if (!latest || btrfs_super_generation(super) > transid) {
3777 if (latest)
3778 btrfs_release_disk_super(super);
3779
3780 latest = super;
3781 transid = btrfs_super_generation(super);
3782 }
3783 }
3784
3785 return super;
3786}
3787
3788/*
3789 * Write superblock @sb to the @device. Do not wait for completion, all the
3790 * pages we use for writing are locked.
3791 *
3792 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3793 * the expected device size at commit time. Note that max_mirrors must be
3794 * same for write and wait phases.
3795 *
3796 * Return number of errors when page is not found or submission fails.
3797 */
3798static int write_dev_supers(struct btrfs_device *device,
3799 struct btrfs_super_block *sb, int max_mirrors)
3800{
3801 struct btrfs_fs_info *fs_info = device->fs_info;
3802 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3803 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3804 int i;
3805 int errors = 0;
3806 int ret;
3807 u64 bytenr, bytenr_orig;
3808
3809 if (max_mirrors == 0)
3810 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3811
3812 shash->tfm = fs_info->csum_shash;
3813
3814 for (i = 0; i < max_mirrors; i++) {
3815 struct page *page;
3816 struct bio *bio;
3817 struct btrfs_super_block *disk_super;
3818
3819 bytenr_orig = btrfs_sb_offset(i);
3820 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3821 if (ret == -ENOENT) {
3822 continue;
3823 } else if (ret < 0) {
3824 btrfs_err(device->fs_info,
3825 "couldn't get super block location for mirror %d",
3826 i);
3827 errors++;
3828 continue;
3829 }
3830 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3831 device->commit_total_bytes)
3832 break;
3833
3834 btrfs_set_super_bytenr(sb, bytenr_orig);
3835
3836 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3837 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3838 sb->csum);
3839
3840 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3841 GFP_NOFS);
3842 if (!page) {
3843 btrfs_err(device->fs_info,
3844 "couldn't get super block page for bytenr %llu",
3845 bytenr);
3846 errors++;
3847 continue;
3848 }
3849
3850 /* Bump the refcount for wait_dev_supers() */
3851 get_page(page);
3852
3853 disk_super = page_address(page);
3854 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3855
3856 /*
3857 * Directly use bios here instead of relying on the page cache
3858 * to do I/O, so we don't lose the ability to do integrity
3859 * checking.
3860 */
3861 bio = bio_alloc(GFP_NOFS, 1);
3862 bio_set_dev(bio, device->bdev);
3863 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3864 bio->bi_private = device;
3865 bio->bi_end_io = btrfs_end_super_write;
3866 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3867 offset_in_page(bytenr));
3868
3869 /*
3870 * We FUA only the first super block. The others we allow to
3871 * go down lazy and there's a short window where the on-disk
3872 * copies might still contain the older version.
3873 */
3874 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3875 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3876 bio->bi_opf |= REQ_FUA;
3877
3878 btrfsic_submit_bio(bio);
3879 btrfs_advance_sb_log(device, i);
3880 }
3881 return errors < i ? 0 : -1;
3882}
3883
3884/*
3885 * Wait for write completion of superblocks done by write_dev_supers,
3886 * @max_mirrors same for write and wait phases.
3887 *
3888 * Return number of errors when page is not found or not marked up to
3889 * date.
3890 */
3891static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3892{
3893 int i;
3894 int errors = 0;
3895 bool primary_failed = false;
3896 int ret;
3897 u64 bytenr;
3898
3899 if (max_mirrors == 0)
3900 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3901
3902 for (i = 0; i < max_mirrors; i++) {
3903 struct page *page;
3904
3905 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3906 if (ret == -ENOENT) {
3907 break;
3908 } else if (ret < 0) {
3909 errors++;
3910 if (i == 0)
3911 primary_failed = true;
3912 continue;
3913 }
3914 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3915 device->commit_total_bytes)
3916 break;
3917
3918 page = find_get_page(device->bdev->bd_inode->i_mapping,
3919 bytenr >> PAGE_SHIFT);
3920 if (!page) {
3921 errors++;
3922 if (i == 0)
3923 primary_failed = true;
3924 continue;
3925 }
3926 /* Page is submitted locked and unlocked once the IO completes */
3927 wait_on_page_locked(page);
3928 if (PageError(page)) {
3929 errors++;
3930 if (i == 0)
3931 primary_failed = true;
3932 }
3933
3934 /* Drop our reference */
3935 put_page(page);
3936
3937 /* Drop the reference from the writing run */
3938 put_page(page);
3939 }
3940
3941 /* log error, force error return */
3942 if (primary_failed) {
3943 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3944 device->devid);
3945 return -1;
3946 }
3947
3948 return errors < i ? 0 : -1;
3949}
3950
3951/*
3952 * endio for the write_dev_flush, this will wake anyone waiting
3953 * for the barrier when it is done
3954 */
3955static void btrfs_end_empty_barrier(struct bio *bio)
3956{
3957 complete(bio->bi_private);
3958}
3959
3960/*
3961 * Submit a flush request to the device if it supports it. Error handling is
3962 * done in the waiting counterpart.
3963 */
3964static void write_dev_flush(struct btrfs_device *device)
3965{
3966 struct request_queue *q = bdev_get_queue(device->bdev);
3967 struct bio *bio = device->flush_bio;
3968
3969 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3970 return;
3971
3972 bio_reset(bio);
3973 bio->bi_end_io = btrfs_end_empty_barrier;
3974 bio_set_dev(bio, device->bdev);
3975 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3976 init_completion(&device->flush_wait);
3977 bio->bi_private = &device->flush_wait;
3978
3979 btrfsic_submit_bio(bio);
3980 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3981}
3982
3983/*
3984 * If the flush bio has been submitted by write_dev_flush, wait for it.
3985 */
3986static blk_status_t wait_dev_flush(struct btrfs_device *device)
3987{
3988 struct bio *bio = device->flush_bio;
3989
3990 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3991 return BLK_STS_OK;
3992
3993 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3994 wait_for_completion_io(&device->flush_wait);
3995
3996 return bio->bi_status;
3997}
3998
3999static int check_barrier_error(struct btrfs_fs_info *fs_info)
4000{
4001 if (!btrfs_check_rw_degradable(fs_info, NULL))
4002 return -EIO;
4003 return 0;
4004}
4005
4006/*
4007 * send an empty flush down to each device in parallel,
4008 * then wait for them
4009 */
4010static int barrier_all_devices(struct btrfs_fs_info *info)
4011{
4012 struct list_head *head;
4013 struct btrfs_device *dev;
4014 int errors_wait = 0;
4015 blk_status_t ret;
4016
4017 lockdep_assert_held(&info->fs_devices->device_list_mutex);
4018 /* send down all the barriers */
4019 head = &info->fs_devices->devices;
4020 list_for_each_entry(dev, head, dev_list) {
4021 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4022 continue;
4023 if (!dev->bdev)
4024 continue;
4025 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4026 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4027 continue;
4028
4029 write_dev_flush(dev);
4030 dev->last_flush_error = BLK_STS_OK;
4031 }
4032
4033 /* wait for all the barriers */
4034 list_for_each_entry(dev, head, dev_list) {
4035 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4036 continue;
4037 if (!dev->bdev) {
4038 errors_wait++;
4039 continue;
4040 }
4041 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4042 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4043 continue;
4044
4045 ret = wait_dev_flush(dev);
4046 if (ret) {
4047 dev->last_flush_error = ret;
4048 btrfs_dev_stat_inc_and_print(dev,
4049 BTRFS_DEV_STAT_FLUSH_ERRS);
4050 errors_wait++;
4051 }
4052 }
4053
4054 if (errors_wait) {
4055 /*
4056 * At some point we need the status of all disks
4057 * to arrive at the volume status. So error checking
4058 * is being pushed to a separate loop.
4059 */
4060 return check_barrier_error(info);
4061 }
4062 return 0;
4063}
4064
4065int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4066{
4067 int raid_type;
4068 int min_tolerated = INT_MAX;
4069
4070 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4071 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4072 min_tolerated = min_t(int, min_tolerated,
4073 btrfs_raid_array[BTRFS_RAID_SINGLE].
4074 tolerated_failures);
4075
4076 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4077 if (raid_type == BTRFS_RAID_SINGLE)
4078 continue;
4079 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4080 continue;
4081 min_tolerated = min_t(int, min_tolerated,
4082 btrfs_raid_array[raid_type].
4083 tolerated_failures);
4084 }
4085
4086 if (min_tolerated == INT_MAX) {
4087 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4088 min_tolerated = 0;
4089 }
4090
4091 return min_tolerated;
4092}
4093
4094int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4095{
4096 struct list_head *head;
4097 struct btrfs_device *dev;
4098 struct btrfs_super_block *sb;
4099 struct btrfs_dev_item *dev_item;
4100 int ret;
4101 int do_barriers;
4102 int max_errors;
4103 int total_errors = 0;
4104 u64 flags;
4105
4106 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4107
4108 /*
4109 * max_mirrors == 0 indicates we're from commit_transaction,
4110 * not from fsync where the tree roots in fs_info have not
4111 * been consistent on disk.
4112 */
4113 if (max_mirrors == 0)
4114 backup_super_roots(fs_info);
4115
4116 sb = fs_info->super_for_commit;
4117 dev_item = &sb->dev_item;
4118
4119 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4120 head = &fs_info->fs_devices->devices;
4121 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4122
4123 if (do_barriers) {
4124 ret = barrier_all_devices(fs_info);
4125 if (ret) {
4126 mutex_unlock(
4127 &fs_info->fs_devices->device_list_mutex);
4128 btrfs_handle_fs_error(fs_info, ret,
4129 "errors while submitting device barriers.");
4130 return ret;
4131 }
4132 }
4133
4134 list_for_each_entry(dev, head, dev_list) {
4135 if (!dev->bdev) {
4136 total_errors++;
4137 continue;
4138 }
4139 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4140 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4141 continue;
4142
4143 btrfs_set_stack_device_generation(dev_item, 0);
4144 btrfs_set_stack_device_type(dev_item, dev->type);
4145 btrfs_set_stack_device_id(dev_item, dev->devid);
4146 btrfs_set_stack_device_total_bytes(dev_item,
4147 dev->commit_total_bytes);
4148 btrfs_set_stack_device_bytes_used(dev_item,
4149 dev->commit_bytes_used);
4150 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4151 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4152 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4153 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4154 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4155 BTRFS_FSID_SIZE);
4156
4157 flags = btrfs_super_flags(sb);
4158 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4159
4160 ret = btrfs_validate_write_super(fs_info, sb);
4161 if (ret < 0) {
4162 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4163 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4164 "unexpected superblock corruption detected");
4165 return -EUCLEAN;
4166 }
4167
4168 ret = write_dev_supers(dev, sb, max_mirrors);
4169 if (ret)
4170 total_errors++;
4171 }
4172 if (total_errors > max_errors) {
4173 btrfs_err(fs_info, "%d errors while writing supers",
4174 total_errors);
4175 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4176
4177 /* FUA is masked off if unsupported and can't be the reason */
4178 btrfs_handle_fs_error(fs_info, -EIO,
4179 "%d errors while writing supers",
4180 total_errors);
4181 return -EIO;
4182 }
4183
4184 total_errors = 0;
4185 list_for_each_entry(dev, head, dev_list) {
4186 if (!dev->bdev)
4187 continue;
4188 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4189 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4190 continue;
4191
4192 ret = wait_dev_supers(dev, max_mirrors);
4193 if (ret)
4194 total_errors++;
4195 }
4196 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4197 if (total_errors > max_errors) {
4198 btrfs_handle_fs_error(fs_info, -EIO,
4199 "%d errors while writing supers",
4200 total_errors);
4201 return -EIO;
4202 }
4203 return 0;
4204}
4205
4206/* Drop a fs root from the radix tree and free it. */
4207void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4208 struct btrfs_root *root)
4209{
4210 bool drop_ref = false;
4211
4212 spin_lock(&fs_info->fs_roots_radix_lock);
4213 radix_tree_delete(&fs_info->fs_roots_radix,
4214 (unsigned long)root->root_key.objectid);
4215 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4216 drop_ref = true;
4217 spin_unlock(&fs_info->fs_roots_radix_lock);
4218
4219 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
4220 ASSERT(root->log_root == NULL);
4221 if (root->reloc_root) {
4222 btrfs_put_root(root->reloc_root);
4223 root->reloc_root = NULL;
4224 }
4225 }
4226
4227 if (drop_ref)
4228 btrfs_put_root(root);
4229}
4230
4231int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4232{
4233 u64 root_objectid = 0;
4234 struct btrfs_root *gang[8];
4235 int i = 0;
4236 int err = 0;
4237 unsigned int ret = 0;
4238
4239 while (1) {
4240 spin_lock(&fs_info->fs_roots_radix_lock);
4241 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4242 (void **)gang, root_objectid,
4243 ARRAY_SIZE(gang));
4244 if (!ret) {
4245 spin_unlock(&fs_info->fs_roots_radix_lock);
4246 break;
4247 }
4248 root_objectid = gang[ret - 1]->root_key.objectid + 1;
4249
4250 for (i = 0; i < ret; i++) {
4251 /* Avoid to grab roots in dead_roots */
4252 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4253 gang[i] = NULL;
4254 continue;
4255 }
4256 /* grab all the search result for later use */
4257 gang[i] = btrfs_grab_root(gang[i]);
4258 }
4259 spin_unlock(&fs_info->fs_roots_radix_lock);
4260
4261 for (i = 0; i < ret; i++) {
4262 if (!gang[i])
4263 continue;
4264 root_objectid = gang[i]->root_key.objectid;
4265 err = btrfs_orphan_cleanup(gang[i]);
4266 if (err)
4267 break;
4268 btrfs_put_root(gang[i]);
4269 }
4270 root_objectid++;
4271 }
4272
4273 /* release the uncleaned roots due to error */
4274 for (; i < ret; i++) {
4275 if (gang[i])
4276 btrfs_put_root(gang[i]);
4277 }
4278 return err;
4279}
4280
4281int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4282{
4283 struct btrfs_root *root = fs_info->tree_root;
4284 struct btrfs_trans_handle *trans;
4285
4286 mutex_lock(&fs_info->cleaner_mutex);
4287 btrfs_run_delayed_iputs(fs_info);
4288 mutex_unlock(&fs_info->cleaner_mutex);
4289 wake_up_process(fs_info->cleaner_kthread);
4290
4291 /* wait until ongoing cleanup work done */
4292 down_write(&fs_info->cleanup_work_sem);
4293 up_write(&fs_info->cleanup_work_sem);
4294
4295 trans = btrfs_join_transaction(root);
4296 if (IS_ERR(trans))
4297 return PTR_ERR(trans);
4298 return btrfs_commit_transaction(trans);
4299}
4300
4301void __cold close_ctree(struct btrfs_fs_info *fs_info)
4302{
4303 int ret;
4304
4305 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4306 /*
4307 * We don't want the cleaner to start new transactions, add more delayed
4308 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4309 * because that frees the task_struct, and the transaction kthread might
4310 * still try to wake up the cleaner.
4311 */
4312 kthread_park(fs_info->cleaner_kthread);
4313
4314 /* wait for the qgroup rescan worker to stop */
4315 btrfs_qgroup_wait_for_completion(fs_info, false);
4316
4317 /* wait for the uuid_scan task to finish */
4318 down(&fs_info->uuid_tree_rescan_sem);
4319 /* avoid complains from lockdep et al., set sem back to initial state */
4320 up(&fs_info->uuid_tree_rescan_sem);
4321
4322 /* pause restriper - we want to resume on mount */
4323 btrfs_pause_balance(fs_info);
4324
4325 btrfs_dev_replace_suspend_for_unmount(fs_info);
4326
4327 btrfs_scrub_cancel(fs_info);
4328
4329 /* wait for any defraggers to finish */
4330 wait_event(fs_info->transaction_wait,
4331 (atomic_read(&fs_info->defrag_running) == 0));
4332
4333 /* clear out the rbtree of defraggable inodes */
4334 btrfs_cleanup_defrag_inodes(fs_info);
4335
4336 cancel_work_sync(&fs_info->async_reclaim_work);
4337 cancel_work_sync(&fs_info->async_data_reclaim_work);
4338 cancel_work_sync(&fs_info->preempt_reclaim_work);
4339
4340 cancel_work_sync(&fs_info->reclaim_bgs_work);
4341
4342 /* Cancel or finish ongoing discard work */
4343 btrfs_discard_cleanup(fs_info);
4344
4345 if (!sb_rdonly(fs_info->sb)) {
4346 /*
4347 * The cleaner kthread is stopped, so do one final pass over
4348 * unused block groups.
4349 */
4350 btrfs_delete_unused_bgs(fs_info);
4351
4352 /*
4353 * There might be existing delayed inode workers still running
4354 * and holding an empty delayed inode item. We must wait for
4355 * them to complete first because they can create a transaction.
4356 * This happens when someone calls btrfs_balance_delayed_items()
4357 * and then a transaction commit runs the same delayed nodes
4358 * before any delayed worker has done something with the nodes.
4359 * We must wait for any worker here and not at transaction
4360 * commit time since that could cause a deadlock.
4361 * This is a very rare case.
4362 */
4363 btrfs_flush_workqueue(fs_info->delayed_workers);
4364
4365 ret = btrfs_commit_super(fs_info);
4366 if (ret)
4367 btrfs_err(fs_info, "commit super ret %d", ret);
4368 }
4369
4370 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4371 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4372 btrfs_error_commit_super(fs_info);
4373
4374 kthread_stop(fs_info->transaction_kthread);
4375 kthread_stop(fs_info->cleaner_kthread);
4376
4377 ASSERT(list_empty(&fs_info->delayed_iputs));
4378 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4379
4380 if (btrfs_check_quota_leak(fs_info)) {
4381 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4382 btrfs_err(fs_info, "qgroup reserved space leaked");
4383 }
4384
4385 btrfs_free_qgroup_config(fs_info);
4386 ASSERT(list_empty(&fs_info->delalloc_roots));
4387
4388 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4389 btrfs_info(fs_info, "at unmount delalloc count %lld",
4390 percpu_counter_sum(&fs_info->delalloc_bytes));
4391 }
4392
4393 if (percpu_counter_sum(&fs_info->ordered_bytes))
4394 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4395 percpu_counter_sum(&fs_info->ordered_bytes));
4396
4397 btrfs_sysfs_remove_mounted(fs_info);
4398 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4399
4400 btrfs_put_block_group_cache(fs_info);
4401
4402 /*
4403 * we must make sure there is not any read request to
4404 * submit after we stopping all workers.
4405 */
4406 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4407 btrfs_stop_all_workers(fs_info);
4408
4409 /* We shouldn't have any transaction open at this point */
4410 ASSERT(list_empty(&fs_info->trans_list));
4411
4412 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4413 free_root_pointers(fs_info, true);
4414 btrfs_free_fs_roots(fs_info);
4415
4416 /*
4417 * We must free the block groups after dropping the fs_roots as we could
4418 * have had an IO error and have left over tree log blocks that aren't
4419 * cleaned up until the fs roots are freed. This makes the block group
4420 * accounting appear to be wrong because there's pending reserved bytes,
4421 * so make sure we do the block group cleanup afterwards.
4422 */
4423 btrfs_free_block_groups(fs_info);
4424
4425 iput(fs_info->btree_inode);
4426
4427#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4428 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4429 btrfsic_unmount(fs_info->fs_devices);
4430#endif
4431
4432 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4433 btrfs_close_devices(fs_info->fs_devices);
4434}
4435
4436int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4437 int atomic)
4438{
4439 int ret;
4440 struct inode *btree_inode = buf->pages[0]->mapping->host;
4441
4442 ret = extent_buffer_uptodate(buf);
4443 if (!ret)
4444 return ret;
4445
4446 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4447 parent_transid, atomic);
4448 if (ret == -EAGAIN)
4449 return ret;
4450 return !ret;
4451}
4452
4453void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4454{
4455 struct btrfs_fs_info *fs_info = buf->fs_info;
4456 u64 transid = btrfs_header_generation(buf);
4457 int was_dirty;
4458
4459#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4460 /*
4461 * This is a fast path so only do this check if we have sanity tests
4462 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4463 * outside of the sanity tests.
4464 */
4465 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4466 return;
4467#endif
4468 btrfs_assert_tree_locked(buf);
4469 if (transid != fs_info->generation)
4470 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4471 buf->start, transid, fs_info->generation);
4472 was_dirty = set_extent_buffer_dirty(buf);
4473 if (!was_dirty)
4474 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4475 buf->len,
4476 fs_info->dirty_metadata_batch);
4477#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4478 /*
4479 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4480 * but item data not updated.
4481 * So here we should only check item pointers, not item data.
4482 */
4483 if (btrfs_header_level(buf) == 0 &&
4484 btrfs_check_leaf_relaxed(buf)) {
4485 btrfs_print_leaf(buf);
4486 ASSERT(0);
4487 }
4488#endif
4489}
4490
4491static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4492 int flush_delayed)
4493{
4494 /*
4495 * looks as though older kernels can get into trouble with
4496 * this code, they end up stuck in balance_dirty_pages forever
4497 */
4498 int ret;
4499
4500 if (current->flags & PF_MEMALLOC)
4501 return;
4502
4503 if (flush_delayed)
4504 btrfs_balance_delayed_items(fs_info);
4505
4506 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4507 BTRFS_DIRTY_METADATA_THRESH,
4508 fs_info->dirty_metadata_batch);
4509 if (ret > 0) {
4510 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4511 }
4512}
4513
4514void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4515{
4516 __btrfs_btree_balance_dirty(fs_info, 1);
4517}
4518
4519void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4520{
4521 __btrfs_btree_balance_dirty(fs_info, 0);
4522}
4523
4524int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4525 struct btrfs_key *first_key)
4526{
4527 return btree_read_extent_buffer_pages(buf, parent_transid,
4528 level, first_key);
4529}
4530
4531static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4532{
4533 /* cleanup FS via transaction */
4534 btrfs_cleanup_transaction(fs_info);
4535
4536 mutex_lock(&fs_info->cleaner_mutex);
4537 btrfs_run_delayed_iputs(fs_info);
4538 mutex_unlock(&fs_info->cleaner_mutex);
4539
4540 down_write(&fs_info->cleanup_work_sem);
4541 up_write(&fs_info->cleanup_work_sem);
4542}
4543
4544static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4545{
4546 struct btrfs_root *gang[8];
4547 u64 root_objectid = 0;
4548 int ret;
4549
4550 spin_lock(&fs_info->fs_roots_radix_lock);
4551 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4552 (void **)gang, root_objectid,
4553 ARRAY_SIZE(gang))) != 0) {
4554 int i;
4555
4556 for (i = 0; i < ret; i++)
4557 gang[i] = btrfs_grab_root(gang[i]);
4558 spin_unlock(&fs_info->fs_roots_radix_lock);
4559
4560 for (i = 0; i < ret; i++) {
4561 if (!gang[i])
4562 continue;
4563 root_objectid = gang[i]->root_key.objectid;
4564 btrfs_free_log(NULL, gang[i]);
4565 btrfs_put_root(gang[i]);
4566 }
4567 root_objectid++;
4568 spin_lock(&fs_info->fs_roots_radix_lock);
4569 }
4570 spin_unlock(&fs_info->fs_roots_radix_lock);
4571 btrfs_free_log_root_tree(NULL, fs_info);
4572}
4573
4574static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4575{
4576 struct btrfs_ordered_extent *ordered;
4577
4578 spin_lock(&root->ordered_extent_lock);
4579 /*
4580 * This will just short circuit the ordered completion stuff which will
4581 * make sure the ordered extent gets properly cleaned up.
4582 */
4583 list_for_each_entry(ordered, &root->ordered_extents,
4584 root_extent_list)
4585 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4586 spin_unlock(&root->ordered_extent_lock);
4587}
4588
4589static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4590{
4591 struct btrfs_root *root;
4592 struct list_head splice;
4593
4594 INIT_LIST_HEAD(&splice);
4595
4596 spin_lock(&fs_info->ordered_root_lock);
4597 list_splice_init(&fs_info->ordered_roots, &splice);
4598 while (!list_empty(&splice)) {
4599 root = list_first_entry(&splice, struct btrfs_root,
4600 ordered_root);
4601 list_move_tail(&root->ordered_root,
4602 &fs_info->ordered_roots);
4603
4604 spin_unlock(&fs_info->ordered_root_lock);
4605 btrfs_destroy_ordered_extents(root);
4606
4607 cond_resched();
4608 spin_lock(&fs_info->ordered_root_lock);
4609 }
4610 spin_unlock(&fs_info->ordered_root_lock);
4611
4612 /*
4613 * We need this here because if we've been flipped read-only we won't
4614 * get sync() from the umount, so we need to make sure any ordered
4615 * extents that haven't had their dirty pages IO start writeout yet
4616 * actually get run and error out properly.
4617 */
4618 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4619}
4620
4621static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4622 struct btrfs_fs_info *fs_info)
4623{
4624 struct rb_node *node;
4625 struct btrfs_delayed_ref_root *delayed_refs;
4626 struct btrfs_delayed_ref_node *ref;
4627 int ret = 0;
4628
4629 delayed_refs = &trans->delayed_refs;
4630
4631 spin_lock(&delayed_refs->lock);
4632 if (atomic_read(&delayed_refs->num_entries) == 0) {
4633 spin_unlock(&delayed_refs->lock);
4634 btrfs_debug(fs_info, "delayed_refs has NO entry");
4635 return ret;
4636 }
4637
4638 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4639 struct btrfs_delayed_ref_head *head;
4640 struct rb_node *n;
4641 bool pin_bytes = false;
4642
4643 head = rb_entry(node, struct btrfs_delayed_ref_head,
4644 href_node);
4645 if (btrfs_delayed_ref_lock(delayed_refs, head))
4646 continue;
4647
4648 spin_lock(&head->lock);
4649 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4650 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4651 ref_node);
4652 ref->in_tree = 0;
4653 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4654 RB_CLEAR_NODE(&ref->ref_node);
4655 if (!list_empty(&ref->add_list))
4656 list_del(&ref->add_list);
4657 atomic_dec(&delayed_refs->num_entries);
4658 btrfs_put_delayed_ref(ref);
4659 }
4660 if (head->must_insert_reserved)
4661 pin_bytes = true;
4662 btrfs_free_delayed_extent_op(head->extent_op);
4663 btrfs_delete_ref_head(delayed_refs, head);
4664 spin_unlock(&head->lock);
4665 spin_unlock(&delayed_refs->lock);
4666 mutex_unlock(&head->mutex);
4667
4668 if (pin_bytes) {
4669 struct btrfs_block_group *cache;
4670
4671 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4672 BUG_ON(!cache);
4673
4674 spin_lock(&cache->space_info->lock);
4675 spin_lock(&cache->lock);
4676 cache->pinned += head->num_bytes;
4677 btrfs_space_info_update_bytes_pinned(fs_info,
4678 cache->space_info, head->num_bytes);
4679 cache->reserved -= head->num_bytes;
4680 cache->space_info->bytes_reserved -= head->num_bytes;
4681 spin_unlock(&cache->lock);
4682 spin_unlock(&cache->space_info->lock);
4683
4684 btrfs_put_block_group(cache);
4685
4686 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4687 head->bytenr + head->num_bytes - 1);
4688 }
4689 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4690 btrfs_put_delayed_ref_head(head);
4691 cond_resched();
4692 spin_lock(&delayed_refs->lock);
4693 }
4694 btrfs_qgroup_destroy_extent_records(trans);
4695
4696 spin_unlock(&delayed_refs->lock);
4697
4698 return ret;
4699}
4700
4701static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4702{
4703 struct btrfs_inode *btrfs_inode;
4704 struct list_head splice;
4705
4706 INIT_LIST_HEAD(&splice);
4707
4708 spin_lock(&root->delalloc_lock);
4709 list_splice_init(&root->delalloc_inodes, &splice);
4710
4711 while (!list_empty(&splice)) {
4712 struct inode *inode = NULL;
4713 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4714 delalloc_inodes);
4715 __btrfs_del_delalloc_inode(root, btrfs_inode);
4716 spin_unlock(&root->delalloc_lock);
4717
4718 /*
4719 * Make sure we get a live inode and that it'll not disappear
4720 * meanwhile.
4721 */
4722 inode = igrab(&btrfs_inode->vfs_inode);
4723 if (inode) {
4724 invalidate_inode_pages2(inode->i_mapping);
4725 iput(inode);
4726 }
4727 spin_lock(&root->delalloc_lock);
4728 }
4729 spin_unlock(&root->delalloc_lock);
4730}
4731
4732static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4733{
4734 struct btrfs_root *root;
4735 struct list_head splice;
4736
4737 INIT_LIST_HEAD(&splice);
4738
4739 spin_lock(&fs_info->delalloc_root_lock);
4740 list_splice_init(&fs_info->delalloc_roots, &splice);
4741 while (!list_empty(&splice)) {
4742 root = list_first_entry(&splice, struct btrfs_root,
4743 delalloc_root);
4744 root = btrfs_grab_root(root);
4745 BUG_ON(!root);
4746 spin_unlock(&fs_info->delalloc_root_lock);
4747
4748 btrfs_destroy_delalloc_inodes(root);
4749 btrfs_put_root(root);
4750
4751 spin_lock(&fs_info->delalloc_root_lock);
4752 }
4753 spin_unlock(&fs_info->delalloc_root_lock);
4754}
4755
4756static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4757 struct extent_io_tree *dirty_pages,
4758 int mark)
4759{
4760 int ret;
4761 struct extent_buffer *eb;
4762 u64 start = 0;
4763 u64 end;
4764
4765 while (1) {
4766 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4767 mark, NULL);
4768 if (ret)
4769 break;
4770
4771 clear_extent_bits(dirty_pages, start, end, mark);
4772 while (start <= end) {
4773 eb = find_extent_buffer(fs_info, start);
4774 start += fs_info->nodesize;
4775 if (!eb)
4776 continue;
4777 wait_on_extent_buffer_writeback(eb);
4778
4779 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4780 &eb->bflags))
4781 clear_extent_buffer_dirty(eb);
4782 free_extent_buffer_stale(eb);
4783 }
4784 }
4785
4786 return ret;
4787}
4788
4789static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4790 struct extent_io_tree *unpin)
4791{
4792 u64 start;
4793 u64 end;
4794 int ret;
4795
4796 while (1) {
4797 struct extent_state *cached_state = NULL;
4798
4799 /*
4800 * The btrfs_finish_extent_commit() may get the same range as
4801 * ours between find_first_extent_bit and clear_extent_dirty.
4802 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4803 * the same extent range.
4804 */
4805 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4806 ret = find_first_extent_bit(unpin, 0, &start, &end,
4807 EXTENT_DIRTY, &cached_state);
4808 if (ret) {
4809 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4810 break;
4811 }
4812
4813 clear_extent_dirty(unpin, start, end, &cached_state);
4814 free_extent_state(cached_state);
4815 btrfs_error_unpin_extent_range(fs_info, start, end);
4816 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4817 cond_resched();
4818 }
4819
4820 return 0;
4821}
4822
4823static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4824{
4825 struct inode *inode;
4826
4827 inode = cache->io_ctl.inode;
4828 if (inode) {
4829 invalidate_inode_pages2(inode->i_mapping);
4830 BTRFS_I(inode)->generation = 0;
4831 cache->io_ctl.inode = NULL;
4832 iput(inode);
4833 }
4834 ASSERT(cache->io_ctl.pages == NULL);
4835 btrfs_put_block_group(cache);
4836}
4837
4838void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4839 struct btrfs_fs_info *fs_info)
4840{
4841 struct btrfs_block_group *cache;
4842
4843 spin_lock(&cur_trans->dirty_bgs_lock);
4844 while (!list_empty(&cur_trans->dirty_bgs)) {
4845 cache = list_first_entry(&cur_trans->dirty_bgs,
4846 struct btrfs_block_group,
4847 dirty_list);
4848
4849 if (!list_empty(&cache->io_list)) {
4850 spin_unlock(&cur_trans->dirty_bgs_lock);
4851 list_del_init(&cache->io_list);
4852 btrfs_cleanup_bg_io(cache);
4853 spin_lock(&cur_trans->dirty_bgs_lock);
4854 }
4855
4856 list_del_init(&cache->dirty_list);
4857 spin_lock(&cache->lock);
4858 cache->disk_cache_state = BTRFS_DC_ERROR;
4859 spin_unlock(&cache->lock);
4860
4861 spin_unlock(&cur_trans->dirty_bgs_lock);
4862 btrfs_put_block_group(cache);
4863 btrfs_delayed_refs_rsv_release(fs_info, 1);
4864 spin_lock(&cur_trans->dirty_bgs_lock);
4865 }
4866 spin_unlock(&cur_trans->dirty_bgs_lock);
4867
4868 /*
4869 * Refer to the definition of io_bgs member for details why it's safe
4870 * to use it without any locking
4871 */
4872 while (!list_empty(&cur_trans->io_bgs)) {
4873 cache = list_first_entry(&cur_trans->io_bgs,
4874 struct btrfs_block_group,
4875 io_list);
4876
4877 list_del_init(&cache->io_list);
4878 spin_lock(&cache->lock);
4879 cache->disk_cache_state = BTRFS_DC_ERROR;
4880 spin_unlock(&cache->lock);
4881 btrfs_cleanup_bg_io(cache);
4882 }
4883}
4884
4885void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4886 struct btrfs_fs_info *fs_info)
4887{
4888 struct btrfs_device *dev, *tmp;
4889
4890 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4891 ASSERT(list_empty(&cur_trans->dirty_bgs));
4892 ASSERT(list_empty(&cur_trans->io_bgs));
4893
4894 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4895 post_commit_list) {
4896 list_del_init(&dev->post_commit_list);
4897 }
4898
4899 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4900
4901 cur_trans->state = TRANS_STATE_COMMIT_START;
4902 wake_up(&fs_info->transaction_blocked_wait);
4903
4904 cur_trans->state = TRANS_STATE_UNBLOCKED;
4905 wake_up(&fs_info->transaction_wait);
4906
4907 btrfs_destroy_delayed_inodes(fs_info);
4908
4909 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4910 EXTENT_DIRTY);
4911 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4912
4913 btrfs_free_redirty_list(cur_trans);
4914
4915 cur_trans->state =TRANS_STATE_COMPLETED;
4916 wake_up(&cur_trans->commit_wait);
4917}
4918
4919static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4920{
4921 struct btrfs_transaction *t;
4922
4923 mutex_lock(&fs_info->transaction_kthread_mutex);
4924
4925 spin_lock(&fs_info->trans_lock);
4926 while (!list_empty(&fs_info->trans_list)) {
4927 t = list_first_entry(&fs_info->trans_list,
4928 struct btrfs_transaction, list);
4929 if (t->state >= TRANS_STATE_COMMIT_START) {
4930 refcount_inc(&t->use_count);
4931 spin_unlock(&fs_info->trans_lock);
4932 btrfs_wait_for_commit(fs_info, t->transid);
4933 btrfs_put_transaction(t);
4934 spin_lock(&fs_info->trans_lock);
4935 continue;
4936 }
4937 if (t == fs_info->running_transaction) {
4938 t->state = TRANS_STATE_COMMIT_DOING;
4939 spin_unlock(&fs_info->trans_lock);
4940 /*
4941 * We wait for 0 num_writers since we don't hold a trans
4942 * handle open currently for this transaction.
4943 */
4944 wait_event(t->writer_wait,
4945 atomic_read(&t->num_writers) == 0);
4946 } else {
4947 spin_unlock(&fs_info->trans_lock);
4948 }
4949 btrfs_cleanup_one_transaction(t, fs_info);
4950
4951 spin_lock(&fs_info->trans_lock);
4952 if (t == fs_info->running_transaction)
4953 fs_info->running_transaction = NULL;
4954 list_del_init(&t->list);
4955 spin_unlock(&fs_info->trans_lock);
4956
4957 btrfs_put_transaction(t);
4958 trace_btrfs_transaction_commit(fs_info->tree_root);
4959 spin_lock(&fs_info->trans_lock);
4960 }
4961 spin_unlock(&fs_info->trans_lock);
4962 btrfs_destroy_all_ordered_extents(fs_info);
4963 btrfs_destroy_delayed_inodes(fs_info);
4964 btrfs_assert_delayed_root_empty(fs_info);
4965 btrfs_destroy_all_delalloc_inodes(fs_info);
4966 btrfs_drop_all_logs(fs_info);
4967 mutex_unlock(&fs_info->transaction_kthread_mutex);
4968
4969 return 0;
4970}
4971
4972int btrfs_init_root_free_objectid(struct btrfs_root *root)
4973{
4974 struct btrfs_path *path;
4975 int ret;
4976 struct extent_buffer *l;
4977 struct btrfs_key search_key;
4978 struct btrfs_key found_key;
4979 int slot;
4980
4981 path = btrfs_alloc_path();
4982 if (!path)
4983 return -ENOMEM;
4984
4985 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4986 search_key.type = -1;
4987 search_key.offset = (u64)-1;
4988 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4989 if (ret < 0)
4990 goto error;
4991 BUG_ON(ret == 0); /* Corruption */
4992 if (path->slots[0] > 0) {
4993 slot = path->slots[0] - 1;
4994 l = path->nodes[0];
4995 btrfs_item_key_to_cpu(l, &found_key, slot);
4996 root->free_objectid = max_t(u64, found_key.objectid + 1,
4997 BTRFS_FIRST_FREE_OBJECTID);
4998 } else {
4999 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
5000 }
5001 ret = 0;
5002error:
5003 btrfs_free_path(path);
5004 return ret;
5005}
5006
5007int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
5008{
5009 int ret;
5010 mutex_lock(&root->objectid_mutex);
5011
5012 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
5013 btrfs_warn(root->fs_info,
5014 "the objectid of root %llu reaches its highest value",
5015 root->root_key.objectid);
5016 ret = -ENOSPC;
5017 goto out;
5018 }
5019
5020 *objectid = root->free_objectid++;
5021 ret = 0;
5022out:
5023 mutex_unlock(&root->objectid_mutex);
5024 return ret;
5025}
1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/fs.h>
20#include <linux/blkdev.h>
21#include <linux/scatterlist.h>
22#include <linux/swap.h>
23#include <linux/radix-tree.h>
24#include <linux/writeback.h>
25#include <linux/buffer_head.h>
26#include <linux/workqueue.h>
27#include <linux/kthread.h>
28#include <linux/slab.h>
29#include <linux/migrate.h>
30#include <linux/ratelimit.h>
31#include <linux/uuid.h>
32#include <linux/semaphore.h>
33#include <asm/unaligned.h>
34#include "ctree.h"
35#include "disk-io.h"
36#include "hash.h"
37#include "transaction.h"
38#include "btrfs_inode.h"
39#include "volumes.h"
40#include "print-tree.h"
41#include "locking.h"
42#include "tree-log.h"
43#include "free-space-cache.h"
44#include "free-space-tree.h"
45#include "inode-map.h"
46#include "check-integrity.h"
47#include "rcu-string.h"
48#include "dev-replace.h"
49#include "raid56.h"
50#include "sysfs.h"
51#include "qgroup.h"
52#include "compression.h"
53
54#ifdef CONFIG_X86
55#include <asm/cpufeature.h>
56#endif
57
58#define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
59 BTRFS_HEADER_FLAG_RELOC |\
60 BTRFS_SUPER_FLAG_ERROR |\
61 BTRFS_SUPER_FLAG_SEEDING |\
62 BTRFS_SUPER_FLAG_METADUMP)
63
64static const struct extent_io_ops btree_extent_io_ops;
65static void end_workqueue_fn(struct btrfs_work *work);
66static void free_fs_root(struct btrfs_root *root);
67static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
68 int read_only);
69static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
70static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
71 struct btrfs_fs_info *fs_info);
72static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
73static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
74 struct extent_io_tree *dirty_pages,
75 int mark);
76static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
77 struct extent_io_tree *pinned_extents);
78static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
79static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
80
81/*
82 * btrfs_end_io_wq structs are used to do processing in task context when an IO
83 * is complete. This is used during reads to verify checksums, and it is used
84 * by writes to insert metadata for new file extents after IO is complete.
85 */
86struct btrfs_end_io_wq {
87 struct bio *bio;
88 bio_end_io_t *end_io;
89 void *private;
90 struct btrfs_fs_info *info;
91 int error;
92 enum btrfs_wq_endio_type metadata;
93 struct list_head list;
94 struct btrfs_work work;
95};
96
97static struct kmem_cache *btrfs_end_io_wq_cache;
98
99int __init btrfs_end_io_wq_init(void)
100{
101 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
102 sizeof(struct btrfs_end_io_wq),
103 0,
104 SLAB_MEM_SPREAD,
105 NULL);
106 if (!btrfs_end_io_wq_cache)
107 return -ENOMEM;
108 return 0;
109}
110
111void btrfs_end_io_wq_exit(void)
112{
113 kmem_cache_destroy(btrfs_end_io_wq_cache);
114}
115
116/*
117 * async submit bios are used to offload expensive checksumming
118 * onto the worker threads. They checksum file and metadata bios
119 * just before they are sent down the IO stack.
120 */
121struct async_submit_bio {
122 struct inode *inode;
123 struct bio *bio;
124 struct list_head list;
125 extent_submit_bio_hook_t *submit_bio_start;
126 extent_submit_bio_hook_t *submit_bio_done;
127 int mirror_num;
128 unsigned long bio_flags;
129 /*
130 * bio_offset is optional, can be used if the pages in the bio
131 * can't tell us where in the file the bio should go
132 */
133 u64 bio_offset;
134 struct btrfs_work work;
135 int error;
136};
137
138/*
139 * Lockdep class keys for extent_buffer->lock's in this root. For a given
140 * eb, the lockdep key is determined by the btrfs_root it belongs to and
141 * the level the eb occupies in the tree.
142 *
143 * Different roots are used for different purposes and may nest inside each
144 * other and they require separate keysets. As lockdep keys should be
145 * static, assign keysets according to the purpose of the root as indicated
146 * by btrfs_root->objectid. This ensures that all special purpose roots
147 * have separate keysets.
148 *
149 * Lock-nesting across peer nodes is always done with the immediate parent
150 * node locked thus preventing deadlock. As lockdep doesn't know this, use
151 * subclass to avoid triggering lockdep warning in such cases.
152 *
153 * The key is set by the readpage_end_io_hook after the buffer has passed
154 * csum validation but before the pages are unlocked. It is also set by
155 * btrfs_init_new_buffer on freshly allocated blocks.
156 *
157 * We also add a check to make sure the highest level of the tree is the
158 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
159 * needs update as well.
160 */
161#ifdef CONFIG_DEBUG_LOCK_ALLOC
162# if BTRFS_MAX_LEVEL != 8
163# error
164# endif
165
166static struct btrfs_lockdep_keyset {
167 u64 id; /* root objectid */
168 const char *name_stem; /* lock name stem */
169 char names[BTRFS_MAX_LEVEL + 1][20];
170 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
171} btrfs_lockdep_keysets[] = {
172 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
173 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
174 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
175 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
176 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
177 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
178 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
179 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
180 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
181 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
182 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
183 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
184 { .id = 0, .name_stem = "tree" },
185};
186
187void __init btrfs_init_lockdep(void)
188{
189 int i, j;
190
191 /* initialize lockdep class names */
192 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
193 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
194
195 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
196 snprintf(ks->names[j], sizeof(ks->names[j]),
197 "btrfs-%s-%02d", ks->name_stem, j);
198 }
199}
200
201void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
202 int level)
203{
204 struct btrfs_lockdep_keyset *ks;
205
206 BUG_ON(level >= ARRAY_SIZE(ks->keys));
207
208 /* find the matching keyset, id 0 is the default entry */
209 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
210 if (ks->id == objectid)
211 break;
212
213 lockdep_set_class_and_name(&eb->lock,
214 &ks->keys[level], ks->names[level]);
215}
216
217#endif
218
219/*
220 * extents on the btree inode are pretty simple, there's one extent
221 * that covers the entire device
222 */
223static struct extent_map *btree_get_extent(struct inode *inode,
224 struct page *page, size_t pg_offset, u64 start, u64 len,
225 int create)
226{
227 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
228 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
229 struct extent_map *em;
230 int ret;
231
232 read_lock(&em_tree->lock);
233 em = lookup_extent_mapping(em_tree, start, len);
234 if (em) {
235 em->bdev = fs_info->fs_devices->latest_bdev;
236 read_unlock(&em_tree->lock);
237 goto out;
238 }
239 read_unlock(&em_tree->lock);
240
241 em = alloc_extent_map();
242 if (!em) {
243 em = ERR_PTR(-ENOMEM);
244 goto out;
245 }
246 em->start = 0;
247 em->len = (u64)-1;
248 em->block_len = (u64)-1;
249 em->block_start = 0;
250 em->bdev = fs_info->fs_devices->latest_bdev;
251
252 write_lock(&em_tree->lock);
253 ret = add_extent_mapping(em_tree, em, 0);
254 if (ret == -EEXIST) {
255 free_extent_map(em);
256 em = lookup_extent_mapping(em_tree, start, len);
257 if (!em)
258 em = ERR_PTR(-EIO);
259 } else if (ret) {
260 free_extent_map(em);
261 em = ERR_PTR(ret);
262 }
263 write_unlock(&em_tree->lock);
264
265out:
266 return em;
267}
268
269u32 btrfs_csum_data(char *data, u32 seed, size_t len)
270{
271 return btrfs_crc32c(seed, data, len);
272}
273
274void btrfs_csum_final(u32 crc, u8 *result)
275{
276 put_unaligned_le32(~crc, result);
277}
278
279/*
280 * compute the csum for a btree block, and either verify it or write it
281 * into the csum field of the block.
282 */
283static int csum_tree_block(struct btrfs_fs_info *fs_info,
284 struct extent_buffer *buf,
285 int verify)
286{
287 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
288 char *result = NULL;
289 unsigned long len;
290 unsigned long cur_len;
291 unsigned long offset = BTRFS_CSUM_SIZE;
292 char *kaddr;
293 unsigned long map_start;
294 unsigned long map_len;
295 int err;
296 u32 crc = ~(u32)0;
297 unsigned long inline_result;
298
299 len = buf->len - offset;
300 while (len > 0) {
301 err = map_private_extent_buffer(buf, offset, 32,
302 &kaddr, &map_start, &map_len);
303 if (err)
304 return err;
305 cur_len = min(len, map_len - (offset - map_start));
306 crc = btrfs_csum_data(kaddr + offset - map_start,
307 crc, cur_len);
308 len -= cur_len;
309 offset += cur_len;
310 }
311 if (csum_size > sizeof(inline_result)) {
312 result = kzalloc(csum_size, GFP_NOFS);
313 if (!result)
314 return -ENOMEM;
315 } else {
316 result = (char *)&inline_result;
317 }
318
319 btrfs_csum_final(crc, result);
320
321 if (verify) {
322 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
323 u32 val;
324 u32 found = 0;
325 memcpy(&found, result, csum_size);
326
327 read_extent_buffer(buf, &val, 0, csum_size);
328 btrfs_warn_rl(fs_info,
329 "%s checksum verify failed on %llu wanted %X found %X level %d",
330 fs_info->sb->s_id, buf->start,
331 val, found, btrfs_header_level(buf));
332 if (result != (char *)&inline_result)
333 kfree(result);
334 return -EUCLEAN;
335 }
336 } else {
337 write_extent_buffer(buf, result, 0, csum_size);
338 }
339 if (result != (char *)&inline_result)
340 kfree(result);
341 return 0;
342}
343
344/*
345 * we can't consider a given block up to date unless the transid of the
346 * block matches the transid in the parent node's pointer. This is how we
347 * detect blocks that either didn't get written at all or got written
348 * in the wrong place.
349 */
350static int verify_parent_transid(struct extent_io_tree *io_tree,
351 struct extent_buffer *eb, u64 parent_transid,
352 int atomic)
353{
354 struct extent_state *cached_state = NULL;
355 int ret;
356 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
357
358 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
359 return 0;
360
361 if (atomic)
362 return -EAGAIN;
363
364 if (need_lock) {
365 btrfs_tree_read_lock(eb);
366 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
367 }
368
369 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
370 &cached_state);
371 if (extent_buffer_uptodate(eb) &&
372 btrfs_header_generation(eb) == parent_transid) {
373 ret = 0;
374 goto out;
375 }
376 btrfs_err_rl(eb->fs_info,
377 "parent transid verify failed on %llu wanted %llu found %llu",
378 eb->start,
379 parent_transid, btrfs_header_generation(eb));
380 ret = 1;
381
382 /*
383 * Things reading via commit roots that don't have normal protection,
384 * like send, can have a really old block in cache that may point at a
385 * block that has been freed and re-allocated. So don't clear uptodate
386 * if we find an eb that is under IO (dirty/writeback) because we could
387 * end up reading in the stale data and then writing it back out and
388 * making everybody very sad.
389 */
390 if (!extent_buffer_under_io(eb))
391 clear_extent_buffer_uptodate(eb);
392out:
393 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
394 &cached_state, GFP_NOFS);
395 if (need_lock)
396 btrfs_tree_read_unlock_blocking(eb);
397 return ret;
398}
399
400/*
401 * Return 0 if the superblock checksum type matches the checksum value of that
402 * algorithm. Pass the raw disk superblock data.
403 */
404static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
405 char *raw_disk_sb)
406{
407 struct btrfs_super_block *disk_sb =
408 (struct btrfs_super_block *)raw_disk_sb;
409 u16 csum_type = btrfs_super_csum_type(disk_sb);
410 int ret = 0;
411
412 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
413 u32 crc = ~(u32)0;
414 const int csum_size = sizeof(crc);
415 char result[csum_size];
416
417 /*
418 * The super_block structure does not span the whole
419 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
420 * is filled with zeros and is included in the checksum.
421 */
422 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
423 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
424 btrfs_csum_final(crc, result);
425
426 if (memcmp(raw_disk_sb, result, csum_size))
427 ret = 1;
428 }
429
430 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
431 btrfs_err(fs_info, "unsupported checksum algorithm %u",
432 csum_type);
433 ret = 1;
434 }
435
436 return ret;
437}
438
439/*
440 * helper to read a given tree block, doing retries as required when
441 * the checksums don't match and we have alternate mirrors to try.
442 */
443static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
444 struct extent_buffer *eb,
445 u64 parent_transid)
446{
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 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
455 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
456 while (1) {
457 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
458 btree_get_extent, mirror_num);
459 if (!ret) {
460 if (!verify_parent_transid(io_tree, eb,
461 parent_transid, 0))
462 break;
463 else
464 ret = -EIO;
465 }
466
467 /*
468 * This buffer's crc is fine, but its contents are corrupted, so
469 * there is no reason to read the other copies, they won't be
470 * any less wrong.
471 */
472 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
473 break;
474
475 num_copies = btrfs_num_copies(fs_info,
476 eb->start, eb->len);
477 if (num_copies == 1)
478 break;
479
480 if (!failed_mirror) {
481 failed = 1;
482 failed_mirror = eb->read_mirror;
483 }
484
485 mirror_num++;
486 if (mirror_num == failed_mirror)
487 mirror_num++;
488
489 if (mirror_num > num_copies)
490 break;
491 }
492
493 if (failed && !ret && failed_mirror)
494 repair_eb_io_failure(fs_info, eb, failed_mirror);
495
496 return ret;
497}
498
499/*
500 * checksum a dirty tree block before IO. This has extra checks to make sure
501 * we only fill in the checksum field in the first page of a multi-page block
502 */
503
504static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
505{
506 u64 start = page_offset(page);
507 u64 found_start;
508 struct extent_buffer *eb;
509
510 eb = (struct extent_buffer *)page->private;
511 if (page != eb->pages[0])
512 return 0;
513
514 found_start = btrfs_header_bytenr(eb);
515 /*
516 * Please do not consolidate these warnings into a single if.
517 * It is useful to know what went wrong.
518 */
519 if (WARN_ON(found_start != start))
520 return -EUCLEAN;
521 if (WARN_ON(!PageUptodate(page)))
522 return -EUCLEAN;
523
524 ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
525 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
526
527 return csum_tree_block(fs_info, eb, 0);
528}
529
530static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
531 struct extent_buffer *eb)
532{
533 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
534 u8 fsid[BTRFS_UUID_SIZE];
535 int ret = 1;
536
537 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
538 while (fs_devices) {
539 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
540 ret = 0;
541 break;
542 }
543 fs_devices = fs_devices->seed;
544 }
545 return ret;
546}
547
548#define CORRUPT(reason, eb, root, slot) \
549 btrfs_crit(root->fs_info, \
550 "corrupt %s, %s: block=%llu, root=%llu, slot=%d", \
551 btrfs_header_level(eb) == 0 ? "leaf" : "node", \
552 reason, btrfs_header_bytenr(eb), root->objectid, slot)
553
554static noinline int check_leaf(struct btrfs_root *root,
555 struct extent_buffer *leaf)
556{
557 struct btrfs_fs_info *fs_info = root->fs_info;
558 struct btrfs_key key;
559 struct btrfs_key leaf_key;
560 u32 nritems = btrfs_header_nritems(leaf);
561 int slot;
562
563 /*
564 * Extent buffers from a relocation tree have a owner field that
565 * corresponds to the subvolume tree they are based on. So just from an
566 * extent buffer alone we can not find out what is the id of the
567 * corresponding subvolume tree, so we can not figure out if the extent
568 * buffer corresponds to the root of the relocation tree or not. So skip
569 * this check for relocation trees.
570 */
571 if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) {
572 struct btrfs_root *check_root;
573
574 key.objectid = btrfs_header_owner(leaf);
575 key.type = BTRFS_ROOT_ITEM_KEY;
576 key.offset = (u64)-1;
577
578 check_root = btrfs_get_fs_root(fs_info, &key, false);
579 /*
580 * The only reason we also check NULL here is that during
581 * open_ctree() some roots has not yet been set up.
582 */
583 if (!IS_ERR_OR_NULL(check_root)) {
584 struct extent_buffer *eb;
585
586 eb = btrfs_root_node(check_root);
587 /* if leaf is the root, then it's fine */
588 if (leaf != eb) {
589 CORRUPT("non-root leaf's nritems is 0",
590 leaf, check_root, 0);
591 free_extent_buffer(eb);
592 return -EIO;
593 }
594 free_extent_buffer(eb);
595 }
596 return 0;
597 }
598
599 if (nritems == 0)
600 return 0;
601
602 /* Check the 0 item */
603 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
604 BTRFS_LEAF_DATA_SIZE(fs_info)) {
605 CORRUPT("invalid item offset size pair", leaf, root, 0);
606 return -EIO;
607 }
608
609 /*
610 * Check to make sure each items keys are in the correct order and their
611 * offsets make sense. We only have to loop through nritems-1 because
612 * we check the current slot against the next slot, which verifies the
613 * next slot's offset+size makes sense and that the current's slot
614 * offset is correct.
615 */
616 for (slot = 0; slot < nritems - 1; slot++) {
617 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
618 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
619
620 /* Make sure the keys are in the right order */
621 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
622 CORRUPT("bad key order", leaf, root, slot);
623 return -EIO;
624 }
625
626 /*
627 * Make sure the offset and ends are right, remember that the
628 * item data starts at the end of the leaf and grows towards the
629 * front.
630 */
631 if (btrfs_item_offset_nr(leaf, slot) !=
632 btrfs_item_end_nr(leaf, slot + 1)) {
633 CORRUPT("slot offset bad", leaf, root, slot);
634 return -EIO;
635 }
636
637 /*
638 * Check to make sure that we don't point outside of the leaf,
639 * just in case all the items are consistent to each other, but
640 * all point outside of the leaf.
641 */
642 if (btrfs_item_end_nr(leaf, slot) >
643 BTRFS_LEAF_DATA_SIZE(fs_info)) {
644 CORRUPT("slot end outside of leaf", leaf, root, slot);
645 return -EIO;
646 }
647 }
648
649 return 0;
650}
651
652static int check_node(struct btrfs_root *root, struct extent_buffer *node)
653{
654 unsigned long nr = btrfs_header_nritems(node);
655 struct btrfs_key key, next_key;
656 int slot;
657 u64 bytenr;
658 int ret = 0;
659
660 if (nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(root->fs_info)) {
661 btrfs_crit(root->fs_info,
662 "corrupt node: block %llu root %llu nritems %lu",
663 node->start, root->objectid, nr);
664 return -EIO;
665 }
666
667 for (slot = 0; slot < nr - 1; slot++) {
668 bytenr = btrfs_node_blockptr(node, slot);
669 btrfs_node_key_to_cpu(node, &key, slot);
670 btrfs_node_key_to_cpu(node, &next_key, slot + 1);
671
672 if (!bytenr) {
673 CORRUPT("invalid item slot", node, root, slot);
674 ret = -EIO;
675 goto out;
676 }
677
678 if (btrfs_comp_cpu_keys(&key, &next_key) >= 0) {
679 CORRUPT("bad key order", node, root, slot);
680 ret = -EIO;
681 goto out;
682 }
683 }
684out:
685 return ret;
686}
687
688static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
689 u64 phy_offset, struct page *page,
690 u64 start, u64 end, int mirror)
691{
692 u64 found_start;
693 int found_level;
694 struct extent_buffer *eb;
695 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
696 struct btrfs_fs_info *fs_info = root->fs_info;
697 int ret = 0;
698 int reads_done;
699
700 if (!page->private)
701 goto out;
702
703 eb = (struct extent_buffer *)page->private;
704
705 /* the pending IO might have been the only thing that kept this buffer
706 * in memory. Make sure we have a ref for all this other checks
707 */
708 extent_buffer_get(eb);
709
710 reads_done = atomic_dec_and_test(&eb->io_pages);
711 if (!reads_done)
712 goto err;
713
714 eb->read_mirror = mirror;
715 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
716 ret = -EIO;
717 goto err;
718 }
719
720 found_start = btrfs_header_bytenr(eb);
721 if (found_start != eb->start) {
722 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
723 found_start, eb->start);
724 ret = -EIO;
725 goto err;
726 }
727 if (check_tree_block_fsid(fs_info, eb)) {
728 btrfs_err_rl(fs_info, "bad fsid on block %llu",
729 eb->start);
730 ret = -EIO;
731 goto err;
732 }
733 found_level = btrfs_header_level(eb);
734 if (found_level >= BTRFS_MAX_LEVEL) {
735 btrfs_err(fs_info, "bad tree block level %d",
736 (int)btrfs_header_level(eb));
737 ret = -EIO;
738 goto err;
739 }
740
741 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
742 eb, found_level);
743
744 ret = csum_tree_block(fs_info, eb, 1);
745 if (ret)
746 goto err;
747
748 /*
749 * If this is a leaf block and it is corrupt, set the corrupt bit so
750 * that we don't try and read the other copies of this block, just
751 * return -EIO.
752 */
753 if (found_level == 0 && check_leaf(root, eb)) {
754 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
755 ret = -EIO;
756 }
757
758 if (found_level > 0 && check_node(root, eb))
759 ret = -EIO;
760
761 if (!ret)
762 set_extent_buffer_uptodate(eb);
763err:
764 if (reads_done &&
765 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
766 btree_readahead_hook(fs_info, eb, ret);
767
768 if (ret) {
769 /*
770 * our io error hook is going to dec the io pages
771 * again, we have to make sure it has something
772 * to decrement
773 */
774 atomic_inc(&eb->io_pages);
775 clear_extent_buffer_uptodate(eb);
776 }
777 free_extent_buffer(eb);
778out:
779 return ret;
780}
781
782static int btree_io_failed_hook(struct page *page, int failed_mirror)
783{
784 struct extent_buffer *eb;
785
786 eb = (struct extent_buffer *)page->private;
787 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
788 eb->read_mirror = failed_mirror;
789 atomic_dec(&eb->io_pages);
790 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
791 btree_readahead_hook(eb->fs_info, eb, -EIO);
792 return -EIO; /* we fixed nothing */
793}
794
795static void end_workqueue_bio(struct bio *bio)
796{
797 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
798 struct btrfs_fs_info *fs_info;
799 struct btrfs_workqueue *wq;
800 btrfs_work_func_t func;
801
802 fs_info = end_io_wq->info;
803 end_io_wq->error = bio->bi_error;
804
805 if (bio_op(bio) == REQ_OP_WRITE) {
806 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
807 wq = fs_info->endio_meta_write_workers;
808 func = btrfs_endio_meta_write_helper;
809 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
810 wq = fs_info->endio_freespace_worker;
811 func = btrfs_freespace_write_helper;
812 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
813 wq = fs_info->endio_raid56_workers;
814 func = btrfs_endio_raid56_helper;
815 } else {
816 wq = fs_info->endio_write_workers;
817 func = btrfs_endio_write_helper;
818 }
819 } else {
820 if (unlikely(end_io_wq->metadata ==
821 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
822 wq = fs_info->endio_repair_workers;
823 func = btrfs_endio_repair_helper;
824 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
825 wq = fs_info->endio_raid56_workers;
826 func = btrfs_endio_raid56_helper;
827 } else if (end_io_wq->metadata) {
828 wq = fs_info->endio_meta_workers;
829 func = btrfs_endio_meta_helper;
830 } else {
831 wq = fs_info->endio_workers;
832 func = btrfs_endio_helper;
833 }
834 }
835
836 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
837 btrfs_queue_work(wq, &end_io_wq->work);
838}
839
840int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
841 enum btrfs_wq_endio_type metadata)
842{
843 struct btrfs_end_io_wq *end_io_wq;
844
845 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
846 if (!end_io_wq)
847 return -ENOMEM;
848
849 end_io_wq->private = bio->bi_private;
850 end_io_wq->end_io = bio->bi_end_io;
851 end_io_wq->info = info;
852 end_io_wq->error = 0;
853 end_io_wq->bio = bio;
854 end_io_wq->metadata = metadata;
855
856 bio->bi_private = end_io_wq;
857 bio->bi_end_io = end_workqueue_bio;
858 return 0;
859}
860
861unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
862{
863 unsigned long limit = min_t(unsigned long,
864 info->thread_pool_size,
865 info->fs_devices->open_devices);
866 return 256 * limit;
867}
868
869static void run_one_async_start(struct btrfs_work *work)
870{
871 struct async_submit_bio *async;
872 int ret;
873
874 async = container_of(work, struct async_submit_bio, work);
875 ret = async->submit_bio_start(async->inode, async->bio,
876 async->mirror_num, async->bio_flags,
877 async->bio_offset);
878 if (ret)
879 async->error = ret;
880}
881
882static void run_one_async_done(struct btrfs_work *work)
883{
884 struct btrfs_fs_info *fs_info;
885 struct async_submit_bio *async;
886 int limit;
887
888 async = container_of(work, struct async_submit_bio, work);
889 fs_info = BTRFS_I(async->inode)->root->fs_info;
890
891 limit = btrfs_async_submit_limit(fs_info);
892 limit = limit * 2 / 3;
893
894 /*
895 * atomic_dec_return implies a barrier for waitqueue_active
896 */
897 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
898 waitqueue_active(&fs_info->async_submit_wait))
899 wake_up(&fs_info->async_submit_wait);
900
901 /* If an error occurred we just want to clean up the bio and move on */
902 if (async->error) {
903 async->bio->bi_error = async->error;
904 bio_endio(async->bio);
905 return;
906 }
907
908 async->submit_bio_done(async->inode, async->bio, async->mirror_num,
909 async->bio_flags, async->bio_offset);
910}
911
912static void run_one_async_free(struct btrfs_work *work)
913{
914 struct async_submit_bio *async;
915
916 async = container_of(work, struct async_submit_bio, work);
917 kfree(async);
918}
919
920int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
921 struct bio *bio, int mirror_num,
922 unsigned long bio_flags,
923 u64 bio_offset,
924 extent_submit_bio_hook_t *submit_bio_start,
925 extent_submit_bio_hook_t *submit_bio_done)
926{
927 struct async_submit_bio *async;
928
929 async = kmalloc(sizeof(*async), GFP_NOFS);
930 if (!async)
931 return -ENOMEM;
932
933 async->inode = inode;
934 async->bio = bio;
935 async->mirror_num = mirror_num;
936 async->submit_bio_start = submit_bio_start;
937 async->submit_bio_done = submit_bio_done;
938
939 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
940 run_one_async_done, run_one_async_free);
941
942 async->bio_flags = bio_flags;
943 async->bio_offset = bio_offset;
944
945 async->error = 0;
946
947 atomic_inc(&fs_info->nr_async_submits);
948
949 if (op_is_sync(bio->bi_opf))
950 btrfs_set_work_high_priority(&async->work);
951
952 btrfs_queue_work(fs_info->workers, &async->work);
953
954 while (atomic_read(&fs_info->async_submit_draining) &&
955 atomic_read(&fs_info->nr_async_submits)) {
956 wait_event(fs_info->async_submit_wait,
957 (atomic_read(&fs_info->nr_async_submits) == 0));
958 }
959
960 return 0;
961}
962
963static int btree_csum_one_bio(struct bio *bio)
964{
965 struct bio_vec *bvec;
966 struct btrfs_root *root;
967 int i, ret = 0;
968
969 bio_for_each_segment_all(bvec, bio, i) {
970 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
971 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
972 if (ret)
973 break;
974 }
975
976 return ret;
977}
978
979static int __btree_submit_bio_start(struct inode *inode, struct bio *bio,
980 int mirror_num, unsigned long bio_flags,
981 u64 bio_offset)
982{
983 /*
984 * when we're called for a write, we're already in the async
985 * submission context. Just jump into btrfs_map_bio
986 */
987 return btree_csum_one_bio(bio);
988}
989
990static int __btree_submit_bio_done(struct inode *inode, struct bio *bio,
991 int mirror_num, unsigned long bio_flags,
992 u64 bio_offset)
993{
994 int ret;
995
996 /*
997 * when we're called for a write, we're already in the async
998 * submission context. Just jump into btrfs_map_bio
999 */
1000 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
1001 if (ret) {
1002 bio->bi_error = ret;
1003 bio_endio(bio);
1004 }
1005 return ret;
1006}
1007
1008static int check_async_write(struct inode *inode, unsigned long bio_flags)
1009{
1010 if (bio_flags & EXTENT_BIO_TREE_LOG)
1011 return 0;
1012#ifdef CONFIG_X86
1013 if (static_cpu_has(X86_FEATURE_XMM4_2))
1014 return 0;
1015#endif
1016 return 1;
1017}
1018
1019static int btree_submit_bio_hook(struct inode *inode, struct bio *bio,
1020 int mirror_num, unsigned long bio_flags,
1021 u64 bio_offset)
1022{
1023 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1024 int async = check_async_write(inode, bio_flags);
1025 int ret;
1026
1027 if (bio_op(bio) != REQ_OP_WRITE) {
1028 /*
1029 * called for a read, do the setup so that checksum validation
1030 * can happen in the async kernel threads
1031 */
1032 ret = btrfs_bio_wq_end_io(fs_info, bio,
1033 BTRFS_WQ_ENDIO_METADATA);
1034 if (ret)
1035 goto out_w_error;
1036 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1037 } else if (!async) {
1038 ret = btree_csum_one_bio(bio);
1039 if (ret)
1040 goto out_w_error;
1041 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1042 } else {
1043 /*
1044 * kthread helpers are used to submit writes so that
1045 * checksumming can happen in parallel across all CPUs
1046 */
1047 ret = btrfs_wq_submit_bio(fs_info, inode, bio, mirror_num, 0,
1048 bio_offset,
1049 __btree_submit_bio_start,
1050 __btree_submit_bio_done);
1051 }
1052
1053 if (ret)
1054 goto out_w_error;
1055 return 0;
1056
1057out_w_error:
1058 bio->bi_error = ret;
1059 bio_endio(bio);
1060 return ret;
1061}
1062
1063#ifdef CONFIG_MIGRATION
1064static int btree_migratepage(struct address_space *mapping,
1065 struct page *newpage, struct page *page,
1066 enum migrate_mode mode)
1067{
1068 /*
1069 * we can't safely write a btree page from here,
1070 * we haven't done the locking hook
1071 */
1072 if (PageDirty(page))
1073 return -EAGAIN;
1074 /*
1075 * Buffers may be managed in a filesystem specific way.
1076 * We must have no buffers or drop them.
1077 */
1078 if (page_has_private(page) &&
1079 !try_to_release_page(page, GFP_KERNEL))
1080 return -EAGAIN;
1081 return migrate_page(mapping, newpage, page, mode);
1082}
1083#endif
1084
1085
1086static int btree_writepages(struct address_space *mapping,
1087 struct writeback_control *wbc)
1088{
1089 struct btrfs_fs_info *fs_info;
1090 int ret;
1091
1092 if (wbc->sync_mode == WB_SYNC_NONE) {
1093
1094 if (wbc->for_kupdate)
1095 return 0;
1096
1097 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1098 /* this is a bit racy, but that's ok */
1099 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1100 BTRFS_DIRTY_METADATA_THRESH);
1101 if (ret < 0)
1102 return 0;
1103 }
1104 return btree_write_cache_pages(mapping, wbc);
1105}
1106
1107static int btree_readpage(struct file *file, struct page *page)
1108{
1109 struct extent_io_tree *tree;
1110 tree = &BTRFS_I(page->mapping->host)->io_tree;
1111 return extent_read_full_page(tree, page, btree_get_extent, 0);
1112}
1113
1114static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1115{
1116 if (PageWriteback(page) || PageDirty(page))
1117 return 0;
1118
1119 return try_release_extent_buffer(page);
1120}
1121
1122static void btree_invalidatepage(struct page *page, unsigned int offset,
1123 unsigned int length)
1124{
1125 struct extent_io_tree *tree;
1126 tree = &BTRFS_I(page->mapping->host)->io_tree;
1127 extent_invalidatepage(tree, page, offset);
1128 btree_releasepage(page, GFP_NOFS);
1129 if (PagePrivate(page)) {
1130 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1131 "page private not zero on page %llu",
1132 (unsigned long long)page_offset(page));
1133 ClearPagePrivate(page);
1134 set_page_private(page, 0);
1135 put_page(page);
1136 }
1137}
1138
1139static int btree_set_page_dirty(struct page *page)
1140{
1141#ifdef DEBUG
1142 struct extent_buffer *eb;
1143
1144 BUG_ON(!PagePrivate(page));
1145 eb = (struct extent_buffer *)page->private;
1146 BUG_ON(!eb);
1147 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1148 BUG_ON(!atomic_read(&eb->refs));
1149 btrfs_assert_tree_locked(eb);
1150#endif
1151 return __set_page_dirty_nobuffers(page);
1152}
1153
1154static const struct address_space_operations btree_aops = {
1155 .readpage = btree_readpage,
1156 .writepages = btree_writepages,
1157 .releasepage = btree_releasepage,
1158 .invalidatepage = btree_invalidatepage,
1159#ifdef CONFIG_MIGRATION
1160 .migratepage = btree_migratepage,
1161#endif
1162 .set_page_dirty = btree_set_page_dirty,
1163};
1164
1165void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1166{
1167 struct extent_buffer *buf = NULL;
1168 struct inode *btree_inode = fs_info->btree_inode;
1169
1170 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1171 if (IS_ERR(buf))
1172 return;
1173 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1174 buf, WAIT_NONE, btree_get_extent, 0);
1175 free_extent_buffer(buf);
1176}
1177
1178int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1179 int mirror_num, struct extent_buffer **eb)
1180{
1181 struct extent_buffer *buf = NULL;
1182 struct inode *btree_inode = fs_info->btree_inode;
1183 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1184 int ret;
1185
1186 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1187 if (IS_ERR(buf))
1188 return 0;
1189
1190 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1191
1192 ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1193 btree_get_extent, mirror_num);
1194 if (ret) {
1195 free_extent_buffer(buf);
1196 return ret;
1197 }
1198
1199 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1200 free_extent_buffer(buf);
1201 return -EIO;
1202 } else if (extent_buffer_uptodate(buf)) {
1203 *eb = buf;
1204 } else {
1205 free_extent_buffer(buf);
1206 }
1207 return 0;
1208}
1209
1210struct extent_buffer *btrfs_find_create_tree_block(
1211 struct btrfs_fs_info *fs_info,
1212 u64 bytenr)
1213{
1214 if (btrfs_is_testing(fs_info))
1215 return alloc_test_extent_buffer(fs_info, bytenr);
1216 return alloc_extent_buffer(fs_info, bytenr);
1217}
1218
1219
1220int btrfs_write_tree_block(struct extent_buffer *buf)
1221{
1222 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1223 buf->start + buf->len - 1);
1224}
1225
1226int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1227{
1228 return filemap_fdatawait_range(buf->pages[0]->mapping,
1229 buf->start, buf->start + buf->len - 1);
1230}
1231
1232struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1233 u64 parent_transid)
1234{
1235 struct extent_buffer *buf = NULL;
1236 int ret;
1237
1238 buf = btrfs_find_create_tree_block(fs_info, bytenr);
1239 if (IS_ERR(buf))
1240 return buf;
1241
1242 ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
1243 if (ret) {
1244 free_extent_buffer(buf);
1245 return ERR_PTR(ret);
1246 }
1247 return buf;
1248
1249}
1250
1251void clean_tree_block(struct btrfs_trans_handle *trans,
1252 struct btrfs_fs_info *fs_info,
1253 struct extent_buffer *buf)
1254{
1255 if (btrfs_header_generation(buf) ==
1256 fs_info->running_transaction->transid) {
1257 btrfs_assert_tree_locked(buf);
1258
1259 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1260 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1261 -buf->len,
1262 fs_info->dirty_metadata_batch);
1263 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1264 btrfs_set_lock_blocking(buf);
1265 clear_extent_buffer_dirty(buf);
1266 }
1267 }
1268}
1269
1270static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1271{
1272 struct btrfs_subvolume_writers *writers;
1273 int ret;
1274
1275 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1276 if (!writers)
1277 return ERR_PTR(-ENOMEM);
1278
1279 ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1280 if (ret < 0) {
1281 kfree(writers);
1282 return ERR_PTR(ret);
1283 }
1284
1285 init_waitqueue_head(&writers->wait);
1286 return writers;
1287}
1288
1289static void
1290btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1291{
1292 percpu_counter_destroy(&writers->counter);
1293 kfree(writers);
1294}
1295
1296static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1297 u64 objectid)
1298{
1299 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1300 root->node = NULL;
1301 root->commit_root = NULL;
1302 root->state = 0;
1303 root->orphan_cleanup_state = 0;
1304
1305 root->objectid = objectid;
1306 root->last_trans = 0;
1307 root->highest_objectid = 0;
1308 root->nr_delalloc_inodes = 0;
1309 root->nr_ordered_extents = 0;
1310 root->name = NULL;
1311 root->inode_tree = RB_ROOT;
1312 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1313 root->block_rsv = NULL;
1314 root->orphan_block_rsv = NULL;
1315
1316 INIT_LIST_HEAD(&root->dirty_list);
1317 INIT_LIST_HEAD(&root->root_list);
1318 INIT_LIST_HEAD(&root->delalloc_inodes);
1319 INIT_LIST_HEAD(&root->delalloc_root);
1320 INIT_LIST_HEAD(&root->ordered_extents);
1321 INIT_LIST_HEAD(&root->ordered_root);
1322 INIT_LIST_HEAD(&root->logged_list[0]);
1323 INIT_LIST_HEAD(&root->logged_list[1]);
1324 spin_lock_init(&root->orphan_lock);
1325 spin_lock_init(&root->inode_lock);
1326 spin_lock_init(&root->delalloc_lock);
1327 spin_lock_init(&root->ordered_extent_lock);
1328 spin_lock_init(&root->accounting_lock);
1329 spin_lock_init(&root->log_extents_lock[0]);
1330 spin_lock_init(&root->log_extents_lock[1]);
1331 mutex_init(&root->objectid_mutex);
1332 mutex_init(&root->log_mutex);
1333 mutex_init(&root->ordered_extent_mutex);
1334 mutex_init(&root->delalloc_mutex);
1335 init_waitqueue_head(&root->log_writer_wait);
1336 init_waitqueue_head(&root->log_commit_wait[0]);
1337 init_waitqueue_head(&root->log_commit_wait[1]);
1338 INIT_LIST_HEAD(&root->log_ctxs[0]);
1339 INIT_LIST_HEAD(&root->log_ctxs[1]);
1340 atomic_set(&root->log_commit[0], 0);
1341 atomic_set(&root->log_commit[1], 0);
1342 atomic_set(&root->log_writers, 0);
1343 atomic_set(&root->log_batch, 0);
1344 atomic_set(&root->orphan_inodes, 0);
1345 atomic_set(&root->refs, 1);
1346 atomic_set(&root->will_be_snapshoted, 0);
1347 atomic_set(&root->qgroup_meta_rsv, 0);
1348 root->log_transid = 0;
1349 root->log_transid_committed = -1;
1350 root->last_log_commit = 0;
1351 if (!dummy)
1352 extent_io_tree_init(&root->dirty_log_pages,
1353 fs_info->btree_inode->i_mapping);
1354
1355 memset(&root->root_key, 0, sizeof(root->root_key));
1356 memset(&root->root_item, 0, sizeof(root->root_item));
1357 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1358 if (!dummy)
1359 root->defrag_trans_start = fs_info->generation;
1360 else
1361 root->defrag_trans_start = 0;
1362 root->root_key.objectid = objectid;
1363 root->anon_dev = 0;
1364
1365 spin_lock_init(&root->root_item_lock);
1366}
1367
1368static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1369 gfp_t flags)
1370{
1371 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1372 if (root)
1373 root->fs_info = fs_info;
1374 return root;
1375}
1376
1377#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1378/* Should only be used by the testing infrastructure */
1379struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1380{
1381 struct btrfs_root *root;
1382
1383 if (!fs_info)
1384 return ERR_PTR(-EINVAL);
1385
1386 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1387 if (!root)
1388 return ERR_PTR(-ENOMEM);
1389
1390 /* We don't use the stripesize in selftest, set it as sectorsize */
1391 __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1392 root->alloc_bytenr = 0;
1393
1394 return root;
1395}
1396#endif
1397
1398struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1399 struct btrfs_fs_info *fs_info,
1400 u64 objectid)
1401{
1402 struct extent_buffer *leaf;
1403 struct btrfs_root *tree_root = fs_info->tree_root;
1404 struct btrfs_root *root;
1405 struct btrfs_key key;
1406 int ret = 0;
1407 uuid_le uuid;
1408
1409 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1410 if (!root)
1411 return ERR_PTR(-ENOMEM);
1412
1413 __setup_root(root, fs_info, objectid);
1414 root->root_key.objectid = objectid;
1415 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1416 root->root_key.offset = 0;
1417
1418 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1419 if (IS_ERR(leaf)) {
1420 ret = PTR_ERR(leaf);
1421 leaf = NULL;
1422 goto fail;
1423 }
1424
1425 memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1426 btrfs_set_header_bytenr(leaf, leaf->start);
1427 btrfs_set_header_generation(leaf, trans->transid);
1428 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1429 btrfs_set_header_owner(leaf, objectid);
1430 root->node = leaf;
1431
1432 write_extent_buffer_fsid(leaf, fs_info->fsid);
1433 write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1434 btrfs_mark_buffer_dirty(leaf);
1435
1436 root->commit_root = btrfs_root_node(root);
1437 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1438
1439 root->root_item.flags = 0;
1440 root->root_item.byte_limit = 0;
1441 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1442 btrfs_set_root_generation(&root->root_item, trans->transid);
1443 btrfs_set_root_level(&root->root_item, 0);
1444 btrfs_set_root_refs(&root->root_item, 1);
1445 btrfs_set_root_used(&root->root_item, leaf->len);
1446 btrfs_set_root_last_snapshot(&root->root_item, 0);
1447 btrfs_set_root_dirid(&root->root_item, 0);
1448 uuid_le_gen(&uuid);
1449 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1450 root->root_item.drop_level = 0;
1451
1452 key.objectid = objectid;
1453 key.type = BTRFS_ROOT_ITEM_KEY;
1454 key.offset = 0;
1455 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1456 if (ret)
1457 goto fail;
1458
1459 btrfs_tree_unlock(leaf);
1460
1461 return root;
1462
1463fail:
1464 if (leaf) {
1465 btrfs_tree_unlock(leaf);
1466 free_extent_buffer(root->commit_root);
1467 free_extent_buffer(leaf);
1468 }
1469 kfree(root);
1470
1471 return ERR_PTR(ret);
1472}
1473
1474static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1475 struct btrfs_fs_info *fs_info)
1476{
1477 struct btrfs_root *root;
1478 struct extent_buffer *leaf;
1479
1480 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1481 if (!root)
1482 return ERR_PTR(-ENOMEM);
1483
1484 __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1485
1486 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1487 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1488 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1489
1490 /*
1491 * DON'T set REF_COWS for log trees
1492 *
1493 * log trees do not get reference counted because they go away
1494 * before a real commit is actually done. They do store pointers
1495 * to file data extents, and those reference counts still get
1496 * updated (along with back refs to the log tree).
1497 */
1498
1499 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1500 NULL, 0, 0, 0);
1501 if (IS_ERR(leaf)) {
1502 kfree(root);
1503 return ERR_CAST(leaf);
1504 }
1505
1506 memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1507 btrfs_set_header_bytenr(leaf, leaf->start);
1508 btrfs_set_header_generation(leaf, trans->transid);
1509 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1510 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1511 root->node = leaf;
1512
1513 write_extent_buffer_fsid(root->node, fs_info->fsid);
1514 btrfs_mark_buffer_dirty(root->node);
1515 btrfs_tree_unlock(root->node);
1516 return root;
1517}
1518
1519int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1520 struct btrfs_fs_info *fs_info)
1521{
1522 struct btrfs_root *log_root;
1523
1524 log_root = alloc_log_tree(trans, fs_info);
1525 if (IS_ERR(log_root))
1526 return PTR_ERR(log_root);
1527 WARN_ON(fs_info->log_root_tree);
1528 fs_info->log_root_tree = log_root;
1529 return 0;
1530}
1531
1532int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1533 struct btrfs_root *root)
1534{
1535 struct btrfs_fs_info *fs_info = root->fs_info;
1536 struct btrfs_root *log_root;
1537 struct btrfs_inode_item *inode_item;
1538
1539 log_root = alloc_log_tree(trans, fs_info);
1540 if (IS_ERR(log_root))
1541 return PTR_ERR(log_root);
1542
1543 log_root->last_trans = trans->transid;
1544 log_root->root_key.offset = root->root_key.objectid;
1545
1546 inode_item = &log_root->root_item.inode;
1547 btrfs_set_stack_inode_generation(inode_item, 1);
1548 btrfs_set_stack_inode_size(inode_item, 3);
1549 btrfs_set_stack_inode_nlink(inode_item, 1);
1550 btrfs_set_stack_inode_nbytes(inode_item,
1551 fs_info->nodesize);
1552 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1553
1554 btrfs_set_root_node(&log_root->root_item, log_root->node);
1555
1556 WARN_ON(root->log_root);
1557 root->log_root = log_root;
1558 root->log_transid = 0;
1559 root->log_transid_committed = -1;
1560 root->last_log_commit = 0;
1561 return 0;
1562}
1563
1564static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1565 struct btrfs_key *key)
1566{
1567 struct btrfs_root *root;
1568 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1569 struct btrfs_path *path;
1570 u64 generation;
1571 int ret;
1572
1573 path = btrfs_alloc_path();
1574 if (!path)
1575 return ERR_PTR(-ENOMEM);
1576
1577 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1578 if (!root) {
1579 ret = -ENOMEM;
1580 goto alloc_fail;
1581 }
1582
1583 __setup_root(root, fs_info, key->objectid);
1584
1585 ret = btrfs_find_root(tree_root, key, path,
1586 &root->root_item, &root->root_key);
1587 if (ret) {
1588 if (ret > 0)
1589 ret = -ENOENT;
1590 goto find_fail;
1591 }
1592
1593 generation = btrfs_root_generation(&root->root_item);
1594 root->node = read_tree_block(fs_info,
1595 btrfs_root_bytenr(&root->root_item),
1596 generation);
1597 if (IS_ERR(root->node)) {
1598 ret = PTR_ERR(root->node);
1599 goto find_fail;
1600 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1601 ret = -EIO;
1602 free_extent_buffer(root->node);
1603 goto find_fail;
1604 }
1605 root->commit_root = btrfs_root_node(root);
1606out:
1607 btrfs_free_path(path);
1608 return root;
1609
1610find_fail:
1611 kfree(root);
1612alloc_fail:
1613 root = ERR_PTR(ret);
1614 goto out;
1615}
1616
1617struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1618 struct btrfs_key *location)
1619{
1620 struct btrfs_root *root;
1621
1622 root = btrfs_read_tree_root(tree_root, location);
1623 if (IS_ERR(root))
1624 return root;
1625
1626 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1627 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1628 btrfs_check_and_init_root_item(&root->root_item);
1629 }
1630
1631 return root;
1632}
1633
1634int btrfs_init_fs_root(struct btrfs_root *root)
1635{
1636 int ret;
1637 struct btrfs_subvolume_writers *writers;
1638
1639 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1640 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1641 GFP_NOFS);
1642 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1643 ret = -ENOMEM;
1644 goto fail;
1645 }
1646
1647 writers = btrfs_alloc_subvolume_writers();
1648 if (IS_ERR(writers)) {
1649 ret = PTR_ERR(writers);
1650 goto fail;
1651 }
1652 root->subv_writers = writers;
1653
1654 btrfs_init_free_ino_ctl(root);
1655 spin_lock_init(&root->ino_cache_lock);
1656 init_waitqueue_head(&root->ino_cache_wait);
1657
1658 ret = get_anon_bdev(&root->anon_dev);
1659 if (ret)
1660 goto fail;
1661
1662 mutex_lock(&root->objectid_mutex);
1663 ret = btrfs_find_highest_objectid(root,
1664 &root->highest_objectid);
1665 if (ret) {
1666 mutex_unlock(&root->objectid_mutex);
1667 goto fail;
1668 }
1669
1670 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1671
1672 mutex_unlock(&root->objectid_mutex);
1673
1674 return 0;
1675fail:
1676 /* the caller is responsible to call free_fs_root */
1677 return ret;
1678}
1679
1680struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1681 u64 root_id)
1682{
1683 struct btrfs_root *root;
1684
1685 spin_lock(&fs_info->fs_roots_radix_lock);
1686 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1687 (unsigned long)root_id);
1688 spin_unlock(&fs_info->fs_roots_radix_lock);
1689 return root;
1690}
1691
1692int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1693 struct btrfs_root *root)
1694{
1695 int ret;
1696
1697 ret = radix_tree_preload(GFP_NOFS);
1698 if (ret)
1699 return ret;
1700
1701 spin_lock(&fs_info->fs_roots_radix_lock);
1702 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1703 (unsigned long)root->root_key.objectid,
1704 root);
1705 if (ret == 0)
1706 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1707 spin_unlock(&fs_info->fs_roots_radix_lock);
1708 radix_tree_preload_end();
1709
1710 return ret;
1711}
1712
1713struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1714 struct btrfs_key *location,
1715 bool check_ref)
1716{
1717 struct btrfs_root *root;
1718 struct btrfs_path *path;
1719 struct btrfs_key key;
1720 int ret;
1721
1722 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1723 return fs_info->tree_root;
1724 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1725 return fs_info->extent_root;
1726 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1727 return fs_info->chunk_root;
1728 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1729 return fs_info->dev_root;
1730 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1731 return fs_info->csum_root;
1732 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1733 return fs_info->quota_root ? fs_info->quota_root :
1734 ERR_PTR(-ENOENT);
1735 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1736 return fs_info->uuid_root ? fs_info->uuid_root :
1737 ERR_PTR(-ENOENT);
1738 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1739 return fs_info->free_space_root ? fs_info->free_space_root :
1740 ERR_PTR(-ENOENT);
1741again:
1742 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1743 if (root) {
1744 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1745 return ERR_PTR(-ENOENT);
1746 return root;
1747 }
1748
1749 root = btrfs_read_fs_root(fs_info->tree_root, location);
1750 if (IS_ERR(root))
1751 return root;
1752
1753 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1754 ret = -ENOENT;
1755 goto fail;
1756 }
1757
1758 ret = btrfs_init_fs_root(root);
1759 if (ret)
1760 goto fail;
1761
1762 path = btrfs_alloc_path();
1763 if (!path) {
1764 ret = -ENOMEM;
1765 goto fail;
1766 }
1767 key.objectid = BTRFS_ORPHAN_OBJECTID;
1768 key.type = BTRFS_ORPHAN_ITEM_KEY;
1769 key.offset = location->objectid;
1770
1771 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1772 btrfs_free_path(path);
1773 if (ret < 0)
1774 goto fail;
1775 if (ret == 0)
1776 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1777
1778 ret = btrfs_insert_fs_root(fs_info, root);
1779 if (ret) {
1780 if (ret == -EEXIST) {
1781 free_fs_root(root);
1782 goto again;
1783 }
1784 goto fail;
1785 }
1786 return root;
1787fail:
1788 free_fs_root(root);
1789 return ERR_PTR(ret);
1790}
1791
1792static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1793{
1794 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1795 int ret = 0;
1796 struct btrfs_device *device;
1797 struct backing_dev_info *bdi;
1798
1799 rcu_read_lock();
1800 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1801 if (!device->bdev)
1802 continue;
1803 bdi = blk_get_backing_dev_info(device->bdev);
1804 if (bdi_congested(bdi, bdi_bits)) {
1805 ret = 1;
1806 break;
1807 }
1808 }
1809 rcu_read_unlock();
1810 return ret;
1811}
1812
1813static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1814{
1815 int err;
1816
1817 err = bdi_setup_and_register(bdi, "btrfs");
1818 if (err)
1819 return err;
1820
1821 bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
1822 bdi->congested_fn = btrfs_congested_fn;
1823 bdi->congested_data = info;
1824 bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1825 return 0;
1826}
1827
1828/*
1829 * called by the kthread helper functions to finally call the bio end_io
1830 * functions. This is where read checksum verification actually happens
1831 */
1832static void end_workqueue_fn(struct btrfs_work *work)
1833{
1834 struct bio *bio;
1835 struct btrfs_end_io_wq *end_io_wq;
1836
1837 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1838 bio = end_io_wq->bio;
1839
1840 bio->bi_error = end_io_wq->error;
1841 bio->bi_private = end_io_wq->private;
1842 bio->bi_end_io = end_io_wq->end_io;
1843 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1844 bio_endio(bio);
1845}
1846
1847static int cleaner_kthread(void *arg)
1848{
1849 struct btrfs_root *root = arg;
1850 struct btrfs_fs_info *fs_info = root->fs_info;
1851 int again;
1852 struct btrfs_trans_handle *trans;
1853
1854 do {
1855 again = 0;
1856
1857 /* Make the cleaner go to sleep early. */
1858 if (btrfs_need_cleaner_sleep(fs_info))
1859 goto sleep;
1860
1861 /*
1862 * Do not do anything if we might cause open_ctree() to block
1863 * before we have finished mounting the filesystem.
1864 */
1865 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1866 goto sleep;
1867
1868 if (!mutex_trylock(&fs_info->cleaner_mutex))
1869 goto sleep;
1870
1871 /*
1872 * Avoid the problem that we change the status of the fs
1873 * during the above check and trylock.
1874 */
1875 if (btrfs_need_cleaner_sleep(fs_info)) {
1876 mutex_unlock(&fs_info->cleaner_mutex);
1877 goto sleep;
1878 }
1879
1880 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1881 btrfs_run_delayed_iputs(fs_info);
1882 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1883
1884 again = btrfs_clean_one_deleted_snapshot(root);
1885 mutex_unlock(&fs_info->cleaner_mutex);
1886
1887 /*
1888 * The defragger has dealt with the R/O remount and umount,
1889 * needn't do anything special here.
1890 */
1891 btrfs_run_defrag_inodes(fs_info);
1892
1893 /*
1894 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1895 * with relocation (btrfs_relocate_chunk) and relocation
1896 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1897 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1898 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1899 * unused block groups.
1900 */
1901 btrfs_delete_unused_bgs(fs_info);
1902sleep:
1903 if (!again) {
1904 set_current_state(TASK_INTERRUPTIBLE);
1905 if (!kthread_should_stop())
1906 schedule();
1907 __set_current_state(TASK_RUNNING);
1908 }
1909 } while (!kthread_should_stop());
1910
1911 /*
1912 * Transaction kthread is stopped before us and wakes us up.
1913 * However we might have started a new transaction and COWed some
1914 * tree blocks when deleting unused block groups for example. So
1915 * make sure we commit the transaction we started to have a clean
1916 * shutdown when evicting the btree inode - if it has dirty pages
1917 * when we do the final iput() on it, eviction will trigger a
1918 * writeback for it which will fail with null pointer dereferences
1919 * since work queues and other resources were already released and
1920 * destroyed by the time the iput/eviction/writeback is made.
1921 */
1922 trans = btrfs_attach_transaction(root);
1923 if (IS_ERR(trans)) {
1924 if (PTR_ERR(trans) != -ENOENT)
1925 btrfs_err(fs_info,
1926 "cleaner transaction attach returned %ld",
1927 PTR_ERR(trans));
1928 } else {
1929 int ret;
1930
1931 ret = btrfs_commit_transaction(trans);
1932 if (ret)
1933 btrfs_err(fs_info,
1934 "cleaner open transaction commit returned %d",
1935 ret);
1936 }
1937
1938 return 0;
1939}
1940
1941static int transaction_kthread(void *arg)
1942{
1943 struct btrfs_root *root = arg;
1944 struct btrfs_fs_info *fs_info = root->fs_info;
1945 struct btrfs_trans_handle *trans;
1946 struct btrfs_transaction *cur;
1947 u64 transid;
1948 unsigned long now;
1949 unsigned long delay;
1950 bool cannot_commit;
1951
1952 do {
1953 cannot_commit = false;
1954 delay = HZ * fs_info->commit_interval;
1955 mutex_lock(&fs_info->transaction_kthread_mutex);
1956
1957 spin_lock(&fs_info->trans_lock);
1958 cur = fs_info->running_transaction;
1959 if (!cur) {
1960 spin_unlock(&fs_info->trans_lock);
1961 goto sleep;
1962 }
1963
1964 now = get_seconds();
1965 if (cur->state < TRANS_STATE_BLOCKED &&
1966 (now < cur->start_time ||
1967 now - cur->start_time < fs_info->commit_interval)) {
1968 spin_unlock(&fs_info->trans_lock);
1969 delay = HZ * 5;
1970 goto sleep;
1971 }
1972 transid = cur->transid;
1973 spin_unlock(&fs_info->trans_lock);
1974
1975 /* If the file system is aborted, this will always fail. */
1976 trans = btrfs_attach_transaction(root);
1977 if (IS_ERR(trans)) {
1978 if (PTR_ERR(trans) != -ENOENT)
1979 cannot_commit = true;
1980 goto sleep;
1981 }
1982 if (transid == trans->transid) {
1983 btrfs_commit_transaction(trans);
1984 } else {
1985 btrfs_end_transaction(trans);
1986 }
1987sleep:
1988 wake_up_process(fs_info->cleaner_kthread);
1989 mutex_unlock(&fs_info->transaction_kthread_mutex);
1990
1991 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1992 &fs_info->fs_state)))
1993 btrfs_cleanup_transaction(fs_info);
1994 set_current_state(TASK_INTERRUPTIBLE);
1995 if (!kthread_should_stop() &&
1996 (!btrfs_transaction_blocked(fs_info) ||
1997 cannot_commit))
1998 schedule_timeout(delay);
1999 __set_current_state(TASK_RUNNING);
2000 } while (!kthread_should_stop());
2001 return 0;
2002}
2003
2004/*
2005 * this will find the highest generation in the array of
2006 * root backups. The index of the highest array is returned,
2007 * or -1 if we can't find anything.
2008 *
2009 * We check to make sure the array is valid by comparing the
2010 * generation of the latest root in the array with the generation
2011 * in the super block. If they don't match we pitch it.
2012 */
2013static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
2014{
2015 u64 cur;
2016 int newest_index = -1;
2017 struct btrfs_root_backup *root_backup;
2018 int i;
2019
2020 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2021 root_backup = info->super_copy->super_roots + i;
2022 cur = btrfs_backup_tree_root_gen(root_backup);
2023 if (cur == newest_gen)
2024 newest_index = i;
2025 }
2026
2027 /* check to see if we actually wrapped around */
2028 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
2029 root_backup = info->super_copy->super_roots;
2030 cur = btrfs_backup_tree_root_gen(root_backup);
2031 if (cur == newest_gen)
2032 newest_index = 0;
2033 }
2034 return newest_index;
2035}
2036
2037
2038/*
2039 * find the oldest backup so we know where to store new entries
2040 * in the backup array. This will set the backup_root_index
2041 * field in the fs_info struct
2042 */
2043static void find_oldest_super_backup(struct btrfs_fs_info *info,
2044 u64 newest_gen)
2045{
2046 int newest_index = -1;
2047
2048 newest_index = find_newest_super_backup(info, newest_gen);
2049 /* if there was garbage in there, just move along */
2050 if (newest_index == -1) {
2051 info->backup_root_index = 0;
2052 } else {
2053 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
2054 }
2055}
2056
2057/*
2058 * copy all the root pointers into the super backup array.
2059 * this will bump the backup pointer by one when it is
2060 * done
2061 */
2062static void backup_super_roots(struct btrfs_fs_info *info)
2063{
2064 int next_backup;
2065 struct btrfs_root_backup *root_backup;
2066 int last_backup;
2067
2068 next_backup = info->backup_root_index;
2069 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2070 BTRFS_NUM_BACKUP_ROOTS;
2071
2072 /*
2073 * just overwrite the last backup if we're at the same generation
2074 * this happens only at umount
2075 */
2076 root_backup = info->super_for_commit->super_roots + last_backup;
2077 if (btrfs_backup_tree_root_gen(root_backup) ==
2078 btrfs_header_generation(info->tree_root->node))
2079 next_backup = last_backup;
2080
2081 root_backup = info->super_for_commit->super_roots + next_backup;
2082
2083 /*
2084 * make sure all of our padding and empty slots get zero filled
2085 * regardless of which ones we use today
2086 */
2087 memset(root_backup, 0, sizeof(*root_backup));
2088
2089 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2090
2091 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2092 btrfs_set_backup_tree_root_gen(root_backup,
2093 btrfs_header_generation(info->tree_root->node));
2094
2095 btrfs_set_backup_tree_root_level(root_backup,
2096 btrfs_header_level(info->tree_root->node));
2097
2098 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2099 btrfs_set_backup_chunk_root_gen(root_backup,
2100 btrfs_header_generation(info->chunk_root->node));
2101 btrfs_set_backup_chunk_root_level(root_backup,
2102 btrfs_header_level(info->chunk_root->node));
2103
2104 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2105 btrfs_set_backup_extent_root_gen(root_backup,
2106 btrfs_header_generation(info->extent_root->node));
2107 btrfs_set_backup_extent_root_level(root_backup,
2108 btrfs_header_level(info->extent_root->node));
2109
2110 /*
2111 * we might commit during log recovery, which happens before we set
2112 * the fs_root. Make sure it is valid before we fill it in.
2113 */
2114 if (info->fs_root && info->fs_root->node) {
2115 btrfs_set_backup_fs_root(root_backup,
2116 info->fs_root->node->start);
2117 btrfs_set_backup_fs_root_gen(root_backup,
2118 btrfs_header_generation(info->fs_root->node));
2119 btrfs_set_backup_fs_root_level(root_backup,
2120 btrfs_header_level(info->fs_root->node));
2121 }
2122
2123 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2124 btrfs_set_backup_dev_root_gen(root_backup,
2125 btrfs_header_generation(info->dev_root->node));
2126 btrfs_set_backup_dev_root_level(root_backup,
2127 btrfs_header_level(info->dev_root->node));
2128
2129 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2130 btrfs_set_backup_csum_root_gen(root_backup,
2131 btrfs_header_generation(info->csum_root->node));
2132 btrfs_set_backup_csum_root_level(root_backup,
2133 btrfs_header_level(info->csum_root->node));
2134
2135 btrfs_set_backup_total_bytes(root_backup,
2136 btrfs_super_total_bytes(info->super_copy));
2137 btrfs_set_backup_bytes_used(root_backup,
2138 btrfs_super_bytes_used(info->super_copy));
2139 btrfs_set_backup_num_devices(root_backup,
2140 btrfs_super_num_devices(info->super_copy));
2141
2142 /*
2143 * if we don't copy this out to the super_copy, it won't get remembered
2144 * for the next commit
2145 */
2146 memcpy(&info->super_copy->super_roots,
2147 &info->super_for_commit->super_roots,
2148 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2149}
2150
2151/*
2152 * this copies info out of the root backup array and back into
2153 * the in-memory super block. It is meant to help iterate through
2154 * the array, so you send it the number of backups you've already
2155 * tried and the last backup index you used.
2156 *
2157 * this returns -1 when it has tried all the backups
2158 */
2159static noinline int next_root_backup(struct btrfs_fs_info *info,
2160 struct btrfs_super_block *super,
2161 int *num_backups_tried, int *backup_index)
2162{
2163 struct btrfs_root_backup *root_backup;
2164 int newest = *backup_index;
2165
2166 if (*num_backups_tried == 0) {
2167 u64 gen = btrfs_super_generation(super);
2168
2169 newest = find_newest_super_backup(info, gen);
2170 if (newest == -1)
2171 return -1;
2172
2173 *backup_index = newest;
2174 *num_backups_tried = 1;
2175 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2176 /* we've tried all the backups, all done */
2177 return -1;
2178 } else {
2179 /* jump to the next oldest backup */
2180 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2181 BTRFS_NUM_BACKUP_ROOTS;
2182 *backup_index = newest;
2183 *num_backups_tried += 1;
2184 }
2185 root_backup = super->super_roots + newest;
2186
2187 btrfs_set_super_generation(super,
2188 btrfs_backup_tree_root_gen(root_backup));
2189 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2190 btrfs_set_super_root_level(super,
2191 btrfs_backup_tree_root_level(root_backup));
2192 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2193
2194 /*
2195 * fixme: the total bytes and num_devices need to match or we should
2196 * need a fsck
2197 */
2198 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2199 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2200 return 0;
2201}
2202
2203/* helper to cleanup workers */
2204static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2205{
2206 btrfs_destroy_workqueue(fs_info->fixup_workers);
2207 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2208 btrfs_destroy_workqueue(fs_info->workers);
2209 btrfs_destroy_workqueue(fs_info->endio_workers);
2210 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2211 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2212 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2213 btrfs_destroy_workqueue(fs_info->rmw_workers);
2214 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2215 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2216 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2217 btrfs_destroy_workqueue(fs_info->submit_workers);
2218 btrfs_destroy_workqueue(fs_info->delayed_workers);
2219 btrfs_destroy_workqueue(fs_info->caching_workers);
2220 btrfs_destroy_workqueue(fs_info->readahead_workers);
2221 btrfs_destroy_workqueue(fs_info->flush_workers);
2222 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2223 btrfs_destroy_workqueue(fs_info->extent_workers);
2224}
2225
2226static void free_root_extent_buffers(struct btrfs_root *root)
2227{
2228 if (root) {
2229 free_extent_buffer(root->node);
2230 free_extent_buffer(root->commit_root);
2231 root->node = NULL;
2232 root->commit_root = NULL;
2233 }
2234}
2235
2236/* helper to cleanup tree roots */
2237static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2238{
2239 free_root_extent_buffers(info->tree_root);
2240
2241 free_root_extent_buffers(info->dev_root);
2242 free_root_extent_buffers(info->extent_root);
2243 free_root_extent_buffers(info->csum_root);
2244 free_root_extent_buffers(info->quota_root);
2245 free_root_extent_buffers(info->uuid_root);
2246 if (chunk_root)
2247 free_root_extent_buffers(info->chunk_root);
2248 free_root_extent_buffers(info->free_space_root);
2249}
2250
2251void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2252{
2253 int ret;
2254 struct btrfs_root *gang[8];
2255 int i;
2256
2257 while (!list_empty(&fs_info->dead_roots)) {
2258 gang[0] = list_entry(fs_info->dead_roots.next,
2259 struct btrfs_root, root_list);
2260 list_del(&gang[0]->root_list);
2261
2262 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2263 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2264 } else {
2265 free_extent_buffer(gang[0]->node);
2266 free_extent_buffer(gang[0]->commit_root);
2267 btrfs_put_fs_root(gang[0]);
2268 }
2269 }
2270
2271 while (1) {
2272 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2273 (void **)gang, 0,
2274 ARRAY_SIZE(gang));
2275 if (!ret)
2276 break;
2277 for (i = 0; i < ret; i++)
2278 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2279 }
2280
2281 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2282 btrfs_free_log_root_tree(NULL, fs_info);
2283 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2284 }
2285}
2286
2287static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2288{
2289 mutex_init(&fs_info->scrub_lock);
2290 atomic_set(&fs_info->scrubs_running, 0);
2291 atomic_set(&fs_info->scrub_pause_req, 0);
2292 atomic_set(&fs_info->scrubs_paused, 0);
2293 atomic_set(&fs_info->scrub_cancel_req, 0);
2294 init_waitqueue_head(&fs_info->scrub_pause_wait);
2295 fs_info->scrub_workers_refcnt = 0;
2296}
2297
2298static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2299{
2300 spin_lock_init(&fs_info->balance_lock);
2301 mutex_init(&fs_info->balance_mutex);
2302 atomic_set(&fs_info->balance_running, 0);
2303 atomic_set(&fs_info->balance_pause_req, 0);
2304 atomic_set(&fs_info->balance_cancel_req, 0);
2305 fs_info->balance_ctl = NULL;
2306 init_waitqueue_head(&fs_info->balance_wait_q);
2307}
2308
2309static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2310{
2311 struct inode *inode = fs_info->btree_inode;
2312
2313 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2314 set_nlink(inode, 1);
2315 /*
2316 * we set the i_size on the btree inode to the max possible int.
2317 * the real end of the address space is determined by all of
2318 * the devices in the system
2319 */
2320 inode->i_size = OFFSET_MAX;
2321 inode->i_mapping->a_ops = &btree_aops;
2322
2323 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2324 extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode->i_mapping);
2325 BTRFS_I(inode)->io_tree.track_uptodate = 0;
2326 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2327
2328 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2329
2330 BTRFS_I(inode)->root = fs_info->tree_root;
2331 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2332 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2333 btrfs_insert_inode_hash(inode);
2334}
2335
2336static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2337{
2338 fs_info->dev_replace.lock_owner = 0;
2339 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2340 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2341 rwlock_init(&fs_info->dev_replace.lock);
2342 atomic_set(&fs_info->dev_replace.read_locks, 0);
2343 atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2344 init_waitqueue_head(&fs_info->replace_wait);
2345 init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2346}
2347
2348static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2349{
2350 spin_lock_init(&fs_info->qgroup_lock);
2351 mutex_init(&fs_info->qgroup_ioctl_lock);
2352 fs_info->qgroup_tree = RB_ROOT;
2353 fs_info->qgroup_op_tree = RB_ROOT;
2354 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2355 fs_info->qgroup_seq = 1;
2356 fs_info->qgroup_ulist = NULL;
2357 fs_info->qgroup_rescan_running = false;
2358 mutex_init(&fs_info->qgroup_rescan_lock);
2359}
2360
2361static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2362 struct btrfs_fs_devices *fs_devices)
2363{
2364 int max_active = fs_info->thread_pool_size;
2365 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2366
2367 fs_info->workers =
2368 btrfs_alloc_workqueue(fs_info, "worker",
2369 flags | WQ_HIGHPRI, max_active, 16);
2370
2371 fs_info->delalloc_workers =
2372 btrfs_alloc_workqueue(fs_info, "delalloc",
2373 flags, max_active, 2);
2374
2375 fs_info->flush_workers =
2376 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2377 flags, max_active, 0);
2378
2379 fs_info->caching_workers =
2380 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2381
2382 /*
2383 * a higher idle thresh on the submit workers makes it much more
2384 * likely that bios will be send down in a sane order to the
2385 * devices
2386 */
2387 fs_info->submit_workers =
2388 btrfs_alloc_workqueue(fs_info, "submit", flags,
2389 min_t(u64, fs_devices->num_devices,
2390 max_active), 64);
2391
2392 fs_info->fixup_workers =
2393 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2394
2395 /*
2396 * endios are largely parallel and should have a very
2397 * low idle thresh
2398 */
2399 fs_info->endio_workers =
2400 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2401 fs_info->endio_meta_workers =
2402 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2403 max_active, 4);
2404 fs_info->endio_meta_write_workers =
2405 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2406 max_active, 2);
2407 fs_info->endio_raid56_workers =
2408 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2409 max_active, 4);
2410 fs_info->endio_repair_workers =
2411 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2412 fs_info->rmw_workers =
2413 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2414 fs_info->endio_write_workers =
2415 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2416 max_active, 2);
2417 fs_info->endio_freespace_worker =
2418 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2419 max_active, 0);
2420 fs_info->delayed_workers =
2421 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2422 max_active, 0);
2423 fs_info->readahead_workers =
2424 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2425 max_active, 2);
2426 fs_info->qgroup_rescan_workers =
2427 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2428 fs_info->extent_workers =
2429 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2430 min_t(u64, fs_devices->num_devices,
2431 max_active), 8);
2432
2433 if (!(fs_info->workers && fs_info->delalloc_workers &&
2434 fs_info->submit_workers && fs_info->flush_workers &&
2435 fs_info->endio_workers && fs_info->endio_meta_workers &&
2436 fs_info->endio_meta_write_workers &&
2437 fs_info->endio_repair_workers &&
2438 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2439 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2440 fs_info->caching_workers && fs_info->readahead_workers &&
2441 fs_info->fixup_workers && fs_info->delayed_workers &&
2442 fs_info->extent_workers &&
2443 fs_info->qgroup_rescan_workers)) {
2444 return -ENOMEM;
2445 }
2446
2447 return 0;
2448}
2449
2450static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2451 struct btrfs_fs_devices *fs_devices)
2452{
2453 int ret;
2454 struct btrfs_root *log_tree_root;
2455 struct btrfs_super_block *disk_super = fs_info->super_copy;
2456 u64 bytenr = btrfs_super_log_root(disk_super);
2457
2458 if (fs_devices->rw_devices == 0) {
2459 btrfs_warn(fs_info, "log replay required on RO media");
2460 return -EIO;
2461 }
2462
2463 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2464 if (!log_tree_root)
2465 return -ENOMEM;
2466
2467 __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2468
2469 log_tree_root->node = read_tree_block(fs_info, bytenr,
2470 fs_info->generation + 1);
2471 if (IS_ERR(log_tree_root->node)) {
2472 btrfs_warn(fs_info, "failed to read log tree");
2473 ret = PTR_ERR(log_tree_root->node);
2474 kfree(log_tree_root);
2475 return ret;
2476 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2477 btrfs_err(fs_info, "failed to read log tree");
2478 free_extent_buffer(log_tree_root->node);
2479 kfree(log_tree_root);
2480 return -EIO;
2481 }
2482 /* returns with log_tree_root freed on success */
2483 ret = btrfs_recover_log_trees(log_tree_root);
2484 if (ret) {
2485 btrfs_handle_fs_error(fs_info, ret,
2486 "Failed to recover log tree");
2487 free_extent_buffer(log_tree_root->node);
2488 kfree(log_tree_root);
2489 return ret;
2490 }
2491
2492 if (fs_info->sb->s_flags & MS_RDONLY) {
2493 ret = btrfs_commit_super(fs_info);
2494 if (ret)
2495 return ret;
2496 }
2497
2498 return 0;
2499}
2500
2501static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2502{
2503 struct btrfs_root *tree_root = fs_info->tree_root;
2504 struct btrfs_root *root;
2505 struct btrfs_key location;
2506 int ret;
2507
2508 BUG_ON(!fs_info->tree_root);
2509
2510 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2511 location.type = BTRFS_ROOT_ITEM_KEY;
2512 location.offset = 0;
2513
2514 root = btrfs_read_tree_root(tree_root, &location);
2515 if (IS_ERR(root))
2516 return PTR_ERR(root);
2517 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2518 fs_info->extent_root = root;
2519
2520 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2521 root = btrfs_read_tree_root(tree_root, &location);
2522 if (IS_ERR(root))
2523 return PTR_ERR(root);
2524 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2525 fs_info->dev_root = root;
2526 btrfs_init_devices_late(fs_info);
2527
2528 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2529 root = btrfs_read_tree_root(tree_root, &location);
2530 if (IS_ERR(root))
2531 return PTR_ERR(root);
2532 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2533 fs_info->csum_root = root;
2534
2535 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2536 root = btrfs_read_tree_root(tree_root, &location);
2537 if (!IS_ERR(root)) {
2538 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2539 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2540 fs_info->quota_root = root;
2541 }
2542
2543 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2544 root = btrfs_read_tree_root(tree_root, &location);
2545 if (IS_ERR(root)) {
2546 ret = PTR_ERR(root);
2547 if (ret != -ENOENT)
2548 return ret;
2549 } else {
2550 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2551 fs_info->uuid_root = root;
2552 }
2553
2554 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2555 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2556 root = btrfs_read_tree_root(tree_root, &location);
2557 if (IS_ERR(root))
2558 return PTR_ERR(root);
2559 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2560 fs_info->free_space_root = root;
2561 }
2562
2563 return 0;
2564}
2565
2566int open_ctree(struct super_block *sb,
2567 struct btrfs_fs_devices *fs_devices,
2568 char *options)
2569{
2570 u32 sectorsize;
2571 u32 nodesize;
2572 u32 stripesize;
2573 u64 generation;
2574 u64 features;
2575 struct btrfs_key location;
2576 struct buffer_head *bh;
2577 struct btrfs_super_block *disk_super;
2578 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2579 struct btrfs_root *tree_root;
2580 struct btrfs_root *chunk_root;
2581 int ret;
2582 int err = -EINVAL;
2583 int num_backups_tried = 0;
2584 int backup_index = 0;
2585 int max_active;
2586 int clear_free_space_tree = 0;
2587
2588 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2589 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2590 if (!tree_root || !chunk_root) {
2591 err = -ENOMEM;
2592 goto fail;
2593 }
2594
2595 ret = init_srcu_struct(&fs_info->subvol_srcu);
2596 if (ret) {
2597 err = ret;
2598 goto fail;
2599 }
2600
2601 ret = setup_bdi(fs_info, &fs_info->bdi);
2602 if (ret) {
2603 err = ret;
2604 goto fail_srcu;
2605 }
2606
2607 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2608 if (ret) {
2609 err = ret;
2610 goto fail_bdi;
2611 }
2612 fs_info->dirty_metadata_batch = PAGE_SIZE *
2613 (1 + ilog2(nr_cpu_ids));
2614
2615 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2616 if (ret) {
2617 err = ret;
2618 goto fail_dirty_metadata_bytes;
2619 }
2620
2621 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2622 if (ret) {
2623 err = ret;
2624 goto fail_delalloc_bytes;
2625 }
2626
2627 fs_info->btree_inode = new_inode(sb);
2628 if (!fs_info->btree_inode) {
2629 err = -ENOMEM;
2630 goto fail_bio_counter;
2631 }
2632
2633 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2634
2635 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2636 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2637 INIT_LIST_HEAD(&fs_info->trans_list);
2638 INIT_LIST_HEAD(&fs_info->dead_roots);
2639 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2640 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2641 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2642 spin_lock_init(&fs_info->delalloc_root_lock);
2643 spin_lock_init(&fs_info->trans_lock);
2644 spin_lock_init(&fs_info->fs_roots_radix_lock);
2645 spin_lock_init(&fs_info->delayed_iput_lock);
2646 spin_lock_init(&fs_info->defrag_inodes_lock);
2647 spin_lock_init(&fs_info->free_chunk_lock);
2648 spin_lock_init(&fs_info->tree_mod_seq_lock);
2649 spin_lock_init(&fs_info->super_lock);
2650 spin_lock_init(&fs_info->qgroup_op_lock);
2651 spin_lock_init(&fs_info->buffer_lock);
2652 spin_lock_init(&fs_info->unused_bgs_lock);
2653 rwlock_init(&fs_info->tree_mod_log_lock);
2654 mutex_init(&fs_info->unused_bg_unpin_mutex);
2655 mutex_init(&fs_info->delete_unused_bgs_mutex);
2656 mutex_init(&fs_info->reloc_mutex);
2657 mutex_init(&fs_info->delalloc_root_mutex);
2658 mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2659 seqlock_init(&fs_info->profiles_lock);
2660
2661 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2662 INIT_LIST_HEAD(&fs_info->space_info);
2663 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2664 INIT_LIST_HEAD(&fs_info->unused_bgs);
2665 btrfs_mapping_init(&fs_info->mapping_tree);
2666 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2667 BTRFS_BLOCK_RSV_GLOBAL);
2668 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2669 BTRFS_BLOCK_RSV_DELALLOC);
2670 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2671 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2672 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2673 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2674 BTRFS_BLOCK_RSV_DELOPS);
2675 atomic_set(&fs_info->nr_async_submits, 0);
2676 atomic_set(&fs_info->async_delalloc_pages, 0);
2677 atomic_set(&fs_info->async_submit_draining, 0);
2678 atomic_set(&fs_info->nr_async_bios, 0);
2679 atomic_set(&fs_info->defrag_running, 0);
2680 atomic_set(&fs_info->qgroup_op_seq, 0);
2681 atomic_set(&fs_info->reada_works_cnt, 0);
2682 atomic64_set(&fs_info->tree_mod_seq, 0);
2683 fs_info->fs_frozen = 0;
2684 fs_info->sb = sb;
2685 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2686 fs_info->metadata_ratio = 0;
2687 fs_info->defrag_inodes = RB_ROOT;
2688 fs_info->free_chunk_space = 0;
2689 fs_info->tree_mod_log = RB_ROOT;
2690 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2691 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2692 /* readahead state */
2693 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2694 spin_lock_init(&fs_info->reada_lock);
2695
2696 fs_info->thread_pool_size = min_t(unsigned long,
2697 num_online_cpus() + 2, 8);
2698
2699 INIT_LIST_HEAD(&fs_info->ordered_roots);
2700 spin_lock_init(&fs_info->ordered_root_lock);
2701 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2702 GFP_KERNEL);
2703 if (!fs_info->delayed_root) {
2704 err = -ENOMEM;
2705 goto fail_iput;
2706 }
2707 btrfs_init_delayed_root(fs_info->delayed_root);
2708
2709 btrfs_init_scrub(fs_info);
2710#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2711 fs_info->check_integrity_print_mask = 0;
2712#endif
2713 btrfs_init_balance(fs_info);
2714 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2715
2716 sb->s_blocksize = 4096;
2717 sb->s_blocksize_bits = blksize_bits(4096);
2718 sb->s_bdi = &fs_info->bdi;
2719
2720 btrfs_init_btree_inode(fs_info);
2721
2722 spin_lock_init(&fs_info->block_group_cache_lock);
2723 fs_info->block_group_cache_tree = RB_ROOT;
2724 fs_info->first_logical_byte = (u64)-1;
2725
2726 extent_io_tree_init(&fs_info->freed_extents[0],
2727 fs_info->btree_inode->i_mapping);
2728 extent_io_tree_init(&fs_info->freed_extents[1],
2729 fs_info->btree_inode->i_mapping);
2730 fs_info->pinned_extents = &fs_info->freed_extents[0];
2731 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2732
2733 mutex_init(&fs_info->ordered_operations_mutex);
2734 mutex_init(&fs_info->tree_log_mutex);
2735 mutex_init(&fs_info->chunk_mutex);
2736 mutex_init(&fs_info->transaction_kthread_mutex);
2737 mutex_init(&fs_info->cleaner_mutex);
2738 mutex_init(&fs_info->volume_mutex);
2739 mutex_init(&fs_info->ro_block_group_mutex);
2740 init_rwsem(&fs_info->commit_root_sem);
2741 init_rwsem(&fs_info->cleanup_work_sem);
2742 init_rwsem(&fs_info->subvol_sem);
2743 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2744
2745 btrfs_init_dev_replace_locks(fs_info);
2746 btrfs_init_qgroup(fs_info);
2747
2748 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2749 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2750
2751 init_waitqueue_head(&fs_info->transaction_throttle);
2752 init_waitqueue_head(&fs_info->transaction_wait);
2753 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2754 init_waitqueue_head(&fs_info->async_submit_wait);
2755
2756 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2757
2758 /* Usable values until the real ones are cached from the superblock */
2759 fs_info->nodesize = 4096;
2760 fs_info->sectorsize = 4096;
2761 fs_info->stripesize = 4096;
2762
2763 ret = btrfs_alloc_stripe_hash_table(fs_info);
2764 if (ret) {
2765 err = ret;
2766 goto fail_alloc;
2767 }
2768
2769 __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2770
2771 invalidate_bdev(fs_devices->latest_bdev);
2772
2773 /*
2774 * Read super block and check the signature bytes only
2775 */
2776 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2777 if (IS_ERR(bh)) {
2778 err = PTR_ERR(bh);
2779 goto fail_alloc;
2780 }
2781
2782 /*
2783 * We want to check superblock checksum, the type is stored inside.
2784 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2785 */
2786 if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2787 btrfs_err(fs_info, "superblock checksum mismatch");
2788 err = -EINVAL;
2789 brelse(bh);
2790 goto fail_alloc;
2791 }
2792
2793 /*
2794 * super_copy is zeroed at allocation time and we never touch the
2795 * following bytes up to INFO_SIZE, the checksum is calculated from
2796 * the whole block of INFO_SIZE
2797 */
2798 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2799 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2800 sizeof(*fs_info->super_for_commit));
2801 brelse(bh);
2802
2803 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2804
2805 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2806 if (ret) {
2807 btrfs_err(fs_info, "superblock contains fatal errors");
2808 err = -EINVAL;
2809 goto fail_alloc;
2810 }
2811
2812 disk_super = fs_info->super_copy;
2813 if (!btrfs_super_root(disk_super))
2814 goto fail_alloc;
2815
2816 /* check FS state, whether FS is broken. */
2817 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2818 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2819
2820 /*
2821 * run through our array of backup supers and setup
2822 * our ring pointer to the oldest one
2823 */
2824 generation = btrfs_super_generation(disk_super);
2825 find_oldest_super_backup(fs_info, generation);
2826
2827 /*
2828 * In the long term, we'll store the compression type in the super
2829 * block, and it'll be used for per file compression control.
2830 */
2831 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2832
2833 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2834 if (ret) {
2835 err = ret;
2836 goto fail_alloc;
2837 }
2838
2839 features = btrfs_super_incompat_flags(disk_super) &
2840 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2841 if (features) {
2842 btrfs_err(fs_info,
2843 "cannot mount because of unsupported optional features (%llx)",
2844 features);
2845 err = -EINVAL;
2846 goto fail_alloc;
2847 }
2848
2849 features = btrfs_super_incompat_flags(disk_super);
2850 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2851 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2852 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2853
2854 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2855 btrfs_info(fs_info, "has skinny extents");
2856
2857 /*
2858 * flag our filesystem as having big metadata blocks if
2859 * they are bigger than the page size
2860 */
2861 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2862 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2863 btrfs_info(fs_info,
2864 "flagging fs with big metadata feature");
2865 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2866 }
2867
2868 nodesize = btrfs_super_nodesize(disk_super);
2869 sectorsize = btrfs_super_sectorsize(disk_super);
2870 stripesize = sectorsize;
2871 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2872 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2873
2874 /* Cache block sizes */
2875 fs_info->nodesize = nodesize;
2876 fs_info->sectorsize = sectorsize;
2877 fs_info->stripesize = stripesize;
2878
2879 /*
2880 * mixed block groups end up with duplicate but slightly offset
2881 * extent buffers for the same range. It leads to corruptions
2882 */
2883 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2884 (sectorsize != nodesize)) {
2885 btrfs_err(fs_info,
2886"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2887 nodesize, sectorsize);
2888 goto fail_alloc;
2889 }
2890
2891 /*
2892 * Needn't use the lock because there is no other task which will
2893 * update the flag.
2894 */
2895 btrfs_set_super_incompat_flags(disk_super, features);
2896
2897 features = btrfs_super_compat_ro_flags(disk_super) &
2898 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2899 if (!(sb->s_flags & MS_RDONLY) && features) {
2900 btrfs_err(fs_info,
2901 "cannot mount read-write because of unsupported optional features (%llx)",
2902 features);
2903 err = -EINVAL;
2904 goto fail_alloc;
2905 }
2906
2907 max_active = fs_info->thread_pool_size;
2908
2909 ret = btrfs_init_workqueues(fs_info, fs_devices);
2910 if (ret) {
2911 err = ret;
2912 goto fail_sb_buffer;
2913 }
2914
2915 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2916 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2917 SZ_4M / PAGE_SIZE);
2918
2919 sb->s_blocksize = sectorsize;
2920 sb->s_blocksize_bits = blksize_bits(sectorsize);
2921
2922 mutex_lock(&fs_info->chunk_mutex);
2923 ret = btrfs_read_sys_array(fs_info);
2924 mutex_unlock(&fs_info->chunk_mutex);
2925 if (ret) {
2926 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2927 goto fail_sb_buffer;
2928 }
2929
2930 generation = btrfs_super_chunk_root_generation(disk_super);
2931
2932 __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2933
2934 chunk_root->node = read_tree_block(fs_info,
2935 btrfs_super_chunk_root(disk_super),
2936 generation);
2937 if (IS_ERR(chunk_root->node) ||
2938 !extent_buffer_uptodate(chunk_root->node)) {
2939 btrfs_err(fs_info, "failed to read chunk root");
2940 if (!IS_ERR(chunk_root->node))
2941 free_extent_buffer(chunk_root->node);
2942 chunk_root->node = NULL;
2943 goto fail_tree_roots;
2944 }
2945 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2946 chunk_root->commit_root = btrfs_root_node(chunk_root);
2947
2948 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2949 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2950
2951 ret = btrfs_read_chunk_tree(fs_info);
2952 if (ret) {
2953 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2954 goto fail_tree_roots;
2955 }
2956
2957 /*
2958 * keep the device that is marked to be the target device for the
2959 * dev_replace procedure
2960 */
2961 btrfs_close_extra_devices(fs_devices, 0);
2962
2963 if (!fs_devices->latest_bdev) {
2964 btrfs_err(fs_info, "failed to read devices");
2965 goto fail_tree_roots;
2966 }
2967
2968retry_root_backup:
2969 generation = btrfs_super_generation(disk_super);
2970
2971 tree_root->node = read_tree_block(fs_info,
2972 btrfs_super_root(disk_super),
2973 generation);
2974 if (IS_ERR(tree_root->node) ||
2975 !extent_buffer_uptodate(tree_root->node)) {
2976 btrfs_warn(fs_info, "failed to read tree root");
2977 if (!IS_ERR(tree_root->node))
2978 free_extent_buffer(tree_root->node);
2979 tree_root->node = NULL;
2980 goto recovery_tree_root;
2981 }
2982
2983 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2984 tree_root->commit_root = btrfs_root_node(tree_root);
2985 btrfs_set_root_refs(&tree_root->root_item, 1);
2986
2987 mutex_lock(&tree_root->objectid_mutex);
2988 ret = btrfs_find_highest_objectid(tree_root,
2989 &tree_root->highest_objectid);
2990 if (ret) {
2991 mutex_unlock(&tree_root->objectid_mutex);
2992 goto recovery_tree_root;
2993 }
2994
2995 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2996
2997 mutex_unlock(&tree_root->objectid_mutex);
2998
2999 ret = btrfs_read_roots(fs_info);
3000 if (ret)
3001 goto recovery_tree_root;
3002
3003 fs_info->generation = generation;
3004 fs_info->last_trans_committed = generation;
3005
3006 ret = btrfs_recover_balance(fs_info);
3007 if (ret) {
3008 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3009 goto fail_block_groups;
3010 }
3011
3012 ret = btrfs_init_dev_stats(fs_info);
3013 if (ret) {
3014 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3015 goto fail_block_groups;
3016 }
3017
3018 ret = btrfs_init_dev_replace(fs_info);
3019 if (ret) {
3020 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3021 goto fail_block_groups;
3022 }
3023
3024 btrfs_close_extra_devices(fs_devices, 1);
3025
3026 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3027 if (ret) {
3028 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3029 ret);
3030 goto fail_block_groups;
3031 }
3032
3033 ret = btrfs_sysfs_add_device(fs_devices);
3034 if (ret) {
3035 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3036 ret);
3037 goto fail_fsdev_sysfs;
3038 }
3039
3040 ret = btrfs_sysfs_add_mounted(fs_info);
3041 if (ret) {
3042 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3043 goto fail_fsdev_sysfs;
3044 }
3045
3046 ret = btrfs_init_space_info(fs_info);
3047 if (ret) {
3048 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3049 goto fail_sysfs;
3050 }
3051
3052 ret = btrfs_read_block_groups(fs_info);
3053 if (ret) {
3054 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3055 goto fail_sysfs;
3056 }
3057 fs_info->num_tolerated_disk_barrier_failures =
3058 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3059 if (fs_info->fs_devices->missing_devices >
3060 fs_info->num_tolerated_disk_barrier_failures &&
3061 !(sb->s_flags & MS_RDONLY)) {
3062 btrfs_warn(fs_info,
3063"missing devices (%llu) exceeds the limit (%d), writeable mount is not allowed",
3064 fs_info->fs_devices->missing_devices,
3065 fs_info->num_tolerated_disk_barrier_failures);
3066 goto fail_sysfs;
3067 }
3068
3069 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3070 "btrfs-cleaner");
3071 if (IS_ERR(fs_info->cleaner_kthread))
3072 goto fail_sysfs;
3073
3074 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3075 tree_root,
3076 "btrfs-transaction");
3077 if (IS_ERR(fs_info->transaction_kthread))
3078 goto fail_cleaner;
3079
3080 if (!btrfs_test_opt(fs_info, SSD) &&
3081 !btrfs_test_opt(fs_info, NOSSD) &&
3082 !fs_info->fs_devices->rotating) {
3083 btrfs_info(fs_info, "detected SSD devices, enabling SSD mode");
3084 btrfs_set_opt(fs_info->mount_opt, SSD);
3085 }
3086
3087 /*
3088 * Mount does not set all options immediately, we can do it now and do
3089 * not have to wait for transaction commit
3090 */
3091 btrfs_apply_pending_changes(fs_info);
3092
3093#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3094 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3095 ret = btrfsic_mount(fs_info, fs_devices,
3096 btrfs_test_opt(fs_info,
3097 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3098 1 : 0,
3099 fs_info->check_integrity_print_mask);
3100 if (ret)
3101 btrfs_warn(fs_info,
3102 "failed to initialize integrity check module: %d",
3103 ret);
3104 }
3105#endif
3106 ret = btrfs_read_qgroup_config(fs_info);
3107 if (ret)
3108 goto fail_trans_kthread;
3109
3110 /* do not make disk changes in broken FS or nologreplay is given */
3111 if (btrfs_super_log_root(disk_super) != 0 &&
3112 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3113 ret = btrfs_replay_log(fs_info, fs_devices);
3114 if (ret) {
3115 err = ret;
3116 goto fail_qgroup;
3117 }
3118 }
3119
3120 ret = btrfs_find_orphan_roots(fs_info);
3121 if (ret)
3122 goto fail_qgroup;
3123
3124 if (!(sb->s_flags & MS_RDONLY)) {
3125 ret = btrfs_cleanup_fs_roots(fs_info);
3126 if (ret)
3127 goto fail_qgroup;
3128
3129 mutex_lock(&fs_info->cleaner_mutex);
3130 ret = btrfs_recover_relocation(tree_root);
3131 mutex_unlock(&fs_info->cleaner_mutex);
3132 if (ret < 0) {
3133 btrfs_warn(fs_info, "failed to recover relocation: %d",
3134 ret);
3135 err = -EINVAL;
3136 goto fail_qgroup;
3137 }
3138 }
3139
3140 location.objectid = BTRFS_FS_TREE_OBJECTID;
3141 location.type = BTRFS_ROOT_ITEM_KEY;
3142 location.offset = 0;
3143
3144 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3145 if (IS_ERR(fs_info->fs_root)) {
3146 err = PTR_ERR(fs_info->fs_root);
3147 goto fail_qgroup;
3148 }
3149
3150 if (sb->s_flags & MS_RDONLY)
3151 return 0;
3152
3153 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3154 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3155 clear_free_space_tree = 1;
3156 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3157 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3158 btrfs_warn(fs_info, "free space tree is invalid");
3159 clear_free_space_tree = 1;
3160 }
3161
3162 if (clear_free_space_tree) {
3163 btrfs_info(fs_info, "clearing free space tree");
3164 ret = btrfs_clear_free_space_tree(fs_info);
3165 if (ret) {
3166 btrfs_warn(fs_info,
3167 "failed to clear free space tree: %d", ret);
3168 close_ctree(fs_info);
3169 return ret;
3170 }
3171 }
3172
3173 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3174 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3175 btrfs_info(fs_info, "creating free space tree");
3176 ret = btrfs_create_free_space_tree(fs_info);
3177 if (ret) {
3178 btrfs_warn(fs_info,
3179 "failed to create free space tree: %d", ret);
3180 close_ctree(fs_info);
3181 return ret;
3182 }
3183 }
3184
3185 down_read(&fs_info->cleanup_work_sem);
3186 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3187 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3188 up_read(&fs_info->cleanup_work_sem);
3189 close_ctree(fs_info);
3190 return ret;
3191 }
3192 up_read(&fs_info->cleanup_work_sem);
3193
3194 ret = btrfs_resume_balance_async(fs_info);
3195 if (ret) {
3196 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3197 close_ctree(fs_info);
3198 return ret;
3199 }
3200
3201 ret = btrfs_resume_dev_replace_async(fs_info);
3202 if (ret) {
3203 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3204 close_ctree(fs_info);
3205 return ret;
3206 }
3207
3208 btrfs_qgroup_rescan_resume(fs_info);
3209
3210 if (!fs_info->uuid_root) {
3211 btrfs_info(fs_info, "creating UUID tree");
3212 ret = btrfs_create_uuid_tree(fs_info);
3213 if (ret) {
3214 btrfs_warn(fs_info,
3215 "failed to create the UUID tree: %d", ret);
3216 close_ctree(fs_info);
3217 return ret;
3218 }
3219 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3220 fs_info->generation !=
3221 btrfs_super_uuid_tree_generation(disk_super)) {
3222 btrfs_info(fs_info, "checking UUID tree");
3223 ret = btrfs_check_uuid_tree(fs_info);
3224 if (ret) {
3225 btrfs_warn(fs_info,
3226 "failed to check the UUID tree: %d", ret);
3227 close_ctree(fs_info);
3228 return ret;
3229 }
3230 } else {
3231 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3232 }
3233 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3234
3235 /*
3236 * backuproot only affect mount behavior, and if open_ctree succeeded,
3237 * no need to keep the flag
3238 */
3239 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3240
3241 return 0;
3242
3243fail_qgroup:
3244 btrfs_free_qgroup_config(fs_info);
3245fail_trans_kthread:
3246 kthread_stop(fs_info->transaction_kthread);
3247 btrfs_cleanup_transaction(fs_info);
3248 btrfs_free_fs_roots(fs_info);
3249fail_cleaner:
3250 kthread_stop(fs_info->cleaner_kthread);
3251
3252 /*
3253 * make sure we're done with the btree inode before we stop our
3254 * kthreads
3255 */
3256 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3257
3258fail_sysfs:
3259 btrfs_sysfs_remove_mounted(fs_info);
3260
3261fail_fsdev_sysfs:
3262 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3263
3264fail_block_groups:
3265 btrfs_put_block_group_cache(fs_info);
3266 btrfs_free_block_groups(fs_info);
3267
3268fail_tree_roots:
3269 free_root_pointers(fs_info, 1);
3270 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3271
3272fail_sb_buffer:
3273 btrfs_stop_all_workers(fs_info);
3274fail_alloc:
3275fail_iput:
3276 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3277
3278 iput(fs_info->btree_inode);
3279fail_bio_counter:
3280 percpu_counter_destroy(&fs_info->bio_counter);
3281fail_delalloc_bytes:
3282 percpu_counter_destroy(&fs_info->delalloc_bytes);
3283fail_dirty_metadata_bytes:
3284 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3285fail_bdi:
3286 bdi_destroy(&fs_info->bdi);
3287fail_srcu:
3288 cleanup_srcu_struct(&fs_info->subvol_srcu);
3289fail:
3290 btrfs_free_stripe_hash_table(fs_info);
3291 btrfs_close_devices(fs_info->fs_devices);
3292 return err;
3293
3294recovery_tree_root:
3295 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3296 goto fail_tree_roots;
3297
3298 free_root_pointers(fs_info, 0);
3299
3300 /* don't use the log in recovery mode, it won't be valid */
3301 btrfs_set_super_log_root(disk_super, 0);
3302
3303 /* we can't trust the free space cache either */
3304 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3305
3306 ret = next_root_backup(fs_info, fs_info->super_copy,
3307 &num_backups_tried, &backup_index);
3308 if (ret == -1)
3309 goto fail_block_groups;
3310 goto retry_root_backup;
3311}
3312
3313static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3314{
3315 if (uptodate) {
3316 set_buffer_uptodate(bh);
3317 } else {
3318 struct btrfs_device *device = (struct btrfs_device *)
3319 bh->b_private;
3320
3321 btrfs_warn_rl_in_rcu(device->fs_info,
3322 "lost page write due to IO error on %s",
3323 rcu_str_deref(device->name));
3324 /* note, we don't set_buffer_write_io_error because we have
3325 * our own ways of dealing with the IO errors
3326 */
3327 clear_buffer_uptodate(bh);
3328 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3329 }
3330 unlock_buffer(bh);
3331 put_bh(bh);
3332}
3333
3334int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3335 struct buffer_head **bh_ret)
3336{
3337 struct buffer_head *bh;
3338 struct btrfs_super_block *super;
3339 u64 bytenr;
3340
3341 bytenr = btrfs_sb_offset(copy_num);
3342 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3343 return -EINVAL;
3344
3345 bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3346 /*
3347 * If we fail to read from the underlying devices, as of now
3348 * the best option we have is to mark it EIO.
3349 */
3350 if (!bh)
3351 return -EIO;
3352
3353 super = (struct btrfs_super_block *)bh->b_data;
3354 if (btrfs_super_bytenr(super) != bytenr ||
3355 btrfs_super_magic(super) != BTRFS_MAGIC) {
3356 brelse(bh);
3357 return -EINVAL;
3358 }
3359
3360 *bh_ret = bh;
3361 return 0;
3362}
3363
3364
3365struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3366{
3367 struct buffer_head *bh;
3368 struct buffer_head *latest = NULL;
3369 struct btrfs_super_block *super;
3370 int i;
3371 u64 transid = 0;
3372 int ret = -EINVAL;
3373
3374 /* we would like to check all the supers, but that would make
3375 * a btrfs mount succeed after a mkfs from a different FS.
3376 * So, we need to add a special mount option to scan for
3377 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3378 */
3379 for (i = 0; i < 1; i++) {
3380 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3381 if (ret)
3382 continue;
3383
3384 super = (struct btrfs_super_block *)bh->b_data;
3385
3386 if (!latest || btrfs_super_generation(super) > transid) {
3387 brelse(latest);
3388 latest = bh;
3389 transid = btrfs_super_generation(super);
3390 } else {
3391 brelse(bh);
3392 }
3393 }
3394
3395 if (!latest)
3396 return ERR_PTR(ret);
3397
3398 return latest;
3399}
3400
3401/*
3402 * this should be called twice, once with wait == 0 and
3403 * once with wait == 1. When wait == 0 is done, all the buffer heads
3404 * we write are pinned.
3405 *
3406 * They are released when wait == 1 is done.
3407 * max_mirrors must be the same for both runs, and it indicates how
3408 * many supers on this one device should be written.
3409 *
3410 * max_mirrors == 0 means to write them all.
3411 */
3412static int write_dev_supers(struct btrfs_device *device,
3413 struct btrfs_super_block *sb,
3414 int do_barriers, int wait, int max_mirrors)
3415{
3416 struct buffer_head *bh;
3417 int i;
3418 int ret;
3419 int errors = 0;
3420 u32 crc;
3421 u64 bytenr;
3422
3423 if (max_mirrors == 0)
3424 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3425
3426 for (i = 0; i < max_mirrors; i++) {
3427 bytenr = btrfs_sb_offset(i);
3428 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3429 device->commit_total_bytes)
3430 break;
3431
3432 if (wait) {
3433 bh = __find_get_block(device->bdev, bytenr / 4096,
3434 BTRFS_SUPER_INFO_SIZE);
3435 if (!bh) {
3436 errors++;
3437 continue;
3438 }
3439 wait_on_buffer(bh);
3440 if (!buffer_uptodate(bh))
3441 errors++;
3442
3443 /* drop our reference */
3444 brelse(bh);
3445
3446 /* drop the reference from the wait == 0 run */
3447 brelse(bh);
3448 continue;
3449 } else {
3450 btrfs_set_super_bytenr(sb, bytenr);
3451
3452 crc = ~(u32)0;
3453 crc = btrfs_csum_data((char *)sb +
3454 BTRFS_CSUM_SIZE, crc,
3455 BTRFS_SUPER_INFO_SIZE -
3456 BTRFS_CSUM_SIZE);
3457 btrfs_csum_final(crc, sb->csum);
3458
3459 /*
3460 * one reference for us, and we leave it for the
3461 * caller
3462 */
3463 bh = __getblk(device->bdev, bytenr / 4096,
3464 BTRFS_SUPER_INFO_SIZE);
3465 if (!bh) {
3466 btrfs_err(device->fs_info,
3467 "couldn't get super buffer head for bytenr %llu",
3468 bytenr);
3469 errors++;
3470 continue;
3471 }
3472
3473 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3474
3475 /* one reference for submit_bh */
3476 get_bh(bh);
3477
3478 set_buffer_uptodate(bh);
3479 lock_buffer(bh);
3480 bh->b_end_io = btrfs_end_buffer_write_sync;
3481 bh->b_private = device;
3482 }
3483
3484 /*
3485 * we fua the first super. The others we allow
3486 * to go down lazy.
3487 */
3488 if (i == 0)
3489 ret = btrfsic_submit_bh(REQ_OP_WRITE, REQ_FUA, bh);
3490 else
3491 ret = btrfsic_submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
3492 if (ret)
3493 errors++;
3494 }
3495 return errors < i ? 0 : -1;
3496}
3497
3498/*
3499 * endio for the write_dev_flush, this will wake anyone waiting
3500 * for the barrier when it is done
3501 */
3502static void btrfs_end_empty_barrier(struct bio *bio)
3503{
3504 if (bio->bi_private)
3505 complete(bio->bi_private);
3506 bio_put(bio);
3507}
3508
3509/*
3510 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3511 * sent down. With wait == 1, it waits for the previous flush.
3512 *
3513 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3514 * capable
3515 */
3516static int write_dev_flush(struct btrfs_device *device, int wait)
3517{
3518 struct bio *bio;
3519 int ret = 0;
3520
3521 if (device->nobarriers)
3522 return 0;
3523
3524 if (wait) {
3525 bio = device->flush_bio;
3526 if (!bio)
3527 return 0;
3528
3529 wait_for_completion(&device->flush_wait);
3530
3531 if (bio->bi_error) {
3532 ret = bio->bi_error;
3533 btrfs_dev_stat_inc_and_print(device,
3534 BTRFS_DEV_STAT_FLUSH_ERRS);
3535 }
3536
3537 /* drop the reference from the wait == 0 run */
3538 bio_put(bio);
3539 device->flush_bio = NULL;
3540
3541 return ret;
3542 }
3543
3544 /*
3545 * one reference for us, and we leave it for the
3546 * caller
3547 */
3548 device->flush_bio = NULL;
3549 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3550 if (!bio)
3551 return -ENOMEM;
3552
3553 bio->bi_end_io = btrfs_end_empty_barrier;
3554 bio->bi_bdev = device->bdev;
3555 bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
3556 init_completion(&device->flush_wait);
3557 bio->bi_private = &device->flush_wait;
3558 device->flush_bio = bio;
3559
3560 bio_get(bio);
3561 btrfsic_submit_bio(bio);
3562
3563 return 0;
3564}
3565
3566/*
3567 * send an empty flush down to each device in parallel,
3568 * then wait for them
3569 */
3570static int barrier_all_devices(struct btrfs_fs_info *info)
3571{
3572 struct list_head *head;
3573 struct btrfs_device *dev;
3574 int errors_send = 0;
3575 int errors_wait = 0;
3576 int ret;
3577
3578 /* send down all the barriers */
3579 head = &info->fs_devices->devices;
3580 list_for_each_entry_rcu(dev, head, dev_list) {
3581 if (dev->missing)
3582 continue;
3583 if (!dev->bdev) {
3584 errors_send++;
3585 continue;
3586 }
3587 if (!dev->in_fs_metadata || !dev->writeable)
3588 continue;
3589
3590 ret = write_dev_flush(dev, 0);
3591 if (ret)
3592 errors_send++;
3593 }
3594
3595 /* wait for all the barriers */
3596 list_for_each_entry_rcu(dev, head, dev_list) {
3597 if (dev->missing)
3598 continue;
3599 if (!dev->bdev) {
3600 errors_wait++;
3601 continue;
3602 }
3603 if (!dev->in_fs_metadata || !dev->writeable)
3604 continue;
3605
3606 ret = write_dev_flush(dev, 1);
3607 if (ret)
3608 errors_wait++;
3609 }
3610 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3611 errors_wait > info->num_tolerated_disk_barrier_failures)
3612 return -EIO;
3613 return 0;
3614}
3615
3616int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3617{
3618 int raid_type;
3619 int min_tolerated = INT_MAX;
3620
3621 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3622 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3623 min_tolerated = min(min_tolerated,
3624 btrfs_raid_array[BTRFS_RAID_SINGLE].
3625 tolerated_failures);
3626
3627 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3628 if (raid_type == BTRFS_RAID_SINGLE)
3629 continue;
3630 if (!(flags & btrfs_raid_group[raid_type]))
3631 continue;
3632 min_tolerated = min(min_tolerated,
3633 btrfs_raid_array[raid_type].
3634 tolerated_failures);
3635 }
3636
3637 if (min_tolerated == INT_MAX) {
3638 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3639 min_tolerated = 0;
3640 }
3641
3642 return min_tolerated;
3643}
3644
3645int btrfs_calc_num_tolerated_disk_barrier_failures(
3646 struct btrfs_fs_info *fs_info)
3647{
3648 struct btrfs_ioctl_space_info space;
3649 struct btrfs_space_info *sinfo;
3650 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3651 BTRFS_BLOCK_GROUP_SYSTEM,
3652 BTRFS_BLOCK_GROUP_METADATA,
3653 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3654 int i;
3655 int c;
3656 int num_tolerated_disk_barrier_failures =
3657 (int)fs_info->fs_devices->num_devices;
3658
3659 for (i = 0; i < ARRAY_SIZE(types); i++) {
3660 struct btrfs_space_info *tmp;
3661
3662 sinfo = NULL;
3663 rcu_read_lock();
3664 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3665 if (tmp->flags == types[i]) {
3666 sinfo = tmp;
3667 break;
3668 }
3669 }
3670 rcu_read_unlock();
3671
3672 if (!sinfo)
3673 continue;
3674
3675 down_read(&sinfo->groups_sem);
3676 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3677 u64 flags;
3678
3679 if (list_empty(&sinfo->block_groups[c]))
3680 continue;
3681
3682 btrfs_get_block_group_info(&sinfo->block_groups[c],
3683 &space);
3684 if (space.total_bytes == 0 || space.used_bytes == 0)
3685 continue;
3686 flags = space.flags;
3687
3688 num_tolerated_disk_barrier_failures = min(
3689 num_tolerated_disk_barrier_failures,
3690 btrfs_get_num_tolerated_disk_barrier_failures(
3691 flags));
3692 }
3693 up_read(&sinfo->groups_sem);
3694 }
3695
3696 return num_tolerated_disk_barrier_failures;
3697}
3698
3699static int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3700{
3701 struct list_head *head;
3702 struct btrfs_device *dev;
3703 struct btrfs_super_block *sb;
3704 struct btrfs_dev_item *dev_item;
3705 int ret;
3706 int do_barriers;
3707 int max_errors;
3708 int total_errors = 0;
3709 u64 flags;
3710
3711 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3712 backup_super_roots(fs_info);
3713
3714 sb = fs_info->super_for_commit;
3715 dev_item = &sb->dev_item;
3716
3717 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3718 head = &fs_info->fs_devices->devices;
3719 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3720
3721 if (do_barriers) {
3722 ret = barrier_all_devices(fs_info);
3723 if (ret) {
3724 mutex_unlock(
3725 &fs_info->fs_devices->device_list_mutex);
3726 btrfs_handle_fs_error(fs_info, ret,
3727 "errors while submitting device barriers.");
3728 return ret;
3729 }
3730 }
3731
3732 list_for_each_entry_rcu(dev, head, dev_list) {
3733 if (!dev->bdev) {
3734 total_errors++;
3735 continue;
3736 }
3737 if (!dev->in_fs_metadata || !dev->writeable)
3738 continue;
3739
3740 btrfs_set_stack_device_generation(dev_item, 0);
3741 btrfs_set_stack_device_type(dev_item, dev->type);
3742 btrfs_set_stack_device_id(dev_item, dev->devid);
3743 btrfs_set_stack_device_total_bytes(dev_item,
3744 dev->commit_total_bytes);
3745 btrfs_set_stack_device_bytes_used(dev_item,
3746 dev->commit_bytes_used);
3747 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3748 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3749 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3750 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3751 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3752
3753 flags = btrfs_super_flags(sb);
3754 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3755
3756 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3757 if (ret)
3758 total_errors++;
3759 }
3760 if (total_errors > max_errors) {
3761 btrfs_err(fs_info, "%d errors while writing supers",
3762 total_errors);
3763 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3764
3765 /* FUA is masked off if unsupported and can't be the reason */
3766 btrfs_handle_fs_error(fs_info, -EIO,
3767 "%d errors while writing supers",
3768 total_errors);
3769 return -EIO;
3770 }
3771
3772 total_errors = 0;
3773 list_for_each_entry_rcu(dev, head, dev_list) {
3774 if (!dev->bdev)
3775 continue;
3776 if (!dev->in_fs_metadata || !dev->writeable)
3777 continue;
3778
3779 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3780 if (ret)
3781 total_errors++;
3782 }
3783 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3784 if (total_errors > max_errors) {
3785 btrfs_handle_fs_error(fs_info, -EIO,
3786 "%d errors while writing supers",
3787 total_errors);
3788 return -EIO;
3789 }
3790 return 0;
3791}
3792
3793int write_ctree_super(struct btrfs_trans_handle *trans,
3794 struct btrfs_fs_info *fs_info, int max_mirrors)
3795{
3796 return write_all_supers(fs_info, max_mirrors);
3797}
3798
3799/* Drop a fs root from the radix tree and free it. */
3800void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3801 struct btrfs_root *root)
3802{
3803 spin_lock(&fs_info->fs_roots_radix_lock);
3804 radix_tree_delete(&fs_info->fs_roots_radix,
3805 (unsigned long)root->root_key.objectid);
3806 spin_unlock(&fs_info->fs_roots_radix_lock);
3807
3808 if (btrfs_root_refs(&root->root_item) == 0)
3809 synchronize_srcu(&fs_info->subvol_srcu);
3810
3811 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3812 btrfs_free_log(NULL, root);
3813 if (root->reloc_root) {
3814 free_extent_buffer(root->reloc_root->node);
3815 free_extent_buffer(root->reloc_root->commit_root);
3816 btrfs_put_fs_root(root->reloc_root);
3817 root->reloc_root = NULL;
3818 }
3819 }
3820
3821 if (root->free_ino_pinned)
3822 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3823 if (root->free_ino_ctl)
3824 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3825 free_fs_root(root);
3826}
3827
3828static void free_fs_root(struct btrfs_root *root)
3829{
3830 iput(root->ino_cache_inode);
3831 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3832 btrfs_free_block_rsv(root->fs_info, root->orphan_block_rsv);
3833 root->orphan_block_rsv = NULL;
3834 if (root->anon_dev)
3835 free_anon_bdev(root->anon_dev);
3836 if (root->subv_writers)
3837 btrfs_free_subvolume_writers(root->subv_writers);
3838 free_extent_buffer(root->node);
3839 free_extent_buffer(root->commit_root);
3840 kfree(root->free_ino_ctl);
3841 kfree(root->free_ino_pinned);
3842 kfree(root->name);
3843 btrfs_put_fs_root(root);
3844}
3845
3846void btrfs_free_fs_root(struct btrfs_root *root)
3847{
3848 free_fs_root(root);
3849}
3850
3851int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3852{
3853 u64 root_objectid = 0;
3854 struct btrfs_root *gang[8];
3855 int i = 0;
3856 int err = 0;
3857 unsigned int ret = 0;
3858 int index;
3859
3860 while (1) {
3861 index = srcu_read_lock(&fs_info->subvol_srcu);
3862 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3863 (void **)gang, root_objectid,
3864 ARRAY_SIZE(gang));
3865 if (!ret) {
3866 srcu_read_unlock(&fs_info->subvol_srcu, index);
3867 break;
3868 }
3869 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3870
3871 for (i = 0; i < ret; i++) {
3872 /* Avoid to grab roots in dead_roots */
3873 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3874 gang[i] = NULL;
3875 continue;
3876 }
3877 /* grab all the search result for later use */
3878 gang[i] = btrfs_grab_fs_root(gang[i]);
3879 }
3880 srcu_read_unlock(&fs_info->subvol_srcu, index);
3881
3882 for (i = 0; i < ret; i++) {
3883 if (!gang[i])
3884 continue;
3885 root_objectid = gang[i]->root_key.objectid;
3886 err = btrfs_orphan_cleanup(gang[i]);
3887 if (err)
3888 break;
3889 btrfs_put_fs_root(gang[i]);
3890 }
3891 root_objectid++;
3892 }
3893
3894 /* release the uncleaned roots due to error */
3895 for (; i < ret; i++) {
3896 if (gang[i])
3897 btrfs_put_fs_root(gang[i]);
3898 }
3899 return err;
3900}
3901
3902int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3903{
3904 struct btrfs_root *root = fs_info->tree_root;
3905 struct btrfs_trans_handle *trans;
3906
3907 mutex_lock(&fs_info->cleaner_mutex);
3908 btrfs_run_delayed_iputs(fs_info);
3909 mutex_unlock(&fs_info->cleaner_mutex);
3910 wake_up_process(fs_info->cleaner_kthread);
3911
3912 /* wait until ongoing cleanup work done */
3913 down_write(&fs_info->cleanup_work_sem);
3914 up_write(&fs_info->cleanup_work_sem);
3915
3916 trans = btrfs_join_transaction(root);
3917 if (IS_ERR(trans))
3918 return PTR_ERR(trans);
3919 return btrfs_commit_transaction(trans);
3920}
3921
3922void close_ctree(struct btrfs_fs_info *fs_info)
3923{
3924 struct btrfs_root *root = fs_info->tree_root;
3925 int ret;
3926
3927 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3928
3929 /* wait for the qgroup rescan worker to stop */
3930 btrfs_qgroup_wait_for_completion(fs_info, false);
3931
3932 /* wait for the uuid_scan task to finish */
3933 down(&fs_info->uuid_tree_rescan_sem);
3934 /* avoid complains from lockdep et al., set sem back to initial state */
3935 up(&fs_info->uuid_tree_rescan_sem);
3936
3937 /* pause restriper - we want to resume on mount */
3938 btrfs_pause_balance(fs_info);
3939
3940 btrfs_dev_replace_suspend_for_unmount(fs_info);
3941
3942 btrfs_scrub_cancel(fs_info);
3943
3944 /* wait for any defraggers to finish */
3945 wait_event(fs_info->transaction_wait,
3946 (atomic_read(&fs_info->defrag_running) == 0));
3947
3948 /* clear out the rbtree of defraggable inodes */
3949 btrfs_cleanup_defrag_inodes(fs_info);
3950
3951 cancel_work_sync(&fs_info->async_reclaim_work);
3952
3953 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3954 /*
3955 * If the cleaner thread is stopped and there are
3956 * block groups queued for removal, the deletion will be
3957 * skipped when we quit the cleaner thread.
3958 */
3959 btrfs_delete_unused_bgs(fs_info);
3960
3961 ret = btrfs_commit_super(fs_info);
3962 if (ret)
3963 btrfs_err(fs_info, "commit super ret %d", ret);
3964 }
3965
3966 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3967 btrfs_error_commit_super(fs_info);
3968
3969 kthread_stop(fs_info->transaction_kthread);
3970 kthread_stop(fs_info->cleaner_kthread);
3971
3972 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
3973
3974 btrfs_free_qgroup_config(fs_info);
3975
3976 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3977 btrfs_info(fs_info, "at unmount delalloc count %lld",
3978 percpu_counter_sum(&fs_info->delalloc_bytes));
3979 }
3980
3981 btrfs_sysfs_remove_mounted(fs_info);
3982 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3983
3984 btrfs_free_fs_roots(fs_info);
3985
3986 btrfs_put_block_group_cache(fs_info);
3987
3988 btrfs_free_block_groups(fs_info);
3989
3990 /*
3991 * we must make sure there is not any read request to
3992 * submit after we stopping all workers.
3993 */
3994 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3995 btrfs_stop_all_workers(fs_info);
3996
3997 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
3998 free_root_pointers(fs_info, 1);
3999
4000 iput(fs_info->btree_inode);
4001
4002#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4003 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4004 btrfsic_unmount(fs_info->fs_devices);
4005#endif
4006
4007 btrfs_close_devices(fs_info->fs_devices);
4008 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4009
4010 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4011 percpu_counter_destroy(&fs_info->delalloc_bytes);
4012 percpu_counter_destroy(&fs_info->bio_counter);
4013 bdi_destroy(&fs_info->bdi);
4014 cleanup_srcu_struct(&fs_info->subvol_srcu);
4015
4016 btrfs_free_stripe_hash_table(fs_info);
4017
4018 __btrfs_free_block_rsv(root->orphan_block_rsv);
4019 root->orphan_block_rsv = NULL;
4020
4021 mutex_lock(&fs_info->chunk_mutex);
4022 while (!list_empty(&fs_info->pinned_chunks)) {
4023 struct extent_map *em;
4024
4025 em = list_first_entry(&fs_info->pinned_chunks,
4026 struct extent_map, list);
4027 list_del_init(&em->list);
4028 free_extent_map(em);
4029 }
4030 mutex_unlock(&fs_info->chunk_mutex);
4031}
4032
4033int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4034 int atomic)
4035{
4036 int ret;
4037 struct inode *btree_inode = buf->pages[0]->mapping->host;
4038
4039 ret = extent_buffer_uptodate(buf);
4040 if (!ret)
4041 return ret;
4042
4043 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4044 parent_transid, atomic);
4045 if (ret == -EAGAIN)
4046 return ret;
4047 return !ret;
4048}
4049
4050void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4051{
4052 struct btrfs_fs_info *fs_info;
4053 struct btrfs_root *root;
4054 u64 transid = btrfs_header_generation(buf);
4055 int was_dirty;
4056
4057#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4058 /*
4059 * This is a fast path so only do this check if we have sanity tests
4060 * enabled. Normal people shouldn't be marking dummy buffers as dirty
4061 * outside of the sanity tests.
4062 */
4063 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
4064 return;
4065#endif
4066 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4067 fs_info = root->fs_info;
4068 btrfs_assert_tree_locked(buf);
4069 if (transid != fs_info->generation)
4070 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4071 buf->start, transid, fs_info->generation);
4072 was_dirty = set_extent_buffer_dirty(buf);
4073 if (!was_dirty)
4074 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
4075 buf->len,
4076 fs_info->dirty_metadata_batch);
4077#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4078 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
4079 btrfs_print_leaf(fs_info, buf);
4080 ASSERT(0);
4081 }
4082#endif
4083}
4084
4085static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4086 int flush_delayed)
4087{
4088 /*
4089 * looks as though older kernels can get into trouble with
4090 * this code, they end up stuck in balance_dirty_pages forever
4091 */
4092 int ret;
4093
4094 if (current->flags & PF_MEMALLOC)
4095 return;
4096
4097 if (flush_delayed)
4098 btrfs_balance_delayed_items(fs_info);
4099
4100 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4101 BTRFS_DIRTY_METADATA_THRESH);
4102 if (ret > 0) {
4103 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4104 }
4105}
4106
4107void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4108{
4109 __btrfs_btree_balance_dirty(fs_info, 1);
4110}
4111
4112void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4113{
4114 __btrfs_btree_balance_dirty(fs_info, 0);
4115}
4116
4117int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4118{
4119 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4120 struct btrfs_fs_info *fs_info = root->fs_info;
4121
4122 return btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
4123}
4124
4125static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
4126 int read_only)
4127{
4128 struct btrfs_super_block *sb = fs_info->super_copy;
4129 u64 nodesize = btrfs_super_nodesize(sb);
4130 u64 sectorsize = btrfs_super_sectorsize(sb);
4131 int ret = 0;
4132
4133 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4134 btrfs_err(fs_info, "no valid FS found");
4135 ret = -EINVAL;
4136 }
4137 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4138 btrfs_warn(fs_info, "unrecognized super flag: %llu",
4139 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4140 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4141 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
4142 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4143 ret = -EINVAL;
4144 }
4145 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4146 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
4147 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4148 ret = -EINVAL;
4149 }
4150 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4151 btrfs_err(fs_info, "log_root level too big: %d >= %d",
4152 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4153 ret = -EINVAL;
4154 }
4155
4156 /*
4157 * Check sectorsize and nodesize first, other check will need it.
4158 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4159 */
4160 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4161 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4162 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
4163 ret = -EINVAL;
4164 }
4165 /* Only PAGE SIZE is supported yet */
4166 if (sectorsize != PAGE_SIZE) {
4167 btrfs_err(fs_info,
4168 "sectorsize %llu not supported yet, only support %lu",
4169 sectorsize, PAGE_SIZE);
4170 ret = -EINVAL;
4171 }
4172 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4173 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4174 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
4175 ret = -EINVAL;
4176 }
4177 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4178 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
4179 le32_to_cpu(sb->__unused_leafsize), nodesize);
4180 ret = -EINVAL;
4181 }
4182
4183 /* Root alignment check */
4184 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4185 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
4186 btrfs_super_root(sb));
4187 ret = -EINVAL;
4188 }
4189 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4190 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
4191 btrfs_super_chunk_root(sb));
4192 ret = -EINVAL;
4193 }
4194 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4195 btrfs_warn(fs_info, "log_root block unaligned: %llu",
4196 btrfs_super_log_root(sb));
4197 ret = -EINVAL;
4198 }
4199
4200 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4201 btrfs_err(fs_info,
4202 "dev_item UUID does not match fsid: %pU != %pU",
4203 fs_info->fsid, sb->dev_item.fsid);
4204 ret = -EINVAL;
4205 }
4206
4207 /*
4208 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4209 * done later
4210 */
4211 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
4212 btrfs_err(fs_info, "bytes_used is too small %llu",
4213 btrfs_super_bytes_used(sb));
4214 ret = -EINVAL;
4215 }
4216 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
4217 btrfs_err(fs_info, "invalid stripesize %u",
4218 btrfs_super_stripesize(sb));
4219 ret = -EINVAL;
4220 }
4221 if (btrfs_super_num_devices(sb) > (1UL << 31))
4222 btrfs_warn(fs_info, "suspicious number of devices: %llu",
4223 btrfs_super_num_devices(sb));
4224 if (btrfs_super_num_devices(sb) == 0) {
4225 btrfs_err(fs_info, "number of devices is 0");
4226 ret = -EINVAL;
4227 }
4228
4229 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4230 btrfs_err(fs_info, "super offset mismatch %llu != %u",
4231 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4232 ret = -EINVAL;
4233 }
4234
4235 /*
4236 * Obvious sys_chunk_array corruptions, it must hold at least one key
4237 * and one chunk
4238 */
4239 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4240 btrfs_err(fs_info, "system chunk array too big %u > %u",
4241 btrfs_super_sys_array_size(sb),
4242 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4243 ret = -EINVAL;
4244 }
4245 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4246 + sizeof(struct btrfs_chunk)) {
4247 btrfs_err(fs_info, "system chunk array too small %u < %zu",
4248 btrfs_super_sys_array_size(sb),
4249 sizeof(struct btrfs_disk_key)
4250 + sizeof(struct btrfs_chunk));
4251 ret = -EINVAL;
4252 }
4253
4254 /*
4255 * The generation is a global counter, we'll trust it more than the others
4256 * but it's still possible that it's the one that's wrong.
4257 */
4258 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4259 btrfs_warn(fs_info,
4260 "suspicious: generation < chunk_root_generation: %llu < %llu",
4261 btrfs_super_generation(sb),
4262 btrfs_super_chunk_root_generation(sb));
4263 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4264 && btrfs_super_cache_generation(sb) != (u64)-1)
4265 btrfs_warn(fs_info,
4266 "suspicious: generation < cache_generation: %llu < %llu",
4267 btrfs_super_generation(sb),
4268 btrfs_super_cache_generation(sb));
4269
4270 return ret;
4271}
4272
4273static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4274{
4275 mutex_lock(&fs_info->cleaner_mutex);
4276 btrfs_run_delayed_iputs(fs_info);
4277 mutex_unlock(&fs_info->cleaner_mutex);
4278
4279 down_write(&fs_info->cleanup_work_sem);
4280 up_write(&fs_info->cleanup_work_sem);
4281
4282 /* cleanup FS via transaction */
4283 btrfs_cleanup_transaction(fs_info);
4284}
4285
4286static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4287{
4288 struct btrfs_ordered_extent *ordered;
4289
4290 spin_lock(&root->ordered_extent_lock);
4291 /*
4292 * This will just short circuit the ordered completion stuff which will
4293 * make sure the ordered extent gets properly cleaned up.
4294 */
4295 list_for_each_entry(ordered, &root->ordered_extents,
4296 root_extent_list)
4297 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4298 spin_unlock(&root->ordered_extent_lock);
4299}
4300
4301static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4302{
4303 struct btrfs_root *root;
4304 struct list_head splice;
4305
4306 INIT_LIST_HEAD(&splice);
4307
4308 spin_lock(&fs_info->ordered_root_lock);
4309 list_splice_init(&fs_info->ordered_roots, &splice);
4310 while (!list_empty(&splice)) {
4311 root = list_first_entry(&splice, struct btrfs_root,
4312 ordered_root);
4313 list_move_tail(&root->ordered_root,
4314 &fs_info->ordered_roots);
4315
4316 spin_unlock(&fs_info->ordered_root_lock);
4317 btrfs_destroy_ordered_extents(root);
4318
4319 cond_resched();
4320 spin_lock(&fs_info->ordered_root_lock);
4321 }
4322 spin_unlock(&fs_info->ordered_root_lock);
4323}
4324
4325static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4326 struct btrfs_fs_info *fs_info)
4327{
4328 struct rb_node *node;
4329 struct btrfs_delayed_ref_root *delayed_refs;
4330 struct btrfs_delayed_ref_node *ref;
4331 int ret = 0;
4332
4333 delayed_refs = &trans->delayed_refs;
4334
4335 spin_lock(&delayed_refs->lock);
4336 if (atomic_read(&delayed_refs->num_entries) == 0) {
4337 spin_unlock(&delayed_refs->lock);
4338 btrfs_info(fs_info, "delayed_refs has NO entry");
4339 return ret;
4340 }
4341
4342 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4343 struct btrfs_delayed_ref_head *head;
4344 struct btrfs_delayed_ref_node *tmp;
4345 bool pin_bytes = false;
4346
4347 head = rb_entry(node, struct btrfs_delayed_ref_head,
4348 href_node);
4349 if (!mutex_trylock(&head->mutex)) {
4350 atomic_inc(&head->node.refs);
4351 spin_unlock(&delayed_refs->lock);
4352
4353 mutex_lock(&head->mutex);
4354 mutex_unlock(&head->mutex);
4355 btrfs_put_delayed_ref(&head->node);
4356 spin_lock(&delayed_refs->lock);
4357 continue;
4358 }
4359 spin_lock(&head->lock);
4360 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4361 list) {
4362 ref->in_tree = 0;
4363 list_del(&ref->list);
4364 if (!list_empty(&ref->add_list))
4365 list_del(&ref->add_list);
4366 atomic_dec(&delayed_refs->num_entries);
4367 btrfs_put_delayed_ref(ref);
4368 }
4369 if (head->must_insert_reserved)
4370 pin_bytes = true;
4371 btrfs_free_delayed_extent_op(head->extent_op);
4372 delayed_refs->num_heads--;
4373 if (head->processing == 0)
4374 delayed_refs->num_heads_ready--;
4375 atomic_dec(&delayed_refs->num_entries);
4376 head->node.in_tree = 0;
4377 rb_erase(&head->href_node, &delayed_refs->href_root);
4378 spin_unlock(&head->lock);
4379 spin_unlock(&delayed_refs->lock);
4380 mutex_unlock(&head->mutex);
4381
4382 if (pin_bytes)
4383 btrfs_pin_extent(fs_info, head->node.bytenr,
4384 head->node.num_bytes, 1);
4385 btrfs_put_delayed_ref(&head->node);
4386 cond_resched();
4387 spin_lock(&delayed_refs->lock);
4388 }
4389
4390 spin_unlock(&delayed_refs->lock);
4391
4392 return ret;
4393}
4394
4395static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4396{
4397 struct btrfs_inode *btrfs_inode;
4398 struct list_head splice;
4399
4400 INIT_LIST_HEAD(&splice);
4401
4402 spin_lock(&root->delalloc_lock);
4403 list_splice_init(&root->delalloc_inodes, &splice);
4404
4405 while (!list_empty(&splice)) {
4406 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4407 delalloc_inodes);
4408
4409 list_del_init(&btrfs_inode->delalloc_inodes);
4410 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4411 &btrfs_inode->runtime_flags);
4412 spin_unlock(&root->delalloc_lock);
4413
4414 btrfs_invalidate_inodes(btrfs_inode->root);
4415
4416 spin_lock(&root->delalloc_lock);
4417 }
4418
4419 spin_unlock(&root->delalloc_lock);
4420}
4421
4422static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4423{
4424 struct btrfs_root *root;
4425 struct list_head splice;
4426
4427 INIT_LIST_HEAD(&splice);
4428
4429 spin_lock(&fs_info->delalloc_root_lock);
4430 list_splice_init(&fs_info->delalloc_roots, &splice);
4431 while (!list_empty(&splice)) {
4432 root = list_first_entry(&splice, struct btrfs_root,
4433 delalloc_root);
4434 list_del_init(&root->delalloc_root);
4435 root = btrfs_grab_fs_root(root);
4436 BUG_ON(!root);
4437 spin_unlock(&fs_info->delalloc_root_lock);
4438
4439 btrfs_destroy_delalloc_inodes(root);
4440 btrfs_put_fs_root(root);
4441
4442 spin_lock(&fs_info->delalloc_root_lock);
4443 }
4444 spin_unlock(&fs_info->delalloc_root_lock);
4445}
4446
4447static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4448 struct extent_io_tree *dirty_pages,
4449 int mark)
4450{
4451 int ret;
4452 struct extent_buffer *eb;
4453 u64 start = 0;
4454 u64 end;
4455
4456 while (1) {
4457 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4458 mark, NULL);
4459 if (ret)
4460 break;
4461
4462 clear_extent_bits(dirty_pages, start, end, mark);
4463 while (start <= end) {
4464 eb = find_extent_buffer(fs_info, start);
4465 start += fs_info->nodesize;
4466 if (!eb)
4467 continue;
4468 wait_on_extent_buffer_writeback(eb);
4469
4470 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4471 &eb->bflags))
4472 clear_extent_buffer_dirty(eb);
4473 free_extent_buffer_stale(eb);
4474 }
4475 }
4476
4477 return ret;
4478}
4479
4480static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4481 struct extent_io_tree *pinned_extents)
4482{
4483 struct extent_io_tree *unpin;
4484 u64 start;
4485 u64 end;
4486 int ret;
4487 bool loop = true;
4488
4489 unpin = pinned_extents;
4490again:
4491 while (1) {
4492 ret = find_first_extent_bit(unpin, 0, &start, &end,
4493 EXTENT_DIRTY, NULL);
4494 if (ret)
4495 break;
4496
4497 clear_extent_dirty(unpin, start, end);
4498 btrfs_error_unpin_extent_range(fs_info, start, end);
4499 cond_resched();
4500 }
4501
4502 if (loop) {
4503 if (unpin == &fs_info->freed_extents[0])
4504 unpin = &fs_info->freed_extents[1];
4505 else
4506 unpin = &fs_info->freed_extents[0];
4507 loop = false;
4508 goto again;
4509 }
4510
4511 return 0;
4512}
4513
4514static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4515{
4516 struct inode *inode;
4517
4518 inode = cache->io_ctl.inode;
4519 if (inode) {
4520 invalidate_inode_pages2(inode->i_mapping);
4521 BTRFS_I(inode)->generation = 0;
4522 cache->io_ctl.inode = NULL;
4523 iput(inode);
4524 }
4525 btrfs_put_block_group(cache);
4526}
4527
4528void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4529 struct btrfs_fs_info *fs_info)
4530{
4531 struct btrfs_block_group_cache *cache;
4532
4533 spin_lock(&cur_trans->dirty_bgs_lock);
4534 while (!list_empty(&cur_trans->dirty_bgs)) {
4535 cache = list_first_entry(&cur_trans->dirty_bgs,
4536 struct btrfs_block_group_cache,
4537 dirty_list);
4538 if (!cache) {
4539 btrfs_err(fs_info, "orphan block group dirty_bgs list");
4540 spin_unlock(&cur_trans->dirty_bgs_lock);
4541 return;
4542 }
4543
4544 if (!list_empty(&cache->io_list)) {
4545 spin_unlock(&cur_trans->dirty_bgs_lock);
4546 list_del_init(&cache->io_list);
4547 btrfs_cleanup_bg_io(cache);
4548 spin_lock(&cur_trans->dirty_bgs_lock);
4549 }
4550
4551 list_del_init(&cache->dirty_list);
4552 spin_lock(&cache->lock);
4553 cache->disk_cache_state = BTRFS_DC_ERROR;
4554 spin_unlock(&cache->lock);
4555
4556 spin_unlock(&cur_trans->dirty_bgs_lock);
4557 btrfs_put_block_group(cache);
4558 spin_lock(&cur_trans->dirty_bgs_lock);
4559 }
4560 spin_unlock(&cur_trans->dirty_bgs_lock);
4561
4562 while (!list_empty(&cur_trans->io_bgs)) {
4563 cache = list_first_entry(&cur_trans->io_bgs,
4564 struct btrfs_block_group_cache,
4565 io_list);
4566 if (!cache) {
4567 btrfs_err(fs_info, "orphan block group on io_bgs list");
4568 return;
4569 }
4570
4571 list_del_init(&cache->io_list);
4572 spin_lock(&cache->lock);
4573 cache->disk_cache_state = BTRFS_DC_ERROR;
4574 spin_unlock(&cache->lock);
4575 btrfs_cleanup_bg_io(cache);
4576 }
4577}
4578
4579void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4580 struct btrfs_fs_info *fs_info)
4581{
4582 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4583 ASSERT(list_empty(&cur_trans->dirty_bgs));
4584 ASSERT(list_empty(&cur_trans->io_bgs));
4585
4586 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4587
4588 cur_trans->state = TRANS_STATE_COMMIT_START;
4589 wake_up(&fs_info->transaction_blocked_wait);
4590
4591 cur_trans->state = TRANS_STATE_UNBLOCKED;
4592 wake_up(&fs_info->transaction_wait);
4593
4594 btrfs_destroy_delayed_inodes(fs_info);
4595 btrfs_assert_delayed_root_empty(fs_info);
4596
4597 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4598 EXTENT_DIRTY);
4599 btrfs_destroy_pinned_extent(fs_info,
4600 fs_info->pinned_extents);
4601
4602 cur_trans->state =TRANS_STATE_COMPLETED;
4603 wake_up(&cur_trans->commit_wait);
4604
4605 /*
4606 memset(cur_trans, 0, sizeof(*cur_trans));
4607 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4608 */
4609}
4610
4611static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4612{
4613 struct btrfs_transaction *t;
4614
4615 mutex_lock(&fs_info->transaction_kthread_mutex);
4616
4617 spin_lock(&fs_info->trans_lock);
4618 while (!list_empty(&fs_info->trans_list)) {
4619 t = list_first_entry(&fs_info->trans_list,
4620 struct btrfs_transaction, list);
4621 if (t->state >= TRANS_STATE_COMMIT_START) {
4622 atomic_inc(&t->use_count);
4623 spin_unlock(&fs_info->trans_lock);
4624 btrfs_wait_for_commit(fs_info, t->transid);
4625 btrfs_put_transaction(t);
4626 spin_lock(&fs_info->trans_lock);
4627 continue;
4628 }
4629 if (t == fs_info->running_transaction) {
4630 t->state = TRANS_STATE_COMMIT_DOING;
4631 spin_unlock(&fs_info->trans_lock);
4632 /*
4633 * We wait for 0 num_writers since we don't hold a trans
4634 * handle open currently for this transaction.
4635 */
4636 wait_event(t->writer_wait,
4637 atomic_read(&t->num_writers) == 0);
4638 } else {
4639 spin_unlock(&fs_info->trans_lock);
4640 }
4641 btrfs_cleanup_one_transaction(t, fs_info);
4642
4643 spin_lock(&fs_info->trans_lock);
4644 if (t == fs_info->running_transaction)
4645 fs_info->running_transaction = NULL;
4646 list_del_init(&t->list);
4647 spin_unlock(&fs_info->trans_lock);
4648
4649 btrfs_put_transaction(t);
4650 trace_btrfs_transaction_commit(fs_info->tree_root);
4651 spin_lock(&fs_info->trans_lock);
4652 }
4653 spin_unlock(&fs_info->trans_lock);
4654 btrfs_destroy_all_ordered_extents(fs_info);
4655 btrfs_destroy_delayed_inodes(fs_info);
4656 btrfs_assert_delayed_root_empty(fs_info);
4657 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4658 btrfs_destroy_all_delalloc_inodes(fs_info);
4659 mutex_unlock(&fs_info->transaction_kthread_mutex);
4660
4661 return 0;
4662}
4663
4664static const struct extent_io_ops btree_extent_io_ops = {
4665 .readpage_end_io_hook = btree_readpage_end_io_hook,
4666 .readpage_io_failed_hook = btree_io_failed_hook,
4667 .submit_bio_hook = btree_submit_bio_hook,
4668 /* note we're sharing with inode.c for the merge bio hook */
4669 .merge_bio_hook = btrfs_merge_bio_hook,
4670};