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
54 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
55 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
56 struct btrfs_fs_info *fs_info);
57 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
58 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
59 struct extent_io_tree *dirty_pages,
60 int mark);
61 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
62 struct extent_io_tree *pinned_extents);
63 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
64 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
65
btrfs_free_csum_hash(struct btrfs_fs_info * fs_info)66 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
67 {
68 if (fs_info->csum_shash)
69 crypto_free_shash(fs_info->csum_shash);
70 }
71
72 /*
73 * async submit bios are used to offload expensive checksumming
74 * onto the worker threads. They checksum file and metadata bios
75 * just before they are sent down the IO stack.
76 */
77 struct async_submit_bio {
78 struct inode *inode;
79 struct bio *bio;
80 extent_submit_bio_start_t *submit_bio_start;
81 int mirror_num;
82
83 /* Optional parameter for submit_bio_start used by direct io */
84 u64 dio_file_offset;
85 struct btrfs_work work;
86 blk_status_t status;
87 };
88
89 /*
90 * Compute the csum of a btree block and store the result to provided buffer.
91 */
csum_tree_block(struct extent_buffer * buf,u8 * result)92 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
93 {
94 struct btrfs_fs_info *fs_info = buf->fs_info;
95 const int num_pages = num_extent_pages(buf);
96 const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
97 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
98 char *kaddr;
99 int i;
100
101 shash->tfm = fs_info->csum_shash;
102 crypto_shash_init(shash);
103 kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
104 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
105 first_page_part - BTRFS_CSUM_SIZE);
106
107 for (i = 1; i < num_pages; i++) {
108 kaddr = page_address(buf->pages[i]);
109 crypto_shash_update(shash, kaddr, PAGE_SIZE);
110 }
111 memset(result, 0, BTRFS_CSUM_SIZE);
112 crypto_shash_final(shash, result);
113 }
114
115 /*
116 * we can't consider a given block up to date unless the transid of the
117 * block matches the transid in the parent node's pointer. This is how we
118 * detect blocks that either didn't get written at all or got written
119 * in the wrong place.
120 */
verify_parent_transid(struct extent_io_tree * io_tree,struct extent_buffer * eb,u64 parent_transid,int atomic)121 static int verify_parent_transid(struct extent_io_tree *io_tree,
122 struct extent_buffer *eb, u64 parent_transid,
123 int atomic)
124 {
125 struct extent_state *cached_state = NULL;
126 int ret;
127
128 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
129 return 0;
130
131 if (atomic)
132 return -EAGAIN;
133
134 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, &cached_state);
135 if (extent_buffer_uptodate(eb) &&
136 btrfs_header_generation(eb) == parent_transid) {
137 ret = 0;
138 goto out;
139 }
140 btrfs_err_rl(eb->fs_info,
141 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
142 eb->start, eb->read_mirror,
143 parent_transid, btrfs_header_generation(eb));
144 ret = 1;
145 clear_extent_buffer_uptodate(eb);
146 out:
147 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
148 &cached_state);
149 return ret;
150 }
151
btrfs_supported_super_csum(u16 csum_type)152 static bool btrfs_supported_super_csum(u16 csum_type)
153 {
154 switch (csum_type) {
155 case BTRFS_CSUM_TYPE_CRC32:
156 case BTRFS_CSUM_TYPE_XXHASH:
157 case BTRFS_CSUM_TYPE_SHA256:
158 case BTRFS_CSUM_TYPE_BLAKE2:
159 return true;
160 default:
161 return false;
162 }
163 }
164
165 /*
166 * Return 0 if the superblock checksum type matches the checksum value of that
167 * algorithm. Pass the raw disk superblock data.
168 */
btrfs_check_super_csum(struct btrfs_fs_info * fs_info,const struct btrfs_super_block * disk_sb)169 int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
170 const struct btrfs_super_block *disk_sb)
171 {
172 char result[BTRFS_CSUM_SIZE];
173 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
174
175 shash->tfm = fs_info->csum_shash;
176
177 /*
178 * The super_block structure does not span the whole
179 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
180 * filled with zeros and is included in the checksum.
181 */
182 crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
183 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
184
185 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
186 return 1;
187
188 return 0;
189 }
190
btrfs_verify_level_key(struct extent_buffer * eb,int level,struct btrfs_key * first_key,u64 parent_transid)191 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
192 struct btrfs_key *first_key, u64 parent_transid)
193 {
194 struct btrfs_fs_info *fs_info = eb->fs_info;
195 int found_level;
196 struct btrfs_key found_key;
197 int ret;
198
199 found_level = btrfs_header_level(eb);
200 if (found_level != level) {
201 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
202 KERN_ERR "BTRFS: tree level check failed\n");
203 btrfs_err(fs_info,
204 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
205 eb->start, level, found_level);
206 return -EIO;
207 }
208
209 if (!first_key)
210 return 0;
211
212 /*
213 * For live tree block (new tree blocks in current transaction),
214 * we need proper lock context to avoid race, which is impossible here.
215 * So we only checks tree blocks which is read from disk, whose
216 * generation <= fs_info->last_trans_committed.
217 */
218 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
219 return 0;
220
221 /* We have @first_key, so this @eb must have at least one item */
222 if (btrfs_header_nritems(eb) == 0) {
223 btrfs_err(fs_info,
224 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
225 eb->start);
226 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
227 return -EUCLEAN;
228 }
229
230 if (found_level)
231 btrfs_node_key_to_cpu(eb, &found_key, 0);
232 else
233 btrfs_item_key_to_cpu(eb, &found_key, 0);
234 ret = btrfs_comp_cpu_keys(first_key, &found_key);
235
236 if (ret) {
237 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
238 KERN_ERR "BTRFS: tree first key check failed\n");
239 btrfs_err(fs_info,
240 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
241 eb->start, parent_transid, first_key->objectid,
242 first_key->type, first_key->offset,
243 found_key.objectid, found_key.type,
244 found_key.offset);
245 }
246 return ret;
247 }
248
249 /*
250 * helper to read a given tree block, doing retries as required when
251 * the checksums don't match and we have alternate mirrors to try.
252 *
253 * @parent_transid: expected transid, skip check if 0
254 * @level: expected level, mandatory check
255 * @first_key: expected key of first slot, skip check if NULL
256 */
btrfs_read_extent_buffer(struct extent_buffer * eb,u64 parent_transid,int level,struct btrfs_key * first_key)257 int btrfs_read_extent_buffer(struct extent_buffer *eb,
258 u64 parent_transid, int level,
259 struct btrfs_key *first_key)
260 {
261 struct btrfs_fs_info *fs_info = eb->fs_info;
262 struct extent_io_tree *io_tree;
263 int failed = 0;
264 int ret;
265 int num_copies = 0;
266 int mirror_num = 0;
267 int failed_mirror = 0;
268
269 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
270 while (1) {
271 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
272 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
273 if (!ret) {
274 if (verify_parent_transid(io_tree, eb,
275 parent_transid, 0))
276 ret = -EIO;
277 else if (btrfs_verify_level_key(eb, level,
278 first_key, parent_transid))
279 ret = -EUCLEAN;
280 else
281 break;
282 }
283
284 num_copies = btrfs_num_copies(fs_info,
285 eb->start, eb->len);
286 if (num_copies == 1)
287 break;
288
289 if (!failed_mirror) {
290 failed = 1;
291 failed_mirror = eb->read_mirror;
292 }
293
294 mirror_num++;
295 if (mirror_num == failed_mirror)
296 mirror_num++;
297
298 if (mirror_num > num_copies)
299 break;
300 }
301
302 if (failed && !ret && failed_mirror)
303 btrfs_repair_eb_io_failure(eb, failed_mirror);
304
305 return ret;
306 }
307
csum_one_extent_buffer(struct extent_buffer * eb)308 static int csum_one_extent_buffer(struct extent_buffer *eb)
309 {
310 struct btrfs_fs_info *fs_info = eb->fs_info;
311 u8 result[BTRFS_CSUM_SIZE];
312 int ret;
313
314 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
315 offsetof(struct btrfs_header, fsid),
316 BTRFS_FSID_SIZE) == 0);
317 csum_tree_block(eb, result);
318
319 if (btrfs_header_level(eb))
320 ret = btrfs_check_node(eb);
321 else
322 ret = btrfs_check_leaf_full(eb);
323
324 if (ret < 0)
325 goto error;
326
327 /*
328 * Also check the generation, the eb reached here must be newer than
329 * last committed. Or something seriously wrong happened.
330 */
331 if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) {
332 ret = -EUCLEAN;
333 btrfs_err(fs_info,
334 "block=%llu bad generation, have %llu expect > %llu",
335 eb->start, btrfs_header_generation(eb),
336 fs_info->last_trans_committed);
337 goto error;
338 }
339 write_extent_buffer(eb, result, 0, fs_info->csum_size);
340
341 return 0;
342
343 error:
344 btrfs_print_tree(eb, 0);
345 btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
346 eb->start);
347 /*
348 * Be noisy if this is an extent buffer from a log tree. We don't abort
349 * a transaction in case there's a bad log tree extent buffer, we just
350 * fallback to a transaction commit. Still we want to know when there is
351 * a bad log tree extent buffer, as that may signal a bug somewhere.
352 */
353 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
354 btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
355 return ret;
356 }
357
358 /* Checksum all dirty extent buffers in one bio_vec */
csum_dirty_subpage_buffers(struct btrfs_fs_info * fs_info,struct bio_vec * bvec)359 static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info,
360 struct bio_vec *bvec)
361 {
362 struct page *page = bvec->bv_page;
363 u64 bvec_start = page_offset(page) + bvec->bv_offset;
364 u64 cur;
365 int ret = 0;
366
367 for (cur = bvec_start; cur < bvec_start + bvec->bv_len;
368 cur += fs_info->nodesize) {
369 struct extent_buffer *eb;
370 bool uptodate;
371
372 eb = find_extent_buffer(fs_info, cur);
373 uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur,
374 fs_info->nodesize);
375
376 /* A dirty eb shouldn't disappear from buffer_radix */
377 if (WARN_ON(!eb))
378 return -EUCLEAN;
379
380 if (WARN_ON(cur != btrfs_header_bytenr(eb))) {
381 free_extent_buffer(eb);
382 return -EUCLEAN;
383 }
384 if (WARN_ON(!uptodate)) {
385 free_extent_buffer(eb);
386 return -EUCLEAN;
387 }
388
389 ret = csum_one_extent_buffer(eb);
390 free_extent_buffer(eb);
391 if (ret < 0)
392 return ret;
393 }
394 return ret;
395 }
396
397 /*
398 * Checksum a dirty tree block before IO. This has extra checks to make sure
399 * we only fill in the checksum field in the first page of a multi-page block.
400 * For subpage extent buffers we need bvec to also read the offset in the page.
401 */
csum_dirty_buffer(struct btrfs_fs_info * fs_info,struct bio_vec * bvec)402 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
403 {
404 struct page *page = bvec->bv_page;
405 u64 start = page_offset(page);
406 u64 found_start;
407 struct extent_buffer *eb;
408
409 if (fs_info->nodesize < PAGE_SIZE)
410 return csum_dirty_subpage_buffers(fs_info, bvec);
411
412 eb = (struct extent_buffer *)page->private;
413 if (page != eb->pages[0])
414 return 0;
415
416 found_start = btrfs_header_bytenr(eb);
417
418 if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
419 WARN_ON(found_start != 0);
420 return 0;
421 }
422
423 /*
424 * Please do not consolidate these warnings into a single if.
425 * It is useful to know what went wrong.
426 */
427 if (WARN_ON(found_start != start))
428 return -EUCLEAN;
429 if (WARN_ON(!PageUptodate(page)))
430 return -EUCLEAN;
431
432 return csum_one_extent_buffer(eb);
433 }
434
check_tree_block_fsid(struct extent_buffer * eb)435 static int check_tree_block_fsid(struct extent_buffer *eb)
436 {
437 struct btrfs_fs_info *fs_info = eb->fs_info;
438 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
439 u8 fsid[BTRFS_FSID_SIZE];
440 u8 *metadata_uuid;
441
442 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
443 BTRFS_FSID_SIZE);
444 /*
445 * Checking the incompat flag is only valid for the current fs. For
446 * seed devices it's forbidden to have their uuid changed so reading
447 * ->fsid in this case is fine
448 */
449 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
450 metadata_uuid = fs_devices->metadata_uuid;
451 else
452 metadata_uuid = fs_devices->fsid;
453
454 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
455 return 0;
456
457 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
458 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
459 return 0;
460
461 return 1;
462 }
463
464 /* Do basic extent buffer checks at read time */
validate_extent_buffer(struct extent_buffer * eb)465 static int validate_extent_buffer(struct extent_buffer *eb)
466 {
467 struct btrfs_fs_info *fs_info = eb->fs_info;
468 u64 found_start;
469 const u32 csum_size = fs_info->csum_size;
470 u8 found_level;
471 u8 result[BTRFS_CSUM_SIZE];
472 const u8 *header_csum;
473 int ret = 0;
474
475 found_start = btrfs_header_bytenr(eb);
476 if (found_start != eb->start) {
477 btrfs_err_rl(fs_info,
478 "bad tree block start, mirror %u want %llu have %llu",
479 eb->read_mirror, eb->start, found_start);
480 ret = -EIO;
481 goto out;
482 }
483 if (check_tree_block_fsid(eb)) {
484 btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
485 eb->start, eb->read_mirror);
486 ret = -EIO;
487 goto out;
488 }
489 found_level = btrfs_header_level(eb);
490 if (found_level >= BTRFS_MAX_LEVEL) {
491 btrfs_err(fs_info,
492 "bad tree block level, mirror %u level %d on logical %llu",
493 eb->read_mirror, btrfs_header_level(eb), eb->start);
494 ret = -EIO;
495 goto out;
496 }
497
498 csum_tree_block(eb, result);
499 header_csum = page_address(eb->pages[0]) +
500 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
501
502 if (memcmp(result, header_csum, csum_size) != 0) {
503 btrfs_warn_rl(fs_info,
504 "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d",
505 eb->start, eb->read_mirror,
506 CSUM_FMT_VALUE(csum_size, header_csum),
507 CSUM_FMT_VALUE(csum_size, result),
508 btrfs_header_level(eb));
509 ret = -EUCLEAN;
510 goto out;
511 }
512
513 /*
514 * If this is a leaf block and it is corrupt, set the corrupt bit so
515 * that we don't try and read the other copies of this block, just
516 * return -EIO.
517 */
518 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
519 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
520 ret = -EIO;
521 }
522
523 if (found_level > 0 && btrfs_check_node(eb))
524 ret = -EIO;
525
526 if (!ret)
527 set_extent_buffer_uptodate(eb);
528 else
529 btrfs_err(fs_info,
530 "read time tree block corruption detected on logical %llu mirror %u",
531 eb->start, eb->read_mirror);
532 out:
533 return ret;
534 }
535
validate_subpage_buffer(struct page * page,u64 start,u64 end,int mirror)536 static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
537 int mirror)
538 {
539 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
540 struct extent_buffer *eb;
541 bool reads_done;
542 int ret = 0;
543
544 /*
545 * We don't allow bio merge for subpage metadata read, so we should
546 * only get one eb for each endio hook.
547 */
548 ASSERT(end == start + fs_info->nodesize - 1);
549 ASSERT(PagePrivate(page));
550
551 eb = find_extent_buffer(fs_info, start);
552 /*
553 * When we are reading one tree block, eb must have been inserted into
554 * the radix tree. If not, something is wrong.
555 */
556 ASSERT(eb);
557
558 reads_done = atomic_dec_and_test(&eb->io_pages);
559 /* Subpage read must finish in page read */
560 ASSERT(reads_done);
561
562 eb->read_mirror = mirror;
563 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
564 ret = -EIO;
565 goto err;
566 }
567 ret = validate_extent_buffer(eb);
568 if (ret < 0)
569 goto err;
570
571 set_extent_buffer_uptodate(eb);
572
573 free_extent_buffer(eb);
574 return ret;
575 err:
576 /*
577 * end_bio_extent_readpage decrements io_pages in case of error,
578 * make sure it has something to decrement.
579 */
580 atomic_inc(&eb->io_pages);
581 clear_extent_buffer_uptodate(eb);
582 free_extent_buffer(eb);
583 return ret;
584 }
585
btrfs_validate_metadata_buffer(struct btrfs_bio * bbio,struct page * page,u64 start,u64 end,int mirror)586 int btrfs_validate_metadata_buffer(struct btrfs_bio *bbio,
587 struct page *page, u64 start, u64 end,
588 int mirror)
589 {
590 struct extent_buffer *eb;
591 int ret = 0;
592 int reads_done;
593
594 ASSERT(page->private);
595
596 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
597 return validate_subpage_buffer(page, start, end, mirror);
598
599 eb = (struct extent_buffer *)page->private;
600
601 /*
602 * The pending IO might have been the only thing that kept this buffer
603 * in memory. Make sure we have a ref for all this other checks
604 */
605 atomic_inc(&eb->refs);
606
607 reads_done = atomic_dec_and_test(&eb->io_pages);
608 if (!reads_done)
609 goto err;
610
611 eb->read_mirror = mirror;
612 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
613 ret = -EIO;
614 goto err;
615 }
616 ret = validate_extent_buffer(eb);
617 err:
618 if (ret) {
619 /*
620 * our io error hook is going to dec the io pages
621 * again, we have to make sure it has something
622 * to decrement
623 */
624 atomic_inc(&eb->io_pages);
625 clear_extent_buffer_uptodate(eb);
626 }
627 free_extent_buffer(eb);
628
629 return ret;
630 }
631
run_one_async_start(struct btrfs_work * work)632 static void run_one_async_start(struct btrfs_work *work)
633 {
634 struct async_submit_bio *async;
635 blk_status_t ret;
636
637 async = container_of(work, struct async_submit_bio, work);
638 ret = async->submit_bio_start(async->inode, async->bio,
639 async->dio_file_offset);
640 if (ret)
641 async->status = ret;
642 }
643
644 /*
645 * In order to insert checksums into the metadata in large chunks, we wait
646 * until bio submission time. All the pages in the bio are checksummed and
647 * sums are attached onto the ordered extent record.
648 *
649 * At IO completion time the csums attached on the ordered extent record are
650 * inserted into the tree.
651 */
run_one_async_done(struct btrfs_work * work)652 static void run_one_async_done(struct btrfs_work *work)
653 {
654 struct async_submit_bio *async =
655 container_of(work, struct async_submit_bio, work);
656 struct inode *inode = async->inode;
657 struct btrfs_bio *bbio = btrfs_bio(async->bio);
658
659 /* If an error occurred we just want to clean up the bio and move on */
660 if (async->status) {
661 btrfs_bio_end_io(bbio, async->status);
662 return;
663 }
664
665 /*
666 * All of the bios that pass through here are from async helpers.
667 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
668 * This changes nothing when cgroups aren't in use.
669 */
670 async->bio->bi_opf |= REQ_CGROUP_PUNT;
671 btrfs_submit_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
672 }
673
run_one_async_free(struct btrfs_work * work)674 static void run_one_async_free(struct btrfs_work *work)
675 {
676 struct async_submit_bio *async;
677
678 async = container_of(work, struct async_submit_bio, work);
679 kfree(async);
680 }
681
682 /*
683 * Submit bio to an async queue.
684 *
685 * Retrun:
686 * - true if the work has been succesfuly submitted
687 * - false in case of error
688 */
btrfs_wq_submit_bio(struct inode * inode,struct bio * bio,int mirror_num,u64 dio_file_offset,extent_submit_bio_start_t * submit_bio_start)689 bool btrfs_wq_submit_bio(struct inode *inode, struct bio *bio, int mirror_num,
690 u64 dio_file_offset,
691 extent_submit_bio_start_t *submit_bio_start)
692 {
693 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
694 struct async_submit_bio *async;
695
696 async = kmalloc(sizeof(*async), GFP_NOFS);
697 if (!async)
698 return false;
699
700 async->inode = inode;
701 async->bio = bio;
702 async->mirror_num = mirror_num;
703 async->submit_bio_start = submit_bio_start;
704
705 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
706 run_one_async_free);
707
708 async->dio_file_offset = dio_file_offset;
709
710 async->status = 0;
711
712 if (op_is_sync(bio->bi_opf))
713 btrfs_queue_work(fs_info->hipri_workers, &async->work);
714 else
715 btrfs_queue_work(fs_info->workers, &async->work);
716 return true;
717 }
718
btree_csum_one_bio(struct bio * bio)719 static blk_status_t btree_csum_one_bio(struct bio *bio)
720 {
721 struct bio_vec *bvec;
722 struct btrfs_root *root;
723 int ret = 0;
724 struct bvec_iter_all iter_all;
725
726 ASSERT(!bio_flagged(bio, BIO_CLONED));
727 bio_for_each_segment_all(bvec, bio, iter_all) {
728 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
729 ret = csum_dirty_buffer(root->fs_info, bvec);
730 if (ret)
731 break;
732 }
733
734 return errno_to_blk_status(ret);
735 }
736
btree_submit_bio_start(struct inode * inode,struct bio * bio,u64 dio_file_offset)737 static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio,
738 u64 dio_file_offset)
739 {
740 /*
741 * when we're called for a write, we're already in the async
742 * submission context. Just jump into btrfs_submit_bio.
743 */
744 return btree_csum_one_bio(bio);
745 }
746
should_async_write(struct btrfs_fs_info * fs_info,struct btrfs_inode * bi)747 static bool should_async_write(struct btrfs_fs_info *fs_info,
748 struct btrfs_inode *bi)
749 {
750 if (btrfs_is_zoned(fs_info))
751 return false;
752 if (atomic_read(&bi->sync_writers))
753 return false;
754 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
755 return false;
756 return true;
757 }
758
btrfs_submit_metadata_bio(struct inode * inode,struct bio * bio,int mirror_num)759 void btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio, int mirror_num)
760 {
761 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
762 struct btrfs_bio *bbio = btrfs_bio(bio);
763 blk_status_t ret;
764
765 bio->bi_opf |= REQ_META;
766
767 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
768 btrfs_submit_bio(fs_info, bio, mirror_num);
769 return;
770 }
771
772 /*
773 * Kthread helpers are used to submit writes so that checksumming can
774 * happen in parallel across all CPUs.
775 */
776 if (should_async_write(fs_info, BTRFS_I(inode)) &&
777 btrfs_wq_submit_bio(inode, bio, mirror_num, 0, btree_submit_bio_start))
778 return;
779
780 ret = btree_csum_one_bio(bio);
781 if (ret) {
782 btrfs_bio_end_io(bbio, ret);
783 return;
784 }
785
786 btrfs_submit_bio(fs_info, bio, mirror_num);
787 }
788
789 #ifdef CONFIG_MIGRATION
btree_migrate_folio(struct address_space * mapping,struct folio * dst,struct folio * src,enum migrate_mode mode)790 static int btree_migrate_folio(struct address_space *mapping,
791 struct folio *dst, struct folio *src, enum migrate_mode mode)
792 {
793 /*
794 * we can't safely write a btree page from here,
795 * we haven't done the locking hook
796 */
797 if (folio_test_dirty(src))
798 return -EAGAIN;
799 /*
800 * Buffers may be managed in a filesystem specific way.
801 * We must have no buffers or drop them.
802 */
803 if (folio_get_private(src) &&
804 !filemap_release_folio(src, GFP_KERNEL))
805 return -EAGAIN;
806 return migrate_folio(mapping, dst, src, mode);
807 }
808 #else
809 #define btree_migrate_folio NULL
810 #endif
811
btree_writepages(struct address_space * mapping,struct writeback_control * wbc)812 static int btree_writepages(struct address_space *mapping,
813 struct writeback_control *wbc)
814 {
815 struct btrfs_fs_info *fs_info;
816 int ret;
817
818 if (wbc->sync_mode == WB_SYNC_NONE) {
819
820 if (wbc->for_kupdate)
821 return 0;
822
823 fs_info = BTRFS_I(mapping->host)->root->fs_info;
824 /* this is a bit racy, but that's ok */
825 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
826 BTRFS_DIRTY_METADATA_THRESH,
827 fs_info->dirty_metadata_batch);
828 if (ret < 0)
829 return 0;
830 }
831 return btree_write_cache_pages(mapping, wbc);
832 }
833
btree_release_folio(struct folio * folio,gfp_t gfp_flags)834 static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
835 {
836 if (folio_test_writeback(folio) || folio_test_dirty(folio))
837 return false;
838
839 return try_release_extent_buffer(&folio->page);
840 }
841
btree_invalidate_folio(struct folio * folio,size_t offset,size_t length)842 static void btree_invalidate_folio(struct folio *folio, size_t offset,
843 size_t length)
844 {
845 struct extent_io_tree *tree;
846 tree = &BTRFS_I(folio->mapping->host)->io_tree;
847 extent_invalidate_folio(tree, folio, offset);
848 btree_release_folio(folio, GFP_NOFS);
849 if (folio_get_private(folio)) {
850 btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info,
851 "folio private not zero on folio %llu",
852 (unsigned long long)folio_pos(folio));
853 folio_detach_private(folio);
854 }
855 }
856
857 #ifdef DEBUG
btree_dirty_folio(struct address_space * mapping,struct folio * folio)858 static bool btree_dirty_folio(struct address_space *mapping,
859 struct folio *folio)
860 {
861 struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
862 struct btrfs_subpage *subpage;
863 struct extent_buffer *eb;
864 int cur_bit = 0;
865 u64 page_start = folio_pos(folio);
866
867 if (fs_info->sectorsize == PAGE_SIZE) {
868 eb = folio_get_private(folio);
869 BUG_ON(!eb);
870 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
871 BUG_ON(!atomic_read(&eb->refs));
872 btrfs_assert_tree_write_locked(eb);
873 return filemap_dirty_folio(mapping, folio);
874 }
875 subpage = folio_get_private(folio);
876
877 ASSERT(subpage->dirty_bitmap);
878 while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) {
879 unsigned long flags;
880 u64 cur;
881 u16 tmp = (1 << cur_bit);
882
883 spin_lock_irqsave(&subpage->lock, flags);
884 if (!(tmp & subpage->dirty_bitmap)) {
885 spin_unlock_irqrestore(&subpage->lock, flags);
886 cur_bit++;
887 continue;
888 }
889 spin_unlock_irqrestore(&subpage->lock, flags);
890 cur = page_start + cur_bit * fs_info->sectorsize;
891
892 eb = find_extent_buffer(fs_info, cur);
893 ASSERT(eb);
894 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
895 ASSERT(atomic_read(&eb->refs));
896 btrfs_assert_tree_write_locked(eb);
897 free_extent_buffer(eb);
898
899 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits);
900 }
901 return filemap_dirty_folio(mapping, folio);
902 }
903 #else
904 #define btree_dirty_folio filemap_dirty_folio
905 #endif
906
907 static const struct address_space_operations btree_aops = {
908 .writepages = btree_writepages,
909 .release_folio = btree_release_folio,
910 .invalidate_folio = btree_invalidate_folio,
911 .migrate_folio = btree_migrate_folio,
912 .dirty_folio = btree_dirty_folio,
913 };
914
btrfs_find_create_tree_block(struct btrfs_fs_info * fs_info,u64 bytenr,u64 owner_root,int level)915 struct extent_buffer *btrfs_find_create_tree_block(
916 struct btrfs_fs_info *fs_info,
917 u64 bytenr, u64 owner_root,
918 int level)
919 {
920 if (btrfs_is_testing(fs_info))
921 return alloc_test_extent_buffer(fs_info, bytenr);
922 return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
923 }
924
925 /*
926 * Read tree block at logical address @bytenr and do variant basic but critical
927 * verification.
928 *
929 * @owner_root: the objectid of the root owner for this block.
930 * @parent_transid: expected transid of this tree block, skip check if 0
931 * @level: expected level, mandatory check
932 * @first_key: expected key in slot 0, skip check if NULL
933 */
read_tree_block(struct btrfs_fs_info * fs_info,u64 bytenr,u64 owner_root,u64 parent_transid,int level,struct btrfs_key * first_key)934 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
935 u64 owner_root, u64 parent_transid,
936 int level, struct btrfs_key *first_key)
937 {
938 struct extent_buffer *buf = NULL;
939 int ret;
940
941 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
942 if (IS_ERR(buf))
943 return buf;
944
945 ret = btrfs_read_extent_buffer(buf, parent_transid, level, first_key);
946 if (ret) {
947 free_extent_buffer_stale(buf);
948 return ERR_PTR(ret);
949 }
950 if (btrfs_check_eb_owner(buf, owner_root)) {
951 free_extent_buffer_stale(buf);
952 return ERR_PTR(-EUCLEAN);
953 }
954 return buf;
955
956 }
957
btrfs_clean_tree_block(struct extent_buffer * buf)958 void btrfs_clean_tree_block(struct extent_buffer *buf)
959 {
960 struct btrfs_fs_info *fs_info = buf->fs_info;
961 if (btrfs_header_generation(buf) ==
962 fs_info->running_transaction->transid) {
963 btrfs_assert_tree_write_locked(buf);
964
965 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
966 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
967 -buf->len,
968 fs_info->dirty_metadata_batch);
969 clear_extent_buffer_dirty(buf);
970 }
971 }
972 }
973
__setup_root(struct btrfs_root * root,struct btrfs_fs_info * fs_info,u64 objectid)974 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
975 u64 objectid)
976 {
977 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
978
979 memset(&root->root_key, 0, sizeof(root->root_key));
980 memset(&root->root_item, 0, sizeof(root->root_item));
981 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
982 root->fs_info = fs_info;
983 root->root_key.objectid = objectid;
984 root->node = NULL;
985 root->commit_root = NULL;
986 root->state = 0;
987 RB_CLEAR_NODE(&root->rb_node);
988
989 root->last_trans = 0;
990 root->free_objectid = 0;
991 root->nr_delalloc_inodes = 0;
992 root->nr_ordered_extents = 0;
993 root->inode_tree = RB_ROOT;
994 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
995
996 btrfs_init_root_block_rsv(root);
997
998 INIT_LIST_HEAD(&root->dirty_list);
999 INIT_LIST_HEAD(&root->root_list);
1000 INIT_LIST_HEAD(&root->delalloc_inodes);
1001 INIT_LIST_HEAD(&root->delalloc_root);
1002 INIT_LIST_HEAD(&root->ordered_extents);
1003 INIT_LIST_HEAD(&root->ordered_root);
1004 INIT_LIST_HEAD(&root->reloc_dirty_list);
1005 INIT_LIST_HEAD(&root->logged_list[0]);
1006 INIT_LIST_HEAD(&root->logged_list[1]);
1007 spin_lock_init(&root->inode_lock);
1008 spin_lock_init(&root->delalloc_lock);
1009 spin_lock_init(&root->ordered_extent_lock);
1010 spin_lock_init(&root->accounting_lock);
1011 spin_lock_init(&root->log_extents_lock[0]);
1012 spin_lock_init(&root->log_extents_lock[1]);
1013 spin_lock_init(&root->qgroup_meta_rsv_lock);
1014 mutex_init(&root->objectid_mutex);
1015 mutex_init(&root->log_mutex);
1016 mutex_init(&root->ordered_extent_mutex);
1017 mutex_init(&root->delalloc_mutex);
1018 init_waitqueue_head(&root->qgroup_flush_wait);
1019 init_waitqueue_head(&root->log_writer_wait);
1020 init_waitqueue_head(&root->log_commit_wait[0]);
1021 init_waitqueue_head(&root->log_commit_wait[1]);
1022 INIT_LIST_HEAD(&root->log_ctxs[0]);
1023 INIT_LIST_HEAD(&root->log_ctxs[1]);
1024 atomic_set(&root->log_commit[0], 0);
1025 atomic_set(&root->log_commit[1], 0);
1026 atomic_set(&root->log_writers, 0);
1027 atomic_set(&root->log_batch, 0);
1028 refcount_set(&root->refs, 1);
1029 atomic_set(&root->snapshot_force_cow, 0);
1030 atomic_set(&root->nr_swapfiles, 0);
1031 root->log_transid = 0;
1032 root->log_transid_committed = -1;
1033 root->last_log_commit = 0;
1034 root->anon_dev = 0;
1035 if (!dummy) {
1036 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1037 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1038 extent_io_tree_init(fs_info, &root->log_csum_range,
1039 IO_TREE_LOG_CSUM_RANGE, NULL);
1040 }
1041
1042 spin_lock_init(&root->root_item_lock);
1043 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1044 #ifdef CONFIG_BTRFS_DEBUG
1045 INIT_LIST_HEAD(&root->leak_list);
1046 spin_lock(&fs_info->fs_roots_radix_lock);
1047 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1048 spin_unlock(&fs_info->fs_roots_radix_lock);
1049 #endif
1050 }
1051
btrfs_alloc_root(struct btrfs_fs_info * fs_info,u64 objectid,gfp_t flags)1052 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1053 u64 objectid, gfp_t flags)
1054 {
1055 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1056 if (root)
1057 __setup_root(root, fs_info, objectid);
1058 return root;
1059 }
1060
1061 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1062 /* Should only be used by the testing infrastructure */
btrfs_alloc_dummy_root(struct btrfs_fs_info * fs_info)1063 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1064 {
1065 struct btrfs_root *root;
1066
1067 if (!fs_info)
1068 return ERR_PTR(-EINVAL);
1069
1070 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1071 if (!root)
1072 return ERR_PTR(-ENOMEM);
1073
1074 /* We don't use the stripesize in selftest, set it as sectorsize */
1075 root->alloc_bytenr = 0;
1076
1077 return root;
1078 }
1079 #endif
1080
global_root_cmp(struct rb_node * a_node,const struct rb_node * b_node)1081 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
1082 {
1083 const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
1084 const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
1085
1086 return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
1087 }
1088
global_root_key_cmp(const void * k,const struct rb_node * node)1089 static int global_root_key_cmp(const void *k, const struct rb_node *node)
1090 {
1091 const struct btrfs_key *key = k;
1092 const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
1093
1094 return btrfs_comp_cpu_keys(key, &root->root_key);
1095 }
1096
btrfs_global_root_insert(struct btrfs_root * root)1097 int btrfs_global_root_insert(struct btrfs_root *root)
1098 {
1099 struct btrfs_fs_info *fs_info = root->fs_info;
1100 struct rb_node *tmp;
1101
1102 write_lock(&fs_info->global_root_lock);
1103 tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
1104 write_unlock(&fs_info->global_root_lock);
1105 ASSERT(!tmp);
1106
1107 return tmp ? -EEXIST : 0;
1108 }
1109
btrfs_global_root_delete(struct btrfs_root * root)1110 void btrfs_global_root_delete(struct btrfs_root *root)
1111 {
1112 struct btrfs_fs_info *fs_info = root->fs_info;
1113
1114 write_lock(&fs_info->global_root_lock);
1115 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1116 write_unlock(&fs_info->global_root_lock);
1117 }
1118
btrfs_global_root(struct btrfs_fs_info * fs_info,struct btrfs_key * key)1119 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
1120 struct btrfs_key *key)
1121 {
1122 struct rb_node *node;
1123 struct btrfs_root *root = NULL;
1124
1125 read_lock(&fs_info->global_root_lock);
1126 node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
1127 if (node)
1128 root = container_of(node, struct btrfs_root, rb_node);
1129 read_unlock(&fs_info->global_root_lock);
1130
1131 return root;
1132 }
1133
btrfs_global_root_id(struct btrfs_fs_info * fs_info,u64 bytenr)1134 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
1135 {
1136 struct btrfs_block_group *block_group;
1137 u64 ret;
1138
1139 if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
1140 return 0;
1141
1142 if (bytenr)
1143 block_group = btrfs_lookup_block_group(fs_info, bytenr);
1144 else
1145 block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
1146 ASSERT(block_group);
1147 if (!block_group)
1148 return 0;
1149 ret = block_group->global_root_id;
1150 btrfs_put_block_group(block_group);
1151
1152 return ret;
1153 }
1154
btrfs_csum_root(struct btrfs_fs_info * fs_info,u64 bytenr)1155 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1156 {
1157 struct btrfs_key key = {
1158 .objectid = BTRFS_CSUM_TREE_OBJECTID,
1159 .type = BTRFS_ROOT_ITEM_KEY,
1160 .offset = btrfs_global_root_id(fs_info, bytenr),
1161 };
1162
1163 return btrfs_global_root(fs_info, &key);
1164 }
1165
btrfs_extent_root(struct btrfs_fs_info * fs_info,u64 bytenr)1166 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1167 {
1168 struct btrfs_key key = {
1169 .objectid = BTRFS_EXTENT_TREE_OBJECTID,
1170 .type = BTRFS_ROOT_ITEM_KEY,
1171 .offset = btrfs_global_root_id(fs_info, bytenr),
1172 };
1173
1174 return btrfs_global_root(fs_info, &key);
1175 }
1176
btrfs_create_tree(struct btrfs_trans_handle * trans,u64 objectid)1177 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1178 u64 objectid)
1179 {
1180 struct btrfs_fs_info *fs_info = trans->fs_info;
1181 struct extent_buffer *leaf;
1182 struct btrfs_root *tree_root = fs_info->tree_root;
1183 struct btrfs_root *root;
1184 struct btrfs_key key;
1185 unsigned int nofs_flag;
1186 int ret = 0;
1187
1188 /*
1189 * We're holding a transaction handle, so use a NOFS memory allocation
1190 * context to avoid deadlock if reclaim happens.
1191 */
1192 nofs_flag = memalloc_nofs_save();
1193 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1194 memalloc_nofs_restore(nofs_flag);
1195 if (!root)
1196 return ERR_PTR(-ENOMEM);
1197
1198 root->root_key.objectid = objectid;
1199 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1200 root->root_key.offset = 0;
1201
1202 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1203 BTRFS_NESTING_NORMAL);
1204 if (IS_ERR(leaf)) {
1205 ret = PTR_ERR(leaf);
1206 leaf = NULL;
1207 goto fail_unlock;
1208 }
1209
1210 root->node = leaf;
1211 btrfs_mark_buffer_dirty(leaf);
1212
1213 root->commit_root = btrfs_root_node(root);
1214 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1215
1216 btrfs_set_root_flags(&root->root_item, 0);
1217 btrfs_set_root_limit(&root->root_item, 0);
1218 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1219 btrfs_set_root_generation(&root->root_item, trans->transid);
1220 btrfs_set_root_level(&root->root_item, 0);
1221 btrfs_set_root_refs(&root->root_item, 1);
1222 btrfs_set_root_used(&root->root_item, leaf->len);
1223 btrfs_set_root_last_snapshot(&root->root_item, 0);
1224 btrfs_set_root_dirid(&root->root_item, 0);
1225 if (is_fstree(objectid))
1226 generate_random_guid(root->root_item.uuid);
1227 else
1228 export_guid(root->root_item.uuid, &guid_null);
1229 btrfs_set_root_drop_level(&root->root_item, 0);
1230
1231 btrfs_tree_unlock(leaf);
1232
1233 key.objectid = objectid;
1234 key.type = BTRFS_ROOT_ITEM_KEY;
1235 key.offset = 0;
1236 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1237 if (ret)
1238 goto fail;
1239
1240 return root;
1241
1242 fail_unlock:
1243 if (leaf)
1244 btrfs_tree_unlock(leaf);
1245 fail:
1246 btrfs_put_root(root);
1247
1248 return ERR_PTR(ret);
1249 }
1250
alloc_log_tree(struct btrfs_trans_handle * trans,struct btrfs_fs_info * fs_info)1251 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1252 struct btrfs_fs_info *fs_info)
1253 {
1254 struct btrfs_root *root;
1255
1256 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1257 if (!root)
1258 return ERR_PTR(-ENOMEM);
1259
1260 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1261 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1262 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1263
1264 return root;
1265 }
1266
btrfs_alloc_log_tree_node(struct btrfs_trans_handle * trans,struct btrfs_root * root)1267 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
1268 struct btrfs_root *root)
1269 {
1270 struct extent_buffer *leaf;
1271
1272 /*
1273 * DON'T set SHAREABLE bit for log trees.
1274 *
1275 * Log trees are not exposed to user space thus can't be snapshotted,
1276 * and they go away before a real commit is actually done.
1277 *
1278 * They do store pointers to file data extents, and those reference
1279 * counts still get updated (along with back refs to the log tree).
1280 */
1281
1282 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1283 NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1284 if (IS_ERR(leaf))
1285 return PTR_ERR(leaf);
1286
1287 root->node = leaf;
1288
1289 btrfs_mark_buffer_dirty(root->node);
1290 btrfs_tree_unlock(root->node);
1291
1292 return 0;
1293 }
1294
btrfs_init_log_root_tree(struct btrfs_trans_handle * trans,struct btrfs_fs_info * fs_info)1295 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1296 struct btrfs_fs_info *fs_info)
1297 {
1298 struct btrfs_root *log_root;
1299
1300 log_root = alloc_log_tree(trans, fs_info);
1301 if (IS_ERR(log_root))
1302 return PTR_ERR(log_root);
1303
1304 if (!btrfs_is_zoned(fs_info)) {
1305 int ret = btrfs_alloc_log_tree_node(trans, log_root);
1306
1307 if (ret) {
1308 btrfs_put_root(log_root);
1309 return ret;
1310 }
1311 }
1312
1313 WARN_ON(fs_info->log_root_tree);
1314 fs_info->log_root_tree = log_root;
1315 return 0;
1316 }
1317
btrfs_add_log_tree(struct btrfs_trans_handle * trans,struct btrfs_root * root)1318 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1319 struct btrfs_root *root)
1320 {
1321 struct btrfs_fs_info *fs_info = root->fs_info;
1322 struct btrfs_root *log_root;
1323 struct btrfs_inode_item *inode_item;
1324 int ret;
1325
1326 log_root = alloc_log_tree(trans, fs_info);
1327 if (IS_ERR(log_root))
1328 return PTR_ERR(log_root);
1329
1330 ret = btrfs_alloc_log_tree_node(trans, log_root);
1331 if (ret) {
1332 btrfs_put_root(log_root);
1333 return ret;
1334 }
1335
1336 log_root->last_trans = trans->transid;
1337 log_root->root_key.offset = root->root_key.objectid;
1338
1339 inode_item = &log_root->root_item.inode;
1340 btrfs_set_stack_inode_generation(inode_item, 1);
1341 btrfs_set_stack_inode_size(inode_item, 3);
1342 btrfs_set_stack_inode_nlink(inode_item, 1);
1343 btrfs_set_stack_inode_nbytes(inode_item,
1344 fs_info->nodesize);
1345 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1346
1347 btrfs_set_root_node(&log_root->root_item, log_root->node);
1348
1349 WARN_ON(root->log_root);
1350 root->log_root = log_root;
1351 root->log_transid = 0;
1352 root->log_transid_committed = -1;
1353 root->last_log_commit = 0;
1354 return 0;
1355 }
1356
read_tree_root_path(struct btrfs_root * tree_root,struct btrfs_path * path,struct btrfs_key * key)1357 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1358 struct btrfs_path *path,
1359 struct btrfs_key *key)
1360 {
1361 struct btrfs_root *root;
1362 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1363 u64 generation;
1364 int ret;
1365 int level;
1366
1367 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1368 if (!root)
1369 return ERR_PTR(-ENOMEM);
1370
1371 ret = btrfs_find_root(tree_root, key, path,
1372 &root->root_item, &root->root_key);
1373 if (ret) {
1374 if (ret > 0)
1375 ret = -ENOENT;
1376 goto fail;
1377 }
1378
1379 generation = btrfs_root_generation(&root->root_item);
1380 level = btrfs_root_level(&root->root_item);
1381 root->node = read_tree_block(fs_info,
1382 btrfs_root_bytenr(&root->root_item),
1383 key->objectid, generation, level, NULL);
1384 if (IS_ERR(root->node)) {
1385 ret = PTR_ERR(root->node);
1386 root->node = NULL;
1387 goto fail;
1388 }
1389 if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1390 ret = -EIO;
1391 goto fail;
1392 }
1393
1394 /*
1395 * For real fs, and not log/reloc trees, root owner must
1396 * match its root node owner
1397 */
1398 if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) &&
1399 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1400 root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1401 root->root_key.objectid != btrfs_header_owner(root->node)) {
1402 btrfs_crit(fs_info,
1403 "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1404 root->root_key.objectid, root->node->start,
1405 btrfs_header_owner(root->node),
1406 root->root_key.objectid);
1407 ret = -EUCLEAN;
1408 goto fail;
1409 }
1410 root->commit_root = btrfs_root_node(root);
1411 return root;
1412 fail:
1413 btrfs_put_root(root);
1414 return ERR_PTR(ret);
1415 }
1416
btrfs_read_tree_root(struct btrfs_root * tree_root,struct btrfs_key * key)1417 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1418 struct btrfs_key *key)
1419 {
1420 struct btrfs_root *root;
1421 struct btrfs_path *path;
1422
1423 path = btrfs_alloc_path();
1424 if (!path)
1425 return ERR_PTR(-ENOMEM);
1426 root = read_tree_root_path(tree_root, path, key);
1427 btrfs_free_path(path);
1428
1429 return root;
1430 }
1431
1432 /*
1433 * Initialize subvolume root in-memory structure
1434 *
1435 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1436 */
btrfs_init_fs_root(struct btrfs_root * root,dev_t anon_dev)1437 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1438 {
1439 int ret;
1440 unsigned int nofs_flag;
1441
1442 /*
1443 * We might be called under a transaction (e.g. indirect backref
1444 * resolution) which could deadlock if it triggers memory reclaim
1445 */
1446 nofs_flag = memalloc_nofs_save();
1447 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1448 memalloc_nofs_restore(nofs_flag);
1449 if (ret)
1450 goto fail;
1451
1452 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1453 !btrfs_is_data_reloc_root(root)) {
1454 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1455 btrfs_check_and_init_root_item(&root->root_item);
1456 }
1457
1458 /*
1459 * Don't assign anonymous block device to roots that are not exposed to
1460 * userspace, the id pool is limited to 1M
1461 */
1462 if (is_fstree(root->root_key.objectid) &&
1463 btrfs_root_refs(&root->root_item) > 0) {
1464 if (!anon_dev) {
1465 ret = get_anon_bdev(&root->anon_dev);
1466 if (ret)
1467 goto fail;
1468 } else {
1469 root->anon_dev = anon_dev;
1470 }
1471 }
1472
1473 mutex_lock(&root->objectid_mutex);
1474 ret = btrfs_init_root_free_objectid(root);
1475 if (ret) {
1476 mutex_unlock(&root->objectid_mutex);
1477 goto fail;
1478 }
1479
1480 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1481
1482 mutex_unlock(&root->objectid_mutex);
1483
1484 return 0;
1485 fail:
1486 /* The caller is responsible to call btrfs_free_fs_root */
1487 return ret;
1488 }
1489
btrfs_lookup_fs_root(struct btrfs_fs_info * fs_info,u64 root_id)1490 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1491 u64 root_id)
1492 {
1493 struct btrfs_root *root;
1494
1495 spin_lock(&fs_info->fs_roots_radix_lock);
1496 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1497 (unsigned long)root_id);
1498 if (root)
1499 root = btrfs_grab_root(root);
1500 spin_unlock(&fs_info->fs_roots_radix_lock);
1501 return root;
1502 }
1503
btrfs_get_global_root(struct btrfs_fs_info * fs_info,u64 objectid)1504 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1505 u64 objectid)
1506 {
1507 struct btrfs_key key = {
1508 .objectid = objectid,
1509 .type = BTRFS_ROOT_ITEM_KEY,
1510 .offset = 0,
1511 };
1512
1513 if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1514 return btrfs_grab_root(fs_info->tree_root);
1515 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1516 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1517 if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1518 return btrfs_grab_root(fs_info->chunk_root);
1519 if (objectid == BTRFS_DEV_TREE_OBJECTID)
1520 return btrfs_grab_root(fs_info->dev_root);
1521 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1522 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1523 if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1524 return btrfs_grab_root(fs_info->quota_root) ?
1525 fs_info->quota_root : ERR_PTR(-ENOENT);
1526 if (objectid == BTRFS_UUID_TREE_OBJECTID)
1527 return btrfs_grab_root(fs_info->uuid_root) ?
1528 fs_info->uuid_root : ERR_PTR(-ENOENT);
1529 if (objectid == BTRFS_BLOCK_GROUP_TREE_OBJECTID)
1530 return btrfs_grab_root(fs_info->block_group_root) ?
1531 fs_info->block_group_root : ERR_PTR(-ENOENT);
1532 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) {
1533 struct btrfs_root *root = btrfs_global_root(fs_info, &key);
1534
1535 return btrfs_grab_root(root) ? root : ERR_PTR(-ENOENT);
1536 }
1537 return NULL;
1538 }
1539
btrfs_insert_fs_root(struct btrfs_fs_info * fs_info,struct btrfs_root * root)1540 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1541 struct btrfs_root *root)
1542 {
1543 int ret;
1544
1545 ret = radix_tree_preload(GFP_NOFS);
1546 if (ret)
1547 return ret;
1548
1549 spin_lock(&fs_info->fs_roots_radix_lock);
1550 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1551 (unsigned long)root->root_key.objectid,
1552 root);
1553 if (ret == 0) {
1554 btrfs_grab_root(root);
1555 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1556 }
1557 spin_unlock(&fs_info->fs_roots_radix_lock);
1558 radix_tree_preload_end();
1559
1560 return ret;
1561 }
1562
btrfs_check_leaked_roots(struct btrfs_fs_info * fs_info)1563 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1564 {
1565 #ifdef CONFIG_BTRFS_DEBUG
1566 struct btrfs_root *root;
1567
1568 while (!list_empty(&fs_info->allocated_roots)) {
1569 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1570
1571 root = list_first_entry(&fs_info->allocated_roots,
1572 struct btrfs_root, leak_list);
1573 btrfs_err(fs_info, "leaked root %s refcount %d",
1574 btrfs_root_name(&root->root_key, buf),
1575 refcount_read(&root->refs));
1576 while (refcount_read(&root->refs) > 1)
1577 btrfs_put_root(root);
1578 btrfs_put_root(root);
1579 }
1580 #endif
1581 }
1582
free_global_roots(struct btrfs_fs_info * fs_info)1583 static void free_global_roots(struct btrfs_fs_info *fs_info)
1584 {
1585 struct btrfs_root *root;
1586 struct rb_node *node;
1587
1588 while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1589 root = rb_entry(node, struct btrfs_root, rb_node);
1590 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1591 btrfs_put_root(root);
1592 }
1593 }
1594
btrfs_free_fs_info(struct btrfs_fs_info * fs_info)1595 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1596 {
1597 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1598 percpu_counter_destroy(&fs_info->delalloc_bytes);
1599 percpu_counter_destroy(&fs_info->ordered_bytes);
1600 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1601 btrfs_free_csum_hash(fs_info);
1602 btrfs_free_stripe_hash_table(fs_info);
1603 btrfs_free_ref_cache(fs_info);
1604 kfree(fs_info->balance_ctl);
1605 kfree(fs_info->delayed_root);
1606 free_global_roots(fs_info);
1607 btrfs_put_root(fs_info->tree_root);
1608 btrfs_put_root(fs_info->chunk_root);
1609 btrfs_put_root(fs_info->dev_root);
1610 btrfs_put_root(fs_info->quota_root);
1611 btrfs_put_root(fs_info->uuid_root);
1612 btrfs_put_root(fs_info->fs_root);
1613 btrfs_put_root(fs_info->data_reloc_root);
1614 btrfs_put_root(fs_info->block_group_root);
1615 btrfs_check_leaked_roots(fs_info);
1616 btrfs_extent_buffer_leak_debug_check(fs_info);
1617 kfree(fs_info->super_copy);
1618 kfree(fs_info->super_for_commit);
1619 kfree(fs_info->subpage_info);
1620 kvfree(fs_info);
1621 }
1622
1623
1624 /*
1625 * Get an in-memory reference of a root structure.
1626 *
1627 * For essential trees like root/extent tree, we grab it from fs_info directly.
1628 * For subvolume trees, we check the cached filesystem roots first. If not
1629 * found, then read it from disk and add it to cached fs roots.
1630 *
1631 * Caller should release the root by calling btrfs_put_root() after the usage.
1632 *
1633 * NOTE: Reloc and log trees can't be read by this function as they share the
1634 * same root objectid.
1635 *
1636 * @objectid: root id
1637 * @anon_dev: preallocated anonymous block device number for new roots,
1638 * pass 0 for new allocation.
1639 * @check_ref: whether to check root item references, If true, return -ENOENT
1640 * for orphan roots
1641 */
btrfs_get_root_ref(struct btrfs_fs_info * fs_info,u64 objectid,dev_t anon_dev,bool check_ref)1642 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1643 u64 objectid, dev_t anon_dev,
1644 bool check_ref)
1645 {
1646 struct btrfs_root *root;
1647 struct btrfs_path *path;
1648 struct btrfs_key key;
1649 int ret;
1650
1651 root = btrfs_get_global_root(fs_info, objectid);
1652 if (root)
1653 return root;
1654 again:
1655 root = btrfs_lookup_fs_root(fs_info, objectid);
1656 if (root) {
1657 /* Shouldn't get preallocated anon_dev for cached roots */
1658 ASSERT(!anon_dev);
1659 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1660 btrfs_put_root(root);
1661 return ERR_PTR(-ENOENT);
1662 }
1663 return root;
1664 }
1665
1666 key.objectid = objectid;
1667 key.type = BTRFS_ROOT_ITEM_KEY;
1668 key.offset = (u64)-1;
1669 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1670 if (IS_ERR(root))
1671 return root;
1672
1673 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1674 ret = -ENOENT;
1675 goto fail;
1676 }
1677
1678 ret = btrfs_init_fs_root(root, anon_dev);
1679 if (ret)
1680 goto fail;
1681
1682 path = btrfs_alloc_path();
1683 if (!path) {
1684 ret = -ENOMEM;
1685 goto fail;
1686 }
1687 key.objectid = BTRFS_ORPHAN_OBJECTID;
1688 key.type = BTRFS_ORPHAN_ITEM_KEY;
1689 key.offset = objectid;
1690
1691 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1692 btrfs_free_path(path);
1693 if (ret < 0)
1694 goto fail;
1695 if (ret == 0)
1696 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1697
1698 ret = btrfs_insert_fs_root(fs_info, root);
1699 if (ret) {
1700 if (ret == -EEXIST) {
1701 btrfs_put_root(root);
1702 goto again;
1703 }
1704 goto fail;
1705 }
1706 return root;
1707 fail:
1708 /*
1709 * If our caller provided us an anonymous device, then it's his
1710 * responsibility to free it in case we fail. So we have to set our
1711 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1712 * and once again by our caller.
1713 */
1714 if (anon_dev)
1715 root->anon_dev = 0;
1716 btrfs_put_root(root);
1717 return ERR_PTR(ret);
1718 }
1719
1720 /*
1721 * Get in-memory reference of a root structure
1722 *
1723 * @objectid: tree objectid
1724 * @check_ref: if set, verify that the tree exists and the item has at least
1725 * one reference
1726 */
btrfs_get_fs_root(struct btrfs_fs_info * fs_info,u64 objectid,bool check_ref)1727 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1728 u64 objectid, bool check_ref)
1729 {
1730 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1731 }
1732
1733 /*
1734 * Get in-memory reference of a root structure, created as new, optionally pass
1735 * the anonymous block device id
1736 *
1737 * @objectid: tree objectid
1738 * @anon_dev: if zero, allocate a new anonymous block device or use the
1739 * parameter value
1740 */
btrfs_get_new_fs_root(struct btrfs_fs_info * fs_info,u64 objectid,dev_t anon_dev)1741 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1742 u64 objectid, dev_t anon_dev)
1743 {
1744 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1745 }
1746
1747 /*
1748 * btrfs_get_fs_root_commit_root - return a root for the given objectid
1749 * @fs_info: the fs_info
1750 * @objectid: the objectid we need to lookup
1751 *
1752 * This is exclusively used for backref walking, and exists specifically because
1753 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1754 * creation time, which means we may have to read the tree_root in order to look
1755 * up a fs root that is not in memory. If the root is not in memory we will
1756 * read the tree root commit root and look up the fs root from there. This is a
1757 * temporary root, it will not be inserted into the radix tree as it doesn't
1758 * have the most uptodate information, it'll simply be discarded once the
1759 * backref code is finished using the root.
1760 */
btrfs_get_fs_root_commit_root(struct btrfs_fs_info * fs_info,struct btrfs_path * path,u64 objectid)1761 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1762 struct btrfs_path *path,
1763 u64 objectid)
1764 {
1765 struct btrfs_root *root;
1766 struct btrfs_key key;
1767
1768 ASSERT(path->search_commit_root && path->skip_locking);
1769
1770 /*
1771 * This can return -ENOENT if we ask for a root that doesn't exist, but
1772 * since this is called via the backref walking code we won't be looking
1773 * up a root that doesn't exist, unless there's corruption. So if root
1774 * != NULL just return it.
1775 */
1776 root = btrfs_get_global_root(fs_info, objectid);
1777 if (root)
1778 return root;
1779
1780 root = btrfs_lookup_fs_root(fs_info, objectid);
1781 if (root)
1782 return root;
1783
1784 key.objectid = objectid;
1785 key.type = BTRFS_ROOT_ITEM_KEY;
1786 key.offset = (u64)-1;
1787 root = read_tree_root_path(fs_info->tree_root, path, &key);
1788 btrfs_release_path(path);
1789
1790 return root;
1791 }
1792
cleaner_kthread(void * arg)1793 static int cleaner_kthread(void *arg)
1794 {
1795 struct btrfs_fs_info *fs_info = arg;
1796 int again;
1797
1798 while (1) {
1799 again = 0;
1800
1801 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1802
1803 /* Make the cleaner go to sleep early. */
1804 if (btrfs_need_cleaner_sleep(fs_info))
1805 goto sleep;
1806
1807 /*
1808 * Do not do anything if we might cause open_ctree() to block
1809 * before we have finished mounting the filesystem.
1810 */
1811 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1812 goto sleep;
1813
1814 if (!mutex_trylock(&fs_info->cleaner_mutex))
1815 goto sleep;
1816
1817 /*
1818 * Avoid the problem that we change the status of the fs
1819 * during the above check and trylock.
1820 */
1821 if (btrfs_need_cleaner_sleep(fs_info)) {
1822 mutex_unlock(&fs_info->cleaner_mutex);
1823 goto sleep;
1824 }
1825
1826 btrfs_run_delayed_iputs(fs_info);
1827
1828 again = btrfs_clean_one_deleted_snapshot(fs_info);
1829 mutex_unlock(&fs_info->cleaner_mutex);
1830
1831 /*
1832 * The defragger has dealt with the R/O remount and umount,
1833 * needn't do anything special here.
1834 */
1835 btrfs_run_defrag_inodes(fs_info);
1836
1837 /*
1838 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1839 * with relocation (btrfs_relocate_chunk) and relocation
1840 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1841 * after acquiring fs_info->reclaim_bgs_lock. So we
1842 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1843 * unused block groups.
1844 */
1845 btrfs_delete_unused_bgs(fs_info);
1846
1847 /*
1848 * Reclaim block groups in the reclaim_bgs list after we deleted
1849 * all unused block_groups. This possibly gives us some more free
1850 * space.
1851 */
1852 btrfs_reclaim_bgs(fs_info);
1853 sleep:
1854 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1855 if (kthread_should_park())
1856 kthread_parkme();
1857 if (kthread_should_stop())
1858 return 0;
1859 if (!again) {
1860 set_current_state(TASK_INTERRUPTIBLE);
1861 schedule();
1862 __set_current_state(TASK_RUNNING);
1863 }
1864 }
1865 }
1866
transaction_kthread(void * arg)1867 static int transaction_kthread(void *arg)
1868 {
1869 struct btrfs_root *root = arg;
1870 struct btrfs_fs_info *fs_info = root->fs_info;
1871 struct btrfs_trans_handle *trans;
1872 struct btrfs_transaction *cur;
1873 u64 transid;
1874 time64_t delta;
1875 unsigned long delay;
1876 bool cannot_commit;
1877
1878 do {
1879 cannot_commit = false;
1880 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1881 mutex_lock(&fs_info->transaction_kthread_mutex);
1882
1883 spin_lock(&fs_info->trans_lock);
1884 cur = fs_info->running_transaction;
1885 if (!cur) {
1886 spin_unlock(&fs_info->trans_lock);
1887 goto sleep;
1888 }
1889
1890 delta = ktime_get_seconds() - cur->start_time;
1891 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1892 cur->state < TRANS_STATE_COMMIT_START &&
1893 delta < fs_info->commit_interval) {
1894 spin_unlock(&fs_info->trans_lock);
1895 delay -= msecs_to_jiffies((delta - 1) * 1000);
1896 delay = min(delay,
1897 msecs_to_jiffies(fs_info->commit_interval * 1000));
1898 goto sleep;
1899 }
1900 transid = cur->transid;
1901 spin_unlock(&fs_info->trans_lock);
1902
1903 /* If the file system is aborted, this will always fail. */
1904 trans = btrfs_attach_transaction(root);
1905 if (IS_ERR(trans)) {
1906 if (PTR_ERR(trans) != -ENOENT)
1907 cannot_commit = true;
1908 goto sleep;
1909 }
1910 if (transid == trans->transid) {
1911 btrfs_commit_transaction(trans);
1912 } else {
1913 btrfs_end_transaction(trans);
1914 }
1915 sleep:
1916 wake_up_process(fs_info->cleaner_kthread);
1917 mutex_unlock(&fs_info->transaction_kthread_mutex);
1918
1919 if (BTRFS_FS_ERROR(fs_info))
1920 btrfs_cleanup_transaction(fs_info);
1921 if (!kthread_should_stop() &&
1922 (!btrfs_transaction_blocked(fs_info) ||
1923 cannot_commit))
1924 schedule_timeout_interruptible(delay);
1925 } while (!kthread_should_stop());
1926 return 0;
1927 }
1928
1929 /*
1930 * This will find the highest generation in the array of root backups. The
1931 * index of the highest array is returned, or -EINVAL if we can't find
1932 * anything.
1933 *
1934 * We check to make sure the array is valid by comparing the
1935 * generation of the latest root in the array with the generation
1936 * in the super block. If they don't match we pitch it.
1937 */
find_newest_super_backup(struct btrfs_fs_info * info)1938 static int find_newest_super_backup(struct btrfs_fs_info *info)
1939 {
1940 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1941 u64 cur;
1942 struct btrfs_root_backup *root_backup;
1943 int i;
1944
1945 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1946 root_backup = info->super_copy->super_roots + i;
1947 cur = btrfs_backup_tree_root_gen(root_backup);
1948 if (cur == newest_gen)
1949 return i;
1950 }
1951
1952 return -EINVAL;
1953 }
1954
1955 /*
1956 * copy all the root pointers into the super backup array.
1957 * this will bump the backup pointer by one when it is
1958 * done
1959 */
backup_super_roots(struct btrfs_fs_info * info)1960 static void backup_super_roots(struct btrfs_fs_info *info)
1961 {
1962 const int next_backup = info->backup_root_index;
1963 struct btrfs_root_backup *root_backup;
1964
1965 root_backup = info->super_for_commit->super_roots + next_backup;
1966
1967 /*
1968 * make sure all of our padding and empty slots get zero filled
1969 * regardless of which ones we use today
1970 */
1971 memset(root_backup, 0, sizeof(*root_backup));
1972
1973 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1974
1975 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1976 btrfs_set_backup_tree_root_gen(root_backup,
1977 btrfs_header_generation(info->tree_root->node));
1978
1979 btrfs_set_backup_tree_root_level(root_backup,
1980 btrfs_header_level(info->tree_root->node));
1981
1982 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1983 btrfs_set_backup_chunk_root_gen(root_backup,
1984 btrfs_header_generation(info->chunk_root->node));
1985 btrfs_set_backup_chunk_root_level(root_backup,
1986 btrfs_header_level(info->chunk_root->node));
1987
1988 if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
1989 struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
1990 struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
1991
1992 btrfs_set_backup_extent_root(root_backup,
1993 extent_root->node->start);
1994 btrfs_set_backup_extent_root_gen(root_backup,
1995 btrfs_header_generation(extent_root->node));
1996 btrfs_set_backup_extent_root_level(root_backup,
1997 btrfs_header_level(extent_root->node));
1998
1999 btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
2000 btrfs_set_backup_csum_root_gen(root_backup,
2001 btrfs_header_generation(csum_root->node));
2002 btrfs_set_backup_csum_root_level(root_backup,
2003 btrfs_header_level(csum_root->node));
2004 }
2005
2006 /*
2007 * we might commit during log recovery, which happens before we set
2008 * the fs_root. Make sure it is valid before we fill it in.
2009 */
2010 if (info->fs_root && info->fs_root->node) {
2011 btrfs_set_backup_fs_root(root_backup,
2012 info->fs_root->node->start);
2013 btrfs_set_backup_fs_root_gen(root_backup,
2014 btrfs_header_generation(info->fs_root->node));
2015 btrfs_set_backup_fs_root_level(root_backup,
2016 btrfs_header_level(info->fs_root->node));
2017 }
2018
2019 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2020 btrfs_set_backup_dev_root_gen(root_backup,
2021 btrfs_header_generation(info->dev_root->node));
2022 btrfs_set_backup_dev_root_level(root_backup,
2023 btrfs_header_level(info->dev_root->node));
2024
2025 btrfs_set_backup_total_bytes(root_backup,
2026 btrfs_super_total_bytes(info->super_copy));
2027 btrfs_set_backup_bytes_used(root_backup,
2028 btrfs_super_bytes_used(info->super_copy));
2029 btrfs_set_backup_num_devices(root_backup,
2030 btrfs_super_num_devices(info->super_copy));
2031
2032 /*
2033 * if we don't copy this out to the super_copy, it won't get remembered
2034 * for the next commit
2035 */
2036 memcpy(&info->super_copy->super_roots,
2037 &info->super_for_commit->super_roots,
2038 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2039 }
2040
2041 /*
2042 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
2043 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
2044 *
2045 * fs_info - filesystem whose backup roots need to be read
2046 * priority - priority of backup root required
2047 *
2048 * Returns backup root index on success and -EINVAL otherwise.
2049 */
read_backup_root(struct btrfs_fs_info * fs_info,u8 priority)2050 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
2051 {
2052 int backup_index = find_newest_super_backup(fs_info);
2053 struct btrfs_super_block *super = fs_info->super_copy;
2054 struct btrfs_root_backup *root_backup;
2055
2056 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
2057 if (priority == 0)
2058 return backup_index;
2059
2060 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
2061 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
2062 } else {
2063 return -EINVAL;
2064 }
2065
2066 root_backup = super->super_roots + backup_index;
2067
2068 btrfs_set_super_generation(super,
2069 btrfs_backup_tree_root_gen(root_backup));
2070 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2071 btrfs_set_super_root_level(super,
2072 btrfs_backup_tree_root_level(root_backup));
2073 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2074
2075 /*
2076 * Fixme: the total bytes and num_devices need to match or we should
2077 * need a fsck
2078 */
2079 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2080 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2081
2082 return backup_index;
2083 }
2084
2085 /* helper to cleanup workers */
btrfs_stop_all_workers(struct btrfs_fs_info * fs_info)2086 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2087 {
2088 btrfs_destroy_workqueue(fs_info->fixup_workers);
2089 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2090 btrfs_destroy_workqueue(fs_info->hipri_workers);
2091 btrfs_destroy_workqueue(fs_info->workers);
2092 if (fs_info->endio_workers)
2093 destroy_workqueue(fs_info->endio_workers);
2094 if (fs_info->endio_raid56_workers)
2095 destroy_workqueue(fs_info->endio_raid56_workers);
2096 if (fs_info->rmw_workers)
2097 destroy_workqueue(fs_info->rmw_workers);
2098 if (fs_info->compressed_write_workers)
2099 destroy_workqueue(fs_info->compressed_write_workers);
2100 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2101 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2102 btrfs_destroy_workqueue(fs_info->delayed_workers);
2103 btrfs_destroy_workqueue(fs_info->caching_workers);
2104 btrfs_destroy_workqueue(fs_info->flush_workers);
2105 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2106 if (fs_info->discard_ctl.discard_workers)
2107 destroy_workqueue(fs_info->discard_ctl.discard_workers);
2108 /*
2109 * Now that all other work queues are destroyed, we can safely destroy
2110 * the queues used for metadata I/O, since tasks from those other work
2111 * queues can do metadata I/O operations.
2112 */
2113 if (fs_info->endio_meta_workers)
2114 destroy_workqueue(fs_info->endio_meta_workers);
2115 }
2116
free_root_extent_buffers(struct btrfs_root * root)2117 static void free_root_extent_buffers(struct btrfs_root *root)
2118 {
2119 if (root) {
2120 free_extent_buffer(root->node);
2121 free_extent_buffer(root->commit_root);
2122 root->node = NULL;
2123 root->commit_root = NULL;
2124 }
2125 }
2126
free_global_root_pointers(struct btrfs_fs_info * fs_info)2127 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
2128 {
2129 struct btrfs_root *root, *tmp;
2130
2131 rbtree_postorder_for_each_entry_safe(root, tmp,
2132 &fs_info->global_root_tree,
2133 rb_node)
2134 free_root_extent_buffers(root);
2135 }
2136
2137 /* helper to cleanup tree roots */
free_root_pointers(struct btrfs_fs_info * info,bool free_chunk_root)2138 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2139 {
2140 free_root_extent_buffers(info->tree_root);
2141
2142 free_global_root_pointers(info);
2143 free_root_extent_buffers(info->dev_root);
2144 free_root_extent_buffers(info->quota_root);
2145 free_root_extent_buffers(info->uuid_root);
2146 free_root_extent_buffers(info->fs_root);
2147 free_root_extent_buffers(info->data_reloc_root);
2148 free_root_extent_buffers(info->block_group_root);
2149 if (free_chunk_root)
2150 free_root_extent_buffers(info->chunk_root);
2151 }
2152
btrfs_put_root(struct btrfs_root * root)2153 void btrfs_put_root(struct btrfs_root *root)
2154 {
2155 if (!root)
2156 return;
2157
2158 if (refcount_dec_and_test(&root->refs)) {
2159 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2160 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2161 if (root->anon_dev)
2162 free_anon_bdev(root->anon_dev);
2163 btrfs_drew_lock_destroy(&root->snapshot_lock);
2164 free_root_extent_buffers(root);
2165 #ifdef CONFIG_BTRFS_DEBUG
2166 spin_lock(&root->fs_info->fs_roots_radix_lock);
2167 list_del_init(&root->leak_list);
2168 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2169 #endif
2170 kfree(root);
2171 }
2172 }
2173
btrfs_free_fs_roots(struct btrfs_fs_info * fs_info)2174 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2175 {
2176 int ret;
2177 struct btrfs_root *gang[8];
2178 int i;
2179
2180 while (!list_empty(&fs_info->dead_roots)) {
2181 gang[0] = list_entry(fs_info->dead_roots.next,
2182 struct btrfs_root, root_list);
2183 list_del(&gang[0]->root_list);
2184
2185 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2186 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2187 btrfs_put_root(gang[0]);
2188 }
2189
2190 while (1) {
2191 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2192 (void **)gang, 0,
2193 ARRAY_SIZE(gang));
2194 if (!ret)
2195 break;
2196 for (i = 0; i < ret; i++)
2197 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2198 }
2199 }
2200
btrfs_init_scrub(struct btrfs_fs_info * fs_info)2201 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2202 {
2203 mutex_init(&fs_info->scrub_lock);
2204 atomic_set(&fs_info->scrubs_running, 0);
2205 atomic_set(&fs_info->scrub_pause_req, 0);
2206 atomic_set(&fs_info->scrubs_paused, 0);
2207 atomic_set(&fs_info->scrub_cancel_req, 0);
2208 init_waitqueue_head(&fs_info->scrub_pause_wait);
2209 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2210 }
2211
btrfs_init_balance(struct btrfs_fs_info * fs_info)2212 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2213 {
2214 spin_lock_init(&fs_info->balance_lock);
2215 mutex_init(&fs_info->balance_mutex);
2216 atomic_set(&fs_info->balance_pause_req, 0);
2217 atomic_set(&fs_info->balance_cancel_req, 0);
2218 fs_info->balance_ctl = NULL;
2219 init_waitqueue_head(&fs_info->balance_wait_q);
2220 atomic_set(&fs_info->reloc_cancel_req, 0);
2221 }
2222
btrfs_init_btree_inode(struct btrfs_fs_info * fs_info)2223 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2224 {
2225 struct inode *inode = fs_info->btree_inode;
2226 unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
2227 fs_info->tree_root);
2228
2229 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2230 set_nlink(inode, 1);
2231 /*
2232 * we set the i_size on the btree inode to the max possible int.
2233 * the real end of the address space is determined by all of
2234 * the devices in the system
2235 */
2236 inode->i_size = OFFSET_MAX;
2237 inode->i_mapping->a_ops = &btree_aops;
2238
2239 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2240 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2241 IO_TREE_BTREE_INODE_IO, NULL);
2242 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2243
2244 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2245 BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID;
2246 BTRFS_I(inode)->location.type = 0;
2247 BTRFS_I(inode)->location.offset = 0;
2248 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2249 __insert_inode_hash(inode, hash);
2250 }
2251
btrfs_init_dev_replace_locks(struct btrfs_fs_info * fs_info)2252 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2253 {
2254 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2255 init_rwsem(&fs_info->dev_replace.rwsem);
2256 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2257 }
2258
btrfs_init_qgroup(struct btrfs_fs_info * fs_info)2259 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2260 {
2261 spin_lock_init(&fs_info->qgroup_lock);
2262 mutex_init(&fs_info->qgroup_ioctl_lock);
2263 fs_info->qgroup_tree = RB_ROOT;
2264 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2265 fs_info->qgroup_seq = 1;
2266 fs_info->qgroup_ulist = NULL;
2267 fs_info->qgroup_rescan_running = false;
2268 fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
2269 mutex_init(&fs_info->qgroup_rescan_lock);
2270 }
2271
btrfs_init_workqueues(struct btrfs_fs_info * fs_info)2272 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
2273 {
2274 u32 max_active = fs_info->thread_pool_size;
2275 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2276
2277 fs_info->workers =
2278 btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
2279 fs_info->hipri_workers =
2280 btrfs_alloc_workqueue(fs_info, "worker-high",
2281 flags | WQ_HIGHPRI, max_active, 16);
2282
2283 fs_info->delalloc_workers =
2284 btrfs_alloc_workqueue(fs_info, "delalloc",
2285 flags, max_active, 2);
2286
2287 fs_info->flush_workers =
2288 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2289 flags, max_active, 0);
2290
2291 fs_info->caching_workers =
2292 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2293
2294 fs_info->fixup_workers =
2295 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2296
2297 fs_info->endio_workers =
2298 alloc_workqueue("btrfs-endio", flags, max_active);
2299 fs_info->endio_meta_workers =
2300 alloc_workqueue("btrfs-endio-meta", flags, max_active);
2301 fs_info->endio_raid56_workers =
2302 alloc_workqueue("btrfs-endio-raid56", flags, max_active);
2303 fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
2304 fs_info->endio_write_workers =
2305 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2306 max_active, 2);
2307 fs_info->compressed_write_workers =
2308 alloc_workqueue("btrfs-compressed-write", flags, max_active);
2309 fs_info->endio_freespace_worker =
2310 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2311 max_active, 0);
2312 fs_info->delayed_workers =
2313 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2314 max_active, 0);
2315 fs_info->qgroup_rescan_workers =
2316 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2317 fs_info->discard_ctl.discard_workers =
2318 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2319
2320 if (!(fs_info->workers && fs_info->hipri_workers &&
2321 fs_info->delalloc_workers && fs_info->flush_workers &&
2322 fs_info->endio_workers && fs_info->endio_meta_workers &&
2323 fs_info->compressed_write_workers &&
2324 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2325 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2326 fs_info->caching_workers && fs_info->fixup_workers &&
2327 fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2328 fs_info->discard_ctl.discard_workers)) {
2329 return -ENOMEM;
2330 }
2331
2332 return 0;
2333 }
2334
btrfs_init_csum_hash(struct btrfs_fs_info * fs_info,u16 csum_type)2335 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2336 {
2337 struct crypto_shash *csum_shash;
2338 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2339
2340 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2341
2342 if (IS_ERR(csum_shash)) {
2343 btrfs_err(fs_info, "error allocating %s hash for checksum",
2344 csum_driver);
2345 return PTR_ERR(csum_shash);
2346 }
2347
2348 fs_info->csum_shash = csum_shash;
2349
2350 btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2351 btrfs_super_csum_name(csum_type),
2352 crypto_shash_driver_name(csum_shash));
2353 return 0;
2354 }
2355
btrfs_replay_log(struct btrfs_fs_info * fs_info,struct btrfs_fs_devices * fs_devices)2356 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2357 struct btrfs_fs_devices *fs_devices)
2358 {
2359 int ret;
2360 struct btrfs_root *log_tree_root;
2361 struct btrfs_super_block *disk_super = fs_info->super_copy;
2362 u64 bytenr = btrfs_super_log_root(disk_super);
2363 int level = btrfs_super_log_root_level(disk_super);
2364
2365 if (fs_devices->rw_devices == 0) {
2366 btrfs_warn(fs_info, "log replay required on RO media");
2367 return -EIO;
2368 }
2369
2370 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2371 GFP_KERNEL);
2372 if (!log_tree_root)
2373 return -ENOMEM;
2374
2375 log_tree_root->node = read_tree_block(fs_info, bytenr,
2376 BTRFS_TREE_LOG_OBJECTID,
2377 fs_info->generation + 1, level,
2378 NULL);
2379 if (IS_ERR(log_tree_root->node)) {
2380 btrfs_warn(fs_info, "failed to read log tree");
2381 ret = PTR_ERR(log_tree_root->node);
2382 log_tree_root->node = NULL;
2383 btrfs_put_root(log_tree_root);
2384 return ret;
2385 }
2386 if (!extent_buffer_uptodate(log_tree_root->node)) {
2387 btrfs_err(fs_info, "failed to read log tree");
2388 btrfs_put_root(log_tree_root);
2389 return -EIO;
2390 }
2391
2392 /* returns with log_tree_root freed on success */
2393 ret = btrfs_recover_log_trees(log_tree_root);
2394 if (ret) {
2395 btrfs_handle_fs_error(fs_info, ret,
2396 "Failed to recover log tree");
2397 btrfs_put_root(log_tree_root);
2398 return ret;
2399 }
2400
2401 if (sb_rdonly(fs_info->sb)) {
2402 ret = btrfs_commit_super(fs_info);
2403 if (ret)
2404 return ret;
2405 }
2406
2407 return 0;
2408 }
2409
load_global_roots_objectid(struct btrfs_root * tree_root,struct btrfs_path * path,u64 objectid,const char * name)2410 static int load_global_roots_objectid(struct btrfs_root *tree_root,
2411 struct btrfs_path *path, u64 objectid,
2412 const char *name)
2413 {
2414 struct btrfs_fs_info *fs_info = tree_root->fs_info;
2415 struct btrfs_root *root;
2416 u64 max_global_id = 0;
2417 int ret;
2418 struct btrfs_key key = {
2419 .objectid = objectid,
2420 .type = BTRFS_ROOT_ITEM_KEY,
2421 .offset = 0,
2422 };
2423 bool found = false;
2424
2425 /* If we have IGNOREDATACSUMS skip loading these roots. */
2426 if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2427 btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2428 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2429 return 0;
2430 }
2431
2432 while (1) {
2433 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2434 if (ret < 0)
2435 break;
2436
2437 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2438 ret = btrfs_next_leaf(tree_root, path);
2439 if (ret) {
2440 if (ret > 0)
2441 ret = 0;
2442 break;
2443 }
2444 }
2445 ret = 0;
2446
2447 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2448 if (key.objectid != objectid)
2449 break;
2450 btrfs_release_path(path);
2451
2452 /*
2453 * Just worry about this for extent tree, it'll be the same for
2454 * everybody.
2455 */
2456 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2457 max_global_id = max(max_global_id, key.offset);
2458
2459 found = true;
2460 root = read_tree_root_path(tree_root, path, &key);
2461 if (IS_ERR(root)) {
2462 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2463 ret = PTR_ERR(root);
2464 break;
2465 }
2466 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2467 ret = btrfs_global_root_insert(root);
2468 if (ret) {
2469 btrfs_put_root(root);
2470 break;
2471 }
2472 key.offset++;
2473 }
2474 btrfs_release_path(path);
2475
2476 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2477 fs_info->nr_global_roots = max_global_id + 1;
2478
2479 if (!found || ret) {
2480 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2481 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2482
2483 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2484 ret = ret ? ret : -ENOENT;
2485 else
2486 ret = 0;
2487 btrfs_err(fs_info, "failed to load root %s", name);
2488 }
2489 return ret;
2490 }
2491
load_global_roots(struct btrfs_root * tree_root)2492 static int load_global_roots(struct btrfs_root *tree_root)
2493 {
2494 struct btrfs_path *path;
2495 int ret = 0;
2496
2497 path = btrfs_alloc_path();
2498 if (!path)
2499 return -ENOMEM;
2500
2501 ret = load_global_roots_objectid(tree_root, path,
2502 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2503 if (ret)
2504 goto out;
2505 ret = load_global_roots_objectid(tree_root, path,
2506 BTRFS_CSUM_TREE_OBJECTID, "csum");
2507 if (ret)
2508 goto out;
2509 if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2510 goto out;
2511 ret = load_global_roots_objectid(tree_root, path,
2512 BTRFS_FREE_SPACE_TREE_OBJECTID,
2513 "free space");
2514 out:
2515 btrfs_free_path(path);
2516 return ret;
2517 }
2518
btrfs_read_roots(struct btrfs_fs_info * fs_info)2519 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2520 {
2521 struct btrfs_root *tree_root = fs_info->tree_root;
2522 struct btrfs_root *root;
2523 struct btrfs_key location;
2524 int ret;
2525
2526 BUG_ON(!fs_info->tree_root);
2527
2528 ret = load_global_roots(tree_root);
2529 if (ret)
2530 return ret;
2531
2532 location.type = BTRFS_ROOT_ITEM_KEY;
2533 location.offset = 0;
2534
2535 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2536 location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2537 root = btrfs_read_tree_root(tree_root, &location);
2538 if (IS_ERR(root)) {
2539 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2540 ret = PTR_ERR(root);
2541 goto out;
2542 }
2543 } else {
2544 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2545 fs_info->block_group_root = root;
2546 }
2547 }
2548
2549 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2550 root = btrfs_read_tree_root(tree_root, &location);
2551 if (IS_ERR(root)) {
2552 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2553 ret = PTR_ERR(root);
2554 goto out;
2555 }
2556 } else {
2557 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2558 fs_info->dev_root = root;
2559 }
2560 /* Initialize fs_info for all devices in any case */
2561 ret = btrfs_init_devices_late(fs_info);
2562 if (ret)
2563 goto out;
2564
2565 /*
2566 * This tree can share blocks with some other fs tree during relocation
2567 * and we need a proper setup by btrfs_get_fs_root
2568 */
2569 root = btrfs_get_fs_root(tree_root->fs_info,
2570 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2571 if (IS_ERR(root)) {
2572 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2573 ret = PTR_ERR(root);
2574 goto out;
2575 }
2576 } else {
2577 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2578 fs_info->data_reloc_root = root;
2579 }
2580
2581 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2582 root = btrfs_read_tree_root(tree_root, &location);
2583 if (!IS_ERR(root)) {
2584 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2585 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2586 fs_info->quota_root = root;
2587 }
2588
2589 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2590 root = btrfs_read_tree_root(tree_root, &location);
2591 if (IS_ERR(root)) {
2592 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2593 ret = PTR_ERR(root);
2594 if (ret != -ENOENT)
2595 goto out;
2596 }
2597 } else {
2598 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2599 fs_info->uuid_root = root;
2600 }
2601
2602 return 0;
2603 out:
2604 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2605 location.objectid, ret);
2606 return ret;
2607 }
2608
2609 /*
2610 * Real super block validation
2611 * NOTE: super csum type and incompat features will not be checked here.
2612 *
2613 * @sb: super block to check
2614 * @mirror_num: the super block number to check its bytenr:
2615 * 0 the primary (1st) sb
2616 * 1, 2 2nd and 3rd backup copy
2617 * -1 skip bytenr check
2618 */
btrfs_validate_super(struct btrfs_fs_info * fs_info,struct btrfs_super_block * sb,int mirror_num)2619 int btrfs_validate_super(struct btrfs_fs_info *fs_info,
2620 struct btrfs_super_block *sb, int mirror_num)
2621 {
2622 u64 nodesize = btrfs_super_nodesize(sb);
2623 u64 sectorsize = btrfs_super_sectorsize(sb);
2624 int ret = 0;
2625
2626 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2627 btrfs_err(fs_info, "no valid FS found");
2628 ret = -EINVAL;
2629 }
2630 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2631 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2632 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2633 ret = -EINVAL;
2634 }
2635 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2636 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2637 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2638 ret = -EINVAL;
2639 }
2640 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2641 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2642 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2643 ret = -EINVAL;
2644 }
2645 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2646 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2647 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2648 ret = -EINVAL;
2649 }
2650
2651 /*
2652 * Check sectorsize and nodesize first, other check will need it.
2653 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2654 */
2655 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2656 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2657 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2658 ret = -EINVAL;
2659 }
2660
2661 /*
2662 * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2663 *
2664 * We can support 16K sectorsize with 64K page size without problem,
2665 * but such sectorsize/pagesize combination doesn't make much sense.
2666 * 4K will be our future standard, PAGE_SIZE is supported from the very
2667 * beginning.
2668 */
2669 if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2670 btrfs_err(fs_info,
2671 "sectorsize %llu not yet supported for page size %lu",
2672 sectorsize, PAGE_SIZE);
2673 ret = -EINVAL;
2674 }
2675
2676 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2677 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2678 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2679 ret = -EINVAL;
2680 }
2681 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2682 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2683 le32_to_cpu(sb->__unused_leafsize), nodesize);
2684 ret = -EINVAL;
2685 }
2686
2687 /* Root alignment check */
2688 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2689 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2690 btrfs_super_root(sb));
2691 ret = -EINVAL;
2692 }
2693 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2694 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2695 btrfs_super_chunk_root(sb));
2696 ret = -EINVAL;
2697 }
2698 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2699 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2700 btrfs_super_log_root(sb));
2701 ret = -EINVAL;
2702 }
2703
2704 if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2705 BTRFS_FSID_SIZE)) {
2706 btrfs_err(fs_info,
2707 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2708 fs_info->super_copy->fsid, fs_info->fs_devices->fsid);
2709 ret = -EINVAL;
2710 }
2711
2712 if (btrfs_fs_incompat(fs_info, METADATA_UUID) &&
2713 memcmp(fs_info->fs_devices->metadata_uuid,
2714 fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) {
2715 btrfs_err(fs_info,
2716 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2717 fs_info->super_copy->metadata_uuid,
2718 fs_info->fs_devices->metadata_uuid);
2719 ret = -EINVAL;
2720 }
2721
2722 /*
2723 * Artificial requirement for block-group-tree to force newer features
2724 * (free-space-tree, no-holes) so the test matrix is smaller.
2725 */
2726 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2727 (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2728 !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2729 btrfs_err(fs_info,
2730 "block-group-tree feature requires fres-space-tree and no-holes");
2731 ret = -EINVAL;
2732 }
2733
2734 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2735 BTRFS_FSID_SIZE) != 0) {
2736 btrfs_err(fs_info,
2737 "dev_item UUID does not match metadata fsid: %pU != %pU",
2738 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2739 ret = -EINVAL;
2740 }
2741
2742 /*
2743 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2744 * done later
2745 */
2746 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2747 btrfs_err(fs_info, "bytes_used is too small %llu",
2748 btrfs_super_bytes_used(sb));
2749 ret = -EINVAL;
2750 }
2751 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2752 btrfs_err(fs_info, "invalid stripesize %u",
2753 btrfs_super_stripesize(sb));
2754 ret = -EINVAL;
2755 }
2756 if (btrfs_super_num_devices(sb) > (1UL << 31))
2757 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2758 btrfs_super_num_devices(sb));
2759 if (btrfs_super_num_devices(sb) == 0) {
2760 btrfs_err(fs_info, "number of devices is 0");
2761 ret = -EINVAL;
2762 }
2763
2764 if (mirror_num >= 0 &&
2765 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2766 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2767 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2768 ret = -EINVAL;
2769 }
2770
2771 /*
2772 * Obvious sys_chunk_array corruptions, it must hold at least one key
2773 * and one chunk
2774 */
2775 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2776 btrfs_err(fs_info, "system chunk array too big %u > %u",
2777 btrfs_super_sys_array_size(sb),
2778 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2779 ret = -EINVAL;
2780 }
2781 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2782 + sizeof(struct btrfs_chunk)) {
2783 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2784 btrfs_super_sys_array_size(sb),
2785 sizeof(struct btrfs_disk_key)
2786 + sizeof(struct btrfs_chunk));
2787 ret = -EINVAL;
2788 }
2789
2790 /*
2791 * The generation is a global counter, we'll trust it more than the others
2792 * but it's still possible that it's the one that's wrong.
2793 */
2794 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2795 btrfs_warn(fs_info,
2796 "suspicious: generation < chunk_root_generation: %llu < %llu",
2797 btrfs_super_generation(sb),
2798 btrfs_super_chunk_root_generation(sb));
2799 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2800 && btrfs_super_cache_generation(sb) != (u64)-1)
2801 btrfs_warn(fs_info,
2802 "suspicious: generation < cache_generation: %llu < %llu",
2803 btrfs_super_generation(sb),
2804 btrfs_super_cache_generation(sb));
2805
2806 return ret;
2807 }
2808
2809 /*
2810 * Validation of super block at mount time.
2811 * Some checks already done early at mount time, like csum type and incompat
2812 * flags will be skipped.
2813 */
btrfs_validate_mount_super(struct btrfs_fs_info * fs_info)2814 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2815 {
2816 return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2817 }
2818
2819 /*
2820 * Validation of super block at write time.
2821 * Some checks like bytenr check will be skipped as their values will be
2822 * overwritten soon.
2823 * Extra checks like csum type and incompat flags will be done here.
2824 */
btrfs_validate_write_super(struct btrfs_fs_info * fs_info,struct btrfs_super_block * sb)2825 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2826 struct btrfs_super_block *sb)
2827 {
2828 int ret;
2829
2830 ret = btrfs_validate_super(fs_info, sb, -1);
2831 if (ret < 0)
2832 goto out;
2833 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2834 ret = -EUCLEAN;
2835 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2836 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2837 goto out;
2838 }
2839 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2840 ret = -EUCLEAN;
2841 btrfs_err(fs_info,
2842 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2843 btrfs_super_incompat_flags(sb),
2844 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2845 goto out;
2846 }
2847 out:
2848 if (ret < 0)
2849 btrfs_err(fs_info,
2850 "super block corruption detected before writing it to disk");
2851 return ret;
2852 }
2853
load_super_root(struct btrfs_root * root,u64 bytenr,u64 gen,int level)2854 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2855 {
2856 int ret = 0;
2857
2858 root->node = read_tree_block(root->fs_info, bytenr,
2859 root->root_key.objectid, gen, level, NULL);
2860 if (IS_ERR(root->node)) {
2861 ret = PTR_ERR(root->node);
2862 root->node = NULL;
2863 return ret;
2864 }
2865 if (!extent_buffer_uptodate(root->node)) {
2866 free_extent_buffer(root->node);
2867 root->node = NULL;
2868 return -EIO;
2869 }
2870
2871 btrfs_set_root_node(&root->root_item, root->node);
2872 root->commit_root = btrfs_root_node(root);
2873 btrfs_set_root_refs(&root->root_item, 1);
2874 return ret;
2875 }
2876
load_important_roots(struct btrfs_fs_info * fs_info)2877 static int load_important_roots(struct btrfs_fs_info *fs_info)
2878 {
2879 struct btrfs_super_block *sb = fs_info->super_copy;
2880 u64 gen, bytenr;
2881 int level, ret;
2882
2883 bytenr = btrfs_super_root(sb);
2884 gen = btrfs_super_generation(sb);
2885 level = btrfs_super_root_level(sb);
2886 ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2887 if (ret) {
2888 btrfs_warn(fs_info, "couldn't read tree root");
2889 return ret;
2890 }
2891 return 0;
2892 }
2893
init_tree_roots(struct btrfs_fs_info * fs_info)2894 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2895 {
2896 int backup_index = find_newest_super_backup(fs_info);
2897 struct btrfs_super_block *sb = fs_info->super_copy;
2898 struct btrfs_root *tree_root = fs_info->tree_root;
2899 bool handle_error = false;
2900 int ret = 0;
2901 int i;
2902
2903 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2904 if (handle_error) {
2905 if (!IS_ERR(tree_root->node))
2906 free_extent_buffer(tree_root->node);
2907 tree_root->node = NULL;
2908
2909 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2910 break;
2911
2912 free_root_pointers(fs_info, 0);
2913
2914 /*
2915 * Don't use the log in recovery mode, it won't be
2916 * valid
2917 */
2918 btrfs_set_super_log_root(sb, 0);
2919
2920 /* We can't trust the free space cache either */
2921 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2922
2923 ret = read_backup_root(fs_info, i);
2924 backup_index = ret;
2925 if (ret < 0)
2926 return ret;
2927 }
2928
2929 ret = load_important_roots(fs_info);
2930 if (ret) {
2931 handle_error = true;
2932 continue;
2933 }
2934
2935 /*
2936 * No need to hold btrfs_root::objectid_mutex since the fs
2937 * hasn't been fully initialised and we are the only user
2938 */
2939 ret = btrfs_init_root_free_objectid(tree_root);
2940 if (ret < 0) {
2941 handle_error = true;
2942 continue;
2943 }
2944
2945 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2946
2947 ret = btrfs_read_roots(fs_info);
2948 if (ret < 0) {
2949 handle_error = true;
2950 continue;
2951 }
2952
2953 /* All successful */
2954 fs_info->generation = btrfs_header_generation(tree_root->node);
2955 fs_info->last_trans_committed = fs_info->generation;
2956 fs_info->last_reloc_trans = 0;
2957
2958 /* Always begin writing backup roots after the one being used */
2959 if (backup_index < 0) {
2960 fs_info->backup_root_index = 0;
2961 } else {
2962 fs_info->backup_root_index = backup_index + 1;
2963 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2964 }
2965 break;
2966 }
2967
2968 return ret;
2969 }
2970
btrfs_init_fs_info(struct btrfs_fs_info * fs_info)2971 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2972 {
2973 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2974 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2975 INIT_LIST_HEAD(&fs_info->trans_list);
2976 INIT_LIST_HEAD(&fs_info->dead_roots);
2977 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2978 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2979 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2980 spin_lock_init(&fs_info->delalloc_root_lock);
2981 spin_lock_init(&fs_info->trans_lock);
2982 spin_lock_init(&fs_info->fs_roots_radix_lock);
2983 spin_lock_init(&fs_info->delayed_iput_lock);
2984 spin_lock_init(&fs_info->defrag_inodes_lock);
2985 spin_lock_init(&fs_info->super_lock);
2986 spin_lock_init(&fs_info->buffer_lock);
2987 spin_lock_init(&fs_info->unused_bgs_lock);
2988 spin_lock_init(&fs_info->treelog_bg_lock);
2989 spin_lock_init(&fs_info->zone_active_bgs_lock);
2990 spin_lock_init(&fs_info->relocation_bg_lock);
2991 rwlock_init(&fs_info->tree_mod_log_lock);
2992 rwlock_init(&fs_info->global_root_lock);
2993 mutex_init(&fs_info->unused_bg_unpin_mutex);
2994 mutex_init(&fs_info->reclaim_bgs_lock);
2995 mutex_init(&fs_info->reloc_mutex);
2996 mutex_init(&fs_info->delalloc_root_mutex);
2997 mutex_init(&fs_info->zoned_meta_io_lock);
2998 mutex_init(&fs_info->zoned_data_reloc_io_lock);
2999 seqlock_init(&fs_info->profiles_lock);
3000
3001 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
3002 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
3003 btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
3004 btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
3005 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_start,
3006 BTRFS_LOCKDEP_TRANS_COMMIT_START);
3007 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
3008 BTRFS_LOCKDEP_TRANS_UNBLOCKED);
3009 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
3010 BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
3011 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
3012 BTRFS_LOCKDEP_TRANS_COMPLETED);
3013
3014 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
3015 INIT_LIST_HEAD(&fs_info->space_info);
3016 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
3017 INIT_LIST_HEAD(&fs_info->unused_bgs);
3018 INIT_LIST_HEAD(&fs_info->reclaim_bgs);
3019 INIT_LIST_HEAD(&fs_info->zone_active_bgs);
3020 #ifdef CONFIG_BTRFS_DEBUG
3021 INIT_LIST_HEAD(&fs_info->allocated_roots);
3022 INIT_LIST_HEAD(&fs_info->allocated_ebs);
3023 spin_lock_init(&fs_info->eb_leak_lock);
3024 #endif
3025 extent_map_tree_init(&fs_info->mapping_tree);
3026 btrfs_init_block_rsv(&fs_info->global_block_rsv,
3027 BTRFS_BLOCK_RSV_GLOBAL);
3028 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
3029 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
3030 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
3031 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
3032 BTRFS_BLOCK_RSV_DELOPS);
3033 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
3034 BTRFS_BLOCK_RSV_DELREFS);
3035
3036 atomic_set(&fs_info->async_delalloc_pages, 0);
3037 atomic_set(&fs_info->defrag_running, 0);
3038 atomic_set(&fs_info->nr_delayed_iputs, 0);
3039 atomic64_set(&fs_info->tree_mod_seq, 0);
3040 fs_info->global_root_tree = RB_ROOT;
3041 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
3042 fs_info->metadata_ratio = 0;
3043 fs_info->defrag_inodes = RB_ROOT;
3044 atomic64_set(&fs_info->free_chunk_space, 0);
3045 fs_info->tree_mod_log = RB_ROOT;
3046 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
3047 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
3048 btrfs_init_ref_verify(fs_info);
3049
3050 fs_info->thread_pool_size = min_t(unsigned long,
3051 num_online_cpus() + 2, 8);
3052
3053 INIT_LIST_HEAD(&fs_info->ordered_roots);
3054 spin_lock_init(&fs_info->ordered_root_lock);
3055
3056 btrfs_init_scrub(fs_info);
3057 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3058 fs_info->check_integrity_print_mask = 0;
3059 #endif
3060 btrfs_init_balance(fs_info);
3061 btrfs_init_async_reclaim_work(fs_info);
3062
3063 rwlock_init(&fs_info->block_group_cache_lock);
3064 fs_info->block_group_cache_tree = RB_ROOT_CACHED;
3065
3066 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
3067 IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
3068
3069 mutex_init(&fs_info->ordered_operations_mutex);
3070 mutex_init(&fs_info->tree_log_mutex);
3071 mutex_init(&fs_info->chunk_mutex);
3072 mutex_init(&fs_info->transaction_kthread_mutex);
3073 mutex_init(&fs_info->cleaner_mutex);
3074 mutex_init(&fs_info->ro_block_group_mutex);
3075 init_rwsem(&fs_info->commit_root_sem);
3076 init_rwsem(&fs_info->cleanup_work_sem);
3077 init_rwsem(&fs_info->subvol_sem);
3078 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
3079
3080 btrfs_init_dev_replace_locks(fs_info);
3081 btrfs_init_qgroup(fs_info);
3082 btrfs_discard_init(fs_info);
3083
3084 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
3085 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
3086
3087 init_waitqueue_head(&fs_info->transaction_throttle);
3088 init_waitqueue_head(&fs_info->transaction_wait);
3089 init_waitqueue_head(&fs_info->transaction_blocked_wait);
3090 init_waitqueue_head(&fs_info->async_submit_wait);
3091 init_waitqueue_head(&fs_info->delayed_iputs_wait);
3092
3093 /* Usable values until the real ones are cached from the superblock */
3094 fs_info->nodesize = 4096;
3095 fs_info->sectorsize = 4096;
3096 fs_info->sectorsize_bits = ilog2(4096);
3097 fs_info->stripesize = 4096;
3098
3099 fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
3100
3101 spin_lock_init(&fs_info->swapfile_pins_lock);
3102 fs_info->swapfile_pins = RB_ROOT;
3103
3104 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
3105 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
3106 }
3107
init_mount_fs_info(struct btrfs_fs_info * fs_info,struct super_block * sb)3108 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
3109 {
3110 int ret;
3111
3112 fs_info->sb = sb;
3113 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
3114 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
3115
3116 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
3117 if (ret)
3118 return ret;
3119
3120 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
3121 if (ret)
3122 return ret;
3123
3124 fs_info->dirty_metadata_batch = PAGE_SIZE *
3125 (1 + ilog2(nr_cpu_ids));
3126
3127 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
3128 if (ret)
3129 return ret;
3130
3131 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
3132 GFP_KERNEL);
3133 if (ret)
3134 return ret;
3135
3136 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
3137 GFP_KERNEL);
3138 if (!fs_info->delayed_root)
3139 return -ENOMEM;
3140 btrfs_init_delayed_root(fs_info->delayed_root);
3141
3142 if (sb_rdonly(sb))
3143 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
3144
3145 return btrfs_alloc_stripe_hash_table(fs_info);
3146 }
3147
btrfs_uuid_rescan_kthread(void * data)3148 static int btrfs_uuid_rescan_kthread(void *data)
3149 {
3150 struct btrfs_fs_info *fs_info = data;
3151 int ret;
3152
3153 /*
3154 * 1st step is to iterate through the existing UUID tree and
3155 * to delete all entries that contain outdated data.
3156 * 2nd step is to add all missing entries to the UUID tree.
3157 */
3158 ret = btrfs_uuid_tree_iterate(fs_info);
3159 if (ret < 0) {
3160 if (ret != -EINTR)
3161 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
3162 ret);
3163 up(&fs_info->uuid_tree_rescan_sem);
3164 return ret;
3165 }
3166 return btrfs_uuid_scan_kthread(data);
3167 }
3168
btrfs_check_uuid_tree(struct btrfs_fs_info * fs_info)3169 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3170 {
3171 struct task_struct *task;
3172
3173 down(&fs_info->uuid_tree_rescan_sem);
3174 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3175 if (IS_ERR(task)) {
3176 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3177 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3178 up(&fs_info->uuid_tree_rescan_sem);
3179 return PTR_ERR(task);
3180 }
3181
3182 return 0;
3183 }
3184
3185 /*
3186 * Some options only have meaning at mount time and shouldn't persist across
3187 * remounts, or be displayed. Clear these at the end of mount and remount
3188 * code paths.
3189 */
btrfs_clear_oneshot_options(struct btrfs_fs_info * fs_info)3190 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
3191 {
3192 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3193 btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
3194 }
3195
3196 /*
3197 * Mounting logic specific to read-write file systems. Shared by open_ctree
3198 * and btrfs_remount when remounting from read-only to read-write.
3199 */
btrfs_start_pre_rw_mount(struct btrfs_fs_info * fs_info)3200 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3201 {
3202 int ret;
3203 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3204 bool clear_free_space_tree = false;
3205
3206 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3207 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3208 clear_free_space_tree = true;
3209 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3210 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3211 btrfs_warn(fs_info, "free space tree is invalid");
3212 clear_free_space_tree = true;
3213 }
3214
3215 if (clear_free_space_tree) {
3216 btrfs_info(fs_info, "clearing free space tree");
3217 ret = btrfs_clear_free_space_tree(fs_info);
3218 if (ret) {
3219 btrfs_warn(fs_info,
3220 "failed to clear free space tree: %d", ret);
3221 goto out;
3222 }
3223 }
3224
3225 /*
3226 * btrfs_find_orphan_roots() is responsible for finding all the dead
3227 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3228 * them into the fs_info->fs_roots_radix tree. This must be done before
3229 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3230 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3231 * item before the root's tree is deleted - this means that if we unmount
3232 * or crash before the deletion completes, on the next mount we will not
3233 * delete what remains of the tree because the orphan item does not
3234 * exists anymore, which is what tells us we have a pending deletion.
3235 */
3236 ret = btrfs_find_orphan_roots(fs_info);
3237 if (ret)
3238 goto out;
3239
3240 ret = btrfs_cleanup_fs_roots(fs_info);
3241 if (ret)
3242 goto out;
3243
3244 down_read(&fs_info->cleanup_work_sem);
3245 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3246 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3247 up_read(&fs_info->cleanup_work_sem);
3248 goto out;
3249 }
3250 up_read(&fs_info->cleanup_work_sem);
3251
3252 mutex_lock(&fs_info->cleaner_mutex);
3253 ret = btrfs_recover_relocation(fs_info);
3254 mutex_unlock(&fs_info->cleaner_mutex);
3255 if (ret < 0) {
3256 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3257 goto out;
3258 }
3259
3260 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3261 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3262 btrfs_info(fs_info, "creating free space tree");
3263 ret = btrfs_create_free_space_tree(fs_info);
3264 if (ret) {
3265 btrfs_warn(fs_info,
3266 "failed to create free space tree: %d", ret);
3267 goto out;
3268 }
3269 }
3270
3271 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3272 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3273 if (ret)
3274 goto out;
3275 }
3276
3277 ret = btrfs_resume_balance_async(fs_info);
3278 if (ret)
3279 goto out;
3280
3281 ret = btrfs_resume_dev_replace_async(fs_info);
3282 if (ret) {
3283 btrfs_warn(fs_info, "failed to resume dev_replace");
3284 goto out;
3285 }
3286
3287 btrfs_qgroup_rescan_resume(fs_info);
3288
3289 if (!fs_info->uuid_root) {
3290 btrfs_info(fs_info, "creating UUID tree");
3291 ret = btrfs_create_uuid_tree(fs_info);
3292 if (ret) {
3293 btrfs_warn(fs_info,
3294 "failed to create the UUID tree %d", ret);
3295 goto out;
3296 }
3297 }
3298
3299 out:
3300 return ret;
3301 }
3302
3303 /*
3304 * Do various sanity and dependency checks of different features.
3305 *
3306 * @is_rw_mount: If the mount is read-write.
3307 *
3308 * This is the place for less strict checks (like for subpage or artificial
3309 * feature dependencies).
3310 *
3311 * For strict checks or possible corruption detection, see
3312 * btrfs_validate_super().
3313 *
3314 * This should be called after btrfs_parse_options(), as some mount options
3315 * (space cache related) can modify on-disk format like free space tree and
3316 * screw up certain feature dependencies.
3317 */
btrfs_check_features(struct btrfs_fs_info * fs_info,bool is_rw_mount)3318 int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3319 {
3320 struct btrfs_super_block *disk_super = fs_info->super_copy;
3321 u64 incompat = btrfs_super_incompat_flags(disk_super);
3322 const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3323 const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3324
3325 if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3326 btrfs_err(fs_info,
3327 "cannot mount because of unknown incompat features (0x%llx)",
3328 incompat);
3329 return -EINVAL;
3330 }
3331
3332 /* Runtime limitation for mixed block groups. */
3333 if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3334 (fs_info->sectorsize != fs_info->nodesize)) {
3335 btrfs_err(fs_info,
3336 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3337 fs_info->nodesize, fs_info->sectorsize);
3338 return -EINVAL;
3339 }
3340
3341 /* Mixed backref is an always-enabled feature. */
3342 incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3343
3344 /* Set compression related flags just in case. */
3345 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3346 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3347 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3348 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3349
3350 /*
3351 * An ancient flag, which should really be marked deprecated.
3352 * Such runtime limitation doesn't really need a incompat flag.
3353 */
3354 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3355 incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3356
3357 if (compat_ro_unsupp && is_rw_mount) {
3358 btrfs_err(fs_info,
3359 "cannot mount read-write because of unknown compat_ro features (0x%llx)",
3360 compat_ro);
3361 return -EINVAL;
3362 }
3363
3364 /*
3365 * We have unsupported RO compat features, although RO mounted, we
3366 * should not cause any metadata writes, including log replay.
3367 * Or we could screw up whatever the new feature requires.
3368 */
3369 if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3370 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3371 btrfs_err(fs_info,
3372 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3373 compat_ro);
3374 return -EINVAL;
3375 }
3376
3377 /*
3378 * Artificial limitations for block group tree, to force
3379 * block-group-tree to rely on no-holes and free-space-tree.
3380 */
3381 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3382 (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3383 !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3384 btrfs_err(fs_info,
3385 "block-group-tree feature requires no-holes and free-space-tree features");
3386 return -EINVAL;
3387 }
3388
3389 /*
3390 * Subpage runtime limitation on v1 cache.
3391 *
3392 * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3393 * we're already defaulting to v2 cache, no need to bother v1 as it's
3394 * going to be deprecated anyway.
3395 */
3396 if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3397 btrfs_warn(fs_info,
3398 "v1 space cache is not supported for page size %lu with sectorsize %u",
3399 PAGE_SIZE, fs_info->sectorsize);
3400 return -EINVAL;
3401 }
3402
3403 /* This can be called by remount, we need to protect the super block. */
3404 spin_lock(&fs_info->super_lock);
3405 btrfs_set_super_incompat_flags(disk_super, incompat);
3406 spin_unlock(&fs_info->super_lock);
3407
3408 return 0;
3409 }
3410
open_ctree(struct super_block * sb,struct btrfs_fs_devices * fs_devices,char * options)3411 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3412 char *options)
3413 {
3414 u32 sectorsize;
3415 u32 nodesize;
3416 u32 stripesize;
3417 u64 generation;
3418 u64 features;
3419 u16 csum_type;
3420 struct btrfs_super_block *disk_super;
3421 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3422 struct btrfs_root *tree_root;
3423 struct btrfs_root *chunk_root;
3424 int ret;
3425 int err = -EINVAL;
3426 int level;
3427
3428 ret = init_mount_fs_info(fs_info, sb);
3429 if (ret) {
3430 err = ret;
3431 goto fail;
3432 }
3433
3434 /* These need to be init'ed before we start creating inodes and such. */
3435 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3436 GFP_KERNEL);
3437 fs_info->tree_root = tree_root;
3438 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3439 GFP_KERNEL);
3440 fs_info->chunk_root = chunk_root;
3441 if (!tree_root || !chunk_root) {
3442 err = -ENOMEM;
3443 goto fail;
3444 }
3445
3446 fs_info->btree_inode = new_inode(sb);
3447 if (!fs_info->btree_inode) {
3448 err = -ENOMEM;
3449 goto fail;
3450 }
3451 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
3452 btrfs_init_btree_inode(fs_info);
3453
3454 invalidate_bdev(fs_devices->latest_dev->bdev);
3455
3456 /*
3457 * Read super block and check the signature bytes only
3458 */
3459 disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3460 if (IS_ERR(disk_super)) {
3461 err = PTR_ERR(disk_super);
3462 goto fail_alloc;
3463 }
3464
3465 /*
3466 * Verify the type first, if that or the checksum value are
3467 * corrupted, we'll find out
3468 */
3469 csum_type = btrfs_super_csum_type(disk_super);
3470 if (!btrfs_supported_super_csum(csum_type)) {
3471 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3472 csum_type);
3473 err = -EINVAL;
3474 btrfs_release_disk_super(disk_super);
3475 goto fail_alloc;
3476 }
3477
3478 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3479
3480 ret = btrfs_init_csum_hash(fs_info, csum_type);
3481 if (ret) {
3482 err = ret;
3483 btrfs_release_disk_super(disk_super);
3484 goto fail_alloc;
3485 }
3486
3487 /*
3488 * We want to check superblock checksum, the type is stored inside.
3489 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3490 */
3491 if (btrfs_check_super_csum(fs_info, disk_super)) {
3492 btrfs_err(fs_info, "superblock checksum mismatch");
3493 err = -EINVAL;
3494 btrfs_release_disk_super(disk_super);
3495 goto fail_alloc;
3496 }
3497
3498 /*
3499 * super_copy is zeroed at allocation time and we never touch the
3500 * following bytes up to INFO_SIZE, the checksum is calculated from
3501 * the whole block of INFO_SIZE
3502 */
3503 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3504 btrfs_release_disk_super(disk_super);
3505
3506 disk_super = fs_info->super_copy;
3507
3508
3509 features = btrfs_super_flags(disk_super);
3510 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3511 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3512 btrfs_set_super_flags(disk_super, features);
3513 btrfs_info(fs_info,
3514 "found metadata UUID change in progress flag, clearing");
3515 }
3516
3517 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3518 sizeof(*fs_info->super_for_commit));
3519
3520 ret = btrfs_validate_mount_super(fs_info);
3521 if (ret) {
3522 btrfs_err(fs_info, "superblock contains fatal errors");
3523 err = -EINVAL;
3524 goto fail_alloc;
3525 }
3526
3527 if (!btrfs_super_root(disk_super))
3528 goto fail_alloc;
3529
3530 /* check FS state, whether FS is broken. */
3531 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3532 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3533
3534 /*
3535 * In the long term, we'll store the compression type in the super
3536 * block, and it'll be used for per file compression control.
3537 */
3538 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3539
3540
3541 /* Set up fs_info before parsing mount options */
3542 nodesize = btrfs_super_nodesize(disk_super);
3543 sectorsize = btrfs_super_sectorsize(disk_super);
3544 stripesize = sectorsize;
3545 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3546 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3547
3548 fs_info->nodesize = nodesize;
3549 fs_info->sectorsize = sectorsize;
3550 fs_info->sectorsize_bits = ilog2(sectorsize);
3551 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3552 fs_info->stripesize = stripesize;
3553
3554 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3555 if (ret) {
3556 err = ret;
3557 goto fail_alloc;
3558 }
3559
3560 ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3561 if (ret < 0) {
3562 err = ret;
3563 goto fail_alloc;
3564 }
3565
3566 if (sectorsize < PAGE_SIZE) {
3567 struct btrfs_subpage_info *subpage_info;
3568
3569 /*
3570 * V1 space cache has some hardcoded PAGE_SIZE usage, and is
3571 * going to be deprecated.
3572 *
3573 * Force to use v2 cache for subpage case.
3574 */
3575 btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
3576 btrfs_set_and_info(fs_info, FREE_SPACE_TREE,
3577 "forcing free space tree for sector size %u with page size %lu",
3578 sectorsize, PAGE_SIZE);
3579
3580 btrfs_warn(fs_info,
3581 "read-write for sector size %u with page size %lu is experimental",
3582 sectorsize, PAGE_SIZE);
3583 subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3584 if (!subpage_info)
3585 goto fail_alloc;
3586 btrfs_init_subpage_info(subpage_info, sectorsize);
3587 fs_info->subpage_info = subpage_info;
3588 }
3589
3590 ret = btrfs_init_workqueues(fs_info);
3591 if (ret) {
3592 err = ret;
3593 goto fail_sb_buffer;
3594 }
3595
3596 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3597 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3598
3599 sb->s_blocksize = sectorsize;
3600 sb->s_blocksize_bits = blksize_bits(sectorsize);
3601 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3602
3603 mutex_lock(&fs_info->chunk_mutex);
3604 ret = btrfs_read_sys_array(fs_info);
3605 mutex_unlock(&fs_info->chunk_mutex);
3606 if (ret) {
3607 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3608 goto fail_sb_buffer;
3609 }
3610
3611 generation = btrfs_super_chunk_root_generation(disk_super);
3612 level = btrfs_super_chunk_root_level(disk_super);
3613 ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3614 generation, level);
3615 if (ret) {
3616 btrfs_err(fs_info, "failed to read chunk root");
3617 goto fail_tree_roots;
3618 }
3619
3620 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3621 offsetof(struct btrfs_header, chunk_tree_uuid),
3622 BTRFS_UUID_SIZE);
3623
3624 ret = btrfs_read_chunk_tree(fs_info);
3625 if (ret) {
3626 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3627 goto fail_tree_roots;
3628 }
3629
3630 /*
3631 * At this point we know all the devices that make this filesystem,
3632 * including the seed devices but we don't know yet if the replace
3633 * target is required. So free devices that are not part of this
3634 * filesystem but skip the replace target device which is checked
3635 * below in btrfs_init_dev_replace().
3636 */
3637 btrfs_free_extra_devids(fs_devices);
3638 if (!fs_devices->latest_dev->bdev) {
3639 btrfs_err(fs_info, "failed to read devices");
3640 goto fail_tree_roots;
3641 }
3642
3643 ret = init_tree_roots(fs_info);
3644 if (ret)
3645 goto fail_tree_roots;
3646
3647 /*
3648 * Get zone type information of zoned block devices. This will also
3649 * handle emulation of a zoned filesystem if a regular device has the
3650 * zoned incompat feature flag set.
3651 */
3652 ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3653 if (ret) {
3654 btrfs_err(fs_info,
3655 "zoned: failed to read device zone info: %d",
3656 ret);
3657 goto fail_block_groups;
3658 }
3659
3660 /*
3661 * If we have a uuid root and we're not being told to rescan we need to
3662 * check the generation here so we can set the
3663 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3664 * transaction during a balance or the log replay without updating the
3665 * uuid generation, and then if we crash we would rescan the uuid tree,
3666 * even though it was perfectly fine.
3667 */
3668 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3669 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3670 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3671
3672 ret = btrfs_verify_dev_extents(fs_info);
3673 if (ret) {
3674 btrfs_err(fs_info,
3675 "failed to verify dev extents against chunks: %d",
3676 ret);
3677 goto fail_block_groups;
3678 }
3679 ret = btrfs_recover_balance(fs_info);
3680 if (ret) {
3681 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3682 goto fail_block_groups;
3683 }
3684
3685 ret = btrfs_init_dev_stats(fs_info);
3686 if (ret) {
3687 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3688 goto fail_block_groups;
3689 }
3690
3691 ret = btrfs_init_dev_replace(fs_info);
3692 if (ret) {
3693 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3694 goto fail_block_groups;
3695 }
3696
3697 ret = btrfs_check_zoned_mode(fs_info);
3698 if (ret) {
3699 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3700 ret);
3701 goto fail_block_groups;
3702 }
3703
3704 ret = btrfs_sysfs_add_fsid(fs_devices);
3705 if (ret) {
3706 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3707 ret);
3708 goto fail_block_groups;
3709 }
3710
3711 ret = btrfs_sysfs_add_mounted(fs_info);
3712 if (ret) {
3713 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3714 goto fail_fsdev_sysfs;
3715 }
3716
3717 ret = btrfs_init_space_info(fs_info);
3718 if (ret) {
3719 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3720 goto fail_sysfs;
3721 }
3722
3723 ret = btrfs_read_block_groups(fs_info);
3724 if (ret) {
3725 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3726 goto fail_sysfs;
3727 }
3728
3729 btrfs_free_zone_cache(fs_info);
3730
3731 if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3732 !btrfs_check_rw_degradable(fs_info, NULL)) {
3733 btrfs_warn(fs_info,
3734 "writable mount is not allowed due to too many missing devices");
3735 goto fail_sysfs;
3736 }
3737
3738 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3739 "btrfs-cleaner");
3740 if (IS_ERR(fs_info->cleaner_kthread))
3741 goto fail_sysfs;
3742
3743 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3744 tree_root,
3745 "btrfs-transaction");
3746 if (IS_ERR(fs_info->transaction_kthread))
3747 goto fail_cleaner;
3748
3749 if (!btrfs_test_opt(fs_info, NOSSD) &&
3750 !fs_info->fs_devices->rotating) {
3751 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3752 }
3753
3754 /*
3755 * Mount does not set all options immediately, we can do it now and do
3756 * not have to wait for transaction commit
3757 */
3758 btrfs_apply_pending_changes(fs_info);
3759
3760 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3761 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3762 ret = btrfsic_mount(fs_info, fs_devices,
3763 btrfs_test_opt(fs_info,
3764 CHECK_INTEGRITY_DATA) ? 1 : 0,
3765 fs_info->check_integrity_print_mask);
3766 if (ret)
3767 btrfs_warn(fs_info,
3768 "failed to initialize integrity check module: %d",
3769 ret);
3770 }
3771 #endif
3772 ret = btrfs_read_qgroup_config(fs_info);
3773 if (ret)
3774 goto fail_trans_kthread;
3775
3776 if (btrfs_build_ref_tree(fs_info))
3777 btrfs_err(fs_info, "couldn't build ref tree");
3778
3779 /* do not make disk changes in broken FS or nologreplay is given */
3780 if (btrfs_super_log_root(disk_super) != 0 &&
3781 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3782 btrfs_info(fs_info, "start tree-log replay");
3783 ret = btrfs_replay_log(fs_info, fs_devices);
3784 if (ret) {
3785 err = ret;
3786 goto fail_qgroup;
3787 }
3788 }
3789
3790 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3791 if (IS_ERR(fs_info->fs_root)) {
3792 err = PTR_ERR(fs_info->fs_root);
3793 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3794 fs_info->fs_root = NULL;
3795 goto fail_qgroup;
3796 }
3797
3798 if (sb_rdonly(sb))
3799 goto clear_oneshot;
3800
3801 ret = btrfs_start_pre_rw_mount(fs_info);
3802 if (ret) {
3803 close_ctree(fs_info);
3804 return ret;
3805 }
3806 btrfs_discard_resume(fs_info);
3807
3808 if (fs_info->uuid_root &&
3809 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3810 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3811 btrfs_info(fs_info, "checking UUID tree");
3812 ret = btrfs_check_uuid_tree(fs_info);
3813 if (ret) {
3814 btrfs_warn(fs_info,
3815 "failed to check the UUID tree: %d", ret);
3816 close_ctree(fs_info);
3817 return ret;
3818 }
3819 }
3820
3821 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3822
3823 /* Kick the cleaner thread so it'll start deleting snapshots. */
3824 if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3825 wake_up_process(fs_info->cleaner_kthread);
3826
3827 clear_oneshot:
3828 btrfs_clear_oneshot_options(fs_info);
3829 return 0;
3830
3831 fail_qgroup:
3832 btrfs_free_qgroup_config(fs_info);
3833 fail_trans_kthread:
3834 kthread_stop(fs_info->transaction_kthread);
3835 btrfs_cleanup_transaction(fs_info);
3836 btrfs_free_fs_roots(fs_info);
3837 fail_cleaner:
3838 kthread_stop(fs_info->cleaner_kthread);
3839
3840 /*
3841 * make sure we're done with the btree inode before we stop our
3842 * kthreads
3843 */
3844 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3845
3846 fail_sysfs:
3847 btrfs_sysfs_remove_mounted(fs_info);
3848
3849 fail_fsdev_sysfs:
3850 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3851
3852 fail_block_groups:
3853 btrfs_put_block_group_cache(fs_info);
3854
3855 fail_tree_roots:
3856 if (fs_info->data_reloc_root)
3857 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3858 free_root_pointers(fs_info, true);
3859 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3860
3861 fail_sb_buffer:
3862 btrfs_stop_all_workers(fs_info);
3863 btrfs_free_block_groups(fs_info);
3864 fail_alloc:
3865 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3866
3867 iput(fs_info->btree_inode);
3868 fail:
3869 btrfs_close_devices(fs_info->fs_devices);
3870 return err;
3871 }
3872 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3873
btrfs_end_super_write(struct bio * bio)3874 static void btrfs_end_super_write(struct bio *bio)
3875 {
3876 struct btrfs_device *device = bio->bi_private;
3877 struct bio_vec *bvec;
3878 struct bvec_iter_all iter_all;
3879 struct page *page;
3880
3881 bio_for_each_segment_all(bvec, bio, iter_all) {
3882 page = bvec->bv_page;
3883
3884 if (bio->bi_status) {
3885 btrfs_warn_rl_in_rcu(device->fs_info,
3886 "lost page write due to IO error on %s (%d)",
3887 rcu_str_deref(device->name),
3888 blk_status_to_errno(bio->bi_status));
3889 ClearPageUptodate(page);
3890 SetPageError(page);
3891 btrfs_dev_stat_inc_and_print(device,
3892 BTRFS_DEV_STAT_WRITE_ERRS);
3893 } else {
3894 SetPageUptodate(page);
3895 }
3896
3897 put_page(page);
3898 unlock_page(page);
3899 }
3900
3901 bio_put(bio);
3902 }
3903
btrfs_read_dev_one_super(struct block_device * bdev,int copy_num,bool drop_cache)3904 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3905 int copy_num, bool drop_cache)
3906 {
3907 struct btrfs_super_block *super;
3908 struct page *page;
3909 u64 bytenr, bytenr_orig;
3910 struct address_space *mapping = bdev->bd_inode->i_mapping;
3911 int ret;
3912
3913 bytenr_orig = btrfs_sb_offset(copy_num);
3914 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3915 if (ret == -ENOENT)
3916 return ERR_PTR(-EINVAL);
3917 else if (ret)
3918 return ERR_PTR(ret);
3919
3920 if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3921 return ERR_PTR(-EINVAL);
3922
3923 if (drop_cache) {
3924 /* This should only be called with the primary sb. */
3925 ASSERT(copy_num == 0);
3926
3927 /*
3928 * Drop the page of the primary superblock, so later read will
3929 * always read from the device.
3930 */
3931 invalidate_inode_pages2_range(mapping,
3932 bytenr >> PAGE_SHIFT,
3933 (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3934 }
3935
3936 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3937 if (IS_ERR(page))
3938 return ERR_CAST(page);
3939
3940 super = page_address(page);
3941 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3942 btrfs_release_disk_super(super);
3943 return ERR_PTR(-ENODATA);
3944 }
3945
3946 if (btrfs_super_bytenr(super) != bytenr_orig) {
3947 btrfs_release_disk_super(super);
3948 return ERR_PTR(-EINVAL);
3949 }
3950
3951 return super;
3952 }
3953
3954
btrfs_read_dev_super(struct block_device * bdev)3955 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3956 {
3957 struct btrfs_super_block *super, *latest = NULL;
3958 int i;
3959 u64 transid = 0;
3960
3961 /* we would like to check all the supers, but that would make
3962 * a btrfs mount succeed after a mkfs from a different FS.
3963 * So, we need to add a special mount option to scan for
3964 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3965 */
3966 for (i = 0; i < 1; i++) {
3967 super = btrfs_read_dev_one_super(bdev, i, false);
3968 if (IS_ERR(super))
3969 continue;
3970
3971 if (!latest || btrfs_super_generation(super) > transid) {
3972 if (latest)
3973 btrfs_release_disk_super(super);
3974
3975 latest = super;
3976 transid = btrfs_super_generation(super);
3977 }
3978 }
3979
3980 return super;
3981 }
3982
3983 /*
3984 * Write superblock @sb to the @device. Do not wait for completion, all the
3985 * pages we use for writing are locked.
3986 *
3987 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3988 * the expected device size at commit time. Note that max_mirrors must be
3989 * same for write and wait phases.
3990 *
3991 * Return number of errors when page is not found or submission fails.
3992 */
write_dev_supers(struct btrfs_device * device,struct btrfs_super_block * sb,int max_mirrors)3993 static int write_dev_supers(struct btrfs_device *device,
3994 struct btrfs_super_block *sb, int max_mirrors)
3995 {
3996 struct btrfs_fs_info *fs_info = device->fs_info;
3997 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3998 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3999 int i;
4000 int errors = 0;
4001 int ret;
4002 u64 bytenr, bytenr_orig;
4003
4004 if (max_mirrors == 0)
4005 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4006
4007 shash->tfm = fs_info->csum_shash;
4008
4009 for (i = 0; i < max_mirrors; i++) {
4010 struct page *page;
4011 struct bio *bio;
4012 struct btrfs_super_block *disk_super;
4013
4014 bytenr_orig = btrfs_sb_offset(i);
4015 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
4016 if (ret == -ENOENT) {
4017 continue;
4018 } else if (ret < 0) {
4019 btrfs_err(device->fs_info,
4020 "couldn't get super block location for mirror %d",
4021 i);
4022 errors++;
4023 continue;
4024 }
4025 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4026 device->commit_total_bytes)
4027 break;
4028
4029 btrfs_set_super_bytenr(sb, bytenr_orig);
4030
4031 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
4032 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
4033 sb->csum);
4034
4035 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
4036 GFP_NOFS);
4037 if (!page) {
4038 btrfs_err(device->fs_info,
4039 "couldn't get super block page for bytenr %llu",
4040 bytenr);
4041 errors++;
4042 continue;
4043 }
4044
4045 /* Bump the refcount for wait_dev_supers() */
4046 get_page(page);
4047
4048 disk_super = page_address(page);
4049 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
4050
4051 /*
4052 * Directly use bios here instead of relying on the page cache
4053 * to do I/O, so we don't lose the ability to do integrity
4054 * checking.
4055 */
4056 bio = bio_alloc(device->bdev, 1,
4057 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
4058 GFP_NOFS);
4059 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
4060 bio->bi_private = device;
4061 bio->bi_end_io = btrfs_end_super_write;
4062 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
4063 offset_in_page(bytenr));
4064
4065 /*
4066 * We FUA only the first super block. The others we allow to
4067 * go down lazy and there's a short window where the on-disk
4068 * copies might still contain the older version.
4069 */
4070 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
4071 bio->bi_opf |= REQ_FUA;
4072
4073 btrfsic_check_bio(bio);
4074 submit_bio(bio);
4075
4076 if (btrfs_advance_sb_log(device, i))
4077 errors++;
4078 }
4079 return errors < i ? 0 : -1;
4080 }
4081
4082 /*
4083 * Wait for write completion of superblocks done by write_dev_supers,
4084 * @max_mirrors same for write and wait phases.
4085 *
4086 * Return number of errors when page is not found or not marked up to
4087 * date.
4088 */
wait_dev_supers(struct btrfs_device * device,int max_mirrors)4089 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
4090 {
4091 int i;
4092 int errors = 0;
4093 bool primary_failed = false;
4094 int ret;
4095 u64 bytenr;
4096
4097 if (max_mirrors == 0)
4098 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4099
4100 for (i = 0; i < max_mirrors; i++) {
4101 struct page *page;
4102
4103 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
4104 if (ret == -ENOENT) {
4105 break;
4106 } else if (ret < 0) {
4107 errors++;
4108 if (i == 0)
4109 primary_failed = true;
4110 continue;
4111 }
4112 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4113 device->commit_total_bytes)
4114 break;
4115
4116 page = find_get_page(device->bdev->bd_inode->i_mapping,
4117 bytenr >> PAGE_SHIFT);
4118 if (!page) {
4119 errors++;
4120 if (i == 0)
4121 primary_failed = true;
4122 continue;
4123 }
4124 /* Page is submitted locked and unlocked once the IO completes */
4125 wait_on_page_locked(page);
4126 if (PageError(page)) {
4127 errors++;
4128 if (i == 0)
4129 primary_failed = true;
4130 }
4131
4132 /* Drop our reference */
4133 put_page(page);
4134
4135 /* Drop the reference from the writing run */
4136 put_page(page);
4137 }
4138
4139 /* log error, force error return */
4140 if (primary_failed) {
4141 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
4142 device->devid);
4143 return -1;
4144 }
4145
4146 return errors < i ? 0 : -1;
4147 }
4148
4149 /*
4150 * endio for the write_dev_flush, this will wake anyone waiting
4151 * for the barrier when it is done
4152 */
btrfs_end_empty_barrier(struct bio * bio)4153 static void btrfs_end_empty_barrier(struct bio *bio)
4154 {
4155 bio_uninit(bio);
4156 complete(bio->bi_private);
4157 }
4158
4159 /*
4160 * Submit a flush request to the device if it supports it. Error handling is
4161 * done in the waiting counterpart.
4162 */
write_dev_flush(struct btrfs_device * device)4163 static void write_dev_flush(struct btrfs_device *device)
4164 {
4165 struct bio *bio = &device->flush_bio;
4166
4167 #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4168 /*
4169 * When a disk has write caching disabled, we skip submission of a bio
4170 * with flush and sync requests before writing the superblock, since
4171 * it's not needed. However when the integrity checker is enabled, this
4172 * results in reports that there are metadata blocks referred by a
4173 * superblock that were not properly flushed. So don't skip the bio
4174 * submission only when the integrity checker is enabled for the sake
4175 * of simplicity, since this is a debug tool and not meant for use in
4176 * non-debug builds.
4177 */
4178 if (!bdev_write_cache(device->bdev))
4179 return;
4180 #endif
4181
4182 bio_init(bio, device->bdev, NULL, 0,
4183 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
4184 bio->bi_end_io = btrfs_end_empty_barrier;
4185 init_completion(&device->flush_wait);
4186 bio->bi_private = &device->flush_wait;
4187
4188 btrfsic_check_bio(bio);
4189 submit_bio(bio);
4190 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4191 }
4192
4193 /*
4194 * If the flush bio has been submitted by write_dev_flush, wait for it.
4195 */
wait_dev_flush(struct btrfs_device * device)4196 static blk_status_t wait_dev_flush(struct btrfs_device *device)
4197 {
4198 struct bio *bio = &device->flush_bio;
4199
4200 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
4201 return BLK_STS_OK;
4202
4203 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4204 wait_for_completion_io(&device->flush_wait);
4205
4206 return bio->bi_status;
4207 }
4208
check_barrier_error(struct btrfs_fs_info * fs_info)4209 static int check_barrier_error(struct btrfs_fs_info *fs_info)
4210 {
4211 if (!btrfs_check_rw_degradable(fs_info, NULL))
4212 return -EIO;
4213 return 0;
4214 }
4215
4216 /*
4217 * send an empty flush down to each device in parallel,
4218 * then wait for them
4219 */
barrier_all_devices(struct btrfs_fs_info * info)4220 static int barrier_all_devices(struct btrfs_fs_info *info)
4221 {
4222 struct list_head *head;
4223 struct btrfs_device *dev;
4224 int errors_wait = 0;
4225 blk_status_t ret;
4226
4227 lockdep_assert_held(&info->fs_devices->device_list_mutex);
4228 /* send down all the barriers */
4229 head = &info->fs_devices->devices;
4230 list_for_each_entry(dev, head, dev_list) {
4231 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4232 continue;
4233 if (!dev->bdev)
4234 continue;
4235 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4236 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4237 continue;
4238
4239 write_dev_flush(dev);
4240 dev->last_flush_error = BLK_STS_OK;
4241 }
4242
4243 /* wait for all the barriers */
4244 list_for_each_entry(dev, head, dev_list) {
4245 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4246 continue;
4247 if (!dev->bdev) {
4248 errors_wait++;
4249 continue;
4250 }
4251 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4252 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4253 continue;
4254
4255 ret = wait_dev_flush(dev);
4256 if (ret) {
4257 dev->last_flush_error = ret;
4258 btrfs_dev_stat_inc_and_print(dev,
4259 BTRFS_DEV_STAT_FLUSH_ERRS);
4260 errors_wait++;
4261 }
4262 }
4263
4264 if (errors_wait) {
4265 /*
4266 * At some point we need the status of all disks
4267 * to arrive at the volume status. So error checking
4268 * is being pushed to a separate loop.
4269 */
4270 return check_barrier_error(info);
4271 }
4272 return 0;
4273 }
4274
btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)4275 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4276 {
4277 int raid_type;
4278 int min_tolerated = INT_MAX;
4279
4280 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4281 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4282 min_tolerated = min_t(int, min_tolerated,
4283 btrfs_raid_array[BTRFS_RAID_SINGLE].
4284 tolerated_failures);
4285
4286 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4287 if (raid_type == BTRFS_RAID_SINGLE)
4288 continue;
4289 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4290 continue;
4291 min_tolerated = min_t(int, min_tolerated,
4292 btrfs_raid_array[raid_type].
4293 tolerated_failures);
4294 }
4295
4296 if (min_tolerated == INT_MAX) {
4297 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4298 min_tolerated = 0;
4299 }
4300
4301 return min_tolerated;
4302 }
4303
write_all_supers(struct btrfs_fs_info * fs_info,int max_mirrors)4304 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4305 {
4306 struct list_head *head;
4307 struct btrfs_device *dev;
4308 struct btrfs_super_block *sb;
4309 struct btrfs_dev_item *dev_item;
4310 int ret;
4311 int do_barriers;
4312 int max_errors;
4313 int total_errors = 0;
4314 u64 flags;
4315
4316 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4317
4318 /*
4319 * max_mirrors == 0 indicates we're from commit_transaction,
4320 * not from fsync where the tree roots in fs_info have not
4321 * been consistent on disk.
4322 */
4323 if (max_mirrors == 0)
4324 backup_super_roots(fs_info);
4325
4326 sb = fs_info->super_for_commit;
4327 dev_item = &sb->dev_item;
4328
4329 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4330 head = &fs_info->fs_devices->devices;
4331 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4332
4333 if (do_barriers) {
4334 ret = barrier_all_devices(fs_info);
4335 if (ret) {
4336 mutex_unlock(
4337 &fs_info->fs_devices->device_list_mutex);
4338 btrfs_handle_fs_error(fs_info, ret,
4339 "errors while submitting device barriers.");
4340 return ret;
4341 }
4342 }
4343
4344 list_for_each_entry(dev, head, dev_list) {
4345 if (!dev->bdev) {
4346 total_errors++;
4347 continue;
4348 }
4349 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4350 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4351 continue;
4352
4353 btrfs_set_stack_device_generation(dev_item, 0);
4354 btrfs_set_stack_device_type(dev_item, dev->type);
4355 btrfs_set_stack_device_id(dev_item, dev->devid);
4356 btrfs_set_stack_device_total_bytes(dev_item,
4357 dev->commit_total_bytes);
4358 btrfs_set_stack_device_bytes_used(dev_item,
4359 dev->commit_bytes_used);
4360 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4361 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4362 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4363 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4364 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4365 BTRFS_FSID_SIZE);
4366
4367 flags = btrfs_super_flags(sb);
4368 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4369
4370 ret = btrfs_validate_write_super(fs_info, sb);
4371 if (ret < 0) {
4372 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4373 btrfs_handle_fs_error(fs_info, -EUCLEAN,
4374 "unexpected superblock corruption detected");
4375 return -EUCLEAN;
4376 }
4377
4378 ret = write_dev_supers(dev, sb, max_mirrors);
4379 if (ret)
4380 total_errors++;
4381 }
4382 if (total_errors > max_errors) {
4383 btrfs_err(fs_info, "%d errors while writing supers",
4384 total_errors);
4385 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4386
4387 /* FUA is masked off if unsupported and can't be the reason */
4388 btrfs_handle_fs_error(fs_info, -EIO,
4389 "%d errors while writing supers",
4390 total_errors);
4391 return -EIO;
4392 }
4393
4394 total_errors = 0;
4395 list_for_each_entry(dev, head, dev_list) {
4396 if (!dev->bdev)
4397 continue;
4398 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4399 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4400 continue;
4401
4402 ret = wait_dev_supers(dev, max_mirrors);
4403 if (ret)
4404 total_errors++;
4405 }
4406 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4407 if (total_errors > max_errors) {
4408 btrfs_handle_fs_error(fs_info, -EIO,
4409 "%d errors while writing supers",
4410 total_errors);
4411 return -EIO;
4412 }
4413 return 0;
4414 }
4415
4416 /* Drop a fs root from the radix tree and free it. */
btrfs_drop_and_free_fs_root(struct btrfs_fs_info * fs_info,struct btrfs_root * root)4417 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4418 struct btrfs_root *root)
4419 {
4420 bool drop_ref = false;
4421
4422 spin_lock(&fs_info->fs_roots_radix_lock);
4423 radix_tree_delete(&fs_info->fs_roots_radix,
4424 (unsigned long)root->root_key.objectid);
4425 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4426 drop_ref = true;
4427 spin_unlock(&fs_info->fs_roots_radix_lock);
4428
4429 if (BTRFS_FS_ERROR(fs_info)) {
4430 ASSERT(root->log_root == NULL);
4431 if (root->reloc_root) {
4432 btrfs_put_root(root->reloc_root);
4433 root->reloc_root = NULL;
4434 }
4435 }
4436
4437 if (drop_ref)
4438 btrfs_put_root(root);
4439 }
4440
btrfs_cleanup_fs_roots(struct btrfs_fs_info * fs_info)4441 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4442 {
4443 u64 root_objectid = 0;
4444 struct btrfs_root *gang[8];
4445 int i = 0;
4446 int err = 0;
4447 unsigned int ret = 0;
4448
4449 while (1) {
4450 spin_lock(&fs_info->fs_roots_radix_lock);
4451 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4452 (void **)gang, root_objectid,
4453 ARRAY_SIZE(gang));
4454 if (!ret) {
4455 spin_unlock(&fs_info->fs_roots_radix_lock);
4456 break;
4457 }
4458 root_objectid = gang[ret - 1]->root_key.objectid + 1;
4459
4460 for (i = 0; i < ret; i++) {
4461 /* Avoid to grab roots in dead_roots */
4462 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4463 gang[i] = NULL;
4464 continue;
4465 }
4466 /* grab all the search result for later use */
4467 gang[i] = btrfs_grab_root(gang[i]);
4468 }
4469 spin_unlock(&fs_info->fs_roots_radix_lock);
4470
4471 for (i = 0; i < ret; i++) {
4472 if (!gang[i])
4473 continue;
4474 root_objectid = gang[i]->root_key.objectid;
4475 err = btrfs_orphan_cleanup(gang[i]);
4476 if (err)
4477 break;
4478 btrfs_put_root(gang[i]);
4479 }
4480 root_objectid++;
4481 }
4482
4483 /* release the uncleaned roots due to error */
4484 for (; i < ret; i++) {
4485 if (gang[i])
4486 btrfs_put_root(gang[i]);
4487 }
4488 return err;
4489 }
4490
btrfs_commit_super(struct btrfs_fs_info * fs_info)4491 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4492 {
4493 struct btrfs_root *root = fs_info->tree_root;
4494 struct btrfs_trans_handle *trans;
4495
4496 mutex_lock(&fs_info->cleaner_mutex);
4497 btrfs_run_delayed_iputs(fs_info);
4498 mutex_unlock(&fs_info->cleaner_mutex);
4499 wake_up_process(fs_info->cleaner_kthread);
4500
4501 /* wait until ongoing cleanup work done */
4502 down_write(&fs_info->cleanup_work_sem);
4503 up_write(&fs_info->cleanup_work_sem);
4504
4505 trans = btrfs_join_transaction(root);
4506 if (IS_ERR(trans))
4507 return PTR_ERR(trans);
4508 return btrfs_commit_transaction(trans);
4509 }
4510
warn_about_uncommitted_trans(struct btrfs_fs_info * fs_info)4511 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4512 {
4513 struct btrfs_transaction *trans;
4514 struct btrfs_transaction *tmp;
4515 bool found = false;
4516
4517 if (list_empty(&fs_info->trans_list))
4518 return;
4519
4520 /*
4521 * This function is only called at the very end of close_ctree(),
4522 * thus no other running transaction, no need to take trans_lock.
4523 */
4524 ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4525 list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4526 struct extent_state *cached = NULL;
4527 u64 dirty_bytes = 0;
4528 u64 cur = 0;
4529 u64 found_start;
4530 u64 found_end;
4531
4532 found = true;
4533 while (!find_first_extent_bit(&trans->dirty_pages, cur,
4534 &found_start, &found_end, EXTENT_DIRTY, &cached)) {
4535 dirty_bytes += found_end + 1 - found_start;
4536 cur = found_end + 1;
4537 }
4538 btrfs_warn(fs_info,
4539 "transaction %llu (with %llu dirty metadata bytes) is not committed",
4540 trans->transid, dirty_bytes);
4541 btrfs_cleanup_one_transaction(trans, fs_info);
4542
4543 if (trans == fs_info->running_transaction)
4544 fs_info->running_transaction = NULL;
4545 list_del_init(&trans->list);
4546
4547 btrfs_put_transaction(trans);
4548 trace_btrfs_transaction_commit(fs_info);
4549 }
4550 ASSERT(!found);
4551 }
4552
close_ctree(struct btrfs_fs_info * fs_info)4553 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4554 {
4555 int ret;
4556
4557 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4558
4559 /*
4560 * If we had UNFINISHED_DROPS we could still be processing them, so
4561 * clear that bit and wake up relocation so it can stop.
4562 * We must do this before stopping the block group reclaim task, because
4563 * at btrfs_relocate_block_group() we wait for this bit, and after the
4564 * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4565 * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4566 * return 1.
4567 */
4568 btrfs_wake_unfinished_drop(fs_info);
4569
4570 /*
4571 * We may have the reclaim task running and relocating a data block group,
4572 * in which case it may create delayed iputs. So stop it before we park
4573 * the cleaner kthread otherwise we can get new delayed iputs after
4574 * parking the cleaner, and that can make the async reclaim task to hang
4575 * if it's waiting for delayed iputs to complete, since the cleaner is
4576 * parked and can not run delayed iputs - this will make us hang when
4577 * trying to stop the async reclaim task.
4578 */
4579 cancel_work_sync(&fs_info->reclaim_bgs_work);
4580 /*
4581 * We don't want the cleaner to start new transactions, add more delayed
4582 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4583 * because that frees the task_struct, and the transaction kthread might
4584 * still try to wake up the cleaner.
4585 */
4586 kthread_park(fs_info->cleaner_kthread);
4587
4588 /* wait for the qgroup rescan worker to stop */
4589 btrfs_qgroup_wait_for_completion(fs_info, false);
4590
4591 /* wait for the uuid_scan task to finish */
4592 down(&fs_info->uuid_tree_rescan_sem);
4593 /* avoid complains from lockdep et al., set sem back to initial state */
4594 up(&fs_info->uuid_tree_rescan_sem);
4595
4596 /* pause restriper - we want to resume on mount */
4597 btrfs_pause_balance(fs_info);
4598
4599 btrfs_dev_replace_suspend_for_unmount(fs_info);
4600
4601 btrfs_scrub_cancel(fs_info);
4602
4603 /* wait for any defraggers to finish */
4604 wait_event(fs_info->transaction_wait,
4605 (atomic_read(&fs_info->defrag_running) == 0));
4606
4607 /* clear out the rbtree of defraggable inodes */
4608 btrfs_cleanup_defrag_inodes(fs_info);
4609
4610 /*
4611 * After we parked the cleaner kthread, ordered extents may have
4612 * completed and created new delayed iputs. If one of the async reclaim
4613 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4614 * can hang forever trying to stop it, because if a delayed iput is
4615 * added after it ran btrfs_run_delayed_iputs() and before it called
4616 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4617 * no one else to run iputs.
4618 *
4619 * So wait for all ongoing ordered extents to complete and then run
4620 * delayed iputs. This works because once we reach this point no one
4621 * can either create new ordered extents nor create delayed iputs
4622 * through some other means.
4623 *
4624 * Also note that btrfs_wait_ordered_roots() is not safe here, because
4625 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4626 * but the delayed iput for the respective inode is made only when doing
4627 * the final btrfs_put_ordered_extent() (which must happen at
4628 * btrfs_finish_ordered_io() when we are unmounting).
4629 */
4630 btrfs_flush_workqueue(fs_info->endio_write_workers);
4631 /* Ordered extents for free space inodes. */
4632 btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4633 btrfs_run_delayed_iputs(fs_info);
4634
4635 cancel_work_sync(&fs_info->async_reclaim_work);
4636 cancel_work_sync(&fs_info->async_data_reclaim_work);
4637 cancel_work_sync(&fs_info->preempt_reclaim_work);
4638
4639 /* Cancel or finish ongoing discard work */
4640 btrfs_discard_cleanup(fs_info);
4641
4642 if (!sb_rdonly(fs_info->sb)) {
4643 /*
4644 * The cleaner kthread is stopped, so do one final pass over
4645 * unused block groups.
4646 */
4647 btrfs_delete_unused_bgs(fs_info);
4648
4649 /*
4650 * There might be existing delayed inode workers still running
4651 * and holding an empty delayed inode item. We must wait for
4652 * them to complete first because they can create a transaction.
4653 * This happens when someone calls btrfs_balance_delayed_items()
4654 * and then a transaction commit runs the same delayed nodes
4655 * before any delayed worker has done something with the nodes.
4656 * We must wait for any worker here and not at transaction
4657 * commit time since that could cause a deadlock.
4658 * This is a very rare case.
4659 */
4660 btrfs_flush_workqueue(fs_info->delayed_workers);
4661
4662 ret = btrfs_commit_super(fs_info);
4663 if (ret)
4664 btrfs_err(fs_info, "commit super ret %d", ret);
4665 }
4666
4667 if (BTRFS_FS_ERROR(fs_info))
4668 btrfs_error_commit_super(fs_info);
4669
4670 kthread_stop(fs_info->transaction_kthread);
4671 kthread_stop(fs_info->cleaner_kthread);
4672
4673 ASSERT(list_empty(&fs_info->delayed_iputs));
4674 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4675
4676 if (btrfs_check_quota_leak(fs_info)) {
4677 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4678 btrfs_err(fs_info, "qgroup reserved space leaked");
4679 }
4680
4681 btrfs_free_qgroup_config(fs_info);
4682 ASSERT(list_empty(&fs_info->delalloc_roots));
4683
4684 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4685 btrfs_info(fs_info, "at unmount delalloc count %lld",
4686 percpu_counter_sum(&fs_info->delalloc_bytes));
4687 }
4688
4689 if (percpu_counter_sum(&fs_info->ordered_bytes))
4690 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4691 percpu_counter_sum(&fs_info->ordered_bytes));
4692
4693 btrfs_sysfs_remove_mounted(fs_info);
4694 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4695
4696 btrfs_put_block_group_cache(fs_info);
4697
4698 /*
4699 * we must make sure there is not any read request to
4700 * submit after we stopping all workers.
4701 */
4702 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4703 btrfs_stop_all_workers(fs_info);
4704
4705 /* We shouldn't have any transaction open at this point */
4706 warn_about_uncommitted_trans(fs_info);
4707
4708 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4709 free_root_pointers(fs_info, true);
4710 btrfs_free_fs_roots(fs_info);
4711
4712 /*
4713 * We must free the block groups after dropping the fs_roots as we could
4714 * have had an IO error and have left over tree log blocks that aren't
4715 * cleaned up until the fs roots are freed. This makes the block group
4716 * accounting appear to be wrong because there's pending reserved bytes,
4717 * so make sure we do the block group cleanup afterwards.
4718 */
4719 btrfs_free_block_groups(fs_info);
4720
4721 iput(fs_info->btree_inode);
4722
4723 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4724 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4725 btrfsic_unmount(fs_info->fs_devices);
4726 #endif
4727
4728 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4729 btrfs_close_devices(fs_info->fs_devices);
4730 }
4731
btrfs_buffer_uptodate(struct extent_buffer * buf,u64 parent_transid,int atomic)4732 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4733 int atomic)
4734 {
4735 int ret;
4736 struct inode *btree_inode = buf->pages[0]->mapping->host;
4737
4738 ret = extent_buffer_uptodate(buf);
4739 if (!ret)
4740 return ret;
4741
4742 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4743 parent_transid, atomic);
4744 if (ret == -EAGAIN)
4745 return ret;
4746 return !ret;
4747 }
4748
btrfs_mark_buffer_dirty(struct extent_buffer * buf)4749 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4750 {
4751 struct btrfs_fs_info *fs_info = buf->fs_info;
4752 u64 transid = btrfs_header_generation(buf);
4753 int was_dirty;
4754
4755 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4756 /*
4757 * This is a fast path so only do this check if we have sanity tests
4758 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4759 * outside of the sanity tests.
4760 */
4761 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4762 return;
4763 #endif
4764 btrfs_assert_tree_write_locked(buf);
4765 if (transid != fs_info->generation)
4766 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4767 buf->start, transid, fs_info->generation);
4768 was_dirty = set_extent_buffer_dirty(buf);
4769 if (!was_dirty)
4770 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4771 buf->len,
4772 fs_info->dirty_metadata_batch);
4773 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4774 /*
4775 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4776 * but item data not updated.
4777 * So here we should only check item pointers, not item data.
4778 */
4779 if (btrfs_header_level(buf) == 0 &&
4780 btrfs_check_leaf_relaxed(buf)) {
4781 btrfs_print_leaf(buf);
4782 ASSERT(0);
4783 }
4784 #endif
4785 }
4786
__btrfs_btree_balance_dirty(struct btrfs_fs_info * fs_info,int flush_delayed)4787 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4788 int flush_delayed)
4789 {
4790 /*
4791 * looks as though older kernels can get into trouble with
4792 * this code, they end up stuck in balance_dirty_pages forever
4793 */
4794 int ret;
4795
4796 if (current->flags & PF_MEMALLOC)
4797 return;
4798
4799 if (flush_delayed)
4800 btrfs_balance_delayed_items(fs_info);
4801
4802 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4803 BTRFS_DIRTY_METADATA_THRESH,
4804 fs_info->dirty_metadata_batch);
4805 if (ret > 0) {
4806 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4807 }
4808 }
4809
btrfs_btree_balance_dirty(struct btrfs_fs_info * fs_info)4810 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4811 {
4812 __btrfs_btree_balance_dirty(fs_info, 1);
4813 }
4814
btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info * fs_info)4815 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4816 {
4817 __btrfs_btree_balance_dirty(fs_info, 0);
4818 }
4819
btrfs_error_commit_super(struct btrfs_fs_info * fs_info)4820 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4821 {
4822 /* cleanup FS via transaction */
4823 btrfs_cleanup_transaction(fs_info);
4824
4825 mutex_lock(&fs_info->cleaner_mutex);
4826 btrfs_run_delayed_iputs(fs_info);
4827 mutex_unlock(&fs_info->cleaner_mutex);
4828
4829 down_write(&fs_info->cleanup_work_sem);
4830 up_write(&fs_info->cleanup_work_sem);
4831 }
4832
btrfs_drop_all_logs(struct btrfs_fs_info * fs_info)4833 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4834 {
4835 struct btrfs_root *gang[8];
4836 u64 root_objectid = 0;
4837 int ret;
4838
4839 spin_lock(&fs_info->fs_roots_radix_lock);
4840 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4841 (void **)gang, root_objectid,
4842 ARRAY_SIZE(gang))) != 0) {
4843 int i;
4844
4845 for (i = 0; i < ret; i++)
4846 gang[i] = btrfs_grab_root(gang[i]);
4847 spin_unlock(&fs_info->fs_roots_radix_lock);
4848
4849 for (i = 0; i < ret; i++) {
4850 if (!gang[i])
4851 continue;
4852 root_objectid = gang[i]->root_key.objectid;
4853 btrfs_free_log(NULL, gang[i]);
4854 btrfs_put_root(gang[i]);
4855 }
4856 root_objectid++;
4857 spin_lock(&fs_info->fs_roots_radix_lock);
4858 }
4859 spin_unlock(&fs_info->fs_roots_radix_lock);
4860 btrfs_free_log_root_tree(NULL, fs_info);
4861 }
4862
btrfs_destroy_ordered_extents(struct btrfs_root * root)4863 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4864 {
4865 struct btrfs_ordered_extent *ordered;
4866
4867 spin_lock(&root->ordered_extent_lock);
4868 /*
4869 * This will just short circuit the ordered completion stuff which will
4870 * make sure the ordered extent gets properly cleaned up.
4871 */
4872 list_for_each_entry(ordered, &root->ordered_extents,
4873 root_extent_list)
4874 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4875 spin_unlock(&root->ordered_extent_lock);
4876 }
4877
btrfs_destroy_all_ordered_extents(struct btrfs_fs_info * fs_info)4878 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4879 {
4880 struct btrfs_root *root;
4881 struct list_head splice;
4882
4883 INIT_LIST_HEAD(&splice);
4884
4885 spin_lock(&fs_info->ordered_root_lock);
4886 list_splice_init(&fs_info->ordered_roots, &splice);
4887 while (!list_empty(&splice)) {
4888 root = list_first_entry(&splice, struct btrfs_root,
4889 ordered_root);
4890 list_move_tail(&root->ordered_root,
4891 &fs_info->ordered_roots);
4892
4893 spin_unlock(&fs_info->ordered_root_lock);
4894 btrfs_destroy_ordered_extents(root);
4895
4896 cond_resched();
4897 spin_lock(&fs_info->ordered_root_lock);
4898 }
4899 spin_unlock(&fs_info->ordered_root_lock);
4900
4901 /*
4902 * We need this here because if we've been flipped read-only we won't
4903 * get sync() from the umount, so we need to make sure any ordered
4904 * extents that haven't had their dirty pages IO start writeout yet
4905 * actually get run and error out properly.
4906 */
4907 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4908 }
4909
btrfs_destroy_delayed_refs(struct btrfs_transaction * trans,struct btrfs_fs_info * fs_info)4910 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4911 struct btrfs_fs_info *fs_info)
4912 {
4913 struct rb_node *node;
4914 struct btrfs_delayed_ref_root *delayed_refs;
4915 struct btrfs_delayed_ref_node *ref;
4916 int ret = 0;
4917
4918 delayed_refs = &trans->delayed_refs;
4919
4920 spin_lock(&delayed_refs->lock);
4921 if (atomic_read(&delayed_refs->num_entries) == 0) {
4922 spin_unlock(&delayed_refs->lock);
4923 btrfs_debug(fs_info, "delayed_refs has NO entry");
4924 return ret;
4925 }
4926
4927 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4928 struct btrfs_delayed_ref_head *head;
4929 struct rb_node *n;
4930 bool pin_bytes = false;
4931
4932 head = rb_entry(node, struct btrfs_delayed_ref_head,
4933 href_node);
4934 if (btrfs_delayed_ref_lock(delayed_refs, head))
4935 continue;
4936
4937 spin_lock(&head->lock);
4938 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4939 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4940 ref_node);
4941 ref->in_tree = 0;
4942 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4943 RB_CLEAR_NODE(&ref->ref_node);
4944 if (!list_empty(&ref->add_list))
4945 list_del(&ref->add_list);
4946 atomic_dec(&delayed_refs->num_entries);
4947 btrfs_put_delayed_ref(ref);
4948 }
4949 if (head->must_insert_reserved)
4950 pin_bytes = true;
4951 btrfs_free_delayed_extent_op(head->extent_op);
4952 btrfs_delete_ref_head(delayed_refs, head);
4953 spin_unlock(&head->lock);
4954 spin_unlock(&delayed_refs->lock);
4955 mutex_unlock(&head->mutex);
4956
4957 if (pin_bytes) {
4958 struct btrfs_block_group *cache;
4959
4960 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4961 BUG_ON(!cache);
4962
4963 spin_lock(&cache->space_info->lock);
4964 spin_lock(&cache->lock);
4965 cache->pinned += head->num_bytes;
4966 btrfs_space_info_update_bytes_pinned(fs_info,
4967 cache->space_info, head->num_bytes);
4968 cache->reserved -= head->num_bytes;
4969 cache->space_info->bytes_reserved -= head->num_bytes;
4970 spin_unlock(&cache->lock);
4971 spin_unlock(&cache->space_info->lock);
4972
4973 btrfs_put_block_group(cache);
4974
4975 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4976 head->bytenr + head->num_bytes - 1);
4977 }
4978 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4979 btrfs_put_delayed_ref_head(head);
4980 cond_resched();
4981 spin_lock(&delayed_refs->lock);
4982 }
4983 btrfs_qgroup_destroy_extent_records(trans);
4984
4985 spin_unlock(&delayed_refs->lock);
4986
4987 return ret;
4988 }
4989
btrfs_destroy_delalloc_inodes(struct btrfs_root * root)4990 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4991 {
4992 struct btrfs_inode *btrfs_inode;
4993 struct list_head splice;
4994
4995 INIT_LIST_HEAD(&splice);
4996
4997 spin_lock(&root->delalloc_lock);
4998 list_splice_init(&root->delalloc_inodes, &splice);
4999
5000 while (!list_empty(&splice)) {
5001 struct inode *inode = NULL;
5002 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
5003 delalloc_inodes);
5004 __btrfs_del_delalloc_inode(root, btrfs_inode);
5005 spin_unlock(&root->delalloc_lock);
5006
5007 /*
5008 * Make sure we get a live inode and that it'll not disappear
5009 * meanwhile.
5010 */
5011 inode = igrab(&btrfs_inode->vfs_inode);
5012 if (inode) {
5013 invalidate_inode_pages2(inode->i_mapping);
5014 iput(inode);
5015 }
5016 spin_lock(&root->delalloc_lock);
5017 }
5018 spin_unlock(&root->delalloc_lock);
5019 }
5020
btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info * fs_info)5021 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
5022 {
5023 struct btrfs_root *root;
5024 struct list_head splice;
5025
5026 INIT_LIST_HEAD(&splice);
5027
5028 spin_lock(&fs_info->delalloc_root_lock);
5029 list_splice_init(&fs_info->delalloc_roots, &splice);
5030 while (!list_empty(&splice)) {
5031 root = list_first_entry(&splice, struct btrfs_root,
5032 delalloc_root);
5033 root = btrfs_grab_root(root);
5034 BUG_ON(!root);
5035 spin_unlock(&fs_info->delalloc_root_lock);
5036
5037 btrfs_destroy_delalloc_inodes(root);
5038 btrfs_put_root(root);
5039
5040 spin_lock(&fs_info->delalloc_root_lock);
5041 }
5042 spin_unlock(&fs_info->delalloc_root_lock);
5043 }
5044
btrfs_destroy_marked_extents(struct btrfs_fs_info * fs_info,struct extent_io_tree * dirty_pages,int mark)5045 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
5046 struct extent_io_tree *dirty_pages,
5047 int mark)
5048 {
5049 int ret;
5050 struct extent_buffer *eb;
5051 u64 start = 0;
5052 u64 end;
5053
5054 while (1) {
5055 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
5056 mark, NULL);
5057 if (ret)
5058 break;
5059
5060 clear_extent_bits(dirty_pages, start, end, mark);
5061 while (start <= end) {
5062 eb = find_extent_buffer(fs_info, start);
5063 start += fs_info->nodesize;
5064 if (!eb)
5065 continue;
5066 wait_on_extent_buffer_writeback(eb);
5067
5068 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
5069 &eb->bflags))
5070 clear_extent_buffer_dirty(eb);
5071 free_extent_buffer_stale(eb);
5072 }
5073 }
5074
5075 return ret;
5076 }
5077
btrfs_destroy_pinned_extent(struct btrfs_fs_info * fs_info,struct extent_io_tree * unpin)5078 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
5079 struct extent_io_tree *unpin)
5080 {
5081 u64 start;
5082 u64 end;
5083 int ret;
5084
5085 while (1) {
5086 struct extent_state *cached_state = NULL;
5087
5088 /*
5089 * The btrfs_finish_extent_commit() may get the same range as
5090 * ours between find_first_extent_bit and clear_extent_dirty.
5091 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
5092 * the same extent range.
5093 */
5094 mutex_lock(&fs_info->unused_bg_unpin_mutex);
5095 ret = find_first_extent_bit(unpin, 0, &start, &end,
5096 EXTENT_DIRTY, &cached_state);
5097 if (ret) {
5098 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5099 break;
5100 }
5101
5102 clear_extent_dirty(unpin, start, end, &cached_state);
5103 free_extent_state(cached_state);
5104 btrfs_error_unpin_extent_range(fs_info, start, end);
5105 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5106 cond_resched();
5107 }
5108
5109 return 0;
5110 }
5111
btrfs_cleanup_bg_io(struct btrfs_block_group * cache)5112 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
5113 {
5114 struct inode *inode;
5115
5116 inode = cache->io_ctl.inode;
5117 if (inode) {
5118 invalidate_inode_pages2(inode->i_mapping);
5119 BTRFS_I(inode)->generation = 0;
5120 cache->io_ctl.inode = NULL;
5121 iput(inode);
5122 }
5123 ASSERT(cache->io_ctl.pages == NULL);
5124 btrfs_put_block_group(cache);
5125 }
5126
btrfs_cleanup_dirty_bgs(struct btrfs_transaction * cur_trans,struct btrfs_fs_info * fs_info)5127 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
5128 struct btrfs_fs_info *fs_info)
5129 {
5130 struct btrfs_block_group *cache;
5131
5132 spin_lock(&cur_trans->dirty_bgs_lock);
5133 while (!list_empty(&cur_trans->dirty_bgs)) {
5134 cache = list_first_entry(&cur_trans->dirty_bgs,
5135 struct btrfs_block_group,
5136 dirty_list);
5137
5138 if (!list_empty(&cache->io_list)) {
5139 spin_unlock(&cur_trans->dirty_bgs_lock);
5140 list_del_init(&cache->io_list);
5141 btrfs_cleanup_bg_io(cache);
5142 spin_lock(&cur_trans->dirty_bgs_lock);
5143 }
5144
5145 list_del_init(&cache->dirty_list);
5146 spin_lock(&cache->lock);
5147 cache->disk_cache_state = BTRFS_DC_ERROR;
5148 spin_unlock(&cache->lock);
5149
5150 spin_unlock(&cur_trans->dirty_bgs_lock);
5151 btrfs_put_block_group(cache);
5152 btrfs_delayed_refs_rsv_release(fs_info, 1);
5153 spin_lock(&cur_trans->dirty_bgs_lock);
5154 }
5155 spin_unlock(&cur_trans->dirty_bgs_lock);
5156
5157 /*
5158 * Refer to the definition of io_bgs member for details why it's safe
5159 * to use it without any locking
5160 */
5161 while (!list_empty(&cur_trans->io_bgs)) {
5162 cache = list_first_entry(&cur_trans->io_bgs,
5163 struct btrfs_block_group,
5164 io_list);
5165
5166 list_del_init(&cache->io_list);
5167 spin_lock(&cache->lock);
5168 cache->disk_cache_state = BTRFS_DC_ERROR;
5169 spin_unlock(&cache->lock);
5170 btrfs_cleanup_bg_io(cache);
5171 }
5172 }
5173
btrfs_cleanup_one_transaction(struct btrfs_transaction * cur_trans,struct btrfs_fs_info * fs_info)5174 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
5175 struct btrfs_fs_info *fs_info)
5176 {
5177 struct btrfs_device *dev, *tmp;
5178
5179 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
5180 ASSERT(list_empty(&cur_trans->dirty_bgs));
5181 ASSERT(list_empty(&cur_trans->io_bgs));
5182
5183 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
5184 post_commit_list) {
5185 list_del_init(&dev->post_commit_list);
5186 }
5187
5188 btrfs_destroy_delayed_refs(cur_trans, fs_info);
5189
5190 cur_trans->state = TRANS_STATE_COMMIT_START;
5191 wake_up(&fs_info->transaction_blocked_wait);
5192
5193 cur_trans->state = TRANS_STATE_UNBLOCKED;
5194 wake_up(&fs_info->transaction_wait);
5195
5196 btrfs_destroy_delayed_inodes(fs_info);
5197
5198 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
5199 EXTENT_DIRTY);
5200 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
5201
5202 btrfs_free_redirty_list(cur_trans);
5203
5204 cur_trans->state =TRANS_STATE_COMPLETED;
5205 wake_up(&cur_trans->commit_wait);
5206 }
5207
btrfs_cleanup_transaction(struct btrfs_fs_info * fs_info)5208 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
5209 {
5210 struct btrfs_transaction *t;
5211
5212 mutex_lock(&fs_info->transaction_kthread_mutex);
5213
5214 spin_lock(&fs_info->trans_lock);
5215 while (!list_empty(&fs_info->trans_list)) {
5216 t = list_first_entry(&fs_info->trans_list,
5217 struct btrfs_transaction, list);
5218 if (t->state >= TRANS_STATE_COMMIT_START) {
5219 refcount_inc(&t->use_count);
5220 spin_unlock(&fs_info->trans_lock);
5221 btrfs_wait_for_commit(fs_info, t->transid);
5222 btrfs_put_transaction(t);
5223 spin_lock(&fs_info->trans_lock);
5224 continue;
5225 }
5226 if (t == fs_info->running_transaction) {
5227 t->state = TRANS_STATE_COMMIT_DOING;
5228 spin_unlock(&fs_info->trans_lock);
5229 /*
5230 * We wait for 0 num_writers since we don't hold a trans
5231 * handle open currently for this transaction.
5232 */
5233 wait_event(t->writer_wait,
5234 atomic_read(&t->num_writers) == 0);
5235 } else {
5236 spin_unlock(&fs_info->trans_lock);
5237 }
5238 btrfs_cleanup_one_transaction(t, fs_info);
5239
5240 spin_lock(&fs_info->trans_lock);
5241 if (t == fs_info->running_transaction)
5242 fs_info->running_transaction = NULL;
5243 list_del_init(&t->list);
5244 spin_unlock(&fs_info->trans_lock);
5245
5246 btrfs_put_transaction(t);
5247 trace_btrfs_transaction_commit(fs_info);
5248 spin_lock(&fs_info->trans_lock);
5249 }
5250 spin_unlock(&fs_info->trans_lock);
5251 btrfs_destroy_all_ordered_extents(fs_info);
5252 btrfs_destroy_delayed_inodes(fs_info);
5253 btrfs_assert_delayed_root_empty(fs_info);
5254 btrfs_destroy_all_delalloc_inodes(fs_info);
5255 btrfs_drop_all_logs(fs_info);
5256 mutex_unlock(&fs_info->transaction_kthread_mutex);
5257
5258 return 0;
5259 }
5260
btrfs_init_root_free_objectid(struct btrfs_root * root)5261 int btrfs_init_root_free_objectid(struct btrfs_root *root)
5262 {
5263 struct btrfs_path *path;
5264 int ret;
5265 struct extent_buffer *l;
5266 struct btrfs_key search_key;
5267 struct btrfs_key found_key;
5268 int slot;
5269
5270 path = btrfs_alloc_path();
5271 if (!path)
5272 return -ENOMEM;
5273
5274 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
5275 search_key.type = -1;
5276 search_key.offset = (u64)-1;
5277 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5278 if (ret < 0)
5279 goto error;
5280 BUG_ON(ret == 0); /* Corruption */
5281 if (path->slots[0] > 0) {
5282 slot = path->slots[0] - 1;
5283 l = path->nodes[0];
5284 btrfs_item_key_to_cpu(l, &found_key, slot);
5285 root->free_objectid = max_t(u64, found_key.objectid + 1,
5286 BTRFS_FIRST_FREE_OBJECTID);
5287 } else {
5288 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
5289 }
5290 ret = 0;
5291 error:
5292 btrfs_free_path(path);
5293 return ret;
5294 }
5295
btrfs_get_free_objectid(struct btrfs_root * root,u64 * objectid)5296 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
5297 {
5298 int ret;
5299 mutex_lock(&root->objectid_mutex);
5300
5301 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
5302 btrfs_warn(root->fs_info,
5303 "the objectid of root %llu reaches its highest value",
5304 root->root_key.objectid);
5305 ret = -ENOSPC;
5306 goto out;
5307 }
5308
5309 *objectid = root->free_objectid++;
5310 ret = 0;
5311 out:
5312 mutex_unlock(&root->objectid_mutex);
5313 return ret;
5314 }
5315