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 "bio.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 #include "fs.h"
47 #include "accessors.h"
48 #include "extent-tree.h"
49 #include "root-tree.h"
50 #include "defrag.h"
51 #include "uuid-tree.h"
52 #include "relocation.h"
53 #include "scrub.h"
54 #include "super.h"
55 
56 #define BTRFS_SUPER_FLAG_SUPP	(BTRFS_HEADER_FLAG_WRITTEN |\
57 				 BTRFS_HEADER_FLAG_RELOC |\
58 				 BTRFS_SUPER_FLAG_ERROR |\
59 				 BTRFS_SUPER_FLAG_SEEDING |\
60 				 BTRFS_SUPER_FLAG_METADUMP |\
61 				 BTRFS_SUPER_FLAG_METADUMP_V2)
62 
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  * Compute the csum of a btree block and store the result to provided buffer.
74  */
csum_tree_block(struct extent_buffer * buf,u8 * result)75 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
76 {
77 	struct btrfs_fs_info *fs_info = buf->fs_info;
78 	const int num_pages = num_extent_pages(buf);
79 	const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
80 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
81 	char *kaddr;
82 	int i;
83 
84 	shash->tfm = fs_info->csum_shash;
85 	crypto_shash_init(shash);
86 	kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
87 	crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
88 			    first_page_part - BTRFS_CSUM_SIZE);
89 
90 	for (i = 1; i < num_pages && INLINE_EXTENT_BUFFER_PAGES > 1; i++) {
91 		kaddr = page_address(buf->pages[i]);
92 		crypto_shash_update(shash, kaddr, PAGE_SIZE);
93 	}
94 	memset(result, 0, BTRFS_CSUM_SIZE);
95 	crypto_shash_final(shash, result);
96 }
97 
98 /*
99  * we can't consider a given block up to date unless the transid of the
100  * block matches the transid in the parent node's pointer.  This is how we
101  * detect blocks that either didn't get written at all or got written
102  * in the wrong place.
103  */
btrfs_buffer_uptodate(struct extent_buffer * eb,u64 parent_transid,int atomic)104 int btrfs_buffer_uptodate(struct extent_buffer *eb, u64 parent_transid, int atomic)
105 {
106 	if (!extent_buffer_uptodate(eb))
107 		return 0;
108 
109 	if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
110 		return 1;
111 
112 	if (atomic)
113 		return -EAGAIN;
114 
115 	if (!extent_buffer_uptodate(eb) ||
116 	    btrfs_header_generation(eb) != parent_transid) {
117 		btrfs_err_rl(eb->fs_info,
118 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
119 			eb->start, eb->read_mirror,
120 			parent_transid, btrfs_header_generation(eb));
121 		clear_extent_buffer_uptodate(eb);
122 		return 0;
123 	}
124 	return 1;
125 }
126 
btrfs_supported_super_csum(u16 csum_type)127 static bool btrfs_supported_super_csum(u16 csum_type)
128 {
129 	switch (csum_type) {
130 	case BTRFS_CSUM_TYPE_CRC32:
131 	case BTRFS_CSUM_TYPE_XXHASH:
132 	case BTRFS_CSUM_TYPE_SHA256:
133 	case BTRFS_CSUM_TYPE_BLAKE2:
134 		return true;
135 	default:
136 		return false;
137 	}
138 }
139 
140 /*
141  * Return 0 if the superblock checksum type matches the checksum value of that
142  * algorithm. Pass the raw disk superblock data.
143  */
btrfs_check_super_csum(struct btrfs_fs_info * fs_info,const struct btrfs_super_block * disk_sb)144 int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
145 			   const struct btrfs_super_block *disk_sb)
146 {
147 	char result[BTRFS_CSUM_SIZE];
148 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
149 
150 	shash->tfm = fs_info->csum_shash;
151 
152 	/*
153 	 * The super_block structure does not span the whole
154 	 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
155 	 * filled with zeros and is included in the checksum.
156 	 */
157 	crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
158 			    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
159 
160 	if (memcmp(disk_sb->csum, result, fs_info->csum_size))
161 		return 1;
162 
163 	return 0;
164 }
165 
btrfs_repair_eb_io_failure(const struct extent_buffer * eb,int mirror_num)166 static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
167 				      int mirror_num)
168 {
169 	struct btrfs_fs_info *fs_info = eb->fs_info;
170 	int i, num_pages = num_extent_pages(eb);
171 	int ret = 0;
172 
173 	if (sb_rdonly(fs_info->sb))
174 		return -EROFS;
175 
176 	for (i = 0; i < num_pages; i++) {
177 		struct page *p = eb->pages[i];
178 		u64 start = max_t(u64, eb->start, page_offset(p));
179 		u64 end = min_t(u64, eb->start + eb->len, page_offset(p) + PAGE_SIZE);
180 		u32 len = end - start;
181 
182 		ret = btrfs_repair_io_failure(fs_info, 0, start, len,
183 				start, p, offset_in_page(start), mirror_num);
184 		if (ret)
185 			break;
186 	}
187 
188 	return ret;
189 }
190 
191 /*
192  * helper to read a given tree block, doing retries as required when
193  * the checksums don't match and we have alternate mirrors to try.
194  *
195  * @check:		expected tree parentness check, see the comments of the
196  *			structure for details.
197  */
btrfs_read_extent_buffer(struct extent_buffer * eb,struct btrfs_tree_parent_check * check)198 int btrfs_read_extent_buffer(struct extent_buffer *eb,
199 			     struct btrfs_tree_parent_check *check)
200 {
201 	struct btrfs_fs_info *fs_info = eb->fs_info;
202 	int failed = 0;
203 	int ret;
204 	int num_copies = 0;
205 	int mirror_num = 0;
206 	int failed_mirror = 0;
207 
208 	ASSERT(check);
209 
210 	while (1) {
211 		clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
212 		ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check);
213 		if (!ret)
214 			break;
215 
216 		num_copies = btrfs_num_copies(fs_info,
217 					      eb->start, eb->len);
218 		if (num_copies == 1)
219 			break;
220 
221 		if (!failed_mirror) {
222 			failed = 1;
223 			failed_mirror = eb->read_mirror;
224 		}
225 
226 		mirror_num++;
227 		if (mirror_num == failed_mirror)
228 			mirror_num++;
229 
230 		if (mirror_num > num_copies)
231 			break;
232 	}
233 
234 	if (failed && !ret && failed_mirror)
235 		btrfs_repair_eb_io_failure(eb, failed_mirror);
236 
237 	return ret;
238 }
239 
240 /*
241  * Checksum a dirty tree block before IO.
242  */
btree_csum_one_bio(struct btrfs_bio * bbio)243 blk_status_t btree_csum_one_bio(struct btrfs_bio *bbio)
244 {
245 	struct extent_buffer *eb = bbio->private;
246 	struct btrfs_fs_info *fs_info = eb->fs_info;
247 	u64 found_start = btrfs_header_bytenr(eb);
248 	u8 result[BTRFS_CSUM_SIZE];
249 	int ret;
250 
251 	/* Btree blocks are always contiguous on disk. */
252 	if (WARN_ON_ONCE(bbio->file_offset != eb->start))
253 		return BLK_STS_IOERR;
254 	if (WARN_ON_ONCE(bbio->bio.bi_iter.bi_size != eb->len))
255 		return BLK_STS_IOERR;
256 
257 	if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
258 		WARN_ON_ONCE(found_start != 0);
259 		return BLK_STS_OK;
260 	}
261 
262 	if (WARN_ON_ONCE(found_start != eb->start))
263 		return BLK_STS_IOERR;
264 	if (WARN_ON(!btrfs_page_test_uptodate(fs_info, eb->pages[0], eb->start,
265 					      eb->len)))
266 		return BLK_STS_IOERR;
267 
268 	ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
269 				    offsetof(struct btrfs_header, fsid),
270 				    BTRFS_FSID_SIZE) == 0);
271 	csum_tree_block(eb, result);
272 
273 	if (btrfs_header_level(eb))
274 		ret = btrfs_check_node(eb);
275 	else
276 		ret = btrfs_check_leaf(eb);
277 
278 	if (ret < 0)
279 		goto error;
280 
281 	/*
282 	 * Also check the generation, the eb reached here must be newer than
283 	 * last committed. Or something seriously wrong happened.
284 	 */
285 	if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) {
286 		ret = -EUCLEAN;
287 		btrfs_err(fs_info,
288 			"block=%llu bad generation, have %llu expect > %llu",
289 			  eb->start, btrfs_header_generation(eb),
290 			  fs_info->last_trans_committed);
291 		goto error;
292 	}
293 	write_extent_buffer(eb, result, 0, fs_info->csum_size);
294 	return BLK_STS_OK;
295 
296 error:
297 	btrfs_print_tree(eb, 0);
298 	btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
299 		  eb->start);
300 	/*
301 	 * Be noisy if this is an extent buffer from a log tree. We don't abort
302 	 * a transaction in case there's a bad log tree extent buffer, we just
303 	 * fallback to a transaction commit. Still we want to know when there is
304 	 * a bad log tree extent buffer, as that may signal a bug somewhere.
305 	 */
306 	WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
307 		btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
308 	return errno_to_blk_status(ret);
309 }
310 
check_tree_block_fsid(struct extent_buffer * eb)311 static bool check_tree_block_fsid(struct extent_buffer *eb)
312 {
313 	struct btrfs_fs_info *fs_info = eb->fs_info;
314 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
315 	u8 fsid[BTRFS_FSID_SIZE];
316 
317 	read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
318 			   BTRFS_FSID_SIZE);
319 
320 	/*
321 	 * alloc_fs_devices() copies the fsid into metadata_uuid if the
322 	 * metadata_uuid is unset in the superblock, including for a seed device.
323 	 * So, we can use fs_devices->metadata_uuid.
324 	 */
325 	if (memcmp(fsid, fs_info->fs_devices->metadata_uuid, BTRFS_FSID_SIZE) == 0)
326 		return false;
327 
328 	list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
329 		if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
330 			return false;
331 
332 	return true;
333 }
334 
335 /* Do basic extent buffer checks at read time */
btrfs_validate_extent_buffer(struct extent_buffer * eb,struct btrfs_tree_parent_check * check)336 int btrfs_validate_extent_buffer(struct extent_buffer *eb,
337 				 struct btrfs_tree_parent_check *check)
338 {
339 	struct btrfs_fs_info *fs_info = eb->fs_info;
340 	u64 found_start;
341 	const u32 csum_size = fs_info->csum_size;
342 	u8 found_level;
343 	u8 result[BTRFS_CSUM_SIZE];
344 	const u8 *header_csum;
345 	int ret = 0;
346 
347 	ASSERT(check);
348 
349 	found_start = btrfs_header_bytenr(eb);
350 	if (found_start != eb->start) {
351 		btrfs_err_rl(fs_info,
352 			"bad tree block start, mirror %u want %llu have %llu",
353 			     eb->read_mirror, eb->start, found_start);
354 		ret = -EIO;
355 		goto out;
356 	}
357 	if (check_tree_block_fsid(eb)) {
358 		btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
359 			     eb->start, eb->read_mirror);
360 		ret = -EIO;
361 		goto out;
362 	}
363 	found_level = btrfs_header_level(eb);
364 	if (found_level >= BTRFS_MAX_LEVEL) {
365 		btrfs_err(fs_info,
366 			"bad tree block level, mirror %u level %d on logical %llu",
367 			eb->read_mirror, btrfs_header_level(eb), eb->start);
368 		ret = -EIO;
369 		goto out;
370 	}
371 
372 	csum_tree_block(eb, result);
373 	header_csum = page_address(eb->pages[0]) +
374 		get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
375 
376 	if (memcmp(result, header_csum, csum_size) != 0) {
377 		btrfs_warn_rl(fs_info,
378 "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d",
379 			      eb->start, eb->read_mirror,
380 			      CSUM_FMT_VALUE(csum_size, header_csum),
381 			      CSUM_FMT_VALUE(csum_size, result),
382 			      btrfs_header_level(eb));
383 		ret = -EUCLEAN;
384 		goto out;
385 	}
386 
387 	if (found_level != check->level) {
388 		btrfs_err(fs_info,
389 		"level verify failed on logical %llu mirror %u wanted %u found %u",
390 			  eb->start, eb->read_mirror, check->level, found_level);
391 		ret = -EIO;
392 		goto out;
393 	}
394 	if (unlikely(check->transid &&
395 		     btrfs_header_generation(eb) != check->transid)) {
396 		btrfs_err_rl(eb->fs_info,
397 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
398 				eb->start, eb->read_mirror, check->transid,
399 				btrfs_header_generation(eb));
400 		ret = -EIO;
401 		goto out;
402 	}
403 	if (check->has_first_key) {
404 		struct btrfs_key *expect_key = &check->first_key;
405 		struct btrfs_key found_key;
406 
407 		if (found_level)
408 			btrfs_node_key_to_cpu(eb, &found_key, 0);
409 		else
410 			btrfs_item_key_to_cpu(eb, &found_key, 0);
411 		if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
412 			btrfs_err(fs_info,
413 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
414 				  eb->start, check->transid,
415 				  expect_key->objectid,
416 				  expect_key->type, expect_key->offset,
417 				  found_key.objectid, found_key.type,
418 				  found_key.offset);
419 			ret = -EUCLEAN;
420 			goto out;
421 		}
422 	}
423 	if (check->owner_root) {
424 		ret = btrfs_check_eb_owner(eb, check->owner_root);
425 		if (ret < 0)
426 			goto out;
427 	}
428 
429 	/*
430 	 * If this is a leaf block and it is corrupt, set the corrupt bit so
431 	 * that we don't try and read the other copies of this block, just
432 	 * return -EIO.
433 	 */
434 	if (found_level == 0 && btrfs_check_leaf(eb)) {
435 		set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
436 		ret = -EIO;
437 	}
438 
439 	if (found_level > 0 && btrfs_check_node(eb))
440 		ret = -EIO;
441 
442 	if (ret)
443 		btrfs_err(fs_info,
444 		"read time tree block corruption detected on logical %llu mirror %u",
445 			  eb->start, eb->read_mirror);
446 out:
447 	return ret;
448 }
449 
450 #ifdef CONFIG_MIGRATION
btree_migrate_folio(struct address_space * mapping,struct folio * dst,struct folio * src,enum migrate_mode mode)451 static int btree_migrate_folio(struct address_space *mapping,
452 		struct folio *dst, struct folio *src, enum migrate_mode mode)
453 {
454 	/*
455 	 * we can't safely write a btree page from here,
456 	 * we haven't done the locking hook
457 	 */
458 	if (folio_test_dirty(src))
459 		return -EAGAIN;
460 	/*
461 	 * Buffers may be managed in a filesystem specific way.
462 	 * We must have no buffers or drop them.
463 	 */
464 	if (folio_get_private(src) &&
465 	    !filemap_release_folio(src, GFP_KERNEL))
466 		return -EAGAIN;
467 	return migrate_folio(mapping, dst, src, mode);
468 }
469 #else
470 #define btree_migrate_folio NULL
471 #endif
472 
btree_writepages(struct address_space * mapping,struct writeback_control * wbc)473 static int btree_writepages(struct address_space *mapping,
474 			    struct writeback_control *wbc)
475 {
476 	struct btrfs_fs_info *fs_info;
477 	int ret;
478 
479 	if (wbc->sync_mode == WB_SYNC_NONE) {
480 
481 		if (wbc->for_kupdate)
482 			return 0;
483 
484 		fs_info = BTRFS_I(mapping->host)->root->fs_info;
485 		/* this is a bit racy, but that's ok */
486 		ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
487 					     BTRFS_DIRTY_METADATA_THRESH,
488 					     fs_info->dirty_metadata_batch);
489 		if (ret < 0)
490 			return 0;
491 	}
492 	return btree_write_cache_pages(mapping, wbc);
493 }
494 
btree_release_folio(struct folio * folio,gfp_t gfp_flags)495 static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
496 {
497 	if (folio_test_writeback(folio) || folio_test_dirty(folio))
498 		return false;
499 
500 	return try_release_extent_buffer(&folio->page);
501 }
502 
btree_invalidate_folio(struct folio * folio,size_t offset,size_t length)503 static void btree_invalidate_folio(struct folio *folio, size_t offset,
504 				 size_t length)
505 {
506 	struct extent_io_tree *tree;
507 	tree = &BTRFS_I(folio->mapping->host)->io_tree;
508 	extent_invalidate_folio(tree, folio, offset);
509 	btree_release_folio(folio, GFP_NOFS);
510 	if (folio_get_private(folio)) {
511 		btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info,
512 			   "folio private not zero on folio %llu",
513 			   (unsigned long long)folio_pos(folio));
514 		folio_detach_private(folio);
515 	}
516 }
517 
518 #ifdef DEBUG
btree_dirty_folio(struct address_space * mapping,struct folio * folio)519 static bool btree_dirty_folio(struct address_space *mapping,
520 		struct folio *folio)
521 {
522 	struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
523 	struct btrfs_subpage_info *spi = fs_info->subpage_info;
524 	struct btrfs_subpage *subpage;
525 	struct extent_buffer *eb;
526 	int cur_bit = 0;
527 	u64 page_start = folio_pos(folio);
528 
529 	if (fs_info->sectorsize == PAGE_SIZE) {
530 		eb = folio_get_private(folio);
531 		BUG_ON(!eb);
532 		BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
533 		BUG_ON(!atomic_read(&eb->refs));
534 		btrfs_assert_tree_write_locked(eb);
535 		return filemap_dirty_folio(mapping, folio);
536 	}
537 
538 	ASSERT(spi);
539 	subpage = folio_get_private(folio);
540 
541 	for (cur_bit = spi->dirty_offset;
542 	     cur_bit < spi->dirty_offset + spi->bitmap_nr_bits;
543 	     cur_bit++) {
544 		unsigned long flags;
545 		u64 cur;
546 
547 		spin_lock_irqsave(&subpage->lock, flags);
548 		if (!test_bit(cur_bit, subpage->bitmaps)) {
549 			spin_unlock_irqrestore(&subpage->lock, flags);
550 			continue;
551 		}
552 		spin_unlock_irqrestore(&subpage->lock, flags);
553 		cur = page_start + cur_bit * fs_info->sectorsize;
554 
555 		eb = find_extent_buffer(fs_info, cur);
556 		ASSERT(eb);
557 		ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
558 		ASSERT(atomic_read(&eb->refs));
559 		btrfs_assert_tree_write_locked(eb);
560 		free_extent_buffer(eb);
561 
562 		cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits) - 1;
563 	}
564 	return filemap_dirty_folio(mapping, folio);
565 }
566 #else
567 #define btree_dirty_folio filemap_dirty_folio
568 #endif
569 
570 static const struct address_space_operations btree_aops = {
571 	.writepages	= btree_writepages,
572 	.release_folio	= btree_release_folio,
573 	.invalidate_folio = btree_invalidate_folio,
574 	.migrate_folio	= btree_migrate_folio,
575 	.dirty_folio	= btree_dirty_folio,
576 };
577 
btrfs_find_create_tree_block(struct btrfs_fs_info * fs_info,u64 bytenr,u64 owner_root,int level)578 struct extent_buffer *btrfs_find_create_tree_block(
579 						struct btrfs_fs_info *fs_info,
580 						u64 bytenr, u64 owner_root,
581 						int level)
582 {
583 	if (btrfs_is_testing(fs_info))
584 		return alloc_test_extent_buffer(fs_info, bytenr);
585 	return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
586 }
587 
588 /*
589  * Read tree block at logical address @bytenr and do variant basic but critical
590  * verification.
591  *
592  * @check:		expected tree parentness check, see comments of the
593  *			structure for details.
594  */
read_tree_block(struct btrfs_fs_info * fs_info,u64 bytenr,struct btrfs_tree_parent_check * check)595 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
596 				      struct btrfs_tree_parent_check *check)
597 {
598 	struct extent_buffer *buf = NULL;
599 	int ret;
600 
601 	ASSERT(check);
602 
603 	buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
604 					   check->level);
605 	if (IS_ERR(buf))
606 		return buf;
607 
608 	ret = btrfs_read_extent_buffer(buf, check);
609 	if (ret) {
610 		free_extent_buffer_stale(buf);
611 		return ERR_PTR(ret);
612 	}
613 	if (btrfs_check_eb_owner(buf, check->owner_root)) {
614 		free_extent_buffer_stale(buf);
615 		return ERR_PTR(-EUCLEAN);
616 	}
617 	return buf;
618 
619 }
620 
__setup_root(struct btrfs_root * root,struct btrfs_fs_info * fs_info,u64 objectid)621 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
622 			 u64 objectid)
623 {
624 	bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
625 
626 	memset(&root->root_key, 0, sizeof(root->root_key));
627 	memset(&root->root_item, 0, sizeof(root->root_item));
628 	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
629 	root->fs_info = fs_info;
630 	root->root_key.objectid = objectid;
631 	root->node = NULL;
632 	root->commit_root = NULL;
633 	root->state = 0;
634 	RB_CLEAR_NODE(&root->rb_node);
635 
636 	root->last_trans = 0;
637 	root->free_objectid = 0;
638 	root->nr_delalloc_inodes = 0;
639 	root->nr_ordered_extents = 0;
640 	root->inode_tree = RB_ROOT;
641 	INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
642 
643 	btrfs_init_root_block_rsv(root);
644 
645 	INIT_LIST_HEAD(&root->dirty_list);
646 	INIT_LIST_HEAD(&root->root_list);
647 	INIT_LIST_HEAD(&root->delalloc_inodes);
648 	INIT_LIST_HEAD(&root->delalloc_root);
649 	INIT_LIST_HEAD(&root->ordered_extents);
650 	INIT_LIST_HEAD(&root->ordered_root);
651 	INIT_LIST_HEAD(&root->reloc_dirty_list);
652 	INIT_LIST_HEAD(&root->logged_list[0]);
653 	INIT_LIST_HEAD(&root->logged_list[1]);
654 	spin_lock_init(&root->inode_lock);
655 	spin_lock_init(&root->delalloc_lock);
656 	spin_lock_init(&root->ordered_extent_lock);
657 	spin_lock_init(&root->accounting_lock);
658 	spin_lock_init(&root->log_extents_lock[0]);
659 	spin_lock_init(&root->log_extents_lock[1]);
660 	spin_lock_init(&root->qgroup_meta_rsv_lock);
661 	mutex_init(&root->objectid_mutex);
662 	mutex_init(&root->log_mutex);
663 	mutex_init(&root->ordered_extent_mutex);
664 	mutex_init(&root->delalloc_mutex);
665 	init_waitqueue_head(&root->qgroup_flush_wait);
666 	init_waitqueue_head(&root->log_writer_wait);
667 	init_waitqueue_head(&root->log_commit_wait[0]);
668 	init_waitqueue_head(&root->log_commit_wait[1]);
669 	INIT_LIST_HEAD(&root->log_ctxs[0]);
670 	INIT_LIST_HEAD(&root->log_ctxs[1]);
671 	atomic_set(&root->log_commit[0], 0);
672 	atomic_set(&root->log_commit[1], 0);
673 	atomic_set(&root->log_writers, 0);
674 	atomic_set(&root->log_batch, 0);
675 	refcount_set(&root->refs, 1);
676 	atomic_set(&root->snapshot_force_cow, 0);
677 	atomic_set(&root->nr_swapfiles, 0);
678 	root->log_transid = 0;
679 	root->log_transid_committed = -1;
680 	root->last_log_commit = 0;
681 	root->anon_dev = 0;
682 	if (!dummy) {
683 		extent_io_tree_init(fs_info, &root->dirty_log_pages,
684 				    IO_TREE_ROOT_DIRTY_LOG_PAGES);
685 		extent_io_tree_init(fs_info, &root->log_csum_range,
686 				    IO_TREE_LOG_CSUM_RANGE);
687 	}
688 
689 	spin_lock_init(&root->root_item_lock);
690 	btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
691 #ifdef CONFIG_BTRFS_DEBUG
692 	INIT_LIST_HEAD(&root->leak_list);
693 	spin_lock(&fs_info->fs_roots_radix_lock);
694 	list_add_tail(&root->leak_list, &fs_info->allocated_roots);
695 	spin_unlock(&fs_info->fs_roots_radix_lock);
696 #endif
697 }
698 
btrfs_alloc_root(struct btrfs_fs_info * fs_info,u64 objectid,gfp_t flags)699 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
700 					   u64 objectid, gfp_t flags)
701 {
702 	struct btrfs_root *root = kzalloc(sizeof(*root), flags);
703 	if (root)
704 		__setup_root(root, fs_info, objectid);
705 	return root;
706 }
707 
708 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
709 /* Should only be used by the testing infrastructure */
btrfs_alloc_dummy_root(struct btrfs_fs_info * fs_info)710 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
711 {
712 	struct btrfs_root *root;
713 
714 	if (!fs_info)
715 		return ERR_PTR(-EINVAL);
716 
717 	root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
718 	if (!root)
719 		return ERR_PTR(-ENOMEM);
720 
721 	/* We don't use the stripesize in selftest, set it as sectorsize */
722 	root->alloc_bytenr = 0;
723 
724 	return root;
725 }
726 #endif
727 
global_root_cmp(struct rb_node * a_node,const struct rb_node * b_node)728 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
729 {
730 	const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
731 	const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
732 
733 	return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
734 }
735 
global_root_key_cmp(const void * k,const struct rb_node * node)736 static int global_root_key_cmp(const void *k, const struct rb_node *node)
737 {
738 	const struct btrfs_key *key = k;
739 	const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
740 
741 	return btrfs_comp_cpu_keys(key, &root->root_key);
742 }
743 
btrfs_global_root_insert(struct btrfs_root * root)744 int btrfs_global_root_insert(struct btrfs_root *root)
745 {
746 	struct btrfs_fs_info *fs_info = root->fs_info;
747 	struct rb_node *tmp;
748 	int ret = 0;
749 
750 	write_lock(&fs_info->global_root_lock);
751 	tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
752 	write_unlock(&fs_info->global_root_lock);
753 
754 	if (tmp) {
755 		ret = -EEXIST;
756 		btrfs_warn(fs_info, "global root %llu %llu already exists",
757 				root->root_key.objectid, root->root_key.offset);
758 	}
759 	return ret;
760 }
761 
btrfs_global_root_delete(struct btrfs_root * root)762 void btrfs_global_root_delete(struct btrfs_root *root)
763 {
764 	struct btrfs_fs_info *fs_info = root->fs_info;
765 
766 	write_lock(&fs_info->global_root_lock);
767 	rb_erase(&root->rb_node, &fs_info->global_root_tree);
768 	write_unlock(&fs_info->global_root_lock);
769 }
770 
btrfs_global_root(struct btrfs_fs_info * fs_info,struct btrfs_key * key)771 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
772 				     struct btrfs_key *key)
773 {
774 	struct rb_node *node;
775 	struct btrfs_root *root = NULL;
776 
777 	read_lock(&fs_info->global_root_lock);
778 	node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
779 	if (node)
780 		root = container_of(node, struct btrfs_root, rb_node);
781 	read_unlock(&fs_info->global_root_lock);
782 
783 	return root;
784 }
785 
btrfs_global_root_id(struct btrfs_fs_info * fs_info,u64 bytenr)786 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
787 {
788 	struct btrfs_block_group *block_group;
789 	u64 ret;
790 
791 	if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
792 		return 0;
793 
794 	if (bytenr)
795 		block_group = btrfs_lookup_block_group(fs_info, bytenr);
796 	else
797 		block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
798 	ASSERT(block_group);
799 	if (!block_group)
800 		return 0;
801 	ret = block_group->global_root_id;
802 	btrfs_put_block_group(block_group);
803 
804 	return ret;
805 }
806 
btrfs_csum_root(struct btrfs_fs_info * fs_info,u64 bytenr)807 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
808 {
809 	struct btrfs_key key = {
810 		.objectid = BTRFS_CSUM_TREE_OBJECTID,
811 		.type = BTRFS_ROOT_ITEM_KEY,
812 		.offset = btrfs_global_root_id(fs_info, bytenr),
813 	};
814 
815 	return btrfs_global_root(fs_info, &key);
816 }
817 
btrfs_extent_root(struct btrfs_fs_info * fs_info,u64 bytenr)818 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
819 {
820 	struct btrfs_key key = {
821 		.objectid = BTRFS_EXTENT_TREE_OBJECTID,
822 		.type = BTRFS_ROOT_ITEM_KEY,
823 		.offset = btrfs_global_root_id(fs_info, bytenr),
824 	};
825 
826 	return btrfs_global_root(fs_info, &key);
827 }
828 
btrfs_block_group_root(struct btrfs_fs_info * fs_info)829 struct btrfs_root *btrfs_block_group_root(struct btrfs_fs_info *fs_info)
830 {
831 	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE))
832 		return fs_info->block_group_root;
833 	return btrfs_extent_root(fs_info, 0);
834 }
835 
btrfs_create_tree(struct btrfs_trans_handle * trans,u64 objectid)836 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
837 				     u64 objectid)
838 {
839 	struct btrfs_fs_info *fs_info = trans->fs_info;
840 	struct extent_buffer *leaf;
841 	struct btrfs_root *tree_root = fs_info->tree_root;
842 	struct btrfs_root *root;
843 	struct btrfs_key key;
844 	unsigned int nofs_flag;
845 	int ret = 0;
846 
847 	/*
848 	 * We're holding a transaction handle, so use a NOFS memory allocation
849 	 * context to avoid deadlock if reclaim happens.
850 	 */
851 	nofs_flag = memalloc_nofs_save();
852 	root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
853 	memalloc_nofs_restore(nofs_flag);
854 	if (!root)
855 		return ERR_PTR(-ENOMEM);
856 
857 	root->root_key.objectid = objectid;
858 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
859 	root->root_key.offset = 0;
860 
861 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
862 				      BTRFS_NESTING_NORMAL);
863 	if (IS_ERR(leaf)) {
864 		ret = PTR_ERR(leaf);
865 		leaf = NULL;
866 		goto fail;
867 	}
868 
869 	root->node = leaf;
870 	btrfs_mark_buffer_dirty(trans, leaf);
871 
872 	root->commit_root = btrfs_root_node(root);
873 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
874 
875 	btrfs_set_root_flags(&root->root_item, 0);
876 	btrfs_set_root_limit(&root->root_item, 0);
877 	btrfs_set_root_bytenr(&root->root_item, leaf->start);
878 	btrfs_set_root_generation(&root->root_item, trans->transid);
879 	btrfs_set_root_level(&root->root_item, 0);
880 	btrfs_set_root_refs(&root->root_item, 1);
881 	btrfs_set_root_used(&root->root_item, leaf->len);
882 	btrfs_set_root_last_snapshot(&root->root_item, 0);
883 	btrfs_set_root_dirid(&root->root_item, 0);
884 	if (is_fstree(objectid))
885 		generate_random_guid(root->root_item.uuid);
886 	else
887 		export_guid(root->root_item.uuid, &guid_null);
888 	btrfs_set_root_drop_level(&root->root_item, 0);
889 
890 	btrfs_tree_unlock(leaf);
891 
892 	key.objectid = objectid;
893 	key.type = BTRFS_ROOT_ITEM_KEY;
894 	key.offset = 0;
895 	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
896 	if (ret)
897 		goto fail;
898 
899 	return root;
900 
901 fail:
902 	btrfs_put_root(root);
903 
904 	return ERR_PTR(ret);
905 }
906 
alloc_log_tree(struct btrfs_trans_handle * trans,struct btrfs_fs_info * fs_info)907 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
908 					 struct btrfs_fs_info *fs_info)
909 {
910 	struct btrfs_root *root;
911 
912 	root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
913 	if (!root)
914 		return ERR_PTR(-ENOMEM);
915 
916 	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
917 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
918 	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
919 
920 	return root;
921 }
922 
btrfs_alloc_log_tree_node(struct btrfs_trans_handle * trans,struct btrfs_root * root)923 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
924 			      struct btrfs_root *root)
925 {
926 	struct extent_buffer *leaf;
927 
928 	/*
929 	 * DON'T set SHAREABLE bit for log trees.
930 	 *
931 	 * Log trees are not exposed to user space thus can't be snapshotted,
932 	 * and they go away before a real commit is actually done.
933 	 *
934 	 * They do store pointers to file data extents, and those reference
935 	 * counts still get updated (along with back refs to the log tree).
936 	 */
937 
938 	leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
939 			NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
940 	if (IS_ERR(leaf))
941 		return PTR_ERR(leaf);
942 
943 	root->node = leaf;
944 
945 	btrfs_mark_buffer_dirty(trans, root->node);
946 	btrfs_tree_unlock(root->node);
947 
948 	return 0;
949 }
950 
btrfs_init_log_root_tree(struct btrfs_trans_handle * trans,struct btrfs_fs_info * fs_info)951 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
952 			     struct btrfs_fs_info *fs_info)
953 {
954 	struct btrfs_root *log_root;
955 
956 	log_root = alloc_log_tree(trans, fs_info);
957 	if (IS_ERR(log_root))
958 		return PTR_ERR(log_root);
959 
960 	if (!btrfs_is_zoned(fs_info)) {
961 		int ret = btrfs_alloc_log_tree_node(trans, log_root);
962 
963 		if (ret) {
964 			btrfs_put_root(log_root);
965 			return ret;
966 		}
967 	}
968 
969 	WARN_ON(fs_info->log_root_tree);
970 	fs_info->log_root_tree = log_root;
971 	return 0;
972 }
973 
btrfs_add_log_tree(struct btrfs_trans_handle * trans,struct btrfs_root * root)974 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
975 		       struct btrfs_root *root)
976 {
977 	struct btrfs_fs_info *fs_info = root->fs_info;
978 	struct btrfs_root *log_root;
979 	struct btrfs_inode_item *inode_item;
980 	int ret;
981 
982 	log_root = alloc_log_tree(trans, fs_info);
983 	if (IS_ERR(log_root))
984 		return PTR_ERR(log_root);
985 
986 	ret = btrfs_alloc_log_tree_node(trans, log_root);
987 	if (ret) {
988 		btrfs_put_root(log_root);
989 		return ret;
990 	}
991 
992 	log_root->last_trans = trans->transid;
993 	log_root->root_key.offset = root->root_key.objectid;
994 
995 	inode_item = &log_root->root_item.inode;
996 	btrfs_set_stack_inode_generation(inode_item, 1);
997 	btrfs_set_stack_inode_size(inode_item, 3);
998 	btrfs_set_stack_inode_nlink(inode_item, 1);
999 	btrfs_set_stack_inode_nbytes(inode_item,
1000 				     fs_info->nodesize);
1001 	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1002 
1003 	btrfs_set_root_node(&log_root->root_item, log_root->node);
1004 
1005 	WARN_ON(root->log_root);
1006 	root->log_root = log_root;
1007 	root->log_transid = 0;
1008 	root->log_transid_committed = -1;
1009 	root->last_log_commit = 0;
1010 	return 0;
1011 }
1012 
read_tree_root_path(struct btrfs_root * tree_root,struct btrfs_path * path,struct btrfs_key * key)1013 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1014 					      struct btrfs_path *path,
1015 					      struct btrfs_key *key)
1016 {
1017 	struct btrfs_root *root;
1018 	struct btrfs_tree_parent_check check = { 0 };
1019 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
1020 	u64 generation;
1021 	int ret;
1022 	int level;
1023 
1024 	root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1025 	if (!root)
1026 		return ERR_PTR(-ENOMEM);
1027 
1028 	ret = btrfs_find_root(tree_root, key, path,
1029 			      &root->root_item, &root->root_key);
1030 	if (ret) {
1031 		if (ret > 0)
1032 			ret = -ENOENT;
1033 		goto fail;
1034 	}
1035 
1036 	generation = btrfs_root_generation(&root->root_item);
1037 	level = btrfs_root_level(&root->root_item);
1038 	check.level = level;
1039 	check.transid = generation;
1040 	check.owner_root = key->objectid;
1041 	root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item),
1042 				     &check);
1043 	if (IS_ERR(root->node)) {
1044 		ret = PTR_ERR(root->node);
1045 		root->node = NULL;
1046 		goto fail;
1047 	}
1048 	if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1049 		ret = -EIO;
1050 		goto fail;
1051 	}
1052 
1053 	/*
1054 	 * For real fs, and not log/reloc trees, root owner must
1055 	 * match its root node owner
1056 	 */
1057 	if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) &&
1058 	    root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1059 	    root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1060 	    root->root_key.objectid != btrfs_header_owner(root->node)) {
1061 		btrfs_crit(fs_info,
1062 "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1063 			   root->root_key.objectid, root->node->start,
1064 			   btrfs_header_owner(root->node),
1065 			   root->root_key.objectid);
1066 		ret = -EUCLEAN;
1067 		goto fail;
1068 	}
1069 	root->commit_root = btrfs_root_node(root);
1070 	return root;
1071 fail:
1072 	btrfs_put_root(root);
1073 	return ERR_PTR(ret);
1074 }
1075 
btrfs_read_tree_root(struct btrfs_root * tree_root,struct btrfs_key * key)1076 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1077 					struct btrfs_key *key)
1078 {
1079 	struct btrfs_root *root;
1080 	struct btrfs_path *path;
1081 
1082 	path = btrfs_alloc_path();
1083 	if (!path)
1084 		return ERR_PTR(-ENOMEM);
1085 	root = read_tree_root_path(tree_root, path, key);
1086 	btrfs_free_path(path);
1087 
1088 	return root;
1089 }
1090 
1091 /*
1092  * Initialize subvolume root in-memory structure
1093  *
1094  * @anon_dev:	anonymous device to attach to the root, if zero, allocate new
1095  */
btrfs_init_fs_root(struct btrfs_root * root,dev_t anon_dev)1096 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1097 {
1098 	int ret;
1099 
1100 	btrfs_drew_lock_init(&root->snapshot_lock);
1101 
1102 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1103 	    !btrfs_is_data_reloc_root(root) &&
1104 	    is_fstree(root->root_key.objectid)) {
1105 		set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1106 		btrfs_check_and_init_root_item(&root->root_item);
1107 	}
1108 
1109 	/*
1110 	 * Don't assign anonymous block device to roots that are not exposed to
1111 	 * userspace, the id pool is limited to 1M
1112 	 */
1113 	if (is_fstree(root->root_key.objectid) &&
1114 	    btrfs_root_refs(&root->root_item) > 0) {
1115 		if (!anon_dev) {
1116 			ret = get_anon_bdev(&root->anon_dev);
1117 			if (ret)
1118 				goto fail;
1119 		} else {
1120 			root->anon_dev = anon_dev;
1121 		}
1122 	}
1123 
1124 	mutex_lock(&root->objectid_mutex);
1125 	ret = btrfs_init_root_free_objectid(root);
1126 	if (ret) {
1127 		mutex_unlock(&root->objectid_mutex);
1128 		goto fail;
1129 	}
1130 
1131 	ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1132 
1133 	mutex_unlock(&root->objectid_mutex);
1134 
1135 	return 0;
1136 fail:
1137 	/* The caller is responsible to call btrfs_free_fs_root */
1138 	return ret;
1139 }
1140 
btrfs_lookup_fs_root(struct btrfs_fs_info * fs_info,u64 root_id)1141 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1142 					       u64 root_id)
1143 {
1144 	struct btrfs_root *root;
1145 
1146 	spin_lock(&fs_info->fs_roots_radix_lock);
1147 	root = radix_tree_lookup(&fs_info->fs_roots_radix,
1148 				 (unsigned long)root_id);
1149 	root = btrfs_grab_root(root);
1150 	spin_unlock(&fs_info->fs_roots_radix_lock);
1151 	return root;
1152 }
1153 
btrfs_get_global_root(struct btrfs_fs_info * fs_info,u64 objectid)1154 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1155 						u64 objectid)
1156 {
1157 	struct btrfs_key key = {
1158 		.objectid = objectid,
1159 		.type = BTRFS_ROOT_ITEM_KEY,
1160 		.offset = 0,
1161 	};
1162 
1163 	switch (objectid) {
1164 	case BTRFS_ROOT_TREE_OBJECTID:
1165 		return btrfs_grab_root(fs_info->tree_root);
1166 	case BTRFS_EXTENT_TREE_OBJECTID:
1167 		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1168 	case BTRFS_CHUNK_TREE_OBJECTID:
1169 		return btrfs_grab_root(fs_info->chunk_root);
1170 	case BTRFS_DEV_TREE_OBJECTID:
1171 		return btrfs_grab_root(fs_info->dev_root);
1172 	case BTRFS_CSUM_TREE_OBJECTID:
1173 		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1174 	case BTRFS_QUOTA_TREE_OBJECTID:
1175 		return btrfs_grab_root(fs_info->quota_root);
1176 	case BTRFS_UUID_TREE_OBJECTID:
1177 		return btrfs_grab_root(fs_info->uuid_root);
1178 	case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
1179 		return btrfs_grab_root(fs_info->block_group_root);
1180 	case BTRFS_FREE_SPACE_TREE_OBJECTID:
1181 		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1182 	default:
1183 		return NULL;
1184 	}
1185 }
1186 
btrfs_insert_fs_root(struct btrfs_fs_info * fs_info,struct btrfs_root * root)1187 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1188 			 struct btrfs_root *root)
1189 {
1190 	int ret;
1191 
1192 	ret = radix_tree_preload(GFP_NOFS);
1193 	if (ret)
1194 		return ret;
1195 
1196 	spin_lock(&fs_info->fs_roots_radix_lock);
1197 	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1198 				(unsigned long)root->root_key.objectid,
1199 				root);
1200 	if (ret == 0) {
1201 		btrfs_grab_root(root);
1202 		set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1203 	}
1204 	spin_unlock(&fs_info->fs_roots_radix_lock);
1205 	radix_tree_preload_end();
1206 
1207 	return ret;
1208 }
1209 
btrfs_check_leaked_roots(struct btrfs_fs_info * fs_info)1210 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1211 {
1212 #ifdef CONFIG_BTRFS_DEBUG
1213 	struct btrfs_root *root;
1214 
1215 	while (!list_empty(&fs_info->allocated_roots)) {
1216 		char buf[BTRFS_ROOT_NAME_BUF_LEN];
1217 
1218 		root = list_first_entry(&fs_info->allocated_roots,
1219 					struct btrfs_root, leak_list);
1220 		btrfs_err(fs_info, "leaked root %s refcount %d",
1221 			  btrfs_root_name(&root->root_key, buf),
1222 			  refcount_read(&root->refs));
1223 		while (refcount_read(&root->refs) > 1)
1224 			btrfs_put_root(root);
1225 		btrfs_put_root(root);
1226 	}
1227 #endif
1228 }
1229 
free_global_roots(struct btrfs_fs_info * fs_info)1230 static void free_global_roots(struct btrfs_fs_info *fs_info)
1231 {
1232 	struct btrfs_root *root;
1233 	struct rb_node *node;
1234 
1235 	while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1236 		root = rb_entry(node, struct btrfs_root, rb_node);
1237 		rb_erase(&root->rb_node, &fs_info->global_root_tree);
1238 		btrfs_put_root(root);
1239 	}
1240 }
1241 
btrfs_free_fs_info(struct btrfs_fs_info * fs_info)1242 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1243 {
1244 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1245 	percpu_counter_destroy(&fs_info->delalloc_bytes);
1246 	percpu_counter_destroy(&fs_info->ordered_bytes);
1247 	percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1248 	btrfs_free_csum_hash(fs_info);
1249 	btrfs_free_stripe_hash_table(fs_info);
1250 	btrfs_free_ref_cache(fs_info);
1251 	kfree(fs_info->balance_ctl);
1252 	kfree(fs_info->delayed_root);
1253 	free_global_roots(fs_info);
1254 	btrfs_put_root(fs_info->tree_root);
1255 	btrfs_put_root(fs_info->chunk_root);
1256 	btrfs_put_root(fs_info->dev_root);
1257 	btrfs_put_root(fs_info->quota_root);
1258 	btrfs_put_root(fs_info->uuid_root);
1259 	btrfs_put_root(fs_info->fs_root);
1260 	btrfs_put_root(fs_info->data_reloc_root);
1261 	btrfs_put_root(fs_info->block_group_root);
1262 	btrfs_check_leaked_roots(fs_info);
1263 	btrfs_extent_buffer_leak_debug_check(fs_info);
1264 	kfree(fs_info->super_copy);
1265 	kfree(fs_info->super_for_commit);
1266 	kfree(fs_info->subpage_info);
1267 	kvfree(fs_info);
1268 }
1269 
1270 
1271 /*
1272  * Get an in-memory reference of a root structure.
1273  *
1274  * For essential trees like root/extent tree, we grab it from fs_info directly.
1275  * For subvolume trees, we check the cached filesystem roots first. If not
1276  * found, then read it from disk and add it to cached fs roots.
1277  *
1278  * Caller should release the root by calling btrfs_put_root() after the usage.
1279  *
1280  * NOTE: Reloc and log trees can't be read by this function as they share the
1281  *	 same root objectid.
1282  *
1283  * @objectid:	root id
1284  * @anon_dev:	preallocated anonymous block device number for new roots,
1285  *		pass NULL for a new allocation.
1286  * @check_ref:	whether to check root item references, If true, return -ENOENT
1287  *		for orphan roots
1288  */
btrfs_get_root_ref(struct btrfs_fs_info * fs_info,u64 objectid,dev_t * anon_dev,bool check_ref)1289 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1290 					     u64 objectid, dev_t *anon_dev,
1291 					     bool check_ref)
1292 {
1293 	struct btrfs_root *root;
1294 	struct btrfs_path *path;
1295 	struct btrfs_key key;
1296 	int ret;
1297 
1298 	root = btrfs_get_global_root(fs_info, objectid);
1299 	if (root)
1300 		return root;
1301 
1302 	/*
1303 	 * If we're called for non-subvolume trees, and above function didn't
1304 	 * find one, do not try to read it from disk.
1305 	 *
1306 	 * This is namely for free-space-tree and quota tree, which can change
1307 	 * at runtime and should only be grabbed from fs_info.
1308 	 */
1309 	if (!is_fstree(objectid) && objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1310 		return ERR_PTR(-ENOENT);
1311 again:
1312 	root = btrfs_lookup_fs_root(fs_info, objectid);
1313 	if (root) {
1314 		/*
1315 		 * Some other caller may have read out the newly inserted
1316 		 * subvolume already (for things like backref walk etc).  Not
1317 		 * that common but still possible.  In that case, we just need
1318 		 * to free the anon_dev.
1319 		 */
1320 		if (unlikely(anon_dev && *anon_dev)) {
1321 			free_anon_bdev(*anon_dev);
1322 			*anon_dev = 0;
1323 		}
1324 
1325 		if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1326 			btrfs_put_root(root);
1327 			return ERR_PTR(-ENOENT);
1328 		}
1329 		return root;
1330 	}
1331 
1332 	key.objectid = objectid;
1333 	key.type = BTRFS_ROOT_ITEM_KEY;
1334 	key.offset = (u64)-1;
1335 	root = btrfs_read_tree_root(fs_info->tree_root, &key);
1336 	if (IS_ERR(root))
1337 		return root;
1338 
1339 	if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1340 		ret = -ENOENT;
1341 		goto fail;
1342 	}
1343 
1344 	ret = btrfs_init_fs_root(root, anon_dev ? *anon_dev : 0);
1345 	if (ret)
1346 		goto fail;
1347 
1348 	path = btrfs_alloc_path();
1349 	if (!path) {
1350 		ret = -ENOMEM;
1351 		goto fail;
1352 	}
1353 	key.objectid = BTRFS_ORPHAN_OBJECTID;
1354 	key.type = BTRFS_ORPHAN_ITEM_KEY;
1355 	key.offset = objectid;
1356 
1357 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1358 	btrfs_free_path(path);
1359 	if (ret < 0)
1360 		goto fail;
1361 	if (ret == 0)
1362 		set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1363 
1364 	ret = btrfs_insert_fs_root(fs_info, root);
1365 	if (ret) {
1366 		if (ret == -EEXIST) {
1367 			btrfs_put_root(root);
1368 			goto again;
1369 		}
1370 		goto fail;
1371 	}
1372 	return root;
1373 fail:
1374 	/*
1375 	 * If our caller provided us an anonymous device, then it's his
1376 	 * responsibility to free it in case we fail. So we have to set our
1377 	 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1378 	 * and once again by our caller.
1379 	 */
1380 	if (anon_dev && *anon_dev)
1381 		root->anon_dev = 0;
1382 	btrfs_put_root(root);
1383 	return ERR_PTR(ret);
1384 }
1385 
1386 /*
1387  * Get in-memory reference of a root structure
1388  *
1389  * @objectid:	tree objectid
1390  * @check_ref:	if set, verify that the tree exists and the item has at least
1391  *		one reference
1392  */
btrfs_get_fs_root(struct btrfs_fs_info * fs_info,u64 objectid,bool check_ref)1393 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1394 				     u64 objectid, bool check_ref)
1395 {
1396 	return btrfs_get_root_ref(fs_info, objectid, NULL, check_ref);
1397 }
1398 
1399 /*
1400  * Get in-memory reference of a root structure, created as new, optionally pass
1401  * the anonymous block device id
1402  *
1403  * @objectid:	tree objectid
1404  * @anon_dev:	if NULL, allocate a new anonymous block device or use the
1405  *		parameter value if not NULL
1406  */
btrfs_get_new_fs_root(struct btrfs_fs_info * fs_info,u64 objectid,dev_t * anon_dev)1407 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1408 					 u64 objectid, dev_t *anon_dev)
1409 {
1410 	return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1411 }
1412 
1413 /*
1414  * btrfs_get_fs_root_commit_root - return a root for the given objectid
1415  * @fs_info:	the fs_info
1416  * @objectid:	the objectid we need to lookup
1417  *
1418  * This is exclusively used for backref walking, and exists specifically because
1419  * of how qgroups does lookups.  Qgroups will do a backref lookup at delayed ref
1420  * creation time, which means we may have to read the tree_root in order to look
1421  * up a fs root that is not in memory.  If the root is not in memory we will
1422  * read the tree root commit root and look up the fs root from there.  This is a
1423  * temporary root, it will not be inserted into the radix tree as it doesn't
1424  * have the most uptodate information, it'll simply be discarded once the
1425  * backref code is finished using the root.
1426  */
btrfs_get_fs_root_commit_root(struct btrfs_fs_info * fs_info,struct btrfs_path * path,u64 objectid)1427 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1428 						 struct btrfs_path *path,
1429 						 u64 objectid)
1430 {
1431 	struct btrfs_root *root;
1432 	struct btrfs_key key;
1433 
1434 	ASSERT(path->search_commit_root && path->skip_locking);
1435 
1436 	/*
1437 	 * This can return -ENOENT if we ask for a root that doesn't exist, but
1438 	 * since this is called via the backref walking code we won't be looking
1439 	 * up a root that doesn't exist, unless there's corruption.  So if root
1440 	 * != NULL just return it.
1441 	 */
1442 	root = btrfs_get_global_root(fs_info, objectid);
1443 	if (root)
1444 		return root;
1445 
1446 	root = btrfs_lookup_fs_root(fs_info, objectid);
1447 	if (root)
1448 		return root;
1449 
1450 	key.objectid = objectid;
1451 	key.type = BTRFS_ROOT_ITEM_KEY;
1452 	key.offset = (u64)-1;
1453 	root = read_tree_root_path(fs_info->tree_root, path, &key);
1454 	btrfs_release_path(path);
1455 
1456 	return root;
1457 }
1458 
cleaner_kthread(void * arg)1459 static int cleaner_kthread(void *arg)
1460 {
1461 	struct btrfs_fs_info *fs_info = arg;
1462 	int again;
1463 
1464 	while (1) {
1465 		again = 0;
1466 
1467 		set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1468 
1469 		/* Make the cleaner go to sleep early. */
1470 		if (btrfs_need_cleaner_sleep(fs_info))
1471 			goto sleep;
1472 
1473 		/*
1474 		 * Do not do anything if we might cause open_ctree() to block
1475 		 * before we have finished mounting the filesystem.
1476 		 */
1477 		if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1478 			goto sleep;
1479 
1480 		if (!mutex_trylock(&fs_info->cleaner_mutex))
1481 			goto sleep;
1482 
1483 		/*
1484 		 * Avoid the problem that we change the status of the fs
1485 		 * during the above check and trylock.
1486 		 */
1487 		if (btrfs_need_cleaner_sleep(fs_info)) {
1488 			mutex_unlock(&fs_info->cleaner_mutex);
1489 			goto sleep;
1490 		}
1491 
1492 		if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags))
1493 			btrfs_sysfs_feature_update(fs_info);
1494 
1495 		btrfs_run_delayed_iputs(fs_info);
1496 
1497 		again = btrfs_clean_one_deleted_snapshot(fs_info);
1498 		mutex_unlock(&fs_info->cleaner_mutex);
1499 
1500 		/*
1501 		 * The defragger has dealt with the R/O remount and umount,
1502 		 * needn't do anything special here.
1503 		 */
1504 		btrfs_run_defrag_inodes(fs_info);
1505 
1506 		/*
1507 		 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1508 		 * with relocation (btrfs_relocate_chunk) and relocation
1509 		 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1510 		 * after acquiring fs_info->reclaim_bgs_lock. So we
1511 		 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1512 		 * unused block groups.
1513 		 */
1514 		btrfs_delete_unused_bgs(fs_info);
1515 
1516 		/*
1517 		 * Reclaim block groups in the reclaim_bgs list after we deleted
1518 		 * all unused block_groups. This possibly gives us some more free
1519 		 * space.
1520 		 */
1521 		btrfs_reclaim_bgs(fs_info);
1522 sleep:
1523 		clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1524 		if (kthread_should_park())
1525 			kthread_parkme();
1526 		if (kthread_should_stop())
1527 			return 0;
1528 		if (!again) {
1529 			set_current_state(TASK_INTERRUPTIBLE);
1530 			schedule();
1531 			__set_current_state(TASK_RUNNING);
1532 		}
1533 	}
1534 }
1535 
transaction_kthread(void * arg)1536 static int transaction_kthread(void *arg)
1537 {
1538 	struct btrfs_root *root = arg;
1539 	struct btrfs_fs_info *fs_info = root->fs_info;
1540 	struct btrfs_trans_handle *trans;
1541 	struct btrfs_transaction *cur;
1542 	u64 transid;
1543 	time64_t delta;
1544 	unsigned long delay;
1545 	bool cannot_commit;
1546 
1547 	do {
1548 		cannot_commit = false;
1549 		delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1550 		mutex_lock(&fs_info->transaction_kthread_mutex);
1551 
1552 		spin_lock(&fs_info->trans_lock);
1553 		cur = fs_info->running_transaction;
1554 		if (!cur) {
1555 			spin_unlock(&fs_info->trans_lock);
1556 			goto sleep;
1557 		}
1558 
1559 		delta = ktime_get_seconds() - cur->start_time;
1560 		if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1561 		    cur->state < TRANS_STATE_COMMIT_PREP &&
1562 		    delta < fs_info->commit_interval) {
1563 			spin_unlock(&fs_info->trans_lock);
1564 			delay -= msecs_to_jiffies((delta - 1) * 1000);
1565 			delay = min(delay,
1566 				    msecs_to_jiffies(fs_info->commit_interval * 1000));
1567 			goto sleep;
1568 		}
1569 		transid = cur->transid;
1570 		spin_unlock(&fs_info->trans_lock);
1571 
1572 		/* If the file system is aborted, this will always fail. */
1573 		trans = btrfs_attach_transaction(root);
1574 		if (IS_ERR(trans)) {
1575 			if (PTR_ERR(trans) != -ENOENT)
1576 				cannot_commit = true;
1577 			goto sleep;
1578 		}
1579 		if (transid == trans->transid) {
1580 			btrfs_commit_transaction(trans);
1581 		} else {
1582 			btrfs_end_transaction(trans);
1583 		}
1584 sleep:
1585 		wake_up_process(fs_info->cleaner_kthread);
1586 		mutex_unlock(&fs_info->transaction_kthread_mutex);
1587 
1588 		if (BTRFS_FS_ERROR(fs_info))
1589 			btrfs_cleanup_transaction(fs_info);
1590 		if (!kthread_should_stop() &&
1591 				(!btrfs_transaction_blocked(fs_info) ||
1592 				 cannot_commit))
1593 			schedule_timeout_interruptible(delay);
1594 	} while (!kthread_should_stop());
1595 	return 0;
1596 }
1597 
1598 /*
1599  * This will find the highest generation in the array of root backups.  The
1600  * index of the highest array is returned, or -EINVAL if we can't find
1601  * anything.
1602  *
1603  * We check to make sure the array is valid by comparing the
1604  * generation of the latest  root in the array with the generation
1605  * in the super block.  If they don't match we pitch it.
1606  */
find_newest_super_backup(struct btrfs_fs_info * info)1607 static int find_newest_super_backup(struct btrfs_fs_info *info)
1608 {
1609 	const u64 newest_gen = btrfs_super_generation(info->super_copy);
1610 	u64 cur;
1611 	struct btrfs_root_backup *root_backup;
1612 	int i;
1613 
1614 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1615 		root_backup = info->super_copy->super_roots + i;
1616 		cur = btrfs_backup_tree_root_gen(root_backup);
1617 		if (cur == newest_gen)
1618 			return i;
1619 	}
1620 
1621 	return -EINVAL;
1622 }
1623 
1624 /*
1625  * copy all the root pointers into the super backup array.
1626  * this will bump the backup pointer by one when it is
1627  * done
1628  */
backup_super_roots(struct btrfs_fs_info * info)1629 static void backup_super_roots(struct btrfs_fs_info *info)
1630 {
1631 	const int next_backup = info->backup_root_index;
1632 	struct btrfs_root_backup *root_backup;
1633 
1634 	root_backup = info->super_for_commit->super_roots + next_backup;
1635 
1636 	/*
1637 	 * make sure all of our padding and empty slots get zero filled
1638 	 * regardless of which ones we use today
1639 	 */
1640 	memset(root_backup, 0, sizeof(*root_backup));
1641 
1642 	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1643 
1644 	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1645 	btrfs_set_backup_tree_root_gen(root_backup,
1646 			       btrfs_header_generation(info->tree_root->node));
1647 
1648 	btrfs_set_backup_tree_root_level(root_backup,
1649 			       btrfs_header_level(info->tree_root->node));
1650 
1651 	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1652 	btrfs_set_backup_chunk_root_gen(root_backup,
1653 			       btrfs_header_generation(info->chunk_root->node));
1654 	btrfs_set_backup_chunk_root_level(root_backup,
1655 			       btrfs_header_level(info->chunk_root->node));
1656 
1657 	if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
1658 		struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
1659 		struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
1660 
1661 		btrfs_set_backup_extent_root(root_backup,
1662 					     extent_root->node->start);
1663 		btrfs_set_backup_extent_root_gen(root_backup,
1664 				btrfs_header_generation(extent_root->node));
1665 		btrfs_set_backup_extent_root_level(root_backup,
1666 					btrfs_header_level(extent_root->node));
1667 
1668 		btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
1669 		btrfs_set_backup_csum_root_gen(root_backup,
1670 					       btrfs_header_generation(csum_root->node));
1671 		btrfs_set_backup_csum_root_level(root_backup,
1672 						 btrfs_header_level(csum_root->node));
1673 	}
1674 
1675 	/*
1676 	 * we might commit during log recovery, which happens before we set
1677 	 * the fs_root.  Make sure it is valid before we fill it in.
1678 	 */
1679 	if (info->fs_root && info->fs_root->node) {
1680 		btrfs_set_backup_fs_root(root_backup,
1681 					 info->fs_root->node->start);
1682 		btrfs_set_backup_fs_root_gen(root_backup,
1683 			       btrfs_header_generation(info->fs_root->node));
1684 		btrfs_set_backup_fs_root_level(root_backup,
1685 			       btrfs_header_level(info->fs_root->node));
1686 	}
1687 
1688 	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1689 	btrfs_set_backup_dev_root_gen(root_backup,
1690 			       btrfs_header_generation(info->dev_root->node));
1691 	btrfs_set_backup_dev_root_level(root_backup,
1692 				       btrfs_header_level(info->dev_root->node));
1693 
1694 	btrfs_set_backup_total_bytes(root_backup,
1695 			     btrfs_super_total_bytes(info->super_copy));
1696 	btrfs_set_backup_bytes_used(root_backup,
1697 			     btrfs_super_bytes_used(info->super_copy));
1698 	btrfs_set_backup_num_devices(root_backup,
1699 			     btrfs_super_num_devices(info->super_copy));
1700 
1701 	/*
1702 	 * if we don't copy this out to the super_copy, it won't get remembered
1703 	 * for the next commit
1704 	 */
1705 	memcpy(&info->super_copy->super_roots,
1706 	       &info->super_for_commit->super_roots,
1707 	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1708 }
1709 
1710 /*
1711  * read_backup_root - Reads a backup root based on the passed priority. Prio 0
1712  * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1713  *
1714  * fs_info - filesystem whose backup roots need to be read
1715  * priority - priority of backup root required
1716  *
1717  * Returns backup root index on success and -EINVAL otherwise.
1718  */
read_backup_root(struct btrfs_fs_info * fs_info,u8 priority)1719 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1720 {
1721 	int backup_index = find_newest_super_backup(fs_info);
1722 	struct btrfs_super_block *super = fs_info->super_copy;
1723 	struct btrfs_root_backup *root_backup;
1724 
1725 	if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1726 		if (priority == 0)
1727 			return backup_index;
1728 
1729 		backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1730 		backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1731 	} else {
1732 		return -EINVAL;
1733 	}
1734 
1735 	root_backup = super->super_roots + backup_index;
1736 
1737 	btrfs_set_super_generation(super,
1738 				   btrfs_backup_tree_root_gen(root_backup));
1739 	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1740 	btrfs_set_super_root_level(super,
1741 				   btrfs_backup_tree_root_level(root_backup));
1742 	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1743 
1744 	/*
1745 	 * Fixme: the total bytes and num_devices need to match or we should
1746 	 * need a fsck
1747 	 */
1748 	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1749 	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1750 
1751 	return backup_index;
1752 }
1753 
1754 /* helper to cleanup workers */
btrfs_stop_all_workers(struct btrfs_fs_info * fs_info)1755 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1756 {
1757 	btrfs_destroy_workqueue(fs_info->fixup_workers);
1758 	btrfs_destroy_workqueue(fs_info->delalloc_workers);
1759 	btrfs_destroy_workqueue(fs_info->workers);
1760 	if (fs_info->endio_workers)
1761 		destroy_workqueue(fs_info->endio_workers);
1762 	if (fs_info->rmw_workers)
1763 		destroy_workqueue(fs_info->rmw_workers);
1764 	if (fs_info->compressed_write_workers)
1765 		destroy_workqueue(fs_info->compressed_write_workers);
1766 	btrfs_destroy_workqueue(fs_info->endio_write_workers);
1767 	btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1768 	btrfs_destroy_workqueue(fs_info->delayed_workers);
1769 	btrfs_destroy_workqueue(fs_info->caching_workers);
1770 	btrfs_destroy_workqueue(fs_info->flush_workers);
1771 	btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1772 	if (fs_info->discard_ctl.discard_workers)
1773 		destroy_workqueue(fs_info->discard_ctl.discard_workers);
1774 	/*
1775 	 * Now that all other work queues are destroyed, we can safely destroy
1776 	 * the queues used for metadata I/O, since tasks from those other work
1777 	 * queues can do metadata I/O operations.
1778 	 */
1779 	if (fs_info->endio_meta_workers)
1780 		destroy_workqueue(fs_info->endio_meta_workers);
1781 }
1782 
free_root_extent_buffers(struct btrfs_root * root)1783 static void free_root_extent_buffers(struct btrfs_root *root)
1784 {
1785 	if (root) {
1786 		free_extent_buffer(root->node);
1787 		free_extent_buffer(root->commit_root);
1788 		root->node = NULL;
1789 		root->commit_root = NULL;
1790 	}
1791 }
1792 
free_global_root_pointers(struct btrfs_fs_info * fs_info)1793 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
1794 {
1795 	struct btrfs_root *root, *tmp;
1796 
1797 	rbtree_postorder_for_each_entry_safe(root, tmp,
1798 					     &fs_info->global_root_tree,
1799 					     rb_node)
1800 		free_root_extent_buffers(root);
1801 }
1802 
1803 /* helper to cleanup tree roots */
free_root_pointers(struct btrfs_fs_info * info,bool free_chunk_root)1804 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1805 {
1806 	free_root_extent_buffers(info->tree_root);
1807 
1808 	free_global_root_pointers(info);
1809 	free_root_extent_buffers(info->dev_root);
1810 	free_root_extent_buffers(info->quota_root);
1811 	free_root_extent_buffers(info->uuid_root);
1812 	free_root_extent_buffers(info->fs_root);
1813 	free_root_extent_buffers(info->data_reloc_root);
1814 	free_root_extent_buffers(info->block_group_root);
1815 	if (free_chunk_root)
1816 		free_root_extent_buffers(info->chunk_root);
1817 }
1818 
btrfs_put_root(struct btrfs_root * root)1819 void btrfs_put_root(struct btrfs_root *root)
1820 {
1821 	if (!root)
1822 		return;
1823 
1824 	if (refcount_dec_and_test(&root->refs)) {
1825 		WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1826 		WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
1827 		if (root->anon_dev)
1828 			free_anon_bdev(root->anon_dev);
1829 		free_root_extent_buffers(root);
1830 #ifdef CONFIG_BTRFS_DEBUG
1831 		spin_lock(&root->fs_info->fs_roots_radix_lock);
1832 		list_del_init(&root->leak_list);
1833 		spin_unlock(&root->fs_info->fs_roots_radix_lock);
1834 #endif
1835 		kfree(root);
1836 	}
1837 }
1838 
btrfs_free_fs_roots(struct btrfs_fs_info * fs_info)1839 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
1840 {
1841 	int ret;
1842 	struct btrfs_root *gang[8];
1843 	int i;
1844 
1845 	while (!list_empty(&fs_info->dead_roots)) {
1846 		gang[0] = list_entry(fs_info->dead_roots.next,
1847 				     struct btrfs_root, root_list);
1848 		list_del(&gang[0]->root_list);
1849 
1850 		if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
1851 			btrfs_drop_and_free_fs_root(fs_info, gang[0]);
1852 		btrfs_put_root(gang[0]);
1853 	}
1854 
1855 	while (1) {
1856 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1857 					     (void **)gang, 0,
1858 					     ARRAY_SIZE(gang));
1859 		if (!ret)
1860 			break;
1861 		for (i = 0; i < ret; i++)
1862 			btrfs_drop_and_free_fs_root(fs_info, gang[i]);
1863 	}
1864 }
1865 
btrfs_init_scrub(struct btrfs_fs_info * fs_info)1866 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
1867 {
1868 	mutex_init(&fs_info->scrub_lock);
1869 	atomic_set(&fs_info->scrubs_running, 0);
1870 	atomic_set(&fs_info->scrub_pause_req, 0);
1871 	atomic_set(&fs_info->scrubs_paused, 0);
1872 	atomic_set(&fs_info->scrub_cancel_req, 0);
1873 	init_waitqueue_head(&fs_info->scrub_pause_wait);
1874 	refcount_set(&fs_info->scrub_workers_refcnt, 0);
1875 }
1876 
btrfs_init_balance(struct btrfs_fs_info * fs_info)1877 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
1878 {
1879 	spin_lock_init(&fs_info->balance_lock);
1880 	mutex_init(&fs_info->balance_mutex);
1881 	atomic_set(&fs_info->balance_pause_req, 0);
1882 	atomic_set(&fs_info->balance_cancel_req, 0);
1883 	fs_info->balance_ctl = NULL;
1884 	init_waitqueue_head(&fs_info->balance_wait_q);
1885 	atomic_set(&fs_info->reloc_cancel_req, 0);
1886 }
1887 
btrfs_init_btree_inode(struct super_block * sb)1888 static int btrfs_init_btree_inode(struct super_block *sb)
1889 {
1890 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1891 	unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
1892 					      fs_info->tree_root);
1893 	struct inode *inode;
1894 
1895 	inode = new_inode(sb);
1896 	if (!inode)
1897 		return -ENOMEM;
1898 
1899 	inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1900 	set_nlink(inode, 1);
1901 	/*
1902 	 * we set the i_size on the btree inode to the max possible int.
1903 	 * the real end of the address space is determined by all of
1904 	 * the devices in the system
1905 	 */
1906 	inode->i_size = OFFSET_MAX;
1907 	inode->i_mapping->a_ops = &btree_aops;
1908 	mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
1909 
1910 	RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
1911 	extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
1912 			    IO_TREE_BTREE_INODE_IO);
1913 	extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
1914 
1915 	BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
1916 	BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID;
1917 	BTRFS_I(inode)->location.type = 0;
1918 	BTRFS_I(inode)->location.offset = 0;
1919 	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
1920 	__insert_inode_hash(inode, hash);
1921 	fs_info->btree_inode = inode;
1922 
1923 	return 0;
1924 }
1925 
btrfs_init_dev_replace_locks(struct btrfs_fs_info * fs_info)1926 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
1927 {
1928 	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
1929 	init_rwsem(&fs_info->dev_replace.rwsem);
1930 	init_waitqueue_head(&fs_info->dev_replace.replace_wait);
1931 }
1932 
btrfs_init_qgroup(struct btrfs_fs_info * fs_info)1933 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
1934 {
1935 	spin_lock_init(&fs_info->qgroup_lock);
1936 	mutex_init(&fs_info->qgroup_ioctl_lock);
1937 	fs_info->qgroup_tree = RB_ROOT;
1938 	INIT_LIST_HEAD(&fs_info->dirty_qgroups);
1939 	fs_info->qgroup_seq = 1;
1940 	fs_info->qgroup_ulist = NULL;
1941 	fs_info->qgroup_rescan_running = false;
1942 	fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
1943 	mutex_init(&fs_info->qgroup_rescan_lock);
1944 }
1945 
btrfs_init_workqueues(struct btrfs_fs_info * fs_info)1946 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
1947 {
1948 	u32 max_active = fs_info->thread_pool_size;
1949 	unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
1950 	unsigned int ordered_flags = WQ_MEM_RECLAIM | WQ_FREEZABLE;
1951 
1952 	fs_info->workers =
1953 		btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
1954 
1955 	fs_info->delalloc_workers =
1956 		btrfs_alloc_workqueue(fs_info, "delalloc",
1957 				      flags, max_active, 2);
1958 
1959 	fs_info->flush_workers =
1960 		btrfs_alloc_workqueue(fs_info, "flush_delalloc",
1961 				      flags, max_active, 0);
1962 
1963 	fs_info->caching_workers =
1964 		btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
1965 
1966 	fs_info->fixup_workers =
1967 		btrfs_alloc_ordered_workqueue(fs_info, "fixup", ordered_flags);
1968 
1969 	fs_info->endio_workers =
1970 		alloc_workqueue("btrfs-endio", flags, max_active);
1971 	fs_info->endio_meta_workers =
1972 		alloc_workqueue("btrfs-endio-meta", flags, max_active);
1973 	fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
1974 	fs_info->endio_write_workers =
1975 		btrfs_alloc_workqueue(fs_info, "endio-write", flags,
1976 				      max_active, 2);
1977 	fs_info->compressed_write_workers =
1978 		alloc_workqueue("btrfs-compressed-write", flags, max_active);
1979 	fs_info->endio_freespace_worker =
1980 		btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
1981 				      max_active, 0);
1982 	fs_info->delayed_workers =
1983 		btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
1984 				      max_active, 0);
1985 	fs_info->qgroup_rescan_workers =
1986 		btrfs_alloc_ordered_workqueue(fs_info, "qgroup-rescan",
1987 					      ordered_flags);
1988 	fs_info->discard_ctl.discard_workers =
1989 		alloc_ordered_workqueue("btrfs_discard", WQ_FREEZABLE);
1990 
1991 	if (!(fs_info->workers &&
1992 	      fs_info->delalloc_workers && fs_info->flush_workers &&
1993 	      fs_info->endio_workers && fs_info->endio_meta_workers &&
1994 	      fs_info->compressed_write_workers &&
1995 	      fs_info->endio_write_workers &&
1996 	      fs_info->endio_freespace_worker && fs_info->rmw_workers &&
1997 	      fs_info->caching_workers && fs_info->fixup_workers &&
1998 	      fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
1999 	      fs_info->discard_ctl.discard_workers)) {
2000 		return -ENOMEM;
2001 	}
2002 
2003 	return 0;
2004 }
2005 
btrfs_init_csum_hash(struct btrfs_fs_info * fs_info,u16 csum_type)2006 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2007 {
2008 	struct crypto_shash *csum_shash;
2009 	const char *csum_driver = btrfs_super_csum_driver(csum_type);
2010 
2011 	csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2012 
2013 	if (IS_ERR(csum_shash)) {
2014 		btrfs_err(fs_info, "error allocating %s hash for checksum",
2015 			  csum_driver);
2016 		return PTR_ERR(csum_shash);
2017 	}
2018 
2019 	fs_info->csum_shash = csum_shash;
2020 
2021 	/*
2022 	 * Check if the checksum implementation is a fast accelerated one.
2023 	 * As-is this is a bit of a hack and should be replaced once the csum
2024 	 * implementations provide that information themselves.
2025 	 */
2026 	switch (csum_type) {
2027 	case BTRFS_CSUM_TYPE_CRC32:
2028 		if (!strstr(crypto_shash_driver_name(csum_shash), "generic"))
2029 			set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2030 		break;
2031 	case BTRFS_CSUM_TYPE_XXHASH:
2032 		set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2033 		break;
2034 	default:
2035 		break;
2036 	}
2037 
2038 	btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2039 			btrfs_super_csum_name(csum_type),
2040 			crypto_shash_driver_name(csum_shash));
2041 	return 0;
2042 }
2043 
btrfs_replay_log(struct btrfs_fs_info * fs_info,struct btrfs_fs_devices * fs_devices)2044 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2045 			    struct btrfs_fs_devices *fs_devices)
2046 {
2047 	int ret;
2048 	struct btrfs_tree_parent_check check = { 0 };
2049 	struct btrfs_root *log_tree_root;
2050 	struct btrfs_super_block *disk_super = fs_info->super_copy;
2051 	u64 bytenr = btrfs_super_log_root(disk_super);
2052 	int level = btrfs_super_log_root_level(disk_super);
2053 
2054 	if (fs_devices->rw_devices == 0) {
2055 		btrfs_warn(fs_info, "log replay required on RO media");
2056 		return -EIO;
2057 	}
2058 
2059 	log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2060 					 GFP_KERNEL);
2061 	if (!log_tree_root)
2062 		return -ENOMEM;
2063 
2064 	check.level = level;
2065 	check.transid = fs_info->generation + 1;
2066 	check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2067 	log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
2068 	if (IS_ERR(log_tree_root->node)) {
2069 		btrfs_warn(fs_info, "failed to read log tree");
2070 		ret = PTR_ERR(log_tree_root->node);
2071 		log_tree_root->node = NULL;
2072 		btrfs_put_root(log_tree_root);
2073 		return ret;
2074 	}
2075 	if (!extent_buffer_uptodate(log_tree_root->node)) {
2076 		btrfs_err(fs_info, "failed to read log tree");
2077 		btrfs_put_root(log_tree_root);
2078 		return -EIO;
2079 	}
2080 
2081 	/* returns with log_tree_root freed on success */
2082 	ret = btrfs_recover_log_trees(log_tree_root);
2083 	if (ret) {
2084 		btrfs_handle_fs_error(fs_info, ret,
2085 				      "Failed to recover log tree");
2086 		btrfs_put_root(log_tree_root);
2087 		return ret;
2088 	}
2089 
2090 	if (sb_rdonly(fs_info->sb)) {
2091 		ret = btrfs_commit_super(fs_info);
2092 		if (ret)
2093 			return ret;
2094 	}
2095 
2096 	return 0;
2097 }
2098 
load_global_roots_objectid(struct btrfs_root * tree_root,struct btrfs_path * path,u64 objectid,const char * name)2099 static int load_global_roots_objectid(struct btrfs_root *tree_root,
2100 				      struct btrfs_path *path, u64 objectid,
2101 				      const char *name)
2102 {
2103 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
2104 	struct btrfs_root *root;
2105 	u64 max_global_id = 0;
2106 	int ret;
2107 	struct btrfs_key key = {
2108 		.objectid = objectid,
2109 		.type = BTRFS_ROOT_ITEM_KEY,
2110 		.offset = 0,
2111 	};
2112 	bool found = false;
2113 
2114 	/* If we have IGNOREDATACSUMS skip loading these roots. */
2115 	if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2116 	    btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2117 		set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2118 		return 0;
2119 	}
2120 
2121 	while (1) {
2122 		ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2123 		if (ret < 0)
2124 			break;
2125 
2126 		if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2127 			ret = btrfs_next_leaf(tree_root, path);
2128 			if (ret) {
2129 				if (ret > 0)
2130 					ret = 0;
2131 				break;
2132 			}
2133 		}
2134 		ret = 0;
2135 
2136 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2137 		if (key.objectid != objectid)
2138 			break;
2139 		btrfs_release_path(path);
2140 
2141 		/*
2142 		 * Just worry about this for extent tree, it'll be the same for
2143 		 * everybody.
2144 		 */
2145 		if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2146 			max_global_id = max(max_global_id, key.offset);
2147 
2148 		found = true;
2149 		root = read_tree_root_path(tree_root, path, &key);
2150 		if (IS_ERR(root)) {
2151 			if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2152 				ret = PTR_ERR(root);
2153 			break;
2154 		}
2155 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2156 		ret = btrfs_global_root_insert(root);
2157 		if (ret) {
2158 			btrfs_put_root(root);
2159 			break;
2160 		}
2161 		key.offset++;
2162 	}
2163 	btrfs_release_path(path);
2164 
2165 	if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2166 		fs_info->nr_global_roots = max_global_id + 1;
2167 
2168 	if (!found || ret) {
2169 		if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2170 			set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2171 
2172 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2173 			ret = ret ? ret : -ENOENT;
2174 		else
2175 			ret = 0;
2176 		btrfs_err(fs_info, "failed to load root %s", name);
2177 	}
2178 	return ret;
2179 }
2180 
load_global_roots(struct btrfs_root * tree_root)2181 static int load_global_roots(struct btrfs_root *tree_root)
2182 {
2183 	struct btrfs_path *path;
2184 	int ret = 0;
2185 
2186 	path = btrfs_alloc_path();
2187 	if (!path)
2188 		return -ENOMEM;
2189 
2190 	ret = load_global_roots_objectid(tree_root, path,
2191 					 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2192 	if (ret)
2193 		goto out;
2194 	ret = load_global_roots_objectid(tree_root, path,
2195 					 BTRFS_CSUM_TREE_OBJECTID, "csum");
2196 	if (ret)
2197 		goto out;
2198 	if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2199 		goto out;
2200 	ret = load_global_roots_objectid(tree_root, path,
2201 					 BTRFS_FREE_SPACE_TREE_OBJECTID,
2202 					 "free space");
2203 out:
2204 	btrfs_free_path(path);
2205 	return ret;
2206 }
2207 
btrfs_read_roots(struct btrfs_fs_info * fs_info)2208 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2209 {
2210 	struct btrfs_root *tree_root = fs_info->tree_root;
2211 	struct btrfs_root *root;
2212 	struct btrfs_key location;
2213 	int ret;
2214 
2215 	BUG_ON(!fs_info->tree_root);
2216 
2217 	ret = load_global_roots(tree_root);
2218 	if (ret)
2219 		return ret;
2220 
2221 	location.type = BTRFS_ROOT_ITEM_KEY;
2222 	location.offset = 0;
2223 
2224 	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2225 		location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2226 		root = btrfs_read_tree_root(tree_root, &location);
2227 		if (IS_ERR(root)) {
2228 			if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2229 				ret = PTR_ERR(root);
2230 				goto out;
2231 			}
2232 		} else {
2233 			set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2234 			fs_info->block_group_root = root;
2235 		}
2236 	}
2237 
2238 	location.objectid = BTRFS_DEV_TREE_OBJECTID;
2239 	root = btrfs_read_tree_root(tree_root, &location);
2240 	if (IS_ERR(root)) {
2241 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2242 			ret = PTR_ERR(root);
2243 			goto out;
2244 		}
2245 	} else {
2246 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2247 		fs_info->dev_root = root;
2248 	}
2249 	/* Initialize fs_info for all devices in any case */
2250 	ret = btrfs_init_devices_late(fs_info);
2251 	if (ret)
2252 		goto out;
2253 
2254 	/*
2255 	 * This tree can share blocks with some other fs tree during relocation
2256 	 * and we need a proper setup by btrfs_get_fs_root
2257 	 */
2258 	root = btrfs_get_fs_root(tree_root->fs_info,
2259 				 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2260 	if (IS_ERR(root)) {
2261 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2262 			ret = PTR_ERR(root);
2263 			goto out;
2264 		}
2265 	} else {
2266 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2267 		fs_info->data_reloc_root = root;
2268 	}
2269 
2270 	location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2271 	root = btrfs_read_tree_root(tree_root, &location);
2272 	if (!IS_ERR(root)) {
2273 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2274 		set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2275 		fs_info->quota_root = root;
2276 	}
2277 
2278 	location.objectid = BTRFS_UUID_TREE_OBJECTID;
2279 	root = btrfs_read_tree_root(tree_root, &location);
2280 	if (IS_ERR(root)) {
2281 		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2282 			ret = PTR_ERR(root);
2283 			if (ret != -ENOENT)
2284 				goto out;
2285 		}
2286 	} else {
2287 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2288 		fs_info->uuid_root = root;
2289 	}
2290 
2291 	return 0;
2292 out:
2293 	btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2294 		   location.objectid, ret);
2295 	return ret;
2296 }
2297 
2298 /*
2299  * Real super block validation
2300  * NOTE: super csum type and incompat features will not be checked here.
2301  *
2302  * @sb:		super block to check
2303  * @mirror_num:	the super block number to check its bytenr:
2304  * 		0	the primary (1st) sb
2305  * 		1, 2	2nd and 3rd backup copy
2306  * 	       -1	skip bytenr check
2307  */
btrfs_validate_super(struct btrfs_fs_info * fs_info,struct btrfs_super_block * sb,int mirror_num)2308 int btrfs_validate_super(struct btrfs_fs_info *fs_info,
2309 			 struct btrfs_super_block *sb, int mirror_num)
2310 {
2311 	u64 nodesize = btrfs_super_nodesize(sb);
2312 	u64 sectorsize = btrfs_super_sectorsize(sb);
2313 	int ret = 0;
2314 
2315 	if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2316 		btrfs_err(fs_info, "no valid FS found");
2317 		ret = -EINVAL;
2318 	}
2319 	if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2320 		btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2321 				btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2322 		ret = -EINVAL;
2323 	}
2324 	if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2325 		btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2326 				btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2327 		ret = -EINVAL;
2328 	}
2329 	if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2330 		btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2331 				btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2332 		ret = -EINVAL;
2333 	}
2334 	if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2335 		btrfs_err(fs_info, "log_root level too big: %d >= %d",
2336 				btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2337 		ret = -EINVAL;
2338 	}
2339 
2340 	/*
2341 	 * Check sectorsize and nodesize first, other check will need it.
2342 	 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2343 	 */
2344 	if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2345 	    sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2346 		btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2347 		ret = -EINVAL;
2348 	}
2349 
2350 	/*
2351 	 * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2352 	 *
2353 	 * We can support 16K sectorsize with 64K page size without problem,
2354 	 * but such sectorsize/pagesize combination doesn't make much sense.
2355 	 * 4K will be our future standard, PAGE_SIZE is supported from the very
2356 	 * beginning.
2357 	 */
2358 	if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2359 		btrfs_err(fs_info,
2360 			"sectorsize %llu not yet supported for page size %lu",
2361 			sectorsize, PAGE_SIZE);
2362 		ret = -EINVAL;
2363 	}
2364 
2365 	if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2366 	    nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2367 		btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2368 		ret = -EINVAL;
2369 	}
2370 	if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2371 		btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2372 			  le32_to_cpu(sb->__unused_leafsize), nodesize);
2373 		ret = -EINVAL;
2374 	}
2375 
2376 	/* Root alignment check */
2377 	if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2378 		btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2379 			   btrfs_super_root(sb));
2380 		ret = -EINVAL;
2381 	}
2382 	if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2383 		btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2384 			   btrfs_super_chunk_root(sb));
2385 		ret = -EINVAL;
2386 	}
2387 	if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2388 		btrfs_warn(fs_info, "log_root block unaligned: %llu",
2389 			   btrfs_super_log_root(sb));
2390 		ret = -EINVAL;
2391 	}
2392 
2393 	if (memcmp(fs_info->fs_devices->fsid, sb->fsid, BTRFS_FSID_SIZE) != 0) {
2394 		btrfs_err(fs_info,
2395 		"superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2396 			  sb->fsid, fs_info->fs_devices->fsid);
2397 		ret = -EINVAL;
2398 	}
2399 
2400 	if (memcmp(fs_info->fs_devices->metadata_uuid, btrfs_sb_fsid_ptr(sb),
2401 		   BTRFS_FSID_SIZE) != 0) {
2402 		btrfs_err(fs_info,
2403 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2404 			  btrfs_sb_fsid_ptr(sb), fs_info->fs_devices->metadata_uuid);
2405 		ret = -EINVAL;
2406 	}
2407 
2408 	if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2409 		   BTRFS_FSID_SIZE) != 0) {
2410 		btrfs_err(fs_info,
2411 			"dev_item UUID does not match metadata fsid: %pU != %pU",
2412 			fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2413 		ret = -EINVAL;
2414 	}
2415 
2416 	/*
2417 	 * Artificial requirement for block-group-tree to force newer features
2418 	 * (free-space-tree, no-holes) so the test matrix is smaller.
2419 	 */
2420 	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2421 	    (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2422 	     !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2423 		btrfs_err(fs_info,
2424 		"block-group-tree feature requires fres-space-tree and no-holes");
2425 		ret = -EINVAL;
2426 	}
2427 
2428 	/*
2429 	 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2430 	 * done later
2431 	 */
2432 	if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2433 		btrfs_err(fs_info, "bytes_used is too small %llu",
2434 			  btrfs_super_bytes_used(sb));
2435 		ret = -EINVAL;
2436 	}
2437 	if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2438 		btrfs_err(fs_info, "invalid stripesize %u",
2439 			  btrfs_super_stripesize(sb));
2440 		ret = -EINVAL;
2441 	}
2442 	if (btrfs_super_num_devices(sb) > (1UL << 31))
2443 		btrfs_warn(fs_info, "suspicious number of devices: %llu",
2444 			   btrfs_super_num_devices(sb));
2445 	if (btrfs_super_num_devices(sb) == 0) {
2446 		btrfs_err(fs_info, "number of devices is 0");
2447 		ret = -EINVAL;
2448 	}
2449 
2450 	if (mirror_num >= 0 &&
2451 	    btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2452 		btrfs_err(fs_info, "super offset mismatch %llu != %u",
2453 			  btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2454 		ret = -EINVAL;
2455 	}
2456 
2457 	/*
2458 	 * Obvious sys_chunk_array corruptions, it must hold at least one key
2459 	 * and one chunk
2460 	 */
2461 	if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2462 		btrfs_err(fs_info, "system chunk array too big %u > %u",
2463 			  btrfs_super_sys_array_size(sb),
2464 			  BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2465 		ret = -EINVAL;
2466 	}
2467 	if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2468 			+ sizeof(struct btrfs_chunk)) {
2469 		btrfs_err(fs_info, "system chunk array too small %u < %zu",
2470 			  btrfs_super_sys_array_size(sb),
2471 			  sizeof(struct btrfs_disk_key)
2472 			  + sizeof(struct btrfs_chunk));
2473 		ret = -EINVAL;
2474 	}
2475 
2476 	/*
2477 	 * The generation is a global counter, we'll trust it more than the others
2478 	 * but it's still possible that it's the one that's wrong.
2479 	 */
2480 	if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2481 		btrfs_warn(fs_info,
2482 			"suspicious: generation < chunk_root_generation: %llu < %llu",
2483 			btrfs_super_generation(sb),
2484 			btrfs_super_chunk_root_generation(sb));
2485 	if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2486 	    && btrfs_super_cache_generation(sb) != (u64)-1)
2487 		btrfs_warn(fs_info,
2488 			"suspicious: generation < cache_generation: %llu < %llu",
2489 			btrfs_super_generation(sb),
2490 			btrfs_super_cache_generation(sb));
2491 
2492 	return ret;
2493 }
2494 
2495 /*
2496  * Validation of super block at mount time.
2497  * Some checks already done early at mount time, like csum type and incompat
2498  * flags will be skipped.
2499  */
btrfs_validate_mount_super(struct btrfs_fs_info * fs_info)2500 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2501 {
2502 	return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2503 }
2504 
2505 /*
2506  * Validation of super block at write time.
2507  * Some checks like bytenr check will be skipped as their values will be
2508  * overwritten soon.
2509  * Extra checks like csum type and incompat flags will be done here.
2510  */
btrfs_validate_write_super(struct btrfs_fs_info * fs_info,struct btrfs_super_block * sb)2511 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2512 				      struct btrfs_super_block *sb)
2513 {
2514 	int ret;
2515 
2516 	ret = btrfs_validate_super(fs_info, sb, -1);
2517 	if (ret < 0)
2518 		goto out;
2519 	if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2520 		ret = -EUCLEAN;
2521 		btrfs_err(fs_info, "invalid csum type, has %u want %u",
2522 			  btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2523 		goto out;
2524 	}
2525 	if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2526 		ret = -EUCLEAN;
2527 		btrfs_err(fs_info,
2528 		"invalid incompat flags, has 0x%llx valid mask 0x%llx",
2529 			  btrfs_super_incompat_flags(sb),
2530 			  (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2531 		goto out;
2532 	}
2533 out:
2534 	if (ret < 0)
2535 		btrfs_err(fs_info,
2536 		"super block corruption detected before writing it to disk");
2537 	return ret;
2538 }
2539 
load_super_root(struct btrfs_root * root,u64 bytenr,u64 gen,int level)2540 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2541 {
2542 	struct btrfs_tree_parent_check check = {
2543 		.level = level,
2544 		.transid = gen,
2545 		.owner_root = root->root_key.objectid
2546 	};
2547 	int ret = 0;
2548 
2549 	root->node = read_tree_block(root->fs_info, bytenr, &check);
2550 	if (IS_ERR(root->node)) {
2551 		ret = PTR_ERR(root->node);
2552 		root->node = NULL;
2553 		return ret;
2554 	}
2555 	if (!extent_buffer_uptodate(root->node)) {
2556 		free_extent_buffer(root->node);
2557 		root->node = NULL;
2558 		return -EIO;
2559 	}
2560 
2561 	btrfs_set_root_node(&root->root_item, root->node);
2562 	root->commit_root = btrfs_root_node(root);
2563 	btrfs_set_root_refs(&root->root_item, 1);
2564 	return ret;
2565 }
2566 
load_important_roots(struct btrfs_fs_info * fs_info)2567 static int load_important_roots(struct btrfs_fs_info *fs_info)
2568 {
2569 	struct btrfs_super_block *sb = fs_info->super_copy;
2570 	u64 gen, bytenr;
2571 	int level, ret;
2572 
2573 	bytenr = btrfs_super_root(sb);
2574 	gen = btrfs_super_generation(sb);
2575 	level = btrfs_super_root_level(sb);
2576 	ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2577 	if (ret) {
2578 		btrfs_warn(fs_info, "couldn't read tree root");
2579 		return ret;
2580 	}
2581 	return 0;
2582 }
2583 
init_tree_roots(struct btrfs_fs_info * fs_info)2584 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2585 {
2586 	int backup_index = find_newest_super_backup(fs_info);
2587 	struct btrfs_super_block *sb = fs_info->super_copy;
2588 	struct btrfs_root *tree_root = fs_info->tree_root;
2589 	bool handle_error = false;
2590 	int ret = 0;
2591 	int i;
2592 
2593 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2594 		if (handle_error) {
2595 			if (!IS_ERR(tree_root->node))
2596 				free_extent_buffer(tree_root->node);
2597 			tree_root->node = NULL;
2598 
2599 			if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2600 				break;
2601 
2602 			free_root_pointers(fs_info, 0);
2603 
2604 			/*
2605 			 * Don't use the log in recovery mode, it won't be
2606 			 * valid
2607 			 */
2608 			btrfs_set_super_log_root(sb, 0);
2609 
2610 			/* We can't trust the free space cache either */
2611 			btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2612 
2613 			btrfs_warn(fs_info, "try to load backup roots slot %d", i);
2614 			ret = read_backup_root(fs_info, i);
2615 			backup_index = ret;
2616 			if (ret < 0)
2617 				return ret;
2618 		}
2619 
2620 		ret = load_important_roots(fs_info);
2621 		if (ret) {
2622 			handle_error = true;
2623 			continue;
2624 		}
2625 
2626 		/*
2627 		 * No need to hold btrfs_root::objectid_mutex since the fs
2628 		 * hasn't been fully initialised and we are the only user
2629 		 */
2630 		ret = btrfs_init_root_free_objectid(tree_root);
2631 		if (ret < 0) {
2632 			handle_error = true;
2633 			continue;
2634 		}
2635 
2636 		ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2637 
2638 		ret = btrfs_read_roots(fs_info);
2639 		if (ret < 0) {
2640 			handle_error = true;
2641 			continue;
2642 		}
2643 
2644 		/* All successful */
2645 		fs_info->generation = btrfs_header_generation(tree_root->node);
2646 		fs_info->last_trans_committed = fs_info->generation;
2647 		fs_info->last_reloc_trans = 0;
2648 
2649 		/* Always begin writing backup roots after the one being used */
2650 		if (backup_index < 0) {
2651 			fs_info->backup_root_index = 0;
2652 		} else {
2653 			fs_info->backup_root_index = backup_index + 1;
2654 			fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2655 		}
2656 		break;
2657 	}
2658 
2659 	return ret;
2660 }
2661 
btrfs_init_fs_info(struct btrfs_fs_info * fs_info)2662 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2663 {
2664 	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2665 	INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2666 	INIT_LIST_HEAD(&fs_info->trans_list);
2667 	INIT_LIST_HEAD(&fs_info->dead_roots);
2668 	INIT_LIST_HEAD(&fs_info->delayed_iputs);
2669 	INIT_LIST_HEAD(&fs_info->delalloc_roots);
2670 	INIT_LIST_HEAD(&fs_info->caching_block_groups);
2671 	spin_lock_init(&fs_info->delalloc_root_lock);
2672 	spin_lock_init(&fs_info->trans_lock);
2673 	spin_lock_init(&fs_info->fs_roots_radix_lock);
2674 	spin_lock_init(&fs_info->delayed_iput_lock);
2675 	spin_lock_init(&fs_info->defrag_inodes_lock);
2676 	spin_lock_init(&fs_info->super_lock);
2677 	spin_lock_init(&fs_info->buffer_lock);
2678 	spin_lock_init(&fs_info->unused_bgs_lock);
2679 	spin_lock_init(&fs_info->treelog_bg_lock);
2680 	spin_lock_init(&fs_info->zone_active_bgs_lock);
2681 	spin_lock_init(&fs_info->relocation_bg_lock);
2682 	rwlock_init(&fs_info->tree_mod_log_lock);
2683 	rwlock_init(&fs_info->global_root_lock);
2684 	mutex_init(&fs_info->unused_bg_unpin_mutex);
2685 	mutex_init(&fs_info->reclaim_bgs_lock);
2686 	mutex_init(&fs_info->reloc_mutex);
2687 	mutex_init(&fs_info->delalloc_root_mutex);
2688 	mutex_init(&fs_info->zoned_meta_io_lock);
2689 	mutex_init(&fs_info->zoned_data_reloc_io_lock);
2690 	seqlock_init(&fs_info->profiles_lock);
2691 
2692 	btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
2693 	btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
2694 	btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
2695 	btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
2696 	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_prep,
2697 				     BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2698 	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
2699 				     BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2700 	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
2701 				     BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2702 	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
2703 				     BTRFS_LOCKDEP_TRANS_COMPLETED);
2704 
2705 	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2706 	INIT_LIST_HEAD(&fs_info->space_info);
2707 	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2708 	INIT_LIST_HEAD(&fs_info->unused_bgs);
2709 	INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2710 	INIT_LIST_HEAD(&fs_info->zone_active_bgs);
2711 #ifdef CONFIG_BTRFS_DEBUG
2712 	INIT_LIST_HEAD(&fs_info->allocated_roots);
2713 	INIT_LIST_HEAD(&fs_info->allocated_ebs);
2714 	spin_lock_init(&fs_info->eb_leak_lock);
2715 #endif
2716 	extent_map_tree_init(&fs_info->mapping_tree);
2717 	btrfs_init_block_rsv(&fs_info->global_block_rsv,
2718 			     BTRFS_BLOCK_RSV_GLOBAL);
2719 	btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2720 	btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2721 	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2722 	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2723 			     BTRFS_BLOCK_RSV_DELOPS);
2724 	btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2725 			     BTRFS_BLOCK_RSV_DELREFS);
2726 
2727 	atomic_set(&fs_info->async_delalloc_pages, 0);
2728 	atomic_set(&fs_info->defrag_running, 0);
2729 	atomic_set(&fs_info->nr_delayed_iputs, 0);
2730 	atomic64_set(&fs_info->tree_mod_seq, 0);
2731 	fs_info->global_root_tree = RB_ROOT;
2732 	fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2733 	fs_info->metadata_ratio = 0;
2734 	fs_info->defrag_inodes = RB_ROOT;
2735 	atomic64_set(&fs_info->free_chunk_space, 0);
2736 	fs_info->tree_mod_log = RB_ROOT;
2737 	fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2738 	btrfs_init_ref_verify(fs_info);
2739 
2740 	fs_info->thread_pool_size = min_t(unsigned long,
2741 					  num_online_cpus() + 2, 8);
2742 
2743 	INIT_LIST_HEAD(&fs_info->ordered_roots);
2744 	spin_lock_init(&fs_info->ordered_root_lock);
2745 
2746 	btrfs_init_scrub(fs_info);
2747 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2748 	fs_info->check_integrity_print_mask = 0;
2749 #endif
2750 	btrfs_init_balance(fs_info);
2751 	btrfs_init_async_reclaim_work(fs_info);
2752 
2753 	rwlock_init(&fs_info->block_group_cache_lock);
2754 	fs_info->block_group_cache_tree = RB_ROOT_CACHED;
2755 
2756 	extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2757 			    IO_TREE_FS_EXCLUDED_EXTENTS);
2758 
2759 	mutex_init(&fs_info->ordered_operations_mutex);
2760 	mutex_init(&fs_info->tree_log_mutex);
2761 	mutex_init(&fs_info->chunk_mutex);
2762 	mutex_init(&fs_info->transaction_kthread_mutex);
2763 	mutex_init(&fs_info->cleaner_mutex);
2764 	mutex_init(&fs_info->ro_block_group_mutex);
2765 	init_rwsem(&fs_info->commit_root_sem);
2766 	init_rwsem(&fs_info->cleanup_work_sem);
2767 	init_rwsem(&fs_info->subvol_sem);
2768 	sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2769 
2770 	btrfs_init_dev_replace_locks(fs_info);
2771 	btrfs_init_qgroup(fs_info);
2772 	btrfs_discard_init(fs_info);
2773 
2774 	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2775 	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2776 
2777 	init_waitqueue_head(&fs_info->transaction_throttle);
2778 	init_waitqueue_head(&fs_info->transaction_wait);
2779 	init_waitqueue_head(&fs_info->transaction_blocked_wait);
2780 	init_waitqueue_head(&fs_info->async_submit_wait);
2781 	init_waitqueue_head(&fs_info->delayed_iputs_wait);
2782 
2783 	/* Usable values until the real ones are cached from the superblock */
2784 	fs_info->nodesize = 4096;
2785 	fs_info->sectorsize = 4096;
2786 	fs_info->sectorsize_bits = ilog2(4096);
2787 	fs_info->stripesize = 4096;
2788 
2789 	fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
2790 
2791 	spin_lock_init(&fs_info->swapfile_pins_lock);
2792 	fs_info->swapfile_pins = RB_ROOT;
2793 
2794 	fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2795 	INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2796 }
2797 
init_mount_fs_info(struct btrfs_fs_info * fs_info,struct super_block * sb)2798 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2799 {
2800 	int ret;
2801 
2802 	fs_info->sb = sb;
2803 	sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2804 	sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2805 
2806 	ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2807 	if (ret)
2808 		return ret;
2809 
2810 	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2811 	if (ret)
2812 		return ret;
2813 
2814 	fs_info->dirty_metadata_batch = PAGE_SIZE *
2815 					(1 + ilog2(nr_cpu_ids));
2816 
2817 	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2818 	if (ret)
2819 		return ret;
2820 
2821 	ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2822 			GFP_KERNEL);
2823 	if (ret)
2824 		return ret;
2825 
2826 	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2827 					GFP_KERNEL);
2828 	if (!fs_info->delayed_root)
2829 		return -ENOMEM;
2830 	btrfs_init_delayed_root(fs_info->delayed_root);
2831 
2832 	if (sb_rdonly(sb))
2833 		set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2834 
2835 	return btrfs_alloc_stripe_hash_table(fs_info);
2836 }
2837 
btrfs_uuid_rescan_kthread(void * data)2838 static int btrfs_uuid_rescan_kthread(void *data)
2839 {
2840 	struct btrfs_fs_info *fs_info = data;
2841 	int ret;
2842 
2843 	/*
2844 	 * 1st step is to iterate through the existing UUID tree and
2845 	 * to delete all entries that contain outdated data.
2846 	 * 2nd step is to add all missing entries to the UUID tree.
2847 	 */
2848 	ret = btrfs_uuid_tree_iterate(fs_info);
2849 	if (ret < 0) {
2850 		if (ret != -EINTR)
2851 			btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2852 				   ret);
2853 		up(&fs_info->uuid_tree_rescan_sem);
2854 		return ret;
2855 	}
2856 	return btrfs_uuid_scan_kthread(data);
2857 }
2858 
btrfs_check_uuid_tree(struct btrfs_fs_info * fs_info)2859 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2860 {
2861 	struct task_struct *task;
2862 
2863 	down(&fs_info->uuid_tree_rescan_sem);
2864 	task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2865 	if (IS_ERR(task)) {
2866 		/* fs_info->update_uuid_tree_gen remains 0 in all error case */
2867 		btrfs_warn(fs_info, "failed to start uuid_rescan task");
2868 		up(&fs_info->uuid_tree_rescan_sem);
2869 		return PTR_ERR(task);
2870 	}
2871 
2872 	return 0;
2873 }
2874 
btrfs_cleanup_fs_roots(struct btrfs_fs_info * fs_info)2875 static int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2876 {
2877 	u64 root_objectid = 0;
2878 	struct btrfs_root *gang[8];
2879 	int i = 0;
2880 	int err = 0;
2881 	unsigned int ret = 0;
2882 
2883 	while (1) {
2884 		spin_lock(&fs_info->fs_roots_radix_lock);
2885 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2886 					     (void **)gang, root_objectid,
2887 					     ARRAY_SIZE(gang));
2888 		if (!ret) {
2889 			spin_unlock(&fs_info->fs_roots_radix_lock);
2890 			break;
2891 		}
2892 		root_objectid = gang[ret - 1]->root_key.objectid + 1;
2893 
2894 		for (i = 0; i < ret; i++) {
2895 			/* Avoid to grab roots in dead_roots. */
2896 			if (btrfs_root_refs(&gang[i]->root_item) == 0) {
2897 				gang[i] = NULL;
2898 				continue;
2899 			}
2900 			/* Grab all the search result for later use. */
2901 			gang[i] = btrfs_grab_root(gang[i]);
2902 		}
2903 		spin_unlock(&fs_info->fs_roots_radix_lock);
2904 
2905 		for (i = 0; i < ret; i++) {
2906 			if (!gang[i])
2907 				continue;
2908 			root_objectid = gang[i]->root_key.objectid;
2909 			err = btrfs_orphan_cleanup(gang[i]);
2910 			if (err)
2911 				goto out;
2912 			btrfs_put_root(gang[i]);
2913 		}
2914 		root_objectid++;
2915 	}
2916 out:
2917 	/* Release the uncleaned roots due to error. */
2918 	for (; i < ret; i++) {
2919 		if (gang[i])
2920 			btrfs_put_root(gang[i]);
2921 	}
2922 	return err;
2923 }
2924 
2925 /*
2926  * Some options only have meaning at mount time and shouldn't persist across
2927  * remounts, or be displayed. Clear these at the end of mount and remount
2928  * code paths.
2929  */
btrfs_clear_oneshot_options(struct btrfs_fs_info * fs_info)2930 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
2931 {
2932 	btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
2933 	btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
2934 }
2935 
2936 /*
2937  * Mounting logic specific to read-write file systems. Shared by open_ctree
2938  * and btrfs_remount when remounting from read-only to read-write.
2939  */
btrfs_start_pre_rw_mount(struct btrfs_fs_info * fs_info)2940 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
2941 {
2942 	int ret;
2943 	const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
2944 	bool rebuild_free_space_tree = false;
2945 
2946 	if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2947 	    btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2948 		rebuild_free_space_tree = true;
2949 	} else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2950 		   !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2951 		btrfs_warn(fs_info, "free space tree is invalid");
2952 		rebuild_free_space_tree = true;
2953 	}
2954 
2955 	if (rebuild_free_space_tree) {
2956 		btrfs_info(fs_info, "rebuilding free space tree");
2957 		ret = btrfs_rebuild_free_space_tree(fs_info);
2958 		if (ret) {
2959 			btrfs_warn(fs_info,
2960 				   "failed to rebuild free space tree: %d", ret);
2961 			goto out;
2962 		}
2963 	}
2964 
2965 	if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2966 	    !btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
2967 		btrfs_info(fs_info, "disabling free space tree");
2968 		ret = btrfs_delete_free_space_tree(fs_info);
2969 		if (ret) {
2970 			btrfs_warn(fs_info,
2971 				   "failed to disable free space tree: %d", ret);
2972 			goto out;
2973 		}
2974 	}
2975 
2976 	/*
2977 	 * btrfs_find_orphan_roots() is responsible for finding all the dead
2978 	 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
2979 	 * them into the fs_info->fs_roots_radix tree. This must be done before
2980 	 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
2981 	 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
2982 	 * item before the root's tree is deleted - this means that if we unmount
2983 	 * or crash before the deletion completes, on the next mount we will not
2984 	 * delete what remains of the tree because the orphan item does not
2985 	 * exists anymore, which is what tells us we have a pending deletion.
2986 	 */
2987 	ret = btrfs_find_orphan_roots(fs_info);
2988 	if (ret)
2989 		goto out;
2990 
2991 	ret = btrfs_cleanup_fs_roots(fs_info);
2992 	if (ret)
2993 		goto out;
2994 
2995 	down_read(&fs_info->cleanup_work_sem);
2996 	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2997 	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2998 		up_read(&fs_info->cleanup_work_sem);
2999 		goto out;
3000 	}
3001 	up_read(&fs_info->cleanup_work_sem);
3002 
3003 	mutex_lock(&fs_info->cleaner_mutex);
3004 	ret = btrfs_recover_relocation(fs_info);
3005 	mutex_unlock(&fs_info->cleaner_mutex);
3006 	if (ret < 0) {
3007 		btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3008 		goto out;
3009 	}
3010 
3011 	if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3012 	    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3013 		btrfs_info(fs_info, "creating free space tree");
3014 		ret = btrfs_create_free_space_tree(fs_info);
3015 		if (ret) {
3016 			btrfs_warn(fs_info,
3017 				"failed to create free space tree: %d", ret);
3018 			goto out;
3019 		}
3020 	}
3021 
3022 	if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3023 		ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3024 		if (ret)
3025 			goto out;
3026 	}
3027 
3028 	ret = btrfs_resume_balance_async(fs_info);
3029 	if (ret)
3030 		goto out;
3031 
3032 	ret = btrfs_resume_dev_replace_async(fs_info);
3033 	if (ret) {
3034 		btrfs_warn(fs_info, "failed to resume dev_replace");
3035 		goto out;
3036 	}
3037 
3038 	btrfs_qgroup_rescan_resume(fs_info);
3039 
3040 	if (!fs_info->uuid_root) {
3041 		btrfs_info(fs_info, "creating UUID tree");
3042 		ret = btrfs_create_uuid_tree(fs_info);
3043 		if (ret) {
3044 			btrfs_warn(fs_info,
3045 				   "failed to create the UUID tree %d", ret);
3046 			goto out;
3047 		}
3048 	}
3049 
3050 out:
3051 	return ret;
3052 }
3053 
3054 /*
3055  * Do various sanity and dependency checks of different features.
3056  *
3057  * @is_rw_mount:	If the mount is read-write.
3058  *
3059  * This is the place for less strict checks (like for subpage or artificial
3060  * feature dependencies).
3061  *
3062  * For strict checks or possible corruption detection, see
3063  * btrfs_validate_super().
3064  *
3065  * This should be called after btrfs_parse_options(), as some mount options
3066  * (space cache related) can modify on-disk format like free space tree and
3067  * screw up certain feature dependencies.
3068  */
btrfs_check_features(struct btrfs_fs_info * fs_info,bool is_rw_mount)3069 int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3070 {
3071 	struct btrfs_super_block *disk_super = fs_info->super_copy;
3072 	u64 incompat = btrfs_super_incompat_flags(disk_super);
3073 	const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3074 	const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3075 
3076 	if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3077 		btrfs_err(fs_info,
3078 		"cannot mount because of unknown incompat features (0x%llx)",
3079 		    incompat);
3080 		return -EINVAL;
3081 	}
3082 
3083 	/* Runtime limitation for mixed block groups. */
3084 	if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3085 	    (fs_info->sectorsize != fs_info->nodesize)) {
3086 		btrfs_err(fs_info,
3087 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3088 			fs_info->nodesize, fs_info->sectorsize);
3089 		return -EINVAL;
3090 	}
3091 
3092 	/* Mixed backref is an always-enabled feature. */
3093 	incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3094 
3095 	/* Set compression related flags just in case. */
3096 	if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3097 		incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3098 	else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3099 		incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3100 
3101 	/*
3102 	 * An ancient flag, which should really be marked deprecated.
3103 	 * Such runtime limitation doesn't really need a incompat flag.
3104 	 */
3105 	if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3106 		incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3107 
3108 	if (compat_ro_unsupp && is_rw_mount) {
3109 		btrfs_err(fs_info,
3110 	"cannot mount read-write because of unknown compat_ro features (0x%llx)",
3111 		       compat_ro);
3112 		return -EINVAL;
3113 	}
3114 
3115 	/*
3116 	 * We have unsupported RO compat features, although RO mounted, we
3117 	 * should not cause any metadata writes, including log replay.
3118 	 * Or we could screw up whatever the new feature requires.
3119 	 */
3120 	if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3121 	    !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3122 		btrfs_err(fs_info,
3123 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3124 			  compat_ro);
3125 		return -EINVAL;
3126 	}
3127 
3128 	/*
3129 	 * Artificial limitations for block group tree, to force
3130 	 * block-group-tree to rely on no-holes and free-space-tree.
3131 	 */
3132 	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3133 	    (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3134 	     !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3135 		btrfs_err(fs_info,
3136 "block-group-tree feature requires no-holes and free-space-tree features");
3137 		return -EINVAL;
3138 	}
3139 
3140 	/*
3141 	 * Subpage runtime limitation on v1 cache.
3142 	 *
3143 	 * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3144 	 * we're already defaulting to v2 cache, no need to bother v1 as it's
3145 	 * going to be deprecated anyway.
3146 	 */
3147 	if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3148 		btrfs_warn(fs_info,
3149 	"v1 space cache is not supported for page size %lu with sectorsize %u",
3150 			   PAGE_SIZE, fs_info->sectorsize);
3151 		return -EINVAL;
3152 	}
3153 
3154 	/* This can be called by remount, we need to protect the super block. */
3155 	spin_lock(&fs_info->super_lock);
3156 	btrfs_set_super_incompat_flags(disk_super, incompat);
3157 	spin_unlock(&fs_info->super_lock);
3158 
3159 	return 0;
3160 }
3161 
open_ctree(struct super_block * sb,struct btrfs_fs_devices * fs_devices,char * options)3162 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3163 		      char *options)
3164 {
3165 	u32 sectorsize;
3166 	u32 nodesize;
3167 	u32 stripesize;
3168 	u64 generation;
3169 	u64 features;
3170 	u16 csum_type;
3171 	struct btrfs_super_block *disk_super;
3172 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3173 	struct btrfs_root *tree_root;
3174 	struct btrfs_root *chunk_root;
3175 	int ret;
3176 	int level;
3177 
3178 	ret = init_mount_fs_info(fs_info, sb);
3179 	if (ret)
3180 		goto fail;
3181 
3182 	/* These need to be init'ed before we start creating inodes and such. */
3183 	tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3184 				     GFP_KERNEL);
3185 	fs_info->tree_root = tree_root;
3186 	chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3187 				      GFP_KERNEL);
3188 	fs_info->chunk_root = chunk_root;
3189 	if (!tree_root || !chunk_root) {
3190 		ret = -ENOMEM;
3191 		goto fail;
3192 	}
3193 
3194 	ret = btrfs_init_btree_inode(sb);
3195 	if (ret)
3196 		goto fail;
3197 
3198 	invalidate_bdev(fs_devices->latest_dev->bdev);
3199 
3200 	/*
3201 	 * Read super block and check the signature bytes only
3202 	 */
3203 	disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3204 	if (IS_ERR(disk_super)) {
3205 		ret = PTR_ERR(disk_super);
3206 		goto fail_alloc;
3207 	}
3208 
3209 	btrfs_info(fs_info, "first mount of filesystem %pU", disk_super->fsid);
3210 	/*
3211 	 * Verify the type first, if that or the checksum value are
3212 	 * corrupted, we'll find out
3213 	 */
3214 	csum_type = btrfs_super_csum_type(disk_super);
3215 	if (!btrfs_supported_super_csum(csum_type)) {
3216 		btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3217 			  csum_type);
3218 		ret = -EINVAL;
3219 		btrfs_release_disk_super(disk_super);
3220 		goto fail_alloc;
3221 	}
3222 
3223 	fs_info->csum_size = btrfs_super_csum_size(disk_super);
3224 
3225 	ret = btrfs_init_csum_hash(fs_info, csum_type);
3226 	if (ret) {
3227 		btrfs_release_disk_super(disk_super);
3228 		goto fail_alloc;
3229 	}
3230 
3231 	/*
3232 	 * We want to check superblock checksum, the type is stored inside.
3233 	 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3234 	 */
3235 	if (btrfs_check_super_csum(fs_info, disk_super)) {
3236 		btrfs_err(fs_info, "superblock checksum mismatch");
3237 		ret = -EINVAL;
3238 		btrfs_release_disk_super(disk_super);
3239 		goto fail_alloc;
3240 	}
3241 
3242 	/*
3243 	 * super_copy is zeroed at allocation time and we never touch the
3244 	 * following bytes up to INFO_SIZE, the checksum is calculated from
3245 	 * the whole block of INFO_SIZE
3246 	 */
3247 	memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3248 	btrfs_release_disk_super(disk_super);
3249 
3250 	disk_super = fs_info->super_copy;
3251 
3252 
3253 	features = btrfs_super_flags(disk_super);
3254 	if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3255 		features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3256 		btrfs_set_super_flags(disk_super, features);
3257 		btrfs_info(fs_info,
3258 			"found metadata UUID change in progress flag, clearing");
3259 	}
3260 
3261 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
3262 	       sizeof(*fs_info->super_for_commit));
3263 
3264 	ret = btrfs_validate_mount_super(fs_info);
3265 	if (ret) {
3266 		btrfs_err(fs_info, "superblock contains fatal errors");
3267 		ret = -EINVAL;
3268 		goto fail_alloc;
3269 	}
3270 
3271 	if (!btrfs_super_root(disk_super)) {
3272 		btrfs_err(fs_info, "invalid superblock tree root bytenr");
3273 		ret = -EINVAL;
3274 		goto fail_alloc;
3275 	}
3276 
3277 	/* check FS state, whether FS is broken. */
3278 	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3279 		WRITE_ONCE(fs_info->fs_error, -EUCLEAN);
3280 
3281 	/*
3282 	 * In the long term, we'll store the compression type in the super
3283 	 * block, and it'll be used for per file compression control.
3284 	 */
3285 	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3286 
3287 
3288 	/* Set up fs_info before parsing mount options */
3289 	nodesize = btrfs_super_nodesize(disk_super);
3290 	sectorsize = btrfs_super_sectorsize(disk_super);
3291 	stripesize = sectorsize;
3292 	fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3293 	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3294 
3295 	fs_info->nodesize = nodesize;
3296 	fs_info->sectorsize = sectorsize;
3297 	fs_info->sectorsize_bits = ilog2(sectorsize);
3298 	fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3299 	fs_info->stripesize = stripesize;
3300 
3301 	ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3302 	if (ret)
3303 		goto fail_alloc;
3304 
3305 	ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3306 	if (ret < 0)
3307 		goto fail_alloc;
3308 
3309 	if (sectorsize < PAGE_SIZE) {
3310 		struct btrfs_subpage_info *subpage_info;
3311 
3312 		/*
3313 		 * V1 space cache has some hardcoded PAGE_SIZE usage, and is
3314 		 * going to be deprecated.
3315 		 *
3316 		 * Force to use v2 cache for subpage case.
3317 		 */
3318 		btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
3319 		btrfs_set_and_info(fs_info, FREE_SPACE_TREE,
3320 			"forcing free space tree for sector size %u with page size %lu",
3321 			sectorsize, PAGE_SIZE);
3322 
3323 		btrfs_warn(fs_info,
3324 		"read-write for sector size %u with page size %lu is experimental",
3325 			   sectorsize, PAGE_SIZE);
3326 		subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3327 		if (!subpage_info) {
3328 			ret = -ENOMEM;
3329 			goto fail_alloc;
3330 		}
3331 		btrfs_init_subpage_info(subpage_info, sectorsize);
3332 		fs_info->subpage_info = subpage_info;
3333 	}
3334 
3335 	ret = btrfs_init_workqueues(fs_info);
3336 	if (ret)
3337 		goto fail_sb_buffer;
3338 
3339 	sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3340 	sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3341 
3342 	sb->s_blocksize = sectorsize;
3343 	sb->s_blocksize_bits = blksize_bits(sectorsize);
3344 	memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3345 
3346 	mutex_lock(&fs_info->chunk_mutex);
3347 	ret = btrfs_read_sys_array(fs_info);
3348 	mutex_unlock(&fs_info->chunk_mutex);
3349 	if (ret) {
3350 		btrfs_err(fs_info, "failed to read the system array: %d", ret);
3351 		goto fail_sb_buffer;
3352 	}
3353 
3354 	generation = btrfs_super_chunk_root_generation(disk_super);
3355 	level = btrfs_super_chunk_root_level(disk_super);
3356 	ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3357 			      generation, level);
3358 	if (ret) {
3359 		btrfs_err(fs_info, "failed to read chunk root");
3360 		goto fail_tree_roots;
3361 	}
3362 
3363 	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3364 			   offsetof(struct btrfs_header, chunk_tree_uuid),
3365 			   BTRFS_UUID_SIZE);
3366 
3367 	ret = btrfs_read_chunk_tree(fs_info);
3368 	if (ret) {
3369 		btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3370 		goto fail_tree_roots;
3371 	}
3372 
3373 	/*
3374 	 * At this point we know all the devices that make this filesystem,
3375 	 * including the seed devices but we don't know yet if the replace
3376 	 * target is required. So free devices that are not part of this
3377 	 * filesystem but skip the replace target device which is checked
3378 	 * below in btrfs_init_dev_replace().
3379 	 */
3380 	btrfs_free_extra_devids(fs_devices);
3381 	if (!fs_devices->latest_dev->bdev) {
3382 		btrfs_err(fs_info, "failed to read devices");
3383 		ret = -EIO;
3384 		goto fail_tree_roots;
3385 	}
3386 
3387 	ret = init_tree_roots(fs_info);
3388 	if (ret)
3389 		goto fail_tree_roots;
3390 
3391 	/*
3392 	 * Get zone type information of zoned block devices. This will also
3393 	 * handle emulation of a zoned filesystem if a regular device has the
3394 	 * zoned incompat feature flag set.
3395 	 */
3396 	ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3397 	if (ret) {
3398 		btrfs_err(fs_info,
3399 			  "zoned: failed to read device zone info: %d", ret);
3400 		goto fail_block_groups;
3401 	}
3402 
3403 	/*
3404 	 * If we have a uuid root and we're not being told to rescan we need to
3405 	 * check the generation here so we can set the
3406 	 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3407 	 * transaction during a balance or the log replay without updating the
3408 	 * uuid generation, and then if we crash we would rescan the uuid tree,
3409 	 * even though it was perfectly fine.
3410 	 */
3411 	if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3412 	    fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3413 		set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3414 
3415 	ret = btrfs_verify_dev_extents(fs_info);
3416 	if (ret) {
3417 		btrfs_err(fs_info,
3418 			  "failed to verify dev extents against chunks: %d",
3419 			  ret);
3420 		goto fail_block_groups;
3421 	}
3422 	ret = btrfs_recover_balance(fs_info);
3423 	if (ret) {
3424 		btrfs_err(fs_info, "failed to recover balance: %d", ret);
3425 		goto fail_block_groups;
3426 	}
3427 
3428 	ret = btrfs_init_dev_stats(fs_info);
3429 	if (ret) {
3430 		btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3431 		goto fail_block_groups;
3432 	}
3433 
3434 	ret = btrfs_init_dev_replace(fs_info);
3435 	if (ret) {
3436 		btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3437 		goto fail_block_groups;
3438 	}
3439 
3440 	ret = btrfs_check_zoned_mode(fs_info);
3441 	if (ret) {
3442 		btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3443 			  ret);
3444 		goto fail_block_groups;
3445 	}
3446 
3447 	ret = btrfs_sysfs_add_fsid(fs_devices);
3448 	if (ret) {
3449 		btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3450 				ret);
3451 		goto fail_block_groups;
3452 	}
3453 
3454 	ret = btrfs_sysfs_add_mounted(fs_info);
3455 	if (ret) {
3456 		btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3457 		goto fail_fsdev_sysfs;
3458 	}
3459 
3460 	ret = btrfs_init_space_info(fs_info);
3461 	if (ret) {
3462 		btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3463 		goto fail_sysfs;
3464 	}
3465 
3466 	ret = btrfs_read_block_groups(fs_info);
3467 	if (ret) {
3468 		btrfs_err(fs_info, "failed to read block groups: %d", ret);
3469 		goto fail_sysfs;
3470 	}
3471 
3472 	btrfs_free_zone_cache(fs_info);
3473 
3474 	btrfs_check_active_zone_reservation(fs_info);
3475 
3476 	if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3477 	    !btrfs_check_rw_degradable(fs_info, NULL)) {
3478 		btrfs_warn(fs_info,
3479 		"writable mount is not allowed due to too many missing devices");
3480 		ret = -EINVAL;
3481 		goto fail_sysfs;
3482 	}
3483 
3484 	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3485 					       "btrfs-cleaner");
3486 	if (IS_ERR(fs_info->cleaner_kthread)) {
3487 		ret = PTR_ERR(fs_info->cleaner_kthread);
3488 		goto fail_sysfs;
3489 	}
3490 
3491 	fs_info->transaction_kthread = kthread_run(transaction_kthread,
3492 						   tree_root,
3493 						   "btrfs-transaction");
3494 	if (IS_ERR(fs_info->transaction_kthread)) {
3495 		ret = PTR_ERR(fs_info->transaction_kthread);
3496 		goto fail_cleaner;
3497 	}
3498 
3499 	if (!btrfs_test_opt(fs_info, NOSSD) &&
3500 	    !fs_info->fs_devices->rotating) {
3501 		btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3502 	}
3503 
3504 	/*
3505 	 * For devices supporting discard turn on discard=async automatically,
3506 	 * unless it's already set or disabled. This could be turned off by
3507 	 * nodiscard for the same mount.
3508 	 *
3509 	 * The zoned mode piggy backs on the discard functionality for
3510 	 * resetting a zone. There is no reason to delay the zone reset as it is
3511 	 * fast enough. So, do not enable async discard for zoned mode.
3512 	 */
3513 	if (!(btrfs_test_opt(fs_info, DISCARD_SYNC) ||
3514 	      btrfs_test_opt(fs_info, DISCARD_ASYNC) ||
3515 	      btrfs_test_opt(fs_info, NODISCARD)) &&
3516 	    fs_info->fs_devices->discardable &&
3517 	    !btrfs_is_zoned(fs_info)) {
3518 		btrfs_set_and_info(fs_info, DISCARD_ASYNC,
3519 				   "auto enabling async discard");
3520 	}
3521 
3522 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3523 	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3524 		ret = btrfsic_mount(fs_info, fs_devices,
3525 				    btrfs_test_opt(fs_info,
3526 					CHECK_INTEGRITY_DATA) ? 1 : 0,
3527 				    fs_info->check_integrity_print_mask);
3528 		if (ret)
3529 			btrfs_warn(fs_info,
3530 				"failed to initialize integrity check module: %d",
3531 				ret);
3532 	}
3533 #endif
3534 	ret = btrfs_read_qgroup_config(fs_info);
3535 	if (ret)
3536 		goto fail_trans_kthread;
3537 
3538 	if (btrfs_build_ref_tree(fs_info))
3539 		btrfs_err(fs_info, "couldn't build ref tree");
3540 
3541 	/* do not make disk changes in broken FS or nologreplay is given */
3542 	if (btrfs_super_log_root(disk_super) != 0 &&
3543 	    !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3544 		btrfs_info(fs_info, "start tree-log replay");
3545 		ret = btrfs_replay_log(fs_info, fs_devices);
3546 		if (ret)
3547 			goto fail_qgroup;
3548 	}
3549 
3550 	fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3551 	if (IS_ERR(fs_info->fs_root)) {
3552 		ret = PTR_ERR(fs_info->fs_root);
3553 		btrfs_warn(fs_info, "failed to read fs tree: %d", ret);
3554 		fs_info->fs_root = NULL;
3555 		goto fail_qgroup;
3556 	}
3557 
3558 	if (sb_rdonly(sb))
3559 		goto clear_oneshot;
3560 
3561 	ret = btrfs_start_pre_rw_mount(fs_info);
3562 	if (ret) {
3563 		close_ctree(fs_info);
3564 		return ret;
3565 	}
3566 	btrfs_discard_resume(fs_info);
3567 
3568 	if (fs_info->uuid_root &&
3569 	    (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3570 	     fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3571 		btrfs_info(fs_info, "checking UUID tree");
3572 		ret = btrfs_check_uuid_tree(fs_info);
3573 		if (ret) {
3574 			btrfs_warn(fs_info,
3575 				"failed to check the UUID tree: %d", ret);
3576 			close_ctree(fs_info);
3577 			return ret;
3578 		}
3579 	}
3580 
3581 	set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3582 
3583 	/* Kick the cleaner thread so it'll start deleting snapshots. */
3584 	if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3585 		wake_up_process(fs_info->cleaner_kthread);
3586 
3587 clear_oneshot:
3588 	btrfs_clear_oneshot_options(fs_info);
3589 	return 0;
3590 
3591 fail_qgroup:
3592 	btrfs_free_qgroup_config(fs_info);
3593 fail_trans_kthread:
3594 	kthread_stop(fs_info->transaction_kthread);
3595 	btrfs_cleanup_transaction(fs_info);
3596 	btrfs_free_fs_roots(fs_info);
3597 fail_cleaner:
3598 	kthread_stop(fs_info->cleaner_kthread);
3599 
3600 	/*
3601 	 * make sure we're done with the btree inode before we stop our
3602 	 * kthreads
3603 	 */
3604 	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3605 
3606 fail_sysfs:
3607 	btrfs_sysfs_remove_mounted(fs_info);
3608 
3609 fail_fsdev_sysfs:
3610 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3611 
3612 fail_block_groups:
3613 	btrfs_put_block_group_cache(fs_info);
3614 
3615 fail_tree_roots:
3616 	if (fs_info->data_reloc_root)
3617 		btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3618 	free_root_pointers(fs_info, true);
3619 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3620 
3621 fail_sb_buffer:
3622 	btrfs_stop_all_workers(fs_info);
3623 	btrfs_free_block_groups(fs_info);
3624 fail_alloc:
3625 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
3626 
3627 	iput(fs_info->btree_inode);
3628 fail:
3629 	btrfs_close_devices(fs_info->fs_devices);
3630 	ASSERT(ret < 0);
3631 	return ret;
3632 }
3633 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3634 
btrfs_end_super_write(struct bio * bio)3635 static void btrfs_end_super_write(struct bio *bio)
3636 {
3637 	struct btrfs_device *device = bio->bi_private;
3638 	struct bio_vec *bvec;
3639 	struct bvec_iter_all iter_all;
3640 	struct page *page;
3641 
3642 	bio_for_each_segment_all(bvec, bio, iter_all) {
3643 		page = bvec->bv_page;
3644 
3645 		if (bio->bi_status) {
3646 			btrfs_warn_rl_in_rcu(device->fs_info,
3647 				"lost page write due to IO error on %s (%d)",
3648 				btrfs_dev_name(device),
3649 				blk_status_to_errno(bio->bi_status));
3650 			ClearPageUptodate(page);
3651 			SetPageError(page);
3652 			btrfs_dev_stat_inc_and_print(device,
3653 						     BTRFS_DEV_STAT_WRITE_ERRS);
3654 		} else {
3655 			SetPageUptodate(page);
3656 		}
3657 
3658 		put_page(page);
3659 		unlock_page(page);
3660 	}
3661 
3662 	bio_put(bio);
3663 }
3664 
btrfs_read_dev_one_super(struct block_device * bdev,int copy_num,bool drop_cache)3665 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3666 						   int copy_num, bool drop_cache)
3667 {
3668 	struct btrfs_super_block *super;
3669 	struct page *page;
3670 	u64 bytenr, bytenr_orig;
3671 	struct address_space *mapping = bdev->bd_inode->i_mapping;
3672 	int ret;
3673 
3674 	bytenr_orig = btrfs_sb_offset(copy_num);
3675 	ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3676 	if (ret == -ENOENT)
3677 		return ERR_PTR(-EINVAL);
3678 	else if (ret)
3679 		return ERR_PTR(ret);
3680 
3681 	if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3682 		return ERR_PTR(-EINVAL);
3683 
3684 	if (drop_cache) {
3685 		/* This should only be called with the primary sb. */
3686 		ASSERT(copy_num == 0);
3687 
3688 		/*
3689 		 * Drop the page of the primary superblock, so later read will
3690 		 * always read from the device.
3691 		 */
3692 		invalidate_inode_pages2_range(mapping,
3693 				bytenr >> PAGE_SHIFT,
3694 				(bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3695 	}
3696 
3697 	page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3698 	if (IS_ERR(page))
3699 		return ERR_CAST(page);
3700 
3701 	super = page_address(page);
3702 	if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3703 		btrfs_release_disk_super(super);
3704 		return ERR_PTR(-ENODATA);
3705 	}
3706 
3707 	if (btrfs_super_bytenr(super) != bytenr_orig) {
3708 		btrfs_release_disk_super(super);
3709 		return ERR_PTR(-EINVAL);
3710 	}
3711 
3712 	return super;
3713 }
3714 
3715 
btrfs_read_dev_super(struct block_device * bdev)3716 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3717 {
3718 	struct btrfs_super_block *super, *latest = NULL;
3719 	int i;
3720 	u64 transid = 0;
3721 
3722 	/* we would like to check all the supers, but that would make
3723 	 * a btrfs mount succeed after a mkfs from a different FS.
3724 	 * So, we need to add a special mount option to scan for
3725 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3726 	 */
3727 	for (i = 0; i < 1; i++) {
3728 		super = btrfs_read_dev_one_super(bdev, i, false);
3729 		if (IS_ERR(super))
3730 			continue;
3731 
3732 		if (!latest || btrfs_super_generation(super) > transid) {
3733 			if (latest)
3734 				btrfs_release_disk_super(super);
3735 
3736 			latest = super;
3737 			transid = btrfs_super_generation(super);
3738 		}
3739 	}
3740 
3741 	return super;
3742 }
3743 
3744 /*
3745  * Write superblock @sb to the @device. Do not wait for completion, all the
3746  * pages we use for writing are locked.
3747  *
3748  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3749  * the expected device size at commit time. Note that max_mirrors must be
3750  * same for write and wait phases.
3751  *
3752  * Return number of errors when page is not found or submission fails.
3753  */
write_dev_supers(struct btrfs_device * device,struct btrfs_super_block * sb,int max_mirrors)3754 static int write_dev_supers(struct btrfs_device *device,
3755 			    struct btrfs_super_block *sb, int max_mirrors)
3756 {
3757 	struct btrfs_fs_info *fs_info = device->fs_info;
3758 	struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3759 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3760 	int i;
3761 	int errors = 0;
3762 	int ret;
3763 	u64 bytenr, bytenr_orig;
3764 
3765 	if (max_mirrors == 0)
3766 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3767 
3768 	shash->tfm = fs_info->csum_shash;
3769 
3770 	for (i = 0; i < max_mirrors; i++) {
3771 		struct page *page;
3772 		struct bio *bio;
3773 		struct btrfs_super_block *disk_super;
3774 
3775 		bytenr_orig = btrfs_sb_offset(i);
3776 		ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3777 		if (ret == -ENOENT) {
3778 			continue;
3779 		} else if (ret < 0) {
3780 			btrfs_err(device->fs_info,
3781 				"couldn't get super block location for mirror %d",
3782 				i);
3783 			errors++;
3784 			continue;
3785 		}
3786 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3787 		    device->commit_total_bytes)
3788 			break;
3789 
3790 		btrfs_set_super_bytenr(sb, bytenr_orig);
3791 
3792 		crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3793 				    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3794 				    sb->csum);
3795 
3796 		page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3797 					   GFP_NOFS);
3798 		if (!page) {
3799 			btrfs_err(device->fs_info,
3800 			    "couldn't get super block page for bytenr %llu",
3801 			    bytenr);
3802 			errors++;
3803 			continue;
3804 		}
3805 
3806 		/* Bump the refcount for wait_dev_supers() */
3807 		get_page(page);
3808 
3809 		disk_super = page_address(page);
3810 		memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3811 
3812 		/*
3813 		 * Directly use bios here instead of relying on the page cache
3814 		 * to do I/O, so we don't lose the ability to do integrity
3815 		 * checking.
3816 		 */
3817 		bio = bio_alloc(device->bdev, 1,
3818 				REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
3819 				GFP_NOFS);
3820 		bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3821 		bio->bi_private = device;
3822 		bio->bi_end_io = btrfs_end_super_write;
3823 		__bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3824 			       offset_in_page(bytenr));
3825 
3826 		/*
3827 		 * We FUA only the first super block.  The others we allow to
3828 		 * go down lazy and there's a short window where the on-disk
3829 		 * copies might still contain the older version.
3830 		 */
3831 		if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3832 			bio->bi_opf |= REQ_FUA;
3833 
3834 		btrfsic_check_bio(bio);
3835 		submit_bio(bio);
3836 
3837 		if (btrfs_advance_sb_log(device, i))
3838 			errors++;
3839 	}
3840 	return errors < i ? 0 : -1;
3841 }
3842 
3843 /*
3844  * Wait for write completion of superblocks done by write_dev_supers,
3845  * @max_mirrors same for write and wait phases.
3846  *
3847  * Return number of errors when page is not found or not marked up to
3848  * date.
3849  */
wait_dev_supers(struct btrfs_device * device,int max_mirrors)3850 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3851 {
3852 	int i;
3853 	int errors = 0;
3854 	bool primary_failed = false;
3855 	int ret;
3856 	u64 bytenr;
3857 
3858 	if (max_mirrors == 0)
3859 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3860 
3861 	for (i = 0; i < max_mirrors; i++) {
3862 		struct page *page;
3863 
3864 		ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3865 		if (ret == -ENOENT) {
3866 			break;
3867 		} else if (ret < 0) {
3868 			errors++;
3869 			if (i == 0)
3870 				primary_failed = true;
3871 			continue;
3872 		}
3873 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3874 		    device->commit_total_bytes)
3875 			break;
3876 
3877 		page = find_get_page(device->bdev->bd_inode->i_mapping,
3878 				     bytenr >> PAGE_SHIFT);
3879 		if (!page) {
3880 			errors++;
3881 			if (i == 0)
3882 				primary_failed = true;
3883 			continue;
3884 		}
3885 		/* Page is submitted locked and unlocked once the IO completes */
3886 		wait_on_page_locked(page);
3887 		if (PageError(page)) {
3888 			errors++;
3889 			if (i == 0)
3890 				primary_failed = true;
3891 		}
3892 
3893 		/* Drop our reference */
3894 		put_page(page);
3895 
3896 		/* Drop the reference from the writing run */
3897 		put_page(page);
3898 	}
3899 
3900 	/* log error, force error return */
3901 	if (primary_failed) {
3902 		btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3903 			  device->devid);
3904 		return -1;
3905 	}
3906 
3907 	return errors < i ? 0 : -1;
3908 }
3909 
3910 /*
3911  * endio for the write_dev_flush, this will wake anyone waiting
3912  * for the barrier when it is done
3913  */
btrfs_end_empty_barrier(struct bio * bio)3914 static void btrfs_end_empty_barrier(struct bio *bio)
3915 {
3916 	bio_uninit(bio);
3917 	complete(bio->bi_private);
3918 }
3919 
3920 /*
3921  * Submit a flush request to the device if it supports it. Error handling is
3922  * done in the waiting counterpart.
3923  */
write_dev_flush(struct btrfs_device * device)3924 static void write_dev_flush(struct btrfs_device *device)
3925 {
3926 	struct bio *bio = &device->flush_bio;
3927 
3928 	device->last_flush_error = BLK_STS_OK;
3929 
3930 #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3931 	/*
3932 	 * When a disk has write caching disabled, we skip submission of a bio
3933 	 * with flush and sync requests before writing the superblock, since
3934 	 * it's not needed. However when the integrity checker is enabled, this
3935 	 * results in reports that there are metadata blocks referred by a
3936 	 * superblock that were not properly flushed. So don't skip the bio
3937 	 * submission only when the integrity checker is enabled for the sake
3938 	 * of simplicity, since this is a debug tool and not meant for use in
3939 	 * non-debug builds.
3940 	 */
3941 	if (!bdev_write_cache(device->bdev))
3942 		return;
3943 #endif
3944 
3945 	bio_init(bio, device->bdev, NULL, 0,
3946 		 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
3947 	bio->bi_end_io = btrfs_end_empty_barrier;
3948 	init_completion(&device->flush_wait);
3949 	bio->bi_private = &device->flush_wait;
3950 
3951 	btrfsic_check_bio(bio);
3952 	submit_bio(bio);
3953 	set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3954 }
3955 
3956 /*
3957  * If the flush bio has been submitted by write_dev_flush, wait for it.
3958  * Return true for any error, and false otherwise.
3959  */
wait_dev_flush(struct btrfs_device * device)3960 static bool wait_dev_flush(struct btrfs_device *device)
3961 {
3962 	struct bio *bio = &device->flush_bio;
3963 
3964 	if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3965 		return false;
3966 
3967 	wait_for_completion_io(&device->flush_wait);
3968 
3969 	if (bio->bi_status) {
3970 		device->last_flush_error = bio->bi_status;
3971 		btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_FLUSH_ERRS);
3972 		return true;
3973 	}
3974 
3975 	return false;
3976 }
3977 
3978 /*
3979  * send an empty flush down to each device in parallel,
3980  * then wait for them
3981  */
barrier_all_devices(struct btrfs_fs_info * info)3982 static int barrier_all_devices(struct btrfs_fs_info *info)
3983 {
3984 	struct list_head *head;
3985 	struct btrfs_device *dev;
3986 	int errors_wait = 0;
3987 
3988 	lockdep_assert_held(&info->fs_devices->device_list_mutex);
3989 	/* send down all the barriers */
3990 	head = &info->fs_devices->devices;
3991 	list_for_each_entry(dev, head, dev_list) {
3992 		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3993 			continue;
3994 		if (!dev->bdev)
3995 			continue;
3996 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3997 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3998 			continue;
3999 
4000 		write_dev_flush(dev);
4001 	}
4002 
4003 	/* wait for all the barriers */
4004 	list_for_each_entry(dev, head, dev_list) {
4005 		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4006 			continue;
4007 		if (!dev->bdev) {
4008 			errors_wait++;
4009 			continue;
4010 		}
4011 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4012 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4013 			continue;
4014 
4015 		if (wait_dev_flush(dev))
4016 			errors_wait++;
4017 	}
4018 
4019 	/*
4020 	 * Checks last_flush_error of disks in order to determine the device
4021 	 * state.
4022 	 */
4023 	if (errors_wait && !btrfs_check_rw_degradable(info, NULL))
4024 		return -EIO;
4025 
4026 	return 0;
4027 }
4028 
btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)4029 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4030 {
4031 	int raid_type;
4032 	int min_tolerated = INT_MAX;
4033 
4034 	if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4035 	    (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4036 		min_tolerated = min_t(int, min_tolerated,
4037 				    btrfs_raid_array[BTRFS_RAID_SINGLE].
4038 				    tolerated_failures);
4039 
4040 	for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4041 		if (raid_type == BTRFS_RAID_SINGLE)
4042 			continue;
4043 		if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4044 			continue;
4045 		min_tolerated = min_t(int, min_tolerated,
4046 				    btrfs_raid_array[raid_type].
4047 				    tolerated_failures);
4048 	}
4049 
4050 	if (min_tolerated == INT_MAX) {
4051 		pr_warn("BTRFS: unknown raid flag: %llu", flags);
4052 		min_tolerated = 0;
4053 	}
4054 
4055 	return min_tolerated;
4056 }
4057 
write_all_supers(struct btrfs_fs_info * fs_info,int max_mirrors)4058 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4059 {
4060 	struct list_head *head;
4061 	struct btrfs_device *dev;
4062 	struct btrfs_super_block *sb;
4063 	struct btrfs_dev_item *dev_item;
4064 	int ret;
4065 	int do_barriers;
4066 	int max_errors;
4067 	int total_errors = 0;
4068 	u64 flags;
4069 
4070 	do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4071 
4072 	/*
4073 	 * max_mirrors == 0 indicates we're from commit_transaction,
4074 	 * not from fsync where the tree roots in fs_info have not
4075 	 * been consistent on disk.
4076 	 */
4077 	if (max_mirrors == 0)
4078 		backup_super_roots(fs_info);
4079 
4080 	sb = fs_info->super_for_commit;
4081 	dev_item = &sb->dev_item;
4082 
4083 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
4084 	head = &fs_info->fs_devices->devices;
4085 	max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4086 
4087 	if (do_barriers) {
4088 		ret = barrier_all_devices(fs_info);
4089 		if (ret) {
4090 			mutex_unlock(
4091 				&fs_info->fs_devices->device_list_mutex);
4092 			btrfs_handle_fs_error(fs_info, ret,
4093 					      "errors while submitting device barriers.");
4094 			return ret;
4095 		}
4096 	}
4097 
4098 	list_for_each_entry(dev, head, dev_list) {
4099 		if (!dev->bdev) {
4100 			total_errors++;
4101 			continue;
4102 		}
4103 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4104 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4105 			continue;
4106 
4107 		btrfs_set_stack_device_generation(dev_item, 0);
4108 		btrfs_set_stack_device_type(dev_item, dev->type);
4109 		btrfs_set_stack_device_id(dev_item, dev->devid);
4110 		btrfs_set_stack_device_total_bytes(dev_item,
4111 						   dev->commit_total_bytes);
4112 		btrfs_set_stack_device_bytes_used(dev_item,
4113 						  dev->commit_bytes_used);
4114 		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4115 		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4116 		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4117 		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4118 		memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4119 		       BTRFS_FSID_SIZE);
4120 
4121 		flags = btrfs_super_flags(sb);
4122 		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4123 
4124 		ret = btrfs_validate_write_super(fs_info, sb);
4125 		if (ret < 0) {
4126 			mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4127 			btrfs_handle_fs_error(fs_info, -EUCLEAN,
4128 				"unexpected superblock corruption detected");
4129 			return -EUCLEAN;
4130 		}
4131 
4132 		ret = write_dev_supers(dev, sb, max_mirrors);
4133 		if (ret)
4134 			total_errors++;
4135 	}
4136 	if (total_errors > max_errors) {
4137 		btrfs_err(fs_info, "%d errors while writing supers",
4138 			  total_errors);
4139 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4140 
4141 		/* FUA is masked off if unsupported and can't be the reason */
4142 		btrfs_handle_fs_error(fs_info, -EIO,
4143 				      "%d errors while writing supers",
4144 				      total_errors);
4145 		return -EIO;
4146 	}
4147 
4148 	total_errors = 0;
4149 	list_for_each_entry(dev, head, dev_list) {
4150 		if (!dev->bdev)
4151 			continue;
4152 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4153 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4154 			continue;
4155 
4156 		ret = wait_dev_supers(dev, max_mirrors);
4157 		if (ret)
4158 			total_errors++;
4159 	}
4160 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4161 	if (total_errors > max_errors) {
4162 		btrfs_handle_fs_error(fs_info, -EIO,
4163 				      "%d errors while writing supers",
4164 				      total_errors);
4165 		return -EIO;
4166 	}
4167 	return 0;
4168 }
4169 
4170 /* 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)4171 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4172 				  struct btrfs_root *root)
4173 {
4174 	bool drop_ref = false;
4175 
4176 	spin_lock(&fs_info->fs_roots_radix_lock);
4177 	radix_tree_delete(&fs_info->fs_roots_radix,
4178 			  (unsigned long)root->root_key.objectid);
4179 	if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4180 		drop_ref = true;
4181 	spin_unlock(&fs_info->fs_roots_radix_lock);
4182 
4183 	if (BTRFS_FS_ERROR(fs_info)) {
4184 		ASSERT(root->log_root == NULL);
4185 		if (root->reloc_root) {
4186 			btrfs_put_root(root->reloc_root);
4187 			root->reloc_root = NULL;
4188 		}
4189 	}
4190 
4191 	if (drop_ref)
4192 		btrfs_put_root(root);
4193 }
4194 
btrfs_commit_super(struct btrfs_fs_info * fs_info)4195 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4196 {
4197 	struct btrfs_root *root = fs_info->tree_root;
4198 	struct btrfs_trans_handle *trans;
4199 
4200 	mutex_lock(&fs_info->cleaner_mutex);
4201 	btrfs_run_delayed_iputs(fs_info);
4202 	mutex_unlock(&fs_info->cleaner_mutex);
4203 	wake_up_process(fs_info->cleaner_kthread);
4204 
4205 	/* wait until ongoing cleanup work done */
4206 	down_write(&fs_info->cleanup_work_sem);
4207 	up_write(&fs_info->cleanup_work_sem);
4208 
4209 	trans = btrfs_join_transaction(root);
4210 	if (IS_ERR(trans))
4211 		return PTR_ERR(trans);
4212 	return btrfs_commit_transaction(trans);
4213 }
4214 
warn_about_uncommitted_trans(struct btrfs_fs_info * fs_info)4215 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4216 {
4217 	struct btrfs_transaction *trans;
4218 	struct btrfs_transaction *tmp;
4219 	bool found = false;
4220 
4221 	if (list_empty(&fs_info->trans_list))
4222 		return;
4223 
4224 	/*
4225 	 * This function is only called at the very end of close_ctree(),
4226 	 * thus no other running transaction, no need to take trans_lock.
4227 	 */
4228 	ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4229 	list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4230 		struct extent_state *cached = NULL;
4231 		u64 dirty_bytes = 0;
4232 		u64 cur = 0;
4233 		u64 found_start;
4234 		u64 found_end;
4235 
4236 		found = true;
4237 		while (find_first_extent_bit(&trans->dirty_pages, cur,
4238 			&found_start, &found_end, EXTENT_DIRTY, &cached)) {
4239 			dirty_bytes += found_end + 1 - found_start;
4240 			cur = found_end + 1;
4241 		}
4242 		btrfs_warn(fs_info,
4243 	"transaction %llu (with %llu dirty metadata bytes) is not committed",
4244 			   trans->transid, dirty_bytes);
4245 		btrfs_cleanup_one_transaction(trans, fs_info);
4246 
4247 		if (trans == fs_info->running_transaction)
4248 			fs_info->running_transaction = NULL;
4249 		list_del_init(&trans->list);
4250 
4251 		btrfs_put_transaction(trans);
4252 		trace_btrfs_transaction_commit(fs_info);
4253 	}
4254 	ASSERT(!found);
4255 }
4256 
close_ctree(struct btrfs_fs_info * fs_info)4257 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4258 {
4259 	int ret;
4260 
4261 	set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4262 
4263 	/*
4264 	 * If we had UNFINISHED_DROPS we could still be processing them, so
4265 	 * clear that bit and wake up relocation so it can stop.
4266 	 * We must do this before stopping the block group reclaim task, because
4267 	 * at btrfs_relocate_block_group() we wait for this bit, and after the
4268 	 * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4269 	 * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4270 	 * return 1.
4271 	 */
4272 	btrfs_wake_unfinished_drop(fs_info);
4273 
4274 	/*
4275 	 * We may have the reclaim task running and relocating a data block group,
4276 	 * in which case it may create delayed iputs. So stop it before we park
4277 	 * the cleaner kthread otherwise we can get new delayed iputs after
4278 	 * parking the cleaner, and that can make the async reclaim task to hang
4279 	 * if it's waiting for delayed iputs to complete, since the cleaner is
4280 	 * parked and can not run delayed iputs - this will make us hang when
4281 	 * trying to stop the async reclaim task.
4282 	 */
4283 	cancel_work_sync(&fs_info->reclaim_bgs_work);
4284 	/*
4285 	 * We don't want the cleaner to start new transactions, add more delayed
4286 	 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4287 	 * because that frees the task_struct, and the transaction kthread might
4288 	 * still try to wake up the cleaner.
4289 	 */
4290 	kthread_park(fs_info->cleaner_kthread);
4291 
4292 	/* wait for the qgroup rescan worker to stop */
4293 	btrfs_qgroup_wait_for_completion(fs_info, false);
4294 
4295 	/* wait for the uuid_scan task to finish */
4296 	down(&fs_info->uuid_tree_rescan_sem);
4297 	/* avoid complains from lockdep et al., set sem back to initial state */
4298 	up(&fs_info->uuid_tree_rescan_sem);
4299 
4300 	/* pause restriper - we want to resume on mount */
4301 	btrfs_pause_balance(fs_info);
4302 
4303 	btrfs_dev_replace_suspend_for_unmount(fs_info);
4304 
4305 	btrfs_scrub_cancel(fs_info);
4306 
4307 	/* wait for any defraggers to finish */
4308 	wait_event(fs_info->transaction_wait,
4309 		   (atomic_read(&fs_info->defrag_running) == 0));
4310 
4311 	/* clear out the rbtree of defraggable inodes */
4312 	btrfs_cleanup_defrag_inodes(fs_info);
4313 
4314 	/*
4315 	 * After we parked the cleaner kthread, ordered extents may have
4316 	 * completed and created new delayed iputs. If one of the async reclaim
4317 	 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4318 	 * can hang forever trying to stop it, because if a delayed iput is
4319 	 * added after it ran btrfs_run_delayed_iputs() and before it called
4320 	 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4321 	 * no one else to run iputs.
4322 	 *
4323 	 * So wait for all ongoing ordered extents to complete and then run
4324 	 * delayed iputs. This works because once we reach this point no one
4325 	 * can either create new ordered extents nor create delayed iputs
4326 	 * through some other means.
4327 	 *
4328 	 * Also note that btrfs_wait_ordered_roots() is not safe here, because
4329 	 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4330 	 * but the delayed iput for the respective inode is made only when doing
4331 	 * the final btrfs_put_ordered_extent() (which must happen at
4332 	 * btrfs_finish_ordered_io() when we are unmounting).
4333 	 */
4334 	btrfs_flush_workqueue(fs_info->endio_write_workers);
4335 	/* Ordered extents for free space inodes. */
4336 	btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4337 	btrfs_run_delayed_iputs(fs_info);
4338 
4339 	cancel_work_sync(&fs_info->async_reclaim_work);
4340 	cancel_work_sync(&fs_info->async_data_reclaim_work);
4341 	cancel_work_sync(&fs_info->preempt_reclaim_work);
4342 
4343 	/* Cancel or finish ongoing discard work */
4344 	btrfs_discard_cleanup(fs_info);
4345 
4346 	if (!sb_rdonly(fs_info->sb)) {
4347 		/*
4348 		 * The cleaner kthread is stopped, so do one final pass over
4349 		 * unused block groups.
4350 		 */
4351 		btrfs_delete_unused_bgs(fs_info);
4352 
4353 		/*
4354 		 * There might be existing delayed inode workers still running
4355 		 * and holding an empty delayed inode item. We must wait for
4356 		 * them to complete first because they can create a transaction.
4357 		 * This happens when someone calls btrfs_balance_delayed_items()
4358 		 * and then a transaction commit runs the same delayed nodes
4359 		 * before any delayed worker has done something with the nodes.
4360 		 * We must wait for any worker here and not at transaction
4361 		 * commit time since that could cause a deadlock.
4362 		 * This is a very rare case.
4363 		 */
4364 		btrfs_flush_workqueue(fs_info->delayed_workers);
4365 
4366 		ret = btrfs_commit_super(fs_info);
4367 		if (ret)
4368 			btrfs_err(fs_info, "commit super ret %d", ret);
4369 	}
4370 
4371 	if (BTRFS_FS_ERROR(fs_info))
4372 		btrfs_error_commit_super(fs_info);
4373 
4374 	kthread_stop(fs_info->transaction_kthread);
4375 	kthread_stop(fs_info->cleaner_kthread);
4376 
4377 	ASSERT(list_empty(&fs_info->delayed_iputs));
4378 	set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4379 
4380 	if (btrfs_check_quota_leak(fs_info)) {
4381 		WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4382 		btrfs_err(fs_info, "qgroup reserved space leaked");
4383 	}
4384 
4385 	btrfs_free_qgroup_config(fs_info);
4386 	ASSERT(list_empty(&fs_info->delalloc_roots));
4387 
4388 	if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4389 		btrfs_info(fs_info, "at unmount delalloc count %lld",
4390 		       percpu_counter_sum(&fs_info->delalloc_bytes));
4391 	}
4392 
4393 	if (percpu_counter_sum(&fs_info->ordered_bytes))
4394 		btrfs_info(fs_info, "at unmount dio bytes count %lld",
4395 			   percpu_counter_sum(&fs_info->ordered_bytes));
4396 
4397 	btrfs_sysfs_remove_mounted(fs_info);
4398 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4399 
4400 	btrfs_put_block_group_cache(fs_info);
4401 
4402 	/*
4403 	 * we must make sure there is not any read request to
4404 	 * submit after we stopping all workers.
4405 	 */
4406 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4407 	btrfs_stop_all_workers(fs_info);
4408 
4409 	/* We shouldn't have any transaction open at this point */
4410 	warn_about_uncommitted_trans(fs_info);
4411 
4412 	clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4413 	free_root_pointers(fs_info, true);
4414 	btrfs_free_fs_roots(fs_info);
4415 
4416 	/*
4417 	 * We must free the block groups after dropping the fs_roots as we could
4418 	 * have had an IO error and have left over tree log blocks that aren't
4419 	 * cleaned up until the fs roots are freed.  This makes the block group
4420 	 * accounting appear to be wrong because there's pending reserved bytes,
4421 	 * so make sure we do the block group cleanup afterwards.
4422 	 */
4423 	btrfs_free_block_groups(fs_info);
4424 
4425 	iput(fs_info->btree_inode);
4426 
4427 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4428 	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4429 		btrfsic_unmount(fs_info->fs_devices);
4430 #endif
4431 
4432 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
4433 	btrfs_close_devices(fs_info->fs_devices);
4434 }
4435 
btrfs_mark_buffer_dirty(struct btrfs_trans_handle * trans,struct extent_buffer * buf)4436 void btrfs_mark_buffer_dirty(struct btrfs_trans_handle *trans,
4437 			     struct extent_buffer *buf)
4438 {
4439 	struct btrfs_fs_info *fs_info = buf->fs_info;
4440 	u64 transid = btrfs_header_generation(buf);
4441 
4442 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4443 	/*
4444 	 * This is a fast path so only do this check if we have sanity tests
4445 	 * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4446 	 * outside of the sanity tests.
4447 	 */
4448 	if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4449 		return;
4450 #endif
4451 	/* This is an active transaction (its state < TRANS_STATE_UNBLOCKED). */
4452 	ASSERT(trans->transid == fs_info->generation);
4453 	btrfs_assert_tree_write_locked(buf);
4454 	if (transid != fs_info->generation) {
4455 		WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4456 			buf->start, transid, fs_info->generation);
4457 		btrfs_abort_transaction(trans, -EUCLEAN);
4458 	}
4459 	set_extent_buffer_dirty(buf);
4460 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4461 	/*
4462 	 * btrfs_check_leaf() won't check item data if we don't have WRITTEN
4463 	 * set, so this will only validate the basic structure of the items.
4464 	 */
4465 	if (btrfs_header_level(buf) == 0 && btrfs_check_leaf(buf)) {
4466 		btrfs_print_leaf(buf);
4467 		ASSERT(0);
4468 	}
4469 #endif
4470 }
4471 
__btrfs_btree_balance_dirty(struct btrfs_fs_info * fs_info,int flush_delayed)4472 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4473 					int flush_delayed)
4474 {
4475 	/*
4476 	 * looks as though older kernels can get into trouble with
4477 	 * this code, they end up stuck in balance_dirty_pages forever
4478 	 */
4479 	int ret;
4480 
4481 	if (current->flags & PF_MEMALLOC)
4482 		return;
4483 
4484 	if (flush_delayed)
4485 		btrfs_balance_delayed_items(fs_info);
4486 
4487 	ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4488 				     BTRFS_DIRTY_METADATA_THRESH,
4489 				     fs_info->dirty_metadata_batch);
4490 	if (ret > 0) {
4491 		balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4492 	}
4493 }
4494 
btrfs_btree_balance_dirty(struct btrfs_fs_info * fs_info)4495 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4496 {
4497 	__btrfs_btree_balance_dirty(fs_info, 1);
4498 }
4499 
btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info * fs_info)4500 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4501 {
4502 	__btrfs_btree_balance_dirty(fs_info, 0);
4503 }
4504 
btrfs_error_commit_super(struct btrfs_fs_info * fs_info)4505 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4506 {
4507 	/* cleanup FS via transaction */
4508 	btrfs_cleanup_transaction(fs_info);
4509 
4510 	mutex_lock(&fs_info->cleaner_mutex);
4511 	btrfs_run_delayed_iputs(fs_info);
4512 	mutex_unlock(&fs_info->cleaner_mutex);
4513 
4514 	down_write(&fs_info->cleanup_work_sem);
4515 	up_write(&fs_info->cleanup_work_sem);
4516 }
4517 
btrfs_drop_all_logs(struct btrfs_fs_info * fs_info)4518 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4519 {
4520 	struct btrfs_root *gang[8];
4521 	u64 root_objectid = 0;
4522 	int ret;
4523 
4524 	spin_lock(&fs_info->fs_roots_radix_lock);
4525 	while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4526 					     (void **)gang, root_objectid,
4527 					     ARRAY_SIZE(gang))) != 0) {
4528 		int i;
4529 
4530 		for (i = 0; i < ret; i++)
4531 			gang[i] = btrfs_grab_root(gang[i]);
4532 		spin_unlock(&fs_info->fs_roots_radix_lock);
4533 
4534 		for (i = 0; i < ret; i++) {
4535 			if (!gang[i])
4536 				continue;
4537 			root_objectid = gang[i]->root_key.objectid;
4538 			btrfs_free_log(NULL, gang[i]);
4539 			btrfs_put_root(gang[i]);
4540 		}
4541 		root_objectid++;
4542 		spin_lock(&fs_info->fs_roots_radix_lock);
4543 	}
4544 	spin_unlock(&fs_info->fs_roots_radix_lock);
4545 	btrfs_free_log_root_tree(NULL, fs_info);
4546 }
4547 
btrfs_destroy_ordered_extents(struct btrfs_root * root)4548 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4549 {
4550 	struct btrfs_ordered_extent *ordered;
4551 
4552 	spin_lock(&root->ordered_extent_lock);
4553 	/*
4554 	 * This will just short circuit the ordered completion stuff which will
4555 	 * make sure the ordered extent gets properly cleaned up.
4556 	 */
4557 	list_for_each_entry(ordered, &root->ordered_extents,
4558 			    root_extent_list)
4559 		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4560 	spin_unlock(&root->ordered_extent_lock);
4561 }
4562 
btrfs_destroy_all_ordered_extents(struct btrfs_fs_info * fs_info)4563 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4564 {
4565 	struct btrfs_root *root;
4566 	LIST_HEAD(splice);
4567 
4568 	spin_lock(&fs_info->ordered_root_lock);
4569 	list_splice_init(&fs_info->ordered_roots, &splice);
4570 	while (!list_empty(&splice)) {
4571 		root = list_first_entry(&splice, struct btrfs_root,
4572 					ordered_root);
4573 		list_move_tail(&root->ordered_root,
4574 			       &fs_info->ordered_roots);
4575 
4576 		spin_unlock(&fs_info->ordered_root_lock);
4577 		btrfs_destroy_ordered_extents(root);
4578 
4579 		cond_resched();
4580 		spin_lock(&fs_info->ordered_root_lock);
4581 	}
4582 	spin_unlock(&fs_info->ordered_root_lock);
4583 
4584 	/*
4585 	 * We need this here because if we've been flipped read-only we won't
4586 	 * get sync() from the umount, so we need to make sure any ordered
4587 	 * extents that haven't had their dirty pages IO start writeout yet
4588 	 * actually get run and error out properly.
4589 	 */
4590 	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4591 }
4592 
btrfs_destroy_delayed_refs(struct btrfs_transaction * trans,struct btrfs_fs_info * fs_info)4593 static void btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4594 				       struct btrfs_fs_info *fs_info)
4595 {
4596 	struct rb_node *node;
4597 	struct btrfs_delayed_ref_root *delayed_refs;
4598 	struct btrfs_delayed_ref_node *ref;
4599 
4600 	delayed_refs = &trans->delayed_refs;
4601 
4602 	spin_lock(&delayed_refs->lock);
4603 	if (atomic_read(&delayed_refs->num_entries) == 0) {
4604 		spin_unlock(&delayed_refs->lock);
4605 		btrfs_debug(fs_info, "delayed_refs has NO entry");
4606 		return;
4607 	}
4608 
4609 	while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4610 		struct btrfs_delayed_ref_head *head;
4611 		struct rb_node *n;
4612 		bool pin_bytes = false;
4613 
4614 		head = rb_entry(node, struct btrfs_delayed_ref_head,
4615 				href_node);
4616 		if (btrfs_delayed_ref_lock(delayed_refs, head))
4617 			continue;
4618 
4619 		spin_lock(&head->lock);
4620 		while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4621 			ref = rb_entry(n, struct btrfs_delayed_ref_node,
4622 				       ref_node);
4623 			rb_erase_cached(&ref->ref_node, &head->ref_tree);
4624 			RB_CLEAR_NODE(&ref->ref_node);
4625 			if (!list_empty(&ref->add_list))
4626 				list_del(&ref->add_list);
4627 			atomic_dec(&delayed_refs->num_entries);
4628 			btrfs_put_delayed_ref(ref);
4629 		}
4630 		if (head->must_insert_reserved)
4631 			pin_bytes = true;
4632 		btrfs_free_delayed_extent_op(head->extent_op);
4633 		btrfs_delete_ref_head(delayed_refs, head);
4634 		spin_unlock(&head->lock);
4635 		spin_unlock(&delayed_refs->lock);
4636 		mutex_unlock(&head->mutex);
4637 
4638 		if (pin_bytes) {
4639 			struct btrfs_block_group *cache;
4640 
4641 			cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4642 			BUG_ON(!cache);
4643 
4644 			spin_lock(&cache->space_info->lock);
4645 			spin_lock(&cache->lock);
4646 			cache->pinned += head->num_bytes;
4647 			btrfs_space_info_update_bytes_pinned(fs_info,
4648 				cache->space_info, head->num_bytes);
4649 			cache->reserved -= head->num_bytes;
4650 			cache->space_info->bytes_reserved -= head->num_bytes;
4651 			spin_unlock(&cache->lock);
4652 			spin_unlock(&cache->space_info->lock);
4653 
4654 			btrfs_put_block_group(cache);
4655 
4656 			btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4657 				head->bytenr + head->num_bytes - 1);
4658 		}
4659 		btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4660 		btrfs_put_delayed_ref_head(head);
4661 		cond_resched();
4662 		spin_lock(&delayed_refs->lock);
4663 	}
4664 	btrfs_qgroup_destroy_extent_records(trans);
4665 
4666 	spin_unlock(&delayed_refs->lock);
4667 }
4668 
btrfs_destroy_delalloc_inodes(struct btrfs_root * root)4669 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4670 {
4671 	struct btrfs_inode *btrfs_inode;
4672 	LIST_HEAD(splice);
4673 
4674 	spin_lock(&root->delalloc_lock);
4675 	list_splice_init(&root->delalloc_inodes, &splice);
4676 
4677 	while (!list_empty(&splice)) {
4678 		struct inode *inode = NULL;
4679 		btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4680 					       delalloc_inodes);
4681 		__btrfs_del_delalloc_inode(root, btrfs_inode);
4682 		spin_unlock(&root->delalloc_lock);
4683 
4684 		/*
4685 		 * Make sure we get a live inode and that it'll not disappear
4686 		 * meanwhile.
4687 		 */
4688 		inode = igrab(&btrfs_inode->vfs_inode);
4689 		if (inode) {
4690 			unsigned int nofs_flag;
4691 
4692 			nofs_flag = memalloc_nofs_save();
4693 			invalidate_inode_pages2(inode->i_mapping);
4694 			memalloc_nofs_restore(nofs_flag);
4695 			iput(inode);
4696 		}
4697 		spin_lock(&root->delalloc_lock);
4698 	}
4699 	spin_unlock(&root->delalloc_lock);
4700 }
4701 
btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info * fs_info)4702 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4703 {
4704 	struct btrfs_root *root;
4705 	LIST_HEAD(splice);
4706 
4707 	spin_lock(&fs_info->delalloc_root_lock);
4708 	list_splice_init(&fs_info->delalloc_roots, &splice);
4709 	while (!list_empty(&splice)) {
4710 		root = list_first_entry(&splice, struct btrfs_root,
4711 					 delalloc_root);
4712 		root = btrfs_grab_root(root);
4713 		BUG_ON(!root);
4714 		spin_unlock(&fs_info->delalloc_root_lock);
4715 
4716 		btrfs_destroy_delalloc_inodes(root);
4717 		btrfs_put_root(root);
4718 
4719 		spin_lock(&fs_info->delalloc_root_lock);
4720 	}
4721 	spin_unlock(&fs_info->delalloc_root_lock);
4722 }
4723 
btrfs_destroy_marked_extents(struct btrfs_fs_info * fs_info,struct extent_io_tree * dirty_pages,int mark)4724 static void btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4725 					 struct extent_io_tree *dirty_pages,
4726 					 int mark)
4727 {
4728 	struct extent_buffer *eb;
4729 	u64 start = 0;
4730 	u64 end;
4731 
4732 	while (find_first_extent_bit(dirty_pages, start, &start, &end,
4733 				     mark, NULL)) {
4734 		clear_extent_bits(dirty_pages, start, end, mark);
4735 		while (start <= end) {
4736 			eb = find_extent_buffer(fs_info, start);
4737 			start += fs_info->nodesize;
4738 			if (!eb)
4739 				continue;
4740 
4741 			btrfs_tree_lock(eb);
4742 			wait_on_extent_buffer_writeback(eb);
4743 			btrfs_clear_buffer_dirty(NULL, eb);
4744 			btrfs_tree_unlock(eb);
4745 
4746 			free_extent_buffer_stale(eb);
4747 		}
4748 	}
4749 }
4750 
btrfs_destroy_pinned_extent(struct btrfs_fs_info * fs_info,struct extent_io_tree * unpin)4751 static void btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4752 					struct extent_io_tree *unpin)
4753 {
4754 	u64 start;
4755 	u64 end;
4756 
4757 	while (1) {
4758 		struct extent_state *cached_state = NULL;
4759 
4760 		/*
4761 		 * The btrfs_finish_extent_commit() may get the same range as
4762 		 * ours between find_first_extent_bit and clear_extent_dirty.
4763 		 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4764 		 * the same extent range.
4765 		 */
4766 		mutex_lock(&fs_info->unused_bg_unpin_mutex);
4767 		if (!find_first_extent_bit(unpin, 0, &start, &end,
4768 					   EXTENT_DIRTY, &cached_state)) {
4769 			mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4770 			break;
4771 		}
4772 
4773 		clear_extent_dirty(unpin, start, end, &cached_state);
4774 		free_extent_state(cached_state);
4775 		btrfs_error_unpin_extent_range(fs_info, start, end);
4776 		mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4777 		cond_resched();
4778 	}
4779 }
4780 
btrfs_cleanup_bg_io(struct btrfs_block_group * cache)4781 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4782 {
4783 	struct inode *inode;
4784 
4785 	inode = cache->io_ctl.inode;
4786 	if (inode) {
4787 		unsigned int nofs_flag;
4788 
4789 		nofs_flag = memalloc_nofs_save();
4790 		invalidate_inode_pages2(inode->i_mapping);
4791 		memalloc_nofs_restore(nofs_flag);
4792 
4793 		BTRFS_I(inode)->generation = 0;
4794 		cache->io_ctl.inode = NULL;
4795 		iput(inode);
4796 	}
4797 	ASSERT(cache->io_ctl.pages == NULL);
4798 	btrfs_put_block_group(cache);
4799 }
4800 
btrfs_cleanup_dirty_bgs(struct btrfs_transaction * cur_trans,struct btrfs_fs_info * fs_info)4801 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4802 			     struct btrfs_fs_info *fs_info)
4803 {
4804 	struct btrfs_block_group *cache;
4805 
4806 	spin_lock(&cur_trans->dirty_bgs_lock);
4807 	while (!list_empty(&cur_trans->dirty_bgs)) {
4808 		cache = list_first_entry(&cur_trans->dirty_bgs,
4809 					 struct btrfs_block_group,
4810 					 dirty_list);
4811 
4812 		if (!list_empty(&cache->io_list)) {
4813 			spin_unlock(&cur_trans->dirty_bgs_lock);
4814 			list_del_init(&cache->io_list);
4815 			btrfs_cleanup_bg_io(cache);
4816 			spin_lock(&cur_trans->dirty_bgs_lock);
4817 		}
4818 
4819 		list_del_init(&cache->dirty_list);
4820 		spin_lock(&cache->lock);
4821 		cache->disk_cache_state = BTRFS_DC_ERROR;
4822 		spin_unlock(&cache->lock);
4823 
4824 		spin_unlock(&cur_trans->dirty_bgs_lock);
4825 		btrfs_put_block_group(cache);
4826 		btrfs_delayed_refs_rsv_release(fs_info, 1);
4827 		spin_lock(&cur_trans->dirty_bgs_lock);
4828 	}
4829 	spin_unlock(&cur_trans->dirty_bgs_lock);
4830 
4831 	/*
4832 	 * Refer to the definition of io_bgs member for details why it's safe
4833 	 * to use it without any locking
4834 	 */
4835 	while (!list_empty(&cur_trans->io_bgs)) {
4836 		cache = list_first_entry(&cur_trans->io_bgs,
4837 					 struct btrfs_block_group,
4838 					 io_list);
4839 
4840 		list_del_init(&cache->io_list);
4841 		spin_lock(&cache->lock);
4842 		cache->disk_cache_state = BTRFS_DC_ERROR;
4843 		spin_unlock(&cache->lock);
4844 		btrfs_cleanup_bg_io(cache);
4845 	}
4846 }
4847 
btrfs_free_all_qgroup_pertrans(struct btrfs_fs_info * fs_info)4848 static void btrfs_free_all_qgroup_pertrans(struct btrfs_fs_info *fs_info)
4849 {
4850 	struct btrfs_root *gang[8];
4851 	int i;
4852 	int ret;
4853 
4854 	spin_lock(&fs_info->fs_roots_radix_lock);
4855 	while (1) {
4856 		ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
4857 						 (void **)gang, 0,
4858 						 ARRAY_SIZE(gang),
4859 						 BTRFS_ROOT_TRANS_TAG);
4860 		if (ret == 0)
4861 			break;
4862 		for (i = 0; i < ret; i++) {
4863 			struct btrfs_root *root = gang[i];
4864 
4865 			btrfs_qgroup_free_meta_all_pertrans(root);
4866 			radix_tree_tag_clear(&fs_info->fs_roots_radix,
4867 					(unsigned long)root->root_key.objectid,
4868 					BTRFS_ROOT_TRANS_TAG);
4869 		}
4870 	}
4871 	spin_unlock(&fs_info->fs_roots_radix_lock);
4872 }
4873 
btrfs_cleanup_one_transaction(struct btrfs_transaction * cur_trans,struct btrfs_fs_info * fs_info)4874 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4875 				   struct btrfs_fs_info *fs_info)
4876 {
4877 	struct btrfs_device *dev, *tmp;
4878 
4879 	btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4880 	ASSERT(list_empty(&cur_trans->dirty_bgs));
4881 	ASSERT(list_empty(&cur_trans->io_bgs));
4882 
4883 	list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4884 				 post_commit_list) {
4885 		list_del_init(&dev->post_commit_list);
4886 	}
4887 
4888 	btrfs_destroy_delayed_refs(cur_trans, fs_info);
4889 
4890 	cur_trans->state = TRANS_STATE_COMMIT_START;
4891 	wake_up(&fs_info->transaction_blocked_wait);
4892 
4893 	cur_trans->state = TRANS_STATE_UNBLOCKED;
4894 	wake_up(&fs_info->transaction_wait);
4895 
4896 	btrfs_destroy_delayed_inodes(fs_info);
4897 
4898 	btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4899 				     EXTENT_DIRTY);
4900 	btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4901 
4902 	btrfs_free_all_qgroup_pertrans(fs_info);
4903 
4904 	cur_trans->state =TRANS_STATE_COMPLETED;
4905 	wake_up(&cur_trans->commit_wait);
4906 }
4907 
btrfs_cleanup_transaction(struct btrfs_fs_info * fs_info)4908 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4909 {
4910 	struct btrfs_transaction *t;
4911 
4912 	mutex_lock(&fs_info->transaction_kthread_mutex);
4913 
4914 	spin_lock(&fs_info->trans_lock);
4915 	while (!list_empty(&fs_info->trans_list)) {
4916 		t = list_first_entry(&fs_info->trans_list,
4917 				     struct btrfs_transaction, list);
4918 		if (t->state >= TRANS_STATE_COMMIT_PREP) {
4919 			refcount_inc(&t->use_count);
4920 			spin_unlock(&fs_info->trans_lock);
4921 			btrfs_wait_for_commit(fs_info, t->transid);
4922 			btrfs_put_transaction(t);
4923 			spin_lock(&fs_info->trans_lock);
4924 			continue;
4925 		}
4926 		if (t == fs_info->running_transaction) {
4927 			t->state = TRANS_STATE_COMMIT_DOING;
4928 			spin_unlock(&fs_info->trans_lock);
4929 			/*
4930 			 * We wait for 0 num_writers since we don't hold a trans
4931 			 * handle open currently for this transaction.
4932 			 */
4933 			wait_event(t->writer_wait,
4934 				   atomic_read(&t->num_writers) == 0);
4935 		} else {
4936 			spin_unlock(&fs_info->trans_lock);
4937 		}
4938 		btrfs_cleanup_one_transaction(t, fs_info);
4939 
4940 		spin_lock(&fs_info->trans_lock);
4941 		if (t == fs_info->running_transaction)
4942 			fs_info->running_transaction = NULL;
4943 		list_del_init(&t->list);
4944 		spin_unlock(&fs_info->trans_lock);
4945 
4946 		btrfs_put_transaction(t);
4947 		trace_btrfs_transaction_commit(fs_info);
4948 		spin_lock(&fs_info->trans_lock);
4949 	}
4950 	spin_unlock(&fs_info->trans_lock);
4951 	btrfs_destroy_all_ordered_extents(fs_info);
4952 	btrfs_destroy_delayed_inodes(fs_info);
4953 	btrfs_assert_delayed_root_empty(fs_info);
4954 	btrfs_destroy_all_delalloc_inodes(fs_info);
4955 	btrfs_drop_all_logs(fs_info);
4956 	mutex_unlock(&fs_info->transaction_kthread_mutex);
4957 
4958 	return 0;
4959 }
4960 
btrfs_init_root_free_objectid(struct btrfs_root * root)4961 int btrfs_init_root_free_objectid(struct btrfs_root *root)
4962 {
4963 	struct btrfs_path *path;
4964 	int ret;
4965 	struct extent_buffer *l;
4966 	struct btrfs_key search_key;
4967 	struct btrfs_key found_key;
4968 	int slot;
4969 
4970 	path = btrfs_alloc_path();
4971 	if (!path)
4972 		return -ENOMEM;
4973 
4974 	search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4975 	search_key.type = -1;
4976 	search_key.offset = (u64)-1;
4977 	ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4978 	if (ret < 0)
4979 		goto error;
4980 	BUG_ON(ret == 0); /* Corruption */
4981 	if (path->slots[0] > 0) {
4982 		slot = path->slots[0] - 1;
4983 		l = path->nodes[0];
4984 		btrfs_item_key_to_cpu(l, &found_key, slot);
4985 		root->free_objectid = max_t(u64, found_key.objectid + 1,
4986 					    BTRFS_FIRST_FREE_OBJECTID);
4987 	} else {
4988 		root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4989 	}
4990 	ret = 0;
4991 error:
4992 	btrfs_free_path(path);
4993 	return ret;
4994 }
4995 
btrfs_get_free_objectid(struct btrfs_root * root,u64 * objectid)4996 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4997 {
4998 	int ret;
4999 	mutex_lock(&root->objectid_mutex);
5000 
5001 	if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
5002 		btrfs_warn(root->fs_info,
5003 			   "the objectid of root %llu reaches its highest value",
5004 			   root->root_key.objectid);
5005 		ret = -ENOSPC;
5006 		goto out;
5007 	}
5008 
5009 	*objectid = root->free_objectid++;
5010 	ret = 0;
5011 out:
5012 	mutex_unlock(&root->objectid_mutex);
5013 	return ret;
5014 }
5015