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 <linux/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 "dev-replace.h"
33	#include "raid56.h"
34	#include "sysfs.h"
35	#include "qgroup.h"
36	#include "compression.h"
37	#include "tree-checker.h"
38	#include "ref-verify.h"
39	#include "block-group.h"
40	#include "discard.h"
41	#include "space-info.h"
42	#include "zoned.h"
43	#include "subpage.h"
44	#include "fs.h"
45	#include "accessors.h"
46	#include "extent-tree.h"
47	#include "root-tree.h"
48	#include "defrag.h"
49	#include "uuid-tree.h"
50	#include "relocation.h"
51	#include "scrub.h"
52	#include "super.h"
53
54	#define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
55	BTRFS_HEADER_FLAG_RELOC |\
56	BTRFS_SUPER_FLAG_ERROR |\
57	BTRFS_SUPER_FLAG_SEEDING |\
58	BTRFS_SUPER_FLAG_METADUMP |\
59	BTRFS_SUPER_FLAG_METADUMP_V2)
60
61	static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
62	static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
63
64	static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
65	{
66	if (fs_info->csum_shash)
67	crypto_free_shash(tfm: fs_info->csum_shash);
68	}
69
70	/*
71	* Compute the csum of a btree block and store the result to provided buffer.
72	*/
73	static void csum_tree_block(struct extent_buffer *buf, u8 *result)
74	{
75	struct btrfs_fs_info *fs_info = buf->fs_info;
76	int num_pages;
77	u32 first_page_part;
78	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
79	char *kaddr;
80	int i;
81
82	shash->tfm = fs_info->csum_shash;
83	crypto_shash_init(desc: shash);
84
85	if (buf->addr) {
86	/* Pages are contiguous, handle them as a big one. */
87	kaddr = buf->addr;
88	first_page_part = fs_info->nodesize;
89	num_pages = 1;
90	} else {
91	kaddr = folio_address(folio: buf->folios[0]);
92	first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
93	num_pages = num_extent_pages(eb: buf);
94	}
95
96	crypto_shash_update(desc: shash, data: kaddr + BTRFS_CSUM_SIZE,
97	len: first_page_part - BTRFS_CSUM_SIZE);
98
99	/*
100	* Multiple single-page folios case would reach here.
101	*
102	* nodesize <= PAGE_SIZE and large folio all handled by above
103	* crypto_shash_update() already.
104	*/
105	for (i = 1; i < num_pages && INLINE_EXTENT_BUFFER_PAGES > 1; i++) {
106	kaddr = folio_address(folio: buf->folios[i]);
107	crypto_shash_update(desc: shash, data: kaddr, PAGE_SIZE);
108	}
109	memset(result, 0, BTRFS_CSUM_SIZE);
110	crypto_shash_final(desc: shash, out: result);
111	}
112
113	/*
114	* we can't consider a given block up to date unless the transid of the
115	* block matches the transid in the parent node's pointer. This is how we
116	* detect blocks that either didn't get written at all or got written
117	* in the wrong place.
118	*/
119	int btrfs_buffer_uptodate(struct extent_buffer *eb, u64 parent_transid, int atomic)
120	{
121	if (!extent_buffer_uptodate(eb))
122	return 0;
123
124	if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
125	return 1;
126
127	if (atomic)
128	return -EAGAIN;
129
130	if (!extent_buffer_uptodate(eb) ||
131	btrfs_header_generation(eb) != parent_transid) {
132	btrfs_err_rl(eb->fs_info,
133	"parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
134	eb->start, eb->read_mirror,
135	parent_transid, btrfs_header_generation(eb));
136	clear_extent_buffer_uptodate(eb);
137	return 0;
138	}
139	return 1;
140	}
141
142	static bool btrfs_supported_super_csum(u16 csum_type)
143	{
144	switch (csum_type) {
145	case BTRFS_CSUM_TYPE_CRC32:
146	case BTRFS_CSUM_TYPE_XXHASH:
147	case BTRFS_CSUM_TYPE_SHA256:
148	case BTRFS_CSUM_TYPE_BLAKE2:
149	return true;
150	default:
151	return false;
152	}
153	}
154
155	/*
156	* Return 0 if the superblock checksum type matches the checksum value of that
157	* algorithm. Pass the raw disk superblock data.
158	*/
159	int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
160	const struct btrfs_super_block *disk_sb)
161	{
162	char result[BTRFS_CSUM_SIZE];
163	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
164
165	shash->tfm = fs_info->csum_shash;
166
167	/*
168	* The super_block structure does not span the whole
169	* BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
170	* filled with zeros and is included in the checksum.
171	*/
172	crypto_shash_digest(desc: shash, data: (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
173	BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, out: result);
174
175	if (memcmp(p: disk_sb->csum, q: result, size: fs_info->csum_size))
176	return 1;
177
178	return 0;
179	}
180
181	static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
182	int mirror_num)
183	{
184	struct btrfs_fs_info *fs_info = eb->fs_info;
185	int ret = 0;
186
187	if (sb_rdonly(sb: fs_info->sb))
188	return -EROFS;
189
190	for (int i = 0; i < num_extent_folios(eb); i++) {
191	struct folio *folio = eb->folios[i];
192	u64 start = max_t(u64, eb->start, folio_pos(folio));
193	u64 end = min_t(u64, eb->start + eb->len,
194	folio_pos(folio) + eb->folio_size);
195	u32 len = end - start;
196	phys_addr_t paddr = PFN_PHYS(folio_pfn(folio)) +
197	offset_in_folio(folio, start);
198
199	ret = btrfs_repair_io_failure(fs_info, ino: 0, start, length: len, logical: start,
200	paddr, mirror_num);
201	if (ret)
202	break;
203	}
204
205	return ret;
206	}
207
208	/*
209	* helper to read a given tree block, doing retries as required when
210	* the checksums don't match and we have alternate mirrors to try.
211	*
212	* @check: expected tree parentness check, see the comments of the
213	* structure for details.
214	*/
215	int btrfs_read_extent_buffer(struct extent_buffer *eb,
216	const struct btrfs_tree_parent_check *check)
217	{
218	struct btrfs_fs_info *fs_info = eb->fs_info;
219	int failed = 0;
220	int ret;
221	int num_copies = 0;
222	int mirror_num = 0;
223	int failed_mirror = 0;
224
225	ASSERT(check);
226
227	while (1) {
228	ret = read_extent_buffer_pages(eb, mirror_num, parent_check: check);
229	if (!ret)
230	break;
231
232	num_copies = btrfs_num_copies(fs_info,
233	logical: eb->start, len: eb->len);
234	if (num_copies == 1)
235	break;
236
237	if (!failed_mirror) {
238	failed = 1;
239	failed_mirror = eb->read_mirror;
240	}
241
242	mirror_num++;
243	if (mirror_num == failed_mirror)
244	mirror_num++;
245
246	if (mirror_num > num_copies)
247	break;
248	}
249
250	if (failed && !ret && failed_mirror)
251	btrfs_repair_eb_io_failure(eb, mirror_num: failed_mirror);
252
253	return ret;
254	}
255
256	/*
257	* Checksum a dirty tree block before IO.
258	*/
259	int btree_csum_one_bio(struct btrfs_bio *bbio)
260	{
261	struct extent_buffer *eb = bbio->private;
262	struct btrfs_fs_info *fs_info = eb->fs_info;
263	u64 found_start = btrfs_header_bytenr(eb);
264	u64 last_trans;
265	u8 result[BTRFS_CSUM_SIZE];
266	int ret;
267
268	/* Btree blocks are always contiguous on disk. */
269	if (WARN_ON_ONCE(bbio->file_offset != eb->start))
270	return -EIO;
271	if (WARN_ON_ONCE(bbio->bio.bi_iter.bi_size != eb->len))
272	return -EIO;
273
274	/*
275	* If an extent_buffer is marked as EXTENT_BUFFER_ZONED_ZEROOUT, don't
276	* checksum it but zero-out its content. This is done to preserve
277	* ordering of I/O without unnecessarily writing out data.
278	*/
279	if (test_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags)) {
280	memzero_extent_buffer(eb, start: 0, len: eb->len);
281	return 0;
282	}
283
284	if (WARN_ON_ONCE(found_start != eb->start))
285	return -EIO;
286	if (WARN_ON(!btrfs_meta_folio_test_uptodate(eb->folios[0], eb)))
287	return -EIO;
288
289	ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
290	offsetof(struct btrfs_header, fsid),
291	BTRFS_FSID_SIZE) == 0);
292	csum_tree_block(buf: eb, result);
293
294	if (btrfs_header_level(eb))
295	ret = btrfs_check_node(node: eb);
296	else
297	ret = btrfs_check_leaf(leaf: eb);
298
299	if (ret < 0)
300	goto error;
301
302	/*
303	* Also check the generation, the eb reached here must be newer than
304	* last committed. Or something seriously wrong happened.
305	*/
306	last_trans = btrfs_get_last_trans_committed(fs_info);
307	if (unlikely(btrfs_header_generation(eb) <= last_trans)) {
308	ret = -EUCLEAN;
309	btrfs_err(fs_info,
310	"block=%llu bad generation, have %llu expect > %llu",
311	eb->start, btrfs_header_generation(eb), last_trans);
312	goto error;
313	}
314	write_extent_buffer(eb, src: result, start: 0, len: fs_info->csum_size);
315	return 0;
316
317	error:
318	btrfs_print_tree(c: eb, follow: 0);
319	btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
320	eb->start);
321	/*
322	* Be noisy if this is an extent buffer from a log tree. We don't abort
323	* a transaction in case there's a bad log tree extent buffer, we just
324	* fallback to a transaction commit. Still we want to know when there is
325	* a bad log tree extent buffer, as that may signal a bug somewhere.
326	*/
327	WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
328	btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
329	return ret;
330	}
331
332	static bool check_tree_block_fsid(struct extent_buffer *eb)
333	{
334	struct btrfs_fs_info *fs_info = eb->fs_info;
335	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
336	u8 fsid[BTRFS_FSID_SIZE];
337
338	read_extent_buffer(eb, dst: fsid, offsetof(struct btrfs_header, fsid),
339	BTRFS_FSID_SIZE);
340
341	/*
342	* alloc_fsid_devices() copies the fsid into fs_devices::metadata_uuid.
343	* This is then overwritten by metadata_uuid if it is present in the
344	* device_list_add(). The same true for a seed device as well. So use of
345	* fs_devices::metadata_uuid is appropriate here.
346	*/
347	if (memcmp(p: fsid, q: fs_info->fs_devices->metadata_uuid, BTRFS_FSID_SIZE) == 0)
348	return false;
349
350	list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
351	if (!memcmp(p: fsid, q: seed_devs->fsid, BTRFS_FSID_SIZE))
352	return false;
353
354	return true;
355	}
356
357	/* Do basic extent buffer checks at read time */
358	int btrfs_validate_extent_buffer(struct extent_buffer *eb,
359	const struct btrfs_tree_parent_check *check)
360	{
361	struct btrfs_fs_info *fs_info = eb->fs_info;
362	u64 found_start;
363	const u32 csum_size = fs_info->csum_size;
364	u8 found_level;
365	u8 result[BTRFS_CSUM_SIZE];
366	const u8 *header_csum;
367	int ret = 0;
368	const bool ignore_csum = btrfs_test_opt(fs_info, IGNOREMETACSUMS);
369
370	ASSERT(check);
371
372	found_start = btrfs_header_bytenr(eb);
373	if (found_start != eb->start) {
374	btrfs_err_rl(fs_info,
375	"bad tree block start, mirror %u want %llu have %llu",
376	eb->read_mirror, eb->start, found_start);
377	ret = -EIO;
378	goto out;
379	}
380	if (check_tree_block_fsid(eb)) {
381	btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
382	eb->start, eb->read_mirror);
383	ret = -EIO;
384	goto out;
385	}
386	found_level = btrfs_header_level(eb);
387	if (found_level >= BTRFS_MAX_LEVEL) {
388	btrfs_err(fs_info,
389	"bad tree block level, mirror %u level %d on logical %llu",
390	eb->read_mirror, btrfs_header_level(eb), eb->start);
391	ret = -EIO;
392	goto out;
393	}
394
395	csum_tree_block(buf: eb, result);
396	header_csum = folio_address(folio: eb->folios[0]) +
397	get_eb_offset_in_folio(eb, offsetof(struct btrfs_header, csum));
398
399	if (memcmp(p: result, q: header_csum, size: csum_size) != 0) {
400	btrfs_warn_rl(fs_info,
401	"checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d%s",
402	eb->start, eb->read_mirror,
403	CSUM_FMT_VALUE(csum_size, header_csum),
404	CSUM_FMT_VALUE(csum_size, result),
405	btrfs_header_level(eb),
406	ignore_csum ? ", ignored" : "");
407	if (!ignore_csum) {
408	ret = -EUCLEAN;
409	goto out;
410	}
411	}
412
413	if (found_level != check->level) {
414	btrfs_err(fs_info,
415	"level verify failed on logical %llu mirror %u wanted %u found %u",
416	eb->start, eb->read_mirror, check->level, found_level);
417	ret = -EIO;
418	goto out;
419	}
420	if (unlikely(check->transid &&
421	btrfs_header_generation(eb) != check->transid)) {
422	btrfs_err_rl(eb->fs_info,
423	"parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
424	eb->start, eb->read_mirror, check->transid,
425	btrfs_header_generation(eb));
426	ret = -EIO;
427	goto out;
428	}
429	if (check->has_first_key) {
430	const struct btrfs_key *expect_key = &check->first_key;
431	struct btrfs_key found_key;
432
433	if (found_level)
434	btrfs_node_key_to_cpu(eb, cpu_key: &found_key, nr: 0);
435	else
436	btrfs_item_key_to_cpu(eb, cpu_key: &found_key, nr: 0);
437	if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
438	btrfs_err(fs_info,
439	"tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
440	eb->start, check->transid,
441	expect_key->objectid,
442	expect_key->type, expect_key->offset,
443	found_key.objectid, found_key.type,
444	found_key.offset);
445	ret = -EUCLEAN;
446	goto out;
447	}
448	}
449	if (check->owner_root) {
450	ret = btrfs_check_eb_owner(eb, root_owner: check->owner_root);
451	if (ret < 0)
452	goto out;
453	}
454
455	/* If this is a leaf block and it is corrupt, just return -EIO. */
456	if (found_level == 0 && btrfs_check_leaf(leaf: eb))
457	ret = -EIO;
458
459	if (found_level > 0 && btrfs_check_node(node: eb))
460	ret = -EIO;
461
462	if (ret)
463	btrfs_err(fs_info,
464	"read time tree block corruption detected on logical %llu mirror %u",
465	eb->start, eb->read_mirror);
466	out:
467	return ret;
468	}
469
470	#ifdef CONFIG_MIGRATION
471	static int btree_migrate_folio(struct address_space *mapping,
472	struct folio *dst, struct folio *src, enum migrate_mode mode)
473	{
474	/*
475	* we can't safely write a btree page from here,
476	* we haven't done the locking hook
477	*/
478	if (folio_test_dirty(folio: src))
479	return -EAGAIN;
480	/*
481	* Buffers may be managed in a filesystem specific way.
482	* We must have no buffers or drop them.
483	*/
484	if (folio_get_private(folio: src) &&
485	!filemap_release_folio(folio: src, GFP_KERNEL))
486	return -EAGAIN;
487	return migrate_folio(mapping, dst, src, mode);
488	}
489	#else
490	#define btree_migrate_folio NULL
491	#endif
492
493	static int btree_writepages(struct address_space *mapping,
494	struct writeback_control *wbc)
495	{
496	int ret;
497
498	if (wbc->sync_mode == WB_SYNC_NONE) {
499	struct btrfs_fs_info *fs_info;
500
501	if (wbc->for_kupdate)
502	return 0;
503
504	fs_info = inode_to_fs_info(mapping->host);
505	/* this is a bit racy, but that's ok */
506	ret = __percpu_counter_compare(fbc: &fs_info->dirty_metadata_bytes,
507	BTRFS_DIRTY_METADATA_THRESH,
508	batch: fs_info->dirty_metadata_batch);
509	if (ret < 0)
510	return 0;
511	}
512	return btree_write_cache_pages(mapping, wbc);
513	}
514
515	static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
516	{
517	if (folio_test_writeback(folio) || folio_test_dirty(folio))
518	return false;
519
520	return try_release_extent_buffer(folio);
521	}
522
523	static void btree_invalidate_folio(struct folio *folio, size_t offset,
524	size_t length)
525	{
526	struct extent_io_tree *tree;
527
528	tree = &folio_to_inode(folio)->io_tree;
529	extent_invalidate_folio(tree, folio, offset);
530	btree_release_folio(folio, GFP_NOFS);
531	if (folio_get_private(folio)) {
532	btrfs_warn(folio_to_fs_info(folio),
533	"folio private not zero on folio %llu",
534	(unsigned long long)folio_pos(folio));
535	folio_detach_private(folio);
536	}
537	}
538
539	#ifdef DEBUG
540	static bool btree_dirty_folio(struct address_space *mapping,
541	struct folio *folio)
542	{
543	struct btrfs_fs_info *fs_info = inode_to_fs_info(mapping->host);
544	struct btrfs_subpage_info *spi = fs_info->subpage_info;
545	struct btrfs_subpage *subpage;
546	struct extent_buffer *eb;
547	int cur_bit = 0;
548	u64 page_start = folio_pos(folio);
549
550	if (fs_info->sectorsize == PAGE_SIZE) {
551	eb = folio_get_private(folio);
552	BUG_ON(!eb);
553	BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
554	BUG_ON(!atomic_read(&eb->refs));
555	btrfs_assert_tree_write_locked(eb);
556	return filemap_dirty_folio(mapping, folio);
557	}
558
559	ASSERT(spi);
560	subpage = folio_get_private(folio);
561
562	for (cur_bit = spi->dirty_offset;
563	cur_bit < spi->dirty_offset + spi->bitmap_nr_bits;
564	cur_bit++) {
565	unsigned long flags;
566	u64 cur;
567
568	spin_lock_irqsave(&subpage->lock, flags);
569	if (!test_bit(cur_bit, subpage->bitmaps)) {
570	spin_unlock_irqrestore(&subpage->lock, flags);
571	continue;
572	}
573	spin_unlock_irqrestore(&subpage->lock, flags);
574	cur = page_start + cur_bit * fs_info->sectorsize;
575
576	eb = find_extent_buffer(fs_info, cur);
577	ASSERT(eb);
578	ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
579	ASSERT(atomic_read(&eb->refs));
580	btrfs_assert_tree_write_locked(eb);
581	free_extent_buffer(eb);
582
583	cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits) - 1;
584	}
585	return filemap_dirty_folio(mapping, folio);
586	}
587	#else
588	#define btree_dirty_folio filemap_dirty_folio
589	#endif
590
591	static const struct address_space_operations btree_aops = {
592	.writepages = btree_writepages,
593	.release_folio = btree_release_folio,
594	.invalidate_folio = btree_invalidate_folio,
595	.migrate_folio = btree_migrate_folio,
596	.dirty_folio = btree_dirty_folio,
597	};
598
599	struct extent_buffer *btrfs_find_create_tree_block(
600	struct btrfs_fs_info *fs_info,
601	u64 bytenr, u64 owner_root,
602	int level)
603	{
604	if (btrfs_is_testing(fs_info))
605	return alloc_test_extent_buffer(fs_info, start: bytenr);
606	return alloc_extent_buffer(fs_info, start: bytenr, owner_root, level);
607	}
608
609	/*
610	* Read tree block at logical address @bytenr and do variant basic but critical
611	* verification.
612	*
613	* @check: expected tree parentness check, see comments of the
614	* structure for details.
615	*/
616	struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
617	struct btrfs_tree_parent_check *check)
618	{
619	struct extent_buffer *buf = NULL;
620	int ret;
621
622	ASSERT(check);
623
624	buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root: check->owner_root,
625	level: check->level);
626	if (IS_ERR(ptr: buf))
627	return buf;
628
629	ret = btrfs_read_extent_buffer(eb: buf, check);
630	if (ret) {
631	free_extent_buffer_stale(eb: buf);
632	return ERR_PTR(error: ret);
633	}
634	return buf;
635
636	}
637
638	static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
639	u64 objectid, gfp_t flags)
640	{
641	struct btrfs_root *root;
642	bool dummy = btrfs_is_testing(fs_info);
643
644	root = kzalloc(sizeof(*root), flags);
645	if (!root)
646	return NULL;
647
648	memset(&root->root_key, 0, sizeof(root->root_key));
649	memset(&root->root_item, 0, sizeof(root->root_item));
650	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
651	root->fs_info = fs_info;
652	root->root_key.objectid = objectid;
653	root->node = NULL;
654	root->commit_root = NULL;
655	root->state = 0;
656	RB_CLEAR_NODE(&root->rb_node);
657
658	btrfs_set_root_last_trans(root, transid: 0);
659	root->free_objectid = 0;
660	root->nr_delalloc_inodes = 0;
661	root->nr_ordered_extents = 0;
662	xa_init(xa: &root->inodes);
663	xa_init(xa: &root->delayed_nodes);
664
665	btrfs_init_root_block_rsv(root);
666
667	INIT_LIST_HEAD(list: &root->dirty_list);
668	INIT_LIST_HEAD(list: &root->root_list);
669	INIT_LIST_HEAD(list: &root->delalloc_inodes);
670	INIT_LIST_HEAD(list: &root->delalloc_root);
671	INIT_LIST_HEAD(list: &root->ordered_extents);
672	INIT_LIST_HEAD(list: &root->ordered_root);
673	INIT_LIST_HEAD(list: &root->reloc_dirty_list);
674	spin_lock_init(&root->delalloc_lock);
675	spin_lock_init(&root->ordered_extent_lock);
676	spin_lock_init(&root->accounting_lock);
677	spin_lock_init(&root->qgroup_meta_rsv_lock);
678	mutex_init(&root->objectid_mutex);
679	mutex_init(&root->log_mutex);
680	mutex_init(&root->ordered_extent_mutex);
681	mutex_init(&root->delalloc_mutex);
682	init_waitqueue_head(&root->qgroup_flush_wait);
683	init_waitqueue_head(&root->log_writer_wait);
684	init_waitqueue_head(&root->log_commit_wait[0]);
685	init_waitqueue_head(&root->log_commit_wait[1]);
686	INIT_LIST_HEAD(list: &root->log_ctxs[0]);
687	INIT_LIST_HEAD(list: &root->log_ctxs[1]);
688	atomic_set(v: &root->log_commit[0], i: 0);
689	atomic_set(v: &root->log_commit[1], i: 0);
690	atomic_set(v: &root->log_writers, i: 0);
691	atomic_set(v: &root->log_batch, i: 0);
692	refcount_set(r: &root->refs, n: 1);
693	atomic_set(v: &root->snapshot_force_cow, i: 0);
694	atomic_set(v: &root->nr_swapfiles, i: 0);
695	btrfs_set_root_log_transid(root, log_transid: 0);
696	root->log_transid_committed = -1;
697	btrfs_set_root_last_log_commit(root, commit_id: 0);
698	root->anon_dev = 0;
699	if (!dummy) {
700	btrfs_extent_io_tree_init(fs_info, tree: &root->dirty_log_pages,
701	owner: IO_TREE_ROOT_DIRTY_LOG_PAGES);
702	btrfs_extent_io_tree_init(fs_info, tree: &root->log_csum_range,
703	owner: IO_TREE_LOG_CSUM_RANGE);
704	}
705
706	spin_lock_init(&root->root_item_lock);
707	btrfs_qgroup_init_swapped_blocks(swapped_blocks: &root->swapped_blocks);
708	#ifdef CONFIG_BTRFS_DEBUG
709	INIT_LIST_HEAD(list: &root->leak_list);
710	spin_lock(lock: &fs_info->fs_roots_radix_lock);
711	list_add_tail(new: &root->leak_list, head: &fs_info->allocated_roots);
712	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
713	#endif
714
715	return root;
716	}
717
718	#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
719	/* Should only be used by the testing infrastructure */
720	struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
721	{
722	struct btrfs_root *root;
723
724	if (!fs_info)
725	return ERR_PTR(error: -EINVAL);
726
727	root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
728	if (!root)
729	return ERR_PTR(error: -ENOMEM);
730
731	/* We don't use the stripesize in selftest, set it as sectorsize */
732	root->alloc_bytenr = 0;
733
734	return root;
735	}
736	#endif
737
738	static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
739	{
740	const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
741	const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
742
743	return btrfs_comp_cpu_keys(k1: &a->root_key, k2: &b->root_key);
744	}
745
746	static int global_root_key_cmp(const void *k, const struct rb_node *node)
747	{
748	const struct btrfs_key *key = k;
749	const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
750
751	return btrfs_comp_cpu_keys(k1: key, k2: &root->root_key);
752	}
753
754	int btrfs_global_root_insert(struct btrfs_root *root)
755	{
756	struct btrfs_fs_info *fs_info = root->fs_info;
757	struct rb_node *tmp;
758	int ret = 0;
759
760	write_lock(&fs_info->global_root_lock);
761	tmp = rb_find_add(node: &root->rb_node, tree: &fs_info->global_root_tree, cmp: global_root_cmp);
762	write_unlock(&fs_info->global_root_lock);
763
764	if (tmp) {
765	ret = -EEXIST;
766	btrfs_warn(fs_info, "global root %llu %llu already exists",
767	btrfs_root_id(root), root->root_key.offset);
768	}
769	return ret;
770	}
771
772	void btrfs_global_root_delete(struct btrfs_root *root)
773	{
774	struct btrfs_fs_info *fs_info = root->fs_info;
775
776	write_lock(&fs_info->global_root_lock);
777	rb_erase(&root->rb_node, &fs_info->global_root_tree);
778	write_unlock(&fs_info->global_root_lock);
779	}
780
781	struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
782	struct btrfs_key *key)
783	{
784	struct rb_node *node;
785	struct btrfs_root *root = NULL;
786
787	read_lock(&fs_info->global_root_lock);
788	node = rb_find(key, tree: &fs_info->global_root_tree, cmp: global_root_key_cmp);
789	if (node)
790	root = container_of(node, struct btrfs_root, rb_node);
791	read_unlock(&fs_info->global_root_lock);
792
793	return root;
794	}
795
796	static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
797	{
798	struct btrfs_block_group *block_group;
799	u64 ret;
800
801	if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
802	return 0;
803
804	if (bytenr)
805	block_group = btrfs_lookup_block_group(info: fs_info, bytenr);
806	else
807	block_group = btrfs_lookup_first_block_group(info: fs_info, bytenr);
808	ASSERT(block_group);
809	if (!block_group)
810	return 0;
811	ret = block_group->global_root_id;
812	btrfs_put_block_group(cache: block_group);
813
814	return ret;
815	}
816
817	struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
818	{
819	struct btrfs_key key = {
820	.objectid = BTRFS_CSUM_TREE_OBJECTID,
821	.type = BTRFS_ROOT_ITEM_KEY,
822	.offset = btrfs_global_root_id(fs_info, bytenr),
823	};
824
825	return btrfs_global_root(fs_info, key: &key);
826	}
827
828	struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
829	{
830	struct btrfs_key key = {
831	.objectid = BTRFS_EXTENT_TREE_OBJECTID,
832	.type = BTRFS_ROOT_ITEM_KEY,
833	.offset = btrfs_global_root_id(fs_info, bytenr),
834	};
835
836	return btrfs_global_root(fs_info, key: &key);
837	}
838
839	struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
840	u64 objectid)
841	{
842	struct btrfs_fs_info *fs_info = trans->fs_info;
843	struct extent_buffer *leaf;
844	struct btrfs_root *tree_root = fs_info->tree_root;
845	struct btrfs_root *root;
846	struct btrfs_key key;
847	unsigned int nofs_flag;
848	int ret = 0;
849
850	/*
851	* We're holding a transaction handle, so use a NOFS memory allocation
852	* context to avoid deadlock if reclaim happens.
853	*/
854	nofs_flag = memalloc_nofs_save();
855	root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
856	memalloc_nofs_restore(flags: nofs_flag);
857	if (!root)
858	return ERR_PTR(error: -ENOMEM);
859
860	root->root_key.objectid = objectid;
861	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
862	root->root_key.offset = 0;
863
864	leaf = btrfs_alloc_tree_block(trans, root, parent: 0, root_objectid: objectid, NULL, level: 0, hint: 0, empty_size: 0,
865	reloc_src_root: 0, nest: BTRFS_NESTING_NORMAL);
866	if (IS_ERR(ptr: leaf)) {
867	ret = PTR_ERR(ptr: leaf);
868	leaf = NULL;
869	goto fail;
870	}
871
872	root->node = leaf;
873	btrfs_mark_buffer_dirty(trans, buf: leaf);
874
875	root->commit_root = btrfs_root_node(root);
876	set_bit(nr: BTRFS_ROOT_TRACK_DIRTY, addr: &root->state);
877
878	btrfs_set_root_flags(s: &root->root_item, val: 0);
879	btrfs_set_root_limit(s: &root->root_item, val: 0);
880	btrfs_set_root_bytenr(s: &root->root_item, val: leaf->start);
881	btrfs_set_root_generation(s: &root->root_item, val: trans->transid);
882	btrfs_set_root_level(s: &root->root_item, val: 0);
883	btrfs_set_root_refs(s: &root->root_item, val: 1);
884	btrfs_set_root_used(s: &root->root_item, val: leaf->len);
885	btrfs_set_root_last_snapshot(s: &root->root_item, val: 0);
886	btrfs_set_root_dirid(s: &root->root_item, val: 0);
887	if (is_fstree(rootid: objectid))
888	generate_random_guid(guid: root->root_item.uuid);
889	else
890	export_guid(dst: root->root_item.uuid, src: &guid_null);
891	btrfs_set_root_drop_level(s: &root->root_item, val: 0);
892
893	btrfs_tree_unlock(eb: leaf);
894
895	key.objectid = objectid;
896	key.type = BTRFS_ROOT_ITEM_KEY;
897	key.offset = 0;
898	ret = btrfs_insert_root(trans, root: tree_root, key: &key, item: &root->root_item);
899	if (ret)
900	goto fail;
901
902	return root;
903
904	fail:
905	btrfs_put_root(root);
906
907	return ERR_PTR(error: ret);
908	}
909
910	static struct btrfs_root *alloc_log_tree(struct btrfs_fs_info *fs_info)
911	{
912	struct btrfs_root *root;
913
914	root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
915	if (!root)
916	return ERR_PTR(error: -ENOMEM);
917
918	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
919	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
920	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
921
922	return root;
923	}
924
925	int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
926	struct btrfs_root *root)
927	{
928	struct extent_buffer *leaf;
929
930	/*
931	* DON'T set SHAREABLE bit for log trees.
932	*
933	* Log trees are not exposed to user space thus can't be snapshotted,
934	* and they go away before a real commit is actually done.
935	*
936	* They do store pointers to file data extents, and those reference
937	* counts still get updated (along with back refs to the log tree).
938	*/
939
940	leaf = btrfs_alloc_tree_block(trans, root, parent: 0, BTRFS_TREE_LOG_OBJECTID,
941	NULL, level: 0, hint: 0, empty_size: 0, reloc_src_root: 0, nest: BTRFS_NESTING_NORMAL);
942	if (IS_ERR(ptr: leaf))
943	return PTR_ERR(ptr: leaf);
944
945	root->node = leaf;
946
947	btrfs_mark_buffer_dirty(trans, buf: root->node);
948	btrfs_tree_unlock(eb: root->node);
949
950	return 0;
951	}
952
953	int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
954	struct btrfs_fs_info *fs_info)
955	{
956	struct btrfs_root *log_root;
957
958	log_root = alloc_log_tree(fs_info);
959	if (IS_ERR(ptr: log_root))
960	return PTR_ERR(ptr: log_root);
961
962	if (!btrfs_is_zoned(fs_info)) {
963	int ret = btrfs_alloc_log_tree_node(trans, root: log_root);
964
965	if (ret) {
966	btrfs_put_root(root: log_root);
967	return ret;
968	}
969	}
970
971	WARN_ON(fs_info->log_root_tree);
972	fs_info->log_root_tree = log_root;
973	return 0;
974	}
975
976	int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
977	struct btrfs_root *root)
978	{
979	struct btrfs_fs_info *fs_info = root->fs_info;
980	struct btrfs_root *log_root;
981	struct btrfs_inode_item *inode_item;
982	int ret;
983
984	log_root = alloc_log_tree(fs_info);
985	if (IS_ERR(ptr: log_root))
986	return PTR_ERR(ptr: log_root);
987
988	ret = btrfs_alloc_log_tree_node(trans, root: log_root);
989	if (ret) {
990	btrfs_put_root(root: log_root);
991	return ret;
992	}
993
994	btrfs_set_root_last_trans(root: log_root, transid: trans->transid);
995	log_root->root_key.offset = btrfs_root_id(root);
996
997	inode_item = &log_root->root_item.inode;
998	btrfs_set_stack_inode_generation(s: inode_item, val: 1);
999	btrfs_set_stack_inode_size(s: inode_item, val: 3);
1000	btrfs_set_stack_inode_nlink(s: inode_item, val: 1);
1001	btrfs_set_stack_inode_nbytes(s: inode_item,
1002	val: fs_info->nodesize);
1003	btrfs_set_stack_inode_mode(s: inode_item, S_IFDIR | 0755);
1004
1005	btrfs_set_root_node(item: &log_root->root_item, node: log_root->node);
1006
1007	WARN_ON(root->log_root);
1008	root->log_root = log_root;
1009	btrfs_set_root_log_transid(root, log_transid: 0);
1010	root->log_transid_committed = -1;
1011	btrfs_set_root_last_log_commit(root, commit_id: 0);
1012	return 0;
1013	}
1014
1015	static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1016	struct btrfs_path *path,
1017	const struct btrfs_key *key)
1018	{
1019	struct btrfs_root *root;
1020	struct btrfs_tree_parent_check check = { 0 };
1021	struct btrfs_fs_info *fs_info = tree_root->fs_info;
1022	u64 generation;
1023	int ret;
1024	int level;
1025
1026	root = btrfs_alloc_root(fs_info, objectid: key->objectid, GFP_NOFS);
1027	if (!root)
1028	return ERR_PTR(error: -ENOMEM);
1029
1030	ret = btrfs_find_root(root: tree_root, search_key: key, path,
1031	root_item: &root->root_item, root_key: &root->root_key);
1032	if (ret) {
1033	if (ret > 0)
1034	ret = -ENOENT;
1035	goto fail;
1036	}
1037
1038	generation = btrfs_root_generation(s: &root->root_item);
1039	level = btrfs_root_level(s: &root->root_item);
1040	check.level = level;
1041	check.transid = generation;
1042	check.owner_root = key->objectid;
1043	root->node = read_tree_block(fs_info, bytenr: btrfs_root_bytenr(s: &root->root_item),
1044	check: &check);
1045	if (IS_ERR(ptr: root->node)) {
1046	ret = PTR_ERR(ptr: root->node);
1047	root->node = NULL;
1048	goto fail;
1049	}
1050	if (!btrfs_buffer_uptodate(eb: root->node, parent_transid: generation, atomic: 0)) {
1051	ret = -EIO;
1052	goto fail;
1053	}
1054
1055	/*
1056	* For real fs, and not log/reloc trees, root owner must
1057	* match its root node owner
1058	*/
1059	if (!btrfs_is_testing(fs_info) &&
1060	btrfs_root_id(root) != BTRFS_TREE_LOG_OBJECTID &&
1061	btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID &&
1062	btrfs_root_id(root) != btrfs_header_owner(eb: root->node)) {
1063	btrfs_crit(fs_info,
1064	"root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1065	btrfs_root_id(root), root->node->start,
1066	btrfs_header_owner(root->node),
1067	btrfs_root_id(root));
1068	ret = -EUCLEAN;
1069	goto fail;
1070	}
1071	root->commit_root = btrfs_root_node(root);
1072	return root;
1073	fail:
1074	btrfs_put_root(root);
1075	return ERR_PTR(error: ret);
1076	}
1077
1078	struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1079	const struct btrfs_key *key)
1080	{
1081	struct btrfs_root *root;
1082	BTRFS_PATH_AUTO_FREE(path);
1083
1084	path = btrfs_alloc_path();
1085	if (!path)
1086	return ERR_PTR(error: -ENOMEM);
1087	root = read_tree_root_path(tree_root, path, key);
1088
1089	return root;
1090	}
1091
1092	/*
1093	* Initialize subvolume root in-memory structure.
1094	*
1095	* @anon_dev: anonymous device to attach to the root, if zero, allocate new
1096	*
1097	* In case of failure the caller is responsible to call btrfs_free_fs_root()
1098	*/
1099	static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1100	{
1101	int ret;
1102
1103	btrfs_drew_lock_init(lock: &root->snapshot_lock);
1104
1105	if (btrfs_root_id(root) != BTRFS_TREE_LOG_OBJECTID &&
1106	!btrfs_is_data_reloc_root(root) &&
1107	is_fstree(rootid: btrfs_root_id(root))) {
1108	set_bit(nr: BTRFS_ROOT_SHAREABLE, addr: &root->state);
1109	btrfs_check_and_init_root_item(item: &root->root_item);
1110	}
1111
1112	/*
1113	* Don't assign anonymous block device to roots that are not exposed to
1114	* userspace, the id pool is limited to 1M
1115	*/
1116	if (is_fstree(rootid: btrfs_root_id(root)) &&
1117	btrfs_root_refs(s: &root->root_item) > 0) {
1118	if (!anon_dev) {
1119	ret = get_anon_bdev(&root->anon_dev);
1120	if (ret)
1121	return ret;
1122	} else {
1123	root->anon_dev = anon_dev;
1124	}
1125	}
1126
1127	mutex_lock(&root->objectid_mutex);
1128	ret = btrfs_init_root_free_objectid(root);
1129	if (ret) {
1130	mutex_unlock(lock: &root->objectid_mutex);
1131	return ret;
1132	}
1133
1134	ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1135
1136	mutex_unlock(lock: &root->objectid_mutex);
1137
1138	return 0;
1139	}
1140
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(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(lock: &fs_info->fs_roots_radix_lock);
1151	return root;
1152	}
1153
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(root: fs_info->tree_root);
1166	case BTRFS_EXTENT_TREE_OBJECTID:
1167	return btrfs_grab_root(root: btrfs_global_root(fs_info, key: &key));
1168	case BTRFS_CHUNK_TREE_OBJECTID:
1169	return btrfs_grab_root(root: fs_info->chunk_root);
1170	case BTRFS_DEV_TREE_OBJECTID:
1171	return btrfs_grab_root(root: fs_info->dev_root);
1172	case BTRFS_CSUM_TREE_OBJECTID:
1173	return btrfs_grab_root(root: btrfs_global_root(fs_info, key: &key));
1174	case BTRFS_QUOTA_TREE_OBJECTID:
1175	return btrfs_grab_root(root: fs_info->quota_root);
1176	case BTRFS_UUID_TREE_OBJECTID:
1177	return btrfs_grab_root(root: fs_info->uuid_root);
1178	case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
1179	return btrfs_grab_root(root: fs_info->block_group_root);
1180	case BTRFS_FREE_SPACE_TREE_OBJECTID:
1181	return btrfs_grab_root(root: btrfs_global_root(fs_info, key: &key));
1182	case BTRFS_RAID_STRIPE_TREE_OBJECTID:
1183	return btrfs_grab_root(root: fs_info->stripe_root);
1184	default:
1185	return NULL;
1186	}
1187	}
1188
1189	int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1190	struct btrfs_root *root)
1191	{
1192	int ret;
1193
1194	ret = radix_tree_preload(GFP_NOFS);
1195	if (ret)
1196	return ret;
1197
1198	spin_lock(lock: &fs_info->fs_roots_radix_lock);
1199	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1200	index: (unsigned long)btrfs_root_id(root),
1201	root);
1202	if (ret == 0) {
1203	btrfs_grab_root(root);
1204	set_bit(nr: BTRFS_ROOT_IN_RADIX, addr: &root->state);
1205	}
1206	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
1207	radix_tree_preload_end();
1208
1209	return ret;
1210	}
1211
1212	void btrfs_check_leaked_roots(const struct btrfs_fs_info *fs_info)
1213	{
1214	#ifdef CONFIG_BTRFS_DEBUG
1215	struct btrfs_root *root;
1216
1217	while (!list_empty(head: &fs_info->allocated_roots)) {
1218	char buf[BTRFS_ROOT_NAME_BUF_LEN];
1219
1220	root = list_first_entry(&fs_info->allocated_roots,
1221	struct btrfs_root, leak_list);
1222	btrfs_err(fs_info, "leaked root %s refcount %d",
1223	btrfs_root_name(&root->root_key, buf),
1224	refcount_read(&root->refs));
1225	WARN_ON_ONCE(1);
1226	while (refcount_read(r: &root->refs) > 1)
1227	btrfs_put_root(root);
1228	btrfs_put_root(root);
1229	}
1230	#endif
1231	}
1232
1233	static void free_global_roots(struct btrfs_fs_info *fs_info)
1234	{
1235	struct btrfs_root *root;
1236	struct rb_node *node;
1237
1238	while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1239	root = rb_entry(node, struct btrfs_root, rb_node);
1240	rb_erase(&root->rb_node, &fs_info->global_root_tree);
1241	btrfs_put_root(root);
1242	}
1243	}
1244
1245	void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1246	{
1247	struct percpu_counter *em_counter = &fs_info->evictable_extent_maps;
1248
1249	percpu_counter_destroy(fbc: &fs_info->stats_read_blocks);
1250	percpu_counter_destroy(fbc: &fs_info->dirty_metadata_bytes);
1251	percpu_counter_destroy(fbc: &fs_info->delalloc_bytes);
1252	percpu_counter_destroy(fbc: &fs_info->ordered_bytes);
1253	if (percpu_counter_initialized(fbc: em_counter))
1254	ASSERT(percpu_counter_sum_positive(em_counter) == 0);
1255	percpu_counter_destroy(fbc: em_counter);
1256	percpu_counter_destroy(fbc: &fs_info->dev_replace.bio_counter);
1257	btrfs_free_csum_hash(fs_info);
1258	btrfs_free_stripe_hash_table(info: fs_info);
1259	btrfs_free_ref_cache(fs_info);
1260	kfree(objp: fs_info->balance_ctl);
1261	kfree(objp: fs_info->delayed_root);
1262	free_global_roots(fs_info);
1263	btrfs_put_root(root: fs_info->tree_root);
1264	btrfs_put_root(root: fs_info->chunk_root);
1265	btrfs_put_root(root: fs_info->dev_root);
1266	btrfs_put_root(root: fs_info->quota_root);
1267	btrfs_put_root(root: fs_info->uuid_root);
1268	btrfs_put_root(root: fs_info->fs_root);
1269	btrfs_put_root(root: fs_info->data_reloc_root);
1270	btrfs_put_root(root: fs_info->block_group_root);
1271	btrfs_put_root(root: fs_info->stripe_root);
1272	btrfs_check_leaked_roots(fs_info);
1273	btrfs_extent_buffer_leak_debug_check(fs_info);
1274	kfree(objp: fs_info->super_copy);
1275	kfree(objp: fs_info->super_for_commit);
1276	kvfree(addr: fs_info);
1277	}
1278
1279
1280	/*
1281	* Get an in-memory reference of a root structure.
1282	*
1283	* For essential trees like root/extent tree, we grab it from fs_info directly.
1284	* For subvolume trees, we check the cached filesystem roots first. If not
1285	* found, then read it from disk and add it to cached fs roots.
1286	*
1287	* Caller should release the root by calling btrfs_put_root() after the usage.
1288	*
1289	* NOTE: Reloc and log trees can't be read by this function as they share the
1290	* same root objectid.
1291	*
1292	* @objectid: root id
1293	* @anon_dev: preallocated anonymous block device number for new roots,
1294	* pass NULL for a new allocation.
1295	* @check_ref: whether to check root item references, If true, return -ENOENT
1296	* for orphan roots
1297	*/
1298	static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1299	u64 objectid, dev_t *anon_dev,
1300	bool check_ref)
1301	{
1302	struct btrfs_root *root;
1303	struct btrfs_path *path;
1304	struct btrfs_key key;
1305	int ret;
1306
1307	root = btrfs_get_global_root(fs_info, objectid);
1308	if (root)
1309	return root;
1310
1311	/*
1312	* If we're called for non-subvolume trees, and above function didn't
1313	* find one, do not try to read it from disk.
1314	*
1315	* This is namely for free-space-tree and quota tree, which can change
1316	* at runtime and should only be grabbed from fs_info.
1317	*/
1318	if (!is_fstree(rootid: objectid) && objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1319	return ERR_PTR(error: -ENOENT);
1320	again:
1321	root = btrfs_lookup_fs_root(fs_info, root_id: objectid);
1322	if (root) {
1323	/*
1324	* Some other caller may have read out the newly inserted
1325	* subvolume already (for things like backref walk etc). Not
1326	* that common but still possible. In that case, we just need
1327	* to free the anon_dev.
1328	*/
1329	if (unlikely(anon_dev && *anon_dev)) {
1330	free_anon_bdev(*anon_dev);
1331	*anon_dev = 0;
1332	}
1333
1334	if (check_ref && btrfs_root_refs(s: &root->root_item) == 0) {
1335	btrfs_put_root(root);
1336	return ERR_PTR(error: -ENOENT);
1337	}
1338	return root;
1339	}
1340
1341	key.objectid = objectid;
1342	key.type = BTRFS_ROOT_ITEM_KEY;
1343	key.offset = (u64)-1;
1344	root = btrfs_read_tree_root(tree_root: fs_info->tree_root, key: &key);
1345	if (IS_ERR(ptr: root))
1346	return root;
1347
1348	if (check_ref && btrfs_root_refs(s: &root->root_item) == 0) {
1349	ret = -ENOENT;
1350	goto fail;
1351	}
1352
1353	ret = btrfs_init_fs_root(root, anon_dev: anon_dev ? *anon_dev : 0);
1354	if (ret)
1355	goto fail;
1356
1357	path = btrfs_alloc_path();
1358	if (!path) {
1359	ret = -ENOMEM;
1360	goto fail;
1361	}
1362	key.objectid = BTRFS_ORPHAN_OBJECTID;
1363	key.type = BTRFS_ORPHAN_ITEM_KEY;
1364	key.offset = objectid;
1365
1366	ret = btrfs_search_slot(NULL, root: fs_info->tree_root, key: &key, p: path, ins_len: 0, cow: 0);
1367	btrfs_free_path(p: path);
1368	if (ret < 0)
1369	goto fail;
1370	if (ret == 0)
1371	set_bit(nr: BTRFS_ROOT_ORPHAN_ITEM_INSERTED, addr: &root->state);
1372
1373	ret = btrfs_insert_fs_root(fs_info, root);
1374	if (ret) {
1375	if (ret == -EEXIST) {
1376	btrfs_put_root(root);
1377	goto again;
1378	}
1379	goto fail;
1380	}
1381	return root;
1382	fail:
1383	/*
1384	* If our caller provided us an anonymous device, then it's his
1385	* responsibility to free it in case we fail. So we have to set our
1386	* root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1387	* and once again by our caller.
1388	*/
1389	if (anon_dev && *anon_dev)
1390	root->anon_dev = 0;
1391	btrfs_put_root(root);
1392	return ERR_PTR(error: ret);
1393	}
1394
1395	/*
1396	* Get in-memory reference of a root structure
1397	*
1398	* @objectid: tree objectid
1399	* @check_ref: if set, verify that the tree exists and the item has at least
1400	* one reference
1401	*/
1402	struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1403	u64 objectid, bool check_ref)
1404	{
1405	return btrfs_get_root_ref(fs_info, objectid, NULL, check_ref);
1406	}
1407
1408	/*
1409	* Get in-memory reference of a root structure, created as new, optionally pass
1410	* the anonymous block device id
1411	*
1412	* @objectid: tree objectid
1413	* @anon_dev: if NULL, allocate a new anonymous block device or use the
1414	* parameter value if not NULL
1415	*/
1416	struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1417	u64 objectid, dev_t *anon_dev)
1418	{
1419	return btrfs_get_root_ref(fs_info, objectid, anon_dev, check_ref: true);
1420	}
1421
1422	/*
1423	* Return a root for the given objectid.
1424	*
1425	* @fs_info: the fs_info
1426	* @objectid: the objectid we need to lookup
1427	*
1428	* This is exclusively used for backref walking, and exists specifically because
1429	* of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1430	* creation time, which means we may have to read the tree_root in order to look
1431	* up a fs root that is not in memory. If the root is not in memory we will
1432	* read the tree root commit root and look up the fs root from there. This is a
1433	* temporary root, it will not be inserted into the radix tree as it doesn't
1434	* have the most uptodate information, it'll simply be discarded once the
1435	* backref code is finished using the root.
1436	*/
1437	struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1438	struct btrfs_path *path,
1439	u64 objectid)
1440	{
1441	struct btrfs_root *root;
1442	struct btrfs_key key;
1443
1444	ASSERT(path->search_commit_root && path->skip_locking);
1445
1446	/*
1447	* This can return -ENOENT if we ask for a root that doesn't exist, but
1448	* since this is called via the backref walking code we won't be looking
1449	* up a root that doesn't exist, unless there's corruption. So if root
1450	* != NULL just return it.
1451	*/
1452	root = btrfs_get_global_root(fs_info, objectid);
1453	if (root)
1454	return root;
1455
1456	root = btrfs_lookup_fs_root(fs_info, root_id: objectid);
1457	if (root)
1458	return root;
1459
1460	key.objectid = objectid;
1461	key.type = BTRFS_ROOT_ITEM_KEY;
1462	key.offset = (u64)-1;
1463	root = read_tree_root_path(tree_root: fs_info->tree_root, path, key: &key);
1464	btrfs_release_path(p: path);
1465
1466	return root;
1467	}
1468
1469	static int cleaner_kthread(void *arg)
1470	{
1471	struct btrfs_fs_info *fs_info = arg;
1472	int again;
1473
1474	while (1) {
1475	again = 0;
1476
1477	set_bit(nr: BTRFS_FS_CLEANER_RUNNING, addr: &fs_info->flags);
1478
1479	/* Make the cleaner go to sleep early. */
1480	if (btrfs_need_cleaner_sleep(fs_info))
1481	goto sleep;
1482
1483	/*
1484	* Do not do anything if we might cause open_ctree() to block
1485	* before we have finished mounting the filesystem.
1486	*/
1487	if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1488	goto sleep;
1489
1490	if (!mutex_trylock(&fs_info->cleaner_mutex))
1491	goto sleep;
1492
1493	/*
1494	* Avoid the problem that we change the status of the fs
1495	* during the above check and trylock.
1496	*/
1497	if (btrfs_need_cleaner_sleep(fs_info)) {
1498	mutex_unlock(lock: &fs_info->cleaner_mutex);
1499	goto sleep;
1500	}
1501
1502	if (test_and_clear_bit(nr: BTRFS_FS_FEATURE_CHANGED, addr: &fs_info->flags))
1503	btrfs_sysfs_feature_update(fs_info);
1504
1505	btrfs_run_delayed_iputs(fs_info);
1506
1507	again = btrfs_clean_one_deleted_snapshot(fs_info);
1508	mutex_unlock(lock: &fs_info->cleaner_mutex);
1509
1510	/*
1511	* The defragger has dealt with the R/O remount and umount,
1512	* needn't do anything special here.
1513	*/
1514	btrfs_run_defrag_inodes(fs_info);
1515
1516	/*
1517	* Acquires fs_info->reclaim_bgs_lock to avoid racing
1518	* with relocation (btrfs_relocate_chunk) and relocation
1519	* acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1520	* after acquiring fs_info->reclaim_bgs_lock. So we
1521	* can't hold, nor need to, fs_info->cleaner_mutex when deleting
1522	* unused block groups.
1523	*/
1524	btrfs_delete_unused_bgs(fs_info);
1525
1526	/*
1527	* Reclaim block groups in the reclaim_bgs list after we deleted
1528	* all unused block_groups. This possibly gives us some more free
1529	* space.
1530	*/
1531	btrfs_reclaim_bgs(fs_info);
1532	sleep:
1533	clear_and_wake_up_bit(bit: BTRFS_FS_CLEANER_RUNNING, word: &fs_info->flags);
1534	if (kthread_should_park())
1535	kthread_parkme();
1536	if (kthread_should_stop())
1537	return 0;
1538	if (!again) {
1539	set_current_state(TASK_INTERRUPTIBLE);
1540	schedule();
1541	__set_current_state(TASK_RUNNING);
1542	}
1543	}
1544	}
1545
1546	static int transaction_kthread(void *arg)
1547	{
1548	struct btrfs_root *root = arg;
1549	struct btrfs_fs_info *fs_info = root->fs_info;
1550	struct btrfs_trans_handle *trans;
1551	struct btrfs_transaction *cur;
1552	u64 transid;
1553	time64_t delta;
1554	unsigned long delay;
1555	bool cannot_commit;
1556
1557	do {
1558	cannot_commit = false;
1559	delay = secs_to_jiffies(fs_info->commit_interval);
1560	mutex_lock(&fs_info->transaction_kthread_mutex);
1561
1562	spin_lock(lock: &fs_info->trans_lock);
1563	cur = fs_info->running_transaction;
1564	if (!cur) {
1565	spin_unlock(lock: &fs_info->trans_lock);
1566	goto sleep;
1567	}
1568
1569	delta = ktime_get_seconds() - cur->start_time;
1570	if (!test_and_clear_bit(nr: BTRFS_FS_COMMIT_TRANS, addr: &fs_info->flags) &&
1571	cur->state < TRANS_STATE_COMMIT_PREP &&
1572	delta < fs_info->commit_interval) {
1573	spin_unlock(lock: &fs_info->trans_lock);
1574	delay -= secs_to_jiffies(delta - 1);
1575	delay = min(delay,
1576	secs_to_jiffies(fs_info->commit_interval));
1577	goto sleep;
1578	}
1579	transid = cur->transid;
1580	spin_unlock(lock: &fs_info->trans_lock);
1581
1582	/* If the file system is aborted, this will always fail. */
1583	trans = btrfs_attach_transaction(root);
1584	if (IS_ERR(ptr: trans)) {
1585	if (PTR_ERR(ptr: trans) != -ENOENT)
1586	cannot_commit = true;
1587	goto sleep;
1588	}
1589	if (transid == trans->transid) {
1590	btrfs_commit_transaction(trans);
1591	} else {
1592	btrfs_end_transaction(trans);
1593	}
1594	sleep:
1595	wake_up_process(tsk: fs_info->cleaner_kthread);
1596	mutex_unlock(lock: &fs_info->transaction_kthread_mutex);
1597
1598	if (BTRFS_FS_ERROR(fs_info))
1599	btrfs_cleanup_transaction(fs_info);
1600	if (!kthread_should_stop() &&
1601	(!btrfs_transaction_blocked(info: fs_info) ||
1602	cannot_commit))
1603	schedule_timeout_interruptible(timeout: delay);
1604	} while (!kthread_should_stop());
1605	return 0;
1606	}
1607
1608	/*
1609	* This will find the highest generation in the array of root backups. The
1610	* index of the highest array is returned, or -EINVAL if we can't find
1611	* anything.
1612	*
1613	* We check to make sure the array is valid by comparing the
1614	* generation of the latest root in the array with the generation
1615	* in the super block. If they don't match we pitch it.
1616	*/
1617	static int find_newest_super_backup(struct btrfs_fs_info *info)
1618	{
1619	const u64 newest_gen = btrfs_super_generation(s: info->super_copy);
1620	u64 cur;
1621	struct btrfs_root_backup *root_backup;
1622	int i;
1623
1624	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1625	root_backup = info->super_copy->super_roots + i;
1626	cur = btrfs_backup_tree_root_gen(s: root_backup);
1627	if (cur == newest_gen)
1628	return i;
1629	}
1630
1631	return -EINVAL;
1632	}
1633
1634	/*
1635	* copy all the root pointers into the super backup array.
1636	* this will bump the backup pointer by one when it is
1637	* done
1638	*/
1639	static void backup_super_roots(struct btrfs_fs_info *info)
1640	{
1641	const int next_backup = info->backup_root_index;
1642	struct btrfs_root_backup *root_backup;
1643
1644	root_backup = info->super_for_commit->super_roots + next_backup;
1645
1646	/*
1647	* make sure all of our padding and empty slots get zero filled
1648	* regardless of which ones we use today
1649	*/
1650	memset(root_backup, 0, sizeof(*root_backup));
1651
1652	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1653
1654	btrfs_set_backup_tree_root(s: root_backup, val: info->tree_root->node->start);
1655	btrfs_set_backup_tree_root_gen(s: root_backup,
1656	val: btrfs_header_generation(eb: info->tree_root->node));
1657
1658	btrfs_set_backup_tree_root_level(s: root_backup,
1659	val: btrfs_header_level(eb: info->tree_root->node));
1660
1661	btrfs_set_backup_chunk_root(s: root_backup, val: info->chunk_root->node->start);
1662	btrfs_set_backup_chunk_root_gen(s: root_backup,
1663	val: btrfs_header_generation(eb: info->chunk_root->node));
1664	btrfs_set_backup_chunk_root_level(s: root_backup,
1665	val: btrfs_header_level(eb: info->chunk_root->node));
1666
1667	if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
1668	struct btrfs_root *extent_root = btrfs_extent_root(fs_info: info, bytenr: 0);
1669	struct btrfs_root *csum_root = btrfs_csum_root(fs_info: info, bytenr: 0);
1670
1671	btrfs_set_backup_extent_root(s: root_backup,
1672	val: extent_root->node->start);
1673	btrfs_set_backup_extent_root_gen(s: root_backup,
1674	val: btrfs_header_generation(eb: extent_root->node));
1675	btrfs_set_backup_extent_root_level(s: root_backup,
1676	val: btrfs_header_level(eb: extent_root->node));
1677
1678	btrfs_set_backup_csum_root(s: root_backup, val: csum_root->node->start);
1679	btrfs_set_backup_csum_root_gen(s: root_backup,
1680	val: btrfs_header_generation(eb: csum_root->node));
1681	btrfs_set_backup_csum_root_level(s: root_backup,
1682	val: btrfs_header_level(eb: csum_root->node));
1683	}
1684
1685	/*
1686	* we might commit during log recovery, which happens before we set
1687	* the fs_root. Make sure it is valid before we fill it in.
1688	*/
1689	if (info->fs_root && info->fs_root->node) {
1690	btrfs_set_backup_fs_root(s: root_backup,
1691	val: info->fs_root->node->start);
1692	btrfs_set_backup_fs_root_gen(s: root_backup,
1693	val: btrfs_header_generation(eb: info->fs_root->node));
1694	btrfs_set_backup_fs_root_level(s: root_backup,
1695	val: btrfs_header_level(eb: info->fs_root->node));
1696	}
1697
1698	btrfs_set_backup_dev_root(s: root_backup, val: info->dev_root->node->start);
1699	btrfs_set_backup_dev_root_gen(s: root_backup,
1700	val: btrfs_header_generation(eb: info->dev_root->node));
1701	btrfs_set_backup_dev_root_level(s: root_backup,
1702	val: btrfs_header_level(eb: info->dev_root->node));
1703
1704	btrfs_set_backup_total_bytes(s: root_backup,
1705	val: btrfs_super_total_bytes(s: info->super_copy));
1706	btrfs_set_backup_bytes_used(s: root_backup,
1707	val: btrfs_super_bytes_used(s: info->super_copy));
1708	btrfs_set_backup_num_devices(s: root_backup,
1709	val: btrfs_super_num_devices(s: info->super_copy));
1710
1711	/*
1712	* if we don't copy this out to the super_copy, it won't get remembered
1713	* for the next commit
1714	*/
1715	memcpy(&info->super_copy->super_roots,
1716	&info->super_for_commit->super_roots,
1717	sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1718	}
1719
1720	/*
1721	* Reads a backup root based on the passed priority. Prio 0 is the newest, prio
1722	* 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1723	*
1724	* @fs_info: filesystem whose backup roots need to be read
1725	* @priority: priority of backup root required
1726	*
1727	* Returns backup root index on success and -EINVAL otherwise.
1728	*/
1729	static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1730	{
1731	int backup_index = find_newest_super_backup(info: fs_info);
1732	struct btrfs_super_block *super = fs_info->super_copy;
1733	struct btrfs_root_backup *root_backup;
1734
1735	if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1736	if (priority == 0)
1737	return backup_index;
1738
1739	backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1740	backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1741	} else {
1742	return -EINVAL;
1743	}
1744
1745	root_backup = super->super_roots + backup_index;
1746
1747	btrfs_set_super_generation(s: super,
1748	val: btrfs_backup_tree_root_gen(s: root_backup));
1749	btrfs_set_super_root(s: super, val: btrfs_backup_tree_root(s: root_backup));
1750	btrfs_set_super_root_level(s: super,
1751	val: btrfs_backup_tree_root_level(s: root_backup));
1752	btrfs_set_super_bytes_used(s: super, val: btrfs_backup_bytes_used(s: root_backup));
1753
1754	/*
1755	* Fixme: the total bytes and num_devices need to match or we should
1756	* need a fsck
1757	*/
1758	btrfs_set_super_total_bytes(s: super, val: btrfs_backup_total_bytes(s: root_backup));
1759	btrfs_set_super_num_devices(s: super, val: btrfs_backup_num_devices(s: root_backup));
1760
1761	return backup_index;
1762	}
1763
1764	/* helper to cleanup workers */
1765	static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1766	{
1767	btrfs_destroy_workqueue(wq: fs_info->fixup_workers);
1768	btrfs_destroy_workqueue(wq: fs_info->delalloc_workers);
1769	btrfs_destroy_workqueue(wq: fs_info->workers);
1770	if (fs_info->endio_workers)
1771	destroy_workqueue(wq: fs_info->endio_workers);
1772	if (fs_info->rmw_workers)
1773	destroy_workqueue(wq: fs_info->rmw_workers);
1774	if (fs_info->compressed_write_workers)
1775	destroy_workqueue(wq: fs_info->compressed_write_workers);
1776	btrfs_destroy_workqueue(wq: fs_info->endio_write_workers);
1777	btrfs_destroy_workqueue(wq: fs_info->endio_freespace_worker);
1778	btrfs_destroy_workqueue(wq: fs_info->delayed_workers);
1779	btrfs_destroy_workqueue(wq: fs_info->caching_workers);
1780	btrfs_destroy_workqueue(wq: fs_info->flush_workers);
1781	btrfs_destroy_workqueue(wq: fs_info->qgroup_rescan_workers);
1782	if (fs_info->discard_ctl.discard_workers)
1783	destroy_workqueue(wq: fs_info->discard_ctl.discard_workers);
1784	/*
1785	* Now that all other work queues are destroyed, we can safely destroy
1786	* the queues used for metadata I/O, since tasks from those other work
1787	* queues can do metadata I/O operations.
1788	*/
1789	if (fs_info->endio_meta_workers)
1790	destroy_workqueue(wq: fs_info->endio_meta_workers);
1791	}
1792
1793	static void free_root_extent_buffers(struct btrfs_root *root)
1794	{
1795	if (root) {
1796	free_extent_buffer(eb: root->node);
1797	free_extent_buffer(eb: root->commit_root);
1798	root->node = NULL;
1799	root->commit_root = NULL;
1800	}
1801	}
1802
1803	static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
1804	{
1805	struct btrfs_root *root, *tmp;
1806
1807	rbtree_postorder_for_each_entry_safe(root, tmp,
1808	&fs_info->global_root_tree,
1809	rb_node)
1810	free_root_extent_buffers(root);
1811	}
1812
1813	/* helper to cleanup tree roots */
1814	static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1815	{
1816	free_root_extent_buffers(root: info->tree_root);
1817
1818	free_global_root_pointers(fs_info: info);
1819	free_root_extent_buffers(root: info->dev_root);
1820	free_root_extent_buffers(root: info->quota_root);
1821	free_root_extent_buffers(root: info->uuid_root);
1822	free_root_extent_buffers(root: info->fs_root);
1823	free_root_extent_buffers(root: info->data_reloc_root);
1824	free_root_extent_buffers(root: info->block_group_root);
1825	free_root_extent_buffers(root: info->stripe_root);
1826	if (free_chunk_root)
1827	free_root_extent_buffers(root: info->chunk_root);
1828	}
1829
1830	void btrfs_put_root(struct btrfs_root *root)
1831	{
1832	if (!root)
1833	return;
1834
1835	if (refcount_dec_and_test(r: &root->refs)) {
1836	if (WARN_ON(!xa_empty(&root->inodes)))
1837	xa_destroy(&root->inodes);
1838	WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
1839	if (root->anon_dev)
1840	free_anon_bdev(root->anon_dev);
1841	free_root_extent_buffers(root);
1842	#ifdef CONFIG_BTRFS_DEBUG
1843	spin_lock(lock: &root->fs_info->fs_roots_radix_lock);
1844	list_del_init(entry: &root->leak_list);
1845	spin_unlock(lock: &root->fs_info->fs_roots_radix_lock);
1846	#endif
1847	kfree(objp: root);
1848	}
1849	}
1850
1851	void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
1852	{
1853	int ret;
1854	struct btrfs_root *gang[8];
1855	int i;
1856
1857	while (!list_empty(head: &fs_info->dead_roots)) {
1858	gang[0] = list_first_entry(&fs_info->dead_roots,
1859	struct btrfs_root, root_list);
1860	list_del(entry: &gang[0]->root_list);
1861
1862	if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
1863	btrfs_drop_and_free_fs_root(fs_info, root: gang[0]);
1864	btrfs_put_root(root: gang[0]);
1865	}
1866
1867	while (1) {
1868	ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1869	results: (void **)gang, first_index: 0,
1870	ARRAY_SIZE(gang));
1871	if (!ret)
1872	break;
1873	for (i = 0; i < ret; i++)
1874	btrfs_drop_and_free_fs_root(fs_info, root: gang[i]);
1875	}
1876	}
1877
1878	static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
1879	{
1880	mutex_init(&fs_info->scrub_lock);
1881	atomic_set(v: &fs_info->scrubs_running, i: 0);
1882	atomic_set(v: &fs_info->scrub_pause_req, i: 0);
1883	atomic_set(v: &fs_info->scrubs_paused, i: 0);
1884	atomic_set(v: &fs_info->scrub_cancel_req, i: 0);
1885	init_waitqueue_head(&fs_info->scrub_pause_wait);
1886	refcount_set(r: &fs_info->scrub_workers_refcnt, n: 0);
1887	}
1888
1889	static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
1890	{
1891	spin_lock_init(&fs_info->balance_lock);
1892	mutex_init(&fs_info->balance_mutex);
1893	atomic_set(v: &fs_info->balance_pause_req, i: 0);
1894	atomic_set(v: &fs_info->balance_cancel_req, i: 0);
1895	fs_info->balance_ctl = NULL;
1896	init_waitqueue_head(&fs_info->balance_wait_q);
1897	atomic_set(v: &fs_info->reloc_cancel_req, i: 0);
1898	}
1899
1900	static int btrfs_init_btree_inode(struct super_block *sb)
1901	{
1902	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1903	unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
1904	root: fs_info->tree_root);
1905	struct inode *inode;
1906
1907	inode = new_inode(sb);
1908	if (!inode)
1909	return -ENOMEM;
1910
1911	btrfs_set_inode_number(BTRFS_I(inode), BTRFS_BTREE_INODE_OBJECTID);
1912	set_nlink(inode, nlink: 1);
1913	/*
1914	* we set the i_size on the btree inode to the max possible int.
1915	* the real end of the address space is determined by all of
1916	* the devices in the system
1917	*/
1918	inode->i_size = OFFSET_MAX;
1919	inode->i_mapping->a_ops = &btree_aops;
1920	mapping_set_gfp_mask(m: inode->i_mapping, GFP_NOFS);
1921
1922	btrfs_extent_io_tree_init(fs_info, tree: &BTRFS_I(inode)->io_tree,
1923	owner: IO_TREE_BTREE_INODE_IO);
1924	btrfs_extent_map_tree_init(tree: &BTRFS_I(inode)->extent_tree);
1925
1926	BTRFS_I(inode)->root = btrfs_grab_root(root: fs_info->tree_root);
1927	set_bit(nr: BTRFS_INODE_DUMMY, addr: &BTRFS_I(inode)->runtime_flags);
1928	__insert_inode_hash(inode, hashval: hash);
1929	fs_info->btree_inode = inode;
1930
1931	return 0;
1932	}
1933
1934	static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
1935	{
1936	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
1937	init_rwsem(&fs_info->dev_replace.rwsem);
1938	init_waitqueue_head(&fs_info->dev_replace.replace_wait);
1939	}
1940
1941	static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
1942	{
1943	spin_lock_init(&fs_info->qgroup_lock);
1944	mutex_init(&fs_info->qgroup_ioctl_lock);
1945	fs_info->qgroup_tree = RB_ROOT;
1946	INIT_LIST_HEAD(list: &fs_info->dirty_qgroups);
1947	fs_info->qgroup_seq = 1;
1948	fs_info->qgroup_ulist = NULL;
1949	fs_info->qgroup_rescan_running = false;
1950	fs_info->qgroup_drop_subtree_thres = BTRFS_QGROUP_DROP_SUBTREE_THRES_DEFAULT;
1951	mutex_init(&fs_info->qgroup_rescan_lock);
1952	}
1953
1954	static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
1955	{
1956	u32 max_active = fs_info->thread_pool_size;
1957	unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
1958	unsigned int ordered_flags = WQ_MEM_RECLAIM | WQ_FREEZABLE;
1959
1960	fs_info->workers =
1961	btrfs_alloc_workqueue(fs_info, name: "worker", flags, limit_active: max_active, thresh: 16);
1962
1963	fs_info->delalloc_workers =
1964	btrfs_alloc_workqueue(fs_info, name: "delalloc",
1965	flags, limit_active: max_active, thresh: 2);
1966
1967	fs_info->flush_workers =
1968	btrfs_alloc_workqueue(fs_info, name: "flush_delalloc",
1969	flags, limit_active: max_active, thresh: 0);
1970
1971	fs_info->caching_workers =
1972	btrfs_alloc_workqueue(fs_info, name: "cache", flags, limit_active: max_active, thresh: 0);
1973
1974	fs_info->fixup_workers =
1975	btrfs_alloc_ordered_workqueue(fs_info, name: "fixup", flags: ordered_flags);
1976
1977	fs_info->endio_workers =
1978	alloc_workqueue(fmt: "btrfs-endio", flags, max_active);
1979	fs_info->endio_meta_workers =
1980	alloc_workqueue(fmt: "btrfs-endio-meta", flags, max_active);
1981	fs_info->rmw_workers = alloc_workqueue(fmt: "btrfs-rmw", flags, max_active);
1982	fs_info->endio_write_workers =
1983	btrfs_alloc_workqueue(fs_info, name: "endio-write", flags,
1984	limit_active: max_active, thresh: 2);
1985	fs_info->compressed_write_workers =
1986	alloc_workqueue(fmt: "btrfs-compressed-write", flags, max_active);
1987	fs_info->endio_freespace_worker =
1988	btrfs_alloc_workqueue(fs_info, name: "freespace-write", flags,
1989	limit_active: max_active, thresh: 0);
1990	fs_info->delayed_workers =
1991	btrfs_alloc_workqueue(fs_info, name: "delayed-meta", flags,
1992	limit_active: max_active, thresh: 0);
1993	fs_info->qgroup_rescan_workers =
1994	btrfs_alloc_ordered_workqueue(fs_info, name: "qgroup-rescan",
1995	flags: ordered_flags);
1996	fs_info->discard_ctl.discard_workers =
1997	alloc_ordered_workqueue("btrfs-discard", WQ_FREEZABLE);
1998
1999	if (!(fs_info->workers &&
2000	fs_info->delalloc_workers && fs_info->flush_workers &&
2001	fs_info->endio_workers && fs_info->endio_meta_workers &&
2002	fs_info->compressed_write_workers &&
2003	fs_info->endio_write_workers &&
2004	fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2005	fs_info->caching_workers && fs_info->fixup_workers &&
2006	fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2007	fs_info->discard_ctl.discard_workers)) {
2008	return -ENOMEM;
2009	}
2010
2011	return 0;
2012	}
2013
2014	static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2015	{
2016	struct crypto_shash *csum_shash;
2017	const char *csum_driver = btrfs_super_csum_driver(csum_type);
2018
2019	csum_shash = crypto_alloc_shash(alg_name: csum_driver, type: 0, mask: 0);
2020
2021	if (IS_ERR(ptr: csum_shash)) {
2022	btrfs_err(fs_info, "error allocating %s hash for checksum",
2023	csum_driver);
2024	return PTR_ERR(ptr: csum_shash);
2025	}
2026
2027	fs_info->csum_shash = csum_shash;
2028
2029	/*
2030	* Check if the checksum implementation is a fast accelerated one.
2031	* As-is this is a bit of a hack and should be replaced once the csum
2032	* implementations provide that information themselves.
2033	*/
2034	switch (csum_type) {
2035	case BTRFS_CSUM_TYPE_CRC32:
2036	if (!strstr(crypto_shash_driver_name(tfm: csum_shash), "generic"))
2037	set_bit(nr: BTRFS_FS_CSUM_IMPL_FAST, addr: &fs_info->flags);
2038	break;
2039	case BTRFS_CSUM_TYPE_XXHASH:
2040	set_bit(nr: BTRFS_FS_CSUM_IMPL_FAST, addr: &fs_info->flags);
2041	break;
2042	default:
2043	break;
2044	}
2045
2046	btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2047	btrfs_super_csum_name(csum_type),
2048	crypto_shash_driver_name(csum_shash));
2049	return 0;
2050	}
2051
2052	static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2053	struct btrfs_fs_devices *fs_devices)
2054	{
2055	int ret;
2056	struct btrfs_tree_parent_check check = { 0 };
2057	struct btrfs_root *log_tree_root;
2058	struct btrfs_super_block *disk_super = fs_info->super_copy;
2059	u64 bytenr = btrfs_super_log_root(s: disk_super);
2060	int level = btrfs_super_log_root_level(s: disk_super);
2061
2062	if (fs_devices->rw_devices == 0) {
2063	btrfs_warn(fs_info, "log replay required on RO media");
2064	return -EIO;
2065	}
2066
2067	log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2068	GFP_KERNEL);
2069	if (!log_tree_root)
2070	return -ENOMEM;
2071
2072	check.level = level;
2073	check.transid = fs_info->generation + 1;
2074	check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2075	log_tree_root->node = read_tree_block(fs_info, bytenr, check: &check);
2076	if (IS_ERR(ptr: log_tree_root->node)) {
2077	btrfs_warn(fs_info, "failed to read log tree");
2078	ret = PTR_ERR(ptr: log_tree_root->node);
2079	log_tree_root->node = NULL;
2080	btrfs_put_root(root: log_tree_root);
2081	return ret;
2082	}
2083	if (!extent_buffer_uptodate(eb: log_tree_root->node)) {
2084	btrfs_err(fs_info, "failed to read log tree");
2085	btrfs_put_root(root: log_tree_root);
2086	return -EIO;
2087	}
2088
2089	/* returns with log_tree_root freed on success */
2090	ret = btrfs_recover_log_trees(tree_root: log_tree_root);
2091	if (ret) {
2092	btrfs_handle_fs_error(fs_info, ret,
2093	"Failed to recover log tree");
2094	btrfs_put_root(root: log_tree_root);
2095	return ret;
2096	}
2097
2098	if (sb_rdonly(sb: fs_info->sb)) {
2099	ret = btrfs_commit_super(fs_info);
2100	if (ret)
2101	return ret;
2102	}
2103
2104	return 0;
2105	}
2106
2107	static int load_global_roots_objectid(struct btrfs_root *tree_root,
2108	struct btrfs_path *path, u64 objectid,
2109	const char *name)
2110	{
2111	struct btrfs_fs_info *fs_info = tree_root->fs_info;
2112	struct btrfs_root *root;
2113	u64 max_global_id = 0;
2114	int ret;
2115	struct btrfs_key key = {
2116	.objectid = objectid,
2117	.type = BTRFS_ROOT_ITEM_KEY,
2118	.offset = 0,
2119	};
2120	bool found = false;
2121
2122	/* If we have IGNOREDATACSUMS skip loading these roots. */
2123	if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2124	btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2125	set_bit(nr: BTRFS_FS_STATE_NO_DATA_CSUMS, addr: &fs_info->fs_state);
2126	return 0;
2127	}
2128
2129	while (1) {
2130	ret = btrfs_search_slot(NULL, root: tree_root, key: &key, p: path, ins_len: 0, cow: 0);
2131	if (ret < 0)
2132	break;
2133
2134	if (path->slots[0] >= btrfs_header_nritems(eb: path->nodes[0])) {
2135	ret = btrfs_next_leaf(root: tree_root, path);
2136	if (ret) {
2137	if (ret > 0)
2138	ret = 0;
2139	break;
2140	}
2141	}
2142	ret = 0;
2143
2144	btrfs_item_key_to_cpu(eb: path->nodes[0], cpu_key: &key, nr: path->slots[0]);
2145	if (key.objectid != objectid)
2146	break;
2147	btrfs_release_path(p: path);
2148
2149	/*
2150	* Just worry about this for extent tree, it'll be the same for
2151	* everybody.
2152	*/
2153	if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2154	max_global_id = max(max_global_id, key.offset);
2155
2156	found = true;
2157	root = read_tree_root_path(tree_root, path, key: &key);
2158	if (IS_ERR(ptr: root)) {
2159	if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2160	ret = PTR_ERR(ptr: root);
2161	break;
2162	}
2163	set_bit(nr: BTRFS_ROOT_TRACK_DIRTY, addr: &root->state);
2164	ret = btrfs_global_root_insert(root);
2165	if (ret) {
2166	btrfs_put_root(root);
2167	break;
2168	}
2169	key.offset++;
2170	}
2171	btrfs_release_path(p: path);
2172
2173	if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2174	fs_info->nr_global_roots = max_global_id + 1;
2175
2176	if (!found || ret) {
2177	if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2178	set_bit(nr: BTRFS_FS_STATE_NO_DATA_CSUMS, addr: &fs_info->fs_state);
2179
2180	if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2181	ret = ret ? ret : -ENOENT;
2182	else
2183	ret = 0;
2184	btrfs_err(fs_info, "failed to load root %s", name);
2185	}
2186	return ret;
2187	}
2188
2189	static int load_global_roots(struct btrfs_root *tree_root)
2190	{
2191	BTRFS_PATH_AUTO_FREE(path);
2192	int ret;
2193
2194	path = btrfs_alloc_path();
2195	if (!path)
2196	return -ENOMEM;
2197
2198	ret = load_global_roots_objectid(tree_root, path,
2199	BTRFS_EXTENT_TREE_OBJECTID, name: "extent");
2200	if (ret)
2201	return ret;
2202	ret = load_global_roots_objectid(tree_root, path,
2203	BTRFS_CSUM_TREE_OBJECTID, name: "csum");
2204	if (ret)
2205	return ret;
2206	if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2207	return ret;
2208	ret = load_global_roots_objectid(tree_root, path,
2209	BTRFS_FREE_SPACE_TREE_OBJECTID,
2210	name: "free space");
2211
2212	return ret;
2213	}
2214
2215	static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2216	{
2217	struct btrfs_root *tree_root = fs_info->tree_root;
2218	struct btrfs_root *root;
2219	struct btrfs_key location;
2220	int ret;
2221
2222	ASSERT(fs_info->tree_root);
2223
2224	ret = load_global_roots(tree_root);
2225	if (ret)
2226	return ret;
2227
2228	location.type = BTRFS_ROOT_ITEM_KEY;
2229	location.offset = 0;
2230
2231	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2232	location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2233	root = btrfs_read_tree_root(tree_root, key: &location);
2234	if (IS_ERR(ptr: root)) {
2235	if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2236	ret = PTR_ERR(ptr: root);
2237	goto out;
2238	}
2239	} else {
2240	set_bit(nr: BTRFS_ROOT_TRACK_DIRTY, addr: &root->state);
2241	fs_info->block_group_root = root;
2242	}
2243	}
2244
2245	location.objectid = BTRFS_DEV_TREE_OBJECTID;
2246	root = btrfs_read_tree_root(tree_root, key: &location);
2247	if (IS_ERR(ptr: root)) {
2248	if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2249	ret = PTR_ERR(ptr: root);
2250	goto out;
2251	}
2252	} else {
2253	set_bit(nr: BTRFS_ROOT_TRACK_DIRTY, addr: &root->state);
2254	fs_info->dev_root = root;
2255	}
2256	/* Initialize fs_info for all devices in any case */
2257	ret = btrfs_init_devices_late(fs_info);
2258	if (ret)
2259	goto out;
2260
2261	/*
2262	* This tree can share blocks with some other fs tree during relocation
2263	* and we need a proper setup by btrfs_get_fs_root
2264	*/
2265	root = btrfs_get_fs_root(fs_info: tree_root->fs_info,
2266	BTRFS_DATA_RELOC_TREE_OBJECTID, check_ref: true);
2267	if (IS_ERR(ptr: root)) {
2268	if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2269	ret = PTR_ERR(ptr: root);
2270	goto out;
2271	}
2272	} else {
2273	set_bit(nr: BTRFS_ROOT_TRACK_DIRTY, addr: &root->state);
2274	fs_info->data_reloc_root = root;
2275	}
2276
2277	location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2278	root = btrfs_read_tree_root(tree_root, key: &location);
2279	if (!IS_ERR(ptr: root)) {
2280	set_bit(nr: BTRFS_ROOT_TRACK_DIRTY, addr: &root->state);
2281	fs_info->quota_root = root;
2282	}
2283
2284	location.objectid = BTRFS_UUID_TREE_OBJECTID;
2285	root = btrfs_read_tree_root(tree_root, key: &location);
2286	if (IS_ERR(ptr: root)) {
2287	if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2288	ret = PTR_ERR(ptr: root);
2289	if (ret != -ENOENT)
2290	goto out;
2291	}
2292	} else {
2293	set_bit(nr: BTRFS_ROOT_TRACK_DIRTY, addr: &root->state);
2294	fs_info->uuid_root = root;
2295	}
2296
2297	if (btrfs_fs_incompat(fs_info, RAID_STRIPE_TREE)) {
2298	location.objectid = BTRFS_RAID_STRIPE_TREE_OBJECTID;
2299	root = btrfs_read_tree_root(tree_root, key: &location);
2300	if (IS_ERR(ptr: root)) {
2301	if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2302	ret = PTR_ERR(ptr: root);
2303	goto out;
2304	}
2305	} else {
2306	set_bit(nr: BTRFS_ROOT_TRACK_DIRTY, addr: &root->state);
2307	fs_info->stripe_root = root;
2308	}
2309	}
2310
2311	return 0;
2312	out:
2313	btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2314	location.objectid, ret);
2315	return ret;
2316	}
2317
2318	static int validate_sys_chunk_array(const struct btrfs_fs_info *fs_info,
2319	const struct btrfs_super_block *sb)
2320	{
2321	unsigned int cur = 0; /* Offset inside the sys chunk array */
2322	/*
2323	* At sb read time, fs_info is not fully initialized. Thus we have
2324	* to use super block sectorsize, which should have been validated.
2325	*/
2326	const u32 sectorsize = btrfs_super_sectorsize(s: sb);
2327	u32 sys_array_size = btrfs_super_sys_array_size(s: sb);
2328
2329	if (sys_array_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2330	btrfs_err(fs_info, "system chunk array too big %u > %u",
2331	sys_array_size, BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2332	return -EUCLEAN;
2333	}
2334
2335	while (cur < sys_array_size) {
2336	struct btrfs_disk_key *disk_key;
2337	struct btrfs_chunk *chunk;
2338	struct btrfs_key key;
2339	u64 type;
2340	u16 num_stripes;
2341	u32 len;
2342	int ret;
2343
2344	disk_key = (struct btrfs_disk_key *)(sb->sys_chunk_array + cur);
2345	len = sizeof(*disk_key);
2346
2347	if (cur + len > sys_array_size)
2348	goto short_read;
2349	cur += len;
2350
2351	btrfs_disk_key_to_cpu(cpu_key: &key, disk_key);
2352	if (key.type != BTRFS_CHUNK_ITEM_KEY) {
2353	btrfs_err(fs_info,
2354	"unexpected item type %u in sys_array at offset %u",
2355	key.type, cur);
2356	return -EUCLEAN;
2357	}
2358	chunk = (struct btrfs_chunk *)(sb->sys_chunk_array + cur);
2359	num_stripes = btrfs_stack_chunk_num_stripes(s: chunk);
2360	if (cur + btrfs_chunk_item_size(num_stripes) > sys_array_size)
2361	goto short_read;
2362	type = btrfs_stack_chunk_type(s: chunk);
2363	if (!(type & BTRFS_BLOCK_GROUP_SYSTEM)) {
2364	btrfs_err(fs_info,
2365	"invalid chunk type %llu in sys_array at offset %u",
2366	type, cur);
2367	return -EUCLEAN;
2368	}
2369	ret = btrfs_check_chunk_valid(fs_info, NULL, chunk, logical: key.offset,
2370	sectorsize);
2371	if (ret < 0)
2372	return ret;
2373	cur += btrfs_chunk_item_size(num_stripes);
2374	}
2375	return 0;
2376	short_read:
2377	btrfs_err(fs_info,
2378	"super block sys chunk array short read, cur=%u sys_array_size=%u",
2379	cur, sys_array_size);
2380	return -EUCLEAN;
2381	}
2382
2383	/*
2384	* Real super block validation
2385	* NOTE: super csum type and incompat features will not be checked here.
2386	*
2387	* @sb: super block to check
2388	* @mirror_num: the super block number to check its bytenr:
2389	* 0 the primary (1st) sb
2390	* 1, 2 2nd and 3rd backup copy
2391	* -1 skip bytenr check
2392	*/
2393	int btrfs_validate_super(const struct btrfs_fs_info *fs_info,
2394	const struct btrfs_super_block *sb, int mirror_num)
2395	{
2396	u64 nodesize = btrfs_super_nodesize(s: sb);
2397	u64 sectorsize = btrfs_super_sectorsize(s: sb);
2398	int ret = 0;
2399	const bool ignore_flags = btrfs_test_opt(fs_info, IGNORESUPERFLAGS);
2400
2401	if (btrfs_super_magic(s: sb) != BTRFS_MAGIC) {
2402	btrfs_err(fs_info, "no valid FS found");
2403	ret = -EINVAL;
2404	}
2405	if ((btrfs_super_flags(s: sb) & ~BTRFS_SUPER_FLAG_SUPP)) {
2406	if (!ignore_flags) {
2407	btrfs_err(fs_info,
2408	"unrecognized or unsupported super flag 0x%llx",
2409	btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2410	ret = -EINVAL;
2411	} else {
2412	btrfs_info(fs_info,
2413	"unrecognized or unsupported super flags: 0x%llx, ignored",
2414	btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2415	}
2416	}
2417	if (btrfs_super_root_level(s: sb) >= BTRFS_MAX_LEVEL) {
2418	btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2419	btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2420	ret = -EINVAL;
2421	}
2422	if (btrfs_super_chunk_root_level(s: sb) >= BTRFS_MAX_LEVEL) {
2423	btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2424	btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2425	ret = -EINVAL;
2426	}
2427	if (btrfs_super_log_root_level(s: sb) >= BTRFS_MAX_LEVEL) {
2428	btrfs_err(fs_info, "log_root level too big: %d >= %d",
2429	btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2430	ret = -EINVAL;
2431	}
2432
2433	/*
2434	* Check sectorsize and nodesize first, other check will need it.
2435	* Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2436	*/
2437	if (!is_power_of_2(n: sectorsize) || sectorsize < BTRFS_MIN_BLOCKSIZE ||
2438	sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2439	btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2440	ret = -EINVAL;
2441	}
2442
2443	/*
2444	* We only support at most 3 sectorsizes: 4K, PAGE_SIZE, MIN_BLOCKSIZE.
2445	*
2446	* For 4K page sized systems with non-debug builds, all 3 matches (4K).
2447	* For 4K page sized systems with debug builds, there are two block sizes
2448	* supported. (4K and 2K)
2449	*
2450	* We can support 16K sectorsize with 64K page size without problem,
2451	* but such sectorsize/pagesize combination doesn't make much sense.
2452	* 4K will be our future standard, PAGE_SIZE is supported from the very
2453	* beginning.
2454	*/
2455	if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K &&
2456	sectorsize != PAGE_SIZE &&
2457	sectorsize != BTRFS_MIN_BLOCKSIZE)) {
2458	btrfs_err(fs_info,
2459	"sectorsize %llu not yet supported for page size %lu",
2460	sectorsize, PAGE_SIZE);
2461	ret = -EINVAL;
2462	}
2463
2464	if (!is_power_of_2(n: nodesize) || nodesize < sectorsize ||
2465	nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2466	btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2467	ret = -EINVAL;
2468	}
2469	if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2470	btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2471	le32_to_cpu(sb->__unused_leafsize), nodesize);
2472	ret = -EINVAL;
2473	}
2474
2475	/* Root alignment check */
2476	if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2477	btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2478	btrfs_super_root(sb));
2479	ret = -EINVAL;
2480	}
2481	if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2482	btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2483	btrfs_super_chunk_root(sb));
2484	ret = -EINVAL;
2485	}
2486	if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2487	btrfs_warn(fs_info, "log_root block unaligned: %llu",
2488	btrfs_super_log_root(sb));
2489	ret = -EINVAL;
2490	}
2491
2492	if (!fs_info->fs_devices->temp_fsid &&
2493	memcmp(p: fs_info->fs_devices->fsid, q: sb->fsid, BTRFS_FSID_SIZE) != 0) {
2494	btrfs_err(fs_info,
2495	"superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2496	sb->fsid, fs_info->fs_devices->fsid);
2497	ret = -EINVAL;
2498	}
2499
2500	if (memcmp(p: fs_info->fs_devices->metadata_uuid, q: btrfs_sb_fsid_ptr(sb),
2501	BTRFS_FSID_SIZE) != 0) {
2502	btrfs_err(fs_info,
2503	"superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2504	btrfs_sb_fsid_ptr(sb), fs_info->fs_devices->metadata_uuid);
2505	ret = -EINVAL;
2506	}
2507
2508	if (memcmp(p: fs_info->fs_devices->metadata_uuid, q: sb->dev_item.fsid,
2509	BTRFS_FSID_SIZE) != 0) {
2510	btrfs_err(fs_info,
2511	"dev_item UUID does not match metadata fsid: %pU != %pU",
2512	fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2513	ret = -EINVAL;
2514	}
2515
2516	/*
2517	* Artificial requirement for block-group-tree to force newer features
2518	* (free-space-tree, no-holes) so the test matrix is smaller.
2519	*/
2520	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2521	(!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2522	!btrfs_fs_incompat(fs_info, NO_HOLES))) {
2523	btrfs_err(fs_info,
2524	"block-group-tree feature requires free-space-tree and no-holes");
2525	ret = -EINVAL;
2526	}
2527
2528	/*
2529	* Hint to catch really bogus numbers, bitflips or so, more exact checks are
2530	* done later
2531	*/
2532	if (btrfs_super_bytes_used(s: sb) < 6 * btrfs_super_nodesize(s: sb)) {
2533	btrfs_err(fs_info, "bytes_used is too small %llu",
2534	btrfs_super_bytes_used(sb));
2535	ret = -EINVAL;
2536	}
2537	if (!is_power_of_2(n: btrfs_super_stripesize(s: sb))) {
2538	btrfs_err(fs_info, "invalid stripesize %u",
2539	btrfs_super_stripesize(sb));
2540	ret = -EINVAL;
2541	}
2542	if (btrfs_super_num_devices(s: sb) > (1UL << 31))
2543	btrfs_warn(fs_info, "suspicious number of devices: %llu",
2544	btrfs_super_num_devices(sb));
2545	if (btrfs_super_num_devices(s: sb) == 0) {
2546	btrfs_err(fs_info, "number of devices is 0");
2547	ret = -EINVAL;
2548	}
2549
2550	if (mirror_num >= 0 &&
2551	btrfs_super_bytenr(s: sb) != btrfs_sb_offset(mirror: mirror_num)) {
2552	btrfs_err(fs_info, "super offset mismatch %llu != %u",
2553	btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2554	ret = -EINVAL;
2555	}
2556
2557	if (ret)
2558	return ret;
2559
2560	ret = validate_sys_chunk_array(fs_info, sb);
2561
2562	/*
2563	* Obvious sys_chunk_array corruptions, it must hold at least one key
2564	* and one chunk
2565	*/
2566	if (btrfs_super_sys_array_size(s: sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2567	btrfs_err(fs_info, "system chunk array too big %u > %u",
2568	btrfs_super_sys_array_size(sb),
2569	BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2570	ret = -EINVAL;
2571	}
2572	if (btrfs_super_sys_array_size(s: sb) < sizeof(struct btrfs_disk_key)
2573	+ sizeof(struct btrfs_chunk)) {
2574	btrfs_err(fs_info, "system chunk array too small %u < %zu",
2575	btrfs_super_sys_array_size(sb),
2576	sizeof(struct btrfs_disk_key)
2577	+ sizeof(struct btrfs_chunk));
2578	ret = -EINVAL;
2579	}
2580
2581	/*
2582	* The generation is a global counter, we'll trust it more than the others
2583	* but it's still possible that it's the one that's wrong.
2584	*/
2585	if (btrfs_super_generation(s: sb) < btrfs_super_chunk_root_generation(s: sb))
2586	btrfs_warn(fs_info,
2587	"suspicious: generation < chunk_root_generation: %llu < %llu",
2588	btrfs_super_generation(sb),
2589	btrfs_super_chunk_root_generation(sb));
2590	if (btrfs_super_generation(s: sb) < btrfs_super_cache_generation(s: sb)
2591	&& btrfs_super_cache_generation(s: sb) != (u64)-1)
2592	btrfs_warn(fs_info,
2593	"suspicious: generation < cache_generation: %llu < %llu",
2594	btrfs_super_generation(sb),
2595	btrfs_super_cache_generation(sb));
2596
2597	return ret;
2598	}
2599
2600	/*
2601	* Validation of super block at mount time.
2602	* Some checks already done early at mount time, like csum type and incompat
2603	* flags will be skipped.
2604	*/
2605	static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2606	{
2607	return btrfs_validate_super(fs_info, sb: fs_info->super_copy, mirror_num: 0);
2608	}
2609
2610	/*
2611	* Validation of super block at write time.
2612	* Some checks like bytenr check will be skipped as their values will be
2613	* overwritten soon.
2614	* Extra checks like csum type and incompat flags will be done here.
2615	*/
2616	static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2617	struct btrfs_super_block *sb)
2618	{
2619	int ret;
2620
2621	ret = btrfs_validate_super(fs_info, sb, mirror_num: -1);
2622	if (ret < 0)
2623	goto out;
2624	if (!btrfs_supported_super_csum(csum_type: btrfs_super_csum_type(s: sb))) {
2625	ret = -EUCLEAN;
2626	btrfs_err(fs_info, "invalid csum type, has %u want %u",
2627	btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2628	goto out;
2629	}
2630	if (btrfs_super_incompat_flags(s: sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2631	ret = -EUCLEAN;
2632	btrfs_err(fs_info,
2633	"invalid incompat flags, has 0x%llx valid mask 0x%llx",
2634	btrfs_super_incompat_flags(sb),
2635	(unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2636	goto out;
2637	}
2638	out:
2639	if (ret < 0)
2640	btrfs_err(fs_info,
2641	"super block corruption detected before writing it to disk");
2642	return ret;
2643	}
2644
2645	static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2646	{
2647	struct btrfs_tree_parent_check check = {
2648	.level = level,
2649	.transid = gen,
2650	.owner_root = btrfs_root_id(root)
2651	};
2652	int ret = 0;
2653
2654	root->node = read_tree_block(fs_info: root->fs_info, bytenr, check: &check);
2655	if (IS_ERR(ptr: root->node)) {
2656	ret = PTR_ERR(ptr: root->node);
2657	root->node = NULL;
2658	return ret;
2659	}
2660	if (!extent_buffer_uptodate(eb: root->node)) {
2661	free_extent_buffer(eb: root->node);
2662	root->node = NULL;
2663	return -EIO;
2664	}
2665
2666	btrfs_set_root_node(item: &root->root_item, node: root->node);
2667	root->commit_root = btrfs_root_node(root);
2668	btrfs_set_root_refs(s: &root->root_item, val: 1);
2669	return ret;
2670	}
2671
2672	static int load_important_roots(struct btrfs_fs_info *fs_info)
2673	{
2674	struct btrfs_super_block *sb = fs_info->super_copy;
2675	u64 gen, bytenr;
2676	int level, ret;
2677
2678	bytenr = btrfs_super_root(s: sb);
2679	gen = btrfs_super_generation(s: sb);
2680	level = btrfs_super_root_level(s: sb);
2681	ret = load_super_root(root: fs_info->tree_root, bytenr, gen, level);
2682	if (ret) {
2683	btrfs_warn(fs_info, "couldn't read tree root");
2684	return ret;
2685	}
2686	return 0;
2687	}
2688
2689	static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2690	{
2691	int backup_index = find_newest_super_backup(info: fs_info);
2692	struct btrfs_super_block *sb = fs_info->super_copy;
2693	struct btrfs_root *tree_root = fs_info->tree_root;
2694	bool handle_error = false;
2695	int ret = 0;
2696	int i;
2697
2698	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2699	if (handle_error) {
2700	if (!IS_ERR(ptr: tree_root->node))
2701	free_extent_buffer(eb: tree_root->node);
2702	tree_root->node = NULL;
2703
2704	if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2705	break;
2706
2707	free_root_pointers(info: fs_info, free_chunk_root: 0);
2708
2709	/*
2710	* Don't use the log in recovery mode, it won't be
2711	* valid
2712	*/
2713	btrfs_set_super_log_root(s: sb, val: 0);
2714
2715	btrfs_warn(fs_info, "try to load backup roots slot %d", i);
2716	ret = read_backup_root(fs_info, priority: i);
2717	backup_index = ret;
2718	if (ret < 0)
2719	return ret;
2720	}
2721
2722	ret = load_important_roots(fs_info);
2723	if (ret) {
2724	handle_error = true;
2725	continue;
2726	}
2727
2728	/*
2729	* No need to hold btrfs_root::objectid_mutex since the fs
2730	* hasn't been fully initialised and we are the only user
2731	*/
2732	ret = btrfs_init_root_free_objectid(root: tree_root);
2733	if (ret < 0) {
2734	handle_error = true;
2735	continue;
2736	}
2737
2738	ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2739
2740	ret = btrfs_read_roots(fs_info);
2741	if (ret < 0) {
2742	handle_error = true;
2743	continue;
2744	}
2745
2746	/* All successful */
2747	fs_info->generation = btrfs_header_generation(eb: tree_root->node);
2748	btrfs_set_last_trans_committed(fs_info, gen: fs_info->generation);
2749	fs_info->last_reloc_trans = 0;
2750
2751	/* Always begin writing backup roots after the one being used */
2752	if (backup_index < 0) {
2753	fs_info->backup_root_index = 0;
2754	} else {
2755	fs_info->backup_root_index = backup_index + 1;
2756	fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2757	}
2758	break;
2759	}
2760
2761	return ret;
2762	}
2763
2764	/*
2765	* Lockdep gets confused between our buffer_tree which requires IRQ locking because
2766	* we modify marks in the IRQ context, and our delayed inode xarray which doesn't
2767	* have these requirements. Use a class key so lockdep doesn't get them mixed up.
2768	*/
2769	static struct lock_class_key buffer_xa_class;
2770
2771	void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2772	{
2773	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2774
2775	/* Use the same flags as mapping->i_pages. */
2776	xa_init_flags(xa: &fs_info->buffer_tree, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
2777	lockdep_set_class(&fs_info->buffer_tree.xa_lock, &buffer_xa_class);
2778
2779	INIT_LIST_HEAD(list: &fs_info->trans_list);
2780	INIT_LIST_HEAD(list: &fs_info->dead_roots);
2781	INIT_LIST_HEAD(list: &fs_info->delayed_iputs);
2782	INIT_LIST_HEAD(list: &fs_info->delalloc_roots);
2783	INIT_LIST_HEAD(list: &fs_info->caching_block_groups);
2784	spin_lock_init(&fs_info->delalloc_root_lock);
2785	spin_lock_init(&fs_info->trans_lock);
2786	spin_lock_init(&fs_info->fs_roots_radix_lock);
2787	spin_lock_init(&fs_info->delayed_iput_lock);
2788	spin_lock_init(&fs_info->defrag_inodes_lock);
2789	spin_lock_init(&fs_info->super_lock);
2790	spin_lock_init(&fs_info->unused_bgs_lock);
2791	spin_lock_init(&fs_info->treelog_bg_lock);
2792	spin_lock_init(&fs_info->zone_active_bgs_lock);
2793	spin_lock_init(&fs_info->relocation_bg_lock);
2794	rwlock_init(&fs_info->tree_mod_log_lock);
2795	rwlock_init(&fs_info->global_root_lock);
2796	mutex_init(&fs_info->unused_bg_unpin_mutex);
2797	mutex_init(&fs_info->reclaim_bgs_lock);
2798	mutex_init(&fs_info->reloc_mutex);
2799	mutex_init(&fs_info->delalloc_root_mutex);
2800	mutex_init(&fs_info->zoned_meta_io_lock);
2801	mutex_init(&fs_info->zoned_data_reloc_io_lock);
2802	seqlock_init(&fs_info->profiles_lock);
2803
2804	btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
2805	btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
2806	btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
2807	btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
2808	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_prep,
2809	BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2810	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
2811	BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2812	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
2813	BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2814	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
2815	BTRFS_LOCKDEP_TRANS_COMPLETED);
2816
2817	INIT_LIST_HEAD(list: &fs_info->dirty_cowonly_roots);
2818	INIT_LIST_HEAD(list: &fs_info->space_info);
2819	INIT_LIST_HEAD(list: &fs_info->tree_mod_seq_list);
2820	INIT_LIST_HEAD(list: &fs_info->unused_bgs);
2821	INIT_LIST_HEAD(list: &fs_info->reclaim_bgs);
2822	INIT_LIST_HEAD(list: &fs_info->zone_active_bgs);
2823	#ifdef CONFIG_BTRFS_DEBUG
2824	INIT_LIST_HEAD(list: &fs_info->allocated_roots);
2825	INIT_LIST_HEAD(list: &fs_info->allocated_ebs);
2826	spin_lock_init(&fs_info->eb_leak_lock);
2827	#endif
2828	fs_info->mapping_tree = RB_ROOT_CACHED;
2829	rwlock_init(&fs_info->mapping_tree_lock);
2830	btrfs_init_block_rsv(rsv: &fs_info->global_block_rsv,
2831	type: BTRFS_BLOCK_RSV_GLOBAL);
2832	btrfs_init_block_rsv(rsv: &fs_info->trans_block_rsv, type: BTRFS_BLOCK_RSV_TRANS);
2833	btrfs_init_block_rsv(rsv: &fs_info->chunk_block_rsv, type: BTRFS_BLOCK_RSV_CHUNK);
2834	btrfs_init_block_rsv(rsv: &fs_info->treelog_rsv, type: BTRFS_BLOCK_RSV_TREELOG);
2835	btrfs_init_block_rsv(rsv: &fs_info->empty_block_rsv, type: BTRFS_BLOCK_RSV_EMPTY);
2836	btrfs_init_block_rsv(rsv: &fs_info->delayed_block_rsv,
2837	type: BTRFS_BLOCK_RSV_DELOPS);
2838	btrfs_init_block_rsv(rsv: &fs_info->delayed_refs_rsv,
2839	type: BTRFS_BLOCK_RSV_DELREFS);
2840
2841	atomic_set(v: &fs_info->async_delalloc_pages, i: 0);
2842	atomic_set(v: &fs_info->defrag_running, i: 0);
2843	atomic_set(v: &fs_info->nr_delayed_iputs, i: 0);
2844	atomic64_set(v: &fs_info->tree_mod_seq, i: 0);
2845	fs_info->global_root_tree = RB_ROOT;
2846	fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2847	fs_info->metadata_ratio = 0;
2848	fs_info->defrag_inodes = RB_ROOT;
2849	atomic64_set(v: &fs_info->free_chunk_space, i: 0);
2850	fs_info->tree_mod_log = RB_ROOT;
2851	fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2852	btrfs_init_ref_verify(fs_info);
2853
2854	fs_info->thread_pool_size = min_t(unsigned long,
2855	num_online_cpus() + 2, 8);
2856
2857	INIT_LIST_HEAD(list: &fs_info->ordered_roots);
2858	spin_lock_init(&fs_info->ordered_root_lock);
2859
2860	btrfs_init_scrub(fs_info);
2861	btrfs_init_balance(fs_info);
2862	btrfs_init_async_reclaim_work(fs_info);
2863	btrfs_init_extent_map_shrinker_work(fs_info);
2864
2865	rwlock_init(&fs_info->block_group_cache_lock);
2866	fs_info->block_group_cache_tree = RB_ROOT_CACHED;
2867
2868	btrfs_extent_io_tree_init(fs_info, tree: &fs_info->excluded_extents,
2869	owner: IO_TREE_FS_EXCLUDED_EXTENTS);
2870
2871	mutex_init(&fs_info->ordered_operations_mutex);
2872	mutex_init(&fs_info->tree_log_mutex);
2873	mutex_init(&fs_info->chunk_mutex);
2874	mutex_init(&fs_info->transaction_kthread_mutex);
2875	mutex_init(&fs_info->cleaner_mutex);
2876	mutex_init(&fs_info->ro_block_group_mutex);
2877	init_rwsem(&fs_info->commit_root_sem);
2878	init_rwsem(&fs_info->cleanup_work_sem);
2879	init_rwsem(&fs_info->subvol_sem);
2880	sema_init(sem: &fs_info->uuid_tree_rescan_sem, val: 1);
2881
2882	btrfs_init_dev_replace_locks(fs_info);
2883	btrfs_init_qgroup(fs_info);
2884	btrfs_discard_init(fs_info);
2885
2886	btrfs_init_free_cluster(cluster: &fs_info->meta_alloc_cluster);
2887	btrfs_init_free_cluster(cluster: &fs_info->data_alloc_cluster);
2888
2889	init_waitqueue_head(&fs_info->transaction_throttle);
2890	init_waitqueue_head(&fs_info->transaction_wait);
2891	init_waitqueue_head(&fs_info->transaction_blocked_wait);
2892	init_waitqueue_head(&fs_info->async_submit_wait);
2893	init_waitqueue_head(&fs_info->delayed_iputs_wait);
2894
2895	/* Usable values until the real ones are cached from the superblock */
2896	fs_info->nodesize = 4096;
2897	fs_info->sectorsize = 4096;
2898	fs_info->sectorsize_bits = ilog2(4096);
2899	fs_info->stripesize = 4096;
2900
2901	/* Default compress algorithm when user does -o compress */
2902	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2903
2904	fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
2905
2906	spin_lock_init(&fs_info->swapfile_pins_lock);
2907	fs_info->swapfile_pins = RB_ROOT;
2908
2909	fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2910	INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2911	}
2912
2913	static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2914	{
2915	int ret;
2916
2917	fs_info->sb = sb;
2918	/* Temporary fixed values for block size until we read the superblock. */
2919	sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2920	sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2921
2922	ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2923	if (ret)
2924	return ret;
2925
2926	ret = percpu_counter_init(&fs_info->evictable_extent_maps, 0, GFP_KERNEL);
2927	if (ret)
2928	return ret;
2929
2930	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2931	if (ret)
2932	return ret;
2933
2934	ret = percpu_counter_init(&fs_info->stats_read_blocks, 0, GFP_KERNEL);
2935	if (ret)
2936	return ret;
2937
2938	fs_info->dirty_metadata_batch = PAGE_SIZE *
2939	(1 + ilog2(nr_cpu_ids));
2940
2941	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2942	if (ret)
2943	return ret;
2944
2945	ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2946	GFP_KERNEL);
2947	if (ret)
2948	return ret;
2949
2950	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2951	GFP_KERNEL);
2952	if (!fs_info->delayed_root)
2953	return -ENOMEM;
2954	btrfs_init_delayed_root(delayed_root: fs_info->delayed_root);
2955
2956	if (sb_rdonly(sb))
2957	set_bit(nr: BTRFS_FS_STATE_RO, addr: &fs_info->fs_state);
2958	if (btrfs_test_opt(fs_info, IGNOREMETACSUMS))
2959	set_bit(nr: BTRFS_FS_STATE_SKIP_META_CSUMS, addr: &fs_info->fs_state);
2960
2961	return btrfs_alloc_stripe_hash_table(info: fs_info);
2962	}
2963
2964	static int btrfs_uuid_rescan_kthread(void *data)
2965	{
2966	struct btrfs_fs_info *fs_info = data;
2967	int ret;
2968
2969	/*
2970	* 1st step is to iterate through the existing UUID tree and
2971	* to delete all entries that contain outdated data.
2972	* 2nd step is to add all missing entries to the UUID tree.
2973	*/
2974	ret = btrfs_uuid_tree_iterate(fs_info);
2975	if (ret < 0) {
2976	if (ret != -EINTR)
2977	btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2978	ret);
2979	up(sem: &fs_info->uuid_tree_rescan_sem);
2980	return ret;
2981	}
2982	return btrfs_uuid_scan_kthread(data);
2983	}
2984
2985	static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2986	{
2987	struct task_struct *task;
2988
2989	down(sem: &fs_info->uuid_tree_rescan_sem);
2990	task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2991	if (IS_ERR(ptr: task)) {
2992	/* fs_info->update_uuid_tree_gen remains 0 in all error case */
2993	btrfs_warn(fs_info, "failed to start uuid_rescan task");
2994	up(sem: &fs_info->uuid_tree_rescan_sem);
2995	return PTR_ERR(ptr: task);
2996	}
2997
2998	return 0;
2999	}
3000
3001	static int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3002	{
3003	u64 root_objectid = 0;
3004	struct btrfs_root *gang[8];
3005	int ret = 0;
3006
3007	while (1) {
3008	unsigned int found;
3009
3010	spin_lock(lock: &fs_info->fs_roots_radix_lock);
3011	found = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3012	results: (void **)gang, first_index: root_objectid,
3013	ARRAY_SIZE(gang));
3014	if (!found) {
3015	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
3016	break;
3017	}
3018	root_objectid = btrfs_root_id(root: gang[found - 1]) + 1;
3019
3020	for (int i = 0; i < found; i++) {
3021	/* Avoid to grab roots in dead_roots. */
3022	if (btrfs_root_refs(s: &gang[i]->root_item) == 0) {
3023	gang[i] = NULL;
3024	continue;
3025	}
3026	/* Grab all the search result for later use. */
3027	gang[i] = btrfs_grab_root(root: gang[i]);
3028	}
3029	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
3030
3031	for (int i = 0; i < found; i++) {
3032	if (!gang[i])
3033	continue;
3034	root_objectid = btrfs_root_id(root: gang[i]);
3035	/*
3036	* Continue to release the remaining roots after the first
3037	* error without cleanup and preserve the first error
3038	* for the return.
3039	*/
3040	if (!ret)
3041	ret = btrfs_orphan_cleanup(root: gang[i]);
3042	btrfs_put_root(root: gang[i]);
3043	}
3044	if (ret)
3045	break;
3046
3047	root_objectid++;
3048	}
3049	return ret;
3050	}
3051
3052	/*
3053	* Mounting logic specific to read-write file systems. Shared by open_ctree
3054	* and btrfs_remount when remounting from read-only to read-write.
3055	*/
3056	int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3057	{
3058	int ret;
3059	const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3060	bool rebuild_free_space_tree = false;
3061
3062	if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3063	btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3064	if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
3065	btrfs_warn(fs_info,
3066	"'clear_cache' option is ignored with extent tree v2");
3067	else
3068	rebuild_free_space_tree = true;
3069	} else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3070	!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3071	btrfs_warn(fs_info, "free space tree is invalid");
3072	rebuild_free_space_tree = true;
3073	}
3074
3075	if (rebuild_free_space_tree) {
3076	btrfs_info(fs_info, "rebuilding free space tree");
3077	ret = btrfs_rebuild_free_space_tree(fs_info);
3078	if (ret) {
3079	btrfs_warn(fs_info,
3080	"failed to rebuild free space tree: %d", ret);
3081	goto out;
3082	}
3083	}
3084
3085	if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3086	!btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
3087	btrfs_info(fs_info, "disabling free space tree");
3088	ret = btrfs_delete_free_space_tree(fs_info);
3089	if (ret) {
3090	btrfs_warn(fs_info,
3091	"failed to disable free space tree: %d", ret);
3092	goto out;
3093	}
3094	}
3095
3096	/*
3097	* btrfs_find_orphan_roots() is responsible for finding all the dead
3098	* roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3099	* them into the fs_info->fs_roots_radix tree. This must be done before
3100	* calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3101	* first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3102	* item before the root's tree is deleted - this means that if we unmount
3103	* or crash before the deletion completes, on the next mount we will not
3104	* delete what remains of the tree because the orphan item does not
3105	* exists anymore, which is what tells us we have a pending deletion.
3106	*/
3107	ret = btrfs_find_orphan_roots(fs_info);
3108	if (ret)
3109	goto out;
3110
3111	ret = btrfs_cleanup_fs_roots(fs_info);
3112	if (ret)
3113	goto out;
3114
3115	down_read(sem: &fs_info->cleanup_work_sem);
3116	if ((ret = btrfs_orphan_cleanup(root: fs_info->fs_root)) ||
3117	(ret = btrfs_orphan_cleanup(root: fs_info->tree_root))) {
3118	up_read(sem: &fs_info->cleanup_work_sem);
3119	goto out;
3120	}
3121	up_read(sem: &fs_info->cleanup_work_sem);
3122
3123	mutex_lock(&fs_info->cleaner_mutex);
3124	ret = btrfs_recover_relocation(fs_info);
3125	mutex_unlock(lock: &fs_info->cleaner_mutex);
3126	if (ret < 0) {
3127	btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3128	goto out;
3129	}
3130
3131	if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3132	!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3133	btrfs_info(fs_info, "creating free space tree");
3134	ret = btrfs_create_free_space_tree(fs_info);
3135	if (ret) {
3136	btrfs_warn(fs_info,
3137	"failed to create free space tree: %d", ret);
3138	goto out;
3139	}
3140	}
3141
3142	if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3143	ret = btrfs_set_free_space_cache_v1_active(fs_info, active: cache_opt);
3144	if (ret)
3145	goto out;
3146	}
3147
3148	ret = btrfs_resume_balance_async(fs_info);
3149	if (ret)
3150	goto out;
3151
3152	ret = btrfs_resume_dev_replace_async(fs_info);
3153	if (ret) {
3154	btrfs_warn(fs_info, "failed to resume dev_replace");
3155	goto out;
3156	}
3157
3158	btrfs_qgroup_rescan_resume(fs_info);
3159
3160	if (!fs_info->uuid_root) {
3161	btrfs_info(fs_info, "creating UUID tree");
3162	ret = btrfs_create_uuid_tree(fs_info);
3163	if (ret) {
3164	btrfs_warn(fs_info,
3165	"failed to create the UUID tree %d", ret);
3166	goto out;
3167	}
3168	}
3169
3170	out:
3171	return ret;
3172	}
3173
3174	/*
3175	* Do various sanity and dependency checks of different features.
3176	*
3177	* @is_rw_mount: If the mount is read-write.
3178	*
3179	* This is the place for less strict checks (like for subpage or artificial
3180	* feature dependencies).
3181	*
3182	* For strict checks or possible corruption detection, see
3183	* btrfs_validate_super().
3184	*
3185	* This should be called after btrfs_parse_options(), as some mount options
3186	* (space cache related) can modify on-disk format like free space tree and
3187	* screw up certain feature dependencies.
3188	*/
3189	int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3190	{
3191	struct btrfs_super_block *disk_super = fs_info->super_copy;
3192	u64 incompat = btrfs_super_incompat_flags(s: disk_super);
3193	const u64 compat_ro = btrfs_super_compat_ro_flags(s: disk_super);
3194	const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3195
3196	if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3197	btrfs_err(fs_info,
3198	"cannot mount because of unknown incompat features (0x%llx)",
3199	incompat);
3200	return -EINVAL;
3201	}
3202
3203	/* Runtime limitation for mixed block groups. */
3204	if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3205	(fs_info->sectorsize != fs_info->nodesize)) {
3206	btrfs_err(fs_info,
3207	"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3208	fs_info->nodesize, fs_info->sectorsize);
3209	return -EINVAL;
3210	}
3211
3212	/* Mixed backref is an always-enabled feature. */
3213	incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3214
3215	/* Set compression related flags just in case. */
3216	if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3217	incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3218	else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3219	incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3220
3221	/*
3222	* An ancient flag, which should really be marked deprecated.
3223	* Such runtime limitation doesn't really need a incompat flag.
3224	*/
3225	if (btrfs_super_nodesize(s: disk_super) > PAGE_SIZE)
3226	incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3227
3228	if (compat_ro_unsupp && is_rw_mount) {
3229	btrfs_err(fs_info,
3230	"cannot mount read-write because of unknown compat_ro features (0x%llx)",
3231	compat_ro);
3232	return -EINVAL;
3233	}
3234
3235	/*
3236	* We have unsupported RO compat features, although RO mounted, we
3237	* should not cause any metadata writes, including log replay.
3238	* Or we could screw up whatever the new feature requires.
3239	*/
3240	if (compat_ro_unsupp && btrfs_super_log_root(s: disk_super) &&
3241	!btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3242	btrfs_err(fs_info,
3243	"cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3244	compat_ro);
3245	return -EINVAL;
3246	}
3247
3248	/*
3249	* Artificial limitations for block group tree, to force
3250	* block-group-tree to rely on no-holes and free-space-tree.
3251	*/
3252	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3253	(!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3254	!btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3255	btrfs_err(fs_info,
3256	"block-group-tree feature requires no-holes and free-space-tree features");
3257	return -EINVAL;
3258	}
3259
3260	/*
3261	* Subpage runtime limitation on v1 cache.
3262	*
3263	* V1 space cache still has some hard codeed PAGE_SIZE usage, while
3264	* we're already defaulting to v2 cache, no need to bother v1 as it's
3265	* going to be deprecated anyway.
3266	*/
3267	if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3268	btrfs_warn(fs_info,
3269	"v1 space cache is not supported for page size %lu with sectorsize %u",
3270	PAGE_SIZE, fs_info->sectorsize);
3271	return -EINVAL;
3272	}
3273
3274	/* This can be called by remount, we need to protect the super block. */
3275	spin_lock(lock: &fs_info->super_lock);
3276	btrfs_set_super_incompat_flags(s: disk_super, val: incompat);
3277	spin_unlock(lock: &fs_info->super_lock);
3278
3279	return 0;
3280	}
3281
3282	int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices)
3283	{
3284	u32 sectorsize;
3285	u32 nodesize;
3286	u32 stripesize;
3287	u64 generation;
3288	u16 csum_type;
3289	struct btrfs_super_block *disk_super;
3290	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3291	struct btrfs_root *tree_root;
3292	struct btrfs_root *chunk_root;
3293	int ret;
3294	int level;
3295
3296	ret = init_mount_fs_info(fs_info, sb);
3297	if (ret)
3298	goto fail;
3299
3300	/* These need to be init'ed before we start creating inodes and such. */
3301	tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3302	GFP_KERNEL);
3303	fs_info->tree_root = tree_root;
3304	chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3305	GFP_KERNEL);
3306	fs_info->chunk_root = chunk_root;
3307	if (!tree_root || !chunk_root) {
3308	ret = -ENOMEM;
3309	goto fail;
3310	}
3311
3312	ret = btrfs_init_btree_inode(sb);
3313	if (ret)
3314	goto fail;
3315
3316	invalidate_bdev(bdev: fs_devices->latest_dev->bdev);
3317
3318	/*
3319	* Read super block and check the signature bytes only
3320	*/
3321	disk_super = btrfs_read_disk_super(bdev: fs_devices->latest_dev->bdev, copy_num: 0, drop_cache: false);
3322	if (IS_ERR(ptr: disk_super)) {
3323	ret = PTR_ERR(ptr: disk_super);
3324	goto fail_alloc;
3325	}
3326
3327	btrfs_info(fs_info, "first mount of filesystem %pU", disk_super->fsid);
3328	/*
3329	* Verify the type first, if that or the checksum value are
3330	* corrupted, we'll find out
3331	*/
3332	csum_type = btrfs_super_csum_type(s: disk_super);
3333	if (!btrfs_supported_super_csum(csum_type)) {
3334	btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3335	csum_type);
3336	ret = -EINVAL;
3337	btrfs_release_disk_super(super: disk_super);
3338	goto fail_alloc;
3339	}
3340
3341	fs_info->csum_size = btrfs_super_csum_size(s: disk_super);
3342
3343	ret = btrfs_init_csum_hash(fs_info, csum_type);
3344	if (ret) {
3345	btrfs_release_disk_super(super: disk_super);
3346	goto fail_alloc;
3347	}
3348
3349	/*
3350	* We want to check superblock checksum, the type is stored inside.
3351	* Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3352	*/
3353	if (btrfs_check_super_csum(fs_info, disk_sb: disk_super)) {
3354	btrfs_err(fs_info, "superblock checksum mismatch");
3355	ret = -EINVAL;
3356	btrfs_release_disk_super(super: disk_super);
3357	goto fail_alloc;
3358	}
3359
3360	/*
3361	* super_copy is zeroed at allocation time and we never touch the
3362	* following bytes up to INFO_SIZE, the checksum is calculated from
3363	* the whole block of INFO_SIZE
3364	*/
3365	memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3366	btrfs_release_disk_super(super: disk_super);
3367
3368	disk_super = fs_info->super_copy;
3369
3370	memcpy(fs_info->super_for_commit, fs_info->super_copy,
3371	sizeof(*fs_info->super_for_commit));
3372
3373	ret = btrfs_validate_mount_super(fs_info);
3374	if (ret) {
3375	btrfs_err(fs_info, "superblock contains fatal errors");
3376	ret = -EINVAL;
3377	goto fail_alloc;
3378	}
3379
3380	if (!btrfs_super_root(s: disk_super)) {
3381	btrfs_err(fs_info, "invalid superblock tree root bytenr");
3382	ret = -EINVAL;
3383	goto fail_alloc;
3384	}
3385
3386	/* check FS state, whether FS is broken. */
3387	if (btrfs_super_flags(s: disk_super) & BTRFS_SUPER_FLAG_ERROR)
3388	WRITE_ONCE(fs_info->fs_error, -EUCLEAN);
3389
3390	/* Set up fs_info before parsing mount options */
3391	nodesize = btrfs_super_nodesize(s: disk_super);
3392	sectorsize = btrfs_super_sectorsize(s: disk_super);
3393	stripesize = sectorsize;
3394	fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3395	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3396
3397	fs_info->nodesize = nodesize;
3398	fs_info->sectorsize = sectorsize;
3399	fs_info->sectorsize_bits = ilog2(sectorsize);
3400	fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(info: fs_info) / fs_info->csum_size;
3401	fs_info->stripesize = stripesize;
3402	fs_info->fs_devices->fs_info = fs_info;
3403
3404	/*
3405	* Handle the space caching options appropriately now that we have the
3406	* super block loaded and validated.
3407	*/
3408	btrfs_set_free_space_cache_settings(fs_info);
3409
3410	if (!btrfs_check_options(info: fs_info, mount_opt: &fs_info->mount_opt, flags: sb->s_flags)) {
3411	ret = -EINVAL;
3412	goto fail_alloc;
3413	}
3414
3415	ret = btrfs_check_features(fs_info, is_rw_mount: !sb_rdonly(sb));
3416	if (ret < 0)
3417	goto fail_alloc;
3418
3419	/*
3420	* At this point our mount options are validated, if we set ->max_inline
3421	* to something non-standard make sure we truncate it to sectorsize.
3422	*/
3423	fs_info->max_inline = min_t(u64, fs_info->max_inline, fs_info->sectorsize);
3424
3425	ret = btrfs_init_workqueues(fs_info);
3426	if (ret)
3427	goto fail_sb_buffer;
3428
3429	sb->s_bdi->ra_pages *= btrfs_super_num_devices(s: disk_super);
3430	sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3431
3432	/* Update the values for the current filesystem. */
3433	sb->s_blocksize = sectorsize;
3434	sb->s_blocksize_bits = blksize_bits(size: sectorsize);
3435	memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3436
3437	mutex_lock(&fs_info->chunk_mutex);
3438	ret = btrfs_read_sys_array(fs_info);
3439	mutex_unlock(lock: &fs_info->chunk_mutex);
3440	if (ret) {
3441	btrfs_err(fs_info, "failed to read the system array: %d", ret);
3442	goto fail_sb_buffer;
3443	}
3444
3445	generation = btrfs_super_chunk_root_generation(s: disk_super);
3446	level = btrfs_super_chunk_root_level(s: disk_super);
3447	ret = load_super_root(root: chunk_root, bytenr: btrfs_super_chunk_root(s: disk_super),
3448	gen: generation, level);
3449	if (ret) {
3450	btrfs_err(fs_info, "failed to read chunk root");
3451	goto fail_tree_roots;
3452	}
3453
3454	read_extent_buffer(eb: chunk_root->node, dst: fs_info->chunk_tree_uuid,
3455	offsetof(struct btrfs_header, chunk_tree_uuid),
3456	BTRFS_UUID_SIZE);
3457
3458	ret = btrfs_read_chunk_tree(fs_info);
3459	if (ret) {
3460	btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3461	goto fail_tree_roots;
3462	}
3463
3464	/*
3465	* At this point we know all the devices that make this filesystem,
3466	* including the seed devices but we don't know yet if the replace
3467	* target is required. So free devices that are not part of this
3468	* filesystem but skip the replace target device which is checked
3469	* below in btrfs_init_dev_replace().
3470	*/
3471	btrfs_free_extra_devids(fs_devices);
3472	if (!fs_devices->latest_dev->bdev) {
3473	btrfs_err(fs_info, "failed to read devices");
3474	ret = -EIO;
3475	goto fail_tree_roots;
3476	}
3477
3478	ret = init_tree_roots(fs_info);
3479	if (ret)
3480	goto fail_tree_roots;
3481
3482	/*
3483	* Get zone type information of zoned block devices. This will also
3484	* handle emulation of a zoned filesystem if a regular device has the
3485	* zoned incompat feature flag set.
3486	*/
3487	ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3488	if (ret) {
3489	btrfs_err(fs_info,
3490	"zoned: failed to read device zone info: %d", ret);
3491	goto fail_block_groups;
3492	}
3493
3494	/*
3495	* If we have a uuid root and we're not being told to rescan we need to
3496	* check the generation here so we can set the
3497	* BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3498	* transaction during a balance or the log replay without updating the
3499	* uuid generation, and then if we crash we would rescan the uuid tree,
3500	* even though it was perfectly fine.
3501	*/
3502	if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3503	fs_info->generation == btrfs_super_uuid_tree_generation(s: disk_super))
3504	set_bit(nr: BTRFS_FS_UPDATE_UUID_TREE_GEN, addr: &fs_info->flags);
3505
3506	ret = btrfs_verify_dev_extents(fs_info);
3507	if (ret) {
3508	btrfs_err(fs_info,
3509	"failed to verify dev extents against chunks: %d",
3510	ret);
3511	goto fail_block_groups;
3512	}
3513	ret = btrfs_recover_balance(fs_info);
3514	if (ret) {
3515	btrfs_err(fs_info, "failed to recover balance: %d", ret);
3516	goto fail_block_groups;
3517	}
3518
3519	ret = btrfs_init_dev_stats(fs_info);
3520	if (ret) {
3521	btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3522	goto fail_block_groups;
3523	}
3524
3525	ret = btrfs_init_dev_replace(fs_info);
3526	if (ret) {
3527	btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3528	goto fail_block_groups;
3529	}
3530
3531	ret = btrfs_check_zoned_mode(fs_info);
3532	if (ret) {
3533	btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3534	ret);
3535	goto fail_block_groups;
3536	}
3537
3538	ret = btrfs_sysfs_add_fsid(fs_devs: fs_devices);
3539	if (ret) {
3540	btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3541	ret);
3542	goto fail_block_groups;
3543	}
3544
3545	ret = btrfs_sysfs_add_mounted(fs_info);
3546	if (ret) {
3547	btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3548	goto fail_fsdev_sysfs;
3549	}
3550
3551	ret = btrfs_init_space_info(fs_info);
3552	if (ret) {
3553	btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3554	goto fail_sysfs;
3555	}
3556
3557	ret = btrfs_read_block_groups(info: fs_info);
3558	if (ret) {
3559	btrfs_err(fs_info, "failed to read block groups: %d", ret);
3560	goto fail_sysfs;
3561	}
3562
3563	btrfs_free_zone_cache(fs_info);
3564
3565	btrfs_check_active_zone_reservation(fs_info);
3566
3567	if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3568	!btrfs_check_rw_degradable(fs_info, NULL)) {
3569	btrfs_warn(fs_info,
3570	"writable mount is not allowed due to too many missing devices");
3571	ret = -EINVAL;
3572	goto fail_sysfs;
3573	}
3574
3575	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3576	"btrfs-cleaner");
3577	if (IS_ERR(ptr: fs_info->cleaner_kthread)) {
3578	ret = PTR_ERR(ptr: fs_info->cleaner_kthread);
3579	goto fail_sysfs;
3580	}
3581
3582	fs_info->transaction_kthread = kthread_run(transaction_kthread,
3583	tree_root,
3584	"btrfs-transaction");
3585	if (IS_ERR(ptr: fs_info->transaction_kthread)) {
3586	ret = PTR_ERR(ptr: fs_info->transaction_kthread);
3587	goto fail_cleaner;
3588	}
3589
3590	ret = btrfs_read_qgroup_config(fs_info);
3591	if (ret)
3592	goto fail_trans_kthread;
3593
3594	if (btrfs_build_ref_tree(fs_info))
3595	btrfs_err(fs_info, "couldn't build ref tree");
3596
3597	/* do not make disk changes in broken FS or nologreplay is given */
3598	if (btrfs_super_log_root(s: disk_super) != 0 &&
3599	!btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3600	btrfs_info(fs_info, "start tree-log replay");
3601	ret = btrfs_replay_log(fs_info, fs_devices);
3602	if (ret)
3603	goto fail_qgroup;
3604	}
3605
3606	fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, check_ref: true);
3607	if (IS_ERR(ptr: fs_info->fs_root)) {
3608	ret = PTR_ERR(ptr: fs_info->fs_root);
3609	btrfs_warn(fs_info, "failed to read fs tree: %d", ret);
3610	fs_info->fs_root = NULL;
3611	goto fail_qgroup;
3612	}
3613
3614	if (sb_rdonly(sb))
3615	return 0;
3616
3617	ret = btrfs_start_pre_rw_mount(fs_info);
3618	if (ret) {
3619	close_ctree(fs_info);
3620	return ret;
3621	}
3622	btrfs_discard_resume(fs_info);
3623
3624	if (fs_info->uuid_root &&
3625	(btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3626	fs_info->generation != btrfs_super_uuid_tree_generation(s: disk_super))) {
3627	btrfs_info(fs_info, "checking UUID tree");
3628	ret = btrfs_check_uuid_tree(fs_info);
3629	if (ret) {
3630	btrfs_warn(fs_info,
3631	"failed to check the UUID tree: %d", ret);
3632	close_ctree(fs_info);
3633	return ret;
3634	}
3635	}
3636
3637	set_bit(nr: BTRFS_FS_OPEN, addr: &fs_info->flags);
3638
3639	/* Kick the cleaner thread so it'll start deleting snapshots. */
3640	if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3641	wake_up_process(tsk: fs_info->cleaner_kthread);
3642
3643	return 0;
3644
3645	fail_qgroup:
3646	btrfs_free_qgroup_config(fs_info);
3647	fail_trans_kthread:
3648	kthread_stop(k: fs_info->transaction_kthread);
3649	btrfs_cleanup_transaction(fs_info);
3650	btrfs_free_fs_roots(fs_info);
3651	fail_cleaner:
3652	kthread_stop(k: fs_info->cleaner_kthread);
3653
3654	/*
3655	* make sure we're done with the btree inode before we stop our
3656	* kthreads
3657	*/
3658	filemap_write_and_wait(mapping: fs_info->btree_inode->i_mapping);
3659
3660	fail_sysfs:
3661	btrfs_sysfs_remove_mounted(fs_info);
3662
3663	fail_fsdev_sysfs:
3664	btrfs_sysfs_remove_fsid(fs_devs: fs_info->fs_devices);
3665
3666	fail_block_groups:
3667	btrfs_put_block_group_cache(info: fs_info);
3668
3669	fail_tree_roots:
3670	if (fs_info->data_reloc_root)
3671	btrfs_drop_and_free_fs_root(fs_info, root: fs_info->data_reloc_root);
3672	free_root_pointers(info: fs_info, free_chunk_root: true);
3673	invalidate_inode_pages2(mapping: fs_info->btree_inode->i_mapping);
3674
3675	fail_sb_buffer:
3676	btrfs_stop_all_workers(fs_info);
3677	btrfs_free_block_groups(info: fs_info);
3678	fail_alloc:
3679	btrfs_mapping_tree_free(fs_info);
3680
3681	iput(fs_info->btree_inode);
3682	fail:
3683	btrfs_close_devices(fs_devices: fs_info->fs_devices);
3684	ASSERT(ret < 0);
3685	return ret;
3686	}
3687	ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3688
3689	static void btrfs_end_super_write(struct bio *bio)
3690	{
3691	struct btrfs_device *device = bio->bi_private;
3692	struct folio_iter fi;
3693
3694	bio_for_each_folio_all(fi, bio) {
3695	if (bio->bi_status) {
3696	btrfs_warn_rl_in_rcu(device->fs_info,
3697	"lost super block write due to IO error on %s (%d)",
3698	btrfs_dev_name(device),
3699	blk_status_to_errno(bio->bi_status));
3700	btrfs_dev_stat_inc_and_print(dev: device,
3701	index: BTRFS_DEV_STAT_WRITE_ERRS);
3702	/* Ensure failure if the primary sb fails. */
3703	if (bio->bi_opf & REQ_FUA)
3704	atomic_add(BTRFS_SUPER_PRIMARY_WRITE_ERROR,
3705	v: &device->sb_write_errors);
3706	else
3707	atomic_inc(v: &device->sb_write_errors);
3708	}
3709	folio_unlock(folio: fi.folio);
3710	folio_put(folio: fi.folio);
3711	}
3712
3713	bio_put(bio);
3714	}
3715
3716	/*
3717	* Write superblock @sb to the @device. Do not wait for completion, all the
3718	* folios we use for writing are locked.
3719	*
3720	* Write @max_mirrors copies of the superblock, where 0 means default that fit
3721	* the expected device size at commit time. Note that max_mirrors must be
3722	* same for write and wait phases.
3723	*
3724	* Return number of errors when folio is not found or submission fails.
3725	*/
3726	static int write_dev_supers(struct btrfs_device *device,
3727	struct btrfs_super_block *sb, int max_mirrors)
3728	{
3729	struct btrfs_fs_info *fs_info = device->fs_info;
3730	struct address_space *mapping = device->bdev->bd_mapping;
3731	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3732	int i;
3733	int ret;
3734	u64 bytenr, bytenr_orig;
3735
3736	atomic_set(v: &device->sb_write_errors, i: 0);
3737
3738	if (max_mirrors == 0)
3739	max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3740
3741	shash->tfm = fs_info->csum_shash;
3742
3743	for (i = 0; i < max_mirrors; i++) {
3744	struct folio *folio;
3745	struct bio *bio;
3746	struct btrfs_super_block *disk_super;
3747	size_t offset;
3748
3749	bytenr_orig = btrfs_sb_offset(mirror: i);
3750	ret = btrfs_sb_log_location(device, mirror: i, WRITE, bytenr_ret: &bytenr);
3751	if (ret == -ENOENT) {
3752	continue;
3753	} else if (ret < 0) {
3754	btrfs_err(device->fs_info,
3755	"couldn't get super block location for mirror %d error %d",
3756	i, ret);
3757	atomic_inc(v: &device->sb_write_errors);
3758	continue;
3759	}
3760	if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3761	device->commit_total_bytes)
3762	break;
3763
3764	btrfs_set_super_bytenr(s: sb, val: bytenr_orig);
3765
3766	crypto_shash_digest(desc: shash, data: (const char *)sb + BTRFS_CSUM_SIZE,
3767	BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3768	out: sb->csum);
3769
3770	folio = __filemap_get_folio(mapping, index: bytenr >> PAGE_SHIFT,
3771	FGP_LOCK | FGP_ACCESSED | FGP_CREAT,
3772	GFP_NOFS);
3773	if (IS_ERR(ptr: folio)) {
3774	btrfs_err(device->fs_info,
3775	"couldn't get super block page for bytenr %llu error %ld",
3776	bytenr, PTR_ERR(folio));
3777	atomic_inc(v: &device->sb_write_errors);
3778	continue;
3779	}
3780
3781	offset = offset_in_folio(folio, bytenr);
3782	disk_super = folio_address(folio) + offset;
3783	memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3784
3785	/*
3786	* Directly use bios here instead of relying on the page cache
3787	* to do I/O, so we don't lose the ability to do integrity
3788	* checking.
3789	*/
3790	bio = bio_alloc(bdev: device->bdev, nr_vecs: 1,
3791	opf: REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
3792	GFP_NOFS);
3793	bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3794	bio->bi_private = device;
3795	bio->bi_end_io = btrfs_end_super_write;
3796	bio_add_folio_nofail(bio, folio, BTRFS_SUPER_INFO_SIZE, off: offset);
3797
3798	/*
3799	* We FUA only the first super block. The others we allow to
3800	* go down lazy and there's a short window where the on-disk
3801	* copies might still contain the older version.
3802	*/
3803	if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3804	bio->bi_opf |= REQ_FUA;
3805	submit_bio(bio);
3806
3807	if (btrfs_advance_sb_log(device, mirror: i))
3808	atomic_inc(v: &device->sb_write_errors);
3809	}
3810	return atomic_read(v: &device->sb_write_errors) < i ? 0 : -1;
3811	}
3812
3813	/*
3814	* Wait for write completion of superblocks done by write_dev_supers,
3815	* @max_mirrors same for write and wait phases.
3816	*
3817	* Return -1 if primary super block write failed or when there were no super block
3818	* copies written. Otherwise 0.
3819	*/
3820	static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3821	{
3822	int i;
3823	int errors = 0;
3824	bool primary_failed = false;
3825	int ret;
3826	u64 bytenr;
3827
3828	if (max_mirrors == 0)
3829	max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3830
3831	for (i = 0; i < max_mirrors; i++) {
3832	struct folio *folio;
3833
3834	ret = btrfs_sb_log_location(device, mirror: i, READ, bytenr_ret: &bytenr);
3835	if (ret == -ENOENT) {
3836	break;
3837	} else if (ret < 0) {
3838	errors++;
3839	if (i == 0)
3840	primary_failed = true;
3841	continue;
3842	}
3843	if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3844	device->commit_total_bytes)
3845	break;
3846
3847	folio = filemap_get_folio(mapping: device->bdev->bd_mapping,
3848	index: bytenr >> PAGE_SHIFT);
3849	/* If the folio has been removed, then we know it completed. */
3850	if (IS_ERR(ptr: folio))
3851	continue;
3852
3853	/* Folio will be unlocked once the write completes. */
3854	folio_wait_locked(folio);
3855	folio_put(folio);
3856	}
3857
3858	errors += atomic_read(v: &device->sb_write_errors);
3859	if (errors >= BTRFS_SUPER_PRIMARY_WRITE_ERROR)
3860	primary_failed = true;
3861	if (primary_failed) {
3862	btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3863	device->devid);
3864	return -1;
3865	}
3866
3867	return errors < i ? 0 : -1;
3868	}
3869
3870	/*
3871	* endio for the write_dev_flush, this will wake anyone waiting
3872	* for the barrier when it is done
3873	*/
3874	static void btrfs_end_empty_barrier(struct bio *bio)
3875	{
3876	bio_uninit(bio);
3877	complete(bio->bi_private);
3878	}
3879
3880	/*
3881	* Submit a flush request to the device if it supports it. Error handling is
3882	* done in the waiting counterpart.
3883	*/
3884	static void write_dev_flush(struct btrfs_device *device)
3885	{
3886	struct bio *bio = &device->flush_bio;
3887
3888	device->last_flush_error = BLK_STS_OK;
3889
3890	bio_init(bio, bdev: device->bdev, NULL, max_vecs: 0,
3891	opf: REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
3892	bio->bi_end_io = btrfs_end_empty_barrier;
3893	init_completion(x: &device->flush_wait);
3894	bio->bi_private = &device->flush_wait;
3895	submit_bio(bio);
3896	set_bit(BTRFS_DEV_STATE_FLUSH_SENT, addr: &device->dev_state);
3897	}
3898
3899	/*
3900	* If the flush bio has been submitted by write_dev_flush, wait for it.
3901	* Return true for any error, and false otherwise.
3902	*/
3903	static bool wait_dev_flush(struct btrfs_device *device)
3904	{
3905	struct bio *bio = &device->flush_bio;
3906
3907	if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, addr: &device->dev_state))
3908	return false;
3909
3910	wait_for_completion_io(&device->flush_wait);
3911
3912	if (bio->bi_status) {
3913	device->last_flush_error = bio->bi_status;
3914	btrfs_dev_stat_inc_and_print(dev: device, index: BTRFS_DEV_STAT_FLUSH_ERRS);
3915	return true;
3916	}
3917
3918	return false;
3919	}
3920
3921	/*
3922	* send an empty flush down to each device in parallel,
3923	* then wait for them
3924	*/
3925	static int barrier_all_devices(struct btrfs_fs_info *info)
3926	{
3927	struct list_head *head;
3928	struct btrfs_device *dev;
3929	int errors_wait = 0;
3930
3931	lockdep_assert_held(&info->fs_devices->device_list_mutex);
3932	/* send down all the barriers */
3933	head = &info->fs_devices->devices;
3934	list_for_each_entry(dev, head, dev_list) {
3935	if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3936	continue;
3937	if (!dev->bdev)
3938	continue;
3939	if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3940	!test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3941	continue;
3942
3943	write_dev_flush(device: dev);
3944	}
3945
3946	/* wait for all the barriers */
3947	list_for_each_entry(dev, head, dev_list) {
3948	if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3949	continue;
3950	if (!dev->bdev) {
3951	errors_wait++;
3952	continue;
3953	}
3954	if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3955	!test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3956	continue;
3957
3958	if (wait_dev_flush(device: dev))
3959	errors_wait++;
3960	}
3961
3962	/*
3963	* Checks last_flush_error of disks in order to determine the device
3964	* state.
3965	*/
3966	if (errors_wait && !btrfs_check_rw_degradable(fs_info: info, NULL))
3967	return -EIO;
3968
3969	return 0;
3970	}
3971
3972	int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3973	{
3974	int raid_type;
3975	int min_tolerated = INT_MAX;
3976
3977	if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3978	(flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3979	min_tolerated = min_t(int, min_tolerated,
3980	btrfs_raid_array[BTRFS_RAID_SINGLE].
3981	tolerated_failures);
3982
3983	for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3984	if (raid_type == BTRFS_RAID_SINGLE)
3985	continue;
3986	if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3987	continue;
3988	min_tolerated = min_t(int, min_tolerated,
3989	btrfs_raid_array[raid_type].
3990	tolerated_failures);
3991	}
3992
3993	if (min_tolerated == INT_MAX) {
3994	pr_warn("BTRFS: unknown raid flag: %llu", flags);
3995	min_tolerated = 0;
3996	}
3997
3998	return min_tolerated;
3999	}
4000
4001	int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4002	{
4003	struct list_head *head;
4004	struct btrfs_device *dev;
4005	struct btrfs_super_block *sb;
4006	struct btrfs_dev_item *dev_item;
4007	int ret;
4008	int do_barriers;
4009	int max_errors;
4010	int total_errors = 0;
4011	u64 flags;
4012
4013	do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4014
4015	/*
4016	* max_mirrors == 0 indicates we're from commit_transaction,
4017	* not from fsync where the tree roots in fs_info have not
4018	* been consistent on disk.
4019	*/
4020	if (max_mirrors == 0)
4021	backup_super_roots(info: fs_info);
4022
4023	sb = fs_info->super_for_commit;
4024	dev_item = &sb->dev_item;
4025
4026	mutex_lock(&fs_info->fs_devices->device_list_mutex);
4027	head = &fs_info->fs_devices->devices;
4028	max_errors = btrfs_super_num_devices(s: fs_info->super_copy) - 1;
4029
4030	if (do_barriers) {
4031	ret = barrier_all_devices(info: fs_info);
4032	if (ret) {
4033	mutex_unlock(
4034	lock: &fs_info->fs_devices->device_list_mutex);
4035	btrfs_handle_fs_error(fs_info, ret,
4036	"errors while submitting device barriers.");
4037	return ret;
4038	}
4039	}
4040
4041	list_for_each_entry(dev, head, dev_list) {
4042	if (!dev->bdev) {
4043	total_errors++;
4044	continue;
4045	}
4046	if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4047	!test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4048	continue;
4049
4050	btrfs_set_stack_device_generation(s: dev_item, val: 0);
4051	btrfs_set_stack_device_type(s: dev_item, val: dev->type);
4052	btrfs_set_stack_device_id(s: dev_item, val: dev->devid);
4053	btrfs_set_stack_device_total_bytes(s: dev_item,
4054	val: dev->commit_total_bytes);
4055	btrfs_set_stack_device_bytes_used(s: dev_item,
4056	val: dev->commit_bytes_used);
4057	btrfs_set_stack_device_io_align(s: dev_item, val: dev->io_align);
4058	btrfs_set_stack_device_io_width(s: dev_item, val: dev->io_width);
4059	btrfs_set_stack_device_sector_size(s: dev_item, val: dev->sector_size);
4060	memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4061	memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4062	BTRFS_FSID_SIZE);
4063
4064	flags = btrfs_super_flags(s: sb);
4065	btrfs_set_super_flags(s: sb, val: flags | BTRFS_HEADER_FLAG_WRITTEN);
4066
4067	ret = btrfs_validate_write_super(fs_info, sb);
4068	if (ret < 0) {
4069	mutex_unlock(lock: &fs_info->fs_devices->device_list_mutex);
4070	btrfs_handle_fs_error(fs_info, -EUCLEAN,
4071	"unexpected superblock corruption detected");
4072	return -EUCLEAN;
4073	}
4074
4075	ret = write_dev_supers(device: dev, sb, max_mirrors);
4076	if (ret)
4077	total_errors++;
4078	}
4079	if (total_errors > max_errors) {
4080	btrfs_err(fs_info, "%d errors while writing supers",
4081	total_errors);
4082	mutex_unlock(lock: &fs_info->fs_devices->device_list_mutex);
4083
4084	/* FUA is masked off if unsupported and can't be the reason */
4085	btrfs_handle_fs_error(fs_info, -EIO,
4086	"%d errors while writing supers",
4087	total_errors);
4088	return -EIO;
4089	}
4090
4091	total_errors = 0;
4092	list_for_each_entry(dev, head, dev_list) {
4093	if (!dev->bdev)
4094	continue;
4095	if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4096	!test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4097	continue;
4098
4099	ret = wait_dev_supers(device: dev, max_mirrors);
4100	if (ret)
4101	total_errors++;
4102	}
4103	mutex_unlock(lock: &fs_info->fs_devices->device_list_mutex);
4104	if (total_errors > max_errors) {
4105	btrfs_handle_fs_error(fs_info, -EIO,
4106	"%d errors while writing supers",
4107	total_errors);
4108	return -EIO;
4109	}
4110	return 0;
4111	}
4112
4113	/* Drop a fs root from the radix tree and free it. */
4114	void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4115	struct btrfs_root *root)
4116	{
4117	bool drop_ref = false;
4118
4119	spin_lock(lock: &fs_info->fs_roots_radix_lock);
4120	radix_tree_delete(&fs_info->fs_roots_radix,
4121	(unsigned long)btrfs_root_id(root));
4122	if (test_and_clear_bit(nr: BTRFS_ROOT_IN_RADIX, addr: &root->state))
4123	drop_ref = true;
4124	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
4125
4126	if (BTRFS_FS_ERROR(fs_info)) {
4127	ASSERT(root->log_root == NULL);
4128	if (root->reloc_root) {
4129	btrfs_put_root(root: root->reloc_root);
4130	root->reloc_root = NULL;
4131	}
4132	}
4133
4134	if (drop_ref)
4135	btrfs_put_root(root);
4136	}
4137
4138	int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4139	{
4140	mutex_lock(&fs_info->cleaner_mutex);
4141	btrfs_run_delayed_iputs(fs_info);
4142	mutex_unlock(lock: &fs_info->cleaner_mutex);
4143	wake_up_process(tsk: fs_info->cleaner_kthread);
4144
4145	/* wait until ongoing cleanup work done */
4146	down_write(sem: &fs_info->cleanup_work_sem);
4147	up_write(sem: &fs_info->cleanup_work_sem);
4148
4149	return btrfs_commit_current_transaction(root: fs_info->tree_root);
4150	}
4151
4152	static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4153	{
4154	struct btrfs_transaction *trans;
4155	struct btrfs_transaction *tmp;
4156	bool found = false;
4157
4158	/*
4159	* This function is only called at the very end of close_ctree(),
4160	* thus no other running transaction, no need to take trans_lock.
4161	*/
4162	ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4163	list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4164	struct extent_state *cached = NULL;
4165	u64 dirty_bytes = 0;
4166	u64 cur = 0;
4167	u64 found_start;
4168	u64 found_end;
4169
4170	found = true;
4171	while (btrfs_find_first_extent_bit(tree: &trans->dirty_pages, start: cur,
4172	start_ret: &found_start, end_ret: &found_end,
4173	bits: EXTENT_DIRTY, cached_state: &cached)) {
4174	dirty_bytes += found_end + 1 - found_start;
4175	cur = found_end + 1;
4176	}
4177	btrfs_warn(fs_info,
4178	"transaction %llu (with %llu dirty metadata bytes) is not committed",
4179	trans->transid, dirty_bytes);
4180	btrfs_cleanup_one_transaction(trans);
4181
4182	if (trans == fs_info->running_transaction)
4183	fs_info->running_transaction = NULL;
4184	list_del_init(entry: &trans->list);
4185
4186	btrfs_put_transaction(transaction: trans);
4187	trace_btrfs_transaction_commit(fs_info);
4188	}
4189	ASSERT(!found);
4190	}
4191
4192	void __cold close_ctree(struct btrfs_fs_info *fs_info)
4193	{
4194	int ret;
4195
4196	set_bit(nr: BTRFS_FS_CLOSING_START, addr: &fs_info->flags);
4197
4198	/*
4199	* If we had UNFINISHED_DROPS we could still be processing them, so
4200	* clear that bit and wake up relocation so it can stop.
4201	* We must do this before stopping the block group reclaim task, because
4202	* at btrfs_relocate_block_group() we wait for this bit, and after the
4203	* wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4204	* have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4205	* return 1.
4206	*/
4207	btrfs_wake_unfinished_drop(fs_info);
4208
4209	/*
4210	* We may have the reclaim task running and relocating a data block group,
4211	* in which case it may create delayed iputs. So stop it before we park
4212	* the cleaner kthread otherwise we can get new delayed iputs after
4213	* parking the cleaner, and that can make the async reclaim task to hang
4214	* if it's waiting for delayed iputs to complete, since the cleaner is
4215	* parked and can not run delayed iputs - this will make us hang when
4216	* trying to stop the async reclaim task.
4217	*/
4218	cancel_work_sync(work: &fs_info->reclaim_bgs_work);
4219	/*
4220	* We don't want the cleaner to start new transactions, add more delayed
4221	* iputs, etc. while we're closing. We can't use kthread_stop() yet
4222	* because that frees the task_struct, and the transaction kthread might
4223	* still try to wake up the cleaner.
4224	*/
4225	kthread_park(k: fs_info->cleaner_kthread);
4226
4227	/* wait for the qgroup rescan worker to stop */
4228	btrfs_qgroup_wait_for_completion(fs_info, interruptible: false);
4229
4230	/* wait for the uuid_scan task to finish */
4231	down(sem: &fs_info->uuid_tree_rescan_sem);
4232	/* avoid complains from lockdep et al., set sem back to initial state */
4233	up(sem: &fs_info->uuid_tree_rescan_sem);
4234
4235	/* pause restriper - we want to resume on mount */
4236	btrfs_pause_balance(fs_info);
4237
4238	btrfs_dev_replace_suspend_for_unmount(fs_info);
4239
4240	btrfs_scrub_cancel(info: fs_info);
4241
4242	/* wait for any defraggers to finish */
4243	wait_event(fs_info->transaction_wait,
4244	(atomic_read(&fs_info->defrag_running) == 0));
4245
4246	/* clear out the rbtree of defraggable inodes */
4247	btrfs_cleanup_defrag_inodes(fs_info);
4248
4249	/*
4250	* Handle the error fs first, as it will flush and wait for all ordered
4251	* extents. This will generate delayed iputs, thus we want to handle
4252	* it first.
4253	*/
4254	if (unlikely(BTRFS_FS_ERROR(fs_info)))
4255	btrfs_error_commit_super(fs_info);
4256
4257	/*
4258	* Wait for any fixup workers to complete.
4259	* If we don't wait for them here and they are still running by the time
4260	* we call kthread_stop() against the cleaner kthread further below, we
4261	* get an use-after-free on the cleaner because the fixup worker adds an
4262	* inode to the list of delayed iputs and then attempts to wakeup the
4263	* cleaner kthread, which was already stopped and destroyed. We parked
4264	* already the cleaner, but below we run all pending delayed iputs.
4265	*/
4266	btrfs_flush_workqueue(wq: fs_info->fixup_workers);
4267	/*
4268	* Similar case here, we have to wait for delalloc workers before we
4269	* proceed below and stop the cleaner kthread, otherwise we trigger a
4270	* use-after-tree on the cleaner kthread task_struct when a delalloc
4271	* worker running submit_compressed_extents() adds a delayed iput, which
4272	* does a wake up on the cleaner kthread, which was already freed below
4273	* when we call kthread_stop().
4274	*/
4275	btrfs_flush_workqueue(wq: fs_info->delalloc_workers);
4276
4277	/*
4278	* We can have ordered extents getting their last reference dropped from
4279	* the fs_info->workers queue because for async writes for data bios we
4280	* queue a work for that queue, at btrfs_wq_submit_bio(), that runs
4281	* run_one_async_done() which calls btrfs_bio_end_io() in case the bio
4282	* has an error, and that later function can do the final
4283	* btrfs_put_ordered_extent() on the ordered extent attached to the bio,
4284	* which adds a delayed iput for the inode. So we must flush the queue
4285	* so that we don't have delayed iputs after committing the current
4286	* transaction below and stopping the cleaner and transaction kthreads.
4287	*/
4288	btrfs_flush_workqueue(wq: fs_info->workers);
4289
4290	/*
4291	* When finishing a compressed write bio we schedule a work queue item
4292	* to finish an ordered extent - btrfs_finish_compressed_write_work()
4293	* calls btrfs_finish_ordered_extent() which in turns does a call to
4294	* btrfs_queue_ordered_fn(), and that queues the ordered extent
4295	* completion either in the endio_write_workers work queue or in the
4296	* fs_info->endio_freespace_worker work queue. We flush those queues
4297	* below, so before we flush them we must flush this queue for the
4298	* workers of compressed writes.
4299	*/
4300	flush_workqueue(fs_info->compressed_write_workers);
4301
4302	/*
4303	* After we parked the cleaner kthread, ordered extents may have
4304	* completed and created new delayed iputs. If one of the async reclaim
4305	* tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4306	* can hang forever trying to stop it, because if a delayed iput is
4307	* added after it ran btrfs_run_delayed_iputs() and before it called
4308	* btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4309	* no one else to run iputs.
4310	*
4311	* So wait for all ongoing ordered extents to complete and then run
4312	* delayed iputs. This works because once we reach this point no one
4313	* can either create new ordered extents nor create delayed iputs
4314	* through some other means.
4315	*
4316	* Also note that btrfs_wait_ordered_roots() is not safe here, because
4317	* it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4318	* but the delayed iput for the respective inode is made only when doing
4319	* the final btrfs_put_ordered_extent() (which must happen at
4320	* btrfs_finish_ordered_io() when we are unmounting).
4321	*/
4322	btrfs_flush_workqueue(wq: fs_info->endio_write_workers);
4323	/* Ordered extents for free space inodes. */
4324	btrfs_flush_workqueue(wq: fs_info->endio_freespace_worker);
4325	btrfs_run_delayed_iputs(fs_info);
4326	/* There should be no more workload to generate new delayed iputs. */
4327	set_bit(nr: BTRFS_FS_STATE_NO_DELAYED_IPUT, addr: &fs_info->fs_state);
4328
4329	cancel_work_sync(work: &fs_info->async_reclaim_work);
4330	cancel_work_sync(work: &fs_info->async_data_reclaim_work);
4331	cancel_work_sync(work: &fs_info->preempt_reclaim_work);
4332	cancel_work_sync(work: &fs_info->em_shrinker_work);
4333
4334	/* Cancel or finish ongoing discard work */
4335	btrfs_discard_cleanup(fs_info);
4336
4337	if (!sb_rdonly(sb: fs_info->sb)) {
4338	/*
4339	* The cleaner kthread is stopped, so do one final pass over
4340	* unused block groups.
4341	*/
4342	btrfs_delete_unused_bgs(fs_info);
4343
4344	/*
4345	* There might be existing delayed inode workers still running
4346	* and holding an empty delayed inode item. We must wait for
4347	* them to complete first because they can create a transaction.
4348	* This happens when someone calls btrfs_balance_delayed_items()
4349	* and then a transaction commit runs the same delayed nodes
4350	* before any delayed worker has done something with the nodes.
4351	* We must wait for any worker here and not at transaction
4352	* commit time since that could cause a deadlock.
4353	* This is a very rare case.
4354	*/
4355	btrfs_flush_workqueue(wq: fs_info->delayed_workers);
4356
4357	ret = btrfs_commit_super(fs_info);
4358	if (ret)
4359	btrfs_err(fs_info, "commit super ret %d", ret);
4360	}
4361
4362	kthread_stop(k: fs_info->transaction_kthread);
4363	kthread_stop(k: fs_info->cleaner_kthread);
4364
4365	ASSERT(list_empty(&fs_info->delayed_iputs));
4366	set_bit(nr: BTRFS_FS_CLOSING_DONE, addr: &fs_info->flags);
4367
4368	if (btrfs_check_quota_leak(fs_info)) {
4369	DEBUG_WARN("qgroup reserved space leaked");
4370	btrfs_err(fs_info, "qgroup reserved space leaked");
4371	}
4372
4373	btrfs_free_qgroup_config(fs_info);
4374	ASSERT(list_empty(&fs_info->delalloc_roots));
4375
4376	if (percpu_counter_sum(fbc: &fs_info->delalloc_bytes)) {
4377	btrfs_info(fs_info, "at unmount delalloc count %lld",
4378	percpu_counter_sum(&fs_info->delalloc_bytes));
4379	}
4380
4381	if (percpu_counter_sum(fbc: &fs_info->ordered_bytes))
4382	btrfs_info(fs_info, "at unmount dio bytes count %lld",
4383	percpu_counter_sum(&fs_info->ordered_bytes));
4384
4385	btrfs_sysfs_remove_mounted(fs_info);
4386	btrfs_sysfs_remove_fsid(fs_devs: fs_info->fs_devices);
4387
4388	btrfs_put_block_group_cache(info: fs_info);
4389
4390	/*
4391	* we must make sure there is not any read request to
4392	* submit after we stopping all workers.
4393	*/
4394	invalidate_inode_pages2(mapping: fs_info->btree_inode->i_mapping);
4395	btrfs_stop_all_workers(fs_info);
4396
4397	/* We shouldn't have any transaction open at this point */
4398	warn_about_uncommitted_trans(fs_info);
4399
4400	clear_bit(nr: BTRFS_FS_OPEN, addr: &fs_info->flags);
4401	free_root_pointers(info: fs_info, free_chunk_root: true);
4402	btrfs_free_fs_roots(fs_info);
4403
4404	/*
4405	* We must free the block groups after dropping the fs_roots as we could
4406	* have had an IO error and have left over tree log blocks that aren't
4407	* cleaned up until the fs roots are freed. This makes the block group
4408	* accounting appear to be wrong because there's pending reserved bytes,
4409	* so make sure we do the block group cleanup afterwards.
4410	*/
4411	btrfs_free_block_groups(info: fs_info);
4412
4413	iput(fs_info->btree_inode);
4414
4415	btrfs_mapping_tree_free(fs_info);
4416	btrfs_close_devices(fs_devices: fs_info->fs_devices);
4417	}
4418
4419	void btrfs_mark_buffer_dirty(struct btrfs_trans_handle *trans,
4420	struct extent_buffer *buf)
4421	{
4422	struct btrfs_fs_info *fs_info = buf->fs_info;
4423	u64 transid = btrfs_header_generation(eb: buf);
4424
4425	#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4426	/*
4427	* This is a fast path so only do this check if we have sanity tests
4428	* enabled. Normal people shouldn't be using unmapped buffers as dirty
4429	* outside of the sanity tests.
4430	*/
4431	if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4432	return;
4433	#endif
4434	/* This is an active transaction (its state < TRANS_STATE_UNBLOCKED). */
4435	ASSERT(trans->transid == fs_info->generation);
4436	btrfs_assert_tree_write_locked(eb: buf);
4437	if (unlikely(transid != fs_info->generation)) {
4438	btrfs_abort_transaction(trans, -EUCLEAN);
4439	btrfs_crit(fs_info,
4440	"dirty buffer transid mismatch, logical %llu found transid %llu running transid %llu",
4441	buf->start, transid, fs_info->generation);
4442	}
4443	set_extent_buffer_dirty(buf);
4444	}
4445
4446	static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4447	int flush_delayed)
4448	{
4449	/*
4450	* looks as though older kernels can get into trouble with
4451	* this code, they end up stuck in balance_dirty_pages forever
4452	*/
4453	int ret;
4454
4455	if (current->flags & PF_MEMALLOC)
4456	return;
4457
4458	if (flush_delayed)
4459	btrfs_balance_delayed_items(fs_info);
4460
4461	ret = __percpu_counter_compare(fbc: &fs_info->dirty_metadata_bytes,
4462	BTRFS_DIRTY_METADATA_THRESH,
4463	batch: fs_info->dirty_metadata_batch);
4464	if (ret > 0) {
4465	balance_dirty_pages_ratelimited(mapping: fs_info->btree_inode->i_mapping);
4466	}
4467	}
4468
4469	void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4470	{
4471	__btrfs_btree_balance_dirty(fs_info, flush_delayed: 1);
4472	}
4473
4474	void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4475	{
4476	__btrfs_btree_balance_dirty(fs_info, flush_delayed: 0);
4477	}
4478
4479	static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4480	{
4481	/* cleanup FS via transaction */
4482	btrfs_cleanup_transaction(fs_info);
4483
4484	down_write(sem: &fs_info->cleanup_work_sem);
4485	up_write(sem: &fs_info->cleanup_work_sem);
4486	}
4487
4488	static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4489	{
4490	struct btrfs_root *gang[8];
4491	u64 root_objectid = 0;
4492	int ret;
4493
4494	spin_lock(lock: &fs_info->fs_roots_radix_lock);
4495	while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4496	results: (void **)gang, first_index: root_objectid,
4497	ARRAY_SIZE(gang))) != 0) {
4498	int i;
4499
4500	for (i = 0; i < ret; i++)
4501	gang[i] = btrfs_grab_root(root: gang[i]);
4502	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
4503
4504	for (i = 0; i < ret; i++) {
4505	if (!gang[i])
4506	continue;
4507	root_objectid = btrfs_root_id(root: gang[i]);
4508	btrfs_free_log(NULL, root: gang[i]);
4509	btrfs_put_root(root: gang[i]);
4510	}
4511	root_objectid++;
4512	spin_lock(lock: &fs_info->fs_roots_radix_lock);
4513	}
4514	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
4515	btrfs_free_log_root_tree(NULL, fs_info);
4516	}
4517
4518	static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4519	{
4520	struct btrfs_ordered_extent *ordered;
4521
4522	spin_lock(lock: &root->ordered_extent_lock);
4523	/*
4524	* This will just short circuit the ordered completion stuff which will
4525	* make sure the ordered extent gets properly cleaned up.
4526	*/
4527	list_for_each_entry(ordered, &root->ordered_extents,
4528	root_extent_list)
4529	set_bit(nr: BTRFS_ORDERED_IOERR, addr: &ordered->flags);
4530	spin_unlock(lock: &root->ordered_extent_lock);
4531	}
4532
4533	static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4534	{
4535	struct btrfs_root *root;
4536	LIST_HEAD(splice);
4537
4538	spin_lock(lock: &fs_info->ordered_root_lock);
4539	list_splice_init(list: &fs_info->ordered_roots, head: &splice);
4540	while (!list_empty(head: &splice)) {
4541	root = list_first_entry(&splice, struct btrfs_root,
4542	ordered_root);
4543	list_move_tail(list: &root->ordered_root,
4544	head: &fs_info->ordered_roots);
4545
4546	spin_unlock(lock: &fs_info->ordered_root_lock);
4547	btrfs_destroy_ordered_extents(root);
4548
4549	cond_resched();
4550	spin_lock(lock: &fs_info->ordered_root_lock);
4551	}
4552	spin_unlock(lock: &fs_info->ordered_root_lock);
4553
4554	/*
4555	* We need this here because if we've been flipped read-only we won't
4556	* get sync() from the umount, so we need to make sure any ordered
4557	* extents that haven't had their dirty pages IO start writeout yet
4558	* actually get run and error out properly.
4559	*/
4560	btrfs_wait_ordered_roots(fs_info, U64_MAX, NULL);
4561	}
4562
4563	static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4564	{
4565	struct btrfs_inode *btrfs_inode;
4566	LIST_HEAD(splice);
4567
4568	spin_lock(lock: &root->delalloc_lock);
4569	list_splice_init(list: &root->delalloc_inodes, head: &splice);
4570
4571	while (!list_empty(head: &splice)) {
4572	struct inode *inode = NULL;
4573	btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4574	delalloc_inodes);
4575	btrfs_del_delalloc_inode(inode: btrfs_inode);
4576	spin_unlock(lock: &root->delalloc_lock);
4577
4578	/*
4579	* Make sure we get a live inode and that it'll not disappear
4580	* meanwhile.
4581	*/
4582	inode = igrab(&btrfs_inode->vfs_inode);
4583	if (inode) {
4584	unsigned int nofs_flag;
4585
4586	nofs_flag = memalloc_nofs_save();
4587	invalidate_inode_pages2(mapping: inode->i_mapping);
4588	memalloc_nofs_restore(flags: nofs_flag);
4589	iput(inode);
4590	}
4591	spin_lock(lock: &root->delalloc_lock);
4592	}
4593	spin_unlock(lock: &root->delalloc_lock);
4594	}
4595
4596	static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4597	{
4598	struct btrfs_root *root;
4599	LIST_HEAD(splice);
4600
4601	spin_lock(lock: &fs_info->delalloc_root_lock);
4602	list_splice_init(list: &fs_info->delalloc_roots, head: &splice);
4603	while (!list_empty(head: &splice)) {
4604	root = list_first_entry(&splice, struct btrfs_root,
4605	delalloc_root);
4606	root = btrfs_grab_root(root);
4607	BUG_ON(!root);
4608	spin_unlock(lock: &fs_info->delalloc_root_lock);
4609
4610	btrfs_destroy_delalloc_inodes(root);
4611	btrfs_put_root(root);
4612
4613	spin_lock(lock: &fs_info->delalloc_root_lock);
4614	}
4615	spin_unlock(lock: &fs_info->delalloc_root_lock);
4616	}
4617
4618	static void btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4619	struct extent_io_tree *dirty_pages,
4620	int mark)
4621	{
4622	struct extent_buffer *eb;
4623	u64 start = 0;
4624	u64 end;
4625
4626	while (btrfs_find_first_extent_bit(tree: dirty_pages, start, start_ret: &start, end_ret: &end,
4627	bits: mark, NULL)) {
4628	btrfs_clear_extent_bits(tree: dirty_pages, start, end, bits: mark);
4629	while (start <= end) {
4630	eb = find_extent_buffer(fs_info, start);
4631	start += fs_info->nodesize;
4632	if (!eb)
4633	continue;
4634
4635	btrfs_tree_lock(eb);
4636	wait_on_extent_buffer_writeback(eb);
4637	btrfs_clear_buffer_dirty(NULL, buf: eb);
4638	btrfs_tree_unlock(eb);
4639
4640	free_extent_buffer_stale(eb);
4641	}
4642	}
4643	}
4644
4645	static void btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4646	struct extent_io_tree *unpin)
4647	{
4648	u64 start;
4649	u64 end;
4650
4651	while (1) {
4652	struct extent_state *cached_state = NULL;
4653
4654	/*
4655	* The btrfs_finish_extent_commit() may get the same range as
4656	* ours between find_first_extent_bit and clear_extent_dirty.
4657	* Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4658	* the same extent range.
4659	*/
4660	mutex_lock(&fs_info->unused_bg_unpin_mutex);
4661	if (!btrfs_find_first_extent_bit(tree: unpin, start: 0, start_ret: &start, end_ret: &end,
4662	bits: EXTENT_DIRTY, cached_state: &cached_state)) {
4663	mutex_unlock(lock: &fs_info->unused_bg_unpin_mutex);
4664	break;
4665	}
4666
4667	btrfs_clear_extent_dirty(tree: unpin, start, end, cached: &cached_state);
4668	btrfs_free_extent_state(state: cached_state);
4669	btrfs_error_unpin_extent_range(fs_info, start, end);
4670	mutex_unlock(lock: &fs_info->unused_bg_unpin_mutex);
4671	cond_resched();
4672	}
4673	}
4674
4675	static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4676	{
4677	struct inode *inode;
4678
4679	inode = cache->io_ctl.inode;
4680	if (inode) {
4681	unsigned int nofs_flag;
4682
4683	nofs_flag = memalloc_nofs_save();
4684	invalidate_inode_pages2(mapping: inode->i_mapping);
4685	memalloc_nofs_restore(flags: nofs_flag);
4686
4687	BTRFS_I(inode)->generation = 0;
4688	cache->io_ctl.inode = NULL;
4689	iput(inode);
4690	}
4691	ASSERT(cache->io_ctl.pages == NULL);
4692	btrfs_put_block_group(cache);
4693	}
4694
4695	void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4696	struct btrfs_fs_info *fs_info)
4697	{
4698	struct btrfs_block_group *cache;
4699
4700	spin_lock(lock: &cur_trans->dirty_bgs_lock);
4701	while (!list_empty(head: &cur_trans->dirty_bgs)) {
4702	cache = list_first_entry(&cur_trans->dirty_bgs,
4703	struct btrfs_block_group,
4704	dirty_list);
4705
4706	if (!list_empty(head: &cache->io_list)) {
4707	spin_unlock(lock: &cur_trans->dirty_bgs_lock);
4708	list_del_init(entry: &cache->io_list);
4709	btrfs_cleanup_bg_io(cache);
4710	spin_lock(lock: &cur_trans->dirty_bgs_lock);
4711	}
4712
4713	list_del_init(entry: &cache->dirty_list);
4714	spin_lock(lock: &cache->lock);
4715	cache->disk_cache_state = BTRFS_DC_ERROR;
4716	spin_unlock(lock: &cache->lock);
4717
4718	spin_unlock(lock: &cur_trans->dirty_bgs_lock);
4719	btrfs_put_block_group(cache);
4720	btrfs_dec_delayed_refs_rsv_bg_updates(fs_info);
4721	spin_lock(lock: &cur_trans->dirty_bgs_lock);
4722	}
4723	spin_unlock(lock: &cur_trans->dirty_bgs_lock);
4724
4725	/*
4726	* Refer to the definition of io_bgs member for details why it's safe
4727	* to use it without any locking
4728	*/
4729	while (!list_empty(head: &cur_trans->io_bgs)) {
4730	cache = list_first_entry(&cur_trans->io_bgs,
4731	struct btrfs_block_group,
4732	io_list);
4733
4734	list_del_init(entry: &cache->io_list);
4735	spin_lock(lock: &cache->lock);
4736	cache->disk_cache_state = BTRFS_DC_ERROR;
4737	spin_unlock(lock: &cache->lock);
4738	btrfs_cleanup_bg_io(cache);
4739	}
4740	}
4741
4742	static void btrfs_free_all_qgroup_pertrans(struct btrfs_fs_info *fs_info)
4743	{
4744	struct btrfs_root *gang[8];
4745	int i;
4746	int ret;
4747
4748	spin_lock(lock: &fs_info->fs_roots_radix_lock);
4749	while (1) {
4750	ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
4751	results: (void **)gang, first_index: 0,
4752	ARRAY_SIZE(gang),
4753	BTRFS_ROOT_TRANS_TAG);
4754	if (ret == 0)
4755	break;
4756	for (i = 0; i < ret; i++) {
4757	struct btrfs_root *root = gang[i];
4758
4759	btrfs_qgroup_free_meta_all_pertrans(root);
4760	radix_tree_tag_clear(&fs_info->fs_roots_radix,
4761	index: (unsigned long)btrfs_root_id(root),
4762	BTRFS_ROOT_TRANS_TAG);
4763	}
4764	}
4765	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
4766	}
4767
4768	void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans)
4769	{
4770	struct btrfs_fs_info *fs_info = cur_trans->fs_info;
4771	struct btrfs_device *dev, *tmp;
4772
4773	btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4774	ASSERT(list_empty(&cur_trans->dirty_bgs));
4775	ASSERT(list_empty(&cur_trans->io_bgs));
4776
4777	list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4778	post_commit_list) {
4779	list_del_init(entry: &dev->post_commit_list);
4780	}
4781
4782	btrfs_destroy_delayed_refs(trans: cur_trans);
4783
4784	cur_trans->state = TRANS_STATE_COMMIT_START;
4785	wake_up(&fs_info->transaction_blocked_wait);
4786
4787	cur_trans->state = TRANS_STATE_UNBLOCKED;
4788	wake_up(&fs_info->transaction_wait);
4789
4790	btrfs_destroy_marked_extents(fs_info, dirty_pages: &cur_trans->dirty_pages,
4791	mark: EXTENT_DIRTY);
4792	btrfs_destroy_pinned_extent(fs_info, unpin: &cur_trans->pinned_extents);
4793
4794	cur_trans->state =TRANS_STATE_COMPLETED;
4795	wake_up(&cur_trans->commit_wait);
4796	}
4797
4798	static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4799	{
4800	struct btrfs_transaction *t;
4801
4802	mutex_lock(&fs_info->transaction_kthread_mutex);
4803
4804	spin_lock(lock: &fs_info->trans_lock);
4805	while (!list_empty(head: &fs_info->trans_list)) {
4806	t = list_first_entry(&fs_info->trans_list,
4807	struct btrfs_transaction, list);
4808	if (t->state >= TRANS_STATE_COMMIT_PREP) {
4809	refcount_inc(r: &t->use_count);
4810	spin_unlock(lock: &fs_info->trans_lock);
4811	btrfs_wait_for_commit(fs_info, transid: t->transid);
4812	btrfs_put_transaction(transaction: t);
4813	spin_lock(lock: &fs_info->trans_lock);
4814	continue;
4815	}
4816	if (t == fs_info->running_transaction) {
4817	t->state = TRANS_STATE_COMMIT_DOING;
4818	spin_unlock(lock: &fs_info->trans_lock);
4819	/*
4820	* We wait for 0 num_writers since we don't hold a trans
4821	* handle open currently for this transaction.
4822	*/
4823	wait_event(t->writer_wait,
4824	atomic_read(&t->num_writers) == 0);
4825	} else {
4826	spin_unlock(lock: &fs_info->trans_lock);
4827	}
4828	btrfs_cleanup_one_transaction(cur_trans: t);
4829
4830	spin_lock(lock: &fs_info->trans_lock);
4831	if (t == fs_info->running_transaction)
4832	fs_info->running_transaction = NULL;
4833	list_del_init(entry: &t->list);
4834	spin_unlock(lock: &fs_info->trans_lock);
4835
4836	btrfs_put_transaction(transaction: t);
4837	trace_btrfs_transaction_commit(fs_info);
4838	spin_lock(lock: &fs_info->trans_lock);
4839	}
4840	spin_unlock(lock: &fs_info->trans_lock);
4841	btrfs_destroy_all_ordered_extents(fs_info);
4842	btrfs_destroy_delayed_inodes(fs_info);
4843	btrfs_assert_delayed_root_empty(fs_info);
4844	btrfs_destroy_all_delalloc_inodes(fs_info);
4845	btrfs_drop_all_logs(fs_info);
4846	btrfs_free_all_qgroup_pertrans(fs_info);
4847	mutex_unlock(lock: &fs_info->transaction_kthread_mutex);
4848
4849	return 0;
4850	}
4851
4852	int btrfs_init_root_free_objectid(struct btrfs_root *root)
4853	{
4854	BTRFS_PATH_AUTO_FREE(path);
4855	int ret;
4856	struct extent_buffer *l;
4857	struct btrfs_key search_key;
4858	struct btrfs_key found_key;
4859	int slot;
4860
4861	path = btrfs_alloc_path();
4862	if (!path)
4863	return -ENOMEM;
4864
4865	search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4866	search_key.type = -1;
4867	search_key.offset = (u64)-1;
4868	ret = btrfs_search_slot(NULL, root, key: &search_key, p: path, ins_len: 0, cow: 0);
4869	if (ret < 0)
4870	return ret;
4871	if (ret == 0) {
4872	/*
4873	* Key with offset -1 found, there would have to exist a root
4874	* with such id, but this is out of valid range.
4875	*/
4876	return -EUCLEAN;
4877	}
4878	if (path->slots[0] > 0) {
4879	slot = path->slots[0] - 1;
4880	l = path->nodes[0];
4881	btrfs_item_key_to_cpu(eb: l, cpu_key: &found_key, nr: slot);
4882	root->free_objectid = max_t(u64, found_key.objectid + 1,
4883	BTRFS_FIRST_FREE_OBJECTID);
4884	} else {
4885	root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4886	}
4887
4888	return 0;
4889	}
4890
4891	int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4892	{
4893	int ret;
4894	mutex_lock(&root->objectid_mutex);
4895
4896	if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4897	btrfs_warn(root->fs_info,
4898	"the objectid of root %llu reaches its highest value",
4899	btrfs_root_id(root));
4900	ret = -ENOSPC;
4901	goto out;
4902	}
4903
4904	*objectid = root->free_objectid++;
4905	ret = 0;
4906	out:
4907	mutex_unlock(lock: &root->objectid_mutex);
4908	return ret;
4909	}
4910