1	// SPDX-License-Identifier: GPL-2.0
2	/*
3	* Copyright (C) 2007 Oracle. All rights reserved.
4	*/
5
6	#include <linux/fs.h>
7	#include <linux/blkdev.h>
8	#include <linux/radix-tree.h>
9	#include <linux/writeback.h>
10	#include <linux/workqueue.h>
11	#include <linux/kthread.h>
12	#include <linux/slab.h>
13	#include <linux/migrate.h>
14	#include <linux/ratelimit.h>
15	#include <linux/uuid.h>
16	#include <linux/semaphore.h>
17	#include <linux/error-injection.h>
18	#include <linux/crc32c.h>
19	#include <linux/sched/mm.h>
20	#include <asm/unaligned.h>
21	#include <crypto/hash.h>
22	#include "ctree.h"
23	#include "disk-io.h"
24	#include "transaction.h"
25	#include "btrfs_inode.h"
26	#include "bio.h"
27	#include "print-tree.h"
28	#include "locking.h"
29	#include "tree-log.h"
30	#include "free-space-cache.h"
31	#include "free-space-tree.h"
32	#include "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 num_folios = num_extent_folios(eb);
186	int ret = 0;
187
188	if (sb_rdonly(sb: fs_info->sb))
189	return -EROFS;
190
191	for (int i = 0; i < num_folios; i++) {
192	struct folio *folio = eb->folios[i];
193	u64 start = max_t(u64, eb->start, folio_pos(folio));
194	u64 end = min_t(u64, eb->start + eb->len,
195	folio_pos(folio) + eb->folio_size);
196	u32 len = end - start;
197
198	ret = btrfs_repair_io_failure(fs_info, ino: 0, start, length: len,
199	logical: start, folio, offset_in_folio(folio, start),
200	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	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	clear_bit(nr: EXTENT_BUFFER_CORRUPT, addr: &eb->bflags);
229	ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, parent_check: check);
230	if (!ret)
231	break;
232
233	num_copies = btrfs_num_copies(fs_info,
234	logical: eb->start, len: eb->len);
235	if (num_copies == 1)
236	break;
237
238	if (!failed_mirror) {
239	failed = 1;
240	failed_mirror = eb->read_mirror;
241	}
242
243	mirror_num++;
244	if (mirror_num == failed_mirror)
245	mirror_num++;
246
247	if (mirror_num > num_copies)
248	break;
249	}
250
251	if (failed && !ret && failed_mirror)
252	btrfs_repair_eb_io_failure(eb, mirror_num: failed_mirror);
253
254	return ret;
255	}
256
257	/*
258	* Checksum a dirty tree block before IO.
259	*/
260	blk_status_t btree_csum_one_bio(struct btrfs_bio *bbio)
261	{
262	struct extent_buffer *eb = bbio->private;
263	struct btrfs_fs_info *fs_info = eb->fs_info;
264	u64 found_start = btrfs_header_bytenr(eb);
265	u64 last_trans;
266	u8 result[BTRFS_CSUM_SIZE];
267	int ret;
268
269	/* Btree blocks are always contiguous on disk. */
270	if (WARN_ON_ONCE(bbio->file_offset != eb->start))
271	return BLK_STS_IOERR;
272	if (WARN_ON_ONCE(bbio->bio.bi_iter.bi_size != eb->len))
273	return BLK_STS_IOERR;
274
275	/*
276	* If an extent_buffer is marked as EXTENT_BUFFER_ZONED_ZEROOUT, don't
277	* checksum it but zero-out its content. This is done to preserve
278	* ordering of I/O without unnecessarily writing out data.
279	*/
280	if (test_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags)) {
281	memzero_extent_buffer(eb, start: 0, len: eb->len);
282	return BLK_STS_OK;
283	}
284
285	if (WARN_ON_ONCE(found_start != eb->start))
286	return BLK_STS_IOERR;
287	if (WARN_ON(!btrfs_folio_test_uptodate(fs_info, eb->folios[0],
288	eb->start, eb->len)))
289	return BLK_STS_IOERR;
290
291	ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
292	offsetof(struct btrfs_header, fsid),
293	BTRFS_FSID_SIZE) == 0);
294	csum_tree_block(buf: eb, result);
295
296	if (btrfs_header_level(eb))
297	ret = btrfs_check_node(node: eb);
298	else
299	ret = btrfs_check_leaf(leaf: eb);
300
301	if (ret < 0)
302	goto error;
303
304	/*
305	* Also check the generation, the eb reached here must be newer than
306	* last committed. Or something seriously wrong happened.
307	*/
308	last_trans = btrfs_get_last_trans_committed(fs_info);
309	if (unlikely(btrfs_header_generation(eb) <= last_trans)) {
310	ret = -EUCLEAN;
311	btrfs_err(fs_info,
312	"block=%llu bad generation, have %llu expect > %llu",
313	eb->start, btrfs_header_generation(eb), last_trans);
314	goto error;
315	}
316	write_extent_buffer(eb, src: result, start: 0, len: fs_info->csum_size);
317	return BLK_STS_OK;
318
319	error:
320	btrfs_print_tree(c: eb, follow: 0);
321	btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
322	eb->start);
323	/*
324	* Be noisy if this is an extent buffer from a log tree. We don't abort
325	* a transaction in case there's a bad log tree extent buffer, we just
326	* fallback to a transaction commit. Still we want to know when there is
327	* a bad log tree extent buffer, as that may signal a bug somewhere.
328	*/
329	WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
330	btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
331	return errno_to_blk_status(errno: ret);
332	}
333
334	static bool check_tree_block_fsid(struct extent_buffer *eb)
335	{
336	struct btrfs_fs_info *fs_info = eb->fs_info;
337	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
338	u8 fsid[BTRFS_FSID_SIZE];
339
340	read_extent_buffer(eb, dst: fsid, offsetof(struct btrfs_header, fsid),
341	BTRFS_FSID_SIZE);
342
343	/*
344	* alloc_fsid_devices() copies the fsid into fs_devices::metadata_uuid.
345	* This is then overwritten by metadata_uuid if it is present in the
346	* device_list_add(). The same true for a seed device as well. So use of
347	* fs_devices::metadata_uuid is appropriate here.
348	*/
349	if (memcmp(p: fsid, q: fs_info->fs_devices->metadata_uuid, BTRFS_FSID_SIZE) == 0)
350	return false;
351
352	list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
353	if (!memcmp(p: fsid, q: seed_devs->fsid, BTRFS_FSID_SIZE))
354	return false;
355
356	return true;
357	}
358
359	/* Do basic extent buffer checks at read time */
360	int btrfs_validate_extent_buffer(struct extent_buffer *eb,
361	struct btrfs_tree_parent_check *check)
362	{
363	struct btrfs_fs_info *fs_info = eb->fs_info;
364	u64 found_start;
365	const u32 csum_size = fs_info->csum_size;
366	u8 found_level;
367	u8 result[BTRFS_CSUM_SIZE];
368	const u8 *header_csum;
369	int ret = 0;
370
371	ASSERT(check);
372
373	found_start = btrfs_header_bytenr(eb);
374	if (found_start != eb->start) {
375	btrfs_err_rl(fs_info,
376	"bad tree block start, mirror %u want %llu have %llu",
377	eb->read_mirror, eb->start, found_start);
378	ret = -EIO;
379	goto out;
380	}
381	if (check_tree_block_fsid(eb)) {
382	btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
383	eb->start, eb->read_mirror);
384	ret = -EIO;
385	goto out;
386	}
387	found_level = btrfs_header_level(eb);
388	if (found_level >= BTRFS_MAX_LEVEL) {
389	btrfs_err(fs_info,
390	"bad tree block level, mirror %u level %d on logical %llu",
391	eb->read_mirror, btrfs_header_level(eb), eb->start);
392	ret = -EIO;
393	goto out;
394	}
395
396	csum_tree_block(buf: eb, result);
397	header_csum = folio_address(folio: eb->folios[0]) +
398	get_eb_offset_in_folio(eb, offsetof(struct btrfs_header, csum));
399
400	if (memcmp(p: result, q: header_csum, size: csum_size) != 0) {
401	btrfs_warn_rl(fs_info,
402	"checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d",
403	eb->start, eb->read_mirror,
404	CSUM_FMT_VALUE(csum_size, header_csum),
405	CSUM_FMT_VALUE(csum_size, result),
406	btrfs_header_level(eb));
407	ret = -EUCLEAN;
408	goto out;
409	}
410
411	if (found_level != check->level) {
412	btrfs_err(fs_info,
413	"level verify failed on logical %llu mirror %u wanted %u found %u",
414	eb->start, eb->read_mirror, check->level, found_level);
415	ret = -EIO;
416	goto out;
417	}
418	if (unlikely(check->transid &&
419	btrfs_header_generation(eb) != check->transid)) {
420	btrfs_err_rl(eb->fs_info,
421	"parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
422	eb->start, eb->read_mirror, check->transid,
423	btrfs_header_generation(eb));
424	ret = -EIO;
425	goto out;
426	}
427	if (check->has_first_key) {
428	struct btrfs_key *expect_key = &check->first_key;
429	struct btrfs_key found_key;
430
431	if (found_level)
432	btrfs_node_key_to_cpu(eb, cpu_key: &found_key, nr: 0);
433	else
434	btrfs_item_key_to_cpu(eb, cpu_key: &found_key, nr: 0);
435	if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
436	btrfs_err(fs_info,
437	"tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
438	eb->start, check->transid,
439	expect_key->objectid,
440	expect_key->type, expect_key->offset,
441	found_key.objectid, found_key.type,
442	found_key.offset);
443	ret = -EUCLEAN;
444	goto out;
445	}
446	}
447	if (check->owner_root) {
448	ret = btrfs_check_eb_owner(eb, root_owner: check->owner_root);
449	if (ret < 0)
450	goto out;
451	}
452
453	/*
454	* If this is a leaf block and it is corrupt, set the corrupt bit so
455	* that we don't try and read the other copies of this block, just
456	* return -EIO.
457	*/
458	if (found_level == 0 && btrfs_check_leaf(leaf: eb)) {
459	set_bit(nr: EXTENT_BUFFER_CORRUPT, addr: &eb->bflags);
460	ret = -EIO;
461	}
462
463	if (found_level > 0 && btrfs_check_node(node: eb))
464	ret = -EIO;
465
466	if (ret)
467	btrfs_err(fs_info,
468	"read time tree block corruption detected on logical %llu mirror %u",
469	eb->start, eb->read_mirror);
470	out:
471	return ret;
472	}
473
474	#ifdef CONFIG_MIGRATION
475	static int btree_migrate_folio(struct address_space *mapping,
476	struct folio *dst, struct folio *src, enum migrate_mode mode)
477	{
478	/*
479	* we can't safely write a btree page from here,
480	* we haven't done the locking hook
481	*/
482	if (folio_test_dirty(folio: src))
483	return -EAGAIN;
484	/*
485	* Buffers may be managed in a filesystem specific way.
486	* We must have no buffers or drop them.
487	*/
488	if (folio_get_private(folio: src) &&
489	!filemap_release_folio(folio: src, GFP_KERNEL))
490	return -EAGAIN;
491	return migrate_folio(mapping, dst, src, mode);
492	}
493	#else
494	#define btree_migrate_folio NULL
495	#endif
496
497	static int btree_writepages(struct address_space *mapping,
498	struct writeback_control *wbc)
499	{
500	int ret;
501
502	if (wbc->sync_mode == WB_SYNC_NONE) {
503	struct btrfs_fs_info *fs_info;
504
505	if (wbc->for_kupdate)
506	return 0;
507
508	fs_info = inode_to_fs_info(mapping->host);
509	/* this is a bit racy, but that's ok */
510	ret = __percpu_counter_compare(fbc: &fs_info->dirty_metadata_bytes,
511	BTRFS_DIRTY_METADATA_THRESH,
512	batch: fs_info->dirty_metadata_batch);
513	if (ret < 0)
514	return 0;
515	}
516	return btree_write_cache_pages(mapping, wbc);
517	}
518
519	static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
520	{
521	if (folio_test_writeback(folio) || folio_test_dirty(folio))
522	return false;
523
524	return try_release_extent_buffer(page: &folio->page);
525	}
526
527	static void btree_invalidate_folio(struct folio *folio, size_t offset,
528	size_t length)
529	{
530	struct extent_io_tree *tree;
531
532	tree = &folio_to_inode(folio)->io_tree;
533	extent_invalidate_folio(tree, folio, offset);
534	btree_release_folio(folio, GFP_NOFS);
535	if (folio_get_private(folio)) {
536	btrfs_warn(folio_to_fs_info(folio),
537	"folio private not zero on folio %llu",
538	(unsigned long long)folio_pos(folio));
539	folio_detach_private(folio);
540	}
541	}
542
543	#ifdef DEBUG
544	static bool btree_dirty_folio(struct address_space *mapping,
545	struct folio *folio)
546	{
547	struct btrfs_fs_info *fs_info = inode_to_fs_info(mapping->host);
548	struct btrfs_subpage_info *spi = fs_info->subpage_info;
549	struct btrfs_subpage *subpage;
550	struct extent_buffer *eb;
551	int cur_bit = 0;
552	u64 page_start = folio_pos(folio);
553
554	if (fs_info->sectorsize == PAGE_SIZE) {
555	eb = folio_get_private(folio);
556	BUG_ON(!eb);
557	BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
558	BUG_ON(!atomic_read(&eb->refs));
559	btrfs_assert_tree_write_locked(eb);
560	return filemap_dirty_folio(mapping, folio);
561	}
562
563	ASSERT(spi);
564	subpage = folio_get_private(folio);
565
566	for (cur_bit = spi->dirty_offset;
567	cur_bit < spi->dirty_offset + spi->bitmap_nr_bits;
568	cur_bit++) {
569	unsigned long flags;
570	u64 cur;
571
572	spin_lock_irqsave(&subpage->lock, flags);
573	if (!test_bit(cur_bit, subpage->bitmaps)) {
574	spin_unlock_irqrestore(&subpage->lock, flags);
575	continue;
576	}
577	spin_unlock_irqrestore(&subpage->lock, flags);
578	cur = page_start + cur_bit * fs_info->sectorsize;
579
580	eb = find_extent_buffer(fs_info, cur);
581	ASSERT(eb);
582	ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
583	ASSERT(atomic_read(&eb->refs));
584	btrfs_assert_tree_write_locked(eb);
585	free_extent_buffer(eb);
586
587	cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits) - 1;
588	}
589	return filemap_dirty_folio(mapping, folio);
590	}
591	#else
592	#define btree_dirty_folio filemap_dirty_folio
593	#endif
594
595	static const struct address_space_operations btree_aops = {
596	.writepages = btree_writepages,
597	.release_folio = btree_release_folio,
598	.invalidate_folio = btree_invalidate_folio,
599	.migrate_folio = btree_migrate_folio,
600	.dirty_folio = btree_dirty_folio,
601	};
602
603	struct extent_buffer *btrfs_find_create_tree_block(
604	struct btrfs_fs_info *fs_info,
605	u64 bytenr, u64 owner_root,
606	int level)
607	{
608	if (btrfs_is_testing(fs_info))
609	return alloc_test_extent_buffer(fs_info, start: bytenr);
610	return alloc_extent_buffer(fs_info, start: bytenr, owner_root, level);
611	}
612
613	/*
614	* Read tree block at logical address @bytenr and do variant basic but critical
615	* verification.
616	*
617	* @check: expected tree parentness check, see comments of the
618	* structure for details.
619	*/
620	struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
621	struct btrfs_tree_parent_check *check)
622	{
623	struct extent_buffer *buf = NULL;
624	int ret;
625
626	ASSERT(check);
627
628	buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root: check->owner_root,
629	level: check->level);
630	if (IS_ERR(ptr: buf))
631	return buf;
632
633	ret = btrfs_read_extent_buffer(eb: buf, check);
634	if (ret) {
635	free_extent_buffer_stale(eb: buf);
636	return ERR_PTR(error: ret);
637	}
638	if (btrfs_check_eb_owner(eb: buf, root_owner: check->owner_root)) {
639	free_extent_buffer_stale(eb: buf);
640	return ERR_PTR(error: -EUCLEAN);
641	}
642	return buf;
643
644	}
645
646	static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
647	u64 objectid)
648	{
649	bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
650
651	memset(&root->root_key, 0, sizeof(root->root_key));
652	memset(&root->root_item, 0, sizeof(root->root_item));
653	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
654	root->fs_info = fs_info;
655	root->root_key.objectid = objectid;
656	root->node = NULL;
657	root->commit_root = NULL;
658	root->state = 0;
659	RB_CLEAR_NODE(&root->rb_node);
660
661	root->last_trans = 0;
662	root->free_objectid = 0;
663	root->nr_delalloc_inodes = 0;
664	root->nr_ordered_extents = 0;
665	root->inode_tree = RB_ROOT;
666	/* GFP flags are compatible with XA_FLAGS_*. */
667	xa_init_flags(xa: &root->delayed_nodes, GFP_ATOMIC);
668
669	btrfs_init_root_block_rsv(root);
670
671	INIT_LIST_HEAD(list: &root->dirty_list);
672	INIT_LIST_HEAD(list: &root->root_list);
673	INIT_LIST_HEAD(list: &root->delalloc_inodes);
674	INIT_LIST_HEAD(list: &root->delalloc_root);
675	INIT_LIST_HEAD(list: &root->ordered_extents);
676	INIT_LIST_HEAD(list: &root->ordered_root);
677	INIT_LIST_HEAD(list: &root->reloc_dirty_list);
678	spin_lock_init(&root->inode_lock);
679	spin_lock_init(&root->delalloc_lock);
680	spin_lock_init(&root->ordered_extent_lock);
681	spin_lock_init(&root->accounting_lock);
682	spin_lock_init(&root->qgroup_meta_rsv_lock);
683	mutex_init(&root->objectid_mutex);
684	mutex_init(&root->log_mutex);
685	mutex_init(&root->ordered_extent_mutex);
686	mutex_init(&root->delalloc_mutex);
687	init_waitqueue_head(&root->qgroup_flush_wait);
688	init_waitqueue_head(&root->log_writer_wait);
689	init_waitqueue_head(&root->log_commit_wait[0]);
690	init_waitqueue_head(&root->log_commit_wait[1]);
691	INIT_LIST_HEAD(list: &root->log_ctxs[0]);
692	INIT_LIST_HEAD(list: &root->log_ctxs[1]);
693	atomic_set(v: &root->log_commit[0], i: 0);
694	atomic_set(v: &root->log_commit[1], i: 0);
695	atomic_set(v: &root->log_writers, i: 0);
696	atomic_set(v: &root->log_batch, i: 0);
697	refcount_set(r: &root->refs, n: 1);
698	atomic_set(v: &root->snapshot_force_cow, i: 0);
699	atomic_set(v: &root->nr_swapfiles, i: 0);
700	btrfs_set_root_log_transid(root, log_transid: 0);
701	root->log_transid_committed = -1;
702	btrfs_set_root_last_log_commit(root, commit_id: 0);
703	root->anon_dev = 0;
704	if (!dummy) {
705	extent_io_tree_init(fs_info, tree: &root->dirty_log_pages,
706	owner: IO_TREE_ROOT_DIRTY_LOG_PAGES);
707	extent_io_tree_init(fs_info, tree: &root->log_csum_range,
708	owner: IO_TREE_LOG_CSUM_RANGE);
709	}
710
711	spin_lock_init(&root->root_item_lock);
712	btrfs_qgroup_init_swapped_blocks(swapped_blocks: &root->swapped_blocks);
713	#ifdef CONFIG_BTRFS_DEBUG
714	INIT_LIST_HEAD(list: &root->leak_list);
715	spin_lock(lock: &fs_info->fs_roots_radix_lock);
716	list_add_tail(new: &root->leak_list, head: &fs_info->allocated_roots);
717	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
718	#endif
719	}
720
721	static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
722	u64 objectid, gfp_t flags)
723	{
724	struct btrfs_root *root = kzalloc(size: sizeof(*root), flags);
725	if (root)
726	__setup_root(root, fs_info, objectid);
727	return root;
728	}
729
730	#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
731	/* Should only be used by the testing infrastructure */
732	struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
733	{
734	struct btrfs_root *root;
735
736	if (!fs_info)
737	return ERR_PTR(error: -EINVAL);
738
739	root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
740	if (!root)
741	return ERR_PTR(error: -ENOMEM);
742
743	/* We don't use the stripesize in selftest, set it as sectorsize */
744	root->alloc_bytenr = 0;
745
746	return root;
747	}
748	#endif
749
750	static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
751	{
752	const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
753	const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
754
755	return btrfs_comp_cpu_keys(k1: &a->root_key, k2: &b->root_key);
756	}
757
758	static int global_root_key_cmp(const void *k, const struct rb_node *node)
759	{
760	const struct btrfs_key *key = k;
761	const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
762
763	return btrfs_comp_cpu_keys(k1: key, k2: &root->root_key);
764	}
765
766	int btrfs_global_root_insert(struct btrfs_root *root)
767	{
768	struct btrfs_fs_info *fs_info = root->fs_info;
769	struct rb_node *tmp;
770	int ret = 0;
771
772	write_lock(&fs_info->global_root_lock);
773	tmp = rb_find_add(node: &root->rb_node, tree: &fs_info->global_root_tree, cmp: global_root_cmp);
774	write_unlock(&fs_info->global_root_lock);
775
776	if (tmp) {
777	ret = -EEXIST;
778	btrfs_warn(fs_info, "global root %llu %llu already exists",
779	root->root_key.objectid, root->root_key.offset);
780	}
781	return ret;
782	}
783
784	void btrfs_global_root_delete(struct btrfs_root *root)
785	{
786	struct btrfs_fs_info *fs_info = root->fs_info;
787
788	write_lock(&fs_info->global_root_lock);
789	rb_erase(&root->rb_node, &fs_info->global_root_tree);
790	write_unlock(&fs_info->global_root_lock);
791	}
792
793	struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
794	struct btrfs_key *key)
795	{
796	struct rb_node *node;
797	struct btrfs_root *root = NULL;
798
799	read_lock(&fs_info->global_root_lock);
800	node = rb_find(key, tree: &fs_info->global_root_tree, cmp: global_root_key_cmp);
801	if (node)
802	root = container_of(node, struct btrfs_root, rb_node);
803	read_unlock(&fs_info->global_root_lock);
804
805	return root;
806	}
807
808	static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
809	{
810	struct btrfs_block_group *block_group;
811	u64 ret;
812
813	if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
814	return 0;
815
816	if (bytenr)
817	block_group = btrfs_lookup_block_group(info: fs_info, bytenr);
818	else
819	block_group = btrfs_lookup_first_block_group(info: fs_info, bytenr);
820	ASSERT(block_group);
821	if (!block_group)
822	return 0;
823	ret = block_group->global_root_id;
824	btrfs_put_block_group(cache: block_group);
825
826	return ret;
827	}
828
829	struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
830	{
831	struct btrfs_key key = {
832	.objectid = BTRFS_CSUM_TREE_OBJECTID,
833	.type = BTRFS_ROOT_ITEM_KEY,
834	.offset = btrfs_global_root_id(fs_info, bytenr),
835	};
836
837	return btrfs_global_root(fs_info, key: &key);
838	}
839
840	struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
841	{
842	struct btrfs_key key = {
843	.objectid = BTRFS_EXTENT_TREE_OBJECTID,
844	.type = BTRFS_ROOT_ITEM_KEY,
845	.offset = btrfs_global_root_id(fs_info, bytenr),
846	};
847
848	return btrfs_global_root(fs_info, key: &key);
849	}
850
851	struct btrfs_root *btrfs_block_group_root(struct btrfs_fs_info *fs_info)
852	{
853	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE))
854	return fs_info->block_group_root;
855	return btrfs_extent_root(fs_info, bytenr: 0);
856	}
857
858	struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
859	u64 objectid)
860	{
861	struct btrfs_fs_info *fs_info = trans->fs_info;
862	struct extent_buffer *leaf;
863	struct btrfs_root *tree_root = fs_info->tree_root;
864	struct btrfs_root *root;
865	struct btrfs_key key;
866	unsigned int nofs_flag;
867	int ret = 0;
868
869	/*
870	* We're holding a transaction handle, so use a NOFS memory allocation
871	* context to avoid deadlock if reclaim happens.
872	*/
873	nofs_flag = memalloc_nofs_save();
874	root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
875	memalloc_nofs_restore(flags: nofs_flag);
876	if (!root)
877	return ERR_PTR(error: -ENOMEM);
878
879	root->root_key.objectid = objectid;
880	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
881	root->root_key.offset = 0;
882
883	leaf = btrfs_alloc_tree_block(trans, root, parent: 0, root_objectid: objectid, NULL, level: 0, hint: 0, empty_size: 0,
884	reloc_src_root: 0, nest: BTRFS_NESTING_NORMAL);
885	if (IS_ERR(ptr: leaf)) {
886	ret = PTR_ERR(ptr: leaf);
887	leaf = NULL;
888	goto fail;
889	}
890
891	root->node = leaf;
892	btrfs_mark_buffer_dirty(trans, buf: leaf);
893
894	root->commit_root = btrfs_root_node(root);
895	set_bit(nr: BTRFS_ROOT_TRACK_DIRTY, addr: &root->state);
896
897	btrfs_set_root_flags(s: &root->root_item, val: 0);
898	btrfs_set_root_limit(s: &root->root_item, val: 0);
899	btrfs_set_root_bytenr(s: &root->root_item, val: leaf->start);
900	btrfs_set_root_generation(s: &root->root_item, val: trans->transid);
901	btrfs_set_root_level(s: &root->root_item, val: 0);
902	btrfs_set_root_refs(s: &root->root_item, val: 1);
903	btrfs_set_root_used(s: &root->root_item, val: leaf->len);
904	btrfs_set_root_last_snapshot(s: &root->root_item, val: 0);
905	btrfs_set_root_dirid(s: &root->root_item, val: 0);
906	if (is_fstree(rootid: objectid))
907	generate_random_guid(guid: root->root_item.uuid);
908	else
909	export_guid(dst: root->root_item.uuid, src: &guid_null);
910	btrfs_set_root_drop_level(s: &root->root_item, val: 0);
911
912	btrfs_tree_unlock(eb: leaf);
913
914	key.objectid = objectid;
915	key.type = BTRFS_ROOT_ITEM_KEY;
916	key.offset = 0;
917	ret = btrfs_insert_root(trans, root: tree_root, key: &key, item: &root->root_item);
918	if (ret)
919	goto fail;
920
921	return root;
922
923	fail:
924	btrfs_put_root(root);
925
926	return ERR_PTR(error: ret);
927	}
928
929	static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
930	struct btrfs_fs_info *fs_info)
931	{
932	struct btrfs_root *root;
933
934	root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
935	if (!root)
936	return ERR_PTR(error: -ENOMEM);
937
938	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
939	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
940	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
941
942	return root;
943	}
944
945	int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
946	struct btrfs_root *root)
947	{
948	struct extent_buffer *leaf;
949
950	/*
951	* DON'T set SHAREABLE bit for log trees.
952	*
953	* Log trees are not exposed to user space thus can't be snapshotted,
954	* and they go away before a real commit is actually done.
955	*
956	* They do store pointers to file data extents, and those reference
957	* counts still get updated (along with back refs to the log tree).
958	*/
959
960	leaf = btrfs_alloc_tree_block(trans, root, parent: 0, BTRFS_TREE_LOG_OBJECTID,
961	NULL, level: 0, hint: 0, empty_size: 0, reloc_src_root: 0, nest: BTRFS_NESTING_NORMAL);
962	if (IS_ERR(ptr: leaf))
963	return PTR_ERR(ptr: leaf);
964
965	root->node = leaf;
966
967	btrfs_mark_buffer_dirty(trans, buf: root->node);
968	btrfs_tree_unlock(eb: root->node);
969
970	return 0;
971	}
972
973	int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
974	struct btrfs_fs_info *fs_info)
975	{
976	struct btrfs_root *log_root;
977
978	log_root = alloc_log_tree(trans, fs_info);
979	if (IS_ERR(ptr: log_root))
980	return PTR_ERR(ptr: log_root);
981
982	if (!btrfs_is_zoned(fs_info)) {
983	int ret = btrfs_alloc_log_tree_node(trans, root: log_root);
984
985	if (ret) {
986	btrfs_put_root(root: log_root);
987	return ret;
988	}
989	}
990
991	WARN_ON(fs_info->log_root_tree);
992	fs_info->log_root_tree = log_root;
993	return 0;
994	}
995
996	int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
997	struct btrfs_root *root)
998	{
999	struct btrfs_fs_info *fs_info = root->fs_info;
1000	struct btrfs_root *log_root;
1001	struct btrfs_inode_item *inode_item;
1002	int ret;
1003
1004	log_root = alloc_log_tree(trans, fs_info);
1005	if (IS_ERR(ptr: log_root))
1006	return PTR_ERR(ptr: log_root);
1007
1008	ret = btrfs_alloc_log_tree_node(trans, root: log_root);
1009	if (ret) {
1010	btrfs_put_root(root: log_root);
1011	return ret;
1012	}
1013
1014	log_root->last_trans = trans->transid;
1015	log_root->root_key.offset = root->root_key.objectid;
1016
1017	inode_item = &log_root->root_item.inode;
1018	btrfs_set_stack_inode_generation(s: inode_item, val: 1);
1019	btrfs_set_stack_inode_size(s: inode_item, val: 3);
1020	btrfs_set_stack_inode_nlink(s: inode_item, val: 1);
1021	btrfs_set_stack_inode_nbytes(s: inode_item,
1022	val: fs_info->nodesize);
1023	btrfs_set_stack_inode_mode(s: inode_item, S_IFDIR | 0755);
1024
1025	btrfs_set_root_node(item: &log_root->root_item, node: log_root->node);
1026
1027	WARN_ON(root->log_root);
1028	root->log_root = log_root;
1029	btrfs_set_root_log_transid(root, log_transid: 0);
1030	root->log_transid_committed = -1;
1031	btrfs_set_root_last_log_commit(root, commit_id: 0);
1032	return 0;
1033	}
1034
1035	static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1036	struct btrfs_path *path,
1037	struct btrfs_key *key)
1038	{
1039	struct btrfs_root *root;
1040	struct btrfs_tree_parent_check check = { 0 };
1041	struct btrfs_fs_info *fs_info = tree_root->fs_info;
1042	u64 generation;
1043	int ret;
1044	int level;
1045
1046	root = btrfs_alloc_root(fs_info, objectid: key->objectid, GFP_NOFS);
1047	if (!root)
1048	return ERR_PTR(error: -ENOMEM);
1049
1050	ret = btrfs_find_root(root: tree_root, search_key: key, path,
1051	root_item: &root->root_item, root_key: &root->root_key);
1052	if (ret) {
1053	if (ret > 0)
1054	ret = -ENOENT;
1055	goto fail;
1056	}
1057
1058	generation = btrfs_root_generation(s: &root->root_item);
1059	level = btrfs_root_level(s: &root->root_item);
1060	check.level = level;
1061	check.transid = generation;
1062	check.owner_root = key->objectid;
1063	root->node = read_tree_block(fs_info, bytenr: btrfs_root_bytenr(s: &root->root_item),
1064	check: &check);
1065	if (IS_ERR(ptr: root->node)) {
1066	ret = PTR_ERR(ptr: root->node);
1067	root->node = NULL;
1068	goto fail;
1069	}
1070	if (!btrfs_buffer_uptodate(eb: root->node, parent_transid: generation, atomic: 0)) {
1071	ret = -EIO;
1072	goto fail;
1073	}
1074
1075	/*
1076	* For real fs, and not log/reloc trees, root owner must
1077	* match its root node owner
1078	*/
1079	if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) &&
1080	root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1081	root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1082	root->root_key.objectid != btrfs_header_owner(eb: root->node)) {
1083	btrfs_crit(fs_info,
1084	"root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1085	root->root_key.objectid, root->node->start,
1086	btrfs_header_owner(root->node),
1087	root->root_key.objectid);
1088	ret = -EUCLEAN;
1089	goto fail;
1090	}
1091	root->commit_root = btrfs_root_node(root);
1092	return root;
1093	fail:
1094	btrfs_put_root(root);
1095	return ERR_PTR(error: ret);
1096	}
1097
1098	struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1099	struct btrfs_key *key)
1100	{
1101	struct btrfs_root *root;
1102	struct btrfs_path *path;
1103
1104	path = btrfs_alloc_path();
1105	if (!path)
1106	return ERR_PTR(error: -ENOMEM);
1107	root = read_tree_root_path(tree_root, path, key);
1108	btrfs_free_path(p: path);
1109
1110	return root;
1111	}
1112
1113	/*
1114	* Initialize subvolume root in-memory structure
1115	*
1116	* @anon_dev: anonymous device to attach to the root, if zero, allocate new
1117	*/
1118	static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1119	{
1120	int ret;
1121
1122	btrfs_drew_lock_init(lock: &root->snapshot_lock);
1123
1124	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1125	!btrfs_is_data_reloc_root(root) &&
1126	is_fstree(rootid: root->root_key.objectid)) {
1127	set_bit(nr: BTRFS_ROOT_SHAREABLE, addr: &root->state);
1128	btrfs_check_and_init_root_item(item: &root->root_item);
1129	}
1130
1131	/*
1132	* Don't assign anonymous block device to roots that are not exposed to
1133	* userspace, the id pool is limited to 1M
1134	*/
1135	if (is_fstree(rootid: root->root_key.objectid) &&
1136	btrfs_root_refs(s: &root->root_item) > 0) {
1137	if (!anon_dev) {
1138	ret = get_anon_bdev(&root->anon_dev);
1139	if (ret)
1140	goto fail;
1141	} else {
1142	root->anon_dev = anon_dev;
1143	}
1144	}
1145
1146	mutex_lock(&root->objectid_mutex);
1147	ret = btrfs_init_root_free_objectid(root);
1148	if (ret) {
1149	mutex_unlock(lock: &root->objectid_mutex);
1150	goto fail;
1151	}
1152
1153	ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1154
1155	mutex_unlock(lock: &root->objectid_mutex);
1156
1157	return 0;
1158	fail:
1159	/* The caller is responsible to call btrfs_free_fs_root */
1160	return ret;
1161	}
1162
1163	static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1164	u64 root_id)
1165	{
1166	struct btrfs_root *root;
1167
1168	spin_lock(lock: &fs_info->fs_roots_radix_lock);
1169	root = radix_tree_lookup(&fs_info->fs_roots_radix,
1170	(unsigned long)root_id);
1171	root = btrfs_grab_root(root);
1172	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
1173	return root;
1174	}
1175
1176	static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1177	u64 objectid)
1178	{
1179	struct btrfs_key key = {
1180	.objectid = objectid,
1181	.type = BTRFS_ROOT_ITEM_KEY,
1182	.offset = 0,
1183	};
1184
1185	switch (objectid) {
1186	case BTRFS_ROOT_TREE_OBJECTID:
1187	return btrfs_grab_root(root: fs_info->tree_root);
1188	case BTRFS_EXTENT_TREE_OBJECTID:
1189	return btrfs_grab_root(root: btrfs_global_root(fs_info, key: &key));
1190	case BTRFS_CHUNK_TREE_OBJECTID:
1191	return btrfs_grab_root(root: fs_info->chunk_root);
1192	case BTRFS_DEV_TREE_OBJECTID:
1193	return btrfs_grab_root(root: fs_info->dev_root);
1194	case BTRFS_CSUM_TREE_OBJECTID:
1195	return btrfs_grab_root(root: btrfs_global_root(fs_info, key: &key));
1196	case BTRFS_QUOTA_TREE_OBJECTID:
1197	return btrfs_grab_root(root: fs_info->quota_root);
1198	case BTRFS_UUID_TREE_OBJECTID:
1199	return btrfs_grab_root(root: fs_info->uuid_root);
1200	case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
1201	return btrfs_grab_root(root: fs_info->block_group_root);
1202	case BTRFS_FREE_SPACE_TREE_OBJECTID:
1203	return btrfs_grab_root(root: btrfs_global_root(fs_info, key: &key));
1204	case BTRFS_RAID_STRIPE_TREE_OBJECTID:
1205	return btrfs_grab_root(root: fs_info->stripe_root);
1206	default:
1207	return NULL;
1208	}
1209	}
1210
1211	int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1212	struct btrfs_root *root)
1213	{
1214	int ret;
1215
1216	ret = radix_tree_preload(GFP_NOFS);
1217	if (ret)
1218	return ret;
1219
1220	spin_lock(lock: &fs_info->fs_roots_radix_lock);
1221	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1222	index: (unsigned long)root->root_key.objectid,
1223	root);
1224	if (ret == 0) {
1225	btrfs_grab_root(root);
1226	set_bit(nr: BTRFS_ROOT_IN_RADIX, addr: &root->state);
1227	}
1228	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
1229	radix_tree_preload_end();
1230
1231	return ret;
1232	}
1233
1234	void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1235	{
1236	#ifdef CONFIG_BTRFS_DEBUG
1237	struct btrfs_root *root;
1238
1239	while (!list_empty(head: &fs_info->allocated_roots)) {
1240	char buf[BTRFS_ROOT_NAME_BUF_LEN];
1241
1242	root = list_first_entry(&fs_info->allocated_roots,
1243	struct btrfs_root, leak_list);
1244	btrfs_err(fs_info, "leaked root %s refcount %d",
1245	btrfs_root_name(&root->root_key, buf),
1246	refcount_read(&root->refs));
1247	WARN_ON_ONCE(1);
1248	while (refcount_read(r: &root->refs) > 1)
1249	btrfs_put_root(root);
1250	btrfs_put_root(root);
1251	}
1252	#endif
1253	}
1254
1255	static void free_global_roots(struct btrfs_fs_info *fs_info)
1256	{
1257	struct btrfs_root *root;
1258	struct rb_node *node;
1259
1260	while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1261	root = rb_entry(node, struct btrfs_root, rb_node);
1262	rb_erase(&root->rb_node, &fs_info->global_root_tree);
1263	btrfs_put_root(root);
1264	}
1265	}
1266
1267	void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1268	{
1269	percpu_counter_destroy(fbc: &fs_info->dirty_metadata_bytes);
1270	percpu_counter_destroy(fbc: &fs_info->delalloc_bytes);
1271	percpu_counter_destroy(fbc: &fs_info->ordered_bytes);
1272	percpu_counter_destroy(fbc: &fs_info->dev_replace.bio_counter);
1273	btrfs_free_csum_hash(fs_info);
1274	btrfs_free_stripe_hash_table(info: fs_info);
1275	btrfs_free_ref_cache(fs_info);
1276	kfree(objp: fs_info->balance_ctl);
1277	kfree(objp: fs_info->delayed_root);
1278	free_global_roots(fs_info);
1279	btrfs_put_root(root: fs_info->tree_root);
1280	btrfs_put_root(root: fs_info->chunk_root);
1281	btrfs_put_root(root: fs_info->dev_root);
1282	btrfs_put_root(root: fs_info->quota_root);
1283	btrfs_put_root(root: fs_info->uuid_root);
1284	btrfs_put_root(root: fs_info->fs_root);
1285	btrfs_put_root(root: fs_info->data_reloc_root);
1286	btrfs_put_root(root: fs_info->block_group_root);
1287	btrfs_put_root(root: fs_info->stripe_root);
1288	btrfs_check_leaked_roots(fs_info);
1289	btrfs_extent_buffer_leak_debug_check(fs_info);
1290	kfree(objp: fs_info->super_copy);
1291	kfree(objp: fs_info->super_for_commit);
1292	kfree(objp: fs_info->subpage_info);
1293	kvfree(addr: fs_info);
1294	}
1295
1296
1297	/*
1298	* Get an in-memory reference of a root structure.
1299	*
1300	* For essential trees like root/extent tree, we grab it from fs_info directly.
1301	* For subvolume trees, we check the cached filesystem roots first. If not
1302	* found, then read it from disk and add it to cached fs roots.
1303	*
1304	* Caller should release the root by calling btrfs_put_root() after the usage.
1305	*
1306	* NOTE: Reloc and log trees can't be read by this function as they share the
1307	* same root objectid.
1308	*
1309	* @objectid: root id
1310	* @anon_dev: preallocated anonymous block device number for new roots,
1311	* pass NULL for a new allocation.
1312	* @check_ref: whether to check root item references, If true, return -ENOENT
1313	* for orphan roots
1314	*/
1315	static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1316	u64 objectid, dev_t *anon_dev,
1317	bool check_ref)
1318	{
1319	struct btrfs_root *root;
1320	struct btrfs_path *path;
1321	struct btrfs_key key;
1322	int ret;
1323
1324	root = btrfs_get_global_root(fs_info, objectid);
1325	if (root)
1326	return root;
1327
1328	/*
1329	* If we're called for non-subvolume trees, and above function didn't
1330	* find one, do not try to read it from disk.
1331	*
1332	* This is namely for free-space-tree and quota tree, which can change
1333	* at runtime and should only be grabbed from fs_info.
1334	*/
1335	if (!is_fstree(rootid: objectid) && objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1336	return ERR_PTR(error: -ENOENT);
1337	again:
1338	root = btrfs_lookup_fs_root(fs_info, root_id: objectid);
1339	if (root) {
1340	/*
1341	* Some other caller may have read out the newly inserted
1342	* subvolume already (for things like backref walk etc). Not
1343	* that common but still possible. In that case, we just need
1344	* to free the anon_dev.
1345	*/
1346	if (unlikely(anon_dev && *anon_dev)) {
1347	free_anon_bdev(*anon_dev);
1348	*anon_dev = 0;
1349	}
1350
1351	if (check_ref && btrfs_root_refs(s: &root->root_item) == 0) {
1352	btrfs_put_root(root);
1353	return ERR_PTR(error: -ENOENT);
1354	}
1355	return root;
1356	}
1357
1358	key.objectid = objectid;
1359	key.type = BTRFS_ROOT_ITEM_KEY;
1360	key.offset = (u64)-1;
1361	root = btrfs_read_tree_root(tree_root: fs_info->tree_root, key: &key);
1362	if (IS_ERR(ptr: root))
1363	return root;
1364
1365	if (check_ref && btrfs_root_refs(s: &root->root_item) == 0) {
1366	ret = -ENOENT;
1367	goto fail;
1368	}
1369
1370	ret = btrfs_init_fs_root(root, anon_dev: anon_dev ? *anon_dev : 0);
1371	if (ret)
1372	goto fail;
1373
1374	path = btrfs_alloc_path();
1375	if (!path) {
1376	ret = -ENOMEM;
1377	goto fail;
1378	}
1379	key.objectid = BTRFS_ORPHAN_OBJECTID;
1380	key.type = BTRFS_ORPHAN_ITEM_KEY;
1381	key.offset = objectid;
1382
1383	ret = btrfs_search_slot(NULL, root: fs_info->tree_root, key: &key, p: path, ins_len: 0, cow: 0);
1384	btrfs_free_path(p: path);
1385	if (ret < 0)
1386	goto fail;
1387	if (ret == 0)
1388	set_bit(nr: BTRFS_ROOT_ORPHAN_ITEM_INSERTED, addr: &root->state);
1389
1390	ret = btrfs_insert_fs_root(fs_info, root);
1391	if (ret) {
1392	if (ret == -EEXIST) {
1393	btrfs_put_root(root);
1394	goto again;
1395	}
1396	goto fail;
1397	}
1398	return root;
1399	fail:
1400	/*
1401	* If our caller provided us an anonymous device, then it's his
1402	* responsibility to free it in case we fail. So we have to set our
1403	* root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1404	* and once again by our caller.
1405	*/
1406	if (anon_dev && *anon_dev)
1407	root->anon_dev = 0;
1408	btrfs_put_root(root);
1409	return ERR_PTR(error: ret);
1410	}
1411
1412	/*
1413	* Get in-memory reference of a root structure
1414	*
1415	* @objectid: tree objectid
1416	* @check_ref: if set, verify that the tree exists and the item has at least
1417	* one reference
1418	*/
1419	struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1420	u64 objectid, bool check_ref)
1421	{
1422	return btrfs_get_root_ref(fs_info, objectid, NULL, check_ref);
1423	}
1424
1425	/*
1426	* Get in-memory reference of a root structure, created as new, optionally pass
1427	* the anonymous block device id
1428	*
1429	* @objectid: tree objectid
1430	* @anon_dev: if NULL, allocate a new anonymous block device or use the
1431	* parameter value if not NULL
1432	*/
1433	struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1434	u64 objectid, dev_t *anon_dev)
1435	{
1436	return btrfs_get_root_ref(fs_info, objectid, anon_dev, check_ref: true);
1437	}
1438
1439	/*
1440	* Return a root for the given objectid.
1441	*
1442	* @fs_info: the fs_info
1443	* @objectid: the objectid we need to lookup
1444	*
1445	* This is exclusively used for backref walking, and exists specifically because
1446	* of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1447	* creation time, which means we may have to read the tree_root in order to look
1448	* up a fs root that is not in memory. If the root is not in memory we will
1449	* read the tree root commit root and look up the fs root from there. This is a
1450	* temporary root, it will not be inserted into the radix tree as it doesn't
1451	* have the most uptodate information, it'll simply be discarded once the
1452	* backref code is finished using the root.
1453	*/
1454	struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1455	struct btrfs_path *path,
1456	u64 objectid)
1457	{
1458	struct btrfs_root *root;
1459	struct btrfs_key key;
1460
1461	ASSERT(path->search_commit_root && path->skip_locking);
1462
1463	/*
1464	* This can return -ENOENT if we ask for a root that doesn't exist, but
1465	* since this is called via the backref walking code we won't be looking
1466	* up a root that doesn't exist, unless there's corruption. So if root
1467	* != NULL just return it.
1468	*/
1469	root = btrfs_get_global_root(fs_info, objectid);
1470	if (root)
1471	return root;
1472
1473	root = btrfs_lookup_fs_root(fs_info, root_id: objectid);
1474	if (root)
1475	return root;
1476
1477	key.objectid = objectid;
1478	key.type = BTRFS_ROOT_ITEM_KEY;
1479	key.offset = (u64)-1;
1480	root = read_tree_root_path(tree_root: fs_info->tree_root, path, key: &key);
1481	btrfs_release_path(p: path);
1482
1483	return root;
1484	}
1485
1486	static int cleaner_kthread(void *arg)
1487	{
1488	struct btrfs_fs_info *fs_info = arg;
1489	int again;
1490
1491	while (1) {
1492	again = 0;
1493
1494	set_bit(nr: BTRFS_FS_CLEANER_RUNNING, addr: &fs_info->flags);
1495
1496	/* Make the cleaner go to sleep early. */
1497	if (btrfs_need_cleaner_sleep(fs_info))
1498	goto sleep;
1499
1500	/*
1501	* Do not do anything if we might cause open_ctree() to block
1502	* before we have finished mounting the filesystem.
1503	*/
1504	if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1505	goto sleep;
1506
1507	if (!mutex_trylock(lock: &fs_info->cleaner_mutex))
1508	goto sleep;
1509
1510	/*
1511	* Avoid the problem that we change the status of the fs
1512	* during the above check and trylock.
1513	*/
1514	if (btrfs_need_cleaner_sleep(fs_info)) {
1515	mutex_unlock(lock: &fs_info->cleaner_mutex);
1516	goto sleep;
1517	}
1518
1519	if (test_and_clear_bit(nr: BTRFS_FS_FEATURE_CHANGED, addr: &fs_info->flags))
1520	btrfs_sysfs_feature_update(fs_info);
1521
1522	btrfs_run_delayed_iputs(fs_info);
1523
1524	again = btrfs_clean_one_deleted_snapshot(fs_info);
1525	mutex_unlock(lock: &fs_info->cleaner_mutex);
1526
1527	/*
1528	* The defragger has dealt with the R/O remount and umount,
1529	* needn't do anything special here.
1530	*/
1531	btrfs_run_defrag_inodes(fs_info);
1532
1533	/*
1534	* Acquires fs_info->reclaim_bgs_lock to avoid racing
1535	* with relocation (btrfs_relocate_chunk) and relocation
1536	* acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1537	* after acquiring fs_info->reclaim_bgs_lock. So we
1538	* can't hold, nor need to, fs_info->cleaner_mutex when deleting
1539	* unused block groups.
1540	*/
1541	btrfs_delete_unused_bgs(fs_info);
1542
1543	/*
1544	* Reclaim block groups in the reclaim_bgs list after we deleted
1545	* all unused block_groups. This possibly gives us some more free
1546	* space.
1547	*/
1548	btrfs_reclaim_bgs(fs_info);
1549	sleep:
1550	clear_and_wake_up_bit(bit: BTRFS_FS_CLEANER_RUNNING, word: &fs_info->flags);
1551	if (kthread_should_park())
1552	kthread_parkme();
1553	if (kthread_should_stop())
1554	return 0;
1555	if (!again) {
1556	set_current_state(TASK_INTERRUPTIBLE);
1557	schedule();
1558	__set_current_state(TASK_RUNNING);
1559	}
1560	}
1561	}
1562
1563	static int transaction_kthread(void *arg)
1564	{
1565	struct btrfs_root *root = arg;
1566	struct btrfs_fs_info *fs_info = root->fs_info;
1567	struct btrfs_trans_handle *trans;
1568	struct btrfs_transaction *cur;
1569	u64 transid;
1570	time64_t delta;
1571	unsigned long delay;
1572	bool cannot_commit;
1573
1574	do {
1575	cannot_commit = false;
1576	delay = msecs_to_jiffies(m: fs_info->commit_interval * 1000);
1577	mutex_lock(&fs_info->transaction_kthread_mutex);
1578
1579	spin_lock(lock: &fs_info->trans_lock);
1580	cur = fs_info->running_transaction;
1581	if (!cur) {
1582	spin_unlock(lock: &fs_info->trans_lock);
1583	goto sleep;
1584	}
1585
1586	delta = ktime_get_seconds() - cur->start_time;
1587	if (!test_and_clear_bit(nr: BTRFS_FS_COMMIT_TRANS, addr: &fs_info->flags) &&
1588	cur->state < TRANS_STATE_COMMIT_PREP &&
1589	delta < fs_info->commit_interval) {
1590	spin_unlock(lock: &fs_info->trans_lock);
1591	delay -= msecs_to_jiffies(m: (delta - 1) * 1000);
1592	delay = min(delay,
1593	msecs_to_jiffies(fs_info->commit_interval * 1000));
1594	goto sleep;
1595	}
1596	transid = cur->transid;
1597	spin_unlock(lock: &fs_info->trans_lock);
1598
1599	/* If the file system is aborted, this will always fail. */
1600	trans = btrfs_attach_transaction(root);
1601	if (IS_ERR(ptr: trans)) {
1602	if (PTR_ERR(ptr: trans) != -ENOENT)
1603	cannot_commit = true;
1604	goto sleep;
1605	}
1606	if (transid == trans->transid) {
1607	btrfs_commit_transaction(trans);
1608	} else {
1609	btrfs_end_transaction(trans);
1610	}
1611	sleep:
1612	wake_up_process(tsk: fs_info->cleaner_kthread);
1613	mutex_unlock(lock: &fs_info->transaction_kthread_mutex);
1614
1615	if (BTRFS_FS_ERROR(fs_info))
1616	btrfs_cleanup_transaction(fs_info);
1617	if (!kthread_should_stop() &&
1618	(!btrfs_transaction_blocked(info: fs_info) ||
1619	cannot_commit))
1620	schedule_timeout_interruptible(timeout: delay);
1621	} while (!kthread_should_stop());
1622	return 0;
1623	}
1624
1625	/*
1626	* This will find the highest generation in the array of root backups. The
1627	* index of the highest array is returned, or -EINVAL if we can't find
1628	* anything.
1629	*
1630	* We check to make sure the array is valid by comparing the
1631	* generation of the latest root in the array with the generation
1632	* in the super block. If they don't match we pitch it.
1633	*/
1634	static int find_newest_super_backup(struct btrfs_fs_info *info)
1635	{
1636	const u64 newest_gen = btrfs_super_generation(s: info->super_copy);
1637	u64 cur;
1638	struct btrfs_root_backup *root_backup;
1639	int i;
1640
1641	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1642	root_backup = info->super_copy->super_roots + i;
1643	cur = btrfs_backup_tree_root_gen(s: root_backup);
1644	if (cur == newest_gen)
1645	return i;
1646	}
1647
1648	return -EINVAL;
1649	}
1650
1651	/*
1652	* copy all the root pointers into the super backup array.
1653	* this will bump the backup pointer by one when it is
1654	* done
1655	*/
1656	static void backup_super_roots(struct btrfs_fs_info *info)
1657	{
1658	const int next_backup = info->backup_root_index;
1659	struct btrfs_root_backup *root_backup;
1660
1661	root_backup = info->super_for_commit->super_roots + next_backup;
1662
1663	/*
1664	* make sure all of our padding and empty slots get zero filled
1665	* regardless of which ones we use today
1666	*/
1667	memset(root_backup, 0, sizeof(*root_backup));
1668
1669	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1670
1671	btrfs_set_backup_tree_root(s: root_backup, val: info->tree_root->node->start);
1672	btrfs_set_backup_tree_root_gen(s: root_backup,
1673	val: btrfs_header_generation(eb: info->tree_root->node));
1674
1675	btrfs_set_backup_tree_root_level(s: root_backup,
1676	val: btrfs_header_level(eb: info->tree_root->node));
1677
1678	btrfs_set_backup_chunk_root(s: root_backup, val: info->chunk_root->node->start);
1679	btrfs_set_backup_chunk_root_gen(s: root_backup,
1680	val: btrfs_header_generation(eb: info->chunk_root->node));
1681	btrfs_set_backup_chunk_root_level(s: root_backup,
1682	val: btrfs_header_level(eb: info->chunk_root->node));
1683
1684	if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
1685	struct btrfs_root *extent_root = btrfs_extent_root(fs_info: info, bytenr: 0);
1686	struct btrfs_root *csum_root = btrfs_csum_root(fs_info: info, bytenr: 0);
1687
1688	btrfs_set_backup_extent_root(s: root_backup,
1689	val: extent_root->node->start);
1690	btrfs_set_backup_extent_root_gen(s: root_backup,
1691	val: btrfs_header_generation(eb: extent_root->node));
1692	btrfs_set_backup_extent_root_level(s: root_backup,
1693	val: btrfs_header_level(eb: extent_root->node));
1694
1695	btrfs_set_backup_csum_root(s: root_backup, val: csum_root->node->start);
1696	btrfs_set_backup_csum_root_gen(s: root_backup,
1697	val: btrfs_header_generation(eb: csum_root->node));
1698	btrfs_set_backup_csum_root_level(s: root_backup,
1699	val: btrfs_header_level(eb: csum_root->node));
1700	}
1701
1702	/*
1703	* we might commit during log recovery, which happens before we set
1704	* the fs_root. Make sure it is valid before we fill it in.
1705	*/
1706	if (info->fs_root && info->fs_root->node) {
1707	btrfs_set_backup_fs_root(s: root_backup,
1708	val: info->fs_root->node->start);
1709	btrfs_set_backup_fs_root_gen(s: root_backup,
1710	val: btrfs_header_generation(eb: info->fs_root->node));
1711	btrfs_set_backup_fs_root_level(s: root_backup,
1712	val: btrfs_header_level(eb: info->fs_root->node));
1713	}
1714
1715	btrfs_set_backup_dev_root(s: root_backup, val: info->dev_root->node->start);
1716	btrfs_set_backup_dev_root_gen(s: root_backup,
1717	val: btrfs_header_generation(eb: info->dev_root->node));
1718	btrfs_set_backup_dev_root_level(s: root_backup,
1719	val: btrfs_header_level(eb: info->dev_root->node));
1720
1721	btrfs_set_backup_total_bytes(s: root_backup,
1722	val: btrfs_super_total_bytes(s: info->super_copy));
1723	btrfs_set_backup_bytes_used(s: root_backup,
1724	val: btrfs_super_bytes_used(s: info->super_copy));
1725	btrfs_set_backup_num_devices(s: root_backup,
1726	val: btrfs_super_num_devices(s: info->super_copy));
1727
1728	/*
1729	* if we don't copy this out to the super_copy, it won't get remembered
1730	* for the next commit
1731	*/
1732	memcpy(&info->super_copy->super_roots,
1733	&info->super_for_commit->super_roots,
1734	sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1735	}
1736
1737	/*
1738	* Reads a backup root based on the passed priority. Prio 0 is the newest, prio
1739	* 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1740	*
1741	* @fs_info: filesystem whose backup roots need to be read
1742	* @priority: priority of backup root required
1743	*
1744	* Returns backup root index on success and -EINVAL otherwise.
1745	*/
1746	static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1747	{
1748	int backup_index = find_newest_super_backup(info: fs_info);
1749	struct btrfs_super_block *super = fs_info->super_copy;
1750	struct btrfs_root_backup *root_backup;
1751
1752	if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1753	if (priority == 0)
1754	return backup_index;
1755
1756	backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1757	backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1758	} else {
1759	return -EINVAL;
1760	}
1761
1762	root_backup = super->super_roots + backup_index;
1763
1764	btrfs_set_super_generation(s: super,
1765	val: btrfs_backup_tree_root_gen(s: root_backup));
1766	btrfs_set_super_root(s: super, val: btrfs_backup_tree_root(s: root_backup));
1767	btrfs_set_super_root_level(s: super,
1768	val: btrfs_backup_tree_root_level(s: root_backup));
1769	btrfs_set_super_bytes_used(s: super, val: btrfs_backup_bytes_used(s: root_backup));
1770
1771	/*
1772	* Fixme: the total bytes and num_devices need to match or we should
1773	* need a fsck
1774	*/
1775	btrfs_set_super_total_bytes(s: super, val: btrfs_backup_total_bytes(s: root_backup));
1776	btrfs_set_super_num_devices(s: super, val: btrfs_backup_num_devices(s: root_backup));
1777
1778	return backup_index;
1779	}
1780
1781	/* helper to cleanup workers */
1782	static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1783	{
1784	btrfs_destroy_workqueue(wq: fs_info->fixup_workers);
1785	btrfs_destroy_workqueue(wq: fs_info->delalloc_workers);
1786	btrfs_destroy_workqueue(wq: fs_info->workers);
1787	if (fs_info->endio_workers)
1788	destroy_workqueue(wq: fs_info->endio_workers);
1789	if (fs_info->rmw_workers)
1790	destroy_workqueue(wq: fs_info->rmw_workers);
1791	if (fs_info->compressed_write_workers)
1792	destroy_workqueue(wq: fs_info->compressed_write_workers);
1793	btrfs_destroy_workqueue(wq: fs_info->endio_write_workers);
1794	btrfs_destroy_workqueue(wq: fs_info->endio_freespace_worker);
1795	btrfs_destroy_workqueue(wq: fs_info->delayed_workers);
1796	btrfs_destroy_workqueue(wq: fs_info->caching_workers);
1797	btrfs_destroy_workqueue(wq: fs_info->flush_workers);
1798	btrfs_destroy_workqueue(wq: fs_info->qgroup_rescan_workers);
1799	if (fs_info->discard_ctl.discard_workers)
1800	destroy_workqueue(wq: fs_info->discard_ctl.discard_workers);
1801	/*
1802	* Now that all other work queues are destroyed, we can safely destroy
1803	* the queues used for metadata I/O, since tasks from those other work
1804	* queues can do metadata I/O operations.
1805	*/
1806	if (fs_info->endio_meta_workers)
1807	destroy_workqueue(wq: fs_info->endio_meta_workers);
1808	}
1809
1810	static void free_root_extent_buffers(struct btrfs_root *root)
1811	{
1812	if (root) {
1813	free_extent_buffer(eb: root->node);
1814	free_extent_buffer(eb: root->commit_root);
1815	root->node = NULL;
1816	root->commit_root = NULL;
1817	}
1818	}
1819
1820	static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
1821	{
1822	struct btrfs_root *root, *tmp;
1823
1824	rbtree_postorder_for_each_entry_safe(root, tmp,
1825	&fs_info->global_root_tree,
1826	rb_node)
1827	free_root_extent_buffers(root);
1828	}
1829
1830	/* helper to cleanup tree roots */
1831	static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1832	{
1833	free_root_extent_buffers(root: info->tree_root);
1834
1835	free_global_root_pointers(fs_info: info);
1836	free_root_extent_buffers(root: info->dev_root);
1837	free_root_extent_buffers(root: info->quota_root);
1838	free_root_extent_buffers(root: info->uuid_root);
1839	free_root_extent_buffers(root: info->fs_root);
1840	free_root_extent_buffers(root: info->data_reloc_root);
1841	free_root_extent_buffers(root: info->block_group_root);
1842	free_root_extent_buffers(root: info->stripe_root);
1843	if (free_chunk_root)
1844	free_root_extent_buffers(root: info->chunk_root);
1845	}
1846
1847	void btrfs_put_root(struct btrfs_root *root)
1848	{
1849	if (!root)
1850	return;
1851
1852	if (refcount_dec_and_test(r: &root->refs)) {
1853	WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1854	WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
1855	if (root->anon_dev)
1856	free_anon_bdev(root->anon_dev);
1857	free_root_extent_buffers(root);
1858	#ifdef CONFIG_BTRFS_DEBUG
1859	spin_lock(lock: &root->fs_info->fs_roots_radix_lock);
1860	list_del_init(entry: &root->leak_list);
1861	spin_unlock(lock: &root->fs_info->fs_roots_radix_lock);
1862	#endif
1863	kfree(objp: root);
1864	}
1865	}
1866
1867	void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
1868	{
1869	int ret;
1870	struct btrfs_root *gang[8];
1871	int i;
1872
1873	while (!list_empty(head: &fs_info->dead_roots)) {
1874	gang[0] = list_entry(fs_info->dead_roots.next,
1875	struct btrfs_root, root_list);
1876	list_del(entry: &gang[0]->root_list);
1877
1878	if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
1879	btrfs_drop_and_free_fs_root(fs_info, root: gang[0]);
1880	btrfs_put_root(root: gang[0]);
1881	}
1882
1883	while (1) {
1884	ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1885	results: (void **)gang, first_index: 0,
1886	ARRAY_SIZE(gang));
1887	if (!ret)
1888	break;
1889	for (i = 0; i < ret; i++)
1890	btrfs_drop_and_free_fs_root(fs_info, root: gang[i]);
1891	}
1892	}
1893
1894	static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
1895	{
1896	mutex_init(&fs_info->scrub_lock);
1897	atomic_set(v: &fs_info->scrubs_running, i: 0);
1898	atomic_set(v: &fs_info->scrub_pause_req, i: 0);
1899	atomic_set(v: &fs_info->scrubs_paused, i: 0);
1900	atomic_set(v: &fs_info->scrub_cancel_req, i: 0);
1901	init_waitqueue_head(&fs_info->scrub_pause_wait);
1902	refcount_set(r: &fs_info->scrub_workers_refcnt, n: 0);
1903	}
1904
1905	static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
1906	{
1907	spin_lock_init(&fs_info->balance_lock);
1908	mutex_init(&fs_info->balance_mutex);
1909	atomic_set(v: &fs_info->balance_pause_req, i: 0);
1910	atomic_set(v: &fs_info->balance_cancel_req, i: 0);
1911	fs_info->balance_ctl = NULL;
1912	init_waitqueue_head(&fs_info->balance_wait_q);
1913	atomic_set(v: &fs_info->reloc_cancel_req, i: 0);
1914	}
1915
1916	static int btrfs_init_btree_inode(struct super_block *sb)
1917	{
1918	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1919	unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
1920	root: fs_info->tree_root);
1921	struct inode *inode;
1922
1923	inode = new_inode(sb);
1924	if (!inode)
1925	return -ENOMEM;
1926
1927	inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1928	set_nlink(inode, nlink: 1);
1929	/*
1930	* we set the i_size on the btree inode to the max possible int.
1931	* the real end of the address space is determined by all of
1932	* the devices in the system
1933	*/
1934	inode->i_size = OFFSET_MAX;
1935	inode->i_mapping->a_ops = &btree_aops;
1936	mapping_set_gfp_mask(m: inode->i_mapping, GFP_NOFS);
1937
1938	RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
1939	extent_io_tree_init(fs_info, tree: &BTRFS_I(inode)->io_tree,
1940	owner: IO_TREE_BTREE_INODE_IO);
1941	extent_map_tree_init(tree: &BTRFS_I(inode)->extent_tree);
1942
1943	BTRFS_I(inode)->root = btrfs_grab_root(root: fs_info->tree_root);
1944	BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID;
1945	BTRFS_I(inode)->location.type = 0;
1946	BTRFS_I(inode)->location.offset = 0;
1947	set_bit(nr: BTRFS_INODE_DUMMY, addr: &BTRFS_I(inode)->runtime_flags);
1948	__insert_inode_hash(inode, hashval: hash);
1949	fs_info->btree_inode = inode;
1950
1951	return 0;
1952	}
1953
1954	static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
1955	{
1956	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
1957	init_rwsem(&fs_info->dev_replace.rwsem);
1958	init_waitqueue_head(&fs_info->dev_replace.replace_wait);
1959	}
1960
1961	static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
1962	{
1963	spin_lock_init(&fs_info->qgroup_lock);
1964	mutex_init(&fs_info->qgroup_ioctl_lock);
1965	fs_info->qgroup_tree = RB_ROOT;
1966	INIT_LIST_HEAD(list: &fs_info->dirty_qgroups);
1967	fs_info->qgroup_seq = 1;
1968	fs_info->qgroup_ulist = NULL;
1969	fs_info->qgroup_rescan_running = false;
1970	fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
1971	mutex_init(&fs_info->qgroup_rescan_lock);
1972	}
1973
1974	static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
1975	{
1976	u32 max_active = fs_info->thread_pool_size;
1977	unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
1978	unsigned int ordered_flags = WQ_MEM_RECLAIM | WQ_FREEZABLE;
1979
1980	fs_info->workers =
1981	btrfs_alloc_workqueue(fs_info, name: "worker", flags, limit_active: max_active, thresh: 16);
1982
1983	fs_info->delalloc_workers =
1984	btrfs_alloc_workqueue(fs_info, name: "delalloc",
1985	flags, limit_active: max_active, thresh: 2);
1986
1987	fs_info->flush_workers =
1988	btrfs_alloc_workqueue(fs_info, name: "flush_delalloc",
1989	flags, limit_active: max_active, thresh: 0);
1990
1991	fs_info->caching_workers =
1992	btrfs_alloc_workqueue(fs_info, name: "cache", flags, limit_active: max_active, thresh: 0);
1993
1994	fs_info->fixup_workers =
1995	btrfs_alloc_ordered_workqueue(fs_info, name: "fixup", flags: ordered_flags);
1996
1997	fs_info->endio_workers =
1998	alloc_workqueue(fmt: "btrfs-endio", flags, max_active);
1999	fs_info->endio_meta_workers =
2000	alloc_workqueue(fmt: "btrfs-endio-meta", flags, max_active);
2001	fs_info->rmw_workers = alloc_workqueue(fmt: "btrfs-rmw", flags, max_active);
2002	fs_info->endio_write_workers =
2003	btrfs_alloc_workqueue(fs_info, name: "endio-write", flags,
2004	limit_active: max_active, thresh: 2);
2005	fs_info->compressed_write_workers =
2006	alloc_workqueue(fmt: "btrfs-compressed-write", flags, max_active);
2007	fs_info->endio_freespace_worker =
2008	btrfs_alloc_workqueue(fs_info, name: "freespace-write", flags,
2009	limit_active: max_active, thresh: 0);
2010	fs_info->delayed_workers =
2011	btrfs_alloc_workqueue(fs_info, name: "delayed-meta", flags,
2012	limit_active: max_active, thresh: 0);
2013	fs_info->qgroup_rescan_workers =
2014	btrfs_alloc_ordered_workqueue(fs_info, name: "qgroup-rescan",
2015	flags: ordered_flags);
2016	fs_info->discard_ctl.discard_workers =
2017	alloc_ordered_workqueue("btrfs_discard", WQ_FREEZABLE);
2018
2019	if (!(fs_info->workers &&
2020	fs_info->delalloc_workers && fs_info->flush_workers &&
2021	fs_info->endio_workers && fs_info->endio_meta_workers &&
2022	fs_info->compressed_write_workers &&
2023	fs_info->endio_write_workers &&
2024	fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2025	fs_info->caching_workers && fs_info->fixup_workers &&
2026	fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2027	fs_info->discard_ctl.discard_workers)) {
2028	return -ENOMEM;
2029	}
2030
2031	return 0;
2032	}
2033
2034	static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2035	{
2036	struct crypto_shash *csum_shash;
2037	const char *csum_driver = btrfs_super_csum_driver(csum_type);
2038
2039	csum_shash = crypto_alloc_shash(alg_name: csum_driver, type: 0, mask: 0);
2040
2041	if (IS_ERR(ptr: csum_shash)) {
2042	btrfs_err(fs_info, "error allocating %s hash for checksum",
2043	csum_driver);
2044	return PTR_ERR(ptr: csum_shash);
2045	}
2046
2047	fs_info->csum_shash = csum_shash;
2048
2049	/*
2050	* Check if the checksum implementation is a fast accelerated one.
2051	* As-is this is a bit of a hack and should be replaced once the csum
2052	* implementations provide that information themselves.
2053	*/
2054	switch (csum_type) {
2055	case BTRFS_CSUM_TYPE_CRC32:
2056	if (!strstr(crypto_shash_driver_name(tfm: csum_shash), "generic"))
2057	set_bit(nr: BTRFS_FS_CSUM_IMPL_FAST, addr: &fs_info->flags);
2058	break;
2059	case BTRFS_CSUM_TYPE_XXHASH:
2060	set_bit(nr: BTRFS_FS_CSUM_IMPL_FAST, addr: &fs_info->flags);
2061	break;
2062	default:
2063	break;
2064	}
2065
2066	btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2067	btrfs_super_csum_name(csum_type),
2068	crypto_shash_driver_name(csum_shash));
2069	return 0;
2070	}
2071
2072	static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2073	struct btrfs_fs_devices *fs_devices)
2074	{
2075	int ret;
2076	struct btrfs_tree_parent_check check = { 0 };
2077	struct btrfs_root *log_tree_root;
2078	struct btrfs_super_block *disk_super = fs_info->super_copy;
2079	u64 bytenr = btrfs_super_log_root(s: disk_super);
2080	int level = btrfs_super_log_root_level(s: disk_super);
2081
2082	if (fs_devices->rw_devices == 0) {
2083	btrfs_warn(fs_info, "log replay required on RO media");
2084	return -EIO;
2085	}
2086
2087	log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2088	GFP_KERNEL);
2089	if (!log_tree_root)
2090	return -ENOMEM;
2091
2092	check.level = level;
2093	check.transid = fs_info->generation + 1;
2094	check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2095	log_tree_root->node = read_tree_block(fs_info, bytenr, check: &check);
2096	if (IS_ERR(ptr: log_tree_root->node)) {
2097	btrfs_warn(fs_info, "failed to read log tree");
2098	ret = PTR_ERR(ptr: log_tree_root->node);
2099	log_tree_root->node = NULL;
2100	btrfs_put_root(root: log_tree_root);
2101	return ret;
2102	}
2103	if (!extent_buffer_uptodate(eb: log_tree_root->node)) {
2104	btrfs_err(fs_info, "failed to read log tree");
2105	btrfs_put_root(root: log_tree_root);
2106	return -EIO;
2107	}
2108
2109	/* returns with log_tree_root freed on success */
2110	ret = btrfs_recover_log_trees(tree_root: log_tree_root);
2111	if (ret) {
2112	btrfs_handle_fs_error(fs_info, ret,
2113	"Failed to recover log tree");
2114	btrfs_put_root(root: log_tree_root);
2115	return ret;
2116	}
2117
2118	if (sb_rdonly(sb: fs_info->sb)) {
2119	ret = btrfs_commit_super(fs_info);
2120	if (ret)
2121	return ret;
2122	}
2123
2124	return 0;
2125	}
2126
2127	static int load_global_roots_objectid(struct btrfs_root *tree_root,
2128	struct btrfs_path *path, u64 objectid,
2129	const char *name)
2130	{
2131	struct btrfs_fs_info *fs_info = tree_root->fs_info;
2132	struct btrfs_root *root;
2133	u64 max_global_id = 0;
2134	int ret;
2135	struct btrfs_key key = {
2136	.objectid = objectid,
2137	.type = BTRFS_ROOT_ITEM_KEY,
2138	.offset = 0,
2139	};
2140	bool found = false;
2141
2142	/* If we have IGNOREDATACSUMS skip loading these roots. */
2143	if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2144	btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2145	set_bit(nr: BTRFS_FS_STATE_NO_CSUMS, addr: &fs_info->fs_state);
2146	return 0;
2147	}
2148
2149	while (1) {
2150	ret = btrfs_search_slot(NULL, root: tree_root, key: &key, p: path, ins_len: 0, cow: 0);
2151	if (ret < 0)
2152	break;
2153
2154	if (path->slots[0] >= btrfs_header_nritems(eb: path->nodes[0])) {
2155	ret = btrfs_next_leaf(root: tree_root, path);
2156	if (ret) {
2157	if (ret > 0)
2158	ret = 0;
2159	break;
2160	}
2161	}
2162	ret = 0;
2163
2164	btrfs_item_key_to_cpu(eb: path->nodes[0], cpu_key: &key, nr: path->slots[0]);
2165	if (key.objectid != objectid)
2166	break;
2167	btrfs_release_path(p: path);
2168
2169	/*
2170	* Just worry about this for extent tree, it'll be the same for
2171	* everybody.
2172	*/
2173	if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2174	max_global_id = max(max_global_id, key.offset);
2175
2176	found = true;
2177	root = read_tree_root_path(tree_root, path, key: &key);
2178	if (IS_ERR(ptr: root)) {
2179	if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2180	ret = PTR_ERR(ptr: root);
2181	break;
2182	}
2183	set_bit(nr: BTRFS_ROOT_TRACK_DIRTY, addr: &root->state);
2184	ret = btrfs_global_root_insert(root);
2185	if (ret) {
2186	btrfs_put_root(root);
2187	break;
2188	}
2189	key.offset++;
2190	}
2191	btrfs_release_path(p: path);
2192
2193	if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2194	fs_info->nr_global_roots = max_global_id + 1;
2195
2196	if (!found || ret) {
2197	if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2198	set_bit(nr: BTRFS_FS_STATE_NO_CSUMS, addr: &fs_info->fs_state);
2199
2200	if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2201	ret = ret ? ret : -ENOENT;
2202	else
2203	ret = 0;
2204	btrfs_err(fs_info, "failed to load root %s", name);
2205	}
2206	return ret;
2207	}
2208
2209	static int load_global_roots(struct btrfs_root *tree_root)
2210	{
2211	struct btrfs_path *path;
2212	int ret = 0;
2213
2214	path = btrfs_alloc_path();
2215	if (!path)
2216	return -ENOMEM;
2217
2218	ret = load_global_roots_objectid(tree_root, path,
2219	BTRFS_EXTENT_TREE_OBJECTID, name: "extent");
2220	if (ret)
2221	goto out;
2222	ret = load_global_roots_objectid(tree_root, path,
2223	BTRFS_CSUM_TREE_OBJECTID, name: "csum");
2224	if (ret)
2225	goto out;
2226	if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2227	goto out;
2228	ret = load_global_roots_objectid(tree_root, path,
2229	BTRFS_FREE_SPACE_TREE_OBJECTID,
2230	name: "free space");
2231	out:
2232	btrfs_free_path(p: path);
2233	return ret;
2234	}
2235
2236	static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2237	{
2238	struct btrfs_root *tree_root = fs_info->tree_root;
2239	struct btrfs_root *root;
2240	struct btrfs_key location;
2241	int ret;
2242
2243	ASSERT(fs_info->tree_root);
2244
2245	ret = load_global_roots(tree_root);
2246	if (ret)
2247	return ret;
2248
2249	location.type = BTRFS_ROOT_ITEM_KEY;
2250	location.offset = 0;
2251
2252	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2253	location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2254	root = btrfs_read_tree_root(tree_root, key: &location);
2255	if (IS_ERR(ptr: root)) {
2256	if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2257	ret = PTR_ERR(ptr: root);
2258	goto out;
2259	}
2260	} else {
2261	set_bit(nr: BTRFS_ROOT_TRACK_DIRTY, addr: &root->state);
2262	fs_info->block_group_root = root;
2263	}
2264	}
2265
2266	location.objectid = BTRFS_DEV_TREE_OBJECTID;
2267	root = btrfs_read_tree_root(tree_root, key: &location);
2268	if (IS_ERR(ptr: root)) {
2269	if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2270	ret = PTR_ERR(ptr: root);
2271	goto out;
2272	}
2273	} else {
2274	set_bit(nr: BTRFS_ROOT_TRACK_DIRTY, addr: &root->state);
2275	fs_info->dev_root = root;
2276	}
2277	/* Initialize fs_info for all devices in any case */
2278	ret = btrfs_init_devices_late(fs_info);
2279	if (ret)
2280	goto out;
2281
2282	/*
2283	* This tree can share blocks with some other fs tree during relocation
2284	* and we need a proper setup by btrfs_get_fs_root
2285	*/
2286	root = btrfs_get_fs_root(fs_info: tree_root->fs_info,
2287	BTRFS_DATA_RELOC_TREE_OBJECTID, check_ref: true);
2288	if (IS_ERR(ptr: root)) {
2289	if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2290	ret = PTR_ERR(ptr: root);
2291	goto out;
2292	}
2293	} else {
2294	set_bit(nr: BTRFS_ROOT_TRACK_DIRTY, addr: &root->state);
2295	fs_info->data_reloc_root = root;
2296	}
2297
2298	location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2299	root = btrfs_read_tree_root(tree_root, key: &location);
2300	if (!IS_ERR(ptr: root)) {
2301	set_bit(nr: BTRFS_ROOT_TRACK_DIRTY, addr: &root->state);
2302	fs_info->quota_root = root;
2303	}
2304
2305	location.objectid = BTRFS_UUID_TREE_OBJECTID;
2306	root = btrfs_read_tree_root(tree_root, key: &location);
2307	if (IS_ERR(ptr: root)) {
2308	if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2309	ret = PTR_ERR(ptr: root);
2310	if (ret != -ENOENT)
2311	goto out;
2312	}
2313	} else {
2314	set_bit(nr: BTRFS_ROOT_TRACK_DIRTY, addr: &root->state);
2315	fs_info->uuid_root = root;
2316	}
2317
2318	if (btrfs_fs_incompat(fs_info, RAID_STRIPE_TREE)) {
2319	location.objectid = BTRFS_RAID_STRIPE_TREE_OBJECTID;
2320	root = btrfs_read_tree_root(tree_root, key: &location);
2321	if (IS_ERR(ptr: root)) {
2322	if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2323	ret = PTR_ERR(ptr: root);
2324	goto out;
2325	}
2326	} else {
2327	set_bit(nr: BTRFS_ROOT_TRACK_DIRTY, addr: &root->state);
2328	fs_info->stripe_root = root;
2329	}
2330	}
2331
2332	return 0;
2333	out:
2334	btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2335	location.objectid, ret);
2336	return ret;
2337	}
2338
2339	/*
2340	* Real super block validation
2341	* NOTE: super csum type and incompat features will not be checked here.
2342	*
2343	* @sb: super block to check
2344	* @mirror_num: the super block number to check its bytenr:
2345	* 0 the primary (1st) sb
2346	* 1, 2 2nd and 3rd backup copy
2347	* -1 skip bytenr check
2348	*/
2349	int btrfs_validate_super(struct btrfs_fs_info *fs_info,
2350	struct btrfs_super_block *sb, int mirror_num)
2351	{
2352	u64 nodesize = btrfs_super_nodesize(s: sb);
2353	u64 sectorsize = btrfs_super_sectorsize(s: sb);
2354	int ret = 0;
2355
2356	if (btrfs_super_magic(s: sb) != BTRFS_MAGIC) {
2357	btrfs_err(fs_info, "no valid FS found");
2358	ret = -EINVAL;
2359	}
2360	if (btrfs_super_flags(s: sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2361	btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2362	btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2363	ret = -EINVAL;
2364	}
2365	if (btrfs_super_root_level(s: sb) >= BTRFS_MAX_LEVEL) {
2366	btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2367	btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2368	ret = -EINVAL;
2369	}
2370	if (btrfs_super_chunk_root_level(s: sb) >= BTRFS_MAX_LEVEL) {
2371	btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2372	btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2373	ret = -EINVAL;
2374	}
2375	if (btrfs_super_log_root_level(s: sb) >= BTRFS_MAX_LEVEL) {
2376	btrfs_err(fs_info, "log_root level too big: %d >= %d",
2377	btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2378	ret = -EINVAL;
2379	}
2380
2381	/*
2382	* Check sectorsize and nodesize first, other check will need it.
2383	* Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2384	*/
2385	if (!is_power_of_2(n: sectorsize) || sectorsize < 4096 ||
2386	sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2387	btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2388	ret = -EINVAL;
2389	}
2390
2391	/*
2392	* We only support at most two sectorsizes: 4K and PAGE_SIZE.
2393	*
2394	* We can support 16K sectorsize with 64K page size without problem,
2395	* but such sectorsize/pagesize combination doesn't make much sense.
2396	* 4K will be our future standard, PAGE_SIZE is supported from the very
2397	* beginning.
2398	*/
2399	if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2400	btrfs_err(fs_info,
2401	"sectorsize %llu not yet supported for page size %lu",
2402	sectorsize, PAGE_SIZE);
2403	ret = -EINVAL;
2404	}
2405
2406	if (!is_power_of_2(n: nodesize) || nodesize < sectorsize ||
2407	nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2408	btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2409	ret = -EINVAL;
2410	}
2411	if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2412	btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2413	le32_to_cpu(sb->__unused_leafsize), nodesize);
2414	ret = -EINVAL;
2415	}
2416
2417	/* Root alignment check */
2418	if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2419	btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2420	btrfs_super_root(sb));
2421	ret = -EINVAL;
2422	}
2423	if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2424	btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2425	btrfs_super_chunk_root(sb));
2426	ret = -EINVAL;
2427	}
2428	if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2429	btrfs_warn(fs_info, "log_root block unaligned: %llu",
2430	btrfs_super_log_root(sb));
2431	ret = -EINVAL;
2432	}
2433
2434	if (!fs_info->fs_devices->temp_fsid &&
2435	memcmp(p: fs_info->fs_devices->fsid, q: sb->fsid, BTRFS_FSID_SIZE) != 0) {
2436	btrfs_err(fs_info,
2437	"superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2438	sb->fsid, fs_info->fs_devices->fsid);
2439	ret = -EINVAL;
2440	}
2441
2442	if (memcmp(p: fs_info->fs_devices->metadata_uuid, q: btrfs_sb_fsid_ptr(sb),
2443	BTRFS_FSID_SIZE) != 0) {
2444	btrfs_err(fs_info,
2445	"superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2446	btrfs_sb_fsid_ptr(sb), fs_info->fs_devices->metadata_uuid);
2447	ret = -EINVAL;
2448	}
2449
2450	if (memcmp(p: fs_info->fs_devices->metadata_uuid, q: sb->dev_item.fsid,
2451	BTRFS_FSID_SIZE) != 0) {
2452	btrfs_err(fs_info,
2453	"dev_item UUID does not match metadata fsid: %pU != %pU",
2454	fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2455	ret = -EINVAL;
2456	}
2457
2458	/*
2459	* Artificial requirement for block-group-tree to force newer features
2460	* (free-space-tree, no-holes) so the test matrix is smaller.
2461	*/
2462	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2463	(!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2464	!btrfs_fs_incompat(fs_info, NO_HOLES))) {
2465	btrfs_err(fs_info,
2466	"block-group-tree feature requires fres-space-tree and no-holes");
2467	ret = -EINVAL;
2468	}
2469
2470	/*
2471	* Hint to catch really bogus numbers, bitflips or so, more exact checks are
2472	* done later
2473	*/
2474	if (btrfs_super_bytes_used(s: sb) < 6 * btrfs_super_nodesize(s: sb)) {
2475	btrfs_err(fs_info, "bytes_used is too small %llu",
2476	btrfs_super_bytes_used(sb));
2477	ret = -EINVAL;
2478	}
2479	if (!is_power_of_2(n: btrfs_super_stripesize(s: sb))) {
2480	btrfs_err(fs_info, "invalid stripesize %u",
2481	btrfs_super_stripesize(sb));
2482	ret = -EINVAL;
2483	}
2484	if (btrfs_super_num_devices(s: sb) > (1UL << 31))
2485	btrfs_warn(fs_info, "suspicious number of devices: %llu",
2486	btrfs_super_num_devices(sb));
2487	if (btrfs_super_num_devices(s: sb) == 0) {
2488	btrfs_err(fs_info, "number of devices is 0");
2489	ret = -EINVAL;
2490	}
2491
2492	if (mirror_num >= 0 &&
2493	btrfs_super_bytenr(s: sb) != btrfs_sb_offset(mirror: mirror_num)) {
2494	btrfs_err(fs_info, "super offset mismatch %llu != %u",
2495	btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2496	ret = -EINVAL;
2497	}
2498
2499	/*
2500	* Obvious sys_chunk_array corruptions, it must hold at least one key
2501	* and one chunk
2502	*/
2503	if (btrfs_super_sys_array_size(s: sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2504	btrfs_err(fs_info, "system chunk array too big %u > %u",
2505	btrfs_super_sys_array_size(sb),
2506	BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2507	ret = -EINVAL;
2508	}
2509	if (btrfs_super_sys_array_size(s: sb) < sizeof(struct btrfs_disk_key)
2510	+ sizeof(struct btrfs_chunk)) {
2511	btrfs_err(fs_info, "system chunk array too small %u < %zu",
2512	btrfs_super_sys_array_size(sb),
2513	sizeof(struct btrfs_disk_key)
2514	+ sizeof(struct btrfs_chunk));
2515	ret = -EINVAL;
2516	}
2517
2518	/*
2519	* The generation is a global counter, we'll trust it more than the others
2520	* but it's still possible that it's the one that's wrong.
2521	*/
2522	if (btrfs_super_generation(s: sb) < btrfs_super_chunk_root_generation(s: sb))
2523	btrfs_warn(fs_info,
2524	"suspicious: generation < chunk_root_generation: %llu < %llu",
2525	btrfs_super_generation(sb),
2526	btrfs_super_chunk_root_generation(sb));
2527	if (btrfs_super_generation(s: sb) < btrfs_super_cache_generation(s: sb)
2528	&& btrfs_super_cache_generation(s: sb) != (u64)-1)
2529	btrfs_warn(fs_info,
2530	"suspicious: generation < cache_generation: %llu < %llu",
2531	btrfs_super_generation(sb),
2532	btrfs_super_cache_generation(sb));
2533
2534	return ret;
2535	}
2536
2537	/*
2538	* Validation of super block at mount time.
2539	* Some checks already done early at mount time, like csum type and incompat
2540	* flags will be skipped.
2541	*/
2542	static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2543	{
2544	return btrfs_validate_super(fs_info, sb: fs_info->super_copy, mirror_num: 0);
2545	}
2546
2547	/*
2548	* Validation of super block at write time.
2549	* Some checks like bytenr check will be skipped as their values will be
2550	* overwritten soon.
2551	* Extra checks like csum type and incompat flags will be done here.
2552	*/
2553	static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2554	struct btrfs_super_block *sb)
2555	{
2556	int ret;
2557
2558	ret = btrfs_validate_super(fs_info, sb, mirror_num: -1);
2559	if (ret < 0)
2560	goto out;
2561	if (!btrfs_supported_super_csum(csum_type: btrfs_super_csum_type(s: sb))) {
2562	ret = -EUCLEAN;
2563	btrfs_err(fs_info, "invalid csum type, has %u want %u",
2564	btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2565	goto out;
2566	}
2567	if (btrfs_super_incompat_flags(s: sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2568	ret = -EUCLEAN;
2569	btrfs_err(fs_info,
2570	"invalid incompat flags, has 0x%llx valid mask 0x%llx",
2571	btrfs_super_incompat_flags(sb),
2572	(unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2573	goto out;
2574	}
2575	out:
2576	if (ret < 0)
2577	btrfs_err(fs_info,
2578	"super block corruption detected before writing it to disk");
2579	return ret;
2580	}
2581
2582	static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2583	{
2584	struct btrfs_tree_parent_check check = {
2585	.level = level,
2586	.transid = gen,
2587	.owner_root = root->root_key.objectid
2588	};
2589	int ret = 0;
2590
2591	root->node = read_tree_block(fs_info: root->fs_info, bytenr, check: &check);
2592	if (IS_ERR(ptr: root->node)) {
2593	ret = PTR_ERR(ptr: root->node);
2594	root->node = NULL;
2595	return ret;
2596	}
2597	if (!extent_buffer_uptodate(eb: root->node)) {
2598	free_extent_buffer(eb: root->node);
2599	root->node = NULL;
2600	return -EIO;
2601	}
2602
2603	btrfs_set_root_node(item: &root->root_item, node: root->node);
2604	root->commit_root = btrfs_root_node(root);
2605	btrfs_set_root_refs(s: &root->root_item, val: 1);
2606	return ret;
2607	}
2608
2609	static int load_important_roots(struct btrfs_fs_info *fs_info)
2610	{
2611	struct btrfs_super_block *sb = fs_info->super_copy;
2612	u64 gen, bytenr;
2613	int level, ret;
2614
2615	bytenr = btrfs_super_root(s: sb);
2616	gen = btrfs_super_generation(s: sb);
2617	level = btrfs_super_root_level(s: sb);
2618	ret = load_super_root(root: fs_info->tree_root, bytenr, gen, level);
2619	if (ret) {
2620	btrfs_warn(fs_info, "couldn't read tree root");
2621	return ret;
2622	}
2623	return 0;
2624	}
2625
2626	static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2627	{
2628	int backup_index = find_newest_super_backup(info: fs_info);
2629	struct btrfs_super_block *sb = fs_info->super_copy;
2630	struct btrfs_root *tree_root = fs_info->tree_root;
2631	bool handle_error = false;
2632	int ret = 0;
2633	int i;
2634
2635	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2636	if (handle_error) {
2637	if (!IS_ERR(ptr: tree_root->node))
2638	free_extent_buffer(eb: tree_root->node);
2639	tree_root->node = NULL;
2640
2641	if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2642	break;
2643
2644	free_root_pointers(info: fs_info, free_chunk_root: 0);
2645
2646	/*
2647	* Don't use the log in recovery mode, it won't be
2648	* valid
2649	*/
2650	btrfs_set_super_log_root(s: sb, val: 0);
2651
2652	btrfs_warn(fs_info, "try to load backup roots slot %d", i);
2653	ret = read_backup_root(fs_info, priority: i);
2654	backup_index = ret;
2655	if (ret < 0)
2656	return ret;
2657	}
2658
2659	ret = load_important_roots(fs_info);
2660	if (ret) {
2661	handle_error = true;
2662	continue;
2663	}
2664
2665	/*
2666	* No need to hold btrfs_root::objectid_mutex since the fs
2667	* hasn't been fully initialised and we are the only user
2668	*/
2669	ret = btrfs_init_root_free_objectid(root: tree_root);
2670	if (ret < 0) {
2671	handle_error = true;
2672	continue;
2673	}
2674
2675	ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2676
2677	ret = btrfs_read_roots(fs_info);
2678	if (ret < 0) {
2679	handle_error = true;
2680	continue;
2681	}
2682
2683	/* All successful */
2684	fs_info->generation = btrfs_header_generation(eb: tree_root->node);
2685	btrfs_set_last_trans_committed(fs_info, gen: fs_info->generation);
2686	fs_info->last_reloc_trans = 0;
2687
2688	/* Always begin writing backup roots after the one being used */
2689	if (backup_index < 0) {
2690	fs_info->backup_root_index = 0;
2691	} else {
2692	fs_info->backup_root_index = backup_index + 1;
2693	fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2694	}
2695	break;
2696	}
2697
2698	return ret;
2699	}
2700
2701	void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2702	{
2703	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2704	INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2705	INIT_LIST_HEAD(list: &fs_info->trans_list);
2706	INIT_LIST_HEAD(list: &fs_info->dead_roots);
2707	INIT_LIST_HEAD(list: &fs_info->delayed_iputs);
2708	INIT_LIST_HEAD(list: &fs_info->delalloc_roots);
2709	INIT_LIST_HEAD(list: &fs_info->caching_block_groups);
2710	spin_lock_init(&fs_info->delalloc_root_lock);
2711	spin_lock_init(&fs_info->trans_lock);
2712	spin_lock_init(&fs_info->fs_roots_radix_lock);
2713	spin_lock_init(&fs_info->delayed_iput_lock);
2714	spin_lock_init(&fs_info->defrag_inodes_lock);
2715	spin_lock_init(&fs_info->super_lock);
2716	spin_lock_init(&fs_info->buffer_lock);
2717	spin_lock_init(&fs_info->unused_bgs_lock);
2718	spin_lock_init(&fs_info->treelog_bg_lock);
2719	spin_lock_init(&fs_info->zone_active_bgs_lock);
2720	spin_lock_init(&fs_info->relocation_bg_lock);
2721	rwlock_init(&fs_info->tree_mod_log_lock);
2722	rwlock_init(&fs_info->global_root_lock);
2723	mutex_init(&fs_info->unused_bg_unpin_mutex);
2724	mutex_init(&fs_info->reclaim_bgs_lock);
2725	mutex_init(&fs_info->reloc_mutex);
2726	mutex_init(&fs_info->delalloc_root_mutex);
2727	mutex_init(&fs_info->zoned_meta_io_lock);
2728	mutex_init(&fs_info->zoned_data_reloc_io_lock);
2729	seqlock_init(&fs_info->profiles_lock);
2730
2731	btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
2732	btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
2733	btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
2734	btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
2735	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_prep,
2736	BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2737	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
2738	BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2739	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
2740	BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2741	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
2742	BTRFS_LOCKDEP_TRANS_COMPLETED);
2743
2744	INIT_LIST_HEAD(list: &fs_info->dirty_cowonly_roots);
2745	INIT_LIST_HEAD(list: &fs_info->space_info);
2746	INIT_LIST_HEAD(list: &fs_info->tree_mod_seq_list);
2747	INIT_LIST_HEAD(list: &fs_info->unused_bgs);
2748	INIT_LIST_HEAD(list: &fs_info->reclaim_bgs);
2749	INIT_LIST_HEAD(list: &fs_info->zone_active_bgs);
2750	#ifdef CONFIG_BTRFS_DEBUG
2751	INIT_LIST_HEAD(list: &fs_info->allocated_roots);
2752	INIT_LIST_HEAD(list: &fs_info->allocated_ebs);
2753	spin_lock_init(&fs_info->eb_leak_lock);
2754	#endif
2755	fs_info->mapping_tree = RB_ROOT_CACHED;
2756	rwlock_init(&fs_info->mapping_tree_lock);
2757	btrfs_init_block_rsv(rsv: &fs_info->global_block_rsv,
2758	type: BTRFS_BLOCK_RSV_GLOBAL);
2759	btrfs_init_block_rsv(rsv: &fs_info->trans_block_rsv, type: BTRFS_BLOCK_RSV_TRANS);
2760	btrfs_init_block_rsv(rsv: &fs_info->chunk_block_rsv, type: BTRFS_BLOCK_RSV_CHUNK);
2761	btrfs_init_block_rsv(rsv: &fs_info->empty_block_rsv, type: BTRFS_BLOCK_RSV_EMPTY);
2762	btrfs_init_block_rsv(rsv: &fs_info->delayed_block_rsv,
2763	type: BTRFS_BLOCK_RSV_DELOPS);
2764	btrfs_init_block_rsv(rsv: &fs_info->delayed_refs_rsv,
2765	type: BTRFS_BLOCK_RSV_DELREFS);
2766
2767	atomic_set(v: &fs_info->async_delalloc_pages, i: 0);
2768	atomic_set(v: &fs_info->defrag_running, i: 0);
2769	atomic_set(v: &fs_info->nr_delayed_iputs, i: 0);
2770	atomic64_set(v: &fs_info->tree_mod_seq, i: 0);
2771	fs_info->global_root_tree = RB_ROOT;
2772	fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2773	fs_info->metadata_ratio = 0;
2774	fs_info->defrag_inodes = RB_ROOT;
2775	atomic64_set(v: &fs_info->free_chunk_space, i: 0);
2776	fs_info->tree_mod_log = RB_ROOT;
2777	fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2778	btrfs_init_ref_verify(fs_info);
2779
2780	fs_info->thread_pool_size = min_t(unsigned long,
2781	num_online_cpus() + 2, 8);
2782
2783	INIT_LIST_HEAD(list: &fs_info->ordered_roots);
2784	spin_lock_init(&fs_info->ordered_root_lock);
2785
2786	btrfs_init_scrub(fs_info);
2787	btrfs_init_balance(fs_info);
2788	btrfs_init_async_reclaim_work(fs_info);
2789
2790	rwlock_init(&fs_info->block_group_cache_lock);
2791	fs_info->block_group_cache_tree = RB_ROOT_CACHED;
2792
2793	extent_io_tree_init(fs_info, tree: &fs_info->excluded_extents,
2794	owner: IO_TREE_FS_EXCLUDED_EXTENTS);
2795
2796	mutex_init(&fs_info->ordered_operations_mutex);
2797	mutex_init(&fs_info->tree_log_mutex);
2798	mutex_init(&fs_info->chunk_mutex);
2799	mutex_init(&fs_info->transaction_kthread_mutex);
2800	mutex_init(&fs_info->cleaner_mutex);
2801	mutex_init(&fs_info->ro_block_group_mutex);
2802	init_rwsem(&fs_info->commit_root_sem);
2803	init_rwsem(&fs_info->cleanup_work_sem);
2804	init_rwsem(&fs_info->subvol_sem);
2805	sema_init(sem: &fs_info->uuid_tree_rescan_sem, val: 1);
2806
2807	btrfs_init_dev_replace_locks(fs_info);
2808	btrfs_init_qgroup(fs_info);
2809	btrfs_discard_init(fs_info);
2810
2811	btrfs_init_free_cluster(cluster: &fs_info->meta_alloc_cluster);
2812	btrfs_init_free_cluster(cluster: &fs_info->data_alloc_cluster);
2813
2814	init_waitqueue_head(&fs_info->transaction_throttle);
2815	init_waitqueue_head(&fs_info->transaction_wait);
2816	init_waitqueue_head(&fs_info->transaction_blocked_wait);
2817	init_waitqueue_head(&fs_info->async_submit_wait);
2818	init_waitqueue_head(&fs_info->delayed_iputs_wait);
2819
2820	/* Usable values until the real ones are cached from the superblock */
2821	fs_info->nodesize = 4096;
2822	fs_info->sectorsize = 4096;
2823	fs_info->sectorsize_bits = ilog2(4096);
2824	fs_info->stripesize = 4096;
2825
2826	/* Default compress algorithm when user does -o compress */
2827	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2828
2829	fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
2830
2831	spin_lock_init(&fs_info->swapfile_pins_lock);
2832	fs_info->swapfile_pins = RB_ROOT;
2833
2834	fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2835	INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2836	}
2837
2838	static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2839	{
2840	int ret;
2841
2842	fs_info->sb = sb;
2843	/* Temporary fixed values for block size until we read the superblock. */
2844	sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2845	sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2846
2847	ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2848	if (ret)
2849	return ret;
2850
2851	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2852	if (ret)
2853	return ret;
2854
2855	fs_info->dirty_metadata_batch = PAGE_SIZE *
2856	(1 + ilog2(nr_cpu_ids));
2857
2858	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2859	if (ret)
2860	return ret;
2861
2862	ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2863	GFP_KERNEL);
2864	if (ret)
2865	return ret;
2866
2867	fs_info->delayed_root = kmalloc(size: sizeof(struct btrfs_delayed_root),
2868	GFP_KERNEL);
2869	if (!fs_info->delayed_root)
2870	return -ENOMEM;
2871	btrfs_init_delayed_root(delayed_root: fs_info->delayed_root);
2872
2873	if (sb_rdonly(sb))
2874	set_bit(nr: BTRFS_FS_STATE_RO, addr: &fs_info->fs_state);
2875
2876	return btrfs_alloc_stripe_hash_table(info: fs_info);
2877	}
2878
2879	static int btrfs_uuid_rescan_kthread(void *data)
2880	{
2881	struct btrfs_fs_info *fs_info = data;
2882	int ret;
2883
2884	/*
2885	* 1st step is to iterate through the existing UUID tree and
2886	* to delete all entries that contain outdated data.
2887	* 2nd step is to add all missing entries to the UUID tree.
2888	*/
2889	ret = btrfs_uuid_tree_iterate(fs_info);
2890	if (ret < 0) {
2891	if (ret != -EINTR)
2892	btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2893	ret);
2894	up(sem: &fs_info->uuid_tree_rescan_sem);
2895	return ret;
2896	}
2897	return btrfs_uuid_scan_kthread(data);
2898	}
2899
2900	static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2901	{
2902	struct task_struct *task;
2903
2904	down(sem: &fs_info->uuid_tree_rescan_sem);
2905	task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2906	if (IS_ERR(ptr: task)) {
2907	/* fs_info->update_uuid_tree_gen remains 0 in all error case */
2908	btrfs_warn(fs_info, "failed to start uuid_rescan task");
2909	up(sem: &fs_info->uuid_tree_rescan_sem);
2910	return PTR_ERR(ptr: task);
2911	}
2912
2913	return 0;
2914	}
2915
2916	static int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2917	{
2918	u64 root_objectid = 0;
2919	struct btrfs_root *gang[8];
2920	int i = 0;
2921	int err = 0;
2922	unsigned int ret = 0;
2923
2924	while (1) {
2925	spin_lock(lock: &fs_info->fs_roots_radix_lock);
2926	ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2927	results: (void **)gang, first_index: root_objectid,
2928	ARRAY_SIZE(gang));
2929	if (!ret) {
2930	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
2931	break;
2932	}
2933	root_objectid = gang[ret - 1]->root_key.objectid + 1;
2934
2935	for (i = 0; i < ret; i++) {
2936	/* Avoid to grab roots in dead_roots. */
2937	if (btrfs_root_refs(s: &gang[i]->root_item) == 0) {
2938	gang[i] = NULL;
2939	continue;
2940	}
2941	/* Grab all the search result for later use. */
2942	gang[i] = btrfs_grab_root(root: gang[i]);
2943	}
2944	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
2945
2946	for (i = 0; i < ret; i++) {
2947	if (!gang[i])
2948	continue;
2949	root_objectid = gang[i]->root_key.objectid;
2950	err = btrfs_orphan_cleanup(root: gang[i]);
2951	if (err)
2952	goto out;
2953	btrfs_put_root(root: gang[i]);
2954	}
2955	root_objectid++;
2956	}
2957	out:
2958	/* Release the uncleaned roots due to error. */
2959	for (; i < ret; i++) {
2960	if (gang[i])
2961	btrfs_put_root(root: gang[i]);
2962	}
2963	return err;
2964	}
2965
2966	/*
2967	* Mounting logic specific to read-write file systems. Shared by open_ctree
2968	* and btrfs_remount when remounting from read-only to read-write.
2969	*/
2970	int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
2971	{
2972	int ret;
2973	const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
2974	bool rebuild_free_space_tree = false;
2975
2976	if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2977	btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2978	if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
2979	btrfs_warn(fs_info,
2980	"'clear_cache' option is ignored with extent tree v2");
2981	else
2982	rebuild_free_space_tree = true;
2983	} else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2984	!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2985	btrfs_warn(fs_info, "free space tree is invalid");
2986	rebuild_free_space_tree = true;
2987	}
2988
2989	if (rebuild_free_space_tree) {
2990	btrfs_info(fs_info, "rebuilding free space tree");
2991	ret = btrfs_rebuild_free_space_tree(fs_info);
2992	if (ret) {
2993	btrfs_warn(fs_info,
2994	"failed to rebuild free space tree: %d", ret);
2995	goto out;
2996	}
2997	}
2998
2999	if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3000	!btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
3001	btrfs_info(fs_info, "disabling free space tree");
3002	ret = btrfs_delete_free_space_tree(fs_info);
3003	if (ret) {
3004	btrfs_warn(fs_info,
3005	"failed to disable free space tree: %d", ret);
3006	goto out;
3007	}
3008	}
3009
3010	/*
3011	* btrfs_find_orphan_roots() is responsible for finding all the dead
3012	* roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3013	* them into the fs_info->fs_roots_radix tree. This must be done before
3014	* calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3015	* first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3016	* item before the root's tree is deleted - this means that if we unmount
3017	* or crash before the deletion completes, on the next mount we will not
3018	* delete what remains of the tree because the orphan item does not
3019	* exists anymore, which is what tells us we have a pending deletion.
3020	*/
3021	ret = btrfs_find_orphan_roots(fs_info);
3022	if (ret)
3023	goto out;
3024
3025	ret = btrfs_cleanup_fs_roots(fs_info);
3026	if (ret)
3027	goto out;
3028
3029	down_read(sem: &fs_info->cleanup_work_sem);
3030	if ((ret = btrfs_orphan_cleanup(root: fs_info->fs_root)) ||
3031	(ret = btrfs_orphan_cleanup(root: fs_info->tree_root))) {
3032	up_read(sem: &fs_info->cleanup_work_sem);
3033	goto out;
3034	}
3035	up_read(sem: &fs_info->cleanup_work_sem);
3036
3037	mutex_lock(&fs_info->cleaner_mutex);
3038	ret = btrfs_recover_relocation(fs_info);
3039	mutex_unlock(lock: &fs_info->cleaner_mutex);
3040	if (ret < 0) {
3041	btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3042	goto out;
3043	}
3044
3045	if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3046	!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3047	btrfs_info(fs_info, "creating free space tree");
3048	ret = btrfs_create_free_space_tree(fs_info);
3049	if (ret) {
3050	btrfs_warn(fs_info,
3051	"failed to create free space tree: %d", ret);
3052	goto out;
3053	}
3054	}
3055
3056	if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3057	ret = btrfs_set_free_space_cache_v1_active(fs_info, active: cache_opt);
3058	if (ret)
3059	goto out;
3060	}
3061
3062	ret = btrfs_resume_balance_async(fs_info);
3063	if (ret)
3064	goto out;
3065
3066	ret = btrfs_resume_dev_replace_async(fs_info);
3067	if (ret) {
3068	btrfs_warn(fs_info, "failed to resume dev_replace");
3069	goto out;
3070	}
3071
3072	btrfs_qgroup_rescan_resume(fs_info);
3073
3074	if (!fs_info->uuid_root) {
3075	btrfs_info(fs_info, "creating UUID tree");
3076	ret = btrfs_create_uuid_tree(fs_info);
3077	if (ret) {
3078	btrfs_warn(fs_info,
3079	"failed to create the UUID tree %d", ret);
3080	goto out;
3081	}
3082	}
3083
3084	out:
3085	return ret;
3086	}
3087
3088	/*
3089	* Do various sanity and dependency checks of different features.
3090	*
3091	* @is_rw_mount: If the mount is read-write.
3092	*
3093	* This is the place for less strict checks (like for subpage or artificial
3094	* feature dependencies).
3095	*
3096	* For strict checks or possible corruption detection, see
3097	* btrfs_validate_super().
3098	*
3099	* This should be called after btrfs_parse_options(), as some mount options
3100	* (space cache related) can modify on-disk format like free space tree and
3101	* screw up certain feature dependencies.
3102	*/
3103	int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3104	{
3105	struct btrfs_super_block *disk_super = fs_info->super_copy;
3106	u64 incompat = btrfs_super_incompat_flags(s: disk_super);
3107	const u64 compat_ro = btrfs_super_compat_ro_flags(s: disk_super);
3108	const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3109
3110	if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3111	btrfs_err(fs_info,
3112	"cannot mount because of unknown incompat features (0x%llx)",
3113	incompat);
3114	return -EINVAL;
3115	}
3116
3117	/* Runtime limitation for mixed block groups. */
3118	if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3119	(fs_info->sectorsize != fs_info->nodesize)) {
3120	btrfs_err(fs_info,
3121	"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3122	fs_info->nodesize, fs_info->sectorsize);
3123	return -EINVAL;
3124	}
3125
3126	/* Mixed backref is an always-enabled feature. */
3127	incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3128
3129	/* Set compression related flags just in case. */
3130	if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3131	incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3132	else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3133	incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3134
3135	/*
3136	* An ancient flag, which should really be marked deprecated.
3137	* Such runtime limitation doesn't really need a incompat flag.
3138	*/
3139	if (btrfs_super_nodesize(s: disk_super) > PAGE_SIZE)
3140	incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3141
3142	if (compat_ro_unsupp && is_rw_mount) {
3143	btrfs_err(fs_info,
3144	"cannot mount read-write because of unknown compat_ro features (0x%llx)",
3145	compat_ro);
3146	return -EINVAL;
3147	}
3148
3149	/*
3150	* We have unsupported RO compat features, although RO mounted, we
3151	* should not cause any metadata writes, including log replay.
3152	* Or we could screw up whatever the new feature requires.
3153	*/
3154	if (compat_ro_unsupp && btrfs_super_log_root(s: disk_super) &&
3155	!btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3156	btrfs_err(fs_info,
3157	"cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3158	compat_ro);
3159	return -EINVAL;
3160	}
3161
3162	/*
3163	* Artificial limitations for block group tree, to force
3164	* block-group-tree to rely on no-holes and free-space-tree.
3165	*/
3166	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3167	(!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3168	!btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3169	btrfs_err(fs_info,
3170	"block-group-tree feature requires no-holes and free-space-tree features");
3171	return -EINVAL;
3172	}
3173
3174	/*
3175	* Subpage runtime limitation on v1 cache.
3176	*
3177	* V1 space cache still has some hard codeed PAGE_SIZE usage, while
3178	* we're already defaulting to v2 cache, no need to bother v1 as it's
3179	* going to be deprecated anyway.
3180	*/
3181	if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3182	btrfs_warn(fs_info,
3183	"v1 space cache is not supported for page size %lu with sectorsize %u",
3184	PAGE_SIZE, fs_info->sectorsize);
3185	return -EINVAL;
3186	}
3187
3188	/* This can be called by remount, we need to protect the super block. */
3189	spin_lock(lock: &fs_info->super_lock);
3190	btrfs_set_super_incompat_flags(s: disk_super, val: incompat);
3191	spin_unlock(lock: &fs_info->super_lock);
3192
3193	return 0;
3194	}
3195
3196	int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3197	char *options)
3198	{
3199	u32 sectorsize;
3200	u32 nodesize;
3201	u32 stripesize;
3202	u64 generation;
3203	u16 csum_type;
3204	struct btrfs_super_block *disk_super;
3205	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3206	struct btrfs_root *tree_root;
3207	struct btrfs_root *chunk_root;
3208	int ret;
3209	int level;
3210
3211	ret = init_mount_fs_info(fs_info, sb);
3212	if (ret)
3213	goto fail;
3214
3215	/* These need to be init'ed before we start creating inodes and such. */
3216	tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3217	GFP_KERNEL);
3218	fs_info->tree_root = tree_root;
3219	chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3220	GFP_KERNEL);
3221	fs_info->chunk_root = chunk_root;
3222	if (!tree_root || !chunk_root) {
3223	ret = -ENOMEM;
3224	goto fail;
3225	}
3226
3227	ret = btrfs_init_btree_inode(sb);
3228	if (ret)
3229	goto fail;
3230
3231	invalidate_bdev(bdev: fs_devices->latest_dev->bdev);
3232
3233	/*
3234	* Read super block and check the signature bytes only
3235	*/
3236	disk_super = btrfs_read_dev_super(bdev: fs_devices->latest_dev->bdev);
3237	if (IS_ERR(ptr: disk_super)) {
3238	ret = PTR_ERR(ptr: disk_super);
3239	goto fail_alloc;
3240	}
3241
3242	btrfs_info(fs_info, "first mount of filesystem %pU", disk_super->fsid);
3243	/*
3244	* Verify the type first, if that or the checksum value are
3245	* corrupted, we'll find out
3246	*/
3247	csum_type = btrfs_super_csum_type(s: disk_super);
3248	if (!btrfs_supported_super_csum(csum_type)) {
3249	btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3250	csum_type);
3251	ret = -EINVAL;
3252	btrfs_release_disk_super(super: disk_super);
3253	goto fail_alloc;
3254	}
3255
3256	fs_info->csum_size = btrfs_super_csum_size(s: disk_super);
3257
3258	ret = btrfs_init_csum_hash(fs_info, csum_type);
3259	if (ret) {
3260	btrfs_release_disk_super(super: disk_super);
3261	goto fail_alloc;
3262	}
3263
3264	/*
3265	* We want to check superblock checksum, the type is stored inside.
3266	* Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3267	*/
3268	if (btrfs_check_super_csum(fs_info, disk_sb: disk_super)) {
3269	btrfs_err(fs_info, "superblock checksum mismatch");
3270	ret = -EINVAL;
3271	btrfs_release_disk_super(super: disk_super);
3272	goto fail_alloc;
3273	}
3274
3275	/*
3276	* super_copy is zeroed at allocation time and we never touch the
3277	* following bytes up to INFO_SIZE, the checksum is calculated from
3278	* the whole block of INFO_SIZE
3279	*/
3280	memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3281	btrfs_release_disk_super(super: disk_super);
3282
3283	disk_super = fs_info->super_copy;
3284
3285	memcpy(fs_info->super_for_commit, fs_info->super_copy,
3286	sizeof(*fs_info->super_for_commit));
3287
3288	ret = btrfs_validate_mount_super(fs_info);
3289	if (ret) {
3290	btrfs_err(fs_info, "superblock contains fatal errors");
3291	ret = -EINVAL;
3292	goto fail_alloc;
3293	}
3294
3295	if (!btrfs_super_root(s: disk_super)) {
3296	btrfs_err(fs_info, "invalid superblock tree root bytenr");
3297	ret = -EINVAL;
3298	goto fail_alloc;
3299	}
3300
3301	/* check FS state, whether FS is broken. */
3302	if (btrfs_super_flags(s: disk_super) & BTRFS_SUPER_FLAG_ERROR)
3303	WRITE_ONCE(fs_info->fs_error, -EUCLEAN);
3304
3305	/* Set up fs_info before parsing mount options */
3306	nodesize = btrfs_super_nodesize(s: disk_super);
3307	sectorsize = btrfs_super_sectorsize(s: disk_super);
3308	stripesize = sectorsize;
3309	fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3310	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3311
3312	fs_info->nodesize = nodesize;
3313	fs_info->sectorsize = sectorsize;
3314	fs_info->sectorsize_bits = ilog2(sectorsize);
3315	fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(info: fs_info) / fs_info->csum_size;
3316	fs_info->stripesize = stripesize;
3317
3318	/*
3319	* Handle the space caching options appropriately now that we have the
3320	* super block loaded and validated.
3321	*/
3322	btrfs_set_free_space_cache_settings(fs_info);
3323
3324	if (!btrfs_check_options(info: fs_info, mount_opt: &fs_info->mount_opt, flags: sb->s_flags)) {
3325	ret = -EINVAL;
3326	goto fail_alloc;
3327	}
3328
3329	ret = btrfs_check_features(fs_info, is_rw_mount: !sb_rdonly(sb));
3330	if (ret < 0)
3331	goto fail_alloc;
3332
3333	/*
3334	* At this point our mount options are validated, if we set ->max_inline
3335	* to something non-standard make sure we truncate it to sectorsize.
3336	*/
3337	fs_info->max_inline = min_t(u64, fs_info->max_inline, fs_info->sectorsize);
3338
3339	if (sectorsize < PAGE_SIZE) {
3340	struct btrfs_subpage_info *subpage_info;
3341
3342	btrfs_warn(fs_info,
3343	"read-write for sector size %u with page size %lu is experimental",
3344	sectorsize, PAGE_SIZE);
3345	subpage_info = kzalloc(size: sizeof(*subpage_info), GFP_KERNEL);
3346	if (!subpage_info) {
3347	ret = -ENOMEM;
3348	goto fail_alloc;
3349	}
3350	btrfs_init_subpage_info(subpage_info, sectorsize);
3351	fs_info->subpage_info = subpage_info;
3352	}
3353
3354	ret = btrfs_init_workqueues(fs_info);
3355	if (ret)
3356	goto fail_sb_buffer;
3357
3358	sb->s_bdi->ra_pages *= btrfs_super_num_devices(s: disk_super);
3359	sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3360
3361	/* Update the values for the current filesystem. */
3362	sb->s_blocksize = sectorsize;
3363	sb->s_blocksize_bits = blksize_bits(size: sectorsize);
3364	memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3365
3366	mutex_lock(&fs_info->chunk_mutex);
3367	ret = btrfs_read_sys_array(fs_info);
3368	mutex_unlock(lock: &fs_info->chunk_mutex);
3369	if (ret) {
3370	btrfs_err(fs_info, "failed to read the system array: %d", ret);
3371	goto fail_sb_buffer;
3372	}
3373
3374	generation = btrfs_super_chunk_root_generation(s: disk_super);
3375	level = btrfs_super_chunk_root_level(s: disk_super);
3376	ret = load_super_root(root: chunk_root, bytenr: btrfs_super_chunk_root(s: disk_super),
3377	gen: generation, level);
3378	if (ret) {
3379	btrfs_err(fs_info, "failed to read chunk root");
3380	goto fail_tree_roots;
3381	}
3382
3383	read_extent_buffer(eb: chunk_root->node, dst: fs_info->chunk_tree_uuid,
3384	offsetof(struct btrfs_header, chunk_tree_uuid),
3385	BTRFS_UUID_SIZE);
3386
3387	ret = btrfs_read_chunk_tree(fs_info);
3388	if (ret) {
3389	btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3390	goto fail_tree_roots;
3391	}
3392
3393	/*
3394	* At this point we know all the devices that make this filesystem,
3395	* including the seed devices but we don't know yet if the replace
3396	* target is required. So free devices that are not part of this
3397	* filesystem but skip the replace target device which is checked
3398	* below in btrfs_init_dev_replace().
3399	*/
3400	btrfs_free_extra_devids(fs_devices);
3401	if (!fs_devices->latest_dev->bdev) {
3402	btrfs_err(fs_info, "failed to read devices");
3403	ret = -EIO;
3404	goto fail_tree_roots;
3405	}
3406
3407	ret = init_tree_roots(fs_info);
3408	if (ret)
3409	goto fail_tree_roots;
3410
3411	/*
3412	* Get zone type information of zoned block devices. This will also
3413	* handle emulation of a zoned filesystem if a regular device has the
3414	* zoned incompat feature flag set.
3415	*/
3416	ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3417	if (ret) {
3418	btrfs_err(fs_info,
3419	"zoned: failed to read device zone info: %d", ret);
3420	goto fail_block_groups;
3421	}
3422
3423	/*
3424	* If we have a uuid root and we're not being told to rescan we need to
3425	* check the generation here so we can set the
3426	* BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3427	* transaction during a balance or the log replay without updating the
3428	* uuid generation, and then if we crash we would rescan the uuid tree,
3429	* even though it was perfectly fine.
3430	*/
3431	if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3432	fs_info->generation == btrfs_super_uuid_tree_generation(s: disk_super))
3433	set_bit(nr: BTRFS_FS_UPDATE_UUID_TREE_GEN, addr: &fs_info->flags);
3434
3435	ret = btrfs_verify_dev_extents(fs_info);
3436	if (ret) {
3437	btrfs_err(fs_info,
3438	"failed to verify dev extents against chunks: %d",
3439	ret);
3440	goto fail_block_groups;
3441	}
3442	ret = btrfs_recover_balance(fs_info);
3443	if (ret) {
3444	btrfs_err(fs_info, "failed to recover balance: %d", ret);
3445	goto fail_block_groups;
3446	}
3447
3448	ret = btrfs_init_dev_stats(fs_info);
3449	if (ret) {
3450	btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3451	goto fail_block_groups;
3452	}
3453
3454	ret = btrfs_init_dev_replace(fs_info);
3455	if (ret) {
3456	btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3457	goto fail_block_groups;
3458	}
3459
3460	ret = btrfs_check_zoned_mode(fs_info);
3461	if (ret) {
3462	btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3463	ret);
3464	goto fail_block_groups;
3465	}
3466
3467	ret = btrfs_sysfs_add_fsid(fs_devs: fs_devices);
3468	if (ret) {
3469	btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3470	ret);
3471	goto fail_block_groups;
3472	}
3473
3474	ret = btrfs_sysfs_add_mounted(fs_info);
3475	if (ret) {
3476	btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3477	goto fail_fsdev_sysfs;
3478	}
3479
3480	ret = btrfs_init_space_info(fs_info);
3481	if (ret) {
3482	btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3483	goto fail_sysfs;
3484	}
3485
3486	ret = btrfs_read_block_groups(info: fs_info);
3487	if (ret) {
3488	btrfs_err(fs_info, "failed to read block groups: %d", ret);
3489	goto fail_sysfs;
3490	}
3491
3492	btrfs_free_zone_cache(fs_info);
3493
3494	btrfs_check_active_zone_reservation(fs_info);
3495
3496	if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3497	!btrfs_check_rw_degradable(fs_info, NULL)) {
3498	btrfs_warn(fs_info,
3499	"writable mount is not allowed due to too many missing devices");
3500	ret = -EINVAL;
3501	goto fail_sysfs;
3502	}
3503
3504	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3505	"btrfs-cleaner");
3506	if (IS_ERR(ptr: fs_info->cleaner_kthread)) {
3507	ret = PTR_ERR(ptr: fs_info->cleaner_kthread);
3508	goto fail_sysfs;
3509	}
3510
3511	fs_info->transaction_kthread = kthread_run(transaction_kthread,
3512	tree_root,
3513	"btrfs-transaction");
3514	if (IS_ERR(ptr: fs_info->transaction_kthread)) {
3515	ret = PTR_ERR(ptr: fs_info->transaction_kthread);
3516	goto fail_cleaner;
3517	}
3518
3519	ret = btrfs_read_qgroup_config(fs_info);
3520	if (ret)
3521	goto fail_trans_kthread;
3522
3523	if (btrfs_build_ref_tree(fs_info))
3524	btrfs_err(fs_info, "couldn't build ref tree");
3525
3526	/* do not make disk changes in broken FS or nologreplay is given */
3527	if (btrfs_super_log_root(s: disk_super) != 0 &&
3528	!btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3529	btrfs_info(fs_info, "start tree-log replay");
3530	ret = btrfs_replay_log(fs_info, fs_devices);
3531	if (ret)
3532	goto fail_qgroup;
3533	}
3534
3535	fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, check_ref: true);
3536	if (IS_ERR(ptr: fs_info->fs_root)) {
3537	ret = PTR_ERR(ptr: fs_info->fs_root);
3538	btrfs_warn(fs_info, "failed to read fs tree: %d", ret);
3539	fs_info->fs_root = NULL;
3540	goto fail_qgroup;
3541	}
3542
3543	if (sb_rdonly(sb))
3544	return 0;
3545
3546	ret = btrfs_start_pre_rw_mount(fs_info);
3547	if (ret) {
3548	close_ctree(fs_info);
3549	return ret;
3550	}
3551	btrfs_discard_resume(fs_info);
3552
3553	if (fs_info->uuid_root &&
3554	(btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3555	fs_info->generation != btrfs_super_uuid_tree_generation(s: disk_super))) {
3556	btrfs_info(fs_info, "checking UUID tree");
3557	ret = btrfs_check_uuid_tree(fs_info);
3558	if (ret) {
3559	btrfs_warn(fs_info,
3560	"failed to check the UUID tree: %d", ret);
3561	close_ctree(fs_info);
3562	return ret;
3563	}
3564	}
3565
3566	set_bit(nr: BTRFS_FS_OPEN, addr: &fs_info->flags);
3567
3568	/* Kick the cleaner thread so it'll start deleting snapshots. */
3569	if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3570	wake_up_process(tsk: fs_info->cleaner_kthread);
3571
3572	return 0;
3573
3574	fail_qgroup:
3575	btrfs_free_qgroup_config(fs_info);
3576	fail_trans_kthread:
3577	kthread_stop(k: fs_info->transaction_kthread);
3578	btrfs_cleanup_transaction(fs_info);
3579	btrfs_free_fs_roots(fs_info);
3580	fail_cleaner:
3581	kthread_stop(k: fs_info->cleaner_kthread);
3582
3583	/*
3584	* make sure we're done with the btree inode before we stop our
3585	* kthreads
3586	*/
3587	filemap_write_and_wait(mapping: fs_info->btree_inode->i_mapping);
3588
3589	fail_sysfs:
3590	btrfs_sysfs_remove_mounted(fs_info);
3591
3592	fail_fsdev_sysfs:
3593	btrfs_sysfs_remove_fsid(fs_devs: fs_info->fs_devices);
3594
3595	fail_block_groups:
3596	btrfs_put_block_group_cache(info: fs_info);
3597
3598	fail_tree_roots:
3599	if (fs_info->data_reloc_root)
3600	btrfs_drop_and_free_fs_root(fs_info, root: fs_info->data_reloc_root);
3601	free_root_pointers(info: fs_info, free_chunk_root: true);
3602	invalidate_inode_pages2(mapping: fs_info->btree_inode->i_mapping);
3603
3604	fail_sb_buffer:
3605	btrfs_stop_all_workers(fs_info);
3606	btrfs_free_block_groups(info: fs_info);
3607	fail_alloc:
3608	btrfs_mapping_tree_free(fs_info);
3609
3610	iput(fs_info->btree_inode);
3611	fail:
3612	btrfs_close_devices(fs_devices: fs_info->fs_devices);
3613	ASSERT(ret < 0);
3614	return ret;
3615	}
3616	ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3617
3618	static void btrfs_end_super_write(struct bio *bio)
3619	{
3620	struct btrfs_device *device = bio->bi_private;
3621	struct bio_vec *bvec;
3622	struct bvec_iter_all iter_all;
3623	struct page *page;
3624
3625	bio_for_each_segment_all(bvec, bio, iter_all) {
3626	page = bvec->bv_page;
3627
3628	if (bio->bi_status) {
3629	btrfs_warn_rl_in_rcu(device->fs_info,
3630	"lost page write due to IO error on %s (%d)",
3631	btrfs_dev_name(device),
3632	blk_status_to_errno(bio->bi_status));
3633	ClearPageUptodate(page);
3634	SetPageError(page);
3635	btrfs_dev_stat_inc_and_print(dev: device,
3636	index: BTRFS_DEV_STAT_WRITE_ERRS);
3637	} else {
3638	SetPageUptodate(page);
3639	}
3640
3641	put_page(page);
3642	unlock_page(page);
3643	}
3644
3645	bio_put(bio);
3646	}
3647
3648	struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3649	int copy_num, bool drop_cache)
3650	{
3651	struct btrfs_super_block *super;
3652	struct page *page;
3653	u64 bytenr, bytenr_orig;
3654	struct address_space *mapping = bdev->bd_inode->i_mapping;
3655	int ret;
3656
3657	bytenr_orig = btrfs_sb_offset(mirror: copy_num);
3658	ret = btrfs_sb_log_location_bdev(bdev, mirror: copy_num, READ, bytenr_ret: &bytenr);
3659	if (ret == -ENOENT)
3660	return ERR_PTR(error: -EINVAL);
3661	else if (ret)
3662	return ERR_PTR(error: ret);
3663
3664	if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3665	return ERR_PTR(error: -EINVAL);
3666
3667	if (drop_cache) {
3668	/* This should only be called with the primary sb. */
3669	ASSERT(copy_num == 0);
3670
3671	/*
3672	* Drop the page of the primary superblock, so later read will
3673	* always read from the device.
3674	*/
3675	invalidate_inode_pages2_range(mapping,
3676	start: bytenr >> PAGE_SHIFT,
3677	end: (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3678	}
3679
3680	page = read_cache_page_gfp(mapping, index: bytenr >> PAGE_SHIFT, GFP_NOFS);
3681	if (IS_ERR(ptr: page))
3682	return ERR_CAST(ptr: page);
3683
3684	super = page_address(page);
3685	if (btrfs_super_magic(s: super) != BTRFS_MAGIC) {
3686	btrfs_release_disk_super(super);
3687	return ERR_PTR(error: -ENODATA);
3688	}
3689
3690	if (btrfs_super_bytenr(s: super) != bytenr_orig) {
3691	btrfs_release_disk_super(super);
3692	return ERR_PTR(error: -EINVAL);
3693	}
3694
3695	return super;
3696	}
3697
3698
3699	struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3700	{
3701	struct btrfs_super_block *super, *latest = NULL;
3702	int i;
3703	u64 transid = 0;
3704
3705	/* we would like to check all the supers, but that would make
3706	* a btrfs mount succeed after a mkfs from a different FS.
3707	* So, we need to add a special mount option to scan for
3708	* later supers, using BTRFS_SUPER_MIRROR_MAX instead
3709	*/
3710	for (i = 0; i < 1; i++) {
3711	super = btrfs_read_dev_one_super(bdev, copy_num: i, drop_cache: false);
3712	if (IS_ERR(ptr: super))
3713	continue;
3714
3715	if (!latest || btrfs_super_generation(s: super) > transid) {
3716	if (latest)
3717	btrfs_release_disk_super(super);
3718
3719	latest = super;
3720	transid = btrfs_super_generation(s: super);
3721	}
3722	}
3723
3724	return super;
3725	}
3726
3727	/*
3728	* Write superblock @sb to the @device. Do not wait for completion, all the
3729	* pages we use for writing are locked.
3730	*
3731	* Write @max_mirrors copies of the superblock, where 0 means default that fit
3732	* the expected device size at commit time. Note that max_mirrors must be
3733	* same for write and wait phases.
3734	*
3735	* Return number of errors when page is not found or submission fails.
3736	*/
3737	static int write_dev_supers(struct btrfs_device *device,
3738	struct btrfs_super_block *sb, int max_mirrors)
3739	{
3740	struct btrfs_fs_info *fs_info = device->fs_info;
3741	struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3742	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3743	int i;
3744	int errors = 0;
3745	int ret;
3746	u64 bytenr, bytenr_orig;
3747
3748	if (max_mirrors == 0)
3749	max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3750
3751	shash->tfm = fs_info->csum_shash;
3752
3753	for (i = 0; i < max_mirrors; i++) {
3754	struct page *page;
3755	struct bio *bio;
3756	struct btrfs_super_block *disk_super;
3757
3758	bytenr_orig = btrfs_sb_offset(mirror: i);
3759	ret = btrfs_sb_log_location(device, mirror: i, WRITE, bytenr_ret: &bytenr);
3760	if (ret == -ENOENT) {
3761	continue;
3762	} else if (ret < 0) {
3763	btrfs_err(device->fs_info,
3764	"couldn't get super block location for mirror %d",
3765	i);
3766	errors++;
3767	continue;
3768	}
3769	if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3770	device->commit_total_bytes)
3771	break;
3772
3773	btrfs_set_super_bytenr(s: sb, val: bytenr_orig);
3774
3775	crypto_shash_digest(desc: shash, data: (const char *)sb + BTRFS_CSUM_SIZE,
3776	BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3777	out: sb->csum);
3778
3779	page = find_or_create_page(mapping, index: bytenr >> PAGE_SHIFT,
3780	GFP_NOFS);
3781	if (!page) {
3782	btrfs_err(device->fs_info,
3783	"couldn't get super block page for bytenr %llu",
3784	bytenr);
3785	errors++;
3786	continue;
3787	}
3788
3789	/* Bump the refcount for wait_dev_supers() */
3790	get_page(page);
3791
3792	disk_super = page_address(page);
3793	memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3794
3795	/*
3796	* Directly use bios here instead of relying on the page cache
3797	* to do I/O, so we don't lose the ability to do integrity
3798	* checking.
3799	*/
3800	bio = bio_alloc(bdev: device->bdev, nr_vecs: 1,
3801	opf: REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
3802	GFP_NOFS);
3803	bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3804	bio->bi_private = device;
3805	bio->bi_end_io = btrfs_end_super_write;
3806	__bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3807	offset_in_page(bytenr));
3808
3809	/*
3810	* We FUA only the first super block. The others we allow to
3811	* go down lazy and there's a short window where the on-disk
3812	* copies might still contain the older version.
3813	*/
3814	if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3815	bio->bi_opf |= REQ_FUA;
3816	submit_bio(bio);
3817
3818	if (btrfs_advance_sb_log(device, mirror: i))
3819	errors++;
3820	}
3821	return errors < i ? 0 : -1;
3822	}
3823
3824	/*
3825	* Wait for write completion of superblocks done by write_dev_supers,
3826	* @max_mirrors same for write and wait phases.
3827	*
3828	* Return number of errors when page is not found or not marked up to
3829	* date.
3830	*/
3831	static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3832	{
3833	int i;
3834	int errors = 0;
3835	bool primary_failed = false;
3836	int ret;
3837	u64 bytenr;
3838
3839	if (max_mirrors == 0)
3840	max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3841
3842	for (i = 0; i < max_mirrors; i++) {
3843	struct page *page;
3844
3845	ret = btrfs_sb_log_location(device, mirror: i, READ, bytenr_ret: &bytenr);
3846	if (ret == -ENOENT) {
3847	break;
3848	} else if (ret < 0) {
3849	errors++;
3850	if (i == 0)
3851	primary_failed = true;
3852	continue;
3853	}
3854	if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3855	device->commit_total_bytes)
3856	break;
3857
3858	page = find_get_page(mapping: device->bdev->bd_inode->i_mapping,
3859	offset: bytenr >> PAGE_SHIFT);
3860	if (!page) {
3861	errors++;
3862	if (i == 0)
3863	primary_failed = true;
3864	continue;
3865	}
3866	/* Page is submitted locked and unlocked once the IO completes */
3867	wait_on_page_locked(page);
3868	if (PageError(page)) {
3869	errors++;
3870	if (i == 0)
3871	primary_failed = true;
3872	}
3873
3874	/* Drop our reference */
3875	put_page(page);
3876
3877	/* Drop the reference from the writing run */
3878	put_page(page);
3879	}
3880
3881	/* log error, force error return */
3882	if (primary_failed) {
3883	btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3884	device->devid);
3885	return -1;
3886	}
3887
3888	return errors < i ? 0 : -1;
3889	}
3890
3891	/*
3892	* endio for the write_dev_flush, this will wake anyone waiting
3893	* for the barrier when it is done
3894	*/
3895	static void btrfs_end_empty_barrier(struct bio *bio)
3896	{
3897	bio_uninit(bio);
3898	complete(bio->bi_private);
3899	}
3900
3901	/*
3902	* Submit a flush request to the device if it supports it. Error handling is
3903	* done in the waiting counterpart.
3904	*/
3905	static void write_dev_flush(struct btrfs_device *device)
3906	{
3907	struct bio *bio = &device->flush_bio;
3908
3909	device->last_flush_error = BLK_STS_OK;
3910
3911	bio_init(bio, bdev: device->bdev, NULL, max_vecs: 0,
3912	opf: REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
3913	bio->bi_end_io = btrfs_end_empty_barrier;
3914	init_completion(x: &device->flush_wait);
3915	bio->bi_private = &device->flush_wait;
3916	submit_bio(bio);
3917	set_bit(BTRFS_DEV_STATE_FLUSH_SENT, addr: &device->dev_state);
3918	}
3919
3920	/*
3921	* If the flush bio has been submitted by write_dev_flush, wait for it.
3922	* Return true for any error, and false otherwise.
3923	*/
3924	static bool wait_dev_flush(struct btrfs_device *device)
3925	{
3926	struct bio *bio = &device->flush_bio;
3927
3928	if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, addr: &device->dev_state))
3929	return false;
3930
3931	wait_for_completion_io(&device->flush_wait);
3932
3933	if (bio->bi_status) {
3934	device->last_flush_error = bio->bi_status;
3935	btrfs_dev_stat_inc_and_print(dev: device, index: BTRFS_DEV_STAT_FLUSH_ERRS);
3936	return true;
3937	}
3938
3939	return false;
3940	}
3941
3942	/*
3943	* send an empty flush down to each device in parallel,
3944	* then wait for them
3945	*/
3946	static int barrier_all_devices(struct btrfs_fs_info *info)
3947	{
3948	struct list_head *head;
3949	struct btrfs_device *dev;
3950	int errors_wait = 0;
3951
3952	lockdep_assert_held(&info->fs_devices->device_list_mutex);
3953	/* send down all the barriers */
3954	head = &info->fs_devices->devices;
3955	list_for_each_entry(dev, head, dev_list) {
3956	if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3957	continue;
3958	if (!dev->bdev)
3959	continue;
3960	if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3961	!test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3962	continue;
3963
3964	write_dev_flush(device: dev);
3965	}
3966
3967	/* wait for all the barriers */
3968	list_for_each_entry(dev, head, dev_list) {
3969	if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3970	continue;
3971	if (!dev->bdev) {
3972	errors_wait++;
3973	continue;
3974	}
3975	if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3976	!test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3977	continue;
3978
3979	if (wait_dev_flush(device: dev))
3980	errors_wait++;
3981	}
3982
3983	/*
3984	* Checks last_flush_error of disks in order to determine the device
3985	* state.
3986	*/
3987	if (errors_wait && !btrfs_check_rw_degradable(fs_info: info, NULL))
3988	return -EIO;
3989
3990	return 0;
3991	}
3992
3993	int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3994	{
3995	int raid_type;
3996	int min_tolerated = INT_MAX;
3997
3998	if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3999	(flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4000	min_tolerated = min_t(int, min_tolerated,
4001	btrfs_raid_array[BTRFS_RAID_SINGLE].
4002	tolerated_failures);
4003
4004	for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4005	if (raid_type == BTRFS_RAID_SINGLE)
4006	continue;
4007	if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4008	continue;
4009	min_tolerated = min_t(int, min_tolerated,
4010	btrfs_raid_array[raid_type].
4011	tolerated_failures);
4012	}
4013
4014	if (min_tolerated == INT_MAX) {
4015	pr_warn("BTRFS: unknown raid flag: %llu", flags);
4016	min_tolerated = 0;
4017	}
4018
4019	return min_tolerated;
4020	}
4021
4022	int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4023	{
4024	struct list_head *head;
4025	struct btrfs_device *dev;
4026	struct btrfs_super_block *sb;
4027	struct btrfs_dev_item *dev_item;
4028	int ret;
4029	int do_barriers;
4030	int max_errors;
4031	int total_errors = 0;
4032	u64 flags;
4033
4034	do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4035
4036	/*
4037	* max_mirrors == 0 indicates we're from commit_transaction,
4038	* not from fsync where the tree roots in fs_info have not
4039	* been consistent on disk.
4040	*/
4041	if (max_mirrors == 0)
4042	backup_super_roots(info: fs_info);
4043
4044	sb = fs_info->super_for_commit;
4045	dev_item = &sb->dev_item;
4046
4047	mutex_lock(&fs_info->fs_devices->device_list_mutex);
4048	head = &fs_info->fs_devices->devices;
4049	max_errors = btrfs_super_num_devices(s: fs_info->super_copy) - 1;
4050
4051	if (do_barriers) {
4052	ret = barrier_all_devices(info: fs_info);
4053	if (ret) {
4054	mutex_unlock(
4055	lock: &fs_info->fs_devices->device_list_mutex);
4056	btrfs_handle_fs_error(fs_info, ret,
4057	"errors while submitting device barriers.");
4058	return ret;
4059	}
4060	}
4061
4062	list_for_each_entry(dev, head, dev_list) {
4063	if (!dev->bdev) {
4064	total_errors++;
4065	continue;
4066	}
4067	if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4068	!test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4069	continue;
4070
4071	btrfs_set_stack_device_generation(s: dev_item, val: 0);
4072	btrfs_set_stack_device_type(s: dev_item, val: dev->type);
4073	btrfs_set_stack_device_id(s: dev_item, val: dev->devid);
4074	btrfs_set_stack_device_total_bytes(s: dev_item,
4075	val: dev->commit_total_bytes);
4076	btrfs_set_stack_device_bytes_used(s: dev_item,
4077	val: dev->commit_bytes_used);
4078	btrfs_set_stack_device_io_align(s: dev_item, val: dev->io_align);
4079	btrfs_set_stack_device_io_width(s: dev_item, val: dev->io_width);
4080	btrfs_set_stack_device_sector_size(s: dev_item, val: dev->sector_size);
4081	memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4082	memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4083	BTRFS_FSID_SIZE);
4084
4085	flags = btrfs_super_flags(s: sb);
4086	btrfs_set_super_flags(s: sb, val: flags | BTRFS_HEADER_FLAG_WRITTEN);
4087
4088	ret = btrfs_validate_write_super(fs_info, sb);
4089	if (ret < 0) {
4090	mutex_unlock(lock: &fs_info->fs_devices->device_list_mutex);
4091	btrfs_handle_fs_error(fs_info, -EUCLEAN,
4092	"unexpected superblock corruption detected");
4093	return -EUCLEAN;
4094	}
4095
4096	ret = write_dev_supers(device: dev, sb, max_mirrors);
4097	if (ret)
4098	total_errors++;
4099	}
4100	if (total_errors > max_errors) {
4101	btrfs_err(fs_info, "%d errors while writing supers",
4102	total_errors);
4103	mutex_unlock(lock: &fs_info->fs_devices->device_list_mutex);
4104
4105	/* FUA is masked off if unsupported and can't be the reason */
4106	btrfs_handle_fs_error(fs_info, -EIO,
4107	"%d errors while writing supers",
4108	total_errors);
4109	return -EIO;
4110	}
4111
4112	total_errors = 0;
4113	list_for_each_entry(dev, head, dev_list) {
4114	if (!dev->bdev)
4115	continue;
4116	if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4117	!test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4118	continue;
4119
4120	ret = wait_dev_supers(device: dev, max_mirrors);
4121	if (ret)
4122	total_errors++;
4123	}
4124	mutex_unlock(lock: &fs_info->fs_devices->device_list_mutex);
4125	if (total_errors > max_errors) {
4126	btrfs_handle_fs_error(fs_info, -EIO,
4127	"%d errors while writing supers",
4128	total_errors);
4129	return -EIO;
4130	}
4131	return 0;
4132	}
4133
4134	/* Drop a fs root from the radix tree and free it. */
4135	void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4136	struct btrfs_root *root)
4137	{
4138	bool drop_ref = false;
4139
4140	spin_lock(lock: &fs_info->fs_roots_radix_lock);
4141	radix_tree_delete(&fs_info->fs_roots_radix,
4142	(unsigned long)root->root_key.objectid);
4143	if (test_and_clear_bit(nr: BTRFS_ROOT_IN_RADIX, addr: &root->state))
4144	drop_ref = true;
4145	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
4146
4147	if (BTRFS_FS_ERROR(fs_info)) {
4148	ASSERT(root->log_root == NULL);
4149	if (root->reloc_root) {
4150	btrfs_put_root(root: root->reloc_root);
4151	root->reloc_root = NULL;
4152	}
4153	}
4154
4155	if (drop_ref)
4156	btrfs_put_root(root);
4157	}
4158
4159	int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4160	{
4161	struct btrfs_root *root = fs_info->tree_root;
4162	struct btrfs_trans_handle *trans;
4163
4164	mutex_lock(&fs_info->cleaner_mutex);
4165	btrfs_run_delayed_iputs(fs_info);
4166	mutex_unlock(lock: &fs_info->cleaner_mutex);
4167	wake_up_process(tsk: fs_info->cleaner_kthread);
4168
4169	/* wait until ongoing cleanup work done */
4170	down_write(sem: &fs_info->cleanup_work_sem);
4171	up_write(sem: &fs_info->cleanup_work_sem);
4172
4173	trans = btrfs_join_transaction(root);
4174	if (IS_ERR(ptr: trans))
4175	return PTR_ERR(ptr: trans);
4176	return btrfs_commit_transaction(trans);
4177	}
4178
4179	static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4180	{
4181	struct btrfs_transaction *trans;
4182	struct btrfs_transaction *tmp;
4183	bool found = false;
4184
4185	if (list_empty(head: &fs_info->trans_list))
4186	return;
4187
4188	/*
4189	* This function is only called at the very end of close_ctree(),
4190	* thus no other running transaction, no need to take trans_lock.
4191	*/
4192	ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4193	list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4194	struct extent_state *cached = NULL;
4195	u64 dirty_bytes = 0;
4196	u64 cur = 0;
4197	u64 found_start;
4198	u64 found_end;
4199
4200	found = true;
4201	while (find_first_extent_bit(tree: &trans->dirty_pages, start: cur,
4202	start_ret: &found_start, end_ret: &found_end, bits: EXTENT_DIRTY, cached_state: &cached)) {
4203	dirty_bytes += found_end + 1 - found_start;
4204	cur = found_end + 1;
4205	}
4206	btrfs_warn(fs_info,
4207	"transaction %llu (with %llu dirty metadata bytes) is not committed",
4208	trans->transid, dirty_bytes);
4209	btrfs_cleanup_one_transaction(trans, fs_info);
4210
4211	if (trans == fs_info->running_transaction)
4212	fs_info->running_transaction = NULL;
4213	list_del_init(entry: &trans->list);
4214
4215	btrfs_put_transaction(transaction: trans);
4216	trace_btrfs_transaction_commit(fs_info);
4217	}
4218	ASSERT(!found);
4219	}
4220
4221	void __cold close_ctree(struct btrfs_fs_info *fs_info)
4222	{
4223	int ret;
4224
4225	set_bit(nr: BTRFS_FS_CLOSING_START, addr: &fs_info->flags);
4226
4227	/*
4228	* If we had UNFINISHED_DROPS we could still be processing them, so
4229	* clear that bit and wake up relocation so it can stop.
4230	* We must do this before stopping the block group reclaim task, because
4231	* at btrfs_relocate_block_group() we wait for this bit, and after the
4232	* wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4233	* have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4234	* return 1.
4235	*/
4236	btrfs_wake_unfinished_drop(fs_info);
4237
4238	/*
4239	* We may have the reclaim task running and relocating a data block group,
4240	* in which case it may create delayed iputs. So stop it before we park
4241	* the cleaner kthread otherwise we can get new delayed iputs after
4242	* parking the cleaner, and that can make the async reclaim task to hang
4243	* if it's waiting for delayed iputs to complete, since the cleaner is
4244	* parked and can not run delayed iputs - this will make us hang when
4245	* trying to stop the async reclaim task.
4246	*/
4247	cancel_work_sync(work: &fs_info->reclaim_bgs_work);
4248	/*
4249	* We don't want the cleaner to start new transactions, add more delayed
4250	* iputs, etc. while we're closing. We can't use kthread_stop() yet
4251	* because that frees the task_struct, and the transaction kthread might
4252	* still try to wake up the cleaner.
4253	*/
4254	kthread_park(k: fs_info->cleaner_kthread);
4255
4256	/* wait for the qgroup rescan worker to stop */
4257	btrfs_qgroup_wait_for_completion(fs_info, interruptible: false);
4258
4259	/* wait for the uuid_scan task to finish */
4260	down(sem: &fs_info->uuid_tree_rescan_sem);
4261	/* avoid complains from lockdep et al., set sem back to initial state */
4262	up(sem: &fs_info->uuid_tree_rescan_sem);
4263
4264	/* pause restriper - we want to resume on mount */
4265	btrfs_pause_balance(fs_info);
4266
4267	btrfs_dev_replace_suspend_for_unmount(fs_info);
4268
4269	btrfs_scrub_cancel(info: fs_info);
4270
4271	/* wait for any defraggers to finish */
4272	wait_event(fs_info->transaction_wait,
4273	(atomic_read(&fs_info->defrag_running) == 0));
4274
4275	/* clear out the rbtree of defraggable inodes */
4276	btrfs_cleanup_defrag_inodes(fs_info);
4277
4278	/*
4279	* After we parked the cleaner kthread, ordered extents may have
4280	* completed and created new delayed iputs. If one of the async reclaim
4281	* tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4282	* can hang forever trying to stop it, because if a delayed iput is
4283	* added after it ran btrfs_run_delayed_iputs() and before it called
4284	* btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4285	* no one else to run iputs.
4286	*
4287	* So wait for all ongoing ordered extents to complete and then run
4288	* delayed iputs. This works because once we reach this point no one
4289	* can either create new ordered extents nor create delayed iputs
4290	* through some other means.
4291	*
4292	* Also note that btrfs_wait_ordered_roots() is not safe here, because
4293	* it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4294	* but the delayed iput for the respective inode is made only when doing
4295	* the final btrfs_put_ordered_extent() (which must happen at
4296	* btrfs_finish_ordered_io() when we are unmounting).
4297	*/
4298	btrfs_flush_workqueue(wq: fs_info->endio_write_workers);
4299	/* Ordered extents for free space inodes. */
4300	btrfs_flush_workqueue(wq: fs_info->endio_freespace_worker);
4301	btrfs_run_delayed_iputs(fs_info);
4302
4303	cancel_work_sync(work: &fs_info->async_reclaim_work);
4304	cancel_work_sync(work: &fs_info->async_data_reclaim_work);
4305	cancel_work_sync(work: &fs_info->preempt_reclaim_work);
4306
4307	/* Cancel or finish ongoing discard work */
4308	btrfs_discard_cleanup(fs_info);
4309
4310	if (!sb_rdonly(sb: fs_info->sb)) {
4311	/*
4312	* The cleaner kthread is stopped, so do one final pass over
4313	* unused block groups.
4314	*/
4315	btrfs_delete_unused_bgs(fs_info);
4316
4317	/*
4318	* There might be existing delayed inode workers still running
4319	* and holding an empty delayed inode item. We must wait for
4320	* them to complete first because they can create a transaction.
4321	* This happens when someone calls btrfs_balance_delayed_items()
4322	* and then a transaction commit runs the same delayed nodes
4323	* before any delayed worker has done something with the nodes.
4324	* We must wait for any worker here and not at transaction
4325	* commit time since that could cause a deadlock.
4326	* This is a very rare case.
4327	*/
4328	btrfs_flush_workqueue(wq: fs_info->delayed_workers);
4329
4330	ret = btrfs_commit_super(fs_info);
4331	if (ret)
4332	btrfs_err(fs_info, "commit super ret %d", ret);
4333	}
4334
4335	if (BTRFS_FS_ERROR(fs_info))
4336	btrfs_error_commit_super(fs_info);
4337
4338	kthread_stop(k: fs_info->transaction_kthread);
4339	kthread_stop(k: fs_info->cleaner_kthread);
4340
4341	ASSERT(list_empty(&fs_info->delayed_iputs));
4342	set_bit(nr: BTRFS_FS_CLOSING_DONE, addr: &fs_info->flags);
4343
4344	if (btrfs_check_quota_leak(fs_info)) {
4345	WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4346	btrfs_err(fs_info, "qgroup reserved space leaked");
4347	}
4348
4349	btrfs_free_qgroup_config(fs_info);
4350	ASSERT(list_empty(&fs_info->delalloc_roots));
4351
4352	if (percpu_counter_sum(fbc: &fs_info->delalloc_bytes)) {
4353	btrfs_info(fs_info, "at unmount delalloc count %lld",
4354	percpu_counter_sum(&fs_info->delalloc_bytes));
4355	}
4356
4357	if (percpu_counter_sum(fbc: &fs_info->ordered_bytes))
4358	btrfs_info(fs_info, "at unmount dio bytes count %lld",
4359	percpu_counter_sum(&fs_info->ordered_bytes));
4360
4361	btrfs_sysfs_remove_mounted(fs_info);
4362	btrfs_sysfs_remove_fsid(fs_devs: fs_info->fs_devices);
4363
4364	btrfs_put_block_group_cache(info: fs_info);
4365
4366	/*
4367	* we must make sure there is not any read request to
4368	* submit after we stopping all workers.
4369	*/
4370	invalidate_inode_pages2(mapping: fs_info->btree_inode->i_mapping);
4371	btrfs_stop_all_workers(fs_info);
4372
4373	/* We shouldn't have any transaction open at this point */
4374	warn_about_uncommitted_trans(fs_info);
4375
4376	clear_bit(nr: BTRFS_FS_OPEN, addr: &fs_info->flags);
4377	free_root_pointers(info: fs_info, free_chunk_root: true);
4378	btrfs_free_fs_roots(fs_info);
4379
4380	/*
4381	* We must free the block groups after dropping the fs_roots as we could
4382	* have had an IO error and have left over tree log blocks that aren't
4383	* cleaned up until the fs roots are freed. This makes the block group
4384	* accounting appear to be wrong because there's pending reserved bytes,
4385	* so make sure we do the block group cleanup afterwards.
4386	*/
4387	btrfs_free_block_groups(info: fs_info);
4388
4389	iput(fs_info->btree_inode);
4390
4391	btrfs_mapping_tree_free(fs_info);
4392	btrfs_close_devices(fs_devices: fs_info->fs_devices);
4393	}
4394
4395	void btrfs_mark_buffer_dirty(struct btrfs_trans_handle *trans,
4396	struct extent_buffer *buf)
4397	{
4398	struct btrfs_fs_info *fs_info = buf->fs_info;
4399	u64 transid = btrfs_header_generation(eb: buf);
4400
4401	#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4402	/*
4403	* This is a fast path so only do this check if we have sanity tests
4404	* enabled. Normal people shouldn't be using unmapped buffers as dirty
4405	* outside of the sanity tests.
4406	*/
4407	if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4408	return;
4409	#endif
4410	/* This is an active transaction (its state < TRANS_STATE_UNBLOCKED). */
4411	ASSERT(trans->transid == fs_info->generation);
4412	btrfs_assert_tree_write_locked(eb: buf);
4413	if (unlikely(transid != fs_info->generation)) {
4414	btrfs_abort_transaction(trans, -EUCLEAN);
4415	btrfs_crit(fs_info,
4416	"dirty buffer transid mismatch, logical %llu found transid %llu running transid %llu",
4417	buf->start, transid, fs_info->generation);
4418	}
4419	set_extent_buffer_dirty(buf);
4420	}
4421
4422	static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4423	int flush_delayed)
4424	{
4425	/*
4426	* looks as though older kernels can get into trouble with
4427	* this code, they end up stuck in balance_dirty_pages forever
4428	*/
4429	int ret;
4430
4431	if (current->flags & PF_MEMALLOC)
4432	return;
4433
4434	if (flush_delayed)
4435	btrfs_balance_delayed_items(fs_info);
4436
4437	ret = __percpu_counter_compare(fbc: &fs_info->dirty_metadata_bytes,
4438	BTRFS_DIRTY_METADATA_THRESH,
4439	batch: fs_info->dirty_metadata_batch);
4440	if (ret > 0) {
4441	balance_dirty_pages_ratelimited(mapping: fs_info->btree_inode->i_mapping);
4442	}
4443	}
4444
4445	void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4446	{
4447	__btrfs_btree_balance_dirty(fs_info, flush_delayed: 1);
4448	}
4449
4450	void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4451	{
4452	__btrfs_btree_balance_dirty(fs_info, flush_delayed: 0);
4453	}
4454
4455	static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4456	{
4457	/* cleanup FS via transaction */
4458	btrfs_cleanup_transaction(fs_info);
4459
4460	mutex_lock(&fs_info->cleaner_mutex);
4461	btrfs_run_delayed_iputs(fs_info);
4462	mutex_unlock(lock: &fs_info->cleaner_mutex);
4463
4464	down_write(sem: &fs_info->cleanup_work_sem);
4465	up_write(sem: &fs_info->cleanup_work_sem);
4466	}
4467
4468	static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4469	{
4470	struct btrfs_root *gang[8];
4471	u64 root_objectid = 0;
4472	int ret;
4473
4474	spin_lock(lock: &fs_info->fs_roots_radix_lock);
4475	while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4476	results: (void **)gang, first_index: root_objectid,
4477	ARRAY_SIZE(gang))) != 0) {
4478	int i;
4479
4480	for (i = 0; i < ret; i++)
4481	gang[i] = btrfs_grab_root(root: gang[i]);
4482	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
4483
4484	for (i = 0; i < ret; i++) {
4485	if (!gang[i])
4486	continue;
4487	root_objectid = gang[i]->root_key.objectid;
4488	btrfs_free_log(NULL, root: gang[i]);
4489	btrfs_put_root(root: gang[i]);
4490	}
4491	root_objectid++;
4492	spin_lock(lock: &fs_info->fs_roots_radix_lock);
4493	}
4494	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
4495	btrfs_free_log_root_tree(NULL, fs_info);
4496	}
4497
4498	static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4499	{
4500	struct btrfs_ordered_extent *ordered;
4501
4502	spin_lock(lock: &root->ordered_extent_lock);
4503	/*
4504	* This will just short circuit the ordered completion stuff which will
4505	* make sure the ordered extent gets properly cleaned up.
4506	*/
4507	list_for_each_entry(ordered, &root->ordered_extents,
4508	root_extent_list)
4509	set_bit(nr: BTRFS_ORDERED_IOERR, addr: &ordered->flags);
4510	spin_unlock(lock: &root->ordered_extent_lock);
4511	}
4512
4513	static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4514	{
4515	struct btrfs_root *root;
4516	LIST_HEAD(splice);
4517
4518	spin_lock(lock: &fs_info->ordered_root_lock);
4519	list_splice_init(list: &fs_info->ordered_roots, head: &splice);
4520	while (!list_empty(head: &splice)) {
4521	root = list_first_entry(&splice, struct btrfs_root,
4522	ordered_root);
4523	list_move_tail(list: &root->ordered_root,
4524	head: &fs_info->ordered_roots);
4525
4526	spin_unlock(lock: &fs_info->ordered_root_lock);
4527	btrfs_destroy_ordered_extents(root);
4528
4529	cond_resched();
4530	spin_lock(lock: &fs_info->ordered_root_lock);
4531	}
4532	spin_unlock(lock: &fs_info->ordered_root_lock);
4533
4534	/*
4535	* We need this here because if we've been flipped read-only we won't
4536	* get sync() from the umount, so we need to make sure any ordered
4537	* extents that haven't had their dirty pages IO start writeout yet
4538	* actually get run and error out properly.
4539	*/
4540	btrfs_wait_ordered_roots(fs_info, U64_MAX, range_start: 0, range_len: (u64)-1);
4541	}
4542
4543	static void btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4544	struct btrfs_fs_info *fs_info)
4545	{
4546	struct rb_node *node;
4547	struct btrfs_delayed_ref_root *delayed_refs;
4548	struct btrfs_delayed_ref_node *ref;
4549
4550	delayed_refs = &trans->delayed_refs;
4551
4552	spin_lock(lock: &delayed_refs->lock);
4553	if (atomic_read(v: &delayed_refs->num_entries) == 0) {
4554	spin_unlock(lock: &delayed_refs->lock);
4555	btrfs_debug(fs_info, "delayed_refs has NO entry");
4556	return;
4557	}
4558
4559	while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4560	struct btrfs_delayed_ref_head *head;
4561	struct rb_node *n;
4562	bool pin_bytes = false;
4563
4564	head = rb_entry(node, struct btrfs_delayed_ref_head,
4565	href_node);
4566	if (btrfs_delayed_ref_lock(delayed_refs, head))
4567	continue;
4568
4569	spin_lock(lock: &head->lock);
4570	while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4571	ref = rb_entry(n, struct btrfs_delayed_ref_node,
4572	ref_node);
4573	rb_erase_cached(node: &ref->ref_node, root: &head->ref_tree);
4574	RB_CLEAR_NODE(&ref->ref_node);
4575	if (!list_empty(head: &ref->add_list))
4576	list_del(entry: &ref->add_list);
4577	atomic_dec(v: &delayed_refs->num_entries);
4578	btrfs_put_delayed_ref(ref);
4579	btrfs_delayed_refs_rsv_release(fs_info, nr_refs: 1, nr_csums: 0);
4580	}
4581	if (head->must_insert_reserved)
4582	pin_bytes = true;
4583	btrfs_free_delayed_extent_op(op: head->extent_op);
4584	btrfs_delete_ref_head(delayed_refs, head);
4585	spin_unlock(lock: &head->lock);
4586	spin_unlock(lock: &delayed_refs->lock);
4587	mutex_unlock(lock: &head->mutex);
4588
4589	if (pin_bytes) {
4590	struct btrfs_block_group *cache;
4591
4592	cache = btrfs_lookup_block_group(info: fs_info, bytenr: head->bytenr);
4593	BUG_ON(!cache);
4594
4595	spin_lock(lock: &cache->space_info->lock);
4596	spin_lock(lock: &cache->lock);
4597	cache->pinned += head->num_bytes;
4598	btrfs_space_info_update_bytes_pinned(fs_info,
4599	sinfo: cache->space_info, bytes: head->num_bytes);
4600	cache->reserved -= head->num_bytes;
4601	cache->space_info->bytes_reserved -= head->num_bytes;
4602	spin_unlock(lock: &cache->lock);
4603	spin_unlock(lock: &cache->space_info->lock);
4604
4605	btrfs_put_block_group(cache);
4606
4607	btrfs_error_unpin_extent_range(fs_info, start: head->bytenr,
4608	end: head->bytenr + head->num_bytes - 1);
4609	}
4610	btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4611	btrfs_put_delayed_ref_head(head);
4612	cond_resched();
4613	spin_lock(lock: &delayed_refs->lock);
4614	}
4615	btrfs_qgroup_destroy_extent_records(trans);
4616
4617	spin_unlock(lock: &delayed_refs->lock);
4618	}
4619
4620	static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4621	{
4622	struct btrfs_inode *btrfs_inode;
4623	LIST_HEAD(splice);
4624
4625	spin_lock(lock: &root->delalloc_lock);
4626	list_splice_init(list: &root->delalloc_inodes, head: &splice);
4627
4628	while (!list_empty(head: &splice)) {
4629	struct inode *inode = NULL;
4630	btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4631	delalloc_inodes);
4632	btrfs_del_delalloc_inode(inode: btrfs_inode);
4633	spin_unlock(lock: &root->delalloc_lock);
4634
4635	/*
4636	* Make sure we get a live inode and that it'll not disappear
4637	* meanwhile.
4638	*/
4639	inode = igrab(&btrfs_inode->vfs_inode);
4640	if (inode) {
4641	unsigned int nofs_flag;
4642
4643	nofs_flag = memalloc_nofs_save();
4644	invalidate_inode_pages2(mapping: inode->i_mapping);
4645	memalloc_nofs_restore(flags: nofs_flag);
4646	iput(inode);
4647	}
4648	spin_lock(lock: &root->delalloc_lock);
4649	}
4650	spin_unlock(lock: &root->delalloc_lock);
4651	}
4652
4653	static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4654	{
4655	struct btrfs_root *root;
4656	LIST_HEAD(splice);
4657
4658	spin_lock(lock: &fs_info->delalloc_root_lock);
4659	list_splice_init(list: &fs_info->delalloc_roots, head: &splice);
4660	while (!list_empty(head: &splice)) {
4661	root = list_first_entry(&splice, struct btrfs_root,
4662	delalloc_root);
4663	root = btrfs_grab_root(root);
4664	BUG_ON(!root);
4665	spin_unlock(lock: &fs_info->delalloc_root_lock);
4666
4667	btrfs_destroy_delalloc_inodes(root);
4668	btrfs_put_root(root);
4669
4670	spin_lock(lock: &fs_info->delalloc_root_lock);
4671	}
4672	spin_unlock(lock: &fs_info->delalloc_root_lock);
4673	}
4674
4675	static void btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4676	struct extent_io_tree *dirty_pages,
4677	int mark)
4678	{
4679	struct extent_buffer *eb;
4680	u64 start = 0;
4681	u64 end;
4682
4683	while (find_first_extent_bit(tree: dirty_pages, start, start_ret: &start, end_ret: &end,
4684	bits: mark, NULL)) {
4685	clear_extent_bits(tree: dirty_pages, start, end, bits: mark);
4686	while (start <= end) {
4687	eb = find_extent_buffer(fs_info, start);
4688	start += fs_info->nodesize;
4689	if (!eb)
4690	continue;
4691
4692	btrfs_tree_lock(eb);
4693	wait_on_extent_buffer_writeback(eb);
4694	btrfs_clear_buffer_dirty(NULL, buf: eb);
4695	btrfs_tree_unlock(eb);
4696
4697	free_extent_buffer_stale(eb);
4698	}
4699	}
4700	}
4701
4702	static void btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4703	struct extent_io_tree *unpin)
4704	{
4705	u64 start;
4706	u64 end;
4707
4708	while (1) {
4709	struct extent_state *cached_state = NULL;
4710
4711	/*
4712	* The btrfs_finish_extent_commit() may get the same range as
4713	* ours between find_first_extent_bit and clear_extent_dirty.
4714	* Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4715	* the same extent range.
4716	*/
4717	mutex_lock(&fs_info->unused_bg_unpin_mutex);
4718	if (!find_first_extent_bit(tree: unpin, start: 0, start_ret: &start, end_ret: &end,
4719	bits: EXTENT_DIRTY, cached_state: &cached_state)) {
4720	mutex_unlock(lock: &fs_info->unused_bg_unpin_mutex);
4721	break;
4722	}
4723
4724	clear_extent_dirty(tree: unpin, start, end, cached: &cached_state);
4725	free_extent_state(state: cached_state);
4726	btrfs_error_unpin_extent_range(fs_info, start, end);
4727	mutex_unlock(lock: &fs_info->unused_bg_unpin_mutex);
4728	cond_resched();
4729	}
4730	}
4731
4732	static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4733	{
4734	struct inode *inode;
4735
4736	inode = cache->io_ctl.inode;
4737	if (inode) {
4738	unsigned int nofs_flag;
4739
4740	nofs_flag = memalloc_nofs_save();
4741	invalidate_inode_pages2(mapping: inode->i_mapping);
4742	memalloc_nofs_restore(flags: nofs_flag);
4743
4744	BTRFS_I(inode)->generation = 0;
4745	cache->io_ctl.inode = NULL;
4746	iput(inode);
4747	}
4748	ASSERT(cache->io_ctl.pages == NULL);
4749	btrfs_put_block_group(cache);
4750	}
4751
4752	void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4753	struct btrfs_fs_info *fs_info)
4754	{
4755	struct btrfs_block_group *cache;
4756
4757	spin_lock(lock: &cur_trans->dirty_bgs_lock);
4758	while (!list_empty(head: &cur_trans->dirty_bgs)) {
4759	cache = list_first_entry(&cur_trans->dirty_bgs,
4760	struct btrfs_block_group,
4761	dirty_list);
4762
4763	if (!list_empty(head: &cache->io_list)) {
4764	spin_unlock(lock: &cur_trans->dirty_bgs_lock);
4765	list_del_init(entry: &cache->io_list);
4766	btrfs_cleanup_bg_io(cache);
4767	spin_lock(lock: &cur_trans->dirty_bgs_lock);
4768	}
4769
4770	list_del_init(entry: &cache->dirty_list);
4771	spin_lock(lock: &cache->lock);
4772	cache->disk_cache_state = BTRFS_DC_ERROR;
4773	spin_unlock(lock: &cache->lock);
4774
4775	spin_unlock(lock: &cur_trans->dirty_bgs_lock);
4776	btrfs_put_block_group(cache);
4777	btrfs_dec_delayed_refs_rsv_bg_updates(fs_info);
4778	spin_lock(lock: &cur_trans->dirty_bgs_lock);
4779	}
4780	spin_unlock(lock: &cur_trans->dirty_bgs_lock);
4781
4782	/*
4783	* Refer to the definition of io_bgs member for details why it's safe
4784	* to use it without any locking
4785	*/
4786	while (!list_empty(head: &cur_trans->io_bgs)) {
4787	cache = list_first_entry(&cur_trans->io_bgs,
4788	struct btrfs_block_group,
4789	io_list);
4790
4791	list_del_init(entry: &cache->io_list);
4792	spin_lock(lock: &cache->lock);
4793	cache->disk_cache_state = BTRFS_DC_ERROR;
4794	spin_unlock(lock: &cache->lock);
4795	btrfs_cleanup_bg_io(cache);
4796	}
4797	}
4798
4799	static void btrfs_free_all_qgroup_pertrans(struct btrfs_fs_info *fs_info)
4800	{
4801	struct btrfs_root *gang[8];
4802	int i;
4803	int ret;
4804
4805	spin_lock(lock: &fs_info->fs_roots_radix_lock);
4806	while (1) {
4807	ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
4808	results: (void **)gang, first_index: 0,
4809	ARRAY_SIZE(gang),
4810	BTRFS_ROOT_TRANS_TAG);
4811	if (ret == 0)
4812	break;
4813	for (i = 0; i < ret; i++) {
4814	struct btrfs_root *root = gang[i];
4815
4816	btrfs_qgroup_free_meta_all_pertrans(root);
4817	radix_tree_tag_clear(&fs_info->fs_roots_radix,
4818	index: (unsigned long)root->root_key.objectid,
4819	BTRFS_ROOT_TRANS_TAG);
4820	}
4821	}
4822	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
4823	}
4824
4825	void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4826	struct btrfs_fs_info *fs_info)
4827	{
4828	struct btrfs_device *dev, *tmp;
4829
4830	btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4831	ASSERT(list_empty(&cur_trans->dirty_bgs));
4832	ASSERT(list_empty(&cur_trans->io_bgs));
4833
4834	list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4835	post_commit_list) {
4836	list_del_init(entry: &dev->post_commit_list);
4837	}
4838
4839	btrfs_destroy_delayed_refs(trans: cur_trans, fs_info);
4840
4841	cur_trans->state = TRANS_STATE_COMMIT_START;
4842	wake_up(&fs_info->transaction_blocked_wait);
4843
4844	cur_trans->state = TRANS_STATE_UNBLOCKED;
4845	wake_up(&fs_info->transaction_wait);
4846
4847	btrfs_destroy_delayed_inodes(fs_info);
4848
4849	btrfs_destroy_marked_extents(fs_info, dirty_pages: &cur_trans->dirty_pages,
4850	mark: EXTENT_DIRTY);
4851	btrfs_destroy_pinned_extent(fs_info, unpin: &cur_trans->pinned_extents);
4852
4853	btrfs_free_all_qgroup_pertrans(fs_info);
4854
4855	cur_trans->state =TRANS_STATE_COMPLETED;
4856	wake_up(&cur_trans->commit_wait);
4857	}
4858
4859	static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4860	{
4861	struct btrfs_transaction *t;
4862
4863	mutex_lock(&fs_info->transaction_kthread_mutex);
4864
4865	spin_lock(lock: &fs_info->trans_lock);
4866	while (!list_empty(head: &fs_info->trans_list)) {
4867	t = list_first_entry(&fs_info->trans_list,
4868	struct btrfs_transaction, list);
4869	if (t->state >= TRANS_STATE_COMMIT_PREP) {
4870	refcount_inc(r: &t->use_count);
4871	spin_unlock(lock: &fs_info->trans_lock);
4872	btrfs_wait_for_commit(fs_info, transid: t->transid);
4873	btrfs_put_transaction(transaction: t);
4874	spin_lock(lock: &fs_info->trans_lock);
4875	continue;
4876	}
4877	if (t == fs_info->running_transaction) {
4878	t->state = TRANS_STATE_COMMIT_DOING;
4879	spin_unlock(lock: &fs_info->trans_lock);
4880	/*
4881	* We wait for 0 num_writers since we don't hold a trans
4882	* handle open currently for this transaction.
4883	*/
4884	wait_event(t->writer_wait,
4885	atomic_read(&t->num_writers) == 0);
4886	} else {
4887	spin_unlock(lock: &fs_info->trans_lock);
4888	}
4889	btrfs_cleanup_one_transaction(cur_trans: t, fs_info);
4890
4891	spin_lock(lock: &fs_info->trans_lock);
4892	if (t == fs_info->running_transaction)
4893	fs_info->running_transaction = NULL;
4894	list_del_init(entry: &t->list);
4895	spin_unlock(lock: &fs_info->trans_lock);
4896
4897	btrfs_put_transaction(transaction: t);
4898	trace_btrfs_transaction_commit(fs_info);
4899	spin_lock(lock: &fs_info->trans_lock);
4900	}
4901	spin_unlock(lock: &fs_info->trans_lock);
4902	btrfs_destroy_all_ordered_extents(fs_info);
4903	btrfs_destroy_delayed_inodes(fs_info);
4904	btrfs_assert_delayed_root_empty(fs_info);
4905	btrfs_destroy_all_delalloc_inodes(fs_info);
4906	btrfs_drop_all_logs(fs_info);
4907	mutex_unlock(lock: &fs_info->transaction_kthread_mutex);
4908
4909	return 0;
4910	}
4911
4912	int btrfs_init_root_free_objectid(struct btrfs_root *root)
4913	{
4914	struct btrfs_path *path;
4915	int ret;
4916	struct extent_buffer *l;
4917	struct btrfs_key search_key;
4918	struct btrfs_key found_key;
4919	int slot;
4920
4921	path = btrfs_alloc_path();
4922	if (!path)
4923	return -ENOMEM;
4924
4925	search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4926	search_key.type = -1;
4927	search_key.offset = (u64)-1;
4928	ret = btrfs_search_slot(NULL, root, key: &search_key, p: path, ins_len: 0, cow: 0);
4929	if (ret < 0)
4930	goto error;
4931	if (ret == 0) {
4932	/*
4933	* Key with offset -1 found, there would have to exist a root
4934	* with such id, but this is out of valid range.
4935	*/
4936	ret = -EUCLEAN;
4937	goto error;
4938	}
4939	if (path->slots[0] > 0) {
4940	slot = path->slots[0] - 1;
4941	l = path->nodes[0];
4942	btrfs_item_key_to_cpu(eb: l, cpu_key: &found_key, nr: slot);
4943	root->free_objectid = max_t(u64, found_key.objectid + 1,
4944	BTRFS_FIRST_FREE_OBJECTID);
4945	} else {
4946	root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4947	}
4948	ret = 0;
4949	error:
4950	btrfs_free_path(p: path);
4951	return ret;
4952	}
4953
4954	int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4955	{
4956	int ret;
4957	mutex_lock(&root->objectid_mutex);
4958
4959	if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4960	btrfs_warn(root->fs_info,
4961	"the objectid of root %llu reaches its highest value",
4962	root->root_key.objectid);
4963	ret = -ENOSPC;
4964	goto out;
4965	}
4966
4967	*objectid = root->free_objectid++;
4968	ret = 0;
4969	out:
4970	mutex_unlock(lock: &root->objectid_mutex);
4971	return ret;
4972	}
4973