disk-io.c

   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 2007 Oracle.  All rights reserved.
   4 */
   5
   6#include <linux/fs.h>
   7#include <linux/blkdev.h>
   8#include <linux/radix-tree.h>
   9#include <linux/writeback.h>
  10#include <linux/workqueue.h>
  11#include <linux/kthread.h>
  12#include <linux/slab.h>
  13#include <linux/migrate.h>
  14#include <linux/ratelimit.h>
  15#include <linux/uuid.h>
  16#include <linux/semaphore.h>
  17#include <linux/error-injection.h>
  18#include <linux/crc32c.h>
  19#include <linux/sched/mm.h>
  20#include <linux/unaligned.h>
  21#include <crypto/hash.h>
  22#include "ctree.h"
  23#include "disk-io.h"
  24#include "transaction.h"
  25#include "btrfs_inode.h"
  26#include "bio.h"
  27#include "print-tree.h"
  28#include "locking.h"
  29#include "tree-log.h"
  30#include "free-space-cache.h"
  31#include "free-space-tree.h"
  32#include "dev-replace.h"
  33#include "raid56.h"
  34#include "sysfs.h"
  35#include "qgroup.h"
  36#include "compression.h"
  37#include "tree-checker.h"
  38#include "ref-verify.h"
  39#include "block-group.h"
  40#include "discard.h"
  41#include "space-info.h"
  42#include "zoned.h"
  43#include "subpage.h"
  44#include "fs.h"
  45#include "accessors.h"
  46#include "extent-tree.h"
  47#include "root-tree.h"
  48#include "defrag.h"
  49#include "uuid-tree.h"
  50#include "relocation.h"
  51#include "scrub.h"
  52#include "super.h"
  53#include "delayed-inode.h"
  54
  55#define BTRFS_SUPER_FLAG_SUPP	(BTRFS_HEADER_FLAG_WRITTEN |\
  56				 BTRFS_HEADER_FLAG_RELOC |\
  57				 BTRFS_SUPER_FLAG_ERROR |\
  58				 BTRFS_SUPER_FLAG_SEEDING |\
  59				 BTRFS_SUPER_FLAG_METADUMP |\
  60				 BTRFS_SUPER_FLAG_METADUMP_V2)
  61
  62static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
  63static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
  64
  65static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
  66{
  67	if (fs_info->csum_shash)
  68		crypto_free_shash(fs_info->csum_shash);
  69}
  70
  71/*
  72 * Compute the csum of a btree block and store the result to provided buffer.
  73 */
  74static void csum_tree_block(struct extent_buffer *buf, u8 *result)
  75{
  76	struct btrfs_fs_info *fs_info = buf->fs_info;
  77	int num_pages;
  78	u32 first_page_part;
  79	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
  80	char *kaddr;
  81	int i;
  82
  83	shash->tfm = fs_info->csum_shash;
  84	crypto_shash_init(shash);
  85
  86	if (buf->addr) {
  87		/* Pages are contiguous, handle them as a big one. */
  88		kaddr = buf->addr;
  89		first_page_part = fs_info->nodesize;
  90		num_pages = 1;
  91	} else {
  92		kaddr = folio_address(buf->folios[0]);
  93		first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
  94		num_pages = num_extent_pages(buf);
  95	}
  96
  97	crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
  98			    first_page_part - BTRFS_CSUM_SIZE);
  99
 100	/*
 101	 * Multiple single-page folios case would reach here.
 102	 *
 103	 * nodesize <= PAGE_SIZE and large folio all handled by above
 104	 * crypto_shash_update() already.
 105	 */
 106	for (i = 1; i < num_pages && INLINE_EXTENT_BUFFER_PAGES > 1; i++) {
 107		kaddr = folio_address(buf->folios[i]);
 108		crypto_shash_update(shash, kaddr, PAGE_SIZE);
 109	}
 110	memset(result, 0, BTRFS_CSUM_SIZE);
 111	crypto_shash_final(shash, result);
 112}
 113
 114/*
 115 * we can't consider a given block up to date unless the transid of the
 116 * block matches the transid in the parent node's pointer.  This is how we
 117 * detect blocks that either didn't get written at all or got written
 118 * in the wrong place.
 119 */
 120int btrfs_buffer_uptodate(struct extent_buffer *eb, u64 parent_transid, bool atomic)
 121{
 122	if (!extent_buffer_uptodate(eb))
 123		return 0;
 124
 125	if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 126		return 1;
 127
 128	if (atomic)
 129		return -EAGAIN;
 130
 131	if (!extent_buffer_uptodate(eb) ||
 132	    btrfs_header_generation(eb) != parent_transid) {
 133		btrfs_err_rl(eb->fs_info,
 134"parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
 135			eb->start, eb->read_mirror,
 136			parent_transid, btrfs_header_generation(eb));
 137		clear_extent_buffer_uptodate(eb);
 138		return 0;
 139	}
 140	return 1;
 141}
 142
 143static bool btrfs_supported_super_csum(u16 csum_type)
 144{
 145	switch (csum_type) {
 146	case BTRFS_CSUM_TYPE_CRC32:
 147	case BTRFS_CSUM_TYPE_XXHASH:
 148	case BTRFS_CSUM_TYPE_SHA256:
 149	case BTRFS_CSUM_TYPE_BLAKE2:
 150		return true;
 151	default:
 152		return false;
 153	}
 154}
 155
 156/*
 157 * Return 0 if the superblock checksum type matches the checksum value of that
 158 * algorithm. Pass the raw disk superblock data.
 159 */
 160int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 161			   const struct btrfs_super_block *disk_sb)
 162{
 163	char result[BTRFS_CSUM_SIZE];
 164	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 165
 166	shash->tfm = fs_info->csum_shash;
 167
 168	/*
 169	 * The super_block structure does not span the whole
 170	 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
 171	 * filled with zeros and is included in the checksum.
 172	 */
 173	crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
 174			    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
 175
 176	if (memcmp(disk_sb->csum, result, fs_info->csum_size))
 177		return 1;
 178
 179	return 0;
 180}
 181
 182static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
 183				      int mirror_num)
 184{
 185	struct btrfs_fs_info *fs_info = eb->fs_info;
 186	const u32 step = min(fs_info->nodesize, PAGE_SIZE);
 187	const u32 nr_steps = eb->len / step;
 188	phys_addr_t paddrs[BTRFS_MAX_BLOCKSIZE / PAGE_SIZE];
 189	int ret = 0;
 190
 191	if (sb_rdonly(fs_info->sb))
 192		return -EROFS;
 193
 194	for (int i = 0; i < num_extent_pages(eb); i++) {
 195		struct folio *folio = eb->folios[i];
 196
 197		/* No large folio support yet. */
 198		ASSERT(folio_order(folio) == 0);
 199		ASSERT(i < nr_steps);
 200
 201		/*
 202		 * For nodesize < page size, there is just one paddr, with some
 203		 * offset inside the page.
 204		 *
 205		 * For nodesize >= page size, it's one or more paddrs, and eb->start
 206		 * must be aligned to page boundary.
 207		 */
 208		paddrs[i] = page_to_phys(&folio->page) + offset_in_page(eb->start);
 209	}
 210
 211	ret = btrfs_repair_io_failure(fs_info, 0, eb->start, eb->len, eb->start,
 212				      paddrs, step, mirror_num);
 213	return ret;
 214}
 215
 216/*
 217 * helper to read a given tree block, doing retries as required when
 218 * the checksums don't match and we have alternate mirrors to try.
 219 *
 220 * @check:		expected tree parentness check, see the comments of the
 221 *			structure for details.
 222 */
 223int btrfs_read_extent_buffer(struct extent_buffer *eb,
 224			     const struct btrfs_tree_parent_check *check)
 225{
 226	struct btrfs_fs_info *fs_info = eb->fs_info;
 227	int failed = 0;
 228	int ret;
 229	int num_copies = 0;
 230	int mirror_num = 0;
 231	int failed_mirror = 0;
 232
 233	ASSERT(check);
 234
 235	while (1) {
 236		ret = read_extent_buffer_pages(eb, mirror_num, check);
 237		if (!ret)
 238			break;
 239
 240		num_copies = btrfs_num_copies(fs_info,
 241					      eb->start, eb->len);
 242		if (num_copies == 1)
 243			break;
 244
 245		if (!failed_mirror) {
 246			failed = 1;
 247			failed_mirror = eb->read_mirror;
 248		}
 249
 250		mirror_num++;
 251		if (mirror_num == failed_mirror)
 252			mirror_num++;
 253
 254		if (mirror_num > num_copies)
 255			break;
 256	}
 257
 258	if (failed && !ret && failed_mirror)
 259		btrfs_repair_eb_io_failure(eb, failed_mirror);
 260
 261	return ret;
 262}
 263
 264/*
 265 * Checksum a dirty tree block before IO.
 266 */
 267int btree_csum_one_bio(struct btrfs_bio *bbio)
 268{
 269	struct extent_buffer *eb = bbio->private;
 270	struct btrfs_fs_info *fs_info = eb->fs_info;
 271	u64 found_start = btrfs_header_bytenr(eb);
 272	u64 last_trans;
 273	u8 result[BTRFS_CSUM_SIZE];
 274	int ret;
 275
 276	/* Btree blocks are always contiguous on disk. */
 277	if (WARN_ON_ONCE(bbio->file_offset != eb->start))
 278		return -EIO;
 279	if (WARN_ON_ONCE(bbio->bio.bi_iter.bi_size != eb->len))
 280		return -EIO;
 281
 282	/*
 283	 * If an extent_buffer is marked as EXTENT_BUFFER_ZONED_ZEROOUT, don't
 284	 * checksum it but zero-out its content. This is done to preserve
 285	 * ordering of I/O without unnecessarily writing out data.
 286	 */
 287	if (test_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags)) {
 288		memzero_extent_buffer(eb, 0, eb->len);
 289		return 0;
 290	}
 291
 292	if (WARN_ON_ONCE(found_start != eb->start))
 293		return -EIO;
 294	if (WARN_ON(!btrfs_meta_folio_test_uptodate(eb->folios[0], eb)))
 295		return -EIO;
 296
 297	ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
 298				    offsetof(struct btrfs_header, fsid),
 299				    BTRFS_FSID_SIZE) == 0);
 300	csum_tree_block(eb, result);
 301
 302	if (btrfs_header_level(eb))
 303		ret = btrfs_check_node(eb);
 304	else
 305		ret = btrfs_check_leaf(eb);
 306
 307	if (ret < 0)
 308		goto error;
 309
 310	/*
 311	 * Also check the generation, the eb reached here must be newer than
 312	 * last committed. Or something seriously wrong happened.
 313	 */
 314	last_trans = btrfs_get_last_trans_committed(fs_info);
 315	if (unlikely(btrfs_header_generation(eb) <= last_trans)) {
 316		ret = -EUCLEAN;
 317		btrfs_err(fs_info,
 318			"block=%llu bad generation, have %llu expect > %llu",
 319			  eb->start, btrfs_header_generation(eb), last_trans);
 320		goto error;
 321	}
 322	write_extent_buffer(eb, result, 0, fs_info->csum_size);
 323	return 0;
 324
 325error:
 326	btrfs_print_tree(eb, 0);
 327	btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
 328		  eb->start);
 329	/*
 330	 * Be noisy if this is an extent buffer from a log tree. We don't abort
 331	 * a transaction in case there's a bad log tree extent buffer, we just
 332	 * fallback to a transaction commit. Still we want to know when there is
 333	 * a bad log tree extent buffer, as that may signal a bug somewhere.
 334	 */
 335	WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
 336		btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
 337	return ret;
 338}
 339
 340static bool check_tree_block_fsid(struct extent_buffer *eb)
 341{
 342	struct btrfs_fs_info *fs_info = eb->fs_info;
 343	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
 344	u8 fsid[BTRFS_FSID_SIZE];
 345
 346	read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
 347			   BTRFS_FSID_SIZE);
 348
 349	/*
 350	 * alloc_fsid_devices() copies the fsid into fs_devices::metadata_uuid.
 351	 * This is then overwritten by metadata_uuid if it is present in the
 352	 * device_list_add(). The same true for a seed device as well. So use of
 353	 * fs_devices::metadata_uuid is appropriate here.
 354	 */
 355	if (memcmp(fsid, fs_info->fs_devices->metadata_uuid, BTRFS_FSID_SIZE) == 0)
 356		return false;
 357
 358	list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
 359		if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
 360			return false;
 361
 362	return true;
 363}
 364
 365/* Do basic extent buffer checks at read time */
 366int btrfs_validate_extent_buffer(struct extent_buffer *eb,
 367				 const struct btrfs_tree_parent_check *check)
 368{
 369	struct btrfs_fs_info *fs_info = eb->fs_info;
 370	u64 found_start;
 371	const u32 csum_size = fs_info->csum_size;
 372	u8 found_level;
 373	u8 result[BTRFS_CSUM_SIZE];
 374	const u8 *header_csum;
 375	int ret = 0;
 376	const bool ignore_csum = btrfs_test_opt(fs_info, IGNOREMETACSUMS);
 377
 378	ASSERT(check);
 379
 380	found_start = btrfs_header_bytenr(eb);
 381	if (unlikely(found_start != eb->start)) {
 382		btrfs_err_rl(fs_info,
 383			"bad tree block start, mirror %u want %llu have %llu",
 384			     eb->read_mirror, eb->start, found_start);
 385		ret = -EIO;
 386		goto out;
 387	}
 388	if (unlikely(check_tree_block_fsid(eb))) {
 389		btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
 390			     eb->start, eb->read_mirror);
 391		ret = -EIO;
 392		goto out;
 393	}
 394	found_level = btrfs_header_level(eb);
 395	if (unlikely(found_level >= BTRFS_MAX_LEVEL)) {
 396		btrfs_err(fs_info,
 397			"bad tree block level, mirror %u level %d on logical %llu",
 398			eb->read_mirror, btrfs_header_level(eb), eb->start);
 399		ret = -EIO;
 400		goto out;
 401	}
 402
 403	csum_tree_block(eb, result);
 404	header_csum = folio_address(eb->folios[0]) +
 405		get_eb_offset_in_folio(eb, offsetof(struct btrfs_header, csum));
 406
 407	if (memcmp(result, header_csum, csum_size) != 0) {
 408		btrfs_warn_rl(fs_info,
 409"checksum verify failed on logical %llu mirror %u wanted " BTRFS_CSUM_FMT " found " BTRFS_CSUM_FMT " level %d%s",
 410			      eb->start, eb->read_mirror,
 411			      BTRFS_CSUM_FMT_VALUE(csum_size, header_csum),
 412			      BTRFS_CSUM_FMT_VALUE(csum_size, result),
 413			      btrfs_header_level(eb),
 414			      ignore_csum ? ", ignored" : "");
 415		if (unlikely(!ignore_csum)) {
 416			ret = -EUCLEAN;
 417			goto out;
 418		}
 419	}
 420
 421	if (unlikely(found_level != check->level)) {
 422		btrfs_err(fs_info,
 423		"level verify failed on logical %llu mirror %u wanted %u found %u",
 424			  eb->start, eb->read_mirror, check->level, found_level);
 425		ret = -EIO;
 426		goto out;
 427	}
 428	if (unlikely(check->transid &&
 429		     btrfs_header_generation(eb) != check->transid)) {
 430		btrfs_err_rl(eb->fs_info,
 431"parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
 432				eb->start, eb->read_mirror, check->transid,
 433				btrfs_header_generation(eb));
 434		ret = -EIO;
 435		goto out;
 436	}
 437	if (check->has_first_key) {
 438		const struct btrfs_key *expect_key = &check->first_key;
 439		struct btrfs_key found_key;
 440
 441		if (found_level)
 442			btrfs_node_key_to_cpu(eb, &found_key, 0);
 443		else
 444			btrfs_item_key_to_cpu(eb, &found_key, 0);
 445		if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
 446			btrfs_err(fs_info,
 447"tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
 448				  eb->start, check->transid,
 449				  expect_key->objectid,
 450				  expect_key->type, expect_key->offset,
 451				  found_key.objectid, found_key.type,
 452				  found_key.offset);
 453			ret = -EUCLEAN;
 454			goto out;
 455		}
 456	}
 457	if (check->owner_root) {
 458		ret = btrfs_check_eb_owner(eb, check->owner_root);
 459		if (ret < 0)
 460			goto out;
 461	}
 462
 463	/* If this is a leaf block and it is corrupt, just return -EIO. */
 464	if (found_level == 0 && btrfs_check_leaf(eb))
 465		ret = -EIO;
 466
 467	if (found_level > 0 && btrfs_check_node(eb))
 468		ret = -EIO;
 469
 470	if (ret)
 471		btrfs_err(fs_info,
 472		"read time tree block corruption detected on logical %llu mirror %u",
 473			  eb->start, eb->read_mirror);
 474out:
 475	return ret;
 476}
 477
 478#ifdef CONFIG_MIGRATION
 479static int btree_migrate_folio(struct address_space *mapping,
 480		struct folio *dst, struct folio *src, enum migrate_mode mode)
 481{
 482	/*
 483	 * we can't safely write a btree page from here,
 484	 * we haven't done the locking hook
 485	 */
 486	if (folio_test_dirty(src))
 487		return -EAGAIN;
 488	/*
 489	 * Buffers may be managed in a filesystem specific way.
 490	 * We must have no buffers or drop them.
 491	 */
 492	if (folio_get_private(src) &&
 493	    !filemap_release_folio(src, GFP_KERNEL))
 494		return -EAGAIN;
 495	return migrate_folio(mapping, dst, src, mode);
 496}
 497#else
 498#define btree_migrate_folio NULL
 499#endif
 500
 501static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
 502{
 503	if (folio_test_writeback(folio) || folio_test_dirty(folio))
 504		return false;
 505
 506	return try_release_extent_buffer(folio);
 507}
 508
 509static void btree_invalidate_folio(struct folio *folio, size_t offset,
 510				 size_t length)
 511{
 512	struct extent_io_tree *tree;
 513
 514	tree = &folio_to_inode(folio)->io_tree;
 515	extent_invalidate_folio(tree, folio, offset);
 516	btree_release_folio(folio, GFP_NOFS);
 517	if (folio_get_private(folio)) {
 518		btrfs_warn(folio_to_fs_info(folio),
 519			   "folio private not zero on folio %llu",
 520			   (unsigned long long)folio_pos(folio));
 521		folio_detach_private(folio);
 522	}
 523}
 524
 525#ifdef DEBUG
 526static bool btree_dirty_folio(struct address_space *mapping,
 527		struct folio *folio)
 528{
 529	struct btrfs_fs_info *fs_info = inode_to_fs_info(mapping->host);
 530	struct btrfs_subpage_info *spi = fs_info->subpage_info;
 531	struct btrfs_subpage *subpage;
 532	struct extent_buffer *eb;
 533	int cur_bit = 0;
 534	u64 page_start = folio_pos(folio);
 535
 536	if (fs_info->sectorsize == PAGE_SIZE) {
 537		eb = folio_get_private(folio);
 538		BUG_ON(!eb);
 539		BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 540		BUG_ON(!atomic_read(&eb->refs));
 541		btrfs_assert_tree_write_locked(eb);
 542		return filemap_dirty_folio(mapping, folio);
 543	}
 544
 545	ASSERT(spi);
 546	subpage = folio_get_private(folio);
 547
 548	for (cur_bit = spi->dirty_offset;
 549	     cur_bit < spi->dirty_offset + spi->bitmap_nr_bits;
 550	     cur_bit++) {
 551		unsigned long flags;
 552		u64 cur;
 553
 554		spin_lock_irqsave(&subpage->lock, flags);
 555		if (!test_bit(cur_bit, subpage->bitmaps)) {
 556			spin_unlock_irqrestore(&subpage->lock, flags);
 557			continue;
 558		}
 559		spin_unlock_irqrestore(&subpage->lock, flags);
 560		cur = page_start + cur_bit * fs_info->sectorsize;
 561
 562		eb = find_extent_buffer(fs_info, cur);
 563		ASSERT(eb);
 564		ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 565		ASSERT(atomic_read(&eb->refs));
 566		btrfs_assert_tree_write_locked(eb);
 567		free_extent_buffer(eb);
 568
 569		cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits) - 1;
 570	}
 571	return filemap_dirty_folio(mapping, folio);
 572}
 573#else
 574#define btree_dirty_folio filemap_dirty_folio
 575#endif
 576
 577static const struct address_space_operations btree_aops = {
 578	.writepages	= btree_writepages,
 579	.release_folio	= btree_release_folio,
 580	.invalidate_folio = btree_invalidate_folio,
 581	.migrate_folio	= btree_migrate_folio,
 582	.dirty_folio	= btree_dirty_folio,
 583};
 584
 585struct extent_buffer *btrfs_find_create_tree_block(
 586						struct btrfs_fs_info *fs_info,
 587						u64 bytenr, u64 owner_root,
 588						int level)
 589{
 590	if (btrfs_is_testing(fs_info))
 591		return alloc_test_extent_buffer(fs_info, bytenr);
 592	return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
 593}
 594
 595/*
 596 * Read tree block at logical address @bytenr and do variant basic but critical
 597 * verification.
 598 *
 599 * @check:		expected tree parentness check, see comments of the
 600 *			structure for details.
 601 */
 602struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
 603				      struct btrfs_tree_parent_check *check)
 604{
 605	struct extent_buffer *buf = NULL;
 606	int ret;
 607
 608	ASSERT(check);
 609
 610	buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
 611					   check->level);
 612	if (IS_ERR(buf))
 613		return buf;
 614
 615	ret = btrfs_read_extent_buffer(buf, check);
 616	if (ret) {
 617		free_extent_buffer_stale(buf);
 618		return ERR_PTR(ret);
 619	}
 620	return buf;
 621
 622}
 623
 624static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
 625					   u64 objectid, gfp_t flags)
 626{
 627	struct btrfs_root *root;
 628
 629	root = kzalloc(sizeof(*root), flags);
 630	if (!root)
 631		return NULL;
 632
 633	root->fs_info = fs_info;
 634	root->root_key.objectid = objectid;
 635	RB_CLEAR_NODE(&root->rb_node);
 636
 637	xa_init(&root->inodes);
 638	xa_init(&root->delayed_nodes);
 639
 640	btrfs_init_root_block_rsv(root);
 641
 642	INIT_LIST_HEAD(&root->dirty_list);
 643	INIT_LIST_HEAD(&root->root_list);
 644	INIT_LIST_HEAD(&root->delalloc_inodes);
 645	INIT_LIST_HEAD(&root->delalloc_root);
 646	INIT_LIST_HEAD(&root->ordered_extents);
 647	INIT_LIST_HEAD(&root->ordered_root);
 648	INIT_LIST_HEAD(&root->reloc_dirty_list);
 649	spin_lock_init(&root->delalloc_lock);
 650	spin_lock_init(&root->ordered_extent_lock);
 651	spin_lock_init(&root->accounting_lock);
 652	spin_lock_init(&root->qgroup_meta_rsv_lock);
 653	mutex_init(&root->objectid_mutex);
 654	mutex_init(&root->log_mutex);
 655	mutex_init(&root->ordered_extent_mutex);
 656	mutex_init(&root->delalloc_mutex);
 657	init_waitqueue_head(&root->qgroup_flush_wait);
 658	init_waitqueue_head(&root->log_writer_wait);
 659	init_waitqueue_head(&root->log_commit_wait[0]);
 660	init_waitqueue_head(&root->log_commit_wait[1]);
 661	INIT_LIST_HEAD(&root->log_ctxs[0]);
 662	INIT_LIST_HEAD(&root->log_ctxs[1]);
 663	atomic_set(&root->log_commit[0], 0);
 664	atomic_set(&root->log_commit[1], 0);
 665	atomic_set(&root->log_writers, 0);
 666	atomic_set(&root->log_batch, 0);
 667	refcount_set(&root->refs, 1);
 668	atomic_set(&root->snapshot_force_cow, 0);
 669	atomic_set(&root->nr_swapfiles, 0);
 670	root->log_transid_committed = -1;
 671	if (!btrfs_is_testing(fs_info)) {
 672		btrfs_extent_io_tree_init(fs_info, &root->dirty_log_pages,
 673					  IO_TREE_ROOT_DIRTY_LOG_PAGES);
 674		btrfs_extent_io_tree_init(fs_info, &root->log_csum_range,
 675					  IO_TREE_LOG_CSUM_RANGE);
 676	}
 677
 678	spin_lock_init(&root->root_item_lock);
 679	btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
 680#ifdef CONFIG_BTRFS_DEBUG
 681	INIT_LIST_HEAD(&root->leak_list);
 682	spin_lock(&fs_info->fs_roots_radix_lock);
 683	list_add_tail(&root->leak_list, &fs_info->allocated_roots);
 684	spin_unlock(&fs_info->fs_roots_radix_lock);
 685#endif
 686
 687	return root;
 688}
 689
 690#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 691/* Should only be used by the testing infrastructure */
 692struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
 693{
 694	struct btrfs_root *root;
 695
 696	if (!fs_info)
 697		return ERR_PTR(-EINVAL);
 698
 699	root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
 700	if (!root)
 701		return ERR_PTR(-ENOMEM);
 702
 703	/* We don't use the stripesize in selftest, set it as sectorsize */
 704	root->alloc_bytenr = 0;
 705
 706	return root;
 707}
 708#endif
 709
 710static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
 711{
 712	const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
 713	const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
 714
 715	return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
 716}
 717
 718static int global_root_key_cmp(const void *k, const struct rb_node *node)
 719{
 720	const struct btrfs_key *key = k;
 721	const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
 722
 723	return btrfs_comp_cpu_keys(key, &root->root_key);
 724}
 725
 726int btrfs_global_root_insert(struct btrfs_root *root)
 727{
 728	struct btrfs_fs_info *fs_info = root->fs_info;
 729	struct rb_node *tmp;
 730	int ret = 0;
 731
 732	write_lock(&fs_info->global_root_lock);
 733	tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
 734	write_unlock(&fs_info->global_root_lock);
 735
 736	if (tmp) {
 737		ret = -EEXIST;
 738		btrfs_warn(fs_info, "global root %llu %llu already exists",
 739			   btrfs_root_id(root), root->root_key.offset);
 740	}
 741	return ret;
 742}
 743
 744void btrfs_global_root_delete(struct btrfs_root *root)
 745{
 746	struct btrfs_fs_info *fs_info = root->fs_info;
 747
 748	write_lock(&fs_info->global_root_lock);
 749	rb_erase(&root->rb_node, &fs_info->global_root_tree);
 750	write_unlock(&fs_info->global_root_lock);
 751}
 752
 753struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
 754				     struct btrfs_key *key)
 755{
 756	struct rb_node *node;
 757	struct btrfs_root *root = NULL;
 758
 759	read_lock(&fs_info->global_root_lock);
 760	node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
 761	if (node)
 762		root = container_of(node, struct btrfs_root, rb_node);
 763	read_unlock(&fs_info->global_root_lock);
 764
 765	return root;
 766}
 767
 768static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
 769{
 770	struct btrfs_block_group *block_group;
 771	u64 ret;
 772
 773	if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
 774		return 0;
 775
 776	if (bytenr)
 777		block_group = btrfs_lookup_block_group(fs_info, bytenr);
 778	else
 779		block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
 780	ASSERT(block_group);
 781	if (!block_group)
 782		return 0;
 783	ret = block_group->global_root_id;
 784	btrfs_put_block_group(block_group);
 785
 786	return ret;
 787}
 788
 789struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
 790{
 791	struct btrfs_key key = {
 792		.objectid = BTRFS_CSUM_TREE_OBJECTID,
 793		.type = BTRFS_ROOT_ITEM_KEY,
 794		.offset = btrfs_global_root_id(fs_info, bytenr),
 795	};
 796
 797	return btrfs_global_root(fs_info, &key);
 798}
 799
 800struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
 801{
 802	struct btrfs_key key = {
 803		.objectid = BTRFS_EXTENT_TREE_OBJECTID,
 804		.type = BTRFS_ROOT_ITEM_KEY,
 805		.offset = btrfs_global_root_id(fs_info, bytenr),
 806	};
 807
 808	return btrfs_global_root(fs_info, &key);
 809}
 810
 811struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
 812				     u64 objectid)
 813{
 814	struct btrfs_fs_info *fs_info = trans->fs_info;
 815	struct extent_buffer *leaf;
 816	struct btrfs_root *tree_root = fs_info->tree_root;
 817	struct btrfs_root *root;
 818	struct btrfs_key key;
 819	unsigned int nofs_flag;
 820	int ret = 0;
 821
 822	/*
 823	 * We're holding a transaction handle, so use a NOFS memory allocation
 824	 * context to avoid deadlock if reclaim happens.
 825	 */
 826	nofs_flag = memalloc_nofs_save();
 827	root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
 828	memalloc_nofs_restore(nofs_flag);
 829	if (!root)
 830		return ERR_PTR(-ENOMEM);
 831
 832	root->root_key.objectid = objectid;
 833	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
 834	root->root_key.offset = 0;
 835
 836	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
 837				      0, BTRFS_NESTING_NORMAL);
 838	if (IS_ERR(leaf)) {
 839		ret = PTR_ERR(leaf);
 840		leaf = NULL;
 841		goto fail;
 842	}
 843
 844	root->node = leaf;
 845	btrfs_mark_buffer_dirty(trans, leaf);
 846
 847	root->commit_root = btrfs_root_node(root);
 848	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
 849
 850	btrfs_set_root_flags(&root->root_item, 0);
 851	btrfs_set_root_limit(&root->root_item, 0);
 852	btrfs_set_root_bytenr(&root->root_item, leaf->start);
 853	btrfs_set_root_generation(&root->root_item, trans->transid);
 854	btrfs_set_root_level(&root->root_item, 0);
 855	btrfs_set_root_refs(&root->root_item, 1);
 856	btrfs_set_root_used(&root->root_item, leaf->len);
 857	btrfs_set_root_last_snapshot(&root->root_item, 0);
 858	btrfs_set_root_dirid(&root->root_item, 0);
 859	if (btrfs_is_fstree(objectid))
 860		generate_random_guid(root->root_item.uuid);
 861	else
 862		export_guid(root->root_item.uuid, &guid_null);
 863	btrfs_set_root_drop_level(&root->root_item, 0);
 864
 865	btrfs_tree_unlock(leaf);
 866
 867	key.objectid = objectid;
 868	key.type = BTRFS_ROOT_ITEM_KEY;
 869	key.offset = 0;
 870	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
 871	if (ret)
 872		goto fail;
 873
 874	return root;
 875
 876fail:
 877	btrfs_put_root(root);
 878
 879	return ERR_PTR(ret);
 880}
 881
 882static struct btrfs_root *alloc_log_tree(struct btrfs_fs_info *fs_info)
 883{
 884	struct btrfs_root *root;
 885
 886	root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
 887	if (!root)
 888		return ERR_PTR(-ENOMEM);
 889
 890	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
 891	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
 892	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
 893
 894	return root;
 895}
 896
 897int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
 898			      struct btrfs_root *root)
 899{
 900	struct extent_buffer *leaf;
 901
 902	/*
 903	 * DON'T set SHAREABLE bit for log trees.
 904	 *
 905	 * Log trees are not exposed to user space thus can't be snapshotted,
 906	 * and they go away before a real commit is actually done.
 907	 *
 908	 * They do store pointers to file data extents, and those reference
 909	 * counts still get updated (along with back refs to the log tree).
 910	 */
 911
 912	leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
 913			NULL, 0, 0, 0, 0, BTRFS_NESTING_NORMAL);
 914	if (IS_ERR(leaf))
 915		return PTR_ERR(leaf);
 916
 917	root->node = leaf;
 918
 919	btrfs_mark_buffer_dirty(trans, root->node);
 920	btrfs_tree_unlock(root->node);
 921
 922	return 0;
 923}
 924
 925int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
 926			     struct btrfs_fs_info *fs_info)
 927{
 928	struct btrfs_root *log_root;
 929
 930	log_root = alloc_log_tree(fs_info);
 931	if (IS_ERR(log_root))
 932		return PTR_ERR(log_root);
 933
 934	if (!btrfs_is_zoned(fs_info)) {
 935		int ret = btrfs_alloc_log_tree_node(trans, log_root);
 936
 937		if (ret) {
 938			btrfs_put_root(log_root);
 939			return ret;
 940		}
 941	}
 942
 943	WARN_ON(fs_info->log_root_tree);
 944	fs_info->log_root_tree = log_root;
 945	return 0;
 946}
 947
 948int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
 949		       struct btrfs_root *root)
 950{
 951	struct btrfs_fs_info *fs_info = root->fs_info;
 952	struct btrfs_root *log_root;
 953	struct btrfs_inode_item *inode_item;
 954	int ret;
 955
 956	log_root = alloc_log_tree(fs_info);
 957	if (IS_ERR(log_root))
 958		return PTR_ERR(log_root);
 959
 960	ret = btrfs_alloc_log_tree_node(trans, log_root);
 961	if (ret) {
 962		btrfs_put_root(log_root);
 963		return ret;
 964	}
 965
 966	btrfs_set_root_last_trans(log_root, trans->transid);
 967	log_root->root_key.offset = btrfs_root_id(root);
 968
 969	inode_item = &log_root->root_item.inode;
 970	btrfs_set_stack_inode_generation(inode_item, 1);
 971	btrfs_set_stack_inode_size(inode_item, 3);
 972	btrfs_set_stack_inode_nlink(inode_item, 1);
 973	btrfs_set_stack_inode_nbytes(inode_item,
 974				     fs_info->nodesize);
 975	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 976
 977	btrfs_set_root_node(&log_root->root_item, log_root->node);
 978
 979	WARN_ON(root->log_root);
 980	root->log_root = log_root;
 981	btrfs_set_root_log_transid(root, 0);
 982	root->log_transid_committed = -1;
 983	btrfs_set_root_last_log_commit(root, 0);
 984	return 0;
 985}
 986
 987static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
 988					      struct btrfs_path *path,
 989					      const struct btrfs_key *key)
 990{
 991	struct btrfs_root *root;
 992	struct btrfs_tree_parent_check check = { 0 };
 993	struct btrfs_fs_info *fs_info = tree_root->fs_info;
 994	u64 generation;
 995	int ret;
 996	int level;
 997
 998	root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
 999	if (!root)
1000		return ERR_PTR(-ENOMEM);
1001
1002	ret = btrfs_find_root(tree_root, key, path,
1003			      &root->root_item, &root->root_key);
1004	if (ret) {
1005		if (ret > 0)
1006			ret = -ENOENT;
1007		goto fail;
1008	}
1009
1010	generation = btrfs_root_generation(&root->root_item);
1011	level = btrfs_root_level(&root->root_item);
1012	check.level = level;
1013	check.transid = generation;
1014	check.owner_root = key->objectid;
1015	root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item),
1016				     &check);
1017	if (IS_ERR(root->node)) {
1018		ret = PTR_ERR(root->node);
1019		root->node = NULL;
1020		goto fail;
1021	}
1022	if (unlikely(!btrfs_buffer_uptodate(root->node, generation, false))) {
1023		ret = -EIO;
1024		goto fail;
1025	}
1026
1027	/*
1028	 * For real fs, and not log/reloc trees, root owner must
1029	 * match its root node owner
1030	 */
1031	if (unlikely(!btrfs_is_testing(fs_info) &&
1032		     btrfs_root_id(root) != BTRFS_TREE_LOG_OBJECTID &&
1033		     btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID &&
1034		     btrfs_root_id(root) != btrfs_header_owner(root->node))) {
1035		btrfs_crit(fs_info,
1036"root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1037			   btrfs_root_id(root), root->node->start,
1038			   btrfs_header_owner(root->node),
1039			   btrfs_root_id(root));
1040		ret = -EUCLEAN;
1041		goto fail;
1042	}
1043	root->commit_root = btrfs_root_node(root);
1044	return root;
1045fail:
1046	btrfs_put_root(root);
1047	return ERR_PTR(ret);
1048}
1049
1050struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1051					const struct btrfs_key *key)
1052{
1053	struct btrfs_root *root;
1054	BTRFS_PATH_AUTO_FREE(path);
1055
1056	path = btrfs_alloc_path();
1057	if (!path)
1058		return ERR_PTR(-ENOMEM);
1059	root = read_tree_root_path(tree_root, path, key);
1060
1061	return root;
1062}
1063
1064/*
1065 * Initialize subvolume root in-memory structure.
1066 *
1067 * @anon_dev:	anonymous device to attach to the root, if zero, allocate new
1068 *
1069 * In case of failure the caller is responsible to call btrfs_free_fs_root()
1070 */
1071static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1072{
1073	int ret;
1074
1075	btrfs_drew_lock_init(&root->snapshot_lock);
1076
1077	if (btrfs_root_id(root) != BTRFS_TREE_LOG_OBJECTID &&
1078	    !btrfs_is_data_reloc_root(root) &&
1079	    btrfs_is_fstree(btrfs_root_id(root))) {
1080		set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1081		btrfs_check_and_init_root_item(&root->root_item);
1082	}
1083
1084	/*
1085	 * Don't assign anonymous block device to roots that are not exposed to
1086	 * userspace, the id pool is limited to 1M
1087	 */
1088	if (btrfs_is_fstree(btrfs_root_id(root)) &&
1089	    btrfs_root_refs(&root->root_item) > 0) {
1090		if (!anon_dev) {
1091			ret = get_anon_bdev(&root->anon_dev);
1092			if (ret)
1093				return ret;
1094		} else {
1095			root->anon_dev = anon_dev;
1096		}
1097	}
1098
1099	mutex_lock(&root->objectid_mutex);
1100	ret = btrfs_init_root_free_objectid(root);
1101	if (ret) {
1102		mutex_unlock(&root->objectid_mutex);
1103		return ret;
1104	}
1105
1106	ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1107
1108	mutex_unlock(&root->objectid_mutex);
1109
1110	return 0;
1111}
1112
1113static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1114					       u64 root_id)
1115{
1116	struct btrfs_root *root;
1117
1118	spin_lock(&fs_info->fs_roots_radix_lock);
1119	root = radix_tree_lookup(&fs_info->fs_roots_radix,
1120				 (unsigned long)root_id);
1121	root = btrfs_grab_root(root);
1122	spin_unlock(&fs_info->fs_roots_radix_lock);
1123	return root;
1124}
1125
1126static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1127						u64 objectid)
1128{
1129	struct btrfs_key key = {
1130		.objectid = objectid,
1131		.type = BTRFS_ROOT_ITEM_KEY,
1132		.offset = 0,
1133	};
1134
1135	switch (objectid) {
1136	case BTRFS_ROOT_TREE_OBJECTID:
1137		return btrfs_grab_root(fs_info->tree_root);
1138	case BTRFS_EXTENT_TREE_OBJECTID:
1139		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1140	case BTRFS_CHUNK_TREE_OBJECTID:
1141		return btrfs_grab_root(fs_info->chunk_root);
1142	case BTRFS_DEV_TREE_OBJECTID:
1143		return btrfs_grab_root(fs_info->dev_root);
1144	case BTRFS_CSUM_TREE_OBJECTID:
1145		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1146	case BTRFS_QUOTA_TREE_OBJECTID:
1147		return btrfs_grab_root(fs_info->quota_root);
1148	case BTRFS_UUID_TREE_OBJECTID:
1149		return btrfs_grab_root(fs_info->uuid_root);
1150	case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
1151		return btrfs_grab_root(fs_info->block_group_root);
1152	case BTRFS_FREE_SPACE_TREE_OBJECTID:
1153		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1154	case BTRFS_RAID_STRIPE_TREE_OBJECTID:
1155		return btrfs_grab_root(fs_info->stripe_root);
1156	default:
1157		return NULL;
1158	}
1159}
1160
1161int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1162			 struct btrfs_root *root)
1163{
1164	int ret;
1165
1166	ret = radix_tree_preload(GFP_NOFS);
1167	if (ret)
1168		return ret;
1169
1170	spin_lock(&fs_info->fs_roots_radix_lock);
1171	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1172				(unsigned long)btrfs_root_id(root),
1173				root);
1174	if (ret == 0) {
1175		btrfs_grab_root(root);
1176		set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1177	}
1178	spin_unlock(&fs_info->fs_roots_radix_lock);
1179	radix_tree_preload_end();
1180
1181	return ret;
1182}
1183
1184void btrfs_check_leaked_roots(const struct btrfs_fs_info *fs_info)
1185{
1186#ifdef CONFIG_BTRFS_DEBUG
1187	struct btrfs_root *root;
1188
1189	while (!list_empty(&fs_info->allocated_roots)) {
1190		char buf[BTRFS_ROOT_NAME_BUF_LEN];
1191
1192		root = list_first_entry(&fs_info->allocated_roots,
1193					struct btrfs_root, leak_list);
1194		btrfs_err(fs_info, "leaked root %s refcount %d",
1195			  btrfs_root_name(&root->root_key, buf),
1196			  refcount_read(&root->refs));
1197		WARN_ON_ONCE(1);
1198		while (refcount_read(&root->refs) > 1)
1199			btrfs_put_root(root);
1200		btrfs_put_root(root);
1201	}
1202#endif
1203}
1204
1205static void free_global_roots(struct btrfs_fs_info *fs_info)
1206{
1207	struct btrfs_root *root;
1208	struct rb_node *node;
1209
1210	while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1211		root = rb_entry(node, struct btrfs_root, rb_node);
1212		rb_erase(&root->rb_node, &fs_info->global_root_tree);
1213		btrfs_put_root(root);
1214	}
1215}
1216
1217void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1218{
1219	struct percpu_counter *em_counter = &fs_info->evictable_extent_maps;
1220
1221	if (fs_info->fs_devices)
1222		btrfs_close_devices(fs_info->fs_devices);
1223	btrfs_free_compress_wsm(fs_info);
1224	percpu_counter_destroy(&fs_info->stats_read_blocks);
1225	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1226	percpu_counter_destroy(&fs_info->delalloc_bytes);
1227	percpu_counter_destroy(&fs_info->ordered_bytes);
1228	if (percpu_counter_initialized(em_counter))
1229		ASSERT(percpu_counter_sum_positive(em_counter) == 0);
1230	percpu_counter_destroy(em_counter);
1231	percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1232	btrfs_free_csum_hash(fs_info);
1233	btrfs_free_stripe_hash_table(fs_info);
1234	btrfs_free_ref_cache(fs_info);
1235	kfree(fs_info->balance_ctl);
1236	kfree(fs_info->delayed_root);
1237	free_global_roots(fs_info);
1238	btrfs_put_root(fs_info->tree_root);
1239	btrfs_put_root(fs_info->chunk_root);
1240	btrfs_put_root(fs_info->dev_root);
1241	btrfs_put_root(fs_info->quota_root);
1242	btrfs_put_root(fs_info->uuid_root);
1243	btrfs_put_root(fs_info->fs_root);
1244	btrfs_put_root(fs_info->data_reloc_root);
1245	btrfs_put_root(fs_info->block_group_root);
1246	btrfs_put_root(fs_info->stripe_root);
1247	btrfs_check_leaked_roots(fs_info);
1248	btrfs_extent_buffer_leak_debug_check(fs_info);
1249	kfree(fs_info->super_copy);
1250	kfree(fs_info->super_for_commit);
1251	kvfree(fs_info);
1252}
1253
1254
1255/*
1256 * Get an in-memory reference of a root structure.
1257 *
1258 * For essential trees like root/extent tree, we grab it from fs_info directly.
1259 * For subvolume trees, we check the cached filesystem roots first. If not
1260 * found, then read it from disk and add it to cached fs roots.
1261 *
1262 * Caller should release the root by calling btrfs_put_root() after the usage.
1263 *
1264 * NOTE: Reloc and log trees can't be read by this function as they share the
1265 *	 same root objectid.
1266 *
1267 * @objectid:	root id
1268 * @anon_dev:	preallocated anonymous block device number for new roots,
1269 *		pass NULL for a new allocation.
1270 * @check_ref:	whether to check root item references, If true, return -ENOENT
1271 *		for orphan roots
1272 */
1273static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1274					     u64 objectid, dev_t *anon_dev,
1275					     bool check_ref)
1276{
1277	struct btrfs_root *root;
1278	struct btrfs_path *path;
1279	struct btrfs_key key;
1280	int ret;
1281
1282	root = btrfs_get_global_root(fs_info, objectid);
1283	if (root)
1284		return root;
1285
1286	/*
1287	 * If we're called for non-subvolume trees, and above function didn't
1288	 * find one, do not try to read it from disk.
1289	 *
1290	 * This is namely for free-space-tree and quota tree, which can change
1291	 * at runtime and should only be grabbed from fs_info.
1292	 */
1293	if (!btrfs_is_fstree(objectid) && objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1294		return ERR_PTR(-ENOENT);
1295again:
1296	root = btrfs_lookup_fs_root(fs_info, objectid);
1297	if (root) {
1298		/*
1299		 * Some other caller may have read out the newly inserted
1300		 * subvolume already (for things like backref walk etc).  Not
1301		 * that common but still possible.  In that case, we just need
1302		 * to free the anon_dev.
1303		 */
1304		if (unlikely(anon_dev && *anon_dev)) {
1305			free_anon_bdev(*anon_dev);
1306			*anon_dev = 0;
1307		}
1308
1309		if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1310			btrfs_put_root(root);
1311			return ERR_PTR(-ENOENT);
1312		}
1313		return root;
1314	}
1315
1316	key.objectid = objectid;
1317	key.type = BTRFS_ROOT_ITEM_KEY;
1318	key.offset = (u64)-1;
1319	root = btrfs_read_tree_root(fs_info->tree_root, &key);
1320	if (IS_ERR(root))
1321		return root;
1322
1323	if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1324		ret = -ENOENT;
1325		goto fail;
1326	}
1327
1328	ret = btrfs_init_fs_root(root, anon_dev ? *anon_dev : 0);
1329	if (ret)
1330		goto fail;
1331
1332	path = btrfs_alloc_path();
1333	if (!path) {
1334		ret = -ENOMEM;
1335		goto fail;
1336	}
1337	key.objectid = BTRFS_ORPHAN_OBJECTID;
1338	key.type = BTRFS_ORPHAN_ITEM_KEY;
1339	key.offset = objectid;
1340
1341	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1342	btrfs_free_path(path);
1343	if (ret < 0)
1344		goto fail;
1345	if (ret == 0)
1346		set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1347
1348	ret = btrfs_insert_fs_root(fs_info, root);
1349	if (ret) {
1350		if (ret == -EEXIST) {
1351			btrfs_put_root(root);
1352			goto again;
1353		}
1354		goto fail;
1355	}
1356	return root;
1357fail:
1358	/*
1359	 * If our caller provided us an anonymous device, then it's his
1360	 * responsibility to free it in case we fail. So we have to set our
1361	 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1362	 * and once again by our caller.
1363	 */
1364	if (anon_dev && *anon_dev)
1365		root->anon_dev = 0;
1366	btrfs_put_root(root);
1367	return ERR_PTR(ret);
1368}
1369
1370/*
1371 * Get in-memory reference of a root structure
1372 *
1373 * @objectid:	tree objectid
1374 * @check_ref:	if set, verify that the tree exists and the item has at least
1375 *		one reference
1376 */
1377struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1378				     u64 objectid, bool check_ref)
1379{
1380	return btrfs_get_root_ref(fs_info, objectid, NULL, check_ref);
1381}
1382
1383/*
1384 * Get in-memory reference of a root structure, created as new, optionally pass
1385 * the anonymous block device id
1386 *
1387 * @objectid:	tree objectid
1388 * @anon_dev:	if NULL, allocate a new anonymous block device or use the
1389 *		parameter value if not NULL
1390 */
1391struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1392					 u64 objectid, dev_t *anon_dev)
1393{
1394	return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1395}
1396
1397/*
1398 * Return a root for the given objectid.
1399 *
1400 * @fs_info:	the fs_info
1401 * @objectid:	the objectid we need to lookup
1402 *
1403 * This is exclusively used for backref walking, and exists specifically because
1404 * of how qgroups does lookups.  Qgroups will do a backref lookup at delayed ref
1405 * creation time, which means we may have to read the tree_root in order to look
1406 * up a fs root that is not in memory.  If the root is not in memory we will
1407 * read the tree root commit root and look up the fs root from there.  This is a
1408 * temporary root, it will not be inserted into the radix tree as it doesn't
1409 * have the most uptodate information, it'll simply be discarded once the
1410 * backref code is finished using the root.
1411 */
1412struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1413						 struct btrfs_path *path,
1414						 u64 objectid)
1415{
1416	struct btrfs_root *root;
1417	struct btrfs_key key;
1418
1419	ASSERT(path->search_commit_root && path->skip_locking);
1420
1421	/*
1422	 * This can return -ENOENT if we ask for a root that doesn't exist, but
1423	 * since this is called via the backref walking code we won't be looking
1424	 * up a root that doesn't exist, unless there's corruption.  So if root
1425	 * != NULL just return it.
1426	 */
1427	root = btrfs_get_global_root(fs_info, objectid);
1428	if (root)
1429		return root;
1430
1431	root = btrfs_lookup_fs_root(fs_info, objectid);
1432	if (root)
1433		return root;
1434
1435	key.objectid = objectid;
1436	key.type = BTRFS_ROOT_ITEM_KEY;
1437	key.offset = (u64)-1;
1438	root = read_tree_root_path(fs_info->tree_root, path, &key);
1439	btrfs_release_path(path);
1440
1441	return root;
1442}
1443
1444static int cleaner_kthread(void *arg)
1445{
1446	struct btrfs_fs_info *fs_info = arg;
1447	int again;
1448
1449	while (1) {
1450		again = 0;
1451
1452		set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1453
1454		/* Make the cleaner go to sleep early. */
1455		if (btrfs_need_cleaner_sleep(fs_info))
1456			goto sleep;
1457
1458		/*
1459		 * Do not do anything if we might cause open_ctree() to block
1460		 * before we have finished mounting the filesystem.
1461		 */
1462		if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1463			goto sleep;
1464
1465		if (!mutex_trylock(&fs_info->cleaner_mutex))
1466			goto sleep;
1467
1468		/*
1469		 * Avoid the problem that we change the status of the fs
1470		 * during the above check and trylock.
1471		 */
1472		if (btrfs_need_cleaner_sleep(fs_info)) {
1473			mutex_unlock(&fs_info->cleaner_mutex);
1474			goto sleep;
1475		}
1476
1477		if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags))
1478			btrfs_sysfs_feature_update(fs_info);
1479
1480		btrfs_run_delayed_iputs(fs_info);
1481
1482		again = btrfs_clean_one_deleted_snapshot(fs_info);
1483		mutex_unlock(&fs_info->cleaner_mutex);
1484
1485		/*
1486		 * The defragger has dealt with the R/O remount and umount,
1487		 * needn't do anything special here.
1488		 */
1489		btrfs_run_defrag_inodes(fs_info);
1490
1491		/*
1492		 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1493		 * with relocation (btrfs_relocate_chunk) and relocation
1494		 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1495		 * after acquiring fs_info->reclaim_bgs_lock. So we
1496		 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1497		 * unused block groups.
1498		 */
1499		btrfs_delete_unused_bgs(fs_info);
1500
1501		/*
1502		 * Reclaim block groups in the reclaim_bgs list after we deleted
1503		 * all unused block_groups. This possibly gives us some more free
1504		 * space.
1505		 */
1506		btrfs_reclaim_bgs(fs_info);
1507sleep:
1508		clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1509		if (kthread_should_park())
1510			kthread_parkme();
1511		if (kthread_should_stop())
1512			return 0;
1513		if (!again) {
1514			set_current_state(TASK_INTERRUPTIBLE);
1515			schedule();
1516			__set_current_state(TASK_RUNNING);
1517		}
1518	}
1519}
1520
1521static int transaction_kthread(void *arg)
1522{
1523	struct btrfs_root *root = arg;
1524	struct btrfs_fs_info *fs_info = root->fs_info;
1525	struct btrfs_trans_handle *trans;
1526	struct btrfs_transaction *cur;
1527	u64 transid;
1528	time64_t delta;
1529	unsigned long delay;
1530	bool cannot_commit;
1531
1532	do {
1533		cannot_commit = false;
1534		delay = secs_to_jiffies(fs_info->commit_interval);
1535		mutex_lock(&fs_info->transaction_kthread_mutex);
1536
1537		spin_lock(&fs_info->trans_lock);
1538		cur = fs_info->running_transaction;
1539		if (!cur) {
1540			spin_unlock(&fs_info->trans_lock);
1541			goto sleep;
1542		}
1543
1544		delta = ktime_get_seconds() - cur->start_time;
1545		if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1546		    cur->state < TRANS_STATE_COMMIT_PREP &&
1547		    delta < fs_info->commit_interval) {
1548			spin_unlock(&fs_info->trans_lock);
1549			delay -= secs_to_jiffies(delta - 1);
1550			delay = min(delay,
1551				    secs_to_jiffies(fs_info->commit_interval));
1552			goto sleep;
1553		}
1554		transid = cur->transid;
1555		spin_unlock(&fs_info->trans_lock);
1556
1557		/* If the file system is aborted, this will always fail. */
1558		trans = btrfs_attach_transaction(root);
1559		if (IS_ERR(trans)) {
1560			if (PTR_ERR(trans) != -ENOENT)
1561				cannot_commit = true;
1562			goto sleep;
1563		}
1564		if (transid == trans->transid) {
1565			btrfs_commit_transaction(trans);
1566		} else {
1567			btrfs_end_transaction(trans);
1568		}
1569sleep:
1570		wake_up_process(fs_info->cleaner_kthread);
1571		mutex_unlock(&fs_info->transaction_kthread_mutex);
1572
1573		if (BTRFS_FS_ERROR(fs_info))
1574			btrfs_cleanup_transaction(fs_info);
1575		if (!kthread_should_stop() &&
1576				(!btrfs_transaction_blocked(fs_info) ||
1577				 cannot_commit))
1578			schedule_timeout_interruptible(delay);
1579	} while (!kthread_should_stop());
1580	return 0;
1581}
1582
1583/*
1584 * This will find the highest generation in the array of root backups.  The
1585 * index of the highest array is returned, or -EINVAL if we can't find
1586 * anything.
1587 *
1588 * We check to make sure the array is valid by comparing the
1589 * generation of the latest  root in the array with the generation
1590 * in the super block.  If they don't match we pitch it.
1591 */
1592static int find_newest_super_backup(struct btrfs_fs_info *info)
1593{
1594	const u64 newest_gen = btrfs_super_generation(info->super_copy);
1595	u64 cur;
1596	struct btrfs_root_backup *root_backup;
1597	int i;
1598
1599	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1600		root_backup = info->super_copy->super_roots + i;
1601		cur = btrfs_backup_tree_root_gen(root_backup);
1602		if (cur == newest_gen)
1603			return i;
1604	}
1605
1606	return -EINVAL;
1607}
1608
1609/*
1610 * copy all the root pointers into the super backup array.
1611 * this will bump the backup pointer by one when it is
1612 * done
1613 */
1614static void backup_super_roots(struct btrfs_fs_info *info)
1615{
1616	const int next_backup = info->backup_root_index;
1617	struct btrfs_root_backup *root_backup;
1618
1619	root_backup = info->super_for_commit->super_roots + next_backup;
1620
1621	/*
1622	 * make sure all of our padding and empty slots get zero filled
1623	 * regardless of which ones we use today
1624	 */
1625	memset(root_backup, 0, sizeof(*root_backup));
1626
1627	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1628
1629	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1630	btrfs_set_backup_tree_root_gen(root_backup,
1631			       btrfs_header_generation(info->tree_root->node));
1632
1633	btrfs_set_backup_tree_root_level(root_backup,
1634			       btrfs_header_level(info->tree_root->node));
1635
1636	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1637	btrfs_set_backup_chunk_root_gen(root_backup,
1638			       btrfs_header_generation(info->chunk_root->node));
1639	btrfs_set_backup_chunk_root_level(root_backup,
1640			       btrfs_header_level(info->chunk_root->node));
1641
1642	if (!btrfs_fs_incompat(info, EXTENT_TREE_V2)) {
1643		struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
1644		struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
1645
1646		btrfs_set_backup_extent_root(root_backup,
1647					     extent_root->node->start);
1648		btrfs_set_backup_extent_root_gen(root_backup,
1649				btrfs_header_generation(extent_root->node));
1650		btrfs_set_backup_extent_root_level(root_backup,
1651					btrfs_header_level(extent_root->node));
1652
1653		btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
1654		btrfs_set_backup_csum_root_gen(root_backup,
1655					       btrfs_header_generation(csum_root->node));
1656		btrfs_set_backup_csum_root_level(root_backup,
1657						 btrfs_header_level(csum_root->node));
1658	}
1659
1660	/*
1661	 * we might commit during log recovery, which happens before we set
1662	 * the fs_root.  Make sure it is valid before we fill it in.
1663	 */
1664	if (info->fs_root && info->fs_root->node) {
1665		btrfs_set_backup_fs_root(root_backup,
1666					 info->fs_root->node->start);
1667		btrfs_set_backup_fs_root_gen(root_backup,
1668			       btrfs_header_generation(info->fs_root->node));
1669		btrfs_set_backup_fs_root_level(root_backup,
1670			       btrfs_header_level(info->fs_root->node));
1671	}
1672
1673	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1674	btrfs_set_backup_dev_root_gen(root_backup,
1675			       btrfs_header_generation(info->dev_root->node));
1676	btrfs_set_backup_dev_root_level(root_backup,
1677				       btrfs_header_level(info->dev_root->node));
1678
1679	btrfs_set_backup_total_bytes(root_backup,
1680			     btrfs_super_total_bytes(info->super_copy));
1681	btrfs_set_backup_bytes_used(root_backup,
1682			     btrfs_super_bytes_used(info->super_copy));
1683	btrfs_set_backup_num_devices(root_backup,
1684			     btrfs_super_num_devices(info->super_copy));
1685
1686	/*
1687	 * if we don't copy this out to the super_copy, it won't get remembered
1688	 * for the next commit
1689	 */
1690	memcpy(&info->super_copy->super_roots,
1691	       &info->super_for_commit->super_roots,
1692	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1693}
1694
1695/*
1696 * Reads a backup root based on the passed priority. Prio 0 is the newest, prio
1697 * 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1698 *
1699 * @fs_info:  filesystem whose backup roots need to be read
1700 * @priority: priority of backup root required
1701 *
1702 * Returns backup root index on success and -EINVAL otherwise.
1703 */
1704static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1705{
1706	int backup_index = find_newest_super_backup(fs_info);
1707	struct btrfs_super_block *super = fs_info->super_copy;
1708	struct btrfs_root_backup *root_backup;
1709
1710	if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1711		if (priority == 0)
1712			return backup_index;
1713
1714		backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1715		backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1716	} else {
1717		return -EINVAL;
1718	}
1719
1720	root_backup = super->super_roots + backup_index;
1721
1722	btrfs_set_super_generation(super,
1723				   btrfs_backup_tree_root_gen(root_backup));
1724	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1725	btrfs_set_super_root_level(super,
1726				   btrfs_backup_tree_root_level(root_backup));
1727	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1728
1729	/*
1730	 * Fixme: the total bytes and num_devices need to match or we should
1731	 * need a fsck
1732	 */
1733	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1734	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1735
1736	return backup_index;
1737}
1738
1739/* helper to cleanup workers */
1740static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1741{
1742	btrfs_destroy_workqueue(fs_info->fixup_workers);
1743	btrfs_destroy_workqueue(fs_info->delalloc_workers);
1744	btrfs_destroy_workqueue(fs_info->workers);
1745	if (fs_info->endio_workers)
1746		destroy_workqueue(fs_info->endio_workers);
1747	if (fs_info->rmw_workers)
1748		destroy_workqueue(fs_info->rmw_workers);
1749	btrfs_destroy_workqueue(fs_info->endio_write_workers);
1750	btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1751	btrfs_destroy_workqueue(fs_info->delayed_workers);
1752	btrfs_destroy_workqueue(fs_info->caching_workers);
1753	btrfs_destroy_workqueue(fs_info->flush_workers);
1754	btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1755	if (fs_info->discard_ctl.discard_workers)
1756		destroy_workqueue(fs_info->discard_ctl.discard_workers);
1757	/*
1758	 * Now that all other work queues are destroyed, we can safely destroy
1759	 * the queues used for metadata I/O, since tasks from those other work
1760	 * queues can do metadata I/O operations.
1761	 */
1762	if (fs_info->endio_meta_workers)
1763		destroy_workqueue(fs_info->endio_meta_workers);
1764}
1765
1766static void free_root_extent_buffers(struct btrfs_root *root)
1767{
1768	if (root) {
1769		free_extent_buffer(root->node);
1770		free_extent_buffer(root->commit_root);
1771		root->node = NULL;
1772		root->commit_root = NULL;
1773	}
1774}
1775
1776static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
1777{
1778	struct btrfs_root *root, *tmp;
1779
1780	rbtree_postorder_for_each_entry_safe(root, tmp,
1781					     &fs_info->global_root_tree,
1782					     rb_node)
1783		free_root_extent_buffers(root);
1784}
1785
1786/* helper to cleanup tree roots */
1787static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1788{
1789	free_root_extent_buffers(info->tree_root);
1790
1791	free_global_root_pointers(info);
1792	free_root_extent_buffers(info->dev_root);
1793	free_root_extent_buffers(info->quota_root);
1794	free_root_extent_buffers(info->uuid_root);
1795	free_root_extent_buffers(info->fs_root);
1796	free_root_extent_buffers(info->data_reloc_root);
1797	free_root_extent_buffers(info->block_group_root);
1798	free_root_extent_buffers(info->stripe_root);
1799	if (free_chunk_root)
1800		free_root_extent_buffers(info->chunk_root);
1801}
1802
1803void btrfs_put_root(struct btrfs_root *root)
1804{
1805	if (!root)
1806		return;
1807
1808	if (refcount_dec_and_test(&root->refs)) {
1809		if (WARN_ON(!xa_empty(&root->inodes)))
1810			xa_destroy(&root->inodes);
1811		if (WARN_ON(!xa_empty(&root->delayed_nodes)))
1812			xa_destroy(&root->delayed_nodes);
1813		WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
1814		if (root->anon_dev)
1815			free_anon_bdev(root->anon_dev);
1816		free_root_extent_buffers(root);
1817#ifdef CONFIG_BTRFS_DEBUG
1818		spin_lock(&root->fs_info->fs_roots_radix_lock);
1819		list_del_init(&root->leak_list);
1820		spin_unlock(&root->fs_info->fs_roots_radix_lock);
1821#endif
1822		kfree(root);
1823	}
1824}
1825
1826void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
1827{
1828	int ret;
1829	struct btrfs_root *gang[8];
1830	int i;
1831
1832	while (!list_empty(&fs_info->dead_roots)) {
1833		gang[0] = list_first_entry(&fs_info->dead_roots,
1834					   struct btrfs_root, root_list);
1835		list_del(&gang[0]->root_list);
1836
1837		if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
1838			btrfs_drop_and_free_fs_root(fs_info, gang[0]);
1839		btrfs_put_root(gang[0]);
1840	}
1841
1842	while (1) {
1843		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1844					     (void **)gang, 0,
1845					     ARRAY_SIZE(gang));
1846		if (!ret)
1847			break;
1848		for (i = 0; i < ret; i++)
1849			btrfs_drop_and_free_fs_root(fs_info, gang[i]);
1850	}
1851}
1852
1853static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
1854{
1855	mutex_init(&fs_info->scrub_lock);
1856	atomic_set(&fs_info->scrubs_running, 0);
1857	atomic_set(&fs_info->scrub_pause_req, 0);
1858	atomic_set(&fs_info->scrubs_paused, 0);
1859	atomic_set(&fs_info->scrub_cancel_req, 0);
1860	init_waitqueue_head(&fs_info->scrub_pause_wait);
1861	refcount_set(&fs_info->scrub_workers_refcnt, 0);
1862}
1863
1864static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
1865{
1866	spin_lock_init(&fs_info->balance_lock);
1867	mutex_init(&fs_info->balance_mutex);
1868	atomic_set(&fs_info->balance_pause_req, 0);
1869	atomic_set(&fs_info->balance_cancel_req, 0);
1870	fs_info->balance_ctl = NULL;
1871	init_waitqueue_head(&fs_info->balance_wait_q);
1872	atomic_set(&fs_info->reloc_cancel_req, 0);
1873}
1874
1875static int btrfs_init_btree_inode(struct super_block *sb)
1876{
1877	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1878	unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
1879					      fs_info->tree_root);
1880	struct inode *inode;
1881
1882	inode = new_inode(sb);
1883	if (!inode)
1884		return -ENOMEM;
1885
1886	btrfs_set_inode_number(BTRFS_I(inode), BTRFS_BTREE_INODE_OBJECTID);
1887	set_nlink(inode, 1);
1888	/*
1889	 * we set the i_size on the btree inode to the max possible int.
1890	 * the real end of the address space is determined by all of
1891	 * the devices in the system
1892	 */
1893	inode->i_size = OFFSET_MAX;
1894	inode->i_mapping->a_ops = &btree_aops;
1895	mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
1896
1897	btrfs_extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
1898				  IO_TREE_BTREE_INODE_IO);
1899	btrfs_extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
1900
1901	BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
1902	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
1903	__insert_inode_hash(inode, hash);
1904	set_bit(AS_KERNEL_FILE, &inode->i_mapping->flags);
1905	fs_info->btree_inode = inode;
1906
1907	return 0;
1908}
1909
1910static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
1911{
1912	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
1913	init_rwsem(&fs_info->dev_replace.rwsem);
1914	init_waitqueue_head(&fs_info->dev_replace.replace_wait);
1915}
1916
1917static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
1918{
1919	spin_lock_init(&fs_info->qgroup_lock);
1920	mutex_init(&fs_info->qgroup_ioctl_lock);
1921	fs_info->qgroup_tree = RB_ROOT;
1922	INIT_LIST_HEAD(&fs_info->dirty_qgroups);
1923	fs_info->qgroup_seq = 1;
1924	fs_info->qgroup_rescan_running = false;
1925	fs_info->qgroup_drop_subtree_thres = BTRFS_QGROUP_DROP_SUBTREE_THRES_DEFAULT;
1926	mutex_init(&fs_info->qgroup_rescan_lock);
1927}
1928
1929static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
1930{
1931	u32 max_active = fs_info->thread_pool_size;
1932	unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
1933	unsigned int ordered_flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_PERCPU;
1934
1935	fs_info->workers =
1936		btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
1937
1938	fs_info->delalloc_workers =
1939		btrfs_alloc_workqueue(fs_info, "delalloc",
1940				      flags, max_active, 2);
1941
1942	fs_info->flush_workers =
1943		btrfs_alloc_workqueue(fs_info, "flush_delalloc",
1944				      flags, max_active, 0);
1945
1946	fs_info->caching_workers =
1947		btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
1948
1949	fs_info->fixup_workers =
1950		btrfs_alloc_ordered_workqueue(fs_info, "fixup", ordered_flags);
1951
1952	fs_info->endio_workers =
1953		alloc_workqueue("btrfs-endio", flags, max_active);
1954	fs_info->endio_meta_workers =
1955		alloc_workqueue("btrfs-endio-meta", flags, max_active);
1956	fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
1957	fs_info->endio_write_workers =
1958		btrfs_alloc_workqueue(fs_info, "endio-write", flags,
1959				      max_active, 2);
1960	fs_info->endio_freespace_worker =
1961		btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
1962				      max_active, 0);
1963	fs_info->delayed_workers =
1964		btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
1965				      max_active, 0);
1966	fs_info->qgroup_rescan_workers =
1967		btrfs_alloc_ordered_workqueue(fs_info, "qgroup-rescan",
1968					      ordered_flags);
1969	fs_info->discard_ctl.discard_workers =
1970		alloc_ordered_workqueue("btrfs-discard", WQ_FREEZABLE);
1971
1972	if (!(fs_info->workers &&
1973	      fs_info->delalloc_workers && fs_info->flush_workers &&
1974	      fs_info->endio_workers && fs_info->endio_meta_workers &&
1975	      fs_info->endio_write_workers &&
1976	      fs_info->endio_freespace_worker && fs_info->rmw_workers &&
1977	      fs_info->caching_workers && fs_info->fixup_workers &&
1978	      fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
1979	      fs_info->discard_ctl.discard_workers)) {
1980		return -ENOMEM;
1981	}
1982
1983	return 0;
1984}
1985
1986static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
1987{
1988	struct crypto_shash *csum_shash;
1989	const char *csum_driver = btrfs_super_csum_driver(csum_type);
1990
1991	csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
1992
1993	if (IS_ERR(csum_shash)) {
1994		btrfs_err(fs_info, "error allocating %s hash for checksum",
1995			  csum_driver);
1996		return PTR_ERR(csum_shash);
1997	}
1998
1999	fs_info->csum_shash = csum_shash;
2000
2001	/* Check if the checksum implementation is a fast accelerated one. */
2002	switch (csum_type) {
2003	case BTRFS_CSUM_TYPE_CRC32:
2004		if (crc32_optimizations() & CRC32C_OPTIMIZATION)
2005			set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2006		break;
2007	case BTRFS_CSUM_TYPE_XXHASH:
2008		set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2009		break;
2010	default:
2011		break;
2012	}
2013
2014	btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2015			btrfs_super_csum_name(csum_type),
2016			crypto_shash_driver_name(csum_shash));
2017	return 0;
2018}
2019
2020static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2021			    struct btrfs_fs_devices *fs_devices)
2022{
2023	int ret;
2024	struct btrfs_tree_parent_check check = { 0 };
2025	struct btrfs_root *log_tree_root;
2026	struct btrfs_super_block *disk_super = fs_info->super_copy;
2027	u64 bytenr = btrfs_super_log_root(disk_super);
2028	int level = btrfs_super_log_root_level(disk_super);
2029
2030	if (unlikely(fs_devices->rw_devices == 0)) {
2031		btrfs_warn(fs_info, "log replay required on RO media");
2032		return -EIO;
2033	}
2034
2035	log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2036					 GFP_KERNEL);
2037	if (!log_tree_root)
2038		return -ENOMEM;
2039
2040	check.level = level;
2041	check.transid = fs_info->generation + 1;
2042	check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2043	log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
2044	if (IS_ERR(log_tree_root->node)) {
2045		btrfs_warn(fs_info, "failed to read log tree");
2046		ret = PTR_ERR(log_tree_root->node);
2047		log_tree_root->node = NULL;
2048		btrfs_put_root(log_tree_root);
2049		return ret;
2050	}
2051	if (unlikely(!extent_buffer_uptodate(log_tree_root->node))) {
2052		btrfs_err(fs_info, "failed to read log tree");
2053		btrfs_put_root(log_tree_root);
2054		return -EIO;
2055	}
2056
2057	/* returns with log_tree_root freed on success */
2058	ret = btrfs_recover_log_trees(log_tree_root);
2059	btrfs_put_root(log_tree_root);
2060	if (ret) {
2061		btrfs_handle_fs_error(fs_info, ret,
2062				      "Failed to recover log tree");
2063		return ret;
2064	}
2065
2066	if (sb_rdonly(fs_info->sb)) {
2067		ret = btrfs_commit_super(fs_info);
2068		if (ret)
2069			return ret;
2070	}
2071
2072	return 0;
2073}
2074
2075static int load_global_roots_objectid(struct btrfs_root *tree_root,
2076				      struct btrfs_path *path, u64 objectid,
2077				      const char *name)
2078{
2079	struct btrfs_fs_info *fs_info = tree_root->fs_info;
2080	struct btrfs_root *root;
2081	u64 max_global_id = 0;
2082	int ret;
2083	struct btrfs_key key = {
2084		.objectid = objectid,
2085		.type = BTRFS_ROOT_ITEM_KEY,
2086		.offset = 0,
2087	};
2088	bool found = false;
2089
2090	/* If we have IGNOREDATACSUMS skip loading these roots. */
2091	if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2092	    btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2093		set_bit(BTRFS_FS_STATE_NO_DATA_CSUMS, &fs_info->fs_state);
2094		return 0;
2095	}
2096
2097	while (1) {
2098		ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2099		if (ret < 0)
2100			break;
2101
2102		if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2103			ret = btrfs_next_leaf(tree_root, path);
2104			if (ret) {
2105				if (ret > 0)
2106					ret = 0;
2107				break;
2108			}
2109		}
2110		ret = 0;
2111
2112		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2113		if (key.objectid != objectid)
2114			break;
2115		btrfs_release_path(path);
2116
2117		/*
2118		 * Just worry about this for extent tree, it'll be the same for
2119		 * everybody.
2120		 */
2121		if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2122			max_global_id = max(max_global_id, key.offset);
2123
2124		found = true;
2125		root = read_tree_root_path(tree_root, path, &key);
2126		if (IS_ERR(root)) {
2127			ret = PTR_ERR(root);
2128			break;
2129		}
2130		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2131		ret = btrfs_global_root_insert(root);
2132		if (ret) {
2133			btrfs_put_root(root);
2134			break;
2135		}
2136		key.offset++;
2137	}
2138	btrfs_release_path(path);
2139
2140	if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2141		fs_info->nr_global_roots = max_global_id + 1;
2142
2143	if (!found || ret) {
2144		if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2145			set_bit(BTRFS_FS_STATE_NO_DATA_CSUMS, &fs_info->fs_state);
2146
2147		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2148			ret = ret ? ret : -ENOENT;
2149		else
2150			ret = 0;
2151		btrfs_err(fs_info, "failed to load root %s", name);
2152	}
2153	return ret;
2154}
2155
2156static int load_global_roots(struct btrfs_root *tree_root)
2157{
2158	BTRFS_PATH_AUTO_FREE(path);
2159	int ret;
2160
2161	path = btrfs_alloc_path();
2162	if (!path)
2163		return -ENOMEM;
2164
2165	ret = load_global_roots_objectid(tree_root, path,
2166					 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2167	if (ret)
2168		return ret;
2169	ret = load_global_roots_objectid(tree_root, path,
2170					 BTRFS_CSUM_TREE_OBJECTID, "csum");
2171	if (ret)
2172		return ret;
2173	if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2174		return ret;
2175	ret = load_global_roots_objectid(tree_root, path,
2176					 BTRFS_FREE_SPACE_TREE_OBJECTID,
2177					 "free space");
2178
2179	return ret;
2180}
2181
2182static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2183{
2184	struct btrfs_root *tree_root = fs_info->tree_root;
2185	struct btrfs_root *root;
2186	struct btrfs_key location;
2187	int ret;
2188
2189	ASSERT(fs_info->tree_root);
2190
2191	ret = load_global_roots(tree_root);
2192	if (ret)
2193		return ret;
2194
2195	location.type = BTRFS_ROOT_ITEM_KEY;
2196	location.offset = 0;
2197
2198	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2199		location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2200		root = btrfs_read_tree_root(tree_root, &location);
2201		if (IS_ERR(root)) {
2202			if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2203				ret = PTR_ERR(root);
2204				goto out;
2205			}
2206		} else {
2207			set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2208			fs_info->block_group_root = root;
2209		}
2210	}
2211
2212	location.objectid = BTRFS_DEV_TREE_OBJECTID;
2213	root = btrfs_read_tree_root(tree_root, &location);
2214	if (IS_ERR(root)) {
2215		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2216			ret = PTR_ERR(root);
2217			goto out;
2218		}
2219	} else {
2220		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2221		fs_info->dev_root = root;
2222	}
2223	/* Initialize fs_info for all devices in any case */
2224	ret = btrfs_init_devices_late(fs_info);
2225	if (ret)
2226		goto out;
2227
2228	/*
2229	 * This tree can share blocks with some other fs tree during relocation
2230	 * and we need a proper setup by btrfs_get_fs_root
2231	 */
2232	root = btrfs_get_fs_root(tree_root->fs_info,
2233				 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2234	if (IS_ERR(root)) {
2235		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2236			location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
2237			ret = PTR_ERR(root);
2238			goto out;
2239		}
2240	} else {
2241		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2242		fs_info->data_reloc_root = root;
2243	}
2244
2245	location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2246	root = btrfs_read_tree_root(tree_root, &location);
2247	if (!IS_ERR(root)) {
2248		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2249		fs_info->quota_root = root;
2250	}
2251
2252	location.objectid = BTRFS_UUID_TREE_OBJECTID;
2253	root = btrfs_read_tree_root(tree_root, &location);
2254	if (IS_ERR(root)) {
2255		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2256			ret = PTR_ERR(root);
2257			if (ret != -ENOENT)
2258				goto out;
2259		}
2260	} else {
2261		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2262		fs_info->uuid_root = root;
2263	}
2264
2265	if (btrfs_fs_incompat(fs_info, RAID_STRIPE_TREE)) {
2266		location.objectid = BTRFS_RAID_STRIPE_TREE_OBJECTID;
2267		root = btrfs_read_tree_root(tree_root, &location);
2268		if (IS_ERR(root)) {
2269			if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2270				ret = PTR_ERR(root);
2271				goto out;
2272			}
2273		} else {
2274			set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2275			fs_info->stripe_root = root;
2276		}
2277	}
2278
2279	return 0;
2280out:
2281	btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2282		   location.objectid, ret);
2283	return ret;
2284}
2285
2286static int validate_sys_chunk_array(const struct btrfs_fs_info *fs_info,
2287				    const struct btrfs_super_block *sb)
2288{
2289	unsigned int cur = 0; /* Offset inside the sys chunk array */
2290	/*
2291	 * At sb read time, fs_info is not fully initialized. Thus we have
2292	 * to use super block sectorsize, which should have been validated.
2293	 */
2294	const u32 sectorsize = btrfs_super_sectorsize(sb);
2295	u32 sys_array_size = btrfs_super_sys_array_size(sb);
2296
2297	if (unlikely(sys_array_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)) {
2298		btrfs_err(fs_info, "system chunk array too big %u > %u",
2299			  sys_array_size, BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2300		return -EUCLEAN;
2301	}
2302
2303	while (cur < sys_array_size) {
2304		struct btrfs_disk_key *disk_key;
2305		struct btrfs_chunk *chunk;
2306		struct btrfs_key key;
2307		u64 type;
2308		u16 num_stripes;
2309		u32 len;
2310		int ret;
2311
2312		disk_key = (struct btrfs_disk_key *)(sb->sys_chunk_array + cur);
2313		len = sizeof(*disk_key);
2314
2315		if (unlikely(cur + len > sys_array_size))
2316			goto short_read;
2317		cur += len;
2318
2319		btrfs_disk_key_to_cpu(&key, disk_key);
2320		if (unlikely(key.type != BTRFS_CHUNK_ITEM_KEY)) {
2321			btrfs_err(fs_info,
2322			    "unexpected item type %u in sys_array at offset %u",
2323				  key.type, cur);
2324			return -EUCLEAN;
2325		}
2326		chunk = (struct btrfs_chunk *)(sb->sys_chunk_array + cur);
2327		num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2328		if (unlikely(cur + btrfs_chunk_item_size(num_stripes) > sys_array_size))
2329			goto short_read;
2330		type = btrfs_stack_chunk_type(chunk);
2331		if (unlikely(!(type & BTRFS_BLOCK_GROUP_SYSTEM))) {
2332			btrfs_err(fs_info,
2333			"invalid chunk type %llu in sys_array at offset %u",
2334				  type, cur);
2335			return -EUCLEAN;
2336		}
2337		ret = btrfs_check_chunk_valid(fs_info, NULL, chunk, key.offset,
2338					      sectorsize);
2339		if (ret < 0)
2340			return ret;
2341		cur += btrfs_chunk_item_size(num_stripes);
2342	}
2343	return 0;
2344short_read:
2345	btrfs_err(fs_info,
2346	"super block sys chunk array short read, cur=%u sys_array_size=%u",
2347		  cur, sys_array_size);
2348	return -EUCLEAN;
2349}
2350
2351/*
2352 * Real super block validation
2353 * NOTE: super csum type and incompat features will not be checked here.
2354 *
2355 * @sb:		super block to check
2356 * @mirror_num:	the super block number to check its bytenr:
2357 * 		0	the primary (1st) sb
2358 * 		1, 2	2nd and 3rd backup copy
2359 * 	       -1	skip bytenr check
2360 */
2361int btrfs_validate_super(const struct btrfs_fs_info *fs_info,
2362			 const struct btrfs_super_block *sb, int mirror_num)
2363{
2364	u64 nodesize = btrfs_super_nodesize(sb);
2365	u64 sectorsize = btrfs_super_sectorsize(sb);
2366	int ret = 0;
2367	const bool ignore_flags = btrfs_test_opt(fs_info, IGNORESUPERFLAGS);
2368
2369	if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2370		btrfs_err(fs_info, "no valid FS found");
2371		ret = -EINVAL;
2372	}
2373	if ((btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)) {
2374		if (!ignore_flags) {
2375			btrfs_err(fs_info,
2376			"unrecognized or unsupported super flag 0x%llx",
2377				  btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2378			ret = -EINVAL;
2379		} else {
2380			btrfs_info(fs_info,
2381			"unrecognized or unsupported super flags: 0x%llx, ignored",
2382				   btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2383		}
2384	}
2385	if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2386		btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2387				btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2388		ret = -EINVAL;
2389	}
2390	if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2391		btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2392				btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2393		ret = -EINVAL;
2394	}
2395	if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2396		btrfs_err(fs_info, "log_root level too big: %d >= %d",
2397				btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2398		ret = -EINVAL;
2399	}
2400
2401	/*
2402	 * Check sectorsize and nodesize first, other check will need it.
2403	 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2404	 */
2405	if (!is_power_of_2(sectorsize) || sectorsize < BTRFS_MIN_BLOCKSIZE ||
2406	    sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2407		btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2408		ret = -EINVAL;
2409	}
2410
2411	if (!btrfs_supported_blocksize(sectorsize)) {
2412		btrfs_err(fs_info,
2413			"sectorsize %llu not yet supported for page size %lu",
2414			sectorsize, PAGE_SIZE);
2415		ret = -EINVAL;
2416	}
2417
2418	if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2419	    nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2420		btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2421		ret = -EINVAL;
2422	}
2423	if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2424		btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2425			  le32_to_cpu(sb->__unused_leafsize), nodesize);
2426		ret = -EINVAL;
2427	}
2428
2429	/* Root alignment check */
2430	if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2431		btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2432			   btrfs_super_root(sb));
2433		ret = -EINVAL;
2434	}
2435	if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2436		btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2437			   btrfs_super_chunk_root(sb));
2438		ret = -EINVAL;
2439	}
2440	if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2441		btrfs_warn(fs_info, "log_root block unaligned: %llu",
2442			   btrfs_super_log_root(sb));
2443		ret = -EINVAL;
2444	}
2445
2446	if (!fs_info->fs_devices->temp_fsid &&
2447	    memcmp(fs_info->fs_devices->fsid, sb->fsid, BTRFS_FSID_SIZE) != 0) {
2448		btrfs_err(fs_info,
2449		"superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2450			  sb->fsid, fs_info->fs_devices->fsid);
2451		ret = -EINVAL;
2452	}
2453
2454	if (memcmp(fs_info->fs_devices->metadata_uuid, btrfs_sb_fsid_ptr(sb),
2455		   BTRFS_FSID_SIZE) != 0) {
2456		btrfs_err(fs_info,
2457"superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2458			  btrfs_sb_fsid_ptr(sb), fs_info->fs_devices->metadata_uuid);
2459		ret = -EINVAL;
2460	}
2461
2462	if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2463		   BTRFS_FSID_SIZE) != 0) {
2464		btrfs_err(fs_info,
2465			"dev_item UUID does not match metadata fsid: %pU != %pU",
2466			fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2467		ret = -EINVAL;
2468	}
2469
2470	/*
2471	 * Artificial requirement for block-group-tree to force newer features
2472	 * (free-space-tree, no-holes) so the test matrix is smaller.
2473	 */
2474	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2475	    (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2476	     !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2477		btrfs_err(fs_info,
2478		"block-group-tree feature requires free-space-tree and no-holes");
2479		ret = -EINVAL;
2480	}
2481
2482	/*
2483	 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2484	 * done later
2485	 */
2486	if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2487		btrfs_err(fs_info, "bytes_used is too small %llu",
2488			  btrfs_super_bytes_used(sb));
2489		ret = -EINVAL;
2490	}
2491	if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2492		btrfs_err(fs_info, "invalid stripesize %u",
2493			  btrfs_super_stripesize(sb));
2494		ret = -EINVAL;
2495	}
2496	if (btrfs_super_num_devices(sb) > (1UL << 31))
2497		btrfs_warn(fs_info, "suspicious number of devices: %llu",
2498			   btrfs_super_num_devices(sb));
2499	if (btrfs_super_num_devices(sb) == 0) {
2500		btrfs_err(fs_info, "number of devices is 0");
2501		ret = -EINVAL;
2502	}
2503
2504	if (mirror_num >= 0 &&
2505	    btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2506		btrfs_err(fs_info, "super offset mismatch %llu != %llu",
2507			  btrfs_super_bytenr(sb), btrfs_sb_offset(mirror_num));
2508		ret = -EINVAL;
2509	}
2510
2511	if (ret)
2512		return ret;
2513
2514	ret = validate_sys_chunk_array(fs_info, sb);
2515
2516	/*
2517	 * Obvious sys_chunk_array corruptions, it must hold at least one key
2518	 * and one chunk
2519	 */
2520	if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2521		btrfs_err(fs_info, "system chunk array too big %u > %u",
2522			  btrfs_super_sys_array_size(sb),
2523			  BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2524		ret = -EINVAL;
2525	}
2526	if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2527			+ sizeof(struct btrfs_chunk)) {
2528		btrfs_err(fs_info, "system chunk array too small %u < %zu",
2529			  btrfs_super_sys_array_size(sb),
2530			  sizeof(struct btrfs_disk_key)
2531			  + sizeof(struct btrfs_chunk));
2532		ret = -EINVAL;
2533	}
2534
2535	/*
2536	 * The generation is a global counter, we'll trust it more than the others
2537	 * but it's still possible that it's the one that's wrong.
2538	 */
2539	if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2540		btrfs_warn(fs_info,
2541			"suspicious: generation < chunk_root_generation: %llu < %llu",
2542			btrfs_super_generation(sb),
2543			btrfs_super_chunk_root_generation(sb));
2544	if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2545	    && btrfs_super_cache_generation(sb) != (u64)-1)
2546		btrfs_warn(fs_info,
2547			"suspicious: generation < cache_generation: %llu < %llu",
2548			btrfs_super_generation(sb),
2549			btrfs_super_cache_generation(sb));
2550
2551	return ret;
2552}
2553
2554/*
2555 * Validation of super block at mount time.
2556 * Some checks already done early at mount time, like csum type and incompat
2557 * flags will be skipped.
2558 */
2559static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2560{
2561	return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2562}
2563
2564/*
2565 * Validation of super block at write time.
2566 * Some checks like bytenr check will be skipped as their values will be
2567 * overwritten soon.
2568 * Extra checks like csum type and incompat flags will be done here.
2569 */
2570static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2571				      struct btrfs_super_block *sb)
2572{
2573	int ret;
2574
2575	ret = btrfs_validate_super(fs_info, sb, -1);
2576	if (ret < 0)
2577		goto out;
2578	if (unlikely(!btrfs_supported_super_csum(btrfs_super_csum_type(sb)))) {
2579		ret = -EUCLEAN;
2580		btrfs_err(fs_info, "invalid csum type, has %u want %u",
2581			  btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2582		goto out;
2583	}
2584	if (unlikely(btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP)) {
2585		ret = -EUCLEAN;
2586		btrfs_err(fs_info,
2587		"invalid incompat flags, has 0x%llx valid mask 0x%llx",
2588			  btrfs_super_incompat_flags(sb),
2589			  (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2590		goto out;
2591	}
2592out:
2593	if (ret < 0)
2594		btrfs_err(fs_info,
2595		"super block corruption detected before writing it to disk");
2596	return ret;
2597}
2598
2599static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2600{
2601	struct btrfs_tree_parent_check check = {
2602		.level = level,
2603		.transid = gen,
2604		.owner_root = btrfs_root_id(root)
2605	};
2606	int ret = 0;
2607
2608	root->node = read_tree_block(root->fs_info, bytenr, &check);
2609	if (IS_ERR(root->node)) {
2610		ret = PTR_ERR(root->node);
2611		root->node = NULL;
2612		return ret;
2613	}
2614	if (unlikely(!extent_buffer_uptodate(root->node))) {
2615		free_extent_buffer(root->node);
2616		root->node = NULL;
2617		return -EIO;
2618	}
2619
2620	btrfs_set_root_node(&root->root_item, root->node);
2621	root->commit_root = btrfs_root_node(root);
2622	btrfs_set_root_refs(&root->root_item, 1);
2623	return ret;
2624}
2625
2626static int load_important_roots(struct btrfs_fs_info *fs_info)
2627{
2628	struct btrfs_super_block *sb = fs_info->super_copy;
2629	u64 gen, bytenr;
2630	int level, ret;
2631
2632	bytenr = btrfs_super_root(sb);
2633	gen = btrfs_super_generation(sb);
2634	level = btrfs_super_root_level(sb);
2635	ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2636	if (ret) {
2637		btrfs_warn(fs_info, "couldn't read tree root");
2638		return ret;
2639	}
2640	return 0;
2641}
2642
2643static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2644{
2645	int backup_index = find_newest_super_backup(fs_info);
2646	struct btrfs_super_block *sb = fs_info->super_copy;
2647	struct btrfs_root *tree_root = fs_info->tree_root;
2648	bool handle_error = false;
2649	int ret = 0;
2650	int i;
2651
2652	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2653		if (handle_error) {
2654			if (!IS_ERR(tree_root->node))
2655				free_extent_buffer(tree_root->node);
2656			tree_root->node = NULL;
2657
2658			if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2659				break;
2660
2661			free_root_pointers(fs_info, 0);
2662
2663			/*
2664			 * Don't use the log in recovery mode, it won't be
2665			 * valid
2666			 */
2667			btrfs_set_super_log_root(sb, 0);
2668
2669			btrfs_warn(fs_info, "try to load backup roots slot %d", i);
2670			ret = read_backup_root(fs_info, i);
2671			backup_index = ret;
2672			if (ret < 0)
2673				return ret;
2674		}
2675
2676		ret = load_important_roots(fs_info);
2677		if (ret) {
2678			handle_error = true;
2679			continue;
2680		}
2681
2682		/*
2683		 * No need to hold btrfs_root::objectid_mutex since the fs
2684		 * hasn't been fully initialised and we are the only user
2685		 */
2686		ret = btrfs_init_root_free_objectid(tree_root);
2687		if (ret < 0) {
2688			handle_error = true;
2689			continue;
2690		}
2691
2692		ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2693
2694		ret = btrfs_read_roots(fs_info);
2695		if (ret < 0) {
2696			handle_error = true;
2697			continue;
2698		}
2699
2700		/* All successful */
2701		fs_info->generation = btrfs_header_generation(tree_root->node);
2702		btrfs_set_last_trans_committed(fs_info, fs_info->generation);
2703		fs_info->last_reloc_trans = 0;
2704
2705		/* Always begin writing backup roots after the one being used */
2706		if (backup_index < 0) {
2707			fs_info->backup_root_index = 0;
2708		} else {
2709			fs_info->backup_root_index = backup_index + 1;
2710			fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2711		}
2712		break;
2713	}
2714
2715	return ret;
2716}
2717
2718/*
2719 * Lockdep gets confused between our buffer_tree which requires IRQ locking because
2720 * we modify marks in the IRQ context, and our delayed inode xarray which doesn't
2721 * have these requirements. Use a class key so lockdep doesn't get them mixed up.
2722 */
2723static struct lock_class_key buffer_xa_class;
2724
2725void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2726{
2727	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2728
2729	/* Use the same flags as mapping->i_pages. */
2730	xa_init_flags(&fs_info->buffer_tree, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
2731	lockdep_set_class(&fs_info->buffer_tree.xa_lock, &buffer_xa_class);
2732
2733	INIT_LIST_HEAD(&fs_info->trans_list);
2734	INIT_LIST_HEAD(&fs_info->dead_roots);
2735	INIT_LIST_HEAD(&fs_info->delayed_iputs);
2736	INIT_LIST_HEAD(&fs_info->delalloc_roots);
2737	INIT_LIST_HEAD(&fs_info->caching_block_groups);
2738	spin_lock_init(&fs_info->delalloc_root_lock);
2739	spin_lock_init(&fs_info->trans_lock);
2740	spin_lock_init(&fs_info->fs_roots_radix_lock);
2741	spin_lock_init(&fs_info->delayed_iput_lock);
2742	spin_lock_init(&fs_info->defrag_inodes_lock);
2743	spin_lock_init(&fs_info->super_lock);
2744	spin_lock_init(&fs_info->unused_bgs_lock);
2745	spin_lock_init(&fs_info->treelog_bg_lock);
2746	spin_lock_init(&fs_info->zone_active_bgs_lock);
2747	spin_lock_init(&fs_info->relocation_bg_lock);
2748	rwlock_init(&fs_info->tree_mod_log_lock);
2749	rwlock_init(&fs_info->global_root_lock);
2750	mutex_init(&fs_info->unused_bg_unpin_mutex);
2751	mutex_init(&fs_info->reclaim_bgs_lock);
2752	mutex_init(&fs_info->reloc_mutex);
2753	mutex_init(&fs_info->delalloc_root_mutex);
2754	mutex_init(&fs_info->zoned_meta_io_lock);
2755	mutex_init(&fs_info->zoned_data_reloc_io_lock);
2756	seqlock_init(&fs_info->profiles_lock);
2757
2758	btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
2759	btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
2760	btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
2761	btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
2762	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_prep,
2763				     BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2764	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
2765				     BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2766	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
2767				     BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2768	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
2769				     BTRFS_LOCKDEP_TRANS_COMPLETED);
2770
2771	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2772	INIT_LIST_HEAD(&fs_info->space_info);
2773	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2774	INIT_LIST_HEAD(&fs_info->unused_bgs);
2775	INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2776	INIT_LIST_HEAD(&fs_info->zone_active_bgs);
2777#ifdef CONFIG_BTRFS_DEBUG
2778	INIT_LIST_HEAD(&fs_info->allocated_roots);
2779	INIT_LIST_HEAD(&fs_info->allocated_ebs);
2780	spin_lock_init(&fs_info->eb_leak_lock);
2781#endif
2782	fs_info->mapping_tree = RB_ROOT_CACHED;
2783	rwlock_init(&fs_info->mapping_tree_lock);
2784	btrfs_init_block_rsv(&fs_info->global_block_rsv,
2785			     BTRFS_BLOCK_RSV_GLOBAL);
2786	btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2787	btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2788	btrfs_init_block_rsv(&fs_info->treelog_rsv, BTRFS_BLOCK_RSV_TREELOG);
2789	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2790	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2791			     BTRFS_BLOCK_RSV_DELOPS);
2792	btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2793			     BTRFS_BLOCK_RSV_DELREFS);
2794
2795	atomic_set(&fs_info->async_delalloc_pages, 0);
2796	atomic_set(&fs_info->defrag_running, 0);
2797	atomic_set(&fs_info->nr_delayed_iputs, 0);
2798	atomic64_set(&fs_info->tree_mod_seq, 0);
2799	fs_info->global_root_tree = RB_ROOT;
2800	fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2801	fs_info->metadata_ratio = 0;
2802	fs_info->defrag_inodes = RB_ROOT;
2803	atomic64_set(&fs_info->free_chunk_space, 0);
2804	fs_info->tree_mod_log = RB_ROOT;
2805	fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2806	btrfs_init_ref_verify(fs_info);
2807
2808	fs_info->thread_pool_size = min_t(unsigned long,
2809					  num_online_cpus() + 2, 8);
2810
2811	INIT_LIST_HEAD(&fs_info->ordered_roots);
2812	spin_lock_init(&fs_info->ordered_root_lock);
2813
2814	btrfs_init_scrub(fs_info);
2815	btrfs_init_balance(fs_info);
2816	btrfs_init_async_reclaim_work(fs_info);
2817	btrfs_init_extent_map_shrinker_work(fs_info);
2818
2819	rwlock_init(&fs_info->block_group_cache_lock);
2820	fs_info->block_group_cache_tree = RB_ROOT_CACHED;
2821
2822	btrfs_extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2823				  IO_TREE_FS_EXCLUDED_EXTENTS);
2824
2825	mutex_init(&fs_info->ordered_operations_mutex);
2826	mutex_init(&fs_info->tree_log_mutex);
2827	mutex_init(&fs_info->chunk_mutex);
2828	mutex_init(&fs_info->transaction_kthread_mutex);
2829	mutex_init(&fs_info->cleaner_mutex);
2830	mutex_init(&fs_info->ro_block_group_mutex);
2831	init_rwsem(&fs_info->commit_root_sem);
2832	init_rwsem(&fs_info->cleanup_work_sem);
2833	init_rwsem(&fs_info->subvol_sem);
2834	sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2835
2836	btrfs_init_dev_replace_locks(fs_info);
2837	btrfs_init_qgroup(fs_info);
2838	btrfs_discard_init(fs_info);
2839
2840	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2841	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2842
2843	init_waitqueue_head(&fs_info->transaction_throttle);
2844	init_waitqueue_head(&fs_info->transaction_wait);
2845	init_waitqueue_head(&fs_info->transaction_blocked_wait);
2846	init_waitqueue_head(&fs_info->async_submit_wait);
2847	init_waitqueue_head(&fs_info->delayed_iputs_wait);
2848
2849	/* Usable values until the real ones are cached from the superblock */
2850	fs_info->nodesize = 4096;
2851	fs_info->sectorsize = 4096;
2852	fs_info->sectorsize_bits = ilog2(4096);
2853	fs_info->stripesize = 4096;
2854
2855	/* Default compress algorithm when user does -o compress */
2856	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2857
2858	fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
2859
2860	spin_lock_init(&fs_info->swapfile_pins_lock);
2861	fs_info->swapfile_pins = RB_ROOT;
2862
2863	fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2864	INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2865}
2866
2867static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2868{
2869	int ret;
2870
2871	fs_info->sb = sb;
2872	/* Temporary fixed values for block size until we read the superblock. */
2873	sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2874	sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2875
2876	ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2877	if (ret)
2878		return ret;
2879
2880	ret = percpu_counter_init(&fs_info->evictable_extent_maps, 0, GFP_KERNEL);
2881	if (ret)
2882		return ret;
2883
2884	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2885	if (ret)
2886		return ret;
2887
2888	ret = percpu_counter_init(&fs_info->stats_read_blocks, 0, GFP_KERNEL);
2889	if (ret)
2890		return ret;
2891
2892	fs_info->dirty_metadata_batch = PAGE_SIZE *
2893					(1 + ilog2(nr_cpu_ids));
2894
2895	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2896	if (ret)
2897		return ret;
2898
2899	ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2900			GFP_KERNEL);
2901	if (ret)
2902		return ret;
2903
2904	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2905					GFP_KERNEL);
2906	if (!fs_info->delayed_root)
2907		return -ENOMEM;
2908	btrfs_init_delayed_root(fs_info->delayed_root);
2909
2910	if (sb_rdonly(sb))
2911		set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2912	if (btrfs_test_opt(fs_info, IGNOREMETACSUMS))
2913		set_bit(BTRFS_FS_STATE_SKIP_META_CSUMS, &fs_info->fs_state);
2914
2915	return btrfs_alloc_stripe_hash_table(fs_info);
2916}
2917
2918static int btrfs_uuid_rescan_kthread(void *data)
2919{
2920	struct btrfs_fs_info *fs_info = data;
2921	int ret;
2922
2923	/*
2924	 * 1st step is to iterate through the existing UUID tree and
2925	 * to delete all entries that contain outdated data.
2926	 * 2nd step is to add all missing entries to the UUID tree.
2927	 */
2928	ret = btrfs_uuid_tree_iterate(fs_info);
2929	if (ret < 0) {
2930		if (ret != -EINTR)
2931			btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2932				   ret);
2933		up(&fs_info->uuid_tree_rescan_sem);
2934		return ret;
2935	}
2936	return btrfs_uuid_scan_kthread(data);
2937}
2938
2939static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2940{
2941	struct task_struct *task;
2942
2943	down(&fs_info->uuid_tree_rescan_sem);
2944	task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2945	if (IS_ERR(task)) {
2946		/* fs_info->update_uuid_tree_gen remains 0 in all error case */
2947		btrfs_warn(fs_info, "failed to start uuid_rescan task");
2948		up(&fs_info->uuid_tree_rescan_sem);
2949		return PTR_ERR(task);
2950	}
2951
2952	return 0;
2953}
2954
2955static int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2956{
2957	u64 root_objectid = 0;
2958	struct btrfs_root *gang[8];
2959	int ret = 0;
2960
2961	while (1) {
2962		unsigned int found;
2963
2964		spin_lock(&fs_info->fs_roots_radix_lock);
2965		found = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2966					     (void **)gang, root_objectid,
2967					     ARRAY_SIZE(gang));
2968		if (!found) {
2969			spin_unlock(&fs_info->fs_roots_radix_lock);
2970			break;
2971		}
2972		root_objectid = btrfs_root_id(gang[found - 1]) + 1;
2973
2974		for (int i = 0; i < found; i++) {
2975			/* Avoid to grab roots in dead_roots. */
2976			if (btrfs_root_refs(&gang[i]->root_item) == 0) {
2977				gang[i] = NULL;
2978				continue;
2979			}
2980			/* Grab all the search result for later use. */
2981			gang[i] = btrfs_grab_root(gang[i]);
2982		}
2983		spin_unlock(&fs_info->fs_roots_radix_lock);
2984
2985		for (int i = 0; i < found; i++) {
2986			if (!gang[i])
2987				continue;
2988			root_objectid = btrfs_root_id(gang[i]);
2989			/*
2990			 * Continue to release the remaining roots after the first
2991			 * error without cleanup and preserve the first error
2992			 * for the return.
2993			 */
2994			if (!ret)
2995				ret = btrfs_orphan_cleanup(gang[i]);
2996			btrfs_put_root(gang[i]);
2997		}
2998		if (ret)
2999			break;
3000
3001		root_objectid++;
3002	}
3003	return ret;
3004}
3005
3006/*
3007 * Mounting logic specific to read-write file systems. Shared by open_ctree
3008 * and btrfs_remount when remounting from read-only to read-write.
3009 */
3010int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3011{
3012	int ret;
3013	const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3014	bool rebuild_free_space_tree = false;
3015
3016	if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3017	    btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3018		if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
3019			btrfs_warn(fs_info,
3020				   "'clear_cache' option is ignored with extent tree v2");
3021		else
3022			rebuild_free_space_tree = true;
3023	} else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3024		   !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3025		btrfs_warn(fs_info, "free space tree is invalid");
3026		rebuild_free_space_tree = true;
3027	}
3028
3029	if (rebuild_free_space_tree) {
3030		btrfs_info(fs_info, "rebuilding free space tree");
3031		ret = btrfs_rebuild_free_space_tree(fs_info);
3032		if (ret) {
3033			btrfs_warn(fs_info,
3034				   "failed to rebuild free space tree: %d", ret);
3035			goto out;
3036		}
3037	}
3038
3039	if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3040	    !btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
3041		btrfs_info(fs_info, "disabling free space tree");
3042		ret = btrfs_delete_free_space_tree(fs_info);
3043		if (ret) {
3044			btrfs_warn(fs_info,
3045				   "failed to disable free space tree: %d", ret);
3046			goto out;
3047		}
3048	}
3049
3050	/*
3051	 * btrfs_find_orphan_roots() is responsible for finding all the dead
3052	 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3053	 * them into the fs_info->fs_roots_radix tree. This must be done before
3054	 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3055	 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3056	 * item before the root's tree is deleted - this means that if we unmount
3057	 * or crash before the deletion completes, on the next mount we will not
3058	 * delete what remains of the tree because the orphan item does not
3059	 * exists anymore, which is what tells us we have a pending deletion.
3060	 */
3061	ret = btrfs_find_orphan_roots(fs_info);
3062	if (ret)
3063		goto out;
3064
3065	ret = btrfs_cleanup_fs_roots(fs_info);
3066	if (ret)
3067		goto out;
3068
3069	down_read(&fs_info->cleanup_work_sem);
3070	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3071	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3072		up_read(&fs_info->cleanup_work_sem);
3073		goto out;
3074	}
3075	up_read(&fs_info->cleanup_work_sem);
3076
3077	mutex_lock(&fs_info->cleaner_mutex);
3078	ret = btrfs_recover_relocation(fs_info);
3079	mutex_unlock(&fs_info->cleaner_mutex);
3080	if (ret < 0) {
3081		btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3082		goto out;
3083	}
3084
3085	if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3086	    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3087		btrfs_info(fs_info, "creating free space tree");
3088		ret = btrfs_create_free_space_tree(fs_info);
3089		if (ret) {
3090			btrfs_warn(fs_info,
3091				"failed to create free space tree: %d", ret);
3092			goto out;
3093		}
3094	}
3095
3096	if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3097		ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3098		if (ret)
3099			goto out;
3100	}
3101
3102	ret = btrfs_resume_balance_async(fs_info);
3103	if (ret)
3104		goto out;
3105
3106	ret = btrfs_resume_dev_replace_async(fs_info);
3107	if (ret) {
3108		btrfs_warn(fs_info, "failed to resume dev_replace");
3109		goto out;
3110	}
3111
3112	btrfs_qgroup_rescan_resume(fs_info);
3113
3114	if (!fs_info->uuid_root) {
3115		btrfs_info(fs_info, "creating UUID tree");
3116		ret = btrfs_create_uuid_tree(fs_info);
3117		if (ret) {
3118			btrfs_warn(fs_info,
3119				   "failed to create the UUID tree %d", ret);
3120			goto out;
3121		}
3122	}
3123
3124out:
3125	return ret;
3126}
3127
3128/*
3129 * Do various sanity and dependency checks of different features.
3130 *
3131 * @is_rw_mount:	If the mount is read-write.
3132 *
3133 * This is the place for less strict checks (like for subpage or artificial
3134 * feature dependencies).
3135 *
3136 * For strict checks or possible corruption detection, see
3137 * btrfs_validate_super().
3138 *
3139 * This should be called after btrfs_parse_options(), as some mount options
3140 * (space cache related) can modify on-disk format like free space tree and
3141 * screw up certain feature dependencies.
3142 */
3143int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3144{
3145	struct btrfs_super_block *disk_super = fs_info->super_copy;
3146	u64 incompat = btrfs_super_incompat_flags(disk_super);
3147	const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3148	const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3149
3150	if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3151		btrfs_err(fs_info,
3152		"cannot mount because of unknown incompat features (0x%llx)",
3153		    incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP);
3154		return -EINVAL;
3155	}
3156
3157	/* Runtime limitation for mixed block groups. */
3158	if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3159	    (fs_info->sectorsize != fs_info->nodesize)) {
3160		btrfs_err(fs_info,
3161"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3162			fs_info->nodesize, fs_info->sectorsize);
3163		return -EINVAL;
3164	}
3165
3166	/* Mixed backref is an always-enabled feature. */
3167	incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3168
3169	/* Set compression related flags just in case. */
3170	if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3171		incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3172	else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3173		incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3174
3175	/*
3176	 * An ancient flag, which should really be marked deprecated.
3177	 * Such runtime limitation doesn't really need a incompat flag.
3178	 */
3179	if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3180		incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3181
3182	if (compat_ro_unsupp && is_rw_mount) {
3183		btrfs_err(fs_info,
3184	"cannot mount read-write because of unknown compat_ro features (0x%llx)",
3185		       compat_ro_unsupp);
3186		return -EINVAL;
3187	}
3188
3189	/*
3190	 * We have unsupported RO compat features, although RO mounted, we
3191	 * should not cause any metadata writes, including log replay.
3192	 * Or we could screw up whatever the new feature requires.
3193	 */
3194	if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3195	    !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3196		btrfs_err(fs_info,
3197"cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3198			  compat_ro_unsupp);
3199		return -EINVAL;
3200	}
3201
3202	/*
3203	 * Artificial limitations for block group tree, to force
3204	 * block-group-tree to rely on no-holes and free-space-tree.
3205	 */
3206	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3207	    (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3208	     !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3209		btrfs_err(fs_info,
3210"block-group-tree feature requires no-holes and free-space-tree features");
3211		return -EINVAL;
3212	}
3213
3214	/*
3215	 * Subpage/bs > ps runtime limitation on v1 cache.
3216	 *
3217	 * V1 space cache still has some hard coded PAGE_SIZE usage, while
3218	 * we're already defaulting to v2 cache, no need to bother v1 as it's
3219	 * going to be deprecated anyway.
3220	 */
3221	if (fs_info->sectorsize != PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3222		btrfs_warn(fs_info,
3223	"v1 space cache is not supported for page size %lu with sectorsize %u",
3224			   PAGE_SIZE, fs_info->sectorsize);
3225		return -EINVAL;
3226	}
3227
3228	/* This can be called by remount, we need to protect the super block. */
3229	spin_lock(&fs_info->super_lock);
3230	btrfs_set_super_incompat_flags(disk_super, incompat);
3231	spin_unlock(&fs_info->super_lock);
3232
3233	return 0;
3234}
3235
3236static bool fs_is_full_ro(const struct btrfs_fs_info *fs_info)
3237{
3238	if (!sb_rdonly(fs_info->sb))
3239		return false;
3240	if (unlikely(fs_info->mount_opt & BTRFS_MOUNT_FULL_RO_MASK))
3241		return true;
3242	return false;
3243}
3244
3245int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices)
3246{
3247	u32 sectorsize;
3248	u32 nodesize;
3249	u32 stripesize;
3250	u64 generation;
3251	u16 csum_type;
3252	struct btrfs_super_block *disk_super;
3253	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3254	struct btrfs_root *tree_root;
3255	struct btrfs_root *chunk_root;
3256	int ret;
3257	int level;
3258
3259	ret = init_mount_fs_info(fs_info, sb);
3260	if (ret)
3261		goto fail;
3262
3263	/* These need to be init'ed before we start creating inodes and such. */
3264	tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3265				     GFP_KERNEL);
3266	fs_info->tree_root = tree_root;
3267	chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3268				      GFP_KERNEL);
3269	fs_info->chunk_root = chunk_root;
3270	if (!tree_root || !chunk_root) {
3271		ret = -ENOMEM;
3272		goto fail;
3273	}
3274
3275	ret = btrfs_init_btree_inode(sb);
3276	if (ret)
3277		goto fail;
3278
3279	invalidate_bdev(fs_devices->latest_dev->bdev);
3280
3281	/*
3282	 * Read super block and check the signature bytes only
3283	 */
3284	disk_super = btrfs_read_disk_super(fs_devices->latest_dev->bdev, 0, false);
3285	if (IS_ERR(disk_super)) {
3286		ret = PTR_ERR(disk_super);
3287		goto fail_alloc;
3288	}
3289
3290	btrfs_info(fs_info, "first mount of filesystem %pU", disk_super->fsid);
3291	/*
3292	 * Verify the type first, if that or the checksum value are
3293	 * corrupted, we'll find out
3294	 */
3295	csum_type = btrfs_super_csum_type(disk_super);
3296	if (!btrfs_supported_super_csum(csum_type)) {
3297		btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3298			  csum_type);
3299		ret = -EINVAL;
3300		btrfs_release_disk_super(disk_super);
3301		goto fail_alloc;
3302	}
3303
3304	fs_info->csum_size = btrfs_super_csum_size(disk_super);
3305
3306	ret = btrfs_init_csum_hash(fs_info, csum_type);
3307	if (ret) {
3308		btrfs_release_disk_super(disk_super);
3309		goto fail_alloc;
3310	}
3311
3312	/*
3313	 * We want to check superblock checksum, the type is stored inside.
3314	 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3315	 */
3316	if (btrfs_check_super_csum(fs_info, disk_super)) {
3317		btrfs_err(fs_info, "superblock checksum mismatch");
3318		ret = -EINVAL;
3319		btrfs_release_disk_super(disk_super);
3320		goto fail_alloc;
3321	}
3322
3323	/*
3324	 * super_copy is zeroed at allocation time and we never touch the
3325	 * following bytes up to INFO_SIZE, the checksum is calculated from
3326	 * the whole block of INFO_SIZE
3327	 */
3328	memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3329	btrfs_release_disk_super(disk_super);
3330
3331	disk_super = fs_info->super_copy;
3332
3333	memcpy(fs_info->super_for_commit, fs_info->super_copy,
3334	       sizeof(*fs_info->super_for_commit));
3335
3336	ret = btrfs_validate_mount_super(fs_info);
3337	if (ret) {
3338		btrfs_err(fs_info, "superblock contains fatal errors");
3339		ret = -EINVAL;
3340		goto fail_alloc;
3341	}
3342
3343	if (!btrfs_super_root(disk_super)) {
3344		btrfs_err(fs_info, "invalid superblock tree root bytenr");
3345		ret = -EINVAL;
3346		goto fail_alloc;
3347	}
3348
3349	/* check FS state, whether FS is broken. */
3350	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3351		WRITE_ONCE(fs_info->fs_error, -EUCLEAN);
3352
3353	/* If the fs has any rescue options, no transaction is allowed. */
3354	if (fs_is_full_ro(fs_info))
3355		WRITE_ONCE(fs_info->fs_error, -EROFS);
3356
3357	/* Set up fs_info before parsing mount options */
3358	nodesize = btrfs_super_nodesize(disk_super);
3359	sectorsize = btrfs_super_sectorsize(disk_super);
3360	stripesize = sectorsize;
3361	fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3362	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3363
3364	fs_info->nodesize = nodesize;
3365	fs_info->nodesize_bits = ilog2(nodesize);
3366	fs_info->sectorsize = sectorsize;
3367	fs_info->sectorsize_bits = ilog2(sectorsize);
3368	fs_info->block_min_order = ilog2(round_up(sectorsize, PAGE_SIZE) >> PAGE_SHIFT);
3369	fs_info->block_max_order = ilog2((BITS_PER_LONG << fs_info->sectorsize_bits) >> PAGE_SHIFT);
3370	fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3371	fs_info->stripesize = stripesize;
3372	fs_info->fs_devices->fs_info = fs_info;
3373
3374	if (fs_info->sectorsize > PAGE_SIZE)
3375		btrfs_warn(fs_info,
3376			   "support for block size %u with page size %lu is experimental, some features may be missing",
3377			   fs_info->sectorsize, PAGE_SIZE);
3378	/*
3379	 * Handle the space caching options appropriately now that we have the
3380	 * super block loaded and validated.
3381	 */
3382	btrfs_set_free_space_cache_settings(fs_info);
3383
3384	if (!btrfs_check_options(fs_info, &fs_info->mount_opt, sb->s_flags)) {
3385		ret = -EINVAL;
3386		goto fail_alloc;
3387	}
3388
3389	ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3390	if (ret < 0)
3391		goto fail_alloc;
3392
3393	/*
3394	 * At this point our mount options are validated, if we set ->max_inline
3395	 * to something non-standard make sure we truncate it to sectorsize.
3396	 */
3397	fs_info->max_inline = min_t(u64, fs_info->max_inline, fs_info->sectorsize);
3398
3399	ret = btrfs_alloc_compress_wsm(fs_info);
3400	if (ret)
3401		goto fail_sb_buffer;
3402	ret = btrfs_init_workqueues(fs_info);
3403	if (ret)
3404		goto fail_sb_buffer;
3405
3406	sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3407	sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3408
3409	/* Update the values for the current filesystem. */
3410	sb->s_blocksize = sectorsize;
3411	sb->s_blocksize_bits = blksize_bits(sectorsize);
3412	memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3413
3414	mutex_lock(&fs_info->chunk_mutex);
3415	ret = btrfs_read_sys_array(fs_info);
3416	mutex_unlock(&fs_info->chunk_mutex);
3417	if (ret) {
3418		btrfs_err(fs_info, "failed to read the system array: %d", ret);
3419		goto fail_sb_buffer;
3420	}
3421
3422	generation = btrfs_super_chunk_root_generation(disk_super);
3423	level = btrfs_super_chunk_root_level(disk_super);
3424	ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3425			      generation, level);
3426	if (ret) {
3427		btrfs_err(fs_info, "failed to read chunk root");
3428		goto fail_tree_roots;
3429	}
3430
3431	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3432			   offsetof(struct btrfs_header, chunk_tree_uuid),
3433			   BTRFS_UUID_SIZE);
3434
3435	ret = btrfs_read_chunk_tree(fs_info);
3436	if (ret) {
3437		btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3438		goto fail_tree_roots;
3439	}
3440
3441	/*
3442	 * At this point we know all the devices that make this filesystem,
3443	 * including the seed devices but we don't know yet if the replace
3444	 * target is required. So free devices that are not part of this
3445	 * filesystem but skip the replace target device which is checked
3446	 * below in btrfs_init_dev_replace().
3447	 */
3448	btrfs_free_extra_devids(fs_devices);
3449	if (unlikely(!fs_devices->latest_dev->bdev)) {
3450		btrfs_err(fs_info, "failed to read devices");
3451		ret = -EIO;
3452		goto fail_tree_roots;
3453	}
3454
3455	ret = init_tree_roots(fs_info);
3456	if (ret)
3457		goto fail_tree_roots;
3458
3459	/*
3460	 * Get zone type information of zoned block devices. This will also
3461	 * handle emulation of a zoned filesystem if a regular device has the
3462	 * zoned incompat feature flag set.
3463	 */
3464	ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3465	if (ret) {
3466		btrfs_err(fs_info,
3467			  "zoned: failed to read device zone info: %d", ret);
3468		goto fail_block_groups;
3469	}
3470
3471	/*
3472	 * If we have a uuid root and we're not being told to rescan we need to
3473	 * check the generation here so we can set the
3474	 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3475	 * transaction during a balance or the log replay without updating the
3476	 * uuid generation, and then if we crash we would rescan the uuid tree,
3477	 * even though it was perfectly fine.
3478	 */
3479	if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3480	    fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3481		set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3482
3483	if (unlikely(btrfs_verify_dev_items(fs_info))) {
3484		ret = -EUCLEAN;
3485		goto fail_block_groups;
3486	}
3487	ret = btrfs_verify_dev_extents(fs_info);
3488	if (ret) {
3489		btrfs_err(fs_info,
3490			  "failed to verify dev extents against chunks: %d",
3491			  ret);
3492		goto fail_block_groups;
3493	}
3494	ret = btrfs_recover_balance(fs_info);
3495	if (ret) {
3496		btrfs_err(fs_info, "failed to recover balance: %d", ret);
3497		goto fail_block_groups;
3498	}
3499
3500	ret = btrfs_init_dev_stats(fs_info);
3501	if (ret) {
3502		btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3503		goto fail_block_groups;
3504	}
3505
3506	ret = btrfs_init_dev_replace(fs_info);
3507	if (ret) {
3508		btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3509		goto fail_block_groups;
3510	}
3511
3512	ret = btrfs_check_zoned_mode(fs_info);
3513	if (ret) {
3514		btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3515			  ret);
3516		goto fail_block_groups;
3517	}
3518
3519	ret = btrfs_sysfs_add_fsid(fs_devices);
3520	if (ret) {
3521		btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3522				ret);
3523		goto fail_block_groups;
3524	}
3525
3526	ret = btrfs_sysfs_add_mounted(fs_info);
3527	if (ret) {
3528		btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3529		goto fail_fsdev_sysfs;
3530	}
3531
3532	ret = btrfs_init_space_info(fs_info);
3533	if (ret) {
3534		btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3535		goto fail_sysfs;
3536	}
3537
3538	ret = btrfs_read_block_groups(fs_info);
3539	if (ret) {
3540		btrfs_err(fs_info, "failed to read block groups: %d", ret);
3541		goto fail_sysfs;
3542	}
3543
3544	btrfs_zoned_reserve_data_reloc_bg(fs_info);
3545	btrfs_free_zone_cache(fs_info);
3546
3547	btrfs_check_active_zone_reservation(fs_info);
3548
3549	if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3550	    !btrfs_check_rw_degradable(fs_info, NULL)) {
3551		btrfs_warn(fs_info,
3552		"writable mount is not allowed due to too many missing devices");
3553		ret = -EINVAL;
3554		goto fail_sysfs;
3555	}
3556
3557	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3558					       "btrfs-cleaner");
3559	if (IS_ERR(fs_info->cleaner_kthread)) {
3560		ret = PTR_ERR(fs_info->cleaner_kthread);
3561		goto fail_sysfs;
3562	}
3563
3564	fs_info->transaction_kthread = kthread_run(transaction_kthread,
3565						   tree_root,
3566						   "btrfs-transaction");
3567	if (IS_ERR(fs_info->transaction_kthread)) {
3568		ret = PTR_ERR(fs_info->transaction_kthread);
3569		goto fail_cleaner;
3570	}
3571
3572	ret = btrfs_read_qgroup_config(fs_info);
3573	if (ret)
3574		goto fail_trans_kthread;
3575
3576	if (btrfs_build_ref_tree(fs_info))
3577		btrfs_err(fs_info, "couldn't build ref tree");
3578
3579	/* do not make disk changes in broken FS or nologreplay is given */
3580	if (btrfs_super_log_root(disk_super) != 0 &&
3581	    !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3582		btrfs_info(fs_info, "start tree-log replay");
3583		ret = btrfs_replay_log(fs_info, fs_devices);
3584		if (ret)
3585			goto fail_qgroup;
3586	}
3587
3588	fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3589	if (IS_ERR(fs_info->fs_root)) {
3590		ret = PTR_ERR(fs_info->fs_root);
3591		btrfs_warn(fs_info, "failed to read fs tree: %d", ret);
3592		fs_info->fs_root = NULL;
3593		goto fail_qgroup;
3594	}
3595
3596	if (sb_rdonly(sb))
3597		return 0;
3598
3599	ret = btrfs_start_pre_rw_mount(fs_info);
3600	if (ret) {
3601		close_ctree(fs_info);
3602		return ret;
3603	}
3604	btrfs_discard_resume(fs_info);
3605
3606	if (fs_info->uuid_root &&
3607	    (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3608	     fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3609		btrfs_info(fs_info, "checking UUID tree");
3610		ret = btrfs_check_uuid_tree(fs_info);
3611		if (ret) {
3612			btrfs_warn(fs_info,
3613				"failed to check the UUID tree: %d", ret);
3614			close_ctree(fs_info);
3615			return ret;
3616		}
3617	}
3618
3619	set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3620
3621	/* Kick the cleaner thread so it'll start deleting snapshots. */
3622	if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3623		wake_up_process(fs_info->cleaner_kthread);
3624
3625	return 0;
3626
3627fail_qgroup:
3628	btrfs_free_qgroup_config(fs_info);
3629fail_trans_kthread:
3630	kthread_stop(fs_info->transaction_kthread);
3631	btrfs_cleanup_transaction(fs_info);
3632	btrfs_free_fs_roots(fs_info);
3633fail_cleaner:
3634	kthread_stop(fs_info->cleaner_kthread);
3635
3636	/*
3637	 * make sure we're done with the btree inode before we stop our
3638	 * kthreads
3639	 */
3640	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3641
3642fail_sysfs:
3643	btrfs_sysfs_remove_mounted(fs_info);
3644
3645fail_fsdev_sysfs:
3646	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3647
3648fail_block_groups:
3649	btrfs_put_block_group_cache(fs_info);
3650
3651fail_tree_roots:
3652	if (fs_info->data_reloc_root)
3653		btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3654	free_root_pointers(fs_info, true);
3655	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3656
3657fail_sb_buffer:
3658	btrfs_stop_all_workers(fs_info);
3659	btrfs_free_block_groups(fs_info);
3660fail_alloc:
3661	btrfs_mapping_tree_free(fs_info);
3662
3663	iput(fs_info->btree_inode);
3664fail:
3665	ASSERT(ret < 0);
3666	return ret;
3667}
3668ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3669
3670static void btrfs_end_super_write(struct bio *bio)
3671{
3672	struct btrfs_device *device = bio->bi_private;
3673	struct folio_iter fi;
3674
3675	bio_for_each_folio_all(fi, bio) {
3676		if (bio->bi_status) {
3677			btrfs_warn_rl(device->fs_info,
3678				"lost super block write due to IO error on %s (%d)",
3679				btrfs_dev_name(device),
3680				blk_status_to_errno(bio->bi_status));
3681			btrfs_dev_stat_inc_and_print(device,
3682						     BTRFS_DEV_STAT_WRITE_ERRS);
3683			/* Ensure failure if the primary sb fails. */
3684			if (bio->bi_opf & REQ_FUA)
3685				atomic_add(BTRFS_SUPER_PRIMARY_WRITE_ERROR,
3686					   &device->sb_write_errors);
3687			else
3688				atomic_inc(&device->sb_write_errors);
3689		}
3690		folio_unlock(fi.folio);
3691		folio_put(fi.folio);
3692	}
3693
3694	bio_put(bio);
3695}
3696
3697/*
3698 * Write superblock @sb to the @device. Do not wait for completion, all the
3699 * folios we use for writing are locked.
3700 *
3701 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3702 * the expected device size at commit time. Note that max_mirrors must be
3703 * same for write and wait phases.
3704 *
3705 * Return number of errors when folio is not found or submission fails.
3706 */
3707static int write_dev_supers(struct btrfs_device *device,
3708			    struct btrfs_super_block *sb, int max_mirrors)
3709{
3710	struct btrfs_fs_info *fs_info = device->fs_info;
3711	struct address_space *mapping = device->bdev->bd_mapping;
3712	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3713	int i;
3714	int ret;
3715	u64 bytenr, bytenr_orig;
3716
3717	atomic_set(&device->sb_write_errors, 0);
3718
3719	if (max_mirrors == 0)
3720		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3721
3722	shash->tfm = fs_info->csum_shash;
3723
3724	for (i = 0; i < max_mirrors; i++) {
3725		struct folio *folio;
3726		struct bio *bio;
3727		struct btrfs_super_block *disk_super;
3728		size_t offset;
3729
3730		bytenr_orig = btrfs_sb_offset(i);
3731		ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3732		if (ret == -ENOENT) {
3733			continue;
3734		} else if (ret < 0) {
3735			btrfs_err(device->fs_info,
3736			  "couldn't get super block location for mirror %d error %d",
3737			  i, ret);
3738			atomic_inc(&device->sb_write_errors);
3739			continue;
3740		}
3741		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3742		    device->commit_total_bytes)
3743			break;
3744
3745		btrfs_set_super_bytenr(sb, bytenr_orig);
3746
3747		crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3748				    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3749				    sb->csum);
3750
3751		folio = __filemap_get_folio(mapping, bytenr >> PAGE_SHIFT,
3752					    FGP_LOCK | FGP_ACCESSED | FGP_CREAT,
3753					    GFP_NOFS);
3754		if (IS_ERR(folio)) {
3755			btrfs_err(device->fs_info,
3756			  "couldn't get super block page for bytenr %llu error %ld",
3757			  bytenr, PTR_ERR(folio));
3758			atomic_inc(&device->sb_write_errors);
3759			continue;
3760		}
3761
3762		offset = offset_in_folio(folio, bytenr);
3763		disk_super = folio_address(folio) + offset;
3764		memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3765
3766		/*
3767		 * Directly use bios here instead of relying on the page cache
3768		 * to do I/O, so we don't lose the ability to do integrity
3769		 * checking.
3770		 */
3771		bio = bio_alloc(device->bdev, 1,
3772				REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
3773				GFP_NOFS);
3774		bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3775		bio->bi_private = device;
3776		bio->bi_end_io = btrfs_end_super_write;
3777		bio_add_folio_nofail(bio, folio, BTRFS_SUPER_INFO_SIZE, offset);
3778
3779		/*
3780		 * We FUA only the first super block.  The others we allow to
3781		 * go down lazy and there's a short window where the on-disk
3782		 * copies might still contain the older version.
3783		 */
3784		if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3785			bio->bi_opf |= REQ_FUA;
3786		submit_bio(bio);
3787
3788		if (btrfs_advance_sb_log(device, i))
3789			atomic_inc(&device->sb_write_errors);
3790	}
3791	return atomic_read(&device->sb_write_errors) < i ? 0 : -1;
3792}
3793
3794/*
3795 * Wait for write completion of superblocks done by write_dev_supers,
3796 * @max_mirrors same for write and wait phases.
3797 *
3798 * Return -1 if primary super block write failed or when there were no super block
3799 * copies written. Otherwise 0.
3800 */
3801static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3802{
3803	int i;
3804	int errors = 0;
3805	bool primary_failed = false;
3806	int ret;
3807	u64 bytenr;
3808
3809	if (max_mirrors == 0)
3810		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3811
3812	for (i = 0; i < max_mirrors; i++) {
3813		struct folio *folio;
3814
3815		ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3816		if (ret == -ENOENT) {
3817			break;
3818		} else if (ret < 0) {
3819			errors++;
3820			if (i == 0)
3821				primary_failed = true;
3822			continue;
3823		}
3824		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3825		    device->commit_total_bytes)
3826			break;
3827
3828		folio = filemap_get_folio(device->bdev->bd_mapping,
3829					  bytenr >> PAGE_SHIFT);
3830		/* If the folio has been removed, then we know it completed. */
3831		if (IS_ERR(folio))
3832			continue;
3833
3834		/* Folio will be unlocked once the write completes. */
3835		folio_wait_locked(folio);
3836		folio_put(folio);
3837	}
3838
3839	errors += atomic_read(&device->sb_write_errors);
3840	if (errors >= BTRFS_SUPER_PRIMARY_WRITE_ERROR)
3841		primary_failed = true;
3842	if (primary_failed) {
3843		btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3844			  device->devid);
3845		return -1;
3846	}
3847
3848	return errors < i ? 0 : -1;
3849}
3850
3851/*
3852 * endio for the write_dev_flush, this will wake anyone waiting
3853 * for the barrier when it is done
3854 */
3855static void btrfs_end_empty_barrier(struct bio *bio)
3856{
3857	bio_uninit(bio);
3858	complete(bio->bi_private);
3859}
3860
3861/*
3862 * Submit a flush request to the device if it supports it. Error handling is
3863 * done in the waiting counterpart.
3864 */
3865static void write_dev_flush(struct btrfs_device *device)
3866{
3867	struct bio *bio = &device->flush_bio;
3868
3869	device->last_flush_error = BLK_STS_OK;
3870
3871	bio_init(bio, device->bdev, NULL, 0,
3872		 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
3873	bio->bi_end_io = btrfs_end_empty_barrier;
3874	init_completion(&device->flush_wait);
3875	bio->bi_private = &device->flush_wait;
3876	submit_bio(bio);
3877	set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3878}
3879
3880/*
3881 * If the flush bio has been submitted by write_dev_flush, wait for it.
3882 * Return true for any error, and false otherwise.
3883 */
3884static bool wait_dev_flush(struct btrfs_device *device)
3885{
3886	struct bio *bio = &device->flush_bio;
3887
3888	if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3889		return false;
3890
3891	wait_for_completion_io(&device->flush_wait);
3892
3893	if (bio->bi_status) {
3894		device->last_flush_error = bio->bi_status;
3895		btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_FLUSH_ERRS);
3896		return true;
3897	}
3898
3899	return false;
3900}
3901
3902/*
3903 * send an empty flush down to each device in parallel,
3904 * then wait for them
3905 */
3906static int barrier_all_devices(struct btrfs_fs_info *info)
3907{
3908	struct list_head *head;
3909	struct btrfs_device *dev;
3910	int errors_wait = 0;
3911
3912	lockdep_assert_held(&info->fs_devices->device_list_mutex);
3913	/* send down all the barriers */
3914	head = &info->fs_devices->devices;
3915	list_for_each_entry(dev, head, dev_list) {
3916		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3917			continue;
3918		if (!dev->bdev)
3919			continue;
3920		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3921		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3922			continue;
3923
3924		write_dev_flush(dev);
3925	}
3926
3927	/* wait for all the barriers */
3928	list_for_each_entry(dev, head, dev_list) {
3929		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3930			continue;
3931		if (!dev->bdev) {
3932			errors_wait++;
3933			continue;
3934		}
3935		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3936		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3937			continue;
3938
3939		if (wait_dev_flush(dev))
3940			errors_wait++;
3941	}
3942
3943	/*
3944	 * Checks last_flush_error of disks in order to determine the device
3945	 * state.
3946	 */
3947	if (unlikely(errors_wait && !btrfs_check_rw_degradable(info, NULL)))
3948		return -EIO;
3949
3950	return 0;
3951}
3952
3953int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3954{
3955	int raid_type;
3956	int min_tolerated = INT_MAX;
3957
3958	if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3959	    (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3960		min_tolerated = min_t(int, min_tolerated,
3961				    btrfs_raid_array[BTRFS_RAID_SINGLE].
3962				    tolerated_failures);
3963
3964	for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3965		if (raid_type == BTRFS_RAID_SINGLE)
3966			continue;
3967		if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3968			continue;
3969		min_tolerated = min_t(int, min_tolerated,
3970				    btrfs_raid_array[raid_type].
3971				    tolerated_failures);
3972	}
3973
3974	if (min_tolerated == INT_MAX) {
3975		btrfs_warn(NULL, "unknown raid flag: %llu", flags);
3976		min_tolerated = 0;
3977	}
3978
3979	return min_tolerated;
3980}
3981
3982int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3983{
3984	struct list_head *head;
3985	struct btrfs_device *dev;
3986	struct btrfs_super_block *sb;
3987	struct btrfs_dev_item *dev_item;
3988	int ret;
3989	int do_barriers;
3990	int max_errors;
3991	int total_errors = 0;
3992	u64 flags;
3993
3994	do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3995
3996	/*
3997	 * max_mirrors == 0 indicates we're from commit_transaction,
3998	 * not from fsync where the tree roots in fs_info have not
3999	 * been consistent on disk.
4000	 */
4001	if (max_mirrors == 0)
4002		backup_super_roots(fs_info);
4003
4004	sb = fs_info->super_for_commit;
4005	dev_item = &sb->dev_item;
4006
4007	mutex_lock(&fs_info->fs_devices->device_list_mutex);
4008	head = &fs_info->fs_devices->devices;
4009	max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4010
4011	if (do_barriers) {
4012		ret = barrier_all_devices(fs_info);
4013		if (ret) {
4014			mutex_unlock(
4015				&fs_info->fs_devices->device_list_mutex);
4016			btrfs_handle_fs_error(fs_info, ret,
4017					      "errors while submitting device barriers.");
4018			return ret;
4019		}
4020	}
4021
4022	list_for_each_entry(dev, head, dev_list) {
4023		if (!dev->bdev) {
4024			total_errors++;
4025			continue;
4026		}
4027		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4028		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4029			continue;
4030
4031		btrfs_set_stack_device_generation(dev_item, 0);
4032		btrfs_set_stack_device_type(dev_item, dev->type);
4033		btrfs_set_stack_device_id(dev_item, dev->devid);
4034		btrfs_set_stack_device_total_bytes(dev_item,
4035						   dev->commit_total_bytes);
4036		btrfs_set_stack_device_bytes_used(dev_item,
4037						  dev->commit_bytes_used);
4038		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4039		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4040		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4041		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4042		memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4043		       BTRFS_FSID_SIZE);
4044
4045		flags = btrfs_super_flags(sb);
4046		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4047
4048		ret = btrfs_validate_write_super(fs_info, sb);
4049		if (unlikely(ret < 0)) {
4050			mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4051			btrfs_handle_fs_error(fs_info, -EUCLEAN,
4052				"unexpected superblock corruption detected");
4053			return -EUCLEAN;
4054		}
4055
4056		ret = write_dev_supers(dev, sb, max_mirrors);
4057		if (ret)
4058			total_errors++;
4059	}
4060	if (unlikely(total_errors > max_errors)) {
4061		btrfs_err(fs_info, "%d errors while writing supers",
4062			  total_errors);
4063		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4064
4065		/* FUA is masked off if unsupported and can't be the reason */
4066		btrfs_handle_fs_error(fs_info, -EIO,
4067				      "%d errors while writing supers",
4068				      total_errors);
4069		return -EIO;
4070	}
4071
4072	total_errors = 0;
4073	list_for_each_entry(dev, head, dev_list) {
4074		if (!dev->bdev)
4075			continue;
4076		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4077		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4078			continue;
4079
4080		ret = wait_dev_supers(dev, max_mirrors);
4081		if (ret)
4082			total_errors++;
4083	}
4084	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4085	if (unlikely(total_errors > max_errors)) {
4086		btrfs_handle_fs_error(fs_info, -EIO,
4087				      "%d errors while writing supers",
4088				      total_errors);
4089		return -EIO;
4090	}
4091	return 0;
4092}
4093
4094/* Drop a fs root from the radix tree and free it. */
4095void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4096				  struct btrfs_root *root)
4097{
4098	bool drop_ref = false;
4099
4100	spin_lock(&fs_info->fs_roots_radix_lock);
4101	radix_tree_delete(&fs_info->fs_roots_radix,
4102			  (unsigned long)btrfs_root_id(root));
4103	if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4104		drop_ref = true;
4105	spin_unlock(&fs_info->fs_roots_radix_lock);
4106
4107	if (BTRFS_FS_ERROR(fs_info)) {
4108		ASSERT(root->log_root == NULL);
4109		if (root->reloc_root) {
4110			btrfs_put_root(root->reloc_root);
4111			root->reloc_root = NULL;
4112		}
4113	}
4114
4115	if (drop_ref)
4116		btrfs_put_root(root);
4117}
4118
4119int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4120{
4121	mutex_lock(&fs_info->cleaner_mutex);
4122	btrfs_run_delayed_iputs(fs_info);
4123	mutex_unlock(&fs_info->cleaner_mutex);
4124	wake_up_process(fs_info->cleaner_kthread);
4125
4126	/* wait until ongoing cleanup work done */
4127	down_write(&fs_info->cleanup_work_sem);
4128	up_write(&fs_info->cleanup_work_sem);
4129
4130	return btrfs_commit_current_transaction(fs_info->tree_root);
4131}
4132
4133static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4134{
4135	struct btrfs_transaction *trans;
4136	struct btrfs_transaction *tmp;
4137	bool found = false;
4138
4139	/*
4140	 * This function is only called at the very end of close_ctree(),
4141	 * thus no other running transaction, no need to take trans_lock.
4142	 */
4143	ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4144	list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4145		struct extent_state *cached = NULL;
4146		u64 dirty_bytes = 0;
4147		u64 cur = 0;
4148		u64 found_start;
4149		u64 found_end;
4150
4151		found = true;
4152		while (btrfs_find_first_extent_bit(&trans->dirty_pages, cur,
4153						   &found_start, &found_end,
4154						   EXTENT_DIRTY, &cached)) {
4155			dirty_bytes += found_end + 1 - found_start;
4156			cur = found_end + 1;
4157		}
4158		btrfs_warn(fs_info,
4159	"transaction %llu (with %llu dirty metadata bytes) is not committed",
4160			   trans->transid, dirty_bytes);
4161		btrfs_cleanup_one_transaction(trans);
4162
4163		if (trans == fs_info->running_transaction)
4164			fs_info->running_transaction = NULL;
4165		list_del_init(&trans->list);
4166
4167		btrfs_put_transaction(trans);
4168		trace_btrfs_transaction_commit(fs_info);
4169	}
4170	ASSERT(!found);
4171}
4172
4173void __cold close_ctree(struct btrfs_fs_info *fs_info)
4174{
4175	int ret;
4176
4177	set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4178
4179	/*
4180	 * If we had UNFINISHED_DROPS we could still be processing them, so
4181	 * clear that bit and wake up relocation so it can stop.
4182	 * We must do this before stopping the block group reclaim task, because
4183	 * at btrfs_relocate_block_group() we wait for this bit, and after the
4184	 * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4185	 * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4186	 * return 1.
4187	 */
4188	btrfs_wake_unfinished_drop(fs_info);
4189
4190	/*
4191	 * We may have the reclaim task running and relocating a data block group,
4192	 * in which case it may create delayed iputs. So stop it before we park
4193	 * the cleaner kthread otherwise we can get new delayed iputs after
4194	 * parking the cleaner, and that can make the async reclaim task to hang
4195	 * if it's waiting for delayed iputs to complete, since the cleaner is
4196	 * parked and can not run delayed iputs - this will make us hang when
4197	 * trying to stop the async reclaim task.
4198	 */
4199	cancel_work_sync(&fs_info->reclaim_bgs_work);
4200	/*
4201	 * We don't want the cleaner to start new transactions, add more delayed
4202	 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4203	 * because that frees the task_struct, and the transaction kthread might
4204	 * still try to wake up the cleaner.
4205	 */
4206	kthread_park(fs_info->cleaner_kthread);
4207
4208	/* wait for the qgroup rescan worker to stop */
4209	btrfs_qgroup_wait_for_completion(fs_info, false);
4210
4211	/* wait for the uuid_scan task to finish */
4212	down(&fs_info->uuid_tree_rescan_sem);
4213	/* avoid complains from lockdep et al., set sem back to initial state */
4214	up(&fs_info->uuid_tree_rescan_sem);
4215
4216	/* pause restriper - we want to resume on mount */
4217	btrfs_pause_balance(fs_info);
4218
4219	btrfs_dev_replace_suspend_for_unmount(fs_info);
4220
4221	btrfs_scrub_cancel(fs_info);
4222
4223	/* wait for any defraggers to finish */
4224	wait_event(fs_info->transaction_wait,
4225		   (atomic_read(&fs_info->defrag_running) == 0));
4226
4227	/* clear out the rbtree of defraggable inodes */
4228	btrfs_cleanup_defrag_inodes(fs_info);
4229
4230	/*
4231	 * Handle the error fs first, as it will flush and wait for all ordered
4232	 * extents.  This will generate delayed iputs, thus we want to handle
4233	 * it first.
4234	 */
4235	if (unlikely(BTRFS_FS_ERROR(fs_info)))
4236		btrfs_error_commit_super(fs_info);
4237
4238	/*
4239	 * Wait for any fixup workers to complete.
4240	 * If we don't wait for them here and they are still running by the time
4241	 * we call kthread_stop() against the cleaner kthread further below, we
4242	 * get an use-after-free on the cleaner because the fixup worker adds an
4243	 * inode to the list of delayed iputs and then attempts to wakeup the
4244	 * cleaner kthread, which was already stopped and destroyed. We parked
4245	 * already the cleaner, but below we run all pending delayed iputs.
4246	 */
4247	btrfs_flush_workqueue(fs_info->fixup_workers);
4248	/*
4249	 * Similar case here, we have to wait for delalloc workers before we
4250	 * proceed below and stop the cleaner kthread, otherwise we trigger a
4251	 * use-after-tree on the cleaner kthread task_struct when a delalloc
4252	 * worker running submit_compressed_extents() adds a delayed iput, which
4253	 * does a wake up on the cleaner kthread, which was already freed below
4254	 * when we call kthread_stop().
4255	 */
4256	btrfs_flush_workqueue(fs_info->delalloc_workers);
4257
4258	/*
4259	 * We can have ordered extents getting their last reference dropped from
4260	 * the fs_info->workers queue because for async writes for data bios we
4261	 * queue a work for that queue, at btrfs_wq_submit_bio(), that runs
4262	 * run_one_async_done() which calls btrfs_bio_end_io() in case the bio
4263	 * has an error, and that later function can do the final
4264	 * btrfs_put_ordered_extent() on the ordered extent attached to the bio,
4265	 * which adds a delayed iput for the inode. So we must flush the queue
4266	 * so that we don't have delayed iputs after committing the current
4267	 * transaction below and stopping the cleaner and transaction kthreads.
4268	 */
4269	btrfs_flush_workqueue(fs_info->workers);
4270
4271	/*
4272	 * When finishing a compressed write bio we schedule a work queue item
4273	 * to finish an ordered extent - end_bbio_compressed_write()
4274	 * calls btrfs_finish_ordered_extent() which in turns does a call to
4275	 * btrfs_queue_ordered_fn(), and that queues the ordered extent
4276	 * completion either in the endio_write_workers work queue or in the
4277	 * fs_info->endio_freespace_worker work queue. We flush those queues
4278	 * below, so before we flush them we must flush this queue for the
4279	 * workers of compressed writes.
4280	 */
4281	flush_workqueue(fs_info->endio_workers);
4282
4283	/*
4284	 * After we parked the cleaner kthread, ordered extents may have
4285	 * completed and created new delayed iputs. If one of the async reclaim
4286	 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4287	 * can hang forever trying to stop it, because if a delayed iput is
4288	 * added after it ran btrfs_run_delayed_iputs() and before it called
4289	 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4290	 * no one else to run iputs.
4291	 *
4292	 * So wait for all ongoing ordered extents to complete and then run
4293	 * delayed iputs. This works because once we reach this point no one
4294	 * can create new ordered extents, but delayed iputs can still be added
4295	 * by a reclaim worker (see comments further below).
4296	 *
4297	 * Also note that btrfs_wait_ordered_roots() is not safe here, because
4298	 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4299	 * but the delayed iput for the respective inode is made only when doing
4300	 * the final btrfs_put_ordered_extent() (which must happen at
4301	 * btrfs_finish_ordered_io() when we are unmounting).
4302	 */
4303	btrfs_flush_workqueue(fs_info->endio_write_workers);
4304	/* Ordered extents for free space inodes. */
4305	btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4306	/*
4307	 * Run delayed iputs in case an async reclaim worker is waiting for them
4308	 * to be run as mentioned above.
4309	 */
4310	btrfs_run_delayed_iputs(fs_info);
4311
4312	cancel_work_sync(&fs_info->async_reclaim_work);
4313	cancel_work_sync(&fs_info->async_data_reclaim_work);
4314	cancel_work_sync(&fs_info->preempt_reclaim_work);
4315	cancel_work_sync(&fs_info->em_shrinker_work);
4316
4317	/*
4318	 * Run delayed iputs again because an async reclaim worker may have
4319	 * added new ones if it was flushing delalloc:
4320	 *
4321	 * shrink_delalloc() -> btrfs_start_delalloc_roots() ->
4322	 *    start_delalloc_inodes() -> btrfs_add_delayed_iput()
4323	 */
4324	btrfs_run_delayed_iputs(fs_info);
4325
4326	/* There should be no more workload to generate new delayed iputs. */
4327	set_bit(BTRFS_FS_STATE_NO_DELAYED_IPUT, &fs_info->fs_state);
4328
4329	/* Cancel or finish ongoing discard work */
4330	btrfs_discard_cleanup(fs_info);
4331
4332	if (!sb_rdonly(fs_info->sb)) {
4333		/*
4334		 * The cleaner kthread is stopped, so do one final pass over
4335		 * unused block groups.
4336		 */
4337		btrfs_delete_unused_bgs(fs_info);
4338
4339		/*
4340		 * There might be existing delayed inode workers still running
4341		 * and holding an empty delayed inode item. We must wait for
4342		 * them to complete first because they can create a transaction.
4343		 * This happens when someone calls btrfs_balance_delayed_items()
4344		 * and then a transaction commit runs the same delayed nodes
4345		 * before any delayed worker has done something with the nodes.
4346		 * We must wait for any worker here and not at transaction
4347		 * commit time since that could cause a deadlock.
4348		 * This is a very rare case.
4349		 */
4350		btrfs_flush_workqueue(fs_info->delayed_workers);
4351
4352		ret = btrfs_commit_super(fs_info);
4353		if (ret)
4354			btrfs_err(fs_info, "commit super ret %d", ret);
4355	}
4356
4357	kthread_stop(fs_info->transaction_kthread);
4358	kthread_stop(fs_info->cleaner_kthread);
4359
4360	ASSERT(list_empty(&fs_info->delayed_iputs));
4361	set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4362
4363	if (btrfs_check_quota_leak(fs_info)) {
4364		DEBUG_WARN("qgroup reserved space leaked");
4365		btrfs_err(fs_info, "qgroup reserved space leaked");
4366	}
4367
4368	btrfs_free_qgroup_config(fs_info);
4369	ASSERT(list_empty(&fs_info->delalloc_roots));
4370
4371	if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4372		btrfs_info(fs_info, "at unmount delalloc count %lld",
4373		       percpu_counter_sum(&fs_info->delalloc_bytes));
4374	}
4375
4376	if (percpu_counter_sum(&fs_info->ordered_bytes))
4377		btrfs_info(fs_info, "at unmount dio bytes count %lld",
4378			   percpu_counter_sum(&fs_info->ordered_bytes));
4379
4380	btrfs_sysfs_remove_mounted(fs_info);
4381	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4382
4383	btrfs_put_block_group_cache(fs_info);
4384
4385	/*
4386	 * we must make sure there is not any read request to
4387	 * submit after we stopping all workers.
4388	 */
4389	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4390	btrfs_stop_all_workers(fs_info);
4391
4392	/* We shouldn't have any transaction open at this point */
4393	warn_about_uncommitted_trans(fs_info);
4394
4395	clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4396	free_root_pointers(fs_info, true);
4397	btrfs_free_fs_roots(fs_info);
4398
4399	/*
4400	 * We must free the block groups after dropping the fs_roots as we could
4401	 * have had an IO error and have left over tree log blocks that aren't
4402	 * cleaned up until the fs roots are freed.  This makes the block group
4403	 * accounting appear to be wrong because there's pending reserved bytes,
4404	 * so make sure we do the block group cleanup afterwards.
4405	 */
4406	btrfs_free_block_groups(fs_info);
4407
4408	iput(fs_info->btree_inode);
4409
4410	btrfs_mapping_tree_free(fs_info);
4411}
4412
4413void btrfs_mark_buffer_dirty(struct btrfs_trans_handle *trans,
4414			     struct extent_buffer *buf)
4415{
4416	struct btrfs_fs_info *fs_info = buf->fs_info;
4417	u64 transid = btrfs_header_generation(buf);
4418
4419#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4420	/*
4421	 * This is a fast path so only do this check if we have sanity tests
4422	 * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4423	 * outside of the sanity tests.
4424	 */
4425	if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4426		return;
4427#endif
4428	/* This is an active transaction (its state < TRANS_STATE_UNBLOCKED). */
4429	ASSERT(trans->transid == fs_info->generation);
4430	btrfs_assert_tree_write_locked(buf);
4431	if (unlikely(transid != fs_info->generation)) {
4432		btrfs_abort_transaction(trans, -EUCLEAN);
4433		btrfs_crit(fs_info,
4434"dirty buffer transid mismatch, logical %llu found transid %llu running transid %llu",
4435			   buf->start, transid, fs_info->generation);
4436	}
4437	set_extent_buffer_dirty(buf);
4438}
4439
4440static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4441					int flush_delayed)
4442{
4443	/*
4444	 * looks as though older kernels can get into trouble with
4445	 * this code, they end up stuck in balance_dirty_pages forever
4446	 */
4447	int ret;
4448
4449	if (current->flags & PF_MEMALLOC)
4450		return;
4451
4452	if (flush_delayed)
4453		btrfs_balance_delayed_items(fs_info);
4454
4455	ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4456				     BTRFS_DIRTY_METADATA_THRESH,
4457				     fs_info->dirty_metadata_batch);
4458	if (ret > 0) {
4459		balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4460	}
4461}
4462
4463void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4464{
4465	__btrfs_btree_balance_dirty(fs_info, 1);
4466}
4467
4468void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4469{
4470	__btrfs_btree_balance_dirty(fs_info, 0);
4471}
4472
4473static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4474{
4475	/* cleanup FS via transaction */
4476	btrfs_cleanup_transaction(fs_info);
4477
4478	down_write(&fs_info->cleanup_work_sem);
4479	up_write(&fs_info->cleanup_work_sem);
4480}
4481
4482static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4483{
4484	struct btrfs_root *gang[8];
4485	u64 root_objectid = 0;
4486	int ret;
4487
4488	spin_lock(&fs_info->fs_roots_radix_lock);
4489	while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4490					     (void **)gang, root_objectid,
4491					     ARRAY_SIZE(gang))) != 0) {
4492		int i;
4493
4494		for (i = 0; i < ret; i++)
4495			gang[i] = btrfs_grab_root(gang[i]);
4496		spin_unlock(&fs_info->fs_roots_radix_lock);
4497
4498		for (i = 0; i < ret; i++) {
4499			if (!gang[i])
4500				continue;
4501			root_objectid = btrfs_root_id(gang[i]);
4502			btrfs_free_log(NULL, gang[i]);
4503			btrfs_put_root(gang[i]);
4504		}
4505		root_objectid++;
4506		spin_lock(&fs_info->fs_roots_radix_lock);
4507	}
4508	spin_unlock(&fs_info->fs_roots_radix_lock);
4509	btrfs_free_log_root_tree(NULL, fs_info);
4510}
4511
4512static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4513{
4514	struct btrfs_ordered_extent *ordered;
4515
4516	spin_lock(&root->ordered_extent_lock);
4517	/*
4518	 * This will just short circuit the ordered completion stuff which will
4519	 * make sure the ordered extent gets properly cleaned up.
4520	 */
4521	list_for_each_entry(ordered, &root->ordered_extents,
4522			    root_extent_list)
4523		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4524	spin_unlock(&root->ordered_extent_lock);
4525}
4526
4527static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4528{
4529	struct btrfs_root *root;
4530	LIST_HEAD(splice);
4531
4532	spin_lock(&fs_info->ordered_root_lock);
4533	list_splice_init(&fs_info->ordered_roots, &splice);
4534	while (!list_empty(&splice)) {
4535		root = list_first_entry(&splice, struct btrfs_root,
4536					ordered_root);
4537		list_move_tail(&root->ordered_root,
4538			       &fs_info->ordered_roots);
4539
4540		spin_unlock(&fs_info->ordered_root_lock);
4541		btrfs_destroy_ordered_extents(root);
4542
4543		cond_resched();
4544		spin_lock(&fs_info->ordered_root_lock);
4545	}
4546	spin_unlock(&fs_info->ordered_root_lock);
4547
4548	/*
4549	 * We need this here because if we've been flipped read-only we won't
4550	 * get sync() from the umount, so we need to make sure any ordered
4551	 * extents that haven't had their dirty pages IO start writeout yet
4552	 * actually get run and error out properly.
4553	 */
4554	btrfs_wait_ordered_roots(fs_info, U64_MAX, NULL);
4555}
4556
4557static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4558{
4559	struct btrfs_inode *btrfs_inode;
4560	LIST_HEAD(splice);
4561
4562	spin_lock(&root->delalloc_lock);
4563	list_splice_init(&root->delalloc_inodes, &splice);
4564
4565	while (!list_empty(&splice)) {
4566		struct inode *inode = NULL;
4567		btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4568					       delalloc_inodes);
4569		btrfs_del_delalloc_inode(btrfs_inode);
4570		spin_unlock(&root->delalloc_lock);
4571
4572		/*
4573		 * Make sure we get a live inode and that it'll not disappear
4574		 * meanwhile.
4575		 */
4576		inode = igrab(&btrfs_inode->vfs_inode);
4577		if (inode) {
4578			unsigned int nofs_flag;
4579
4580			nofs_flag = memalloc_nofs_save();
4581			invalidate_inode_pages2(inode->i_mapping);
4582			memalloc_nofs_restore(nofs_flag);
4583			iput(inode);
4584		}
4585		spin_lock(&root->delalloc_lock);
4586	}
4587	spin_unlock(&root->delalloc_lock);
4588}
4589
4590static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4591{
4592	struct btrfs_root *root;
4593	LIST_HEAD(splice);
4594
4595	spin_lock(&fs_info->delalloc_root_lock);
4596	list_splice_init(&fs_info->delalloc_roots, &splice);
4597	while (!list_empty(&splice)) {
4598		root = list_first_entry(&splice, struct btrfs_root,
4599					 delalloc_root);
4600		root = btrfs_grab_root(root);
4601		BUG_ON(!root);
4602		spin_unlock(&fs_info->delalloc_root_lock);
4603
4604		btrfs_destroy_delalloc_inodes(root);
4605		btrfs_put_root(root);
4606
4607		spin_lock(&fs_info->delalloc_root_lock);
4608	}
4609	spin_unlock(&fs_info->delalloc_root_lock);
4610}
4611
4612static void btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4613					 struct extent_io_tree *dirty_pages,
4614					 int mark)
4615{
4616	struct extent_buffer *eb;
4617	u64 start = 0;
4618	u64 end;
4619
4620	while (btrfs_find_first_extent_bit(dirty_pages, start, &start, &end,
4621					   mark, NULL)) {
4622		btrfs_clear_extent_bit(dirty_pages, start, end, mark, NULL);
4623		while (start <= end) {
4624			eb = find_extent_buffer(fs_info, start);
4625			start += fs_info->nodesize;
4626			if (!eb)
4627				continue;
4628
4629			btrfs_tree_lock(eb);
4630			wait_on_extent_buffer_writeback(eb);
4631			btrfs_clear_buffer_dirty(NULL, eb);
4632			btrfs_tree_unlock(eb);
4633
4634			free_extent_buffer_stale(eb);
4635		}
4636	}
4637}
4638
4639static void btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4640					struct extent_io_tree *unpin)
4641{
4642	u64 start;
4643	u64 end;
4644
4645	while (1) {
4646		struct extent_state *cached_state = NULL;
4647
4648		/*
4649		 * The btrfs_finish_extent_commit() may get the same range as
4650		 * ours between find_first_extent_bit and clear_extent_dirty.
4651		 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4652		 * the same extent range.
4653		 */
4654		mutex_lock(&fs_info->unused_bg_unpin_mutex);
4655		if (!btrfs_find_first_extent_bit(unpin, 0, &start, &end,
4656						 EXTENT_DIRTY, &cached_state)) {
4657			mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4658			break;
4659		}
4660
4661		btrfs_clear_extent_dirty(unpin, start, end, &cached_state);
4662		btrfs_free_extent_state(cached_state);
4663		btrfs_error_unpin_extent_range(fs_info, start, end);
4664		mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4665		cond_resched();
4666	}
4667}
4668
4669static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4670{
4671	struct inode *inode;
4672
4673	inode = cache->io_ctl.inode;
4674	if (inode) {
4675		unsigned int nofs_flag;
4676
4677		nofs_flag = memalloc_nofs_save();
4678		invalidate_inode_pages2(inode->i_mapping);
4679		memalloc_nofs_restore(nofs_flag);
4680
4681		BTRFS_I(inode)->generation = 0;
4682		cache->io_ctl.inode = NULL;
4683		iput(inode);
4684	}
4685	ASSERT(cache->io_ctl.pages == NULL);
4686	btrfs_put_block_group(cache);
4687}
4688
4689void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4690			     struct btrfs_fs_info *fs_info)
4691{
4692	struct btrfs_block_group *cache;
4693
4694	spin_lock(&cur_trans->dirty_bgs_lock);
4695	while (!list_empty(&cur_trans->dirty_bgs)) {
4696		cache = list_first_entry(&cur_trans->dirty_bgs,
4697					 struct btrfs_block_group,
4698					 dirty_list);
4699
4700		if (!list_empty(&cache->io_list)) {
4701			spin_unlock(&cur_trans->dirty_bgs_lock);
4702			list_del_init(&cache->io_list);
4703			btrfs_cleanup_bg_io(cache);
4704			spin_lock(&cur_trans->dirty_bgs_lock);
4705		}
4706
4707		list_del_init(&cache->dirty_list);
4708		spin_lock(&cache->lock);
4709		cache->disk_cache_state = BTRFS_DC_ERROR;
4710		spin_unlock(&cache->lock);
4711
4712		spin_unlock(&cur_trans->dirty_bgs_lock);
4713		btrfs_put_block_group(cache);
4714		btrfs_dec_delayed_refs_rsv_bg_updates(fs_info);
4715		spin_lock(&cur_trans->dirty_bgs_lock);
4716	}
4717	spin_unlock(&cur_trans->dirty_bgs_lock);
4718
4719	/*
4720	 * Refer to the definition of io_bgs member for details why it's safe
4721	 * to use it without any locking
4722	 */
4723	while (!list_empty(&cur_trans->io_bgs)) {
4724		cache = list_first_entry(&cur_trans->io_bgs,
4725					 struct btrfs_block_group,
4726					 io_list);
4727
4728		list_del_init(&cache->io_list);
4729		spin_lock(&cache->lock);
4730		cache->disk_cache_state = BTRFS_DC_ERROR;
4731		spin_unlock(&cache->lock);
4732		btrfs_cleanup_bg_io(cache);
4733	}
4734}
4735
4736static void btrfs_free_all_qgroup_pertrans(struct btrfs_fs_info *fs_info)
4737{
4738	struct btrfs_root *gang[8];
4739	int i;
4740	int ret;
4741
4742	spin_lock(&fs_info->fs_roots_radix_lock);
4743	while (1) {
4744		ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
4745						 (void **)gang, 0,
4746						 ARRAY_SIZE(gang),
4747						 BTRFS_ROOT_TRANS_TAG);
4748		if (ret == 0)
4749			break;
4750		for (i = 0; i < ret; i++) {
4751			struct btrfs_root *root = gang[i];
4752
4753			btrfs_qgroup_free_meta_all_pertrans(root);
4754			radix_tree_tag_clear(&fs_info->fs_roots_radix,
4755					(unsigned long)btrfs_root_id(root),
4756					BTRFS_ROOT_TRANS_TAG);
4757		}
4758	}
4759	spin_unlock(&fs_info->fs_roots_radix_lock);
4760}
4761
4762void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans)
4763{
4764	struct btrfs_fs_info *fs_info = cur_trans->fs_info;
4765	struct btrfs_device *dev, *tmp;
4766
4767	btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4768	ASSERT(list_empty(&cur_trans->dirty_bgs));
4769	ASSERT(list_empty(&cur_trans->io_bgs));
4770
4771	list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4772				 post_commit_list) {
4773		list_del_init(&dev->post_commit_list);
4774	}
4775
4776	btrfs_destroy_delayed_refs(cur_trans);
4777
4778	cur_trans->state = TRANS_STATE_COMMIT_START;
4779	wake_up(&fs_info->transaction_blocked_wait);
4780
4781	cur_trans->state = TRANS_STATE_UNBLOCKED;
4782	wake_up(&fs_info->transaction_wait);
4783
4784	btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4785				     EXTENT_DIRTY);
4786	btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4787
4788	cur_trans->state =TRANS_STATE_COMPLETED;
4789	wake_up(&cur_trans->commit_wait);
4790}
4791
4792static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4793{
4794	struct btrfs_transaction *t;
4795
4796	mutex_lock(&fs_info->transaction_kthread_mutex);
4797
4798	spin_lock(&fs_info->trans_lock);
4799	while (!list_empty(&fs_info->trans_list)) {
4800		t = list_first_entry(&fs_info->trans_list,
4801				     struct btrfs_transaction, list);
4802		if (t->state >= TRANS_STATE_COMMIT_PREP) {
4803			refcount_inc(&t->use_count);
4804			spin_unlock(&fs_info->trans_lock);
4805			btrfs_wait_for_commit(fs_info, t->transid);
4806			btrfs_put_transaction(t);
4807			spin_lock(&fs_info->trans_lock);
4808			continue;
4809		}
4810		if (t == fs_info->running_transaction) {
4811			t->state = TRANS_STATE_COMMIT_DOING;
4812			spin_unlock(&fs_info->trans_lock);
4813			/*
4814			 * We wait for 0 num_writers since we don't hold a trans
4815			 * handle open currently for this transaction.
4816			 */
4817			wait_event(t->writer_wait,
4818				   atomic_read(&t->num_writers) == 0);
4819		} else {
4820			spin_unlock(&fs_info->trans_lock);
4821		}
4822		btrfs_cleanup_one_transaction(t);
4823
4824		spin_lock(&fs_info->trans_lock);
4825		if (t == fs_info->running_transaction)
4826			fs_info->running_transaction = NULL;
4827		list_del_init(&t->list);
4828		spin_unlock(&fs_info->trans_lock);
4829
4830		btrfs_put_transaction(t);
4831		trace_btrfs_transaction_commit(fs_info);
4832		spin_lock(&fs_info->trans_lock);
4833	}
4834	spin_unlock(&fs_info->trans_lock);
4835	btrfs_destroy_all_ordered_extents(fs_info);
4836	btrfs_destroy_delayed_inodes(fs_info);
4837	btrfs_assert_delayed_root_empty(fs_info);
4838	btrfs_destroy_all_delalloc_inodes(fs_info);
4839	btrfs_drop_all_logs(fs_info);
4840	btrfs_free_all_qgroup_pertrans(fs_info);
4841	mutex_unlock(&fs_info->transaction_kthread_mutex);
4842
4843	return 0;
4844}
4845
4846int btrfs_init_root_free_objectid(struct btrfs_root *root)
4847{
4848	BTRFS_PATH_AUTO_FREE(path);
4849	int ret;
4850	struct extent_buffer *l;
4851	struct btrfs_key search_key;
4852	struct btrfs_key found_key;
4853	int slot;
4854
4855	path = btrfs_alloc_path();
4856	if (!path)
4857		return -ENOMEM;
4858
4859	search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4860	search_key.type = -1;
4861	search_key.offset = (u64)-1;
4862	ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4863	if (ret < 0)
4864		return ret;
4865	if (unlikely(ret == 0)) {
4866		/*
4867		 * Key with offset -1 found, there would have to exist a root
4868		 * with such id, but this is out of valid range.
4869		 */
4870		return -EUCLEAN;
4871	}
4872	if (path->slots[0] > 0) {
4873		slot = path->slots[0] - 1;
4874		l = path->nodes[0];
4875		btrfs_item_key_to_cpu(l, &found_key, slot);
4876		root->free_objectid = max_t(u64, found_key.objectid + 1,
4877					    BTRFS_FIRST_FREE_OBJECTID);
4878	} else {
4879		root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4880	}
4881
4882	return 0;
4883}
4884
4885int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4886{
4887	int ret;
4888	mutex_lock(&root->objectid_mutex);
4889
4890	if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4891		btrfs_warn(root->fs_info,
4892			   "the objectid of root %llu reaches its highest value",
4893			   btrfs_root_id(root));
4894		ret = -ENOSPC;
4895		goto out;
4896	}
4897
4898	*objectid = root->free_objectid++;
4899	ret = 0;
4900out:
4901	mutex_unlock(&root->objectid_mutex);
4902	return ret;
4903}