Linux Audio

Check our new training course

Loading...
v5.4
   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/buffer_head.h>
  11#include <linux/workqueue.h>
  12#include <linux/kthread.h>
  13#include <linux/slab.h>
  14#include <linux/migrate.h>
  15#include <linux/ratelimit.h>
  16#include <linux/uuid.h>
  17#include <linux/semaphore.h>
  18#include <linux/error-injection.h>
  19#include <linux/crc32c.h>
  20#include <linux/sched/mm.h>
  21#include <asm/unaligned.h>
  22#include <crypto/hash.h>
  23#include "ctree.h"
  24#include "disk-io.h"
  25#include "transaction.h"
  26#include "btrfs_inode.h"
  27#include "volumes.h"
  28#include "print-tree.h"
  29#include "locking.h"
  30#include "tree-log.h"
  31#include "free-space-cache.h"
  32#include "free-space-tree.h"
  33#include "inode-map.h"
  34#include "check-integrity.h"
  35#include "rcu-string.h"
  36#include "dev-replace.h"
  37#include "raid56.h"
  38#include "sysfs.h"
  39#include "qgroup.h"
  40#include "compression.h"
  41#include "tree-checker.h"
  42#include "ref-verify.h"
  43#include "block-group.h"
 
 
 
 
 
 
 
 
 
 
 
 
 
  44
  45#define BTRFS_SUPER_FLAG_SUPP	(BTRFS_HEADER_FLAG_WRITTEN |\
  46				 BTRFS_HEADER_FLAG_RELOC |\
  47				 BTRFS_SUPER_FLAG_ERROR |\
  48				 BTRFS_SUPER_FLAG_SEEDING |\
  49				 BTRFS_SUPER_FLAG_METADUMP |\
  50				 BTRFS_SUPER_FLAG_METADUMP_V2)
  51
  52static const struct extent_io_ops btree_extent_io_ops;
  53static void end_workqueue_fn(struct btrfs_work *work);
  54static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  55static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  56				      struct btrfs_fs_info *fs_info);
  57static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  58static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
  59					struct extent_io_tree *dirty_pages,
  60					int mark);
  61static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
  62				       struct extent_io_tree *pinned_extents);
  63static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
  64static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
  65
  66/*
  67 * btrfs_end_io_wq structs are used to do processing in task context when an IO
  68 * is complete.  This is used during reads to verify checksums, and it is used
  69 * by writes to insert metadata for new file extents after IO is complete.
  70 */
  71struct btrfs_end_io_wq {
  72	struct bio *bio;
  73	bio_end_io_t *end_io;
  74	void *private;
  75	struct btrfs_fs_info *info;
  76	blk_status_t status;
  77	enum btrfs_wq_endio_type metadata;
  78	struct btrfs_work work;
  79};
  80
  81static struct kmem_cache *btrfs_end_io_wq_cache;
  82
  83int __init btrfs_end_io_wq_init(void)
  84{
  85	btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
  86					sizeof(struct btrfs_end_io_wq),
  87					0,
  88					SLAB_MEM_SPREAD,
  89					NULL);
  90	if (!btrfs_end_io_wq_cache)
  91		return -ENOMEM;
  92	return 0;
  93}
  94
  95void __cold btrfs_end_io_wq_exit(void)
  96{
  97	kmem_cache_destroy(btrfs_end_io_wq_cache);
  98}
  99
 100/*
 101 * async submit bios are used to offload expensive checksumming
 102 * onto the worker threads.  They checksum file and metadata bios
 103 * just before they are sent down the IO stack.
 104 */
 105struct async_submit_bio {
 106	void *private_data;
 107	struct bio *bio;
 108	extent_submit_bio_start_t *submit_bio_start;
 109	int mirror_num;
 110	/*
 111	 * bio_offset is optional, can be used if the pages in the bio
 112	 * can't tell us where in the file the bio should go
 113	 */
 114	u64 bio_offset;
 115	struct btrfs_work work;
 116	blk_status_t status;
 117};
 118
 119/*
 120 * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 121 * eb, the lockdep key is determined by the btrfs_root it belongs to and
 122 * the level the eb occupies in the tree.
 123 *
 124 * Different roots are used for different purposes and may nest inside each
 125 * other and they require separate keysets.  As lockdep keys should be
 126 * static, assign keysets according to the purpose of the root as indicated
 127 * by btrfs_root->root_key.objectid.  This ensures that all special purpose
 128 * roots have separate keysets.
 129 *
 130 * Lock-nesting across peer nodes is always done with the immediate parent
 131 * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 132 * subclass to avoid triggering lockdep warning in such cases.
 133 *
 134 * The key is set by the readpage_end_io_hook after the buffer has passed
 135 * csum validation but before the pages are unlocked.  It is also set by
 136 * btrfs_init_new_buffer on freshly allocated blocks.
 137 *
 138 * We also add a check to make sure the highest level of the tree is the
 139 * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 140 * needs update as well.
 141 */
 142#ifdef CONFIG_DEBUG_LOCK_ALLOC
 143# if BTRFS_MAX_LEVEL != 8
 144#  error
 145# endif
 146
 147static struct btrfs_lockdep_keyset {
 148	u64			id;		/* root objectid */
 149	const char		*name_stem;	/* lock name stem */
 150	char			names[BTRFS_MAX_LEVEL + 1][20];
 151	struct lock_class_key	keys[BTRFS_MAX_LEVEL + 1];
 152} btrfs_lockdep_keysets[] = {
 153	{ .id = BTRFS_ROOT_TREE_OBJECTID,	.name_stem = "root"	},
 154	{ .id = BTRFS_EXTENT_TREE_OBJECTID,	.name_stem = "extent"	},
 155	{ .id = BTRFS_CHUNK_TREE_OBJECTID,	.name_stem = "chunk"	},
 156	{ .id = BTRFS_DEV_TREE_OBJECTID,	.name_stem = "dev"	},
 157	{ .id = BTRFS_FS_TREE_OBJECTID,		.name_stem = "fs"	},
 158	{ .id = BTRFS_CSUM_TREE_OBJECTID,	.name_stem = "csum"	},
 159	{ .id = BTRFS_QUOTA_TREE_OBJECTID,	.name_stem = "quota"	},
 160	{ .id = BTRFS_TREE_LOG_OBJECTID,	.name_stem = "log"	},
 161	{ .id = BTRFS_TREE_RELOC_OBJECTID,	.name_stem = "treloc"	},
 162	{ .id = BTRFS_DATA_RELOC_TREE_OBJECTID,	.name_stem = "dreloc"	},
 163	{ .id = BTRFS_UUID_TREE_OBJECTID,	.name_stem = "uuid"	},
 164	{ .id = BTRFS_FREE_SPACE_TREE_OBJECTID,	.name_stem = "free-space" },
 165	{ .id = 0,				.name_stem = "tree"	},
 166};
 167
 168void __init btrfs_init_lockdep(void)
 169{
 170	int i, j;
 171
 172	/* initialize lockdep class names */
 173	for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 174		struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 175
 176		for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 177			snprintf(ks->names[j], sizeof(ks->names[j]),
 178				 "btrfs-%s-%02d", ks->name_stem, j);
 179	}
 180}
 181
 182void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 183				    int level)
 184{
 185	struct btrfs_lockdep_keyset *ks;
 186
 187	BUG_ON(level >= ARRAY_SIZE(ks->keys));
 188
 189	/* find the matching keyset, id 0 is the default entry */
 190	for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 191		if (ks->id == objectid)
 192			break;
 193
 194	lockdep_set_class_and_name(&eb->lock,
 195				   &ks->keys[level], ks->names[level]);
 196}
 197
 198#endif
 199
 200/*
 201 * extents on the btree inode are pretty simple, there's one extent
 202 * that covers the entire device
 203 */
 204struct extent_map *btree_get_extent(struct btrfs_inode *inode,
 205		struct page *page, size_t pg_offset, u64 start, u64 len,
 206		int create)
 207{
 208	struct btrfs_fs_info *fs_info = inode->root->fs_info;
 209	struct extent_map_tree *em_tree = &inode->extent_tree;
 210	struct extent_map *em;
 211	int ret;
 212
 213	read_lock(&em_tree->lock);
 214	em = lookup_extent_mapping(em_tree, start, len);
 215	if (em) {
 216		em->bdev = fs_info->fs_devices->latest_bdev;
 217		read_unlock(&em_tree->lock);
 218		goto out;
 219	}
 220	read_unlock(&em_tree->lock);
 221
 222	em = alloc_extent_map();
 223	if (!em) {
 224		em = ERR_PTR(-ENOMEM);
 225		goto out;
 226	}
 227	em->start = 0;
 228	em->len = (u64)-1;
 229	em->block_len = (u64)-1;
 230	em->block_start = 0;
 231	em->bdev = fs_info->fs_devices->latest_bdev;
 232
 233	write_lock(&em_tree->lock);
 234	ret = add_extent_mapping(em_tree, em, 0);
 235	if (ret == -EEXIST) {
 236		free_extent_map(em);
 237		em = lookup_extent_mapping(em_tree, start, len);
 238		if (!em)
 239			em = ERR_PTR(-EIO);
 240	} else if (ret) {
 241		free_extent_map(em);
 242		em = ERR_PTR(ret);
 243	}
 244	write_unlock(&em_tree->lock);
 245
 246out:
 247	return em;
 248}
 249
 250/*
 251 * Compute the csum of a btree block and store the result to provided buffer.
 252 *
 253 * Returns error if the extent buffer cannot be mapped.
 254 */
 255static int csum_tree_block(struct extent_buffer *buf, u8 *result)
 256{
 257	struct btrfs_fs_info *fs_info = buf->fs_info;
 
 
 258	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 259	unsigned long len;
 260	unsigned long cur_len;
 261	unsigned long offset = BTRFS_CSUM_SIZE;
 262	char *kaddr;
 263	unsigned long map_start;
 264	unsigned long map_len;
 265	int err;
 266
 267	shash->tfm = fs_info->csum_shash;
 268	crypto_shash_init(shash);
 269
 270	len = buf->len - offset;
 
 
 
 
 
 
 
 
 
 271
 272	while (len > 0) {
 273		/*
 274		 * Note: we don't need to check for the err == 1 case here, as
 275		 * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
 276		 * and 'min_len = 32' and the currently implemented mapping
 277		 * algorithm we cannot cross a page boundary.
 278		 */
 279		err = map_private_extent_buffer(buf, offset, 32,
 280					&kaddr, &map_start, &map_len);
 281		if (WARN_ON(err))
 282			return err;
 283		cur_len = min(len, map_len - (offset - map_start));
 284		crypto_shash_update(shash, kaddr + offset - map_start, cur_len);
 285		len -= cur_len;
 286		offset += cur_len;
 287	}
 288	memset(result, 0, BTRFS_CSUM_SIZE);
 289
 290	crypto_shash_final(shash, result);
 291
 292	return 0;
 293}
 294
 295/*
 296 * we can't consider a given block up to date unless the transid of the
 297 * block matches the transid in the parent node's pointer.  This is how we
 298 * detect blocks that either didn't get written at all or got written
 299 * in the wrong place.
 300 */
 301static int verify_parent_transid(struct extent_io_tree *io_tree,
 302				 struct extent_buffer *eb, u64 parent_transid,
 303				 int atomic)
 304{
 305	struct extent_state *cached_state = NULL;
 306	int ret;
 307	bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 308
 309	if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 310		return 0;
 311
 312	if (atomic)
 313		return -EAGAIN;
 314
 315	if (need_lock) {
 316		btrfs_tree_read_lock(eb);
 317		btrfs_set_lock_blocking_read(eb);
 318	}
 319
 320	lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 321			 &cached_state);
 322	if (extent_buffer_uptodate(eb) &&
 323	    btrfs_header_generation(eb) == parent_transid) {
 324		ret = 0;
 325		goto out;
 326	}
 327	btrfs_err_rl(eb->fs_info,
 328		"parent transid verify failed on %llu wanted %llu found %llu",
 329			eb->start,
 330			parent_transid, btrfs_header_generation(eb));
 331	ret = 1;
 332
 333	/*
 334	 * Things reading via commit roots that don't have normal protection,
 335	 * like send, can have a really old block in cache that may point at a
 336	 * block that has been freed and re-allocated.  So don't clear uptodate
 337	 * if we find an eb that is under IO (dirty/writeback) because we could
 338	 * end up reading in the stale data and then writing it back out and
 339	 * making everybody very sad.
 340	 */
 341	if (!extent_buffer_under_io(eb))
 342		clear_extent_buffer_uptodate(eb);
 343out:
 344	unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 345			     &cached_state);
 346	if (need_lock)
 347		btrfs_tree_read_unlock_blocking(eb);
 348	return ret;
 349}
 350
 351static bool btrfs_supported_super_csum(u16 csum_type)
 352{
 353	switch (csum_type) {
 354	case BTRFS_CSUM_TYPE_CRC32:
 
 
 
 355		return true;
 356	default:
 357		return false;
 358	}
 359}
 360
 361/*
 362 * Return 0 if the superblock checksum type matches the checksum value of that
 363 * algorithm. Pass the raw disk superblock data.
 364 */
 365static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 366				  char *raw_disk_sb)
 367{
 368	struct btrfs_super_block *disk_sb =
 369		(struct btrfs_super_block *)raw_disk_sb;
 370	char result[BTRFS_CSUM_SIZE];
 371	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 372
 373	shash->tfm = fs_info->csum_shash;
 374	crypto_shash_init(shash);
 375
 376	/*
 377	 * The super_block structure does not span the whole
 378	 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
 379	 * filled with zeros and is included in the checksum.
 380	 */
 381	crypto_shash_update(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
 382			    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 383	crypto_shash_final(shash, result);
 384
 385	if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb)))
 386		return 1;
 387
 388	return 0;
 389}
 390
 391int btrfs_verify_level_key(struct extent_buffer *eb, int level,
 392			   struct btrfs_key *first_key, u64 parent_transid)
 393{
 394	struct btrfs_fs_info *fs_info = eb->fs_info;
 395	int found_level;
 396	struct btrfs_key found_key;
 397	int ret;
 398
 399	found_level = btrfs_header_level(eb);
 400	if (found_level != level) {
 401		WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
 402		     KERN_ERR "BTRFS: tree level check failed\n");
 403		btrfs_err(fs_info,
 404"tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
 405			  eb->start, level, found_level);
 406		return -EIO;
 407	}
 408
 409	if (!first_key)
 410		return 0;
 411
 412	/*
 413	 * For live tree block (new tree blocks in current transaction),
 414	 * we need proper lock context to avoid race, which is impossible here.
 415	 * So we only checks tree blocks which is read from disk, whose
 416	 * generation <= fs_info->last_trans_committed.
 417	 */
 418	if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
 419		return 0;
 420
 421	/* We have @first_key, so this @eb must have at least one item */
 422	if (btrfs_header_nritems(eb) == 0) {
 423		btrfs_err(fs_info,
 424		"invalid tree nritems, bytenr=%llu nritems=0 expect >0",
 425			  eb->start);
 426		WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
 427		return -EUCLEAN;
 
 
 
 
 
 428	}
 429
 430	if (found_level)
 431		btrfs_node_key_to_cpu(eb, &found_key, 0);
 432	else
 433		btrfs_item_key_to_cpu(eb, &found_key, 0);
 434	ret = btrfs_comp_cpu_keys(first_key, &found_key);
 435
 436	if (ret) {
 437		WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
 438		     KERN_ERR "BTRFS: tree first key check failed\n");
 439		btrfs_err(fs_info,
 440"tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
 441			  eb->start, parent_transid, first_key->objectid,
 442			  first_key->type, first_key->offset,
 443			  found_key.objectid, found_key.type,
 444			  found_key.offset);
 445	}
 446	return ret;
 447}
 448
 449/*
 450 * helper to read a given tree block, doing retries as required when
 451 * the checksums don't match and we have alternate mirrors to try.
 452 *
 453 * @parent_transid:	expected transid, skip check if 0
 454 * @level:		expected level, mandatory check
 455 * @first_key:		expected key of first slot, skip check if NULL
 456 */
 457static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
 458					  u64 parent_transid, int level,
 459					  struct btrfs_key *first_key)
 460{
 461	struct btrfs_fs_info *fs_info = eb->fs_info;
 462	struct extent_io_tree *io_tree;
 463	int failed = 0;
 464	int ret;
 465	int num_copies = 0;
 466	int mirror_num = 0;
 467	int failed_mirror = 0;
 468
 469	io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
 
 470	while (1) {
 471		clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 472		ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
 473		if (!ret) {
 474			if (verify_parent_transid(io_tree, eb,
 475						   parent_transid, 0))
 476				ret = -EIO;
 477			else if (btrfs_verify_level_key(eb, level,
 478						first_key, parent_transid))
 479				ret = -EUCLEAN;
 480			else
 481				break;
 482		}
 483
 484		num_copies = btrfs_num_copies(fs_info,
 485					      eb->start, eb->len);
 486		if (num_copies == 1)
 487			break;
 488
 489		if (!failed_mirror) {
 490			failed = 1;
 491			failed_mirror = eb->read_mirror;
 492		}
 493
 494		mirror_num++;
 495		if (mirror_num == failed_mirror)
 496			mirror_num++;
 497
 498		if (mirror_num > num_copies)
 499			break;
 500	}
 501
 502	if (failed && !ret && failed_mirror)
 503		btrfs_repair_eb_io_failure(eb, failed_mirror);
 504
 505	return ret;
 506}
 507
 508/*
 509 * checksum a dirty tree block before IO.  This has extra checks to make sure
 510 * we only fill in the checksum field in the first page of a multi-page block
 511 */
 512
 513static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 514{
 515	u64 start = page_offset(page);
 516	u64 found_start;
 
 
 517	u8 result[BTRFS_CSUM_SIZE];
 518	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 519	struct extent_buffer *eb;
 520	int ret;
 521
 522	eb = (struct extent_buffer *)page->private;
 523	if (page != eb->pages[0])
 524		return 0;
 
 
 525
 526	found_start = btrfs_header_bytenr(eb);
 527	/*
 528	 * Please do not consolidate these warnings into a single if.
 529	 * It is useful to know what went wrong.
 
 530	 */
 531	if (WARN_ON(found_start != start))
 532		return -EUCLEAN;
 533	if (WARN_ON(!PageUptodate(page)))
 534		return -EUCLEAN;
 535
 536	ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
 537			btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
 
 
 
 538
 539	if (csum_tree_block(eb, result))
 540		return -EINVAL;
 
 
 541
 542	if (btrfs_header_level(eb))
 543		ret = btrfs_check_node(eb);
 544	else
 545		ret = btrfs_check_leaf_full(eb);
 546
 547	if (ret < 0) {
 548		btrfs_err(fs_info,
 549		"block=%llu write time tree block corruption detected",
 550			  eb->start);
 551		return ret;
 552	}
 553	write_extent_buffer(eb, result, 0, csum_size);
 554
 555	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 556}
 557
 558static int check_tree_block_fsid(struct extent_buffer *eb)
 559{
 560	struct btrfs_fs_info *fs_info = eb->fs_info;
 561	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 562	u8 fsid[BTRFS_FSID_SIZE];
 563	int ret = 1;
 564
 565	read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 566	while (fs_devices) {
 567		u8 *metadata_uuid;
 568
 569		/*
 570		 * Checking the incompat flag is only valid for the current
 571		 * fs. For seed devices it's forbidden to have their uuid
 572		 * changed so reading ->fsid in this case is fine
 573		 */
 574		if (fs_devices == fs_info->fs_devices &&
 575		    btrfs_fs_incompat(fs_info, METADATA_UUID))
 576			metadata_uuid = fs_devices->metadata_uuid;
 577		else
 578			metadata_uuid = fs_devices->fsid;
 579
 580		if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
 581			ret = 0;
 582			break;
 583		}
 584		fs_devices = fs_devices->seed;
 585	}
 586	return ret;
 587}
 588
 589static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 590				      u64 phy_offset, struct page *page,
 591				      u64 start, u64 end, int mirror)
 592{
 
 593	u64 found_start;
 594	int found_level;
 595	struct extent_buffer *eb;
 596	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 597	struct btrfs_fs_info *fs_info = root->fs_info;
 598	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 599	int ret = 0;
 600	u8 result[BTRFS_CSUM_SIZE];
 601	int reads_done;
 602
 603	if (!page->private)
 604		goto out;
 605
 606	eb = (struct extent_buffer *)page->private;
 607
 608	/* the pending IO might have been the only thing that kept this buffer
 609	 * in memory.  Make sure we have a ref for all this other checks
 610	 */
 611	extent_buffer_get(eb);
 612
 613	reads_done = atomic_dec_and_test(&eb->io_pages);
 614	if (!reads_done)
 615		goto err;
 616
 617	eb->read_mirror = mirror;
 618	if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 619		ret = -EIO;
 620		goto err;
 621	}
 622
 623	found_start = btrfs_header_bytenr(eb);
 624	if (found_start != eb->start) {
 625		btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
 626			     eb->start, found_start);
 
 627		ret = -EIO;
 628		goto err;
 629	}
 630	if (check_tree_block_fsid(eb)) {
 631		btrfs_err_rl(fs_info, "bad fsid on block %llu",
 632			     eb->start);
 633		ret = -EIO;
 634		goto err;
 635	}
 636	found_level = btrfs_header_level(eb);
 637	if (found_level >= BTRFS_MAX_LEVEL) {
 638		btrfs_err(fs_info, "bad tree block level %d on %llu",
 639			  (int)btrfs_header_level(eb), eb->start);
 
 640		ret = -EIO;
 641		goto err;
 642	}
 643
 644	btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 645				       eb, found_level);
 646
 647	ret = csum_tree_block(eb, result);
 648	if (ret)
 649		goto err;
 650
 651	if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
 652		u32 val;
 653		u32 found = 0;
 654
 655		memcpy(&found, result, csum_size);
 656
 657		read_extent_buffer(eb, &val, 0, csum_size);
 658		btrfs_warn_rl(fs_info,
 659		"%s checksum verify failed on %llu wanted %x found %x level %d",
 660			      fs_info->sb->s_id, eb->start,
 661			      val, found, btrfs_header_level(eb));
 
 
 662		ret = -EUCLEAN;
 663		goto err;
 664	}
 665
 666	/*
 667	 * If this is a leaf block and it is corrupt, set the corrupt bit so
 668	 * that we don't try and read the other copies of this block, just
 669	 * return -EIO.
 670	 */
 671	if (found_level == 0 && btrfs_check_leaf_full(eb)) {
 672		set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 673		ret = -EIO;
 
 674	}
 675
 676	if (found_level > 0 && btrfs_check_node(eb))
 
 
 
 
 677		ret = -EIO;
 678
 679	if (!ret)
 680		set_extent_buffer_uptodate(eb);
 681	else
 682		btrfs_err(fs_info,
 683			  "block=%llu read time tree block corruption detected",
 684			  eb->start);
 685err:
 686	if (reads_done &&
 687	    test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 688		btree_readahead_hook(eb, ret);
 689
 690	if (ret) {
 691		/*
 692		 * our io error hook is going to dec the io pages
 693		 * again, we have to make sure it has something
 694		 * to decrement
 695		 */
 696		atomic_inc(&eb->io_pages);
 697		clear_extent_buffer_uptodate(eb);
 698	}
 699	free_extent_buffer(eb);
 700out:
 701	return ret;
 702}
 703
 704static void end_workqueue_bio(struct bio *bio)
 705{
 706	struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 707	struct btrfs_fs_info *fs_info;
 708	struct btrfs_workqueue *wq;
 709	btrfs_work_func_t func;
 710
 711	fs_info = end_io_wq->info;
 712	end_io_wq->status = bio->bi_status;
 713
 714	if (bio_op(bio) == REQ_OP_WRITE) {
 715		if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 716			wq = fs_info->endio_meta_write_workers;
 717			func = btrfs_endio_meta_write_helper;
 718		} else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 719			wq = fs_info->endio_freespace_worker;
 720			func = btrfs_freespace_write_helper;
 721		} else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 722			wq = fs_info->endio_raid56_workers;
 723			func = btrfs_endio_raid56_helper;
 724		} else {
 725			wq = fs_info->endio_write_workers;
 726			func = btrfs_endio_write_helper;
 727		}
 728	} else {
 729		if (unlikely(end_io_wq->metadata ==
 730			     BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 731			wq = fs_info->endio_repair_workers;
 732			func = btrfs_endio_repair_helper;
 733		} else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 734			wq = fs_info->endio_raid56_workers;
 735			func = btrfs_endio_raid56_helper;
 736		} else if (end_io_wq->metadata) {
 737			wq = fs_info->endio_meta_workers;
 738			func = btrfs_endio_meta_helper;
 739		} else {
 740			wq = fs_info->endio_workers;
 741			func = btrfs_endio_helper;
 742		}
 743	}
 744
 745	btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 746	btrfs_queue_work(wq, &end_io_wq->work);
 747}
 748
 749blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 750			enum btrfs_wq_endio_type metadata)
 751{
 752	struct btrfs_end_io_wq *end_io_wq;
 753
 754	end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 755	if (!end_io_wq)
 756		return BLK_STS_RESOURCE;
 757
 758	end_io_wq->private = bio->bi_private;
 759	end_io_wq->end_io = bio->bi_end_io;
 760	end_io_wq->info = info;
 761	end_io_wq->status = 0;
 762	end_io_wq->bio = bio;
 763	end_io_wq->metadata = metadata;
 764
 765	bio->bi_private = end_io_wq;
 766	bio->bi_end_io = end_workqueue_bio;
 767	return 0;
 768}
 769
 770static void run_one_async_start(struct btrfs_work *work)
 771{
 772	struct async_submit_bio *async;
 773	blk_status_t ret;
 774
 775	async = container_of(work, struct  async_submit_bio, work);
 776	ret = async->submit_bio_start(async->private_data, async->bio,
 777				      async->bio_offset);
 778	if (ret)
 779		async->status = ret;
 780}
 781
 782/*
 783 * In order to insert checksums into the metadata in large chunks, we wait
 784 * until bio submission time.   All the pages in the bio are checksummed and
 785 * sums are attached onto the ordered extent record.
 786 *
 787 * At IO completion time the csums attached on the ordered extent record are
 788 * inserted into the tree.
 789 */
 790static void run_one_async_done(struct btrfs_work *work)
 791{
 792	struct async_submit_bio *async;
 793	struct inode *inode;
 794	blk_status_t ret;
 795
 796	async = container_of(work, struct  async_submit_bio, work);
 797	inode = async->private_data;
 798
 799	/* If an error occurred we just want to clean up the bio and move on */
 800	if (async->status) {
 801		async->bio->bi_status = async->status;
 802		bio_endio(async->bio);
 803		return;
 804	}
 805
 806	ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio,
 807			async->mirror_num, 1);
 808	if (ret) {
 809		async->bio->bi_status = ret;
 810		bio_endio(async->bio);
 811	}
 812}
 813
 814static void run_one_async_free(struct btrfs_work *work)
 815{
 816	struct async_submit_bio *async;
 817
 818	async = container_of(work, struct  async_submit_bio, work);
 819	kfree(async);
 820}
 821
 822blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
 823				 int mirror_num, unsigned long bio_flags,
 824				 u64 bio_offset, void *private_data,
 825				 extent_submit_bio_start_t *submit_bio_start)
 826{
 827	struct async_submit_bio *async;
 828
 829	async = kmalloc(sizeof(*async), GFP_NOFS);
 830	if (!async)
 831		return BLK_STS_RESOURCE;
 832
 833	async->private_data = private_data;
 834	async->bio = bio;
 835	async->mirror_num = mirror_num;
 836	async->submit_bio_start = submit_bio_start;
 837
 838	btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 839			run_one_async_done, run_one_async_free);
 840
 841	async->bio_offset = bio_offset;
 842
 843	async->status = 0;
 844
 845	if (op_is_sync(bio->bi_opf))
 846		btrfs_set_work_high_priority(&async->work);
 847
 848	btrfs_queue_work(fs_info->workers, &async->work);
 849	return 0;
 850}
 851
 852static blk_status_t btree_csum_one_bio(struct bio *bio)
 853{
 854	struct bio_vec *bvec;
 855	struct btrfs_root *root;
 856	int ret = 0;
 857	struct bvec_iter_all iter_all;
 858
 859	ASSERT(!bio_flagged(bio, BIO_CLONED));
 860	bio_for_each_segment_all(bvec, bio, iter_all) {
 861		root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 862		ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 863		if (ret)
 864			break;
 865	}
 866
 867	return errno_to_blk_status(ret);
 868}
 869
 870static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
 871					     u64 bio_offset)
 872{
 873	/*
 874	 * when we're called for a write, we're already in the async
 875	 * submission context.  Just jump into btrfs_map_bio
 
 876	 */
 877	return btree_csum_one_bio(bio);
 878}
 879
 880static int check_async_write(struct btrfs_fs_info *fs_info,
 881			     struct btrfs_inode *bi)
 882{
 883	if (atomic_read(&bi->sync_writers))
 884		return 0;
 885	if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
 886		return 0;
 887	return 1;
 888}
 889
 890static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
 891					  int mirror_num,
 892					  unsigned long bio_flags)
 893{
 894	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 895	int async = check_async_write(fs_info, BTRFS_I(inode));
 896	blk_status_t ret;
 897
 898	if (bio_op(bio) != REQ_OP_WRITE) {
 899		/*
 900		 * called for a read, do the setup so that checksum validation
 901		 * can happen in the async kernel threads
 902		 */
 903		ret = btrfs_bio_wq_end_io(fs_info, bio,
 904					  BTRFS_WQ_ENDIO_METADATA);
 905		if (ret)
 906			goto out_w_error;
 907		ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
 908	} else if (!async) {
 909		ret = btree_csum_one_bio(bio);
 910		if (ret)
 911			goto out_w_error;
 912		ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
 913	} else {
 914		/*
 915		 * kthread helpers are used to submit writes so that
 916		 * checksumming can happen in parallel across all CPUs
 917		 */
 918		ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
 919					  0, inode, btree_submit_bio_start);
 920	}
 921
 922	if (ret)
 923		goto out_w_error;
 924	return 0;
 925
 926out_w_error:
 927	bio->bi_status = ret;
 928	bio_endio(bio);
 
 
 929	return ret;
 930}
 931
 932#ifdef CONFIG_MIGRATION
 933static int btree_migratepage(struct address_space *mapping,
 934			struct page *newpage, struct page *page,
 935			enum migrate_mode mode)
 936{
 937	/*
 938	 * we can't safely write a btree page from here,
 939	 * we haven't done the locking hook
 940	 */
 941	if (PageDirty(page))
 942		return -EAGAIN;
 943	/*
 944	 * Buffers may be managed in a filesystem specific way.
 945	 * We must have no buffers or drop them.
 946	 */
 947	if (page_has_private(page) &&
 948	    !try_to_release_page(page, GFP_KERNEL))
 949		return -EAGAIN;
 950	return migrate_page(mapping, newpage, page, mode);
 951}
 
 
 952#endif
 953
 954
 955static int btree_writepages(struct address_space *mapping,
 956			    struct writeback_control *wbc)
 957{
 958	struct btrfs_fs_info *fs_info;
 959	int ret;
 960
 961	if (wbc->sync_mode == WB_SYNC_NONE) {
 962
 963		if (wbc->for_kupdate)
 964			return 0;
 965
 966		fs_info = BTRFS_I(mapping->host)->root->fs_info;
 967		/* this is a bit racy, but that's ok */
 968		ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 969					     BTRFS_DIRTY_METADATA_THRESH,
 970					     fs_info->dirty_metadata_batch);
 971		if (ret < 0)
 972			return 0;
 973	}
 974	return btree_write_cache_pages(mapping, wbc);
 975}
 976
 977static int btree_readpage(struct file *file, struct page *page)
 978{
 979	struct extent_io_tree *tree;
 980	tree = &BTRFS_I(page->mapping->host)->io_tree;
 981	return extent_read_full_page(tree, page, btree_get_extent, 0);
 982}
 983
 984static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 985{
 986	if (PageWriteback(page) || PageDirty(page))
 987		return 0;
 988
 989	return try_release_extent_buffer(page);
 990}
 991
 992static void btree_invalidatepage(struct page *page, unsigned int offset,
 993				 unsigned int length)
 994{
 995	struct extent_io_tree *tree;
 996	tree = &BTRFS_I(page->mapping->host)->io_tree;
 997	extent_invalidatepage(tree, page, offset);
 998	btree_releasepage(page, GFP_NOFS);
 999	if (PagePrivate(page)) {
1000		btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1001			   "page private not zero on page %llu",
1002			   (unsigned long long)page_offset(page));
1003		ClearPagePrivate(page);
1004		set_page_private(page, 0);
1005		put_page(page);
1006	}
1007}
1008
1009static int btree_set_page_dirty(struct page *page)
1010{
1011#ifdef DEBUG
 
 
 
 
 
 
1012	struct extent_buffer *eb;
 
 
1013
1014	BUG_ON(!PagePrivate(page));
1015	eb = (struct extent_buffer *)page->private;
1016	BUG_ON(!eb);
1017	BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1018	BUG_ON(!atomic_read(&eb->refs));
1019	btrfs_assert_tree_locked(eb);
1020#endif
1021	return __set_page_dirty_nobuffers(page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1022}
 
 
 
1023
1024static const struct address_space_operations btree_aops = {
1025	.readpage	= btree_readpage,
1026	.writepages	= btree_writepages,
1027	.releasepage	= btree_releasepage,
1028	.invalidatepage = btree_invalidatepage,
1029#ifdef CONFIG_MIGRATION
1030	.migratepage	= btree_migratepage,
1031#endif
1032	.set_page_dirty = btree_set_page_dirty,
1033};
1034
1035void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1036{
1037	struct extent_buffer *buf = NULL;
1038	int ret;
1039
1040	buf = btrfs_find_create_tree_block(fs_info, bytenr);
1041	if (IS_ERR(buf))
1042		return;
1043
1044	ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1045	if (ret < 0)
1046		free_extent_buffer_stale(buf);
1047	else
1048		free_extent_buffer(buf);
1049}
1050
1051struct extent_buffer *btrfs_find_create_tree_block(
1052						struct btrfs_fs_info *fs_info,
1053						u64 bytenr)
 
1054{
1055	if (btrfs_is_testing(fs_info))
1056		return alloc_test_extent_buffer(fs_info, bytenr);
1057	return alloc_extent_buffer(fs_info, bytenr);
1058}
1059
1060/*
1061 * Read tree block at logical address @bytenr and do variant basic but critical
1062 * verification.
1063 *
1064 * @parent_transid:	expected transid of this tree block, skip check if 0
1065 * @level:		expected level, mandatory check
1066 * @first_key:		expected key in slot 0, skip check if NULL
1067 */
1068struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1069				      u64 parent_transid, int level,
1070				      struct btrfs_key *first_key)
1071{
1072	struct extent_buffer *buf = NULL;
1073	int ret;
1074
1075	buf = btrfs_find_create_tree_block(fs_info, bytenr);
 
 
 
1076	if (IS_ERR(buf))
1077		return buf;
1078
1079	ret = btree_read_extent_buffer_pages(buf, parent_transid,
1080					     level, first_key);
1081	if (ret) {
1082		free_extent_buffer_stale(buf);
1083		return ERR_PTR(ret);
1084	}
1085	return buf;
1086
1087}
1088
1089void btrfs_clean_tree_block(struct extent_buffer *buf)
1090{
1091	struct btrfs_fs_info *fs_info = buf->fs_info;
1092	if (btrfs_header_generation(buf) ==
1093	    fs_info->running_transaction->transid) {
1094		btrfs_assert_tree_locked(buf);
1095
1096		if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1097			percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1098						 -buf->len,
1099						 fs_info->dirty_metadata_batch);
1100			/* ugh, clear_extent_buffer_dirty needs to lock the page */
1101			btrfs_set_lock_blocking_write(buf);
1102			clear_extent_buffer_dirty(buf);
1103		}
1104	}
1105}
1106
1107static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1108{
1109	struct btrfs_subvolume_writers *writers;
1110	int ret;
1111
1112	writers = kmalloc(sizeof(*writers), GFP_NOFS);
1113	if (!writers)
1114		return ERR_PTR(-ENOMEM);
1115
1116	ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1117	if (ret < 0) {
1118		kfree(writers);
1119		return ERR_PTR(ret);
1120	}
 
1121
1122	init_waitqueue_head(&writers->wait);
1123	return writers;
1124}
1125
1126static void
1127btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1128{
1129	percpu_counter_destroy(&writers->counter);
1130	kfree(writers);
1131}
1132
1133static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1134			 u64 objectid)
1135{
1136	bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
 
 
 
 
 
 
1137	root->node = NULL;
1138	root->commit_root = NULL;
1139	root->state = 0;
1140	root->orphan_cleanup_state = 0;
1141
1142	root->last_trans = 0;
1143	root->highest_objectid = 0;
1144	root->nr_delalloc_inodes = 0;
1145	root->nr_ordered_extents = 0;
1146	root->inode_tree = RB_ROOT;
1147	INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1148	root->block_rsv = NULL;
 
 
1149
1150	INIT_LIST_HEAD(&root->dirty_list);
1151	INIT_LIST_HEAD(&root->root_list);
1152	INIT_LIST_HEAD(&root->delalloc_inodes);
1153	INIT_LIST_HEAD(&root->delalloc_root);
1154	INIT_LIST_HEAD(&root->ordered_extents);
1155	INIT_LIST_HEAD(&root->ordered_root);
1156	INIT_LIST_HEAD(&root->reloc_dirty_list);
1157	INIT_LIST_HEAD(&root->logged_list[0]);
1158	INIT_LIST_HEAD(&root->logged_list[1]);
1159	spin_lock_init(&root->inode_lock);
1160	spin_lock_init(&root->delalloc_lock);
1161	spin_lock_init(&root->ordered_extent_lock);
1162	spin_lock_init(&root->accounting_lock);
1163	spin_lock_init(&root->log_extents_lock[0]);
1164	spin_lock_init(&root->log_extents_lock[1]);
1165	spin_lock_init(&root->qgroup_meta_rsv_lock);
1166	mutex_init(&root->objectid_mutex);
1167	mutex_init(&root->log_mutex);
1168	mutex_init(&root->ordered_extent_mutex);
1169	mutex_init(&root->delalloc_mutex);
 
1170	init_waitqueue_head(&root->log_writer_wait);
1171	init_waitqueue_head(&root->log_commit_wait[0]);
1172	init_waitqueue_head(&root->log_commit_wait[1]);
1173	INIT_LIST_HEAD(&root->log_ctxs[0]);
1174	INIT_LIST_HEAD(&root->log_ctxs[1]);
1175	atomic_set(&root->log_commit[0], 0);
1176	atomic_set(&root->log_commit[1], 0);
1177	atomic_set(&root->log_writers, 0);
1178	atomic_set(&root->log_batch, 0);
1179	refcount_set(&root->refs, 1);
1180	atomic_set(&root->will_be_snapshotted, 0);
1181	atomic_set(&root->snapshot_force_cow, 0);
1182	atomic_set(&root->nr_swapfiles, 0);
1183	root->log_transid = 0;
1184	root->log_transid_committed = -1;
1185	root->last_log_commit = 0;
1186	if (!dummy)
1187		extent_io_tree_init(fs_info, &root->dirty_log_pages,
1188				    IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1189
1190	memset(&root->root_key, 0, sizeof(root->root_key));
1191	memset(&root->root_item, 0, sizeof(root->root_item));
1192	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1193	if (!dummy)
1194		root->defrag_trans_start = fs_info->generation;
1195	else
1196		root->defrag_trans_start = 0;
1197	root->root_key.objectid = objectid;
1198	root->anon_dev = 0;
 
 
 
 
 
 
1199
1200	spin_lock_init(&root->root_item_lock);
1201	btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
 
 
 
 
 
 
1202}
1203
1204static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1205		gfp_t flags)
1206{
1207	struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1208	if (root)
1209		root->fs_info = fs_info;
1210	return root;
1211}
1212
1213#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1214/* Should only be used by the testing infrastructure */
1215struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1216{
1217	struct btrfs_root *root;
1218
1219	if (!fs_info)
1220		return ERR_PTR(-EINVAL);
1221
1222	root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1223	if (!root)
1224		return ERR_PTR(-ENOMEM);
1225
1226	/* We don't use the stripesize in selftest, set it as sectorsize */
1227	__setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1228	root->alloc_bytenr = 0;
1229
1230	return root;
1231}
1232#endif
1233
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1234struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1235				     u64 objectid)
1236{
1237	struct btrfs_fs_info *fs_info = trans->fs_info;
1238	struct extent_buffer *leaf;
1239	struct btrfs_root *tree_root = fs_info->tree_root;
1240	struct btrfs_root *root;
1241	struct btrfs_key key;
1242	unsigned int nofs_flag;
1243	int ret = 0;
1244	uuid_le uuid = NULL_UUID_LE;
1245
1246	/*
1247	 * We're holding a transaction handle, so use a NOFS memory allocation
1248	 * context to avoid deadlock if reclaim happens.
1249	 */
1250	nofs_flag = memalloc_nofs_save();
1251	root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1252	memalloc_nofs_restore(nofs_flag);
1253	if (!root)
1254		return ERR_PTR(-ENOMEM);
1255
1256	__setup_root(root, fs_info, objectid);
1257	root->root_key.objectid = objectid;
1258	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1259	root->root_key.offset = 0;
1260
1261	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 
1262	if (IS_ERR(leaf)) {
1263		ret = PTR_ERR(leaf);
1264		leaf = NULL;
1265		goto fail;
1266	}
1267
1268	root->node = leaf;
1269	btrfs_mark_buffer_dirty(leaf);
1270
1271	root->commit_root = btrfs_root_node(root);
1272	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1273
1274	root->root_item.flags = 0;
1275	root->root_item.byte_limit = 0;
1276	btrfs_set_root_bytenr(&root->root_item, leaf->start);
1277	btrfs_set_root_generation(&root->root_item, trans->transid);
1278	btrfs_set_root_level(&root->root_item, 0);
1279	btrfs_set_root_refs(&root->root_item, 1);
1280	btrfs_set_root_used(&root->root_item, leaf->len);
1281	btrfs_set_root_last_snapshot(&root->root_item, 0);
1282	btrfs_set_root_dirid(&root->root_item, 0);
1283	if (is_fstree(objectid))
1284		uuid_le_gen(&uuid);
1285	memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1286	root->root_item.drop_level = 0;
 
 
 
1287
1288	key.objectid = objectid;
1289	key.type = BTRFS_ROOT_ITEM_KEY;
1290	key.offset = 0;
1291	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1292	if (ret)
1293		goto fail;
1294
1295	btrfs_tree_unlock(leaf);
1296
1297	return root;
1298
1299fail:
1300	if (leaf) {
1301		btrfs_tree_unlock(leaf);
1302		free_extent_buffer(root->commit_root);
1303		free_extent_buffer(leaf);
1304	}
1305	kfree(root);
1306
1307	return ERR_PTR(ret);
1308}
1309
1310static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1311					 struct btrfs_fs_info *fs_info)
1312{
1313	struct btrfs_root *root;
1314	struct extent_buffer *leaf;
1315
1316	root = btrfs_alloc_root(fs_info, GFP_NOFS);
1317	if (!root)
1318		return ERR_PTR(-ENOMEM);
1319
1320	__setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1321
1322	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1323	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1324	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1325
 
 
 
 
 
 
 
 
1326	/*
1327	 * DON'T set REF_COWS for log trees
1328	 *
1329	 * log trees do not get reference counted because they go away
1330	 * before a real commit is actually done.  They do store pointers
1331	 * to file data extents, and those reference counts still get
1332	 * updated (along with back refs to the log tree).
 
1333	 */
1334
1335	leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1336			NULL, 0, 0, 0);
1337	if (IS_ERR(leaf)) {
1338		kfree(root);
1339		return ERR_CAST(leaf);
1340	}
1341
1342	root->node = leaf;
1343
1344	btrfs_mark_buffer_dirty(root->node);
1345	btrfs_tree_unlock(root->node);
1346	return root;
 
1347}
1348
1349int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1350			     struct btrfs_fs_info *fs_info)
1351{
1352	struct btrfs_root *log_root;
1353
1354	log_root = alloc_log_tree(trans, fs_info);
1355	if (IS_ERR(log_root))
1356		return PTR_ERR(log_root);
 
 
 
 
 
 
 
 
 
 
1357	WARN_ON(fs_info->log_root_tree);
1358	fs_info->log_root_tree = log_root;
1359	return 0;
1360}
1361
1362int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1363		       struct btrfs_root *root)
1364{
1365	struct btrfs_fs_info *fs_info = root->fs_info;
1366	struct btrfs_root *log_root;
1367	struct btrfs_inode_item *inode_item;
 
1368
1369	log_root = alloc_log_tree(trans, fs_info);
1370	if (IS_ERR(log_root))
1371		return PTR_ERR(log_root);
1372
 
 
 
 
 
 
1373	log_root->last_trans = trans->transid;
1374	log_root->root_key.offset = root->root_key.objectid;
1375
1376	inode_item = &log_root->root_item.inode;
1377	btrfs_set_stack_inode_generation(inode_item, 1);
1378	btrfs_set_stack_inode_size(inode_item, 3);
1379	btrfs_set_stack_inode_nlink(inode_item, 1);
1380	btrfs_set_stack_inode_nbytes(inode_item,
1381				     fs_info->nodesize);
1382	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1383
1384	btrfs_set_root_node(&log_root->root_item, log_root->node);
1385
1386	WARN_ON(root->log_root);
1387	root->log_root = log_root;
1388	root->log_transid = 0;
1389	root->log_transid_committed = -1;
1390	root->last_log_commit = 0;
1391	return 0;
1392}
1393
1394static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1395					       struct btrfs_key *key)
 
1396{
1397	struct btrfs_root *root;
 
1398	struct btrfs_fs_info *fs_info = tree_root->fs_info;
1399	struct btrfs_path *path;
1400	u64 generation;
1401	int ret;
1402	int level;
1403
1404	path = btrfs_alloc_path();
1405	if (!path)
1406		return ERR_PTR(-ENOMEM);
1407
1408	root = btrfs_alloc_root(fs_info, GFP_NOFS);
1409	if (!root) {
1410		ret = -ENOMEM;
1411		goto alloc_fail;
1412	}
1413
1414	__setup_root(root, fs_info, key->objectid);
1415
1416	ret = btrfs_find_root(tree_root, key, path,
1417			      &root->root_item, &root->root_key);
1418	if (ret) {
1419		if (ret > 0)
1420			ret = -ENOENT;
1421		goto find_fail;
1422	}
1423
1424	generation = btrfs_root_generation(&root->root_item);
1425	level = btrfs_root_level(&root->root_item);
1426	root->node = read_tree_block(fs_info,
1427				     btrfs_root_bytenr(&root->root_item),
1428				     generation, level, NULL);
 
 
1429	if (IS_ERR(root->node)) {
1430		ret = PTR_ERR(root->node);
1431		goto find_fail;
1432	} else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
 
 
1433		ret = -EIO;
1434		free_extent_buffer(root->node);
1435		goto find_fail;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1436	}
1437	root->commit_root = btrfs_root_node(root);
1438out:
1439	btrfs_free_path(path);
1440	return root;
1441
1442find_fail:
1443	kfree(root);
1444alloc_fail:
1445	root = ERR_PTR(ret);
1446	goto out;
1447}
1448
1449struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1450				      struct btrfs_key *location)
1451{
1452	struct btrfs_root *root;
 
1453
1454	root = btrfs_read_tree_root(tree_root, location);
1455	if (IS_ERR(root))
1456		return root;
1457
1458	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1459		set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1460		btrfs_check_and_init_root_item(&root->root_item);
1461	}
1462
1463	return root;
1464}
1465
1466int btrfs_init_fs_root(struct btrfs_root *root)
 
 
 
 
 
1467{
1468	int ret;
1469	struct btrfs_subvolume_writers *writers;
1470
1471	root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1472	root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1473					GFP_NOFS);
1474	if (!root->free_ino_pinned || !root->free_ino_ctl) {
1475		ret = -ENOMEM;
1476		goto fail;
1477	}
1478
1479	writers = btrfs_alloc_subvolume_writers();
1480	if (IS_ERR(writers)) {
1481		ret = PTR_ERR(writers);
1482		goto fail;
 
1483	}
1484	root->subv_writers = writers;
1485
1486	btrfs_init_free_ino_ctl(root);
1487	spin_lock_init(&root->ino_cache_lock);
1488	init_waitqueue_head(&root->ino_cache_wait);
1489
1490	ret = get_anon_bdev(&root->anon_dev);
1491	if (ret)
1492		goto fail;
 
 
 
 
 
 
 
 
 
 
 
1493
1494	mutex_lock(&root->objectid_mutex);
1495	ret = btrfs_find_highest_objectid(root,
1496					&root->highest_objectid);
1497	if (ret) {
1498		mutex_unlock(&root->objectid_mutex);
1499		goto fail;
1500	}
1501
1502	ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1503
1504	mutex_unlock(&root->objectid_mutex);
1505
1506	return 0;
1507fail:
1508	/* The caller is responsible to call btrfs_free_fs_root */
1509	return ret;
1510}
1511
1512struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1513					u64 root_id)
1514{
1515	struct btrfs_root *root;
1516
1517	spin_lock(&fs_info->fs_roots_radix_lock);
1518	root = radix_tree_lookup(&fs_info->fs_roots_radix,
1519				 (unsigned long)root_id);
 
1520	spin_unlock(&fs_info->fs_roots_radix_lock);
1521	return root;
1522}
1523
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1524int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1525			 struct btrfs_root *root)
1526{
1527	int ret;
1528
1529	ret = radix_tree_preload(GFP_NOFS);
1530	if (ret)
1531		return ret;
1532
1533	spin_lock(&fs_info->fs_roots_radix_lock);
1534	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1535				(unsigned long)root->root_key.objectid,
1536				root);
1537	if (ret == 0)
 
1538		set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
 
1539	spin_unlock(&fs_info->fs_roots_radix_lock);
1540	radix_tree_preload_end();
1541
1542	return ret;
1543}
1544
1545struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1546				     struct btrfs_key *location,
1547				     bool check_ref)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1548{
1549	struct btrfs_root *root;
1550	struct btrfs_path *path;
1551	struct btrfs_key key;
1552	int ret;
1553
1554	if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1555		return fs_info->tree_root;
1556	if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1557		return fs_info->extent_root;
1558	if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1559		return fs_info->chunk_root;
1560	if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1561		return fs_info->dev_root;
1562	if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1563		return fs_info->csum_root;
1564	if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1565		return fs_info->quota_root ? fs_info->quota_root :
1566					     ERR_PTR(-ENOENT);
1567	if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1568		return fs_info->uuid_root ? fs_info->uuid_root :
1569					    ERR_PTR(-ENOENT);
1570	if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1571		return fs_info->free_space_root ? fs_info->free_space_root :
1572						  ERR_PTR(-ENOENT);
1573again:
1574	root = btrfs_lookup_fs_root(fs_info, location->objectid);
1575	if (root) {
1576		if (check_ref && btrfs_root_refs(&root->root_item) == 0)
 
 
 
 
 
 
 
 
 
 
 
 
1577			return ERR_PTR(-ENOENT);
 
1578		return root;
1579	}
1580
1581	root = btrfs_read_fs_root(fs_info->tree_root, location);
 
 
 
1582	if (IS_ERR(root))
1583		return root;
1584
1585	if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1586		ret = -ENOENT;
1587		goto fail;
1588	}
1589
1590	ret = btrfs_init_fs_root(root);
1591	if (ret)
1592		goto fail;
1593
1594	path = btrfs_alloc_path();
1595	if (!path) {
1596		ret = -ENOMEM;
1597		goto fail;
1598	}
1599	key.objectid = BTRFS_ORPHAN_OBJECTID;
1600	key.type = BTRFS_ORPHAN_ITEM_KEY;
1601	key.offset = location->objectid;
1602
1603	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1604	btrfs_free_path(path);
1605	if (ret < 0)
1606		goto fail;
1607	if (ret == 0)
1608		set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1609
1610	ret = btrfs_insert_fs_root(fs_info, root);
1611	if (ret) {
1612		if (ret == -EEXIST) {
1613			btrfs_free_fs_root(root);
1614			goto again;
1615		}
1616		goto fail;
1617	}
1618	return root;
1619fail:
1620	btrfs_free_fs_root(root);
 
 
 
 
 
 
 
 
1621	return ERR_PTR(ret);
1622}
1623
1624static int btrfs_congested_fn(void *congested_data, int bdi_bits)
 
 
 
 
 
 
 
 
1625{
1626	struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1627	int ret = 0;
1628	struct btrfs_device *device;
1629	struct backing_dev_info *bdi;
1630
1631	rcu_read_lock();
1632	list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1633		if (!device->bdev)
1634			continue;
1635		bdi = device->bdev->bd_bdi;
1636		if (bdi_congested(bdi, bdi_bits)) {
1637			ret = 1;
1638			break;
1639		}
1640	}
1641	rcu_read_unlock();
1642	return ret;
1643}
1644
1645/*
1646 * called by the kthread helper functions to finally call the bio end_io
1647 * functions.  This is where read checksum verification actually happens
 
 
 
 
 
 
 
 
 
 
 
1648 */
1649static void end_workqueue_fn(struct btrfs_work *work)
 
 
1650{
1651	struct bio *bio;
1652	struct btrfs_end_io_wq *end_io_wq;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1653
1654	end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1655	bio = end_io_wq->bio;
 
 
 
1656
1657	bio->bi_status = end_io_wq->status;
1658	bio->bi_private = end_io_wq->private;
1659	bio->bi_end_io = end_io_wq->end_io;
1660	kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1661	bio_endio(bio);
1662}
1663
1664static int cleaner_kthread(void *arg)
1665{
1666	struct btrfs_root *root = arg;
1667	struct btrfs_fs_info *fs_info = root->fs_info;
1668	int again;
1669
1670	while (1) {
1671		again = 0;
1672
1673		set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1674
1675		/* Make the cleaner go to sleep early. */
1676		if (btrfs_need_cleaner_sleep(fs_info))
1677			goto sleep;
1678
1679		/*
1680		 * Do not do anything if we might cause open_ctree() to block
1681		 * before we have finished mounting the filesystem.
1682		 */
1683		if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1684			goto sleep;
1685
1686		if (!mutex_trylock(&fs_info->cleaner_mutex))
1687			goto sleep;
1688
1689		/*
1690		 * Avoid the problem that we change the status of the fs
1691		 * during the above check and trylock.
1692		 */
1693		if (btrfs_need_cleaner_sleep(fs_info)) {
1694			mutex_unlock(&fs_info->cleaner_mutex);
1695			goto sleep;
1696		}
1697
 
 
 
1698		btrfs_run_delayed_iputs(fs_info);
1699
1700		again = btrfs_clean_one_deleted_snapshot(root);
1701		mutex_unlock(&fs_info->cleaner_mutex);
1702
1703		/*
1704		 * The defragger has dealt with the R/O remount and umount,
1705		 * needn't do anything special here.
1706		 */
1707		btrfs_run_defrag_inodes(fs_info);
1708
1709		/*
1710		 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1711		 * with relocation (btrfs_relocate_chunk) and relocation
1712		 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1713		 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1714		 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1715		 * unused block groups.
1716		 */
1717		btrfs_delete_unused_bgs(fs_info);
 
 
 
 
 
 
 
1718sleep:
1719		clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1720		if (kthread_should_park())
1721			kthread_parkme();
1722		if (kthread_should_stop())
1723			return 0;
1724		if (!again) {
1725			set_current_state(TASK_INTERRUPTIBLE);
1726			schedule();
1727			__set_current_state(TASK_RUNNING);
1728		}
1729	}
1730}
1731
1732static int transaction_kthread(void *arg)
1733{
1734	struct btrfs_root *root = arg;
1735	struct btrfs_fs_info *fs_info = root->fs_info;
1736	struct btrfs_trans_handle *trans;
1737	struct btrfs_transaction *cur;
1738	u64 transid;
1739	time64_t now;
1740	unsigned long delay;
1741	bool cannot_commit;
1742
1743	do {
1744		cannot_commit = false;
1745		delay = HZ * fs_info->commit_interval;
1746		mutex_lock(&fs_info->transaction_kthread_mutex);
1747
1748		spin_lock(&fs_info->trans_lock);
1749		cur = fs_info->running_transaction;
1750		if (!cur) {
1751			spin_unlock(&fs_info->trans_lock);
1752			goto sleep;
1753		}
1754
1755		now = ktime_get_seconds();
1756		if (cur->state < TRANS_STATE_BLOCKED &&
1757		    !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1758		    (now < cur->start_time ||
1759		     now - cur->start_time < fs_info->commit_interval)) {
1760			spin_unlock(&fs_info->trans_lock);
1761			delay = HZ * 5;
 
 
1762			goto sleep;
1763		}
1764		transid = cur->transid;
1765		spin_unlock(&fs_info->trans_lock);
1766
1767		/* If the file system is aborted, this will always fail. */
1768		trans = btrfs_attach_transaction(root);
1769		if (IS_ERR(trans)) {
1770			if (PTR_ERR(trans) != -ENOENT)
1771				cannot_commit = true;
1772			goto sleep;
1773		}
1774		if (transid == trans->transid) {
1775			btrfs_commit_transaction(trans);
1776		} else {
1777			btrfs_end_transaction(trans);
1778		}
1779sleep:
1780		wake_up_process(fs_info->cleaner_kthread);
1781		mutex_unlock(&fs_info->transaction_kthread_mutex);
1782
1783		if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1784				      &fs_info->fs_state)))
1785			btrfs_cleanup_transaction(fs_info);
1786		if (!kthread_should_stop() &&
1787				(!btrfs_transaction_blocked(fs_info) ||
1788				 cannot_commit))
1789			schedule_timeout_interruptible(delay);
1790	} while (!kthread_should_stop());
1791	return 0;
1792}
1793
1794/*
1795 * this will find the highest generation in the array of
1796 * root backups.  The index of the highest array is returned,
1797 * or -1 if we can't find anything.
1798 *
1799 * We check to make sure the array is valid by comparing the
1800 * generation of the latest  root in the array with the generation
1801 * in the super block.  If they don't match we pitch it.
1802 */
1803static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1804{
 
1805	u64 cur;
1806	int newest_index = -1;
1807	struct btrfs_root_backup *root_backup;
1808	int i;
1809
1810	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1811		root_backup = info->super_copy->super_roots + i;
1812		cur = btrfs_backup_tree_root_gen(root_backup);
1813		if (cur == newest_gen)
1814			newest_index = i;
1815	}
1816
1817	/* check to see if we actually wrapped around */
1818	if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1819		root_backup = info->super_copy->super_roots;
1820		cur = btrfs_backup_tree_root_gen(root_backup);
1821		if (cur == newest_gen)
1822			newest_index = 0;
1823	}
1824	return newest_index;
1825}
1826
1827
1828/*
1829 * find the oldest backup so we know where to store new entries
1830 * in the backup array.  This will set the backup_root_index
1831 * field in the fs_info struct
1832 */
1833static void find_oldest_super_backup(struct btrfs_fs_info *info,
1834				     u64 newest_gen)
1835{
1836	int newest_index = -1;
1837
1838	newest_index = find_newest_super_backup(info, newest_gen);
1839	/* if there was garbage in there, just move along */
1840	if (newest_index == -1) {
1841		info->backup_root_index = 0;
1842	} else {
1843		info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1844	}
1845}
1846
1847/*
1848 * copy all the root pointers into the super backup array.
1849 * this will bump the backup pointer by one when it is
1850 * done
1851 */
1852static void backup_super_roots(struct btrfs_fs_info *info)
1853{
1854	int next_backup;
1855	struct btrfs_root_backup *root_backup;
1856	int last_backup;
1857
1858	next_backup = info->backup_root_index;
1859	last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1860		BTRFS_NUM_BACKUP_ROOTS;
1861
1862	/*
1863	 * just overwrite the last backup if we're at the same generation
1864	 * this happens only at umount
1865	 */
1866	root_backup = info->super_for_commit->super_roots + last_backup;
1867	if (btrfs_backup_tree_root_gen(root_backup) ==
1868	    btrfs_header_generation(info->tree_root->node))
1869		next_backup = last_backup;
1870
1871	root_backup = info->super_for_commit->super_roots + next_backup;
1872
1873	/*
1874	 * make sure all of our padding and empty slots get zero filled
1875	 * regardless of which ones we use today
1876	 */
1877	memset(root_backup, 0, sizeof(*root_backup));
1878
1879	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1880
1881	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1882	btrfs_set_backup_tree_root_gen(root_backup,
1883			       btrfs_header_generation(info->tree_root->node));
1884
1885	btrfs_set_backup_tree_root_level(root_backup,
1886			       btrfs_header_level(info->tree_root->node));
1887
1888	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1889	btrfs_set_backup_chunk_root_gen(root_backup,
1890			       btrfs_header_generation(info->chunk_root->node));
1891	btrfs_set_backup_chunk_root_level(root_backup,
1892			       btrfs_header_level(info->chunk_root->node));
1893
1894	btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1895	btrfs_set_backup_extent_root_gen(root_backup,
1896			       btrfs_header_generation(info->extent_root->node));
1897	btrfs_set_backup_extent_root_level(root_backup,
1898			       btrfs_header_level(info->extent_root->node));
 
 
 
 
 
 
 
 
 
 
 
 
1899
1900	/*
1901	 * we might commit during log recovery, which happens before we set
1902	 * the fs_root.  Make sure it is valid before we fill it in.
1903	 */
1904	if (info->fs_root && info->fs_root->node) {
1905		btrfs_set_backup_fs_root(root_backup,
1906					 info->fs_root->node->start);
1907		btrfs_set_backup_fs_root_gen(root_backup,
1908			       btrfs_header_generation(info->fs_root->node));
1909		btrfs_set_backup_fs_root_level(root_backup,
1910			       btrfs_header_level(info->fs_root->node));
1911	}
1912
1913	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1914	btrfs_set_backup_dev_root_gen(root_backup,
1915			       btrfs_header_generation(info->dev_root->node));
1916	btrfs_set_backup_dev_root_level(root_backup,
1917				       btrfs_header_level(info->dev_root->node));
1918
1919	btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1920	btrfs_set_backup_csum_root_gen(root_backup,
1921			       btrfs_header_generation(info->csum_root->node));
1922	btrfs_set_backup_csum_root_level(root_backup,
1923			       btrfs_header_level(info->csum_root->node));
1924
1925	btrfs_set_backup_total_bytes(root_backup,
1926			     btrfs_super_total_bytes(info->super_copy));
1927	btrfs_set_backup_bytes_used(root_backup,
1928			     btrfs_super_bytes_used(info->super_copy));
1929	btrfs_set_backup_num_devices(root_backup,
1930			     btrfs_super_num_devices(info->super_copy));
1931
1932	/*
1933	 * if we don't copy this out to the super_copy, it won't get remembered
1934	 * for the next commit
1935	 */
1936	memcpy(&info->super_copy->super_roots,
1937	       &info->super_for_commit->super_roots,
1938	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1939}
1940
1941/*
1942 * this copies info out of the root backup array and back into
1943 * the in-memory super block.  It is meant to help iterate through
1944 * the array, so you send it the number of backups you've already
1945 * tried and the last backup index you used.
1946 *
1947 * this returns -1 when it has tried all the backups
 
 
 
1948 */
1949static noinline int next_root_backup(struct btrfs_fs_info *info,
1950				     struct btrfs_super_block *super,
1951				     int *num_backups_tried, int *backup_index)
1952{
 
 
1953	struct btrfs_root_backup *root_backup;
1954	int newest = *backup_index;
1955
1956	if (*num_backups_tried == 0) {
1957		u64 gen = btrfs_super_generation(super);
 
1958
1959		newest = find_newest_super_backup(info, gen);
1960		if (newest == -1)
1961			return -1;
1962
1963		*backup_index = newest;
1964		*num_backups_tried = 1;
1965	} else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1966		/* we've tried all the backups, all done */
1967		return -1;
1968	} else {
1969		/* jump to the next oldest backup */
1970		newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1971			BTRFS_NUM_BACKUP_ROOTS;
1972		*backup_index = newest;
1973		*num_backups_tried += 1;
1974	}
1975	root_backup = super->super_roots + newest;
 
1976
1977	btrfs_set_super_generation(super,
1978				   btrfs_backup_tree_root_gen(root_backup));
1979	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1980	btrfs_set_super_root_level(super,
1981				   btrfs_backup_tree_root_level(root_backup));
1982	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1983
1984	/*
1985	 * fixme: the total bytes and num_devices need to match or we should
1986	 * need a fsck
1987	 */
1988	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1989	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1990	return 0;
 
1991}
1992
1993/* helper to cleanup workers */
1994static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1995{
1996	btrfs_destroy_workqueue(fs_info->fixup_workers);
1997	btrfs_destroy_workqueue(fs_info->delalloc_workers);
1998	btrfs_destroy_workqueue(fs_info->workers);
1999	btrfs_destroy_workqueue(fs_info->endio_workers);
2000	btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2001	btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2002	btrfs_destroy_workqueue(fs_info->rmw_workers);
 
 
2003	btrfs_destroy_workqueue(fs_info->endio_write_workers);
2004	btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2005	btrfs_destroy_workqueue(fs_info->submit_workers);
2006	btrfs_destroy_workqueue(fs_info->delayed_workers);
2007	btrfs_destroy_workqueue(fs_info->caching_workers);
2008	btrfs_destroy_workqueue(fs_info->readahead_workers);
2009	btrfs_destroy_workqueue(fs_info->flush_workers);
2010	btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
 
 
2011	/*
2012	 * Now that all other work queues are destroyed, we can safely destroy
2013	 * the queues used for metadata I/O, since tasks from those other work
2014	 * queues can do metadata I/O operations.
2015	 */
2016	btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2017	btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2018}
2019
2020static void free_root_extent_buffers(struct btrfs_root *root)
2021{
2022	if (root) {
2023		free_extent_buffer(root->node);
2024		free_extent_buffer(root->commit_root);
2025		root->node = NULL;
2026		root->commit_root = NULL;
2027	}
2028}
2029
 
 
 
 
 
 
 
 
 
 
2030/* helper to cleanup tree roots */
2031static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2032{
2033	free_root_extent_buffers(info->tree_root);
2034
 
2035	free_root_extent_buffers(info->dev_root);
2036	free_root_extent_buffers(info->extent_root);
2037	free_root_extent_buffers(info->csum_root);
2038	free_root_extent_buffers(info->quota_root);
2039	free_root_extent_buffers(info->uuid_root);
2040	if (chunk_root)
 
 
 
 
2041		free_root_extent_buffers(info->chunk_root);
2042	free_root_extent_buffers(info->free_space_root);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2043}
2044
2045void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2046{
2047	int ret;
2048	struct btrfs_root *gang[8];
2049	int i;
2050
2051	while (!list_empty(&fs_info->dead_roots)) {
2052		gang[0] = list_entry(fs_info->dead_roots.next,
2053				     struct btrfs_root, root_list);
2054		list_del(&gang[0]->root_list);
2055
2056		if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2057			btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2058		} else {
2059			free_extent_buffer(gang[0]->node);
2060			free_extent_buffer(gang[0]->commit_root);
2061			btrfs_put_fs_root(gang[0]);
2062		}
2063	}
2064
2065	while (1) {
2066		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2067					     (void **)gang, 0,
2068					     ARRAY_SIZE(gang));
2069		if (!ret)
2070			break;
2071		for (i = 0; i < ret; i++)
2072			btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2073	}
2074
2075	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2076		btrfs_free_log_root_tree(NULL, fs_info);
2077		btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2078	}
2079}
2080
2081static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2082{
2083	mutex_init(&fs_info->scrub_lock);
2084	atomic_set(&fs_info->scrubs_running, 0);
2085	atomic_set(&fs_info->scrub_pause_req, 0);
2086	atomic_set(&fs_info->scrubs_paused, 0);
2087	atomic_set(&fs_info->scrub_cancel_req, 0);
2088	init_waitqueue_head(&fs_info->scrub_pause_wait);
2089	refcount_set(&fs_info->scrub_workers_refcnt, 0);
2090}
2091
2092static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2093{
2094	spin_lock_init(&fs_info->balance_lock);
2095	mutex_init(&fs_info->balance_mutex);
2096	atomic_set(&fs_info->balance_pause_req, 0);
2097	atomic_set(&fs_info->balance_cancel_req, 0);
2098	fs_info->balance_ctl = NULL;
2099	init_waitqueue_head(&fs_info->balance_wait_q);
 
2100}
2101
2102static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2103{
2104	struct inode *inode = fs_info->btree_inode;
 
 
 
 
 
 
 
2105
2106	inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2107	set_nlink(inode, 1);
2108	/*
2109	 * we set the i_size on the btree inode to the max possible int.
2110	 * the real end of the address space is determined by all of
2111	 * the devices in the system
2112	 */
2113	inode->i_size = OFFSET_MAX;
2114	inode->i_mapping->a_ops = &btree_aops;
 
2115
2116	RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2117	extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2118			    IO_TREE_INODE_IO, inode);
2119	BTRFS_I(inode)->io_tree.track_uptodate = false;
2120	extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2121
2122	BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2123
2124	BTRFS_I(inode)->root = fs_info->tree_root;
2125	memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2126	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2127	btrfs_insert_inode_hash(inode);
 
 
 
2128}
2129
2130static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2131{
2132	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2133	init_rwsem(&fs_info->dev_replace.rwsem);
2134	init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2135}
2136
2137static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2138{
2139	spin_lock_init(&fs_info->qgroup_lock);
2140	mutex_init(&fs_info->qgroup_ioctl_lock);
2141	fs_info->qgroup_tree = RB_ROOT;
2142	INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2143	fs_info->qgroup_seq = 1;
2144	fs_info->qgroup_ulist = NULL;
2145	fs_info->qgroup_rescan_running = false;
 
2146	mutex_init(&fs_info->qgroup_rescan_lock);
2147}
2148
2149static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2150		struct btrfs_fs_devices *fs_devices)
2151{
2152	u32 max_active = fs_info->thread_pool_size;
2153	unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
 
2154
2155	fs_info->workers =
2156		btrfs_alloc_workqueue(fs_info, "worker",
2157				      flags | WQ_HIGHPRI, max_active, 16);
2158
2159	fs_info->delalloc_workers =
2160		btrfs_alloc_workqueue(fs_info, "delalloc",
2161				      flags, max_active, 2);
2162
2163	fs_info->flush_workers =
2164		btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2165				      flags, max_active, 0);
2166
2167	fs_info->caching_workers =
2168		btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2169
2170	/*
2171	 * a higher idle thresh on the submit workers makes it much more
2172	 * likely that bios will be send down in a sane order to the
2173	 * devices
2174	 */
2175	fs_info->submit_workers =
2176		btrfs_alloc_workqueue(fs_info, "submit", flags,
2177				      min_t(u64, fs_devices->num_devices,
2178					    max_active), 64);
2179
2180	fs_info->fixup_workers =
2181		btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2182
2183	/*
2184	 * endios are largely parallel and should have a very
2185	 * low idle thresh
2186	 */
2187	fs_info->endio_workers =
2188		btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2189	fs_info->endio_meta_workers =
2190		btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2191				      max_active, 4);
2192	fs_info->endio_meta_write_workers =
2193		btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2194				      max_active, 2);
2195	fs_info->endio_raid56_workers =
2196		btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2197				      max_active, 4);
2198	fs_info->endio_repair_workers =
2199		btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2200	fs_info->rmw_workers =
2201		btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2202	fs_info->endio_write_workers =
2203		btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2204				      max_active, 2);
 
 
2205	fs_info->endio_freespace_worker =
2206		btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2207				      max_active, 0);
2208	fs_info->delayed_workers =
2209		btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2210				      max_active, 0);
2211	fs_info->readahead_workers =
2212		btrfs_alloc_workqueue(fs_info, "readahead", flags,
2213				      max_active, 2);
2214	fs_info->qgroup_rescan_workers =
2215		btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
 
 
 
2216
2217	if (!(fs_info->workers && fs_info->delalloc_workers &&
2218	      fs_info->submit_workers && fs_info->flush_workers &&
2219	      fs_info->endio_workers && fs_info->endio_meta_workers &&
2220	      fs_info->endio_meta_write_workers &&
2221	      fs_info->endio_repair_workers &&
2222	      fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2223	      fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2224	      fs_info->caching_workers && fs_info->readahead_workers &&
2225	      fs_info->fixup_workers && fs_info->delayed_workers &&
2226	      fs_info->qgroup_rescan_workers)) {
2227		return -ENOMEM;
2228	}
2229
2230	return 0;
2231}
2232
2233static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2234{
2235	struct crypto_shash *csum_shash;
2236	const char *csum_name = btrfs_super_csum_name(csum_type);
2237
2238	csum_shash = crypto_alloc_shash(csum_name, 0, 0);
2239
2240	if (IS_ERR(csum_shash)) {
2241		btrfs_err(fs_info, "error allocating %s hash for checksum",
2242			  csum_name);
2243		return PTR_ERR(csum_shash);
2244	}
2245
2246	fs_info->csum_shash = csum_shash;
2247
2248	return 0;
2249}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2250
2251static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
2252{
2253	crypto_free_shash(fs_info->csum_shash);
 
2254}
2255
2256static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2257			    struct btrfs_fs_devices *fs_devices)
2258{
2259	int ret;
 
2260	struct btrfs_root *log_tree_root;
2261	struct btrfs_super_block *disk_super = fs_info->super_copy;
2262	u64 bytenr = btrfs_super_log_root(disk_super);
2263	int level = btrfs_super_log_root_level(disk_super);
2264
2265	if (fs_devices->rw_devices == 0) {
2266		btrfs_warn(fs_info, "log replay required on RO media");
2267		return -EIO;
2268	}
2269
2270	log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
 
2271	if (!log_tree_root)
2272		return -ENOMEM;
2273
2274	__setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2275
2276	log_tree_root->node = read_tree_block(fs_info, bytenr,
2277					      fs_info->generation + 1,
2278					      level, NULL);
2279	if (IS_ERR(log_tree_root->node)) {
2280		btrfs_warn(fs_info, "failed to read log tree");
2281		ret = PTR_ERR(log_tree_root->node);
2282		kfree(log_tree_root);
 
2283		return ret;
2284	} else if (!extent_buffer_uptodate(log_tree_root->node)) {
 
2285		btrfs_err(fs_info, "failed to read log tree");
2286		free_extent_buffer(log_tree_root->node);
2287		kfree(log_tree_root);
2288		return -EIO;
2289	}
 
2290	/* returns with log_tree_root freed on success */
2291	ret = btrfs_recover_log_trees(log_tree_root);
2292	if (ret) {
2293		btrfs_handle_fs_error(fs_info, ret,
2294				      "Failed to recover log tree");
2295		free_extent_buffer(log_tree_root->node);
2296		kfree(log_tree_root);
2297		return ret;
2298	}
2299
2300	if (sb_rdonly(fs_info->sb)) {
2301		ret = btrfs_commit_super(fs_info);
2302		if (ret)
2303			return ret;
2304	}
2305
2306	return 0;
2307}
2308
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2309static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2310{
2311	struct btrfs_root *tree_root = fs_info->tree_root;
2312	struct btrfs_root *root;
2313	struct btrfs_key location;
2314	int ret;
2315
2316	BUG_ON(!fs_info->tree_root);
2317
2318	location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
 
 
 
2319	location.type = BTRFS_ROOT_ITEM_KEY;
2320	location.offset = 0;
2321
2322	root = btrfs_read_tree_root(tree_root, &location);
2323	if (IS_ERR(root)) {
2324		ret = PTR_ERR(root);
2325		goto out;
 
 
 
 
 
 
 
 
2326	}
2327	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2328	fs_info->extent_root = root;
2329
2330	location.objectid = BTRFS_DEV_TREE_OBJECTID;
2331	root = btrfs_read_tree_root(tree_root, &location);
2332	if (IS_ERR(root)) {
2333		ret = PTR_ERR(root);
2334		goto out;
 
 
 
 
 
2335	}
2336	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2337	fs_info->dev_root = root;
2338	btrfs_init_devices_late(fs_info);
 
2339
2340	location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2341	root = btrfs_read_tree_root(tree_root, &location);
 
 
 
 
2342	if (IS_ERR(root)) {
2343		ret = PTR_ERR(root);
2344		goto out;
 
 
 
 
 
2345	}
2346	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2347	fs_info->csum_root = root;
2348
2349	location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2350	root = btrfs_read_tree_root(tree_root, &location);
2351	if (!IS_ERR(root)) {
2352		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2353		set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2354		fs_info->quota_root = root;
2355	}
2356
2357	location.objectid = BTRFS_UUID_TREE_OBJECTID;
2358	root = btrfs_read_tree_root(tree_root, &location);
2359	if (IS_ERR(root)) {
2360		ret = PTR_ERR(root);
2361		if (ret != -ENOENT)
2362			goto out;
 
 
2363	} else {
2364		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2365		fs_info->uuid_root = root;
2366	}
2367
2368	if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2369		location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2370		root = btrfs_read_tree_root(tree_root, &location);
2371		if (IS_ERR(root)) {
2372			ret = PTR_ERR(root);
2373			goto out;
 
 
 
 
 
2374		}
2375		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2376		fs_info->free_space_root = root;
2377	}
2378
2379	return 0;
2380out:
2381	btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2382		   location.objectid, ret);
2383	return ret;
2384}
2385
2386/*
2387 * Real super block validation
2388 * NOTE: super csum type and incompat features will not be checked here.
2389 *
2390 * @sb:		super block to check
2391 * @mirror_num:	the super block number to check its bytenr:
2392 * 		0	the primary (1st) sb
2393 * 		1, 2	2nd and 3rd backup copy
2394 * 	       -1	skip bytenr check
2395 */
2396static int validate_super(struct btrfs_fs_info *fs_info,
2397			    struct btrfs_super_block *sb, int mirror_num)
2398{
2399	u64 nodesize = btrfs_super_nodesize(sb);
2400	u64 sectorsize = btrfs_super_sectorsize(sb);
2401	int ret = 0;
2402
2403	if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2404		btrfs_err(fs_info, "no valid FS found");
2405		ret = -EINVAL;
2406	}
2407	if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2408		btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2409				btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2410		ret = -EINVAL;
2411	}
2412	if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2413		btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2414				btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2415		ret = -EINVAL;
2416	}
2417	if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2418		btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2419				btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2420		ret = -EINVAL;
2421	}
2422	if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2423		btrfs_err(fs_info, "log_root level too big: %d >= %d",
2424				btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2425		ret = -EINVAL;
2426	}
2427
2428	/*
2429	 * Check sectorsize and nodesize first, other check will need it.
2430	 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2431	 */
2432	if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2433	    sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2434		btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2435		ret = -EINVAL;
2436	}
2437	/* Only PAGE SIZE is supported yet */
2438	if (sectorsize != PAGE_SIZE) {
 
 
 
 
 
 
 
 
2439		btrfs_err(fs_info,
2440			"sectorsize %llu not supported yet, only support %lu",
2441			sectorsize, PAGE_SIZE);
2442		ret = -EINVAL;
2443	}
 
2444	if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2445	    nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2446		btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2447		ret = -EINVAL;
2448	}
2449	if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2450		btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2451			  le32_to_cpu(sb->__unused_leafsize), nodesize);
2452		ret = -EINVAL;
2453	}
2454
2455	/* Root alignment check */
2456	if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2457		btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2458			   btrfs_super_root(sb));
2459		ret = -EINVAL;
2460	}
2461	if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2462		btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2463			   btrfs_super_chunk_root(sb));
2464		ret = -EINVAL;
2465	}
2466	if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2467		btrfs_warn(fs_info, "log_root block unaligned: %llu",
2468			   btrfs_super_log_root(sb));
2469		ret = -EINVAL;
2470	}
2471
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2472	if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2473		   BTRFS_FSID_SIZE) != 0) {
2474		btrfs_err(fs_info,
2475			"dev_item UUID does not match metadata fsid: %pU != %pU",
2476			fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2477		ret = -EINVAL;
2478	}
2479
2480	/*
 
 
 
 
 
 
 
 
 
 
 
 
2481	 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2482	 * done later
2483	 */
2484	if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2485		btrfs_err(fs_info, "bytes_used is too small %llu",
2486			  btrfs_super_bytes_used(sb));
2487		ret = -EINVAL;
2488	}
2489	if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2490		btrfs_err(fs_info, "invalid stripesize %u",
2491			  btrfs_super_stripesize(sb));
2492		ret = -EINVAL;
2493	}
2494	if (btrfs_super_num_devices(sb) > (1UL << 31))
2495		btrfs_warn(fs_info, "suspicious number of devices: %llu",
2496			   btrfs_super_num_devices(sb));
2497	if (btrfs_super_num_devices(sb) == 0) {
2498		btrfs_err(fs_info, "number of devices is 0");
2499		ret = -EINVAL;
2500	}
2501
2502	if (mirror_num >= 0 &&
2503	    btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2504		btrfs_err(fs_info, "super offset mismatch %llu != %u",
2505			  btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2506		ret = -EINVAL;
2507	}
2508
2509	/*
2510	 * Obvious sys_chunk_array corruptions, it must hold at least one key
2511	 * and one chunk
2512	 */
2513	if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2514		btrfs_err(fs_info, "system chunk array too big %u > %u",
2515			  btrfs_super_sys_array_size(sb),
2516			  BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2517		ret = -EINVAL;
2518	}
2519	if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2520			+ sizeof(struct btrfs_chunk)) {
2521		btrfs_err(fs_info, "system chunk array too small %u < %zu",
2522			  btrfs_super_sys_array_size(sb),
2523			  sizeof(struct btrfs_disk_key)
2524			  + sizeof(struct btrfs_chunk));
2525		ret = -EINVAL;
2526	}
2527
2528	/*
2529	 * The generation is a global counter, we'll trust it more than the others
2530	 * but it's still possible that it's the one that's wrong.
2531	 */
2532	if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2533		btrfs_warn(fs_info,
2534			"suspicious: generation < chunk_root_generation: %llu < %llu",
2535			btrfs_super_generation(sb),
2536			btrfs_super_chunk_root_generation(sb));
2537	if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2538	    && btrfs_super_cache_generation(sb) != (u64)-1)
2539		btrfs_warn(fs_info,
2540			"suspicious: generation < cache_generation: %llu < %llu",
2541			btrfs_super_generation(sb),
2542			btrfs_super_cache_generation(sb));
2543
2544	return ret;
2545}
2546
2547/*
2548 * Validation of super block at mount time.
2549 * Some checks already done early at mount time, like csum type and incompat
2550 * flags will be skipped.
2551 */
2552static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2553{
2554	return validate_super(fs_info, fs_info->super_copy, 0);
2555}
2556
2557/*
2558 * Validation of super block at write time.
2559 * Some checks like bytenr check will be skipped as their values will be
2560 * overwritten soon.
2561 * Extra checks like csum type and incompat flags will be done here.
2562 */
2563static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2564				      struct btrfs_super_block *sb)
2565{
2566	int ret;
2567
2568	ret = validate_super(fs_info, sb, -1);
2569	if (ret < 0)
2570		goto out;
2571	if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2572		ret = -EUCLEAN;
2573		btrfs_err(fs_info, "invalid csum type, has %u want %u",
2574			  btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2575		goto out;
2576	}
2577	if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2578		ret = -EUCLEAN;
2579		btrfs_err(fs_info,
2580		"invalid incompat flags, has 0x%llx valid mask 0x%llx",
2581			  btrfs_super_incompat_flags(sb),
2582			  (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2583		goto out;
2584	}
2585out:
2586	if (ret < 0)
2587		btrfs_err(fs_info,
2588		"super block corruption detected before writing it to disk");
2589	return ret;
2590}
2591
2592int open_ctree(struct super_block *sb,
2593	       struct btrfs_fs_devices *fs_devices,
2594	       char *options)
2595{
2596	u32 sectorsize;
2597	u32 nodesize;
2598	u32 stripesize;
2599	u64 generation;
2600	u64 features;
2601	u16 csum_type;
2602	struct btrfs_key location;
2603	struct buffer_head *bh;
2604	struct btrfs_super_block *disk_super;
2605	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2606	struct btrfs_root *tree_root;
2607	struct btrfs_root *chunk_root;
2608	int ret;
2609	int err = -EINVAL;
2610	int num_backups_tried = 0;
2611	int backup_index = 0;
2612	int clear_free_space_tree = 0;
2613	int level;
2614
2615	tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2616	chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2617	if (!tree_root || !chunk_root) {
2618		err = -ENOMEM;
2619		goto fail;
2620	}
2621
2622	ret = init_srcu_struct(&fs_info->subvol_srcu);
2623	if (ret) {
2624		err = ret;
2625		goto fail;
2626	}
2627
2628	ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2629	if (ret) {
2630		err = ret;
2631		goto fail_srcu;
2632	}
2633
2634	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
 
 
 
 
 
 
 
 
 
2635	if (ret) {
2636		err = ret;
2637		goto fail_dio_bytes;
2638	}
2639	fs_info->dirty_metadata_batch = PAGE_SIZE *
2640					(1 + ilog2(nr_cpu_ids));
2641
2642	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2643	if (ret) {
2644		err = ret;
2645		goto fail_dirty_metadata_bytes;
2646	}
 
 
 
2647
2648	ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2649			GFP_KERNEL);
2650	if (ret) {
2651		err = ret;
2652		goto fail_delalloc_bytes;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2653	}
2654
 
 
 
 
 
2655	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2656	INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2657	INIT_LIST_HEAD(&fs_info->trans_list);
2658	INIT_LIST_HEAD(&fs_info->dead_roots);
2659	INIT_LIST_HEAD(&fs_info->delayed_iputs);
2660	INIT_LIST_HEAD(&fs_info->delalloc_roots);
2661	INIT_LIST_HEAD(&fs_info->caching_block_groups);
2662	spin_lock_init(&fs_info->delalloc_root_lock);
2663	spin_lock_init(&fs_info->trans_lock);
2664	spin_lock_init(&fs_info->fs_roots_radix_lock);
2665	spin_lock_init(&fs_info->delayed_iput_lock);
2666	spin_lock_init(&fs_info->defrag_inodes_lock);
2667	spin_lock_init(&fs_info->tree_mod_seq_lock);
2668	spin_lock_init(&fs_info->super_lock);
2669	spin_lock_init(&fs_info->buffer_lock);
2670	spin_lock_init(&fs_info->unused_bgs_lock);
 
 
 
2671	rwlock_init(&fs_info->tree_mod_log_lock);
 
2672	mutex_init(&fs_info->unused_bg_unpin_mutex);
2673	mutex_init(&fs_info->delete_unused_bgs_mutex);
2674	mutex_init(&fs_info->reloc_mutex);
2675	mutex_init(&fs_info->delalloc_root_mutex);
 
 
2676	seqlock_init(&fs_info->profiles_lock);
2677
 
 
 
 
 
 
 
 
 
 
 
 
 
2678	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2679	INIT_LIST_HEAD(&fs_info->space_info);
2680	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2681	INIT_LIST_HEAD(&fs_info->unused_bgs);
2682	extent_map_tree_init(&fs_info->mapping_tree);
 
 
 
 
 
 
 
 
2683	btrfs_init_block_rsv(&fs_info->global_block_rsv,
2684			     BTRFS_BLOCK_RSV_GLOBAL);
2685	btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2686	btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2687	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2688	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2689			     BTRFS_BLOCK_RSV_DELOPS);
2690	btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2691			     BTRFS_BLOCK_RSV_DELREFS);
2692
2693	atomic_set(&fs_info->async_delalloc_pages, 0);
2694	atomic_set(&fs_info->defrag_running, 0);
2695	atomic_set(&fs_info->reada_works_cnt, 0);
2696	atomic_set(&fs_info->nr_delayed_iputs, 0);
2697	atomic64_set(&fs_info->tree_mod_seq, 0);
2698	fs_info->sb = sb;
2699	fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2700	fs_info->metadata_ratio = 0;
2701	fs_info->defrag_inodes = RB_ROOT;
2702	atomic64_set(&fs_info->free_chunk_space, 0);
2703	fs_info->tree_mod_log = RB_ROOT;
2704	fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2705	fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2706	/* readahead state */
2707	INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2708	spin_lock_init(&fs_info->reada_lock);
2709	btrfs_init_ref_verify(fs_info);
2710
2711	fs_info->thread_pool_size = min_t(unsigned long,
2712					  num_online_cpus() + 2, 8);
2713
2714	INIT_LIST_HEAD(&fs_info->ordered_roots);
2715	spin_lock_init(&fs_info->ordered_root_lock);
2716
2717	fs_info->btree_inode = new_inode(sb);
2718	if (!fs_info->btree_inode) {
2719		err = -ENOMEM;
2720		goto fail_bio_counter;
2721	}
2722	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2723
2724	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2725					GFP_KERNEL);
2726	if (!fs_info->delayed_root) {
2727		err = -ENOMEM;
2728		goto fail_iput;
2729	}
2730	btrfs_init_delayed_root(fs_info->delayed_root);
2731
2732	btrfs_init_scrub(fs_info);
2733#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2734	fs_info->check_integrity_print_mask = 0;
2735#endif
2736	btrfs_init_balance(fs_info);
2737	btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2738
2739	sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2740	sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2741
2742	btrfs_init_btree_inode(fs_info);
 
2743
2744	spin_lock_init(&fs_info->block_group_cache_lock);
2745	fs_info->block_group_cache_tree = RB_ROOT;
2746	fs_info->first_logical_byte = (u64)-1;
2747
2748	extent_io_tree_init(fs_info, &fs_info->freed_extents[0],
2749			    IO_TREE_FS_INFO_FREED_EXTENTS0, NULL);
2750	extent_io_tree_init(fs_info, &fs_info->freed_extents[1],
2751			    IO_TREE_FS_INFO_FREED_EXTENTS1, NULL);
2752	fs_info->pinned_extents = &fs_info->freed_extents[0];
2753	set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2754
2755	mutex_init(&fs_info->ordered_operations_mutex);
2756	mutex_init(&fs_info->tree_log_mutex);
2757	mutex_init(&fs_info->chunk_mutex);
2758	mutex_init(&fs_info->transaction_kthread_mutex);
2759	mutex_init(&fs_info->cleaner_mutex);
2760	mutex_init(&fs_info->ro_block_group_mutex);
2761	init_rwsem(&fs_info->commit_root_sem);
2762	init_rwsem(&fs_info->cleanup_work_sem);
2763	init_rwsem(&fs_info->subvol_sem);
2764	sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2765
2766	btrfs_init_dev_replace_locks(fs_info);
2767	btrfs_init_qgroup(fs_info);
 
2768
2769	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2770	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2771
2772	init_waitqueue_head(&fs_info->transaction_throttle);
2773	init_waitqueue_head(&fs_info->transaction_wait);
2774	init_waitqueue_head(&fs_info->transaction_blocked_wait);
2775	init_waitqueue_head(&fs_info->async_submit_wait);
2776	init_waitqueue_head(&fs_info->delayed_iputs_wait);
2777
2778	/* Usable values until the real ones are cached from the superblock */
2779	fs_info->nodesize = 4096;
2780	fs_info->sectorsize = 4096;
 
2781	fs_info->stripesize = 4096;
2782
 
 
 
 
 
2783	spin_lock_init(&fs_info->swapfile_pins_lock);
2784	fs_info->swapfile_pins = RB_ROOT;
2785
2786	fs_info->send_in_progress = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2787
2788	ret = btrfs_alloc_stripe_hash_table(fs_info);
 
 
 
 
2789	if (ret) {
2790		err = ret;
2791		goto fail_alloc;
2792	}
2793
2794	__setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2795
2796	invalidate_bdev(fs_devices->latest_bdev);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2797
2798	/*
2799	 * Read super block and check the signature bytes only
2800	 */
2801	bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2802	if (IS_ERR(bh)) {
2803		err = PTR_ERR(bh);
2804		goto fail_alloc;
2805	}
2806
 
2807	/*
2808	 * Verify the type first, if that or the the checksum value are
2809	 * corrupted, we'll find out
2810	 */
2811	csum_type = btrfs_super_csum_type((struct btrfs_super_block *)bh->b_data);
2812	if (!btrfs_supported_super_csum(csum_type)) {
2813		btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2814			  csum_type);
2815		err = -EINVAL;
2816		brelse(bh);
2817		goto fail_alloc;
2818	}
2819
 
 
2820	ret = btrfs_init_csum_hash(fs_info, csum_type);
2821	if (ret) {
2822		err = ret;
2823		goto fail_alloc;
2824	}
2825
2826	/*
2827	 * We want to check superblock checksum, the type is stored inside.
2828	 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2829	 */
2830	if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2831		btrfs_err(fs_info, "superblock checksum mismatch");
2832		err = -EINVAL;
2833		brelse(bh);
2834		goto fail_csum;
2835	}
2836
2837	/*
2838	 * super_copy is zeroed at allocation time and we never touch the
2839	 * following bytes up to INFO_SIZE, the checksum is calculated from
2840	 * the whole block of INFO_SIZE
2841	 */
2842	memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2843	brelse(bh);
2844
2845	disk_super = fs_info->super_copy;
2846
2847	ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2848		       BTRFS_FSID_SIZE));
2849
2850	if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2851		ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2852				fs_info->super_copy->metadata_uuid,
2853				BTRFS_FSID_SIZE));
2854	}
2855
2856	features = btrfs_super_flags(disk_super);
2857	if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2858		features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2859		btrfs_set_super_flags(disk_super, features);
2860		btrfs_info(fs_info,
2861			"found metadata UUID change in progress flag, clearing");
2862	}
2863
2864	memcpy(fs_info->super_for_commit, fs_info->super_copy,
2865	       sizeof(*fs_info->super_for_commit));
2866
2867	ret = btrfs_validate_mount_super(fs_info);
2868	if (ret) {
2869		btrfs_err(fs_info, "superblock contains fatal errors");
2870		err = -EINVAL;
2871		goto fail_csum;
2872	}
2873
2874	if (!btrfs_super_root(disk_super))
2875		goto fail_csum;
 
 
 
2876
2877	/* check FS state, whether FS is broken. */
2878	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2879		set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2880
2881	/*
2882	 * run through our array of backup supers and setup
2883	 * our ring pointer to the oldest one
2884	 */
2885	generation = btrfs_super_generation(disk_super);
2886	find_oldest_super_backup(fs_info, generation);
2887
2888	/*
2889	 * In the long term, we'll store the compression type in the super
2890	 * block, and it'll be used for per file compression control.
2891	 */
2892	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2893
2894	ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2895	if (ret) {
2896		err = ret;
2897		goto fail_csum;
2898	}
2899
2900	features = btrfs_super_incompat_flags(disk_super) &
2901		~BTRFS_FEATURE_INCOMPAT_SUPP;
2902	if (features) {
2903		btrfs_err(fs_info,
2904		    "cannot mount because of unsupported optional features (%llx)",
2905		    features);
2906		err = -EINVAL;
2907		goto fail_csum;
2908	}
2909
2910	features = btrfs_super_incompat_flags(disk_super);
2911	features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2912	if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2913		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2914	else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2915		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2916
2917	if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2918		btrfs_info(fs_info, "has skinny extents");
2919
2920	/*
2921	 * flag our filesystem as having big metadata blocks if
2922	 * they are bigger than the page size
2923	 */
2924	if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2925		if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2926			btrfs_info(fs_info,
2927				"flagging fs with big metadata feature");
2928		features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2929	}
2930
 
2931	nodesize = btrfs_super_nodesize(disk_super);
2932	sectorsize = btrfs_super_sectorsize(disk_super);
2933	stripesize = sectorsize;
2934	fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2935	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2936
2937	/* Cache block sizes */
2938	fs_info->nodesize = nodesize;
2939	fs_info->sectorsize = sectorsize;
 
 
2940	fs_info->stripesize = stripesize;
2941
2942	/*
2943	 * mixed block groups end up with duplicate but slightly offset
2944	 * extent buffers for the same range.  It leads to corruptions
2945	 */
2946	if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2947	    (sectorsize != nodesize)) {
2948		btrfs_err(fs_info,
2949"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2950			nodesize, sectorsize);
2951		goto fail_csum;
2952	}
2953
 
 
 
 
2954	/*
2955	 * Needn't use the lock because there is no other task which will
2956	 * update the flag.
2957	 */
2958	btrfs_set_super_incompat_flags(disk_super, features);
2959
2960	features = btrfs_super_compat_ro_flags(disk_super) &
2961		~BTRFS_FEATURE_COMPAT_RO_SUPP;
2962	if (!sb_rdonly(sb) && features) {
2963		btrfs_err(fs_info,
2964	"cannot mount read-write because of unsupported optional features (%llx)",
2965		       features);
2966		err = -EINVAL;
2967		goto fail_csum;
 
 
 
 
 
2968	}
2969
2970	ret = btrfs_init_workqueues(fs_info, fs_devices);
2971	if (ret) {
2972		err = ret;
2973		goto fail_sb_buffer;
2974	}
2975
2976	sb->s_bdi->congested_fn = btrfs_congested_fn;
2977	sb->s_bdi->congested_data = fs_info;
2978	sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2979	sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
2980	sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2981	sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2982
2983	sb->s_blocksize = sectorsize;
2984	sb->s_blocksize_bits = blksize_bits(sectorsize);
2985	memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
2986
2987	mutex_lock(&fs_info->chunk_mutex);
2988	ret = btrfs_read_sys_array(fs_info);
2989	mutex_unlock(&fs_info->chunk_mutex);
2990	if (ret) {
2991		btrfs_err(fs_info, "failed to read the system array: %d", ret);
2992		goto fail_sb_buffer;
2993	}
2994
2995	generation = btrfs_super_chunk_root_generation(disk_super);
2996	level = btrfs_super_chunk_root_level(disk_super);
2997
2998	__setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2999
3000	chunk_root->node = read_tree_block(fs_info,
3001					   btrfs_super_chunk_root(disk_super),
3002					   generation, level, NULL);
3003	if (IS_ERR(chunk_root->node) ||
3004	    !extent_buffer_uptodate(chunk_root->node)) {
3005		btrfs_err(fs_info, "failed to read chunk root");
3006		if (!IS_ERR(chunk_root->node))
3007			free_extent_buffer(chunk_root->node);
3008		chunk_root->node = NULL;
3009		goto fail_tree_roots;
3010	}
3011	btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3012	chunk_root->commit_root = btrfs_root_node(chunk_root);
3013
3014	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3015	   btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
 
3016
3017	ret = btrfs_read_chunk_tree(fs_info);
3018	if (ret) {
3019		btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3020		goto fail_tree_roots;
3021	}
3022
3023	/*
3024	 * Keep the devid that is marked to be the target device for the
3025	 * device replace procedure
 
 
 
3026	 */
3027	btrfs_free_extra_devids(fs_devices, 0);
3028
3029	if (!fs_devices->latest_bdev) {
3030		btrfs_err(fs_info, "failed to read devices");
 
3031		goto fail_tree_roots;
3032	}
3033
3034retry_root_backup:
3035	generation = btrfs_super_generation(disk_super);
3036	level = btrfs_super_root_level(disk_super);
3037
3038	tree_root->node = read_tree_block(fs_info,
3039					  btrfs_super_root(disk_super),
3040					  generation, level, NULL);
3041	if (IS_ERR(tree_root->node) ||
3042	    !extent_buffer_uptodate(tree_root->node)) {
3043		btrfs_warn(fs_info, "failed to read tree root");
3044		if (!IS_ERR(tree_root->node))
3045			free_extent_buffer(tree_root->node);
3046		tree_root->node = NULL;
3047		goto recovery_tree_root;
3048	}
3049
3050	btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3051	tree_root->commit_root = btrfs_root_node(tree_root);
3052	btrfs_set_root_refs(&tree_root->root_item, 1);
3053
3054	mutex_lock(&tree_root->objectid_mutex);
3055	ret = btrfs_find_highest_objectid(tree_root,
3056					&tree_root->highest_objectid);
3057	if (ret) {
3058		mutex_unlock(&tree_root->objectid_mutex);
3059		goto recovery_tree_root;
 
3060	}
3061
3062	ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3063
3064	mutex_unlock(&tree_root->objectid_mutex);
3065
3066	ret = btrfs_read_roots(fs_info);
3067	if (ret)
3068		goto recovery_tree_root;
3069
3070	fs_info->generation = generation;
3071	fs_info->last_trans_committed = generation;
 
3072
3073	ret = btrfs_verify_dev_extents(fs_info);
3074	if (ret) {
3075		btrfs_err(fs_info,
3076			  "failed to verify dev extents against chunks: %d",
3077			  ret);
3078		goto fail_block_groups;
3079	}
3080	ret = btrfs_recover_balance(fs_info);
3081	if (ret) {
3082		btrfs_err(fs_info, "failed to recover balance: %d", ret);
3083		goto fail_block_groups;
3084	}
3085
3086	ret = btrfs_init_dev_stats(fs_info);
3087	if (ret) {
3088		btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3089		goto fail_block_groups;
3090	}
3091
3092	ret = btrfs_init_dev_replace(fs_info);
3093	if (ret) {
3094		btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3095		goto fail_block_groups;
3096	}
3097
3098	btrfs_free_extra_devids(fs_devices, 1);
3099
3100	ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3101	if (ret) {
3102		btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3103				ret);
3104		goto fail_block_groups;
3105	}
3106
3107	ret = btrfs_sysfs_add_device(fs_devices);
3108	if (ret) {
3109		btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3110				ret);
3111		goto fail_fsdev_sysfs;
3112	}
3113
3114	ret = btrfs_sysfs_add_mounted(fs_info);
3115	if (ret) {
3116		btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3117		goto fail_fsdev_sysfs;
3118	}
3119
3120	ret = btrfs_init_space_info(fs_info);
3121	if (ret) {
3122		btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3123		goto fail_sysfs;
3124	}
3125
3126	ret = btrfs_read_block_groups(fs_info);
3127	if (ret) {
3128		btrfs_err(fs_info, "failed to read block groups: %d", ret);
3129		goto fail_sysfs;
3130	}
3131
3132	if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
 
 
 
 
 
3133		btrfs_warn(fs_info,
3134		"writable mount is not allowed due to too many missing devices");
 
3135		goto fail_sysfs;
3136	}
3137
3138	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3139					       "btrfs-cleaner");
3140	if (IS_ERR(fs_info->cleaner_kthread))
 
3141		goto fail_sysfs;
 
3142
3143	fs_info->transaction_kthread = kthread_run(transaction_kthread,
3144						   tree_root,
3145						   "btrfs-transaction");
3146	if (IS_ERR(fs_info->transaction_kthread))
 
3147		goto fail_cleaner;
3148
3149	if (!btrfs_test_opt(fs_info, NOSSD) &&
3150	    !fs_info->fs_devices->rotating) {
3151		btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3152	}
3153
3154	/*
3155	 * Mount does not set all options immediately, we can do it now and do
3156	 * not have to wait for transaction commit
3157	 */
3158	btrfs_apply_pending_changes(fs_info);
3159
3160#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3161	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3162		ret = btrfsic_mount(fs_info, fs_devices,
3163				    btrfs_test_opt(fs_info,
3164					CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3165				    1 : 0,
3166				    fs_info->check_integrity_print_mask);
3167		if (ret)
3168			btrfs_warn(fs_info,
3169				"failed to initialize integrity check module: %d",
3170				ret);
3171	}
3172#endif
3173	ret = btrfs_read_qgroup_config(fs_info);
3174	if (ret)
3175		goto fail_trans_kthread;
3176
3177	if (btrfs_build_ref_tree(fs_info))
3178		btrfs_err(fs_info, "couldn't build ref tree");
3179
3180	/* do not make disk changes in broken FS or nologreplay is given */
3181	if (btrfs_super_log_root(disk_super) != 0 &&
3182	    !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
 
3183		ret = btrfs_replay_log(fs_info, fs_devices);
3184		if (ret) {
3185			err = ret;
3186			goto fail_qgroup;
3187		}
3188	}
3189
3190	ret = btrfs_find_orphan_roots(fs_info);
3191	if (ret)
3192		goto fail_qgroup;
3193
3194	if (!sb_rdonly(sb)) {
3195		ret = btrfs_cleanup_fs_roots(fs_info);
3196		if (ret)
3197			goto fail_qgroup;
3198
3199		mutex_lock(&fs_info->cleaner_mutex);
3200		ret = btrfs_recover_relocation(tree_root);
3201		mutex_unlock(&fs_info->cleaner_mutex);
3202		if (ret < 0) {
3203			btrfs_warn(fs_info, "failed to recover relocation: %d",
3204					ret);
3205			err = -EINVAL;
3206			goto fail_qgroup;
3207		}
3208	}
3209
3210	location.objectid = BTRFS_FS_TREE_OBJECTID;
3211	location.type = BTRFS_ROOT_ITEM_KEY;
3212	location.offset = 0;
3213
3214	fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3215	if (IS_ERR(fs_info->fs_root)) {
3216		err = PTR_ERR(fs_info->fs_root);
3217		btrfs_warn(fs_info, "failed to read fs tree: %d", err);
 
3218		goto fail_qgroup;
3219	}
3220
3221	if (sb_rdonly(sb))
3222		return 0;
3223
3224	if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3225	    btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3226		clear_free_space_tree = 1;
3227	} else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3228		   !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3229		btrfs_warn(fs_info, "free space tree is invalid");
3230		clear_free_space_tree = 1;
3231	}
3232
3233	if (clear_free_space_tree) {
3234		btrfs_info(fs_info, "clearing free space tree");
3235		ret = btrfs_clear_free_space_tree(fs_info);
3236		if (ret) {
3237			btrfs_warn(fs_info,
3238				   "failed to clear free space tree: %d", ret);
3239			close_ctree(fs_info);
3240			return ret;
3241		}
3242	}
3243
3244	if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3245	    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3246		btrfs_info(fs_info, "creating free space tree");
3247		ret = btrfs_create_free_space_tree(fs_info);
3248		if (ret) {
3249			btrfs_warn(fs_info,
3250				"failed to create free space tree: %d", ret);
3251			close_ctree(fs_info);
3252			return ret;
3253		}
3254	}
3255
3256	down_read(&fs_info->cleanup_work_sem);
3257	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3258	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3259		up_read(&fs_info->cleanup_work_sem);
3260		close_ctree(fs_info);
3261		return ret;
3262	}
3263	up_read(&fs_info->cleanup_work_sem);
3264
3265	ret = btrfs_resume_balance_async(fs_info);
3266	if (ret) {
3267		btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3268		close_ctree(fs_info);
3269		return ret;
3270	}
3271
3272	ret = btrfs_resume_dev_replace_async(fs_info);
3273	if (ret) {
3274		btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3275		close_ctree(fs_info);
3276		return ret;
3277	}
 
3278
3279	btrfs_qgroup_rescan_resume(fs_info);
3280
3281	if (!fs_info->uuid_root) {
3282		btrfs_info(fs_info, "creating UUID tree");
3283		ret = btrfs_create_uuid_tree(fs_info);
3284		if (ret) {
3285			btrfs_warn(fs_info,
3286				"failed to create the UUID tree: %d", ret);
3287			close_ctree(fs_info);
3288			return ret;
3289		}
3290	} else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3291		   fs_info->generation !=
3292				btrfs_super_uuid_tree_generation(disk_super)) {
3293		btrfs_info(fs_info, "checking UUID tree");
3294		ret = btrfs_check_uuid_tree(fs_info);
3295		if (ret) {
3296			btrfs_warn(fs_info,
3297				"failed to check the UUID tree: %d", ret);
3298			close_ctree(fs_info);
3299			return ret;
3300		}
3301	} else {
3302		set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3303	}
 
3304	set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3305
3306	/*
3307	 * backuproot only affect mount behavior, and if open_ctree succeeded,
3308	 * no need to keep the flag
3309	 */
3310	btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3311
3312	return 0;
3313
3314fail_qgroup:
3315	btrfs_free_qgroup_config(fs_info);
3316fail_trans_kthread:
3317	kthread_stop(fs_info->transaction_kthread);
3318	btrfs_cleanup_transaction(fs_info);
3319	btrfs_free_fs_roots(fs_info);
3320fail_cleaner:
3321	kthread_stop(fs_info->cleaner_kthread);
3322
3323	/*
3324	 * make sure we're done with the btree inode before we stop our
3325	 * kthreads
3326	 */
3327	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3328
3329fail_sysfs:
3330	btrfs_sysfs_remove_mounted(fs_info);
3331
3332fail_fsdev_sysfs:
3333	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3334
3335fail_block_groups:
3336	btrfs_put_block_group_cache(fs_info);
3337
3338fail_tree_roots:
3339	free_root_pointers(fs_info, 1);
 
 
3340	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3341
3342fail_sb_buffer:
3343	btrfs_stop_all_workers(fs_info);
3344	btrfs_free_block_groups(fs_info);
3345fail_csum:
3346	btrfs_free_csum_hash(fs_info);
3347fail_alloc:
3348fail_iput:
3349	btrfs_mapping_tree_free(&fs_info->mapping_tree);
3350
3351	iput(fs_info->btree_inode);
3352fail_bio_counter:
3353	percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
3354fail_delalloc_bytes:
3355	percpu_counter_destroy(&fs_info->delalloc_bytes);
3356fail_dirty_metadata_bytes:
3357	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3358fail_dio_bytes:
3359	percpu_counter_destroy(&fs_info->dio_bytes);
3360fail_srcu:
3361	cleanup_srcu_struct(&fs_info->subvol_srcu);
3362fail:
3363	btrfs_free_stripe_hash_table(fs_info);
3364	btrfs_close_devices(fs_info->fs_devices);
3365	return err;
 
 
 
3366
3367recovery_tree_root:
3368	if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3369		goto fail_tree_roots;
 
 
 
3370
3371	free_root_pointers(fs_info, 0);
 
3372
3373	/* don't use the log in recovery mode, it won't be valid */
3374	btrfs_set_super_log_root(disk_super, 0);
 
 
 
 
 
 
 
 
 
 
3375
3376	/* we can't trust the free space cache either */
3377	btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
 
3378
3379	ret = next_root_backup(fs_info, fs_info->super_copy,
3380			       &num_backups_tried, &backup_index);
3381	if (ret == -1)
3382		goto fail_block_groups;
3383	goto retry_root_backup;
3384}
3385ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3386
3387static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
 
3388{
3389	if (uptodate) {
3390		set_buffer_uptodate(bh);
3391	} else {
3392		struct btrfs_device *device = (struct btrfs_device *)
3393			bh->b_private;
 
 
 
 
 
 
 
 
 
 
3394
3395		btrfs_warn_rl_in_rcu(device->fs_info,
3396				"lost page write due to IO error on %s",
3397					  rcu_str_deref(device->name));
3398		/* note, we don't set_buffer_write_io_error because we have
3399		 * our own ways of dealing with the IO errors
 
 
3400		 */
3401		clear_buffer_uptodate(bh);
3402		btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
 
3403	}
3404	unlock_buffer(bh);
3405	put_bh(bh);
3406}
3407
3408int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3409			struct buffer_head **bh_ret)
3410{
3411	struct buffer_head *bh;
3412	struct btrfs_super_block *super;
3413	u64 bytenr;
3414
3415	bytenr = btrfs_sb_offset(copy_num);
3416	if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3417		return -EINVAL;
3418
3419	bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3420	/*
3421	 * If we fail to read from the underlying devices, as of now
3422	 * the best option we have is to mark it EIO.
3423	 */
3424	if (!bh)
3425		return -EIO;
3426
3427	super = (struct btrfs_super_block *)bh->b_data;
3428	if (btrfs_super_bytenr(super) != bytenr ||
3429		    btrfs_super_magic(super) != BTRFS_MAGIC) {
3430		brelse(bh);
3431		return -EINVAL;
3432	}
3433
3434	*bh_ret = bh;
3435	return 0;
3436}
3437
3438
3439struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3440{
3441	struct buffer_head *bh;
3442	struct buffer_head *latest = NULL;
3443	struct btrfs_super_block *super;
3444	int i;
3445	u64 transid = 0;
3446	int ret = -EINVAL;
3447
3448	/* we would like to check all the supers, but that would make
3449	 * a btrfs mount succeed after a mkfs from a different FS.
3450	 * So, we need to add a special mount option to scan for
3451	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3452	 */
3453	for (i = 0; i < 1; i++) {
3454		ret = btrfs_read_dev_one_super(bdev, i, &bh);
3455		if (ret)
3456			continue;
3457
3458		super = (struct btrfs_super_block *)bh->b_data;
3459
3460		if (!latest || btrfs_super_generation(super) > transid) {
3461			brelse(latest);
3462			latest = bh;
 
 
3463			transid = btrfs_super_generation(super);
3464		} else {
3465			brelse(bh);
3466		}
3467	}
3468
3469	if (!latest)
3470		return ERR_PTR(ret);
3471
3472	return latest;
3473}
3474
3475/*
3476 * Write superblock @sb to the @device. Do not wait for completion, all the
3477 * buffer heads we write are pinned.
3478 *
3479 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3480 * the expected device size at commit time. Note that max_mirrors must be
3481 * same for write and wait phases.
3482 *
3483 * Return number of errors when buffer head is not found or submission fails.
3484 */
3485static int write_dev_supers(struct btrfs_device *device,
3486			    struct btrfs_super_block *sb, int max_mirrors)
3487{
3488	struct btrfs_fs_info *fs_info = device->fs_info;
 
3489	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3490	struct buffer_head *bh;
3491	int i;
3492	int ret;
3493	int errors = 0;
3494	u64 bytenr;
3495	int op_flags;
3496
3497	if (max_mirrors == 0)
3498		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3499
3500	shash->tfm = fs_info->csum_shash;
3501
3502	for (i = 0; i < max_mirrors; i++) {
3503		bytenr = btrfs_sb_offset(i);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3504		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3505		    device->commit_total_bytes)
3506			break;
3507
3508		btrfs_set_super_bytenr(sb, bytenr);
3509
3510		crypto_shash_init(shash);
3511		crypto_shash_update(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3512				    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3513		crypto_shash_final(shash, sb->csum);
3514
3515		/* One reference for us, and we leave it for the caller */
3516		bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3517			      BTRFS_SUPER_INFO_SIZE);
3518		if (!bh) {
3519			btrfs_err(device->fs_info,
3520			    "couldn't get super buffer head for bytenr %llu",
3521			    bytenr);
3522			errors++;
3523			continue;
3524		}
3525
3526		memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
 
3527
3528		/* one reference for submit_bh */
3529		get_bh(bh);
3530
3531		set_buffer_uptodate(bh);
3532		lock_buffer(bh);
3533		bh->b_end_io = btrfs_end_buffer_write_sync;
3534		bh->b_private = device;
 
 
 
 
 
 
 
 
 
3535
3536		/*
3537		 * we fua the first super.  The others we allow
3538		 * to go down lazy.
 
3539		 */
3540		op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3541		if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3542			op_flags |= REQ_FUA;
3543		ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3544		if (ret)
 
3545			errors++;
3546	}
3547	return errors < i ? 0 : -1;
3548}
3549
3550/*
3551 * Wait for write completion of superblocks done by write_dev_supers,
3552 * @max_mirrors same for write and wait phases.
3553 *
3554 * Return number of errors when buffer head is not found or not marked up to
3555 * date.
3556 */
3557static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3558{
3559	struct buffer_head *bh;
3560	int i;
3561	int errors = 0;
3562	bool primary_failed = false;
 
3563	u64 bytenr;
3564
3565	if (max_mirrors == 0)
3566		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3567
3568	for (i = 0; i < max_mirrors; i++) {
3569		bytenr = btrfs_sb_offset(i);
 
 
 
 
 
 
 
 
 
 
3570		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3571		    device->commit_total_bytes)
3572			break;
3573
3574		bh = __find_get_block(device->bdev,
3575				      bytenr / BTRFS_BDEV_BLOCKSIZE,
3576				      BTRFS_SUPER_INFO_SIZE);
3577		if (!bh) {
3578			errors++;
3579			if (i == 0)
3580				primary_failed = true;
3581			continue;
3582		}
3583		wait_on_buffer(bh);
3584		if (!buffer_uptodate(bh)) {
 
3585			errors++;
3586			if (i == 0)
3587				primary_failed = true;
3588		}
3589
3590		/* drop our reference */
3591		brelse(bh);
3592
3593		/* drop the reference from the writing run */
3594		brelse(bh);
3595	}
3596
3597	/* log error, force error return */
3598	if (primary_failed) {
3599		btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3600			  device->devid);
3601		return -1;
3602	}
3603
3604	return errors < i ? 0 : -1;
3605}
3606
3607/*
3608 * endio for the write_dev_flush, this will wake anyone waiting
3609 * for the barrier when it is done
3610 */
3611static void btrfs_end_empty_barrier(struct bio *bio)
3612{
 
3613	complete(bio->bi_private);
3614}
3615
3616/*
3617 * Submit a flush request to the device if it supports it. Error handling is
3618 * done in the waiting counterpart.
3619 */
3620static void write_dev_flush(struct btrfs_device *device)
3621{
3622	struct request_queue *q = bdev_get_queue(device->bdev);
3623	struct bio *bio = device->flush_bio;
3624
3625	if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3626		return;
3627
3628	bio_reset(bio);
 
3629	bio->bi_end_io = btrfs_end_empty_barrier;
3630	bio_set_dev(bio, device->bdev);
3631	bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3632	init_completion(&device->flush_wait);
3633	bio->bi_private = &device->flush_wait;
3634
3635	btrfsic_submit_bio(bio);
3636	set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3637}
3638
3639/*
3640 * If the flush bio has been submitted by write_dev_flush, wait for it.
 
3641 */
3642static blk_status_t wait_dev_flush(struct btrfs_device *device)
3643{
3644	struct bio *bio = device->flush_bio;
3645
3646	if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3647		return BLK_STS_OK;
3648
3649	clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3650	wait_for_completion_io(&device->flush_wait);
3651
3652	return bio->bi_status;
3653}
 
 
 
3654
3655static int check_barrier_error(struct btrfs_fs_info *fs_info)
3656{
3657	if (!btrfs_check_rw_degradable(fs_info, NULL))
3658		return -EIO;
3659	return 0;
3660}
3661
3662/*
3663 * send an empty flush down to each device in parallel,
3664 * then wait for them
3665 */
3666static int barrier_all_devices(struct btrfs_fs_info *info)
3667{
3668	struct list_head *head;
3669	struct btrfs_device *dev;
3670	int errors_wait = 0;
3671	blk_status_t ret;
3672
3673	lockdep_assert_held(&info->fs_devices->device_list_mutex);
3674	/* send down all the barriers */
3675	head = &info->fs_devices->devices;
3676	list_for_each_entry(dev, head, dev_list) {
3677		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3678			continue;
3679		if (!dev->bdev)
3680			continue;
3681		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3682		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3683			continue;
3684
3685		write_dev_flush(dev);
3686		dev->last_flush_error = BLK_STS_OK;
3687	}
3688
3689	/* wait for all the barriers */
3690	list_for_each_entry(dev, head, dev_list) {
3691		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3692			continue;
3693		if (!dev->bdev) {
3694			errors_wait++;
3695			continue;
3696		}
3697		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3698		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3699			continue;
3700
3701		ret = wait_dev_flush(dev);
3702		if (ret) {
3703			dev->last_flush_error = ret;
3704			btrfs_dev_stat_inc_and_print(dev,
3705					BTRFS_DEV_STAT_FLUSH_ERRS);
3706			errors_wait++;
3707		}
3708	}
3709
3710	if (errors_wait) {
3711		/*
3712		 * At some point we need the status of all disks
3713		 * to arrive at the volume status. So error checking
3714		 * is being pushed to a separate loop.
3715		 */
3716		return check_barrier_error(info);
3717	}
3718	return 0;
3719}
3720
3721int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3722{
3723	int raid_type;
3724	int min_tolerated = INT_MAX;
3725
3726	if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3727	    (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3728		min_tolerated = min_t(int, min_tolerated,
3729				    btrfs_raid_array[BTRFS_RAID_SINGLE].
3730				    tolerated_failures);
3731
3732	for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3733		if (raid_type == BTRFS_RAID_SINGLE)
3734			continue;
3735		if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3736			continue;
3737		min_tolerated = min_t(int, min_tolerated,
3738				    btrfs_raid_array[raid_type].
3739				    tolerated_failures);
3740	}
3741
3742	if (min_tolerated == INT_MAX) {
3743		pr_warn("BTRFS: unknown raid flag: %llu", flags);
3744		min_tolerated = 0;
3745	}
3746
3747	return min_tolerated;
3748}
3749
3750int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3751{
3752	struct list_head *head;
3753	struct btrfs_device *dev;
3754	struct btrfs_super_block *sb;
3755	struct btrfs_dev_item *dev_item;
3756	int ret;
3757	int do_barriers;
3758	int max_errors;
3759	int total_errors = 0;
3760	u64 flags;
3761
3762	do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3763
3764	/*
3765	 * max_mirrors == 0 indicates we're from commit_transaction,
3766	 * not from fsync where the tree roots in fs_info have not
3767	 * been consistent on disk.
3768	 */
3769	if (max_mirrors == 0)
3770		backup_super_roots(fs_info);
3771
3772	sb = fs_info->super_for_commit;
3773	dev_item = &sb->dev_item;
3774
3775	mutex_lock(&fs_info->fs_devices->device_list_mutex);
3776	head = &fs_info->fs_devices->devices;
3777	max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3778
3779	if (do_barriers) {
3780		ret = barrier_all_devices(fs_info);
3781		if (ret) {
3782			mutex_unlock(
3783				&fs_info->fs_devices->device_list_mutex);
3784			btrfs_handle_fs_error(fs_info, ret,
3785					      "errors while submitting device barriers.");
3786			return ret;
3787		}
3788	}
3789
3790	list_for_each_entry(dev, head, dev_list) {
3791		if (!dev->bdev) {
3792			total_errors++;
3793			continue;
3794		}
3795		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3796		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3797			continue;
3798
3799		btrfs_set_stack_device_generation(dev_item, 0);
3800		btrfs_set_stack_device_type(dev_item, dev->type);
3801		btrfs_set_stack_device_id(dev_item, dev->devid);
3802		btrfs_set_stack_device_total_bytes(dev_item,
3803						   dev->commit_total_bytes);
3804		btrfs_set_stack_device_bytes_used(dev_item,
3805						  dev->commit_bytes_used);
3806		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3807		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3808		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3809		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3810		memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3811		       BTRFS_FSID_SIZE);
3812
3813		flags = btrfs_super_flags(sb);
3814		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3815
3816		ret = btrfs_validate_write_super(fs_info, sb);
3817		if (ret < 0) {
3818			mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3819			btrfs_handle_fs_error(fs_info, -EUCLEAN,
3820				"unexpected superblock corruption detected");
3821			return -EUCLEAN;
3822		}
3823
3824		ret = write_dev_supers(dev, sb, max_mirrors);
3825		if (ret)
3826			total_errors++;
3827	}
3828	if (total_errors > max_errors) {
3829		btrfs_err(fs_info, "%d errors while writing supers",
3830			  total_errors);
3831		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3832
3833		/* FUA is masked off if unsupported and can't be the reason */
3834		btrfs_handle_fs_error(fs_info, -EIO,
3835				      "%d errors while writing supers",
3836				      total_errors);
3837		return -EIO;
3838	}
3839
3840	total_errors = 0;
3841	list_for_each_entry(dev, head, dev_list) {
3842		if (!dev->bdev)
3843			continue;
3844		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3845		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3846			continue;
3847
3848		ret = wait_dev_supers(dev, max_mirrors);
3849		if (ret)
3850			total_errors++;
3851	}
3852	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3853	if (total_errors > max_errors) {
3854		btrfs_handle_fs_error(fs_info, -EIO,
3855				      "%d errors while writing supers",
3856				      total_errors);
3857		return -EIO;
3858	}
3859	return 0;
3860}
3861
3862/* Drop a fs root from the radix tree and free it. */
3863void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3864				  struct btrfs_root *root)
3865{
 
 
3866	spin_lock(&fs_info->fs_roots_radix_lock);
3867	radix_tree_delete(&fs_info->fs_roots_radix,
3868			  (unsigned long)root->root_key.objectid);
 
 
3869	spin_unlock(&fs_info->fs_roots_radix_lock);
3870
3871	if (btrfs_root_refs(&root->root_item) == 0)
3872		synchronize_srcu(&fs_info->subvol_srcu);
3873
3874	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3875		btrfs_free_log(NULL, root);
3876		if (root->reloc_root) {
3877			free_extent_buffer(root->reloc_root->node);
3878			free_extent_buffer(root->reloc_root->commit_root);
3879			btrfs_put_fs_root(root->reloc_root);
3880			root->reloc_root = NULL;
3881		}
3882	}
3883
3884	if (root->free_ino_pinned)
3885		__btrfs_remove_free_space_cache(root->free_ino_pinned);
3886	if (root->free_ino_ctl)
3887		__btrfs_remove_free_space_cache(root->free_ino_ctl);
3888	btrfs_free_fs_root(root);
3889}
3890
3891void btrfs_free_fs_root(struct btrfs_root *root)
3892{
3893	iput(root->ino_cache_inode);
3894	WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3895	if (root->anon_dev)
3896		free_anon_bdev(root->anon_dev);
3897	if (root->subv_writers)
3898		btrfs_free_subvolume_writers(root->subv_writers);
3899	free_extent_buffer(root->node);
3900	free_extent_buffer(root->commit_root);
3901	kfree(root->free_ino_ctl);
3902	kfree(root->free_ino_pinned);
3903	btrfs_put_fs_root(root);
3904}
3905
3906int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3907{
3908	u64 root_objectid = 0;
3909	struct btrfs_root *gang[8];
3910	int i = 0;
3911	int err = 0;
3912	unsigned int ret = 0;
3913	int index;
3914
3915	while (1) {
3916		index = srcu_read_lock(&fs_info->subvol_srcu);
3917		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3918					     (void **)gang, root_objectid,
3919					     ARRAY_SIZE(gang));
3920		if (!ret) {
3921			srcu_read_unlock(&fs_info->subvol_srcu, index);
3922			break;
3923		}
3924		root_objectid = gang[ret - 1]->root_key.objectid + 1;
3925
3926		for (i = 0; i < ret; i++) {
3927			/* Avoid to grab roots in dead_roots */
3928			if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3929				gang[i] = NULL;
3930				continue;
3931			}
3932			/* grab all the search result for later use */
3933			gang[i] = btrfs_grab_fs_root(gang[i]);
3934		}
3935		srcu_read_unlock(&fs_info->subvol_srcu, index);
3936
3937		for (i = 0; i < ret; i++) {
3938			if (!gang[i])
3939				continue;
3940			root_objectid = gang[i]->root_key.objectid;
3941			err = btrfs_orphan_cleanup(gang[i]);
3942			if (err)
3943				break;
3944			btrfs_put_fs_root(gang[i]);
3945		}
3946		root_objectid++;
3947	}
3948
3949	/* release the uncleaned roots due to error */
3950	for (; i < ret; i++) {
3951		if (gang[i])
3952			btrfs_put_fs_root(gang[i]);
3953	}
3954	return err;
3955}
3956
3957int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3958{
3959	struct btrfs_root *root = fs_info->tree_root;
3960	struct btrfs_trans_handle *trans;
3961
3962	mutex_lock(&fs_info->cleaner_mutex);
3963	btrfs_run_delayed_iputs(fs_info);
3964	mutex_unlock(&fs_info->cleaner_mutex);
3965	wake_up_process(fs_info->cleaner_kthread);
3966
3967	/* wait until ongoing cleanup work done */
3968	down_write(&fs_info->cleanup_work_sem);
3969	up_write(&fs_info->cleanup_work_sem);
3970
3971	trans = btrfs_join_transaction(root);
3972	if (IS_ERR(trans))
3973		return PTR_ERR(trans);
3974	return btrfs_commit_transaction(trans);
3975}
3976
3977void close_ctree(struct btrfs_fs_info *fs_info)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3978{
3979	int ret;
3980
3981	set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3982	/*
3983	 * We don't want the cleaner to start new transactions, add more delayed
3984	 * iputs, etc. while we're closing. We can't use kthread_stop() yet
3985	 * because that frees the task_struct, and the transaction kthread might
3986	 * still try to wake up the cleaner.
3987	 */
3988	kthread_park(fs_info->cleaner_kthread);
3989
3990	/* wait for the qgroup rescan worker to stop */
3991	btrfs_qgroup_wait_for_completion(fs_info, false);
3992
3993	/* wait for the uuid_scan task to finish */
3994	down(&fs_info->uuid_tree_rescan_sem);
3995	/* avoid complains from lockdep et al., set sem back to initial state */
3996	up(&fs_info->uuid_tree_rescan_sem);
3997
3998	/* pause restriper - we want to resume on mount */
3999	btrfs_pause_balance(fs_info);
4000
4001	btrfs_dev_replace_suspend_for_unmount(fs_info);
4002
4003	btrfs_scrub_cancel(fs_info);
4004
4005	/* wait for any defraggers to finish */
4006	wait_event(fs_info->transaction_wait,
4007		   (atomic_read(&fs_info->defrag_running) == 0));
4008
4009	/* clear out the rbtree of defraggable inodes */
4010	btrfs_cleanup_defrag_inodes(fs_info);
4011
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4012	cancel_work_sync(&fs_info->async_reclaim_work);
 
 
 
 
 
4013
4014	if (!sb_rdonly(fs_info->sb)) {
4015		/*
4016		 * The cleaner kthread is stopped, so do one final pass over
4017		 * unused block groups.
4018		 */
4019		btrfs_delete_unused_bgs(fs_info);
4020
 
 
 
 
 
 
 
 
 
 
 
 
 
4021		ret = btrfs_commit_super(fs_info);
4022		if (ret)
4023			btrfs_err(fs_info, "commit super ret %d", ret);
4024	}
4025
4026	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4027	    test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4028		btrfs_error_commit_super(fs_info);
4029
4030	kthread_stop(fs_info->transaction_kthread);
4031	kthread_stop(fs_info->cleaner_kthread);
4032
4033	ASSERT(list_empty(&fs_info->delayed_iputs));
4034	set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4035
 
 
 
 
 
4036	btrfs_free_qgroup_config(fs_info);
4037	ASSERT(list_empty(&fs_info->delalloc_roots));
4038
4039	if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4040		btrfs_info(fs_info, "at unmount delalloc count %lld",
4041		       percpu_counter_sum(&fs_info->delalloc_bytes));
4042	}
4043
4044	if (percpu_counter_sum(&fs_info->dio_bytes))
4045		btrfs_info(fs_info, "at unmount dio bytes count %lld",
4046			   percpu_counter_sum(&fs_info->dio_bytes));
4047
4048	btrfs_sysfs_remove_mounted(fs_info);
4049	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4050
4051	btrfs_free_fs_roots(fs_info);
4052
4053	btrfs_put_block_group_cache(fs_info);
4054
4055	/*
4056	 * we must make sure there is not any read request to
4057	 * submit after we stopping all workers.
4058	 */
4059	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4060	btrfs_stop_all_workers(fs_info);
4061
4062	btrfs_free_block_groups(fs_info);
 
4063
4064	clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4065	free_root_pointers(fs_info, 1);
 
4066
4067	iput(fs_info->btree_inode);
 
 
 
 
 
 
 
4068
4069#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4070	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4071		btrfsic_unmount(fs_info->fs_devices);
4072#endif
4073
4074	btrfs_mapping_tree_free(&fs_info->mapping_tree);
4075	btrfs_close_devices(fs_info->fs_devices);
4076
4077	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4078	percpu_counter_destroy(&fs_info->delalloc_bytes);
4079	percpu_counter_destroy(&fs_info->dio_bytes);
4080	percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
4081	cleanup_srcu_struct(&fs_info->subvol_srcu);
4082
4083	btrfs_free_csum_hash(fs_info);
4084	btrfs_free_stripe_hash_table(fs_info);
4085	btrfs_free_ref_cache(fs_info);
4086}
4087
4088int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4089			  int atomic)
4090{
4091	int ret;
4092	struct inode *btree_inode = buf->pages[0]->mapping->host;
4093
4094	ret = extent_buffer_uptodate(buf);
4095	if (!ret)
4096		return ret;
4097
4098	ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4099				    parent_transid, atomic);
4100	if (ret == -EAGAIN)
4101		return ret;
4102	return !ret;
4103}
4104
4105void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
 
4106{
4107	struct btrfs_fs_info *fs_info;
4108	struct btrfs_root *root;
4109	u64 transid = btrfs_header_generation(buf);
4110	int was_dirty;
4111
4112#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4113	/*
4114	 * This is a fast path so only do this check if we have sanity tests
4115	 * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4116	 * outside of the sanity tests.
4117	 */
4118	if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4119		return;
4120#endif
4121	root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4122	fs_info = root->fs_info;
4123	btrfs_assert_tree_locked(buf);
4124	if (transid != fs_info->generation)
4125		WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4126			buf->start, transid, fs_info->generation);
4127	was_dirty = set_extent_buffer_dirty(buf);
4128	if (!was_dirty)
4129		percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4130					 buf->len,
4131					 fs_info->dirty_metadata_batch);
4132#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4133	/*
4134	 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4135	 * but item data not updated.
4136	 * So here we should only check item pointers, not item data.
4137	 */
4138	if (btrfs_header_level(buf) == 0 &&
4139	    btrfs_check_leaf_relaxed(buf)) {
4140		btrfs_print_leaf(buf);
4141		ASSERT(0);
4142	}
4143#endif
4144}
4145
4146static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4147					int flush_delayed)
4148{
4149	/*
4150	 * looks as though older kernels can get into trouble with
4151	 * this code, they end up stuck in balance_dirty_pages forever
4152	 */
4153	int ret;
4154
4155	if (current->flags & PF_MEMALLOC)
4156		return;
4157
4158	if (flush_delayed)
4159		btrfs_balance_delayed_items(fs_info);
4160
4161	ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4162				     BTRFS_DIRTY_METADATA_THRESH,
4163				     fs_info->dirty_metadata_batch);
4164	if (ret > 0) {
4165		balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4166	}
4167}
4168
4169void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4170{
4171	__btrfs_btree_balance_dirty(fs_info, 1);
4172}
4173
4174void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4175{
4176	__btrfs_btree_balance_dirty(fs_info, 0);
4177}
4178
4179int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4180		      struct btrfs_key *first_key)
4181{
4182	return btree_read_extent_buffer_pages(buf, parent_transid,
4183					      level, first_key);
4184}
4185
4186static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4187{
4188	/* cleanup FS via transaction */
4189	btrfs_cleanup_transaction(fs_info);
4190
4191	mutex_lock(&fs_info->cleaner_mutex);
4192	btrfs_run_delayed_iputs(fs_info);
4193	mutex_unlock(&fs_info->cleaner_mutex);
4194
4195	down_write(&fs_info->cleanup_work_sem);
4196	up_write(&fs_info->cleanup_work_sem);
4197}
4198
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4199static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4200{
4201	struct btrfs_ordered_extent *ordered;
4202
4203	spin_lock(&root->ordered_extent_lock);
4204	/*
4205	 * This will just short circuit the ordered completion stuff which will
4206	 * make sure the ordered extent gets properly cleaned up.
4207	 */
4208	list_for_each_entry(ordered, &root->ordered_extents,
4209			    root_extent_list)
4210		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4211	spin_unlock(&root->ordered_extent_lock);
4212}
4213
4214static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4215{
4216	struct btrfs_root *root;
4217	struct list_head splice;
4218
4219	INIT_LIST_HEAD(&splice);
4220
4221	spin_lock(&fs_info->ordered_root_lock);
4222	list_splice_init(&fs_info->ordered_roots, &splice);
4223	while (!list_empty(&splice)) {
4224		root = list_first_entry(&splice, struct btrfs_root,
4225					ordered_root);
4226		list_move_tail(&root->ordered_root,
4227			       &fs_info->ordered_roots);
4228
4229		spin_unlock(&fs_info->ordered_root_lock);
4230		btrfs_destroy_ordered_extents(root);
4231
4232		cond_resched();
4233		spin_lock(&fs_info->ordered_root_lock);
4234	}
4235	spin_unlock(&fs_info->ordered_root_lock);
4236
4237	/*
4238	 * We need this here because if we've been flipped read-only we won't
4239	 * get sync() from the umount, so we need to make sure any ordered
4240	 * extents that haven't had their dirty pages IO start writeout yet
4241	 * actually get run and error out properly.
4242	 */
4243	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4244}
4245
4246static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4247				      struct btrfs_fs_info *fs_info)
4248{
4249	struct rb_node *node;
4250	struct btrfs_delayed_ref_root *delayed_refs;
4251	struct btrfs_delayed_ref_node *ref;
4252	int ret = 0;
4253
4254	delayed_refs = &trans->delayed_refs;
4255
4256	spin_lock(&delayed_refs->lock);
4257	if (atomic_read(&delayed_refs->num_entries) == 0) {
4258		spin_unlock(&delayed_refs->lock);
4259		btrfs_info(fs_info, "delayed_refs has NO entry");
4260		return ret;
4261	}
4262
4263	while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4264		struct btrfs_delayed_ref_head *head;
4265		struct rb_node *n;
4266		bool pin_bytes = false;
4267
4268		head = rb_entry(node, struct btrfs_delayed_ref_head,
4269				href_node);
4270		if (btrfs_delayed_ref_lock(delayed_refs, head))
4271			continue;
4272
4273		spin_lock(&head->lock);
4274		while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4275			ref = rb_entry(n, struct btrfs_delayed_ref_node,
4276				       ref_node);
4277			ref->in_tree = 0;
4278			rb_erase_cached(&ref->ref_node, &head->ref_tree);
4279			RB_CLEAR_NODE(&ref->ref_node);
4280			if (!list_empty(&ref->add_list))
4281				list_del(&ref->add_list);
4282			atomic_dec(&delayed_refs->num_entries);
4283			btrfs_put_delayed_ref(ref);
 
4284		}
4285		if (head->must_insert_reserved)
4286			pin_bytes = true;
4287		btrfs_free_delayed_extent_op(head->extent_op);
4288		btrfs_delete_ref_head(delayed_refs, head);
4289		spin_unlock(&head->lock);
4290		spin_unlock(&delayed_refs->lock);
4291		mutex_unlock(&head->mutex);
4292
4293		if (pin_bytes)
4294			btrfs_pin_extent(fs_info, head->bytenr,
4295					 head->num_bytes, 1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4296		btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4297		btrfs_put_delayed_ref_head(head);
4298		cond_resched();
4299		spin_lock(&delayed_refs->lock);
4300	}
 
4301
4302	spin_unlock(&delayed_refs->lock);
4303
4304	return ret;
4305}
4306
4307static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4308{
4309	struct btrfs_inode *btrfs_inode;
4310	struct list_head splice;
4311
4312	INIT_LIST_HEAD(&splice);
4313
4314	spin_lock(&root->delalloc_lock);
4315	list_splice_init(&root->delalloc_inodes, &splice);
4316
4317	while (!list_empty(&splice)) {
4318		struct inode *inode = NULL;
4319		btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4320					       delalloc_inodes);
4321		__btrfs_del_delalloc_inode(root, btrfs_inode);
4322		spin_unlock(&root->delalloc_lock);
4323
4324		/*
4325		 * Make sure we get a live inode and that it'll not disappear
4326		 * meanwhile.
4327		 */
4328		inode = igrab(&btrfs_inode->vfs_inode);
4329		if (inode) {
 
 
 
4330			invalidate_inode_pages2(inode->i_mapping);
 
4331			iput(inode);
4332		}
4333		spin_lock(&root->delalloc_lock);
4334	}
4335	spin_unlock(&root->delalloc_lock);
4336}
4337
4338static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4339{
4340	struct btrfs_root *root;
4341	struct list_head splice;
4342
4343	INIT_LIST_HEAD(&splice);
4344
4345	spin_lock(&fs_info->delalloc_root_lock);
4346	list_splice_init(&fs_info->delalloc_roots, &splice);
4347	while (!list_empty(&splice)) {
4348		root = list_first_entry(&splice, struct btrfs_root,
4349					 delalloc_root);
4350		root = btrfs_grab_fs_root(root);
4351		BUG_ON(!root);
4352		spin_unlock(&fs_info->delalloc_root_lock);
4353
4354		btrfs_destroy_delalloc_inodes(root);
4355		btrfs_put_fs_root(root);
4356
4357		spin_lock(&fs_info->delalloc_root_lock);
4358	}
4359	spin_unlock(&fs_info->delalloc_root_lock);
4360}
4361
4362static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4363					struct extent_io_tree *dirty_pages,
4364					int mark)
4365{
4366	int ret;
4367	struct extent_buffer *eb;
4368	u64 start = 0;
4369	u64 end;
4370
4371	while (1) {
4372		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4373					    mark, NULL);
4374		if (ret)
4375			break;
4376
4377		clear_extent_bits(dirty_pages, start, end, mark);
4378		while (start <= end) {
4379			eb = find_extent_buffer(fs_info, start);
4380			start += fs_info->nodesize;
4381			if (!eb)
4382				continue;
 
 
4383			wait_on_extent_buffer_writeback(eb);
 
 
4384
4385			if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4386					       &eb->bflags))
4387				clear_extent_buffer_dirty(eb);
4388			free_extent_buffer_stale(eb);
4389		}
4390	}
4391
4392	return ret;
4393}
4394
4395static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4396				       struct extent_io_tree *pinned_extents)
4397{
4398	struct extent_io_tree *unpin;
4399	u64 start;
4400	u64 end;
4401	int ret;
4402	bool loop = true;
4403
4404	unpin = pinned_extents;
4405again:
4406	while (1) {
4407		struct extent_state *cached_state = NULL;
4408
4409		/*
4410		 * The btrfs_finish_extent_commit() may get the same range as
4411		 * ours between find_first_extent_bit and clear_extent_dirty.
4412		 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4413		 * the same extent range.
4414		 */
4415		mutex_lock(&fs_info->unused_bg_unpin_mutex);
4416		ret = find_first_extent_bit(unpin, 0, &start, &end,
4417					    EXTENT_DIRTY, &cached_state);
4418		if (ret) {
4419			mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4420			break;
4421		}
4422
4423		clear_extent_dirty(unpin, start, end, &cached_state);
4424		free_extent_state(cached_state);
4425		btrfs_error_unpin_extent_range(fs_info, start, end);
4426		mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4427		cond_resched();
4428	}
4429
4430	if (loop) {
4431		if (unpin == &fs_info->freed_extents[0])
4432			unpin = &fs_info->freed_extents[1];
4433		else
4434			unpin = &fs_info->freed_extents[0];
4435		loop = false;
4436		goto again;
4437	}
4438
4439	return 0;
4440}
4441
4442static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4443{
4444	struct inode *inode;
4445
4446	inode = cache->io_ctl.inode;
4447	if (inode) {
 
 
 
4448		invalidate_inode_pages2(inode->i_mapping);
 
 
4449		BTRFS_I(inode)->generation = 0;
4450		cache->io_ctl.inode = NULL;
4451		iput(inode);
4452	}
 
4453	btrfs_put_block_group(cache);
4454}
4455
4456void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4457			     struct btrfs_fs_info *fs_info)
4458{
4459	struct btrfs_block_group_cache *cache;
4460
4461	spin_lock(&cur_trans->dirty_bgs_lock);
4462	while (!list_empty(&cur_trans->dirty_bgs)) {
4463		cache = list_first_entry(&cur_trans->dirty_bgs,
4464					 struct btrfs_block_group_cache,
4465					 dirty_list);
4466
4467		if (!list_empty(&cache->io_list)) {
4468			spin_unlock(&cur_trans->dirty_bgs_lock);
4469			list_del_init(&cache->io_list);
4470			btrfs_cleanup_bg_io(cache);
4471			spin_lock(&cur_trans->dirty_bgs_lock);
4472		}
4473
4474		list_del_init(&cache->dirty_list);
4475		spin_lock(&cache->lock);
4476		cache->disk_cache_state = BTRFS_DC_ERROR;
4477		spin_unlock(&cache->lock);
4478
4479		spin_unlock(&cur_trans->dirty_bgs_lock);
4480		btrfs_put_block_group(cache);
4481		btrfs_delayed_refs_rsv_release(fs_info, 1);
4482		spin_lock(&cur_trans->dirty_bgs_lock);
4483	}
4484	spin_unlock(&cur_trans->dirty_bgs_lock);
4485
4486	/*
4487	 * Refer to the definition of io_bgs member for details why it's safe
4488	 * to use it without any locking
4489	 */
4490	while (!list_empty(&cur_trans->io_bgs)) {
4491		cache = list_first_entry(&cur_trans->io_bgs,
4492					 struct btrfs_block_group_cache,
4493					 io_list);
4494
4495		list_del_init(&cache->io_list);
4496		spin_lock(&cache->lock);
4497		cache->disk_cache_state = BTRFS_DC_ERROR;
4498		spin_unlock(&cache->lock);
4499		btrfs_cleanup_bg_io(cache);
4500	}
4501}
4502
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4503void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4504				   struct btrfs_fs_info *fs_info)
4505{
4506	struct btrfs_device *dev, *tmp;
4507
4508	btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4509	ASSERT(list_empty(&cur_trans->dirty_bgs));
4510	ASSERT(list_empty(&cur_trans->io_bgs));
4511
4512	list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4513				 post_commit_list) {
4514		list_del_init(&dev->post_commit_list);
4515	}
4516
4517	btrfs_destroy_delayed_refs(cur_trans, fs_info);
4518
4519	cur_trans->state = TRANS_STATE_COMMIT_START;
4520	wake_up(&fs_info->transaction_blocked_wait);
4521
4522	cur_trans->state = TRANS_STATE_UNBLOCKED;
4523	wake_up(&fs_info->transaction_wait);
4524
4525	btrfs_destroy_delayed_inodes(fs_info);
4526	btrfs_assert_delayed_root_empty(fs_info);
4527
4528	btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4529				     EXTENT_DIRTY);
4530	btrfs_destroy_pinned_extent(fs_info,
4531				    fs_info->pinned_extents);
 
4532
4533	cur_trans->state =TRANS_STATE_COMPLETED;
4534	wake_up(&cur_trans->commit_wait);
4535}
4536
4537static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4538{
4539	struct btrfs_transaction *t;
4540
4541	mutex_lock(&fs_info->transaction_kthread_mutex);
4542
4543	spin_lock(&fs_info->trans_lock);
4544	while (!list_empty(&fs_info->trans_list)) {
4545		t = list_first_entry(&fs_info->trans_list,
4546				     struct btrfs_transaction, list);
4547		if (t->state >= TRANS_STATE_COMMIT_START) {
4548			refcount_inc(&t->use_count);
4549			spin_unlock(&fs_info->trans_lock);
4550			btrfs_wait_for_commit(fs_info, t->transid);
4551			btrfs_put_transaction(t);
4552			spin_lock(&fs_info->trans_lock);
4553			continue;
4554		}
4555		if (t == fs_info->running_transaction) {
4556			t->state = TRANS_STATE_COMMIT_DOING;
4557			spin_unlock(&fs_info->trans_lock);
4558			/*
4559			 * We wait for 0 num_writers since we don't hold a trans
4560			 * handle open currently for this transaction.
4561			 */
4562			wait_event(t->writer_wait,
4563				   atomic_read(&t->num_writers) == 0);
4564		} else {
4565			spin_unlock(&fs_info->trans_lock);
4566		}
4567		btrfs_cleanup_one_transaction(t, fs_info);
4568
4569		spin_lock(&fs_info->trans_lock);
4570		if (t == fs_info->running_transaction)
4571			fs_info->running_transaction = NULL;
4572		list_del_init(&t->list);
4573		spin_unlock(&fs_info->trans_lock);
4574
4575		btrfs_put_transaction(t);
4576		trace_btrfs_transaction_commit(fs_info->tree_root);
4577		spin_lock(&fs_info->trans_lock);
4578	}
4579	spin_unlock(&fs_info->trans_lock);
4580	btrfs_destroy_all_ordered_extents(fs_info);
4581	btrfs_destroy_delayed_inodes(fs_info);
4582	btrfs_assert_delayed_root_empty(fs_info);
4583	btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4584	btrfs_destroy_all_delalloc_inodes(fs_info);
 
4585	mutex_unlock(&fs_info->transaction_kthread_mutex);
4586
4587	return 0;
4588}
4589
4590static const struct extent_io_ops btree_extent_io_ops = {
4591	/* mandatory callbacks */
4592	.submit_bio_hook = btree_submit_bio_hook,
4593	.readpage_end_io_hook = btree_readpage_end_io_hook,
4594};
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
v6.8
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 2007 Oracle.  All rights reserved.
   4 */
   5
   6#include <linux/fs.h>
   7#include <linux/blkdev.h>
   8#include <linux/radix-tree.h>
   9#include <linux/writeback.h>
 
  10#include <linux/workqueue.h>
  11#include <linux/kthread.h>
  12#include <linux/slab.h>
  13#include <linux/migrate.h>
  14#include <linux/ratelimit.h>
  15#include <linux/uuid.h>
  16#include <linux/semaphore.h>
  17#include <linux/error-injection.h>
  18#include <linux/crc32c.h>
  19#include <linux/sched/mm.h>
  20#include <asm/unaligned.h>
  21#include <crypto/hash.h>
  22#include "ctree.h"
  23#include "disk-io.h"
  24#include "transaction.h"
  25#include "btrfs_inode.h"
  26#include "bio.h"
  27#include "print-tree.h"
  28#include "locking.h"
  29#include "tree-log.h"
  30#include "free-space-cache.h"
  31#include "free-space-tree.h"
 
 
  32#include "rcu-string.h"
  33#include "dev-replace.h"
  34#include "raid56.h"
  35#include "sysfs.h"
  36#include "qgroup.h"
  37#include "compression.h"
  38#include "tree-checker.h"
  39#include "ref-verify.h"
  40#include "block-group.h"
  41#include "discard.h"
  42#include "space-info.h"
  43#include "zoned.h"
  44#include "subpage.h"
  45#include "fs.h"
  46#include "accessors.h"
  47#include "extent-tree.h"
  48#include "root-tree.h"
  49#include "defrag.h"
  50#include "uuid-tree.h"
  51#include "relocation.h"
  52#include "scrub.h"
  53#include "super.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, int 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	int num_folios = num_extent_folios(eb);
 187	int ret = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 188
 189	if (sb_rdonly(fs_info->sb))
 190		return -EROFS;
 
 
 
 
 
 
 191
 192	for (int i = 0; i < num_folios; i++) {
 193		struct folio *folio = eb->folios[i];
 194		u64 start = max_t(u64, eb->start, folio_pos(folio));
 195		u64 end = min_t(u64, eb->start + eb->len,
 196				folio_pos(folio) + folio_size(folio));
 197		u32 len = end - start;
 198
 199		ret = btrfs_repair_io_failure(fs_info, 0, start, len,
 200					      start, folio, offset_in_folio(folio, start),
 201					      mirror_num);
 202		if (ret)
 203			break;
 204	}
 205
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 206	return ret;
 207}
 208
 209/*
 210 * helper to read a given tree block, doing retries as required when
 211 * the checksums don't match and we have alternate mirrors to try.
 212 *
 213 * @check:		expected tree parentness check, see the comments of the
 214 *			structure for details.
 
 215 */
 216int btrfs_read_extent_buffer(struct extent_buffer *eb,
 217			     struct btrfs_tree_parent_check *check)
 
 218{
 219	struct btrfs_fs_info *fs_info = eb->fs_info;
 
 220	int failed = 0;
 221	int ret;
 222	int num_copies = 0;
 223	int mirror_num = 0;
 224	int failed_mirror = 0;
 225
 226	ASSERT(check);
 227
 228	while (1) {
 229		clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 230		ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check);
 231		if (!ret)
 232			break;
 
 
 
 
 
 
 
 
 233
 234		num_copies = btrfs_num_copies(fs_info,
 235					      eb->start, eb->len);
 236		if (num_copies == 1)
 237			break;
 238
 239		if (!failed_mirror) {
 240			failed = 1;
 241			failed_mirror = eb->read_mirror;
 242		}
 243
 244		mirror_num++;
 245		if (mirror_num == failed_mirror)
 246			mirror_num++;
 247
 248		if (mirror_num > num_copies)
 249			break;
 250	}
 251
 252	if (failed && !ret && failed_mirror)
 253		btrfs_repair_eb_io_failure(eb, failed_mirror);
 254
 255	return ret;
 256}
 257
 258/*
 259 * Checksum a dirty tree block before IO.
 
 260 */
 261blk_status_t btree_csum_one_bio(struct btrfs_bio *bbio)
 
 262{
 263	struct extent_buffer *eb = bbio->private;
 264	struct btrfs_fs_info *fs_info = eb->fs_info;
 265	u64 found_start = btrfs_header_bytenr(eb);
 266	u64 last_trans;
 267	u8 result[BTRFS_CSUM_SIZE];
 
 
 268	int ret;
 269
 270	/* Btree blocks are always contiguous on disk. */
 271	if (WARN_ON_ONCE(bbio->file_offset != eb->start))
 272		return BLK_STS_IOERR;
 273	if (WARN_ON_ONCE(bbio->bio.bi_iter.bi_size != eb->len))
 274		return BLK_STS_IOERR;
 275
 
 276	/*
 277	 * If an extent_buffer is marked as EXTENT_BUFFER_ZONED_ZEROOUT, don't
 278	 * checksum it but zero-out its content. This is done to preserve
 279	 * ordering of I/O without unnecessarily writing out data.
 280	 */
 281	if (test_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags)) {
 282		memzero_extent_buffer(eb, 0, eb->len);
 283		return BLK_STS_OK;
 284	}
 285
 286	if (WARN_ON_ONCE(found_start != eb->start))
 287		return BLK_STS_IOERR;
 288	if (WARN_ON(!btrfs_folio_test_uptodate(fs_info, eb->folios[0],
 289					       eb->start, eb->len)))
 290		return BLK_STS_IOERR;
 291
 292	ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
 293				    offsetof(struct btrfs_header, fsid),
 294				    BTRFS_FSID_SIZE) == 0);
 295	csum_tree_block(eb, result);
 296
 297	if (btrfs_header_level(eb))
 298		ret = btrfs_check_node(eb);
 299	else
 300		ret = btrfs_check_leaf(eb);
 301
 302	if (ret < 0)
 303		goto error;
 
 
 
 
 
 304
 305	/*
 306	 * Also check the generation, the eb reached here must be newer than
 307	 * last committed. Or something seriously wrong happened.
 308	 */
 309	last_trans = btrfs_get_last_trans_committed(fs_info);
 310	if (unlikely(btrfs_header_generation(eb) <= last_trans)) {
 311		ret = -EUCLEAN;
 312		btrfs_err(fs_info,
 313			"block=%llu bad generation, have %llu expect > %llu",
 314			  eb->start, btrfs_header_generation(eb), last_trans);
 315		goto error;
 316	}
 317	write_extent_buffer(eb, result, 0, fs_info->csum_size);
 318	return BLK_STS_OK;
 319
 320error:
 321	btrfs_print_tree(eb, 0);
 322	btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
 323		  eb->start);
 324	/*
 325	 * Be noisy if this is an extent buffer from a log tree. We don't abort
 326	 * a transaction in case there's a bad log tree extent buffer, we just
 327	 * fallback to a transaction commit. Still we want to know when there is
 328	 * a bad log tree extent buffer, as that may signal a bug somewhere.
 329	 */
 330	WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
 331		btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
 332	return errno_to_blk_status(ret);
 333}
 334
 335static bool check_tree_block_fsid(struct extent_buffer *eb)
 336{
 337	struct btrfs_fs_info *fs_info = eb->fs_info;
 338	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
 339	u8 fsid[BTRFS_FSID_SIZE];
 
 340
 341	read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
 342			   BTRFS_FSID_SIZE);
 
 343
 344	/*
 345	 * alloc_fsid_devices() copies the fsid into fs_devices::metadata_uuid.
 346	 * This is then overwritten by metadata_uuid if it is present in the
 347	 * device_list_add(). The same true for a seed device as well. So use of
 348	 * fs_devices::metadata_uuid is appropriate here.
 349	 */
 350	if (memcmp(fsid, fs_info->fs_devices->metadata_uuid, BTRFS_FSID_SIZE) == 0)
 351		return false;
 
 
 352
 353	list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
 354		if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
 355			return false;
 356
 357	return true;
 
 
 358}
 359
 360/* Do basic extent buffer checks at read time */
 361int btrfs_validate_extent_buffer(struct extent_buffer *eb,
 362				 struct btrfs_tree_parent_check *check)
 363{
 364	struct btrfs_fs_info *fs_info = eb->fs_info;
 365	u64 found_start;
 366	const u32 csum_size = fs_info->csum_size;
 367	u8 found_level;
 
 
 
 
 368	u8 result[BTRFS_CSUM_SIZE];
 369	const u8 *header_csum;
 370	int ret = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 371
 372	ASSERT(check);
 
 
 
 
 373
 374	found_start = btrfs_header_bytenr(eb);
 375	if (found_start != eb->start) {
 376		btrfs_err_rl(fs_info,
 377			"bad tree block start, mirror %u want %llu have %llu",
 378			     eb->read_mirror, eb->start, found_start);
 379		ret = -EIO;
 380		goto out;
 381	}
 382	if (check_tree_block_fsid(eb)) {
 383		btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
 384			     eb->start, eb->read_mirror);
 385		ret = -EIO;
 386		goto out;
 387	}
 388	found_level = btrfs_header_level(eb);
 389	if (found_level >= BTRFS_MAX_LEVEL) {
 390		btrfs_err(fs_info,
 391			"bad tree block level, mirror %u level %d on logical %llu",
 392			eb->read_mirror, btrfs_header_level(eb), eb->start);
 393		ret = -EIO;
 394		goto out;
 395	}
 396
 397	csum_tree_block(eb, result);
 398	header_csum = folio_address(eb->folios[0]) +
 399		get_eb_offset_in_folio(eb, offsetof(struct btrfs_header, csum));
 
 
 
 
 
 
 
 400
 401	if (memcmp(result, header_csum, csum_size) != 0) {
 
 
 402		btrfs_warn_rl(fs_info,
 403"checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d",
 404			      eb->start, eb->read_mirror,
 405			      CSUM_FMT_VALUE(csum_size, header_csum),
 406			      CSUM_FMT_VALUE(csum_size, result),
 407			      btrfs_header_level(eb));
 408		ret = -EUCLEAN;
 409		goto out;
 410	}
 411
 412	if (found_level != check->level) {
 413		btrfs_err(fs_info,
 414		"level verify failed on logical %llu mirror %u wanted %u found %u",
 415			  eb->start, eb->read_mirror, check->level, found_level);
 
 
 
 416		ret = -EIO;
 417		goto out;
 418	}
 419	if (unlikely(check->transid &&
 420		     btrfs_header_generation(eb) != check->transid)) {
 421		btrfs_err_rl(eb->fs_info,
 422"parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
 423				eb->start, eb->read_mirror, check->transid,
 424				btrfs_header_generation(eb));
 425		ret = -EIO;
 426		goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 427	}
 428	if (check->has_first_key) {
 429		struct btrfs_key *expect_key = &check->first_key;
 430		struct btrfs_key found_key;
 
 431
 432		if (found_level)
 433			btrfs_node_key_to_cpu(eb, &found_key, 0);
 434		else
 435			btrfs_item_key_to_cpu(eb, &found_key, 0);
 436		if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
 437			btrfs_err(fs_info,
 438"tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
 439				  eb->start, check->transid,
 440				  expect_key->objectid,
 441				  expect_key->type, expect_key->offset,
 442				  found_key.objectid, found_key.type,
 443				  found_key.offset);
 444			ret = -EUCLEAN;
 445			goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 446		}
 447	}
 448	if (check->owner_root) {
 449		ret = btrfs_check_eb_owner(eb, check->owner_root);
 450		if (ret < 0)
 451			goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 452	}
 453
 
 
 
 
 
 
 454	/*
 455	 * If this is a leaf block and it is corrupt, set the corrupt bit so
 456	 * that we don't try and read the other copies of this block, just
 457	 * return -EIO.
 458	 */
 459	if (found_level == 0 && btrfs_check_leaf(eb)) {
 460		set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 461		ret = -EIO;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 462	}
 463
 464	if (found_level > 0 && btrfs_check_node(eb))
 465		ret = -EIO;
 
 466
 467	if (ret)
 468		btrfs_err(fs_info,
 469		"read time tree block corruption detected on logical %llu mirror %u",
 470			  eb->start, eb->read_mirror);
 471out:
 472	return ret;
 473}
 474
 475#ifdef CONFIG_MIGRATION
 476static int btree_migrate_folio(struct address_space *mapping,
 477		struct folio *dst, struct folio *src, enum migrate_mode mode)
 
 478{
 479	/*
 480	 * we can't safely write a btree page from here,
 481	 * we haven't done the locking hook
 482	 */
 483	if (folio_test_dirty(src))
 484		return -EAGAIN;
 485	/*
 486	 * Buffers may be managed in a filesystem specific way.
 487	 * We must have no buffers or drop them.
 488	 */
 489	if (folio_get_private(src) &&
 490	    !filemap_release_folio(src, GFP_KERNEL))
 491		return -EAGAIN;
 492	return migrate_folio(mapping, dst, src, mode);
 493}
 494#else
 495#define btree_migrate_folio NULL
 496#endif
 497
 
 498static int btree_writepages(struct address_space *mapping,
 499			    struct writeback_control *wbc)
 500{
 501	struct btrfs_fs_info *fs_info;
 502	int ret;
 503
 504	if (wbc->sync_mode == WB_SYNC_NONE) {
 505
 506		if (wbc->for_kupdate)
 507			return 0;
 508
 509		fs_info = BTRFS_I(mapping->host)->root->fs_info;
 510		/* this is a bit racy, but that's ok */
 511		ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 512					     BTRFS_DIRTY_METADATA_THRESH,
 513					     fs_info->dirty_metadata_batch);
 514		if (ret < 0)
 515			return 0;
 516	}
 517	return btree_write_cache_pages(mapping, wbc);
 518}
 519
 520static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
 521{
 522	if (folio_test_writeback(folio) || folio_test_dirty(folio))
 523		return false;
 
 
 
 
 
 
 
 524
 525	return try_release_extent_buffer(&folio->page);
 526}
 527
 528static void btree_invalidate_folio(struct folio *folio, size_t offset,
 529				 size_t length)
 530{
 531	struct extent_io_tree *tree;
 532	tree = &BTRFS_I(folio->mapping->host)->io_tree;
 533	extent_invalidate_folio(tree, folio, offset);
 534	btree_release_folio(folio, GFP_NOFS);
 535	if (folio_get_private(folio)) {
 536		btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info,
 537			   "folio private not zero on folio %llu",
 538			   (unsigned long long)folio_pos(folio));
 539		folio_detach_private(folio);
 
 
 540	}
 541}
 542
 
 
 543#ifdef DEBUG
 544static bool btree_dirty_folio(struct address_space *mapping,
 545		struct folio *folio)
 546{
 547	struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
 548	struct btrfs_subpage_info *spi = fs_info->subpage_info;
 549	struct btrfs_subpage *subpage;
 550	struct extent_buffer *eb;
 551	int cur_bit = 0;
 552	u64 page_start = folio_pos(folio);
 553
 554	if (fs_info->sectorsize == PAGE_SIZE) {
 555		eb = folio_get_private(folio);
 556		BUG_ON(!eb);
 557		BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 558		BUG_ON(!atomic_read(&eb->refs));
 559		btrfs_assert_tree_write_locked(eb);
 560		return filemap_dirty_folio(mapping, folio);
 561	}
 562
 563	ASSERT(spi);
 564	subpage = folio_get_private(folio);
 565
 566	for (cur_bit = spi->dirty_offset;
 567	     cur_bit < spi->dirty_offset + spi->bitmap_nr_bits;
 568	     cur_bit++) {
 569		unsigned long flags;
 570		u64 cur;
 571
 572		spin_lock_irqsave(&subpage->lock, flags);
 573		if (!test_bit(cur_bit, subpage->bitmaps)) {
 574			spin_unlock_irqrestore(&subpage->lock, flags);
 575			continue;
 576		}
 577		spin_unlock_irqrestore(&subpage->lock, flags);
 578		cur = page_start + cur_bit * fs_info->sectorsize;
 579
 580		eb = find_extent_buffer(fs_info, cur);
 581		ASSERT(eb);
 582		ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 583		ASSERT(atomic_read(&eb->refs));
 584		btrfs_assert_tree_write_locked(eb);
 585		free_extent_buffer(eb);
 586
 587		cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits) - 1;
 588	}
 589	return filemap_dirty_folio(mapping, folio);
 590}
 591#else
 592#define btree_dirty_folio filemap_dirty_folio
 593#endif
 594
 595static const struct address_space_operations btree_aops = {
 
 596	.writepages	= btree_writepages,
 597	.release_folio	= btree_release_folio,
 598	.invalidate_folio = btree_invalidate_folio,
 599	.migrate_folio	= btree_migrate_folio,
 600	.dirty_folio	= btree_dirty_folio,
 
 
 601};
 602
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 603struct extent_buffer *btrfs_find_create_tree_block(
 604						struct btrfs_fs_info *fs_info,
 605						u64 bytenr, u64 owner_root,
 606						int level)
 607{
 608	if (btrfs_is_testing(fs_info))
 609		return alloc_test_extent_buffer(fs_info, bytenr);
 610	return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
 611}
 612
 613/*
 614 * Read tree block at logical address @bytenr and do variant basic but critical
 615 * verification.
 616 *
 617 * @check:		expected tree parentness check, see comments of the
 618 *			structure for details.
 
 619 */
 620struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
 621				      struct btrfs_tree_parent_check *check)
 
 622{
 623	struct extent_buffer *buf = NULL;
 624	int ret;
 625
 626	ASSERT(check);
 627
 628	buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
 629					   check->level);
 630	if (IS_ERR(buf))
 631		return buf;
 632
 633	ret = btrfs_read_extent_buffer(buf, check);
 
 634	if (ret) {
 635		free_extent_buffer_stale(buf);
 636		return ERR_PTR(ret);
 637	}
 638	if (btrfs_check_eb_owner(buf, check->owner_root)) {
 639		free_extent_buffer_stale(buf);
 640		return ERR_PTR(-EUCLEAN);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 641	}
 642	return buf;
 643
 
 
 
 
 
 
 
 
 
 644}
 645
 646static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
 647			 u64 objectid)
 648{
 649	bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
 650
 651	memset(&root->root_key, 0, sizeof(root->root_key));
 652	memset(&root->root_item, 0, sizeof(root->root_item));
 653	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 654	root->fs_info = fs_info;
 655	root->root_key.objectid = objectid;
 656	root->node = NULL;
 657	root->commit_root = NULL;
 658	root->state = 0;
 659	RB_CLEAR_NODE(&root->rb_node);
 660
 661	root->last_trans = 0;
 662	root->free_objectid = 0;
 663	root->nr_delalloc_inodes = 0;
 664	root->nr_ordered_extents = 0;
 665	root->inode_tree = RB_ROOT;
 666	/* GFP flags are compatible with XA_FLAGS_*. */
 667	xa_init_flags(&root->delayed_nodes, GFP_ATOMIC);
 668
 669	btrfs_init_root_block_rsv(root);
 670
 671	INIT_LIST_HEAD(&root->dirty_list);
 672	INIT_LIST_HEAD(&root->root_list);
 673	INIT_LIST_HEAD(&root->delalloc_inodes);
 674	INIT_LIST_HEAD(&root->delalloc_root);
 675	INIT_LIST_HEAD(&root->ordered_extents);
 676	INIT_LIST_HEAD(&root->ordered_root);
 677	INIT_LIST_HEAD(&root->reloc_dirty_list);
 
 
 678	spin_lock_init(&root->inode_lock);
 679	spin_lock_init(&root->delalloc_lock);
 680	spin_lock_init(&root->ordered_extent_lock);
 681	spin_lock_init(&root->accounting_lock);
 
 
 682	spin_lock_init(&root->qgroup_meta_rsv_lock);
 683	mutex_init(&root->objectid_mutex);
 684	mutex_init(&root->log_mutex);
 685	mutex_init(&root->ordered_extent_mutex);
 686	mutex_init(&root->delalloc_mutex);
 687	init_waitqueue_head(&root->qgroup_flush_wait);
 688	init_waitqueue_head(&root->log_writer_wait);
 689	init_waitqueue_head(&root->log_commit_wait[0]);
 690	init_waitqueue_head(&root->log_commit_wait[1]);
 691	INIT_LIST_HEAD(&root->log_ctxs[0]);
 692	INIT_LIST_HEAD(&root->log_ctxs[1]);
 693	atomic_set(&root->log_commit[0], 0);
 694	atomic_set(&root->log_commit[1], 0);
 695	atomic_set(&root->log_writers, 0);
 696	atomic_set(&root->log_batch, 0);
 697	refcount_set(&root->refs, 1);
 
 698	atomic_set(&root->snapshot_force_cow, 0);
 699	atomic_set(&root->nr_swapfiles, 0);
 700	btrfs_set_root_log_transid(root, 0);
 701	root->log_transid_committed = -1;
 702	btrfs_set_root_last_log_commit(root, 0);
 
 
 
 
 
 
 
 
 
 
 
 
 703	root->anon_dev = 0;
 704	if (!dummy) {
 705		extent_io_tree_init(fs_info, &root->dirty_log_pages,
 706				    IO_TREE_ROOT_DIRTY_LOG_PAGES);
 707		extent_io_tree_init(fs_info, &root->log_csum_range,
 708				    IO_TREE_LOG_CSUM_RANGE);
 709	}
 710
 711	spin_lock_init(&root->root_item_lock);
 712	btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
 713#ifdef CONFIG_BTRFS_DEBUG
 714	INIT_LIST_HEAD(&root->leak_list);
 715	spin_lock(&fs_info->fs_roots_radix_lock);
 716	list_add_tail(&root->leak_list, &fs_info->allocated_roots);
 717	spin_unlock(&fs_info->fs_roots_radix_lock);
 718#endif
 719}
 720
 721static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
 722					   u64 objectid, gfp_t flags)
 723{
 724	struct btrfs_root *root = kzalloc(sizeof(*root), flags);
 725	if (root)
 726		__setup_root(root, fs_info, objectid);
 727	return root;
 728}
 729
 730#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 731/* Should only be used by the testing infrastructure */
 732struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
 733{
 734	struct btrfs_root *root;
 735
 736	if (!fs_info)
 737		return ERR_PTR(-EINVAL);
 738
 739	root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
 740	if (!root)
 741		return ERR_PTR(-ENOMEM);
 742
 743	/* We don't use the stripesize in selftest, set it as sectorsize */
 
 744	root->alloc_bytenr = 0;
 745
 746	return root;
 747}
 748#endif
 749
 750static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
 751{
 752	const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
 753	const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
 754
 755	return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
 756}
 757
 758static int global_root_key_cmp(const void *k, const struct rb_node *node)
 759{
 760	const struct btrfs_key *key = k;
 761	const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
 762
 763	return btrfs_comp_cpu_keys(key, &root->root_key);
 764}
 765
 766int btrfs_global_root_insert(struct btrfs_root *root)
 767{
 768	struct btrfs_fs_info *fs_info = root->fs_info;
 769	struct rb_node *tmp;
 770	int ret = 0;
 771
 772	write_lock(&fs_info->global_root_lock);
 773	tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
 774	write_unlock(&fs_info->global_root_lock);
 775
 776	if (tmp) {
 777		ret = -EEXIST;
 778		btrfs_warn(fs_info, "global root %llu %llu already exists",
 779				root->root_key.objectid, root->root_key.offset);
 780	}
 781	return ret;
 782}
 783
 784void btrfs_global_root_delete(struct btrfs_root *root)
 785{
 786	struct btrfs_fs_info *fs_info = root->fs_info;
 787
 788	write_lock(&fs_info->global_root_lock);
 789	rb_erase(&root->rb_node, &fs_info->global_root_tree);
 790	write_unlock(&fs_info->global_root_lock);
 791}
 792
 793struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
 794				     struct btrfs_key *key)
 795{
 796	struct rb_node *node;
 797	struct btrfs_root *root = NULL;
 798
 799	read_lock(&fs_info->global_root_lock);
 800	node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
 801	if (node)
 802		root = container_of(node, struct btrfs_root, rb_node);
 803	read_unlock(&fs_info->global_root_lock);
 804
 805	return root;
 806}
 807
 808static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
 809{
 810	struct btrfs_block_group *block_group;
 811	u64 ret;
 812
 813	if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
 814		return 0;
 815
 816	if (bytenr)
 817		block_group = btrfs_lookup_block_group(fs_info, bytenr);
 818	else
 819		block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
 820	ASSERT(block_group);
 821	if (!block_group)
 822		return 0;
 823	ret = block_group->global_root_id;
 824	btrfs_put_block_group(block_group);
 825
 826	return ret;
 827}
 828
 829struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
 830{
 831	struct btrfs_key key = {
 832		.objectid = BTRFS_CSUM_TREE_OBJECTID,
 833		.type = BTRFS_ROOT_ITEM_KEY,
 834		.offset = btrfs_global_root_id(fs_info, bytenr),
 835	};
 836
 837	return btrfs_global_root(fs_info, &key);
 838}
 839
 840struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
 841{
 842	struct btrfs_key key = {
 843		.objectid = BTRFS_EXTENT_TREE_OBJECTID,
 844		.type = BTRFS_ROOT_ITEM_KEY,
 845		.offset = btrfs_global_root_id(fs_info, bytenr),
 846	};
 847
 848	return btrfs_global_root(fs_info, &key);
 849}
 850
 851struct btrfs_root *btrfs_block_group_root(struct btrfs_fs_info *fs_info)
 852{
 853	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE))
 854		return fs_info->block_group_root;
 855	return btrfs_extent_root(fs_info, 0);
 856}
 857
 858struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
 859				     u64 objectid)
 860{
 861	struct btrfs_fs_info *fs_info = trans->fs_info;
 862	struct extent_buffer *leaf;
 863	struct btrfs_root *tree_root = fs_info->tree_root;
 864	struct btrfs_root *root;
 865	struct btrfs_key key;
 866	unsigned int nofs_flag;
 867	int ret = 0;
 
 868
 869	/*
 870	 * We're holding a transaction handle, so use a NOFS memory allocation
 871	 * context to avoid deadlock if reclaim happens.
 872	 */
 873	nofs_flag = memalloc_nofs_save();
 874	root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
 875	memalloc_nofs_restore(nofs_flag);
 876	if (!root)
 877		return ERR_PTR(-ENOMEM);
 878
 
 879	root->root_key.objectid = objectid;
 880	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
 881	root->root_key.offset = 0;
 882
 883	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
 884				      0, BTRFS_NESTING_NORMAL);
 885	if (IS_ERR(leaf)) {
 886		ret = PTR_ERR(leaf);
 887		leaf = NULL;
 888		goto fail;
 889	}
 890
 891	root->node = leaf;
 892	btrfs_mark_buffer_dirty(trans, leaf);
 893
 894	root->commit_root = btrfs_root_node(root);
 895	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
 896
 897	btrfs_set_root_flags(&root->root_item, 0);
 898	btrfs_set_root_limit(&root->root_item, 0);
 899	btrfs_set_root_bytenr(&root->root_item, leaf->start);
 900	btrfs_set_root_generation(&root->root_item, trans->transid);
 901	btrfs_set_root_level(&root->root_item, 0);
 902	btrfs_set_root_refs(&root->root_item, 1);
 903	btrfs_set_root_used(&root->root_item, leaf->len);
 904	btrfs_set_root_last_snapshot(&root->root_item, 0);
 905	btrfs_set_root_dirid(&root->root_item, 0);
 906	if (is_fstree(objectid))
 907		generate_random_guid(root->root_item.uuid);
 908	else
 909		export_guid(root->root_item.uuid, &guid_null);
 910	btrfs_set_root_drop_level(&root->root_item, 0);
 911
 912	btrfs_tree_unlock(leaf);
 913
 914	key.objectid = objectid;
 915	key.type = BTRFS_ROOT_ITEM_KEY;
 916	key.offset = 0;
 917	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
 918	if (ret)
 919		goto fail;
 920
 
 
 921	return root;
 922
 923fail:
 924	btrfs_put_root(root);
 
 
 
 
 
 925
 926	return ERR_PTR(ret);
 927}
 928
 929static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
 930					 struct btrfs_fs_info *fs_info)
 931{
 932	struct btrfs_root *root;
 
 933
 934	root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
 935	if (!root)
 936		return ERR_PTR(-ENOMEM);
 937
 
 
 938	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
 939	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
 940	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
 941
 942	return root;
 943}
 944
 945int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
 946			      struct btrfs_root *root)
 947{
 948	struct extent_buffer *leaf;
 949
 950	/*
 951	 * DON'T set SHAREABLE bit for log trees.
 952	 *
 953	 * Log trees are not exposed to user space thus can't be snapshotted,
 954	 * and they go away before a real commit is actually done.
 955	 *
 956	 * They do store pointers to file data extents, and those reference
 957	 * counts still get updated (along with back refs to the log tree).
 958	 */
 959
 960	leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
 961			NULL, 0, 0, 0, 0, BTRFS_NESTING_NORMAL);
 962	if (IS_ERR(leaf))
 963		return PTR_ERR(leaf);
 
 
 964
 965	root->node = leaf;
 966
 967	btrfs_mark_buffer_dirty(trans, root->node);
 968	btrfs_tree_unlock(root->node);
 969
 970	return 0;
 971}
 972
 973int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
 974			     struct btrfs_fs_info *fs_info)
 975{
 976	struct btrfs_root *log_root;
 977
 978	log_root = alloc_log_tree(trans, fs_info);
 979	if (IS_ERR(log_root))
 980		return PTR_ERR(log_root);
 981
 982	if (!btrfs_is_zoned(fs_info)) {
 983		int ret = btrfs_alloc_log_tree_node(trans, log_root);
 984
 985		if (ret) {
 986			btrfs_put_root(log_root);
 987			return ret;
 988		}
 989	}
 990
 991	WARN_ON(fs_info->log_root_tree);
 992	fs_info->log_root_tree = log_root;
 993	return 0;
 994}
 995
 996int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
 997		       struct btrfs_root *root)
 998{
 999	struct btrfs_fs_info *fs_info = root->fs_info;
1000	struct btrfs_root *log_root;
1001	struct btrfs_inode_item *inode_item;
1002	int ret;
1003
1004	log_root = alloc_log_tree(trans, fs_info);
1005	if (IS_ERR(log_root))
1006		return PTR_ERR(log_root);
1007
1008	ret = btrfs_alloc_log_tree_node(trans, log_root);
1009	if (ret) {
1010		btrfs_put_root(log_root);
1011		return ret;
1012	}
1013
1014	log_root->last_trans = trans->transid;
1015	log_root->root_key.offset = root->root_key.objectid;
1016
1017	inode_item = &log_root->root_item.inode;
1018	btrfs_set_stack_inode_generation(inode_item, 1);
1019	btrfs_set_stack_inode_size(inode_item, 3);
1020	btrfs_set_stack_inode_nlink(inode_item, 1);
1021	btrfs_set_stack_inode_nbytes(inode_item,
1022				     fs_info->nodesize);
1023	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1024
1025	btrfs_set_root_node(&log_root->root_item, log_root->node);
1026
1027	WARN_ON(root->log_root);
1028	root->log_root = log_root;
1029	btrfs_set_root_log_transid(root, 0);
1030	root->log_transid_committed = -1;
1031	btrfs_set_root_last_log_commit(root, 0);
1032	return 0;
1033}
1034
1035static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1036					      struct btrfs_path *path,
1037					      struct btrfs_key *key)
1038{
1039	struct btrfs_root *root;
1040	struct btrfs_tree_parent_check check = { 0 };
1041	struct btrfs_fs_info *fs_info = tree_root->fs_info;
 
1042	u64 generation;
1043	int ret;
1044	int level;
1045
1046	root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1047	if (!root)
1048		return ERR_PTR(-ENOMEM);
1049
 
 
 
 
 
 
 
 
1050	ret = btrfs_find_root(tree_root, key, path,
1051			      &root->root_item, &root->root_key);
1052	if (ret) {
1053		if (ret > 0)
1054			ret = -ENOENT;
1055		goto fail;
1056	}
1057
1058	generation = btrfs_root_generation(&root->root_item);
1059	level = btrfs_root_level(&root->root_item);
1060	check.level = level;
1061	check.transid = generation;
1062	check.owner_root = key->objectid;
1063	root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item),
1064				     &check);
1065	if (IS_ERR(root->node)) {
1066		ret = PTR_ERR(root->node);
1067		root->node = NULL;
1068		goto fail;
1069	}
1070	if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1071		ret = -EIO;
1072		goto fail;
1073	}
1074
1075	/*
1076	 * For real fs, and not log/reloc trees, root owner must
1077	 * match its root node owner
1078	 */
1079	if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) &&
1080	    root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1081	    root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1082	    root->root_key.objectid != btrfs_header_owner(root->node)) {
1083		btrfs_crit(fs_info,
1084"root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1085			   root->root_key.objectid, root->node->start,
1086			   btrfs_header_owner(root->node),
1087			   root->root_key.objectid);
1088		ret = -EUCLEAN;
1089		goto fail;
1090	}
1091	root->commit_root = btrfs_root_node(root);
 
 
1092	return root;
1093fail:
1094	btrfs_put_root(root);
1095	return ERR_PTR(ret);
 
 
 
1096}
1097
1098struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1099					struct btrfs_key *key)
1100{
1101	struct btrfs_root *root;
1102	struct btrfs_path *path;
1103
1104	path = btrfs_alloc_path();
1105	if (!path)
1106		return ERR_PTR(-ENOMEM);
1107	root = read_tree_root_path(tree_root, path, key);
1108	btrfs_free_path(path);
 
 
 
1109
1110	return root;
1111}
1112
1113/*
1114 * Initialize subvolume root in-memory structure
1115 *
1116 * @anon_dev:	anonymous device to attach to the root, if zero, allocate new
1117 */
1118static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1119{
1120	int ret;
 
1121
1122	btrfs_drew_lock_init(&root->snapshot_lock);
 
 
 
 
 
 
1123
1124	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1125	    !btrfs_is_data_reloc_root(root) &&
1126	    is_fstree(root->root_key.objectid)) {
1127		set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1128		btrfs_check_and_init_root_item(&root->root_item);
1129	}
 
 
 
 
 
1130
1131	/*
1132	 * Don't assign anonymous block device to roots that are not exposed to
1133	 * userspace, the id pool is limited to 1M
1134	 */
1135	if (is_fstree(root->root_key.objectid) &&
1136	    btrfs_root_refs(&root->root_item) > 0) {
1137		if (!anon_dev) {
1138			ret = get_anon_bdev(&root->anon_dev);
1139			if (ret)
1140				goto fail;
1141		} else {
1142			root->anon_dev = anon_dev;
1143		}
1144	}
1145
1146	mutex_lock(&root->objectid_mutex);
1147	ret = btrfs_init_root_free_objectid(root);
 
1148	if (ret) {
1149		mutex_unlock(&root->objectid_mutex);
1150		goto fail;
1151	}
1152
1153	ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1154
1155	mutex_unlock(&root->objectid_mutex);
1156
1157	return 0;
1158fail:
1159	/* The caller is responsible to call btrfs_free_fs_root */
1160	return ret;
1161}
1162
1163static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1164					       u64 root_id)
1165{
1166	struct btrfs_root *root;
1167
1168	spin_lock(&fs_info->fs_roots_radix_lock);
1169	root = radix_tree_lookup(&fs_info->fs_roots_radix,
1170				 (unsigned long)root_id);
1171	root = btrfs_grab_root(root);
1172	spin_unlock(&fs_info->fs_roots_radix_lock);
1173	return root;
1174}
1175
1176static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1177						u64 objectid)
1178{
1179	struct btrfs_key key = {
1180		.objectid = objectid,
1181		.type = BTRFS_ROOT_ITEM_KEY,
1182		.offset = 0,
1183	};
1184
1185	switch (objectid) {
1186	case BTRFS_ROOT_TREE_OBJECTID:
1187		return btrfs_grab_root(fs_info->tree_root);
1188	case BTRFS_EXTENT_TREE_OBJECTID:
1189		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1190	case BTRFS_CHUNK_TREE_OBJECTID:
1191		return btrfs_grab_root(fs_info->chunk_root);
1192	case BTRFS_DEV_TREE_OBJECTID:
1193		return btrfs_grab_root(fs_info->dev_root);
1194	case BTRFS_CSUM_TREE_OBJECTID:
1195		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1196	case BTRFS_QUOTA_TREE_OBJECTID:
1197		return btrfs_grab_root(fs_info->quota_root);
1198	case BTRFS_UUID_TREE_OBJECTID:
1199		return btrfs_grab_root(fs_info->uuid_root);
1200	case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
1201		return btrfs_grab_root(fs_info->block_group_root);
1202	case BTRFS_FREE_SPACE_TREE_OBJECTID:
1203		return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1204	case BTRFS_RAID_STRIPE_TREE_OBJECTID:
1205		return btrfs_grab_root(fs_info->stripe_root);
1206	default:
1207		return NULL;
1208	}
1209}
1210
1211int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1212			 struct btrfs_root *root)
1213{
1214	int ret;
1215
1216	ret = radix_tree_preload(GFP_NOFS);
1217	if (ret)
1218		return ret;
1219
1220	spin_lock(&fs_info->fs_roots_radix_lock);
1221	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1222				(unsigned long)root->root_key.objectid,
1223				root);
1224	if (ret == 0) {
1225		btrfs_grab_root(root);
1226		set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1227	}
1228	spin_unlock(&fs_info->fs_roots_radix_lock);
1229	radix_tree_preload_end();
1230
1231	return ret;
1232}
1233
1234void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1235{
1236#ifdef CONFIG_BTRFS_DEBUG
1237	struct btrfs_root *root;
1238
1239	while (!list_empty(&fs_info->allocated_roots)) {
1240		char buf[BTRFS_ROOT_NAME_BUF_LEN];
1241
1242		root = list_first_entry(&fs_info->allocated_roots,
1243					struct btrfs_root, leak_list);
1244		btrfs_err(fs_info, "leaked root %s refcount %d",
1245			  btrfs_root_name(&root->root_key, buf),
1246			  refcount_read(&root->refs));
1247		while (refcount_read(&root->refs) > 1)
1248			btrfs_put_root(root);
1249		btrfs_put_root(root);
1250	}
1251#endif
1252}
1253
1254static void free_global_roots(struct btrfs_fs_info *fs_info)
1255{
1256	struct btrfs_root *root;
1257	struct rb_node *node;
1258
1259	while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1260		root = rb_entry(node, struct btrfs_root, rb_node);
1261		rb_erase(&root->rb_node, &fs_info->global_root_tree);
1262		btrfs_put_root(root);
1263	}
1264}
1265
1266void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1267{
1268	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1269	percpu_counter_destroy(&fs_info->delalloc_bytes);
1270	percpu_counter_destroy(&fs_info->ordered_bytes);
1271	percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1272	btrfs_free_csum_hash(fs_info);
1273	btrfs_free_stripe_hash_table(fs_info);
1274	btrfs_free_ref_cache(fs_info);
1275	kfree(fs_info->balance_ctl);
1276	kfree(fs_info->delayed_root);
1277	free_global_roots(fs_info);
1278	btrfs_put_root(fs_info->tree_root);
1279	btrfs_put_root(fs_info->chunk_root);
1280	btrfs_put_root(fs_info->dev_root);
1281	btrfs_put_root(fs_info->quota_root);
1282	btrfs_put_root(fs_info->uuid_root);
1283	btrfs_put_root(fs_info->fs_root);
1284	btrfs_put_root(fs_info->data_reloc_root);
1285	btrfs_put_root(fs_info->block_group_root);
1286	btrfs_put_root(fs_info->stripe_root);
1287	btrfs_check_leaked_roots(fs_info);
1288	btrfs_extent_buffer_leak_debug_check(fs_info);
1289	kfree(fs_info->super_copy);
1290	kfree(fs_info->super_for_commit);
1291	kfree(fs_info->subpage_info);
1292	kvfree(fs_info);
1293}
1294
1295
1296/*
1297 * Get an in-memory reference of a root structure.
1298 *
1299 * For essential trees like root/extent tree, we grab it from fs_info directly.
1300 * For subvolume trees, we check the cached filesystem roots first. If not
1301 * found, then read it from disk and add it to cached fs roots.
1302 *
1303 * Caller should release the root by calling btrfs_put_root() after the usage.
1304 *
1305 * NOTE: Reloc and log trees can't be read by this function as they share the
1306 *	 same root objectid.
1307 *
1308 * @objectid:	root id
1309 * @anon_dev:	preallocated anonymous block device number for new roots,
1310 *		pass NULL for a new allocation.
1311 * @check_ref:	whether to check root item references, If true, return -ENOENT
1312 *		for orphan roots
1313 */
1314static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1315					     u64 objectid, dev_t *anon_dev,
1316					     bool check_ref)
1317{
1318	struct btrfs_root *root;
1319	struct btrfs_path *path;
1320	struct btrfs_key key;
1321	int ret;
1322
1323	root = btrfs_get_global_root(fs_info, objectid);
1324	if (root)
1325		return root;
1326
1327	/*
1328	 * If we're called for non-subvolume trees, and above function didn't
1329	 * find one, do not try to read it from disk.
1330	 *
1331	 * This is namely for free-space-tree and quota tree, which can change
1332	 * at runtime and should only be grabbed from fs_info.
1333	 */
1334	if (!is_fstree(objectid) && objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1335		return ERR_PTR(-ENOENT);
 
 
 
 
 
 
1336again:
1337	root = btrfs_lookup_fs_root(fs_info, objectid);
1338	if (root) {
1339		/*
1340		 * Some other caller may have read out the newly inserted
1341		 * subvolume already (for things like backref walk etc).  Not
1342		 * that common but still possible.  In that case, we just need
1343		 * to free the anon_dev.
1344		 */
1345		if (unlikely(anon_dev && *anon_dev)) {
1346			free_anon_bdev(*anon_dev);
1347			*anon_dev = 0;
1348		}
1349
1350		if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1351			btrfs_put_root(root);
1352			return ERR_PTR(-ENOENT);
1353		}
1354		return root;
1355	}
1356
1357	key.objectid = objectid;
1358	key.type = BTRFS_ROOT_ITEM_KEY;
1359	key.offset = (u64)-1;
1360	root = btrfs_read_tree_root(fs_info->tree_root, &key);
1361	if (IS_ERR(root))
1362		return root;
1363
1364	if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1365		ret = -ENOENT;
1366		goto fail;
1367	}
1368
1369	ret = btrfs_init_fs_root(root, anon_dev ? *anon_dev : 0);
1370	if (ret)
1371		goto fail;
1372
1373	path = btrfs_alloc_path();
1374	if (!path) {
1375		ret = -ENOMEM;
1376		goto fail;
1377	}
1378	key.objectid = BTRFS_ORPHAN_OBJECTID;
1379	key.type = BTRFS_ORPHAN_ITEM_KEY;
1380	key.offset = objectid;
1381
1382	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1383	btrfs_free_path(path);
1384	if (ret < 0)
1385		goto fail;
1386	if (ret == 0)
1387		set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1388
1389	ret = btrfs_insert_fs_root(fs_info, root);
1390	if (ret) {
1391		if (ret == -EEXIST) {
1392			btrfs_put_root(root);
1393			goto again;
1394		}
1395		goto fail;
1396	}
1397	return root;
1398fail:
1399	/*
1400	 * If our caller provided us an anonymous device, then it's his
1401	 * responsibility to free it in case we fail. So we have to set our
1402	 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1403	 * and once again by our caller.
1404	 */
1405	if (anon_dev && *anon_dev)
1406		root->anon_dev = 0;
1407	btrfs_put_root(root);
1408	return ERR_PTR(ret);
1409}
1410
1411/*
1412 * Get in-memory reference of a root structure
1413 *
1414 * @objectid:	tree objectid
1415 * @check_ref:	if set, verify that the tree exists and the item has at least
1416 *		one reference
1417 */
1418struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1419				     u64 objectid, bool check_ref)
1420{
1421	return btrfs_get_root_ref(fs_info, objectid, NULL, check_ref);
1422}
 
 
1423
1424/*
1425 * Get in-memory reference of a root structure, created as new, optionally pass
1426 * the anonymous block device id
1427 *
1428 * @objectid:	tree objectid
1429 * @anon_dev:	if NULL, allocate a new anonymous block device or use the
1430 *		parameter value if not NULL
1431 */
1432struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1433					 u64 objectid, dev_t *anon_dev)
1434{
1435	return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1436}
1437
1438/*
1439 * Return a root for the given objectid.
1440 *
1441 * @fs_info:	the fs_info
1442 * @objectid:	the objectid we need to lookup
1443 *
1444 * This is exclusively used for backref walking, and exists specifically because
1445 * of how qgroups does lookups.  Qgroups will do a backref lookup at delayed ref
1446 * creation time, which means we may have to read the tree_root in order to look
1447 * up a fs root that is not in memory.  If the root is not in memory we will
1448 * read the tree root commit root and look up the fs root from there.  This is a
1449 * temporary root, it will not be inserted into the radix tree as it doesn't
1450 * have the most uptodate information, it'll simply be discarded once the
1451 * backref code is finished using the root.
1452 */
1453struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1454						 struct btrfs_path *path,
1455						 u64 objectid)
1456{
1457	struct btrfs_root *root;
1458	struct btrfs_key key;
1459
1460	ASSERT(path->search_commit_root && path->skip_locking);
1461
1462	/*
1463	 * This can return -ENOENT if we ask for a root that doesn't exist, but
1464	 * since this is called via the backref walking code we won't be looking
1465	 * up a root that doesn't exist, unless there's corruption.  So if root
1466	 * != NULL just return it.
1467	 */
1468	root = btrfs_get_global_root(fs_info, objectid);
1469	if (root)
1470		return root;
1471
1472	root = btrfs_lookup_fs_root(fs_info, objectid);
1473	if (root)
1474		return root;
1475
1476	key.objectid = objectid;
1477	key.type = BTRFS_ROOT_ITEM_KEY;
1478	key.offset = (u64)-1;
1479	root = read_tree_root_path(fs_info->tree_root, path, &key);
1480	btrfs_release_path(path);
1481
1482	return root;
 
 
 
 
1483}
1484
1485static int cleaner_kthread(void *arg)
1486{
1487	struct btrfs_fs_info *fs_info = arg;
 
1488	int again;
1489
1490	while (1) {
1491		again = 0;
1492
1493		set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1494
1495		/* Make the cleaner go to sleep early. */
1496		if (btrfs_need_cleaner_sleep(fs_info))
1497			goto sleep;
1498
1499		/*
1500		 * Do not do anything if we might cause open_ctree() to block
1501		 * before we have finished mounting the filesystem.
1502		 */
1503		if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1504			goto sleep;
1505
1506		if (!mutex_trylock(&fs_info->cleaner_mutex))
1507			goto sleep;
1508
1509		/*
1510		 * Avoid the problem that we change the status of the fs
1511		 * during the above check and trylock.
1512		 */
1513		if (btrfs_need_cleaner_sleep(fs_info)) {
1514			mutex_unlock(&fs_info->cleaner_mutex);
1515			goto sleep;
1516		}
1517
1518		if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags))
1519			btrfs_sysfs_feature_update(fs_info);
1520
1521		btrfs_run_delayed_iputs(fs_info);
1522
1523		again = btrfs_clean_one_deleted_snapshot(fs_info);
1524		mutex_unlock(&fs_info->cleaner_mutex);
1525
1526		/*
1527		 * The defragger has dealt with the R/O remount and umount,
1528		 * needn't do anything special here.
1529		 */
1530		btrfs_run_defrag_inodes(fs_info);
1531
1532		/*
1533		 * Acquires fs_info->reclaim_bgs_lock to avoid racing
1534		 * with relocation (btrfs_relocate_chunk) and relocation
1535		 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1536		 * after acquiring fs_info->reclaim_bgs_lock. So we
1537		 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1538		 * unused block groups.
1539		 */
1540		btrfs_delete_unused_bgs(fs_info);
1541
1542		/*
1543		 * Reclaim block groups in the reclaim_bgs list after we deleted
1544		 * all unused block_groups. This possibly gives us some more free
1545		 * space.
1546		 */
1547		btrfs_reclaim_bgs(fs_info);
1548sleep:
1549		clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1550		if (kthread_should_park())
1551			kthread_parkme();
1552		if (kthread_should_stop())
1553			return 0;
1554		if (!again) {
1555			set_current_state(TASK_INTERRUPTIBLE);
1556			schedule();
1557			__set_current_state(TASK_RUNNING);
1558		}
1559	}
1560}
1561
1562static int transaction_kthread(void *arg)
1563{
1564	struct btrfs_root *root = arg;
1565	struct btrfs_fs_info *fs_info = root->fs_info;
1566	struct btrfs_trans_handle *trans;
1567	struct btrfs_transaction *cur;
1568	u64 transid;
1569	time64_t delta;
1570	unsigned long delay;
1571	bool cannot_commit;
1572
1573	do {
1574		cannot_commit = false;
1575		delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1576		mutex_lock(&fs_info->transaction_kthread_mutex);
1577
1578		spin_lock(&fs_info->trans_lock);
1579		cur = fs_info->running_transaction;
1580		if (!cur) {
1581			spin_unlock(&fs_info->trans_lock);
1582			goto sleep;
1583		}
1584
1585		delta = ktime_get_seconds() - cur->start_time;
1586		if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1587		    cur->state < TRANS_STATE_COMMIT_PREP &&
1588		    delta < fs_info->commit_interval) {
 
1589			spin_unlock(&fs_info->trans_lock);
1590			delay -= msecs_to_jiffies((delta - 1) * 1000);
1591			delay = min(delay,
1592				    msecs_to_jiffies(fs_info->commit_interval * 1000));
1593			goto sleep;
1594		}
1595		transid = cur->transid;
1596		spin_unlock(&fs_info->trans_lock);
1597
1598		/* If the file system is aborted, this will always fail. */
1599		trans = btrfs_attach_transaction(root);
1600		if (IS_ERR(trans)) {
1601			if (PTR_ERR(trans) != -ENOENT)
1602				cannot_commit = true;
1603			goto sleep;
1604		}
1605		if (transid == trans->transid) {
1606			btrfs_commit_transaction(trans);
1607		} else {
1608			btrfs_end_transaction(trans);
1609		}
1610sleep:
1611		wake_up_process(fs_info->cleaner_kthread);
1612		mutex_unlock(&fs_info->transaction_kthread_mutex);
1613
1614		if (BTRFS_FS_ERROR(fs_info))
 
1615			btrfs_cleanup_transaction(fs_info);
1616		if (!kthread_should_stop() &&
1617				(!btrfs_transaction_blocked(fs_info) ||
1618				 cannot_commit))
1619			schedule_timeout_interruptible(delay);
1620	} while (!kthread_should_stop());
1621	return 0;
1622}
1623
1624/*
1625 * This will find the highest generation in the array of root backups.  The
1626 * index of the highest array is returned, or -EINVAL if we can't find
1627 * anything.
1628 *
1629 * We check to make sure the array is valid by comparing the
1630 * generation of the latest  root in the array with the generation
1631 * in the super block.  If they don't match we pitch it.
1632 */
1633static int find_newest_super_backup(struct btrfs_fs_info *info)
1634{
1635	const u64 newest_gen = btrfs_super_generation(info->super_copy);
1636	u64 cur;
 
1637	struct btrfs_root_backup *root_backup;
1638	int i;
1639
1640	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1641		root_backup = info->super_copy->super_roots + i;
1642		cur = btrfs_backup_tree_root_gen(root_backup);
1643		if (cur == newest_gen)
1644			return i;
 
 
 
 
 
 
 
 
1645	}
 
 
1646
1647	return -EINVAL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1648}
1649
1650/*
1651 * copy all the root pointers into the super backup array.
1652 * this will bump the backup pointer by one when it is
1653 * done
1654 */
1655static void backup_super_roots(struct btrfs_fs_info *info)
1656{
1657	const int next_backup = info->backup_root_index;
1658	struct btrfs_root_backup *root_backup;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1659
1660	root_backup = info->super_for_commit->super_roots + next_backup;
1661
1662	/*
1663	 * make sure all of our padding and empty slots get zero filled
1664	 * regardless of which ones we use today
1665	 */
1666	memset(root_backup, 0, sizeof(*root_backup));
1667
1668	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1669
1670	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1671	btrfs_set_backup_tree_root_gen(root_backup,
1672			       btrfs_header_generation(info->tree_root->node));
1673
1674	btrfs_set_backup_tree_root_level(root_backup,
1675			       btrfs_header_level(info->tree_root->node));
1676
1677	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1678	btrfs_set_backup_chunk_root_gen(root_backup,
1679			       btrfs_header_generation(info->chunk_root->node));
1680	btrfs_set_backup_chunk_root_level(root_backup,
1681			       btrfs_header_level(info->chunk_root->node));
1682
1683	if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
1684		struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
1685		struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
1686
1687		btrfs_set_backup_extent_root(root_backup,
1688					     extent_root->node->start);
1689		btrfs_set_backup_extent_root_gen(root_backup,
1690				btrfs_header_generation(extent_root->node));
1691		btrfs_set_backup_extent_root_level(root_backup,
1692					btrfs_header_level(extent_root->node));
1693
1694		btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
1695		btrfs_set_backup_csum_root_gen(root_backup,
1696					       btrfs_header_generation(csum_root->node));
1697		btrfs_set_backup_csum_root_level(root_backup,
1698						 btrfs_header_level(csum_root->node));
1699	}
1700
1701	/*
1702	 * we might commit during log recovery, which happens before we set
1703	 * the fs_root.  Make sure it is valid before we fill it in.
1704	 */
1705	if (info->fs_root && info->fs_root->node) {
1706		btrfs_set_backup_fs_root(root_backup,
1707					 info->fs_root->node->start);
1708		btrfs_set_backup_fs_root_gen(root_backup,
1709			       btrfs_header_generation(info->fs_root->node));
1710		btrfs_set_backup_fs_root_level(root_backup,
1711			       btrfs_header_level(info->fs_root->node));
1712	}
1713
1714	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1715	btrfs_set_backup_dev_root_gen(root_backup,
1716			       btrfs_header_generation(info->dev_root->node));
1717	btrfs_set_backup_dev_root_level(root_backup,
1718				       btrfs_header_level(info->dev_root->node));
1719
 
 
 
 
 
 
1720	btrfs_set_backup_total_bytes(root_backup,
1721			     btrfs_super_total_bytes(info->super_copy));
1722	btrfs_set_backup_bytes_used(root_backup,
1723			     btrfs_super_bytes_used(info->super_copy));
1724	btrfs_set_backup_num_devices(root_backup,
1725			     btrfs_super_num_devices(info->super_copy));
1726
1727	/*
1728	 * if we don't copy this out to the super_copy, it won't get remembered
1729	 * for the next commit
1730	 */
1731	memcpy(&info->super_copy->super_roots,
1732	       &info->super_for_commit->super_roots,
1733	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1734}
1735
1736/*
1737 * Reads a backup root based on the passed priority. Prio 0 is the newest, prio
1738 * 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
 
 
1739 *
1740 * @fs_info:  filesystem whose backup roots need to be read
1741 * @priority: priority of backup root required
1742 *
1743 * Returns backup root index on success and -EINVAL otherwise.
1744 */
1745static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
 
 
1746{
1747	int backup_index = find_newest_super_backup(fs_info);
1748	struct btrfs_super_block *super = fs_info->super_copy;
1749	struct btrfs_root_backup *root_backup;
 
1750
1751	if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1752		if (priority == 0)
1753			return backup_index;
1754
1755		backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1756		backup_index %= BTRFS_NUM_BACKUP_ROOTS;
 
 
 
 
 
 
 
1757	} else {
1758		return -EINVAL;
 
 
 
 
1759	}
1760
1761	root_backup = super->super_roots + backup_index;
1762
1763	btrfs_set_super_generation(super,
1764				   btrfs_backup_tree_root_gen(root_backup));
1765	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1766	btrfs_set_super_root_level(super,
1767				   btrfs_backup_tree_root_level(root_backup));
1768	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1769
1770	/*
1771	 * Fixme: the total bytes and num_devices need to match or we should
1772	 * need a fsck
1773	 */
1774	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1775	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1776
1777	return backup_index;
1778}
1779
1780/* helper to cleanup workers */
1781static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1782{
1783	btrfs_destroy_workqueue(fs_info->fixup_workers);
1784	btrfs_destroy_workqueue(fs_info->delalloc_workers);
1785	btrfs_destroy_workqueue(fs_info->workers);
1786	if (fs_info->endio_workers)
1787		destroy_workqueue(fs_info->endio_workers);
1788	if (fs_info->rmw_workers)
1789		destroy_workqueue(fs_info->rmw_workers);
1790	if (fs_info->compressed_write_workers)
1791		destroy_workqueue(fs_info->compressed_write_workers);
1792	btrfs_destroy_workqueue(fs_info->endio_write_workers);
1793	btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
 
1794	btrfs_destroy_workqueue(fs_info->delayed_workers);
1795	btrfs_destroy_workqueue(fs_info->caching_workers);
 
1796	btrfs_destroy_workqueue(fs_info->flush_workers);
1797	btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1798	if (fs_info->discard_ctl.discard_workers)
1799		destroy_workqueue(fs_info->discard_ctl.discard_workers);
1800	/*
1801	 * Now that all other work queues are destroyed, we can safely destroy
1802	 * the queues used for metadata I/O, since tasks from those other work
1803	 * queues can do metadata I/O operations.
1804	 */
1805	if (fs_info->endio_meta_workers)
1806		destroy_workqueue(fs_info->endio_meta_workers);
1807}
1808
1809static void free_root_extent_buffers(struct btrfs_root *root)
1810{
1811	if (root) {
1812		free_extent_buffer(root->node);
1813		free_extent_buffer(root->commit_root);
1814		root->node = NULL;
1815		root->commit_root = NULL;
1816	}
1817}
1818
1819static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
1820{
1821	struct btrfs_root *root, *tmp;
1822
1823	rbtree_postorder_for_each_entry_safe(root, tmp,
1824					     &fs_info->global_root_tree,
1825					     rb_node)
1826		free_root_extent_buffers(root);
1827}
1828
1829/* helper to cleanup tree roots */
1830static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1831{
1832	free_root_extent_buffers(info->tree_root);
1833
1834	free_global_root_pointers(info);
1835	free_root_extent_buffers(info->dev_root);
 
 
1836	free_root_extent_buffers(info->quota_root);
1837	free_root_extent_buffers(info->uuid_root);
1838	free_root_extent_buffers(info->fs_root);
1839	free_root_extent_buffers(info->data_reloc_root);
1840	free_root_extent_buffers(info->block_group_root);
1841	free_root_extent_buffers(info->stripe_root);
1842	if (free_chunk_root)
1843		free_root_extent_buffers(info->chunk_root);
1844}
1845
1846void btrfs_put_root(struct btrfs_root *root)
1847{
1848	if (!root)
1849		return;
1850
1851	if (refcount_dec_and_test(&root->refs)) {
1852		WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1853		WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
1854		if (root->anon_dev)
1855			free_anon_bdev(root->anon_dev);
1856		free_root_extent_buffers(root);
1857#ifdef CONFIG_BTRFS_DEBUG
1858		spin_lock(&root->fs_info->fs_roots_radix_lock);
1859		list_del_init(&root->leak_list);
1860		spin_unlock(&root->fs_info->fs_roots_radix_lock);
1861#endif
1862		kfree(root);
1863	}
1864}
1865
1866void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
1867{
1868	int ret;
1869	struct btrfs_root *gang[8];
1870	int i;
1871
1872	while (!list_empty(&fs_info->dead_roots)) {
1873		gang[0] = list_entry(fs_info->dead_roots.next,
1874				     struct btrfs_root, root_list);
1875		list_del(&gang[0]->root_list);
1876
1877		if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
1878			btrfs_drop_and_free_fs_root(fs_info, gang[0]);
1879		btrfs_put_root(gang[0]);
 
 
 
 
1880	}
1881
1882	while (1) {
1883		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1884					     (void **)gang, 0,
1885					     ARRAY_SIZE(gang));
1886		if (!ret)
1887			break;
1888		for (i = 0; i < ret; i++)
1889			btrfs_drop_and_free_fs_root(fs_info, gang[i]);
1890	}
 
 
 
 
 
1891}
1892
1893static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
1894{
1895	mutex_init(&fs_info->scrub_lock);
1896	atomic_set(&fs_info->scrubs_running, 0);
1897	atomic_set(&fs_info->scrub_pause_req, 0);
1898	atomic_set(&fs_info->scrubs_paused, 0);
1899	atomic_set(&fs_info->scrub_cancel_req, 0);
1900	init_waitqueue_head(&fs_info->scrub_pause_wait);
1901	refcount_set(&fs_info->scrub_workers_refcnt, 0);
1902}
1903
1904static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
1905{
1906	spin_lock_init(&fs_info->balance_lock);
1907	mutex_init(&fs_info->balance_mutex);
1908	atomic_set(&fs_info->balance_pause_req, 0);
1909	atomic_set(&fs_info->balance_cancel_req, 0);
1910	fs_info->balance_ctl = NULL;
1911	init_waitqueue_head(&fs_info->balance_wait_q);
1912	atomic_set(&fs_info->reloc_cancel_req, 0);
1913}
1914
1915static int btrfs_init_btree_inode(struct super_block *sb)
1916{
1917	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1918	unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
1919					      fs_info->tree_root);
1920	struct inode *inode;
1921
1922	inode = new_inode(sb);
1923	if (!inode)
1924		return -ENOMEM;
1925
1926	inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1927	set_nlink(inode, 1);
1928	/*
1929	 * we set the i_size on the btree inode to the max possible int.
1930	 * the real end of the address space is determined by all of
1931	 * the devices in the system
1932	 */
1933	inode->i_size = OFFSET_MAX;
1934	inode->i_mapping->a_ops = &btree_aops;
1935	mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
1936
1937	RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
1938	extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
1939			    IO_TREE_BTREE_INODE_IO);
 
1940	extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
1941
1942	BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
1943	BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID;
1944	BTRFS_I(inode)->location.type = 0;
1945	BTRFS_I(inode)->location.offset = 0;
1946	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
1947	__insert_inode_hash(inode, hash);
1948	fs_info->btree_inode = inode;
1949
1950	return 0;
1951}
1952
1953static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
1954{
1955	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
1956	init_rwsem(&fs_info->dev_replace.rwsem);
1957	init_waitqueue_head(&fs_info->dev_replace.replace_wait);
1958}
1959
1960static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
1961{
1962	spin_lock_init(&fs_info->qgroup_lock);
1963	mutex_init(&fs_info->qgroup_ioctl_lock);
1964	fs_info->qgroup_tree = RB_ROOT;
1965	INIT_LIST_HEAD(&fs_info->dirty_qgroups);
1966	fs_info->qgroup_seq = 1;
1967	fs_info->qgroup_ulist = NULL;
1968	fs_info->qgroup_rescan_running = false;
1969	fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
1970	mutex_init(&fs_info->qgroup_rescan_lock);
1971}
1972
1973static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
 
1974{
1975	u32 max_active = fs_info->thread_pool_size;
1976	unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
1977	unsigned int ordered_flags = WQ_MEM_RECLAIM | WQ_FREEZABLE;
1978
1979	fs_info->workers =
1980		btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
 
1981
1982	fs_info->delalloc_workers =
1983		btrfs_alloc_workqueue(fs_info, "delalloc",
1984				      flags, max_active, 2);
1985
1986	fs_info->flush_workers =
1987		btrfs_alloc_workqueue(fs_info, "flush_delalloc",
1988				      flags, max_active, 0);
1989
1990	fs_info->caching_workers =
1991		btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
1992
 
 
 
 
 
 
 
 
 
 
1993	fs_info->fixup_workers =
1994		btrfs_alloc_ordered_workqueue(fs_info, "fixup", ordered_flags);
1995
 
 
 
 
1996	fs_info->endio_workers =
1997		alloc_workqueue("btrfs-endio", flags, max_active);
1998	fs_info->endio_meta_workers =
1999		alloc_workqueue("btrfs-endio-meta", flags, max_active);
2000	fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
 
 
 
 
 
 
 
 
 
 
2001	fs_info->endio_write_workers =
2002		btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2003				      max_active, 2);
2004	fs_info->compressed_write_workers =
2005		alloc_workqueue("btrfs-compressed-write", flags, max_active);
2006	fs_info->endio_freespace_worker =
2007		btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2008				      max_active, 0);
2009	fs_info->delayed_workers =
2010		btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2011				      max_active, 0);
 
 
 
2012	fs_info->qgroup_rescan_workers =
2013		btrfs_alloc_ordered_workqueue(fs_info, "qgroup-rescan",
2014					      ordered_flags);
2015	fs_info->discard_ctl.discard_workers =
2016		alloc_ordered_workqueue("btrfs_discard", WQ_FREEZABLE);
2017
2018	if (!(fs_info->workers &&
2019	      fs_info->delalloc_workers && fs_info->flush_workers &&
2020	      fs_info->endio_workers && fs_info->endio_meta_workers &&
2021	      fs_info->compressed_write_workers &&
2022	      fs_info->endio_write_workers &&
 
2023	      fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2024	      fs_info->caching_workers && fs_info->fixup_workers &&
2025	      fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2026	      fs_info->discard_ctl.discard_workers)) {
2027		return -ENOMEM;
2028	}
2029
2030	return 0;
2031}
2032
2033static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2034{
2035	struct crypto_shash *csum_shash;
2036	const char *csum_driver = btrfs_super_csum_driver(csum_type);
2037
2038	csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2039
2040	if (IS_ERR(csum_shash)) {
2041		btrfs_err(fs_info, "error allocating %s hash for checksum",
2042			  csum_driver);
2043		return PTR_ERR(csum_shash);
2044	}
2045
2046	fs_info->csum_shash = csum_shash;
2047
2048	/*
2049	 * Check if the checksum implementation is a fast accelerated one.
2050	 * As-is this is a bit of a hack and should be replaced once the csum
2051	 * implementations provide that information themselves.
2052	 */
2053	switch (csum_type) {
2054	case BTRFS_CSUM_TYPE_CRC32:
2055		if (!strstr(crypto_shash_driver_name(csum_shash), "generic"))
2056			set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2057		break;
2058	case BTRFS_CSUM_TYPE_XXHASH:
2059		set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2060		break;
2061	default:
2062		break;
2063	}
2064
2065	btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2066			btrfs_super_csum_name(csum_type),
2067			crypto_shash_driver_name(csum_shash));
2068	return 0;
2069}
2070
2071static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2072			    struct btrfs_fs_devices *fs_devices)
2073{
2074	int ret;
2075	struct btrfs_tree_parent_check check = { 0 };
2076	struct btrfs_root *log_tree_root;
2077	struct btrfs_super_block *disk_super = fs_info->super_copy;
2078	u64 bytenr = btrfs_super_log_root(disk_super);
2079	int level = btrfs_super_log_root_level(disk_super);
2080
2081	if (fs_devices->rw_devices == 0) {
2082		btrfs_warn(fs_info, "log replay required on RO media");
2083		return -EIO;
2084	}
2085
2086	log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2087					 GFP_KERNEL);
2088	if (!log_tree_root)
2089		return -ENOMEM;
2090
2091	check.level = level;
2092	check.transid = fs_info->generation + 1;
2093	check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2094	log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
 
2095	if (IS_ERR(log_tree_root->node)) {
2096		btrfs_warn(fs_info, "failed to read log tree");
2097		ret = PTR_ERR(log_tree_root->node);
2098		log_tree_root->node = NULL;
2099		btrfs_put_root(log_tree_root);
2100		return ret;
2101	}
2102	if (!extent_buffer_uptodate(log_tree_root->node)) {
2103		btrfs_err(fs_info, "failed to read log tree");
2104		btrfs_put_root(log_tree_root);
 
2105		return -EIO;
2106	}
2107
2108	/* returns with log_tree_root freed on success */
2109	ret = btrfs_recover_log_trees(log_tree_root);
2110	if (ret) {
2111		btrfs_handle_fs_error(fs_info, ret,
2112				      "Failed to recover log tree");
2113		btrfs_put_root(log_tree_root);
 
2114		return ret;
2115	}
2116
2117	if (sb_rdonly(fs_info->sb)) {
2118		ret = btrfs_commit_super(fs_info);
2119		if (ret)
2120			return ret;
2121	}
2122
2123	return 0;
2124}
2125
2126static int load_global_roots_objectid(struct btrfs_root *tree_root,
2127				      struct btrfs_path *path, u64 objectid,
2128				      const char *name)
2129{
2130	struct btrfs_fs_info *fs_info = tree_root->fs_info;
2131	struct btrfs_root *root;
2132	u64 max_global_id = 0;
2133	int ret;
2134	struct btrfs_key key = {
2135		.objectid = objectid,
2136		.type = BTRFS_ROOT_ITEM_KEY,
2137		.offset = 0,
2138	};
2139	bool found = false;
2140
2141	/* If we have IGNOREDATACSUMS skip loading these roots. */
2142	if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2143	    btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2144		set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2145		return 0;
2146	}
2147
2148	while (1) {
2149		ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2150		if (ret < 0)
2151			break;
2152
2153		if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2154			ret = btrfs_next_leaf(tree_root, path);
2155			if (ret) {
2156				if (ret > 0)
2157					ret = 0;
2158				break;
2159			}
2160		}
2161		ret = 0;
2162
2163		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2164		if (key.objectid != objectid)
2165			break;
2166		btrfs_release_path(path);
2167
2168		/*
2169		 * Just worry about this for extent tree, it'll be the same for
2170		 * everybody.
2171		 */
2172		if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2173			max_global_id = max(max_global_id, key.offset);
2174
2175		found = true;
2176		root = read_tree_root_path(tree_root, path, &key);
2177		if (IS_ERR(root)) {
2178			if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2179				ret = PTR_ERR(root);
2180			break;
2181		}
2182		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2183		ret = btrfs_global_root_insert(root);
2184		if (ret) {
2185			btrfs_put_root(root);
2186			break;
2187		}
2188		key.offset++;
2189	}
2190	btrfs_release_path(path);
2191
2192	if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2193		fs_info->nr_global_roots = max_global_id + 1;
2194
2195	if (!found || ret) {
2196		if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2197			set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2198
2199		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2200			ret = ret ? ret : -ENOENT;
2201		else
2202			ret = 0;
2203		btrfs_err(fs_info, "failed to load root %s", name);
2204	}
2205	return ret;
2206}
2207
2208static int load_global_roots(struct btrfs_root *tree_root)
2209{
2210	struct btrfs_path *path;
2211	int ret = 0;
2212
2213	path = btrfs_alloc_path();
2214	if (!path)
2215		return -ENOMEM;
2216
2217	ret = load_global_roots_objectid(tree_root, path,
2218					 BTRFS_EXTENT_TREE_OBJECTID, "extent");
2219	if (ret)
2220		goto out;
2221	ret = load_global_roots_objectid(tree_root, path,
2222					 BTRFS_CSUM_TREE_OBJECTID, "csum");
2223	if (ret)
2224		goto out;
2225	if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2226		goto out;
2227	ret = load_global_roots_objectid(tree_root, path,
2228					 BTRFS_FREE_SPACE_TREE_OBJECTID,
2229					 "free space");
2230out:
2231	btrfs_free_path(path);
2232	return ret;
2233}
2234
2235static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2236{
2237	struct btrfs_root *tree_root = fs_info->tree_root;
2238	struct btrfs_root *root;
2239	struct btrfs_key location;
2240	int ret;
2241
2242	BUG_ON(!fs_info->tree_root);
2243
2244	ret = load_global_roots(tree_root);
2245	if (ret)
2246		return ret;
2247
2248	location.type = BTRFS_ROOT_ITEM_KEY;
2249	location.offset = 0;
2250
2251	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2252		location.objectid = BTRFS_BLOCK_GROUP_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				goto out;
2258			}
2259		} else {
2260			set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2261			fs_info->block_group_root = root;
2262		}
2263	}
 
 
2264
2265	location.objectid = BTRFS_DEV_TREE_OBJECTID;
2266	root = btrfs_read_tree_root(tree_root, &location);
2267	if (IS_ERR(root)) {
2268		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2269			ret = PTR_ERR(root);
2270			goto out;
2271		}
2272	} else {
2273		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2274		fs_info->dev_root = root;
2275	}
2276	/* Initialize fs_info for all devices in any case */
2277	ret = btrfs_init_devices_late(fs_info);
2278	if (ret)
2279		goto out;
2280
2281	/*
2282	 * This tree can share blocks with some other fs tree during relocation
2283	 * and we need a proper setup by btrfs_get_fs_root
2284	 */
2285	root = btrfs_get_fs_root(tree_root->fs_info,
2286				 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2287	if (IS_ERR(root)) {
2288		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2289			ret = PTR_ERR(root);
2290			goto out;
2291		}
2292	} else {
2293		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2294		fs_info->data_reloc_root = root;
2295	}
 
 
2296
2297	location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2298	root = btrfs_read_tree_root(tree_root, &location);
2299	if (!IS_ERR(root)) {
2300		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
 
2301		fs_info->quota_root = root;
2302	}
2303
2304	location.objectid = BTRFS_UUID_TREE_OBJECTID;
2305	root = btrfs_read_tree_root(tree_root, &location);
2306	if (IS_ERR(root)) {
2307		if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2308			ret = PTR_ERR(root);
2309			if (ret != -ENOENT)
2310				goto out;
2311		}
2312	} else {
2313		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2314		fs_info->uuid_root = root;
2315	}
2316
2317	if (btrfs_fs_incompat(fs_info, RAID_STRIPE_TREE)) {
2318		location.objectid = BTRFS_RAID_STRIPE_TREE_OBJECTID;
2319		root = btrfs_read_tree_root(tree_root, &location);
2320		if (IS_ERR(root)) {
2321			if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2322				ret = PTR_ERR(root);
2323				goto out;
2324			}
2325		} else {
2326			set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2327			fs_info->stripe_root = root;
2328		}
 
 
2329	}
2330
2331	return 0;
2332out:
2333	btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2334		   location.objectid, ret);
2335	return ret;
2336}
2337
2338/*
2339 * Real super block validation
2340 * NOTE: super csum type and incompat features will not be checked here.
2341 *
2342 * @sb:		super block to check
2343 * @mirror_num:	the super block number to check its bytenr:
2344 * 		0	the primary (1st) sb
2345 * 		1, 2	2nd and 3rd backup copy
2346 * 	       -1	skip bytenr check
2347 */
2348int btrfs_validate_super(struct btrfs_fs_info *fs_info,
2349			 struct btrfs_super_block *sb, int mirror_num)
2350{
2351	u64 nodesize = btrfs_super_nodesize(sb);
2352	u64 sectorsize = btrfs_super_sectorsize(sb);
2353	int ret = 0;
2354
2355	if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2356		btrfs_err(fs_info, "no valid FS found");
2357		ret = -EINVAL;
2358	}
2359	if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2360		btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2361				btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2362		ret = -EINVAL;
2363	}
2364	if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2365		btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2366				btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2367		ret = -EINVAL;
2368	}
2369	if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2370		btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2371				btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2372		ret = -EINVAL;
2373	}
2374	if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2375		btrfs_err(fs_info, "log_root level too big: %d >= %d",
2376				btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2377		ret = -EINVAL;
2378	}
2379
2380	/*
2381	 * Check sectorsize and nodesize first, other check will need it.
2382	 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2383	 */
2384	if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2385	    sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2386		btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2387		ret = -EINVAL;
2388	}
2389
2390	/*
2391	 * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2392	 *
2393	 * We can support 16K sectorsize with 64K page size without problem,
2394	 * but such sectorsize/pagesize combination doesn't make much sense.
2395	 * 4K will be our future standard, PAGE_SIZE is supported from the very
2396	 * beginning.
2397	 */
2398	if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2399		btrfs_err(fs_info,
2400			"sectorsize %llu not yet supported for page size %lu",
2401			sectorsize, PAGE_SIZE);
2402		ret = -EINVAL;
2403	}
2404
2405	if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2406	    nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2407		btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2408		ret = -EINVAL;
2409	}
2410	if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2411		btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2412			  le32_to_cpu(sb->__unused_leafsize), nodesize);
2413		ret = -EINVAL;
2414	}
2415
2416	/* Root alignment check */
2417	if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2418		btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2419			   btrfs_super_root(sb));
2420		ret = -EINVAL;
2421	}
2422	if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2423		btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2424			   btrfs_super_chunk_root(sb));
2425		ret = -EINVAL;
2426	}
2427	if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2428		btrfs_warn(fs_info, "log_root block unaligned: %llu",
2429			   btrfs_super_log_root(sb));
2430		ret = -EINVAL;
2431	}
2432
2433	if (!fs_info->fs_devices->temp_fsid &&
2434	    memcmp(fs_info->fs_devices->fsid, sb->fsid, BTRFS_FSID_SIZE) != 0) {
2435		btrfs_err(fs_info,
2436		"superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2437			  sb->fsid, fs_info->fs_devices->fsid);
2438		ret = -EINVAL;
2439	}
2440
2441	if (memcmp(fs_info->fs_devices->metadata_uuid, btrfs_sb_fsid_ptr(sb),
2442		   BTRFS_FSID_SIZE) != 0) {
2443		btrfs_err(fs_info,
2444"superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2445			  btrfs_sb_fsid_ptr(sb), fs_info->fs_devices->metadata_uuid);
2446		ret = -EINVAL;
2447	}
2448
2449	if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2450		   BTRFS_FSID_SIZE) != 0) {
2451		btrfs_err(fs_info,
2452			"dev_item UUID does not match metadata fsid: %pU != %pU",
2453			fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2454		ret = -EINVAL;
2455	}
2456
2457	/*
2458	 * Artificial requirement for block-group-tree to force newer features
2459	 * (free-space-tree, no-holes) so the test matrix is smaller.
2460	 */
2461	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2462	    (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2463	     !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2464		btrfs_err(fs_info,
2465		"block-group-tree feature requires fres-space-tree and no-holes");
2466		ret = -EINVAL;
2467	}
2468
2469	/*
2470	 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2471	 * done later
2472	 */
2473	if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2474		btrfs_err(fs_info, "bytes_used is too small %llu",
2475			  btrfs_super_bytes_used(sb));
2476		ret = -EINVAL;
2477	}
2478	if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2479		btrfs_err(fs_info, "invalid stripesize %u",
2480			  btrfs_super_stripesize(sb));
2481		ret = -EINVAL;
2482	}
2483	if (btrfs_super_num_devices(sb) > (1UL << 31))
2484		btrfs_warn(fs_info, "suspicious number of devices: %llu",
2485			   btrfs_super_num_devices(sb));
2486	if (btrfs_super_num_devices(sb) == 0) {
2487		btrfs_err(fs_info, "number of devices is 0");
2488		ret = -EINVAL;
2489	}
2490
2491	if (mirror_num >= 0 &&
2492	    btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2493		btrfs_err(fs_info, "super offset mismatch %llu != %u",
2494			  btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2495		ret = -EINVAL;
2496	}
2497
2498	/*
2499	 * Obvious sys_chunk_array corruptions, it must hold at least one key
2500	 * and one chunk
2501	 */
2502	if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2503		btrfs_err(fs_info, "system chunk array too big %u > %u",
2504			  btrfs_super_sys_array_size(sb),
2505			  BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2506		ret = -EINVAL;
2507	}
2508	if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2509			+ sizeof(struct btrfs_chunk)) {
2510		btrfs_err(fs_info, "system chunk array too small %u < %zu",
2511			  btrfs_super_sys_array_size(sb),
2512			  sizeof(struct btrfs_disk_key)
2513			  + sizeof(struct btrfs_chunk));
2514		ret = -EINVAL;
2515	}
2516
2517	/*
2518	 * The generation is a global counter, we'll trust it more than the others
2519	 * but it's still possible that it's the one that's wrong.
2520	 */
2521	if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2522		btrfs_warn(fs_info,
2523			"suspicious: generation < chunk_root_generation: %llu < %llu",
2524			btrfs_super_generation(sb),
2525			btrfs_super_chunk_root_generation(sb));
2526	if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2527	    && btrfs_super_cache_generation(sb) != (u64)-1)
2528		btrfs_warn(fs_info,
2529			"suspicious: generation < cache_generation: %llu < %llu",
2530			btrfs_super_generation(sb),
2531			btrfs_super_cache_generation(sb));
2532
2533	return ret;
2534}
2535
2536/*
2537 * Validation of super block at mount time.
2538 * Some checks already done early at mount time, like csum type and incompat
2539 * flags will be skipped.
2540 */
2541static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2542{
2543	return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2544}
2545
2546/*
2547 * Validation of super block at write time.
2548 * Some checks like bytenr check will be skipped as their values will be
2549 * overwritten soon.
2550 * Extra checks like csum type and incompat flags will be done here.
2551 */
2552static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2553				      struct btrfs_super_block *sb)
2554{
2555	int ret;
2556
2557	ret = btrfs_validate_super(fs_info, sb, -1);
2558	if (ret < 0)
2559		goto out;
2560	if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2561		ret = -EUCLEAN;
2562		btrfs_err(fs_info, "invalid csum type, has %u want %u",
2563			  btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2564		goto out;
2565	}
2566	if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2567		ret = -EUCLEAN;
2568		btrfs_err(fs_info,
2569		"invalid incompat flags, has 0x%llx valid mask 0x%llx",
2570			  btrfs_super_incompat_flags(sb),
2571			  (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2572		goto out;
2573	}
2574out:
2575	if (ret < 0)
2576		btrfs_err(fs_info,
2577		"super block corruption detected before writing it to disk");
2578	return ret;
2579}
2580
2581static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
 
 
2582{
2583	struct btrfs_tree_parent_check check = {
2584		.level = level,
2585		.transid = gen,
2586		.owner_root = root->root_key.objectid
2587	};
2588	int ret = 0;
 
 
 
 
 
 
 
 
 
 
 
 
2589
2590	root->node = read_tree_block(root->fs_info, bytenr, &check);
2591	if (IS_ERR(root->node)) {
2592		ret = PTR_ERR(root->node);
2593		root->node = NULL;
2594		return ret;
2595	}
2596	if (!extent_buffer_uptodate(root->node)) {
2597		free_extent_buffer(root->node);
2598		root->node = NULL;
2599		return -EIO;
 
2600	}
2601
2602	btrfs_set_root_node(&root->root_item, root->node);
2603	root->commit_root = btrfs_root_node(root);
2604	btrfs_set_root_refs(&root->root_item, 1);
2605	return ret;
2606}
2607
2608static int load_important_roots(struct btrfs_fs_info *fs_info)
2609{
2610	struct btrfs_super_block *sb = fs_info->super_copy;
2611	u64 gen, bytenr;
2612	int level, ret;
2613
2614	bytenr = btrfs_super_root(sb);
2615	gen = btrfs_super_generation(sb);
2616	level = btrfs_super_root_level(sb);
2617	ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2618	if (ret) {
2619		btrfs_warn(fs_info, "couldn't read tree root");
2620		return ret;
2621	}
2622	return 0;
2623}
2624
2625static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2626{
2627	int backup_index = find_newest_super_backup(fs_info);
2628	struct btrfs_super_block *sb = fs_info->super_copy;
2629	struct btrfs_root *tree_root = fs_info->tree_root;
2630	bool handle_error = false;
2631	int ret = 0;
2632	int i;
2633
2634	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2635		if (handle_error) {
2636			if (!IS_ERR(tree_root->node))
2637				free_extent_buffer(tree_root->node);
2638			tree_root->node = NULL;
2639
2640			if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2641				break;
2642
2643			free_root_pointers(fs_info, 0);
2644
2645			/*
2646			 * Don't use the log in recovery mode, it won't be
2647			 * valid
2648			 */
2649			btrfs_set_super_log_root(sb, 0);
2650
2651			btrfs_warn(fs_info, "try to load backup roots slot %d", i);
2652			ret = read_backup_root(fs_info, i);
2653			backup_index = ret;
2654			if (ret < 0)
2655				return ret;
2656		}
2657
2658		ret = load_important_roots(fs_info);
2659		if (ret) {
2660			handle_error = true;
2661			continue;
2662		}
2663
2664		/*
2665		 * No need to hold btrfs_root::objectid_mutex since the fs
2666		 * hasn't been fully initialised and we are the only user
2667		 */
2668		ret = btrfs_init_root_free_objectid(tree_root);
2669		if (ret < 0) {
2670			handle_error = true;
2671			continue;
2672		}
2673
2674		ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2675
2676		ret = btrfs_read_roots(fs_info);
2677		if (ret < 0) {
2678			handle_error = true;
2679			continue;
2680		}
2681
2682		/* All successful */
2683		fs_info->generation = btrfs_header_generation(tree_root->node);
2684		btrfs_set_last_trans_committed(fs_info, fs_info->generation);
2685		fs_info->last_reloc_trans = 0;
2686
2687		/* Always begin writing backup roots after the one being used */
2688		if (backup_index < 0) {
2689			fs_info->backup_root_index = 0;
2690		} else {
2691			fs_info->backup_root_index = backup_index + 1;
2692			fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2693		}
2694		break;
2695	}
2696
2697	return ret;
2698}
2699
2700void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2701{
2702	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2703	INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2704	INIT_LIST_HEAD(&fs_info->trans_list);
2705	INIT_LIST_HEAD(&fs_info->dead_roots);
2706	INIT_LIST_HEAD(&fs_info->delayed_iputs);
2707	INIT_LIST_HEAD(&fs_info->delalloc_roots);
2708	INIT_LIST_HEAD(&fs_info->caching_block_groups);
2709	spin_lock_init(&fs_info->delalloc_root_lock);
2710	spin_lock_init(&fs_info->trans_lock);
2711	spin_lock_init(&fs_info->fs_roots_radix_lock);
2712	spin_lock_init(&fs_info->delayed_iput_lock);
2713	spin_lock_init(&fs_info->defrag_inodes_lock);
 
2714	spin_lock_init(&fs_info->super_lock);
2715	spin_lock_init(&fs_info->buffer_lock);
2716	spin_lock_init(&fs_info->unused_bgs_lock);
2717	spin_lock_init(&fs_info->treelog_bg_lock);
2718	spin_lock_init(&fs_info->zone_active_bgs_lock);
2719	spin_lock_init(&fs_info->relocation_bg_lock);
2720	rwlock_init(&fs_info->tree_mod_log_lock);
2721	rwlock_init(&fs_info->global_root_lock);
2722	mutex_init(&fs_info->unused_bg_unpin_mutex);
2723	mutex_init(&fs_info->reclaim_bgs_lock);
2724	mutex_init(&fs_info->reloc_mutex);
2725	mutex_init(&fs_info->delalloc_root_mutex);
2726	mutex_init(&fs_info->zoned_meta_io_lock);
2727	mutex_init(&fs_info->zoned_data_reloc_io_lock);
2728	seqlock_init(&fs_info->profiles_lock);
2729
2730	btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
2731	btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
2732	btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
2733	btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
2734	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_prep,
2735				     BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2736	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
2737				     BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2738	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
2739				     BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2740	btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
2741				     BTRFS_LOCKDEP_TRANS_COMPLETED);
2742
2743	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2744	INIT_LIST_HEAD(&fs_info->space_info);
2745	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2746	INIT_LIST_HEAD(&fs_info->unused_bgs);
2747	INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2748	INIT_LIST_HEAD(&fs_info->zone_active_bgs);
2749#ifdef CONFIG_BTRFS_DEBUG
2750	INIT_LIST_HEAD(&fs_info->allocated_roots);
2751	INIT_LIST_HEAD(&fs_info->allocated_ebs);
2752	spin_lock_init(&fs_info->eb_leak_lock);
2753#endif
2754	fs_info->mapping_tree = RB_ROOT_CACHED;
2755	rwlock_init(&fs_info->mapping_tree_lock);
2756	btrfs_init_block_rsv(&fs_info->global_block_rsv,
2757			     BTRFS_BLOCK_RSV_GLOBAL);
2758	btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2759	btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2760	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2761	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2762			     BTRFS_BLOCK_RSV_DELOPS);
2763	btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2764			     BTRFS_BLOCK_RSV_DELREFS);
2765
2766	atomic_set(&fs_info->async_delalloc_pages, 0);
2767	atomic_set(&fs_info->defrag_running, 0);
 
2768	atomic_set(&fs_info->nr_delayed_iputs, 0);
2769	atomic64_set(&fs_info->tree_mod_seq, 0);
2770	fs_info->global_root_tree = RB_ROOT;
2771	fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2772	fs_info->metadata_ratio = 0;
2773	fs_info->defrag_inodes = RB_ROOT;
2774	atomic64_set(&fs_info->free_chunk_space, 0);
2775	fs_info->tree_mod_log = RB_ROOT;
2776	fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
 
 
 
 
2777	btrfs_init_ref_verify(fs_info);
2778
2779	fs_info->thread_pool_size = min_t(unsigned long,
2780					  num_online_cpus() + 2, 8);
2781
2782	INIT_LIST_HEAD(&fs_info->ordered_roots);
2783	spin_lock_init(&fs_info->ordered_root_lock);
2784
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2785	btrfs_init_scrub(fs_info);
 
 
 
2786	btrfs_init_balance(fs_info);
2787	btrfs_init_async_reclaim_work(fs_info);
 
 
 
2788
2789	rwlock_init(&fs_info->block_group_cache_lock);
2790	fs_info->block_group_cache_tree = RB_ROOT_CACHED;
2791
2792	extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2793			    IO_TREE_FS_EXCLUDED_EXTENTS);
 
 
 
 
 
 
 
 
2794
2795	mutex_init(&fs_info->ordered_operations_mutex);
2796	mutex_init(&fs_info->tree_log_mutex);
2797	mutex_init(&fs_info->chunk_mutex);
2798	mutex_init(&fs_info->transaction_kthread_mutex);
2799	mutex_init(&fs_info->cleaner_mutex);
2800	mutex_init(&fs_info->ro_block_group_mutex);
2801	init_rwsem(&fs_info->commit_root_sem);
2802	init_rwsem(&fs_info->cleanup_work_sem);
2803	init_rwsem(&fs_info->subvol_sem);
2804	sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2805
2806	btrfs_init_dev_replace_locks(fs_info);
2807	btrfs_init_qgroup(fs_info);
2808	btrfs_discard_init(fs_info);
2809
2810	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2811	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2812
2813	init_waitqueue_head(&fs_info->transaction_throttle);
2814	init_waitqueue_head(&fs_info->transaction_wait);
2815	init_waitqueue_head(&fs_info->transaction_blocked_wait);
2816	init_waitqueue_head(&fs_info->async_submit_wait);
2817	init_waitqueue_head(&fs_info->delayed_iputs_wait);
2818
2819	/* Usable values until the real ones are cached from the superblock */
2820	fs_info->nodesize = 4096;
2821	fs_info->sectorsize = 4096;
2822	fs_info->sectorsize_bits = ilog2(4096);
2823	fs_info->stripesize = 4096;
2824
2825	/* Default compress algorithm when user does -o compress */
2826	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2827
2828	fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
2829
2830	spin_lock_init(&fs_info->swapfile_pins_lock);
2831	fs_info->swapfile_pins = RB_ROOT;
2832
2833	fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2834	INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2835}
2836
2837static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2838{
2839	int ret;
2840
2841	fs_info->sb = sb;
2842	sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2843	sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2844
2845	ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2846	if (ret)
2847		return ret;
2848
2849	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2850	if (ret)
2851		return ret;
2852
2853	fs_info->dirty_metadata_batch = PAGE_SIZE *
2854					(1 + ilog2(nr_cpu_ids));
2855
2856	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2857	if (ret)
2858		return ret;
2859
2860	ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2861			GFP_KERNEL);
2862	if (ret)
2863		return ret;
2864
2865	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2866					GFP_KERNEL);
2867	if (!fs_info->delayed_root)
2868		return -ENOMEM;
2869	btrfs_init_delayed_root(fs_info->delayed_root);
2870
2871	if (sb_rdonly(sb))
2872		set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2873
2874	return btrfs_alloc_stripe_hash_table(fs_info);
2875}
2876
2877static int btrfs_uuid_rescan_kthread(void *data)
2878{
2879	struct btrfs_fs_info *fs_info = data;
2880	int ret;
2881
2882	/*
2883	 * 1st step is to iterate through the existing UUID tree and
2884	 * to delete all entries that contain outdated data.
2885	 * 2nd step is to add all missing entries to the UUID tree.
2886	 */
2887	ret = btrfs_uuid_tree_iterate(fs_info);
2888	if (ret < 0) {
2889		if (ret != -EINTR)
2890			btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2891				   ret);
2892		up(&fs_info->uuid_tree_rescan_sem);
2893		return ret;
2894	}
2895	return btrfs_uuid_scan_kthread(data);
2896}
2897
2898static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2899{
2900	struct task_struct *task;
2901
2902	down(&fs_info->uuid_tree_rescan_sem);
2903	task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2904	if (IS_ERR(task)) {
2905		/* fs_info->update_uuid_tree_gen remains 0 in all error case */
2906		btrfs_warn(fs_info, "failed to start uuid_rescan task");
2907		up(&fs_info->uuid_tree_rescan_sem);
2908		return PTR_ERR(task);
2909	}
2910
2911	return 0;
2912}
2913
2914static int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2915{
2916	u64 root_objectid = 0;
2917	struct btrfs_root *gang[8];
2918	int i = 0;
2919	int err = 0;
2920	unsigned int ret = 0;
2921
2922	while (1) {
2923		spin_lock(&fs_info->fs_roots_radix_lock);
2924		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2925					     (void **)gang, root_objectid,
2926					     ARRAY_SIZE(gang));
2927		if (!ret) {
2928			spin_unlock(&fs_info->fs_roots_radix_lock);
2929			break;
2930		}
2931		root_objectid = gang[ret - 1]->root_key.objectid + 1;
2932
2933		for (i = 0; i < ret; i++) {
2934			/* Avoid to grab roots in dead_roots. */
2935			if (btrfs_root_refs(&gang[i]->root_item) == 0) {
2936				gang[i] = NULL;
2937				continue;
2938			}
2939			/* Grab all the search result for later use. */
2940			gang[i] = btrfs_grab_root(gang[i]);
2941		}
2942		spin_unlock(&fs_info->fs_roots_radix_lock);
2943
2944		for (i = 0; i < ret; i++) {
2945			if (!gang[i])
2946				continue;
2947			root_objectid = gang[i]->root_key.objectid;
2948			err = btrfs_orphan_cleanup(gang[i]);
2949			if (err)
2950				goto out;
2951			btrfs_put_root(gang[i]);
2952		}
2953		root_objectid++;
2954	}
2955out:
2956	/* Release the uncleaned roots due to error. */
2957	for (; i < ret; i++) {
2958		if (gang[i])
2959			btrfs_put_root(gang[i]);
2960	}
2961	return err;
2962}
2963
2964/*
2965 * Mounting logic specific to read-write file systems. Shared by open_ctree
2966 * and btrfs_remount when remounting from read-only to read-write.
2967 */
2968int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
2969{
2970	int ret;
2971	const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
2972	bool rebuild_free_space_tree = false;
2973
2974	if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2975	    btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2976		if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
2977			btrfs_warn(fs_info,
2978				   "'clear_cache' option is ignored with extent tree v2");
2979		else
2980			rebuild_free_space_tree = true;
2981	} else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2982		   !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2983		btrfs_warn(fs_info, "free space tree is invalid");
2984		rebuild_free_space_tree = true;
2985	}
2986
2987	if (rebuild_free_space_tree) {
2988		btrfs_info(fs_info, "rebuilding free space tree");
2989		ret = btrfs_rebuild_free_space_tree(fs_info);
2990		if (ret) {
2991			btrfs_warn(fs_info,
2992				   "failed to rebuild free space tree: %d", ret);
2993			goto out;
2994		}
2995	}
2996
2997	if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2998	    !btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
2999		btrfs_info(fs_info, "disabling free space tree");
3000		ret = btrfs_delete_free_space_tree(fs_info);
3001		if (ret) {
3002			btrfs_warn(fs_info,
3003				   "failed to disable free space tree: %d", ret);
3004			goto out;
3005		}
3006	}
3007
3008	/*
3009	 * btrfs_find_orphan_roots() is responsible for finding all the dead
3010	 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3011	 * them into the fs_info->fs_roots_radix tree. This must be done before
3012	 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3013	 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3014	 * item before the root's tree is deleted - this means that if we unmount
3015	 * or crash before the deletion completes, on the next mount we will not
3016	 * delete what remains of the tree because the orphan item does not
3017	 * exists anymore, which is what tells us we have a pending deletion.
3018	 */
3019	ret = btrfs_find_orphan_roots(fs_info);
3020	if (ret)
3021		goto out;
3022
3023	ret = btrfs_cleanup_fs_roots(fs_info);
3024	if (ret)
3025		goto out;
3026
3027	down_read(&fs_info->cleanup_work_sem);
3028	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3029	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3030		up_read(&fs_info->cleanup_work_sem);
3031		goto out;
3032	}
3033	up_read(&fs_info->cleanup_work_sem);
3034
3035	mutex_lock(&fs_info->cleaner_mutex);
3036	ret = btrfs_recover_relocation(fs_info);
3037	mutex_unlock(&fs_info->cleaner_mutex);
3038	if (ret < 0) {
3039		btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3040		goto out;
3041	}
3042
3043	if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3044	    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3045		btrfs_info(fs_info, "creating free space tree");
3046		ret = btrfs_create_free_space_tree(fs_info);
3047		if (ret) {
3048			btrfs_warn(fs_info,
3049				"failed to create free space tree: %d", ret);
3050			goto out;
3051		}
3052	}
3053
3054	if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3055		ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3056		if (ret)
3057			goto out;
3058	}
3059
3060	ret = btrfs_resume_balance_async(fs_info);
3061	if (ret)
3062		goto out;
3063
3064	ret = btrfs_resume_dev_replace_async(fs_info);
3065	if (ret) {
3066		btrfs_warn(fs_info, "failed to resume dev_replace");
3067		goto out;
3068	}
3069
3070	btrfs_qgroup_rescan_resume(fs_info);
3071
3072	if (!fs_info->uuid_root) {
3073		btrfs_info(fs_info, "creating UUID tree");
3074		ret = btrfs_create_uuid_tree(fs_info);
3075		if (ret) {
3076			btrfs_warn(fs_info,
3077				   "failed to create the UUID tree %d", ret);
3078			goto out;
3079		}
3080	}
3081
3082out:
3083	return ret;
3084}
3085
3086/*
3087 * Do various sanity and dependency checks of different features.
3088 *
3089 * @is_rw_mount:	If the mount is read-write.
3090 *
3091 * This is the place for less strict checks (like for subpage or artificial
3092 * feature dependencies).
3093 *
3094 * For strict checks or possible corruption detection, see
3095 * btrfs_validate_super().
3096 *
3097 * This should be called after btrfs_parse_options(), as some mount options
3098 * (space cache related) can modify on-disk format like free space tree and
3099 * screw up certain feature dependencies.
3100 */
3101int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3102{
3103	struct btrfs_super_block *disk_super = fs_info->super_copy;
3104	u64 incompat = btrfs_super_incompat_flags(disk_super);
3105	const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3106	const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3107
3108	if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3109		btrfs_err(fs_info,
3110		"cannot mount because of unknown incompat features (0x%llx)",
3111		    incompat);
3112		return -EINVAL;
3113	}
3114
3115	/* Runtime limitation for mixed block groups. */
3116	if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3117	    (fs_info->sectorsize != fs_info->nodesize)) {
3118		btrfs_err(fs_info,
3119"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3120			fs_info->nodesize, fs_info->sectorsize);
3121		return -EINVAL;
3122	}
3123
3124	/* Mixed backref is an always-enabled feature. */
3125	incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3126
3127	/* Set compression related flags just in case. */
3128	if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3129		incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3130	else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3131		incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3132
3133	/*
3134	 * An ancient flag, which should really be marked deprecated.
3135	 * Such runtime limitation doesn't really need a incompat flag.
3136	 */
3137	if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3138		incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3139
3140	if (compat_ro_unsupp && is_rw_mount) {
3141		btrfs_err(fs_info,
3142	"cannot mount read-write because of unknown compat_ro features (0x%llx)",
3143		       compat_ro);
3144		return -EINVAL;
3145	}
3146
3147	/*
3148	 * We have unsupported RO compat features, although RO mounted, we
3149	 * should not cause any metadata writes, including log replay.
3150	 * Or we could screw up whatever the new feature requires.
3151	 */
3152	if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3153	    !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3154		btrfs_err(fs_info,
3155"cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3156			  compat_ro);
3157		return -EINVAL;
3158	}
3159
3160	/*
3161	 * Artificial limitations for block group tree, to force
3162	 * block-group-tree to rely on no-holes and free-space-tree.
3163	 */
3164	if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3165	    (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3166	     !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3167		btrfs_err(fs_info,
3168"block-group-tree feature requires no-holes and free-space-tree features");
3169		return -EINVAL;
3170	}
3171
3172	/*
3173	 * Subpage runtime limitation on v1 cache.
3174	 *
3175	 * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3176	 * we're already defaulting to v2 cache, no need to bother v1 as it's
3177	 * going to be deprecated anyway.
3178	 */
3179	if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3180		btrfs_warn(fs_info,
3181	"v1 space cache is not supported for page size %lu with sectorsize %u",
3182			   PAGE_SIZE, fs_info->sectorsize);
3183		return -EINVAL;
3184	}
3185
3186	/* This can be called by remount, we need to protect the super block. */
3187	spin_lock(&fs_info->super_lock);
3188	btrfs_set_super_incompat_flags(disk_super, incompat);
3189	spin_unlock(&fs_info->super_lock);
3190
3191	return 0;
3192}
3193
3194int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3195		      char *options)
3196{
3197	u32 sectorsize;
3198	u32 nodesize;
3199	u32 stripesize;
3200	u64 generation;
3201	u16 csum_type;
3202	struct btrfs_super_block *disk_super;
3203	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3204	struct btrfs_root *tree_root;
3205	struct btrfs_root *chunk_root;
3206	int ret;
3207	int level;
3208
3209	ret = init_mount_fs_info(fs_info, sb);
3210	if (ret)
3211		goto fail;
3212
3213	/* These need to be init'ed before we start creating inodes and such. */
3214	tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3215				     GFP_KERNEL);
3216	fs_info->tree_root = tree_root;
3217	chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3218				      GFP_KERNEL);
3219	fs_info->chunk_root = chunk_root;
3220	if (!tree_root || !chunk_root) {
3221		ret = -ENOMEM;
3222		goto fail;
3223	}
3224
3225	ret = btrfs_init_btree_inode(sb);
3226	if (ret)
3227		goto fail;
3228
3229	invalidate_bdev(fs_devices->latest_dev->bdev);
3230
3231	/*
3232	 * Read super block and check the signature bytes only
3233	 */
3234	disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3235	if (IS_ERR(disk_super)) {
3236		ret = PTR_ERR(disk_super);
3237		goto fail_alloc;
3238	}
3239
3240	btrfs_info(fs_info, "first mount of filesystem %pU", disk_super->fsid);
3241	/*
3242	 * Verify the type first, if that or the checksum value are
3243	 * corrupted, we'll find out
3244	 */
3245	csum_type = btrfs_super_csum_type(disk_super);
3246	if (!btrfs_supported_super_csum(csum_type)) {
3247		btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3248			  csum_type);
3249		ret = -EINVAL;
3250		btrfs_release_disk_super(disk_super);
3251		goto fail_alloc;
3252	}
3253
3254	fs_info->csum_size = btrfs_super_csum_size(disk_super);
3255
3256	ret = btrfs_init_csum_hash(fs_info, csum_type);
3257	if (ret) {
3258		btrfs_release_disk_super(disk_super);
3259		goto fail_alloc;
3260	}
3261
3262	/*
3263	 * We want to check superblock checksum, the type is stored inside.
3264	 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3265	 */
3266	if (btrfs_check_super_csum(fs_info, disk_super)) {
3267		btrfs_err(fs_info, "superblock checksum mismatch");
3268		ret = -EINVAL;
3269		btrfs_release_disk_super(disk_super);
3270		goto fail_alloc;
3271	}
3272
3273	/*
3274	 * super_copy is zeroed at allocation time and we never touch the
3275	 * following bytes up to INFO_SIZE, the checksum is calculated from
3276	 * the whole block of INFO_SIZE
3277	 */
3278	memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3279	btrfs_release_disk_super(disk_super);
3280
3281	disk_super = fs_info->super_copy;
3282
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3283	memcpy(fs_info->super_for_commit, fs_info->super_copy,
3284	       sizeof(*fs_info->super_for_commit));
3285
3286	ret = btrfs_validate_mount_super(fs_info);
3287	if (ret) {
3288		btrfs_err(fs_info, "superblock contains fatal errors");
3289		ret = -EINVAL;
3290		goto fail_alloc;
3291	}
3292
3293	if (!btrfs_super_root(disk_super)) {
3294		btrfs_err(fs_info, "invalid superblock tree root bytenr");
3295		ret = -EINVAL;
3296		goto fail_alloc;
3297	}
3298
3299	/* check FS state, whether FS is broken. */
3300	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3301		WRITE_ONCE(fs_info->fs_error, -EUCLEAN);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3302
3303	/* Set up fs_info before parsing mount options */
3304	nodesize = btrfs_super_nodesize(disk_super);
3305	sectorsize = btrfs_super_sectorsize(disk_super);
3306	stripesize = sectorsize;
3307	fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3308	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3309
 
3310	fs_info->nodesize = nodesize;
3311	fs_info->sectorsize = sectorsize;
3312	fs_info->sectorsize_bits = ilog2(sectorsize);
3313	fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3314	fs_info->stripesize = stripesize;
3315
3316	/*
3317	 * Handle the space caching options appropriately now that we have the
3318	 * super block loaded and validated.
3319	 */
3320	btrfs_set_free_space_cache_settings(fs_info);
3321
3322	if (!btrfs_check_options(fs_info, &fs_info->mount_opt, sb->s_flags)) {
3323		ret = -EINVAL;
3324		goto fail_alloc;
 
3325	}
3326
3327	ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3328	if (ret < 0)
3329		goto fail_alloc;
3330
3331	/*
3332	 * At this point our mount options are validated, if we set ->max_inline
3333	 * to something non-standard make sure we truncate it to sectorsize.
3334	 */
3335	fs_info->max_inline = min_t(u64, fs_info->max_inline, fs_info->sectorsize);
3336
3337	if (sectorsize < PAGE_SIZE) {
3338		struct btrfs_subpage_info *subpage_info;
3339
3340		btrfs_warn(fs_info,
3341		"read-write for sector size %u with page size %lu is experimental",
3342			   sectorsize, PAGE_SIZE);
3343		subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3344		if (!subpage_info) {
3345			ret = -ENOMEM;
3346			goto fail_alloc;
3347		}
3348		btrfs_init_subpage_info(subpage_info, sectorsize);
3349		fs_info->subpage_info = subpage_info;
3350	}
3351
3352	ret = btrfs_init_workqueues(fs_info);
3353	if (ret)
 
3354		goto fail_sb_buffer;
 
3355
 
 
 
 
3356	sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3357	sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3358
3359	sb->s_blocksize = sectorsize;
3360	sb->s_blocksize_bits = blksize_bits(sectorsize);
3361	memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3362
3363	mutex_lock(&fs_info->chunk_mutex);
3364	ret = btrfs_read_sys_array(fs_info);
3365	mutex_unlock(&fs_info->chunk_mutex);
3366	if (ret) {
3367		btrfs_err(fs_info, "failed to read the system array: %d", ret);
3368		goto fail_sb_buffer;
3369	}
3370
3371	generation = btrfs_super_chunk_root_generation(disk_super);
3372	level = btrfs_super_chunk_root_level(disk_super);
3373	ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3374			      generation, level);
3375	if (ret) {
 
 
 
 
 
3376		btrfs_err(fs_info, "failed to read chunk root");
 
 
 
3377		goto fail_tree_roots;
3378	}
 
 
3379
3380	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3381			   offsetof(struct btrfs_header, chunk_tree_uuid),
3382			   BTRFS_UUID_SIZE);
3383
3384	ret = btrfs_read_chunk_tree(fs_info);
3385	if (ret) {
3386		btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3387		goto fail_tree_roots;
3388	}
3389
3390	/*
3391	 * At this point we know all the devices that make this filesystem,
3392	 * including the seed devices but we don't know yet if the replace
3393	 * target is required. So free devices that are not part of this
3394	 * filesystem but skip the replace target device which is checked
3395	 * below in btrfs_init_dev_replace().
3396	 */
3397	btrfs_free_extra_devids(fs_devices);
3398	if (!fs_devices->latest_dev->bdev) {
 
3399		btrfs_err(fs_info, "failed to read devices");
3400		ret = -EIO;
3401		goto fail_tree_roots;
3402	}
3403
3404	ret = init_tree_roots(fs_info);
3405	if (ret)
3406		goto fail_tree_roots;
3407
3408	/*
3409	 * Get zone type information of zoned block devices. This will also
3410	 * handle emulation of a zoned filesystem if a regular device has the
3411	 * zoned incompat feature flag set.
3412	 */
3413	ret = btrfs_get_dev_zone_info_all_devices(fs_info);
 
 
 
 
 
 
 
 
 
 
 
 
 
3414	if (ret) {
3415		btrfs_err(fs_info,
3416			  "zoned: failed to read device zone info: %d", ret);
3417		goto fail_block_groups;
3418	}
3419
3420	/*
3421	 * If we have a uuid root and we're not being told to rescan we need to
3422	 * check the generation here so we can set the
3423	 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3424	 * transaction during a balance or the log replay without updating the
3425	 * uuid generation, and then if we crash we would rescan the uuid tree,
3426	 * even though it was perfectly fine.
3427	 */
3428	if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3429	    fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3430		set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3431
3432	ret = btrfs_verify_dev_extents(fs_info);
3433	if (ret) {
3434		btrfs_err(fs_info,
3435			  "failed to verify dev extents against chunks: %d",
3436			  ret);
3437		goto fail_block_groups;
3438	}
3439	ret = btrfs_recover_balance(fs_info);
3440	if (ret) {
3441		btrfs_err(fs_info, "failed to recover balance: %d", ret);
3442		goto fail_block_groups;
3443	}
3444
3445	ret = btrfs_init_dev_stats(fs_info);
3446	if (ret) {
3447		btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3448		goto fail_block_groups;
3449	}
3450
3451	ret = btrfs_init_dev_replace(fs_info);
3452	if (ret) {
3453		btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3454		goto fail_block_groups;
3455	}
3456
3457	ret = btrfs_check_zoned_mode(fs_info);
 
 
3458	if (ret) {
3459		btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3460			  ret);
3461		goto fail_block_groups;
3462	}
3463
3464	ret = btrfs_sysfs_add_fsid(fs_devices);
3465	if (ret) {
3466		btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3467				ret);
3468		goto fail_block_groups;
3469	}
3470
3471	ret = btrfs_sysfs_add_mounted(fs_info);
3472	if (ret) {
3473		btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3474		goto fail_fsdev_sysfs;
3475	}
3476
3477	ret = btrfs_init_space_info(fs_info);
3478	if (ret) {
3479		btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3480		goto fail_sysfs;
3481	}
3482
3483	ret = btrfs_read_block_groups(fs_info);
3484	if (ret) {
3485		btrfs_err(fs_info, "failed to read block groups: %d", ret);
3486		goto fail_sysfs;
3487	}
3488
3489	btrfs_free_zone_cache(fs_info);
3490
3491	btrfs_check_active_zone_reservation(fs_info);
3492
3493	if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3494	    !btrfs_check_rw_degradable(fs_info, NULL)) {
3495		btrfs_warn(fs_info,
3496		"writable mount is not allowed due to too many missing devices");
3497		ret = -EINVAL;
3498		goto fail_sysfs;
3499	}
3500
3501	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3502					       "btrfs-cleaner");
3503	if (IS_ERR(fs_info->cleaner_kthread)) {
3504		ret = PTR_ERR(fs_info->cleaner_kthread);
3505		goto fail_sysfs;
3506	}
3507
3508	fs_info->transaction_kthread = kthread_run(transaction_kthread,
3509						   tree_root,
3510						   "btrfs-transaction");
3511	if (IS_ERR(fs_info->transaction_kthread)) {
3512		ret = PTR_ERR(fs_info->transaction_kthread);
3513		goto fail_cleaner;
 
 
 
 
3514	}
3515
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3516	ret = btrfs_read_qgroup_config(fs_info);
3517	if (ret)
3518		goto fail_trans_kthread;
3519
3520	if (btrfs_build_ref_tree(fs_info))
3521		btrfs_err(fs_info, "couldn't build ref tree");
3522
3523	/* do not make disk changes in broken FS or nologreplay is given */
3524	if (btrfs_super_log_root(disk_super) != 0 &&
3525	    !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3526		btrfs_info(fs_info, "start tree-log replay");
3527		ret = btrfs_replay_log(fs_info, fs_devices);
 
 
 
 
 
 
 
 
 
 
 
 
3528		if (ret)
3529			goto fail_qgroup;
 
 
 
 
 
 
 
 
 
 
3530	}
3531
3532	fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
 
 
 
 
3533	if (IS_ERR(fs_info->fs_root)) {
3534		ret = PTR_ERR(fs_info->fs_root);
3535		btrfs_warn(fs_info, "failed to read fs tree: %d", ret);
3536		fs_info->fs_root = NULL;
3537		goto fail_qgroup;
3538	}
3539
3540	if (sb_rdonly(sb))
3541		return 0;
3542
3543	ret = btrfs_start_pre_rw_mount(fs_info);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3544	if (ret) {
 
3545		close_ctree(fs_info);
3546		return ret;
3547	}
3548	btrfs_discard_resume(fs_info);
3549
3550	if (fs_info->uuid_root &&
3551	    (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3552	     fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
 
 
 
 
 
 
 
 
 
 
 
3553		btrfs_info(fs_info, "checking UUID tree");
3554		ret = btrfs_check_uuid_tree(fs_info);
3555		if (ret) {
3556			btrfs_warn(fs_info,
3557				"failed to check the UUID tree: %d", ret);
3558			close_ctree(fs_info);
3559			return ret;
3560		}
 
 
3561	}
3562
3563	set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3564
3565	/* Kick the cleaner thread so it'll start deleting snapshots. */
3566	if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3567		wake_up_process(fs_info->cleaner_kthread);
 
 
3568
3569	return 0;
3570
3571fail_qgroup:
3572	btrfs_free_qgroup_config(fs_info);
3573fail_trans_kthread:
3574	kthread_stop(fs_info->transaction_kthread);
3575	btrfs_cleanup_transaction(fs_info);
3576	btrfs_free_fs_roots(fs_info);
3577fail_cleaner:
3578	kthread_stop(fs_info->cleaner_kthread);
3579
3580	/*
3581	 * make sure we're done with the btree inode before we stop our
3582	 * kthreads
3583	 */
3584	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3585
3586fail_sysfs:
3587	btrfs_sysfs_remove_mounted(fs_info);
3588
3589fail_fsdev_sysfs:
3590	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3591
3592fail_block_groups:
3593	btrfs_put_block_group_cache(fs_info);
3594
3595fail_tree_roots:
3596	if (fs_info->data_reloc_root)
3597		btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3598	free_root_pointers(fs_info, true);
3599	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3600
3601fail_sb_buffer:
3602	btrfs_stop_all_workers(fs_info);
3603	btrfs_free_block_groups(fs_info);
 
 
3604fail_alloc:
3605	btrfs_mapping_tree_free(fs_info);
 
3606
3607	iput(fs_info->btree_inode);
 
 
 
 
 
 
 
 
 
 
3608fail:
 
3609	btrfs_close_devices(fs_info->fs_devices);
3610	ASSERT(ret < 0);
3611	return ret;
3612}
3613ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3614
3615static void btrfs_end_super_write(struct bio *bio)
3616{
3617	struct btrfs_device *device = bio->bi_private;
3618	struct bio_vec *bvec;
3619	struct bvec_iter_all iter_all;
3620	struct page *page;
3621
3622	bio_for_each_segment_all(bvec, bio, iter_all) {
3623		page = bvec->bv_page;
3624
3625		if (bio->bi_status) {
3626			btrfs_warn_rl_in_rcu(device->fs_info,
3627				"lost page write due to IO error on %s (%d)",
3628				btrfs_dev_name(device),
3629				blk_status_to_errno(bio->bi_status));
3630			ClearPageUptodate(page);
3631			SetPageError(page);
3632			btrfs_dev_stat_inc_and_print(device,
3633						     BTRFS_DEV_STAT_WRITE_ERRS);
3634		} else {
3635			SetPageUptodate(page);
3636		}
3637
3638		put_page(page);
3639		unlock_page(page);
3640	}
3641
3642	bio_put(bio);
 
 
 
 
3643}
 
3644
3645struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3646						   int copy_num, bool drop_cache)
3647{
3648	struct btrfs_super_block *super;
3649	struct page *page;
3650	u64 bytenr, bytenr_orig;
3651	struct address_space *mapping = bdev->bd_inode->i_mapping;
3652	int ret;
3653
3654	bytenr_orig = btrfs_sb_offset(copy_num);
3655	ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3656	if (ret == -ENOENT)
3657		return ERR_PTR(-EINVAL);
3658	else if (ret)
3659		return ERR_PTR(ret);
3660
3661	if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3662		return ERR_PTR(-EINVAL);
3663
3664	if (drop_cache) {
3665		/* This should only be called with the primary sb. */
3666		ASSERT(copy_num == 0);
3667
3668		/*
3669		 * Drop the page of the primary superblock, so later read will
3670		 * always read from the device.
3671		 */
3672		invalidate_inode_pages2_range(mapping,
3673				bytenr >> PAGE_SHIFT,
3674				(bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3675	}
 
 
 
 
 
 
 
 
 
 
3676
3677	page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3678	if (IS_ERR(page))
3679		return ERR_CAST(page);
3680
3681	super = page_address(page);
3682	if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3683		btrfs_release_disk_super(super);
3684		return ERR_PTR(-ENODATA);
3685	}
 
 
3686
3687	if (btrfs_super_bytenr(super) != bytenr_orig) {
3688		btrfs_release_disk_super(super);
3689		return ERR_PTR(-EINVAL);
 
 
3690	}
3691
3692	return super;
 
3693}
3694
3695
3696struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3697{
3698	struct btrfs_super_block *super, *latest = NULL;
 
 
3699	int i;
3700	u64 transid = 0;
 
3701
3702	/* we would like to check all the supers, but that would make
3703	 * a btrfs mount succeed after a mkfs from a different FS.
3704	 * So, we need to add a special mount option to scan for
3705	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3706	 */
3707	for (i = 0; i < 1; i++) {
3708		super = btrfs_read_dev_one_super(bdev, i, false);
3709		if (IS_ERR(super))
3710			continue;
3711
 
 
3712		if (!latest || btrfs_super_generation(super) > transid) {
3713			if (latest)
3714				btrfs_release_disk_super(super);
3715
3716			latest = super;
3717			transid = btrfs_super_generation(super);
 
 
3718		}
3719	}
3720
3721	return super;
 
 
 
3722}
3723
3724/*
3725 * Write superblock @sb to the @device. Do not wait for completion, all the
3726 * pages we use for writing are locked.
3727 *
3728 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3729 * the expected device size at commit time. Note that max_mirrors must be
3730 * same for write and wait phases.
3731 *
3732 * Return number of errors when page is not found or submission fails.
3733 */
3734static int write_dev_supers(struct btrfs_device *device,
3735			    struct btrfs_super_block *sb, int max_mirrors)
3736{
3737	struct btrfs_fs_info *fs_info = device->fs_info;
3738	struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3739	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 
3740	int i;
 
3741	int errors = 0;
3742	int ret;
3743	u64 bytenr, bytenr_orig;
3744
3745	if (max_mirrors == 0)
3746		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3747
3748	shash->tfm = fs_info->csum_shash;
3749
3750	for (i = 0; i < max_mirrors; i++) {
3751		struct page *page;
3752		struct bio *bio;
3753		struct btrfs_super_block *disk_super;
3754
3755		bytenr_orig = btrfs_sb_offset(i);
3756		ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3757		if (ret == -ENOENT) {
3758			continue;
3759		} else if (ret < 0) {
3760			btrfs_err(device->fs_info,
3761				"couldn't get super block location for mirror %d",
3762				i);
3763			errors++;
3764			continue;
3765		}
3766		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3767		    device->commit_total_bytes)
3768			break;
3769
3770		btrfs_set_super_bytenr(sb, bytenr_orig);
3771
3772		crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3773				    BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3774				    sb->csum);
3775
3776		page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3777					   GFP_NOFS);
3778		if (!page) {
 
 
3779			btrfs_err(device->fs_info,
3780			    "couldn't get super block page for bytenr %llu",
3781			    bytenr);
3782			errors++;
3783			continue;
3784		}
3785
3786		/* Bump the refcount for wait_dev_supers() */
3787		get_page(page);
3788
3789		disk_super = page_address(page);
3790		memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3791
3792		/*
3793		 * Directly use bios here instead of relying on the page cache
3794		 * to do I/O, so we don't lose the ability to do integrity
3795		 * checking.
3796		 */
3797		bio = bio_alloc(device->bdev, 1,
3798				REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
3799				GFP_NOFS);
3800		bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3801		bio->bi_private = device;
3802		bio->bi_end_io = btrfs_end_super_write;
3803		__bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3804			       offset_in_page(bytenr));
3805
3806		/*
3807		 * We FUA only the first super block.  The others we allow to
3808		 * go down lazy and there's a short window where the on-disk
3809		 * copies might still contain the older version.
3810		 */
 
3811		if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3812			bio->bi_opf |= REQ_FUA;
3813		submit_bio(bio);
3814
3815		if (btrfs_advance_sb_log(device, i))
3816			errors++;
3817	}
3818	return errors < i ? 0 : -1;
3819}
3820
3821/*
3822 * Wait for write completion of superblocks done by write_dev_supers,
3823 * @max_mirrors same for write and wait phases.
3824 *
3825 * Return number of errors when page is not found or not marked up to
3826 * date.
3827 */
3828static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3829{
 
3830	int i;
3831	int errors = 0;
3832	bool primary_failed = false;
3833	int ret;
3834	u64 bytenr;
3835
3836	if (max_mirrors == 0)
3837		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3838
3839	for (i = 0; i < max_mirrors; i++) {
3840		struct page *page;
3841
3842		ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3843		if (ret == -ENOENT) {
3844			break;
3845		} else if (ret < 0) {
3846			errors++;
3847			if (i == 0)
3848				primary_failed = true;
3849			continue;
3850		}
3851		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3852		    device->commit_total_bytes)
3853			break;
3854
3855		page = find_get_page(device->bdev->bd_inode->i_mapping,
3856				     bytenr >> PAGE_SHIFT);
3857		if (!page) {
 
3858			errors++;
3859			if (i == 0)
3860				primary_failed = true;
3861			continue;
3862		}
3863		/* Page is submitted locked and unlocked once the IO completes */
3864		wait_on_page_locked(page);
3865		if (PageError(page)) {
3866			errors++;
3867			if (i == 0)
3868				primary_failed = true;
3869		}
3870
3871		/* Drop our reference */
3872		put_page(page);
3873
3874		/* Drop the reference from the writing run */
3875		put_page(page);
3876	}
3877
3878	/* log error, force error return */
3879	if (primary_failed) {
3880		btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3881			  device->devid);
3882		return -1;
3883	}
3884
3885	return errors < i ? 0 : -1;
3886}
3887
3888/*
3889 * endio for the write_dev_flush, this will wake anyone waiting
3890 * for the barrier when it is done
3891 */
3892static void btrfs_end_empty_barrier(struct bio *bio)
3893{
3894	bio_uninit(bio);
3895	complete(bio->bi_private);
3896}
3897
3898/*
3899 * Submit a flush request to the device if it supports it. Error handling is
3900 * done in the waiting counterpart.
3901 */
3902static void write_dev_flush(struct btrfs_device *device)
3903{
3904	struct bio *bio = &device->flush_bio;
 
3905
3906	device->last_flush_error = BLK_STS_OK;
 
3907
3908	bio_init(bio, device->bdev, NULL, 0,
3909		 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
3910	bio->bi_end_io = btrfs_end_empty_barrier;
 
 
3911	init_completion(&device->flush_wait);
3912	bio->bi_private = &device->flush_wait;
3913	submit_bio(bio);
 
3914	set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3915}
3916
3917/*
3918 * If the flush bio has been submitted by write_dev_flush, wait for it.
3919 * Return true for any error, and false otherwise.
3920 */
3921static bool wait_dev_flush(struct btrfs_device *device)
3922{
3923	struct bio *bio = &device->flush_bio;
3924
3925	if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3926		return false;
3927
 
3928	wait_for_completion_io(&device->flush_wait);
3929
3930	if (bio->bi_status) {
3931		device->last_flush_error = bio->bi_status;
3932		btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_FLUSH_ERRS);
3933		return true;
3934	}
3935
3936	return false;
 
 
 
 
3937}
3938
3939/*
3940 * send an empty flush down to each device in parallel,
3941 * then wait for them
3942 */
3943static int barrier_all_devices(struct btrfs_fs_info *info)
3944{
3945	struct list_head *head;
3946	struct btrfs_device *dev;
3947	int errors_wait = 0;
 
3948
3949	lockdep_assert_held(&info->fs_devices->device_list_mutex);
3950	/* send down all the barriers */
3951	head = &info->fs_devices->devices;
3952	list_for_each_entry(dev, head, dev_list) {
3953		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3954			continue;
3955		if (!dev->bdev)
3956			continue;
3957		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3958		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3959			continue;
3960
3961		write_dev_flush(dev);
 
3962	}
3963
3964	/* wait for all the barriers */
3965	list_for_each_entry(dev, head, dev_list) {
3966		if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3967			continue;
3968		if (!dev->bdev) {
3969			errors_wait++;
3970			continue;
3971		}
3972		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3973		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3974			continue;
3975
3976		if (wait_dev_flush(dev))
 
 
 
 
3977			errors_wait++;
 
3978	}
3979
3980	/*
3981	 * Checks last_flush_error of disks in order to determine the device
3982	 * state.
3983	 */
3984	if (errors_wait && !btrfs_check_rw_degradable(info, NULL))
3985		return -EIO;
3986
 
3987	return 0;
3988}
3989
3990int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3991{
3992	int raid_type;
3993	int min_tolerated = INT_MAX;
3994
3995	if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3996	    (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3997		min_tolerated = min_t(int, min_tolerated,
3998				    btrfs_raid_array[BTRFS_RAID_SINGLE].
3999				    tolerated_failures);
4000
4001	for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4002		if (raid_type == BTRFS_RAID_SINGLE)
4003			continue;
4004		if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4005			continue;
4006		min_tolerated = min_t(int, min_tolerated,
4007				    btrfs_raid_array[raid_type].
4008				    tolerated_failures);
4009	}
4010
4011	if (min_tolerated == INT_MAX) {
4012		pr_warn("BTRFS: unknown raid flag: %llu", flags);
4013		min_tolerated = 0;
4014	}
4015
4016	return min_tolerated;
4017}
4018
4019int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4020{
4021	struct list_head *head;
4022	struct btrfs_device *dev;
4023	struct btrfs_super_block *sb;
4024	struct btrfs_dev_item *dev_item;
4025	int ret;
4026	int do_barriers;
4027	int max_errors;
4028	int total_errors = 0;
4029	u64 flags;
4030
4031	do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4032
4033	/*
4034	 * max_mirrors == 0 indicates we're from commit_transaction,
4035	 * not from fsync where the tree roots in fs_info have not
4036	 * been consistent on disk.
4037	 */
4038	if (max_mirrors == 0)
4039		backup_super_roots(fs_info);
4040
4041	sb = fs_info->super_for_commit;
4042	dev_item = &sb->dev_item;
4043
4044	mutex_lock(&fs_info->fs_devices->device_list_mutex);
4045	head = &fs_info->fs_devices->devices;
4046	max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4047
4048	if (do_barriers) {
4049		ret = barrier_all_devices(fs_info);
4050		if (ret) {
4051			mutex_unlock(
4052				&fs_info->fs_devices->device_list_mutex);
4053			btrfs_handle_fs_error(fs_info, ret,
4054					      "errors while submitting device barriers.");
4055			return ret;
4056		}
4057	}
4058
4059	list_for_each_entry(dev, head, dev_list) {
4060		if (!dev->bdev) {
4061			total_errors++;
4062			continue;
4063		}
4064		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4065		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4066			continue;
4067
4068		btrfs_set_stack_device_generation(dev_item, 0);
4069		btrfs_set_stack_device_type(dev_item, dev->type);
4070		btrfs_set_stack_device_id(dev_item, dev->devid);
4071		btrfs_set_stack_device_total_bytes(dev_item,
4072						   dev->commit_total_bytes);
4073		btrfs_set_stack_device_bytes_used(dev_item,
4074						  dev->commit_bytes_used);
4075		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4076		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4077		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4078		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4079		memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4080		       BTRFS_FSID_SIZE);
4081
4082		flags = btrfs_super_flags(sb);
4083		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4084
4085		ret = btrfs_validate_write_super(fs_info, sb);
4086		if (ret < 0) {
4087			mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4088			btrfs_handle_fs_error(fs_info, -EUCLEAN,
4089				"unexpected superblock corruption detected");
4090			return -EUCLEAN;
4091		}
4092
4093		ret = write_dev_supers(dev, sb, max_mirrors);
4094		if (ret)
4095			total_errors++;
4096	}
4097	if (total_errors > max_errors) {
4098		btrfs_err(fs_info, "%d errors while writing supers",
4099			  total_errors);
4100		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4101
4102		/* FUA is masked off if unsupported and can't be the reason */
4103		btrfs_handle_fs_error(fs_info, -EIO,
4104				      "%d errors while writing supers",
4105				      total_errors);
4106		return -EIO;
4107	}
4108
4109	total_errors = 0;
4110	list_for_each_entry(dev, head, dev_list) {
4111		if (!dev->bdev)
4112			continue;
4113		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4114		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4115			continue;
4116
4117		ret = wait_dev_supers(dev, max_mirrors);
4118		if (ret)
4119			total_errors++;
4120	}
4121	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4122	if (total_errors > max_errors) {
4123		btrfs_handle_fs_error(fs_info, -EIO,
4124				      "%d errors while writing supers",
4125				      total_errors);
4126		return -EIO;
4127	}
4128	return 0;
4129}
4130
4131/* Drop a fs root from the radix tree and free it. */
4132void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4133				  struct btrfs_root *root)
4134{
4135	bool drop_ref = false;
4136
4137	spin_lock(&fs_info->fs_roots_radix_lock);
4138	radix_tree_delete(&fs_info->fs_roots_radix,
4139			  (unsigned long)root->root_key.objectid);
4140	if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4141		drop_ref = true;
4142	spin_unlock(&fs_info->fs_roots_radix_lock);
4143
4144	if (BTRFS_FS_ERROR(fs_info)) {
4145		ASSERT(root->log_root == NULL);
 
 
 
4146		if (root->reloc_root) {
4147			btrfs_put_root(root->reloc_root);
 
 
4148			root->reloc_root = NULL;
4149		}
4150	}
4151
4152	if (drop_ref)
4153		btrfs_put_root(root);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4154}
4155
4156int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4157{
4158	struct btrfs_root *root = fs_info->tree_root;
4159	struct btrfs_trans_handle *trans;
4160
4161	mutex_lock(&fs_info->cleaner_mutex);
4162	btrfs_run_delayed_iputs(fs_info);
4163	mutex_unlock(&fs_info->cleaner_mutex);
4164	wake_up_process(fs_info->cleaner_kthread);
4165
4166	/* wait until ongoing cleanup work done */
4167	down_write(&fs_info->cleanup_work_sem);
4168	up_write(&fs_info->cleanup_work_sem);
4169
4170	trans = btrfs_join_transaction(root);
4171	if (IS_ERR(trans))
4172		return PTR_ERR(trans);
4173	return btrfs_commit_transaction(trans);
4174}
4175
4176static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4177{
4178	struct btrfs_transaction *trans;
4179	struct btrfs_transaction *tmp;
4180	bool found = false;
4181
4182	if (list_empty(&fs_info->trans_list))
4183		return;
4184
4185	/*
4186	 * This function is only called at the very end of close_ctree(),
4187	 * thus no other running transaction, no need to take trans_lock.
4188	 */
4189	ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4190	list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4191		struct extent_state *cached = NULL;
4192		u64 dirty_bytes = 0;
4193		u64 cur = 0;
4194		u64 found_start;
4195		u64 found_end;
4196
4197		found = true;
4198		while (find_first_extent_bit(&trans->dirty_pages, cur,
4199			&found_start, &found_end, EXTENT_DIRTY, &cached)) {
4200			dirty_bytes += found_end + 1 - found_start;
4201			cur = found_end + 1;
4202		}
4203		btrfs_warn(fs_info,
4204	"transaction %llu (with %llu dirty metadata bytes) is not committed",
4205			   trans->transid, dirty_bytes);
4206		btrfs_cleanup_one_transaction(trans, fs_info);
4207
4208		if (trans == fs_info->running_transaction)
4209			fs_info->running_transaction = NULL;
4210		list_del_init(&trans->list);
4211
4212		btrfs_put_transaction(trans);
4213		trace_btrfs_transaction_commit(fs_info);
4214	}
4215	ASSERT(!found);
4216}
4217
4218void __cold close_ctree(struct btrfs_fs_info *fs_info)
4219{
4220	int ret;
4221
4222	set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4223
4224	/*
4225	 * If we had UNFINISHED_DROPS we could still be processing them, so
4226	 * clear that bit and wake up relocation so it can stop.
4227	 * We must do this before stopping the block group reclaim task, because
4228	 * at btrfs_relocate_block_group() we wait for this bit, and after the
4229	 * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4230	 * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4231	 * return 1.
4232	 */
4233	btrfs_wake_unfinished_drop(fs_info);
4234
4235	/*
4236	 * We may have the reclaim task running and relocating a data block group,
4237	 * in which case it may create delayed iputs. So stop it before we park
4238	 * the cleaner kthread otherwise we can get new delayed iputs after
4239	 * parking the cleaner, and that can make the async reclaim task to hang
4240	 * if it's waiting for delayed iputs to complete, since the cleaner is
4241	 * parked and can not run delayed iputs - this will make us hang when
4242	 * trying to stop the async reclaim task.
4243	 */
4244	cancel_work_sync(&fs_info->reclaim_bgs_work);
4245	/*
4246	 * We don't want the cleaner to start new transactions, add more delayed
4247	 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4248	 * because that frees the task_struct, and the transaction kthread might
4249	 * still try to wake up the cleaner.
4250	 */
4251	kthread_park(fs_info->cleaner_kthread);
4252
4253	/* wait for the qgroup rescan worker to stop */
4254	btrfs_qgroup_wait_for_completion(fs_info, false);
4255
4256	/* wait for the uuid_scan task to finish */
4257	down(&fs_info->uuid_tree_rescan_sem);
4258	/* avoid complains from lockdep et al., set sem back to initial state */
4259	up(&fs_info->uuid_tree_rescan_sem);
4260
4261	/* pause restriper - we want to resume on mount */
4262	btrfs_pause_balance(fs_info);
4263
4264	btrfs_dev_replace_suspend_for_unmount(fs_info);
4265
4266	btrfs_scrub_cancel(fs_info);
4267
4268	/* wait for any defraggers to finish */
4269	wait_event(fs_info->transaction_wait,
4270		   (atomic_read(&fs_info->defrag_running) == 0));
4271
4272	/* clear out the rbtree of defraggable inodes */
4273	btrfs_cleanup_defrag_inodes(fs_info);
4274
4275	/*
4276	 * After we parked the cleaner kthread, ordered extents may have
4277	 * completed and created new delayed iputs. If one of the async reclaim
4278	 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4279	 * can hang forever trying to stop it, because if a delayed iput is
4280	 * added after it ran btrfs_run_delayed_iputs() and before it called
4281	 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4282	 * no one else to run iputs.
4283	 *
4284	 * So wait for all ongoing ordered extents to complete and then run
4285	 * delayed iputs. This works because once we reach this point no one
4286	 * can either create new ordered extents nor create delayed iputs
4287	 * through some other means.
4288	 *
4289	 * Also note that btrfs_wait_ordered_roots() is not safe here, because
4290	 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4291	 * but the delayed iput for the respective inode is made only when doing
4292	 * the final btrfs_put_ordered_extent() (which must happen at
4293	 * btrfs_finish_ordered_io() when we are unmounting).
4294	 */
4295	btrfs_flush_workqueue(fs_info->endio_write_workers);
4296	/* Ordered extents for free space inodes. */
4297	btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4298	btrfs_run_delayed_iputs(fs_info);
4299
4300	cancel_work_sync(&fs_info->async_reclaim_work);
4301	cancel_work_sync(&fs_info->async_data_reclaim_work);
4302	cancel_work_sync(&fs_info->preempt_reclaim_work);
4303
4304	/* Cancel or finish ongoing discard work */
4305	btrfs_discard_cleanup(fs_info);
4306
4307	if (!sb_rdonly(fs_info->sb)) {
4308		/*
4309		 * The cleaner kthread is stopped, so do one final pass over
4310		 * unused block groups.
4311		 */
4312		btrfs_delete_unused_bgs(fs_info);
4313
4314		/*
4315		 * There might be existing delayed inode workers still running
4316		 * and holding an empty delayed inode item. We must wait for
4317		 * them to complete first because they can create a transaction.
4318		 * This happens when someone calls btrfs_balance_delayed_items()
4319		 * and then a transaction commit runs the same delayed nodes
4320		 * before any delayed worker has done something with the nodes.
4321		 * We must wait for any worker here and not at transaction
4322		 * commit time since that could cause a deadlock.
4323		 * This is a very rare case.
4324		 */
4325		btrfs_flush_workqueue(fs_info->delayed_workers);
4326
4327		ret = btrfs_commit_super(fs_info);
4328		if (ret)
4329			btrfs_err(fs_info, "commit super ret %d", ret);
4330	}
4331
4332	if (BTRFS_FS_ERROR(fs_info))
 
4333		btrfs_error_commit_super(fs_info);
4334
4335	kthread_stop(fs_info->transaction_kthread);
4336	kthread_stop(fs_info->cleaner_kthread);
4337
4338	ASSERT(list_empty(&fs_info->delayed_iputs));
4339	set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4340
4341	if (btrfs_check_quota_leak(fs_info)) {
4342		WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4343		btrfs_err(fs_info, "qgroup reserved space leaked");
4344	}
4345
4346	btrfs_free_qgroup_config(fs_info);
4347	ASSERT(list_empty(&fs_info->delalloc_roots));
4348
4349	if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4350		btrfs_info(fs_info, "at unmount delalloc count %lld",
4351		       percpu_counter_sum(&fs_info->delalloc_bytes));
4352	}
4353
4354	if (percpu_counter_sum(&fs_info->ordered_bytes))
4355		btrfs_info(fs_info, "at unmount dio bytes count %lld",
4356			   percpu_counter_sum(&fs_info->ordered_bytes));
4357
4358	btrfs_sysfs_remove_mounted(fs_info);
4359	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4360
 
 
4361	btrfs_put_block_group_cache(fs_info);
4362
4363	/*
4364	 * we must make sure there is not any read request to
4365	 * submit after we stopping all workers.
4366	 */
4367	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4368	btrfs_stop_all_workers(fs_info);
4369
4370	/* We shouldn't have any transaction open at this point */
4371	warn_about_uncommitted_trans(fs_info);
4372
4373	clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4374	free_root_pointers(fs_info, true);
4375	btrfs_free_fs_roots(fs_info);
4376
4377	/*
4378	 * We must free the block groups after dropping the fs_roots as we could
4379	 * have had an IO error and have left over tree log blocks that aren't
4380	 * cleaned up until the fs roots are freed.  This makes the block group
4381	 * accounting appear to be wrong because there's pending reserved bytes,
4382	 * so make sure we do the block group cleanup afterwards.
4383	 */
4384	btrfs_free_block_groups(fs_info);
4385
4386	iput(fs_info->btree_inode);
 
 
 
4387
4388	btrfs_mapping_tree_free(fs_info);
4389	btrfs_close_devices(fs_info->fs_devices);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4390}
4391
4392void btrfs_mark_buffer_dirty(struct btrfs_trans_handle *trans,
4393			     struct extent_buffer *buf)
4394{
4395	struct btrfs_fs_info *fs_info = buf->fs_info;
 
4396	u64 transid = btrfs_header_generation(buf);
 
4397
4398#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4399	/*
4400	 * This is a fast path so only do this check if we have sanity tests
4401	 * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4402	 * outside of the sanity tests.
4403	 */
4404	if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4405		return;
4406#endif
4407	/* This is an active transaction (its state < TRANS_STATE_UNBLOCKED). */
4408	ASSERT(trans->transid == fs_info->generation);
4409	btrfs_assert_tree_write_locked(buf);
4410	if (unlikely(transid != fs_info->generation)) {
4411		btrfs_abort_transaction(trans, -EUCLEAN);
4412		btrfs_crit(fs_info,
4413"dirty buffer transid mismatch, logical %llu found transid %llu running transid %llu",
4414			   buf->start, transid, fs_info->generation);
 
 
 
 
 
 
 
 
 
 
 
 
 
4415	}
4416	set_extent_buffer_dirty(buf);
4417}
4418
4419static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4420					int flush_delayed)
4421{
4422	/*
4423	 * looks as though older kernels can get into trouble with
4424	 * this code, they end up stuck in balance_dirty_pages forever
4425	 */
4426	int ret;
4427
4428	if (current->flags & PF_MEMALLOC)
4429		return;
4430
4431	if (flush_delayed)
4432		btrfs_balance_delayed_items(fs_info);
4433
4434	ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4435				     BTRFS_DIRTY_METADATA_THRESH,
4436				     fs_info->dirty_metadata_batch);
4437	if (ret > 0) {
4438		balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4439	}
4440}
4441
4442void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4443{
4444	__btrfs_btree_balance_dirty(fs_info, 1);
4445}
4446
4447void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4448{
4449	__btrfs_btree_balance_dirty(fs_info, 0);
4450}
4451
 
 
 
 
 
 
 
4452static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4453{
4454	/* cleanup FS via transaction */
4455	btrfs_cleanup_transaction(fs_info);
4456
4457	mutex_lock(&fs_info->cleaner_mutex);
4458	btrfs_run_delayed_iputs(fs_info);
4459	mutex_unlock(&fs_info->cleaner_mutex);
4460
4461	down_write(&fs_info->cleanup_work_sem);
4462	up_write(&fs_info->cleanup_work_sem);
4463}
4464
4465static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4466{
4467	struct btrfs_root *gang[8];
4468	u64 root_objectid = 0;
4469	int ret;
4470
4471	spin_lock(&fs_info->fs_roots_radix_lock);
4472	while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4473					     (void **)gang, root_objectid,
4474					     ARRAY_SIZE(gang))) != 0) {
4475		int i;
4476
4477		for (i = 0; i < ret; i++)
4478			gang[i] = btrfs_grab_root(gang[i]);
4479		spin_unlock(&fs_info->fs_roots_radix_lock);
4480
4481		for (i = 0; i < ret; i++) {
4482			if (!gang[i])
4483				continue;
4484			root_objectid = gang[i]->root_key.objectid;
4485			btrfs_free_log(NULL, gang[i]);
4486			btrfs_put_root(gang[i]);
4487		}
4488		root_objectid++;
4489		spin_lock(&fs_info->fs_roots_radix_lock);
4490	}
4491	spin_unlock(&fs_info->fs_roots_radix_lock);
4492	btrfs_free_log_root_tree(NULL, fs_info);
4493}
4494
4495static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4496{
4497	struct btrfs_ordered_extent *ordered;
4498
4499	spin_lock(&root->ordered_extent_lock);
4500	/*
4501	 * This will just short circuit the ordered completion stuff which will
4502	 * make sure the ordered extent gets properly cleaned up.
4503	 */
4504	list_for_each_entry(ordered, &root->ordered_extents,
4505			    root_extent_list)
4506		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4507	spin_unlock(&root->ordered_extent_lock);
4508}
4509
4510static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4511{
4512	struct btrfs_root *root;
4513	LIST_HEAD(splice);
 
 
4514
4515	spin_lock(&fs_info->ordered_root_lock);
4516	list_splice_init(&fs_info->ordered_roots, &splice);
4517	while (!list_empty(&splice)) {
4518		root = list_first_entry(&splice, struct btrfs_root,
4519					ordered_root);
4520		list_move_tail(&root->ordered_root,
4521			       &fs_info->ordered_roots);
4522
4523		spin_unlock(&fs_info->ordered_root_lock);
4524		btrfs_destroy_ordered_extents(root);
4525
4526		cond_resched();
4527		spin_lock(&fs_info->ordered_root_lock);
4528	}
4529	spin_unlock(&fs_info->ordered_root_lock);
4530
4531	/*
4532	 * We need this here because if we've been flipped read-only we won't
4533	 * get sync() from the umount, so we need to make sure any ordered
4534	 * extents that haven't had their dirty pages IO start writeout yet
4535	 * actually get run and error out properly.
4536	 */
4537	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4538}
4539
4540static void btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4541				       struct btrfs_fs_info *fs_info)
4542{
4543	struct rb_node *node;
4544	struct btrfs_delayed_ref_root *delayed_refs;
4545	struct btrfs_delayed_ref_node *ref;
 
4546
4547	delayed_refs = &trans->delayed_refs;
4548
4549	spin_lock(&delayed_refs->lock);
4550	if (atomic_read(&delayed_refs->num_entries) == 0) {
4551		spin_unlock(&delayed_refs->lock);
4552		btrfs_debug(fs_info, "delayed_refs has NO entry");
4553		return;
4554	}
4555
4556	while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4557		struct btrfs_delayed_ref_head *head;
4558		struct rb_node *n;
4559		bool pin_bytes = false;
4560
4561		head = rb_entry(node, struct btrfs_delayed_ref_head,
4562				href_node);
4563		if (btrfs_delayed_ref_lock(delayed_refs, head))
4564			continue;
4565
4566		spin_lock(&head->lock);
4567		while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4568			ref = rb_entry(n, struct btrfs_delayed_ref_node,
4569				       ref_node);
 
4570			rb_erase_cached(&ref->ref_node, &head->ref_tree);
4571			RB_CLEAR_NODE(&ref->ref_node);
4572			if (!list_empty(&ref->add_list))
4573				list_del(&ref->add_list);
4574			atomic_dec(&delayed_refs->num_entries);
4575			btrfs_put_delayed_ref(ref);
4576			btrfs_delayed_refs_rsv_release(fs_info, 1, 0);
4577		}
4578		if (head->must_insert_reserved)
4579			pin_bytes = true;
4580		btrfs_free_delayed_extent_op(head->extent_op);
4581		btrfs_delete_ref_head(delayed_refs, head);
4582		spin_unlock(&head->lock);
4583		spin_unlock(&delayed_refs->lock);
4584		mutex_unlock(&head->mutex);
4585
4586		if (pin_bytes) {
4587			struct btrfs_block_group *cache;
4588
4589			cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4590			BUG_ON(!cache);
4591
4592			spin_lock(&cache->space_info->lock);
4593			spin_lock(&cache->lock);
4594			cache->pinned += head->num_bytes;
4595			btrfs_space_info_update_bytes_pinned(fs_info,
4596				cache->space_info, head->num_bytes);
4597			cache->reserved -= head->num_bytes;
4598			cache->space_info->bytes_reserved -= head->num_bytes;
4599			spin_unlock(&cache->lock);
4600			spin_unlock(&cache->space_info->lock);
4601
4602			btrfs_put_block_group(cache);
4603
4604			btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4605				head->bytenr + head->num_bytes - 1);
4606		}
4607		btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4608		btrfs_put_delayed_ref_head(head);
4609		cond_resched();
4610		spin_lock(&delayed_refs->lock);
4611	}
4612	btrfs_qgroup_destroy_extent_records(trans);
4613
4614	spin_unlock(&delayed_refs->lock);
 
 
4615}
4616
4617static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4618{
4619	struct btrfs_inode *btrfs_inode;
4620	LIST_HEAD(splice);
 
 
4621
4622	spin_lock(&root->delalloc_lock);
4623	list_splice_init(&root->delalloc_inodes, &splice);
4624
4625	while (!list_empty(&splice)) {
4626		struct inode *inode = NULL;
4627		btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4628					       delalloc_inodes);
4629		__btrfs_del_delalloc_inode(root, btrfs_inode);
4630		spin_unlock(&root->delalloc_lock);
4631
4632		/*
4633		 * Make sure we get a live inode and that it'll not disappear
4634		 * meanwhile.
4635		 */
4636		inode = igrab(&btrfs_inode->vfs_inode);
4637		if (inode) {
4638			unsigned int nofs_flag;
4639
4640			nofs_flag = memalloc_nofs_save();
4641			invalidate_inode_pages2(inode->i_mapping);
4642			memalloc_nofs_restore(nofs_flag);
4643			iput(inode);
4644		}
4645		spin_lock(&root->delalloc_lock);
4646	}
4647	spin_unlock(&root->delalloc_lock);
4648}
4649
4650static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4651{
4652	struct btrfs_root *root;
4653	LIST_HEAD(splice);
 
 
4654
4655	spin_lock(&fs_info->delalloc_root_lock);
4656	list_splice_init(&fs_info->delalloc_roots, &splice);
4657	while (!list_empty(&splice)) {
4658		root = list_first_entry(&splice, struct btrfs_root,
4659					 delalloc_root);
4660		root = btrfs_grab_root(root);
4661		BUG_ON(!root);
4662		spin_unlock(&fs_info->delalloc_root_lock);
4663
4664		btrfs_destroy_delalloc_inodes(root);
4665		btrfs_put_root(root);
4666
4667		spin_lock(&fs_info->delalloc_root_lock);
4668	}
4669	spin_unlock(&fs_info->delalloc_root_lock);
4670}
4671
4672static void btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4673					 struct extent_io_tree *dirty_pages,
4674					 int mark)
4675{
 
4676	struct extent_buffer *eb;
4677	u64 start = 0;
4678	u64 end;
4679
4680	while (find_first_extent_bit(dirty_pages, start, &start, &end,
4681				     mark, NULL)) {
 
 
 
 
4682		clear_extent_bits(dirty_pages, start, end, mark);
4683		while (start <= end) {
4684			eb = find_extent_buffer(fs_info, start);
4685			start += fs_info->nodesize;
4686			if (!eb)
4687				continue;
4688
4689			btrfs_tree_lock(eb);
4690			wait_on_extent_buffer_writeback(eb);
4691			btrfs_clear_buffer_dirty(NULL, eb);
4692			btrfs_tree_unlock(eb);
4693
 
 
 
4694			free_extent_buffer_stale(eb);
4695		}
4696	}
 
 
4697}
4698
4699static void btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4700					struct extent_io_tree *unpin)
4701{
 
4702	u64 start;
4703	u64 end;
 
 
4704
 
 
4705	while (1) {
4706		struct extent_state *cached_state = NULL;
4707
4708		/*
4709		 * The btrfs_finish_extent_commit() may get the same range as
4710		 * ours between find_first_extent_bit and clear_extent_dirty.
4711		 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4712		 * the same extent range.
4713		 */
4714		mutex_lock(&fs_info->unused_bg_unpin_mutex);
4715		if (!find_first_extent_bit(unpin, 0, &start, &end,
4716					   EXTENT_DIRTY, &cached_state)) {
 
4717			mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4718			break;
4719		}
4720
4721		clear_extent_dirty(unpin, start, end, &cached_state);
4722		free_extent_state(cached_state);
4723		btrfs_error_unpin_extent_range(fs_info, start, end);
4724		mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4725		cond_resched();
4726	}
 
 
 
 
 
 
 
 
 
 
 
4727}
4728
4729static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4730{
4731	struct inode *inode;
4732
4733	inode = cache->io_ctl.inode;
4734	if (inode) {
4735		unsigned int nofs_flag;
4736
4737		nofs_flag = memalloc_nofs_save();
4738		invalidate_inode_pages2(inode->i_mapping);
4739		memalloc_nofs_restore(nofs_flag);
4740
4741		BTRFS_I(inode)->generation = 0;
4742		cache->io_ctl.inode = NULL;
4743		iput(inode);
4744	}
4745	ASSERT(cache->io_ctl.pages == NULL);
4746	btrfs_put_block_group(cache);
4747}
4748
4749void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4750			     struct btrfs_fs_info *fs_info)
4751{
4752	struct btrfs_block_group *cache;
4753
4754	spin_lock(&cur_trans->dirty_bgs_lock);
4755	while (!list_empty(&cur_trans->dirty_bgs)) {
4756		cache = list_first_entry(&cur_trans->dirty_bgs,
4757					 struct btrfs_block_group,
4758					 dirty_list);
4759
4760		if (!list_empty(&cache->io_list)) {
4761			spin_unlock(&cur_trans->dirty_bgs_lock);
4762			list_del_init(&cache->io_list);
4763			btrfs_cleanup_bg_io(cache);
4764			spin_lock(&cur_trans->dirty_bgs_lock);
4765		}
4766
4767		list_del_init(&cache->dirty_list);
4768		spin_lock(&cache->lock);
4769		cache->disk_cache_state = BTRFS_DC_ERROR;
4770		spin_unlock(&cache->lock);
4771
4772		spin_unlock(&cur_trans->dirty_bgs_lock);
4773		btrfs_put_block_group(cache);
4774		btrfs_dec_delayed_refs_rsv_bg_updates(fs_info);
4775		spin_lock(&cur_trans->dirty_bgs_lock);
4776	}
4777	spin_unlock(&cur_trans->dirty_bgs_lock);
4778
4779	/*
4780	 * Refer to the definition of io_bgs member for details why it's safe
4781	 * to use it without any locking
4782	 */
4783	while (!list_empty(&cur_trans->io_bgs)) {
4784		cache = list_first_entry(&cur_trans->io_bgs,
4785					 struct btrfs_block_group,
4786					 io_list);
4787
4788		list_del_init(&cache->io_list);
4789		spin_lock(&cache->lock);
4790		cache->disk_cache_state = BTRFS_DC_ERROR;
4791		spin_unlock(&cache->lock);
4792		btrfs_cleanup_bg_io(cache);
4793	}
4794}
4795
4796static void btrfs_free_all_qgroup_pertrans(struct btrfs_fs_info *fs_info)
4797{
4798	struct btrfs_root *gang[8];
4799	int i;
4800	int ret;
4801
4802	spin_lock(&fs_info->fs_roots_radix_lock);
4803	while (1) {
4804		ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
4805						 (void **)gang, 0,
4806						 ARRAY_SIZE(gang),
4807						 BTRFS_ROOT_TRANS_TAG);
4808		if (ret == 0)
4809			break;
4810		for (i = 0; i < ret; i++) {
4811			struct btrfs_root *root = gang[i];
4812
4813			btrfs_qgroup_free_meta_all_pertrans(root);
4814			radix_tree_tag_clear(&fs_info->fs_roots_radix,
4815					(unsigned long)root->root_key.objectid,
4816					BTRFS_ROOT_TRANS_TAG);
4817		}
4818	}
4819	spin_unlock(&fs_info->fs_roots_radix_lock);
4820}
4821
4822void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4823				   struct btrfs_fs_info *fs_info)
4824{
4825	struct btrfs_device *dev, *tmp;
4826
4827	btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4828	ASSERT(list_empty(&cur_trans->dirty_bgs));
4829	ASSERT(list_empty(&cur_trans->io_bgs));
4830
4831	list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4832				 post_commit_list) {
4833		list_del_init(&dev->post_commit_list);
4834	}
4835
4836	btrfs_destroy_delayed_refs(cur_trans, fs_info);
4837
4838	cur_trans->state = TRANS_STATE_COMMIT_START;
4839	wake_up(&fs_info->transaction_blocked_wait);
4840
4841	cur_trans->state = TRANS_STATE_UNBLOCKED;
4842	wake_up(&fs_info->transaction_wait);
4843
4844	btrfs_destroy_delayed_inodes(fs_info);
 
4845
4846	btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4847				     EXTENT_DIRTY);
4848	btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4849
4850	btrfs_free_all_qgroup_pertrans(fs_info);
4851
4852	cur_trans->state =TRANS_STATE_COMPLETED;
4853	wake_up(&cur_trans->commit_wait);
4854}
4855
4856static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4857{
4858	struct btrfs_transaction *t;
4859
4860	mutex_lock(&fs_info->transaction_kthread_mutex);
4861
4862	spin_lock(&fs_info->trans_lock);
4863	while (!list_empty(&fs_info->trans_list)) {
4864		t = list_first_entry(&fs_info->trans_list,
4865				     struct btrfs_transaction, list);
4866		if (t->state >= TRANS_STATE_COMMIT_PREP) {
4867			refcount_inc(&t->use_count);
4868			spin_unlock(&fs_info->trans_lock);
4869			btrfs_wait_for_commit(fs_info, t->transid);
4870			btrfs_put_transaction(t);
4871			spin_lock(&fs_info->trans_lock);
4872			continue;
4873		}
4874		if (t == fs_info->running_transaction) {
4875			t->state = TRANS_STATE_COMMIT_DOING;
4876			spin_unlock(&fs_info->trans_lock);
4877			/*
4878			 * We wait for 0 num_writers since we don't hold a trans
4879			 * handle open currently for this transaction.
4880			 */
4881			wait_event(t->writer_wait,
4882				   atomic_read(&t->num_writers) == 0);
4883		} else {
4884			spin_unlock(&fs_info->trans_lock);
4885		}
4886		btrfs_cleanup_one_transaction(t, fs_info);
4887
4888		spin_lock(&fs_info->trans_lock);
4889		if (t == fs_info->running_transaction)
4890			fs_info->running_transaction = NULL;
4891		list_del_init(&t->list);
4892		spin_unlock(&fs_info->trans_lock);
4893
4894		btrfs_put_transaction(t);
4895		trace_btrfs_transaction_commit(fs_info);
4896		spin_lock(&fs_info->trans_lock);
4897	}
4898	spin_unlock(&fs_info->trans_lock);
4899	btrfs_destroy_all_ordered_extents(fs_info);
4900	btrfs_destroy_delayed_inodes(fs_info);
4901	btrfs_assert_delayed_root_empty(fs_info);
 
4902	btrfs_destroy_all_delalloc_inodes(fs_info);
4903	btrfs_drop_all_logs(fs_info);
4904	mutex_unlock(&fs_info->transaction_kthread_mutex);
4905
4906	return 0;
4907}
4908
4909int btrfs_init_root_free_objectid(struct btrfs_root *root)
4910{
4911	struct btrfs_path *path;
4912	int ret;
4913	struct extent_buffer *l;
4914	struct btrfs_key search_key;
4915	struct btrfs_key found_key;
4916	int slot;
4917
4918	path = btrfs_alloc_path();
4919	if (!path)
4920		return -ENOMEM;
4921
4922	search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4923	search_key.type = -1;
4924	search_key.offset = (u64)-1;
4925	ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4926	if (ret < 0)
4927		goto error;
4928	BUG_ON(ret == 0); /* Corruption */
4929	if (path->slots[0] > 0) {
4930		slot = path->slots[0] - 1;
4931		l = path->nodes[0];
4932		btrfs_item_key_to_cpu(l, &found_key, slot);
4933		root->free_objectid = max_t(u64, found_key.objectid + 1,
4934					    BTRFS_FIRST_FREE_OBJECTID);
4935	} else {
4936		root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4937	}
4938	ret = 0;
4939error:
4940	btrfs_free_path(path);
4941	return ret;
4942}
4943
4944int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4945{
4946	int ret;
4947	mutex_lock(&root->objectid_mutex);
4948
4949	if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4950		btrfs_warn(root->fs_info,
4951			   "the objectid of root %llu reaches its highest value",
4952			   root->root_key.objectid);
4953		ret = -ENOSPC;
4954		goto out;
4955	}
4956
4957	*objectid = root->free_objectid++;
4958	ret = 0;
4959out:
4960	mutex_unlock(&root->objectid_mutex);
4961	return ret;
4962}