Linux Audio

Check our new training course

Loading...
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}
v4.10.11
 
   1/*
   2 * Copyright (C) 2007 Oracle.  All rights reserved.
   3 *
   4 * This program is free software; you can redistribute it and/or
   5 * modify it under the terms of the GNU General Public
   6 * License v2 as published by the Free Software Foundation.
   7 *
   8 * This program is distributed in the hope that it will be useful,
   9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11 * General Public License for more details.
  12 *
  13 * You should have received a copy of the GNU General Public
  14 * License along with this program; if not, write to the
  15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16 * Boston, MA 021110-1307, USA.
  17 */
  18
  19#include <linux/fs.h>
  20#include <linux/blkdev.h>
  21#include <linux/scatterlist.h>
  22#include <linux/swap.h>
  23#include <linux/radix-tree.h>
  24#include <linux/writeback.h>
  25#include <linux/buffer_head.h>
  26#include <linux/workqueue.h>
  27#include <linux/kthread.h>
  28#include <linux/slab.h>
  29#include <linux/migrate.h>
  30#include <linux/ratelimit.h>
  31#include <linux/uuid.h>
  32#include <linux/semaphore.h>
 
 
 
  33#include <asm/unaligned.h>
 
  34#include "ctree.h"
  35#include "disk-io.h"
  36#include "hash.h"
  37#include "transaction.h"
  38#include "btrfs_inode.h"
  39#include "volumes.h"
  40#include "print-tree.h"
  41#include "locking.h"
  42#include "tree-log.h"
  43#include "free-space-cache.h"
  44#include "free-space-tree.h"
  45#include "inode-map.h"
  46#include "check-integrity.h"
  47#include "rcu-string.h"
  48#include "dev-replace.h"
  49#include "raid56.h"
  50#include "sysfs.h"
  51#include "qgroup.h"
  52#include "compression.h"
  53
  54#ifdef CONFIG_X86
  55#include <asm/cpufeature.h>
  56#endif
 
 
 
 
 
 
 
 
 
 
 
 
  57
  58#define BTRFS_SUPER_FLAG_SUPP	(BTRFS_HEADER_FLAG_WRITTEN |\
  59				 BTRFS_HEADER_FLAG_RELOC |\
  60				 BTRFS_SUPER_FLAG_ERROR |\
  61				 BTRFS_SUPER_FLAG_SEEDING |\
  62				 BTRFS_SUPER_FLAG_METADUMP)
 
  63
  64static const struct extent_io_ops btree_extent_io_ops;
  65static void end_workqueue_fn(struct btrfs_work *work);
  66static void free_fs_root(struct btrfs_root *root);
  67static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  68				    int read_only);
  69static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  70static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  71				      struct btrfs_fs_info *fs_info);
  72static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  73static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
  74					struct extent_io_tree *dirty_pages,
  75					int mark);
  76static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
  77				       struct extent_io_tree *pinned_extents);
  78static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
  79static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
  80
  81/*
  82 * btrfs_end_io_wq structs are used to do processing in task context when an IO
  83 * is complete.  This is used during reads to verify checksums, and it is used
  84 * by writes to insert metadata for new file extents after IO is complete.
  85 */
  86struct btrfs_end_io_wq {
  87	struct bio *bio;
  88	bio_end_io_t *end_io;
  89	void *private;
  90	struct btrfs_fs_info *info;
  91	int error;
  92	enum btrfs_wq_endio_type metadata;
  93	struct list_head list;
  94	struct btrfs_work work;
  95};
  96
  97static struct kmem_cache *btrfs_end_io_wq_cache;
  98
  99int __init btrfs_end_io_wq_init(void)
 100{
 101	btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
 102					sizeof(struct btrfs_end_io_wq),
 103					0,
 104					SLAB_MEM_SPREAD,
 105					NULL);
 106	if (!btrfs_end_io_wq_cache)
 107		return -ENOMEM;
 108	return 0;
 109}
 110
 111void btrfs_end_io_wq_exit(void)
 112{
 113	kmem_cache_destroy(btrfs_end_io_wq_cache);
 
 114}
 115
 116/*
 117 * async submit bios are used to offload expensive checksumming
 118 * onto the worker threads.  They checksum file and metadata bios
 119 * just before they are sent down the IO stack.
 120 */
 121struct async_submit_bio {
 122	struct inode *inode;
 123	struct bio *bio;
 124	struct list_head list;
 125	extent_submit_bio_hook_t *submit_bio_start;
 126	extent_submit_bio_hook_t *submit_bio_done;
 127	int mirror_num;
 128	unsigned long bio_flags;
 129	/*
 130	 * bio_offset is optional, can be used if the pages in the bio
 131	 * can't tell us where in the file the bio should go
 132	 */
 133	u64 bio_offset;
 134	struct btrfs_work work;
 135	int error;
 136};
 137
 138/*
 139 * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 140 * eb, the lockdep key is determined by the btrfs_root it belongs to and
 141 * the level the eb occupies in the tree.
 142 *
 143 * Different roots are used for different purposes and may nest inside each
 144 * other and they require separate keysets.  As lockdep keys should be
 145 * static, assign keysets according to the purpose of the root as indicated
 146 * by btrfs_root->objectid.  This ensures that all special purpose roots
 147 * have separate keysets.
 148 *
 149 * Lock-nesting across peer nodes is always done with the immediate parent
 150 * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 151 * subclass to avoid triggering lockdep warning in such cases.
 152 *
 153 * The key is set by the readpage_end_io_hook after the buffer has passed
 154 * csum validation but before the pages are unlocked.  It is also set by
 155 * btrfs_init_new_buffer on freshly allocated blocks.
 156 *
 157 * We also add a check to make sure the highest level of the tree is the
 158 * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 159 * needs update as well.
 160 */
 161#ifdef CONFIG_DEBUG_LOCK_ALLOC
 162# if BTRFS_MAX_LEVEL != 8
 163#  error
 164# endif
 165
 166static struct btrfs_lockdep_keyset {
 167	u64			id;		/* root objectid */
 168	const char		*name_stem;	/* lock name stem */
 169	char			names[BTRFS_MAX_LEVEL + 1][20];
 170	struct lock_class_key	keys[BTRFS_MAX_LEVEL + 1];
 171} btrfs_lockdep_keysets[] = {
 172	{ .id = BTRFS_ROOT_TREE_OBJECTID,	.name_stem = "root"	},
 173	{ .id = BTRFS_EXTENT_TREE_OBJECTID,	.name_stem = "extent"	},
 174	{ .id = BTRFS_CHUNK_TREE_OBJECTID,	.name_stem = "chunk"	},
 175	{ .id = BTRFS_DEV_TREE_OBJECTID,	.name_stem = "dev"	},
 176	{ .id = BTRFS_FS_TREE_OBJECTID,		.name_stem = "fs"	},
 177	{ .id = BTRFS_CSUM_TREE_OBJECTID,	.name_stem = "csum"	},
 178	{ .id = BTRFS_QUOTA_TREE_OBJECTID,	.name_stem = "quota"	},
 179	{ .id = BTRFS_TREE_LOG_OBJECTID,	.name_stem = "log"	},
 180	{ .id = BTRFS_TREE_RELOC_OBJECTID,	.name_stem = "treloc"	},
 181	{ .id = BTRFS_DATA_RELOC_TREE_OBJECTID,	.name_stem = "dreloc"	},
 182	{ .id = BTRFS_UUID_TREE_OBJECTID,	.name_stem = "uuid"	},
 183	{ .id = BTRFS_FREE_SPACE_TREE_OBJECTID,	.name_stem = "free-space" },
 184	{ .id = 0,				.name_stem = "tree"	},
 185};
 186
 187void __init btrfs_init_lockdep(void)
 188{
 189	int i, j;
 190
 191	/* initialize lockdep class names */
 192	for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 193		struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 
 194
 195		for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 196			snprintf(ks->names[j], sizeof(ks->names[j]),
 197				 "btrfs-%s-%02d", ks->name_stem, j);
 198	}
 199}
 200
 201void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 202				    int level)
 203{
 204	struct btrfs_lockdep_keyset *ks;
 205
 206	BUG_ON(level >= ARRAY_SIZE(ks->keys));
 207
 208	/* find the matching keyset, id 0 is the default entry */
 209	for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 210		if (ks->id == objectid)
 211			break;
 212
 213	lockdep_set_class_and_name(&eb->lock,
 214				   &ks->keys[level], ks->names[level]);
 215}
 216
 217#endif
 218
 219/*
 220 * extents on the btree inode are pretty simple, there's one extent
 221 * that covers the entire device
 222 */
 223static struct extent_map *btree_get_extent(struct inode *inode,
 224		struct page *page, size_t pg_offset, u64 start, u64 len,
 225		int create)
 226{
 227	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 228	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 229	struct extent_map *em;
 230	int ret;
 231
 232	read_lock(&em_tree->lock);
 233	em = lookup_extent_mapping(em_tree, start, len);
 234	if (em) {
 235		em->bdev = fs_info->fs_devices->latest_bdev;
 236		read_unlock(&em_tree->lock);
 237		goto out;
 238	}
 239	read_unlock(&em_tree->lock);
 240
 241	em = alloc_extent_map();
 242	if (!em) {
 243		em = ERR_PTR(-ENOMEM);
 244		goto out;
 245	}
 246	em->start = 0;
 247	em->len = (u64)-1;
 248	em->block_len = (u64)-1;
 249	em->block_start = 0;
 250	em->bdev = fs_info->fs_devices->latest_bdev;
 251
 252	write_lock(&em_tree->lock);
 253	ret = add_extent_mapping(em_tree, em, 0);
 254	if (ret == -EEXIST) {
 255		free_extent_map(em);
 256		em = lookup_extent_mapping(em_tree, start, len);
 257		if (!em)
 258			em = ERR_PTR(-EIO);
 259	} else if (ret) {
 260		free_extent_map(em);
 261		em = ERR_PTR(ret);
 262	}
 263	write_unlock(&em_tree->lock);
 264
 265out:
 266	return em;
 267}
 268
 269u32 btrfs_csum_data(char *data, u32 seed, size_t len)
 270{
 271	return btrfs_crc32c(seed, data, len);
 272}
 273
 274void btrfs_csum_final(u32 crc, u8 *result)
 275{
 276	put_unaligned_le32(~crc, result);
 277}
 278
 279/*
 280 * compute the csum for a btree block, and either verify it or write it
 281 * into the csum field of the block.
 282 */
 283static int csum_tree_block(struct btrfs_fs_info *fs_info,
 284			   struct extent_buffer *buf,
 285			   int verify)
 286{
 287	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 288	char *result = NULL;
 289	unsigned long len;
 290	unsigned long cur_len;
 291	unsigned long offset = BTRFS_CSUM_SIZE;
 292	char *kaddr;
 293	unsigned long map_start;
 294	unsigned long map_len;
 295	int err;
 296	u32 crc = ~(u32)0;
 297	unsigned long inline_result;
 298
 299	len = buf->len - offset;
 300	while (len > 0) {
 301		err = map_private_extent_buffer(buf, offset, 32,
 302					&kaddr, &map_start, &map_len);
 303		if (err)
 304			return err;
 305		cur_len = min(len, map_len - (offset - map_start));
 306		crc = btrfs_csum_data(kaddr + offset - map_start,
 307				      crc, cur_len);
 308		len -= cur_len;
 309		offset += cur_len;
 310	}
 311	if (csum_size > sizeof(inline_result)) {
 312		result = kzalloc(csum_size, GFP_NOFS);
 313		if (!result)
 314			return -ENOMEM;
 315	} else {
 316		result = (char *)&inline_result;
 
 
 317	}
 318
 319	btrfs_csum_final(crc, result);
 
 320
 321	if (verify) {
 322		if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 323			u32 val;
 324			u32 found = 0;
 325			memcpy(&found, result, csum_size);
 326
 327			read_extent_buffer(buf, &val, 0, csum_size);
 328			btrfs_warn_rl(fs_info,
 329				"%s checksum verify failed on %llu wanted %X found %X level %d",
 330				fs_info->sb->s_id, buf->start,
 331				val, found, btrfs_header_level(buf));
 332			if (result != (char *)&inline_result)
 333				kfree(result);
 334			return -EUCLEAN;
 335		}
 336	} else {
 337		write_extent_buffer(buf, result, 0, csum_size);
 338	}
 339	if (result != (char *)&inline_result)
 340		kfree(result);
 341	return 0;
 342}
 343
 344/*
 345 * we can't consider a given block up to date unless the transid of the
 346 * block matches the transid in the parent node's pointer.  This is how we
 347 * detect blocks that either didn't get written at all or got written
 348 * in the wrong place.
 349 */
 350static int verify_parent_transid(struct extent_io_tree *io_tree,
 351				 struct extent_buffer *eb, u64 parent_transid,
 352				 int atomic)
 353{
 354	struct extent_state *cached_state = NULL;
 355	int ret;
 356	bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 357
 358	if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 359		return 0;
 360
 361	if (atomic)
 362		return -EAGAIN;
 363
 364	if (need_lock) {
 365		btrfs_tree_read_lock(eb);
 366		btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 
 
 
 
 
 367	}
 
 
 368
 369	lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 370			 &cached_state);
 371	if (extent_buffer_uptodate(eb) &&
 372	    btrfs_header_generation(eb) == parent_transid) {
 373		ret = 0;
 374		goto out;
 
 
 
 
 375	}
 376	btrfs_err_rl(eb->fs_info,
 377		"parent transid verify failed on %llu wanted %llu found %llu",
 378			eb->start,
 379			parent_transid, btrfs_header_generation(eb));
 380	ret = 1;
 381
 382	/*
 383	 * Things reading via commit roots that don't have normal protection,
 384	 * like send, can have a really old block in cache that may point at a
 385	 * block that has been freed and re-allocated.  So don't clear uptodate
 386	 * if we find an eb that is under IO (dirty/writeback) because we could
 387	 * end up reading in the stale data and then writing it back out and
 388	 * making everybody very sad.
 389	 */
 390	if (!extent_buffer_under_io(eb))
 391		clear_extent_buffer_uptodate(eb);
 392out:
 393	unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 394			     &cached_state, GFP_NOFS);
 395	if (need_lock)
 396		btrfs_tree_read_unlock_blocking(eb);
 397	return ret;
 398}
 399
 400/*
 401 * Return 0 if the superblock checksum type matches the checksum value of that
 402 * algorithm. Pass the raw disk superblock data.
 403 */
 404static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 405				  char *raw_disk_sb)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 406{
 407	struct btrfs_super_block *disk_sb =
 408		(struct btrfs_super_block *)raw_disk_sb;
 409	u16 csum_type = btrfs_super_csum_type(disk_sb);
 410	int ret = 0;
 411
 412	if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 413		u32 crc = ~(u32)0;
 414		const int csum_size = sizeof(crc);
 415		char result[csum_size];
 416
 417		/*
 418		 * The super_block structure does not span the whole
 419		 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 420		 * is filled with zeros and is included in the checksum.
 421		 */
 422		crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 423				crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 424		btrfs_csum_final(crc, result);
 425
 426		if (memcmp(raw_disk_sb, result, csum_size))
 427			ret = 1;
 428	}
 429
 430	if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 431		btrfs_err(fs_info, "unsupported checksum algorithm %u",
 432				csum_type);
 433		ret = 1;
 434	}
 435
 436	return ret;
 437}
 438
 439/*
 440 * helper to read a given tree block, doing retries as required when
 441 * the checksums don't match and we have alternate mirrors to try.
 
 
 
 442 */
 443static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
 444					  struct extent_buffer *eb,
 445					  u64 parent_transid)
 446{
 447	struct extent_io_tree *io_tree;
 448	int failed = 0;
 449	int ret;
 450	int num_copies = 0;
 451	int mirror_num = 0;
 452	int failed_mirror = 0;
 453
 454	clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 455	io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
 456	while (1) {
 457		ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
 458					       btree_get_extent, mirror_num);
 459		if (!ret) {
 460			if (!verify_parent_transid(io_tree, eb,
 461						   parent_transid, 0))
 462				break;
 463			else
 464				ret = -EIO;
 465		}
 466
 467		/*
 468		 * This buffer's crc is fine, but its contents are corrupted, so
 469		 * there is no reason to read the other copies, they won't be
 470		 * any less wrong.
 471		 */
 472		if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 473			break;
 474
 475		num_copies = btrfs_num_copies(fs_info,
 476					      eb->start, eb->len);
 477		if (num_copies == 1)
 478			break;
 479
 480		if (!failed_mirror) {
 481			failed = 1;
 482			failed_mirror = eb->read_mirror;
 483		}
 484
 485		mirror_num++;
 486		if (mirror_num == failed_mirror)
 487			mirror_num++;
 488
 489		if (mirror_num > num_copies)
 490			break;
 491	}
 492
 493	if (failed && !ret && failed_mirror)
 494		repair_eb_io_failure(fs_info, eb, failed_mirror);
 495
 496	return ret;
 497}
 498
 499/*
 500 * checksum a dirty tree block before IO.  This has extra checks to make sure
 501 * we only fill in the checksum field in the first page of a multi-page block
 502 */
 503
 504static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 505{
 506	u64 start = page_offset(page);
 507	u64 found_start;
 508	struct extent_buffer *eb;
 
 
 
 509
 510	eb = (struct extent_buffer *)page->private;
 511	if (page != eb->pages[0])
 512		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 513
 514	found_start = btrfs_header_bytenr(eb);
 515	/*
 516	 * Please do not consolidate these warnings into a single if.
 517	 * It is useful to know what went wrong.
 518	 */
 519	if (WARN_ON(found_start != start))
 520		return -EUCLEAN;
 521	if (WARN_ON(!PageUptodate(page)))
 522		return -EUCLEAN;
 523
 524	ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
 525			btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
 526
 527	return csum_tree_block(fs_info, eb, 0);
 528}
 529
 530static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
 531				 struct extent_buffer *eb)
 532{
 533	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 534	u8 fsid[BTRFS_UUID_SIZE];
 535	int ret = 1;
 536
 537	read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 538	while (fs_devices) {
 539		if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 540			ret = 0;
 541			break;
 542		}
 543		fs_devices = fs_devices->seed;
 544	}
 545	return ret;
 546}
 547
 548#define CORRUPT(reason, eb, root, slot)					\
 549	btrfs_crit(root->fs_info,					\
 550		   "corrupt %s, %s: block=%llu, root=%llu, slot=%d",	\
 551		   btrfs_header_level(eb) == 0 ? "leaf" : "node",	\
 552		   reason, btrfs_header_bytenr(eb), root->objectid, slot)
 553
 554static noinline int check_leaf(struct btrfs_root *root,
 555			       struct extent_buffer *leaf)
 556{
 557	struct btrfs_fs_info *fs_info = root->fs_info;
 558	struct btrfs_key key;
 559	struct btrfs_key leaf_key;
 560	u32 nritems = btrfs_header_nritems(leaf);
 561	int slot;
 562
 563	/*
 564	 * Extent buffers from a relocation tree have a owner field that
 565	 * corresponds to the subvolume tree they are based on. So just from an
 566	 * extent buffer alone we can not find out what is the id of the
 567	 * corresponding subvolume tree, so we can not figure out if the extent
 568	 * buffer corresponds to the root of the relocation tree or not. So skip
 569	 * this check for relocation trees.
 570	 */
 571	if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) {
 572		struct btrfs_root *check_root;
 573
 574		key.objectid = btrfs_header_owner(leaf);
 575		key.type = BTRFS_ROOT_ITEM_KEY;
 576		key.offset = (u64)-1;
 577
 578		check_root = btrfs_get_fs_root(fs_info, &key, false);
 579		/*
 580		 * The only reason we also check NULL here is that during
 581		 * open_ctree() some roots has not yet been set up.
 582		 */
 583		if (!IS_ERR_OR_NULL(check_root)) {
 584			struct extent_buffer *eb;
 585
 586			eb = btrfs_root_node(check_root);
 587			/* if leaf is the root, then it's fine */
 588			if (leaf != eb) {
 589				CORRUPT("non-root leaf's nritems is 0",
 590					leaf, check_root, 0);
 591				free_extent_buffer(eb);
 592				return -EIO;
 593			}
 594			free_extent_buffer(eb);
 595		}
 596		return 0;
 597	}
 598
 599	if (nritems == 0)
 600		return 0;
 601
 602	/* Check the 0 item */
 603	if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 604	    BTRFS_LEAF_DATA_SIZE(fs_info)) {
 605		CORRUPT("invalid item offset size pair", leaf, root, 0);
 606		return -EIO;
 607	}
 
 
 608
 
 
 
 
 609	/*
 610	 * Check to make sure each items keys are in the correct order and their
 611	 * offsets make sense.  We only have to loop through nritems-1 because
 612	 * we check the current slot against the next slot, which verifies the
 613	 * next slot's offset+size makes sense and that the current's slot
 614	 * offset is correct.
 615	 */
 616	for (slot = 0; slot < nritems - 1; slot++) {
 617		btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 618		btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 619
 620		/* Make sure the keys are in the right order */
 621		if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 622			CORRUPT("bad key order", leaf, root, slot);
 623			return -EIO;
 624		}
 625
 626		/*
 627		 * Make sure the offset and ends are right, remember that the
 628		 * item data starts at the end of the leaf and grows towards the
 629		 * front.
 630		 */
 631		if (btrfs_item_offset_nr(leaf, slot) !=
 632			btrfs_item_end_nr(leaf, slot + 1)) {
 633			CORRUPT("slot offset bad", leaf, root, slot);
 634			return -EIO;
 635		}
 636
 637		/*
 638		 * Check to make sure that we don't point outside of the leaf,
 639		 * just in case all the items are consistent to each other, but
 640		 * all point outside of the leaf.
 641		 */
 642		if (btrfs_item_end_nr(leaf, slot) >
 643		    BTRFS_LEAF_DATA_SIZE(fs_info)) {
 644			CORRUPT("slot end outside of leaf", leaf, root, slot);
 645			return -EIO;
 646		}
 647	}
 648
 649	return 0;
 650}
 651
 652static int check_node(struct btrfs_root *root, struct extent_buffer *node)
 653{
 654	unsigned long nr = btrfs_header_nritems(node);
 655	struct btrfs_key key, next_key;
 656	int slot;
 657	u64 bytenr;
 658	int ret = 0;
 
 659
 660	if (nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(root->fs_info)) {
 661		btrfs_crit(root->fs_info,
 662			   "corrupt node: block %llu root %llu nritems %lu",
 663			   node->start, root->objectid, nr);
 664		return -EIO;
 665	}
 
 
 666
 667	for (slot = 0; slot < nr - 1; slot++) {
 668		bytenr = btrfs_node_blockptr(node, slot);
 669		btrfs_node_key_to_cpu(node, &key, slot);
 670		btrfs_node_key_to_cpu(node, &next_key, slot + 1);
 671
 672		if (!bytenr) {
 673			CORRUPT("invalid item slot", node, root, slot);
 674			ret = -EIO;
 675			goto out;
 676		}
 677
 678		if (btrfs_comp_cpu_keys(&key, &next_key) >= 0) {
 679			CORRUPT("bad key order", node, root, slot);
 680			ret = -EIO;
 681			goto out;
 682		}
 683	}
 684out:
 685	return ret;
 686}
 687
 688static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 689				      u64 phy_offset, struct page *page,
 690				      u64 start, u64 end, int mirror)
 691{
 
 692	u64 found_start;
 693	int found_level;
 694	struct extent_buffer *eb;
 695	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 696	struct btrfs_fs_info *fs_info = root->fs_info;
 697	int ret = 0;
 698	int reads_done;
 699
 700	if (!page->private)
 701		goto out;
 702
 703	eb = (struct extent_buffer *)page->private;
 704
 705	/* the pending IO might have been the only thing that kept this buffer
 706	 * in memory.  Make sure we have a ref for all this other checks
 707	 */
 708	extent_buffer_get(eb);
 709
 710	reads_done = atomic_dec_and_test(&eb->io_pages);
 711	if (!reads_done)
 712		goto err;
 713
 714	eb->read_mirror = mirror;
 715	if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 716		ret = -EIO;
 717		goto err;
 718	}
 719
 720	found_start = btrfs_header_bytenr(eb);
 721	if (found_start != eb->start) {
 722		btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
 723			     found_start, eb->start);
 
 724		ret = -EIO;
 725		goto err;
 726	}
 727	if (check_tree_block_fsid(fs_info, eb)) {
 728		btrfs_err_rl(fs_info, "bad fsid on block %llu",
 729			     eb->start);
 730		ret = -EIO;
 731		goto err;
 732	}
 733	found_level = btrfs_header_level(eb);
 734	if (found_level >= BTRFS_MAX_LEVEL) {
 735		btrfs_err(fs_info, "bad tree block level %d",
 736			  (int)btrfs_header_level(eb));
 
 737		ret = -EIO;
 738		goto err;
 739	}
 740
 741	btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 742				       eb, found_level);
 
 
 
 
 
 
 
 
 
 
 
 
 743
 744	ret = csum_tree_block(fs_info, eb, 1);
 745	if (ret)
 746		goto err;
 747
 748	/*
 749	 * If this is a leaf block and it is corrupt, set the corrupt bit so
 750	 * that we don't try and read the other copies of this block, just
 751	 * return -EIO.
 752	 */
 753	if (found_level == 0 && check_leaf(root, eb)) {
 754		set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 755		ret = -EIO;
 
 756	}
 757
 758	if (found_level > 0 && check_node(root, eb))
 
 
 
 
 759		ret = -EIO;
 760
 761	if (!ret)
 762		set_extent_buffer_uptodate(eb);
 763err:
 764	if (reads_done &&
 765	    test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 766		btree_readahead_hook(fs_info, eb, ret);
 767
 768	if (ret) {
 769		/*
 770		 * our io error hook is going to dec the io pages
 771		 * again, we have to make sure it has something
 772		 * to decrement
 773		 */
 774		atomic_inc(&eb->io_pages);
 775		clear_extent_buffer_uptodate(eb);
 776	}
 777	free_extent_buffer(eb);
 778out:
 779	return ret;
 780}
 781
 782static int btree_io_failed_hook(struct page *page, int failed_mirror)
 783{
 784	struct extent_buffer *eb;
 785
 786	eb = (struct extent_buffer *)page->private;
 787	set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
 788	eb->read_mirror = failed_mirror;
 789	atomic_dec(&eb->io_pages);
 790	if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 791		btree_readahead_hook(eb->fs_info, eb, -EIO);
 792	return -EIO;	/* we fixed nothing */
 793}
 794
 795static void end_workqueue_bio(struct bio *bio)
 796{
 797	struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 798	struct btrfs_fs_info *fs_info;
 799	struct btrfs_workqueue *wq;
 800	btrfs_work_func_t func;
 801
 802	fs_info = end_io_wq->info;
 803	end_io_wq->error = bio->bi_error;
 804
 805	if (bio_op(bio) == REQ_OP_WRITE) {
 806		if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 807			wq = fs_info->endio_meta_write_workers;
 808			func = btrfs_endio_meta_write_helper;
 809		} else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 810			wq = fs_info->endio_freespace_worker;
 811			func = btrfs_freespace_write_helper;
 812		} else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 813			wq = fs_info->endio_raid56_workers;
 814			func = btrfs_endio_raid56_helper;
 815		} else {
 816			wq = fs_info->endio_write_workers;
 817			func = btrfs_endio_write_helper;
 818		}
 819	} else {
 820		if (unlikely(end_io_wq->metadata ==
 821			     BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 822			wq = fs_info->endio_repair_workers;
 823			func = btrfs_endio_repair_helper;
 824		} else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 825			wq = fs_info->endio_raid56_workers;
 826			func = btrfs_endio_raid56_helper;
 827		} else if (end_io_wq->metadata) {
 828			wq = fs_info->endio_meta_workers;
 829			func = btrfs_endio_meta_helper;
 830		} else {
 831			wq = fs_info->endio_workers;
 832			func = btrfs_endio_helper;
 833		}
 834	}
 835
 836	btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 837	btrfs_queue_work(wq, &end_io_wq->work);
 838}
 839
 840int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 841			enum btrfs_wq_endio_type metadata)
 842{
 843	struct btrfs_end_io_wq *end_io_wq;
 844
 845	end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 846	if (!end_io_wq)
 847		return -ENOMEM;
 848
 849	end_io_wq->private = bio->bi_private;
 850	end_io_wq->end_io = bio->bi_end_io;
 851	end_io_wq->info = info;
 852	end_io_wq->error = 0;
 853	end_io_wq->bio = bio;
 854	end_io_wq->metadata = metadata;
 855
 856	bio->bi_private = end_io_wq;
 857	bio->bi_end_io = end_workqueue_bio;
 858	return 0;
 859}
 860
 861unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 862{
 863	unsigned long limit = min_t(unsigned long,
 864				    info->thread_pool_size,
 865				    info->fs_devices->open_devices);
 866	return 256 * limit;
 867}
 868
 869static void run_one_async_start(struct btrfs_work *work)
 870{
 871	struct async_submit_bio *async;
 872	int ret;
 873
 874	async = container_of(work, struct  async_submit_bio, work);
 875	ret = async->submit_bio_start(async->inode, async->bio,
 876				      async->mirror_num, async->bio_flags,
 877				      async->bio_offset);
 878	if (ret)
 879		async->error = ret;
 880}
 881
 882static void run_one_async_done(struct btrfs_work *work)
 883{
 884	struct btrfs_fs_info *fs_info;
 885	struct async_submit_bio *async;
 886	int limit;
 887
 888	async = container_of(work, struct  async_submit_bio, work);
 889	fs_info = BTRFS_I(async->inode)->root->fs_info;
 890
 891	limit = btrfs_async_submit_limit(fs_info);
 892	limit = limit * 2 / 3;
 893
 894	/*
 895	 * atomic_dec_return implies a barrier for waitqueue_active
 896	 */
 897	if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
 898	    waitqueue_active(&fs_info->async_submit_wait))
 899		wake_up(&fs_info->async_submit_wait);
 900
 901	/* If an error occurred we just want to clean up the bio and move on */
 902	if (async->error) {
 903		async->bio->bi_error = async->error;
 904		bio_endio(async->bio);
 905		return;
 906	}
 907
 908	async->submit_bio_done(async->inode, async->bio, async->mirror_num,
 909			       async->bio_flags, async->bio_offset);
 910}
 911
 912static void run_one_async_free(struct btrfs_work *work)
 913{
 914	struct async_submit_bio *async;
 915
 916	async = container_of(work, struct  async_submit_bio, work);
 917	kfree(async);
 918}
 919
 920int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 921			struct bio *bio, int mirror_num,
 922			unsigned long bio_flags,
 923			u64 bio_offset,
 924			extent_submit_bio_hook_t *submit_bio_start,
 925			extent_submit_bio_hook_t *submit_bio_done)
 926{
 927	struct async_submit_bio *async;
 928
 929	async = kmalloc(sizeof(*async), GFP_NOFS);
 930	if (!async)
 931		return -ENOMEM;
 932
 933	async->inode = inode;
 934	async->bio = bio;
 935	async->mirror_num = mirror_num;
 936	async->submit_bio_start = submit_bio_start;
 937	async->submit_bio_done = submit_bio_done;
 938
 939	btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 940			run_one_async_done, run_one_async_free);
 941
 942	async->bio_flags = bio_flags;
 943	async->bio_offset = bio_offset;
 944
 945	async->error = 0;
 946
 947	atomic_inc(&fs_info->nr_async_submits);
 948
 949	if (op_is_sync(bio->bi_opf))
 950		btrfs_set_work_high_priority(&async->work);
 951
 952	btrfs_queue_work(fs_info->workers, &async->work);
 953
 954	while (atomic_read(&fs_info->async_submit_draining) &&
 955	      atomic_read(&fs_info->nr_async_submits)) {
 956		wait_event(fs_info->async_submit_wait,
 957			   (atomic_read(&fs_info->nr_async_submits) == 0));
 958	}
 959
 960	return 0;
 961}
 962
 963static int btree_csum_one_bio(struct bio *bio)
 964{
 965	struct bio_vec *bvec;
 966	struct btrfs_root *root;
 967	int i, ret = 0;
 968
 969	bio_for_each_segment_all(bvec, bio, i) {
 970		root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 971		ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 972		if (ret)
 973			break;
 974	}
 975
 976	return ret;
 977}
 978
 979static int __btree_submit_bio_start(struct inode *inode, struct bio *bio,
 980				    int mirror_num, unsigned long bio_flags,
 981				    u64 bio_offset)
 982{
 983	/*
 984	 * when we're called for a write, we're already in the async
 985	 * submission context.  Just jump into btrfs_map_bio
 
 986	 */
 987	return btree_csum_one_bio(bio);
 988}
 989
 990static int __btree_submit_bio_done(struct inode *inode, struct bio *bio,
 991				 int mirror_num, unsigned long bio_flags,
 992				 u64 bio_offset)
 993{
 994	int ret;
 995
 996	/*
 997	 * when we're called for a write, we're already in the async
 998	 * submission context.  Just jump into btrfs_map_bio
 999	 */
1000	ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
1001	if (ret) {
1002		bio->bi_error = ret;
1003		bio_endio(bio);
1004	}
1005	return ret;
1006}
1007
1008static int check_async_write(struct inode *inode, unsigned long bio_flags)
1009{
1010	if (bio_flags & EXTENT_BIO_TREE_LOG)
1011		return 0;
1012#ifdef CONFIG_X86
1013	if (static_cpu_has(X86_FEATURE_XMM4_2))
1014		return 0;
1015#endif
1016	return 1;
1017}
1018
1019static int btree_submit_bio_hook(struct inode *inode, struct bio *bio,
1020				 int mirror_num, unsigned long bio_flags,
1021				 u64 bio_offset)
1022{
1023	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1024	int async = check_async_write(inode, bio_flags);
1025	int ret;
1026
1027	if (bio_op(bio) != REQ_OP_WRITE) {
1028		/*
1029		 * called for a read, do the setup so that checksum validation
1030		 * can happen in the async kernel threads
1031		 */
1032		ret = btrfs_bio_wq_end_io(fs_info, bio,
1033					  BTRFS_WQ_ENDIO_METADATA);
1034		if (ret)
1035			goto out_w_error;
1036		ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1037	} else if (!async) {
1038		ret = btree_csum_one_bio(bio);
1039		if (ret)
1040			goto out_w_error;
1041		ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1042	} else {
1043		/*
1044		 * kthread helpers are used to submit writes so that
1045		 * checksumming can happen in parallel across all CPUs
1046		 */
1047		ret = btrfs_wq_submit_bio(fs_info, inode, bio, mirror_num, 0,
1048					  bio_offset,
1049					  __btree_submit_bio_start,
1050					  __btree_submit_bio_done);
1051	}
1052
1053	if (ret)
1054		goto out_w_error;
1055	return 0;
1056
1057out_w_error:
1058	bio->bi_error = ret;
1059	bio_endio(bio);
1060	return ret;
1061}
1062
1063#ifdef CONFIG_MIGRATION
1064static int btree_migratepage(struct address_space *mapping,
1065			struct page *newpage, struct page *page,
1066			enum migrate_mode mode)
1067{
1068	/*
1069	 * we can't safely write a btree page from here,
1070	 * we haven't done the locking hook
1071	 */
1072	if (PageDirty(page))
1073		return -EAGAIN;
1074	/*
1075	 * Buffers may be managed in a filesystem specific way.
1076	 * We must have no buffers or drop them.
1077	 */
1078	if (page_has_private(page) &&
1079	    !try_to_release_page(page, GFP_KERNEL))
1080		return -EAGAIN;
1081	return migrate_page(mapping, newpage, page, mode);
1082}
 
 
1083#endif
1084
1085
1086static int btree_writepages(struct address_space *mapping,
1087			    struct writeback_control *wbc)
1088{
1089	struct btrfs_fs_info *fs_info;
1090	int ret;
1091
1092	if (wbc->sync_mode == WB_SYNC_NONE) {
1093
1094		if (wbc->for_kupdate)
1095			return 0;
1096
1097		fs_info = BTRFS_I(mapping->host)->root->fs_info;
1098		/* this is a bit racy, but that's ok */
1099		ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1100					     BTRFS_DIRTY_METADATA_THRESH);
 
1101		if (ret < 0)
1102			return 0;
1103	}
1104	return btree_write_cache_pages(mapping, wbc);
1105}
1106
1107static int btree_readpage(struct file *file, struct page *page)
1108{
1109	struct extent_io_tree *tree;
1110	tree = &BTRFS_I(page->mapping->host)->io_tree;
1111	return extent_read_full_page(tree, page, btree_get_extent, 0);
1112}
1113
1114static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1115{
1116	if (PageWriteback(page) || PageDirty(page))
1117		return 0;
1118
1119	return try_release_extent_buffer(page);
1120}
1121
1122static void btree_invalidatepage(struct page *page, unsigned int offset,
1123				 unsigned int length)
1124{
1125	struct extent_io_tree *tree;
1126	tree = &BTRFS_I(page->mapping->host)->io_tree;
1127	extent_invalidatepage(tree, page, offset);
1128	btree_releasepage(page, GFP_NOFS);
1129	if (PagePrivate(page)) {
1130		btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1131			   "page private not zero on page %llu",
1132			   (unsigned long long)page_offset(page));
1133		ClearPagePrivate(page);
1134		set_page_private(page, 0);
1135		put_page(page);
1136	}
1137}
1138
1139static int btree_set_page_dirty(struct page *page)
 
 
1140{
1141#ifdef DEBUG
 
 
1142	struct extent_buffer *eb;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1143
1144	BUG_ON(!PagePrivate(page));
1145	eb = (struct extent_buffer *)page->private;
1146	BUG_ON(!eb);
1147	BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1148	BUG_ON(!atomic_read(&eb->refs));
1149	btrfs_assert_tree_locked(eb);
1150#endif
1151	return __set_page_dirty_nobuffers(page);
1152}
1153
1154static const struct address_space_operations btree_aops = {
1155	.readpage	= btree_readpage,
1156	.writepages	= btree_writepages,
1157	.releasepage	= btree_releasepage,
1158	.invalidatepage = btree_invalidatepage,
1159#ifdef CONFIG_MIGRATION
1160	.migratepage	= btree_migratepage,
1161#endif
1162	.set_page_dirty = btree_set_page_dirty,
1163};
1164
1165void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1166{
1167	struct extent_buffer *buf = NULL;
1168	struct inode *btree_inode = fs_info->btree_inode;
1169
1170	buf = btrfs_find_create_tree_block(fs_info, bytenr);
1171	if (IS_ERR(buf))
1172		return;
1173	read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1174				 buf, WAIT_NONE, btree_get_extent, 0);
1175	free_extent_buffer(buf);
1176}
1177
1178int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1179			 int mirror_num, struct extent_buffer **eb)
1180{
1181	struct extent_buffer *buf = NULL;
1182	struct inode *btree_inode = fs_info->btree_inode;
1183	struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1184	int ret;
1185
1186	buf = btrfs_find_create_tree_block(fs_info, bytenr);
1187	if (IS_ERR(buf))
1188		return 0;
1189
1190	set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1191
1192	ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1193				       btree_get_extent, mirror_num);
1194	if (ret) {
1195		free_extent_buffer(buf);
1196		return ret;
1197	}
1198
1199	if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1200		free_extent_buffer(buf);
1201		return -EIO;
1202	} else if (extent_buffer_uptodate(buf)) {
1203		*eb = buf;
1204	} else {
1205		free_extent_buffer(buf);
1206	}
1207	return 0;
1208}
1209
1210struct extent_buffer *btrfs_find_create_tree_block(
1211						struct btrfs_fs_info *fs_info,
1212						u64 bytenr)
 
1213{
1214	if (btrfs_is_testing(fs_info))
1215		return alloc_test_extent_buffer(fs_info, bytenr);
1216	return alloc_extent_buffer(fs_info, bytenr);
1217}
1218
1219
1220int btrfs_write_tree_block(struct extent_buffer *buf)
1221{
1222	return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1223					buf->start + buf->len - 1);
1224}
1225
1226int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1227{
1228	return filemap_fdatawait_range(buf->pages[0]->mapping,
1229				       buf->start, buf->start + buf->len - 1);
1230}
1231
 
 
 
 
 
 
 
1232struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1233				      u64 parent_transid)
1234{
1235	struct extent_buffer *buf = NULL;
1236	int ret;
1237
1238	buf = btrfs_find_create_tree_block(fs_info, bytenr);
 
 
 
1239	if (IS_ERR(buf))
1240		return buf;
1241
1242	ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
1243	if (ret) {
1244		free_extent_buffer(buf);
1245		return ERR_PTR(ret);
1246	}
 
 
 
 
1247	return buf;
1248
1249}
1250
1251void clean_tree_block(struct btrfs_trans_handle *trans,
1252		      struct btrfs_fs_info *fs_info,
1253		      struct extent_buffer *buf)
1254{
1255	if (btrfs_header_generation(buf) ==
1256	    fs_info->running_transaction->transid) {
1257		btrfs_assert_tree_locked(buf);
1258
1259		if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1260			__percpu_counter_add(&fs_info->dirty_metadata_bytes,
1261					     -buf->len,
1262					     fs_info->dirty_metadata_batch);
1263			/* ugh, clear_extent_buffer_dirty needs to lock the page */
1264			btrfs_set_lock_blocking(buf);
1265			clear_extent_buffer_dirty(buf);
1266		}
1267	}
1268}
1269
1270static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1271{
1272	struct btrfs_subvolume_writers *writers;
1273	int ret;
1274
1275	writers = kmalloc(sizeof(*writers), GFP_NOFS);
1276	if (!writers)
1277		return ERR_PTR(-ENOMEM);
1278
1279	ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1280	if (ret < 0) {
1281		kfree(writers);
1282		return ERR_PTR(ret);
1283	}
1284
1285	init_waitqueue_head(&writers->wait);
1286	return writers;
1287}
1288
1289static void
1290btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1291{
1292	percpu_counter_destroy(&writers->counter);
1293	kfree(writers);
1294}
1295
1296static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1297			 u64 objectid)
1298{
1299	bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
 
 
 
 
 
 
1300	root->node = NULL;
1301	root->commit_root = NULL;
1302	root->state = 0;
1303	root->orphan_cleanup_state = 0;
1304
1305	root->objectid = objectid;
1306	root->last_trans = 0;
1307	root->highest_objectid = 0;
1308	root->nr_delalloc_inodes = 0;
1309	root->nr_ordered_extents = 0;
1310	root->name = NULL;
1311	root->inode_tree = RB_ROOT;
1312	INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1313	root->block_rsv = NULL;
1314	root->orphan_block_rsv = NULL;
 
1315
1316	INIT_LIST_HEAD(&root->dirty_list);
1317	INIT_LIST_HEAD(&root->root_list);
1318	INIT_LIST_HEAD(&root->delalloc_inodes);
1319	INIT_LIST_HEAD(&root->delalloc_root);
1320	INIT_LIST_HEAD(&root->ordered_extents);
1321	INIT_LIST_HEAD(&root->ordered_root);
1322	INIT_LIST_HEAD(&root->logged_list[0]);
1323	INIT_LIST_HEAD(&root->logged_list[1]);
1324	spin_lock_init(&root->orphan_lock);
1325	spin_lock_init(&root->inode_lock);
1326	spin_lock_init(&root->delalloc_lock);
1327	spin_lock_init(&root->ordered_extent_lock);
1328	spin_lock_init(&root->accounting_lock);
1329	spin_lock_init(&root->log_extents_lock[0]);
1330	spin_lock_init(&root->log_extents_lock[1]);
1331	mutex_init(&root->objectid_mutex);
1332	mutex_init(&root->log_mutex);
1333	mutex_init(&root->ordered_extent_mutex);
1334	mutex_init(&root->delalloc_mutex);
 
1335	init_waitqueue_head(&root->log_writer_wait);
1336	init_waitqueue_head(&root->log_commit_wait[0]);
1337	init_waitqueue_head(&root->log_commit_wait[1]);
1338	INIT_LIST_HEAD(&root->log_ctxs[0]);
1339	INIT_LIST_HEAD(&root->log_ctxs[1]);
1340	atomic_set(&root->log_commit[0], 0);
1341	atomic_set(&root->log_commit[1], 0);
1342	atomic_set(&root->log_writers, 0);
1343	atomic_set(&root->log_batch, 0);
1344	atomic_set(&root->orphan_inodes, 0);
1345	atomic_set(&root->refs, 1);
1346	atomic_set(&root->will_be_snapshoted, 0);
1347	atomic_set(&root->qgroup_meta_rsv, 0);
1348	root->log_transid = 0;
1349	root->log_transid_committed = -1;
1350	root->last_log_commit = 0;
1351	if (!dummy)
1352		extent_io_tree_init(&root->dirty_log_pages,
1353				     fs_info->btree_inode->i_mapping);
1354
1355	memset(&root->root_key, 0, sizeof(root->root_key));
1356	memset(&root->root_item, 0, sizeof(root->root_item));
1357	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1358	if (!dummy)
1359		root->defrag_trans_start = fs_info->generation;
1360	else
1361		root->defrag_trans_start = 0;
1362	root->root_key.objectid = objectid;
1363	root->anon_dev = 0;
 
 
 
 
 
 
1364
1365	spin_lock_init(&root->root_item_lock);
 
 
 
 
 
 
 
1366}
1367
1368static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1369		gfp_t flags)
1370{
1371	struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1372	if (root)
1373		root->fs_info = fs_info;
1374	return root;
1375}
1376
1377#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1378/* Should only be used by the testing infrastructure */
1379struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1380{
1381	struct btrfs_root *root;
1382
1383	if (!fs_info)
1384		return ERR_PTR(-EINVAL);
1385
1386	root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1387	if (!root)
1388		return ERR_PTR(-ENOMEM);
1389
1390	/* We don't use the stripesize in selftest, set it as sectorsize */
1391	__setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1392	root->alloc_bytenr = 0;
1393
1394	return root;
1395}
1396#endif
1397
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1398struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1399				     struct btrfs_fs_info *fs_info,
1400				     u64 objectid)
1401{
 
1402	struct extent_buffer *leaf;
1403	struct btrfs_root *tree_root = fs_info->tree_root;
1404	struct btrfs_root *root;
1405	struct btrfs_key key;
 
1406	int ret = 0;
1407	uuid_le uuid;
1408
1409	root = btrfs_alloc_root(fs_info, GFP_KERNEL);
 
 
 
 
 
 
1410	if (!root)
1411		return ERR_PTR(-ENOMEM);
1412
1413	__setup_root(root, fs_info, objectid);
1414	root->root_key.objectid = objectid;
1415	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1416	root->root_key.offset = 0;
1417
1418	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 
1419	if (IS_ERR(leaf)) {
1420		ret = PTR_ERR(leaf);
1421		leaf = NULL;
1422		goto fail;
1423	}
1424
1425	memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1426	btrfs_set_header_bytenr(leaf, leaf->start);
1427	btrfs_set_header_generation(leaf, trans->transid);
1428	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1429	btrfs_set_header_owner(leaf, objectid);
1430	root->node = leaf;
1431
1432	write_extent_buffer_fsid(leaf, fs_info->fsid);
1433	write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1434	btrfs_mark_buffer_dirty(leaf);
1435
1436	root->commit_root = btrfs_root_node(root);
1437	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1438
1439	root->root_item.flags = 0;
1440	root->root_item.byte_limit = 0;
1441	btrfs_set_root_bytenr(&root->root_item, leaf->start);
1442	btrfs_set_root_generation(&root->root_item, trans->transid);
1443	btrfs_set_root_level(&root->root_item, 0);
1444	btrfs_set_root_refs(&root->root_item, 1);
1445	btrfs_set_root_used(&root->root_item, leaf->len);
1446	btrfs_set_root_last_snapshot(&root->root_item, 0);
1447	btrfs_set_root_dirid(&root->root_item, 0);
1448	uuid_le_gen(&uuid);
1449	memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1450	root->root_item.drop_level = 0;
 
 
 
 
1451
1452	key.objectid = objectid;
1453	key.type = BTRFS_ROOT_ITEM_KEY;
1454	key.offset = 0;
1455	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1456	if (ret)
1457		goto fail;
1458
1459	btrfs_tree_unlock(leaf);
1460
1461	return root;
1462
1463fail:
1464	if (leaf) {
1465		btrfs_tree_unlock(leaf);
1466		free_extent_buffer(root->commit_root);
1467		free_extent_buffer(leaf);
1468	}
1469	kfree(root);
1470
1471	return ERR_PTR(ret);
1472}
1473
1474static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1475					 struct btrfs_fs_info *fs_info)
1476{
1477	struct btrfs_root *root;
1478	struct extent_buffer *leaf;
1479
1480	root = btrfs_alloc_root(fs_info, GFP_NOFS);
1481	if (!root)
1482		return ERR_PTR(-ENOMEM);
1483
1484	__setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1485
1486	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1487	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1488	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1489
 
 
 
 
 
 
 
 
1490	/*
1491	 * DON'T set REF_COWS for log trees
1492	 *
1493	 * log trees do not get reference counted because they go away
1494	 * before a real commit is actually done.  They do store pointers
1495	 * to file data extents, and those reference counts still get
1496	 * updated (along with back refs to the log tree).
 
1497	 */
1498
1499	leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1500			NULL, 0, 0, 0);
1501	if (IS_ERR(leaf)) {
1502		kfree(root);
1503		return ERR_CAST(leaf);
1504	}
1505
1506	memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1507	btrfs_set_header_bytenr(leaf, leaf->start);
1508	btrfs_set_header_generation(leaf, trans->transid);
1509	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1510	btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1511	root->node = leaf;
1512
1513	write_extent_buffer_fsid(root->node, fs_info->fsid);
1514	btrfs_mark_buffer_dirty(root->node);
1515	btrfs_tree_unlock(root->node);
1516	return root;
 
1517}
1518
1519int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1520			     struct btrfs_fs_info *fs_info)
1521{
1522	struct btrfs_root *log_root;
1523
1524	log_root = alloc_log_tree(trans, fs_info);
1525	if (IS_ERR(log_root))
1526		return PTR_ERR(log_root);
 
 
 
 
 
 
 
 
 
 
1527	WARN_ON(fs_info->log_root_tree);
1528	fs_info->log_root_tree = log_root;
1529	return 0;
1530}
1531
1532int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1533		       struct btrfs_root *root)
1534{
1535	struct btrfs_fs_info *fs_info = root->fs_info;
1536	struct btrfs_root *log_root;
1537	struct btrfs_inode_item *inode_item;
 
1538
1539	log_root = alloc_log_tree(trans, fs_info);
1540	if (IS_ERR(log_root))
1541		return PTR_ERR(log_root);
1542
 
 
 
 
 
 
1543	log_root->last_trans = trans->transid;
1544	log_root->root_key.offset = root->root_key.objectid;
1545
1546	inode_item = &log_root->root_item.inode;
1547	btrfs_set_stack_inode_generation(inode_item, 1);
1548	btrfs_set_stack_inode_size(inode_item, 3);
1549	btrfs_set_stack_inode_nlink(inode_item, 1);
1550	btrfs_set_stack_inode_nbytes(inode_item,
1551				     fs_info->nodesize);
1552	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1553
1554	btrfs_set_root_node(&log_root->root_item, log_root->node);
1555
1556	WARN_ON(root->log_root);
1557	root->log_root = log_root;
1558	root->log_transid = 0;
1559	root->log_transid_committed = -1;
1560	root->last_log_commit = 0;
1561	return 0;
1562}
1563
1564static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1565					       struct btrfs_key *key)
 
1566{
1567	struct btrfs_root *root;
 
1568	struct btrfs_fs_info *fs_info = tree_root->fs_info;
1569	struct btrfs_path *path;
1570	u64 generation;
1571	int ret;
 
1572
1573	path = btrfs_alloc_path();
1574	if (!path)
1575		return ERR_PTR(-ENOMEM);
1576
1577	root = btrfs_alloc_root(fs_info, GFP_NOFS);
1578	if (!root) {
1579		ret = -ENOMEM;
1580		goto alloc_fail;
1581	}
1582
1583	__setup_root(root, fs_info, key->objectid);
1584
1585	ret = btrfs_find_root(tree_root, key, path,
1586			      &root->root_item, &root->root_key);
1587	if (ret) {
1588		if (ret > 0)
1589			ret = -ENOENT;
1590		goto find_fail;
1591	}
1592
1593	generation = btrfs_root_generation(&root->root_item);
1594	root->node = read_tree_block(fs_info,
1595				     btrfs_root_bytenr(&root->root_item),
1596				     generation);
 
 
 
1597	if (IS_ERR(root->node)) {
1598		ret = PTR_ERR(root->node);
1599		goto find_fail;
1600	} else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
 
 
1601		ret = -EIO;
1602		free_extent_buffer(root->node);
1603		goto find_fail;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1604	}
1605	root->commit_root = btrfs_root_node(root);
1606out:
1607	btrfs_free_path(path);
1608	return root;
1609
1610find_fail:
1611	kfree(root);
1612alloc_fail:
1613	root = ERR_PTR(ret);
1614	goto out;
1615}
1616
1617struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1618				      struct btrfs_key *location)
1619{
1620	struct btrfs_root *root;
 
1621
1622	root = btrfs_read_tree_root(tree_root, location);
1623	if (IS_ERR(root))
1624		return root;
1625
1626	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1627		set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1628		btrfs_check_and_init_root_item(&root->root_item);
1629	}
1630
1631	return root;
1632}
1633
1634int btrfs_init_fs_root(struct btrfs_root *root)
 
 
 
 
 
1635{
1636	int ret;
1637	struct btrfs_subvolume_writers *writers;
1638
1639	root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1640	root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1641					GFP_NOFS);
1642	if (!root->free_ino_pinned || !root->free_ino_ctl) {
1643		ret = -ENOMEM;
1644		goto fail;
 
1645	}
1646
1647	writers = btrfs_alloc_subvolume_writers();
1648	if (IS_ERR(writers)) {
1649		ret = PTR_ERR(writers);
1650		goto fail;
 
 
 
 
 
 
 
 
 
1651	}
1652	root->subv_writers = writers;
1653
1654	btrfs_init_free_ino_ctl(root);
1655	spin_lock_init(&root->ino_cache_lock);
1656	init_waitqueue_head(&root->ino_cache_wait);
1657
1658	ret = get_anon_bdev(&root->anon_dev);
1659	if (ret)
1660		goto fail;
1661
1662	mutex_lock(&root->objectid_mutex);
1663	ret = btrfs_find_highest_objectid(root,
1664					&root->highest_objectid);
1665	if (ret) {
1666		mutex_unlock(&root->objectid_mutex);
1667		goto fail;
1668	}
1669
1670	ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1671
1672	mutex_unlock(&root->objectid_mutex);
1673
1674	return 0;
1675fail:
1676	/* the caller is responsible to call free_fs_root */
1677	return ret;
1678}
1679
1680struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1681					u64 root_id)
1682{
1683	struct btrfs_root *root;
1684
1685	spin_lock(&fs_info->fs_roots_radix_lock);
1686	root = radix_tree_lookup(&fs_info->fs_roots_radix,
1687				 (unsigned long)root_id);
 
1688	spin_unlock(&fs_info->fs_roots_radix_lock);
1689	return root;
1690}
1691
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1692int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1693			 struct btrfs_root *root)
1694{
1695	int ret;
1696
1697	ret = radix_tree_preload(GFP_NOFS);
1698	if (ret)
1699		return ret;
1700
1701	spin_lock(&fs_info->fs_roots_radix_lock);
1702	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1703				(unsigned long)root->root_key.objectid,
1704				root);
1705	if (ret == 0)
 
1706		set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
 
1707	spin_unlock(&fs_info->fs_roots_radix_lock);
1708	radix_tree_preload_end();
1709
1710	return ret;
1711}
1712
1713struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1714				     struct btrfs_key *location,
1715				     bool check_ref)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1716{
1717	struct btrfs_root *root;
1718	struct btrfs_path *path;
1719	struct btrfs_key key;
1720	int ret;
1721
1722	if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1723		return fs_info->tree_root;
1724	if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1725		return fs_info->extent_root;
1726	if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1727		return fs_info->chunk_root;
1728	if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1729		return fs_info->dev_root;
1730	if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1731		return fs_info->csum_root;
1732	if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1733		return fs_info->quota_root ? fs_info->quota_root :
1734					     ERR_PTR(-ENOENT);
1735	if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1736		return fs_info->uuid_root ? fs_info->uuid_root :
1737					    ERR_PTR(-ENOENT);
1738	if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1739		return fs_info->free_space_root ? fs_info->free_space_root :
1740						  ERR_PTR(-ENOENT);
1741again:
1742	root = btrfs_lookup_fs_root(fs_info, location->objectid);
1743	if (root) {
1744		if (check_ref && btrfs_root_refs(&root->root_item) == 0)
 
 
 
 
 
 
 
 
 
 
 
 
1745			return ERR_PTR(-ENOENT);
 
1746		return root;
1747	}
1748
1749	root = btrfs_read_fs_root(fs_info->tree_root, location);
 
 
 
1750	if (IS_ERR(root))
1751		return root;
1752
1753	if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1754		ret = -ENOENT;
1755		goto fail;
1756	}
1757
1758	ret = btrfs_init_fs_root(root);
1759	if (ret)
1760		goto fail;
1761
1762	path = btrfs_alloc_path();
1763	if (!path) {
1764		ret = -ENOMEM;
1765		goto fail;
1766	}
1767	key.objectid = BTRFS_ORPHAN_OBJECTID;
1768	key.type = BTRFS_ORPHAN_ITEM_KEY;
1769	key.offset = location->objectid;
1770
1771	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1772	btrfs_free_path(path);
1773	if (ret < 0)
1774		goto fail;
1775	if (ret == 0)
1776		set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1777
1778	ret = btrfs_insert_fs_root(fs_info, root);
1779	if (ret) {
1780		if (ret == -EEXIST) {
1781			free_fs_root(root);
1782			goto again;
1783		}
1784		goto fail;
1785	}
1786	return root;
1787fail:
1788	free_fs_root(root);
 
 
 
 
 
 
 
 
1789	return ERR_PTR(ret);
1790}
1791
1792static int btrfs_congested_fn(void *congested_data, int bdi_bits)
 
 
 
 
 
 
 
 
1793{
1794	struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1795	int ret = 0;
1796	struct btrfs_device *device;
1797	struct backing_dev_info *bdi;
1798
1799	rcu_read_lock();
1800	list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1801		if (!device->bdev)
1802			continue;
1803		bdi = blk_get_backing_dev_info(device->bdev);
1804		if (bdi_congested(bdi, bdi_bits)) {
1805			ret = 1;
1806			break;
1807		}
1808	}
1809	rcu_read_unlock();
1810	return ret;
1811}
1812
1813static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
 
 
 
 
 
 
 
 
 
1814{
1815	int err;
1816
1817	err = bdi_setup_and_register(bdi, "btrfs");
1818	if (err)
1819		return err;
1820
1821	bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
1822	bdi->congested_fn	= btrfs_congested_fn;
1823	bdi->congested_data	= info;
1824	bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1825	return 0;
1826}
1827
1828/*
1829 * called by the kthread helper functions to finally call the bio end_io
1830 * functions.  This is where read checksum verification actually happens
 
 
 
 
 
 
 
 
 
 
 
1831 */
1832static void end_workqueue_fn(struct btrfs_work *work)
 
 
1833{
1834	struct bio *bio;
1835	struct btrfs_end_io_wq *end_io_wq;
1836
1837	end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1838	bio = end_io_wq->bio;
1839
1840	bio->bi_error = end_io_wq->error;
1841	bio->bi_private = end_io_wq->private;
1842	bio->bi_end_io = end_io_wq->end_io;
1843	kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1844	bio_endio(bio);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1845}
1846
1847static int cleaner_kthread(void *arg)
1848{
1849	struct btrfs_root *root = arg;
1850	struct btrfs_fs_info *fs_info = root->fs_info;
1851	int again;
1852	struct btrfs_trans_handle *trans;
1853
1854	do {
1855		again = 0;
1856
 
 
1857		/* Make the cleaner go to sleep early. */
1858		if (btrfs_need_cleaner_sleep(fs_info))
1859			goto sleep;
1860
1861		/*
1862		 * Do not do anything if we might cause open_ctree() to block
1863		 * before we have finished mounting the filesystem.
1864		 */
1865		if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1866			goto sleep;
1867
1868		if (!mutex_trylock(&fs_info->cleaner_mutex))
1869			goto sleep;
1870
1871		/*
1872		 * Avoid the problem that we change the status of the fs
1873		 * during the above check and trylock.
1874		 */
1875		if (btrfs_need_cleaner_sleep(fs_info)) {
1876			mutex_unlock(&fs_info->cleaner_mutex);
1877			goto sleep;
1878		}
1879
1880		mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
 
 
1881		btrfs_run_delayed_iputs(fs_info);
1882		mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1883
1884		again = btrfs_clean_one_deleted_snapshot(root);
1885		mutex_unlock(&fs_info->cleaner_mutex);
1886
1887		/*
1888		 * The defragger has dealt with the R/O remount and umount,
1889		 * needn't do anything special here.
1890		 */
1891		btrfs_run_defrag_inodes(fs_info);
1892
1893		/*
1894		 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1895		 * with relocation (btrfs_relocate_chunk) and relocation
1896		 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1897		 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1898		 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1899		 * unused block groups.
1900		 */
1901		btrfs_delete_unused_bgs(fs_info);
 
 
 
 
 
 
 
1902sleep:
 
 
 
 
 
1903		if (!again) {
1904			set_current_state(TASK_INTERRUPTIBLE);
1905			if (!kthread_should_stop())
1906				schedule();
1907			__set_current_state(TASK_RUNNING);
1908		}
1909	} while (!kthread_should_stop());
1910
1911	/*
1912	 * Transaction kthread is stopped before us and wakes us up.
1913	 * However we might have started a new transaction and COWed some
1914	 * tree blocks when deleting unused block groups for example. So
1915	 * make sure we commit the transaction we started to have a clean
1916	 * shutdown when evicting the btree inode - if it has dirty pages
1917	 * when we do the final iput() on it, eviction will trigger a
1918	 * writeback for it which will fail with null pointer dereferences
1919	 * since work queues and other resources were already released and
1920	 * destroyed by the time the iput/eviction/writeback is made.
1921	 */
1922	trans = btrfs_attach_transaction(root);
1923	if (IS_ERR(trans)) {
1924		if (PTR_ERR(trans) != -ENOENT)
1925			btrfs_err(fs_info,
1926				  "cleaner transaction attach returned %ld",
1927				  PTR_ERR(trans));
1928	} else {
1929		int ret;
1930
1931		ret = btrfs_commit_transaction(trans);
1932		if (ret)
1933			btrfs_err(fs_info,
1934				  "cleaner open transaction commit returned %d",
1935				  ret);
1936	}
1937
1938	return 0;
1939}
1940
1941static int transaction_kthread(void *arg)
1942{
1943	struct btrfs_root *root = arg;
1944	struct btrfs_fs_info *fs_info = root->fs_info;
1945	struct btrfs_trans_handle *trans;
1946	struct btrfs_transaction *cur;
1947	u64 transid;
1948	unsigned long now;
1949	unsigned long delay;
1950	bool cannot_commit;
1951
1952	do {
1953		cannot_commit = false;
1954		delay = HZ * fs_info->commit_interval;
1955		mutex_lock(&fs_info->transaction_kthread_mutex);
1956
1957		spin_lock(&fs_info->trans_lock);
1958		cur = fs_info->running_transaction;
1959		if (!cur) {
1960			spin_unlock(&fs_info->trans_lock);
1961			goto sleep;
1962		}
1963
1964		now = get_seconds();
1965		if (cur->state < TRANS_STATE_BLOCKED &&
1966		    (now < cur->start_time ||
1967		     now - cur->start_time < fs_info->commit_interval)) {
1968			spin_unlock(&fs_info->trans_lock);
1969			delay = HZ * 5;
 
 
1970			goto sleep;
1971		}
1972		transid = cur->transid;
1973		spin_unlock(&fs_info->trans_lock);
1974
1975		/* If the file system is aborted, this will always fail. */
1976		trans = btrfs_attach_transaction(root);
1977		if (IS_ERR(trans)) {
1978			if (PTR_ERR(trans) != -ENOENT)
1979				cannot_commit = true;
1980			goto sleep;
1981		}
1982		if (transid == trans->transid) {
1983			btrfs_commit_transaction(trans);
1984		} else {
1985			btrfs_end_transaction(trans);
1986		}
1987sleep:
1988		wake_up_process(fs_info->cleaner_kthread);
1989		mutex_unlock(&fs_info->transaction_kthread_mutex);
1990
1991		if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1992				      &fs_info->fs_state)))
1993			btrfs_cleanup_transaction(fs_info);
1994		set_current_state(TASK_INTERRUPTIBLE);
1995		if (!kthread_should_stop() &&
1996				(!btrfs_transaction_blocked(fs_info) ||
1997				 cannot_commit))
1998			schedule_timeout(delay);
1999		__set_current_state(TASK_RUNNING);
2000	} while (!kthread_should_stop());
2001	return 0;
2002}
2003
2004/*
2005 * this will find the highest generation in the array of
2006 * root backups.  The index of the highest array is returned,
2007 * or -1 if we can't find anything.
2008 *
2009 * We check to make sure the array is valid by comparing the
2010 * generation of the latest  root in the array with the generation
2011 * in the super block.  If they don't match we pitch it.
2012 */
2013static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
2014{
 
2015	u64 cur;
2016	int newest_index = -1;
2017	struct btrfs_root_backup *root_backup;
2018	int i;
2019
2020	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2021		root_backup = info->super_copy->super_roots + i;
2022		cur = btrfs_backup_tree_root_gen(root_backup);
2023		if (cur == newest_gen)
2024			newest_index = i;
2025	}
2026
2027	/* check to see if we actually wrapped around */
2028	if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
2029		root_backup = info->super_copy->super_roots;
2030		cur = btrfs_backup_tree_root_gen(root_backup);
2031		if (cur == newest_gen)
2032			newest_index = 0;
2033	}
2034	return newest_index;
2035}
2036
2037
2038/*
2039 * find the oldest backup so we know where to store new entries
2040 * in the backup array.  This will set the backup_root_index
2041 * field in the fs_info struct
2042 */
2043static void find_oldest_super_backup(struct btrfs_fs_info *info,
2044				     u64 newest_gen)
2045{
2046	int newest_index = -1;
2047
2048	newest_index = find_newest_super_backup(info, newest_gen);
2049	/* if there was garbage in there, just move along */
2050	if (newest_index == -1) {
2051		info->backup_root_index = 0;
2052	} else {
2053		info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
2054	}
2055}
2056
2057/*
2058 * copy all the root pointers into the super backup array.
2059 * this will bump the backup pointer by one when it is
2060 * done
2061 */
2062static void backup_super_roots(struct btrfs_fs_info *info)
2063{
2064	int next_backup;
2065	struct btrfs_root_backup *root_backup;
2066	int last_backup;
2067
2068	next_backup = info->backup_root_index;
2069	last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2070		BTRFS_NUM_BACKUP_ROOTS;
2071
2072	/*
2073	 * just overwrite the last backup if we're at the same generation
2074	 * this happens only at umount
2075	 */
2076	root_backup = info->super_for_commit->super_roots + last_backup;
2077	if (btrfs_backup_tree_root_gen(root_backup) ==
2078	    btrfs_header_generation(info->tree_root->node))
2079		next_backup = last_backup;
2080
2081	root_backup = info->super_for_commit->super_roots + next_backup;
2082
2083	/*
2084	 * make sure all of our padding and empty slots get zero filled
2085	 * regardless of which ones we use today
2086	 */
2087	memset(root_backup, 0, sizeof(*root_backup));
2088
2089	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2090
2091	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2092	btrfs_set_backup_tree_root_gen(root_backup,
2093			       btrfs_header_generation(info->tree_root->node));
2094
2095	btrfs_set_backup_tree_root_level(root_backup,
2096			       btrfs_header_level(info->tree_root->node));
2097
2098	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2099	btrfs_set_backup_chunk_root_gen(root_backup,
2100			       btrfs_header_generation(info->chunk_root->node));
2101	btrfs_set_backup_chunk_root_level(root_backup,
2102			       btrfs_header_level(info->chunk_root->node));
2103
2104	btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2105	btrfs_set_backup_extent_root_gen(root_backup,
2106			       btrfs_header_generation(info->extent_root->node));
2107	btrfs_set_backup_extent_root_level(root_backup,
2108			       btrfs_header_level(info->extent_root->node));
 
 
 
 
 
 
 
 
 
 
 
 
2109
2110	/*
2111	 * we might commit during log recovery, which happens before we set
2112	 * the fs_root.  Make sure it is valid before we fill it in.
2113	 */
2114	if (info->fs_root && info->fs_root->node) {
2115		btrfs_set_backup_fs_root(root_backup,
2116					 info->fs_root->node->start);
2117		btrfs_set_backup_fs_root_gen(root_backup,
2118			       btrfs_header_generation(info->fs_root->node));
2119		btrfs_set_backup_fs_root_level(root_backup,
2120			       btrfs_header_level(info->fs_root->node));
2121	}
2122
2123	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2124	btrfs_set_backup_dev_root_gen(root_backup,
2125			       btrfs_header_generation(info->dev_root->node));
2126	btrfs_set_backup_dev_root_level(root_backup,
2127				       btrfs_header_level(info->dev_root->node));
2128
2129	btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2130	btrfs_set_backup_csum_root_gen(root_backup,
2131			       btrfs_header_generation(info->csum_root->node));
2132	btrfs_set_backup_csum_root_level(root_backup,
2133			       btrfs_header_level(info->csum_root->node));
2134
2135	btrfs_set_backup_total_bytes(root_backup,
2136			     btrfs_super_total_bytes(info->super_copy));
2137	btrfs_set_backup_bytes_used(root_backup,
2138			     btrfs_super_bytes_used(info->super_copy));
2139	btrfs_set_backup_num_devices(root_backup,
2140			     btrfs_super_num_devices(info->super_copy));
2141
2142	/*
2143	 * if we don't copy this out to the super_copy, it won't get remembered
2144	 * for the next commit
2145	 */
2146	memcpy(&info->super_copy->super_roots,
2147	       &info->super_for_commit->super_roots,
2148	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2149}
2150
2151/*
2152 * this copies info out of the root backup array and back into
2153 * the in-memory super block.  It is meant to help iterate through
2154 * the array, so you send it the number of backups you've already
2155 * tried and the last backup index you used.
2156 *
2157 * this returns -1 when it has tried all the backups
 
 
 
2158 */
2159static noinline int next_root_backup(struct btrfs_fs_info *info,
2160				     struct btrfs_super_block *super,
2161				     int *num_backups_tried, int *backup_index)
2162{
 
 
2163	struct btrfs_root_backup *root_backup;
2164	int newest = *backup_index;
2165
2166	if (*num_backups_tried == 0) {
2167		u64 gen = btrfs_super_generation(super);
 
2168
2169		newest = find_newest_super_backup(info, gen);
2170		if (newest == -1)
2171			return -1;
2172
2173		*backup_index = newest;
2174		*num_backups_tried = 1;
2175	} else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2176		/* we've tried all the backups, all done */
2177		return -1;
2178	} else {
2179		/* jump to the next oldest backup */
2180		newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2181			BTRFS_NUM_BACKUP_ROOTS;
2182		*backup_index = newest;
2183		*num_backups_tried += 1;
2184	}
2185	root_backup = super->super_roots + newest;
 
2186
2187	btrfs_set_super_generation(super,
2188				   btrfs_backup_tree_root_gen(root_backup));
2189	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2190	btrfs_set_super_root_level(super,
2191				   btrfs_backup_tree_root_level(root_backup));
2192	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2193
2194	/*
2195	 * fixme: the total bytes and num_devices need to match or we should
2196	 * need a fsck
2197	 */
2198	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2199	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2200	return 0;
 
2201}
2202
2203/* helper to cleanup workers */
2204static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2205{
2206	btrfs_destroy_workqueue(fs_info->fixup_workers);
2207	btrfs_destroy_workqueue(fs_info->delalloc_workers);
2208	btrfs_destroy_workqueue(fs_info->workers);
2209	btrfs_destroy_workqueue(fs_info->endio_workers);
2210	btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2211	btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2212	btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2213	btrfs_destroy_workqueue(fs_info->rmw_workers);
2214	btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2215	btrfs_destroy_workqueue(fs_info->endio_write_workers);
2216	btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2217	btrfs_destroy_workqueue(fs_info->submit_workers);
2218	btrfs_destroy_workqueue(fs_info->delayed_workers);
2219	btrfs_destroy_workqueue(fs_info->caching_workers);
2220	btrfs_destroy_workqueue(fs_info->readahead_workers);
2221	btrfs_destroy_workqueue(fs_info->flush_workers);
2222	btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2223	btrfs_destroy_workqueue(fs_info->extent_workers);
 
 
 
 
 
 
 
 
2224}
2225
2226static void free_root_extent_buffers(struct btrfs_root *root)
2227{
2228	if (root) {
2229		free_extent_buffer(root->node);
2230		free_extent_buffer(root->commit_root);
2231		root->node = NULL;
2232		root->commit_root = NULL;
2233	}
2234}
2235
 
 
 
 
 
 
 
 
 
 
2236/* helper to cleanup tree roots */
2237static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2238{
2239	free_root_extent_buffers(info->tree_root);
2240
 
2241	free_root_extent_buffers(info->dev_root);
2242	free_root_extent_buffers(info->extent_root);
2243	free_root_extent_buffers(info->csum_root);
2244	free_root_extent_buffers(info->quota_root);
2245	free_root_extent_buffers(info->uuid_root);
2246	if (chunk_root)
 
 
 
 
2247		free_root_extent_buffers(info->chunk_root);
2248	free_root_extent_buffers(info->free_space_root);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2249}
2250
2251void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2252{
2253	int ret;
2254	struct btrfs_root *gang[8];
2255	int i;
2256
2257	while (!list_empty(&fs_info->dead_roots)) {
2258		gang[0] = list_entry(fs_info->dead_roots.next,
2259				     struct btrfs_root, root_list);
2260		list_del(&gang[0]->root_list);
2261
2262		if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2263			btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2264		} else {
2265			free_extent_buffer(gang[0]->node);
2266			free_extent_buffer(gang[0]->commit_root);
2267			btrfs_put_fs_root(gang[0]);
2268		}
2269	}
2270
2271	while (1) {
2272		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2273					     (void **)gang, 0,
2274					     ARRAY_SIZE(gang));
2275		if (!ret)
2276			break;
2277		for (i = 0; i < ret; i++)
2278			btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2279	}
2280
2281	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2282		btrfs_free_log_root_tree(NULL, fs_info);
2283		btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2284	}
2285}
2286
2287static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2288{
2289	mutex_init(&fs_info->scrub_lock);
2290	atomic_set(&fs_info->scrubs_running, 0);
2291	atomic_set(&fs_info->scrub_pause_req, 0);
2292	atomic_set(&fs_info->scrubs_paused, 0);
2293	atomic_set(&fs_info->scrub_cancel_req, 0);
2294	init_waitqueue_head(&fs_info->scrub_pause_wait);
2295	fs_info->scrub_workers_refcnt = 0;
2296}
2297
2298static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2299{
2300	spin_lock_init(&fs_info->balance_lock);
2301	mutex_init(&fs_info->balance_mutex);
2302	atomic_set(&fs_info->balance_running, 0);
2303	atomic_set(&fs_info->balance_pause_req, 0);
2304	atomic_set(&fs_info->balance_cancel_req, 0);
2305	fs_info->balance_ctl = NULL;
2306	init_waitqueue_head(&fs_info->balance_wait_q);
 
2307}
2308
2309static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2310{
2311	struct inode *inode = fs_info->btree_inode;
 
 
 
 
 
 
 
2312
2313	inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2314	set_nlink(inode, 1);
2315	/*
2316	 * we set the i_size on the btree inode to the max possible int.
2317	 * the real end of the address space is determined by all of
2318	 * the devices in the system
2319	 */
2320	inode->i_size = OFFSET_MAX;
2321	inode->i_mapping->a_ops = &btree_aops;
 
2322
2323	RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2324	extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode->i_mapping);
2325	BTRFS_I(inode)->io_tree.track_uptodate = 0;
2326	extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2327
2328	BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
 
 
 
 
 
 
2329
2330	BTRFS_I(inode)->root = fs_info->tree_root;
2331	memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2332	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2333	btrfs_insert_inode_hash(inode);
2334}
2335
2336static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2337{
2338	fs_info->dev_replace.lock_owner = 0;
2339	atomic_set(&fs_info->dev_replace.nesting_level, 0);
2340	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2341	rwlock_init(&fs_info->dev_replace.lock);
2342	atomic_set(&fs_info->dev_replace.read_locks, 0);
2343	atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2344	init_waitqueue_head(&fs_info->replace_wait);
2345	init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2346}
2347
2348static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2349{
2350	spin_lock_init(&fs_info->qgroup_lock);
2351	mutex_init(&fs_info->qgroup_ioctl_lock);
2352	fs_info->qgroup_tree = RB_ROOT;
2353	fs_info->qgroup_op_tree = RB_ROOT;
2354	INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2355	fs_info->qgroup_seq = 1;
2356	fs_info->qgroup_ulist = NULL;
2357	fs_info->qgroup_rescan_running = false;
 
2358	mutex_init(&fs_info->qgroup_rescan_lock);
2359}
2360
2361static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2362		struct btrfs_fs_devices *fs_devices)
2363{
2364	int max_active = fs_info->thread_pool_size;
2365	unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
 
2366
2367	fs_info->workers =
2368		btrfs_alloc_workqueue(fs_info, "worker",
2369				      flags | WQ_HIGHPRI, max_active, 16);
2370
2371	fs_info->delalloc_workers =
2372		btrfs_alloc_workqueue(fs_info, "delalloc",
2373				      flags, max_active, 2);
2374
2375	fs_info->flush_workers =
2376		btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2377				      flags, max_active, 0);
2378
2379	fs_info->caching_workers =
2380		btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2381
2382	/*
2383	 * a higher idle thresh on the submit workers makes it much more
2384	 * likely that bios will be send down in a sane order to the
2385	 * devices
2386	 */
2387	fs_info->submit_workers =
2388		btrfs_alloc_workqueue(fs_info, "submit", flags,
2389				      min_t(u64, fs_devices->num_devices,
2390					    max_active), 64);
2391
2392	fs_info->fixup_workers =
2393		btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2394
2395	/*
2396	 * endios are largely parallel and should have a very
2397	 * low idle thresh
2398	 */
2399	fs_info->endio_workers =
2400		btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2401	fs_info->endio_meta_workers =
2402		btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2403				      max_active, 4);
2404	fs_info->endio_meta_write_workers =
2405		btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2406				      max_active, 2);
2407	fs_info->endio_raid56_workers =
2408		btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2409				      max_active, 4);
2410	fs_info->endio_repair_workers =
2411		btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2412	fs_info->rmw_workers =
2413		btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2414	fs_info->endio_write_workers =
2415		btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2416				      max_active, 2);
 
 
2417	fs_info->endio_freespace_worker =
2418		btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2419				      max_active, 0);
2420	fs_info->delayed_workers =
2421		btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2422				      max_active, 0);
2423	fs_info->readahead_workers =
2424		btrfs_alloc_workqueue(fs_info, "readahead", flags,
2425				      max_active, 2);
2426	fs_info->qgroup_rescan_workers =
2427		btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2428	fs_info->extent_workers =
2429		btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2430				      min_t(u64, fs_devices->num_devices,
2431					    max_active), 8);
2432
2433	if (!(fs_info->workers && fs_info->delalloc_workers &&
2434	      fs_info->submit_workers && fs_info->flush_workers &&
2435	      fs_info->endio_workers && fs_info->endio_meta_workers &&
2436	      fs_info->endio_meta_write_workers &&
2437	      fs_info->endio_repair_workers &&
2438	      fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2439	      fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2440	      fs_info->caching_workers && fs_info->readahead_workers &&
2441	      fs_info->fixup_workers && fs_info->delayed_workers &&
2442	      fs_info->extent_workers &&
2443	      fs_info->qgroup_rescan_workers)) {
2444		return -ENOMEM;
2445	}
2446
2447	return 0;
2448}
2449
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2450static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2451			    struct btrfs_fs_devices *fs_devices)
2452{
2453	int ret;
 
2454	struct btrfs_root *log_tree_root;
2455	struct btrfs_super_block *disk_super = fs_info->super_copy;
2456	u64 bytenr = btrfs_super_log_root(disk_super);
 
2457
2458	if (fs_devices->rw_devices == 0) {
2459		btrfs_warn(fs_info, "log replay required on RO media");
2460		return -EIO;
2461	}
2462
2463	log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
 
2464	if (!log_tree_root)
2465		return -ENOMEM;
2466
2467	__setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2468
2469	log_tree_root->node = read_tree_block(fs_info, bytenr,
2470					      fs_info->generation + 1);
2471	if (IS_ERR(log_tree_root->node)) {
2472		btrfs_warn(fs_info, "failed to read log tree");
2473		ret = PTR_ERR(log_tree_root->node);
2474		kfree(log_tree_root);
 
2475		return ret;
2476	} else if (!extent_buffer_uptodate(log_tree_root->node)) {
 
2477		btrfs_err(fs_info, "failed to read log tree");
2478		free_extent_buffer(log_tree_root->node);
2479		kfree(log_tree_root);
2480		return -EIO;
2481	}
 
2482	/* returns with log_tree_root freed on success */
2483	ret = btrfs_recover_log_trees(log_tree_root);
2484	if (ret) {
2485		btrfs_handle_fs_error(fs_info, ret,
2486				      "Failed to recover log tree");
2487		free_extent_buffer(log_tree_root->node);
2488		kfree(log_tree_root);
2489		return ret;
2490	}
2491
2492	if (fs_info->sb->s_flags & MS_RDONLY) {
2493		ret = btrfs_commit_super(fs_info);
2494		if (ret)
2495			return ret;
2496	}
2497
2498	return 0;
2499}
2500
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2501static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2502{
2503	struct btrfs_root *tree_root = fs_info->tree_root;
2504	struct btrfs_root *root;
2505	struct btrfs_key location;
2506	int ret;
2507
2508	BUG_ON(!fs_info->tree_root);
2509
2510	location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
 
 
 
2511	location.type = BTRFS_ROOT_ITEM_KEY;
2512	location.offset = 0;
2513
2514	root = btrfs_read_tree_root(tree_root, &location);
2515	if (IS_ERR(root))
2516		return PTR_ERR(root);
2517	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2518	fs_info->extent_root = root;
 
 
 
 
 
 
 
 
2519
2520	location.objectid = BTRFS_DEV_TREE_OBJECTID;
2521	root = btrfs_read_tree_root(tree_root, &location);
2522	if (IS_ERR(root))
2523		return PTR_ERR(root);
2524	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2525	fs_info->dev_root = root;
2526	btrfs_init_devices_late(fs_info);
 
 
 
 
 
 
 
 
2527
2528	location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2529	root = btrfs_read_tree_root(tree_root, &location);
2530	if (IS_ERR(root))
2531		return PTR_ERR(root);
2532	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2533	fs_info->csum_root = root;
 
 
 
 
 
 
 
 
 
2534
2535	location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2536	root = btrfs_read_tree_root(tree_root, &location);
2537	if (!IS_ERR(root)) {
2538		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2539		set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2540		fs_info->quota_root = root;
2541	}
2542
2543	location.objectid = BTRFS_UUID_TREE_OBJECTID;
2544	root = btrfs_read_tree_root(tree_root, &location);
2545	if (IS_ERR(root)) {
2546		ret = PTR_ERR(root);
2547		if (ret != -ENOENT)
2548			return ret;
 
 
2549	} else {
2550		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2551		fs_info->uuid_root = root;
2552	}
2553
2554	if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2555		location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2556		root = btrfs_read_tree_root(tree_root, &location);
2557		if (IS_ERR(root))
2558			return PTR_ERR(root);
2559		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2560		fs_info->free_space_root = root;
 
 
 
 
 
2561	}
2562
2563	return 0;
 
 
 
 
2564}
2565
2566int open_ctree(struct super_block *sb,
2567	       struct btrfs_fs_devices *fs_devices,
2568	       char *options)
 
 
 
 
 
 
 
 
 
2569{
2570	u32 sectorsize;
2571	u32 nodesize;
2572	u32 stripesize;
2573	u64 generation;
2574	u64 features;
2575	struct btrfs_key location;
2576	struct buffer_head *bh;
2577	struct btrfs_super_block *disk_super;
2578	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2579	struct btrfs_root *tree_root;
2580	struct btrfs_root *chunk_root;
2581	int ret;
2582	int err = -EINVAL;
2583	int num_backups_tried = 0;
2584	int backup_index = 0;
2585	int max_active;
2586	int clear_free_space_tree = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2587
2588	tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2589	chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2590	if (!tree_root || !chunk_root) {
2591		err = -ENOMEM;
2592		goto fail;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2593	}
2594
2595	ret = init_srcu_struct(&fs_info->subvol_srcu);
2596	if (ret) {
2597		err = ret;
2598		goto fail;
 
2599	}
2600
2601	ret = setup_bdi(fs_info, &fs_info->bdi);
2602	if (ret) {
2603		err = ret;
2604		goto fail_srcu;
 
 
 
 
 
 
 
 
 
 
 
 
 
2605	}
2606
2607	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2608	if (ret) {
2609		err = ret;
2610		goto fail_bdi;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2611	}
2612	fs_info->dirty_metadata_batch = PAGE_SIZE *
2613					(1 + ilog2(nr_cpu_ids));
 
 
 
 
 
 
 
 
 
 
 
 
 
2614
2615	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2616	if (ret) {
2617		err = ret;
2618		goto fail_dirty_metadata_bytes;
 
 
 
 
 
 
2619	}
2620
2621	ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2622	if (ret) {
2623		err = ret;
2624		goto fail_delalloc_bytes;
2625	}
 
 
 
 
 
 
 
 
 
 
 
2626
2627	fs_info->btree_inode = new_inode(sb);
2628	if (!fs_info->btree_inode) {
2629		err = -ENOMEM;
2630		goto fail_bio_counter;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2631	}
2632
2633	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
 
2634
 
 
2635	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2636	INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2637	INIT_LIST_HEAD(&fs_info->trans_list);
2638	INIT_LIST_HEAD(&fs_info->dead_roots);
2639	INIT_LIST_HEAD(&fs_info->delayed_iputs);
2640	INIT_LIST_HEAD(&fs_info->delalloc_roots);
2641	INIT_LIST_HEAD(&fs_info->caching_block_groups);
2642	spin_lock_init(&fs_info->delalloc_root_lock);
2643	spin_lock_init(&fs_info->trans_lock);
2644	spin_lock_init(&fs_info->fs_roots_radix_lock);
2645	spin_lock_init(&fs_info->delayed_iput_lock);
2646	spin_lock_init(&fs_info->defrag_inodes_lock);
2647	spin_lock_init(&fs_info->free_chunk_lock);
2648	spin_lock_init(&fs_info->tree_mod_seq_lock);
2649	spin_lock_init(&fs_info->super_lock);
2650	spin_lock_init(&fs_info->qgroup_op_lock);
2651	spin_lock_init(&fs_info->buffer_lock);
2652	spin_lock_init(&fs_info->unused_bgs_lock);
 
 
 
2653	rwlock_init(&fs_info->tree_mod_log_lock);
 
2654	mutex_init(&fs_info->unused_bg_unpin_mutex);
2655	mutex_init(&fs_info->delete_unused_bgs_mutex);
2656	mutex_init(&fs_info->reloc_mutex);
2657	mutex_init(&fs_info->delalloc_root_mutex);
2658	mutex_init(&fs_info->cleaner_delayed_iput_mutex);
 
2659	seqlock_init(&fs_info->profiles_lock);
2660
 
 
 
 
 
 
 
 
 
 
 
 
 
2661	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2662	INIT_LIST_HEAD(&fs_info->space_info);
2663	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2664	INIT_LIST_HEAD(&fs_info->unused_bgs);
2665	btrfs_mapping_init(&fs_info->mapping_tree);
 
 
 
 
 
 
 
 
2666	btrfs_init_block_rsv(&fs_info->global_block_rsv,
2667			     BTRFS_BLOCK_RSV_GLOBAL);
2668	btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2669			     BTRFS_BLOCK_RSV_DELALLOC);
2670	btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2671	btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2672	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2673	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2674			     BTRFS_BLOCK_RSV_DELOPS);
2675	atomic_set(&fs_info->nr_async_submits, 0);
 
 
2676	atomic_set(&fs_info->async_delalloc_pages, 0);
2677	atomic_set(&fs_info->async_submit_draining, 0);
2678	atomic_set(&fs_info->nr_async_bios, 0);
2679	atomic_set(&fs_info->defrag_running, 0);
2680	atomic_set(&fs_info->qgroup_op_seq, 0);
2681	atomic_set(&fs_info->reada_works_cnt, 0);
2682	atomic64_set(&fs_info->tree_mod_seq, 0);
2683	fs_info->fs_frozen = 0;
2684	fs_info->sb = sb;
2685	fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2686	fs_info->metadata_ratio = 0;
2687	fs_info->defrag_inodes = RB_ROOT;
2688	fs_info->free_chunk_space = 0;
2689	fs_info->tree_mod_log = RB_ROOT;
2690	fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2691	fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2692	/* readahead state */
2693	INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2694	spin_lock_init(&fs_info->reada_lock);
2695
2696	fs_info->thread_pool_size = min_t(unsigned long,
2697					  num_online_cpus() + 2, 8);
2698
2699	INIT_LIST_HEAD(&fs_info->ordered_roots);
2700	spin_lock_init(&fs_info->ordered_root_lock);
2701	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2702					GFP_KERNEL);
2703	if (!fs_info->delayed_root) {
2704		err = -ENOMEM;
2705		goto fail_iput;
2706	}
2707	btrfs_init_delayed_root(fs_info->delayed_root);
2708
2709	btrfs_init_scrub(fs_info);
2710#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2711	fs_info->check_integrity_print_mask = 0;
2712#endif
2713	btrfs_init_balance(fs_info);
2714	btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
 
 
 
2715
2716	sb->s_blocksize = 4096;
2717	sb->s_blocksize_bits = blksize_bits(4096);
2718	sb->s_bdi = &fs_info->bdi;
2719
2720	btrfs_init_btree_inode(fs_info);
2721
2722	spin_lock_init(&fs_info->block_group_cache_lock);
2723	fs_info->block_group_cache_tree = RB_ROOT;
2724	fs_info->first_logical_byte = (u64)-1;
2725
2726	extent_io_tree_init(&fs_info->freed_extents[0],
2727			     fs_info->btree_inode->i_mapping);
2728	extent_io_tree_init(&fs_info->freed_extents[1],
2729			     fs_info->btree_inode->i_mapping);
2730	fs_info->pinned_extents = &fs_info->freed_extents[0];
2731	set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2732
2733	mutex_init(&fs_info->ordered_operations_mutex);
2734	mutex_init(&fs_info->tree_log_mutex);
2735	mutex_init(&fs_info->chunk_mutex);
2736	mutex_init(&fs_info->transaction_kthread_mutex);
2737	mutex_init(&fs_info->cleaner_mutex);
2738	mutex_init(&fs_info->volume_mutex);
2739	mutex_init(&fs_info->ro_block_group_mutex);
2740	init_rwsem(&fs_info->commit_root_sem);
2741	init_rwsem(&fs_info->cleanup_work_sem);
2742	init_rwsem(&fs_info->subvol_sem);
2743	sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2744
2745	btrfs_init_dev_replace_locks(fs_info);
2746	btrfs_init_qgroup(fs_info);
 
2747
2748	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2749	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2750
2751	init_waitqueue_head(&fs_info->transaction_throttle);
2752	init_waitqueue_head(&fs_info->transaction_wait);
2753	init_waitqueue_head(&fs_info->transaction_blocked_wait);
2754	init_waitqueue_head(&fs_info->async_submit_wait);
2755
2756	INIT_LIST_HEAD(&fs_info->pinned_chunks);
2757
2758	/* Usable values until the real ones are cached from the superblock */
2759	fs_info->nodesize = 4096;
2760	fs_info->sectorsize = 4096;
 
2761	fs_info->stripesize = 4096;
2762
2763	ret = btrfs_alloc_stripe_hash_table(fs_info);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2764	if (ret) {
2765		err = ret;
2766		goto fail_alloc;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2767	}
2768
2769	__setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2770
2771	invalidate_bdev(fs_devices->latest_bdev);
 
 
 
 
 
2772
2773	/*
2774	 * Read super block and check the signature bytes only
 
 
2775	 */
2776	bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2777	if (IS_ERR(bh)) {
2778		err = PTR_ERR(bh);
2779		goto fail_alloc;
 
 
2780	}
2781
2782	/*
2783	 * We want to check superblock checksum, the type is stored inside.
2784	 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2785	 */
2786	if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2787		btrfs_err(fs_info, "superblock checksum mismatch");
2788		err = -EINVAL;
2789		brelse(bh);
2790		goto fail_alloc;
 
2791	}
2792
2793	/*
2794	 * super_copy is zeroed at allocation time and we never touch the
2795	 * following bytes up to INFO_SIZE, the checksum is calculated from
2796	 * the whole block of INFO_SIZE
 
 
2797	 */
2798	memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2799	memcpy(fs_info->super_for_commit, fs_info->super_copy,
2800	       sizeof(*fs_info->super_for_commit));
2801	brelse(bh);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2802
2803	memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
 
 
2804
2805	ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2806	if (ret) {
2807		btrfs_err(fs_info, "superblock contains fatal errors");
2808		err = -EINVAL;
2809		goto fail_alloc;
 
 
 
 
 
2810	}
2811
2812	disk_super = fs_info->super_copy;
2813	if (!btrfs_super_root(disk_super))
2814		goto fail_alloc;
2815
2816	/* check FS state, whether FS is broken. */
2817	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2818		set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2819
2820	/*
2821	 * run through our array of backup supers and setup
2822	 * our ring pointer to the oldest one
2823	 */
2824	generation = btrfs_super_generation(disk_super);
2825	find_oldest_super_backup(fs_info, generation);
 
 
 
2826
 
2827	/*
2828	 * In the long term, we'll store the compression type in the super
2829	 * block, and it'll be used for per file compression control.
2830	 */
2831	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
 
 
 
 
 
 
 
 
 
2832
2833	ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2834	if (ret) {
2835		err = ret;
2836		goto fail_alloc;
2837	}
2838
2839	features = btrfs_super_incompat_flags(disk_super) &
2840		~BTRFS_FEATURE_INCOMPAT_SUPP;
2841	if (features) {
2842		btrfs_err(fs_info,
2843		    "cannot mount because of unsupported optional features (%llx)",
2844		    features);
2845		err = -EINVAL;
 
2846		goto fail_alloc;
2847	}
2848
2849	features = btrfs_super_incompat_flags(disk_super);
2850	features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2851	if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2852		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
 
 
 
 
 
2853
2854	if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2855		btrfs_info(fs_info, "has skinny extents");
2856
2857	/*
2858	 * flag our filesystem as having big metadata blocks if
2859	 * they are bigger than the page size
2860	 */
2861	if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2862		if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2863			btrfs_info(fs_info,
2864				"flagging fs with big metadata feature");
2865		features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
 
 
2866	}
2867
 
 
 
 
 
2868	nodesize = btrfs_super_nodesize(disk_super);
2869	sectorsize = btrfs_super_sectorsize(disk_super);
2870	stripesize = sectorsize;
2871	fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2872	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2873
2874	/* Cache block sizes */
2875	fs_info->nodesize = nodesize;
2876	fs_info->sectorsize = sectorsize;
 
 
2877	fs_info->stripesize = stripesize;
2878
2879	/*
2880	 * mixed block groups end up with duplicate but slightly offset
2881	 * extent buffers for the same range.  It leads to corruptions
2882	 */
2883	if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2884	    (sectorsize != nodesize)) {
2885		btrfs_err(fs_info,
2886"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2887			nodesize, sectorsize);
2888		goto fail_alloc;
2889	}
2890
 
 
 
 
2891	/*
2892	 * Needn't use the lock because there is no other task which will
2893	 * update the flag.
2894	 */
2895	btrfs_set_super_incompat_flags(disk_super, features);
 
 
 
2896
2897	features = btrfs_super_compat_ro_flags(disk_super) &
2898		~BTRFS_FEATURE_COMPAT_RO_SUPP;
2899	if (!(sb->s_flags & MS_RDONLY) && features) {
2900		btrfs_err(fs_info,
2901	"cannot mount read-write because of unsupported optional features (%llx)",
2902		       features);
2903		err = -EINVAL;
2904		goto fail_alloc;
 
 
2905	}
2906
2907	max_active = fs_info->thread_pool_size;
2908
2909	ret = btrfs_init_workqueues(fs_info, fs_devices);
2910	if (ret) {
2911		err = ret;
2912		goto fail_sb_buffer;
2913	}
2914
2915	fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2916	fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2917				    SZ_4M / PAGE_SIZE);
2918
2919	sb->s_blocksize = sectorsize;
2920	sb->s_blocksize_bits = blksize_bits(sectorsize);
 
2921
2922	mutex_lock(&fs_info->chunk_mutex);
2923	ret = btrfs_read_sys_array(fs_info);
2924	mutex_unlock(&fs_info->chunk_mutex);
2925	if (ret) {
2926		btrfs_err(fs_info, "failed to read the system array: %d", ret);
2927		goto fail_sb_buffer;
2928	}
2929
2930	generation = btrfs_super_chunk_root_generation(disk_super);
2931
2932	__setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2933
2934	chunk_root->node = read_tree_block(fs_info,
2935					   btrfs_super_chunk_root(disk_super),
2936					   generation);
2937	if (IS_ERR(chunk_root->node) ||
2938	    !extent_buffer_uptodate(chunk_root->node)) {
2939		btrfs_err(fs_info, "failed to read chunk root");
2940		if (!IS_ERR(chunk_root->node))
2941			free_extent_buffer(chunk_root->node);
2942		chunk_root->node = NULL;
2943		goto fail_tree_roots;
2944	}
2945	btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2946	chunk_root->commit_root = btrfs_root_node(chunk_root);
2947
2948	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2949	   btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
 
2950
2951	ret = btrfs_read_chunk_tree(fs_info);
2952	if (ret) {
2953		btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2954		goto fail_tree_roots;
2955	}
2956
2957	/*
2958	 * keep the device that is marked to be the target device for the
2959	 * dev_replace procedure
 
 
 
2960	 */
2961	btrfs_close_extra_devices(fs_devices, 0);
2962
2963	if (!fs_devices->latest_bdev) {
2964		btrfs_err(fs_info, "failed to read devices");
 
2965		goto fail_tree_roots;
2966	}
2967
2968retry_root_backup:
2969	generation = btrfs_super_generation(disk_super);
 
2970
2971	tree_root->node = read_tree_block(fs_info,
2972					  btrfs_super_root(disk_super),
2973					  generation);
2974	if (IS_ERR(tree_root->node) ||
2975	    !extent_buffer_uptodate(tree_root->node)) {
2976		btrfs_warn(fs_info, "failed to read tree root");
2977		if (!IS_ERR(tree_root->node))
2978			free_extent_buffer(tree_root->node);
2979		tree_root->node = NULL;
2980		goto recovery_tree_root;
2981	}
2982
2983	btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2984	tree_root->commit_root = btrfs_root_node(tree_root);
2985	btrfs_set_root_refs(&tree_root->root_item, 1);
2986
2987	mutex_lock(&tree_root->objectid_mutex);
2988	ret = btrfs_find_highest_objectid(tree_root,
2989					&tree_root->highest_objectid);
2990	if (ret) {
2991		mutex_unlock(&tree_root->objectid_mutex);
2992		goto recovery_tree_root;
 
2993	}
2994
2995	ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2996
2997	mutex_unlock(&tree_root->objectid_mutex);
2998
2999	ret = btrfs_read_roots(fs_info);
3000	if (ret)
3001		goto recovery_tree_root;
3002
3003	fs_info->generation = generation;
3004	fs_info->last_trans_committed = generation;
 
3005
 
 
 
 
 
 
 
3006	ret = btrfs_recover_balance(fs_info);
3007	if (ret) {
3008		btrfs_err(fs_info, "failed to recover balance: %d", ret);
3009		goto fail_block_groups;
3010	}
3011
3012	ret = btrfs_init_dev_stats(fs_info);
3013	if (ret) {
3014		btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3015		goto fail_block_groups;
3016	}
3017
3018	ret = btrfs_init_dev_replace(fs_info);
3019	if (ret) {
3020		btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3021		goto fail_block_groups;
3022	}
3023
3024	btrfs_close_extra_devices(fs_devices, 1);
3025
3026	ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3027	if (ret) {
3028		btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3029				ret);
3030		goto fail_block_groups;
3031	}
3032
3033	ret = btrfs_sysfs_add_device(fs_devices);
3034	if (ret) {
3035		btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3036				ret);
3037		goto fail_fsdev_sysfs;
3038	}
3039
3040	ret = btrfs_sysfs_add_mounted(fs_info);
3041	if (ret) {
3042		btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3043		goto fail_fsdev_sysfs;
3044	}
3045
3046	ret = btrfs_init_space_info(fs_info);
3047	if (ret) {
3048		btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3049		goto fail_sysfs;
3050	}
3051
3052	ret = btrfs_read_block_groups(fs_info);
3053	if (ret) {
3054		btrfs_err(fs_info, "failed to read block groups: %d", ret);
3055		goto fail_sysfs;
3056	}
3057	fs_info->num_tolerated_disk_barrier_failures =
3058		btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3059	if (fs_info->fs_devices->missing_devices >
3060	     fs_info->num_tolerated_disk_barrier_failures &&
3061	    !(sb->s_flags & MS_RDONLY)) {
 
 
3062		btrfs_warn(fs_info,
3063"missing devices (%llu) exceeds the limit (%d), writeable mount is not allowed",
3064			fs_info->fs_devices->missing_devices,
3065			fs_info->num_tolerated_disk_barrier_failures);
3066		goto fail_sysfs;
3067	}
3068
3069	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3070					       "btrfs-cleaner");
3071	if (IS_ERR(fs_info->cleaner_kthread))
 
3072		goto fail_sysfs;
 
3073
3074	fs_info->transaction_kthread = kthread_run(transaction_kthread,
3075						   tree_root,
3076						   "btrfs-transaction");
3077	if (IS_ERR(fs_info->transaction_kthread))
 
3078		goto fail_cleaner;
3079
3080	if (!btrfs_test_opt(fs_info, SSD) &&
3081	    !btrfs_test_opt(fs_info, NOSSD) &&
3082	    !fs_info->fs_devices->rotating) {
3083		btrfs_info(fs_info, "detected SSD devices, enabling SSD mode");
3084		btrfs_set_opt(fs_info->mount_opt, SSD);
3085	}
3086
3087	/*
3088	 * Mount does not set all options immediately, we can do it now and do
3089	 * not have to wait for transaction commit
3090	 */
3091	btrfs_apply_pending_changes(fs_info);
3092
3093#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3094	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3095		ret = btrfsic_mount(fs_info, fs_devices,
3096				    btrfs_test_opt(fs_info,
3097					CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3098				    1 : 0,
3099				    fs_info->check_integrity_print_mask);
3100		if (ret)
3101			btrfs_warn(fs_info,
3102				"failed to initialize integrity check module: %d",
3103				ret);
3104	}
3105#endif
3106	ret = btrfs_read_qgroup_config(fs_info);
3107	if (ret)
3108		goto fail_trans_kthread;
3109
 
 
 
3110	/* do not make disk changes in broken FS or nologreplay is given */
3111	if (btrfs_super_log_root(disk_super) != 0 &&
3112	    !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
 
3113		ret = btrfs_replay_log(fs_info, fs_devices);
3114		if (ret) {
3115			err = ret;
3116			goto fail_qgroup;
3117		}
3118	}
3119
3120	ret = btrfs_find_orphan_roots(fs_info);
3121	if (ret)
3122		goto fail_qgroup;
3123
3124	if (!(sb->s_flags & MS_RDONLY)) {
3125		ret = btrfs_cleanup_fs_roots(fs_info);
3126		if (ret)
3127			goto fail_qgroup;
3128
3129		mutex_lock(&fs_info->cleaner_mutex);
3130		ret = btrfs_recover_relocation(tree_root);
3131		mutex_unlock(&fs_info->cleaner_mutex);
3132		if (ret < 0) {
3133			btrfs_warn(fs_info, "failed to recover relocation: %d",
3134					ret);
3135			err = -EINVAL;
3136			goto fail_qgroup;
3137		}
3138	}
3139
3140	location.objectid = BTRFS_FS_TREE_OBJECTID;
3141	location.type = BTRFS_ROOT_ITEM_KEY;
3142	location.offset = 0;
3143
3144	fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3145	if (IS_ERR(fs_info->fs_root)) {
3146		err = PTR_ERR(fs_info->fs_root);
 
 
3147		goto fail_qgroup;
3148	}
3149
3150	if (sb->s_flags & MS_RDONLY)
3151		return 0;
3152
3153	if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3154	    btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3155		clear_free_space_tree = 1;
3156	} else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3157		   !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3158		btrfs_warn(fs_info, "free space tree is invalid");
3159		clear_free_space_tree = 1;
3160	}
3161
3162	if (clear_free_space_tree) {
3163		btrfs_info(fs_info, "clearing free space tree");
3164		ret = btrfs_clear_free_space_tree(fs_info);
3165		if (ret) {
3166			btrfs_warn(fs_info,
3167				   "failed to clear free space tree: %d", ret);
3168			close_ctree(fs_info);
3169			return ret;
3170		}
3171	}
3172
3173	if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3174	    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3175		btrfs_info(fs_info, "creating free space tree");
3176		ret = btrfs_create_free_space_tree(fs_info);
3177		if (ret) {
3178			btrfs_warn(fs_info,
3179				"failed to create free space tree: %d", ret);
3180			close_ctree(fs_info);
3181			return ret;
3182		}
3183	}
3184
3185	down_read(&fs_info->cleanup_work_sem);
3186	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3187	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3188		up_read(&fs_info->cleanup_work_sem);
3189		close_ctree(fs_info);
3190		return ret;
3191	}
3192	up_read(&fs_info->cleanup_work_sem);
3193
3194	ret = btrfs_resume_balance_async(fs_info);
3195	if (ret) {
3196		btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3197		close_ctree(fs_info);
3198		return ret;
3199	}
 
3200
3201	ret = btrfs_resume_dev_replace_async(fs_info);
3202	if (ret) {
3203		btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3204		close_ctree(fs_info);
3205		return ret;
3206	}
3207
3208	btrfs_qgroup_rescan_resume(fs_info);
3209
3210	if (!fs_info->uuid_root) {
3211		btrfs_info(fs_info, "creating UUID tree");
3212		ret = btrfs_create_uuid_tree(fs_info);
3213		if (ret) {
3214			btrfs_warn(fs_info,
3215				"failed to create the UUID tree: %d", ret);
3216			close_ctree(fs_info);
3217			return ret;
3218		}
3219	} else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3220		   fs_info->generation !=
3221				btrfs_super_uuid_tree_generation(disk_super)) {
3222		btrfs_info(fs_info, "checking UUID tree");
3223		ret = btrfs_check_uuid_tree(fs_info);
3224		if (ret) {
3225			btrfs_warn(fs_info,
3226				"failed to check the UUID tree: %d", ret);
3227			close_ctree(fs_info);
3228			return ret;
3229		}
3230	} else {
3231		set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3232	}
 
3233	set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3234
3235	/*
3236	 * backuproot only affect mount behavior, and if open_ctree succeeded,
3237	 * no need to keep the flag
3238	 */
3239	btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3240
3241	return 0;
3242
3243fail_qgroup:
3244	btrfs_free_qgroup_config(fs_info);
3245fail_trans_kthread:
3246	kthread_stop(fs_info->transaction_kthread);
3247	btrfs_cleanup_transaction(fs_info);
3248	btrfs_free_fs_roots(fs_info);
3249fail_cleaner:
3250	kthread_stop(fs_info->cleaner_kthread);
3251
3252	/*
3253	 * make sure we're done with the btree inode before we stop our
3254	 * kthreads
3255	 */
3256	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3257
3258fail_sysfs:
3259	btrfs_sysfs_remove_mounted(fs_info);
3260
3261fail_fsdev_sysfs:
3262	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3263
3264fail_block_groups:
3265	btrfs_put_block_group_cache(fs_info);
3266	btrfs_free_block_groups(fs_info);
3267
3268fail_tree_roots:
3269	free_root_pointers(fs_info, 1);
 
 
3270	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3271
3272fail_sb_buffer:
3273	btrfs_stop_all_workers(fs_info);
 
3274fail_alloc:
3275fail_iput:
3276	btrfs_mapping_tree_free(&fs_info->mapping_tree);
3277
3278	iput(fs_info->btree_inode);
3279fail_bio_counter:
3280	percpu_counter_destroy(&fs_info->bio_counter);
3281fail_delalloc_bytes:
3282	percpu_counter_destroy(&fs_info->delalloc_bytes);
3283fail_dirty_metadata_bytes:
3284	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3285fail_bdi:
3286	bdi_destroy(&fs_info->bdi);
3287fail_srcu:
3288	cleanup_srcu_struct(&fs_info->subvol_srcu);
3289fail:
3290	btrfs_free_stripe_hash_table(fs_info);
3291	btrfs_close_devices(fs_info->fs_devices);
3292	return err;
 
 
 
3293
3294recovery_tree_root:
3295	if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3296		goto fail_tree_roots;
 
 
 
3297
3298	free_root_pointers(fs_info, 0);
 
3299
3300	/* don't use the log in recovery mode, it won't be valid */
3301	btrfs_set_super_log_root(disk_super, 0);
 
 
 
 
 
 
 
 
 
 
3302
3303	/* we can't trust the free space cache either */
3304	btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
 
3305
3306	ret = next_root_backup(fs_info, fs_info->super_copy,
3307			       &num_backups_tried, &backup_index);
3308	if (ret == -1)
3309		goto fail_block_groups;
3310	goto retry_root_backup;
3311}
3312
3313static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
 
3314{
3315	if (uptodate) {
3316		set_buffer_uptodate(bh);
3317	} else {
3318		struct btrfs_device *device = (struct btrfs_device *)
3319			bh->b_private;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3320
3321		btrfs_warn_rl_in_rcu(device->fs_info,
3322				"lost page write due to IO error on %s",
3323					  rcu_str_deref(device->name));
3324		/* note, we don't set_buffer_write_io_error because we have
3325		 * our own ways of dealing with the IO errors
3326		 */
3327		clear_buffer_uptodate(bh);
3328		btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
 
3329	}
3330	unlock_buffer(bh);
3331	put_bh(bh);
3332}
3333
3334int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3335			struct buffer_head **bh_ret)
3336{
3337	struct buffer_head *bh;
3338	struct btrfs_super_block *super;
3339	u64 bytenr;
3340
3341	bytenr = btrfs_sb_offset(copy_num);
3342	if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3343		return -EINVAL;
 
 
3344
3345	bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3346	/*
3347	 * If we fail to read from the underlying devices, as of now
3348	 * the best option we have is to mark it EIO.
3349	 */
3350	if (!bh)
3351		return -EIO;
3352
3353	super = (struct btrfs_super_block *)bh->b_data;
3354	if (btrfs_super_bytenr(super) != bytenr ||
3355		    btrfs_super_magic(super) != BTRFS_MAGIC) {
3356		brelse(bh);
3357		return -EINVAL;
3358	}
3359
3360	*bh_ret = bh;
3361	return 0;
3362}
3363
3364
3365struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3366{
3367	struct buffer_head *bh;
3368	struct buffer_head *latest = NULL;
3369	struct btrfs_super_block *super;
3370	int i;
3371	u64 transid = 0;
3372	int ret = -EINVAL;
3373
3374	/* we would like to check all the supers, but that would make
3375	 * a btrfs mount succeed after a mkfs from a different FS.
3376	 * So, we need to add a special mount option to scan for
3377	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3378	 */
3379	for (i = 0; i < 1; i++) {
3380		ret = btrfs_read_dev_one_super(bdev, i, &bh);
3381		if (ret)
3382			continue;
3383
3384		super = (struct btrfs_super_block *)bh->b_data;
 
 
3385
3386		if (!latest || btrfs_super_generation(super) > transid) {
3387			brelse(latest);
3388			latest = bh;
3389			transid = btrfs_super_generation(super);
3390		} else {
3391			brelse(bh);
3392		}
3393	}
3394
3395	if (!latest)
3396		return ERR_PTR(ret);
3397
3398	return latest;
3399}
3400
3401/*
3402 * this should be called twice, once with wait == 0 and
3403 * once with wait == 1.  When wait == 0 is done, all the buffer heads
3404 * we write are pinned.
3405 *
3406 * They are released when wait == 1 is done.
3407 * max_mirrors must be the same for both runs, and it indicates how
3408 * many supers on this one device should be written.
3409 *
3410 * max_mirrors == 0 means to write them all.
3411 */
3412static int write_dev_supers(struct btrfs_device *device,
3413			    struct btrfs_super_block *sb,
3414			    int do_barriers, int wait, int max_mirrors)
3415{
3416	struct buffer_head *bh;
 
 
3417	int i;
 
3418	int ret;
3419	int errors = 0;
3420	u32 crc;
3421	u64 bytenr;
3422
3423	if (max_mirrors == 0)
3424		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3425
 
 
3426	for (i = 0; i < max_mirrors; i++) {
3427		bytenr = btrfs_sb_offset(i);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3428		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3429		    device->commit_total_bytes)
3430			break;
3431
3432		if (wait) {
3433			bh = __find_get_block(device->bdev, bytenr / 4096,
3434					      BTRFS_SUPER_INFO_SIZE);
3435			if (!bh) {
3436				errors++;
3437				continue;
3438			}
3439			wait_on_buffer(bh);
3440			if (!buffer_uptodate(bh))
3441				errors++;
 
 
 
 
 
 
 
 
3442
3443			/* drop our reference */
3444			brelse(bh);
3445
3446			/* drop the reference from the wait == 0 run */
3447			brelse(bh);
3448			continue;
3449		} else {
3450			btrfs_set_super_bytenr(sb, bytenr);
 
 
 
 
 
 
 
 
3451
3452			crc = ~(u32)0;
3453			crc = btrfs_csum_data((char *)sb +
3454					      BTRFS_CSUM_SIZE, crc,
3455					      BTRFS_SUPER_INFO_SIZE -
3456					      BTRFS_CSUM_SIZE);
3457			btrfs_csum_final(crc, sb->csum);
 
 
3458
3459			/*
3460			 * one reference for us, and we leave it for the
3461			 * caller
3462			 */
3463			bh = __getblk(device->bdev, bytenr / 4096,
3464				      BTRFS_SUPER_INFO_SIZE);
3465			if (!bh) {
3466				btrfs_err(device->fs_info,
3467				    "couldn't get super buffer head for bytenr %llu",
3468				    bytenr);
3469				errors++;
3470				continue;
3471			}
 
 
 
 
 
 
 
3472
3473			memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
 
3474
3475			/* one reference for submit_bh */
3476			get_bh(bh);
3477
3478			set_buffer_uptodate(bh);
3479			lock_buffer(bh);
3480			bh->b_end_io = btrfs_end_buffer_write_sync;
3481			bh->b_private = device;
 
 
 
 
3482		}
 
 
 
3483
3484		/*
3485		 * we fua the first super.  The others we allow
3486		 * to go down lazy.
3487		 */
3488		if (i == 0)
3489			ret = btrfsic_submit_bh(REQ_OP_WRITE, REQ_FUA, bh);
3490		else
3491			ret = btrfsic_submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
3492		if (ret)
 
 
3493			errors++;
 
 
 
 
 
 
 
 
 
3494	}
 
 
 
 
 
 
 
 
3495	return errors < i ? 0 : -1;
3496}
3497
3498/*
3499 * endio for the write_dev_flush, this will wake anyone waiting
3500 * for the barrier when it is done
3501 */
3502static void btrfs_end_empty_barrier(struct bio *bio)
3503{
3504	if (bio->bi_private)
3505		complete(bio->bi_private);
3506	bio_put(bio);
3507}
3508
3509/*
3510 * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3511 * sent down.  With wait == 1, it waits for the previous flush.
3512 *
3513 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3514 * capable
3515 */
3516static int write_dev_flush(struct btrfs_device *device, int wait)
3517{
3518	struct bio *bio;
3519	int ret = 0;
3520
3521	if (device->nobarriers)
3522		return 0;
3523
3524	if (wait) {
3525		bio = device->flush_bio;
3526		if (!bio)
3527			return 0;
 
 
 
 
3528
3529		wait_for_completion(&device->flush_wait);
 
 
 
 
 
 
3530
3531		if (bio->bi_error) {
3532			ret = bio->bi_error;
3533			btrfs_dev_stat_inc_and_print(device,
3534				BTRFS_DEV_STAT_FLUSH_ERRS);
3535		}
3536
3537		/* drop the reference from the wait == 0 run */
3538		bio_put(bio);
3539		device->flush_bio = NULL;
3540
3541		return ret;
 
 
 
3542	}
3543
3544	/*
3545	 * one reference for us, and we leave it for the
3546	 * caller
3547	 */
3548	device->flush_bio = NULL;
3549	bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3550	if (!bio)
3551		return -ENOMEM;
3552
3553	bio->bi_end_io = btrfs_end_empty_barrier;
3554	bio->bi_bdev = device->bdev;
3555	bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
3556	init_completion(&device->flush_wait);
3557	bio->bi_private = &device->flush_wait;
3558	device->flush_bio = bio;
3559
3560	bio_get(bio);
3561	btrfsic_submit_bio(bio);
3562
3563	return 0;
3564}
3565
3566/*
3567 * send an empty flush down to each device in parallel,
3568 * then wait for them
3569 */
3570static int barrier_all_devices(struct btrfs_fs_info *info)
3571{
3572	struct list_head *head;
3573	struct btrfs_device *dev;
3574	int errors_send = 0;
3575	int errors_wait = 0;
3576	int ret;
3577
 
3578	/* send down all the barriers */
3579	head = &info->fs_devices->devices;
3580	list_for_each_entry_rcu(dev, head, dev_list) {
3581		if (dev->missing)
3582			continue;
3583		if (!dev->bdev) {
3584			errors_send++;
3585			continue;
3586		}
3587		if (!dev->in_fs_metadata || !dev->writeable)
3588			continue;
3589
3590		ret = write_dev_flush(dev, 0);
3591		if (ret)
3592			errors_send++;
3593	}
3594
3595	/* wait for all the barriers */
3596	list_for_each_entry_rcu(dev, head, dev_list) {
3597		if (dev->missing)
3598			continue;
3599		if (!dev->bdev) {
3600			errors_wait++;
3601			continue;
3602		}
3603		if (!dev->in_fs_metadata || !dev->writeable)
 
3604			continue;
3605
3606		ret = write_dev_flush(dev, 1);
3607		if (ret)
3608			errors_wait++;
3609	}
3610	if (errors_send > info->num_tolerated_disk_barrier_failures ||
3611	    errors_wait > info->num_tolerated_disk_barrier_failures)
 
 
 
 
3612		return -EIO;
 
3613	return 0;
3614}
3615
3616int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3617{
3618	int raid_type;
3619	int min_tolerated = INT_MAX;
3620
3621	if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3622	    (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3623		min_tolerated = min(min_tolerated,
3624				    btrfs_raid_array[BTRFS_RAID_SINGLE].
3625				    tolerated_failures);
3626
3627	for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3628		if (raid_type == BTRFS_RAID_SINGLE)
3629			continue;
3630		if (!(flags & btrfs_raid_group[raid_type]))
3631			continue;
3632		min_tolerated = min(min_tolerated,
3633				    btrfs_raid_array[raid_type].
3634				    tolerated_failures);
3635	}
3636
3637	if (min_tolerated == INT_MAX) {
3638		pr_warn("BTRFS: unknown raid flag: %llu", flags);
3639		min_tolerated = 0;
3640	}
3641
3642	return min_tolerated;
3643}
3644
3645int btrfs_calc_num_tolerated_disk_barrier_failures(
3646	struct btrfs_fs_info *fs_info)
3647{
3648	struct btrfs_ioctl_space_info space;
3649	struct btrfs_space_info *sinfo;
3650	u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3651		       BTRFS_BLOCK_GROUP_SYSTEM,
3652		       BTRFS_BLOCK_GROUP_METADATA,
3653		       BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3654	int i;
3655	int c;
3656	int num_tolerated_disk_barrier_failures =
3657		(int)fs_info->fs_devices->num_devices;
3658
3659	for (i = 0; i < ARRAY_SIZE(types); i++) {
3660		struct btrfs_space_info *tmp;
3661
3662		sinfo = NULL;
3663		rcu_read_lock();
3664		list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3665			if (tmp->flags == types[i]) {
3666				sinfo = tmp;
3667				break;
3668			}
3669		}
3670		rcu_read_unlock();
3671
3672		if (!sinfo)
3673			continue;
3674
3675		down_read(&sinfo->groups_sem);
3676		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3677			u64 flags;
3678
3679			if (list_empty(&sinfo->block_groups[c]))
3680				continue;
3681
3682			btrfs_get_block_group_info(&sinfo->block_groups[c],
3683						   &space);
3684			if (space.total_bytes == 0 || space.used_bytes == 0)
3685				continue;
3686			flags = space.flags;
3687
3688			num_tolerated_disk_barrier_failures = min(
3689				num_tolerated_disk_barrier_failures,
3690				btrfs_get_num_tolerated_disk_barrier_failures(
3691					flags));
3692		}
3693		up_read(&sinfo->groups_sem);
3694	}
3695
3696	return num_tolerated_disk_barrier_failures;
3697}
3698
3699static int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3700{
3701	struct list_head *head;
3702	struct btrfs_device *dev;
3703	struct btrfs_super_block *sb;
3704	struct btrfs_dev_item *dev_item;
3705	int ret;
3706	int do_barriers;
3707	int max_errors;
3708	int total_errors = 0;
3709	u64 flags;
3710
3711	do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3712	backup_super_roots(fs_info);
 
 
 
 
 
 
 
3713
3714	sb = fs_info->super_for_commit;
3715	dev_item = &sb->dev_item;
3716
3717	mutex_lock(&fs_info->fs_devices->device_list_mutex);
3718	head = &fs_info->fs_devices->devices;
3719	max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3720
3721	if (do_barriers) {
3722		ret = barrier_all_devices(fs_info);
3723		if (ret) {
3724			mutex_unlock(
3725				&fs_info->fs_devices->device_list_mutex);
3726			btrfs_handle_fs_error(fs_info, ret,
3727					      "errors while submitting device barriers.");
3728			return ret;
3729		}
3730	}
3731
3732	list_for_each_entry_rcu(dev, head, dev_list) {
3733		if (!dev->bdev) {
3734			total_errors++;
3735			continue;
3736		}
3737		if (!dev->in_fs_metadata || !dev->writeable)
 
3738			continue;
3739
3740		btrfs_set_stack_device_generation(dev_item, 0);
3741		btrfs_set_stack_device_type(dev_item, dev->type);
3742		btrfs_set_stack_device_id(dev_item, dev->devid);
3743		btrfs_set_stack_device_total_bytes(dev_item,
3744						   dev->commit_total_bytes);
3745		btrfs_set_stack_device_bytes_used(dev_item,
3746						  dev->commit_bytes_used);
3747		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3748		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3749		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3750		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3751		memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
 
3752
3753		flags = btrfs_super_flags(sb);
3754		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3755
3756		ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
 
 
 
 
 
 
 
 
3757		if (ret)
3758			total_errors++;
3759	}
3760	if (total_errors > max_errors) {
3761		btrfs_err(fs_info, "%d errors while writing supers",
3762			  total_errors);
3763		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3764
3765		/* FUA is masked off if unsupported and can't be the reason */
3766		btrfs_handle_fs_error(fs_info, -EIO,
3767				      "%d errors while writing supers",
3768				      total_errors);
3769		return -EIO;
3770	}
3771
3772	total_errors = 0;
3773	list_for_each_entry_rcu(dev, head, dev_list) {
3774		if (!dev->bdev)
3775			continue;
3776		if (!dev->in_fs_metadata || !dev->writeable)
 
3777			continue;
3778
3779		ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3780		if (ret)
3781			total_errors++;
3782	}
3783	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3784	if (total_errors > max_errors) {
3785		btrfs_handle_fs_error(fs_info, -EIO,
3786				      "%d errors while writing supers",
3787				      total_errors);
3788		return -EIO;
3789	}
3790	return 0;
3791}
3792
3793int write_ctree_super(struct btrfs_trans_handle *trans,
3794		      struct btrfs_fs_info *fs_info, int max_mirrors)
3795{
3796	return write_all_supers(fs_info, max_mirrors);
3797}
3798
3799/* Drop a fs root from the radix tree and free it. */
3800void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3801				  struct btrfs_root *root)
3802{
 
 
3803	spin_lock(&fs_info->fs_roots_radix_lock);
3804	radix_tree_delete(&fs_info->fs_roots_radix,
3805			  (unsigned long)root->root_key.objectid);
 
 
3806	spin_unlock(&fs_info->fs_roots_radix_lock);
3807
3808	if (btrfs_root_refs(&root->root_item) == 0)
3809		synchronize_srcu(&fs_info->subvol_srcu);
3810
3811	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3812		btrfs_free_log(NULL, root);
3813		if (root->reloc_root) {
3814			free_extent_buffer(root->reloc_root->node);
3815			free_extent_buffer(root->reloc_root->commit_root);
3816			btrfs_put_fs_root(root->reloc_root);
3817			root->reloc_root = NULL;
3818		}
3819	}
3820
3821	if (root->free_ino_pinned)
3822		__btrfs_remove_free_space_cache(root->free_ino_pinned);
3823	if (root->free_ino_ctl)
3824		__btrfs_remove_free_space_cache(root->free_ino_ctl);
3825	free_fs_root(root);
3826}
3827
3828static void free_fs_root(struct btrfs_root *root)
3829{
3830	iput(root->ino_cache_inode);
3831	WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3832	btrfs_free_block_rsv(root->fs_info, root->orphan_block_rsv);
3833	root->orphan_block_rsv = NULL;
3834	if (root->anon_dev)
3835		free_anon_bdev(root->anon_dev);
3836	if (root->subv_writers)
3837		btrfs_free_subvolume_writers(root->subv_writers);
3838	free_extent_buffer(root->node);
3839	free_extent_buffer(root->commit_root);
3840	kfree(root->free_ino_ctl);
3841	kfree(root->free_ino_pinned);
3842	kfree(root->name);
3843	btrfs_put_fs_root(root);
3844}
3845
3846void btrfs_free_fs_root(struct btrfs_root *root)
3847{
3848	free_fs_root(root);
3849}
3850
3851int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3852{
3853	u64 root_objectid = 0;
3854	struct btrfs_root *gang[8];
3855	int i = 0;
3856	int err = 0;
3857	unsigned int ret = 0;
3858	int index;
3859
3860	while (1) {
3861		index = srcu_read_lock(&fs_info->subvol_srcu);
3862		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3863					     (void **)gang, root_objectid,
3864					     ARRAY_SIZE(gang));
3865		if (!ret) {
3866			srcu_read_unlock(&fs_info->subvol_srcu, index);
3867			break;
3868		}
3869		root_objectid = gang[ret - 1]->root_key.objectid + 1;
3870
3871		for (i = 0; i < ret; i++) {
3872			/* Avoid to grab roots in dead_roots */
3873			if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3874				gang[i] = NULL;
3875				continue;
3876			}
3877			/* grab all the search result for later use */
3878			gang[i] = btrfs_grab_fs_root(gang[i]);
3879		}
3880		srcu_read_unlock(&fs_info->subvol_srcu, index);
3881
3882		for (i = 0; i < ret; i++) {
3883			if (!gang[i])
3884				continue;
3885			root_objectid = gang[i]->root_key.objectid;
3886			err = btrfs_orphan_cleanup(gang[i]);
3887			if (err)
3888				break;
3889			btrfs_put_fs_root(gang[i]);
3890		}
3891		root_objectid++;
3892	}
3893
3894	/* release the uncleaned roots due to error */
3895	for (; i < ret; i++) {
3896		if (gang[i])
3897			btrfs_put_fs_root(gang[i]);
3898	}
3899	return err;
3900}
3901
3902int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3903{
3904	struct btrfs_root *root = fs_info->tree_root;
3905	struct btrfs_trans_handle *trans;
3906
3907	mutex_lock(&fs_info->cleaner_mutex);
3908	btrfs_run_delayed_iputs(fs_info);
3909	mutex_unlock(&fs_info->cleaner_mutex);
3910	wake_up_process(fs_info->cleaner_kthread);
3911
3912	/* wait until ongoing cleanup work done */
3913	down_write(&fs_info->cleanup_work_sem);
3914	up_write(&fs_info->cleanup_work_sem);
3915
3916	trans = btrfs_join_transaction(root);
3917	if (IS_ERR(trans))
3918		return PTR_ERR(trans);
3919	return btrfs_commit_transaction(trans);
3920}
3921
3922void close_ctree(struct btrfs_fs_info *fs_info)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3923{
3924	struct btrfs_root *root = fs_info->tree_root;
3925	int ret;
3926
3927	set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3928
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3929	/* wait for the qgroup rescan worker to stop */
3930	btrfs_qgroup_wait_for_completion(fs_info, false);
3931
3932	/* wait for the uuid_scan task to finish */
3933	down(&fs_info->uuid_tree_rescan_sem);
3934	/* avoid complains from lockdep et al., set sem back to initial state */
3935	up(&fs_info->uuid_tree_rescan_sem);
3936
3937	/* pause restriper - we want to resume on mount */
3938	btrfs_pause_balance(fs_info);
3939
3940	btrfs_dev_replace_suspend_for_unmount(fs_info);
3941
3942	btrfs_scrub_cancel(fs_info);
3943
3944	/* wait for any defraggers to finish */
3945	wait_event(fs_info->transaction_wait,
3946		   (atomic_read(&fs_info->defrag_running) == 0));
3947
3948	/* clear out the rbtree of defraggable inodes */
3949	btrfs_cleanup_defrag_inodes(fs_info);
3950
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3951	cancel_work_sync(&fs_info->async_reclaim_work);
 
 
3952
3953	if (!(fs_info->sb->s_flags & MS_RDONLY)) {
 
 
 
3954		/*
3955		 * If the cleaner thread is stopped and there are
3956		 * block groups queued for removal, the deletion will be
3957		 * skipped when we quit the cleaner thread.
3958		 */
3959		btrfs_delete_unused_bgs(fs_info);
3960
 
 
 
 
 
 
 
 
 
 
 
 
 
3961		ret = btrfs_commit_super(fs_info);
3962		if (ret)
3963			btrfs_err(fs_info, "commit super ret %d", ret);
3964	}
3965
3966	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3967		btrfs_error_commit_super(fs_info);
3968
3969	kthread_stop(fs_info->transaction_kthread);
3970	kthread_stop(fs_info->cleaner_kthread);
3971
 
3972	set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
3973
 
 
 
 
 
3974	btrfs_free_qgroup_config(fs_info);
 
3975
3976	if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3977		btrfs_info(fs_info, "at unmount delalloc count %lld",
3978		       percpu_counter_sum(&fs_info->delalloc_bytes));
3979	}
3980
 
 
 
 
3981	btrfs_sysfs_remove_mounted(fs_info);
3982	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3983
3984	btrfs_free_fs_roots(fs_info);
3985
3986	btrfs_put_block_group_cache(fs_info);
3987
3988	btrfs_free_block_groups(fs_info);
3989
3990	/*
3991	 * we must make sure there is not any read request to
3992	 * submit after we stopping all workers.
3993	 */
3994	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3995	btrfs_stop_all_workers(fs_info);
3996
 
 
 
3997	clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
3998	free_root_pointers(fs_info, 1);
 
 
 
 
 
 
 
 
 
 
3999
4000	iput(fs_info->btree_inode);
4001
4002#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4003	if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4004		btrfsic_unmount(fs_info->fs_devices);
4005#endif
4006
4007	btrfs_close_devices(fs_info->fs_devices);
4008	btrfs_mapping_tree_free(&fs_info->mapping_tree);
4009
4010	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4011	percpu_counter_destroy(&fs_info->delalloc_bytes);
4012	percpu_counter_destroy(&fs_info->bio_counter);
4013	bdi_destroy(&fs_info->bdi);
4014	cleanup_srcu_struct(&fs_info->subvol_srcu);
4015
4016	btrfs_free_stripe_hash_table(fs_info);
4017
4018	__btrfs_free_block_rsv(root->orphan_block_rsv);
4019	root->orphan_block_rsv = NULL;
4020
4021	mutex_lock(&fs_info->chunk_mutex);
4022	while (!list_empty(&fs_info->pinned_chunks)) {
4023		struct extent_map *em;
4024
4025		em = list_first_entry(&fs_info->pinned_chunks,
4026				      struct extent_map, list);
4027		list_del_init(&em->list);
4028		free_extent_map(em);
4029	}
4030	mutex_unlock(&fs_info->chunk_mutex);
4031}
4032
4033int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4034			  int atomic)
4035{
4036	int ret;
4037	struct inode *btree_inode = buf->pages[0]->mapping->host;
4038
4039	ret = extent_buffer_uptodate(buf);
4040	if (!ret)
4041		return ret;
4042
4043	ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4044				    parent_transid, atomic);
4045	if (ret == -EAGAIN)
4046		return ret;
4047	return !ret;
4048}
4049
4050void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
 
4051{
4052	struct btrfs_fs_info *fs_info;
4053	struct btrfs_root *root;
4054	u64 transid = btrfs_header_generation(buf);
4055	int was_dirty;
4056
4057#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4058	/*
4059	 * This is a fast path so only do this check if we have sanity tests
4060	 * enabled.  Normal people shouldn't be marking dummy buffers as dirty
4061	 * outside of the sanity tests.
4062	 */
4063	if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
4064		return;
4065#endif
4066	root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4067	fs_info = root->fs_info;
4068	btrfs_assert_tree_locked(buf);
4069	if (transid != fs_info->generation)
4070		WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4071			buf->start, transid, fs_info->generation);
4072	was_dirty = set_extent_buffer_dirty(buf);
4073	if (!was_dirty)
4074		__percpu_counter_add(&fs_info->dirty_metadata_bytes,
4075				     buf->len,
4076				     fs_info->dirty_metadata_batch);
4077#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4078	if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
4079		btrfs_print_leaf(fs_info, buf);
4080		ASSERT(0);
4081	}
4082#endif
4083}
4084
4085static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4086					int flush_delayed)
4087{
4088	/*
4089	 * looks as though older kernels can get into trouble with
4090	 * this code, they end up stuck in balance_dirty_pages forever
4091	 */
4092	int ret;
4093
4094	if (current->flags & PF_MEMALLOC)
4095		return;
4096
4097	if (flush_delayed)
4098		btrfs_balance_delayed_items(fs_info);
4099
4100	ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4101				     BTRFS_DIRTY_METADATA_THRESH);
 
4102	if (ret > 0) {
4103		balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4104	}
4105}
4106
4107void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4108{
4109	__btrfs_btree_balance_dirty(fs_info, 1);
4110}
4111
4112void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4113{
4114	__btrfs_btree_balance_dirty(fs_info, 0);
4115}
4116
4117int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4118{
4119	struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4120	struct btrfs_fs_info *fs_info = root->fs_info;
 
 
 
 
4121
4122	return btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
 
4123}
4124
4125static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
4126			      int read_only)
4127{
4128	struct btrfs_super_block *sb = fs_info->super_copy;
4129	u64 nodesize = btrfs_super_nodesize(sb);
4130	u64 sectorsize = btrfs_super_sectorsize(sb);
4131	int ret = 0;
4132
4133	if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4134		btrfs_err(fs_info, "no valid FS found");
4135		ret = -EINVAL;
4136	}
4137	if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4138		btrfs_warn(fs_info, "unrecognized super flag: %llu",
4139				btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4140	if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4141		btrfs_err(fs_info, "tree_root level too big: %d >= %d",
4142				btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4143		ret = -EINVAL;
4144	}
4145	if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4146		btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
4147				btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4148		ret = -EINVAL;
4149	}
4150	if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4151		btrfs_err(fs_info, "log_root level too big: %d >= %d",
4152				btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4153		ret = -EINVAL;
4154	}
4155
4156	/*
4157	 * Check sectorsize and nodesize first, other check will need it.
4158	 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4159	 */
4160	if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4161	    sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4162		btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
4163		ret = -EINVAL;
4164	}
4165	/* Only PAGE SIZE is supported yet */
4166	if (sectorsize != PAGE_SIZE) {
4167		btrfs_err(fs_info,
4168			"sectorsize %llu not supported yet, only support %lu",
4169			sectorsize, PAGE_SIZE);
4170		ret = -EINVAL;
4171	}
4172	if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4173	    nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4174		btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
4175		ret = -EINVAL;
4176	}
4177	if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4178		btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
4179			  le32_to_cpu(sb->__unused_leafsize), nodesize);
4180		ret = -EINVAL;
4181	}
4182
4183	/* Root alignment check */
4184	if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4185		btrfs_warn(fs_info, "tree_root block unaligned: %llu",
4186			   btrfs_super_root(sb));
4187		ret = -EINVAL;
 
 
 
 
4188	}
4189	if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4190		btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
4191			   btrfs_super_chunk_root(sb));
4192		ret = -EINVAL;
4193	}
4194	if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4195		btrfs_warn(fs_info, "log_root block unaligned: %llu",
4196			   btrfs_super_log_root(sb));
4197		ret = -EINVAL;
4198	}
4199
4200	if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4201		btrfs_err(fs_info,
4202			   "dev_item UUID does not match fsid: %pU != %pU",
4203			   fs_info->fsid, sb->dev_item.fsid);
4204		ret = -EINVAL;
4205	}
4206
4207	/*
4208	 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4209	 * done later
4210	 */
4211	if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
4212		btrfs_err(fs_info, "bytes_used is too small %llu",
4213			  btrfs_super_bytes_used(sb));
4214		ret = -EINVAL;
4215	}
4216	if (!is_power_of_2(btrfs_super_stripesize(sb))) {
4217		btrfs_err(fs_info, "invalid stripesize %u",
4218			  btrfs_super_stripesize(sb));
4219		ret = -EINVAL;
4220	}
4221	if (btrfs_super_num_devices(sb) > (1UL << 31))
4222		btrfs_warn(fs_info, "suspicious number of devices: %llu",
4223			   btrfs_super_num_devices(sb));
4224	if (btrfs_super_num_devices(sb) == 0) {
4225		btrfs_err(fs_info, "number of devices is 0");
4226		ret = -EINVAL;
4227	}
4228
4229	if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4230		btrfs_err(fs_info, "super offset mismatch %llu != %u",
4231			  btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4232		ret = -EINVAL;
4233	}
4234
4235	/*
4236	 * Obvious sys_chunk_array corruptions, it must hold at least one key
4237	 * and one chunk
4238	 */
4239	if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4240		btrfs_err(fs_info, "system chunk array too big %u > %u",
4241			  btrfs_super_sys_array_size(sb),
4242			  BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4243		ret = -EINVAL;
4244	}
4245	if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4246			+ sizeof(struct btrfs_chunk)) {
4247		btrfs_err(fs_info, "system chunk array too small %u < %zu",
4248			  btrfs_super_sys_array_size(sb),
4249			  sizeof(struct btrfs_disk_key)
4250			  + sizeof(struct btrfs_chunk));
4251		ret = -EINVAL;
4252	}
4253
4254	/*
4255	 * The generation is a global counter, we'll trust it more than the others
4256	 * but it's still possible that it's the one that's wrong.
4257	 */
4258	if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4259		btrfs_warn(fs_info,
4260			"suspicious: generation < chunk_root_generation: %llu < %llu",
4261			btrfs_super_generation(sb),
4262			btrfs_super_chunk_root_generation(sb));
4263	if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4264	    && btrfs_super_cache_generation(sb) != (u64)-1)
4265		btrfs_warn(fs_info,
4266			"suspicious: generation < cache_generation: %llu < %llu",
4267			btrfs_super_generation(sb),
4268			btrfs_super_cache_generation(sb));
4269
4270	return ret;
4271}
4272
4273static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4274{
4275	mutex_lock(&fs_info->cleaner_mutex);
4276	btrfs_run_delayed_iputs(fs_info);
4277	mutex_unlock(&fs_info->cleaner_mutex);
4278
4279	down_write(&fs_info->cleanup_work_sem);
4280	up_write(&fs_info->cleanup_work_sem);
4281
4282	/* cleanup FS via transaction */
4283	btrfs_cleanup_transaction(fs_info);
4284}
4285
4286static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4287{
4288	struct btrfs_ordered_extent *ordered;
4289
4290	spin_lock(&root->ordered_extent_lock);
4291	/*
4292	 * This will just short circuit the ordered completion stuff which will
4293	 * make sure the ordered extent gets properly cleaned up.
4294	 */
4295	list_for_each_entry(ordered, &root->ordered_extents,
4296			    root_extent_list)
4297		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4298	spin_unlock(&root->ordered_extent_lock);
4299}
4300
4301static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4302{
4303	struct btrfs_root *root;
4304	struct list_head splice;
4305
4306	INIT_LIST_HEAD(&splice);
4307
4308	spin_lock(&fs_info->ordered_root_lock);
4309	list_splice_init(&fs_info->ordered_roots, &splice);
4310	while (!list_empty(&splice)) {
4311		root = list_first_entry(&splice, struct btrfs_root,
4312					ordered_root);
4313		list_move_tail(&root->ordered_root,
4314			       &fs_info->ordered_roots);
4315
4316		spin_unlock(&fs_info->ordered_root_lock);
4317		btrfs_destroy_ordered_extents(root);
4318
4319		cond_resched();
4320		spin_lock(&fs_info->ordered_root_lock);
4321	}
4322	spin_unlock(&fs_info->ordered_root_lock);
 
 
 
 
 
 
 
 
4323}
4324
4325static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4326				      struct btrfs_fs_info *fs_info)
4327{
4328	struct rb_node *node;
4329	struct btrfs_delayed_ref_root *delayed_refs;
4330	struct btrfs_delayed_ref_node *ref;
4331	int ret = 0;
4332
4333	delayed_refs = &trans->delayed_refs;
4334
4335	spin_lock(&delayed_refs->lock);
4336	if (atomic_read(&delayed_refs->num_entries) == 0) {
4337		spin_unlock(&delayed_refs->lock);
4338		btrfs_info(fs_info, "delayed_refs has NO entry");
4339		return ret;
4340	}
4341
4342	while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4343		struct btrfs_delayed_ref_head *head;
4344		struct btrfs_delayed_ref_node *tmp;
4345		bool pin_bytes = false;
4346
4347		head = rb_entry(node, struct btrfs_delayed_ref_head,
4348				href_node);
4349		if (!mutex_trylock(&head->mutex)) {
4350			atomic_inc(&head->node.refs);
4351			spin_unlock(&delayed_refs->lock);
4352
4353			mutex_lock(&head->mutex);
4354			mutex_unlock(&head->mutex);
4355			btrfs_put_delayed_ref(&head->node);
4356			spin_lock(&delayed_refs->lock);
4357			continue;
4358		}
4359		spin_lock(&head->lock);
4360		list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4361						 list) {
4362			ref->in_tree = 0;
4363			list_del(&ref->list);
 
4364			if (!list_empty(&ref->add_list))
4365				list_del(&ref->add_list);
4366			atomic_dec(&delayed_refs->num_entries);
4367			btrfs_put_delayed_ref(ref);
 
4368		}
4369		if (head->must_insert_reserved)
4370			pin_bytes = true;
4371		btrfs_free_delayed_extent_op(head->extent_op);
4372		delayed_refs->num_heads--;
4373		if (head->processing == 0)
4374			delayed_refs->num_heads_ready--;
4375		atomic_dec(&delayed_refs->num_entries);
4376		head->node.in_tree = 0;
4377		rb_erase(&head->href_node, &delayed_refs->href_root);
4378		spin_unlock(&head->lock);
4379		spin_unlock(&delayed_refs->lock);
4380		mutex_unlock(&head->mutex);
4381
4382		if (pin_bytes)
4383			btrfs_pin_extent(fs_info, head->node.bytenr,
4384					 head->node.num_bytes, 1);
4385		btrfs_put_delayed_ref(&head->node);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4386		cond_resched();
4387		spin_lock(&delayed_refs->lock);
4388	}
 
4389
4390	spin_unlock(&delayed_refs->lock);
4391
4392	return ret;
4393}
4394
4395static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4396{
4397	struct btrfs_inode *btrfs_inode;
4398	struct list_head splice;
4399
4400	INIT_LIST_HEAD(&splice);
4401
4402	spin_lock(&root->delalloc_lock);
4403	list_splice_init(&root->delalloc_inodes, &splice);
4404
4405	while (!list_empty(&splice)) {
 
4406		btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4407					       delalloc_inodes);
4408
4409		list_del_init(&btrfs_inode->delalloc_inodes);
4410		clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4411			  &btrfs_inode->runtime_flags);
4412		spin_unlock(&root->delalloc_lock);
4413
4414		btrfs_invalidate_inodes(btrfs_inode->root);
4415
 
 
 
 
 
 
 
 
 
 
 
4416		spin_lock(&root->delalloc_lock);
4417	}
4418
4419	spin_unlock(&root->delalloc_lock);
4420}
4421
4422static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4423{
4424	struct btrfs_root *root;
4425	struct list_head splice;
4426
4427	INIT_LIST_HEAD(&splice);
4428
4429	spin_lock(&fs_info->delalloc_root_lock);
4430	list_splice_init(&fs_info->delalloc_roots, &splice);
4431	while (!list_empty(&splice)) {
4432		root = list_first_entry(&splice, struct btrfs_root,
4433					 delalloc_root);
4434		list_del_init(&root->delalloc_root);
4435		root = btrfs_grab_fs_root(root);
4436		BUG_ON(!root);
4437		spin_unlock(&fs_info->delalloc_root_lock);
4438
4439		btrfs_destroy_delalloc_inodes(root);
4440		btrfs_put_fs_root(root);
4441
4442		spin_lock(&fs_info->delalloc_root_lock);
4443	}
4444	spin_unlock(&fs_info->delalloc_root_lock);
4445}
4446
4447static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4448					struct extent_io_tree *dirty_pages,
4449					int mark)
4450{
4451	int ret;
4452	struct extent_buffer *eb;
4453	u64 start = 0;
4454	u64 end;
4455
4456	while (1) {
4457		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4458					    mark, NULL);
4459		if (ret)
4460			break;
4461
4462		clear_extent_bits(dirty_pages, start, end, mark);
4463		while (start <= end) {
4464			eb = find_extent_buffer(fs_info, start);
4465			start += fs_info->nodesize;
4466			if (!eb)
4467				continue;
 
 
4468			wait_on_extent_buffer_writeback(eb);
 
 
4469
4470			if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4471					       &eb->bflags))
4472				clear_extent_buffer_dirty(eb);
4473			free_extent_buffer_stale(eb);
4474		}
4475	}
4476
4477	return ret;
4478}
4479
4480static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4481				       struct extent_io_tree *pinned_extents)
4482{
4483	struct extent_io_tree *unpin;
4484	u64 start;
4485	u64 end;
4486	int ret;
4487	bool loop = true;
4488
4489	unpin = pinned_extents;
4490again:
4491	while (1) {
4492		ret = find_first_extent_bit(unpin, 0, &start, &end,
4493					    EXTENT_DIRTY, NULL);
4494		if (ret)
 
 
 
 
 
 
 
 
 
4495			break;
 
4496
4497		clear_extent_dirty(unpin, start, end);
 
4498		btrfs_error_unpin_extent_range(fs_info, start, end);
 
4499		cond_resched();
4500	}
4501
4502	if (loop) {
4503		if (unpin == &fs_info->freed_extents[0])
4504			unpin = &fs_info->freed_extents[1];
4505		else
4506			unpin = &fs_info->freed_extents[0];
4507		loop = false;
4508		goto again;
4509	}
4510
4511	return 0;
4512}
4513
4514static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4515{
4516	struct inode *inode;
4517
4518	inode = cache->io_ctl.inode;
4519	if (inode) {
 
 
 
4520		invalidate_inode_pages2(inode->i_mapping);
 
 
4521		BTRFS_I(inode)->generation = 0;
4522		cache->io_ctl.inode = NULL;
4523		iput(inode);
4524	}
 
4525	btrfs_put_block_group(cache);
4526}
4527
4528void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4529			     struct btrfs_fs_info *fs_info)
4530{
4531	struct btrfs_block_group_cache *cache;
4532
4533	spin_lock(&cur_trans->dirty_bgs_lock);
4534	while (!list_empty(&cur_trans->dirty_bgs)) {
4535		cache = list_first_entry(&cur_trans->dirty_bgs,
4536					 struct btrfs_block_group_cache,
4537					 dirty_list);
4538		if (!cache) {
4539			btrfs_err(fs_info, "orphan block group dirty_bgs list");
4540			spin_unlock(&cur_trans->dirty_bgs_lock);
4541			return;
4542		}
4543
4544		if (!list_empty(&cache->io_list)) {
4545			spin_unlock(&cur_trans->dirty_bgs_lock);
4546			list_del_init(&cache->io_list);
4547			btrfs_cleanup_bg_io(cache);
4548			spin_lock(&cur_trans->dirty_bgs_lock);
4549		}
4550
4551		list_del_init(&cache->dirty_list);
4552		spin_lock(&cache->lock);
4553		cache->disk_cache_state = BTRFS_DC_ERROR;
4554		spin_unlock(&cache->lock);
4555
4556		spin_unlock(&cur_trans->dirty_bgs_lock);
4557		btrfs_put_block_group(cache);
 
4558		spin_lock(&cur_trans->dirty_bgs_lock);
4559	}
4560	spin_unlock(&cur_trans->dirty_bgs_lock);
4561
 
 
 
 
4562	while (!list_empty(&cur_trans->io_bgs)) {
4563		cache = list_first_entry(&cur_trans->io_bgs,
4564					 struct btrfs_block_group_cache,
4565					 io_list);
4566		if (!cache) {
4567			btrfs_err(fs_info, "orphan block group on io_bgs list");
4568			return;
4569		}
4570
4571		list_del_init(&cache->io_list);
4572		spin_lock(&cache->lock);
4573		cache->disk_cache_state = BTRFS_DC_ERROR;
4574		spin_unlock(&cache->lock);
4575		btrfs_cleanup_bg_io(cache);
4576	}
4577}
4578
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4579void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4580				   struct btrfs_fs_info *fs_info)
4581{
 
 
4582	btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4583	ASSERT(list_empty(&cur_trans->dirty_bgs));
4584	ASSERT(list_empty(&cur_trans->io_bgs));
4585
 
 
 
 
 
4586	btrfs_destroy_delayed_refs(cur_trans, fs_info);
4587
4588	cur_trans->state = TRANS_STATE_COMMIT_START;
4589	wake_up(&fs_info->transaction_blocked_wait);
4590
4591	cur_trans->state = TRANS_STATE_UNBLOCKED;
4592	wake_up(&fs_info->transaction_wait);
4593
4594	btrfs_destroy_delayed_inodes(fs_info);
4595	btrfs_assert_delayed_root_empty(fs_info);
4596
4597	btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4598				     EXTENT_DIRTY);
4599	btrfs_destroy_pinned_extent(fs_info,
4600				    fs_info->pinned_extents);
 
4601
4602	cur_trans->state =TRANS_STATE_COMPLETED;
4603	wake_up(&cur_trans->commit_wait);
4604
4605	/*
4606	memset(cur_trans, 0, sizeof(*cur_trans));
4607	kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4608	*/
4609}
4610
4611static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4612{
4613	struct btrfs_transaction *t;
4614
4615	mutex_lock(&fs_info->transaction_kthread_mutex);
4616
4617	spin_lock(&fs_info->trans_lock);
4618	while (!list_empty(&fs_info->trans_list)) {
4619		t = list_first_entry(&fs_info->trans_list,
4620				     struct btrfs_transaction, list);
4621		if (t->state >= TRANS_STATE_COMMIT_START) {
4622			atomic_inc(&t->use_count);
4623			spin_unlock(&fs_info->trans_lock);
4624			btrfs_wait_for_commit(fs_info, t->transid);
4625			btrfs_put_transaction(t);
4626			spin_lock(&fs_info->trans_lock);
4627			continue;
4628		}
4629		if (t == fs_info->running_transaction) {
4630			t->state = TRANS_STATE_COMMIT_DOING;
4631			spin_unlock(&fs_info->trans_lock);
4632			/*
4633			 * We wait for 0 num_writers since we don't hold a trans
4634			 * handle open currently for this transaction.
4635			 */
4636			wait_event(t->writer_wait,
4637				   atomic_read(&t->num_writers) == 0);
4638		} else {
4639			spin_unlock(&fs_info->trans_lock);
4640		}
4641		btrfs_cleanup_one_transaction(t, fs_info);
4642
4643		spin_lock(&fs_info->trans_lock);
4644		if (t == fs_info->running_transaction)
4645			fs_info->running_transaction = NULL;
4646		list_del_init(&t->list);
4647		spin_unlock(&fs_info->trans_lock);
4648
4649		btrfs_put_transaction(t);
4650		trace_btrfs_transaction_commit(fs_info->tree_root);
4651		spin_lock(&fs_info->trans_lock);
4652	}
4653	spin_unlock(&fs_info->trans_lock);
4654	btrfs_destroy_all_ordered_extents(fs_info);
4655	btrfs_destroy_delayed_inodes(fs_info);
4656	btrfs_assert_delayed_root_empty(fs_info);
4657	btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4658	btrfs_destroy_all_delalloc_inodes(fs_info);
 
4659	mutex_unlock(&fs_info->transaction_kthread_mutex);
4660
4661	return 0;
4662}
4663
4664static const struct extent_io_ops btree_extent_io_ops = {
4665	.readpage_end_io_hook = btree_readpage_end_io_hook,
4666	.readpage_io_failed_hook = btree_io_failed_hook,
4667	.submit_bio_hook = btree_submit_bio_hook,
4668	/* note we're sharing with inode.c for the merge bio hook */
4669	.merge_bio_hook = btrfs_merge_bio_hook,
4670};