1c1d7c514SDavid Sterba // SPDX-License-Identifier: GPL-2.0 26cbd5570SChris Mason /* 36cbd5570SChris Mason * Copyright (C) 2007 Oracle. All rights reserved. 46cbd5570SChris Mason */ 56cbd5570SChris Mason 6e20d96d6SChris Mason #include <linux/fs.h> 7d98237b3SChris Mason #include <linux/blkdev.h> 8fc7cbcd4SDavid Sterba #include <linux/radix-tree.h> 935b7e476SChris Mason #include <linux/writeback.h> 10ce9adaa5SChris Mason #include <linux/workqueue.h> 11a74a4b97SChris Mason #include <linux/kthread.h> 125a0e3ad6STejun Heo #include <linux/slab.h> 13784b4e29SChris Mason #include <linux/migrate.h> 147a36ddecSDavid Sterba #include <linux/ratelimit.h> 156463fe58SStefan Behrens #include <linux/uuid.h> 16803b2f54SStefan Behrens #include <linux/semaphore.h> 17540adea3SMasami Hiramatsu #include <linux/error-injection.h> 189678c543SNikolay Borisov #include <linux/crc32c.h> 19b89f6d1fSFilipe Manana #include <linux/sched/mm.h> 207e75bf3fSDavid Sterba #include <asm/unaligned.h> 216d97c6e3SJohannes Thumshirn #include <crypto/hash.h> 22eb60ceacSChris Mason #include "ctree.h" 23eb60ceacSChris Mason #include "disk-io.h" 24e089f05cSChris Mason #include "transaction.h" 250f7d52f4SChris Mason #include "btrfs_inode.h" 26103c1972SChristoph Hellwig #include "bio.h" 27db94535dSChris Mason #include "print-tree.h" 28925baeddSChris Mason #include "locking.h" 29e02119d5SChris Mason #include "tree-log.h" 30fa9c0d79SChris Mason #include "free-space-cache.h" 3170f6d82eSOmar Sandoval #include "free-space-tree.h" 3221adbd5cSStefan Behrens #include "check-integrity.h" 33606686eeSJosef Bacik #include "rcu-string.h" 348dabb742SStefan Behrens #include "dev-replace.h" 3553b381b3SDavid Woodhouse #include "raid56.h" 365ac1d209SJeff Mahoney #include "sysfs.h" 37fcebe456SJosef Bacik #include "qgroup.h" 38ebb8765bSAnand Jain #include "compression.h" 39557ea5ddSQu Wenruo #include "tree-checker.h" 40fd708b81SJosef Bacik #include "ref-verify.h" 41aac0023cSJosef Bacik #include "block-group.h" 42b0643e59SDennis Zhou #include "discard.h" 43f603bb94SNikolay Borisov #include "space-info.h" 44b70f5097SNaohiro Aota #include "zoned.h" 45139e8cd3SQu Wenruo #include "subpage.h" 46c7f13d42SJosef Bacik #include "fs.h" 4707e81dc9SJosef Bacik #include "accessors.h" 48a0231804SJosef Bacik #include "extent-tree.h" 4945c40c8fSJosef Bacik #include "root-tree.h" 5059b818e0SJosef Bacik #include "defrag.h" 51c7a03b52SJosef Bacik #include "uuid-tree.h" 5267707479SJosef Bacik #include "relocation.h" 532fc6822cSJosef Bacik #include "scrub.h" 54c03b2207SJosef Bacik #include "super.h" 55eb60ceacSChris Mason 56319e4d06SQu Wenruo #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\ 57319e4d06SQu Wenruo BTRFS_HEADER_FLAG_RELOC |\ 58319e4d06SQu Wenruo BTRFS_SUPER_FLAG_ERROR |\ 59319e4d06SQu Wenruo BTRFS_SUPER_FLAG_SEEDING |\ 60e2731e55SAnand Jain BTRFS_SUPER_FLAG_METADUMP |\ 61e2731e55SAnand Jain BTRFS_SUPER_FLAG_METADUMP_V2) 62319e4d06SQu Wenruo 63143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root); 64acce952bSliubo static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 652ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info); 66143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root); 672ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 68acce952bSliubo struct extent_io_tree *dirty_pages, 69acce952bSliubo int mark); 702ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 71acce952bSliubo struct extent_io_tree *pinned_extents); 722ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info); 732ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info); 74ce9adaa5SChris Mason 75141386e1SJosef Bacik static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info) 76141386e1SJosef Bacik { 77141386e1SJosef Bacik if (fs_info->csum_shash) 78141386e1SJosef Bacik crypto_free_shash(fs_info->csum_shash); 79141386e1SJosef Bacik } 80141386e1SJosef Bacik 81d352ac68SChris Mason /* 822996e1f8SJohannes Thumshirn * Compute the csum of a btree block and store the result to provided buffer. 83d352ac68SChris Mason */ 84c67b3892SDavid Sterba static void csum_tree_block(struct extent_buffer *buf, u8 *result) 8519c00ddcSChris Mason { 86d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = buf->fs_info; 877280305eSDavid Sterba const int num_pages = num_extent_pages(buf); 88a26663e7SQu Wenruo const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize); 89d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 9019c00ddcSChris Mason char *kaddr; 91e9be5a30SDavid Sterba int i; 92d5178578SJohannes Thumshirn 93d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 94d5178578SJohannes Thumshirn crypto_shash_init(shash); 95a26663e7SQu Wenruo kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start); 96e9be5a30SDavid Sterba crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE, 97a26663e7SQu Wenruo first_page_part - BTRFS_CSUM_SIZE); 9819c00ddcSChris Mason 99e9be5a30SDavid Sterba for (i = 1; i < num_pages; i++) { 100e9be5a30SDavid Sterba kaddr = page_address(buf->pages[i]); 101e9be5a30SDavid Sterba crypto_shash_update(shash, kaddr, PAGE_SIZE); 10219c00ddcSChris Mason } 10371a63551SDavid Sterba memset(result, 0, BTRFS_CSUM_SIZE); 104d5178578SJohannes Thumshirn crypto_shash_final(shash, result); 10519c00ddcSChris Mason } 10619c00ddcSChris Mason 107d352ac68SChris Mason /* 108d352ac68SChris Mason * we can't consider a given block up to date unless the transid of the 109d352ac68SChris Mason * block matches the transid in the parent node's pointer. This is how we 110d352ac68SChris Mason * detect blocks that either didn't get written at all or got written 111d352ac68SChris Mason * in the wrong place. 112d352ac68SChris Mason */ 1131259ab75SChris Mason static int verify_parent_transid(struct extent_io_tree *io_tree, 114b9fab919SChris Mason struct extent_buffer *eb, u64 parent_transid, 115b9fab919SChris Mason int atomic) 1161259ab75SChris Mason { 1172ac55d41SJosef Bacik struct extent_state *cached_state = NULL; 1181259ab75SChris Mason int ret; 1191259ab75SChris Mason 1201259ab75SChris Mason if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 1211259ab75SChris Mason return 0; 1221259ab75SChris Mason 123b9fab919SChris Mason if (atomic) 124b9fab919SChris Mason return -EAGAIN; 125b9fab919SChris Mason 126570eb97bSJosef Bacik lock_extent(io_tree, eb->start, eb->start + eb->len - 1, &cached_state); 1270b32f4bbSJosef Bacik if (extent_buffer_uptodate(eb) && 1281259ab75SChris Mason btrfs_header_generation(eb) == parent_transid) { 1291259ab75SChris Mason ret = 0; 1301259ab75SChris Mason goto out; 1311259ab75SChris Mason } 13294647322SDavid Sterba btrfs_err_rl(eb->fs_info, 1338f0ed7d4SQu Wenruo "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu", 1348f0ed7d4SQu Wenruo eb->start, eb->read_mirror, 13529549aecSWang Shilong parent_transid, btrfs_header_generation(eb)); 1361259ab75SChris Mason ret = 1; 1370b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 13833958dc6SChris Mason out: 139570eb97bSJosef Bacik unlock_extent(io_tree, eb->start, eb->start + eb->len - 1, 140e43bbe5eSDavid Sterba &cached_state); 1411259ab75SChris Mason return ret; 1421259ab75SChris Mason } 1431259ab75SChris Mason 144e7e16f48SJohannes Thumshirn static bool btrfs_supported_super_csum(u16 csum_type) 145e7e16f48SJohannes Thumshirn { 146e7e16f48SJohannes Thumshirn switch (csum_type) { 147e7e16f48SJohannes Thumshirn case BTRFS_CSUM_TYPE_CRC32: 1483951e7f0SJohannes Thumshirn case BTRFS_CSUM_TYPE_XXHASH: 1493831bf00SJohannes Thumshirn case BTRFS_CSUM_TYPE_SHA256: 150352ae07bSDavid Sterba case BTRFS_CSUM_TYPE_BLAKE2: 151e7e16f48SJohannes Thumshirn return true; 152e7e16f48SJohannes Thumshirn default: 153e7e16f48SJohannes Thumshirn return false; 154e7e16f48SJohannes Thumshirn } 155e7e16f48SJohannes Thumshirn } 156e7e16f48SJohannes Thumshirn 157d352ac68SChris Mason /* 1581104a885SDavid Sterba * Return 0 if the superblock checksum type matches the checksum value of that 1591104a885SDavid Sterba * algorithm. Pass the raw disk superblock data. 1601104a885SDavid Sterba */ 1613d17adeaSQu Wenruo int btrfs_check_super_csum(struct btrfs_fs_info *fs_info, 1623d17adeaSQu Wenruo const struct btrfs_super_block *disk_sb) 1631104a885SDavid Sterba { 16451bce6c9SJohannes Thumshirn char result[BTRFS_CSUM_SIZE]; 165d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 166d5178578SJohannes Thumshirn 167d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 1681104a885SDavid Sterba 1691104a885SDavid Sterba /* 1701104a885SDavid Sterba * The super_block structure does not span the whole 17151bce6c9SJohannes Thumshirn * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is 17251bce6c9SJohannes Thumshirn * filled with zeros and is included in the checksum. 1731104a885SDavid Sterba */ 1743d17adeaSQu Wenruo crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE, 175fd08001fSEric Biggers BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result); 1761104a885SDavid Sterba 17755fc29beSDavid Sterba if (memcmp(disk_sb->csum, result, fs_info->csum_size)) 178e7e16f48SJohannes Thumshirn return 1; 1791104a885SDavid Sterba 180e7e16f48SJohannes Thumshirn return 0; 1811104a885SDavid Sterba } 1821104a885SDavid Sterba 183e064d5e9SDavid Sterba int btrfs_verify_level_key(struct extent_buffer *eb, int level, 184ff76a864SLiu Bo struct btrfs_key *first_key, u64 parent_transid) 185581c1760SQu Wenruo { 186e064d5e9SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 187581c1760SQu Wenruo int found_level; 188581c1760SQu Wenruo struct btrfs_key found_key; 189581c1760SQu Wenruo int ret; 190581c1760SQu Wenruo 191581c1760SQu Wenruo found_level = btrfs_header_level(eb); 192581c1760SQu Wenruo if (found_level != level) { 19363489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 19463489055SQu Wenruo KERN_ERR "BTRFS: tree level check failed\n"); 195581c1760SQu Wenruo btrfs_err(fs_info, 196581c1760SQu Wenruo "tree level mismatch detected, bytenr=%llu level expected=%u has=%u", 197581c1760SQu Wenruo eb->start, level, found_level); 198581c1760SQu Wenruo return -EIO; 199581c1760SQu Wenruo } 200581c1760SQu Wenruo 201581c1760SQu Wenruo if (!first_key) 202581c1760SQu Wenruo return 0; 203581c1760SQu Wenruo 2045d41be6fSQu Wenruo /* 2055d41be6fSQu Wenruo * For live tree block (new tree blocks in current transaction), 2065d41be6fSQu Wenruo * we need proper lock context to avoid race, which is impossible here. 2075d41be6fSQu Wenruo * So we only checks tree blocks which is read from disk, whose 2085d41be6fSQu Wenruo * generation <= fs_info->last_trans_committed. 2095d41be6fSQu Wenruo */ 2105d41be6fSQu Wenruo if (btrfs_header_generation(eb) > fs_info->last_trans_committed) 2115d41be6fSQu Wenruo return 0; 21262fdaa52SQu Wenruo 21362fdaa52SQu Wenruo /* We have @first_key, so this @eb must have at least one item */ 21462fdaa52SQu Wenruo if (btrfs_header_nritems(eb) == 0) { 21562fdaa52SQu Wenruo btrfs_err(fs_info, 21662fdaa52SQu Wenruo "invalid tree nritems, bytenr=%llu nritems=0 expect >0", 21762fdaa52SQu Wenruo eb->start); 21862fdaa52SQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 21962fdaa52SQu Wenruo return -EUCLEAN; 22062fdaa52SQu Wenruo } 22162fdaa52SQu Wenruo 222581c1760SQu Wenruo if (found_level) 223581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &found_key, 0); 224581c1760SQu Wenruo else 225581c1760SQu Wenruo btrfs_item_key_to_cpu(eb, &found_key, 0); 226581c1760SQu Wenruo ret = btrfs_comp_cpu_keys(first_key, &found_key); 227581c1760SQu Wenruo 228581c1760SQu Wenruo if (ret) { 22963489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 23063489055SQu Wenruo KERN_ERR "BTRFS: tree first key check failed\n"); 231581c1760SQu Wenruo btrfs_err(fs_info, 232ff76a864SLiu Bo "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", 233ff76a864SLiu Bo eb->start, parent_transid, first_key->objectid, 234ff76a864SLiu Bo first_key->type, first_key->offset, 235ff76a864SLiu Bo found_key.objectid, found_key.type, 236ff76a864SLiu Bo found_key.offset); 237581c1760SQu Wenruo } 238581c1760SQu Wenruo return ret; 239581c1760SQu Wenruo } 240581c1760SQu Wenruo 241bacf60e5SChristoph Hellwig static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb, 242bacf60e5SChristoph Hellwig int mirror_num) 243bacf60e5SChristoph Hellwig { 244bacf60e5SChristoph Hellwig struct btrfs_fs_info *fs_info = eb->fs_info; 245bacf60e5SChristoph Hellwig u64 start = eb->start; 246bacf60e5SChristoph Hellwig int i, num_pages = num_extent_pages(eb); 247bacf60e5SChristoph Hellwig int ret = 0; 248bacf60e5SChristoph Hellwig 249bacf60e5SChristoph Hellwig if (sb_rdonly(fs_info->sb)) 250bacf60e5SChristoph Hellwig return -EROFS; 251bacf60e5SChristoph Hellwig 252bacf60e5SChristoph Hellwig for (i = 0; i < num_pages; i++) { 253bacf60e5SChristoph Hellwig struct page *p = eb->pages[i]; 254bacf60e5SChristoph Hellwig 255bacf60e5SChristoph Hellwig ret = btrfs_repair_io_failure(fs_info, 0, start, PAGE_SIZE, 256bacf60e5SChristoph Hellwig start, p, start - page_offset(p), mirror_num); 257bacf60e5SChristoph Hellwig if (ret) 258bacf60e5SChristoph Hellwig break; 259bacf60e5SChristoph Hellwig start += PAGE_SIZE; 260bacf60e5SChristoph Hellwig } 261bacf60e5SChristoph Hellwig 262bacf60e5SChristoph Hellwig return ret; 263bacf60e5SChristoph Hellwig } 264bacf60e5SChristoph Hellwig 2651104a885SDavid Sterba /* 266d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 267d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 268581c1760SQu Wenruo * 269789d6a3aSQu Wenruo * @check: expected tree parentness check, see the comments of the 270789d6a3aSQu Wenruo * structure for details. 271d352ac68SChris Mason */ 2726a2e9dc4SFilipe Manana int btrfs_read_extent_buffer(struct extent_buffer *eb, 273789d6a3aSQu Wenruo struct btrfs_tree_parent_check *check) 274f188591eSChris Mason { 2755ab12d1fSDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 276ea466794SJosef Bacik int failed = 0; 277f188591eSChris Mason int ret; 278f188591eSChris Mason int num_copies = 0; 279f188591eSChris Mason int mirror_num = 0; 280ea466794SJosef Bacik int failed_mirror = 0; 281f188591eSChris Mason 282789d6a3aSQu Wenruo ASSERT(check); 283789d6a3aSQu Wenruo 284f188591eSChris Mason while (1) { 285f8397d69SNikolay Borisov clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 286947a6299SQu Wenruo ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check); 287947a6299SQu Wenruo if (!ret) 288581c1760SQu Wenruo break; 289d397712bSChris Mason 2900b246afaSJeff Mahoney num_copies = btrfs_num_copies(fs_info, 291f188591eSChris Mason eb->start, eb->len); 2924235298eSChris Mason if (num_copies == 1) 293ea466794SJosef Bacik break; 2944235298eSChris Mason 2955cf1ab56SJosef Bacik if (!failed_mirror) { 2965cf1ab56SJosef Bacik failed = 1; 2975cf1ab56SJosef Bacik failed_mirror = eb->read_mirror; 2985cf1ab56SJosef Bacik } 2995cf1ab56SJosef Bacik 300f188591eSChris Mason mirror_num++; 301ea466794SJosef Bacik if (mirror_num == failed_mirror) 302ea466794SJosef Bacik mirror_num++; 303ea466794SJosef Bacik 3044235298eSChris Mason if (mirror_num > num_copies) 305ea466794SJosef Bacik break; 306f188591eSChris Mason } 307ea466794SJosef Bacik 308c0901581SStefan Behrens if (failed && !ret && failed_mirror) 30920a1fbf9SDavid Sterba btrfs_repair_eb_io_failure(eb, failed_mirror); 310ea466794SJosef Bacik 311ea466794SJosef Bacik return ret; 312f188591eSChris Mason } 31319c00ddcSChris Mason 314eca0f6f6SQu Wenruo static int csum_one_extent_buffer(struct extent_buffer *eb) 315eca0f6f6SQu Wenruo { 316eca0f6f6SQu Wenruo struct btrfs_fs_info *fs_info = eb->fs_info; 317eca0f6f6SQu Wenruo u8 result[BTRFS_CSUM_SIZE]; 318eca0f6f6SQu Wenruo int ret; 319eca0f6f6SQu Wenruo 320eca0f6f6SQu Wenruo ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid, 321eca0f6f6SQu Wenruo offsetof(struct btrfs_header, fsid), 322eca0f6f6SQu Wenruo BTRFS_FSID_SIZE) == 0); 323eca0f6f6SQu Wenruo csum_tree_block(eb, result); 324eca0f6f6SQu Wenruo 325eca0f6f6SQu Wenruo if (btrfs_header_level(eb)) 326eca0f6f6SQu Wenruo ret = btrfs_check_node(eb); 327eca0f6f6SQu Wenruo else 328eca0f6f6SQu Wenruo ret = btrfs_check_leaf_full(eb); 329eca0f6f6SQu Wenruo 3303777369fSQu Wenruo if (ret < 0) 3313777369fSQu Wenruo goto error; 3323777369fSQu Wenruo 3333777369fSQu Wenruo /* 3343777369fSQu Wenruo * Also check the generation, the eb reached here must be newer than 3353777369fSQu Wenruo * last committed. Or something seriously wrong happened. 3363777369fSQu Wenruo */ 3373777369fSQu Wenruo if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) { 3383777369fSQu Wenruo ret = -EUCLEAN; 339eca0f6f6SQu Wenruo btrfs_err(fs_info, 3403777369fSQu Wenruo "block=%llu bad generation, have %llu expect > %llu", 3413777369fSQu Wenruo eb->start, btrfs_header_generation(eb), 3423777369fSQu Wenruo fs_info->last_trans_committed); 3433777369fSQu Wenruo goto error; 344eca0f6f6SQu Wenruo } 345eca0f6f6SQu Wenruo write_extent_buffer(eb, result, 0, fs_info->csum_size); 346eca0f6f6SQu Wenruo 347eca0f6f6SQu Wenruo return 0; 3483777369fSQu Wenruo 3493777369fSQu Wenruo error: 3503777369fSQu Wenruo btrfs_print_tree(eb, 0); 3513777369fSQu Wenruo btrfs_err(fs_info, "block=%llu write time tree block corruption detected", 3523777369fSQu Wenruo eb->start); 35316199ad9SFilipe Manana /* 35416199ad9SFilipe Manana * Be noisy if this is an extent buffer from a log tree. We don't abort 35516199ad9SFilipe Manana * a transaction in case there's a bad log tree extent buffer, we just 35616199ad9SFilipe Manana * fallback to a transaction commit. Still we want to know when there is 35716199ad9SFilipe Manana * a bad log tree extent buffer, as that may signal a bug somewhere. 35816199ad9SFilipe Manana */ 35916199ad9SFilipe Manana WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) || 36016199ad9SFilipe Manana btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID); 3613777369fSQu Wenruo return ret; 362eca0f6f6SQu Wenruo } 363eca0f6f6SQu Wenruo 364eca0f6f6SQu Wenruo /* Checksum all dirty extent buffers in one bio_vec */ 365eca0f6f6SQu Wenruo static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info, 366eca0f6f6SQu Wenruo struct bio_vec *bvec) 367eca0f6f6SQu Wenruo { 368eca0f6f6SQu Wenruo struct page *page = bvec->bv_page; 369eca0f6f6SQu Wenruo u64 bvec_start = page_offset(page) + bvec->bv_offset; 370eca0f6f6SQu Wenruo u64 cur; 371eca0f6f6SQu Wenruo int ret = 0; 372eca0f6f6SQu Wenruo 373eca0f6f6SQu Wenruo for (cur = bvec_start; cur < bvec_start + bvec->bv_len; 374eca0f6f6SQu Wenruo cur += fs_info->nodesize) { 375eca0f6f6SQu Wenruo struct extent_buffer *eb; 376eca0f6f6SQu Wenruo bool uptodate; 377eca0f6f6SQu Wenruo 378eca0f6f6SQu Wenruo eb = find_extent_buffer(fs_info, cur); 379eca0f6f6SQu Wenruo uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur, 380eca0f6f6SQu Wenruo fs_info->nodesize); 381eca0f6f6SQu Wenruo 38201cd3909SDavid Sterba /* A dirty eb shouldn't disappear from buffer_radix */ 383eca0f6f6SQu Wenruo if (WARN_ON(!eb)) 384eca0f6f6SQu Wenruo return -EUCLEAN; 385eca0f6f6SQu Wenruo 386eca0f6f6SQu Wenruo if (WARN_ON(cur != btrfs_header_bytenr(eb))) { 387eca0f6f6SQu Wenruo free_extent_buffer(eb); 388eca0f6f6SQu Wenruo return -EUCLEAN; 389eca0f6f6SQu Wenruo } 390eca0f6f6SQu Wenruo if (WARN_ON(!uptodate)) { 391eca0f6f6SQu Wenruo free_extent_buffer(eb); 392eca0f6f6SQu Wenruo return -EUCLEAN; 393eca0f6f6SQu Wenruo } 394eca0f6f6SQu Wenruo 395eca0f6f6SQu Wenruo ret = csum_one_extent_buffer(eb); 396eca0f6f6SQu Wenruo free_extent_buffer(eb); 397eca0f6f6SQu Wenruo if (ret < 0) 398eca0f6f6SQu Wenruo return ret; 399eca0f6f6SQu Wenruo } 400eca0f6f6SQu Wenruo return ret; 401eca0f6f6SQu Wenruo } 402eca0f6f6SQu Wenruo 403d352ac68SChris Mason /* 404ac303b69SQu Wenruo * Checksum a dirty tree block before IO. This has extra checks to make sure 405ac303b69SQu Wenruo * we only fill in the checksum field in the first page of a multi-page block. 406ac303b69SQu Wenruo * For subpage extent buffers we need bvec to also read the offset in the page. 407d352ac68SChris Mason */ 408ac303b69SQu Wenruo static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec) 40919c00ddcSChris Mason { 410ac303b69SQu Wenruo struct page *page = bvec->bv_page; 4114eee4fa4SMiao Xie u64 start = page_offset(page); 41219c00ddcSChris Mason u64 found_start; 41319c00ddcSChris Mason struct extent_buffer *eb; 414eca0f6f6SQu Wenruo 415fbca46ebSQu Wenruo if (fs_info->nodesize < PAGE_SIZE) 416eca0f6f6SQu Wenruo return csum_dirty_subpage_buffers(fs_info, bvec); 417f188591eSChris Mason 4184f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 4194f2de97aSJosef Bacik if (page != eb->pages[0]) 4204f2de97aSJosef Bacik return 0; 4210f805531SAlex Lyakas 42219c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 423d3575156SNaohiro Aota 424d3575156SNaohiro Aota if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) { 425d3575156SNaohiro Aota WARN_ON(found_start != 0); 426d3575156SNaohiro Aota return 0; 427d3575156SNaohiro Aota } 428d3575156SNaohiro Aota 4290f805531SAlex Lyakas /* 4300f805531SAlex Lyakas * Please do not consolidate these warnings into a single if. 4310f805531SAlex Lyakas * It is useful to know what went wrong. 4320f805531SAlex Lyakas */ 4330f805531SAlex Lyakas if (WARN_ON(found_start != start)) 4340f805531SAlex Lyakas return -EUCLEAN; 4350f805531SAlex Lyakas if (WARN_ON(!PageUptodate(page))) 4360f805531SAlex Lyakas return -EUCLEAN; 4370f805531SAlex Lyakas 438eca0f6f6SQu Wenruo return csum_one_extent_buffer(eb); 43919c00ddcSChris Mason } 44019c00ddcSChris Mason 441542e300eSChristoph Hellwig blk_status_t btree_csum_one_bio(struct btrfs_bio *bbio) 442deb6216fSChristoph Hellwig { 443542e300eSChristoph Hellwig struct btrfs_fs_info *fs_info = bbio->inode->root->fs_info; 444542e300eSChristoph Hellwig struct bvec_iter iter; 445542e300eSChristoph Hellwig struct bio_vec bv; 446deb6216fSChristoph Hellwig int ret = 0; 447deb6216fSChristoph Hellwig 448542e300eSChristoph Hellwig bio_for_each_segment(bv, &bbio->bio, iter) { 449542e300eSChristoph Hellwig ret = csum_dirty_buffer(fs_info, &bv); 450deb6216fSChristoph Hellwig if (ret) 451deb6216fSChristoph Hellwig break; 452deb6216fSChristoph Hellwig } 453deb6216fSChristoph Hellwig 454deb6216fSChristoph Hellwig return errno_to_blk_status(ret); 455deb6216fSChristoph Hellwig } 456deb6216fSChristoph Hellwig 457b0c9b3b0SDavid Sterba static int check_tree_block_fsid(struct extent_buffer *eb) 4582b82032cSYan Zheng { 459b0c9b3b0SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 460944d3f9fSNikolay Borisov struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 46144880fdcSAnand Jain u8 fsid[BTRFS_FSID_SIZE]; 462944d3f9fSNikolay Borisov u8 *metadata_uuid; 4632b82032cSYan Zheng 4649a8658e3SDavid Sterba read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid), 4659a8658e3SDavid Sterba BTRFS_FSID_SIZE); 4667239ff4bSNikolay Borisov /* 467944d3f9fSNikolay Borisov * Checking the incompat flag is only valid for the current fs. For 468944d3f9fSNikolay Borisov * seed devices it's forbidden to have their uuid changed so reading 469944d3f9fSNikolay Borisov * ->fsid in this case is fine 4707239ff4bSNikolay Borisov */ 471944d3f9fSNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID)) 4727239ff4bSNikolay Borisov metadata_uuid = fs_devices->metadata_uuid; 4737239ff4bSNikolay Borisov else 4747239ff4bSNikolay Borisov metadata_uuid = fs_devices->fsid; 4757239ff4bSNikolay Borisov 476944d3f9fSNikolay Borisov if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) 477944d3f9fSNikolay Borisov return 0; 478944d3f9fSNikolay Borisov 479944d3f9fSNikolay Borisov list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) 480944d3f9fSNikolay Borisov if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE)) 481944d3f9fSNikolay Borisov return 0; 482944d3f9fSNikolay Borisov 483944d3f9fSNikolay Borisov return 1; 4842b82032cSYan Zheng } 4852b82032cSYan Zheng 48677bf40a2SQu Wenruo /* Do basic extent buffer checks at read time */ 487947a6299SQu Wenruo static int validate_extent_buffer(struct extent_buffer *eb, 488947a6299SQu Wenruo struct btrfs_tree_parent_check *check) 489ce9adaa5SChris Mason { 49077bf40a2SQu Wenruo struct btrfs_fs_info *fs_info = eb->fs_info; 491ce9adaa5SChris Mason u64 found_start; 49277bf40a2SQu Wenruo const u32 csum_size = fs_info->csum_size; 49377bf40a2SQu Wenruo u8 found_level; 4942996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 495dfd29eedSDavid Sterba const u8 *header_csum; 49677bf40a2SQu Wenruo int ret = 0; 497ea466794SJosef Bacik 498947a6299SQu Wenruo ASSERT(check); 499947a6299SQu Wenruo 500ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 501727011e0SChris Mason if (found_start != eb->start) { 5028f0ed7d4SQu Wenruo btrfs_err_rl(fs_info, 5038f0ed7d4SQu Wenruo "bad tree block start, mirror %u want %llu have %llu", 5048f0ed7d4SQu Wenruo eb->read_mirror, eb->start, found_start); 505f188591eSChris Mason ret = -EIO; 50677bf40a2SQu Wenruo goto out; 507ce9adaa5SChris Mason } 508b0c9b3b0SDavid Sterba if (check_tree_block_fsid(eb)) { 5098f0ed7d4SQu Wenruo btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u", 5108f0ed7d4SQu Wenruo eb->start, eb->read_mirror); 5111259ab75SChris Mason ret = -EIO; 51277bf40a2SQu Wenruo goto out; 5131259ab75SChris Mason } 514ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 5151c24c3ceSJosef Bacik if (found_level >= BTRFS_MAX_LEVEL) { 5168f0ed7d4SQu Wenruo btrfs_err(fs_info, 5178f0ed7d4SQu Wenruo "bad tree block level, mirror %u level %d on logical %llu", 5188f0ed7d4SQu Wenruo eb->read_mirror, btrfs_header_level(eb), eb->start); 5191c24c3ceSJosef Bacik ret = -EIO; 52077bf40a2SQu Wenruo goto out; 5211c24c3ceSJosef Bacik } 522ce9adaa5SChris Mason 523c67b3892SDavid Sterba csum_tree_block(eb, result); 524dfd29eedSDavid Sterba header_csum = page_address(eb->pages[0]) + 525dfd29eedSDavid Sterba get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum)); 526a826d6dcSJosef Bacik 527dfd29eedSDavid Sterba if (memcmp(result, header_csum, csum_size) != 0) { 5282996e1f8SJohannes Thumshirn btrfs_warn_rl(fs_info, 5298f0ed7d4SQu Wenruo "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d", 5308f0ed7d4SQu Wenruo eb->start, eb->read_mirror, 531dfd29eedSDavid Sterba CSUM_FMT_VALUE(csum_size, header_csum), 53235be8851SJohannes Thumshirn CSUM_FMT_VALUE(csum_size, result), 53335be8851SJohannes Thumshirn btrfs_header_level(eb)); 5342996e1f8SJohannes Thumshirn ret = -EUCLEAN; 53577bf40a2SQu Wenruo goto out; 5362996e1f8SJohannes Thumshirn } 5372996e1f8SJohannes Thumshirn 538947a6299SQu Wenruo if (found_level != check->level) { 53977177ed1SQu Wenruo btrfs_err(fs_info, 54077177ed1SQu Wenruo "level verify failed on logical %llu mirror %u wanted %u found %u", 54177177ed1SQu Wenruo eb->start, eb->read_mirror, check->level, found_level); 542947a6299SQu Wenruo ret = -EIO; 543947a6299SQu Wenruo goto out; 544947a6299SQu Wenruo } 545947a6299SQu Wenruo if (unlikely(check->transid && 546947a6299SQu Wenruo btrfs_header_generation(eb) != check->transid)) { 547947a6299SQu Wenruo btrfs_err_rl(eb->fs_info, 548947a6299SQu Wenruo "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu", 549947a6299SQu Wenruo eb->start, eb->read_mirror, check->transid, 550947a6299SQu Wenruo btrfs_header_generation(eb)); 551947a6299SQu Wenruo ret = -EIO; 552947a6299SQu Wenruo goto out; 553947a6299SQu Wenruo } 554947a6299SQu Wenruo if (check->has_first_key) { 555947a6299SQu Wenruo struct btrfs_key *expect_key = &check->first_key; 556947a6299SQu Wenruo struct btrfs_key found_key; 557947a6299SQu Wenruo 558947a6299SQu Wenruo if (found_level) 559947a6299SQu Wenruo btrfs_node_key_to_cpu(eb, &found_key, 0); 560947a6299SQu Wenruo else 561947a6299SQu Wenruo btrfs_item_key_to_cpu(eb, &found_key, 0); 562947a6299SQu Wenruo if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) { 563947a6299SQu Wenruo btrfs_err(fs_info, 564947a6299SQu Wenruo "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", 565947a6299SQu Wenruo eb->start, check->transid, 566947a6299SQu Wenruo expect_key->objectid, 567947a6299SQu Wenruo expect_key->type, expect_key->offset, 568947a6299SQu Wenruo found_key.objectid, found_key.type, 569947a6299SQu Wenruo found_key.offset); 570947a6299SQu Wenruo ret = -EUCLEAN; 571947a6299SQu Wenruo goto out; 572947a6299SQu Wenruo } 573947a6299SQu Wenruo } 574947a6299SQu Wenruo if (check->owner_root) { 575947a6299SQu Wenruo ret = btrfs_check_eb_owner(eb, check->owner_root); 576947a6299SQu Wenruo if (ret < 0) 577947a6299SQu Wenruo goto out; 578947a6299SQu Wenruo } 579947a6299SQu Wenruo 580a826d6dcSJosef Bacik /* 581a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 582a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 583a826d6dcSJosef Bacik * return -EIO. 584a826d6dcSJosef Bacik */ 5851c4360eeSDavid Sterba if (found_level == 0 && btrfs_check_leaf_full(eb)) { 586a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 587a826d6dcSJosef Bacik ret = -EIO; 588a826d6dcSJosef Bacik } 589ce9adaa5SChris Mason 590813fd1dcSDavid Sterba if (found_level > 0 && btrfs_check_node(eb)) 591053ab70fSLiu Bo ret = -EIO; 592053ab70fSLiu Bo 5930b32f4bbSJosef Bacik if (!ret) 5940b32f4bbSJosef Bacik set_extent_buffer_uptodate(eb); 59575391f0dSQu Wenruo else 59675391f0dSQu Wenruo btrfs_err(fs_info, 5978f0ed7d4SQu Wenruo "read time tree block corruption detected on logical %llu mirror %u", 5988f0ed7d4SQu Wenruo eb->start, eb->read_mirror); 59977bf40a2SQu Wenruo out: 60077bf40a2SQu Wenruo return ret; 60177bf40a2SQu Wenruo } 60277bf40a2SQu Wenruo 603371cdc07SQu Wenruo static int validate_subpage_buffer(struct page *page, u64 start, u64 end, 604947a6299SQu Wenruo int mirror, struct btrfs_tree_parent_check *check) 605371cdc07SQu Wenruo { 606371cdc07SQu Wenruo struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); 607371cdc07SQu Wenruo struct extent_buffer *eb; 608371cdc07SQu Wenruo bool reads_done; 609371cdc07SQu Wenruo int ret = 0; 610371cdc07SQu Wenruo 611947a6299SQu Wenruo ASSERT(check); 612947a6299SQu Wenruo 613371cdc07SQu Wenruo /* 614371cdc07SQu Wenruo * We don't allow bio merge for subpage metadata read, so we should 615371cdc07SQu Wenruo * only get one eb for each endio hook. 616371cdc07SQu Wenruo */ 617371cdc07SQu Wenruo ASSERT(end == start + fs_info->nodesize - 1); 618371cdc07SQu Wenruo ASSERT(PagePrivate(page)); 619371cdc07SQu Wenruo 620371cdc07SQu Wenruo eb = find_extent_buffer(fs_info, start); 621371cdc07SQu Wenruo /* 622371cdc07SQu Wenruo * When we are reading one tree block, eb must have been inserted into 623371cdc07SQu Wenruo * the radix tree. If not, something is wrong. 624371cdc07SQu Wenruo */ 625371cdc07SQu Wenruo ASSERT(eb); 626371cdc07SQu Wenruo 627371cdc07SQu Wenruo reads_done = atomic_dec_and_test(&eb->io_pages); 628371cdc07SQu Wenruo /* Subpage read must finish in page read */ 629371cdc07SQu Wenruo ASSERT(reads_done); 630371cdc07SQu Wenruo 631371cdc07SQu Wenruo eb->read_mirror = mirror; 632371cdc07SQu Wenruo if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 633371cdc07SQu Wenruo ret = -EIO; 634371cdc07SQu Wenruo goto err; 635371cdc07SQu Wenruo } 636947a6299SQu Wenruo ret = validate_extent_buffer(eb, check); 637371cdc07SQu Wenruo if (ret < 0) 638371cdc07SQu Wenruo goto err; 639371cdc07SQu Wenruo 640371cdc07SQu Wenruo set_extent_buffer_uptodate(eb); 641371cdc07SQu Wenruo 642371cdc07SQu Wenruo free_extent_buffer(eb); 643371cdc07SQu Wenruo return ret; 644371cdc07SQu Wenruo err: 645371cdc07SQu Wenruo /* 646371cdc07SQu Wenruo * end_bio_extent_readpage decrements io_pages in case of error, 647371cdc07SQu Wenruo * make sure it has something to decrement. 648371cdc07SQu Wenruo */ 649371cdc07SQu Wenruo atomic_inc(&eb->io_pages); 650371cdc07SQu Wenruo clear_extent_buffer_uptodate(eb); 651371cdc07SQu Wenruo free_extent_buffer(eb); 652371cdc07SQu Wenruo return ret; 653371cdc07SQu Wenruo } 654371cdc07SQu Wenruo 655c3a3b19bSQu Wenruo int btrfs_validate_metadata_buffer(struct btrfs_bio *bbio, 65677bf40a2SQu Wenruo struct page *page, u64 start, u64 end, 65777bf40a2SQu Wenruo int mirror) 65877bf40a2SQu Wenruo { 65977bf40a2SQu Wenruo struct extent_buffer *eb; 66077bf40a2SQu Wenruo int ret = 0; 66177bf40a2SQu Wenruo int reads_done; 66277bf40a2SQu Wenruo 66377bf40a2SQu Wenruo ASSERT(page->private); 664371cdc07SQu Wenruo 665fbca46ebSQu Wenruo if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE) 666947a6299SQu Wenruo return validate_subpage_buffer(page, start, end, mirror, 667947a6299SQu Wenruo &bbio->parent_check); 668371cdc07SQu Wenruo 66977bf40a2SQu Wenruo eb = (struct extent_buffer *)page->private; 67077bf40a2SQu Wenruo 67177bf40a2SQu Wenruo /* 67277bf40a2SQu Wenruo * The pending IO might have been the only thing that kept this buffer 67377bf40a2SQu Wenruo * in memory. Make sure we have a ref for all this other checks 67477bf40a2SQu Wenruo */ 67577bf40a2SQu Wenruo atomic_inc(&eb->refs); 67677bf40a2SQu Wenruo 67777bf40a2SQu Wenruo reads_done = atomic_dec_and_test(&eb->io_pages); 67877bf40a2SQu Wenruo if (!reads_done) 67977bf40a2SQu Wenruo goto err; 68077bf40a2SQu Wenruo 68177bf40a2SQu Wenruo eb->read_mirror = mirror; 68277bf40a2SQu Wenruo if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 68377bf40a2SQu Wenruo ret = -EIO; 68477bf40a2SQu Wenruo goto err; 68577bf40a2SQu Wenruo } 686947a6299SQu Wenruo ret = validate_extent_buffer(eb, &bbio->parent_check); 687ce9adaa5SChris Mason err: 68853b381b3SDavid Woodhouse if (ret) { 68953b381b3SDavid Woodhouse /* 69053b381b3SDavid Woodhouse * our io error hook is going to dec the io pages 69153b381b3SDavid Woodhouse * again, we have to make sure it has something 69253b381b3SDavid Woodhouse * to decrement 69353b381b3SDavid Woodhouse */ 69453b381b3SDavid Woodhouse atomic_inc(&eb->io_pages); 6950b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 69653b381b3SDavid Woodhouse } 6970b32f4bbSJosef Bacik free_extent_buffer(eb); 69877bf40a2SQu Wenruo 699f188591eSChris Mason return ret; 700ce9adaa5SChris Mason } 701ce9adaa5SChris Mason 7023dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 7038958b551SMatthew Wilcox (Oracle) static int btree_migrate_folio(struct address_space *mapping, 7048958b551SMatthew Wilcox (Oracle) struct folio *dst, struct folio *src, enum migrate_mode mode) 705784b4e29SChris Mason { 706784b4e29SChris Mason /* 707784b4e29SChris Mason * we can't safely write a btree page from here, 708784b4e29SChris Mason * we haven't done the locking hook 709784b4e29SChris Mason */ 7108958b551SMatthew Wilcox (Oracle) if (folio_test_dirty(src)) 711784b4e29SChris Mason return -EAGAIN; 712784b4e29SChris Mason /* 713784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 714784b4e29SChris Mason * We must have no buffers or drop them. 715784b4e29SChris Mason */ 7168958b551SMatthew Wilcox (Oracle) if (folio_get_private(src) && 7178958b551SMatthew Wilcox (Oracle) !filemap_release_folio(src, GFP_KERNEL)) 718784b4e29SChris Mason return -EAGAIN; 71954184650SMatthew Wilcox (Oracle) return migrate_folio(mapping, dst, src, mode); 720784b4e29SChris Mason } 7218958b551SMatthew Wilcox (Oracle) #else 7228958b551SMatthew Wilcox (Oracle) #define btree_migrate_folio NULL 7233dd1462eSJan Beulich #endif 724784b4e29SChris Mason 7250da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 7260da5468fSChris Mason struct writeback_control *wbc) 7270da5468fSChris Mason { 728e2d84521SMiao Xie struct btrfs_fs_info *fs_info; 729e2d84521SMiao Xie int ret; 730e2d84521SMiao Xie 731d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 732448d640bSChris Mason 733448d640bSChris Mason if (wbc->for_kupdate) 734448d640bSChris Mason return 0; 735448d640bSChris Mason 736e2d84521SMiao Xie fs_info = BTRFS_I(mapping->host)->root->fs_info; 737b9473439SChris Mason /* this is a bit racy, but that's ok */ 738d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 739d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 740d814a491SEthan Lien fs_info->dirty_metadata_batch); 741e2d84521SMiao Xie if (ret < 0) 742793955bcSChris Mason return 0; 743793955bcSChris Mason } 7440b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 7450da5468fSChris Mason } 7460da5468fSChris Mason 747f913cff3SMatthew Wilcox (Oracle) static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags) 7485f39d397SChris Mason { 749f913cff3SMatthew Wilcox (Oracle) if (folio_test_writeback(folio) || folio_test_dirty(folio)) 750f913cff3SMatthew Wilcox (Oracle) return false; 7510c4e538bSDavid Sterba 752f913cff3SMatthew Wilcox (Oracle) return try_release_extent_buffer(&folio->page); 753d98237b3SChris Mason } 754d98237b3SChris Mason 755895586ebSMatthew Wilcox (Oracle) static void btree_invalidate_folio(struct folio *folio, size_t offset, 756895586ebSMatthew Wilcox (Oracle) size_t length) 757d98237b3SChris Mason { 758d1310b2eSChris Mason struct extent_io_tree *tree; 759895586ebSMatthew Wilcox (Oracle) tree = &BTRFS_I(folio->mapping->host)->io_tree; 760895586ebSMatthew Wilcox (Oracle) extent_invalidate_folio(tree, folio, offset); 761f913cff3SMatthew Wilcox (Oracle) btree_release_folio(folio, GFP_NOFS); 762895586ebSMatthew Wilcox (Oracle) if (folio_get_private(folio)) { 763895586ebSMatthew Wilcox (Oracle) btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info, 764895586ebSMatthew Wilcox (Oracle) "folio private not zero on folio %llu", 765895586ebSMatthew Wilcox (Oracle) (unsigned long long)folio_pos(folio)); 766895586ebSMatthew Wilcox (Oracle) folio_detach_private(folio); 7679ad6b7bcSChris Mason } 768d98237b3SChris Mason } 769d98237b3SChris Mason 770bb146eb2SJosef Bacik #ifdef DEBUG 7710079c3b1SMatthew Wilcox (Oracle) static bool btree_dirty_folio(struct address_space *mapping, 7720079c3b1SMatthew Wilcox (Oracle) struct folio *folio) 7730079c3b1SMatthew Wilcox (Oracle) { 7740079c3b1SMatthew Wilcox (Oracle) struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb); 775139e8cd3SQu Wenruo struct btrfs_subpage *subpage; 7760b32f4bbSJosef Bacik struct extent_buffer *eb; 777139e8cd3SQu Wenruo int cur_bit = 0; 7780079c3b1SMatthew Wilcox (Oracle) u64 page_start = folio_pos(folio); 7790b32f4bbSJosef Bacik 780139e8cd3SQu Wenruo if (fs_info->sectorsize == PAGE_SIZE) { 7810079c3b1SMatthew Wilcox (Oracle) eb = folio_get_private(folio); 7820b32f4bbSJosef Bacik BUG_ON(!eb); 7830b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 7840b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 78549d0c642SFilipe Manana btrfs_assert_tree_write_locked(eb); 7860079c3b1SMatthew Wilcox (Oracle) return filemap_dirty_folio(mapping, folio); 787139e8cd3SQu Wenruo } 7880079c3b1SMatthew Wilcox (Oracle) subpage = folio_get_private(folio); 789139e8cd3SQu Wenruo 790139e8cd3SQu Wenruo ASSERT(subpage->dirty_bitmap); 791139e8cd3SQu Wenruo while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) { 792139e8cd3SQu Wenruo unsigned long flags; 793139e8cd3SQu Wenruo u64 cur; 794139e8cd3SQu Wenruo u16 tmp = (1 << cur_bit); 795139e8cd3SQu Wenruo 796139e8cd3SQu Wenruo spin_lock_irqsave(&subpage->lock, flags); 797139e8cd3SQu Wenruo if (!(tmp & subpage->dirty_bitmap)) { 798139e8cd3SQu Wenruo spin_unlock_irqrestore(&subpage->lock, flags); 799139e8cd3SQu Wenruo cur_bit++; 800139e8cd3SQu Wenruo continue; 801139e8cd3SQu Wenruo } 802139e8cd3SQu Wenruo spin_unlock_irqrestore(&subpage->lock, flags); 803139e8cd3SQu Wenruo cur = page_start + cur_bit * fs_info->sectorsize; 804139e8cd3SQu Wenruo 805139e8cd3SQu Wenruo eb = find_extent_buffer(fs_info, cur); 806139e8cd3SQu Wenruo ASSERT(eb); 807139e8cd3SQu Wenruo ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 808139e8cd3SQu Wenruo ASSERT(atomic_read(&eb->refs)); 80949d0c642SFilipe Manana btrfs_assert_tree_write_locked(eb); 810139e8cd3SQu Wenruo free_extent_buffer(eb); 811139e8cd3SQu Wenruo 812139e8cd3SQu Wenruo cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits); 813139e8cd3SQu Wenruo } 8140079c3b1SMatthew Wilcox (Oracle) return filemap_dirty_folio(mapping, folio); 8150b32f4bbSJosef Bacik } 8160079c3b1SMatthew Wilcox (Oracle) #else 8170079c3b1SMatthew Wilcox (Oracle) #define btree_dirty_folio filemap_dirty_folio 8180079c3b1SMatthew Wilcox (Oracle) #endif 8190b32f4bbSJosef Bacik 8207f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 8210da5468fSChris Mason .writepages = btree_writepages, 822f913cff3SMatthew Wilcox (Oracle) .release_folio = btree_release_folio, 823895586ebSMatthew Wilcox (Oracle) .invalidate_folio = btree_invalidate_folio, 8248958b551SMatthew Wilcox (Oracle) .migrate_folio = btree_migrate_folio, 8250079c3b1SMatthew Wilcox (Oracle) .dirty_folio = btree_dirty_folio, 826d98237b3SChris Mason }; 827123abc88SChris Mason 8282ff7e61eSJeff Mahoney struct extent_buffer *btrfs_find_create_tree_block( 8292ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info, 8303fbaf258SJosef Bacik u64 bytenr, u64 owner_root, 8313fbaf258SJosef Bacik int level) 8320999df54SChris Mason { 8330b246afaSJeff Mahoney if (btrfs_is_testing(fs_info)) 8340b246afaSJeff Mahoney return alloc_test_extent_buffer(fs_info, bytenr); 8353fbaf258SJosef Bacik return alloc_extent_buffer(fs_info, bytenr, owner_root, level); 8360999df54SChris Mason } 8370999df54SChris Mason 838581c1760SQu Wenruo /* 839581c1760SQu Wenruo * Read tree block at logical address @bytenr and do variant basic but critical 840581c1760SQu Wenruo * verification. 841581c1760SQu Wenruo * 842789d6a3aSQu Wenruo * @check: expected tree parentness check, see comments of the 843789d6a3aSQu Wenruo * structure for details. 844581c1760SQu Wenruo */ 8452ff7e61eSJeff Mahoney struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, 846789d6a3aSQu Wenruo struct btrfs_tree_parent_check *check) 847e20d96d6SChris Mason { 8485f39d397SChris Mason struct extent_buffer *buf = NULL; 84919c00ddcSChris Mason int ret; 85019c00ddcSChris Mason 851789d6a3aSQu Wenruo ASSERT(check); 852789d6a3aSQu Wenruo 853789d6a3aSQu Wenruo buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root, 854789d6a3aSQu Wenruo check->level); 855c871b0f2SLiu Bo if (IS_ERR(buf)) 856c871b0f2SLiu Bo return buf; 857e4204dedSChris Mason 858789d6a3aSQu Wenruo ret = btrfs_read_extent_buffer(buf, check); 8590f0fe8f7SFilipe David Borba Manana if (ret) { 860537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 86164c043deSLiu Bo return ERR_PTR(ret); 8620f0fe8f7SFilipe David Borba Manana } 863789d6a3aSQu Wenruo if (btrfs_check_eb_owner(buf, check->owner_root)) { 86488c602abSQu Wenruo free_extent_buffer_stale(buf); 86588c602abSQu Wenruo return ERR_PTR(-EUCLEAN); 86688c602abSQu Wenruo } 8675f39d397SChris Mason return buf; 868ce9adaa5SChris Mason 869eb60ceacSChris Mason } 870eb60ceacSChris Mason 871da17066cSJeff Mahoney static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, 872e20d96d6SChris Mason u64 objectid) 873d97e63b6SChris Mason { 8747c0260eeSJeff Mahoney bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); 8752e608bd1SJosef Bacik 8762e608bd1SJosef Bacik memset(&root->root_key, 0, sizeof(root->root_key)); 8772e608bd1SJosef Bacik memset(&root->root_item, 0, sizeof(root->root_item)); 8782e608bd1SJosef Bacik memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 87996dfcb46SJosef Bacik root->fs_info = fs_info; 8802e608bd1SJosef Bacik root->root_key.objectid = objectid; 881cfaa7295SChris Mason root->node = NULL; 882a28ec197SChris Mason root->commit_root = NULL; 88327cdeb70SMiao Xie root->state = 0; 884abed4aaaSJosef Bacik RB_CLEAR_NODE(&root->rb_node); 8850b86a832SChris Mason 8860f7d52f4SChris Mason root->last_trans = 0; 8876b8fad57SNikolay Borisov root->free_objectid = 0; 888eb73c1b7SMiao Xie root->nr_delalloc_inodes = 0; 889199c2a9cSMiao Xie root->nr_ordered_extents = 0; 8906bef4d31SEric Paris root->inode_tree = RB_ROOT; 891088aea3bSDavid Sterba INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 8922e608bd1SJosef Bacik 8932e608bd1SJosef Bacik btrfs_init_root_block_rsv(root); 8940b86a832SChris Mason 8950b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 8965d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 897eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_inodes); 898eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_root); 899199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_extents); 900199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_root); 901d2311e69SQu Wenruo INIT_LIST_HEAD(&root->reloc_dirty_list); 9022ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 9032ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 9045d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 905eb73c1b7SMiao Xie spin_lock_init(&root->delalloc_lock); 906199c2a9cSMiao Xie spin_lock_init(&root->ordered_extent_lock); 907f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 9082ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 9092ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 9108287475aSQu Wenruo spin_lock_init(&root->qgroup_meta_rsv_lock); 911a2135011SChris Mason mutex_init(&root->objectid_mutex); 912e02119d5SChris Mason mutex_init(&root->log_mutex); 91331f3d255SMiao Xie mutex_init(&root->ordered_extent_mutex); 914573bfb72SMiao Xie mutex_init(&root->delalloc_mutex); 915c53e9653SQu Wenruo init_waitqueue_head(&root->qgroup_flush_wait); 9167237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 9177237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 9187237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 9198b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[0]); 9208b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[1]); 9217237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 9227237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 9237237f183SYan Zheng atomic_set(&root->log_writers, 0); 9242ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 9250700cea7SElena Reshetova refcount_set(&root->refs, 1); 9268ecebf4dSRobbie Ko atomic_set(&root->snapshot_force_cow, 0); 927eede2bf3SOmar Sandoval atomic_set(&root->nr_swapfiles, 0); 9287237f183SYan Zheng root->log_transid = 0; 929d1433debSMiao Xie root->log_transid_committed = -1; 930257c62e1SChris Mason root->last_log_commit = 0; 9312e608bd1SJosef Bacik root->anon_dev = 0; 932e289f03eSFilipe Manana if (!dummy) { 93343eb5f29SQu Wenruo extent_io_tree_init(fs_info, &root->dirty_log_pages, 93435da5a7eSDavid Sterba IO_TREE_ROOT_DIRTY_LOG_PAGES); 935e289f03eSFilipe Manana extent_io_tree_init(fs_info, &root->log_csum_range, 93635da5a7eSDavid Sterba IO_TREE_LOG_CSUM_RANGE); 937e289f03eSFilipe Manana } 938017e5369SChris Mason 9395f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 940370a11b8SQu Wenruo btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks); 941bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 942bd647ce3SJosef Bacik INIT_LIST_HEAD(&root->leak_list); 943fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 944bd647ce3SJosef Bacik list_add_tail(&root->leak_list, &fs_info->allocated_roots); 945fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 946bd647ce3SJosef Bacik #endif 9473768f368SChris Mason } 9483768f368SChris Mason 94974e4d827SDavid Sterba static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, 95096dfcb46SJosef Bacik u64 objectid, gfp_t flags) 9516f07e42eSAl Viro { 95274e4d827SDavid Sterba struct btrfs_root *root = kzalloc(sizeof(*root), flags); 9536f07e42eSAl Viro if (root) 95496dfcb46SJosef Bacik __setup_root(root, fs_info, objectid); 9556f07e42eSAl Viro return root; 9566f07e42eSAl Viro } 9576f07e42eSAl Viro 95806ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 95906ea65a3SJosef Bacik /* Should only be used by the testing infrastructure */ 960da17066cSJeff Mahoney struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) 96106ea65a3SJosef Bacik { 96206ea65a3SJosef Bacik struct btrfs_root *root; 96306ea65a3SJosef Bacik 9647c0260eeSJeff Mahoney if (!fs_info) 9657c0260eeSJeff Mahoney return ERR_PTR(-EINVAL); 9667c0260eeSJeff Mahoney 96796dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL); 96806ea65a3SJosef Bacik if (!root) 96906ea65a3SJosef Bacik return ERR_PTR(-ENOMEM); 970da17066cSJeff Mahoney 971b9ef22deSFeifei Xu /* We don't use the stripesize in selftest, set it as sectorsize */ 972faa2dbf0SJosef Bacik root->alloc_bytenr = 0; 97306ea65a3SJosef Bacik 97406ea65a3SJosef Bacik return root; 97506ea65a3SJosef Bacik } 97606ea65a3SJosef Bacik #endif 97706ea65a3SJosef Bacik 978abed4aaaSJosef Bacik static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node) 979abed4aaaSJosef Bacik { 980abed4aaaSJosef Bacik const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node); 981abed4aaaSJosef Bacik const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node); 982abed4aaaSJosef Bacik 983abed4aaaSJosef Bacik return btrfs_comp_cpu_keys(&a->root_key, &b->root_key); 984abed4aaaSJosef Bacik } 985abed4aaaSJosef Bacik 986abed4aaaSJosef Bacik static int global_root_key_cmp(const void *k, const struct rb_node *node) 987abed4aaaSJosef Bacik { 988abed4aaaSJosef Bacik const struct btrfs_key *key = k; 989abed4aaaSJosef Bacik const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node); 990abed4aaaSJosef Bacik 991abed4aaaSJosef Bacik return btrfs_comp_cpu_keys(key, &root->root_key); 992abed4aaaSJosef Bacik } 993abed4aaaSJosef Bacik 994abed4aaaSJosef Bacik int btrfs_global_root_insert(struct btrfs_root *root) 995abed4aaaSJosef Bacik { 996abed4aaaSJosef Bacik struct btrfs_fs_info *fs_info = root->fs_info; 997abed4aaaSJosef Bacik struct rb_node *tmp; 998abed4aaaSJosef Bacik 999abed4aaaSJosef Bacik write_lock(&fs_info->global_root_lock); 1000abed4aaaSJosef Bacik tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp); 1001abed4aaaSJosef Bacik write_unlock(&fs_info->global_root_lock); 1002abed4aaaSJosef Bacik ASSERT(!tmp); 1003abed4aaaSJosef Bacik 1004abed4aaaSJosef Bacik return tmp ? -EEXIST : 0; 1005abed4aaaSJosef Bacik } 1006abed4aaaSJosef Bacik 1007abed4aaaSJosef Bacik void btrfs_global_root_delete(struct btrfs_root *root) 1008abed4aaaSJosef Bacik { 1009abed4aaaSJosef Bacik struct btrfs_fs_info *fs_info = root->fs_info; 1010abed4aaaSJosef Bacik 1011abed4aaaSJosef Bacik write_lock(&fs_info->global_root_lock); 1012abed4aaaSJosef Bacik rb_erase(&root->rb_node, &fs_info->global_root_tree); 1013abed4aaaSJosef Bacik write_unlock(&fs_info->global_root_lock); 1014abed4aaaSJosef Bacik } 1015abed4aaaSJosef Bacik 1016abed4aaaSJosef Bacik struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info, 1017abed4aaaSJosef Bacik struct btrfs_key *key) 1018abed4aaaSJosef Bacik { 1019abed4aaaSJosef Bacik struct rb_node *node; 1020abed4aaaSJosef Bacik struct btrfs_root *root = NULL; 1021abed4aaaSJosef Bacik 1022abed4aaaSJosef Bacik read_lock(&fs_info->global_root_lock); 1023abed4aaaSJosef Bacik node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp); 1024abed4aaaSJosef Bacik if (node) 1025abed4aaaSJosef Bacik root = container_of(node, struct btrfs_root, rb_node); 1026abed4aaaSJosef Bacik read_unlock(&fs_info->global_root_lock); 1027abed4aaaSJosef Bacik 1028abed4aaaSJosef Bacik return root; 1029abed4aaaSJosef Bacik } 1030abed4aaaSJosef Bacik 1031f7238e50SJosef Bacik static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr) 1032f7238e50SJosef Bacik { 1033f7238e50SJosef Bacik struct btrfs_block_group *block_group; 1034f7238e50SJosef Bacik u64 ret; 1035f7238e50SJosef Bacik 1036f7238e50SJosef Bacik if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) 1037f7238e50SJosef Bacik return 0; 1038f7238e50SJosef Bacik 1039f7238e50SJosef Bacik if (bytenr) 1040f7238e50SJosef Bacik block_group = btrfs_lookup_block_group(fs_info, bytenr); 1041f7238e50SJosef Bacik else 1042f7238e50SJosef Bacik block_group = btrfs_lookup_first_block_group(fs_info, bytenr); 1043f7238e50SJosef Bacik ASSERT(block_group); 1044f7238e50SJosef Bacik if (!block_group) 1045f7238e50SJosef Bacik return 0; 1046f7238e50SJosef Bacik ret = block_group->global_root_id; 1047f7238e50SJosef Bacik btrfs_put_block_group(block_group); 1048f7238e50SJosef Bacik 1049f7238e50SJosef Bacik return ret; 1050f7238e50SJosef Bacik } 1051f7238e50SJosef Bacik 1052abed4aaaSJosef Bacik struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr) 1053abed4aaaSJosef Bacik { 1054abed4aaaSJosef Bacik struct btrfs_key key = { 1055abed4aaaSJosef Bacik .objectid = BTRFS_CSUM_TREE_OBJECTID, 1056abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 1057f7238e50SJosef Bacik .offset = btrfs_global_root_id(fs_info, bytenr), 1058abed4aaaSJosef Bacik }; 1059abed4aaaSJosef Bacik 1060abed4aaaSJosef Bacik return btrfs_global_root(fs_info, &key); 1061abed4aaaSJosef Bacik } 1062abed4aaaSJosef Bacik 1063abed4aaaSJosef Bacik struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr) 1064abed4aaaSJosef Bacik { 1065abed4aaaSJosef Bacik struct btrfs_key key = { 1066abed4aaaSJosef Bacik .objectid = BTRFS_EXTENT_TREE_OBJECTID, 1067abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 1068f7238e50SJosef Bacik .offset = btrfs_global_root_id(fs_info, bytenr), 1069abed4aaaSJosef Bacik }; 1070abed4aaaSJosef Bacik 1071abed4aaaSJosef Bacik return btrfs_global_root(fs_info, &key); 1072abed4aaaSJosef Bacik } 1073abed4aaaSJosef Bacik 107451129b33SJosef Bacik struct btrfs_root *btrfs_block_group_root(struct btrfs_fs_info *fs_info) 107551129b33SJosef Bacik { 107651129b33SJosef Bacik if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) 107751129b33SJosef Bacik return fs_info->block_group_root; 107851129b33SJosef Bacik return btrfs_extent_root(fs_info, 0); 107951129b33SJosef Bacik } 108051129b33SJosef Bacik 108120897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 108220897f5cSArne Jansen u64 objectid) 108320897f5cSArne Jansen { 10849b7a2440SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 108520897f5cSArne Jansen struct extent_buffer *leaf; 108620897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 108720897f5cSArne Jansen struct btrfs_root *root; 108820897f5cSArne Jansen struct btrfs_key key; 1089b89f6d1fSFilipe Manana unsigned int nofs_flag; 109020897f5cSArne Jansen int ret = 0; 109120897f5cSArne Jansen 1092b89f6d1fSFilipe Manana /* 1093b89f6d1fSFilipe Manana * We're holding a transaction handle, so use a NOFS memory allocation 1094b89f6d1fSFilipe Manana * context to avoid deadlock if reclaim happens. 1095b89f6d1fSFilipe Manana */ 1096b89f6d1fSFilipe Manana nofs_flag = memalloc_nofs_save(); 109796dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL); 1098b89f6d1fSFilipe Manana memalloc_nofs_restore(nofs_flag); 109920897f5cSArne Jansen if (!root) 110020897f5cSArne Jansen return ERR_PTR(-ENOMEM); 110120897f5cSArne Jansen 110220897f5cSArne Jansen root->root_key.objectid = objectid; 110320897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 110420897f5cSArne Jansen root->root_key.offset = 0; 110520897f5cSArne Jansen 11069631e4ccSJosef Bacik leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0, 11079631e4ccSJosef Bacik BTRFS_NESTING_NORMAL); 110820897f5cSArne Jansen if (IS_ERR(leaf)) { 110920897f5cSArne Jansen ret = PTR_ERR(leaf); 11101dd05682STsutomu Itoh leaf = NULL; 1111c1b07854SPeng Hao goto fail; 111220897f5cSArne Jansen } 111320897f5cSArne Jansen 111420897f5cSArne Jansen root->node = leaf; 111520897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 111620897f5cSArne Jansen 111720897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 111827cdeb70SMiao Xie set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 111920897f5cSArne Jansen 1120f944d2cbSDavid Sterba btrfs_set_root_flags(&root->root_item, 0); 1121f944d2cbSDavid Sterba btrfs_set_root_limit(&root->root_item, 0); 112220897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 112320897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 112420897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 112520897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 112620897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 112720897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 112820897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 112933d85fdaSQu Wenruo if (is_fstree(objectid)) 1130807fc790SAndy Shevchenko generate_random_guid(root->root_item.uuid); 1131807fc790SAndy Shevchenko else 1132807fc790SAndy Shevchenko export_guid(root->root_item.uuid, &guid_null); 1133c8422684SDavid Sterba btrfs_set_root_drop_level(&root->root_item, 0); 113420897f5cSArne Jansen 11358a6a87cdSBoris Burkov btrfs_tree_unlock(leaf); 11368a6a87cdSBoris Burkov 113720897f5cSArne Jansen key.objectid = objectid; 113820897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 113920897f5cSArne Jansen key.offset = 0; 114020897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 114120897f5cSArne Jansen if (ret) 114220897f5cSArne Jansen goto fail; 114320897f5cSArne Jansen 114420897f5cSArne Jansen return root; 11451dd05682STsutomu Itoh 11468a6a87cdSBoris Burkov fail: 114700246528SJosef Bacik btrfs_put_root(root); 11481dd05682STsutomu Itoh 11491dd05682STsutomu Itoh return ERR_PTR(ret); 115020897f5cSArne Jansen } 115120897f5cSArne Jansen 11527237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1153e02119d5SChris Mason struct btrfs_fs_info *fs_info) 11540f7d52f4SChris Mason { 11550f7d52f4SChris Mason struct btrfs_root *root; 1156e02119d5SChris Mason 115796dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS); 1158e02119d5SChris Mason if (!root) 11597237f183SYan Zheng return ERR_PTR(-ENOMEM); 1160e02119d5SChris Mason 1161e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1162e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1163e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 116427cdeb70SMiao Xie 11656ab6ebb7SNaohiro Aota return root; 11666ab6ebb7SNaohiro Aota } 11676ab6ebb7SNaohiro Aota 11686ab6ebb7SNaohiro Aota int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans, 11696ab6ebb7SNaohiro Aota struct btrfs_root *root) 11706ab6ebb7SNaohiro Aota { 11716ab6ebb7SNaohiro Aota struct extent_buffer *leaf; 11726ab6ebb7SNaohiro Aota 11737237f183SYan Zheng /* 117492a7cc42SQu Wenruo * DON'T set SHAREABLE bit for log trees. 117527cdeb70SMiao Xie * 117692a7cc42SQu Wenruo * Log trees are not exposed to user space thus can't be snapshotted, 117792a7cc42SQu Wenruo * and they go away before a real commit is actually done. 117892a7cc42SQu Wenruo * 117992a7cc42SQu Wenruo * They do store pointers to file data extents, and those reference 118092a7cc42SQu Wenruo * counts still get updated (along with back refs to the log tree). 11817237f183SYan Zheng */ 1182e02119d5SChris Mason 11834d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 11849631e4ccSJosef Bacik NULL, 0, 0, 0, BTRFS_NESTING_NORMAL); 11856ab6ebb7SNaohiro Aota if (IS_ERR(leaf)) 11866ab6ebb7SNaohiro Aota return PTR_ERR(leaf); 1187e02119d5SChris Mason 11887237f183SYan Zheng root->node = leaf; 1189e02119d5SChris Mason 1190e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1191e02119d5SChris Mason btrfs_tree_unlock(root->node); 11926ab6ebb7SNaohiro Aota 11936ab6ebb7SNaohiro Aota return 0; 11947237f183SYan Zheng } 11957237f183SYan Zheng 11967237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 11977237f183SYan Zheng struct btrfs_fs_info *fs_info) 11987237f183SYan Zheng { 11997237f183SYan Zheng struct btrfs_root *log_root; 12007237f183SYan Zheng 12017237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 12027237f183SYan Zheng if (IS_ERR(log_root)) 12037237f183SYan Zheng return PTR_ERR(log_root); 12046ab6ebb7SNaohiro Aota 12053ddebf27SNaohiro Aota if (!btrfs_is_zoned(fs_info)) { 12063ddebf27SNaohiro Aota int ret = btrfs_alloc_log_tree_node(trans, log_root); 12073ddebf27SNaohiro Aota 12086ab6ebb7SNaohiro Aota if (ret) { 12096ab6ebb7SNaohiro Aota btrfs_put_root(log_root); 12106ab6ebb7SNaohiro Aota return ret; 12116ab6ebb7SNaohiro Aota } 12123ddebf27SNaohiro Aota } 12136ab6ebb7SNaohiro Aota 12147237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 12157237f183SYan Zheng fs_info->log_root_tree = log_root; 12167237f183SYan Zheng return 0; 12177237f183SYan Zheng } 12187237f183SYan Zheng 12197237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 12207237f183SYan Zheng struct btrfs_root *root) 12217237f183SYan Zheng { 12220b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12237237f183SYan Zheng struct btrfs_root *log_root; 12247237f183SYan Zheng struct btrfs_inode_item *inode_item; 12256ab6ebb7SNaohiro Aota int ret; 12267237f183SYan Zheng 12270b246afaSJeff Mahoney log_root = alloc_log_tree(trans, fs_info); 12287237f183SYan Zheng if (IS_ERR(log_root)) 12297237f183SYan Zheng return PTR_ERR(log_root); 12307237f183SYan Zheng 12316ab6ebb7SNaohiro Aota ret = btrfs_alloc_log_tree_node(trans, log_root); 12326ab6ebb7SNaohiro Aota if (ret) { 12336ab6ebb7SNaohiro Aota btrfs_put_root(log_root); 12346ab6ebb7SNaohiro Aota return ret; 12356ab6ebb7SNaohiro Aota } 12366ab6ebb7SNaohiro Aota 12377237f183SYan Zheng log_root->last_trans = trans->transid; 12387237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 12397237f183SYan Zheng 12407237f183SYan Zheng inode_item = &log_root->root_item.inode; 12413cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 12423cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 12433cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 1244da17066cSJeff Mahoney btrfs_set_stack_inode_nbytes(inode_item, 12450b246afaSJeff Mahoney fs_info->nodesize); 12463cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 12477237f183SYan Zheng 12485d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 12497237f183SYan Zheng 12507237f183SYan Zheng WARN_ON(root->log_root); 12517237f183SYan Zheng root->log_root = log_root; 12527237f183SYan Zheng root->log_transid = 0; 1253d1433debSMiao Xie root->log_transid_committed = -1; 1254257c62e1SChris Mason root->last_log_commit = 0; 1255e02119d5SChris Mason return 0; 1256e02119d5SChris Mason } 1257e02119d5SChris Mason 125849d11beaSJosef Bacik static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root, 125949d11beaSJosef Bacik struct btrfs_path *path, 1260cb517eabSMiao Xie struct btrfs_key *key) 1261e02119d5SChris Mason { 1262e02119d5SChris Mason struct btrfs_root *root; 1263789d6a3aSQu Wenruo struct btrfs_tree_parent_check check = { 0 }; 1264e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 126584234f3aSYan Zheng u64 generation; 1266cb517eabSMiao Xie int ret; 1267581c1760SQu Wenruo int level; 1268cb517eabSMiao Xie 126996dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS); 127049d11beaSJosef Bacik if (!root) 127149d11beaSJosef Bacik return ERR_PTR(-ENOMEM); 12720f7d52f4SChris Mason 1273cb517eabSMiao Xie ret = btrfs_find_root(tree_root, key, path, 1274cb517eabSMiao Xie &root->root_item, &root->root_key); 12750f7d52f4SChris Mason if (ret) { 127613a8a7c8SYan, Zheng if (ret > 0) 127713a8a7c8SYan, Zheng ret = -ENOENT; 127849d11beaSJosef Bacik goto fail; 12790f7d52f4SChris Mason } 128013a8a7c8SYan, Zheng 128184234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1282581c1760SQu Wenruo level = btrfs_root_level(&root->root_item); 1283789d6a3aSQu Wenruo check.level = level; 1284789d6a3aSQu Wenruo check.transid = generation; 1285789d6a3aSQu Wenruo check.owner_root = key->objectid; 1286789d6a3aSQu Wenruo root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item), 1287789d6a3aSQu Wenruo &check); 128864c043deSLiu Bo if (IS_ERR(root->node)) { 128964c043deSLiu Bo ret = PTR_ERR(root->node); 12908c38938cSJosef Bacik root->node = NULL; 129149d11beaSJosef Bacik goto fail; 12924eb150d6SQu Wenruo } 12934eb150d6SQu Wenruo if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1294cb517eabSMiao Xie ret = -EIO; 129549d11beaSJosef Bacik goto fail; 1296416bc658SJosef Bacik } 129788c602abSQu Wenruo 129888c602abSQu Wenruo /* 129988c602abSQu Wenruo * For real fs, and not log/reloc trees, root owner must 130088c602abSQu Wenruo * match its root node owner 130188c602abSQu Wenruo */ 130288c602abSQu Wenruo if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) && 130388c602abSQu Wenruo root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID && 130488c602abSQu Wenruo root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && 130588c602abSQu Wenruo root->root_key.objectid != btrfs_header_owner(root->node)) { 130688c602abSQu Wenruo btrfs_crit(fs_info, 130788c602abSQu Wenruo "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu", 130888c602abSQu Wenruo root->root_key.objectid, root->node->start, 130988c602abSQu Wenruo btrfs_header_owner(root->node), 131088c602abSQu Wenruo root->root_key.objectid); 131188c602abSQu Wenruo ret = -EUCLEAN; 131288c602abSQu Wenruo goto fail; 131388c602abSQu Wenruo } 13145d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 1315cb517eabSMiao Xie return root; 131649d11beaSJosef Bacik fail: 131700246528SJosef Bacik btrfs_put_root(root); 131849d11beaSJosef Bacik return ERR_PTR(ret); 131949d11beaSJosef Bacik } 132049d11beaSJosef Bacik 132149d11beaSJosef Bacik struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 132249d11beaSJosef Bacik struct btrfs_key *key) 132349d11beaSJosef Bacik { 132449d11beaSJosef Bacik struct btrfs_root *root; 132549d11beaSJosef Bacik struct btrfs_path *path; 132649d11beaSJosef Bacik 132749d11beaSJosef Bacik path = btrfs_alloc_path(); 132849d11beaSJosef Bacik if (!path) 132949d11beaSJosef Bacik return ERR_PTR(-ENOMEM); 133049d11beaSJosef Bacik root = read_tree_root_path(tree_root, path, key); 133149d11beaSJosef Bacik btrfs_free_path(path); 133249d11beaSJosef Bacik 133349d11beaSJosef Bacik return root; 1334cb517eabSMiao Xie } 1335cb517eabSMiao Xie 13362dfb1e43SQu Wenruo /* 13372dfb1e43SQu Wenruo * Initialize subvolume root in-memory structure 13382dfb1e43SQu Wenruo * 13392dfb1e43SQu Wenruo * @anon_dev: anonymous device to attach to the root, if zero, allocate new 13402dfb1e43SQu Wenruo */ 13412dfb1e43SQu Wenruo static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev) 1342cb517eabSMiao Xie { 1343cb517eabSMiao Xie int ret; 1344cb517eabSMiao Xie 1345*0b548539SDavid Sterba btrfs_drew_lock_init(&root->snapshot_lock); 13468257b2dcSMiao Xie 1347aeb935a4SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID && 134837f00a6dSJohannes Thumshirn !btrfs_is_data_reloc_root(root)) { 134992a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &root->state); 1350f39e4571SJosef Bacik btrfs_check_and_init_root_item(&root->root_item); 1351f39e4571SJosef Bacik } 1352f39e4571SJosef Bacik 1353851fd730SQu Wenruo /* 1354851fd730SQu Wenruo * Don't assign anonymous block device to roots that are not exposed to 1355851fd730SQu Wenruo * userspace, the id pool is limited to 1M 1356851fd730SQu Wenruo */ 1357851fd730SQu Wenruo if (is_fstree(root->root_key.objectid) && 1358851fd730SQu Wenruo btrfs_root_refs(&root->root_item) > 0) { 13592dfb1e43SQu Wenruo if (!anon_dev) { 1360cb517eabSMiao Xie ret = get_anon_bdev(&root->anon_dev); 1361cb517eabSMiao Xie if (ret) 1362876d2cf1SLiu Bo goto fail; 13632dfb1e43SQu Wenruo } else { 13642dfb1e43SQu Wenruo root->anon_dev = anon_dev; 13652dfb1e43SQu Wenruo } 1366851fd730SQu Wenruo } 1367f32e48e9SChandan Rajendra 1368f32e48e9SChandan Rajendra mutex_lock(&root->objectid_mutex); 1369453e4873SNikolay Borisov ret = btrfs_init_root_free_objectid(root); 1370f32e48e9SChandan Rajendra if (ret) { 1371f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1372876d2cf1SLiu Bo goto fail; 1373f32e48e9SChandan Rajendra } 1374f32e48e9SChandan Rajendra 13756b8fad57SNikolay Borisov ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID); 1376f32e48e9SChandan Rajendra 1377f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1378f32e48e9SChandan Rajendra 1379cb517eabSMiao Xie return 0; 1380cb517eabSMiao Xie fail: 138184db5ccfSDavid Sterba /* The caller is responsible to call btrfs_free_fs_root */ 1382cb517eabSMiao Xie return ret; 1383cb517eabSMiao Xie } 1384cb517eabSMiao Xie 1385a98db0f3SJosef Bacik static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1386cb517eabSMiao Xie u64 root_id) 1387cb517eabSMiao Xie { 1388cb517eabSMiao Xie struct btrfs_root *root; 1389cb517eabSMiao Xie 1390fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1391fc7cbcd4SDavid Sterba root = radix_tree_lookup(&fs_info->fs_roots_radix, 1392fc7cbcd4SDavid Sterba (unsigned long)root_id); 1393bc44d7c4SJosef Bacik if (root) 139400246528SJosef Bacik root = btrfs_grab_root(root); 1395fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 1396cb517eabSMiao Xie return root; 1397cb517eabSMiao Xie } 1398cb517eabSMiao Xie 139949d11beaSJosef Bacik static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info, 140049d11beaSJosef Bacik u64 objectid) 140149d11beaSJosef Bacik { 1402abed4aaaSJosef Bacik struct btrfs_key key = { 1403abed4aaaSJosef Bacik .objectid = objectid, 1404abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 1405abed4aaaSJosef Bacik .offset = 0, 1406abed4aaaSJosef Bacik }; 1407abed4aaaSJosef Bacik 140849d11beaSJosef Bacik if (objectid == BTRFS_ROOT_TREE_OBJECTID) 140949d11beaSJosef Bacik return btrfs_grab_root(fs_info->tree_root); 141049d11beaSJosef Bacik if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 1411abed4aaaSJosef Bacik return btrfs_grab_root(btrfs_global_root(fs_info, &key)); 141249d11beaSJosef Bacik if (objectid == BTRFS_CHUNK_TREE_OBJECTID) 141349d11beaSJosef Bacik return btrfs_grab_root(fs_info->chunk_root); 141449d11beaSJosef Bacik if (objectid == BTRFS_DEV_TREE_OBJECTID) 141549d11beaSJosef Bacik return btrfs_grab_root(fs_info->dev_root); 141649d11beaSJosef Bacik if (objectid == BTRFS_CSUM_TREE_OBJECTID) 1417abed4aaaSJosef Bacik return btrfs_grab_root(btrfs_global_root(fs_info, &key)); 141849d11beaSJosef Bacik if (objectid == BTRFS_QUOTA_TREE_OBJECTID) 141949d11beaSJosef Bacik return btrfs_grab_root(fs_info->quota_root) ? 142049d11beaSJosef Bacik fs_info->quota_root : ERR_PTR(-ENOENT); 142149d11beaSJosef Bacik if (objectid == BTRFS_UUID_TREE_OBJECTID) 142249d11beaSJosef Bacik return btrfs_grab_root(fs_info->uuid_root) ? 142349d11beaSJosef Bacik fs_info->uuid_root : ERR_PTR(-ENOENT); 142414033b08SQu Wenruo if (objectid == BTRFS_BLOCK_GROUP_TREE_OBJECTID) 142514033b08SQu Wenruo return btrfs_grab_root(fs_info->block_group_root) ? 142614033b08SQu Wenruo fs_info->block_group_root : ERR_PTR(-ENOENT); 1427abed4aaaSJosef Bacik if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) { 1428abed4aaaSJosef Bacik struct btrfs_root *root = btrfs_global_root(fs_info, &key); 1429abed4aaaSJosef Bacik 1430abed4aaaSJosef Bacik return btrfs_grab_root(root) ? root : ERR_PTR(-ENOENT); 1431abed4aaaSJosef Bacik } 143249d11beaSJosef Bacik return NULL; 143349d11beaSJosef Bacik } 143449d11beaSJosef Bacik 1435cb517eabSMiao Xie int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1436cb517eabSMiao Xie struct btrfs_root *root) 1437cb517eabSMiao Xie { 1438cb517eabSMiao Xie int ret; 1439cb517eabSMiao Xie 1440fc7cbcd4SDavid Sterba ret = radix_tree_preload(GFP_NOFS); 1441fc7cbcd4SDavid Sterba if (ret) 1442fc7cbcd4SDavid Sterba return ret; 1443fc7cbcd4SDavid Sterba 1444fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1445fc7cbcd4SDavid Sterba ret = radix_tree_insert(&fs_info->fs_roots_radix, 1446fc7cbcd4SDavid Sterba (unsigned long)root->root_key.objectid, 1447fc7cbcd4SDavid Sterba root); 1448af01d2e5SJosef Bacik if (ret == 0) { 144900246528SJosef Bacik btrfs_grab_root(root); 1450fc7cbcd4SDavid Sterba set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1451af01d2e5SJosef Bacik } 1452fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 1453fc7cbcd4SDavid Sterba radix_tree_preload_end(); 1454cb517eabSMiao Xie 1455cb517eabSMiao Xie return ret; 1456cb517eabSMiao Xie } 1457cb517eabSMiao Xie 1458bd647ce3SJosef Bacik void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info) 1459bd647ce3SJosef Bacik { 1460bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 1461bd647ce3SJosef Bacik struct btrfs_root *root; 1462bd647ce3SJosef Bacik 1463bd647ce3SJosef Bacik while (!list_empty(&fs_info->allocated_roots)) { 1464457f1864SJosef Bacik char buf[BTRFS_ROOT_NAME_BUF_LEN]; 1465457f1864SJosef Bacik 1466bd647ce3SJosef Bacik root = list_first_entry(&fs_info->allocated_roots, 1467bd647ce3SJosef Bacik struct btrfs_root, leak_list); 1468457f1864SJosef Bacik btrfs_err(fs_info, "leaked root %s refcount %d", 146971008734SJosef Bacik btrfs_root_name(&root->root_key, buf), 1470bd647ce3SJosef Bacik refcount_read(&root->refs)); 1471bd647ce3SJosef Bacik while (refcount_read(&root->refs) > 1) 147200246528SJosef Bacik btrfs_put_root(root); 147300246528SJosef Bacik btrfs_put_root(root); 1474bd647ce3SJosef Bacik } 1475bd647ce3SJosef Bacik #endif 1476bd647ce3SJosef Bacik } 1477bd647ce3SJosef Bacik 1478abed4aaaSJosef Bacik static void free_global_roots(struct btrfs_fs_info *fs_info) 1479abed4aaaSJosef Bacik { 1480abed4aaaSJosef Bacik struct btrfs_root *root; 1481abed4aaaSJosef Bacik struct rb_node *node; 1482abed4aaaSJosef Bacik 1483abed4aaaSJosef Bacik while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) { 1484abed4aaaSJosef Bacik root = rb_entry(node, struct btrfs_root, rb_node); 1485abed4aaaSJosef Bacik rb_erase(&root->rb_node, &fs_info->global_root_tree); 1486abed4aaaSJosef Bacik btrfs_put_root(root); 1487abed4aaaSJosef Bacik } 1488abed4aaaSJosef Bacik } 1489abed4aaaSJosef Bacik 14900d4b0463SJosef Bacik void btrfs_free_fs_info(struct btrfs_fs_info *fs_info) 14910d4b0463SJosef Bacik { 1492141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 1493141386e1SJosef Bacik percpu_counter_destroy(&fs_info->delalloc_bytes); 14945deb17e1SJosef Bacik percpu_counter_destroy(&fs_info->ordered_bytes); 1495141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 1496141386e1SJosef Bacik btrfs_free_csum_hash(fs_info); 1497141386e1SJosef Bacik btrfs_free_stripe_hash_table(fs_info); 1498141386e1SJosef Bacik btrfs_free_ref_cache(fs_info); 14990d4b0463SJosef Bacik kfree(fs_info->balance_ctl); 15000d4b0463SJosef Bacik kfree(fs_info->delayed_root); 1501abed4aaaSJosef Bacik free_global_roots(fs_info); 150200246528SJosef Bacik btrfs_put_root(fs_info->tree_root); 150300246528SJosef Bacik btrfs_put_root(fs_info->chunk_root); 150400246528SJosef Bacik btrfs_put_root(fs_info->dev_root); 150500246528SJosef Bacik btrfs_put_root(fs_info->quota_root); 150600246528SJosef Bacik btrfs_put_root(fs_info->uuid_root); 150700246528SJosef Bacik btrfs_put_root(fs_info->fs_root); 1508aeb935a4SQu Wenruo btrfs_put_root(fs_info->data_reloc_root); 15099c54e80dSJosef Bacik btrfs_put_root(fs_info->block_group_root); 1510bd647ce3SJosef Bacik btrfs_check_leaked_roots(fs_info); 15113fd63727SJosef Bacik btrfs_extent_buffer_leak_debug_check(fs_info); 15120d4b0463SJosef Bacik kfree(fs_info->super_copy); 15130d4b0463SJosef Bacik kfree(fs_info->super_for_commit); 15148481dd80SQu Wenruo kfree(fs_info->subpage_info); 15150d4b0463SJosef Bacik kvfree(fs_info); 15160d4b0463SJosef Bacik } 15170d4b0463SJosef Bacik 15180d4b0463SJosef Bacik 15192dfb1e43SQu Wenruo /* 15202dfb1e43SQu Wenruo * Get an in-memory reference of a root structure. 15212dfb1e43SQu Wenruo * 15222dfb1e43SQu Wenruo * For essential trees like root/extent tree, we grab it from fs_info directly. 15232dfb1e43SQu Wenruo * For subvolume trees, we check the cached filesystem roots first. If not 15242dfb1e43SQu Wenruo * found, then read it from disk and add it to cached fs roots. 15252dfb1e43SQu Wenruo * 15262dfb1e43SQu Wenruo * Caller should release the root by calling btrfs_put_root() after the usage. 15272dfb1e43SQu Wenruo * 15282dfb1e43SQu Wenruo * NOTE: Reloc and log trees can't be read by this function as they share the 15292dfb1e43SQu Wenruo * same root objectid. 15302dfb1e43SQu Wenruo * 15312dfb1e43SQu Wenruo * @objectid: root id 15322dfb1e43SQu Wenruo * @anon_dev: preallocated anonymous block device number for new roots, 15332dfb1e43SQu Wenruo * pass 0 for new allocation. 15342dfb1e43SQu Wenruo * @check_ref: whether to check root item references, If true, return -ENOENT 15352dfb1e43SQu Wenruo * for orphan roots 15362dfb1e43SQu Wenruo */ 15372dfb1e43SQu Wenruo static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info, 15382dfb1e43SQu Wenruo u64 objectid, dev_t anon_dev, 15392dfb1e43SQu Wenruo bool check_ref) 15405eda7b5eSChris Mason { 15415eda7b5eSChris Mason struct btrfs_root *root; 1542381cf658SDavid Sterba struct btrfs_path *path; 15431d4c08e0SDavid Sterba struct btrfs_key key; 15445eda7b5eSChris Mason int ret; 15455eda7b5eSChris Mason 154649d11beaSJosef Bacik root = btrfs_get_global_root(fs_info, objectid); 154749d11beaSJosef Bacik if (root) 154849d11beaSJosef Bacik return root; 15494df27c4dSYan, Zheng again: 155056e9357aSDavid Sterba root = btrfs_lookup_fs_root(fs_info, objectid); 155148475471SStefan Behrens if (root) { 15522dfb1e43SQu Wenruo /* Shouldn't get preallocated anon_dev for cached roots */ 15532dfb1e43SQu Wenruo ASSERT(!anon_dev); 1554bc44d7c4SJosef Bacik if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 155500246528SJosef Bacik btrfs_put_root(root); 155648475471SStefan Behrens return ERR_PTR(-ENOENT); 1557bc44d7c4SJosef Bacik } 15585eda7b5eSChris Mason return root; 155948475471SStefan Behrens } 15605eda7b5eSChris Mason 156156e9357aSDavid Sterba key.objectid = objectid; 156256e9357aSDavid Sterba key.type = BTRFS_ROOT_ITEM_KEY; 156356e9357aSDavid Sterba key.offset = (u64)-1; 156456e9357aSDavid Sterba root = btrfs_read_tree_root(fs_info->tree_root, &key); 15655eda7b5eSChris Mason if (IS_ERR(root)) 15665eda7b5eSChris Mason return root; 15673394e160SChris Mason 1568c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1569d68fc57bSYan, Zheng ret = -ENOENT; 1570d68fc57bSYan, Zheng goto fail; 1571d68fc57bSYan, Zheng } 1572d68fc57bSYan, Zheng 15732dfb1e43SQu Wenruo ret = btrfs_init_fs_root(root, anon_dev); 1574cb517eabSMiao Xie if (ret) 1575cb517eabSMiao Xie goto fail; 1576cb517eabSMiao Xie 1577381cf658SDavid Sterba path = btrfs_alloc_path(); 1578381cf658SDavid Sterba if (!path) { 1579381cf658SDavid Sterba ret = -ENOMEM; 1580381cf658SDavid Sterba goto fail; 1581381cf658SDavid Sterba } 15821d4c08e0SDavid Sterba key.objectid = BTRFS_ORPHAN_OBJECTID; 15831d4c08e0SDavid Sterba key.type = BTRFS_ORPHAN_ITEM_KEY; 158456e9357aSDavid Sterba key.offset = objectid; 15851d4c08e0SDavid Sterba 15861d4c08e0SDavid Sterba ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1587381cf658SDavid Sterba btrfs_free_path(path); 1588d68fc57bSYan, Zheng if (ret < 0) 1589d68fc57bSYan, Zheng goto fail; 1590d68fc57bSYan, Zheng if (ret == 0) 159127cdeb70SMiao Xie set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1592d68fc57bSYan, Zheng 1593cb517eabSMiao Xie ret = btrfs_insert_fs_root(fs_info, root); 15940f7d52f4SChris Mason if (ret) { 1595168a2f77SJia-Ju Bai if (ret == -EEXIST) { 159600246528SJosef Bacik btrfs_put_root(root); 15974df27c4dSYan, Zheng goto again; 1598168a2f77SJia-Ju Bai } 15994df27c4dSYan, Zheng goto fail; 16004df27c4dSYan, Zheng } 1601edbd8d4eSChris Mason return root; 16024df27c4dSYan, Zheng fail: 160333fab972SFilipe Manana /* 160433fab972SFilipe Manana * If our caller provided us an anonymous device, then it's his 1605143823cfSDavid Sterba * responsibility to free it in case we fail. So we have to set our 160633fab972SFilipe Manana * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root() 160733fab972SFilipe Manana * and once again by our caller. 160833fab972SFilipe Manana */ 160933fab972SFilipe Manana if (anon_dev) 161033fab972SFilipe Manana root->anon_dev = 0; 16118c38938cSJosef Bacik btrfs_put_root(root); 16124df27c4dSYan, Zheng return ERR_PTR(ret); 1613edbd8d4eSChris Mason } 1614edbd8d4eSChris Mason 16152dfb1e43SQu Wenruo /* 16162dfb1e43SQu Wenruo * Get in-memory reference of a root structure 16172dfb1e43SQu Wenruo * 16182dfb1e43SQu Wenruo * @objectid: tree objectid 16192dfb1e43SQu Wenruo * @check_ref: if set, verify that the tree exists and the item has at least 16202dfb1e43SQu Wenruo * one reference 16212dfb1e43SQu Wenruo */ 16222dfb1e43SQu Wenruo struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 16232dfb1e43SQu Wenruo u64 objectid, bool check_ref) 16242dfb1e43SQu Wenruo { 16252dfb1e43SQu Wenruo return btrfs_get_root_ref(fs_info, objectid, 0, check_ref); 16262dfb1e43SQu Wenruo } 16272dfb1e43SQu Wenruo 16282dfb1e43SQu Wenruo /* 16292dfb1e43SQu Wenruo * Get in-memory reference of a root structure, created as new, optionally pass 16302dfb1e43SQu Wenruo * the anonymous block device id 16312dfb1e43SQu Wenruo * 16322dfb1e43SQu Wenruo * @objectid: tree objectid 16332dfb1e43SQu Wenruo * @anon_dev: if zero, allocate a new anonymous block device or use the 16342dfb1e43SQu Wenruo * parameter value 16352dfb1e43SQu Wenruo */ 16362dfb1e43SQu Wenruo struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info, 16372dfb1e43SQu Wenruo u64 objectid, dev_t anon_dev) 16382dfb1e43SQu Wenruo { 16392dfb1e43SQu Wenruo return btrfs_get_root_ref(fs_info, objectid, anon_dev, true); 16402dfb1e43SQu Wenruo } 16412dfb1e43SQu Wenruo 16428b712842SChris Mason /* 164349d11beaSJosef Bacik * btrfs_get_fs_root_commit_root - return a root for the given objectid 164449d11beaSJosef Bacik * @fs_info: the fs_info 164549d11beaSJosef Bacik * @objectid: the objectid we need to lookup 164649d11beaSJosef Bacik * 164749d11beaSJosef Bacik * This is exclusively used for backref walking, and exists specifically because 164849d11beaSJosef Bacik * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref 164949d11beaSJosef Bacik * creation time, which means we may have to read the tree_root in order to look 165049d11beaSJosef Bacik * up a fs root that is not in memory. If the root is not in memory we will 165149d11beaSJosef Bacik * read the tree root commit root and look up the fs root from there. This is a 165249d11beaSJosef Bacik * temporary root, it will not be inserted into the radix tree as it doesn't 165349d11beaSJosef Bacik * have the most uptodate information, it'll simply be discarded once the 165449d11beaSJosef Bacik * backref code is finished using the root. 165549d11beaSJosef Bacik */ 165649d11beaSJosef Bacik struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info, 165749d11beaSJosef Bacik struct btrfs_path *path, 165849d11beaSJosef Bacik u64 objectid) 165949d11beaSJosef Bacik { 166049d11beaSJosef Bacik struct btrfs_root *root; 166149d11beaSJosef Bacik struct btrfs_key key; 166249d11beaSJosef Bacik 166349d11beaSJosef Bacik ASSERT(path->search_commit_root && path->skip_locking); 166449d11beaSJosef Bacik 166549d11beaSJosef Bacik /* 166649d11beaSJosef Bacik * This can return -ENOENT if we ask for a root that doesn't exist, but 166749d11beaSJosef Bacik * since this is called via the backref walking code we won't be looking 166849d11beaSJosef Bacik * up a root that doesn't exist, unless there's corruption. So if root 166949d11beaSJosef Bacik * != NULL just return it. 167049d11beaSJosef Bacik */ 167149d11beaSJosef Bacik root = btrfs_get_global_root(fs_info, objectid); 167249d11beaSJosef Bacik if (root) 167349d11beaSJosef Bacik return root; 167449d11beaSJosef Bacik 167549d11beaSJosef Bacik root = btrfs_lookup_fs_root(fs_info, objectid); 167649d11beaSJosef Bacik if (root) 167749d11beaSJosef Bacik return root; 167849d11beaSJosef Bacik 167949d11beaSJosef Bacik key.objectid = objectid; 168049d11beaSJosef Bacik key.type = BTRFS_ROOT_ITEM_KEY; 168149d11beaSJosef Bacik key.offset = (u64)-1; 168249d11beaSJosef Bacik root = read_tree_root_path(fs_info->tree_root, path, &key); 168349d11beaSJosef Bacik btrfs_release_path(path); 168449d11beaSJosef Bacik 168549d11beaSJosef Bacik return root; 168649d11beaSJosef Bacik } 168749d11beaSJosef Bacik 1688a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1689a74a4b97SChris Mason { 16900d031dc4SYu Zhe struct btrfs_fs_info *fs_info = arg; 1691d0278245SMiao Xie int again; 1692a74a4b97SChris Mason 1693d6fd0ae2SOmar Sandoval while (1) { 1694d0278245SMiao Xie again = 0; 16959d1a2a3aSDavid Sterba 1696fd340d0fSJosef Bacik set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1697fd340d0fSJosef Bacik 1698d0278245SMiao Xie /* Make the cleaner go to sleep early. */ 16992ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) 1700d0278245SMiao Xie goto sleep; 1701d0278245SMiao Xie 170290c711abSZygo Blaxell /* 170390c711abSZygo Blaxell * Do not do anything if we might cause open_ctree() to block 170490c711abSZygo Blaxell * before we have finished mounting the filesystem. 170590c711abSZygo Blaxell */ 17060b246afaSJeff Mahoney if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 170790c711abSZygo Blaxell goto sleep; 170890c711abSZygo Blaxell 17090b246afaSJeff Mahoney if (!mutex_trylock(&fs_info->cleaner_mutex)) 1710d0278245SMiao Xie goto sleep; 1711d0278245SMiao Xie 1712dc7f370cSMiao Xie /* 1713dc7f370cSMiao Xie * Avoid the problem that we change the status of the fs 1714dc7f370cSMiao Xie * during the above check and trylock. 1715dc7f370cSMiao Xie */ 17162ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) { 17170b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1718dc7f370cSMiao Xie goto sleep; 1719dc7f370cSMiao Xie } 1720dc7f370cSMiao Xie 1721b7625f46SQu Wenruo if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags)) 1722b7625f46SQu Wenruo btrfs_sysfs_feature_update(fs_info); 1723b7625f46SQu Wenruo 17242ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 1725c2d6cb16SFilipe Manana 172633c44184SJosef Bacik again = btrfs_clean_one_deleted_snapshot(fs_info); 17270b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1728a74a4b97SChris Mason 1729d0278245SMiao Xie /* 173005323cd1SMiao Xie * The defragger has dealt with the R/O remount and umount, 173105323cd1SMiao Xie * needn't do anything special here. 1732d0278245SMiao Xie */ 17330b246afaSJeff Mahoney btrfs_run_defrag_inodes(fs_info); 173467c5e7d4SFilipe Manana 173567c5e7d4SFilipe Manana /* 1736f3372065SJohannes Thumshirn * Acquires fs_info->reclaim_bgs_lock to avoid racing 173767c5e7d4SFilipe Manana * with relocation (btrfs_relocate_chunk) and relocation 173867c5e7d4SFilipe Manana * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 1739f3372065SJohannes Thumshirn * after acquiring fs_info->reclaim_bgs_lock. So we 174067c5e7d4SFilipe Manana * can't hold, nor need to, fs_info->cleaner_mutex when deleting 174167c5e7d4SFilipe Manana * unused block groups. 174267c5e7d4SFilipe Manana */ 17430b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 174418bb8bbfSJohannes Thumshirn 174518bb8bbfSJohannes Thumshirn /* 174618bb8bbfSJohannes Thumshirn * Reclaim block groups in the reclaim_bgs list after we deleted 174718bb8bbfSJohannes Thumshirn * all unused block_groups. This possibly gives us some more free 174818bb8bbfSJohannes Thumshirn * space. 174918bb8bbfSJohannes Thumshirn */ 175018bb8bbfSJohannes Thumshirn btrfs_reclaim_bgs(fs_info); 1751d0278245SMiao Xie sleep: 1752a0a1db70SFilipe Manana clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1753d6fd0ae2SOmar Sandoval if (kthread_should_park()) 1754d6fd0ae2SOmar Sandoval kthread_parkme(); 1755d6fd0ae2SOmar Sandoval if (kthread_should_stop()) 1756d6fd0ae2SOmar Sandoval return 0; 1757838fe188SJiri Kosina if (!again) { 1758a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 1759a74a4b97SChris Mason schedule(); 1760a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1761a74a4b97SChris Mason } 1762da288d28SFilipe Manana } 1763a74a4b97SChris Mason } 1764a74a4b97SChris Mason 1765a74a4b97SChris Mason static int transaction_kthread(void *arg) 1766a74a4b97SChris Mason { 1767a74a4b97SChris Mason struct btrfs_root *root = arg; 17680b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1769a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1770a74a4b97SChris Mason struct btrfs_transaction *cur; 17718929ecfaSYan, Zheng u64 transid; 1772643900beSNikolay Borisov time64_t delta; 1773a74a4b97SChris Mason unsigned long delay; 1774914b2007SJan Kara bool cannot_commit; 1775a74a4b97SChris Mason 1776a74a4b97SChris Mason do { 1777914b2007SJan Kara cannot_commit = false; 1778ba1bc00fSNikolay Borisov delay = msecs_to_jiffies(fs_info->commit_interval * 1000); 17790b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 1780a74a4b97SChris Mason 17810b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 17820b246afaSJeff Mahoney cur = fs_info->running_transaction; 1783a74a4b97SChris Mason if (!cur) { 17840b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1785a74a4b97SChris Mason goto sleep; 1786a74a4b97SChris Mason } 178731153d81SYan Zheng 1788643900beSNikolay Borisov delta = ktime_get_seconds() - cur->start_time; 1789fdfbf020SJosef Bacik if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) && 1790fdfbf020SJosef Bacik cur->state < TRANS_STATE_COMMIT_START && 1791643900beSNikolay Borisov delta < fs_info->commit_interval) { 17920b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1793fb8a7e94SNikolay Borisov delay -= msecs_to_jiffies((delta - 1) * 1000); 1794fb8a7e94SNikolay Borisov delay = min(delay, 1795fb8a7e94SNikolay Borisov msecs_to_jiffies(fs_info->commit_interval * 1000)); 1796a74a4b97SChris Mason goto sleep; 1797a74a4b97SChris Mason } 17988929ecfaSYan, Zheng transid = cur->transid; 17990b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 180056bec294SChris Mason 180179787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 1802354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 1803914b2007SJan Kara if (IS_ERR(trans)) { 1804354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 1805914b2007SJan Kara cannot_commit = true; 180679787eaaSJeff Mahoney goto sleep; 1807914b2007SJan Kara } 18088929ecfaSYan, Zheng if (transid == trans->transid) { 18093a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 18108929ecfaSYan, Zheng } else { 18113a45bb20SJeff Mahoney btrfs_end_transaction(trans); 18128929ecfaSYan, Zheng } 1813a74a4b97SChris Mason sleep: 18140b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 18150b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 1816a74a4b97SChris Mason 181784961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) 18182ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 18198929ecfaSYan, Zheng if (!kthread_should_stop() && 18200b246afaSJeff Mahoney (!btrfs_transaction_blocked(fs_info) || 1821914b2007SJan Kara cannot_commit)) 1822bc5511d0SNikolay Borisov schedule_timeout_interruptible(delay); 1823a74a4b97SChris Mason } while (!kthread_should_stop()); 1824a74a4b97SChris Mason return 0; 1825a74a4b97SChris Mason } 1826a74a4b97SChris Mason 1827af31f5e5SChris Mason /* 182801f0f9daSNikolay Borisov * This will find the highest generation in the array of root backups. The 182901f0f9daSNikolay Borisov * index of the highest array is returned, or -EINVAL if we can't find 183001f0f9daSNikolay Borisov * anything. 1831af31f5e5SChris Mason * 1832af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1833af31f5e5SChris Mason * generation of the latest root in the array with the generation 1834af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1835af31f5e5SChris Mason */ 183601f0f9daSNikolay Borisov static int find_newest_super_backup(struct btrfs_fs_info *info) 1837af31f5e5SChris Mason { 183801f0f9daSNikolay Borisov const u64 newest_gen = btrfs_super_generation(info->super_copy); 1839af31f5e5SChris Mason u64 cur; 1840af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1841af31f5e5SChris Mason int i; 1842af31f5e5SChris Mason 1843af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1844af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1845af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1846af31f5e5SChris Mason if (cur == newest_gen) 184701f0f9daSNikolay Borisov return i; 1848af31f5e5SChris Mason } 1849af31f5e5SChris Mason 185001f0f9daSNikolay Borisov return -EINVAL; 1851af31f5e5SChris Mason } 1852af31f5e5SChris Mason 1853af31f5e5SChris Mason /* 1854af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1855af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1856af31f5e5SChris Mason * done 1857af31f5e5SChris Mason */ 1858af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1859af31f5e5SChris Mason { 18606ef108ddSNikolay Borisov const int next_backup = info->backup_root_index; 1861af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1862af31f5e5SChris Mason 1863af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1864af31f5e5SChris Mason 1865af31f5e5SChris Mason /* 1866af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1867af31f5e5SChris Mason * regardless of which ones we use today 1868af31f5e5SChris Mason */ 1869af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1870af31f5e5SChris Mason 1871af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1872af31f5e5SChris Mason 1873af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1874af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1875af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1876af31f5e5SChris Mason 1877af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1878af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1879af31f5e5SChris Mason 1880af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1881af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1882af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 1883af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 1884af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 1885af31f5e5SChris Mason 18861c56ab99SQu Wenruo if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) { 18879c54e80dSJosef Bacik struct btrfs_root *extent_root = btrfs_extent_root(info, 0); 1888f7238e50SJosef Bacik struct btrfs_root *csum_root = btrfs_csum_root(info, 0); 18899c54e80dSJosef Bacik 18909c54e80dSJosef Bacik btrfs_set_backup_extent_root(root_backup, 18919c54e80dSJosef Bacik extent_root->node->start); 1892af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 189329cbcf40SJosef Bacik btrfs_header_generation(extent_root->node)); 1894af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 189529cbcf40SJosef Bacik btrfs_header_level(extent_root->node)); 1896af31f5e5SChris Mason 1897f7238e50SJosef Bacik btrfs_set_backup_csum_root(root_backup, csum_root->node->start); 1898f7238e50SJosef Bacik btrfs_set_backup_csum_root_gen(root_backup, 1899f7238e50SJosef Bacik btrfs_header_generation(csum_root->node)); 1900f7238e50SJosef Bacik btrfs_set_backup_csum_root_level(root_backup, 1901f7238e50SJosef Bacik btrfs_header_level(csum_root->node)); 19029c54e80dSJosef Bacik } 1903af31f5e5SChris Mason 19047c7e82a7SChris Mason /* 19057c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 19067c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 19077c7e82a7SChris Mason */ 19087c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 19097c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 19107c7e82a7SChris Mason info->fs_root->node->start); 1911af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 1912af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 1913af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 1914af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 19157c7e82a7SChris Mason } 1916af31f5e5SChris Mason 1917af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 1918af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 1919af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 1920af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 1921af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 1922af31f5e5SChris Mason 1923af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 1924af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 1925af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 1926af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 1927af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 1928af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 1929af31f5e5SChris Mason 1930af31f5e5SChris Mason /* 1931af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 1932af31f5e5SChris Mason * for the next commit 1933af31f5e5SChris Mason */ 1934af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 1935af31f5e5SChris Mason &info->super_for_commit->super_roots, 1936af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 1937af31f5e5SChris Mason } 1938af31f5e5SChris Mason 1939af31f5e5SChris Mason /* 1940bd2336b2SNikolay Borisov * read_backup_root - Reads a backup root based on the passed priority. Prio 0 1941bd2336b2SNikolay Borisov * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots 1942bd2336b2SNikolay Borisov * 1943bd2336b2SNikolay Borisov * fs_info - filesystem whose backup roots need to be read 1944bd2336b2SNikolay Borisov * priority - priority of backup root required 1945bd2336b2SNikolay Borisov * 1946bd2336b2SNikolay Borisov * Returns backup root index on success and -EINVAL otherwise. 1947bd2336b2SNikolay Borisov */ 1948bd2336b2SNikolay Borisov static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority) 1949bd2336b2SNikolay Borisov { 1950bd2336b2SNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 1951bd2336b2SNikolay Borisov struct btrfs_super_block *super = fs_info->super_copy; 1952bd2336b2SNikolay Borisov struct btrfs_root_backup *root_backup; 1953bd2336b2SNikolay Borisov 1954bd2336b2SNikolay Borisov if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) { 1955bd2336b2SNikolay Borisov if (priority == 0) 1956bd2336b2SNikolay Borisov return backup_index; 1957bd2336b2SNikolay Borisov 1958bd2336b2SNikolay Borisov backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority; 1959bd2336b2SNikolay Borisov backup_index %= BTRFS_NUM_BACKUP_ROOTS; 1960bd2336b2SNikolay Borisov } else { 1961bd2336b2SNikolay Borisov return -EINVAL; 1962bd2336b2SNikolay Borisov } 1963bd2336b2SNikolay Borisov 1964bd2336b2SNikolay Borisov root_backup = super->super_roots + backup_index; 1965bd2336b2SNikolay Borisov 1966bd2336b2SNikolay Borisov btrfs_set_super_generation(super, 1967bd2336b2SNikolay Borisov btrfs_backup_tree_root_gen(root_backup)); 1968bd2336b2SNikolay Borisov btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 1969bd2336b2SNikolay Borisov btrfs_set_super_root_level(super, 1970bd2336b2SNikolay Borisov btrfs_backup_tree_root_level(root_backup)); 1971bd2336b2SNikolay Borisov btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 1972bd2336b2SNikolay Borisov 1973bd2336b2SNikolay Borisov /* 1974bd2336b2SNikolay Borisov * Fixme: the total bytes and num_devices need to match or we should 1975bd2336b2SNikolay Borisov * need a fsck 1976bd2336b2SNikolay Borisov */ 1977bd2336b2SNikolay Borisov btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 1978bd2336b2SNikolay Borisov btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 1979bd2336b2SNikolay Borisov 1980bd2336b2SNikolay Borisov return backup_index; 1981bd2336b2SNikolay Borisov } 1982bd2336b2SNikolay Borisov 19837abadb64SLiu Bo /* helper to cleanup workers */ 19847abadb64SLiu Bo static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 19857abadb64SLiu Bo { 1986dc6e3209SQu Wenruo btrfs_destroy_workqueue(fs_info->fixup_workers); 1987afe3d242SQu Wenruo btrfs_destroy_workqueue(fs_info->delalloc_workers); 1988a31b4a43SChristoph Hellwig btrfs_destroy_workqueue(fs_info->hipri_workers); 19895cdc7ad3SQu Wenruo btrfs_destroy_workqueue(fs_info->workers); 1990d7b9416fSChristoph Hellwig if (fs_info->endio_workers) 1991d7b9416fSChristoph Hellwig destroy_workqueue(fs_info->endio_workers); 1992385de0efSChristoph Hellwig if (fs_info->rmw_workers) 1993385de0efSChristoph Hellwig destroy_workqueue(fs_info->rmw_workers); 1994fed8a72dSChristoph Hellwig if (fs_info->compressed_write_workers) 1995fed8a72dSChristoph Hellwig destroy_workqueue(fs_info->compressed_write_workers); 1996fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_write_workers); 1997fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 19985b3bc44eSQu Wenruo btrfs_destroy_workqueue(fs_info->delayed_workers); 1999e66f0bb1SQu Wenruo btrfs_destroy_workqueue(fs_info->caching_workers); 2000a44903abSQu Wenruo btrfs_destroy_workqueue(fs_info->flush_workers); 2001fc97fab0SQu Wenruo btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2002b0643e59SDennis Zhou if (fs_info->discard_ctl.discard_workers) 2003b0643e59SDennis Zhou destroy_workqueue(fs_info->discard_ctl.discard_workers); 2004a9b9477dSFilipe Manana /* 2005a9b9477dSFilipe Manana * Now that all other work queues are destroyed, we can safely destroy 2006a9b9477dSFilipe Manana * the queues used for metadata I/O, since tasks from those other work 2007a9b9477dSFilipe Manana * queues can do metadata I/O operations. 2008a9b9477dSFilipe Manana */ 2009d7b9416fSChristoph Hellwig if (fs_info->endio_meta_workers) 2010d7b9416fSChristoph Hellwig destroy_workqueue(fs_info->endio_meta_workers); 20117abadb64SLiu Bo } 20127abadb64SLiu Bo 20132e9f5954SRashika static void free_root_extent_buffers(struct btrfs_root *root) 20142e9f5954SRashika { 20152e9f5954SRashika if (root) { 20162e9f5954SRashika free_extent_buffer(root->node); 20172e9f5954SRashika free_extent_buffer(root->commit_root); 20182e9f5954SRashika root->node = NULL; 20192e9f5954SRashika root->commit_root = NULL; 20202e9f5954SRashika } 20212e9f5954SRashika } 20222e9f5954SRashika 2023abed4aaaSJosef Bacik static void free_global_root_pointers(struct btrfs_fs_info *fs_info) 2024abed4aaaSJosef Bacik { 2025abed4aaaSJosef Bacik struct btrfs_root *root, *tmp; 2026abed4aaaSJosef Bacik 2027abed4aaaSJosef Bacik rbtree_postorder_for_each_entry_safe(root, tmp, 2028abed4aaaSJosef Bacik &fs_info->global_root_tree, 2029abed4aaaSJosef Bacik rb_node) 2030abed4aaaSJosef Bacik free_root_extent_buffers(root); 2031abed4aaaSJosef Bacik } 2032abed4aaaSJosef Bacik 2033af31f5e5SChris Mason /* helper to cleanup tree roots */ 20344273eaffSAnand Jain static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root) 2035af31f5e5SChris Mason { 20362e9f5954SRashika free_root_extent_buffers(info->tree_root); 2037655b09feSJosef Bacik 2038abed4aaaSJosef Bacik free_global_root_pointers(info); 20392e9f5954SRashika free_root_extent_buffers(info->dev_root); 20402e9f5954SRashika free_root_extent_buffers(info->quota_root); 20412e9f5954SRashika free_root_extent_buffers(info->uuid_root); 20428c38938cSJosef Bacik free_root_extent_buffers(info->fs_root); 2043aeb935a4SQu Wenruo free_root_extent_buffers(info->data_reloc_root); 20449c54e80dSJosef Bacik free_root_extent_buffers(info->block_group_root); 20454273eaffSAnand Jain if (free_chunk_root) 20462e9f5954SRashika free_root_extent_buffers(info->chunk_root); 2047af31f5e5SChris Mason } 2048af31f5e5SChris Mason 20498c38938cSJosef Bacik void btrfs_put_root(struct btrfs_root *root) 20508c38938cSJosef Bacik { 20518c38938cSJosef Bacik if (!root) 20528c38938cSJosef Bacik return; 20538c38938cSJosef Bacik 20548c38938cSJosef Bacik if (refcount_dec_and_test(&root->refs)) { 20558c38938cSJosef Bacik WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 20561dae7e0eSQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)); 20578c38938cSJosef Bacik if (root->anon_dev) 20588c38938cSJosef Bacik free_anon_bdev(root->anon_dev); 2059923eb523SJohannes Thumshirn free_root_extent_buffers(root); 20608c38938cSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 2061fc7cbcd4SDavid Sterba spin_lock(&root->fs_info->fs_roots_radix_lock); 20628c38938cSJosef Bacik list_del_init(&root->leak_list); 2063fc7cbcd4SDavid Sterba spin_unlock(&root->fs_info->fs_roots_radix_lock); 20648c38938cSJosef Bacik #endif 20658c38938cSJosef Bacik kfree(root); 20668c38938cSJosef Bacik } 20678c38938cSJosef Bacik } 20688c38938cSJosef Bacik 2069faa2dbf0SJosef Bacik void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2070171f6537SJosef Bacik { 2071fc7cbcd4SDavid Sterba int ret; 2072fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 2073fc7cbcd4SDavid Sterba int i; 2074171f6537SJosef Bacik 2075171f6537SJosef Bacik while (!list_empty(&fs_info->dead_roots)) { 2076fc7cbcd4SDavid Sterba gang[0] = list_entry(fs_info->dead_roots.next, 2077171f6537SJosef Bacik struct btrfs_root, root_list); 2078fc7cbcd4SDavid Sterba list_del(&gang[0]->root_list); 2079171f6537SJosef Bacik 2080fc7cbcd4SDavid Sterba if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) 2081fc7cbcd4SDavid Sterba btrfs_drop_and_free_fs_root(fs_info, gang[0]); 2082fc7cbcd4SDavid Sterba btrfs_put_root(gang[0]); 2083171f6537SJosef Bacik } 2084171f6537SJosef Bacik 2085fc7cbcd4SDavid Sterba while (1) { 2086fc7cbcd4SDavid Sterba ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2087fc7cbcd4SDavid Sterba (void **)gang, 0, 2088fc7cbcd4SDavid Sterba ARRAY_SIZE(gang)); 2089fc7cbcd4SDavid Sterba if (!ret) 2090fc7cbcd4SDavid Sterba break; 2091fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) 2092fc7cbcd4SDavid Sterba btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2093171f6537SJosef Bacik } 2094171f6537SJosef Bacik } 2095af31f5e5SChris Mason 2096638aa7edSEric Sandeen static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2097638aa7edSEric Sandeen { 2098638aa7edSEric Sandeen mutex_init(&fs_info->scrub_lock); 2099638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_running, 0); 2100638aa7edSEric Sandeen atomic_set(&fs_info->scrub_pause_req, 0); 2101638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_paused, 0); 2102638aa7edSEric Sandeen atomic_set(&fs_info->scrub_cancel_req, 0); 2103638aa7edSEric Sandeen init_waitqueue_head(&fs_info->scrub_pause_wait); 2104ff09c4caSAnand Jain refcount_set(&fs_info->scrub_workers_refcnt, 0); 2105638aa7edSEric Sandeen } 2106638aa7edSEric Sandeen 2107779a65a4SEric Sandeen static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2108779a65a4SEric Sandeen { 2109779a65a4SEric Sandeen spin_lock_init(&fs_info->balance_lock); 2110779a65a4SEric Sandeen mutex_init(&fs_info->balance_mutex); 2111779a65a4SEric Sandeen atomic_set(&fs_info->balance_pause_req, 0); 2112779a65a4SEric Sandeen atomic_set(&fs_info->balance_cancel_req, 0); 2113779a65a4SEric Sandeen fs_info->balance_ctl = NULL; 2114779a65a4SEric Sandeen init_waitqueue_head(&fs_info->balance_wait_q); 2115907d2710SDavid Sterba atomic_set(&fs_info->reloc_cancel_req, 0); 2116779a65a4SEric Sandeen } 2117779a65a4SEric Sandeen 2118dcb2137cSChristoph Hellwig static int btrfs_init_btree_inode(struct super_block *sb) 2119f37938e0SEric Sandeen { 2120dcb2137cSChristoph Hellwig struct btrfs_fs_info *fs_info = btrfs_sb(sb); 2121e256927bSJosef Bacik unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID, 2122e256927bSJosef Bacik fs_info->tree_root); 2123dcb2137cSChristoph Hellwig struct inode *inode; 2124dcb2137cSChristoph Hellwig 2125dcb2137cSChristoph Hellwig inode = new_inode(sb); 2126dcb2137cSChristoph Hellwig if (!inode) 2127dcb2137cSChristoph Hellwig return -ENOMEM; 21282ff7e61eSJeff Mahoney 21292ff7e61eSJeff Mahoney inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 21302ff7e61eSJeff Mahoney set_nlink(inode, 1); 2131f37938e0SEric Sandeen /* 2132f37938e0SEric Sandeen * we set the i_size on the btree inode to the max possible int. 2133f37938e0SEric Sandeen * the real end of the address space is determined by all of 2134f37938e0SEric Sandeen * the devices in the system 2135f37938e0SEric Sandeen */ 21362ff7e61eSJeff Mahoney inode->i_size = OFFSET_MAX; 21372ff7e61eSJeff Mahoney inode->i_mapping->a_ops = &btree_aops; 2138dcb2137cSChristoph Hellwig mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS); 2139f37938e0SEric Sandeen 21402ff7e61eSJeff Mahoney RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 214143eb5f29SQu Wenruo extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree, 214235da5a7eSDavid Sterba IO_TREE_BTREE_INODE_IO); 21432ff7e61eSJeff Mahoney extent_map_tree_init(&BTRFS_I(inode)->extent_tree); 2144f37938e0SEric Sandeen 21455c8fd99fSJosef Bacik BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root); 2146adac5584SFilipe Manana BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID; 2147adac5584SFilipe Manana BTRFS_I(inode)->location.type = 0; 2148adac5584SFilipe Manana BTRFS_I(inode)->location.offset = 0; 21492ff7e61eSJeff Mahoney set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); 2150e256927bSJosef Bacik __insert_inode_hash(inode, hash); 2151dcb2137cSChristoph Hellwig fs_info->btree_inode = inode; 2152dcb2137cSChristoph Hellwig 2153dcb2137cSChristoph Hellwig return 0; 2154f37938e0SEric Sandeen } 2155f37938e0SEric Sandeen 2156ad618368SEric Sandeen static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2157ad618368SEric Sandeen { 2158ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2159129827e3SDavid Sterba init_rwsem(&fs_info->dev_replace.rwsem); 21607f8d236aSDavid Sterba init_waitqueue_head(&fs_info->dev_replace.replace_wait); 2161ad618368SEric Sandeen } 2162ad618368SEric Sandeen 2163f9e92e40SEric Sandeen static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2164f9e92e40SEric Sandeen { 2165f9e92e40SEric Sandeen spin_lock_init(&fs_info->qgroup_lock); 2166f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_ioctl_lock); 2167f9e92e40SEric Sandeen fs_info->qgroup_tree = RB_ROOT; 2168f9e92e40SEric Sandeen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2169f9e92e40SEric Sandeen fs_info->qgroup_seq = 1; 2170f9e92e40SEric Sandeen fs_info->qgroup_ulist = NULL; 2171d2c609b8SJeff Mahoney fs_info->qgroup_rescan_running = false; 2172011b46c3SQu Wenruo fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL; 2173f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_rescan_lock); 2174f9e92e40SEric Sandeen } 2175f9e92e40SEric Sandeen 2176d21deec5SSu Yue static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info) 21772a458198SEric Sandeen { 2178f7b885beSAnand Jain u32 max_active = fs_info->thread_pool_size; 21796f011058SDavid Sterba unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 21802a458198SEric Sandeen 21812a458198SEric Sandeen fs_info->workers = 2182a31b4a43SChristoph Hellwig btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16); 2183a31b4a43SChristoph Hellwig fs_info->hipri_workers = 2184a31b4a43SChristoph Hellwig btrfs_alloc_workqueue(fs_info, "worker-high", 2185cb001095SJeff Mahoney flags | WQ_HIGHPRI, max_active, 16); 21862a458198SEric Sandeen 21872a458198SEric Sandeen fs_info->delalloc_workers = 2188cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delalloc", 2189cb001095SJeff Mahoney flags, max_active, 2); 21902a458198SEric Sandeen 21912a458198SEric Sandeen fs_info->flush_workers = 2192cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "flush_delalloc", 2193cb001095SJeff Mahoney flags, max_active, 0); 21942a458198SEric Sandeen 21952a458198SEric Sandeen fs_info->caching_workers = 2196cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); 21972a458198SEric Sandeen 21982a458198SEric Sandeen fs_info->fixup_workers = 2199cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); 22002a458198SEric Sandeen 22012a458198SEric Sandeen fs_info->endio_workers = 2202d7b9416fSChristoph Hellwig alloc_workqueue("btrfs-endio", flags, max_active); 22032a458198SEric Sandeen fs_info->endio_meta_workers = 2204d7b9416fSChristoph Hellwig alloc_workqueue("btrfs-endio-meta", flags, max_active); 2205385de0efSChristoph Hellwig fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active); 22062a458198SEric Sandeen fs_info->endio_write_workers = 2207cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-write", flags, 2208cb001095SJeff Mahoney max_active, 2); 2209fed8a72dSChristoph Hellwig fs_info->compressed_write_workers = 2210fed8a72dSChristoph Hellwig alloc_workqueue("btrfs-compressed-write", flags, max_active); 22112a458198SEric Sandeen fs_info->endio_freespace_worker = 2212cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "freespace-write", flags, 2213cb001095SJeff Mahoney max_active, 0); 22142a458198SEric Sandeen fs_info->delayed_workers = 2215cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, 2216cb001095SJeff Mahoney max_active, 0); 22172a458198SEric Sandeen fs_info->qgroup_rescan_workers = 2218cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); 2219b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers = 2220b0643e59SDennis Zhou alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1); 22212a458198SEric Sandeen 2222a31b4a43SChristoph Hellwig if (!(fs_info->workers && fs_info->hipri_workers && 2223a31b4a43SChristoph Hellwig fs_info->delalloc_workers && fs_info->flush_workers && 22242a458198SEric Sandeen fs_info->endio_workers && fs_info->endio_meta_workers && 2225fed8a72dSChristoph Hellwig fs_info->compressed_write_workers && 22261a1a2851SQu Wenruo fs_info->endio_write_workers && 22272a458198SEric Sandeen fs_info->endio_freespace_worker && fs_info->rmw_workers && 2228f26c9238SQu Wenruo fs_info->caching_workers && fs_info->fixup_workers && 2229f26c9238SQu Wenruo fs_info->delayed_workers && fs_info->qgroup_rescan_workers && 2230b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers)) { 22312a458198SEric Sandeen return -ENOMEM; 22322a458198SEric Sandeen } 22332a458198SEric Sandeen 22342a458198SEric Sandeen return 0; 22352a458198SEric Sandeen } 22362a458198SEric Sandeen 22376d97c6e3SJohannes Thumshirn static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type) 22386d97c6e3SJohannes Thumshirn { 22396d97c6e3SJohannes Thumshirn struct crypto_shash *csum_shash; 2240b4e967beSDavid Sterba const char *csum_driver = btrfs_super_csum_driver(csum_type); 22416d97c6e3SJohannes Thumshirn 2242b4e967beSDavid Sterba csum_shash = crypto_alloc_shash(csum_driver, 0, 0); 22436d97c6e3SJohannes Thumshirn 22446d97c6e3SJohannes Thumshirn if (IS_ERR(csum_shash)) { 22456d97c6e3SJohannes Thumshirn btrfs_err(fs_info, "error allocating %s hash for checksum", 2246b4e967beSDavid Sterba csum_driver); 22476d97c6e3SJohannes Thumshirn return PTR_ERR(csum_shash); 22486d97c6e3SJohannes Thumshirn } 22496d97c6e3SJohannes Thumshirn 22506d97c6e3SJohannes Thumshirn fs_info->csum_shash = csum_shash; 22516d97c6e3SJohannes Thumshirn 225268d99ab0SChristoph Hellwig /* 225368d99ab0SChristoph Hellwig * Check if the checksum implementation is a fast accelerated one. 225468d99ab0SChristoph Hellwig * As-is this is a bit of a hack and should be replaced once the csum 225568d99ab0SChristoph Hellwig * implementations provide that information themselves. 225668d99ab0SChristoph Hellwig */ 225768d99ab0SChristoph Hellwig switch (csum_type) { 225868d99ab0SChristoph Hellwig case BTRFS_CSUM_TYPE_CRC32: 225968d99ab0SChristoph Hellwig if (!strstr(crypto_shash_driver_name(csum_shash), "generic")) 226068d99ab0SChristoph Hellwig set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags); 226168d99ab0SChristoph Hellwig break; 226268d99ab0SChristoph Hellwig default: 226368d99ab0SChristoph Hellwig break; 226468d99ab0SChristoph Hellwig } 226568d99ab0SChristoph Hellwig 2266c8a5f8caSDavid Sterba btrfs_info(fs_info, "using %s (%s) checksum algorithm", 2267c8a5f8caSDavid Sterba btrfs_super_csum_name(csum_type), 2268c8a5f8caSDavid Sterba crypto_shash_driver_name(csum_shash)); 22696d97c6e3SJohannes Thumshirn return 0; 22706d97c6e3SJohannes Thumshirn } 22716d97c6e3SJohannes Thumshirn 227263443bf5SEric Sandeen static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 227363443bf5SEric Sandeen struct btrfs_fs_devices *fs_devices) 227463443bf5SEric Sandeen { 227563443bf5SEric Sandeen int ret; 2276789d6a3aSQu Wenruo struct btrfs_tree_parent_check check = { 0 }; 227763443bf5SEric Sandeen struct btrfs_root *log_tree_root; 227863443bf5SEric Sandeen struct btrfs_super_block *disk_super = fs_info->super_copy; 227963443bf5SEric Sandeen u64 bytenr = btrfs_super_log_root(disk_super); 2280581c1760SQu Wenruo int level = btrfs_super_log_root_level(disk_super); 228163443bf5SEric Sandeen 228263443bf5SEric Sandeen if (fs_devices->rw_devices == 0) { 2283f14d104dSDavid Sterba btrfs_warn(fs_info, "log replay required on RO media"); 228463443bf5SEric Sandeen return -EIO; 228563443bf5SEric Sandeen } 228663443bf5SEric Sandeen 228796dfcb46SJosef Bacik log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, 228896dfcb46SJosef Bacik GFP_KERNEL); 228963443bf5SEric Sandeen if (!log_tree_root) 229063443bf5SEric Sandeen return -ENOMEM; 229163443bf5SEric Sandeen 2292789d6a3aSQu Wenruo check.level = level; 2293789d6a3aSQu Wenruo check.transid = fs_info->generation + 1; 2294789d6a3aSQu Wenruo check.owner_root = BTRFS_TREE_LOG_OBJECTID; 2295789d6a3aSQu Wenruo log_tree_root->node = read_tree_block(fs_info, bytenr, &check); 229664c043deSLiu Bo if (IS_ERR(log_tree_root->node)) { 2297f14d104dSDavid Sterba btrfs_warn(fs_info, "failed to read log tree"); 22980eeff236SLiu Bo ret = PTR_ERR(log_tree_root->node); 22998c38938cSJosef Bacik log_tree_root->node = NULL; 230000246528SJosef Bacik btrfs_put_root(log_tree_root); 23010eeff236SLiu Bo return ret; 23024eb150d6SQu Wenruo } 23034eb150d6SQu Wenruo if (!extent_buffer_uptodate(log_tree_root->node)) { 2304f14d104dSDavid Sterba btrfs_err(fs_info, "failed to read log tree"); 230500246528SJosef Bacik btrfs_put_root(log_tree_root); 230663443bf5SEric Sandeen return -EIO; 230763443bf5SEric Sandeen } 23084eb150d6SQu Wenruo 230963443bf5SEric Sandeen /* returns with log_tree_root freed on success */ 231063443bf5SEric Sandeen ret = btrfs_recover_log_trees(log_tree_root); 231163443bf5SEric Sandeen if (ret) { 23120b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 231363443bf5SEric Sandeen "Failed to recover log tree"); 231400246528SJosef Bacik btrfs_put_root(log_tree_root); 231563443bf5SEric Sandeen return ret; 231663443bf5SEric Sandeen } 231763443bf5SEric Sandeen 2318bc98a42cSDavid Howells if (sb_rdonly(fs_info->sb)) { 23196bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 232063443bf5SEric Sandeen if (ret) 232163443bf5SEric Sandeen return ret; 232263443bf5SEric Sandeen } 232363443bf5SEric Sandeen 232463443bf5SEric Sandeen return 0; 232563443bf5SEric Sandeen } 232663443bf5SEric Sandeen 2327abed4aaaSJosef Bacik static int load_global_roots_objectid(struct btrfs_root *tree_root, 2328abed4aaaSJosef Bacik struct btrfs_path *path, u64 objectid, 2329abed4aaaSJosef Bacik const char *name) 2330abed4aaaSJosef Bacik { 2331abed4aaaSJosef Bacik struct btrfs_fs_info *fs_info = tree_root->fs_info; 2332abed4aaaSJosef Bacik struct btrfs_root *root; 2333f7238e50SJosef Bacik u64 max_global_id = 0; 2334abed4aaaSJosef Bacik int ret; 2335abed4aaaSJosef Bacik struct btrfs_key key = { 2336abed4aaaSJosef Bacik .objectid = objectid, 2337abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 2338abed4aaaSJosef Bacik .offset = 0, 2339abed4aaaSJosef Bacik }; 2340abed4aaaSJosef Bacik bool found = false; 2341abed4aaaSJosef Bacik 2342abed4aaaSJosef Bacik /* If we have IGNOREDATACSUMS skip loading these roots. */ 2343abed4aaaSJosef Bacik if (objectid == BTRFS_CSUM_TREE_OBJECTID && 2344abed4aaaSJosef Bacik btrfs_test_opt(fs_info, IGNOREDATACSUMS)) { 2345abed4aaaSJosef Bacik set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state); 2346abed4aaaSJosef Bacik return 0; 2347abed4aaaSJosef Bacik } 2348abed4aaaSJosef Bacik 2349abed4aaaSJosef Bacik while (1) { 2350abed4aaaSJosef Bacik ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0); 2351abed4aaaSJosef Bacik if (ret < 0) 2352abed4aaaSJosef Bacik break; 2353abed4aaaSJosef Bacik 2354abed4aaaSJosef Bacik if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 2355abed4aaaSJosef Bacik ret = btrfs_next_leaf(tree_root, path); 2356abed4aaaSJosef Bacik if (ret) { 2357abed4aaaSJosef Bacik if (ret > 0) 2358abed4aaaSJosef Bacik ret = 0; 2359abed4aaaSJosef Bacik break; 2360abed4aaaSJosef Bacik } 2361abed4aaaSJosef Bacik } 2362abed4aaaSJosef Bacik ret = 0; 2363abed4aaaSJosef Bacik 2364abed4aaaSJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 2365abed4aaaSJosef Bacik if (key.objectid != objectid) 2366abed4aaaSJosef Bacik break; 2367abed4aaaSJosef Bacik btrfs_release_path(path); 2368abed4aaaSJosef Bacik 2369f7238e50SJosef Bacik /* 2370f7238e50SJosef Bacik * Just worry about this for extent tree, it'll be the same for 2371f7238e50SJosef Bacik * everybody. 2372f7238e50SJosef Bacik */ 2373f7238e50SJosef Bacik if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 2374f7238e50SJosef Bacik max_global_id = max(max_global_id, key.offset); 2375f7238e50SJosef Bacik 2376abed4aaaSJosef Bacik found = true; 2377abed4aaaSJosef Bacik root = read_tree_root_path(tree_root, path, &key); 2378abed4aaaSJosef Bacik if (IS_ERR(root)) { 2379abed4aaaSJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) 2380abed4aaaSJosef Bacik ret = PTR_ERR(root); 2381abed4aaaSJosef Bacik break; 2382abed4aaaSJosef Bacik } 2383abed4aaaSJosef Bacik set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2384abed4aaaSJosef Bacik ret = btrfs_global_root_insert(root); 2385abed4aaaSJosef Bacik if (ret) { 2386abed4aaaSJosef Bacik btrfs_put_root(root); 2387abed4aaaSJosef Bacik break; 2388abed4aaaSJosef Bacik } 2389abed4aaaSJosef Bacik key.offset++; 2390abed4aaaSJosef Bacik } 2391abed4aaaSJosef Bacik btrfs_release_path(path); 2392abed4aaaSJosef Bacik 2393f7238e50SJosef Bacik if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 2394f7238e50SJosef Bacik fs_info->nr_global_roots = max_global_id + 1; 2395f7238e50SJosef Bacik 2396abed4aaaSJosef Bacik if (!found || ret) { 2397abed4aaaSJosef Bacik if (objectid == BTRFS_CSUM_TREE_OBJECTID) 2398abed4aaaSJosef Bacik set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state); 2399abed4aaaSJosef Bacik 2400abed4aaaSJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) 2401abed4aaaSJosef Bacik ret = ret ? ret : -ENOENT; 2402abed4aaaSJosef Bacik else 2403abed4aaaSJosef Bacik ret = 0; 2404abed4aaaSJosef Bacik btrfs_err(fs_info, "failed to load root %s", name); 2405abed4aaaSJosef Bacik } 2406abed4aaaSJosef Bacik return ret; 2407abed4aaaSJosef Bacik } 2408abed4aaaSJosef Bacik 2409abed4aaaSJosef Bacik static int load_global_roots(struct btrfs_root *tree_root) 2410abed4aaaSJosef Bacik { 2411abed4aaaSJosef Bacik struct btrfs_path *path; 2412abed4aaaSJosef Bacik int ret = 0; 2413abed4aaaSJosef Bacik 2414abed4aaaSJosef Bacik path = btrfs_alloc_path(); 2415abed4aaaSJosef Bacik if (!path) 2416abed4aaaSJosef Bacik return -ENOMEM; 2417abed4aaaSJosef Bacik 2418abed4aaaSJosef Bacik ret = load_global_roots_objectid(tree_root, path, 2419abed4aaaSJosef Bacik BTRFS_EXTENT_TREE_OBJECTID, "extent"); 2420abed4aaaSJosef Bacik if (ret) 2421abed4aaaSJosef Bacik goto out; 2422abed4aaaSJosef Bacik ret = load_global_roots_objectid(tree_root, path, 2423abed4aaaSJosef Bacik BTRFS_CSUM_TREE_OBJECTID, "csum"); 2424abed4aaaSJosef Bacik if (ret) 2425abed4aaaSJosef Bacik goto out; 2426abed4aaaSJosef Bacik if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE)) 2427abed4aaaSJosef Bacik goto out; 2428abed4aaaSJosef Bacik ret = load_global_roots_objectid(tree_root, path, 2429abed4aaaSJosef Bacik BTRFS_FREE_SPACE_TREE_OBJECTID, 2430abed4aaaSJosef Bacik "free space"); 2431abed4aaaSJosef Bacik out: 2432abed4aaaSJosef Bacik btrfs_free_path(path); 2433abed4aaaSJosef Bacik return ret; 2434abed4aaaSJosef Bacik } 2435abed4aaaSJosef Bacik 24366bccf3abSJeff Mahoney static int btrfs_read_roots(struct btrfs_fs_info *fs_info) 24374bbcaa64SEric Sandeen { 24386bccf3abSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 2439a4f3d2c4SDavid Sterba struct btrfs_root *root; 24404bbcaa64SEric Sandeen struct btrfs_key location; 24414bbcaa64SEric Sandeen int ret; 24424bbcaa64SEric Sandeen 24436bccf3abSJeff Mahoney BUG_ON(!fs_info->tree_root); 24446bccf3abSJeff Mahoney 2445abed4aaaSJosef Bacik ret = load_global_roots(tree_root); 2446abed4aaaSJosef Bacik if (ret) 2447abed4aaaSJosef Bacik return ret; 2448abed4aaaSJosef Bacik 24494bbcaa64SEric Sandeen location.type = BTRFS_ROOT_ITEM_KEY; 24504bbcaa64SEric Sandeen location.offset = 0; 24514bbcaa64SEric Sandeen 24521c56ab99SQu Wenruo if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) { 245314033b08SQu Wenruo location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID; 245414033b08SQu Wenruo root = btrfs_read_tree_root(tree_root, &location); 245514033b08SQu Wenruo if (IS_ERR(root)) { 245614033b08SQu Wenruo if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 245714033b08SQu Wenruo ret = PTR_ERR(root); 245814033b08SQu Wenruo goto out; 245914033b08SQu Wenruo } 246014033b08SQu Wenruo } else { 246114033b08SQu Wenruo set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 246214033b08SQu Wenruo fs_info->block_group_root = root; 246314033b08SQu Wenruo } 246414033b08SQu Wenruo } 246514033b08SQu Wenruo 246614033b08SQu Wenruo location.objectid = BTRFS_DEV_TREE_OBJECTID; 2467a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2468f50f4353SLiu Bo if (IS_ERR(root)) { 246942437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2470f50f4353SLiu Bo ret = PTR_ERR(root); 2471f50f4353SLiu Bo goto out; 2472f50f4353SLiu Bo } 247342437a63SJosef Bacik } else { 2474a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2475a4f3d2c4SDavid Sterba fs_info->dev_root = root; 247642437a63SJosef Bacik } 2477820a49daSJosef Bacik /* Initialize fs_info for all devices in any case */ 2478a8d1b164SJohannes Thumshirn ret = btrfs_init_devices_late(fs_info); 2479a8d1b164SJohannes Thumshirn if (ret) 2480a8d1b164SJohannes Thumshirn goto out; 24814bbcaa64SEric Sandeen 2482aeb935a4SQu Wenruo /* 2483aeb935a4SQu Wenruo * This tree can share blocks with some other fs tree during relocation 2484aeb935a4SQu Wenruo * and we need a proper setup by btrfs_get_fs_root 2485aeb935a4SQu Wenruo */ 248656e9357aSDavid Sterba root = btrfs_get_fs_root(tree_root->fs_info, 248756e9357aSDavid Sterba BTRFS_DATA_RELOC_TREE_OBJECTID, true); 2488aeb935a4SQu Wenruo if (IS_ERR(root)) { 248942437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2490aeb935a4SQu Wenruo ret = PTR_ERR(root); 2491aeb935a4SQu Wenruo goto out; 2492aeb935a4SQu Wenruo } 249342437a63SJosef Bacik } else { 2494aeb935a4SQu Wenruo set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2495aeb935a4SQu Wenruo fs_info->data_reloc_root = root; 249642437a63SJosef Bacik } 2497aeb935a4SQu Wenruo 24984bbcaa64SEric Sandeen location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2499a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2500a4f3d2c4SDavid Sterba if (!IS_ERR(root)) { 2501a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2502afcdd129SJosef Bacik set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); 2503a4f3d2c4SDavid Sterba fs_info->quota_root = root; 25044bbcaa64SEric Sandeen } 25054bbcaa64SEric Sandeen 25064bbcaa64SEric Sandeen location.objectid = BTRFS_UUID_TREE_OBJECTID; 2507a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2508a4f3d2c4SDavid Sterba if (IS_ERR(root)) { 250942437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2510a4f3d2c4SDavid Sterba ret = PTR_ERR(root); 25114bbcaa64SEric Sandeen if (ret != -ENOENT) 2512f50f4353SLiu Bo goto out; 251342437a63SJosef Bacik } 25144bbcaa64SEric Sandeen } else { 2515a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2516a4f3d2c4SDavid Sterba fs_info->uuid_root = root; 25174bbcaa64SEric Sandeen } 25184bbcaa64SEric Sandeen 25194bbcaa64SEric Sandeen return 0; 2520f50f4353SLiu Bo out: 2521f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", 2522f50f4353SLiu Bo location.objectid, ret); 2523f50f4353SLiu Bo return ret; 25244bbcaa64SEric Sandeen } 25254bbcaa64SEric Sandeen 2526069ec957SQu Wenruo /* 2527069ec957SQu Wenruo * Real super block validation 2528069ec957SQu Wenruo * NOTE: super csum type and incompat features will not be checked here. 2529069ec957SQu Wenruo * 2530069ec957SQu Wenruo * @sb: super block to check 2531069ec957SQu Wenruo * @mirror_num: the super block number to check its bytenr: 2532069ec957SQu Wenruo * 0 the primary (1st) sb 2533069ec957SQu Wenruo * 1, 2 2nd and 3rd backup copy 2534069ec957SQu Wenruo * -1 skip bytenr check 2535069ec957SQu Wenruo */ 2536a05d3c91SQu Wenruo int btrfs_validate_super(struct btrfs_fs_info *fs_info, 2537069ec957SQu Wenruo struct btrfs_super_block *sb, int mirror_num) 253821a852b0SQu Wenruo { 253921a852b0SQu Wenruo u64 nodesize = btrfs_super_nodesize(sb); 254021a852b0SQu Wenruo u64 sectorsize = btrfs_super_sectorsize(sb); 254121a852b0SQu Wenruo int ret = 0; 254221a852b0SQu Wenruo 254321a852b0SQu Wenruo if (btrfs_super_magic(sb) != BTRFS_MAGIC) { 254421a852b0SQu Wenruo btrfs_err(fs_info, "no valid FS found"); 254521a852b0SQu Wenruo ret = -EINVAL; 254621a852b0SQu Wenruo } 254721a852b0SQu Wenruo if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { 254821a852b0SQu Wenruo btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", 254921a852b0SQu Wenruo btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); 255021a852b0SQu Wenruo ret = -EINVAL; 255121a852b0SQu Wenruo } 255221a852b0SQu Wenruo if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 255321a852b0SQu Wenruo btrfs_err(fs_info, "tree_root level too big: %d >= %d", 255421a852b0SQu Wenruo btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 255521a852b0SQu Wenruo ret = -EINVAL; 255621a852b0SQu Wenruo } 255721a852b0SQu Wenruo if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 255821a852b0SQu Wenruo btrfs_err(fs_info, "chunk_root level too big: %d >= %d", 255921a852b0SQu Wenruo btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 256021a852b0SQu Wenruo ret = -EINVAL; 256121a852b0SQu Wenruo } 256221a852b0SQu Wenruo if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 256321a852b0SQu Wenruo btrfs_err(fs_info, "log_root level too big: %d >= %d", 256421a852b0SQu Wenruo btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 256521a852b0SQu Wenruo ret = -EINVAL; 256621a852b0SQu Wenruo } 256721a852b0SQu Wenruo 256821a852b0SQu Wenruo /* 256921a852b0SQu Wenruo * Check sectorsize and nodesize first, other check will need it. 257021a852b0SQu Wenruo * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. 257121a852b0SQu Wenruo */ 257221a852b0SQu Wenruo if (!is_power_of_2(sectorsize) || sectorsize < 4096 || 257321a852b0SQu Wenruo sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { 257421a852b0SQu Wenruo btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); 257521a852b0SQu Wenruo ret = -EINVAL; 257621a852b0SQu Wenruo } 25770bb3eb3eSQu Wenruo 25780bb3eb3eSQu Wenruo /* 25791a42daabSQu Wenruo * We only support at most two sectorsizes: 4K and PAGE_SIZE. 25801a42daabSQu Wenruo * 25811a42daabSQu Wenruo * We can support 16K sectorsize with 64K page size without problem, 25821a42daabSQu Wenruo * but such sectorsize/pagesize combination doesn't make much sense. 25831a42daabSQu Wenruo * 4K will be our future standard, PAGE_SIZE is supported from the very 25841a42daabSQu Wenruo * beginning. 25850bb3eb3eSQu Wenruo */ 25861a42daabSQu Wenruo if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) { 258721a852b0SQu Wenruo btrfs_err(fs_info, 25880bb3eb3eSQu Wenruo "sectorsize %llu not yet supported for page size %lu", 258921a852b0SQu Wenruo sectorsize, PAGE_SIZE); 259021a852b0SQu Wenruo ret = -EINVAL; 259121a852b0SQu Wenruo } 25920bb3eb3eSQu Wenruo 259321a852b0SQu Wenruo if (!is_power_of_2(nodesize) || nodesize < sectorsize || 259421a852b0SQu Wenruo nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { 259521a852b0SQu Wenruo btrfs_err(fs_info, "invalid nodesize %llu", nodesize); 259621a852b0SQu Wenruo ret = -EINVAL; 259721a852b0SQu Wenruo } 259821a852b0SQu Wenruo if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { 259921a852b0SQu Wenruo btrfs_err(fs_info, "invalid leafsize %u, should be %llu", 260021a852b0SQu Wenruo le32_to_cpu(sb->__unused_leafsize), nodesize); 260121a852b0SQu Wenruo ret = -EINVAL; 260221a852b0SQu Wenruo } 260321a852b0SQu Wenruo 260421a852b0SQu Wenruo /* Root alignment check */ 260521a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { 260621a852b0SQu Wenruo btrfs_warn(fs_info, "tree_root block unaligned: %llu", 260721a852b0SQu Wenruo btrfs_super_root(sb)); 260821a852b0SQu Wenruo ret = -EINVAL; 260921a852b0SQu Wenruo } 261021a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { 261121a852b0SQu Wenruo btrfs_warn(fs_info, "chunk_root block unaligned: %llu", 261221a852b0SQu Wenruo btrfs_super_chunk_root(sb)); 261321a852b0SQu Wenruo ret = -EINVAL; 261421a852b0SQu Wenruo } 261521a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { 261621a852b0SQu Wenruo btrfs_warn(fs_info, "log_root block unaligned: %llu", 261721a852b0SQu Wenruo btrfs_super_log_root(sb)); 261821a852b0SQu Wenruo ret = -EINVAL; 261921a852b0SQu Wenruo } 262021a852b0SQu Wenruo 2621aefd7f70SNikolay Borisov if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, 2622aefd7f70SNikolay Borisov BTRFS_FSID_SIZE)) { 2623aefd7f70SNikolay Borisov btrfs_err(fs_info, 2624aefd7f70SNikolay Borisov "superblock fsid doesn't match fsid of fs_devices: %pU != %pU", 2625aefd7f70SNikolay Borisov fs_info->super_copy->fsid, fs_info->fs_devices->fsid); 2626aefd7f70SNikolay Borisov ret = -EINVAL; 2627aefd7f70SNikolay Borisov } 2628aefd7f70SNikolay Borisov 2629aefd7f70SNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID) && 2630aefd7f70SNikolay Borisov memcmp(fs_info->fs_devices->metadata_uuid, 2631aefd7f70SNikolay Borisov fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) { 2632aefd7f70SNikolay Borisov btrfs_err(fs_info, 2633aefd7f70SNikolay Borisov "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU", 2634aefd7f70SNikolay Borisov fs_info->super_copy->metadata_uuid, 2635aefd7f70SNikolay Borisov fs_info->fs_devices->metadata_uuid); 2636aefd7f70SNikolay Borisov ret = -EINVAL; 2637aefd7f70SNikolay Borisov } 2638aefd7f70SNikolay Borisov 26391c56ab99SQu Wenruo /* 26401c56ab99SQu Wenruo * Artificial requirement for block-group-tree to force newer features 26411c56ab99SQu Wenruo * (free-space-tree, no-holes) so the test matrix is smaller. 26421c56ab99SQu Wenruo */ 26431c56ab99SQu Wenruo if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) && 26441c56ab99SQu Wenruo (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) || 26451c56ab99SQu Wenruo !btrfs_fs_incompat(fs_info, NO_HOLES))) { 26461c56ab99SQu Wenruo btrfs_err(fs_info, 26471c56ab99SQu Wenruo "block-group-tree feature requires fres-space-tree and no-holes"); 26481c56ab99SQu Wenruo ret = -EINVAL; 26491c56ab99SQu Wenruo } 26501c56ab99SQu Wenruo 2651de37aa51SNikolay Borisov if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, 26527239ff4bSNikolay Borisov BTRFS_FSID_SIZE) != 0) { 265321a852b0SQu Wenruo btrfs_err(fs_info, 26547239ff4bSNikolay Borisov "dev_item UUID does not match metadata fsid: %pU != %pU", 2655de37aa51SNikolay Borisov fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); 265621a852b0SQu Wenruo ret = -EINVAL; 265721a852b0SQu Wenruo } 265821a852b0SQu Wenruo 265921a852b0SQu Wenruo /* 266021a852b0SQu Wenruo * Hint to catch really bogus numbers, bitflips or so, more exact checks are 266121a852b0SQu Wenruo * done later 266221a852b0SQu Wenruo */ 266321a852b0SQu Wenruo if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { 266421a852b0SQu Wenruo btrfs_err(fs_info, "bytes_used is too small %llu", 266521a852b0SQu Wenruo btrfs_super_bytes_used(sb)); 266621a852b0SQu Wenruo ret = -EINVAL; 266721a852b0SQu Wenruo } 266821a852b0SQu Wenruo if (!is_power_of_2(btrfs_super_stripesize(sb))) { 266921a852b0SQu Wenruo btrfs_err(fs_info, "invalid stripesize %u", 267021a852b0SQu Wenruo btrfs_super_stripesize(sb)); 267121a852b0SQu Wenruo ret = -EINVAL; 267221a852b0SQu Wenruo } 267321a852b0SQu Wenruo if (btrfs_super_num_devices(sb) > (1UL << 31)) 267421a852b0SQu Wenruo btrfs_warn(fs_info, "suspicious number of devices: %llu", 267521a852b0SQu Wenruo btrfs_super_num_devices(sb)); 267621a852b0SQu Wenruo if (btrfs_super_num_devices(sb) == 0) { 267721a852b0SQu Wenruo btrfs_err(fs_info, "number of devices is 0"); 267821a852b0SQu Wenruo ret = -EINVAL; 267921a852b0SQu Wenruo } 268021a852b0SQu Wenruo 2681069ec957SQu Wenruo if (mirror_num >= 0 && 2682069ec957SQu Wenruo btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { 268321a852b0SQu Wenruo btrfs_err(fs_info, "super offset mismatch %llu != %u", 268421a852b0SQu Wenruo btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 268521a852b0SQu Wenruo ret = -EINVAL; 268621a852b0SQu Wenruo } 268721a852b0SQu Wenruo 268821a852b0SQu Wenruo /* 268921a852b0SQu Wenruo * Obvious sys_chunk_array corruptions, it must hold at least one key 269021a852b0SQu Wenruo * and one chunk 269121a852b0SQu Wenruo */ 269221a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 269321a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too big %u > %u", 269421a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 269521a852b0SQu Wenruo BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 269621a852b0SQu Wenruo ret = -EINVAL; 269721a852b0SQu Wenruo } 269821a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 269921a852b0SQu Wenruo + sizeof(struct btrfs_chunk)) { 270021a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too small %u < %zu", 270121a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 270221a852b0SQu Wenruo sizeof(struct btrfs_disk_key) 270321a852b0SQu Wenruo + sizeof(struct btrfs_chunk)); 270421a852b0SQu Wenruo ret = -EINVAL; 270521a852b0SQu Wenruo } 270621a852b0SQu Wenruo 270721a852b0SQu Wenruo /* 270821a852b0SQu Wenruo * The generation is a global counter, we'll trust it more than the others 270921a852b0SQu Wenruo * but it's still possible that it's the one that's wrong. 271021a852b0SQu Wenruo */ 271121a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 271221a852b0SQu Wenruo btrfs_warn(fs_info, 271321a852b0SQu Wenruo "suspicious: generation < chunk_root_generation: %llu < %llu", 271421a852b0SQu Wenruo btrfs_super_generation(sb), 271521a852b0SQu Wenruo btrfs_super_chunk_root_generation(sb)); 271621a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 271721a852b0SQu Wenruo && btrfs_super_cache_generation(sb) != (u64)-1) 271821a852b0SQu Wenruo btrfs_warn(fs_info, 271921a852b0SQu Wenruo "suspicious: generation < cache_generation: %llu < %llu", 272021a852b0SQu Wenruo btrfs_super_generation(sb), 272121a852b0SQu Wenruo btrfs_super_cache_generation(sb)); 272221a852b0SQu Wenruo 272321a852b0SQu Wenruo return ret; 272421a852b0SQu Wenruo } 272521a852b0SQu Wenruo 2726069ec957SQu Wenruo /* 2727069ec957SQu Wenruo * Validation of super block at mount time. 2728069ec957SQu Wenruo * Some checks already done early at mount time, like csum type and incompat 2729069ec957SQu Wenruo * flags will be skipped. 2730069ec957SQu Wenruo */ 2731069ec957SQu Wenruo static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) 2732069ec957SQu Wenruo { 2733a05d3c91SQu Wenruo return btrfs_validate_super(fs_info, fs_info->super_copy, 0); 2734069ec957SQu Wenruo } 2735069ec957SQu Wenruo 273675cb857dSQu Wenruo /* 273775cb857dSQu Wenruo * Validation of super block at write time. 273875cb857dSQu Wenruo * Some checks like bytenr check will be skipped as their values will be 273975cb857dSQu Wenruo * overwritten soon. 274075cb857dSQu Wenruo * Extra checks like csum type and incompat flags will be done here. 274175cb857dSQu Wenruo */ 274275cb857dSQu Wenruo static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, 274375cb857dSQu Wenruo struct btrfs_super_block *sb) 274475cb857dSQu Wenruo { 274575cb857dSQu Wenruo int ret; 274675cb857dSQu Wenruo 2747a05d3c91SQu Wenruo ret = btrfs_validate_super(fs_info, sb, -1); 274875cb857dSQu Wenruo if (ret < 0) 274975cb857dSQu Wenruo goto out; 2750e7e16f48SJohannes Thumshirn if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) { 275175cb857dSQu Wenruo ret = -EUCLEAN; 275275cb857dSQu Wenruo btrfs_err(fs_info, "invalid csum type, has %u want %u", 275375cb857dSQu Wenruo btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); 275475cb857dSQu Wenruo goto out; 275575cb857dSQu Wenruo } 275675cb857dSQu Wenruo if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 275775cb857dSQu Wenruo ret = -EUCLEAN; 275875cb857dSQu Wenruo btrfs_err(fs_info, 275975cb857dSQu Wenruo "invalid incompat flags, has 0x%llx valid mask 0x%llx", 276075cb857dSQu Wenruo btrfs_super_incompat_flags(sb), 276175cb857dSQu Wenruo (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); 276275cb857dSQu Wenruo goto out; 276375cb857dSQu Wenruo } 276475cb857dSQu Wenruo out: 276575cb857dSQu Wenruo if (ret < 0) 276675cb857dSQu Wenruo btrfs_err(fs_info, 276775cb857dSQu Wenruo "super block corruption detected before writing it to disk"); 276875cb857dSQu Wenruo return ret; 276975cb857dSQu Wenruo } 277075cb857dSQu Wenruo 2771bd676446SJosef Bacik static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level) 2772bd676446SJosef Bacik { 2773789d6a3aSQu Wenruo struct btrfs_tree_parent_check check = { 2774789d6a3aSQu Wenruo .level = level, 2775789d6a3aSQu Wenruo .transid = gen, 2776789d6a3aSQu Wenruo .owner_root = root->root_key.objectid 2777789d6a3aSQu Wenruo }; 2778bd676446SJosef Bacik int ret = 0; 2779bd676446SJosef Bacik 2780789d6a3aSQu Wenruo root->node = read_tree_block(root->fs_info, bytenr, &check); 2781bd676446SJosef Bacik if (IS_ERR(root->node)) { 2782bd676446SJosef Bacik ret = PTR_ERR(root->node); 2783bd676446SJosef Bacik root->node = NULL; 27844eb150d6SQu Wenruo return ret; 27854eb150d6SQu Wenruo } 27864eb150d6SQu Wenruo if (!extent_buffer_uptodate(root->node)) { 2787bd676446SJosef Bacik free_extent_buffer(root->node); 2788bd676446SJosef Bacik root->node = NULL; 27894eb150d6SQu Wenruo return -EIO; 2790bd676446SJosef Bacik } 2791bd676446SJosef Bacik 2792bd676446SJosef Bacik btrfs_set_root_node(&root->root_item, root->node); 2793bd676446SJosef Bacik root->commit_root = btrfs_root_node(root); 2794bd676446SJosef Bacik btrfs_set_root_refs(&root->root_item, 1); 2795bd676446SJosef Bacik return ret; 2796bd676446SJosef Bacik } 2797bd676446SJosef Bacik 2798bd676446SJosef Bacik static int load_important_roots(struct btrfs_fs_info *fs_info) 2799bd676446SJosef Bacik { 2800bd676446SJosef Bacik struct btrfs_super_block *sb = fs_info->super_copy; 2801bd676446SJosef Bacik u64 gen, bytenr; 2802bd676446SJosef Bacik int level, ret; 2803bd676446SJosef Bacik 2804bd676446SJosef Bacik bytenr = btrfs_super_root(sb); 2805bd676446SJosef Bacik gen = btrfs_super_generation(sb); 2806bd676446SJosef Bacik level = btrfs_super_root_level(sb); 2807bd676446SJosef Bacik ret = load_super_root(fs_info->tree_root, bytenr, gen, level); 28089c54e80dSJosef Bacik if (ret) { 2809bd676446SJosef Bacik btrfs_warn(fs_info, "couldn't read tree root"); 2810bd676446SJosef Bacik return ret; 2811bd676446SJosef Bacik } 28129c54e80dSJosef Bacik return 0; 28139c54e80dSJosef Bacik } 28149c54e80dSJosef Bacik 28156ef108ddSNikolay Borisov static int __cold init_tree_roots(struct btrfs_fs_info *fs_info) 2816b8522a1eSNikolay Borisov { 28176ef108ddSNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 2818b8522a1eSNikolay Borisov struct btrfs_super_block *sb = fs_info->super_copy; 2819b8522a1eSNikolay Borisov struct btrfs_root *tree_root = fs_info->tree_root; 2820b8522a1eSNikolay Borisov bool handle_error = false; 2821b8522a1eSNikolay Borisov int ret = 0; 2822b8522a1eSNikolay Borisov int i; 2823b8522a1eSNikolay Borisov 2824b8522a1eSNikolay Borisov for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 2825b8522a1eSNikolay Borisov if (handle_error) { 2826b8522a1eSNikolay Borisov if (!IS_ERR(tree_root->node)) 2827b8522a1eSNikolay Borisov free_extent_buffer(tree_root->node); 2828b8522a1eSNikolay Borisov tree_root->node = NULL; 2829b8522a1eSNikolay Borisov 2830b8522a1eSNikolay Borisov if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) 2831b8522a1eSNikolay Borisov break; 2832b8522a1eSNikolay Borisov 2833b8522a1eSNikolay Borisov free_root_pointers(fs_info, 0); 2834b8522a1eSNikolay Borisov 2835b8522a1eSNikolay Borisov /* 2836b8522a1eSNikolay Borisov * Don't use the log in recovery mode, it won't be 2837b8522a1eSNikolay Borisov * valid 2838b8522a1eSNikolay Borisov */ 2839b8522a1eSNikolay Borisov btrfs_set_super_log_root(sb, 0); 2840b8522a1eSNikolay Borisov 2841b8522a1eSNikolay Borisov /* We can't trust the free space cache either */ 2842b8522a1eSNikolay Borisov btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 2843b8522a1eSNikolay Borisov 2844b8522a1eSNikolay Borisov ret = read_backup_root(fs_info, i); 28456ef108ddSNikolay Borisov backup_index = ret; 2846b8522a1eSNikolay Borisov if (ret < 0) 2847b8522a1eSNikolay Borisov return ret; 2848b8522a1eSNikolay Borisov } 2849b8522a1eSNikolay Borisov 2850bd676446SJosef Bacik ret = load_important_roots(fs_info); 2851bd676446SJosef Bacik if (ret) { 2852217f5004SNikolay Borisov handle_error = true; 2853b8522a1eSNikolay Borisov continue; 2854b8522a1eSNikolay Borisov } 2855b8522a1eSNikolay Borisov 2856336a0d8dSNikolay Borisov /* 2857336a0d8dSNikolay Borisov * No need to hold btrfs_root::objectid_mutex since the fs 2858336a0d8dSNikolay Borisov * hasn't been fully initialised and we are the only user 2859336a0d8dSNikolay Borisov */ 2860453e4873SNikolay Borisov ret = btrfs_init_root_free_objectid(tree_root); 2861b8522a1eSNikolay Borisov if (ret < 0) { 2862b8522a1eSNikolay Borisov handle_error = true; 2863b8522a1eSNikolay Borisov continue; 2864b8522a1eSNikolay Borisov } 2865b8522a1eSNikolay Borisov 28666b8fad57SNikolay Borisov ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID); 2867b8522a1eSNikolay Borisov 2868b8522a1eSNikolay Borisov ret = btrfs_read_roots(fs_info); 2869b8522a1eSNikolay Borisov if (ret < 0) { 2870b8522a1eSNikolay Borisov handle_error = true; 2871b8522a1eSNikolay Borisov continue; 2872b8522a1eSNikolay Borisov } 2873b8522a1eSNikolay Borisov 2874b8522a1eSNikolay Borisov /* All successful */ 2875bd676446SJosef Bacik fs_info->generation = btrfs_header_generation(tree_root->node); 2876bd676446SJosef Bacik fs_info->last_trans_committed = fs_info->generation; 2877d96b3424SFilipe Manana fs_info->last_reloc_trans = 0; 28786ef108ddSNikolay Borisov 28796ef108ddSNikolay Borisov /* Always begin writing backup roots after the one being used */ 28806ef108ddSNikolay Borisov if (backup_index < 0) { 28816ef108ddSNikolay Borisov fs_info->backup_root_index = 0; 28826ef108ddSNikolay Borisov } else { 28836ef108ddSNikolay Borisov fs_info->backup_root_index = backup_index + 1; 28846ef108ddSNikolay Borisov fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS; 28856ef108ddSNikolay Borisov } 2886b8522a1eSNikolay Borisov break; 2887b8522a1eSNikolay Borisov } 2888b8522a1eSNikolay Borisov 2889b8522a1eSNikolay Borisov return ret; 2890b8522a1eSNikolay Borisov } 2891b8522a1eSNikolay Borisov 28928260edbaSJosef Bacik void btrfs_init_fs_info(struct btrfs_fs_info *fs_info) 2893eb60ceacSChris Mason { 2894fc7cbcd4SDavid Sterba INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 289501cd3909SDavid Sterba INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 28968fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 2897facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 289824bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 2899eb73c1b7SMiao Xie INIT_LIST_HEAD(&fs_info->delalloc_roots); 290011833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 2901eb73c1b7SMiao Xie spin_lock_init(&fs_info->delalloc_root_lock); 2902a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 2903fc7cbcd4SDavid Sterba spin_lock_init(&fs_info->fs_roots_radix_lock); 290424bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 29054cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 2906ceda0864SMiao Xie spin_lock_init(&fs_info->super_lock); 2907f28491e0SJosef Bacik spin_lock_init(&fs_info->buffer_lock); 290847ab2a6cSJosef Bacik spin_lock_init(&fs_info->unused_bgs_lock); 290940ab3be1SNaohiro Aota spin_lock_init(&fs_info->treelog_bg_lock); 2910afba2bc0SNaohiro Aota spin_lock_init(&fs_info->zone_active_bgs_lock); 2911c2707a25SJohannes Thumshirn spin_lock_init(&fs_info->relocation_bg_lock); 2912f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 2913abed4aaaSJosef Bacik rwlock_init(&fs_info->global_root_lock); 2914d7c15171SZhao Lei mutex_init(&fs_info->unused_bg_unpin_mutex); 2915f3372065SJohannes Thumshirn mutex_init(&fs_info->reclaim_bgs_lock); 29167585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 2917573bfb72SMiao Xie mutex_init(&fs_info->delalloc_root_mutex); 29180bc09ca1SNaohiro Aota mutex_init(&fs_info->zoned_meta_io_lock); 29195f0addf7SNaohiro Aota mutex_init(&fs_info->zoned_data_reloc_io_lock); 2920de98ced9SMiao Xie seqlock_init(&fs_info->profiles_lock); 292119c00ddcSChris Mason 2922e1489b4fSIoannis Angelakopoulos btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers); 29235a9ba670SIoannis Angelakopoulos btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters); 29248b53779eSIoannis Angelakopoulos btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered); 29255f4403e1SIoannis Angelakopoulos btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent); 29263e738c53SIoannis Angelakopoulos btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_start, 29273e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_COMMIT_START); 29283e738c53SIoannis Angelakopoulos btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked, 29293e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_UNBLOCKED); 29303e738c53SIoannis Angelakopoulos btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed, 29313e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 29323e738c53SIoannis Angelakopoulos btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed, 29333e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_COMPLETED); 2934e1489b4fSIoannis Angelakopoulos 29350b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 29366324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 2937f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 293847ab2a6cSJosef Bacik INIT_LIST_HEAD(&fs_info->unused_bgs); 293918bb8bbfSJohannes Thumshirn INIT_LIST_HEAD(&fs_info->reclaim_bgs); 2940afba2bc0SNaohiro Aota INIT_LIST_HEAD(&fs_info->zone_active_bgs); 2941bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 2942bd647ce3SJosef Bacik INIT_LIST_HEAD(&fs_info->allocated_roots); 29433fd63727SJosef Bacik INIT_LIST_HEAD(&fs_info->allocated_ebs); 29443fd63727SJosef Bacik spin_lock_init(&fs_info->eb_leak_lock); 2945bd647ce3SJosef Bacik #endif 2946c8bf1b67SDavid Sterba extent_map_tree_init(&fs_info->mapping_tree); 294766d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 294866d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 294966d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 295066d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 295166d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 295266d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 295366d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 2954ba2c4d4eSJosef Bacik btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, 2955ba2c4d4eSJosef Bacik BTRFS_BLOCK_RSV_DELREFS); 2956ba2c4d4eSJosef Bacik 2957771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 29584cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 2959034f784dSJosef Bacik atomic_set(&fs_info->nr_delayed_iputs, 0); 2960fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 2961abed4aaaSJosef Bacik fs_info->global_root_tree = RB_ROOT; 296295ac567aSFilipe David Borba Manana fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 29639ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 29644cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 2965a5ed45f8SNikolay Borisov atomic64_set(&fs_info->free_chunk_space, 0); 2966f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 29678b87dc17SDavid Sterba fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 2968f8c269d7SDavid Sterba fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 2969fd708b81SJosef Bacik btrfs_init_ref_verify(fs_info); 2970c8b97818SChris Mason 2971b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 2972b34b086cSChris Mason num_online_cpus() + 2, 8); 29730afbaf8cSChris Mason 2974199c2a9cSMiao Xie INIT_LIST_HEAD(&fs_info->ordered_roots); 2975199c2a9cSMiao Xie spin_lock_init(&fs_info->ordered_root_lock); 297669fe2d75SJosef Bacik 2977638aa7edSEric Sandeen btrfs_init_scrub(fs_info); 297821adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 297921adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 298021adbd5cSStefan Behrens #endif 2981779a65a4SEric Sandeen btrfs_init_balance(fs_info); 298257056740SJosef Bacik btrfs_init_async_reclaim_work(fs_info); 2983a2de733cSArne Jansen 298416b0c258SFilipe Manana rwlock_init(&fs_info->block_group_cache_lock); 298508dddb29SFilipe Manana fs_info->block_group_cache_tree = RB_ROOT_CACHED; 29860f9dd46cSJosef Bacik 2987fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &fs_info->excluded_extents, 298835da5a7eSDavid Sterba IO_TREE_FS_EXCLUDED_EXTENTS); 298939279cc3SChris Mason 29905a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 2991e02119d5SChris Mason mutex_init(&fs_info->tree_log_mutex); 2992925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 2993a74a4b97SChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 2994a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 29951bbc621eSChris Mason mutex_init(&fs_info->ro_block_group_mutex); 29969e351cc8SJosef Bacik init_rwsem(&fs_info->commit_root_sem); 2997c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 299876dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 2999803b2f54SStefan Behrens sema_init(&fs_info->uuid_tree_rescan_sem, 1); 3000fa9c0d79SChris Mason 3001ad618368SEric Sandeen btrfs_init_dev_replace_locks(fs_info); 3002f9e92e40SEric Sandeen btrfs_init_qgroup(fs_info); 3003b0643e59SDennis Zhou btrfs_discard_init(fs_info); 3004416ac51dSArne Jansen 3005fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 3006fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 3007fa9c0d79SChris Mason 3008e6dcd2dcSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 3009f9295749SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 3010bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 30114854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 3012034f784dSJosef Bacik init_waitqueue_head(&fs_info->delayed_iputs_wait); 30133768f368SChris Mason 3014da17066cSJeff Mahoney /* Usable values until the real ones are cached from the superblock */ 3015da17066cSJeff Mahoney fs_info->nodesize = 4096; 3016da17066cSJeff Mahoney fs_info->sectorsize = 4096; 3017ab108d99SDavid Sterba fs_info->sectorsize_bits = ilog2(4096); 3018da17066cSJeff Mahoney fs_info->stripesize = 4096; 3019da17066cSJeff Mahoney 3020f7b12a62SNaohiro Aota fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE; 3021f7b12a62SNaohiro Aota 3022eede2bf3SOmar Sandoval spin_lock_init(&fs_info->swapfile_pins_lock); 3023eede2bf3SOmar Sandoval fs_info->swapfile_pins = RB_ROOT; 3024eede2bf3SOmar Sandoval 302518bb8bbfSJohannes Thumshirn fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH; 302618bb8bbfSJohannes Thumshirn INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work); 30278260edbaSJosef Bacik } 30288260edbaSJosef Bacik 30298260edbaSJosef Bacik static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb) 30308260edbaSJosef Bacik { 30318260edbaSJosef Bacik int ret; 30328260edbaSJosef Bacik 30338260edbaSJosef Bacik fs_info->sb = sb; 30348260edbaSJosef Bacik sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; 30358260edbaSJosef Bacik sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); 30369e967495SFilipe Manana 30375deb17e1SJosef Bacik ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL); 3038ae18c37aSJosef Bacik if (ret) 3039ae18c37aSJosef Bacik return ret; 3040ae18c37aSJosef Bacik 3041ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 3042ae18c37aSJosef Bacik if (ret) 3043c75e8394SJosef Bacik return ret; 3044ae18c37aSJosef Bacik 3045ae18c37aSJosef Bacik fs_info->dirty_metadata_batch = PAGE_SIZE * 3046ae18c37aSJosef Bacik (1 + ilog2(nr_cpu_ids)); 3047ae18c37aSJosef Bacik 3048ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 3049ae18c37aSJosef Bacik if (ret) 3050c75e8394SJosef Bacik return ret; 3051ae18c37aSJosef Bacik 3052ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, 3053ae18c37aSJosef Bacik GFP_KERNEL); 3054ae18c37aSJosef Bacik if (ret) 3055c75e8394SJosef Bacik return ret; 3056ae18c37aSJosef Bacik 3057ae18c37aSJosef Bacik fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 3058ae18c37aSJosef Bacik GFP_KERNEL); 3059c75e8394SJosef Bacik if (!fs_info->delayed_root) 3060c75e8394SJosef Bacik return -ENOMEM; 3061ae18c37aSJosef Bacik btrfs_init_delayed_root(fs_info->delayed_root); 3062ae18c37aSJosef Bacik 3063a0a1db70SFilipe Manana if (sb_rdonly(sb)) 3064a0a1db70SFilipe Manana set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state); 3065a0a1db70SFilipe Manana 3066c75e8394SJosef Bacik return btrfs_alloc_stripe_hash_table(fs_info); 3067ae18c37aSJosef Bacik } 3068ae18c37aSJosef Bacik 306997f4dd09SNikolay Borisov static int btrfs_uuid_rescan_kthread(void *data) 307097f4dd09SNikolay Borisov { 30710d031dc4SYu Zhe struct btrfs_fs_info *fs_info = data; 307297f4dd09SNikolay Borisov int ret; 307397f4dd09SNikolay Borisov 307497f4dd09SNikolay Borisov /* 307597f4dd09SNikolay Borisov * 1st step is to iterate through the existing UUID tree and 307697f4dd09SNikolay Borisov * to delete all entries that contain outdated data. 307797f4dd09SNikolay Borisov * 2nd step is to add all missing entries to the UUID tree. 307897f4dd09SNikolay Borisov */ 307997f4dd09SNikolay Borisov ret = btrfs_uuid_tree_iterate(fs_info); 308097f4dd09SNikolay Borisov if (ret < 0) { 3081c94bec2cSJosef Bacik if (ret != -EINTR) 3082c94bec2cSJosef Bacik btrfs_warn(fs_info, "iterating uuid_tree failed %d", 3083c94bec2cSJosef Bacik ret); 308497f4dd09SNikolay Borisov up(&fs_info->uuid_tree_rescan_sem); 308597f4dd09SNikolay Borisov return ret; 308697f4dd09SNikolay Borisov } 308797f4dd09SNikolay Borisov return btrfs_uuid_scan_kthread(data); 308897f4dd09SNikolay Borisov } 308997f4dd09SNikolay Borisov 309097f4dd09SNikolay Borisov static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) 309197f4dd09SNikolay Borisov { 309297f4dd09SNikolay Borisov struct task_struct *task; 309397f4dd09SNikolay Borisov 309497f4dd09SNikolay Borisov down(&fs_info->uuid_tree_rescan_sem); 309597f4dd09SNikolay Borisov task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); 309697f4dd09SNikolay Borisov if (IS_ERR(task)) { 309797f4dd09SNikolay Borisov /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 309897f4dd09SNikolay Borisov btrfs_warn(fs_info, "failed to start uuid_rescan task"); 309997f4dd09SNikolay Borisov up(&fs_info->uuid_tree_rescan_sem); 310097f4dd09SNikolay Borisov return PTR_ERR(task); 310197f4dd09SNikolay Borisov } 310297f4dd09SNikolay Borisov 310397f4dd09SNikolay Borisov return 0; 310497f4dd09SNikolay Borisov } 310597f4dd09SNikolay Borisov 310644c0ca21SBoris Burkov /* 31078cd29088SBoris Burkov * Some options only have meaning at mount time and shouldn't persist across 31088cd29088SBoris Burkov * remounts, or be displayed. Clear these at the end of mount and remount 31098cd29088SBoris Burkov * code paths. 31108cd29088SBoris Burkov */ 31118cd29088SBoris Burkov void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info) 31128cd29088SBoris Burkov { 31138cd29088SBoris Burkov btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT); 31148b228324SBoris Burkov btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE); 31158cd29088SBoris Burkov } 31168cd29088SBoris Burkov 31178cd29088SBoris Burkov /* 311844c0ca21SBoris Burkov * Mounting logic specific to read-write file systems. Shared by open_ctree 311944c0ca21SBoris Burkov * and btrfs_remount when remounting from read-only to read-write. 312044c0ca21SBoris Burkov */ 312144c0ca21SBoris Burkov int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info) 312244c0ca21SBoris Burkov { 312344c0ca21SBoris Burkov int ret; 312494846229SBoris Burkov const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE); 31258b228324SBoris Burkov bool clear_free_space_tree = false; 31268b228324SBoris Burkov 31278b228324SBoris Burkov if (btrfs_test_opt(fs_info, CLEAR_CACHE) && 31288b228324SBoris Burkov btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 31298b228324SBoris Burkov clear_free_space_tree = true; 31308b228324SBoris Burkov } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 31318b228324SBoris Burkov !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { 31328b228324SBoris Burkov btrfs_warn(fs_info, "free space tree is invalid"); 31338b228324SBoris Burkov clear_free_space_tree = true; 31348b228324SBoris Burkov } 31358b228324SBoris Burkov 31368b228324SBoris Burkov if (clear_free_space_tree) { 31378b228324SBoris Burkov btrfs_info(fs_info, "clearing free space tree"); 31388b228324SBoris Burkov ret = btrfs_clear_free_space_tree(fs_info); 31398b228324SBoris Burkov if (ret) { 31408b228324SBoris Burkov btrfs_warn(fs_info, 31418b228324SBoris Burkov "failed to clear free space tree: %d", ret); 31428b228324SBoris Burkov goto out; 31438b228324SBoris Burkov } 31448b228324SBoris Burkov } 314544c0ca21SBoris Burkov 31468d488a8cSFilipe Manana /* 31478d488a8cSFilipe Manana * btrfs_find_orphan_roots() is responsible for finding all the dead 31488d488a8cSFilipe Manana * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load 3149fc7cbcd4SDavid Sterba * them into the fs_info->fs_roots_radix tree. This must be done before 31508d488a8cSFilipe Manana * calling btrfs_orphan_cleanup() on the tree root. If we don't do it 31518d488a8cSFilipe Manana * first, then btrfs_orphan_cleanup() will delete a dead root's orphan 31528d488a8cSFilipe Manana * item before the root's tree is deleted - this means that if we unmount 31538d488a8cSFilipe Manana * or crash before the deletion completes, on the next mount we will not 31548d488a8cSFilipe Manana * delete what remains of the tree because the orphan item does not 31558d488a8cSFilipe Manana * exists anymore, which is what tells us we have a pending deletion. 31568d488a8cSFilipe Manana */ 31578d488a8cSFilipe Manana ret = btrfs_find_orphan_roots(fs_info); 31588d488a8cSFilipe Manana if (ret) 31598d488a8cSFilipe Manana goto out; 31608d488a8cSFilipe Manana 316144c0ca21SBoris Burkov ret = btrfs_cleanup_fs_roots(fs_info); 316244c0ca21SBoris Burkov if (ret) 316344c0ca21SBoris Burkov goto out; 316444c0ca21SBoris Burkov 31658f1c21d7SBoris Burkov down_read(&fs_info->cleanup_work_sem); 31668f1c21d7SBoris Burkov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 31678f1c21d7SBoris Burkov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 31688f1c21d7SBoris Burkov up_read(&fs_info->cleanup_work_sem); 31698f1c21d7SBoris Burkov goto out; 31708f1c21d7SBoris Burkov } 31718f1c21d7SBoris Burkov up_read(&fs_info->cleanup_work_sem); 31728f1c21d7SBoris Burkov 317344c0ca21SBoris Burkov mutex_lock(&fs_info->cleaner_mutex); 31747eefae6bSJosef Bacik ret = btrfs_recover_relocation(fs_info); 317544c0ca21SBoris Burkov mutex_unlock(&fs_info->cleaner_mutex); 317644c0ca21SBoris Burkov if (ret < 0) { 317744c0ca21SBoris Burkov btrfs_warn(fs_info, "failed to recover relocation: %d", ret); 317844c0ca21SBoris Burkov goto out; 317944c0ca21SBoris Burkov } 318044c0ca21SBoris Burkov 31815011139aSBoris Burkov if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && 31825011139aSBoris Burkov !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 31835011139aSBoris Burkov btrfs_info(fs_info, "creating free space tree"); 31845011139aSBoris Burkov ret = btrfs_create_free_space_tree(fs_info); 31855011139aSBoris Burkov if (ret) { 31865011139aSBoris Burkov btrfs_warn(fs_info, 31875011139aSBoris Burkov "failed to create free space tree: %d", ret); 31885011139aSBoris Burkov goto out; 31895011139aSBoris Burkov } 31905011139aSBoris Burkov } 31915011139aSBoris Burkov 319294846229SBoris Burkov if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) { 319394846229SBoris Burkov ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt); 319494846229SBoris Burkov if (ret) 319594846229SBoris Burkov goto out; 319694846229SBoris Burkov } 319794846229SBoris Burkov 319844c0ca21SBoris Burkov ret = btrfs_resume_balance_async(fs_info); 319944c0ca21SBoris Burkov if (ret) 320044c0ca21SBoris Burkov goto out; 320144c0ca21SBoris Burkov 320244c0ca21SBoris Burkov ret = btrfs_resume_dev_replace_async(fs_info); 320344c0ca21SBoris Burkov if (ret) { 320444c0ca21SBoris Burkov btrfs_warn(fs_info, "failed to resume dev_replace"); 320544c0ca21SBoris Burkov goto out; 320644c0ca21SBoris Burkov } 320744c0ca21SBoris Burkov 320844c0ca21SBoris Burkov btrfs_qgroup_rescan_resume(fs_info); 320944c0ca21SBoris Burkov 321044c0ca21SBoris Burkov if (!fs_info->uuid_root) { 321144c0ca21SBoris Burkov btrfs_info(fs_info, "creating UUID tree"); 321244c0ca21SBoris Burkov ret = btrfs_create_uuid_tree(fs_info); 321344c0ca21SBoris Burkov if (ret) { 321444c0ca21SBoris Burkov btrfs_warn(fs_info, 321544c0ca21SBoris Burkov "failed to create the UUID tree %d", ret); 321644c0ca21SBoris Burkov goto out; 321744c0ca21SBoris Burkov } 321844c0ca21SBoris Burkov } 321944c0ca21SBoris Burkov 322044c0ca21SBoris Burkov out: 322144c0ca21SBoris Burkov return ret; 322244c0ca21SBoris Burkov } 322344c0ca21SBoris Burkov 3224d7f67ac9SQu Wenruo /* 3225d7f67ac9SQu Wenruo * Do various sanity and dependency checks of different features. 3226d7f67ac9SQu Wenruo * 32272ba48b20SQu Wenruo * @is_rw_mount: If the mount is read-write. 32282ba48b20SQu Wenruo * 3229d7f67ac9SQu Wenruo * This is the place for less strict checks (like for subpage or artificial 3230d7f67ac9SQu Wenruo * feature dependencies). 3231d7f67ac9SQu Wenruo * 3232d7f67ac9SQu Wenruo * For strict checks or possible corruption detection, see 3233d7f67ac9SQu Wenruo * btrfs_validate_super(). 3234d7f67ac9SQu Wenruo * 3235d7f67ac9SQu Wenruo * This should be called after btrfs_parse_options(), as some mount options 3236d7f67ac9SQu Wenruo * (space cache related) can modify on-disk format like free space tree and 3237d7f67ac9SQu Wenruo * screw up certain feature dependencies. 3238d7f67ac9SQu Wenruo */ 32392ba48b20SQu Wenruo int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount) 3240d7f67ac9SQu Wenruo { 3241d7f67ac9SQu Wenruo struct btrfs_super_block *disk_super = fs_info->super_copy; 3242d7f67ac9SQu Wenruo u64 incompat = btrfs_super_incompat_flags(disk_super); 3243d7f67ac9SQu Wenruo const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super); 3244d7f67ac9SQu Wenruo const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP); 3245d7f67ac9SQu Wenruo 3246d7f67ac9SQu Wenruo if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 3247d7f67ac9SQu Wenruo btrfs_err(fs_info, 3248d7f67ac9SQu Wenruo "cannot mount because of unknown incompat features (0x%llx)", 3249d7f67ac9SQu Wenruo incompat); 3250d7f67ac9SQu Wenruo return -EINVAL; 3251d7f67ac9SQu Wenruo } 3252d7f67ac9SQu Wenruo 3253d7f67ac9SQu Wenruo /* Runtime limitation for mixed block groups. */ 3254d7f67ac9SQu Wenruo if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 3255d7f67ac9SQu Wenruo (fs_info->sectorsize != fs_info->nodesize)) { 3256d7f67ac9SQu Wenruo btrfs_err(fs_info, 3257d7f67ac9SQu Wenruo "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", 3258d7f67ac9SQu Wenruo fs_info->nodesize, fs_info->sectorsize); 3259d7f67ac9SQu Wenruo return -EINVAL; 3260d7f67ac9SQu Wenruo } 3261d7f67ac9SQu Wenruo 3262d7f67ac9SQu Wenruo /* Mixed backref is an always-enabled feature. */ 3263d7f67ac9SQu Wenruo incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 3264d7f67ac9SQu Wenruo 3265d7f67ac9SQu Wenruo /* Set compression related flags just in case. */ 3266d7f67ac9SQu Wenruo if (fs_info->compress_type == BTRFS_COMPRESS_LZO) 3267d7f67ac9SQu Wenruo incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 3268d7f67ac9SQu Wenruo else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) 3269d7f67ac9SQu Wenruo incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; 3270d7f67ac9SQu Wenruo 3271d7f67ac9SQu Wenruo /* 3272d7f67ac9SQu Wenruo * An ancient flag, which should really be marked deprecated. 3273d7f67ac9SQu Wenruo * Such runtime limitation doesn't really need a incompat flag. 3274d7f67ac9SQu Wenruo */ 3275d7f67ac9SQu Wenruo if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) 3276d7f67ac9SQu Wenruo incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 3277d7f67ac9SQu Wenruo 32782ba48b20SQu Wenruo if (compat_ro_unsupp && is_rw_mount) { 3279d7f67ac9SQu Wenruo btrfs_err(fs_info, 3280d7f67ac9SQu Wenruo "cannot mount read-write because of unknown compat_ro features (0x%llx)", 3281d7f67ac9SQu Wenruo compat_ro); 3282d7f67ac9SQu Wenruo return -EINVAL; 3283d7f67ac9SQu Wenruo } 3284d7f67ac9SQu Wenruo 3285d7f67ac9SQu Wenruo /* 3286d7f67ac9SQu Wenruo * We have unsupported RO compat features, although RO mounted, we 3287d7f67ac9SQu Wenruo * should not cause any metadata writes, including log replay. 3288d7f67ac9SQu Wenruo * Or we could screw up whatever the new feature requires. 3289d7f67ac9SQu Wenruo */ 3290d7f67ac9SQu Wenruo if (compat_ro_unsupp && btrfs_super_log_root(disk_super) && 3291d7f67ac9SQu Wenruo !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 3292d7f67ac9SQu Wenruo btrfs_err(fs_info, 3293d7f67ac9SQu Wenruo "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay", 3294d7f67ac9SQu Wenruo compat_ro); 3295d7f67ac9SQu Wenruo return -EINVAL; 3296d7f67ac9SQu Wenruo } 3297d7f67ac9SQu Wenruo 3298d7f67ac9SQu Wenruo /* 3299d7f67ac9SQu Wenruo * Artificial limitations for block group tree, to force 3300d7f67ac9SQu Wenruo * block-group-tree to rely on no-holes and free-space-tree. 3301d7f67ac9SQu Wenruo */ 3302d7f67ac9SQu Wenruo if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) && 3303d7f67ac9SQu Wenruo (!btrfs_fs_incompat(fs_info, NO_HOLES) || 3304d7f67ac9SQu Wenruo !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) { 3305d7f67ac9SQu Wenruo btrfs_err(fs_info, 3306d7f67ac9SQu Wenruo "block-group-tree feature requires no-holes and free-space-tree features"); 3307d7f67ac9SQu Wenruo return -EINVAL; 3308d7f67ac9SQu Wenruo } 3309d7f67ac9SQu Wenruo 3310d7f67ac9SQu Wenruo /* 3311d7f67ac9SQu Wenruo * Subpage runtime limitation on v1 cache. 3312d7f67ac9SQu Wenruo * 3313d7f67ac9SQu Wenruo * V1 space cache still has some hard codeed PAGE_SIZE usage, while 3314d7f67ac9SQu Wenruo * we're already defaulting to v2 cache, no need to bother v1 as it's 3315d7f67ac9SQu Wenruo * going to be deprecated anyway. 3316d7f67ac9SQu Wenruo */ 3317d7f67ac9SQu Wenruo if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) { 3318d7f67ac9SQu Wenruo btrfs_warn(fs_info, 3319d7f67ac9SQu Wenruo "v1 space cache is not supported for page size %lu with sectorsize %u", 3320d7f67ac9SQu Wenruo PAGE_SIZE, fs_info->sectorsize); 3321d7f67ac9SQu Wenruo return -EINVAL; 3322d7f67ac9SQu Wenruo } 3323d7f67ac9SQu Wenruo 3324d7f67ac9SQu Wenruo /* This can be called by remount, we need to protect the super block. */ 3325d7f67ac9SQu Wenruo spin_lock(&fs_info->super_lock); 3326d7f67ac9SQu Wenruo btrfs_set_super_incompat_flags(disk_super, incompat); 3327d7f67ac9SQu Wenruo spin_unlock(&fs_info->super_lock); 3328d7f67ac9SQu Wenruo 3329d7f67ac9SQu Wenruo return 0; 3330d7f67ac9SQu Wenruo } 3331d7f67ac9SQu Wenruo 3332ae18c37aSJosef Bacik int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices, 3333ae18c37aSJosef Bacik char *options) 3334ae18c37aSJosef Bacik { 3335ae18c37aSJosef Bacik u32 sectorsize; 3336ae18c37aSJosef Bacik u32 nodesize; 3337ae18c37aSJosef Bacik u32 stripesize; 3338ae18c37aSJosef Bacik u64 generation; 3339ae18c37aSJosef Bacik u64 features; 3340ae18c37aSJosef Bacik u16 csum_type; 3341ae18c37aSJosef Bacik struct btrfs_super_block *disk_super; 3342ae18c37aSJosef Bacik struct btrfs_fs_info *fs_info = btrfs_sb(sb); 3343ae18c37aSJosef Bacik struct btrfs_root *tree_root; 3344ae18c37aSJosef Bacik struct btrfs_root *chunk_root; 3345ae18c37aSJosef Bacik int ret; 3346ae18c37aSJosef Bacik int level; 3347ae18c37aSJosef Bacik 33488260edbaSJosef Bacik ret = init_mount_fs_info(fs_info, sb); 33494871c33bSQu Wenruo if (ret) 3350ae18c37aSJosef Bacik goto fail; 335153b381b3SDavid Woodhouse 3352ae18c37aSJosef Bacik /* These need to be init'ed before we start creating inodes and such. */ 3353ae18c37aSJosef Bacik tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, 3354ae18c37aSJosef Bacik GFP_KERNEL); 3355ae18c37aSJosef Bacik fs_info->tree_root = tree_root; 3356ae18c37aSJosef Bacik chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID, 3357ae18c37aSJosef Bacik GFP_KERNEL); 3358ae18c37aSJosef Bacik fs_info->chunk_root = chunk_root; 3359ae18c37aSJosef Bacik if (!tree_root || !chunk_root) { 33604871c33bSQu Wenruo ret = -ENOMEM; 3361c75e8394SJosef Bacik goto fail; 3362ae18c37aSJosef Bacik } 3363ae18c37aSJosef Bacik 3364dcb2137cSChristoph Hellwig ret = btrfs_init_btree_inode(sb); 33654871c33bSQu Wenruo if (ret) 3366c75e8394SJosef Bacik goto fail; 3367ae18c37aSJosef Bacik 3368d24fa5c1SAnand Jain invalidate_bdev(fs_devices->latest_dev->bdev); 33691104a885SDavid Sterba 33701104a885SDavid Sterba /* 33711104a885SDavid Sterba * Read super block and check the signature bytes only 33721104a885SDavid Sterba */ 3373d24fa5c1SAnand Jain disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev); 33748f32380dSJohannes Thumshirn if (IS_ERR(disk_super)) { 33754871c33bSQu Wenruo ret = PTR_ERR(disk_super); 337616cdcec7SMiao Xie goto fail_alloc; 337720b45077SDave Young } 337839279cc3SChris Mason 33791104a885SDavid Sterba /* 3380260db43cSRandy Dunlap * Verify the type first, if that or the checksum value are 33818dc3f22cSJohannes Thumshirn * corrupted, we'll find out 33828dc3f22cSJohannes Thumshirn */ 33838f32380dSJohannes Thumshirn csum_type = btrfs_super_csum_type(disk_super); 338451bce6c9SJohannes Thumshirn if (!btrfs_supported_super_csum(csum_type)) { 33858dc3f22cSJohannes Thumshirn btrfs_err(fs_info, "unsupported checksum algorithm: %u", 338651bce6c9SJohannes Thumshirn csum_type); 33874871c33bSQu Wenruo ret = -EINVAL; 33888f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 33898dc3f22cSJohannes Thumshirn goto fail_alloc; 33908dc3f22cSJohannes Thumshirn } 33918dc3f22cSJohannes Thumshirn 339283c68bbcSSu Yue fs_info->csum_size = btrfs_super_csum_size(disk_super); 339383c68bbcSSu Yue 33946d97c6e3SJohannes Thumshirn ret = btrfs_init_csum_hash(fs_info, csum_type); 33956d97c6e3SJohannes Thumshirn if (ret) { 33968f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 33976d97c6e3SJohannes Thumshirn goto fail_alloc; 33986d97c6e3SJohannes Thumshirn } 33996d97c6e3SJohannes Thumshirn 34008dc3f22cSJohannes Thumshirn /* 34011104a885SDavid Sterba * We want to check superblock checksum, the type is stored inside. 34021104a885SDavid Sterba * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 34031104a885SDavid Sterba */ 34043d17adeaSQu Wenruo if (btrfs_check_super_csum(fs_info, disk_super)) { 340505135f59SDavid Sterba btrfs_err(fs_info, "superblock checksum mismatch"); 34064871c33bSQu Wenruo ret = -EINVAL; 34078f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 3408141386e1SJosef Bacik goto fail_alloc; 34091104a885SDavid Sterba } 34101104a885SDavid Sterba 34111104a885SDavid Sterba /* 34121104a885SDavid Sterba * super_copy is zeroed at allocation time and we never touch the 34131104a885SDavid Sterba * following bytes up to INFO_SIZE, the checksum is calculated from 34141104a885SDavid Sterba * the whole block of INFO_SIZE 34151104a885SDavid Sterba */ 34168f32380dSJohannes Thumshirn memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy)); 34178f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 34185f39d397SChris Mason 3419fbc6feaeSNikolay Borisov disk_super = fs_info->super_copy; 3420fbc6feaeSNikolay Borisov 34210b86a832SChris Mason 3422fbc6feaeSNikolay Borisov features = btrfs_super_flags(disk_super); 3423fbc6feaeSNikolay Borisov if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { 3424fbc6feaeSNikolay Borisov features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; 3425fbc6feaeSNikolay Borisov btrfs_set_super_flags(disk_super, features); 3426fbc6feaeSNikolay Borisov btrfs_info(fs_info, 3427fbc6feaeSNikolay Borisov "found metadata UUID change in progress flag, clearing"); 3428fbc6feaeSNikolay Borisov } 3429fbc6feaeSNikolay Borisov 3430fbc6feaeSNikolay Borisov memcpy(fs_info->super_for_commit, fs_info->super_copy, 3431fbc6feaeSNikolay Borisov sizeof(*fs_info->super_for_commit)); 3432de37aa51SNikolay Borisov 3433069ec957SQu Wenruo ret = btrfs_validate_mount_super(fs_info); 34341104a885SDavid Sterba if (ret) { 343505135f59SDavid Sterba btrfs_err(fs_info, "superblock contains fatal errors"); 34364871c33bSQu Wenruo ret = -EINVAL; 3437141386e1SJosef Bacik goto fail_alloc; 34381104a885SDavid Sterba } 34391104a885SDavid Sterba 34404871c33bSQu Wenruo if (!btrfs_super_root(disk_super)) { 34414871c33bSQu Wenruo btrfs_err(fs_info, "invalid superblock tree root bytenr"); 34424871c33bSQu Wenruo ret = -EINVAL; 3443141386e1SJosef Bacik goto fail_alloc; 34444871c33bSQu Wenruo } 34450f7d52f4SChris Mason 3446acce952bSliubo /* check FS state, whether FS is broken. */ 344787533c47SMiao Xie if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 344887533c47SMiao Xie set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 3449acce952bSliubo 345075e7cb7fSLiu Bo /* 345175e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 345275e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 345375e7cb7fSLiu Bo */ 345475e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 345575e7cb7fSLiu Bo 34566f93e834SAnand Jain 34576f93e834SAnand Jain /* Set up fs_info before parsing mount options */ 34586f93e834SAnand Jain nodesize = btrfs_super_nodesize(disk_super); 34596f93e834SAnand Jain sectorsize = btrfs_super_sectorsize(disk_super); 34606f93e834SAnand Jain stripesize = sectorsize; 34616f93e834SAnand Jain fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 34626f93e834SAnand Jain fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 34636f93e834SAnand Jain 34646f93e834SAnand Jain fs_info->nodesize = nodesize; 34656f93e834SAnand Jain fs_info->sectorsize = sectorsize; 34666f93e834SAnand Jain fs_info->sectorsize_bits = ilog2(sectorsize); 34676f93e834SAnand Jain fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size; 34686f93e834SAnand Jain fs_info->stripesize = stripesize; 34696f93e834SAnand Jain 34702ff7e61eSJeff Mahoney ret = btrfs_parse_options(fs_info, options, sb->s_flags); 34714871c33bSQu Wenruo if (ret) 3472141386e1SJosef Bacik goto fail_alloc; 3473dfe25020SChris Mason 34742ba48b20SQu Wenruo ret = btrfs_check_features(fs_info, !sb_rdonly(sb)); 34754871c33bSQu Wenruo if (ret < 0) 3476141386e1SJosef Bacik goto fail_alloc; 3477f2b636e8SJosef Bacik 34788481dd80SQu Wenruo if (sectorsize < PAGE_SIZE) { 34798481dd80SQu Wenruo struct btrfs_subpage_info *subpage_info; 34808481dd80SQu Wenruo 34819f73f1aeSQu Wenruo /* 34829f73f1aeSQu Wenruo * V1 space cache has some hardcoded PAGE_SIZE usage, and is 34839f73f1aeSQu Wenruo * going to be deprecated. 34849f73f1aeSQu Wenruo * 34859f73f1aeSQu Wenruo * Force to use v2 cache for subpage case. 34869f73f1aeSQu Wenruo */ 34879f73f1aeSQu Wenruo btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE); 34889f73f1aeSQu Wenruo btrfs_set_and_info(fs_info, FREE_SPACE_TREE, 34899f73f1aeSQu Wenruo "forcing free space tree for sector size %u with page size %lu", 34909f73f1aeSQu Wenruo sectorsize, PAGE_SIZE); 34919f73f1aeSQu Wenruo 349295ea0486SQu Wenruo btrfs_warn(fs_info, 349395ea0486SQu Wenruo "read-write for sector size %u with page size %lu is experimental", 34940bb3eb3eSQu Wenruo sectorsize, PAGE_SIZE); 34958481dd80SQu Wenruo subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL); 34964871c33bSQu Wenruo if (!subpage_info) { 34974871c33bSQu Wenruo ret = -ENOMEM; 34988481dd80SQu Wenruo goto fail_alloc; 34994871c33bSQu Wenruo } 35008481dd80SQu Wenruo btrfs_init_subpage_info(subpage_info, sectorsize); 35018481dd80SQu Wenruo fs_info->subpage_info = subpage_info; 3502c8050b3bSQu Wenruo } 35030bb3eb3eSQu Wenruo 3504d21deec5SSu Yue ret = btrfs_init_workqueues(fs_info); 35054871c33bSQu Wenruo if (ret) 35060dc3b84aSJosef Bacik goto fail_sb_buffer; 35074543df7eSChris Mason 35089e11ceeeSJan Kara sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); 35099e11ceeeSJan Kara sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); 35104575c9ccSChris Mason 3511a061fc8dSChris Mason sb->s_blocksize = sectorsize; 3512a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 3513de37aa51SNikolay Borisov memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); 3514db94535dSChris Mason 3515925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 35166bccf3abSJeff Mahoney ret = btrfs_read_sys_array(fs_info); 3517925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 351884eed90fSChris Mason if (ret) { 351905135f59SDavid Sterba btrfs_err(fs_info, "failed to read the system array: %d", ret); 35205d4f98a2SYan Zheng goto fail_sb_buffer; 352184eed90fSChris Mason } 35220b86a832SChris Mason 352384234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 3524581c1760SQu Wenruo level = btrfs_super_chunk_root_level(disk_super); 3525bd676446SJosef Bacik ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super), 3526bd676446SJosef Bacik generation, level); 3527bd676446SJosef Bacik if (ret) { 352805135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk root"); 3529af31f5e5SChris Mason goto fail_tree_roots; 353083121942SDavid Woodhouse } 35310b86a832SChris Mason 3532e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 3533c4ac7541SDavid Sterba offsetof(struct btrfs_header, chunk_tree_uuid), 3534c4ac7541SDavid Sterba BTRFS_UUID_SIZE); 3535e17cade2SChris Mason 35365b4aacefSJeff Mahoney ret = btrfs_read_chunk_tree(fs_info); 35372b82032cSYan Zheng if (ret) { 353805135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk tree: %d", ret); 3539af31f5e5SChris Mason goto fail_tree_roots; 35402b82032cSYan Zheng } 35410b86a832SChris Mason 35428dabb742SStefan Behrens /* 3543bacce86aSAnand Jain * At this point we know all the devices that make this filesystem, 3544bacce86aSAnand Jain * including the seed devices but we don't know yet if the replace 3545bacce86aSAnand Jain * target is required. So free devices that are not part of this 35461a9fd417SDavid Sterba * filesystem but skip the replace target device which is checked 3547bacce86aSAnand Jain * below in btrfs_init_dev_replace(). 35488dabb742SStefan Behrens */ 3549bacce86aSAnand Jain btrfs_free_extra_devids(fs_devices); 3550d24fa5c1SAnand Jain if (!fs_devices->latest_dev->bdev) { 355105135f59SDavid Sterba btrfs_err(fs_info, "failed to read devices"); 35524871c33bSQu Wenruo ret = -EIO; 3553a6b0d5c8SChris Mason goto fail_tree_roots; 3554a6b0d5c8SChris Mason } 3555a6b0d5c8SChris Mason 3556b8522a1eSNikolay Borisov ret = init_tree_roots(fs_info); 35574bbcaa64SEric Sandeen if (ret) 3558b8522a1eSNikolay Borisov goto fail_tree_roots; 35598929ecfaSYan, Zheng 356075ec1db8SJosef Bacik /* 356173651042SNaohiro Aota * Get zone type information of zoned block devices. This will also 356273651042SNaohiro Aota * handle emulation of a zoned filesystem if a regular device has the 356373651042SNaohiro Aota * zoned incompat feature flag set. 356473651042SNaohiro Aota */ 356573651042SNaohiro Aota ret = btrfs_get_dev_zone_info_all_devices(fs_info); 356673651042SNaohiro Aota if (ret) { 356773651042SNaohiro Aota btrfs_err(fs_info, 35684871c33bSQu Wenruo "zoned: failed to read device zone info: %d", ret); 356973651042SNaohiro Aota goto fail_block_groups; 357073651042SNaohiro Aota } 357173651042SNaohiro Aota 357273651042SNaohiro Aota /* 357375ec1db8SJosef Bacik * If we have a uuid root and we're not being told to rescan we need to 357475ec1db8SJosef Bacik * check the generation here so we can set the 357575ec1db8SJosef Bacik * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the 357675ec1db8SJosef Bacik * transaction during a balance or the log replay without updating the 357775ec1db8SJosef Bacik * uuid generation, and then if we crash we would rescan the uuid tree, 357875ec1db8SJosef Bacik * even though it was perfectly fine. 357975ec1db8SJosef Bacik */ 358075ec1db8SJosef Bacik if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) && 358175ec1db8SJosef Bacik fs_info->generation == btrfs_super_uuid_tree_generation(disk_super)) 358275ec1db8SJosef Bacik set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 358375ec1db8SJosef Bacik 3584cf90d884SQu Wenruo ret = btrfs_verify_dev_extents(fs_info); 3585cf90d884SQu Wenruo if (ret) { 3586cf90d884SQu Wenruo btrfs_err(fs_info, 3587cf90d884SQu Wenruo "failed to verify dev extents against chunks: %d", 3588cf90d884SQu Wenruo ret); 3589cf90d884SQu Wenruo goto fail_block_groups; 3590cf90d884SQu Wenruo } 359168310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 359268310a5eSIlya Dryomov if (ret) { 359305135f59SDavid Sterba btrfs_err(fs_info, "failed to recover balance: %d", ret); 359468310a5eSIlya Dryomov goto fail_block_groups; 359568310a5eSIlya Dryomov } 359668310a5eSIlya Dryomov 3597733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 3598733f4fbbSStefan Behrens if (ret) { 359905135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_stats: %d", ret); 3600733f4fbbSStefan Behrens goto fail_block_groups; 3601733f4fbbSStefan Behrens } 3602733f4fbbSStefan Behrens 36038dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 36048dabb742SStefan Behrens if (ret) { 360505135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_replace: %d", ret); 36068dabb742SStefan Behrens goto fail_block_groups; 36078dabb742SStefan Behrens } 36088dabb742SStefan Behrens 3609b70f5097SNaohiro Aota ret = btrfs_check_zoned_mode(fs_info); 3610b70f5097SNaohiro Aota if (ret) { 3611b70f5097SNaohiro Aota btrfs_err(fs_info, "failed to initialize zoned mode: %d", 3612b70f5097SNaohiro Aota ret); 3613b70f5097SNaohiro Aota goto fail_block_groups; 3614b70f5097SNaohiro Aota } 3615b70f5097SNaohiro Aota 3616c6761a9eSAnand Jain ret = btrfs_sysfs_add_fsid(fs_devices); 3617b7c35e81SAnand Jain if (ret) { 361805135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", 361905135f59SDavid Sterba ret); 3620b7c35e81SAnand Jain goto fail_block_groups; 3621b7c35e81SAnand Jain } 3622b7c35e81SAnand Jain 362396f3136eSAnand Jain ret = btrfs_sysfs_add_mounted(fs_info); 36245ac1d209SJeff Mahoney if (ret) { 362505135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); 3626b7c35e81SAnand Jain goto fail_fsdev_sysfs; 36275ac1d209SJeff Mahoney } 36285ac1d209SJeff Mahoney 3629c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 3630c59021f8Sliubo if (ret) { 363105135f59SDavid Sterba btrfs_err(fs_info, "failed to initialize space info: %d", ret); 36322365dd3cSAnand Jain goto fail_sysfs; 3633c59021f8Sliubo } 3634c59021f8Sliubo 36355b4aacefSJeff Mahoney ret = btrfs_read_block_groups(fs_info); 36361b1d1f66SJosef Bacik if (ret) { 363705135f59SDavid Sterba btrfs_err(fs_info, "failed to read block groups: %d", ret); 36382365dd3cSAnand Jain goto fail_sysfs; 36391b1d1f66SJosef Bacik } 36404330e183SQu Wenruo 364116beac87SNaohiro Aota btrfs_free_zone_cache(fs_info); 364216beac87SNaohiro Aota 36435c78a5e7SAnand Jain if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices && 36445c78a5e7SAnand Jain !btrfs_check_rw_degradable(fs_info, NULL)) { 364505135f59SDavid Sterba btrfs_warn(fs_info, 364652042d8eSAndrea Gelmini "writable mount is not allowed due to too many missing devices"); 36474871c33bSQu Wenruo ret = -EINVAL; 36482365dd3cSAnand Jain goto fail_sysfs; 3649292fd7fcSStefan Behrens } 36509078a3e1SChris Mason 365133c44184SJosef Bacik fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info, 3652a74a4b97SChris Mason "btrfs-cleaner"); 36534871c33bSQu Wenruo if (IS_ERR(fs_info->cleaner_kthread)) { 36544871c33bSQu Wenruo ret = PTR_ERR(fs_info->cleaner_kthread); 36552365dd3cSAnand Jain goto fail_sysfs; 36564871c33bSQu Wenruo } 3657a74a4b97SChris Mason 3658a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 3659a74a4b97SChris Mason tree_root, 3660a74a4b97SChris Mason "btrfs-transaction"); 36614871c33bSQu Wenruo if (IS_ERR(fs_info->transaction_kthread)) { 36624871c33bSQu Wenruo ret = PTR_ERR(fs_info->transaction_kthread); 36633f157a2fSChris Mason goto fail_cleaner; 36644871c33bSQu Wenruo } 3665a74a4b97SChris Mason 3666583b7231SHans van Kranenburg if (!btrfs_test_opt(fs_info, NOSSD) && 3667c289811cSChris Mason !fs_info->fs_devices->rotating) { 3668583b7231SHans van Kranenburg btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); 3669c289811cSChris Mason } 3670c289811cSChris Mason 3671572d9ab7SDavid Sterba /* 367263a7cb13SDavid Sterba * For devices supporting discard turn on discard=async automatically, 367363a7cb13SDavid Sterba * unless it's already set or disabled. This could be turned off by 367463a7cb13SDavid Sterba * nodiscard for the same mount. 367563a7cb13SDavid Sterba */ 367663a7cb13SDavid Sterba if (!(btrfs_test_opt(fs_info, DISCARD_SYNC) || 367763a7cb13SDavid Sterba btrfs_test_opt(fs_info, DISCARD_ASYNC) || 367863a7cb13SDavid Sterba btrfs_test_opt(fs_info, NODISCARD)) && 367963a7cb13SDavid Sterba fs_info->fs_devices->discardable) { 368063a7cb13SDavid Sterba btrfs_set_and_info(fs_info, DISCARD_ASYNC, 368163a7cb13SDavid Sterba "auto enabling async discard"); 368263a7cb13SDavid Sterba } 368363a7cb13SDavid Sterba 368421adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 36850b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { 36862ff7e61eSJeff Mahoney ret = btrfsic_mount(fs_info, fs_devices, 36870b246afaSJeff Mahoney btrfs_test_opt(fs_info, 3688cbeaae4fSDavid Sterba CHECK_INTEGRITY_DATA) ? 1 : 0, 368921adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 369021adbd5cSStefan Behrens if (ret) 369105135f59SDavid Sterba btrfs_warn(fs_info, 369205135f59SDavid Sterba "failed to initialize integrity check module: %d", 369305135f59SDavid Sterba ret); 369421adbd5cSStefan Behrens } 369521adbd5cSStefan Behrens #endif 3696bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 3697bcef60f2SArne Jansen if (ret) 3698bcef60f2SArne Jansen goto fail_trans_kthread; 369921adbd5cSStefan Behrens 3700fd708b81SJosef Bacik if (btrfs_build_ref_tree(fs_info)) 3701fd708b81SJosef Bacik btrfs_err(fs_info, "couldn't build ref tree"); 3702fd708b81SJosef Bacik 370396da0919SQu Wenruo /* do not make disk changes in broken FS or nologreplay is given */ 370496da0919SQu Wenruo if (btrfs_super_log_root(disk_super) != 0 && 37050b246afaSJeff Mahoney !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 3706e8294f2fSDavid Sterba btrfs_info(fs_info, "start tree-log replay"); 370763443bf5SEric Sandeen ret = btrfs_replay_log(fs_info, fs_devices); 37084871c33bSQu Wenruo if (ret) 370928c16cbbSWang Shilong goto fail_qgroup; 3710e556ce2cSYan Zheng } 37111a40e23bSZheng Yan 371256e9357aSDavid Sterba fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true); 37133140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 37144871c33bSQu Wenruo ret = PTR_ERR(fs_info->fs_root); 37154871c33bSQu Wenruo btrfs_warn(fs_info, "failed to read fs tree: %d", ret); 3716315bf8efSJosef Bacik fs_info->fs_root = NULL; 3717bcef60f2SArne Jansen goto fail_qgroup; 37183140c9a3SDan Carpenter } 3719c289811cSChris Mason 3720bc98a42cSDavid Howells if (sb_rdonly(sb)) 37218cd29088SBoris Burkov goto clear_oneshot; 37222b6ba629SIlya Dryomov 372344c0ca21SBoris Burkov ret = btrfs_start_pre_rw_mount(fs_info); 37242b6ba629SIlya Dryomov if (ret) { 37256bccf3abSJeff Mahoney close_ctree(fs_info); 37262b6ba629SIlya Dryomov return ret; 3727e3acc2a6SJosef Bacik } 3728b0643e59SDennis Zhou btrfs_discard_resume(fs_info); 3729b382a324SJan Schmidt 373044c0ca21SBoris Burkov if (fs_info->uuid_root && 373144c0ca21SBoris Burkov (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || 373244c0ca21SBoris Burkov fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) { 373305135f59SDavid Sterba btrfs_info(fs_info, "checking UUID tree"); 373470f80175SStefan Behrens ret = btrfs_check_uuid_tree(fs_info); 373570f80175SStefan Behrens if (ret) { 373605135f59SDavid Sterba btrfs_warn(fs_info, 373705135f59SDavid Sterba "failed to check the UUID tree: %d", ret); 37386bccf3abSJeff Mahoney close_ctree(fs_info); 373970f80175SStefan Behrens return ret; 374070f80175SStefan Behrens } 3741f7a81ea4SStefan Behrens } 374294846229SBoris Burkov 3743afcdd129SJosef Bacik set_bit(BTRFS_FS_OPEN, &fs_info->flags); 374447ab2a6cSJosef Bacik 3745b4be6aefSJosef Bacik /* Kick the cleaner thread so it'll start deleting snapshots. */ 3746b4be6aefSJosef Bacik if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags)) 3747b4be6aefSJosef Bacik wake_up_process(fs_info->cleaner_kthread); 3748b4be6aefSJosef Bacik 37498cd29088SBoris Burkov clear_oneshot: 37508cd29088SBoris Burkov btrfs_clear_oneshot_options(fs_info); 3751ad2b2c80SAl Viro return 0; 375239279cc3SChris Mason 3753bcef60f2SArne Jansen fail_qgroup: 3754bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 37557c2ca468SChris Mason fail_trans_kthread: 37567c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 37572ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 3758faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 37593f157a2fSChris Mason fail_cleaner: 3760a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 37617c2ca468SChris Mason 37627c2ca468SChris Mason /* 37637c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 37647c2ca468SChris Mason * kthreads 37657c2ca468SChris Mason */ 37667c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 37677c2ca468SChris Mason 37682365dd3cSAnand Jain fail_sysfs: 37696618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 37702365dd3cSAnand Jain 3771b7c35e81SAnand Jain fail_fsdev_sysfs: 3772b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3773b7c35e81SAnand Jain 37741b1d1f66SJosef Bacik fail_block_groups: 377554067ae9SJosef Bacik btrfs_put_block_group_cache(fs_info); 3776af31f5e5SChris Mason 3777af31f5e5SChris Mason fail_tree_roots: 37789e3aa805SJosef Bacik if (fs_info->data_reloc_root) 37799e3aa805SJosef Bacik btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root); 37804273eaffSAnand Jain free_root_pointers(fs_info, true); 37812b8195bbSMiao Xie invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3782af31f5e5SChris Mason 378339279cc3SChris Mason fail_sb_buffer: 37847abadb64SLiu Bo btrfs_stop_all_workers(fs_info); 37855cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 378616cdcec7SMiao Xie fail_alloc: 3787586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 3788586e46e2SIlya Dryomov 37894543df7eSChris Mason iput(fs_info->btree_inode); 37907e662854SQinghuang Feng fail: 3791586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 37924871c33bSQu Wenruo ASSERT(ret < 0); 37934871c33bSQu Wenruo return ret; 3794eb60ceacSChris Mason } 3795663faf9fSMasami Hiramatsu ALLOW_ERROR_INJECTION(open_ctree, ERRNO); 3796eb60ceacSChris Mason 3797314b6dd0SJohannes Thumshirn static void btrfs_end_super_write(struct bio *bio) 3798f2984462SChris Mason { 3799314b6dd0SJohannes Thumshirn struct btrfs_device *device = bio->bi_private; 3800314b6dd0SJohannes Thumshirn struct bio_vec *bvec; 3801314b6dd0SJohannes Thumshirn struct bvec_iter_all iter_all; 3802314b6dd0SJohannes Thumshirn struct page *page; 3803442a4f63SStefan Behrens 3804314b6dd0SJohannes Thumshirn bio_for_each_segment_all(bvec, bio, iter_all) { 3805314b6dd0SJohannes Thumshirn page = bvec->bv_page; 3806314b6dd0SJohannes Thumshirn 3807314b6dd0SJohannes Thumshirn if (bio->bi_status) { 3808fb456252SJeff Mahoney btrfs_warn_rl_in_rcu(device->fs_info, 3809314b6dd0SJohannes Thumshirn "lost page write due to IO error on %s (%d)", 3810cb3e217bSQu Wenruo btrfs_dev_name(device), 3811314b6dd0SJohannes Thumshirn blk_status_to_errno(bio->bi_status)); 3812314b6dd0SJohannes Thumshirn ClearPageUptodate(page); 3813314b6dd0SJohannes Thumshirn SetPageError(page); 3814314b6dd0SJohannes Thumshirn btrfs_dev_stat_inc_and_print(device, 3815314b6dd0SJohannes Thumshirn BTRFS_DEV_STAT_WRITE_ERRS); 3816314b6dd0SJohannes Thumshirn } else { 3817314b6dd0SJohannes Thumshirn SetPageUptodate(page); 3818f2984462SChris Mason } 3819314b6dd0SJohannes Thumshirn 3820314b6dd0SJohannes Thumshirn put_page(page); 3821314b6dd0SJohannes Thumshirn unlock_page(page); 3822314b6dd0SJohannes Thumshirn } 3823314b6dd0SJohannes Thumshirn 3824314b6dd0SJohannes Thumshirn bio_put(bio); 3825f2984462SChris Mason } 3826f2984462SChris Mason 38278f32380dSJohannes Thumshirn struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev, 3828a05d3c91SQu Wenruo int copy_num, bool drop_cache) 382929c36d72SAnand Jain { 383029c36d72SAnand Jain struct btrfs_super_block *super; 38318f32380dSJohannes Thumshirn struct page *page; 383212659251SNaohiro Aota u64 bytenr, bytenr_orig; 38338f32380dSJohannes Thumshirn struct address_space *mapping = bdev->bd_inode->i_mapping; 383412659251SNaohiro Aota int ret; 383529c36d72SAnand Jain 383612659251SNaohiro Aota bytenr_orig = btrfs_sb_offset(copy_num); 383712659251SNaohiro Aota ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr); 383812659251SNaohiro Aota if (ret == -ENOENT) 383912659251SNaohiro Aota return ERR_PTR(-EINVAL); 384012659251SNaohiro Aota else if (ret) 384112659251SNaohiro Aota return ERR_PTR(ret); 384212659251SNaohiro Aota 3843cda00ebaSChristoph Hellwig if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev)) 38448f32380dSJohannes Thumshirn return ERR_PTR(-EINVAL); 384529c36d72SAnand Jain 3846a05d3c91SQu Wenruo if (drop_cache) { 3847a05d3c91SQu Wenruo /* This should only be called with the primary sb. */ 3848a05d3c91SQu Wenruo ASSERT(copy_num == 0); 3849a05d3c91SQu Wenruo 3850a05d3c91SQu Wenruo /* 3851a05d3c91SQu Wenruo * Drop the page of the primary superblock, so later read will 3852a05d3c91SQu Wenruo * always read from the device. 3853a05d3c91SQu Wenruo */ 3854a05d3c91SQu Wenruo invalidate_inode_pages2_range(mapping, 3855a05d3c91SQu Wenruo bytenr >> PAGE_SHIFT, 3856a05d3c91SQu Wenruo (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT); 3857a05d3c91SQu Wenruo } 3858a05d3c91SQu Wenruo 38598f32380dSJohannes Thumshirn page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS); 38608f32380dSJohannes Thumshirn if (IS_ERR(page)) 38618f32380dSJohannes Thumshirn return ERR_CAST(page); 386229c36d72SAnand Jain 38638f32380dSJohannes Thumshirn super = page_address(page); 386496c2e067SAnand Jain if (btrfs_super_magic(super) != BTRFS_MAGIC) { 386596c2e067SAnand Jain btrfs_release_disk_super(super); 386696c2e067SAnand Jain return ERR_PTR(-ENODATA); 386796c2e067SAnand Jain } 386896c2e067SAnand Jain 386912659251SNaohiro Aota if (btrfs_super_bytenr(super) != bytenr_orig) { 38708f32380dSJohannes Thumshirn btrfs_release_disk_super(super); 38718f32380dSJohannes Thumshirn return ERR_PTR(-EINVAL); 387229c36d72SAnand Jain } 387329c36d72SAnand Jain 38748f32380dSJohannes Thumshirn return super; 387529c36d72SAnand Jain } 387629c36d72SAnand Jain 387729c36d72SAnand Jain 38788f32380dSJohannes Thumshirn struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev) 3879a512bbf8SYan Zheng { 38808f32380dSJohannes Thumshirn struct btrfs_super_block *super, *latest = NULL; 3881a512bbf8SYan Zheng int i; 3882a512bbf8SYan Zheng u64 transid = 0; 3883a512bbf8SYan Zheng 3884a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 3885a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 3886a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 3887a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3888a512bbf8SYan Zheng */ 3889a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 3890a05d3c91SQu Wenruo super = btrfs_read_dev_one_super(bdev, i, false); 38918f32380dSJohannes Thumshirn if (IS_ERR(super)) 3892a512bbf8SYan Zheng continue; 3893a512bbf8SYan Zheng 3894a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 38958f32380dSJohannes Thumshirn if (latest) 38968f32380dSJohannes Thumshirn btrfs_release_disk_super(super); 38978f32380dSJohannes Thumshirn 38988f32380dSJohannes Thumshirn latest = super; 3899a512bbf8SYan Zheng transid = btrfs_super_generation(super); 3900a512bbf8SYan Zheng } 3901a512bbf8SYan Zheng } 390292fc03fbSAnand Jain 39038f32380dSJohannes Thumshirn return super; 3904a512bbf8SYan Zheng } 3905a512bbf8SYan Zheng 39064eedeb75SHisashi Hifumi /* 3907abbb3b8eSDavid Sterba * Write superblock @sb to the @device. Do not wait for completion, all the 3908314b6dd0SJohannes Thumshirn * pages we use for writing are locked. 39094eedeb75SHisashi Hifumi * 3910abbb3b8eSDavid Sterba * Write @max_mirrors copies of the superblock, where 0 means default that fit 3911abbb3b8eSDavid Sterba * the expected device size at commit time. Note that max_mirrors must be 3912abbb3b8eSDavid Sterba * same for write and wait phases. 39134eedeb75SHisashi Hifumi * 3914314b6dd0SJohannes Thumshirn * Return number of errors when page is not found or submission fails. 39154eedeb75SHisashi Hifumi */ 3916a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 3917abbb3b8eSDavid Sterba struct btrfs_super_block *sb, int max_mirrors) 3918a512bbf8SYan Zheng { 3919d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = device->fs_info; 3920314b6dd0SJohannes Thumshirn struct address_space *mapping = device->bdev->bd_inode->i_mapping; 3921d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 3922a512bbf8SYan Zheng int i; 3923a512bbf8SYan Zheng int errors = 0; 392412659251SNaohiro Aota int ret; 392512659251SNaohiro Aota u64 bytenr, bytenr_orig; 3926a512bbf8SYan Zheng 3927a512bbf8SYan Zheng if (max_mirrors == 0) 3928a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3929a512bbf8SYan Zheng 3930d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 3931d5178578SJohannes Thumshirn 3932a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 3933314b6dd0SJohannes Thumshirn struct page *page; 3934314b6dd0SJohannes Thumshirn struct bio *bio; 3935314b6dd0SJohannes Thumshirn struct btrfs_super_block *disk_super; 3936314b6dd0SJohannes Thumshirn 393712659251SNaohiro Aota bytenr_orig = btrfs_sb_offset(i); 393812659251SNaohiro Aota ret = btrfs_sb_log_location(device, i, WRITE, &bytenr); 393912659251SNaohiro Aota if (ret == -ENOENT) { 394012659251SNaohiro Aota continue; 394112659251SNaohiro Aota } else if (ret < 0) { 394212659251SNaohiro Aota btrfs_err(device->fs_info, 394312659251SNaohiro Aota "couldn't get super block location for mirror %d", 394412659251SNaohiro Aota i); 394512659251SNaohiro Aota errors++; 394612659251SNaohiro Aota continue; 394712659251SNaohiro Aota } 3948935e5cc9SMiao Xie if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3949935e5cc9SMiao Xie device->commit_total_bytes) 3950a512bbf8SYan Zheng break; 3951a512bbf8SYan Zheng 395212659251SNaohiro Aota btrfs_set_super_bytenr(sb, bytenr_orig); 3953a512bbf8SYan Zheng 3954fd08001fSEric Biggers crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE, 3955fd08001fSEric Biggers BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, 3956fd08001fSEric Biggers sb->csum); 3957a512bbf8SYan Zheng 3958314b6dd0SJohannes Thumshirn page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT, 3959314b6dd0SJohannes Thumshirn GFP_NOFS); 3960314b6dd0SJohannes Thumshirn if (!page) { 3961fb456252SJeff Mahoney btrfs_err(device->fs_info, 3962314b6dd0SJohannes Thumshirn "couldn't get super block page for bytenr %llu", 3963f14d104dSDavid Sterba bytenr); 3964634554dcSJosef Bacik errors++; 3965634554dcSJosef Bacik continue; 3966634554dcSJosef Bacik } 3967634554dcSJosef Bacik 3968314b6dd0SJohannes Thumshirn /* Bump the refcount for wait_dev_supers() */ 3969314b6dd0SJohannes Thumshirn get_page(page); 3970a512bbf8SYan Zheng 3971314b6dd0SJohannes Thumshirn disk_super = page_address(page); 3972314b6dd0SJohannes Thumshirn memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE); 3973a512bbf8SYan Zheng 3974387125fcSChris Mason /* 3975314b6dd0SJohannes Thumshirn * Directly use bios here instead of relying on the page cache 3976314b6dd0SJohannes Thumshirn * to do I/O, so we don't lose the ability to do integrity 3977314b6dd0SJohannes Thumshirn * checking. 3978387125fcSChris Mason */ 397907888c66SChristoph Hellwig bio = bio_alloc(device->bdev, 1, 398007888c66SChristoph Hellwig REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO, 398107888c66SChristoph Hellwig GFP_NOFS); 3982314b6dd0SJohannes Thumshirn bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT; 3983314b6dd0SJohannes Thumshirn bio->bi_private = device; 3984314b6dd0SJohannes Thumshirn bio->bi_end_io = btrfs_end_super_write; 3985314b6dd0SJohannes Thumshirn __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE, 3986314b6dd0SJohannes Thumshirn offset_in_page(bytenr)); 3987314b6dd0SJohannes Thumshirn 3988314b6dd0SJohannes Thumshirn /* 3989314b6dd0SJohannes Thumshirn * We FUA only the first super block. The others we allow to 3990314b6dd0SJohannes Thumshirn * go down lazy and there's a short window where the on-disk 3991314b6dd0SJohannes Thumshirn * copies might still contain the older version. 3992314b6dd0SJohannes Thumshirn */ 39931b9e619cSOmar Sandoval if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) 3994314b6dd0SJohannes Thumshirn bio->bi_opf |= REQ_FUA; 3995314b6dd0SJohannes Thumshirn 399658ff51f1SChristoph Hellwig btrfsic_check_bio(bio); 399758ff51f1SChristoph Hellwig submit_bio(bio); 39988376d9e1SNaohiro Aota 39998376d9e1SNaohiro Aota if (btrfs_advance_sb_log(device, i)) 40008376d9e1SNaohiro Aota errors++; 4001a512bbf8SYan Zheng } 4002a512bbf8SYan Zheng return errors < i ? 0 : -1; 4003a512bbf8SYan Zheng } 4004a512bbf8SYan Zheng 4005387125fcSChris Mason /* 4006abbb3b8eSDavid Sterba * Wait for write completion of superblocks done by write_dev_supers, 4007abbb3b8eSDavid Sterba * @max_mirrors same for write and wait phases. 4008abbb3b8eSDavid Sterba * 4009314b6dd0SJohannes Thumshirn * Return number of errors when page is not found or not marked up to 4010abbb3b8eSDavid Sterba * date. 4011abbb3b8eSDavid Sterba */ 4012abbb3b8eSDavid Sterba static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) 4013abbb3b8eSDavid Sterba { 4014abbb3b8eSDavid Sterba int i; 4015abbb3b8eSDavid Sterba int errors = 0; 4016b6a535faSHoward McLauchlan bool primary_failed = false; 401712659251SNaohiro Aota int ret; 4018abbb3b8eSDavid Sterba u64 bytenr; 4019abbb3b8eSDavid Sterba 4020abbb3b8eSDavid Sterba if (max_mirrors == 0) 4021abbb3b8eSDavid Sterba max_mirrors = BTRFS_SUPER_MIRROR_MAX; 4022abbb3b8eSDavid Sterba 4023abbb3b8eSDavid Sterba for (i = 0; i < max_mirrors; i++) { 4024314b6dd0SJohannes Thumshirn struct page *page; 4025314b6dd0SJohannes Thumshirn 402612659251SNaohiro Aota ret = btrfs_sb_log_location(device, i, READ, &bytenr); 402712659251SNaohiro Aota if (ret == -ENOENT) { 402812659251SNaohiro Aota break; 402912659251SNaohiro Aota } else if (ret < 0) { 403012659251SNaohiro Aota errors++; 403112659251SNaohiro Aota if (i == 0) 403212659251SNaohiro Aota primary_failed = true; 403312659251SNaohiro Aota continue; 403412659251SNaohiro Aota } 4035abbb3b8eSDavid Sterba if (bytenr + BTRFS_SUPER_INFO_SIZE >= 4036abbb3b8eSDavid Sterba device->commit_total_bytes) 4037abbb3b8eSDavid Sterba break; 4038abbb3b8eSDavid Sterba 4039314b6dd0SJohannes Thumshirn page = find_get_page(device->bdev->bd_inode->i_mapping, 4040314b6dd0SJohannes Thumshirn bytenr >> PAGE_SHIFT); 4041314b6dd0SJohannes Thumshirn if (!page) { 4042abbb3b8eSDavid Sterba errors++; 4043b6a535faSHoward McLauchlan if (i == 0) 4044b6a535faSHoward McLauchlan primary_failed = true; 4045abbb3b8eSDavid Sterba continue; 4046abbb3b8eSDavid Sterba } 4047314b6dd0SJohannes Thumshirn /* Page is submitted locked and unlocked once the IO completes */ 4048314b6dd0SJohannes Thumshirn wait_on_page_locked(page); 4049314b6dd0SJohannes Thumshirn if (PageError(page)) { 4050abbb3b8eSDavid Sterba errors++; 4051b6a535faSHoward McLauchlan if (i == 0) 4052b6a535faSHoward McLauchlan primary_failed = true; 4053b6a535faSHoward McLauchlan } 4054abbb3b8eSDavid Sterba 4055314b6dd0SJohannes Thumshirn /* Drop our reference */ 4056314b6dd0SJohannes Thumshirn put_page(page); 4057abbb3b8eSDavid Sterba 4058314b6dd0SJohannes Thumshirn /* Drop the reference from the writing run */ 4059314b6dd0SJohannes Thumshirn put_page(page); 4060abbb3b8eSDavid Sterba } 4061abbb3b8eSDavid Sterba 4062b6a535faSHoward McLauchlan /* log error, force error return */ 4063b6a535faSHoward McLauchlan if (primary_failed) { 4064b6a535faSHoward McLauchlan btrfs_err(device->fs_info, "error writing primary super block to device %llu", 4065b6a535faSHoward McLauchlan device->devid); 4066b6a535faSHoward McLauchlan return -1; 4067b6a535faSHoward McLauchlan } 4068b6a535faSHoward McLauchlan 4069abbb3b8eSDavid Sterba return errors < i ? 0 : -1; 4070abbb3b8eSDavid Sterba } 4071abbb3b8eSDavid Sterba 4072abbb3b8eSDavid Sterba /* 4073387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 4074387125fcSChris Mason * for the barrier when it is done 4075387125fcSChris Mason */ 40764246a0b6SChristoph Hellwig static void btrfs_end_empty_barrier(struct bio *bio) 4077387125fcSChris Mason { 4078f9e69aa9SChristoph Hellwig bio_uninit(bio); 4079387125fcSChris Mason complete(bio->bi_private); 4080387125fcSChris Mason } 4081387125fcSChris Mason 4082387125fcSChris Mason /* 40834fc6441aSAnand Jain * Submit a flush request to the device if it supports it. Error handling is 40844fc6441aSAnand Jain * done in the waiting counterpart. 4085387125fcSChris Mason */ 40864fc6441aSAnand Jain static void write_dev_flush(struct btrfs_device *device) 4087387125fcSChris Mason { 4088f9e69aa9SChristoph Hellwig struct bio *bio = &device->flush_bio; 4089387125fcSChris Mason 4090a91cf0ffSWang Yugui #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY 4091a91cf0ffSWang Yugui /* 4092a91cf0ffSWang Yugui * When a disk has write caching disabled, we skip submission of a bio 4093a91cf0ffSWang Yugui * with flush and sync requests before writing the superblock, since 4094a91cf0ffSWang Yugui * it's not needed. However when the integrity checker is enabled, this 4095a91cf0ffSWang Yugui * results in reports that there are metadata blocks referred by a 4096a91cf0ffSWang Yugui * superblock that were not properly flushed. So don't skip the bio 4097a91cf0ffSWang Yugui * submission only when the integrity checker is enabled for the sake 4098a91cf0ffSWang Yugui * of simplicity, since this is a debug tool and not meant for use in 4099a91cf0ffSWang Yugui * non-debug builds. 4100a91cf0ffSWang Yugui */ 410108e688fdSChristoph Hellwig if (!bdev_write_cache(device->bdev)) 41024fc6441aSAnand Jain return; 4103a91cf0ffSWang Yugui #endif 4104387125fcSChris Mason 4105f9e69aa9SChristoph Hellwig bio_init(bio, device->bdev, NULL, 0, 4106f9e69aa9SChristoph Hellwig REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH); 4107387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 4108387125fcSChris Mason init_completion(&device->flush_wait); 4109387125fcSChris Mason bio->bi_private = &device->flush_wait; 4110387125fcSChris Mason 411158ff51f1SChristoph Hellwig btrfsic_check_bio(bio); 411258ff51f1SChristoph Hellwig submit_bio(bio); 41131c3063b6SAnand Jain set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 41144fc6441aSAnand Jain } 4115387125fcSChris Mason 41164fc6441aSAnand Jain /* 41174fc6441aSAnand Jain * If the flush bio has been submitted by write_dev_flush, wait for it. 41184fc6441aSAnand Jain */ 41198c27cb35SLinus Torvalds static blk_status_t wait_dev_flush(struct btrfs_device *device) 41204fc6441aSAnand Jain { 4121f9e69aa9SChristoph Hellwig struct bio *bio = &device->flush_bio; 41224fc6441aSAnand Jain 41231c3063b6SAnand Jain if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) 412458efbc9fSOmar Sandoval return BLK_STS_OK; 41254fc6441aSAnand Jain 41261c3063b6SAnand Jain clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 41272980d574SDavid Sterba wait_for_completion_io(&device->flush_wait); 41284fc6441aSAnand Jain 41298c27cb35SLinus Torvalds return bio->bi_status; 4130387125fcSChris Mason } 4131387125fcSChris Mason 4132d10b82feSQu Wenruo static int check_barrier_error(struct btrfs_fs_info *fs_info) 4133401b41e5SAnand Jain { 41346528b99dSAnand Jain if (!btrfs_check_rw_degradable(fs_info, NULL)) 4135401b41e5SAnand Jain return -EIO; 4136387125fcSChris Mason return 0; 4137387125fcSChris Mason } 4138387125fcSChris Mason 4139387125fcSChris Mason /* 4140387125fcSChris Mason * send an empty flush down to each device in parallel, 4141387125fcSChris Mason * then wait for them 4142387125fcSChris Mason */ 4143387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 4144387125fcSChris Mason { 4145387125fcSChris Mason struct list_head *head; 4146387125fcSChris Mason struct btrfs_device *dev; 41475af3e8ccSStefan Behrens int errors_wait = 0; 41484e4cbee9SChristoph Hellwig blk_status_t ret; 4149387125fcSChris Mason 41501538e6c5SDavid Sterba lockdep_assert_held(&info->fs_devices->device_list_mutex); 4151387125fcSChris Mason /* send down all the barriers */ 4152387125fcSChris Mason head = &info->fs_devices->devices; 41531538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4154e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 4155f88ba6a2SHidetoshi Seto continue; 4156cea7c8bfSAnand Jain if (!dev->bdev) 4157387125fcSChris Mason continue; 4158e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4159ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4160387125fcSChris Mason continue; 4161387125fcSChris Mason 41624fc6441aSAnand Jain write_dev_flush(dev); 416358efbc9fSOmar Sandoval dev->last_flush_error = BLK_STS_OK; 4164387125fcSChris Mason } 4165387125fcSChris Mason 4166387125fcSChris Mason /* wait for all the barriers */ 41671538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4168e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 4169f88ba6a2SHidetoshi Seto continue; 4170387125fcSChris Mason if (!dev->bdev) { 41715af3e8ccSStefan Behrens errors_wait++; 4172387125fcSChris Mason continue; 4173387125fcSChris Mason } 4174e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4175ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4176387125fcSChris Mason continue; 4177387125fcSChris Mason 41784fc6441aSAnand Jain ret = wait_dev_flush(dev); 4179401b41e5SAnand Jain if (ret) { 4180401b41e5SAnand Jain dev->last_flush_error = ret; 418166b4993eSDavid Sterba btrfs_dev_stat_inc_and_print(dev, 418266b4993eSDavid Sterba BTRFS_DEV_STAT_FLUSH_ERRS); 41835af3e8ccSStefan Behrens errors_wait++; 4184387125fcSChris Mason } 4185401b41e5SAnand Jain } 4186401b41e5SAnand Jain 4187cea7c8bfSAnand Jain if (errors_wait) { 4188401b41e5SAnand Jain /* 4189401b41e5SAnand Jain * At some point we need the status of all disks 4190401b41e5SAnand Jain * to arrive at the volume status. So error checking 4191401b41e5SAnand Jain * is being pushed to a separate loop. 4192401b41e5SAnand Jain */ 4193d10b82feSQu Wenruo return check_barrier_error(info); 4194401b41e5SAnand Jain } 4195387125fcSChris Mason return 0; 4196387125fcSChris Mason } 4197387125fcSChris Mason 4198943c6e99SZhao Lei int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 4199943c6e99SZhao Lei { 42008789f4feSZhao Lei int raid_type; 42018789f4feSZhao Lei int min_tolerated = INT_MAX; 4202943c6e99SZhao Lei 42038789f4feSZhao Lei if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 42048789f4feSZhao Lei (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 42058c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 42068789f4feSZhao Lei btrfs_raid_array[BTRFS_RAID_SINGLE]. 42078789f4feSZhao Lei tolerated_failures); 4208943c6e99SZhao Lei 42098789f4feSZhao Lei for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 42108789f4feSZhao Lei if (raid_type == BTRFS_RAID_SINGLE) 42118789f4feSZhao Lei continue; 421241a6e891SAnand Jain if (!(flags & btrfs_raid_array[raid_type].bg_flag)) 42138789f4feSZhao Lei continue; 42148c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 42158789f4feSZhao Lei btrfs_raid_array[raid_type]. 42168789f4feSZhao Lei tolerated_failures); 42178789f4feSZhao Lei } 4218943c6e99SZhao Lei 42198789f4feSZhao Lei if (min_tolerated == INT_MAX) { 4220ab8d0fc4SJeff Mahoney pr_warn("BTRFS: unknown raid flag: %llu", flags); 42218789f4feSZhao Lei min_tolerated = 0; 42228789f4feSZhao Lei } 42238789f4feSZhao Lei 42248789f4feSZhao Lei return min_tolerated; 4225943c6e99SZhao Lei } 4226943c6e99SZhao Lei 4227eece6a9cSDavid Sterba int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) 4228f2984462SChris Mason { 4229e5e9a520SChris Mason struct list_head *head; 4230f2984462SChris Mason struct btrfs_device *dev; 4231a061fc8dSChris Mason struct btrfs_super_block *sb; 4232f2984462SChris Mason struct btrfs_dev_item *dev_item; 4233f2984462SChris Mason int ret; 4234f2984462SChris Mason int do_barriers; 4235a236aed1SChris Mason int max_errors; 4236a236aed1SChris Mason int total_errors = 0; 4237a061fc8dSChris Mason u64 flags; 4238f2984462SChris Mason 42390b246afaSJeff Mahoney do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); 4240fed3b381SLiu Bo 4241fed3b381SLiu Bo /* 4242fed3b381SLiu Bo * max_mirrors == 0 indicates we're from commit_transaction, 4243fed3b381SLiu Bo * not from fsync where the tree roots in fs_info have not 4244fed3b381SLiu Bo * been consistent on disk. 4245fed3b381SLiu Bo */ 4246fed3b381SLiu Bo if (max_mirrors == 0) 42470b246afaSJeff Mahoney backup_super_roots(fs_info); 4248f2984462SChris Mason 42490b246afaSJeff Mahoney sb = fs_info->super_for_commit; 4250a061fc8dSChris Mason dev_item = &sb->dev_item; 4251e5e9a520SChris Mason 42520b246afaSJeff Mahoney mutex_lock(&fs_info->fs_devices->device_list_mutex); 42530b246afaSJeff Mahoney head = &fs_info->fs_devices->devices; 42540b246afaSJeff Mahoney max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; 4255387125fcSChris Mason 42565af3e8ccSStefan Behrens if (do_barriers) { 42570b246afaSJeff Mahoney ret = barrier_all_devices(fs_info); 42585af3e8ccSStefan Behrens if (ret) { 42595af3e8ccSStefan Behrens mutex_unlock( 42600b246afaSJeff Mahoney &fs_info->fs_devices->device_list_mutex); 42610b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 42625af3e8ccSStefan Behrens "errors while submitting device barriers."); 42635af3e8ccSStefan Behrens return ret; 42645af3e8ccSStefan Behrens } 42655af3e8ccSStefan Behrens } 4266387125fcSChris Mason 42671538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4268dfe25020SChris Mason if (!dev->bdev) { 4269dfe25020SChris Mason total_errors++; 4270dfe25020SChris Mason continue; 4271dfe25020SChris Mason } 4272e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4273ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4274dfe25020SChris Mason continue; 4275dfe25020SChris Mason 42762b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 4277a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 4278a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 42797df69d3eSMiao Xie btrfs_set_stack_device_total_bytes(dev_item, 4280935e5cc9SMiao Xie dev->commit_total_bytes); 4281ce7213c7SMiao Xie btrfs_set_stack_device_bytes_used(dev_item, 4282ce7213c7SMiao Xie dev->commit_bytes_used); 4283a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 4284a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 4285a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 4286a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 42877239ff4bSNikolay Borisov memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, 42887239ff4bSNikolay Borisov BTRFS_FSID_SIZE); 4289a512bbf8SYan Zheng 4290a061fc8dSChris Mason flags = btrfs_super_flags(sb); 4291a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 4292f2984462SChris Mason 429375cb857dSQu Wenruo ret = btrfs_validate_write_super(fs_info, sb); 429475cb857dSQu Wenruo if (ret < 0) { 429575cb857dSQu Wenruo mutex_unlock(&fs_info->fs_devices->device_list_mutex); 429675cb857dSQu Wenruo btrfs_handle_fs_error(fs_info, -EUCLEAN, 429775cb857dSQu Wenruo "unexpected superblock corruption detected"); 429875cb857dSQu Wenruo return -EUCLEAN; 429975cb857dSQu Wenruo } 430075cb857dSQu Wenruo 4301abbb3b8eSDavid Sterba ret = write_dev_supers(dev, sb, max_mirrors); 4302a236aed1SChris Mason if (ret) 4303a236aed1SChris Mason total_errors++; 4304f2984462SChris Mason } 4305a236aed1SChris Mason if (total_errors > max_errors) { 43060b246afaSJeff Mahoney btrfs_err(fs_info, "%d errors while writing supers", 4307d397712bSChris Mason total_errors); 43080b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 430979787eaaSJeff Mahoney 43109d565ba4SStefan Behrens /* FUA is masked off if unsupported and can't be the reason */ 43110b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 43120b246afaSJeff Mahoney "%d errors while writing supers", 43130b246afaSJeff Mahoney total_errors); 43149d565ba4SStefan Behrens return -EIO; 4315a236aed1SChris Mason } 4316f2984462SChris Mason 4317a512bbf8SYan Zheng total_errors = 0; 43181538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4319dfe25020SChris Mason if (!dev->bdev) 4320dfe25020SChris Mason continue; 4321e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4322ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4323dfe25020SChris Mason continue; 4324dfe25020SChris Mason 4325abbb3b8eSDavid Sterba ret = wait_dev_supers(dev, max_mirrors); 4326a512bbf8SYan Zheng if (ret) 43271259ab75SChris Mason total_errors++; 4328f2984462SChris Mason } 43290b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 4330a236aed1SChris Mason if (total_errors > max_errors) { 43310b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 43320b246afaSJeff Mahoney "%d errors while writing supers", 43330b246afaSJeff Mahoney total_errors); 433479787eaaSJeff Mahoney return -EIO; 4335a236aed1SChris Mason } 4336f2984462SChris Mason return 0; 4337f2984462SChris Mason } 4338f2984462SChris Mason 4339cb517eabSMiao Xie /* Drop a fs root from the radix tree and free it. */ 4340cb517eabSMiao Xie void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 4341cb517eabSMiao Xie struct btrfs_root *root) 43422619ba1fSChris Mason { 43434785e24fSJosef Bacik bool drop_ref = false; 43444785e24fSJosef Bacik 4345fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4346fc7cbcd4SDavid Sterba radix_tree_delete(&fs_info->fs_roots_radix, 4347fc7cbcd4SDavid Sterba (unsigned long)root->root_key.objectid); 4348fc7cbcd4SDavid Sterba if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state)) 43494785e24fSJosef Bacik drop_ref = true; 4350fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 435176dda93cSYan, Zheng 435284961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 4353ef67963dSJosef Bacik ASSERT(root->log_root == NULL); 43541c1ea4f7SLiu Bo if (root->reloc_root) { 435500246528SJosef Bacik btrfs_put_root(root->reloc_root); 43561c1ea4f7SLiu Bo root->reloc_root = NULL; 43571c1ea4f7SLiu Bo } 43581c1ea4f7SLiu Bo } 43593321719eSLiu Bo 43604785e24fSJosef Bacik if (drop_ref) 436100246528SJosef Bacik btrfs_put_root(root); 43622619ba1fSChris Mason } 43632619ba1fSChris Mason 4364c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 4365c146afadSYan Zheng { 4366fc7cbcd4SDavid Sterba u64 root_objectid = 0; 4367fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 4368fc7cbcd4SDavid Sterba int i = 0; 436965d33fd7SQu Wenruo int err = 0; 4370fc7cbcd4SDavid Sterba unsigned int ret = 0; 4371c146afadSYan Zheng 4372c146afadSYan Zheng while (1) { 4373fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4374fc7cbcd4SDavid Sterba ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4375fc7cbcd4SDavid Sterba (void **)gang, root_objectid, 4376fc7cbcd4SDavid Sterba ARRAY_SIZE(gang)); 4377fc7cbcd4SDavid Sterba if (!ret) { 4378fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4379c146afadSYan Zheng break; 438065d33fd7SQu Wenruo } 4381fc7cbcd4SDavid Sterba root_objectid = gang[ret - 1]->root_key.objectid + 1; 438266b4ffd1SJosef Bacik 4383fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 4384fc7cbcd4SDavid Sterba /* Avoid to grab roots in dead_roots */ 4385fc7cbcd4SDavid Sterba if (btrfs_root_refs(&gang[i]->root_item) == 0) { 4386fc7cbcd4SDavid Sterba gang[i] = NULL; 438765d33fd7SQu Wenruo continue; 4388c146afadSYan Zheng } 4389fc7cbcd4SDavid Sterba /* grab all the search result for later use */ 4390fc7cbcd4SDavid Sterba gang[i] = btrfs_grab_root(gang[i]); 4391fc7cbcd4SDavid Sterba } 4392fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4393fc7cbcd4SDavid Sterba 4394fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 4395fc7cbcd4SDavid Sterba if (!gang[i]) 4396fc7cbcd4SDavid Sterba continue; 4397fc7cbcd4SDavid Sterba root_objectid = gang[i]->root_key.objectid; 4398fc7cbcd4SDavid Sterba err = btrfs_orphan_cleanup(gang[i]); 4399fc7cbcd4SDavid Sterba if (err) 44006989627dSJosef Bacik goto out; 4401fc7cbcd4SDavid Sterba btrfs_put_root(gang[i]); 4402fc7cbcd4SDavid Sterba } 4403fc7cbcd4SDavid Sterba root_objectid++; 4404c146afadSYan Zheng } 44056989627dSJosef Bacik out: 4406fc7cbcd4SDavid Sterba /* release the uncleaned roots due to error */ 4407fc7cbcd4SDavid Sterba for (; i < ret; i++) { 4408fc7cbcd4SDavid Sterba if (gang[i]) 4409fc7cbcd4SDavid Sterba btrfs_put_root(gang[i]); 441065d33fd7SQu Wenruo } 441165d33fd7SQu Wenruo return err; 4412c146afadSYan Zheng } 4413c146afadSYan Zheng 44146bccf3abSJeff Mahoney int btrfs_commit_super(struct btrfs_fs_info *fs_info) 4415c146afadSYan Zheng { 44166bccf3abSJeff Mahoney struct btrfs_root *root = fs_info->tree_root; 4417c146afadSYan Zheng struct btrfs_trans_handle *trans; 4418c146afadSYan Zheng 44190b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 44202ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 44210b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 44220b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 4423c71bf099SYan, Zheng 4424c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 44250b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 44260b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4427c71bf099SYan, Zheng 44287a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 44293612b495STsutomu Itoh if (IS_ERR(trans)) 44303612b495STsutomu Itoh return PTR_ERR(trans); 44313a45bb20SJeff Mahoney return btrfs_commit_transaction(trans); 4432c146afadSYan Zheng } 4433c146afadSYan Zheng 443436c86a9eSQu Wenruo static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info) 443536c86a9eSQu Wenruo { 443636c86a9eSQu Wenruo struct btrfs_transaction *trans; 443736c86a9eSQu Wenruo struct btrfs_transaction *tmp; 443836c86a9eSQu Wenruo bool found = false; 443936c86a9eSQu Wenruo 444036c86a9eSQu Wenruo if (list_empty(&fs_info->trans_list)) 444136c86a9eSQu Wenruo return; 444236c86a9eSQu Wenruo 444336c86a9eSQu Wenruo /* 444436c86a9eSQu Wenruo * This function is only called at the very end of close_ctree(), 444536c86a9eSQu Wenruo * thus no other running transaction, no need to take trans_lock. 444636c86a9eSQu Wenruo */ 444736c86a9eSQu Wenruo ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags)); 444836c86a9eSQu Wenruo list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) { 444936c86a9eSQu Wenruo struct extent_state *cached = NULL; 445036c86a9eSQu Wenruo u64 dirty_bytes = 0; 445136c86a9eSQu Wenruo u64 cur = 0; 445236c86a9eSQu Wenruo u64 found_start; 445336c86a9eSQu Wenruo u64 found_end; 445436c86a9eSQu Wenruo 445536c86a9eSQu Wenruo found = true; 445636c86a9eSQu Wenruo while (!find_first_extent_bit(&trans->dirty_pages, cur, 445736c86a9eSQu Wenruo &found_start, &found_end, EXTENT_DIRTY, &cached)) { 445836c86a9eSQu Wenruo dirty_bytes += found_end + 1 - found_start; 445936c86a9eSQu Wenruo cur = found_end + 1; 446036c86a9eSQu Wenruo } 446136c86a9eSQu Wenruo btrfs_warn(fs_info, 446236c86a9eSQu Wenruo "transaction %llu (with %llu dirty metadata bytes) is not committed", 446336c86a9eSQu Wenruo trans->transid, dirty_bytes); 446436c86a9eSQu Wenruo btrfs_cleanup_one_transaction(trans, fs_info); 446536c86a9eSQu Wenruo 446636c86a9eSQu Wenruo if (trans == fs_info->running_transaction) 446736c86a9eSQu Wenruo fs_info->running_transaction = NULL; 446836c86a9eSQu Wenruo list_del_init(&trans->list); 446936c86a9eSQu Wenruo 447036c86a9eSQu Wenruo btrfs_put_transaction(trans); 447136c86a9eSQu Wenruo trace_btrfs_transaction_commit(fs_info); 447236c86a9eSQu Wenruo } 447336c86a9eSQu Wenruo ASSERT(!found); 447436c86a9eSQu Wenruo } 447536c86a9eSQu Wenruo 4476b105e927SDavid Sterba void __cold close_ctree(struct btrfs_fs_info *fs_info) 4477eb60ceacSChris Mason { 4478c146afadSYan Zheng int ret; 4479e089f05cSChris Mason 4480afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); 448131e70e52SFilipe Manana 448231e70e52SFilipe Manana /* 44838a1f1e3dSFilipe Manana * If we had UNFINISHED_DROPS we could still be processing them, so 44848a1f1e3dSFilipe Manana * clear that bit and wake up relocation so it can stop. 44858a1f1e3dSFilipe Manana * We must do this before stopping the block group reclaim task, because 44868a1f1e3dSFilipe Manana * at btrfs_relocate_block_group() we wait for this bit, and after the 44878a1f1e3dSFilipe Manana * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we 44888a1f1e3dSFilipe Manana * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will 44898a1f1e3dSFilipe Manana * return 1. 44908a1f1e3dSFilipe Manana */ 44918a1f1e3dSFilipe Manana btrfs_wake_unfinished_drop(fs_info); 44928a1f1e3dSFilipe Manana 44938a1f1e3dSFilipe Manana /* 449431e70e52SFilipe Manana * We may have the reclaim task running and relocating a data block group, 449531e70e52SFilipe Manana * in which case it may create delayed iputs. So stop it before we park 449631e70e52SFilipe Manana * the cleaner kthread otherwise we can get new delayed iputs after 449731e70e52SFilipe Manana * parking the cleaner, and that can make the async reclaim task to hang 449831e70e52SFilipe Manana * if it's waiting for delayed iputs to complete, since the cleaner is 449931e70e52SFilipe Manana * parked and can not run delayed iputs - this will make us hang when 450031e70e52SFilipe Manana * trying to stop the async reclaim task. 450131e70e52SFilipe Manana */ 450231e70e52SFilipe Manana cancel_work_sync(&fs_info->reclaim_bgs_work); 4503d6fd0ae2SOmar Sandoval /* 4504d6fd0ae2SOmar Sandoval * We don't want the cleaner to start new transactions, add more delayed 4505d6fd0ae2SOmar Sandoval * iputs, etc. while we're closing. We can't use kthread_stop() yet 4506d6fd0ae2SOmar Sandoval * because that frees the task_struct, and the transaction kthread might 4507d6fd0ae2SOmar Sandoval * still try to wake up the cleaner. 4508d6fd0ae2SOmar Sandoval */ 4509d6fd0ae2SOmar Sandoval kthread_park(fs_info->cleaner_kthread); 4510a2135011SChris Mason 45117343dd61SJustin Maggard /* wait for the qgroup rescan worker to stop */ 4512d06f23d6SJeff Mahoney btrfs_qgroup_wait_for_completion(fs_info, false); 45137343dd61SJustin Maggard 4514803b2f54SStefan Behrens /* wait for the uuid_scan task to finish */ 4515803b2f54SStefan Behrens down(&fs_info->uuid_tree_rescan_sem); 4516803b2f54SStefan Behrens /* avoid complains from lockdep et al., set sem back to initial state */ 4517803b2f54SStefan Behrens up(&fs_info->uuid_tree_rescan_sem); 4518803b2f54SStefan Behrens 4519837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 4520aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 4521837d5b6eSIlya Dryomov 45228dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 45238dabb742SStefan Behrens 4524aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 45254cb5300bSChris Mason 45264cb5300bSChris Mason /* wait for any defraggers to finish */ 45274cb5300bSChris Mason wait_event(fs_info->transaction_wait, 45284cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 45294cb5300bSChris Mason 45304cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 453126176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 45324cb5300bSChris Mason 4533a362bb86SFilipe Manana /* 4534a362bb86SFilipe Manana * After we parked the cleaner kthread, ordered extents may have 4535a362bb86SFilipe Manana * completed and created new delayed iputs. If one of the async reclaim 4536a362bb86SFilipe Manana * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we 4537a362bb86SFilipe Manana * can hang forever trying to stop it, because if a delayed iput is 4538a362bb86SFilipe Manana * added after it ran btrfs_run_delayed_iputs() and before it called 4539a362bb86SFilipe Manana * btrfs_wait_on_delayed_iputs(), it will hang forever since there is 4540a362bb86SFilipe Manana * no one else to run iputs. 4541a362bb86SFilipe Manana * 4542a362bb86SFilipe Manana * So wait for all ongoing ordered extents to complete and then run 4543a362bb86SFilipe Manana * delayed iputs. This works because once we reach this point no one 4544a362bb86SFilipe Manana * can either create new ordered extents nor create delayed iputs 4545a362bb86SFilipe Manana * through some other means. 4546a362bb86SFilipe Manana * 4547a362bb86SFilipe Manana * Also note that btrfs_wait_ordered_roots() is not safe here, because 4548a362bb86SFilipe Manana * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent, 4549a362bb86SFilipe Manana * but the delayed iput for the respective inode is made only when doing 4550a362bb86SFilipe Manana * the final btrfs_put_ordered_extent() (which must happen at 4551a362bb86SFilipe Manana * btrfs_finish_ordered_io() when we are unmounting). 4552a362bb86SFilipe Manana */ 4553a362bb86SFilipe Manana btrfs_flush_workqueue(fs_info->endio_write_workers); 4554a362bb86SFilipe Manana /* Ordered extents for free space inodes. */ 4555a362bb86SFilipe Manana btrfs_flush_workqueue(fs_info->endio_freespace_worker); 4556a362bb86SFilipe Manana btrfs_run_delayed_iputs(fs_info); 4557a362bb86SFilipe Manana 455821c7e756SMiao Xie cancel_work_sync(&fs_info->async_reclaim_work); 455957056740SJosef Bacik cancel_work_sync(&fs_info->async_data_reclaim_work); 4560576fa348SJosef Bacik cancel_work_sync(&fs_info->preempt_reclaim_work); 456121c7e756SMiao Xie 4562b0643e59SDennis Zhou /* Cancel or finish ongoing discard work */ 4563b0643e59SDennis Zhou btrfs_discard_cleanup(fs_info); 4564b0643e59SDennis Zhou 4565bc98a42cSDavid Howells if (!sb_rdonly(fs_info->sb)) { 4566e44163e1SJeff Mahoney /* 4567d6fd0ae2SOmar Sandoval * The cleaner kthread is stopped, so do one final pass over 4568d6fd0ae2SOmar Sandoval * unused block groups. 4569e44163e1SJeff Mahoney */ 45700b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 4571e44163e1SJeff Mahoney 4572f0cc2cd7SFilipe Manana /* 4573f0cc2cd7SFilipe Manana * There might be existing delayed inode workers still running 4574f0cc2cd7SFilipe Manana * and holding an empty delayed inode item. We must wait for 4575f0cc2cd7SFilipe Manana * them to complete first because they can create a transaction. 4576f0cc2cd7SFilipe Manana * This happens when someone calls btrfs_balance_delayed_items() 4577f0cc2cd7SFilipe Manana * and then a transaction commit runs the same delayed nodes 4578f0cc2cd7SFilipe Manana * before any delayed worker has done something with the nodes. 4579f0cc2cd7SFilipe Manana * We must wait for any worker here and not at transaction 4580f0cc2cd7SFilipe Manana * commit time since that could cause a deadlock. 4581f0cc2cd7SFilipe Manana * This is a very rare case. 4582f0cc2cd7SFilipe Manana */ 4583f0cc2cd7SFilipe Manana btrfs_flush_workqueue(fs_info->delayed_workers); 4584f0cc2cd7SFilipe Manana 45856bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 4586d397712bSChris Mason if (ret) 458704892340SEric Sandeen btrfs_err(fs_info, "commit super ret %d", ret); 4588c146afadSYan Zheng } 4589ed2ff2cbSChris Mason 459084961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) 45912ff7e61eSJeff Mahoney btrfs_error_commit_super(fs_info); 4592acce952bSliubo 4593e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 4594e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 45958929ecfaSYan, Zheng 4596e187831eSJosef Bacik ASSERT(list_empty(&fs_info->delayed_iputs)); 4597afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); 4598f25784b3SYan Zheng 45995958253cSQu Wenruo if (btrfs_check_quota_leak(fs_info)) { 46005958253cSQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 46015958253cSQu Wenruo btrfs_err(fs_info, "qgroup reserved space leaked"); 46025958253cSQu Wenruo } 46035958253cSQu Wenruo 460404892340SEric Sandeen btrfs_free_qgroup_config(fs_info); 4605fe816d0fSNikolay Borisov ASSERT(list_empty(&fs_info->delalloc_roots)); 4606bcef60f2SArne Jansen 4607963d678bSMiao Xie if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 460804892340SEric Sandeen btrfs_info(fs_info, "at unmount delalloc count %lld", 4609963d678bSMiao Xie percpu_counter_sum(&fs_info->delalloc_bytes)); 4610b0c68f8bSChris Mason } 461131153d81SYan Zheng 46125deb17e1SJosef Bacik if (percpu_counter_sum(&fs_info->ordered_bytes)) 46134297ff84SJosef Bacik btrfs_info(fs_info, "at unmount dio bytes count %lld", 46145deb17e1SJosef Bacik percpu_counter_sum(&fs_info->ordered_bytes)); 46154297ff84SJosef Bacik 46166618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 4617b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 46185ac1d209SJeff Mahoney 46191a4319ccSLiu Bo btrfs_put_block_group_cache(fs_info); 46201a4319ccSLiu Bo 4621de348ee0SWang Shilong /* 4622de348ee0SWang Shilong * we must make sure there is not any read request to 4623de348ee0SWang Shilong * submit after we stopping all workers. 4624de348ee0SWang Shilong */ 4625de348ee0SWang Shilong invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 462696192499SJosef Bacik btrfs_stop_all_workers(fs_info); 462796192499SJosef Bacik 46280a31daa4SFilipe Manana /* We shouldn't have any transaction open at this point */ 462936c86a9eSQu Wenruo warn_about_uncommitted_trans(fs_info); 46300a31daa4SFilipe Manana 4631afcdd129SJosef Bacik clear_bit(BTRFS_FS_OPEN, &fs_info->flags); 46324273eaffSAnand Jain free_root_pointers(fs_info, true); 46338c38938cSJosef Bacik btrfs_free_fs_roots(fs_info); 46349ad6b7bcSChris Mason 46354e19443dSJosef Bacik /* 46364e19443dSJosef Bacik * We must free the block groups after dropping the fs_roots as we could 46374e19443dSJosef Bacik * have had an IO error and have left over tree log blocks that aren't 46384e19443dSJosef Bacik * cleaned up until the fs roots are freed. This makes the block group 46394e19443dSJosef Bacik * accounting appear to be wrong because there's pending reserved bytes, 46404e19443dSJosef Bacik * so make sure we do the block group cleanup afterwards. 46414e19443dSJosef Bacik */ 46424e19443dSJosef Bacik btrfs_free_block_groups(fs_info); 46434e19443dSJosef Bacik 464413e6c37bSJosef Bacik iput(fs_info->btree_inode); 4645d6bfde87SChris Mason 464621adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 46470b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) 46482ff7e61eSJeff Mahoney btrfsic_unmount(fs_info->fs_devices); 464921adbd5cSStefan Behrens #endif 465021adbd5cSStefan Behrens 46510b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 465268c94e55SNikolay Borisov btrfs_close_devices(fs_info->fs_devices); 4653eb60ceacSChris Mason } 4654eb60ceacSChris Mason 4655b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 4656b9fab919SChris Mason int atomic) 4657ccd467d6SChris Mason { 46581259ab75SChris Mason int ret; 4659727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 46601259ab75SChris Mason 46610b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 46621259ab75SChris Mason if (!ret) 46631259ab75SChris Mason return ret; 46641259ab75SChris Mason 46651259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 4666b9fab919SChris Mason parent_transid, atomic); 4667b9fab919SChris Mason if (ret == -EAGAIN) 4668b9fab919SChris Mason return ret; 46691259ab75SChris Mason return !ret; 46705f39d397SChris Mason } 46716702ed49SChris Mason 46725f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 46735f39d397SChris Mason { 46742f4d60dfSQu Wenruo struct btrfs_fs_info *fs_info = buf->fs_info; 46755f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 4676b9473439SChris Mason int was_dirty; 4677b4ce94deSChris Mason 467806ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 467906ea65a3SJosef Bacik /* 468006ea65a3SJosef Bacik * This is a fast path so only do this check if we have sanity tests 468152042d8eSAndrea Gelmini * enabled. Normal people shouldn't be using unmapped buffers as dirty 468206ea65a3SJosef Bacik * outside of the sanity tests. 468306ea65a3SJosef Bacik */ 4684b0132a3bSNikolay Borisov if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) 468506ea65a3SJosef Bacik return; 468606ea65a3SJosef Bacik #endif 468749d0c642SFilipe Manana btrfs_assert_tree_write_locked(buf); 46880b246afaSJeff Mahoney if (transid != fs_info->generation) 46895d163e0eSJeff Mahoney WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", 46900b246afaSJeff Mahoney buf->start, transid, fs_info->generation); 46910b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 4692e2d84521SMiao Xie if (!was_dirty) 4693104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 4694e2d84521SMiao Xie buf->len, 46950b246afaSJeff Mahoney fs_info->dirty_metadata_batch); 46961f21ef0aSFilipe Manana #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 469769fc6cbbSQu Wenruo /* 469869fc6cbbSQu Wenruo * Since btrfs_mark_buffer_dirty() can be called with item pointer set 469969fc6cbbSQu Wenruo * but item data not updated. 470069fc6cbbSQu Wenruo * So here we should only check item pointers, not item data. 470169fc6cbbSQu Wenruo */ 470269fc6cbbSQu Wenruo if (btrfs_header_level(buf) == 0 && 4703cfdaad5eSDavid Sterba btrfs_check_leaf_relaxed(buf)) { 4704a4f78750SDavid Sterba btrfs_print_leaf(buf); 47051f21ef0aSFilipe Manana ASSERT(0); 47061f21ef0aSFilipe Manana } 47071f21ef0aSFilipe Manana #endif 4708eb60ceacSChris Mason } 4709eb60ceacSChris Mason 47102ff7e61eSJeff Mahoney static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, 4711b53d3f5dSLiu Bo int flush_delayed) 471235b7e476SChris Mason { 4713188de649SChris Mason /* 4714188de649SChris Mason * looks as though older kernels can get into trouble with 4715188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 4716188de649SChris Mason */ 4717e2d84521SMiao Xie int ret; 4718d6bfde87SChris Mason 47196933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 4720d6bfde87SChris Mason return; 4721d6bfde87SChris Mason 4722b53d3f5dSLiu Bo if (flush_delayed) 47232ff7e61eSJeff Mahoney btrfs_balance_delayed_items(fs_info); 472416cdcec7SMiao Xie 4725d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 4726d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 4727d814a491SEthan Lien fs_info->dirty_metadata_batch); 4728e2d84521SMiao Xie if (ret > 0) { 47290b246afaSJeff Mahoney balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); 473016cdcec7SMiao Xie } 473116cdcec7SMiao Xie } 473216cdcec7SMiao Xie 47332ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) 473416cdcec7SMiao Xie { 47352ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 1); 473635b7e476SChris Mason } 4737b53d3f5dSLiu Bo 47382ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) 4739b53d3f5dSLiu Bo { 47402ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 0); 4741d6bfde87SChris Mason } 47426b80053dSChris Mason 47432ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) 4744acce952bSliubo { 4745fe816d0fSNikolay Borisov /* cleanup FS via transaction */ 4746fe816d0fSNikolay Borisov btrfs_cleanup_transaction(fs_info); 4747fe816d0fSNikolay Borisov 47480b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 47492ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 47500b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 4751acce952bSliubo 47520b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 47530b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4754acce952bSliubo } 4755acce952bSliubo 4756ef67963dSJosef Bacik static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info) 4757ef67963dSJosef Bacik { 4758fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 4759fc7cbcd4SDavid Sterba u64 root_objectid = 0; 4760fc7cbcd4SDavid Sterba int ret; 4761ef67963dSJosef Bacik 4762fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4763fc7cbcd4SDavid Sterba while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4764fc7cbcd4SDavid Sterba (void **)gang, root_objectid, 4765fc7cbcd4SDavid Sterba ARRAY_SIZE(gang))) != 0) { 4766fc7cbcd4SDavid Sterba int i; 4767ef67963dSJosef Bacik 4768fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) 4769fc7cbcd4SDavid Sterba gang[i] = btrfs_grab_root(gang[i]); 4770fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4771fc7cbcd4SDavid Sterba 4772fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 4773fc7cbcd4SDavid Sterba if (!gang[i]) 4774ef67963dSJosef Bacik continue; 4775fc7cbcd4SDavid Sterba root_objectid = gang[i]->root_key.objectid; 4776fc7cbcd4SDavid Sterba btrfs_free_log(NULL, gang[i]); 4777fc7cbcd4SDavid Sterba btrfs_put_root(gang[i]); 4778ef67963dSJosef Bacik } 4779fc7cbcd4SDavid Sterba root_objectid++; 4780fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4781ef67963dSJosef Bacik } 4782fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4783ef67963dSJosef Bacik btrfs_free_log_root_tree(NULL, fs_info); 4784ef67963dSJosef Bacik } 4785ef67963dSJosef Bacik 4786143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4787acce952bSliubo { 4788acce952bSliubo struct btrfs_ordered_extent *ordered; 4789acce952bSliubo 4790199c2a9cSMiao Xie spin_lock(&root->ordered_extent_lock); 4791779880efSJosef Bacik /* 4792779880efSJosef Bacik * This will just short circuit the ordered completion stuff which will 4793779880efSJosef Bacik * make sure the ordered extent gets properly cleaned up. 4794779880efSJosef Bacik */ 4795199c2a9cSMiao Xie list_for_each_entry(ordered, &root->ordered_extents, 4796779880efSJosef Bacik root_extent_list) 4797779880efSJosef Bacik set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4798199c2a9cSMiao Xie spin_unlock(&root->ordered_extent_lock); 4799199c2a9cSMiao Xie } 4800199c2a9cSMiao Xie 4801199c2a9cSMiao Xie static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4802199c2a9cSMiao Xie { 4803199c2a9cSMiao Xie struct btrfs_root *root; 4804199c2a9cSMiao Xie struct list_head splice; 4805199c2a9cSMiao Xie 4806199c2a9cSMiao Xie INIT_LIST_HEAD(&splice); 4807199c2a9cSMiao Xie 4808199c2a9cSMiao Xie spin_lock(&fs_info->ordered_root_lock); 4809199c2a9cSMiao Xie list_splice_init(&fs_info->ordered_roots, &splice); 4810199c2a9cSMiao Xie while (!list_empty(&splice)) { 4811199c2a9cSMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4812199c2a9cSMiao Xie ordered_root); 48131de2cfdeSJosef Bacik list_move_tail(&root->ordered_root, 48141de2cfdeSJosef Bacik &fs_info->ordered_roots); 4815199c2a9cSMiao Xie 48162a85d9caSLiu Bo spin_unlock(&fs_info->ordered_root_lock); 4817199c2a9cSMiao Xie btrfs_destroy_ordered_extents(root); 4818199c2a9cSMiao Xie 48192a85d9caSLiu Bo cond_resched(); 48202a85d9caSLiu Bo spin_lock(&fs_info->ordered_root_lock); 4821199c2a9cSMiao Xie } 4822199c2a9cSMiao Xie spin_unlock(&fs_info->ordered_root_lock); 482374d5d229SJosef Bacik 482474d5d229SJosef Bacik /* 482574d5d229SJosef Bacik * We need this here because if we've been flipped read-only we won't 482674d5d229SJosef Bacik * get sync() from the umount, so we need to make sure any ordered 482774d5d229SJosef Bacik * extents that haven't had their dirty pages IO start writeout yet 482874d5d229SJosef Bacik * actually get run and error out properly. 482974d5d229SJosef Bacik */ 483074d5d229SJosef Bacik btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 4831acce952bSliubo } 4832acce952bSliubo 483335a3621bSStefan Behrens static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 48342ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4835acce952bSliubo { 4836acce952bSliubo struct rb_node *node; 4837acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 4838acce952bSliubo struct btrfs_delayed_ref_node *ref; 4839acce952bSliubo int ret = 0; 4840acce952bSliubo 4841acce952bSliubo delayed_refs = &trans->delayed_refs; 4842acce952bSliubo 4843acce952bSliubo spin_lock(&delayed_refs->lock); 4844d7df2c79SJosef Bacik if (atomic_read(&delayed_refs->num_entries) == 0) { 4845cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 4846b79ce3ddSDavid Sterba btrfs_debug(fs_info, "delayed_refs has NO entry"); 4847acce952bSliubo return ret; 4848acce952bSliubo } 4849acce952bSliubo 48505c9d028bSLiu Bo while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { 4851d7df2c79SJosef Bacik struct btrfs_delayed_ref_head *head; 48520e0adbcfSJosef Bacik struct rb_node *n; 4853e78417d1SJosef Bacik bool pin_bytes = false; 4854acce952bSliubo 4855d7df2c79SJosef Bacik head = rb_entry(node, struct btrfs_delayed_ref_head, 4856d7df2c79SJosef Bacik href_node); 48573069bd26SJosef Bacik if (btrfs_delayed_ref_lock(delayed_refs, head)) 4858b939d1abSJosef Bacik continue; 48593069bd26SJosef Bacik 4860d7df2c79SJosef Bacik spin_lock(&head->lock); 4861e3d03965SLiu Bo while ((n = rb_first_cached(&head->ref_tree)) != NULL) { 48620e0adbcfSJosef Bacik ref = rb_entry(n, struct btrfs_delayed_ref_node, 48630e0adbcfSJosef Bacik ref_node); 4864d7df2c79SJosef Bacik ref->in_tree = 0; 4865e3d03965SLiu Bo rb_erase_cached(&ref->ref_node, &head->ref_tree); 48660e0adbcfSJosef Bacik RB_CLEAR_NODE(&ref->ref_node); 48671d57ee94SWang Xiaoguang if (!list_empty(&ref->add_list)) 48681d57ee94SWang Xiaoguang list_del(&ref->add_list); 4869d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4870d7df2c79SJosef Bacik btrfs_put_delayed_ref(ref); 4871d7df2c79SJosef Bacik } 487254067ae9SJosef Bacik if (head->must_insert_reserved) 4873e78417d1SJosef Bacik pin_bytes = true; 487478a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 4875fa781ceaSJosef Bacik btrfs_delete_ref_head(delayed_refs, head); 4876d7df2c79SJosef Bacik spin_unlock(&head->lock); 4877acce952bSliubo spin_unlock(&delayed_refs->lock); 4878e78417d1SJosef Bacik mutex_unlock(&head->mutex); 4879acce952bSliubo 4880f603bb94SNikolay Borisov if (pin_bytes) { 4881f603bb94SNikolay Borisov struct btrfs_block_group *cache; 4882f603bb94SNikolay Borisov 4883f603bb94SNikolay Borisov cache = btrfs_lookup_block_group(fs_info, head->bytenr); 4884f603bb94SNikolay Borisov BUG_ON(!cache); 4885f603bb94SNikolay Borisov 4886f603bb94SNikolay Borisov spin_lock(&cache->space_info->lock); 4887f603bb94SNikolay Borisov spin_lock(&cache->lock); 4888f603bb94SNikolay Borisov cache->pinned += head->num_bytes; 4889f603bb94SNikolay Borisov btrfs_space_info_update_bytes_pinned(fs_info, 4890f603bb94SNikolay Borisov cache->space_info, head->num_bytes); 4891f603bb94SNikolay Borisov cache->reserved -= head->num_bytes; 4892f603bb94SNikolay Borisov cache->space_info->bytes_reserved -= head->num_bytes; 4893f603bb94SNikolay Borisov spin_unlock(&cache->lock); 4894f603bb94SNikolay Borisov spin_unlock(&cache->space_info->lock); 4895f603bb94SNikolay Borisov 4896f603bb94SNikolay Borisov btrfs_put_block_group(cache); 4897f603bb94SNikolay Borisov 4898f603bb94SNikolay Borisov btrfs_error_unpin_extent_range(fs_info, head->bytenr, 4899f603bb94SNikolay Borisov head->bytenr + head->num_bytes - 1); 4900f603bb94SNikolay Borisov } 490131890da0SJosef Bacik btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); 4902d278850eSJosef Bacik btrfs_put_delayed_ref_head(head); 4903acce952bSliubo cond_resched(); 4904acce952bSliubo spin_lock(&delayed_refs->lock); 4905acce952bSliubo } 490681f7eb00SJeff Mahoney btrfs_qgroup_destroy_extent_records(trans); 4907acce952bSliubo 4908acce952bSliubo spin_unlock(&delayed_refs->lock); 4909acce952bSliubo 4910acce952bSliubo return ret; 4911acce952bSliubo } 4912acce952bSliubo 4913143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 4914acce952bSliubo { 4915acce952bSliubo struct btrfs_inode *btrfs_inode; 4916acce952bSliubo struct list_head splice; 4917acce952bSliubo 4918acce952bSliubo INIT_LIST_HEAD(&splice); 4919acce952bSliubo 4920eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4921eb73c1b7SMiao Xie list_splice_init(&root->delalloc_inodes, &splice); 4922acce952bSliubo 4923acce952bSliubo while (!list_empty(&splice)) { 4924fe816d0fSNikolay Borisov struct inode *inode = NULL; 4925eb73c1b7SMiao Xie btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 4926acce952bSliubo delalloc_inodes); 4927fe816d0fSNikolay Borisov __btrfs_del_delalloc_inode(root, btrfs_inode); 4928eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4929acce952bSliubo 4930fe816d0fSNikolay Borisov /* 4931fe816d0fSNikolay Borisov * Make sure we get a live inode and that it'll not disappear 4932fe816d0fSNikolay Borisov * meanwhile. 4933fe816d0fSNikolay Borisov */ 4934fe816d0fSNikolay Borisov inode = igrab(&btrfs_inode->vfs_inode); 4935fe816d0fSNikolay Borisov if (inode) { 4936fe816d0fSNikolay Borisov invalidate_inode_pages2(inode->i_mapping); 4937fe816d0fSNikolay Borisov iput(inode); 4938fe816d0fSNikolay Borisov } 4939eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4940acce952bSliubo } 4941eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4942eb73c1b7SMiao Xie } 4943eb73c1b7SMiao Xie 4944eb73c1b7SMiao Xie static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 4945eb73c1b7SMiao Xie { 4946eb73c1b7SMiao Xie struct btrfs_root *root; 4947eb73c1b7SMiao Xie struct list_head splice; 4948eb73c1b7SMiao Xie 4949eb73c1b7SMiao Xie INIT_LIST_HEAD(&splice); 4950eb73c1b7SMiao Xie 4951eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4952eb73c1b7SMiao Xie list_splice_init(&fs_info->delalloc_roots, &splice); 4953eb73c1b7SMiao Xie while (!list_empty(&splice)) { 4954eb73c1b7SMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4955eb73c1b7SMiao Xie delalloc_root); 495600246528SJosef Bacik root = btrfs_grab_root(root); 4957eb73c1b7SMiao Xie BUG_ON(!root); 4958eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4959eb73c1b7SMiao Xie 4960eb73c1b7SMiao Xie btrfs_destroy_delalloc_inodes(root); 496100246528SJosef Bacik btrfs_put_root(root); 4962eb73c1b7SMiao Xie 4963eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4964eb73c1b7SMiao Xie } 4965eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4966acce952bSliubo } 4967acce952bSliubo 49682ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 4969acce952bSliubo struct extent_io_tree *dirty_pages, 4970acce952bSliubo int mark) 4971acce952bSliubo { 4972acce952bSliubo int ret; 4973acce952bSliubo struct extent_buffer *eb; 4974acce952bSliubo u64 start = 0; 4975acce952bSliubo u64 end; 4976acce952bSliubo 4977acce952bSliubo while (1) { 4978acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 4979e6138876SJosef Bacik mark, NULL); 4980acce952bSliubo if (ret) 4981acce952bSliubo break; 4982acce952bSliubo 498391166212SDavid Sterba clear_extent_bits(dirty_pages, start, end, mark); 4984acce952bSliubo while (start <= end) { 49850b246afaSJeff Mahoney eb = find_extent_buffer(fs_info, start); 49860b246afaSJeff Mahoney start += fs_info->nodesize; 4987fd8b2b61SJosef Bacik if (!eb) 4988acce952bSliubo continue; 4989acce952bSliubo 4990c4e54a65SJosef Bacik btrfs_tree_lock(eb); 4991c4e54a65SJosef Bacik wait_on_extent_buffer_writeback(eb); 4992190a8339SJosef Bacik btrfs_clear_buffer_dirty(NULL, eb); 4993c4e54a65SJosef Bacik btrfs_tree_unlock(eb); 4994c4e54a65SJosef Bacik 4995fd8b2b61SJosef Bacik free_extent_buffer_stale(eb); 4996acce952bSliubo } 4997acce952bSliubo } 4998acce952bSliubo 4999acce952bSliubo return ret; 5000acce952bSliubo } 5001acce952bSliubo 50022ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 5003fe119a6eSNikolay Borisov struct extent_io_tree *unpin) 5004acce952bSliubo { 5005acce952bSliubo u64 start; 5006acce952bSliubo u64 end; 5007acce952bSliubo int ret; 5008acce952bSliubo 5009acce952bSliubo while (1) { 50100e6ec385SFilipe Manana struct extent_state *cached_state = NULL; 50110e6ec385SFilipe Manana 5012fcd5e742SLu Fengqi /* 5013fcd5e742SLu Fengqi * The btrfs_finish_extent_commit() may get the same range as 5014fcd5e742SLu Fengqi * ours between find_first_extent_bit and clear_extent_dirty. 5015fcd5e742SLu Fengqi * Hence, hold the unused_bg_unpin_mutex to avoid double unpin 5016fcd5e742SLu Fengqi * the same extent range. 5017fcd5e742SLu Fengqi */ 5018fcd5e742SLu Fengqi mutex_lock(&fs_info->unused_bg_unpin_mutex); 5019acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 50200e6ec385SFilipe Manana EXTENT_DIRTY, &cached_state); 5021fcd5e742SLu Fengqi if (ret) { 5022fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 5023acce952bSliubo break; 5024fcd5e742SLu Fengqi } 5025acce952bSliubo 50260e6ec385SFilipe Manana clear_extent_dirty(unpin, start, end, &cached_state); 50270e6ec385SFilipe Manana free_extent_state(cached_state); 50282ff7e61eSJeff Mahoney btrfs_error_unpin_extent_range(fs_info, start, end); 5029fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 5030acce952bSliubo cond_resched(); 5031acce952bSliubo } 5032acce952bSliubo 5033acce952bSliubo return 0; 5034acce952bSliubo } 5035acce952bSliubo 503632da5386SDavid Sterba static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache) 5037c79a1751SLiu Bo { 5038c79a1751SLiu Bo struct inode *inode; 5039c79a1751SLiu Bo 5040c79a1751SLiu Bo inode = cache->io_ctl.inode; 5041c79a1751SLiu Bo if (inode) { 5042c79a1751SLiu Bo invalidate_inode_pages2(inode->i_mapping); 5043c79a1751SLiu Bo BTRFS_I(inode)->generation = 0; 5044c79a1751SLiu Bo cache->io_ctl.inode = NULL; 5045c79a1751SLiu Bo iput(inode); 5046c79a1751SLiu Bo } 5047bbc37d6eSFilipe Manana ASSERT(cache->io_ctl.pages == NULL); 5048c79a1751SLiu Bo btrfs_put_block_group(cache); 5049c79a1751SLiu Bo } 5050c79a1751SLiu Bo 5051c79a1751SLiu Bo void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, 50522ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 5053c79a1751SLiu Bo { 505432da5386SDavid Sterba struct btrfs_block_group *cache; 5055c79a1751SLiu Bo 5056c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 5057c79a1751SLiu Bo while (!list_empty(&cur_trans->dirty_bgs)) { 5058c79a1751SLiu Bo cache = list_first_entry(&cur_trans->dirty_bgs, 505932da5386SDavid Sterba struct btrfs_block_group, 5060c79a1751SLiu Bo dirty_list); 5061c79a1751SLiu Bo 5062c79a1751SLiu Bo if (!list_empty(&cache->io_list)) { 5063c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 5064c79a1751SLiu Bo list_del_init(&cache->io_list); 5065c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 5066c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 5067c79a1751SLiu Bo } 5068c79a1751SLiu Bo 5069c79a1751SLiu Bo list_del_init(&cache->dirty_list); 5070c79a1751SLiu Bo spin_lock(&cache->lock); 5071c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 5072c79a1751SLiu Bo spin_unlock(&cache->lock); 5073c79a1751SLiu Bo 5074c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 5075c79a1751SLiu Bo btrfs_put_block_group(cache); 5076ba2c4d4eSJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 5077c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 5078c79a1751SLiu Bo } 5079c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 5080c79a1751SLiu Bo 508145ae2c18SNikolay Borisov /* 508245ae2c18SNikolay Borisov * Refer to the definition of io_bgs member for details why it's safe 508345ae2c18SNikolay Borisov * to use it without any locking 508445ae2c18SNikolay Borisov */ 5085c79a1751SLiu Bo while (!list_empty(&cur_trans->io_bgs)) { 5086c79a1751SLiu Bo cache = list_first_entry(&cur_trans->io_bgs, 508732da5386SDavid Sterba struct btrfs_block_group, 5088c79a1751SLiu Bo io_list); 5089c79a1751SLiu Bo 5090c79a1751SLiu Bo list_del_init(&cache->io_list); 5091c79a1751SLiu Bo spin_lock(&cache->lock); 5092c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 5093c79a1751SLiu Bo spin_unlock(&cache->lock); 5094c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 5095c79a1751SLiu Bo } 5096c79a1751SLiu Bo } 5097c79a1751SLiu Bo 509849b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 50992ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 510049b25e05SJeff Mahoney { 5101bbbf7243SNikolay Borisov struct btrfs_device *dev, *tmp; 5102bbbf7243SNikolay Borisov 51032ff7e61eSJeff Mahoney btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 5104c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->dirty_bgs)); 5105c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->io_bgs)); 5106c79a1751SLiu Bo 5107bbbf7243SNikolay Borisov list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list, 5108bbbf7243SNikolay Borisov post_commit_list) { 5109bbbf7243SNikolay Borisov list_del_init(&dev->post_commit_list); 5110bbbf7243SNikolay Borisov } 5111bbbf7243SNikolay Borisov 51122ff7e61eSJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, fs_info); 511349b25e05SJeff Mahoney 51144a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 51150b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 511649b25e05SJeff Mahoney 51174a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 51180b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 511949b25e05SJeff Mahoney 5120ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 512167cde344SMiao Xie 51222ff7e61eSJeff Mahoney btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, 512349b25e05SJeff Mahoney EXTENT_DIRTY); 5124fe119a6eSNikolay Borisov btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents); 512549b25e05SJeff Mahoney 5126d3575156SNaohiro Aota btrfs_free_redirty_list(cur_trans); 5127d3575156SNaohiro Aota 51284a9d8bdeSMiao Xie cur_trans->state =TRANS_STATE_COMPLETED; 51294a9d8bdeSMiao Xie wake_up(&cur_trans->commit_wait); 513049b25e05SJeff Mahoney } 513149b25e05SJeff Mahoney 51322ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) 5133acce952bSliubo { 5134acce952bSliubo struct btrfs_transaction *t; 5135acce952bSliubo 51360b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 5137acce952bSliubo 51380b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 51390b246afaSJeff Mahoney while (!list_empty(&fs_info->trans_list)) { 51400b246afaSJeff Mahoney t = list_first_entry(&fs_info->trans_list, 5141724e2315SJosef Bacik struct btrfs_transaction, list); 5142724e2315SJosef Bacik if (t->state >= TRANS_STATE_COMMIT_START) { 51439b64f57dSElena Reshetova refcount_inc(&t->use_count); 51440b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 51452ff7e61eSJeff Mahoney btrfs_wait_for_commit(fs_info, t->transid); 5146724e2315SJosef Bacik btrfs_put_transaction(t); 51470b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5148724e2315SJosef Bacik continue; 5149724e2315SJosef Bacik } 51500b246afaSJeff Mahoney if (t == fs_info->running_transaction) { 5151724e2315SJosef Bacik t->state = TRANS_STATE_COMMIT_DOING; 51520b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 5153724e2315SJosef Bacik /* 5154724e2315SJosef Bacik * We wait for 0 num_writers since we don't hold a trans 5155724e2315SJosef Bacik * handle open currently for this transaction. 5156724e2315SJosef Bacik */ 5157724e2315SJosef Bacik wait_event(t->writer_wait, 5158724e2315SJosef Bacik atomic_read(&t->num_writers) == 0); 5159724e2315SJosef Bacik } else { 51600b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 5161724e2315SJosef Bacik } 51622ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(t, fs_info); 5163724e2315SJosef Bacik 51640b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 51650b246afaSJeff Mahoney if (t == fs_info->running_transaction) 51660b246afaSJeff Mahoney fs_info->running_transaction = NULL; 5167724e2315SJosef Bacik list_del_init(&t->list); 51680b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 5169a4abeea4SJosef Bacik 5170724e2315SJosef Bacik btrfs_put_transaction(t); 51712e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 51720b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5173724e2315SJosef Bacik } 51740b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 51750b246afaSJeff Mahoney btrfs_destroy_all_ordered_extents(fs_info); 5176ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 5177ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 51780b246afaSJeff Mahoney btrfs_destroy_all_delalloc_inodes(fs_info); 5179ef67963dSJosef Bacik btrfs_drop_all_logs(fs_info); 51800b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 5181acce952bSliubo 5182acce952bSliubo return 0; 5183acce952bSliubo } 5184ec7d6dfdSNikolay Borisov 5185453e4873SNikolay Borisov int btrfs_init_root_free_objectid(struct btrfs_root *root) 5186ec7d6dfdSNikolay Borisov { 5187ec7d6dfdSNikolay Borisov struct btrfs_path *path; 5188ec7d6dfdSNikolay Borisov int ret; 5189ec7d6dfdSNikolay Borisov struct extent_buffer *l; 5190ec7d6dfdSNikolay Borisov struct btrfs_key search_key; 5191ec7d6dfdSNikolay Borisov struct btrfs_key found_key; 5192ec7d6dfdSNikolay Borisov int slot; 5193ec7d6dfdSNikolay Borisov 5194ec7d6dfdSNikolay Borisov path = btrfs_alloc_path(); 5195ec7d6dfdSNikolay Borisov if (!path) 5196ec7d6dfdSNikolay Borisov return -ENOMEM; 5197ec7d6dfdSNikolay Borisov 5198ec7d6dfdSNikolay Borisov search_key.objectid = BTRFS_LAST_FREE_OBJECTID; 5199ec7d6dfdSNikolay Borisov search_key.type = -1; 5200ec7d6dfdSNikolay Borisov search_key.offset = (u64)-1; 5201ec7d6dfdSNikolay Borisov ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); 5202ec7d6dfdSNikolay Borisov if (ret < 0) 5203ec7d6dfdSNikolay Borisov goto error; 5204ec7d6dfdSNikolay Borisov BUG_ON(ret == 0); /* Corruption */ 5205ec7d6dfdSNikolay Borisov if (path->slots[0] > 0) { 5206ec7d6dfdSNikolay Borisov slot = path->slots[0] - 1; 5207ec7d6dfdSNikolay Borisov l = path->nodes[0]; 5208ec7d6dfdSNikolay Borisov btrfs_item_key_to_cpu(l, &found_key, slot); 520923125104SNikolay Borisov root->free_objectid = max_t(u64, found_key.objectid + 1, 521023125104SNikolay Borisov BTRFS_FIRST_FREE_OBJECTID); 5211ec7d6dfdSNikolay Borisov } else { 521223125104SNikolay Borisov root->free_objectid = BTRFS_FIRST_FREE_OBJECTID; 5213ec7d6dfdSNikolay Borisov } 5214ec7d6dfdSNikolay Borisov ret = 0; 5215ec7d6dfdSNikolay Borisov error: 5216ec7d6dfdSNikolay Borisov btrfs_free_path(path); 5217ec7d6dfdSNikolay Borisov return ret; 5218ec7d6dfdSNikolay Borisov } 5219ec7d6dfdSNikolay Borisov 5220543068a2SNikolay Borisov int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid) 5221ec7d6dfdSNikolay Borisov { 5222ec7d6dfdSNikolay Borisov int ret; 5223ec7d6dfdSNikolay Borisov mutex_lock(&root->objectid_mutex); 5224ec7d6dfdSNikolay Borisov 52256b8fad57SNikolay Borisov if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) { 5226ec7d6dfdSNikolay Borisov btrfs_warn(root->fs_info, 5227ec7d6dfdSNikolay Borisov "the objectid of root %llu reaches its highest value", 5228ec7d6dfdSNikolay Borisov root->root_key.objectid); 5229ec7d6dfdSNikolay Borisov ret = -ENOSPC; 5230ec7d6dfdSNikolay Borisov goto out; 5231ec7d6dfdSNikolay Borisov } 5232ec7d6dfdSNikolay Borisov 523323125104SNikolay Borisov *objectid = root->free_objectid++; 5234ec7d6dfdSNikolay Borisov ret = 0; 5235ec7d6dfdSNikolay Borisov out: 5236ec7d6dfdSNikolay Borisov mutex_unlock(&root->objectid_mutex); 5237ec7d6dfdSNikolay Borisov return ret; 5238ec7d6dfdSNikolay Borisov } 5239