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 871*190a8339SJosef Bacik void btrfs_clear_buffer_dirty(struct btrfs_trans_handle *trans, 872ed25dab3SJosef Bacik struct extent_buffer *buf) 873ed2ff2cbSChris Mason { 8746a884d7dSDavid Sterba struct btrfs_fs_info *fs_info = buf->fs_info; 875c4e54a65SJosef Bacik if (!trans || btrfs_header_generation(buf) == trans->transid) { 87649d0c642SFilipe Manana btrfs_assert_tree_write_locked(buf); 877b4ce94deSChris Mason 878b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 879104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 880e2d84521SMiao Xie -buf->len, 881e2d84521SMiao Xie fs_info->dirty_metadata_batch); 8820b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 883925baeddSChris Mason } 8845f39d397SChris Mason } 885ed7b63ebSJosef Bacik } 8865f39d397SChris Mason 887da17066cSJeff Mahoney static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, 888e20d96d6SChris Mason u64 objectid) 889d97e63b6SChris Mason { 8907c0260eeSJeff Mahoney bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); 8912e608bd1SJosef Bacik 8922e608bd1SJosef Bacik memset(&root->root_key, 0, sizeof(root->root_key)); 8932e608bd1SJosef Bacik memset(&root->root_item, 0, sizeof(root->root_item)); 8942e608bd1SJosef Bacik memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 89596dfcb46SJosef Bacik root->fs_info = fs_info; 8962e608bd1SJosef Bacik root->root_key.objectid = objectid; 897cfaa7295SChris Mason root->node = NULL; 898a28ec197SChris Mason root->commit_root = NULL; 89927cdeb70SMiao Xie root->state = 0; 900abed4aaaSJosef Bacik RB_CLEAR_NODE(&root->rb_node); 9010b86a832SChris Mason 9020f7d52f4SChris Mason root->last_trans = 0; 9036b8fad57SNikolay Borisov root->free_objectid = 0; 904eb73c1b7SMiao Xie root->nr_delalloc_inodes = 0; 905199c2a9cSMiao Xie root->nr_ordered_extents = 0; 9066bef4d31SEric Paris root->inode_tree = RB_ROOT; 907088aea3bSDavid Sterba INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 9082e608bd1SJosef Bacik 9092e608bd1SJosef Bacik btrfs_init_root_block_rsv(root); 9100b86a832SChris Mason 9110b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 9125d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 913eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_inodes); 914eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_root); 915199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_extents); 916199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_root); 917d2311e69SQu Wenruo INIT_LIST_HEAD(&root->reloc_dirty_list); 9182ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 9192ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 9205d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 921eb73c1b7SMiao Xie spin_lock_init(&root->delalloc_lock); 922199c2a9cSMiao Xie spin_lock_init(&root->ordered_extent_lock); 923f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 9242ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 9252ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 9268287475aSQu Wenruo spin_lock_init(&root->qgroup_meta_rsv_lock); 927a2135011SChris Mason mutex_init(&root->objectid_mutex); 928e02119d5SChris Mason mutex_init(&root->log_mutex); 92931f3d255SMiao Xie mutex_init(&root->ordered_extent_mutex); 930573bfb72SMiao Xie mutex_init(&root->delalloc_mutex); 931c53e9653SQu Wenruo init_waitqueue_head(&root->qgroup_flush_wait); 9327237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 9337237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 9347237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 9358b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[0]); 9368b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[1]); 9377237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 9387237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 9397237f183SYan Zheng atomic_set(&root->log_writers, 0); 9402ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 9410700cea7SElena Reshetova refcount_set(&root->refs, 1); 9428ecebf4dSRobbie Ko atomic_set(&root->snapshot_force_cow, 0); 943eede2bf3SOmar Sandoval atomic_set(&root->nr_swapfiles, 0); 9447237f183SYan Zheng root->log_transid = 0; 945d1433debSMiao Xie root->log_transid_committed = -1; 946257c62e1SChris Mason root->last_log_commit = 0; 9472e608bd1SJosef Bacik root->anon_dev = 0; 948e289f03eSFilipe Manana if (!dummy) { 94943eb5f29SQu Wenruo extent_io_tree_init(fs_info, &root->dirty_log_pages, 95035da5a7eSDavid Sterba IO_TREE_ROOT_DIRTY_LOG_PAGES); 951e289f03eSFilipe Manana extent_io_tree_init(fs_info, &root->log_csum_range, 95235da5a7eSDavid Sterba IO_TREE_LOG_CSUM_RANGE); 953e289f03eSFilipe Manana } 954017e5369SChris Mason 9555f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 956370a11b8SQu Wenruo btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks); 957bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 958bd647ce3SJosef Bacik INIT_LIST_HEAD(&root->leak_list); 959fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 960bd647ce3SJosef Bacik list_add_tail(&root->leak_list, &fs_info->allocated_roots); 961fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 962bd647ce3SJosef Bacik #endif 9633768f368SChris Mason } 9643768f368SChris Mason 96574e4d827SDavid Sterba static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, 96696dfcb46SJosef Bacik u64 objectid, gfp_t flags) 9676f07e42eSAl Viro { 96874e4d827SDavid Sterba struct btrfs_root *root = kzalloc(sizeof(*root), flags); 9696f07e42eSAl Viro if (root) 97096dfcb46SJosef Bacik __setup_root(root, fs_info, objectid); 9716f07e42eSAl Viro return root; 9726f07e42eSAl Viro } 9736f07e42eSAl Viro 97406ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 97506ea65a3SJosef Bacik /* Should only be used by the testing infrastructure */ 976da17066cSJeff Mahoney struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) 97706ea65a3SJosef Bacik { 97806ea65a3SJosef Bacik struct btrfs_root *root; 97906ea65a3SJosef Bacik 9807c0260eeSJeff Mahoney if (!fs_info) 9817c0260eeSJeff Mahoney return ERR_PTR(-EINVAL); 9827c0260eeSJeff Mahoney 98396dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL); 98406ea65a3SJosef Bacik if (!root) 98506ea65a3SJosef Bacik return ERR_PTR(-ENOMEM); 986da17066cSJeff Mahoney 987b9ef22deSFeifei Xu /* We don't use the stripesize in selftest, set it as sectorsize */ 988faa2dbf0SJosef Bacik root->alloc_bytenr = 0; 98906ea65a3SJosef Bacik 99006ea65a3SJosef Bacik return root; 99106ea65a3SJosef Bacik } 99206ea65a3SJosef Bacik #endif 99306ea65a3SJosef Bacik 994abed4aaaSJosef Bacik static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node) 995abed4aaaSJosef Bacik { 996abed4aaaSJosef Bacik const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node); 997abed4aaaSJosef Bacik const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node); 998abed4aaaSJosef Bacik 999abed4aaaSJosef Bacik return btrfs_comp_cpu_keys(&a->root_key, &b->root_key); 1000abed4aaaSJosef Bacik } 1001abed4aaaSJosef Bacik 1002abed4aaaSJosef Bacik static int global_root_key_cmp(const void *k, const struct rb_node *node) 1003abed4aaaSJosef Bacik { 1004abed4aaaSJosef Bacik const struct btrfs_key *key = k; 1005abed4aaaSJosef Bacik const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node); 1006abed4aaaSJosef Bacik 1007abed4aaaSJosef Bacik return btrfs_comp_cpu_keys(key, &root->root_key); 1008abed4aaaSJosef Bacik } 1009abed4aaaSJosef Bacik 1010abed4aaaSJosef Bacik int btrfs_global_root_insert(struct btrfs_root *root) 1011abed4aaaSJosef Bacik { 1012abed4aaaSJosef Bacik struct btrfs_fs_info *fs_info = root->fs_info; 1013abed4aaaSJosef Bacik struct rb_node *tmp; 1014abed4aaaSJosef Bacik 1015abed4aaaSJosef Bacik write_lock(&fs_info->global_root_lock); 1016abed4aaaSJosef Bacik tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp); 1017abed4aaaSJosef Bacik write_unlock(&fs_info->global_root_lock); 1018abed4aaaSJosef Bacik ASSERT(!tmp); 1019abed4aaaSJosef Bacik 1020abed4aaaSJosef Bacik return tmp ? -EEXIST : 0; 1021abed4aaaSJosef Bacik } 1022abed4aaaSJosef Bacik 1023abed4aaaSJosef Bacik void btrfs_global_root_delete(struct btrfs_root *root) 1024abed4aaaSJosef Bacik { 1025abed4aaaSJosef Bacik struct btrfs_fs_info *fs_info = root->fs_info; 1026abed4aaaSJosef Bacik 1027abed4aaaSJosef Bacik write_lock(&fs_info->global_root_lock); 1028abed4aaaSJosef Bacik rb_erase(&root->rb_node, &fs_info->global_root_tree); 1029abed4aaaSJosef Bacik write_unlock(&fs_info->global_root_lock); 1030abed4aaaSJosef Bacik } 1031abed4aaaSJosef Bacik 1032abed4aaaSJosef Bacik struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info, 1033abed4aaaSJosef Bacik struct btrfs_key *key) 1034abed4aaaSJosef Bacik { 1035abed4aaaSJosef Bacik struct rb_node *node; 1036abed4aaaSJosef Bacik struct btrfs_root *root = NULL; 1037abed4aaaSJosef Bacik 1038abed4aaaSJosef Bacik read_lock(&fs_info->global_root_lock); 1039abed4aaaSJosef Bacik node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp); 1040abed4aaaSJosef Bacik if (node) 1041abed4aaaSJosef Bacik root = container_of(node, struct btrfs_root, rb_node); 1042abed4aaaSJosef Bacik read_unlock(&fs_info->global_root_lock); 1043abed4aaaSJosef Bacik 1044abed4aaaSJosef Bacik return root; 1045abed4aaaSJosef Bacik } 1046abed4aaaSJosef Bacik 1047f7238e50SJosef Bacik static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr) 1048f7238e50SJosef Bacik { 1049f7238e50SJosef Bacik struct btrfs_block_group *block_group; 1050f7238e50SJosef Bacik u64 ret; 1051f7238e50SJosef Bacik 1052f7238e50SJosef Bacik if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) 1053f7238e50SJosef Bacik return 0; 1054f7238e50SJosef Bacik 1055f7238e50SJosef Bacik if (bytenr) 1056f7238e50SJosef Bacik block_group = btrfs_lookup_block_group(fs_info, bytenr); 1057f7238e50SJosef Bacik else 1058f7238e50SJosef Bacik block_group = btrfs_lookup_first_block_group(fs_info, bytenr); 1059f7238e50SJosef Bacik ASSERT(block_group); 1060f7238e50SJosef Bacik if (!block_group) 1061f7238e50SJosef Bacik return 0; 1062f7238e50SJosef Bacik ret = block_group->global_root_id; 1063f7238e50SJosef Bacik btrfs_put_block_group(block_group); 1064f7238e50SJosef Bacik 1065f7238e50SJosef Bacik return ret; 1066f7238e50SJosef Bacik } 1067f7238e50SJosef Bacik 1068abed4aaaSJosef Bacik struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr) 1069abed4aaaSJosef Bacik { 1070abed4aaaSJosef Bacik struct btrfs_key key = { 1071abed4aaaSJosef Bacik .objectid = BTRFS_CSUM_TREE_OBJECTID, 1072abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 1073f7238e50SJosef Bacik .offset = btrfs_global_root_id(fs_info, bytenr), 1074abed4aaaSJosef Bacik }; 1075abed4aaaSJosef Bacik 1076abed4aaaSJosef Bacik return btrfs_global_root(fs_info, &key); 1077abed4aaaSJosef Bacik } 1078abed4aaaSJosef Bacik 1079abed4aaaSJosef Bacik struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr) 1080abed4aaaSJosef Bacik { 1081abed4aaaSJosef Bacik struct btrfs_key key = { 1082abed4aaaSJosef Bacik .objectid = BTRFS_EXTENT_TREE_OBJECTID, 1083abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 1084f7238e50SJosef Bacik .offset = btrfs_global_root_id(fs_info, bytenr), 1085abed4aaaSJosef Bacik }; 1086abed4aaaSJosef Bacik 1087abed4aaaSJosef Bacik return btrfs_global_root(fs_info, &key); 1088abed4aaaSJosef Bacik } 1089abed4aaaSJosef Bacik 109051129b33SJosef Bacik struct btrfs_root *btrfs_block_group_root(struct btrfs_fs_info *fs_info) 109151129b33SJosef Bacik { 109251129b33SJosef Bacik if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) 109351129b33SJosef Bacik return fs_info->block_group_root; 109451129b33SJosef Bacik return btrfs_extent_root(fs_info, 0); 109551129b33SJosef Bacik } 109651129b33SJosef Bacik 109720897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 109820897f5cSArne Jansen u64 objectid) 109920897f5cSArne Jansen { 11009b7a2440SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 110120897f5cSArne Jansen struct extent_buffer *leaf; 110220897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 110320897f5cSArne Jansen struct btrfs_root *root; 110420897f5cSArne Jansen struct btrfs_key key; 1105b89f6d1fSFilipe Manana unsigned int nofs_flag; 110620897f5cSArne Jansen int ret = 0; 110720897f5cSArne Jansen 1108b89f6d1fSFilipe Manana /* 1109b89f6d1fSFilipe Manana * We're holding a transaction handle, so use a NOFS memory allocation 1110b89f6d1fSFilipe Manana * context to avoid deadlock if reclaim happens. 1111b89f6d1fSFilipe Manana */ 1112b89f6d1fSFilipe Manana nofs_flag = memalloc_nofs_save(); 111396dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL); 1114b89f6d1fSFilipe Manana memalloc_nofs_restore(nofs_flag); 111520897f5cSArne Jansen if (!root) 111620897f5cSArne Jansen return ERR_PTR(-ENOMEM); 111720897f5cSArne Jansen 111820897f5cSArne Jansen root->root_key.objectid = objectid; 111920897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 112020897f5cSArne Jansen root->root_key.offset = 0; 112120897f5cSArne Jansen 11229631e4ccSJosef Bacik leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0, 11239631e4ccSJosef Bacik BTRFS_NESTING_NORMAL); 112420897f5cSArne Jansen if (IS_ERR(leaf)) { 112520897f5cSArne Jansen ret = PTR_ERR(leaf); 11261dd05682STsutomu Itoh leaf = NULL; 1127c1b07854SPeng Hao goto fail; 112820897f5cSArne Jansen } 112920897f5cSArne Jansen 113020897f5cSArne Jansen root->node = leaf; 113120897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 113220897f5cSArne Jansen 113320897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 113427cdeb70SMiao Xie set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 113520897f5cSArne Jansen 1136f944d2cbSDavid Sterba btrfs_set_root_flags(&root->root_item, 0); 1137f944d2cbSDavid Sterba btrfs_set_root_limit(&root->root_item, 0); 113820897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 113920897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 114020897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 114120897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 114220897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 114320897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 114420897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 114533d85fdaSQu Wenruo if (is_fstree(objectid)) 1146807fc790SAndy Shevchenko generate_random_guid(root->root_item.uuid); 1147807fc790SAndy Shevchenko else 1148807fc790SAndy Shevchenko export_guid(root->root_item.uuid, &guid_null); 1149c8422684SDavid Sterba btrfs_set_root_drop_level(&root->root_item, 0); 115020897f5cSArne Jansen 11518a6a87cdSBoris Burkov btrfs_tree_unlock(leaf); 11528a6a87cdSBoris Burkov 115320897f5cSArne Jansen key.objectid = objectid; 115420897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 115520897f5cSArne Jansen key.offset = 0; 115620897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 115720897f5cSArne Jansen if (ret) 115820897f5cSArne Jansen goto fail; 115920897f5cSArne Jansen 116020897f5cSArne Jansen return root; 11611dd05682STsutomu Itoh 11628a6a87cdSBoris Burkov fail: 116300246528SJosef Bacik btrfs_put_root(root); 11641dd05682STsutomu Itoh 11651dd05682STsutomu Itoh return ERR_PTR(ret); 116620897f5cSArne Jansen } 116720897f5cSArne Jansen 11687237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1169e02119d5SChris Mason struct btrfs_fs_info *fs_info) 11700f7d52f4SChris Mason { 11710f7d52f4SChris Mason struct btrfs_root *root; 1172e02119d5SChris Mason 117396dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS); 1174e02119d5SChris Mason if (!root) 11757237f183SYan Zheng return ERR_PTR(-ENOMEM); 1176e02119d5SChris Mason 1177e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1178e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1179e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 118027cdeb70SMiao Xie 11816ab6ebb7SNaohiro Aota return root; 11826ab6ebb7SNaohiro Aota } 11836ab6ebb7SNaohiro Aota 11846ab6ebb7SNaohiro Aota int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans, 11856ab6ebb7SNaohiro Aota struct btrfs_root *root) 11866ab6ebb7SNaohiro Aota { 11876ab6ebb7SNaohiro Aota struct extent_buffer *leaf; 11886ab6ebb7SNaohiro Aota 11897237f183SYan Zheng /* 119092a7cc42SQu Wenruo * DON'T set SHAREABLE bit for log trees. 119127cdeb70SMiao Xie * 119292a7cc42SQu Wenruo * Log trees are not exposed to user space thus can't be snapshotted, 119392a7cc42SQu Wenruo * and they go away before a real commit is actually done. 119492a7cc42SQu Wenruo * 119592a7cc42SQu Wenruo * They do store pointers to file data extents, and those reference 119692a7cc42SQu Wenruo * counts still get updated (along with back refs to the log tree). 11977237f183SYan Zheng */ 1198e02119d5SChris Mason 11994d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 12009631e4ccSJosef Bacik NULL, 0, 0, 0, BTRFS_NESTING_NORMAL); 12016ab6ebb7SNaohiro Aota if (IS_ERR(leaf)) 12026ab6ebb7SNaohiro Aota return PTR_ERR(leaf); 1203e02119d5SChris Mason 12047237f183SYan Zheng root->node = leaf; 1205e02119d5SChris Mason 1206e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1207e02119d5SChris Mason btrfs_tree_unlock(root->node); 12086ab6ebb7SNaohiro Aota 12096ab6ebb7SNaohiro Aota return 0; 12107237f183SYan Zheng } 12117237f183SYan Zheng 12127237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 12137237f183SYan Zheng struct btrfs_fs_info *fs_info) 12147237f183SYan Zheng { 12157237f183SYan Zheng struct btrfs_root *log_root; 12167237f183SYan Zheng 12177237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 12187237f183SYan Zheng if (IS_ERR(log_root)) 12197237f183SYan Zheng return PTR_ERR(log_root); 12206ab6ebb7SNaohiro Aota 12213ddebf27SNaohiro Aota if (!btrfs_is_zoned(fs_info)) { 12223ddebf27SNaohiro Aota int ret = btrfs_alloc_log_tree_node(trans, log_root); 12233ddebf27SNaohiro Aota 12246ab6ebb7SNaohiro Aota if (ret) { 12256ab6ebb7SNaohiro Aota btrfs_put_root(log_root); 12266ab6ebb7SNaohiro Aota return ret; 12276ab6ebb7SNaohiro Aota } 12283ddebf27SNaohiro Aota } 12296ab6ebb7SNaohiro Aota 12307237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 12317237f183SYan Zheng fs_info->log_root_tree = log_root; 12327237f183SYan Zheng return 0; 12337237f183SYan Zheng } 12347237f183SYan Zheng 12357237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 12367237f183SYan Zheng struct btrfs_root *root) 12377237f183SYan Zheng { 12380b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12397237f183SYan Zheng struct btrfs_root *log_root; 12407237f183SYan Zheng struct btrfs_inode_item *inode_item; 12416ab6ebb7SNaohiro Aota int ret; 12427237f183SYan Zheng 12430b246afaSJeff Mahoney log_root = alloc_log_tree(trans, fs_info); 12447237f183SYan Zheng if (IS_ERR(log_root)) 12457237f183SYan Zheng return PTR_ERR(log_root); 12467237f183SYan Zheng 12476ab6ebb7SNaohiro Aota ret = btrfs_alloc_log_tree_node(trans, log_root); 12486ab6ebb7SNaohiro Aota if (ret) { 12496ab6ebb7SNaohiro Aota btrfs_put_root(log_root); 12506ab6ebb7SNaohiro Aota return ret; 12516ab6ebb7SNaohiro Aota } 12526ab6ebb7SNaohiro Aota 12537237f183SYan Zheng log_root->last_trans = trans->transid; 12547237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 12557237f183SYan Zheng 12567237f183SYan Zheng inode_item = &log_root->root_item.inode; 12573cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 12583cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 12593cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 1260da17066cSJeff Mahoney btrfs_set_stack_inode_nbytes(inode_item, 12610b246afaSJeff Mahoney fs_info->nodesize); 12623cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 12637237f183SYan Zheng 12645d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 12657237f183SYan Zheng 12667237f183SYan Zheng WARN_ON(root->log_root); 12677237f183SYan Zheng root->log_root = log_root; 12687237f183SYan Zheng root->log_transid = 0; 1269d1433debSMiao Xie root->log_transid_committed = -1; 1270257c62e1SChris Mason root->last_log_commit = 0; 1271e02119d5SChris Mason return 0; 1272e02119d5SChris Mason } 1273e02119d5SChris Mason 127449d11beaSJosef Bacik static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root, 127549d11beaSJosef Bacik struct btrfs_path *path, 1276cb517eabSMiao Xie struct btrfs_key *key) 1277e02119d5SChris Mason { 1278e02119d5SChris Mason struct btrfs_root *root; 1279789d6a3aSQu Wenruo struct btrfs_tree_parent_check check = { 0 }; 1280e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 128184234f3aSYan Zheng u64 generation; 1282cb517eabSMiao Xie int ret; 1283581c1760SQu Wenruo int level; 1284cb517eabSMiao Xie 128596dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS); 128649d11beaSJosef Bacik if (!root) 128749d11beaSJosef Bacik return ERR_PTR(-ENOMEM); 12880f7d52f4SChris Mason 1289cb517eabSMiao Xie ret = btrfs_find_root(tree_root, key, path, 1290cb517eabSMiao Xie &root->root_item, &root->root_key); 12910f7d52f4SChris Mason if (ret) { 129213a8a7c8SYan, Zheng if (ret > 0) 129313a8a7c8SYan, Zheng ret = -ENOENT; 129449d11beaSJosef Bacik goto fail; 12950f7d52f4SChris Mason } 129613a8a7c8SYan, Zheng 129784234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1298581c1760SQu Wenruo level = btrfs_root_level(&root->root_item); 1299789d6a3aSQu Wenruo check.level = level; 1300789d6a3aSQu Wenruo check.transid = generation; 1301789d6a3aSQu Wenruo check.owner_root = key->objectid; 1302789d6a3aSQu Wenruo root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item), 1303789d6a3aSQu Wenruo &check); 130464c043deSLiu Bo if (IS_ERR(root->node)) { 130564c043deSLiu Bo ret = PTR_ERR(root->node); 13068c38938cSJosef Bacik root->node = NULL; 130749d11beaSJosef Bacik goto fail; 13084eb150d6SQu Wenruo } 13094eb150d6SQu Wenruo if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1310cb517eabSMiao Xie ret = -EIO; 131149d11beaSJosef Bacik goto fail; 1312416bc658SJosef Bacik } 131388c602abSQu Wenruo 131488c602abSQu Wenruo /* 131588c602abSQu Wenruo * For real fs, and not log/reloc trees, root owner must 131688c602abSQu Wenruo * match its root node owner 131788c602abSQu Wenruo */ 131888c602abSQu Wenruo if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) && 131988c602abSQu Wenruo root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID && 132088c602abSQu Wenruo root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && 132188c602abSQu Wenruo root->root_key.objectid != btrfs_header_owner(root->node)) { 132288c602abSQu Wenruo btrfs_crit(fs_info, 132388c602abSQu Wenruo "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu", 132488c602abSQu Wenruo root->root_key.objectid, root->node->start, 132588c602abSQu Wenruo btrfs_header_owner(root->node), 132688c602abSQu Wenruo root->root_key.objectid); 132788c602abSQu Wenruo ret = -EUCLEAN; 132888c602abSQu Wenruo goto fail; 132988c602abSQu Wenruo } 13305d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 1331cb517eabSMiao Xie return root; 133249d11beaSJosef Bacik fail: 133300246528SJosef Bacik btrfs_put_root(root); 133449d11beaSJosef Bacik return ERR_PTR(ret); 133549d11beaSJosef Bacik } 133649d11beaSJosef Bacik 133749d11beaSJosef Bacik struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 133849d11beaSJosef Bacik struct btrfs_key *key) 133949d11beaSJosef Bacik { 134049d11beaSJosef Bacik struct btrfs_root *root; 134149d11beaSJosef Bacik struct btrfs_path *path; 134249d11beaSJosef Bacik 134349d11beaSJosef Bacik path = btrfs_alloc_path(); 134449d11beaSJosef Bacik if (!path) 134549d11beaSJosef Bacik return ERR_PTR(-ENOMEM); 134649d11beaSJosef Bacik root = read_tree_root_path(tree_root, path, key); 134749d11beaSJosef Bacik btrfs_free_path(path); 134849d11beaSJosef Bacik 134949d11beaSJosef Bacik return root; 1350cb517eabSMiao Xie } 1351cb517eabSMiao Xie 13522dfb1e43SQu Wenruo /* 13532dfb1e43SQu Wenruo * Initialize subvolume root in-memory structure 13542dfb1e43SQu Wenruo * 13552dfb1e43SQu Wenruo * @anon_dev: anonymous device to attach to the root, if zero, allocate new 13562dfb1e43SQu Wenruo */ 13572dfb1e43SQu Wenruo static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev) 1358cb517eabSMiao Xie { 1359cb517eabSMiao Xie int ret; 1360dcc3eb96SNikolay Borisov unsigned int nofs_flag; 1361cb517eabSMiao Xie 1362dcc3eb96SNikolay Borisov /* 1363dcc3eb96SNikolay Borisov * We might be called under a transaction (e.g. indirect backref 1364dcc3eb96SNikolay Borisov * resolution) which could deadlock if it triggers memory reclaim 1365dcc3eb96SNikolay Borisov */ 1366dcc3eb96SNikolay Borisov nofs_flag = memalloc_nofs_save(); 1367dcc3eb96SNikolay Borisov ret = btrfs_drew_lock_init(&root->snapshot_lock); 1368dcc3eb96SNikolay Borisov memalloc_nofs_restore(nofs_flag); 1369dcc3eb96SNikolay Borisov if (ret) 13708257b2dcSMiao Xie goto fail; 13718257b2dcSMiao Xie 1372aeb935a4SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID && 137337f00a6dSJohannes Thumshirn !btrfs_is_data_reloc_root(root)) { 137492a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &root->state); 1375f39e4571SJosef Bacik btrfs_check_and_init_root_item(&root->root_item); 1376f39e4571SJosef Bacik } 1377f39e4571SJosef Bacik 1378851fd730SQu Wenruo /* 1379851fd730SQu Wenruo * Don't assign anonymous block device to roots that are not exposed to 1380851fd730SQu Wenruo * userspace, the id pool is limited to 1M 1381851fd730SQu Wenruo */ 1382851fd730SQu Wenruo if (is_fstree(root->root_key.objectid) && 1383851fd730SQu Wenruo btrfs_root_refs(&root->root_item) > 0) { 13842dfb1e43SQu Wenruo if (!anon_dev) { 1385cb517eabSMiao Xie ret = get_anon_bdev(&root->anon_dev); 1386cb517eabSMiao Xie if (ret) 1387876d2cf1SLiu Bo goto fail; 13882dfb1e43SQu Wenruo } else { 13892dfb1e43SQu Wenruo root->anon_dev = anon_dev; 13902dfb1e43SQu Wenruo } 1391851fd730SQu Wenruo } 1392f32e48e9SChandan Rajendra 1393f32e48e9SChandan Rajendra mutex_lock(&root->objectid_mutex); 1394453e4873SNikolay Borisov ret = btrfs_init_root_free_objectid(root); 1395f32e48e9SChandan Rajendra if (ret) { 1396f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1397876d2cf1SLiu Bo goto fail; 1398f32e48e9SChandan Rajendra } 1399f32e48e9SChandan Rajendra 14006b8fad57SNikolay Borisov ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID); 1401f32e48e9SChandan Rajendra 1402f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1403f32e48e9SChandan Rajendra 1404cb517eabSMiao Xie return 0; 1405cb517eabSMiao Xie fail: 140684db5ccfSDavid Sterba /* The caller is responsible to call btrfs_free_fs_root */ 1407cb517eabSMiao Xie return ret; 1408cb517eabSMiao Xie } 1409cb517eabSMiao Xie 1410a98db0f3SJosef Bacik static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1411cb517eabSMiao Xie u64 root_id) 1412cb517eabSMiao Xie { 1413cb517eabSMiao Xie struct btrfs_root *root; 1414cb517eabSMiao Xie 1415fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1416fc7cbcd4SDavid Sterba root = radix_tree_lookup(&fs_info->fs_roots_radix, 1417fc7cbcd4SDavid Sterba (unsigned long)root_id); 1418bc44d7c4SJosef Bacik if (root) 141900246528SJosef Bacik root = btrfs_grab_root(root); 1420fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 1421cb517eabSMiao Xie return root; 1422cb517eabSMiao Xie } 1423cb517eabSMiao Xie 142449d11beaSJosef Bacik static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info, 142549d11beaSJosef Bacik u64 objectid) 142649d11beaSJosef Bacik { 1427abed4aaaSJosef Bacik struct btrfs_key key = { 1428abed4aaaSJosef Bacik .objectid = objectid, 1429abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 1430abed4aaaSJosef Bacik .offset = 0, 1431abed4aaaSJosef Bacik }; 1432abed4aaaSJosef Bacik 143349d11beaSJosef Bacik if (objectid == BTRFS_ROOT_TREE_OBJECTID) 143449d11beaSJosef Bacik return btrfs_grab_root(fs_info->tree_root); 143549d11beaSJosef Bacik if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 1436abed4aaaSJosef Bacik return btrfs_grab_root(btrfs_global_root(fs_info, &key)); 143749d11beaSJosef Bacik if (objectid == BTRFS_CHUNK_TREE_OBJECTID) 143849d11beaSJosef Bacik return btrfs_grab_root(fs_info->chunk_root); 143949d11beaSJosef Bacik if (objectid == BTRFS_DEV_TREE_OBJECTID) 144049d11beaSJosef Bacik return btrfs_grab_root(fs_info->dev_root); 144149d11beaSJosef Bacik if (objectid == BTRFS_CSUM_TREE_OBJECTID) 1442abed4aaaSJosef Bacik return btrfs_grab_root(btrfs_global_root(fs_info, &key)); 144349d11beaSJosef Bacik if (objectid == BTRFS_QUOTA_TREE_OBJECTID) 144449d11beaSJosef Bacik return btrfs_grab_root(fs_info->quota_root) ? 144549d11beaSJosef Bacik fs_info->quota_root : ERR_PTR(-ENOENT); 144649d11beaSJosef Bacik if (objectid == BTRFS_UUID_TREE_OBJECTID) 144749d11beaSJosef Bacik return btrfs_grab_root(fs_info->uuid_root) ? 144849d11beaSJosef Bacik fs_info->uuid_root : ERR_PTR(-ENOENT); 144914033b08SQu Wenruo if (objectid == BTRFS_BLOCK_GROUP_TREE_OBJECTID) 145014033b08SQu Wenruo return btrfs_grab_root(fs_info->block_group_root) ? 145114033b08SQu Wenruo fs_info->block_group_root : ERR_PTR(-ENOENT); 1452abed4aaaSJosef Bacik if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) { 1453abed4aaaSJosef Bacik struct btrfs_root *root = btrfs_global_root(fs_info, &key); 1454abed4aaaSJosef Bacik 1455abed4aaaSJosef Bacik return btrfs_grab_root(root) ? root : ERR_PTR(-ENOENT); 1456abed4aaaSJosef Bacik } 145749d11beaSJosef Bacik return NULL; 145849d11beaSJosef Bacik } 145949d11beaSJosef Bacik 1460cb517eabSMiao Xie int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1461cb517eabSMiao Xie struct btrfs_root *root) 1462cb517eabSMiao Xie { 1463cb517eabSMiao Xie int ret; 1464cb517eabSMiao Xie 1465fc7cbcd4SDavid Sterba ret = radix_tree_preload(GFP_NOFS); 1466fc7cbcd4SDavid Sterba if (ret) 1467fc7cbcd4SDavid Sterba return ret; 1468fc7cbcd4SDavid Sterba 1469fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1470fc7cbcd4SDavid Sterba ret = radix_tree_insert(&fs_info->fs_roots_radix, 1471fc7cbcd4SDavid Sterba (unsigned long)root->root_key.objectid, 1472fc7cbcd4SDavid Sterba root); 1473af01d2e5SJosef Bacik if (ret == 0) { 147400246528SJosef Bacik btrfs_grab_root(root); 1475fc7cbcd4SDavid Sterba set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1476af01d2e5SJosef Bacik } 1477fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 1478fc7cbcd4SDavid Sterba radix_tree_preload_end(); 1479cb517eabSMiao Xie 1480cb517eabSMiao Xie return ret; 1481cb517eabSMiao Xie } 1482cb517eabSMiao Xie 1483bd647ce3SJosef Bacik void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info) 1484bd647ce3SJosef Bacik { 1485bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 1486bd647ce3SJosef Bacik struct btrfs_root *root; 1487bd647ce3SJosef Bacik 1488bd647ce3SJosef Bacik while (!list_empty(&fs_info->allocated_roots)) { 1489457f1864SJosef Bacik char buf[BTRFS_ROOT_NAME_BUF_LEN]; 1490457f1864SJosef Bacik 1491bd647ce3SJosef Bacik root = list_first_entry(&fs_info->allocated_roots, 1492bd647ce3SJosef Bacik struct btrfs_root, leak_list); 1493457f1864SJosef Bacik btrfs_err(fs_info, "leaked root %s refcount %d", 149471008734SJosef Bacik btrfs_root_name(&root->root_key, buf), 1495bd647ce3SJosef Bacik refcount_read(&root->refs)); 1496bd647ce3SJosef Bacik while (refcount_read(&root->refs) > 1) 149700246528SJosef Bacik btrfs_put_root(root); 149800246528SJosef Bacik btrfs_put_root(root); 1499bd647ce3SJosef Bacik } 1500bd647ce3SJosef Bacik #endif 1501bd647ce3SJosef Bacik } 1502bd647ce3SJosef Bacik 1503abed4aaaSJosef Bacik static void free_global_roots(struct btrfs_fs_info *fs_info) 1504abed4aaaSJosef Bacik { 1505abed4aaaSJosef Bacik struct btrfs_root *root; 1506abed4aaaSJosef Bacik struct rb_node *node; 1507abed4aaaSJosef Bacik 1508abed4aaaSJosef Bacik while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) { 1509abed4aaaSJosef Bacik root = rb_entry(node, struct btrfs_root, rb_node); 1510abed4aaaSJosef Bacik rb_erase(&root->rb_node, &fs_info->global_root_tree); 1511abed4aaaSJosef Bacik btrfs_put_root(root); 1512abed4aaaSJosef Bacik } 1513abed4aaaSJosef Bacik } 1514abed4aaaSJosef Bacik 15150d4b0463SJosef Bacik void btrfs_free_fs_info(struct btrfs_fs_info *fs_info) 15160d4b0463SJosef Bacik { 1517141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 1518141386e1SJosef Bacik percpu_counter_destroy(&fs_info->delalloc_bytes); 15195deb17e1SJosef Bacik percpu_counter_destroy(&fs_info->ordered_bytes); 1520141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 1521141386e1SJosef Bacik btrfs_free_csum_hash(fs_info); 1522141386e1SJosef Bacik btrfs_free_stripe_hash_table(fs_info); 1523141386e1SJosef Bacik btrfs_free_ref_cache(fs_info); 15240d4b0463SJosef Bacik kfree(fs_info->balance_ctl); 15250d4b0463SJosef Bacik kfree(fs_info->delayed_root); 1526abed4aaaSJosef Bacik free_global_roots(fs_info); 152700246528SJosef Bacik btrfs_put_root(fs_info->tree_root); 152800246528SJosef Bacik btrfs_put_root(fs_info->chunk_root); 152900246528SJosef Bacik btrfs_put_root(fs_info->dev_root); 153000246528SJosef Bacik btrfs_put_root(fs_info->quota_root); 153100246528SJosef Bacik btrfs_put_root(fs_info->uuid_root); 153200246528SJosef Bacik btrfs_put_root(fs_info->fs_root); 1533aeb935a4SQu Wenruo btrfs_put_root(fs_info->data_reloc_root); 15349c54e80dSJosef Bacik btrfs_put_root(fs_info->block_group_root); 1535bd647ce3SJosef Bacik btrfs_check_leaked_roots(fs_info); 15363fd63727SJosef Bacik btrfs_extent_buffer_leak_debug_check(fs_info); 15370d4b0463SJosef Bacik kfree(fs_info->super_copy); 15380d4b0463SJosef Bacik kfree(fs_info->super_for_commit); 15398481dd80SQu Wenruo kfree(fs_info->subpage_info); 15400d4b0463SJosef Bacik kvfree(fs_info); 15410d4b0463SJosef Bacik } 15420d4b0463SJosef Bacik 15430d4b0463SJosef Bacik 15442dfb1e43SQu Wenruo /* 15452dfb1e43SQu Wenruo * Get an in-memory reference of a root structure. 15462dfb1e43SQu Wenruo * 15472dfb1e43SQu Wenruo * For essential trees like root/extent tree, we grab it from fs_info directly. 15482dfb1e43SQu Wenruo * For subvolume trees, we check the cached filesystem roots first. If not 15492dfb1e43SQu Wenruo * found, then read it from disk and add it to cached fs roots. 15502dfb1e43SQu Wenruo * 15512dfb1e43SQu Wenruo * Caller should release the root by calling btrfs_put_root() after the usage. 15522dfb1e43SQu Wenruo * 15532dfb1e43SQu Wenruo * NOTE: Reloc and log trees can't be read by this function as they share the 15542dfb1e43SQu Wenruo * same root objectid. 15552dfb1e43SQu Wenruo * 15562dfb1e43SQu Wenruo * @objectid: root id 15572dfb1e43SQu Wenruo * @anon_dev: preallocated anonymous block device number for new roots, 15582dfb1e43SQu Wenruo * pass 0 for new allocation. 15592dfb1e43SQu Wenruo * @check_ref: whether to check root item references, If true, return -ENOENT 15602dfb1e43SQu Wenruo * for orphan roots 15612dfb1e43SQu Wenruo */ 15622dfb1e43SQu Wenruo static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info, 15632dfb1e43SQu Wenruo u64 objectid, dev_t anon_dev, 15642dfb1e43SQu Wenruo bool check_ref) 15655eda7b5eSChris Mason { 15665eda7b5eSChris Mason struct btrfs_root *root; 1567381cf658SDavid Sterba struct btrfs_path *path; 15681d4c08e0SDavid Sterba struct btrfs_key key; 15695eda7b5eSChris Mason int ret; 15705eda7b5eSChris Mason 157149d11beaSJosef Bacik root = btrfs_get_global_root(fs_info, objectid); 157249d11beaSJosef Bacik if (root) 157349d11beaSJosef Bacik return root; 15744df27c4dSYan, Zheng again: 157556e9357aSDavid Sterba root = btrfs_lookup_fs_root(fs_info, objectid); 157648475471SStefan Behrens if (root) { 15772dfb1e43SQu Wenruo /* Shouldn't get preallocated anon_dev for cached roots */ 15782dfb1e43SQu Wenruo ASSERT(!anon_dev); 1579bc44d7c4SJosef Bacik if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 158000246528SJosef Bacik btrfs_put_root(root); 158148475471SStefan Behrens return ERR_PTR(-ENOENT); 1582bc44d7c4SJosef Bacik } 15835eda7b5eSChris Mason return root; 158448475471SStefan Behrens } 15855eda7b5eSChris Mason 158656e9357aSDavid Sterba key.objectid = objectid; 158756e9357aSDavid Sterba key.type = BTRFS_ROOT_ITEM_KEY; 158856e9357aSDavid Sterba key.offset = (u64)-1; 158956e9357aSDavid Sterba root = btrfs_read_tree_root(fs_info->tree_root, &key); 15905eda7b5eSChris Mason if (IS_ERR(root)) 15915eda7b5eSChris Mason return root; 15923394e160SChris Mason 1593c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1594d68fc57bSYan, Zheng ret = -ENOENT; 1595d68fc57bSYan, Zheng goto fail; 1596d68fc57bSYan, Zheng } 1597d68fc57bSYan, Zheng 15982dfb1e43SQu Wenruo ret = btrfs_init_fs_root(root, anon_dev); 1599cb517eabSMiao Xie if (ret) 1600cb517eabSMiao Xie goto fail; 1601cb517eabSMiao Xie 1602381cf658SDavid Sterba path = btrfs_alloc_path(); 1603381cf658SDavid Sterba if (!path) { 1604381cf658SDavid Sterba ret = -ENOMEM; 1605381cf658SDavid Sterba goto fail; 1606381cf658SDavid Sterba } 16071d4c08e0SDavid Sterba key.objectid = BTRFS_ORPHAN_OBJECTID; 16081d4c08e0SDavid Sterba key.type = BTRFS_ORPHAN_ITEM_KEY; 160956e9357aSDavid Sterba key.offset = objectid; 16101d4c08e0SDavid Sterba 16111d4c08e0SDavid Sterba ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1612381cf658SDavid Sterba btrfs_free_path(path); 1613d68fc57bSYan, Zheng if (ret < 0) 1614d68fc57bSYan, Zheng goto fail; 1615d68fc57bSYan, Zheng if (ret == 0) 161627cdeb70SMiao Xie set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1617d68fc57bSYan, Zheng 1618cb517eabSMiao Xie ret = btrfs_insert_fs_root(fs_info, root); 16190f7d52f4SChris Mason if (ret) { 1620168a2f77SJia-Ju Bai if (ret == -EEXIST) { 162100246528SJosef Bacik btrfs_put_root(root); 16224df27c4dSYan, Zheng goto again; 1623168a2f77SJia-Ju Bai } 16244df27c4dSYan, Zheng goto fail; 16254df27c4dSYan, Zheng } 1626edbd8d4eSChris Mason return root; 16274df27c4dSYan, Zheng fail: 162833fab972SFilipe Manana /* 162933fab972SFilipe Manana * If our caller provided us an anonymous device, then it's his 1630143823cfSDavid Sterba * responsibility to free it in case we fail. So we have to set our 163133fab972SFilipe Manana * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root() 163233fab972SFilipe Manana * and once again by our caller. 163333fab972SFilipe Manana */ 163433fab972SFilipe Manana if (anon_dev) 163533fab972SFilipe Manana root->anon_dev = 0; 16368c38938cSJosef Bacik btrfs_put_root(root); 16374df27c4dSYan, Zheng return ERR_PTR(ret); 1638edbd8d4eSChris Mason } 1639edbd8d4eSChris Mason 16402dfb1e43SQu Wenruo /* 16412dfb1e43SQu Wenruo * Get in-memory reference of a root structure 16422dfb1e43SQu Wenruo * 16432dfb1e43SQu Wenruo * @objectid: tree objectid 16442dfb1e43SQu Wenruo * @check_ref: if set, verify that the tree exists and the item has at least 16452dfb1e43SQu Wenruo * one reference 16462dfb1e43SQu Wenruo */ 16472dfb1e43SQu Wenruo struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 16482dfb1e43SQu Wenruo u64 objectid, bool check_ref) 16492dfb1e43SQu Wenruo { 16502dfb1e43SQu Wenruo return btrfs_get_root_ref(fs_info, objectid, 0, check_ref); 16512dfb1e43SQu Wenruo } 16522dfb1e43SQu Wenruo 16532dfb1e43SQu Wenruo /* 16542dfb1e43SQu Wenruo * Get in-memory reference of a root structure, created as new, optionally pass 16552dfb1e43SQu Wenruo * the anonymous block device id 16562dfb1e43SQu Wenruo * 16572dfb1e43SQu Wenruo * @objectid: tree objectid 16582dfb1e43SQu Wenruo * @anon_dev: if zero, allocate a new anonymous block device or use the 16592dfb1e43SQu Wenruo * parameter value 16602dfb1e43SQu Wenruo */ 16612dfb1e43SQu Wenruo struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info, 16622dfb1e43SQu Wenruo u64 objectid, dev_t anon_dev) 16632dfb1e43SQu Wenruo { 16642dfb1e43SQu Wenruo return btrfs_get_root_ref(fs_info, objectid, anon_dev, true); 16652dfb1e43SQu Wenruo } 16662dfb1e43SQu Wenruo 16678b712842SChris Mason /* 166849d11beaSJosef Bacik * btrfs_get_fs_root_commit_root - return a root for the given objectid 166949d11beaSJosef Bacik * @fs_info: the fs_info 167049d11beaSJosef Bacik * @objectid: the objectid we need to lookup 167149d11beaSJosef Bacik * 167249d11beaSJosef Bacik * This is exclusively used for backref walking, and exists specifically because 167349d11beaSJosef Bacik * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref 167449d11beaSJosef Bacik * creation time, which means we may have to read the tree_root in order to look 167549d11beaSJosef Bacik * up a fs root that is not in memory. If the root is not in memory we will 167649d11beaSJosef Bacik * read the tree root commit root and look up the fs root from there. This is a 167749d11beaSJosef Bacik * temporary root, it will not be inserted into the radix tree as it doesn't 167849d11beaSJosef Bacik * have the most uptodate information, it'll simply be discarded once the 167949d11beaSJosef Bacik * backref code is finished using the root. 168049d11beaSJosef Bacik */ 168149d11beaSJosef Bacik struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info, 168249d11beaSJosef Bacik struct btrfs_path *path, 168349d11beaSJosef Bacik u64 objectid) 168449d11beaSJosef Bacik { 168549d11beaSJosef Bacik struct btrfs_root *root; 168649d11beaSJosef Bacik struct btrfs_key key; 168749d11beaSJosef Bacik 168849d11beaSJosef Bacik ASSERT(path->search_commit_root && path->skip_locking); 168949d11beaSJosef Bacik 169049d11beaSJosef Bacik /* 169149d11beaSJosef Bacik * This can return -ENOENT if we ask for a root that doesn't exist, but 169249d11beaSJosef Bacik * since this is called via the backref walking code we won't be looking 169349d11beaSJosef Bacik * up a root that doesn't exist, unless there's corruption. So if root 169449d11beaSJosef Bacik * != NULL just return it. 169549d11beaSJosef Bacik */ 169649d11beaSJosef Bacik root = btrfs_get_global_root(fs_info, objectid); 169749d11beaSJosef Bacik if (root) 169849d11beaSJosef Bacik return root; 169949d11beaSJosef Bacik 170049d11beaSJosef Bacik root = btrfs_lookup_fs_root(fs_info, objectid); 170149d11beaSJosef Bacik if (root) 170249d11beaSJosef Bacik return root; 170349d11beaSJosef Bacik 170449d11beaSJosef Bacik key.objectid = objectid; 170549d11beaSJosef Bacik key.type = BTRFS_ROOT_ITEM_KEY; 170649d11beaSJosef Bacik key.offset = (u64)-1; 170749d11beaSJosef Bacik root = read_tree_root_path(fs_info->tree_root, path, &key); 170849d11beaSJosef Bacik btrfs_release_path(path); 170949d11beaSJosef Bacik 171049d11beaSJosef Bacik return root; 171149d11beaSJosef Bacik } 171249d11beaSJosef Bacik 1713a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1714a74a4b97SChris Mason { 17150d031dc4SYu Zhe struct btrfs_fs_info *fs_info = arg; 1716d0278245SMiao Xie int again; 1717a74a4b97SChris Mason 1718d6fd0ae2SOmar Sandoval while (1) { 1719d0278245SMiao Xie again = 0; 17209d1a2a3aSDavid Sterba 1721fd340d0fSJosef Bacik set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1722fd340d0fSJosef Bacik 1723d0278245SMiao Xie /* Make the cleaner go to sleep early. */ 17242ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) 1725d0278245SMiao Xie goto sleep; 1726d0278245SMiao Xie 172790c711abSZygo Blaxell /* 172890c711abSZygo Blaxell * Do not do anything if we might cause open_ctree() to block 172990c711abSZygo Blaxell * before we have finished mounting the filesystem. 173090c711abSZygo Blaxell */ 17310b246afaSJeff Mahoney if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 173290c711abSZygo Blaxell goto sleep; 173390c711abSZygo Blaxell 17340b246afaSJeff Mahoney if (!mutex_trylock(&fs_info->cleaner_mutex)) 1735d0278245SMiao Xie goto sleep; 1736d0278245SMiao Xie 1737dc7f370cSMiao Xie /* 1738dc7f370cSMiao Xie * Avoid the problem that we change the status of the fs 1739dc7f370cSMiao Xie * during the above check and trylock. 1740dc7f370cSMiao Xie */ 17412ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) { 17420b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1743dc7f370cSMiao Xie goto sleep; 1744dc7f370cSMiao Xie } 1745dc7f370cSMiao Xie 1746b7625f46SQu Wenruo if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags)) 1747b7625f46SQu Wenruo btrfs_sysfs_feature_update(fs_info); 1748b7625f46SQu Wenruo 17492ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 1750c2d6cb16SFilipe Manana 175133c44184SJosef Bacik again = btrfs_clean_one_deleted_snapshot(fs_info); 17520b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1753a74a4b97SChris Mason 1754d0278245SMiao Xie /* 175505323cd1SMiao Xie * The defragger has dealt with the R/O remount and umount, 175605323cd1SMiao Xie * needn't do anything special here. 1757d0278245SMiao Xie */ 17580b246afaSJeff Mahoney btrfs_run_defrag_inodes(fs_info); 175967c5e7d4SFilipe Manana 176067c5e7d4SFilipe Manana /* 1761f3372065SJohannes Thumshirn * Acquires fs_info->reclaim_bgs_lock to avoid racing 176267c5e7d4SFilipe Manana * with relocation (btrfs_relocate_chunk) and relocation 176367c5e7d4SFilipe Manana * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 1764f3372065SJohannes Thumshirn * after acquiring fs_info->reclaim_bgs_lock. So we 176567c5e7d4SFilipe Manana * can't hold, nor need to, fs_info->cleaner_mutex when deleting 176667c5e7d4SFilipe Manana * unused block groups. 176767c5e7d4SFilipe Manana */ 17680b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 176918bb8bbfSJohannes Thumshirn 177018bb8bbfSJohannes Thumshirn /* 177118bb8bbfSJohannes Thumshirn * Reclaim block groups in the reclaim_bgs list after we deleted 177218bb8bbfSJohannes Thumshirn * all unused block_groups. This possibly gives us some more free 177318bb8bbfSJohannes Thumshirn * space. 177418bb8bbfSJohannes Thumshirn */ 177518bb8bbfSJohannes Thumshirn btrfs_reclaim_bgs(fs_info); 1776d0278245SMiao Xie sleep: 1777a0a1db70SFilipe Manana clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1778d6fd0ae2SOmar Sandoval if (kthread_should_park()) 1779d6fd0ae2SOmar Sandoval kthread_parkme(); 1780d6fd0ae2SOmar Sandoval if (kthread_should_stop()) 1781d6fd0ae2SOmar Sandoval return 0; 1782838fe188SJiri Kosina if (!again) { 1783a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 1784a74a4b97SChris Mason schedule(); 1785a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1786a74a4b97SChris Mason } 1787da288d28SFilipe Manana } 1788a74a4b97SChris Mason } 1789a74a4b97SChris Mason 1790a74a4b97SChris Mason static int transaction_kthread(void *arg) 1791a74a4b97SChris Mason { 1792a74a4b97SChris Mason struct btrfs_root *root = arg; 17930b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1794a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1795a74a4b97SChris Mason struct btrfs_transaction *cur; 17968929ecfaSYan, Zheng u64 transid; 1797643900beSNikolay Borisov time64_t delta; 1798a74a4b97SChris Mason unsigned long delay; 1799914b2007SJan Kara bool cannot_commit; 1800a74a4b97SChris Mason 1801a74a4b97SChris Mason do { 1802914b2007SJan Kara cannot_commit = false; 1803ba1bc00fSNikolay Borisov delay = msecs_to_jiffies(fs_info->commit_interval * 1000); 18040b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 1805a74a4b97SChris Mason 18060b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 18070b246afaSJeff Mahoney cur = fs_info->running_transaction; 1808a74a4b97SChris Mason if (!cur) { 18090b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1810a74a4b97SChris Mason goto sleep; 1811a74a4b97SChris Mason } 181231153d81SYan Zheng 1813643900beSNikolay Borisov delta = ktime_get_seconds() - cur->start_time; 1814fdfbf020SJosef Bacik if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) && 1815fdfbf020SJosef Bacik cur->state < TRANS_STATE_COMMIT_START && 1816643900beSNikolay Borisov delta < fs_info->commit_interval) { 18170b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1818fb8a7e94SNikolay Borisov delay -= msecs_to_jiffies((delta - 1) * 1000); 1819fb8a7e94SNikolay Borisov delay = min(delay, 1820fb8a7e94SNikolay Borisov msecs_to_jiffies(fs_info->commit_interval * 1000)); 1821a74a4b97SChris Mason goto sleep; 1822a74a4b97SChris Mason } 18238929ecfaSYan, Zheng transid = cur->transid; 18240b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 182556bec294SChris Mason 182679787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 1827354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 1828914b2007SJan Kara if (IS_ERR(trans)) { 1829354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 1830914b2007SJan Kara cannot_commit = true; 183179787eaaSJeff Mahoney goto sleep; 1832914b2007SJan Kara } 18338929ecfaSYan, Zheng if (transid == trans->transid) { 18343a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 18358929ecfaSYan, Zheng } else { 18363a45bb20SJeff Mahoney btrfs_end_transaction(trans); 18378929ecfaSYan, Zheng } 1838a74a4b97SChris Mason sleep: 18390b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 18400b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 1841a74a4b97SChris Mason 184284961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) 18432ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 18448929ecfaSYan, Zheng if (!kthread_should_stop() && 18450b246afaSJeff Mahoney (!btrfs_transaction_blocked(fs_info) || 1846914b2007SJan Kara cannot_commit)) 1847bc5511d0SNikolay Borisov schedule_timeout_interruptible(delay); 1848a74a4b97SChris Mason } while (!kthread_should_stop()); 1849a74a4b97SChris Mason return 0; 1850a74a4b97SChris Mason } 1851a74a4b97SChris Mason 1852af31f5e5SChris Mason /* 185301f0f9daSNikolay Borisov * This will find the highest generation in the array of root backups. The 185401f0f9daSNikolay Borisov * index of the highest array is returned, or -EINVAL if we can't find 185501f0f9daSNikolay Borisov * anything. 1856af31f5e5SChris Mason * 1857af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1858af31f5e5SChris Mason * generation of the latest root in the array with the generation 1859af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1860af31f5e5SChris Mason */ 186101f0f9daSNikolay Borisov static int find_newest_super_backup(struct btrfs_fs_info *info) 1862af31f5e5SChris Mason { 186301f0f9daSNikolay Borisov const u64 newest_gen = btrfs_super_generation(info->super_copy); 1864af31f5e5SChris Mason u64 cur; 1865af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1866af31f5e5SChris Mason int i; 1867af31f5e5SChris Mason 1868af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1869af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1870af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1871af31f5e5SChris Mason if (cur == newest_gen) 187201f0f9daSNikolay Borisov return i; 1873af31f5e5SChris Mason } 1874af31f5e5SChris Mason 187501f0f9daSNikolay Borisov return -EINVAL; 1876af31f5e5SChris Mason } 1877af31f5e5SChris Mason 1878af31f5e5SChris Mason /* 1879af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1880af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1881af31f5e5SChris Mason * done 1882af31f5e5SChris Mason */ 1883af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1884af31f5e5SChris Mason { 18856ef108ddSNikolay Borisov const int next_backup = info->backup_root_index; 1886af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1887af31f5e5SChris Mason 1888af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1889af31f5e5SChris Mason 1890af31f5e5SChris Mason /* 1891af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1892af31f5e5SChris Mason * regardless of which ones we use today 1893af31f5e5SChris Mason */ 1894af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1895af31f5e5SChris Mason 1896af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1897af31f5e5SChris Mason 1898af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1899af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1900af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1901af31f5e5SChris Mason 1902af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1903af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1904af31f5e5SChris Mason 1905af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1906af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1907af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 1908af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 1909af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 1910af31f5e5SChris Mason 19111c56ab99SQu Wenruo if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) { 19129c54e80dSJosef Bacik struct btrfs_root *extent_root = btrfs_extent_root(info, 0); 1913f7238e50SJosef Bacik struct btrfs_root *csum_root = btrfs_csum_root(info, 0); 19149c54e80dSJosef Bacik 19159c54e80dSJosef Bacik btrfs_set_backup_extent_root(root_backup, 19169c54e80dSJosef Bacik extent_root->node->start); 1917af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 191829cbcf40SJosef Bacik btrfs_header_generation(extent_root->node)); 1919af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 192029cbcf40SJosef Bacik btrfs_header_level(extent_root->node)); 1921af31f5e5SChris Mason 1922f7238e50SJosef Bacik btrfs_set_backup_csum_root(root_backup, csum_root->node->start); 1923f7238e50SJosef Bacik btrfs_set_backup_csum_root_gen(root_backup, 1924f7238e50SJosef Bacik btrfs_header_generation(csum_root->node)); 1925f7238e50SJosef Bacik btrfs_set_backup_csum_root_level(root_backup, 1926f7238e50SJosef Bacik btrfs_header_level(csum_root->node)); 19279c54e80dSJosef Bacik } 1928af31f5e5SChris Mason 19297c7e82a7SChris Mason /* 19307c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 19317c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 19327c7e82a7SChris Mason */ 19337c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 19347c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 19357c7e82a7SChris Mason info->fs_root->node->start); 1936af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 1937af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 1938af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 1939af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 19407c7e82a7SChris Mason } 1941af31f5e5SChris Mason 1942af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 1943af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 1944af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 1945af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 1946af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 1947af31f5e5SChris Mason 1948af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 1949af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 1950af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 1951af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 1952af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 1953af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 1954af31f5e5SChris Mason 1955af31f5e5SChris Mason /* 1956af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 1957af31f5e5SChris Mason * for the next commit 1958af31f5e5SChris Mason */ 1959af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 1960af31f5e5SChris Mason &info->super_for_commit->super_roots, 1961af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 1962af31f5e5SChris Mason } 1963af31f5e5SChris Mason 1964af31f5e5SChris Mason /* 1965bd2336b2SNikolay Borisov * read_backup_root - Reads a backup root based on the passed priority. Prio 0 1966bd2336b2SNikolay Borisov * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots 1967bd2336b2SNikolay Borisov * 1968bd2336b2SNikolay Borisov * fs_info - filesystem whose backup roots need to be read 1969bd2336b2SNikolay Borisov * priority - priority of backup root required 1970bd2336b2SNikolay Borisov * 1971bd2336b2SNikolay Borisov * Returns backup root index on success and -EINVAL otherwise. 1972bd2336b2SNikolay Borisov */ 1973bd2336b2SNikolay Borisov static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority) 1974bd2336b2SNikolay Borisov { 1975bd2336b2SNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 1976bd2336b2SNikolay Borisov struct btrfs_super_block *super = fs_info->super_copy; 1977bd2336b2SNikolay Borisov struct btrfs_root_backup *root_backup; 1978bd2336b2SNikolay Borisov 1979bd2336b2SNikolay Borisov if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) { 1980bd2336b2SNikolay Borisov if (priority == 0) 1981bd2336b2SNikolay Borisov return backup_index; 1982bd2336b2SNikolay Borisov 1983bd2336b2SNikolay Borisov backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority; 1984bd2336b2SNikolay Borisov backup_index %= BTRFS_NUM_BACKUP_ROOTS; 1985bd2336b2SNikolay Borisov } else { 1986bd2336b2SNikolay Borisov return -EINVAL; 1987bd2336b2SNikolay Borisov } 1988bd2336b2SNikolay Borisov 1989bd2336b2SNikolay Borisov root_backup = super->super_roots + backup_index; 1990bd2336b2SNikolay Borisov 1991bd2336b2SNikolay Borisov btrfs_set_super_generation(super, 1992bd2336b2SNikolay Borisov btrfs_backup_tree_root_gen(root_backup)); 1993bd2336b2SNikolay Borisov btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 1994bd2336b2SNikolay Borisov btrfs_set_super_root_level(super, 1995bd2336b2SNikolay Borisov btrfs_backup_tree_root_level(root_backup)); 1996bd2336b2SNikolay Borisov btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 1997bd2336b2SNikolay Borisov 1998bd2336b2SNikolay Borisov /* 1999bd2336b2SNikolay Borisov * Fixme: the total bytes and num_devices need to match or we should 2000bd2336b2SNikolay Borisov * need a fsck 2001bd2336b2SNikolay Borisov */ 2002bd2336b2SNikolay Borisov btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 2003bd2336b2SNikolay Borisov btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 2004bd2336b2SNikolay Borisov 2005bd2336b2SNikolay Borisov return backup_index; 2006bd2336b2SNikolay Borisov } 2007bd2336b2SNikolay Borisov 20087abadb64SLiu Bo /* helper to cleanup workers */ 20097abadb64SLiu Bo static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 20107abadb64SLiu Bo { 2011dc6e3209SQu Wenruo btrfs_destroy_workqueue(fs_info->fixup_workers); 2012afe3d242SQu Wenruo btrfs_destroy_workqueue(fs_info->delalloc_workers); 2013a31b4a43SChristoph Hellwig btrfs_destroy_workqueue(fs_info->hipri_workers); 20145cdc7ad3SQu Wenruo btrfs_destroy_workqueue(fs_info->workers); 2015d7b9416fSChristoph Hellwig if (fs_info->endio_workers) 2016d7b9416fSChristoph Hellwig destroy_workqueue(fs_info->endio_workers); 2017385de0efSChristoph Hellwig if (fs_info->rmw_workers) 2018385de0efSChristoph Hellwig destroy_workqueue(fs_info->rmw_workers); 2019fed8a72dSChristoph Hellwig if (fs_info->compressed_write_workers) 2020fed8a72dSChristoph Hellwig destroy_workqueue(fs_info->compressed_write_workers); 2021fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_write_workers); 2022fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 20235b3bc44eSQu Wenruo btrfs_destroy_workqueue(fs_info->delayed_workers); 2024e66f0bb1SQu Wenruo btrfs_destroy_workqueue(fs_info->caching_workers); 2025a44903abSQu Wenruo btrfs_destroy_workqueue(fs_info->flush_workers); 2026fc97fab0SQu Wenruo btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2027b0643e59SDennis Zhou if (fs_info->discard_ctl.discard_workers) 2028b0643e59SDennis Zhou destroy_workqueue(fs_info->discard_ctl.discard_workers); 2029a9b9477dSFilipe Manana /* 2030a9b9477dSFilipe Manana * Now that all other work queues are destroyed, we can safely destroy 2031a9b9477dSFilipe Manana * the queues used for metadata I/O, since tasks from those other work 2032a9b9477dSFilipe Manana * queues can do metadata I/O operations. 2033a9b9477dSFilipe Manana */ 2034d7b9416fSChristoph Hellwig if (fs_info->endio_meta_workers) 2035d7b9416fSChristoph Hellwig destroy_workqueue(fs_info->endio_meta_workers); 20367abadb64SLiu Bo } 20377abadb64SLiu Bo 20382e9f5954SRashika static void free_root_extent_buffers(struct btrfs_root *root) 20392e9f5954SRashika { 20402e9f5954SRashika if (root) { 20412e9f5954SRashika free_extent_buffer(root->node); 20422e9f5954SRashika free_extent_buffer(root->commit_root); 20432e9f5954SRashika root->node = NULL; 20442e9f5954SRashika root->commit_root = NULL; 20452e9f5954SRashika } 20462e9f5954SRashika } 20472e9f5954SRashika 2048abed4aaaSJosef Bacik static void free_global_root_pointers(struct btrfs_fs_info *fs_info) 2049abed4aaaSJosef Bacik { 2050abed4aaaSJosef Bacik struct btrfs_root *root, *tmp; 2051abed4aaaSJosef Bacik 2052abed4aaaSJosef Bacik rbtree_postorder_for_each_entry_safe(root, tmp, 2053abed4aaaSJosef Bacik &fs_info->global_root_tree, 2054abed4aaaSJosef Bacik rb_node) 2055abed4aaaSJosef Bacik free_root_extent_buffers(root); 2056abed4aaaSJosef Bacik } 2057abed4aaaSJosef Bacik 2058af31f5e5SChris Mason /* helper to cleanup tree roots */ 20594273eaffSAnand Jain static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root) 2060af31f5e5SChris Mason { 20612e9f5954SRashika free_root_extent_buffers(info->tree_root); 2062655b09feSJosef Bacik 2063abed4aaaSJosef Bacik free_global_root_pointers(info); 20642e9f5954SRashika free_root_extent_buffers(info->dev_root); 20652e9f5954SRashika free_root_extent_buffers(info->quota_root); 20662e9f5954SRashika free_root_extent_buffers(info->uuid_root); 20678c38938cSJosef Bacik free_root_extent_buffers(info->fs_root); 2068aeb935a4SQu Wenruo free_root_extent_buffers(info->data_reloc_root); 20699c54e80dSJosef Bacik free_root_extent_buffers(info->block_group_root); 20704273eaffSAnand Jain if (free_chunk_root) 20712e9f5954SRashika free_root_extent_buffers(info->chunk_root); 2072af31f5e5SChris Mason } 2073af31f5e5SChris Mason 20748c38938cSJosef Bacik void btrfs_put_root(struct btrfs_root *root) 20758c38938cSJosef Bacik { 20768c38938cSJosef Bacik if (!root) 20778c38938cSJosef Bacik return; 20788c38938cSJosef Bacik 20798c38938cSJosef Bacik if (refcount_dec_and_test(&root->refs)) { 20808c38938cSJosef Bacik WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 20811dae7e0eSQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)); 20828c38938cSJosef Bacik if (root->anon_dev) 20838c38938cSJosef Bacik free_anon_bdev(root->anon_dev); 20848c38938cSJosef Bacik btrfs_drew_lock_destroy(&root->snapshot_lock); 2085923eb523SJohannes Thumshirn free_root_extent_buffers(root); 20868c38938cSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 2087fc7cbcd4SDavid Sterba spin_lock(&root->fs_info->fs_roots_radix_lock); 20888c38938cSJosef Bacik list_del_init(&root->leak_list); 2089fc7cbcd4SDavid Sterba spin_unlock(&root->fs_info->fs_roots_radix_lock); 20908c38938cSJosef Bacik #endif 20918c38938cSJosef Bacik kfree(root); 20928c38938cSJosef Bacik } 20938c38938cSJosef Bacik } 20948c38938cSJosef Bacik 2095faa2dbf0SJosef Bacik void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2096171f6537SJosef Bacik { 2097fc7cbcd4SDavid Sterba int ret; 2098fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 2099fc7cbcd4SDavid Sterba int i; 2100171f6537SJosef Bacik 2101171f6537SJosef Bacik while (!list_empty(&fs_info->dead_roots)) { 2102fc7cbcd4SDavid Sterba gang[0] = list_entry(fs_info->dead_roots.next, 2103171f6537SJosef Bacik struct btrfs_root, root_list); 2104fc7cbcd4SDavid Sterba list_del(&gang[0]->root_list); 2105171f6537SJosef Bacik 2106fc7cbcd4SDavid Sterba if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) 2107fc7cbcd4SDavid Sterba btrfs_drop_and_free_fs_root(fs_info, gang[0]); 2108fc7cbcd4SDavid Sterba btrfs_put_root(gang[0]); 2109171f6537SJosef Bacik } 2110171f6537SJosef Bacik 2111fc7cbcd4SDavid Sterba while (1) { 2112fc7cbcd4SDavid Sterba ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2113fc7cbcd4SDavid Sterba (void **)gang, 0, 2114fc7cbcd4SDavid Sterba ARRAY_SIZE(gang)); 2115fc7cbcd4SDavid Sterba if (!ret) 2116fc7cbcd4SDavid Sterba break; 2117fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) 2118fc7cbcd4SDavid Sterba btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2119171f6537SJosef Bacik } 2120171f6537SJosef Bacik } 2121af31f5e5SChris Mason 2122638aa7edSEric Sandeen static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2123638aa7edSEric Sandeen { 2124638aa7edSEric Sandeen mutex_init(&fs_info->scrub_lock); 2125638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_running, 0); 2126638aa7edSEric Sandeen atomic_set(&fs_info->scrub_pause_req, 0); 2127638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_paused, 0); 2128638aa7edSEric Sandeen atomic_set(&fs_info->scrub_cancel_req, 0); 2129638aa7edSEric Sandeen init_waitqueue_head(&fs_info->scrub_pause_wait); 2130ff09c4caSAnand Jain refcount_set(&fs_info->scrub_workers_refcnt, 0); 2131638aa7edSEric Sandeen } 2132638aa7edSEric Sandeen 2133779a65a4SEric Sandeen static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2134779a65a4SEric Sandeen { 2135779a65a4SEric Sandeen spin_lock_init(&fs_info->balance_lock); 2136779a65a4SEric Sandeen mutex_init(&fs_info->balance_mutex); 2137779a65a4SEric Sandeen atomic_set(&fs_info->balance_pause_req, 0); 2138779a65a4SEric Sandeen atomic_set(&fs_info->balance_cancel_req, 0); 2139779a65a4SEric Sandeen fs_info->balance_ctl = NULL; 2140779a65a4SEric Sandeen init_waitqueue_head(&fs_info->balance_wait_q); 2141907d2710SDavid Sterba atomic_set(&fs_info->reloc_cancel_req, 0); 2142779a65a4SEric Sandeen } 2143779a65a4SEric Sandeen 21446bccf3abSJeff Mahoney static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info) 2145f37938e0SEric Sandeen { 21462ff7e61eSJeff Mahoney struct inode *inode = fs_info->btree_inode; 2147e256927bSJosef Bacik unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID, 2148e256927bSJosef Bacik fs_info->tree_root); 21492ff7e61eSJeff Mahoney 21502ff7e61eSJeff Mahoney inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 21512ff7e61eSJeff Mahoney set_nlink(inode, 1); 2152f37938e0SEric Sandeen /* 2153f37938e0SEric Sandeen * we set the i_size on the btree inode to the max possible int. 2154f37938e0SEric Sandeen * the real end of the address space is determined by all of 2155f37938e0SEric Sandeen * the devices in the system 2156f37938e0SEric Sandeen */ 21572ff7e61eSJeff Mahoney inode->i_size = OFFSET_MAX; 21582ff7e61eSJeff Mahoney inode->i_mapping->a_ops = &btree_aops; 2159f37938e0SEric Sandeen 21602ff7e61eSJeff Mahoney RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 216143eb5f29SQu Wenruo extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree, 216235da5a7eSDavid Sterba IO_TREE_BTREE_INODE_IO); 21632ff7e61eSJeff Mahoney extent_map_tree_init(&BTRFS_I(inode)->extent_tree); 2164f37938e0SEric Sandeen 21655c8fd99fSJosef Bacik BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root); 2166adac5584SFilipe Manana BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID; 2167adac5584SFilipe Manana BTRFS_I(inode)->location.type = 0; 2168adac5584SFilipe Manana BTRFS_I(inode)->location.offset = 0; 21692ff7e61eSJeff Mahoney set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); 2170e256927bSJosef Bacik __insert_inode_hash(inode, hash); 2171f37938e0SEric Sandeen } 2172f37938e0SEric Sandeen 2173ad618368SEric Sandeen static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2174ad618368SEric Sandeen { 2175ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2176129827e3SDavid Sterba init_rwsem(&fs_info->dev_replace.rwsem); 21777f8d236aSDavid Sterba init_waitqueue_head(&fs_info->dev_replace.replace_wait); 2178ad618368SEric Sandeen } 2179ad618368SEric Sandeen 2180f9e92e40SEric Sandeen static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2181f9e92e40SEric Sandeen { 2182f9e92e40SEric Sandeen spin_lock_init(&fs_info->qgroup_lock); 2183f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_ioctl_lock); 2184f9e92e40SEric Sandeen fs_info->qgroup_tree = RB_ROOT; 2185f9e92e40SEric Sandeen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2186f9e92e40SEric Sandeen fs_info->qgroup_seq = 1; 2187f9e92e40SEric Sandeen fs_info->qgroup_ulist = NULL; 2188d2c609b8SJeff Mahoney fs_info->qgroup_rescan_running = false; 2189011b46c3SQu Wenruo fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL; 2190f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_rescan_lock); 2191f9e92e40SEric Sandeen } 2192f9e92e40SEric Sandeen 2193d21deec5SSu Yue static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info) 21942a458198SEric Sandeen { 2195f7b885beSAnand Jain u32 max_active = fs_info->thread_pool_size; 21966f011058SDavid Sterba unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 21972a458198SEric Sandeen 21982a458198SEric Sandeen fs_info->workers = 2199a31b4a43SChristoph Hellwig btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16); 2200a31b4a43SChristoph Hellwig fs_info->hipri_workers = 2201a31b4a43SChristoph Hellwig btrfs_alloc_workqueue(fs_info, "worker-high", 2202cb001095SJeff Mahoney flags | WQ_HIGHPRI, max_active, 16); 22032a458198SEric Sandeen 22042a458198SEric Sandeen fs_info->delalloc_workers = 2205cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delalloc", 2206cb001095SJeff Mahoney flags, max_active, 2); 22072a458198SEric Sandeen 22082a458198SEric Sandeen fs_info->flush_workers = 2209cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "flush_delalloc", 2210cb001095SJeff Mahoney flags, max_active, 0); 22112a458198SEric Sandeen 22122a458198SEric Sandeen fs_info->caching_workers = 2213cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); 22142a458198SEric Sandeen 22152a458198SEric Sandeen fs_info->fixup_workers = 2216cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); 22172a458198SEric Sandeen 22182a458198SEric Sandeen fs_info->endio_workers = 2219d7b9416fSChristoph Hellwig alloc_workqueue("btrfs-endio", flags, max_active); 22202a458198SEric Sandeen fs_info->endio_meta_workers = 2221d7b9416fSChristoph Hellwig alloc_workqueue("btrfs-endio-meta", flags, max_active); 2222385de0efSChristoph Hellwig fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active); 22232a458198SEric Sandeen fs_info->endio_write_workers = 2224cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-write", flags, 2225cb001095SJeff Mahoney max_active, 2); 2226fed8a72dSChristoph Hellwig fs_info->compressed_write_workers = 2227fed8a72dSChristoph Hellwig alloc_workqueue("btrfs-compressed-write", flags, max_active); 22282a458198SEric Sandeen fs_info->endio_freespace_worker = 2229cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "freespace-write", flags, 2230cb001095SJeff Mahoney max_active, 0); 22312a458198SEric Sandeen fs_info->delayed_workers = 2232cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, 2233cb001095SJeff Mahoney max_active, 0); 22342a458198SEric Sandeen fs_info->qgroup_rescan_workers = 2235cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); 2236b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers = 2237b0643e59SDennis Zhou alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1); 22382a458198SEric Sandeen 2239a31b4a43SChristoph Hellwig if (!(fs_info->workers && fs_info->hipri_workers && 2240a31b4a43SChristoph Hellwig fs_info->delalloc_workers && fs_info->flush_workers && 22412a458198SEric Sandeen fs_info->endio_workers && fs_info->endio_meta_workers && 2242fed8a72dSChristoph Hellwig fs_info->compressed_write_workers && 22431a1a2851SQu Wenruo fs_info->endio_write_workers && 22442a458198SEric Sandeen fs_info->endio_freespace_worker && fs_info->rmw_workers && 2245f26c9238SQu Wenruo fs_info->caching_workers && fs_info->fixup_workers && 2246f26c9238SQu Wenruo fs_info->delayed_workers && fs_info->qgroup_rescan_workers && 2247b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers)) { 22482a458198SEric Sandeen return -ENOMEM; 22492a458198SEric Sandeen } 22502a458198SEric Sandeen 22512a458198SEric Sandeen return 0; 22522a458198SEric Sandeen } 22532a458198SEric Sandeen 22546d97c6e3SJohannes Thumshirn static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type) 22556d97c6e3SJohannes Thumshirn { 22566d97c6e3SJohannes Thumshirn struct crypto_shash *csum_shash; 2257b4e967beSDavid Sterba const char *csum_driver = btrfs_super_csum_driver(csum_type); 22586d97c6e3SJohannes Thumshirn 2259b4e967beSDavid Sterba csum_shash = crypto_alloc_shash(csum_driver, 0, 0); 22606d97c6e3SJohannes Thumshirn 22616d97c6e3SJohannes Thumshirn if (IS_ERR(csum_shash)) { 22626d97c6e3SJohannes Thumshirn btrfs_err(fs_info, "error allocating %s hash for checksum", 2263b4e967beSDavid Sterba csum_driver); 22646d97c6e3SJohannes Thumshirn return PTR_ERR(csum_shash); 22656d97c6e3SJohannes Thumshirn } 22666d97c6e3SJohannes Thumshirn 22676d97c6e3SJohannes Thumshirn fs_info->csum_shash = csum_shash; 22686d97c6e3SJohannes Thumshirn 2269c8a5f8caSDavid Sterba btrfs_info(fs_info, "using %s (%s) checksum algorithm", 2270c8a5f8caSDavid Sterba btrfs_super_csum_name(csum_type), 2271c8a5f8caSDavid Sterba crypto_shash_driver_name(csum_shash)); 22726d97c6e3SJohannes Thumshirn return 0; 22736d97c6e3SJohannes Thumshirn } 22746d97c6e3SJohannes Thumshirn 227563443bf5SEric Sandeen static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 227663443bf5SEric Sandeen struct btrfs_fs_devices *fs_devices) 227763443bf5SEric Sandeen { 227863443bf5SEric Sandeen int ret; 2279789d6a3aSQu Wenruo struct btrfs_tree_parent_check check = { 0 }; 228063443bf5SEric Sandeen struct btrfs_root *log_tree_root; 228163443bf5SEric Sandeen struct btrfs_super_block *disk_super = fs_info->super_copy; 228263443bf5SEric Sandeen u64 bytenr = btrfs_super_log_root(disk_super); 2283581c1760SQu Wenruo int level = btrfs_super_log_root_level(disk_super); 228463443bf5SEric Sandeen 228563443bf5SEric Sandeen if (fs_devices->rw_devices == 0) { 2286f14d104dSDavid Sterba btrfs_warn(fs_info, "log replay required on RO media"); 228763443bf5SEric Sandeen return -EIO; 228863443bf5SEric Sandeen } 228963443bf5SEric Sandeen 229096dfcb46SJosef Bacik log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, 229196dfcb46SJosef Bacik GFP_KERNEL); 229263443bf5SEric Sandeen if (!log_tree_root) 229363443bf5SEric Sandeen return -ENOMEM; 229463443bf5SEric Sandeen 2295789d6a3aSQu Wenruo check.level = level; 2296789d6a3aSQu Wenruo check.transid = fs_info->generation + 1; 2297789d6a3aSQu Wenruo check.owner_root = BTRFS_TREE_LOG_OBJECTID; 2298789d6a3aSQu Wenruo log_tree_root->node = read_tree_block(fs_info, bytenr, &check); 229964c043deSLiu Bo if (IS_ERR(log_tree_root->node)) { 2300f14d104dSDavid Sterba btrfs_warn(fs_info, "failed to read log tree"); 23010eeff236SLiu Bo ret = PTR_ERR(log_tree_root->node); 23028c38938cSJosef Bacik log_tree_root->node = NULL; 230300246528SJosef Bacik btrfs_put_root(log_tree_root); 23040eeff236SLiu Bo return ret; 23054eb150d6SQu Wenruo } 23064eb150d6SQu Wenruo if (!extent_buffer_uptodate(log_tree_root->node)) { 2307f14d104dSDavid Sterba btrfs_err(fs_info, "failed to read log tree"); 230800246528SJosef Bacik btrfs_put_root(log_tree_root); 230963443bf5SEric Sandeen return -EIO; 231063443bf5SEric Sandeen } 23114eb150d6SQu Wenruo 231263443bf5SEric Sandeen /* returns with log_tree_root freed on success */ 231363443bf5SEric Sandeen ret = btrfs_recover_log_trees(log_tree_root); 231463443bf5SEric Sandeen if (ret) { 23150b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 231663443bf5SEric Sandeen "Failed to recover log tree"); 231700246528SJosef Bacik btrfs_put_root(log_tree_root); 231863443bf5SEric Sandeen return ret; 231963443bf5SEric Sandeen } 232063443bf5SEric Sandeen 2321bc98a42cSDavid Howells if (sb_rdonly(fs_info->sb)) { 23226bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 232363443bf5SEric Sandeen if (ret) 232463443bf5SEric Sandeen return ret; 232563443bf5SEric Sandeen } 232663443bf5SEric Sandeen 232763443bf5SEric Sandeen return 0; 232863443bf5SEric Sandeen } 232963443bf5SEric Sandeen 2330abed4aaaSJosef Bacik static int load_global_roots_objectid(struct btrfs_root *tree_root, 2331abed4aaaSJosef Bacik struct btrfs_path *path, u64 objectid, 2332abed4aaaSJosef Bacik const char *name) 2333abed4aaaSJosef Bacik { 2334abed4aaaSJosef Bacik struct btrfs_fs_info *fs_info = tree_root->fs_info; 2335abed4aaaSJosef Bacik struct btrfs_root *root; 2336f7238e50SJosef Bacik u64 max_global_id = 0; 2337abed4aaaSJosef Bacik int ret; 2338abed4aaaSJosef Bacik struct btrfs_key key = { 2339abed4aaaSJosef Bacik .objectid = objectid, 2340abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 2341abed4aaaSJosef Bacik .offset = 0, 2342abed4aaaSJosef Bacik }; 2343abed4aaaSJosef Bacik bool found = false; 2344abed4aaaSJosef Bacik 2345abed4aaaSJosef Bacik /* If we have IGNOREDATACSUMS skip loading these roots. */ 2346abed4aaaSJosef Bacik if (objectid == BTRFS_CSUM_TREE_OBJECTID && 2347abed4aaaSJosef Bacik btrfs_test_opt(fs_info, IGNOREDATACSUMS)) { 2348abed4aaaSJosef Bacik set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state); 2349abed4aaaSJosef Bacik return 0; 2350abed4aaaSJosef Bacik } 2351abed4aaaSJosef Bacik 2352abed4aaaSJosef Bacik while (1) { 2353abed4aaaSJosef Bacik ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0); 2354abed4aaaSJosef Bacik if (ret < 0) 2355abed4aaaSJosef Bacik break; 2356abed4aaaSJosef Bacik 2357abed4aaaSJosef Bacik if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 2358abed4aaaSJosef Bacik ret = btrfs_next_leaf(tree_root, path); 2359abed4aaaSJosef Bacik if (ret) { 2360abed4aaaSJosef Bacik if (ret > 0) 2361abed4aaaSJosef Bacik ret = 0; 2362abed4aaaSJosef Bacik break; 2363abed4aaaSJosef Bacik } 2364abed4aaaSJosef Bacik } 2365abed4aaaSJosef Bacik ret = 0; 2366abed4aaaSJosef Bacik 2367abed4aaaSJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 2368abed4aaaSJosef Bacik if (key.objectid != objectid) 2369abed4aaaSJosef Bacik break; 2370abed4aaaSJosef Bacik btrfs_release_path(path); 2371abed4aaaSJosef Bacik 2372f7238e50SJosef Bacik /* 2373f7238e50SJosef Bacik * Just worry about this for extent tree, it'll be the same for 2374f7238e50SJosef Bacik * everybody. 2375f7238e50SJosef Bacik */ 2376f7238e50SJosef Bacik if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 2377f7238e50SJosef Bacik max_global_id = max(max_global_id, key.offset); 2378f7238e50SJosef Bacik 2379abed4aaaSJosef Bacik found = true; 2380abed4aaaSJosef Bacik root = read_tree_root_path(tree_root, path, &key); 2381abed4aaaSJosef Bacik if (IS_ERR(root)) { 2382abed4aaaSJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) 2383abed4aaaSJosef Bacik ret = PTR_ERR(root); 2384abed4aaaSJosef Bacik break; 2385abed4aaaSJosef Bacik } 2386abed4aaaSJosef Bacik set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2387abed4aaaSJosef Bacik ret = btrfs_global_root_insert(root); 2388abed4aaaSJosef Bacik if (ret) { 2389abed4aaaSJosef Bacik btrfs_put_root(root); 2390abed4aaaSJosef Bacik break; 2391abed4aaaSJosef Bacik } 2392abed4aaaSJosef Bacik key.offset++; 2393abed4aaaSJosef Bacik } 2394abed4aaaSJosef Bacik btrfs_release_path(path); 2395abed4aaaSJosef Bacik 2396f7238e50SJosef Bacik if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 2397f7238e50SJosef Bacik fs_info->nr_global_roots = max_global_id + 1; 2398f7238e50SJosef Bacik 2399abed4aaaSJosef Bacik if (!found || ret) { 2400abed4aaaSJosef Bacik if (objectid == BTRFS_CSUM_TREE_OBJECTID) 2401abed4aaaSJosef Bacik set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state); 2402abed4aaaSJosef Bacik 2403abed4aaaSJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) 2404abed4aaaSJosef Bacik ret = ret ? ret : -ENOENT; 2405abed4aaaSJosef Bacik else 2406abed4aaaSJosef Bacik ret = 0; 2407abed4aaaSJosef Bacik btrfs_err(fs_info, "failed to load root %s", name); 2408abed4aaaSJosef Bacik } 2409abed4aaaSJosef Bacik return ret; 2410abed4aaaSJosef Bacik } 2411abed4aaaSJosef Bacik 2412abed4aaaSJosef Bacik static int load_global_roots(struct btrfs_root *tree_root) 2413abed4aaaSJosef Bacik { 2414abed4aaaSJosef Bacik struct btrfs_path *path; 2415abed4aaaSJosef Bacik int ret = 0; 2416abed4aaaSJosef Bacik 2417abed4aaaSJosef Bacik path = btrfs_alloc_path(); 2418abed4aaaSJosef Bacik if (!path) 2419abed4aaaSJosef Bacik return -ENOMEM; 2420abed4aaaSJosef Bacik 2421abed4aaaSJosef Bacik ret = load_global_roots_objectid(tree_root, path, 2422abed4aaaSJosef Bacik BTRFS_EXTENT_TREE_OBJECTID, "extent"); 2423abed4aaaSJosef Bacik if (ret) 2424abed4aaaSJosef Bacik goto out; 2425abed4aaaSJosef Bacik ret = load_global_roots_objectid(tree_root, path, 2426abed4aaaSJosef Bacik BTRFS_CSUM_TREE_OBJECTID, "csum"); 2427abed4aaaSJosef Bacik if (ret) 2428abed4aaaSJosef Bacik goto out; 2429abed4aaaSJosef Bacik if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE)) 2430abed4aaaSJosef Bacik goto out; 2431abed4aaaSJosef Bacik ret = load_global_roots_objectid(tree_root, path, 2432abed4aaaSJosef Bacik BTRFS_FREE_SPACE_TREE_OBJECTID, 2433abed4aaaSJosef Bacik "free space"); 2434abed4aaaSJosef Bacik out: 2435abed4aaaSJosef Bacik btrfs_free_path(path); 2436abed4aaaSJosef Bacik return ret; 2437abed4aaaSJosef Bacik } 2438abed4aaaSJosef Bacik 24396bccf3abSJeff Mahoney static int btrfs_read_roots(struct btrfs_fs_info *fs_info) 24404bbcaa64SEric Sandeen { 24416bccf3abSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 2442a4f3d2c4SDavid Sterba struct btrfs_root *root; 24434bbcaa64SEric Sandeen struct btrfs_key location; 24444bbcaa64SEric Sandeen int ret; 24454bbcaa64SEric Sandeen 24466bccf3abSJeff Mahoney BUG_ON(!fs_info->tree_root); 24476bccf3abSJeff Mahoney 2448abed4aaaSJosef Bacik ret = load_global_roots(tree_root); 2449abed4aaaSJosef Bacik if (ret) 2450abed4aaaSJosef Bacik return ret; 2451abed4aaaSJosef Bacik 24524bbcaa64SEric Sandeen location.type = BTRFS_ROOT_ITEM_KEY; 24534bbcaa64SEric Sandeen location.offset = 0; 24544bbcaa64SEric Sandeen 24551c56ab99SQu Wenruo if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) { 245614033b08SQu Wenruo location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID; 245714033b08SQu Wenruo root = btrfs_read_tree_root(tree_root, &location); 245814033b08SQu Wenruo if (IS_ERR(root)) { 245914033b08SQu Wenruo if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 246014033b08SQu Wenruo ret = PTR_ERR(root); 246114033b08SQu Wenruo goto out; 246214033b08SQu Wenruo } 246314033b08SQu Wenruo } else { 246414033b08SQu Wenruo set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 246514033b08SQu Wenruo fs_info->block_group_root = root; 246614033b08SQu Wenruo } 246714033b08SQu Wenruo } 246814033b08SQu Wenruo 246914033b08SQu Wenruo location.objectid = BTRFS_DEV_TREE_OBJECTID; 2470a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2471f50f4353SLiu Bo if (IS_ERR(root)) { 247242437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2473f50f4353SLiu Bo ret = PTR_ERR(root); 2474f50f4353SLiu Bo goto out; 2475f50f4353SLiu Bo } 247642437a63SJosef Bacik } else { 2477a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2478a4f3d2c4SDavid Sterba fs_info->dev_root = root; 247942437a63SJosef Bacik } 2480820a49daSJosef Bacik /* Initialize fs_info for all devices in any case */ 2481a8d1b164SJohannes Thumshirn ret = btrfs_init_devices_late(fs_info); 2482a8d1b164SJohannes Thumshirn if (ret) 2483a8d1b164SJohannes Thumshirn goto out; 24844bbcaa64SEric Sandeen 2485aeb935a4SQu Wenruo /* 2486aeb935a4SQu Wenruo * This tree can share blocks with some other fs tree during relocation 2487aeb935a4SQu Wenruo * and we need a proper setup by btrfs_get_fs_root 2488aeb935a4SQu Wenruo */ 248956e9357aSDavid Sterba root = btrfs_get_fs_root(tree_root->fs_info, 249056e9357aSDavid Sterba BTRFS_DATA_RELOC_TREE_OBJECTID, true); 2491aeb935a4SQu Wenruo if (IS_ERR(root)) { 249242437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2493aeb935a4SQu Wenruo ret = PTR_ERR(root); 2494aeb935a4SQu Wenruo goto out; 2495aeb935a4SQu Wenruo } 249642437a63SJosef Bacik } else { 2497aeb935a4SQu Wenruo set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2498aeb935a4SQu Wenruo fs_info->data_reloc_root = root; 249942437a63SJosef Bacik } 2500aeb935a4SQu Wenruo 25014bbcaa64SEric Sandeen location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2502a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2503a4f3d2c4SDavid Sterba if (!IS_ERR(root)) { 2504a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2505afcdd129SJosef Bacik set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); 2506a4f3d2c4SDavid Sterba fs_info->quota_root = root; 25074bbcaa64SEric Sandeen } 25084bbcaa64SEric Sandeen 25094bbcaa64SEric Sandeen location.objectid = BTRFS_UUID_TREE_OBJECTID; 2510a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2511a4f3d2c4SDavid Sterba if (IS_ERR(root)) { 251242437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2513a4f3d2c4SDavid Sterba ret = PTR_ERR(root); 25144bbcaa64SEric Sandeen if (ret != -ENOENT) 2515f50f4353SLiu Bo goto out; 251642437a63SJosef Bacik } 25174bbcaa64SEric Sandeen } else { 2518a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2519a4f3d2c4SDavid Sterba fs_info->uuid_root = root; 25204bbcaa64SEric Sandeen } 25214bbcaa64SEric Sandeen 25224bbcaa64SEric Sandeen return 0; 2523f50f4353SLiu Bo out: 2524f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", 2525f50f4353SLiu Bo location.objectid, ret); 2526f50f4353SLiu Bo return ret; 25274bbcaa64SEric Sandeen } 25284bbcaa64SEric Sandeen 2529069ec957SQu Wenruo /* 2530069ec957SQu Wenruo * Real super block validation 2531069ec957SQu Wenruo * NOTE: super csum type and incompat features will not be checked here. 2532069ec957SQu Wenruo * 2533069ec957SQu Wenruo * @sb: super block to check 2534069ec957SQu Wenruo * @mirror_num: the super block number to check its bytenr: 2535069ec957SQu Wenruo * 0 the primary (1st) sb 2536069ec957SQu Wenruo * 1, 2 2nd and 3rd backup copy 2537069ec957SQu Wenruo * -1 skip bytenr check 2538069ec957SQu Wenruo */ 2539a05d3c91SQu Wenruo int btrfs_validate_super(struct btrfs_fs_info *fs_info, 2540069ec957SQu Wenruo struct btrfs_super_block *sb, int mirror_num) 254121a852b0SQu Wenruo { 254221a852b0SQu Wenruo u64 nodesize = btrfs_super_nodesize(sb); 254321a852b0SQu Wenruo u64 sectorsize = btrfs_super_sectorsize(sb); 254421a852b0SQu Wenruo int ret = 0; 254521a852b0SQu Wenruo 254621a852b0SQu Wenruo if (btrfs_super_magic(sb) != BTRFS_MAGIC) { 254721a852b0SQu Wenruo btrfs_err(fs_info, "no valid FS found"); 254821a852b0SQu Wenruo ret = -EINVAL; 254921a852b0SQu Wenruo } 255021a852b0SQu Wenruo if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { 255121a852b0SQu Wenruo btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", 255221a852b0SQu Wenruo btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); 255321a852b0SQu Wenruo ret = -EINVAL; 255421a852b0SQu Wenruo } 255521a852b0SQu Wenruo if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 255621a852b0SQu Wenruo btrfs_err(fs_info, "tree_root level too big: %d >= %d", 255721a852b0SQu Wenruo btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 255821a852b0SQu Wenruo ret = -EINVAL; 255921a852b0SQu Wenruo } 256021a852b0SQu Wenruo if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 256121a852b0SQu Wenruo btrfs_err(fs_info, "chunk_root level too big: %d >= %d", 256221a852b0SQu Wenruo btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 256321a852b0SQu Wenruo ret = -EINVAL; 256421a852b0SQu Wenruo } 256521a852b0SQu Wenruo if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 256621a852b0SQu Wenruo btrfs_err(fs_info, "log_root level too big: %d >= %d", 256721a852b0SQu Wenruo btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 256821a852b0SQu Wenruo ret = -EINVAL; 256921a852b0SQu Wenruo } 257021a852b0SQu Wenruo 257121a852b0SQu Wenruo /* 257221a852b0SQu Wenruo * Check sectorsize and nodesize first, other check will need it. 257321a852b0SQu Wenruo * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. 257421a852b0SQu Wenruo */ 257521a852b0SQu Wenruo if (!is_power_of_2(sectorsize) || sectorsize < 4096 || 257621a852b0SQu Wenruo sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { 257721a852b0SQu Wenruo btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); 257821a852b0SQu Wenruo ret = -EINVAL; 257921a852b0SQu Wenruo } 25800bb3eb3eSQu Wenruo 25810bb3eb3eSQu Wenruo /* 25821a42daabSQu Wenruo * We only support at most two sectorsizes: 4K and PAGE_SIZE. 25831a42daabSQu Wenruo * 25841a42daabSQu Wenruo * We can support 16K sectorsize with 64K page size without problem, 25851a42daabSQu Wenruo * but such sectorsize/pagesize combination doesn't make much sense. 25861a42daabSQu Wenruo * 4K will be our future standard, PAGE_SIZE is supported from the very 25871a42daabSQu Wenruo * beginning. 25880bb3eb3eSQu Wenruo */ 25891a42daabSQu Wenruo if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) { 259021a852b0SQu Wenruo btrfs_err(fs_info, 25910bb3eb3eSQu Wenruo "sectorsize %llu not yet supported for page size %lu", 259221a852b0SQu Wenruo sectorsize, PAGE_SIZE); 259321a852b0SQu Wenruo ret = -EINVAL; 259421a852b0SQu Wenruo } 25950bb3eb3eSQu Wenruo 259621a852b0SQu Wenruo if (!is_power_of_2(nodesize) || nodesize < sectorsize || 259721a852b0SQu Wenruo nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { 259821a852b0SQu Wenruo btrfs_err(fs_info, "invalid nodesize %llu", nodesize); 259921a852b0SQu Wenruo ret = -EINVAL; 260021a852b0SQu Wenruo } 260121a852b0SQu Wenruo if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { 260221a852b0SQu Wenruo btrfs_err(fs_info, "invalid leafsize %u, should be %llu", 260321a852b0SQu Wenruo le32_to_cpu(sb->__unused_leafsize), nodesize); 260421a852b0SQu Wenruo ret = -EINVAL; 260521a852b0SQu Wenruo } 260621a852b0SQu Wenruo 260721a852b0SQu Wenruo /* Root alignment check */ 260821a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { 260921a852b0SQu Wenruo btrfs_warn(fs_info, "tree_root block unaligned: %llu", 261021a852b0SQu Wenruo btrfs_super_root(sb)); 261121a852b0SQu Wenruo ret = -EINVAL; 261221a852b0SQu Wenruo } 261321a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { 261421a852b0SQu Wenruo btrfs_warn(fs_info, "chunk_root block unaligned: %llu", 261521a852b0SQu Wenruo btrfs_super_chunk_root(sb)); 261621a852b0SQu Wenruo ret = -EINVAL; 261721a852b0SQu Wenruo } 261821a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { 261921a852b0SQu Wenruo btrfs_warn(fs_info, "log_root block unaligned: %llu", 262021a852b0SQu Wenruo btrfs_super_log_root(sb)); 262121a852b0SQu Wenruo ret = -EINVAL; 262221a852b0SQu Wenruo } 262321a852b0SQu Wenruo 2624aefd7f70SNikolay Borisov if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, 2625aefd7f70SNikolay Borisov BTRFS_FSID_SIZE)) { 2626aefd7f70SNikolay Borisov btrfs_err(fs_info, 2627aefd7f70SNikolay Borisov "superblock fsid doesn't match fsid of fs_devices: %pU != %pU", 2628aefd7f70SNikolay Borisov fs_info->super_copy->fsid, fs_info->fs_devices->fsid); 2629aefd7f70SNikolay Borisov ret = -EINVAL; 2630aefd7f70SNikolay Borisov } 2631aefd7f70SNikolay Borisov 2632aefd7f70SNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID) && 2633aefd7f70SNikolay Borisov memcmp(fs_info->fs_devices->metadata_uuid, 2634aefd7f70SNikolay Borisov fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) { 2635aefd7f70SNikolay Borisov btrfs_err(fs_info, 2636aefd7f70SNikolay Borisov "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU", 2637aefd7f70SNikolay Borisov fs_info->super_copy->metadata_uuid, 2638aefd7f70SNikolay Borisov fs_info->fs_devices->metadata_uuid); 2639aefd7f70SNikolay Borisov ret = -EINVAL; 2640aefd7f70SNikolay Borisov } 2641aefd7f70SNikolay Borisov 26421c56ab99SQu Wenruo /* 26431c56ab99SQu Wenruo * Artificial requirement for block-group-tree to force newer features 26441c56ab99SQu Wenruo * (free-space-tree, no-holes) so the test matrix is smaller. 26451c56ab99SQu Wenruo */ 26461c56ab99SQu Wenruo if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) && 26471c56ab99SQu Wenruo (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) || 26481c56ab99SQu Wenruo !btrfs_fs_incompat(fs_info, NO_HOLES))) { 26491c56ab99SQu Wenruo btrfs_err(fs_info, 26501c56ab99SQu Wenruo "block-group-tree feature requires fres-space-tree and no-holes"); 26511c56ab99SQu Wenruo ret = -EINVAL; 26521c56ab99SQu Wenruo } 26531c56ab99SQu Wenruo 2654de37aa51SNikolay Borisov if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, 26557239ff4bSNikolay Borisov BTRFS_FSID_SIZE) != 0) { 265621a852b0SQu Wenruo btrfs_err(fs_info, 26577239ff4bSNikolay Borisov "dev_item UUID does not match metadata fsid: %pU != %pU", 2658de37aa51SNikolay Borisov fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); 265921a852b0SQu Wenruo ret = -EINVAL; 266021a852b0SQu Wenruo } 266121a852b0SQu Wenruo 266221a852b0SQu Wenruo /* 266321a852b0SQu Wenruo * Hint to catch really bogus numbers, bitflips or so, more exact checks are 266421a852b0SQu Wenruo * done later 266521a852b0SQu Wenruo */ 266621a852b0SQu Wenruo if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { 266721a852b0SQu Wenruo btrfs_err(fs_info, "bytes_used is too small %llu", 266821a852b0SQu Wenruo btrfs_super_bytes_used(sb)); 266921a852b0SQu Wenruo ret = -EINVAL; 267021a852b0SQu Wenruo } 267121a852b0SQu Wenruo if (!is_power_of_2(btrfs_super_stripesize(sb))) { 267221a852b0SQu Wenruo btrfs_err(fs_info, "invalid stripesize %u", 267321a852b0SQu Wenruo btrfs_super_stripesize(sb)); 267421a852b0SQu Wenruo ret = -EINVAL; 267521a852b0SQu Wenruo } 267621a852b0SQu Wenruo if (btrfs_super_num_devices(sb) > (1UL << 31)) 267721a852b0SQu Wenruo btrfs_warn(fs_info, "suspicious number of devices: %llu", 267821a852b0SQu Wenruo btrfs_super_num_devices(sb)); 267921a852b0SQu Wenruo if (btrfs_super_num_devices(sb) == 0) { 268021a852b0SQu Wenruo btrfs_err(fs_info, "number of devices is 0"); 268121a852b0SQu Wenruo ret = -EINVAL; 268221a852b0SQu Wenruo } 268321a852b0SQu Wenruo 2684069ec957SQu Wenruo if (mirror_num >= 0 && 2685069ec957SQu Wenruo btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { 268621a852b0SQu Wenruo btrfs_err(fs_info, "super offset mismatch %llu != %u", 268721a852b0SQu Wenruo btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 268821a852b0SQu Wenruo ret = -EINVAL; 268921a852b0SQu Wenruo } 269021a852b0SQu Wenruo 269121a852b0SQu Wenruo /* 269221a852b0SQu Wenruo * Obvious sys_chunk_array corruptions, it must hold at least one key 269321a852b0SQu Wenruo * and one chunk 269421a852b0SQu Wenruo */ 269521a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 269621a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too big %u > %u", 269721a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 269821a852b0SQu Wenruo BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 269921a852b0SQu Wenruo ret = -EINVAL; 270021a852b0SQu Wenruo } 270121a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 270221a852b0SQu Wenruo + sizeof(struct btrfs_chunk)) { 270321a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too small %u < %zu", 270421a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 270521a852b0SQu Wenruo sizeof(struct btrfs_disk_key) 270621a852b0SQu Wenruo + sizeof(struct btrfs_chunk)); 270721a852b0SQu Wenruo ret = -EINVAL; 270821a852b0SQu Wenruo } 270921a852b0SQu Wenruo 271021a852b0SQu Wenruo /* 271121a852b0SQu Wenruo * The generation is a global counter, we'll trust it more than the others 271221a852b0SQu Wenruo * but it's still possible that it's the one that's wrong. 271321a852b0SQu Wenruo */ 271421a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 271521a852b0SQu Wenruo btrfs_warn(fs_info, 271621a852b0SQu Wenruo "suspicious: generation < chunk_root_generation: %llu < %llu", 271721a852b0SQu Wenruo btrfs_super_generation(sb), 271821a852b0SQu Wenruo btrfs_super_chunk_root_generation(sb)); 271921a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 272021a852b0SQu Wenruo && btrfs_super_cache_generation(sb) != (u64)-1) 272121a852b0SQu Wenruo btrfs_warn(fs_info, 272221a852b0SQu Wenruo "suspicious: generation < cache_generation: %llu < %llu", 272321a852b0SQu Wenruo btrfs_super_generation(sb), 272421a852b0SQu Wenruo btrfs_super_cache_generation(sb)); 272521a852b0SQu Wenruo 272621a852b0SQu Wenruo return ret; 272721a852b0SQu Wenruo } 272821a852b0SQu Wenruo 2729069ec957SQu Wenruo /* 2730069ec957SQu Wenruo * Validation of super block at mount time. 2731069ec957SQu Wenruo * Some checks already done early at mount time, like csum type and incompat 2732069ec957SQu Wenruo * flags will be skipped. 2733069ec957SQu Wenruo */ 2734069ec957SQu Wenruo static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) 2735069ec957SQu Wenruo { 2736a05d3c91SQu Wenruo return btrfs_validate_super(fs_info, fs_info->super_copy, 0); 2737069ec957SQu Wenruo } 2738069ec957SQu Wenruo 273975cb857dSQu Wenruo /* 274075cb857dSQu Wenruo * Validation of super block at write time. 274175cb857dSQu Wenruo * Some checks like bytenr check will be skipped as their values will be 274275cb857dSQu Wenruo * overwritten soon. 274375cb857dSQu Wenruo * Extra checks like csum type and incompat flags will be done here. 274475cb857dSQu Wenruo */ 274575cb857dSQu Wenruo static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, 274675cb857dSQu Wenruo struct btrfs_super_block *sb) 274775cb857dSQu Wenruo { 274875cb857dSQu Wenruo int ret; 274975cb857dSQu Wenruo 2750a05d3c91SQu Wenruo ret = btrfs_validate_super(fs_info, sb, -1); 275175cb857dSQu Wenruo if (ret < 0) 275275cb857dSQu Wenruo goto out; 2753e7e16f48SJohannes Thumshirn if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) { 275475cb857dSQu Wenruo ret = -EUCLEAN; 275575cb857dSQu Wenruo btrfs_err(fs_info, "invalid csum type, has %u want %u", 275675cb857dSQu Wenruo btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); 275775cb857dSQu Wenruo goto out; 275875cb857dSQu Wenruo } 275975cb857dSQu Wenruo if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 276075cb857dSQu Wenruo ret = -EUCLEAN; 276175cb857dSQu Wenruo btrfs_err(fs_info, 276275cb857dSQu Wenruo "invalid incompat flags, has 0x%llx valid mask 0x%llx", 276375cb857dSQu Wenruo btrfs_super_incompat_flags(sb), 276475cb857dSQu Wenruo (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); 276575cb857dSQu Wenruo goto out; 276675cb857dSQu Wenruo } 276775cb857dSQu Wenruo out: 276875cb857dSQu Wenruo if (ret < 0) 276975cb857dSQu Wenruo btrfs_err(fs_info, 277075cb857dSQu Wenruo "super block corruption detected before writing it to disk"); 277175cb857dSQu Wenruo return ret; 277275cb857dSQu Wenruo } 277375cb857dSQu Wenruo 2774bd676446SJosef Bacik static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level) 2775bd676446SJosef Bacik { 2776789d6a3aSQu Wenruo struct btrfs_tree_parent_check check = { 2777789d6a3aSQu Wenruo .level = level, 2778789d6a3aSQu Wenruo .transid = gen, 2779789d6a3aSQu Wenruo .owner_root = root->root_key.objectid 2780789d6a3aSQu Wenruo }; 2781bd676446SJosef Bacik int ret = 0; 2782bd676446SJosef Bacik 2783789d6a3aSQu Wenruo root->node = read_tree_block(root->fs_info, bytenr, &check); 2784bd676446SJosef Bacik if (IS_ERR(root->node)) { 2785bd676446SJosef Bacik ret = PTR_ERR(root->node); 2786bd676446SJosef Bacik root->node = NULL; 27874eb150d6SQu Wenruo return ret; 27884eb150d6SQu Wenruo } 27894eb150d6SQu Wenruo if (!extent_buffer_uptodate(root->node)) { 2790bd676446SJosef Bacik free_extent_buffer(root->node); 2791bd676446SJosef Bacik root->node = NULL; 27924eb150d6SQu Wenruo return -EIO; 2793bd676446SJosef Bacik } 2794bd676446SJosef Bacik 2795bd676446SJosef Bacik btrfs_set_root_node(&root->root_item, root->node); 2796bd676446SJosef Bacik root->commit_root = btrfs_root_node(root); 2797bd676446SJosef Bacik btrfs_set_root_refs(&root->root_item, 1); 2798bd676446SJosef Bacik return ret; 2799bd676446SJosef Bacik } 2800bd676446SJosef Bacik 2801bd676446SJosef Bacik static int load_important_roots(struct btrfs_fs_info *fs_info) 2802bd676446SJosef Bacik { 2803bd676446SJosef Bacik struct btrfs_super_block *sb = fs_info->super_copy; 2804bd676446SJosef Bacik u64 gen, bytenr; 2805bd676446SJosef Bacik int level, ret; 2806bd676446SJosef Bacik 2807bd676446SJosef Bacik bytenr = btrfs_super_root(sb); 2808bd676446SJosef Bacik gen = btrfs_super_generation(sb); 2809bd676446SJosef Bacik level = btrfs_super_root_level(sb); 2810bd676446SJosef Bacik ret = load_super_root(fs_info->tree_root, bytenr, gen, level); 28119c54e80dSJosef Bacik if (ret) { 2812bd676446SJosef Bacik btrfs_warn(fs_info, "couldn't read tree root"); 2813bd676446SJosef Bacik return ret; 2814bd676446SJosef Bacik } 28159c54e80dSJosef Bacik return 0; 28169c54e80dSJosef Bacik } 28179c54e80dSJosef Bacik 28186ef108ddSNikolay Borisov static int __cold init_tree_roots(struct btrfs_fs_info *fs_info) 2819b8522a1eSNikolay Borisov { 28206ef108ddSNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 2821b8522a1eSNikolay Borisov struct btrfs_super_block *sb = fs_info->super_copy; 2822b8522a1eSNikolay Borisov struct btrfs_root *tree_root = fs_info->tree_root; 2823b8522a1eSNikolay Borisov bool handle_error = false; 2824b8522a1eSNikolay Borisov int ret = 0; 2825b8522a1eSNikolay Borisov int i; 2826b8522a1eSNikolay Borisov 2827b8522a1eSNikolay Borisov for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 2828b8522a1eSNikolay Borisov if (handle_error) { 2829b8522a1eSNikolay Borisov if (!IS_ERR(tree_root->node)) 2830b8522a1eSNikolay Borisov free_extent_buffer(tree_root->node); 2831b8522a1eSNikolay Borisov tree_root->node = NULL; 2832b8522a1eSNikolay Borisov 2833b8522a1eSNikolay Borisov if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) 2834b8522a1eSNikolay Borisov break; 2835b8522a1eSNikolay Borisov 2836b8522a1eSNikolay Borisov free_root_pointers(fs_info, 0); 2837b8522a1eSNikolay Borisov 2838b8522a1eSNikolay Borisov /* 2839b8522a1eSNikolay Borisov * Don't use the log in recovery mode, it won't be 2840b8522a1eSNikolay Borisov * valid 2841b8522a1eSNikolay Borisov */ 2842b8522a1eSNikolay Borisov btrfs_set_super_log_root(sb, 0); 2843b8522a1eSNikolay Borisov 2844b8522a1eSNikolay Borisov /* We can't trust the free space cache either */ 2845b8522a1eSNikolay Borisov btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 2846b8522a1eSNikolay Borisov 2847b8522a1eSNikolay Borisov ret = read_backup_root(fs_info, i); 28486ef108ddSNikolay Borisov backup_index = ret; 2849b8522a1eSNikolay Borisov if (ret < 0) 2850b8522a1eSNikolay Borisov return ret; 2851b8522a1eSNikolay Borisov } 2852b8522a1eSNikolay Borisov 2853bd676446SJosef Bacik ret = load_important_roots(fs_info); 2854bd676446SJosef Bacik if (ret) { 2855217f5004SNikolay Borisov handle_error = true; 2856b8522a1eSNikolay Borisov continue; 2857b8522a1eSNikolay Borisov } 2858b8522a1eSNikolay Borisov 2859336a0d8dSNikolay Borisov /* 2860336a0d8dSNikolay Borisov * No need to hold btrfs_root::objectid_mutex since the fs 2861336a0d8dSNikolay Borisov * hasn't been fully initialised and we are the only user 2862336a0d8dSNikolay Borisov */ 2863453e4873SNikolay Borisov ret = btrfs_init_root_free_objectid(tree_root); 2864b8522a1eSNikolay Borisov if (ret < 0) { 2865b8522a1eSNikolay Borisov handle_error = true; 2866b8522a1eSNikolay Borisov continue; 2867b8522a1eSNikolay Borisov } 2868b8522a1eSNikolay Borisov 28696b8fad57SNikolay Borisov ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID); 2870b8522a1eSNikolay Borisov 2871b8522a1eSNikolay Borisov ret = btrfs_read_roots(fs_info); 2872b8522a1eSNikolay Borisov if (ret < 0) { 2873b8522a1eSNikolay Borisov handle_error = true; 2874b8522a1eSNikolay Borisov continue; 2875b8522a1eSNikolay Borisov } 2876b8522a1eSNikolay Borisov 2877b8522a1eSNikolay Borisov /* All successful */ 2878bd676446SJosef Bacik fs_info->generation = btrfs_header_generation(tree_root->node); 2879bd676446SJosef Bacik fs_info->last_trans_committed = fs_info->generation; 2880d96b3424SFilipe Manana fs_info->last_reloc_trans = 0; 28816ef108ddSNikolay Borisov 28826ef108ddSNikolay Borisov /* Always begin writing backup roots after the one being used */ 28836ef108ddSNikolay Borisov if (backup_index < 0) { 28846ef108ddSNikolay Borisov fs_info->backup_root_index = 0; 28856ef108ddSNikolay Borisov } else { 28866ef108ddSNikolay Borisov fs_info->backup_root_index = backup_index + 1; 28876ef108ddSNikolay Borisov fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS; 28886ef108ddSNikolay Borisov } 2889b8522a1eSNikolay Borisov break; 2890b8522a1eSNikolay Borisov } 2891b8522a1eSNikolay Borisov 2892b8522a1eSNikolay Borisov return ret; 2893b8522a1eSNikolay Borisov } 2894b8522a1eSNikolay Borisov 28958260edbaSJosef Bacik void btrfs_init_fs_info(struct btrfs_fs_info *fs_info) 2896eb60ceacSChris Mason { 2897fc7cbcd4SDavid Sterba INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 289801cd3909SDavid Sterba INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 28998fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 2900facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 290124bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 2902eb73c1b7SMiao Xie INIT_LIST_HEAD(&fs_info->delalloc_roots); 290311833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 2904eb73c1b7SMiao Xie spin_lock_init(&fs_info->delalloc_root_lock); 2905a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 2906fc7cbcd4SDavid Sterba spin_lock_init(&fs_info->fs_roots_radix_lock); 290724bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 29084cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 2909ceda0864SMiao Xie spin_lock_init(&fs_info->super_lock); 2910f28491e0SJosef Bacik spin_lock_init(&fs_info->buffer_lock); 291147ab2a6cSJosef Bacik spin_lock_init(&fs_info->unused_bgs_lock); 291240ab3be1SNaohiro Aota spin_lock_init(&fs_info->treelog_bg_lock); 2913afba2bc0SNaohiro Aota spin_lock_init(&fs_info->zone_active_bgs_lock); 2914c2707a25SJohannes Thumshirn spin_lock_init(&fs_info->relocation_bg_lock); 2915f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 2916abed4aaaSJosef Bacik rwlock_init(&fs_info->global_root_lock); 2917d7c15171SZhao Lei mutex_init(&fs_info->unused_bg_unpin_mutex); 2918f3372065SJohannes Thumshirn mutex_init(&fs_info->reclaim_bgs_lock); 29197585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 2920573bfb72SMiao Xie mutex_init(&fs_info->delalloc_root_mutex); 29210bc09ca1SNaohiro Aota mutex_init(&fs_info->zoned_meta_io_lock); 29225f0addf7SNaohiro Aota mutex_init(&fs_info->zoned_data_reloc_io_lock); 2923de98ced9SMiao Xie seqlock_init(&fs_info->profiles_lock); 292419c00ddcSChris Mason 2925e1489b4fSIoannis Angelakopoulos btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers); 29265a9ba670SIoannis Angelakopoulos btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters); 29278b53779eSIoannis Angelakopoulos btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered); 29285f4403e1SIoannis Angelakopoulos btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent); 29293e738c53SIoannis Angelakopoulos btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_start, 29303e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_COMMIT_START); 29313e738c53SIoannis Angelakopoulos btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked, 29323e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_UNBLOCKED); 29333e738c53SIoannis Angelakopoulos btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed, 29343e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 29353e738c53SIoannis Angelakopoulos btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed, 29363e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_COMPLETED); 2937e1489b4fSIoannis Angelakopoulos 29380b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 29396324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 2940f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 294147ab2a6cSJosef Bacik INIT_LIST_HEAD(&fs_info->unused_bgs); 294218bb8bbfSJohannes Thumshirn INIT_LIST_HEAD(&fs_info->reclaim_bgs); 2943afba2bc0SNaohiro Aota INIT_LIST_HEAD(&fs_info->zone_active_bgs); 2944bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 2945bd647ce3SJosef Bacik INIT_LIST_HEAD(&fs_info->allocated_roots); 29463fd63727SJosef Bacik INIT_LIST_HEAD(&fs_info->allocated_ebs); 29473fd63727SJosef Bacik spin_lock_init(&fs_info->eb_leak_lock); 2948bd647ce3SJosef Bacik #endif 2949c8bf1b67SDavid Sterba extent_map_tree_init(&fs_info->mapping_tree); 295066d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 295166d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 295266d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 295366d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 295466d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 295566d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 295666d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 2957ba2c4d4eSJosef Bacik btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, 2958ba2c4d4eSJosef Bacik BTRFS_BLOCK_RSV_DELREFS); 2959ba2c4d4eSJosef Bacik 2960771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 29614cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 2962034f784dSJosef Bacik atomic_set(&fs_info->nr_delayed_iputs, 0); 2963fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 2964abed4aaaSJosef Bacik fs_info->global_root_tree = RB_ROOT; 296595ac567aSFilipe David Borba Manana fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 29669ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 29674cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 2968a5ed45f8SNikolay Borisov atomic64_set(&fs_info->free_chunk_space, 0); 2969f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 29708b87dc17SDavid Sterba fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 2971f8c269d7SDavid Sterba fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 2972fd708b81SJosef Bacik btrfs_init_ref_verify(fs_info); 2973c8b97818SChris Mason 2974b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 2975b34b086cSChris Mason num_online_cpus() + 2, 8); 29760afbaf8cSChris Mason 2977199c2a9cSMiao Xie INIT_LIST_HEAD(&fs_info->ordered_roots); 2978199c2a9cSMiao Xie spin_lock_init(&fs_info->ordered_root_lock); 297969fe2d75SJosef Bacik 2980638aa7edSEric Sandeen btrfs_init_scrub(fs_info); 298121adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 298221adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 298321adbd5cSStefan Behrens #endif 2984779a65a4SEric Sandeen btrfs_init_balance(fs_info); 298557056740SJosef Bacik btrfs_init_async_reclaim_work(fs_info); 2986a2de733cSArne Jansen 298716b0c258SFilipe Manana rwlock_init(&fs_info->block_group_cache_lock); 298808dddb29SFilipe Manana fs_info->block_group_cache_tree = RB_ROOT_CACHED; 29890f9dd46cSJosef Bacik 2990fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &fs_info->excluded_extents, 299135da5a7eSDavid Sterba IO_TREE_FS_EXCLUDED_EXTENTS); 299239279cc3SChris Mason 29935a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 2994e02119d5SChris Mason mutex_init(&fs_info->tree_log_mutex); 2995925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 2996a74a4b97SChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 2997a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 29981bbc621eSChris Mason mutex_init(&fs_info->ro_block_group_mutex); 29999e351cc8SJosef Bacik init_rwsem(&fs_info->commit_root_sem); 3000c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 300176dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 3002803b2f54SStefan Behrens sema_init(&fs_info->uuid_tree_rescan_sem, 1); 3003fa9c0d79SChris Mason 3004ad618368SEric Sandeen btrfs_init_dev_replace_locks(fs_info); 3005f9e92e40SEric Sandeen btrfs_init_qgroup(fs_info); 3006b0643e59SDennis Zhou btrfs_discard_init(fs_info); 3007416ac51dSArne Jansen 3008fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 3009fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 3010fa9c0d79SChris Mason 3011e6dcd2dcSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 3012f9295749SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 3013bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 30144854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 3015034f784dSJosef Bacik init_waitqueue_head(&fs_info->delayed_iputs_wait); 30163768f368SChris Mason 3017da17066cSJeff Mahoney /* Usable values until the real ones are cached from the superblock */ 3018da17066cSJeff Mahoney fs_info->nodesize = 4096; 3019da17066cSJeff Mahoney fs_info->sectorsize = 4096; 3020ab108d99SDavid Sterba fs_info->sectorsize_bits = ilog2(4096); 3021da17066cSJeff Mahoney fs_info->stripesize = 4096; 3022da17066cSJeff Mahoney 3023f7b12a62SNaohiro Aota fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE; 3024f7b12a62SNaohiro Aota 3025eede2bf3SOmar Sandoval spin_lock_init(&fs_info->swapfile_pins_lock); 3026eede2bf3SOmar Sandoval fs_info->swapfile_pins = RB_ROOT; 3027eede2bf3SOmar Sandoval 302818bb8bbfSJohannes Thumshirn fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH; 302918bb8bbfSJohannes Thumshirn INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work); 30308260edbaSJosef Bacik } 30318260edbaSJosef Bacik 30328260edbaSJosef Bacik static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb) 30338260edbaSJosef Bacik { 30348260edbaSJosef Bacik int ret; 30358260edbaSJosef Bacik 30368260edbaSJosef Bacik fs_info->sb = sb; 30378260edbaSJosef Bacik sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; 30388260edbaSJosef Bacik sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); 30399e967495SFilipe Manana 30405deb17e1SJosef Bacik ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL); 3041ae18c37aSJosef Bacik if (ret) 3042ae18c37aSJosef Bacik return ret; 3043ae18c37aSJosef Bacik 3044ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 3045ae18c37aSJosef Bacik if (ret) 3046c75e8394SJosef Bacik return ret; 3047ae18c37aSJosef Bacik 3048ae18c37aSJosef Bacik fs_info->dirty_metadata_batch = PAGE_SIZE * 3049ae18c37aSJosef Bacik (1 + ilog2(nr_cpu_ids)); 3050ae18c37aSJosef Bacik 3051ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 3052ae18c37aSJosef Bacik if (ret) 3053c75e8394SJosef Bacik return ret; 3054ae18c37aSJosef Bacik 3055ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, 3056ae18c37aSJosef Bacik GFP_KERNEL); 3057ae18c37aSJosef Bacik if (ret) 3058c75e8394SJosef Bacik return ret; 3059ae18c37aSJosef Bacik 3060ae18c37aSJosef Bacik fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 3061ae18c37aSJosef Bacik GFP_KERNEL); 3062c75e8394SJosef Bacik if (!fs_info->delayed_root) 3063c75e8394SJosef Bacik return -ENOMEM; 3064ae18c37aSJosef Bacik btrfs_init_delayed_root(fs_info->delayed_root); 3065ae18c37aSJosef Bacik 3066a0a1db70SFilipe Manana if (sb_rdonly(sb)) 3067a0a1db70SFilipe Manana set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state); 3068a0a1db70SFilipe Manana 3069c75e8394SJosef Bacik return btrfs_alloc_stripe_hash_table(fs_info); 3070ae18c37aSJosef Bacik } 3071ae18c37aSJosef Bacik 307297f4dd09SNikolay Borisov static int btrfs_uuid_rescan_kthread(void *data) 307397f4dd09SNikolay Borisov { 30740d031dc4SYu Zhe struct btrfs_fs_info *fs_info = data; 307597f4dd09SNikolay Borisov int ret; 307697f4dd09SNikolay Borisov 307797f4dd09SNikolay Borisov /* 307897f4dd09SNikolay Borisov * 1st step is to iterate through the existing UUID tree and 307997f4dd09SNikolay Borisov * to delete all entries that contain outdated data. 308097f4dd09SNikolay Borisov * 2nd step is to add all missing entries to the UUID tree. 308197f4dd09SNikolay Borisov */ 308297f4dd09SNikolay Borisov ret = btrfs_uuid_tree_iterate(fs_info); 308397f4dd09SNikolay Borisov if (ret < 0) { 3084c94bec2cSJosef Bacik if (ret != -EINTR) 3085c94bec2cSJosef Bacik btrfs_warn(fs_info, "iterating uuid_tree failed %d", 3086c94bec2cSJosef Bacik ret); 308797f4dd09SNikolay Borisov up(&fs_info->uuid_tree_rescan_sem); 308897f4dd09SNikolay Borisov return ret; 308997f4dd09SNikolay Borisov } 309097f4dd09SNikolay Borisov return btrfs_uuid_scan_kthread(data); 309197f4dd09SNikolay Borisov } 309297f4dd09SNikolay Borisov 309397f4dd09SNikolay Borisov static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) 309497f4dd09SNikolay Borisov { 309597f4dd09SNikolay Borisov struct task_struct *task; 309697f4dd09SNikolay Borisov 309797f4dd09SNikolay Borisov down(&fs_info->uuid_tree_rescan_sem); 309897f4dd09SNikolay Borisov task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); 309997f4dd09SNikolay Borisov if (IS_ERR(task)) { 310097f4dd09SNikolay Borisov /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 310197f4dd09SNikolay Borisov btrfs_warn(fs_info, "failed to start uuid_rescan task"); 310297f4dd09SNikolay Borisov up(&fs_info->uuid_tree_rescan_sem); 310397f4dd09SNikolay Borisov return PTR_ERR(task); 310497f4dd09SNikolay Borisov } 310597f4dd09SNikolay Borisov 310697f4dd09SNikolay Borisov return 0; 310797f4dd09SNikolay Borisov } 310897f4dd09SNikolay Borisov 310944c0ca21SBoris Burkov /* 31108cd29088SBoris Burkov * Some options only have meaning at mount time and shouldn't persist across 31118cd29088SBoris Burkov * remounts, or be displayed. Clear these at the end of mount and remount 31128cd29088SBoris Burkov * code paths. 31138cd29088SBoris Burkov */ 31148cd29088SBoris Burkov void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info) 31158cd29088SBoris Burkov { 31168cd29088SBoris Burkov btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT); 31178b228324SBoris Burkov btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE); 31188cd29088SBoris Burkov } 31198cd29088SBoris Burkov 31208cd29088SBoris Burkov /* 312144c0ca21SBoris Burkov * Mounting logic specific to read-write file systems. Shared by open_ctree 312244c0ca21SBoris Burkov * and btrfs_remount when remounting from read-only to read-write. 312344c0ca21SBoris Burkov */ 312444c0ca21SBoris Burkov int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info) 312544c0ca21SBoris Burkov { 312644c0ca21SBoris Burkov int ret; 312794846229SBoris Burkov const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE); 31288b228324SBoris Burkov bool clear_free_space_tree = false; 31298b228324SBoris Burkov 31308b228324SBoris Burkov if (btrfs_test_opt(fs_info, CLEAR_CACHE) && 31318b228324SBoris Burkov btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 31328b228324SBoris Burkov clear_free_space_tree = true; 31338b228324SBoris Burkov } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 31348b228324SBoris Burkov !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { 31358b228324SBoris Burkov btrfs_warn(fs_info, "free space tree is invalid"); 31368b228324SBoris Burkov clear_free_space_tree = true; 31378b228324SBoris Burkov } 31388b228324SBoris Burkov 31398b228324SBoris Burkov if (clear_free_space_tree) { 31408b228324SBoris Burkov btrfs_info(fs_info, "clearing free space tree"); 31418b228324SBoris Burkov ret = btrfs_clear_free_space_tree(fs_info); 31428b228324SBoris Burkov if (ret) { 31438b228324SBoris Burkov btrfs_warn(fs_info, 31448b228324SBoris Burkov "failed to clear free space tree: %d", ret); 31458b228324SBoris Burkov goto out; 31468b228324SBoris Burkov } 31478b228324SBoris Burkov } 314844c0ca21SBoris Burkov 31498d488a8cSFilipe Manana /* 31508d488a8cSFilipe Manana * btrfs_find_orphan_roots() is responsible for finding all the dead 31518d488a8cSFilipe Manana * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load 3152fc7cbcd4SDavid Sterba * them into the fs_info->fs_roots_radix tree. This must be done before 31538d488a8cSFilipe Manana * calling btrfs_orphan_cleanup() on the tree root. If we don't do it 31548d488a8cSFilipe Manana * first, then btrfs_orphan_cleanup() will delete a dead root's orphan 31558d488a8cSFilipe Manana * item before the root's tree is deleted - this means that if we unmount 31568d488a8cSFilipe Manana * or crash before the deletion completes, on the next mount we will not 31578d488a8cSFilipe Manana * delete what remains of the tree because the orphan item does not 31588d488a8cSFilipe Manana * exists anymore, which is what tells us we have a pending deletion. 31598d488a8cSFilipe Manana */ 31608d488a8cSFilipe Manana ret = btrfs_find_orphan_roots(fs_info); 31618d488a8cSFilipe Manana if (ret) 31628d488a8cSFilipe Manana goto out; 31638d488a8cSFilipe Manana 316444c0ca21SBoris Burkov ret = btrfs_cleanup_fs_roots(fs_info); 316544c0ca21SBoris Burkov if (ret) 316644c0ca21SBoris Burkov goto out; 316744c0ca21SBoris Burkov 31688f1c21d7SBoris Burkov down_read(&fs_info->cleanup_work_sem); 31698f1c21d7SBoris Burkov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 31708f1c21d7SBoris Burkov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 31718f1c21d7SBoris Burkov up_read(&fs_info->cleanup_work_sem); 31728f1c21d7SBoris Burkov goto out; 31738f1c21d7SBoris Burkov } 31748f1c21d7SBoris Burkov up_read(&fs_info->cleanup_work_sem); 31758f1c21d7SBoris Burkov 317644c0ca21SBoris Burkov mutex_lock(&fs_info->cleaner_mutex); 31777eefae6bSJosef Bacik ret = btrfs_recover_relocation(fs_info); 317844c0ca21SBoris Burkov mutex_unlock(&fs_info->cleaner_mutex); 317944c0ca21SBoris Burkov if (ret < 0) { 318044c0ca21SBoris Burkov btrfs_warn(fs_info, "failed to recover relocation: %d", ret); 318144c0ca21SBoris Burkov goto out; 318244c0ca21SBoris Burkov } 318344c0ca21SBoris Burkov 31845011139aSBoris Burkov if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && 31855011139aSBoris Burkov !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 31865011139aSBoris Burkov btrfs_info(fs_info, "creating free space tree"); 31875011139aSBoris Burkov ret = btrfs_create_free_space_tree(fs_info); 31885011139aSBoris Burkov if (ret) { 31895011139aSBoris Burkov btrfs_warn(fs_info, 31905011139aSBoris Burkov "failed to create free space tree: %d", ret); 31915011139aSBoris Burkov goto out; 31925011139aSBoris Burkov } 31935011139aSBoris Burkov } 31945011139aSBoris Burkov 319594846229SBoris Burkov if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) { 319694846229SBoris Burkov ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt); 319794846229SBoris Burkov if (ret) 319894846229SBoris Burkov goto out; 319994846229SBoris Burkov } 320094846229SBoris Burkov 320144c0ca21SBoris Burkov ret = btrfs_resume_balance_async(fs_info); 320244c0ca21SBoris Burkov if (ret) 320344c0ca21SBoris Burkov goto out; 320444c0ca21SBoris Burkov 320544c0ca21SBoris Burkov ret = btrfs_resume_dev_replace_async(fs_info); 320644c0ca21SBoris Burkov if (ret) { 320744c0ca21SBoris Burkov btrfs_warn(fs_info, "failed to resume dev_replace"); 320844c0ca21SBoris Burkov goto out; 320944c0ca21SBoris Burkov } 321044c0ca21SBoris Burkov 321144c0ca21SBoris Burkov btrfs_qgroup_rescan_resume(fs_info); 321244c0ca21SBoris Burkov 321344c0ca21SBoris Burkov if (!fs_info->uuid_root) { 321444c0ca21SBoris Burkov btrfs_info(fs_info, "creating UUID tree"); 321544c0ca21SBoris Burkov ret = btrfs_create_uuid_tree(fs_info); 321644c0ca21SBoris Burkov if (ret) { 321744c0ca21SBoris Burkov btrfs_warn(fs_info, 321844c0ca21SBoris Burkov "failed to create the UUID tree %d", ret); 321944c0ca21SBoris Burkov goto out; 322044c0ca21SBoris Burkov } 322144c0ca21SBoris Burkov } 322244c0ca21SBoris Burkov 322344c0ca21SBoris Burkov out: 322444c0ca21SBoris Burkov return ret; 322544c0ca21SBoris Burkov } 322644c0ca21SBoris Burkov 3227d7f67ac9SQu Wenruo /* 3228d7f67ac9SQu Wenruo * Do various sanity and dependency checks of different features. 3229d7f67ac9SQu Wenruo * 32302ba48b20SQu Wenruo * @is_rw_mount: If the mount is read-write. 32312ba48b20SQu Wenruo * 3232d7f67ac9SQu Wenruo * This is the place for less strict checks (like for subpage or artificial 3233d7f67ac9SQu Wenruo * feature dependencies). 3234d7f67ac9SQu Wenruo * 3235d7f67ac9SQu Wenruo * For strict checks or possible corruption detection, see 3236d7f67ac9SQu Wenruo * btrfs_validate_super(). 3237d7f67ac9SQu Wenruo * 3238d7f67ac9SQu Wenruo * This should be called after btrfs_parse_options(), as some mount options 3239d7f67ac9SQu Wenruo * (space cache related) can modify on-disk format like free space tree and 3240d7f67ac9SQu Wenruo * screw up certain feature dependencies. 3241d7f67ac9SQu Wenruo */ 32422ba48b20SQu Wenruo int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount) 3243d7f67ac9SQu Wenruo { 3244d7f67ac9SQu Wenruo struct btrfs_super_block *disk_super = fs_info->super_copy; 3245d7f67ac9SQu Wenruo u64 incompat = btrfs_super_incompat_flags(disk_super); 3246d7f67ac9SQu Wenruo const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super); 3247d7f67ac9SQu Wenruo const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP); 3248d7f67ac9SQu Wenruo 3249d7f67ac9SQu Wenruo if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 3250d7f67ac9SQu Wenruo btrfs_err(fs_info, 3251d7f67ac9SQu Wenruo "cannot mount because of unknown incompat features (0x%llx)", 3252d7f67ac9SQu Wenruo incompat); 3253d7f67ac9SQu Wenruo return -EINVAL; 3254d7f67ac9SQu Wenruo } 3255d7f67ac9SQu Wenruo 3256d7f67ac9SQu Wenruo /* Runtime limitation for mixed block groups. */ 3257d7f67ac9SQu Wenruo if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 3258d7f67ac9SQu Wenruo (fs_info->sectorsize != fs_info->nodesize)) { 3259d7f67ac9SQu Wenruo btrfs_err(fs_info, 3260d7f67ac9SQu Wenruo "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", 3261d7f67ac9SQu Wenruo fs_info->nodesize, fs_info->sectorsize); 3262d7f67ac9SQu Wenruo return -EINVAL; 3263d7f67ac9SQu Wenruo } 3264d7f67ac9SQu Wenruo 3265d7f67ac9SQu Wenruo /* Mixed backref is an always-enabled feature. */ 3266d7f67ac9SQu Wenruo incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 3267d7f67ac9SQu Wenruo 3268d7f67ac9SQu Wenruo /* Set compression related flags just in case. */ 3269d7f67ac9SQu Wenruo if (fs_info->compress_type == BTRFS_COMPRESS_LZO) 3270d7f67ac9SQu Wenruo incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 3271d7f67ac9SQu Wenruo else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) 3272d7f67ac9SQu Wenruo incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; 3273d7f67ac9SQu Wenruo 3274d7f67ac9SQu Wenruo /* 3275d7f67ac9SQu Wenruo * An ancient flag, which should really be marked deprecated. 3276d7f67ac9SQu Wenruo * Such runtime limitation doesn't really need a incompat flag. 3277d7f67ac9SQu Wenruo */ 3278d7f67ac9SQu Wenruo if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) 3279d7f67ac9SQu Wenruo incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 3280d7f67ac9SQu Wenruo 32812ba48b20SQu Wenruo if (compat_ro_unsupp && is_rw_mount) { 3282d7f67ac9SQu Wenruo btrfs_err(fs_info, 3283d7f67ac9SQu Wenruo "cannot mount read-write because of unknown compat_ro features (0x%llx)", 3284d7f67ac9SQu Wenruo compat_ro); 3285d7f67ac9SQu Wenruo return -EINVAL; 3286d7f67ac9SQu Wenruo } 3287d7f67ac9SQu Wenruo 3288d7f67ac9SQu Wenruo /* 3289d7f67ac9SQu Wenruo * We have unsupported RO compat features, although RO mounted, we 3290d7f67ac9SQu Wenruo * should not cause any metadata writes, including log replay. 3291d7f67ac9SQu Wenruo * Or we could screw up whatever the new feature requires. 3292d7f67ac9SQu Wenruo */ 3293d7f67ac9SQu Wenruo if (compat_ro_unsupp && btrfs_super_log_root(disk_super) && 3294d7f67ac9SQu Wenruo !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 3295d7f67ac9SQu Wenruo btrfs_err(fs_info, 3296d7f67ac9SQu Wenruo "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay", 3297d7f67ac9SQu Wenruo compat_ro); 3298d7f67ac9SQu Wenruo return -EINVAL; 3299d7f67ac9SQu Wenruo } 3300d7f67ac9SQu Wenruo 3301d7f67ac9SQu Wenruo /* 3302d7f67ac9SQu Wenruo * Artificial limitations for block group tree, to force 3303d7f67ac9SQu Wenruo * block-group-tree to rely on no-holes and free-space-tree. 3304d7f67ac9SQu Wenruo */ 3305d7f67ac9SQu Wenruo if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) && 3306d7f67ac9SQu Wenruo (!btrfs_fs_incompat(fs_info, NO_HOLES) || 3307d7f67ac9SQu Wenruo !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) { 3308d7f67ac9SQu Wenruo btrfs_err(fs_info, 3309d7f67ac9SQu Wenruo "block-group-tree feature requires no-holes and free-space-tree features"); 3310d7f67ac9SQu Wenruo return -EINVAL; 3311d7f67ac9SQu Wenruo } 3312d7f67ac9SQu Wenruo 3313d7f67ac9SQu Wenruo /* 3314d7f67ac9SQu Wenruo * Subpage runtime limitation on v1 cache. 3315d7f67ac9SQu Wenruo * 3316d7f67ac9SQu Wenruo * V1 space cache still has some hard codeed PAGE_SIZE usage, while 3317d7f67ac9SQu Wenruo * we're already defaulting to v2 cache, no need to bother v1 as it's 3318d7f67ac9SQu Wenruo * going to be deprecated anyway. 3319d7f67ac9SQu Wenruo */ 3320d7f67ac9SQu Wenruo if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) { 3321d7f67ac9SQu Wenruo btrfs_warn(fs_info, 3322d7f67ac9SQu Wenruo "v1 space cache is not supported for page size %lu with sectorsize %u", 3323d7f67ac9SQu Wenruo PAGE_SIZE, fs_info->sectorsize); 3324d7f67ac9SQu Wenruo return -EINVAL; 3325d7f67ac9SQu Wenruo } 3326d7f67ac9SQu Wenruo 3327d7f67ac9SQu Wenruo /* This can be called by remount, we need to protect the super block. */ 3328d7f67ac9SQu Wenruo spin_lock(&fs_info->super_lock); 3329d7f67ac9SQu Wenruo btrfs_set_super_incompat_flags(disk_super, incompat); 3330d7f67ac9SQu Wenruo spin_unlock(&fs_info->super_lock); 3331d7f67ac9SQu Wenruo 3332d7f67ac9SQu Wenruo return 0; 3333d7f67ac9SQu Wenruo } 3334d7f67ac9SQu Wenruo 3335ae18c37aSJosef Bacik int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices, 3336ae18c37aSJosef Bacik char *options) 3337ae18c37aSJosef Bacik { 3338ae18c37aSJosef Bacik u32 sectorsize; 3339ae18c37aSJosef Bacik u32 nodesize; 3340ae18c37aSJosef Bacik u32 stripesize; 3341ae18c37aSJosef Bacik u64 generation; 3342ae18c37aSJosef Bacik u64 features; 3343ae18c37aSJosef Bacik u16 csum_type; 3344ae18c37aSJosef Bacik struct btrfs_super_block *disk_super; 3345ae18c37aSJosef Bacik struct btrfs_fs_info *fs_info = btrfs_sb(sb); 3346ae18c37aSJosef Bacik struct btrfs_root *tree_root; 3347ae18c37aSJosef Bacik struct btrfs_root *chunk_root; 3348ae18c37aSJosef Bacik int ret; 3349ae18c37aSJosef Bacik int err = -EINVAL; 3350ae18c37aSJosef Bacik int level; 3351ae18c37aSJosef Bacik 33528260edbaSJosef Bacik ret = init_mount_fs_info(fs_info, sb); 335353b381b3SDavid Woodhouse if (ret) { 335483c8266aSDavid Sterba err = ret; 3355ae18c37aSJosef Bacik goto fail; 335653b381b3SDavid Woodhouse } 335753b381b3SDavid Woodhouse 3358ae18c37aSJosef Bacik /* These need to be init'ed before we start creating inodes and such. */ 3359ae18c37aSJosef Bacik tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, 3360ae18c37aSJosef Bacik GFP_KERNEL); 3361ae18c37aSJosef Bacik fs_info->tree_root = tree_root; 3362ae18c37aSJosef Bacik chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID, 3363ae18c37aSJosef Bacik GFP_KERNEL); 3364ae18c37aSJosef Bacik fs_info->chunk_root = chunk_root; 3365ae18c37aSJosef Bacik if (!tree_root || !chunk_root) { 3366ae18c37aSJosef Bacik err = -ENOMEM; 3367c75e8394SJosef Bacik goto fail; 3368ae18c37aSJosef Bacik } 3369ae18c37aSJosef Bacik 3370ae18c37aSJosef Bacik fs_info->btree_inode = new_inode(sb); 3371ae18c37aSJosef Bacik if (!fs_info->btree_inode) { 3372ae18c37aSJosef Bacik err = -ENOMEM; 3373c75e8394SJosef Bacik goto fail; 3374ae18c37aSJosef Bacik } 3375ae18c37aSJosef Bacik mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 3376ae18c37aSJosef Bacik btrfs_init_btree_inode(fs_info); 3377ae18c37aSJosef Bacik 3378d24fa5c1SAnand Jain invalidate_bdev(fs_devices->latest_dev->bdev); 33791104a885SDavid Sterba 33801104a885SDavid Sterba /* 33811104a885SDavid Sterba * Read super block and check the signature bytes only 33821104a885SDavid Sterba */ 3383d24fa5c1SAnand Jain disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev); 33848f32380dSJohannes Thumshirn if (IS_ERR(disk_super)) { 33858f32380dSJohannes Thumshirn err = PTR_ERR(disk_super); 338616cdcec7SMiao Xie goto fail_alloc; 338720b45077SDave Young } 338839279cc3SChris Mason 33891104a885SDavid Sterba /* 3390260db43cSRandy Dunlap * Verify the type first, if that or the checksum value are 33918dc3f22cSJohannes Thumshirn * corrupted, we'll find out 33928dc3f22cSJohannes Thumshirn */ 33938f32380dSJohannes Thumshirn csum_type = btrfs_super_csum_type(disk_super); 339451bce6c9SJohannes Thumshirn if (!btrfs_supported_super_csum(csum_type)) { 33958dc3f22cSJohannes Thumshirn btrfs_err(fs_info, "unsupported checksum algorithm: %u", 339651bce6c9SJohannes Thumshirn csum_type); 33978dc3f22cSJohannes Thumshirn err = -EINVAL; 33988f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 33998dc3f22cSJohannes Thumshirn goto fail_alloc; 34008dc3f22cSJohannes Thumshirn } 34018dc3f22cSJohannes Thumshirn 340283c68bbcSSu Yue fs_info->csum_size = btrfs_super_csum_size(disk_super); 340383c68bbcSSu Yue 34046d97c6e3SJohannes Thumshirn ret = btrfs_init_csum_hash(fs_info, csum_type); 34056d97c6e3SJohannes Thumshirn if (ret) { 34066d97c6e3SJohannes Thumshirn err = ret; 34078f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 34086d97c6e3SJohannes Thumshirn goto fail_alloc; 34096d97c6e3SJohannes Thumshirn } 34106d97c6e3SJohannes Thumshirn 34118dc3f22cSJohannes Thumshirn /* 34121104a885SDavid Sterba * We want to check superblock checksum, the type is stored inside. 34131104a885SDavid Sterba * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 34141104a885SDavid Sterba */ 34153d17adeaSQu Wenruo if (btrfs_check_super_csum(fs_info, disk_super)) { 341605135f59SDavid Sterba btrfs_err(fs_info, "superblock checksum mismatch"); 34171104a885SDavid Sterba err = -EINVAL; 34188f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 3419141386e1SJosef Bacik goto fail_alloc; 34201104a885SDavid Sterba } 34211104a885SDavid Sterba 34221104a885SDavid Sterba /* 34231104a885SDavid Sterba * super_copy is zeroed at allocation time and we never touch the 34241104a885SDavid Sterba * following bytes up to INFO_SIZE, the checksum is calculated from 34251104a885SDavid Sterba * the whole block of INFO_SIZE 34261104a885SDavid Sterba */ 34278f32380dSJohannes Thumshirn memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy)); 34288f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 34295f39d397SChris Mason 3430fbc6feaeSNikolay Borisov disk_super = fs_info->super_copy; 3431fbc6feaeSNikolay Borisov 34320b86a832SChris Mason 3433fbc6feaeSNikolay Borisov features = btrfs_super_flags(disk_super); 3434fbc6feaeSNikolay Borisov if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { 3435fbc6feaeSNikolay Borisov features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; 3436fbc6feaeSNikolay Borisov btrfs_set_super_flags(disk_super, features); 3437fbc6feaeSNikolay Borisov btrfs_info(fs_info, 3438fbc6feaeSNikolay Borisov "found metadata UUID change in progress flag, clearing"); 3439fbc6feaeSNikolay Borisov } 3440fbc6feaeSNikolay Borisov 3441fbc6feaeSNikolay Borisov memcpy(fs_info->super_for_commit, fs_info->super_copy, 3442fbc6feaeSNikolay Borisov sizeof(*fs_info->super_for_commit)); 3443de37aa51SNikolay Borisov 3444069ec957SQu Wenruo ret = btrfs_validate_mount_super(fs_info); 34451104a885SDavid Sterba if (ret) { 344605135f59SDavid Sterba btrfs_err(fs_info, "superblock contains fatal errors"); 34471104a885SDavid Sterba err = -EINVAL; 3448141386e1SJosef Bacik goto fail_alloc; 34491104a885SDavid Sterba } 34501104a885SDavid Sterba 34510f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 3452141386e1SJosef Bacik goto fail_alloc; 34530f7d52f4SChris Mason 3454acce952bSliubo /* check FS state, whether FS is broken. */ 345587533c47SMiao Xie if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 345687533c47SMiao Xie set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 3457acce952bSliubo 345875e7cb7fSLiu Bo /* 345975e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 346075e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 346175e7cb7fSLiu Bo */ 346275e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 346375e7cb7fSLiu Bo 34646f93e834SAnand Jain 34656f93e834SAnand Jain /* Set up fs_info before parsing mount options */ 34666f93e834SAnand Jain nodesize = btrfs_super_nodesize(disk_super); 34676f93e834SAnand Jain sectorsize = btrfs_super_sectorsize(disk_super); 34686f93e834SAnand Jain stripesize = sectorsize; 34696f93e834SAnand Jain fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 34706f93e834SAnand Jain fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 34716f93e834SAnand Jain 34726f93e834SAnand Jain fs_info->nodesize = nodesize; 34736f93e834SAnand Jain fs_info->sectorsize = sectorsize; 34746f93e834SAnand Jain fs_info->sectorsize_bits = ilog2(sectorsize); 34756f93e834SAnand Jain fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size; 34766f93e834SAnand Jain fs_info->stripesize = stripesize; 34776f93e834SAnand Jain 34782ff7e61eSJeff Mahoney ret = btrfs_parse_options(fs_info, options, sb->s_flags); 34792b82032cSYan Zheng if (ret) { 34802b82032cSYan Zheng err = ret; 3481141386e1SJosef Bacik goto fail_alloc; 34822b82032cSYan Zheng } 3483dfe25020SChris Mason 34842ba48b20SQu Wenruo ret = btrfs_check_features(fs_info, !sb_rdonly(sb)); 3485d7f67ac9SQu Wenruo if (ret < 0) { 3486d7f67ac9SQu Wenruo err = ret; 3487141386e1SJosef Bacik goto fail_alloc; 3488f2b636e8SJosef Bacik } 3489f2b636e8SJosef Bacik 34908481dd80SQu Wenruo if (sectorsize < PAGE_SIZE) { 34918481dd80SQu Wenruo struct btrfs_subpage_info *subpage_info; 34928481dd80SQu Wenruo 34939f73f1aeSQu Wenruo /* 34949f73f1aeSQu Wenruo * V1 space cache has some hardcoded PAGE_SIZE usage, and is 34959f73f1aeSQu Wenruo * going to be deprecated. 34969f73f1aeSQu Wenruo * 34979f73f1aeSQu Wenruo * Force to use v2 cache for subpage case. 34989f73f1aeSQu Wenruo */ 34999f73f1aeSQu Wenruo btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE); 35009f73f1aeSQu Wenruo btrfs_set_and_info(fs_info, FREE_SPACE_TREE, 35019f73f1aeSQu Wenruo "forcing free space tree for sector size %u with page size %lu", 35029f73f1aeSQu Wenruo sectorsize, PAGE_SIZE); 35039f73f1aeSQu Wenruo 350495ea0486SQu Wenruo btrfs_warn(fs_info, 350595ea0486SQu Wenruo "read-write for sector size %u with page size %lu is experimental", 35060bb3eb3eSQu Wenruo sectorsize, PAGE_SIZE); 35078481dd80SQu Wenruo subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL); 35088481dd80SQu Wenruo if (!subpage_info) 35098481dd80SQu Wenruo goto fail_alloc; 35108481dd80SQu Wenruo btrfs_init_subpage_info(subpage_info, sectorsize); 35118481dd80SQu Wenruo fs_info->subpage_info = subpage_info; 3512c8050b3bSQu Wenruo } 35130bb3eb3eSQu Wenruo 3514d21deec5SSu Yue ret = btrfs_init_workqueues(fs_info); 35152a458198SEric Sandeen if (ret) { 35162a458198SEric Sandeen err = ret; 35170dc3b84aSJosef Bacik goto fail_sb_buffer; 35180dc3b84aSJosef Bacik } 35194543df7eSChris Mason 35209e11ceeeSJan Kara sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); 35219e11ceeeSJan Kara sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); 35224575c9ccSChris Mason 3523a061fc8dSChris Mason sb->s_blocksize = sectorsize; 3524a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 3525de37aa51SNikolay Borisov memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); 3526db94535dSChris Mason 3527925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 35286bccf3abSJeff Mahoney ret = btrfs_read_sys_array(fs_info); 3529925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 353084eed90fSChris Mason if (ret) { 353105135f59SDavid Sterba btrfs_err(fs_info, "failed to read the system array: %d", ret); 35325d4f98a2SYan Zheng goto fail_sb_buffer; 353384eed90fSChris Mason } 35340b86a832SChris Mason 353584234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 3536581c1760SQu Wenruo level = btrfs_super_chunk_root_level(disk_super); 3537bd676446SJosef Bacik ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super), 3538bd676446SJosef Bacik generation, level); 3539bd676446SJosef Bacik if (ret) { 354005135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk root"); 3541af31f5e5SChris Mason goto fail_tree_roots; 354283121942SDavid Woodhouse } 35430b86a832SChris Mason 3544e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 3545c4ac7541SDavid Sterba offsetof(struct btrfs_header, chunk_tree_uuid), 3546c4ac7541SDavid Sterba BTRFS_UUID_SIZE); 3547e17cade2SChris Mason 35485b4aacefSJeff Mahoney ret = btrfs_read_chunk_tree(fs_info); 35492b82032cSYan Zheng if (ret) { 355005135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk tree: %d", ret); 3551af31f5e5SChris Mason goto fail_tree_roots; 35522b82032cSYan Zheng } 35530b86a832SChris Mason 35548dabb742SStefan Behrens /* 3555bacce86aSAnand Jain * At this point we know all the devices that make this filesystem, 3556bacce86aSAnand Jain * including the seed devices but we don't know yet if the replace 3557bacce86aSAnand Jain * target is required. So free devices that are not part of this 35581a9fd417SDavid Sterba * filesystem but skip the replace target device which is checked 3559bacce86aSAnand Jain * below in btrfs_init_dev_replace(). 35608dabb742SStefan Behrens */ 3561bacce86aSAnand Jain btrfs_free_extra_devids(fs_devices); 3562d24fa5c1SAnand Jain if (!fs_devices->latest_dev->bdev) { 356305135f59SDavid Sterba btrfs_err(fs_info, "failed to read devices"); 3564a6b0d5c8SChris Mason goto fail_tree_roots; 3565a6b0d5c8SChris Mason } 3566a6b0d5c8SChris Mason 3567b8522a1eSNikolay Borisov ret = init_tree_roots(fs_info); 35684bbcaa64SEric Sandeen if (ret) 3569b8522a1eSNikolay Borisov goto fail_tree_roots; 35708929ecfaSYan, Zheng 357175ec1db8SJosef Bacik /* 357273651042SNaohiro Aota * Get zone type information of zoned block devices. This will also 357373651042SNaohiro Aota * handle emulation of a zoned filesystem if a regular device has the 357473651042SNaohiro Aota * zoned incompat feature flag set. 357573651042SNaohiro Aota */ 357673651042SNaohiro Aota ret = btrfs_get_dev_zone_info_all_devices(fs_info); 357773651042SNaohiro Aota if (ret) { 357873651042SNaohiro Aota btrfs_err(fs_info, 357973651042SNaohiro Aota "zoned: failed to read device zone info: %d", 358073651042SNaohiro Aota ret); 358173651042SNaohiro Aota goto fail_block_groups; 358273651042SNaohiro Aota } 358373651042SNaohiro Aota 358473651042SNaohiro Aota /* 358575ec1db8SJosef Bacik * If we have a uuid root and we're not being told to rescan we need to 358675ec1db8SJosef Bacik * check the generation here so we can set the 358775ec1db8SJosef Bacik * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the 358875ec1db8SJosef Bacik * transaction during a balance or the log replay without updating the 358975ec1db8SJosef Bacik * uuid generation, and then if we crash we would rescan the uuid tree, 359075ec1db8SJosef Bacik * even though it was perfectly fine. 359175ec1db8SJosef Bacik */ 359275ec1db8SJosef Bacik if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) && 359375ec1db8SJosef Bacik fs_info->generation == btrfs_super_uuid_tree_generation(disk_super)) 359475ec1db8SJosef Bacik set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 359575ec1db8SJosef Bacik 3596cf90d884SQu Wenruo ret = btrfs_verify_dev_extents(fs_info); 3597cf90d884SQu Wenruo if (ret) { 3598cf90d884SQu Wenruo btrfs_err(fs_info, 3599cf90d884SQu Wenruo "failed to verify dev extents against chunks: %d", 3600cf90d884SQu Wenruo ret); 3601cf90d884SQu Wenruo goto fail_block_groups; 3602cf90d884SQu Wenruo } 360368310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 360468310a5eSIlya Dryomov if (ret) { 360505135f59SDavid Sterba btrfs_err(fs_info, "failed to recover balance: %d", ret); 360668310a5eSIlya Dryomov goto fail_block_groups; 360768310a5eSIlya Dryomov } 360868310a5eSIlya Dryomov 3609733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 3610733f4fbbSStefan Behrens if (ret) { 361105135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_stats: %d", ret); 3612733f4fbbSStefan Behrens goto fail_block_groups; 3613733f4fbbSStefan Behrens } 3614733f4fbbSStefan Behrens 36158dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 36168dabb742SStefan Behrens if (ret) { 361705135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_replace: %d", ret); 36188dabb742SStefan Behrens goto fail_block_groups; 36198dabb742SStefan Behrens } 36208dabb742SStefan Behrens 3621b70f5097SNaohiro Aota ret = btrfs_check_zoned_mode(fs_info); 3622b70f5097SNaohiro Aota if (ret) { 3623b70f5097SNaohiro Aota btrfs_err(fs_info, "failed to initialize zoned mode: %d", 3624b70f5097SNaohiro Aota ret); 3625b70f5097SNaohiro Aota goto fail_block_groups; 3626b70f5097SNaohiro Aota } 3627b70f5097SNaohiro Aota 3628c6761a9eSAnand Jain ret = btrfs_sysfs_add_fsid(fs_devices); 3629b7c35e81SAnand Jain if (ret) { 363005135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", 363105135f59SDavid Sterba ret); 3632b7c35e81SAnand Jain goto fail_block_groups; 3633b7c35e81SAnand Jain } 3634b7c35e81SAnand Jain 363596f3136eSAnand Jain ret = btrfs_sysfs_add_mounted(fs_info); 36365ac1d209SJeff Mahoney if (ret) { 363705135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); 3638b7c35e81SAnand Jain goto fail_fsdev_sysfs; 36395ac1d209SJeff Mahoney } 36405ac1d209SJeff Mahoney 3641c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 3642c59021f8Sliubo if (ret) { 364305135f59SDavid Sterba btrfs_err(fs_info, "failed to initialize space info: %d", ret); 36442365dd3cSAnand Jain goto fail_sysfs; 3645c59021f8Sliubo } 3646c59021f8Sliubo 36475b4aacefSJeff Mahoney ret = btrfs_read_block_groups(fs_info); 36481b1d1f66SJosef Bacik if (ret) { 364905135f59SDavid Sterba btrfs_err(fs_info, "failed to read block groups: %d", ret); 36502365dd3cSAnand Jain goto fail_sysfs; 36511b1d1f66SJosef Bacik } 36524330e183SQu Wenruo 365316beac87SNaohiro Aota btrfs_free_zone_cache(fs_info); 365416beac87SNaohiro Aota 36555c78a5e7SAnand Jain if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices && 36565c78a5e7SAnand Jain !btrfs_check_rw_degradable(fs_info, NULL)) { 365705135f59SDavid Sterba btrfs_warn(fs_info, 365852042d8eSAndrea Gelmini "writable mount is not allowed due to too many missing devices"); 36592365dd3cSAnand Jain goto fail_sysfs; 3660292fd7fcSStefan Behrens } 36619078a3e1SChris Mason 366233c44184SJosef Bacik fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info, 3663a74a4b97SChris Mason "btrfs-cleaner"); 366457506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 36652365dd3cSAnand Jain goto fail_sysfs; 3666a74a4b97SChris Mason 3667a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 3668a74a4b97SChris Mason tree_root, 3669a74a4b97SChris Mason "btrfs-transaction"); 367057506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 36713f157a2fSChris Mason goto fail_cleaner; 3672a74a4b97SChris Mason 3673583b7231SHans van Kranenburg if (!btrfs_test_opt(fs_info, NOSSD) && 3674c289811cSChris Mason !fs_info->fs_devices->rotating) { 3675583b7231SHans van Kranenburg btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); 3676c289811cSChris Mason } 3677c289811cSChris Mason 3678572d9ab7SDavid Sterba /* 367963a7cb13SDavid Sterba * For devices supporting discard turn on discard=async automatically, 368063a7cb13SDavid Sterba * unless it's already set or disabled. This could be turned off by 368163a7cb13SDavid Sterba * nodiscard for the same mount. 368263a7cb13SDavid Sterba */ 368363a7cb13SDavid Sterba if (!(btrfs_test_opt(fs_info, DISCARD_SYNC) || 368463a7cb13SDavid Sterba btrfs_test_opt(fs_info, DISCARD_ASYNC) || 368563a7cb13SDavid Sterba btrfs_test_opt(fs_info, NODISCARD)) && 368663a7cb13SDavid Sterba fs_info->fs_devices->discardable) { 368763a7cb13SDavid Sterba btrfs_set_and_info(fs_info, DISCARD_ASYNC, 368863a7cb13SDavid Sterba "auto enabling async discard"); 368963a7cb13SDavid Sterba btrfs_clear_opt(fs_info->mount_opt, NODISCARD); 369063a7cb13SDavid Sterba } 369163a7cb13SDavid Sterba 369221adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 36930b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { 36942ff7e61eSJeff Mahoney ret = btrfsic_mount(fs_info, fs_devices, 36950b246afaSJeff Mahoney btrfs_test_opt(fs_info, 3696cbeaae4fSDavid Sterba CHECK_INTEGRITY_DATA) ? 1 : 0, 369721adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 369821adbd5cSStefan Behrens if (ret) 369905135f59SDavid Sterba btrfs_warn(fs_info, 370005135f59SDavid Sterba "failed to initialize integrity check module: %d", 370105135f59SDavid Sterba ret); 370221adbd5cSStefan Behrens } 370321adbd5cSStefan Behrens #endif 3704bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 3705bcef60f2SArne Jansen if (ret) 3706bcef60f2SArne Jansen goto fail_trans_kthread; 370721adbd5cSStefan Behrens 3708fd708b81SJosef Bacik if (btrfs_build_ref_tree(fs_info)) 3709fd708b81SJosef Bacik btrfs_err(fs_info, "couldn't build ref tree"); 3710fd708b81SJosef Bacik 371196da0919SQu Wenruo /* do not make disk changes in broken FS or nologreplay is given */ 371296da0919SQu Wenruo if (btrfs_super_log_root(disk_super) != 0 && 37130b246afaSJeff Mahoney !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 3714e8294f2fSDavid Sterba btrfs_info(fs_info, "start tree-log replay"); 371563443bf5SEric Sandeen ret = btrfs_replay_log(fs_info, fs_devices); 371679787eaaSJeff Mahoney if (ret) { 371763443bf5SEric Sandeen err = ret; 371828c16cbbSWang Shilong goto fail_qgroup; 3719e556ce2cSYan Zheng } 3720e02119d5SChris Mason } 37211a40e23bSZheng Yan 372256e9357aSDavid Sterba fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true); 37233140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 37243140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 3725f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read fs tree: %d", err); 3726315bf8efSJosef Bacik fs_info->fs_root = NULL; 3727bcef60f2SArne Jansen goto fail_qgroup; 37283140c9a3SDan Carpenter } 3729c289811cSChris Mason 3730bc98a42cSDavid Howells if (sb_rdonly(sb)) 37318cd29088SBoris Burkov goto clear_oneshot; 37322b6ba629SIlya Dryomov 373344c0ca21SBoris Burkov ret = btrfs_start_pre_rw_mount(fs_info); 37342b6ba629SIlya Dryomov if (ret) { 37356bccf3abSJeff Mahoney close_ctree(fs_info); 37362b6ba629SIlya Dryomov return ret; 3737e3acc2a6SJosef Bacik } 3738b0643e59SDennis Zhou btrfs_discard_resume(fs_info); 3739b382a324SJan Schmidt 374044c0ca21SBoris Burkov if (fs_info->uuid_root && 374144c0ca21SBoris Burkov (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || 374244c0ca21SBoris Burkov fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) { 374305135f59SDavid Sterba btrfs_info(fs_info, "checking UUID tree"); 374470f80175SStefan Behrens ret = btrfs_check_uuid_tree(fs_info); 374570f80175SStefan Behrens if (ret) { 374605135f59SDavid Sterba btrfs_warn(fs_info, 374705135f59SDavid Sterba "failed to check the UUID tree: %d", ret); 37486bccf3abSJeff Mahoney close_ctree(fs_info); 374970f80175SStefan Behrens return ret; 375070f80175SStefan Behrens } 3751f7a81ea4SStefan Behrens } 375294846229SBoris Burkov 3753afcdd129SJosef Bacik set_bit(BTRFS_FS_OPEN, &fs_info->flags); 375447ab2a6cSJosef Bacik 3755b4be6aefSJosef Bacik /* Kick the cleaner thread so it'll start deleting snapshots. */ 3756b4be6aefSJosef Bacik if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags)) 3757b4be6aefSJosef Bacik wake_up_process(fs_info->cleaner_kthread); 3758b4be6aefSJosef Bacik 37598cd29088SBoris Burkov clear_oneshot: 37608cd29088SBoris Burkov btrfs_clear_oneshot_options(fs_info); 3761ad2b2c80SAl Viro return 0; 376239279cc3SChris Mason 3763bcef60f2SArne Jansen fail_qgroup: 3764bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 37657c2ca468SChris Mason fail_trans_kthread: 37667c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 37672ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 3768faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 37693f157a2fSChris Mason fail_cleaner: 3770a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 37717c2ca468SChris Mason 37727c2ca468SChris Mason /* 37737c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 37747c2ca468SChris Mason * kthreads 37757c2ca468SChris Mason */ 37767c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 37777c2ca468SChris Mason 37782365dd3cSAnand Jain fail_sysfs: 37796618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 37802365dd3cSAnand Jain 3781b7c35e81SAnand Jain fail_fsdev_sysfs: 3782b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3783b7c35e81SAnand Jain 37841b1d1f66SJosef Bacik fail_block_groups: 378554067ae9SJosef Bacik btrfs_put_block_group_cache(fs_info); 3786af31f5e5SChris Mason 3787af31f5e5SChris Mason fail_tree_roots: 37889e3aa805SJosef Bacik if (fs_info->data_reloc_root) 37899e3aa805SJosef Bacik btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root); 37904273eaffSAnand Jain free_root_pointers(fs_info, true); 37912b8195bbSMiao Xie invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3792af31f5e5SChris Mason 379339279cc3SChris Mason fail_sb_buffer: 37947abadb64SLiu Bo btrfs_stop_all_workers(fs_info); 37955cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 379616cdcec7SMiao Xie fail_alloc: 3797586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 3798586e46e2SIlya Dryomov 37994543df7eSChris Mason iput(fs_info->btree_inode); 38007e662854SQinghuang Feng fail: 3801586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 3802ad2b2c80SAl Viro return err; 3803eb60ceacSChris Mason } 3804663faf9fSMasami Hiramatsu ALLOW_ERROR_INJECTION(open_ctree, ERRNO); 3805eb60ceacSChris Mason 3806314b6dd0SJohannes Thumshirn static void btrfs_end_super_write(struct bio *bio) 3807f2984462SChris Mason { 3808314b6dd0SJohannes Thumshirn struct btrfs_device *device = bio->bi_private; 3809314b6dd0SJohannes Thumshirn struct bio_vec *bvec; 3810314b6dd0SJohannes Thumshirn struct bvec_iter_all iter_all; 3811314b6dd0SJohannes Thumshirn struct page *page; 3812442a4f63SStefan Behrens 3813314b6dd0SJohannes Thumshirn bio_for_each_segment_all(bvec, bio, iter_all) { 3814314b6dd0SJohannes Thumshirn page = bvec->bv_page; 3815314b6dd0SJohannes Thumshirn 3816314b6dd0SJohannes Thumshirn if (bio->bi_status) { 3817fb456252SJeff Mahoney btrfs_warn_rl_in_rcu(device->fs_info, 3818314b6dd0SJohannes Thumshirn "lost page write due to IO error on %s (%d)", 3819cb3e217bSQu Wenruo btrfs_dev_name(device), 3820314b6dd0SJohannes Thumshirn blk_status_to_errno(bio->bi_status)); 3821314b6dd0SJohannes Thumshirn ClearPageUptodate(page); 3822314b6dd0SJohannes Thumshirn SetPageError(page); 3823314b6dd0SJohannes Thumshirn btrfs_dev_stat_inc_and_print(device, 3824314b6dd0SJohannes Thumshirn BTRFS_DEV_STAT_WRITE_ERRS); 3825314b6dd0SJohannes Thumshirn } else { 3826314b6dd0SJohannes Thumshirn SetPageUptodate(page); 3827f2984462SChris Mason } 3828314b6dd0SJohannes Thumshirn 3829314b6dd0SJohannes Thumshirn put_page(page); 3830314b6dd0SJohannes Thumshirn unlock_page(page); 3831314b6dd0SJohannes Thumshirn } 3832314b6dd0SJohannes Thumshirn 3833314b6dd0SJohannes Thumshirn bio_put(bio); 3834f2984462SChris Mason } 3835f2984462SChris Mason 38368f32380dSJohannes Thumshirn struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev, 3837a05d3c91SQu Wenruo int copy_num, bool drop_cache) 383829c36d72SAnand Jain { 383929c36d72SAnand Jain struct btrfs_super_block *super; 38408f32380dSJohannes Thumshirn struct page *page; 384112659251SNaohiro Aota u64 bytenr, bytenr_orig; 38428f32380dSJohannes Thumshirn struct address_space *mapping = bdev->bd_inode->i_mapping; 384312659251SNaohiro Aota int ret; 384429c36d72SAnand Jain 384512659251SNaohiro Aota bytenr_orig = btrfs_sb_offset(copy_num); 384612659251SNaohiro Aota ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr); 384712659251SNaohiro Aota if (ret == -ENOENT) 384812659251SNaohiro Aota return ERR_PTR(-EINVAL); 384912659251SNaohiro Aota else if (ret) 385012659251SNaohiro Aota return ERR_PTR(ret); 385112659251SNaohiro Aota 3852cda00ebaSChristoph Hellwig if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev)) 38538f32380dSJohannes Thumshirn return ERR_PTR(-EINVAL); 385429c36d72SAnand Jain 3855a05d3c91SQu Wenruo if (drop_cache) { 3856a05d3c91SQu Wenruo /* This should only be called with the primary sb. */ 3857a05d3c91SQu Wenruo ASSERT(copy_num == 0); 3858a05d3c91SQu Wenruo 3859a05d3c91SQu Wenruo /* 3860a05d3c91SQu Wenruo * Drop the page of the primary superblock, so later read will 3861a05d3c91SQu Wenruo * always read from the device. 3862a05d3c91SQu Wenruo */ 3863a05d3c91SQu Wenruo invalidate_inode_pages2_range(mapping, 3864a05d3c91SQu Wenruo bytenr >> PAGE_SHIFT, 3865a05d3c91SQu Wenruo (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT); 3866a05d3c91SQu Wenruo } 3867a05d3c91SQu Wenruo 38688f32380dSJohannes Thumshirn page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS); 38698f32380dSJohannes Thumshirn if (IS_ERR(page)) 38708f32380dSJohannes Thumshirn return ERR_CAST(page); 387129c36d72SAnand Jain 38728f32380dSJohannes Thumshirn super = page_address(page); 387396c2e067SAnand Jain if (btrfs_super_magic(super) != BTRFS_MAGIC) { 387496c2e067SAnand Jain btrfs_release_disk_super(super); 387596c2e067SAnand Jain return ERR_PTR(-ENODATA); 387696c2e067SAnand Jain } 387796c2e067SAnand Jain 387812659251SNaohiro Aota if (btrfs_super_bytenr(super) != bytenr_orig) { 38798f32380dSJohannes Thumshirn btrfs_release_disk_super(super); 38808f32380dSJohannes Thumshirn return ERR_PTR(-EINVAL); 388129c36d72SAnand Jain } 388229c36d72SAnand Jain 38838f32380dSJohannes Thumshirn return super; 388429c36d72SAnand Jain } 388529c36d72SAnand Jain 388629c36d72SAnand Jain 38878f32380dSJohannes Thumshirn struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev) 3888a512bbf8SYan Zheng { 38898f32380dSJohannes Thumshirn struct btrfs_super_block *super, *latest = NULL; 3890a512bbf8SYan Zheng int i; 3891a512bbf8SYan Zheng u64 transid = 0; 3892a512bbf8SYan Zheng 3893a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 3894a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 3895a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 3896a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3897a512bbf8SYan Zheng */ 3898a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 3899a05d3c91SQu Wenruo super = btrfs_read_dev_one_super(bdev, i, false); 39008f32380dSJohannes Thumshirn if (IS_ERR(super)) 3901a512bbf8SYan Zheng continue; 3902a512bbf8SYan Zheng 3903a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 39048f32380dSJohannes Thumshirn if (latest) 39058f32380dSJohannes Thumshirn btrfs_release_disk_super(super); 39068f32380dSJohannes Thumshirn 39078f32380dSJohannes Thumshirn latest = super; 3908a512bbf8SYan Zheng transid = btrfs_super_generation(super); 3909a512bbf8SYan Zheng } 3910a512bbf8SYan Zheng } 391192fc03fbSAnand Jain 39128f32380dSJohannes Thumshirn return super; 3913a512bbf8SYan Zheng } 3914a512bbf8SYan Zheng 39154eedeb75SHisashi Hifumi /* 3916abbb3b8eSDavid Sterba * Write superblock @sb to the @device. Do not wait for completion, all the 3917314b6dd0SJohannes Thumshirn * pages we use for writing are locked. 39184eedeb75SHisashi Hifumi * 3919abbb3b8eSDavid Sterba * Write @max_mirrors copies of the superblock, where 0 means default that fit 3920abbb3b8eSDavid Sterba * the expected device size at commit time. Note that max_mirrors must be 3921abbb3b8eSDavid Sterba * same for write and wait phases. 39224eedeb75SHisashi Hifumi * 3923314b6dd0SJohannes Thumshirn * Return number of errors when page is not found or submission fails. 39244eedeb75SHisashi Hifumi */ 3925a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 3926abbb3b8eSDavid Sterba struct btrfs_super_block *sb, int max_mirrors) 3927a512bbf8SYan Zheng { 3928d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = device->fs_info; 3929314b6dd0SJohannes Thumshirn struct address_space *mapping = device->bdev->bd_inode->i_mapping; 3930d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 3931a512bbf8SYan Zheng int i; 3932a512bbf8SYan Zheng int errors = 0; 393312659251SNaohiro Aota int ret; 393412659251SNaohiro Aota u64 bytenr, bytenr_orig; 3935a512bbf8SYan Zheng 3936a512bbf8SYan Zheng if (max_mirrors == 0) 3937a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3938a512bbf8SYan Zheng 3939d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 3940d5178578SJohannes Thumshirn 3941a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 3942314b6dd0SJohannes Thumshirn struct page *page; 3943314b6dd0SJohannes Thumshirn struct bio *bio; 3944314b6dd0SJohannes Thumshirn struct btrfs_super_block *disk_super; 3945314b6dd0SJohannes Thumshirn 394612659251SNaohiro Aota bytenr_orig = btrfs_sb_offset(i); 394712659251SNaohiro Aota ret = btrfs_sb_log_location(device, i, WRITE, &bytenr); 394812659251SNaohiro Aota if (ret == -ENOENT) { 394912659251SNaohiro Aota continue; 395012659251SNaohiro Aota } else if (ret < 0) { 395112659251SNaohiro Aota btrfs_err(device->fs_info, 395212659251SNaohiro Aota "couldn't get super block location for mirror %d", 395312659251SNaohiro Aota i); 395412659251SNaohiro Aota errors++; 395512659251SNaohiro Aota continue; 395612659251SNaohiro Aota } 3957935e5cc9SMiao Xie if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3958935e5cc9SMiao Xie device->commit_total_bytes) 3959a512bbf8SYan Zheng break; 3960a512bbf8SYan Zheng 396112659251SNaohiro Aota btrfs_set_super_bytenr(sb, bytenr_orig); 3962a512bbf8SYan Zheng 3963fd08001fSEric Biggers crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE, 3964fd08001fSEric Biggers BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, 3965fd08001fSEric Biggers sb->csum); 3966a512bbf8SYan Zheng 3967314b6dd0SJohannes Thumshirn page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT, 3968314b6dd0SJohannes Thumshirn GFP_NOFS); 3969314b6dd0SJohannes Thumshirn if (!page) { 3970fb456252SJeff Mahoney btrfs_err(device->fs_info, 3971314b6dd0SJohannes Thumshirn "couldn't get super block page for bytenr %llu", 3972f14d104dSDavid Sterba bytenr); 3973634554dcSJosef Bacik errors++; 3974634554dcSJosef Bacik continue; 3975634554dcSJosef Bacik } 3976634554dcSJosef Bacik 3977314b6dd0SJohannes Thumshirn /* Bump the refcount for wait_dev_supers() */ 3978314b6dd0SJohannes Thumshirn get_page(page); 3979a512bbf8SYan Zheng 3980314b6dd0SJohannes Thumshirn disk_super = page_address(page); 3981314b6dd0SJohannes Thumshirn memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE); 3982a512bbf8SYan Zheng 3983387125fcSChris Mason /* 3984314b6dd0SJohannes Thumshirn * Directly use bios here instead of relying on the page cache 3985314b6dd0SJohannes Thumshirn * to do I/O, so we don't lose the ability to do integrity 3986314b6dd0SJohannes Thumshirn * checking. 3987387125fcSChris Mason */ 398807888c66SChristoph Hellwig bio = bio_alloc(device->bdev, 1, 398907888c66SChristoph Hellwig REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO, 399007888c66SChristoph Hellwig GFP_NOFS); 3991314b6dd0SJohannes Thumshirn bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT; 3992314b6dd0SJohannes Thumshirn bio->bi_private = device; 3993314b6dd0SJohannes Thumshirn bio->bi_end_io = btrfs_end_super_write; 3994314b6dd0SJohannes Thumshirn __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE, 3995314b6dd0SJohannes Thumshirn offset_in_page(bytenr)); 3996314b6dd0SJohannes Thumshirn 3997314b6dd0SJohannes Thumshirn /* 3998314b6dd0SJohannes Thumshirn * We FUA only the first super block. The others we allow to 3999314b6dd0SJohannes Thumshirn * go down lazy and there's a short window where the on-disk 4000314b6dd0SJohannes Thumshirn * copies might still contain the older version. 4001314b6dd0SJohannes Thumshirn */ 40021b9e619cSOmar Sandoval if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) 4003314b6dd0SJohannes Thumshirn bio->bi_opf |= REQ_FUA; 4004314b6dd0SJohannes Thumshirn 400558ff51f1SChristoph Hellwig btrfsic_check_bio(bio); 400658ff51f1SChristoph Hellwig submit_bio(bio); 40078376d9e1SNaohiro Aota 40088376d9e1SNaohiro Aota if (btrfs_advance_sb_log(device, i)) 40098376d9e1SNaohiro Aota errors++; 4010a512bbf8SYan Zheng } 4011a512bbf8SYan Zheng return errors < i ? 0 : -1; 4012a512bbf8SYan Zheng } 4013a512bbf8SYan Zheng 4014387125fcSChris Mason /* 4015abbb3b8eSDavid Sterba * Wait for write completion of superblocks done by write_dev_supers, 4016abbb3b8eSDavid Sterba * @max_mirrors same for write and wait phases. 4017abbb3b8eSDavid Sterba * 4018314b6dd0SJohannes Thumshirn * Return number of errors when page is not found or not marked up to 4019abbb3b8eSDavid Sterba * date. 4020abbb3b8eSDavid Sterba */ 4021abbb3b8eSDavid Sterba static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) 4022abbb3b8eSDavid Sterba { 4023abbb3b8eSDavid Sterba int i; 4024abbb3b8eSDavid Sterba int errors = 0; 4025b6a535faSHoward McLauchlan bool primary_failed = false; 402612659251SNaohiro Aota int ret; 4027abbb3b8eSDavid Sterba u64 bytenr; 4028abbb3b8eSDavid Sterba 4029abbb3b8eSDavid Sterba if (max_mirrors == 0) 4030abbb3b8eSDavid Sterba max_mirrors = BTRFS_SUPER_MIRROR_MAX; 4031abbb3b8eSDavid Sterba 4032abbb3b8eSDavid Sterba for (i = 0; i < max_mirrors; i++) { 4033314b6dd0SJohannes Thumshirn struct page *page; 4034314b6dd0SJohannes Thumshirn 403512659251SNaohiro Aota ret = btrfs_sb_log_location(device, i, READ, &bytenr); 403612659251SNaohiro Aota if (ret == -ENOENT) { 403712659251SNaohiro Aota break; 403812659251SNaohiro Aota } else if (ret < 0) { 403912659251SNaohiro Aota errors++; 404012659251SNaohiro Aota if (i == 0) 404112659251SNaohiro Aota primary_failed = true; 404212659251SNaohiro Aota continue; 404312659251SNaohiro Aota } 4044abbb3b8eSDavid Sterba if (bytenr + BTRFS_SUPER_INFO_SIZE >= 4045abbb3b8eSDavid Sterba device->commit_total_bytes) 4046abbb3b8eSDavid Sterba break; 4047abbb3b8eSDavid Sterba 4048314b6dd0SJohannes Thumshirn page = find_get_page(device->bdev->bd_inode->i_mapping, 4049314b6dd0SJohannes Thumshirn bytenr >> PAGE_SHIFT); 4050314b6dd0SJohannes Thumshirn if (!page) { 4051abbb3b8eSDavid Sterba errors++; 4052b6a535faSHoward McLauchlan if (i == 0) 4053b6a535faSHoward McLauchlan primary_failed = true; 4054abbb3b8eSDavid Sterba continue; 4055abbb3b8eSDavid Sterba } 4056314b6dd0SJohannes Thumshirn /* Page is submitted locked and unlocked once the IO completes */ 4057314b6dd0SJohannes Thumshirn wait_on_page_locked(page); 4058314b6dd0SJohannes Thumshirn if (PageError(page)) { 4059abbb3b8eSDavid Sterba errors++; 4060b6a535faSHoward McLauchlan if (i == 0) 4061b6a535faSHoward McLauchlan primary_failed = true; 4062b6a535faSHoward McLauchlan } 4063abbb3b8eSDavid Sterba 4064314b6dd0SJohannes Thumshirn /* Drop our reference */ 4065314b6dd0SJohannes Thumshirn put_page(page); 4066abbb3b8eSDavid Sterba 4067314b6dd0SJohannes Thumshirn /* Drop the reference from the writing run */ 4068314b6dd0SJohannes Thumshirn put_page(page); 4069abbb3b8eSDavid Sterba } 4070abbb3b8eSDavid Sterba 4071b6a535faSHoward McLauchlan /* log error, force error return */ 4072b6a535faSHoward McLauchlan if (primary_failed) { 4073b6a535faSHoward McLauchlan btrfs_err(device->fs_info, "error writing primary super block to device %llu", 4074b6a535faSHoward McLauchlan device->devid); 4075b6a535faSHoward McLauchlan return -1; 4076b6a535faSHoward McLauchlan } 4077b6a535faSHoward McLauchlan 4078abbb3b8eSDavid Sterba return errors < i ? 0 : -1; 4079abbb3b8eSDavid Sterba } 4080abbb3b8eSDavid Sterba 4081abbb3b8eSDavid Sterba /* 4082387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 4083387125fcSChris Mason * for the barrier when it is done 4084387125fcSChris Mason */ 40854246a0b6SChristoph Hellwig static void btrfs_end_empty_barrier(struct bio *bio) 4086387125fcSChris Mason { 4087f9e69aa9SChristoph Hellwig bio_uninit(bio); 4088387125fcSChris Mason complete(bio->bi_private); 4089387125fcSChris Mason } 4090387125fcSChris Mason 4091387125fcSChris Mason /* 40924fc6441aSAnand Jain * Submit a flush request to the device if it supports it. Error handling is 40934fc6441aSAnand Jain * done in the waiting counterpart. 4094387125fcSChris Mason */ 40954fc6441aSAnand Jain static void write_dev_flush(struct btrfs_device *device) 4096387125fcSChris Mason { 4097f9e69aa9SChristoph Hellwig struct bio *bio = &device->flush_bio; 4098387125fcSChris Mason 4099a91cf0ffSWang Yugui #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY 4100a91cf0ffSWang Yugui /* 4101a91cf0ffSWang Yugui * When a disk has write caching disabled, we skip submission of a bio 4102a91cf0ffSWang Yugui * with flush and sync requests before writing the superblock, since 4103a91cf0ffSWang Yugui * it's not needed. However when the integrity checker is enabled, this 4104a91cf0ffSWang Yugui * results in reports that there are metadata blocks referred by a 4105a91cf0ffSWang Yugui * superblock that were not properly flushed. So don't skip the bio 4106a91cf0ffSWang Yugui * submission only when the integrity checker is enabled for the sake 4107a91cf0ffSWang Yugui * of simplicity, since this is a debug tool and not meant for use in 4108a91cf0ffSWang Yugui * non-debug builds. 4109a91cf0ffSWang Yugui */ 411008e688fdSChristoph Hellwig if (!bdev_write_cache(device->bdev)) 41114fc6441aSAnand Jain return; 4112a91cf0ffSWang Yugui #endif 4113387125fcSChris Mason 4114f9e69aa9SChristoph Hellwig bio_init(bio, device->bdev, NULL, 0, 4115f9e69aa9SChristoph Hellwig REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH); 4116387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 4117387125fcSChris Mason init_completion(&device->flush_wait); 4118387125fcSChris Mason bio->bi_private = &device->flush_wait; 4119387125fcSChris Mason 412058ff51f1SChristoph Hellwig btrfsic_check_bio(bio); 412158ff51f1SChristoph Hellwig submit_bio(bio); 41221c3063b6SAnand Jain set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 41234fc6441aSAnand Jain } 4124387125fcSChris Mason 41254fc6441aSAnand Jain /* 41264fc6441aSAnand Jain * If the flush bio has been submitted by write_dev_flush, wait for it. 41274fc6441aSAnand Jain */ 41288c27cb35SLinus Torvalds static blk_status_t wait_dev_flush(struct btrfs_device *device) 41294fc6441aSAnand Jain { 4130f9e69aa9SChristoph Hellwig struct bio *bio = &device->flush_bio; 41314fc6441aSAnand Jain 41321c3063b6SAnand Jain if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) 413358efbc9fSOmar Sandoval return BLK_STS_OK; 41344fc6441aSAnand Jain 41351c3063b6SAnand Jain clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 41362980d574SDavid Sterba wait_for_completion_io(&device->flush_wait); 41374fc6441aSAnand Jain 41388c27cb35SLinus Torvalds return bio->bi_status; 4139387125fcSChris Mason } 4140387125fcSChris Mason 4141d10b82feSQu Wenruo static int check_barrier_error(struct btrfs_fs_info *fs_info) 4142401b41e5SAnand Jain { 41436528b99dSAnand Jain if (!btrfs_check_rw_degradable(fs_info, NULL)) 4144401b41e5SAnand Jain return -EIO; 4145387125fcSChris Mason return 0; 4146387125fcSChris Mason } 4147387125fcSChris Mason 4148387125fcSChris Mason /* 4149387125fcSChris Mason * send an empty flush down to each device in parallel, 4150387125fcSChris Mason * then wait for them 4151387125fcSChris Mason */ 4152387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 4153387125fcSChris Mason { 4154387125fcSChris Mason struct list_head *head; 4155387125fcSChris Mason struct btrfs_device *dev; 41565af3e8ccSStefan Behrens int errors_wait = 0; 41574e4cbee9SChristoph Hellwig blk_status_t ret; 4158387125fcSChris Mason 41591538e6c5SDavid Sterba lockdep_assert_held(&info->fs_devices->device_list_mutex); 4160387125fcSChris Mason /* send down all the barriers */ 4161387125fcSChris Mason head = &info->fs_devices->devices; 41621538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4163e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 4164f88ba6a2SHidetoshi Seto continue; 4165cea7c8bfSAnand Jain if (!dev->bdev) 4166387125fcSChris Mason continue; 4167e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4168ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4169387125fcSChris Mason continue; 4170387125fcSChris Mason 41714fc6441aSAnand Jain write_dev_flush(dev); 417258efbc9fSOmar Sandoval dev->last_flush_error = BLK_STS_OK; 4173387125fcSChris Mason } 4174387125fcSChris Mason 4175387125fcSChris Mason /* wait for all the barriers */ 41761538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4177e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 4178f88ba6a2SHidetoshi Seto continue; 4179387125fcSChris Mason if (!dev->bdev) { 41805af3e8ccSStefan Behrens errors_wait++; 4181387125fcSChris Mason continue; 4182387125fcSChris Mason } 4183e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4184ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4185387125fcSChris Mason continue; 4186387125fcSChris Mason 41874fc6441aSAnand Jain ret = wait_dev_flush(dev); 4188401b41e5SAnand Jain if (ret) { 4189401b41e5SAnand Jain dev->last_flush_error = ret; 419066b4993eSDavid Sterba btrfs_dev_stat_inc_and_print(dev, 419166b4993eSDavid Sterba BTRFS_DEV_STAT_FLUSH_ERRS); 41925af3e8ccSStefan Behrens errors_wait++; 4193387125fcSChris Mason } 4194401b41e5SAnand Jain } 4195401b41e5SAnand Jain 4196cea7c8bfSAnand Jain if (errors_wait) { 4197401b41e5SAnand Jain /* 4198401b41e5SAnand Jain * At some point we need the status of all disks 4199401b41e5SAnand Jain * to arrive at the volume status. So error checking 4200401b41e5SAnand Jain * is being pushed to a separate loop. 4201401b41e5SAnand Jain */ 4202d10b82feSQu Wenruo return check_barrier_error(info); 4203401b41e5SAnand Jain } 4204387125fcSChris Mason return 0; 4205387125fcSChris Mason } 4206387125fcSChris Mason 4207943c6e99SZhao Lei int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 4208943c6e99SZhao Lei { 42098789f4feSZhao Lei int raid_type; 42108789f4feSZhao Lei int min_tolerated = INT_MAX; 4211943c6e99SZhao Lei 42128789f4feSZhao Lei if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 42138789f4feSZhao Lei (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 42148c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 42158789f4feSZhao Lei btrfs_raid_array[BTRFS_RAID_SINGLE]. 42168789f4feSZhao Lei tolerated_failures); 4217943c6e99SZhao Lei 42188789f4feSZhao Lei for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 42198789f4feSZhao Lei if (raid_type == BTRFS_RAID_SINGLE) 42208789f4feSZhao Lei continue; 422141a6e891SAnand Jain if (!(flags & btrfs_raid_array[raid_type].bg_flag)) 42228789f4feSZhao Lei continue; 42238c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 42248789f4feSZhao Lei btrfs_raid_array[raid_type]. 42258789f4feSZhao Lei tolerated_failures); 42268789f4feSZhao Lei } 4227943c6e99SZhao Lei 42288789f4feSZhao Lei if (min_tolerated == INT_MAX) { 4229ab8d0fc4SJeff Mahoney pr_warn("BTRFS: unknown raid flag: %llu", flags); 42308789f4feSZhao Lei min_tolerated = 0; 42318789f4feSZhao Lei } 42328789f4feSZhao Lei 42338789f4feSZhao Lei return min_tolerated; 4234943c6e99SZhao Lei } 4235943c6e99SZhao Lei 4236eece6a9cSDavid Sterba int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) 4237f2984462SChris Mason { 4238e5e9a520SChris Mason struct list_head *head; 4239f2984462SChris Mason struct btrfs_device *dev; 4240a061fc8dSChris Mason struct btrfs_super_block *sb; 4241f2984462SChris Mason struct btrfs_dev_item *dev_item; 4242f2984462SChris Mason int ret; 4243f2984462SChris Mason int do_barriers; 4244a236aed1SChris Mason int max_errors; 4245a236aed1SChris Mason int total_errors = 0; 4246a061fc8dSChris Mason u64 flags; 4247f2984462SChris Mason 42480b246afaSJeff Mahoney do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); 4249fed3b381SLiu Bo 4250fed3b381SLiu Bo /* 4251fed3b381SLiu Bo * max_mirrors == 0 indicates we're from commit_transaction, 4252fed3b381SLiu Bo * not from fsync where the tree roots in fs_info have not 4253fed3b381SLiu Bo * been consistent on disk. 4254fed3b381SLiu Bo */ 4255fed3b381SLiu Bo if (max_mirrors == 0) 42560b246afaSJeff Mahoney backup_super_roots(fs_info); 4257f2984462SChris Mason 42580b246afaSJeff Mahoney sb = fs_info->super_for_commit; 4259a061fc8dSChris Mason dev_item = &sb->dev_item; 4260e5e9a520SChris Mason 42610b246afaSJeff Mahoney mutex_lock(&fs_info->fs_devices->device_list_mutex); 42620b246afaSJeff Mahoney head = &fs_info->fs_devices->devices; 42630b246afaSJeff Mahoney max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; 4264387125fcSChris Mason 42655af3e8ccSStefan Behrens if (do_barriers) { 42660b246afaSJeff Mahoney ret = barrier_all_devices(fs_info); 42675af3e8ccSStefan Behrens if (ret) { 42685af3e8ccSStefan Behrens mutex_unlock( 42690b246afaSJeff Mahoney &fs_info->fs_devices->device_list_mutex); 42700b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 42715af3e8ccSStefan Behrens "errors while submitting device barriers."); 42725af3e8ccSStefan Behrens return ret; 42735af3e8ccSStefan Behrens } 42745af3e8ccSStefan Behrens } 4275387125fcSChris Mason 42761538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4277dfe25020SChris Mason if (!dev->bdev) { 4278dfe25020SChris Mason total_errors++; 4279dfe25020SChris Mason continue; 4280dfe25020SChris Mason } 4281e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4282ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4283dfe25020SChris Mason continue; 4284dfe25020SChris Mason 42852b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 4286a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 4287a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 42887df69d3eSMiao Xie btrfs_set_stack_device_total_bytes(dev_item, 4289935e5cc9SMiao Xie dev->commit_total_bytes); 4290ce7213c7SMiao Xie btrfs_set_stack_device_bytes_used(dev_item, 4291ce7213c7SMiao Xie dev->commit_bytes_used); 4292a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 4293a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 4294a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 4295a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 42967239ff4bSNikolay Borisov memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, 42977239ff4bSNikolay Borisov BTRFS_FSID_SIZE); 4298a512bbf8SYan Zheng 4299a061fc8dSChris Mason flags = btrfs_super_flags(sb); 4300a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 4301f2984462SChris Mason 430275cb857dSQu Wenruo ret = btrfs_validate_write_super(fs_info, sb); 430375cb857dSQu Wenruo if (ret < 0) { 430475cb857dSQu Wenruo mutex_unlock(&fs_info->fs_devices->device_list_mutex); 430575cb857dSQu Wenruo btrfs_handle_fs_error(fs_info, -EUCLEAN, 430675cb857dSQu Wenruo "unexpected superblock corruption detected"); 430775cb857dSQu Wenruo return -EUCLEAN; 430875cb857dSQu Wenruo } 430975cb857dSQu Wenruo 4310abbb3b8eSDavid Sterba ret = write_dev_supers(dev, sb, max_mirrors); 4311a236aed1SChris Mason if (ret) 4312a236aed1SChris Mason total_errors++; 4313f2984462SChris Mason } 4314a236aed1SChris Mason if (total_errors > max_errors) { 43150b246afaSJeff Mahoney btrfs_err(fs_info, "%d errors while writing supers", 4316d397712bSChris Mason total_errors); 43170b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 431879787eaaSJeff Mahoney 43199d565ba4SStefan Behrens /* FUA is masked off if unsupported and can't be the reason */ 43200b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 43210b246afaSJeff Mahoney "%d errors while writing supers", 43220b246afaSJeff Mahoney total_errors); 43239d565ba4SStefan Behrens return -EIO; 4324a236aed1SChris Mason } 4325f2984462SChris Mason 4326a512bbf8SYan Zheng total_errors = 0; 43271538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4328dfe25020SChris Mason if (!dev->bdev) 4329dfe25020SChris Mason continue; 4330e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4331ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4332dfe25020SChris Mason continue; 4333dfe25020SChris Mason 4334abbb3b8eSDavid Sterba ret = wait_dev_supers(dev, max_mirrors); 4335a512bbf8SYan Zheng if (ret) 43361259ab75SChris Mason total_errors++; 4337f2984462SChris Mason } 43380b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 4339a236aed1SChris Mason if (total_errors > max_errors) { 43400b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 43410b246afaSJeff Mahoney "%d errors while writing supers", 43420b246afaSJeff Mahoney total_errors); 434379787eaaSJeff Mahoney return -EIO; 4344a236aed1SChris Mason } 4345f2984462SChris Mason return 0; 4346f2984462SChris Mason } 4347f2984462SChris Mason 4348cb517eabSMiao Xie /* Drop a fs root from the radix tree and free it. */ 4349cb517eabSMiao Xie void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 4350cb517eabSMiao Xie struct btrfs_root *root) 43512619ba1fSChris Mason { 43524785e24fSJosef Bacik bool drop_ref = false; 43534785e24fSJosef Bacik 4354fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4355fc7cbcd4SDavid Sterba radix_tree_delete(&fs_info->fs_roots_radix, 4356fc7cbcd4SDavid Sterba (unsigned long)root->root_key.objectid); 4357fc7cbcd4SDavid Sterba if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state)) 43584785e24fSJosef Bacik drop_ref = true; 4359fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 436076dda93cSYan, Zheng 436184961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 4362ef67963dSJosef Bacik ASSERT(root->log_root == NULL); 43631c1ea4f7SLiu Bo if (root->reloc_root) { 436400246528SJosef Bacik btrfs_put_root(root->reloc_root); 43651c1ea4f7SLiu Bo root->reloc_root = NULL; 43661c1ea4f7SLiu Bo } 43671c1ea4f7SLiu Bo } 43683321719eSLiu Bo 43694785e24fSJosef Bacik if (drop_ref) 437000246528SJosef Bacik btrfs_put_root(root); 43712619ba1fSChris Mason } 43722619ba1fSChris Mason 4373c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 4374c146afadSYan Zheng { 4375fc7cbcd4SDavid Sterba u64 root_objectid = 0; 4376fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 4377fc7cbcd4SDavid Sterba int i = 0; 437865d33fd7SQu Wenruo int err = 0; 4379fc7cbcd4SDavid Sterba unsigned int ret = 0; 4380c146afadSYan Zheng 4381c146afadSYan Zheng while (1) { 4382fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4383fc7cbcd4SDavid Sterba ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4384fc7cbcd4SDavid Sterba (void **)gang, root_objectid, 4385fc7cbcd4SDavid Sterba ARRAY_SIZE(gang)); 4386fc7cbcd4SDavid Sterba if (!ret) { 4387fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4388c146afadSYan Zheng break; 438965d33fd7SQu Wenruo } 4390fc7cbcd4SDavid Sterba root_objectid = gang[ret - 1]->root_key.objectid + 1; 439166b4ffd1SJosef Bacik 4392fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 4393fc7cbcd4SDavid Sterba /* Avoid to grab roots in dead_roots */ 4394fc7cbcd4SDavid Sterba if (btrfs_root_refs(&gang[i]->root_item) == 0) { 4395fc7cbcd4SDavid Sterba gang[i] = NULL; 439665d33fd7SQu Wenruo continue; 4397c146afadSYan Zheng } 4398fc7cbcd4SDavid Sterba /* grab all the search result for later use */ 4399fc7cbcd4SDavid Sterba gang[i] = btrfs_grab_root(gang[i]); 4400fc7cbcd4SDavid Sterba } 4401fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4402fc7cbcd4SDavid Sterba 4403fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 4404fc7cbcd4SDavid Sterba if (!gang[i]) 4405fc7cbcd4SDavid Sterba continue; 4406fc7cbcd4SDavid Sterba root_objectid = gang[i]->root_key.objectid; 4407fc7cbcd4SDavid Sterba err = btrfs_orphan_cleanup(gang[i]); 4408fc7cbcd4SDavid Sterba if (err) 4409fc7cbcd4SDavid Sterba break; 4410fc7cbcd4SDavid Sterba btrfs_put_root(gang[i]); 4411fc7cbcd4SDavid Sterba } 4412fc7cbcd4SDavid Sterba root_objectid++; 4413c146afadSYan Zheng } 441465d33fd7SQu Wenruo 4415fc7cbcd4SDavid Sterba /* release the uncleaned roots due to error */ 4416fc7cbcd4SDavid Sterba for (; i < ret; i++) { 4417fc7cbcd4SDavid Sterba if (gang[i]) 4418fc7cbcd4SDavid Sterba btrfs_put_root(gang[i]); 441965d33fd7SQu Wenruo } 442065d33fd7SQu Wenruo return err; 4421c146afadSYan Zheng } 4422c146afadSYan Zheng 44236bccf3abSJeff Mahoney int btrfs_commit_super(struct btrfs_fs_info *fs_info) 4424c146afadSYan Zheng { 44256bccf3abSJeff Mahoney struct btrfs_root *root = fs_info->tree_root; 4426c146afadSYan Zheng struct btrfs_trans_handle *trans; 4427c146afadSYan Zheng 44280b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 44292ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 44300b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 44310b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 4432c71bf099SYan, Zheng 4433c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 44340b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 44350b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4436c71bf099SYan, Zheng 44377a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 44383612b495STsutomu Itoh if (IS_ERR(trans)) 44393612b495STsutomu Itoh return PTR_ERR(trans); 44403a45bb20SJeff Mahoney return btrfs_commit_transaction(trans); 4441c146afadSYan Zheng } 4442c146afadSYan Zheng 444336c86a9eSQu Wenruo static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info) 444436c86a9eSQu Wenruo { 444536c86a9eSQu Wenruo struct btrfs_transaction *trans; 444636c86a9eSQu Wenruo struct btrfs_transaction *tmp; 444736c86a9eSQu Wenruo bool found = false; 444836c86a9eSQu Wenruo 444936c86a9eSQu Wenruo if (list_empty(&fs_info->trans_list)) 445036c86a9eSQu Wenruo return; 445136c86a9eSQu Wenruo 445236c86a9eSQu Wenruo /* 445336c86a9eSQu Wenruo * This function is only called at the very end of close_ctree(), 445436c86a9eSQu Wenruo * thus no other running transaction, no need to take trans_lock. 445536c86a9eSQu Wenruo */ 445636c86a9eSQu Wenruo ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags)); 445736c86a9eSQu Wenruo list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) { 445836c86a9eSQu Wenruo struct extent_state *cached = NULL; 445936c86a9eSQu Wenruo u64 dirty_bytes = 0; 446036c86a9eSQu Wenruo u64 cur = 0; 446136c86a9eSQu Wenruo u64 found_start; 446236c86a9eSQu Wenruo u64 found_end; 446336c86a9eSQu Wenruo 446436c86a9eSQu Wenruo found = true; 446536c86a9eSQu Wenruo while (!find_first_extent_bit(&trans->dirty_pages, cur, 446636c86a9eSQu Wenruo &found_start, &found_end, EXTENT_DIRTY, &cached)) { 446736c86a9eSQu Wenruo dirty_bytes += found_end + 1 - found_start; 446836c86a9eSQu Wenruo cur = found_end + 1; 446936c86a9eSQu Wenruo } 447036c86a9eSQu Wenruo btrfs_warn(fs_info, 447136c86a9eSQu Wenruo "transaction %llu (with %llu dirty metadata bytes) is not committed", 447236c86a9eSQu Wenruo trans->transid, dirty_bytes); 447336c86a9eSQu Wenruo btrfs_cleanup_one_transaction(trans, fs_info); 447436c86a9eSQu Wenruo 447536c86a9eSQu Wenruo if (trans == fs_info->running_transaction) 447636c86a9eSQu Wenruo fs_info->running_transaction = NULL; 447736c86a9eSQu Wenruo list_del_init(&trans->list); 447836c86a9eSQu Wenruo 447936c86a9eSQu Wenruo btrfs_put_transaction(trans); 448036c86a9eSQu Wenruo trace_btrfs_transaction_commit(fs_info); 448136c86a9eSQu Wenruo } 448236c86a9eSQu Wenruo ASSERT(!found); 448336c86a9eSQu Wenruo } 448436c86a9eSQu Wenruo 4485b105e927SDavid Sterba void __cold close_ctree(struct btrfs_fs_info *fs_info) 4486eb60ceacSChris Mason { 4487c146afadSYan Zheng int ret; 4488e089f05cSChris Mason 4489afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); 449031e70e52SFilipe Manana 449131e70e52SFilipe Manana /* 44928a1f1e3dSFilipe Manana * If we had UNFINISHED_DROPS we could still be processing them, so 44938a1f1e3dSFilipe Manana * clear that bit and wake up relocation so it can stop. 44948a1f1e3dSFilipe Manana * We must do this before stopping the block group reclaim task, because 44958a1f1e3dSFilipe Manana * at btrfs_relocate_block_group() we wait for this bit, and after the 44968a1f1e3dSFilipe Manana * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we 44978a1f1e3dSFilipe Manana * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will 44988a1f1e3dSFilipe Manana * return 1. 44998a1f1e3dSFilipe Manana */ 45008a1f1e3dSFilipe Manana btrfs_wake_unfinished_drop(fs_info); 45018a1f1e3dSFilipe Manana 45028a1f1e3dSFilipe Manana /* 450331e70e52SFilipe Manana * We may have the reclaim task running and relocating a data block group, 450431e70e52SFilipe Manana * in which case it may create delayed iputs. So stop it before we park 450531e70e52SFilipe Manana * the cleaner kthread otherwise we can get new delayed iputs after 450631e70e52SFilipe Manana * parking the cleaner, and that can make the async reclaim task to hang 450731e70e52SFilipe Manana * if it's waiting for delayed iputs to complete, since the cleaner is 450831e70e52SFilipe Manana * parked and can not run delayed iputs - this will make us hang when 450931e70e52SFilipe Manana * trying to stop the async reclaim task. 451031e70e52SFilipe Manana */ 451131e70e52SFilipe Manana cancel_work_sync(&fs_info->reclaim_bgs_work); 4512d6fd0ae2SOmar Sandoval /* 4513d6fd0ae2SOmar Sandoval * We don't want the cleaner to start new transactions, add more delayed 4514d6fd0ae2SOmar Sandoval * iputs, etc. while we're closing. We can't use kthread_stop() yet 4515d6fd0ae2SOmar Sandoval * because that frees the task_struct, and the transaction kthread might 4516d6fd0ae2SOmar Sandoval * still try to wake up the cleaner. 4517d6fd0ae2SOmar Sandoval */ 4518d6fd0ae2SOmar Sandoval kthread_park(fs_info->cleaner_kthread); 4519a2135011SChris Mason 45207343dd61SJustin Maggard /* wait for the qgroup rescan worker to stop */ 4521d06f23d6SJeff Mahoney btrfs_qgroup_wait_for_completion(fs_info, false); 45227343dd61SJustin Maggard 4523803b2f54SStefan Behrens /* wait for the uuid_scan task to finish */ 4524803b2f54SStefan Behrens down(&fs_info->uuid_tree_rescan_sem); 4525803b2f54SStefan Behrens /* avoid complains from lockdep et al., set sem back to initial state */ 4526803b2f54SStefan Behrens up(&fs_info->uuid_tree_rescan_sem); 4527803b2f54SStefan Behrens 4528837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 4529aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 4530837d5b6eSIlya Dryomov 45318dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 45328dabb742SStefan Behrens 4533aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 45344cb5300bSChris Mason 45354cb5300bSChris Mason /* wait for any defraggers to finish */ 45364cb5300bSChris Mason wait_event(fs_info->transaction_wait, 45374cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 45384cb5300bSChris Mason 45394cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 454026176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 45414cb5300bSChris Mason 4542a362bb86SFilipe Manana /* 4543a362bb86SFilipe Manana * After we parked the cleaner kthread, ordered extents may have 4544a362bb86SFilipe Manana * completed and created new delayed iputs. If one of the async reclaim 4545a362bb86SFilipe Manana * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we 4546a362bb86SFilipe Manana * can hang forever trying to stop it, because if a delayed iput is 4547a362bb86SFilipe Manana * added after it ran btrfs_run_delayed_iputs() and before it called 4548a362bb86SFilipe Manana * btrfs_wait_on_delayed_iputs(), it will hang forever since there is 4549a362bb86SFilipe Manana * no one else to run iputs. 4550a362bb86SFilipe Manana * 4551a362bb86SFilipe Manana * So wait for all ongoing ordered extents to complete and then run 4552a362bb86SFilipe Manana * delayed iputs. This works because once we reach this point no one 4553a362bb86SFilipe Manana * can either create new ordered extents nor create delayed iputs 4554a362bb86SFilipe Manana * through some other means. 4555a362bb86SFilipe Manana * 4556a362bb86SFilipe Manana * Also note that btrfs_wait_ordered_roots() is not safe here, because 4557a362bb86SFilipe Manana * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent, 4558a362bb86SFilipe Manana * but the delayed iput for the respective inode is made only when doing 4559a362bb86SFilipe Manana * the final btrfs_put_ordered_extent() (which must happen at 4560a362bb86SFilipe Manana * btrfs_finish_ordered_io() when we are unmounting). 4561a362bb86SFilipe Manana */ 4562a362bb86SFilipe Manana btrfs_flush_workqueue(fs_info->endio_write_workers); 4563a362bb86SFilipe Manana /* Ordered extents for free space inodes. */ 4564a362bb86SFilipe Manana btrfs_flush_workqueue(fs_info->endio_freespace_worker); 4565a362bb86SFilipe Manana btrfs_run_delayed_iputs(fs_info); 4566a362bb86SFilipe Manana 456721c7e756SMiao Xie cancel_work_sync(&fs_info->async_reclaim_work); 456857056740SJosef Bacik cancel_work_sync(&fs_info->async_data_reclaim_work); 4569576fa348SJosef Bacik cancel_work_sync(&fs_info->preempt_reclaim_work); 457021c7e756SMiao Xie 4571b0643e59SDennis Zhou /* Cancel or finish ongoing discard work */ 4572b0643e59SDennis Zhou btrfs_discard_cleanup(fs_info); 4573b0643e59SDennis Zhou 4574bc98a42cSDavid Howells if (!sb_rdonly(fs_info->sb)) { 4575e44163e1SJeff Mahoney /* 4576d6fd0ae2SOmar Sandoval * The cleaner kthread is stopped, so do one final pass over 4577d6fd0ae2SOmar Sandoval * unused block groups. 4578e44163e1SJeff Mahoney */ 45790b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 4580e44163e1SJeff Mahoney 4581f0cc2cd7SFilipe Manana /* 4582f0cc2cd7SFilipe Manana * There might be existing delayed inode workers still running 4583f0cc2cd7SFilipe Manana * and holding an empty delayed inode item. We must wait for 4584f0cc2cd7SFilipe Manana * them to complete first because they can create a transaction. 4585f0cc2cd7SFilipe Manana * This happens when someone calls btrfs_balance_delayed_items() 4586f0cc2cd7SFilipe Manana * and then a transaction commit runs the same delayed nodes 4587f0cc2cd7SFilipe Manana * before any delayed worker has done something with the nodes. 4588f0cc2cd7SFilipe Manana * We must wait for any worker here and not at transaction 4589f0cc2cd7SFilipe Manana * commit time since that could cause a deadlock. 4590f0cc2cd7SFilipe Manana * This is a very rare case. 4591f0cc2cd7SFilipe Manana */ 4592f0cc2cd7SFilipe Manana btrfs_flush_workqueue(fs_info->delayed_workers); 4593f0cc2cd7SFilipe Manana 45946bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 4595d397712bSChris Mason if (ret) 459604892340SEric Sandeen btrfs_err(fs_info, "commit super ret %d", ret); 4597c146afadSYan Zheng } 4598ed2ff2cbSChris Mason 459984961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) 46002ff7e61eSJeff Mahoney btrfs_error_commit_super(fs_info); 4601acce952bSliubo 4602e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 4603e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 46048929ecfaSYan, Zheng 4605e187831eSJosef Bacik ASSERT(list_empty(&fs_info->delayed_iputs)); 4606afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); 4607f25784b3SYan Zheng 46085958253cSQu Wenruo if (btrfs_check_quota_leak(fs_info)) { 46095958253cSQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 46105958253cSQu Wenruo btrfs_err(fs_info, "qgroup reserved space leaked"); 46115958253cSQu Wenruo } 46125958253cSQu Wenruo 461304892340SEric Sandeen btrfs_free_qgroup_config(fs_info); 4614fe816d0fSNikolay Borisov ASSERT(list_empty(&fs_info->delalloc_roots)); 4615bcef60f2SArne Jansen 4616963d678bSMiao Xie if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 461704892340SEric Sandeen btrfs_info(fs_info, "at unmount delalloc count %lld", 4618963d678bSMiao Xie percpu_counter_sum(&fs_info->delalloc_bytes)); 4619b0c68f8bSChris Mason } 462031153d81SYan Zheng 46215deb17e1SJosef Bacik if (percpu_counter_sum(&fs_info->ordered_bytes)) 46224297ff84SJosef Bacik btrfs_info(fs_info, "at unmount dio bytes count %lld", 46235deb17e1SJosef Bacik percpu_counter_sum(&fs_info->ordered_bytes)); 46244297ff84SJosef Bacik 46256618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 4626b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 46275ac1d209SJeff Mahoney 46281a4319ccSLiu Bo btrfs_put_block_group_cache(fs_info); 46291a4319ccSLiu Bo 4630de348ee0SWang Shilong /* 4631de348ee0SWang Shilong * we must make sure there is not any read request to 4632de348ee0SWang Shilong * submit after we stopping all workers. 4633de348ee0SWang Shilong */ 4634de348ee0SWang Shilong invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 463596192499SJosef Bacik btrfs_stop_all_workers(fs_info); 463696192499SJosef Bacik 46370a31daa4SFilipe Manana /* We shouldn't have any transaction open at this point */ 463836c86a9eSQu Wenruo warn_about_uncommitted_trans(fs_info); 46390a31daa4SFilipe Manana 4640afcdd129SJosef Bacik clear_bit(BTRFS_FS_OPEN, &fs_info->flags); 46414273eaffSAnand Jain free_root_pointers(fs_info, true); 46428c38938cSJosef Bacik btrfs_free_fs_roots(fs_info); 46439ad6b7bcSChris Mason 46444e19443dSJosef Bacik /* 46454e19443dSJosef Bacik * We must free the block groups after dropping the fs_roots as we could 46464e19443dSJosef Bacik * have had an IO error and have left over tree log blocks that aren't 46474e19443dSJosef Bacik * cleaned up until the fs roots are freed. This makes the block group 46484e19443dSJosef Bacik * accounting appear to be wrong because there's pending reserved bytes, 46494e19443dSJosef Bacik * so make sure we do the block group cleanup afterwards. 46504e19443dSJosef Bacik */ 46514e19443dSJosef Bacik btrfs_free_block_groups(fs_info); 46524e19443dSJosef Bacik 465313e6c37bSJosef Bacik iput(fs_info->btree_inode); 4654d6bfde87SChris Mason 465521adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 46560b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) 46572ff7e61eSJeff Mahoney btrfsic_unmount(fs_info->fs_devices); 465821adbd5cSStefan Behrens #endif 465921adbd5cSStefan Behrens 46600b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 466168c94e55SNikolay Borisov btrfs_close_devices(fs_info->fs_devices); 4662eb60ceacSChris Mason } 4663eb60ceacSChris Mason 4664b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 4665b9fab919SChris Mason int atomic) 4666ccd467d6SChris Mason { 46671259ab75SChris Mason int ret; 4668727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 46691259ab75SChris Mason 46700b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 46711259ab75SChris Mason if (!ret) 46721259ab75SChris Mason return ret; 46731259ab75SChris Mason 46741259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 4675b9fab919SChris Mason parent_transid, atomic); 4676b9fab919SChris Mason if (ret == -EAGAIN) 4677b9fab919SChris Mason return ret; 46781259ab75SChris Mason return !ret; 46795f39d397SChris Mason } 46806702ed49SChris Mason 46815f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 46825f39d397SChris Mason { 46832f4d60dfSQu Wenruo struct btrfs_fs_info *fs_info = buf->fs_info; 46845f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 4685b9473439SChris Mason int was_dirty; 4686b4ce94deSChris Mason 468706ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 468806ea65a3SJosef Bacik /* 468906ea65a3SJosef Bacik * This is a fast path so only do this check if we have sanity tests 469052042d8eSAndrea Gelmini * enabled. Normal people shouldn't be using unmapped buffers as dirty 469106ea65a3SJosef Bacik * outside of the sanity tests. 469206ea65a3SJosef Bacik */ 4693b0132a3bSNikolay Borisov if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) 469406ea65a3SJosef Bacik return; 469506ea65a3SJosef Bacik #endif 469649d0c642SFilipe Manana btrfs_assert_tree_write_locked(buf); 46970b246afaSJeff Mahoney if (transid != fs_info->generation) 46985d163e0eSJeff Mahoney WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", 46990b246afaSJeff Mahoney buf->start, transid, fs_info->generation); 47000b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 4701e2d84521SMiao Xie if (!was_dirty) 4702104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 4703e2d84521SMiao Xie buf->len, 47040b246afaSJeff Mahoney fs_info->dirty_metadata_batch); 47051f21ef0aSFilipe Manana #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 470669fc6cbbSQu Wenruo /* 470769fc6cbbSQu Wenruo * Since btrfs_mark_buffer_dirty() can be called with item pointer set 470869fc6cbbSQu Wenruo * but item data not updated. 470969fc6cbbSQu Wenruo * So here we should only check item pointers, not item data. 471069fc6cbbSQu Wenruo */ 471169fc6cbbSQu Wenruo if (btrfs_header_level(buf) == 0 && 4712cfdaad5eSDavid Sterba btrfs_check_leaf_relaxed(buf)) { 4713a4f78750SDavid Sterba btrfs_print_leaf(buf); 47141f21ef0aSFilipe Manana ASSERT(0); 47151f21ef0aSFilipe Manana } 47161f21ef0aSFilipe Manana #endif 4717eb60ceacSChris Mason } 4718eb60ceacSChris Mason 47192ff7e61eSJeff Mahoney static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, 4720b53d3f5dSLiu Bo int flush_delayed) 472135b7e476SChris Mason { 4722188de649SChris Mason /* 4723188de649SChris Mason * looks as though older kernels can get into trouble with 4724188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 4725188de649SChris Mason */ 4726e2d84521SMiao Xie int ret; 4727d6bfde87SChris Mason 47286933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 4729d6bfde87SChris Mason return; 4730d6bfde87SChris Mason 4731b53d3f5dSLiu Bo if (flush_delayed) 47322ff7e61eSJeff Mahoney btrfs_balance_delayed_items(fs_info); 473316cdcec7SMiao Xie 4734d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 4735d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 4736d814a491SEthan Lien fs_info->dirty_metadata_batch); 4737e2d84521SMiao Xie if (ret > 0) { 47380b246afaSJeff Mahoney balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); 473916cdcec7SMiao Xie } 474016cdcec7SMiao Xie } 474116cdcec7SMiao Xie 47422ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) 474316cdcec7SMiao Xie { 47442ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 1); 474535b7e476SChris Mason } 4746b53d3f5dSLiu Bo 47472ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) 4748b53d3f5dSLiu Bo { 47492ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 0); 4750d6bfde87SChris Mason } 47516b80053dSChris Mason 47522ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) 4753acce952bSliubo { 4754fe816d0fSNikolay Borisov /* cleanup FS via transaction */ 4755fe816d0fSNikolay Borisov btrfs_cleanup_transaction(fs_info); 4756fe816d0fSNikolay Borisov 47570b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 47582ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 47590b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 4760acce952bSliubo 47610b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 47620b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4763acce952bSliubo } 4764acce952bSliubo 4765ef67963dSJosef Bacik static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info) 4766ef67963dSJosef Bacik { 4767fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 4768fc7cbcd4SDavid Sterba u64 root_objectid = 0; 4769fc7cbcd4SDavid Sterba int ret; 4770ef67963dSJosef Bacik 4771fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4772fc7cbcd4SDavid Sterba while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4773fc7cbcd4SDavid Sterba (void **)gang, root_objectid, 4774fc7cbcd4SDavid Sterba ARRAY_SIZE(gang))) != 0) { 4775fc7cbcd4SDavid Sterba int i; 4776ef67963dSJosef Bacik 4777fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) 4778fc7cbcd4SDavid Sterba gang[i] = btrfs_grab_root(gang[i]); 4779fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4780fc7cbcd4SDavid Sterba 4781fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 4782fc7cbcd4SDavid Sterba if (!gang[i]) 4783ef67963dSJosef Bacik continue; 4784fc7cbcd4SDavid Sterba root_objectid = gang[i]->root_key.objectid; 4785fc7cbcd4SDavid Sterba btrfs_free_log(NULL, gang[i]); 4786fc7cbcd4SDavid Sterba btrfs_put_root(gang[i]); 4787ef67963dSJosef Bacik } 4788fc7cbcd4SDavid Sterba root_objectid++; 4789fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4790ef67963dSJosef Bacik } 4791fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4792ef67963dSJosef Bacik btrfs_free_log_root_tree(NULL, fs_info); 4793ef67963dSJosef Bacik } 4794ef67963dSJosef Bacik 4795143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4796acce952bSliubo { 4797acce952bSliubo struct btrfs_ordered_extent *ordered; 4798acce952bSliubo 4799199c2a9cSMiao Xie spin_lock(&root->ordered_extent_lock); 4800779880efSJosef Bacik /* 4801779880efSJosef Bacik * This will just short circuit the ordered completion stuff which will 4802779880efSJosef Bacik * make sure the ordered extent gets properly cleaned up. 4803779880efSJosef Bacik */ 4804199c2a9cSMiao Xie list_for_each_entry(ordered, &root->ordered_extents, 4805779880efSJosef Bacik root_extent_list) 4806779880efSJosef Bacik set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4807199c2a9cSMiao Xie spin_unlock(&root->ordered_extent_lock); 4808199c2a9cSMiao Xie } 4809199c2a9cSMiao Xie 4810199c2a9cSMiao Xie static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4811199c2a9cSMiao Xie { 4812199c2a9cSMiao Xie struct btrfs_root *root; 4813199c2a9cSMiao Xie struct list_head splice; 4814199c2a9cSMiao Xie 4815199c2a9cSMiao Xie INIT_LIST_HEAD(&splice); 4816199c2a9cSMiao Xie 4817199c2a9cSMiao Xie spin_lock(&fs_info->ordered_root_lock); 4818199c2a9cSMiao Xie list_splice_init(&fs_info->ordered_roots, &splice); 4819199c2a9cSMiao Xie while (!list_empty(&splice)) { 4820199c2a9cSMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4821199c2a9cSMiao Xie ordered_root); 48221de2cfdeSJosef Bacik list_move_tail(&root->ordered_root, 48231de2cfdeSJosef Bacik &fs_info->ordered_roots); 4824199c2a9cSMiao Xie 48252a85d9caSLiu Bo spin_unlock(&fs_info->ordered_root_lock); 4826199c2a9cSMiao Xie btrfs_destroy_ordered_extents(root); 4827199c2a9cSMiao Xie 48282a85d9caSLiu Bo cond_resched(); 48292a85d9caSLiu Bo spin_lock(&fs_info->ordered_root_lock); 4830199c2a9cSMiao Xie } 4831199c2a9cSMiao Xie spin_unlock(&fs_info->ordered_root_lock); 483274d5d229SJosef Bacik 483374d5d229SJosef Bacik /* 483474d5d229SJosef Bacik * We need this here because if we've been flipped read-only we won't 483574d5d229SJosef Bacik * get sync() from the umount, so we need to make sure any ordered 483674d5d229SJosef Bacik * extents that haven't had their dirty pages IO start writeout yet 483774d5d229SJosef Bacik * actually get run and error out properly. 483874d5d229SJosef Bacik */ 483974d5d229SJosef Bacik btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 4840acce952bSliubo } 4841acce952bSliubo 484235a3621bSStefan Behrens static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 48432ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4844acce952bSliubo { 4845acce952bSliubo struct rb_node *node; 4846acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 4847acce952bSliubo struct btrfs_delayed_ref_node *ref; 4848acce952bSliubo int ret = 0; 4849acce952bSliubo 4850acce952bSliubo delayed_refs = &trans->delayed_refs; 4851acce952bSliubo 4852acce952bSliubo spin_lock(&delayed_refs->lock); 4853d7df2c79SJosef Bacik if (atomic_read(&delayed_refs->num_entries) == 0) { 4854cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 4855b79ce3ddSDavid Sterba btrfs_debug(fs_info, "delayed_refs has NO entry"); 4856acce952bSliubo return ret; 4857acce952bSliubo } 4858acce952bSliubo 48595c9d028bSLiu Bo while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { 4860d7df2c79SJosef Bacik struct btrfs_delayed_ref_head *head; 48610e0adbcfSJosef Bacik struct rb_node *n; 4862e78417d1SJosef Bacik bool pin_bytes = false; 4863acce952bSliubo 4864d7df2c79SJosef Bacik head = rb_entry(node, struct btrfs_delayed_ref_head, 4865d7df2c79SJosef Bacik href_node); 48663069bd26SJosef Bacik if (btrfs_delayed_ref_lock(delayed_refs, head)) 4867b939d1abSJosef Bacik continue; 48683069bd26SJosef Bacik 4869d7df2c79SJosef Bacik spin_lock(&head->lock); 4870e3d03965SLiu Bo while ((n = rb_first_cached(&head->ref_tree)) != NULL) { 48710e0adbcfSJosef Bacik ref = rb_entry(n, struct btrfs_delayed_ref_node, 48720e0adbcfSJosef Bacik ref_node); 4873d7df2c79SJosef Bacik ref->in_tree = 0; 4874e3d03965SLiu Bo rb_erase_cached(&ref->ref_node, &head->ref_tree); 48750e0adbcfSJosef Bacik RB_CLEAR_NODE(&ref->ref_node); 48761d57ee94SWang Xiaoguang if (!list_empty(&ref->add_list)) 48771d57ee94SWang Xiaoguang list_del(&ref->add_list); 4878d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4879d7df2c79SJosef Bacik btrfs_put_delayed_ref(ref); 4880d7df2c79SJosef Bacik } 488154067ae9SJosef Bacik if (head->must_insert_reserved) 4882e78417d1SJosef Bacik pin_bytes = true; 488378a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 4884fa781ceaSJosef Bacik btrfs_delete_ref_head(delayed_refs, head); 4885d7df2c79SJosef Bacik spin_unlock(&head->lock); 4886acce952bSliubo spin_unlock(&delayed_refs->lock); 4887e78417d1SJosef Bacik mutex_unlock(&head->mutex); 4888acce952bSliubo 4889f603bb94SNikolay Borisov if (pin_bytes) { 4890f603bb94SNikolay Borisov struct btrfs_block_group *cache; 4891f603bb94SNikolay Borisov 4892f603bb94SNikolay Borisov cache = btrfs_lookup_block_group(fs_info, head->bytenr); 4893f603bb94SNikolay Borisov BUG_ON(!cache); 4894f603bb94SNikolay Borisov 4895f603bb94SNikolay Borisov spin_lock(&cache->space_info->lock); 4896f603bb94SNikolay Borisov spin_lock(&cache->lock); 4897f603bb94SNikolay Borisov cache->pinned += head->num_bytes; 4898f603bb94SNikolay Borisov btrfs_space_info_update_bytes_pinned(fs_info, 4899f603bb94SNikolay Borisov cache->space_info, head->num_bytes); 4900f603bb94SNikolay Borisov cache->reserved -= head->num_bytes; 4901f603bb94SNikolay Borisov cache->space_info->bytes_reserved -= head->num_bytes; 4902f603bb94SNikolay Borisov spin_unlock(&cache->lock); 4903f603bb94SNikolay Borisov spin_unlock(&cache->space_info->lock); 4904f603bb94SNikolay Borisov 4905f603bb94SNikolay Borisov btrfs_put_block_group(cache); 4906f603bb94SNikolay Borisov 4907f603bb94SNikolay Borisov btrfs_error_unpin_extent_range(fs_info, head->bytenr, 4908f603bb94SNikolay Borisov head->bytenr + head->num_bytes - 1); 4909f603bb94SNikolay Borisov } 491031890da0SJosef Bacik btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); 4911d278850eSJosef Bacik btrfs_put_delayed_ref_head(head); 4912acce952bSliubo cond_resched(); 4913acce952bSliubo spin_lock(&delayed_refs->lock); 4914acce952bSliubo } 491581f7eb00SJeff Mahoney btrfs_qgroup_destroy_extent_records(trans); 4916acce952bSliubo 4917acce952bSliubo spin_unlock(&delayed_refs->lock); 4918acce952bSliubo 4919acce952bSliubo return ret; 4920acce952bSliubo } 4921acce952bSliubo 4922143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 4923acce952bSliubo { 4924acce952bSliubo struct btrfs_inode *btrfs_inode; 4925acce952bSliubo struct list_head splice; 4926acce952bSliubo 4927acce952bSliubo INIT_LIST_HEAD(&splice); 4928acce952bSliubo 4929eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4930eb73c1b7SMiao Xie list_splice_init(&root->delalloc_inodes, &splice); 4931acce952bSliubo 4932acce952bSliubo while (!list_empty(&splice)) { 4933fe816d0fSNikolay Borisov struct inode *inode = NULL; 4934eb73c1b7SMiao Xie btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 4935acce952bSliubo delalloc_inodes); 4936fe816d0fSNikolay Borisov __btrfs_del_delalloc_inode(root, btrfs_inode); 4937eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4938acce952bSliubo 4939fe816d0fSNikolay Borisov /* 4940fe816d0fSNikolay Borisov * Make sure we get a live inode and that it'll not disappear 4941fe816d0fSNikolay Borisov * meanwhile. 4942fe816d0fSNikolay Borisov */ 4943fe816d0fSNikolay Borisov inode = igrab(&btrfs_inode->vfs_inode); 4944fe816d0fSNikolay Borisov if (inode) { 4945fe816d0fSNikolay Borisov invalidate_inode_pages2(inode->i_mapping); 4946fe816d0fSNikolay Borisov iput(inode); 4947fe816d0fSNikolay Borisov } 4948eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4949acce952bSliubo } 4950eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4951eb73c1b7SMiao Xie } 4952eb73c1b7SMiao Xie 4953eb73c1b7SMiao Xie static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 4954eb73c1b7SMiao Xie { 4955eb73c1b7SMiao Xie struct btrfs_root *root; 4956eb73c1b7SMiao Xie struct list_head splice; 4957eb73c1b7SMiao Xie 4958eb73c1b7SMiao Xie INIT_LIST_HEAD(&splice); 4959eb73c1b7SMiao Xie 4960eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4961eb73c1b7SMiao Xie list_splice_init(&fs_info->delalloc_roots, &splice); 4962eb73c1b7SMiao Xie while (!list_empty(&splice)) { 4963eb73c1b7SMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4964eb73c1b7SMiao Xie delalloc_root); 496500246528SJosef Bacik root = btrfs_grab_root(root); 4966eb73c1b7SMiao Xie BUG_ON(!root); 4967eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4968eb73c1b7SMiao Xie 4969eb73c1b7SMiao Xie btrfs_destroy_delalloc_inodes(root); 497000246528SJosef Bacik btrfs_put_root(root); 4971eb73c1b7SMiao Xie 4972eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4973eb73c1b7SMiao Xie } 4974eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4975acce952bSliubo } 4976acce952bSliubo 49772ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 4978acce952bSliubo struct extent_io_tree *dirty_pages, 4979acce952bSliubo int mark) 4980acce952bSliubo { 4981acce952bSliubo int ret; 4982acce952bSliubo struct extent_buffer *eb; 4983acce952bSliubo u64 start = 0; 4984acce952bSliubo u64 end; 4985acce952bSliubo 4986acce952bSliubo while (1) { 4987acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 4988e6138876SJosef Bacik mark, NULL); 4989acce952bSliubo if (ret) 4990acce952bSliubo break; 4991acce952bSliubo 499291166212SDavid Sterba clear_extent_bits(dirty_pages, start, end, mark); 4993acce952bSliubo while (start <= end) { 49940b246afaSJeff Mahoney eb = find_extent_buffer(fs_info, start); 49950b246afaSJeff Mahoney start += fs_info->nodesize; 4996fd8b2b61SJosef Bacik if (!eb) 4997acce952bSliubo continue; 4998acce952bSliubo 4999c4e54a65SJosef Bacik btrfs_tree_lock(eb); 5000c4e54a65SJosef Bacik wait_on_extent_buffer_writeback(eb); 5001*190a8339SJosef Bacik btrfs_clear_buffer_dirty(NULL, eb); 5002c4e54a65SJosef Bacik btrfs_tree_unlock(eb); 5003c4e54a65SJosef Bacik 5004fd8b2b61SJosef Bacik free_extent_buffer_stale(eb); 5005acce952bSliubo } 5006acce952bSliubo } 5007acce952bSliubo 5008acce952bSliubo return ret; 5009acce952bSliubo } 5010acce952bSliubo 50112ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 5012fe119a6eSNikolay Borisov struct extent_io_tree *unpin) 5013acce952bSliubo { 5014acce952bSliubo u64 start; 5015acce952bSliubo u64 end; 5016acce952bSliubo int ret; 5017acce952bSliubo 5018acce952bSliubo while (1) { 50190e6ec385SFilipe Manana struct extent_state *cached_state = NULL; 50200e6ec385SFilipe Manana 5021fcd5e742SLu Fengqi /* 5022fcd5e742SLu Fengqi * The btrfs_finish_extent_commit() may get the same range as 5023fcd5e742SLu Fengqi * ours between find_first_extent_bit and clear_extent_dirty. 5024fcd5e742SLu Fengqi * Hence, hold the unused_bg_unpin_mutex to avoid double unpin 5025fcd5e742SLu Fengqi * the same extent range. 5026fcd5e742SLu Fengqi */ 5027fcd5e742SLu Fengqi mutex_lock(&fs_info->unused_bg_unpin_mutex); 5028acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 50290e6ec385SFilipe Manana EXTENT_DIRTY, &cached_state); 5030fcd5e742SLu Fengqi if (ret) { 5031fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 5032acce952bSliubo break; 5033fcd5e742SLu Fengqi } 5034acce952bSliubo 50350e6ec385SFilipe Manana clear_extent_dirty(unpin, start, end, &cached_state); 50360e6ec385SFilipe Manana free_extent_state(cached_state); 50372ff7e61eSJeff Mahoney btrfs_error_unpin_extent_range(fs_info, start, end); 5038fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 5039acce952bSliubo cond_resched(); 5040acce952bSliubo } 5041acce952bSliubo 5042acce952bSliubo return 0; 5043acce952bSliubo } 5044acce952bSliubo 504532da5386SDavid Sterba static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache) 5046c79a1751SLiu Bo { 5047c79a1751SLiu Bo struct inode *inode; 5048c79a1751SLiu Bo 5049c79a1751SLiu Bo inode = cache->io_ctl.inode; 5050c79a1751SLiu Bo if (inode) { 5051c79a1751SLiu Bo invalidate_inode_pages2(inode->i_mapping); 5052c79a1751SLiu Bo BTRFS_I(inode)->generation = 0; 5053c79a1751SLiu Bo cache->io_ctl.inode = NULL; 5054c79a1751SLiu Bo iput(inode); 5055c79a1751SLiu Bo } 5056bbc37d6eSFilipe Manana ASSERT(cache->io_ctl.pages == NULL); 5057c79a1751SLiu Bo btrfs_put_block_group(cache); 5058c79a1751SLiu Bo } 5059c79a1751SLiu Bo 5060c79a1751SLiu Bo void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, 50612ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 5062c79a1751SLiu Bo { 506332da5386SDavid Sterba struct btrfs_block_group *cache; 5064c79a1751SLiu Bo 5065c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 5066c79a1751SLiu Bo while (!list_empty(&cur_trans->dirty_bgs)) { 5067c79a1751SLiu Bo cache = list_first_entry(&cur_trans->dirty_bgs, 506832da5386SDavid Sterba struct btrfs_block_group, 5069c79a1751SLiu Bo dirty_list); 5070c79a1751SLiu Bo 5071c79a1751SLiu Bo if (!list_empty(&cache->io_list)) { 5072c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 5073c79a1751SLiu Bo list_del_init(&cache->io_list); 5074c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 5075c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 5076c79a1751SLiu Bo } 5077c79a1751SLiu Bo 5078c79a1751SLiu Bo list_del_init(&cache->dirty_list); 5079c79a1751SLiu Bo spin_lock(&cache->lock); 5080c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 5081c79a1751SLiu Bo spin_unlock(&cache->lock); 5082c79a1751SLiu Bo 5083c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 5084c79a1751SLiu Bo btrfs_put_block_group(cache); 5085ba2c4d4eSJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 5086c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 5087c79a1751SLiu Bo } 5088c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 5089c79a1751SLiu Bo 509045ae2c18SNikolay Borisov /* 509145ae2c18SNikolay Borisov * Refer to the definition of io_bgs member for details why it's safe 509245ae2c18SNikolay Borisov * to use it without any locking 509345ae2c18SNikolay Borisov */ 5094c79a1751SLiu Bo while (!list_empty(&cur_trans->io_bgs)) { 5095c79a1751SLiu Bo cache = list_first_entry(&cur_trans->io_bgs, 509632da5386SDavid Sterba struct btrfs_block_group, 5097c79a1751SLiu Bo io_list); 5098c79a1751SLiu Bo 5099c79a1751SLiu Bo list_del_init(&cache->io_list); 5100c79a1751SLiu Bo spin_lock(&cache->lock); 5101c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 5102c79a1751SLiu Bo spin_unlock(&cache->lock); 5103c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 5104c79a1751SLiu Bo } 5105c79a1751SLiu Bo } 5106c79a1751SLiu Bo 510749b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 51082ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 510949b25e05SJeff Mahoney { 5110bbbf7243SNikolay Borisov struct btrfs_device *dev, *tmp; 5111bbbf7243SNikolay Borisov 51122ff7e61eSJeff Mahoney btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 5113c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->dirty_bgs)); 5114c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->io_bgs)); 5115c79a1751SLiu Bo 5116bbbf7243SNikolay Borisov list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list, 5117bbbf7243SNikolay Borisov post_commit_list) { 5118bbbf7243SNikolay Borisov list_del_init(&dev->post_commit_list); 5119bbbf7243SNikolay Borisov } 5120bbbf7243SNikolay Borisov 51212ff7e61eSJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, fs_info); 512249b25e05SJeff Mahoney 51234a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 51240b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 512549b25e05SJeff Mahoney 51264a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 51270b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 512849b25e05SJeff Mahoney 5129ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 513067cde344SMiao Xie 51312ff7e61eSJeff Mahoney btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, 513249b25e05SJeff Mahoney EXTENT_DIRTY); 5133fe119a6eSNikolay Borisov btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents); 513449b25e05SJeff Mahoney 5135d3575156SNaohiro Aota btrfs_free_redirty_list(cur_trans); 5136d3575156SNaohiro Aota 51374a9d8bdeSMiao Xie cur_trans->state =TRANS_STATE_COMPLETED; 51384a9d8bdeSMiao Xie wake_up(&cur_trans->commit_wait); 513949b25e05SJeff Mahoney } 514049b25e05SJeff Mahoney 51412ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) 5142acce952bSliubo { 5143acce952bSliubo struct btrfs_transaction *t; 5144acce952bSliubo 51450b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 5146acce952bSliubo 51470b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 51480b246afaSJeff Mahoney while (!list_empty(&fs_info->trans_list)) { 51490b246afaSJeff Mahoney t = list_first_entry(&fs_info->trans_list, 5150724e2315SJosef Bacik struct btrfs_transaction, list); 5151724e2315SJosef Bacik if (t->state >= TRANS_STATE_COMMIT_START) { 51529b64f57dSElena Reshetova refcount_inc(&t->use_count); 51530b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 51542ff7e61eSJeff Mahoney btrfs_wait_for_commit(fs_info, t->transid); 5155724e2315SJosef Bacik btrfs_put_transaction(t); 51560b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5157724e2315SJosef Bacik continue; 5158724e2315SJosef Bacik } 51590b246afaSJeff Mahoney if (t == fs_info->running_transaction) { 5160724e2315SJosef Bacik t->state = TRANS_STATE_COMMIT_DOING; 51610b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 5162724e2315SJosef Bacik /* 5163724e2315SJosef Bacik * We wait for 0 num_writers since we don't hold a trans 5164724e2315SJosef Bacik * handle open currently for this transaction. 5165724e2315SJosef Bacik */ 5166724e2315SJosef Bacik wait_event(t->writer_wait, 5167724e2315SJosef Bacik atomic_read(&t->num_writers) == 0); 5168724e2315SJosef Bacik } else { 51690b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 5170724e2315SJosef Bacik } 51712ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(t, fs_info); 5172724e2315SJosef Bacik 51730b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 51740b246afaSJeff Mahoney if (t == fs_info->running_transaction) 51750b246afaSJeff Mahoney fs_info->running_transaction = NULL; 5176724e2315SJosef Bacik list_del_init(&t->list); 51770b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 5178a4abeea4SJosef Bacik 5179724e2315SJosef Bacik btrfs_put_transaction(t); 51802e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 51810b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5182724e2315SJosef Bacik } 51830b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 51840b246afaSJeff Mahoney btrfs_destroy_all_ordered_extents(fs_info); 5185ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 5186ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 51870b246afaSJeff Mahoney btrfs_destroy_all_delalloc_inodes(fs_info); 5188ef67963dSJosef Bacik btrfs_drop_all_logs(fs_info); 51890b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 5190acce952bSliubo 5191acce952bSliubo return 0; 5192acce952bSliubo } 5193ec7d6dfdSNikolay Borisov 5194453e4873SNikolay Borisov int btrfs_init_root_free_objectid(struct btrfs_root *root) 5195ec7d6dfdSNikolay Borisov { 5196ec7d6dfdSNikolay Borisov struct btrfs_path *path; 5197ec7d6dfdSNikolay Borisov int ret; 5198ec7d6dfdSNikolay Borisov struct extent_buffer *l; 5199ec7d6dfdSNikolay Borisov struct btrfs_key search_key; 5200ec7d6dfdSNikolay Borisov struct btrfs_key found_key; 5201ec7d6dfdSNikolay Borisov int slot; 5202ec7d6dfdSNikolay Borisov 5203ec7d6dfdSNikolay Borisov path = btrfs_alloc_path(); 5204ec7d6dfdSNikolay Borisov if (!path) 5205ec7d6dfdSNikolay Borisov return -ENOMEM; 5206ec7d6dfdSNikolay Borisov 5207ec7d6dfdSNikolay Borisov search_key.objectid = BTRFS_LAST_FREE_OBJECTID; 5208ec7d6dfdSNikolay Borisov search_key.type = -1; 5209ec7d6dfdSNikolay Borisov search_key.offset = (u64)-1; 5210ec7d6dfdSNikolay Borisov ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); 5211ec7d6dfdSNikolay Borisov if (ret < 0) 5212ec7d6dfdSNikolay Borisov goto error; 5213ec7d6dfdSNikolay Borisov BUG_ON(ret == 0); /* Corruption */ 5214ec7d6dfdSNikolay Borisov if (path->slots[0] > 0) { 5215ec7d6dfdSNikolay Borisov slot = path->slots[0] - 1; 5216ec7d6dfdSNikolay Borisov l = path->nodes[0]; 5217ec7d6dfdSNikolay Borisov btrfs_item_key_to_cpu(l, &found_key, slot); 521823125104SNikolay Borisov root->free_objectid = max_t(u64, found_key.objectid + 1, 521923125104SNikolay Borisov BTRFS_FIRST_FREE_OBJECTID); 5220ec7d6dfdSNikolay Borisov } else { 522123125104SNikolay Borisov root->free_objectid = BTRFS_FIRST_FREE_OBJECTID; 5222ec7d6dfdSNikolay Borisov } 5223ec7d6dfdSNikolay Borisov ret = 0; 5224ec7d6dfdSNikolay Borisov error: 5225ec7d6dfdSNikolay Borisov btrfs_free_path(path); 5226ec7d6dfdSNikolay Borisov return ret; 5227ec7d6dfdSNikolay Borisov } 5228ec7d6dfdSNikolay Borisov 5229543068a2SNikolay Borisov int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid) 5230ec7d6dfdSNikolay Borisov { 5231ec7d6dfdSNikolay Borisov int ret; 5232ec7d6dfdSNikolay Borisov mutex_lock(&root->objectid_mutex); 5233ec7d6dfdSNikolay Borisov 52346b8fad57SNikolay Borisov if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) { 5235ec7d6dfdSNikolay Borisov btrfs_warn(root->fs_info, 5236ec7d6dfdSNikolay Borisov "the objectid of root %llu reaches its highest value", 5237ec7d6dfdSNikolay Borisov root->root_key.objectid); 5238ec7d6dfdSNikolay Borisov ret = -ENOSPC; 5239ec7d6dfdSNikolay Borisov goto out; 5240ec7d6dfdSNikolay Borisov } 5241ec7d6dfdSNikolay Borisov 524223125104SNikolay Borisov *objectid = root->free_objectid++; 5243ec7d6dfdSNikolay Borisov ret = 0; 5244ec7d6dfdSNikolay Borisov out: 5245ec7d6dfdSNikolay Borisov mutex_unlock(&root->objectid_mutex); 5246ec7d6dfdSNikolay Borisov return ret; 5247ec7d6dfdSNikolay Borisov } 5248