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> 80f7d52f4SChris Mason #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" 260b86a832SChris Mason #include "volumes.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" 32581bb050SLi Zefan #include "inode-map.h" 3321adbd5cSStefan Behrens #include "check-integrity.h" 34606686eeSJosef Bacik #include "rcu-string.h" 358dabb742SStefan Behrens #include "dev-replace.h" 3653b381b3SDavid Woodhouse #include "raid56.h" 375ac1d209SJeff Mahoney #include "sysfs.h" 38fcebe456SJosef Bacik #include "qgroup.h" 39ebb8765bSAnand Jain #include "compression.h" 40557ea5ddSQu Wenruo #include "tree-checker.h" 41fd708b81SJosef Bacik #include "ref-verify.h" 42aac0023cSJosef Bacik #include "block-group.h" 43b0643e59SDennis Zhou #include "discard.h" 44f603bb94SNikolay Borisov #include "space-info.h" 45eb60ceacSChris Mason 46319e4d06SQu Wenruo #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\ 47319e4d06SQu Wenruo BTRFS_HEADER_FLAG_RELOC |\ 48319e4d06SQu Wenruo BTRFS_SUPER_FLAG_ERROR |\ 49319e4d06SQu Wenruo BTRFS_SUPER_FLAG_SEEDING |\ 50e2731e55SAnand Jain BTRFS_SUPER_FLAG_METADUMP |\ 51e2731e55SAnand Jain BTRFS_SUPER_FLAG_METADUMP_V2) 52319e4d06SQu Wenruo 538b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work); 54143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root); 55acce952bSliubo static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 562ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info); 57143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root); 582ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 59acce952bSliubo struct extent_io_tree *dirty_pages, 60acce952bSliubo int mark); 612ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 62acce952bSliubo struct extent_io_tree *pinned_extents); 632ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info); 642ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info); 65ce9adaa5SChris Mason 66d352ac68SChris Mason /* 6797eb6b69SDavid Sterba * btrfs_end_io_wq structs are used to do processing in task context when an IO 6897eb6b69SDavid Sterba * is complete. This is used during reads to verify checksums, and it is used 69d352ac68SChris Mason * by writes to insert metadata for new file extents after IO is complete. 70d352ac68SChris Mason */ 7197eb6b69SDavid Sterba struct btrfs_end_io_wq { 72ce9adaa5SChris Mason struct bio *bio; 73ce9adaa5SChris Mason bio_end_io_t *end_io; 74ce9adaa5SChris Mason void *private; 75ce9adaa5SChris Mason struct btrfs_fs_info *info; 764e4cbee9SChristoph Hellwig blk_status_t status; 77bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata; 788b712842SChris Mason struct btrfs_work work; 79ce9adaa5SChris Mason }; 800da5468fSChris Mason 8197eb6b69SDavid Sterba static struct kmem_cache *btrfs_end_io_wq_cache; 8297eb6b69SDavid Sterba 8397eb6b69SDavid Sterba int __init btrfs_end_io_wq_init(void) 8497eb6b69SDavid Sterba { 8597eb6b69SDavid Sterba btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq", 8697eb6b69SDavid Sterba sizeof(struct btrfs_end_io_wq), 8797eb6b69SDavid Sterba 0, 88fba4b697SNikolay Borisov SLAB_MEM_SPREAD, 8997eb6b69SDavid Sterba NULL); 9097eb6b69SDavid Sterba if (!btrfs_end_io_wq_cache) 9197eb6b69SDavid Sterba return -ENOMEM; 9297eb6b69SDavid Sterba return 0; 9397eb6b69SDavid Sterba } 9497eb6b69SDavid Sterba 95e67c718bSDavid Sterba void __cold btrfs_end_io_wq_exit(void) 9697eb6b69SDavid Sterba { 9797eb6b69SDavid Sterba kmem_cache_destroy(btrfs_end_io_wq_cache); 9897eb6b69SDavid Sterba } 9997eb6b69SDavid Sterba 100141386e1SJosef Bacik static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info) 101141386e1SJosef Bacik { 102141386e1SJosef Bacik if (fs_info->csum_shash) 103141386e1SJosef Bacik crypto_free_shash(fs_info->csum_shash); 104141386e1SJosef Bacik } 105141386e1SJosef Bacik 106d352ac68SChris Mason /* 107d352ac68SChris Mason * async submit bios are used to offload expensive checksumming 108d352ac68SChris Mason * onto the worker threads. They checksum file and metadata bios 109d352ac68SChris Mason * just before they are sent down the IO stack. 110d352ac68SChris Mason */ 11144b8bd7eSChris Mason struct async_submit_bio { 112c6100a4bSJosef Bacik void *private_data; 11344b8bd7eSChris Mason struct bio *bio; 114a758781dSDavid Sterba extent_submit_bio_start_t *submit_bio_start; 11544b8bd7eSChris Mason int mirror_num; 116eaf25d93SChris Mason /* 117eaf25d93SChris Mason * bio_offset is optional, can be used if the pages in the bio 118eaf25d93SChris Mason * can't tell us where in the file the bio should go 119eaf25d93SChris Mason */ 120eaf25d93SChris Mason u64 bio_offset; 1218b712842SChris Mason struct btrfs_work work; 1224e4cbee9SChristoph Hellwig blk_status_t status; 12344b8bd7eSChris Mason }; 12444b8bd7eSChris Mason 12585d4e461SChris Mason /* 12685d4e461SChris Mason * Lockdep class keys for extent_buffer->lock's in this root. For a given 12785d4e461SChris Mason * eb, the lockdep key is determined by the btrfs_root it belongs to and 12885d4e461SChris Mason * the level the eb occupies in the tree. 1294008c04aSChris Mason * 13085d4e461SChris Mason * Different roots are used for different purposes and may nest inside each 13185d4e461SChris Mason * other and they require separate keysets. As lockdep keys should be 13285d4e461SChris Mason * static, assign keysets according to the purpose of the root as indicated 1334fd786e6SMisono Tomohiro * by btrfs_root->root_key.objectid. This ensures that all special purpose 1344fd786e6SMisono Tomohiro * roots have separate keysets. 1354008c04aSChris Mason * 13685d4e461SChris Mason * Lock-nesting across peer nodes is always done with the immediate parent 13785d4e461SChris Mason * node locked thus preventing deadlock. As lockdep doesn't know this, use 13885d4e461SChris Mason * subclass to avoid triggering lockdep warning in such cases. 1394008c04aSChris Mason * 14085d4e461SChris Mason * The key is set by the readpage_end_io_hook after the buffer has passed 14185d4e461SChris Mason * csum validation but before the pages are unlocked. It is also set by 14285d4e461SChris Mason * btrfs_init_new_buffer on freshly allocated blocks. 14385d4e461SChris Mason * 14485d4e461SChris Mason * We also add a check to make sure the highest level of the tree is the 14585d4e461SChris Mason * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code 14685d4e461SChris Mason * needs update as well. 1474008c04aSChris Mason */ 1484008c04aSChris Mason #ifdef CONFIG_DEBUG_LOCK_ALLOC 1494008c04aSChris Mason # if BTRFS_MAX_LEVEL != 8 1504008c04aSChris Mason # error 1514008c04aSChris Mason # endif 15285d4e461SChris Mason 153ab1405aaSDavid Sterba #define DEFINE_LEVEL(stem, level) \ 154ab1405aaSDavid Sterba .names[level] = "btrfs-" stem "-0" #level, 155ab1405aaSDavid Sterba 156ab1405aaSDavid Sterba #define DEFINE_NAME(stem) \ 157ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 0) \ 158ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 1) \ 159ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 2) \ 160ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 3) \ 161ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 4) \ 162ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 5) \ 163ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 6) \ 164ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 7) 165ab1405aaSDavid Sterba 16685d4e461SChris Mason static struct btrfs_lockdep_keyset { 16785d4e461SChris Mason u64 id; /* root objectid */ 168ab1405aaSDavid Sterba /* Longest entry: btrfs-free-space-00 */ 169387824afSDavid Sterba char names[BTRFS_MAX_LEVEL][20]; 170387824afSDavid Sterba struct lock_class_key keys[BTRFS_MAX_LEVEL]; 17185d4e461SChris Mason } btrfs_lockdep_keysets[] = { 172ab1405aaSDavid Sterba { .id = BTRFS_ROOT_TREE_OBJECTID, DEFINE_NAME("root") }, 173ab1405aaSDavid Sterba { .id = BTRFS_EXTENT_TREE_OBJECTID, DEFINE_NAME("extent") }, 174ab1405aaSDavid Sterba { .id = BTRFS_CHUNK_TREE_OBJECTID, DEFINE_NAME("chunk") }, 175ab1405aaSDavid Sterba { .id = BTRFS_DEV_TREE_OBJECTID, DEFINE_NAME("dev") }, 176ab1405aaSDavid Sterba { .id = BTRFS_FS_TREE_OBJECTID, DEFINE_NAME("fs") }, 177ab1405aaSDavid Sterba { .id = BTRFS_CSUM_TREE_OBJECTID, DEFINE_NAME("csum") }, 178ab1405aaSDavid Sterba { .id = BTRFS_QUOTA_TREE_OBJECTID, DEFINE_NAME("quota") }, 179ab1405aaSDavid Sterba { .id = BTRFS_TREE_LOG_OBJECTID, DEFINE_NAME("log") }, 180ab1405aaSDavid Sterba { .id = BTRFS_TREE_RELOC_OBJECTID, DEFINE_NAME("treloc") }, 181ab1405aaSDavid Sterba { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, DEFINE_NAME("dreloc") }, 182ab1405aaSDavid Sterba { .id = BTRFS_UUID_TREE_OBJECTID, DEFINE_NAME("uuid") }, 183ab1405aaSDavid Sterba { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, DEFINE_NAME("free-space") }, 184ab1405aaSDavid Sterba { .id = 0, DEFINE_NAME("tree") }, 1854008c04aSChris Mason }; 18685d4e461SChris Mason 187ab1405aaSDavid Sterba #undef DEFINE_LEVEL 188ab1405aaSDavid Sterba #undef DEFINE_NAME 18985d4e461SChris Mason 19085d4e461SChris Mason void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb, 19185d4e461SChris Mason int level) 19285d4e461SChris Mason { 19385d4e461SChris Mason struct btrfs_lockdep_keyset *ks; 19485d4e461SChris Mason 19585d4e461SChris Mason BUG_ON(level >= ARRAY_SIZE(ks->keys)); 19685d4e461SChris Mason 19785d4e461SChris Mason /* find the matching keyset, id 0 is the default entry */ 19885d4e461SChris Mason for (ks = btrfs_lockdep_keysets; ks->id; ks++) 19985d4e461SChris Mason if (ks->id == objectid) 20085d4e461SChris Mason break; 20185d4e461SChris Mason 20285d4e461SChris Mason lockdep_set_class_and_name(&eb->lock, 20385d4e461SChris Mason &ks->keys[level], ks->names[level]); 20485d4e461SChris Mason } 20585d4e461SChris Mason 2064008c04aSChris Mason #endif 2074008c04aSChris Mason 208d352ac68SChris Mason /* 2092996e1f8SJohannes Thumshirn * Compute the csum of a btree block and store the result to provided buffer. 210d352ac68SChris Mason */ 211c67b3892SDavid Sterba static void csum_tree_block(struct extent_buffer *buf, u8 *result) 21219c00ddcSChris Mason { 213d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = buf->fs_info; 214e9be5a30SDavid Sterba const int num_pages = fs_info->nodesize >> PAGE_SHIFT; 215d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 21619c00ddcSChris Mason char *kaddr; 217e9be5a30SDavid Sterba int i; 218d5178578SJohannes Thumshirn 219d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 220d5178578SJohannes Thumshirn crypto_shash_init(shash); 221e9be5a30SDavid Sterba kaddr = page_address(buf->pages[0]); 222e9be5a30SDavid Sterba crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE, 223e9be5a30SDavid Sterba PAGE_SIZE - BTRFS_CSUM_SIZE); 22419c00ddcSChris Mason 225e9be5a30SDavid Sterba for (i = 1; i < num_pages; i++) { 226e9be5a30SDavid Sterba kaddr = page_address(buf->pages[i]); 227e9be5a30SDavid Sterba crypto_shash_update(shash, kaddr, PAGE_SIZE); 22819c00ddcSChris Mason } 22971a63551SDavid Sterba memset(result, 0, BTRFS_CSUM_SIZE); 230d5178578SJohannes Thumshirn crypto_shash_final(shash, result); 23119c00ddcSChris Mason } 23219c00ddcSChris Mason 233d352ac68SChris Mason /* 234d352ac68SChris Mason * we can't consider a given block up to date unless the transid of the 235d352ac68SChris Mason * block matches the transid in the parent node's pointer. This is how we 236d352ac68SChris Mason * detect blocks that either didn't get written at all or got written 237d352ac68SChris Mason * in the wrong place. 238d352ac68SChris Mason */ 2391259ab75SChris Mason static int verify_parent_transid(struct extent_io_tree *io_tree, 240b9fab919SChris Mason struct extent_buffer *eb, u64 parent_transid, 241b9fab919SChris Mason int atomic) 2421259ab75SChris Mason { 2432ac55d41SJosef Bacik struct extent_state *cached_state = NULL; 2441259ab75SChris Mason int ret; 2452755a0deSDavid Sterba bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB); 2461259ab75SChris Mason 2471259ab75SChris Mason if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 2481259ab75SChris Mason return 0; 2491259ab75SChris Mason 250b9fab919SChris Mason if (atomic) 251b9fab919SChris Mason return -EAGAIN; 252b9fab919SChris Mason 253a26e8c9fSJosef Bacik if (need_lock) { 254a26e8c9fSJosef Bacik btrfs_tree_read_lock(eb); 255300aa896SDavid Sterba btrfs_set_lock_blocking_read(eb); 256a26e8c9fSJosef Bacik } 257a26e8c9fSJosef Bacik 2582ac55d41SJosef Bacik lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 259ff13db41SDavid Sterba &cached_state); 2600b32f4bbSJosef Bacik if (extent_buffer_uptodate(eb) && 2611259ab75SChris Mason btrfs_header_generation(eb) == parent_transid) { 2621259ab75SChris Mason ret = 0; 2631259ab75SChris Mason goto out; 2641259ab75SChris Mason } 26594647322SDavid Sterba btrfs_err_rl(eb->fs_info, 26694647322SDavid Sterba "parent transid verify failed on %llu wanted %llu found %llu", 26794647322SDavid Sterba eb->start, 26829549aecSWang Shilong parent_transid, btrfs_header_generation(eb)); 2691259ab75SChris Mason ret = 1; 270a26e8c9fSJosef Bacik 271a26e8c9fSJosef Bacik /* 272a26e8c9fSJosef Bacik * Things reading via commit roots that don't have normal protection, 273a26e8c9fSJosef Bacik * like send, can have a really old block in cache that may point at a 27401327610SNicholas D Steeves * block that has been freed and re-allocated. So don't clear uptodate 275a26e8c9fSJosef Bacik * if we find an eb that is under IO (dirty/writeback) because we could 276a26e8c9fSJosef Bacik * end up reading in the stale data and then writing it back out and 277a26e8c9fSJosef Bacik * making everybody very sad. 278a26e8c9fSJosef Bacik */ 279a26e8c9fSJosef Bacik if (!extent_buffer_under_io(eb)) 2800b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 28133958dc6SChris Mason out: 2822ac55d41SJosef Bacik unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 283e43bbe5eSDavid Sterba &cached_state); 284472b909fSJosef Bacik if (need_lock) 285a26e8c9fSJosef Bacik btrfs_tree_read_unlock_blocking(eb); 2861259ab75SChris Mason return ret; 2871259ab75SChris Mason } 2881259ab75SChris Mason 289e7e16f48SJohannes Thumshirn static bool btrfs_supported_super_csum(u16 csum_type) 290e7e16f48SJohannes Thumshirn { 291e7e16f48SJohannes Thumshirn switch (csum_type) { 292e7e16f48SJohannes Thumshirn case BTRFS_CSUM_TYPE_CRC32: 2933951e7f0SJohannes Thumshirn case BTRFS_CSUM_TYPE_XXHASH: 2943831bf00SJohannes Thumshirn case BTRFS_CSUM_TYPE_SHA256: 295352ae07bSDavid Sterba case BTRFS_CSUM_TYPE_BLAKE2: 296e7e16f48SJohannes Thumshirn return true; 297e7e16f48SJohannes Thumshirn default: 298e7e16f48SJohannes Thumshirn return false; 299e7e16f48SJohannes Thumshirn } 300e7e16f48SJohannes Thumshirn } 301e7e16f48SJohannes Thumshirn 302d352ac68SChris Mason /* 3031104a885SDavid Sterba * Return 0 if the superblock checksum type matches the checksum value of that 3041104a885SDavid Sterba * algorithm. Pass the raw disk superblock data. 3051104a885SDavid Sterba */ 306ab8d0fc4SJeff Mahoney static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info, 307ab8d0fc4SJeff Mahoney char *raw_disk_sb) 3081104a885SDavid Sterba { 3091104a885SDavid Sterba struct btrfs_super_block *disk_sb = 3101104a885SDavid Sterba (struct btrfs_super_block *)raw_disk_sb; 31151bce6c9SJohannes Thumshirn char result[BTRFS_CSUM_SIZE]; 312d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 313d5178578SJohannes Thumshirn 314d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 3151104a885SDavid Sterba 3161104a885SDavid Sterba /* 3171104a885SDavid Sterba * The super_block structure does not span the whole 31851bce6c9SJohannes Thumshirn * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is 31951bce6c9SJohannes Thumshirn * filled with zeros and is included in the checksum. 3201104a885SDavid Sterba */ 321fd08001fSEric Biggers crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE, 322fd08001fSEric Biggers BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result); 3231104a885SDavid Sterba 32451bce6c9SJohannes Thumshirn if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb))) 325e7e16f48SJohannes Thumshirn return 1; 3261104a885SDavid Sterba 327e7e16f48SJohannes Thumshirn return 0; 3281104a885SDavid Sterba } 3291104a885SDavid Sterba 330e064d5e9SDavid Sterba int btrfs_verify_level_key(struct extent_buffer *eb, int level, 331ff76a864SLiu Bo struct btrfs_key *first_key, u64 parent_transid) 332581c1760SQu Wenruo { 333e064d5e9SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 334581c1760SQu Wenruo int found_level; 335581c1760SQu Wenruo struct btrfs_key found_key; 336581c1760SQu Wenruo int ret; 337581c1760SQu Wenruo 338581c1760SQu Wenruo found_level = btrfs_header_level(eb); 339581c1760SQu Wenruo if (found_level != level) { 34063489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 34163489055SQu Wenruo KERN_ERR "BTRFS: tree level check failed\n"); 342581c1760SQu Wenruo btrfs_err(fs_info, 343581c1760SQu Wenruo "tree level mismatch detected, bytenr=%llu level expected=%u has=%u", 344581c1760SQu Wenruo eb->start, level, found_level); 345581c1760SQu Wenruo return -EIO; 346581c1760SQu Wenruo } 347581c1760SQu Wenruo 348581c1760SQu Wenruo if (!first_key) 349581c1760SQu Wenruo return 0; 350581c1760SQu Wenruo 3515d41be6fSQu Wenruo /* 3525d41be6fSQu Wenruo * For live tree block (new tree blocks in current transaction), 3535d41be6fSQu Wenruo * we need proper lock context to avoid race, which is impossible here. 3545d41be6fSQu Wenruo * So we only checks tree blocks which is read from disk, whose 3555d41be6fSQu Wenruo * generation <= fs_info->last_trans_committed. 3565d41be6fSQu Wenruo */ 3575d41be6fSQu Wenruo if (btrfs_header_generation(eb) > fs_info->last_trans_committed) 3585d41be6fSQu Wenruo return 0; 35962fdaa52SQu Wenruo 36062fdaa52SQu Wenruo /* We have @first_key, so this @eb must have at least one item */ 36162fdaa52SQu Wenruo if (btrfs_header_nritems(eb) == 0) { 36262fdaa52SQu Wenruo btrfs_err(fs_info, 36362fdaa52SQu Wenruo "invalid tree nritems, bytenr=%llu nritems=0 expect >0", 36462fdaa52SQu Wenruo eb->start); 36562fdaa52SQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 36662fdaa52SQu Wenruo return -EUCLEAN; 36762fdaa52SQu Wenruo } 36862fdaa52SQu Wenruo 369581c1760SQu Wenruo if (found_level) 370581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &found_key, 0); 371581c1760SQu Wenruo else 372581c1760SQu Wenruo btrfs_item_key_to_cpu(eb, &found_key, 0); 373581c1760SQu Wenruo ret = btrfs_comp_cpu_keys(first_key, &found_key); 374581c1760SQu Wenruo 375581c1760SQu Wenruo if (ret) { 37663489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 37763489055SQu Wenruo KERN_ERR "BTRFS: tree first key check failed\n"); 378581c1760SQu Wenruo btrfs_err(fs_info, 379ff76a864SLiu Bo "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", 380ff76a864SLiu Bo eb->start, parent_transid, first_key->objectid, 381ff76a864SLiu Bo first_key->type, first_key->offset, 382ff76a864SLiu Bo found_key.objectid, found_key.type, 383ff76a864SLiu Bo found_key.offset); 384581c1760SQu Wenruo } 385581c1760SQu Wenruo return ret; 386581c1760SQu Wenruo } 387581c1760SQu Wenruo 3881104a885SDavid Sterba /* 389d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 390d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 391581c1760SQu Wenruo * 392581c1760SQu Wenruo * @parent_transid: expected transid, skip check if 0 393581c1760SQu Wenruo * @level: expected level, mandatory check 394581c1760SQu Wenruo * @first_key: expected key of first slot, skip check if NULL 395d352ac68SChris Mason */ 3965ab12d1fSDavid Sterba static int btree_read_extent_buffer_pages(struct extent_buffer *eb, 397581c1760SQu Wenruo u64 parent_transid, int level, 398581c1760SQu Wenruo struct btrfs_key *first_key) 399f188591eSChris Mason { 4005ab12d1fSDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 401f188591eSChris Mason struct extent_io_tree *io_tree; 402ea466794SJosef Bacik int failed = 0; 403f188591eSChris Mason int ret; 404f188591eSChris Mason int num_copies = 0; 405f188591eSChris Mason int mirror_num = 0; 406ea466794SJosef Bacik int failed_mirror = 0; 407f188591eSChris Mason 4080b246afaSJeff Mahoney io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; 409f188591eSChris Mason while (1) { 410f8397d69SNikolay Borisov clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 411c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num); 412256dd1bbSStefan Behrens if (!ret) { 413581c1760SQu Wenruo if (verify_parent_transid(io_tree, eb, 414b9fab919SChris Mason parent_transid, 0)) 415256dd1bbSStefan Behrens ret = -EIO; 416e064d5e9SDavid Sterba else if (btrfs_verify_level_key(eb, level, 417ff76a864SLiu Bo first_key, parent_transid)) 418581c1760SQu Wenruo ret = -EUCLEAN; 419581c1760SQu Wenruo else 420581c1760SQu Wenruo break; 421256dd1bbSStefan Behrens } 422d397712bSChris Mason 4230b246afaSJeff Mahoney num_copies = btrfs_num_copies(fs_info, 424f188591eSChris Mason eb->start, eb->len); 4254235298eSChris Mason if (num_copies == 1) 426ea466794SJosef Bacik break; 4274235298eSChris Mason 4285cf1ab56SJosef Bacik if (!failed_mirror) { 4295cf1ab56SJosef Bacik failed = 1; 4305cf1ab56SJosef Bacik failed_mirror = eb->read_mirror; 4315cf1ab56SJosef Bacik } 4325cf1ab56SJosef Bacik 433f188591eSChris Mason mirror_num++; 434ea466794SJosef Bacik if (mirror_num == failed_mirror) 435ea466794SJosef Bacik mirror_num++; 436ea466794SJosef Bacik 4374235298eSChris Mason if (mirror_num > num_copies) 438ea466794SJosef Bacik break; 439f188591eSChris Mason } 440ea466794SJosef Bacik 441c0901581SStefan Behrens if (failed && !ret && failed_mirror) 44220a1fbf9SDavid Sterba btrfs_repair_eb_io_failure(eb, failed_mirror); 443ea466794SJosef Bacik 444ea466794SJosef Bacik return ret; 445f188591eSChris Mason } 44619c00ddcSChris Mason 447d352ac68SChris Mason /* 448d397712bSChris Mason * checksum a dirty tree block before IO. This has extra checks to make sure 449d397712bSChris Mason * we only fill in the checksum field in the first page of a multi-page block 450d352ac68SChris Mason */ 451d397712bSChris Mason 45201d58472SDaniel Dressler static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page) 45319c00ddcSChris Mason { 4544eee4fa4SMiao Xie u64 start = page_offset(page); 45519c00ddcSChris Mason u64 found_start; 4562996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 4572996e1f8SJohannes Thumshirn u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); 45819c00ddcSChris Mason struct extent_buffer *eb; 4598d47a0d8SQu Wenruo int ret; 460f188591eSChris Mason 4614f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 4624f2de97aSJosef Bacik if (page != eb->pages[0]) 4634f2de97aSJosef Bacik return 0; 4640f805531SAlex Lyakas 46519c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 4660f805531SAlex Lyakas /* 4670f805531SAlex Lyakas * Please do not consolidate these warnings into a single if. 4680f805531SAlex Lyakas * It is useful to know what went wrong. 4690f805531SAlex Lyakas */ 4700f805531SAlex Lyakas if (WARN_ON(found_start != start)) 4710f805531SAlex Lyakas return -EUCLEAN; 4720f805531SAlex Lyakas if (WARN_ON(!PageUptodate(page))) 4730f805531SAlex Lyakas return -EUCLEAN; 4740f805531SAlex Lyakas 475de37aa51SNikolay Borisov ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid, 4769a8658e3SDavid Sterba offsetof(struct btrfs_header, fsid), 4779a8658e3SDavid Sterba BTRFS_FSID_SIZE) == 0); 4780f805531SAlex Lyakas 479c67b3892SDavid Sterba csum_tree_block(eb, result); 4802996e1f8SJohannes Thumshirn 4818d47a0d8SQu Wenruo if (btrfs_header_level(eb)) 4828d47a0d8SQu Wenruo ret = btrfs_check_node(eb); 4838d47a0d8SQu Wenruo else 4848d47a0d8SQu Wenruo ret = btrfs_check_leaf_full(eb); 4858d47a0d8SQu Wenruo 4868d47a0d8SQu Wenruo if (ret < 0) { 487c06631b0SQu Wenruo btrfs_print_tree(eb, 0); 4888d47a0d8SQu Wenruo btrfs_err(fs_info, 4898d47a0d8SQu Wenruo "block=%llu write time tree block corruption detected", 4908d47a0d8SQu Wenruo eb->start); 491c06631b0SQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 4928d47a0d8SQu Wenruo return ret; 4938d47a0d8SQu Wenruo } 4942996e1f8SJohannes Thumshirn write_extent_buffer(eb, result, 0, csum_size); 4958d47a0d8SQu Wenruo 4962996e1f8SJohannes Thumshirn return 0; 49719c00ddcSChris Mason } 49819c00ddcSChris Mason 499b0c9b3b0SDavid Sterba static int check_tree_block_fsid(struct extent_buffer *eb) 5002b82032cSYan Zheng { 501b0c9b3b0SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 502944d3f9fSNikolay Borisov struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 50344880fdcSAnand Jain u8 fsid[BTRFS_FSID_SIZE]; 504944d3f9fSNikolay Borisov u8 *metadata_uuid; 5052b82032cSYan Zheng 5069a8658e3SDavid Sterba read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid), 5079a8658e3SDavid Sterba BTRFS_FSID_SIZE); 5087239ff4bSNikolay Borisov /* 509944d3f9fSNikolay Borisov * Checking the incompat flag is only valid for the current fs. For 510944d3f9fSNikolay Borisov * seed devices it's forbidden to have their uuid changed so reading 511944d3f9fSNikolay Borisov * ->fsid in this case is fine 5127239ff4bSNikolay Borisov */ 513944d3f9fSNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID)) 5147239ff4bSNikolay Borisov metadata_uuid = fs_devices->metadata_uuid; 5157239ff4bSNikolay Borisov else 5167239ff4bSNikolay Borisov metadata_uuid = fs_devices->fsid; 5177239ff4bSNikolay Borisov 518944d3f9fSNikolay Borisov if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) 519944d3f9fSNikolay Borisov return 0; 520944d3f9fSNikolay Borisov 521944d3f9fSNikolay Borisov list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) 522944d3f9fSNikolay Borisov if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE)) 523944d3f9fSNikolay Borisov return 0; 524944d3f9fSNikolay Borisov 525944d3f9fSNikolay Borisov return 1; 5262b82032cSYan Zheng } 5272b82032cSYan Zheng 5289a446d6aSNikolay Borisov int btrfs_validate_metadata_buffer(struct btrfs_io_bio *io_bio, u64 phy_offset, 5299a446d6aSNikolay Borisov struct page *page, u64 start, u64 end, 5309a446d6aSNikolay Borisov int mirror) 531ce9adaa5SChris Mason { 532ce9adaa5SChris Mason u64 found_start; 533ce9adaa5SChris Mason int found_level; 534ce9adaa5SChris Mason struct extent_buffer *eb; 53515b6e8a8SDavid Sterba struct btrfs_fs_info *fs_info; 53615b6e8a8SDavid Sterba u16 csum_size; 537f188591eSChris Mason int ret = 0; 5382996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 539727011e0SChris Mason int reads_done; 540ce9adaa5SChris Mason 541ce9adaa5SChris Mason if (!page->private) 542ce9adaa5SChris Mason goto out; 543d397712bSChris Mason 5444f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 54515b6e8a8SDavid Sterba fs_info = eb->fs_info; 54615b6e8a8SDavid Sterba csum_size = btrfs_super_csum_size(fs_info->super_copy); 547d397712bSChris Mason 5480b32f4bbSJosef Bacik /* the pending IO might have been the only thing that kept this buffer 5490b32f4bbSJosef Bacik * in memory. Make sure we have a ref for all this other checks 5500b32f4bbSJosef Bacik */ 55167439dadSDavid Sterba atomic_inc(&eb->refs); 5520b32f4bbSJosef Bacik 5530b32f4bbSJosef Bacik reads_done = atomic_dec_and_test(&eb->io_pages); 554727011e0SChris Mason if (!reads_done) 555727011e0SChris Mason goto err; 556f188591eSChris Mason 5575cf1ab56SJosef Bacik eb->read_mirror = mirror; 558656f30dbSFilipe Manana if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 559ea466794SJosef Bacik ret = -EIO; 560ea466794SJosef Bacik goto err; 561ea466794SJosef Bacik } 562ea466794SJosef Bacik 563ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 564727011e0SChris Mason if (found_start != eb->start) { 565893bf4b1SSu Yue btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu", 566893bf4b1SSu Yue eb->start, found_start); 567f188591eSChris Mason ret = -EIO; 568ce9adaa5SChris Mason goto err; 569ce9adaa5SChris Mason } 570b0c9b3b0SDavid Sterba if (check_tree_block_fsid(eb)) { 57102873e43SZhao Lei btrfs_err_rl(fs_info, "bad fsid on block %llu", 57294647322SDavid Sterba eb->start); 5731259ab75SChris Mason ret = -EIO; 5741259ab75SChris Mason goto err; 5751259ab75SChris Mason } 576ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 5771c24c3ceSJosef Bacik if (found_level >= BTRFS_MAX_LEVEL) { 578893bf4b1SSu Yue btrfs_err(fs_info, "bad tree block level %d on %llu", 579893bf4b1SSu Yue (int)btrfs_header_level(eb), eb->start); 5801c24c3ceSJosef Bacik ret = -EIO; 5811c24c3ceSJosef Bacik goto err; 5821c24c3ceSJosef Bacik } 583ce9adaa5SChris Mason 58485d4e461SChris Mason btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), 58585d4e461SChris Mason eb, found_level); 5864008c04aSChris Mason 587c67b3892SDavid Sterba csum_tree_block(eb, result); 588a826d6dcSJosef Bacik 5892996e1f8SJohannes Thumshirn if (memcmp_extent_buffer(eb, result, 0, csum_size)) { 59035be8851SJohannes Thumshirn u8 val[BTRFS_CSUM_SIZE] = { 0 }; 5912996e1f8SJohannes Thumshirn 5922996e1f8SJohannes Thumshirn read_extent_buffer(eb, &val, 0, csum_size); 5932996e1f8SJohannes Thumshirn btrfs_warn_rl(fs_info, 59435be8851SJohannes Thumshirn "%s checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d", 5952996e1f8SJohannes Thumshirn fs_info->sb->s_id, eb->start, 59635be8851SJohannes Thumshirn CSUM_FMT_VALUE(csum_size, val), 59735be8851SJohannes Thumshirn CSUM_FMT_VALUE(csum_size, result), 59835be8851SJohannes Thumshirn btrfs_header_level(eb)); 5992996e1f8SJohannes Thumshirn ret = -EUCLEAN; 6002996e1f8SJohannes Thumshirn goto err; 6012996e1f8SJohannes Thumshirn } 6022996e1f8SJohannes Thumshirn 603a826d6dcSJosef Bacik /* 604a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 605a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 606a826d6dcSJosef Bacik * return -EIO. 607a826d6dcSJosef Bacik */ 6081c4360eeSDavid Sterba if (found_level == 0 && btrfs_check_leaf_full(eb)) { 609a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 610a826d6dcSJosef Bacik ret = -EIO; 611a826d6dcSJosef Bacik } 612ce9adaa5SChris Mason 613813fd1dcSDavid Sterba if (found_level > 0 && btrfs_check_node(eb)) 614053ab70fSLiu Bo ret = -EIO; 615053ab70fSLiu Bo 6160b32f4bbSJosef Bacik if (!ret) 6170b32f4bbSJosef Bacik set_extent_buffer_uptodate(eb); 61875391f0dSQu Wenruo else 61975391f0dSQu Wenruo btrfs_err(fs_info, 62075391f0dSQu Wenruo "block=%llu read time tree block corruption detected", 62175391f0dSQu Wenruo eb->start); 622ce9adaa5SChris Mason err: 62379fb65a1SJosef Bacik if (reads_done && 62479fb65a1SJosef Bacik test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 625d48d71aaSDavid Sterba btree_readahead_hook(eb, ret); 6264bb31e92SArne Jansen 62753b381b3SDavid Woodhouse if (ret) { 62853b381b3SDavid Woodhouse /* 62953b381b3SDavid Woodhouse * our io error hook is going to dec the io pages 63053b381b3SDavid Woodhouse * again, we have to make sure it has something 63153b381b3SDavid Woodhouse * to decrement 63253b381b3SDavid Woodhouse */ 63353b381b3SDavid Woodhouse atomic_inc(&eb->io_pages); 6340b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 63553b381b3SDavid Woodhouse } 6360b32f4bbSJosef Bacik free_extent_buffer(eb); 637ce9adaa5SChris Mason out: 638f188591eSChris Mason return ret; 639ce9adaa5SChris Mason } 640ce9adaa5SChris Mason 6414246a0b6SChristoph Hellwig static void end_workqueue_bio(struct bio *bio) 642ce9adaa5SChris Mason { 64397eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq = bio->bi_private; 644ce9adaa5SChris Mason struct btrfs_fs_info *fs_info; 6459e0af237SLiu Bo struct btrfs_workqueue *wq; 646ce9adaa5SChris Mason 647ce9adaa5SChris Mason fs_info = end_io_wq->info; 6484e4cbee9SChristoph Hellwig end_io_wq->status = bio->bi_status; 649d20f7043SChris Mason 65037226b21SMike Christie if (bio_op(bio) == REQ_OP_WRITE) { 651a0cac0ecSOmar Sandoval if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) 6529e0af237SLiu Bo wq = fs_info->endio_meta_write_workers; 653a0cac0ecSOmar Sandoval else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) 6549e0af237SLiu Bo wq = fs_info->endio_freespace_worker; 655a0cac0ecSOmar Sandoval else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 6569e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 657a0cac0ecSOmar Sandoval else 6589e0af237SLiu Bo wq = fs_info->endio_write_workers; 6599e0af237SLiu Bo } else { 6605c047a69SOmar Sandoval if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 6619e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 662a0cac0ecSOmar Sandoval else if (end_io_wq->metadata) 6639e0af237SLiu Bo wq = fs_info->endio_meta_workers; 664a0cac0ecSOmar Sandoval else 6659e0af237SLiu Bo wq = fs_info->endio_workers; 6669e0af237SLiu Bo } 6679e0af237SLiu Bo 668a0cac0ecSOmar Sandoval btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL); 6699e0af237SLiu Bo btrfs_queue_work(wq, &end_io_wq->work); 670ce9adaa5SChris Mason } 671ce9adaa5SChris Mason 6724e4cbee9SChristoph Hellwig blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 673bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata) 6740b86a832SChris Mason { 67597eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 6768b110e39SMiao Xie 67797eb6b69SDavid Sterba end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS); 678ce9adaa5SChris Mason if (!end_io_wq) 6794e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 680ce9adaa5SChris Mason 681ce9adaa5SChris Mason end_io_wq->private = bio->bi_private; 682ce9adaa5SChris Mason end_io_wq->end_io = bio->bi_end_io; 68322c59948SChris Mason end_io_wq->info = info; 6844e4cbee9SChristoph Hellwig end_io_wq->status = 0; 685ce9adaa5SChris Mason end_io_wq->bio = bio; 68622c59948SChris Mason end_io_wq->metadata = metadata; 687ce9adaa5SChris Mason 688ce9adaa5SChris Mason bio->bi_private = end_io_wq; 689ce9adaa5SChris Mason bio->bi_end_io = end_workqueue_bio; 69022c59948SChris Mason return 0; 69122c59948SChris Mason } 69222c59948SChris Mason 6934a69a410SChris Mason static void run_one_async_start(struct btrfs_work *work) 6944a69a410SChris Mason { 6954a69a410SChris Mason struct async_submit_bio *async; 6964e4cbee9SChristoph Hellwig blk_status_t ret; 6974a69a410SChris Mason 6984a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 699c6100a4bSJosef Bacik ret = async->submit_bio_start(async->private_data, async->bio, 700eaf25d93SChris Mason async->bio_offset); 70179787eaaSJeff Mahoney if (ret) 7024e4cbee9SChristoph Hellwig async->status = ret; 7034a69a410SChris Mason } 7044a69a410SChris Mason 70506ea01b1SDavid Sterba /* 70606ea01b1SDavid Sterba * In order to insert checksums into the metadata in large chunks, we wait 70706ea01b1SDavid Sterba * until bio submission time. All the pages in the bio are checksummed and 70806ea01b1SDavid Sterba * sums are attached onto the ordered extent record. 70906ea01b1SDavid Sterba * 71006ea01b1SDavid Sterba * At IO completion time the csums attached on the ordered extent record are 71106ea01b1SDavid Sterba * inserted into the tree. 71206ea01b1SDavid Sterba */ 7134a69a410SChris Mason static void run_one_async_done(struct btrfs_work *work) 7148b712842SChris Mason { 7158b712842SChris Mason struct async_submit_bio *async; 71606ea01b1SDavid Sterba struct inode *inode; 71706ea01b1SDavid Sterba blk_status_t ret; 7188b712842SChris Mason 7198b712842SChris Mason async = container_of(work, struct async_submit_bio, work); 72006ea01b1SDavid Sterba inode = async->private_data; 7214854ddd0SChris Mason 722bb7ab3b9SAdam Buchbinder /* If an error occurred we just want to clean up the bio and move on */ 7234e4cbee9SChristoph Hellwig if (async->status) { 7244e4cbee9SChristoph Hellwig async->bio->bi_status = async->status; 7254246a0b6SChristoph Hellwig bio_endio(async->bio); 72679787eaaSJeff Mahoney return; 72779787eaaSJeff Mahoney } 72879787eaaSJeff Mahoney 729ec39f769SChris Mason /* 730ec39f769SChris Mason * All of the bios that pass through here are from async helpers. 731ec39f769SChris Mason * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context. 732ec39f769SChris Mason * This changes nothing when cgroups aren't in use. 733ec39f769SChris Mason */ 734ec39f769SChris Mason async->bio->bi_opf |= REQ_CGROUP_PUNT; 73508635baeSChris Mason ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num); 73606ea01b1SDavid Sterba if (ret) { 73706ea01b1SDavid Sterba async->bio->bi_status = ret; 73806ea01b1SDavid Sterba bio_endio(async->bio); 73906ea01b1SDavid Sterba } 7404a69a410SChris Mason } 7414a69a410SChris Mason 7424a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 7434a69a410SChris Mason { 7444a69a410SChris Mason struct async_submit_bio *async; 7454a69a410SChris Mason 7464a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 7478b712842SChris Mason kfree(async); 7488b712842SChris Mason } 7498b712842SChris Mason 7508c27cb35SLinus Torvalds blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio, 751c6100a4bSJosef Bacik int mirror_num, unsigned long bio_flags, 752c6100a4bSJosef Bacik u64 bio_offset, void *private_data, 753e288c080SDavid Sterba extent_submit_bio_start_t *submit_bio_start) 75444b8bd7eSChris Mason { 75544b8bd7eSChris Mason struct async_submit_bio *async; 75644b8bd7eSChris Mason 75744b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 75844b8bd7eSChris Mason if (!async) 7594e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 76044b8bd7eSChris Mason 761c6100a4bSJosef Bacik async->private_data = private_data; 76244b8bd7eSChris Mason async->bio = bio; 76344b8bd7eSChris Mason async->mirror_num = mirror_num; 7644a69a410SChris Mason async->submit_bio_start = submit_bio_start; 7654a69a410SChris Mason 766a0cac0ecSOmar Sandoval btrfs_init_work(&async->work, run_one_async_start, run_one_async_done, 767a0cac0ecSOmar Sandoval run_one_async_free); 7684a69a410SChris Mason 769eaf25d93SChris Mason async->bio_offset = bio_offset; 7708c8bee1dSChris Mason 7714e4cbee9SChristoph Hellwig async->status = 0; 77279787eaaSJeff Mahoney 77367f055c7SChristoph Hellwig if (op_is_sync(bio->bi_opf)) 7745cdc7ad3SQu Wenruo btrfs_set_work_high_priority(&async->work); 775d313d7a3SChris Mason 7765cdc7ad3SQu Wenruo btrfs_queue_work(fs_info->workers, &async->work); 77744b8bd7eSChris Mason return 0; 77844b8bd7eSChris Mason } 77944b8bd7eSChris Mason 7804e4cbee9SChristoph Hellwig static blk_status_t btree_csum_one_bio(struct bio *bio) 781ce3ed71aSChris Mason { 7822c30c71bSKent Overstreet struct bio_vec *bvec; 783ce3ed71aSChris Mason struct btrfs_root *root; 7842b070cfeSChristoph Hellwig int ret = 0; 7856dc4f100SMing Lei struct bvec_iter_all iter_all; 786ce3ed71aSChris Mason 787c09abff8SDavid Sterba ASSERT(!bio_flagged(bio, BIO_CLONED)); 7882b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 789ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 79001d58472SDaniel Dressler ret = csum_dirty_buffer(root->fs_info, bvec->bv_page); 79179787eaaSJeff Mahoney if (ret) 79279787eaaSJeff Mahoney break; 793ce3ed71aSChris Mason } 7942c30c71bSKent Overstreet 7954e4cbee9SChristoph Hellwig return errno_to_blk_status(ret); 796ce3ed71aSChris Mason } 797ce3ed71aSChris Mason 798d0ee3934SDavid Sterba static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio, 799eaf25d93SChris Mason u64 bio_offset) 80022c59948SChris Mason { 8018b712842SChris Mason /* 8028b712842SChris Mason * when we're called for a write, we're already in the async 8035443be45SChris Mason * submission context. Just jump into btrfs_map_bio 8048b712842SChris Mason */ 80579787eaaSJeff Mahoney return btree_csum_one_bio(bio); 80622c59948SChris Mason } 80722c59948SChris Mason 8089b4e675aSDavid Sterba static int check_async_write(struct btrfs_fs_info *fs_info, 8099b4e675aSDavid Sterba struct btrfs_inode *bi) 810de0022b9SJosef Bacik { 8116300463bSLiu Bo if (atomic_read(&bi->sync_writers)) 8126300463bSLiu Bo return 0; 8139b4e675aSDavid Sterba if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags)) 814de0022b9SJosef Bacik return 0; 815de0022b9SJosef Bacik return 1; 816de0022b9SJosef Bacik } 817de0022b9SJosef Bacik 8181b36294aSNikolay Borisov blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio, 8191b36294aSNikolay Borisov int mirror_num, unsigned long bio_flags) 82044b8bd7eSChris Mason { 8210b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 8229b4e675aSDavid Sterba int async = check_async_write(fs_info, BTRFS_I(inode)); 8234e4cbee9SChristoph Hellwig blk_status_t ret; 824cad321adSChris Mason 82537226b21SMike Christie if (bio_op(bio) != REQ_OP_WRITE) { 826cad321adSChris Mason /* 827cad321adSChris Mason * called for a read, do the setup so that checksum validation 828cad321adSChris Mason * can happen in the async kernel threads 829cad321adSChris Mason */ 8300b246afaSJeff Mahoney ret = btrfs_bio_wq_end_io(fs_info, bio, 8310b246afaSJeff Mahoney BTRFS_WQ_ENDIO_METADATA); 8321d4284bdSChris Mason if (ret) 83361891923SStefan Behrens goto out_w_error; 83408635baeSChris Mason ret = btrfs_map_bio(fs_info, bio, mirror_num); 835de0022b9SJosef Bacik } else if (!async) { 836de0022b9SJosef Bacik ret = btree_csum_one_bio(bio); 837de0022b9SJosef Bacik if (ret) 83861891923SStefan Behrens goto out_w_error; 83908635baeSChris Mason ret = btrfs_map_bio(fs_info, bio, mirror_num); 84061891923SStefan Behrens } else { 841cad321adSChris Mason /* 84261891923SStefan Behrens * kthread helpers are used to submit writes so that 84361891923SStefan Behrens * checksumming can happen in parallel across all CPUs 844cad321adSChris Mason */ 845c6100a4bSJosef Bacik ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0, 846e68f2ee7SNikolay Borisov 0, inode, btree_submit_bio_start); 84744b8bd7eSChris Mason } 84844b8bd7eSChris Mason 8494246a0b6SChristoph Hellwig if (ret) 8504246a0b6SChristoph Hellwig goto out_w_error; 8514246a0b6SChristoph Hellwig return 0; 8524246a0b6SChristoph Hellwig 85361891923SStefan Behrens out_w_error: 8544e4cbee9SChristoph Hellwig bio->bi_status = ret; 8554246a0b6SChristoph Hellwig bio_endio(bio); 85661891923SStefan Behrens return ret; 85761891923SStefan Behrens } 85861891923SStefan Behrens 8593dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 860784b4e29SChris Mason static int btree_migratepage(struct address_space *mapping, 861a6bc32b8SMel Gorman struct page *newpage, struct page *page, 862a6bc32b8SMel Gorman enum migrate_mode mode) 863784b4e29SChris Mason { 864784b4e29SChris Mason /* 865784b4e29SChris Mason * we can't safely write a btree page from here, 866784b4e29SChris Mason * we haven't done the locking hook 867784b4e29SChris Mason */ 868784b4e29SChris Mason if (PageDirty(page)) 869784b4e29SChris Mason return -EAGAIN; 870784b4e29SChris Mason /* 871784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 872784b4e29SChris Mason * We must have no buffers or drop them. 873784b4e29SChris Mason */ 874784b4e29SChris Mason if (page_has_private(page) && 875784b4e29SChris Mason !try_to_release_page(page, GFP_KERNEL)) 876784b4e29SChris Mason return -EAGAIN; 877a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 878784b4e29SChris Mason } 8793dd1462eSJan Beulich #endif 880784b4e29SChris Mason 8810da5468fSChris Mason 8820da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 8830da5468fSChris Mason struct writeback_control *wbc) 8840da5468fSChris Mason { 885e2d84521SMiao Xie struct btrfs_fs_info *fs_info; 886e2d84521SMiao Xie int ret; 887e2d84521SMiao Xie 888d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 889448d640bSChris Mason 890448d640bSChris Mason if (wbc->for_kupdate) 891448d640bSChris Mason return 0; 892448d640bSChris Mason 893e2d84521SMiao Xie fs_info = BTRFS_I(mapping->host)->root->fs_info; 894b9473439SChris Mason /* this is a bit racy, but that's ok */ 895d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 896d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 897d814a491SEthan Lien fs_info->dirty_metadata_batch); 898e2d84521SMiao Xie if (ret < 0) 899793955bcSChris Mason return 0; 900793955bcSChris Mason } 9010b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 9020da5468fSChris Mason } 9030da5468fSChris Mason 90470dec807SChris Mason static int btree_releasepage(struct page *page, gfp_t gfp_flags) 9055f39d397SChris Mason { 90698509cfcSChris Mason if (PageWriteback(page) || PageDirty(page)) 90798509cfcSChris Mason return 0; 9080c4e538bSDavid Sterba 909f7a52a40SDavid Sterba return try_release_extent_buffer(page); 910d98237b3SChris Mason } 911d98237b3SChris Mason 912d47992f8SLukas Czerner static void btree_invalidatepage(struct page *page, unsigned int offset, 913d47992f8SLukas Czerner unsigned int length) 914d98237b3SChris Mason { 915d1310b2eSChris Mason struct extent_io_tree *tree; 916d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9175f39d397SChris Mason extent_invalidatepage(tree, page, offset); 9185f39d397SChris Mason btree_releasepage(page, GFP_NOFS); 9199ad6b7bcSChris Mason if (PagePrivate(page)) { 920efe120a0SFrank Holton btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info, 921efe120a0SFrank Holton "page private not zero on page %llu", 922efe120a0SFrank Holton (unsigned long long)page_offset(page)); 923d1b89bc0SGuoqing Jiang detach_page_private(page); 9249ad6b7bcSChris Mason } 925d98237b3SChris Mason } 926d98237b3SChris Mason 9270b32f4bbSJosef Bacik static int btree_set_page_dirty(struct page *page) 9280b32f4bbSJosef Bacik { 929bb146eb2SJosef Bacik #ifdef DEBUG 9300b32f4bbSJosef Bacik struct extent_buffer *eb; 9310b32f4bbSJosef Bacik 9320b32f4bbSJosef Bacik BUG_ON(!PagePrivate(page)); 9330b32f4bbSJosef Bacik eb = (struct extent_buffer *)page->private; 9340b32f4bbSJosef Bacik BUG_ON(!eb); 9350b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 9360b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 9370b32f4bbSJosef Bacik btrfs_assert_tree_locked(eb); 938bb146eb2SJosef Bacik #endif 9390b32f4bbSJosef Bacik return __set_page_dirty_nobuffers(page); 9400b32f4bbSJosef Bacik } 9410b32f4bbSJosef Bacik 9427f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 9430da5468fSChris Mason .writepages = btree_writepages, 9445f39d397SChris Mason .releasepage = btree_releasepage, 9455f39d397SChris Mason .invalidatepage = btree_invalidatepage, 9465a92bc88SChris Mason #ifdef CONFIG_MIGRATION 947784b4e29SChris Mason .migratepage = btree_migratepage, 9485a92bc88SChris Mason #endif 9490b32f4bbSJosef Bacik .set_page_dirty = btree_set_page_dirty, 950d98237b3SChris Mason }; 951123abc88SChris Mason 9522ff7e61eSJeff Mahoney void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr) 953090d1875SChris Mason { 9545f39d397SChris Mason struct extent_buffer *buf = NULL; 955537f38f0SNikolay Borisov int ret; 956090d1875SChris Mason 9572ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 958c871b0f2SLiu Bo if (IS_ERR(buf)) 9596197d86eSDavid Sterba return; 960537f38f0SNikolay Borisov 961c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(buf, WAIT_NONE, 0); 962537f38f0SNikolay Borisov if (ret < 0) 963537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 964537f38f0SNikolay Borisov else 9655f39d397SChris Mason free_extent_buffer(buf); 966090d1875SChris Mason } 967090d1875SChris Mason 9682ff7e61eSJeff Mahoney struct extent_buffer *btrfs_find_create_tree_block( 9692ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info, 9700308af44SDavid Sterba u64 bytenr) 9710999df54SChris Mason { 9720b246afaSJeff Mahoney if (btrfs_is_testing(fs_info)) 9730b246afaSJeff Mahoney return alloc_test_extent_buffer(fs_info, bytenr); 9740b246afaSJeff Mahoney return alloc_extent_buffer(fs_info, bytenr); 9750999df54SChris Mason } 9760999df54SChris Mason 977581c1760SQu Wenruo /* 978581c1760SQu Wenruo * Read tree block at logical address @bytenr and do variant basic but critical 979581c1760SQu Wenruo * verification. 980581c1760SQu Wenruo * 981581c1760SQu Wenruo * @parent_transid: expected transid of this tree block, skip check if 0 982581c1760SQu Wenruo * @level: expected level, mandatory check 983581c1760SQu Wenruo * @first_key: expected key in slot 0, skip check if NULL 984581c1760SQu Wenruo */ 9852ff7e61eSJeff Mahoney struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, 986581c1760SQu Wenruo u64 parent_transid, int level, 987581c1760SQu Wenruo struct btrfs_key *first_key) 988e20d96d6SChris Mason { 9895f39d397SChris Mason struct extent_buffer *buf = NULL; 99019c00ddcSChris Mason int ret; 99119c00ddcSChris Mason 9922ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 993c871b0f2SLiu Bo if (IS_ERR(buf)) 994c871b0f2SLiu Bo return buf; 995e4204dedSChris Mason 9965ab12d1fSDavid Sterba ret = btree_read_extent_buffer_pages(buf, parent_transid, 997581c1760SQu Wenruo level, first_key); 9980f0fe8f7SFilipe David Borba Manana if (ret) { 999537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 100064c043deSLiu Bo return ERR_PTR(ret); 10010f0fe8f7SFilipe David Borba Manana } 10025f39d397SChris Mason return buf; 1003ce9adaa5SChris Mason 1004eb60ceacSChris Mason } 1005eb60ceacSChris Mason 10066a884d7dSDavid Sterba void btrfs_clean_tree_block(struct extent_buffer *buf) 1007ed2ff2cbSChris Mason { 10086a884d7dSDavid Sterba struct btrfs_fs_info *fs_info = buf->fs_info; 100955c69072SChris Mason if (btrfs_header_generation(buf) == 1010e2d84521SMiao Xie fs_info->running_transaction->transid) { 1011b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1012b4ce94deSChris Mason 1013b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1014104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 1015e2d84521SMiao Xie -buf->len, 1016e2d84521SMiao Xie fs_info->dirty_metadata_batch); 1017b9473439SChris Mason /* ugh, clear_extent_buffer_dirty needs to lock the page */ 10188bead258SDavid Sterba btrfs_set_lock_blocking_write(buf); 10190b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 1020925baeddSChris Mason } 10215f39d397SChris Mason } 1022ed7b63ebSJosef Bacik } 10235f39d397SChris Mason 1024da17066cSJeff Mahoney static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, 1025e20d96d6SChris Mason u64 objectid) 1026d97e63b6SChris Mason { 10277c0260eeSJeff Mahoney bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); 102896dfcb46SJosef Bacik root->fs_info = fs_info; 1029cfaa7295SChris Mason root->node = NULL; 1030a28ec197SChris Mason root->commit_root = NULL; 103127cdeb70SMiao Xie root->state = 0; 1032d68fc57bSYan, Zheng root->orphan_cleanup_state = 0; 10330b86a832SChris Mason 10340f7d52f4SChris Mason root->last_trans = 0; 103513a8a7c8SYan, Zheng root->highest_objectid = 0; 1036eb73c1b7SMiao Xie root->nr_delalloc_inodes = 0; 1037199c2a9cSMiao Xie root->nr_ordered_extents = 0; 10386bef4d31SEric Paris root->inode_tree = RB_ROOT; 103916cdcec7SMiao Xie INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1040f0486c68SYan, Zheng root->block_rsv = NULL; 10410b86a832SChris Mason 10420b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 10435d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 1044eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_inodes); 1045eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_root); 1046199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_extents); 1047199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_root); 1048d2311e69SQu Wenruo INIT_LIST_HEAD(&root->reloc_dirty_list); 10492ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 10502ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 10515d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1052eb73c1b7SMiao Xie spin_lock_init(&root->delalloc_lock); 1053199c2a9cSMiao Xie spin_lock_init(&root->ordered_extent_lock); 1054f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 10552ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 10562ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 10578287475aSQu Wenruo spin_lock_init(&root->qgroup_meta_rsv_lock); 1058a2135011SChris Mason mutex_init(&root->objectid_mutex); 1059e02119d5SChris Mason mutex_init(&root->log_mutex); 106031f3d255SMiao Xie mutex_init(&root->ordered_extent_mutex); 1061573bfb72SMiao Xie mutex_init(&root->delalloc_mutex); 1062c53e9653SQu Wenruo init_waitqueue_head(&root->qgroup_flush_wait); 10637237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 10647237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 10657237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 10668b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[0]); 10678b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[1]); 10687237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 10697237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 10707237f183SYan Zheng atomic_set(&root->log_writers, 0); 10712ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 10720700cea7SElena Reshetova refcount_set(&root->refs, 1); 10738ecebf4dSRobbie Ko atomic_set(&root->snapshot_force_cow, 0); 1074eede2bf3SOmar Sandoval atomic_set(&root->nr_swapfiles, 0); 10757237f183SYan Zheng root->log_transid = 0; 1076d1433debSMiao Xie root->log_transid_committed = -1; 1077257c62e1SChris Mason root->last_log_commit = 0; 1078e289f03eSFilipe Manana if (!dummy) { 107943eb5f29SQu Wenruo extent_io_tree_init(fs_info, &root->dirty_log_pages, 108043eb5f29SQu Wenruo IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL); 1081e289f03eSFilipe Manana extent_io_tree_init(fs_info, &root->log_csum_range, 1082e289f03eSFilipe Manana IO_TREE_LOG_CSUM_RANGE, NULL); 1083e289f03eSFilipe Manana } 1084017e5369SChris Mason 10853768f368SChris Mason memset(&root->root_key, 0, sizeof(root->root_key)); 10863768f368SChris Mason memset(&root->root_item, 0, sizeof(root->root_item)); 10876702ed49SChris Mason memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 10884d775673SChris Mason root->root_key.objectid = objectid; 10890ee5dc67SAl Viro root->anon_dev = 0; 10908ea05e3aSAlexander Block 10915f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 1092370a11b8SQu Wenruo btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks); 1093bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 1094bd647ce3SJosef Bacik INIT_LIST_HEAD(&root->leak_list); 1095bd647ce3SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 1096bd647ce3SJosef Bacik list_add_tail(&root->leak_list, &fs_info->allocated_roots); 1097bd647ce3SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 1098bd647ce3SJosef Bacik #endif 10993768f368SChris Mason } 11003768f368SChris Mason 110174e4d827SDavid Sterba static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, 110296dfcb46SJosef Bacik u64 objectid, gfp_t flags) 11036f07e42eSAl Viro { 110474e4d827SDavid Sterba struct btrfs_root *root = kzalloc(sizeof(*root), flags); 11056f07e42eSAl Viro if (root) 110696dfcb46SJosef Bacik __setup_root(root, fs_info, objectid); 11076f07e42eSAl Viro return root; 11086f07e42eSAl Viro } 11096f07e42eSAl Viro 111006ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 111106ea65a3SJosef Bacik /* Should only be used by the testing infrastructure */ 1112da17066cSJeff Mahoney struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) 111306ea65a3SJosef Bacik { 111406ea65a3SJosef Bacik struct btrfs_root *root; 111506ea65a3SJosef Bacik 11167c0260eeSJeff Mahoney if (!fs_info) 11177c0260eeSJeff Mahoney return ERR_PTR(-EINVAL); 11187c0260eeSJeff Mahoney 111996dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL); 112006ea65a3SJosef Bacik if (!root) 112106ea65a3SJosef Bacik return ERR_PTR(-ENOMEM); 1122da17066cSJeff Mahoney 1123b9ef22deSFeifei Xu /* We don't use the stripesize in selftest, set it as sectorsize */ 1124faa2dbf0SJosef Bacik root->alloc_bytenr = 0; 112506ea65a3SJosef Bacik 112606ea65a3SJosef Bacik return root; 112706ea65a3SJosef Bacik } 112806ea65a3SJosef Bacik #endif 112906ea65a3SJosef Bacik 113020897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 113120897f5cSArne Jansen u64 objectid) 113220897f5cSArne Jansen { 11339b7a2440SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 113420897f5cSArne Jansen struct extent_buffer *leaf; 113520897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 113620897f5cSArne Jansen struct btrfs_root *root; 113720897f5cSArne Jansen struct btrfs_key key; 1138b89f6d1fSFilipe Manana unsigned int nofs_flag; 113920897f5cSArne Jansen int ret = 0; 114020897f5cSArne Jansen 1141b89f6d1fSFilipe Manana /* 1142b89f6d1fSFilipe Manana * We're holding a transaction handle, so use a NOFS memory allocation 1143b89f6d1fSFilipe Manana * context to avoid deadlock if reclaim happens. 1144b89f6d1fSFilipe Manana */ 1145b89f6d1fSFilipe Manana nofs_flag = memalloc_nofs_save(); 114696dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL); 1147b89f6d1fSFilipe Manana memalloc_nofs_restore(nofs_flag); 114820897f5cSArne Jansen if (!root) 114920897f5cSArne Jansen return ERR_PTR(-ENOMEM); 115020897f5cSArne Jansen 115120897f5cSArne Jansen root->root_key.objectid = objectid; 115220897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 115320897f5cSArne Jansen root->root_key.offset = 0; 115420897f5cSArne Jansen 11559631e4ccSJosef Bacik leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0, 11569631e4ccSJosef Bacik BTRFS_NESTING_NORMAL); 115720897f5cSArne Jansen if (IS_ERR(leaf)) { 115820897f5cSArne Jansen ret = PTR_ERR(leaf); 11591dd05682STsutomu Itoh leaf = NULL; 116020897f5cSArne Jansen goto fail; 116120897f5cSArne Jansen } 116220897f5cSArne Jansen 116320897f5cSArne Jansen root->node = leaf; 116420897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 116520897f5cSArne Jansen 116620897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 116727cdeb70SMiao Xie set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 116820897f5cSArne Jansen 1169*f944d2cbSDavid Sterba btrfs_set_root_flags(&root->root_item, 0); 1170*f944d2cbSDavid Sterba btrfs_set_root_limit(&root->root_item, 0); 117120897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 117220897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 117320897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 117420897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 117520897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 117620897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 117720897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 117833d85fdaSQu Wenruo if (is_fstree(objectid)) 1179807fc790SAndy Shevchenko generate_random_guid(root->root_item.uuid); 1180807fc790SAndy Shevchenko else 1181807fc790SAndy Shevchenko export_guid(root->root_item.uuid, &guid_null); 118220897f5cSArne Jansen root->root_item.drop_level = 0; 118320897f5cSArne Jansen 118420897f5cSArne Jansen key.objectid = objectid; 118520897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 118620897f5cSArne Jansen key.offset = 0; 118720897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 118820897f5cSArne Jansen if (ret) 118920897f5cSArne Jansen goto fail; 119020897f5cSArne Jansen 119120897f5cSArne Jansen btrfs_tree_unlock(leaf); 119220897f5cSArne Jansen 119320897f5cSArne Jansen return root; 11941dd05682STsutomu Itoh 11951dd05682STsutomu Itoh fail: 11968c38938cSJosef Bacik if (leaf) 11971dd05682STsutomu Itoh btrfs_tree_unlock(leaf); 119800246528SJosef Bacik btrfs_put_root(root); 11991dd05682STsutomu Itoh 12001dd05682STsutomu Itoh return ERR_PTR(ret); 120120897f5cSArne Jansen } 120220897f5cSArne Jansen 12037237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1204e02119d5SChris Mason struct btrfs_fs_info *fs_info) 12050f7d52f4SChris Mason { 12060f7d52f4SChris Mason struct btrfs_root *root; 12077237f183SYan Zheng struct extent_buffer *leaf; 1208e02119d5SChris Mason 120996dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS); 1210e02119d5SChris Mason if (!root) 12117237f183SYan Zheng return ERR_PTR(-ENOMEM); 1212e02119d5SChris Mason 1213e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1214e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1215e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 121627cdeb70SMiao Xie 12177237f183SYan Zheng /* 121892a7cc42SQu Wenruo * DON'T set SHAREABLE bit for log trees. 121927cdeb70SMiao Xie * 122092a7cc42SQu Wenruo * Log trees are not exposed to user space thus can't be snapshotted, 122192a7cc42SQu Wenruo * and they go away before a real commit is actually done. 122292a7cc42SQu Wenruo * 122392a7cc42SQu Wenruo * They do store pointers to file data extents, and those reference 122492a7cc42SQu Wenruo * counts still get updated (along with back refs to the log tree). 12257237f183SYan Zheng */ 1226e02119d5SChris Mason 12274d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 12289631e4ccSJosef Bacik NULL, 0, 0, 0, BTRFS_NESTING_NORMAL); 12297237f183SYan Zheng if (IS_ERR(leaf)) { 123000246528SJosef Bacik btrfs_put_root(root); 12317237f183SYan Zheng return ERR_CAST(leaf); 12327237f183SYan Zheng } 1233e02119d5SChris Mason 12347237f183SYan Zheng root->node = leaf; 1235e02119d5SChris Mason 1236e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1237e02119d5SChris Mason btrfs_tree_unlock(root->node); 12387237f183SYan Zheng return root; 12397237f183SYan Zheng } 12407237f183SYan Zheng 12417237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 12427237f183SYan Zheng struct btrfs_fs_info *fs_info) 12437237f183SYan Zheng { 12447237f183SYan Zheng struct btrfs_root *log_root; 12457237f183SYan Zheng 12467237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 12477237f183SYan Zheng if (IS_ERR(log_root)) 12487237f183SYan Zheng return PTR_ERR(log_root); 12497237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 12507237f183SYan Zheng fs_info->log_root_tree = log_root; 12517237f183SYan Zheng return 0; 12527237f183SYan Zheng } 12537237f183SYan Zheng 12547237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 12557237f183SYan Zheng struct btrfs_root *root) 12567237f183SYan Zheng { 12570b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12587237f183SYan Zheng struct btrfs_root *log_root; 12597237f183SYan Zheng struct btrfs_inode_item *inode_item; 12607237f183SYan Zheng 12610b246afaSJeff Mahoney log_root = alloc_log_tree(trans, fs_info); 12627237f183SYan Zheng if (IS_ERR(log_root)) 12637237f183SYan Zheng return PTR_ERR(log_root); 12647237f183SYan Zheng 12657237f183SYan Zheng log_root->last_trans = trans->transid; 12667237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 12677237f183SYan Zheng 12687237f183SYan Zheng inode_item = &log_root->root_item.inode; 12693cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 12703cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 12713cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 1272da17066cSJeff Mahoney btrfs_set_stack_inode_nbytes(inode_item, 12730b246afaSJeff Mahoney fs_info->nodesize); 12743cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 12757237f183SYan Zheng 12765d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 12777237f183SYan Zheng 12787237f183SYan Zheng WARN_ON(root->log_root); 12797237f183SYan Zheng root->log_root = log_root; 12807237f183SYan Zheng root->log_transid = 0; 1281d1433debSMiao Xie root->log_transid_committed = -1; 1282257c62e1SChris Mason root->last_log_commit = 0; 1283e02119d5SChris Mason return 0; 1284e02119d5SChris Mason } 1285e02119d5SChris Mason 128649d11beaSJosef Bacik static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root, 128749d11beaSJosef Bacik struct btrfs_path *path, 1288cb517eabSMiao Xie struct btrfs_key *key) 1289e02119d5SChris Mason { 1290e02119d5SChris Mason struct btrfs_root *root; 1291e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 129284234f3aSYan Zheng u64 generation; 1293cb517eabSMiao Xie int ret; 1294581c1760SQu Wenruo int level; 1295cb517eabSMiao Xie 129696dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS); 129749d11beaSJosef Bacik if (!root) 129849d11beaSJosef Bacik return ERR_PTR(-ENOMEM); 12990f7d52f4SChris Mason 1300cb517eabSMiao Xie ret = btrfs_find_root(tree_root, key, path, 1301cb517eabSMiao Xie &root->root_item, &root->root_key); 13020f7d52f4SChris Mason if (ret) { 130313a8a7c8SYan, Zheng if (ret > 0) 130413a8a7c8SYan, Zheng ret = -ENOENT; 130549d11beaSJosef Bacik goto fail; 13060f7d52f4SChris Mason } 130713a8a7c8SYan, Zheng 130884234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1309581c1760SQu Wenruo level = btrfs_root_level(&root->root_item); 13102ff7e61eSJeff Mahoney root->node = read_tree_block(fs_info, 13112ff7e61eSJeff Mahoney btrfs_root_bytenr(&root->root_item), 1312581c1760SQu Wenruo generation, level, NULL); 131364c043deSLiu Bo if (IS_ERR(root->node)) { 131464c043deSLiu Bo ret = PTR_ERR(root->node); 13158c38938cSJosef Bacik root->node = NULL; 131649d11beaSJosef Bacik goto fail; 1317cb517eabSMiao Xie } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1318cb517eabSMiao Xie ret = -EIO; 131949d11beaSJosef Bacik goto fail; 1320416bc658SJosef Bacik } 13215d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 1322cb517eabSMiao Xie return root; 132349d11beaSJosef Bacik fail: 132400246528SJosef Bacik btrfs_put_root(root); 132549d11beaSJosef Bacik return ERR_PTR(ret); 132649d11beaSJosef Bacik } 132749d11beaSJosef Bacik 132849d11beaSJosef Bacik struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 132949d11beaSJosef Bacik struct btrfs_key *key) 133049d11beaSJosef Bacik { 133149d11beaSJosef Bacik struct btrfs_root *root; 133249d11beaSJosef Bacik struct btrfs_path *path; 133349d11beaSJosef Bacik 133449d11beaSJosef Bacik path = btrfs_alloc_path(); 133549d11beaSJosef Bacik if (!path) 133649d11beaSJosef Bacik return ERR_PTR(-ENOMEM); 133749d11beaSJosef Bacik root = read_tree_root_path(tree_root, path, key); 133849d11beaSJosef Bacik btrfs_free_path(path); 133949d11beaSJosef Bacik 134049d11beaSJosef Bacik return root; 1341cb517eabSMiao Xie } 1342cb517eabSMiao Xie 13432dfb1e43SQu Wenruo /* 13442dfb1e43SQu Wenruo * Initialize subvolume root in-memory structure 13452dfb1e43SQu Wenruo * 13462dfb1e43SQu Wenruo * @anon_dev: anonymous device to attach to the root, if zero, allocate new 13472dfb1e43SQu Wenruo */ 13482dfb1e43SQu Wenruo static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev) 1349cb517eabSMiao Xie { 1350cb517eabSMiao Xie int ret; 1351dcc3eb96SNikolay Borisov unsigned int nofs_flag; 1352cb517eabSMiao Xie 1353cb517eabSMiao Xie root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); 1354cb517eabSMiao Xie root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), 1355cb517eabSMiao Xie GFP_NOFS); 1356cb517eabSMiao Xie if (!root->free_ino_pinned || !root->free_ino_ctl) { 1357cb517eabSMiao Xie ret = -ENOMEM; 1358cb517eabSMiao Xie goto fail; 1359cb517eabSMiao Xie } 1360cb517eabSMiao Xie 1361dcc3eb96SNikolay Borisov /* 1362dcc3eb96SNikolay Borisov * We might be called under a transaction (e.g. indirect backref 1363dcc3eb96SNikolay Borisov * resolution) which could deadlock if it triggers memory reclaim 1364dcc3eb96SNikolay Borisov */ 1365dcc3eb96SNikolay Borisov nofs_flag = memalloc_nofs_save(); 1366dcc3eb96SNikolay Borisov ret = btrfs_drew_lock_init(&root->snapshot_lock); 1367dcc3eb96SNikolay Borisov memalloc_nofs_restore(nofs_flag); 1368dcc3eb96SNikolay Borisov if (ret) 13698257b2dcSMiao Xie goto fail; 13708257b2dcSMiao Xie 1371aeb935a4SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID && 1372aeb935a4SQu Wenruo root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { 137392a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &root->state); 1374f39e4571SJosef Bacik btrfs_check_and_init_root_item(&root->root_item); 1375f39e4571SJosef Bacik } 1376f39e4571SJosef Bacik 1377cb517eabSMiao Xie btrfs_init_free_ino_ctl(root); 137857cdc8dbSDavid Sterba spin_lock_init(&root->ino_cache_lock); 137957cdc8dbSDavid Sterba init_waitqueue_head(&root->ino_cache_wait); 1380cb517eabSMiao Xie 1381851fd730SQu Wenruo /* 1382851fd730SQu Wenruo * Don't assign anonymous block device to roots that are not exposed to 1383851fd730SQu Wenruo * userspace, the id pool is limited to 1M 1384851fd730SQu Wenruo */ 1385851fd730SQu Wenruo if (is_fstree(root->root_key.objectid) && 1386851fd730SQu Wenruo btrfs_root_refs(&root->root_item) > 0) { 13872dfb1e43SQu Wenruo if (!anon_dev) { 1388cb517eabSMiao Xie ret = get_anon_bdev(&root->anon_dev); 1389cb517eabSMiao Xie if (ret) 1390876d2cf1SLiu Bo goto fail; 13912dfb1e43SQu Wenruo } else { 13922dfb1e43SQu Wenruo root->anon_dev = anon_dev; 13932dfb1e43SQu Wenruo } 1394851fd730SQu Wenruo } 1395f32e48e9SChandan Rajendra 1396f32e48e9SChandan Rajendra mutex_lock(&root->objectid_mutex); 1397f32e48e9SChandan Rajendra ret = btrfs_find_highest_objectid(root, 1398f32e48e9SChandan Rajendra &root->highest_objectid); 1399f32e48e9SChandan Rajendra if (ret) { 1400f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1401876d2cf1SLiu Bo goto fail; 1402f32e48e9SChandan Rajendra } 1403f32e48e9SChandan Rajendra 1404f32e48e9SChandan Rajendra ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 1405f32e48e9SChandan Rajendra 1406f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1407f32e48e9SChandan Rajendra 1408cb517eabSMiao Xie return 0; 1409cb517eabSMiao Xie fail: 141084db5ccfSDavid Sterba /* The caller is responsible to call btrfs_free_fs_root */ 1411cb517eabSMiao Xie return ret; 1412cb517eabSMiao Xie } 1413cb517eabSMiao Xie 1414a98db0f3SJosef Bacik static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1415cb517eabSMiao Xie u64 root_id) 1416cb517eabSMiao Xie { 1417cb517eabSMiao Xie struct btrfs_root *root; 1418cb517eabSMiao Xie 1419cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1420cb517eabSMiao Xie root = radix_tree_lookup(&fs_info->fs_roots_radix, 1421cb517eabSMiao Xie (unsigned long)root_id); 1422bc44d7c4SJosef Bacik if (root) 142300246528SJosef Bacik root = btrfs_grab_root(root); 1424cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1425cb517eabSMiao Xie return root; 1426cb517eabSMiao Xie } 1427cb517eabSMiao Xie 142849d11beaSJosef Bacik static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info, 142949d11beaSJosef Bacik u64 objectid) 143049d11beaSJosef Bacik { 143149d11beaSJosef Bacik if (objectid == BTRFS_ROOT_TREE_OBJECTID) 143249d11beaSJosef Bacik return btrfs_grab_root(fs_info->tree_root); 143349d11beaSJosef Bacik if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 143449d11beaSJosef Bacik return btrfs_grab_root(fs_info->extent_root); 143549d11beaSJosef Bacik if (objectid == BTRFS_CHUNK_TREE_OBJECTID) 143649d11beaSJosef Bacik return btrfs_grab_root(fs_info->chunk_root); 143749d11beaSJosef Bacik if (objectid == BTRFS_DEV_TREE_OBJECTID) 143849d11beaSJosef Bacik return btrfs_grab_root(fs_info->dev_root); 143949d11beaSJosef Bacik if (objectid == BTRFS_CSUM_TREE_OBJECTID) 144049d11beaSJosef Bacik return btrfs_grab_root(fs_info->csum_root); 144149d11beaSJosef Bacik if (objectid == BTRFS_QUOTA_TREE_OBJECTID) 144249d11beaSJosef Bacik return btrfs_grab_root(fs_info->quota_root) ? 144349d11beaSJosef Bacik fs_info->quota_root : ERR_PTR(-ENOENT); 144449d11beaSJosef Bacik if (objectid == BTRFS_UUID_TREE_OBJECTID) 144549d11beaSJosef Bacik return btrfs_grab_root(fs_info->uuid_root) ? 144649d11beaSJosef Bacik fs_info->uuid_root : ERR_PTR(-ENOENT); 144749d11beaSJosef Bacik if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) 144849d11beaSJosef Bacik return btrfs_grab_root(fs_info->free_space_root) ? 144949d11beaSJosef Bacik fs_info->free_space_root : ERR_PTR(-ENOENT); 145049d11beaSJosef Bacik return NULL; 145149d11beaSJosef Bacik } 145249d11beaSJosef Bacik 1453cb517eabSMiao Xie int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1454cb517eabSMiao Xie struct btrfs_root *root) 1455cb517eabSMiao Xie { 1456cb517eabSMiao Xie int ret; 1457cb517eabSMiao Xie 1458e1860a77SDavid Sterba ret = radix_tree_preload(GFP_NOFS); 1459cb517eabSMiao Xie if (ret) 1460cb517eabSMiao Xie return ret; 1461cb517eabSMiao Xie 1462cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1463cb517eabSMiao Xie ret = radix_tree_insert(&fs_info->fs_roots_radix, 1464cb517eabSMiao Xie (unsigned long)root->root_key.objectid, 1465cb517eabSMiao Xie root); 1466af01d2e5SJosef Bacik if (ret == 0) { 146700246528SJosef Bacik btrfs_grab_root(root); 146827cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1469af01d2e5SJosef Bacik } 1470cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1471cb517eabSMiao Xie radix_tree_preload_end(); 1472cb517eabSMiao Xie 1473cb517eabSMiao Xie return ret; 1474cb517eabSMiao Xie } 1475cb517eabSMiao Xie 1476bd647ce3SJosef Bacik void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info) 1477bd647ce3SJosef Bacik { 1478bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 1479bd647ce3SJosef Bacik struct btrfs_root *root; 1480bd647ce3SJosef Bacik 1481bd647ce3SJosef Bacik while (!list_empty(&fs_info->allocated_roots)) { 1482457f1864SJosef Bacik char buf[BTRFS_ROOT_NAME_BUF_LEN]; 1483457f1864SJosef Bacik 1484bd647ce3SJosef Bacik root = list_first_entry(&fs_info->allocated_roots, 1485bd647ce3SJosef Bacik struct btrfs_root, leak_list); 1486457f1864SJosef Bacik btrfs_err(fs_info, "leaked root %s refcount %d", 1487457f1864SJosef Bacik btrfs_root_name(root->root_key.objectid, buf), 1488bd647ce3SJosef Bacik refcount_read(&root->refs)); 1489bd647ce3SJosef Bacik while (refcount_read(&root->refs) > 1) 149000246528SJosef Bacik btrfs_put_root(root); 149100246528SJosef Bacik btrfs_put_root(root); 1492bd647ce3SJosef Bacik } 1493bd647ce3SJosef Bacik #endif 1494bd647ce3SJosef Bacik } 1495bd647ce3SJosef Bacik 14960d4b0463SJosef Bacik void btrfs_free_fs_info(struct btrfs_fs_info *fs_info) 14970d4b0463SJosef Bacik { 1498141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 1499141386e1SJosef Bacik percpu_counter_destroy(&fs_info->delalloc_bytes); 1500141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dio_bytes); 1501141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 1502141386e1SJosef Bacik btrfs_free_csum_hash(fs_info); 1503141386e1SJosef Bacik btrfs_free_stripe_hash_table(fs_info); 1504141386e1SJosef Bacik btrfs_free_ref_cache(fs_info); 15050d4b0463SJosef Bacik kfree(fs_info->balance_ctl); 15060d4b0463SJosef Bacik kfree(fs_info->delayed_root); 150700246528SJosef Bacik btrfs_put_root(fs_info->extent_root); 150800246528SJosef Bacik btrfs_put_root(fs_info->tree_root); 150900246528SJosef Bacik btrfs_put_root(fs_info->chunk_root); 151000246528SJosef Bacik btrfs_put_root(fs_info->dev_root); 151100246528SJosef Bacik btrfs_put_root(fs_info->csum_root); 151200246528SJosef Bacik btrfs_put_root(fs_info->quota_root); 151300246528SJosef Bacik btrfs_put_root(fs_info->uuid_root); 151400246528SJosef Bacik btrfs_put_root(fs_info->free_space_root); 151500246528SJosef Bacik btrfs_put_root(fs_info->fs_root); 1516aeb935a4SQu Wenruo btrfs_put_root(fs_info->data_reloc_root); 1517bd647ce3SJosef Bacik btrfs_check_leaked_roots(fs_info); 15183fd63727SJosef Bacik btrfs_extent_buffer_leak_debug_check(fs_info); 15190d4b0463SJosef Bacik kfree(fs_info->super_copy); 15200d4b0463SJosef Bacik kfree(fs_info->super_for_commit); 15210d4b0463SJosef Bacik kvfree(fs_info); 15220d4b0463SJosef Bacik } 15230d4b0463SJosef Bacik 15240d4b0463SJosef Bacik 15252dfb1e43SQu Wenruo /* 15262dfb1e43SQu Wenruo * Get an in-memory reference of a root structure. 15272dfb1e43SQu Wenruo * 15282dfb1e43SQu Wenruo * For essential trees like root/extent tree, we grab it from fs_info directly. 15292dfb1e43SQu Wenruo * For subvolume trees, we check the cached filesystem roots first. If not 15302dfb1e43SQu Wenruo * found, then read it from disk and add it to cached fs roots. 15312dfb1e43SQu Wenruo * 15322dfb1e43SQu Wenruo * Caller should release the root by calling btrfs_put_root() after the usage. 15332dfb1e43SQu Wenruo * 15342dfb1e43SQu Wenruo * NOTE: Reloc and log trees can't be read by this function as they share the 15352dfb1e43SQu Wenruo * same root objectid. 15362dfb1e43SQu Wenruo * 15372dfb1e43SQu Wenruo * @objectid: root id 15382dfb1e43SQu Wenruo * @anon_dev: preallocated anonymous block device number for new roots, 15392dfb1e43SQu Wenruo * pass 0 for new allocation. 15402dfb1e43SQu Wenruo * @check_ref: whether to check root item references, If true, return -ENOENT 15412dfb1e43SQu Wenruo * for orphan roots 15422dfb1e43SQu Wenruo */ 15432dfb1e43SQu Wenruo static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info, 15442dfb1e43SQu Wenruo u64 objectid, dev_t anon_dev, 15452dfb1e43SQu Wenruo bool check_ref) 15465eda7b5eSChris Mason { 15475eda7b5eSChris Mason struct btrfs_root *root; 1548381cf658SDavid Sterba struct btrfs_path *path; 15491d4c08e0SDavid Sterba struct btrfs_key key; 15505eda7b5eSChris Mason int ret; 15515eda7b5eSChris Mason 155249d11beaSJosef Bacik root = btrfs_get_global_root(fs_info, objectid); 155349d11beaSJosef Bacik if (root) 155449d11beaSJosef Bacik return root; 15554df27c4dSYan, Zheng again: 155656e9357aSDavid Sterba root = btrfs_lookup_fs_root(fs_info, objectid); 155748475471SStefan Behrens if (root) { 15582dfb1e43SQu Wenruo /* Shouldn't get preallocated anon_dev for cached roots */ 15592dfb1e43SQu Wenruo ASSERT(!anon_dev); 1560bc44d7c4SJosef Bacik if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 156100246528SJosef Bacik btrfs_put_root(root); 156248475471SStefan Behrens return ERR_PTR(-ENOENT); 1563bc44d7c4SJosef Bacik } 15645eda7b5eSChris Mason return root; 156548475471SStefan Behrens } 15665eda7b5eSChris Mason 156756e9357aSDavid Sterba key.objectid = objectid; 156856e9357aSDavid Sterba key.type = BTRFS_ROOT_ITEM_KEY; 156956e9357aSDavid Sterba key.offset = (u64)-1; 157056e9357aSDavid Sterba root = btrfs_read_tree_root(fs_info->tree_root, &key); 15715eda7b5eSChris Mason if (IS_ERR(root)) 15725eda7b5eSChris Mason return root; 15733394e160SChris Mason 1574c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1575d68fc57bSYan, Zheng ret = -ENOENT; 1576d68fc57bSYan, Zheng goto fail; 1577d68fc57bSYan, Zheng } 1578d68fc57bSYan, Zheng 15792dfb1e43SQu Wenruo ret = btrfs_init_fs_root(root, anon_dev); 1580cb517eabSMiao Xie if (ret) 1581cb517eabSMiao Xie goto fail; 1582cb517eabSMiao Xie 1583381cf658SDavid Sterba path = btrfs_alloc_path(); 1584381cf658SDavid Sterba if (!path) { 1585381cf658SDavid Sterba ret = -ENOMEM; 1586381cf658SDavid Sterba goto fail; 1587381cf658SDavid Sterba } 15881d4c08e0SDavid Sterba key.objectid = BTRFS_ORPHAN_OBJECTID; 15891d4c08e0SDavid Sterba key.type = BTRFS_ORPHAN_ITEM_KEY; 159056e9357aSDavid Sterba key.offset = objectid; 15911d4c08e0SDavid Sterba 15921d4c08e0SDavid Sterba ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1593381cf658SDavid Sterba btrfs_free_path(path); 1594d68fc57bSYan, Zheng if (ret < 0) 1595d68fc57bSYan, Zheng goto fail; 1596d68fc57bSYan, Zheng if (ret == 0) 159727cdeb70SMiao Xie set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1598d68fc57bSYan, Zheng 1599cb517eabSMiao Xie ret = btrfs_insert_fs_root(fs_info, root); 16000f7d52f4SChris Mason if (ret) { 160100246528SJosef Bacik btrfs_put_root(root); 16024785e24fSJosef Bacik if (ret == -EEXIST) 16034df27c4dSYan, Zheng goto again; 16044df27c4dSYan, Zheng goto fail; 16054df27c4dSYan, Zheng } 1606edbd8d4eSChris Mason return root; 16074df27c4dSYan, Zheng fail: 16088c38938cSJosef Bacik btrfs_put_root(root); 16094df27c4dSYan, Zheng return ERR_PTR(ret); 1610edbd8d4eSChris Mason } 1611edbd8d4eSChris Mason 16122dfb1e43SQu Wenruo /* 16132dfb1e43SQu Wenruo * Get in-memory reference of a root structure 16142dfb1e43SQu Wenruo * 16152dfb1e43SQu Wenruo * @objectid: tree objectid 16162dfb1e43SQu Wenruo * @check_ref: if set, verify that the tree exists and the item has at least 16172dfb1e43SQu Wenruo * one reference 16182dfb1e43SQu Wenruo */ 16192dfb1e43SQu Wenruo struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 16202dfb1e43SQu Wenruo u64 objectid, bool check_ref) 16212dfb1e43SQu Wenruo { 16222dfb1e43SQu Wenruo return btrfs_get_root_ref(fs_info, objectid, 0, check_ref); 16232dfb1e43SQu Wenruo } 16242dfb1e43SQu Wenruo 16252dfb1e43SQu Wenruo /* 16262dfb1e43SQu Wenruo * Get in-memory reference of a root structure, created as new, optionally pass 16272dfb1e43SQu Wenruo * the anonymous block device id 16282dfb1e43SQu Wenruo * 16292dfb1e43SQu Wenruo * @objectid: tree objectid 16302dfb1e43SQu Wenruo * @anon_dev: if zero, allocate a new anonymous block device or use the 16312dfb1e43SQu Wenruo * parameter value 16322dfb1e43SQu Wenruo */ 16332dfb1e43SQu Wenruo struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info, 16342dfb1e43SQu Wenruo u64 objectid, dev_t anon_dev) 16352dfb1e43SQu Wenruo { 16362dfb1e43SQu Wenruo return btrfs_get_root_ref(fs_info, objectid, anon_dev, true); 16372dfb1e43SQu Wenruo } 16382dfb1e43SQu Wenruo 16398b712842SChris Mason /* 164049d11beaSJosef Bacik * btrfs_get_fs_root_commit_root - return a root for the given objectid 164149d11beaSJosef Bacik * @fs_info: the fs_info 164249d11beaSJosef Bacik * @objectid: the objectid we need to lookup 164349d11beaSJosef Bacik * 164449d11beaSJosef Bacik * This is exclusively used for backref walking, and exists specifically because 164549d11beaSJosef Bacik * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref 164649d11beaSJosef Bacik * creation time, which means we may have to read the tree_root in order to look 164749d11beaSJosef Bacik * up a fs root that is not in memory. If the root is not in memory we will 164849d11beaSJosef Bacik * read the tree root commit root and look up the fs root from there. This is a 164949d11beaSJosef Bacik * temporary root, it will not be inserted into the radix tree as it doesn't 165049d11beaSJosef Bacik * have the most uptodate information, it'll simply be discarded once the 165149d11beaSJosef Bacik * backref code is finished using the root. 165249d11beaSJosef Bacik */ 165349d11beaSJosef Bacik struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info, 165449d11beaSJosef Bacik struct btrfs_path *path, 165549d11beaSJosef Bacik u64 objectid) 165649d11beaSJosef Bacik { 165749d11beaSJosef Bacik struct btrfs_root *root; 165849d11beaSJosef Bacik struct btrfs_key key; 165949d11beaSJosef Bacik 166049d11beaSJosef Bacik ASSERT(path->search_commit_root && path->skip_locking); 166149d11beaSJosef Bacik 166249d11beaSJosef Bacik /* 166349d11beaSJosef Bacik * This can return -ENOENT if we ask for a root that doesn't exist, but 166449d11beaSJosef Bacik * since this is called via the backref walking code we won't be looking 166549d11beaSJosef Bacik * up a root that doesn't exist, unless there's corruption. So if root 166649d11beaSJosef Bacik * != NULL just return it. 166749d11beaSJosef Bacik */ 166849d11beaSJosef Bacik root = btrfs_get_global_root(fs_info, objectid); 166949d11beaSJosef Bacik if (root) 167049d11beaSJosef Bacik return root; 167149d11beaSJosef Bacik 167249d11beaSJosef Bacik root = btrfs_lookup_fs_root(fs_info, objectid); 167349d11beaSJosef Bacik if (root) 167449d11beaSJosef Bacik return root; 167549d11beaSJosef Bacik 167649d11beaSJosef Bacik key.objectid = objectid; 167749d11beaSJosef Bacik key.type = BTRFS_ROOT_ITEM_KEY; 167849d11beaSJosef Bacik key.offset = (u64)-1; 167949d11beaSJosef Bacik root = read_tree_root_path(fs_info->tree_root, path, &key); 168049d11beaSJosef Bacik btrfs_release_path(path); 168149d11beaSJosef Bacik 168249d11beaSJosef Bacik return root; 168349d11beaSJosef Bacik } 168449d11beaSJosef Bacik 168549d11beaSJosef Bacik /* 16868b712842SChris Mason * called by the kthread helper functions to finally call the bio end_io 16878b712842SChris Mason * functions. This is where read checksum verification actually happens 16888b712842SChris Mason */ 16898b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work) 1690ce9adaa5SChris Mason { 1691ce9adaa5SChris Mason struct bio *bio; 169297eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 1693ce9adaa5SChris Mason 169497eb6b69SDavid Sterba end_io_wq = container_of(work, struct btrfs_end_io_wq, work); 1695ce9adaa5SChris Mason bio = end_io_wq->bio; 16968b712842SChris Mason 16974e4cbee9SChristoph Hellwig bio->bi_status = end_io_wq->status; 1698ce9adaa5SChris Mason bio->bi_private = end_io_wq->private; 1699ce9adaa5SChris Mason bio->bi_end_io = end_io_wq->end_io; 17004246a0b6SChristoph Hellwig bio_endio(bio); 17019be490f1SOmar Sandoval kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq); 1702ce9adaa5SChris Mason } 170344b8bd7eSChris Mason 1704a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1705a74a4b97SChris Mason { 1706a74a4b97SChris Mason struct btrfs_root *root = arg; 17070b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1708d0278245SMiao Xie int again; 1709a74a4b97SChris Mason 1710d6fd0ae2SOmar Sandoval while (1) { 1711d0278245SMiao Xie again = 0; 17129d1a2a3aSDavid Sterba 1713fd340d0fSJosef Bacik set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1714fd340d0fSJosef Bacik 1715d0278245SMiao Xie /* Make the cleaner go to sleep early. */ 17162ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) 1717d0278245SMiao Xie goto sleep; 1718d0278245SMiao Xie 171990c711abSZygo Blaxell /* 172090c711abSZygo Blaxell * Do not do anything if we might cause open_ctree() to block 172190c711abSZygo Blaxell * before we have finished mounting the filesystem. 172290c711abSZygo Blaxell */ 17230b246afaSJeff Mahoney if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 172490c711abSZygo Blaxell goto sleep; 172590c711abSZygo Blaxell 17260b246afaSJeff Mahoney if (!mutex_trylock(&fs_info->cleaner_mutex)) 1727d0278245SMiao Xie goto sleep; 1728d0278245SMiao Xie 1729dc7f370cSMiao Xie /* 1730dc7f370cSMiao Xie * Avoid the problem that we change the status of the fs 1731dc7f370cSMiao Xie * during the above check and trylock. 1732dc7f370cSMiao Xie */ 17332ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) { 17340b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1735dc7f370cSMiao Xie goto sleep; 1736dc7f370cSMiao Xie } 1737dc7f370cSMiao Xie 17382ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 1739c2d6cb16SFilipe Manana 17409d1a2a3aSDavid Sterba again = btrfs_clean_one_deleted_snapshot(root); 17410b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1742a74a4b97SChris Mason 1743d0278245SMiao Xie /* 174405323cd1SMiao Xie * The defragger has dealt with the R/O remount and umount, 174505323cd1SMiao Xie * needn't do anything special here. 1746d0278245SMiao Xie */ 17470b246afaSJeff Mahoney btrfs_run_defrag_inodes(fs_info); 174867c5e7d4SFilipe Manana 174967c5e7d4SFilipe Manana /* 175067c5e7d4SFilipe Manana * Acquires fs_info->delete_unused_bgs_mutex to avoid racing 175167c5e7d4SFilipe Manana * with relocation (btrfs_relocate_chunk) and relocation 175267c5e7d4SFilipe Manana * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 175367c5e7d4SFilipe Manana * after acquiring fs_info->delete_unused_bgs_mutex. So we 175467c5e7d4SFilipe Manana * can't hold, nor need to, fs_info->cleaner_mutex when deleting 175567c5e7d4SFilipe Manana * unused block groups. 175667c5e7d4SFilipe Manana */ 17570b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 1758d0278245SMiao Xie sleep: 1759fd340d0fSJosef Bacik clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1760d6fd0ae2SOmar Sandoval if (kthread_should_park()) 1761d6fd0ae2SOmar Sandoval kthread_parkme(); 1762d6fd0ae2SOmar Sandoval if (kthread_should_stop()) 1763d6fd0ae2SOmar Sandoval return 0; 1764838fe188SJiri Kosina if (!again) { 1765a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 1766a74a4b97SChris Mason schedule(); 1767a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1768a74a4b97SChris Mason } 1769da288d28SFilipe Manana } 1770a74a4b97SChris Mason } 1771a74a4b97SChris Mason 1772a74a4b97SChris Mason static int transaction_kthread(void *arg) 1773a74a4b97SChris Mason { 1774a74a4b97SChris Mason struct btrfs_root *root = arg; 17750b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1776a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1777a74a4b97SChris Mason struct btrfs_transaction *cur; 17788929ecfaSYan, Zheng u64 transid; 1779643900beSNikolay Borisov time64_t delta; 1780a74a4b97SChris Mason unsigned long delay; 1781914b2007SJan Kara bool cannot_commit; 1782a74a4b97SChris Mason 1783a74a4b97SChris Mason do { 1784914b2007SJan Kara cannot_commit = false; 1785ba1bc00fSNikolay Borisov delay = msecs_to_jiffies(fs_info->commit_interval * 1000); 17860b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 1787a74a4b97SChris Mason 17880b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 17890b246afaSJeff Mahoney cur = fs_info->running_transaction; 1790a74a4b97SChris Mason if (!cur) { 17910b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1792a74a4b97SChris Mason goto sleep; 1793a74a4b97SChris Mason } 179431153d81SYan Zheng 1795643900beSNikolay Borisov delta = ktime_get_seconds() - cur->start_time; 17963296bf56SQu Wenruo if (cur->state < TRANS_STATE_COMMIT_START && 1797643900beSNikolay Borisov delta < fs_info->commit_interval) { 17980b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1799fb8a7e94SNikolay Borisov delay -= msecs_to_jiffies((delta - 1) * 1000); 1800fb8a7e94SNikolay Borisov delay = min(delay, 1801fb8a7e94SNikolay Borisov msecs_to_jiffies(fs_info->commit_interval * 1000)); 1802a74a4b97SChris Mason goto sleep; 1803a74a4b97SChris Mason } 18048929ecfaSYan, Zheng transid = cur->transid; 18050b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 180656bec294SChris Mason 180779787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 1808354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 1809914b2007SJan Kara if (IS_ERR(trans)) { 1810354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 1811914b2007SJan Kara cannot_commit = true; 181279787eaaSJeff Mahoney goto sleep; 1813914b2007SJan Kara } 18148929ecfaSYan, Zheng if (transid == trans->transid) { 18153a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 18168929ecfaSYan, Zheng } else { 18173a45bb20SJeff Mahoney btrfs_end_transaction(trans); 18188929ecfaSYan, Zheng } 1819a74a4b97SChris Mason sleep: 18200b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 18210b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 1822a74a4b97SChris Mason 18234e121c06SJosef Bacik if (unlikely(test_bit(BTRFS_FS_STATE_ERROR, 18240b246afaSJeff Mahoney &fs_info->fs_state))) 18252ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 18268929ecfaSYan, Zheng if (!kthread_should_stop() && 18270b246afaSJeff Mahoney (!btrfs_transaction_blocked(fs_info) || 1828914b2007SJan Kara cannot_commit)) 1829bc5511d0SNikolay Borisov schedule_timeout_interruptible(delay); 1830a74a4b97SChris Mason } while (!kthread_should_stop()); 1831a74a4b97SChris Mason return 0; 1832a74a4b97SChris Mason } 1833a74a4b97SChris Mason 1834af31f5e5SChris Mason /* 183501f0f9daSNikolay Borisov * This will find the highest generation in the array of root backups. The 183601f0f9daSNikolay Borisov * index of the highest array is returned, or -EINVAL if we can't find 183701f0f9daSNikolay Borisov * anything. 1838af31f5e5SChris Mason * 1839af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1840af31f5e5SChris Mason * generation of the latest root in the array with the generation 1841af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1842af31f5e5SChris Mason */ 184301f0f9daSNikolay Borisov static int find_newest_super_backup(struct btrfs_fs_info *info) 1844af31f5e5SChris Mason { 184501f0f9daSNikolay Borisov const u64 newest_gen = btrfs_super_generation(info->super_copy); 1846af31f5e5SChris Mason u64 cur; 1847af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1848af31f5e5SChris Mason int i; 1849af31f5e5SChris Mason 1850af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1851af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1852af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1853af31f5e5SChris Mason if (cur == newest_gen) 185401f0f9daSNikolay Borisov return i; 1855af31f5e5SChris Mason } 1856af31f5e5SChris Mason 185701f0f9daSNikolay Borisov return -EINVAL; 1858af31f5e5SChris Mason } 1859af31f5e5SChris Mason 1860af31f5e5SChris Mason /* 1861af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1862af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1863af31f5e5SChris Mason * done 1864af31f5e5SChris Mason */ 1865af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1866af31f5e5SChris Mason { 18676ef108ddSNikolay Borisov const int next_backup = info->backup_root_index; 1868af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1869af31f5e5SChris Mason 1870af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1871af31f5e5SChris Mason 1872af31f5e5SChris Mason /* 1873af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1874af31f5e5SChris Mason * regardless of which ones we use today 1875af31f5e5SChris Mason */ 1876af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1877af31f5e5SChris Mason 1878af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1879af31f5e5SChris Mason 1880af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1881af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1882af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1883af31f5e5SChris Mason 1884af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1885af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1886af31f5e5SChris Mason 1887af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1888af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1889af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 1890af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 1891af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 1892af31f5e5SChris Mason 1893af31f5e5SChris Mason btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); 1894af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 1895af31f5e5SChris Mason btrfs_header_generation(info->extent_root->node)); 1896af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 1897af31f5e5SChris Mason btrfs_header_level(info->extent_root->node)); 1898af31f5e5SChris Mason 18997c7e82a7SChris Mason /* 19007c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 19017c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 19027c7e82a7SChris Mason */ 19037c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 19047c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 19057c7e82a7SChris Mason info->fs_root->node->start); 1906af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 1907af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 1908af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 1909af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 19107c7e82a7SChris Mason } 1911af31f5e5SChris Mason 1912af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 1913af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 1914af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 1915af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 1916af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 1917af31f5e5SChris Mason 1918af31f5e5SChris Mason btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); 1919af31f5e5SChris Mason btrfs_set_backup_csum_root_gen(root_backup, 1920af31f5e5SChris Mason btrfs_header_generation(info->csum_root->node)); 1921af31f5e5SChris Mason btrfs_set_backup_csum_root_level(root_backup, 1922af31f5e5SChris Mason btrfs_header_level(info->csum_root->node)); 1923af31f5e5SChris Mason 1924af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 1925af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 1926af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 1927af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 1928af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 1929af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 1930af31f5e5SChris Mason 1931af31f5e5SChris Mason /* 1932af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 1933af31f5e5SChris Mason * for the next commit 1934af31f5e5SChris Mason */ 1935af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 1936af31f5e5SChris Mason &info->super_for_commit->super_roots, 1937af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 1938af31f5e5SChris Mason } 1939af31f5e5SChris Mason 1940af31f5e5SChris Mason /* 1941bd2336b2SNikolay Borisov * read_backup_root - Reads a backup root based on the passed priority. Prio 0 1942bd2336b2SNikolay Borisov * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots 1943bd2336b2SNikolay Borisov * 1944bd2336b2SNikolay Borisov * fs_info - filesystem whose backup roots need to be read 1945bd2336b2SNikolay Borisov * priority - priority of backup root required 1946bd2336b2SNikolay Borisov * 1947bd2336b2SNikolay Borisov * Returns backup root index on success and -EINVAL otherwise. 1948bd2336b2SNikolay Borisov */ 1949bd2336b2SNikolay Borisov static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority) 1950bd2336b2SNikolay Borisov { 1951bd2336b2SNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 1952bd2336b2SNikolay Borisov struct btrfs_super_block *super = fs_info->super_copy; 1953bd2336b2SNikolay Borisov struct btrfs_root_backup *root_backup; 1954bd2336b2SNikolay Borisov 1955bd2336b2SNikolay Borisov if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) { 1956bd2336b2SNikolay Borisov if (priority == 0) 1957bd2336b2SNikolay Borisov return backup_index; 1958bd2336b2SNikolay Borisov 1959bd2336b2SNikolay Borisov backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority; 1960bd2336b2SNikolay Borisov backup_index %= BTRFS_NUM_BACKUP_ROOTS; 1961bd2336b2SNikolay Borisov } else { 1962bd2336b2SNikolay Borisov return -EINVAL; 1963bd2336b2SNikolay Borisov } 1964bd2336b2SNikolay Borisov 1965bd2336b2SNikolay Borisov root_backup = super->super_roots + backup_index; 1966bd2336b2SNikolay Borisov 1967bd2336b2SNikolay Borisov btrfs_set_super_generation(super, 1968bd2336b2SNikolay Borisov btrfs_backup_tree_root_gen(root_backup)); 1969bd2336b2SNikolay Borisov btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 1970bd2336b2SNikolay Borisov btrfs_set_super_root_level(super, 1971bd2336b2SNikolay Borisov btrfs_backup_tree_root_level(root_backup)); 1972bd2336b2SNikolay Borisov btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 1973bd2336b2SNikolay Borisov 1974bd2336b2SNikolay Borisov /* 1975bd2336b2SNikolay Borisov * Fixme: the total bytes and num_devices need to match or we should 1976bd2336b2SNikolay Borisov * need a fsck 1977bd2336b2SNikolay Borisov */ 1978bd2336b2SNikolay Borisov btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 1979bd2336b2SNikolay Borisov btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 1980bd2336b2SNikolay Borisov 1981bd2336b2SNikolay Borisov return backup_index; 1982bd2336b2SNikolay Borisov } 1983bd2336b2SNikolay Borisov 19847abadb64SLiu Bo /* helper to cleanup workers */ 19857abadb64SLiu Bo static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 19867abadb64SLiu Bo { 1987dc6e3209SQu Wenruo btrfs_destroy_workqueue(fs_info->fixup_workers); 1988afe3d242SQu Wenruo btrfs_destroy_workqueue(fs_info->delalloc_workers); 19895cdc7ad3SQu Wenruo btrfs_destroy_workqueue(fs_info->workers); 1990fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_workers); 1991fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_raid56_workers); 1992d05a33acSQu Wenruo btrfs_destroy_workqueue(fs_info->rmw_workers); 1993fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_write_workers); 1994fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 19955b3bc44eSQu Wenruo btrfs_destroy_workqueue(fs_info->delayed_workers); 1996e66f0bb1SQu Wenruo btrfs_destroy_workqueue(fs_info->caching_workers); 1997736cfa15SQu Wenruo btrfs_destroy_workqueue(fs_info->readahead_workers); 1998a44903abSQu Wenruo btrfs_destroy_workqueue(fs_info->flush_workers); 1999fc97fab0SQu Wenruo btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2000b0643e59SDennis Zhou if (fs_info->discard_ctl.discard_workers) 2001b0643e59SDennis Zhou destroy_workqueue(fs_info->discard_ctl.discard_workers); 2002a9b9477dSFilipe Manana /* 2003a9b9477dSFilipe Manana * Now that all other work queues are destroyed, we can safely destroy 2004a9b9477dSFilipe Manana * the queues used for metadata I/O, since tasks from those other work 2005a9b9477dSFilipe Manana * queues can do metadata I/O operations. 2006a9b9477dSFilipe Manana */ 2007a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_workers); 2008a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_write_workers); 20097abadb64SLiu Bo } 20107abadb64SLiu Bo 20112e9f5954SRashika static void free_root_extent_buffers(struct btrfs_root *root) 20122e9f5954SRashika { 20132e9f5954SRashika if (root) { 20142e9f5954SRashika free_extent_buffer(root->node); 20152e9f5954SRashika free_extent_buffer(root->commit_root); 20162e9f5954SRashika root->node = NULL; 20172e9f5954SRashika root->commit_root = NULL; 20182e9f5954SRashika } 20192e9f5954SRashika } 20202e9f5954SRashika 2021af31f5e5SChris Mason /* helper to cleanup tree roots */ 20224273eaffSAnand Jain static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root) 2023af31f5e5SChris Mason { 20242e9f5954SRashika free_root_extent_buffers(info->tree_root); 2025655b09feSJosef Bacik 20262e9f5954SRashika free_root_extent_buffers(info->dev_root); 20272e9f5954SRashika free_root_extent_buffers(info->extent_root); 20282e9f5954SRashika free_root_extent_buffers(info->csum_root); 20292e9f5954SRashika free_root_extent_buffers(info->quota_root); 20302e9f5954SRashika free_root_extent_buffers(info->uuid_root); 20318c38938cSJosef Bacik free_root_extent_buffers(info->fs_root); 2032aeb935a4SQu Wenruo free_root_extent_buffers(info->data_reloc_root); 20334273eaffSAnand Jain if (free_chunk_root) 20342e9f5954SRashika free_root_extent_buffers(info->chunk_root); 203570f6d82eSOmar Sandoval free_root_extent_buffers(info->free_space_root); 2036af31f5e5SChris Mason } 2037af31f5e5SChris Mason 20388c38938cSJosef Bacik void btrfs_put_root(struct btrfs_root *root) 20398c38938cSJosef Bacik { 20408c38938cSJosef Bacik if (!root) 20418c38938cSJosef Bacik return; 20428c38938cSJosef Bacik 20438c38938cSJosef Bacik if (refcount_dec_and_test(&root->refs)) { 20448c38938cSJosef Bacik WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 20451dae7e0eSQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)); 20468c38938cSJosef Bacik if (root->anon_dev) 20478c38938cSJosef Bacik free_anon_bdev(root->anon_dev); 20488c38938cSJosef Bacik btrfs_drew_lock_destroy(&root->snapshot_lock); 2049923eb523SJohannes Thumshirn free_root_extent_buffers(root); 20508c38938cSJosef Bacik kfree(root->free_ino_ctl); 20518c38938cSJosef Bacik kfree(root->free_ino_pinned); 20528c38938cSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 20538c38938cSJosef Bacik spin_lock(&root->fs_info->fs_roots_radix_lock); 20548c38938cSJosef Bacik list_del_init(&root->leak_list); 20558c38938cSJosef Bacik spin_unlock(&root->fs_info->fs_roots_radix_lock); 20568c38938cSJosef Bacik #endif 20578c38938cSJosef Bacik kfree(root); 20588c38938cSJosef Bacik } 20598c38938cSJosef Bacik } 20608c38938cSJosef Bacik 2061faa2dbf0SJosef Bacik void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2062171f6537SJosef Bacik { 2063171f6537SJosef Bacik int ret; 2064171f6537SJosef Bacik struct btrfs_root *gang[8]; 2065171f6537SJosef Bacik int i; 2066171f6537SJosef Bacik 2067171f6537SJosef Bacik while (!list_empty(&fs_info->dead_roots)) { 2068171f6537SJosef Bacik gang[0] = list_entry(fs_info->dead_roots.next, 2069171f6537SJosef Bacik struct btrfs_root, root_list); 2070171f6537SJosef Bacik list_del(&gang[0]->root_list); 2071171f6537SJosef Bacik 20728c38938cSJosef Bacik if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) 2073cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[0]); 207400246528SJosef Bacik btrfs_put_root(gang[0]); 2075171f6537SJosef Bacik } 2076171f6537SJosef Bacik 2077171f6537SJosef Bacik while (1) { 2078171f6537SJosef Bacik ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2079171f6537SJosef Bacik (void **)gang, 0, 2080171f6537SJosef Bacik ARRAY_SIZE(gang)); 2081171f6537SJosef Bacik if (!ret) 2082171f6537SJosef Bacik break; 2083171f6537SJosef Bacik for (i = 0; i < ret; i++) 2084cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2085171f6537SJosef Bacik } 2086171f6537SJosef Bacik } 2087af31f5e5SChris Mason 2088638aa7edSEric Sandeen static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2089638aa7edSEric Sandeen { 2090638aa7edSEric Sandeen mutex_init(&fs_info->scrub_lock); 2091638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_running, 0); 2092638aa7edSEric Sandeen atomic_set(&fs_info->scrub_pause_req, 0); 2093638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_paused, 0); 2094638aa7edSEric Sandeen atomic_set(&fs_info->scrub_cancel_req, 0); 2095638aa7edSEric Sandeen init_waitqueue_head(&fs_info->scrub_pause_wait); 2096ff09c4caSAnand Jain refcount_set(&fs_info->scrub_workers_refcnt, 0); 2097638aa7edSEric Sandeen } 2098638aa7edSEric Sandeen 2099779a65a4SEric Sandeen static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2100779a65a4SEric Sandeen { 2101779a65a4SEric Sandeen spin_lock_init(&fs_info->balance_lock); 2102779a65a4SEric Sandeen mutex_init(&fs_info->balance_mutex); 2103779a65a4SEric Sandeen atomic_set(&fs_info->balance_pause_req, 0); 2104779a65a4SEric Sandeen atomic_set(&fs_info->balance_cancel_req, 0); 2105779a65a4SEric Sandeen fs_info->balance_ctl = NULL; 2106779a65a4SEric Sandeen init_waitqueue_head(&fs_info->balance_wait_q); 2107779a65a4SEric Sandeen } 2108779a65a4SEric Sandeen 21096bccf3abSJeff Mahoney static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info) 2110f37938e0SEric Sandeen { 21112ff7e61eSJeff Mahoney struct inode *inode = fs_info->btree_inode; 21122ff7e61eSJeff Mahoney 21132ff7e61eSJeff Mahoney inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 21142ff7e61eSJeff Mahoney set_nlink(inode, 1); 2115f37938e0SEric Sandeen /* 2116f37938e0SEric Sandeen * we set the i_size on the btree inode to the max possible int. 2117f37938e0SEric Sandeen * the real end of the address space is determined by all of 2118f37938e0SEric Sandeen * the devices in the system 2119f37938e0SEric Sandeen */ 21202ff7e61eSJeff Mahoney inode->i_size = OFFSET_MAX; 21212ff7e61eSJeff Mahoney inode->i_mapping->a_ops = &btree_aops; 2122f37938e0SEric Sandeen 21232ff7e61eSJeff Mahoney RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 212443eb5f29SQu Wenruo extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree, 21252c53a14dSQu Wenruo IO_TREE_BTREE_INODE_IO, inode); 21267b439738SDavid Sterba BTRFS_I(inode)->io_tree.track_uptodate = false; 21272ff7e61eSJeff Mahoney extent_map_tree_init(&BTRFS_I(inode)->extent_tree); 2128f37938e0SEric Sandeen 21295c8fd99fSJosef Bacik BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root); 21302ff7e61eSJeff Mahoney memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key)); 21312ff7e61eSJeff Mahoney set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); 21322ff7e61eSJeff Mahoney btrfs_insert_inode_hash(inode); 2133f37938e0SEric Sandeen } 2134f37938e0SEric Sandeen 2135ad618368SEric Sandeen static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2136ad618368SEric Sandeen { 2137ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2138129827e3SDavid Sterba init_rwsem(&fs_info->dev_replace.rwsem); 21397f8d236aSDavid Sterba init_waitqueue_head(&fs_info->dev_replace.replace_wait); 2140ad618368SEric Sandeen } 2141ad618368SEric Sandeen 2142f9e92e40SEric Sandeen static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2143f9e92e40SEric Sandeen { 2144f9e92e40SEric Sandeen spin_lock_init(&fs_info->qgroup_lock); 2145f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_ioctl_lock); 2146f9e92e40SEric Sandeen fs_info->qgroup_tree = RB_ROOT; 2147f9e92e40SEric Sandeen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2148f9e92e40SEric Sandeen fs_info->qgroup_seq = 1; 2149f9e92e40SEric Sandeen fs_info->qgroup_ulist = NULL; 2150d2c609b8SJeff Mahoney fs_info->qgroup_rescan_running = false; 2151f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_rescan_lock); 2152f9e92e40SEric Sandeen } 2153f9e92e40SEric Sandeen 21542a458198SEric Sandeen static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info, 21552a458198SEric Sandeen struct btrfs_fs_devices *fs_devices) 21562a458198SEric Sandeen { 2157f7b885beSAnand Jain u32 max_active = fs_info->thread_pool_size; 21586f011058SDavid Sterba unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 21592a458198SEric Sandeen 21602a458198SEric Sandeen fs_info->workers = 2161cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "worker", 2162cb001095SJeff Mahoney flags | WQ_HIGHPRI, max_active, 16); 21632a458198SEric Sandeen 21642a458198SEric Sandeen fs_info->delalloc_workers = 2165cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delalloc", 2166cb001095SJeff Mahoney flags, max_active, 2); 21672a458198SEric Sandeen 21682a458198SEric Sandeen fs_info->flush_workers = 2169cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "flush_delalloc", 2170cb001095SJeff Mahoney flags, max_active, 0); 21712a458198SEric Sandeen 21722a458198SEric Sandeen fs_info->caching_workers = 2173cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); 21742a458198SEric Sandeen 21752a458198SEric Sandeen fs_info->fixup_workers = 2176cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); 21772a458198SEric Sandeen 21782a458198SEric Sandeen /* 21792a458198SEric Sandeen * endios are largely parallel and should have a very 21802a458198SEric Sandeen * low idle thresh 21812a458198SEric Sandeen */ 21822a458198SEric Sandeen fs_info->endio_workers = 2183cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4); 21842a458198SEric Sandeen fs_info->endio_meta_workers = 2185cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta", flags, 2186cb001095SJeff Mahoney max_active, 4); 21872a458198SEric Sandeen fs_info->endio_meta_write_workers = 2188cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags, 2189cb001095SJeff Mahoney max_active, 2); 21902a458198SEric Sandeen fs_info->endio_raid56_workers = 2191cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-raid56", flags, 2192cb001095SJeff Mahoney max_active, 4); 21932a458198SEric Sandeen fs_info->rmw_workers = 2194cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2); 21952a458198SEric Sandeen fs_info->endio_write_workers = 2196cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-write", flags, 2197cb001095SJeff Mahoney max_active, 2); 21982a458198SEric Sandeen fs_info->endio_freespace_worker = 2199cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "freespace-write", flags, 2200cb001095SJeff Mahoney max_active, 0); 22012a458198SEric Sandeen fs_info->delayed_workers = 2202cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, 2203cb001095SJeff Mahoney max_active, 0); 22042a458198SEric Sandeen fs_info->readahead_workers = 2205cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "readahead", flags, 2206cb001095SJeff Mahoney max_active, 2); 22072a458198SEric Sandeen fs_info->qgroup_rescan_workers = 2208cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); 2209b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers = 2210b0643e59SDennis Zhou alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1); 22112a458198SEric Sandeen 22122a458198SEric Sandeen if (!(fs_info->workers && fs_info->delalloc_workers && 2213ba8a9d07SChris Mason fs_info->flush_workers && 22142a458198SEric Sandeen fs_info->endio_workers && fs_info->endio_meta_workers && 22152a458198SEric Sandeen fs_info->endio_meta_write_workers && 22162a458198SEric Sandeen fs_info->endio_write_workers && fs_info->endio_raid56_workers && 22172a458198SEric Sandeen fs_info->endio_freespace_worker && fs_info->rmw_workers && 22182a458198SEric Sandeen fs_info->caching_workers && fs_info->readahead_workers && 22192a458198SEric Sandeen fs_info->fixup_workers && fs_info->delayed_workers && 2220b0643e59SDennis Zhou fs_info->qgroup_rescan_workers && 2221b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers)) { 22222a458198SEric Sandeen return -ENOMEM; 22232a458198SEric Sandeen } 22242a458198SEric Sandeen 22252a458198SEric Sandeen return 0; 22262a458198SEric Sandeen } 22272a458198SEric Sandeen 22286d97c6e3SJohannes Thumshirn static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type) 22296d97c6e3SJohannes Thumshirn { 22306d97c6e3SJohannes Thumshirn struct crypto_shash *csum_shash; 2231b4e967beSDavid Sterba const char *csum_driver = btrfs_super_csum_driver(csum_type); 22326d97c6e3SJohannes Thumshirn 2233b4e967beSDavid Sterba csum_shash = crypto_alloc_shash(csum_driver, 0, 0); 22346d97c6e3SJohannes Thumshirn 22356d97c6e3SJohannes Thumshirn if (IS_ERR(csum_shash)) { 22366d97c6e3SJohannes Thumshirn btrfs_err(fs_info, "error allocating %s hash for checksum", 2237b4e967beSDavid Sterba csum_driver); 22386d97c6e3SJohannes Thumshirn return PTR_ERR(csum_shash); 22396d97c6e3SJohannes Thumshirn } 22406d97c6e3SJohannes Thumshirn 22416d97c6e3SJohannes Thumshirn fs_info->csum_shash = csum_shash; 22426d97c6e3SJohannes Thumshirn 22436d97c6e3SJohannes Thumshirn return 0; 22446d97c6e3SJohannes Thumshirn } 22456d97c6e3SJohannes Thumshirn 224663443bf5SEric Sandeen static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 224763443bf5SEric Sandeen struct btrfs_fs_devices *fs_devices) 224863443bf5SEric Sandeen { 224963443bf5SEric Sandeen int ret; 225063443bf5SEric Sandeen struct btrfs_root *log_tree_root; 225163443bf5SEric Sandeen struct btrfs_super_block *disk_super = fs_info->super_copy; 225263443bf5SEric Sandeen u64 bytenr = btrfs_super_log_root(disk_super); 2253581c1760SQu Wenruo int level = btrfs_super_log_root_level(disk_super); 225463443bf5SEric Sandeen 225563443bf5SEric Sandeen if (fs_devices->rw_devices == 0) { 2256f14d104dSDavid Sterba btrfs_warn(fs_info, "log replay required on RO media"); 225763443bf5SEric Sandeen return -EIO; 225863443bf5SEric Sandeen } 225963443bf5SEric Sandeen 226096dfcb46SJosef Bacik log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, 226196dfcb46SJosef Bacik GFP_KERNEL); 226263443bf5SEric Sandeen if (!log_tree_root) 226363443bf5SEric Sandeen return -ENOMEM; 226463443bf5SEric Sandeen 22652ff7e61eSJeff Mahoney log_tree_root->node = read_tree_block(fs_info, bytenr, 2266581c1760SQu Wenruo fs_info->generation + 1, 2267581c1760SQu Wenruo level, NULL); 226864c043deSLiu Bo if (IS_ERR(log_tree_root->node)) { 2269f14d104dSDavid Sterba btrfs_warn(fs_info, "failed to read log tree"); 22700eeff236SLiu Bo ret = PTR_ERR(log_tree_root->node); 22718c38938cSJosef Bacik log_tree_root->node = NULL; 227200246528SJosef Bacik btrfs_put_root(log_tree_root); 22730eeff236SLiu Bo return ret; 227464c043deSLiu Bo } else if (!extent_buffer_uptodate(log_tree_root->node)) { 2275f14d104dSDavid Sterba btrfs_err(fs_info, "failed to read log tree"); 227600246528SJosef Bacik btrfs_put_root(log_tree_root); 227763443bf5SEric Sandeen return -EIO; 227863443bf5SEric Sandeen } 227963443bf5SEric Sandeen /* returns with log_tree_root freed on success */ 228063443bf5SEric Sandeen ret = btrfs_recover_log_trees(log_tree_root); 228163443bf5SEric Sandeen if (ret) { 22820b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 228363443bf5SEric Sandeen "Failed to recover log tree"); 228400246528SJosef Bacik btrfs_put_root(log_tree_root); 228563443bf5SEric Sandeen return ret; 228663443bf5SEric Sandeen } 228763443bf5SEric Sandeen 2288bc98a42cSDavid Howells if (sb_rdonly(fs_info->sb)) { 22896bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 229063443bf5SEric Sandeen if (ret) 229163443bf5SEric Sandeen return ret; 229263443bf5SEric Sandeen } 229363443bf5SEric Sandeen 229463443bf5SEric Sandeen return 0; 229563443bf5SEric Sandeen } 229663443bf5SEric Sandeen 22976bccf3abSJeff Mahoney static int btrfs_read_roots(struct btrfs_fs_info *fs_info) 22984bbcaa64SEric Sandeen { 22996bccf3abSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 2300a4f3d2c4SDavid Sterba struct btrfs_root *root; 23014bbcaa64SEric Sandeen struct btrfs_key location; 23024bbcaa64SEric Sandeen int ret; 23034bbcaa64SEric Sandeen 23046bccf3abSJeff Mahoney BUG_ON(!fs_info->tree_root); 23056bccf3abSJeff Mahoney 23064bbcaa64SEric Sandeen location.objectid = BTRFS_EXTENT_TREE_OBJECTID; 23074bbcaa64SEric Sandeen location.type = BTRFS_ROOT_ITEM_KEY; 23084bbcaa64SEric Sandeen location.offset = 0; 23094bbcaa64SEric Sandeen 2310a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2311f50f4353SLiu Bo if (IS_ERR(root)) { 231242437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2313f50f4353SLiu Bo ret = PTR_ERR(root); 2314f50f4353SLiu Bo goto out; 2315f50f4353SLiu Bo } 231642437a63SJosef Bacik } else { 2317a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2318a4f3d2c4SDavid Sterba fs_info->extent_root = root; 231942437a63SJosef Bacik } 23204bbcaa64SEric Sandeen 23214bbcaa64SEric Sandeen location.objectid = BTRFS_DEV_TREE_OBJECTID; 2322a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2323f50f4353SLiu Bo if (IS_ERR(root)) { 232442437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2325f50f4353SLiu Bo ret = PTR_ERR(root); 2326f50f4353SLiu Bo goto out; 2327f50f4353SLiu Bo } 232842437a63SJosef Bacik } else { 2329a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2330a4f3d2c4SDavid Sterba fs_info->dev_root = root; 23314bbcaa64SEric Sandeen btrfs_init_devices_late(fs_info); 233242437a63SJosef Bacik } 23334bbcaa64SEric Sandeen 2334882dbe0cSJosef Bacik /* If IGNOREDATACSUMS is set don't bother reading the csum root. */ 2335882dbe0cSJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREDATACSUMS)) { 23364bbcaa64SEric Sandeen location.objectid = BTRFS_CSUM_TREE_OBJECTID; 2337a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2338f50f4353SLiu Bo if (IS_ERR(root)) { 233942437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2340f50f4353SLiu Bo ret = PTR_ERR(root); 2341f50f4353SLiu Bo goto out; 2342f50f4353SLiu Bo } 234342437a63SJosef Bacik } else { 2344a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2345a4f3d2c4SDavid Sterba fs_info->csum_root = root; 234642437a63SJosef Bacik } 2347882dbe0cSJosef Bacik } 23484bbcaa64SEric Sandeen 2349aeb935a4SQu Wenruo /* 2350aeb935a4SQu Wenruo * This tree can share blocks with some other fs tree during relocation 2351aeb935a4SQu Wenruo * and we need a proper setup by btrfs_get_fs_root 2352aeb935a4SQu Wenruo */ 235356e9357aSDavid Sterba root = btrfs_get_fs_root(tree_root->fs_info, 235456e9357aSDavid Sterba BTRFS_DATA_RELOC_TREE_OBJECTID, true); 2355aeb935a4SQu Wenruo if (IS_ERR(root)) { 235642437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2357aeb935a4SQu Wenruo ret = PTR_ERR(root); 2358aeb935a4SQu Wenruo goto out; 2359aeb935a4SQu Wenruo } 236042437a63SJosef Bacik } else { 2361aeb935a4SQu Wenruo set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2362aeb935a4SQu Wenruo fs_info->data_reloc_root = root; 236342437a63SJosef Bacik } 2364aeb935a4SQu Wenruo 23654bbcaa64SEric Sandeen location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2366a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2367a4f3d2c4SDavid Sterba if (!IS_ERR(root)) { 2368a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2369afcdd129SJosef Bacik set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); 2370a4f3d2c4SDavid Sterba fs_info->quota_root = root; 23714bbcaa64SEric Sandeen } 23724bbcaa64SEric Sandeen 23734bbcaa64SEric Sandeen location.objectid = BTRFS_UUID_TREE_OBJECTID; 2374a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2375a4f3d2c4SDavid Sterba if (IS_ERR(root)) { 237642437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2377a4f3d2c4SDavid Sterba ret = PTR_ERR(root); 23784bbcaa64SEric Sandeen if (ret != -ENOENT) 2379f50f4353SLiu Bo goto out; 238042437a63SJosef Bacik } 23814bbcaa64SEric Sandeen } else { 2382a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2383a4f3d2c4SDavid Sterba fs_info->uuid_root = root; 23844bbcaa64SEric Sandeen } 23854bbcaa64SEric Sandeen 238670f6d82eSOmar Sandoval if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 238770f6d82eSOmar Sandoval location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID; 238870f6d82eSOmar Sandoval root = btrfs_read_tree_root(tree_root, &location); 2389f50f4353SLiu Bo if (IS_ERR(root)) { 239042437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2391f50f4353SLiu Bo ret = PTR_ERR(root); 2392f50f4353SLiu Bo goto out; 2393f50f4353SLiu Bo } 239442437a63SJosef Bacik } else { 239570f6d82eSOmar Sandoval set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 239670f6d82eSOmar Sandoval fs_info->free_space_root = root; 239770f6d82eSOmar Sandoval } 239842437a63SJosef Bacik } 239970f6d82eSOmar Sandoval 24004bbcaa64SEric Sandeen return 0; 2401f50f4353SLiu Bo out: 2402f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", 2403f50f4353SLiu Bo location.objectid, ret); 2404f50f4353SLiu Bo return ret; 24054bbcaa64SEric Sandeen } 24064bbcaa64SEric Sandeen 2407069ec957SQu Wenruo /* 2408069ec957SQu Wenruo * Real super block validation 2409069ec957SQu Wenruo * NOTE: super csum type and incompat features will not be checked here. 2410069ec957SQu Wenruo * 2411069ec957SQu Wenruo * @sb: super block to check 2412069ec957SQu Wenruo * @mirror_num: the super block number to check its bytenr: 2413069ec957SQu Wenruo * 0 the primary (1st) sb 2414069ec957SQu Wenruo * 1, 2 2nd and 3rd backup copy 2415069ec957SQu Wenruo * -1 skip bytenr check 2416069ec957SQu Wenruo */ 2417069ec957SQu Wenruo static int validate_super(struct btrfs_fs_info *fs_info, 2418069ec957SQu Wenruo struct btrfs_super_block *sb, int mirror_num) 241921a852b0SQu Wenruo { 242021a852b0SQu Wenruo u64 nodesize = btrfs_super_nodesize(sb); 242121a852b0SQu Wenruo u64 sectorsize = btrfs_super_sectorsize(sb); 242221a852b0SQu Wenruo int ret = 0; 242321a852b0SQu Wenruo 242421a852b0SQu Wenruo if (btrfs_super_magic(sb) != BTRFS_MAGIC) { 242521a852b0SQu Wenruo btrfs_err(fs_info, "no valid FS found"); 242621a852b0SQu Wenruo ret = -EINVAL; 242721a852b0SQu Wenruo } 242821a852b0SQu Wenruo if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { 242921a852b0SQu Wenruo btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", 243021a852b0SQu Wenruo btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); 243121a852b0SQu Wenruo ret = -EINVAL; 243221a852b0SQu Wenruo } 243321a852b0SQu Wenruo if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 243421a852b0SQu Wenruo btrfs_err(fs_info, "tree_root level too big: %d >= %d", 243521a852b0SQu Wenruo btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 243621a852b0SQu Wenruo ret = -EINVAL; 243721a852b0SQu Wenruo } 243821a852b0SQu Wenruo if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 243921a852b0SQu Wenruo btrfs_err(fs_info, "chunk_root level too big: %d >= %d", 244021a852b0SQu Wenruo btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 244121a852b0SQu Wenruo ret = -EINVAL; 244221a852b0SQu Wenruo } 244321a852b0SQu Wenruo if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 244421a852b0SQu Wenruo btrfs_err(fs_info, "log_root level too big: %d >= %d", 244521a852b0SQu Wenruo btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 244621a852b0SQu Wenruo ret = -EINVAL; 244721a852b0SQu Wenruo } 244821a852b0SQu Wenruo 244921a852b0SQu Wenruo /* 245021a852b0SQu Wenruo * Check sectorsize and nodesize first, other check will need it. 245121a852b0SQu Wenruo * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. 245221a852b0SQu Wenruo */ 245321a852b0SQu Wenruo if (!is_power_of_2(sectorsize) || sectorsize < 4096 || 245421a852b0SQu Wenruo sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { 245521a852b0SQu Wenruo btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); 245621a852b0SQu Wenruo ret = -EINVAL; 245721a852b0SQu Wenruo } 245821a852b0SQu Wenruo /* Only PAGE SIZE is supported yet */ 245921a852b0SQu Wenruo if (sectorsize != PAGE_SIZE) { 246021a852b0SQu Wenruo btrfs_err(fs_info, 246121a852b0SQu Wenruo "sectorsize %llu not supported yet, only support %lu", 246221a852b0SQu Wenruo sectorsize, PAGE_SIZE); 246321a852b0SQu Wenruo ret = -EINVAL; 246421a852b0SQu Wenruo } 246521a852b0SQu Wenruo if (!is_power_of_2(nodesize) || nodesize < sectorsize || 246621a852b0SQu Wenruo nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { 246721a852b0SQu Wenruo btrfs_err(fs_info, "invalid nodesize %llu", nodesize); 246821a852b0SQu Wenruo ret = -EINVAL; 246921a852b0SQu Wenruo } 247021a852b0SQu Wenruo if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { 247121a852b0SQu Wenruo btrfs_err(fs_info, "invalid leafsize %u, should be %llu", 247221a852b0SQu Wenruo le32_to_cpu(sb->__unused_leafsize), nodesize); 247321a852b0SQu Wenruo ret = -EINVAL; 247421a852b0SQu Wenruo } 247521a852b0SQu Wenruo 247621a852b0SQu Wenruo /* Root alignment check */ 247721a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { 247821a852b0SQu Wenruo btrfs_warn(fs_info, "tree_root block unaligned: %llu", 247921a852b0SQu Wenruo btrfs_super_root(sb)); 248021a852b0SQu Wenruo ret = -EINVAL; 248121a852b0SQu Wenruo } 248221a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { 248321a852b0SQu Wenruo btrfs_warn(fs_info, "chunk_root block unaligned: %llu", 248421a852b0SQu Wenruo btrfs_super_chunk_root(sb)); 248521a852b0SQu Wenruo ret = -EINVAL; 248621a852b0SQu Wenruo } 248721a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { 248821a852b0SQu Wenruo btrfs_warn(fs_info, "log_root block unaligned: %llu", 248921a852b0SQu Wenruo btrfs_super_log_root(sb)); 249021a852b0SQu Wenruo ret = -EINVAL; 249121a852b0SQu Wenruo } 249221a852b0SQu Wenruo 2493de37aa51SNikolay Borisov if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, 24947239ff4bSNikolay Borisov BTRFS_FSID_SIZE) != 0) { 249521a852b0SQu Wenruo btrfs_err(fs_info, 24967239ff4bSNikolay Borisov "dev_item UUID does not match metadata fsid: %pU != %pU", 2497de37aa51SNikolay Borisov fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); 249821a852b0SQu Wenruo ret = -EINVAL; 249921a852b0SQu Wenruo } 250021a852b0SQu Wenruo 250121a852b0SQu Wenruo /* 250221a852b0SQu Wenruo * Hint to catch really bogus numbers, bitflips or so, more exact checks are 250321a852b0SQu Wenruo * done later 250421a852b0SQu Wenruo */ 250521a852b0SQu Wenruo if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { 250621a852b0SQu Wenruo btrfs_err(fs_info, "bytes_used is too small %llu", 250721a852b0SQu Wenruo btrfs_super_bytes_used(sb)); 250821a852b0SQu Wenruo ret = -EINVAL; 250921a852b0SQu Wenruo } 251021a852b0SQu Wenruo if (!is_power_of_2(btrfs_super_stripesize(sb))) { 251121a852b0SQu Wenruo btrfs_err(fs_info, "invalid stripesize %u", 251221a852b0SQu Wenruo btrfs_super_stripesize(sb)); 251321a852b0SQu Wenruo ret = -EINVAL; 251421a852b0SQu Wenruo } 251521a852b0SQu Wenruo if (btrfs_super_num_devices(sb) > (1UL << 31)) 251621a852b0SQu Wenruo btrfs_warn(fs_info, "suspicious number of devices: %llu", 251721a852b0SQu Wenruo btrfs_super_num_devices(sb)); 251821a852b0SQu Wenruo if (btrfs_super_num_devices(sb) == 0) { 251921a852b0SQu Wenruo btrfs_err(fs_info, "number of devices is 0"); 252021a852b0SQu Wenruo ret = -EINVAL; 252121a852b0SQu Wenruo } 252221a852b0SQu Wenruo 2523069ec957SQu Wenruo if (mirror_num >= 0 && 2524069ec957SQu Wenruo btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { 252521a852b0SQu Wenruo btrfs_err(fs_info, "super offset mismatch %llu != %u", 252621a852b0SQu Wenruo btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 252721a852b0SQu Wenruo ret = -EINVAL; 252821a852b0SQu Wenruo } 252921a852b0SQu Wenruo 253021a852b0SQu Wenruo /* 253121a852b0SQu Wenruo * Obvious sys_chunk_array corruptions, it must hold at least one key 253221a852b0SQu Wenruo * and one chunk 253321a852b0SQu Wenruo */ 253421a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 253521a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too big %u > %u", 253621a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 253721a852b0SQu Wenruo BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 253821a852b0SQu Wenruo ret = -EINVAL; 253921a852b0SQu Wenruo } 254021a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 254121a852b0SQu Wenruo + sizeof(struct btrfs_chunk)) { 254221a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too small %u < %zu", 254321a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 254421a852b0SQu Wenruo sizeof(struct btrfs_disk_key) 254521a852b0SQu Wenruo + sizeof(struct btrfs_chunk)); 254621a852b0SQu Wenruo ret = -EINVAL; 254721a852b0SQu Wenruo } 254821a852b0SQu Wenruo 254921a852b0SQu Wenruo /* 255021a852b0SQu Wenruo * The generation is a global counter, we'll trust it more than the others 255121a852b0SQu Wenruo * but it's still possible that it's the one that's wrong. 255221a852b0SQu Wenruo */ 255321a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 255421a852b0SQu Wenruo btrfs_warn(fs_info, 255521a852b0SQu Wenruo "suspicious: generation < chunk_root_generation: %llu < %llu", 255621a852b0SQu Wenruo btrfs_super_generation(sb), 255721a852b0SQu Wenruo btrfs_super_chunk_root_generation(sb)); 255821a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 255921a852b0SQu Wenruo && btrfs_super_cache_generation(sb) != (u64)-1) 256021a852b0SQu Wenruo btrfs_warn(fs_info, 256121a852b0SQu Wenruo "suspicious: generation < cache_generation: %llu < %llu", 256221a852b0SQu Wenruo btrfs_super_generation(sb), 256321a852b0SQu Wenruo btrfs_super_cache_generation(sb)); 256421a852b0SQu Wenruo 256521a852b0SQu Wenruo return ret; 256621a852b0SQu Wenruo } 256721a852b0SQu Wenruo 2568069ec957SQu Wenruo /* 2569069ec957SQu Wenruo * Validation of super block at mount time. 2570069ec957SQu Wenruo * Some checks already done early at mount time, like csum type and incompat 2571069ec957SQu Wenruo * flags will be skipped. 2572069ec957SQu Wenruo */ 2573069ec957SQu Wenruo static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) 2574069ec957SQu Wenruo { 2575069ec957SQu Wenruo return validate_super(fs_info, fs_info->super_copy, 0); 2576069ec957SQu Wenruo } 2577069ec957SQu Wenruo 257875cb857dSQu Wenruo /* 257975cb857dSQu Wenruo * Validation of super block at write time. 258075cb857dSQu Wenruo * Some checks like bytenr check will be skipped as their values will be 258175cb857dSQu Wenruo * overwritten soon. 258275cb857dSQu Wenruo * Extra checks like csum type and incompat flags will be done here. 258375cb857dSQu Wenruo */ 258475cb857dSQu Wenruo static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, 258575cb857dSQu Wenruo struct btrfs_super_block *sb) 258675cb857dSQu Wenruo { 258775cb857dSQu Wenruo int ret; 258875cb857dSQu Wenruo 258975cb857dSQu Wenruo ret = validate_super(fs_info, sb, -1); 259075cb857dSQu Wenruo if (ret < 0) 259175cb857dSQu Wenruo goto out; 2592e7e16f48SJohannes Thumshirn if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) { 259375cb857dSQu Wenruo ret = -EUCLEAN; 259475cb857dSQu Wenruo btrfs_err(fs_info, "invalid csum type, has %u want %u", 259575cb857dSQu Wenruo btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); 259675cb857dSQu Wenruo goto out; 259775cb857dSQu Wenruo } 259875cb857dSQu Wenruo if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 259975cb857dSQu Wenruo ret = -EUCLEAN; 260075cb857dSQu Wenruo btrfs_err(fs_info, 260175cb857dSQu Wenruo "invalid incompat flags, has 0x%llx valid mask 0x%llx", 260275cb857dSQu Wenruo btrfs_super_incompat_flags(sb), 260375cb857dSQu Wenruo (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); 260475cb857dSQu Wenruo goto out; 260575cb857dSQu Wenruo } 260675cb857dSQu Wenruo out: 260775cb857dSQu Wenruo if (ret < 0) 260875cb857dSQu Wenruo btrfs_err(fs_info, 260975cb857dSQu Wenruo "super block corruption detected before writing it to disk"); 261075cb857dSQu Wenruo return ret; 261175cb857dSQu Wenruo } 261275cb857dSQu Wenruo 26136ef108ddSNikolay Borisov static int __cold init_tree_roots(struct btrfs_fs_info *fs_info) 2614b8522a1eSNikolay Borisov { 26156ef108ddSNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 2616b8522a1eSNikolay Borisov struct btrfs_super_block *sb = fs_info->super_copy; 2617b8522a1eSNikolay Borisov struct btrfs_root *tree_root = fs_info->tree_root; 2618b8522a1eSNikolay Borisov bool handle_error = false; 2619b8522a1eSNikolay Borisov int ret = 0; 2620b8522a1eSNikolay Borisov int i; 2621b8522a1eSNikolay Borisov 2622b8522a1eSNikolay Borisov for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 2623b8522a1eSNikolay Borisov u64 generation; 2624b8522a1eSNikolay Borisov int level; 2625b8522a1eSNikolay Borisov 2626b8522a1eSNikolay Borisov if (handle_error) { 2627b8522a1eSNikolay Borisov if (!IS_ERR(tree_root->node)) 2628b8522a1eSNikolay Borisov free_extent_buffer(tree_root->node); 2629b8522a1eSNikolay Borisov tree_root->node = NULL; 2630b8522a1eSNikolay Borisov 2631b8522a1eSNikolay Borisov if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) 2632b8522a1eSNikolay Borisov break; 2633b8522a1eSNikolay Borisov 2634b8522a1eSNikolay Borisov free_root_pointers(fs_info, 0); 2635b8522a1eSNikolay Borisov 2636b8522a1eSNikolay Borisov /* 2637b8522a1eSNikolay Borisov * Don't use the log in recovery mode, it won't be 2638b8522a1eSNikolay Borisov * valid 2639b8522a1eSNikolay Borisov */ 2640b8522a1eSNikolay Borisov btrfs_set_super_log_root(sb, 0); 2641b8522a1eSNikolay Borisov 2642b8522a1eSNikolay Borisov /* We can't trust the free space cache either */ 2643b8522a1eSNikolay Borisov btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 2644b8522a1eSNikolay Borisov 2645b8522a1eSNikolay Borisov ret = read_backup_root(fs_info, i); 26466ef108ddSNikolay Borisov backup_index = ret; 2647b8522a1eSNikolay Borisov if (ret < 0) 2648b8522a1eSNikolay Borisov return ret; 2649b8522a1eSNikolay Borisov } 2650b8522a1eSNikolay Borisov generation = btrfs_super_generation(sb); 2651b8522a1eSNikolay Borisov level = btrfs_super_root_level(sb); 2652b8522a1eSNikolay Borisov tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb), 2653b8522a1eSNikolay Borisov generation, level, NULL); 26540465337cSJosef Bacik if (IS_ERR(tree_root->node)) { 2655217f5004SNikolay Borisov handle_error = true; 2656b8522a1eSNikolay Borisov ret = PTR_ERR(tree_root->node); 26570465337cSJosef Bacik tree_root->node = NULL; 2658217f5004SNikolay Borisov btrfs_warn(fs_info, "couldn't read tree root"); 2659217f5004SNikolay Borisov continue; 2660b8522a1eSNikolay Borisov 2661217f5004SNikolay Borisov } else if (!extent_buffer_uptodate(tree_root->node)) { 2662217f5004SNikolay Borisov handle_error = true; 2663217f5004SNikolay Borisov ret = -EIO; 2664217f5004SNikolay Borisov btrfs_warn(fs_info, "error while reading tree root"); 2665b8522a1eSNikolay Borisov continue; 2666b8522a1eSNikolay Borisov } 2667b8522a1eSNikolay Borisov 2668b8522a1eSNikolay Borisov btrfs_set_root_node(&tree_root->root_item, tree_root->node); 2669b8522a1eSNikolay Borisov tree_root->commit_root = btrfs_root_node(tree_root); 2670b8522a1eSNikolay Borisov btrfs_set_root_refs(&tree_root->root_item, 1); 2671b8522a1eSNikolay Borisov 2672336a0d8dSNikolay Borisov /* 2673336a0d8dSNikolay Borisov * No need to hold btrfs_root::objectid_mutex since the fs 2674336a0d8dSNikolay Borisov * hasn't been fully initialised and we are the only user 2675336a0d8dSNikolay Borisov */ 2676b8522a1eSNikolay Borisov ret = btrfs_find_highest_objectid(tree_root, 2677b8522a1eSNikolay Borisov &tree_root->highest_objectid); 2678b8522a1eSNikolay Borisov if (ret < 0) { 2679b8522a1eSNikolay Borisov handle_error = true; 2680b8522a1eSNikolay Borisov continue; 2681b8522a1eSNikolay Borisov } 2682b8522a1eSNikolay Borisov 2683b8522a1eSNikolay Borisov ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 2684b8522a1eSNikolay Borisov 2685b8522a1eSNikolay Borisov ret = btrfs_read_roots(fs_info); 2686b8522a1eSNikolay Borisov if (ret < 0) { 2687b8522a1eSNikolay Borisov handle_error = true; 2688b8522a1eSNikolay Borisov continue; 2689b8522a1eSNikolay Borisov } 2690b8522a1eSNikolay Borisov 2691b8522a1eSNikolay Borisov /* All successful */ 2692b8522a1eSNikolay Borisov fs_info->generation = generation; 2693b8522a1eSNikolay Borisov fs_info->last_trans_committed = generation; 26946ef108ddSNikolay Borisov 26956ef108ddSNikolay Borisov /* Always begin writing backup roots after the one being used */ 26966ef108ddSNikolay Borisov if (backup_index < 0) { 26976ef108ddSNikolay Borisov fs_info->backup_root_index = 0; 26986ef108ddSNikolay Borisov } else { 26996ef108ddSNikolay Borisov fs_info->backup_root_index = backup_index + 1; 27006ef108ddSNikolay Borisov fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS; 27016ef108ddSNikolay Borisov } 2702b8522a1eSNikolay Borisov break; 2703b8522a1eSNikolay Borisov } 2704b8522a1eSNikolay Borisov 2705b8522a1eSNikolay Borisov return ret; 2706b8522a1eSNikolay Borisov } 2707b8522a1eSNikolay Borisov 27088260edbaSJosef Bacik void btrfs_init_fs_info(struct btrfs_fs_info *fs_info) 2709eb60ceacSChris Mason { 271076dda93cSYan, Zheng INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 2711f28491e0SJosef Bacik INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 27128fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 2713facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 271424bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 2715eb73c1b7SMiao Xie INIT_LIST_HEAD(&fs_info->delalloc_roots); 271611833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 2717eb73c1b7SMiao Xie spin_lock_init(&fs_info->delalloc_root_lock); 2718a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 271976dda93cSYan, Zheng spin_lock_init(&fs_info->fs_roots_radix_lock); 272024bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 27214cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 2722ceda0864SMiao Xie spin_lock_init(&fs_info->super_lock); 2723f28491e0SJosef Bacik spin_lock_init(&fs_info->buffer_lock); 272447ab2a6cSJosef Bacik spin_lock_init(&fs_info->unused_bgs_lock); 2725f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 2726d7c15171SZhao Lei mutex_init(&fs_info->unused_bg_unpin_mutex); 272767c5e7d4SFilipe Manana mutex_init(&fs_info->delete_unused_bgs_mutex); 27287585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 2729573bfb72SMiao Xie mutex_init(&fs_info->delalloc_root_mutex); 2730de98ced9SMiao Xie seqlock_init(&fs_info->profiles_lock); 273119c00ddcSChris Mason 27320b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 27336324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 2734f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 273547ab2a6cSJosef Bacik INIT_LIST_HEAD(&fs_info->unused_bgs); 2736bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 2737bd647ce3SJosef Bacik INIT_LIST_HEAD(&fs_info->allocated_roots); 27383fd63727SJosef Bacik INIT_LIST_HEAD(&fs_info->allocated_ebs); 27393fd63727SJosef Bacik spin_lock_init(&fs_info->eb_leak_lock); 2740bd647ce3SJosef Bacik #endif 2741c8bf1b67SDavid Sterba extent_map_tree_init(&fs_info->mapping_tree); 274266d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 274366d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 274466d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 274566d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 274666d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 274766d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 274866d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 2749ba2c4d4eSJosef Bacik btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, 2750ba2c4d4eSJosef Bacik BTRFS_BLOCK_RSV_DELREFS); 2751ba2c4d4eSJosef Bacik 2752771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 27534cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 27542fefd558SZhao Lei atomic_set(&fs_info->reada_works_cnt, 0); 2755034f784dSJosef Bacik atomic_set(&fs_info->nr_delayed_iputs, 0); 2756fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 275795ac567aSFilipe David Borba Manana fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 27589ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 27594cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 2760a5ed45f8SNikolay Borisov atomic64_set(&fs_info->free_chunk_space, 0); 2761f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 27628b87dc17SDavid Sterba fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 2763f8c269d7SDavid Sterba fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 276490519d66SArne Jansen /* readahead state */ 2765d0164adcSMel Gorman INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); 276690519d66SArne Jansen spin_lock_init(&fs_info->reada_lock); 2767fd708b81SJosef Bacik btrfs_init_ref_verify(fs_info); 2768c8b97818SChris Mason 2769b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 2770b34b086cSChris Mason num_online_cpus() + 2, 8); 27710afbaf8cSChris Mason 2772199c2a9cSMiao Xie INIT_LIST_HEAD(&fs_info->ordered_roots); 2773199c2a9cSMiao Xie spin_lock_init(&fs_info->ordered_root_lock); 277469fe2d75SJosef Bacik 2775638aa7edSEric Sandeen btrfs_init_scrub(fs_info); 277621adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 277721adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 277821adbd5cSStefan Behrens #endif 2779779a65a4SEric Sandeen btrfs_init_balance(fs_info); 278057056740SJosef Bacik btrfs_init_async_reclaim_work(fs_info); 2781a2de733cSArne Jansen 27820f9dd46cSJosef Bacik spin_lock_init(&fs_info->block_group_cache_lock); 27836bef4d31SEric Paris fs_info->block_group_cache_tree = RB_ROOT; 2784a1897fddSLiu Bo fs_info->first_logical_byte = (u64)-1; 27850f9dd46cSJosef Bacik 2786fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &fs_info->excluded_extents, 2787fe119a6eSNikolay Borisov IO_TREE_FS_EXCLUDED_EXTENTS, NULL); 2788afcdd129SJosef Bacik set_bit(BTRFS_FS_BARRIER, &fs_info->flags); 278939279cc3SChris Mason 27905a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 2791e02119d5SChris Mason mutex_init(&fs_info->tree_log_mutex); 2792925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 2793a74a4b97SChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 2794a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 27951bbc621eSChris Mason mutex_init(&fs_info->ro_block_group_mutex); 27969e351cc8SJosef Bacik init_rwsem(&fs_info->commit_root_sem); 2797c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 279876dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 2799803b2f54SStefan Behrens sema_init(&fs_info->uuid_tree_rescan_sem, 1); 2800fa9c0d79SChris Mason 2801ad618368SEric Sandeen btrfs_init_dev_replace_locks(fs_info); 2802f9e92e40SEric Sandeen btrfs_init_qgroup(fs_info); 2803b0643e59SDennis Zhou btrfs_discard_init(fs_info); 2804416ac51dSArne Jansen 2805fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2806fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2807fa9c0d79SChris Mason 2808e6dcd2dcSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 2809f9295749SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 2810bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 28114854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 2812034f784dSJosef Bacik init_waitqueue_head(&fs_info->delayed_iputs_wait); 28133768f368SChris Mason 2814da17066cSJeff Mahoney /* Usable values until the real ones are cached from the superblock */ 2815da17066cSJeff Mahoney fs_info->nodesize = 4096; 2816da17066cSJeff Mahoney fs_info->sectorsize = 4096; 2817da17066cSJeff Mahoney fs_info->stripesize = 4096; 2818da17066cSJeff Mahoney 2819eede2bf3SOmar Sandoval spin_lock_init(&fs_info->swapfile_pins_lock); 2820eede2bf3SOmar Sandoval fs_info->swapfile_pins = RB_ROOT; 2821eede2bf3SOmar Sandoval 28229e967495SFilipe Manana fs_info->send_in_progress = 0; 28238260edbaSJosef Bacik } 28248260edbaSJosef Bacik 28258260edbaSJosef Bacik static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb) 28268260edbaSJosef Bacik { 28278260edbaSJosef Bacik int ret; 28288260edbaSJosef Bacik 28298260edbaSJosef Bacik fs_info->sb = sb; 28308260edbaSJosef Bacik sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; 28318260edbaSJosef Bacik sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); 28329e967495SFilipe Manana 2833c75e8394SJosef Bacik ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL); 2834ae18c37aSJosef Bacik if (ret) 2835ae18c37aSJosef Bacik return ret; 2836ae18c37aSJosef Bacik 2837ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 2838ae18c37aSJosef Bacik if (ret) 2839c75e8394SJosef Bacik return ret; 2840ae18c37aSJosef Bacik 2841ae18c37aSJosef Bacik fs_info->dirty_metadata_batch = PAGE_SIZE * 2842ae18c37aSJosef Bacik (1 + ilog2(nr_cpu_ids)); 2843ae18c37aSJosef Bacik 2844ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 2845ae18c37aSJosef Bacik if (ret) 2846c75e8394SJosef Bacik return ret; 2847ae18c37aSJosef Bacik 2848ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, 2849ae18c37aSJosef Bacik GFP_KERNEL); 2850ae18c37aSJosef Bacik if (ret) 2851c75e8394SJosef Bacik return ret; 2852ae18c37aSJosef Bacik 2853ae18c37aSJosef Bacik fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 2854ae18c37aSJosef Bacik GFP_KERNEL); 2855c75e8394SJosef Bacik if (!fs_info->delayed_root) 2856c75e8394SJosef Bacik return -ENOMEM; 2857ae18c37aSJosef Bacik btrfs_init_delayed_root(fs_info->delayed_root); 2858ae18c37aSJosef Bacik 2859c75e8394SJosef Bacik return btrfs_alloc_stripe_hash_table(fs_info); 2860ae18c37aSJosef Bacik } 2861ae18c37aSJosef Bacik 286297f4dd09SNikolay Borisov static int btrfs_uuid_rescan_kthread(void *data) 286397f4dd09SNikolay Borisov { 286497f4dd09SNikolay Borisov struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; 286597f4dd09SNikolay Borisov int ret; 286697f4dd09SNikolay Borisov 286797f4dd09SNikolay Borisov /* 286897f4dd09SNikolay Borisov * 1st step is to iterate through the existing UUID tree and 286997f4dd09SNikolay Borisov * to delete all entries that contain outdated data. 287097f4dd09SNikolay Borisov * 2nd step is to add all missing entries to the UUID tree. 287197f4dd09SNikolay Borisov */ 287297f4dd09SNikolay Borisov ret = btrfs_uuid_tree_iterate(fs_info); 287397f4dd09SNikolay Borisov if (ret < 0) { 2874c94bec2cSJosef Bacik if (ret != -EINTR) 2875c94bec2cSJosef Bacik btrfs_warn(fs_info, "iterating uuid_tree failed %d", 2876c94bec2cSJosef Bacik ret); 287797f4dd09SNikolay Borisov up(&fs_info->uuid_tree_rescan_sem); 287897f4dd09SNikolay Borisov return ret; 287997f4dd09SNikolay Borisov } 288097f4dd09SNikolay Borisov return btrfs_uuid_scan_kthread(data); 288197f4dd09SNikolay Borisov } 288297f4dd09SNikolay Borisov 288397f4dd09SNikolay Borisov static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) 288497f4dd09SNikolay Borisov { 288597f4dd09SNikolay Borisov struct task_struct *task; 288697f4dd09SNikolay Borisov 288797f4dd09SNikolay Borisov down(&fs_info->uuid_tree_rescan_sem); 288897f4dd09SNikolay Borisov task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); 288997f4dd09SNikolay Borisov if (IS_ERR(task)) { 289097f4dd09SNikolay Borisov /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 289197f4dd09SNikolay Borisov btrfs_warn(fs_info, "failed to start uuid_rescan task"); 289297f4dd09SNikolay Borisov up(&fs_info->uuid_tree_rescan_sem); 289397f4dd09SNikolay Borisov return PTR_ERR(task); 289497f4dd09SNikolay Borisov } 289597f4dd09SNikolay Borisov 289697f4dd09SNikolay Borisov return 0; 289797f4dd09SNikolay Borisov } 289897f4dd09SNikolay Borisov 2899ae18c37aSJosef Bacik int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices, 2900ae18c37aSJosef Bacik char *options) 2901ae18c37aSJosef Bacik { 2902ae18c37aSJosef Bacik u32 sectorsize; 2903ae18c37aSJosef Bacik u32 nodesize; 2904ae18c37aSJosef Bacik u32 stripesize; 2905ae18c37aSJosef Bacik u64 generation; 2906ae18c37aSJosef Bacik u64 features; 2907ae18c37aSJosef Bacik u16 csum_type; 2908ae18c37aSJosef Bacik struct btrfs_super_block *disk_super; 2909ae18c37aSJosef Bacik struct btrfs_fs_info *fs_info = btrfs_sb(sb); 2910ae18c37aSJosef Bacik struct btrfs_root *tree_root; 2911ae18c37aSJosef Bacik struct btrfs_root *chunk_root; 2912ae18c37aSJosef Bacik int ret; 2913ae18c37aSJosef Bacik int err = -EINVAL; 2914ae18c37aSJosef Bacik int clear_free_space_tree = 0; 2915ae18c37aSJosef Bacik int level; 2916ae18c37aSJosef Bacik 29178260edbaSJosef Bacik ret = init_mount_fs_info(fs_info, sb); 291853b381b3SDavid Woodhouse if (ret) { 291983c8266aSDavid Sterba err = ret; 2920ae18c37aSJosef Bacik goto fail; 292153b381b3SDavid Woodhouse } 292253b381b3SDavid Woodhouse 2923ae18c37aSJosef Bacik /* These need to be init'ed before we start creating inodes and such. */ 2924ae18c37aSJosef Bacik tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, 2925ae18c37aSJosef Bacik GFP_KERNEL); 2926ae18c37aSJosef Bacik fs_info->tree_root = tree_root; 2927ae18c37aSJosef Bacik chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID, 2928ae18c37aSJosef Bacik GFP_KERNEL); 2929ae18c37aSJosef Bacik fs_info->chunk_root = chunk_root; 2930ae18c37aSJosef Bacik if (!tree_root || !chunk_root) { 2931ae18c37aSJosef Bacik err = -ENOMEM; 2932c75e8394SJosef Bacik goto fail; 2933ae18c37aSJosef Bacik } 2934ae18c37aSJosef Bacik 2935ae18c37aSJosef Bacik fs_info->btree_inode = new_inode(sb); 2936ae18c37aSJosef Bacik if (!fs_info->btree_inode) { 2937ae18c37aSJosef Bacik err = -ENOMEM; 2938c75e8394SJosef Bacik goto fail; 2939ae18c37aSJosef Bacik } 2940ae18c37aSJosef Bacik mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 2941ae18c37aSJosef Bacik btrfs_init_btree_inode(fs_info); 2942ae18c37aSJosef Bacik 29433c4bb26bSChris Mason invalidate_bdev(fs_devices->latest_bdev); 29441104a885SDavid Sterba 29451104a885SDavid Sterba /* 29461104a885SDavid Sterba * Read super block and check the signature bytes only 29471104a885SDavid Sterba */ 29488f32380dSJohannes Thumshirn disk_super = btrfs_read_dev_super(fs_devices->latest_bdev); 29498f32380dSJohannes Thumshirn if (IS_ERR(disk_super)) { 29508f32380dSJohannes Thumshirn err = PTR_ERR(disk_super); 295116cdcec7SMiao Xie goto fail_alloc; 295220b45077SDave Young } 295339279cc3SChris Mason 29541104a885SDavid Sterba /* 2955260db43cSRandy Dunlap * Verify the type first, if that or the checksum value are 29568dc3f22cSJohannes Thumshirn * corrupted, we'll find out 29578dc3f22cSJohannes Thumshirn */ 29588f32380dSJohannes Thumshirn csum_type = btrfs_super_csum_type(disk_super); 295951bce6c9SJohannes Thumshirn if (!btrfs_supported_super_csum(csum_type)) { 29608dc3f22cSJohannes Thumshirn btrfs_err(fs_info, "unsupported checksum algorithm: %u", 296151bce6c9SJohannes Thumshirn csum_type); 29628dc3f22cSJohannes Thumshirn err = -EINVAL; 29638f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 29648dc3f22cSJohannes Thumshirn goto fail_alloc; 29658dc3f22cSJohannes Thumshirn } 29668dc3f22cSJohannes Thumshirn 29676d97c6e3SJohannes Thumshirn ret = btrfs_init_csum_hash(fs_info, csum_type); 29686d97c6e3SJohannes Thumshirn if (ret) { 29696d97c6e3SJohannes Thumshirn err = ret; 29708f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 29716d97c6e3SJohannes Thumshirn goto fail_alloc; 29726d97c6e3SJohannes Thumshirn } 29736d97c6e3SJohannes Thumshirn 29748dc3f22cSJohannes Thumshirn /* 29751104a885SDavid Sterba * We want to check superblock checksum, the type is stored inside. 29761104a885SDavid Sterba * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 29771104a885SDavid Sterba */ 29788f32380dSJohannes Thumshirn if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) { 297905135f59SDavid Sterba btrfs_err(fs_info, "superblock checksum mismatch"); 29801104a885SDavid Sterba err = -EINVAL; 29818f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 2982141386e1SJosef Bacik goto fail_alloc; 29831104a885SDavid Sterba } 29841104a885SDavid Sterba 29851104a885SDavid Sterba /* 29861104a885SDavid Sterba * super_copy is zeroed at allocation time and we never touch the 29871104a885SDavid Sterba * following bytes up to INFO_SIZE, the checksum is calculated from 29881104a885SDavid Sterba * the whole block of INFO_SIZE 29891104a885SDavid Sterba */ 29908f32380dSJohannes Thumshirn memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy)); 29918f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 29925f39d397SChris Mason 2993fbc6feaeSNikolay Borisov disk_super = fs_info->super_copy; 2994fbc6feaeSNikolay Borisov 2995de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, 2996de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 2997de37aa51SNikolay Borisov 29987239ff4bSNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID)) { 2999de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid, 3000de37aa51SNikolay Borisov fs_info->super_copy->metadata_uuid, 3001de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 30027239ff4bSNikolay Borisov } 30030b86a832SChris Mason 3004fbc6feaeSNikolay Borisov features = btrfs_super_flags(disk_super); 3005fbc6feaeSNikolay Borisov if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { 3006fbc6feaeSNikolay Borisov features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; 3007fbc6feaeSNikolay Borisov btrfs_set_super_flags(disk_super, features); 3008fbc6feaeSNikolay Borisov btrfs_info(fs_info, 3009fbc6feaeSNikolay Borisov "found metadata UUID change in progress flag, clearing"); 3010fbc6feaeSNikolay Borisov } 3011fbc6feaeSNikolay Borisov 3012fbc6feaeSNikolay Borisov memcpy(fs_info->super_for_commit, fs_info->super_copy, 3013fbc6feaeSNikolay Borisov sizeof(*fs_info->super_for_commit)); 3014de37aa51SNikolay Borisov 3015069ec957SQu Wenruo ret = btrfs_validate_mount_super(fs_info); 30161104a885SDavid Sterba if (ret) { 301705135f59SDavid Sterba btrfs_err(fs_info, "superblock contains fatal errors"); 30181104a885SDavid Sterba err = -EINVAL; 3019141386e1SJosef Bacik goto fail_alloc; 30201104a885SDavid Sterba } 30211104a885SDavid Sterba 30220f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 3023141386e1SJosef Bacik goto fail_alloc; 30240f7d52f4SChris Mason 3025acce952bSliubo /* check FS state, whether FS is broken. */ 302687533c47SMiao Xie if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 302787533c47SMiao Xie set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 3028acce952bSliubo 302975e7cb7fSLiu Bo /* 303075e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 303175e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 303275e7cb7fSLiu Bo */ 303375e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 303475e7cb7fSLiu Bo 30352ff7e61eSJeff Mahoney ret = btrfs_parse_options(fs_info, options, sb->s_flags); 30362b82032cSYan Zheng if (ret) { 30372b82032cSYan Zheng err = ret; 3038141386e1SJosef Bacik goto fail_alloc; 30392b82032cSYan Zheng } 3040dfe25020SChris Mason 3041f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 3042f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 3043f2b636e8SJosef Bacik if (features) { 304405135f59SDavid Sterba btrfs_err(fs_info, 304505135f59SDavid Sterba "cannot mount because of unsupported optional features (%llx)", 3046c1c9ff7cSGeert Uytterhoeven features); 3047f2b636e8SJosef Bacik err = -EINVAL; 3048141386e1SJosef Bacik goto fail_alloc; 3049f2b636e8SJosef Bacik } 3050f2b636e8SJosef Bacik 30515d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 30525d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 30530b246afaSJeff Mahoney if (fs_info->compress_type == BTRFS_COMPRESS_LZO) 3054a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 30555c1aab1dSNick Terrell else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) 30565c1aab1dSNick Terrell features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; 3057727011e0SChris Mason 30583173a18fSJosef Bacik if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) 305905135f59SDavid Sterba btrfs_info(fs_info, "has skinny extents"); 30603173a18fSJosef Bacik 3061727011e0SChris Mason /* 3062727011e0SChris Mason * flag our filesystem as having big metadata blocks if 3063727011e0SChris Mason * they are bigger than the page size 3064727011e0SChris Mason */ 306509cbfeafSKirill A. Shutemov if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) { 3066727011e0SChris Mason if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 306705135f59SDavid Sterba btrfs_info(fs_info, 306805135f59SDavid Sterba "flagging fs with big metadata feature"); 3069727011e0SChris Mason features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 3070727011e0SChris Mason } 3071727011e0SChris Mason 3072bc3f116fSChris Mason nodesize = btrfs_super_nodesize(disk_super); 3073bc3f116fSChris Mason sectorsize = btrfs_super_sectorsize(disk_super); 3074b7f67055SChandan Rajendra stripesize = sectorsize; 3075707e8a07SDavid Sterba fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 3076963d678bSMiao Xie fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 3077bc3f116fSChris Mason 3078da17066cSJeff Mahoney /* Cache block sizes */ 3079da17066cSJeff Mahoney fs_info->nodesize = nodesize; 3080da17066cSJeff Mahoney fs_info->sectorsize = sectorsize; 3081da17066cSJeff Mahoney fs_info->stripesize = stripesize; 3082da17066cSJeff Mahoney 3083bc3f116fSChris Mason /* 3084bc3f116fSChris Mason * mixed block groups end up with duplicate but slightly offset 3085bc3f116fSChris Mason * extent buffers for the same range. It leads to corruptions 3086bc3f116fSChris Mason */ 3087bc3f116fSChris Mason if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 3088707e8a07SDavid Sterba (sectorsize != nodesize)) { 308905135f59SDavid Sterba btrfs_err(fs_info, 309005135f59SDavid Sterba "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", 309105135f59SDavid Sterba nodesize, sectorsize); 3092141386e1SJosef Bacik goto fail_alloc; 3093bc3f116fSChris Mason } 3094bc3f116fSChris Mason 3095ceda0864SMiao Xie /* 3096ceda0864SMiao Xie * Needn't use the lock because there is no other task which will 3097ceda0864SMiao Xie * update the flag. 3098ceda0864SMiao Xie */ 30995d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 31005d4f98a2SYan Zheng 3101f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 3102f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 3103bc98a42cSDavid Howells if (!sb_rdonly(sb) && features) { 310405135f59SDavid Sterba btrfs_err(fs_info, 310505135f59SDavid Sterba "cannot mount read-write because of unsupported optional features (%llx)", 3106c1c9ff7cSGeert Uytterhoeven features); 3107f2b636e8SJosef Bacik err = -EINVAL; 3108141386e1SJosef Bacik goto fail_alloc; 3109f2b636e8SJosef Bacik } 311061d92c32SChris Mason 31112a458198SEric Sandeen ret = btrfs_init_workqueues(fs_info, fs_devices); 31122a458198SEric Sandeen if (ret) { 31132a458198SEric Sandeen err = ret; 31140dc3b84aSJosef Bacik goto fail_sb_buffer; 31150dc3b84aSJosef Bacik } 31164543df7eSChris Mason 31179e11ceeeSJan Kara sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); 31189e11ceeeSJan Kara sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); 31194575c9ccSChris Mason 3120a061fc8dSChris Mason sb->s_blocksize = sectorsize; 3121a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 3122de37aa51SNikolay Borisov memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); 3123db94535dSChris Mason 3124925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 31256bccf3abSJeff Mahoney ret = btrfs_read_sys_array(fs_info); 3126925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 312784eed90fSChris Mason if (ret) { 312805135f59SDavid Sterba btrfs_err(fs_info, "failed to read the system array: %d", ret); 31295d4f98a2SYan Zheng goto fail_sb_buffer; 313084eed90fSChris Mason } 31310b86a832SChris Mason 313284234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 3133581c1760SQu Wenruo level = btrfs_super_chunk_root_level(disk_super); 31340b86a832SChris Mason 31352ff7e61eSJeff Mahoney chunk_root->node = read_tree_block(fs_info, 31360b86a832SChris Mason btrfs_super_chunk_root(disk_super), 3137581c1760SQu Wenruo generation, level, NULL); 313864c043deSLiu Bo if (IS_ERR(chunk_root->node) || 313964c043deSLiu Bo !extent_buffer_uptodate(chunk_root->node)) { 314005135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk root"); 3141e5fffbacSchandan if (!IS_ERR(chunk_root->node)) 3142e5fffbacSchandan free_extent_buffer(chunk_root->node); 314395ab1f64SZhao Lei chunk_root->node = NULL; 3144af31f5e5SChris Mason goto fail_tree_roots; 314583121942SDavid Woodhouse } 31465d4f98a2SYan Zheng btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 31475d4f98a2SYan Zheng chunk_root->commit_root = btrfs_root_node(chunk_root); 31480b86a832SChris Mason 3149e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 3150c4ac7541SDavid Sterba offsetof(struct btrfs_header, chunk_tree_uuid), 3151c4ac7541SDavid Sterba BTRFS_UUID_SIZE); 3152e17cade2SChris Mason 31535b4aacefSJeff Mahoney ret = btrfs_read_chunk_tree(fs_info); 31542b82032cSYan Zheng if (ret) { 315505135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk tree: %d", ret); 3156af31f5e5SChris Mason goto fail_tree_roots; 31572b82032cSYan Zheng } 31580b86a832SChris Mason 31598dabb742SStefan Behrens /* 31609b99b115SAnand Jain * Keep the devid that is marked to be the target device for the 31619b99b115SAnand Jain * device replace procedure 31628dabb742SStefan Behrens */ 31639b99b115SAnand Jain btrfs_free_extra_devids(fs_devices, 0); 3164dfe25020SChris Mason 3165a6b0d5c8SChris Mason if (!fs_devices->latest_bdev) { 316605135f59SDavid Sterba btrfs_err(fs_info, "failed to read devices"); 3167a6b0d5c8SChris Mason goto fail_tree_roots; 3168a6b0d5c8SChris Mason } 3169a6b0d5c8SChris Mason 3170b8522a1eSNikolay Borisov ret = init_tree_roots(fs_info); 31714bbcaa64SEric Sandeen if (ret) 3172b8522a1eSNikolay Borisov goto fail_tree_roots; 31738929ecfaSYan, Zheng 317475ec1db8SJosef Bacik /* 317575ec1db8SJosef Bacik * If we have a uuid root and we're not being told to rescan we need to 317675ec1db8SJosef Bacik * check the generation here so we can set the 317775ec1db8SJosef Bacik * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the 317875ec1db8SJosef Bacik * transaction during a balance or the log replay without updating the 317975ec1db8SJosef Bacik * uuid generation, and then if we crash we would rescan the uuid tree, 318075ec1db8SJosef Bacik * even though it was perfectly fine. 318175ec1db8SJosef Bacik */ 318275ec1db8SJosef Bacik if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) && 318375ec1db8SJosef Bacik fs_info->generation == btrfs_super_uuid_tree_generation(disk_super)) 318475ec1db8SJosef Bacik set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 318575ec1db8SJosef Bacik 3186cf90d884SQu Wenruo ret = btrfs_verify_dev_extents(fs_info); 3187cf90d884SQu Wenruo if (ret) { 3188cf90d884SQu Wenruo btrfs_err(fs_info, 3189cf90d884SQu Wenruo "failed to verify dev extents against chunks: %d", 3190cf90d884SQu Wenruo ret); 3191cf90d884SQu Wenruo goto fail_block_groups; 3192cf90d884SQu Wenruo } 319368310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 319468310a5eSIlya Dryomov if (ret) { 319505135f59SDavid Sterba btrfs_err(fs_info, "failed to recover balance: %d", ret); 319668310a5eSIlya Dryomov goto fail_block_groups; 319768310a5eSIlya Dryomov } 319868310a5eSIlya Dryomov 3199733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 3200733f4fbbSStefan Behrens if (ret) { 320105135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_stats: %d", ret); 3202733f4fbbSStefan Behrens goto fail_block_groups; 3203733f4fbbSStefan Behrens } 3204733f4fbbSStefan Behrens 32058dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 32068dabb742SStefan Behrens if (ret) { 320705135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_replace: %d", ret); 32088dabb742SStefan Behrens goto fail_block_groups; 32098dabb742SStefan Behrens } 32108dabb742SStefan Behrens 32119b99b115SAnand Jain btrfs_free_extra_devids(fs_devices, 1); 32128dabb742SStefan Behrens 3213c6761a9eSAnand Jain ret = btrfs_sysfs_add_fsid(fs_devices); 3214b7c35e81SAnand Jain if (ret) { 321505135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", 321605135f59SDavid Sterba ret); 3217b7c35e81SAnand Jain goto fail_block_groups; 3218b7c35e81SAnand Jain } 3219b7c35e81SAnand Jain 322096f3136eSAnand Jain ret = btrfs_sysfs_add_mounted(fs_info); 32215ac1d209SJeff Mahoney if (ret) { 322205135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); 3223b7c35e81SAnand Jain goto fail_fsdev_sysfs; 32245ac1d209SJeff Mahoney } 32255ac1d209SJeff Mahoney 3226c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 3227c59021f8Sliubo if (ret) { 322805135f59SDavid Sterba btrfs_err(fs_info, "failed to initialize space info: %d", ret); 32292365dd3cSAnand Jain goto fail_sysfs; 3230c59021f8Sliubo } 3231c59021f8Sliubo 32325b4aacefSJeff Mahoney ret = btrfs_read_block_groups(fs_info); 32331b1d1f66SJosef Bacik if (ret) { 323405135f59SDavid Sterba btrfs_err(fs_info, "failed to read block groups: %d", ret); 32352365dd3cSAnand Jain goto fail_sysfs; 32361b1d1f66SJosef Bacik } 32374330e183SQu Wenruo 32386528b99dSAnand Jain if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) { 323905135f59SDavid Sterba btrfs_warn(fs_info, 324052042d8eSAndrea Gelmini "writable mount is not allowed due to too many missing devices"); 32412365dd3cSAnand Jain goto fail_sysfs; 3242292fd7fcSStefan Behrens } 32439078a3e1SChris Mason 3244a74a4b97SChris Mason fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 3245a74a4b97SChris Mason "btrfs-cleaner"); 324657506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 32472365dd3cSAnand Jain goto fail_sysfs; 3248a74a4b97SChris Mason 3249a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 3250a74a4b97SChris Mason tree_root, 3251a74a4b97SChris Mason "btrfs-transaction"); 325257506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 32533f157a2fSChris Mason goto fail_cleaner; 3254a74a4b97SChris Mason 3255583b7231SHans van Kranenburg if (!btrfs_test_opt(fs_info, NOSSD) && 3256c289811cSChris Mason !fs_info->fs_devices->rotating) { 3257583b7231SHans van Kranenburg btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); 3258c289811cSChris Mason } 3259c289811cSChris Mason 3260572d9ab7SDavid Sterba /* 326101327610SNicholas D Steeves * Mount does not set all options immediately, we can do it now and do 3262572d9ab7SDavid Sterba * not have to wait for transaction commit 3263572d9ab7SDavid Sterba */ 3264572d9ab7SDavid Sterba btrfs_apply_pending_changes(fs_info); 32653818aea2SQu Wenruo 326621adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 32670b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { 32682ff7e61eSJeff Mahoney ret = btrfsic_mount(fs_info, fs_devices, 32690b246afaSJeff Mahoney btrfs_test_opt(fs_info, 327021adbd5cSStefan Behrens CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? 327121adbd5cSStefan Behrens 1 : 0, 327221adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 327321adbd5cSStefan Behrens if (ret) 327405135f59SDavid Sterba btrfs_warn(fs_info, 327505135f59SDavid Sterba "failed to initialize integrity check module: %d", 327605135f59SDavid Sterba ret); 327721adbd5cSStefan Behrens } 327821adbd5cSStefan Behrens #endif 3279bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 3280bcef60f2SArne Jansen if (ret) 3281bcef60f2SArne Jansen goto fail_trans_kthread; 328221adbd5cSStefan Behrens 3283fd708b81SJosef Bacik if (btrfs_build_ref_tree(fs_info)) 3284fd708b81SJosef Bacik btrfs_err(fs_info, "couldn't build ref tree"); 3285fd708b81SJosef Bacik 328696da0919SQu Wenruo /* do not make disk changes in broken FS or nologreplay is given */ 328796da0919SQu Wenruo if (btrfs_super_log_root(disk_super) != 0 && 32880b246afaSJeff Mahoney !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 3289e8294f2fSDavid Sterba btrfs_info(fs_info, "start tree-log replay"); 329063443bf5SEric Sandeen ret = btrfs_replay_log(fs_info, fs_devices); 329179787eaaSJeff Mahoney if (ret) { 329263443bf5SEric Sandeen err = ret; 329328c16cbbSWang Shilong goto fail_qgroup; 3294e556ce2cSYan Zheng } 3295e02119d5SChris Mason } 32961a40e23bSZheng Yan 32976bccf3abSJeff Mahoney ret = btrfs_find_orphan_roots(fs_info); 329879787eaaSJeff Mahoney if (ret) 329928c16cbbSWang Shilong goto fail_qgroup; 330076dda93cSYan, Zheng 3301bc98a42cSDavid Howells if (!sb_rdonly(sb)) { 3302d68fc57bSYan, Zheng ret = btrfs_cleanup_fs_roots(fs_info); 330344c44af2SIlya Dryomov if (ret) 330428c16cbbSWang Shilong goto fail_qgroup; 330590c711abSZygo Blaxell 330690c711abSZygo Blaxell mutex_lock(&fs_info->cleaner_mutex); 33075d4f98a2SYan Zheng ret = btrfs_recover_relocation(tree_root); 330890c711abSZygo Blaxell mutex_unlock(&fs_info->cleaner_mutex); 3309d7ce5843SMiao Xie if (ret < 0) { 331005135f59SDavid Sterba btrfs_warn(fs_info, "failed to recover relocation: %d", 331105135f59SDavid Sterba ret); 3312d7ce5843SMiao Xie err = -EINVAL; 3313bcef60f2SArne Jansen goto fail_qgroup; 3314d7ce5843SMiao Xie } 33157c2ca468SChris Mason } 33161a40e23bSZheng Yan 331756e9357aSDavid Sterba fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true); 33183140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 33193140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 3320f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read fs tree: %d", err); 3321315bf8efSJosef Bacik fs_info->fs_root = NULL; 3322bcef60f2SArne Jansen goto fail_qgroup; 33233140c9a3SDan Carpenter } 3324c289811cSChris Mason 3325bc98a42cSDavid Howells if (sb_rdonly(sb)) 33262b6ba629SIlya Dryomov return 0; 33272b6ba629SIlya Dryomov 3328f8d468a1SOmar Sandoval if (btrfs_test_opt(fs_info, CLEAR_CACHE) && 3329f8d468a1SOmar Sandoval btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 33306675df31SOmar Sandoval clear_free_space_tree = 1; 33316675df31SOmar Sandoval } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 33326675df31SOmar Sandoval !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { 33336675df31SOmar Sandoval btrfs_warn(fs_info, "free space tree is invalid"); 33346675df31SOmar Sandoval clear_free_space_tree = 1; 33356675df31SOmar Sandoval } 33366675df31SOmar Sandoval 33376675df31SOmar Sandoval if (clear_free_space_tree) { 3338f8d468a1SOmar Sandoval btrfs_info(fs_info, "clearing free space tree"); 3339f8d468a1SOmar Sandoval ret = btrfs_clear_free_space_tree(fs_info); 3340f8d468a1SOmar Sandoval if (ret) { 3341f8d468a1SOmar Sandoval btrfs_warn(fs_info, 3342f8d468a1SOmar Sandoval "failed to clear free space tree: %d", ret); 33436bccf3abSJeff Mahoney close_ctree(fs_info); 3344f8d468a1SOmar Sandoval return ret; 3345f8d468a1SOmar Sandoval } 3346f8d468a1SOmar Sandoval } 3347f8d468a1SOmar Sandoval 33480b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && 3349511711afSChris Mason !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 335005135f59SDavid Sterba btrfs_info(fs_info, "creating free space tree"); 3351511711afSChris Mason ret = btrfs_create_free_space_tree(fs_info); 3352511711afSChris Mason if (ret) { 335305135f59SDavid Sterba btrfs_warn(fs_info, 335405135f59SDavid Sterba "failed to create free space tree: %d", ret); 33556bccf3abSJeff Mahoney close_ctree(fs_info); 3356511711afSChris Mason return ret; 3357511711afSChris Mason } 3358511711afSChris Mason } 3359511711afSChris Mason 3360e3acc2a6SJosef Bacik down_read(&fs_info->cleanup_work_sem); 33612b6ba629SIlya Dryomov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 33622b6ba629SIlya Dryomov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 33632b6ba629SIlya Dryomov up_read(&fs_info->cleanup_work_sem); 33646bccf3abSJeff Mahoney close_ctree(fs_info); 33652b6ba629SIlya Dryomov return ret; 33662b6ba629SIlya Dryomov } 3367e3acc2a6SJosef Bacik up_read(&fs_info->cleanup_work_sem); 336859641015SIlya Dryomov 33692b6ba629SIlya Dryomov ret = btrfs_resume_balance_async(fs_info); 33702b6ba629SIlya Dryomov if (ret) { 337105135f59SDavid Sterba btrfs_warn(fs_info, "failed to resume balance: %d", ret); 33726bccf3abSJeff Mahoney close_ctree(fs_info); 33732b6ba629SIlya Dryomov return ret; 3374e3acc2a6SJosef Bacik } 3375e3acc2a6SJosef Bacik 33768dabb742SStefan Behrens ret = btrfs_resume_dev_replace_async(fs_info); 33778dabb742SStefan Behrens if (ret) { 337805135f59SDavid Sterba btrfs_warn(fs_info, "failed to resume device replace: %d", ret); 33796bccf3abSJeff Mahoney close_ctree(fs_info); 33808dabb742SStefan Behrens return ret; 33818dabb742SStefan Behrens } 33828dabb742SStefan Behrens 3383b382a324SJan Schmidt btrfs_qgroup_rescan_resume(fs_info); 3384b0643e59SDennis Zhou btrfs_discard_resume(fs_info); 3385b382a324SJan Schmidt 33864bbcaa64SEric Sandeen if (!fs_info->uuid_root) { 338705135f59SDavid Sterba btrfs_info(fs_info, "creating UUID tree"); 3388f7a81ea4SStefan Behrens ret = btrfs_create_uuid_tree(fs_info); 3389f7a81ea4SStefan Behrens if (ret) { 339005135f59SDavid Sterba btrfs_warn(fs_info, 339105135f59SDavid Sterba "failed to create the UUID tree: %d", ret); 33926bccf3abSJeff Mahoney close_ctree(fs_info); 3393f7a81ea4SStefan Behrens return ret; 3394f7a81ea4SStefan Behrens } 33950b246afaSJeff Mahoney } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || 33964bbcaa64SEric Sandeen fs_info->generation != 33974bbcaa64SEric Sandeen btrfs_super_uuid_tree_generation(disk_super)) { 339805135f59SDavid Sterba btrfs_info(fs_info, "checking UUID tree"); 339970f80175SStefan Behrens ret = btrfs_check_uuid_tree(fs_info); 340070f80175SStefan Behrens if (ret) { 340105135f59SDavid Sterba btrfs_warn(fs_info, 340205135f59SDavid Sterba "failed to check the UUID tree: %d", ret); 34036bccf3abSJeff Mahoney close_ctree(fs_info); 340470f80175SStefan Behrens return ret; 340570f80175SStefan Behrens } 3406f7a81ea4SStefan Behrens } 3407afcdd129SJosef Bacik set_bit(BTRFS_FS_OPEN, &fs_info->flags); 340847ab2a6cSJosef Bacik 3409ad2b2c80SAl Viro return 0; 341039279cc3SChris Mason 3411bcef60f2SArne Jansen fail_qgroup: 3412bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 34137c2ca468SChris Mason fail_trans_kthread: 34147c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 34152ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 3416faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 34173f157a2fSChris Mason fail_cleaner: 3418a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 34197c2ca468SChris Mason 34207c2ca468SChris Mason /* 34217c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 34227c2ca468SChris Mason * kthreads 34237c2ca468SChris Mason */ 34247c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 34257c2ca468SChris Mason 34262365dd3cSAnand Jain fail_sysfs: 34276618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 34282365dd3cSAnand Jain 3429b7c35e81SAnand Jain fail_fsdev_sysfs: 3430b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3431b7c35e81SAnand Jain 34321b1d1f66SJosef Bacik fail_block_groups: 343354067ae9SJosef Bacik btrfs_put_block_group_cache(fs_info); 3434af31f5e5SChris Mason 3435af31f5e5SChris Mason fail_tree_roots: 34369e3aa805SJosef Bacik if (fs_info->data_reloc_root) 34379e3aa805SJosef Bacik btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root); 34384273eaffSAnand Jain free_root_pointers(fs_info, true); 34392b8195bbSMiao Xie invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3440af31f5e5SChris Mason 344139279cc3SChris Mason fail_sb_buffer: 34427abadb64SLiu Bo btrfs_stop_all_workers(fs_info); 34435cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 344416cdcec7SMiao Xie fail_alloc: 3445586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 3446586e46e2SIlya Dryomov 34474543df7eSChris Mason iput(fs_info->btree_inode); 34487e662854SQinghuang Feng fail: 3449586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 3450ad2b2c80SAl Viro return err; 3451eb60ceacSChris Mason } 3452663faf9fSMasami Hiramatsu ALLOW_ERROR_INJECTION(open_ctree, ERRNO); 3453eb60ceacSChris Mason 3454314b6dd0SJohannes Thumshirn static void btrfs_end_super_write(struct bio *bio) 3455f2984462SChris Mason { 3456314b6dd0SJohannes Thumshirn struct btrfs_device *device = bio->bi_private; 3457314b6dd0SJohannes Thumshirn struct bio_vec *bvec; 3458314b6dd0SJohannes Thumshirn struct bvec_iter_all iter_all; 3459314b6dd0SJohannes Thumshirn struct page *page; 3460442a4f63SStefan Behrens 3461314b6dd0SJohannes Thumshirn bio_for_each_segment_all(bvec, bio, iter_all) { 3462314b6dd0SJohannes Thumshirn page = bvec->bv_page; 3463314b6dd0SJohannes Thumshirn 3464314b6dd0SJohannes Thumshirn if (bio->bi_status) { 3465fb456252SJeff Mahoney btrfs_warn_rl_in_rcu(device->fs_info, 3466314b6dd0SJohannes Thumshirn "lost page write due to IO error on %s (%d)", 3467314b6dd0SJohannes Thumshirn rcu_str_deref(device->name), 3468314b6dd0SJohannes Thumshirn blk_status_to_errno(bio->bi_status)); 3469314b6dd0SJohannes Thumshirn ClearPageUptodate(page); 3470314b6dd0SJohannes Thumshirn SetPageError(page); 3471314b6dd0SJohannes Thumshirn btrfs_dev_stat_inc_and_print(device, 3472314b6dd0SJohannes Thumshirn BTRFS_DEV_STAT_WRITE_ERRS); 3473314b6dd0SJohannes Thumshirn } else { 3474314b6dd0SJohannes Thumshirn SetPageUptodate(page); 3475f2984462SChris Mason } 3476314b6dd0SJohannes Thumshirn 3477314b6dd0SJohannes Thumshirn put_page(page); 3478314b6dd0SJohannes Thumshirn unlock_page(page); 3479314b6dd0SJohannes Thumshirn } 3480314b6dd0SJohannes Thumshirn 3481314b6dd0SJohannes Thumshirn bio_put(bio); 3482f2984462SChris Mason } 3483f2984462SChris Mason 34848f32380dSJohannes Thumshirn struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev, 34858f32380dSJohannes Thumshirn int copy_num) 348629c36d72SAnand Jain { 348729c36d72SAnand Jain struct btrfs_super_block *super; 34888f32380dSJohannes Thumshirn struct page *page; 348929c36d72SAnand Jain u64 bytenr; 34908f32380dSJohannes Thumshirn struct address_space *mapping = bdev->bd_inode->i_mapping; 349129c36d72SAnand Jain 349229c36d72SAnand Jain bytenr = btrfs_sb_offset(copy_num); 349329c36d72SAnand Jain if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode)) 34948f32380dSJohannes Thumshirn return ERR_PTR(-EINVAL); 349529c36d72SAnand Jain 34968f32380dSJohannes Thumshirn page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS); 34978f32380dSJohannes Thumshirn if (IS_ERR(page)) 34988f32380dSJohannes Thumshirn return ERR_CAST(page); 349929c36d72SAnand Jain 35008f32380dSJohannes Thumshirn super = page_address(page); 350196c2e067SAnand Jain if (btrfs_super_magic(super) != BTRFS_MAGIC) { 350296c2e067SAnand Jain btrfs_release_disk_super(super); 350396c2e067SAnand Jain return ERR_PTR(-ENODATA); 350496c2e067SAnand Jain } 350596c2e067SAnand Jain 350696c2e067SAnand Jain if (btrfs_super_bytenr(super) != bytenr) { 35078f32380dSJohannes Thumshirn btrfs_release_disk_super(super); 35088f32380dSJohannes Thumshirn return ERR_PTR(-EINVAL); 350929c36d72SAnand Jain } 351029c36d72SAnand Jain 35118f32380dSJohannes Thumshirn return super; 351229c36d72SAnand Jain } 351329c36d72SAnand Jain 351429c36d72SAnand Jain 35158f32380dSJohannes Thumshirn struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev) 3516a512bbf8SYan Zheng { 35178f32380dSJohannes Thumshirn struct btrfs_super_block *super, *latest = NULL; 3518a512bbf8SYan Zheng int i; 3519a512bbf8SYan Zheng u64 transid = 0; 3520a512bbf8SYan Zheng 3521a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 3522a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 3523a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 3524a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3525a512bbf8SYan Zheng */ 3526a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 35278f32380dSJohannes Thumshirn super = btrfs_read_dev_one_super(bdev, i); 35288f32380dSJohannes Thumshirn if (IS_ERR(super)) 3529a512bbf8SYan Zheng continue; 3530a512bbf8SYan Zheng 3531a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 35328f32380dSJohannes Thumshirn if (latest) 35338f32380dSJohannes Thumshirn btrfs_release_disk_super(super); 35348f32380dSJohannes Thumshirn 35358f32380dSJohannes Thumshirn latest = super; 3536a512bbf8SYan Zheng transid = btrfs_super_generation(super); 3537a512bbf8SYan Zheng } 3538a512bbf8SYan Zheng } 353992fc03fbSAnand Jain 35408f32380dSJohannes Thumshirn return super; 3541a512bbf8SYan Zheng } 3542a512bbf8SYan Zheng 35434eedeb75SHisashi Hifumi /* 3544abbb3b8eSDavid Sterba * Write superblock @sb to the @device. Do not wait for completion, all the 3545314b6dd0SJohannes Thumshirn * pages we use for writing are locked. 35464eedeb75SHisashi Hifumi * 3547abbb3b8eSDavid Sterba * Write @max_mirrors copies of the superblock, where 0 means default that fit 3548abbb3b8eSDavid Sterba * the expected device size at commit time. Note that max_mirrors must be 3549abbb3b8eSDavid Sterba * same for write and wait phases. 35504eedeb75SHisashi Hifumi * 3551314b6dd0SJohannes Thumshirn * Return number of errors when page is not found or submission fails. 35524eedeb75SHisashi Hifumi */ 3553a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 3554abbb3b8eSDavid Sterba struct btrfs_super_block *sb, int max_mirrors) 3555a512bbf8SYan Zheng { 3556d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = device->fs_info; 3557314b6dd0SJohannes Thumshirn struct address_space *mapping = device->bdev->bd_inode->i_mapping; 3558d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 3559a512bbf8SYan Zheng int i; 3560a512bbf8SYan Zheng int errors = 0; 3561a512bbf8SYan Zheng u64 bytenr; 3562a512bbf8SYan Zheng 3563a512bbf8SYan Zheng if (max_mirrors == 0) 3564a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3565a512bbf8SYan Zheng 3566d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 3567d5178578SJohannes Thumshirn 3568a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 3569314b6dd0SJohannes Thumshirn struct page *page; 3570314b6dd0SJohannes Thumshirn struct bio *bio; 3571314b6dd0SJohannes Thumshirn struct btrfs_super_block *disk_super; 3572314b6dd0SJohannes Thumshirn 3573a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 3574935e5cc9SMiao Xie if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3575935e5cc9SMiao Xie device->commit_total_bytes) 3576a512bbf8SYan Zheng break; 3577a512bbf8SYan Zheng 3578a512bbf8SYan Zheng btrfs_set_super_bytenr(sb, bytenr); 3579a512bbf8SYan Zheng 3580fd08001fSEric Biggers crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE, 3581fd08001fSEric Biggers BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, 3582fd08001fSEric Biggers sb->csum); 3583a512bbf8SYan Zheng 3584314b6dd0SJohannes Thumshirn page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT, 3585314b6dd0SJohannes Thumshirn GFP_NOFS); 3586314b6dd0SJohannes Thumshirn if (!page) { 3587fb456252SJeff Mahoney btrfs_err(device->fs_info, 3588314b6dd0SJohannes Thumshirn "couldn't get super block page for bytenr %llu", 3589f14d104dSDavid Sterba bytenr); 3590634554dcSJosef Bacik errors++; 3591634554dcSJosef Bacik continue; 3592634554dcSJosef Bacik } 3593634554dcSJosef Bacik 3594314b6dd0SJohannes Thumshirn /* Bump the refcount for wait_dev_supers() */ 3595314b6dd0SJohannes Thumshirn get_page(page); 3596a512bbf8SYan Zheng 3597314b6dd0SJohannes Thumshirn disk_super = page_address(page); 3598314b6dd0SJohannes Thumshirn memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE); 3599a512bbf8SYan Zheng 3600387125fcSChris Mason /* 3601314b6dd0SJohannes Thumshirn * Directly use bios here instead of relying on the page cache 3602314b6dd0SJohannes Thumshirn * to do I/O, so we don't lose the ability to do integrity 3603314b6dd0SJohannes Thumshirn * checking. 3604387125fcSChris Mason */ 3605314b6dd0SJohannes Thumshirn bio = bio_alloc(GFP_NOFS, 1); 3606314b6dd0SJohannes Thumshirn bio_set_dev(bio, device->bdev); 3607314b6dd0SJohannes Thumshirn bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT; 3608314b6dd0SJohannes Thumshirn bio->bi_private = device; 3609314b6dd0SJohannes Thumshirn bio->bi_end_io = btrfs_end_super_write; 3610314b6dd0SJohannes Thumshirn __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE, 3611314b6dd0SJohannes Thumshirn offset_in_page(bytenr)); 3612314b6dd0SJohannes Thumshirn 3613314b6dd0SJohannes Thumshirn /* 3614314b6dd0SJohannes Thumshirn * We FUA only the first super block. The others we allow to 3615314b6dd0SJohannes Thumshirn * go down lazy and there's a short window where the on-disk 3616314b6dd0SJohannes Thumshirn * copies might still contain the older version. 3617314b6dd0SJohannes Thumshirn */ 3618314b6dd0SJohannes Thumshirn bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO; 36191b9e619cSOmar Sandoval if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) 3620314b6dd0SJohannes Thumshirn bio->bi_opf |= REQ_FUA; 3621314b6dd0SJohannes Thumshirn 3622314b6dd0SJohannes Thumshirn btrfsic_submit_bio(bio); 3623a512bbf8SYan Zheng } 3624a512bbf8SYan Zheng return errors < i ? 0 : -1; 3625a512bbf8SYan Zheng } 3626a512bbf8SYan Zheng 3627387125fcSChris Mason /* 3628abbb3b8eSDavid Sterba * Wait for write completion of superblocks done by write_dev_supers, 3629abbb3b8eSDavid Sterba * @max_mirrors same for write and wait phases. 3630abbb3b8eSDavid Sterba * 3631314b6dd0SJohannes Thumshirn * Return number of errors when page is not found or not marked up to 3632abbb3b8eSDavid Sterba * date. 3633abbb3b8eSDavid Sterba */ 3634abbb3b8eSDavid Sterba static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) 3635abbb3b8eSDavid Sterba { 3636abbb3b8eSDavid Sterba int i; 3637abbb3b8eSDavid Sterba int errors = 0; 3638b6a535faSHoward McLauchlan bool primary_failed = false; 3639abbb3b8eSDavid Sterba u64 bytenr; 3640abbb3b8eSDavid Sterba 3641abbb3b8eSDavid Sterba if (max_mirrors == 0) 3642abbb3b8eSDavid Sterba max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3643abbb3b8eSDavid Sterba 3644abbb3b8eSDavid Sterba for (i = 0; i < max_mirrors; i++) { 3645314b6dd0SJohannes Thumshirn struct page *page; 3646314b6dd0SJohannes Thumshirn 3647abbb3b8eSDavid Sterba bytenr = btrfs_sb_offset(i); 3648abbb3b8eSDavid Sterba if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3649abbb3b8eSDavid Sterba device->commit_total_bytes) 3650abbb3b8eSDavid Sterba break; 3651abbb3b8eSDavid Sterba 3652314b6dd0SJohannes Thumshirn page = find_get_page(device->bdev->bd_inode->i_mapping, 3653314b6dd0SJohannes Thumshirn bytenr >> PAGE_SHIFT); 3654314b6dd0SJohannes Thumshirn if (!page) { 3655abbb3b8eSDavid Sterba errors++; 3656b6a535faSHoward McLauchlan if (i == 0) 3657b6a535faSHoward McLauchlan primary_failed = true; 3658abbb3b8eSDavid Sterba continue; 3659abbb3b8eSDavid Sterba } 3660314b6dd0SJohannes Thumshirn /* Page is submitted locked and unlocked once the IO completes */ 3661314b6dd0SJohannes Thumshirn wait_on_page_locked(page); 3662314b6dd0SJohannes Thumshirn if (PageError(page)) { 3663abbb3b8eSDavid Sterba errors++; 3664b6a535faSHoward McLauchlan if (i == 0) 3665b6a535faSHoward McLauchlan primary_failed = true; 3666b6a535faSHoward McLauchlan } 3667abbb3b8eSDavid Sterba 3668314b6dd0SJohannes Thumshirn /* Drop our reference */ 3669314b6dd0SJohannes Thumshirn put_page(page); 3670abbb3b8eSDavid Sterba 3671314b6dd0SJohannes Thumshirn /* Drop the reference from the writing run */ 3672314b6dd0SJohannes Thumshirn put_page(page); 3673abbb3b8eSDavid Sterba } 3674abbb3b8eSDavid Sterba 3675b6a535faSHoward McLauchlan /* log error, force error return */ 3676b6a535faSHoward McLauchlan if (primary_failed) { 3677b6a535faSHoward McLauchlan btrfs_err(device->fs_info, "error writing primary super block to device %llu", 3678b6a535faSHoward McLauchlan device->devid); 3679b6a535faSHoward McLauchlan return -1; 3680b6a535faSHoward McLauchlan } 3681b6a535faSHoward McLauchlan 3682abbb3b8eSDavid Sterba return errors < i ? 0 : -1; 3683abbb3b8eSDavid Sterba } 3684abbb3b8eSDavid Sterba 3685abbb3b8eSDavid Sterba /* 3686387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 3687387125fcSChris Mason * for the barrier when it is done 3688387125fcSChris Mason */ 36894246a0b6SChristoph Hellwig static void btrfs_end_empty_barrier(struct bio *bio) 3690387125fcSChris Mason { 3691387125fcSChris Mason complete(bio->bi_private); 3692387125fcSChris Mason } 3693387125fcSChris Mason 3694387125fcSChris Mason /* 36954fc6441aSAnand Jain * Submit a flush request to the device if it supports it. Error handling is 36964fc6441aSAnand Jain * done in the waiting counterpart. 3697387125fcSChris Mason */ 36984fc6441aSAnand Jain static void write_dev_flush(struct btrfs_device *device) 3699387125fcSChris Mason { 3700c2a9c7abSAnand Jain struct request_queue *q = bdev_get_queue(device->bdev); 3701e0ae9994SDavid Sterba struct bio *bio = device->flush_bio; 3702387125fcSChris Mason 3703c2a9c7abSAnand Jain if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) 37044fc6441aSAnand Jain return; 3705387125fcSChris Mason 3706e0ae9994SDavid Sterba bio_reset(bio); 3707387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 370874d46992SChristoph Hellwig bio_set_dev(bio, device->bdev); 37098d910125SJan Kara bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH; 3710387125fcSChris Mason init_completion(&device->flush_wait); 3711387125fcSChris Mason bio->bi_private = &device->flush_wait; 3712387125fcSChris Mason 371343a01111SLu Fengqi btrfsic_submit_bio(bio); 37141c3063b6SAnand Jain set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 37154fc6441aSAnand Jain } 3716387125fcSChris Mason 37174fc6441aSAnand Jain /* 37184fc6441aSAnand Jain * If the flush bio has been submitted by write_dev_flush, wait for it. 37194fc6441aSAnand Jain */ 37208c27cb35SLinus Torvalds static blk_status_t wait_dev_flush(struct btrfs_device *device) 37214fc6441aSAnand Jain { 37224fc6441aSAnand Jain struct bio *bio = device->flush_bio; 37234fc6441aSAnand Jain 37241c3063b6SAnand Jain if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) 372558efbc9fSOmar Sandoval return BLK_STS_OK; 37264fc6441aSAnand Jain 37271c3063b6SAnand Jain clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 37282980d574SDavid Sterba wait_for_completion_io(&device->flush_wait); 37294fc6441aSAnand Jain 37308c27cb35SLinus Torvalds return bio->bi_status; 3731387125fcSChris Mason } 3732387125fcSChris Mason 3733d10b82feSQu Wenruo static int check_barrier_error(struct btrfs_fs_info *fs_info) 3734401b41e5SAnand Jain { 37356528b99dSAnand Jain if (!btrfs_check_rw_degradable(fs_info, NULL)) 3736401b41e5SAnand Jain return -EIO; 3737387125fcSChris Mason return 0; 3738387125fcSChris Mason } 3739387125fcSChris Mason 3740387125fcSChris Mason /* 3741387125fcSChris Mason * send an empty flush down to each device in parallel, 3742387125fcSChris Mason * then wait for them 3743387125fcSChris Mason */ 3744387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 3745387125fcSChris Mason { 3746387125fcSChris Mason struct list_head *head; 3747387125fcSChris Mason struct btrfs_device *dev; 37485af3e8ccSStefan Behrens int errors_wait = 0; 37494e4cbee9SChristoph Hellwig blk_status_t ret; 3750387125fcSChris Mason 37511538e6c5SDavid Sterba lockdep_assert_held(&info->fs_devices->device_list_mutex); 3752387125fcSChris Mason /* send down all the barriers */ 3753387125fcSChris Mason head = &info->fs_devices->devices; 37541538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3755e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3756f88ba6a2SHidetoshi Seto continue; 3757cea7c8bfSAnand Jain if (!dev->bdev) 3758387125fcSChris Mason continue; 3759e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3760ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3761387125fcSChris Mason continue; 3762387125fcSChris Mason 37634fc6441aSAnand Jain write_dev_flush(dev); 376458efbc9fSOmar Sandoval dev->last_flush_error = BLK_STS_OK; 3765387125fcSChris Mason } 3766387125fcSChris Mason 3767387125fcSChris Mason /* wait for all the barriers */ 37681538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3769e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3770f88ba6a2SHidetoshi Seto continue; 3771387125fcSChris Mason if (!dev->bdev) { 37725af3e8ccSStefan Behrens errors_wait++; 3773387125fcSChris Mason continue; 3774387125fcSChris Mason } 3775e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3776ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3777387125fcSChris Mason continue; 3778387125fcSChris Mason 37794fc6441aSAnand Jain ret = wait_dev_flush(dev); 3780401b41e5SAnand Jain if (ret) { 3781401b41e5SAnand Jain dev->last_flush_error = ret; 378266b4993eSDavid Sterba btrfs_dev_stat_inc_and_print(dev, 378366b4993eSDavid Sterba BTRFS_DEV_STAT_FLUSH_ERRS); 37845af3e8ccSStefan Behrens errors_wait++; 3785387125fcSChris Mason } 3786401b41e5SAnand Jain } 3787401b41e5SAnand Jain 3788cea7c8bfSAnand Jain if (errors_wait) { 3789401b41e5SAnand Jain /* 3790401b41e5SAnand Jain * At some point we need the status of all disks 3791401b41e5SAnand Jain * to arrive at the volume status. So error checking 3792401b41e5SAnand Jain * is being pushed to a separate loop. 3793401b41e5SAnand Jain */ 3794d10b82feSQu Wenruo return check_barrier_error(info); 3795401b41e5SAnand Jain } 3796387125fcSChris Mason return 0; 3797387125fcSChris Mason } 3798387125fcSChris Mason 3799943c6e99SZhao Lei int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 3800943c6e99SZhao Lei { 38018789f4feSZhao Lei int raid_type; 38028789f4feSZhao Lei int min_tolerated = INT_MAX; 3803943c6e99SZhao Lei 38048789f4feSZhao Lei if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 38058789f4feSZhao Lei (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 38068c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 38078789f4feSZhao Lei btrfs_raid_array[BTRFS_RAID_SINGLE]. 38088789f4feSZhao Lei tolerated_failures); 3809943c6e99SZhao Lei 38108789f4feSZhao Lei for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 38118789f4feSZhao Lei if (raid_type == BTRFS_RAID_SINGLE) 38128789f4feSZhao Lei continue; 381341a6e891SAnand Jain if (!(flags & btrfs_raid_array[raid_type].bg_flag)) 38148789f4feSZhao Lei continue; 38158c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 38168789f4feSZhao Lei btrfs_raid_array[raid_type]. 38178789f4feSZhao Lei tolerated_failures); 38188789f4feSZhao Lei } 3819943c6e99SZhao Lei 38208789f4feSZhao Lei if (min_tolerated == INT_MAX) { 3821ab8d0fc4SJeff Mahoney pr_warn("BTRFS: unknown raid flag: %llu", flags); 38228789f4feSZhao Lei min_tolerated = 0; 38238789f4feSZhao Lei } 38248789f4feSZhao Lei 38258789f4feSZhao Lei return min_tolerated; 3826943c6e99SZhao Lei } 3827943c6e99SZhao Lei 3828eece6a9cSDavid Sterba int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) 3829f2984462SChris Mason { 3830e5e9a520SChris Mason struct list_head *head; 3831f2984462SChris Mason struct btrfs_device *dev; 3832a061fc8dSChris Mason struct btrfs_super_block *sb; 3833f2984462SChris Mason struct btrfs_dev_item *dev_item; 3834f2984462SChris Mason int ret; 3835f2984462SChris Mason int do_barriers; 3836a236aed1SChris Mason int max_errors; 3837a236aed1SChris Mason int total_errors = 0; 3838a061fc8dSChris Mason u64 flags; 3839f2984462SChris Mason 38400b246afaSJeff Mahoney do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); 3841fed3b381SLiu Bo 3842fed3b381SLiu Bo /* 3843fed3b381SLiu Bo * max_mirrors == 0 indicates we're from commit_transaction, 3844fed3b381SLiu Bo * not from fsync where the tree roots in fs_info have not 3845fed3b381SLiu Bo * been consistent on disk. 3846fed3b381SLiu Bo */ 3847fed3b381SLiu Bo if (max_mirrors == 0) 38480b246afaSJeff Mahoney backup_super_roots(fs_info); 3849f2984462SChris Mason 38500b246afaSJeff Mahoney sb = fs_info->super_for_commit; 3851a061fc8dSChris Mason dev_item = &sb->dev_item; 3852e5e9a520SChris Mason 38530b246afaSJeff Mahoney mutex_lock(&fs_info->fs_devices->device_list_mutex); 38540b246afaSJeff Mahoney head = &fs_info->fs_devices->devices; 38550b246afaSJeff Mahoney max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; 3856387125fcSChris Mason 38575af3e8ccSStefan Behrens if (do_barriers) { 38580b246afaSJeff Mahoney ret = barrier_all_devices(fs_info); 38595af3e8ccSStefan Behrens if (ret) { 38605af3e8ccSStefan Behrens mutex_unlock( 38610b246afaSJeff Mahoney &fs_info->fs_devices->device_list_mutex); 38620b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 38635af3e8ccSStefan Behrens "errors while submitting device barriers."); 38645af3e8ccSStefan Behrens return ret; 38655af3e8ccSStefan Behrens } 38665af3e8ccSStefan Behrens } 3867387125fcSChris Mason 38681538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3869dfe25020SChris Mason if (!dev->bdev) { 3870dfe25020SChris Mason total_errors++; 3871dfe25020SChris Mason continue; 3872dfe25020SChris Mason } 3873e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3874ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3875dfe25020SChris Mason continue; 3876dfe25020SChris Mason 38772b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 3878a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 3879a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 38807df69d3eSMiao Xie btrfs_set_stack_device_total_bytes(dev_item, 3881935e5cc9SMiao Xie dev->commit_total_bytes); 3882ce7213c7SMiao Xie btrfs_set_stack_device_bytes_used(dev_item, 3883ce7213c7SMiao Xie dev->commit_bytes_used); 3884a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 3885a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 3886a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 3887a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 38887239ff4bSNikolay Borisov memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, 38897239ff4bSNikolay Borisov BTRFS_FSID_SIZE); 3890a512bbf8SYan Zheng 3891a061fc8dSChris Mason flags = btrfs_super_flags(sb); 3892a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 3893f2984462SChris Mason 389475cb857dSQu Wenruo ret = btrfs_validate_write_super(fs_info, sb); 389575cb857dSQu Wenruo if (ret < 0) { 389675cb857dSQu Wenruo mutex_unlock(&fs_info->fs_devices->device_list_mutex); 389775cb857dSQu Wenruo btrfs_handle_fs_error(fs_info, -EUCLEAN, 389875cb857dSQu Wenruo "unexpected superblock corruption detected"); 389975cb857dSQu Wenruo return -EUCLEAN; 390075cb857dSQu Wenruo } 390175cb857dSQu Wenruo 3902abbb3b8eSDavid Sterba ret = write_dev_supers(dev, sb, max_mirrors); 3903a236aed1SChris Mason if (ret) 3904a236aed1SChris Mason total_errors++; 3905f2984462SChris Mason } 3906a236aed1SChris Mason if (total_errors > max_errors) { 39070b246afaSJeff Mahoney btrfs_err(fs_info, "%d errors while writing supers", 3908d397712bSChris Mason total_errors); 39090b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 391079787eaaSJeff Mahoney 39119d565ba4SStefan Behrens /* FUA is masked off if unsupported and can't be the reason */ 39120b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 39130b246afaSJeff Mahoney "%d errors while writing supers", 39140b246afaSJeff Mahoney total_errors); 39159d565ba4SStefan Behrens return -EIO; 3916a236aed1SChris Mason } 3917f2984462SChris Mason 3918a512bbf8SYan Zheng total_errors = 0; 39191538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3920dfe25020SChris Mason if (!dev->bdev) 3921dfe25020SChris Mason continue; 3922e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3923ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3924dfe25020SChris Mason continue; 3925dfe25020SChris Mason 3926abbb3b8eSDavid Sterba ret = wait_dev_supers(dev, max_mirrors); 3927a512bbf8SYan Zheng if (ret) 39281259ab75SChris Mason total_errors++; 3929f2984462SChris Mason } 39300b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 3931a236aed1SChris Mason if (total_errors > max_errors) { 39320b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 39330b246afaSJeff Mahoney "%d errors while writing supers", 39340b246afaSJeff Mahoney total_errors); 393579787eaaSJeff Mahoney return -EIO; 3936a236aed1SChris Mason } 3937f2984462SChris Mason return 0; 3938f2984462SChris Mason } 3939f2984462SChris Mason 3940cb517eabSMiao Xie /* Drop a fs root from the radix tree and free it. */ 3941cb517eabSMiao Xie void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 3942cb517eabSMiao Xie struct btrfs_root *root) 39432619ba1fSChris Mason { 39444785e24fSJosef Bacik bool drop_ref = false; 39454785e24fSJosef Bacik 39464df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 39472619ba1fSChris Mason radix_tree_delete(&fs_info->fs_roots_radix, 39482619ba1fSChris Mason (unsigned long)root->root_key.objectid); 3949af01d2e5SJosef Bacik if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state)) 39504785e24fSJosef Bacik drop_ref = true; 39514df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 395276dda93cSYan, Zheng 39531c1ea4f7SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 3954ef67963dSJosef Bacik ASSERT(root->log_root == NULL); 39551c1ea4f7SLiu Bo if (root->reloc_root) { 395600246528SJosef Bacik btrfs_put_root(root->reloc_root); 39571c1ea4f7SLiu Bo root->reloc_root = NULL; 39581c1ea4f7SLiu Bo } 39591c1ea4f7SLiu Bo } 39603321719eSLiu Bo 3961faa2dbf0SJosef Bacik if (root->free_ino_pinned) 3962581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_pinned); 3963faa2dbf0SJosef Bacik if (root->free_ino_ctl) 3964581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_ctl); 39650e996e7fSJosef Bacik if (root->ino_cache_inode) { 39660e996e7fSJosef Bacik iput(root->ino_cache_inode); 39670e996e7fSJosef Bacik root->ino_cache_inode = NULL; 39680e996e7fSJosef Bacik } 39694785e24fSJosef Bacik if (drop_ref) 397000246528SJosef Bacik btrfs_put_root(root); 39712619ba1fSChris Mason } 39722619ba1fSChris Mason 3973c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 3974c146afadSYan Zheng { 3975c146afadSYan Zheng u64 root_objectid = 0; 3976c146afadSYan Zheng struct btrfs_root *gang[8]; 397765d33fd7SQu Wenruo int i = 0; 397865d33fd7SQu Wenruo int err = 0; 397965d33fd7SQu Wenruo unsigned int ret = 0; 3980c146afadSYan Zheng 3981c146afadSYan Zheng while (1) { 3982efc34534SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 3983c146afadSYan Zheng ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 3984c146afadSYan Zheng (void **)gang, root_objectid, 3985c146afadSYan Zheng ARRAY_SIZE(gang)); 398665d33fd7SQu Wenruo if (!ret) { 3987efc34534SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 3988c146afadSYan Zheng break; 398965d33fd7SQu Wenruo } 39905d4f98a2SYan Zheng root_objectid = gang[ret - 1]->root_key.objectid + 1; 399166b4ffd1SJosef Bacik 399265d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 399365d33fd7SQu Wenruo /* Avoid to grab roots in dead_roots */ 399465d33fd7SQu Wenruo if (btrfs_root_refs(&gang[i]->root_item) == 0) { 399565d33fd7SQu Wenruo gang[i] = NULL; 399665d33fd7SQu Wenruo continue; 399765d33fd7SQu Wenruo } 399865d33fd7SQu Wenruo /* grab all the search result for later use */ 399900246528SJosef Bacik gang[i] = btrfs_grab_root(gang[i]); 400065d33fd7SQu Wenruo } 4001efc34534SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 400265d33fd7SQu Wenruo 400365d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 400465d33fd7SQu Wenruo if (!gang[i]) 400565d33fd7SQu Wenruo continue; 4006c146afadSYan Zheng root_objectid = gang[i]->root_key.objectid; 400766b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(gang[i]); 400866b4ffd1SJosef Bacik if (err) 400965d33fd7SQu Wenruo break; 401000246528SJosef Bacik btrfs_put_root(gang[i]); 4011c146afadSYan Zheng } 4012c146afadSYan Zheng root_objectid++; 4013c146afadSYan Zheng } 401465d33fd7SQu Wenruo 401565d33fd7SQu Wenruo /* release the uncleaned roots due to error */ 401665d33fd7SQu Wenruo for (; i < ret; i++) { 401765d33fd7SQu Wenruo if (gang[i]) 401800246528SJosef Bacik btrfs_put_root(gang[i]); 401965d33fd7SQu Wenruo } 402065d33fd7SQu Wenruo return err; 4021c146afadSYan Zheng } 4022c146afadSYan Zheng 40236bccf3abSJeff Mahoney int btrfs_commit_super(struct btrfs_fs_info *fs_info) 4024c146afadSYan Zheng { 40256bccf3abSJeff Mahoney struct btrfs_root *root = fs_info->tree_root; 4026c146afadSYan Zheng struct btrfs_trans_handle *trans; 4027c146afadSYan Zheng 40280b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 40292ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 40300b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 40310b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 4032c71bf099SYan, Zheng 4033c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 40340b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 40350b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4036c71bf099SYan, Zheng 40377a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 40383612b495STsutomu Itoh if (IS_ERR(trans)) 40393612b495STsutomu Itoh return PTR_ERR(trans); 40403a45bb20SJeff Mahoney return btrfs_commit_transaction(trans); 4041c146afadSYan Zheng } 4042c146afadSYan Zheng 4043b105e927SDavid Sterba void __cold close_ctree(struct btrfs_fs_info *fs_info) 4044eb60ceacSChris Mason { 4045c146afadSYan Zheng int ret; 4046e089f05cSChris Mason 4047afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); 4048d6fd0ae2SOmar Sandoval /* 4049d6fd0ae2SOmar Sandoval * We don't want the cleaner to start new transactions, add more delayed 4050d6fd0ae2SOmar Sandoval * iputs, etc. while we're closing. We can't use kthread_stop() yet 4051d6fd0ae2SOmar Sandoval * because that frees the task_struct, and the transaction kthread might 4052d6fd0ae2SOmar Sandoval * still try to wake up the cleaner. 4053d6fd0ae2SOmar Sandoval */ 4054d6fd0ae2SOmar Sandoval kthread_park(fs_info->cleaner_kthread); 4055a2135011SChris Mason 40567343dd61SJustin Maggard /* wait for the qgroup rescan worker to stop */ 4057d06f23d6SJeff Mahoney btrfs_qgroup_wait_for_completion(fs_info, false); 40587343dd61SJustin Maggard 4059803b2f54SStefan Behrens /* wait for the uuid_scan task to finish */ 4060803b2f54SStefan Behrens down(&fs_info->uuid_tree_rescan_sem); 4061803b2f54SStefan Behrens /* avoid complains from lockdep et al., set sem back to initial state */ 4062803b2f54SStefan Behrens up(&fs_info->uuid_tree_rescan_sem); 4063803b2f54SStefan Behrens 4064837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 4065aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 4066837d5b6eSIlya Dryomov 40678dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 40688dabb742SStefan Behrens 4069aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 40704cb5300bSChris Mason 40714cb5300bSChris Mason /* wait for any defraggers to finish */ 40724cb5300bSChris Mason wait_event(fs_info->transaction_wait, 40734cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 40744cb5300bSChris Mason 40754cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 407626176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 40774cb5300bSChris Mason 407821c7e756SMiao Xie cancel_work_sync(&fs_info->async_reclaim_work); 407957056740SJosef Bacik cancel_work_sync(&fs_info->async_data_reclaim_work); 408021c7e756SMiao Xie 4081b0643e59SDennis Zhou /* Cancel or finish ongoing discard work */ 4082b0643e59SDennis Zhou btrfs_discard_cleanup(fs_info); 4083b0643e59SDennis Zhou 4084bc98a42cSDavid Howells if (!sb_rdonly(fs_info->sb)) { 4085e44163e1SJeff Mahoney /* 4086d6fd0ae2SOmar Sandoval * The cleaner kthread is stopped, so do one final pass over 4087d6fd0ae2SOmar Sandoval * unused block groups. 4088e44163e1SJeff Mahoney */ 40890b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 4090e44163e1SJeff Mahoney 4091f0cc2cd7SFilipe Manana /* 4092f0cc2cd7SFilipe Manana * There might be existing delayed inode workers still running 4093f0cc2cd7SFilipe Manana * and holding an empty delayed inode item. We must wait for 4094f0cc2cd7SFilipe Manana * them to complete first because they can create a transaction. 4095f0cc2cd7SFilipe Manana * This happens when someone calls btrfs_balance_delayed_items() 4096f0cc2cd7SFilipe Manana * and then a transaction commit runs the same delayed nodes 4097f0cc2cd7SFilipe Manana * before any delayed worker has done something with the nodes. 4098f0cc2cd7SFilipe Manana * We must wait for any worker here and not at transaction 4099f0cc2cd7SFilipe Manana * commit time since that could cause a deadlock. 4100f0cc2cd7SFilipe Manana * This is a very rare case. 4101f0cc2cd7SFilipe Manana */ 4102f0cc2cd7SFilipe Manana btrfs_flush_workqueue(fs_info->delayed_workers); 4103f0cc2cd7SFilipe Manana 41046bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 4105d397712bSChris Mason if (ret) 410604892340SEric Sandeen btrfs_err(fs_info, "commit super ret %d", ret); 4107c146afadSYan Zheng } 4108ed2ff2cbSChris Mason 4109af722733SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) || 4110af722733SLiu Bo test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) 41112ff7e61eSJeff Mahoney btrfs_error_commit_super(fs_info); 4112acce952bSliubo 4113e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 4114e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 41158929ecfaSYan, Zheng 4116e187831eSJosef Bacik ASSERT(list_empty(&fs_info->delayed_iputs)); 4117afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); 4118f25784b3SYan Zheng 41195958253cSQu Wenruo if (btrfs_check_quota_leak(fs_info)) { 41205958253cSQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 41215958253cSQu Wenruo btrfs_err(fs_info, "qgroup reserved space leaked"); 41225958253cSQu Wenruo } 41235958253cSQu Wenruo 412404892340SEric Sandeen btrfs_free_qgroup_config(fs_info); 4125fe816d0fSNikolay Borisov ASSERT(list_empty(&fs_info->delalloc_roots)); 4126bcef60f2SArne Jansen 4127963d678bSMiao Xie if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 412804892340SEric Sandeen btrfs_info(fs_info, "at unmount delalloc count %lld", 4129963d678bSMiao Xie percpu_counter_sum(&fs_info->delalloc_bytes)); 4130b0c68f8bSChris Mason } 413131153d81SYan Zheng 41324297ff84SJosef Bacik if (percpu_counter_sum(&fs_info->dio_bytes)) 41334297ff84SJosef Bacik btrfs_info(fs_info, "at unmount dio bytes count %lld", 41344297ff84SJosef Bacik percpu_counter_sum(&fs_info->dio_bytes)); 41354297ff84SJosef Bacik 41366618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 4137b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 41385ac1d209SJeff Mahoney 41391a4319ccSLiu Bo btrfs_put_block_group_cache(fs_info); 41401a4319ccSLiu Bo 4141de348ee0SWang Shilong /* 4142de348ee0SWang Shilong * we must make sure there is not any read request to 4143de348ee0SWang Shilong * submit after we stopping all workers. 4144de348ee0SWang Shilong */ 4145de348ee0SWang Shilong invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 414696192499SJosef Bacik btrfs_stop_all_workers(fs_info); 414796192499SJosef Bacik 4148afcdd129SJosef Bacik clear_bit(BTRFS_FS_OPEN, &fs_info->flags); 41494273eaffSAnand Jain free_root_pointers(fs_info, true); 41508c38938cSJosef Bacik btrfs_free_fs_roots(fs_info); 41519ad6b7bcSChris Mason 41524e19443dSJosef Bacik /* 41534e19443dSJosef Bacik * We must free the block groups after dropping the fs_roots as we could 41544e19443dSJosef Bacik * have had an IO error and have left over tree log blocks that aren't 41554e19443dSJosef Bacik * cleaned up until the fs roots are freed. This makes the block group 41564e19443dSJosef Bacik * accounting appear to be wrong because there's pending reserved bytes, 41574e19443dSJosef Bacik * so make sure we do the block group cleanup afterwards. 41584e19443dSJosef Bacik */ 41594e19443dSJosef Bacik btrfs_free_block_groups(fs_info); 41604e19443dSJosef Bacik 416113e6c37bSJosef Bacik iput(fs_info->btree_inode); 4162d6bfde87SChris Mason 416321adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 41640b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) 41652ff7e61eSJeff Mahoney btrfsic_unmount(fs_info->fs_devices); 416621adbd5cSStefan Behrens #endif 416721adbd5cSStefan Behrens 41680b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 416968c94e55SNikolay Borisov btrfs_close_devices(fs_info->fs_devices); 4170eb60ceacSChris Mason } 4171eb60ceacSChris Mason 4172b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 4173b9fab919SChris Mason int atomic) 4174ccd467d6SChris Mason { 41751259ab75SChris Mason int ret; 4176727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 41771259ab75SChris Mason 41780b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 41791259ab75SChris Mason if (!ret) 41801259ab75SChris Mason return ret; 41811259ab75SChris Mason 41821259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 4183b9fab919SChris Mason parent_transid, atomic); 4184b9fab919SChris Mason if (ret == -EAGAIN) 4185b9fab919SChris Mason return ret; 41861259ab75SChris Mason return !ret; 41875f39d397SChris Mason } 41886702ed49SChris Mason 41895f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 41905f39d397SChris Mason { 41910b246afaSJeff Mahoney struct btrfs_fs_info *fs_info; 419206ea65a3SJosef Bacik struct btrfs_root *root; 41935f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 4194b9473439SChris Mason int was_dirty; 4195b4ce94deSChris Mason 419606ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 419706ea65a3SJosef Bacik /* 419806ea65a3SJosef Bacik * This is a fast path so only do this check if we have sanity tests 419952042d8eSAndrea Gelmini * enabled. Normal people shouldn't be using unmapped buffers as dirty 420006ea65a3SJosef Bacik * outside of the sanity tests. 420106ea65a3SJosef Bacik */ 4202b0132a3bSNikolay Borisov if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) 420306ea65a3SJosef Bacik return; 420406ea65a3SJosef Bacik #endif 420506ea65a3SJosef Bacik root = BTRFS_I(buf->pages[0]->mapping->host)->root; 42060b246afaSJeff Mahoney fs_info = root->fs_info; 4207b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 42080b246afaSJeff Mahoney if (transid != fs_info->generation) 42095d163e0eSJeff Mahoney WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", 42100b246afaSJeff Mahoney buf->start, transid, fs_info->generation); 42110b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 4212e2d84521SMiao Xie if (!was_dirty) 4213104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 4214e2d84521SMiao Xie buf->len, 42150b246afaSJeff Mahoney fs_info->dirty_metadata_batch); 42161f21ef0aSFilipe Manana #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 421769fc6cbbSQu Wenruo /* 421869fc6cbbSQu Wenruo * Since btrfs_mark_buffer_dirty() can be called with item pointer set 421969fc6cbbSQu Wenruo * but item data not updated. 422069fc6cbbSQu Wenruo * So here we should only check item pointers, not item data. 422169fc6cbbSQu Wenruo */ 422269fc6cbbSQu Wenruo if (btrfs_header_level(buf) == 0 && 4223cfdaad5eSDavid Sterba btrfs_check_leaf_relaxed(buf)) { 4224a4f78750SDavid Sterba btrfs_print_leaf(buf); 42251f21ef0aSFilipe Manana ASSERT(0); 42261f21ef0aSFilipe Manana } 42271f21ef0aSFilipe Manana #endif 4228eb60ceacSChris Mason } 4229eb60ceacSChris Mason 42302ff7e61eSJeff Mahoney static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, 4231b53d3f5dSLiu Bo int flush_delayed) 423235b7e476SChris Mason { 4233188de649SChris Mason /* 4234188de649SChris Mason * looks as though older kernels can get into trouble with 4235188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 4236188de649SChris Mason */ 4237e2d84521SMiao Xie int ret; 4238d6bfde87SChris Mason 42396933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 4240d6bfde87SChris Mason return; 4241d6bfde87SChris Mason 4242b53d3f5dSLiu Bo if (flush_delayed) 42432ff7e61eSJeff Mahoney btrfs_balance_delayed_items(fs_info); 424416cdcec7SMiao Xie 4245d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 4246d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 4247d814a491SEthan Lien fs_info->dirty_metadata_batch); 4248e2d84521SMiao Xie if (ret > 0) { 42490b246afaSJeff Mahoney balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); 425016cdcec7SMiao Xie } 425116cdcec7SMiao Xie } 425216cdcec7SMiao Xie 42532ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) 425416cdcec7SMiao Xie { 42552ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 1); 425635b7e476SChris Mason } 4257b53d3f5dSLiu Bo 42582ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) 4259b53d3f5dSLiu Bo { 42602ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 0); 4261d6bfde87SChris Mason } 42626b80053dSChris Mason 4263581c1760SQu Wenruo int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level, 4264581c1760SQu Wenruo struct btrfs_key *first_key) 42656b80053dSChris Mason { 42665ab12d1fSDavid Sterba return btree_read_extent_buffer_pages(buf, parent_transid, 4267581c1760SQu Wenruo level, first_key); 42686b80053dSChris Mason } 42690da5468fSChris Mason 42702ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) 4271acce952bSliubo { 4272fe816d0fSNikolay Borisov /* cleanup FS via transaction */ 4273fe816d0fSNikolay Borisov btrfs_cleanup_transaction(fs_info); 4274fe816d0fSNikolay Borisov 42750b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 42762ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 42770b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 4278acce952bSliubo 42790b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 42800b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4281acce952bSliubo } 4282acce952bSliubo 4283ef67963dSJosef Bacik static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info) 4284ef67963dSJosef Bacik { 4285ef67963dSJosef Bacik struct btrfs_root *gang[8]; 4286ef67963dSJosef Bacik u64 root_objectid = 0; 4287ef67963dSJosef Bacik int ret; 4288ef67963dSJosef Bacik 4289ef67963dSJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 4290ef67963dSJosef Bacik while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4291ef67963dSJosef Bacik (void **)gang, root_objectid, 4292ef67963dSJosef Bacik ARRAY_SIZE(gang))) != 0) { 4293ef67963dSJosef Bacik int i; 4294ef67963dSJosef Bacik 4295ef67963dSJosef Bacik for (i = 0; i < ret; i++) 4296ef67963dSJosef Bacik gang[i] = btrfs_grab_root(gang[i]); 4297ef67963dSJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 4298ef67963dSJosef Bacik 4299ef67963dSJosef Bacik for (i = 0; i < ret; i++) { 4300ef67963dSJosef Bacik if (!gang[i]) 4301ef67963dSJosef Bacik continue; 4302ef67963dSJosef Bacik root_objectid = gang[i]->root_key.objectid; 4303ef67963dSJosef Bacik btrfs_free_log(NULL, gang[i]); 4304ef67963dSJosef Bacik btrfs_put_root(gang[i]); 4305ef67963dSJosef Bacik } 4306ef67963dSJosef Bacik root_objectid++; 4307ef67963dSJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 4308ef67963dSJosef Bacik } 4309ef67963dSJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 4310ef67963dSJosef Bacik btrfs_free_log_root_tree(NULL, fs_info); 4311ef67963dSJosef Bacik } 4312ef67963dSJosef Bacik 4313143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4314acce952bSliubo { 4315acce952bSliubo struct btrfs_ordered_extent *ordered; 4316acce952bSliubo 4317199c2a9cSMiao Xie spin_lock(&root->ordered_extent_lock); 4318779880efSJosef Bacik /* 4319779880efSJosef Bacik * This will just short circuit the ordered completion stuff which will 4320779880efSJosef Bacik * make sure the ordered extent gets properly cleaned up. 4321779880efSJosef Bacik */ 4322199c2a9cSMiao Xie list_for_each_entry(ordered, &root->ordered_extents, 4323779880efSJosef Bacik root_extent_list) 4324779880efSJosef Bacik set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4325199c2a9cSMiao Xie spin_unlock(&root->ordered_extent_lock); 4326199c2a9cSMiao Xie } 4327199c2a9cSMiao Xie 4328199c2a9cSMiao Xie static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4329199c2a9cSMiao Xie { 4330199c2a9cSMiao Xie struct btrfs_root *root; 4331199c2a9cSMiao Xie struct list_head splice; 4332199c2a9cSMiao Xie 4333199c2a9cSMiao Xie INIT_LIST_HEAD(&splice); 4334199c2a9cSMiao Xie 4335199c2a9cSMiao Xie spin_lock(&fs_info->ordered_root_lock); 4336199c2a9cSMiao Xie list_splice_init(&fs_info->ordered_roots, &splice); 4337199c2a9cSMiao Xie while (!list_empty(&splice)) { 4338199c2a9cSMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4339199c2a9cSMiao Xie ordered_root); 43401de2cfdeSJosef Bacik list_move_tail(&root->ordered_root, 43411de2cfdeSJosef Bacik &fs_info->ordered_roots); 4342199c2a9cSMiao Xie 43432a85d9caSLiu Bo spin_unlock(&fs_info->ordered_root_lock); 4344199c2a9cSMiao Xie btrfs_destroy_ordered_extents(root); 4345199c2a9cSMiao Xie 43462a85d9caSLiu Bo cond_resched(); 43472a85d9caSLiu Bo spin_lock(&fs_info->ordered_root_lock); 4348199c2a9cSMiao Xie } 4349199c2a9cSMiao Xie spin_unlock(&fs_info->ordered_root_lock); 435074d5d229SJosef Bacik 435174d5d229SJosef Bacik /* 435274d5d229SJosef Bacik * We need this here because if we've been flipped read-only we won't 435374d5d229SJosef Bacik * get sync() from the umount, so we need to make sure any ordered 435474d5d229SJosef Bacik * extents that haven't had their dirty pages IO start writeout yet 435574d5d229SJosef Bacik * actually get run and error out properly. 435674d5d229SJosef Bacik */ 435774d5d229SJosef Bacik btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 4358acce952bSliubo } 4359acce952bSliubo 436035a3621bSStefan Behrens static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 43612ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4362acce952bSliubo { 4363acce952bSliubo struct rb_node *node; 4364acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 4365acce952bSliubo struct btrfs_delayed_ref_node *ref; 4366acce952bSliubo int ret = 0; 4367acce952bSliubo 4368acce952bSliubo delayed_refs = &trans->delayed_refs; 4369acce952bSliubo 4370acce952bSliubo spin_lock(&delayed_refs->lock); 4371d7df2c79SJosef Bacik if (atomic_read(&delayed_refs->num_entries) == 0) { 4372cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 4373b79ce3ddSDavid Sterba btrfs_debug(fs_info, "delayed_refs has NO entry"); 4374acce952bSliubo return ret; 4375acce952bSliubo } 4376acce952bSliubo 43775c9d028bSLiu Bo while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { 4378d7df2c79SJosef Bacik struct btrfs_delayed_ref_head *head; 43790e0adbcfSJosef Bacik struct rb_node *n; 4380e78417d1SJosef Bacik bool pin_bytes = false; 4381acce952bSliubo 4382d7df2c79SJosef Bacik head = rb_entry(node, struct btrfs_delayed_ref_head, 4383d7df2c79SJosef Bacik href_node); 43843069bd26SJosef Bacik if (btrfs_delayed_ref_lock(delayed_refs, head)) 4385b939d1abSJosef Bacik continue; 43863069bd26SJosef Bacik 4387d7df2c79SJosef Bacik spin_lock(&head->lock); 4388e3d03965SLiu Bo while ((n = rb_first_cached(&head->ref_tree)) != NULL) { 43890e0adbcfSJosef Bacik ref = rb_entry(n, struct btrfs_delayed_ref_node, 43900e0adbcfSJosef Bacik ref_node); 4391d7df2c79SJosef Bacik ref->in_tree = 0; 4392e3d03965SLiu Bo rb_erase_cached(&ref->ref_node, &head->ref_tree); 43930e0adbcfSJosef Bacik RB_CLEAR_NODE(&ref->ref_node); 43941d57ee94SWang Xiaoguang if (!list_empty(&ref->add_list)) 43951d57ee94SWang Xiaoguang list_del(&ref->add_list); 4396d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4397d7df2c79SJosef Bacik btrfs_put_delayed_ref(ref); 4398d7df2c79SJosef Bacik } 439954067ae9SJosef Bacik if (head->must_insert_reserved) 4400e78417d1SJosef Bacik pin_bytes = true; 440178a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 4402fa781ceaSJosef Bacik btrfs_delete_ref_head(delayed_refs, head); 4403d7df2c79SJosef Bacik spin_unlock(&head->lock); 4404acce952bSliubo spin_unlock(&delayed_refs->lock); 4405e78417d1SJosef Bacik mutex_unlock(&head->mutex); 4406acce952bSliubo 4407f603bb94SNikolay Borisov if (pin_bytes) { 4408f603bb94SNikolay Borisov struct btrfs_block_group *cache; 4409f603bb94SNikolay Borisov 4410f603bb94SNikolay Borisov cache = btrfs_lookup_block_group(fs_info, head->bytenr); 4411f603bb94SNikolay Borisov BUG_ON(!cache); 4412f603bb94SNikolay Borisov 4413f603bb94SNikolay Borisov spin_lock(&cache->space_info->lock); 4414f603bb94SNikolay Borisov spin_lock(&cache->lock); 4415f603bb94SNikolay Borisov cache->pinned += head->num_bytes; 4416f603bb94SNikolay Borisov btrfs_space_info_update_bytes_pinned(fs_info, 4417f603bb94SNikolay Borisov cache->space_info, head->num_bytes); 4418f603bb94SNikolay Borisov cache->reserved -= head->num_bytes; 4419f603bb94SNikolay Borisov cache->space_info->bytes_reserved -= head->num_bytes; 4420f603bb94SNikolay Borisov spin_unlock(&cache->lock); 4421f603bb94SNikolay Borisov spin_unlock(&cache->space_info->lock); 4422f603bb94SNikolay Borisov percpu_counter_add_batch( 4423f603bb94SNikolay Borisov &cache->space_info->total_bytes_pinned, 4424f603bb94SNikolay Borisov head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH); 4425f603bb94SNikolay Borisov 4426f603bb94SNikolay Borisov btrfs_put_block_group(cache); 4427f603bb94SNikolay Borisov 4428f603bb94SNikolay Borisov btrfs_error_unpin_extent_range(fs_info, head->bytenr, 4429f603bb94SNikolay Borisov head->bytenr + head->num_bytes - 1); 4430f603bb94SNikolay Borisov } 443131890da0SJosef Bacik btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); 4432d278850eSJosef Bacik btrfs_put_delayed_ref_head(head); 4433acce952bSliubo cond_resched(); 4434acce952bSliubo spin_lock(&delayed_refs->lock); 4435acce952bSliubo } 443681f7eb00SJeff Mahoney btrfs_qgroup_destroy_extent_records(trans); 4437acce952bSliubo 4438acce952bSliubo spin_unlock(&delayed_refs->lock); 4439acce952bSliubo 4440acce952bSliubo return ret; 4441acce952bSliubo } 4442acce952bSliubo 4443143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 4444acce952bSliubo { 4445acce952bSliubo struct btrfs_inode *btrfs_inode; 4446acce952bSliubo struct list_head splice; 4447acce952bSliubo 4448acce952bSliubo INIT_LIST_HEAD(&splice); 4449acce952bSliubo 4450eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4451eb73c1b7SMiao Xie list_splice_init(&root->delalloc_inodes, &splice); 4452acce952bSliubo 4453acce952bSliubo while (!list_empty(&splice)) { 4454fe816d0fSNikolay Borisov struct inode *inode = NULL; 4455eb73c1b7SMiao Xie btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 4456acce952bSliubo delalloc_inodes); 4457fe816d0fSNikolay Borisov __btrfs_del_delalloc_inode(root, btrfs_inode); 4458eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4459acce952bSliubo 4460fe816d0fSNikolay Borisov /* 4461fe816d0fSNikolay Borisov * Make sure we get a live inode and that it'll not disappear 4462fe816d0fSNikolay Borisov * meanwhile. 4463fe816d0fSNikolay Borisov */ 4464fe816d0fSNikolay Borisov inode = igrab(&btrfs_inode->vfs_inode); 4465fe816d0fSNikolay Borisov if (inode) { 4466fe816d0fSNikolay Borisov invalidate_inode_pages2(inode->i_mapping); 4467fe816d0fSNikolay Borisov iput(inode); 4468fe816d0fSNikolay Borisov } 4469eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4470acce952bSliubo } 4471eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4472eb73c1b7SMiao Xie } 4473eb73c1b7SMiao Xie 4474eb73c1b7SMiao Xie static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 4475eb73c1b7SMiao Xie { 4476eb73c1b7SMiao Xie struct btrfs_root *root; 4477eb73c1b7SMiao Xie struct list_head splice; 4478eb73c1b7SMiao Xie 4479eb73c1b7SMiao Xie INIT_LIST_HEAD(&splice); 4480eb73c1b7SMiao Xie 4481eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4482eb73c1b7SMiao Xie list_splice_init(&fs_info->delalloc_roots, &splice); 4483eb73c1b7SMiao Xie while (!list_empty(&splice)) { 4484eb73c1b7SMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4485eb73c1b7SMiao Xie delalloc_root); 448600246528SJosef Bacik root = btrfs_grab_root(root); 4487eb73c1b7SMiao Xie BUG_ON(!root); 4488eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4489eb73c1b7SMiao Xie 4490eb73c1b7SMiao Xie btrfs_destroy_delalloc_inodes(root); 449100246528SJosef Bacik btrfs_put_root(root); 4492eb73c1b7SMiao Xie 4493eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4494eb73c1b7SMiao Xie } 4495eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4496acce952bSliubo } 4497acce952bSliubo 44982ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 4499acce952bSliubo struct extent_io_tree *dirty_pages, 4500acce952bSliubo int mark) 4501acce952bSliubo { 4502acce952bSliubo int ret; 4503acce952bSliubo struct extent_buffer *eb; 4504acce952bSliubo u64 start = 0; 4505acce952bSliubo u64 end; 4506acce952bSliubo 4507acce952bSliubo while (1) { 4508acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 4509e6138876SJosef Bacik mark, NULL); 4510acce952bSliubo if (ret) 4511acce952bSliubo break; 4512acce952bSliubo 451391166212SDavid Sterba clear_extent_bits(dirty_pages, start, end, mark); 4514acce952bSliubo while (start <= end) { 45150b246afaSJeff Mahoney eb = find_extent_buffer(fs_info, start); 45160b246afaSJeff Mahoney start += fs_info->nodesize; 4517fd8b2b61SJosef Bacik if (!eb) 4518acce952bSliubo continue; 4519fd8b2b61SJosef Bacik wait_on_extent_buffer_writeback(eb); 4520acce952bSliubo 4521fd8b2b61SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 4522fd8b2b61SJosef Bacik &eb->bflags)) 4523fd8b2b61SJosef Bacik clear_extent_buffer_dirty(eb); 4524fd8b2b61SJosef Bacik free_extent_buffer_stale(eb); 4525acce952bSliubo } 4526acce952bSliubo } 4527acce952bSliubo 4528acce952bSliubo return ret; 4529acce952bSliubo } 4530acce952bSliubo 45312ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 4532fe119a6eSNikolay Borisov struct extent_io_tree *unpin) 4533acce952bSliubo { 4534acce952bSliubo u64 start; 4535acce952bSliubo u64 end; 4536acce952bSliubo int ret; 4537acce952bSliubo 4538acce952bSliubo while (1) { 45390e6ec385SFilipe Manana struct extent_state *cached_state = NULL; 45400e6ec385SFilipe Manana 4541fcd5e742SLu Fengqi /* 4542fcd5e742SLu Fengqi * The btrfs_finish_extent_commit() may get the same range as 4543fcd5e742SLu Fengqi * ours between find_first_extent_bit and clear_extent_dirty. 4544fcd5e742SLu Fengqi * Hence, hold the unused_bg_unpin_mutex to avoid double unpin 4545fcd5e742SLu Fengqi * the same extent range. 4546fcd5e742SLu Fengqi */ 4547fcd5e742SLu Fengqi mutex_lock(&fs_info->unused_bg_unpin_mutex); 4548acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 45490e6ec385SFilipe Manana EXTENT_DIRTY, &cached_state); 4550fcd5e742SLu Fengqi if (ret) { 4551fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4552acce952bSliubo break; 4553fcd5e742SLu Fengqi } 4554acce952bSliubo 45550e6ec385SFilipe Manana clear_extent_dirty(unpin, start, end, &cached_state); 45560e6ec385SFilipe Manana free_extent_state(cached_state); 45572ff7e61eSJeff Mahoney btrfs_error_unpin_extent_range(fs_info, start, end); 4558fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4559acce952bSliubo cond_resched(); 4560acce952bSliubo } 4561acce952bSliubo 4562acce952bSliubo return 0; 4563acce952bSliubo } 4564acce952bSliubo 456532da5386SDavid Sterba static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache) 4566c79a1751SLiu Bo { 4567c79a1751SLiu Bo struct inode *inode; 4568c79a1751SLiu Bo 4569c79a1751SLiu Bo inode = cache->io_ctl.inode; 4570c79a1751SLiu Bo if (inode) { 4571c79a1751SLiu Bo invalidate_inode_pages2(inode->i_mapping); 4572c79a1751SLiu Bo BTRFS_I(inode)->generation = 0; 4573c79a1751SLiu Bo cache->io_ctl.inode = NULL; 4574c79a1751SLiu Bo iput(inode); 4575c79a1751SLiu Bo } 4576bbc37d6eSFilipe Manana ASSERT(cache->io_ctl.pages == NULL); 4577c79a1751SLiu Bo btrfs_put_block_group(cache); 4578c79a1751SLiu Bo } 4579c79a1751SLiu Bo 4580c79a1751SLiu Bo void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, 45812ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4582c79a1751SLiu Bo { 458332da5386SDavid Sterba struct btrfs_block_group *cache; 4584c79a1751SLiu Bo 4585c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4586c79a1751SLiu Bo while (!list_empty(&cur_trans->dirty_bgs)) { 4587c79a1751SLiu Bo cache = list_first_entry(&cur_trans->dirty_bgs, 458832da5386SDavid Sterba struct btrfs_block_group, 4589c79a1751SLiu Bo dirty_list); 4590c79a1751SLiu Bo 4591c79a1751SLiu Bo if (!list_empty(&cache->io_list)) { 4592c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4593c79a1751SLiu Bo list_del_init(&cache->io_list); 4594c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 4595c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4596c79a1751SLiu Bo } 4597c79a1751SLiu Bo 4598c79a1751SLiu Bo list_del_init(&cache->dirty_list); 4599c79a1751SLiu Bo spin_lock(&cache->lock); 4600c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 4601c79a1751SLiu Bo spin_unlock(&cache->lock); 4602c79a1751SLiu Bo 4603c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4604c79a1751SLiu Bo btrfs_put_block_group(cache); 4605ba2c4d4eSJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 4606c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4607c79a1751SLiu Bo } 4608c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4609c79a1751SLiu Bo 461045ae2c18SNikolay Borisov /* 461145ae2c18SNikolay Borisov * Refer to the definition of io_bgs member for details why it's safe 461245ae2c18SNikolay Borisov * to use it without any locking 461345ae2c18SNikolay Borisov */ 4614c79a1751SLiu Bo while (!list_empty(&cur_trans->io_bgs)) { 4615c79a1751SLiu Bo cache = list_first_entry(&cur_trans->io_bgs, 461632da5386SDavid Sterba struct btrfs_block_group, 4617c79a1751SLiu Bo io_list); 4618c79a1751SLiu Bo 4619c79a1751SLiu Bo list_del_init(&cache->io_list); 4620c79a1751SLiu Bo spin_lock(&cache->lock); 4621c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 4622c79a1751SLiu Bo spin_unlock(&cache->lock); 4623c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 4624c79a1751SLiu Bo } 4625c79a1751SLiu Bo } 4626c79a1751SLiu Bo 462749b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 46282ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 462949b25e05SJeff Mahoney { 4630bbbf7243SNikolay Borisov struct btrfs_device *dev, *tmp; 4631bbbf7243SNikolay Borisov 46322ff7e61eSJeff Mahoney btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 4633c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->dirty_bgs)); 4634c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->io_bgs)); 4635c79a1751SLiu Bo 4636bbbf7243SNikolay Borisov list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list, 4637bbbf7243SNikolay Borisov post_commit_list) { 4638bbbf7243SNikolay Borisov list_del_init(&dev->post_commit_list); 4639bbbf7243SNikolay Borisov } 4640bbbf7243SNikolay Borisov 46412ff7e61eSJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, fs_info); 464249b25e05SJeff Mahoney 46434a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 46440b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 464549b25e05SJeff Mahoney 46464a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 46470b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 464849b25e05SJeff Mahoney 4649ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 465067cde344SMiao Xie 46512ff7e61eSJeff Mahoney btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, 465249b25e05SJeff Mahoney EXTENT_DIRTY); 4653fe119a6eSNikolay Borisov btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents); 465449b25e05SJeff Mahoney 46554a9d8bdeSMiao Xie cur_trans->state =TRANS_STATE_COMPLETED; 46564a9d8bdeSMiao Xie wake_up(&cur_trans->commit_wait); 465749b25e05SJeff Mahoney } 465849b25e05SJeff Mahoney 46592ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) 4660acce952bSliubo { 4661acce952bSliubo struct btrfs_transaction *t; 4662acce952bSliubo 46630b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 4664acce952bSliubo 46650b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 46660b246afaSJeff Mahoney while (!list_empty(&fs_info->trans_list)) { 46670b246afaSJeff Mahoney t = list_first_entry(&fs_info->trans_list, 4668724e2315SJosef Bacik struct btrfs_transaction, list); 4669724e2315SJosef Bacik if (t->state >= TRANS_STATE_COMMIT_START) { 46709b64f57dSElena Reshetova refcount_inc(&t->use_count); 46710b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 46722ff7e61eSJeff Mahoney btrfs_wait_for_commit(fs_info, t->transid); 4673724e2315SJosef Bacik btrfs_put_transaction(t); 46740b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4675724e2315SJosef Bacik continue; 4676724e2315SJosef Bacik } 46770b246afaSJeff Mahoney if (t == fs_info->running_transaction) { 4678724e2315SJosef Bacik t->state = TRANS_STATE_COMMIT_DOING; 46790b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4680724e2315SJosef Bacik /* 4681724e2315SJosef Bacik * We wait for 0 num_writers since we don't hold a trans 4682724e2315SJosef Bacik * handle open currently for this transaction. 4683724e2315SJosef Bacik */ 4684724e2315SJosef Bacik wait_event(t->writer_wait, 4685724e2315SJosef Bacik atomic_read(&t->num_writers) == 0); 4686724e2315SJosef Bacik } else { 46870b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4688724e2315SJosef Bacik } 46892ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(t, fs_info); 4690724e2315SJosef Bacik 46910b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 46920b246afaSJeff Mahoney if (t == fs_info->running_transaction) 46930b246afaSJeff Mahoney fs_info->running_transaction = NULL; 4694724e2315SJosef Bacik list_del_init(&t->list); 46950b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4696a4abeea4SJosef Bacik 4697724e2315SJosef Bacik btrfs_put_transaction(t); 46982ff7e61eSJeff Mahoney trace_btrfs_transaction_commit(fs_info->tree_root); 46990b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4700724e2315SJosef Bacik } 47010b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 47020b246afaSJeff Mahoney btrfs_destroy_all_ordered_extents(fs_info); 4703ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 4704ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 47050b246afaSJeff Mahoney btrfs_destroy_all_delalloc_inodes(fs_info); 4706ef67963dSJosef Bacik btrfs_drop_all_logs(fs_info); 47070b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 4708acce952bSliubo 4709acce952bSliubo return 0; 4710acce952bSliubo } 4711