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 53e8c9f186SDavid Sterba static const struct extent_io_ops btree_extent_io_ops; 548b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work); 55143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root); 56acce952bSliubo static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 572ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info); 58143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root); 592ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 60acce952bSliubo struct extent_io_tree *dirty_pages, 61acce952bSliubo int mark); 622ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 63acce952bSliubo struct extent_io_tree *pinned_extents); 642ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info); 652ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info); 66ce9adaa5SChris Mason 67d352ac68SChris Mason /* 6897eb6b69SDavid Sterba * btrfs_end_io_wq structs are used to do processing in task context when an IO 6997eb6b69SDavid Sterba * is complete. This is used during reads to verify checksums, and it is used 70d352ac68SChris Mason * by writes to insert metadata for new file extents after IO is complete. 71d352ac68SChris Mason */ 7297eb6b69SDavid Sterba struct btrfs_end_io_wq { 73ce9adaa5SChris Mason struct bio *bio; 74ce9adaa5SChris Mason bio_end_io_t *end_io; 75ce9adaa5SChris Mason void *private; 76ce9adaa5SChris Mason struct btrfs_fs_info *info; 774e4cbee9SChristoph Hellwig blk_status_t status; 78bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata; 798b712842SChris Mason struct btrfs_work work; 80ce9adaa5SChris Mason }; 810da5468fSChris Mason 8297eb6b69SDavid Sterba static struct kmem_cache *btrfs_end_io_wq_cache; 8397eb6b69SDavid Sterba 8497eb6b69SDavid Sterba int __init btrfs_end_io_wq_init(void) 8597eb6b69SDavid Sterba { 8697eb6b69SDavid Sterba btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq", 8797eb6b69SDavid Sterba sizeof(struct btrfs_end_io_wq), 8897eb6b69SDavid Sterba 0, 89fba4b697SNikolay Borisov SLAB_MEM_SPREAD, 9097eb6b69SDavid Sterba NULL); 9197eb6b69SDavid Sterba if (!btrfs_end_io_wq_cache) 9297eb6b69SDavid Sterba return -ENOMEM; 9397eb6b69SDavid Sterba return 0; 9497eb6b69SDavid Sterba } 9597eb6b69SDavid Sterba 96e67c718bSDavid Sterba void __cold btrfs_end_io_wq_exit(void) 9797eb6b69SDavid Sterba { 9897eb6b69SDavid Sterba kmem_cache_destroy(btrfs_end_io_wq_cache); 9997eb6b69SDavid Sterba } 10097eb6b69SDavid Sterba 101141386e1SJosef Bacik static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info) 102141386e1SJosef Bacik { 103141386e1SJosef Bacik if (fs_info->csum_shash) 104141386e1SJosef Bacik crypto_free_shash(fs_info->csum_shash); 105141386e1SJosef Bacik } 106141386e1SJosef Bacik 107d352ac68SChris Mason /* 108d352ac68SChris Mason * async submit bios are used to offload expensive checksumming 109d352ac68SChris Mason * onto the worker threads. They checksum file and metadata bios 110d352ac68SChris Mason * just before they are sent down the IO stack. 111d352ac68SChris Mason */ 11244b8bd7eSChris Mason struct async_submit_bio { 113c6100a4bSJosef Bacik void *private_data; 11444b8bd7eSChris Mason struct bio *bio; 115a758781dSDavid Sterba extent_submit_bio_start_t *submit_bio_start; 11644b8bd7eSChris Mason int mirror_num; 117eaf25d93SChris Mason /* 118eaf25d93SChris Mason * bio_offset is optional, can be used if the pages in the bio 119eaf25d93SChris Mason * can't tell us where in the file the bio should go 120eaf25d93SChris Mason */ 121eaf25d93SChris Mason u64 bio_offset; 1228b712842SChris Mason struct btrfs_work work; 1234e4cbee9SChristoph Hellwig blk_status_t status; 12444b8bd7eSChris Mason }; 12544b8bd7eSChris Mason 12685d4e461SChris Mason /* 12785d4e461SChris Mason * Lockdep class keys for extent_buffer->lock's in this root. For a given 12885d4e461SChris Mason * eb, the lockdep key is determined by the btrfs_root it belongs to and 12985d4e461SChris Mason * the level the eb occupies in the tree. 1304008c04aSChris Mason * 13185d4e461SChris Mason * Different roots are used for different purposes and may nest inside each 13285d4e461SChris Mason * other and they require separate keysets. As lockdep keys should be 13385d4e461SChris Mason * static, assign keysets according to the purpose of the root as indicated 1344fd786e6SMisono Tomohiro * by btrfs_root->root_key.objectid. This ensures that all special purpose 1354fd786e6SMisono Tomohiro * roots have separate keysets. 1364008c04aSChris Mason * 13785d4e461SChris Mason * Lock-nesting across peer nodes is always done with the immediate parent 13885d4e461SChris Mason * node locked thus preventing deadlock. As lockdep doesn't know this, use 13985d4e461SChris Mason * subclass to avoid triggering lockdep warning in such cases. 1404008c04aSChris Mason * 14185d4e461SChris Mason * The key is set by the readpage_end_io_hook after the buffer has passed 14285d4e461SChris Mason * csum validation but before the pages are unlocked. It is also set by 14385d4e461SChris Mason * btrfs_init_new_buffer on freshly allocated blocks. 14485d4e461SChris Mason * 14585d4e461SChris Mason * We also add a check to make sure the highest level of the tree is the 14685d4e461SChris Mason * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code 14785d4e461SChris Mason * needs update as well. 1484008c04aSChris Mason */ 1494008c04aSChris Mason #ifdef CONFIG_DEBUG_LOCK_ALLOC 1504008c04aSChris Mason # if BTRFS_MAX_LEVEL != 8 1514008c04aSChris Mason # error 1524008c04aSChris Mason # endif 15385d4e461SChris Mason 15485d4e461SChris Mason static struct btrfs_lockdep_keyset { 15585d4e461SChris Mason u64 id; /* root objectid */ 15685d4e461SChris Mason const char *name_stem; /* lock name stem */ 15785d4e461SChris Mason char names[BTRFS_MAX_LEVEL + 1][20]; 15885d4e461SChris Mason struct lock_class_key keys[BTRFS_MAX_LEVEL + 1]; 15985d4e461SChris Mason } btrfs_lockdep_keysets[] = { 16085d4e461SChris Mason { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" }, 16185d4e461SChris Mason { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" }, 16285d4e461SChris Mason { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" }, 16385d4e461SChris Mason { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" }, 16485d4e461SChris Mason { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" }, 16585d4e461SChris Mason { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" }, 16660b62978SDavid Sterba { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" }, 16785d4e461SChris Mason { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" }, 16885d4e461SChris Mason { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" }, 16985d4e461SChris Mason { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" }, 17013fd8da9SDavid Sterba { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" }, 1716b20e0adSDavid Sterba { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" }, 17285d4e461SChris Mason { .id = 0, .name_stem = "tree" }, 1734008c04aSChris Mason }; 17485d4e461SChris Mason 17585d4e461SChris Mason void __init btrfs_init_lockdep(void) 17685d4e461SChris Mason { 17785d4e461SChris Mason int i, j; 17885d4e461SChris Mason 17985d4e461SChris Mason /* initialize lockdep class names */ 18085d4e461SChris Mason for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) { 18185d4e461SChris Mason struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i]; 18285d4e461SChris Mason 18385d4e461SChris Mason for (j = 0; j < ARRAY_SIZE(ks->names); j++) 18485d4e461SChris Mason snprintf(ks->names[j], sizeof(ks->names[j]), 18585d4e461SChris Mason "btrfs-%s-%02d", ks->name_stem, j); 18685d4e461SChris Mason } 18785d4e461SChris Mason } 18885d4e461SChris Mason 18985d4e461SChris Mason void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb, 19085d4e461SChris Mason int level) 19185d4e461SChris Mason { 19285d4e461SChris Mason struct btrfs_lockdep_keyset *ks; 19385d4e461SChris Mason 19485d4e461SChris Mason BUG_ON(level >= ARRAY_SIZE(ks->keys)); 19585d4e461SChris Mason 19685d4e461SChris Mason /* find the matching keyset, id 0 is the default entry */ 19785d4e461SChris Mason for (ks = btrfs_lockdep_keysets; ks->id; ks++) 19885d4e461SChris Mason if (ks->id == objectid) 19985d4e461SChris Mason break; 20085d4e461SChris Mason 20185d4e461SChris Mason lockdep_set_class_and_name(&eb->lock, 20285d4e461SChris Mason &ks->keys[level], ks->names[level]); 20385d4e461SChris Mason } 20485d4e461SChris Mason 2054008c04aSChris Mason #endif 2064008c04aSChris Mason 207d352ac68SChris Mason /* 208d352ac68SChris Mason * extents on the btree inode are pretty simple, there's one extent 209d352ac68SChris Mason * that covers the entire device 210d352ac68SChris Mason */ 2116af49dbdSDavid Sterba struct extent_map *btree_get_extent(struct btrfs_inode *inode, 21239b07b5dSOmar Sandoval struct page *page, size_t pg_offset, 21339b07b5dSOmar Sandoval u64 start, u64 len) 2145f39d397SChris Mason { 215fc4f21b1SNikolay Borisov struct extent_map_tree *em_tree = &inode->extent_tree; 2165f39d397SChris Mason struct extent_map *em; 2175f39d397SChris Mason int ret; 2185f39d397SChris Mason 219890871beSChris Mason read_lock(&em_tree->lock); 220d1310b2eSChris Mason em = lookup_extent_mapping(em_tree, start, len); 221a061fc8dSChris Mason if (em) { 222890871beSChris Mason read_unlock(&em_tree->lock); 2235f39d397SChris Mason goto out; 224a061fc8dSChris Mason } 225890871beSChris Mason read_unlock(&em_tree->lock); 2267b13b7b1SChris Mason 227172ddd60SDavid Sterba em = alloc_extent_map(); 2285f39d397SChris Mason if (!em) { 2295f39d397SChris Mason em = ERR_PTR(-ENOMEM); 2305f39d397SChris Mason goto out; 2315f39d397SChris Mason } 2325f39d397SChris Mason em->start = 0; 2330afbaf8cSChris Mason em->len = (u64)-1; 234c8b97818SChris Mason em->block_len = (u64)-1; 2355f39d397SChris Mason em->block_start = 0; 236d1310b2eSChris Mason 237890871beSChris Mason write_lock(&em_tree->lock); 23809a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 2395f39d397SChris Mason if (ret == -EEXIST) { 2405f39d397SChris Mason free_extent_map(em); 2417b13b7b1SChris Mason em = lookup_extent_mapping(em_tree, start, len); 242b4f359abSTsutomu Itoh if (!em) 2430433f20dSTsutomu Itoh em = ERR_PTR(-EIO); 2445f39d397SChris Mason } else if (ret) { 2457b13b7b1SChris Mason free_extent_map(em); 2460433f20dSTsutomu Itoh em = ERR_PTR(ret); 2475f39d397SChris Mason } 248890871beSChris Mason write_unlock(&em_tree->lock); 2497b13b7b1SChris Mason 2505f39d397SChris Mason out: 2515f39d397SChris Mason return em; 2525f39d397SChris Mason } 2535f39d397SChris Mason 254d352ac68SChris Mason /* 2552996e1f8SJohannes Thumshirn * Compute the csum of a btree block and store the result to provided buffer. 256d352ac68SChris Mason */ 257c67b3892SDavid Sterba static void csum_tree_block(struct extent_buffer *buf, u8 *result) 25819c00ddcSChris Mason { 259d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = buf->fs_info; 260e9be5a30SDavid Sterba const int num_pages = fs_info->nodesize >> PAGE_SHIFT; 261d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 26219c00ddcSChris Mason char *kaddr; 263e9be5a30SDavid Sterba int i; 264d5178578SJohannes Thumshirn 265d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 266d5178578SJohannes Thumshirn crypto_shash_init(shash); 267e9be5a30SDavid Sterba kaddr = page_address(buf->pages[0]); 268e9be5a30SDavid Sterba crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE, 269e9be5a30SDavid Sterba PAGE_SIZE - BTRFS_CSUM_SIZE); 27019c00ddcSChris Mason 271e9be5a30SDavid Sterba for (i = 1; i < num_pages; i++) { 272e9be5a30SDavid Sterba kaddr = page_address(buf->pages[i]); 273e9be5a30SDavid Sterba crypto_shash_update(shash, kaddr, PAGE_SIZE); 27419c00ddcSChris Mason } 27571a63551SDavid Sterba memset(result, 0, BTRFS_CSUM_SIZE); 276d5178578SJohannes Thumshirn crypto_shash_final(shash, result); 27719c00ddcSChris Mason } 27819c00ddcSChris Mason 279d352ac68SChris Mason /* 280d352ac68SChris Mason * we can't consider a given block up to date unless the transid of the 281d352ac68SChris Mason * block matches the transid in the parent node's pointer. This is how we 282d352ac68SChris Mason * detect blocks that either didn't get written at all or got written 283d352ac68SChris Mason * in the wrong place. 284d352ac68SChris Mason */ 2851259ab75SChris Mason static int verify_parent_transid(struct extent_io_tree *io_tree, 286b9fab919SChris Mason struct extent_buffer *eb, u64 parent_transid, 287b9fab919SChris Mason int atomic) 2881259ab75SChris Mason { 2892ac55d41SJosef Bacik struct extent_state *cached_state = NULL; 2901259ab75SChris Mason int ret; 2912755a0deSDavid Sterba bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB); 2921259ab75SChris Mason 2931259ab75SChris Mason if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 2941259ab75SChris Mason return 0; 2951259ab75SChris Mason 296b9fab919SChris Mason if (atomic) 297b9fab919SChris Mason return -EAGAIN; 298b9fab919SChris Mason 299a26e8c9fSJosef Bacik if (need_lock) { 300a26e8c9fSJosef Bacik btrfs_tree_read_lock(eb); 301300aa896SDavid Sterba btrfs_set_lock_blocking_read(eb); 302a26e8c9fSJosef Bacik } 303a26e8c9fSJosef Bacik 3042ac55d41SJosef Bacik lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 305ff13db41SDavid Sterba &cached_state); 3060b32f4bbSJosef Bacik if (extent_buffer_uptodate(eb) && 3071259ab75SChris Mason btrfs_header_generation(eb) == parent_transid) { 3081259ab75SChris Mason ret = 0; 3091259ab75SChris Mason goto out; 3101259ab75SChris Mason } 31194647322SDavid Sterba btrfs_err_rl(eb->fs_info, 31294647322SDavid Sterba "parent transid verify failed on %llu wanted %llu found %llu", 31394647322SDavid Sterba eb->start, 31429549aecSWang Shilong parent_transid, btrfs_header_generation(eb)); 3151259ab75SChris Mason ret = 1; 316a26e8c9fSJosef Bacik 317a26e8c9fSJosef Bacik /* 318a26e8c9fSJosef Bacik * Things reading via commit roots that don't have normal protection, 319a26e8c9fSJosef Bacik * like send, can have a really old block in cache that may point at a 32001327610SNicholas D Steeves * block that has been freed and re-allocated. So don't clear uptodate 321a26e8c9fSJosef Bacik * if we find an eb that is under IO (dirty/writeback) because we could 322a26e8c9fSJosef Bacik * end up reading in the stale data and then writing it back out and 323a26e8c9fSJosef Bacik * making everybody very sad. 324a26e8c9fSJosef Bacik */ 325a26e8c9fSJosef Bacik if (!extent_buffer_under_io(eb)) 3260b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 32733958dc6SChris Mason out: 3282ac55d41SJosef Bacik unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 329e43bbe5eSDavid Sterba &cached_state); 330472b909fSJosef Bacik if (need_lock) 331a26e8c9fSJosef Bacik btrfs_tree_read_unlock_blocking(eb); 3321259ab75SChris Mason return ret; 3331259ab75SChris Mason } 3341259ab75SChris Mason 335e7e16f48SJohannes Thumshirn static bool btrfs_supported_super_csum(u16 csum_type) 336e7e16f48SJohannes Thumshirn { 337e7e16f48SJohannes Thumshirn switch (csum_type) { 338e7e16f48SJohannes Thumshirn case BTRFS_CSUM_TYPE_CRC32: 3393951e7f0SJohannes Thumshirn case BTRFS_CSUM_TYPE_XXHASH: 3403831bf00SJohannes Thumshirn case BTRFS_CSUM_TYPE_SHA256: 341352ae07bSDavid Sterba case BTRFS_CSUM_TYPE_BLAKE2: 342e7e16f48SJohannes Thumshirn return true; 343e7e16f48SJohannes Thumshirn default: 344e7e16f48SJohannes Thumshirn return false; 345e7e16f48SJohannes Thumshirn } 346e7e16f48SJohannes Thumshirn } 347e7e16f48SJohannes Thumshirn 348d352ac68SChris Mason /* 3491104a885SDavid Sterba * Return 0 if the superblock checksum type matches the checksum value of that 3501104a885SDavid Sterba * algorithm. Pass the raw disk superblock data. 3511104a885SDavid Sterba */ 352ab8d0fc4SJeff Mahoney static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info, 353ab8d0fc4SJeff Mahoney char *raw_disk_sb) 3541104a885SDavid Sterba { 3551104a885SDavid Sterba struct btrfs_super_block *disk_sb = 3561104a885SDavid Sterba (struct btrfs_super_block *)raw_disk_sb; 35751bce6c9SJohannes Thumshirn char result[BTRFS_CSUM_SIZE]; 358d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 359d5178578SJohannes Thumshirn 360d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 3611104a885SDavid Sterba 3621104a885SDavid Sterba /* 3631104a885SDavid Sterba * The super_block structure does not span the whole 36451bce6c9SJohannes Thumshirn * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is 36551bce6c9SJohannes Thumshirn * filled with zeros and is included in the checksum. 3661104a885SDavid Sterba */ 367fd08001fSEric Biggers crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE, 368fd08001fSEric Biggers BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result); 3691104a885SDavid Sterba 37051bce6c9SJohannes Thumshirn if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb))) 371e7e16f48SJohannes Thumshirn return 1; 3721104a885SDavid Sterba 373e7e16f48SJohannes Thumshirn return 0; 3741104a885SDavid Sterba } 3751104a885SDavid Sterba 376e064d5e9SDavid Sterba int btrfs_verify_level_key(struct extent_buffer *eb, int level, 377ff76a864SLiu Bo struct btrfs_key *first_key, u64 parent_transid) 378581c1760SQu Wenruo { 379e064d5e9SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 380581c1760SQu Wenruo int found_level; 381581c1760SQu Wenruo struct btrfs_key found_key; 382581c1760SQu Wenruo int ret; 383581c1760SQu Wenruo 384581c1760SQu Wenruo found_level = btrfs_header_level(eb); 385581c1760SQu Wenruo if (found_level != level) { 38663489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 38763489055SQu Wenruo KERN_ERR "BTRFS: tree level check failed\n"); 388581c1760SQu Wenruo btrfs_err(fs_info, 389581c1760SQu Wenruo "tree level mismatch detected, bytenr=%llu level expected=%u has=%u", 390581c1760SQu Wenruo eb->start, level, found_level); 391581c1760SQu Wenruo return -EIO; 392581c1760SQu Wenruo } 393581c1760SQu Wenruo 394581c1760SQu Wenruo if (!first_key) 395581c1760SQu Wenruo return 0; 396581c1760SQu Wenruo 3975d41be6fSQu Wenruo /* 3985d41be6fSQu Wenruo * For live tree block (new tree blocks in current transaction), 3995d41be6fSQu Wenruo * we need proper lock context to avoid race, which is impossible here. 4005d41be6fSQu Wenruo * So we only checks tree blocks which is read from disk, whose 4015d41be6fSQu Wenruo * generation <= fs_info->last_trans_committed. 4025d41be6fSQu Wenruo */ 4035d41be6fSQu Wenruo if (btrfs_header_generation(eb) > fs_info->last_trans_committed) 4045d41be6fSQu Wenruo return 0; 40562fdaa52SQu Wenruo 40662fdaa52SQu Wenruo /* We have @first_key, so this @eb must have at least one item */ 40762fdaa52SQu Wenruo if (btrfs_header_nritems(eb) == 0) { 40862fdaa52SQu Wenruo btrfs_err(fs_info, 40962fdaa52SQu Wenruo "invalid tree nritems, bytenr=%llu nritems=0 expect >0", 41062fdaa52SQu Wenruo eb->start); 41162fdaa52SQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 41262fdaa52SQu Wenruo return -EUCLEAN; 41362fdaa52SQu Wenruo } 41462fdaa52SQu Wenruo 415581c1760SQu Wenruo if (found_level) 416581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &found_key, 0); 417581c1760SQu Wenruo else 418581c1760SQu Wenruo btrfs_item_key_to_cpu(eb, &found_key, 0); 419581c1760SQu Wenruo ret = btrfs_comp_cpu_keys(first_key, &found_key); 420581c1760SQu Wenruo 421581c1760SQu Wenruo if (ret) { 42263489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 42363489055SQu Wenruo KERN_ERR "BTRFS: tree first key check failed\n"); 424581c1760SQu Wenruo btrfs_err(fs_info, 425ff76a864SLiu Bo "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", 426ff76a864SLiu Bo eb->start, parent_transid, first_key->objectid, 427ff76a864SLiu Bo first_key->type, first_key->offset, 428ff76a864SLiu Bo found_key.objectid, found_key.type, 429ff76a864SLiu Bo found_key.offset); 430581c1760SQu Wenruo } 431581c1760SQu Wenruo return ret; 432581c1760SQu Wenruo } 433581c1760SQu Wenruo 4341104a885SDavid Sterba /* 435d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 436d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 437581c1760SQu Wenruo * 438581c1760SQu Wenruo * @parent_transid: expected transid, skip check if 0 439581c1760SQu Wenruo * @level: expected level, mandatory check 440581c1760SQu Wenruo * @first_key: expected key of first slot, skip check if NULL 441d352ac68SChris Mason */ 4425ab12d1fSDavid Sterba static int btree_read_extent_buffer_pages(struct extent_buffer *eb, 443581c1760SQu Wenruo u64 parent_transid, int level, 444581c1760SQu Wenruo struct btrfs_key *first_key) 445f188591eSChris Mason { 4465ab12d1fSDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 447f188591eSChris Mason struct extent_io_tree *io_tree; 448ea466794SJosef Bacik int failed = 0; 449f188591eSChris Mason int ret; 450f188591eSChris Mason int num_copies = 0; 451f188591eSChris Mason int mirror_num = 0; 452ea466794SJosef Bacik int failed_mirror = 0; 453f188591eSChris Mason 4540b246afaSJeff Mahoney io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; 455f188591eSChris Mason while (1) { 456f8397d69SNikolay Borisov clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 457c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num); 458256dd1bbSStefan Behrens if (!ret) { 459581c1760SQu Wenruo if (verify_parent_transid(io_tree, eb, 460b9fab919SChris Mason parent_transid, 0)) 461256dd1bbSStefan Behrens ret = -EIO; 462e064d5e9SDavid Sterba else if (btrfs_verify_level_key(eb, level, 463ff76a864SLiu Bo first_key, parent_transid)) 464581c1760SQu Wenruo ret = -EUCLEAN; 465581c1760SQu Wenruo else 466581c1760SQu Wenruo break; 467256dd1bbSStefan Behrens } 468d397712bSChris Mason 4690b246afaSJeff Mahoney num_copies = btrfs_num_copies(fs_info, 470f188591eSChris Mason eb->start, eb->len); 4714235298eSChris Mason if (num_copies == 1) 472ea466794SJosef Bacik break; 4734235298eSChris Mason 4745cf1ab56SJosef Bacik if (!failed_mirror) { 4755cf1ab56SJosef Bacik failed = 1; 4765cf1ab56SJosef Bacik failed_mirror = eb->read_mirror; 4775cf1ab56SJosef Bacik } 4785cf1ab56SJosef Bacik 479f188591eSChris Mason mirror_num++; 480ea466794SJosef Bacik if (mirror_num == failed_mirror) 481ea466794SJosef Bacik mirror_num++; 482ea466794SJosef Bacik 4834235298eSChris Mason if (mirror_num > num_copies) 484ea466794SJosef Bacik break; 485f188591eSChris Mason } 486ea466794SJosef Bacik 487c0901581SStefan Behrens if (failed && !ret && failed_mirror) 48820a1fbf9SDavid Sterba btrfs_repair_eb_io_failure(eb, failed_mirror); 489ea466794SJosef Bacik 490ea466794SJosef Bacik return ret; 491f188591eSChris Mason } 49219c00ddcSChris Mason 493d352ac68SChris Mason /* 494d397712bSChris Mason * checksum a dirty tree block before IO. This has extra checks to make sure 495d397712bSChris Mason * we only fill in the checksum field in the first page of a multi-page block 496d352ac68SChris Mason */ 497d397712bSChris Mason 49801d58472SDaniel Dressler static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page) 49919c00ddcSChris Mason { 5004eee4fa4SMiao Xie u64 start = page_offset(page); 50119c00ddcSChris Mason u64 found_start; 5022996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 5032996e1f8SJohannes Thumshirn u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); 50419c00ddcSChris Mason struct extent_buffer *eb; 5058d47a0d8SQu Wenruo int ret; 506f188591eSChris Mason 5074f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 5084f2de97aSJosef Bacik if (page != eb->pages[0]) 5094f2de97aSJosef Bacik return 0; 5100f805531SAlex Lyakas 51119c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 5120f805531SAlex Lyakas /* 5130f805531SAlex Lyakas * Please do not consolidate these warnings into a single if. 5140f805531SAlex Lyakas * It is useful to know what went wrong. 5150f805531SAlex Lyakas */ 5160f805531SAlex Lyakas if (WARN_ON(found_start != start)) 5170f805531SAlex Lyakas return -EUCLEAN; 5180f805531SAlex Lyakas if (WARN_ON(!PageUptodate(page))) 5190f805531SAlex Lyakas return -EUCLEAN; 5200f805531SAlex Lyakas 521de37aa51SNikolay Borisov ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid, 5229a8658e3SDavid Sterba offsetof(struct btrfs_header, fsid), 5239a8658e3SDavid Sterba BTRFS_FSID_SIZE) == 0); 5240f805531SAlex Lyakas 525c67b3892SDavid Sterba csum_tree_block(eb, result); 5262996e1f8SJohannes Thumshirn 5278d47a0d8SQu Wenruo if (btrfs_header_level(eb)) 5288d47a0d8SQu Wenruo ret = btrfs_check_node(eb); 5298d47a0d8SQu Wenruo else 5308d47a0d8SQu Wenruo ret = btrfs_check_leaf_full(eb); 5318d47a0d8SQu Wenruo 5328d47a0d8SQu Wenruo if (ret < 0) { 533c06631b0SQu Wenruo btrfs_print_tree(eb, 0); 5348d47a0d8SQu Wenruo btrfs_err(fs_info, 5358d47a0d8SQu Wenruo "block=%llu write time tree block corruption detected", 5368d47a0d8SQu Wenruo eb->start); 537c06631b0SQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 5388d47a0d8SQu Wenruo return ret; 5398d47a0d8SQu Wenruo } 5402996e1f8SJohannes Thumshirn write_extent_buffer(eb, result, 0, csum_size); 5418d47a0d8SQu Wenruo 5422996e1f8SJohannes Thumshirn return 0; 54319c00ddcSChris Mason } 54419c00ddcSChris Mason 545b0c9b3b0SDavid Sterba static int check_tree_block_fsid(struct extent_buffer *eb) 5462b82032cSYan Zheng { 547b0c9b3b0SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 54801d58472SDaniel Dressler struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 54944880fdcSAnand Jain u8 fsid[BTRFS_FSID_SIZE]; 5502b82032cSYan Zheng int ret = 1; 5512b82032cSYan Zheng 5529a8658e3SDavid Sterba read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid), 5539a8658e3SDavid Sterba BTRFS_FSID_SIZE); 5542b82032cSYan Zheng while (fs_devices) { 5557239ff4bSNikolay Borisov u8 *metadata_uuid; 5567239ff4bSNikolay Borisov 5577239ff4bSNikolay Borisov /* 5587239ff4bSNikolay Borisov * Checking the incompat flag is only valid for the current 5597239ff4bSNikolay Borisov * fs. For seed devices it's forbidden to have their uuid 5607239ff4bSNikolay Borisov * changed so reading ->fsid in this case is fine 5617239ff4bSNikolay Borisov */ 5627239ff4bSNikolay Borisov if (fs_devices == fs_info->fs_devices && 5637239ff4bSNikolay Borisov btrfs_fs_incompat(fs_info, METADATA_UUID)) 5647239ff4bSNikolay Borisov metadata_uuid = fs_devices->metadata_uuid; 5657239ff4bSNikolay Borisov else 5667239ff4bSNikolay Borisov metadata_uuid = fs_devices->fsid; 5677239ff4bSNikolay Borisov 5687239ff4bSNikolay Borisov if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) { 5692b82032cSYan Zheng ret = 0; 5702b82032cSYan Zheng break; 5712b82032cSYan Zheng } 5722b82032cSYan Zheng fs_devices = fs_devices->seed; 5732b82032cSYan Zheng } 5742b82032cSYan Zheng return ret; 5752b82032cSYan Zheng } 5762b82032cSYan Zheng 577facc8a22SMiao Xie static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio, 578facc8a22SMiao Xie u64 phy_offset, struct page *page, 579facc8a22SMiao Xie u64 start, u64 end, int mirror) 580ce9adaa5SChris Mason { 581ce9adaa5SChris Mason u64 found_start; 582ce9adaa5SChris Mason int found_level; 583ce9adaa5SChris Mason struct extent_buffer *eb; 58415b6e8a8SDavid Sterba struct btrfs_fs_info *fs_info; 58515b6e8a8SDavid Sterba u16 csum_size; 586f188591eSChris Mason int ret = 0; 5872996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 588727011e0SChris Mason int reads_done; 589ce9adaa5SChris Mason 590ce9adaa5SChris Mason if (!page->private) 591ce9adaa5SChris Mason goto out; 592d397712bSChris Mason 5934f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 59415b6e8a8SDavid Sterba fs_info = eb->fs_info; 59515b6e8a8SDavid Sterba csum_size = btrfs_super_csum_size(fs_info->super_copy); 596d397712bSChris Mason 5970b32f4bbSJosef Bacik /* the pending IO might have been the only thing that kept this buffer 5980b32f4bbSJosef Bacik * in memory. Make sure we have a ref for all this other checks 5990b32f4bbSJosef Bacik */ 60067439dadSDavid Sterba atomic_inc(&eb->refs); 6010b32f4bbSJosef Bacik 6020b32f4bbSJosef Bacik reads_done = atomic_dec_and_test(&eb->io_pages); 603727011e0SChris Mason if (!reads_done) 604727011e0SChris Mason goto err; 605f188591eSChris Mason 6065cf1ab56SJosef Bacik eb->read_mirror = mirror; 607656f30dbSFilipe Manana if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 608ea466794SJosef Bacik ret = -EIO; 609ea466794SJosef Bacik goto err; 610ea466794SJosef Bacik } 611ea466794SJosef Bacik 612ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 613727011e0SChris Mason if (found_start != eb->start) { 614893bf4b1SSu Yue btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu", 615893bf4b1SSu Yue eb->start, found_start); 616f188591eSChris Mason ret = -EIO; 617ce9adaa5SChris Mason goto err; 618ce9adaa5SChris Mason } 619b0c9b3b0SDavid Sterba if (check_tree_block_fsid(eb)) { 62002873e43SZhao Lei btrfs_err_rl(fs_info, "bad fsid on block %llu", 62194647322SDavid Sterba eb->start); 6221259ab75SChris Mason ret = -EIO; 6231259ab75SChris Mason goto err; 6241259ab75SChris Mason } 625ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 6261c24c3ceSJosef Bacik if (found_level >= BTRFS_MAX_LEVEL) { 627893bf4b1SSu Yue btrfs_err(fs_info, "bad tree block level %d on %llu", 628893bf4b1SSu Yue (int)btrfs_header_level(eb), eb->start); 6291c24c3ceSJosef Bacik ret = -EIO; 6301c24c3ceSJosef Bacik goto err; 6311c24c3ceSJosef Bacik } 632ce9adaa5SChris Mason 63385d4e461SChris Mason btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), 63485d4e461SChris Mason eb, found_level); 6354008c04aSChris Mason 636c67b3892SDavid Sterba csum_tree_block(eb, result); 637a826d6dcSJosef Bacik 6382996e1f8SJohannes Thumshirn if (memcmp_extent_buffer(eb, result, 0, csum_size)) { 6392996e1f8SJohannes Thumshirn u32 val; 6402996e1f8SJohannes Thumshirn u32 found = 0; 6412996e1f8SJohannes Thumshirn 6422996e1f8SJohannes Thumshirn memcpy(&found, result, csum_size); 6432996e1f8SJohannes Thumshirn 6442996e1f8SJohannes Thumshirn read_extent_buffer(eb, &val, 0, csum_size); 6452996e1f8SJohannes Thumshirn btrfs_warn_rl(fs_info, 6462996e1f8SJohannes Thumshirn "%s checksum verify failed on %llu wanted %x found %x level %d", 6472996e1f8SJohannes Thumshirn fs_info->sb->s_id, eb->start, 6482996e1f8SJohannes Thumshirn val, found, btrfs_header_level(eb)); 6492996e1f8SJohannes Thumshirn ret = -EUCLEAN; 6502996e1f8SJohannes Thumshirn goto err; 6512996e1f8SJohannes Thumshirn } 6522996e1f8SJohannes Thumshirn 653a826d6dcSJosef Bacik /* 654a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 655a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 656a826d6dcSJosef Bacik * return -EIO. 657a826d6dcSJosef Bacik */ 6581c4360eeSDavid Sterba if (found_level == 0 && btrfs_check_leaf_full(eb)) { 659a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 660a826d6dcSJosef Bacik ret = -EIO; 661a826d6dcSJosef Bacik } 662ce9adaa5SChris Mason 663813fd1dcSDavid Sterba if (found_level > 0 && btrfs_check_node(eb)) 664053ab70fSLiu Bo ret = -EIO; 665053ab70fSLiu Bo 6660b32f4bbSJosef Bacik if (!ret) 6670b32f4bbSJosef Bacik set_extent_buffer_uptodate(eb); 66875391f0dSQu Wenruo else 66975391f0dSQu Wenruo btrfs_err(fs_info, 67075391f0dSQu Wenruo "block=%llu read time tree block corruption detected", 67175391f0dSQu Wenruo eb->start); 672ce9adaa5SChris Mason err: 67379fb65a1SJosef Bacik if (reads_done && 67479fb65a1SJosef Bacik test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 675d48d71aaSDavid Sterba btree_readahead_hook(eb, ret); 6764bb31e92SArne Jansen 67753b381b3SDavid Woodhouse if (ret) { 67853b381b3SDavid Woodhouse /* 67953b381b3SDavid Woodhouse * our io error hook is going to dec the io pages 68053b381b3SDavid Woodhouse * again, we have to make sure it has something 68153b381b3SDavid Woodhouse * to decrement 68253b381b3SDavid Woodhouse */ 68353b381b3SDavid Woodhouse atomic_inc(&eb->io_pages); 6840b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 68553b381b3SDavid Woodhouse } 6860b32f4bbSJosef Bacik free_extent_buffer(eb); 687ce9adaa5SChris Mason out: 688f188591eSChris Mason return ret; 689ce9adaa5SChris Mason } 690ce9adaa5SChris Mason 6914246a0b6SChristoph Hellwig static void end_workqueue_bio(struct bio *bio) 692ce9adaa5SChris Mason { 69397eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq = bio->bi_private; 694ce9adaa5SChris Mason struct btrfs_fs_info *fs_info; 6959e0af237SLiu Bo struct btrfs_workqueue *wq; 696ce9adaa5SChris Mason 697ce9adaa5SChris Mason fs_info = end_io_wq->info; 6984e4cbee9SChristoph Hellwig end_io_wq->status = bio->bi_status; 699d20f7043SChris Mason 70037226b21SMike Christie if (bio_op(bio) == REQ_OP_WRITE) { 701a0cac0ecSOmar Sandoval if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) 7029e0af237SLiu Bo wq = fs_info->endio_meta_write_workers; 703a0cac0ecSOmar Sandoval else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) 7049e0af237SLiu Bo wq = fs_info->endio_freespace_worker; 705a0cac0ecSOmar Sandoval else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 7069e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 707a0cac0ecSOmar Sandoval else 7089e0af237SLiu Bo wq = fs_info->endio_write_workers; 7099e0af237SLiu Bo } else { 7105c047a69SOmar Sandoval if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 7119e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 712a0cac0ecSOmar Sandoval else if (end_io_wq->metadata) 7139e0af237SLiu Bo wq = fs_info->endio_meta_workers; 714a0cac0ecSOmar Sandoval else 7159e0af237SLiu Bo wq = fs_info->endio_workers; 7169e0af237SLiu Bo } 7179e0af237SLiu Bo 718a0cac0ecSOmar Sandoval btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL); 7199e0af237SLiu Bo btrfs_queue_work(wq, &end_io_wq->work); 720ce9adaa5SChris Mason } 721ce9adaa5SChris Mason 7224e4cbee9SChristoph Hellwig blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 723bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata) 7240b86a832SChris Mason { 72597eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 7268b110e39SMiao Xie 72797eb6b69SDavid Sterba end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS); 728ce9adaa5SChris Mason if (!end_io_wq) 7294e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 730ce9adaa5SChris Mason 731ce9adaa5SChris Mason end_io_wq->private = bio->bi_private; 732ce9adaa5SChris Mason end_io_wq->end_io = bio->bi_end_io; 73322c59948SChris Mason end_io_wq->info = info; 7344e4cbee9SChristoph Hellwig end_io_wq->status = 0; 735ce9adaa5SChris Mason end_io_wq->bio = bio; 73622c59948SChris Mason end_io_wq->metadata = metadata; 737ce9adaa5SChris Mason 738ce9adaa5SChris Mason bio->bi_private = end_io_wq; 739ce9adaa5SChris Mason bio->bi_end_io = end_workqueue_bio; 74022c59948SChris Mason return 0; 74122c59948SChris Mason } 74222c59948SChris Mason 7434a69a410SChris Mason static void run_one_async_start(struct btrfs_work *work) 7444a69a410SChris Mason { 7454a69a410SChris Mason struct async_submit_bio *async; 7464e4cbee9SChristoph Hellwig blk_status_t ret; 7474a69a410SChris Mason 7484a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 749c6100a4bSJosef Bacik ret = async->submit_bio_start(async->private_data, async->bio, 750eaf25d93SChris Mason async->bio_offset); 75179787eaaSJeff Mahoney if (ret) 7524e4cbee9SChristoph Hellwig async->status = ret; 7534a69a410SChris Mason } 7544a69a410SChris Mason 75506ea01b1SDavid Sterba /* 75606ea01b1SDavid Sterba * In order to insert checksums into the metadata in large chunks, we wait 75706ea01b1SDavid Sterba * until bio submission time. All the pages in the bio are checksummed and 75806ea01b1SDavid Sterba * sums are attached onto the ordered extent record. 75906ea01b1SDavid Sterba * 76006ea01b1SDavid Sterba * At IO completion time the csums attached on the ordered extent record are 76106ea01b1SDavid Sterba * inserted into the tree. 76206ea01b1SDavid Sterba */ 7634a69a410SChris Mason static void run_one_async_done(struct btrfs_work *work) 7648b712842SChris Mason { 7658b712842SChris Mason struct async_submit_bio *async; 76606ea01b1SDavid Sterba struct inode *inode; 76706ea01b1SDavid Sterba blk_status_t ret; 7688b712842SChris Mason 7698b712842SChris Mason async = container_of(work, struct async_submit_bio, work); 77006ea01b1SDavid Sterba inode = async->private_data; 7714854ddd0SChris Mason 772bb7ab3b9SAdam Buchbinder /* If an error occurred we just want to clean up the bio and move on */ 7734e4cbee9SChristoph Hellwig if (async->status) { 7744e4cbee9SChristoph Hellwig async->bio->bi_status = async->status; 7754246a0b6SChristoph Hellwig bio_endio(async->bio); 77679787eaaSJeff Mahoney return; 77779787eaaSJeff Mahoney } 77879787eaaSJeff Mahoney 779ec39f769SChris Mason /* 780ec39f769SChris Mason * All of the bios that pass through here are from async helpers. 781ec39f769SChris Mason * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context. 782ec39f769SChris Mason * This changes nothing when cgroups aren't in use. 783ec39f769SChris Mason */ 784ec39f769SChris Mason async->bio->bi_opf |= REQ_CGROUP_PUNT; 78508635baeSChris Mason ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num); 78606ea01b1SDavid Sterba if (ret) { 78706ea01b1SDavid Sterba async->bio->bi_status = ret; 78806ea01b1SDavid Sterba bio_endio(async->bio); 78906ea01b1SDavid Sterba } 7904a69a410SChris Mason } 7914a69a410SChris Mason 7924a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 7934a69a410SChris Mason { 7944a69a410SChris Mason struct async_submit_bio *async; 7954a69a410SChris Mason 7964a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 7978b712842SChris Mason kfree(async); 7988b712842SChris Mason } 7998b712842SChris Mason 8008c27cb35SLinus Torvalds blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio, 801c6100a4bSJosef Bacik int mirror_num, unsigned long bio_flags, 802c6100a4bSJosef Bacik u64 bio_offset, void *private_data, 803e288c080SDavid Sterba extent_submit_bio_start_t *submit_bio_start) 80444b8bd7eSChris Mason { 80544b8bd7eSChris Mason struct async_submit_bio *async; 80644b8bd7eSChris Mason 80744b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 80844b8bd7eSChris Mason if (!async) 8094e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 81044b8bd7eSChris Mason 811c6100a4bSJosef Bacik async->private_data = private_data; 81244b8bd7eSChris Mason async->bio = bio; 81344b8bd7eSChris Mason async->mirror_num = mirror_num; 8144a69a410SChris Mason async->submit_bio_start = submit_bio_start; 8154a69a410SChris Mason 816a0cac0ecSOmar Sandoval btrfs_init_work(&async->work, run_one_async_start, run_one_async_done, 817a0cac0ecSOmar Sandoval run_one_async_free); 8184a69a410SChris Mason 819eaf25d93SChris Mason async->bio_offset = bio_offset; 8208c8bee1dSChris Mason 8214e4cbee9SChristoph Hellwig async->status = 0; 82279787eaaSJeff Mahoney 82367f055c7SChristoph Hellwig if (op_is_sync(bio->bi_opf)) 8245cdc7ad3SQu Wenruo btrfs_set_work_high_priority(&async->work); 825d313d7a3SChris Mason 8265cdc7ad3SQu Wenruo btrfs_queue_work(fs_info->workers, &async->work); 82744b8bd7eSChris Mason return 0; 82844b8bd7eSChris Mason } 82944b8bd7eSChris Mason 8304e4cbee9SChristoph Hellwig static blk_status_t btree_csum_one_bio(struct bio *bio) 831ce3ed71aSChris Mason { 8322c30c71bSKent Overstreet struct bio_vec *bvec; 833ce3ed71aSChris Mason struct btrfs_root *root; 8342b070cfeSChristoph Hellwig int ret = 0; 8356dc4f100SMing Lei struct bvec_iter_all iter_all; 836ce3ed71aSChris Mason 837c09abff8SDavid Sterba ASSERT(!bio_flagged(bio, BIO_CLONED)); 8382b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 839ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 84001d58472SDaniel Dressler ret = csum_dirty_buffer(root->fs_info, bvec->bv_page); 84179787eaaSJeff Mahoney if (ret) 84279787eaaSJeff Mahoney break; 843ce3ed71aSChris Mason } 8442c30c71bSKent Overstreet 8454e4cbee9SChristoph Hellwig return errno_to_blk_status(ret); 846ce3ed71aSChris Mason } 847ce3ed71aSChris Mason 848d0ee3934SDavid Sterba static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio, 849eaf25d93SChris Mason u64 bio_offset) 85022c59948SChris Mason { 8518b712842SChris Mason /* 8528b712842SChris Mason * when we're called for a write, we're already in the async 8535443be45SChris Mason * submission context. Just jump into btrfs_map_bio 8548b712842SChris Mason */ 85579787eaaSJeff Mahoney return btree_csum_one_bio(bio); 85622c59948SChris Mason } 85722c59948SChris Mason 8589b4e675aSDavid Sterba static int check_async_write(struct btrfs_fs_info *fs_info, 8599b4e675aSDavid Sterba struct btrfs_inode *bi) 860de0022b9SJosef Bacik { 8616300463bSLiu Bo if (atomic_read(&bi->sync_writers)) 8626300463bSLiu Bo return 0; 8639b4e675aSDavid Sterba if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags)) 864de0022b9SJosef Bacik return 0; 865de0022b9SJosef Bacik return 1; 866de0022b9SJosef Bacik } 867de0022b9SJosef Bacik 868a56b1c7bSNikolay Borisov static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio, 86950489a57SNikolay Borisov int mirror_num, 87050489a57SNikolay Borisov unsigned long bio_flags) 87144b8bd7eSChris Mason { 8720b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 8739b4e675aSDavid Sterba int async = check_async_write(fs_info, BTRFS_I(inode)); 8744e4cbee9SChristoph Hellwig blk_status_t ret; 875cad321adSChris Mason 87637226b21SMike Christie if (bio_op(bio) != REQ_OP_WRITE) { 877cad321adSChris Mason /* 878cad321adSChris Mason * called for a read, do the setup so that checksum validation 879cad321adSChris Mason * can happen in the async kernel threads 880cad321adSChris Mason */ 8810b246afaSJeff Mahoney ret = btrfs_bio_wq_end_io(fs_info, bio, 8820b246afaSJeff Mahoney BTRFS_WQ_ENDIO_METADATA); 8831d4284bdSChris Mason if (ret) 88461891923SStefan Behrens goto out_w_error; 88508635baeSChris Mason ret = btrfs_map_bio(fs_info, bio, mirror_num); 886de0022b9SJosef Bacik } else if (!async) { 887de0022b9SJosef Bacik ret = btree_csum_one_bio(bio); 888de0022b9SJosef Bacik if (ret) 88961891923SStefan Behrens goto out_w_error; 89008635baeSChris Mason ret = btrfs_map_bio(fs_info, bio, mirror_num); 89161891923SStefan Behrens } else { 892cad321adSChris Mason /* 89361891923SStefan Behrens * kthread helpers are used to submit writes so that 89461891923SStefan Behrens * checksumming can happen in parallel across all CPUs 895cad321adSChris Mason */ 896c6100a4bSJosef Bacik ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0, 897e68f2ee7SNikolay Borisov 0, inode, btree_submit_bio_start); 89844b8bd7eSChris Mason } 89944b8bd7eSChris Mason 9004246a0b6SChristoph Hellwig if (ret) 9014246a0b6SChristoph Hellwig goto out_w_error; 9024246a0b6SChristoph Hellwig return 0; 9034246a0b6SChristoph Hellwig 90461891923SStefan Behrens out_w_error: 9054e4cbee9SChristoph Hellwig bio->bi_status = ret; 9064246a0b6SChristoph Hellwig bio_endio(bio); 90761891923SStefan Behrens return ret; 90861891923SStefan Behrens } 90961891923SStefan Behrens 9103dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 911784b4e29SChris Mason static int btree_migratepage(struct address_space *mapping, 912a6bc32b8SMel Gorman struct page *newpage, struct page *page, 913a6bc32b8SMel Gorman enum migrate_mode mode) 914784b4e29SChris Mason { 915784b4e29SChris Mason /* 916784b4e29SChris Mason * we can't safely write a btree page from here, 917784b4e29SChris Mason * we haven't done the locking hook 918784b4e29SChris Mason */ 919784b4e29SChris Mason if (PageDirty(page)) 920784b4e29SChris Mason return -EAGAIN; 921784b4e29SChris Mason /* 922784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 923784b4e29SChris Mason * We must have no buffers or drop them. 924784b4e29SChris Mason */ 925784b4e29SChris Mason if (page_has_private(page) && 926784b4e29SChris Mason !try_to_release_page(page, GFP_KERNEL)) 927784b4e29SChris Mason return -EAGAIN; 928a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 929784b4e29SChris Mason } 9303dd1462eSJan Beulich #endif 931784b4e29SChris Mason 9320da5468fSChris Mason 9330da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 9340da5468fSChris Mason struct writeback_control *wbc) 9350da5468fSChris Mason { 936e2d84521SMiao Xie struct btrfs_fs_info *fs_info; 937e2d84521SMiao Xie int ret; 938e2d84521SMiao Xie 939d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 940448d640bSChris Mason 941448d640bSChris Mason if (wbc->for_kupdate) 942448d640bSChris Mason return 0; 943448d640bSChris Mason 944e2d84521SMiao Xie fs_info = BTRFS_I(mapping->host)->root->fs_info; 945b9473439SChris Mason /* this is a bit racy, but that's ok */ 946d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 947d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 948d814a491SEthan Lien fs_info->dirty_metadata_batch); 949e2d84521SMiao Xie if (ret < 0) 950793955bcSChris Mason return 0; 951793955bcSChris Mason } 9520b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 9530da5468fSChris Mason } 9540da5468fSChris Mason 955b2950863SChristoph Hellwig static int btree_readpage(struct file *file, struct page *page) 9565f39d397SChris Mason { 95771ad38b4SDavid Sterba return extent_read_full_page(page, btree_get_extent, 0); 9585f39d397SChris Mason } 9595f39d397SChris Mason 96070dec807SChris Mason static int btree_releasepage(struct page *page, gfp_t gfp_flags) 9615f39d397SChris Mason { 96298509cfcSChris Mason if (PageWriteback(page) || PageDirty(page)) 96398509cfcSChris Mason return 0; 9640c4e538bSDavid Sterba 965f7a52a40SDavid Sterba return try_release_extent_buffer(page); 966d98237b3SChris Mason } 967d98237b3SChris Mason 968d47992f8SLukas Czerner static void btree_invalidatepage(struct page *page, unsigned int offset, 969d47992f8SLukas Czerner unsigned int length) 970d98237b3SChris Mason { 971d1310b2eSChris Mason struct extent_io_tree *tree; 972d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9735f39d397SChris Mason extent_invalidatepage(tree, page, offset); 9745f39d397SChris Mason btree_releasepage(page, GFP_NOFS); 9759ad6b7bcSChris Mason if (PagePrivate(page)) { 976efe120a0SFrank Holton btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info, 977efe120a0SFrank Holton "page private not zero on page %llu", 978efe120a0SFrank Holton (unsigned long long)page_offset(page)); 9799ad6b7bcSChris Mason ClearPagePrivate(page); 9809ad6b7bcSChris Mason set_page_private(page, 0); 98109cbfeafSKirill A. Shutemov put_page(page); 9829ad6b7bcSChris Mason } 983d98237b3SChris Mason } 984d98237b3SChris Mason 9850b32f4bbSJosef Bacik static int btree_set_page_dirty(struct page *page) 9860b32f4bbSJosef Bacik { 987bb146eb2SJosef Bacik #ifdef DEBUG 9880b32f4bbSJosef Bacik struct extent_buffer *eb; 9890b32f4bbSJosef Bacik 9900b32f4bbSJosef Bacik BUG_ON(!PagePrivate(page)); 9910b32f4bbSJosef Bacik eb = (struct extent_buffer *)page->private; 9920b32f4bbSJosef Bacik BUG_ON(!eb); 9930b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 9940b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 9950b32f4bbSJosef Bacik btrfs_assert_tree_locked(eb); 996bb146eb2SJosef Bacik #endif 9970b32f4bbSJosef Bacik return __set_page_dirty_nobuffers(page); 9980b32f4bbSJosef Bacik } 9990b32f4bbSJosef Bacik 10007f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 1001d98237b3SChris Mason .readpage = btree_readpage, 10020da5468fSChris Mason .writepages = btree_writepages, 10035f39d397SChris Mason .releasepage = btree_releasepage, 10045f39d397SChris Mason .invalidatepage = btree_invalidatepage, 10055a92bc88SChris Mason #ifdef CONFIG_MIGRATION 1006784b4e29SChris Mason .migratepage = btree_migratepage, 10075a92bc88SChris Mason #endif 10080b32f4bbSJosef Bacik .set_page_dirty = btree_set_page_dirty, 1009d98237b3SChris Mason }; 1010123abc88SChris Mason 10112ff7e61eSJeff Mahoney void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr) 1012090d1875SChris Mason { 10135f39d397SChris Mason struct extent_buffer *buf = NULL; 1014537f38f0SNikolay Borisov int ret; 1015090d1875SChris Mason 10162ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 1017c871b0f2SLiu Bo if (IS_ERR(buf)) 10186197d86eSDavid Sterba return; 1019537f38f0SNikolay Borisov 1020c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(buf, WAIT_NONE, 0); 1021537f38f0SNikolay Borisov if (ret < 0) 1022537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 1023537f38f0SNikolay Borisov else 10245f39d397SChris Mason free_extent_buffer(buf); 1025090d1875SChris Mason } 1026090d1875SChris Mason 10272ff7e61eSJeff Mahoney struct extent_buffer *btrfs_find_create_tree_block( 10282ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info, 10290308af44SDavid Sterba u64 bytenr) 10300999df54SChris Mason { 10310b246afaSJeff Mahoney if (btrfs_is_testing(fs_info)) 10320b246afaSJeff Mahoney return alloc_test_extent_buffer(fs_info, bytenr); 10330b246afaSJeff Mahoney return alloc_extent_buffer(fs_info, bytenr); 10340999df54SChris Mason } 10350999df54SChris Mason 1036581c1760SQu Wenruo /* 1037581c1760SQu Wenruo * Read tree block at logical address @bytenr and do variant basic but critical 1038581c1760SQu Wenruo * verification. 1039581c1760SQu Wenruo * 1040581c1760SQu Wenruo * @parent_transid: expected transid of this tree block, skip check if 0 1041581c1760SQu Wenruo * @level: expected level, mandatory check 1042581c1760SQu Wenruo * @first_key: expected key in slot 0, skip check if NULL 1043581c1760SQu Wenruo */ 10442ff7e61eSJeff Mahoney struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, 1045581c1760SQu Wenruo u64 parent_transid, int level, 1046581c1760SQu Wenruo struct btrfs_key *first_key) 1047e20d96d6SChris Mason { 10485f39d397SChris Mason struct extent_buffer *buf = NULL; 104919c00ddcSChris Mason int ret; 105019c00ddcSChris Mason 10512ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 1052c871b0f2SLiu Bo if (IS_ERR(buf)) 1053c871b0f2SLiu Bo return buf; 1054e4204dedSChris Mason 10555ab12d1fSDavid Sterba ret = btree_read_extent_buffer_pages(buf, parent_transid, 1056581c1760SQu Wenruo level, first_key); 10570f0fe8f7SFilipe David Borba Manana if (ret) { 1058537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 105964c043deSLiu Bo return ERR_PTR(ret); 10600f0fe8f7SFilipe David Borba Manana } 10615f39d397SChris Mason return buf; 1062ce9adaa5SChris Mason 1063eb60ceacSChris Mason } 1064eb60ceacSChris Mason 10656a884d7dSDavid Sterba void btrfs_clean_tree_block(struct extent_buffer *buf) 1066ed2ff2cbSChris Mason { 10676a884d7dSDavid Sterba struct btrfs_fs_info *fs_info = buf->fs_info; 106855c69072SChris Mason if (btrfs_header_generation(buf) == 1069e2d84521SMiao Xie fs_info->running_transaction->transid) { 1070b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1071b4ce94deSChris Mason 1072b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1073104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 1074e2d84521SMiao Xie -buf->len, 1075e2d84521SMiao Xie fs_info->dirty_metadata_batch); 1076b9473439SChris Mason /* ugh, clear_extent_buffer_dirty needs to lock the page */ 10778bead258SDavid Sterba btrfs_set_lock_blocking_write(buf); 10780b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 1079925baeddSChris Mason } 10805f39d397SChris Mason } 1081ed7b63ebSJosef Bacik } 10825f39d397SChris Mason 1083da17066cSJeff Mahoney static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, 1084e20d96d6SChris Mason u64 objectid) 1085d97e63b6SChris Mason { 10867c0260eeSJeff Mahoney bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); 108796dfcb46SJosef Bacik root->fs_info = fs_info; 1088cfaa7295SChris Mason root->node = NULL; 1089a28ec197SChris Mason root->commit_root = NULL; 109027cdeb70SMiao Xie root->state = 0; 1091d68fc57bSYan, Zheng root->orphan_cleanup_state = 0; 10920b86a832SChris Mason 10930f7d52f4SChris Mason root->last_trans = 0; 109413a8a7c8SYan, Zheng root->highest_objectid = 0; 1095eb73c1b7SMiao Xie root->nr_delalloc_inodes = 0; 1096199c2a9cSMiao Xie root->nr_ordered_extents = 0; 10976bef4d31SEric Paris root->inode_tree = RB_ROOT; 109816cdcec7SMiao Xie INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1099f0486c68SYan, Zheng root->block_rsv = NULL; 11000b86a832SChris Mason 11010b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 11025d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 1103eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_inodes); 1104eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_root); 1105199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_extents); 1106199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_root); 1107d2311e69SQu Wenruo INIT_LIST_HEAD(&root->reloc_dirty_list); 11082ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 11092ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 11105d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1111eb73c1b7SMiao Xie spin_lock_init(&root->delalloc_lock); 1112199c2a9cSMiao Xie spin_lock_init(&root->ordered_extent_lock); 1113f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 11142ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 11152ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 11168287475aSQu Wenruo spin_lock_init(&root->qgroup_meta_rsv_lock); 1117a2135011SChris Mason mutex_init(&root->objectid_mutex); 1118e02119d5SChris Mason mutex_init(&root->log_mutex); 111931f3d255SMiao Xie mutex_init(&root->ordered_extent_mutex); 1120573bfb72SMiao Xie mutex_init(&root->delalloc_mutex); 11217237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 11227237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 11237237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 11248b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[0]); 11258b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[1]); 11267237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 11277237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 11287237f183SYan Zheng atomic_set(&root->log_writers, 0); 11292ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 11300700cea7SElena Reshetova refcount_set(&root->refs, 1); 11318ecebf4dSRobbie Ko atomic_set(&root->snapshot_force_cow, 0); 1132eede2bf3SOmar Sandoval atomic_set(&root->nr_swapfiles, 0); 11337237f183SYan Zheng root->log_transid = 0; 1134d1433debSMiao Xie root->log_transid_committed = -1; 1135257c62e1SChris Mason root->last_log_commit = 0; 11367c0260eeSJeff Mahoney if (!dummy) 113743eb5f29SQu Wenruo extent_io_tree_init(fs_info, &root->dirty_log_pages, 113843eb5f29SQu Wenruo IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL); 1139017e5369SChris Mason 11403768f368SChris Mason memset(&root->root_key, 0, sizeof(root->root_key)); 11413768f368SChris Mason memset(&root->root_item, 0, sizeof(root->root_item)); 11426702ed49SChris Mason memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 11437c0260eeSJeff Mahoney if (!dummy) 11443f157a2fSChris Mason root->defrag_trans_start = fs_info->generation; 114506ea65a3SJosef Bacik else 114606ea65a3SJosef Bacik root->defrag_trans_start = 0; 11474d775673SChris Mason root->root_key.objectid = objectid; 11480ee5dc67SAl Viro root->anon_dev = 0; 11498ea05e3aSAlexander Block 11505f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 1151370a11b8SQu Wenruo btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks); 1152bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 1153bd647ce3SJosef Bacik INIT_LIST_HEAD(&root->leak_list); 1154bd647ce3SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 1155bd647ce3SJosef Bacik list_add_tail(&root->leak_list, &fs_info->allocated_roots); 1156bd647ce3SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 1157bd647ce3SJosef Bacik #endif 11583768f368SChris Mason } 11593768f368SChris Mason 116074e4d827SDavid Sterba static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, 116196dfcb46SJosef Bacik u64 objectid, gfp_t flags) 11626f07e42eSAl Viro { 116374e4d827SDavid Sterba struct btrfs_root *root = kzalloc(sizeof(*root), flags); 11646f07e42eSAl Viro if (root) 116596dfcb46SJosef Bacik __setup_root(root, fs_info, objectid); 11666f07e42eSAl Viro return root; 11676f07e42eSAl Viro } 11686f07e42eSAl Viro 116906ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 117006ea65a3SJosef Bacik /* Should only be used by the testing infrastructure */ 1171da17066cSJeff Mahoney struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) 117206ea65a3SJosef Bacik { 117306ea65a3SJosef Bacik struct btrfs_root *root; 117406ea65a3SJosef Bacik 11757c0260eeSJeff Mahoney if (!fs_info) 11767c0260eeSJeff Mahoney return ERR_PTR(-EINVAL); 11777c0260eeSJeff Mahoney 117896dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL); 117906ea65a3SJosef Bacik if (!root) 118006ea65a3SJosef Bacik return ERR_PTR(-ENOMEM); 1181da17066cSJeff Mahoney 1182b9ef22deSFeifei Xu /* We don't use the stripesize in selftest, set it as sectorsize */ 1183faa2dbf0SJosef Bacik root->alloc_bytenr = 0; 118406ea65a3SJosef Bacik 118506ea65a3SJosef Bacik return root; 118606ea65a3SJosef Bacik } 118706ea65a3SJosef Bacik #endif 118806ea65a3SJosef Bacik 118920897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 119020897f5cSArne Jansen u64 objectid) 119120897f5cSArne Jansen { 11929b7a2440SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 119320897f5cSArne Jansen struct extent_buffer *leaf; 119420897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 119520897f5cSArne Jansen struct btrfs_root *root; 119620897f5cSArne Jansen struct btrfs_key key; 1197b89f6d1fSFilipe Manana unsigned int nofs_flag; 119820897f5cSArne Jansen int ret = 0; 119920897f5cSArne Jansen 1200b89f6d1fSFilipe Manana /* 1201b89f6d1fSFilipe Manana * We're holding a transaction handle, so use a NOFS memory allocation 1202b89f6d1fSFilipe Manana * context to avoid deadlock if reclaim happens. 1203b89f6d1fSFilipe Manana */ 1204b89f6d1fSFilipe Manana nofs_flag = memalloc_nofs_save(); 120596dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL); 1206b89f6d1fSFilipe Manana memalloc_nofs_restore(nofs_flag); 120720897f5cSArne Jansen if (!root) 120820897f5cSArne Jansen return ERR_PTR(-ENOMEM); 120920897f5cSArne Jansen 121020897f5cSArne Jansen root->root_key.objectid = objectid; 121120897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 121220897f5cSArne Jansen root->root_key.offset = 0; 121320897f5cSArne Jansen 12144d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0); 121520897f5cSArne Jansen if (IS_ERR(leaf)) { 121620897f5cSArne Jansen ret = PTR_ERR(leaf); 12171dd05682STsutomu Itoh leaf = NULL; 121820897f5cSArne Jansen goto fail; 121920897f5cSArne Jansen } 122020897f5cSArne Jansen 122120897f5cSArne Jansen root->node = leaf; 122220897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 122320897f5cSArne Jansen 122420897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 122527cdeb70SMiao Xie set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 122620897f5cSArne Jansen 122720897f5cSArne Jansen root->root_item.flags = 0; 122820897f5cSArne Jansen root->root_item.byte_limit = 0; 122920897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 123020897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 123120897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 123220897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 123320897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 123420897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 123520897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 123633d85fdaSQu Wenruo if (is_fstree(objectid)) 1237807fc790SAndy Shevchenko generate_random_guid(root->root_item.uuid); 1238807fc790SAndy Shevchenko else 1239807fc790SAndy Shevchenko export_guid(root->root_item.uuid, &guid_null); 124020897f5cSArne Jansen root->root_item.drop_level = 0; 124120897f5cSArne Jansen 124220897f5cSArne Jansen key.objectid = objectid; 124320897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 124420897f5cSArne Jansen key.offset = 0; 124520897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 124620897f5cSArne Jansen if (ret) 124720897f5cSArne Jansen goto fail; 124820897f5cSArne Jansen 124920897f5cSArne Jansen btrfs_tree_unlock(leaf); 125020897f5cSArne Jansen 125120897f5cSArne Jansen return root; 12521dd05682STsutomu Itoh 12531dd05682STsutomu Itoh fail: 12548c38938cSJosef Bacik if (leaf) 12551dd05682STsutomu Itoh btrfs_tree_unlock(leaf); 125600246528SJosef Bacik btrfs_put_root(root); 12571dd05682STsutomu Itoh 12581dd05682STsutomu Itoh return ERR_PTR(ret); 125920897f5cSArne Jansen } 126020897f5cSArne Jansen 12617237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1262e02119d5SChris Mason struct btrfs_fs_info *fs_info) 12630f7d52f4SChris Mason { 12640f7d52f4SChris Mason struct btrfs_root *root; 12657237f183SYan Zheng struct extent_buffer *leaf; 1266e02119d5SChris Mason 126796dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS); 1268e02119d5SChris Mason if (!root) 12697237f183SYan Zheng return ERR_PTR(-ENOMEM); 1270e02119d5SChris Mason 1271e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1272e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1273e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 127427cdeb70SMiao Xie 12757237f183SYan Zheng /* 127692a7cc42SQu Wenruo * DON'T set SHAREABLE bit for log trees. 127727cdeb70SMiao Xie * 127892a7cc42SQu Wenruo * Log trees are not exposed to user space thus can't be snapshotted, 127992a7cc42SQu Wenruo * and they go away before a real commit is actually done. 128092a7cc42SQu Wenruo * 128192a7cc42SQu Wenruo * They do store pointers to file data extents, and those reference 128292a7cc42SQu Wenruo * counts still get updated (along with back refs to the log tree). 12837237f183SYan Zheng */ 1284e02119d5SChris Mason 12854d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 12864d75f8a9SDavid Sterba NULL, 0, 0, 0); 12877237f183SYan Zheng if (IS_ERR(leaf)) { 128800246528SJosef Bacik btrfs_put_root(root); 12897237f183SYan Zheng return ERR_CAST(leaf); 12907237f183SYan Zheng } 1291e02119d5SChris Mason 12927237f183SYan Zheng root->node = leaf; 1293e02119d5SChris Mason 1294e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1295e02119d5SChris Mason btrfs_tree_unlock(root->node); 12967237f183SYan Zheng return root; 12977237f183SYan Zheng } 12987237f183SYan Zheng 12997237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 13007237f183SYan Zheng struct btrfs_fs_info *fs_info) 13017237f183SYan Zheng { 13027237f183SYan Zheng struct btrfs_root *log_root; 13037237f183SYan Zheng 13047237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 13057237f183SYan Zheng if (IS_ERR(log_root)) 13067237f183SYan Zheng return PTR_ERR(log_root); 13077237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 13087237f183SYan Zheng fs_info->log_root_tree = log_root; 13097237f183SYan Zheng return 0; 13107237f183SYan Zheng } 13117237f183SYan Zheng 13127237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 13137237f183SYan Zheng struct btrfs_root *root) 13147237f183SYan Zheng { 13150b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13167237f183SYan Zheng struct btrfs_root *log_root; 13177237f183SYan Zheng struct btrfs_inode_item *inode_item; 13187237f183SYan Zheng 13190b246afaSJeff Mahoney log_root = alloc_log_tree(trans, fs_info); 13207237f183SYan Zheng if (IS_ERR(log_root)) 13217237f183SYan Zheng return PTR_ERR(log_root); 13227237f183SYan Zheng 13237237f183SYan Zheng log_root->last_trans = trans->transid; 13247237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 13257237f183SYan Zheng 13267237f183SYan Zheng inode_item = &log_root->root_item.inode; 13273cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 13283cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 13293cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 1330da17066cSJeff Mahoney btrfs_set_stack_inode_nbytes(inode_item, 13310b246afaSJeff Mahoney fs_info->nodesize); 13323cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 13337237f183SYan Zheng 13345d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 13357237f183SYan Zheng 13367237f183SYan Zheng WARN_ON(root->log_root); 13377237f183SYan Zheng root->log_root = log_root; 13387237f183SYan Zheng root->log_transid = 0; 1339d1433debSMiao Xie root->log_transid_committed = -1; 1340257c62e1SChris Mason root->last_log_commit = 0; 1341e02119d5SChris Mason return 0; 1342e02119d5SChris Mason } 1343e02119d5SChris Mason 134462a2c73eSJosef Bacik struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 1345cb517eabSMiao Xie struct btrfs_key *key) 1346e02119d5SChris Mason { 1347e02119d5SChris Mason struct btrfs_root *root; 1348e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 13490f7d52f4SChris Mason struct btrfs_path *path; 135084234f3aSYan Zheng u64 generation; 1351cb517eabSMiao Xie int ret; 1352581c1760SQu Wenruo int level; 1353cb517eabSMiao Xie 1354cb517eabSMiao Xie path = btrfs_alloc_path(); 1355cb517eabSMiao Xie if (!path) 1356cb517eabSMiao Xie return ERR_PTR(-ENOMEM); 13570f7d52f4SChris Mason 135896dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS); 1359cb517eabSMiao Xie if (!root) { 1360cb517eabSMiao Xie ret = -ENOMEM; 1361cb517eabSMiao Xie goto alloc_fail; 13620f7d52f4SChris Mason } 13630f7d52f4SChris Mason 1364cb517eabSMiao Xie ret = btrfs_find_root(tree_root, key, path, 1365cb517eabSMiao Xie &root->root_item, &root->root_key); 13660f7d52f4SChris Mason if (ret) { 136713a8a7c8SYan, Zheng if (ret > 0) 136813a8a7c8SYan, Zheng ret = -ENOENT; 1369cb517eabSMiao Xie goto find_fail; 13700f7d52f4SChris Mason } 137113a8a7c8SYan, Zheng 137284234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1373581c1760SQu Wenruo level = btrfs_root_level(&root->root_item); 13742ff7e61eSJeff Mahoney root->node = read_tree_block(fs_info, 13752ff7e61eSJeff Mahoney btrfs_root_bytenr(&root->root_item), 1376581c1760SQu Wenruo generation, level, NULL); 137764c043deSLiu Bo if (IS_ERR(root->node)) { 137864c043deSLiu Bo ret = PTR_ERR(root->node); 13798c38938cSJosef Bacik root->node = NULL; 1380cb517eabSMiao Xie goto find_fail; 1381cb517eabSMiao Xie } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1382cb517eabSMiao Xie ret = -EIO; 138364c043deSLiu Bo goto find_fail; 1384416bc658SJosef Bacik } 13855d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 138613a8a7c8SYan, Zheng out: 1387cb517eabSMiao Xie btrfs_free_path(path); 1388cb517eabSMiao Xie return root; 1389cb517eabSMiao Xie 1390cb517eabSMiao Xie find_fail: 139100246528SJosef Bacik btrfs_put_root(root); 1392cb517eabSMiao Xie alloc_fail: 1393cb517eabSMiao Xie root = ERR_PTR(ret); 1394cb517eabSMiao Xie goto out; 1395cb517eabSMiao Xie } 1396cb517eabSMiao Xie 1397a98db0f3SJosef Bacik static int btrfs_init_fs_root(struct btrfs_root *root) 1398cb517eabSMiao Xie { 1399cb517eabSMiao Xie int ret; 1400dcc3eb96SNikolay Borisov unsigned int nofs_flag; 1401cb517eabSMiao Xie 1402cb517eabSMiao Xie root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); 1403cb517eabSMiao Xie root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), 1404cb517eabSMiao Xie GFP_NOFS); 1405cb517eabSMiao Xie if (!root->free_ino_pinned || !root->free_ino_ctl) { 1406cb517eabSMiao Xie ret = -ENOMEM; 1407cb517eabSMiao Xie goto fail; 1408cb517eabSMiao Xie } 1409cb517eabSMiao Xie 1410dcc3eb96SNikolay Borisov /* 1411dcc3eb96SNikolay Borisov * We might be called under a transaction (e.g. indirect backref 1412dcc3eb96SNikolay Borisov * resolution) which could deadlock if it triggers memory reclaim 1413dcc3eb96SNikolay Borisov */ 1414dcc3eb96SNikolay Borisov nofs_flag = memalloc_nofs_save(); 1415dcc3eb96SNikolay Borisov ret = btrfs_drew_lock_init(&root->snapshot_lock); 1416dcc3eb96SNikolay Borisov memalloc_nofs_restore(nofs_flag); 1417dcc3eb96SNikolay Borisov if (ret) 14188257b2dcSMiao Xie goto fail; 14198257b2dcSMiao Xie 1420aeb935a4SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID && 1421aeb935a4SQu Wenruo root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { 142292a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &root->state); 1423f39e4571SJosef Bacik btrfs_check_and_init_root_item(&root->root_item); 1424f39e4571SJosef Bacik } 1425f39e4571SJosef Bacik 1426cb517eabSMiao Xie btrfs_init_free_ino_ctl(root); 142757cdc8dbSDavid Sterba spin_lock_init(&root->ino_cache_lock); 142857cdc8dbSDavid Sterba init_waitqueue_head(&root->ino_cache_wait); 1429cb517eabSMiao Xie 1430cb517eabSMiao Xie ret = get_anon_bdev(&root->anon_dev); 1431cb517eabSMiao Xie if (ret) 1432876d2cf1SLiu Bo goto fail; 1433f32e48e9SChandan Rajendra 1434f32e48e9SChandan Rajendra mutex_lock(&root->objectid_mutex); 1435f32e48e9SChandan Rajendra ret = btrfs_find_highest_objectid(root, 1436f32e48e9SChandan Rajendra &root->highest_objectid); 1437f32e48e9SChandan Rajendra if (ret) { 1438f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1439876d2cf1SLiu Bo goto fail; 1440f32e48e9SChandan Rajendra } 1441f32e48e9SChandan Rajendra 1442f32e48e9SChandan Rajendra ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 1443f32e48e9SChandan Rajendra 1444f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1445f32e48e9SChandan Rajendra 1446cb517eabSMiao Xie return 0; 1447cb517eabSMiao Xie fail: 144884db5ccfSDavid Sterba /* The caller is responsible to call btrfs_free_fs_root */ 1449cb517eabSMiao Xie return ret; 1450cb517eabSMiao Xie } 1451cb517eabSMiao Xie 1452a98db0f3SJosef Bacik static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1453cb517eabSMiao Xie u64 root_id) 1454cb517eabSMiao Xie { 1455cb517eabSMiao Xie struct btrfs_root *root; 1456cb517eabSMiao Xie 1457cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1458cb517eabSMiao Xie root = radix_tree_lookup(&fs_info->fs_roots_radix, 1459cb517eabSMiao Xie (unsigned long)root_id); 1460bc44d7c4SJosef Bacik if (root) 146100246528SJosef Bacik root = btrfs_grab_root(root); 1462cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1463cb517eabSMiao Xie return root; 1464cb517eabSMiao Xie } 1465cb517eabSMiao Xie 1466cb517eabSMiao Xie int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1467cb517eabSMiao Xie struct btrfs_root *root) 1468cb517eabSMiao Xie { 1469cb517eabSMiao Xie int ret; 1470cb517eabSMiao Xie 1471e1860a77SDavid Sterba ret = radix_tree_preload(GFP_NOFS); 1472cb517eabSMiao Xie if (ret) 1473cb517eabSMiao Xie return ret; 1474cb517eabSMiao Xie 1475cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1476cb517eabSMiao Xie ret = radix_tree_insert(&fs_info->fs_roots_radix, 1477cb517eabSMiao Xie (unsigned long)root->root_key.objectid, 1478cb517eabSMiao Xie root); 1479af01d2e5SJosef Bacik if (ret == 0) { 148000246528SJosef Bacik btrfs_grab_root(root); 148127cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1482af01d2e5SJosef Bacik } 1483cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1484cb517eabSMiao Xie radix_tree_preload_end(); 1485cb517eabSMiao Xie 1486cb517eabSMiao Xie return ret; 1487cb517eabSMiao Xie } 1488cb517eabSMiao Xie 1489bd647ce3SJosef Bacik void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info) 1490bd647ce3SJosef Bacik { 1491bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 1492bd647ce3SJosef Bacik struct btrfs_root *root; 1493bd647ce3SJosef Bacik 1494bd647ce3SJosef Bacik while (!list_empty(&fs_info->allocated_roots)) { 1495bd647ce3SJosef Bacik root = list_first_entry(&fs_info->allocated_roots, 1496bd647ce3SJosef Bacik struct btrfs_root, leak_list); 1497bd647ce3SJosef Bacik btrfs_err(fs_info, "leaked root %llu-%llu refcount %d", 1498bd647ce3SJosef Bacik root->root_key.objectid, root->root_key.offset, 1499bd647ce3SJosef Bacik refcount_read(&root->refs)); 1500bd647ce3SJosef Bacik while (refcount_read(&root->refs) > 1) 150100246528SJosef Bacik btrfs_put_root(root); 150200246528SJosef Bacik btrfs_put_root(root); 1503bd647ce3SJosef Bacik } 1504bd647ce3SJosef Bacik #endif 1505bd647ce3SJosef Bacik } 1506bd647ce3SJosef Bacik 15070d4b0463SJosef Bacik void btrfs_free_fs_info(struct btrfs_fs_info *fs_info) 15080d4b0463SJosef Bacik { 1509141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 1510141386e1SJosef Bacik percpu_counter_destroy(&fs_info->delalloc_bytes); 1511141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dio_bytes); 1512141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 1513141386e1SJosef Bacik btrfs_free_csum_hash(fs_info); 1514141386e1SJosef Bacik btrfs_free_stripe_hash_table(fs_info); 1515141386e1SJosef Bacik btrfs_free_ref_cache(fs_info); 15160d4b0463SJosef Bacik kfree(fs_info->balance_ctl); 15170d4b0463SJosef Bacik kfree(fs_info->delayed_root); 151800246528SJosef Bacik btrfs_put_root(fs_info->extent_root); 151900246528SJosef Bacik btrfs_put_root(fs_info->tree_root); 152000246528SJosef Bacik btrfs_put_root(fs_info->chunk_root); 152100246528SJosef Bacik btrfs_put_root(fs_info->dev_root); 152200246528SJosef Bacik btrfs_put_root(fs_info->csum_root); 152300246528SJosef Bacik btrfs_put_root(fs_info->quota_root); 152400246528SJosef Bacik btrfs_put_root(fs_info->uuid_root); 152500246528SJosef Bacik btrfs_put_root(fs_info->free_space_root); 152600246528SJosef Bacik btrfs_put_root(fs_info->fs_root); 1527aeb935a4SQu Wenruo btrfs_put_root(fs_info->data_reloc_root); 1528bd647ce3SJosef Bacik btrfs_check_leaked_roots(fs_info); 15293fd63727SJosef Bacik btrfs_extent_buffer_leak_debug_check(fs_info); 15300d4b0463SJosef Bacik kfree(fs_info->super_copy); 15310d4b0463SJosef Bacik kfree(fs_info->super_for_commit); 15320d4b0463SJosef Bacik kvfree(fs_info); 15330d4b0463SJosef Bacik } 15340d4b0463SJosef Bacik 15350d4b0463SJosef Bacik 1536c00869f1SMiao Xie struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 153756e9357aSDavid Sterba u64 objectid, bool check_ref) 15385eda7b5eSChris Mason { 15395eda7b5eSChris Mason struct btrfs_root *root; 1540381cf658SDavid Sterba struct btrfs_path *path; 15411d4c08e0SDavid Sterba struct btrfs_key key; 15425eda7b5eSChris Mason int ret; 15435eda7b5eSChris Mason 154456e9357aSDavid Sterba if (objectid == BTRFS_ROOT_TREE_OBJECTID) 154500246528SJosef Bacik return btrfs_grab_root(fs_info->tree_root); 154656e9357aSDavid Sterba if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 154700246528SJosef Bacik return btrfs_grab_root(fs_info->extent_root); 154856e9357aSDavid Sterba if (objectid == BTRFS_CHUNK_TREE_OBJECTID) 154900246528SJosef Bacik return btrfs_grab_root(fs_info->chunk_root); 155056e9357aSDavid Sterba if (objectid == BTRFS_DEV_TREE_OBJECTID) 155100246528SJosef Bacik return btrfs_grab_root(fs_info->dev_root); 155256e9357aSDavid Sterba if (objectid == BTRFS_CSUM_TREE_OBJECTID) 155300246528SJosef Bacik return btrfs_grab_root(fs_info->csum_root); 155456e9357aSDavid Sterba if (objectid == BTRFS_QUOTA_TREE_OBJECTID) 155500246528SJosef Bacik return btrfs_grab_root(fs_info->quota_root) ? 1556bc44d7c4SJosef Bacik fs_info->quota_root : ERR_PTR(-ENOENT); 155756e9357aSDavid Sterba if (objectid == BTRFS_UUID_TREE_OBJECTID) 155800246528SJosef Bacik return btrfs_grab_root(fs_info->uuid_root) ? 1559bc44d7c4SJosef Bacik fs_info->uuid_root : ERR_PTR(-ENOENT); 156056e9357aSDavid Sterba if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) 156100246528SJosef Bacik return btrfs_grab_root(fs_info->free_space_root) ? 1562bc44d7c4SJosef Bacik fs_info->free_space_root : ERR_PTR(-ENOENT); 15634df27c4dSYan, Zheng again: 156456e9357aSDavid Sterba root = btrfs_lookup_fs_root(fs_info, objectid); 156548475471SStefan Behrens if (root) { 1566bc44d7c4SJosef Bacik if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 156700246528SJosef Bacik btrfs_put_root(root); 156848475471SStefan Behrens return ERR_PTR(-ENOENT); 1569bc44d7c4SJosef Bacik } 15705eda7b5eSChris Mason return root; 157148475471SStefan Behrens } 15725eda7b5eSChris Mason 157356e9357aSDavid Sterba key.objectid = objectid; 157456e9357aSDavid Sterba key.type = BTRFS_ROOT_ITEM_KEY; 157556e9357aSDavid Sterba key.offset = (u64)-1; 157656e9357aSDavid Sterba root = btrfs_read_tree_root(fs_info->tree_root, &key); 15775eda7b5eSChris Mason if (IS_ERR(root)) 15785eda7b5eSChris Mason return root; 15793394e160SChris Mason 1580c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1581d68fc57bSYan, Zheng ret = -ENOENT; 1582d68fc57bSYan, Zheng goto fail; 1583d68fc57bSYan, Zheng } 1584d68fc57bSYan, Zheng 1585cb517eabSMiao Xie ret = btrfs_init_fs_root(root); 1586cb517eabSMiao Xie if (ret) 1587cb517eabSMiao Xie goto fail; 1588cb517eabSMiao Xie 1589381cf658SDavid Sterba path = btrfs_alloc_path(); 1590381cf658SDavid Sterba if (!path) { 1591381cf658SDavid Sterba ret = -ENOMEM; 1592381cf658SDavid Sterba goto fail; 1593381cf658SDavid Sterba } 15941d4c08e0SDavid Sterba key.objectid = BTRFS_ORPHAN_OBJECTID; 15951d4c08e0SDavid Sterba key.type = BTRFS_ORPHAN_ITEM_KEY; 159656e9357aSDavid Sterba key.offset = objectid; 15971d4c08e0SDavid Sterba 15981d4c08e0SDavid Sterba ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1599381cf658SDavid Sterba btrfs_free_path(path); 1600d68fc57bSYan, Zheng if (ret < 0) 1601d68fc57bSYan, Zheng goto fail; 1602d68fc57bSYan, Zheng if (ret == 0) 160327cdeb70SMiao Xie set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1604d68fc57bSYan, Zheng 1605cb517eabSMiao Xie ret = btrfs_insert_fs_root(fs_info, root); 16060f7d52f4SChris Mason if (ret) { 160700246528SJosef Bacik btrfs_put_root(root); 16084785e24fSJosef Bacik if (ret == -EEXIST) 16094df27c4dSYan, Zheng goto again; 16104df27c4dSYan, Zheng goto fail; 16114df27c4dSYan, Zheng } 1612edbd8d4eSChris Mason return root; 16134df27c4dSYan, Zheng fail: 16148c38938cSJosef Bacik btrfs_put_root(root); 16154df27c4dSYan, Zheng return ERR_PTR(ret); 1616edbd8d4eSChris Mason } 1617edbd8d4eSChris Mason 161804160088SChris Mason static int btrfs_congested_fn(void *congested_data, int bdi_bits) 161904160088SChris Mason { 162004160088SChris Mason struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; 162104160088SChris Mason int ret = 0; 162204160088SChris Mason struct btrfs_device *device; 162304160088SChris Mason struct backing_dev_info *bdi; 1624b7967db7SChris Mason 16251f78160cSXiao Guangrong rcu_read_lock(); 16261f78160cSXiao Guangrong list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) { 1627dfe25020SChris Mason if (!device->bdev) 1628dfe25020SChris Mason continue; 1629efa7c9f9SJan Kara bdi = device->bdev->bd_bdi; 1630ff9ea323STejun Heo if (bdi_congested(bdi, bdi_bits)) { 163104160088SChris Mason ret = 1; 163204160088SChris Mason break; 163304160088SChris Mason } 163404160088SChris Mason } 16351f78160cSXiao Guangrong rcu_read_unlock(); 163604160088SChris Mason return ret; 163704160088SChris Mason } 163804160088SChris Mason 16398b712842SChris Mason /* 16408b712842SChris Mason * called by the kthread helper functions to finally call the bio end_io 16418b712842SChris Mason * functions. This is where read checksum verification actually happens 16428b712842SChris Mason */ 16438b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work) 1644ce9adaa5SChris Mason { 1645ce9adaa5SChris Mason struct bio *bio; 164697eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 1647ce9adaa5SChris Mason 164897eb6b69SDavid Sterba end_io_wq = container_of(work, struct btrfs_end_io_wq, work); 1649ce9adaa5SChris Mason bio = end_io_wq->bio; 16508b712842SChris Mason 16514e4cbee9SChristoph Hellwig bio->bi_status = end_io_wq->status; 1652ce9adaa5SChris Mason bio->bi_private = end_io_wq->private; 1653ce9adaa5SChris Mason bio->bi_end_io = end_io_wq->end_io; 16544246a0b6SChristoph Hellwig bio_endio(bio); 16559be490f1SOmar Sandoval kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq); 1656ce9adaa5SChris Mason } 165744b8bd7eSChris Mason 1658a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1659a74a4b97SChris Mason { 1660a74a4b97SChris Mason struct btrfs_root *root = arg; 16610b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1662d0278245SMiao Xie int again; 1663a74a4b97SChris Mason 1664d6fd0ae2SOmar Sandoval while (1) { 1665d0278245SMiao Xie again = 0; 16669d1a2a3aSDavid Sterba 1667fd340d0fSJosef Bacik set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1668fd340d0fSJosef Bacik 1669d0278245SMiao Xie /* Make the cleaner go to sleep early. */ 16702ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) 1671d0278245SMiao Xie goto sleep; 1672d0278245SMiao Xie 167390c711abSZygo Blaxell /* 167490c711abSZygo Blaxell * Do not do anything if we might cause open_ctree() to block 167590c711abSZygo Blaxell * before we have finished mounting the filesystem. 167690c711abSZygo Blaxell */ 16770b246afaSJeff Mahoney if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 167890c711abSZygo Blaxell goto sleep; 167990c711abSZygo Blaxell 16800b246afaSJeff Mahoney if (!mutex_trylock(&fs_info->cleaner_mutex)) 1681d0278245SMiao Xie goto sleep; 1682d0278245SMiao Xie 1683dc7f370cSMiao Xie /* 1684dc7f370cSMiao Xie * Avoid the problem that we change the status of the fs 1685dc7f370cSMiao Xie * during the above check and trylock. 1686dc7f370cSMiao Xie */ 16872ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) { 16880b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1689dc7f370cSMiao Xie goto sleep; 1690dc7f370cSMiao Xie } 1691dc7f370cSMiao Xie 16922ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 1693c2d6cb16SFilipe Manana 16949d1a2a3aSDavid Sterba again = btrfs_clean_one_deleted_snapshot(root); 16950b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1696a74a4b97SChris Mason 1697d0278245SMiao Xie /* 169805323cd1SMiao Xie * The defragger has dealt with the R/O remount and umount, 169905323cd1SMiao Xie * needn't do anything special here. 1700d0278245SMiao Xie */ 17010b246afaSJeff Mahoney btrfs_run_defrag_inodes(fs_info); 170267c5e7d4SFilipe Manana 170367c5e7d4SFilipe Manana /* 170467c5e7d4SFilipe Manana * Acquires fs_info->delete_unused_bgs_mutex to avoid racing 170567c5e7d4SFilipe Manana * with relocation (btrfs_relocate_chunk) and relocation 170667c5e7d4SFilipe Manana * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 170767c5e7d4SFilipe Manana * after acquiring fs_info->delete_unused_bgs_mutex. So we 170867c5e7d4SFilipe Manana * can't hold, nor need to, fs_info->cleaner_mutex when deleting 170967c5e7d4SFilipe Manana * unused block groups. 171067c5e7d4SFilipe Manana */ 17110b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 1712d0278245SMiao Xie sleep: 1713fd340d0fSJosef Bacik clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1714d6fd0ae2SOmar Sandoval if (kthread_should_park()) 1715d6fd0ae2SOmar Sandoval kthread_parkme(); 1716d6fd0ae2SOmar Sandoval if (kthread_should_stop()) 1717d6fd0ae2SOmar Sandoval return 0; 1718838fe188SJiri Kosina if (!again) { 1719a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 1720a74a4b97SChris Mason schedule(); 1721a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1722a74a4b97SChris Mason } 1723da288d28SFilipe Manana } 1724a74a4b97SChris Mason } 1725a74a4b97SChris Mason 1726a74a4b97SChris Mason static int transaction_kthread(void *arg) 1727a74a4b97SChris Mason { 1728a74a4b97SChris Mason struct btrfs_root *root = arg; 17290b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1730a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1731a74a4b97SChris Mason struct btrfs_transaction *cur; 17328929ecfaSYan, Zheng u64 transid; 1733a944442cSAllen Pais time64_t now; 1734a74a4b97SChris Mason unsigned long delay; 1735914b2007SJan Kara bool cannot_commit; 1736a74a4b97SChris Mason 1737a74a4b97SChris Mason do { 1738914b2007SJan Kara cannot_commit = false; 17390b246afaSJeff Mahoney delay = HZ * fs_info->commit_interval; 17400b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 1741a74a4b97SChris Mason 17420b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 17430b246afaSJeff Mahoney cur = fs_info->running_transaction; 1744a74a4b97SChris Mason if (!cur) { 17450b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1746a74a4b97SChris Mason goto sleep; 1747a74a4b97SChris Mason } 174831153d81SYan Zheng 1749afd48513SArnd Bergmann now = ktime_get_seconds(); 17503296bf56SQu Wenruo if (cur->state < TRANS_STATE_COMMIT_START && 1751a514d638SQu Wenruo !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) && 17528b87dc17SDavid Sterba (now < cur->start_time || 17530b246afaSJeff Mahoney now - cur->start_time < fs_info->commit_interval)) { 17540b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1755a74a4b97SChris Mason delay = HZ * 5; 1756a74a4b97SChris Mason goto sleep; 1757a74a4b97SChris Mason } 17588929ecfaSYan, Zheng transid = cur->transid; 17590b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 176056bec294SChris Mason 176179787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 1762354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 1763914b2007SJan Kara if (IS_ERR(trans)) { 1764354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 1765914b2007SJan Kara cannot_commit = true; 176679787eaaSJeff Mahoney goto sleep; 1767914b2007SJan Kara } 17688929ecfaSYan, Zheng if (transid == trans->transid) { 17693a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 17708929ecfaSYan, Zheng } else { 17713a45bb20SJeff Mahoney btrfs_end_transaction(trans); 17728929ecfaSYan, Zheng } 1773a74a4b97SChris Mason sleep: 17740b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 17750b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 1776a74a4b97SChris Mason 17774e121c06SJosef Bacik if (unlikely(test_bit(BTRFS_FS_STATE_ERROR, 17780b246afaSJeff Mahoney &fs_info->fs_state))) 17792ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 17808929ecfaSYan, Zheng if (!kthread_should_stop() && 17810b246afaSJeff Mahoney (!btrfs_transaction_blocked(fs_info) || 1782914b2007SJan Kara cannot_commit)) 1783bc5511d0SNikolay Borisov schedule_timeout_interruptible(delay); 1784a74a4b97SChris Mason } while (!kthread_should_stop()); 1785a74a4b97SChris Mason return 0; 1786a74a4b97SChris Mason } 1787a74a4b97SChris Mason 1788af31f5e5SChris Mason /* 178901f0f9daSNikolay Borisov * This will find the highest generation in the array of root backups. The 179001f0f9daSNikolay Borisov * index of the highest array is returned, or -EINVAL if we can't find 179101f0f9daSNikolay Borisov * anything. 1792af31f5e5SChris Mason * 1793af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1794af31f5e5SChris Mason * generation of the latest root in the array with the generation 1795af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1796af31f5e5SChris Mason */ 179701f0f9daSNikolay Borisov static int find_newest_super_backup(struct btrfs_fs_info *info) 1798af31f5e5SChris Mason { 179901f0f9daSNikolay Borisov const u64 newest_gen = btrfs_super_generation(info->super_copy); 1800af31f5e5SChris Mason u64 cur; 1801af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1802af31f5e5SChris Mason int i; 1803af31f5e5SChris Mason 1804af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1805af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1806af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1807af31f5e5SChris Mason if (cur == newest_gen) 180801f0f9daSNikolay Borisov return i; 1809af31f5e5SChris Mason } 1810af31f5e5SChris Mason 181101f0f9daSNikolay Borisov return -EINVAL; 1812af31f5e5SChris Mason } 1813af31f5e5SChris Mason 1814af31f5e5SChris Mason /* 1815af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1816af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1817af31f5e5SChris Mason * done 1818af31f5e5SChris Mason */ 1819af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1820af31f5e5SChris Mason { 18216ef108ddSNikolay Borisov const int next_backup = info->backup_root_index; 1822af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1823af31f5e5SChris Mason 1824af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1825af31f5e5SChris Mason 1826af31f5e5SChris Mason /* 1827af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1828af31f5e5SChris Mason * regardless of which ones we use today 1829af31f5e5SChris Mason */ 1830af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1831af31f5e5SChris Mason 1832af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1833af31f5e5SChris Mason 1834af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1835af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1836af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1837af31f5e5SChris Mason 1838af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1839af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1840af31f5e5SChris Mason 1841af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1842af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1843af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 1844af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 1845af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 1846af31f5e5SChris Mason 1847af31f5e5SChris Mason btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); 1848af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 1849af31f5e5SChris Mason btrfs_header_generation(info->extent_root->node)); 1850af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 1851af31f5e5SChris Mason btrfs_header_level(info->extent_root->node)); 1852af31f5e5SChris Mason 18537c7e82a7SChris Mason /* 18547c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 18557c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 18567c7e82a7SChris Mason */ 18577c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 18587c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 18597c7e82a7SChris Mason info->fs_root->node->start); 1860af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 1861af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 1862af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 1863af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 18647c7e82a7SChris Mason } 1865af31f5e5SChris Mason 1866af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 1867af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 1868af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 1869af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 1870af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 1871af31f5e5SChris Mason 1872af31f5e5SChris Mason btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); 1873af31f5e5SChris Mason btrfs_set_backup_csum_root_gen(root_backup, 1874af31f5e5SChris Mason btrfs_header_generation(info->csum_root->node)); 1875af31f5e5SChris Mason btrfs_set_backup_csum_root_level(root_backup, 1876af31f5e5SChris Mason btrfs_header_level(info->csum_root->node)); 1877af31f5e5SChris Mason 1878af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 1879af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 1880af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 1881af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 1882af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 1883af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 1884af31f5e5SChris Mason 1885af31f5e5SChris Mason /* 1886af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 1887af31f5e5SChris Mason * for the next commit 1888af31f5e5SChris Mason */ 1889af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 1890af31f5e5SChris Mason &info->super_for_commit->super_roots, 1891af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 1892af31f5e5SChris Mason } 1893af31f5e5SChris Mason 1894af31f5e5SChris Mason /* 1895bd2336b2SNikolay Borisov * read_backup_root - Reads a backup root based on the passed priority. Prio 0 1896bd2336b2SNikolay Borisov * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots 1897bd2336b2SNikolay Borisov * 1898bd2336b2SNikolay Borisov * fs_info - filesystem whose backup roots need to be read 1899bd2336b2SNikolay Borisov * priority - priority of backup root required 1900bd2336b2SNikolay Borisov * 1901bd2336b2SNikolay Borisov * Returns backup root index on success and -EINVAL otherwise. 1902bd2336b2SNikolay Borisov */ 1903bd2336b2SNikolay Borisov static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority) 1904bd2336b2SNikolay Borisov { 1905bd2336b2SNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 1906bd2336b2SNikolay Borisov struct btrfs_super_block *super = fs_info->super_copy; 1907bd2336b2SNikolay Borisov struct btrfs_root_backup *root_backup; 1908bd2336b2SNikolay Borisov 1909bd2336b2SNikolay Borisov if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) { 1910bd2336b2SNikolay Borisov if (priority == 0) 1911bd2336b2SNikolay Borisov return backup_index; 1912bd2336b2SNikolay Borisov 1913bd2336b2SNikolay Borisov backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority; 1914bd2336b2SNikolay Borisov backup_index %= BTRFS_NUM_BACKUP_ROOTS; 1915bd2336b2SNikolay Borisov } else { 1916bd2336b2SNikolay Borisov return -EINVAL; 1917bd2336b2SNikolay Borisov } 1918bd2336b2SNikolay Borisov 1919bd2336b2SNikolay Borisov root_backup = super->super_roots + backup_index; 1920bd2336b2SNikolay Borisov 1921bd2336b2SNikolay Borisov btrfs_set_super_generation(super, 1922bd2336b2SNikolay Borisov btrfs_backup_tree_root_gen(root_backup)); 1923bd2336b2SNikolay Borisov btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 1924bd2336b2SNikolay Borisov btrfs_set_super_root_level(super, 1925bd2336b2SNikolay Borisov btrfs_backup_tree_root_level(root_backup)); 1926bd2336b2SNikolay Borisov btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 1927bd2336b2SNikolay Borisov 1928bd2336b2SNikolay Borisov /* 1929bd2336b2SNikolay Borisov * Fixme: the total bytes and num_devices need to match or we should 1930bd2336b2SNikolay Borisov * need a fsck 1931bd2336b2SNikolay Borisov */ 1932bd2336b2SNikolay Borisov btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 1933bd2336b2SNikolay Borisov btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 1934bd2336b2SNikolay Borisov 1935bd2336b2SNikolay Borisov return backup_index; 1936bd2336b2SNikolay Borisov } 1937bd2336b2SNikolay Borisov 19387abadb64SLiu Bo /* helper to cleanup workers */ 19397abadb64SLiu Bo static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 19407abadb64SLiu Bo { 1941dc6e3209SQu Wenruo btrfs_destroy_workqueue(fs_info->fixup_workers); 1942afe3d242SQu Wenruo btrfs_destroy_workqueue(fs_info->delalloc_workers); 19435cdc7ad3SQu Wenruo btrfs_destroy_workqueue(fs_info->workers); 1944fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_workers); 1945fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_raid56_workers); 1946d05a33acSQu Wenruo btrfs_destroy_workqueue(fs_info->rmw_workers); 1947fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_write_workers); 1948fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 19495b3bc44eSQu Wenruo btrfs_destroy_workqueue(fs_info->delayed_workers); 1950e66f0bb1SQu Wenruo btrfs_destroy_workqueue(fs_info->caching_workers); 1951736cfa15SQu Wenruo btrfs_destroy_workqueue(fs_info->readahead_workers); 1952a44903abSQu Wenruo btrfs_destroy_workqueue(fs_info->flush_workers); 1953fc97fab0SQu Wenruo btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 1954b0643e59SDennis Zhou if (fs_info->discard_ctl.discard_workers) 1955b0643e59SDennis Zhou destroy_workqueue(fs_info->discard_ctl.discard_workers); 1956a9b9477dSFilipe Manana /* 1957a9b9477dSFilipe Manana * Now that all other work queues are destroyed, we can safely destroy 1958a9b9477dSFilipe Manana * the queues used for metadata I/O, since tasks from those other work 1959a9b9477dSFilipe Manana * queues can do metadata I/O operations. 1960a9b9477dSFilipe Manana */ 1961a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_workers); 1962a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_write_workers); 19637abadb64SLiu Bo } 19647abadb64SLiu Bo 19652e9f5954SRashika static void free_root_extent_buffers(struct btrfs_root *root) 19662e9f5954SRashika { 19672e9f5954SRashika if (root) { 19682e9f5954SRashika free_extent_buffer(root->node); 19692e9f5954SRashika free_extent_buffer(root->commit_root); 19702e9f5954SRashika root->node = NULL; 19712e9f5954SRashika root->commit_root = NULL; 19722e9f5954SRashika } 19732e9f5954SRashika } 19742e9f5954SRashika 1975af31f5e5SChris Mason /* helper to cleanup tree roots */ 19764273eaffSAnand Jain static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root) 1977af31f5e5SChris Mason { 19782e9f5954SRashika free_root_extent_buffers(info->tree_root); 1979655b09feSJosef Bacik 19802e9f5954SRashika free_root_extent_buffers(info->dev_root); 19812e9f5954SRashika free_root_extent_buffers(info->extent_root); 19822e9f5954SRashika free_root_extent_buffers(info->csum_root); 19832e9f5954SRashika free_root_extent_buffers(info->quota_root); 19842e9f5954SRashika free_root_extent_buffers(info->uuid_root); 19858c38938cSJosef Bacik free_root_extent_buffers(info->fs_root); 1986aeb935a4SQu Wenruo free_root_extent_buffers(info->data_reloc_root); 19874273eaffSAnand Jain if (free_chunk_root) 19882e9f5954SRashika free_root_extent_buffers(info->chunk_root); 198970f6d82eSOmar Sandoval free_root_extent_buffers(info->free_space_root); 1990af31f5e5SChris Mason } 1991af31f5e5SChris Mason 19928c38938cSJosef Bacik void btrfs_put_root(struct btrfs_root *root) 19938c38938cSJosef Bacik { 19948c38938cSJosef Bacik if (!root) 19958c38938cSJosef Bacik return; 19968c38938cSJosef Bacik 19978c38938cSJosef Bacik if (refcount_dec_and_test(&root->refs)) { 19988c38938cSJosef Bacik WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 19991dae7e0eSQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)); 20008c38938cSJosef Bacik if (root->anon_dev) 20018c38938cSJosef Bacik free_anon_bdev(root->anon_dev); 20028c38938cSJosef Bacik btrfs_drew_lock_destroy(&root->snapshot_lock); 20038c38938cSJosef Bacik free_extent_buffer(root->node); 20048c38938cSJosef Bacik free_extent_buffer(root->commit_root); 20058c38938cSJosef Bacik kfree(root->free_ino_ctl); 20068c38938cSJosef Bacik kfree(root->free_ino_pinned); 20078c38938cSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 20088c38938cSJosef Bacik spin_lock(&root->fs_info->fs_roots_radix_lock); 20098c38938cSJosef Bacik list_del_init(&root->leak_list); 20108c38938cSJosef Bacik spin_unlock(&root->fs_info->fs_roots_radix_lock); 20118c38938cSJosef Bacik #endif 20128c38938cSJosef Bacik kfree(root); 20138c38938cSJosef Bacik } 20148c38938cSJosef Bacik } 20158c38938cSJosef Bacik 2016faa2dbf0SJosef Bacik void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2017171f6537SJosef Bacik { 2018171f6537SJosef Bacik int ret; 2019171f6537SJosef Bacik struct btrfs_root *gang[8]; 2020171f6537SJosef Bacik int i; 2021171f6537SJosef Bacik 2022171f6537SJosef Bacik while (!list_empty(&fs_info->dead_roots)) { 2023171f6537SJosef Bacik gang[0] = list_entry(fs_info->dead_roots.next, 2024171f6537SJosef Bacik struct btrfs_root, root_list); 2025171f6537SJosef Bacik list_del(&gang[0]->root_list); 2026171f6537SJosef Bacik 20278c38938cSJosef Bacik if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) 2028cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[0]); 202900246528SJosef Bacik btrfs_put_root(gang[0]); 2030171f6537SJosef Bacik } 2031171f6537SJosef Bacik 2032171f6537SJosef Bacik while (1) { 2033171f6537SJosef Bacik ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2034171f6537SJosef Bacik (void **)gang, 0, 2035171f6537SJosef Bacik ARRAY_SIZE(gang)); 2036171f6537SJosef Bacik if (!ret) 2037171f6537SJosef Bacik break; 2038171f6537SJosef Bacik for (i = 0; i < ret; i++) 2039cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2040171f6537SJosef Bacik } 2041171f6537SJosef Bacik } 2042af31f5e5SChris Mason 2043638aa7edSEric Sandeen static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2044638aa7edSEric Sandeen { 2045638aa7edSEric Sandeen mutex_init(&fs_info->scrub_lock); 2046638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_running, 0); 2047638aa7edSEric Sandeen atomic_set(&fs_info->scrub_pause_req, 0); 2048638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_paused, 0); 2049638aa7edSEric Sandeen atomic_set(&fs_info->scrub_cancel_req, 0); 2050638aa7edSEric Sandeen init_waitqueue_head(&fs_info->scrub_pause_wait); 2051ff09c4caSAnand Jain refcount_set(&fs_info->scrub_workers_refcnt, 0); 2052638aa7edSEric Sandeen } 2053638aa7edSEric Sandeen 2054779a65a4SEric Sandeen static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2055779a65a4SEric Sandeen { 2056779a65a4SEric Sandeen spin_lock_init(&fs_info->balance_lock); 2057779a65a4SEric Sandeen mutex_init(&fs_info->balance_mutex); 2058779a65a4SEric Sandeen atomic_set(&fs_info->balance_pause_req, 0); 2059779a65a4SEric Sandeen atomic_set(&fs_info->balance_cancel_req, 0); 2060779a65a4SEric Sandeen fs_info->balance_ctl = NULL; 2061779a65a4SEric Sandeen init_waitqueue_head(&fs_info->balance_wait_q); 2062779a65a4SEric Sandeen } 2063779a65a4SEric Sandeen 20646bccf3abSJeff Mahoney static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info) 2065f37938e0SEric Sandeen { 20662ff7e61eSJeff Mahoney struct inode *inode = fs_info->btree_inode; 20672ff7e61eSJeff Mahoney 20682ff7e61eSJeff Mahoney inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 20692ff7e61eSJeff Mahoney set_nlink(inode, 1); 2070f37938e0SEric Sandeen /* 2071f37938e0SEric Sandeen * we set the i_size on the btree inode to the max possible int. 2072f37938e0SEric Sandeen * the real end of the address space is determined by all of 2073f37938e0SEric Sandeen * the devices in the system 2074f37938e0SEric Sandeen */ 20752ff7e61eSJeff Mahoney inode->i_size = OFFSET_MAX; 20762ff7e61eSJeff Mahoney inode->i_mapping->a_ops = &btree_aops; 2077f37938e0SEric Sandeen 20782ff7e61eSJeff Mahoney RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 207943eb5f29SQu Wenruo extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree, 208043eb5f29SQu Wenruo IO_TREE_INODE_IO, inode); 20817b439738SDavid Sterba BTRFS_I(inode)->io_tree.track_uptodate = false; 20822ff7e61eSJeff Mahoney extent_map_tree_init(&BTRFS_I(inode)->extent_tree); 2083f37938e0SEric Sandeen 20842ff7e61eSJeff Mahoney BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops; 2085f37938e0SEric Sandeen 20865c8fd99fSJosef Bacik BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root); 20872ff7e61eSJeff Mahoney memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key)); 20882ff7e61eSJeff Mahoney set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); 20892ff7e61eSJeff Mahoney btrfs_insert_inode_hash(inode); 2090f37938e0SEric Sandeen } 2091f37938e0SEric Sandeen 2092ad618368SEric Sandeen static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2093ad618368SEric Sandeen { 2094ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2095129827e3SDavid Sterba init_rwsem(&fs_info->dev_replace.rwsem); 20967f8d236aSDavid Sterba init_waitqueue_head(&fs_info->dev_replace.replace_wait); 2097ad618368SEric Sandeen } 2098ad618368SEric Sandeen 2099f9e92e40SEric Sandeen static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2100f9e92e40SEric Sandeen { 2101f9e92e40SEric Sandeen spin_lock_init(&fs_info->qgroup_lock); 2102f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_ioctl_lock); 2103f9e92e40SEric Sandeen fs_info->qgroup_tree = RB_ROOT; 2104f9e92e40SEric Sandeen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2105f9e92e40SEric Sandeen fs_info->qgroup_seq = 1; 2106f9e92e40SEric Sandeen fs_info->qgroup_ulist = NULL; 2107d2c609b8SJeff Mahoney fs_info->qgroup_rescan_running = false; 2108f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_rescan_lock); 2109f9e92e40SEric Sandeen } 2110f9e92e40SEric Sandeen 21112a458198SEric Sandeen static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info, 21122a458198SEric Sandeen struct btrfs_fs_devices *fs_devices) 21132a458198SEric Sandeen { 2114f7b885beSAnand Jain u32 max_active = fs_info->thread_pool_size; 21156f011058SDavid Sterba unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 21162a458198SEric Sandeen 21172a458198SEric Sandeen fs_info->workers = 2118cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "worker", 2119cb001095SJeff Mahoney flags | WQ_HIGHPRI, max_active, 16); 21202a458198SEric Sandeen 21212a458198SEric Sandeen fs_info->delalloc_workers = 2122cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delalloc", 2123cb001095SJeff Mahoney flags, max_active, 2); 21242a458198SEric Sandeen 21252a458198SEric Sandeen fs_info->flush_workers = 2126cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "flush_delalloc", 2127cb001095SJeff Mahoney flags, max_active, 0); 21282a458198SEric Sandeen 21292a458198SEric Sandeen fs_info->caching_workers = 2130cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); 21312a458198SEric Sandeen 21322a458198SEric Sandeen fs_info->fixup_workers = 2133cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); 21342a458198SEric Sandeen 21352a458198SEric Sandeen /* 21362a458198SEric Sandeen * endios are largely parallel and should have a very 21372a458198SEric Sandeen * low idle thresh 21382a458198SEric Sandeen */ 21392a458198SEric Sandeen fs_info->endio_workers = 2140cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4); 21412a458198SEric Sandeen fs_info->endio_meta_workers = 2142cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta", flags, 2143cb001095SJeff Mahoney max_active, 4); 21442a458198SEric Sandeen fs_info->endio_meta_write_workers = 2145cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags, 2146cb001095SJeff Mahoney max_active, 2); 21472a458198SEric Sandeen fs_info->endio_raid56_workers = 2148cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-raid56", flags, 2149cb001095SJeff Mahoney max_active, 4); 21502a458198SEric Sandeen fs_info->rmw_workers = 2151cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2); 21522a458198SEric Sandeen fs_info->endio_write_workers = 2153cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-write", flags, 2154cb001095SJeff Mahoney max_active, 2); 21552a458198SEric Sandeen fs_info->endio_freespace_worker = 2156cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "freespace-write", flags, 2157cb001095SJeff Mahoney max_active, 0); 21582a458198SEric Sandeen fs_info->delayed_workers = 2159cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, 2160cb001095SJeff Mahoney max_active, 0); 21612a458198SEric Sandeen fs_info->readahead_workers = 2162cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "readahead", flags, 2163cb001095SJeff Mahoney max_active, 2); 21642a458198SEric Sandeen fs_info->qgroup_rescan_workers = 2165cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); 2166b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers = 2167b0643e59SDennis Zhou alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1); 21682a458198SEric Sandeen 21692a458198SEric Sandeen if (!(fs_info->workers && fs_info->delalloc_workers && 2170ba8a9d07SChris Mason fs_info->flush_workers && 21712a458198SEric Sandeen fs_info->endio_workers && fs_info->endio_meta_workers && 21722a458198SEric Sandeen fs_info->endio_meta_write_workers && 21732a458198SEric Sandeen fs_info->endio_write_workers && fs_info->endio_raid56_workers && 21742a458198SEric Sandeen fs_info->endio_freespace_worker && fs_info->rmw_workers && 21752a458198SEric Sandeen fs_info->caching_workers && fs_info->readahead_workers && 21762a458198SEric Sandeen fs_info->fixup_workers && fs_info->delayed_workers && 2177b0643e59SDennis Zhou fs_info->qgroup_rescan_workers && 2178b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers)) { 21792a458198SEric Sandeen return -ENOMEM; 21802a458198SEric Sandeen } 21812a458198SEric Sandeen 21822a458198SEric Sandeen return 0; 21832a458198SEric Sandeen } 21842a458198SEric Sandeen 21856d97c6e3SJohannes Thumshirn static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type) 21866d97c6e3SJohannes Thumshirn { 21876d97c6e3SJohannes Thumshirn struct crypto_shash *csum_shash; 2188b4e967beSDavid Sterba const char *csum_driver = btrfs_super_csum_driver(csum_type); 21896d97c6e3SJohannes Thumshirn 2190b4e967beSDavid Sterba csum_shash = crypto_alloc_shash(csum_driver, 0, 0); 21916d97c6e3SJohannes Thumshirn 21926d97c6e3SJohannes Thumshirn if (IS_ERR(csum_shash)) { 21936d97c6e3SJohannes Thumshirn btrfs_err(fs_info, "error allocating %s hash for checksum", 2194b4e967beSDavid Sterba csum_driver); 21956d97c6e3SJohannes Thumshirn return PTR_ERR(csum_shash); 21966d97c6e3SJohannes Thumshirn } 21976d97c6e3SJohannes Thumshirn 21986d97c6e3SJohannes Thumshirn fs_info->csum_shash = csum_shash; 21996d97c6e3SJohannes Thumshirn 22006d97c6e3SJohannes Thumshirn return 0; 22016d97c6e3SJohannes Thumshirn } 22026d97c6e3SJohannes Thumshirn 220363443bf5SEric Sandeen static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 220463443bf5SEric Sandeen struct btrfs_fs_devices *fs_devices) 220563443bf5SEric Sandeen { 220663443bf5SEric Sandeen int ret; 220763443bf5SEric Sandeen struct btrfs_root *log_tree_root; 220863443bf5SEric Sandeen struct btrfs_super_block *disk_super = fs_info->super_copy; 220963443bf5SEric Sandeen u64 bytenr = btrfs_super_log_root(disk_super); 2210581c1760SQu Wenruo int level = btrfs_super_log_root_level(disk_super); 221163443bf5SEric Sandeen 221263443bf5SEric Sandeen if (fs_devices->rw_devices == 0) { 2213f14d104dSDavid Sterba btrfs_warn(fs_info, "log replay required on RO media"); 221463443bf5SEric Sandeen return -EIO; 221563443bf5SEric Sandeen } 221663443bf5SEric Sandeen 221796dfcb46SJosef Bacik log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, 221896dfcb46SJosef Bacik GFP_KERNEL); 221963443bf5SEric Sandeen if (!log_tree_root) 222063443bf5SEric Sandeen return -ENOMEM; 222163443bf5SEric Sandeen 22222ff7e61eSJeff Mahoney log_tree_root->node = read_tree_block(fs_info, bytenr, 2223581c1760SQu Wenruo fs_info->generation + 1, 2224581c1760SQu Wenruo level, NULL); 222564c043deSLiu Bo if (IS_ERR(log_tree_root->node)) { 2226f14d104dSDavid Sterba btrfs_warn(fs_info, "failed to read log tree"); 22270eeff236SLiu Bo ret = PTR_ERR(log_tree_root->node); 22288c38938cSJosef Bacik log_tree_root->node = NULL; 222900246528SJosef Bacik btrfs_put_root(log_tree_root); 22300eeff236SLiu Bo return ret; 223164c043deSLiu Bo } else if (!extent_buffer_uptodate(log_tree_root->node)) { 2232f14d104dSDavid Sterba btrfs_err(fs_info, "failed to read log tree"); 223300246528SJosef Bacik btrfs_put_root(log_tree_root); 223463443bf5SEric Sandeen return -EIO; 223563443bf5SEric Sandeen } 223663443bf5SEric Sandeen /* returns with log_tree_root freed on success */ 223763443bf5SEric Sandeen ret = btrfs_recover_log_trees(log_tree_root); 223863443bf5SEric Sandeen if (ret) { 22390b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 224063443bf5SEric Sandeen "Failed to recover log tree"); 224100246528SJosef Bacik btrfs_put_root(log_tree_root); 224263443bf5SEric Sandeen return ret; 224363443bf5SEric Sandeen } 224463443bf5SEric Sandeen 2245bc98a42cSDavid Howells if (sb_rdonly(fs_info->sb)) { 22466bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 224763443bf5SEric Sandeen if (ret) 224863443bf5SEric Sandeen return ret; 224963443bf5SEric Sandeen } 225063443bf5SEric Sandeen 225163443bf5SEric Sandeen return 0; 225263443bf5SEric Sandeen } 225363443bf5SEric Sandeen 22546bccf3abSJeff Mahoney static int btrfs_read_roots(struct btrfs_fs_info *fs_info) 22554bbcaa64SEric Sandeen { 22566bccf3abSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 2257a4f3d2c4SDavid Sterba struct btrfs_root *root; 22584bbcaa64SEric Sandeen struct btrfs_key location; 22594bbcaa64SEric Sandeen int ret; 22604bbcaa64SEric Sandeen 22616bccf3abSJeff Mahoney BUG_ON(!fs_info->tree_root); 22626bccf3abSJeff Mahoney 22634bbcaa64SEric Sandeen location.objectid = BTRFS_EXTENT_TREE_OBJECTID; 22644bbcaa64SEric Sandeen location.type = BTRFS_ROOT_ITEM_KEY; 22654bbcaa64SEric Sandeen location.offset = 0; 22664bbcaa64SEric Sandeen 2267a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2268f50f4353SLiu Bo if (IS_ERR(root)) { 2269f50f4353SLiu Bo ret = PTR_ERR(root); 2270f50f4353SLiu Bo goto out; 2271f50f4353SLiu Bo } 2272a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2273a4f3d2c4SDavid Sterba fs_info->extent_root = root; 22744bbcaa64SEric Sandeen 22754bbcaa64SEric Sandeen location.objectid = BTRFS_DEV_TREE_OBJECTID; 2276a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2277f50f4353SLiu Bo if (IS_ERR(root)) { 2278f50f4353SLiu Bo ret = PTR_ERR(root); 2279f50f4353SLiu Bo goto out; 2280f50f4353SLiu Bo } 2281a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2282a4f3d2c4SDavid Sterba fs_info->dev_root = root; 22834bbcaa64SEric Sandeen btrfs_init_devices_late(fs_info); 22844bbcaa64SEric Sandeen 22854bbcaa64SEric Sandeen location.objectid = BTRFS_CSUM_TREE_OBJECTID; 2286a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2287f50f4353SLiu Bo if (IS_ERR(root)) { 2288f50f4353SLiu Bo ret = PTR_ERR(root); 2289f50f4353SLiu Bo goto out; 2290f50f4353SLiu Bo } 2291a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2292a4f3d2c4SDavid Sterba fs_info->csum_root = root; 22934bbcaa64SEric Sandeen 2294aeb935a4SQu Wenruo /* 2295aeb935a4SQu Wenruo * This tree can share blocks with some other fs tree during relocation 2296aeb935a4SQu Wenruo * and we need a proper setup by btrfs_get_fs_root 2297aeb935a4SQu Wenruo */ 229856e9357aSDavid Sterba root = btrfs_get_fs_root(tree_root->fs_info, 229956e9357aSDavid Sterba BTRFS_DATA_RELOC_TREE_OBJECTID, true); 2300aeb935a4SQu Wenruo if (IS_ERR(root)) { 2301aeb935a4SQu Wenruo ret = PTR_ERR(root); 2302aeb935a4SQu Wenruo goto out; 2303aeb935a4SQu Wenruo } 2304aeb935a4SQu Wenruo set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2305aeb935a4SQu Wenruo fs_info->data_reloc_root = root; 2306aeb935a4SQu Wenruo 23074bbcaa64SEric Sandeen location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2308a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2309a4f3d2c4SDavid Sterba if (!IS_ERR(root)) { 2310a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2311afcdd129SJosef Bacik set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); 2312a4f3d2c4SDavid Sterba fs_info->quota_root = root; 23134bbcaa64SEric Sandeen } 23144bbcaa64SEric Sandeen 23154bbcaa64SEric Sandeen location.objectid = BTRFS_UUID_TREE_OBJECTID; 2316a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2317a4f3d2c4SDavid Sterba if (IS_ERR(root)) { 2318a4f3d2c4SDavid Sterba ret = PTR_ERR(root); 23194bbcaa64SEric Sandeen if (ret != -ENOENT) 2320f50f4353SLiu Bo goto out; 23214bbcaa64SEric Sandeen } else { 2322a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2323a4f3d2c4SDavid Sterba fs_info->uuid_root = root; 23244bbcaa64SEric Sandeen } 23254bbcaa64SEric Sandeen 232670f6d82eSOmar Sandoval if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 232770f6d82eSOmar Sandoval location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID; 232870f6d82eSOmar Sandoval root = btrfs_read_tree_root(tree_root, &location); 2329f50f4353SLiu Bo if (IS_ERR(root)) { 2330f50f4353SLiu Bo ret = PTR_ERR(root); 2331f50f4353SLiu Bo goto out; 2332f50f4353SLiu Bo } 233370f6d82eSOmar Sandoval set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 233470f6d82eSOmar Sandoval fs_info->free_space_root = root; 233570f6d82eSOmar Sandoval } 233670f6d82eSOmar Sandoval 23374bbcaa64SEric Sandeen return 0; 2338f50f4353SLiu Bo out: 2339f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", 2340f50f4353SLiu Bo location.objectid, ret); 2341f50f4353SLiu Bo return ret; 23424bbcaa64SEric Sandeen } 23434bbcaa64SEric Sandeen 2344069ec957SQu Wenruo /* 2345069ec957SQu Wenruo * Real super block validation 2346069ec957SQu Wenruo * NOTE: super csum type and incompat features will not be checked here. 2347069ec957SQu Wenruo * 2348069ec957SQu Wenruo * @sb: super block to check 2349069ec957SQu Wenruo * @mirror_num: the super block number to check its bytenr: 2350069ec957SQu Wenruo * 0 the primary (1st) sb 2351069ec957SQu Wenruo * 1, 2 2nd and 3rd backup copy 2352069ec957SQu Wenruo * -1 skip bytenr check 2353069ec957SQu Wenruo */ 2354069ec957SQu Wenruo static int validate_super(struct btrfs_fs_info *fs_info, 2355069ec957SQu Wenruo struct btrfs_super_block *sb, int mirror_num) 235621a852b0SQu Wenruo { 235721a852b0SQu Wenruo u64 nodesize = btrfs_super_nodesize(sb); 235821a852b0SQu Wenruo u64 sectorsize = btrfs_super_sectorsize(sb); 235921a852b0SQu Wenruo int ret = 0; 236021a852b0SQu Wenruo 236121a852b0SQu Wenruo if (btrfs_super_magic(sb) != BTRFS_MAGIC) { 236221a852b0SQu Wenruo btrfs_err(fs_info, "no valid FS found"); 236321a852b0SQu Wenruo ret = -EINVAL; 236421a852b0SQu Wenruo } 236521a852b0SQu Wenruo if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { 236621a852b0SQu Wenruo btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", 236721a852b0SQu Wenruo btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); 236821a852b0SQu Wenruo ret = -EINVAL; 236921a852b0SQu Wenruo } 237021a852b0SQu Wenruo if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 237121a852b0SQu Wenruo btrfs_err(fs_info, "tree_root level too big: %d >= %d", 237221a852b0SQu Wenruo btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 237321a852b0SQu Wenruo ret = -EINVAL; 237421a852b0SQu Wenruo } 237521a852b0SQu Wenruo if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 237621a852b0SQu Wenruo btrfs_err(fs_info, "chunk_root level too big: %d >= %d", 237721a852b0SQu Wenruo btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 237821a852b0SQu Wenruo ret = -EINVAL; 237921a852b0SQu Wenruo } 238021a852b0SQu Wenruo if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 238121a852b0SQu Wenruo btrfs_err(fs_info, "log_root level too big: %d >= %d", 238221a852b0SQu Wenruo btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 238321a852b0SQu Wenruo ret = -EINVAL; 238421a852b0SQu Wenruo } 238521a852b0SQu Wenruo 238621a852b0SQu Wenruo /* 238721a852b0SQu Wenruo * Check sectorsize and nodesize first, other check will need it. 238821a852b0SQu Wenruo * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. 238921a852b0SQu Wenruo */ 239021a852b0SQu Wenruo if (!is_power_of_2(sectorsize) || sectorsize < 4096 || 239121a852b0SQu Wenruo sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { 239221a852b0SQu Wenruo btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); 239321a852b0SQu Wenruo ret = -EINVAL; 239421a852b0SQu Wenruo } 239521a852b0SQu Wenruo /* Only PAGE SIZE is supported yet */ 239621a852b0SQu Wenruo if (sectorsize != PAGE_SIZE) { 239721a852b0SQu Wenruo btrfs_err(fs_info, 239821a852b0SQu Wenruo "sectorsize %llu not supported yet, only support %lu", 239921a852b0SQu Wenruo sectorsize, PAGE_SIZE); 240021a852b0SQu Wenruo ret = -EINVAL; 240121a852b0SQu Wenruo } 240221a852b0SQu Wenruo if (!is_power_of_2(nodesize) || nodesize < sectorsize || 240321a852b0SQu Wenruo nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { 240421a852b0SQu Wenruo btrfs_err(fs_info, "invalid nodesize %llu", nodesize); 240521a852b0SQu Wenruo ret = -EINVAL; 240621a852b0SQu Wenruo } 240721a852b0SQu Wenruo if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { 240821a852b0SQu Wenruo btrfs_err(fs_info, "invalid leafsize %u, should be %llu", 240921a852b0SQu Wenruo le32_to_cpu(sb->__unused_leafsize), nodesize); 241021a852b0SQu Wenruo ret = -EINVAL; 241121a852b0SQu Wenruo } 241221a852b0SQu Wenruo 241321a852b0SQu Wenruo /* Root alignment check */ 241421a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { 241521a852b0SQu Wenruo btrfs_warn(fs_info, "tree_root block unaligned: %llu", 241621a852b0SQu Wenruo btrfs_super_root(sb)); 241721a852b0SQu Wenruo ret = -EINVAL; 241821a852b0SQu Wenruo } 241921a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { 242021a852b0SQu Wenruo btrfs_warn(fs_info, "chunk_root block unaligned: %llu", 242121a852b0SQu Wenruo btrfs_super_chunk_root(sb)); 242221a852b0SQu Wenruo ret = -EINVAL; 242321a852b0SQu Wenruo } 242421a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { 242521a852b0SQu Wenruo btrfs_warn(fs_info, "log_root block unaligned: %llu", 242621a852b0SQu Wenruo btrfs_super_log_root(sb)); 242721a852b0SQu Wenruo ret = -EINVAL; 242821a852b0SQu Wenruo } 242921a852b0SQu Wenruo 2430de37aa51SNikolay Borisov if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, 24317239ff4bSNikolay Borisov BTRFS_FSID_SIZE) != 0) { 243221a852b0SQu Wenruo btrfs_err(fs_info, 24337239ff4bSNikolay Borisov "dev_item UUID does not match metadata fsid: %pU != %pU", 2434de37aa51SNikolay Borisov fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); 243521a852b0SQu Wenruo ret = -EINVAL; 243621a852b0SQu Wenruo } 243721a852b0SQu Wenruo 243821a852b0SQu Wenruo /* 243921a852b0SQu Wenruo * Hint to catch really bogus numbers, bitflips or so, more exact checks are 244021a852b0SQu Wenruo * done later 244121a852b0SQu Wenruo */ 244221a852b0SQu Wenruo if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { 244321a852b0SQu Wenruo btrfs_err(fs_info, "bytes_used is too small %llu", 244421a852b0SQu Wenruo btrfs_super_bytes_used(sb)); 244521a852b0SQu Wenruo ret = -EINVAL; 244621a852b0SQu Wenruo } 244721a852b0SQu Wenruo if (!is_power_of_2(btrfs_super_stripesize(sb))) { 244821a852b0SQu Wenruo btrfs_err(fs_info, "invalid stripesize %u", 244921a852b0SQu Wenruo btrfs_super_stripesize(sb)); 245021a852b0SQu Wenruo ret = -EINVAL; 245121a852b0SQu Wenruo } 245221a852b0SQu Wenruo if (btrfs_super_num_devices(sb) > (1UL << 31)) 245321a852b0SQu Wenruo btrfs_warn(fs_info, "suspicious number of devices: %llu", 245421a852b0SQu Wenruo btrfs_super_num_devices(sb)); 245521a852b0SQu Wenruo if (btrfs_super_num_devices(sb) == 0) { 245621a852b0SQu Wenruo btrfs_err(fs_info, "number of devices is 0"); 245721a852b0SQu Wenruo ret = -EINVAL; 245821a852b0SQu Wenruo } 245921a852b0SQu Wenruo 2460069ec957SQu Wenruo if (mirror_num >= 0 && 2461069ec957SQu Wenruo btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { 246221a852b0SQu Wenruo btrfs_err(fs_info, "super offset mismatch %llu != %u", 246321a852b0SQu Wenruo btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 246421a852b0SQu Wenruo ret = -EINVAL; 246521a852b0SQu Wenruo } 246621a852b0SQu Wenruo 246721a852b0SQu Wenruo /* 246821a852b0SQu Wenruo * Obvious sys_chunk_array corruptions, it must hold at least one key 246921a852b0SQu Wenruo * and one chunk 247021a852b0SQu Wenruo */ 247121a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 247221a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too big %u > %u", 247321a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 247421a852b0SQu Wenruo BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 247521a852b0SQu Wenruo ret = -EINVAL; 247621a852b0SQu Wenruo } 247721a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 247821a852b0SQu Wenruo + sizeof(struct btrfs_chunk)) { 247921a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too small %u < %zu", 248021a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 248121a852b0SQu Wenruo sizeof(struct btrfs_disk_key) 248221a852b0SQu Wenruo + sizeof(struct btrfs_chunk)); 248321a852b0SQu Wenruo ret = -EINVAL; 248421a852b0SQu Wenruo } 248521a852b0SQu Wenruo 248621a852b0SQu Wenruo /* 248721a852b0SQu Wenruo * The generation is a global counter, we'll trust it more than the others 248821a852b0SQu Wenruo * but it's still possible that it's the one that's wrong. 248921a852b0SQu Wenruo */ 249021a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 249121a852b0SQu Wenruo btrfs_warn(fs_info, 249221a852b0SQu Wenruo "suspicious: generation < chunk_root_generation: %llu < %llu", 249321a852b0SQu Wenruo btrfs_super_generation(sb), 249421a852b0SQu Wenruo btrfs_super_chunk_root_generation(sb)); 249521a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 249621a852b0SQu Wenruo && btrfs_super_cache_generation(sb) != (u64)-1) 249721a852b0SQu Wenruo btrfs_warn(fs_info, 249821a852b0SQu Wenruo "suspicious: generation < cache_generation: %llu < %llu", 249921a852b0SQu Wenruo btrfs_super_generation(sb), 250021a852b0SQu Wenruo btrfs_super_cache_generation(sb)); 250121a852b0SQu Wenruo 250221a852b0SQu Wenruo return ret; 250321a852b0SQu Wenruo } 250421a852b0SQu Wenruo 2505069ec957SQu Wenruo /* 2506069ec957SQu Wenruo * Validation of super block at mount time. 2507069ec957SQu Wenruo * Some checks already done early at mount time, like csum type and incompat 2508069ec957SQu Wenruo * flags will be skipped. 2509069ec957SQu Wenruo */ 2510069ec957SQu Wenruo static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) 2511069ec957SQu Wenruo { 2512069ec957SQu Wenruo return validate_super(fs_info, fs_info->super_copy, 0); 2513069ec957SQu Wenruo } 2514069ec957SQu Wenruo 251575cb857dSQu Wenruo /* 251675cb857dSQu Wenruo * Validation of super block at write time. 251775cb857dSQu Wenruo * Some checks like bytenr check will be skipped as their values will be 251875cb857dSQu Wenruo * overwritten soon. 251975cb857dSQu Wenruo * Extra checks like csum type and incompat flags will be done here. 252075cb857dSQu Wenruo */ 252175cb857dSQu Wenruo static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, 252275cb857dSQu Wenruo struct btrfs_super_block *sb) 252375cb857dSQu Wenruo { 252475cb857dSQu Wenruo int ret; 252575cb857dSQu Wenruo 252675cb857dSQu Wenruo ret = validate_super(fs_info, sb, -1); 252775cb857dSQu Wenruo if (ret < 0) 252875cb857dSQu Wenruo goto out; 2529e7e16f48SJohannes Thumshirn if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) { 253075cb857dSQu Wenruo ret = -EUCLEAN; 253175cb857dSQu Wenruo btrfs_err(fs_info, "invalid csum type, has %u want %u", 253275cb857dSQu Wenruo btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); 253375cb857dSQu Wenruo goto out; 253475cb857dSQu Wenruo } 253575cb857dSQu Wenruo if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 253675cb857dSQu Wenruo ret = -EUCLEAN; 253775cb857dSQu Wenruo btrfs_err(fs_info, 253875cb857dSQu Wenruo "invalid incompat flags, has 0x%llx valid mask 0x%llx", 253975cb857dSQu Wenruo btrfs_super_incompat_flags(sb), 254075cb857dSQu Wenruo (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); 254175cb857dSQu Wenruo goto out; 254275cb857dSQu Wenruo } 254375cb857dSQu Wenruo out: 254475cb857dSQu Wenruo if (ret < 0) 254575cb857dSQu Wenruo btrfs_err(fs_info, 254675cb857dSQu Wenruo "super block corruption detected before writing it to disk"); 254775cb857dSQu Wenruo return ret; 254875cb857dSQu Wenruo } 254975cb857dSQu Wenruo 25506ef108ddSNikolay Borisov static int __cold init_tree_roots(struct btrfs_fs_info *fs_info) 2551b8522a1eSNikolay Borisov { 25526ef108ddSNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 2553b8522a1eSNikolay Borisov struct btrfs_super_block *sb = fs_info->super_copy; 2554b8522a1eSNikolay Borisov struct btrfs_root *tree_root = fs_info->tree_root; 2555b8522a1eSNikolay Borisov bool handle_error = false; 2556b8522a1eSNikolay Borisov int ret = 0; 2557b8522a1eSNikolay Borisov int i; 2558b8522a1eSNikolay Borisov 2559b8522a1eSNikolay Borisov for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 2560b8522a1eSNikolay Borisov u64 generation; 2561b8522a1eSNikolay Borisov int level; 2562b8522a1eSNikolay Borisov 2563b8522a1eSNikolay Borisov if (handle_error) { 2564b8522a1eSNikolay Borisov if (!IS_ERR(tree_root->node)) 2565b8522a1eSNikolay Borisov free_extent_buffer(tree_root->node); 2566b8522a1eSNikolay Borisov tree_root->node = NULL; 2567b8522a1eSNikolay Borisov 2568b8522a1eSNikolay Borisov if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) 2569b8522a1eSNikolay Borisov break; 2570b8522a1eSNikolay Borisov 2571b8522a1eSNikolay Borisov free_root_pointers(fs_info, 0); 2572b8522a1eSNikolay Borisov 2573b8522a1eSNikolay Borisov /* 2574b8522a1eSNikolay Borisov * Don't use the log in recovery mode, it won't be 2575b8522a1eSNikolay Borisov * valid 2576b8522a1eSNikolay Borisov */ 2577b8522a1eSNikolay Borisov btrfs_set_super_log_root(sb, 0); 2578b8522a1eSNikolay Borisov 2579b8522a1eSNikolay Borisov /* We can't trust the free space cache either */ 2580b8522a1eSNikolay Borisov btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 2581b8522a1eSNikolay Borisov 2582b8522a1eSNikolay Borisov ret = read_backup_root(fs_info, i); 25836ef108ddSNikolay Borisov backup_index = ret; 2584b8522a1eSNikolay Borisov if (ret < 0) 2585b8522a1eSNikolay Borisov return ret; 2586b8522a1eSNikolay Borisov } 2587b8522a1eSNikolay Borisov generation = btrfs_super_generation(sb); 2588b8522a1eSNikolay Borisov level = btrfs_super_root_level(sb); 2589b8522a1eSNikolay Borisov tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb), 2590b8522a1eSNikolay Borisov generation, level, NULL); 2591b8522a1eSNikolay Borisov if (IS_ERR(tree_root->node) || 2592b8522a1eSNikolay Borisov !extent_buffer_uptodate(tree_root->node)) { 2593b8522a1eSNikolay Borisov handle_error = true; 2594b8522a1eSNikolay Borisov 2595b8522a1eSNikolay Borisov if (IS_ERR(tree_root->node)) 2596b8522a1eSNikolay Borisov ret = PTR_ERR(tree_root->node); 2597b8522a1eSNikolay Borisov else if (!extent_buffer_uptodate(tree_root->node)) 2598b8522a1eSNikolay Borisov ret = -EUCLEAN; 2599b8522a1eSNikolay Borisov 2600b8522a1eSNikolay Borisov btrfs_warn(fs_info, "failed to read tree root"); 2601b8522a1eSNikolay Borisov continue; 2602b8522a1eSNikolay Borisov } 2603b8522a1eSNikolay Borisov 2604b8522a1eSNikolay Borisov btrfs_set_root_node(&tree_root->root_item, tree_root->node); 2605b8522a1eSNikolay Borisov tree_root->commit_root = btrfs_root_node(tree_root); 2606b8522a1eSNikolay Borisov btrfs_set_root_refs(&tree_root->root_item, 1); 2607b8522a1eSNikolay Borisov 2608336a0d8dSNikolay Borisov /* 2609336a0d8dSNikolay Borisov * No need to hold btrfs_root::objectid_mutex since the fs 2610336a0d8dSNikolay Borisov * hasn't been fully initialised and we are the only user 2611336a0d8dSNikolay Borisov */ 2612b8522a1eSNikolay Borisov ret = btrfs_find_highest_objectid(tree_root, 2613b8522a1eSNikolay Borisov &tree_root->highest_objectid); 2614b8522a1eSNikolay Borisov if (ret < 0) { 2615b8522a1eSNikolay Borisov handle_error = true; 2616b8522a1eSNikolay Borisov continue; 2617b8522a1eSNikolay Borisov } 2618b8522a1eSNikolay Borisov 2619b8522a1eSNikolay Borisov ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 2620b8522a1eSNikolay Borisov 2621b8522a1eSNikolay Borisov ret = btrfs_read_roots(fs_info); 2622b8522a1eSNikolay Borisov if (ret < 0) { 2623b8522a1eSNikolay Borisov handle_error = true; 2624b8522a1eSNikolay Borisov continue; 2625b8522a1eSNikolay Borisov } 2626b8522a1eSNikolay Borisov 2627b8522a1eSNikolay Borisov /* All successful */ 2628b8522a1eSNikolay Borisov fs_info->generation = generation; 2629b8522a1eSNikolay Borisov fs_info->last_trans_committed = generation; 26306ef108ddSNikolay Borisov 26316ef108ddSNikolay Borisov /* Always begin writing backup roots after the one being used */ 26326ef108ddSNikolay Borisov if (backup_index < 0) { 26336ef108ddSNikolay Borisov fs_info->backup_root_index = 0; 26346ef108ddSNikolay Borisov } else { 26356ef108ddSNikolay Borisov fs_info->backup_root_index = backup_index + 1; 26366ef108ddSNikolay Borisov fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS; 26376ef108ddSNikolay Borisov } 2638b8522a1eSNikolay Borisov break; 2639b8522a1eSNikolay Borisov } 2640b8522a1eSNikolay Borisov 2641b8522a1eSNikolay Borisov return ret; 2642b8522a1eSNikolay Borisov } 2643b8522a1eSNikolay Borisov 26448260edbaSJosef Bacik void btrfs_init_fs_info(struct btrfs_fs_info *fs_info) 2645eb60ceacSChris Mason { 264676dda93cSYan, Zheng INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 2647f28491e0SJosef Bacik INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 26488fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 2649facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 265024bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 2651eb73c1b7SMiao Xie INIT_LIST_HEAD(&fs_info->delalloc_roots); 265211833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 2653eb73c1b7SMiao Xie spin_lock_init(&fs_info->delalloc_root_lock); 2654a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 265576dda93cSYan, Zheng spin_lock_init(&fs_info->fs_roots_radix_lock); 265624bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 26574cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 2658ceda0864SMiao Xie spin_lock_init(&fs_info->super_lock); 2659f28491e0SJosef Bacik spin_lock_init(&fs_info->buffer_lock); 266047ab2a6cSJosef Bacik spin_lock_init(&fs_info->unused_bgs_lock); 2661f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 2662d7c15171SZhao Lei mutex_init(&fs_info->unused_bg_unpin_mutex); 266367c5e7d4SFilipe Manana mutex_init(&fs_info->delete_unused_bgs_mutex); 26647585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 2665573bfb72SMiao Xie mutex_init(&fs_info->delalloc_root_mutex); 2666de98ced9SMiao Xie seqlock_init(&fs_info->profiles_lock); 266719c00ddcSChris Mason 26680b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 26696324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 2670f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 267147ab2a6cSJosef Bacik INIT_LIST_HEAD(&fs_info->unused_bgs); 2672bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 2673bd647ce3SJosef Bacik INIT_LIST_HEAD(&fs_info->allocated_roots); 26743fd63727SJosef Bacik INIT_LIST_HEAD(&fs_info->allocated_ebs); 26753fd63727SJosef Bacik spin_lock_init(&fs_info->eb_leak_lock); 2676bd647ce3SJosef Bacik #endif 2677c8bf1b67SDavid Sterba extent_map_tree_init(&fs_info->mapping_tree); 267866d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 267966d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 268066d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 268166d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 268266d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 268366d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 268466d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 2685ba2c4d4eSJosef Bacik btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, 2686ba2c4d4eSJosef Bacik BTRFS_BLOCK_RSV_DELREFS); 2687ba2c4d4eSJosef Bacik 2688771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 26894cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 26902fefd558SZhao Lei atomic_set(&fs_info->reada_works_cnt, 0); 2691034f784dSJosef Bacik atomic_set(&fs_info->nr_delayed_iputs, 0); 2692fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 269395ac567aSFilipe David Borba Manana fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 26949ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 26954cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 2696a5ed45f8SNikolay Borisov atomic64_set(&fs_info->free_chunk_space, 0); 2697f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 26988b87dc17SDavid Sterba fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 2699f8c269d7SDavid Sterba fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 270090519d66SArne Jansen /* readahead state */ 2701d0164adcSMel Gorman INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); 270290519d66SArne Jansen spin_lock_init(&fs_info->reada_lock); 2703fd708b81SJosef Bacik btrfs_init_ref_verify(fs_info); 2704c8b97818SChris Mason 2705b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 2706b34b086cSChris Mason num_online_cpus() + 2, 8); 27070afbaf8cSChris Mason 2708199c2a9cSMiao Xie INIT_LIST_HEAD(&fs_info->ordered_roots); 2709199c2a9cSMiao Xie spin_lock_init(&fs_info->ordered_root_lock); 271069fe2d75SJosef Bacik 2711638aa7edSEric Sandeen btrfs_init_scrub(fs_info); 271221adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 271321adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 271421adbd5cSStefan Behrens #endif 2715779a65a4SEric Sandeen btrfs_init_balance(fs_info); 271621c7e756SMiao Xie btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work); 2717a2de733cSArne Jansen 27180f9dd46cSJosef Bacik spin_lock_init(&fs_info->block_group_cache_lock); 27196bef4d31SEric Paris fs_info->block_group_cache_tree = RB_ROOT; 2720a1897fddSLiu Bo fs_info->first_logical_byte = (u64)-1; 27210f9dd46cSJosef Bacik 2722fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &fs_info->excluded_extents, 2723fe119a6eSNikolay Borisov IO_TREE_FS_EXCLUDED_EXTENTS, NULL); 2724afcdd129SJosef Bacik set_bit(BTRFS_FS_BARRIER, &fs_info->flags); 272539279cc3SChris Mason 27265a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 2727e02119d5SChris Mason mutex_init(&fs_info->tree_log_mutex); 2728925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 2729a74a4b97SChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 2730a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 27311bbc621eSChris Mason mutex_init(&fs_info->ro_block_group_mutex); 27329e351cc8SJosef Bacik init_rwsem(&fs_info->commit_root_sem); 2733c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 273476dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 2735803b2f54SStefan Behrens sema_init(&fs_info->uuid_tree_rescan_sem, 1); 2736fa9c0d79SChris Mason 2737ad618368SEric Sandeen btrfs_init_dev_replace_locks(fs_info); 2738f9e92e40SEric Sandeen btrfs_init_qgroup(fs_info); 2739b0643e59SDennis Zhou btrfs_discard_init(fs_info); 2740416ac51dSArne Jansen 2741fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2742fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2743fa9c0d79SChris Mason 2744e6dcd2dcSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 2745f9295749SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 2746bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 27474854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 2748034f784dSJosef Bacik init_waitqueue_head(&fs_info->delayed_iputs_wait); 27493768f368SChris Mason 2750da17066cSJeff Mahoney /* Usable values until the real ones are cached from the superblock */ 2751da17066cSJeff Mahoney fs_info->nodesize = 4096; 2752da17066cSJeff Mahoney fs_info->sectorsize = 4096; 2753da17066cSJeff Mahoney fs_info->stripesize = 4096; 2754da17066cSJeff Mahoney 2755eede2bf3SOmar Sandoval spin_lock_init(&fs_info->swapfile_pins_lock); 2756eede2bf3SOmar Sandoval fs_info->swapfile_pins = RB_ROOT; 2757eede2bf3SOmar Sandoval 27589e967495SFilipe Manana fs_info->send_in_progress = 0; 27598260edbaSJosef Bacik } 27608260edbaSJosef Bacik 27618260edbaSJosef Bacik static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb) 27628260edbaSJosef Bacik { 27638260edbaSJosef Bacik int ret; 27648260edbaSJosef Bacik 27658260edbaSJosef Bacik fs_info->sb = sb; 27668260edbaSJosef Bacik sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; 27678260edbaSJosef Bacik sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); 27689e967495SFilipe Manana 2769c75e8394SJosef Bacik ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL); 2770ae18c37aSJosef Bacik if (ret) 2771ae18c37aSJosef Bacik return ret; 2772ae18c37aSJosef Bacik 2773ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 2774ae18c37aSJosef Bacik if (ret) 2775c75e8394SJosef Bacik return ret; 2776ae18c37aSJosef Bacik 2777ae18c37aSJosef Bacik fs_info->dirty_metadata_batch = PAGE_SIZE * 2778ae18c37aSJosef Bacik (1 + ilog2(nr_cpu_ids)); 2779ae18c37aSJosef Bacik 2780ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 2781ae18c37aSJosef Bacik if (ret) 2782c75e8394SJosef Bacik return ret; 2783ae18c37aSJosef Bacik 2784ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, 2785ae18c37aSJosef Bacik GFP_KERNEL); 2786ae18c37aSJosef Bacik if (ret) 2787c75e8394SJosef Bacik return ret; 2788ae18c37aSJosef Bacik 2789ae18c37aSJosef Bacik fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 2790ae18c37aSJosef Bacik GFP_KERNEL); 2791c75e8394SJosef Bacik if (!fs_info->delayed_root) 2792c75e8394SJosef Bacik return -ENOMEM; 2793ae18c37aSJosef Bacik btrfs_init_delayed_root(fs_info->delayed_root); 2794ae18c37aSJosef Bacik 2795c75e8394SJosef Bacik return btrfs_alloc_stripe_hash_table(fs_info); 2796ae18c37aSJosef Bacik } 2797ae18c37aSJosef Bacik 279897f4dd09SNikolay Borisov static int btrfs_uuid_rescan_kthread(void *data) 279997f4dd09SNikolay Borisov { 280097f4dd09SNikolay Borisov struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; 280197f4dd09SNikolay Borisov int ret; 280297f4dd09SNikolay Borisov 280397f4dd09SNikolay Borisov /* 280497f4dd09SNikolay Borisov * 1st step is to iterate through the existing UUID tree and 280597f4dd09SNikolay Borisov * to delete all entries that contain outdated data. 280697f4dd09SNikolay Borisov * 2nd step is to add all missing entries to the UUID tree. 280797f4dd09SNikolay Borisov */ 280897f4dd09SNikolay Borisov ret = btrfs_uuid_tree_iterate(fs_info); 280997f4dd09SNikolay Borisov if (ret < 0) { 2810c94bec2cSJosef Bacik if (ret != -EINTR) 2811c94bec2cSJosef Bacik btrfs_warn(fs_info, "iterating uuid_tree failed %d", 2812c94bec2cSJosef Bacik ret); 281397f4dd09SNikolay Borisov up(&fs_info->uuid_tree_rescan_sem); 281497f4dd09SNikolay Borisov return ret; 281597f4dd09SNikolay Borisov } 281697f4dd09SNikolay Borisov return btrfs_uuid_scan_kthread(data); 281797f4dd09SNikolay Borisov } 281897f4dd09SNikolay Borisov 281997f4dd09SNikolay Borisov static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) 282097f4dd09SNikolay Borisov { 282197f4dd09SNikolay Borisov struct task_struct *task; 282297f4dd09SNikolay Borisov 282397f4dd09SNikolay Borisov down(&fs_info->uuid_tree_rescan_sem); 282497f4dd09SNikolay Borisov task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); 282597f4dd09SNikolay Borisov if (IS_ERR(task)) { 282697f4dd09SNikolay Borisov /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 282797f4dd09SNikolay Borisov btrfs_warn(fs_info, "failed to start uuid_rescan task"); 282897f4dd09SNikolay Borisov up(&fs_info->uuid_tree_rescan_sem); 282997f4dd09SNikolay Borisov return PTR_ERR(task); 283097f4dd09SNikolay Borisov } 283197f4dd09SNikolay Borisov 283297f4dd09SNikolay Borisov return 0; 283397f4dd09SNikolay Borisov } 283497f4dd09SNikolay Borisov 2835ae18c37aSJosef Bacik int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices, 2836ae18c37aSJosef Bacik char *options) 2837ae18c37aSJosef Bacik { 2838ae18c37aSJosef Bacik u32 sectorsize; 2839ae18c37aSJosef Bacik u32 nodesize; 2840ae18c37aSJosef Bacik u32 stripesize; 2841ae18c37aSJosef Bacik u64 generation; 2842ae18c37aSJosef Bacik u64 features; 2843ae18c37aSJosef Bacik u16 csum_type; 2844ae18c37aSJosef Bacik struct btrfs_super_block *disk_super; 2845ae18c37aSJosef Bacik struct btrfs_fs_info *fs_info = btrfs_sb(sb); 2846ae18c37aSJosef Bacik struct btrfs_root *tree_root; 2847ae18c37aSJosef Bacik struct btrfs_root *chunk_root; 2848ae18c37aSJosef Bacik int ret; 2849ae18c37aSJosef Bacik int err = -EINVAL; 2850ae18c37aSJosef Bacik int clear_free_space_tree = 0; 2851ae18c37aSJosef Bacik int level; 2852ae18c37aSJosef Bacik 28538260edbaSJosef Bacik ret = init_mount_fs_info(fs_info, sb); 285453b381b3SDavid Woodhouse if (ret) { 285583c8266aSDavid Sterba err = ret; 2856ae18c37aSJosef Bacik goto fail; 285753b381b3SDavid Woodhouse } 285853b381b3SDavid Woodhouse 2859ae18c37aSJosef Bacik /* These need to be init'ed before we start creating inodes and such. */ 2860ae18c37aSJosef Bacik tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, 2861ae18c37aSJosef Bacik GFP_KERNEL); 2862ae18c37aSJosef Bacik fs_info->tree_root = tree_root; 2863ae18c37aSJosef Bacik chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID, 2864ae18c37aSJosef Bacik GFP_KERNEL); 2865ae18c37aSJosef Bacik fs_info->chunk_root = chunk_root; 2866ae18c37aSJosef Bacik if (!tree_root || !chunk_root) { 2867ae18c37aSJosef Bacik err = -ENOMEM; 2868c75e8394SJosef Bacik goto fail; 2869ae18c37aSJosef Bacik } 2870ae18c37aSJosef Bacik 2871ae18c37aSJosef Bacik fs_info->btree_inode = new_inode(sb); 2872ae18c37aSJosef Bacik if (!fs_info->btree_inode) { 2873ae18c37aSJosef Bacik err = -ENOMEM; 2874c75e8394SJosef Bacik goto fail; 2875ae18c37aSJosef Bacik } 2876ae18c37aSJosef Bacik mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 2877ae18c37aSJosef Bacik btrfs_init_btree_inode(fs_info); 2878ae18c37aSJosef Bacik 28793c4bb26bSChris Mason invalidate_bdev(fs_devices->latest_bdev); 28801104a885SDavid Sterba 28811104a885SDavid Sterba /* 28821104a885SDavid Sterba * Read super block and check the signature bytes only 28831104a885SDavid Sterba */ 28848f32380dSJohannes Thumshirn disk_super = btrfs_read_dev_super(fs_devices->latest_bdev); 28858f32380dSJohannes Thumshirn if (IS_ERR(disk_super)) { 28868f32380dSJohannes Thumshirn err = PTR_ERR(disk_super); 288716cdcec7SMiao Xie goto fail_alloc; 288820b45077SDave Young } 288939279cc3SChris Mason 28901104a885SDavid Sterba /* 28918dc3f22cSJohannes Thumshirn * Verify the type first, if that or the the checksum value are 28928dc3f22cSJohannes Thumshirn * corrupted, we'll find out 28938dc3f22cSJohannes Thumshirn */ 28948f32380dSJohannes Thumshirn csum_type = btrfs_super_csum_type(disk_super); 289551bce6c9SJohannes Thumshirn if (!btrfs_supported_super_csum(csum_type)) { 28968dc3f22cSJohannes Thumshirn btrfs_err(fs_info, "unsupported checksum algorithm: %u", 289751bce6c9SJohannes Thumshirn csum_type); 28988dc3f22cSJohannes Thumshirn err = -EINVAL; 28998f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 29008dc3f22cSJohannes Thumshirn goto fail_alloc; 29018dc3f22cSJohannes Thumshirn } 29028dc3f22cSJohannes Thumshirn 29036d97c6e3SJohannes Thumshirn ret = btrfs_init_csum_hash(fs_info, csum_type); 29046d97c6e3SJohannes Thumshirn if (ret) { 29056d97c6e3SJohannes Thumshirn err = ret; 29068f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 29076d97c6e3SJohannes Thumshirn goto fail_alloc; 29086d97c6e3SJohannes Thumshirn } 29096d97c6e3SJohannes Thumshirn 29108dc3f22cSJohannes Thumshirn /* 29111104a885SDavid Sterba * We want to check superblock checksum, the type is stored inside. 29121104a885SDavid Sterba * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 29131104a885SDavid Sterba */ 29148f32380dSJohannes Thumshirn if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) { 291505135f59SDavid Sterba btrfs_err(fs_info, "superblock checksum mismatch"); 29161104a885SDavid Sterba err = -EINVAL; 29178f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 2918141386e1SJosef Bacik goto fail_alloc; 29191104a885SDavid Sterba } 29201104a885SDavid Sterba 29211104a885SDavid Sterba /* 29221104a885SDavid Sterba * super_copy is zeroed at allocation time and we never touch the 29231104a885SDavid Sterba * following bytes up to INFO_SIZE, the checksum is calculated from 29241104a885SDavid Sterba * the whole block of INFO_SIZE 29251104a885SDavid Sterba */ 29268f32380dSJohannes Thumshirn memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy)); 29278f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 29285f39d397SChris Mason 2929fbc6feaeSNikolay Borisov disk_super = fs_info->super_copy; 2930fbc6feaeSNikolay Borisov 2931de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, 2932de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 2933de37aa51SNikolay Borisov 29347239ff4bSNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID)) { 2935de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid, 2936de37aa51SNikolay Borisov fs_info->super_copy->metadata_uuid, 2937de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 29387239ff4bSNikolay Borisov } 29390b86a832SChris Mason 2940fbc6feaeSNikolay Borisov features = btrfs_super_flags(disk_super); 2941fbc6feaeSNikolay Borisov if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { 2942fbc6feaeSNikolay Borisov features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; 2943fbc6feaeSNikolay Borisov btrfs_set_super_flags(disk_super, features); 2944fbc6feaeSNikolay Borisov btrfs_info(fs_info, 2945fbc6feaeSNikolay Borisov "found metadata UUID change in progress flag, clearing"); 2946fbc6feaeSNikolay Borisov } 2947fbc6feaeSNikolay Borisov 2948fbc6feaeSNikolay Borisov memcpy(fs_info->super_for_commit, fs_info->super_copy, 2949fbc6feaeSNikolay Borisov sizeof(*fs_info->super_for_commit)); 2950de37aa51SNikolay Borisov 2951069ec957SQu Wenruo ret = btrfs_validate_mount_super(fs_info); 29521104a885SDavid Sterba if (ret) { 295305135f59SDavid Sterba btrfs_err(fs_info, "superblock contains fatal errors"); 29541104a885SDavid Sterba err = -EINVAL; 2955141386e1SJosef Bacik goto fail_alloc; 29561104a885SDavid Sterba } 29571104a885SDavid Sterba 29580f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 2959141386e1SJosef Bacik goto fail_alloc; 29600f7d52f4SChris Mason 2961acce952bSliubo /* check FS state, whether FS is broken. */ 296287533c47SMiao Xie if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 296387533c47SMiao Xie set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 2964acce952bSliubo 296575e7cb7fSLiu Bo /* 296675e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 296775e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 296875e7cb7fSLiu Bo */ 296975e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 297075e7cb7fSLiu Bo 29712ff7e61eSJeff Mahoney ret = btrfs_parse_options(fs_info, options, sb->s_flags); 29722b82032cSYan Zheng if (ret) { 29732b82032cSYan Zheng err = ret; 2974141386e1SJosef Bacik goto fail_alloc; 29752b82032cSYan Zheng } 2976dfe25020SChris Mason 2977f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 2978f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 2979f2b636e8SJosef Bacik if (features) { 298005135f59SDavid Sterba btrfs_err(fs_info, 298105135f59SDavid Sterba "cannot mount because of unsupported optional features (%llx)", 2982c1c9ff7cSGeert Uytterhoeven features); 2983f2b636e8SJosef Bacik err = -EINVAL; 2984141386e1SJosef Bacik goto fail_alloc; 2985f2b636e8SJosef Bacik } 2986f2b636e8SJosef Bacik 29875d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 29885d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 29890b246afaSJeff Mahoney if (fs_info->compress_type == BTRFS_COMPRESS_LZO) 2990a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 29915c1aab1dSNick Terrell else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) 29925c1aab1dSNick Terrell features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; 2993727011e0SChris Mason 29943173a18fSJosef Bacik if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) 299505135f59SDavid Sterba btrfs_info(fs_info, "has skinny extents"); 29963173a18fSJosef Bacik 2997727011e0SChris Mason /* 2998727011e0SChris Mason * flag our filesystem as having big metadata blocks if 2999727011e0SChris Mason * they are bigger than the page size 3000727011e0SChris Mason */ 300109cbfeafSKirill A. Shutemov if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) { 3002727011e0SChris Mason if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 300305135f59SDavid Sterba btrfs_info(fs_info, 300405135f59SDavid Sterba "flagging fs with big metadata feature"); 3005727011e0SChris Mason features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 3006727011e0SChris Mason } 3007727011e0SChris Mason 3008bc3f116fSChris Mason nodesize = btrfs_super_nodesize(disk_super); 3009bc3f116fSChris Mason sectorsize = btrfs_super_sectorsize(disk_super); 3010b7f67055SChandan Rajendra stripesize = sectorsize; 3011707e8a07SDavid Sterba fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 3012963d678bSMiao Xie fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 3013bc3f116fSChris Mason 3014da17066cSJeff Mahoney /* Cache block sizes */ 3015da17066cSJeff Mahoney fs_info->nodesize = nodesize; 3016da17066cSJeff Mahoney fs_info->sectorsize = sectorsize; 3017da17066cSJeff Mahoney fs_info->stripesize = stripesize; 3018da17066cSJeff Mahoney 3019bc3f116fSChris Mason /* 3020bc3f116fSChris Mason * mixed block groups end up with duplicate but slightly offset 3021bc3f116fSChris Mason * extent buffers for the same range. It leads to corruptions 3022bc3f116fSChris Mason */ 3023bc3f116fSChris Mason if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 3024707e8a07SDavid Sterba (sectorsize != nodesize)) { 302505135f59SDavid Sterba btrfs_err(fs_info, 302605135f59SDavid Sterba "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", 302705135f59SDavid Sterba nodesize, sectorsize); 3028141386e1SJosef Bacik goto fail_alloc; 3029bc3f116fSChris Mason } 3030bc3f116fSChris Mason 3031ceda0864SMiao Xie /* 3032ceda0864SMiao Xie * Needn't use the lock because there is no other task which will 3033ceda0864SMiao Xie * update the flag. 3034ceda0864SMiao Xie */ 30355d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 30365d4f98a2SYan Zheng 3037f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 3038f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 3039bc98a42cSDavid Howells if (!sb_rdonly(sb) && features) { 304005135f59SDavid Sterba btrfs_err(fs_info, 304105135f59SDavid Sterba "cannot mount read-write because of unsupported optional features (%llx)", 3042c1c9ff7cSGeert Uytterhoeven features); 3043f2b636e8SJosef Bacik err = -EINVAL; 3044141386e1SJosef Bacik goto fail_alloc; 3045f2b636e8SJosef Bacik } 304661d92c32SChris Mason 30472a458198SEric Sandeen ret = btrfs_init_workqueues(fs_info, fs_devices); 30482a458198SEric Sandeen if (ret) { 30492a458198SEric Sandeen err = ret; 30500dc3b84aSJosef Bacik goto fail_sb_buffer; 30510dc3b84aSJosef Bacik } 30524543df7eSChris Mason 30539e11ceeeSJan Kara sb->s_bdi->congested_fn = btrfs_congested_fn; 30549e11ceeeSJan Kara sb->s_bdi->congested_data = fs_info; 30559e11ceeeSJan Kara sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK; 3056b5420237SNikolay Borisov sb->s_bdi->ra_pages = VM_READAHEAD_PAGES; 30579e11ceeeSJan Kara sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); 30589e11ceeeSJan Kara sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); 30594575c9ccSChris Mason 3060a061fc8dSChris Mason sb->s_blocksize = sectorsize; 3061a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 3062de37aa51SNikolay Borisov memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); 3063db94535dSChris Mason 3064925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 30656bccf3abSJeff Mahoney ret = btrfs_read_sys_array(fs_info); 3066925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 306784eed90fSChris Mason if (ret) { 306805135f59SDavid Sterba btrfs_err(fs_info, "failed to read the system array: %d", ret); 30695d4f98a2SYan Zheng goto fail_sb_buffer; 307084eed90fSChris Mason } 30710b86a832SChris Mason 307284234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 3073581c1760SQu Wenruo level = btrfs_super_chunk_root_level(disk_super); 30740b86a832SChris Mason 30752ff7e61eSJeff Mahoney chunk_root->node = read_tree_block(fs_info, 30760b86a832SChris Mason btrfs_super_chunk_root(disk_super), 3077581c1760SQu Wenruo generation, level, NULL); 307864c043deSLiu Bo if (IS_ERR(chunk_root->node) || 307964c043deSLiu Bo !extent_buffer_uptodate(chunk_root->node)) { 308005135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk root"); 3081e5fffbacSchandan if (!IS_ERR(chunk_root->node)) 3082e5fffbacSchandan free_extent_buffer(chunk_root->node); 308395ab1f64SZhao Lei chunk_root->node = NULL; 3084af31f5e5SChris Mason goto fail_tree_roots; 308583121942SDavid Woodhouse } 30865d4f98a2SYan Zheng btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 30875d4f98a2SYan Zheng chunk_root->commit_root = btrfs_root_node(chunk_root); 30880b86a832SChris Mason 3089e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 3090c4ac7541SDavid Sterba offsetof(struct btrfs_header, chunk_tree_uuid), 3091c4ac7541SDavid Sterba BTRFS_UUID_SIZE); 3092e17cade2SChris Mason 30935b4aacefSJeff Mahoney ret = btrfs_read_chunk_tree(fs_info); 30942b82032cSYan Zheng if (ret) { 309505135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk tree: %d", ret); 3096af31f5e5SChris Mason goto fail_tree_roots; 30972b82032cSYan Zheng } 30980b86a832SChris Mason 30998dabb742SStefan Behrens /* 31009b99b115SAnand Jain * Keep the devid that is marked to be the target device for the 31019b99b115SAnand Jain * device replace procedure 31028dabb742SStefan Behrens */ 31039b99b115SAnand Jain btrfs_free_extra_devids(fs_devices, 0); 3104dfe25020SChris Mason 3105a6b0d5c8SChris Mason if (!fs_devices->latest_bdev) { 310605135f59SDavid Sterba btrfs_err(fs_info, "failed to read devices"); 3107a6b0d5c8SChris Mason goto fail_tree_roots; 3108a6b0d5c8SChris Mason } 3109a6b0d5c8SChris Mason 3110b8522a1eSNikolay Borisov ret = init_tree_roots(fs_info); 31114bbcaa64SEric Sandeen if (ret) 3112b8522a1eSNikolay Borisov goto fail_tree_roots; 31138929ecfaSYan, Zheng 311475ec1db8SJosef Bacik /* 311575ec1db8SJosef Bacik * If we have a uuid root and we're not being told to rescan we need to 311675ec1db8SJosef Bacik * check the generation here so we can set the 311775ec1db8SJosef Bacik * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the 311875ec1db8SJosef Bacik * transaction during a balance or the log replay without updating the 311975ec1db8SJosef Bacik * uuid generation, and then if we crash we would rescan the uuid tree, 312075ec1db8SJosef Bacik * even though it was perfectly fine. 312175ec1db8SJosef Bacik */ 312275ec1db8SJosef Bacik if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) && 312375ec1db8SJosef Bacik fs_info->generation == btrfs_super_uuid_tree_generation(disk_super)) 312475ec1db8SJosef Bacik set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 312575ec1db8SJosef Bacik 3126cf90d884SQu Wenruo ret = btrfs_verify_dev_extents(fs_info); 3127cf90d884SQu Wenruo if (ret) { 3128cf90d884SQu Wenruo btrfs_err(fs_info, 3129cf90d884SQu Wenruo "failed to verify dev extents against chunks: %d", 3130cf90d884SQu Wenruo ret); 3131cf90d884SQu Wenruo goto fail_block_groups; 3132cf90d884SQu Wenruo } 313368310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 313468310a5eSIlya Dryomov if (ret) { 313505135f59SDavid Sterba btrfs_err(fs_info, "failed to recover balance: %d", ret); 313668310a5eSIlya Dryomov goto fail_block_groups; 313768310a5eSIlya Dryomov } 313868310a5eSIlya Dryomov 3139733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 3140733f4fbbSStefan Behrens if (ret) { 314105135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_stats: %d", ret); 3142733f4fbbSStefan Behrens goto fail_block_groups; 3143733f4fbbSStefan Behrens } 3144733f4fbbSStefan Behrens 31458dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 31468dabb742SStefan Behrens if (ret) { 314705135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_replace: %d", ret); 31488dabb742SStefan Behrens goto fail_block_groups; 31498dabb742SStefan Behrens } 31508dabb742SStefan Behrens 31519b99b115SAnand Jain btrfs_free_extra_devids(fs_devices, 1); 31528dabb742SStefan Behrens 3153c6761a9eSAnand Jain ret = btrfs_sysfs_add_fsid(fs_devices); 3154b7c35e81SAnand Jain if (ret) { 315505135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", 315605135f59SDavid Sterba ret); 3157b7c35e81SAnand Jain goto fail_block_groups; 3158b7c35e81SAnand Jain } 3159b7c35e81SAnand Jain 316096f3136eSAnand Jain ret = btrfs_sysfs_add_mounted(fs_info); 31615ac1d209SJeff Mahoney if (ret) { 316205135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); 3163b7c35e81SAnand Jain goto fail_fsdev_sysfs; 31645ac1d209SJeff Mahoney } 31655ac1d209SJeff Mahoney 3166c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 3167c59021f8Sliubo if (ret) { 316805135f59SDavid Sterba btrfs_err(fs_info, "failed to initialize space info: %d", ret); 31692365dd3cSAnand Jain goto fail_sysfs; 3170c59021f8Sliubo } 3171c59021f8Sliubo 31725b4aacefSJeff Mahoney ret = btrfs_read_block_groups(fs_info); 31731b1d1f66SJosef Bacik if (ret) { 317405135f59SDavid Sterba btrfs_err(fs_info, "failed to read block groups: %d", ret); 31752365dd3cSAnand Jain goto fail_sysfs; 31761b1d1f66SJosef Bacik } 31774330e183SQu Wenruo 31786528b99dSAnand Jain if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) { 317905135f59SDavid Sterba btrfs_warn(fs_info, 318052042d8eSAndrea Gelmini "writable mount is not allowed due to too many missing devices"); 31812365dd3cSAnand Jain goto fail_sysfs; 3182292fd7fcSStefan Behrens } 31839078a3e1SChris Mason 3184a74a4b97SChris Mason fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 3185a74a4b97SChris Mason "btrfs-cleaner"); 318657506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 31872365dd3cSAnand Jain goto fail_sysfs; 3188a74a4b97SChris Mason 3189a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 3190a74a4b97SChris Mason tree_root, 3191a74a4b97SChris Mason "btrfs-transaction"); 319257506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 31933f157a2fSChris Mason goto fail_cleaner; 3194a74a4b97SChris Mason 3195583b7231SHans van Kranenburg if (!btrfs_test_opt(fs_info, NOSSD) && 3196c289811cSChris Mason !fs_info->fs_devices->rotating) { 3197583b7231SHans van Kranenburg btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); 3198c289811cSChris Mason } 3199c289811cSChris Mason 3200572d9ab7SDavid Sterba /* 320101327610SNicholas D Steeves * Mount does not set all options immediately, we can do it now and do 3202572d9ab7SDavid Sterba * not have to wait for transaction commit 3203572d9ab7SDavid Sterba */ 3204572d9ab7SDavid Sterba btrfs_apply_pending_changes(fs_info); 32053818aea2SQu Wenruo 320621adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 32070b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { 32082ff7e61eSJeff Mahoney ret = btrfsic_mount(fs_info, fs_devices, 32090b246afaSJeff Mahoney btrfs_test_opt(fs_info, 321021adbd5cSStefan Behrens CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? 321121adbd5cSStefan Behrens 1 : 0, 321221adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 321321adbd5cSStefan Behrens if (ret) 321405135f59SDavid Sterba btrfs_warn(fs_info, 321505135f59SDavid Sterba "failed to initialize integrity check module: %d", 321605135f59SDavid Sterba ret); 321721adbd5cSStefan Behrens } 321821adbd5cSStefan Behrens #endif 3219bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 3220bcef60f2SArne Jansen if (ret) 3221bcef60f2SArne Jansen goto fail_trans_kthread; 322221adbd5cSStefan Behrens 3223fd708b81SJosef Bacik if (btrfs_build_ref_tree(fs_info)) 3224fd708b81SJosef Bacik btrfs_err(fs_info, "couldn't build ref tree"); 3225fd708b81SJosef Bacik 322696da0919SQu Wenruo /* do not make disk changes in broken FS or nologreplay is given */ 322796da0919SQu Wenruo if (btrfs_super_log_root(disk_super) != 0 && 32280b246afaSJeff Mahoney !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 3229e8294f2fSDavid Sterba btrfs_info(fs_info, "start tree-log replay"); 323063443bf5SEric Sandeen ret = btrfs_replay_log(fs_info, fs_devices); 323179787eaaSJeff Mahoney if (ret) { 323263443bf5SEric Sandeen err = ret; 323328c16cbbSWang Shilong goto fail_qgroup; 3234e556ce2cSYan Zheng } 3235e02119d5SChris Mason } 32361a40e23bSZheng Yan 32376bccf3abSJeff Mahoney ret = btrfs_find_orphan_roots(fs_info); 323879787eaaSJeff Mahoney if (ret) 323928c16cbbSWang Shilong goto fail_qgroup; 324076dda93cSYan, Zheng 3241bc98a42cSDavid Howells if (!sb_rdonly(sb)) { 3242d68fc57bSYan, Zheng ret = btrfs_cleanup_fs_roots(fs_info); 324344c44af2SIlya Dryomov if (ret) 324428c16cbbSWang Shilong goto fail_qgroup; 324590c711abSZygo Blaxell 324690c711abSZygo Blaxell mutex_lock(&fs_info->cleaner_mutex); 32475d4f98a2SYan Zheng ret = btrfs_recover_relocation(tree_root); 324890c711abSZygo Blaxell mutex_unlock(&fs_info->cleaner_mutex); 3249d7ce5843SMiao Xie if (ret < 0) { 325005135f59SDavid Sterba btrfs_warn(fs_info, "failed to recover relocation: %d", 325105135f59SDavid Sterba ret); 3252d7ce5843SMiao Xie err = -EINVAL; 3253bcef60f2SArne Jansen goto fail_qgroup; 3254d7ce5843SMiao Xie } 32557c2ca468SChris Mason } 32561a40e23bSZheng Yan 325756e9357aSDavid Sterba fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true); 32583140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 32593140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 3260f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read fs tree: %d", err); 3261315bf8efSJosef Bacik fs_info->fs_root = NULL; 3262bcef60f2SArne Jansen goto fail_qgroup; 32633140c9a3SDan Carpenter } 3264c289811cSChris Mason 3265bc98a42cSDavid Howells if (sb_rdonly(sb)) 32662b6ba629SIlya Dryomov return 0; 32672b6ba629SIlya Dryomov 3268f8d468a1SOmar Sandoval if (btrfs_test_opt(fs_info, CLEAR_CACHE) && 3269f8d468a1SOmar Sandoval btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 32706675df31SOmar Sandoval clear_free_space_tree = 1; 32716675df31SOmar Sandoval } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 32726675df31SOmar Sandoval !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { 32736675df31SOmar Sandoval btrfs_warn(fs_info, "free space tree is invalid"); 32746675df31SOmar Sandoval clear_free_space_tree = 1; 32756675df31SOmar Sandoval } 32766675df31SOmar Sandoval 32776675df31SOmar Sandoval if (clear_free_space_tree) { 3278f8d468a1SOmar Sandoval btrfs_info(fs_info, "clearing free space tree"); 3279f8d468a1SOmar Sandoval ret = btrfs_clear_free_space_tree(fs_info); 3280f8d468a1SOmar Sandoval if (ret) { 3281f8d468a1SOmar Sandoval btrfs_warn(fs_info, 3282f8d468a1SOmar Sandoval "failed to clear free space tree: %d", ret); 32836bccf3abSJeff Mahoney close_ctree(fs_info); 3284f8d468a1SOmar Sandoval return ret; 3285f8d468a1SOmar Sandoval } 3286f8d468a1SOmar Sandoval } 3287f8d468a1SOmar Sandoval 32880b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && 3289511711afSChris Mason !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 329005135f59SDavid Sterba btrfs_info(fs_info, "creating free space tree"); 3291511711afSChris Mason ret = btrfs_create_free_space_tree(fs_info); 3292511711afSChris Mason if (ret) { 329305135f59SDavid Sterba btrfs_warn(fs_info, 329405135f59SDavid Sterba "failed to create free space tree: %d", ret); 32956bccf3abSJeff Mahoney close_ctree(fs_info); 3296511711afSChris Mason return ret; 3297511711afSChris Mason } 3298511711afSChris Mason } 3299511711afSChris Mason 3300e3acc2a6SJosef Bacik down_read(&fs_info->cleanup_work_sem); 33012b6ba629SIlya Dryomov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 33022b6ba629SIlya Dryomov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 33032b6ba629SIlya Dryomov up_read(&fs_info->cleanup_work_sem); 33046bccf3abSJeff Mahoney close_ctree(fs_info); 33052b6ba629SIlya Dryomov return ret; 33062b6ba629SIlya Dryomov } 3307e3acc2a6SJosef Bacik up_read(&fs_info->cleanup_work_sem); 330859641015SIlya Dryomov 33092b6ba629SIlya Dryomov ret = btrfs_resume_balance_async(fs_info); 33102b6ba629SIlya Dryomov if (ret) { 331105135f59SDavid Sterba btrfs_warn(fs_info, "failed to resume balance: %d", ret); 33126bccf3abSJeff Mahoney close_ctree(fs_info); 33132b6ba629SIlya Dryomov return ret; 3314e3acc2a6SJosef Bacik } 3315e3acc2a6SJosef Bacik 33168dabb742SStefan Behrens ret = btrfs_resume_dev_replace_async(fs_info); 33178dabb742SStefan Behrens if (ret) { 331805135f59SDavid Sterba btrfs_warn(fs_info, "failed to resume device replace: %d", ret); 33196bccf3abSJeff Mahoney close_ctree(fs_info); 33208dabb742SStefan Behrens return ret; 33218dabb742SStefan Behrens } 33228dabb742SStefan Behrens 3323b382a324SJan Schmidt btrfs_qgroup_rescan_resume(fs_info); 3324b0643e59SDennis Zhou btrfs_discard_resume(fs_info); 3325b382a324SJan Schmidt 33264bbcaa64SEric Sandeen if (!fs_info->uuid_root) { 332705135f59SDavid Sterba btrfs_info(fs_info, "creating UUID tree"); 3328f7a81ea4SStefan Behrens ret = btrfs_create_uuid_tree(fs_info); 3329f7a81ea4SStefan Behrens if (ret) { 333005135f59SDavid Sterba btrfs_warn(fs_info, 333105135f59SDavid Sterba "failed to create the UUID tree: %d", ret); 33326bccf3abSJeff Mahoney close_ctree(fs_info); 3333f7a81ea4SStefan Behrens return ret; 3334f7a81ea4SStefan Behrens } 33350b246afaSJeff Mahoney } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || 33364bbcaa64SEric Sandeen fs_info->generation != 33374bbcaa64SEric Sandeen btrfs_super_uuid_tree_generation(disk_super)) { 333805135f59SDavid Sterba btrfs_info(fs_info, "checking UUID tree"); 333970f80175SStefan Behrens ret = btrfs_check_uuid_tree(fs_info); 334070f80175SStefan Behrens if (ret) { 334105135f59SDavid Sterba btrfs_warn(fs_info, 334205135f59SDavid Sterba "failed to check the UUID tree: %d", ret); 33436bccf3abSJeff Mahoney close_ctree(fs_info); 334470f80175SStefan Behrens return ret; 334570f80175SStefan Behrens } 3346f7a81ea4SStefan Behrens } 3347afcdd129SJosef Bacik set_bit(BTRFS_FS_OPEN, &fs_info->flags); 334847ab2a6cSJosef Bacik 33498dcddfa0SQu Wenruo /* 33508dcddfa0SQu Wenruo * backuproot only affect mount behavior, and if open_ctree succeeded, 33518dcddfa0SQu Wenruo * no need to keep the flag 33528dcddfa0SQu Wenruo */ 33538dcddfa0SQu Wenruo btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT); 33548dcddfa0SQu Wenruo 3355ad2b2c80SAl Viro return 0; 335639279cc3SChris Mason 3357bcef60f2SArne Jansen fail_qgroup: 3358bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 33597c2ca468SChris Mason fail_trans_kthread: 33607c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 33612ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 3362faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 33633f157a2fSChris Mason fail_cleaner: 3364a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 33657c2ca468SChris Mason 33667c2ca468SChris Mason /* 33677c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 33687c2ca468SChris Mason * kthreads 33697c2ca468SChris Mason */ 33707c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 33717c2ca468SChris Mason 33722365dd3cSAnand Jain fail_sysfs: 33736618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 33742365dd3cSAnand Jain 3375b7c35e81SAnand Jain fail_fsdev_sysfs: 3376b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3377b7c35e81SAnand Jain 33781b1d1f66SJosef Bacik fail_block_groups: 337954067ae9SJosef Bacik btrfs_put_block_group_cache(fs_info); 3380af31f5e5SChris Mason 3381af31f5e5SChris Mason fail_tree_roots: 33824273eaffSAnand Jain free_root_pointers(fs_info, true); 33832b8195bbSMiao Xie invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3384af31f5e5SChris Mason 338539279cc3SChris Mason fail_sb_buffer: 33867abadb64SLiu Bo btrfs_stop_all_workers(fs_info); 33875cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 338816cdcec7SMiao Xie fail_alloc: 3389586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 3390586e46e2SIlya Dryomov 33914543df7eSChris Mason iput(fs_info->btree_inode); 33927e662854SQinghuang Feng fail: 3393586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 3394ad2b2c80SAl Viro return err; 3395eb60ceacSChris Mason } 3396663faf9fSMasami Hiramatsu ALLOW_ERROR_INJECTION(open_ctree, ERRNO); 3397eb60ceacSChris Mason 3398314b6dd0SJohannes Thumshirn static void btrfs_end_super_write(struct bio *bio) 3399f2984462SChris Mason { 3400314b6dd0SJohannes Thumshirn struct btrfs_device *device = bio->bi_private; 3401314b6dd0SJohannes Thumshirn struct bio_vec *bvec; 3402314b6dd0SJohannes Thumshirn struct bvec_iter_all iter_all; 3403314b6dd0SJohannes Thumshirn struct page *page; 3404442a4f63SStefan Behrens 3405314b6dd0SJohannes Thumshirn bio_for_each_segment_all(bvec, bio, iter_all) { 3406314b6dd0SJohannes Thumshirn page = bvec->bv_page; 3407314b6dd0SJohannes Thumshirn 3408314b6dd0SJohannes Thumshirn if (bio->bi_status) { 3409fb456252SJeff Mahoney btrfs_warn_rl_in_rcu(device->fs_info, 3410314b6dd0SJohannes Thumshirn "lost page write due to IO error on %s (%d)", 3411314b6dd0SJohannes Thumshirn rcu_str_deref(device->name), 3412314b6dd0SJohannes Thumshirn blk_status_to_errno(bio->bi_status)); 3413314b6dd0SJohannes Thumshirn ClearPageUptodate(page); 3414314b6dd0SJohannes Thumshirn SetPageError(page); 3415314b6dd0SJohannes Thumshirn btrfs_dev_stat_inc_and_print(device, 3416314b6dd0SJohannes Thumshirn BTRFS_DEV_STAT_WRITE_ERRS); 3417314b6dd0SJohannes Thumshirn } else { 3418314b6dd0SJohannes Thumshirn SetPageUptodate(page); 3419f2984462SChris Mason } 3420314b6dd0SJohannes Thumshirn 3421314b6dd0SJohannes Thumshirn put_page(page); 3422314b6dd0SJohannes Thumshirn unlock_page(page); 3423314b6dd0SJohannes Thumshirn } 3424314b6dd0SJohannes Thumshirn 3425314b6dd0SJohannes Thumshirn bio_put(bio); 3426f2984462SChris Mason } 3427f2984462SChris Mason 34288f32380dSJohannes Thumshirn struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev, 34298f32380dSJohannes Thumshirn int copy_num) 343029c36d72SAnand Jain { 343129c36d72SAnand Jain struct btrfs_super_block *super; 34328f32380dSJohannes Thumshirn struct page *page; 343329c36d72SAnand Jain u64 bytenr; 34348f32380dSJohannes Thumshirn struct address_space *mapping = bdev->bd_inode->i_mapping; 343529c36d72SAnand Jain 343629c36d72SAnand Jain bytenr = btrfs_sb_offset(copy_num); 343729c36d72SAnand Jain if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode)) 34388f32380dSJohannes Thumshirn return ERR_PTR(-EINVAL); 343929c36d72SAnand Jain 34408f32380dSJohannes Thumshirn page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS); 34418f32380dSJohannes Thumshirn if (IS_ERR(page)) 34428f32380dSJohannes Thumshirn return ERR_CAST(page); 344329c36d72SAnand Jain 34448f32380dSJohannes Thumshirn super = page_address(page); 344529c36d72SAnand Jain if (btrfs_super_bytenr(super) != bytenr || 344629c36d72SAnand Jain btrfs_super_magic(super) != BTRFS_MAGIC) { 34478f32380dSJohannes Thumshirn btrfs_release_disk_super(super); 34488f32380dSJohannes Thumshirn return ERR_PTR(-EINVAL); 344929c36d72SAnand Jain } 345029c36d72SAnand Jain 34518f32380dSJohannes Thumshirn return super; 345229c36d72SAnand Jain } 345329c36d72SAnand Jain 345429c36d72SAnand Jain 34558f32380dSJohannes Thumshirn struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev) 3456a512bbf8SYan Zheng { 34578f32380dSJohannes Thumshirn struct btrfs_super_block *super, *latest = NULL; 3458a512bbf8SYan Zheng int i; 3459a512bbf8SYan Zheng u64 transid = 0; 3460a512bbf8SYan Zheng 3461a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 3462a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 3463a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 3464a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3465a512bbf8SYan Zheng */ 3466a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 34678f32380dSJohannes Thumshirn super = btrfs_read_dev_one_super(bdev, i); 34688f32380dSJohannes Thumshirn if (IS_ERR(super)) 3469a512bbf8SYan Zheng continue; 3470a512bbf8SYan Zheng 3471a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 34728f32380dSJohannes Thumshirn if (latest) 34738f32380dSJohannes Thumshirn btrfs_release_disk_super(super); 34748f32380dSJohannes Thumshirn 34758f32380dSJohannes Thumshirn latest = super; 3476a512bbf8SYan Zheng transid = btrfs_super_generation(super); 3477a512bbf8SYan Zheng } 3478a512bbf8SYan Zheng } 347992fc03fbSAnand Jain 34808f32380dSJohannes Thumshirn return super; 3481a512bbf8SYan Zheng } 3482a512bbf8SYan Zheng 34834eedeb75SHisashi Hifumi /* 3484abbb3b8eSDavid Sterba * Write superblock @sb to the @device. Do not wait for completion, all the 3485314b6dd0SJohannes Thumshirn * pages we use for writing are locked. 34864eedeb75SHisashi Hifumi * 3487abbb3b8eSDavid Sterba * Write @max_mirrors copies of the superblock, where 0 means default that fit 3488abbb3b8eSDavid Sterba * the expected device size at commit time. Note that max_mirrors must be 3489abbb3b8eSDavid Sterba * same for write and wait phases. 34904eedeb75SHisashi Hifumi * 3491314b6dd0SJohannes Thumshirn * Return number of errors when page is not found or submission fails. 34924eedeb75SHisashi Hifumi */ 3493a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 3494abbb3b8eSDavid Sterba struct btrfs_super_block *sb, int max_mirrors) 3495a512bbf8SYan Zheng { 3496d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = device->fs_info; 3497314b6dd0SJohannes Thumshirn struct address_space *mapping = device->bdev->bd_inode->i_mapping; 3498d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 3499a512bbf8SYan Zheng int i; 3500a512bbf8SYan Zheng int errors = 0; 3501a512bbf8SYan Zheng u64 bytenr; 3502a512bbf8SYan Zheng 3503a512bbf8SYan Zheng if (max_mirrors == 0) 3504a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3505a512bbf8SYan Zheng 3506d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 3507d5178578SJohannes Thumshirn 3508a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 3509314b6dd0SJohannes Thumshirn struct page *page; 3510314b6dd0SJohannes Thumshirn struct bio *bio; 3511314b6dd0SJohannes Thumshirn struct btrfs_super_block *disk_super; 3512314b6dd0SJohannes Thumshirn 3513a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 3514935e5cc9SMiao Xie if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3515935e5cc9SMiao Xie device->commit_total_bytes) 3516a512bbf8SYan Zheng break; 3517a512bbf8SYan Zheng 3518a512bbf8SYan Zheng btrfs_set_super_bytenr(sb, bytenr); 3519a512bbf8SYan Zheng 3520fd08001fSEric Biggers crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE, 3521fd08001fSEric Biggers BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, 3522fd08001fSEric Biggers sb->csum); 3523a512bbf8SYan Zheng 3524314b6dd0SJohannes Thumshirn page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT, 3525314b6dd0SJohannes Thumshirn GFP_NOFS); 3526314b6dd0SJohannes Thumshirn if (!page) { 3527fb456252SJeff Mahoney btrfs_err(device->fs_info, 3528314b6dd0SJohannes Thumshirn "couldn't get super block page for bytenr %llu", 3529f14d104dSDavid Sterba bytenr); 3530634554dcSJosef Bacik errors++; 3531634554dcSJosef Bacik continue; 3532634554dcSJosef Bacik } 3533634554dcSJosef Bacik 3534314b6dd0SJohannes Thumshirn /* Bump the refcount for wait_dev_supers() */ 3535314b6dd0SJohannes Thumshirn get_page(page); 3536a512bbf8SYan Zheng 3537314b6dd0SJohannes Thumshirn disk_super = page_address(page); 3538314b6dd0SJohannes Thumshirn memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE); 3539a512bbf8SYan Zheng 3540387125fcSChris Mason /* 3541314b6dd0SJohannes Thumshirn * Directly use bios here instead of relying on the page cache 3542314b6dd0SJohannes Thumshirn * to do I/O, so we don't lose the ability to do integrity 3543314b6dd0SJohannes Thumshirn * checking. 3544387125fcSChris Mason */ 3545314b6dd0SJohannes Thumshirn bio = bio_alloc(GFP_NOFS, 1); 3546314b6dd0SJohannes Thumshirn bio_set_dev(bio, device->bdev); 3547314b6dd0SJohannes Thumshirn bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT; 3548314b6dd0SJohannes Thumshirn bio->bi_private = device; 3549314b6dd0SJohannes Thumshirn bio->bi_end_io = btrfs_end_super_write; 3550314b6dd0SJohannes Thumshirn __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE, 3551314b6dd0SJohannes Thumshirn offset_in_page(bytenr)); 3552314b6dd0SJohannes Thumshirn 3553314b6dd0SJohannes Thumshirn /* 3554314b6dd0SJohannes Thumshirn * We FUA only the first super block. The others we allow to 3555314b6dd0SJohannes Thumshirn * go down lazy and there's a short window where the on-disk 3556314b6dd0SJohannes Thumshirn * copies might still contain the older version. 3557314b6dd0SJohannes Thumshirn */ 3558314b6dd0SJohannes Thumshirn bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO; 35591b9e619cSOmar Sandoval if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) 3560314b6dd0SJohannes Thumshirn bio->bi_opf |= REQ_FUA; 3561314b6dd0SJohannes Thumshirn 3562314b6dd0SJohannes Thumshirn btrfsic_submit_bio(bio); 3563a512bbf8SYan Zheng } 3564a512bbf8SYan Zheng return errors < i ? 0 : -1; 3565a512bbf8SYan Zheng } 3566a512bbf8SYan Zheng 3567387125fcSChris Mason /* 3568abbb3b8eSDavid Sterba * Wait for write completion of superblocks done by write_dev_supers, 3569abbb3b8eSDavid Sterba * @max_mirrors same for write and wait phases. 3570abbb3b8eSDavid Sterba * 3571314b6dd0SJohannes Thumshirn * Return number of errors when page is not found or not marked up to 3572abbb3b8eSDavid Sterba * date. 3573abbb3b8eSDavid Sterba */ 3574abbb3b8eSDavid Sterba static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) 3575abbb3b8eSDavid Sterba { 3576abbb3b8eSDavid Sterba int i; 3577abbb3b8eSDavid Sterba int errors = 0; 3578b6a535faSHoward McLauchlan bool primary_failed = false; 3579abbb3b8eSDavid Sterba u64 bytenr; 3580abbb3b8eSDavid Sterba 3581abbb3b8eSDavid Sterba if (max_mirrors == 0) 3582abbb3b8eSDavid Sterba max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3583abbb3b8eSDavid Sterba 3584abbb3b8eSDavid Sterba for (i = 0; i < max_mirrors; i++) { 3585314b6dd0SJohannes Thumshirn struct page *page; 3586314b6dd0SJohannes Thumshirn 3587abbb3b8eSDavid Sterba bytenr = btrfs_sb_offset(i); 3588abbb3b8eSDavid Sterba if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3589abbb3b8eSDavid Sterba device->commit_total_bytes) 3590abbb3b8eSDavid Sterba break; 3591abbb3b8eSDavid Sterba 3592314b6dd0SJohannes Thumshirn page = find_get_page(device->bdev->bd_inode->i_mapping, 3593314b6dd0SJohannes Thumshirn bytenr >> PAGE_SHIFT); 3594314b6dd0SJohannes Thumshirn if (!page) { 3595abbb3b8eSDavid Sterba errors++; 3596b6a535faSHoward McLauchlan if (i == 0) 3597b6a535faSHoward McLauchlan primary_failed = true; 3598abbb3b8eSDavid Sterba continue; 3599abbb3b8eSDavid Sterba } 3600314b6dd0SJohannes Thumshirn /* Page is submitted locked and unlocked once the IO completes */ 3601314b6dd0SJohannes Thumshirn wait_on_page_locked(page); 3602314b6dd0SJohannes Thumshirn if (PageError(page)) { 3603abbb3b8eSDavid Sterba errors++; 3604b6a535faSHoward McLauchlan if (i == 0) 3605b6a535faSHoward McLauchlan primary_failed = true; 3606b6a535faSHoward McLauchlan } 3607abbb3b8eSDavid Sterba 3608314b6dd0SJohannes Thumshirn /* Drop our reference */ 3609314b6dd0SJohannes Thumshirn put_page(page); 3610abbb3b8eSDavid Sterba 3611314b6dd0SJohannes Thumshirn /* Drop the reference from the writing run */ 3612314b6dd0SJohannes Thumshirn put_page(page); 3613abbb3b8eSDavid Sterba } 3614abbb3b8eSDavid Sterba 3615b6a535faSHoward McLauchlan /* log error, force error return */ 3616b6a535faSHoward McLauchlan if (primary_failed) { 3617b6a535faSHoward McLauchlan btrfs_err(device->fs_info, "error writing primary super block to device %llu", 3618b6a535faSHoward McLauchlan device->devid); 3619b6a535faSHoward McLauchlan return -1; 3620b6a535faSHoward McLauchlan } 3621b6a535faSHoward McLauchlan 3622abbb3b8eSDavid Sterba return errors < i ? 0 : -1; 3623abbb3b8eSDavid Sterba } 3624abbb3b8eSDavid Sterba 3625abbb3b8eSDavid Sterba /* 3626387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 3627387125fcSChris Mason * for the barrier when it is done 3628387125fcSChris Mason */ 36294246a0b6SChristoph Hellwig static void btrfs_end_empty_barrier(struct bio *bio) 3630387125fcSChris Mason { 3631387125fcSChris Mason complete(bio->bi_private); 3632387125fcSChris Mason } 3633387125fcSChris Mason 3634387125fcSChris Mason /* 36354fc6441aSAnand Jain * Submit a flush request to the device if it supports it. Error handling is 36364fc6441aSAnand Jain * done in the waiting counterpart. 3637387125fcSChris Mason */ 36384fc6441aSAnand Jain static void write_dev_flush(struct btrfs_device *device) 3639387125fcSChris Mason { 3640c2a9c7abSAnand Jain struct request_queue *q = bdev_get_queue(device->bdev); 3641e0ae9994SDavid Sterba struct bio *bio = device->flush_bio; 3642387125fcSChris Mason 3643c2a9c7abSAnand Jain if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) 36444fc6441aSAnand Jain return; 3645387125fcSChris Mason 3646e0ae9994SDavid Sterba bio_reset(bio); 3647387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 364874d46992SChristoph Hellwig bio_set_dev(bio, device->bdev); 36498d910125SJan Kara bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH; 3650387125fcSChris Mason init_completion(&device->flush_wait); 3651387125fcSChris Mason bio->bi_private = &device->flush_wait; 3652387125fcSChris Mason 365343a01111SLu Fengqi btrfsic_submit_bio(bio); 36541c3063b6SAnand Jain set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 36554fc6441aSAnand Jain } 3656387125fcSChris Mason 36574fc6441aSAnand Jain /* 36584fc6441aSAnand Jain * If the flush bio has been submitted by write_dev_flush, wait for it. 36594fc6441aSAnand Jain */ 36608c27cb35SLinus Torvalds static blk_status_t wait_dev_flush(struct btrfs_device *device) 36614fc6441aSAnand Jain { 36624fc6441aSAnand Jain struct bio *bio = device->flush_bio; 36634fc6441aSAnand Jain 36641c3063b6SAnand Jain if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) 366558efbc9fSOmar Sandoval return BLK_STS_OK; 36664fc6441aSAnand Jain 36671c3063b6SAnand Jain clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 36682980d574SDavid Sterba wait_for_completion_io(&device->flush_wait); 36694fc6441aSAnand Jain 36708c27cb35SLinus Torvalds return bio->bi_status; 3671387125fcSChris Mason } 3672387125fcSChris Mason 3673d10b82feSQu Wenruo static int check_barrier_error(struct btrfs_fs_info *fs_info) 3674401b41e5SAnand Jain { 36756528b99dSAnand Jain if (!btrfs_check_rw_degradable(fs_info, NULL)) 3676401b41e5SAnand Jain return -EIO; 3677387125fcSChris Mason return 0; 3678387125fcSChris Mason } 3679387125fcSChris Mason 3680387125fcSChris Mason /* 3681387125fcSChris Mason * send an empty flush down to each device in parallel, 3682387125fcSChris Mason * then wait for them 3683387125fcSChris Mason */ 3684387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 3685387125fcSChris Mason { 3686387125fcSChris Mason struct list_head *head; 3687387125fcSChris Mason struct btrfs_device *dev; 36885af3e8ccSStefan Behrens int errors_wait = 0; 36894e4cbee9SChristoph Hellwig blk_status_t ret; 3690387125fcSChris Mason 36911538e6c5SDavid Sterba lockdep_assert_held(&info->fs_devices->device_list_mutex); 3692387125fcSChris Mason /* send down all the barriers */ 3693387125fcSChris Mason head = &info->fs_devices->devices; 36941538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3695e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3696f88ba6a2SHidetoshi Seto continue; 3697cea7c8bfSAnand Jain if (!dev->bdev) 3698387125fcSChris Mason continue; 3699e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3700ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3701387125fcSChris Mason continue; 3702387125fcSChris Mason 37034fc6441aSAnand Jain write_dev_flush(dev); 370458efbc9fSOmar Sandoval dev->last_flush_error = BLK_STS_OK; 3705387125fcSChris Mason } 3706387125fcSChris Mason 3707387125fcSChris Mason /* wait for all the barriers */ 37081538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3709e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3710f88ba6a2SHidetoshi Seto continue; 3711387125fcSChris Mason if (!dev->bdev) { 37125af3e8ccSStefan Behrens errors_wait++; 3713387125fcSChris Mason continue; 3714387125fcSChris Mason } 3715e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3716ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3717387125fcSChris Mason continue; 3718387125fcSChris Mason 37194fc6441aSAnand Jain ret = wait_dev_flush(dev); 3720401b41e5SAnand Jain if (ret) { 3721401b41e5SAnand Jain dev->last_flush_error = ret; 372266b4993eSDavid Sterba btrfs_dev_stat_inc_and_print(dev, 372366b4993eSDavid Sterba BTRFS_DEV_STAT_FLUSH_ERRS); 37245af3e8ccSStefan Behrens errors_wait++; 3725387125fcSChris Mason } 3726401b41e5SAnand Jain } 3727401b41e5SAnand Jain 3728cea7c8bfSAnand Jain if (errors_wait) { 3729401b41e5SAnand Jain /* 3730401b41e5SAnand Jain * At some point we need the status of all disks 3731401b41e5SAnand Jain * to arrive at the volume status. So error checking 3732401b41e5SAnand Jain * is being pushed to a separate loop. 3733401b41e5SAnand Jain */ 3734d10b82feSQu Wenruo return check_barrier_error(info); 3735401b41e5SAnand Jain } 3736387125fcSChris Mason return 0; 3737387125fcSChris Mason } 3738387125fcSChris Mason 3739943c6e99SZhao Lei int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 3740943c6e99SZhao Lei { 37418789f4feSZhao Lei int raid_type; 37428789f4feSZhao Lei int min_tolerated = INT_MAX; 3743943c6e99SZhao Lei 37448789f4feSZhao Lei if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 37458789f4feSZhao Lei (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 37468c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 37478789f4feSZhao Lei btrfs_raid_array[BTRFS_RAID_SINGLE]. 37488789f4feSZhao Lei tolerated_failures); 3749943c6e99SZhao Lei 37508789f4feSZhao Lei for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 37518789f4feSZhao Lei if (raid_type == BTRFS_RAID_SINGLE) 37528789f4feSZhao Lei continue; 375341a6e891SAnand Jain if (!(flags & btrfs_raid_array[raid_type].bg_flag)) 37548789f4feSZhao Lei continue; 37558c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 37568789f4feSZhao Lei btrfs_raid_array[raid_type]. 37578789f4feSZhao Lei tolerated_failures); 37588789f4feSZhao Lei } 3759943c6e99SZhao Lei 37608789f4feSZhao Lei if (min_tolerated == INT_MAX) { 3761ab8d0fc4SJeff Mahoney pr_warn("BTRFS: unknown raid flag: %llu", flags); 37628789f4feSZhao Lei min_tolerated = 0; 37638789f4feSZhao Lei } 37648789f4feSZhao Lei 37658789f4feSZhao Lei return min_tolerated; 3766943c6e99SZhao Lei } 3767943c6e99SZhao Lei 3768eece6a9cSDavid Sterba int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) 3769f2984462SChris Mason { 3770e5e9a520SChris Mason struct list_head *head; 3771f2984462SChris Mason struct btrfs_device *dev; 3772a061fc8dSChris Mason struct btrfs_super_block *sb; 3773f2984462SChris Mason struct btrfs_dev_item *dev_item; 3774f2984462SChris Mason int ret; 3775f2984462SChris Mason int do_barriers; 3776a236aed1SChris Mason int max_errors; 3777a236aed1SChris Mason int total_errors = 0; 3778a061fc8dSChris Mason u64 flags; 3779f2984462SChris Mason 37800b246afaSJeff Mahoney do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); 3781fed3b381SLiu Bo 3782fed3b381SLiu Bo /* 3783fed3b381SLiu Bo * max_mirrors == 0 indicates we're from commit_transaction, 3784fed3b381SLiu Bo * not from fsync where the tree roots in fs_info have not 3785fed3b381SLiu Bo * been consistent on disk. 3786fed3b381SLiu Bo */ 3787fed3b381SLiu Bo if (max_mirrors == 0) 37880b246afaSJeff Mahoney backup_super_roots(fs_info); 3789f2984462SChris Mason 37900b246afaSJeff Mahoney sb = fs_info->super_for_commit; 3791a061fc8dSChris Mason dev_item = &sb->dev_item; 3792e5e9a520SChris Mason 37930b246afaSJeff Mahoney mutex_lock(&fs_info->fs_devices->device_list_mutex); 37940b246afaSJeff Mahoney head = &fs_info->fs_devices->devices; 37950b246afaSJeff Mahoney max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; 3796387125fcSChris Mason 37975af3e8ccSStefan Behrens if (do_barriers) { 37980b246afaSJeff Mahoney ret = barrier_all_devices(fs_info); 37995af3e8ccSStefan Behrens if (ret) { 38005af3e8ccSStefan Behrens mutex_unlock( 38010b246afaSJeff Mahoney &fs_info->fs_devices->device_list_mutex); 38020b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 38035af3e8ccSStefan Behrens "errors while submitting device barriers."); 38045af3e8ccSStefan Behrens return ret; 38055af3e8ccSStefan Behrens } 38065af3e8ccSStefan Behrens } 3807387125fcSChris Mason 38081538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3809dfe25020SChris Mason if (!dev->bdev) { 3810dfe25020SChris Mason total_errors++; 3811dfe25020SChris Mason continue; 3812dfe25020SChris Mason } 3813e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3814ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3815dfe25020SChris Mason continue; 3816dfe25020SChris Mason 38172b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 3818a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 3819a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 38207df69d3eSMiao Xie btrfs_set_stack_device_total_bytes(dev_item, 3821935e5cc9SMiao Xie dev->commit_total_bytes); 3822ce7213c7SMiao Xie btrfs_set_stack_device_bytes_used(dev_item, 3823ce7213c7SMiao Xie dev->commit_bytes_used); 3824a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 3825a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 3826a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 3827a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 38287239ff4bSNikolay Borisov memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, 38297239ff4bSNikolay Borisov BTRFS_FSID_SIZE); 3830a512bbf8SYan Zheng 3831a061fc8dSChris Mason flags = btrfs_super_flags(sb); 3832a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 3833f2984462SChris Mason 383475cb857dSQu Wenruo ret = btrfs_validate_write_super(fs_info, sb); 383575cb857dSQu Wenruo if (ret < 0) { 383675cb857dSQu Wenruo mutex_unlock(&fs_info->fs_devices->device_list_mutex); 383775cb857dSQu Wenruo btrfs_handle_fs_error(fs_info, -EUCLEAN, 383875cb857dSQu Wenruo "unexpected superblock corruption detected"); 383975cb857dSQu Wenruo return -EUCLEAN; 384075cb857dSQu Wenruo } 384175cb857dSQu Wenruo 3842abbb3b8eSDavid Sterba ret = write_dev_supers(dev, sb, max_mirrors); 3843a236aed1SChris Mason if (ret) 3844a236aed1SChris Mason total_errors++; 3845f2984462SChris Mason } 3846a236aed1SChris Mason if (total_errors > max_errors) { 38470b246afaSJeff Mahoney btrfs_err(fs_info, "%d errors while writing supers", 3848d397712bSChris Mason total_errors); 38490b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 385079787eaaSJeff Mahoney 38519d565ba4SStefan Behrens /* FUA is masked off if unsupported and can't be the reason */ 38520b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 38530b246afaSJeff Mahoney "%d errors while writing supers", 38540b246afaSJeff Mahoney total_errors); 38559d565ba4SStefan Behrens return -EIO; 3856a236aed1SChris Mason } 3857f2984462SChris Mason 3858a512bbf8SYan Zheng total_errors = 0; 38591538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3860dfe25020SChris Mason if (!dev->bdev) 3861dfe25020SChris Mason continue; 3862e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3863ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3864dfe25020SChris Mason continue; 3865dfe25020SChris Mason 3866abbb3b8eSDavid Sterba ret = wait_dev_supers(dev, max_mirrors); 3867a512bbf8SYan Zheng if (ret) 38681259ab75SChris Mason total_errors++; 3869f2984462SChris Mason } 38700b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 3871a236aed1SChris Mason if (total_errors > max_errors) { 38720b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 38730b246afaSJeff Mahoney "%d errors while writing supers", 38740b246afaSJeff Mahoney total_errors); 387579787eaaSJeff Mahoney return -EIO; 3876a236aed1SChris Mason } 3877f2984462SChris Mason return 0; 3878f2984462SChris Mason } 3879f2984462SChris Mason 3880cb517eabSMiao Xie /* Drop a fs root from the radix tree and free it. */ 3881cb517eabSMiao Xie void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 3882cb517eabSMiao Xie struct btrfs_root *root) 38832619ba1fSChris Mason { 38844785e24fSJosef Bacik bool drop_ref = false; 38854785e24fSJosef Bacik 38864df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 38872619ba1fSChris Mason radix_tree_delete(&fs_info->fs_roots_radix, 38882619ba1fSChris Mason (unsigned long)root->root_key.objectid); 3889af01d2e5SJosef Bacik if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state)) 38904785e24fSJosef Bacik drop_ref = true; 38914df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 389276dda93cSYan, Zheng 38931c1ea4f7SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 3894ef67963dSJosef Bacik ASSERT(root->log_root == NULL); 38951c1ea4f7SLiu Bo if (root->reloc_root) { 389600246528SJosef Bacik btrfs_put_root(root->reloc_root); 38971c1ea4f7SLiu Bo root->reloc_root = NULL; 38981c1ea4f7SLiu Bo } 38991c1ea4f7SLiu Bo } 39003321719eSLiu Bo 3901faa2dbf0SJosef Bacik if (root->free_ino_pinned) 3902581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_pinned); 3903faa2dbf0SJosef Bacik if (root->free_ino_ctl) 3904581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_ctl); 39050e996e7fSJosef Bacik if (root->ino_cache_inode) { 39060e996e7fSJosef Bacik iput(root->ino_cache_inode); 39070e996e7fSJosef Bacik root->ino_cache_inode = NULL; 39080e996e7fSJosef Bacik } 39094785e24fSJosef Bacik if (drop_ref) 391000246528SJosef Bacik btrfs_put_root(root); 39112619ba1fSChris Mason } 39122619ba1fSChris Mason 3913c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 3914c146afadSYan Zheng { 3915c146afadSYan Zheng u64 root_objectid = 0; 3916c146afadSYan Zheng struct btrfs_root *gang[8]; 391765d33fd7SQu Wenruo int i = 0; 391865d33fd7SQu Wenruo int err = 0; 391965d33fd7SQu Wenruo unsigned int ret = 0; 3920c146afadSYan Zheng 3921c146afadSYan Zheng while (1) { 3922efc34534SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 3923c146afadSYan Zheng ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 3924c146afadSYan Zheng (void **)gang, root_objectid, 3925c146afadSYan Zheng ARRAY_SIZE(gang)); 392665d33fd7SQu Wenruo if (!ret) { 3927efc34534SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 3928c146afadSYan Zheng break; 392965d33fd7SQu Wenruo } 39305d4f98a2SYan Zheng root_objectid = gang[ret - 1]->root_key.objectid + 1; 393166b4ffd1SJosef Bacik 393265d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 393365d33fd7SQu Wenruo /* Avoid to grab roots in dead_roots */ 393465d33fd7SQu Wenruo if (btrfs_root_refs(&gang[i]->root_item) == 0) { 393565d33fd7SQu Wenruo gang[i] = NULL; 393665d33fd7SQu Wenruo continue; 393765d33fd7SQu Wenruo } 393865d33fd7SQu Wenruo /* grab all the search result for later use */ 393900246528SJosef Bacik gang[i] = btrfs_grab_root(gang[i]); 394065d33fd7SQu Wenruo } 3941efc34534SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 394265d33fd7SQu Wenruo 394365d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 394465d33fd7SQu Wenruo if (!gang[i]) 394565d33fd7SQu Wenruo continue; 3946c146afadSYan Zheng root_objectid = gang[i]->root_key.objectid; 394766b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(gang[i]); 394866b4ffd1SJosef Bacik if (err) 394965d33fd7SQu Wenruo break; 395000246528SJosef Bacik btrfs_put_root(gang[i]); 3951c146afadSYan Zheng } 3952c146afadSYan Zheng root_objectid++; 3953c146afadSYan Zheng } 395465d33fd7SQu Wenruo 395565d33fd7SQu Wenruo /* release the uncleaned roots due to error */ 395665d33fd7SQu Wenruo for (; i < ret; i++) { 395765d33fd7SQu Wenruo if (gang[i]) 395800246528SJosef Bacik btrfs_put_root(gang[i]); 395965d33fd7SQu Wenruo } 396065d33fd7SQu Wenruo return err; 3961c146afadSYan Zheng } 3962c146afadSYan Zheng 39636bccf3abSJeff Mahoney int btrfs_commit_super(struct btrfs_fs_info *fs_info) 3964c146afadSYan Zheng { 39656bccf3abSJeff Mahoney struct btrfs_root *root = fs_info->tree_root; 3966c146afadSYan Zheng struct btrfs_trans_handle *trans; 3967c146afadSYan Zheng 39680b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 39692ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 39700b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 39710b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 3972c71bf099SYan, Zheng 3973c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 39740b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 39750b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 3976c71bf099SYan, Zheng 39777a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 39783612b495STsutomu Itoh if (IS_ERR(trans)) 39793612b495STsutomu Itoh return PTR_ERR(trans); 39803a45bb20SJeff Mahoney return btrfs_commit_transaction(trans); 3981c146afadSYan Zheng } 3982c146afadSYan Zheng 3983b105e927SDavid Sterba void __cold close_ctree(struct btrfs_fs_info *fs_info) 3984eb60ceacSChris Mason { 3985c146afadSYan Zheng int ret; 3986e089f05cSChris Mason 3987afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); 3988d6fd0ae2SOmar Sandoval /* 3989d6fd0ae2SOmar Sandoval * We don't want the cleaner to start new transactions, add more delayed 3990d6fd0ae2SOmar Sandoval * iputs, etc. while we're closing. We can't use kthread_stop() yet 3991d6fd0ae2SOmar Sandoval * because that frees the task_struct, and the transaction kthread might 3992d6fd0ae2SOmar Sandoval * still try to wake up the cleaner. 3993d6fd0ae2SOmar Sandoval */ 3994d6fd0ae2SOmar Sandoval kthread_park(fs_info->cleaner_kthread); 3995a2135011SChris Mason 39967343dd61SJustin Maggard /* wait for the qgroup rescan worker to stop */ 3997d06f23d6SJeff Mahoney btrfs_qgroup_wait_for_completion(fs_info, false); 39987343dd61SJustin Maggard 3999803b2f54SStefan Behrens /* wait for the uuid_scan task to finish */ 4000803b2f54SStefan Behrens down(&fs_info->uuid_tree_rescan_sem); 4001803b2f54SStefan Behrens /* avoid complains from lockdep et al., set sem back to initial state */ 4002803b2f54SStefan Behrens up(&fs_info->uuid_tree_rescan_sem); 4003803b2f54SStefan Behrens 4004837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 4005aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 4006837d5b6eSIlya Dryomov 40078dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 40088dabb742SStefan Behrens 4009aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 40104cb5300bSChris Mason 40114cb5300bSChris Mason /* wait for any defraggers to finish */ 40124cb5300bSChris Mason wait_event(fs_info->transaction_wait, 40134cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 40144cb5300bSChris Mason 40154cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 401626176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 40174cb5300bSChris Mason 401821c7e756SMiao Xie cancel_work_sync(&fs_info->async_reclaim_work); 401921c7e756SMiao Xie 4020b0643e59SDennis Zhou /* Cancel or finish ongoing discard work */ 4021b0643e59SDennis Zhou btrfs_discard_cleanup(fs_info); 4022b0643e59SDennis Zhou 4023bc98a42cSDavid Howells if (!sb_rdonly(fs_info->sb)) { 4024e44163e1SJeff Mahoney /* 4025d6fd0ae2SOmar Sandoval * The cleaner kthread is stopped, so do one final pass over 4026d6fd0ae2SOmar Sandoval * unused block groups. 4027e44163e1SJeff Mahoney */ 40280b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 4029e44163e1SJeff Mahoney 4030f0cc2cd7SFilipe Manana /* 4031f0cc2cd7SFilipe Manana * There might be existing delayed inode workers still running 4032f0cc2cd7SFilipe Manana * and holding an empty delayed inode item. We must wait for 4033f0cc2cd7SFilipe Manana * them to complete first because they can create a transaction. 4034f0cc2cd7SFilipe Manana * This happens when someone calls btrfs_balance_delayed_items() 4035f0cc2cd7SFilipe Manana * and then a transaction commit runs the same delayed nodes 4036f0cc2cd7SFilipe Manana * before any delayed worker has done something with the nodes. 4037f0cc2cd7SFilipe Manana * We must wait for any worker here and not at transaction 4038f0cc2cd7SFilipe Manana * commit time since that could cause a deadlock. 4039f0cc2cd7SFilipe Manana * This is a very rare case. 4040f0cc2cd7SFilipe Manana */ 4041f0cc2cd7SFilipe Manana btrfs_flush_workqueue(fs_info->delayed_workers); 4042f0cc2cd7SFilipe Manana 40436bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 4044d397712bSChris Mason if (ret) 404504892340SEric Sandeen btrfs_err(fs_info, "commit super ret %d", ret); 4046c146afadSYan Zheng } 4047ed2ff2cbSChris Mason 4048af722733SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) || 4049af722733SLiu Bo test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) 40502ff7e61eSJeff Mahoney btrfs_error_commit_super(fs_info); 4051acce952bSliubo 4052e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 4053e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 40548929ecfaSYan, Zheng 4055e187831eSJosef Bacik ASSERT(list_empty(&fs_info->delayed_iputs)); 4056afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); 4057f25784b3SYan Zheng 405804892340SEric Sandeen btrfs_free_qgroup_config(fs_info); 4059fe816d0fSNikolay Borisov ASSERT(list_empty(&fs_info->delalloc_roots)); 4060bcef60f2SArne Jansen 4061963d678bSMiao Xie if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 406204892340SEric Sandeen btrfs_info(fs_info, "at unmount delalloc count %lld", 4063963d678bSMiao Xie percpu_counter_sum(&fs_info->delalloc_bytes)); 4064b0c68f8bSChris Mason } 406531153d81SYan Zheng 40664297ff84SJosef Bacik if (percpu_counter_sum(&fs_info->dio_bytes)) 40674297ff84SJosef Bacik btrfs_info(fs_info, "at unmount dio bytes count %lld", 40684297ff84SJosef Bacik percpu_counter_sum(&fs_info->dio_bytes)); 40694297ff84SJosef Bacik 40706618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 4071b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 40725ac1d209SJeff Mahoney 40731a4319ccSLiu Bo btrfs_put_block_group_cache(fs_info); 40741a4319ccSLiu Bo 4075de348ee0SWang Shilong /* 4076de348ee0SWang Shilong * we must make sure there is not any read request to 4077de348ee0SWang Shilong * submit after we stopping all workers. 4078de348ee0SWang Shilong */ 4079de348ee0SWang Shilong invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 408096192499SJosef Bacik btrfs_stop_all_workers(fs_info); 408196192499SJosef Bacik 4082afcdd129SJosef Bacik clear_bit(BTRFS_FS_OPEN, &fs_info->flags); 40834273eaffSAnand Jain free_root_pointers(fs_info, true); 40848c38938cSJosef Bacik btrfs_free_fs_roots(fs_info); 40859ad6b7bcSChris Mason 40864e19443dSJosef Bacik /* 40874e19443dSJosef Bacik * We must free the block groups after dropping the fs_roots as we could 40884e19443dSJosef Bacik * have had an IO error and have left over tree log blocks that aren't 40894e19443dSJosef Bacik * cleaned up until the fs roots are freed. This makes the block group 40904e19443dSJosef Bacik * accounting appear to be wrong because there's pending reserved bytes, 40914e19443dSJosef Bacik * so make sure we do the block group cleanup afterwards. 40924e19443dSJosef Bacik */ 40934e19443dSJosef Bacik btrfs_free_block_groups(fs_info); 40944e19443dSJosef Bacik 409513e6c37bSJosef Bacik iput(fs_info->btree_inode); 4096d6bfde87SChris Mason 409721adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 40980b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) 40992ff7e61eSJeff Mahoney btrfsic_unmount(fs_info->fs_devices); 410021adbd5cSStefan Behrens #endif 410121adbd5cSStefan Behrens 41020b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 410368c94e55SNikolay Borisov btrfs_close_devices(fs_info->fs_devices); 4104eb60ceacSChris Mason } 4105eb60ceacSChris Mason 4106b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 4107b9fab919SChris Mason int atomic) 4108ccd467d6SChris Mason { 41091259ab75SChris Mason int ret; 4110727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 41111259ab75SChris Mason 41120b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 41131259ab75SChris Mason if (!ret) 41141259ab75SChris Mason return ret; 41151259ab75SChris Mason 41161259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 4117b9fab919SChris Mason parent_transid, atomic); 4118b9fab919SChris Mason if (ret == -EAGAIN) 4119b9fab919SChris Mason return ret; 41201259ab75SChris Mason return !ret; 41215f39d397SChris Mason } 41226702ed49SChris Mason 41235f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 41245f39d397SChris Mason { 41250b246afaSJeff Mahoney struct btrfs_fs_info *fs_info; 412606ea65a3SJosef Bacik struct btrfs_root *root; 41275f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 4128b9473439SChris Mason int was_dirty; 4129b4ce94deSChris Mason 413006ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 413106ea65a3SJosef Bacik /* 413206ea65a3SJosef Bacik * This is a fast path so only do this check if we have sanity tests 413352042d8eSAndrea Gelmini * enabled. Normal people shouldn't be using unmapped buffers as dirty 413406ea65a3SJosef Bacik * outside of the sanity tests. 413506ea65a3SJosef Bacik */ 4136b0132a3bSNikolay Borisov if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) 413706ea65a3SJosef Bacik return; 413806ea65a3SJosef Bacik #endif 413906ea65a3SJosef Bacik root = BTRFS_I(buf->pages[0]->mapping->host)->root; 41400b246afaSJeff Mahoney fs_info = root->fs_info; 4141b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 41420b246afaSJeff Mahoney if (transid != fs_info->generation) 41435d163e0eSJeff Mahoney WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", 41440b246afaSJeff Mahoney buf->start, transid, fs_info->generation); 41450b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 4146e2d84521SMiao Xie if (!was_dirty) 4147104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 4148e2d84521SMiao Xie buf->len, 41490b246afaSJeff Mahoney fs_info->dirty_metadata_batch); 41501f21ef0aSFilipe Manana #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 415169fc6cbbSQu Wenruo /* 415269fc6cbbSQu Wenruo * Since btrfs_mark_buffer_dirty() can be called with item pointer set 415369fc6cbbSQu Wenruo * but item data not updated. 415469fc6cbbSQu Wenruo * So here we should only check item pointers, not item data. 415569fc6cbbSQu Wenruo */ 415669fc6cbbSQu Wenruo if (btrfs_header_level(buf) == 0 && 4157cfdaad5eSDavid Sterba btrfs_check_leaf_relaxed(buf)) { 4158a4f78750SDavid Sterba btrfs_print_leaf(buf); 41591f21ef0aSFilipe Manana ASSERT(0); 41601f21ef0aSFilipe Manana } 41611f21ef0aSFilipe Manana #endif 4162eb60ceacSChris Mason } 4163eb60ceacSChris Mason 41642ff7e61eSJeff Mahoney static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, 4165b53d3f5dSLiu Bo int flush_delayed) 416635b7e476SChris Mason { 4167188de649SChris Mason /* 4168188de649SChris Mason * looks as though older kernels can get into trouble with 4169188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 4170188de649SChris Mason */ 4171e2d84521SMiao Xie int ret; 4172d6bfde87SChris Mason 41736933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 4174d6bfde87SChris Mason return; 4175d6bfde87SChris Mason 4176b53d3f5dSLiu Bo if (flush_delayed) 41772ff7e61eSJeff Mahoney btrfs_balance_delayed_items(fs_info); 417816cdcec7SMiao Xie 4179d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 4180d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 4181d814a491SEthan Lien fs_info->dirty_metadata_batch); 4182e2d84521SMiao Xie if (ret > 0) { 41830b246afaSJeff Mahoney balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); 418416cdcec7SMiao Xie } 418516cdcec7SMiao Xie } 418616cdcec7SMiao Xie 41872ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) 418816cdcec7SMiao Xie { 41892ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 1); 419035b7e476SChris Mason } 4191b53d3f5dSLiu Bo 41922ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) 4193b53d3f5dSLiu Bo { 41942ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 0); 4195d6bfde87SChris Mason } 41966b80053dSChris Mason 4197581c1760SQu Wenruo int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level, 4198581c1760SQu Wenruo struct btrfs_key *first_key) 41996b80053dSChris Mason { 42005ab12d1fSDavid Sterba return btree_read_extent_buffer_pages(buf, parent_transid, 4201581c1760SQu Wenruo level, first_key); 42026b80053dSChris Mason } 42030da5468fSChris Mason 42042ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) 4205acce952bSliubo { 4206fe816d0fSNikolay Borisov /* cleanup FS via transaction */ 4207fe816d0fSNikolay Borisov btrfs_cleanup_transaction(fs_info); 4208fe816d0fSNikolay Borisov 42090b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 42102ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 42110b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 4212acce952bSliubo 42130b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 42140b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4215acce952bSliubo } 4216acce952bSliubo 4217ef67963dSJosef Bacik static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info) 4218ef67963dSJosef Bacik { 4219ef67963dSJosef Bacik struct btrfs_root *gang[8]; 4220ef67963dSJosef Bacik u64 root_objectid = 0; 4221ef67963dSJosef Bacik int ret; 4222ef67963dSJosef Bacik 4223ef67963dSJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 4224ef67963dSJosef Bacik while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4225ef67963dSJosef Bacik (void **)gang, root_objectid, 4226ef67963dSJosef Bacik ARRAY_SIZE(gang))) != 0) { 4227ef67963dSJosef Bacik int i; 4228ef67963dSJosef Bacik 4229ef67963dSJosef Bacik for (i = 0; i < ret; i++) 4230ef67963dSJosef Bacik gang[i] = btrfs_grab_root(gang[i]); 4231ef67963dSJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 4232ef67963dSJosef Bacik 4233ef67963dSJosef Bacik for (i = 0; i < ret; i++) { 4234ef67963dSJosef Bacik if (!gang[i]) 4235ef67963dSJosef Bacik continue; 4236ef67963dSJosef Bacik root_objectid = gang[i]->root_key.objectid; 4237ef67963dSJosef Bacik btrfs_free_log(NULL, gang[i]); 4238ef67963dSJosef Bacik btrfs_put_root(gang[i]); 4239ef67963dSJosef Bacik } 4240ef67963dSJosef Bacik root_objectid++; 4241ef67963dSJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 4242ef67963dSJosef Bacik } 4243ef67963dSJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 4244ef67963dSJosef Bacik btrfs_free_log_root_tree(NULL, fs_info); 4245ef67963dSJosef Bacik } 4246ef67963dSJosef Bacik 4247143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4248acce952bSliubo { 4249acce952bSliubo struct btrfs_ordered_extent *ordered; 4250acce952bSliubo 4251199c2a9cSMiao Xie spin_lock(&root->ordered_extent_lock); 4252779880efSJosef Bacik /* 4253779880efSJosef Bacik * This will just short circuit the ordered completion stuff which will 4254779880efSJosef Bacik * make sure the ordered extent gets properly cleaned up. 4255779880efSJosef Bacik */ 4256199c2a9cSMiao Xie list_for_each_entry(ordered, &root->ordered_extents, 4257779880efSJosef Bacik root_extent_list) 4258779880efSJosef Bacik set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4259199c2a9cSMiao Xie spin_unlock(&root->ordered_extent_lock); 4260199c2a9cSMiao Xie } 4261199c2a9cSMiao Xie 4262199c2a9cSMiao Xie static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4263199c2a9cSMiao Xie { 4264199c2a9cSMiao Xie struct btrfs_root *root; 4265199c2a9cSMiao Xie struct list_head splice; 4266199c2a9cSMiao Xie 4267199c2a9cSMiao Xie INIT_LIST_HEAD(&splice); 4268199c2a9cSMiao Xie 4269199c2a9cSMiao Xie spin_lock(&fs_info->ordered_root_lock); 4270199c2a9cSMiao Xie list_splice_init(&fs_info->ordered_roots, &splice); 4271199c2a9cSMiao Xie while (!list_empty(&splice)) { 4272199c2a9cSMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4273199c2a9cSMiao Xie ordered_root); 42741de2cfdeSJosef Bacik list_move_tail(&root->ordered_root, 42751de2cfdeSJosef Bacik &fs_info->ordered_roots); 4276199c2a9cSMiao Xie 42772a85d9caSLiu Bo spin_unlock(&fs_info->ordered_root_lock); 4278199c2a9cSMiao Xie btrfs_destroy_ordered_extents(root); 4279199c2a9cSMiao Xie 42802a85d9caSLiu Bo cond_resched(); 42812a85d9caSLiu Bo spin_lock(&fs_info->ordered_root_lock); 4282199c2a9cSMiao Xie } 4283199c2a9cSMiao Xie spin_unlock(&fs_info->ordered_root_lock); 428474d5d229SJosef Bacik 428574d5d229SJosef Bacik /* 428674d5d229SJosef Bacik * We need this here because if we've been flipped read-only we won't 428774d5d229SJosef Bacik * get sync() from the umount, so we need to make sure any ordered 428874d5d229SJosef Bacik * extents that haven't had their dirty pages IO start writeout yet 428974d5d229SJosef Bacik * actually get run and error out properly. 429074d5d229SJosef Bacik */ 429174d5d229SJosef Bacik btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 4292acce952bSliubo } 4293acce952bSliubo 429435a3621bSStefan Behrens static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 42952ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4296acce952bSliubo { 4297acce952bSliubo struct rb_node *node; 4298acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 4299acce952bSliubo struct btrfs_delayed_ref_node *ref; 4300acce952bSliubo int ret = 0; 4301acce952bSliubo 4302acce952bSliubo delayed_refs = &trans->delayed_refs; 4303acce952bSliubo 4304acce952bSliubo spin_lock(&delayed_refs->lock); 4305d7df2c79SJosef Bacik if (atomic_read(&delayed_refs->num_entries) == 0) { 4306cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 4307b79ce3ddSDavid Sterba btrfs_debug(fs_info, "delayed_refs has NO entry"); 4308acce952bSliubo return ret; 4309acce952bSliubo } 4310acce952bSliubo 43115c9d028bSLiu Bo while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { 4312d7df2c79SJosef Bacik struct btrfs_delayed_ref_head *head; 43130e0adbcfSJosef Bacik struct rb_node *n; 4314e78417d1SJosef Bacik bool pin_bytes = false; 4315acce952bSliubo 4316d7df2c79SJosef Bacik head = rb_entry(node, struct btrfs_delayed_ref_head, 4317d7df2c79SJosef Bacik href_node); 43183069bd26SJosef Bacik if (btrfs_delayed_ref_lock(delayed_refs, head)) 4319b939d1abSJosef Bacik continue; 43203069bd26SJosef Bacik 4321d7df2c79SJosef Bacik spin_lock(&head->lock); 4322e3d03965SLiu Bo while ((n = rb_first_cached(&head->ref_tree)) != NULL) { 43230e0adbcfSJosef Bacik ref = rb_entry(n, struct btrfs_delayed_ref_node, 43240e0adbcfSJosef Bacik ref_node); 4325d7df2c79SJosef Bacik ref->in_tree = 0; 4326e3d03965SLiu Bo rb_erase_cached(&ref->ref_node, &head->ref_tree); 43270e0adbcfSJosef Bacik RB_CLEAR_NODE(&ref->ref_node); 43281d57ee94SWang Xiaoguang if (!list_empty(&ref->add_list)) 43291d57ee94SWang Xiaoguang list_del(&ref->add_list); 4330d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4331d7df2c79SJosef Bacik btrfs_put_delayed_ref(ref); 4332d7df2c79SJosef Bacik } 433354067ae9SJosef Bacik if (head->must_insert_reserved) 4334e78417d1SJosef Bacik pin_bytes = true; 433578a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 4336fa781ceaSJosef Bacik btrfs_delete_ref_head(delayed_refs, head); 4337d7df2c79SJosef Bacik spin_unlock(&head->lock); 4338acce952bSliubo spin_unlock(&delayed_refs->lock); 4339e78417d1SJosef Bacik mutex_unlock(&head->mutex); 4340acce952bSliubo 4341f603bb94SNikolay Borisov if (pin_bytes) { 4342f603bb94SNikolay Borisov struct btrfs_block_group *cache; 4343f603bb94SNikolay Borisov 4344f603bb94SNikolay Borisov cache = btrfs_lookup_block_group(fs_info, head->bytenr); 4345f603bb94SNikolay Borisov BUG_ON(!cache); 4346f603bb94SNikolay Borisov 4347f603bb94SNikolay Borisov spin_lock(&cache->space_info->lock); 4348f603bb94SNikolay Borisov spin_lock(&cache->lock); 4349f603bb94SNikolay Borisov cache->pinned += head->num_bytes; 4350f603bb94SNikolay Borisov btrfs_space_info_update_bytes_pinned(fs_info, 4351f603bb94SNikolay Borisov cache->space_info, head->num_bytes); 4352f603bb94SNikolay Borisov cache->reserved -= head->num_bytes; 4353f603bb94SNikolay Borisov cache->space_info->bytes_reserved -= head->num_bytes; 4354f603bb94SNikolay Borisov spin_unlock(&cache->lock); 4355f603bb94SNikolay Borisov spin_unlock(&cache->space_info->lock); 4356f603bb94SNikolay Borisov percpu_counter_add_batch( 4357f603bb94SNikolay Borisov &cache->space_info->total_bytes_pinned, 4358f603bb94SNikolay Borisov head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH); 4359f603bb94SNikolay Borisov 4360f603bb94SNikolay Borisov btrfs_put_block_group(cache); 4361f603bb94SNikolay Borisov 4362f603bb94SNikolay Borisov btrfs_error_unpin_extent_range(fs_info, head->bytenr, 4363f603bb94SNikolay Borisov head->bytenr + head->num_bytes - 1); 4364f603bb94SNikolay Borisov } 436531890da0SJosef Bacik btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); 4366d278850eSJosef Bacik btrfs_put_delayed_ref_head(head); 4367acce952bSliubo cond_resched(); 4368acce952bSliubo spin_lock(&delayed_refs->lock); 4369acce952bSliubo } 437081f7eb00SJeff Mahoney btrfs_qgroup_destroy_extent_records(trans); 4371acce952bSliubo 4372acce952bSliubo spin_unlock(&delayed_refs->lock); 4373acce952bSliubo 4374acce952bSliubo return ret; 4375acce952bSliubo } 4376acce952bSliubo 4377143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 4378acce952bSliubo { 4379acce952bSliubo struct btrfs_inode *btrfs_inode; 4380acce952bSliubo struct list_head splice; 4381acce952bSliubo 4382acce952bSliubo INIT_LIST_HEAD(&splice); 4383acce952bSliubo 4384eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4385eb73c1b7SMiao Xie list_splice_init(&root->delalloc_inodes, &splice); 4386acce952bSliubo 4387acce952bSliubo while (!list_empty(&splice)) { 4388fe816d0fSNikolay Borisov struct inode *inode = NULL; 4389eb73c1b7SMiao Xie btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 4390acce952bSliubo delalloc_inodes); 4391fe816d0fSNikolay Borisov __btrfs_del_delalloc_inode(root, btrfs_inode); 4392eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4393acce952bSliubo 4394fe816d0fSNikolay Borisov /* 4395fe816d0fSNikolay Borisov * Make sure we get a live inode and that it'll not disappear 4396fe816d0fSNikolay Borisov * meanwhile. 4397fe816d0fSNikolay Borisov */ 4398fe816d0fSNikolay Borisov inode = igrab(&btrfs_inode->vfs_inode); 4399fe816d0fSNikolay Borisov if (inode) { 4400fe816d0fSNikolay Borisov invalidate_inode_pages2(inode->i_mapping); 4401fe816d0fSNikolay Borisov iput(inode); 4402fe816d0fSNikolay Borisov } 4403eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4404acce952bSliubo } 4405eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4406eb73c1b7SMiao Xie } 4407eb73c1b7SMiao Xie 4408eb73c1b7SMiao Xie static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 4409eb73c1b7SMiao Xie { 4410eb73c1b7SMiao Xie struct btrfs_root *root; 4411eb73c1b7SMiao Xie struct list_head splice; 4412eb73c1b7SMiao Xie 4413eb73c1b7SMiao Xie INIT_LIST_HEAD(&splice); 4414eb73c1b7SMiao Xie 4415eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4416eb73c1b7SMiao Xie list_splice_init(&fs_info->delalloc_roots, &splice); 4417eb73c1b7SMiao Xie while (!list_empty(&splice)) { 4418eb73c1b7SMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4419eb73c1b7SMiao Xie delalloc_root); 442000246528SJosef Bacik root = btrfs_grab_root(root); 4421eb73c1b7SMiao Xie BUG_ON(!root); 4422eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4423eb73c1b7SMiao Xie 4424eb73c1b7SMiao Xie btrfs_destroy_delalloc_inodes(root); 442500246528SJosef Bacik btrfs_put_root(root); 4426eb73c1b7SMiao Xie 4427eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4428eb73c1b7SMiao Xie } 4429eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4430acce952bSliubo } 4431acce952bSliubo 44322ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 4433acce952bSliubo struct extent_io_tree *dirty_pages, 4434acce952bSliubo int mark) 4435acce952bSliubo { 4436acce952bSliubo int ret; 4437acce952bSliubo struct extent_buffer *eb; 4438acce952bSliubo u64 start = 0; 4439acce952bSliubo u64 end; 4440acce952bSliubo 4441acce952bSliubo while (1) { 4442acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 4443e6138876SJosef Bacik mark, NULL); 4444acce952bSliubo if (ret) 4445acce952bSliubo break; 4446acce952bSliubo 444791166212SDavid Sterba clear_extent_bits(dirty_pages, start, end, mark); 4448acce952bSliubo while (start <= end) { 44490b246afaSJeff Mahoney eb = find_extent_buffer(fs_info, start); 44500b246afaSJeff Mahoney start += fs_info->nodesize; 4451fd8b2b61SJosef Bacik if (!eb) 4452acce952bSliubo continue; 4453fd8b2b61SJosef Bacik wait_on_extent_buffer_writeback(eb); 4454acce952bSliubo 4455fd8b2b61SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 4456fd8b2b61SJosef Bacik &eb->bflags)) 4457fd8b2b61SJosef Bacik clear_extent_buffer_dirty(eb); 4458fd8b2b61SJosef Bacik free_extent_buffer_stale(eb); 4459acce952bSliubo } 4460acce952bSliubo } 4461acce952bSliubo 4462acce952bSliubo return ret; 4463acce952bSliubo } 4464acce952bSliubo 44652ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 4466fe119a6eSNikolay Borisov struct extent_io_tree *unpin) 4467acce952bSliubo { 4468acce952bSliubo u64 start; 4469acce952bSliubo u64 end; 4470acce952bSliubo int ret; 4471acce952bSliubo 4472acce952bSliubo while (1) { 44730e6ec385SFilipe Manana struct extent_state *cached_state = NULL; 44740e6ec385SFilipe Manana 4475fcd5e742SLu Fengqi /* 4476fcd5e742SLu Fengqi * The btrfs_finish_extent_commit() may get the same range as 4477fcd5e742SLu Fengqi * ours between find_first_extent_bit and clear_extent_dirty. 4478fcd5e742SLu Fengqi * Hence, hold the unused_bg_unpin_mutex to avoid double unpin 4479fcd5e742SLu Fengqi * the same extent range. 4480fcd5e742SLu Fengqi */ 4481fcd5e742SLu Fengqi mutex_lock(&fs_info->unused_bg_unpin_mutex); 4482acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 44830e6ec385SFilipe Manana EXTENT_DIRTY, &cached_state); 4484fcd5e742SLu Fengqi if (ret) { 4485fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4486acce952bSliubo break; 4487fcd5e742SLu Fengqi } 4488acce952bSliubo 44890e6ec385SFilipe Manana clear_extent_dirty(unpin, start, end, &cached_state); 44900e6ec385SFilipe Manana free_extent_state(cached_state); 44912ff7e61eSJeff Mahoney btrfs_error_unpin_extent_range(fs_info, start, end); 4492fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4493acce952bSliubo cond_resched(); 4494acce952bSliubo } 4495acce952bSliubo 4496acce952bSliubo return 0; 4497acce952bSliubo } 4498acce952bSliubo 449932da5386SDavid Sterba static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache) 4500c79a1751SLiu Bo { 4501c79a1751SLiu Bo struct inode *inode; 4502c79a1751SLiu Bo 4503c79a1751SLiu Bo inode = cache->io_ctl.inode; 4504c79a1751SLiu Bo if (inode) { 4505c79a1751SLiu Bo invalidate_inode_pages2(inode->i_mapping); 4506c79a1751SLiu Bo BTRFS_I(inode)->generation = 0; 4507c79a1751SLiu Bo cache->io_ctl.inode = NULL; 4508c79a1751SLiu Bo iput(inode); 4509c79a1751SLiu Bo } 4510c79a1751SLiu Bo btrfs_put_block_group(cache); 4511c79a1751SLiu Bo } 4512c79a1751SLiu Bo 4513c79a1751SLiu Bo void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, 45142ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4515c79a1751SLiu Bo { 451632da5386SDavid Sterba struct btrfs_block_group *cache; 4517c79a1751SLiu Bo 4518c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4519c79a1751SLiu Bo while (!list_empty(&cur_trans->dirty_bgs)) { 4520c79a1751SLiu Bo cache = list_first_entry(&cur_trans->dirty_bgs, 452132da5386SDavid Sterba struct btrfs_block_group, 4522c79a1751SLiu Bo dirty_list); 4523c79a1751SLiu Bo 4524c79a1751SLiu Bo if (!list_empty(&cache->io_list)) { 4525c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4526c79a1751SLiu Bo list_del_init(&cache->io_list); 4527c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 4528c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4529c79a1751SLiu Bo } 4530c79a1751SLiu Bo 4531c79a1751SLiu Bo list_del_init(&cache->dirty_list); 4532c79a1751SLiu Bo spin_lock(&cache->lock); 4533c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 4534c79a1751SLiu Bo spin_unlock(&cache->lock); 4535c79a1751SLiu Bo 4536c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4537c79a1751SLiu Bo btrfs_put_block_group(cache); 4538ba2c4d4eSJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 4539c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4540c79a1751SLiu Bo } 4541c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4542c79a1751SLiu Bo 454345ae2c18SNikolay Borisov /* 454445ae2c18SNikolay Borisov * Refer to the definition of io_bgs member for details why it's safe 454545ae2c18SNikolay Borisov * to use it without any locking 454645ae2c18SNikolay Borisov */ 4547c79a1751SLiu Bo while (!list_empty(&cur_trans->io_bgs)) { 4548c79a1751SLiu Bo cache = list_first_entry(&cur_trans->io_bgs, 454932da5386SDavid Sterba struct btrfs_block_group, 4550c79a1751SLiu Bo io_list); 4551c79a1751SLiu Bo 4552c79a1751SLiu Bo list_del_init(&cache->io_list); 4553c79a1751SLiu Bo spin_lock(&cache->lock); 4554c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 4555c79a1751SLiu Bo spin_unlock(&cache->lock); 4556c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 4557c79a1751SLiu Bo } 4558c79a1751SLiu Bo } 4559c79a1751SLiu Bo 456049b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 45612ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 456249b25e05SJeff Mahoney { 4563bbbf7243SNikolay Borisov struct btrfs_device *dev, *tmp; 4564bbbf7243SNikolay Borisov 45652ff7e61eSJeff Mahoney btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 4566c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->dirty_bgs)); 4567c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->io_bgs)); 4568c79a1751SLiu Bo 4569bbbf7243SNikolay Borisov list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list, 4570bbbf7243SNikolay Borisov post_commit_list) { 4571bbbf7243SNikolay Borisov list_del_init(&dev->post_commit_list); 4572bbbf7243SNikolay Borisov } 4573bbbf7243SNikolay Borisov 45742ff7e61eSJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, fs_info); 457549b25e05SJeff Mahoney 45764a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 45770b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 457849b25e05SJeff Mahoney 45794a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 45800b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 458149b25e05SJeff Mahoney 4582ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 458367cde344SMiao Xie 45842ff7e61eSJeff Mahoney btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, 458549b25e05SJeff Mahoney EXTENT_DIRTY); 4586fe119a6eSNikolay Borisov btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents); 458749b25e05SJeff Mahoney 45884a9d8bdeSMiao Xie cur_trans->state =TRANS_STATE_COMPLETED; 45894a9d8bdeSMiao Xie wake_up(&cur_trans->commit_wait); 459049b25e05SJeff Mahoney } 459149b25e05SJeff Mahoney 45922ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) 4593acce952bSliubo { 4594acce952bSliubo struct btrfs_transaction *t; 4595acce952bSliubo 45960b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 4597acce952bSliubo 45980b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 45990b246afaSJeff Mahoney while (!list_empty(&fs_info->trans_list)) { 46000b246afaSJeff Mahoney t = list_first_entry(&fs_info->trans_list, 4601724e2315SJosef Bacik struct btrfs_transaction, list); 4602724e2315SJosef Bacik if (t->state >= TRANS_STATE_COMMIT_START) { 46039b64f57dSElena Reshetova refcount_inc(&t->use_count); 46040b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 46052ff7e61eSJeff Mahoney btrfs_wait_for_commit(fs_info, t->transid); 4606724e2315SJosef Bacik btrfs_put_transaction(t); 46070b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4608724e2315SJosef Bacik continue; 4609724e2315SJosef Bacik } 46100b246afaSJeff Mahoney if (t == fs_info->running_transaction) { 4611724e2315SJosef Bacik t->state = TRANS_STATE_COMMIT_DOING; 46120b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4613724e2315SJosef Bacik /* 4614724e2315SJosef Bacik * We wait for 0 num_writers since we don't hold a trans 4615724e2315SJosef Bacik * handle open currently for this transaction. 4616724e2315SJosef Bacik */ 4617724e2315SJosef Bacik wait_event(t->writer_wait, 4618724e2315SJosef Bacik atomic_read(&t->num_writers) == 0); 4619724e2315SJosef Bacik } else { 46200b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4621724e2315SJosef Bacik } 46222ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(t, fs_info); 4623724e2315SJosef Bacik 46240b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 46250b246afaSJeff Mahoney if (t == fs_info->running_transaction) 46260b246afaSJeff Mahoney fs_info->running_transaction = NULL; 4627724e2315SJosef Bacik list_del_init(&t->list); 46280b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4629a4abeea4SJosef Bacik 4630724e2315SJosef Bacik btrfs_put_transaction(t); 46312ff7e61eSJeff Mahoney trace_btrfs_transaction_commit(fs_info->tree_root); 46320b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4633724e2315SJosef Bacik } 46340b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 46350b246afaSJeff Mahoney btrfs_destroy_all_ordered_extents(fs_info); 4636ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 4637ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 46380b246afaSJeff Mahoney btrfs_destroy_all_delalloc_inodes(fs_info); 4639ef67963dSJosef Bacik btrfs_drop_all_logs(fs_info); 46400b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 4641acce952bSliubo 4642acce952bSliubo return 0; 4643acce952bSliubo } 4644acce952bSliubo 4645e8c9f186SDavid Sterba static const struct extent_io_ops btree_extent_io_ops = { 46464d53dddbSDavid Sterba /* mandatory callbacks */ 46470b86a832SChris Mason .submit_bio_hook = btree_submit_bio_hook, 46484d53dddbSDavid Sterba .readpage_end_io_hook = btree_readpage_end_io_hook, 46490da5468fSChris Mason }; 4650