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> 10d397712bSChris Mason #include <linux/buffer_head.h> 11ce9adaa5SChris Mason #include <linux/workqueue.h> 12a74a4b97SChris Mason #include <linux/kthread.h> 135a0e3ad6STejun Heo #include <linux/slab.h> 14784b4e29SChris Mason #include <linux/migrate.h> 157a36ddecSDavid Sterba #include <linux/ratelimit.h> 166463fe58SStefan Behrens #include <linux/uuid.h> 17803b2f54SStefan Behrens #include <linux/semaphore.h> 18540adea3SMasami Hiramatsu #include <linux/error-injection.h> 199678c543SNikolay Borisov #include <linux/crc32c.h> 20b89f6d1fSFilipe Manana #include <linux/sched/mm.h> 217e75bf3fSDavid Sterba #include <asm/unaligned.h> 226d97c6e3SJohannes Thumshirn #include <crypto/hash.h> 23eb60ceacSChris Mason #include "ctree.h" 24eb60ceacSChris Mason #include "disk-io.h" 25e089f05cSChris Mason #include "transaction.h" 260f7d52f4SChris Mason #include "btrfs_inode.h" 270b86a832SChris Mason #include "volumes.h" 28db94535dSChris Mason #include "print-tree.h" 29925baeddSChris Mason #include "locking.h" 30e02119d5SChris Mason #include "tree-log.h" 31fa9c0d79SChris Mason #include "free-space-cache.h" 3270f6d82eSOmar Sandoval #include "free-space-tree.h" 33581bb050SLi Zefan #include "inode-map.h" 3421adbd5cSStefan Behrens #include "check-integrity.h" 35606686eeSJosef Bacik #include "rcu-string.h" 368dabb742SStefan Behrens #include "dev-replace.h" 3753b381b3SDavid Woodhouse #include "raid56.h" 385ac1d209SJeff Mahoney #include "sysfs.h" 39fcebe456SJosef Bacik #include "qgroup.h" 40ebb8765bSAnand Jain #include "compression.h" 41557ea5ddSQu Wenruo #include "tree-checker.h" 42fd708b81SJosef Bacik #include "ref-verify.h" 43aac0023cSJosef Bacik #include "block-group.h" 44b0643e59SDennis Zhou #include "discard.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 101d352ac68SChris Mason /* 102d352ac68SChris Mason * async submit bios are used to offload expensive checksumming 103d352ac68SChris Mason * onto the worker threads. They checksum file and metadata bios 104d352ac68SChris Mason * just before they are sent down the IO stack. 105d352ac68SChris Mason */ 10644b8bd7eSChris Mason struct async_submit_bio { 107c6100a4bSJosef Bacik void *private_data; 10844b8bd7eSChris Mason struct bio *bio; 109a758781dSDavid Sterba extent_submit_bio_start_t *submit_bio_start; 11044b8bd7eSChris Mason int mirror_num; 111eaf25d93SChris Mason /* 112eaf25d93SChris Mason * bio_offset is optional, can be used if the pages in the bio 113eaf25d93SChris Mason * can't tell us where in the file the bio should go 114eaf25d93SChris Mason */ 115eaf25d93SChris Mason u64 bio_offset; 1168b712842SChris Mason struct btrfs_work work; 1174e4cbee9SChristoph Hellwig blk_status_t status; 11844b8bd7eSChris Mason }; 11944b8bd7eSChris Mason 12085d4e461SChris Mason /* 12185d4e461SChris Mason * Lockdep class keys for extent_buffer->lock's in this root. For a given 12285d4e461SChris Mason * eb, the lockdep key is determined by the btrfs_root it belongs to and 12385d4e461SChris Mason * the level the eb occupies in the tree. 1244008c04aSChris Mason * 12585d4e461SChris Mason * Different roots are used for different purposes and may nest inside each 12685d4e461SChris Mason * other and they require separate keysets. As lockdep keys should be 12785d4e461SChris Mason * static, assign keysets according to the purpose of the root as indicated 1284fd786e6SMisono Tomohiro * by btrfs_root->root_key.objectid. This ensures that all special purpose 1294fd786e6SMisono Tomohiro * roots have separate keysets. 1304008c04aSChris Mason * 13185d4e461SChris Mason * Lock-nesting across peer nodes is always done with the immediate parent 13285d4e461SChris Mason * node locked thus preventing deadlock. As lockdep doesn't know this, use 13385d4e461SChris Mason * subclass to avoid triggering lockdep warning in such cases. 1344008c04aSChris Mason * 13585d4e461SChris Mason * The key is set by the readpage_end_io_hook after the buffer has passed 13685d4e461SChris Mason * csum validation but before the pages are unlocked. It is also set by 13785d4e461SChris Mason * btrfs_init_new_buffer on freshly allocated blocks. 13885d4e461SChris Mason * 13985d4e461SChris Mason * We also add a check to make sure the highest level of the tree is the 14085d4e461SChris Mason * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code 14185d4e461SChris Mason * needs update as well. 1424008c04aSChris Mason */ 1434008c04aSChris Mason #ifdef CONFIG_DEBUG_LOCK_ALLOC 1444008c04aSChris Mason # if BTRFS_MAX_LEVEL != 8 1454008c04aSChris Mason # error 1464008c04aSChris Mason # endif 14785d4e461SChris Mason 14885d4e461SChris Mason static struct btrfs_lockdep_keyset { 14985d4e461SChris Mason u64 id; /* root objectid */ 15085d4e461SChris Mason const char *name_stem; /* lock name stem */ 15185d4e461SChris Mason char names[BTRFS_MAX_LEVEL + 1][20]; 15285d4e461SChris Mason struct lock_class_key keys[BTRFS_MAX_LEVEL + 1]; 15385d4e461SChris Mason } btrfs_lockdep_keysets[] = { 15485d4e461SChris Mason { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" }, 15585d4e461SChris Mason { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" }, 15685d4e461SChris Mason { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" }, 15785d4e461SChris Mason { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" }, 15885d4e461SChris Mason { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" }, 15985d4e461SChris Mason { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" }, 16060b62978SDavid Sterba { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" }, 16185d4e461SChris Mason { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" }, 16285d4e461SChris Mason { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" }, 16385d4e461SChris Mason { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" }, 16413fd8da9SDavid Sterba { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" }, 1656b20e0adSDavid Sterba { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" }, 16685d4e461SChris Mason { .id = 0, .name_stem = "tree" }, 1674008c04aSChris Mason }; 16885d4e461SChris Mason 16985d4e461SChris Mason void __init btrfs_init_lockdep(void) 17085d4e461SChris Mason { 17185d4e461SChris Mason int i, j; 17285d4e461SChris Mason 17385d4e461SChris Mason /* initialize lockdep class names */ 17485d4e461SChris Mason for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) { 17585d4e461SChris Mason struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i]; 17685d4e461SChris Mason 17785d4e461SChris Mason for (j = 0; j < ARRAY_SIZE(ks->names); j++) 17885d4e461SChris Mason snprintf(ks->names[j], sizeof(ks->names[j]), 17985d4e461SChris Mason "btrfs-%s-%02d", ks->name_stem, j); 18085d4e461SChris Mason } 18185d4e461SChris Mason } 18285d4e461SChris Mason 18385d4e461SChris Mason void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb, 18485d4e461SChris Mason int level) 18585d4e461SChris Mason { 18685d4e461SChris Mason struct btrfs_lockdep_keyset *ks; 18785d4e461SChris Mason 18885d4e461SChris Mason BUG_ON(level >= ARRAY_SIZE(ks->keys)); 18985d4e461SChris Mason 19085d4e461SChris Mason /* find the matching keyset, id 0 is the default entry */ 19185d4e461SChris Mason for (ks = btrfs_lockdep_keysets; ks->id; ks++) 19285d4e461SChris Mason if (ks->id == objectid) 19385d4e461SChris Mason break; 19485d4e461SChris Mason 19585d4e461SChris Mason lockdep_set_class_and_name(&eb->lock, 19685d4e461SChris Mason &ks->keys[level], ks->names[level]); 19785d4e461SChris Mason } 19885d4e461SChris Mason 1994008c04aSChris Mason #endif 2004008c04aSChris Mason 201d352ac68SChris Mason /* 202d352ac68SChris Mason * extents on the btree inode are pretty simple, there's one extent 203d352ac68SChris Mason * that covers the entire device 204d352ac68SChris Mason */ 2056af49dbdSDavid Sterba struct extent_map *btree_get_extent(struct btrfs_inode *inode, 20639b07b5dSOmar Sandoval struct page *page, size_t pg_offset, 20739b07b5dSOmar Sandoval u64 start, u64 len) 2085f39d397SChris Mason { 209fc4f21b1SNikolay Borisov struct extent_map_tree *em_tree = &inode->extent_tree; 2105f39d397SChris Mason struct extent_map *em; 2115f39d397SChris Mason int ret; 2125f39d397SChris Mason 213890871beSChris Mason read_lock(&em_tree->lock); 214d1310b2eSChris Mason em = lookup_extent_mapping(em_tree, start, len); 215a061fc8dSChris Mason if (em) { 216890871beSChris Mason read_unlock(&em_tree->lock); 2175f39d397SChris Mason goto out; 218a061fc8dSChris Mason } 219890871beSChris Mason read_unlock(&em_tree->lock); 2207b13b7b1SChris Mason 221172ddd60SDavid Sterba em = alloc_extent_map(); 2225f39d397SChris Mason if (!em) { 2235f39d397SChris Mason em = ERR_PTR(-ENOMEM); 2245f39d397SChris Mason goto out; 2255f39d397SChris Mason } 2265f39d397SChris Mason em->start = 0; 2270afbaf8cSChris Mason em->len = (u64)-1; 228c8b97818SChris Mason em->block_len = (u64)-1; 2295f39d397SChris Mason em->block_start = 0; 230d1310b2eSChris Mason 231890871beSChris Mason write_lock(&em_tree->lock); 23209a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 2335f39d397SChris Mason if (ret == -EEXIST) { 2345f39d397SChris Mason free_extent_map(em); 2357b13b7b1SChris Mason em = lookup_extent_mapping(em_tree, start, len); 236b4f359abSTsutomu Itoh if (!em) 2370433f20dSTsutomu Itoh em = ERR_PTR(-EIO); 2385f39d397SChris Mason } else if (ret) { 2397b13b7b1SChris Mason free_extent_map(em); 2400433f20dSTsutomu Itoh em = ERR_PTR(ret); 2415f39d397SChris Mason } 242890871beSChris Mason write_unlock(&em_tree->lock); 2437b13b7b1SChris Mason 2445f39d397SChris Mason out: 2455f39d397SChris Mason return em; 2465f39d397SChris Mason } 2475f39d397SChris Mason 248d352ac68SChris Mason /* 2492996e1f8SJohannes Thumshirn * Compute the csum of a btree block and store the result to provided buffer. 2502996e1f8SJohannes Thumshirn * 2512996e1f8SJohannes Thumshirn * Returns error if the extent buffer cannot be mapped. 252d352ac68SChris Mason */ 2532996e1f8SJohannes Thumshirn static int csum_tree_block(struct extent_buffer *buf, u8 *result) 25419c00ddcSChris Mason { 255d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = buf->fs_info; 256d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 25719c00ddcSChris Mason unsigned long len; 25819c00ddcSChris Mason unsigned long cur_len; 25919c00ddcSChris Mason unsigned long offset = BTRFS_CSUM_SIZE; 26019c00ddcSChris Mason char *kaddr; 26119c00ddcSChris Mason unsigned long map_start; 26219c00ddcSChris Mason unsigned long map_len; 26319c00ddcSChris Mason int err; 264d5178578SJohannes Thumshirn 265d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 266d5178578SJohannes Thumshirn crypto_shash_init(shash); 26719c00ddcSChris Mason 26819c00ddcSChris Mason len = buf->len - offset; 269d5178578SJohannes Thumshirn 27019c00ddcSChris Mason while (len > 0) { 271d2e174d5SJohannes Thumshirn /* 272d2e174d5SJohannes Thumshirn * Note: we don't need to check for the err == 1 case here, as 273d2e174d5SJohannes Thumshirn * with the given combination of 'start = BTRFS_CSUM_SIZE (32)' 274d2e174d5SJohannes Thumshirn * and 'min_len = 32' and the currently implemented mapping 275d2e174d5SJohannes Thumshirn * algorithm we cannot cross a page boundary. 276d2e174d5SJohannes Thumshirn */ 27719c00ddcSChris Mason err = map_private_extent_buffer(buf, offset, 32, 278a6591715SChris Mason &kaddr, &map_start, &map_len); 279c53839fcSJohannes Thumshirn if (WARN_ON(err)) 2808bd98f0eSAlex Lyakas return err; 28119c00ddcSChris Mason cur_len = min(len, map_len - (offset - map_start)); 282d5178578SJohannes Thumshirn crypto_shash_update(shash, kaddr + offset - map_start, cur_len); 28319c00ddcSChris Mason len -= cur_len; 28419c00ddcSChris Mason offset += cur_len; 28519c00ddcSChris Mason } 28671a63551SDavid Sterba memset(result, 0, BTRFS_CSUM_SIZE); 287607d432dSJosef Bacik 288d5178578SJohannes Thumshirn crypto_shash_final(shash, result); 28919c00ddcSChris Mason 29019c00ddcSChris Mason return 0; 29119c00ddcSChris Mason } 29219c00ddcSChris Mason 293d352ac68SChris Mason /* 294d352ac68SChris Mason * we can't consider a given block up to date unless the transid of the 295d352ac68SChris Mason * block matches the transid in the parent node's pointer. This is how we 296d352ac68SChris Mason * detect blocks that either didn't get written at all or got written 297d352ac68SChris Mason * in the wrong place. 298d352ac68SChris Mason */ 2991259ab75SChris Mason static int verify_parent_transid(struct extent_io_tree *io_tree, 300b9fab919SChris Mason struct extent_buffer *eb, u64 parent_transid, 301b9fab919SChris Mason int atomic) 3021259ab75SChris Mason { 3032ac55d41SJosef Bacik struct extent_state *cached_state = NULL; 3041259ab75SChris Mason int ret; 3052755a0deSDavid Sterba bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB); 3061259ab75SChris Mason 3071259ab75SChris Mason if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 3081259ab75SChris Mason return 0; 3091259ab75SChris Mason 310b9fab919SChris Mason if (atomic) 311b9fab919SChris Mason return -EAGAIN; 312b9fab919SChris Mason 313a26e8c9fSJosef Bacik if (need_lock) { 314a26e8c9fSJosef Bacik btrfs_tree_read_lock(eb); 315300aa896SDavid Sterba btrfs_set_lock_blocking_read(eb); 316a26e8c9fSJosef Bacik } 317a26e8c9fSJosef Bacik 3182ac55d41SJosef Bacik lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 319ff13db41SDavid Sterba &cached_state); 3200b32f4bbSJosef Bacik if (extent_buffer_uptodate(eb) && 3211259ab75SChris Mason btrfs_header_generation(eb) == parent_transid) { 3221259ab75SChris Mason ret = 0; 3231259ab75SChris Mason goto out; 3241259ab75SChris Mason } 32594647322SDavid Sterba btrfs_err_rl(eb->fs_info, 32694647322SDavid Sterba "parent transid verify failed on %llu wanted %llu found %llu", 32794647322SDavid Sterba eb->start, 32829549aecSWang Shilong parent_transid, btrfs_header_generation(eb)); 3291259ab75SChris Mason ret = 1; 330a26e8c9fSJosef Bacik 331a26e8c9fSJosef Bacik /* 332a26e8c9fSJosef Bacik * Things reading via commit roots that don't have normal protection, 333a26e8c9fSJosef Bacik * like send, can have a really old block in cache that may point at a 33401327610SNicholas D Steeves * block that has been freed and re-allocated. So don't clear uptodate 335a26e8c9fSJosef Bacik * if we find an eb that is under IO (dirty/writeback) because we could 336a26e8c9fSJosef Bacik * end up reading in the stale data and then writing it back out and 337a26e8c9fSJosef Bacik * making everybody very sad. 338a26e8c9fSJosef Bacik */ 339a26e8c9fSJosef Bacik if (!extent_buffer_under_io(eb)) 3400b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 34133958dc6SChris Mason out: 3422ac55d41SJosef Bacik unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 343e43bbe5eSDavid Sterba &cached_state); 344472b909fSJosef Bacik if (need_lock) 345a26e8c9fSJosef Bacik btrfs_tree_read_unlock_blocking(eb); 3461259ab75SChris Mason return ret; 3471259ab75SChris Mason } 3481259ab75SChris Mason 349e7e16f48SJohannes Thumshirn static bool btrfs_supported_super_csum(u16 csum_type) 350e7e16f48SJohannes Thumshirn { 351e7e16f48SJohannes Thumshirn switch (csum_type) { 352e7e16f48SJohannes Thumshirn case BTRFS_CSUM_TYPE_CRC32: 3533951e7f0SJohannes Thumshirn case BTRFS_CSUM_TYPE_XXHASH: 3543831bf00SJohannes Thumshirn case BTRFS_CSUM_TYPE_SHA256: 355352ae07bSDavid Sterba case BTRFS_CSUM_TYPE_BLAKE2: 356e7e16f48SJohannes Thumshirn return true; 357e7e16f48SJohannes Thumshirn default: 358e7e16f48SJohannes Thumshirn return false; 359e7e16f48SJohannes Thumshirn } 360e7e16f48SJohannes Thumshirn } 361e7e16f48SJohannes Thumshirn 362d352ac68SChris Mason /* 3631104a885SDavid Sterba * Return 0 if the superblock checksum type matches the checksum value of that 3641104a885SDavid Sterba * algorithm. Pass the raw disk superblock data. 3651104a885SDavid Sterba */ 366ab8d0fc4SJeff Mahoney static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info, 367ab8d0fc4SJeff Mahoney char *raw_disk_sb) 3681104a885SDavid Sterba { 3691104a885SDavid Sterba struct btrfs_super_block *disk_sb = 3701104a885SDavid Sterba (struct btrfs_super_block *)raw_disk_sb; 37151bce6c9SJohannes Thumshirn char result[BTRFS_CSUM_SIZE]; 372d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 373d5178578SJohannes Thumshirn 374d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 375d5178578SJohannes Thumshirn crypto_shash_init(shash); 3761104a885SDavid Sterba 3771104a885SDavid Sterba /* 3781104a885SDavid Sterba * The super_block structure does not span the whole 37951bce6c9SJohannes Thumshirn * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is 38051bce6c9SJohannes Thumshirn * filled with zeros and is included in the checksum. 3811104a885SDavid Sterba */ 382d5178578SJohannes Thumshirn crypto_shash_update(shash, raw_disk_sb + BTRFS_CSUM_SIZE, 383d5178578SJohannes Thumshirn BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE); 384d5178578SJohannes Thumshirn crypto_shash_final(shash, result); 3851104a885SDavid Sterba 38651bce6c9SJohannes Thumshirn if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb))) 387e7e16f48SJohannes Thumshirn return 1; 3881104a885SDavid Sterba 389e7e16f48SJohannes Thumshirn return 0; 3901104a885SDavid Sterba } 3911104a885SDavid Sterba 392e064d5e9SDavid Sterba int btrfs_verify_level_key(struct extent_buffer *eb, int level, 393ff76a864SLiu Bo struct btrfs_key *first_key, u64 parent_transid) 394581c1760SQu Wenruo { 395e064d5e9SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 396581c1760SQu Wenruo int found_level; 397581c1760SQu Wenruo struct btrfs_key found_key; 398581c1760SQu Wenruo int ret; 399581c1760SQu Wenruo 400581c1760SQu Wenruo found_level = btrfs_header_level(eb); 401581c1760SQu Wenruo if (found_level != level) { 40263489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 40363489055SQu Wenruo KERN_ERR "BTRFS: tree level check failed\n"); 404581c1760SQu Wenruo btrfs_err(fs_info, 405581c1760SQu Wenruo "tree level mismatch detected, bytenr=%llu level expected=%u has=%u", 406581c1760SQu Wenruo eb->start, level, found_level); 407581c1760SQu Wenruo return -EIO; 408581c1760SQu Wenruo } 409581c1760SQu Wenruo 410581c1760SQu Wenruo if (!first_key) 411581c1760SQu Wenruo return 0; 412581c1760SQu Wenruo 4135d41be6fSQu Wenruo /* 4145d41be6fSQu Wenruo * For live tree block (new tree blocks in current transaction), 4155d41be6fSQu Wenruo * we need proper lock context to avoid race, which is impossible here. 4165d41be6fSQu Wenruo * So we only checks tree blocks which is read from disk, whose 4175d41be6fSQu Wenruo * generation <= fs_info->last_trans_committed. 4185d41be6fSQu Wenruo */ 4195d41be6fSQu Wenruo if (btrfs_header_generation(eb) > fs_info->last_trans_committed) 4205d41be6fSQu Wenruo return 0; 42162fdaa52SQu Wenruo 42262fdaa52SQu Wenruo /* We have @first_key, so this @eb must have at least one item */ 42362fdaa52SQu Wenruo if (btrfs_header_nritems(eb) == 0) { 42462fdaa52SQu Wenruo btrfs_err(fs_info, 42562fdaa52SQu Wenruo "invalid tree nritems, bytenr=%llu nritems=0 expect >0", 42662fdaa52SQu Wenruo eb->start); 42762fdaa52SQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 42862fdaa52SQu Wenruo return -EUCLEAN; 42962fdaa52SQu Wenruo } 43062fdaa52SQu Wenruo 431581c1760SQu Wenruo if (found_level) 432581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &found_key, 0); 433581c1760SQu Wenruo else 434581c1760SQu Wenruo btrfs_item_key_to_cpu(eb, &found_key, 0); 435581c1760SQu Wenruo ret = btrfs_comp_cpu_keys(first_key, &found_key); 436581c1760SQu Wenruo 437581c1760SQu Wenruo if (ret) { 43863489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 43963489055SQu Wenruo KERN_ERR "BTRFS: tree first key check failed\n"); 440581c1760SQu Wenruo btrfs_err(fs_info, 441ff76a864SLiu Bo "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", 442ff76a864SLiu Bo eb->start, parent_transid, first_key->objectid, 443ff76a864SLiu Bo first_key->type, first_key->offset, 444ff76a864SLiu Bo found_key.objectid, found_key.type, 445ff76a864SLiu Bo found_key.offset); 446581c1760SQu Wenruo } 447581c1760SQu Wenruo return ret; 448581c1760SQu Wenruo } 449581c1760SQu Wenruo 4501104a885SDavid Sterba /* 451d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 452d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 453581c1760SQu Wenruo * 454581c1760SQu Wenruo * @parent_transid: expected transid, skip check if 0 455581c1760SQu Wenruo * @level: expected level, mandatory check 456581c1760SQu Wenruo * @first_key: expected key of first slot, skip check if NULL 457d352ac68SChris Mason */ 4585ab12d1fSDavid Sterba static int btree_read_extent_buffer_pages(struct extent_buffer *eb, 459581c1760SQu Wenruo u64 parent_transid, int level, 460581c1760SQu Wenruo struct btrfs_key *first_key) 461f188591eSChris Mason { 4625ab12d1fSDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 463f188591eSChris Mason struct extent_io_tree *io_tree; 464ea466794SJosef Bacik int failed = 0; 465f188591eSChris Mason int ret; 466f188591eSChris Mason int num_copies = 0; 467f188591eSChris Mason int mirror_num = 0; 468ea466794SJosef Bacik int failed_mirror = 0; 469f188591eSChris Mason 4700b246afaSJeff Mahoney io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; 471f188591eSChris Mason while (1) { 472f8397d69SNikolay Borisov clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 473c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num); 474256dd1bbSStefan Behrens if (!ret) { 475581c1760SQu Wenruo if (verify_parent_transid(io_tree, eb, 476b9fab919SChris Mason parent_transid, 0)) 477256dd1bbSStefan Behrens ret = -EIO; 478e064d5e9SDavid Sterba else if (btrfs_verify_level_key(eb, level, 479ff76a864SLiu Bo first_key, parent_transid)) 480581c1760SQu Wenruo ret = -EUCLEAN; 481581c1760SQu Wenruo else 482581c1760SQu Wenruo break; 483256dd1bbSStefan Behrens } 484d397712bSChris Mason 4850b246afaSJeff Mahoney num_copies = btrfs_num_copies(fs_info, 486f188591eSChris Mason eb->start, eb->len); 4874235298eSChris Mason if (num_copies == 1) 488ea466794SJosef Bacik break; 4894235298eSChris Mason 4905cf1ab56SJosef Bacik if (!failed_mirror) { 4915cf1ab56SJosef Bacik failed = 1; 4925cf1ab56SJosef Bacik failed_mirror = eb->read_mirror; 4935cf1ab56SJosef Bacik } 4945cf1ab56SJosef Bacik 495f188591eSChris Mason mirror_num++; 496ea466794SJosef Bacik if (mirror_num == failed_mirror) 497ea466794SJosef Bacik mirror_num++; 498ea466794SJosef Bacik 4994235298eSChris Mason if (mirror_num > num_copies) 500ea466794SJosef Bacik break; 501f188591eSChris Mason } 502ea466794SJosef Bacik 503c0901581SStefan Behrens if (failed && !ret && failed_mirror) 50420a1fbf9SDavid Sterba btrfs_repair_eb_io_failure(eb, failed_mirror); 505ea466794SJosef Bacik 506ea466794SJosef Bacik return ret; 507f188591eSChris Mason } 50819c00ddcSChris Mason 509d352ac68SChris Mason /* 510d397712bSChris Mason * checksum a dirty tree block before IO. This has extra checks to make sure 511d397712bSChris Mason * we only fill in the checksum field in the first page of a multi-page block 512d352ac68SChris Mason */ 513d397712bSChris Mason 51401d58472SDaniel Dressler static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page) 51519c00ddcSChris Mason { 5164eee4fa4SMiao Xie u64 start = page_offset(page); 51719c00ddcSChris Mason u64 found_start; 5182996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 5192996e1f8SJohannes Thumshirn u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); 52019c00ddcSChris Mason struct extent_buffer *eb; 5218d47a0d8SQu Wenruo int ret; 522f188591eSChris Mason 5234f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 5244f2de97aSJosef Bacik if (page != eb->pages[0]) 5254f2de97aSJosef Bacik return 0; 5260f805531SAlex Lyakas 52719c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 5280f805531SAlex Lyakas /* 5290f805531SAlex Lyakas * Please do not consolidate these warnings into a single if. 5300f805531SAlex Lyakas * It is useful to know what went wrong. 5310f805531SAlex Lyakas */ 5320f805531SAlex Lyakas if (WARN_ON(found_start != start)) 5330f805531SAlex Lyakas return -EUCLEAN; 5340f805531SAlex Lyakas if (WARN_ON(!PageUptodate(page))) 5350f805531SAlex Lyakas return -EUCLEAN; 5360f805531SAlex Lyakas 537de37aa51SNikolay Borisov ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid, 5380f805531SAlex Lyakas btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0); 5390f805531SAlex Lyakas 5402996e1f8SJohannes Thumshirn if (csum_tree_block(eb, result)) 5412996e1f8SJohannes Thumshirn return -EINVAL; 5422996e1f8SJohannes Thumshirn 5438d47a0d8SQu Wenruo if (btrfs_header_level(eb)) 5448d47a0d8SQu Wenruo ret = btrfs_check_node(eb); 5458d47a0d8SQu Wenruo else 5468d47a0d8SQu Wenruo ret = btrfs_check_leaf_full(eb); 5478d47a0d8SQu Wenruo 5488d47a0d8SQu Wenruo if (ret < 0) { 549c06631b0SQu Wenruo btrfs_print_tree(eb, 0); 5508d47a0d8SQu Wenruo btrfs_err(fs_info, 5518d47a0d8SQu Wenruo "block=%llu write time tree block corruption detected", 5528d47a0d8SQu Wenruo eb->start); 553c06631b0SQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 5548d47a0d8SQu Wenruo return ret; 5558d47a0d8SQu Wenruo } 5562996e1f8SJohannes Thumshirn write_extent_buffer(eb, result, 0, csum_size); 5578d47a0d8SQu Wenruo 5582996e1f8SJohannes Thumshirn return 0; 55919c00ddcSChris Mason } 56019c00ddcSChris Mason 561b0c9b3b0SDavid Sterba static int check_tree_block_fsid(struct extent_buffer *eb) 5622b82032cSYan Zheng { 563b0c9b3b0SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 56401d58472SDaniel Dressler struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 56544880fdcSAnand Jain u8 fsid[BTRFS_FSID_SIZE]; 5662b82032cSYan Zheng int ret = 1; 5672b82032cSYan Zheng 5680a4e5586SRoss Kirk read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE); 5692b82032cSYan Zheng while (fs_devices) { 5707239ff4bSNikolay Borisov u8 *metadata_uuid; 5717239ff4bSNikolay Borisov 5727239ff4bSNikolay Borisov /* 5737239ff4bSNikolay Borisov * Checking the incompat flag is only valid for the current 5747239ff4bSNikolay Borisov * fs. For seed devices it's forbidden to have their uuid 5757239ff4bSNikolay Borisov * changed so reading ->fsid in this case is fine 5767239ff4bSNikolay Borisov */ 5777239ff4bSNikolay Borisov if (fs_devices == fs_info->fs_devices && 5787239ff4bSNikolay Borisov btrfs_fs_incompat(fs_info, METADATA_UUID)) 5797239ff4bSNikolay Borisov metadata_uuid = fs_devices->metadata_uuid; 5807239ff4bSNikolay Borisov else 5817239ff4bSNikolay Borisov metadata_uuid = fs_devices->fsid; 5827239ff4bSNikolay Borisov 5837239ff4bSNikolay Borisov if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) { 5842b82032cSYan Zheng ret = 0; 5852b82032cSYan Zheng break; 5862b82032cSYan Zheng } 5872b82032cSYan Zheng fs_devices = fs_devices->seed; 5882b82032cSYan Zheng } 5892b82032cSYan Zheng return ret; 5902b82032cSYan Zheng } 5912b82032cSYan Zheng 592facc8a22SMiao Xie static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio, 593facc8a22SMiao Xie u64 phy_offset, struct page *page, 594facc8a22SMiao Xie u64 start, u64 end, int mirror) 595ce9adaa5SChris Mason { 596ce9adaa5SChris Mason u64 found_start; 597ce9adaa5SChris Mason int found_level; 598ce9adaa5SChris Mason struct extent_buffer *eb; 599ce9adaa5SChris Mason struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 60002873e43SZhao Lei struct btrfs_fs_info *fs_info = root->fs_info; 6012996e1f8SJohannes Thumshirn u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); 602f188591eSChris Mason int ret = 0; 6032996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 604727011e0SChris Mason int reads_done; 605ce9adaa5SChris Mason 606ce9adaa5SChris Mason if (!page->private) 607ce9adaa5SChris Mason goto out; 608d397712bSChris Mason 6094f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 610d397712bSChris Mason 6110b32f4bbSJosef Bacik /* the pending IO might have been the only thing that kept this buffer 6120b32f4bbSJosef Bacik * in memory. Make sure we have a ref for all this other checks 6130b32f4bbSJosef Bacik */ 61467439dadSDavid Sterba atomic_inc(&eb->refs); 6150b32f4bbSJosef Bacik 6160b32f4bbSJosef Bacik reads_done = atomic_dec_and_test(&eb->io_pages); 617727011e0SChris Mason if (!reads_done) 618727011e0SChris Mason goto err; 619f188591eSChris Mason 6205cf1ab56SJosef Bacik eb->read_mirror = mirror; 621656f30dbSFilipe Manana if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 622ea466794SJosef Bacik ret = -EIO; 623ea466794SJosef Bacik goto err; 624ea466794SJosef Bacik } 625ea466794SJosef Bacik 626ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 627727011e0SChris Mason if (found_start != eb->start) { 628893bf4b1SSu Yue btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu", 629893bf4b1SSu Yue eb->start, found_start); 630f188591eSChris Mason ret = -EIO; 631ce9adaa5SChris Mason goto err; 632ce9adaa5SChris Mason } 633b0c9b3b0SDavid Sterba if (check_tree_block_fsid(eb)) { 63402873e43SZhao Lei btrfs_err_rl(fs_info, "bad fsid on block %llu", 63594647322SDavid Sterba eb->start); 6361259ab75SChris Mason ret = -EIO; 6371259ab75SChris Mason goto err; 6381259ab75SChris Mason } 639ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 6401c24c3ceSJosef Bacik if (found_level >= BTRFS_MAX_LEVEL) { 641893bf4b1SSu Yue btrfs_err(fs_info, "bad tree block level %d on %llu", 642893bf4b1SSu Yue (int)btrfs_header_level(eb), eb->start); 6431c24c3ceSJosef Bacik ret = -EIO; 6441c24c3ceSJosef Bacik goto err; 6451c24c3ceSJosef Bacik } 646ce9adaa5SChris Mason 64785d4e461SChris Mason btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), 64885d4e461SChris Mason eb, found_level); 6494008c04aSChris Mason 6502996e1f8SJohannes Thumshirn ret = csum_tree_block(eb, result); 6518bd98f0eSAlex Lyakas if (ret) 652a826d6dcSJosef Bacik goto err; 653a826d6dcSJosef Bacik 6542996e1f8SJohannes Thumshirn if (memcmp_extent_buffer(eb, result, 0, csum_size)) { 6552996e1f8SJohannes Thumshirn u32 val; 6562996e1f8SJohannes Thumshirn u32 found = 0; 6572996e1f8SJohannes Thumshirn 6582996e1f8SJohannes Thumshirn memcpy(&found, result, csum_size); 6592996e1f8SJohannes Thumshirn 6602996e1f8SJohannes Thumshirn read_extent_buffer(eb, &val, 0, csum_size); 6612996e1f8SJohannes Thumshirn btrfs_warn_rl(fs_info, 6622996e1f8SJohannes Thumshirn "%s checksum verify failed on %llu wanted %x found %x level %d", 6632996e1f8SJohannes Thumshirn fs_info->sb->s_id, eb->start, 6642996e1f8SJohannes Thumshirn val, found, btrfs_header_level(eb)); 6652996e1f8SJohannes Thumshirn ret = -EUCLEAN; 6662996e1f8SJohannes Thumshirn goto err; 6672996e1f8SJohannes Thumshirn } 6682996e1f8SJohannes Thumshirn 669a826d6dcSJosef Bacik /* 670a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 671a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 672a826d6dcSJosef Bacik * return -EIO. 673a826d6dcSJosef Bacik */ 6741c4360eeSDavid Sterba if (found_level == 0 && btrfs_check_leaf_full(eb)) { 675a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 676a826d6dcSJosef Bacik ret = -EIO; 677a826d6dcSJosef Bacik } 678ce9adaa5SChris Mason 679813fd1dcSDavid Sterba if (found_level > 0 && btrfs_check_node(eb)) 680053ab70fSLiu Bo ret = -EIO; 681053ab70fSLiu Bo 6820b32f4bbSJosef Bacik if (!ret) 6830b32f4bbSJosef Bacik set_extent_buffer_uptodate(eb); 68475391f0dSQu Wenruo else 68575391f0dSQu Wenruo btrfs_err(fs_info, 68675391f0dSQu Wenruo "block=%llu read time tree block corruption detected", 68775391f0dSQu Wenruo eb->start); 688ce9adaa5SChris Mason err: 68979fb65a1SJosef Bacik if (reads_done && 69079fb65a1SJosef Bacik test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 691d48d71aaSDavid Sterba btree_readahead_hook(eb, ret); 6924bb31e92SArne Jansen 69353b381b3SDavid Woodhouse if (ret) { 69453b381b3SDavid Woodhouse /* 69553b381b3SDavid Woodhouse * our io error hook is going to dec the io pages 69653b381b3SDavid Woodhouse * again, we have to make sure it has something 69753b381b3SDavid Woodhouse * to decrement 69853b381b3SDavid Woodhouse */ 69953b381b3SDavid Woodhouse atomic_inc(&eb->io_pages); 7000b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 70153b381b3SDavid Woodhouse } 7020b32f4bbSJosef Bacik free_extent_buffer(eb); 703ce9adaa5SChris Mason out: 704f188591eSChris Mason return ret; 705ce9adaa5SChris Mason } 706ce9adaa5SChris Mason 7074246a0b6SChristoph Hellwig static void end_workqueue_bio(struct bio *bio) 708ce9adaa5SChris Mason { 70997eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq = bio->bi_private; 710ce9adaa5SChris Mason struct btrfs_fs_info *fs_info; 7119e0af237SLiu Bo struct btrfs_workqueue *wq; 712ce9adaa5SChris Mason 713ce9adaa5SChris Mason fs_info = end_io_wq->info; 7144e4cbee9SChristoph Hellwig end_io_wq->status = bio->bi_status; 715d20f7043SChris Mason 71637226b21SMike Christie if (bio_op(bio) == REQ_OP_WRITE) { 717a0cac0ecSOmar Sandoval if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) 7189e0af237SLiu Bo wq = fs_info->endio_meta_write_workers; 719a0cac0ecSOmar Sandoval else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) 7209e0af237SLiu Bo wq = fs_info->endio_freespace_worker; 721a0cac0ecSOmar Sandoval else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 7229e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 723a0cac0ecSOmar Sandoval else 7249e0af237SLiu Bo wq = fs_info->endio_write_workers; 7259e0af237SLiu Bo } else { 726a0cac0ecSOmar Sandoval if (unlikely(end_io_wq->metadata == BTRFS_WQ_ENDIO_DIO_REPAIR)) 7278b110e39SMiao Xie wq = fs_info->endio_repair_workers; 728a0cac0ecSOmar Sandoval else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 7299e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 730a0cac0ecSOmar Sandoval else if (end_io_wq->metadata) 7319e0af237SLiu Bo wq = fs_info->endio_meta_workers; 732a0cac0ecSOmar Sandoval else 7339e0af237SLiu Bo wq = fs_info->endio_workers; 7349e0af237SLiu Bo } 7359e0af237SLiu Bo 736a0cac0ecSOmar Sandoval btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL); 7379e0af237SLiu Bo btrfs_queue_work(wq, &end_io_wq->work); 738ce9adaa5SChris Mason } 739ce9adaa5SChris Mason 7404e4cbee9SChristoph Hellwig blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 741bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata) 7420b86a832SChris Mason { 74397eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 7448b110e39SMiao Xie 74597eb6b69SDavid Sterba end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS); 746ce9adaa5SChris Mason if (!end_io_wq) 7474e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 748ce9adaa5SChris Mason 749ce9adaa5SChris Mason end_io_wq->private = bio->bi_private; 750ce9adaa5SChris Mason end_io_wq->end_io = bio->bi_end_io; 75122c59948SChris Mason end_io_wq->info = info; 7524e4cbee9SChristoph Hellwig end_io_wq->status = 0; 753ce9adaa5SChris Mason end_io_wq->bio = bio; 75422c59948SChris Mason end_io_wq->metadata = metadata; 755ce9adaa5SChris Mason 756ce9adaa5SChris Mason bio->bi_private = end_io_wq; 757ce9adaa5SChris Mason bio->bi_end_io = end_workqueue_bio; 75822c59948SChris Mason return 0; 75922c59948SChris Mason } 76022c59948SChris Mason 7614a69a410SChris Mason static void run_one_async_start(struct btrfs_work *work) 7624a69a410SChris Mason { 7634a69a410SChris Mason struct async_submit_bio *async; 7644e4cbee9SChristoph Hellwig blk_status_t ret; 7654a69a410SChris Mason 7664a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 767c6100a4bSJosef Bacik ret = async->submit_bio_start(async->private_data, async->bio, 768eaf25d93SChris Mason async->bio_offset); 76979787eaaSJeff Mahoney if (ret) 7704e4cbee9SChristoph Hellwig async->status = ret; 7714a69a410SChris Mason } 7724a69a410SChris Mason 77306ea01b1SDavid Sterba /* 77406ea01b1SDavid Sterba * In order to insert checksums into the metadata in large chunks, we wait 77506ea01b1SDavid Sterba * until bio submission time. All the pages in the bio are checksummed and 77606ea01b1SDavid Sterba * sums are attached onto the ordered extent record. 77706ea01b1SDavid Sterba * 77806ea01b1SDavid Sterba * At IO completion time the csums attached on the ordered extent record are 77906ea01b1SDavid Sterba * inserted into the tree. 78006ea01b1SDavid Sterba */ 7814a69a410SChris Mason static void run_one_async_done(struct btrfs_work *work) 7828b712842SChris Mason { 7838b712842SChris Mason struct async_submit_bio *async; 78406ea01b1SDavid Sterba struct inode *inode; 78506ea01b1SDavid Sterba blk_status_t ret; 7868b712842SChris Mason 7878b712842SChris Mason async = container_of(work, struct async_submit_bio, work); 78806ea01b1SDavid Sterba inode = async->private_data; 7894854ddd0SChris Mason 790bb7ab3b9SAdam Buchbinder /* If an error occurred we just want to clean up the bio and move on */ 7914e4cbee9SChristoph Hellwig if (async->status) { 7924e4cbee9SChristoph Hellwig async->bio->bi_status = async->status; 7934246a0b6SChristoph Hellwig bio_endio(async->bio); 79479787eaaSJeff Mahoney return; 79579787eaaSJeff Mahoney } 79679787eaaSJeff Mahoney 797ec39f769SChris Mason /* 798ec39f769SChris Mason * All of the bios that pass through here are from async helpers. 799ec39f769SChris Mason * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context. 800ec39f769SChris Mason * This changes nothing when cgroups aren't in use. 801ec39f769SChris Mason */ 802ec39f769SChris Mason async->bio->bi_opf |= REQ_CGROUP_PUNT; 80308635baeSChris Mason ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num); 80406ea01b1SDavid Sterba if (ret) { 80506ea01b1SDavid Sterba async->bio->bi_status = ret; 80606ea01b1SDavid Sterba bio_endio(async->bio); 80706ea01b1SDavid Sterba } 8084a69a410SChris Mason } 8094a69a410SChris Mason 8104a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 8114a69a410SChris Mason { 8124a69a410SChris Mason struct async_submit_bio *async; 8134a69a410SChris Mason 8144a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 8158b712842SChris Mason kfree(async); 8168b712842SChris Mason } 8178b712842SChris Mason 8188c27cb35SLinus Torvalds blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio, 819c6100a4bSJosef Bacik int mirror_num, unsigned long bio_flags, 820c6100a4bSJosef Bacik u64 bio_offset, void *private_data, 821e288c080SDavid Sterba extent_submit_bio_start_t *submit_bio_start) 82244b8bd7eSChris Mason { 82344b8bd7eSChris Mason struct async_submit_bio *async; 82444b8bd7eSChris Mason 82544b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 82644b8bd7eSChris Mason if (!async) 8274e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 82844b8bd7eSChris Mason 829c6100a4bSJosef Bacik async->private_data = private_data; 83044b8bd7eSChris Mason async->bio = bio; 83144b8bd7eSChris Mason async->mirror_num = mirror_num; 8324a69a410SChris Mason async->submit_bio_start = submit_bio_start; 8334a69a410SChris Mason 834a0cac0ecSOmar Sandoval btrfs_init_work(&async->work, run_one_async_start, run_one_async_done, 835a0cac0ecSOmar Sandoval run_one_async_free); 8364a69a410SChris Mason 837eaf25d93SChris Mason async->bio_offset = bio_offset; 8388c8bee1dSChris Mason 8394e4cbee9SChristoph Hellwig async->status = 0; 84079787eaaSJeff Mahoney 84167f055c7SChristoph Hellwig if (op_is_sync(bio->bi_opf)) 8425cdc7ad3SQu Wenruo btrfs_set_work_high_priority(&async->work); 843d313d7a3SChris Mason 8445cdc7ad3SQu Wenruo btrfs_queue_work(fs_info->workers, &async->work); 84544b8bd7eSChris Mason return 0; 84644b8bd7eSChris Mason } 84744b8bd7eSChris Mason 8484e4cbee9SChristoph Hellwig static blk_status_t btree_csum_one_bio(struct bio *bio) 849ce3ed71aSChris Mason { 8502c30c71bSKent Overstreet struct bio_vec *bvec; 851ce3ed71aSChris Mason struct btrfs_root *root; 8522b070cfeSChristoph Hellwig int ret = 0; 8536dc4f100SMing Lei struct bvec_iter_all iter_all; 854ce3ed71aSChris Mason 855c09abff8SDavid Sterba ASSERT(!bio_flagged(bio, BIO_CLONED)); 8562b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 857ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 85801d58472SDaniel Dressler ret = csum_dirty_buffer(root->fs_info, bvec->bv_page); 85979787eaaSJeff Mahoney if (ret) 86079787eaaSJeff Mahoney break; 861ce3ed71aSChris Mason } 8622c30c71bSKent Overstreet 8634e4cbee9SChristoph Hellwig return errno_to_blk_status(ret); 864ce3ed71aSChris Mason } 865ce3ed71aSChris Mason 866d0ee3934SDavid Sterba static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio, 867eaf25d93SChris Mason u64 bio_offset) 86822c59948SChris Mason { 8698b712842SChris Mason /* 8708b712842SChris Mason * when we're called for a write, we're already in the async 8715443be45SChris Mason * submission context. Just jump into btrfs_map_bio 8728b712842SChris Mason */ 87379787eaaSJeff Mahoney return btree_csum_one_bio(bio); 87422c59948SChris Mason } 87522c59948SChris Mason 8769b4e675aSDavid Sterba static int check_async_write(struct btrfs_fs_info *fs_info, 8779b4e675aSDavid Sterba struct btrfs_inode *bi) 878de0022b9SJosef Bacik { 8796300463bSLiu Bo if (atomic_read(&bi->sync_writers)) 8806300463bSLiu Bo return 0; 8819b4e675aSDavid Sterba if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags)) 882de0022b9SJosef Bacik return 0; 883de0022b9SJosef Bacik return 1; 884de0022b9SJosef Bacik } 885de0022b9SJosef Bacik 886a56b1c7bSNikolay Borisov static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio, 88750489a57SNikolay Borisov int mirror_num, 88850489a57SNikolay Borisov unsigned long bio_flags) 88944b8bd7eSChris Mason { 8900b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 8919b4e675aSDavid Sterba int async = check_async_write(fs_info, BTRFS_I(inode)); 8924e4cbee9SChristoph Hellwig blk_status_t ret; 893cad321adSChris Mason 89437226b21SMike Christie if (bio_op(bio) != REQ_OP_WRITE) { 895cad321adSChris Mason /* 896cad321adSChris Mason * called for a read, do the setup so that checksum validation 897cad321adSChris Mason * can happen in the async kernel threads 898cad321adSChris Mason */ 8990b246afaSJeff Mahoney ret = btrfs_bio_wq_end_io(fs_info, bio, 9000b246afaSJeff Mahoney BTRFS_WQ_ENDIO_METADATA); 9011d4284bdSChris Mason if (ret) 90261891923SStefan Behrens goto out_w_error; 90308635baeSChris Mason ret = btrfs_map_bio(fs_info, bio, mirror_num); 904de0022b9SJosef Bacik } else if (!async) { 905de0022b9SJosef Bacik ret = btree_csum_one_bio(bio); 906de0022b9SJosef Bacik if (ret) 90761891923SStefan Behrens goto out_w_error; 90808635baeSChris Mason ret = btrfs_map_bio(fs_info, bio, mirror_num); 90961891923SStefan Behrens } else { 910cad321adSChris Mason /* 91161891923SStefan Behrens * kthread helpers are used to submit writes so that 91261891923SStefan Behrens * checksumming can happen in parallel across all CPUs 913cad321adSChris Mason */ 914c6100a4bSJosef Bacik ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0, 915e68f2ee7SNikolay Borisov 0, inode, btree_submit_bio_start); 91644b8bd7eSChris Mason } 91744b8bd7eSChris Mason 9184246a0b6SChristoph Hellwig if (ret) 9194246a0b6SChristoph Hellwig goto out_w_error; 9204246a0b6SChristoph Hellwig return 0; 9214246a0b6SChristoph Hellwig 92261891923SStefan Behrens out_w_error: 9234e4cbee9SChristoph Hellwig bio->bi_status = ret; 9244246a0b6SChristoph Hellwig bio_endio(bio); 92561891923SStefan Behrens return ret; 92661891923SStefan Behrens } 92761891923SStefan Behrens 9283dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 929784b4e29SChris Mason static int btree_migratepage(struct address_space *mapping, 930a6bc32b8SMel Gorman struct page *newpage, struct page *page, 931a6bc32b8SMel Gorman enum migrate_mode mode) 932784b4e29SChris Mason { 933784b4e29SChris Mason /* 934784b4e29SChris Mason * we can't safely write a btree page from here, 935784b4e29SChris Mason * we haven't done the locking hook 936784b4e29SChris Mason */ 937784b4e29SChris Mason if (PageDirty(page)) 938784b4e29SChris Mason return -EAGAIN; 939784b4e29SChris Mason /* 940784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 941784b4e29SChris Mason * We must have no buffers or drop them. 942784b4e29SChris Mason */ 943784b4e29SChris Mason if (page_has_private(page) && 944784b4e29SChris Mason !try_to_release_page(page, GFP_KERNEL)) 945784b4e29SChris Mason return -EAGAIN; 946a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 947784b4e29SChris Mason } 9483dd1462eSJan Beulich #endif 949784b4e29SChris Mason 9500da5468fSChris Mason 9510da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 9520da5468fSChris Mason struct writeback_control *wbc) 9530da5468fSChris Mason { 954e2d84521SMiao Xie struct btrfs_fs_info *fs_info; 955e2d84521SMiao Xie int ret; 956e2d84521SMiao Xie 957d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 958448d640bSChris Mason 959448d640bSChris Mason if (wbc->for_kupdate) 960448d640bSChris Mason return 0; 961448d640bSChris Mason 962e2d84521SMiao Xie fs_info = BTRFS_I(mapping->host)->root->fs_info; 963b9473439SChris Mason /* this is a bit racy, but that's ok */ 964d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 965d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 966d814a491SEthan Lien fs_info->dirty_metadata_batch); 967e2d84521SMiao Xie if (ret < 0) 968793955bcSChris Mason return 0; 969793955bcSChris Mason } 9700b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 9710da5468fSChris Mason } 9720da5468fSChris Mason 973b2950863SChristoph Hellwig static int btree_readpage(struct file *file, struct page *page) 9745f39d397SChris Mason { 975d1310b2eSChris Mason struct extent_io_tree *tree; 976d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9778ddc7d9cSJan Schmidt return extent_read_full_page(tree, page, btree_get_extent, 0); 9785f39d397SChris Mason } 9795f39d397SChris Mason 98070dec807SChris Mason static int btree_releasepage(struct page *page, gfp_t gfp_flags) 9815f39d397SChris Mason { 98298509cfcSChris Mason if (PageWriteback(page) || PageDirty(page)) 98398509cfcSChris Mason return 0; 9840c4e538bSDavid Sterba 985f7a52a40SDavid Sterba return try_release_extent_buffer(page); 986d98237b3SChris Mason } 987d98237b3SChris Mason 988d47992f8SLukas Czerner static void btree_invalidatepage(struct page *page, unsigned int offset, 989d47992f8SLukas Czerner unsigned int length) 990d98237b3SChris Mason { 991d1310b2eSChris Mason struct extent_io_tree *tree; 992d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9935f39d397SChris Mason extent_invalidatepage(tree, page, offset); 9945f39d397SChris Mason btree_releasepage(page, GFP_NOFS); 9959ad6b7bcSChris Mason if (PagePrivate(page)) { 996efe120a0SFrank Holton btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info, 997efe120a0SFrank Holton "page private not zero on page %llu", 998efe120a0SFrank Holton (unsigned long long)page_offset(page)); 9999ad6b7bcSChris Mason ClearPagePrivate(page); 10009ad6b7bcSChris Mason set_page_private(page, 0); 100109cbfeafSKirill A. Shutemov put_page(page); 10029ad6b7bcSChris Mason } 1003d98237b3SChris Mason } 1004d98237b3SChris Mason 10050b32f4bbSJosef Bacik static int btree_set_page_dirty(struct page *page) 10060b32f4bbSJosef Bacik { 1007bb146eb2SJosef Bacik #ifdef DEBUG 10080b32f4bbSJosef Bacik struct extent_buffer *eb; 10090b32f4bbSJosef Bacik 10100b32f4bbSJosef Bacik BUG_ON(!PagePrivate(page)); 10110b32f4bbSJosef Bacik eb = (struct extent_buffer *)page->private; 10120b32f4bbSJosef Bacik BUG_ON(!eb); 10130b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 10140b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 10150b32f4bbSJosef Bacik btrfs_assert_tree_locked(eb); 1016bb146eb2SJosef Bacik #endif 10170b32f4bbSJosef Bacik return __set_page_dirty_nobuffers(page); 10180b32f4bbSJosef Bacik } 10190b32f4bbSJosef Bacik 10207f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 1021d98237b3SChris Mason .readpage = btree_readpage, 10220da5468fSChris Mason .writepages = btree_writepages, 10235f39d397SChris Mason .releasepage = btree_releasepage, 10245f39d397SChris Mason .invalidatepage = btree_invalidatepage, 10255a92bc88SChris Mason #ifdef CONFIG_MIGRATION 1026784b4e29SChris Mason .migratepage = btree_migratepage, 10275a92bc88SChris Mason #endif 10280b32f4bbSJosef Bacik .set_page_dirty = btree_set_page_dirty, 1029d98237b3SChris Mason }; 1030123abc88SChris Mason 10312ff7e61eSJeff Mahoney void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr) 1032090d1875SChris Mason { 10335f39d397SChris Mason struct extent_buffer *buf = NULL; 1034537f38f0SNikolay Borisov int ret; 1035090d1875SChris Mason 10362ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 1037c871b0f2SLiu Bo if (IS_ERR(buf)) 10386197d86eSDavid Sterba return; 1039537f38f0SNikolay Borisov 1040c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(buf, WAIT_NONE, 0); 1041537f38f0SNikolay Borisov if (ret < 0) 1042537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 1043537f38f0SNikolay Borisov else 10445f39d397SChris Mason free_extent_buffer(buf); 1045090d1875SChris Mason } 1046090d1875SChris Mason 10472ff7e61eSJeff Mahoney struct extent_buffer *btrfs_find_create_tree_block( 10482ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info, 10490308af44SDavid Sterba u64 bytenr) 10500999df54SChris Mason { 10510b246afaSJeff Mahoney if (btrfs_is_testing(fs_info)) 10520b246afaSJeff Mahoney return alloc_test_extent_buffer(fs_info, bytenr); 10530b246afaSJeff Mahoney return alloc_extent_buffer(fs_info, bytenr); 10540999df54SChris Mason } 10550999df54SChris Mason 1056581c1760SQu Wenruo /* 1057581c1760SQu Wenruo * Read tree block at logical address @bytenr and do variant basic but critical 1058581c1760SQu Wenruo * verification. 1059581c1760SQu Wenruo * 1060581c1760SQu Wenruo * @parent_transid: expected transid of this tree block, skip check if 0 1061581c1760SQu Wenruo * @level: expected level, mandatory check 1062581c1760SQu Wenruo * @first_key: expected key in slot 0, skip check if NULL 1063581c1760SQu Wenruo */ 10642ff7e61eSJeff Mahoney struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, 1065581c1760SQu Wenruo u64 parent_transid, int level, 1066581c1760SQu Wenruo struct btrfs_key *first_key) 1067e20d96d6SChris Mason { 10685f39d397SChris Mason struct extent_buffer *buf = NULL; 106919c00ddcSChris Mason int ret; 107019c00ddcSChris Mason 10712ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 1072c871b0f2SLiu Bo if (IS_ERR(buf)) 1073c871b0f2SLiu Bo return buf; 1074e4204dedSChris Mason 10755ab12d1fSDavid Sterba ret = btree_read_extent_buffer_pages(buf, parent_transid, 1076581c1760SQu Wenruo level, first_key); 10770f0fe8f7SFilipe David Borba Manana if (ret) { 1078537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 107964c043deSLiu Bo return ERR_PTR(ret); 10800f0fe8f7SFilipe David Borba Manana } 10815f39d397SChris Mason return buf; 1082ce9adaa5SChris Mason 1083eb60ceacSChris Mason } 1084eb60ceacSChris Mason 10856a884d7dSDavid Sterba void btrfs_clean_tree_block(struct extent_buffer *buf) 1086ed2ff2cbSChris Mason { 10876a884d7dSDavid Sterba struct btrfs_fs_info *fs_info = buf->fs_info; 108855c69072SChris Mason if (btrfs_header_generation(buf) == 1089e2d84521SMiao Xie fs_info->running_transaction->transid) { 1090b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1091b4ce94deSChris Mason 1092b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1093104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 1094e2d84521SMiao Xie -buf->len, 1095e2d84521SMiao Xie fs_info->dirty_metadata_batch); 1096b9473439SChris Mason /* ugh, clear_extent_buffer_dirty needs to lock the page */ 10978bead258SDavid Sterba btrfs_set_lock_blocking_write(buf); 10980b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 1099925baeddSChris Mason } 11005f39d397SChris Mason } 1101ed7b63ebSJosef Bacik } 11025f39d397SChris Mason 11038257b2dcSMiao Xie static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void) 11048257b2dcSMiao Xie { 11058257b2dcSMiao Xie struct btrfs_subvolume_writers *writers; 11068257b2dcSMiao Xie int ret; 11078257b2dcSMiao Xie 11088257b2dcSMiao Xie writers = kmalloc(sizeof(*writers), GFP_NOFS); 11098257b2dcSMiao Xie if (!writers) 11108257b2dcSMiao Xie return ERR_PTR(-ENOMEM); 11118257b2dcSMiao Xie 11128a5a916dSJeff Mahoney ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS); 11138257b2dcSMiao Xie if (ret < 0) { 11148257b2dcSMiao Xie kfree(writers); 11158257b2dcSMiao Xie return ERR_PTR(ret); 11168257b2dcSMiao Xie } 11178257b2dcSMiao Xie 11188257b2dcSMiao Xie init_waitqueue_head(&writers->wait); 11198257b2dcSMiao Xie return writers; 11208257b2dcSMiao Xie } 11218257b2dcSMiao Xie 11228257b2dcSMiao Xie static void 11238257b2dcSMiao Xie btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers) 11248257b2dcSMiao Xie { 11258257b2dcSMiao Xie percpu_counter_destroy(&writers->counter); 11268257b2dcSMiao Xie kfree(writers); 11278257b2dcSMiao Xie } 11288257b2dcSMiao Xie 1129da17066cSJeff Mahoney static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, 1130e20d96d6SChris Mason u64 objectid) 1131d97e63b6SChris Mason { 11327c0260eeSJeff Mahoney bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); 113396dfcb46SJosef Bacik root->fs_info = fs_info; 1134cfaa7295SChris Mason root->node = NULL; 1135a28ec197SChris Mason root->commit_root = NULL; 113627cdeb70SMiao Xie root->state = 0; 1137d68fc57bSYan, Zheng root->orphan_cleanup_state = 0; 11380b86a832SChris Mason 11390f7d52f4SChris Mason root->last_trans = 0; 114013a8a7c8SYan, Zheng root->highest_objectid = 0; 1141eb73c1b7SMiao Xie root->nr_delalloc_inodes = 0; 1142199c2a9cSMiao Xie root->nr_ordered_extents = 0; 11436bef4d31SEric Paris root->inode_tree = RB_ROOT; 114416cdcec7SMiao Xie INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1145f0486c68SYan, Zheng root->block_rsv = NULL; 11460b86a832SChris Mason 11470b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 11485d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 1149eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_inodes); 1150eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_root); 1151199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_extents); 1152199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_root); 1153d2311e69SQu Wenruo INIT_LIST_HEAD(&root->reloc_dirty_list); 11542ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 11552ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 11565d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1157eb73c1b7SMiao Xie spin_lock_init(&root->delalloc_lock); 1158199c2a9cSMiao Xie spin_lock_init(&root->ordered_extent_lock); 1159f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 11602ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 11612ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 11628287475aSQu Wenruo spin_lock_init(&root->qgroup_meta_rsv_lock); 1163a2135011SChris Mason mutex_init(&root->objectid_mutex); 1164e02119d5SChris Mason mutex_init(&root->log_mutex); 116531f3d255SMiao Xie mutex_init(&root->ordered_extent_mutex); 1166573bfb72SMiao Xie mutex_init(&root->delalloc_mutex); 11677237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 11687237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 11697237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 11708b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[0]); 11718b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[1]); 11727237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 11737237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 11747237f183SYan Zheng atomic_set(&root->log_writers, 0); 11752ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 11760700cea7SElena Reshetova refcount_set(&root->refs, 1); 1177ea14b57fSDavid Sterba atomic_set(&root->will_be_snapshotted, 0); 11788ecebf4dSRobbie Ko atomic_set(&root->snapshot_force_cow, 0); 1179eede2bf3SOmar Sandoval atomic_set(&root->nr_swapfiles, 0); 11807237f183SYan Zheng root->log_transid = 0; 1181d1433debSMiao Xie root->log_transid_committed = -1; 1182257c62e1SChris Mason root->last_log_commit = 0; 11837c0260eeSJeff Mahoney if (!dummy) 118443eb5f29SQu Wenruo extent_io_tree_init(fs_info, &root->dirty_log_pages, 118543eb5f29SQu Wenruo IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL); 1186017e5369SChris Mason 11873768f368SChris Mason memset(&root->root_key, 0, sizeof(root->root_key)); 11883768f368SChris Mason memset(&root->root_item, 0, sizeof(root->root_item)); 11896702ed49SChris Mason memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 11907c0260eeSJeff Mahoney if (!dummy) 11913f157a2fSChris Mason root->defrag_trans_start = fs_info->generation; 119206ea65a3SJosef Bacik else 119306ea65a3SJosef Bacik root->defrag_trans_start = 0; 11944d775673SChris Mason root->root_key.objectid = objectid; 11950ee5dc67SAl Viro root->anon_dev = 0; 11968ea05e3aSAlexander Block 11975f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 1198370a11b8SQu Wenruo btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks); 11993768f368SChris Mason } 12003768f368SChris Mason 120174e4d827SDavid Sterba static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, 120296dfcb46SJosef Bacik u64 objectid, gfp_t flags) 12036f07e42eSAl Viro { 120474e4d827SDavid Sterba struct btrfs_root *root = kzalloc(sizeof(*root), flags); 12056f07e42eSAl Viro if (root) 120696dfcb46SJosef Bacik __setup_root(root, fs_info, objectid); 12076f07e42eSAl Viro return root; 12086f07e42eSAl Viro } 12096f07e42eSAl Viro 121006ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 121106ea65a3SJosef Bacik /* Should only be used by the testing infrastructure */ 1212da17066cSJeff Mahoney struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) 121306ea65a3SJosef Bacik { 121406ea65a3SJosef Bacik struct btrfs_root *root; 121506ea65a3SJosef Bacik 12167c0260eeSJeff Mahoney if (!fs_info) 12177c0260eeSJeff Mahoney return ERR_PTR(-EINVAL); 12187c0260eeSJeff Mahoney 121996dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL); 122006ea65a3SJosef Bacik if (!root) 122106ea65a3SJosef Bacik return ERR_PTR(-ENOMEM); 1222da17066cSJeff Mahoney 1223b9ef22deSFeifei Xu /* We don't use the stripesize in selftest, set it as sectorsize */ 1224faa2dbf0SJosef Bacik root->alloc_bytenr = 0; 122506ea65a3SJosef Bacik 122606ea65a3SJosef Bacik return root; 122706ea65a3SJosef Bacik } 122806ea65a3SJosef Bacik #endif 122906ea65a3SJosef Bacik 123020897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 123120897f5cSArne Jansen u64 objectid) 123220897f5cSArne Jansen { 12339b7a2440SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 123420897f5cSArne Jansen struct extent_buffer *leaf; 123520897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 123620897f5cSArne Jansen struct btrfs_root *root; 123720897f5cSArne Jansen struct btrfs_key key; 1238b89f6d1fSFilipe Manana unsigned int nofs_flag; 123920897f5cSArne Jansen int ret = 0; 124033d85fdaSQu Wenruo uuid_le uuid = NULL_UUID_LE; 124120897f5cSArne Jansen 1242b89f6d1fSFilipe Manana /* 1243b89f6d1fSFilipe Manana * We're holding a transaction handle, so use a NOFS memory allocation 1244b89f6d1fSFilipe Manana * context to avoid deadlock if reclaim happens. 1245b89f6d1fSFilipe Manana */ 1246b89f6d1fSFilipe Manana nofs_flag = memalloc_nofs_save(); 124796dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL); 1248b89f6d1fSFilipe Manana memalloc_nofs_restore(nofs_flag); 124920897f5cSArne Jansen if (!root) 125020897f5cSArne Jansen return ERR_PTR(-ENOMEM); 125120897f5cSArne Jansen 125220897f5cSArne Jansen root->root_key.objectid = objectid; 125320897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 125420897f5cSArne Jansen root->root_key.offset = 0; 125520897f5cSArne Jansen 12564d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0); 125720897f5cSArne Jansen if (IS_ERR(leaf)) { 125820897f5cSArne Jansen ret = PTR_ERR(leaf); 12591dd05682STsutomu Itoh leaf = NULL; 126020897f5cSArne Jansen goto fail; 126120897f5cSArne Jansen } 126220897f5cSArne Jansen 126320897f5cSArne Jansen root->node = leaf; 126420897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 126520897f5cSArne Jansen 126620897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 126727cdeb70SMiao Xie set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 126820897f5cSArne Jansen 126920897f5cSArne Jansen root->root_item.flags = 0; 127020897f5cSArne Jansen root->root_item.byte_limit = 0; 127120897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 127220897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 127320897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 127420897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 127520897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 127620897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 127720897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 127833d85fdaSQu Wenruo if (is_fstree(objectid)) 12796463fe58SStefan Behrens uuid_le_gen(&uuid); 12806463fe58SStefan Behrens memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE); 128120897f5cSArne Jansen root->root_item.drop_level = 0; 128220897f5cSArne Jansen 128320897f5cSArne Jansen key.objectid = objectid; 128420897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 128520897f5cSArne Jansen key.offset = 0; 128620897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 128720897f5cSArne Jansen if (ret) 128820897f5cSArne Jansen goto fail; 128920897f5cSArne Jansen 129020897f5cSArne Jansen btrfs_tree_unlock(leaf); 129120897f5cSArne Jansen 129220897f5cSArne Jansen return root; 12931dd05682STsutomu Itoh 12941dd05682STsutomu Itoh fail: 12951dd05682STsutomu Itoh if (leaf) { 12961dd05682STsutomu Itoh btrfs_tree_unlock(leaf); 129759885b39STsutomu Itoh free_extent_buffer(root->commit_root); 12981dd05682STsutomu Itoh free_extent_buffer(leaf); 12991dd05682STsutomu Itoh } 13001dd05682STsutomu Itoh kfree(root); 13011dd05682STsutomu Itoh 13021dd05682STsutomu Itoh return ERR_PTR(ret); 130320897f5cSArne Jansen } 130420897f5cSArne Jansen 13057237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1306e02119d5SChris Mason struct btrfs_fs_info *fs_info) 13070f7d52f4SChris Mason { 13080f7d52f4SChris Mason struct btrfs_root *root; 13097237f183SYan Zheng struct extent_buffer *leaf; 1310e02119d5SChris Mason 131196dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS); 1312e02119d5SChris Mason if (!root) 13137237f183SYan Zheng return ERR_PTR(-ENOMEM); 1314e02119d5SChris Mason 1315e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1316e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1317e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 131827cdeb70SMiao Xie 13197237f183SYan Zheng /* 132027cdeb70SMiao Xie * DON'T set REF_COWS for log trees 132127cdeb70SMiao Xie * 13227237f183SYan Zheng * log trees do not get reference counted because they go away 13237237f183SYan Zheng * before a real commit is actually done. They do store pointers 13247237f183SYan Zheng * to file data extents, and those reference counts still get 13257237f183SYan Zheng * updated (along with back refs to the log tree). 13267237f183SYan Zheng */ 1327e02119d5SChris Mason 13284d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 13294d75f8a9SDavid Sterba NULL, 0, 0, 0); 13307237f183SYan Zheng if (IS_ERR(leaf)) { 13317237f183SYan Zheng kfree(root); 13327237f183SYan Zheng return ERR_CAST(leaf); 13337237f183SYan Zheng } 1334e02119d5SChris Mason 13357237f183SYan Zheng root->node = leaf; 1336e02119d5SChris Mason 1337e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1338e02119d5SChris Mason btrfs_tree_unlock(root->node); 13397237f183SYan Zheng return root; 13407237f183SYan Zheng } 13417237f183SYan Zheng 13427237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 13437237f183SYan Zheng struct btrfs_fs_info *fs_info) 13447237f183SYan Zheng { 13457237f183SYan Zheng struct btrfs_root *log_root; 13467237f183SYan Zheng 13477237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 13487237f183SYan Zheng if (IS_ERR(log_root)) 13497237f183SYan Zheng return PTR_ERR(log_root); 13507237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 13517237f183SYan Zheng fs_info->log_root_tree = log_root; 13527237f183SYan Zheng return 0; 13537237f183SYan Zheng } 13547237f183SYan Zheng 13557237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 13567237f183SYan Zheng struct btrfs_root *root) 13577237f183SYan Zheng { 13580b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13597237f183SYan Zheng struct btrfs_root *log_root; 13607237f183SYan Zheng struct btrfs_inode_item *inode_item; 13617237f183SYan Zheng 13620b246afaSJeff Mahoney log_root = alloc_log_tree(trans, fs_info); 13637237f183SYan Zheng if (IS_ERR(log_root)) 13647237f183SYan Zheng return PTR_ERR(log_root); 13657237f183SYan Zheng 13667237f183SYan Zheng log_root->last_trans = trans->transid; 13677237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 13687237f183SYan Zheng 13697237f183SYan Zheng inode_item = &log_root->root_item.inode; 13703cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 13713cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 13723cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 1373da17066cSJeff Mahoney btrfs_set_stack_inode_nbytes(inode_item, 13740b246afaSJeff Mahoney fs_info->nodesize); 13753cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 13767237f183SYan Zheng 13775d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 13787237f183SYan Zheng 13797237f183SYan Zheng WARN_ON(root->log_root); 13807237f183SYan Zheng root->log_root = log_root; 13817237f183SYan Zheng root->log_transid = 0; 1382d1433debSMiao Xie root->log_transid_committed = -1; 1383257c62e1SChris Mason root->last_log_commit = 0; 1384e02119d5SChris Mason return 0; 1385e02119d5SChris Mason } 1386e02119d5SChris Mason 138735a3621bSStefan Behrens static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 1388cb517eabSMiao Xie struct btrfs_key *key) 1389e02119d5SChris Mason { 1390e02119d5SChris Mason struct btrfs_root *root; 1391e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 13920f7d52f4SChris Mason struct btrfs_path *path; 139384234f3aSYan Zheng u64 generation; 1394cb517eabSMiao Xie int ret; 1395581c1760SQu Wenruo int level; 1396cb517eabSMiao Xie 1397cb517eabSMiao Xie path = btrfs_alloc_path(); 1398cb517eabSMiao Xie if (!path) 1399cb517eabSMiao Xie return ERR_PTR(-ENOMEM); 14000f7d52f4SChris Mason 140196dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS); 1402cb517eabSMiao Xie if (!root) { 1403cb517eabSMiao Xie ret = -ENOMEM; 1404cb517eabSMiao Xie goto alloc_fail; 14050f7d52f4SChris Mason } 14060f7d52f4SChris Mason 1407cb517eabSMiao Xie ret = btrfs_find_root(tree_root, key, path, 1408cb517eabSMiao Xie &root->root_item, &root->root_key); 14090f7d52f4SChris Mason if (ret) { 141013a8a7c8SYan, Zheng if (ret > 0) 141113a8a7c8SYan, Zheng ret = -ENOENT; 1412cb517eabSMiao Xie goto find_fail; 14130f7d52f4SChris Mason } 141413a8a7c8SYan, Zheng 141584234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1416581c1760SQu Wenruo level = btrfs_root_level(&root->root_item); 14172ff7e61eSJeff Mahoney root->node = read_tree_block(fs_info, 14182ff7e61eSJeff Mahoney btrfs_root_bytenr(&root->root_item), 1419581c1760SQu Wenruo generation, level, NULL); 142064c043deSLiu Bo if (IS_ERR(root->node)) { 142164c043deSLiu Bo ret = PTR_ERR(root->node); 1422cb517eabSMiao Xie goto find_fail; 1423cb517eabSMiao Xie } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1424cb517eabSMiao Xie ret = -EIO; 142564c043deSLiu Bo free_extent_buffer(root->node); 142664c043deSLiu Bo goto find_fail; 1427416bc658SJosef Bacik } 14285d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 142913a8a7c8SYan, Zheng out: 1430cb517eabSMiao Xie btrfs_free_path(path); 1431cb517eabSMiao Xie return root; 1432cb517eabSMiao Xie 1433cb517eabSMiao Xie find_fail: 1434cb517eabSMiao Xie kfree(root); 1435cb517eabSMiao Xie alloc_fail: 1436cb517eabSMiao Xie root = ERR_PTR(ret); 1437cb517eabSMiao Xie goto out; 1438cb517eabSMiao Xie } 1439cb517eabSMiao Xie 1440cb517eabSMiao Xie struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root, 1441cb517eabSMiao Xie struct btrfs_key *location) 1442cb517eabSMiao Xie { 1443cb517eabSMiao Xie struct btrfs_root *root; 1444cb517eabSMiao Xie 1445cb517eabSMiao Xie root = btrfs_read_tree_root(tree_root, location); 1446cb517eabSMiao Xie if (IS_ERR(root)) 1447cb517eabSMiao Xie return root; 1448cb517eabSMiao Xie 1449cb517eabSMiao Xie if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { 145027cdeb70SMiao Xie set_bit(BTRFS_ROOT_REF_COWS, &root->state); 145108fe4db1SLi Zefan btrfs_check_and_init_root_item(&root->root_item); 145208fe4db1SLi Zefan } 145313a8a7c8SYan, Zheng 14545eda7b5eSChris Mason return root; 14555eda7b5eSChris Mason } 14565eda7b5eSChris Mason 1457cb517eabSMiao Xie int btrfs_init_fs_root(struct btrfs_root *root) 1458cb517eabSMiao Xie { 1459cb517eabSMiao Xie int ret; 14608257b2dcSMiao Xie struct btrfs_subvolume_writers *writers; 1461cb517eabSMiao Xie 1462cb517eabSMiao Xie root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); 1463cb517eabSMiao Xie root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), 1464cb517eabSMiao Xie GFP_NOFS); 1465cb517eabSMiao Xie if (!root->free_ino_pinned || !root->free_ino_ctl) { 1466cb517eabSMiao Xie ret = -ENOMEM; 1467cb517eabSMiao Xie goto fail; 1468cb517eabSMiao Xie } 1469cb517eabSMiao Xie 14708257b2dcSMiao Xie writers = btrfs_alloc_subvolume_writers(); 14718257b2dcSMiao Xie if (IS_ERR(writers)) { 14728257b2dcSMiao Xie ret = PTR_ERR(writers); 14738257b2dcSMiao Xie goto fail; 14748257b2dcSMiao Xie } 14758257b2dcSMiao Xie root->subv_writers = writers; 14768257b2dcSMiao Xie 1477cb517eabSMiao Xie btrfs_init_free_ino_ctl(root); 147857cdc8dbSDavid Sterba spin_lock_init(&root->ino_cache_lock); 147957cdc8dbSDavid Sterba init_waitqueue_head(&root->ino_cache_wait); 1480cb517eabSMiao Xie 1481cb517eabSMiao Xie ret = get_anon_bdev(&root->anon_dev); 1482cb517eabSMiao Xie if (ret) 1483876d2cf1SLiu Bo goto fail; 1484f32e48e9SChandan Rajendra 1485f32e48e9SChandan Rajendra mutex_lock(&root->objectid_mutex); 1486f32e48e9SChandan Rajendra ret = btrfs_find_highest_objectid(root, 1487f32e48e9SChandan Rajendra &root->highest_objectid); 1488f32e48e9SChandan Rajendra if (ret) { 1489f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1490876d2cf1SLiu Bo goto fail; 1491f32e48e9SChandan Rajendra } 1492f32e48e9SChandan Rajendra 1493f32e48e9SChandan Rajendra ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 1494f32e48e9SChandan Rajendra 1495f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1496f32e48e9SChandan Rajendra 1497cb517eabSMiao Xie return 0; 1498cb517eabSMiao Xie fail: 149984db5ccfSDavid Sterba /* The caller is responsible to call btrfs_free_fs_root */ 1500cb517eabSMiao Xie return ret; 1501cb517eabSMiao Xie } 1502cb517eabSMiao Xie 150335bbb97fSJeff Mahoney struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1504cb517eabSMiao Xie u64 root_id) 1505cb517eabSMiao Xie { 1506cb517eabSMiao Xie struct btrfs_root *root; 1507cb517eabSMiao Xie 1508cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1509cb517eabSMiao Xie root = radix_tree_lookup(&fs_info->fs_roots_radix, 1510cb517eabSMiao Xie (unsigned long)root_id); 1511cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1512cb517eabSMiao Xie return root; 1513cb517eabSMiao Xie } 1514cb517eabSMiao Xie 1515cb517eabSMiao Xie int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1516cb517eabSMiao Xie struct btrfs_root *root) 1517cb517eabSMiao Xie { 1518cb517eabSMiao Xie int ret; 1519cb517eabSMiao Xie 1520e1860a77SDavid Sterba ret = radix_tree_preload(GFP_NOFS); 1521cb517eabSMiao Xie if (ret) 1522cb517eabSMiao Xie return ret; 1523cb517eabSMiao Xie 1524cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1525cb517eabSMiao Xie ret = radix_tree_insert(&fs_info->fs_roots_radix, 1526cb517eabSMiao Xie (unsigned long)root->root_key.objectid, 1527cb517eabSMiao Xie root); 1528cb517eabSMiao Xie if (ret == 0) 152927cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1530cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1531cb517eabSMiao Xie radix_tree_preload_end(); 1532cb517eabSMiao Xie 1533cb517eabSMiao Xie return ret; 1534cb517eabSMiao Xie } 1535cb517eabSMiao Xie 1536c00869f1SMiao Xie struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 1537c00869f1SMiao Xie struct btrfs_key *location, 1538c00869f1SMiao Xie bool check_ref) 15395eda7b5eSChris Mason { 15405eda7b5eSChris Mason struct btrfs_root *root; 1541381cf658SDavid Sterba struct btrfs_path *path; 15421d4c08e0SDavid Sterba struct btrfs_key key; 15435eda7b5eSChris Mason int ret; 15445eda7b5eSChris Mason 1545edbd8d4eSChris Mason if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 1546edbd8d4eSChris Mason return fs_info->tree_root; 1547edbd8d4eSChris Mason if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID) 1548edbd8d4eSChris Mason return fs_info->extent_root; 15498f18cf13SChris Mason if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID) 15508f18cf13SChris Mason return fs_info->chunk_root; 15518f18cf13SChris Mason if (location->objectid == BTRFS_DEV_TREE_OBJECTID) 15528f18cf13SChris Mason return fs_info->dev_root; 15530403e47eSYan Zheng if (location->objectid == BTRFS_CSUM_TREE_OBJECTID) 15540403e47eSYan Zheng return fs_info->csum_root; 1555bcef60f2SArne Jansen if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID) 1556bcef60f2SArne Jansen return fs_info->quota_root ? fs_info->quota_root : 1557bcef60f2SArne Jansen ERR_PTR(-ENOENT); 1558f7a81ea4SStefan Behrens if (location->objectid == BTRFS_UUID_TREE_OBJECTID) 1559f7a81ea4SStefan Behrens return fs_info->uuid_root ? fs_info->uuid_root : 1560f7a81ea4SStefan Behrens ERR_PTR(-ENOENT); 156170f6d82eSOmar Sandoval if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) 156270f6d82eSOmar Sandoval return fs_info->free_space_root ? fs_info->free_space_root : 156370f6d82eSOmar Sandoval ERR_PTR(-ENOENT); 15644df27c4dSYan, Zheng again: 1565cb517eabSMiao Xie root = btrfs_lookup_fs_root(fs_info, location->objectid); 156648475471SStefan Behrens if (root) { 1567c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) 156848475471SStefan Behrens return ERR_PTR(-ENOENT); 15695eda7b5eSChris Mason return root; 157048475471SStefan Behrens } 15715eda7b5eSChris Mason 1572cb517eabSMiao Xie root = btrfs_read_fs_root(fs_info->tree_root, location); 15735eda7b5eSChris Mason if (IS_ERR(root)) 15745eda7b5eSChris Mason return root; 15753394e160SChris Mason 1576c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1577d68fc57bSYan, Zheng ret = -ENOENT; 1578d68fc57bSYan, Zheng goto fail; 1579d68fc57bSYan, Zheng } 1580d68fc57bSYan, Zheng 1581cb517eabSMiao Xie ret = btrfs_init_fs_root(root); 1582cb517eabSMiao Xie if (ret) 1583cb517eabSMiao Xie goto fail; 1584cb517eabSMiao Xie 1585381cf658SDavid Sterba path = btrfs_alloc_path(); 1586381cf658SDavid Sterba if (!path) { 1587381cf658SDavid Sterba ret = -ENOMEM; 1588381cf658SDavid Sterba goto fail; 1589381cf658SDavid Sterba } 15901d4c08e0SDavid Sterba key.objectid = BTRFS_ORPHAN_OBJECTID; 15911d4c08e0SDavid Sterba key.type = BTRFS_ORPHAN_ITEM_KEY; 15921d4c08e0SDavid Sterba key.offset = location->objectid; 15931d4c08e0SDavid Sterba 15941d4c08e0SDavid Sterba ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1595381cf658SDavid Sterba btrfs_free_path(path); 1596d68fc57bSYan, Zheng if (ret < 0) 1597d68fc57bSYan, Zheng goto fail; 1598d68fc57bSYan, Zheng if (ret == 0) 159927cdeb70SMiao Xie set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1600d68fc57bSYan, Zheng 1601cb517eabSMiao Xie ret = btrfs_insert_fs_root(fs_info, root); 16020f7d52f4SChris Mason if (ret) { 16034df27c4dSYan, Zheng if (ret == -EEXIST) { 160484db5ccfSDavid Sterba btrfs_free_fs_root(root); 16054df27c4dSYan, Zheng goto again; 16060f7d52f4SChris Mason } 16074df27c4dSYan, Zheng goto fail; 16084df27c4dSYan, Zheng } 1609edbd8d4eSChris Mason return root; 16104df27c4dSYan, Zheng fail: 161184db5ccfSDavid Sterba btrfs_free_fs_root(root); 16124df27c4dSYan, Zheng return ERR_PTR(ret); 1613edbd8d4eSChris Mason } 1614edbd8d4eSChris Mason 161504160088SChris Mason static int btrfs_congested_fn(void *congested_data, int bdi_bits) 161604160088SChris Mason { 161704160088SChris Mason struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; 161804160088SChris Mason int ret = 0; 161904160088SChris Mason struct btrfs_device *device; 162004160088SChris Mason struct backing_dev_info *bdi; 1621b7967db7SChris Mason 16221f78160cSXiao Guangrong rcu_read_lock(); 16231f78160cSXiao Guangrong list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) { 1624dfe25020SChris Mason if (!device->bdev) 1625dfe25020SChris Mason continue; 1626efa7c9f9SJan Kara bdi = device->bdev->bd_bdi; 1627ff9ea323STejun Heo if (bdi_congested(bdi, bdi_bits)) { 162804160088SChris Mason ret = 1; 162904160088SChris Mason break; 163004160088SChris Mason } 163104160088SChris Mason } 16321f78160cSXiao Guangrong rcu_read_unlock(); 163304160088SChris Mason return ret; 163404160088SChris Mason } 163504160088SChris Mason 16368b712842SChris Mason /* 16378b712842SChris Mason * called by the kthread helper functions to finally call the bio end_io 16388b712842SChris Mason * functions. This is where read checksum verification actually happens 16398b712842SChris Mason */ 16408b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work) 1641ce9adaa5SChris Mason { 1642ce9adaa5SChris Mason struct bio *bio; 164397eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 1644ce9adaa5SChris Mason 164597eb6b69SDavid Sterba end_io_wq = container_of(work, struct btrfs_end_io_wq, work); 1646ce9adaa5SChris Mason bio = end_io_wq->bio; 16478b712842SChris Mason 16484e4cbee9SChristoph Hellwig bio->bi_status = end_io_wq->status; 1649ce9adaa5SChris Mason bio->bi_private = end_io_wq->private; 1650ce9adaa5SChris Mason bio->bi_end_io = end_io_wq->end_io; 16514246a0b6SChristoph Hellwig bio_endio(bio); 16529be490f1SOmar Sandoval kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq); 1653ce9adaa5SChris Mason } 165444b8bd7eSChris Mason 1655a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1656a74a4b97SChris Mason { 1657a74a4b97SChris Mason struct btrfs_root *root = arg; 16580b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1659d0278245SMiao Xie int again; 1660a74a4b97SChris Mason 1661d6fd0ae2SOmar Sandoval while (1) { 1662d0278245SMiao Xie again = 0; 16639d1a2a3aSDavid Sterba 1664fd340d0fSJosef Bacik set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1665fd340d0fSJosef Bacik 1666d0278245SMiao Xie /* Make the cleaner go to sleep early. */ 16672ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) 1668d0278245SMiao Xie goto sleep; 1669d0278245SMiao Xie 167090c711abSZygo Blaxell /* 167190c711abSZygo Blaxell * Do not do anything if we might cause open_ctree() to block 167290c711abSZygo Blaxell * before we have finished mounting the filesystem. 167390c711abSZygo Blaxell */ 16740b246afaSJeff Mahoney if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 167590c711abSZygo Blaxell goto sleep; 167690c711abSZygo Blaxell 16770b246afaSJeff Mahoney if (!mutex_trylock(&fs_info->cleaner_mutex)) 1678d0278245SMiao Xie goto sleep; 1679d0278245SMiao Xie 1680dc7f370cSMiao Xie /* 1681dc7f370cSMiao Xie * Avoid the problem that we change the status of the fs 1682dc7f370cSMiao Xie * during the above check and trylock. 1683dc7f370cSMiao Xie */ 16842ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) { 16850b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1686dc7f370cSMiao Xie goto sleep; 1687dc7f370cSMiao Xie } 1688dc7f370cSMiao Xie 16892ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 1690c2d6cb16SFilipe Manana 16919d1a2a3aSDavid Sterba again = btrfs_clean_one_deleted_snapshot(root); 16920b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1693a74a4b97SChris Mason 1694d0278245SMiao Xie /* 169505323cd1SMiao Xie * The defragger has dealt with the R/O remount and umount, 169605323cd1SMiao Xie * needn't do anything special here. 1697d0278245SMiao Xie */ 16980b246afaSJeff Mahoney btrfs_run_defrag_inodes(fs_info); 169967c5e7d4SFilipe Manana 170067c5e7d4SFilipe Manana /* 170167c5e7d4SFilipe Manana * Acquires fs_info->delete_unused_bgs_mutex to avoid racing 170267c5e7d4SFilipe Manana * with relocation (btrfs_relocate_chunk) and relocation 170367c5e7d4SFilipe Manana * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 170467c5e7d4SFilipe Manana * after acquiring fs_info->delete_unused_bgs_mutex. So we 170567c5e7d4SFilipe Manana * can't hold, nor need to, fs_info->cleaner_mutex when deleting 170667c5e7d4SFilipe Manana * unused block groups. 170767c5e7d4SFilipe Manana */ 17080b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 1709d0278245SMiao Xie sleep: 1710fd340d0fSJosef Bacik clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1711d6fd0ae2SOmar Sandoval if (kthread_should_park()) 1712d6fd0ae2SOmar Sandoval kthread_parkme(); 1713d6fd0ae2SOmar Sandoval if (kthread_should_stop()) 1714d6fd0ae2SOmar Sandoval return 0; 1715838fe188SJiri Kosina if (!again) { 1716a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 1717a74a4b97SChris Mason schedule(); 1718a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1719a74a4b97SChris Mason } 1720da288d28SFilipe Manana } 1721a74a4b97SChris Mason } 1722a74a4b97SChris Mason 1723a74a4b97SChris Mason static int transaction_kthread(void *arg) 1724a74a4b97SChris Mason { 1725a74a4b97SChris Mason struct btrfs_root *root = arg; 17260b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1727a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1728a74a4b97SChris Mason struct btrfs_transaction *cur; 17298929ecfaSYan, Zheng u64 transid; 1730a944442cSAllen Pais time64_t now; 1731a74a4b97SChris Mason unsigned long delay; 1732914b2007SJan Kara bool cannot_commit; 1733a74a4b97SChris Mason 1734a74a4b97SChris Mason do { 1735914b2007SJan Kara cannot_commit = false; 17360b246afaSJeff Mahoney delay = HZ * fs_info->commit_interval; 17370b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 1738a74a4b97SChris Mason 17390b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 17400b246afaSJeff Mahoney cur = fs_info->running_transaction; 1741a74a4b97SChris Mason if (!cur) { 17420b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1743a74a4b97SChris Mason goto sleep; 1744a74a4b97SChris Mason } 174531153d81SYan Zheng 1746afd48513SArnd Bergmann now = ktime_get_seconds(); 17473296bf56SQu Wenruo if (cur->state < TRANS_STATE_COMMIT_START && 1748a514d638SQu Wenruo !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) && 17498b87dc17SDavid Sterba (now < cur->start_time || 17500b246afaSJeff Mahoney now - cur->start_time < fs_info->commit_interval)) { 17510b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1752a74a4b97SChris Mason delay = HZ * 5; 1753a74a4b97SChris Mason goto sleep; 1754a74a4b97SChris Mason } 17558929ecfaSYan, Zheng transid = cur->transid; 17560b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 175756bec294SChris Mason 175879787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 1759354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 1760914b2007SJan Kara if (IS_ERR(trans)) { 1761354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 1762914b2007SJan Kara cannot_commit = true; 176379787eaaSJeff Mahoney goto sleep; 1764914b2007SJan Kara } 17658929ecfaSYan, Zheng if (transid == trans->transid) { 17663a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 17678929ecfaSYan, Zheng } else { 17683a45bb20SJeff Mahoney btrfs_end_transaction(trans); 17698929ecfaSYan, Zheng } 1770a74a4b97SChris Mason sleep: 17710b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 17720b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 1773a74a4b97SChris Mason 17744e121c06SJosef Bacik if (unlikely(test_bit(BTRFS_FS_STATE_ERROR, 17750b246afaSJeff Mahoney &fs_info->fs_state))) 17762ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 17778929ecfaSYan, Zheng if (!kthread_should_stop() && 17780b246afaSJeff Mahoney (!btrfs_transaction_blocked(fs_info) || 1779914b2007SJan Kara cannot_commit)) 1780bc5511d0SNikolay Borisov schedule_timeout_interruptible(delay); 1781a74a4b97SChris Mason } while (!kthread_should_stop()); 1782a74a4b97SChris Mason return 0; 1783a74a4b97SChris Mason } 1784a74a4b97SChris Mason 1785af31f5e5SChris Mason /* 178601f0f9daSNikolay Borisov * This will find the highest generation in the array of root backups. The 178701f0f9daSNikolay Borisov * index of the highest array is returned, or -EINVAL if we can't find 178801f0f9daSNikolay Borisov * anything. 1789af31f5e5SChris Mason * 1790af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1791af31f5e5SChris Mason * generation of the latest root in the array with the generation 1792af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1793af31f5e5SChris Mason */ 179401f0f9daSNikolay Borisov static int find_newest_super_backup(struct btrfs_fs_info *info) 1795af31f5e5SChris Mason { 179601f0f9daSNikolay Borisov const u64 newest_gen = btrfs_super_generation(info->super_copy); 1797af31f5e5SChris Mason u64 cur; 1798af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1799af31f5e5SChris Mason int i; 1800af31f5e5SChris Mason 1801af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1802af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1803af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1804af31f5e5SChris Mason if (cur == newest_gen) 180501f0f9daSNikolay Borisov return i; 1806af31f5e5SChris Mason } 1807af31f5e5SChris Mason 180801f0f9daSNikolay Borisov return -EINVAL; 1809af31f5e5SChris Mason } 1810af31f5e5SChris Mason 1811af31f5e5SChris Mason /* 1812af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1813af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1814af31f5e5SChris Mason * done 1815af31f5e5SChris Mason */ 1816af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1817af31f5e5SChris Mason { 18186ef108ddSNikolay Borisov const int next_backup = info->backup_root_index; 1819af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1820af31f5e5SChris Mason 1821af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1822af31f5e5SChris Mason 1823af31f5e5SChris Mason /* 1824af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1825af31f5e5SChris Mason * regardless of which ones we use today 1826af31f5e5SChris Mason */ 1827af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1828af31f5e5SChris Mason 1829af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1830af31f5e5SChris Mason 1831af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1832af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1833af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1834af31f5e5SChris Mason 1835af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1836af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1837af31f5e5SChris Mason 1838af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1839af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1840af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 1841af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 1842af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 1843af31f5e5SChris Mason 1844af31f5e5SChris Mason btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); 1845af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 1846af31f5e5SChris Mason btrfs_header_generation(info->extent_root->node)); 1847af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 1848af31f5e5SChris Mason btrfs_header_level(info->extent_root->node)); 1849af31f5e5SChris Mason 18507c7e82a7SChris Mason /* 18517c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 18527c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 18537c7e82a7SChris Mason */ 18547c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 18557c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 18567c7e82a7SChris Mason info->fs_root->node->start); 1857af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 1858af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 1859af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 1860af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 18617c7e82a7SChris Mason } 1862af31f5e5SChris Mason 1863af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 1864af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 1865af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 1866af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 1867af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 1868af31f5e5SChris Mason 1869af31f5e5SChris Mason btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); 1870af31f5e5SChris Mason btrfs_set_backup_csum_root_gen(root_backup, 1871af31f5e5SChris Mason btrfs_header_generation(info->csum_root->node)); 1872af31f5e5SChris Mason btrfs_set_backup_csum_root_level(root_backup, 1873af31f5e5SChris Mason btrfs_header_level(info->csum_root->node)); 1874af31f5e5SChris Mason 1875af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 1876af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 1877af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 1878af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 1879af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 1880af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 1881af31f5e5SChris Mason 1882af31f5e5SChris Mason /* 1883af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 1884af31f5e5SChris Mason * for the next commit 1885af31f5e5SChris Mason */ 1886af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 1887af31f5e5SChris Mason &info->super_for_commit->super_roots, 1888af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 1889af31f5e5SChris Mason } 1890af31f5e5SChris Mason 1891af31f5e5SChris Mason /* 1892bd2336b2SNikolay Borisov * read_backup_root - Reads a backup root based on the passed priority. Prio 0 1893bd2336b2SNikolay Borisov * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots 1894bd2336b2SNikolay Borisov * 1895bd2336b2SNikolay Borisov * fs_info - filesystem whose backup roots need to be read 1896bd2336b2SNikolay Borisov * priority - priority of backup root required 1897bd2336b2SNikolay Borisov * 1898bd2336b2SNikolay Borisov * Returns backup root index on success and -EINVAL otherwise. 1899bd2336b2SNikolay Borisov */ 1900bd2336b2SNikolay Borisov static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority) 1901bd2336b2SNikolay Borisov { 1902bd2336b2SNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 1903bd2336b2SNikolay Borisov struct btrfs_super_block *super = fs_info->super_copy; 1904bd2336b2SNikolay Borisov struct btrfs_root_backup *root_backup; 1905bd2336b2SNikolay Borisov 1906bd2336b2SNikolay Borisov if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) { 1907bd2336b2SNikolay Borisov if (priority == 0) 1908bd2336b2SNikolay Borisov return backup_index; 1909bd2336b2SNikolay Borisov 1910bd2336b2SNikolay Borisov backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority; 1911bd2336b2SNikolay Borisov backup_index %= BTRFS_NUM_BACKUP_ROOTS; 1912bd2336b2SNikolay Borisov } else { 1913bd2336b2SNikolay Borisov return -EINVAL; 1914bd2336b2SNikolay Borisov } 1915bd2336b2SNikolay Borisov 1916bd2336b2SNikolay Borisov root_backup = super->super_roots + backup_index; 1917bd2336b2SNikolay Borisov 1918bd2336b2SNikolay Borisov btrfs_set_super_generation(super, 1919bd2336b2SNikolay Borisov btrfs_backup_tree_root_gen(root_backup)); 1920bd2336b2SNikolay Borisov btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 1921bd2336b2SNikolay Borisov btrfs_set_super_root_level(super, 1922bd2336b2SNikolay Borisov btrfs_backup_tree_root_level(root_backup)); 1923bd2336b2SNikolay Borisov btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 1924bd2336b2SNikolay Borisov 1925bd2336b2SNikolay Borisov /* 1926bd2336b2SNikolay Borisov * Fixme: the total bytes and num_devices need to match or we should 1927bd2336b2SNikolay Borisov * need a fsck 1928bd2336b2SNikolay Borisov */ 1929bd2336b2SNikolay Borisov btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 1930bd2336b2SNikolay Borisov btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 1931bd2336b2SNikolay Borisov 1932bd2336b2SNikolay Borisov return backup_index; 1933bd2336b2SNikolay Borisov } 1934bd2336b2SNikolay Borisov 19357abadb64SLiu Bo /* helper to cleanup workers */ 19367abadb64SLiu Bo static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 19377abadb64SLiu Bo { 1938dc6e3209SQu Wenruo btrfs_destroy_workqueue(fs_info->fixup_workers); 1939afe3d242SQu Wenruo btrfs_destroy_workqueue(fs_info->delalloc_workers); 19405cdc7ad3SQu Wenruo btrfs_destroy_workqueue(fs_info->workers); 1941fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_workers); 1942fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_raid56_workers); 19438b110e39SMiao Xie btrfs_destroy_workqueue(fs_info->endio_repair_workers); 1944d05a33acSQu Wenruo btrfs_destroy_workqueue(fs_info->rmw_workers); 1945fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_write_workers); 1946fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 19475b3bc44eSQu Wenruo btrfs_destroy_workqueue(fs_info->delayed_workers); 1948e66f0bb1SQu Wenruo btrfs_destroy_workqueue(fs_info->caching_workers); 1949736cfa15SQu Wenruo btrfs_destroy_workqueue(fs_info->readahead_workers); 1950a44903abSQu Wenruo btrfs_destroy_workqueue(fs_info->flush_workers); 1951fc97fab0SQu Wenruo btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 1952b0643e59SDennis Zhou if (fs_info->discard_ctl.discard_workers) 1953b0643e59SDennis Zhou destroy_workqueue(fs_info->discard_ctl.discard_workers); 1954a9b9477dSFilipe Manana /* 1955a9b9477dSFilipe Manana * Now that all other work queues are destroyed, we can safely destroy 1956a9b9477dSFilipe Manana * the queues used for metadata I/O, since tasks from those other work 1957a9b9477dSFilipe Manana * queues can do metadata I/O operations. 1958a9b9477dSFilipe Manana */ 1959a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_workers); 1960a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_write_workers); 19617abadb64SLiu Bo } 19627abadb64SLiu Bo 19632e9f5954SRashika static void free_root_extent_buffers(struct btrfs_root *root) 19642e9f5954SRashika { 19652e9f5954SRashika if (root) { 19662e9f5954SRashika free_extent_buffer(root->node); 19672e9f5954SRashika free_extent_buffer(root->commit_root); 19682e9f5954SRashika root->node = NULL; 19692e9f5954SRashika root->commit_root = NULL; 19702e9f5954SRashika } 19712e9f5954SRashika } 19722e9f5954SRashika 1973af31f5e5SChris Mason /* helper to cleanup tree roots */ 19744273eaffSAnand Jain static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root) 1975af31f5e5SChris Mason { 19762e9f5954SRashika free_root_extent_buffers(info->tree_root); 1977655b09feSJosef Bacik 19782e9f5954SRashika free_root_extent_buffers(info->dev_root); 19792e9f5954SRashika free_root_extent_buffers(info->extent_root); 19802e9f5954SRashika free_root_extent_buffers(info->csum_root); 19812e9f5954SRashika free_root_extent_buffers(info->quota_root); 19822e9f5954SRashika free_root_extent_buffers(info->uuid_root); 19834273eaffSAnand Jain if (free_chunk_root) 19842e9f5954SRashika free_root_extent_buffers(info->chunk_root); 198570f6d82eSOmar Sandoval free_root_extent_buffers(info->free_space_root); 1986af31f5e5SChris Mason } 1987af31f5e5SChris Mason 1988faa2dbf0SJosef Bacik void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 1989171f6537SJosef Bacik { 1990171f6537SJosef Bacik int ret; 1991171f6537SJosef Bacik struct btrfs_root *gang[8]; 1992171f6537SJosef Bacik int i; 1993171f6537SJosef Bacik 1994171f6537SJosef Bacik while (!list_empty(&fs_info->dead_roots)) { 1995171f6537SJosef Bacik gang[0] = list_entry(fs_info->dead_roots.next, 1996171f6537SJosef Bacik struct btrfs_root, root_list); 1997171f6537SJosef Bacik list_del(&gang[0]->root_list); 1998171f6537SJosef Bacik 199927cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) { 2000cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[0]); 2001171f6537SJosef Bacik } else { 2002171f6537SJosef Bacik free_extent_buffer(gang[0]->node); 2003171f6537SJosef Bacik free_extent_buffer(gang[0]->commit_root); 2004b0feb9d9SMiao Xie btrfs_put_fs_root(gang[0]); 2005171f6537SJosef Bacik } 2006171f6537SJosef Bacik } 2007171f6537SJosef Bacik 2008171f6537SJosef Bacik while (1) { 2009171f6537SJosef Bacik ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2010171f6537SJosef Bacik (void **)gang, 0, 2011171f6537SJosef Bacik ARRAY_SIZE(gang)); 2012171f6537SJosef Bacik if (!ret) 2013171f6537SJosef Bacik break; 2014171f6537SJosef Bacik for (i = 0; i < ret; i++) 2015cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2016171f6537SJosef Bacik } 20171a4319ccSLiu Bo 20181a4319ccSLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 20191a4319ccSLiu Bo btrfs_free_log_root_tree(NULL, fs_info); 20202ff7e61eSJeff Mahoney btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents); 20211a4319ccSLiu Bo } 2022171f6537SJosef Bacik } 2023af31f5e5SChris Mason 2024638aa7edSEric Sandeen static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2025638aa7edSEric Sandeen { 2026638aa7edSEric Sandeen mutex_init(&fs_info->scrub_lock); 2027638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_running, 0); 2028638aa7edSEric Sandeen atomic_set(&fs_info->scrub_pause_req, 0); 2029638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_paused, 0); 2030638aa7edSEric Sandeen atomic_set(&fs_info->scrub_cancel_req, 0); 2031638aa7edSEric Sandeen init_waitqueue_head(&fs_info->scrub_pause_wait); 2032ff09c4caSAnand Jain refcount_set(&fs_info->scrub_workers_refcnt, 0); 2033638aa7edSEric Sandeen } 2034638aa7edSEric Sandeen 2035779a65a4SEric Sandeen static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2036779a65a4SEric Sandeen { 2037779a65a4SEric Sandeen spin_lock_init(&fs_info->balance_lock); 2038779a65a4SEric Sandeen mutex_init(&fs_info->balance_mutex); 2039779a65a4SEric Sandeen atomic_set(&fs_info->balance_pause_req, 0); 2040779a65a4SEric Sandeen atomic_set(&fs_info->balance_cancel_req, 0); 2041779a65a4SEric Sandeen fs_info->balance_ctl = NULL; 2042779a65a4SEric Sandeen init_waitqueue_head(&fs_info->balance_wait_q); 2043779a65a4SEric Sandeen } 2044779a65a4SEric Sandeen 20456bccf3abSJeff Mahoney static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info) 2046f37938e0SEric Sandeen { 20472ff7e61eSJeff Mahoney struct inode *inode = fs_info->btree_inode; 20482ff7e61eSJeff Mahoney 20492ff7e61eSJeff Mahoney inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 20502ff7e61eSJeff Mahoney set_nlink(inode, 1); 2051f37938e0SEric Sandeen /* 2052f37938e0SEric Sandeen * we set the i_size on the btree inode to the max possible int. 2053f37938e0SEric Sandeen * the real end of the address space is determined by all of 2054f37938e0SEric Sandeen * the devices in the system 2055f37938e0SEric Sandeen */ 20562ff7e61eSJeff Mahoney inode->i_size = OFFSET_MAX; 20572ff7e61eSJeff Mahoney inode->i_mapping->a_ops = &btree_aops; 2058f37938e0SEric Sandeen 20592ff7e61eSJeff Mahoney RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 206043eb5f29SQu Wenruo extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree, 206143eb5f29SQu Wenruo IO_TREE_INODE_IO, inode); 20627b439738SDavid Sterba BTRFS_I(inode)->io_tree.track_uptodate = false; 20632ff7e61eSJeff Mahoney extent_map_tree_init(&BTRFS_I(inode)->extent_tree); 2064f37938e0SEric Sandeen 20652ff7e61eSJeff Mahoney BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops; 2066f37938e0SEric Sandeen 20672ff7e61eSJeff Mahoney BTRFS_I(inode)->root = fs_info->tree_root; 20682ff7e61eSJeff Mahoney memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key)); 20692ff7e61eSJeff Mahoney set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); 20702ff7e61eSJeff Mahoney btrfs_insert_inode_hash(inode); 2071f37938e0SEric Sandeen } 2072f37938e0SEric Sandeen 2073ad618368SEric Sandeen static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2074ad618368SEric Sandeen { 2075ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2076129827e3SDavid Sterba init_rwsem(&fs_info->dev_replace.rwsem); 20777f8d236aSDavid Sterba init_waitqueue_head(&fs_info->dev_replace.replace_wait); 2078ad618368SEric Sandeen } 2079ad618368SEric Sandeen 2080f9e92e40SEric Sandeen static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2081f9e92e40SEric Sandeen { 2082f9e92e40SEric Sandeen spin_lock_init(&fs_info->qgroup_lock); 2083f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_ioctl_lock); 2084f9e92e40SEric Sandeen fs_info->qgroup_tree = RB_ROOT; 2085f9e92e40SEric Sandeen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2086f9e92e40SEric Sandeen fs_info->qgroup_seq = 1; 2087f9e92e40SEric Sandeen fs_info->qgroup_ulist = NULL; 2088d2c609b8SJeff Mahoney fs_info->qgroup_rescan_running = false; 2089f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_rescan_lock); 2090f9e92e40SEric Sandeen } 2091f9e92e40SEric Sandeen 20922a458198SEric Sandeen static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info, 20932a458198SEric Sandeen struct btrfs_fs_devices *fs_devices) 20942a458198SEric Sandeen { 2095f7b885beSAnand Jain u32 max_active = fs_info->thread_pool_size; 20966f011058SDavid Sterba unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 20972a458198SEric Sandeen 20982a458198SEric Sandeen fs_info->workers = 2099cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "worker", 2100cb001095SJeff Mahoney flags | WQ_HIGHPRI, max_active, 16); 21012a458198SEric Sandeen 21022a458198SEric Sandeen fs_info->delalloc_workers = 2103cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delalloc", 2104cb001095SJeff Mahoney flags, max_active, 2); 21052a458198SEric Sandeen 21062a458198SEric Sandeen fs_info->flush_workers = 2107cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "flush_delalloc", 2108cb001095SJeff Mahoney flags, max_active, 0); 21092a458198SEric Sandeen 21102a458198SEric Sandeen fs_info->caching_workers = 2111cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); 21122a458198SEric Sandeen 21132a458198SEric Sandeen fs_info->fixup_workers = 2114cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); 21152a458198SEric Sandeen 21162a458198SEric Sandeen /* 21172a458198SEric Sandeen * endios are largely parallel and should have a very 21182a458198SEric Sandeen * low idle thresh 21192a458198SEric Sandeen */ 21202a458198SEric Sandeen fs_info->endio_workers = 2121cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4); 21222a458198SEric Sandeen fs_info->endio_meta_workers = 2123cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta", flags, 2124cb001095SJeff Mahoney max_active, 4); 21252a458198SEric Sandeen fs_info->endio_meta_write_workers = 2126cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags, 2127cb001095SJeff Mahoney max_active, 2); 21282a458198SEric Sandeen fs_info->endio_raid56_workers = 2129cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-raid56", flags, 2130cb001095SJeff Mahoney max_active, 4); 21312a458198SEric Sandeen fs_info->endio_repair_workers = 2132cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0); 21332a458198SEric Sandeen fs_info->rmw_workers = 2134cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2); 21352a458198SEric Sandeen fs_info->endio_write_workers = 2136cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-write", flags, 2137cb001095SJeff Mahoney max_active, 2); 21382a458198SEric Sandeen fs_info->endio_freespace_worker = 2139cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "freespace-write", flags, 2140cb001095SJeff Mahoney max_active, 0); 21412a458198SEric Sandeen fs_info->delayed_workers = 2142cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, 2143cb001095SJeff Mahoney max_active, 0); 21442a458198SEric Sandeen fs_info->readahead_workers = 2145cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "readahead", flags, 2146cb001095SJeff Mahoney max_active, 2); 21472a458198SEric Sandeen fs_info->qgroup_rescan_workers = 2148cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); 2149b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers = 2150b0643e59SDennis Zhou alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1); 21512a458198SEric Sandeen 21522a458198SEric Sandeen if (!(fs_info->workers && fs_info->delalloc_workers && 2153ba8a9d07SChris Mason fs_info->flush_workers && 21542a458198SEric Sandeen fs_info->endio_workers && fs_info->endio_meta_workers && 21552a458198SEric Sandeen fs_info->endio_meta_write_workers && 21562a458198SEric Sandeen fs_info->endio_repair_workers && 21572a458198SEric Sandeen fs_info->endio_write_workers && fs_info->endio_raid56_workers && 21582a458198SEric Sandeen fs_info->endio_freespace_worker && fs_info->rmw_workers && 21592a458198SEric Sandeen fs_info->caching_workers && fs_info->readahead_workers && 21602a458198SEric Sandeen fs_info->fixup_workers && fs_info->delayed_workers && 2161b0643e59SDennis Zhou fs_info->qgroup_rescan_workers && 2162b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers)) { 21632a458198SEric Sandeen return -ENOMEM; 21642a458198SEric Sandeen } 21652a458198SEric Sandeen 21662a458198SEric Sandeen return 0; 21672a458198SEric Sandeen } 21682a458198SEric Sandeen 21696d97c6e3SJohannes Thumshirn static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type) 21706d97c6e3SJohannes Thumshirn { 21716d97c6e3SJohannes Thumshirn struct crypto_shash *csum_shash; 2172b4e967beSDavid Sterba const char *csum_driver = btrfs_super_csum_driver(csum_type); 21736d97c6e3SJohannes Thumshirn 2174b4e967beSDavid Sterba csum_shash = crypto_alloc_shash(csum_driver, 0, 0); 21756d97c6e3SJohannes Thumshirn 21766d97c6e3SJohannes Thumshirn if (IS_ERR(csum_shash)) { 21776d97c6e3SJohannes Thumshirn btrfs_err(fs_info, "error allocating %s hash for checksum", 2178b4e967beSDavid Sterba csum_driver); 21796d97c6e3SJohannes Thumshirn return PTR_ERR(csum_shash); 21806d97c6e3SJohannes Thumshirn } 21816d97c6e3SJohannes Thumshirn 21826d97c6e3SJohannes Thumshirn fs_info->csum_shash = csum_shash; 21836d97c6e3SJohannes Thumshirn 21846d97c6e3SJohannes Thumshirn return 0; 21856d97c6e3SJohannes Thumshirn } 21866d97c6e3SJohannes Thumshirn 21876d97c6e3SJohannes Thumshirn static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info) 21886d97c6e3SJohannes Thumshirn { 21896d97c6e3SJohannes Thumshirn crypto_free_shash(fs_info->csum_shash); 21906d97c6e3SJohannes Thumshirn } 21916d97c6e3SJohannes Thumshirn 219263443bf5SEric Sandeen static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 219363443bf5SEric Sandeen struct btrfs_fs_devices *fs_devices) 219463443bf5SEric Sandeen { 219563443bf5SEric Sandeen int ret; 219663443bf5SEric Sandeen struct btrfs_root *log_tree_root; 219763443bf5SEric Sandeen struct btrfs_super_block *disk_super = fs_info->super_copy; 219863443bf5SEric Sandeen u64 bytenr = btrfs_super_log_root(disk_super); 2199581c1760SQu Wenruo int level = btrfs_super_log_root_level(disk_super); 220063443bf5SEric Sandeen 220163443bf5SEric Sandeen if (fs_devices->rw_devices == 0) { 2202f14d104dSDavid Sterba btrfs_warn(fs_info, "log replay required on RO media"); 220363443bf5SEric Sandeen return -EIO; 220463443bf5SEric Sandeen } 220563443bf5SEric Sandeen 220696dfcb46SJosef Bacik log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, 220796dfcb46SJosef Bacik GFP_KERNEL); 220863443bf5SEric Sandeen if (!log_tree_root) 220963443bf5SEric Sandeen return -ENOMEM; 221063443bf5SEric Sandeen 22112ff7e61eSJeff Mahoney log_tree_root->node = read_tree_block(fs_info, bytenr, 2212581c1760SQu Wenruo fs_info->generation + 1, 2213581c1760SQu Wenruo level, NULL); 221464c043deSLiu Bo if (IS_ERR(log_tree_root->node)) { 2215f14d104dSDavid Sterba btrfs_warn(fs_info, "failed to read log tree"); 22160eeff236SLiu Bo ret = PTR_ERR(log_tree_root->node); 221764c043deSLiu Bo kfree(log_tree_root); 22180eeff236SLiu Bo return ret; 221964c043deSLiu Bo } else if (!extent_buffer_uptodate(log_tree_root->node)) { 2220f14d104dSDavid Sterba btrfs_err(fs_info, "failed to read log tree"); 222163443bf5SEric Sandeen free_extent_buffer(log_tree_root->node); 222263443bf5SEric Sandeen kfree(log_tree_root); 222363443bf5SEric Sandeen return -EIO; 222463443bf5SEric Sandeen } 222563443bf5SEric Sandeen /* returns with log_tree_root freed on success */ 222663443bf5SEric Sandeen ret = btrfs_recover_log_trees(log_tree_root); 222763443bf5SEric Sandeen if (ret) { 22280b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 222963443bf5SEric Sandeen "Failed to recover log tree"); 223063443bf5SEric Sandeen free_extent_buffer(log_tree_root->node); 223163443bf5SEric Sandeen kfree(log_tree_root); 223263443bf5SEric Sandeen return ret; 223363443bf5SEric Sandeen } 223463443bf5SEric Sandeen 2235bc98a42cSDavid Howells if (sb_rdonly(fs_info->sb)) { 22366bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 223763443bf5SEric Sandeen if (ret) 223863443bf5SEric Sandeen return ret; 223963443bf5SEric Sandeen } 224063443bf5SEric Sandeen 224163443bf5SEric Sandeen return 0; 224263443bf5SEric Sandeen } 224363443bf5SEric Sandeen 22446bccf3abSJeff Mahoney static int btrfs_read_roots(struct btrfs_fs_info *fs_info) 22454bbcaa64SEric Sandeen { 22466bccf3abSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 2247a4f3d2c4SDavid Sterba struct btrfs_root *root; 22484bbcaa64SEric Sandeen struct btrfs_key location; 22494bbcaa64SEric Sandeen int ret; 22504bbcaa64SEric Sandeen 22516bccf3abSJeff Mahoney BUG_ON(!fs_info->tree_root); 22526bccf3abSJeff Mahoney 22534bbcaa64SEric Sandeen location.objectid = BTRFS_EXTENT_TREE_OBJECTID; 22544bbcaa64SEric Sandeen location.type = BTRFS_ROOT_ITEM_KEY; 22554bbcaa64SEric Sandeen location.offset = 0; 22564bbcaa64SEric Sandeen 2257a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2258f50f4353SLiu Bo if (IS_ERR(root)) { 2259f50f4353SLiu Bo ret = PTR_ERR(root); 2260f50f4353SLiu Bo goto out; 2261f50f4353SLiu Bo } 2262a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2263a4f3d2c4SDavid Sterba fs_info->extent_root = root; 22644bbcaa64SEric Sandeen 22654bbcaa64SEric Sandeen location.objectid = BTRFS_DEV_TREE_OBJECTID; 2266a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2267f50f4353SLiu Bo if (IS_ERR(root)) { 2268f50f4353SLiu Bo ret = PTR_ERR(root); 2269f50f4353SLiu Bo goto out; 2270f50f4353SLiu Bo } 2271a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2272a4f3d2c4SDavid Sterba fs_info->dev_root = root; 22734bbcaa64SEric Sandeen btrfs_init_devices_late(fs_info); 22744bbcaa64SEric Sandeen 22754bbcaa64SEric Sandeen location.objectid = BTRFS_CSUM_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->csum_root = root; 22834bbcaa64SEric Sandeen 22844bbcaa64SEric Sandeen location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2285a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2286a4f3d2c4SDavid Sterba if (!IS_ERR(root)) { 2287a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2288afcdd129SJosef Bacik set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); 2289a4f3d2c4SDavid Sterba fs_info->quota_root = root; 22904bbcaa64SEric Sandeen } 22914bbcaa64SEric Sandeen 22924bbcaa64SEric Sandeen location.objectid = BTRFS_UUID_TREE_OBJECTID; 2293a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2294a4f3d2c4SDavid Sterba if (IS_ERR(root)) { 2295a4f3d2c4SDavid Sterba ret = PTR_ERR(root); 22964bbcaa64SEric Sandeen if (ret != -ENOENT) 2297f50f4353SLiu Bo goto out; 22984bbcaa64SEric Sandeen } else { 2299a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2300a4f3d2c4SDavid Sterba fs_info->uuid_root = root; 23014bbcaa64SEric Sandeen } 23024bbcaa64SEric Sandeen 230370f6d82eSOmar Sandoval if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 230470f6d82eSOmar Sandoval location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID; 230570f6d82eSOmar Sandoval root = btrfs_read_tree_root(tree_root, &location); 2306f50f4353SLiu Bo if (IS_ERR(root)) { 2307f50f4353SLiu Bo ret = PTR_ERR(root); 2308f50f4353SLiu Bo goto out; 2309f50f4353SLiu Bo } 231070f6d82eSOmar Sandoval set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 231170f6d82eSOmar Sandoval fs_info->free_space_root = root; 231270f6d82eSOmar Sandoval } 231370f6d82eSOmar Sandoval 23144bbcaa64SEric Sandeen return 0; 2315f50f4353SLiu Bo out: 2316f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", 2317f50f4353SLiu Bo location.objectid, ret); 2318f50f4353SLiu Bo return ret; 23194bbcaa64SEric Sandeen } 23204bbcaa64SEric Sandeen 2321069ec957SQu Wenruo /* 2322069ec957SQu Wenruo * Real super block validation 2323069ec957SQu Wenruo * NOTE: super csum type and incompat features will not be checked here. 2324069ec957SQu Wenruo * 2325069ec957SQu Wenruo * @sb: super block to check 2326069ec957SQu Wenruo * @mirror_num: the super block number to check its bytenr: 2327069ec957SQu Wenruo * 0 the primary (1st) sb 2328069ec957SQu Wenruo * 1, 2 2nd and 3rd backup copy 2329069ec957SQu Wenruo * -1 skip bytenr check 2330069ec957SQu Wenruo */ 2331069ec957SQu Wenruo static int validate_super(struct btrfs_fs_info *fs_info, 2332069ec957SQu Wenruo struct btrfs_super_block *sb, int mirror_num) 233321a852b0SQu Wenruo { 233421a852b0SQu Wenruo u64 nodesize = btrfs_super_nodesize(sb); 233521a852b0SQu Wenruo u64 sectorsize = btrfs_super_sectorsize(sb); 233621a852b0SQu Wenruo int ret = 0; 233721a852b0SQu Wenruo 233821a852b0SQu Wenruo if (btrfs_super_magic(sb) != BTRFS_MAGIC) { 233921a852b0SQu Wenruo btrfs_err(fs_info, "no valid FS found"); 234021a852b0SQu Wenruo ret = -EINVAL; 234121a852b0SQu Wenruo } 234221a852b0SQu Wenruo if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { 234321a852b0SQu Wenruo btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", 234421a852b0SQu Wenruo btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); 234521a852b0SQu Wenruo ret = -EINVAL; 234621a852b0SQu Wenruo } 234721a852b0SQu Wenruo if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 234821a852b0SQu Wenruo btrfs_err(fs_info, "tree_root level too big: %d >= %d", 234921a852b0SQu Wenruo btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 235021a852b0SQu Wenruo ret = -EINVAL; 235121a852b0SQu Wenruo } 235221a852b0SQu Wenruo if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 235321a852b0SQu Wenruo btrfs_err(fs_info, "chunk_root level too big: %d >= %d", 235421a852b0SQu Wenruo btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 235521a852b0SQu Wenruo ret = -EINVAL; 235621a852b0SQu Wenruo } 235721a852b0SQu Wenruo if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 235821a852b0SQu Wenruo btrfs_err(fs_info, "log_root level too big: %d >= %d", 235921a852b0SQu Wenruo btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 236021a852b0SQu Wenruo ret = -EINVAL; 236121a852b0SQu Wenruo } 236221a852b0SQu Wenruo 236321a852b0SQu Wenruo /* 236421a852b0SQu Wenruo * Check sectorsize and nodesize first, other check will need it. 236521a852b0SQu Wenruo * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. 236621a852b0SQu Wenruo */ 236721a852b0SQu Wenruo if (!is_power_of_2(sectorsize) || sectorsize < 4096 || 236821a852b0SQu Wenruo sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { 236921a852b0SQu Wenruo btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); 237021a852b0SQu Wenruo ret = -EINVAL; 237121a852b0SQu Wenruo } 237221a852b0SQu Wenruo /* Only PAGE SIZE is supported yet */ 237321a852b0SQu Wenruo if (sectorsize != PAGE_SIZE) { 237421a852b0SQu Wenruo btrfs_err(fs_info, 237521a852b0SQu Wenruo "sectorsize %llu not supported yet, only support %lu", 237621a852b0SQu Wenruo sectorsize, PAGE_SIZE); 237721a852b0SQu Wenruo ret = -EINVAL; 237821a852b0SQu Wenruo } 237921a852b0SQu Wenruo if (!is_power_of_2(nodesize) || nodesize < sectorsize || 238021a852b0SQu Wenruo nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { 238121a852b0SQu Wenruo btrfs_err(fs_info, "invalid nodesize %llu", nodesize); 238221a852b0SQu Wenruo ret = -EINVAL; 238321a852b0SQu Wenruo } 238421a852b0SQu Wenruo if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { 238521a852b0SQu Wenruo btrfs_err(fs_info, "invalid leafsize %u, should be %llu", 238621a852b0SQu Wenruo le32_to_cpu(sb->__unused_leafsize), nodesize); 238721a852b0SQu Wenruo ret = -EINVAL; 238821a852b0SQu Wenruo } 238921a852b0SQu Wenruo 239021a852b0SQu Wenruo /* Root alignment check */ 239121a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { 239221a852b0SQu Wenruo btrfs_warn(fs_info, "tree_root block unaligned: %llu", 239321a852b0SQu Wenruo btrfs_super_root(sb)); 239421a852b0SQu Wenruo ret = -EINVAL; 239521a852b0SQu Wenruo } 239621a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { 239721a852b0SQu Wenruo btrfs_warn(fs_info, "chunk_root block unaligned: %llu", 239821a852b0SQu Wenruo btrfs_super_chunk_root(sb)); 239921a852b0SQu Wenruo ret = -EINVAL; 240021a852b0SQu Wenruo } 240121a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { 240221a852b0SQu Wenruo btrfs_warn(fs_info, "log_root block unaligned: %llu", 240321a852b0SQu Wenruo btrfs_super_log_root(sb)); 240421a852b0SQu Wenruo ret = -EINVAL; 240521a852b0SQu Wenruo } 240621a852b0SQu Wenruo 2407de37aa51SNikolay Borisov if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, 24087239ff4bSNikolay Borisov BTRFS_FSID_SIZE) != 0) { 240921a852b0SQu Wenruo btrfs_err(fs_info, 24107239ff4bSNikolay Borisov "dev_item UUID does not match metadata fsid: %pU != %pU", 2411de37aa51SNikolay Borisov fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); 241221a852b0SQu Wenruo ret = -EINVAL; 241321a852b0SQu Wenruo } 241421a852b0SQu Wenruo 241521a852b0SQu Wenruo /* 241621a852b0SQu Wenruo * Hint to catch really bogus numbers, bitflips or so, more exact checks are 241721a852b0SQu Wenruo * done later 241821a852b0SQu Wenruo */ 241921a852b0SQu Wenruo if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { 242021a852b0SQu Wenruo btrfs_err(fs_info, "bytes_used is too small %llu", 242121a852b0SQu Wenruo btrfs_super_bytes_used(sb)); 242221a852b0SQu Wenruo ret = -EINVAL; 242321a852b0SQu Wenruo } 242421a852b0SQu Wenruo if (!is_power_of_2(btrfs_super_stripesize(sb))) { 242521a852b0SQu Wenruo btrfs_err(fs_info, "invalid stripesize %u", 242621a852b0SQu Wenruo btrfs_super_stripesize(sb)); 242721a852b0SQu Wenruo ret = -EINVAL; 242821a852b0SQu Wenruo } 242921a852b0SQu Wenruo if (btrfs_super_num_devices(sb) > (1UL << 31)) 243021a852b0SQu Wenruo btrfs_warn(fs_info, "suspicious number of devices: %llu", 243121a852b0SQu Wenruo btrfs_super_num_devices(sb)); 243221a852b0SQu Wenruo if (btrfs_super_num_devices(sb) == 0) { 243321a852b0SQu Wenruo btrfs_err(fs_info, "number of devices is 0"); 243421a852b0SQu Wenruo ret = -EINVAL; 243521a852b0SQu Wenruo } 243621a852b0SQu Wenruo 2437069ec957SQu Wenruo if (mirror_num >= 0 && 2438069ec957SQu Wenruo btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { 243921a852b0SQu Wenruo btrfs_err(fs_info, "super offset mismatch %llu != %u", 244021a852b0SQu Wenruo btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 244121a852b0SQu Wenruo ret = -EINVAL; 244221a852b0SQu Wenruo } 244321a852b0SQu Wenruo 244421a852b0SQu Wenruo /* 244521a852b0SQu Wenruo * Obvious sys_chunk_array corruptions, it must hold at least one key 244621a852b0SQu Wenruo * and one chunk 244721a852b0SQu Wenruo */ 244821a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 244921a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too big %u > %u", 245021a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 245121a852b0SQu Wenruo BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 245221a852b0SQu Wenruo ret = -EINVAL; 245321a852b0SQu Wenruo } 245421a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 245521a852b0SQu Wenruo + sizeof(struct btrfs_chunk)) { 245621a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too small %u < %zu", 245721a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 245821a852b0SQu Wenruo sizeof(struct btrfs_disk_key) 245921a852b0SQu Wenruo + sizeof(struct btrfs_chunk)); 246021a852b0SQu Wenruo ret = -EINVAL; 246121a852b0SQu Wenruo } 246221a852b0SQu Wenruo 246321a852b0SQu Wenruo /* 246421a852b0SQu Wenruo * The generation is a global counter, we'll trust it more than the others 246521a852b0SQu Wenruo * but it's still possible that it's the one that's wrong. 246621a852b0SQu Wenruo */ 246721a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 246821a852b0SQu Wenruo btrfs_warn(fs_info, 246921a852b0SQu Wenruo "suspicious: generation < chunk_root_generation: %llu < %llu", 247021a852b0SQu Wenruo btrfs_super_generation(sb), 247121a852b0SQu Wenruo btrfs_super_chunk_root_generation(sb)); 247221a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 247321a852b0SQu Wenruo && btrfs_super_cache_generation(sb) != (u64)-1) 247421a852b0SQu Wenruo btrfs_warn(fs_info, 247521a852b0SQu Wenruo "suspicious: generation < cache_generation: %llu < %llu", 247621a852b0SQu Wenruo btrfs_super_generation(sb), 247721a852b0SQu Wenruo btrfs_super_cache_generation(sb)); 247821a852b0SQu Wenruo 247921a852b0SQu Wenruo return ret; 248021a852b0SQu Wenruo } 248121a852b0SQu Wenruo 2482069ec957SQu Wenruo /* 2483069ec957SQu Wenruo * Validation of super block at mount time. 2484069ec957SQu Wenruo * Some checks already done early at mount time, like csum type and incompat 2485069ec957SQu Wenruo * flags will be skipped. 2486069ec957SQu Wenruo */ 2487069ec957SQu Wenruo static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) 2488069ec957SQu Wenruo { 2489069ec957SQu Wenruo return validate_super(fs_info, fs_info->super_copy, 0); 2490069ec957SQu Wenruo } 2491069ec957SQu Wenruo 249275cb857dSQu Wenruo /* 249375cb857dSQu Wenruo * Validation of super block at write time. 249475cb857dSQu Wenruo * Some checks like bytenr check will be skipped as their values will be 249575cb857dSQu Wenruo * overwritten soon. 249675cb857dSQu Wenruo * Extra checks like csum type and incompat flags will be done here. 249775cb857dSQu Wenruo */ 249875cb857dSQu Wenruo static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, 249975cb857dSQu Wenruo struct btrfs_super_block *sb) 250075cb857dSQu Wenruo { 250175cb857dSQu Wenruo int ret; 250275cb857dSQu Wenruo 250375cb857dSQu Wenruo ret = validate_super(fs_info, sb, -1); 250475cb857dSQu Wenruo if (ret < 0) 250575cb857dSQu Wenruo goto out; 2506e7e16f48SJohannes Thumshirn if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) { 250775cb857dSQu Wenruo ret = -EUCLEAN; 250875cb857dSQu Wenruo btrfs_err(fs_info, "invalid csum type, has %u want %u", 250975cb857dSQu Wenruo btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); 251075cb857dSQu Wenruo goto out; 251175cb857dSQu Wenruo } 251275cb857dSQu Wenruo if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 251375cb857dSQu Wenruo ret = -EUCLEAN; 251475cb857dSQu Wenruo btrfs_err(fs_info, 251575cb857dSQu Wenruo "invalid incompat flags, has 0x%llx valid mask 0x%llx", 251675cb857dSQu Wenruo btrfs_super_incompat_flags(sb), 251775cb857dSQu Wenruo (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); 251875cb857dSQu Wenruo goto out; 251975cb857dSQu Wenruo } 252075cb857dSQu Wenruo out: 252175cb857dSQu Wenruo if (ret < 0) 252275cb857dSQu Wenruo btrfs_err(fs_info, 252375cb857dSQu Wenruo "super block corruption detected before writing it to disk"); 252475cb857dSQu Wenruo return ret; 252575cb857dSQu Wenruo } 252675cb857dSQu Wenruo 25276ef108ddSNikolay Borisov static int __cold init_tree_roots(struct btrfs_fs_info *fs_info) 2528b8522a1eSNikolay Borisov { 25296ef108ddSNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 2530b8522a1eSNikolay Borisov struct btrfs_super_block *sb = fs_info->super_copy; 2531b8522a1eSNikolay Borisov struct btrfs_root *tree_root = fs_info->tree_root; 2532b8522a1eSNikolay Borisov bool handle_error = false; 2533b8522a1eSNikolay Borisov int ret = 0; 2534b8522a1eSNikolay Borisov int i; 2535b8522a1eSNikolay Borisov 2536b8522a1eSNikolay Borisov for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 2537b8522a1eSNikolay Borisov u64 generation; 2538b8522a1eSNikolay Borisov int level; 2539b8522a1eSNikolay Borisov 2540b8522a1eSNikolay Borisov if (handle_error) { 2541b8522a1eSNikolay Borisov if (!IS_ERR(tree_root->node)) 2542b8522a1eSNikolay Borisov free_extent_buffer(tree_root->node); 2543b8522a1eSNikolay Borisov tree_root->node = NULL; 2544b8522a1eSNikolay Borisov 2545b8522a1eSNikolay Borisov if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) 2546b8522a1eSNikolay Borisov break; 2547b8522a1eSNikolay Borisov 2548b8522a1eSNikolay Borisov free_root_pointers(fs_info, 0); 2549b8522a1eSNikolay Borisov 2550b8522a1eSNikolay Borisov /* 2551b8522a1eSNikolay Borisov * Don't use the log in recovery mode, it won't be 2552b8522a1eSNikolay Borisov * valid 2553b8522a1eSNikolay Borisov */ 2554b8522a1eSNikolay Borisov btrfs_set_super_log_root(sb, 0); 2555b8522a1eSNikolay Borisov 2556b8522a1eSNikolay Borisov /* We can't trust the free space cache either */ 2557b8522a1eSNikolay Borisov btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 2558b8522a1eSNikolay Borisov 2559b8522a1eSNikolay Borisov ret = read_backup_root(fs_info, i); 25606ef108ddSNikolay Borisov backup_index = ret; 2561b8522a1eSNikolay Borisov if (ret < 0) 2562b8522a1eSNikolay Borisov return ret; 2563b8522a1eSNikolay Borisov } 2564b8522a1eSNikolay Borisov generation = btrfs_super_generation(sb); 2565b8522a1eSNikolay Borisov level = btrfs_super_root_level(sb); 2566b8522a1eSNikolay Borisov tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb), 2567b8522a1eSNikolay Borisov generation, level, NULL); 2568b8522a1eSNikolay Borisov if (IS_ERR(tree_root->node) || 2569b8522a1eSNikolay Borisov !extent_buffer_uptodate(tree_root->node)) { 2570b8522a1eSNikolay Borisov handle_error = true; 2571b8522a1eSNikolay Borisov 2572b8522a1eSNikolay Borisov if (IS_ERR(tree_root->node)) 2573b8522a1eSNikolay Borisov ret = PTR_ERR(tree_root->node); 2574b8522a1eSNikolay Borisov else if (!extent_buffer_uptodate(tree_root->node)) 2575b8522a1eSNikolay Borisov ret = -EUCLEAN; 2576b8522a1eSNikolay Borisov 2577b8522a1eSNikolay Borisov btrfs_warn(fs_info, "failed to read tree root"); 2578b8522a1eSNikolay Borisov continue; 2579b8522a1eSNikolay Borisov } 2580b8522a1eSNikolay Borisov 2581b8522a1eSNikolay Borisov btrfs_set_root_node(&tree_root->root_item, tree_root->node); 2582b8522a1eSNikolay Borisov tree_root->commit_root = btrfs_root_node(tree_root); 2583b8522a1eSNikolay Borisov btrfs_set_root_refs(&tree_root->root_item, 1); 2584b8522a1eSNikolay Borisov 2585336a0d8dSNikolay Borisov /* 2586336a0d8dSNikolay Borisov * No need to hold btrfs_root::objectid_mutex since the fs 2587336a0d8dSNikolay Borisov * hasn't been fully initialised and we are the only user 2588336a0d8dSNikolay Borisov */ 2589b8522a1eSNikolay Borisov ret = btrfs_find_highest_objectid(tree_root, 2590b8522a1eSNikolay Borisov &tree_root->highest_objectid); 2591b8522a1eSNikolay Borisov if (ret < 0) { 2592b8522a1eSNikolay Borisov handle_error = true; 2593b8522a1eSNikolay Borisov continue; 2594b8522a1eSNikolay Borisov } 2595b8522a1eSNikolay Borisov 2596b8522a1eSNikolay Borisov ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 2597b8522a1eSNikolay Borisov 2598b8522a1eSNikolay Borisov ret = btrfs_read_roots(fs_info); 2599b8522a1eSNikolay Borisov if (ret < 0) { 2600b8522a1eSNikolay Borisov handle_error = true; 2601b8522a1eSNikolay Borisov continue; 2602b8522a1eSNikolay Borisov } 2603b8522a1eSNikolay Borisov 2604b8522a1eSNikolay Borisov /* All successful */ 2605b8522a1eSNikolay Borisov fs_info->generation = generation; 2606b8522a1eSNikolay Borisov fs_info->last_trans_committed = generation; 26076ef108ddSNikolay Borisov 26086ef108ddSNikolay Borisov /* Always begin writing backup roots after the one being used */ 26096ef108ddSNikolay Borisov if (backup_index < 0) { 26106ef108ddSNikolay Borisov fs_info->backup_root_index = 0; 26116ef108ddSNikolay Borisov } else { 26126ef108ddSNikolay Borisov fs_info->backup_root_index = backup_index + 1; 26136ef108ddSNikolay Borisov fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS; 26146ef108ddSNikolay Borisov } 2615b8522a1eSNikolay Borisov break; 2616b8522a1eSNikolay Borisov } 2617b8522a1eSNikolay Borisov 2618b8522a1eSNikolay Borisov return ret; 2619b8522a1eSNikolay Borisov } 2620b8522a1eSNikolay Borisov 2621b105e927SDavid Sterba int __cold open_ctree(struct super_block *sb, 2622dfe25020SChris Mason struct btrfs_fs_devices *fs_devices, 2623dfe25020SChris Mason char *options) 2624eb60ceacSChris Mason { 2625db94535dSChris Mason u32 sectorsize; 2626db94535dSChris Mason u32 nodesize; 262787ee04ebSChris Mason u32 stripesize; 262884234f3aSYan Zheng u64 generation; 2629f2b636e8SJosef Bacik u64 features; 263051bce6c9SJohannes Thumshirn u16 csum_type; 26313de4586cSChris Mason struct btrfs_key location; 2632a061fc8dSChris Mason struct buffer_head *bh; 26334d34b278SIlya Dryomov struct btrfs_super_block *disk_super; 2634815745cfSAl Viro struct btrfs_fs_info *fs_info = btrfs_sb(sb); 2635f84a8bd6SAl Viro struct btrfs_root *tree_root; 26364d34b278SIlya Dryomov struct btrfs_root *chunk_root; 2637eb60ceacSChris Mason int ret; 2638e58ca020SYan int err = -EINVAL; 26396675df31SOmar Sandoval int clear_free_space_tree = 0; 2640581c1760SQu Wenruo int level; 26414543df7eSChris Mason 264296dfcb46SJosef Bacik tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, 264396dfcb46SJosef Bacik GFP_KERNEL); 264496dfcb46SJosef Bacik fs_info->tree_root = tree_root; 264596dfcb46SJosef Bacik chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID, 264696dfcb46SJosef Bacik GFP_KERNEL); 264796dfcb46SJosef Bacik fs_info->chunk_root = chunk_root; 2648cb517eabSMiao Xie if (!tree_root || !chunk_root) { 264939279cc3SChris Mason err = -ENOMEM; 265039279cc3SChris Mason goto fail; 265139279cc3SChris Mason } 265276dda93cSYan, Zheng 265376dda93cSYan, Zheng ret = init_srcu_struct(&fs_info->subvol_srcu); 265476dda93cSYan, Zheng if (ret) { 265576dda93cSYan, Zheng err = ret; 265676dda93cSYan, Zheng goto fail; 265776dda93cSYan, Zheng } 265876dda93cSYan, Zheng 26594297ff84SJosef Bacik ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL); 2660e2d84521SMiao Xie if (ret) { 2661e2d84521SMiao Xie err = ret; 26629e11ceeeSJan Kara goto fail_srcu; 2663e2d84521SMiao Xie } 26644297ff84SJosef Bacik 26654297ff84SJosef Bacik ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 26664297ff84SJosef Bacik if (ret) { 26674297ff84SJosef Bacik err = ret; 26684297ff84SJosef Bacik goto fail_dio_bytes; 26694297ff84SJosef Bacik } 267009cbfeafSKirill A. Shutemov fs_info->dirty_metadata_batch = PAGE_SIZE * 2671e2d84521SMiao Xie (1 + ilog2(nr_cpu_ids)); 2672e2d84521SMiao Xie 2673908c7f19STejun Heo ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 2674963d678bSMiao Xie if (ret) { 2675963d678bSMiao Xie err = ret; 2676963d678bSMiao Xie goto fail_dirty_metadata_bytes; 2677963d678bSMiao Xie } 2678963d678bSMiao Xie 26797f8d236aSDavid Sterba ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, 26807f8d236aSDavid Sterba GFP_KERNEL); 2681c404e0dcSMiao Xie if (ret) { 2682c404e0dcSMiao Xie err = ret; 2683c404e0dcSMiao Xie goto fail_delalloc_bytes; 2684c404e0dcSMiao Xie } 2685c404e0dcSMiao Xie 268676dda93cSYan, Zheng INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 2687f28491e0SJosef Bacik INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 26888fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 2689facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 269024bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 2691eb73c1b7SMiao Xie INIT_LIST_HEAD(&fs_info->delalloc_roots); 269211833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 2693eb73c1b7SMiao Xie spin_lock_init(&fs_info->delalloc_root_lock); 2694a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 269576dda93cSYan, Zheng spin_lock_init(&fs_info->fs_roots_radix_lock); 269624bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 26974cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 2698ceda0864SMiao Xie spin_lock_init(&fs_info->super_lock); 2699f28491e0SJosef Bacik spin_lock_init(&fs_info->buffer_lock); 270047ab2a6cSJosef Bacik spin_lock_init(&fs_info->unused_bgs_lock); 2701f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 2702d7c15171SZhao Lei mutex_init(&fs_info->unused_bg_unpin_mutex); 270367c5e7d4SFilipe Manana mutex_init(&fs_info->delete_unused_bgs_mutex); 27047585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 2705573bfb72SMiao Xie mutex_init(&fs_info->delalloc_root_mutex); 2706de98ced9SMiao Xie seqlock_init(&fs_info->profiles_lock); 270719c00ddcSChris Mason 27080b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 27096324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 2710f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 271147ab2a6cSJosef Bacik INIT_LIST_HEAD(&fs_info->unused_bgs); 2712c8bf1b67SDavid Sterba extent_map_tree_init(&fs_info->mapping_tree); 271366d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 271466d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 271566d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 271666d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 271766d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 271866d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 271966d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 2720ba2c4d4eSJosef Bacik btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, 2721ba2c4d4eSJosef Bacik BTRFS_BLOCK_RSV_DELREFS); 2722ba2c4d4eSJosef Bacik 2723771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 27244cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 27252fefd558SZhao Lei atomic_set(&fs_info->reada_works_cnt, 0); 2726034f784dSJosef Bacik atomic_set(&fs_info->nr_delayed_iputs, 0); 2727fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 2728e20d96d6SChris Mason fs_info->sb = sb; 272995ac567aSFilipe David Borba Manana fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 27309ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 27314cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 2732a5ed45f8SNikolay Borisov atomic64_set(&fs_info->free_chunk_space, 0); 2733f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 27348b87dc17SDavid Sterba fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 2735f8c269d7SDavid Sterba fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 273690519d66SArne Jansen /* readahead state */ 2737d0164adcSMel Gorman INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); 273890519d66SArne Jansen spin_lock_init(&fs_info->reada_lock); 2739fd708b81SJosef Bacik btrfs_init_ref_verify(fs_info); 2740c8b97818SChris Mason 2741b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 2742b34b086cSChris Mason num_online_cpus() + 2, 8); 27430afbaf8cSChris Mason 2744199c2a9cSMiao Xie INIT_LIST_HEAD(&fs_info->ordered_roots); 2745199c2a9cSMiao Xie spin_lock_init(&fs_info->ordered_root_lock); 274669fe2d75SJosef Bacik 274769fe2d75SJosef Bacik fs_info->btree_inode = new_inode(sb); 274869fe2d75SJosef Bacik if (!fs_info->btree_inode) { 274969fe2d75SJosef Bacik err = -ENOMEM; 275069fe2d75SJosef Bacik goto fail_bio_counter; 275169fe2d75SJosef Bacik } 275269fe2d75SJosef Bacik mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 275369fe2d75SJosef Bacik 275416cdcec7SMiao Xie fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 275574e4d827SDavid Sterba GFP_KERNEL); 275616cdcec7SMiao Xie if (!fs_info->delayed_root) { 275716cdcec7SMiao Xie err = -ENOMEM; 275816cdcec7SMiao Xie goto fail_iput; 275916cdcec7SMiao Xie } 276016cdcec7SMiao Xie btrfs_init_delayed_root(fs_info->delayed_root); 27613eaa2885SChris Mason 2762638aa7edSEric Sandeen btrfs_init_scrub(fs_info); 276321adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 276421adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 276521adbd5cSStefan Behrens #endif 2766779a65a4SEric Sandeen btrfs_init_balance(fs_info); 276721c7e756SMiao Xie btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work); 2768a2de733cSArne Jansen 27699f6d2510SDavid Sterba sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; 27709f6d2510SDavid Sterba sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); 2771a061fc8dSChris Mason 27726bccf3abSJeff Mahoney btrfs_init_btree_inode(fs_info); 277339279cc3SChris Mason 2774e02119d5SChris Mason spin_lock_init(&fs_info->block_group_cache_lock); 27756bef4d31SEric Paris fs_info->block_group_cache_tree = RB_ROOT; 2776a1897fddSLiu Bo fs_info->first_logical_byte = (u64)-1; 2777925baeddSChris Mason 277843eb5f29SQu Wenruo extent_io_tree_init(fs_info, &fs_info->freed_extents[0], 277943eb5f29SQu Wenruo IO_TREE_FS_INFO_FREED_EXTENTS0, NULL); 278043eb5f29SQu Wenruo extent_io_tree_init(fs_info, &fs_info->freed_extents[1], 278143eb5f29SQu Wenruo IO_TREE_FS_INFO_FREED_EXTENTS1, NULL); 278211833d66SYan Zheng fs_info->pinned_extents = &fs_info->freed_extents[0]; 2783afcdd129SJosef Bacik set_bit(BTRFS_FS_BARRIER, &fs_info->flags); 2784509659cdSChris Mason 27855a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 278679154b1bSChris Mason mutex_init(&fs_info->tree_log_mutex); 2787925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 2788a74a4b97SChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 2789a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 27901bbc621eSChris Mason mutex_init(&fs_info->ro_block_group_mutex); 27919e351cc8SJosef Bacik init_rwsem(&fs_info->commit_root_sem); 2792c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 279376dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 2794803b2f54SStefan Behrens sema_init(&fs_info->uuid_tree_rescan_sem, 1); 2795fa9c0d79SChris Mason 2796ad618368SEric Sandeen btrfs_init_dev_replace_locks(fs_info); 2797f9e92e40SEric Sandeen btrfs_init_qgroup(fs_info); 2798b0643e59SDennis Zhou btrfs_discard_init(fs_info); 2799416ac51dSArne Jansen 2800fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2801fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2802fa9c0d79SChris Mason 2803e6dcd2dcSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 2804f9295749SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 2805bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 28064854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 2807034f784dSJosef Bacik init_waitqueue_head(&fs_info->delayed_iputs_wait); 28083768f368SChris Mason 2809da17066cSJeff Mahoney /* Usable values until the real ones are cached from the superblock */ 2810da17066cSJeff Mahoney fs_info->nodesize = 4096; 2811da17066cSJeff Mahoney fs_info->sectorsize = 4096; 2812da17066cSJeff Mahoney fs_info->stripesize = 4096; 2813da17066cSJeff Mahoney 2814eede2bf3SOmar Sandoval spin_lock_init(&fs_info->swapfile_pins_lock); 2815eede2bf3SOmar Sandoval fs_info->swapfile_pins = RB_ROOT; 2816eede2bf3SOmar Sandoval 28179e967495SFilipe Manana fs_info->send_in_progress = 0; 28189e967495SFilipe Manana 281953b381b3SDavid Woodhouse ret = btrfs_alloc_stripe_hash_table(fs_info); 282053b381b3SDavid Woodhouse if (ret) { 282183c8266aSDavid Sterba err = ret; 282253b381b3SDavid Woodhouse goto fail_alloc; 282353b381b3SDavid Woodhouse } 282453b381b3SDavid Woodhouse 28253c4bb26bSChris Mason invalidate_bdev(fs_devices->latest_bdev); 28261104a885SDavid Sterba 28271104a885SDavid Sterba /* 28281104a885SDavid Sterba * Read super block and check the signature bytes only 28291104a885SDavid Sterba */ 2830a512bbf8SYan Zheng bh = btrfs_read_dev_super(fs_devices->latest_bdev); 283192fc03fbSAnand Jain if (IS_ERR(bh)) { 283292fc03fbSAnand Jain err = PTR_ERR(bh); 283316cdcec7SMiao Xie goto fail_alloc; 283420b45077SDave Young } 283539279cc3SChris Mason 28361104a885SDavid Sterba /* 28378dc3f22cSJohannes Thumshirn * Verify the type first, if that or the the checksum value are 28388dc3f22cSJohannes Thumshirn * corrupted, we'll find out 28398dc3f22cSJohannes Thumshirn */ 284051bce6c9SJohannes Thumshirn csum_type = btrfs_super_csum_type((struct btrfs_super_block *)bh->b_data); 284151bce6c9SJohannes Thumshirn if (!btrfs_supported_super_csum(csum_type)) { 28428dc3f22cSJohannes Thumshirn btrfs_err(fs_info, "unsupported checksum algorithm: %u", 284351bce6c9SJohannes Thumshirn csum_type); 28448dc3f22cSJohannes Thumshirn err = -EINVAL; 28458dc3f22cSJohannes Thumshirn brelse(bh); 28468dc3f22cSJohannes Thumshirn goto fail_alloc; 28478dc3f22cSJohannes Thumshirn } 28488dc3f22cSJohannes Thumshirn 28496d97c6e3SJohannes Thumshirn ret = btrfs_init_csum_hash(fs_info, csum_type); 28506d97c6e3SJohannes Thumshirn if (ret) { 28516d97c6e3SJohannes Thumshirn err = ret; 28526d97c6e3SJohannes Thumshirn goto fail_alloc; 28536d97c6e3SJohannes Thumshirn } 28546d97c6e3SJohannes Thumshirn 28558dc3f22cSJohannes Thumshirn /* 28561104a885SDavid Sterba * We want to check superblock checksum, the type is stored inside. 28571104a885SDavid Sterba * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 28581104a885SDavid Sterba */ 2859ab8d0fc4SJeff Mahoney if (btrfs_check_super_csum(fs_info, bh->b_data)) { 286005135f59SDavid Sterba btrfs_err(fs_info, "superblock checksum mismatch"); 28611104a885SDavid Sterba err = -EINVAL; 2862b2acdddfSAnand Jain brelse(bh); 28636d97c6e3SJohannes Thumshirn goto fail_csum; 28641104a885SDavid Sterba } 28651104a885SDavid Sterba 28661104a885SDavid Sterba /* 28671104a885SDavid Sterba * super_copy is zeroed at allocation time and we never touch the 28681104a885SDavid Sterba * following bytes up to INFO_SIZE, the checksum is calculated from 28691104a885SDavid Sterba * the whole block of INFO_SIZE 28701104a885SDavid Sterba */ 28716c41761fSDavid Sterba memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy)); 2872a061fc8dSChris Mason brelse(bh); 28735f39d397SChris Mason 2874fbc6feaeSNikolay Borisov disk_super = fs_info->super_copy; 2875fbc6feaeSNikolay Borisov 2876de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, 2877de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 2878de37aa51SNikolay Borisov 28797239ff4bSNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID)) { 2880de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid, 2881de37aa51SNikolay Borisov fs_info->super_copy->metadata_uuid, 2882de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 28837239ff4bSNikolay Borisov } 28840b86a832SChris Mason 2885fbc6feaeSNikolay Borisov features = btrfs_super_flags(disk_super); 2886fbc6feaeSNikolay Borisov if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { 2887fbc6feaeSNikolay Borisov features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; 2888fbc6feaeSNikolay Borisov btrfs_set_super_flags(disk_super, features); 2889fbc6feaeSNikolay Borisov btrfs_info(fs_info, 2890fbc6feaeSNikolay Borisov "found metadata UUID change in progress flag, clearing"); 2891fbc6feaeSNikolay Borisov } 2892fbc6feaeSNikolay Borisov 2893fbc6feaeSNikolay Borisov memcpy(fs_info->super_for_commit, fs_info->super_copy, 2894fbc6feaeSNikolay Borisov sizeof(*fs_info->super_for_commit)); 2895de37aa51SNikolay Borisov 2896069ec957SQu Wenruo ret = btrfs_validate_mount_super(fs_info); 28971104a885SDavid Sterba if (ret) { 289805135f59SDavid Sterba btrfs_err(fs_info, "superblock contains fatal errors"); 28991104a885SDavid Sterba err = -EINVAL; 29006d97c6e3SJohannes Thumshirn goto fail_csum; 29011104a885SDavid Sterba } 29021104a885SDavid Sterba 29030f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 29046d97c6e3SJohannes Thumshirn goto fail_csum; 29050f7d52f4SChris Mason 2906acce952bSliubo /* check FS state, whether FS is broken. */ 290787533c47SMiao Xie if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 290887533c47SMiao Xie set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 2909acce952bSliubo 291075e7cb7fSLiu Bo /* 291175e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 291275e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 291375e7cb7fSLiu Bo */ 291475e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 291575e7cb7fSLiu Bo 29162ff7e61eSJeff Mahoney ret = btrfs_parse_options(fs_info, options, sb->s_flags); 29172b82032cSYan Zheng if (ret) { 29182b82032cSYan Zheng err = ret; 29196d97c6e3SJohannes Thumshirn goto fail_csum; 29202b82032cSYan Zheng } 2921dfe25020SChris Mason 2922f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 2923f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 2924f2b636e8SJosef Bacik if (features) { 292505135f59SDavid Sterba btrfs_err(fs_info, 292605135f59SDavid Sterba "cannot mount because of unsupported optional features (%llx)", 2927c1c9ff7cSGeert Uytterhoeven features); 2928f2b636e8SJosef Bacik err = -EINVAL; 29296d97c6e3SJohannes Thumshirn goto fail_csum; 2930f2b636e8SJosef Bacik } 2931f2b636e8SJosef Bacik 29325d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 29335d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 29340b246afaSJeff Mahoney if (fs_info->compress_type == BTRFS_COMPRESS_LZO) 2935a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 29365c1aab1dSNick Terrell else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) 29375c1aab1dSNick Terrell features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; 2938727011e0SChris Mason 29393173a18fSJosef Bacik if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) 294005135f59SDavid Sterba btrfs_info(fs_info, "has skinny extents"); 29413173a18fSJosef Bacik 2942727011e0SChris Mason /* 2943727011e0SChris Mason * flag our filesystem as having big metadata blocks if 2944727011e0SChris Mason * they are bigger than the page size 2945727011e0SChris Mason */ 294609cbfeafSKirill A. Shutemov if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) { 2947727011e0SChris Mason if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 294805135f59SDavid Sterba btrfs_info(fs_info, 294905135f59SDavid Sterba "flagging fs with big metadata feature"); 2950727011e0SChris Mason features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 2951727011e0SChris Mason } 2952727011e0SChris Mason 2953bc3f116fSChris Mason nodesize = btrfs_super_nodesize(disk_super); 2954bc3f116fSChris Mason sectorsize = btrfs_super_sectorsize(disk_super); 2955b7f67055SChandan Rajendra stripesize = sectorsize; 2956707e8a07SDavid Sterba fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 2957963d678bSMiao Xie fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 2958bc3f116fSChris Mason 2959da17066cSJeff Mahoney /* Cache block sizes */ 2960da17066cSJeff Mahoney fs_info->nodesize = nodesize; 2961da17066cSJeff Mahoney fs_info->sectorsize = sectorsize; 2962da17066cSJeff Mahoney fs_info->stripesize = stripesize; 2963da17066cSJeff Mahoney 2964bc3f116fSChris Mason /* 2965bc3f116fSChris Mason * mixed block groups end up with duplicate but slightly offset 2966bc3f116fSChris Mason * extent buffers for the same range. It leads to corruptions 2967bc3f116fSChris Mason */ 2968bc3f116fSChris Mason if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 2969707e8a07SDavid Sterba (sectorsize != nodesize)) { 297005135f59SDavid Sterba btrfs_err(fs_info, 297105135f59SDavid Sterba "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", 297205135f59SDavid Sterba nodesize, sectorsize); 29736d97c6e3SJohannes Thumshirn goto fail_csum; 2974bc3f116fSChris Mason } 2975bc3f116fSChris Mason 2976ceda0864SMiao Xie /* 2977ceda0864SMiao Xie * Needn't use the lock because there is no other task which will 2978ceda0864SMiao Xie * update the flag. 2979ceda0864SMiao Xie */ 29805d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 29815d4f98a2SYan Zheng 2982f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 2983f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 2984bc98a42cSDavid Howells if (!sb_rdonly(sb) && features) { 298505135f59SDavid Sterba btrfs_err(fs_info, 298605135f59SDavid Sterba "cannot mount read-write because of unsupported optional features (%llx)", 2987c1c9ff7cSGeert Uytterhoeven features); 2988f2b636e8SJosef Bacik err = -EINVAL; 29896d97c6e3SJohannes Thumshirn goto fail_csum; 2990f2b636e8SJosef Bacik } 299161d92c32SChris Mason 29922a458198SEric Sandeen ret = btrfs_init_workqueues(fs_info, fs_devices); 29932a458198SEric Sandeen if (ret) { 29942a458198SEric Sandeen err = ret; 29950dc3b84aSJosef Bacik goto fail_sb_buffer; 29960dc3b84aSJosef Bacik } 29974543df7eSChris Mason 29989e11ceeeSJan Kara sb->s_bdi->congested_fn = btrfs_congested_fn; 29999e11ceeeSJan Kara sb->s_bdi->congested_data = fs_info; 30009e11ceeeSJan Kara sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK; 3001b5420237SNikolay Borisov sb->s_bdi->ra_pages = VM_READAHEAD_PAGES; 30029e11ceeeSJan Kara sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); 30039e11ceeeSJan Kara sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); 30044575c9ccSChris Mason 3005a061fc8dSChris Mason sb->s_blocksize = sectorsize; 3006a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 3007de37aa51SNikolay Borisov memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); 3008db94535dSChris Mason 3009925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 30106bccf3abSJeff Mahoney ret = btrfs_read_sys_array(fs_info); 3011925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 301284eed90fSChris Mason if (ret) { 301305135f59SDavid Sterba btrfs_err(fs_info, "failed to read the system array: %d", ret); 30145d4f98a2SYan Zheng goto fail_sb_buffer; 301584eed90fSChris Mason } 30160b86a832SChris Mason 301784234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 3018581c1760SQu Wenruo level = btrfs_super_chunk_root_level(disk_super); 30190b86a832SChris Mason 30202ff7e61eSJeff Mahoney chunk_root->node = read_tree_block(fs_info, 30210b86a832SChris Mason btrfs_super_chunk_root(disk_super), 3022581c1760SQu Wenruo generation, level, NULL); 302364c043deSLiu Bo if (IS_ERR(chunk_root->node) || 302464c043deSLiu Bo !extent_buffer_uptodate(chunk_root->node)) { 302505135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk root"); 3026e5fffbacSchandan if (!IS_ERR(chunk_root->node)) 3027e5fffbacSchandan free_extent_buffer(chunk_root->node); 302895ab1f64SZhao Lei chunk_root->node = NULL; 3029af31f5e5SChris Mason goto fail_tree_roots; 303083121942SDavid Woodhouse } 30315d4f98a2SYan Zheng btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 30325d4f98a2SYan Zheng chunk_root->commit_root = btrfs_root_node(chunk_root); 30330b86a832SChris Mason 3034e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 3035b308bc2fSGeert Uytterhoeven btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE); 3036e17cade2SChris Mason 30375b4aacefSJeff Mahoney ret = btrfs_read_chunk_tree(fs_info); 30382b82032cSYan Zheng if (ret) { 303905135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk tree: %d", ret); 3040af31f5e5SChris Mason goto fail_tree_roots; 30412b82032cSYan Zheng } 30420b86a832SChris Mason 30438dabb742SStefan Behrens /* 30449b99b115SAnand Jain * Keep the devid that is marked to be the target device for the 30459b99b115SAnand Jain * device replace procedure 30468dabb742SStefan Behrens */ 30479b99b115SAnand Jain btrfs_free_extra_devids(fs_devices, 0); 3048dfe25020SChris Mason 3049a6b0d5c8SChris Mason if (!fs_devices->latest_bdev) { 305005135f59SDavid Sterba btrfs_err(fs_info, "failed to read devices"); 3051a6b0d5c8SChris Mason goto fail_tree_roots; 3052a6b0d5c8SChris Mason } 3053a6b0d5c8SChris Mason 3054b8522a1eSNikolay Borisov ret = init_tree_roots(fs_info); 30554bbcaa64SEric Sandeen if (ret) 3056b8522a1eSNikolay Borisov goto fail_tree_roots; 30578929ecfaSYan, Zheng 3058cf90d884SQu Wenruo ret = btrfs_verify_dev_extents(fs_info); 3059cf90d884SQu Wenruo if (ret) { 3060cf90d884SQu Wenruo btrfs_err(fs_info, 3061cf90d884SQu Wenruo "failed to verify dev extents against chunks: %d", 3062cf90d884SQu Wenruo ret); 3063cf90d884SQu Wenruo goto fail_block_groups; 3064cf90d884SQu Wenruo } 306568310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 306668310a5eSIlya Dryomov if (ret) { 306705135f59SDavid Sterba btrfs_err(fs_info, "failed to recover balance: %d", ret); 306868310a5eSIlya Dryomov goto fail_block_groups; 306968310a5eSIlya Dryomov } 307068310a5eSIlya Dryomov 3071733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 3072733f4fbbSStefan Behrens if (ret) { 307305135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_stats: %d", ret); 3074733f4fbbSStefan Behrens goto fail_block_groups; 3075733f4fbbSStefan Behrens } 3076733f4fbbSStefan Behrens 30778dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 30788dabb742SStefan Behrens if (ret) { 307905135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_replace: %d", ret); 30808dabb742SStefan Behrens goto fail_block_groups; 30818dabb742SStefan Behrens } 30828dabb742SStefan Behrens 30839b99b115SAnand Jain btrfs_free_extra_devids(fs_devices, 1); 30848dabb742SStefan Behrens 3085c6761a9eSAnand Jain ret = btrfs_sysfs_add_fsid(fs_devices); 3086b7c35e81SAnand Jain if (ret) { 308705135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", 308805135f59SDavid Sterba ret); 3089b7c35e81SAnand Jain goto fail_block_groups; 3090b7c35e81SAnand Jain } 3091b7c35e81SAnand Jain 309296f3136eSAnand Jain ret = btrfs_sysfs_add_mounted(fs_info); 30935ac1d209SJeff Mahoney if (ret) { 309405135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); 3095b7c35e81SAnand Jain goto fail_fsdev_sysfs; 30965ac1d209SJeff Mahoney } 30975ac1d209SJeff Mahoney 3098c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 3099c59021f8Sliubo if (ret) { 310005135f59SDavid Sterba btrfs_err(fs_info, "failed to initialize space info: %d", ret); 31012365dd3cSAnand Jain goto fail_sysfs; 3102c59021f8Sliubo } 3103c59021f8Sliubo 31045b4aacefSJeff Mahoney ret = btrfs_read_block_groups(fs_info); 31051b1d1f66SJosef Bacik if (ret) { 310605135f59SDavid Sterba btrfs_err(fs_info, "failed to read block groups: %d", ret); 31072365dd3cSAnand Jain goto fail_sysfs; 31081b1d1f66SJosef Bacik } 31094330e183SQu Wenruo 31106528b99dSAnand Jain if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) { 311105135f59SDavid Sterba btrfs_warn(fs_info, 311252042d8eSAndrea Gelmini "writable mount is not allowed due to too many missing devices"); 31132365dd3cSAnand Jain goto fail_sysfs; 3114292fd7fcSStefan Behrens } 31159078a3e1SChris Mason 3116a74a4b97SChris Mason fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 3117a74a4b97SChris Mason "btrfs-cleaner"); 311857506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 31192365dd3cSAnand Jain goto fail_sysfs; 3120a74a4b97SChris Mason 3121a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 3122a74a4b97SChris Mason tree_root, 3123a74a4b97SChris Mason "btrfs-transaction"); 312457506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 31253f157a2fSChris Mason goto fail_cleaner; 3126a74a4b97SChris Mason 3127583b7231SHans van Kranenburg if (!btrfs_test_opt(fs_info, NOSSD) && 3128c289811cSChris Mason !fs_info->fs_devices->rotating) { 3129583b7231SHans van Kranenburg btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); 3130c289811cSChris Mason } 3131c289811cSChris Mason 3132572d9ab7SDavid Sterba /* 313301327610SNicholas D Steeves * Mount does not set all options immediately, we can do it now and do 3134572d9ab7SDavid Sterba * not have to wait for transaction commit 3135572d9ab7SDavid Sterba */ 3136572d9ab7SDavid Sterba btrfs_apply_pending_changes(fs_info); 31373818aea2SQu Wenruo 313821adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 31390b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { 31402ff7e61eSJeff Mahoney ret = btrfsic_mount(fs_info, fs_devices, 31410b246afaSJeff Mahoney btrfs_test_opt(fs_info, 314221adbd5cSStefan Behrens CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? 314321adbd5cSStefan Behrens 1 : 0, 314421adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 314521adbd5cSStefan Behrens if (ret) 314605135f59SDavid Sterba btrfs_warn(fs_info, 314705135f59SDavid Sterba "failed to initialize integrity check module: %d", 314805135f59SDavid Sterba ret); 314921adbd5cSStefan Behrens } 315021adbd5cSStefan Behrens #endif 3151bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 3152bcef60f2SArne Jansen if (ret) 3153bcef60f2SArne Jansen goto fail_trans_kthread; 315421adbd5cSStefan Behrens 3155fd708b81SJosef Bacik if (btrfs_build_ref_tree(fs_info)) 3156fd708b81SJosef Bacik btrfs_err(fs_info, "couldn't build ref tree"); 3157fd708b81SJosef Bacik 315896da0919SQu Wenruo /* do not make disk changes in broken FS or nologreplay is given */ 315996da0919SQu Wenruo if (btrfs_super_log_root(disk_super) != 0 && 31600b246afaSJeff Mahoney !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 3161e8294f2fSDavid Sterba btrfs_info(fs_info, "start tree-log replay"); 316263443bf5SEric Sandeen ret = btrfs_replay_log(fs_info, fs_devices); 316379787eaaSJeff Mahoney if (ret) { 316463443bf5SEric Sandeen err = ret; 316528c16cbbSWang Shilong goto fail_qgroup; 3166e556ce2cSYan Zheng } 3167e02119d5SChris Mason } 31681a40e23bSZheng Yan 31696bccf3abSJeff Mahoney ret = btrfs_find_orphan_roots(fs_info); 317079787eaaSJeff Mahoney if (ret) 317128c16cbbSWang Shilong goto fail_qgroup; 317276dda93cSYan, Zheng 3173bc98a42cSDavid Howells if (!sb_rdonly(sb)) { 3174d68fc57bSYan, Zheng ret = btrfs_cleanup_fs_roots(fs_info); 317544c44af2SIlya Dryomov if (ret) 317628c16cbbSWang Shilong goto fail_qgroup; 317790c711abSZygo Blaxell 317890c711abSZygo Blaxell mutex_lock(&fs_info->cleaner_mutex); 31795d4f98a2SYan Zheng ret = btrfs_recover_relocation(tree_root); 318090c711abSZygo Blaxell mutex_unlock(&fs_info->cleaner_mutex); 3181d7ce5843SMiao Xie if (ret < 0) { 318205135f59SDavid Sterba btrfs_warn(fs_info, "failed to recover relocation: %d", 318305135f59SDavid Sterba ret); 3184d7ce5843SMiao Xie err = -EINVAL; 3185bcef60f2SArne Jansen goto fail_qgroup; 3186d7ce5843SMiao Xie } 31877c2ca468SChris Mason } 31881a40e23bSZheng Yan 31893de4586cSChris Mason location.objectid = BTRFS_FS_TREE_OBJECTID; 31903de4586cSChris Mason location.type = BTRFS_ROOT_ITEM_KEY; 3191cb517eabSMiao Xie location.offset = 0; 31923de4586cSChris Mason 31933de4586cSChris Mason fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location); 31943140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 31953140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 3196f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read fs tree: %d", err); 3197315bf8efSJosef Bacik fs_info->fs_root = NULL; 3198bcef60f2SArne Jansen goto fail_qgroup; 31993140c9a3SDan Carpenter } 3200c289811cSChris Mason 3201bc98a42cSDavid Howells if (sb_rdonly(sb)) 32022b6ba629SIlya Dryomov return 0; 32032b6ba629SIlya Dryomov 3204f8d468a1SOmar Sandoval if (btrfs_test_opt(fs_info, CLEAR_CACHE) && 3205f8d468a1SOmar Sandoval btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 32066675df31SOmar Sandoval clear_free_space_tree = 1; 32076675df31SOmar Sandoval } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 32086675df31SOmar Sandoval !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { 32096675df31SOmar Sandoval btrfs_warn(fs_info, "free space tree is invalid"); 32106675df31SOmar Sandoval clear_free_space_tree = 1; 32116675df31SOmar Sandoval } 32126675df31SOmar Sandoval 32136675df31SOmar Sandoval if (clear_free_space_tree) { 3214f8d468a1SOmar Sandoval btrfs_info(fs_info, "clearing free space tree"); 3215f8d468a1SOmar Sandoval ret = btrfs_clear_free_space_tree(fs_info); 3216f8d468a1SOmar Sandoval if (ret) { 3217f8d468a1SOmar Sandoval btrfs_warn(fs_info, 3218f8d468a1SOmar Sandoval "failed to clear free space tree: %d", ret); 32196bccf3abSJeff Mahoney close_ctree(fs_info); 3220f8d468a1SOmar Sandoval return ret; 3221f8d468a1SOmar Sandoval } 3222f8d468a1SOmar Sandoval } 3223f8d468a1SOmar Sandoval 32240b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && 3225511711afSChris Mason !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 322605135f59SDavid Sterba btrfs_info(fs_info, "creating free space tree"); 3227511711afSChris Mason ret = btrfs_create_free_space_tree(fs_info); 3228511711afSChris Mason if (ret) { 322905135f59SDavid Sterba btrfs_warn(fs_info, 323005135f59SDavid Sterba "failed to create free space tree: %d", ret); 32316bccf3abSJeff Mahoney close_ctree(fs_info); 3232511711afSChris Mason return ret; 3233511711afSChris Mason } 3234511711afSChris Mason } 3235511711afSChris Mason 3236e3acc2a6SJosef Bacik down_read(&fs_info->cleanup_work_sem); 32372b6ba629SIlya Dryomov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 32382b6ba629SIlya Dryomov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 32392b6ba629SIlya Dryomov up_read(&fs_info->cleanup_work_sem); 32406bccf3abSJeff Mahoney close_ctree(fs_info); 32412b6ba629SIlya Dryomov return ret; 32422b6ba629SIlya Dryomov } 3243e3acc2a6SJosef Bacik up_read(&fs_info->cleanup_work_sem); 324459641015SIlya Dryomov 32452b6ba629SIlya Dryomov ret = btrfs_resume_balance_async(fs_info); 32462b6ba629SIlya Dryomov if (ret) { 324705135f59SDavid Sterba btrfs_warn(fs_info, "failed to resume balance: %d", ret); 32486bccf3abSJeff Mahoney close_ctree(fs_info); 32492b6ba629SIlya Dryomov return ret; 3250e3acc2a6SJosef Bacik } 3251e3acc2a6SJosef Bacik 32528dabb742SStefan Behrens ret = btrfs_resume_dev_replace_async(fs_info); 32538dabb742SStefan Behrens if (ret) { 325405135f59SDavid Sterba btrfs_warn(fs_info, "failed to resume device replace: %d", ret); 32556bccf3abSJeff Mahoney close_ctree(fs_info); 32568dabb742SStefan Behrens return ret; 32578dabb742SStefan Behrens } 32588dabb742SStefan Behrens 3259b382a324SJan Schmidt btrfs_qgroup_rescan_resume(fs_info); 3260b0643e59SDennis Zhou btrfs_discard_resume(fs_info); 3261b382a324SJan Schmidt 32624bbcaa64SEric Sandeen if (!fs_info->uuid_root) { 326305135f59SDavid Sterba btrfs_info(fs_info, "creating UUID tree"); 3264f7a81ea4SStefan Behrens ret = btrfs_create_uuid_tree(fs_info); 3265f7a81ea4SStefan Behrens if (ret) { 326605135f59SDavid Sterba btrfs_warn(fs_info, 326705135f59SDavid Sterba "failed to create the UUID tree: %d", ret); 32686bccf3abSJeff Mahoney close_ctree(fs_info); 3269f7a81ea4SStefan Behrens return ret; 3270f7a81ea4SStefan Behrens } 32710b246afaSJeff Mahoney } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || 32724bbcaa64SEric Sandeen fs_info->generation != 32734bbcaa64SEric Sandeen btrfs_super_uuid_tree_generation(disk_super)) { 327405135f59SDavid Sterba btrfs_info(fs_info, "checking UUID tree"); 327570f80175SStefan Behrens ret = btrfs_check_uuid_tree(fs_info); 327670f80175SStefan Behrens if (ret) { 327705135f59SDavid Sterba btrfs_warn(fs_info, 327805135f59SDavid Sterba "failed to check the UUID tree: %d", ret); 32796bccf3abSJeff Mahoney close_ctree(fs_info); 328070f80175SStefan Behrens return ret; 328170f80175SStefan Behrens } 328270f80175SStefan Behrens } else { 3283afcdd129SJosef Bacik set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 3284f7a81ea4SStefan Behrens } 3285afcdd129SJosef Bacik set_bit(BTRFS_FS_OPEN, &fs_info->flags); 328647ab2a6cSJosef Bacik 32878dcddfa0SQu Wenruo /* 32888dcddfa0SQu Wenruo * backuproot only affect mount behavior, and if open_ctree succeeded, 32898dcddfa0SQu Wenruo * no need to keep the flag 32908dcddfa0SQu Wenruo */ 32918dcddfa0SQu Wenruo btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT); 32928dcddfa0SQu Wenruo 3293ad2b2c80SAl Viro return 0; 329439279cc3SChris Mason 3295bcef60f2SArne Jansen fail_qgroup: 3296bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 32977c2ca468SChris Mason fail_trans_kthread: 32987c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 32992ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 3300faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 33013f157a2fSChris Mason fail_cleaner: 3302a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 33037c2ca468SChris Mason 33047c2ca468SChris Mason /* 33057c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 33067c2ca468SChris Mason * kthreads 33077c2ca468SChris Mason */ 33087c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 33097c2ca468SChris Mason 33102365dd3cSAnand Jain fail_sysfs: 33116618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 33122365dd3cSAnand Jain 3313b7c35e81SAnand Jain fail_fsdev_sysfs: 3314b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3315b7c35e81SAnand Jain 33161b1d1f66SJosef Bacik fail_block_groups: 331754067ae9SJosef Bacik btrfs_put_block_group_cache(fs_info); 3318af31f5e5SChris Mason 3319af31f5e5SChris Mason fail_tree_roots: 33204273eaffSAnand Jain free_root_pointers(fs_info, true); 33212b8195bbSMiao Xie invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3322af31f5e5SChris Mason 332339279cc3SChris Mason fail_sb_buffer: 33247abadb64SLiu Bo btrfs_stop_all_workers(fs_info); 33255cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 33266d97c6e3SJohannes Thumshirn fail_csum: 33276d97c6e3SJohannes Thumshirn btrfs_free_csum_hash(fs_info); 332816cdcec7SMiao Xie fail_alloc: 33294543df7eSChris Mason fail_iput: 3330586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 3331586e46e2SIlya Dryomov 33324543df7eSChris Mason iput(fs_info->btree_inode); 3333c404e0dcSMiao Xie fail_bio_counter: 33347f8d236aSDavid Sterba percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 3335963d678bSMiao Xie fail_delalloc_bytes: 3336963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 3337e2d84521SMiao Xie fail_dirty_metadata_bytes: 3338e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 33394297ff84SJosef Bacik fail_dio_bytes: 33404297ff84SJosef Bacik percpu_counter_destroy(&fs_info->dio_bytes); 334176dda93cSYan, Zheng fail_srcu: 334276dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 33437e662854SQinghuang Feng fail: 334453b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 3345586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 3346ad2b2c80SAl Viro return err; 3347eb60ceacSChris Mason } 3348663faf9fSMasami Hiramatsu ALLOW_ERROR_INJECTION(open_ctree, ERRNO); 3349eb60ceacSChris Mason 3350f2984462SChris Mason static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate) 3351f2984462SChris Mason { 3352f2984462SChris Mason if (uptodate) { 3353f2984462SChris Mason set_buffer_uptodate(bh); 3354f2984462SChris Mason } else { 3355442a4f63SStefan Behrens struct btrfs_device *device = (struct btrfs_device *) 3356442a4f63SStefan Behrens bh->b_private; 3357442a4f63SStefan Behrens 3358fb456252SJeff Mahoney btrfs_warn_rl_in_rcu(device->fs_info, 3359b14af3b4SDavid Sterba "lost page write due to IO error on %s", 3360606686eeSJosef Bacik rcu_str_deref(device->name)); 336101327610SNicholas D Steeves /* note, we don't set_buffer_write_io_error because we have 33621259ab75SChris Mason * our own ways of dealing with the IO errors 33631259ab75SChris Mason */ 3364f2984462SChris Mason clear_buffer_uptodate(bh); 3365442a4f63SStefan Behrens btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS); 3366f2984462SChris Mason } 3367f2984462SChris Mason unlock_buffer(bh); 3368f2984462SChris Mason put_bh(bh); 3369f2984462SChris Mason } 3370f2984462SChris Mason 337129c36d72SAnand Jain int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num, 337229c36d72SAnand Jain struct buffer_head **bh_ret) 337329c36d72SAnand Jain { 337429c36d72SAnand Jain struct buffer_head *bh; 337529c36d72SAnand Jain struct btrfs_super_block *super; 337629c36d72SAnand Jain u64 bytenr; 337729c36d72SAnand Jain 337829c36d72SAnand Jain bytenr = btrfs_sb_offset(copy_num); 337929c36d72SAnand Jain if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode)) 338029c36d72SAnand Jain return -EINVAL; 338129c36d72SAnand Jain 33829f6d2510SDavid Sterba bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE); 338329c36d72SAnand Jain /* 338429c36d72SAnand Jain * If we fail to read from the underlying devices, as of now 338529c36d72SAnand Jain * the best option we have is to mark it EIO. 338629c36d72SAnand Jain */ 338729c36d72SAnand Jain if (!bh) 338829c36d72SAnand Jain return -EIO; 338929c36d72SAnand Jain 339029c36d72SAnand Jain super = (struct btrfs_super_block *)bh->b_data; 339129c36d72SAnand Jain if (btrfs_super_bytenr(super) != bytenr || 339229c36d72SAnand Jain btrfs_super_magic(super) != BTRFS_MAGIC) { 339329c36d72SAnand Jain brelse(bh); 339429c36d72SAnand Jain return -EINVAL; 339529c36d72SAnand Jain } 339629c36d72SAnand Jain 339729c36d72SAnand Jain *bh_ret = bh; 339829c36d72SAnand Jain return 0; 339929c36d72SAnand Jain } 340029c36d72SAnand Jain 340129c36d72SAnand Jain 3402a512bbf8SYan Zheng struct buffer_head *btrfs_read_dev_super(struct block_device *bdev) 3403a512bbf8SYan Zheng { 3404a512bbf8SYan Zheng struct buffer_head *bh; 3405a512bbf8SYan Zheng struct buffer_head *latest = NULL; 3406a512bbf8SYan Zheng struct btrfs_super_block *super; 3407a512bbf8SYan Zheng int i; 3408a512bbf8SYan Zheng u64 transid = 0; 340992fc03fbSAnand Jain int ret = -EINVAL; 3410a512bbf8SYan Zheng 3411a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 3412a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 3413a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 3414a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3415a512bbf8SYan Zheng */ 3416a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 341729c36d72SAnand Jain ret = btrfs_read_dev_one_super(bdev, i, &bh); 341829c36d72SAnand Jain if (ret) 3419a512bbf8SYan Zheng continue; 3420a512bbf8SYan Zheng 3421a512bbf8SYan Zheng super = (struct btrfs_super_block *)bh->b_data; 3422a512bbf8SYan Zheng 3423a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 3424a512bbf8SYan Zheng brelse(latest); 3425a512bbf8SYan Zheng latest = bh; 3426a512bbf8SYan Zheng transid = btrfs_super_generation(super); 3427a512bbf8SYan Zheng } else { 3428a512bbf8SYan Zheng brelse(bh); 3429a512bbf8SYan Zheng } 3430a512bbf8SYan Zheng } 343192fc03fbSAnand Jain 343292fc03fbSAnand Jain if (!latest) 343392fc03fbSAnand Jain return ERR_PTR(ret); 343492fc03fbSAnand Jain 3435a512bbf8SYan Zheng return latest; 3436a512bbf8SYan Zheng } 3437a512bbf8SYan Zheng 34384eedeb75SHisashi Hifumi /* 3439abbb3b8eSDavid Sterba * Write superblock @sb to the @device. Do not wait for completion, all the 3440abbb3b8eSDavid Sterba * buffer heads we write are pinned. 34414eedeb75SHisashi Hifumi * 3442abbb3b8eSDavid Sterba * Write @max_mirrors copies of the superblock, where 0 means default that fit 3443abbb3b8eSDavid Sterba * the expected device size at commit time. Note that max_mirrors must be 3444abbb3b8eSDavid Sterba * same for write and wait phases. 34454eedeb75SHisashi Hifumi * 3446abbb3b8eSDavid Sterba * Return number of errors when buffer head is not found or submission fails. 34474eedeb75SHisashi Hifumi */ 3448a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 3449abbb3b8eSDavid Sterba struct btrfs_super_block *sb, int max_mirrors) 3450a512bbf8SYan Zheng { 3451d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = device->fs_info; 3452d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 3453a512bbf8SYan Zheng struct buffer_head *bh; 3454a512bbf8SYan Zheng int i; 3455a512bbf8SYan Zheng int ret; 3456a512bbf8SYan Zheng int errors = 0; 3457a512bbf8SYan Zheng u64 bytenr; 34581b9e619cSOmar Sandoval int op_flags; 3459a512bbf8SYan Zheng 3460a512bbf8SYan Zheng if (max_mirrors == 0) 3461a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3462a512bbf8SYan Zheng 3463d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 3464d5178578SJohannes Thumshirn 3465a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 3466a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 3467935e5cc9SMiao Xie if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3468935e5cc9SMiao Xie device->commit_total_bytes) 3469a512bbf8SYan Zheng break; 3470a512bbf8SYan Zheng 3471a512bbf8SYan Zheng btrfs_set_super_bytenr(sb, bytenr); 3472a512bbf8SYan Zheng 3473d5178578SJohannes Thumshirn crypto_shash_init(shash); 3474d5178578SJohannes Thumshirn crypto_shash_update(shash, (const char *)sb + BTRFS_CSUM_SIZE, 3475abbb3b8eSDavid Sterba BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE); 3476d5178578SJohannes Thumshirn crypto_shash_final(shash, sb->csum); 3477a512bbf8SYan Zheng 3478abbb3b8eSDavid Sterba /* One reference for us, and we leave it for the caller */ 34799f6d2510SDavid Sterba bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, 3480a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 3481634554dcSJosef Bacik if (!bh) { 3482fb456252SJeff Mahoney btrfs_err(device->fs_info, 3483f14d104dSDavid Sterba "couldn't get super buffer head for bytenr %llu", 3484f14d104dSDavid Sterba bytenr); 3485634554dcSJosef Bacik errors++; 3486634554dcSJosef Bacik continue; 3487634554dcSJosef Bacik } 3488634554dcSJosef Bacik 3489a512bbf8SYan Zheng memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE); 3490a512bbf8SYan Zheng 34914eedeb75SHisashi Hifumi /* one reference for submit_bh */ 3492a512bbf8SYan Zheng get_bh(bh); 34934eedeb75SHisashi Hifumi 34944eedeb75SHisashi Hifumi set_buffer_uptodate(bh); 3495a512bbf8SYan Zheng lock_buffer(bh); 3496a512bbf8SYan Zheng bh->b_end_io = btrfs_end_buffer_write_sync; 3497442a4f63SStefan Behrens bh->b_private = device; 3498a512bbf8SYan Zheng 3499387125fcSChris Mason /* 3500387125fcSChris Mason * we fua the first super. The others we allow 3501387125fcSChris Mason * to go down lazy. 3502387125fcSChris Mason */ 35031b9e619cSOmar Sandoval op_flags = REQ_SYNC | REQ_META | REQ_PRIO; 35041b9e619cSOmar Sandoval if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) 35051b9e619cSOmar Sandoval op_flags |= REQ_FUA; 35061b9e619cSOmar Sandoval ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh); 35074eedeb75SHisashi Hifumi if (ret) 3508a512bbf8SYan Zheng errors++; 3509a512bbf8SYan Zheng } 3510a512bbf8SYan Zheng return errors < i ? 0 : -1; 3511a512bbf8SYan Zheng } 3512a512bbf8SYan Zheng 3513387125fcSChris Mason /* 3514abbb3b8eSDavid Sterba * Wait for write completion of superblocks done by write_dev_supers, 3515abbb3b8eSDavid Sterba * @max_mirrors same for write and wait phases. 3516abbb3b8eSDavid Sterba * 3517abbb3b8eSDavid Sterba * Return number of errors when buffer head is not found or not marked up to 3518abbb3b8eSDavid Sterba * date. 3519abbb3b8eSDavid Sterba */ 3520abbb3b8eSDavid Sterba static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) 3521abbb3b8eSDavid Sterba { 3522abbb3b8eSDavid Sterba struct buffer_head *bh; 3523abbb3b8eSDavid Sterba int i; 3524abbb3b8eSDavid Sterba int errors = 0; 3525b6a535faSHoward McLauchlan bool primary_failed = false; 3526abbb3b8eSDavid Sterba u64 bytenr; 3527abbb3b8eSDavid Sterba 3528abbb3b8eSDavid Sterba if (max_mirrors == 0) 3529abbb3b8eSDavid Sterba max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3530abbb3b8eSDavid Sterba 3531abbb3b8eSDavid Sterba for (i = 0; i < max_mirrors; i++) { 3532abbb3b8eSDavid Sterba bytenr = btrfs_sb_offset(i); 3533abbb3b8eSDavid Sterba if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3534abbb3b8eSDavid Sterba device->commit_total_bytes) 3535abbb3b8eSDavid Sterba break; 3536abbb3b8eSDavid Sterba 35379f6d2510SDavid Sterba bh = __find_get_block(device->bdev, 35389f6d2510SDavid Sterba bytenr / BTRFS_BDEV_BLOCKSIZE, 3539abbb3b8eSDavid Sterba BTRFS_SUPER_INFO_SIZE); 3540abbb3b8eSDavid Sterba if (!bh) { 3541abbb3b8eSDavid Sterba errors++; 3542b6a535faSHoward McLauchlan if (i == 0) 3543b6a535faSHoward McLauchlan primary_failed = true; 3544abbb3b8eSDavid Sterba continue; 3545abbb3b8eSDavid Sterba } 3546abbb3b8eSDavid Sterba wait_on_buffer(bh); 3547b6a535faSHoward McLauchlan if (!buffer_uptodate(bh)) { 3548abbb3b8eSDavid Sterba errors++; 3549b6a535faSHoward McLauchlan if (i == 0) 3550b6a535faSHoward McLauchlan primary_failed = true; 3551b6a535faSHoward McLauchlan } 3552abbb3b8eSDavid Sterba 3553abbb3b8eSDavid Sterba /* drop our reference */ 3554abbb3b8eSDavid Sterba brelse(bh); 3555abbb3b8eSDavid Sterba 3556abbb3b8eSDavid Sterba /* drop the reference from the writing run */ 3557abbb3b8eSDavid Sterba brelse(bh); 3558abbb3b8eSDavid Sterba } 3559abbb3b8eSDavid Sterba 3560b6a535faSHoward McLauchlan /* log error, force error return */ 3561b6a535faSHoward McLauchlan if (primary_failed) { 3562b6a535faSHoward McLauchlan btrfs_err(device->fs_info, "error writing primary super block to device %llu", 3563b6a535faSHoward McLauchlan device->devid); 3564b6a535faSHoward McLauchlan return -1; 3565b6a535faSHoward McLauchlan } 3566b6a535faSHoward McLauchlan 3567abbb3b8eSDavid Sterba return errors < i ? 0 : -1; 3568abbb3b8eSDavid Sterba } 3569abbb3b8eSDavid Sterba 3570abbb3b8eSDavid Sterba /* 3571387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 3572387125fcSChris Mason * for the barrier when it is done 3573387125fcSChris Mason */ 35744246a0b6SChristoph Hellwig static void btrfs_end_empty_barrier(struct bio *bio) 3575387125fcSChris Mason { 3576387125fcSChris Mason complete(bio->bi_private); 3577387125fcSChris Mason } 3578387125fcSChris Mason 3579387125fcSChris Mason /* 35804fc6441aSAnand Jain * Submit a flush request to the device if it supports it. Error handling is 35814fc6441aSAnand Jain * done in the waiting counterpart. 3582387125fcSChris Mason */ 35834fc6441aSAnand Jain static void write_dev_flush(struct btrfs_device *device) 3584387125fcSChris Mason { 3585c2a9c7abSAnand Jain struct request_queue *q = bdev_get_queue(device->bdev); 3586e0ae9994SDavid Sterba struct bio *bio = device->flush_bio; 3587387125fcSChris Mason 3588c2a9c7abSAnand Jain if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) 35894fc6441aSAnand Jain return; 3590387125fcSChris Mason 3591e0ae9994SDavid Sterba bio_reset(bio); 3592387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 359374d46992SChristoph Hellwig bio_set_dev(bio, device->bdev); 35948d910125SJan Kara bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH; 3595387125fcSChris Mason init_completion(&device->flush_wait); 3596387125fcSChris Mason bio->bi_private = &device->flush_wait; 3597387125fcSChris Mason 359843a01111SLu Fengqi btrfsic_submit_bio(bio); 35991c3063b6SAnand Jain set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 36004fc6441aSAnand Jain } 3601387125fcSChris Mason 36024fc6441aSAnand Jain /* 36034fc6441aSAnand Jain * If the flush bio has been submitted by write_dev_flush, wait for it. 36044fc6441aSAnand Jain */ 36058c27cb35SLinus Torvalds static blk_status_t wait_dev_flush(struct btrfs_device *device) 36064fc6441aSAnand Jain { 36074fc6441aSAnand Jain struct bio *bio = device->flush_bio; 36084fc6441aSAnand Jain 36091c3063b6SAnand Jain if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) 361058efbc9fSOmar Sandoval return BLK_STS_OK; 36114fc6441aSAnand Jain 36121c3063b6SAnand Jain clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 36132980d574SDavid Sterba wait_for_completion_io(&device->flush_wait); 36144fc6441aSAnand Jain 36158c27cb35SLinus Torvalds return bio->bi_status; 3616387125fcSChris Mason } 3617387125fcSChris Mason 3618d10b82feSQu Wenruo static int check_barrier_error(struct btrfs_fs_info *fs_info) 3619401b41e5SAnand Jain { 36206528b99dSAnand Jain if (!btrfs_check_rw_degradable(fs_info, NULL)) 3621401b41e5SAnand Jain return -EIO; 3622387125fcSChris Mason return 0; 3623387125fcSChris Mason } 3624387125fcSChris Mason 3625387125fcSChris Mason /* 3626387125fcSChris Mason * send an empty flush down to each device in parallel, 3627387125fcSChris Mason * then wait for them 3628387125fcSChris Mason */ 3629387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 3630387125fcSChris Mason { 3631387125fcSChris Mason struct list_head *head; 3632387125fcSChris Mason struct btrfs_device *dev; 36335af3e8ccSStefan Behrens int errors_wait = 0; 36344e4cbee9SChristoph Hellwig blk_status_t ret; 3635387125fcSChris Mason 36361538e6c5SDavid Sterba lockdep_assert_held(&info->fs_devices->device_list_mutex); 3637387125fcSChris Mason /* send down all the barriers */ 3638387125fcSChris Mason head = &info->fs_devices->devices; 36391538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3640e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3641f88ba6a2SHidetoshi Seto continue; 3642cea7c8bfSAnand Jain if (!dev->bdev) 3643387125fcSChris Mason continue; 3644e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3645ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3646387125fcSChris Mason continue; 3647387125fcSChris Mason 36484fc6441aSAnand Jain write_dev_flush(dev); 364958efbc9fSOmar Sandoval dev->last_flush_error = BLK_STS_OK; 3650387125fcSChris Mason } 3651387125fcSChris Mason 3652387125fcSChris Mason /* wait for all the barriers */ 36531538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3654e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3655f88ba6a2SHidetoshi Seto continue; 3656387125fcSChris Mason if (!dev->bdev) { 36575af3e8ccSStefan Behrens errors_wait++; 3658387125fcSChris Mason continue; 3659387125fcSChris Mason } 3660e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3661ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3662387125fcSChris Mason continue; 3663387125fcSChris Mason 36644fc6441aSAnand Jain ret = wait_dev_flush(dev); 3665401b41e5SAnand Jain if (ret) { 3666401b41e5SAnand Jain dev->last_flush_error = ret; 366766b4993eSDavid Sterba btrfs_dev_stat_inc_and_print(dev, 366866b4993eSDavid Sterba BTRFS_DEV_STAT_FLUSH_ERRS); 36695af3e8ccSStefan Behrens errors_wait++; 3670387125fcSChris Mason } 3671401b41e5SAnand Jain } 3672401b41e5SAnand Jain 3673cea7c8bfSAnand Jain if (errors_wait) { 3674401b41e5SAnand Jain /* 3675401b41e5SAnand Jain * At some point we need the status of all disks 3676401b41e5SAnand Jain * to arrive at the volume status. So error checking 3677401b41e5SAnand Jain * is being pushed to a separate loop. 3678401b41e5SAnand Jain */ 3679d10b82feSQu Wenruo return check_barrier_error(info); 3680401b41e5SAnand Jain } 3681387125fcSChris Mason return 0; 3682387125fcSChris Mason } 3683387125fcSChris Mason 3684943c6e99SZhao Lei int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 3685943c6e99SZhao Lei { 36868789f4feSZhao Lei int raid_type; 36878789f4feSZhao Lei int min_tolerated = INT_MAX; 3688943c6e99SZhao Lei 36898789f4feSZhao Lei if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 36908789f4feSZhao Lei (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 36918c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 36928789f4feSZhao Lei btrfs_raid_array[BTRFS_RAID_SINGLE]. 36938789f4feSZhao Lei tolerated_failures); 3694943c6e99SZhao Lei 36958789f4feSZhao Lei for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 36968789f4feSZhao Lei if (raid_type == BTRFS_RAID_SINGLE) 36978789f4feSZhao Lei continue; 369841a6e891SAnand Jain if (!(flags & btrfs_raid_array[raid_type].bg_flag)) 36998789f4feSZhao Lei continue; 37008c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 37018789f4feSZhao Lei btrfs_raid_array[raid_type]. 37028789f4feSZhao Lei tolerated_failures); 37038789f4feSZhao Lei } 3704943c6e99SZhao Lei 37058789f4feSZhao Lei if (min_tolerated == INT_MAX) { 3706ab8d0fc4SJeff Mahoney pr_warn("BTRFS: unknown raid flag: %llu", flags); 37078789f4feSZhao Lei min_tolerated = 0; 37088789f4feSZhao Lei } 37098789f4feSZhao Lei 37108789f4feSZhao Lei return min_tolerated; 3711943c6e99SZhao Lei } 3712943c6e99SZhao Lei 3713eece6a9cSDavid Sterba int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) 3714f2984462SChris Mason { 3715e5e9a520SChris Mason struct list_head *head; 3716f2984462SChris Mason struct btrfs_device *dev; 3717a061fc8dSChris Mason struct btrfs_super_block *sb; 3718f2984462SChris Mason struct btrfs_dev_item *dev_item; 3719f2984462SChris Mason int ret; 3720f2984462SChris Mason int do_barriers; 3721a236aed1SChris Mason int max_errors; 3722a236aed1SChris Mason int total_errors = 0; 3723a061fc8dSChris Mason u64 flags; 3724f2984462SChris Mason 37250b246afaSJeff Mahoney do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); 3726fed3b381SLiu Bo 3727fed3b381SLiu Bo /* 3728fed3b381SLiu Bo * max_mirrors == 0 indicates we're from commit_transaction, 3729fed3b381SLiu Bo * not from fsync where the tree roots in fs_info have not 3730fed3b381SLiu Bo * been consistent on disk. 3731fed3b381SLiu Bo */ 3732fed3b381SLiu Bo if (max_mirrors == 0) 37330b246afaSJeff Mahoney backup_super_roots(fs_info); 3734f2984462SChris Mason 37350b246afaSJeff Mahoney sb = fs_info->super_for_commit; 3736a061fc8dSChris Mason dev_item = &sb->dev_item; 3737e5e9a520SChris Mason 37380b246afaSJeff Mahoney mutex_lock(&fs_info->fs_devices->device_list_mutex); 37390b246afaSJeff Mahoney head = &fs_info->fs_devices->devices; 37400b246afaSJeff Mahoney max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; 3741387125fcSChris Mason 37425af3e8ccSStefan Behrens if (do_barriers) { 37430b246afaSJeff Mahoney ret = barrier_all_devices(fs_info); 37445af3e8ccSStefan Behrens if (ret) { 37455af3e8ccSStefan Behrens mutex_unlock( 37460b246afaSJeff Mahoney &fs_info->fs_devices->device_list_mutex); 37470b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 37485af3e8ccSStefan Behrens "errors while submitting device barriers."); 37495af3e8ccSStefan Behrens return ret; 37505af3e8ccSStefan Behrens } 37515af3e8ccSStefan Behrens } 3752387125fcSChris Mason 37531538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3754dfe25020SChris Mason if (!dev->bdev) { 3755dfe25020SChris Mason total_errors++; 3756dfe25020SChris Mason continue; 3757dfe25020SChris Mason } 3758e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3759ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3760dfe25020SChris Mason continue; 3761dfe25020SChris Mason 37622b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 3763a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 3764a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 37657df69d3eSMiao Xie btrfs_set_stack_device_total_bytes(dev_item, 3766935e5cc9SMiao Xie dev->commit_total_bytes); 3767ce7213c7SMiao Xie btrfs_set_stack_device_bytes_used(dev_item, 3768ce7213c7SMiao Xie dev->commit_bytes_used); 3769a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 3770a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 3771a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 3772a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 37737239ff4bSNikolay Borisov memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, 37747239ff4bSNikolay Borisov BTRFS_FSID_SIZE); 3775a512bbf8SYan Zheng 3776a061fc8dSChris Mason flags = btrfs_super_flags(sb); 3777a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 3778f2984462SChris Mason 377975cb857dSQu Wenruo ret = btrfs_validate_write_super(fs_info, sb); 378075cb857dSQu Wenruo if (ret < 0) { 378175cb857dSQu Wenruo mutex_unlock(&fs_info->fs_devices->device_list_mutex); 378275cb857dSQu Wenruo btrfs_handle_fs_error(fs_info, -EUCLEAN, 378375cb857dSQu Wenruo "unexpected superblock corruption detected"); 378475cb857dSQu Wenruo return -EUCLEAN; 378575cb857dSQu Wenruo } 378675cb857dSQu Wenruo 3787abbb3b8eSDavid Sterba ret = write_dev_supers(dev, sb, max_mirrors); 3788a236aed1SChris Mason if (ret) 3789a236aed1SChris Mason total_errors++; 3790f2984462SChris Mason } 3791a236aed1SChris Mason if (total_errors > max_errors) { 37920b246afaSJeff Mahoney btrfs_err(fs_info, "%d errors while writing supers", 3793d397712bSChris Mason total_errors); 37940b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 379579787eaaSJeff Mahoney 37969d565ba4SStefan Behrens /* FUA is masked off if unsupported and can't be the reason */ 37970b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 37980b246afaSJeff Mahoney "%d errors while writing supers", 37990b246afaSJeff Mahoney total_errors); 38009d565ba4SStefan Behrens return -EIO; 3801a236aed1SChris Mason } 3802f2984462SChris Mason 3803a512bbf8SYan Zheng total_errors = 0; 38041538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3805dfe25020SChris Mason if (!dev->bdev) 3806dfe25020SChris Mason continue; 3807e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3808ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3809dfe25020SChris Mason continue; 3810dfe25020SChris Mason 3811abbb3b8eSDavid Sterba ret = wait_dev_supers(dev, max_mirrors); 3812a512bbf8SYan Zheng if (ret) 38131259ab75SChris Mason total_errors++; 3814f2984462SChris Mason } 38150b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 3816a236aed1SChris Mason if (total_errors > max_errors) { 38170b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 38180b246afaSJeff Mahoney "%d errors while writing supers", 38190b246afaSJeff Mahoney total_errors); 382079787eaaSJeff Mahoney return -EIO; 3821a236aed1SChris Mason } 3822f2984462SChris Mason return 0; 3823f2984462SChris Mason } 3824f2984462SChris Mason 3825cb517eabSMiao Xie /* Drop a fs root from the radix tree and free it. */ 3826cb517eabSMiao Xie void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 3827cb517eabSMiao Xie struct btrfs_root *root) 38282619ba1fSChris Mason { 38294df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 38302619ba1fSChris Mason radix_tree_delete(&fs_info->fs_roots_radix, 38312619ba1fSChris Mason (unsigned long)root->root_key.objectid); 38324df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 383376dda93cSYan, Zheng 383476dda93cSYan, Zheng if (btrfs_root_refs(&root->root_item) == 0) 383576dda93cSYan, Zheng synchronize_srcu(&fs_info->subvol_srcu); 383676dda93cSYan, Zheng 38371c1ea4f7SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 38383321719eSLiu Bo btrfs_free_log(NULL, root); 38391c1ea4f7SLiu Bo if (root->reloc_root) { 38401c1ea4f7SLiu Bo free_extent_buffer(root->reloc_root->node); 38411c1ea4f7SLiu Bo free_extent_buffer(root->reloc_root->commit_root); 38421c1ea4f7SLiu Bo btrfs_put_fs_root(root->reloc_root); 38431c1ea4f7SLiu Bo root->reloc_root = NULL; 38441c1ea4f7SLiu Bo } 38451c1ea4f7SLiu Bo } 38463321719eSLiu Bo 3847faa2dbf0SJosef Bacik if (root->free_ino_pinned) 3848581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_pinned); 3849faa2dbf0SJosef Bacik if (root->free_ino_ctl) 3850581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_ctl); 385184db5ccfSDavid Sterba btrfs_free_fs_root(root); 38524df27c4dSYan, Zheng } 38534df27c4dSYan, Zheng 385484db5ccfSDavid Sterba void btrfs_free_fs_root(struct btrfs_root *root) 38554df27c4dSYan, Zheng { 385657cdc8dbSDavid Sterba iput(root->ino_cache_inode); 38574df27c4dSYan, Zheng WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 38580ee5dc67SAl Viro if (root->anon_dev) 38590ee5dc67SAl Viro free_anon_bdev(root->anon_dev); 38608257b2dcSMiao Xie if (root->subv_writers) 38618257b2dcSMiao Xie btrfs_free_subvolume_writers(root->subv_writers); 38625f39d397SChris Mason free_extent_buffer(root->node); 38635f39d397SChris Mason free_extent_buffer(root->commit_root); 3864581bb050SLi Zefan kfree(root->free_ino_ctl); 3865581bb050SLi Zefan kfree(root->free_ino_pinned); 3866b0feb9d9SMiao Xie btrfs_put_fs_root(root); 38672619ba1fSChris Mason } 38682619ba1fSChris Mason 3869c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 3870c146afadSYan Zheng { 3871c146afadSYan Zheng u64 root_objectid = 0; 3872c146afadSYan Zheng struct btrfs_root *gang[8]; 387365d33fd7SQu Wenruo int i = 0; 387465d33fd7SQu Wenruo int err = 0; 387565d33fd7SQu Wenruo unsigned int ret = 0; 387665d33fd7SQu Wenruo int index; 3877c146afadSYan Zheng 3878c146afadSYan Zheng while (1) { 387965d33fd7SQu Wenruo index = srcu_read_lock(&fs_info->subvol_srcu); 3880c146afadSYan Zheng ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 3881c146afadSYan Zheng (void **)gang, root_objectid, 3882c146afadSYan Zheng ARRAY_SIZE(gang)); 388365d33fd7SQu Wenruo if (!ret) { 388465d33fd7SQu Wenruo srcu_read_unlock(&fs_info->subvol_srcu, index); 3885c146afadSYan Zheng break; 388665d33fd7SQu Wenruo } 38875d4f98a2SYan Zheng root_objectid = gang[ret - 1]->root_key.objectid + 1; 388866b4ffd1SJosef Bacik 388965d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 389065d33fd7SQu Wenruo /* Avoid to grab roots in dead_roots */ 389165d33fd7SQu Wenruo if (btrfs_root_refs(&gang[i]->root_item) == 0) { 389265d33fd7SQu Wenruo gang[i] = NULL; 389365d33fd7SQu Wenruo continue; 389465d33fd7SQu Wenruo } 389565d33fd7SQu Wenruo /* grab all the search result for later use */ 389665d33fd7SQu Wenruo gang[i] = btrfs_grab_fs_root(gang[i]); 389765d33fd7SQu Wenruo } 389865d33fd7SQu Wenruo srcu_read_unlock(&fs_info->subvol_srcu, index); 389965d33fd7SQu Wenruo 390065d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 390165d33fd7SQu Wenruo if (!gang[i]) 390265d33fd7SQu Wenruo continue; 3903c146afadSYan Zheng root_objectid = gang[i]->root_key.objectid; 390466b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(gang[i]); 390566b4ffd1SJosef Bacik if (err) 390665d33fd7SQu Wenruo break; 390765d33fd7SQu Wenruo btrfs_put_fs_root(gang[i]); 3908c146afadSYan Zheng } 3909c146afadSYan Zheng root_objectid++; 3910c146afadSYan Zheng } 391165d33fd7SQu Wenruo 391265d33fd7SQu Wenruo /* release the uncleaned roots due to error */ 391365d33fd7SQu Wenruo for (; i < ret; i++) { 391465d33fd7SQu Wenruo if (gang[i]) 391565d33fd7SQu Wenruo btrfs_put_fs_root(gang[i]); 391665d33fd7SQu Wenruo } 391765d33fd7SQu Wenruo return err; 3918c146afadSYan Zheng } 3919c146afadSYan Zheng 39206bccf3abSJeff Mahoney int btrfs_commit_super(struct btrfs_fs_info *fs_info) 3921c146afadSYan Zheng { 39226bccf3abSJeff Mahoney struct btrfs_root *root = fs_info->tree_root; 3923c146afadSYan Zheng struct btrfs_trans_handle *trans; 3924c146afadSYan Zheng 39250b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 39262ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 39270b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 39280b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 3929c71bf099SYan, Zheng 3930c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 39310b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 39320b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 3933c71bf099SYan, Zheng 39347a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 39353612b495STsutomu Itoh if (IS_ERR(trans)) 39363612b495STsutomu Itoh return PTR_ERR(trans); 39373a45bb20SJeff Mahoney return btrfs_commit_transaction(trans); 3938c146afadSYan Zheng } 3939c146afadSYan Zheng 3940b105e927SDavid Sterba void __cold close_ctree(struct btrfs_fs_info *fs_info) 3941eb60ceacSChris Mason { 3942c146afadSYan Zheng int ret; 3943e089f05cSChris Mason 3944afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); 3945d6fd0ae2SOmar Sandoval /* 3946d6fd0ae2SOmar Sandoval * We don't want the cleaner to start new transactions, add more delayed 3947d6fd0ae2SOmar Sandoval * iputs, etc. while we're closing. We can't use kthread_stop() yet 3948d6fd0ae2SOmar Sandoval * because that frees the task_struct, and the transaction kthread might 3949d6fd0ae2SOmar Sandoval * still try to wake up the cleaner. 3950d6fd0ae2SOmar Sandoval */ 3951d6fd0ae2SOmar Sandoval kthread_park(fs_info->cleaner_kthread); 3952a2135011SChris Mason 39537343dd61SJustin Maggard /* wait for the qgroup rescan worker to stop */ 3954d06f23d6SJeff Mahoney btrfs_qgroup_wait_for_completion(fs_info, false); 39557343dd61SJustin Maggard 3956803b2f54SStefan Behrens /* wait for the uuid_scan task to finish */ 3957803b2f54SStefan Behrens down(&fs_info->uuid_tree_rescan_sem); 3958803b2f54SStefan Behrens /* avoid complains from lockdep et al., set sem back to initial state */ 3959803b2f54SStefan Behrens up(&fs_info->uuid_tree_rescan_sem); 3960803b2f54SStefan Behrens 3961837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 3962aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 3963837d5b6eSIlya Dryomov 39648dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 39658dabb742SStefan Behrens 3966aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 39674cb5300bSChris Mason 39684cb5300bSChris Mason /* wait for any defraggers to finish */ 39694cb5300bSChris Mason wait_event(fs_info->transaction_wait, 39704cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 39714cb5300bSChris Mason 39724cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 397326176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 39744cb5300bSChris Mason 397521c7e756SMiao Xie cancel_work_sync(&fs_info->async_reclaim_work); 397621c7e756SMiao Xie 3977b0643e59SDennis Zhou /* Cancel or finish ongoing discard work */ 3978b0643e59SDennis Zhou btrfs_discard_cleanup(fs_info); 3979b0643e59SDennis Zhou 3980bc98a42cSDavid Howells if (!sb_rdonly(fs_info->sb)) { 3981e44163e1SJeff Mahoney /* 3982d6fd0ae2SOmar Sandoval * The cleaner kthread is stopped, so do one final pass over 3983d6fd0ae2SOmar Sandoval * unused block groups. 3984e44163e1SJeff Mahoney */ 39850b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 3986e44163e1SJeff Mahoney 39876bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 3988d397712bSChris Mason if (ret) 398904892340SEric Sandeen btrfs_err(fs_info, "commit super ret %d", ret); 3990c146afadSYan Zheng } 3991ed2ff2cbSChris Mason 3992af722733SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) || 3993af722733SLiu Bo test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) 39942ff7e61eSJeff Mahoney btrfs_error_commit_super(fs_info); 3995acce952bSliubo 3996e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 3997e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 39988929ecfaSYan, Zheng 3999e187831eSJosef Bacik ASSERT(list_empty(&fs_info->delayed_iputs)); 4000afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); 4001f25784b3SYan Zheng 400204892340SEric Sandeen btrfs_free_qgroup_config(fs_info); 4003fe816d0fSNikolay Borisov ASSERT(list_empty(&fs_info->delalloc_roots)); 4004bcef60f2SArne Jansen 4005963d678bSMiao Xie if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 400604892340SEric Sandeen btrfs_info(fs_info, "at unmount delalloc count %lld", 4007963d678bSMiao Xie percpu_counter_sum(&fs_info->delalloc_bytes)); 4008b0c68f8bSChris Mason } 400931153d81SYan Zheng 40104297ff84SJosef Bacik if (percpu_counter_sum(&fs_info->dio_bytes)) 40114297ff84SJosef Bacik btrfs_info(fs_info, "at unmount dio bytes count %lld", 40124297ff84SJosef Bacik percpu_counter_sum(&fs_info->dio_bytes)); 40134297ff84SJosef Bacik 40146618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 4015b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 40165ac1d209SJeff Mahoney 4017faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 4018d10c5f31SChris Mason 40191a4319ccSLiu Bo btrfs_put_block_group_cache(fs_info); 40201a4319ccSLiu Bo 4021de348ee0SWang Shilong /* 4022de348ee0SWang Shilong * we must make sure there is not any read request to 4023de348ee0SWang Shilong * submit after we stopping all workers. 4024de348ee0SWang Shilong */ 4025de348ee0SWang Shilong invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 402696192499SJosef Bacik btrfs_stop_all_workers(fs_info); 402796192499SJosef Bacik 4028afcdd129SJosef Bacik clear_bit(BTRFS_FS_OPEN, &fs_info->flags); 40294273eaffSAnand Jain free_root_pointers(fs_info, true); 40309ad6b7bcSChris Mason 40314e19443dSJosef Bacik /* 40324e19443dSJosef Bacik * We must free the block groups after dropping the fs_roots as we could 40334e19443dSJosef Bacik * have had an IO error and have left over tree log blocks that aren't 40344e19443dSJosef Bacik * cleaned up until the fs roots are freed. This makes the block group 40354e19443dSJosef Bacik * accounting appear to be wrong because there's pending reserved bytes, 40364e19443dSJosef Bacik * so make sure we do the block group cleanup afterwards. 40374e19443dSJosef Bacik */ 40384e19443dSJosef Bacik btrfs_free_block_groups(fs_info); 40394e19443dSJosef Bacik 404013e6c37bSJosef Bacik iput(fs_info->btree_inode); 4041d6bfde87SChris Mason 404221adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 40430b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) 40442ff7e61eSJeff Mahoney btrfsic_unmount(fs_info->fs_devices); 404521adbd5cSStefan Behrens #endif 404621adbd5cSStefan Behrens 40470b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 404868c94e55SNikolay Borisov btrfs_close_devices(fs_info->fs_devices); 4049b248a415SChris Mason 4050e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 4051963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 40524297ff84SJosef Bacik percpu_counter_destroy(&fs_info->dio_bytes); 40537f8d236aSDavid Sterba percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 405476dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 40550b86a832SChris Mason 4056bfcea1c6SJohannes Thumshirn btrfs_free_csum_hash(fs_info); 405753b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 4058fd708b81SJosef Bacik btrfs_free_ref_cache(fs_info); 4059eb60ceacSChris Mason } 4060eb60ceacSChris Mason 4061b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 4062b9fab919SChris Mason int atomic) 4063ccd467d6SChris Mason { 40641259ab75SChris Mason int ret; 4065727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 40661259ab75SChris Mason 40670b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 40681259ab75SChris Mason if (!ret) 40691259ab75SChris Mason return ret; 40701259ab75SChris Mason 40711259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 4072b9fab919SChris Mason parent_transid, atomic); 4073b9fab919SChris Mason if (ret == -EAGAIN) 4074b9fab919SChris Mason return ret; 40751259ab75SChris Mason return !ret; 40765f39d397SChris Mason } 40776702ed49SChris Mason 40785f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 40795f39d397SChris Mason { 40800b246afaSJeff Mahoney struct btrfs_fs_info *fs_info; 408106ea65a3SJosef Bacik struct btrfs_root *root; 40825f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 4083b9473439SChris Mason int was_dirty; 4084b4ce94deSChris Mason 408506ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 408606ea65a3SJosef Bacik /* 408706ea65a3SJosef Bacik * This is a fast path so only do this check if we have sanity tests 408852042d8eSAndrea Gelmini * enabled. Normal people shouldn't be using unmapped buffers as dirty 408906ea65a3SJosef Bacik * outside of the sanity tests. 409006ea65a3SJosef Bacik */ 4091b0132a3bSNikolay Borisov if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) 409206ea65a3SJosef Bacik return; 409306ea65a3SJosef Bacik #endif 409406ea65a3SJosef Bacik root = BTRFS_I(buf->pages[0]->mapping->host)->root; 40950b246afaSJeff Mahoney fs_info = root->fs_info; 4096b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 40970b246afaSJeff Mahoney if (transid != fs_info->generation) 40985d163e0eSJeff Mahoney WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", 40990b246afaSJeff Mahoney buf->start, transid, fs_info->generation); 41000b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 4101e2d84521SMiao Xie if (!was_dirty) 4102104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 4103e2d84521SMiao Xie buf->len, 41040b246afaSJeff Mahoney fs_info->dirty_metadata_batch); 41051f21ef0aSFilipe Manana #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 410669fc6cbbSQu Wenruo /* 410769fc6cbbSQu Wenruo * Since btrfs_mark_buffer_dirty() can be called with item pointer set 410869fc6cbbSQu Wenruo * but item data not updated. 410969fc6cbbSQu Wenruo * So here we should only check item pointers, not item data. 411069fc6cbbSQu Wenruo */ 411169fc6cbbSQu Wenruo if (btrfs_header_level(buf) == 0 && 4112cfdaad5eSDavid Sterba btrfs_check_leaf_relaxed(buf)) { 4113a4f78750SDavid Sterba btrfs_print_leaf(buf); 41141f21ef0aSFilipe Manana ASSERT(0); 41151f21ef0aSFilipe Manana } 41161f21ef0aSFilipe Manana #endif 4117eb60ceacSChris Mason } 4118eb60ceacSChris Mason 41192ff7e61eSJeff Mahoney static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, 4120b53d3f5dSLiu Bo int flush_delayed) 412135b7e476SChris Mason { 4122188de649SChris Mason /* 4123188de649SChris Mason * looks as though older kernels can get into trouble with 4124188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 4125188de649SChris Mason */ 4126e2d84521SMiao Xie int ret; 4127d6bfde87SChris Mason 41286933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 4129d6bfde87SChris Mason return; 4130d6bfde87SChris Mason 4131b53d3f5dSLiu Bo if (flush_delayed) 41322ff7e61eSJeff Mahoney btrfs_balance_delayed_items(fs_info); 413316cdcec7SMiao Xie 4134d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 4135d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 4136d814a491SEthan Lien fs_info->dirty_metadata_batch); 4137e2d84521SMiao Xie if (ret > 0) { 41380b246afaSJeff Mahoney balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); 413916cdcec7SMiao Xie } 414016cdcec7SMiao Xie } 414116cdcec7SMiao Xie 41422ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) 414316cdcec7SMiao Xie { 41442ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 1); 414535b7e476SChris Mason } 4146b53d3f5dSLiu Bo 41472ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) 4148b53d3f5dSLiu Bo { 41492ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 0); 4150d6bfde87SChris Mason } 41516b80053dSChris Mason 4152581c1760SQu Wenruo int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level, 4153581c1760SQu Wenruo struct btrfs_key *first_key) 41546b80053dSChris Mason { 41555ab12d1fSDavid Sterba return btree_read_extent_buffer_pages(buf, parent_transid, 4156581c1760SQu Wenruo level, first_key); 41576b80053dSChris Mason } 41580da5468fSChris Mason 41592ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) 4160acce952bSliubo { 4161fe816d0fSNikolay Borisov /* cleanup FS via transaction */ 4162fe816d0fSNikolay Borisov btrfs_cleanup_transaction(fs_info); 4163fe816d0fSNikolay Borisov 41640b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 41652ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 41660b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 4167acce952bSliubo 41680b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 41690b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4170acce952bSliubo } 4171acce952bSliubo 4172143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4173acce952bSliubo { 4174acce952bSliubo struct btrfs_ordered_extent *ordered; 4175acce952bSliubo 4176199c2a9cSMiao Xie spin_lock(&root->ordered_extent_lock); 4177779880efSJosef Bacik /* 4178779880efSJosef Bacik * This will just short circuit the ordered completion stuff which will 4179779880efSJosef Bacik * make sure the ordered extent gets properly cleaned up. 4180779880efSJosef Bacik */ 4181199c2a9cSMiao Xie list_for_each_entry(ordered, &root->ordered_extents, 4182779880efSJosef Bacik root_extent_list) 4183779880efSJosef Bacik set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4184199c2a9cSMiao Xie spin_unlock(&root->ordered_extent_lock); 4185199c2a9cSMiao Xie } 4186199c2a9cSMiao Xie 4187199c2a9cSMiao Xie static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4188199c2a9cSMiao Xie { 4189199c2a9cSMiao Xie struct btrfs_root *root; 4190199c2a9cSMiao Xie struct list_head splice; 4191199c2a9cSMiao Xie 4192199c2a9cSMiao Xie INIT_LIST_HEAD(&splice); 4193199c2a9cSMiao Xie 4194199c2a9cSMiao Xie spin_lock(&fs_info->ordered_root_lock); 4195199c2a9cSMiao Xie list_splice_init(&fs_info->ordered_roots, &splice); 4196199c2a9cSMiao Xie while (!list_empty(&splice)) { 4197199c2a9cSMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4198199c2a9cSMiao Xie ordered_root); 41991de2cfdeSJosef Bacik list_move_tail(&root->ordered_root, 42001de2cfdeSJosef Bacik &fs_info->ordered_roots); 4201199c2a9cSMiao Xie 42022a85d9caSLiu Bo spin_unlock(&fs_info->ordered_root_lock); 4203199c2a9cSMiao Xie btrfs_destroy_ordered_extents(root); 4204199c2a9cSMiao Xie 42052a85d9caSLiu Bo cond_resched(); 42062a85d9caSLiu Bo spin_lock(&fs_info->ordered_root_lock); 4207199c2a9cSMiao Xie } 4208199c2a9cSMiao Xie spin_unlock(&fs_info->ordered_root_lock); 420974d5d229SJosef Bacik 421074d5d229SJosef Bacik /* 421174d5d229SJosef Bacik * We need this here because if we've been flipped read-only we won't 421274d5d229SJosef Bacik * get sync() from the umount, so we need to make sure any ordered 421374d5d229SJosef Bacik * extents that haven't had their dirty pages IO start writeout yet 421474d5d229SJosef Bacik * actually get run and error out properly. 421574d5d229SJosef Bacik */ 421674d5d229SJosef Bacik btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 4217acce952bSliubo } 4218acce952bSliubo 421935a3621bSStefan Behrens static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 42202ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4221acce952bSliubo { 4222acce952bSliubo struct rb_node *node; 4223acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 4224acce952bSliubo struct btrfs_delayed_ref_node *ref; 4225acce952bSliubo int ret = 0; 4226acce952bSliubo 4227acce952bSliubo delayed_refs = &trans->delayed_refs; 4228acce952bSliubo 4229acce952bSliubo spin_lock(&delayed_refs->lock); 4230d7df2c79SJosef Bacik if (atomic_read(&delayed_refs->num_entries) == 0) { 4231cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 42320b246afaSJeff Mahoney btrfs_info(fs_info, "delayed_refs has NO entry"); 4233acce952bSliubo return ret; 4234acce952bSliubo } 4235acce952bSliubo 42365c9d028bSLiu Bo while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { 4237d7df2c79SJosef Bacik struct btrfs_delayed_ref_head *head; 42380e0adbcfSJosef Bacik struct rb_node *n; 4239e78417d1SJosef Bacik bool pin_bytes = false; 4240acce952bSliubo 4241d7df2c79SJosef Bacik head = rb_entry(node, struct btrfs_delayed_ref_head, 4242d7df2c79SJosef Bacik href_node); 42433069bd26SJosef Bacik if (btrfs_delayed_ref_lock(delayed_refs, head)) 4244b939d1abSJosef Bacik continue; 42453069bd26SJosef Bacik 4246d7df2c79SJosef Bacik spin_lock(&head->lock); 4247e3d03965SLiu Bo while ((n = rb_first_cached(&head->ref_tree)) != NULL) { 42480e0adbcfSJosef Bacik ref = rb_entry(n, struct btrfs_delayed_ref_node, 42490e0adbcfSJosef Bacik ref_node); 4250d7df2c79SJosef Bacik ref->in_tree = 0; 4251e3d03965SLiu Bo rb_erase_cached(&ref->ref_node, &head->ref_tree); 42520e0adbcfSJosef Bacik RB_CLEAR_NODE(&ref->ref_node); 42531d57ee94SWang Xiaoguang if (!list_empty(&ref->add_list)) 42541d57ee94SWang Xiaoguang list_del(&ref->add_list); 4255d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4256d7df2c79SJosef Bacik btrfs_put_delayed_ref(ref); 4257d7df2c79SJosef Bacik } 425854067ae9SJosef Bacik if (head->must_insert_reserved) 4259e78417d1SJosef Bacik pin_bytes = true; 426078a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 4261fa781ceaSJosef Bacik btrfs_delete_ref_head(delayed_refs, head); 4262d7df2c79SJosef Bacik spin_unlock(&head->lock); 4263acce952bSliubo spin_unlock(&delayed_refs->lock); 4264e78417d1SJosef Bacik mutex_unlock(&head->mutex); 4265acce952bSliubo 4266d7df2c79SJosef Bacik if (pin_bytes) 4267d278850eSJosef Bacik btrfs_pin_extent(fs_info, head->bytenr, 4268d278850eSJosef Bacik head->num_bytes, 1); 426931890da0SJosef Bacik btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); 4270d278850eSJosef Bacik btrfs_put_delayed_ref_head(head); 4271acce952bSliubo cond_resched(); 4272acce952bSliubo spin_lock(&delayed_refs->lock); 4273acce952bSliubo } 427481f7eb00SJeff Mahoney btrfs_qgroup_destroy_extent_records(trans); 4275acce952bSliubo 4276acce952bSliubo spin_unlock(&delayed_refs->lock); 4277acce952bSliubo 4278acce952bSliubo return ret; 4279acce952bSliubo } 4280acce952bSliubo 4281143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 4282acce952bSliubo { 4283acce952bSliubo struct btrfs_inode *btrfs_inode; 4284acce952bSliubo struct list_head splice; 4285acce952bSliubo 4286acce952bSliubo INIT_LIST_HEAD(&splice); 4287acce952bSliubo 4288eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4289eb73c1b7SMiao Xie list_splice_init(&root->delalloc_inodes, &splice); 4290acce952bSliubo 4291acce952bSliubo while (!list_empty(&splice)) { 4292fe816d0fSNikolay Borisov struct inode *inode = NULL; 4293eb73c1b7SMiao Xie btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 4294acce952bSliubo delalloc_inodes); 4295fe816d0fSNikolay Borisov __btrfs_del_delalloc_inode(root, btrfs_inode); 4296eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4297acce952bSliubo 4298fe816d0fSNikolay Borisov /* 4299fe816d0fSNikolay Borisov * Make sure we get a live inode and that it'll not disappear 4300fe816d0fSNikolay Borisov * meanwhile. 4301fe816d0fSNikolay Borisov */ 4302fe816d0fSNikolay Borisov inode = igrab(&btrfs_inode->vfs_inode); 4303fe816d0fSNikolay Borisov if (inode) { 4304fe816d0fSNikolay Borisov invalidate_inode_pages2(inode->i_mapping); 4305fe816d0fSNikolay Borisov iput(inode); 4306fe816d0fSNikolay Borisov } 4307eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4308acce952bSliubo } 4309eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4310eb73c1b7SMiao Xie } 4311eb73c1b7SMiao Xie 4312eb73c1b7SMiao Xie static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 4313eb73c1b7SMiao Xie { 4314eb73c1b7SMiao Xie struct btrfs_root *root; 4315eb73c1b7SMiao Xie struct list_head splice; 4316eb73c1b7SMiao Xie 4317eb73c1b7SMiao Xie INIT_LIST_HEAD(&splice); 4318eb73c1b7SMiao Xie 4319eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4320eb73c1b7SMiao Xie list_splice_init(&fs_info->delalloc_roots, &splice); 4321eb73c1b7SMiao Xie while (!list_empty(&splice)) { 4322eb73c1b7SMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4323eb73c1b7SMiao Xie delalloc_root); 4324eb73c1b7SMiao Xie root = btrfs_grab_fs_root(root); 4325eb73c1b7SMiao Xie BUG_ON(!root); 4326eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4327eb73c1b7SMiao Xie 4328eb73c1b7SMiao Xie btrfs_destroy_delalloc_inodes(root); 4329eb73c1b7SMiao Xie btrfs_put_fs_root(root); 4330eb73c1b7SMiao Xie 4331eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4332eb73c1b7SMiao Xie } 4333eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4334acce952bSliubo } 4335acce952bSliubo 43362ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 4337acce952bSliubo struct extent_io_tree *dirty_pages, 4338acce952bSliubo int mark) 4339acce952bSliubo { 4340acce952bSliubo int ret; 4341acce952bSliubo struct extent_buffer *eb; 4342acce952bSliubo u64 start = 0; 4343acce952bSliubo u64 end; 4344acce952bSliubo 4345acce952bSliubo while (1) { 4346acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 4347e6138876SJosef Bacik mark, NULL); 4348acce952bSliubo if (ret) 4349acce952bSliubo break; 4350acce952bSliubo 435191166212SDavid Sterba clear_extent_bits(dirty_pages, start, end, mark); 4352acce952bSliubo while (start <= end) { 43530b246afaSJeff Mahoney eb = find_extent_buffer(fs_info, start); 43540b246afaSJeff Mahoney start += fs_info->nodesize; 4355fd8b2b61SJosef Bacik if (!eb) 4356acce952bSliubo continue; 4357fd8b2b61SJosef Bacik wait_on_extent_buffer_writeback(eb); 4358acce952bSliubo 4359fd8b2b61SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 4360fd8b2b61SJosef Bacik &eb->bflags)) 4361fd8b2b61SJosef Bacik clear_extent_buffer_dirty(eb); 4362fd8b2b61SJosef Bacik free_extent_buffer_stale(eb); 4363acce952bSliubo } 4364acce952bSliubo } 4365acce952bSliubo 4366acce952bSliubo return ret; 4367acce952bSliubo } 4368acce952bSliubo 43692ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 4370acce952bSliubo struct extent_io_tree *pinned_extents) 4371acce952bSliubo { 4372acce952bSliubo struct extent_io_tree *unpin; 4373acce952bSliubo u64 start; 4374acce952bSliubo u64 end; 4375acce952bSliubo int ret; 4376ed0eaa14SLiu Bo bool loop = true; 4377acce952bSliubo 4378acce952bSliubo unpin = pinned_extents; 4379ed0eaa14SLiu Bo again: 4380acce952bSliubo while (1) { 43810e6ec385SFilipe Manana struct extent_state *cached_state = NULL; 43820e6ec385SFilipe Manana 4383fcd5e742SLu Fengqi /* 4384fcd5e742SLu Fengqi * The btrfs_finish_extent_commit() may get the same range as 4385fcd5e742SLu Fengqi * ours between find_first_extent_bit and clear_extent_dirty. 4386fcd5e742SLu Fengqi * Hence, hold the unused_bg_unpin_mutex to avoid double unpin 4387fcd5e742SLu Fengqi * the same extent range. 4388fcd5e742SLu Fengqi */ 4389fcd5e742SLu Fengqi mutex_lock(&fs_info->unused_bg_unpin_mutex); 4390acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 43910e6ec385SFilipe Manana EXTENT_DIRTY, &cached_state); 4392fcd5e742SLu Fengqi if (ret) { 4393fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4394acce952bSliubo break; 4395fcd5e742SLu Fengqi } 4396acce952bSliubo 43970e6ec385SFilipe Manana clear_extent_dirty(unpin, start, end, &cached_state); 43980e6ec385SFilipe Manana free_extent_state(cached_state); 43992ff7e61eSJeff Mahoney btrfs_error_unpin_extent_range(fs_info, start, end); 4400fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4401acce952bSliubo cond_resched(); 4402acce952bSliubo } 4403acce952bSliubo 4404ed0eaa14SLiu Bo if (loop) { 44050b246afaSJeff Mahoney if (unpin == &fs_info->freed_extents[0]) 44060b246afaSJeff Mahoney unpin = &fs_info->freed_extents[1]; 4407ed0eaa14SLiu Bo else 44080b246afaSJeff Mahoney unpin = &fs_info->freed_extents[0]; 4409ed0eaa14SLiu Bo loop = false; 4410ed0eaa14SLiu Bo goto again; 4411ed0eaa14SLiu Bo } 4412ed0eaa14SLiu Bo 4413acce952bSliubo return 0; 4414acce952bSliubo } 4415acce952bSliubo 441632da5386SDavid Sterba static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache) 4417c79a1751SLiu Bo { 4418c79a1751SLiu Bo struct inode *inode; 4419c79a1751SLiu Bo 4420c79a1751SLiu Bo inode = cache->io_ctl.inode; 4421c79a1751SLiu Bo if (inode) { 4422c79a1751SLiu Bo invalidate_inode_pages2(inode->i_mapping); 4423c79a1751SLiu Bo BTRFS_I(inode)->generation = 0; 4424c79a1751SLiu Bo cache->io_ctl.inode = NULL; 4425c79a1751SLiu Bo iput(inode); 4426c79a1751SLiu Bo } 4427c79a1751SLiu Bo btrfs_put_block_group(cache); 4428c79a1751SLiu Bo } 4429c79a1751SLiu Bo 4430c79a1751SLiu Bo void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, 44312ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4432c79a1751SLiu Bo { 443332da5386SDavid Sterba struct btrfs_block_group *cache; 4434c79a1751SLiu Bo 4435c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4436c79a1751SLiu Bo while (!list_empty(&cur_trans->dirty_bgs)) { 4437c79a1751SLiu Bo cache = list_first_entry(&cur_trans->dirty_bgs, 443832da5386SDavid Sterba struct btrfs_block_group, 4439c79a1751SLiu Bo dirty_list); 4440c79a1751SLiu Bo 4441c79a1751SLiu Bo if (!list_empty(&cache->io_list)) { 4442c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4443c79a1751SLiu Bo list_del_init(&cache->io_list); 4444c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 4445c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4446c79a1751SLiu Bo } 4447c79a1751SLiu Bo 4448c79a1751SLiu Bo list_del_init(&cache->dirty_list); 4449c79a1751SLiu Bo spin_lock(&cache->lock); 4450c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 4451c79a1751SLiu Bo spin_unlock(&cache->lock); 4452c79a1751SLiu Bo 4453c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4454c79a1751SLiu Bo btrfs_put_block_group(cache); 4455ba2c4d4eSJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 4456c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4457c79a1751SLiu Bo } 4458c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4459c79a1751SLiu Bo 446045ae2c18SNikolay Borisov /* 446145ae2c18SNikolay Borisov * Refer to the definition of io_bgs member for details why it's safe 446245ae2c18SNikolay Borisov * to use it without any locking 446345ae2c18SNikolay Borisov */ 4464c79a1751SLiu Bo while (!list_empty(&cur_trans->io_bgs)) { 4465c79a1751SLiu Bo cache = list_first_entry(&cur_trans->io_bgs, 446632da5386SDavid Sterba struct btrfs_block_group, 4467c79a1751SLiu Bo io_list); 4468c79a1751SLiu Bo 4469c79a1751SLiu Bo list_del_init(&cache->io_list); 4470c79a1751SLiu Bo spin_lock(&cache->lock); 4471c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 4472c79a1751SLiu Bo spin_unlock(&cache->lock); 4473c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 4474c79a1751SLiu Bo } 4475c79a1751SLiu Bo } 4476c79a1751SLiu Bo 447749b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 44782ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 447949b25e05SJeff Mahoney { 4480bbbf7243SNikolay Borisov struct btrfs_device *dev, *tmp; 4481bbbf7243SNikolay Borisov 44822ff7e61eSJeff Mahoney btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 4483c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->dirty_bgs)); 4484c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->io_bgs)); 4485c79a1751SLiu Bo 4486bbbf7243SNikolay Borisov list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list, 4487bbbf7243SNikolay Borisov post_commit_list) { 4488bbbf7243SNikolay Borisov list_del_init(&dev->post_commit_list); 4489bbbf7243SNikolay Borisov } 4490bbbf7243SNikolay Borisov 44912ff7e61eSJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, fs_info); 449249b25e05SJeff Mahoney 44934a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 44940b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 449549b25e05SJeff Mahoney 44964a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 44970b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 449849b25e05SJeff Mahoney 4499ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 450067cde344SMiao Xie 45012ff7e61eSJeff Mahoney btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, 450249b25e05SJeff Mahoney EXTENT_DIRTY); 45032ff7e61eSJeff Mahoney btrfs_destroy_pinned_extent(fs_info, 45040b246afaSJeff Mahoney fs_info->pinned_extents); 450549b25e05SJeff Mahoney 45064a9d8bdeSMiao Xie cur_trans->state =TRANS_STATE_COMPLETED; 45074a9d8bdeSMiao Xie wake_up(&cur_trans->commit_wait); 450849b25e05SJeff Mahoney } 450949b25e05SJeff Mahoney 45102ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) 4511acce952bSliubo { 4512acce952bSliubo struct btrfs_transaction *t; 4513acce952bSliubo 45140b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 4515acce952bSliubo 45160b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 45170b246afaSJeff Mahoney while (!list_empty(&fs_info->trans_list)) { 45180b246afaSJeff Mahoney t = list_first_entry(&fs_info->trans_list, 4519724e2315SJosef Bacik struct btrfs_transaction, list); 4520724e2315SJosef Bacik if (t->state >= TRANS_STATE_COMMIT_START) { 45219b64f57dSElena Reshetova refcount_inc(&t->use_count); 45220b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 45232ff7e61eSJeff Mahoney btrfs_wait_for_commit(fs_info, t->transid); 4524724e2315SJosef Bacik btrfs_put_transaction(t); 45250b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4526724e2315SJosef Bacik continue; 4527724e2315SJosef Bacik } 45280b246afaSJeff Mahoney if (t == fs_info->running_transaction) { 4529724e2315SJosef Bacik t->state = TRANS_STATE_COMMIT_DOING; 45300b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4531724e2315SJosef Bacik /* 4532724e2315SJosef Bacik * We wait for 0 num_writers since we don't hold a trans 4533724e2315SJosef Bacik * handle open currently for this transaction. 4534724e2315SJosef Bacik */ 4535724e2315SJosef Bacik wait_event(t->writer_wait, 4536724e2315SJosef Bacik atomic_read(&t->num_writers) == 0); 4537724e2315SJosef Bacik } else { 45380b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4539724e2315SJosef Bacik } 45402ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(t, fs_info); 4541724e2315SJosef Bacik 45420b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 45430b246afaSJeff Mahoney if (t == fs_info->running_transaction) 45440b246afaSJeff Mahoney fs_info->running_transaction = NULL; 4545724e2315SJosef Bacik list_del_init(&t->list); 45460b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4547a4abeea4SJosef Bacik 4548724e2315SJosef Bacik btrfs_put_transaction(t); 45492ff7e61eSJeff Mahoney trace_btrfs_transaction_commit(fs_info->tree_root); 45500b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4551724e2315SJosef Bacik } 45520b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 45530b246afaSJeff Mahoney btrfs_destroy_all_ordered_extents(fs_info); 4554ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 4555ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 45562ff7e61eSJeff Mahoney btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents); 45570b246afaSJeff Mahoney btrfs_destroy_all_delalloc_inodes(fs_info); 45580b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 4559acce952bSliubo 4560acce952bSliubo return 0; 4561acce952bSliubo } 4562acce952bSliubo 4563e8c9f186SDavid Sterba static const struct extent_io_ops btree_extent_io_ops = { 45644d53dddbSDavid Sterba /* mandatory callbacks */ 45650b86a832SChris Mason .submit_bio_hook = btree_submit_bio_hook, 45664d53dddbSDavid Sterba .readpage_end_io_hook = btree_readpage_end_io_hook, 45670da5468fSChris Mason }; 4568