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> 207e75bf3fSDavid Sterba #include <asm/unaligned.h> 21eb60ceacSChris Mason #include "ctree.h" 22eb60ceacSChris Mason #include "disk-io.h" 23e089f05cSChris Mason #include "transaction.h" 240f7d52f4SChris Mason #include "btrfs_inode.h" 250b86a832SChris Mason #include "volumes.h" 26db94535dSChris Mason #include "print-tree.h" 27925baeddSChris Mason #include "locking.h" 28e02119d5SChris Mason #include "tree-log.h" 29fa9c0d79SChris Mason #include "free-space-cache.h" 3070f6d82eSOmar Sandoval #include "free-space-tree.h" 31581bb050SLi Zefan #include "inode-map.h" 3221adbd5cSStefan Behrens #include "check-integrity.h" 33606686eeSJosef Bacik #include "rcu-string.h" 348dabb742SStefan Behrens #include "dev-replace.h" 3553b381b3SDavid Woodhouse #include "raid56.h" 365ac1d209SJeff Mahoney #include "sysfs.h" 37fcebe456SJosef Bacik #include "qgroup.h" 38ebb8765bSAnand Jain #include "compression.h" 39557ea5ddSQu Wenruo #include "tree-checker.h" 40fd708b81SJosef Bacik #include "ref-verify.h" 41eb60ceacSChris Mason 42de0022b9SJosef Bacik #ifdef CONFIG_X86 43de0022b9SJosef Bacik #include <asm/cpufeature.h> 44de0022b9SJosef Bacik #endif 45de0022b9SJosef Bacik 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, 206306e16ceSDavid Sterba struct page *page, size_t pg_offset, u64 start, u64 len, 2075f39d397SChris Mason int create) 2085f39d397SChris Mason { 2093ffbd68cSDavid Sterba struct btrfs_fs_info *fs_info = inode->root->fs_info; 210fc4f21b1SNikolay Borisov struct extent_map_tree *em_tree = &inode->extent_tree; 2115f39d397SChris Mason struct extent_map *em; 2125f39d397SChris Mason int ret; 2135f39d397SChris Mason 214890871beSChris Mason read_lock(&em_tree->lock); 215d1310b2eSChris Mason em = lookup_extent_mapping(em_tree, start, len); 216a061fc8dSChris Mason if (em) { 2170b246afaSJeff Mahoney em->bdev = fs_info->fs_devices->latest_bdev; 218890871beSChris Mason read_unlock(&em_tree->lock); 2195f39d397SChris Mason goto out; 220a061fc8dSChris Mason } 221890871beSChris Mason read_unlock(&em_tree->lock); 2227b13b7b1SChris Mason 223172ddd60SDavid Sterba em = alloc_extent_map(); 2245f39d397SChris Mason if (!em) { 2255f39d397SChris Mason em = ERR_PTR(-ENOMEM); 2265f39d397SChris Mason goto out; 2275f39d397SChris Mason } 2285f39d397SChris Mason em->start = 0; 2290afbaf8cSChris Mason em->len = (u64)-1; 230c8b97818SChris Mason em->block_len = (u64)-1; 2315f39d397SChris Mason em->block_start = 0; 2320b246afaSJeff Mahoney em->bdev = fs_info->fs_devices->latest_bdev; 233d1310b2eSChris Mason 234890871beSChris Mason write_lock(&em_tree->lock); 23509a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 2365f39d397SChris Mason if (ret == -EEXIST) { 2375f39d397SChris Mason free_extent_map(em); 2387b13b7b1SChris Mason em = lookup_extent_mapping(em_tree, start, len); 239b4f359abSTsutomu Itoh if (!em) 2400433f20dSTsutomu Itoh em = ERR_PTR(-EIO); 2415f39d397SChris Mason } else if (ret) { 2427b13b7b1SChris Mason free_extent_map(em); 2430433f20dSTsutomu Itoh em = ERR_PTR(ret); 2445f39d397SChris Mason } 245890871beSChris Mason write_unlock(&em_tree->lock); 2467b13b7b1SChris Mason 2475f39d397SChris Mason out: 2485f39d397SChris Mason return em; 2495f39d397SChris Mason } 2505f39d397SChris Mason 2519ed57367SDavid Sterba u32 btrfs_csum_data(const char *data, u32 seed, size_t len) 25219c00ddcSChris Mason { 2539678c543SNikolay Borisov return crc32c(seed, data, len); 25419c00ddcSChris Mason } 25519c00ddcSChris Mason 2560b5e3dafSDomagoj Tršan void btrfs_csum_final(u32 crc, u8 *result) 25719c00ddcSChris Mason { 2587e75bf3fSDavid Sterba put_unaligned_le32(~crc, result); 25919c00ddcSChris Mason } 26019c00ddcSChris Mason 261d352ac68SChris Mason /* 262d352ac68SChris Mason * compute the csum for a btree block, and either verify it or write it 263d352ac68SChris Mason * into the csum field of the block. 264d352ac68SChris Mason */ 26501d58472SDaniel Dressler static int csum_tree_block(struct btrfs_fs_info *fs_info, 26601d58472SDaniel Dressler struct extent_buffer *buf, 26719c00ddcSChris Mason int verify) 26819c00ddcSChris Mason { 26901d58472SDaniel Dressler u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); 27071a63551SDavid Sterba char result[BTRFS_CSUM_SIZE]; 27119c00ddcSChris Mason unsigned long len; 27219c00ddcSChris Mason unsigned long cur_len; 27319c00ddcSChris Mason unsigned long offset = BTRFS_CSUM_SIZE; 27419c00ddcSChris Mason char *kaddr; 27519c00ddcSChris Mason unsigned long map_start; 27619c00ddcSChris Mason unsigned long map_len; 27719c00ddcSChris Mason int err; 27819c00ddcSChris Mason u32 crc = ~(u32)0; 27919c00ddcSChris Mason 28019c00ddcSChris Mason len = buf->len - offset; 28119c00ddcSChris Mason while (len > 0) { 282d2e174d5SJohannes Thumshirn /* 283d2e174d5SJohannes Thumshirn * Note: we don't need to check for the err == 1 case here, as 284d2e174d5SJohannes Thumshirn * with the given combination of 'start = BTRFS_CSUM_SIZE (32)' 285d2e174d5SJohannes Thumshirn * and 'min_len = 32' and the currently implemented mapping 286d2e174d5SJohannes Thumshirn * algorithm we cannot cross a page boundary. 287d2e174d5SJohannes Thumshirn */ 28819c00ddcSChris Mason err = map_private_extent_buffer(buf, offset, 32, 289a6591715SChris Mason &kaddr, &map_start, &map_len); 290d397712bSChris Mason if (err) 2918bd98f0eSAlex Lyakas return err; 29219c00ddcSChris Mason cur_len = min(len, map_len - (offset - map_start)); 293b0496686SLiu Bo crc = btrfs_csum_data(kaddr + offset - map_start, 29419c00ddcSChris Mason crc, cur_len); 29519c00ddcSChris Mason len -= cur_len; 29619c00ddcSChris Mason offset += cur_len; 29719c00ddcSChris Mason } 29871a63551SDavid Sterba memset(result, 0, BTRFS_CSUM_SIZE); 299607d432dSJosef Bacik 30019c00ddcSChris Mason btrfs_csum_final(crc, result); 30119c00ddcSChris Mason 30219c00ddcSChris Mason if (verify) { 303607d432dSJosef Bacik if (memcmp_extent_buffer(buf, result, 0, csum_size)) { 304e4204dedSChris Mason u32 val; 305e4204dedSChris Mason u32 found = 0; 306607d432dSJosef Bacik memcpy(&found, result, csum_size); 307e4204dedSChris Mason 308607d432dSJosef Bacik read_extent_buffer(buf, &val, 0, csum_size); 30994647322SDavid Sterba btrfs_warn_rl(fs_info, 3105d163e0eSJeff Mahoney "%s checksum verify failed on %llu wanted %X found %X level %d", 31101d58472SDaniel Dressler fs_info->sb->s_id, buf->start, 312c1c9ff7cSGeert Uytterhoeven val, found, btrfs_header_level(buf)); 3138bd98f0eSAlex Lyakas return -EUCLEAN; 31419c00ddcSChris Mason } 31519c00ddcSChris Mason } else { 316607d432dSJosef Bacik write_extent_buffer(buf, result, 0, csum_size); 31719c00ddcSChris Mason } 31871a63551SDavid Sterba 31919c00ddcSChris Mason return 0; 32019c00ddcSChris Mason } 32119c00ddcSChris Mason 322d352ac68SChris Mason /* 323d352ac68SChris Mason * we can't consider a given block up to date unless the transid of the 324d352ac68SChris Mason * block matches the transid in the parent node's pointer. This is how we 325d352ac68SChris Mason * detect blocks that either didn't get written at all or got written 326d352ac68SChris Mason * in the wrong place. 327d352ac68SChris Mason */ 3281259ab75SChris Mason static int verify_parent_transid(struct extent_io_tree *io_tree, 329b9fab919SChris Mason struct extent_buffer *eb, u64 parent_transid, 330b9fab919SChris Mason int atomic) 3311259ab75SChris Mason { 3322ac55d41SJosef Bacik struct extent_state *cached_state = NULL; 3331259ab75SChris Mason int ret; 3342755a0deSDavid Sterba bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB); 3351259ab75SChris Mason 3361259ab75SChris Mason if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 3371259ab75SChris Mason return 0; 3381259ab75SChris Mason 339b9fab919SChris Mason if (atomic) 340b9fab919SChris Mason return -EAGAIN; 341b9fab919SChris Mason 342a26e8c9fSJosef Bacik if (need_lock) { 343a26e8c9fSJosef Bacik btrfs_tree_read_lock(eb); 344a26e8c9fSJosef Bacik btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); 345a26e8c9fSJosef Bacik } 346a26e8c9fSJosef Bacik 3472ac55d41SJosef Bacik lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 348ff13db41SDavid Sterba &cached_state); 3490b32f4bbSJosef Bacik if (extent_buffer_uptodate(eb) && 3501259ab75SChris Mason btrfs_header_generation(eb) == parent_transid) { 3511259ab75SChris Mason ret = 0; 3521259ab75SChris Mason goto out; 3531259ab75SChris Mason } 35494647322SDavid Sterba btrfs_err_rl(eb->fs_info, 35594647322SDavid Sterba "parent transid verify failed on %llu wanted %llu found %llu", 35694647322SDavid Sterba eb->start, 35729549aecSWang Shilong parent_transid, btrfs_header_generation(eb)); 3581259ab75SChris Mason ret = 1; 359a26e8c9fSJosef Bacik 360a26e8c9fSJosef Bacik /* 361a26e8c9fSJosef Bacik * Things reading via commit roots that don't have normal protection, 362a26e8c9fSJosef Bacik * like send, can have a really old block in cache that may point at a 36301327610SNicholas D Steeves * block that has been freed and re-allocated. So don't clear uptodate 364a26e8c9fSJosef Bacik * if we find an eb that is under IO (dirty/writeback) because we could 365a26e8c9fSJosef Bacik * end up reading in the stale data and then writing it back out and 366a26e8c9fSJosef Bacik * making everybody very sad. 367a26e8c9fSJosef Bacik */ 368a26e8c9fSJosef Bacik if (!extent_buffer_under_io(eb)) 3690b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 37033958dc6SChris Mason out: 3712ac55d41SJosef Bacik unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 372e43bbe5eSDavid Sterba &cached_state); 373472b909fSJosef Bacik if (need_lock) 374a26e8c9fSJosef Bacik btrfs_tree_read_unlock_blocking(eb); 3751259ab75SChris Mason return ret; 3761259ab75SChris Mason } 3771259ab75SChris Mason 378d352ac68SChris Mason /* 3791104a885SDavid Sterba * Return 0 if the superblock checksum type matches the checksum value of that 3801104a885SDavid Sterba * algorithm. Pass the raw disk superblock data. 3811104a885SDavid Sterba */ 382ab8d0fc4SJeff Mahoney static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info, 383ab8d0fc4SJeff Mahoney char *raw_disk_sb) 3841104a885SDavid Sterba { 3851104a885SDavid Sterba struct btrfs_super_block *disk_sb = 3861104a885SDavid Sterba (struct btrfs_super_block *)raw_disk_sb; 3871104a885SDavid Sterba u16 csum_type = btrfs_super_csum_type(disk_sb); 3881104a885SDavid Sterba int ret = 0; 3891104a885SDavid Sterba 3901104a885SDavid Sterba if (csum_type == BTRFS_CSUM_TYPE_CRC32) { 3911104a885SDavid Sterba u32 crc = ~(u32)0; 392776c4a7cSNikolay Borisov char result[sizeof(crc)]; 3931104a885SDavid Sterba 3941104a885SDavid Sterba /* 3951104a885SDavid Sterba * The super_block structure does not span the whole 3961104a885SDavid Sterba * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space 39701327610SNicholas D Steeves * is filled with zeros and is included in the checksum. 3981104a885SDavid Sterba */ 3991104a885SDavid Sterba crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE, 4001104a885SDavid Sterba crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE); 4011104a885SDavid Sterba btrfs_csum_final(crc, result); 4021104a885SDavid Sterba 403776c4a7cSNikolay Borisov if (memcmp(raw_disk_sb, result, sizeof(result))) 4041104a885SDavid Sterba ret = 1; 4051104a885SDavid Sterba } 4061104a885SDavid Sterba 4071104a885SDavid Sterba if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) { 408ab8d0fc4SJeff Mahoney btrfs_err(fs_info, "unsupported checksum algorithm %u", 4091104a885SDavid Sterba csum_type); 4101104a885SDavid Sterba ret = 1; 4111104a885SDavid Sterba } 4121104a885SDavid Sterba 4131104a885SDavid Sterba return ret; 4141104a885SDavid Sterba } 4151104a885SDavid Sterba 416581c1760SQu Wenruo static int verify_level_key(struct btrfs_fs_info *fs_info, 417581c1760SQu Wenruo struct extent_buffer *eb, int level, 418ff76a864SLiu Bo struct btrfs_key *first_key, u64 parent_transid) 419581c1760SQu Wenruo { 420581c1760SQu Wenruo int found_level; 421581c1760SQu Wenruo struct btrfs_key found_key; 422581c1760SQu Wenruo int ret; 423581c1760SQu Wenruo 424581c1760SQu Wenruo found_level = btrfs_header_level(eb); 425581c1760SQu Wenruo if (found_level != level) { 426581c1760SQu Wenruo #ifdef CONFIG_BTRFS_DEBUG 427581c1760SQu Wenruo WARN_ON(1); 428581c1760SQu Wenruo btrfs_err(fs_info, 429581c1760SQu Wenruo "tree level mismatch detected, bytenr=%llu level expected=%u has=%u", 430581c1760SQu Wenruo eb->start, level, found_level); 431581c1760SQu Wenruo #endif 432581c1760SQu Wenruo return -EIO; 433581c1760SQu Wenruo } 434581c1760SQu Wenruo 435581c1760SQu Wenruo if (!first_key) 436581c1760SQu Wenruo return 0; 437581c1760SQu Wenruo 4385d41be6fSQu Wenruo /* 4395d41be6fSQu Wenruo * For live tree block (new tree blocks in current transaction), 4405d41be6fSQu Wenruo * we need proper lock context to avoid race, which is impossible here. 4415d41be6fSQu Wenruo * So we only checks tree blocks which is read from disk, whose 4425d41be6fSQu Wenruo * generation <= fs_info->last_trans_committed. 4435d41be6fSQu Wenruo */ 4445d41be6fSQu Wenruo if (btrfs_header_generation(eb) > fs_info->last_trans_committed) 4455d41be6fSQu Wenruo return 0; 446581c1760SQu Wenruo if (found_level) 447581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &found_key, 0); 448581c1760SQu Wenruo else 449581c1760SQu Wenruo btrfs_item_key_to_cpu(eb, &found_key, 0); 450581c1760SQu Wenruo ret = btrfs_comp_cpu_keys(first_key, &found_key); 451581c1760SQu Wenruo 452581c1760SQu Wenruo #ifdef CONFIG_BTRFS_DEBUG 453581c1760SQu Wenruo if (ret) { 454581c1760SQu Wenruo WARN_ON(1); 455581c1760SQu Wenruo btrfs_err(fs_info, 456ff76a864SLiu Bo "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", 457ff76a864SLiu Bo eb->start, parent_transid, first_key->objectid, 458ff76a864SLiu Bo first_key->type, first_key->offset, 459ff76a864SLiu Bo found_key.objectid, found_key.type, 460ff76a864SLiu Bo found_key.offset); 461581c1760SQu Wenruo } 462581c1760SQu Wenruo #endif 463581c1760SQu Wenruo return ret; 464581c1760SQu Wenruo } 465581c1760SQu Wenruo 4661104a885SDavid Sterba /* 467d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 468d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 469581c1760SQu Wenruo * 470581c1760SQu Wenruo * @parent_transid: expected transid, skip check if 0 471581c1760SQu Wenruo * @level: expected level, mandatory check 472581c1760SQu Wenruo * @first_key: expected key of first slot, skip check if NULL 473d352ac68SChris Mason */ 4742ff7e61eSJeff Mahoney static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info, 475f188591eSChris Mason struct extent_buffer *eb, 476581c1760SQu Wenruo u64 parent_transid, int level, 477581c1760SQu Wenruo struct btrfs_key *first_key) 478f188591eSChris Mason { 479f188591eSChris Mason struct extent_io_tree *io_tree; 480ea466794SJosef Bacik int failed = 0; 481f188591eSChris Mason int ret; 482f188591eSChris Mason int num_copies = 0; 483f188591eSChris Mason int mirror_num = 0; 484ea466794SJosef Bacik int failed_mirror = 0; 485f188591eSChris Mason 4860b246afaSJeff Mahoney io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; 487f188591eSChris Mason while (1) { 488f8397d69SNikolay Borisov clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 4898436ea91SJosef Bacik ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE, 4906af49dbdSDavid Sterba mirror_num); 491256dd1bbSStefan Behrens if (!ret) { 492581c1760SQu Wenruo if (verify_parent_transid(io_tree, eb, 493b9fab919SChris Mason parent_transid, 0)) 494256dd1bbSStefan Behrens ret = -EIO; 495581c1760SQu Wenruo else if (verify_level_key(fs_info, eb, level, 496ff76a864SLiu Bo first_key, parent_transid)) 497581c1760SQu Wenruo ret = -EUCLEAN; 498581c1760SQu Wenruo else 499581c1760SQu Wenruo break; 500256dd1bbSStefan Behrens } 501d397712bSChris Mason 5020b246afaSJeff Mahoney num_copies = btrfs_num_copies(fs_info, 503f188591eSChris Mason eb->start, eb->len); 5044235298eSChris Mason if (num_copies == 1) 505ea466794SJosef Bacik break; 5064235298eSChris Mason 5075cf1ab56SJosef Bacik if (!failed_mirror) { 5085cf1ab56SJosef Bacik failed = 1; 5095cf1ab56SJosef Bacik failed_mirror = eb->read_mirror; 5105cf1ab56SJosef Bacik } 5115cf1ab56SJosef Bacik 512f188591eSChris Mason mirror_num++; 513ea466794SJosef Bacik if (mirror_num == failed_mirror) 514ea466794SJosef Bacik mirror_num++; 515ea466794SJosef Bacik 5164235298eSChris Mason if (mirror_num > num_copies) 517ea466794SJosef Bacik break; 518f188591eSChris Mason } 519ea466794SJosef Bacik 520c0901581SStefan Behrens if (failed && !ret && failed_mirror) 5212ff7e61eSJeff Mahoney repair_eb_io_failure(fs_info, eb, failed_mirror); 522ea466794SJosef Bacik 523ea466794SJosef Bacik return ret; 524f188591eSChris Mason } 52519c00ddcSChris Mason 526d352ac68SChris Mason /* 527d397712bSChris Mason * checksum a dirty tree block before IO. This has extra checks to make sure 528d397712bSChris Mason * we only fill in the checksum field in the first page of a multi-page block 529d352ac68SChris Mason */ 530d397712bSChris Mason 53101d58472SDaniel Dressler static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page) 53219c00ddcSChris Mason { 5334eee4fa4SMiao Xie u64 start = page_offset(page); 53419c00ddcSChris Mason u64 found_start; 53519c00ddcSChris Mason struct extent_buffer *eb; 536f188591eSChris Mason 5374f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 5384f2de97aSJosef Bacik if (page != eb->pages[0]) 5394f2de97aSJosef Bacik return 0; 5400f805531SAlex Lyakas 54119c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 5420f805531SAlex Lyakas /* 5430f805531SAlex Lyakas * Please do not consolidate these warnings into a single if. 5440f805531SAlex Lyakas * It is useful to know what went wrong. 5450f805531SAlex Lyakas */ 5460f805531SAlex Lyakas if (WARN_ON(found_start != start)) 5470f805531SAlex Lyakas return -EUCLEAN; 5480f805531SAlex Lyakas if (WARN_ON(!PageUptodate(page))) 5490f805531SAlex Lyakas return -EUCLEAN; 5500f805531SAlex Lyakas 551de37aa51SNikolay Borisov ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid, 5520f805531SAlex Lyakas btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0); 5530f805531SAlex Lyakas 5548bd98f0eSAlex Lyakas return csum_tree_block(fs_info, eb, 0); 55519c00ddcSChris Mason } 55619c00ddcSChris Mason 55701d58472SDaniel Dressler static int check_tree_block_fsid(struct btrfs_fs_info *fs_info, 5582b82032cSYan Zheng struct extent_buffer *eb) 5592b82032cSYan Zheng { 56001d58472SDaniel Dressler struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 56144880fdcSAnand Jain u8 fsid[BTRFS_FSID_SIZE]; 5622b82032cSYan Zheng int ret = 1; 5632b82032cSYan Zheng 5640a4e5586SRoss Kirk read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE); 5652b82032cSYan Zheng while (fs_devices) { 5667239ff4bSNikolay Borisov u8 *metadata_uuid; 5677239ff4bSNikolay Borisov 5687239ff4bSNikolay Borisov /* 5697239ff4bSNikolay Borisov * Checking the incompat flag is only valid for the current 5707239ff4bSNikolay Borisov * fs. For seed devices it's forbidden to have their uuid 5717239ff4bSNikolay Borisov * changed so reading ->fsid in this case is fine 5727239ff4bSNikolay Borisov */ 5737239ff4bSNikolay Borisov if (fs_devices == fs_info->fs_devices && 5747239ff4bSNikolay Borisov btrfs_fs_incompat(fs_info, METADATA_UUID)) 5757239ff4bSNikolay Borisov metadata_uuid = fs_devices->metadata_uuid; 5767239ff4bSNikolay Borisov else 5777239ff4bSNikolay Borisov metadata_uuid = fs_devices->fsid; 5787239ff4bSNikolay Borisov 5797239ff4bSNikolay Borisov if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) { 5802b82032cSYan Zheng ret = 0; 5812b82032cSYan Zheng break; 5822b82032cSYan Zheng } 5832b82032cSYan Zheng fs_devices = fs_devices->seed; 5842b82032cSYan Zheng } 5852b82032cSYan Zheng return ret; 5862b82032cSYan Zheng } 5872b82032cSYan Zheng 588facc8a22SMiao Xie static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio, 589facc8a22SMiao Xie u64 phy_offset, struct page *page, 590facc8a22SMiao Xie u64 start, u64 end, int mirror) 591ce9adaa5SChris Mason { 592ce9adaa5SChris Mason u64 found_start; 593ce9adaa5SChris Mason int found_level; 594ce9adaa5SChris Mason struct extent_buffer *eb; 595ce9adaa5SChris Mason struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 59602873e43SZhao Lei struct btrfs_fs_info *fs_info = root->fs_info; 597f188591eSChris Mason int ret = 0; 598727011e0SChris Mason int reads_done; 599ce9adaa5SChris Mason 600ce9adaa5SChris Mason if (!page->private) 601ce9adaa5SChris Mason goto out; 602d397712bSChris Mason 6034f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 604d397712bSChris Mason 6050b32f4bbSJosef Bacik /* the pending IO might have been the only thing that kept this buffer 6060b32f4bbSJosef Bacik * in memory. Make sure we have a ref for all this other checks 6070b32f4bbSJosef Bacik */ 6080b32f4bbSJosef Bacik extent_buffer_get(eb); 6090b32f4bbSJosef Bacik 6100b32f4bbSJosef Bacik reads_done = atomic_dec_and_test(&eb->io_pages); 611727011e0SChris Mason if (!reads_done) 612727011e0SChris Mason goto err; 613f188591eSChris Mason 6145cf1ab56SJosef Bacik eb->read_mirror = mirror; 615656f30dbSFilipe Manana if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 616ea466794SJosef Bacik ret = -EIO; 617ea466794SJosef Bacik goto err; 618ea466794SJosef Bacik } 619ea466794SJosef Bacik 620ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 621727011e0SChris Mason if (found_start != eb->start) { 622893bf4b1SSu Yue btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu", 623893bf4b1SSu Yue eb->start, found_start); 624f188591eSChris Mason ret = -EIO; 625ce9adaa5SChris Mason goto err; 626ce9adaa5SChris Mason } 62702873e43SZhao Lei if (check_tree_block_fsid(fs_info, eb)) { 62802873e43SZhao Lei btrfs_err_rl(fs_info, "bad fsid on block %llu", 62994647322SDavid Sterba eb->start); 6301259ab75SChris Mason ret = -EIO; 6311259ab75SChris Mason goto err; 6321259ab75SChris Mason } 633ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 6341c24c3ceSJosef Bacik if (found_level >= BTRFS_MAX_LEVEL) { 635893bf4b1SSu Yue btrfs_err(fs_info, "bad tree block level %d on %llu", 636893bf4b1SSu Yue (int)btrfs_header_level(eb), eb->start); 6371c24c3ceSJosef Bacik ret = -EIO; 6381c24c3ceSJosef Bacik goto err; 6391c24c3ceSJosef Bacik } 640ce9adaa5SChris Mason 64185d4e461SChris Mason btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), 64285d4e461SChris Mason eb, found_level); 6434008c04aSChris Mason 64402873e43SZhao Lei ret = csum_tree_block(fs_info, eb, 1); 6458bd98f0eSAlex Lyakas if (ret) 646a826d6dcSJosef Bacik goto err; 647a826d6dcSJosef Bacik 648a826d6dcSJosef Bacik /* 649a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 650a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 651a826d6dcSJosef Bacik * return -EIO. 652a826d6dcSJosef Bacik */ 6532f659546SQu Wenruo if (found_level == 0 && btrfs_check_leaf_full(fs_info, eb)) { 654a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 655a826d6dcSJosef Bacik ret = -EIO; 656a826d6dcSJosef Bacik } 657ce9adaa5SChris Mason 6582f659546SQu Wenruo if (found_level > 0 && btrfs_check_node(fs_info, eb)) 659053ab70fSLiu Bo ret = -EIO; 660053ab70fSLiu Bo 6610b32f4bbSJosef Bacik if (!ret) 6620b32f4bbSJosef Bacik set_extent_buffer_uptodate(eb); 663ce9adaa5SChris Mason err: 66479fb65a1SJosef Bacik if (reads_done && 66579fb65a1SJosef Bacik test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 666d48d71aaSDavid Sterba btree_readahead_hook(eb, ret); 6674bb31e92SArne Jansen 66853b381b3SDavid Woodhouse if (ret) { 66953b381b3SDavid Woodhouse /* 67053b381b3SDavid Woodhouse * our io error hook is going to dec the io pages 67153b381b3SDavid Woodhouse * again, we have to make sure it has something 67253b381b3SDavid Woodhouse * to decrement 67353b381b3SDavid Woodhouse */ 67453b381b3SDavid Woodhouse atomic_inc(&eb->io_pages); 6750b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 67653b381b3SDavid Woodhouse } 6770b32f4bbSJosef Bacik free_extent_buffer(eb); 678ce9adaa5SChris Mason out: 679f188591eSChris Mason return ret; 680ce9adaa5SChris Mason } 681ce9adaa5SChris Mason 6824246a0b6SChristoph Hellwig static void end_workqueue_bio(struct bio *bio) 683ce9adaa5SChris Mason { 68497eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq = bio->bi_private; 685ce9adaa5SChris Mason struct btrfs_fs_info *fs_info; 6869e0af237SLiu Bo struct btrfs_workqueue *wq; 6879e0af237SLiu Bo btrfs_work_func_t func; 688ce9adaa5SChris Mason 689ce9adaa5SChris Mason fs_info = end_io_wq->info; 6904e4cbee9SChristoph Hellwig end_io_wq->status = bio->bi_status; 691d20f7043SChris Mason 69237226b21SMike Christie if (bio_op(bio) == REQ_OP_WRITE) { 6939e0af237SLiu Bo if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) { 6949e0af237SLiu Bo wq = fs_info->endio_meta_write_workers; 6959e0af237SLiu Bo func = btrfs_endio_meta_write_helper; 6969e0af237SLiu Bo } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) { 6979e0af237SLiu Bo wq = fs_info->endio_freespace_worker; 6989e0af237SLiu Bo func = btrfs_freespace_write_helper; 6999e0af237SLiu Bo } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) { 7009e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 7019e0af237SLiu Bo func = btrfs_endio_raid56_helper; 702d20f7043SChris Mason } else { 7039e0af237SLiu Bo wq = fs_info->endio_write_workers; 7049e0af237SLiu Bo func = btrfs_endio_write_helper; 705d20f7043SChris Mason } 7069e0af237SLiu Bo } else { 7078b110e39SMiao Xie if (unlikely(end_io_wq->metadata == 7088b110e39SMiao Xie BTRFS_WQ_ENDIO_DIO_REPAIR)) { 7098b110e39SMiao Xie wq = fs_info->endio_repair_workers; 7108b110e39SMiao Xie func = btrfs_endio_repair_helper; 7118b110e39SMiao Xie } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) { 7129e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 7139e0af237SLiu Bo func = btrfs_endio_raid56_helper; 7149e0af237SLiu Bo } else if (end_io_wq->metadata) { 7159e0af237SLiu Bo wq = fs_info->endio_meta_workers; 7169e0af237SLiu Bo func = btrfs_endio_meta_helper; 7179e0af237SLiu Bo } else { 7189e0af237SLiu Bo wq = fs_info->endio_workers; 7199e0af237SLiu Bo func = btrfs_endio_helper; 7209e0af237SLiu Bo } 7219e0af237SLiu Bo } 7229e0af237SLiu Bo 7239e0af237SLiu Bo btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL); 7249e0af237SLiu Bo btrfs_queue_work(wq, &end_io_wq->work); 725ce9adaa5SChris Mason } 726ce9adaa5SChris Mason 7274e4cbee9SChristoph Hellwig blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 728bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata) 7290b86a832SChris Mason { 73097eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 7318b110e39SMiao Xie 73297eb6b69SDavid Sterba end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS); 733ce9adaa5SChris Mason if (!end_io_wq) 7344e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 735ce9adaa5SChris Mason 736ce9adaa5SChris Mason end_io_wq->private = bio->bi_private; 737ce9adaa5SChris Mason end_io_wq->end_io = bio->bi_end_io; 73822c59948SChris Mason end_io_wq->info = info; 7394e4cbee9SChristoph Hellwig end_io_wq->status = 0; 740ce9adaa5SChris Mason end_io_wq->bio = bio; 74122c59948SChris Mason end_io_wq->metadata = metadata; 742ce9adaa5SChris Mason 743ce9adaa5SChris Mason bio->bi_private = end_io_wq; 744ce9adaa5SChris Mason bio->bi_end_io = end_workqueue_bio; 74522c59948SChris Mason return 0; 74622c59948SChris Mason } 74722c59948SChris Mason 7484a69a410SChris Mason static void run_one_async_start(struct btrfs_work *work) 7494a69a410SChris Mason { 7504a69a410SChris Mason struct async_submit_bio *async; 7514e4cbee9SChristoph Hellwig blk_status_t ret; 7524a69a410SChris Mason 7534a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 754c6100a4bSJosef Bacik ret = async->submit_bio_start(async->private_data, async->bio, 755eaf25d93SChris Mason async->bio_offset); 75679787eaaSJeff Mahoney if (ret) 7574e4cbee9SChristoph Hellwig async->status = ret; 7584a69a410SChris Mason } 7594a69a410SChris Mason 76006ea01b1SDavid Sterba /* 76106ea01b1SDavid Sterba * In order to insert checksums into the metadata in large chunks, we wait 76206ea01b1SDavid Sterba * until bio submission time. All the pages in the bio are checksummed and 76306ea01b1SDavid Sterba * sums are attached onto the ordered extent record. 76406ea01b1SDavid Sterba * 76506ea01b1SDavid Sterba * At IO completion time the csums attached on the ordered extent record are 76606ea01b1SDavid Sterba * inserted into the tree. 76706ea01b1SDavid Sterba */ 7684a69a410SChris Mason static void run_one_async_done(struct btrfs_work *work) 7698b712842SChris Mason { 7708b712842SChris Mason struct async_submit_bio *async; 77106ea01b1SDavid Sterba struct inode *inode; 77206ea01b1SDavid Sterba blk_status_t ret; 7738b712842SChris Mason 7748b712842SChris Mason async = container_of(work, struct async_submit_bio, work); 77506ea01b1SDavid Sterba inode = async->private_data; 7764854ddd0SChris Mason 777bb7ab3b9SAdam Buchbinder /* If an error occurred we just want to clean up the bio and move on */ 7784e4cbee9SChristoph Hellwig if (async->status) { 7794e4cbee9SChristoph Hellwig async->bio->bi_status = async->status; 7804246a0b6SChristoph Hellwig bio_endio(async->bio); 78179787eaaSJeff Mahoney return; 78279787eaaSJeff Mahoney } 78379787eaaSJeff Mahoney 78406ea01b1SDavid Sterba ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, 78506ea01b1SDavid Sterba async->mirror_num, 1); 78606ea01b1SDavid Sterba if (ret) { 78706ea01b1SDavid Sterba async->bio->bi_status = ret; 78806ea01b1SDavid Sterba bio_endio(async->bio); 78906ea01b1SDavid Sterba } 7904a69a410SChris Mason } 7914a69a410SChris Mason 7924a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 7934a69a410SChris Mason { 7944a69a410SChris Mason struct async_submit_bio *async; 7954a69a410SChris Mason 7964a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 7978b712842SChris Mason kfree(async); 7988b712842SChris Mason } 7998b712842SChris Mason 8008c27cb35SLinus Torvalds blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio, 801c6100a4bSJosef Bacik int mirror_num, unsigned long bio_flags, 802c6100a4bSJosef Bacik u64 bio_offset, void *private_data, 803e288c080SDavid Sterba extent_submit_bio_start_t *submit_bio_start) 80444b8bd7eSChris Mason { 80544b8bd7eSChris Mason struct async_submit_bio *async; 80644b8bd7eSChris Mason 80744b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 80844b8bd7eSChris Mason if (!async) 8094e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 81044b8bd7eSChris Mason 811c6100a4bSJosef Bacik async->private_data = private_data; 81244b8bd7eSChris Mason async->bio = bio; 81344b8bd7eSChris Mason async->mirror_num = mirror_num; 8144a69a410SChris Mason async->submit_bio_start = submit_bio_start; 8154a69a410SChris Mason 8169e0af237SLiu Bo btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start, 8175cdc7ad3SQu Wenruo run_one_async_done, run_one_async_free); 8184a69a410SChris Mason 819eaf25d93SChris Mason async->bio_offset = bio_offset; 8208c8bee1dSChris Mason 8214e4cbee9SChristoph Hellwig async->status = 0; 82279787eaaSJeff Mahoney 82367f055c7SChristoph Hellwig if (op_is_sync(bio->bi_opf)) 8245cdc7ad3SQu Wenruo btrfs_set_work_high_priority(&async->work); 825d313d7a3SChris Mason 8265cdc7ad3SQu Wenruo btrfs_queue_work(fs_info->workers, &async->work); 82744b8bd7eSChris Mason return 0; 82844b8bd7eSChris Mason } 82944b8bd7eSChris Mason 8304e4cbee9SChristoph Hellwig static blk_status_t btree_csum_one_bio(struct bio *bio) 831ce3ed71aSChris Mason { 8322c30c71bSKent Overstreet struct bio_vec *bvec; 833ce3ed71aSChris Mason struct btrfs_root *root; 8342c30c71bSKent Overstreet int i, ret = 0; 835ce3ed71aSChris Mason 836c09abff8SDavid Sterba ASSERT(!bio_flagged(bio, BIO_CLONED)); 8372c30c71bSKent Overstreet bio_for_each_segment_all(bvec, bio, i) { 838ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 83901d58472SDaniel Dressler ret = csum_dirty_buffer(root->fs_info, bvec->bv_page); 84079787eaaSJeff Mahoney if (ret) 84179787eaaSJeff Mahoney break; 842ce3ed71aSChris Mason } 8432c30c71bSKent Overstreet 8444e4cbee9SChristoph Hellwig return errno_to_blk_status(ret); 845ce3ed71aSChris Mason } 846ce3ed71aSChris Mason 847d0ee3934SDavid Sterba static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio, 848eaf25d93SChris Mason u64 bio_offset) 84922c59948SChris Mason { 8508b712842SChris Mason /* 8518b712842SChris Mason * when we're called for a write, we're already in the async 8525443be45SChris Mason * submission context. Just jump into btrfs_map_bio 8538b712842SChris Mason */ 85479787eaaSJeff Mahoney return btree_csum_one_bio(bio); 85522c59948SChris Mason } 85622c59948SChris Mason 85718fdc679SLiu Bo static int check_async_write(struct btrfs_inode *bi) 858de0022b9SJosef Bacik { 8596300463bSLiu Bo if (atomic_read(&bi->sync_writers)) 8606300463bSLiu Bo return 0; 861de0022b9SJosef Bacik #ifdef CONFIG_X86 862bc696ca0SBorislav Petkov if (static_cpu_has(X86_FEATURE_XMM4_2)) 863de0022b9SJosef Bacik return 0; 864de0022b9SJosef Bacik #endif 865de0022b9SJosef Bacik return 1; 866de0022b9SJosef Bacik } 867de0022b9SJosef Bacik 8688c27cb35SLinus Torvalds static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio, 869eaf25d93SChris Mason int mirror_num, unsigned long bio_flags, 870eaf25d93SChris Mason u64 bio_offset) 87144b8bd7eSChris Mason { 872c6100a4bSJosef Bacik struct inode *inode = private_data; 8730b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 87418fdc679SLiu Bo int async = check_async_write(BTRFS_I(inode)); 8754e4cbee9SChristoph Hellwig blk_status_t ret; 876cad321adSChris Mason 87737226b21SMike Christie if (bio_op(bio) != REQ_OP_WRITE) { 878cad321adSChris Mason /* 879cad321adSChris Mason * called for a read, do the setup so that checksum validation 880cad321adSChris Mason * can happen in the async kernel threads 881cad321adSChris Mason */ 8820b246afaSJeff Mahoney ret = btrfs_bio_wq_end_io(fs_info, bio, 8830b246afaSJeff Mahoney BTRFS_WQ_ENDIO_METADATA); 8841d4284bdSChris Mason if (ret) 88561891923SStefan Behrens goto out_w_error; 8862ff7e61eSJeff Mahoney ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); 887de0022b9SJosef Bacik } else if (!async) { 888de0022b9SJosef Bacik ret = btree_csum_one_bio(bio); 889de0022b9SJosef Bacik if (ret) 89061891923SStefan Behrens goto out_w_error; 8912ff7e61eSJeff Mahoney ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); 89261891923SStefan Behrens } else { 893cad321adSChris Mason /* 89461891923SStefan Behrens * kthread helpers are used to submit writes so that 89561891923SStefan Behrens * checksumming can happen in parallel across all CPUs 896cad321adSChris Mason */ 897c6100a4bSJosef Bacik ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0, 898c6100a4bSJosef Bacik bio_offset, private_data, 899e288c080SDavid Sterba btree_submit_bio_start); 90044b8bd7eSChris Mason } 90144b8bd7eSChris Mason 9024246a0b6SChristoph Hellwig if (ret) 9034246a0b6SChristoph Hellwig goto out_w_error; 9044246a0b6SChristoph Hellwig return 0; 9054246a0b6SChristoph Hellwig 90661891923SStefan Behrens out_w_error: 9074e4cbee9SChristoph Hellwig bio->bi_status = ret; 9084246a0b6SChristoph Hellwig bio_endio(bio); 90961891923SStefan Behrens return ret; 91061891923SStefan Behrens } 91161891923SStefan Behrens 9123dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 913784b4e29SChris Mason static int btree_migratepage(struct address_space *mapping, 914a6bc32b8SMel Gorman struct page *newpage, struct page *page, 915a6bc32b8SMel Gorman enum migrate_mode mode) 916784b4e29SChris Mason { 917784b4e29SChris Mason /* 918784b4e29SChris Mason * we can't safely write a btree page from here, 919784b4e29SChris Mason * we haven't done the locking hook 920784b4e29SChris Mason */ 921784b4e29SChris Mason if (PageDirty(page)) 922784b4e29SChris Mason return -EAGAIN; 923784b4e29SChris Mason /* 924784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 925784b4e29SChris Mason * We must have no buffers or drop them. 926784b4e29SChris Mason */ 927784b4e29SChris Mason if (page_has_private(page) && 928784b4e29SChris Mason !try_to_release_page(page, GFP_KERNEL)) 929784b4e29SChris Mason return -EAGAIN; 930a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 931784b4e29SChris Mason } 9323dd1462eSJan Beulich #endif 933784b4e29SChris Mason 9340da5468fSChris Mason 9350da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 9360da5468fSChris Mason struct writeback_control *wbc) 9370da5468fSChris Mason { 938e2d84521SMiao Xie struct btrfs_fs_info *fs_info; 939e2d84521SMiao Xie int ret; 940e2d84521SMiao Xie 941d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 942448d640bSChris Mason 943448d640bSChris Mason if (wbc->for_kupdate) 944448d640bSChris Mason return 0; 945448d640bSChris Mason 946e2d84521SMiao Xie fs_info = BTRFS_I(mapping->host)->root->fs_info; 947b9473439SChris Mason /* this is a bit racy, but that's ok */ 948d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 949d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 950d814a491SEthan Lien fs_info->dirty_metadata_batch); 951e2d84521SMiao Xie if (ret < 0) 952793955bcSChris Mason return 0; 953793955bcSChris Mason } 9540b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 9550da5468fSChris Mason } 9560da5468fSChris Mason 957b2950863SChristoph Hellwig static int btree_readpage(struct file *file, struct page *page) 9585f39d397SChris Mason { 959d1310b2eSChris Mason struct extent_io_tree *tree; 960d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9618ddc7d9cSJan Schmidt return extent_read_full_page(tree, page, btree_get_extent, 0); 9625f39d397SChris Mason } 9635f39d397SChris Mason 96470dec807SChris Mason static int btree_releasepage(struct page *page, gfp_t gfp_flags) 9655f39d397SChris Mason { 96698509cfcSChris Mason if (PageWriteback(page) || PageDirty(page)) 96798509cfcSChris Mason return 0; 9680c4e538bSDavid Sterba 969f7a52a40SDavid Sterba return try_release_extent_buffer(page); 970d98237b3SChris Mason } 971d98237b3SChris Mason 972d47992f8SLukas Czerner static void btree_invalidatepage(struct page *page, unsigned int offset, 973d47992f8SLukas Czerner unsigned int length) 974d98237b3SChris Mason { 975d1310b2eSChris Mason struct extent_io_tree *tree; 976d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9775f39d397SChris Mason extent_invalidatepage(tree, page, offset); 9785f39d397SChris Mason btree_releasepage(page, GFP_NOFS); 9799ad6b7bcSChris Mason if (PagePrivate(page)) { 980efe120a0SFrank Holton btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info, 981efe120a0SFrank Holton "page private not zero on page %llu", 982efe120a0SFrank Holton (unsigned long long)page_offset(page)); 9839ad6b7bcSChris Mason ClearPagePrivate(page); 9849ad6b7bcSChris Mason set_page_private(page, 0); 98509cbfeafSKirill A. Shutemov put_page(page); 9869ad6b7bcSChris Mason } 987d98237b3SChris Mason } 988d98237b3SChris Mason 9890b32f4bbSJosef Bacik static int btree_set_page_dirty(struct page *page) 9900b32f4bbSJosef Bacik { 991bb146eb2SJosef Bacik #ifdef DEBUG 9920b32f4bbSJosef Bacik struct extent_buffer *eb; 9930b32f4bbSJosef Bacik 9940b32f4bbSJosef Bacik BUG_ON(!PagePrivate(page)); 9950b32f4bbSJosef Bacik eb = (struct extent_buffer *)page->private; 9960b32f4bbSJosef Bacik BUG_ON(!eb); 9970b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 9980b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 9990b32f4bbSJosef Bacik btrfs_assert_tree_locked(eb); 1000bb146eb2SJosef Bacik #endif 10010b32f4bbSJosef Bacik return __set_page_dirty_nobuffers(page); 10020b32f4bbSJosef Bacik } 10030b32f4bbSJosef Bacik 10047f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 1005d98237b3SChris Mason .readpage = btree_readpage, 10060da5468fSChris Mason .writepages = btree_writepages, 10075f39d397SChris Mason .releasepage = btree_releasepage, 10085f39d397SChris Mason .invalidatepage = btree_invalidatepage, 10095a92bc88SChris Mason #ifdef CONFIG_MIGRATION 1010784b4e29SChris Mason .migratepage = btree_migratepage, 10115a92bc88SChris Mason #endif 10120b32f4bbSJosef Bacik .set_page_dirty = btree_set_page_dirty, 1013d98237b3SChris Mason }; 1014123abc88SChris Mason 10152ff7e61eSJeff Mahoney void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr) 1016090d1875SChris Mason { 10175f39d397SChris Mason struct extent_buffer *buf = NULL; 10182ff7e61eSJeff Mahoney struct inode *btree_inode = fs_info->btree_inode; 1019090d1875SChris Mason 10202ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 1021c871b0f2SLiu Bo if (IS_ERR(buf)) 10226197d86eSDavid Sterba return; 1023d1310b2eSChris Mason read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, 10246af49dbdSDavid Sterba buf, WAIT_NONE, 0); 10255f39d397SChris Mason free_extent_buffer(buf); 1026090d1875SChris Mason } 1027090d1875SChris Mason 10282ff7e61eSJeff Mahoney int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr, 1029ab0fff03SArne Jansen int mirror_num, struct extent_buffer **eb) 1030ab0fff03SArne Jansen { 1031ab0fff03SArne Jansen struct extent_buffer *buf = NULL; 10322ff7e61eSJeff Mahoney struct inode *btree_inode = fs_info->btree_inode; 1033ab0fff03SArne Jansen struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree; 1034ab0fff03SArne Jansen int ret; 1035ab0fff03SArne Jansen 10362ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 1037c871b0f2SLiu Bo if (IS_ERR(buf)) 1038ab0fff03SArne Jansen return 0; 1039ab0fff03SArne Jansen 1040ab0fff03SArne Jansen set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags); 1041ab0fff03SArne Jansen 10428436ea91SJosef Bacik ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK, 10436af49dbdSDavid Sterba mirror_num); 1044ab0fff03SArne Jansen if (ret) { 1045ab0fff03SArne Jansen free_extent_buffer(buf); 1046ab0fff03SArne Jansen return ret; 1047ab0fff03SArne Jansen } 1048ab0fff03SArne Jansen 1049ab0fff03SArne Jansen if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) { 1050ab0fff03SArne Jansen free_extent_buffer(buf); 1051ab0fff03SArne Jansen return -EIO; 10520b32f4bbSJosef Bacik } else if (extent_buffer_uptodate(buf)) { 1053ab0fff03SArne Jansen *eb = buf; 1054ab0fff03SArne Jansen } else { 1055ab0fff03SArne Jansen free_extent_buffer(buf); 1056ab0fff03SArne Jansen } 1057ab0fff03SArne Jansen return 0; 1058ab0fff03SArne Jansen } 1059ab0fff03SArne Jansen 10602ff7e61eSJeff Mahoney struct extent_buffer *btrfs_find_create_tree_block( 10612ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info, 10620308af44SDavid Sterba u64 bytenr) 10630999df54SChris Mason { 10640b246afaSJeff Mahoney if (btrfs_is_testing(fs_info)) 10650b246afaSJeff Mahoney return alloc_test_extent_buffer(fs_info, bytenr); 10660b246afaSJeff Mahoney return alloc_extent_buffer(fs_info, bytenr); 10670999df54SChris Mason } 10680999df54SChris Mason 10690999df54SChris Mason 1070e02119d5SChris Mason int btrfs_write_tree_block(struct extent_buffer *buf) 1071e02119d5SChris Mason { 1072727011e0SChris Mason return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start, 10738aa38c31SChristoph Hellwig buf->start + buf->len - 1); 1074e02119d5SChris Mason } 1075e02119d5SChris Mason 10763189ff77SJeff Layton void btrfs_wait_tree_block_writeback(struct extent_buffer *buf) 1077e02119d5SChris Mason { 10783189ff77SJeff Layton filemap_fdatawait_range(buf->pages[0]->mapping, 10798aa38c31SChristoph Hellwig buf->start, buf->start + buf->len - 1); 1080e02119d5SChris Mason } 1081e02119d5SChris Mason 1082581c1760SQu Wenruo /* 1083581c1760SQu Wenruo * Read tree block at logical address @bytenr and do variant basic but critical 1084581c1760SQu Wenruo * verification. 1085581c1760SQu Wenruo * 1086581c1760SQu Wenruo * @parent_transid: expected transid of this tree block, skip check if 0 1087581c1760SQu Wenruo * @level: expected level, mandatory check 1088581c1760SQu Wenruo * @first_key: expected key in slot 0, skip check if NULL 1089581c1760SQu Wenruo */ 10902ff7e61eSJeff Mahoney struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, 1091581c1760SQu Wenruo u64 parent_transid, int level, 1092581c1760SQu Wenruo struct btrfs_key *first_key) 1093e20d96d6SChris Mason { 10945f39d397SChris Mason struct extent_buffer *buf = NULL; 109519c00ddcSChris Mason int ret; 109619c00ddcSChris Mason 10972ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 1098c871b0f2SLiu Bo if (IS_ERR(buf)) 1099c871b0f2SLiu Bo return buf; 1100e4204dedSChris Mason 1101581c1760SQu Wenruo ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid, 1102581c1760SQu Wenruo level, first_key); 11030f0fe8f7SFilipe David Borba Manana if (ret) { 11040f0fe8f7SFilipe David Borba Manana free_extent_buffer(buf); 110564c043deSLiu Bo return ERR_PTR(ret); 11060f0fe8f7SFilipe David Borba Manana } 11075f39d397SChris Mason return buf; 1108ce9adaa5SChris Mason 1109eb60ceacSChris Mason } 1110eb60ceacSChris Mason 11117c302b49SDavid Sterba void clean_tree_block(struct btrfs_fs_info *fs_info, 11125f39d397SChris Mason struct extent_buffer *buf) 1113ed2ff2cbSChris Mason { 111455c69072SChris Mason if (btrfs_header_generation(buf) == 1115e2d84521SMiao Xie fs_info->running_transaction->transid) { 1116b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1117b4ce94deSChris Mason 1118b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1119104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 1120e2d84521SMiao Xie -buf->len, 1121e2d84521SMiao Xie fs_info->dirty_metadata_batch); 1122b9473439SChris Mason /* ugh, clear_extent_buffer_dirty needs to lock the page */ 1123b9473439SChris Mason btrfs_set_lock_blocking(buf); 11240b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 1125925baeddSChris Mason } 11265f39d397SChris Mason } 1127ed7b63ebSJosef Bacik } 11285f39d397SChris Mason 11298257b2dcSMiao Xie static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void) 11308257b2dcSMiao Xie { 11318257b2dcSMiao Xie struct btrfs_subvolume_writers *writers; 11328257b2dcSMiao Xie int ret; 11338257b2dcSMiao Xie 11348257b2dcSMiao Xie writers = kmalloc(sizeof(*writers), GFP_NOFS); 11358257b2dcSMiao Xie if (!writers) 11368257b2dcSMiao Xie return ERR_PTR(-ENOMEM); 11378257b2dcSMiao Xie 11388a5a916dSJeff Mahoney ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS); 11398257b2dcSMiao Xie if (ret < 0) { 11408257b2dcSMiao Xie kfree(writers); 11418257b2dcSMiao Xie return ERR_PTR(ret); 11428257b2dcSMiao Xie } 11438257b2dcSMiao Xie 11448257b2dcSMiao Xie init_waitqueue_head(&writers->wait); 11458257b2dcSMiao Xie return writers; 11468257b2dcSMiao Xie } 11478257b2dcSMiao Xie 11488257b2dcSMiao Xie static void 11498257b2dcSMiao Xie btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers) 11508257b2dcSMiao Xie { 11518257b2dcSMiao Xie percpu_counter_destroy(&writers->counter); 11528257b2dcSMiao Xie kfree(writers); 11538257b2dcSMiao Xie } 11548257b2dcSMiao Xie 1155da17066cSJeff Mahoney static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, 1156e20d96d6SChris Mason u64 objectid) 1157d97e63b6SChris Mason { 11587c0260eeSJeff Mahoney bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); 1159cfaa7295SChris Mason root->node = NULL; 1160a28ec197SChris Mason root->commit_root = NULL; 116127cdeb70SMiao Xie root->state = 0; 1162d68fc57bSYan, Zheng root->orphan_cleanup_state = 0; 11630b86a832SChris Mason 11640f7d52f4SChris Mason root->last_trans = 0; 116513a8a7c8SYan, Zheng root->highest_objectid = 0; 1166eb73c1b7SMiao Xie root->nr_delalloc_inodes = 0; 1167199c2a9cSMiao Xie root->nr_ordered_extents = 0; 11686bef4d31SEric Paris root->inode_tree = RB_ROOT; 116916cdcec7SMiao Xie INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1170f0486c68SYan, Zheng root->block_rsv = NULL; 11710b86a832SChris Mason 11720b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 11735d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 1174eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_inodes); 1175eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_root); 1176199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_extents); 1177199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_root); 11782ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 11792ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 11805d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1181eb73c1b7SMiao Xie spin_lock_init(&root->delalloc_lock); 1182199c2a9cSMiao Xie spin_lock_init(&root->ordered_extent_lock); 1183f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 11842ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 11852ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 11868287475aSQu Wenruo spin_lock_init(&root->qgroup_meta_rsv_lock); 1187a2135011SChris Mason mutex_init(&root->objectid_mutex); 1188e02119d5SChris Mason mutex_init(&root->log_mutex); 118931f3d255SMiao Xie mutex_init(&root->ordered_extent_mutex); 1190573bfb72SMiao Xie mutex_init(&root->delalloc_mutex); 11917237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 11927237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 11937237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 11948b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[0]); 11958b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[1]); 11967237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 11977237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 11987237f183SYan Zheng atomic_set(&root->log_writers, 0); 11992ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 12000700cea7SElena Reshetova refcount_set(&root->refs, 1); 1201ea14b57fSDavid Sterba atomic_set(&root->will_be_snapshotted, 0); 12028ecebf4dSRobbie Ko atomic_set(&root->snapshot_force_cow, 0); 1203eede2bf3SOmar Sandoval atomic_set(&root->nr_swapfiles, 0); 12047237f183SYan Zheng root->log_transid = 0; 1205d1433debSMiao Xie root->log_transid_committed = -1; 1206257c62e1SChris Mason root->last_log_commit = 0; 12077c0260eeSJeff Mahoney if (!dummy) 1208c6100a4bSJosef Bacik extent_io_tree_init(&root->dirty_log_pages, NULL); 1209017e5369SChris Mason 12103768f368SChris Mason memset(&root->root_key, 0, sizeof(root->root_key)); 12113768f368SChris Mason memset(&root->root_item, 0, sizeof(root->root_item)); 12126702ed49SChris Mason memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 12137c0260eeSJeff Mahoney if (!dummy) 12143f157a2fSChris Mason root->defrag_trans_start = fs_info->generation; 121506ea65a3SJosef Bacik else 121606ea65a3SJosef Bacik root->defrag_trans_start = 0; 12174d775673SChris Mason root->root_key.objectid = objectid; 12180ee5dc67SAl Viro root->anon_dev = 0; 12198ea05e3aSAlexander Block 12205f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 12213768f368SChris Mason } 12223768f368SChris Mason 122374e4d827SDavid Sterba static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, 122474e4d827SDavid Sterba gfp_t flags) 12256f07e42eSAl Viro { 122674e4d827SDavid Sterba struct btrfs_root *root = kzalloc(sizeof(*root), flags); 12276f07e42eSAl Viro if (root) 12286f07e42eSAl Viro root->fs_info = fs_info; 12296f07e42eSAl Viro return root; 12306f07e42eSAl Viro } 12316f07e42eSAl Viro 123206ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 123306ea65a3SJosef Bacik /* Should only be used by the testing infrastructure */ 1234da17066cSJeff Mahoney struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) 123506ea65a3SJosef Bacik { 123606ea65a3SJosef Bacik struct btrfs_root *root; 123706ea65a3SJosef Bacik 12387c0260eeSJeff Mahoney if (!fs_info) 12397c0260eeSJeff Mahoney return ERR_PTR(-EINVAL); 12407c0260eeSJeff Mahoney 12417c0260eeSJeff Mahoney root = btrfs_alloc_root(fs_info, GFP_KERNEL); 124206ea65a3SJosef Bacik if (!root) 124306ea65a3SJosef Bacik return ERR_PTR(-ENOMEM); 1244da17066cSJeff Mahoney 1245b9ef22deSFeifei Xu /* We don't use the stripesize in selftest, set it as sectorsize */ 1246da17066cSJeff Mahoney __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID); 1247faa2dbf0SJosef Bacik root->alloc_bytenr = 0; 124806ea65a3SJosef Bacik 124906ea65a3SJosef Bacik return root; 125006ea65a3SJosef Bacik } 125106ea65a3SJosef Bacik #endif 125206ea65a3SJosef Bacik 125320897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 125420897f5cSArne Jansen struct btrfs_fs_info *fs_info, 125520897f5cSArne Jansen u64 objectid) 125620897f5cSArne Jansen { 125720897f5cSArne Jansen struct extent_buffer *leaf; 125820897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 125920897f5cSArne Jansen struct btrfs_root *root; 126020897f5cSArne Jansen struct btrfs_key key; 126120897f5cSArne Jansen int ret = 0; 126233d85fdaSQu Wenruo uuid_le uuid = NULL_UUID_LE; 126320897f5cSArne Jansen 126474e4d827SDavid Sterba root = btrfs_alloc_root(fs_info, GFP_KERNEL); 126520897f5cSArne Jansen if (!root) 126620897f5cSArne Jansen return ERR_PTR(-ENOMEM); 126720897f5cSArne Jansen 1268da17066cSJeff Mahoney __setup_root(root, fs_info, objectid); 126920897f5cSArne Jansen root->root_key.objectid = objectid; 127020897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 127120897f5cSArne Jansen root->root_key.offset = 0; 127220897f5cSArne Jansen 12734d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0); 127420897f5cSArne Jansen if (IS_ERR(leaf)) { 127520897f5cSArne Jansen ret = PTR_ERR(leaf); 12761dd05682STsutomu Itoh leaf = NULL; 127720897f5cSArne Jansen goto fail; 127820897f5cSArne Jansen } 127920897f5cSArne Jansen 128020897f5cSArne Jansen root->node = leaf; 128120897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 128220897f5cSArne Jansen 128320897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 128427cdeb70SMiao Xie set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 128520897f5cSArne Jansen 128620897f5cSArne Jansen root->root_item.flags = 0; 128720897f5cSArne Jansen root->root_item.byte_limit = 0; 128820897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 128920897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 129020897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 129120897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 129220897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 129320897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 129420897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 129533d85fdaSQu Wenruo if (is_fstree(objectid)) 12966463fe58SStefan Behrens uuid_le_gen(&uuid); 12976463fe58SStefan Behrens memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE); 129820897f5cSArne Jansen root->root_item.drop_level = 0; 129920897f5cSArne Jansen 130020897f5cSArne Jansen key.objectid = objectid; 130120897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 130220897f5cSArne Jansen key.offset = 0; 130320897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 130420897f5cSArne Jansen if (ret) 130520897f5cSArne Jansen goto fail; 130620897f5cSArne Jansen 130720897f5cSArne Jansen btrfs_tree_unlock(leaf); 130820897f5cSArne Jansen 130920897f5cSArne Jansen return root; 13101dd05682STsutomu Itoh 13111dd05682STsutomu Itoh fail: 13121dd05682STsutomu Itoh if (leaf) { 13131dd05682STsutomu Itoh btrfs_tree_unlock(leaf); 131459885b39STsutomu Itoh free_extent_buffer(root->commit_root); 13151dd05682STsutomu Itoh free_extent_buffer(leaf); 13161dd05682STsutomu Itoh } 13171dd05682STsutomu Itoh kfree(root); 13181dd05682STsutomu Itoh 13191dd05682STsutomu Itoh return ERR_PTR(ret); 132020897f5cSArne Jansen } 132120897f5cSArne Jansen 13227237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1323e02119d5SChris Mason struct btrfs_fs_info *fs_info) 13240f7d52f4SChris Mason { 13250f7d52f4SChris Mason struct btrfs_root *root; 13267237f183SYan Zheng struct extent_buffer *leaf; 1327e02119d5SChris Mason 132874e4d827SDavid Sterba root = btrfs_alloc_root(fs_info, GFP_NOFS); 1329e02119d5SChris Mason if (!root) 13307237f183SYan Zheng return ERR_PTR(-ENOMEM); 1331e02119d5SChris Mason 1332da17066cSJeff Mahoney __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID); 1333e02119d5SChris Mason 1334e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1335e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1336e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 133727cdeb70SMiao Xie 13387237f183SYan Zheng /* 133927cdeb70SMiao Xie * DON'T set REF_COWS for log trees 134027cdeb70SMiao Xie * 13417237f183SYan Zheng * log trees do not get reference counted because they go away 13427237f183SYan Zheng * before a real commit is actually done. They do store pointers 13437237f183SYan Zheng * to file data extents, and those reference counts still get 13447237f183SYan Zheng * updated (along with back refs to the log tree). 13457237f183SYan Zheng */ 1346e02119d5SChris Mason 13474d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 13484d75f8a9SDavid Sterba NULL, 0, 0, 0); 13497237f183SYan Zheng if (IS_ERR(leaf)) { 13507237f183SYan Zheng kfree(root); 13517237f183SYan Zheng return ERR_CAST(leaf); 13527237f183SYan Zheng } 1353e02119d5SChris Mason 13547237f183SYan Zheng root->node = leaf; 1355e02119d5SChris Mason 1356e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1357e02119d5SChris Mason btrfs_tree_unlock(root->node); 13587237f183SYan Zheng return root; 13597237f183SYan Zheng } 13607237f183SYan Zheng 13617237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 13627237f183SYan Zheng struct btrfs_fs_info *fs_info) 13637237f183SYan Zheng { 13647237f183SYan Zheng struct btrfs_root *log_root; 13657237f183SYan Zheng 13667237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 13677237f183SYan Zheng if (IS_ERR(log_root)) 13687237f183SYan Zheng return PTR_ERR(log_root); 13697237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 13707237f183SYan Zheng fs_info->log_root_tree = log_root; 13717237f183SYan Zheng return 0; 13727237f183SYan Zheng } 13737237f183SYan Zheng 13747237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 13757237f183SYan Zheng struct btrfs_root *root) 13767237f183SYan Zheng { 13770b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13787237f183SYan Zheng struct btrfs_root *log_root; 13797237f183SYan Zheng struct btrfs_inode_item *inode_item; 13807237f183SYan Zheng 13810b246afaSJeff Mahoney log_root = alloc_log_tree(trans, fs_info); 13827237f183SYan Zheng if (IS_ERR(log_root)) 13837237f183SYan Zheng return PTR_ERR(log_root); 13847237f183SYan Zheng 13857237f183SYan Zheng log_root->last_trans = trans->transid; 13867237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 13877237f183SYan Zheng 13887237f183SYan Zheng inode_item = &log_root->root_item.inode; 13893cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 13903cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 13913cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 1392da17066cSJeff Mahoney btrfs_set_stack_inode_nbytes(inode_item, 13930b246afaSJeff Mahoney fs_info->nodesize); 13943cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 13957237f183SYan Zheng 13965d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 13977237f183SYan Zheng 13987237f183SYan Zheng WARN_ON(root->log_root); 13997237f183SYan Zheng root->log_root = log_root; 14007237f183SYan Zheng root->log_transid = 0; 1401d1433debSMiao Xie root->log_transid_committed = -1; 1402257c62e1SChris Mason root->last_log_commit = 0; 1403e02119d5SChris Mason return 0; 1404e02119d5SChris Mason } 1405e02119d5SChris Mason 140635a3621bSStefan Behrens static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 1407cb517eabSMiao Xie struct btrfs_key *key) 1408e02119d5SChris Mason { 1409e02119d5SChris Mason struct btrfs_root *root; 1410e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 14110f7d52f4SChris Mason struct btrfs_path *path; 141284234f3aSYan Zheng u64 generation; 1413cb517eabSMiao Xie int ret; 1414581c1760SQu Wenruo int level; 1415cb517eabSMiao Xie 1416cb517eabSMiao Xie path = btrfs_alloc_path(); 1417cb517eabSMiao Xie if (!path) 1418cb517eabSMiao Xie return ERR_PTR(-ENOMEM); 14190f7d52f4SChris Mason 142074e4d827SDavid Sterba root = btrfs_alloc_root(fs_info, GFP_NOFS); 1421cb517eabSMiao Xie if (!root) { 1422cb517eabSMiao Xie ret = -ENOMEM; 1423cb517eabSMiao Xie goto alloc_fail; 14240f7d52f4SChris Mason } 14250f7d52f4SChris Mason 1426da17066cSJeff Mahoney __setup_root(root, fs_info, key->objectid); 14270f7d52f4SChris Mason 1428cb517eabSMiao Xie ret = btrfs_find_root(tree_root, key, path, 1429cb517eabSMiao Xie &root->root_item, &root->root_key); 14300f7d52f4SChris Mason if (ret) { 143113a8a7c8SYan, Zheng if (ret > 0) 143213a8a7c8SYan, Zheng ret = -ENOENT; 1433cb517eabSMiao Xie goto find_fail; 14340f7d52f4SChris Mason } 143513a8a7c8SYan, Zheng 143684234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1437581c1760SQu Wenruo level = btrfs_root_level(&root->root_item); 14382ff7e61eSJeff Mahoney root->node = read_tree_block(fs_info, 14392ff7e61eSJeff Mahoney btrfs_root_bytenr(&root->root_item), 1440581c1760SQu Wenruo generation, level, NULL); 144164c043deSLiu Bo if (IS_ERR(root->node)) { 144264c043deSLiu Bo ret = PTR_ERR(root->node); 1443cb517eabSMiao Xie goto find_fail; 1444cb517eabSMiao Xie } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1445cb517eabSMiao Xie ret = -EIO; 144664c043deSLiu Bo free_extent_buffer(root->node); 144764c043deSLiu Bo goto find_fail; 1448416bc658SJosef Bacik } 14495d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 145013a8a7c8SYan, Zheng out: 1451cb517eabSMiao Xie btrfs_free_path(path); 1452cb517eabSMiao Xie return root; 1453cb517eabSMiao Xie 1454cb517eabSMiao Xie find_fail: 1455cb517eabSMiao Xie kfree(root); 1456cb517eabSMiao Xie alloc_fail: 1457cb517eabSMiao Xie root = ERR_PTR(ret); 1458cb517eabSMiao Xie goto out; 1459cb517eabSMiao Xie } 1460cb517eabSMiao Xie 1461cb517eabSMiao Xie struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root, 1462cb517eabSMiao Xie struct btrfs_key *location) 1463cb517eabSMiao Xie { 1464cb517eabSMiao Xie struct btrfs_root *root; 1465cb517eabSMiao Xie 1466cb517eabSMiao Xie root = btrfs_read_tree_root(tree_root, location); 1467cb517eabSMiao Xie if (IS_ERR(root)) 1468cb517eabSMiao Xie return root; 1469cb517eabSMiao Xie 1470cb517eabSMiao Xie if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { 147127cdeb70SMiao Xie set_bit(BTRFS_ROOT_REF_COWS, &root->state); 147208fe4db1SLi Zefan btrfs_check_and_init_root_item(&root->root_item); 147308fe4db1SLi Zefan } 147413a8a7c8SYan, Zheng 14755eda7b5eSChris Mason return root; 14765eda7b5eSChris Mason } 14775eda7b5eSChris Mason 1478cb517eabSMiao Xie int btrfs_init_fs_root(struct btrfs_root *root) 1479cb517eabSMiao Xie { 1480cb517eabSMiao Xie int ret; 14818257b2dcSMiao Xie struct btrfs_subvolume_writers *writers; 1482cb517eabSMiao Xie 1483cb517eabSMiao Xie root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); 1484cb517eabSMiao Xie root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), 1485cb517eabSMiao Xie GFP_NOFS); 1486cb517eabSMiao Xie if (!root->free_ino_pinned || !root->free_ino_ctl) { 1487cb517eabSMiao Xie ret = -ENOMEM; 1488cb517eabSMiao Xie goto fail; 1489cb517eabSMiao Xie } 1490cb517eabSMiao Xie 14918257b2dcSMiao Xie writers = btrfs_alloc_subvolume_writers(); 14928257b2dcSMiao Xie if (IS_ERR(writers)) { 14938257b2dcSMiao Xie ret = PTR_ERR(writers); 14948257b2dcSMiao Xie goto fail; 14958257b2dcSMiao Xie } 14968257b2dcSMiao Xie root->subv_writers = writers; 14978257b2dcSMiao Xie 1498cb517eabSMiao Xie btrfs_init_free_ino_ctl(root); 149957cdc8dbSDavid Sterba spin_lock_init(&root->ino_cache_lock); 150057cdc8dbSDavid Sterba init_waitqueue_head(&root->ino_cache_wait); 1501cb517eabSMiao Xie 1502cb517eabSMiao Xie ret = get_anon_bdev(&root->anon_dev); 1503cb517eabSMiao Xie if (ret) 1504876d2cf1SLiu Bo goto fail; 1505f32e48e9SChandan Rajendra 1506f32e48e9SChandan Rajendra mutex_lock(&root->objectid_mutex); 1507f32e48e9SChandan Rajendra ret = btrfs_find_highest_objectid(root, 1508f32e48e9SChandan Rajendra &root->highest_objectid); 1509f32e48e9SChandan Rajendra if (ret) { 1510f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1511876d2cf1SLiu Bo goto fail; 1512f32e48e9SChandan Rajendra } 1513f32e48e9SChandan Rajendra 1514f32e48e9SChandan Rajendra ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 1515f32e48e9SChandan Rajendra 1516f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1517f32e48e9SChandan Rajendra 1518cb517eabSMiao Xie return 0; 1519cb517eabSMiao Xie fail: 152084db5ccfSDavid Sterba /* The caller is responsible to call btrfs_free_fs_root */ 1521cb517eabSMiao Xie return ret; 1522cb517eabSMiao Xie } 1523cb517eabSMiao Xie 152435bbb97fSJeff Mahoney struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1525cb517eabSMiao Xie u64 root_id) 1526cb517eabSMiao Xie { 1527cb517eabSMiao Xie struct btrfs_root *root; 1528cb517eabSMiao Xie 1529cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1530cb517eabSMiao Xie root = radix_tree_lookup(&fs_info->fs_roots_radix, 1531cb517eabSMiao Xie (unsigned long)root_id); 1532cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1533cb517eabSMiao Xie return root; 1534cb517eabSMiao Xie } 1535cb517eabSMiao Xie 1536cb517eabSMiao Xie int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1537cb517eabSMiao Xie struct btrfs_root *root) 1538cb517eabSMiao Xie { 1539cb517eabSMiao Xie int ret; 1540cb517eabSMiao Xie 1541e1860a77SDavid Sterba ret = radix_tree_preload(GFP_NOFS); 1542cb517eabSMiao Xie if (ret) 1543cb517eabSMiao Xie return ret; 1544cb517eabSMiao Xie 1545cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1546cb517eabSMiao Xie ret = radix_tree_insert(&fs_info->fs_roots_radix, 1547cb517eabSMiao Xie (unsigned long)root->root_key.objectid, 1548cb517eabSMiao Xie root); 1549cb517eabSMiao Xie if (ret == 0) 155027cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1551cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1552cb517eabSMiao Xie radix_tree_preload_end(); 1553cb517eabSMiao Xie 1554cb517eabSMiao Xie return ret; 1555cb517eabSMiao Xie } 1556cb517eabSMiao Xie 1557c00869f1SMiao Xie struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 1558c00869f1SMiao Xie struct btrfs_key *location, 1559c00869f1SMiao Xie bool check_ref) 15605eda7b5eSChris Mason { 15615eda7b5eSChris Mason struct btrfs_root *root; 1562381cf658SDavid Sterba struct btrfs_path *path; 15631d4c08e0SDavid Sterba struct btrfs_key key; 15645eda7b5eSChris Mason int ret; 15655eda7b5eSChris Mason 1566edbd8d4eSChris Mason if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 1567edbd8d4eSChris Mason return fs_info->tree_root; 1568edbd8d4eSChris Mason if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID) 1569edbd8d4eSChris Mason return fs_info->extent_root; 15708f18cf13SChris Mason if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID) 15718f18cf13SChris Mason return fs_info->chunk_root; 15728f18cf13SChris Mason if (location->objectid == BTRFS_DEV_TREE_OBJECTID) 15738f18cf13SChris Mason return fs_info->dev_root; 15740403e47eSYan Zheng if (location->objectid == BTRFS_CSUM_TREE_OBJECTID) 15750403e47eSYan Zheng return fs_info->csum_root; 1576bcef60f2SArne Jansen if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID) 1577bcef60f2SArne Jansen return fs_info->quota_root ? fs_info->quota_root : 1578bcef60f2SArne Jansen ERR_PTR(-ENOENT); 1579f7a81ea4SStefan Behrens if (location->objectid == BTRFS_UUID_TREE_OBJECTID) 1580f7a81ea4SStefan Behrens return fs_info->uuid_root ? fs_info->uuid_root : 1581f7a81ea4SStefan Behrens ERR_PTR(-ENOENT); 158270f6d82eSOmar Sandoval if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) 158370f6d82eSOmar Sandoval return fs_info->free_space_root ? fs_info->free_space_root : 158470f6d82eSOmar Sandoval ERR_PTR(-ENOENT); 15854df27c4dSYan, Zheng again: 1586cb517eabSMiao Xie root = btrfs_lookup_fs_root(fs_info, location->objectid); 158748475471SStefan Behrens if (root) { 1588c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) 158948475471SStefan Behrens return ERR_PTR(-ENOENT); 15905eda7b5eSChris Mason return root; 159148475471SStefan Behrens } 15925eda7b5eSChris Mason 1593cb517eabSMiao Xie root = btrfs_read_fs_root(fs_info->tree_root, location); 15945eda7b5eSChris Mason if (IS_ERR(root)) 15955eda7b5eSChris Mason return root; 15963394e160SChris Mason 1597c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1598d68fc57bSYan, Zheng ret = -ENOENT; 1599d68fc57bSYan, Zheng goto fail; 1600d68fc57bSYan, Zheng } 1601d68fc57bSYan, Zheng 1602cb517eabSMiao Xie ret = btrfs_init_fs_root(root); 1603cb517eabSMiao Xie if (ret) 1604cb517eabSMiao Xie goto fail; 1605cb517eabSMiao Xie 1606381cf658SDavid Sterba path = btrfs_alloc_path(); 1607381cf658SDavid Sterba if (!path) { 1608381cf658SDavid Sterba ret = -ENOMEM; 1609381cf658SDavid Sterba goto fail; 1610381cf658SDavid Sterba } 16111d4c08e0SDavid Sterba key.objectid = BTRFS_ORPHAN_OBJECTID; 16121d4c08e0SDavid Sterba key.type = BTRFS_ORPHAN_ITEM_KEY; 16131d4c08e0SDavid Sterba key.offset = location->objectid; 16141d4c08e0SDavid Sterba 16151d4c08e0SDavid Sterba ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1616381cf658SDavid Sterba btrfs_free_path(path); 1617d68fc57bSYan, Zheng if (ret < 0) 1618d68fc57bSYan, Zheng goto fail; 1619d68fc57bSYan, Zheng if (ret == 0) 162027cdeb70SMiao Xie set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1621d68fc57bSYan, Zheng 1622cb517eabSMiao Xie ret = btrfs_insert_fs_root(fs_info, root); 16230f7d52f4SChris Mason if (ret) { 16244df27c4dSYan, Zheng if (ret == -EEXIST) { 162584db5ccfSDavid Sterba btrfs_free_fs_root(root); 16264df27c4dSYan, Zheng goto again; 16270f7d52f4SChris Mason } 16284df27c4dSYan, Zheng goto fail; 16294df27c4dSYan, Zheng } 1630edbd8d4eSChris Mason return root; 16314df27c4dSYan, Zheng fail: 163284db5ccfSDavid Sterba btrfs_free_fs_root(root); 16334df27c4dSYan, Zheng return ERR_PTR(ret); 1634edbd8d4eSChris Mason } 1635edbd8d4eSChris Mason 163604160088SChris Mason static int btrfs_congested_fn(void *congested_data, int bdi_bits) 163704160088SChris Mason { 163804160088SChris Mason struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; 163904160088SChris Mason int ret = 0; 164004160088SChris Mason struct btrfs_device *device; 164104160088SChris Mason struct backing_dev_info *bdi; 1642b7967db7SChris Mason 16431f78160cSXiao Guangrong rcu_read_lock(); 16441f78160cSXiao Guangrong list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) { 1645dfe25020SChris Mason if (!device->bdev) 1646dfe25020SChris Mason continue; 1647efa7c9f9SJan Kara bdi = device->bdev->bd_bdi; 1648ff9ea323STejun Heo if (bdi_congested(bdi, bdi_bits)) { 164904160088SChris Mason ret = 1; 165004160088SChris Mason break; 165104160088SChris Mason } 165204160088SChris Mason } 16531f78160cSXiao Guangrong rcu_read_unlock(); 165404160088SChris Mason return ret; 165504160088SChris Mason } 165604160088SChris Mason 16578b712842SChris Mason /* 16588b712842SChris Mason * called by the kthread helper functions to finally call the bio end_io 16598b712842SChris Mason * functions. This is where read checksum verification actually happens 16608b712842SChris Mason */ 16618b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work) 1662ce9adaa5SChris Mason { 1663ce9adaa5SChris Mason struct bio *bio; 166497eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 1665ce9adaa5SChris Mason 166697eb6b69SDavid Sterba end_io_wq = container_of(work, struct btrfs_end_io_wq, work); 1667ce9adaa5SChris Mason bio = end_io_wq->bio; 16688b712842SChris Mason 16694e4cbee9SChristoph Hellwig bio->bi_status = end_io_wq->status; 1670ce9adaa5SChris Mason bio->bi_private = end_io_wq->private; 1671ce9adaa5SChris Mason bio->bi_end_io = end_io_wq->end_io; 167297eb6b69SDavid Sterba kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq); 16734246a0b6SChristoph Hellwig bio_endio(bio); 1674ce9adaa5SChris Mason } 167544b8bd7eSChris Mason 1676a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1677a74a4b97SChris Mason { 1678a74a4b97SChris Mason struct btrfs_root *root = arg; 16790b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1680d0278245SMiao Xie int again; 1681a74a4b97SChris Mason 1682d6fd0ae2SOmar Sandoval while (1) { 1683d0278245SMiao Xie again = 0; 16849d1a2a3aSDavid Sterba 1685fd340d0fSJosef Bacik set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1686fd340d0fSJosef Bacik 1687d0278245SMiao Xie /* Make the cleaner go to sleep early. */ 16882ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) 1689d0278245SMiao Xie goto sleep; 1690d0278245SMiao Xie 169190c711abSZygo Blaxell /* 169290c711abSZygo Blaxell * Do not do anything if we might cause open_ctree() to block 169390c711abSZygo Blaxell * before we have finished mounting the filesystem. 169490c711abSZygo Blaxell */ 16950b246afaSJeff Mahoney if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 169690c711abSZygo Blaxell goto sleep; 169790c711abSZygo Blaxell 16980b246afaSJeff Mahoney if (!mutex_trylock(&fs_info->cleaner_mutex)) 1699d0278245SMiao Xie goto sleep; 1700d0278245SMiao Xie 1701dc7f370cSMiao Xie /* 1702dc7f370cSMiao Xie * Avoid the problem that we change the status of the fs 1703dc7f370cSMiao Xie * during the above check and trylock. 1704dc7f370cSMiao Xie */ 17052ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) { 17060b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1707dc7f370cSMiao Xie goto sleep; 1708dc7f370cSMiao Xie } 1709dc7f370cSMiao Xie 17100b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_delayed_iput_mutex); 17112ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 17120b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_delayed_iput_mutex); 1713c2d6cb16SFilipe Manana 17149d1a2a3aSDavid Sterba again = btrfs_clean_one_deleted_snapshot(root); 17150b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1716a74a4b97SChris Mason 1717d0278245SMiao Xie /* 171805323cd1SMiao Xie * The defragger has dealt with the R/O remount and umount, 171905323cd1SMiao Xie * needn't do anything special here. 1720d0278245SMiao Xie */ 17210b246afaSJeff Mahoney btrfs_run_defrag_inodes(fs_info); 172267c5e7d4SFilipe Manana 172367c5e7d4SFilipe Manana /* 172467c5e7d4SFilipe Manana * Acquires fs_info->delete_unused_bgs_mutex to avoid racing 172567c5e7d4SFilipe Manana * with relocation (btrfs_relocate_chunk) and relocation 172667c5e7d4SFilipe Manana * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 172767c5e7d4SFilipe Manana * after acquiring fs_info->delete_unused_bgs_mutex. So we 172867c5e7d4SFilipe Manana * can't hold, nor need to, fs_info->cleaner_mutex when deleting 172967c5e7d4SFilipe Manana * unused block groups. 173067c5e7d4SFilipe Manana */ 17310b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 1732d0278245SMiao Xie sleep: 1733fd340d0fSJosef Bacik clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1734d6fd0ae2SOmar Sandoval if (kthread_should_park()) 1735d6fd0ae2SOmar Sandoval kthread_parkme(); 1736d6fd0ae2SOmar Sandoval if (kthread_should_stop()) 1737d6fd0ae2SOmar Sandoval return 0; 1738838fe188SJiri Kosina if (!again) { 1739a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 1740a74a4b97SChris Mason schedule(); 1741a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1742a74a4b97SChris Mason } 1743da288d28SFilipe Manana } 1744a74a4b97SChris Mason } 1745a74a4b97SChris Mason 1746a74a4b97SChris Mason static int transaction_kthread(void *arg) 1747a74a4b97SChris Mason { 1748a74a4b97SChris Mason struct btrfs_root *root = arg; 17490b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1750a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1751a74a4b97SChris Mason struct btrfs_transaction *cur; 17528929ecfaSYan, Zheng u64 transid; 1753a944442cSAllen Pais time64_t now; 1754a74a4b97SChris Mason unsigned long delay; 1755914b2007SJan Kara bool cannot_commit; 1756a74a4b97SChris Mason 1757a74a4b97SChris Mason do { 1758914b2007SJan Kara cannot_commit = false; 17590b246afaSJeff Mahoney delay = HZ * fs_info->commit_interval; 17600b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 1761a74a4b97SChris Mason 17620b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 17630b246afaSJeff Mahoney cur = fs_info->running_transaction; 1764a74a4b97SChris Mason if (!cur) { 17650b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1766a74a4b97SChris Mason goto sleep; 1767a74a4b97SChris Mason } 176831153d81SYan Zheng 1769afd48513SArnd Bergmann now = ktime_get_seconds(); 17704a9d8bdeSMiao Xie if (cur->state < TRANS_STATE_BLOCKED && 1771a514d638SQu Wenruo !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) && 17728b87dc17SDavid Sterba (now < cur->start_time || 17730b246afaSJeff Mahoney now - cur->start_time < fs_info->commit_interval)) { 17740b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1775a74a4b97SChris Mason delay = HZ * 5; 1776a74a4b97SChris Mason goto sleep; 1777a74a4b97SChris Mason } 17788929ecfaSYan, Zheng transid = cur->transid; 17790b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 178056bec294SChris Mason 178179787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 1782354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 1783914b2007SJan Kara if (IS_ERR(trans)) { 1784354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 1785914b2007SJan Kara cannot_commit = true; 178679787eaaSJeff Mahoney goto sleep; 1787914b2007SJan Kara } 17888929ecfaSYan, Zheng if (transid == trans->transid) { 17893a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 17908929ecfaSYan, Zheng } else { 17913a45bb20SJeff Mahoney btrfs_end_transaction(trans); 17928929ecfaSYan, Zheng } 1793a74a4b97SChris Mason sleep: 17940b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 17950b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 1796a74a4b97SChris Mason 17974e121c06SJosef Bacik if (unlikely(test_bit(BTRFS_FS_STATE_ERROR, 17980b246afaSJeff Mahoney &fs_info->fs_state))) 17992ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 18008929ecfaSYan, Zheng if (!kthread_should_stop() && 18010b246afaSJeff Mahoney (!btrfs_transaction_blocked(fs_info) || 1802914b2007SJan Kara cannot_commit)) 1803bc5511d0SNikolay Borisov schedule_timeout_interruptible(delay); 1804a74a4b97SChris Mason } while (!kthread_should_stop()); 1805a74a4b97SChris Mason return 0; 1806a74a4b97SChris Mason } 1807a74a4b97SChris Mason 1808af31f5e5SChris Mason /* 1809af31f5e5SChris Mason * this will find the highest generation in the array of 1810af31f5e5SChris Mason * root backups. The index of the highest array is returned, 1811af31f5e5SChris Mason * or -1 if we can't find anything. 1812af31f5e5SChris Mason * 1813af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1814af31f5e5SChris Mason * generation of the latest root in the array with the generation 1815af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1816af31f5e5SChris Mason */ 1817af31f5e5SChris Mason static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen) 1818af31f5e5SChris Mason { 1819af31f5e5SChris Mason u64 cur; 1820af31f5e5SChris Mason int newest_index = -1; 1821af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1822af31f5e5SChris Mason int i; 1823af31f5e5SChris Mason 1824af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1825af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1826af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1827af31f5e5SChris Mason if (cur == newest_gen) 1828af31f5e5SChris Mason newest_index = i; 1829af31f5e5SChris Mason } 1830af31f5e5SChris Mason 1831af31f5e5SChris Mason /* check to see if we actually wrapped around */ 1832af31f5e5SChris Mason if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) { 1833af31f5e5SChris Mason root_backup = info->super_copy->super_roots; 1834af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1835af31f5e5SChris Mason if (cur == newest_gen) 1836af31f5e5SChris Mason newest_index = 0; 1837af31f5e5SChris Mason } 1838af31f5e5SChris Mason return newest_index; 1839af31f5e5SChris Mason } 1840af31f5e5SChris Mason 1841af31f5e5SChris Mason 1842af31f5e5SChris Mason /* 1843af31f5e5SChris Mason * find the oldest backup so we know where to store new entries 1844af31f5e5SChris Mason * in the backup array. This will set the backup_root_index 1845af31f5e5SChris Mason * field in the fs_info struct 1846af31f5e5SChris Mason */ 1847af31f5e5SChris Mason static void find_oldest_super_backup(struct btrfs_fs_info *info, 1848af31f5e5SChris Mason u64 newest_gen) 1849af31f5e5SChris Mason { 1850af31f5e5SChris Mason int newest_index = -1; 1851af31f5e5SChris Mason 1852af31f5e5SChris Mason newest_index = find_newest_super_backup(info, newest_gen); 1853af31f5e5SChris Mason /* if there was garbage in there, just move along */ 1854af31f5e5SChris Mason if (newest_index == -1) { 1855af31f5e5SChris Mason info->backup_root_index = 0; 1856af31f5e5SChris Mason } else { 1857af31f5e5SChris Mason info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS; 1858af31f5e5SChris Mason } 1859af31f5e5SChris Mason } 1860af31f5e5SChris Mason 1861af31f5e5SChris Mason /* 1862af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1863af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1864af31f5e5SChris Mason * done 1865af31f5e5SChris Mason */ 1866af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1867af31f5e5SChris Mason { 1868af31f5e5SChris Mason int next_backup; 1869af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1870af31f5e5SChris Mason int last_backup; 1871af31f5e5SChris Mason 1872af31f5e5SChris Mason next_backup = info->backup_root_index; 1873af31f5e5SChris Mason last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) % 1874af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 1875af31f5e5SChris Mason 1876af31f5e5SChris Mason /* 1877af31f5e5SChris Mason * just overwrite the last backup if we're at the same generation 1878af31f5e5SChris Mason * this happens only at umount 1879af31f5e5SChris Mason */ 1880af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + last_backup; 1881af31f5e5SChris Mason if (btrfs_backup_tree_root_gen(root_backup) == 1882af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)) 1883af31f5e5SChris Mason next_backup = last_backup; 1884af31f5e5SChris Mason 1885af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1886af31f5e5SChris Mason 1887af31f5e5SChris Mason /* 1888af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1889af31f5e5SChris Mason * regardless of which ones we use today 1890af31f5e5SChris Mason */ 1891af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1892af31f5e5SChris Mason 1893af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1894af31f5e5SChris Mason 1895af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1896af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1897af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1898af31f5e5SChris Mason 1899af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1900af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1901af31f5e5SChris Mason 1902af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1903af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1904af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 1905af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 1906af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 1907af31f5e5SChris Mason 1908af31f5e5SChris Mason btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); 1909af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 1910af31f5e5SChris Mason btrfs_header_generation(info->extent_root->node)); 1911af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 1912af31f5e5SChris Mason btrfs_header_level(info->extent_root->node)); 1913af31f5e5SChris Mason 19147c7e82a7SChris Mason /* 19157c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 19167c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 19177c7e82a7SChris Mason */ 19187c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 19197c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 19207c7e82a7SChris Mason info->fs_root->node->start); 1921af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 1922af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 1923af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 1924af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 19257c7e82a7SChris Mason } 1926af31f5e5SChris Mason 1927af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 1928af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 1929af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 1930af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 1931af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 1932af31f5e5SChris Mason 1933af31f5e5SChris Mason btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); 1934af31f5e5SChris Mason btrfs_set_backup_csum_root_gen(root_backup, 1935af31f5e5SChris Mason btrfs_header_generation(info->csum_root->node)); 1936af31f5e5SChris Mason btrfs_set_backup_csum_root_level(root_backup, 1937af31f5e5SChris Mason btrfs_header_level(info->csum_root->node)); 1938af31f5e5SChris Mason 1939af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 1940af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 1941af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 1942af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 1943af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 1944af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 1945af31f5e5SChris Mason 1946af31f5e5SChris Mason /* 1947af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 1948af31f5e5SChris Mason * for the next commit 1949af31f5e5SChris Mason */ 1950af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 1951af31f5e5SChris Mason &info->super_for_commit->super_roots, 1952af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 1953af31f5e5SChris Mason } 1954af31f5e5SChris Mason 1955af31f5e5SChris Mason /* 1956af31f5e5SChris Mason * this copies info out of the root backup array and back into 1957af31f5e5SChris Mason * the in-memory super block. It is meant to help iterate through 1958af31f5e5SChris Mason * the array, so you send it the number of backups you've already 1959af31f5e5SChris Mason * tried and the last backup index you used. 1960af31f5e5SChris Mason * 1961af31f5e5SChris Mason * this returns -1 when it has tried all the backups 1962af31f5e5SChris Mason */ 1963af31f5e5SChris Mason static noinline int next_root_backup(struct btrfs_fs_info *info, 1964af31f5e5SChris Mason struct btrfs_super_block *super, 1965af31f5e5SChris Mason int *num_backups_tried, int *backup_index) 1966af31f5e5SChris Mason { 1967af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1968af31f5e5SChris Mason int newest = *backup_index; 1969af31f5e5SChris Mason 1970af31f5e5SChris Mason if (*num_backups_tried == 0) { 1971af31f5e5SChris Mason u64 gen = btrfs_super_generation(super); 1972af31f5e5SChris Mason 1973af31f5e5SChris Mason newest = find_newest_super_backup(info, gen); 1974af31f5e5SChris Mason if (newest == -1) 1975af31f5e5SChris Mason return -1; 1976af31f5e5SChris Mason 1977af31f5e5SChris Mason *backup_index = newest; 1978af31f5e5SChris Mason *num_backups_tried = 1; 1979af31f5e5SChris Mason } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) { 1980af31f5e5SChris Mason /* we've tried all the backups, all done */ 1981af31f5e5SChris Mason return -1; 1982af31f5e5SChris Mason } else { 1983af31f5e5SChris Mason /* jump to the next oldest backup */ 1984af31f5e5SChris Mason newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) % 1985af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 1986af31f5e5SChris Mason *backup_index = newest; 1987af31f5e5SChris Mason *num_backups_tried += 1; 1988af31f5e5SChris Mason } 1989af31f5e5SChris Mason root_backup = super->super_roots + newest; 1990af31f5e5SChris Mason 1991af31f5e5SChris Mason btrfs_set_super_generation(super, 1992af31f5e5SChris Mason btrfs_backup_tree_root_gen(root_backup)); 1993af31f5e5SChris Mason btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 1994af31f5e5SChris Mason btrfs_set_super_root_level(super, 1995af31f5e5SChris Mason btrfs_backup_tree_root_level(root_backup)); 1996af31f5e5SChris Mason btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 1997af31f5e5SChris Mason 1998af31f5e5SChris Mason /* 1999af31f5e5SChris Mason * fixme: the total bytes and num_devices need to match or we should 2000af31f5e5SChris Mason * need a fsck 2001af31f5e5SChris Mason */ 2002af31f5e5SChris Mason btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 2003af31f5e5SChris Mason btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 2004af31f5e5SChris Mason return 0; 2005af31f5e5SChris Mason } 2006af31f5e5SChris Mason 20077abadb64SLiu Bo /* helper to cleanup workers */ 20087abadb64SLiu Bo static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 20097abadb64SLiu Bo { 2010dc6e3209SQu Wenruo btrfs_destroy_workqueue(fs_info->fixup_workers); 2011afe3d242SQu Wenruo btrfs_destroy_workqueue(fs_info->delalloc_workers); 20125cdc7ad3SQu Wenruo btrfs_destroy_workqueue(fs_info->workers); 2013fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_workers); 2014fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_raid56_workers); 20158b110e39SMiao Xie btrfs_destroy_workqueue(fs_info->endio_repair_workers); 2016d05a33acSQu Wenruo btrfs_destroy_workqueue(fs_info->rmw_workers); 2017fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_write_workers); 2018fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 2019a8c93d4eSQu Wenruo btrfs_destroy_workqueue(fs_info->submit_workers); 20205b3bc44eSQu Wenruo btrfs_destroy_workqueue(fs_info->delayed_workers); 2021e66f0bb1SQu Wenruo btrfs_destroy_workqueue(fs_info->caching_workers); 2022736cfa15SQu Wenruo btrfs_destroy_workqueue(fs_info->readahead_workers); 2023a44903abSQu Wenruo btrfs_destroy_workqueue(fs_info->flush_workers); 2024fc97fab0SQu Wenruo btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2025a79b7d4bSChris Mason btrfs_destroy_workqueue(fs_info->extent_workers); 2026a9b9477dSFilipe Manana /* 2027a9b9477dSFilipe Manana * Now that all other work queues are destroyed, we can safely destroy 2028a9b9477dSFilipe Manana * the queues used for metadata I/O, since tasks from those other work 2029a9b9477dSFilipe Manana * queues can do metadata I/O operations. 2030a9b9477dSFilipe Manana */ 2031a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_workers); 2032a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_write_workers); 20337abadb64SLiu Bo } 20347abadb64SLiu Bo 20352e9f5954SRashika static void free_root_extent_buffers(struct btrfs_root *root) 20362e9f5954SRashika { 20372e9f5954SRashika if (root) { 20382e9f5954SRashika free_extent_buffer(root->node); 20392e9f5954SRashika free_extent_buffer(root->commit_root); 20402e9f5954SRashika root->node = NULL; 20412e9f5954SRashika root->commit_root = NULL; 20422e9f5954SRashika } 20432e9f5954SRashika } 20442e9f5954SRashika 2045af31f5e5SChris Mason /* helper to cleanup tree roots */ 2046af31f5e5SChris Mason static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root) 2047af31f5e5SChris Mason { 20482e9f5954SRashika free_root_extent_buffers(info->tree_root); 2049655b09feSJosef Bacik 20502e9f5954SRashika free_root_extent_buffers(info->dev_root); 20512e9f5954SRashika free_root_extent_buffers(info->extent_root); 20522e9f5954SRashika free_root_extent_buffers(info->csum_root); 20532e9f5954SRashika free_root_extent_buffers(info->quota_root); 20542e9f5954SRashika free_root_extent_buffers(info->uuid_root); 20552e9f5954SRashika if (chunk_root) 20562e9f5954SRashika free_root_extent_buffers(info->chunk_root); 205770f6d82eSOmar Sandoval free_root_extent_buffers(info->free_space_root); 2058af31f5e5SChris Mason } 2059af31f5e5SChris Mason 2060faa2dbf0SJosef Bacik void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2061171f6537SJosef Bacik { 2062171f6537SJosef Bacik int ret; 2063171f6537SJosef Bacik struct btrfs_root *gang[8]; 2064171f6537SJosef Bacik int i; 2065171f6537SJosef Bacik 2066171f6537SJosef Bacik while (!list_empty(&fs_info->dead_roots)) { 2067171f6537SJosef Bacik gang[0] = list_entry(fs_info->dead_roots.next, 2068171f6537SJosef Bacik struct btrfs_root, root_list); 2069171f6537SJosef Bacik list_del(&gang[0]->root_list); 2070171f6537SJosef Bacik 207127cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) { 2072cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[0]); 2073171f6537SJosef Bacik } else { 2074171f6537SJosef Bacik free_extent_buffer(gang[0]->node); 2075171f6537SJosef Bacik free_extent_buffer(gang[0]->commit_root); 2076b0feb9d9SMiao Xie btrfs_put_fs_root(gang[0]); 2077171f6537SJosef Bacik } 2078171f6537SJosef Bacik } 2079171f6537SJosef Bacik 2080171f6537SJosef Bacik while (1) { 2081171f6537SJosef Bacik ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2082171f6537SJosef Bacik (void **)gang, 0, 2083171f6537SJosef Bacik ARRAY_SIZE(gang)); 2084171f6537SJosef Bacik if (!ret) 2085171f6537SJosef Bacik break; 2086171f6537SJosef Bacik for (i = 0; i < ret; i++) 2087cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2088171f6537SJosef Bacik } 20891a4319ccSLiu Bo 20901a4319ccSLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 20911a4319ccSLiu Bo btrfs_free_log_root_tree(NULL, fs_info); 20922ff7e61eSJeff Mahoney btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents); 20931a4319ccSLiu Bo } 2094171f6537SJosef Bacik } 2095af31f5e5SChris Mason 2096638aa7edSEric Sandeen static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2097638aa7edSEric Sandeen { 2098638aa7edSEric Sandeen mutex_init(&fs_info->scrub_lock); 2099638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_running, 0); 2100638aa7edSEric Sandeen atomic_set(&fs_info->scrub_pause_req, 0); 2101638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_paused, 0); 2102638aa7edSEric Sandeen atomic_set(&fs_info->scrub_cancel_req, 0); 2103638aa7edSEric Sandeen init_waitqueue_head(&fs_info->scrub_pause_wait); 2104638aa7edSEric Sandeen fs_info->scrub_workers_refcnt = 0; 2105638aa7edSEric Sandeen } 2106638aa7edSEric Sandeen 2107779a65a4SEric Sandeen static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2108779a65a4SEric Sandeen { 2109779a65a4SEric Sandeen spin_lock_init(&fs_info->balance_lock); 2110779a65a4SEric Sandeen mutex_init(&fs_info->balance_mutex); 2111779a65a4SEric Sandeen atomic_set(&fs_info->balance_pause_req, 0); 2112779a65a4SEric Sandeen atomic_set(&fs_info->balance_cancel_req, 0); 2113779a65a4SEric Sandeen fs_info->balance_ctl = NULL; 2114779a65a4SEric Sandeen init_waitqueue_head(&fs_info->balance_wait_q); 2115779a65a4SEric Sandeen } 2116779a65a4SEric Sandeen 21176bccf3abSJeff Mahoney static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info) 2118f37938e0SEric Sandeen { 21192ff7e61eSJeff Mahoney struct inode *inode = fs_info->btree_inode; 21202ff7e61eSJeff Mahoney 21212ff7e61eSJeff Mahoney inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 21222ff7e61eSJeff Mahoney set_nlink(inode, 1); 2123f37938e0SEric Sandeen /* 2124f37938e0SEric Sandeen * we set the i_size on the btree inode to the max possible int. 2125f37938e0SEric Sandeen * the real end of the address space is determined by all of 2126f37938e0SEric Sandeen * the devices in the system 2127f37938e0SEric Sandeen */ 21282ff7e61eSJeff Mahoney inode->i_size = OFFSET_MAX; 21292ff7e61eSJeff Mahoney inode->i_mapping->a_ops = &btree_aops; 2130f37938e0SEric Sandeen 21312ff7e61eSJeff Mahoney RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 2132c6100a4bSJosef Bacik extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode); 21332ff7e61eSJeff Mahoney BTRFS_I(inode)->io_tree.track_uptodate = 0; 21342ff7e61eSJeff Mahoney extent_map_tree_init(&BTRFS_I(inode)->extent_tree); 2135f37938e0SEric Sandeen 21362ff7e61eSJeff Mahoney BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops; 2137f37938e0SEric Sandeen 21382ff7e61eSJeff Mahoney BTRFS_I(inode)->root = fs_info->tree_root; 21392ff7e61eSJeff Mahoney memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key)); 21402ff7e61eSJeff Mahoney set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); 21412ff7e61eSJeff Mahoney btrfs_insert_inode_hash(inode); 2142f37938e0SEric Sandeen } 2143f37938e0SEric Sandeen 2144ad618368SEric Sandeen static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2145ad618368SEric Sandeen { 2146ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2147129827e3SDavid Sterba init_rwsem(&fs_info->dev_replace.rwsem); 21487f8d236aSDavid Sterba init_waitqueue_head(&fs_info->dev_replace.replace_wait); 2149ad618368SEric Sandeen } 2150ad618368SEric Sandeen 2151f9e92e40SEric Sandeen static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2152f9e92e40SEric Sandeen { 2153f9e92e40SEric Sandeen spin_lock_init(&fs_info->qgroup_lock); 2154f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_ioctl_lock); 2155f9e92e40SEric Sandeen fs_info->qgroup_tree = RB_ROOT; 2156f9e92e40SEric Sandeen fs_info->qgroup_op_tree = RB_ROOT; 2157f9e92e40SEric Sandeen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2158f9e92e40SEric Sandeen fs_info->qgroup_seq = 1; 2159f9e92e40SEric Sandeen fs_info->qgroup_ulist = NULL; 2160d2c609b8SJeff Mahoney fs_info->qgroup_rescan_running = false; 2161f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_rescan_lock); 2162f9e92e40SEric Sandeen } 2163f9e92e40SEric Sandeen 21642a458198SEric Sandeen static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info, 21652a458198SEric Sandeen struct btrfs_fs_devices *fs_devices) 21662a458198SEric Sandeen { 2167f7b885beSAnand Jain u32 max_active = fs_info->thread_pool_size; 21686f011058SDavid Sterba unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 21692a458198SEric Sandeen 21702a458198SEric Sandeen fs_info->workers = 2171cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "worker", 2172cb001095SJeff Mahoney flags | WQ_HIGHPRI, max_active, 16); 21732a458198SEric Sandeen 21742a458198SEric Sandeen fs_info->delalloc_workers = 2175cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delalloc", 2176cb001095SJeff Mahoney flags, max_active, 2); 21772a458198SEric Sandeen 21782a458198SEric Sandeen fs_info->flush_workers = 2179cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "flush_delalloc", 2180cb001095SJeff Mahoney flags, max_active, 0); 21812a458198SEric Sandeen 21822a458198SEric Sandeen fs_info->caching_workers = 2183cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); 21842a458198SEric Sandeen 21852a458198SEric Sandeen /* 21862a458198SEric Sandeen * a higher idle thresh on the submit workers makes it much more 21872a458198SEric Sandeen * likely that bios will be send down in a sane order to the 21882a458198SEric Sandeen * devices 21892a458198SEric Sandeen */ 21902a458198SEric Sandeen fs_info->submit_workers = 2191cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "submit", flags, 21922a458198SEric Sandeen min_t(u64, fs_devices->num_devices, 21932a458198SEric Sandeen max_active), 64); 21942a458198SEric Sandeen 21952a458198SEric Sandeen fs_info->fixup_workers = 2196cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); 21972a458198SEric Sandeen 21982a458198SEric Sandeen /* 21992a458198SEric Sandeen * endios are largely parallel and should have a very 22002a458198SEric Sandeen * low idle thresh 22012a458198SEric Sandeen */ 22022a458198SEric Sandeen fs_info->endio_workers = 2203cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4); 22042a458198SEric Sandeen fs_info->endio_meta_workers = 2205cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta", flags, 2206cb001095SJeff Mahoney max_active, 4); 22072a458198SEric Sandeen fs_info->endio_meta_write_workers = 2208cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags, 2209cb001095SJeff Mahoney max_active, 2); 22102a458198SEric Sandeen fs_info->endio_raid56_workers = 2211cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-raid56", flags, 2212cb001095SJeff Mahoney max_active, 4); 22132a458198SEric Sandeen fs_info->endio_repair_workers = 2214cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0); 22152a458198SEric Sandeen fs_info->rmw_workers = 2216cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2); 22172a458198SEric Sandeen fs_info->endio_write_workers = 2218cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-write", flags, 2219cb001095SJeff Mahoney max_active, 2); 22202a458198SEric Sandeen fs_info->endio_freespace_worker = 2221cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "freespace-write", flags, 2222cb001095SJeff Mahoney max_active, 0); 22232a458198SEric Sandeen fs_info->delayed_workers = 2224cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, 2225cb001095SJeff Mahoney max_active, 0); 22262a458198SEric Sandeen fs_info->readahead_workers = 2227cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "readahead", flags, 2228cb001095SJeff Mahoney max_active, 2); 22292a458198SEric Sandeen fs_info->qgroup_rescan_workers = 2230cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); 22312a458198SEric Sandeen fs_info->extent_workers = 2232cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "extent-refs", flags, 22332a458198SEric Sandeen min_t(u64, fs_devices->num_devices, 22342a458198SEric Sandeen max_active), 8); 22352a458198SEric Sandeen 22362a458198SEric Sandeen if (!(fs_info->workers && fs_info->delalloc_workers && 22372a458198SEric Sandeen fs_info->submit_workers && fs_info->flush_workers && 22382a458198SEric Sandeen fs_info->endio_workers && fs_info->endio_meta_workers && 22392a458198SEric Sandeen fs_info->endio_meta_write_workers && 22402a458198SEric Sandeen fs_info->endio_repair_workers && 22412a458198SEric Sandeen fs_info->endio_write_workers && fs_info->endio_raid56_workers && 22422a458198SEric Sandeen fs_info->endio_freespace_worker && fs_info->rmw_workers && 22432a458198SEric Sandeen fs_info->caching_workers && fs_info->readahead_workers && 22442a458198SEric Sandeen fs_info->fixup_workers && fs_info->delayed_workers && 22452a458198SEric Sandeen fs_info->extent_workers && 22462a458198SEric Sandeen fs_info->qgroup_rescan_workers)) { 22472a458198SEric Sandeen return -ENOMEM; 22482a458198SEric Sandeen } 22492a458198SEric Sandeen 22502a458198SEric Sandeen return 0; 22512a458198SEric Sandeen } 22522a458198SEric Sandeen 225363443bf5SEric Sandeen static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 225463443bf5SEric Sandeen struct btrfs_fs_devices *fs_devices) 225563443bf5SEric Sandeen { 225663443bf5SEric Sandeen int ret; 225763443bf5SEric Sandeen struct btrfs_root *log_tree_root; 225863443bf5SEric Sandeen struct btrfs_super_block *disk_super = fs_info->super_copy; 225963443bf5SEric Sandeen u64 bytenr = btrfs_super_log_root(disk_super); 2260581c1760SQu Wenruo int level = btrfs_super_log_root_level(disk_super); 226163443bf5SEric Sandeen 226263443bf5SEric Sandeen if (fs_devices->rw_devices == 0) { 2263f14d104dSDavid Sterba btrfs_warn(fs_info, "log replay required on RO media"); 226463443bf5SEric Sandeen return -EIO; 226563443bf5SEric Sandeen } 226663443bf5SEric Sandeen 226774e4d827SDavid Sterba log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL); 226863443bf5SEric Sandeen if (!log_tree_root) 226963443bf5SEric Sandeen return -ENOMEM; 227063443bf5SEric Sandeen 2271da17066cSJeff Mahoney __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID); 227263443bf5SEric Sandeen 22732ff7e61eSJeff Mahoney log_tree_root->node = read_tree_block(fs_info, bytenr, 2274581c1760SQu Wenruo fs_info->generation + 1, 2275581c1760SQu Wenruo level, NULL); 227664c043deSLiu Bo if (IS_ERR(log_tree_root->node)) { 2277f14d104dSDavid Sterba btrfs_warn(fs_info, "failed to read log tree"); 22780eeff236SLiu Bo ret = PTR_ERR(log_tree_root->node); 227964c043deSLiu Bo kfree(log_tree_root); 22800eeff236SLiu Bo return ret; 228164c043deSLiu Bo } else if (!extent_buffer_uptodate(log_tree_root->node)) { 2282f14d104dSDavid Sterba btrfs_err(fs_info, "failed to read log tree"); 228363443bf5SEric Sandeen free_extent_buffer(log_tree_root->node); 228463443bf5SEric Sandeen kfree(log_tree_root); 228563443bf5SEric Sandeen return -EIO; 228663443bf5SEric Sandeen } 228763443bf5SEric Sandeen /* returns with log_tree_root freed on success */ 228863443bf5SEric Sandeen ret = btrfs_recover_log_trees(log_tree_root); 228963443bf5SEric Sandeen if (ret) { 22900b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 229163443bf5SEric Sandeen "Failed to recover log tree"); 229263443bf5SEric Sandeen free_extent_buffer(log_tree_root->node); 229363443bf5SEric Sandeen kfree(log_tree_root); 229463443bf5SEric Sandeen return ret; 229563443bf5SEric Sandeen } 229663443bf5SEric Sandeen 2297bc98a42cSDavid Howells if (sb_rdonly(fs_info->sb)) { 22986bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 229963443bf5SEric Sandeen if (ret) 230063443bf5SEric Sandeen return ret; 230163443bf5SEric Sandeen } 230263443bf5SEric Sandeen 230363443bf5SEric Sandeen return 0; 230463443bf5SEric Sandeen } 230563443bf5SEric Sandeen 23066bccf3abSJeff Mahoney static int btrfs_read_roots(struct btrfs_fs_info *fs_info) 23074bbcaa64SEric Sandeen { 23086bccf3abSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 2309a4f3d2c4SDavid Sterba struct btrfs_root *root; 23104bbcaa64SEric Sandeen struct btrfs_key location; 23114bbcaa64SEric Sandeen int ret; 23124bbcaa64SEric Sandeen 23136bccf3abSJeff Mahoney BUG_ON(!fs_info->tree_root); 23146bccf3abSJeff Mahoney 23154bbcaa64SEric Sandeen location.objectid = BTRFS_EXTENT_TREE_OBJECTID; 23164bbcaa64SEric Sandeen location.type = BTRFS_ROOT_ITEM_KEY; 23174bbcaa64SEric Sandeen location.offset = 0; 23184bbcaa64SEric Sandeen 2319a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2320f50f4353SLiu Bo if (IS_ERR(root)) { 2321f50f4353SLiu Bo ret = PTR_ERR(root); 2322f50f4353SLiu Bo goto out; 2323f50f4353SLiu Bo } 2324a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2325a4f3d2c4SDavid Sterba fs_info->extent_root = root; 23264bbcaa64SEric Sandeen 23274bbcaa64SEric Sandeen location.objectid = BTRFS_DEV_TREE_OBJECTID; 2328a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2329f50f4353SLiu Bo if (IS_ERR(root)) { 2330f50f4353SLiu Bo ret = PTR_ERR(root); 2331f50f4353SLiu Bo goto out; 2332f50f4353SLiu Bo } 2333a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2334a4f3d2c4SDavid Sterba fs_info->dev_root = root; 23354bbcaa64SEric Sandeen btrfs_init_devices_late(fs_info); 23364bbcaa64SEric Sandeen 23374bbcaa64SEric Sandeen location.objectid = BTRFS_CSUM_TREE_OBJECTID; 2338a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2339f50f4353SLiu Bo if (IS_ERR(root)) { 2340f50f4353SLiu Bo ret = PTR_ERR(root); 2341f50f4353SLiu Bo goto out; 2342f50f4353SLiu Bo } 2343a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2344a4f3d2c4SDavid Sterba fs_info->csum_root = root; 23454bbcaa64SEric Sandeen 23464bbcaa64SEric Sandeen location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2347a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2348a4f3d2c4SDavid Sterba if (!IS_ERR(root)) { 2349a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2350afcdd129SJosef Bacik set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); 2351a4f3d2c4SDavid Sterba fs_info->quota_root = root; 23524bbcaa64SEric Sandeen } 23534bbcaa64SEric Sandeen 23544bbcaa64SEric Sandeen location.objectid = BTRFS_UUID_TREE_OBJECTID; 2355a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2356a4f3d2c4SDavid Sterba if (IS_ERR(root)) { 2357a4f3d2c4SDavid Sterba ret = PTR_ERR(root); 23584bbcaa64SEric Sandeen if (ret != -ENOENT) 2359f50f4353SLiu Bo goto out; 23604bbcaa64SEric Sandeen } else { 2361a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2362a4f3d2c4SDavid Sterba fs_info->uuid_root = root; 23634bbcaa64SEric Sandeen } 23644bbcaa64SEric Sandeen 236570f6d82eSOmar Sandoval if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 236670f6d82eSOmar Sandoval location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID; 236770f6d82eSOmar Sandoval root = btrfs_read_tree_root(tree_root, &location); 2368f50f4353SLiu Bo if (IS_ERR(root)) { 2369f50f4353SLiu Bo ret = PTR_ERR(root); 2370f50f4353SLiu Bo goto out; 2371f50f4353SLiu Bo } 237270f6d82eSOmar Sandoval set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 237370f6d82eSOmar Sandoval fs_info->free_space_root = root; 237470f6d82eSOmar Sandoval } 237570f6d82eSOmar Sandoval 23764bbcaa64SEric Sandeen return 0; 2377f50f4353SLiu Bo out: 2378f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", 2379f50f4353SLiu Bo location.objectid, ret); 2380f50f4353SLiu Bo return ret; 23814bbcaa64SEric Sandeen } 23824bbcaa64SEric Sandeen 2383069ec957SQu Wenruo /* 2384069ec957SQu Wenruo * Real super block validation 2385069ec957SQu Wenruo * NOTE: super csum type and incompat features will not be checked here. 2386069ec957SQu Wenruo * 2387069ec957SQu Wenruo * @sb: super block to check 2388069ec957SQu Wenruo * @mirror_num: the super block number to check its bytenr: 2389069ec957SQu Wenruo * 0 the primary (1st) sb 2390069ec957SQu Wenruo * 1, 2 2nd and 3rd backup copy 2391069ec957SQu Wenruo * -1 skip bytenr check 2392069ec957SQu Wenruo */ 2393069ec957SQu Wenruo static int validate_super(struct btrfs_fs_info *fs_info, 2394069ec957SQu Wenruo struct btrfs_super_block *sb, int mirror_num) 239521a852b0SQu Wenruo { 239621a852b0SQu Wenruo u64 nodesize = btrfs_super_nodesize(sb); 239721a852b0SQu Wenruo u64 sectorsize = btrfs_super_sectorsize(sb); 239821a852b0SQu Wenruo int ret = 0; 239921a852b0SQu Wenruo 240021a852b0SQu Wenruo if (btrfs_super_magic(sb) != BTRFS_MAGIC) { 240121a852b0SQu Wenruo btrfs_err(fs_info, "no valid FS found"); 240221a852b0SQu Wenruo ret = -EINVAL; 240321a852b0SQu Wenruo } 240421a852b0SQu Wenruo if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { 240521a852b0SQu Wenruo btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", 240621a852b0SQu Wenruo btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); 240721a852b0SQu Wenruo ret = -EINVAL; 240821a852b0SQu Wenruo } 240921a852b0SQu Wenruo if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 241021a852b0SQu Wenruo btrfs_err(fs_info, "tree_root level too big: %d >= %d", 241121a852b0SQu Wenruo btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 241221a852b0SQu Wenruo ret = -EINVAL; 241321a852b0SQu Wenruo } 241421a852b0SQu Wenruo if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 241521a852b0SQu Wenruo btrfs_err(fs_info, "chunk_root level too big: %d >= %d", 241621a852b0SQu Wenruo btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 241721a852b0SQu Wenruo ret = -EINVAL; 241821a852b0SQu Wenruo } 241921a852b0SQu Wenruo if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 242021a852b0SQu Wenruo btrfs_err(fs_info, "log_root level too big: %d >= %d", 242121a852b0SQu Wenruo btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 242221a852b0SQu Wenruo ret = -EINVAL; 242321a852b0SQu Wenruo } 242421a852b0SQu Wenruo 242521a852b0SQu Wenruo /* 242621a852b0SQu Wenruo * Check sectorsize and nodesize first, other check will need it. 242721a852b0SQu Wenruo * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. 242821a852b0SQu Wenruo */ 242921a852b0SQu Wenruo if (!is_power_of_2(sectorsize) || sectorsize < 4096 || 243021a852b0SQu Wenruo sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { 243121a852b0SQu Wenruo btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); 243221a852b0SQu Wenruo ret = -EINVAL; 243321a852b0SQu Wenruo } 243421a852b0SQu Wenruo /* Only PAGE SIZE is supported yet */ 243521a852b0SQu Wenruo if (sectorsize != PAGE_SIZE) { 243621a852b0SQu Wenruo btrfs_err(fs_info, 243721a852b0SQu Wenruo "sectorsize %llu not supported yet, only support %lu", 243821a852b0SQu Wenruo sectorsize, PAGE_SIZE); 243921a852b0SQu Wenruo ret = -EINVAL; 244021a852b0SQu Wenruo } 244121a852b0SQu Wenruo if (!is_power_of_2(nodesize) || nodesize < sectorsize || 244221a852b0SQu Wenruo nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { 244321a852b0SQu Wenruo btrfs_err(fs_info, "invalid nodesize %llu", nodesize); 244421a852b0SQu Wenruo ret = -EINVAL; 244521a852b0SQu Wenruo } 244621a852b0SQu Wenruo if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { 244721a852b0SQu Wenruo btrfs_err(fs_info, "invalid leafsize %u, should be %llu", 244821a852b0SQu Wenruo le32_to_cpu(sb->__unused_leafsize), nodesize); 244921a852b0SQu Wenruo ret = -EINVAL; 245021a852b0SQu Wenruo } 245121a852b0SQu Wenruo 245221a852b0SQu Wenruo /* Root alignment check */ 245321a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { 245421a852b0SQu Wenruo btrfs_warn(fs_info, "tree_root block unaligned: %llu", 245521a852b0SQu Wenruo btrfs_super_root(sb)); 245621a852b0SQu Wenruo ret = -EINVAL; 245721a852b0SQu Wenruo } 245821a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { 245921a852b0SQu Wenruo btrfs_warn(fs_info, "chunk_root block unaligned: %llu", 246021a852b0SQu Wenruo btrfs_super_chunk_root(sb)); 246121a852b0SQu Wenruo ret = -EINVAL; 246221a852b0SQu Wenruo } 246321a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { 246421a852b0SQu Wenruo btrfs_warn(fs_info, "log_root block unaligned: %llu", 246521a852b0SQu Wenruo btrfs_super_log_root(sb)); 246621a852b0SQu Wenruo ret = -EINVAL; 246721a852b0SQu Wenruo } 246821a852b0SQu Wenruo 2469de37aa51SNikolay Borisov if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, 24707239ff4bSNikolay Borisov BTRFS_FSID_SIZE) != 0) { 247121a852b0SQu Wenruo btrfs_err(fs_info, 24727239ff4bSNikolay Borisov "dev_item UUID does not match metadata fsid: %pU != %pU", 2473de37aa51SNikolay Borisov fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); 247421a852b0SQu Wenruo ret = -EINVAL; 247521a852b0SQu Wenruo } 247621a852b0SQu Wenruo 247721a852b0SQu Wenruo /* 247821a852b0SQu Wenruo * Hint to catch really bogus numbers, bitflips or so, more exact checks are 247921a852b0SQu Wenruo * done later 248021a852b0SQu Wenruo */ 248121a852b0SQu Wenruo if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { 248221a852b0SQu Wenruo btrfs_err(fs_info, "bytes_used is too small %llu", 248321a852b0SQu Wenruo btrfs_super_bytes_used(sb)); 248421a852b0SQu Wenruo ret = -EINVAL; 248521a852b0SQu Wenruo } 248621a852b0SQu Wenruo if (!is_power_of_2(btrfs_super_stripesize(sb))) { 248721a852b0SQu Wenruo btrfs_err(fs_info, "invalid stripesize %u", 248821a852b0SQu Wenruo btrfs_super_stripesize(sb)); 248921a852b0SQu Wenruo ret = -EINVAL; 249021a852b0SQu Wenruo } 249121a852b0SQu Wenruo if (btrfs_super_num_devices(sb) > (1UL << 31)) 249221a852b0SQu Wenruo btrfs_warn(fs_info, "suspicious number of devices: %llu", 249321a852b0SQu Wenruo btrfs_super_num_devices(sb)); 249421a852b0SQu Wenruo if (btrfs_super_num_devices(sb) == 0) { 249521a852b0SQu Wenruo btrfs_err(fs_info, "number of devices is 0"); 249621a852b0SQu Wenruo ret = -EINVAL; 249721a852b0SQu Wenruo } 249821a852b0SQu Wenruo 2499069ec957SQu Wenruo if (mirror_num >= 0 && 2500069ec957SQu Wenruo btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { 250121a852b0SQu Wenruo btrfs_err(fs_info, "super offset mismatch %llu != %u", 250221a852b0SQu Wenruo btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 250321a852b0SQu Wenruo ret = -EINVAL; 250421a852b0SQu Wenruo } 250521a852b0SQu Wenruo 250621a852b0SQu Wenruo /* 250721a852b0SQu Wenruo * Obvious sys_chunk_array corruptions, it must hold at least one key 250821a852b0SQu Wenruo * and one chunk 250921a852b0SQu Wenruo */ 251021a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 251121a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too big %u > %u", 251221a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 251321a852b0SQu Wenruo BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 251421a852b0SQu Wenruo ret = -EINVAL; 251521a852b0SQu Wenruo } 251621a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 251721a852b0SQu Wenruo + sizeof(struct btrfs_chunk)) { 251821a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too small %u < %zu", 251921a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 252021a852b0SQu Wenruo sizeof(struct btrfs_disk_key) 252121a852b0SQu Wenruo + sizeof(struct btrfs_chunk)); 252221a852b0SQu Wenruo ret = -EINVAL; 252321a852b0SQu Wenruo } 252421a852b0SQu Wenruo 252521a852b0SQu Wenruo /* 252621a852b0SQu Wenruo * The generation is a global counter, we'll trust it more than the others 252721a852b0SQu Wenruo * but it's still possible that it's the one that's wrong. 252821a852b0SQu Wenruo */ 252921a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 253021a852b0SQu Wenruo btrfs_warn(fs_info, 253121a852b0SQu Wenruo "suspicious: generation < chunk_root_generation: %llu < %llu", 253221a852b0SQu Wenruo btrfs_super_generation(sb), 253321a852b0SQu Wenruo btrfs_super_chunk_root_generation(sb)); 253421a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 253521a852b0SQu Wenruo && btrfs_super_cache_generation(sb) != (u64)-1) 253621a852b0SQu Wenruo btrfs_warn(fs_info, 253721a852b0SQu Wenruo "suspicious: generation < cache_generation: %llu < %llu", 253821a852b0SQu Wenruo btrfs_super_generation(sb), 253921a852b0SQu Wenruo btrfs_super_cache_generation(sb)); 254021a852b0SQu Wenruo 254121a852b0SQu Wenruo return ret; 254221a852b0SQu Wenruo } 254321a852b0SQu Wenruo 2544069ec957SQu Wenruo /* 2545069ec957SQu Wenruo * Validation of super block at mount time. 2546069ec957SQu Wenruo * Some checks already done early at mount time, like csum type and incompat 2547069ec957SQu Wenruo * flags will be skipped. 2548069ec957SQu Wenruo */ 2549069ec957SQu Wenruo static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) 2550069ec957SQu Wenruo { 2551069ec957SQu Wenruo return validate_super(fs_info, fs_info->super_copy, 0); 2552069ec957SQu Wenruo } 2553069ec957SQu Wenruo 255475cb857dSQu Wenruo /* 255575cb857dSQu Wenruo * Validation of super block at write time. 255675cb857dSQu Wenruo * Some checks like bytenr check will be skipped as their values will be 255775cb857dSQu Wenruo * overwritten soon. 255875cb857dSQu Wenruo * Extra checks like csum type and incompat flags will be done here. 255975cb857dSQu Wenruo */ 256075cb857dSQu Wenruo static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, 256175cb857dSQu Wenruo struct btrfs_super_block *sb) 256275cb857dSQu Wenruo { 256375cb857dSQu Wenruo int ret; 256475cb857dSQu Wenruo 256575cb857dSQu Wenruo ret = validate_super(fs_info, sb, -1); 256675cb857dSQu Wenruo if (ret < 0) 256775cb857dSQu Wenruo goto out; 256875cb857dSQu Wenruo if (btrfs_super_csum_type(sb) != BTRFS_CSUM_TYPE_CRC32) { 256975cb857dSQu Wenruo ret = -EUCLEAN; 257075cb857dSQu Wenruo btrfs_err(fs_info, "invalid csum type, has %u want %u", 257175cb857dSQu Wenruo btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); 257275cb857dSQu Wenruo goto out; 257375cb857dSQu Wenruo } 257475cb857dSQu Wenruo if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 257575cb857dSQu Wenruo ret = -EUCLEAN; 257675cb857dSQu Wenruo btrfs_err(fs_info, 257775cb857dSQu Wenruo "invalid incompat flags, has 0x%llx valid mask 0x%llx", 257875cb857dSQu Wenruo btrfs_super_incompat_flags(sb), 257975cb857dSQu Wenruo (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); 258075cb857dSQu Wenruo goto out; 258175cb857dSQu Wenruo } 258275cb857dSQu Wenruo out: 258375cb857dSQu Wenruo if (ret < 0) 258475cb857dSQu Wenruo btrfs_err(fs_info, 258575cb857dSQu Wenruo "super block corruption detected before writing it to disk"); 258675cb857dSQu Wenruo return ret; 258775cb857dSQu Wenruo } 258875cb857dSQu Wenruo 2589ad2b2c80SAl Viro int open_ctree(struct super_block *sb, 2590dfe25020SChris Mason struct btrfs_fs_devices *fs_devices, 2591dfe25020SChris Mason char *options) 2592eb60ceacSChris Mason { 2593db94535dSChris Mason u32 sectorsize; 2594db94535dSChris Mason u32 nodesize; 259587ee04ebSChris Mason u32 stripesize; 259684234f3aSYan Zheng u64 generation; 2597f2b636e8SJosef Bacik u64 features; 25983de4586cSChris Mason struct btrfs_key location; 2599a061fc8dSChris Mason struct buffer_head *bh; 26004d34b278SIlya Dryomov struct btrfs_super_block *disk_super; 2601815745cfSAl Viro struct btrfs_fs_info *fs_info = btrfs_sb(sb); 2602f84a8bd6SAl Viro struct btrfs_root *tree_root; 26034d34b278SIlya Dryomov struct btrfs_root *chunk_root; 2604eb60ceacSChris Mason int ret; 2605e58ca020SYan int err = -EINVAL; 2606af31f5e5SChris Mason int num_backups_tried = 0; 2607af31f5e5SChris Mason int backup_index = 0; 26086675df31SOmar Sandoval int clear_free_space_tree = 0; 2609581c1760SQu Wenruo int level; 26104543df7eSChris Mason 261174e4d827SDavid Sterba tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL); 261274e4d827SDavid Sterba chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL); 2613cb517eabSMiao Xie if (!tree_root || !chunk_root) { 261439279cc3SChris Mason err = -ENOMEM; 261539279cc3SChris Mason goto fail; 261639279cc3SChris Mason } 261776dda93cSYan, Zheng 261876dda93cSYan, Zheng ret = init_srcu_struct(&fs_info->subvol_srcu); 261976dda93cSYan, Zheng if (ret) { 262076dda93cSYan, Zheng err = ret; 262176dda93cSYan, Zheng goto fail; 262276dda93cSYan, Zheng } 262376dda93cSYan, Zheng 2624908c7f19STejun Heo ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 2625e2d84521SMiao Xie if (ret) { 2626e2d84521SMiao Xie err = ret; 26279e11ceeeSJan Kara goto fail_srcu; 2628e2d84521SMiao Xie } 262909cbfeafSKirill A. Shutemov fs_info->dirty_metadata_batch = PAGE_SIZE * 2630e2d84521SMiao Xie (1 + ilog2(nr_cpu_ids)); 2631e2d84521SMiao Xie 2632908c7f19STejun Heo ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 2633963d678bSMiao Xie if (ret) { 2634963d678bSMiao Xie err = ret; 2635963d678bSMiao Xie goto fail_dirty_metadata_bytes; 2636963d678bSMiao Xie } 2637963d678bSMiao Xie 26387f8d236aSDavid Sterba ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, 26397f8d236aSDavid Sterba GFP_KERNEL); 2640c404e0dcSMiao Xie if (ret) { 2641c404e0dcSMiao Xie err = ret; 2642c404e0dcSMiao Xie goto fail_delalloc_bytes; 2643c404e0dcSMiao Xie } 2644c404e0dcSMiao Xie 264576dda93cSYan, Zheng INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 2646f28491e0SJosef Bacik INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 26478fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 2648facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 264924bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 2650eb73c1b7SMiao Xie INIT_LIST_HEAD(&fs_info->delalloc_roots); 265111833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 265275cb379dSJeff Mahoney INIT_LIST_HEAD(&fs_info->pending_raid_kobjs); 265375cb379dSJeff Mahoney spin_lock_init(&fs_info->pending_raid_kobjs_lock); 2654eb73c1b7SMiao Xie spin_lock_init(&fs_info->delalloc_root_lock); 2655a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 265676dda93cSYan, Zheng spin_lock_init(&fs_info->fs_roots_radix_lock); 265724bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 26584cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 2659f29021b2SJan Schmidt spin_lock_init(&fs_info->tree_mod_seq_lock); 2660ceda0864SMiao Xie spin_lock_init(&fs_info->super_lock); 2661fcebe456SJosef Bacik spin_lock_init(&fs_info->qgroup_op_lock); 2662f28491e0SJosef Bacik spin_lock_init(&fs_info->buffer_lock); 266347ab2a6cSJosef Bacik spin_lock_init(&fs_info->unused_bgs_lock); 2664f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 2665d7c15171SZhao Lei mutex_init(&fs_info->unused_bg_unpin_mutex); 266667c5e7d4SFilipe Manana mutex_init(&fs_info->delete_unused_bgs_mutex); 26677585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 2668573bfb72SMiao Xie mutex_init(&fs_info->delalloc_root_mutex); 2669c2d6cb16SFilipe Manana mutex_init(&fs_info->cleaner_delayed_iput_mutex); 2670de98ced9SMiao Xie seqlock_init(&fs_info->profiles_lock); 267119c00ddcSChris Mason 26720b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 26736324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 2674f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 267547ab2a6cSJosef Bacik INIT_LIST_HEAD(&fs_info->unused_bgs); 26760b86a832SChris Mason btrfs_mapping_init(&fs_info->mapping_tree); 267766d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 267866d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 267966d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 268066d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 268166d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 268266d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 268366d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 2684ba2c4d4eSJosef Bacik btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, 2685ba2c4d4eSJosef Bacik BTRFS_BLOCK_RSV_DELREFS); 2686ba2c4d4eSJosef Bacik 2687771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 26884cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 2689fcebe456SJosef Bacik atomic_set(&fs_info->qgroup_op_seq, 0); 26902fefd558SZhao Lei atomic_set(&fs_info->reada_works_cnt, 0); 2691fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 2692e20d96d6SChris Mason fs_info->sb = sb; 269395ac567aSFilipe David Borba Manana fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 26949ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 26954cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 2696a5ed45f8SNikolay Borisov atomic64_set(&fs_info->free_chunk_space, 0); 2697f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 26988b87dc17SDavid Sterba fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 2699f8c269d7SDavid Sterba fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 270090519d66SArne Jansen /* readahead state */ 2701d0164adcSMel Gorman INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); 270290519d66SArne Jansen spin_lock_init(&fs_info->reada_lock); 2703fd708b81SJosef Bacik btrfs_init_ref_verify(fs_info); 2704c8b97818SChris Mason 2705b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 2706b34b086cSChris Mason num_online_cpus() + 2, 8); 27070afbaf8cSChris Mason 2708199c2a9cSMiao Xie INIT_LIST_HEAD(&fs_info->ordered_roots); 2709199c2a9cSMiao Xie spin_lock_init(&fs_info->ordered_root_lock); 271069fe2d75SJosef Bacik 271169fe2d75SJosef Bacik fs_info->btree_inode = new_inode(sb); 271269fe2d75SJosef Bacik if (!fs_info->btree_inode) { 271369fe2d75SJosef Bacik err = -ENOMEM; 271469fe2d75SJosef Bacik goto fail_bio_counter; 271569fe2d75SJosef Bacik } 271669fe2d75SJosef Bacik mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 271769fe2d75SJosef Bacik 271816cdcec7SMiao Xie fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 271974e4d827SDavid Sterba GFP_KERNEL); 272016cdcec7SMiao Xie if (!fs_info->delayed_root) { 272116cdcec7SMiao Xie err = -ENOMEM; 272216cdcec7SMiao Xie goto fail_iput; 272316cdcec7SMiao Xie } 272416cdcec7SMiao Xie btrfs_init_delayed_root(fs_info->delayed_root); 27253eaa2885SChris Mason 2726638aa7edSEric Sandeen btrfs_init_scrub(fs_info); 272721adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 272821adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 272921adbd5cSStefan Behrens #endif 2730779a65a4SEric Sandeen btrfs_init_balance(fs_info); 273121c7e756SMiao Xie btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work); 2732a2de733cSArne Jansen 27339f6d2510SDavid Sterba sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; 27349f6d2510SDavid Sterba sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); 2735a061fc8dSChris Mason 27366bccf3abSJeff Mahoney btrfs_init_btree_inode(fs_info); 273739279cc3SChris Mason 2738e02119d5SChris Mason spin_lock_init(&fs_info->block_group_cache_lock); 27396bef4d31SEric Paris fs_info->block_group_cache_tree = RB_ROOT; 2740a1897fddSLiu Bo fs_info->first_logical_byte = (u64)-1; 2741925baeddSChris Mason 2742c6100a4bSJosef Bacik extent_io_tree_init(&fs_info->freed_extents[0], NULL); 2743c6100a4bSJosef Bacik extent_io_tree_init(&fs_info->freed_extents[1], NULL); 274411833d66SYan Zheng fs_info->pinned_extents = &fs_info->freed_extents[0]; 2745afcdd129SJosef Bacik set_bit(BTRFS_FS_BARRIER, &fs_info->flags); 2746509659cdSChris Mason 27475a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 274879154b1bSChris Mason mutex_init(&fs_info->tree_log_mutex); 2749925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 2750a74a4b97SChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 2751a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 27521bbc621eSChris Mason mutex_init(&fs_info->ro_block_group_mutex); 27539e351cc8SJosef Bacik init_rwsem(&fs_info->commit_root_sem); 2754c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 275576dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 2756803b2f54SStefan Behrens sema_init(&fs_info->uuid_tree_rescan_sem, 1); 2757fa9c0d79SChris Mason 2758ad618368SEric Sandeen btrfs_init_dev_replace_locks(fs_info); 2759f9e92e40SEric Sandeen btrfs_init_qgroup(fs_info); 2760416ac51dSArne Jansen 2761fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2762fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2763fa9c0d79SChris Mason 2764e6dcd2dcSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 2765f9295749SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 2766bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 27674854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 27683768f368SChris Mason 276904216820SFilipe Manana INIT_LIST_HEAD(&fs_info->pinned_chunks); 277004216820SFilipe Manana 2771da17066cSJeff Mahoney /* Usable values until the real ones are cached from the superblock */ 2772da17066cSJeff Mahoney fs_info->nodesize = 4096; 2773da17066cSJeff Mahoney fs_info->sectorsize = 4096; 2774da17066cSJeff Mahoney fs_info->stripesize = 4096; 2775da17066cSJeff Mahoney 2776eede2bf3SOmar Sandoval spin_lock_init(&fs_info->swapfile_pins_lock); 2777eede2bf3SOmar Sandoval fs_info->swapfile_pins = RB_ROOT; 2778eede2bf3SOmar Sandoval 277953b381b3SDavid Woodhouse ret = btrfs_alloc_stripe_hash_table(fs_info); 278053b381b3SDavid Woodhouse if (ret) { 278183c8266aSDavid Sterba err = ret; 278253b381b3SDavid Woodhouse goto fail_alloc; 278353b381b3SDavid Woodhouse } 278453b381b3SDavid Woodhouse 2785da17066cSJeff Mahoney __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID); 27867eccb903SChris Mason 27873c4bb26bSChris Mason invalidate_bdev(fs_devices->latest_bdev); 27881104a885SDavid Sterba 27891104a885SDavid Sterba /* 27901104a885SDavid Sterba * Read super block and check the signature bytes only 27911104a885SDavid Sterba */ 2792a512bbf8SYan Zheng bh = btrfs_read_dev_super(fs_devices->latest_bdev); 279392fc03fbSAnand Jain if (IS_ERR(bh)) { 279492fc03fbSAnand Jain err = PTR_ERR(bh); 279516cdcec7SMiao Xie goto fail_alloc; 279620b45077SDave Young } 279739279cc3SChris Mason 27981104a885SDavid Sterba /* 27991104a885SDavid Sterba * We want to check superblock checksum, the type is stored inside. 28001104a885SDavid Sterba * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 28011104a885SDavid Sterba */ 2802ab8d0fc4SJeff Mahoney if (btrfs_check_super_csum(fs_info, bh->b_data)) { 280305135f59SDavid Sterba btrfs_err(fs_info, "superblock checksum mismatch"); 28041104a885SDavid Sterba err = -EINVAL; 2805b2acdddfSAnand Jain brelse(bh); 28061104a885SDavid Sterba goto fail_alloc; 28071104a885SDavid Sterba } 28081104a885SDavid Sterba 28091104a885SDavid Sterba /* 28101104a885SDavid Sterba * super_copy is zeroed at allocation time and we never touch the 28111104a885SDavid Sterba * following bytes up to INFO_SIZE, the checksum is calculated from 28121104a885SDavid Sterba * the whole block of INFO_SIZE 28131104a885SDavid Sterba */ 28146c41761fSDavid Sterba memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy)); 2815a061fc8dSChris Mason brelse(bh); 28165f39d397SChris Mason 2817fbc6feaeSNikolay Borisov disk_super = fs_info->super_copy; 2818fbc6feaeSNikolay Borisov 2819de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, 2820de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 2821de37aa51SNikolay Borisov 28227239ff4bSNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID)) { 2823de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid, 2824de37aa51SNikolay Borisov fs_info->super_copy->metadata_uuid, 2825de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 28267239ff4bSNikolay Borisov } 28270b86a832SChris Mason 2828fbc6feaeSNikolay Borisov features = btrfs_super_flags(disk_super); 2829fbc6feaeSNikolay Borisov if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { 2830fbc6feaeSNikolay Borisov features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; 2831fbc6feaeSNikolay Borisov btrfs_set_super_flags(disk_super, features); 2832fbc6feaeSNikolay Borisov btrfs_info(fs_info, 2833fbc6feaeSNikolay Borisov "found metadata UUID change in progress flag, clearing"); 2834fbc6feaeSNikolay Borisov } 2835fbc6feaeSNikolay Borisov 2836fbc6feaeSNikolay Borisov memcpy(fs_info->super_for_commit, fs_info->super_copy, 2837fbc6feaeSNikolay Borisov sizeof(*fs_info->super_for_commit)); 2838de37aa51SNikolay Borisov 2839069ec957SQu Wenruo ret = btrfs_validate_mount_super(fs_info); 28401104a885SDavid Sterba if (ret) { 284105135f59SDavid Sterba btrfs_err(fs_info, "superblock contains fatal errors"); 28421104a885SDavid Sterba err = -EINVAL; 28431104a885SDavid Sterba goto fail_alloc; 28441104a885SDavid Sterba } 28451104a885SDavid Sterba 28460f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 284716cdcec7SMiao Xie goto fail_alloc; 28480f7d52f4SChris Mason 2849acce952bSliubo /* check FS state, whether FS is broken. */ 285087533c47SMiao Xie if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 285187533c47SMiao Xie set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 2852acce952bSliubo 285375e7cb7fSLiu Bo /* 2854af31f5e5SChris Mason * run through our array of backup supers and setup 2855af31f5e5SChris Mason * our ring pointer to the oldest one 2856af31f5e5SChris Mason */ 2857af31f5e5SChris Mason generation = btrfs_super_generation(disk_super); 2858af31f5e5SChris Mason find_oldest_super_backup(fs_info, generation); 2859af31f5e5SChris Mason 2860af31f5e5SChris Mason /* 286175e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 286275e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 286375e7cb7fSLiu Bo */ 286475e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 286575e7cb7fSLiu Bo 28662ff7e61eSJeff Mahoney ret = btrfs_parse_options(fs_info, options, sb->s_flags); 28672b82032cSYan Zheng if (ret) { 28682b82032cSYan Zheng err = ret; 286916cdcec7SMiao Xie goto fail_alloc; 28702b82032cSYan Zheng } 2871dfe25020SChris Mason 2872f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 2873f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 2874f2b636e8SJosef Bacik if (features) { 287505135f59SDavid Sterba btrfs_err(fs_info, 287605135f59SDavid Sterba "cannot mount because of unsupported optional features (%llx)", 2877c1c9ff7cSGeert Uytterhoeven features); 2878f2b636e8SJosef Bacik err = -EINVAL; 287916cdcec7SMiao Xie goto fail_alloc; 2880f2b636e8SJosef Bacik } 2881f2b636e8SJosef Bacik 28825d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 28835d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 28840b246afaSJeff Mahoney if (fs_info->compress_type == BTRFS_COMPRESS_LZO) 2885a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 28865c1aab1dSNick Terrell else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) 28875c1aab1dSNick Terrell features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; 2888727011e0SChris Mason 28893173a18fSJosef Bacik if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) 289005135f59SDavid Sterba btrfs_info(fs_info, "has skinny extents"); 28913173a18fSJosef Bacik 2892727011e0SChris Mason /* 2893727011e0SChris Mason * flag our filesystem as having big metadata blocks if 2894727011e0SChris Mason * they are bigger than the page size 2895727011e0SChris Mason */ 289609cbfeafSKirill A. Shutemov if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) { 2897727011e0SChris Mason if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 289805135f59SDavid Sterba btrfs_info(fs_info, 289905135f59SDavid Sterba "flagging fs with big metadata feature"); 2900727011e0SChris Mason features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 2901727011e0SChris Mason } 2902727011e0SChris Mason 2903bc3f116fSChris Mason nodesize = btrfs_super_nodesize(disk_super); 2904bc3f116fSChris Mason sectorsize = btrfs_super_sectorsize(disk_super); 2905b7f67055SChandan Rajendra stripesize = sectorsize; 2906707e8a07SDavid Sterba fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 2907963d678bSMiao Xie fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 2908bc3f116fSChris Mason 2909da17066cSJeff Mahoney /* Cache block sizes */ 2910da17066cSJeff Mahoney fs_info->nodesize = nodesize; 2911da17066cSJeff Mahoney fs_info->sectorsize = sectorsize; 2912da17066cSJeff Mahoney fs_info->stripesize = stripesize; 2913da17066cSJeff Mahoney 2914bc3f116fSChris Mason /* 2915bc3f116fSChris Mason * mixed block groups end up with duplicate but slightly offset 2916bc3f116fSChris Mason * extent buffers for the same range. It leads to corruptions 2917bc3f116fSChris Mason */ 2918bc3f116fSChris Mason if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 2919707e8a07SDavid Sterba (sectorsize != nodesize)) { 292005135f59SDavid Sterba btrfs_err(fs_info, 292105135f59SDavid Sterba "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", 292205135f59SDavid Sterba nodesize, sectorsize); 2923bc3f116fSChris Mason goto fail_alloc; 2924bc3f116fSChris Mason } 2925bc3f116fSChris Mason 2926ceda0864SMiao Xie /* 2927ceda0864SMiao Xie * Needn't use the lock because there is no other task which will 2928ceda0864SMiao Xie * update the flag. 2929ceda0864SMiao Xie */ 29305d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 29315d4f98a2SYan Zheng 2932f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 2933f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 2934bc98a42cSDavid Howells if (!sb_rdonly(sb) && features) { 293505135f59SDavid Sterba btrfs_err(fs_info, 293605135f59SDavid Sterba "cannot mount read-write because of unsupported optional features (%llx)", 2937c1c9ff7cSGeert Uytterhoeven features); 2938f2b636e8SJosef Bacik err = -EINVAL; 293916cdcec7SMiao Xie goto fail_alloc; 2940f2b636e8SJosef Bacik } 294161d92c32SChris Mason 29422a458198SEric Sandeen ret = btrfs_init_workqueues(fs_info, fs_devices); 29432a458198SEric Sandeen if (ret) { 29442a458198SEric Sandeen err = ret; 29450dc3b84aSJosef Bacik goto fail_sb_buffer; 29460dc3b84aSJosef Bacik } 29474543df7eSChris Mason 29489e11ceeeSJan Kara sb->s_bdi->congested_fn = btrfs_congested_fn; 29499e11ceeeSJan Kara sb->s_bdi->congested_data = fs_info; 29509e11ceeeSJan Kara sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK; 2951d4417e22SNikolay Borisov sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE; 29529e11ceeeSJan Kara sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); 29539e11ceeeSJan Kara sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); 29544575c9ccSChris Mason 2955a061fc8dSChris Mason sb->s_blocksize = sectorsize; 2956a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 2957de37aa51SNikolay Borisov memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); 2958db94535dSChris Mason 2959925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 29606bccf3abSJeff Mahoney ret = btrfs_read_sys_array(fs_info); 2961925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 296284eed90fSChris Mason if (ret) { 296305135f59SDavid Sterba btrfs_err(fs_info, "failed to read the system array: %d", ret); 29645d4f98a2SYan Zheng goto fail_sb_buffer; 296584eed90fSChris Mason } 29660b86a832SChris Mason 296784234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 2968581c1760SQu Wenruo level = btrfs_super_chunk_root_level(disk_super); 29690b86a832SChris Mason 2970da17066cSJeff Mahoney __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID); 29710b86a832SChris Mason 29722ff7e61eSJeff Mahoney chunk_root->node = read_tree_block(fs_info, 29730b86a832SChris Mason btrfs_super_chunk_root(disk_super), 2974581c1760SQu Wenruo generation, level, NULL); 297564c043deSLiu Bo if (IS_ERR(chunk_root->node) || 297664c043deSLiu Bo !extent_buffer_uptodate(chunk_root->node)) { 297705135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk root"); 2978e5fffbacSchandan if (!IS_ERR(chunk_root->node)) 2979e5fffbacSchandan free_extent_buffer(chunk_root->node); 298095ab1f64SZhao Lei chunk_root->node = NULL; 2981af31f5e5SChris Mason goto fail_tree_roots; 298283121942SDavid Woodhouse } 29835d4f98a2SYan Zheng btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 29845d4f98a2SYan Zheng chunk_root->commit_root = btrfs_root_node(chunk_root); 29850b86a832SChris Mason 2986e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 2987b308bc2fSGeert Uytterhoeven btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE); 2988e17cade2SChris Mason 29895b4aacefSJeff Mahoney ret = btrfs_read_chunk_tree(fs_info); 29902b82032cSYan Zheng if (ret) { 299105135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk tree: %d", ret); 2992af31f5e5SChris Mason goto fail_tree_roots; 29932b82032cSYan Zheng } 29940b86a832SChris Mason 29958dabb742SStefan Behrens /* 29969b99b115SAnand Jain * Keep the devid that is marked to be the target device for the 29979b99b115SAnand Jain * device replace procedure 29988dabb742SStefan Behrens */ 29999b99b115SAnand Jain btrfs_free_extra_devids(fs_devices, 0); 3000dfe25020SChris Mason 3001a6b0d5c8SChris Mason if (!fs_devices->latest_bdev) { 300205135f59SDavid Sterba btrfs_err(fs_info, "failed to read devices"); 3003a6b0d5c8SChris Mason goto fail_tree_roots; 3004a6b0d5c8SChris Mason } 3005a6b0d5c8SChris Mason 3006af31f5e5SChris Mason retry_root_backup: 300784234f3aSYan Zheng generation = btrfs_super_generation(disk_super); 3008581c1760SQu Wenruo level = btrfs_super_root_level(disk_super); 30090b86a832SChris Mason 30102ff7e61eSJeff Mahoney tree_root->node = read_tree_block(fs_info, 3011db94535dSChris Mason btrfs_super_root(disk_super), 3012581c1760SQu Wenruo generation, level, NULL); 301364c043deSLiu Bo if (IS_ERR(tree_root->node) || 301464c043deSLiu Bo !extent_buffer_uptodate(tree_root->node)) { 301505135f59SDavid Sterba btrfs_warn(fs_info, "failed to read tree root"); 3016e5fffbacSchandan if (!IS_ERR(tree_root->node)) 3017e5fffbacSchandan free_extent_buffer(tree_root->node); 301895ab1f64SZhao Lei tree_root->node = NULL; 3019af31f5e5SChris Mason goto recovery_tree_root; 302083121942SDavid Woodhouse } 3021af31f5e5SChris Mason 30225d4f98a2SYan Zheng btrfs_set_root_node(&tree_root->root_item, tree_root->node); 30235d4f98a2SYan Zheng tree_root->commit_root = btrfs_root_node(tree_root); 302469e9c6c6SStefan Behrens btrfs_set_root_refs(&tree_root->root_item, 1); 3025db94535dSChris Mason 3026f32e48e9SChandan Rajendra mutex_lock(&tree_root->objectid_mutex); 3027f32e48e9SChandan Rajendra ret = btrfs_find_highest_objectid(tree_root, 3028f32e48e9SChandan Rajendra &tree_root->highest_objectid); 3029f32e48e9SChandan Rajendra if (ret) { 3030f32e48e9SChandan Rajendra mutex_unlock(&tree_root->objectid_mutex); 3031f32e48e9SChandan Rajendra goto recovery_tree_root; 3032f32e48e9SChandan Rajendra } 3033f32e48e9SChandan Rajendra 3034f32e48e9SChandan Rajendra ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 3035f32e48e9SChandan Rajendra 3036f32e48e9SChandan Rajendra mutex_unlock(&tree_root->objectid_mutex); 3037f32e48e9SChandan Rajendra 30386bccf3abSJeff Mahoney ret = btrfs_read_roots(fs_info); 30394bbcaa64SEric Sandeen if (ret) 3040af31f5e5SChris Mason goto recovery_tree_root; 3041f7a81ea4SStefan Behrens 30428929ecfaSYan, Zheng fs_info->generation = generation; 30438929ecfaSYan, Zheng fs_info->last_trans_committed = generation; 30448929ecfaSYan, Zheng 3045cf90d884SQu Wenruo ret = btrfs_verify_dev_extents(fs_info); 3046cf90d884SQu Wenruo if (ret) { 3047cf90d884SQu Wenruo btrfs_err(fs_info, 3048cf90d884SQu Wenruo "failed to verify dev extents against chunks: %d", 3049cf90d884SQu Wenruo ret); 3050cf90d884SQu Wenruo goto fail_block_groups; 3051cf90d884SQu Wenruo } 305268310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 305368310a5eSIlya Dryomov if (ret) { 305405135f59SDavid Sterba btrfs_err(fs_info, "failed to recover balance: %d", ret); 305568310a5eSIlya Dryomov goto fail_block_groups; 305668310a5eSIlya Dryomov } 305768310a5eSIlya Dryomov 3058733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 3059733f4fbbSStefan Behrens if (ret) { 306005135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_stats: %d", ret); 3061733f4fbbSStefan Behrens goto fail_block_groups; 3062733f4fbbSStefan Behrens } 3063733f4fbbSStefan Behrens 30648dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 30658dabb742SStefan Behrens if (ret) { 306605135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_replace: %d", ret); 30678dabb742SStefan Behrens goto fail_block_groups; 30688dabb742SStefan Behrens } 30698dabb742SStefan Behrens 30709b99b115SAnand Jain btrfs_free_extra_devids(fs_devices, 1); 30718dabb742SStefan Behrens 3072b7c35e81SAnand Jain ret = btrfs_sysfs_add_fsid(fs_devices, NULL); 3073b7c35e81SAnand Jain if (ret) { 307405135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", 307505135f59SDavid Sterba ret); 3076b7c35e81SAnand Jain goto fail_block_groups; 3077b7c35e81SAnand Jain } 3078b7c35e81SAnand Jain 3079b7c35e81SAnand Jain ret = btrfs_sysfs_add_device(fs_devices); 3080b7c35e81SAnand Jain if (ret) { 308105135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs device interface: %d", 308205135f59SDavid Sterba ret); 3083b7c35e81SAnand Jain goto fail_fsdev_sysfs; 3084b7c35e81SAnand Jain } 3085b7c35e81SAnand Jain 308696f3136eSAnand Jain ret = btrfs_sysfs_add_mounted(fs_info); 30875ac1d209SJeff Mahoney if (ret) { 308805135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); 3089b7c35e81SAnand Jain goto fail_fsdev_sysfs; 30905ac1d209SJeff Mahoney } 30915ac1d209SJeff Mahoney 3092c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 3093c59021f8Sliubo if (ret) { 309405135f59SDavid Sterba btrfs_err(fs_info, "failed to initialize space info: %d", ret); 30952365dd3cSAnand Jain goto fail_sysfs; 3096c59021f8Sliubo } 3097c59021f8Sliubo 30985b4aacefSJeff Mahoney ret = btrfs_read_block_groups(fs_info); 30991b1d1f66SJosef Bacik if (ret) { 310005135f59SDavid Sterba btrfs_err(fs_info, "failed to read block groups: %d", ret); 31012365dd3cSAnand Jain goto fail_sysfs; 31021b1d1f66SJosef Bacik } 31034330e183SQu Wenruo 31046528b99dSAnand Jain if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) { 310505135f59SDavid Sterba btrfs_warn(fs_info, 310652042d8eSAndrea Gelmini "writable mount is not allowed due to too many missing devices"); 31072365dd3cSAnand Jain goto fail_sysfs; 3108292fd7fcSStefan Behrens } 31099078a3e1SChris Mason 3110a74a4b97SChris Mason fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 3111a74a4b97SChris Mason "btrfs-cleaner"); 311257506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 31132365dd3cSAnand Jain goto fail_sysfs; 3114a74a4b97SChris Mason 3115a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 3116a74a4b97SChris Mason tree_root, 3117a74a4b97SChris Mason "btrfs-transaction"); 311857506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 31193f157a2fSChris Mason goto fail_cleaner; 3120a74a4b97SChris Mason 3121583b7231SHans van Kranenburg if (!btrfs_test_opt(fs_info, NOSSD) && 3122c289811cSChris Mason !fs_info->fs_devices->rotating) { 3123583b7231SHans van Kranenburg btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); 3124c289811cSChris Mason } 3125c289811cSChris Mason 3126572d9ab7SDavid Sterba /* 312701327610SNicholas D Steeves * Mount does not set all options immediately, we can do it now and do 3128572d9ab7SDavid Sterba * not have to wait for transaction commit 3129572d9ab7SDavid Sterba */ 3130572d9ab7SDavid Sterba btrfs_apply_pending_changes(fs_info); 31313818aea2SQu Wenruo 313221adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 31330b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { 31342ff7e61eSJeff Mahoney ret = btrfsic_mount(fs_info, fs_devices, 31350b246afaSJeff Mahoney btrfs_test_opt(fs_info, 313621adbd5cSStefan Behrens CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? 313721adbd5cSStefan Behrens 1 : 0, 313821adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 313921adbd5cSStefan Behrens if (ret) 314005135f59SDavid Sterba btrfs_warn(fs_info, 314105135f59SDavid Sterba "failed to initialize integrity check module: %d", 314205135f59SDavid Sterba ret); 314321adbd5cSStefan Behrens } 314421adbd5cSStefan Behrens #endif 3145bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 3146bcef60f2SArne Jansen if (ret) 3147bcef60f2SArne Jansen goto fail_trans_kthread; 314821adbd5cSStefan Behrens 3149fd708b81SJosef Bacik if (btrfs_build_ref_tree(fs_info)) 3150fd708b81SJosef Bacik btrfs_err(fs_info, "couldn't build ref tree"); 3151fd708b81SJosef Bacik 315296da0919SQu Wenruo /* do not make disk changes in broken FS or nologreplay is given */ 315396da0919SQu Wenruo if (btrfs_super_log_root(disk_super) != 0 && 31540b246afaSJeff Mahoney !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 315563443bf5SEric Sandeen ret = btrfs_replay_log(fs_info, fs_devices); 315679787eaaSJeff Mahoney if (ret) { 315763443bf5SEric Sandeen err = ret; 315828c16cbbSWang Shilong goto fail_qgroup; 3159e556ce2cSYan Zheng } 3160e02119d5SChris Mason } 31611a40e23bSZheng Yan 31626bccf3abSJeff Mahoney ret = btrfs_find_orphan_roots(fs_info); 316379787eaaSJeff Mahoney if (ret) 316428c16cbbSWang Shilong goto fail_qgroup; 316576dda93cSYan, Zheng 3166bc98a42cSDavid Howells if (!sb_rdonly(sb)) { 3167d68fc57bSYan, Zheng ret = btrfs_cleanup_fs_roots(fs_info); 316844c44af2SIlya Dryomov if (ret) 316928c16cbbSWang Shilong goto fail_qgroup; 317090c711abSZygo Blaxell 317190c711abSZygo Blaxell mutex_lock(&fs_info->cleaner_mutex); 31725d4f98a2SYan Zheng ret = btrfs_recover_relocation(tree_root); 317390c711abSZygo Blaxell mutex_unlock(&fs_info->cleaner_mutex); 3174d7ce5843SMiao Xie if (ret < 0) { 317505135f59SDavid Sterba btrfs_warn(fs_info, "failed to recover relocation: %d", 317605135f59SDavid Sterba ret); 3177d7ce5843SMiao Xie err = -EINVAL; 3178bcef60f2SArne Jansen goto fail_qgroup; 3179d7ce5843SMiao Xie } 31807c2ca468SChris Mason } 31811a40e23bSZheng Yan 31823de4586cSChris Mason location.objectid = BTRFS_FS_TREE_OBJECTID; 31833de4586cSChris Mason location.type = BTRFS_ROOT_ITEM_KEY; 3184cb517eabSMiao Xie location.offset = 0; 31853de4586cSChris Mason 31863de4586cSChris Mason fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location); 31873140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 31883140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 3189f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read fs tree: %d", err); 3190bcef60f2SArne Jansen goto fail_qgroup; 31913140c9a3SDan Carpenter } 3192c289811cSChris Mason 3193bc98a42cSDavid Howells if (sb_rdonly(sb)) 31942b6ba629SIlya Dryomov return 0; 31952b6ba629SIlya Dryomov 3196f8d468a1SOmar Sandoval if (btrfs_test_opt(fs_info, CLEAR_CACHE) && 3197f8d468a1SOmar Sandoval btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 31986675df31SOmar Sandoval clear_free_space_tree = 1; 31996675df31SOmar Sandoval } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 32006675df31SOmar Sandoval !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { 32016675df31SOmar Sandoval btrfs_warn(fs_info, "free space tree is invalid"); 32026675df31SOmar Sandoval clear_free_space_tree = 1; 32036675df31SOmar Sandoval } 32046675df31SOmar Sandoval 32056675df31SOmar Sandoval if (clear_free_space_tree) { 3206f8d468a1SOmar Sandoval btrfs_info(fs_info, "clearing free space tree"); 3207f8d468a1SOmar Sandoval ret = btrfs_clear_free_space_tree(fs_info); 3208f8d468a1SOmar Sandoval if (ret) { 3209f8d468a1SOmar Sandoval btrfs_warn(fs_info, 3210f8d468a1SOmar Sandoval "failed to clear free space tree: %d", ret); 32116bccf3abSJeff Mahoney close_ctree(fs_info); 3212f8d468a1SOmar Sandoval return ret; 3213f8d468a1SOmar Sandoval } 3214f8d468a1SOmar Sandoval } 3215f8d468a1SOmar Sandoval 32160b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && 3217511711afSChris Mason !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 321805135f59SDavid Sterba btrfs_info(fs_info, "creating free space tree"); 3219511711afSChris Mason ret = btrfs_create_free_space_tree(fs_info); 3220511711afSChris Mason if (ret) { 322105135f59SDavid Sterba btrfs_warn(fs_info, 322205135f59SDavid Sterba "failed to create free space tree: %d", ret); 32236bccf3abSJeff Mahoney close_ctree(fs_info); 3224511711afSChris Mason return ret; 3225511711afSChris Mason } 3226511711afSChris Mason } 3227511711afSChris Mason 3228e3acc2a6SJosef Bacik down_read(&fs_info->cleanup_work_sem); 32292b6ba629SIlya Dryomov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 32302b6ba629SIlya Dryomov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 32312b6ba629SIlya Dryomov up_read(&fs_info->cleanup_work_sem); 32326bccf3abSJeff Mahoney close_ctree(fs_info); 32332b6ba629SIlya Dryomov return ret; 32342b6ba629SIlya Dryomov } 3235e3acc2a6SJosef Bacik up_read(&fs_info->cleanup_work_sem); 323659641015SIlya Dryomov 32372b6ba629SIlya Dryomov ret = btrfs_resume_balance_async(fs_info); 32382b6ba629SIlya Dryomov if (ret) { 323905135f59SDavid Sterba btrfs_warn(fs_info, "failed to resume balance: %d", ret); 32406bccf3abSJeff Mahoney close_ctree(fs_info); 32412b6ba629SIlya Dryomov return ret; 3242e3acc2a6SJosef Bacik } 3243e3acc2a6SJosef Bacik 32448dabb742SStefan Behrens ret = btrfs_resume_dev_replace_async(fs_info); 32458dabb742SStefan Behrens if (ret) { 324605135f59SDavid Sterba btrfs_warn(fs_info, "failed to resume device replace: %d", ret); 32476bccf3abSJeff Mahoney close_ctree(fs_info); 32488dabb742SStefan Behrens return ret; 32498dabb742SStefan Behrens } 32508dabb742SStefan Behrens 3251b382a324SJan Schmidt btrfs_qgroup_rescan_resume(fs_info); 3252b382a324SJan Schmidt 32534bbcaa64SEric Sandeen if (!fs_info->uuid_root) { 325405135f59SDavid Sterba btrfs_info(fs_info, "creating UUID tree"); 3255f7a81ea4SStefan Behrens ret = btrfs_create_uuid_tree(fs_info); 3256f7a81ea4SStefan Behrens if (ret) { 325705135f59SDavid Sterba btrfs_warn(fs_info, 325805135f59SDavid Sterba "failed to create the UUID tree: %d", ret); 32596bccf3abSJeff Mahoney close_ctree(fs_info); 3260f7a81ea4SStefan Behrens return ret; 3261f7a81ea4SStefan Behrens } 32620b246afaSJeff Mahoney } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || 32634bbcaa64SEric Sandeen fs_info->generation != 32644bbcaa64SEric Sandeen btrfs_super_uuid_tree_generation(disk_super)) { 326505135f59SDavid Sterba btrfs_info(fs_info, "checking UUID tree"); 326670f80175SStefan Behrens ret = btrfs_check_uuid_tree(fs_info); 326770f80175SStefan Behrens if (ret) { 326805135f59SDavid Sterba btrfs_warn(fs_info, 326905135f59SDavid Sterba "failed to check the UUID tree: %d", ret); 32706bccf3abSJeff Mahoney close_ctree(fs_info); 327170f80175SStefan Behrens return ret; 327270f80175SStefan Behrens } 327370f80175SStefan Behrens } else { 3274afcdd129SJosef Bacik set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 3275f7a81ea4SStefan Behrens } 3276afcdd129SJosef Bacik set_bit(BTRFS_FS_OPEN, &fs_info->flags); 327747ab2a6cSJosef Bacik 32788dcddfa0SQu Wenruo /* 32798dcddfa0SQu Wenruo * backuproot only affect mount behavior, and if open_ctree succeeded, 32808dcddfa0SQu Wenruo * no need to keep the flag 32818dcddfa0SQu Wenruo */ 32828dcddfa0SQu Wenruo btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT); 32838dcddfa0SQu Wenruo 3284ad2b2c80SAl Viro return 0; 328539279cc3SChris Mason 3286bcef60f2SArne Jansen fail_qgroup: 3287bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 32887c2ca468SChris Mason fail_trans_kthread: 32897c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 32902ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 3291faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 32923f157a2fSChris Mason fail_cleaner: 3293a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 32947c2ca468SChris Mason 32957c2ca468SChris Mason /* 32967c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 32977c2ca468SChris Mason * kthreads 32987c2ca468SChris Mason */ 32997c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 33007c2ca468SChris Mason 33012365dd3cSAnand Jain fail_sysfs: 33026618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 33032365dd3cSAnand Jain 3304b7c35e81SAnand Jain fail_fsdev_sysfs: 3305b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3306b7c35e81SAnand Jain 33071b1d1f66SJosef Bacik fail_block_groups: 330854067ae9SJosef Bacik btrfs_put_block_group_cache(fs_info); 3309af31f5e5SChris Mason 3310af31f5e5SChris Mason fail_tree_roots: 3311af31f5e5SChris Mason free_root_pointers(fs_info, 1); 33122b8195bbSMiao Xie invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3313af31f5e5SChris Mason 331439279cc3SChris Mason fail_sb_buffer: 33157abadb64SLiu Bo btrfs_stop_all_workers(fs_info); 33165cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 331716cdcec7SMiao Xie fail_alloc: 33184543df7eSChris Mason fail_iput: 3319586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 3320586e46e2SIlya Dryomov 33214543df7eSChris Mason iput(fs_info->btree_inode); 3322c404e0dcSMiao Xie fail_bio_counter: 33237f8d236aSDavid Sterba percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 3324963d678bSMiao Xie fail_delalloc_bytes: 3325963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 3326e2d84521SMiao Xie fail_dirty_metadata_bytes: 3327e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 332876dda93cSYan, Zheng fail_srcu: 332976dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 33307e662854SQinghuang Feng fail: 333153b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 3332586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 3333ad2b2c80SAl Viro return err; 3334af31f5e5SChris Mason 3335af31f5e5SChris Mason recovery_tree_root: 33360b246afaSJeff Mahoney if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) 3337af31f5e5SChris Mason goto fail_tree_roots; 3338af31f5e5SChris Mason 3339af31f5e5SChris Mason free_root_pointers(fs_info, 0); 3340af31f5e5SChris Mason 3341af31f5e5SChris Mason /* don't use the log in recovery mode, it won't be valid */ 3342af31f5e5SChris Mason btrfs_set_super_log_root(disk_super, 0); 3343af31f5e5SChris Mason 3344af31f5e5SChris Mason /* we can't trust the free space cache either */ 3345af31f5e5SChris Mason btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 3346af31f5e5SChris Mason 3347af31f5e5SChris Mason ret = next_root_backup(fs_info, fs_info->super_copy, 3348af31f5e5SChris Mason &num_backups_tried, &backup_index); 3349af31f5e5SChris Mason if (ret == -1) 3350af31f5e5SChris Mason goto fail_block_groups; 3351af31f5e5SChris Mason goto retry_root_backup; 3352eb60ceacSChris Mason } 3353663faf9fSMasami Hiramatsu ALLOW_ERROR_INJECTION(open_ctree, ERRNO); 3354eb60ceacSChris Mason 3355f2984462SChris Mason static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate) 3356f2984462SChris Mason { 3357f2984462SChris Mason if (uptodate) { 3358f2984462SChris Mason set_buffer_uptodate(bh); 3359f2984462SChris Mason } else { 3360442a4f63SStefan Behrens struct btrfs_device *device = (struct btrfs_device *) 3361442a4f63SStefan Behrens bh->b_private; 3362442a4f63SStefan Behrens 3363fb456252SJeff Mahoney btrfs_warn_rl_in_rcu(device->fs_info, 3364b14af3b4SDavid Sterba "lost page write due to IO error on %s", 3365606686eeSJosef Bacik rcu_str_deref(device->name)); 336601327610SNicholas D Steeves /* note, we don't set_buffer_write_io_error because we have 33671259ab75SChris Mason * our own ways of dealing with the IO errors 33681259ab75SChris Mason */ 3369f2984462SChris Mason clear_buffer_uptodate(bh); 3370442a4f63SStefan Behrens btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS); 3371f2984462SChris Mason } 3372f2984462SChris Mason unlock_buffer(bh); 3373f2984462SChris Mason put_bh(bh); 3374f2984462SChris Mason } 3375f2984462SChris Mason 337629c36d72SAnand Jain int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num, 337729c36d72SAnand Jain struct buffer_head **bh_ret) 337829c36d72SAnand Jain { 337929c36d72SAnand Jain struct buffer_head *bh; 338029c36d72SAnand Jain struct btrfs_super_block *super; 338129c36d72SAnand Jain u64 bytenr; 338229c36d72SAnand Jain 338329c36d72SAnand Jain bytenr = btrfs_sb_offset(copy_num); 338429c36d72SAnand Jain if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode)) 338529c36d72SAnand Jain return -EINVAL; 338629c36d72SAnand Jain 33879f6d2510SDavid Sterba bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE); 338829c36d72SAnand Jain /* 338929c36d72SAnand Jain * If we fail to read from the underlying devices, as of now 339029c36d72SAnand Jain * the best option we have is to mark it EIO. 339129c36d72SAnand Jain */ 339229c36d72SAnand Jain if (!bh) 339329c36d72SAnand Jain return -EIO; 339429c36d72SAnand Jain 339529c36d72SAnand Jain super = (struct btrfs_super_block *)bh->b_data; 339629c36d72SAnand Jain if (btrfs_super_bytenr(super) != bytenr || 339729c36d72SAnand Jain btrfs_super_magic(super) != BTRFS_MAGIC) { 339829c36d72SAnand Jain brelse(bh); 339929c36d72SAnand Jain return -EINVAL; 340029c36d72SAnand Jain } 340129c36d72SAnand Jain 340229c36d72SAnand Jain *bh_ret = bh; 340329c36d72SAnand Jain return 0; 340429c36d72SAnand Jain } 340529c36d72SAnand Jain 340629c36d72SAnand Jain 3407a512bbf8SYan Zheng struct buffer_head *btrfs_read_dev_super(struct block_device *bdev) 3408a512bbf8SYan Zheng { 3409a512bbf8SYan Zheng struct buffer_head *bh; 3410a512bbf8SYan Zheng struct buffer_head *latest = NULL; 3411a512bbf8SYan Zheng struct btrfs_super_block *super; 3412a512bbf8SYan Zheng int i; 3413a512bbf8SYan Zheng u64 transid = 0; 341492fc03fbSAnand Jain int ret = -EINVAL; 3415a512bbf8SYan Zheng 3416a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 3417a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 3418a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 3419a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3420a512bbf8SYan Zheng */ 3421a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 342229c36d72SAnand Jain ret = btrfs_read_dev_one_super(bdev, i, &bh); 342329c36d72SAnand Jain if (ret) 3424a512bbf8SYan Zheng continue; 3425a512bbf8SYan Zheng 3426a512bbf8SYan Zheng super = (struct btrfs_super_block *)bh->b_data; 3427a512bbf8SYan Zheng 3428a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 3429a512bbf8SYan Zheng brelse(latest); 3430a512bbf8SYan Zheng latest = bh; 3431a512bbf8SYan Zheng transid = btrfs_super_generation(super); 3432a512bbf8SYan Zheng } else { 3433a512bbf8SYan Zheng brelse(bh); 3434a512bbf8SYan Zheng } 3435a512bbf8SYan Zheng } 343692fc03fbSAnand Jain 343792fc03fbSAnand Jain if (!latest) 343892fc03fbSAnand Jain return ERR_PTR(ret); 343992fc03fbSAnand Jain 3440a512bbf8SYan Zheng return latest; 3441a512bbf8SYan Zheng } 3442a512bbf8SYan Zheng 34434eedeb75SHisashi Hifumi /* 3444abbb3b8eSDavid Sterba * Write superblock @sb to the @device. Do not wait for completion, all the 3445abbb3b8eSDavid Sterba * buffer heads we write are pinned. 34464eedeb75SHisashi Hifumi * 3447abbb3b8eSDavid Sterba * Write @max_mirrors copies of the superblock, where 0 means default that fit 3448abbb3b8eSDavid Sterba * the expected device size at commit time. Note that max_mirrors must be 3449abbb3b8eSDavid Sterba * same for write and wait phases. 34504eedeb75SHisashi Hifumi * 3451abbb3b8eSDavid Sterba * Return number of errors when buffer head is not found or submission fails. 34524eedeb75SHisashi Hifumi */ 3453a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 3454abbb3b8eSDavid Sterba struct btrfs_super_block *sb, int max_mirrors) 3455a512bbf8SYan Zheng { 3456a512bbf8SYan Zheng struct buffer_head *bh; 3457a512bbf8SYan Zheng int i; 3458a512bbf8SYan Zheng int ret; 3459a512bbf8SYan Zheng int errors = 0; 3460a512bbf8SYan Zheng u32 crc; 3461a512bbf8SYan Zheng u64 bytenr; 34621b9e619cSOmar Sandoval int op_flags; 3463a512bbf8SYan Zheng 3464a512bbf8SYan Zheng if (max_mirrors == 0) 3465a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3466a512bbf8SYan Zheng 3467a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 3468a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 3469935e5cc9SMiao Xie if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3470935e5cc9SMiao Xie device->commit_total_bytes) 3471a512bbf8SYan Zheng break; 3472a512bbf8SYan Zheng 3473a512bbf8SYan Zheng btrfs_set_super_bytenr(sb, bytenr); 3474a512bbf8SYan Zheng 3475a512bbf8SYan Zheng crc = ~(u32)0; 3476abbb3b8eSDavid Sterba crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc, 3477abbb3b8eSDavid Sterba BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE); 3478a512bbf8SYan Zheng btrfs_csum_final(crc, sb->csum); 3479a512bbf8SYan Zheng 3480abbb3b8eSDavid Sterba /* One reference for us, and we leave it for the caller */ 34819f6d2510SDavid Sterba bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, 3482a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 3483634554dcSJosef Bacik if (!bh) { 3484fb456252SJeff Mahoney btrfs_err(device->fs_info, 3485f14d104dSDavid Sterba "couldn't get super buffer head for bytenr %llu", 3486f14d104dSDavid Sterba bytenr); 3487634554dcSJosef Bacik errors++; 3488634554dcSJosef Bacik continue; 3489634554dcSJosef Bacik } 3490634554dcSJosef Bacik 3491a512bbf8SYan Zheng memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE); 3492a512bbf8SYan Zheng 34934eedeb75SHisashi Hifumi /* one reference for submit_bh */ 3494a512bbf8SYan Zheng get_bh(bh); 34954eedeb75SHisashi Hifumi 34964eedeb75SHisashi Hifumi set_buffer_uptodate(bh); 3497a512bbf8SYan Zheng lock_buffer(bh); 3498a512bbf8SYan Zheng bh->b_end_io = btrfs_end_buffer_write_sync; 3499442a4f63SStefan Behrens bh->b_private = device; 3500a512bbf8SYan Zheng 3501387125fcSChris Mason /* 3502387125fcSChris Mason * we fua the first super. The others we allow 3503387125fcSChris Mason * to go down lazy. 3504387125fcSChris Mason */ 35051b9e619cSOmar Sandoval op_flags = REQ_SYNC | REQ_META | REQ_PRIO; 35061b9e619cSOmar Sandoval if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) 35071b9e619cSOmar Sandoval op_flags |= REQ_FUA; 35081b9e619cSOmar Sandoval ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh); 35094eedeb75SHisashi Hifumi if (ret) 3510a512bbf8SYan Zheng errors++; 3511a512bbf8SYan Zheng } 3512a512bbf8SYan Zheng return errors < i ? 0 : -1; 3513a512bbf8SYan Zheng } 3514a512bbf8SYan Zheng 3515387125fcSChris Mason /* 3516abbb3b8eSDavid Sterba * Wait for write completion of superblocks done by write_dev_supers, 3517abbb3b8eSDavid Sterba * @max_mirrors same for write and wait phases. 3518abbb3b8eSDavid Sterba * 3519abbb3b8eSDavid Sterba * Return number of errors when buffer head is not found or not marked up to 3520abbb3b8eSDavid Sterba * date. 3521abbb3b8eSDavid Sterba */ 3522abbb3b8eSDavid Sterba static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) 3523abbb3b8eSDavid Sterba { 3524abbb3b8eSDavid Sterba struct buffer_head *bh; 3525abbb3b8eSDavid Sterba int i; 3526abbb3b8eSDavid Sterba int errors = 0; 3527b6a535faSHoward McLauchlan bool primary_failed = false; 3528abbb3b8eSDavid Sterba u64 bytenr; 3529abbb3b8eSDavid Sterba 3530abbb3b8eSDavid Sterba if (max_mirrors == 0) 3531abbb3b8eSDavid Sterba max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3532abbb3b8eSDavid Sterba 3533abbb3b8eSDavid Sterba for (i = 0; i < max_mirrors; i++) { 3534abbb3b8eSDavid Sterba bytenr = btrfs_sb_offset(i); 3535abbb3b8eSDavid Sterba if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3536abbb3b8eSDavid Sterba device->commit_total_bytes) 3537abbb3b8eSDavid Sterba break; 3538abbb3b8eSDavid Sterba 35399f6d2510SDavid Sterba bh = __find_get_block(device->bdev, 35409f6d2510SDavid Sterba bytenr / BTRFS_BDEV_BLOCKSIZE, 3541abbb3b8eSDavid Sterba BTRFS_SUPER_INFO_SIZE); 3542abbb3b8eSDavid Sterba if (!bh) { 3543abbb3b8eSDavid Sterba errors++; 3544b6a535faSHoward McLauchlan if (i == 0) 3545b6a535faSHoward McLauchlan primary_failed = true; 3546abbb3b8eSDavid Sterba continue; 3547abbb3b8eSDavid Sterba } 3548abbb3b8eSDavid Sterba wait_on_buffer(bh); 3549b6a535faSHoward McLauchlan if (!buffer_uptodate(bh)) { 3550abbb3b8eSDavid Sterba errors++; 3551b6a535faSHoward McLauchlan if (i == 0) 3552b6a535faSHoward McLauchlan primary_failed = true; 3553b6a535faSHoward McLauchlan } 3554abbb3b8eSDavid Sterba 3555abbb3b8eSDavid Sterba /* drop our reference */ 3556abbb3b8eSDavid Sterba brelse(bh); 3557abbb3b8eSDavid Sterba 3558abbb3b8eSDavid Sterba /* drop the reference from the writing run */ 3559abbb3b8eSDavid Sterba brelse(bh); 3560abbb3b8eSDavid Sterba } 3561abbb3b8eSDavid Sterba 3562b6a535faSHoward McLauchlan /* log error, force error return */ 3563b6a535faSHoward McLauchlan if (primary_failed) { 3564b6a535faSHoward McLauchlan btrfs_err(device->fs_info, "error writing primary super block to device %llu", 3565b6a535faSHoward McLauchlan device->devid); 3566b6a535faSHoward McLauchlan return -1; 3567b6a535faSHoward McLauchlan } 3568b6a535faSHoward McLauchlan 3569abbb3b8eSDavid Sterba return errors < i ? 0 : -1; 3570abbb3b8eSDavid Sterba } 3571abbb3b8eSDavid Sterba 3572abbb3b8eSDavid Sterba /* 3573387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 3574387125fcSChris Mason * for the barrier when it is done 3575387125fcSChris Mason */ 35764246a0b6SChristoph Hellwig static void btrfs_end_empty_barrier(struct bio *bio) 3577387125fcSChris Mason { 3578387125fcSChris Mason complete(bio->bi_private); 3579387125fcSChris Mason } 3580387125fcSChris Mason 3581387125fcSChris Mason /* 35824fc6441aSAnand Jain * Submit a flush request to the device if it supports it. Error handling is 35834fc6441aSAnand Jain * done in the waiting counterpart. 3584387125fcSChris Mason */ 35854fc6441aSAnand Jain static void write_dev_flush(struct btrfs_device *device) 3586387125fcSChris Mason { 3587c2a9c7abSAnand Jain struct request_queue *q = bdev_get_queue(device->bdev); 3588e0ae9994SDavid Sterba struct bio *bio = device->flush_bio; 3589387125fcSChris Mason 3590c2a9c7abSAnand Jain if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) 35914fc6441aSAnand Jain return; 3592387125fcSChris Mason 3593e0ae9994SDavid Sterba bio_reset(bio); 3594387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 359574d46992SChristoph Hellwig bio_set_dev(bio, device->bdev); 35968d910125SJan Kara bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH; 3597387125fcSChris Mason init_completion(&device->flush_wait); 3598387125fcSChris Mason bio->bi_private = &device->flush_wait; 3599387125fcSChris Mason 360043a01111SLu Fengqi btrfsic_submit_bio(bio); 36011c3063b6SAnand Jain set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 36024fc6441aSAnand Jain } 3603387125fcSChris Mason 36044fc6441aSAnand Jain /* 36054fc6441aSAnand Jain * If the flush bio has been submitted by write_dev_flush, wait for it. 36064fc6441aSAnand Jain */ 36078c27cb35SLinus Torvalds static blk_status_t wait_dev_flush(struct btrfs_device *device) 36084fc6441aSAnand Jain { 36094fc6441aSAnand Jain struct bio *bio = device->flush_bio; 36104fc6441aSAnand Jain 36111c3063b6SAnand Jain if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) 361258efbc9fSOmar Sandoval return BLK_STS_OK; 36134fc6441aSAnand Jain 36141c3063b6SAnand Jain clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 36152980d574SDavid Sterba wait_for_completion_io(&device->flush_wait); 36164fc6441aSAnand Jain 36178c27cb35SLinus Torvalds return bio->bi_status; 3618387125fcSChris Mason } 3619387125fcSChris Mason 3620d10b82feSQu Wenruo static int check_barrier_error(struct btrfs_fs_info *fs_info) 3621401b41e5SAnand Jain { 36226528b99dSAnand Jain if (!btrfs_check_rw_degradable(fs_info, NULL)) 3623401b41e5SAnand Jain return -EIO; 3624387125fcSChris Mason return 0; 3625387125fcSChris Mason } 3626387125fcSChris Mason 3627387125fcSChris Mason /* 3628387125fcSChris Mason * send an empty flush down to each device in parallel, 3629387125fcSChris Mason * then wait for them 3630387125fcSChris Mason */ 3631387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 3632387125fcSChris Mason { 3633387125fcSChris Mason struct list_head *head; 3634387125fcSChris Mason struct btrfs_device *dev; 36355af3e8ccSStefan Behrens int errors_wait = 0; 36364e4cbee9SChristoph Hellwig blk_status_t ret; 3637387125fcSChris Mason 36381538e6c5SDavid Sterba lockdep_assert_held(&info->fs_devices->device_list_mutex); 3639387125fcSChris Mason /* send down all the barriers */ 3640387125fcSChris Mason head = &info->fs_devices->devices; 36411538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3642e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3643f88ba6a2SHidetoshi Seto continue; 3644cea7c8bfSAnand Jain if (!dev->bdev) 3645387125fcSChris Mason continue; 3646e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3647ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3648387125fcSChris Mason continue; 3649387125fcSChris Mason 36504fc6441aSAnand Jain write_dev_flush(dev); 365158efbc9fSOmar Sandoval dev->last_flush_error = BLK_STS_OK; 3652387125fcSChris Mason } 3653387125fcSChris Mason 3654387125fcSChris Mason /* wait for all the barriers */ 36551538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3656e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3657f88ba6a2SHidetoshi Seto continue; 3658387125fcSChris Mason if (!dev->bdev) { 36595af3e8ccSStefan Behrens errors_wait++; 3660387125fcSChris Mason continue; 3661387125fcSChris Mason } 3662e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3663ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3664387125fcSChris Mason continue; 3665387125fcSChris Mason 36664fc6441aSAnand Jain ret = wait_dev_flush(dev); 3667401b41e5SAnand Jain if (ret) { 3668401b41e5SAnand Jain dev->last_flush_error = ret; 366966b4993eSDavid Sterba btrfs_dev_stat_inc_and_print(dev, 367066b4993eSDavid Sterba BTRFS_DEV_STAT_FLUSH_ERRS); 36715af3e8ccSStefan Behrens errors_wait++; 3672387125fcSChris Mason } 3673401b41e5SAnand Jain } 3674401b41e5SAnand Jain 3675cea7c8bfSAnand Jain if (errors_wait) { 3676401b41e5SAnand Jain /* 3677401b41e5SAnand Jain * At some point we need the status of all disks 3678401b41e5SAnand Jain * to arrive at the volume status. So error checking 3679401b41e5SAnand Jain * is being pushed to a separate loop. 3680401b41e5SAnand Jain */ 3681d10b82feSQu Wenruo return check_barrier_error(info); 3682401b41e5SAnand Jain } 3683387125fcSChris Mason return 0; 3684387125fcSChris Mason } 3685387125fcSChris Mason 3686943c6e99SZhao Lei int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 3687943c6e99SZhao Lei { 36888789f4feSZhao Lei int raid_type; 36898789f4feSZhao Lei int min_tolerated = INT_MAX; 3690943c6e99SZhao Lei 36918789f4feSZhao Lei if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 36928789f4feSZhao Lei (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 36938789f4feSZhao Lei min_tolerated = min(min_tolerated, 36948789f4feSZhao Lei btrfs_raid_array[BTRFS_RAID_SINGLE]. 36958789f4feSZhao Lei tolerated_failures); 3696943c6e99SZhao Lei 36978789f4feSZhao Lei for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 36988789f4feSZhao Lei if (raid_type == BTRFS_RAID_SINGLE) 36998789f4feSZhao Lei continue; 370041a6e891SAnand Jain if (!(flags & btrfs_raid_array[raid_type].bg_flag)) 37018789f4feSZhao Lei continue; 37028789f4feSZhao Lei min_tolerated = min(min_tolerated, 37038789f4feSZhao Lei btrfs_raid_array[raid_type]. 37048789f4feSZhao Lei tolerated_failures); 37058789f4feSZhao Lei } 3706943c6e99SZhao Lei 37078789f4feSZhao Lei if (min_tolerated == INT_MAX) { 3708ab8d0fc4SJeff Mahoney pr_warn("BTRFS: unknown raid flag: %llu", flags); 37098789f4feSZhao Lei min_tolerated = 0; 37108789f4feSZhao Lei } 37118789f4feSZhao Lei 37128789f4feSZhao Lei return min_tolerated; 3713943c6e99SZhao Lei } 3714943c6e99SZhao Lei 3715eece6a9cSDavid Sterba int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) 3716f2984462SChris Mason { 3717e5e9a520SChris Mason struct list_head *head; 3718f2984462SChris Mason struct btrfs_device *dev; 3719a061fc8dSChris Mason struct btrfs_super_block *sb; 3720f2984462SChris Mason struct btrfs_dev_item *dev_item; 3721f2984462SChris Mason int ret; 3722f2984462SChris Mason int do_barriers; 3723a236aed1SChris Mason int max_errors; 3724a236aed1SChris Mason int total_errors = 0; 3725a061fc8dSChris Mason u64 flags; 3726f2984462SChris Mason 37270b246afaSJeff Mahoney do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); 3728fed3b381SLiu Bo 3729fed3b381SLiu Bo /* 3730fed3b381SLiu Bo * max_mirrors == 0 indicates we're from commit_transaction, 3731fed3b381SLiu Bo * not from fsync where the tree roots in fs_info have not 3732fed3b381SLiu Bo * been consistent on disk. 3733fed3b381SLiu Bo */ 3734fed3b381SLiu Bo if (max_mirrors == 0) 37350b246afaSJeff Mahoney backup_super_roots(fs_info); 3736f2984462SChris Mason 37370b246afaSJeff Mahoney sb = fs_info->super_for_commit; 3738a061fc8dSChris Mason dev_item = &sb->dev_item; 3739e5e9a520SChris Mason 37400b246afaSJeff Mahoney mutex_lock(&fs_info->fs_devices->device_list_mutex); 37410b246afaSJeff Mahoney head = &fs_info->fs_devices->devices; 37420b246afaSJeff Mahoney max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; 3743387125fcSChris Mason 37445af3e8ccSStefan Behrens if (do_barriers) { 37450b246afaSJeff Mahoney ret = barrier_all_devices(fs_info); 37465af3e8ccSStefan Behrens if (ret) { 37475af3e8ccSStefan Behrens mutex_unlock( 37480b246afaSJeff Mahoney &fs_info->fs_devices->device_list_mutex); 37490b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 37505af3e8ccSStefan Behrens "errors while submitting device barriers."); 37515af3e8ccSStefan Behrens return ret; 37525af3e8ccSStefan Behrens } 37535af3e8ccSStefan Behrens } 3754387125fcSChris Mason 37551538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3756dfe25020SChris Mason if (!dev->bdev) { 3757dfe25020SChris Mason total_errors++; 3758dfe25020SChris Mason continue; 3759dfe25020SChris Mason } 3760e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3761ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3762dfe25020SChris Mason continue; 3763dfe25020SChris Mason 37642b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 3765a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 3766a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 37677df69d3eSMiao Xie btrfs_set_stack_device_total_bytes(dev_item, 3768935e5cc9SMiao Xie dev->commit_total_bytes); 3769ce7213c7SMiao Xie btrfs_set_stack_device_bytes_used(dev_item, 3770ce7213c7SMiao Xie dev->commit_bytes_used); 3771a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 3772a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 3773a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 3774a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 37757239ff4bSNikolay Borisov memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, 37767239ff4bSNikolay Borisov BTRFS_FSID_SIZE); 3777a512bbf8SYan Zheng 3778a061fc8dSChris Mason flags = btrfs_super_flags(sb); 3779a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 3780f2984462SChris Mason 378175cb857dSQu Wenruo ret = btrfs_validate_write_super(fs_info, sb); 378275cb857dSQu Wenruo if (ret < 0) { 378375cb857dSQu Wenruo mutex_unlock(&fs_info->fs_devices->device_list_mutex); 378475cb857dSQu Wenruo btrfs_handle_fs_error(fs_info, -EUCLEAN, 378575cb857dSQu Wenruo "unexpected superblock corruption detected"); 378675cb857dSQu Wenruo return -EUCLEAN; 378775cb857dSQu Wenruo } 378875cb857dSQu Wenruo 3789abbb3b8eSDavid Sterba ret = write_dev_supers(dev, sb, max_mirrors); 3790a236aed1SChris Mason if (ret) 3791a236aed1SChris Mason total_errors++; 3792f2984462SChris Mason } 3793a236aed1SChris Mason if (total_errors > max_errors) { 37940b246afaSJeff Mahoney btrfs_err(fs_info, "%d errors while writing supers", 3795d397712bSChris Mason total_errors); 37960b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 379779787eaaSJeff Mahoney 37989d565ba4SStefan Behrens /* FUA is masked off if unsupported and can't be the reason */ 37990b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 38000b246afaSJeff Mahoney "%d errors while writing supers", 38010b246afaSJeff Mahoney total_errors); 38029d565ba4SStefan Behrens return -EIO; 3803a236aed1SChris Mason } 3804f2984462SChris Mason 3805a512bbf8SYan Zheng total_errors = 0; 38061538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3807dfe25020SChris Mason if (!dev->bdev) 3808dfe25020SChris Mason continue; 3809e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3810ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3811dfe25020SChris Mason continue; 3812dfe25020SChris Mason 3813abbb3b8eSDavid Sterba ret = wait_dev_supers(dev, max_mirrors); 3814a512bbf8SYan Zheng if (ret) 38151259ab75SChris Mason total_errors++; 3816f2984462SChris Mason } 38170b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 3818a236aed1SChris Mason if (total_errors > max_errors) { 38190b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 38200b246afaSJeff Mahoney "%d errors while writing supers", 38210b246afaSJeff Mahoney total_errors); 382279787eaaSJeff Mahoney return -EIO; 3823a236aed1SChris Mason } 3824f2984462SChris Mason return 0; 3825f2984462SChris Mason } 3826f2984462SChris Mason 3827cb517eabSMiao Xie /* Drop a fs root from the radix tree and free it. */ 3828cb517eabSMiao Xie void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 3829cb517eabSMiao Xie struct btrfs_root *root) 38302619ba1fSChris Mason { 38314df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 38322619ba1fSChris Mason radix_tree_delete(&fs_info->fs_roots_radix, 38332619ba1fSChris Mason (unsigned long)root->root_key.objectid); 38344df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 383576dda93cSYan, Zheng 383676dda93cSYan, Zheng if (btrfs_root_refs(&root->root_item) == 0) 383776dda93cSYan, Zheng synchronize_srcu(&fs_info->subvol_srcu); 383876dda93cSYan, Zheng 38391c1ea4f7SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 38403321719eSLiu Bo btrfs_free_log(NULL, root); 38411c1ea4f7SLiu Bo if (root->reloc_root) { 38421c1ea4f7SLiu Bo free_extent_buffer(root->reloc_root->node); 38431c1ea4f7SLiu Bo free_extent_buffer(root->reloc_root->commit_root); 38441c1ea4f7SLiu Bo btrfs_put_fs_root(root->reloc_root); 38451c1ea4f7SLiu Bo root->reloc_root = NULL; 38461c1ea4f7SLiu Bo } 38471c1ea4f7SLiu Bo } 38483321719eSLiu Bo 3849faa2dbf0SJosef Bacik if (root->free_ino_pinned) 3850581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_pinned); 3851faa2dbf0SJosef Bacik if (root->free_ino_ctl) 3852581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_ctl); 385384db5ccfSDavid Sterba btrfs_free_fs_root(root); 38544df27c4dSYan, Zheng } 38554df27c4dSYan, Zheng 385684db5ccfSDavid Sterba void btrfs_free_fs_root(struct btrfs_root *root) 38574df27c4dSYan, Zheng { 385857cdc8dbSDavid Sterba iput(root->ino_cache_inode); 38594df27c4dSYan, Zheng WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 38600ee5dc67SAl Viro if (root->anon_dev) 38610ee5dc67SAl Viro free_anon_bdev(root->anon_dev); 38628257b2dcSMiao Xie if (root->subv_writers) 38638257b2dcSMiao Xie btrfs_free_subvolume_writers(root->subv_writers); 38645f39d397SChris Mason free_extent_buffer(root->node); 38655f39d397SChris Mason free_extent_buffer(root->commit_root); 3866581bb050SLi Zefan kfree(root->free_ino_ctl); 3867581bb050SLi Zefan kfree(root->free_ino_pinned); 3868b0feb9d9SMiao Xie btrfs_put_fs_root(root); 38692619ba1fSChris Mason } 38702619ba1fSChris Mason 3871c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 3872c146afadSYan Zheng { 3873c146afadSYan Zheng u64 root_objectid = 0; 3874c146afadSYan Zheng struct btrfs_root *gang[8]; 387565d33fd7SQu Wenruo int i = 0; 387665d33fd7SQu Wenruo int err = 0; 387765d33fd7SQu Wenruo unsigned int ret = 0; 387865d33fd7SQu Wenruo int index; 3879c146afadSYan Zheng 3880c146afadSYan Zheng while (1) { 388165d33fd7SQu Wenruo index = srcu_read_lock(&fs_info->subvol_srcu); 3882c146afadSYan Zheng ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 3883c146afadSYan Zheng (void **)gang, root_objectid, 3884c146afadSYan Zheng ARRAY_SIZE(gang)); 388565d33fd7SQu Wenruo if (!ret) { 388665d33fd7SQu Wenruo srcu_read_unlock(&fs_info->subvol_srcu, index); 3887c146afadSYan Zheng break; 388865d33fd7SQu Wenruo } 38895d4f98a2SYan Zheng root_objectid = gang[ret - 1]->root_key.objectid + 1; 389066b4ffd1SJosef Bacik 389165d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 389265d33fd7SQu Wenruo /* Avoid to grab roots in dead_roots */ 389365d33fd7SQu Wenruo if (btrfs_root_refs(&gang[i]->root_item) == 0) { 389465d33fd7SQu Wenruo gang[i] = NULL; 389565d33fd7SQu Wenruo continue; 389665d33fd7SQu Wenruo } 389765d33fd7SQu Wenruo /* grab all the search result for later use */ 389865d33fd7SQu Wenruo gang[i] = btrfs_grab_fs_root(gang[i]); 389965d33fd7SQu Wenruo } 390065d33fd7SQu Wenruo srcu_read_unlock(&fs_info->subvol_srcu, index); 390165d33fd7SQu Wenruo 390265d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 390365d33fd7SQu Wenruo if (!gang[i]) 390465d33fd7SQu Wenruo continue; 3905c146afadSYan Zheng root_objectid = gang[i]->root_key.objectid; 390666b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(gang[i]); 390766b4ffd1SJosef Bacik if (err) 390865d33fd7SQu Wenruo break; 390965d33fd7SQu Wenruo btrfs_put_fs_root(gang[i]); 3910c146afadSYan Zheng } 3911c146afadSYan Zheng root_objectid++; 3912c146afadSYan Zheng } 391365d33fd7SQu Wenruo 391465d33fd7SQu Wenruo /* release the uncleaned roots due to error */ 391565d33fd7SQu Wenruo for (; i < ret; i++) { 391665d33fd7SQu Wenruo if (gang[i]) 391765d33fd7SQu Wenruo btrfs_put_fs_root(gang[i]); 391865d33fd7SQu Wenruo } 391965d33fd7SQu Wenruo return err; 3920c146afadSYan Zheng } 3921c146afadSYan Zheng 39226bccf3abSJeff Mahoney int btrfs_commit_super(struct btrfs_fs_info *fs_info) 3923c146afadSYan Zheng { 39246bccf3abSJeff Mahoney struct btrfs_root *root = fs_info->tree_root; 3925c146afadSYan Zheng struct btrfs_trans_handle *trans; 3926c146afadSYan Zheng 39270b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 39282ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 39290b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 39300b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 3931c71bf099SYan, Zheng 3932c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 39330b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 39340b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 3935c71bf099SYan, Zheng 39367a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 39373612b495STsutomu Itoh if (IS_ERR(trans)) 39383612b495STsutomu Itoh return PTR_ERR(trans); 39393a45bb20SJeff Mahoney return btrfs_commit_transaction(trans); 3940c146afadSYan Zheng } 3941c146afadSYan Zheng 39426bccf3abSJeff Mahoney void close_ctree(struct btrfs_fs_info *fs_info) 3943eb60ceacSChris Mason { 3944c146afadSYan Zheng int ret; 3945e089f05cSChris Mason 3946afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); 3947d6fd0ae2SOmar Sandoval /* 3948d6fd0ae2SOmar Sandoval * We don't want the cleaner to start new transactions, add more delayed 3949d6fd0ae2SOmar Sandoval * iputs, etc. while we're closing. We can't use kthread_stop() yet 3950d6fd0ae2SOmar Sandoval * because that frees the task_struct, and the transaction kthread might 3951d6fd0ae2SOmar Sandoval * still try to wake up the cleaner. 3952d6fd0ae2SOmar Sandoval */ 3953d6fd0ae2SOmar Sandoval kthread_park(fs_info->cleaner_kthread); 3954a2135011SChris Mason 39557343dd61SJustin Maggard /* wait for the qgroup rescan worker to stop */ 3956d06f23d6SJeff Mahoney btrfs_qgroup_wait_for_completion(fs_info, false); 39577343dd61SJustin Maggard 3958803b2f54SStefan Behrens /* wait for the uuid_scan task to finish */ 3959803b2f54SStefan Behrens down(&fs_info->uuid_tree_rescan_sem); 3960803b2f54SStefan Behrens /* avoid complains from lockdep et al., set sem back to initial state */ 3961803b2f54SStefan Behrens up(&fs_info->uuid_tree_rescan_sem); 3962803b2f54SStefan Behrens 3963837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 3964aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 3965837d5b6eSIlya Dryomov 39668dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 39678dabb742SStefan Behrens 3968aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 39694cb5300bSChris Mason 39704cb5300bSChris Mason /* wait for any defraggers to finish */ 39714cb5300bSChris Mason wait_event(fs_info->transaction_wait, 39724cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 39734cb5300bSChris Mason 39744cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 397526176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 39764cb5300bSChris Mason 397721c7e756SMiao Xie cancel_work_sync(&fs_info->async_reclaim_work); 397821c7e756SMiao Xie 3979bc98a42cSDavid Howells if (!sb_rdonly(fs_info->sb)) { 3980e44163e1SJeff Mahoney /* 3981d6fd0ae2SOmar Sandoval * The cleaner kthread is stopped, so do one final pass over 3982d6fd0ae2SOmar Sandoval * unused block groups. 3983e44163e1SJeff Mahoney */ 39840b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 3985e44163e1SJeff Mahoney 39866bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 3987d397712bSChris Mason if (ret) 398804892340SEric Sandeen btrfs_err(fs_info, "commit super ret %d", ret); 3989c146afadSYan Zheng } 3990ed2ff2cbSChris Mason 3991af722733SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) || 3992af722733SLiu Bo test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) 39932ff7e61eSJeff Mahoney btrfs_error_commit_super(fs_info); 3994acce952bSliubo 3995e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 3996e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 39978929ecfaSYan, Zheng 3998e187831eSJosef Bacik ASSERT(list_empty(&fs_info->delayed_iputs)); 3999afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); 4000f25784b3SYan Zheng 400104892340SEric Sandeen btrfs_free_qgroup_config(fs_info); 4002fe816d0fSNikolay Borisov ASSERT(list_empty(&fs_info->delalloc_roots)); 4003bcef60f2SArne Jansen 4004963d678bSMiao Xie if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 400504892340SEric Sandeen btrfs_info(fs_info, "at unmount delalloc count %lld", 4006963d678bSMiao Xie percpu_counter_sum(&fs_info->delalloc_bytes)); 4007b0c68f8bSChris Mason } 400831153d81SYan Zheng 40096618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 4010b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 40115ac1d209SJeff Mahoney 4012faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 4013d10c5f31SChris Mason 40141a4319ccSLiu Bo btrfs_put_block_group_cache(fs_info); 40151a4319ccSLiu Bo 4016de348ee0SWang Shilong /* 4017de348ee0SWang Shilong * we must make sure there is not any read request to 4018de348ee0SWang Shilong * submit after we stopping all workers. 4019de348ee0SWang Shilong */ 4020de348ee0SWang Shilong invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 402196192499SJosef Bacik btrfs_stop_all_workers(fs_info); 402296192499SJosef Bacik 40235cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 40245cdd7db6SFilipe Manana 4025afcdd129SJosef Bacik clear_bit(BTRFS_FS_OPEN, &fs_info->flags); 402613e6c37bSJosef Bacik free_root_pointers(fs_info, 1); 40279ad6b7bcSChris Mason 402813e6c37bSJosef Bacik iput(fs_info->btree_inode); 4029d6bfde87SChris Mason 403021adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 40310b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) 40322ff7e61eSJeff Mahoney btrfsic_unmount(fs_info->fs_devices); 403321adbd5cSStefan Behrens #endif 403421adbd5cSStefan Behrens 4035dfe25020SChris Mason btrfs_close_devices(fs_info->fs_devices); 40360b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 4037b248a415SChris Mason 4038e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 4039963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 40407f8d236aSDavid Sterba percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 404176dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 40420b86a832SChris Mason 404353b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 4044fd708b81SJosef Bacik btrfs_free_ref_cache(fs_info); 404553b381b3SDavid Woodhouse 404604216820SFilipe Manana while (!list_empty(&fs_info->pinned_chunks)) { 404704216820SFilipe Manana struct extent_map *em; 404804216820SFilipe Manana 404904216820SFilipe Manana em = list_first_entry(&fs_info->pinned_chunks, 405004216820SFilipe Manana struct extent_map, list); 405104216820SFilipe Manana list_del_init(&em->list); 405204216820SFilipe Manana free_extent_map(em); 405304216820SFilipe Manana } 4054eb60ceacSChris Mason } 4055eb60ceacSChris Mason 4056b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 4057b9fab919SChris Mason int atomic) 4058ccd467d6SChris Mason { 40591259ab75SChris Mason int ret; 4060727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 40611259ab75SChris Mason 40620b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 40631259ab75SChris Mason if (!ret) 40641259ab75SChris Mason return ret; 40651259ab75SChris Mason 40661259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 4067b9fab919SChris Mason parent_transid, atomic); 4068b9fab919SChris Mason if (ret == -EAGAIN) 4069b9fab919SChris Mason return ret; 40701259ab75SChris Mason return !ret; 40715f39d397SChris Mason } 40726702ed49SChris Mason 40735f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 40745f39d397SChris Mason { 40750b246afaSJeff Mahoney struct btrfs_fs_info *fs_info; 407606ea65a3SJosef Bacik struct btrfs_root *root; 40775f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 4078b9473439SChris Mason int was_dirty; 4079b4ce94deSChris Mason 408006ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 408106ea65a3SJosef Bacik /* 408206ea65a3SJosef Bacik * This is a fast path so only do this check if we have sanity tests 408352042d8eSAndrea Gelmini * enabled. Normal people shouldn't be using unmapped buffers as dirty 408406ea65a3SJosef Bacik * outside of the sanity tests. 408506ea65a3SJosef Bacik */ 4086b0132a3bSNikolay Borisov if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) 408706ea65a3SJosef Bacik return; 408806ea65a3SJosef Bacik #endif 408906ea65a3SJosef Bacik root = BTRFS_I(buf->pages[0]->mapping->host)->root; 40900b246afaSJeff Mahoney fs_info = root->fs_info; 4091b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 40920b246afaSJeff Mahoney if (transid != fs_info->generation) 40935d163e0eSJeff Mahoney WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", 40940b246afaSJeff Mahoney buf->start, transid, fs_info->generation); 40950b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 4096e2d84521SMiao Xie if (!was_dirty) 4097104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 4098e2d84521SMiao Xie buf->len, 40990b246afaSJeff Mahoney fs_info->dirty_metadata_batch); 41001f21ef0aSFilipe Manana #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 410169fc6cbbSQu Wenruo /* 410269fc6cbbSQu Wenruo * Since btrfs_mark_buffer_dirty() can be called with item pointer set 410369fc6cbbSQu Wenruo * but item data not updated. 410469fc6cbbSQu Wenruo * So here we should only check item pointers, not item data. 410569fc6cbbSQu Wenruo */ 410669fc6cbbSQu Wenruo if (btrfs_header_level(buf) == 0 && 41072f659546SQu Wenruo btrfs_check_leaf_relaxed(fs_info, buf)) { 4108a4f78750SDavid Sterba btrfs_print_leaf(buf); 41091f21ef0aSFilipe Manana ASSERT(0); 41101f21ef0aSFilipe Manana } 41111f21ef0aSFilipe Manana #endif 4112eb60ceacSChris Mason } 4113eb60ceacSChris Mason 41142ff7e61eSJeff Mahoney static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, 4115b53d3f5dSLiu Bo int flush_delayed) 411635b7e476SChris Mason { 4117188de649SChris Mason /* 4118188de649SChris Mason * looks as though older kernels can get into trouble with 4119188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 4120188de649SChris Mason */ 4121e2d84521SMiao Xie int ret; 4122d6bfde87SChris Mason 41236933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 4124d6bfde87SChris Mason return; 4125d6bfde87SChris Mason 4126b53d3f5dSLiu Bo if (flush_delayed) 41272ff7e61eSJeff Mahoney btrfs_balance_delayed_items(fs_info); 412816cdcec7SMiao Xie 4129d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 4130d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 4131d814a491SEthan Lien fs_info->dirty_metadata_batch); 4132e2d84521SMiao Xie if (ret > 0) { 41330b246afaSJeff Mahoney balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); 413416cdcec7SMiao Xie } 413516cdcec7SMiao Xie } 413616cdcec7SMiao Xie 41372ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) 413816cdcec7SMiao Xie { 41392ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 1); 414035b7e476SChris Mason } 4141b53d3f5dSLiu Bo 41422ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) 4143b53d3f5dSLiu Bo { 41442ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 0); 4145d6bfde87SChris Mason } 41466b80053dSChris Mason 4147581c1760SQu Wenruo int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level, 4148581c1760SQu Wenruo struct btrfs_key *first_key) 41496b80053dSChris Mason { 4150727011e0SChris Mason struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root; 41512ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 41522ff7e61eSJeff Mahoney 4153581c1760SQu Wenruo return btree_read_extent_buffer_pages(fs_info, buf, parent_transid, 4154581c1760SQu Wenruo level, first_key); 41556b80053dSChris Mason } 41560da5468fSChris Mason 41572ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) 4158acce952bSliubo { 4159fe816d0fSNikolay Borisov /* cleanup FS via transaction */ 4160fe816d0fSNikolay Borisov btrfs_cleanup_transaction(fs_info); 4161fe816d0fSNikolay Borisov 41620b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 41632ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 41640b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 4165acce952bSliubo 41660b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 41670b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4168acce952bSliubo } 4169acce952bSliubo 4170143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4171acce952bSliubo { 4172acce952bSliubo struct btrfs_ordered_extent *ordered; 4173acce952bSliubo 4174199c2a9cSMiao Xie spin_lock(&root->ordered_extent_lock); 4175779880efSJosef Bacik /* 4176779880efSJosef Bacik * This will just short circuit the ordered completion stuff which will 4177779880efSJosef Bacik * make sure the ordered extent gets properly cleaned up. 4178779880efSJosef Bacik */ 4179199c2a9cSMiao Xie list_for_each_entry(ordered, &root->ordered_extents, 4180779880efSJosef Bacik root_extent_list) 4181779880efSJosef Bacik set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4182199c2a9cSMiao Xie spin_unlock(&root->ordered_extent_lock); 4183199c2a9cSMiao Xie } 4184199c2a9cSMiao Xie 4185199c2a9cSMiao Xie static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4186199c2a9cSMiao Xie { 4187199c2a9cSMiao Xie struct btrfs_root *root; 4188199c2a9cSMiao Xie struct list_head splice; 4189199c2a9cSMiao Xie 4190199c2a9cSMiao Xie INIT_LIST_HEAD(&splice); 4191199c2a9cSMiao Xie 4192199c2a9cSMiao Xie spin_lock(&fs_info->ordered_root_lock); 4193199c2a9cSMiao Xie list_splice_init(&fs_info->ordered_roots, &splice); 4194199c2a9cSMiao Xie while (!list_empty(&splice)) { 4195199c2a9cSMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4196199c2a9cSMiao Xie ordered_root); 41971de2cfdeSJosef Bacik list_move_tail(&root->ordered_root, 41981de2cfdeSJosef Bacik &fs_info->ordered_roots); 4199199c2a9cSMiao Xie 42002a85d9caSLiu Bo spin_unlock(&fs_info->ordered_root_lock); 4201199c2a9cSMiao Xie btrfs_destroy_ordered_extents(root); 4202199c2a9cSMiao Xie 42032a85d9caSLiu Bo cond_resched(); 42042a85d9caSLiu Bo spin_lock(&fs_info->ordered_root_lock); 4205199c2a9cSMiao Xie } 4206199c2a9cSMiao Xie spin_unlock(&fs_info->ordered_root_lock); 420774d5d229SJosef Bacik 420874d5d229SJosef Bacik /* 420974d5d229SJosef Bacik * We need this here because if we've been flipped read-only we won't 421074d5d229SJosef Bacik * get sync() from the umount, so we need to make sure any ordered 421174d5d229SJosef Bacik * extents that haven't had their dirty pages IO start writeout yet 421274d5d229SJosef Bacik * actually get run and error out properly. 421374d5d229SJosef Bacik */ 421474d5d229SJosef Bacik btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 4215acce952bSliubo } 4216acce952bSliubo 421735a3621bSStefan Behrens static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 42182ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4219acce952bSliubo { 4220acce952bSliubo struct rb_node *node; 4221acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 4222acce952bSliubo struct btrfs_delayed_ref_node *ref; 4223acce952bSliubo int ret = 0; 4224acce952bSliubo 4225acce952bSliubo delayed_refs = &trans->delayed_refs; 4226acce952bSliubo 4227acce952bSliubo spin_lock(&delayed_refs->lock); 4228d7df2c79SJosef Bacik if (atomic_read(&delayed_refs->num_entries) == 0) { 4229cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 42300b246afaSJeff Mahoney btrfs_info(fs_info, "delayed_refs has NO entry"); 4231acce952bSliubo return ret; 4232acce952bSliubo } 4233acce952bSliubo 42345c9d028bSLiu Bo while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { 4235d7df2c79SJosef Bacik struct btrfs_delayed_ref_head *head; 42360e0adbcfSJosef Bacik struct rb_node *n; 4237e78417d1SJosef Bacik bool pin_bytes = false; 4238acce952bSliubo 4239d7df2c79SJosef Bacik head = rb_entry(node, struct btrfs_delayed_ref_head, 4240d7df2c79SJosef Bacik href_node); 4241b939d1abSJosef Bacik if (!mutex_trylock(&head->mutex)) { 4242d278850eSJosef Bacik refcount_inc(&head->refs); 424379787eaaSJeff Mahoney spin_unlock(&delayed_refs->lock); 4244b939d1abSJosef Bacik 4245acce952bSliubo mutex_lock(&head->mutex); 4246b939d1abSJosef Bacik mutex_unlock(&head->mutex); 4247d278850eSJosef Bacik btrfs_put_delayed_ref_head(head); 4248e18fca73SJosef Bacik spin_lock(&delayed_refs->lock); 4249b939d1abSJosef Bacik continue; 4250b939d1abSJosef Bacik } 4251d7df2c79SJosef Bacik spin_lock(&head->lock); 4252e3d03965SLiu Bo while ((n = rb_first_cached(&head->ref_tree)) != NULL) { 42530e0adbcfSJosef Bacik ref = rb_entry(n, struct btrfs_delayed_ref_node, 42540e0adbcfSJosef Bacik ref_node); 4255d7df2c79SJosef Bacik ref->in_tree = 0; 4256e3d03965SLiu Bo rb_erase_cached(&ref->ref_node, &head->ref_tree); 42570e0adbcfSJosef Bacik RB_CLEAR_NODE(&ref->ref_node); 42581d57ee94SWang Xiaoguang if (!list_empty(&ref->add_list)) 42591d57ee94SWang Xiaoguang list_del(&ref->add_list); 4260d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4261d7df2c79SJosef Bacik btrfs_put_delayed_ref(ref); 4262d7df2c79SJosef Bacik } 426354067ae9SJosef Bacik if (head->must_insert_reserved) 4264e78417d1SJosef Bacik pin_bytes = true; 426578a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 4266acce952bSliubo delayed_refs->num_heads--; 4267d7df2c79SJosef Bacik if (head->processing == 0) 4268acce952bSliubo delayed_refs->num_heads_ready--; 4269d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 42705c9d028bSLiu Bo rb_erase_cached(&head->href_node, &delayed_refs->href_root); 4271d278850eSJosef Bacik RB_CLEAR_NODE(&head->href_node); 4272d7df2c79SJosef Bacik spin_unlock(&head->lock); 4273acce952bSliubo spin_unlock(&delayed_refs->lock); 4274e78417d1SJosef Bacik mutex_unlock(&head->mutex); 4275acce952bSliubo 4276d7df2c79SJosef Bacik if (pin_bytes) 4277d278850eSJosef Bacik btrfs_pin_extent(fs_info, head->bytenr, 4278d278850eSJosef Bacik head->num_bytes, 1); 427931890da0SJosef Bacik btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); 4280d278850eSJosef Bacik btrfs_put_delayed_ref_head(head); 4281acce952bSliubo cond_resched(); 4282acce952bSliubo spin_lock(&delayed_refs->lock); 4283acce952bSliubo } 4284acce952bSliubo 4285acce952bSliubo spin_unlock(&delayed_refs->lock); 4286acce952bSliubo 4287acce952bSliubo return ret; 4288acce952bSliubo } 4289acce952bSliubo 4290143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 4291acce952bSliubo { 4292acce952bSliubo struct btrfs_inode *btrfs_inode; 4293acce952bSliubo struct list_head splice; 4294acce952bSliubo 4295acce952bSliubo INIT_LIST_HEAD(&splice); 4296acce952bSliubo 4297eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4298eb73c1b7SMiao Xie list_splice_init(&root->delalloc_inodes, &splice); 4299acce952bSliubo 4300acce952bSliubo while (!list_empty(&splice)) { 4301fe816d0fSNikolay Borisov struct inode *inode = NULL; 4302eb73c1b7SMiao Xie btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 4303acce952bSliubo delalloc_inodes); 4304fe816d0fSNikolay Borisov __btrfs_del_delalloc_inode(root, btrfs_inode); 4305eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4306acce952bSliubo 4307fe816d0fSNikolay Borisov /* 4308fe816d0fSNikolay Borisov * Make sure we get a live inode and that it'll not disappear 4309fe816d0fSNikolay Borisov * meanwhile. 4310fe816d0fSNikolay Borisov */ 4311fe816d0fSNikolay Borisov inode = igrab(&btrfs_inode->vfs_inode); 4312fe816d0fSNikolay Borisov if (inode) { 4313fe816d0fSNikolay Borisov invalidate_inode_pages2(inode->i_mapping); 4314fe816d0fSNikolay Borisov iput(inode); 4315fe816d0fSNikolay Borisov } 4316eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4317acce952bSliubo } 4318eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4319eb73c1b7SMiao Xie } 4320eb73c1b7SMiao Xie 4321eb73c1b7SMiao Xie static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 4322eb73c1b7SMiao Xie { 4323eb73c1b7SMiao Xie struct btrfs_root *root; 4324eb73c1b7SMiao Xie struct list_head splice; 4325eb73c1b7SMiao Xie 4326eb73c1b7SMiao Xie INIT_LIST_HEAD(&splice); 4327eb73c1b7SMiao Xie 4328eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4329eb73c1b7SMiao Xie list_splice_init(&fs_info->delalloc_roots, &splice); 4330eb73c1b7SMiao Xie while (!list_empty(&splice)) { 4331eb73c1b7SMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4332eb73c1b7SMiao Xie delalloc_root); 4333eb73c1b7SMiao Xie root = btrfs_grab_fs_root(root); 4334eb73c1b7SMiao Xie BUG_ON(!root); 4335eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4336eb73c1b7SMiao Xie 4337eb73c1b7SMiao Xie btrfs_destroy_delalloc_inodes(root); 4338eb73c1b7SMiao Xie btrfs_put_fs_root(root); 4339eb73c1b7SMiao Xie 4340eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4341eb73c1b7SMiao Xie } 4342eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4343acce952bSliubo } 4344acce952bSliubo 43452ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 4346acce952bSliubo struct extent_io_tree *dirty_pages, 4347acce952bSliubo int mark) 4348acce952bSliubo { 4349acce952bSliubo int ret; 4350acce952bSliubo struct extent_buffer *eb; 4351acce952bSliubo u64 start = 0; 4352acce952bSliubo u64 end; 4353acce952bSliubo 4354acce952bSliubo while (1) { 4355acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 4356e6138876SJosef Bacik mark, NULL); 4357acce952bSliubo if (ret) 4358acce952bSliubo break; 4359acce952bSliubo 436091166212SDavid Sterba clear_extent_bits(dirty_pages, start, end, mark); 4361acce952bSliubo while (start <= end) { 43620b246afaSJeff Mahoney eb = find_extent_buffer(fs_info, start); 43630b246afaSJeff Mahoney start += fs_info->nodesize; 4364fd8b2b61SJosef Bacik if (!eb) 4365acce952bSliubo continue; 4366fd8b2b61SJosef Bacik wait_on_extent_buffer_writeback(eb); 4367acce952bSliubo 4368fd8b2b61SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 4369fd8b2b61SJosef Bacik &eb->bflags)) 4370fd8b2b61SJosef Bacik clear_extent_buffer_dirty(eb); 4371fd8b2b61SJosef Bacik free_extent_buffer_stale(eb); 4372acce952bSliubo } 4373acce952bSliubo } 4374acce952bSliubo 4375acce952bSliubo return ret; 4376acce952bSliubo } 4377acce952bSliubo 43782ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 4379acce952bSliubo struct extent_io_tree *pinned_extents) 4380acce952bSliubo { 4381acce952bSliubo struct extent_io_tree *unpin; 4382acce952bSliubo u64 start; 4383acce952bSliubo u64 end; 4384acce952bSliubo int ret; 4385ed0eaa14SLiu Bo bool loop = true; 4386acce952bSliubo 4387acce952bSliubo unpin = pinned_extents; 4388ed0eaa14SLiu Bo again: 4389acce952bSliubo while (1) { 43900e6ec385SFilipe Manana struct extent_state *cached_state = NULL; 43910e6ec385SFilipe Manana 4392fcd5e742SLu Fengqi /* 4393fcd5e742SLu Fengqi * The btrfs_finish_extent_commit() may get the same range as 4394fcd5e742SLu Fengqi * ours between find_first_extent_bit and clear_extent_dirty. 4395fcd5e742SLu Fengqi * Hence, hold the unused_bg_unpin_mutex to avoid double unpin 4396fcd5e742SLu Fengqi * the same extent range. 4397fcd5e742SLu Fengqi */ 4398fcd5e742SLu Fengqi mutex_lock(&fs_info->unused_bg_unpin_mutex); 4399acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 44000e6ec385SFilipe Manana EXTENT_DIRTY, &cached_state); 4401fcd5e742SLu Fengqi if (ret) { 4402fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4403acce952bSliubo break; 4404fcd5e742SLu Fengqi } 4405acce952bSliubo 44060e6ec385SFilipe Manana clear_extent_dirty(unpin, start, end, &cached_state); 44070e6ec385SFilipe Manana free_extent_state(cached_state); 44082ff7e61eSJeff Mahoney btrfs_error_unpin_extent_range(fs_info, start, end); 4409fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4410acce952bSliubo cond_resched(); 4411acce952bSliubo } 4412acce952bSliubo 4413ed0eaa14SLiu Bo if (loop) { 44140b246afaSJeff Mahoney if (unpin == &fs_info->freed_extents[0]) 44150b246afaSJeff Mahoney unpin = &fs_info->freed_extents[1]; 4416ed0eaa14SLiu Bo else 44170b246afaSJeff Mahoney unpin = &fs_info->freed_extents[0]; 4418ed0eaa14SLiu Bo loop = false; 4419ed0eaa14SLiu Bo goto again; 4420ed0eaa14SLiu Bo } 4421ed0eaa14SLiu Bo 4422acce952bSliubo return 0; 4423acce952bSliubo } 4424acce952bSliubo 4425c79a1751SLiu Bo static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache) 4426c79a1751SLiu Bo { 4427c79a1751SLiu Bo struct inode *inode; 4428c79a1751SLiu Bo 4429c79a1751SLiu Bo inode = cache->io_ctl.inode; 4430c79a1751SLiu Bo if (inode) { 4431c79a1751SLiu Bo invalidate_inode_pages2(inode->i_mapping); 4432c79a1751SLiu Bo BTRFS_I(inode)->generation = 0; 4433c79a1751SLiu Bo cache->io_ctl.inode = NULL; 4434c79a1751SLiu Bo iput(inode); 4435c79a1751SLiu Bo } 4436c79a1751SLiu Bo btrfs_put_block_group(cache); 4437c79a1751SLiu Bo } 4438c79a1751SLiu Bo 4439c79a1751SLiu Bo void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, 44402ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4441c79a1751SLiu Bo { 4442c79a1751SLiu Bo struct btrfs_block_group_cache *cache; 4443c79a1751SLiu Bo 4444c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4445c79a1751SLiu Bo while (!list_empty(&cur_trans->dirty_bgs)) { 4446c79a1751SLiu Bo cache = list_first_entry(&cur_trans->dirty_bgs, 4447c79a1751SLiu Bo struct btrfs_block_group_cache, 4448c79a1751SLiu Bo dirty_list); 4449c79a1751SLiu Bo 4450c79a1751SLiu Bo if (!list_empty(&cache->io_list)) { 4451c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4452c79a1751SLiu Bo list_del_init(&cache->io_list); 4453c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 4454c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4455c79a1751SLiu Bo } 4456c79a1751SLiu Bo 4457c79a1751SLiu Bo list_del_init(&cache->dirty_list); 4458c79a1751SLiu Bo spin_lock(&cache->lock); 4459c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 4460c79a1751SLiu Bo spin_unlock(&cache->lock); 4461c79a1751SLiu Bo 4462c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4463c79a1751SLiu Bo btrfs_put_block_group(cache); 4464ba2c4d4eSJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 4465c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4466c79a1751SLiu Bo } 4467c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4468c79a1751SLiu Bo 446945ae2c18SNikolay Borisov /* 447045ae2c18SNikolay Borisov * Refer to the definition of io_bgs member for details why it's safe 447145ae2c18SNikolay Borisov * to use it without any locking 447245ae2c18SNikolay Borisov */ 4473c79a1751SLiu Bo while (!list_empty(&cur_trans->io_bgs)) { 4474c79a1751SLiu Bo cache = list_first_entry(&cur_trans->io_bgs, 4475c79a1751SLiu Bo struct btrfs_block_group_cache, 4476c79a1751SLiu Bo io_list); 4477c79a1751SLiu Bo 4478c79a1751SLiu Bo list_del_init(&cache->io_list); 4479c79a1751SLiu Bo spin_lock(&cache->lock); 4480c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 4481c79a1751SLiu Bo spin_unlock(&cache->lock); 4482c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 4483c79a1751SLiu Bo } 4484c79a1751SLiu Bo } 4485c79a1751SLiu Bo 448649b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 44872ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 448849b25e05SJeff Mahoney { 44892ff7e61eSJeff Mahoney btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 4490c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->dirty_bgs)); 4491c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->io_bgs)); 4492c79a1751SLiu Bo 44932ff7e61eSJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, fs_info); 449449b25e05SJeff Mahoney 44954a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 44960b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 449749b25e05SJeff Mahoney 44984a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 44990b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 450049b25e05SJeff Mahoney 4501ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 4502ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 450367cde344SMiao Xie 45042ff7e61eSJeff Mahoney btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, 450549b25e05SJeff Mahoney EXTENT_DIRTY); 45062ff7e61eSJeff Mahoney btrfs_destroy_pinned_extent(fs_info, 45070b246afaSJeff Mahoney fs_info->pinned_extents); 450849b25e05SJeff Mahoney 45094a9d8bdeSMiao Xie cur_trans->state =TRANS_STATE_COMPLETED; 45104a9d8bdeSMiao Xie wake_up(&cur_trans->commit_wait); 451149b25e05SJeff Mahoney } 451249b25e05SJeff Mahoney 45132ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) 4514acce952bSliubo { 4515acce952bSliubo struct btrfs_transaction *t; 4516acce952bSliubo 45170b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 4518acce952bSliubo 45190b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 45200b246afaSJeff Mahoney while (!list_empty(&fs_info->trans_list)) { 45210b246afaSJeff Mahoney t = list_first_entry(&fs_info->trans_list, 4522724e2315SJosef Bacik struct btrfs_transaction, list); 4523724e2315SJosef Bacik if (t->state >= TRANS_STATE_COMMIT_START) { 45249b64f57dSElena Reshetova refcount_inc(&t->use_count); 45250b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 45262ff7e61eSJeff Mahoney btrfs_wait_for_commit(fs_info, t->transid); 4527724e2315SJosef Bacik btrfs_put_transaction(t); 45280b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4529724e2315SJosef Bacik continue; 4530724e2315SJosef Bacik } 45310b246afaSJeff Mahoney if (t == fs_info->running_transaction) { 4532724e2315SJosef Bacik t->state = TRANS_STATE_COMMIT_DOING; 45330b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4534724e2315SJosef Bacik /* 4535724e2315SJosef Bacik * We wait for 0 num_writers since we don't hold a trans 4536724e2315SJosef Bacik * handle open currently for this transaction. 4537724e2315SJosef Bacik */ 4538724e2315SJosef Bacik wait_event(t->writer_wait, 4539724e2315SJosef Bacik atomic_read(&t->num_writers) == 0); 4540724e2315SJosef Bacik } else { 45410b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4542724e2315SJosef Bacik } 45432ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(t, fs_info); 4544724e2315SJosef Bacik 45450b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 45460b246afaSJeff Mahoney if (t == fs_info->running_transaction) 45470b246afaSJeff Mahoney fs_info->running_transaction = NULL; 4548724e2315SJosef Bacik list_del_init(&t->list); 45490b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4550a4abeea4SJosef Bacik 4551724e2315SJosef Bacik btrfs_put_transaction(t); 45522ff7e61eSJeff Mahoney trace_btrfs_transaction_commit(fs_info->tree_root); 45530b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4554724e2315SJosef Bacik } 45550b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 45560b246afaSJeff Mahoney btrfs_destroy_all_ordered_extents(fs_info); 4557ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 4558ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 45592ff7e61eSJeff Mahoney btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents); 45600b246afaSJeff Mahoney btrfs_destroy_all_delalloc_inodes(fs_info); 45610b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 4562acce952bSliubo 4563acce952bSliubo return 0; 4564acce952bSliubo } 4565acce952bSliubo 4566e8c9f186SDavid Sterba static const struct extent_io_ops btree_extent_io_ops = { 45674d53dddbSDavid Sterba /* mandatory callbacks */ 45680b86a832SChris Mason .submit_bio_hook = btree_submit_bio_hook, 45694d53dddbSDavid Sterba .readpage_end_io_hook = btree_readpage_end_io_hook, 45700da5468fSChris Mason }; 4571