1c1d7c514SDavid Sterba // SPDX-License-Identifier: GPL-2.0 26cbd5570SChris Mason /* 36cbd5570SChris Mason * Copyright (C) 2007 Oracle. All rights reserved. 46cbd5570SChris Mason */ 56cbd5570SChris Mason 6e20d96d6SChris Mason #include <linux/fs.h> 7d98237b3SChris Mason #include <linux/blkdev.h> 80f7d52f4SChris Mason #include <linux/radix-tree.h> 935b7e476SChris Mason #include <linux/writeback.h> 10d397712bSChris Mason #include <linux/buffer_head.h> 11ce9adaa5SChris Mason #include <linux/workqueue.h> 12a74a4b97SChris Mason #include <linux/kthread.h> 135a0e3ad6STejun Heo #include <linux/slab.h> 14784b4e29SChris Mason #include <linux/migrate.h> 157a36ddecSDavid Sterba #include <linux/ratelimit.h> 166463fe58SStefan Behrens #include <linux/uuid.h> 17803b2f54SStefan Behrens #include <linux/semaphore.h> 18540adea3SMasami Hiramatsu #include <linux/error-injection.h> 199678c543SNikolay Borisov #include <linux/crc32c.h> 20b89f6d1fSFilipe Manana #include <linux/sched/mm.h> 217e75bf3fSDavid Sterba #include <asm/unaligned.h> 22eb60ceacSChris Mason #include "ctree.h" 23eb60ceacSChris Mason #include "disk-io.h" 24e089f05cSChris Mason #include "transaction.h" 250f7d52f4SChris Mason #include "btrfs_inode.h" 260b86a832SChris Mason #include "volumes.h" 27db94535dSChris Mason #include "print-tree.h" 28925baeddSChris Mason #include "locking.h" 29e02119d5SChris Mason #include "tree-log.h" 30fa9c0d79SChris Mason #include "free-space-cache.h" 3170f6d82eSOmar Sandoval #include "free-space-tree.h" 32581bb050SLi Zefan #include "inode-map.h" 3321adbd5cSStefan Behrens #include "check-integrity.h" 34606686eeSJosef Bacik #include "rcu-string.h" 358dabb742SStefan Behrens #include "dev-replace.h" 3653b381b3SDavid Woodhouse #include "raid56.h" 375ac1d209SJeff Mahoney #include "sysfs.h" 38fcebe456SJosef Bacik #include "qgroup.h" 39ebb8765bSAnand Jain #include "compression.h" 40557ea5ddSQu Wenruo #include "tree-checker.h" 41fd708b81SJosef Bacik #include "ref-verify.h" 42eb60ceacSChris Mason 43de0022b9SJosef Bacik #ifdef CONFIG_X86 44de0022b9SJosef Bacik #include <asm/cpufeature.h> 45de0022b9SJosef Bacik #endif 46de0022b9SJosef Bacik 47319e4d06SQu Wenruo #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\ 48319e4d06SQu Wenruo BTRFS_HEADER_FLAG_RELOC |\ 49319e4d06SQu Wenruo BTRFS_SUPER_FLAG_ERROR |\ 50319e4d06SQu Wenruo BTRFS_SUPER_FLAG_SEEDING |\ 51e2731e55SAnand Jain BTRFS_SUPER_FLAG_METADUMP |\ 52e2731e55SAnand Jain BTRFS_SUPER_FLAG_METADUMP_V2) 53319e4d06SQu Wenruo 54e8c9f186SDavid Sterba static const struct extent_io_ops btree_extent_io_ops; 558b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work); 56143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root); 57acce952bSliubo static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 582ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info); 59143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root); 602ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 61acce952bSliubo struct extent_io_tree *dirty_pages, 62acce952bSliubo int mark); 632ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 64acce952bSliubo struct extent_io_tree *pinned_extents); 652ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info); 662ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info); 67ce9adaa5SChris Mason 68d352ac68SChris Mason /* 6997eb6b69SDavid Sterba * btrfs_end_io_wq structs are used to do processing in task context when an IO 7097eb6b69SDavid Sterba * is complete. This is used during reads to verify checksums, and it is used 71d352ac68SChris Mason * by writes to insert metadata for new file extents after IO is complete. 72d352ac68SChris Mason */ 7397eb6b69SDavid Sterba struct btrfs_end_io_wq { 74ce9adaa5SChris Mason struct bio *bio; 75ce9adaa5SChris Mason bio_end_io_t *end_io; 76ce9adaa5SChris Mason void *private; 77ce9adaa5SChris Mason struct btrfs_fs_info *info; 784e4cbee9SChristoph Hellwig blk_status_t status; 79bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata; 808b712842SChris Mason struct btrfs_work work; 81ce9adaa5SChris Mason }; 820da5468fSChris Mason 8397eb6b69SDavid Sterba static struct kmem_cache *btrfs_end_io_wq_cache; 8497eb6b69SDavid Sterba 8597eb6b69SDavid Sterba int __init btrfs_end_io_wq_init(void) 8697eb6b69SDavid Sterba { 8797eb6b69SDavid Sterba btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq", 8897eb6b69SDavid Sterba sizeof(struct btrfs_end_io_wq), 8997eb6b69SDavid Sterba 0, 90fba4b697SNikolay Borisov SLAB_MEM_SPREAD, 9197eb6b69SDavid Sterba NULL); 9297eb6b69SDavid Sterba if (!btrfs_end_io_wq_cache) 9397eb6b69SDavid Sterba return -ENOMEM; 9497eb6b69SDavid Sterba return 0; 9597eb6b69SDavid Sterba } 9697eb6b69SDavid Sterba 97e67c718bSDavid Sterba void __cold btrfs_end_io_wq_exit(void) 9897eb6b69SDavid Sterba { 9997eb6b69SDavid Sterba kmem_cache_destroy(btrfs_end_io_wq_cache); 10097eb6b69SDavid Sterba } 10197eb6b69SDavid Sterba 102d352ac68SChris Mason /* 103d352ac68SChris Mason * async submit bios are used to offload expensive checksumming 104d352ac68SChris Mason * onto the worker threads. They checksum file and metadata bios 105d352ac68SChris Mason * just before they are sent down the IO stack. 106d352ac68SChris Mason */ 10744b8bd7eSChris Mason struct async_submit_bio { 108c6100a4bSJosef Bacik void *private_data; 10944b8bd7eSChris Mason struct bio *bio; 110a758781dSDavid Sterba extent_submit_bio_start_t *submit_bio_start; 11144b8bd7eSChris Mason int mirror_num; 112eaf25d93SChris Mason /* 113eaf25d93SChris Mason * bio_offset is optional, can be used if the pages in the bio 114eaf25d93SChris Mason * can't tell us where in the file the bio should go 115eaf25d93SChris Mason */ 116eaf25d93SChris Mason u64 bio_offset; 1178b712842SChris Mason struct btrfs_work work; 1184e4cbee9SChristoph Hellwig blk_status_t status; 11944b8bd7eSChris Mason }; 12044b8bd7eSChris Mason 12185d4e461SChris Mason /* 12285d4e461SChris Mason * Lockdep class keys for extent_buffer->lock's in this root. For a given 12385d4e461SChris Mason * eb, the lockdep key is determined by the btrfs_root it belongs to and 12485d4e461SChris Mason * the level the eb occupies in the tree. 1254008c04aSChris Mason * 12685d4e461SChris Mason * Different roots are used for different purposes and may nest inside each 12785d4e461SChris Mason * other and they require separate keysets. As lockdep keys should be 12885d4e461SChris Mason * static, assign keysets according to the purpose of the root as indicated 1294fd786e6SMisono Tomohiro * by btrfs_root->root_key.objectid. This ensures that all special purpose 1304fd786e6SMisono Tomohiro * roots have separate keysets. 1314008c04aSChris Mason * 13285d4e461SChris Mason * Lock-nesting across peer nodes is always done with the immediate parent 13385d4e461SChris Mason * node locked thus preventing deadlock. As lockdep doesn't know this, use 13485d4e461SChris Mason * subclass to avoid triggering lockdep warning in such cases. 1354008c04aSChris Mason * 13685d4e461SChris Mason * The key is set by the readpage_end_io_hook after the buffer has passed 13785d4e461SChris Mason * csum validation but before the pages are unlocked. It is also set by 13885d4e461SChris Mason * btrfs_init_new_buffer on freshly allocated blocks. 13985d4e461SChris Mason * 14085d4e461SChris Mason * We also add a check to make sure the highest level of the tree is the 14185d4e461SChris Mason * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code 14285d4e461SChris Mason * needs update as well. 1434008c04aSChris Mason */ 1444008c04aSChris Mason #ifdef CONFIG_DEBUG_LOCK_ALLOC 1454008c04aSChris Mason # if BTRFS_MAX_LEVEL != 8 1464008c04aSChris Mason # error 1474008c04aSChris Mason # endif 14885d4e461SChris Mason 14985d4e461SChris Mason static struct btrfs_lockdep_keyset { 15085d4e461SChris Mason u64 id; /* root objectid */ 15185d4e461SChris Mason const char *name_stem; /* lock name stem */ 15285d4e461SChris Mason char names[BTRFS_MAX_LEVEL + 1][20]; 15385d4e461SChris Mason struct lock_class_key keys[BTRFS_MAX_LEVEL + 1]; 15485d4e461SChris Mason } btrfs_lockdep_keysets[] = { 15585d4e461SChris Mason { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" }, 15685d4e461SChris Mason { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" }, 15785d4e461SChris Mason { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" }, 15885d4e461SChris Mason { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" }, 15985d4e461SChris Mason { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" }, 16085d4e461SChris Mason { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" }, 16160b62978SDavid Sterba { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" }, 16285d4e461SChris Mason { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" }, 16385d4e461SChris Mason { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" }, 16485d4e461SChris Mason { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" }, 16513fd8da9SDavid Sterba { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" }, 1666b20e0adSDavid Sterba { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" }, 16785d4e461SChris Mason { .id = 0, .name_stem = "tree" }, 1684008c04aSChris Mason }; 16985d4e461SChris Mason 17085d4e461SChris Mason void __init btrfs_init_lockdep(void) 17185d4e461SChris Mason { 17285d4e461SChris Mason int i, j; 17385d4e461SChris Mason 17485d4e461SChris Mason /* initialize lockdep class names */ 17585d4e461SChris Mason for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) { 17685d4e461SChris Mason struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i]; 17785d4e461SChris Mason 17885d4e461SChris Mason for (j = 0; j < ARRAY_SIZE(ks->names); j++) 17985d4e461SChris Mason snprintf(ks->names[j], sizeof(ks->names[j]), 18085d4e461SChris Mason "btrfs-%s-%02d", ks->name_stem, j); 18185d4e461SChris Mason } 18285d4e461SChris Mason } 18385d4e461SChris Mason 18485d4e461SChris Mason void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb, 18585d4e461SChris Mason int level) 18685d4e461SChris Mason { 18785d4e461SChris Mason struct btrfs_lockdep_keyset *ks; 18885d4e461SChris Mason 18985d4e461SChris Mason BUG_ON(level >= ARRAY_SIZE(ks->keys)); 19085d4e461SChris Mason 19185d4e461SChris Mason /* find the matching keyset, id 0 is the default entry */ 19285d4e461SChris Mason for (ks = btrfs_lockdep_keysets; ks->id; ks++) 19385d4e461SChris Mason if (ks->id == objectid) 19485d4e461SChris Mason break; 19585d4e461SChris Mason 19685d4e461SChris Mason lockdep_set_class_and_name(&eb->lock, 19785d4e461SChris Mason &ks->keys[level], ks->names[level]); 19885d4e461SChris Mason } 19985d4e461SChris Mason 2004008c04aSChris Mason #endif 2014008c04aSChris Mason 202d352ac68SChris Mason /* 203d352ac68SChris Mason * extents on the btree inode are pretty simple, there's one extent 204d352ac68SChris Mason * that covers the entire device 205d352ac68SChris Mason */ 2066af49dbdSDavid Sterba struct extent_map *btree_get_extent(struct btrfs_inode *inode, 207306e16ceSDavid Sterba struct page *page, size_t pg_offset, u64 start, u64 len, 2085f39d397SChris Mason int create) 2095f39d397SChris Mason { 2103ffbd68cSDavid Sterba struct btrfs_fs_info *fs_info = inode->root->fs_info; 211fc4f21b1SNikolay Borisov struct extent_map_tree *em_tree = &inode->extent_tree; 2125f39d397SChris Mason struct extent_map *em; 2135f39d397SChris Mason int ret; 2145f39d397SChris Mason 215890871beSChris Mason read_lock(&em_tree->lock); 216d1310b2eSChris Mason em = lookup_extent_mapping(em_tree, start, len); 217a061fc8dSChris Mason if (em) { 2180b246afaSJeff Mahoney em->bdev = fs_info->fs_devices->latest_bdev; 219890871beSChris Mason read_unlock(&em_tree->lock); 2205f39d397SChris Mason goto out; 221a061fc8dSChris Mason } 222890871beSChris Mason read_unlock(&em_tree->lock); 2237b13b7b1SChris Mason 224172ddd60SDavid Sterba em = alloc_extent_map(); 2255f39d397SChris Mason if (!em) { 2265f39d397SChris Mason em = ERR_PTR(-ENOMEM); 2275f39d397SChris Mason goto out; 2285f39d397SChris Mason } 2295f39d397SChris Mason em->start = 0; 2300afbaf8cSChris Mason em->len = (u64)-1; 231c8b97818SChris Mason em->block_len = (u64)-1; 2325f39d397SChris Mason em->block_start = 0; 2330b246afaSJeff Mahoney em->bdev = fs_info->fs_devices->latest_bdev; 234d1310b2eSChris Mason 235890871beSChris Mason write_lock(&em_tree->lock); 23609a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 2375f39d397SChris Mason if (ret == -EEXIST) { 2385f39d397SChris Mason free_extent_map(em); 2397b13b7b1SChris Mason em = lookup_extent_mapping(em_tree, start, len); 240b4f359abSTsutomu Itoh if (!em) 2410433f20dSTsutomu Itoh em = ERR_PTR(-EIO); 2425f39d397SChris Mason } else if (ret) { 2437b13b7b1SChris Mason free_extent_map(em); 2440433f20dSTsutomu Itoh em = ERR_PTR(ret); 2455f39d397SChris Mason } 246890871beSChris Mason write_unlock(&em_tree->lock); 2477b13b7b1SChris Mason 2485f39d397SChris Mason out: 2495f39d397SChris Mason return em; 2505f39d397SChris Mason } 2515f39d397SChris Mason 2529ed57367SDavid Sterba u32 btrfs_csum_data(const char *data, u32 seed, size_t len) 25319c00ddcSChris Mason { 2549678c543SNikolay Borisov return crc32c(seed, data, len); 25519c00ddcSChris Mason } 25619c00ddcSChris Mason 2570b5e3dafSDomagoj Tršan void btrfs_csum_final(u32 crc, u8 *result) 25819c00ddcSChris Mason { 2597e75bf3fSDavid Sterba put_unaligned_le32(~crc, result); 26019c00ddcSChris Mason } 26119c00ddcSChris Mason 262d352ac68SChris Mason /* 2632996e1f8SJohannes Thumshirn * Compute the csum of a btree block and store the result to provided buffer. 2642996e1f8SJohannes Thumshirn * 2652996e1f8SJohannes Thumshirn * Returns error if the extent buffer cannot be mapped. 266d352ac68SChris Mason */ 2672996e1f8SJohannes Thumshirn static int csum_tree_block(struct extent_buffer *buf, u8 *result) 26819c00ddcSChris Mason { 26919c00ddcSChris Mason unsigned long len; 27019c00ddcSChris Mason unsigned long cur_len; 27119c00ddcSChris Mason unsigned long offset = BTRFS_CSUM_SIZE; 27219c00ddcSChris Mason char *kaddr; 27319c00ddcSChris Mason unsigned long map_start; 27419c00ddcSChris Mason unsigned long map_len; 27519c00ddcSChris Mason int err; 27619c00ddcSChris Mason u32 crc = ~(u32)0; 27719c00ddcSChris Mason 27819c00ddcSChris Mason len = buf->len - offset; 27919c00ddcSChris Mason while (len > 0) { 280d2e174d5SJohannes Thumshirn /* 281d2e174d5SJohannes Thumshirn * Note: we don't need to check for the err == 1 case here, as 282d2e174d5SJohannes Thumshirn * with the given combination of 'start = BTRFS_CSUM_SIZE (32)' 283d2e174d5SJohannes Thumshirn * and 'min_len = 32' and the currently implemented mapping 284d2e174d5SJohannes Thumshirn * algorithm we cannot cross a page boundary. 285d2e174d5SJohannes Thumshirn */ 28619c00ddcSChris Mason err = map_private_extent_buffer(buf, offset, 32, 287a6591715SChris Mason &kaddr, &map_start, &map_len); 288c53839fcSJohannes Thumshirn if (WARN_ON(err)) 2898bd98f0eSAlex Lyakas return err; 29019c00ddcSChris Mason cur_len = min(len, map_len - (offset - map_start)); 291b0496686SLiu Bo crc = btrfs_csum_data(kaddr + offset - map_start, 29219c00ddcSChris Mason crc, cur_len); 29319c00ddcSChris Mason len -= cur_len; 29419c00ddcSChris Mason offset += cur_len; 29519c00ddcSChris Mason } 29671a63551SDavid Sterba memset(result, 0, BTRFS_CSUM_SIZE); 297607d432dSJosef Bacik 29819c00ddcSChris Mason btrfs_csum_final(crc, result); 29919c00ddcSChris Mason 30019c00ddcSChris Mason return 0; 30119c00ddcSChris Mason } 30219c00ddcSChris Mason 303d352ac68SChris Mason /* 304d352ac68SChris Mason * we can't consider a given block up to date unless the transid of the 305d352ac68SChris Mason * block matches the transid in the parent node's pointer. This is how we 306d352ac68SChris Mason * detect blocks that either didn't get written at all or got written 307d352ac68SChris Mason * in the wrong place. 308d352ac68SChris Mason */ 3091259ab75SChris Mason static int verify_parent_transid(struct extent_io_tree *io_tree, 310b9fab919SChris Mason struct extent_buffer *eb, u64 parent_transid, 311b9fab919SChris Mason int atomic) 3121259ab75SChris Mason { 3132ac55d41SJosef Bacik struct extent_state *cached_state = NULL; 3141259ab75SChris Mason int ret; 3152755a0deSDavid Sterba bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB); 3161259ab75SChris Mason 3171259ab75SChris Mason if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 3181259ab75SChris Mason return 0; 3191259ab75SChris Mason 320b9fab919SChris Mason if (atomic) 321b9fab919SChris Mason return -EAGAIN; 322b9fab919SChris Mason 323a26e8c9fSJosef Bacik if (need_lock) { 324a26e8c9fSJosef Bacik btrfs_tree_read_lock(eb); 325300aa896SDavid Sterba btrfs_set_lock_blocking_read(eb); 326a26e8c9fSJosef Bacik } 327a26e8c9fSJosef Bacik 3282ac55d41SJosef Bacik lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 329ff13db41SDavid Sterba &cached_state); 3300b32f4bbSJosef Bacik if (extent_buffer_uptodate(eb) && 3311259ab75SChris Mason btrfs_header_generation(eb) == parent_transid) { 3321259ab75SChris Mason ret = 0; 3331259ab75SChris Mason goto out; 3341259ab75SChris Mason } 33594647322SDavid Sterba btrfs_err_rl(eb->fs_info, 33694647322SDavid Sterba "parent transid verify failed on %llu wanted %llu found %llu", 33794647322SDavid Sterba eb->start, 33829549aecSWang Shilong parent_transid, btrfs_header_generation(eb)); 3391259ab75SChris Mason ret = 1; 340a26e8c9fSJosef Bacik 341a26e8c9fSJosef Bacik /* 342a26e8c9fSJosef Bacik * Things reading via commit roots that don't have normal protection, 343a26e8c9fSJosef Bacik * like send, can have a really old block in cache that may point at a 34401327610SNicholas D Steeves * block that has been freed and re-allocated. So don't clear uptodate 345a26e8c9fSJosef Bacik * if we find an eb that is under IO (dirty/writeback) because we could 346a26e8c9fSJosef Bacik * end up reading in the stale data and then writing it back out and 347a26e8c9fSJosef Bacik * making everybody very sad. 348a26e8c9fSJosef Bacik */ 349a26e8c9fSJosef Bacik if (!extent_buffer_under_io(eb)) 3500b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 35133958dc6SChris Mason out: 3522ac55d41SJosef Bacik unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 353e43bbe5eSDavid Sterba &cached_state); 354472b909fSJosef Bacik if (need_lock) 355a26e8c9fSJosef Bacik btrfs_tree_read_unlock_blocking(eb); 3561259ab75SChris Mason return ret; 3571259ab75SChris Mason } 3581259ab75SChris Mason 359d352ac68SChris Mason /* 3601104a885SDavid Sterba * Return 0 if the superblock checksum type matches the checksum value of that 3611104a885SDavid Sterba * algorithm. Pass the raw disk superblock data. 3621104a885SDavid Sterba */ 363ab8d0fc4SJeff Mahoney static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info, 364ab8d0fc4SJeff Mahoney char *raw_disk_sb) 3651104a885SDavid Sterba { 3661104a885SDavid Sterba struct btrfs_super_block *disk_sb = 3671104a885SDavid Sterba (struct btrfs_super_block *)raw_disk_sb; 3681104a885SDavid Sterba u16 csum_type = btrfs_super_csum_type(disk_sb); 3691104a885SDavid Sterba int ret = 0; 3701104a885SDavid Sterba 3711104a885SDavid Sterba if (csum_type == BTRFS_CSUM_TYPE_CRC32) { 3721104a885SDavid Sterba u32 crc = ~(u32)0; 373776c4a7cSNikolay Borisov char result[sizeof(crc)]; 3741104a885SDavid Sterba 3751104a885SDavid Sterba /* 3761104a885SDavid Sterba * The super_block structure does not span the whole 3771104a885SDavid Sterba * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space 37801327610SNicholas D Steeves * is filled with zeros and is included in the checksum. 3791104a885SDavid Sterba */ 3801104a885SDavid Sterba crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE, 3811104a885SDavid Sterba crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE); 3821104a885SDavid Sterba btrfs_csum_final(crc, result); 3831104a885SDavid Sterba 384776c4a7cSNikolay Borisov if (memcmp(raw_disk_sb, result, sizeof(result))) 3851104a885SDavid Sterba ret = 1; 3861104a885SDavid Sterba } 3871104a885SDavid Sterba 3881104a885SDavid Sterba if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) { 389ab8d0fc4SJeff Mahoney btrfs_err(fs_info, "unsupported checksum algorithm %u", 3901104a885SDavid Sterba csum_type); 3911104a885SDavid Sterba ret = 1; 3921104a885SDavid Sterba } 3931104a885SDavid Sterba 3941104a885SDavid Sterba return ret; 3951104a885SDavid Sterba } 3961104a885SDavid Sterba 397e064d5e9SDavid Sterba int btrfs_verify_level_key(struct extent_buffer *eb, int level, 398ff76a864SLiu Bo struct btrfs_key *first_key, u64 parent_transid) 399581c1760SQu Wenruo { 400e064d5e9SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 401581c1760SQu Wenruo int found_level; 402581c1760SQu Wenruo struct btrfs_key found_key; 403581c1760SQu Wenruo int ret; 404581c1760SQu Wenruo 405581c1760SQu Wenruo found_level = btrfs_header_level(eb); 406581c1760SQu Wenruo if (found_level != level) { 40763489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 40863489055SQu Wenruo KERN_ERR "BTRFS: tree level check failed\n"); 409581c1760SQu Wenruo btrfs_err(fs_info, 410581c1760SQu Wenruo "tree level mismatch detected, bytenr=%llu level expected=%u has=%u", 411581c1760SQu Wenruo eb->start, level, found_level); 412581c1760SQu Wenruo return -EIO; 413581c1760SQu Wenruo } 414581c1760SQu Wenruo 415581c1760SQu Wenruo if (!first_key) 416581c1760SQu Wenruo return 0; 417581c1760SQu Wenruo 4185d41be6fSQu Wenruo /* 4195d41be6fSQu Wenruo * For live tree block (new tree blocks in current transaction), 4205d41be6fSQu Wenruo * we need proper lock context to avoid race, which is impossible here. 4215d41be6fSQu Wenruo * So we only checks tree blocks which is read from disk, whose 4225d41be6fSQu Wenruo * generation <= fs_info->last_trans_committed. 4235d41be6fSQu Wenruo */ 4245d41be6fSQu Wenruo if (btrfs_header_generation(eb) > fs_info->last_trans_committed) 4255d41be6fSQu Wenruo return 0; 426581c1760SQu Wenruo if (found_level) 427581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &found_key, 0); 428581c1760SQu Wenruo else 429581c1760SQu Wenruo btrfs_item_key_to_cpu(eb, &found_key, 0); 430581c1760SQu Wenruo ret = btrfs_comp_cpu_keys(first_key, &found_key); 431581c1760SQu Wenruo 432581c1760SQu Wenruo if (ret) { 43363489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 43463489055SQu Wenruo KERN_ERR "BTRFS: tree first key check failed\n"); 435581c1760SQu Wenruo btrfs_err(fs_info, 436ff76a864SLiu Bo "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", 437ff76a864SLiu Bo eb->start, parent_transid, first_key->objectid, 438ff76a864SLiu Bo first_key->type, first_key->offset, 439ff76a864SLiu Bo found_key.objectid, found_key.type, 440ff76a864SLiu Bo found_key.offset); 441581c1760SQu Wenruo } 442581c1760SQu Wenruo return ret; 443581c1760SQu Wenruo } 444581c1760SQu Wenruo 4451104a885SDavid Sterba /* 446d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 447d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 448581c1760SQu Wenruo * 449581c1760SQu Wenruo * @parent_transid: expected transid, skip check if 0 450581c1760SQu Wenruo * @level: expected level, mandatory check 451581c1760SQu Wenruo * @first_key: expected key of first slot, skip check if NULL 452d352ac68SChris Mason */ 4535ab12d1fSDavid Sterba static int btree_read_extent_buffer_pages(struct extent_buffer *eb, 454581c1760SQu Wenruo u64 parent_transid, int level, 455581c1760SQu Wenruo struct btrfs_key *first_key) 456f188591eSChris Mason { 4575ab12d1fSDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 458f188591eSChris Mason struct extent_io_tree *io_tree; 459ea466794SJosef Bacik int failed = 0; 460f188591eSChris Mason int ret; 461f188591eSChris Mason int num_copies = 0; 462f188591eSChris Mason int mirror_num = 0; 463ea466794SJosef Bacik int failed_mirror = 0; 464f188591eSChris Mason 4650b246afaSJeff Mahoney io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; 466f188591eSChris Mason while (1) { 467f8397d69SNikolay Borisov clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 468c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num); 469256dd1bbSStefan Behrens if (!ret) { 470581c1760SQu Wenruo if (verify_parent_transid(io_tree, eb, 471b9fab919SChris Mason parent_transid, 0)) 472256dd1bbSStefan Behrens ret = -EIO; 473e064d5e9SDavid Sterba else if (btrfs_verify_level_key(eb, level, 474ff76a864SLiu Bo first_key, parent_transid)) 475581c1760SQu Wenruo ret = -EUCLEAN; 476581c1760SQu Wenruo else 477581c1760SQu Wenruo break; 478256dd1bbSStefan Behrens } 479d397712bSChris Mason 4800b246afaSJeff Mahoney num_copies = btrfs_num_copies(fs_info, 481f188591eSChris Mason eb->start, eb->len); 4824235298eSChris Mason if (num_copies == 1) 483ea466794SJosef Bacik break; 4844235298eSChris Mason 4855cf1ab56SJosef Bacik if (!failed_mirror) { 4865cf1ab56SJosef Bacik failed = 1; 4875cf1ab56SJosef Bacik failed_mirror = eb->read_mirror; 4885cf1ab56SJosef Bacik } 4895cf1ab56SJosef Bacik 490f188591eSChris Mason mirror_num++; 491ea466794SJosef Bacik if (mirror_num == failed_mirror) 492ea466794SJosef Bacik mirror_num++; 493ea466794SJosef Bacik 4944235298eSChris Mason if (mirror_num > num_copies) 495ea466794SJosef Bacik break; 496f188591eSChris Mason } 497ea466794SJosef Bacik 498c0901581SStefan Behrens if (failed && !ret && failed_mirror) 49920a1fbf9SDavid Sterba btrfs_repair_eb_io_failure(eb, failed_mirror); 500ea466794SJosef Bacik 501ea466794SJosef Bacik return ret; 502f188591eSChris Mason } 50319c00ddcSChris Mason 504d352ac68SChris Mason /* 505d397712bSChris Mason * checksum a dirty tree block before IO. This has extra checks to make sure 506d397712bSChris Mason * we only fill in the checksum field in the first page of a multi-page block 507d352ac68SChris Mason */ 508d397712bSChris Mason 50901d58472SDaniel Dressler static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page) 51019c00ddcSChris Mason { 5114eee4fa4SMiao Xie u64 start = page_offset(page); 51219c00ddcSChris Mason u64 found_start; 5132996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 5142996e1f8SJohannes Thumshirn u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); 51519c00ddcSChris Mason struct extent_buffer *eb; 5168d47a0d8SQu Wenruo int ret; 517f188591eSChris Mason 5184f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 5194f2de97aSJosef Bacik if (page != eb->pages[0]) 5204f2de97aSJosef Bacik return 0; 5210f805531SAlex Lyakas 52219c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 5230f805531SAlex Lyakas /* 5240f805531SAlex Lyakas * Please do not consolidate these warnings into a single if. 5250f805531SAlex Lyakas * It is useful to know what went wrong. 5260f805531SAlex Lyakas */ 5270f805531SAlex Lyakas if (WARN_ON(found_start != start)) 5280f805531SAlex Lyakas return -EUCLEAN; 5290f805531SAlex Lyakas if (WARN_ON(!PageUptodate(page))) 5300f805531SAlex Lyakas return -EUCLEAN; 5310f805531SAlex Lyakas 532de37aa51SNikolay Borisov ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid, 5330f805531SAlex Lyakas btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0); 5340f805531SAlex Lyakas 5352996e1f8SJohannes Thumshirn if (csum_tree_block(eb, result)) 5362996e1f8SJohannes Thumshirn return -EINVAL; 5372996e1f8SJohannes Thumshirn 5388d47a0d8SQu Wenruo if (btrfs_header_level(eb)) 5398d47a0d8SQu Wenruo ret = btrfs_check_node(eb); 5408d47a0d8SQu Wenruo else 5418d47a0d8SQu Wenruo ret = btrfs_check_leaf_full(eb); 5428d47a0d8SQu Wenruo 5438d47a0d8SQu Wenruo if (ret < 0) { 5448d47a0d8SQu Wenruo btrfs_err(fs_info, 5458d47a0d8SQu Wenruo "block=%llu write time tree block corruption detected", 5468d47a0d8SQu Wenruo eb->start); 5478d47a0d8SQu Wenruo return ret; 5488d47a0d8SQu Wenruo } 5492996e1f8SJohannes Thumshirn write_extent_buffer(eb, result, 0, csum_size); 5508d47a0d8SQu Wenruo 5512996e1f8SJohannes Thumshirn return 0; 55219c00ddcSChris Mason } 55319c00ddcSChris Mason 554b0c9b3b0SDavid Sterba static int check_tree_block_fsid(struct extent_buffer *eb) 5552b82032cSYan Zheng { 556b0c9b3b0SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 55701d58472SDaniel Dressler struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 55844880fdcSAnand Jain u8 fsid[BTRFS_FSID_SIZE]; 5592b82032cSYan Zheng int ret = 1; 5602b82032cSYan Zheng 5610a4e5586SRoss Kirk read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE); 5622b82032cSYan Zheng while (fs_devices) { 5637239ff4bSNikolay Borisov u8 *metadata_uuid; 5647239ff4bSNikolay Borisov 5657239ff4bSNikolay Borisov /* 5667239ff4bSNikolay Borisov * Checking the incompat flag is only valid for the current 5677239ff4bSNikolay Borisov * fs. For seed devices it's forbidden to have their uuid 5687239ff4bSNikolay Borisov * changed so reading ->fsid in this case is fine 5697239ff4bSNikolay Borisov */ 5707239ff4bSNikolay Borisov if (fs_devices == fs_info->fs_devices && 5717239ff4bSNikolay Borisov btrfs_fs_incompat(fs_info, METADATA_UUID)) 5727239ff4bSNikolay Borisov metadata_uuid = fs_devices->metadata_uuid; 5737239ff4bSNikolay Borisov else 5747239ff4bSNikolay Borisov metadata_uuid = fs_devices->fsid; 5757239ff4bSNikolay Borisov 5767239ff4bSNikolay Borisov if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) { 5772b82032cSYan Zheng ret = 0; 5782b82032cSYan Zheng break; 5792b82032cSYan Zheng } 5802b82032cSYan Zheng fs_devices = fs_devices->seed; 5812b82032cSYan Zheng } 5822b82032cSYan Zheng return ret; 5832b82032cSYan Zheng } 5842b82032cSYan Zheng 585facc8a22SMiao Xie static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio, 586facc8a22SMiao Xie u64 phy_offset, struct page *page, 587facc8a22SMiao Xie u64 start, u64 end, int mirror) 588ce9adaa5SChris Mason { 589ce9adaa5SChris Mason u64 found_start; 590ce9adaa5SChris Mason int found_level; 591ce9adaa5SChris Mason struct extent_buffer *eb; 592ce9adaa5SChris Mason struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 59302873e43SZhao Lei struct btrfs_fs_info *fs_info = root->fs_info; 5942996e1f8SJohannes Thumshirn u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); 595f188591eSChris Mason int ret = 0; 5962996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 597727011e0SChris Mason int reads_done; 598ce9adaa5SChris Mason 599ce9adaa5SChris Mason if (!page->private) 600ce9adaa5SChris Mason goto out; 601d397712bSChris Mason 6024f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 603d397712bSChris Mason 6040b32f4bbSJosef Bacik /* the pending IO might have been the only thing that kept this buffer 6050b32f4bbSJosef Bacik * in memory. Make sure we have a ref for all this other checks 6060b32f4bbSJosef Bacik */ 6070b32f4bbSJosef Bacik extent_buffer_get(eb); 6080b32f4bbSJosef Bacik 6090b32f4bbSJosef Bacik reads_done = atomic_dec_and_test(&eb->io_pages); 610727011e0SChris Mason if (!reads_done) 611727011e0SChris Mason goto err; 612f188591eSChris Mason 6135cf1ab56SJosef Bacik eb->read_mirror = mirror; 614656f30dbSFilipe Manana if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 615ea466794SJosef Bacik ret = -EIO; 616ea466794SJosef Bacik goto err; 617ea466794SJosef Bacik } 618ea466794SJosef Bacik 619ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 620727011e0SChris Mason if (found_start != eb->start) { 621893bf4b1SSu Yue btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu", 622893bf4b1SSu Yue eb->start, found_start); 623f188591eSChris Mason ret = -EIO; 624ce9adaa5SChris Mason goto err; 625ce9adaa5SChris Mason } 626b0c9b3b0SDavid Sterba if (check_tree_block_fsid(eb)) { 62702873e43SZhao Lei btrfs_err_rl(fs_info, "bad fsid on block %llu", 62894647322SDavid Sterba eb->start); 6291259ab75SChris Mason ret = -EIO; 6301259ab75SChris Mason goto err; 6311259ab75SChris Mason } 632ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 6331c24c3ceSJosef Bacik if (found_level >= BTRFS_MAX_LEVEL) { 634893bf4b1SSu Yue btrfs_err(fs_info, "bad tree block level %d on %llu", 635893bf4b1SSu Yue (int)btrfs_header_level(eb), eb->start); 6361c24c3ceSJosef Bacik ret = -EIO; 6371c24c3ceSJosef Bacik goto err; 6381c24c3ceSJosef Bacik } 639ce9adaa5SChris Mason 64085d4e461SChris Mason btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), 64185d4e461SChris Mason eb, found_level); 6424008c04aSChris Mason 6432996e1f8SJohannes Thumshirn ret = csum_tree_block(eb, result); 6448bd98f0eSAlex Lyakas if (ret) 645a826d6dcSJosef Bacik goto err; 646a826d6dcSJosef Bacik 6472996e1f8SJohannes Thumshirn if (memcmp_extent_buffer(eb, result, 0, csum_size)) { 6482996e1f8SJohannes Thumshirn u32 val; 6492996e1f8SJohannes Thumshirn u32 found = 0; 6502996e1f8SJohannes Thumshirn 6512996e1f8SJohannes Thumshirn memcpy(&found, result, csum_size); 6522996e1f8SJohannes Thumshirn 6532996e1f8SJohannes Thumshirn read_extent_buffer(eb, &val, 0, csum_size); 6542996e1f8SJohannes Thumshirn btrfs_warn_rl(fs_info, 6552996e1f8SJohannes Thumshirn "%s checksum verify failed on %llu wanted %x found %x level %d", 6562996e1f8SJohannes Thumshirn fs_info->sb->s_id, eb->start, 6572996e1f8SJohannes Thumshirn val, found, btrfs_header_level(eb)); 6582996e1f8SJohannes Thumshirn ret = -EUCLEAN; 6592996e1f8SJohannes Thumshirn goto err; 6602996e1f8SJohannes Thumshirn } 6612996e1f8SJohannes Thumshirn 662a826d6dcSJosef Bacik /* 663a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 664a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 665a826d6dcSJosef Bacik * return -EIO. 666a826d6dcSJosef Bacik */ 6671c4360eeSDavid Sterba if (found_level == 0 && btrfs_check_leaf_full(eb)) { 668a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 669a826d6dcSJosef Bacik ret = -EIO; 670a826d6dcSJosef Bacik } 671ce9adaa5SChris Mason 672813fd1dcSDavid Sterba if (found_level > 0 && btrfs_check_node(eb)) 673053ab70fSLiu Bo ret = -EIO; 674053ab70fSLiu Bo 6750b32f4bbSJosef Bacik if (!ret) 6760b32f4bbSJosef Bacik set_extent_buffer_uptodate(eb); 67775391f0dSQu Wenruo else 67875391f0dSQu Wenruo btrfs_err(fs_info, 67975391f0dSQu Wenruo "block=%llu read time tree block corruption detected", 68075391f0dSQu Wenruo eb->start); 681ce9adaa5SChris Mason err: 68279fb65a1SJosef Bacik if (reads_done && 68379fb65a1SJosef Bacik test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 684d48d71aaSDavid Sterba btree_readahead_hook(eb, ret); 6854bb31e92SArne Jansen 68653b381b3SDavid Woodhouse if (ret) { 68753b381b3SDavid Woodhouse /* 68853b381b3SDavid Woodhouse * our io error hook is going to dec the io pages 68953b381b3SDavid Woodhouse * again, we have to make sure it has something 69053b381b3SDavid Woodhouse * to decrement 69153b381b3SDavid Woodhouse */ 69253b381b3SDavid Woodhouse atomic_inc(&eb->io_pages); 6930b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 69453b381b3SDavid Woodhouse } 6950b32f4bbSJosef Bacik free_extent_buffer(eb); 696ce9adaa5SChris Mason out: 697f188591eSChris Mason return ret; 698ce9adaa5SChris Mason } 699ce9adaa5SChris Mason 7004246a0b6SChristoph Hellwig static void end_workqueue_bio(struct bio *bio) 701ce9adaa5SChris Mason { 70297eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq = bio->bi_private; 703ce9adaa5SChris Mason struct btrfs_fs_info *fs_info; 7049e0af237SLiu Bo struct btrfs_workqueue *wq; 7059e0af237SLiu Bo btrfs_work_func_t func; 706ce9adaa5SChris Mason 707ce9adaa5SChris Mason fs_info = end_io_wq->info; 7084e4cbee9SChristoph Hellwig end_io_wq->status = bio->bi_status; 709d20f7043SChris Mason 71037226b21SMike Christie if (bio_op(bio) == REQ_OP_WRITE) { 7119e0af237SLiu Bo if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) { 7129e0af237SLiu Bo wq = fs_info->endio_meta_write_workers; 7139e0af237SLiu Bo func = btrfs_endio_meta_write_helper; 7149e0af237SLiu Bo } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) { 7159e0af237SLiu Bo wq = fs_info->endio_freespace_worker; 7169e0af237SLiu Bo func = btrfs_freespace_write_helper; 7179e0af237SLiu Bo } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) { 7189e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 7199e0af237SLiu Bo func = btrfs_endio_raid56_helper; 720d20f7043SChris Mason } else { 7219e0af237SLiu Bo wq = fs_info->endio_write_workers; 7229e0af237SLiu Bo func = btrfs_endio_write_helper; 723d20f7043SChris Mason } 7249e0af237SLiu Bo } else { 7258b110e39SMiao Xie if (unlikely(end_io_wq->metadata == 7268b110e39SMiao Xie BTRFS_WQ_ENDIO_DIO_REPAIR)) { 7278b110e39SMiao Xie wq = fs_info->endio_repair_workers; 7288b110e39SMiao Xie func = btrfs_endio_repair_helper; 7298b110e39SMiao Xie } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) { 7309e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 7319e0af237SLiu Bo func = btrfs_endio_raid56_helper; 7329e0af237SLiu Bo } else if (end_io_wq->metadata) { 7339e0af237SLiu Bo wq = fs_info->endio_meta_workers; 7349e0af237SLiu Bo func = btrfs_endio_meta_helper; 7359e0af237SLiu Bo } else { 7369e0af237SLiu Bo wq = fs_info->endio_workers; 7379e0af237SLiu Bo func = btrfs_endio_helper; 7389e0af237SLiu Bo } 7399e0af237SLiu Bo } 7409e0af237SLiu Bo 7419e0af237SLiu Bo btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL); 7429e0af237SLiu Bo btrfs_queue_work(wq, &end_io_wq->work); 743ce9adaa5SChris Mason } 744ce9adaa5SChris Mason 7454e4cbee9SChristoph Hellwig blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 746bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata) 7470b86a832SChris Mason { 74897eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 7498b110e39SMiao Xie 75097eb6b69SDavid Sterba end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS); 751ce9adaa5SChris Mason if (!end_io_wq) 7524e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 753ce9adaa5SChris Mason 754ce9adaa5SChris Mason end_io_wq->private = bio->bi_private; 755ce9adaa5SChris Mason end_io_wq->end_io = bio->bi_end_io; 75622c59948SChris Mason end_io_wq->info = info; 7574e4cbee9SChristoph Hellwig end_io_wq->status = 0; 758ce9adaa5SChris Mason end_io_wq->bio = bio; 75922c59948SChris Mason end_io_wq->metadata = metadata; 760ce9adaa5SChris Mason 761ce9adaa5SChris Mason bio->bi_private = end_io_wq; 762ce9adaa5SChris Mason bio->bi_end_io = end_workqueue_bio; 76322c59948SChris Mason return 0; 76422c59948SChris Mason } 76522c59948SChris Mason 7664a69a410SChris Mason static void run_one_async_start(struct btrfs_work *work) 7674a69a410SChris Mason { 7684a69a410SChris Mason struct async_submit_bio *async; 7694e4cbee9SChristoph Hellwig blk_status_t ret; 7704a69a410SChris Mason 7714a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 772c6100a4bSJosef Bacik ret = async->submit_bio_start(async->private_data, async->bio, 773eaf25d93SChris Mason async->bio_offset); 77479787eaaSJeff Mahoney if (ret) 7754e4cbee9SChristoph Hellwig async->status = ret; 7764a69a410SChris Mason } 7774a69a410SChris Mason 77806ea01b1SDavid Sterba /* 77906ea01b1SDavid Sterba * In order to insert checksums into the metadata in large chunks, we wait 78006ea01b1SDavid Sterba * until bio submission time. All the pages in the bio are checksummed and 78106ea01b1SDavid Sterba * sums are attached onto the ordered extent record. 78206ea01b1SDavid Sterba * 78306ea01b1SDavid Sterba * At IO completion time the csums attached on the ordered extent record are 78406ea01b1SDavid Sterba * inserted into the tree. 78506ea01b1SDavid Sterba */ 7864a69a410SChris Mason static void run_one_async_done(struct btrfs_work *work) 7878b712842SChris Mason { 7888b712842SChris Mason struct async_submit_bio *async; 78906ea01b1SDavid Sterba struct inode *inode; 79006ea01b1SDavid Sterba blk_status_t ret; 7918b712842SChris Mason 7928b712842SChris Mason async = container_of(work, struct async_submit_bio, work); 79306ea01b1SDavid Sterba inode = async->private_data; 7944854ddd0SChris Mason 795bb7ab3b9SAdam Buchbinder /* If an error occurred we just want to clean up the bio and move on */ 7964e4cbee9SChristoph Hellwig if (async->status) { 7974e4cbee9SChristoph Hellwig async->bio->bi_status = async->status; 7984246a0b6SChristoph Hellwig bio_endio(async->bio); 79979787eaaSJeff Mahoney return; 80079787eaaSJeff Mahoney } 80179787eaaSJeff Mahoney 80206ea01b1SDavid Sterba ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, 80306ea01b1SDavid Sterba async->mirror_num, 1); 80406ea01b1SDavid Sterba if (ret) { 80506ea01b1SDavid Sterba async->bio->bi_status = ret; 80606ea01b1SDavid Sterba bio_endio(async->bio); 80706ea01b1SDavid Sterba } 8084a69a410SChris Mason } 8094a69a410SChris Mason 8104a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 8114a69a410SChris Mason { 8124a69a410SChris Mason struct async_submit_bio *async; 8134a69a410SChris Mason 8144a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 8158b712842SChris Mason kfree(async); 8168b712842SChris Mason } 8178b712842SChris Mason 8188c27cb35SLinus Torvalds blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio, 819c6100a4bSJosef Bacik int mirror_num, unsigned long bio_flags, 820c6100a4bSJosef Bacik u64 bio_offset, void *private_data, 821e288c080SDavid Sterba extent_submit_bio_start_t *submit_bio_start) 82244b8bd7eSChris Mason { 82344b8bd7eSChris Mason struct async_submit_bio *async; 82444b8bd7eSChris Mason 82544b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 82644b8bd7eSChris Mason if (!async) 8274e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 82844b8bd7eSChris Mason 829c6100a4bSJosef Bacik async->private_data = private_data; 83044b8bd7eSChris Mason async->bio = bio; 83144b8bd7eSChris Mason async->mirror_num = mirror_num; 8324a69a410SChris Mason async->submit_bio_start = submit_bio_start; 8334a69a410SChris Mason 8349e0af237SLiu Bo btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start, 8355cdc7ad3SQu Wenruo run_one_async_done, run_one_async_free); 8364a69a410SChris Mason 837eaf25d93SChris Mason async->bio_offset = bio_offset; 8388c8bee1dSChris Mason 8394e4cbee9SChristoph Hellwig async->status = 0; 84079787eaaSJeff Mahoney 84167f055c7SChristoph Hellwig if (op_is_sync(bio->bi_opf)) 8425cdc7ad3SQu Wenruo btrfs_set_work_high_priority(&async->work); 843d313d7a3SChris Mason 8445cdc7ad3SQu Wenruo btrfs_queue_work(fs_info->workers, &async->work); 84544b8bd7eSChris Mason return 0; 84644b8bd7eSChris Mason } 84744b8bd7eSChris Mason 8484e4cbee9SChristoph Hellwig static blk_status_t btree_csum_one_bio(struct bio *bio) 849ce3ed71aSChris Mason { 8502c30c71bSKent Overstreet struct bio_vec *bvec; 851ce3ed71aSChris Mason struct btrfs_root *root; 8522c30c71bSKent Overstreet int i, ret = 0; 8536dc4f100SMing Lei struct bvec_iter_all iter_all; 854ce3ed71aSChris Mason 855c09abff8SDavid Sterba ASSERT(!bio_flagged(bio, BIO_CLONED)); 8566dc4f100SMing Lei bio_for_each_segment_all(bvec, bio, i, iter_all) { 857ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 85801d58472SDaniel Dressler ret = csum_dirty_buffer(root->fs_info, bvec->bv_page); 85979787eaaSJeff Mahoney if (ret) 86079787eaaSJeff Mahoney break; 861ce3ed71aSChris Mason } 8622c30c71bSKent Overstreet 8634e4cbee9SChristoph Hellwig return errno_to_blk_status(ret); 864ce3ed71aSChris Mason } 865ce3ed71aSChris Mason 866d0ee3934SDavid Sterba static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio, 867eaf25d93SChris Mason u64 bio_offset) 86822c59948SChris Mason { 8698b712842SChris Mason /* 8708b712842SChris Mason * when we're called for a write, we're already in the async 8715443be45SChris Mason * submission context. Just jump into btrfs_map_bio 8728b712842SChris Mason */ 87379787eaaSJeff Mahoney return btree_csum_one_bio(bio); 87422c59948SChris Mason } 87522c59948SChris Mason 87618fdc679SLiu Bo static int check_async_write(struct btrfs_inode *bi) 877de0022b9SJosef Bacik { 8786300463bSLiu Bo if (atomic_read(&bi->sync_writers)) 8796300463bSLiu Bo return 0; 880de0022b9SJosef Bacik #ifdef CONFIG_X86 881bc696ca0SBorislav Petkov if (static_cpu_has(X86_FEATURE_XMM4_2)) 882de0022b9SJosef Bacik return 0; 883de0022b9SJosef Bacik #endif 884de0022b9SJosef Bacik return 1; 885de0022b9SJosef Bacik } 886de0022b9SJosef Bacik 887a56b1c7bSNikolay Borisov static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio, 88850489a57SNikolay Borisov int mirror_num, 88950489a57SNikolay Borisov unsigned long bio_flags) 89044b8bd7eSChris Mason { 8910b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 89218fdc679SLiu Bo int async = check_async_write(BTRFS_I(inode)); 8934e4cbee9SChristoph Hellwig blk_status_t ret; 894cad321adSChris Mason 89537226b21SMike Christie if (bio_op(bio) != REQ_OP_WRITE) { 896cad321adSChris Mason /* 897cad321adSChris Mason * called for a read, do the setup so that checksum validation 898cad321adSChris Mason * can happen in the async kernel threads 899cad321adSChris Mason */ 9000b246afaSJeff Mahoney ret = btrfs_bio_wq_end_io(fs_info, bio, 9010b246afaSJeff Mahoney BTRFS_WQ_ENDIO_METADATA); 9021d4284bdSChris Mason if (ret) 90361891923SStefan Behrens goto out_w_error; 9042ff7e61eSJeff Mahoney ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); 905de0022b9SJosef Bacik } else if (!async) { 906de0022b9SJosef Bacik ret = btree_csum_one_bio(bio); 907de0022b9SJosef Bacik if (ret) 90861891923SStefan Behrens goto out_w_error; 9092ff7e61eSJeff Mahoney ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); 91061891923SStefan Behrens } else { 911cad321adSChris Mason /* 91261891923SStefan Behrens * kthread helpers are used to submit writes so that 91361891923SStefan Behrens * checksumming can happen in parallel across all CPUs 914cad321adSChris Mason */ 915c6100a4bSJosef Bacik ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0, 916e68f2ee7SNikolay Borisov 0, inode, btree_submit_bio_start); 91744b8bd7eSChris Mason } 91844b8bd7eSChris Mason 9194246a0b6SChristoph Hellwig if (ret) 9204246a0b6SChristoph Hellwig goto out_w_error; 9214246a0b6SChristoph Hellwig return 0; 9224246a0b6SChristoph Hellwig 92361891923SStefan Behrens out_w_error: 9244e4cbee9SChristoph Hellwig bio->bi_status = ret; 9254246a0b6SChristoph Hellwig bio_endio(bio); 92661891923SStefan Behrens return ret; 92761891923SStefan Behrens } 92861891923SStefan Behrens 9293dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 930784b4e29SChris Mason static int btree_migratepage(struct address_space *mapping, 931a6bc32b8SMel Gorman struct page *newpage, struct page *page, 932a6bc32b8SMel Gorman enum migrate_mode mode) 933784b4e29SChris Mason { 934784b4e29SChris Mason /* 935784b4e29SChris Mason * we can't safely write a btree page from here, 936784b4e29SChris Mason * we haven't done the locking hook 937784b4e29SChris Mason */ 938784b4e29SChris Mason if (PageDirty(page)) 939784b4e29SChris Mason return -EAGAIN; 940784b4e29SChris Mason /* 941784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 942784b4e29SChris Mason * We must have no buffers or drop them. 943784b4e29SChris Mason */ 944784b4e29SChris Mason if (page_has_private(page) && 945784b4e29SChris Mason !try_to_release_page(page, GFP_KERNEL)) 946784b4e29SChris Mason return -EAGAIN; 947a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 948784b4e29SChris Mason } 9493dd1462eSJan Beulich #endif 950784b4e29SChris Mason 9510da5468fSChris Mason 9520da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 9530da5468fSChris Mason struct writeback_control *wbc) 9540da5468fSChris Mason { 955e2d84521SMiao Xie struct btrfs_fs_info *fs_info; 956e2d84521SMiao Xie int ret; 957e2d84521SMiao Xie 958d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 959448d640bSChris Mason 960448d640bSChris Mason if (wbc->for_kupdate) 961448d640bSChris Mason return 0; 962448d640bSChris Mason 963e2d84521SMiao Xie fs_info = BTRFS_I(mapping->host)->root->fs_info; 964b9473439SChris Mason /* this is a bit racy, but that's ok */ 965d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 966d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 967d814a491SEthan Lien fs_info->dirty_metadata_batch); 968e2d84521SMiao Xie if (ret < 0) 969793955bcSChris Mason return 0; 970793955bcSChris Mason } 9710b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 9720da5468fSChris Mason } 9730da5468fSChris Mason 974b2950863SChristoph Hellwig static int btree_readpage(struct file *file, struct page *page) 9755f39d397SChris Mason { 976d1310b2eSChris Mason struct extent_io_tree *tree; 977d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9788ddc7d9cSJan Schmidt return extent_read_full_page(tree, page, btree_get_extent, 0); 9795f39d397SChris Mason } 9805f39d397SChris Mason 98170dec807SChris Mason static int btree_releasepage(struct page *page, gfp_t gfp_flags) 9825f39d397SChris Mason { 98398509cfcSChris Mason if (PageWriteback(page) || PageDirty(page)) 98498509cfcSChris Mason return 0; 9850c4e538bSDavid Sterba 986f7a52a40SDavid Sterba return try_release_extent_buffer(page); 987d98237b3SChris Mason } 988d98237b3SChris Mason 989d47992f8SLukas Czerner static void btree_invalidatepage(struct page *page, unsigned int offset, 990d47992f8SLukas Czerner unsigned int length) 991d98237b3SChris Mason { 992d1310b2eSChris Mason struct extent_io_tree *tree; 993d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9945f39d397SChris Mason extent_invalidatepage(tree, page, offset); 9955f39d397SChris Mason btree_releasepage(page, GFP_NOFS); 9969ad6b7bcSChris Mason if (PagePrivate(page)) { 997efe120a0SFrank Holton btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info, 998efe120a0SFrank Holton "page private not zero on page %llu", 999efe120a0SFrank Holton (unsigned long long)page_offset(page)); 10009ad6b7bcSChris Mason ClearPagePrivate(page); 10019ad6b7bcSChris Mason set_page_private(page, 0); 100209cbfeafSKirill A. Shutemov put_page(page); 10039ad6b7bcSChris Mason } 1004d98237b3SChris Mason } 1005d98237b3SChris Mason 10060b32f4bbSJosef Bacik static int btree_set_page_dirty(struct page *page) 10070b32f4bbSJosef Bacik { 1008bb146eb2SJosef Bacik #ifdef DEBUG 10090b32f4bbSJosef Bacik struct extent_buffer *eb; 10100b32f4bbSJosef Bacik 10110b32f4bbSJosef Bacik BUG_ON(!PagePrivate(page)); 10120b32f4bbSJosef Bacik eb = (struct extent_buffer *)page->private; 10130b32f4bbSJosef Bacik BUG_ON(!eb); 10140b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 10150b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 10160b32f4bbSJosef Bacik btrfs_assert_tree_locked(eb); 1017bb146eb2SJosef Bacik #endif 10180b32f4bbSJosef Bacik return __set_page_dirty_nobuffers(page); 10190b32f4bbSJosef Bacik } 10200b32f4bbSJosef Bacik 10217f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 1022d98237b3SChris Mason .readpage = btree_readpage, 10230da5468fSChris Mason .writepages = btree_writepages, 10245f39d397SChris Mason .releasepage = btree_releasepage, 10255f39d397SChris Mason .invalidatepage = btree_invalidatepage, 10265a92bc88SChris Mason #ifdef CONFIG_MIGRATION 1027784b4e29SChris Mason .migratepage = btree_migratepage, 10285a92bc88SChris Mason #endif 10290b32f4bbSJosef Bacik .set_page_dirty = btree_set_page_dirty, 1030d98237b3SChris Mason }; 1031123abc88SChris Mason 10322ff7e61eSJeff Mahoney void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr) 1033090d1875SChris Mason { 10345f39d397SChris Mason struct extent_buffer *buf = NULL; 1035537f38f0SNikolay Borisov int ret; 1036090d1875SChris Mason 10372ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 1038c871b0f2SLiu Bo if (IS_ERR(buf)) 10396197d86eSDavid Sterba return; 1040537f38f0SNikolay Borisov 1041c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(buf, WAIT_NONE, 0); 1042537f38f0SNikolay Borisov if (ret < 0) 1043537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 1044537f38f0SNikolay Borisov else 10455f39d397SChris Mason free_extent_buffer(buf); 1046090d1875SChris Mason } 1047090d1875SChris Mason 10482ff7e61eSJeff Mahoney int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr, 1049ab0fff03SArne Jansen int mirror_num, struct extent_buffer **eb) 1050ab0fff03SArne Jansen { 1051ab0fff03SArne Jansen struct extent_buffer *buf = NULL; 1052ab0fff03SArne Jansen int ret; 1053ab0fff03SArne Jansen 10542ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 1055c871b0f2SLiu Bo if (IS_ERR(buf)) 1056ab0fff03SArne Jansen return 0; 1057ab0fff03SArne Jansen 1058ab0fff03SArne Jansen set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags); 1059ab0fff03SArne Jansen 1060c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(buf, WAIT_PAGE_LOCK, mirror_num); 1061ab0fff03SArne Jansen if (ret) { 1062537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 1063ab0fff03SArne Jansen return ret; 1064ab0fff03SArne Jansen } 1065ab0fff03SArne Jansen 1066ab0fff03SArne Jansen if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) { 1067537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 1068ab0fff03SArne Jansen return -EIO; 10690b32f4bbSJosef Bacik } else if (extent_buffer_uptodate(buf)) { 1070ab0fff03SArne Jansen *eb = buf; 1071ab0fff03SArne Jansen } else { 1072ab0fff03SArne Jansen free_extent_buffer(buf); 1073ab0fff03SArne Jansen } 1074ab0fff03SArne Jansen return 0; 1075ab0fff03SArne Jansen } 1076ab0fff03SArne Jansen 10772ff7e61eSJeff Mahoney struct extent_buffer *btrfs_find_create_tree_block( 10782ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info, 10790308af44SDavid Sterba u64 bytenr) 10800999df54SChris Mason { 10810b246afaSJeff Mahoney if (btrfs_is_testing(fs_info)) 10820b246afaSJeff Mahoney return alloc_test_extent_buffer(fs_info, bytenr); 10830b246afaSJeff Mahoney return alloc_extent_buffer(fs_info, bytenr); 10840999df54SChris Mason } 10850999df54SChris Mason 1086581c1760SQu Wenruo /* 1087581c1760SQu Wenruo * Read tree block at logical address @bytenr and do variant basic but critical 1088581c1760SQu Wenruo * verification. 1089581c1760SQu Wenruo * 1090581c1760SQu Wenruo * @parent_transid: expected transid of this tree block, skip check if 0 1091581c1760SQu Wenruo * @level: expected level, mandatory check 1092581c1760SQu Wenruo * @first_key: expected key in slot 0, skip check if NULL 1093581c1760SQu Wenruo */ 10942ff7e61eSJeff Mahoney struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, 1095581c1760SQu Wenruo u64 parent_transid, int level, 1096581c1760SQu Wenruo struct btrfs_key *first_key) 1097e20d96d6SChris Mason { 10985f39d397SChris Mason struct extent_buffer *buf = NULL; 109919c00ddcSChris Mason int ret; 110019c00ddcSChris Mason 11012ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 1102c871b0f2SLiu Bo if (IS_ERR(buf)) 1103c871b0f2SLiu Bo return buf; 1104e4204dedSChris Mason 11055ab12d1fSDavid Sterba ret = btree_read_extent_buffer_pages(buf, parent_transid, 1106581c1760SQu Wenruo level, first_key); 11070f0fe8f7SFilipe David Borba Manana if (ret) { 1108537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 110964c043deSLiu Bo return ERR_PTR(ret); 11100f0fe8f7SFilipe David Borba Manana } 11115f39d397SChris Mason return buf; 1112ce9adaa5SChris Mason 1113eb60ceacSChris Mason } 1114eb60ceacSChris Mason 11156a884d7dSDavid Sterba void btrfs_clean_tree_block(struct extent_buffer *buf) 1116ed2ff2cbSChris Mason { 11176a884d7dSDavid Sterba struct btrfs_fs_info *fs_info = buf->fs_info; 111855c69072SChris Mason if (btrfs_header_generation(buf) == 1119e2d84521SMiao Xie fs_info->running_transaction->transid) { 1120b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1121b4ce94deSChris Mason 1122b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1123104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 1124e2d84521SMiao Xie -buf->len, 1125e2d84521SMiao Xie fs_info->dirty_metadata_batch); 1126b9473439SChris Mason /* ugh, clear_extent_buffer_dirty needs to lock the page */ 11278bead258SDavid Sterba btrfs_set_lock_blocking_write(buf); 11280b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 1129925baeddSChris Mason } 11305f39d397SChris Mason } 1131ed7b63ebSJosef Bacik } 11325f39d397SChris Mason 11338257b2dcSMiao Xie static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void) 11348257b2dcSMiao Xie { 11358257b2dcSMiao Xie struct btrfs_subvolume_writers *writers; 11368257b2dcSMiao Xie int ret; 11378257b2dcSMiao Xie 11388257b2dcSMiao Xie writers = kmalloc(sizeof(*writers), GFP_NOFS); 11398257b2dcSMiao Xie if (!writers) 11408257b2dcSMiao Xie return ERR_PTR(-ENOMEM); 11418257b2dcSMiao Xie 11428a5a916dSJeff Mahoney ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS); 11438257b2dcSMiao Xie if (ret < 0) { 11448257b2dcSMiao Xie kfree(writers); 11458257b2dcSMiao Xie return ERR_PTR(ret); 11468257b2dcSMiao Xie } 11478257b2dcSMiao Xie 11488257b2dcSMiao Xie init_waitqueue_head(&writers->wait); 11498257b2dcSMiao Xie return writers; 11508257b2dcSMiao Xie } 11518257b2dcSMiao Xie 11528257b2dcSMiao Xie static void 11538257b2dcSMiao Xie btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers) 11548257b2dcSMiao Xie { 11558257b2dcSMiao Xie percpu_counter_destroy(&writers->counter); 11568257b2dcSMiao Xie kfree(writers); 11578257b2dcSMiao Xie } 11588257b2dcSMiao Xie 1159da17066cSJeff Mahoney static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, 1160e20d96d6SChris Mason u64 objectid) 1161d97e63b6SChris Mason { 11627c0260eeSJeff Mahoney bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); 1163cfaa7295SChris Mason root->node = NULL; 1164a28ec197SChris Mason root->commit_root = NULL; 116527cdeb70SMiao Xie root->state = 0; 1166d68fc57bSYan, Zheng root->orphan_cleanup_state = 0; 11670b86a832SChris Mason 11680f7d52f4SChris Mason root->last_trans = 0; 116913a8a7c8SYan, Zheng root->highest_objectid = 0; 1170eb73c1b7SMiao Xie root->nr_delalloc_inodes = 0; 1171199c2a9cSMiao Xie root->nr_ordered_extents = 0; 11726bef4d31SEric Paris root->inode_tree = RB_ROOT; 117316cdcec7SMiao Xie INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1174f0486c68SYan, Zheng root->block_rsv = NULL; 11750b86a832SChris Mason 11760b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 11775d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 1178eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_inodes); 1179eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_root); 1180199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_extents); 1181199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_root); 1182d2311e69SQu Wenruo INIT_LIST_HEAD(&root->reloc_dirty_list); 11832ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 11842ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 11855d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1186eb73c1b7SMiao Xie spin_lock_init(&root->delalloc_lock); 1187199c2a9cSMiao Xie spin_lock_init(&root->ordered_extent_lock); 1188f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 11892ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 11902ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 11918287475aSQu Wenruo spin_lock_init(&root->qgroup_meta_rsv_lock); 1192a2135011SChris Mason mutex_init(&root->objectid_mutex); 1193e02119d5SChris Mason mutex_init(&root->log_mutex); 119431f3d255SMiao Xie mutex_init(&root->ordered_extent_mutex); 1195573bfb72SMiao Xie mutex_init(&root->delalloc_mutex); 11967237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 11977237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 11987237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 11998b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[0]); 12008b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[1]); 12017237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 12027237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 12037237f183SYan Zheng atomic_set(&root->log_writers, 0); 12042ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 12050700cea7SElena Reshetova refcount_set(&root->refs, 1); 1206ea14b57fSDavid Sterba atomic_set(&root->will_be_snapshotted, 0); 12078ecebf4dSRobbie Ko atomic_set(&root->snapshot_force_cow, 0); 1208eede2bf3SOmar Sandoval atomic_set(&root->nr_swapfiles, 0); 12097237f183SYan Zheng root->log_transid = 0; 1210d1433debSMiao Xie root->log_transid_committed = -1; 1211257c62e1SChris Mason root->last_log_commit = 0; 12127c0260eeSJeff Mahoney if (!dummy) 121343eb5f29SQu Wenruo extent_io_tree_init(fs_info, &root->dirty_log_pages, 121443eb5f29SQu Wenruo IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL); 1215017e5369SChris Mason 12163768f368SChris Mason memset(&root->root_key, 0, sizeof(root->root_key)); 12173768f368SChris Mason memset(&root->root_item, 0, sizeof(root->root_item)); 12186702ed49SChris Mason memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 12197c0260eeSJeff Mahoney if (!dummy) 12203f157a2fSChris Mason root->defrag_trans_start = fs_info->generation; 122106ea65a3SJosef Bacik else 122206ea65a3SJosef Bacik root->defrag_trans_start = 0; 12234d775673SChris Mason root->root_key.objectid = objectid; 12240ee5dc67SAl Viro root->anon_dev = 0; 12258ea05e3aSAlexander Block 12265f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 1227370a11b8SQu Wenruo btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks); 12283768f368SChris Mason } 12293768f368SChris Mason 123074e4d827SDavid Sterba static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, 123174e4d827SDavid Sterba gfp_t flags) 12326f07e42eSAl Viro { 123374e4d827SDavid Sterba struct btrfs_root *root = kzalloc(sizeof(*root), flags); 12346f07e42eSAl Viro if (root) 12356f07e42eSAl Viro root->fs_info = fs_info; 12366f07e42eSAl Viro return root; 12376f07e42eSAl Viro } 12386f07e42eSAl Viro 123906ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 124006ea65a3SJosef Bacik /* Should only be used by the testing infrastructure */ 1241da17066cSJeff Mahoney struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) 124206ea65a3SJosef Bacik { 124306ea65a3SJosef Bacik struct btrfs_root *root; 124406ea65a3SJosef Bacik 12457c0260eeSJeff Mahoney if (!fs_info) 12467c0260eeSJeff Mahoney return ERR_PTR(-EINVAL); 12477c0260eeSJeff Mahoney 12487c0260eeSJeff Mahoney root = btrfs_alloc_root(fs_info, GFP_KERNEL); 124906ea65a3SJosef Bacik if (!root) 125006ea65a3SJosef Bacik return ERR_PTR(-ENOMEM); 1251da17066cSJeff Mahoney 1252b9ef22deSFeifei Xu /* We don't use the stripesize in selftest, set it as sectorsize */ 1253da17066cSJeff Mahoney __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID); 1254faa2dbf0SJosef Bacik root->alloc_bytenr = 0; 125506ea65a3SJosef Bacik 125606ea65a3SJosef Bacik return root; 125706ea65a3SJosef Bacik } 125806ea65a3SJosef Bacik #endif 125906ea65a3SJosef Bacik 126020897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 126120897f5cSArne Jansen u64 objectid) 126220897f5cSArne Jansen { 12639b7a2440SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 126420897f5cSArne Jansen struct extent_buffer *leaf; 126520897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 126620897f5cSArne Jansen struct btrfs_root *root; 126720897f5cSArne Jansen struct btrfs_key key; 1268b89f6d1fSFilipe Manana unsigned int nofs_flag; 126920897f5cSArne Jansen int ret = 0; 127033d85fdaSQu Wenruo uuid_le uuid = NULL_UUID_LE; 127120897f5cSArne Jansen 1272b89f6d1fSFilipe Manana /* 1273b89f6d1fSFilipe Manana * We're holding a transaction handle, so use a NOFS memory allocation 1274b89f6d1fSFilipe Manana * context to avoid deadlock if reclaim happens. 1275b89f6d1fSFilipe Manana */ 1276b89f6d1fSFilipe Manana nofs_flag = memalloc_nofs_save(); 127774e4d827SDavid Sterba root = btrfs_alloc_root(fs_info, GFP_KERNEL); 1278b89f6d1fSFilipe Manana memalloc_nofs_restore(nofs_flag); 127920897f5cSArne Jansen if (!root) 128020897f5cSArne Jansen return ERR_PTR(-ENOMEM); 128120897f5cSArne Jansen 1282da17066cSJeff Mahoney __setup_root(root, fs_info, objectid); 128320897f5cSArne Jansen root->root_key.objectid = objectid; 128420897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 128520897f5cSArne Jansen root->root_key.offset = 0; 128620897f5cSArne Jansen 12874d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0); 128820897f5cSArne Jansen if (IS_ERR(leaf)) { 128920897f5cSArne Jansen ret = PTR_ERR(leaf); 12901dd05682STsutomu Itoh leaf = NULL; 129120897f5cSArne Jansen goto fail; 129220897f5cSArne Jansen } 129320897f5cSArne Jansen 129420897f5cSArne Jansen root->node = leaf; 129520897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 129620897f5cSArne Jansen 129720897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 129827cdeb70SMiao Xie set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 129920897f5cSArne Jansen 130020897f5cSArne Jansen root->root_item.flags = 0; 130120897f5cSArne Jansen root->root_item.byte_limit = 0; 130220897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 130320897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 130420897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 130520897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 130620897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 130720897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 130820897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 130933d85fdaSQu Wenruo if (is_fstree(objectid)) 13106463fe58SStefan Behrens uuid_le_gen(&uuid); 13116463fe58SStefan Behrens memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE); 131220897f5cSArne Jansen root->root_item.drop_level = 0; 131320897f5cSArne Jansen 131420897f5cSArne Jansen key.objectid = objectid; 131520897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 131620897f5cSArne Jansen key.offset = 0; 131720897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 131820897f5cSArne Jansen if (ret) 131920897f5cSArne Jansen goto fail; 132020897f5cSArne Jansen 132120897f5cSArne Jansen btrfs_tree_unlock(leaf); 132220897f5cSArne Jansen 132320897f5cSArne Jansen return root; 13241dd05682STsutomu Itoh 13251dd05682STsutomu Itoh fail: 13261dd05682STsutomu Itoh if (leaf) { 13271dd05682STsutomu Itoh btrfs_tree_unlock(leaf); 132859885b39STsutomu Itoh free_extent_buffer(root->commit_root); 13291dd05682STsutomu Itoh free_extent_buffer(leaf); 13301dd05682STsutomu Itoh } 13311dd05682STsutomu Itoh kfree(root); 13321dd05682STsutomu Itoh 13331dd05682STsutomu Itoh return ERR_PTR(ret); 133420897f5cSArne Jansen } 133520897f5cSArne Jansen 13367237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1337e02119d5SChris Mason struct btrfs_fs_info *fs_info) 13380f7d52f4SChris Mason { 13390f7d52f4SChris Mason struct btrfs_root *root; 13407237f183SYan Zheng struct extent_buffer *leaf; 1341e02119d5SChris Mason 134274e4d827SDavid Sterba root = btrfs_alloc_root(fs_info, GFP_NOFS); 1343e02119d5SChris Mason if (!root) 13447237f183SYan Zheng return ERR_PTR(-ENOMEM); 1345e02119d5SChris Mason 1346da17066cSJeff Mahoney __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID); 1347e02119d5SChris Mason 1348e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1349e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1350e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 135127cdeb70SMiao Xie 13527237f183SYan Zheng /* 135327cdeb70SMiao Xie * DON'T set REF_COWS for log trees 135427cdeb70SMiao Xie * 13557237f183SYan Zheng * log trees do not get reference counted because they go away 13567237f183SYan Zheng * before a real commit is actually done. They do store pointers 13577237f183SYan Zheng * to file data extents, and those reference counts still get 13587237f183SYan Zheng * updated (along with back refs to the log tree). 13597237f183SYan Zheng */ 1360e02119d5SChris Mason 13614d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 13624d75f8a9SDavid Sterba NULL, 0, 0, 0); 13637237f183SYan Zheng if (IS_ERR(leaf)) { 13647237f183SYan Zheng kfree(root); 13657237f183SYan Zheng return ERR_CAST(leaf); 13667237f183SYan Zheng } 1367e02119d5SChris Mason 13687237f183SYan Zheng root->node = leaf; 1369e02119d5SChris Mason 1370e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1371e02119d5SChris Mason btrfs_tree_unlock(root->node); 13727237f183SYan Zheng return root; 13737237f183SYan Zheng } 13747237f183SYan Zheng 13757237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 13767237f183SYan Zheng struct btrfs_fs_info *fs_info) 13777237f183SYan Zheng { 13787237f183SYan Zheng struct btrfs_root *log_root; 13797237f183SYan Zheng 13807237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 13817237f183SYan Zheng if (IS_ERR(log_root)) 13827237f183SYan Zheng return PTR_ERR(log_root); 13837237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 13847237f183SYan Zheng fs_info->log_root_tree = log_root; 13857237f183SYan Zheng return 0; 13867237f183SYan Zheng } 13877237f183SYan Zheng 13887237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 13897237f183SYan Zheng struct btrfs_root *root) 13907237f183SYan Zheng { 13910b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13927237f183SYan Zheng struct btrfs_root *log_root; 13937237f183SYan Zheng struct btrfs_inode_item *inode_item; 13947237f183SYan Zheng 13950b246afaSJeff Mahoney log_root = alloc_log_tree(trans, fs_info); 13967237f183SYan Zheng if (IS_ERR(log_root)) 13977237f183SYan Zheng return PTR_ERR(log_root); 13987237f183SYan Zheng 13997237f183SYan Zheng log_root->last_trans = trans->transid; 14007237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 14017237f183SYan Zheng 14027237f183SYan Zheng inode_item = &log_root->root_item.inode; 14033cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 14043cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 14053cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 1406da17066cSJeff Mahoney btrfs_set_stack_inode_nbytes(inode_item, 14070b246afaSJeff Mahoney fs_info->nodesize); 14083cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 14097237f183SYan Zheng 14105d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 14117237f183SYan Zheng 14127237f183SYan Zheng WARN_ON(root->log_root); 14137237f183SYan Zheng root->log_root = log_root; 14147237f183SYan Zheng root->log_transid = 0; 1415d1433debSMiao Xie root->log_transid_committed = -1; 1416257c62e1SChris Mason root->last_log_commit = 0; 1417e02119d5SChris Mason return 0; 1418e02119d5SChris Mason } 1419e02119d5SChris Mason 142035a3621bSStefan Behrens static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 1421cb517eabSMiao Xie struct btrfs_key *key) 1422e02119d5SChris Mason { 1423e02119d5SChris Mason struct btrfs_root *root; 1424e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 14250f7d52f4SChris Mason struct btrfs_path *path; 142684234f3aSYan Zheng u64 generation; 1427cb517eabSMiao Xie int ret; 1428581c1760SQu Wenruo int level; 1429cb517eabSMiao Xie 1430cb517eabSMiao Xie path = btrfs_alloc_path(); 1431cb517eabSMiao Xie if (!path) 1432cb517eabSMiao Xie return ERR_PTR(-ENOMEM); 14330f7d52f4SChris Mason 143474e4d827SDavid Sterba root = btrfs_alloc_root(fs_info, GFP_NOFS); 1435cb517eabSMiao Xie if (!root) { 1436cb517eabSMiao Xie ret = -ENOMEM; 1437cb517eabSMiao Xie goto alloc_fail; 14380f7d52f4SChris Mason } 14390f7d52f4SChris Mason 1440da17066cSJeff Mahoney __setup_root(root, fs_info, key->objectid); 14410f7d52f4SChris Mason 1442cb517eabSMiao Xie ret = btrfs_find_root(tree_root, key, path, 1443cb517eabSMiao Xie &root->root_item, &root->root_key); 14440f7d52f4SChris Mason if (ret) { 144513a8a7c8SYan, Zheng if (ret > 0) 144613a8a7c8SYan, Zheng ret = -ENOENT; 1447cb517eabSMiao Xie goto find_fail; 14480f7d52f4SChris Mason } 144913a8a7c8SYan, Zheng 145084234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1451581c1760SQu Wenruo level = btrfs_root_level(&root->root_item); 14522ff7e61eSJeff Mahoney root->node = read_tree_block(fs_info, 14532ff7e61eSJeff Mahoney btrfs_root_bytenr(&root->root_item), 1454581c1760SQu Wenruo generation, level, NULL); 145564c043deSLiu Bo if (IS_ERR(root->node)) { 145664c043deSLiu Bo ret = PTR_ERR(root->node); 1457cb517eabSMiao Xie goto find_fail; 1458cb517eabSMiao Xie } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1459cb517eabSMiao Xie ret = -EIO; 146064c043deSLiu Bo free_extent_buffer(root->node); 146164c043deSLiu Bo goto find_fail; 1462416bc658SJosef Bacik } 14635d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 146413a8a7c8SYan, Zheng out: 1465cb517eabSMiao Xie btrfs_free_path(path); 1466cb517eabSMiao Xie return root; 1467cb517eabSMiao Xie 1468cb517eabSMiao Xie find_fail: 1469cb517eabSMiao Xie kfree(root); 1470cb517eabSMiao Xie alloc_fail: 1471cb517eabSMiao Xie root = ERR_PTR(ret); 1472cb517eabSMiao Xie goto out; 1473cb517eabSMiao Xie } 1474cb517eabSMiao Xie 1475cb517eabSMiao Xie struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root, 1476cb517eabSMiao Xie struct btrfs_key *location) 1477cb517eabSMiao Xie { 1478cb517eabSMiao Xie struct btrfs_root *root; 1479cb517eabSMiao Xie 1480cb517eabSMiao Xie root = btrfs_read_tree_root(tree_root, location); 1481cb517eabSMiao Xie if (IS_ERR(root)) 1482cb517eabSMiao Xie return root; 1483cb517eabSMiao Xie 1484cb517eabSMiao Xie if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { 148527cdeb70SMiao Xie set_bit(BTRFS_ROOT_REF_COWS, &root->state); 148608fe4db1SLi Zefan btrfs_check_and_init_root_item(&root->root_item); 148708fe4db1SLi Zefan } 148813a8a7c8SYan, Zheng 14895eda7b5eSChris Mason return root; 14905eda7b5eSChris Mason } 14915eda7b5eSChris Mason 1492cb517eabSMiao Xie int btrfs_init_fs_root(struct btrfs_root *root) 1493cb517eabSMiao Xie { 1494cb517eabSMiao Xie int ret; 14958257b2dcSMiao Xie struct btrfs_subvolume_writers *writers; 1496cb517eabSMiao Xie 1497cb517eabSMiao Xie root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); 1498cb517eabSMiao Xie root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), 1499cb517eabSMiao Xie GFP_NOFS); 1500cb517eabSMiao Xie if (!root->free_ino_pinned || !root->free_ino_ctl) { 1501cb517eabSMiao Xie ret = -ENOMEM; 1502cb517eabSMiao Xie goto fail; 1503cb517eabSMiao Xie } 1504cb517eabSMiao Xie 15058257b2dcSMiao Xie writers = btrfs_alloc_subvolume_writers(); 15068257b2dcSMiao Xie if (IS_ERR(writers)) { 15078257b2dcSMiao Xie ret = PTR_ERR(writers); 15088257b2dcSMiao Xie goto fail; 15098257b2dcSMiao Xie } 15108257b2dcSMiao Xie root->subv_writers = writers; 15118257b2dcSMiao Xie 1512cb517eabSMiao Xie btrfs_init_free_ino_ctl(root); 151357cdc8dbSDavid Sterba spin_lock_init(&root->ino_cache_lock); 151457cdc8dbSDavid Sterba init_waitqueue_head(&root->ino_cache_wait); 1515cb517eabSMiao Xie 1516cb517eabSMiao Xie ret = get_anon_bdev(&root->anon_dev); 1517cb517eabSMiao Xie if (ret) 1518876d2cf1SLiu Bo goto fail; 1519f32e48e9SChandan Rajendra 1520f32e48e9SChandan Rajendra mutex_lock(&root->objectid_mutex); 1521f32e48e9SChandan Rajendra ret = btrfs_find_highest_objectid(root, 1522f32e48e9SChandan Rajendra &root->highest_objectid); 1523f32e48e9SChandan Rajendra if (ret) { 1524f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1525876d2cf1SLiu Bo goto fail; 1526f32e48e9SChandan Rajendra } 1527f32e48e9SChandan Rajendra 1528f32e48e9SChandan Rajendra ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 1529f32e48e9SChandan Rajendra 1530f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1531f32e48e9SChandan Rajendra 1532cb517eabSMiao Xie return 0; 1533cb517eabSMiao Xie fail: 153484db5ccfSDavid Sterba /* The caller is responsible to call btrfs_free_fs_root */ 1535cb517eabSMiao Xie return ret; 1536cb517eabSMiao Xie } 1537cb517eabSMiao Xie 153835bbb97fSJeff Mahoney struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1539cb517eabSMiao Xie u64 root_id) 1540cb517eabSMiao Xie { 1541cb517eabSMiao Xie struct btrfs_root *root; 1542cb517eabSMiao Xie 1543cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1544cb517eabSMiao Xie root = radix_tree_lookup(&fs_info->fs_roots_radix, 1545cb517eabSMiao Xie (unsigned long)root_id); 1546cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1547cb517eabSMiao Xie return root; 1548cb517eabSMiao Xie } 1549cb517eabSMiao Xie 1550cb517eabSMiao Xie int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1551cb517eabSMiao Xie struct btrfs_root *root) 1552cb517eabSMiao Xie { 1553cb517eabSMiao Xie int ret; 1554cb517eabSMiao Xie 1555e1860a77SDavid Sterba ret = radix_tree_preload(GFP_NOFS); 1556cb517eabSMiao Xie if (ret) 1557cb517eabSMiao Xie return ret; 1558cb517eabSMiao Xie 1559cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1560cb517eabSMiao Xie ret = radix_tree_insert(&fs_info->fs_roots_radix, 1561cb517eabSMiao Xie (unsigned long)root->root_key.objectid, 1562cb517eabSMiao Xie root); 1563cb517eabSMiao Xie if (ret == 0) 156427cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1565cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1566cb517eabSMiao Xie radix_tree_preload_end(); 1567cb517eabSMiao Xie 1568cb517eabSMiao Xie return ret; 1569cb517eabSMiao Xie } 1570cb517eabSMiao Xie 1571c00869f1SMiao Xie struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 1572c00869f1SMiao Xie struct btrfs_key *location, 1573c00869f1SMiao Xie bool check_ref) 15745eda7b5eSChris Mason { 15755eda7b5eSChris Mason struct btrfs_root *root; 1576381cf658SDavid Sterba struct btrfs_path *path; 15771d4c08e0SDavid Sterba struct btrfs_key key; 15785eda7b5eSChris Mason int ret; 15795eda7b5eSChris Mason 1580edbd8d4eSChris Mason if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 1581edbd8d4eSChris Mason return fs_info->tree_root; 1582edbd8d4eSChris Mason if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID) 1583edbd8d4eSChris Mason return fs_info->extent_root; 15848f18cf13SChris Mason if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID) 15858f18cf13SChris Mason return fs_info->chunk_root; 15868f18cf13SChris Mason if (location->objectid == BTRFS_DEV_TREE_OBJECTID) 15878f18cf13SChris Mason return fs_info->dev_root; 15880403e47eSYan Zheng if (location->objectid == BTRFS_CSUM_TREE_OBJECTID) 15890403e47eSYan Zheng return fs_info->csum_root; 1590bcef60f2SArne Jansen if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID) 1591bcef60f2SArne Jansen return fs_info->quota_root ? fs_info->quota_root : 1592bcef60f2SArne Jansen ERR_PTR(-ENOENT); 1593f7a81ea4SStefan Behrens if (location->objectid == BTRFS_UUID_TREE_OBJECTID) 1594f7a81ea4SStefan Behrens return fs_info->uuid_root ? fs_info->uuid_root : 1595f7a81ea4SStefan Behrens ERR_PTR(-ENOENT); 159670f6d82eSOmar Sandoval if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) 159770f6d82eSOmar Sandoval return fs_info->free_space_root ? fs_info->free_space_root : 159870f6d82eSOmar Sandoval ERR_PTR(-ENOENT); 15994df27c4dSYan, Zheng again: 1600cb517eabSMiao Xie root = btrfs_lookup_fs_root(fs_info, location->objectid); 160148475471SStefan Behrens if (root) { 1602c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) 160348475471SStefan Behrens return ERR_PTR(-ENOENT); 16045eda7b5eSChris Mason return root; 160548475471SStefan Behrens } 16065eda7b5eSChris Mason 1607cb517eabSMiao Xie root = btrfs_read_fs_root(fs_info->tree_root, location); 16085eda7b5eSChris Mason if (IS_ERR(root)) 16095eda7b5eSChris Mason return root; 16103394e160SChris Mason 1611c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1612d68fc57bSYan, Zheng ret = -ENOENT; 1613d68fc57bSYan, Zheng goto fail; 1614d68fc57bSYan, Zheng } 1615d68fc57bSYan, Zheng 1616cb517eabSMiao Xie ret = btrfs_init_fs_root(root); 1617cb517eabSMiao Xie if (ret) 1618cb517eabSMiao Xie goto fail; 1619cb517eabSMiao Xie 1620381cf658SDavid Sterba path = btrfs_alloc_path(); 1621381cf658SDavid Sterba if (!path) { 1622381cf658SDavid Sterba ret = -ENOMEM; 1623381cf658SDavid Sterba goto fail; 1624381cf658SDavid Sterba } 16251d4c08e0SDavid Sterba key.objectid = BTRFS_ORPHAN_OBJECTID; 16261d4c08e0SDavid Sterba key.type = BTRFS_ORPHAN_ITEM_KEY; 16271d4c08e0SDavid Sterba key.offset = location->objectid; 16281d4c08e0SDavid Sterba 16291d4c08e0SDavid Sterba ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1630381cf658SDavid Sterba btrfs_free_path(path); 1631d68fc57bSYan, Zheng if (ret < 0) 1632d68fc57bSYan, Zheng goto fail; 1633d68fc57bSYan, Zheng if (ret == 0) 163427cdeb70SMiao Xie set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1635d68fc57bSYan, Zheng 1636cb517eabSMiao Xie ret = btrfs_insert_fs_root(fs_info, root); 16370f7d52f4SChris Mason if (ret) { 16384df27c4dSYan, Zheng if (ret == -EEXIST) { 163984db5ccfSDavid Sterba btrfs_free_fs_root(root); 16404df27c4dSYan, Zheng goto again; 16410f7d52f4SChris Mason } 16424df27c4dSYan, Zheng goto fail; 16434df27c4dSYan, Zheng } 1644edbd8d4eSChris Mason return root; 16454df27c4dSYan, Zheng fail: 164684db5ccfSDavid Sterba btrfs_free_fs_root(root); 16474df27c4dSYan, Zheng return ERR_PTR(ret); 1648edbd8d4eSChris Mason } 1649edbd8d4eSChris Mason 165004160088SChris Mason static int btrfs_congested_fn(void *congested_data, int bdi_bits) 165104160088SChris Mason { 165204160088SChris Mason struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; 165304160088SChris Mason int ret = 0; 165404160088SChris Mason struct btrfs_device *device; 165504160088SChris Mason struct backing_dev_info *bdi; 1656b7967db7SChris Mason 16571f78160cSXiao Guangrong rcu_read_lock(); 16581f78160cSXiao Guangrong list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) { 1659dfe25020SChris Mason if (!device->bdev) 1660dfe25020SChris Mason continue; 1661efa7c9f9SJan Kara bdi = device->bdev->bd_bdi; 1662ff9ea323STejun Heo if (bdi_congested(bdi, bdi_bits)) { 166304160088SChris Mason ret = 1; 166404160088SChris Mason break; 166504160088SChris Mason } 166604160088SChris Mason } 16671f78160cSXiao Guangrong rcu_read_unlock(); 166804160088SChris Mason return ret; 166904160088SChris Mason } 167004160088SChris Mason 16718b712842SChris Mason /* 16728b712842SChris Mason * called by the kthread helper functions to finally call the bio end_io 16738b712842SChris Mason * functions. This is where read checksum verification actually happens 16748b712842SChris Mason */ 16758b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work) 1676ce9adaa5SChris Mason { 1677ce9adaa5SChris Mason struct bio *bio; 167897eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 1679ce9adaa5SChris Mason 168097eb6b69SDavid Sterba end_io_wq = container_of(work, struct btrfs_end_io_wq, work); 1681ce9adaa5SChris Mason bio = end_io_wq->bio; 16828b712842SChris Mason 16834e4cbee9SChristoph Hellwig bio->bi_status = end_io_wq->status; 1684ce9adaa5SChris Mason bio->bi_private = end_io_wq->private; 1685ce9adaa5SChris Mason bio->bi_end_io = end_io_wq->end_io; 168697eb6b69SDavid Sterba kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq); 16874246a0b6SChristoph Hellwig bio_endio(bio); 1688ce9adaa5SChris Mason } 168944b8bd7eSChris Mason 1690a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1691a74a4b97SChris Mason { 1692a74a4b97SChris Mason struct btrfs_root *root = arg; 16930b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1694d0278245SMiao Xie int again; 1695a74a4b97SChris Mason 1696d6fd0ae2SOmar Sandoval while (1) { 1697d0278245SMiao Xie again = 0; 16989d1a2a3aSDavid Sterba 1699fd340d0fSJosef Bacik set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1700fd340d0fSJosef Bacik 1701d0278245SMiao Xie /* Make the cleaner go to sleep early. */ 17022ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) 1703d0278245SMiao Xie goto sleep; 1704d0278245SMiao Xie 170590c711abSZygo Blaxell /* 170690c711abSZygo Blaxell * Do not do anything if we might cause open_ctree() to block 170790c711abSZygo Blaxell * before we have finished mounting the filesystem. 170890c711abSZygo Blaxell */ 17090b246afaSJeff Mahoney if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 171090c711abSZygo Blaxell goto sleep; 171190c711abSZygo Blaxell 17120b246afaSJeff Mahoney if (!mutex_trylock(&fs_info->cleaner_mutex)) 1713d0278245SMiao Xie goto sleep; 1714d0278245SMiao Xie 1715dc7f370cSMiao Xie /* 1716dc7f370cSMiao Xie * Avoid the problem that we change the status of the fs 1717dc7f370cSMiao Xie * during the above check and trylock. 1718dc7f370cSMiao Xie */ 17192ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) { 17200b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1721dc7f370cSMiao Xie goto sleep; 1722dc7f370cSMiao Xie } 1723dc7f370cSMiao Xie 17242ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 1725c2d6cb16SFilipe Manana 17269d1a2a3aSDavid Sterba again = btrfs_clean_one_deleted_snapshot(root); 17270b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1728a74a4b97SChris Mason 1729d0278245SMiao Xie /* 173005323cd1SMiao Xie * The defragger has dealt with the R/O remount and umount, 173105323cd1SMiao Xie * needn't do anything special here. 1732d0278245SMiao Xie */ 17330b246afaSJeff Mahoney btrfs_run_defrag_inodes(fs_info); 173467c5e7d4SFilipe Manana 173567c5e7d4SFilipe Manana /* 173667c5e7d4SFilipe Manana * Acquires fs_info->delete_unused_bgs_mutex to avoid racing 173767c5e7d4SFilipe Manana * with relocation (btrfs_relocate_chunk) and relocation 173867c5e7d4SFilipe Manana * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 173967c5e7d4SFilipe Manana * after acquiring fs_info->delete_unused_bgs_mutex. So we 174067c5e7d4SFilipe Manana * can't hold, nor need to, fs_info->cleaner_mutex when deleting 174167c5e7d4SFilipe Manana * unused block groups. 174267c5e7d4SFilipe Manana */ 17430b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 1744d0278245SMiao Xie sleep: 1745fd340d0fSJosef Bacik clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1746d6fd0ae2SOmar Sandoval if (kthread_should_park()) 1747d6fd0ae2SOmar Sandoval kthread_parkme(); 1748d6fd0ae2SOmar Sandoval if (kthread_should_stop()) 1749d6fd0ae2SOmar Sandoval return 0; 1750838fe188SJiri Kosina if (!again) { 1751a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 1752a74a4b97SChris Mason schedule(); 1753a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1754a74a4b97SChris Mason } 1755da288d28SFilipe Manana } 1756a74a4b97SChris Mason } 1757a74a4b97SChris Mason 1758a74a4b97SChris Mason static int transaction_kthread(void *arg) 1759a74a4b97SChris Mason { 1760a74a4b97SChris Mason struct btrfs_root *root = arg; 17610b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1762a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1763a74a4b97SChris Mason struct btrfs_transaction *cur; 17648929ecfaSYan, Zheng u64 transid; 1765a944442cSAllen Pais time64_t now; 1766a74a4b97SChris Mason unsigned long delay; 1767914b2007SJan Kara bool cannot_commit; 1768a74a4b97SChris Mason 1769a74a4b97SChris Mason do { 1770914b2007SJan Kara cannot_commit = false; 17710b246afaSJeff Mahoney delay = HZ * fs_info->commit_interval; 17720b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 1773a74a4b97SChris Mason 17740b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 17750b246afaSJeff Mahoney cur = fs_info->running_transaction; 1776a74a4b97SChris Mason if (!cur) { 17770b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1778a74a4b97SChris Mason goto sleep; 1779a74a4b97SChris Mason } 178031153d81SYan Zheng 1781afd48513SArnd Bergmann now = ktime_get_seconds(); 17824a9d8bdeSMiao Xie if (cur->state < TRANS_STATE_BLOCKED && 1783a514d638SQu Wenruo !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) && 17848b87dc17SDavid Sterba (now < cur->start_time || 17850b246afaSJeff Mahoney now - cur->start_time < fs_info->commit_interval)) { 17860b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1787a74a4b97SChris Mason delay = HZ * 5; 1788a74a4b97SChris Mason goto sleep; 1789a74a4b97SChris Mason } 17908929ecfaSYan, Zheng transid = cur->transid; 17910b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 179256bec294SChris Mason 179379787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 1794354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 1795914b2007SJan Kara if (IS_ERR(trans)) { 1796354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 1797914b2007SJan Kara cannot_commit = true; 179879787eaaSJeff Mahoney goto sleep; 1799914b2007SJan Kara } 18008929ecfaSYan, Zheng if (transid == trans->transid) { 18013a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 18028929ecfaSYan, Zheng } else { 18033a45bb20SJeff Mahoney btrfs_end_transaction(trans); 18048929ecfaSYan, Zheng } 1805a74a4b97SChris Mason sleep: 18060b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 18070b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 1808a74a4b97SChris Mason 18094e121c06SJosef Bacik if (unlikely(test_bit(BTRFS_FS_STATE_ERROR, 18100b246afaSJeff Mahoney &fs_info->fs_state))) 18112ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 18128929ecfaSYan, Zheng if (!kthread_should_stop() && 18130b246afaSJeff Mahoney (!btrfs_transaction_blocked(fs_info) || 1814914b2007SJan Kara cannot_commit)) 1815bc5511d0SNikolay Borisov schedule_timeout_interruptible(delay); 1816a74a4b97SChris Mason } while (!kthread_should_stop()); 1817a74a4b97SChris Mason return 0; 1818a74a4b97SChris Mason } 1819a74a4b97SChris Mason 1820af31f5e5SChris Mason /* 1821af31f5e5SChris Mason * this will find the highest generation in the array of 1822af31f5e5SChris Mason * root backups. The index of the highest array is returned, 1823af31f5e5SChris Mason * or -1 if we can't find anything. 1824af31f5e5SChris Mason * 1825af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1826af31f5e5SChris Mason * generation of the latest root in the array with the generation 1827af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1828af31f5e5SChris Mason */ 1829af31f5e5SChris Mason static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen) 1830af31f5e5SChris Mason { 1831af31f5e5SChris Mason u64 cur; 1832af31f5e5SChris Mason int newest_index = -1; 1833af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1834af31f5e5SChris Mason int i; 1835af31f5e5SChris Mason 1836af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1837af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1838af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1839af31f5e5SChris Mason if (cur == newest_gen) 1840af31f5e5SChris Mason newest_index = i; 1841af31f5e5SChris Mason } 1842af31f5e5SChris Mason 1843af31f5e5SChris Mason /* check to see if we actually wrapped around */ 1844af31f5e5SChris Mason if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) { 1845af31f5e5SChris Mason root_backup = info->super_copy->super_roots; 1846af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1847af31f5e5SChris Mason if (cur == newest_gen) 1848af31f5e5SChris Mason newest_index = 0; 1849af31f5e5SChris Mason } 1850af31f5e5SChris Mason return newest_index; 1851af31f5e5SChris Mason } 1852af31f5e5SChris Mason 1853af31f5e5SChris Mason 1854af31f5e5SChris Mason /* 1855af31f5e5SChris Mason * find the oldest backup so we know where to store new entries 1856af31f5e5SChris Mason * in the backup array. This will set the backup_root_index 1857af31f5e5SChris Mason * field in the fs_info struct 1858af31f5e5SChris Mason */ 1859af31f5e5SChris Mason static void find_oldest_super_backup(struct btrfs_fs_info *info, 1860af31f5e5SChris Mason u64 newest_gen) 1861af31f5e5SChris Mason { 1862af31f5e5SChris Mason int newest_index = -1; 1863af31f5e5SChris Mason 1864af31f5e5SChris Mason newest_index = find_newest_super_backup(info, newest_gen); 1865af31f5e5SChris Mason /* if there was garbage in there, just move along */ 1866af31f5e5SChris Mason if (newest_index == -1) { 1867af31f5e5SChris Mason info->backup_root_index = 0; 1868af31f5e5SChris Mason } else { 1869af31f5e5SChris Mason info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS; 1870af31f5e5SChris Mason } 1871af31f5e5SChris Mason } 1872af31f5e5SChris Mason 1873af31f5e5SChris Mason /* 1874af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1875af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1876af31f5e5SChris Mason * done 1877af31f5e5SChris Mason */ 1878af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1879af31f5e5SChris Mason { 1880af31f5e5SChris Mason int next_backup; 1881af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1882af31f5e5SChris Mason int last_backup; 1883af31f5e5SChris Mason 1884af31f5e5SChris Mason next_backup = info->backup_root_index; 1885af31f5e5SChris Mason last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) % 1886af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 1887af31f5e5SChris Mason 1888af31f5e5SChris Mason /* 1889af31f5e5SChris Mason * just overwrite the last backup if we're at the same generation 1890af31f5e5SChris Mason * this happens only at umount 1891af31f5e5SChris Mason */ 1892af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + last_backup; 1893af31f5e5SChris Mason if (btrfs_backup_tree_root_gen(root_backup) == 1894af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)) 1895af31f5e5SChris Mason next_backup = last_backup; 1896af31f5e5SChris Mason 1897af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1898af31f5e5SChris Mason 1899af31f5e5SChris Mason /* 1900af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1901af31f5e5SChris Mason * regardless of which ones we use today 1902af31f5e5SChris Mason */ 1903af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1904af31f5e5SChris Mason 1905af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1906af31f5e5SChris Mason 1907af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1908af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1909af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1910af31f5e5SChris Mason 1911af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1912af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1913af31f5e5SChris Mason 1914af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1915af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1916af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 1917af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 1918af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 1919af31f5e5SChris Mason 1920af31f5e5SChris Mason btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); 1921af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 1922af31f5e5SChris Mason btrfs_header_generation(info->extent_root->node)); 1923af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 1924af31f5e5SChris Mason btrfs_header_level(info->extent_root->node)); 1925af31f5e5SChris Mason 19267c7e82a7SChris Mason /* 19277c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 19287c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 19297c7e82a7SChris Mason */ 19307c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 19317c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 19327c7e82a7SChris Mason info->fs_root->node->start); 1933af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 1934af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 1935af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 1936af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 19377c7e82a7SChris Mason } 1938af31f5e5SChris Mason 1939af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 1940af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 1941af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 1942af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 1943af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 1944af31f5e5SChris Mason 1945af31f5e5SChris Mason btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); 1946af31f5e5SChris Mason btrfs_set_backup_csum_root_gen(root_backup, 1947af31f5e5SChris Mason btrfs_header_generation(info->csum_root->node)); 1948af31f5e5SChris Mason btrfs_set_backup_csum_root_level(root_backup, 1949af31f5e5SChris Mason btrfs_header_level(info->csum_root->node)); 1950af31f5e5SChris Mason 1951af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 1952af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 1953af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 1954af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 1955af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 1956af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 1957af31f5e5SChris Mason 1958af31f5e5SChris Mason /* 1959af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 1960af31f5e5SChris Mason * for the next commit 1961af31f5e5SChris Mason */ 1962af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 1963af31f5e5SChris Mason &info->super_for_commit->super_roots, 1964af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 1965af31f5e5SChris Mason } 1966af31f5e5SChris Mason 1967af31f5e5SChris Mason /* 1968af31f5e5SChris Mason * this copies info out of the root backup array and back into 1969af31f5e5SChris Mason * the in-memory super block. It is meant to help iterate through 1970af31f5e5SChris Mason * the array, so you send it the number of backups you've already 1971af31f5e5SChris Mason * tried and the last backup index you used. 1972af31f5e5SChris Mason * 1973af31f5e5SChris Mason * this returns -1 when it has tried all the backups 1974af31f5e5SChris Mason */ 1975af31f5e5SChris Mason static noinline int next_root_backup(struct btrfs_fs_info *info, 1976af31f5e5SChris Mason struct btrfs_super_block *super, 1977af31f5e5SChris Mason int *num_backups_tried, int *backup_index) 1978af31f5e5SChris Mason { 1979af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1980af31f5e5SChris Mason int newest = *backup_index; 1981af31f5e5SChris Mason 1982af31f5e5SChris Mason if (*num_backups_tried == 0) { 1983af31f5e5SChris Mason u64 gen = btrfs_super_generation(super); 1984af31f5e5SChris Mason 1985af31f5e5SChris Mason newest = find_newest_super_backup(info, gen); 1986af31f5e5SChris Mason if (newest == -1) 1987af31f5e5SChris Mason return -1; 1988af31f5e5SChris Mason 1989af31f5e5SChris Mason *backup_index = newest; 1990af31f5e5SChris Mason *num_backups_tried = 1; 1991af31f5e5SChris Mason } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) { 1992af31f5e5SChris Mason /* we've tried all the backups, all done */ 1993af31f5e5SChris Mason return -1; 1994af31f5e5SChris Mason } else { 1995af31f5e5SChris Mason /* jump to the next oldest backup */ 1996af31f5e5SChris Mason newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) % 1997af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 1998af31f5e5SChris Mason *backup_index = newest; 1999af31f5e5SChris Mason *num_backups_tried += 1; 2000af31f5e5SChris Mason } 2001af31f5e5SChris Mason root_backup = super->super_roots + newest; 2002af31f5e5SChris Mason 2003af31f5e5SChris Mason btrfs_set_super_generation(super, 2004af31f5e5SChris Mason btrfs_backup_tree_root_gen(root_backup)); 2005af31f5e5SChris Mason btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 2006af31f5e5SChris Mason btrfs_set_super_root_level(super, 2007af31f5e5SChris Mason btrfs_backup_tree_root_level(root_backup)); 2008af31f5e5SChris Mason btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 2009af31f5e5SChris Mason 2010af31f5e5SChris Mason /* 2011af31f5e5SChris Mason * fixme: the total bytes and num_devices need to match or we should 2012af31f5e5SChris Mason * need a fsck 2013af31f5e5SChris Mason */ 2014af31f5e5SChris Mason btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 2015af31f5e5SChris Mason btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 2016af31f5e5SChris Mason return 0; 2017af31f5e5SChris Mason } 2018af31f5e5SChris Mason 20197abadb64SLiu Bo /* helper to cleanup workers */ 20207abadb64SLiu Bo static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 20217abadb64SLiu Bo { 2022dc6e3209SQu Wenruo btrfs_destroy_workqueue(fs_info->fixup_workers); 2023afe3d242SQu Wenruo btrfs_destroy_workqueue(fs_info->delalloc_workers); 20245cdc7ad3SQu Wenruo btrfs_destroy_workqueue(fs_info->workers); 2025fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_workers); 2026fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_raid56_workers); 20278b110e39SMiao Xie btrfs_destroy_workqueue(fs_info->endio_repair_workers); 2028d05a33acSQu Wenruo btrfs_destroy_workqueue(fs_info->rmw_workers); 2029fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_write_workers); 2030fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 2031a8c93d4eSQu Wenruo btrfs_destroy_workqueue(fs_info->submit_workers); 20325b3bc44eSQu Wenruo btrfs_destroy_workqueue(fs_info->delayed_workers); 2033e66f0bb1SQu Wenruo btrfs_destroy_workqueue(fs_info->caching_workers); 2034736cfa15SQu Wenruo btrfs_destroy_workqueue(fs_info->readahead_workers); 2035a44903abSQu Wenruo btrfs_destroy_workqueue(fs_info->flush_workers); 2036fc97fab0SQu Wenruo btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2037a79b7d4bSChris Mason btrfs_destroy_workqueue(fs_info->extent_workers); 2038a9b9477dSFilipe Manana /* 2039a9b9477dSFilipe Manana * Now that all other work queues are destroyed, we can safely destroy 2040a9b9477dSFilipe Manana * the queues used for metadata I/O, since tasks from those other work 2041a9b9477dSFilipe Manana * queues can do metadata I/O operations. 2042a9b9477dSFilipe Manana */ 2043a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_workers); 2044a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_write_workers); 20457abadb64SLiu Bo } 20467abadb64SLiu Bo 20472e9f5954SRashika static void free_root_extent_buffers(struct btrfs_root *root) 20482e9f5954SRashika { 20492e9f5954SRashika if (root) { 20502e9f5954SRashika free_extent_buffer(root->node); 20512e9f5954SRashika free_extent_buffer(root->commit_root); 20522e9f5954SRashika root->node = NULL; 20532e9f5954SRashika root->commit_root = NULL; 20542e9f5954SRashika } 20552e9f5954SRashika } 20562e9f5954SRashika 2057af31f5e5SChris Mason /* helper to cleanup tree roots */ 2058af31f5e5SChris Mason static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root) 2059af31f5e5SChris Mason { 20602e9f5954SRashika free_root_extent_buffers(info->tree_root); 2061655b09feSJosef Bacik 20622e9f5954SRashika free_root_extent_buffers(info->dev_root); 20632e9f5954SRashika free_root_extent_buffers(info->extent_root); 20642e9f5954SRashika free_root_extent_buffers(info->csum_root); 20652e9f5954SRashika free_root_extent_buffers(info->quota_root); 20662e9f5954SRashika free_root_extent_buffers(info->uuid_root); 20672e9f5954SRashika if (chunk_root) 20682e9f5954SRashika free_root_extent_buffers(info->chunk_root); 206970f6d82eSOmar Sandoval free_root_extent_buffers(info->free_space_root); 2070af31f5e5SChris Mason } 2071af31f5e5SChris Mason 2072faa2dbf0SJosef Bacik void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2073171f6537SJosef Bacik { 2074171f6537SJosef Bacik int ret; 2075171f6537SJosef Bacik struct btrfs_root *gang[8]; 2076171f6537SJosef Bacik int i; 2077171f6537SJosef Bacik 2078171f6537SJosef Bacik while (!list_empty(&fs_info->dead_roots)) { 2079171f6537SJosef Bacik gang[0] = list_entry(fs_info->dead_roots.next, 2080171f6537SJosef Bacik struct btrfs_root, root_list); 2081171f6537SJosef Bacik list_del(&gang[0]->root_list); 2082171f6537SJosef Bacik 208327cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) { 2084cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[0]); 2085171f6537SJosef Bacik } else { 2086171f6537SJosef Bacik free_extent_buffer(gang[0]->node); 2087171f6537SJosef Bacik free_extent_buffer(gang[0]->commit_root); 2088b0feb9d9SMiao Xie btrfs_put_fs_root(gang[0]); 2089171f6537SJosef Bacik } 2090171f6537SJosef Bacik } 2091171f6537SJosef Bacik 2092171f6537SJosef Bacik while (1) { 2093171f6537SJosef Bacik ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2094171f6537SJosef Bacik (void **)gang, 0, 2095171f6537SJosef Bacik ARRAY_SIZE(gang)); 2096171f6537SJosef Bacik if (!ret) 2097171f6537SJosef Bacik break; 2098171f6537SJosef Bacik for (i = 0; i < ret; i++) 2099cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2100171f6537SJosef Bacik } 21011a4319ccSLiu Bo 21021a4319ccSLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 21031a4319ccSLiu Bo btrfs_free_log_root_tree(NULL, fs_info); 21042ff7e61eSJeff Mahoney btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents); 21051a4319ccSLiu Bo } 2106171f6537SJosef Bacik } 2107af31f5e5SChris Mason 2108638aa7edSEric Sandeen static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2109638aa7edSEric Sandeen { 2110638aa7edSEric Sandeen mutex_init(&fs_info->scrub_lock); 2111638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_running, 0); 2112638aa7edSEric Sandeen atomic_set(&fs_info->scrub_pause_req, 0); 2113638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_paused, 0); 2114638aa7edSEric Sandeen atomic_set(&fs_info->scrub_cancel_req, 0); 2115638aa7edSEric Sandeen init_waitqueue_head(&fs_info->scrub_pause_wait); 2116ff09c4caSAnand Jain refcount_set(&fs_info->scrub_workers_refcnt, 0); 2117638aa7edSEric Sandeen } 2118638aa7edSEric Sandeen 2119779a65a4SEric Sandeen static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2120779a65a4SEric Sandeen { 2121779a65a4SEric Sandeen spin_lock_init(&fs_info->balance_lock); 2122779a65a4SEric Sandeen mutex_init(&fs_info->balance_mutex); 2123779a65a4SEric Sandeen atomic_set(&fs_info->balance_pause_req, 0); 2124779a65a4SEric Sandeen atomic_set(&fs_info->balance_cancel_req, 0); 2125779a65a4SEric Sandeen fs_info->balance_ctl = NULL; 2126779a65a4SEric Sandeen init_waitqueue_head(&fs_info->balance_wait_q); 2127779a65a4SEric Sandeen } 2128779a65a4SEric Sandeen 21296bccf3abSJeff Mahoney static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info) 2130f37938e0SEric Sandeen { 21312ff7e61eSJeff Mahoney struct inode *inode = fs_info->btree_inode; 21322ff7e61eSJeff Mahoney 21332ff7e61eSJeff Mahoney inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 21342ff7e61eSJeff Mahoney set_nlink(inode, 1); 2135f37938e0SEric Sandeen /* 2136f37938e0SEric Sandeen * we set the i_size on the btree inode to the max possible int. 2137f37938e0SEric Sandeen * the real end of the address space is determined by all of 2138f37938e0SEric Sandeen * the devices in the system 2139f37938e0SEric Sandeen */ 21402ff7e61eSJeff Mahoney inode->i_size = OFFSET_MAX; 21412ff7e61eSJeff Mahoney inode->i_mapping->a_ops = &btree_aops; 2142f37938e0SEric Sandeen 21432ff7e61eSJeff Mahoney RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 214443eb5f29SQu Wenruo extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree, 214543eb5f29SQu Wenruo IO_TREE_INODE_IO, inode); 21467b439738SDavid Sterba BTRFS_I(inode)->io_tree.track_uptodate = false; 21472ff7e61eSJeff Mahoney extent_map_tree_init(&BTRFS_I(inode)->extent_tree); 2148f37938e0SEric Sandeen 21492ff7e61eSJeff Mahoney BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops; 2150f37938e0SEric Sandeen 21512ff7e61eSJeff Mahoney BTRFS_I(inode)->root = fs_info->tree_root; 21522ff7e61eSJeff Mahoney memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key)); 21532ff7e61eSJeff Mahoney set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); 21542ff7e61eSJeff Mahoney btrfs_insert_inode_hash(inode); 2155f37938e0SEric Sandeen } 2156f37938e0SEric Sandeen 2157ad618368SEric Sandeen static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2158ad618368SEric Sandeen { 2159ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2160129827e3SDavid Sterba init_rwsem(&fs_info->dev_replace.rwsem); 21617f8d236aSDavid Sterba init_waitqueue_head(&fs_info->dev_replace.replace_wait); 2162ad618368SEric Sandeen } 2163ad618368SEric Sandeen 2164f9e92e40SEric Sandeen static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2165f9e92e40SEric Sandeen { 2166f9e92e40SEric Sandeen spin_lock_init(&fs_info->qgroup_lock); 2167f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_ioctl_lock); 2168f9e92e40SEric Sandeen fs_info->qgroup_tree = RB_ROOT; 2169f9e92e40SEric Sandeen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2170f9e92e40SEric Sandeen fs_info->qgroup_seq = 1; 2171f9e92e40SEric Sandeen fs_info->qgroup_ulist = NULL; 2172d2c609b8SJeff Mahoney fs_info->qgroup_rescan_running = false; 2173f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_rescan_lock); 2174f9e92e40SEric Sandeen } 2175f9e92e40SEric Sandeen 21762a458198SEric Sandeen static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info, 21772a458198SEric Sandeen struct btrfs_fs_devices *fs_devices) 21782a458198SEric Sandeen { 2179f7b885beSAnand Jain u32 max_active = fs_info->thread_pool_size; 21806f011058SDavid Sterba unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 21812a458198SEric Sandeen 21822a458198SEric Sandeen fs_info->workers = 2183cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "worker", 2184cb001095SJeff Mahoney flags | WQ_HIGHPRI, max_active, 16); 21852a458198SEric Sandeen 21862a458198SEric Sandeen fs_info->delalloc_workers = 2187cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delalloc", 2188cb001095SJeff Mahoney flags, max_active, 2); 21892a458198SEric Sandeen 21902a458198SEric Sandeen fs_info->flush_workers = 2191cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "flush_delalloc", 2192cb001095SJeff Mahoney flags, max_active, 0); 21932a458198SEric Sandeen 21942a458198SEric Sandeen fs_info->caching_workers = 2195cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); 21962a458198SEric Sandeen 21972a458198SEric Sandeen /* 21982a458198SEric Sandeen * a higher idle thresh on the submit workers makes it much more 21992a458198SEric Sandeen * likely that bios will be send down in a sane order to the 22002a458198SEric Sandeen * devices 22012a458198SEric Sandeen */ 22022a458198SEric Sandeen fs_info->submit_workers = 2203cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "submit", flags, 22042a458198SEric Sandeen min_t(u64, fs_devices->num_devices, 22052a458198SEric Sandeen max_active), 64); 22062a458198SEric Sandeen 22072a458198SEric Sandeen fs_info->fixup_workers = 2208cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); 22092a458198SEric Sandeen 22102a458198SEric Sandeen /* 22112a458198SEric Sandeen * endios are largely parallel and should have a very 22122a458198SEric Sandeen * low idle thresh 22132a458198SEric Sandeen */ 22142a458198SEric Sandeen fs_info->endio_workers = 2215cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4); 22162a458198SEric Sandeen fs_info->endio_meta_workers = 2217cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta", flags, 2218cb001095SJeff Mahoney max_active, 4); 22192a458198SEric Sandeen fs_info->endio_meta_write_workers = 2220cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags, 2221cb001095SJeff Mahoney max_active, 2); 22222a458198SEric Sandeen fs_info->endio_raid56_workers = 2223cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-raid56", flags, 2224cb001095SJeff Mahoney max_active, 4); 22252a458198SEric Sandeen fs_info->endio_repair_workers = 2226cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0); 22272a458198SEric Sandeen fs_info->rmw_workers = 2228cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2); 22292a458198SEric Sandeen fs_info->endio_write_workers = 2230cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-write", flags, 2231cb001095SJeff Mahoney max_active, 2); 22322a458198SEric Sandeen fs_info->endio_freespace_worker = 2233cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "freespace-write", flags, 2234cb001095SJeff Mahoney max_active, 0); 22352a458198SEric Sandeen fs_info->delayed_workers = 2236cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, 2237cb001095SJeff Mahoney max_active, 0); 22382a458198SEric Sandeen fs_info->readahead_workers = 2239cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "readahead", flags, 2240cb001095SJeff Mahoney max_active, 2); 22412a458198SEric Sandeen fs_info->qgroup_rescan_workers = 2242cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); 22432a458198SEric Sandeen fs_info->extent_workers = 2244cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "extent-refs", flags, 22452a458198SEric Sandeen min_t(u64, fs_devices->num_devices, 22462a458198SEric Sandeen max_active), 8); 22472a458198SEric Sandeen 22482a458198SEric Sandeen if (!(fs_info->workers && fs_info->delalloc_workers && 22492a458198SEric Sandeen fs_info->submit_workers && fs_info->flush_workers && 22502a458198SEric Sandeen fs_info->endio_workers && fs_info->endio_meta_workers && 22512a458198SEric Sandeen fs_info->endio_meta_write_workers && 22522a458198SEric Sandeen fs_info->endio_repair_workers && 22532a458198SEric Sandeen fs_info->endio_write_workers && fs_info->endio_raid56_workers && 22542a458198SEric Sandeen fs_info->endio_freespace_worker && fs_info->rmw_workers && 22552a458198SEric Sandeen fs_info->caching_workers && fs_info->readahead_workers && 22562a458198SEric Sandeen fs_info->fixup_workers && fs_info->delayed_workers && 22572a458198SEric Sandeen fs_info->extent_workers && 22582a458198SEric Sandeen fs_info->qgroup_rescan_workers)) { 22592a458198SEric Sandeen return -ENOMEM; 22602a458198SEric Sandeen } 22612a458198SEric Sandeen 22622a458198SEric Sandeen return 0; 22632a458198SEric Sandeen } 22642a458198SEric Sandeen 226563443bf5SEric Sandeen static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 226663443bf5SEric Sandeen struct btrfs_fs_devices *fs_devices) 226763443bf5SEric Sandeen { 226863443bf5SEric Sandeen int ret; 226963443bf5SEric Sandeen struct btrfs_root *log_tree_root; 227063443bf5SEric Sandeen struct btrfs_super_block *disk_super = fs_info->super_copy; 227163443bf5SEric Sandeen u64 bytenr = btrfs_super_log_root(disk_super); 2272581c1760SQu Wenruo int level = btrfs_super_log_root_level(disk_super); 227363443bf5SEric Sandeen 227463443bf5SEric Sandeen if (fs_devices->rw_devices == 0) { 2275f14d104dSDavid Sterba btrfs_warn(fs_info, "log replay required on RO media"); 227663443bf5SEric Sandeen return -EIO; 227763443bf5SEric Sandeen } 227863443bf5SEric Sandeen 227974e4d827SDavid Sterba log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL); 228063443bf5SEric Sandeen if (!log_tree_root) 228163443bf5SEric Sandeen return -ENOMEM; 228263443bf5SEric Sandeen 2283da17066cSJeff Mahoney __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID); 228463443bf5SEric Sandeen 22852ff7e61eSJeff Mahoney log_tree_root->node = read_tree_block(fs_info, bytenr, 2286581c1760SQu Wenruo fs_info->generation + 1, 2287581c1760SQu Wenruo level, NULL); 228864c043deSLiu Bo if (IS_ERR(log_tree_root->node)) { 2289f14d104dSDavid Sterba btrfs_warn(fs_info, "failed to read log tree"); 22900eeff236SLiu Bo ret = PTR_ERR(log_tree_root->node); 229164c043deSLiu Bo kfree(log_tree_root); 22920eeff236SLiu Bo return ret; 229364c043deSLiu Bo } else if (!extent_buffer_uptodate(log_tree_root->node)) { 2294f14d104dSDavid Sterba btrfs_err(fs_info, "failed to read log tree"); 229563443bf5SEric Sandeen free_extent_buffer(log_tree_root->node); 229663443bf5SEric Sandeen kfree(log_tree_root); 229763443bf5SEric Sandeen return -EIO; 229863443bf5SEric Sandeen } 229963443bf5SEric Sandeen /* returns with log_tree_root freed on success */ 230063443bf5SEric Sandeen ret = btrfs_recover_log_trees(log_tree_root); 230163443bf5SEric Sandeen if (ret) { 23020b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 230363443bf5SEric Sandeen "Failed to recover log tree"); 230463443bf5SEric Sandeen free_extent_buffer(log_tree_root->node); 230563443bf5SEric Sandeen kfree(log_tree_root); 230663443bf5SEric Sandeen return ret; 230763443bf5SEric Sandeen } 230863443bf5SEric Sandeen 2309bc98a42cSDavid Howells if (sb_rdonly(fs_info->sb)) { 23106bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 231163443bf5SEric Sandeen if (ret) 231263443bf5SEric Sandeen return ret; 231363443bf5SEric Sandeen } 231463443bf5SEric Sandeen 231563443bf5SEric Sandeen return 0; 231663443bf5SEric Sandeen } 231763443bf5SEric Sandeen 23186bccf3abSJeff Mahoney static int btrfs_read_roots(struct btrfs_fs_info *fs_info) 23194bbcaa64SEric Sandeen { 23206bccf3abSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 2321a4f3d2c4SDavid Sterba struct btrfs_root *root; 23224bbcaa64SEric Sandeen struct btrfs_key location; 23234bbcaa64SEric Sandeen int ret; 23244bbcaa64SEric Sandeen 23256bccf3abSJeff Mahoney BUG_ON(!fs_info->tree_root); 23266bccf3abSJeff Mahoney 23274bbcaa64SEric Sandeen location.objectid = BTRFS_EXTENT_TREE_OBJECTID; 23284bbcaa64SEric Sandeen location.type = BTRFS_ROOT_ITEM_KEY; 23294bbcaa64SEric Sandeen location.offset = 0; 23304bbcaa64SEric Sandeen 2331a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2332f50f4353SLiu Bo if (IS_ERR(root)) { 2333f50f4353SLiu Bo ret = PTR_ERR(root); 2334f50f4353SLiu Bo goto out; 2335f50f4353SLiu Bo } 2336a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2337a4f3d2c4SDavid Sterba fs_info->extent_root = root; 23384bbcaa64SEric Sandeen 23394bbcaa64SEric Sandeen location.objectid = BTRFS_DEV_TREE_OBJECTID; 2340a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2341f50f4353SLiu Bo if (IS_ERR(root)) { 2342f50f4353SLiu Bo ret = PTR_ERR(root); 2343f50f4353SLiu Bo goto out; 2344f50f4353SLiu Bo } 2345a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2346a4f3d2c4SDavid Sterba fs_info->dev_root = root; 23474bbcaa64SEric Sandeen btrfs_init_devices_late(fs_info); 23484bbcaa64SEric Sandeen 23494bbcaa64SEric Sandeen location.objectid = BTRFS_CSUM_TREE_OBJECTID; 2350a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2351f50f4353SLiu Bo if (IS_ERR(root)) { 2352f50f4353SLiu Bo ret = PTR_ERR(root); 2353f50f4353SLiu Bo goto out; 2354f50f4353SLiu Bo } 2355a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2356a4f3d2c4SDavid Sterba fs_info->csum_root = root; 23574bbcaa64SEric Sandeen 23584bbcaa64SEric Sandeen location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2359a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2360a4f3d2c4SDavid Sterba if (!IS_ERR(root)) { 2361a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2362afcdd129SJosef Bacik set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); 2363a4f3d2c4SDavid Sterba fs_info->quota_root = root; 23644bbcaa64SEric Sandeen } 23654bbcaa64SEric Sandeen 23664bbcaa64SEric Sandeen location.objectid = BTRFS_UUID_TREE_OBJECTID; 2367a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2368a4f3d2c4SDavid Sterba if (IS_ERR(root)) { 2369a4f3d2c4SDavid Sterba ret = PTR_ERR(root); 23704bbcaa64SEric Sandeen if (ret != -ENOENT) 2371f50f4353SLiu Bo goto out; 23724bbcaa64SEric Sandeen } else { 2373a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2374a4f3d2c4SDavid Sterba fs_info->uuid_root = root; 23754bbcaa64SEric Sandeen } 23764bbcaa64SEric Sandeen 237770f6d82eSOmar Sandoval if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 237870f6d82eSOmar Sandoval location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID; 237970f6d82eSOmar Sandoval root = btrfs_read_tree_root(tree_root, &location); 2380f50f4353SLiu Bo if (IS_ERR(root)) { 2381f50f4353SLiu Bo ret = PTR_ERR(root); 2382f50f4353SLiu Bo goto out; 2383f50f4353SLiu Bo } 238470f6d82eSOmar Sandoval set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 238570f6d82eSOmar Sandoval fs_info->free_space_root = root; 238670f6d82eSOmar Sandoval } 238770f6d82eSOmar Sandoval 23884bbcaa64SEric Sandeen return 0; 2389f50f4353SLiu Bo out: 2390f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", 2391f50f4353SLiu Bo location.objectid, ret); 2392f50f4353SLiu Bo return ret; 23934bbcaa64SEric Sandeen } 23944bbcaa64SEric Sandeen 2395069ec957SQu Wenruo /* 2396069ec957SQu Wenruo * Real super block validation 2397069ec957SQu Wenruo * NOTE: super csum type and incompat features will not be checked here. 2398069ec957SQu Wenruo * 2399069ec957SQu Wenruo * @sb: super block to check 2400069ec957SQu Wenruo * @mirror_num: the super block number to check its bytenr: 2401069ec957SQu Wenruo * 0 the primary (1st) sb 2402069ec957SQu Wenruo * 1, 2 2nd and 3rd backup copy 2403069ec957SQu Wenruo * -1 skip bytenr check 2404069ec957SQu Wenruo */ 2405069ec957SQu Wenruo static int validate_super(struct btrfs_fs_info *fs_info, 2406069ec957SQu Wenruo struct btrfs_super_block *sb, int mirror_num) 240721a852b0SQu Wenruo { 240821a852b0SQu Wenruo u64 nodesize = btrfs_super_nodesize(sb); 240921a852b0SQu Wenruo u64 sectorsize = btrfs_super_sectorsize(sb); 241021a852b0SQu Wenruo int ret = 0; 241121a852b0SQu Wenruo 241221a852b0SQu Wenruo if (btrfs_super_magic(sb) != BTRFS_MAGIC) { 241321a852b0SQu Wenruo btrfs_err(fs_info, "no valid FS found"); 241421a852b0SQu Wenruo ret = -EINVAL; 241521a852b0SQu Wenruo } 241621a852b0SQu Wenruo if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { 241721a852b0SQu Wenruo btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", 241821a852b0SQu Wenruo btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); 241921a852b0SQu Wenruo ret = -EINVAL; 242021a852b0SQu Wenruo } 242121a852b0SQu Wenruo if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 242221a852b0SQu Wenruo btrfs_err(fs_info, "tree_root level too big: %d >= %d", 242321a852b0SQu Wenruo btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 242421a852b0SQu Wenruo ret = -EINVAL; 242521a852b0SQu Wenruo } 242621a852b0SQu Wenruo if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 242721a852b0SQu Wenruo btrfs_err(fs_info, "chunk_root level too big: %d >= %d", 242821a852b0SQu Wenruo btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 242921a852b0SQu Wenruo ret = -EINVAL; 243021a852b0SQu Wenruo } 243121a852b0SQu Wenruo if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 243221a852b0SQu Wenruo btrfs_err(fs_info, "log_root level too big: %d >= %d", 243321a852b0SQu Wenruo btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 243421a852b0SQu Wenruo ret = -EINVAL; 243521a852b0SQu Wenruo } 243621a852b0SQu Wenruo 243721a852b0SQu Wenruo /* 243821a852b0SQu Wenruo * Check sectorsize and nodesize first, other check will need it. 243921a852b0SQu Wenruo * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. 244021a852b0SQu Wenruo */ 244121a852b0SQu Wenruo if (!is_power_of_2(sectorsize) || sectorsize < 4096 || 244221a852b0SQu Wenruo sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { 244321a852b0SQu Wenruo btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); 244421a852b0SQu Wenruo ret = -EINVAL; 244521a852b0SQu Wenruo } 244621a852b0SQu Wenruo /* Only PAGE SIZE is supported yet */ 244721a852b0SQu Wenruo if (sectorsize != PAGE_SIZE) { 244821a852b0SQu Wenruo btrfs_err(fs_info, 244921a852b0SQu Wenruo "sectorsize %llu not supported yet, only support %lu", 245021a852b0SQu Wenruo sectorsize, PAGE_SIZE); 245121a852b0SQu Wenruo ret = -EINVAL; 245221a852b0SQu Wenruo } 245321a852b0SQu Wenruo if (!is_power_of_2(nodesize) || nodesize < sectorsize || 245421a852b0SQu Wenruo nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { 245521a852b0SQu Wenruo btrfs_err(fs_info, "invalid nodesize %llu", nodesize); 245621a852b0SQu Wenruo ret = -EINVAL; 245721a852b0SQu Wenruo } 245821a852b0SQu Wenruo if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { 245921a852b0SQu Wenruo btrfs_err(fs_info, "invalid leafsize %u, should be %llu", 246021a852b0SQu Wenruo le32_to_cpu(sb->__unused_leafsize), nodesize); 246121a852b0SQu Wenruo ret = -EINVAL; 246221a852b0SQu Wenruo } 246321a852b0SQu Wenruo 246421a852b0SQu Wenruo /* Root alignment check */ 246521a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { 246621a852b0SQu Wenruo btrfs_warn(fs_info, "tree_root block unaligned: %llu", 246721a852b0SQu Wenruo btrfs_super_root(sb)); 246821a852b0SQu Wenruo ret = -EINVAL; 246921a852b0SQu Wenruo } 247021a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { 247121a852b0SQu Wenruo btrfs_warn(fs_info, "chunk_root block unaligned: %llu", 247221a852b0SQu Wenruo btrfs_super_chunk_root(sb)); 247321a852b0SQu Wenruo ret = -EINVAL; 247421a852b0SQu Wenruo } 247521a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { 247621a852b0SQu Wenruo btrfs_warn(fs_info, "log_root block unaligned: %llu", 247721a852b0SQu Wenruo btrfs_super_log_root(sb)); 247821a852b0SQu Wenruo ret = -EINVAL; 247921a852b0SQu Wenruo } 248021a852b0SQu Wenruo 2481de37aa51SNikolay Borisov if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, 24827239ff4bSNikolay Borisov BTRFS_FSID_SIZE) != 0) { 248321a852b0SQu Wenruo btrfs_err(fs_info, 24847239ff4bSNikolay Borisov "dev_item UUID does not match metadata fsid: %pU != %pU", 2485de37aa51SNikolay Borisov fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); 248621a852b0SQu Wenruo ret = -EINVAL; 248721a852b0SQu Wenruo } 248821a852b0SQu Wenruo 248921a852b0SQu Wenruo /* 249021a852b0SQu Wenruo * Hint to catch really bogus numbers, bitflips or so, more exact checks are 249121a852b0SQu Wenruo * done later 249221a852b0SQu Wenruo */ 249321a852b0SQu Wenruo if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { 249421a852b0SQu Wenruo btrfs_err(fs_info, "bytes_used is too small %llu", 249521a852b0SQu Wenruo btrfs_super_bytes_used(sb)); 249621a852b0SQu Wenruo ret = -EINVAL; 249721a852b0SQu Wenruo } 249821a852b0SQu Wenruo if (!is_power_of_2(btrfs_super_stripesize(sb))) { 249921a852b0SQu Wenruo btrfs_err(fs_info, "invalid stripesize %u", 250021a852b0SQu Wenruo btrfs_super_stripesize(sb)); 250121a852b0SQu Wenruo ret = -EINVAL; 250221a852b0SQu Wenruo } 250321a852b0SQu Wenruo if (btrfs_super_num_devices(sb) > (1UL << 31)) 250421a852b0SQu Wenruo btrfs_warn(fs_info, "suspicious number of devices: %llu", 250521a852b0SQu Wenruo btrfs_super_num_devices(sb)); 250621a852b0SQu Wenruo if (btrfs_super_num_devices(sb) == 0) { 250721a852b0SQu Wenruo btrfs_err(fs_info, "number of devices is 0"); 250821a852b0SQu Wenruo ret = -EINVAL; 250921a852b0SQu Wenruo } 251021a852b0SQu Wenruo 2511069ec957SQu Wenruo if (mirror_num >= 0 && 2512069ec957SQu Wenruo btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { 251321a852b0SQu Wenruo btrfs_err(fs_info, "super offset mismatch %llu != %u", 251421a852b0SQu Wenruo btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 251521a852b0SQu Wenruo ret = -EINVAL; 251621a852b0SQu Wenruo } 251721a852b0SQu Wenruo 251821a852b0SQu Wenruo /* 251921a852b0SQu Wenruo * Obvious sys_chunk_array corruptions, it must hold at least one key 252021a852b0SQu Wenruo * and one chunk 252121a852b0SQu Wenruo */ 252221a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 252321a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too big %u > %u", 252421a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 252521a852b0SQu Wenruo BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 252621a852b0SQu Wenruo ret = -EINVAL; 252721a852b0SQu Wenruo } 252821a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 252921a852b0SQu Wenruo + sizeof(struct btrfs_chunk)) { 253021a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too small %u < %zu", 253121a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 253221a852b0SQu Wenruo sizeof(struct btrfs_disk_key) 253321a852b0SQu Wenruo + sizeof(struct btrfs_chunk)); 253421a852b0SQu Wenruo ret = -EINVAL; 253521a852b0SQu Wenruo } 253621a852b0SQu Wenruo 253721a852b0SQu Wenruo /* 253821a852b0SQu Wenruo * The generation is a global counter, we'll trust it more than the others 253921a852b0SQu Wenruo * but it's still possible that it's the one that's wrong. 254021a852b0SQu Wenruo */ 254121a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 254221a852b0SQu Wenruo btrfs_warn(fs_info, 254321a852b0SQu Wenruo "suspicious: generation < chunk_root_generation: %llu < %llu", 254421a852b0SQu Wenruo btrfs_super_generation(sb), 254521a852b0SQu Wenruo btrfs_super_chunk_root_generation(sb)); 254621a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 254721a852b0SQu Wenruo && btrfs_super_cache_generation(sb) != (u64)-1) 254821a852b0SQu Wenruo btrfs_warn(fs_info, 254921a852b0SQu Wenruo "suspicious: generation < cache_generation: %llu < %llu", 255021a852b0SQu Wenruo btrfs_super_generation(sb), 255121a852b0SQu Wenruo btrfs_super_cache_generation(sb)); 255221a852b0SQu Wenruo 255321a852b0SQu Wenruo return ret; 255421a852b0SQu Wenruo } 255521a852b0SQu Wenruo 2556069ec957SQu Wenruo /* 2557069ec957SQu Wenruo * Validation of super block at mount time. 2558069ec957SQu Wenruo * Some checks already done early at mount time, like csum type and incompat 2559069ec957SQu Wenruo * flags will be skipped. 2560069ec957SQu Wenruo */ 2561069ec957SQu Wenruo static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) 2562069ec957SQu Wenruo { 2563069ec957SQu Wenruo return validate_super(fs_info, fs_info->super_copy, 0); 2564069ec957SQu Wenruo } 2565069ec957SQu Wenruo 256675cb857dSQu Wenruo /* 256775cb857dSQu Wenruo * Validation of super block at write time. 256875cb857dSQu Wenruo * Some checks like bytenr check will be skipped as their values will be 256975cb857dSQu Wenruo * overwritten soon. 257075cb857dSQu Wenruo * Extra checks like csum type and incompat flags will be done here. 257175cb857dSQu Wenruo */ 257275cb857dSQu Wenruo static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, 257375cb857dSQu Wenruo struct btrfs_super_block *sb) 257475cb857dSQu Wenruo { 257575cb857dSQu Wenruo int ret; 257675cb857dSQu Wenruo 257775cb857dSQu Wenruo ret = validate_super(fs_info, sb, -1); 257875cb857dSQu Wenruo if (ret < 0) 257975cb857dSQu Wenruo goto out; 258075cb857dSQu Wenruo if (btrfs_super_csum_type(sb) != BTRFS_CSUM_TYPE_CRC32) { 258175cb857dSQu Wenruo ret = -EUCLEAN; 258275cb857dSQu Wenruo btrfs_err(fs_info, "invalid csum type, has %u want %u", 258375cb857dSQu Wenruo btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); 258475cb857dSQu Wenruo goto out; 258575cb857dSQu Wenruo } 258675cb857dSQu Wenruo if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 258775cb857dSQu Wenruo ret = -EUCLEAN; 258875cb857dSQu Wenruo btrfs_err(fs_info, 258975cb857dSQu Wenruo "invalid incompat flags, has 0x%llx valid mask 0x%llx", 259075cb857dSQu Wenruo btrfs_super_incompat_flags(sb), 259175cb857dSQu Wenruo (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); 259275cb857dSQu Wenruo goto out; 259375cb857dSQu Wenruo } 259475cb857dSQu Wenruo out: 259575cb857dSQu Wenruo if (ret < 0) 259675cb857dSQu Wenruo btrfs_err(fs_info, 259775cb857dSQu Wenruo "super block corruption detected before writing it to disk"); 259875cb857dSQu Wenruo return ret; 259975cb857dSQu Wenruo } 260075cb857dSQu Wenruo 2601ad2b2c80SAl Viro int open_ctree(struct super_block *sb, 2602dfe25020SChris Mason struct btrfs_fs_devices *fs_devices, 2603dfe25020SChris Mason char *options) 2604eb60ceacSChris Mason { 2605db94535dSChris Mason u32 sectorsize; 2606db94535dSChris Mason u32 nodesize; 260787ee04ebSChris Mason u32 stripesize; 260884234f3aSYan Zheng u64 generation; 2609f2b636e8SJosef Bacik u64 features; 26103de4586cSChris Mason struct btrfs_key location; 2611a061fc8dSChris Mason struct buffer_head *bh; 26124d34b278SIlya Dryomov struct btrfs_super_block *disk_super; 2613815745cfSAl Viro struct btrfs_fs_info *fs_info = btrfs_sb(sb); 2614f84a8bd6SAl Viro struct btrfs_root *tree_root; 26154d34b278SIlya Dryomov struct btrfs_root *chunk_root; 2616eb60ceacSChris Mason int ret; 2617e58ca020SYan int err = -EINVAL; 2618af31f5e5SChris Mason int num_backups_tried = 0; 2619af31f5e5SChris Mason int backup_index = 0; 26206675df31SOmar Sandoval int clear_free_space_tree = 0; 2621581c1760SQu Wenruo int level; 26224543df7eSChris Mason 262374e4d827SDavid Sterba tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL); 262474e4d827SDavid Sterba chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL); 2625cb517eabSMiao Xie if (!tree_root || !chunk_root) { 262639279cc3SChris Mason err = -ENOMEM; 262739279cc3SChris Mason goto fail; 262839279cc3SChris Mason } 262976dda93cSYan, Zheng 263076dda93cSYan, Zheng ret = init_srcu_struct(&fs_info->subvol_srcu); 263176dda93cSYan, Zheng if (ret) { 263276dda93cSYan, Zheng err = ret; 263376dda93cSYan, Zheng goto fail; 263476dda93cSYan, Zheng } 263576dda93cSYan, Zheng 26364297ff84SJosef Bacik ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL); 2637e2d84521SMiao Xie if (ret) { 2638e2d84521SMiao Xie err = ret; 26399e11ceeeSJan Kara goto fail_srcu; 2640e2d84521SMiao Xie } 26414297ff84SJosef Bacik 26424297ff84SJosef Bacik ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 26434297ff84SJosef Bacik if (ret) { 26444297ff84SJosef Bacik err = ret; 26454297ff84SJosef Bacik goto fail_dio_bytes; 26464297ff84SJosef Bacik } 264709cbfeafSKirill A. Shutemov fs_info->dirty_metadata_batch = PAGE_SIZE * 2648e2d84521SMiao Xie (1 + ilog2(nr_cpu_ids)); 2649e2d84521SMiao Xie 2650908c7f19STejun Heo ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 2651963d678bSMiao Xie if (ret) { 2652963d678bSMiao Xie err = ret; 2653963d678bSMiao Xie goto fail_dirty_metadata_bytes; 2654963d678bSMiao Xie } 2655963d678bSMiao Xie 26567f8d236aSDavid Sterba ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, 26577f8d236aSDavid Sterba GFP_KERNEL); 2658c404e0dcSMiao Xie if (ret) { 2659c404e0dcSMiao Xie err = ret; 2660c404e0dcSMiao Xie goto fail_delalloc_bytes; 2661c404e0dcSMiao Xie } 2662c404e0dcSMiao Xie 266376dda93cSYan, Zheng INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 2664f28491e0SJosef Bacik INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 26658fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 2666facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 266724bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 2668eb73c1b7SMiao Xie INIT_LIST_HEAD(&fs_info->delalloc_roots); 266911833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 267075cb379dSJeff Mahoney INIT_LIST_HEAD(&fs_info->pending_raid_kobjs); 267175cb379dSJeff Mahoney spin_lock_init(&fs_info->pending_raid_kobjs_lock); 2672eb73c1b7SMiao Xie spin_lock_init(&fs_info->delalloc_root_lock); 2673a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 267476dda93cSYan, Zheng spin_lock_init(&fs_info->fs_roots_radix_lock); 267524bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 26764cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 2677f29021b2SJan Schmidt spin_lock_init(&fs_info->tree_mod_seq_lock); 2678ceda0864SMiao Xie spin_lock_init(&fs_info->super_lock); 2679f28491e0SJosef Bacik spin_lock_init(&fs_info->buffer_lock); 268047ab2a6cSJosef Bacik spin_lock_init(&fs_info->unused_bgs_lock); 2681f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 2682d7c15171SZhao Lei mutex_init(&fs_info->unused_bg_unpin_mutex); 268367c5e7d4SFilipe Manana mutex_init(&fs_info->delete_unused_bgs_mutex); 26847585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 2685573bfb72SMiao Xie mutex_init(&fs_info->delalloc_root_mutex); 2686de98ced9SMiao Xie seqlock_init(&fs_info->profiles_lock); 268719c00ddcSChris Mason 26880b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 26896324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 2690f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 269147ab2a6cSJosef Bacik INIT_LIST_HEAD(&fs_info->unused_bgs); 26920b86a832SChris Mason btrfs_mapping_init(&fs_info->mapping_tree); 269366d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 269466d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 269566d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 269666d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 269766d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 269866d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 269966d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 2700ba2c4d4eSJosef Bacik btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, 2701ba2c4d4eSJosef Bacik BTRFS_BLOCK_RSV_DELREFS); 2702ba2c4d4eSJosef Bacik 2703771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 27044cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 27052fefd558SZhao Lei atomic_set(&fs_info->reada_works_cnt, 0); 2706034f784dSJosef Bacik atomic_set(&fs_info->nr_delayed_iputs, 0); 2707fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 2708e20d96d6SChris Mason fs_info->sb = sb; 270995ac567aSFilipe David Borba Manana fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 27109ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 27114cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 2712a5ed45f8SNikolay Borisov atomic64_set(&fs_info->free_chunk_space, 0); 2713f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 27148b87dc17SDavid Sterba fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 2715f8c269d7SDavid Sterba fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 271690519d66SArne Jansen /* readahead state */ 2717d0164adcSMel Gorman INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); 271890519d66SArne Jansen spin_lock_init(&fs_info->reada_lock); 2719fd708b81SJosef Bacik btrfs_init_ref_verify(fs_info); 2720c8b97818SChris Mason 2721b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 2722b34b086cSChris Mason num_online_cpus() + 2, 8); 27230afbaf8cSChris Mason 2724199c2a9cSMiao Xie INIT_LIST_HEAD(&fs_info->ordered_roots); 2725199c2a9cSMiao Xie spin_lock_init(&fs_info->ordered_root_lock); 272669fe2d75SJosef Bacik 272769fe2d75SJosef Bacik fs_info->btree_inode = new_inode(sb); 272869fe2d75SJosef Bacik if (!fs_info->btree_inode) { 272969fe2d75SJosef Bacik err = -ENOMEM; 273069fe2d75SJosef Bacik goto fail_bio_counter; 273169fe2d75SJosef Bacik } 273269fe2d75SJosef Bacik mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 273369fe2d75SJosef Bacik 273416cdcec7SMiao Xie fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 273574e4d827SDavid Sterba GFP_KERNEL); 273616cdcec7SMiao Xie if (!fs_info->delayed_root) { 273716cdcec7SMiao Xie err = -ENOMEM; 273816cdcec7SMiao Xie goto fail_iput; 273916cdcec7SMiao Xie } 274016cdcec7SMiao Xie btrfs_init_delayed_root(fs_info->delayed_root); 27413eaa2885SChris Mason 2742638aa7edSEric Sandeen btrfs_init_scrub(fs_info); 274321adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 274421adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 274521adbd5cSStefan Behrens #endif 2746779a65a4SEric Sandeen btrfs_init_balance(fs_info); 274721c7e756SMiao Xie btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work); 2748a2de733cSArne Jansen 27499f6d2510SDavid Sterba sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; 27509f6d2510SDavid Sterba sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); 2751a061fc8dSChris Mason 27526bccf3abSJeff Mahoney btrfs_init_btree_inode(fs_info); 275339279cc3SChris Mason 2754e02119d5SChris Mason spin_lock_init(&fs_info->block_group_cache_lock); 27556bef4d31SEric Paris fs_info->block_group_cache_tree = RB_ROOT; 2756a1897fddSLiu Bo fs_info->first_logical_byte = (u64)-1; 2757925baeddSChris Mason 275843eb5f29SQu Wenruo extent_io_tree_init(fs_info, &fs_info->freed_extents[0], 275943eb5f29SQu Wenruo IO_TREE_FS_INFO_FREED_EXTENTS0, NULL); 276043eb5f29SQu Wenruo extent_io_tree_init(fs_info, &fs_info->freed_extents[1], 276143eb5f29SQu Wenruo IO_TREE_FS_INFO_FREED_EXTENTS1, NULL); 276211833d66SYan Zheng fs_info->pinned_extents = &fs_info->freed_extents[0]; 2763afcdd129SJosef Bacik set_bit(BTRFS_FS_BARRIER, &fs_info->flags); 2764509659cdSChris Mason 27655a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 276679154b1bSChris Mason mutex_init(&fs_info->tree_log_mutex); 2767925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 2768a74a4b97SChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 2769a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 27701bbc621eSChris Mason mutex_init(&fs_info->ro_block_group_mutex); 27719e351cc8SJosef Bacik init_rwsem(&fs_info->commit_root_sem); 2772c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 277376dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 2774803b2f54SStefan Behrens sema_init(&fs_info->uuid_tree_rescan_sem, 1); 2775fa9c0d79SChris Mason 2776ad618368SEric Sandeen btrfs_init_dev_replace_locks(fs_info); 2777f9e92e40SEric Sandeen btrfs_init_qgroup(fs_info); 2778416ac51dSArne Jansen 2779fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2780fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2781fa9c0d79SChris Mason 2782e6dcd2dcSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 2783f9295749SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 2784bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 27854854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 2786034f784dSJosef Bacik init_waitqueue_head(&fs_info->delayed_iputs_wait); 27873768f368SChris Mason 2788da17066cSJeff Mahoney /* Usable values until the real ones are cached from the superblock */ 2789da17066cSJeff Mahoney fs_info->nodesize = 4096; 2790da17066cSJeff Mahoney fs_info->sectorsize = 4096; 2791da17066cSJeff Mahoney fs_info->stripesize = 4096; 2792da17066cSJeff Mahoney 2793eede2bf3SOmar Sandoval spin_lock_init(&fs_info->swapfile_pins_lock); 2794eede2bf3SOmar Sandoval fs_info->swapfile_pins = RB_ROOT; 2795eede2bf3SOmar Sandoval 279653b381b3SDavid Woodhouse ret = btrfs_alloc_stripe_hash_table(fs_info); 279753b381b3SDavid Woodhouse if (ret) { 279883c8266aSDavid Sterba err = ret; 279953b381b3SDavid Woodhouse goto fail_alloc; 280053b381b3SDavid Woodhouse } 280153b381b3SDavid Woodhouse 2802da17066cSJeff Mahoney __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID); 28037eccb903SChris Mason 28043c4bb26bSChris Mason invalidate_bdev(fs_devices->latest_bdev); 28051104a885SDavid Sterba 28061104a885SDavid Sterba /* 28071104a885SDavid Sterba * Read super block and check the signature bytes only 28081104a885SDavid Sterba */ 2809a512bbf8SYan Zheng bh = btrfs_read_dev_super(fs_devices->latest_bdev); 281092fc03fbSAnand Jain if (IS_ERR(bh)) { 281192fc03fbSAnand Jain err = PTR_ERR(bh); 281216cdcec7SMiao Xie goto fail_alloc; 281320b45077SDave Young } 281439279cc3SChris Mason 28151104a885SDavid Sterba /* 28161104a885SDavid Sterba * We want to check superblock checksum, the type is stored inside. 28171104a885SDavid Sterba * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 28181104a885SDavid Sterba */ 2819ab8d0fc4SJeff Mahoney if (btrfs_check_super_csum(fs_info, bh->b_data)) { 282005135f59SDavid Sterba btrfs_err(fs_info, "superblock checksum mismatch"); 28211104a885SDavid Sterba err = -EINVAL; 2822b2acdddfSAnand Jain brelse(bh); 28231104a885SDavid Sterba goto fail_alloc; 28241104a885SDavid Sterba } 28251104a885SDavid Sterba 28261104a885SDavid Sterba /* 28271104a885SDavid Sterba * super_copy is zeroed at allocation time and we never touch the 28281104a885SDavid Sterba * following bytes up to INFO_SIZE, the checksum is calculated from 28291104a885SDavid Sterba * the whole block of INFO_SIZE 28301104a885SDavid Sterba */ 28316c41761fSDavid Sterba memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy)); 2832a061fc8dSChris Mason brelse(bh); 28335f39d397SChris Mason 2834fbc6feaeSNikolay Borisov disk_super = fs_info->super_copy; 2835fbc6feaeSNikolay Borisov 2836de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, 2837de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 2838de37aa51SNikolay Borisov 28397239ff4bSNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID)) { 2840de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid, 2841de37aa51SNikolay Borisov fs_info->super_copy->metadata_uuid, 2842de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 28437239ff4bSNikolay Borisov } 28440b86a832SChris Mason 2845fbc6feaeSNikolay Borisov features = btrfs_super_flags(disk_super); 2846fbc6feaeSNikolay Borisov if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { 2847fbc6feaeSNikolay Borisov features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; 2848fbc6feaeSNikolay Borisov btrfs_set_super_flags(disk_super, features); 2849fbc6feaeSNikolay Borisov btrfs_info(fs_info, 2850fbc6feaeSNikolay Borisov "found metadata UUID change in progress flag, clearing"); 2851fbc6feaeSNikolay Borisov } 2852fbc6feaeSNikolay Borisov 2853fbc6feaeSNikolay Borisov memcpy(fs_info->super_for_commit, fs_info->super_copy, 2854fbc6feaeSNikolay Borisov sizeof(*fs_info->super_for_commit)); 2855de37aa51SNikolay Borisov 2856069ec957SQu Wenruo ret = btrfs_validate_mount_super(fs_info); 28571104a885SDavid Sterba if (ret) { 285805135f59SDavid Sterba btrfs_err(fs_info, "superblock contains fatal errors"); 28591104a885SDavid Sterba err = -EINVAL; 28601104a885SDavid Sterba goto fail_alloc; 28611104a885SDavid Sterba } 28621104a885SDavid Sterba 28630f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 286416cdcec7SMiao Xie goto fail_alloc; 28650f7d52f4SChris Mason 2866acce952bSliubo /* check FS state, whether FS is broken. */ 286787533c47SMiao Xie if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 286887533c47SMiao Xie set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 2869acce952bSliubo 287075e7cb7fSLiu Bo /* 2871af31f5e5SChris Mason * run through our array of backup supers and setup 2872af31f5e5SChris Mason * our ring pointer to the oldest one 2873af31f5e5SChris Mason */ 2874af31f5e5SChris Mason generation = btrfs_super_generation(disk_super); 2875af31f5e5SChris Mason find_oldest_super_backup(fs_info, generation); 2876af31f5e5SChris Mason 2877af31f5e5SChris Mason /* 287875e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 287975e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 288075e7cb7fSLiu Bo */ 288175e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 288275e7cb7fSLiu Bo 28832ff7e61eSJeff Mahoney ret = btrfs_parse_options(fs_info, options, sb->s_flags); 28842b82032cSYan Zheng if (ret) { 28852b82032cSYan Zheng err = ret; 288616cdcec7SMiao Xie goto fail_alloc; 28872b82032cSYan Zheng } 2888dfe25020SChris Mason 2889f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 2890f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 2891f2b636e8SJosef Bacik if (features) { 289205135f59SDavid Sterba btrfs_err(fs_info, 289305135f59SDavid Sterba "cannot mount because of unsupported optional features (%llx)", 2894c1c9ff7cSGeert Uytterhoeven features); 2895f2b636e8SJosef Bacik err = -EINVAL; 289616cdcec7SMiao Xie goto fail_alloc; 2897f2b636e8SJosef Bacik } 2898f2b636e8SJosef Bacik 28995d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 29005d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 29010b246afaSJeff Mahoney if (fs_info->compress_type == BTRFS_COMPRESS_LZO) 2902a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 29035c1aab1dSNick Terrell else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) 29045c1aab1dSNick Terrell features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; 2905727011e0SChris Mason 29063173a18fSJosef Bacik if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) 290705135f59SDavid Sterba btrfs_info(fs_info, "has skinny extents"); 29083173a18fSJosef Bacik 2909727011e0SChris Mason /* 2910727011e0SChris Mason * flag our filesystem as having big metadata blocks if 2911727011e0SChris Mason * they are bigger than the page size 2912727011e0SChris Mason */ 291309cbfeafSKirill A. Shutemov if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) { 2914727011e0SChris Mason if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 291505135f59SDavid Sterba btrfs_info(fs_info, 291605135f59SDavid Sterba "flagging fs with big metadata feature"); 2917727011e0SChris Mason features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 2918727011e0SChris Mason } 2919727011e0SChris Mason 2920bc3f116fSChris Mason nodesize = btrfs_super_nodesize(disk_super); 2921bc3f116fSChris Mason sectorsize = btrfs_super_sectorsize(disk_super); 2922b7f67055SChandan Rajendra stripesize = sectorsize; 2923707e8a07SDavid Sterba fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 2924963d678bSMiao Xie fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 2925bc3f116fSChris Mason 2926da17066cSJeff Mahoney /* Cache block sizes */ 2927da17066cSJeff Mahoney fs_info->nodesize = nodesize; 2928da17066cSJeff Mahoney fs_info->sectorsize = sectorsize; 2929da17066cSJeff Mahoney fs_info->stripesize = stripesize; 2930da17066cSJeff Mahoney 2931bc3f116fSChris Mason /* 2932bc3f116fSChris Mason * mixed block groups end up with duplicate but slightly offset 2933bc3f116fSChris Mason * extent buffers for the same range. It leads to corruptions 2934bc3f116fSChris Mason */ 2935bc3f116fSChris Mason if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 2936707e8a07SDavid Sterba (sectorsize != nodesize)) { 293705135f59SDavid Sterba btrfs_err(fs_info, 293805135f59SDavid Sterba "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", 293905135f59SDavid Sterba nodesize, sectorsize); 2940bc3f116fSChris Mason goto fail_alloc; 2941bc3f116fSChris Mason } 2942bc3f116fSChris Mason 2943ceda0864SMiao Xie /* 2944ceda0864SMiao Xie * Needn't use the lock because there is no other task which will 2945ceda0864SMiao Xie * update the flag. 2946ceda0864SMiao Xie */ 29475d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 29485d4f98a2SYan Zheng 2949f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 2950f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 2951bc98a42cSDavid Howells if (!sb_rdonly(sb) && features) { 295205135f59SDavid Sterba btrfs_err(fs_info, 295305135f59SDavid Sterba "cannot mount read-write because of unsupported optional features (%llx)", 2954c1c9ff7cSGeert Uytterhoeven features); 2955f2b636e8SJosef Bacik err = -EINVAL; 295616cdcec7SMiao Xie goto fail_alloc; 2957f2b636e8SJosef Bacik } 295861d92c32SChris Mason 29592a458198SEric Sandeen ret = btrfs_init_workqueues(fs_info, fs_devices); 29602a458198SEric Sandeen if (ret) { 29612a458198SEric Sandeen err = ret; 29620dc3b84aSJosef Bacik goto fail_sb_buffer; 29630dc3b84aSJosef Bacik } 29644543df7eSChris Mason 29659e11ceeeSJan Kara sb->s_bdi->congested_fn = btrfs_congested_fn; 29669e11ceeeSJan Kara sb->s_bdi->congested_data = fs_info; 29679e11ceeeSJan Kara sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK; 2968b5420237SNikolay Borisov sb->s_bdi->ra_pages = VM_READAHEAD_PAGES; 29699e11ceeeSJan Kara sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); 29709e11ceeeSJan Kara sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); 29714575c9ccSChris Mason 2972a061fc8dSChris Mason sb->s_blocksize = sectorsize; 2973a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 2974de37aa51SNikolay Borisov memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); 2975db94535dSChris Mason 2976925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 29776bccf3abSJeff Mahoney ret = btrfs_read_sys_array(fs_info); 2978925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 297984eed90fSChris Mason if (ret) { 298005135f59SDavid Sterba btrfs_err(fs_info, "failed to read the system array: %d", ret); 29815d4f98a2SYan Zheng goto fail_sb_buffer; 298284eed90fSChris Mason } 29830b86a832SChris Mason 298484234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 2985581c1760SQu Wenruo level = btrfs_super_chunk_root_level(disk_super); 29860b86a832SChris Mason 2987da17066cSJeff Mahoney __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID); 29880b86a832SChris Mason 29892ff7e61eSJeff Mahoney chunk_root->node = read_tree_block(fs_info, 29900b86a832SChris Mason btrfs_super_chunk_root(disk_super), 2991581c1760SQu Wenruo generation, level, NULL); 299264c043deSLiu Bo if (IS_ERR(chunk_root->node) || 299364c043deSLiu Bo !extent_buffer_uptodate(chunk_root->node)) { 299405135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk root"); 2995e5fffbacSchandan if (!IS_ERR(chunk_root->node)) 2996e5fffbacSchandan free_extent_buffer(chunk_root->node); 299795ab1f64SZhao Lei chunk_root->node = NULL; 2998af31f5e5SChris Mason goto fail_tree_roots; 299983121942SDavid Woodhouse } 30005d4f98a2SYan Zheng btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 30015d4f98a2SYan Zheng chunk_root->commit_root = btrfs_root_node(chunk_root); 30020b86a832SChris Mason 3003e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 3004b308bc2fSGeert Uytterhoeven btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE); 3005e17cade2SChris Mason 30065b4aacefSJeff Mahoney ret = btrfs_read_chunk_tree(fs_info); 30072b82032cSYan Zheng if (ret) { 300805135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk tree: %d", ret); 3009af31f5e5SChris Mason goto fail_tree_roots; 30102b82032cSYan Zheng } 30110b86a832SChris Mason 30128dabb742SStefan Behrens /* 30139b99b115SAnand Jain * Keep the devid that is marked to be the target device for the 30149b99b115SAnand Jain * device replace procedure 30158dabb742SStefan Behrens */ 30169b99b115SAnand Jain btrfs_free_extra_devids(fs_devices, 0); 3017dfe25020SChris Mason 3018a6b0d5c8SChris Mason if (!fs_devices->latest_bdev) { 301905135f59SDavid Sterba btrfs_err(fs_info, "failed to read devices"); 3020a6b0d5c8SChris Mason goto fail_tree_roots; 3021a6b0d5c8SChris Mason } 3022a6b0d5c8SChris Mason 3023af31f5e5SChris Mason retry_root_backup: 302484234f3aSYan Zheng generation = btrfs_super_generation(disk_super); 3025581c1760SQu Wenruo level = btrfs_super_root_level(disk_super); 30260b86a832SChris Mason 30272ff7e61eSJeff Mahoney tree_root->node = read_tree_block(fs_info, 3028db94535dSChris Mason btrfs_super_root(disk_super), 3029581c1760SQu Wenruo generation, level, NULL); 303064c043deSLiu Bo if (IS_ERR(tree_root->node) || 303164c043deSLiu Bo !extent_buffer_uptodate(tree_root->node)) { 303205135f59SDavid Sterba btrfs_warn(fs_info, "failed to read tree root"); 3033e5fffbacSchandan if (!IS_ERR(tree_root->node)) 3034e5fffbacSchandan free_extent_buffer(tree_root->node); 303595ab1f64SZhao Lei tree_root->node = NULL; 3036af31f5e5SChris Mason goto recovery_tree_root; 303783121942SDavid Woodhouse } 3038af31f5e5SChris Mason 30395d4f98a2SYan Zheng btrfs_set_root_node(&tree_root->root_item, tree_root->node); 30405d4f98a2SYan Zheng tree_root->commit_root = btrfs_root_node(tree_root); 304169e9c6c6SStefan Behrens btrfs_set_root_refs(&tree_root->root_item, 1); 3042db94535dSChris Mason 3043f32e48e9SChandan Rajendra mutex_lock(&tree_root->objectid_mutex); 3044f32e48e9SChandan Rajendra ret = btrfs_find_highest_objectid(tree_root, 3045f32e48e9SChandan Rajendra &tree_root->highest_objectid); 3046f32e48e9SChandan Rajendra if (ret) { 3047f32e48e9SChandan Rajendra mutex_unlock(&tree_root->objectid_mutex); 3048f32e48e9SChandan Rajendra goto recovery_tree_root; 3049f32e48e9SChandan Rajendra } 3050f32e48e9SChandan Rajendra 3051f32e48e9SChandan Rajendra ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 3052f32e48e9SChandan Rajendra 3053f32e48e9SChandan Rajendra mutex_unlock(&tree_root->objectid_mutex); 3054f32e48e9SChandan Rajendra 30556bccf3abSJeff Mahoney ret = btrfs_read_roots(fs_info); 30564bbcaa64SEric Sandeen if (ret) 3057af31f5e5SChris Mason goto recovery_tree_root; 3058f7a81ea4SStefan Behrens 30598929ecfaSYan, Zheng fs_info->generation = generation; 30608929ecfaSYan, Zheng fs_info->last_trans_committed = generation; 30618929ecfaSYan, Zheng 3062cf90d884SQu Wenruo ret = btrfs_verify_dev_extents(fs_info); 3063cf90d884SQu Wenruo if (ret) { 3064cf90d884SQu Wenruo btrfs_err(fs_info, 3065cf90d884SQu Wenruo "failed to verify dev extents against chunks: %d", 3066cf90d884SQu Wenruo ret); 3067cf90d884SQu Wenruo goto fail_block_groups; 3068cf90d884SQu Wenruo } 306968310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 307068310a5eSIlya Dryomov if (ret) { 307105135f59SDavid Sterba btrfs_err(fs_info, "failed to recover balance: %d", ret); 307268310a5eSIlya Dryomov goto fail_block_groups; 307368310a5eSIlya Dryomov } 307468310a5eSIlya Dryomov 3075733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 3076733f4fbbSStefan Behrens if (ret) { 307705135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_stats: %d", ret); 3078733f4fbbSStefan Behrens goto fail_block_groups; 3079733f4fbbSStefan Behrens } 3080733f4fbbSStefan Behrens 30818dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 30828dabb742SStefan Behrens if (ret) { 308305135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_replace: %d", ret); 30848dabb742SStefan Behrens goto fail_block_groups; 30858dabb742SStefan Behrens } 30868dabb742SStefan Behrens 30879b99b115SAnand Jain btrfs_free_extra_devids(fs_devices, 1); 30888dabb742SStefan Behrens 3089b7c35e81SAnand Jain ret = btrfs_sysfs_add_fsid(fs_devices, NULL); 3090b7c35e81SAnand Jain if (ret) { 309105135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", 309205135f59SDavid Sterba ret); 3093b7c35e81SAnand Jain goto fail_block_groups; 3094b7c35e81SAnand Jain } 3095b7c35e81SAnand Jain 3096b7c35e81SAnand Jain ret = btrfs_sysfs_add_device(fs_devices); 3097b7c35e81SAnand Jain if (ret) { 309805135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs device interface: %d", 309905135f59SDavid Sterba ret); 3100b7c35e81SAnand Jain goto fail_fsdev_sysfs; 3101b7c35e81SAnand Jain } 3102b7c35e81SAnand Jain 310396f3136eSAnand Jain ret = btrfs_sysfs_add_mounted(fs_info); 31045ac1d209SJeff Mahoney if (ret) { 310505135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); 3106b7c35e81SAnand Jain goto fail_fsdev_sysfs; 31075ac1d209SJeff Mahoney } 31085ac1d209SJeff Mahoney 3109c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 3110c59021f8Sliubo if (ret) { 311105135f59SDavid Sterba btrfs_err(fs_info, "failed to initialize space info: %d", ret); 31122365dd3cSAnand Jain goto fail_sysfs; 3113c59021f8Sliubo } 3114c59021f8Sliubo 31155b4aacefSJeff Mahoney ret = btrfs_read_block_groups(fs_info); 31161b1d1f66SJosef Bacik if (ret) { 311705135f59SDavid Sterba btrfs_err(fs_info, "failed to read block groups: %d", ret); 31182365dd3cSAnand Jain goto fail_sysfs; 31191b1d1f66SJosef Bacik } 31204330e183SQu Wenruo 31216528b99dSAnand Jain if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) { 312205135f59SDavid Sterba btrfs_warn(fs_info, 312352042d8eSAndrea Gelmini "writable mount is not allowed due to too many missing devices"); 31242365dd3cSAnand Jain goto fail_sysfs; 3125292fd7fcSStefan Behrens } 31269078a3e1SChris Mason 3127a74a4b97SChris Mason fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 3128a74a4b97SChris Mason "btrfs-cleaner"); 312957506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 31302365dd3cSAnand Jain goto fail_sysfs; 3131a74a4b97SChris Mason 3132a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 3133a74a4b97SChris Mason tree_root, 3134a74a4b97SChris Mason "btrfs-transaction"); 313557506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 31363f157a2fSChris Mason goto fail_cleaner; 3137a74a4b97SChris Mason 3138583b7231SHans van Kranenburg if (!btrfs_test_opt(fs_info, NOSSD) && 3139c289811cSChris Mason !fs_info->fs_devices->rotating) { 3140583b7231SHans van Kranenburg btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); 3141c289811cSChris Mason } 3142c289811cSChris Mason 3143572d9ab7SDavid Sterba /* 314401327610SNicholas D Steeves * Mount does not set all options immediately, we can do it now and do 3145572d9ab7SDavid Sterba * not have to wait for transaction commit 3146572d9ab7SDavid Sterba */ 3147572d9ab7SDavid Sterba btrfs_apply_pending_changes(fs_info); 31483818aea2SQu Wenruo 314921adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 31500b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { 31512ff7e61eSJeff Mahoney ret = btrfsic_mount(fs_info, fs_devices, 31520b246afaSJeff Mahoney btrfs_test_opt(fs_info, 315321adbd5cSStefan Behrens CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? 315421adbd5cSStefan Behrens 1 : 0, 315521adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 315621adbd5cSStefan Behrens if (ret) 315705135f59SDavid Sterba btrfs_warn(fs_info, 315805135f59SDavid Sterba "failed to initialize integrity check module: %d", 315905135f59SDavid Sterba ret); 316021adbd5cSStefan Behrens } 316121adbd5cSStefan Behrens #endif 3162bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 3163bcef60f2SArne Jansen if (ret) 3164bcef60f2SArne Jansen goto fail_trans_kthread; 316521adbd5cSStefan Behrens 3166fd708b81SJosef Bacik if (btrfs_build_ref_tree(fs_info)) 3167fd708b81SJosef Bacik btrfs_err(fs_info, "couldn't build ref tree"); 3168fd708b81SJosef Bacik 316996da0919SQu Wenruo /* do not make disk changes in broken FS or nologreplay is given */ 317096da0919SQu Wenruo if (btrfs_super_log_root(disk_super) != 0 && 31710b246afaSJeff Mahoney !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 317263443bf5SEric Sandeen ret = btrfs_replay_log(fs_info, fs_devices); 317379787eaaSJeff Mahoney if (ret) { 317463443bf5SEric Sandeen err = ret; 317528c16cbbSWang Shilong goto fail_qgroup; 3176e556ce2cSYan Zheng } 3177e02119d5SChris Mason } 31781a40e23bSZheng Yan 31796bccf3abSJeff Mahoney ret = btrfs_find_orphan_roots(fs_info); 318079787eaaSJeff Mahoney if (ret) 318128c16cbbSWang Shilong goto fail_qgroup; 318276dda93cSYan, Zheng 3183bc98a42cSDavid Howells if (!sb_rdonly(sb)) { 3184d68fc57bSYan, Zheng ret = btrfs_cleanup_fs_roots(fs_info); 318544c44af2SIlya Dryomov if (ret) 318628c16cbbSWang Shilong goto fail_qgroup; 318790c711abSZygo Blaxell 318890c711abSZygo Blaxell mutex_lock(&fs_info->cleaner_mutex); 31895d4f98a2SYan Zheng ret = btrfs_recover_relocation(tree_root); 319090c711abSZygo Blaxell mutex_unlock(&fs_info->cleaner_mutex); 3191d7ce5843SMiao Xie if (ret < 0) { 319205135f59SDavid Sterba btrfs_warn(fs_info, "failed to recover relocation: %d", 319305135f59SDavid Sterba ret); 3194d7ce5843SMiao Xie err = -EINVAL; 3195bcef60f2SArne Jansen goto fail_qgroup; 3196d7ce5843SMiao Xie } 31977c2ca468SChris Mason } 31981a40e23bSZheng Yan 31993de4586cSChris Mason location.objectid = BTRFS_FS_TREE_OBJECTID; 32003de4586cSChris Mason location.type = BTRFS_ROOT_ITEM_KEY; 3201cb517eabSMiao Xie location.offset = 0; 32023de4586cSChris Mason 32033de4586cSChris Mason fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location); 32043140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 32053140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 3206f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read fs tree: %d", err); 3207bcef60f2SArne Jansen goto fail_qgroup; 32083140c9a3SDan Carpenter } 3209c289811cSChris Mason 3210bc98a42cSDavid Howells if (sb_rdonly(sb)) 32112b6ba629SIlya Dryomov return 0; 32122b6ba629SIlya Dryomov 3213f8d468a1SOmar Sandoval if (btrfs_test_opt(fs_info, CLEAR_CACHE) && 3214f8d468a1SOmar Sandoval btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 32156675df31SOmar Sandoval clear_free_space_tree = 1; 32166675df31SOmar Sandoval } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 32176675df31SOmar Sandoval !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { 32186675df31SOmar Sandoval btrfs_warn(fs_info, "free space tree is invalid"); 32196675df31SOmar Sandoval clear_free_space_tree = 1; 32206675df31SOmar Sandoval } 32216675df31SOmar Sandoval 32226675df31SOmar Sandoval if (clear_free_space_tree) { 3223f8d468a1SOmar Sandoval btrfs_info(fs_info, "clearing free space tree"); 3224f8d468a1SOmar Sandoval ret = btrfs_clear_free_space_tree(fs_info); 3225f8d468a1SOmar Sandoval if (ret) { 3226f8d468a1SOmar Sandoval btrfs_warn(fs_info, 3227f8d468a1SOmar Sandoval "failed to clear free space tree: %d", ret); 32286bccf3abSJeff Mahoney close_ctree(fs_info); 3229f8d468a1SOmar Sandoval return ret; 3230f8d468a1SOmar Sandoval } 3231f8d468a1SOmar Sandoval } 3232f8d468a1SOmar Sandoval 32330b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && 3234511711afSChris Mason !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 323505135f59SDavid Sterba btrfs_info(fs_info, "creating free space tree"); 3236511711afSChris Mason ret = btrfs_create_free_space_tree(fs_info); 3237511711afSChris Mason if (ret) { 323805135f59SDavid Sterba btrfs_warn(fs_info, 323905135f59SDavid Sterba "failed to create free space tree: %d", ret); 32406bccf3abSJeff Mahoney close_ctree(fs_info); 3241511711afSChris Mason return ret; 3242511711afSChris Mason } 3243511711afSChris Mason } 3244511711afSChris Mason 3245e3acc2a6SJosef Bacik down_read(&fs_info->cleanup_work_sem); 32462b6ba629SIlya Dryomov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 32472b6ba629SIlya Dryomov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 32482b6ba629SIlya Dryomov up_read(&fs_info->cleanup_work_sem); 32496bccf3abSJeff Mahoney close_ctree(fs_info); 32502b6ba629SIlya Dryomov return ret; 32512b6ba629SIlya Dryomov } 3252e3acc2a6SJosef Bacik up_read(&fs_info->cleanup_work_sem); 325359641015SIlya Dryomov 32542b6ba629SIlya Dryomov ret = btrfs_resume_balance_async(fs_info); 32552b6ba629SIlya Dryomov if (ret) { 325605135f59SDavid Sterba btrfs_warn(fs_info, "failed to resume balance: %d", ret); 32576bccf3abSJeff Mahoney close_ctree(fs_info); 32582b6ba629SIlya Dryomov return ret; 3259e3acc2a6SJosef Bacik } 3260e3acc2a6SJosef Bacik 32618dabb742SStefan Behrens ret = btrfs_resume_dev_replace_async(fs_info); 32628dabb742SStefan Behrens if (ret) { 326305135f59SDavid Sterba btrfs_warn(fs_info, "failed to resume device replace: %d", ret); 32646bccf3abSJeff Mahoney close_ctree(fs_info); 32658dabb742SStefan Behrens return ret; 32668dabb742SStefan Behrens } 32678dabb742SStefan Behrens 3268b382a324SJan Schmidt btrfs_qgroup_rescan_resume(fs_info); 3269b382a324SJan Schmidt 32704bbcaa64SEric Sandeen if (!fs_info->uuid_root) { 327105135f59SDavid Sterba btrfs_info(fs_info, "creating UUID tree"); 3272f7a81ea4SStefan Behrens ret = btrfs_create_uuid_tree(fs_info); 3273f7a81ea4SStefan Behrens if (ret) { 327405135f59SDavid Sterba btrfs_warn(fs_info, 327505135f59SDavid Sterba "failed to create the UUID tree: %d", ret); 32766bccf3abSJeff Mahoney close_ctree(fs_info); 3277f7a81ea4SStefan Behrens return ret; 3278f7a81ea4SStefan Behrens } 32790b246afaSJeff Mahoney } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || 32804bbcaa64SEric Sandeen fs_info->generation != 32814bbcaa64SEric Sandeen btrfs_super_uuid_tree_generation(disk_super)) { 328205135f59SDavid Sterba btrfs_info(fs_info, "checking UUID tree"); 328370f80175SStefan Behrens ret = btrfs_check_uuid_tree(fs_info); 328470f80175SStefan Behrens if (ret) { 328505135f59SDavid Sterba btrfs_warn(fs_info, 328605135f59SDavid Sterba "failed to check the UUID tree: %d", ret); 32876bccf3abSJeff Mahoney close_ctree(fs_info); 328870f80175SStefan Behrens return ret; 328970f80175SStefan Behrens } 329070f80175SStefan Behrens } else { 3291afcdd129SJosef Bacik set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 3292f7a81ea4SStefan Behrens } 3293afcdd129SJosef Bacik set_bit(BTRFS_FS_OPEN, &fs_info->flags); 329447ab2a6cSJosef Bacik 32958dcddfa0SQu Wenruo /* 32968dcddfa0SQu Wenruo * backuproot only affect mount behavior, and if open_ctree succeeded, 32978dcddfa0SQu Wenruo * no need to keep the flag 32988dcddfa0SQu Wenruo */ 32998dcddfa0SQu Wenruo btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT); 33008dcddfa0SQu Wenruo 3301ad2b2c80SAl Viro return 0; 330239279cc3SChris Mason 3303bcef60f2SArne Jansen fail_qgroup: 3304bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 33057c2ca468SChris Mason fail_trans_kthread: 33067c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 33072ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 3308faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 33093f157a2fSChris Mason fail_cleaner: 3310a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 33117c2ca468SChris Mason 33127c2ca468SChris Mason /* 33137c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 33147c2ca468SChris Mason * kthreads 33157c2ca468SChris Mason */ 33167c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 33177c2ca468SChris Mason 33182365dd3cSAnand Jain fail_sysfs: 33196618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 33202365dd3cSAnand Jain 3321b7c35e81SAnand Jain fail_fsdev_sysfs: 3322b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3323b7c35e81SAnand Jain 33241b1d1f66SJosef Bacik fail_block_groups: 332554067ae9SJosef Bacik btrfs_put_block_group_cache(fs_info); 3326af31f5e5SChris Mason 3327af31f5e5SChris Mason fail_tree_roots: 3328af31f5e5SChris Mason free_root_pointers(fs_info, 1); 33292b8195bbSMiao Xie invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3330af31f5e5SChris Mason 333139279cc3SChris Mason fail_sb_buffer: 33327abadb64SLiu Bo btrfs_stop_all_workers(fs_info); 33335cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 333416cdcec7SMiao Xie fail_alloc: 33354543df7eSChris Mason fail_iput: 3336586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 3337586e46e2SIlya Dryomov 33384543df7eSChris Mason iput(fs_info->btree_inode); 3339c404e0dcSMiao Xie fail_bio_counter: 33407f8d236aSDavid Sterba percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 3341963d678bSMiao Xie fail_delalloc_bytes: 3342963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 3343e2d84521SMiao Xie fail_dirty_metadata_bytes: 3344e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 33454297ff84SJosef Bacik fail_dio_bytes: 33464297ff84SJosef Bacik percpu_counter_destroy(&fs_info->dio_bytes); 334776dda93cSYan, Zheng fail_srcu: 334876dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 33497e662854SQinghuang Feng fail: 335053b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 3351586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 3352ad2b2c80SAl Viro return err; 3353af31f5e5SChris Mason 3354af31f5e5SChris Mason recovery_tree_root: 33550b246afaSJeff Mahoney if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) 3356af31f5e5SChris Mason goto fail_tree_roots; 3357af31f5e5SChris Mason 3358af31f5e5SChris Mason free_root_pointers(fs_info, 0); 3359af31f5e5SChris Mason 3360af31f5e5SChris Mason /* don't use the log in recovery mode, it won't be valid */ 3361af31f5e5SChris Mason btrfs_set_super_log_root(disk_super, 0); 3362af31f5e5SChris Mason 3363af31f5e5SChris Mason /* we can't trust the free space cache either */ 3364af31f5e5SChris Mason btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 3365af31f5e5SChris Mason 3366af31f5e5SChris Mason ret = next_root_backup(fs_info, fs_info->super_copy, 3367af31f5e5SChris Mason &num_backups_tried, &backup_index); 3368af31f5e5SChris Mason if (ret == -1) 3369af31f5e5SChris Mason goto fail_block_groups; 3370af31f5e5SChris Mason goto retry_root_backup; 3371eb60ceacSChris Mason } 3372663faf9fSMasami Hiramatsu ALLOW_ERROR_INJECTION(open_ctree, ERRNO); 3373eb60ceacSChris Mason 3374f2984462SChris Mason static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate) 3375f2984462SChris Mason { 3376f2984462SChris Mason if (uptodate) { 3377f2984462SChris Mason set_buffer_uptodate(bh); 3378f2984462SChris Mason } else { 3379442a4f63SStefan Behrens struct btrfs_device *device = (struct btrfs_device *) 3380442a4f63SStefan Behrens bh->b_private; 3381442a4f63SStefan Behrens 3382fb456252SJeff Mahoney btrfs_warn_rl_in_rcu(device->fs_info, 3383b14af3b4SDavid Sterba "lost page write due to IO error on %s", 3384606686eeSJosef Bacik rcu_str_deref(device->name)); 338501327610SNicholas D Steeves /* note, we don't set_buffer_write_io_error because we have 33861259ab75SChris Mason * our own ways of dealing with the IO errors 33871259ab75SChris Mason */ 3388f2984462SChris Mason clear_buffer_uptodate(bh); 3389442a4f63SStefan Behrens btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS); 3390f2984462SChris Mason } 3391f2984462SChris Mason unlock_buffer(bh); 3392f2984462SChris Mason put_bh(bh); 3393f2984462SChris Mason } 3394f2984462SChris Mason 339529c36d72SAnand Jain int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num, 339629c36d72SAnand Jain struct buffer_head **bh_ret) 339729c36d72SAnand Jain { 339829c36d72SAnand Jain struct buffer_head *bh; 339929c36d72SAnand Jain struct btrfs_super_block *super; 340029c36d72SAnand Jain u64 bytenr; 340129c36d72SAnand Jain 340229c36d72SAnand Jain bytenr = btrfs_sb_offset(copy_num); 340329c36d72SAnand Jain if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode)) 340429c36d72SAnand Jain return -EINVAL; 340529c36d72SAnand Jain 34069f6d2510SDavid Sterba bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE); 340729c36d72SAnand Jain /* 340829c36d72SAnand Jain * If we fail to read from the underlying devices, as of now 340929c36d72SAnand Jain * the best option we have is to mark it EIO. 341029c36d72SAnand Jain */ 341129c36d72SAnand Jain if (!bh) 341229c36d72SAnand Jain return -EIO; 341329c36d72SAnand Jain 341429c36d72SAnand Jain super = (struct btrfs_super_block *)bh->b_data; 341529c36d72SAnand Jain if (btrfs_super_bytenr(super) != bytenr || 341629c36d72SAnand Jain btrfs_super_magic(super) != BTRFS_MAGIC) { 341729c36d72SAnand Jain brelse(bh); 341829c36d72SAnand Jain return -EINVAL; 341929c36d72SAnand Jain } 342029c36d72SAnand Jain 342129c36d72SAnand Jain *bh_ret = bh; 342229c36d72SAnand Jain return 0; 342329c36d72SAnand Jain } 342429c36d72SAnand Jain 342529c36d72SAnand Jain 3426a512bbf8SYan Zheng struct buffer_head *btrfs_read_dev_super(struct block_device *bdev) 3427a512bbf8SYan Zheng { 3428a512bbf8SYan Zheng struct buffer_head *bh; 3429a512bbf8SYan Zheng struct buffer_head *latest = NULL; 3430a512bbf8SYan Zheng struct btrfs_super_block *super; 3431a512bbf8SYan Zheng int i; 3432a512bbf8SYan Zheng u64 transid = 0; 343392fc03fbSAnand Jain int ret = -EINVAL; 3434a512bbf8SYan Zheng 3435a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 3436a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 3437a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 3438a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3439a512bbf8SYan Zheng */ 3440a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 344129c36d72SAnand Jain ret = btrfs_read_dev_one_super(bdev, i, &bh); 344229c36d72SAnand Jain if (ret) 3443a512bbf8SYan Zheng continue; 3444a512bbf8SYan Zheng 3445a512bbf8SYan Zheng super = (struct btrfs_super_block *)bh->b_data; 3446a512bbf8SYan Zheng 3447a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 3448a512bbf8SYan Zheng brelse(latest); 3449a512bbf8SYan Zheng latest = bh; 3450a512bbf8SYan Zheng transid = btrfs_super_generation(super); 3451a512bbf8SYan Zheng } else { 3452a512bbf8SYan Zheng brelse(bh); 3453a512bbf8SYan Zheng } 3454a512bbf8SYan Zheng } 345592fc03fbSAnand Jain 345692fc03fbSAnand Jain if (!latest) 345792fc03fbSAnand Jain return ERR_PTR(ret); 345892fc03fbSAnand Jain 3459a512bbf8SYan Zheng return latest; 3460a512bbf8SYan Zheng } 3461a512bbf8SYan Zheng 34624eedeb75SHisashi Hifumi /* 3463abbb3b8eSDavid Sterba * Write superblock @sb to the @device. Do not wait for completion, all the 3464abbb3b8eSDavid Sterba * buffer heads we write are pinned. 34654eedeb75SHisashi Hifumi * 3466abbb3b8eSDavid Sterba * Write @max_mirrors copies of the superblock, where 0 means default that fit 3467abbb3b8eSDavid Sterba * the expected device size at commit time. Note that max_mirrors must be 3468abbb3b8eSDavid Sterba * same for write and wait phases. 34694eedeb75SHisashi Hifumi * 3470abbb3b8eSDavid Sterba * Return number of errors when buffer head is not found or submission fails. 34714eedeb75SHisashi Hifumi */ 3472a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 3473abbb3b8eSDavid Sterba struct btrfs_super_block *sb, int max_mirrors) 3474a512bbf8SYan Zheng { 3475a512bbf8SYan Zheng struct buffer_head *bh; 3476a512bbf8SYan Zheng int i; 3477a512bbf8SYan Zheng int ret; 3478a512bbf8SYan Zheng int errors = 0; 3479a512bbf8SYan Zheng u32 crc; 3480a512bbf8SYan Zheng u64 bytenr; 34811b9e619cSOmar Sandoval int op_flags; 3482a512bbf8SYan Zheng 3483a512bbf8SYan Zheng if (max_mirrors == 0) 3484a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3485a512bbf8SYan Zheng 3486a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 3487a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 3488935e5cc9SMiao Xie if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3489935e5cc9SMiao Xie device->commit_total_bytes) 3490a512bbf8SYan Zheng break; 3491a512bbf8SYan Zheng 3492a512bbf8SYan Zheng btrfs_set_super_bytenr(sb, bytenr); 3493a512bbf8SYan Zheng 3494a512bbf8SYan Zheng crc = ~(u32)0; 3495abbb3b8eSDavid Sterba crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc, 3496abbb3b8eSDavid Sterba BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE); 3497a512bbf8SYan Zheng btrfs_csum_final(crc, sb->csum); 3498a512bbf8SYan Zheng 3499abbb3b8eSDavid Sterba /* One reference for us, and we leave it for the caller */ 35009f6d2510SDavid Sterba bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, 3501a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 3502634554dcSJosef Bacik if (!bh) { 3503fb456252SJeff Mahoney btrfs_err(device->fs_info, 3504f14d104dSDavid Sterba "couldn't get super buffer head for bytenr %llu", 3505f14d104dSDavid Sterba bytenr); 3506634554dcSJosef Bacik errors++; 3507634554dcSJosef Bacik continue; 3508634554dcSJosef Bacik } 3509634554dcSJosef Bacik 3510a512bbf8SYan Zheng memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE); 3511a512bbf8SYan Zheng 35124eedeb75SHisashi Hifumi /* one reference for submit_bh */ 3513a512bbf8SYan Zheng get_bh(bh); 35144eedeb75SHisashi Hifumi 35154eedeb75SHisashi Hifumi set_buffer_uptodate(bh); 3516a512bbf8SYan Zheng lock_buffer(bh); 3517a512bbf8SYan Zheng bh->b_end_io = btrfs_end_buffer_write_sync; 3518442a4f63SStefan Behrens bh->b_private = device; 3519a512bbf8SYan Zheng 3520387125fcSChris Mason /* 3521387125fcSChris Mason * we fua the first super. The others we allow 3522387125fcSChris Mason * to go down lazy. 3523387125fcSChris Mason */ 35241b9e619cSOmar Sandoval op_flags = REQ_SYNC | REQ_META | REQ_PRIO; 35251b9e619cSOmar Sandoval if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) 35261b9e619cSOmar Sandoval op_flags |= REQ_FUA; 35271b9e619cSOmar Sandoval ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh); 35284eedeb75SHisashi Hifumi if (ret) 3529a512bbf8SYan Zheng errors++; 3530a512bbf8SYan Zheng } 3531a512bbf8SYan Zheng return errors < i ? 0 : -1; 3532a512bbf8SYan Zheng } 3533a512bbf8SYan Zheng 3534387125fcSChris Mason /* 3535abbb3b8eSDavid Sterba * Wait for write completion of superblocks done by write_dev_supers, 3536abbb3b8eSDavid Sterba * @max_mirrors same for write and wait phases. 3537abbb3b8eSDavid Sterba * 3538abbb3b8eSDavid Sterba * Return number of errors when buffer head is not found or not marked up to 3539abbb3b8eSDavid Sterba * date. 3540abbb3b8eSDavid Sterba */ 3541abbb3b8eSDavid Sterba static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) 3542abbb3b8eSDavid Sterba { 3543abbb3b8eSDavid Sterba struct buffer_head *bh; 3544abbb3b8eSDavid Sterba int i; 3545abbb3b8eSDavid Sterba int errors = 0; 3546b6a535faSHoward McLauchlan bool primary_failed = false; 3547abbb3b8eSDavid Sterba u64 bytenr; 3548abbb3b8eSDavid Sterba 3549abbb3b8eSDavid Sterba if (max_mirrors == 0) 3550abbb3b8eSDavid Sterba max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3551abbb3b8eSDavid Sterba 3552abbb3b8eSDavid Sterba for (i = 0; i < max_mirrors; i++) { 3553abbb3b8eSDavid Sterba bytenr = btrfs_sb_offset(i); 3554abbb3b8eSDavid Sterba if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3555abbb3b8eSDavid Sterba device->commit_total_bytes) 3556abbb3b8eSDavid Sterba break; 3557abbb3b8eSDavid Sterba 35589f6d2510SDavid Sterba bh = __find_get_block(device->bdev, 35599f6d2510SDavid Sterba bytenr / BTRFS_BDEV_BLOCKSIZE, 3560abbb3b8eSDavid Sterba BTRFS_SUPER_INFO_SIZE); 3561abbb3b8eSDavid Sterba if (!bh) { 3562abbb3b8eSDavid Sterba errors++; 3563b6a535faSHoward McLauchlan if (i == 0) 3564b6a535faSHoward McLauchlan primary_failed = true; 3565abbb3b8eSDavid Sterba continue; 3566abbb3b8eSDavid Sterba } 3567abbb3b8eSDavid Sterba wait_on_buffer(bh); 3568b6a535faSHoward McLauchlan if (!buffer_uptodate(bh)) { 3569abbb3b8eSDavid Sterba errors++; 3570b6a535faSHoward McLauchlan if (i == 0) 3571b6a535faSHoward McLauchlan primary_failed = true; 3572b6a535faSHoward McLauchlan } 3573abbb3b8eSDavid Sterba 3574abbb3b8eSDavid Sterba /* drop our reference */ 3575abbb3b8eSDavid Sterba brelse(bh); 3576abbb3b8eSDavid Sterba 3577abbb3b8eSDavid Sterba /* drop the reference from the writing run */ 3578abbb3b8eSDavid Sterba brelse(bh); 3579abbb3b8eSDavid Sterba } 3580abbb3b8eSDavid Sterba 3581b6a535faSHoward McLauchlan /* log error, force error return */ 3582b6a535faSHoward McLauchlan if (primary_failed) { 3583b6a535faSHoward McLauchlan btrfs_err(device->fs_info, "error writing primary super block to device %llu", 3584b6a535faSHoward McLauchlan device->devid); 3585b6a535faSHoward McLauchlan return -1; 3586b6a535faSHoward McLauchlan } 3587b6a535faSHoward McLauchlan 3588abbb3b8eSDavid Sterba return errors < i ? 0 : -1; 3589abbb3b8eSDavid Sterba } 3590abbb3b8eSDavid Sterba 3591abbb3b8eSDavid Sterba /* 3592387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 3593387125fcSChris Mason * for the barrier when it is done 3594387125fcSChris Mason */ 35954246a0b6SChristoph Hellwig static void btrfs_end_empty_barrier(struct bio *bio) 3596387125fcSChris Mason { 3597387125fcSChris Mason complete(bio->bi_private); 3598387125fcSChris Mason } 3599387125fcSChris Mason 3600387125fcSChris Mason /* 36014fc6441aSAnand Jain * Submit a flush request to the device if it supports it. Error handling is 36024fc6441aSAnand Jain * done in the waiting counterpart. 3603387125fcSChris Mason */ 36044fc6441aSAnand Jain static void write_dev_flush(struct btrfs_device *device) 3605387125fcSChris Mason { 3606c2a9c7abSAnand Jain struct request_queue *q = bdev_get_queue(device->bdev); 3607e0ae9994SDavid Sterba struct bio *bio = device->flush_bio; 3608387125fcSChris Mason 3609c2a9c7abSAnand Jain if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) 36104fc6441aSAnand Jain return; 3611387125fcSChris Mason 3612e0ae9994SDavid Sterba bio_reset(bio); 3613387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 361474d46992SChristoph Hellwig bio_set_dev(bio, device->bdev); 36158d910125SJan Kara bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH; 3616387125fcSChris Mason init_completion(&device->flush_wait); 3617387125fcSChris Mason bio->bi_private = &device->flush_wait; 3618387125fcSChris Mason 361943a01111SLu Fengqi btrfsic_submit_bio(bio); 36201c3063b6SAnand Jain set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 36214fc6441aSAnand Jain } 3622387125fcSChris Mason 36234fc6441aSAnand Jain /* 36244fc6441aSAnand Jain * If the flush bio has been submitted by write_dev_flush, wait for it. 36254fc6441aSAnand Jain */ 36268c27cb35SLinus Torvalds static blk_status_t wait_dev_flush(struct btrfs_device *device) 36274fc6441aSAnand Jain { 36284fc6441aSAnand Jain struct bio *bio = device->flush_bio; 36294fc6441aSAnand Jain 36301c3063b6SAnand Jain if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) 363158efbc9fSOmar Sandoval return BLK_STS_OK; 36324fc6441aSAnand Jain 36331c3063b6SAnand Jain clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 36342980d574SDavid Sterba wait_for_completion_io(&device->flush_wait); 36354fc6441aSAnand Jain 36368c27cb35SLinus Torvalds return bio->bi_status; 3637387125fcSChris Mason } 3638387125fcSChris Mason 3639d10b82feSQu Wenruo static int check_barrier_error(struct btrfs_fs_info *fs_info) 3640401b41e5SAnand Jain { 36416528b99dSAnand Jain if (!btrfs_check_rw_degradable(fs_info, NULL)) 3642401b41e5SAnand Jain return -EIO; 3643387125fcSChris Mason return 0; 3644387125fcSChris Mason } 3645387125fcSChris Mason 3646387125fcSChris Mason /* 3647387125fcSChris Mason * send an empty flush down to each device in parallel, 3648387125fcSChris Mason * then wait for them 3649387125fcSChris Mason */ 3650387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 3651387125fcSChris Mason { 3652387125fcSChris Mason struct list_head *head; 3653387125fcSChris Mason struct btrfs_device *dev; 36545af3e8ccSStefan Behrens int errors_wait = 0; 36554e4cbee9SChristoph Hellwig blk_status_t ret; 3656387125fcSChris Mason 36571538e6c5SDavid Sterba lockdep_assert_held(&info->fs_devices->device_list_mutex); 3658387125fcSChris Mason /* send down all the barriers */ 3659387125fcSChris Mason head = &info->fs_devices->devices; 36601538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3661e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3662f88ba6a2SHidetoshi Seto continue; 3663cea7c8bfSAnand Jain if (!dev->bdev) 3664387125fcSChris Mason continue; 3665e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3666ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3667387125fcSChris Mason continue; 3668387125fcSChris Mason 36694fc6441aSAnand Jain write_dev_flush(dev); 367058efbc9fSOmar Sandoval dev->last_flush_error = BLK_STS_OK; 3671387125fcSChris Mason } 3672387125fcSChris Mason 3673387125fcSChris Mason /* wait for all the barriers */ 36741538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3675e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3676f88ba6a2SHidetoshi Seto continue; 3677387125fcSChris Mason if (!dev->bdev) { 36785af3e8ccSStefan Behrens errors_wait++; 3679387125fcSChris Mason continue; 3680387125fcSChris Mason } 3681e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3682ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3683387125fcSChris Mason continue; 3684387125fcSChris Mason 36854fc6441aSAnand Jain ret = wait_dev_flush(dev); 3686401b41e5SAnand Jain if (ret) { 3687401b41e5SAnand Jain dev->last_flush_error = ret; 368866b4993eSDavid Sterba btrfs_dev_stat_inc_and_print(dev, 368966b4993eSDavid Sterba BTRFS_DEV_STAT_FLUSH_ERRS); 36905af3e8ccSStefan Behrens errors_wait++; 3691387125fcSChris Mason } 3692401b41e5SAnand Jain } 3693401b41e5SAnand Jain 3694cea7c8bfSAnand Jain if (errors_wait) { 3695401b41e5SAnand Jain /* 3696401b41e5SAnand Jain * At some point we need the status of all disks 3697401b41e5SAnand Jain * to arrive at the volume status. So error checking 3698401b41e5SAnand Jain * is being pushed to a separate loop. 3699401b41e5SAnand Jain */ 3700d10b82feSQu Wenruo return check_barrier_error(info); 3701401b41e5SAnand Jain } 3702387125fcSChris Mason return 0; 3703387125fcSChris Mason } 3704387125fcSChris Mason 3705943c6e99SZhao Lei int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 3706943c6e99SZhao Lei { 37078789f4feSZhao Lei int raid_type; 37088789f4feSZhao Lei int min_tolerated = INT_MAX; 3709943c6e99SZhao Lei 37108789f4feSZhao Lei if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 37118789f4feSZhao Lei (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 37128789f4feSZhao Lei min_tolerated = min(min_tolerated, 37138789f4feSZhao Lei btrfs_raid_array[BTRFS_RAID_SINGLE]. 37148789f4feSZhao Lei tolerated_failures); 3715943c6e99SZhao Lei 37168789f4feSZhao Lei for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 37178789f4feSZhao Lei if (raid_type == BTRFS_RAID_SINGLE) 37188789f4feSZhao Lei continue; 371941a6e891SAnand Jain if (!(flags & btrfs_raid_array[raid_type].bg_flag)) 37208789f4feSZhao Lei continue; 37218789f4feSZhao Lei min_tolerated = min(min_tolerated, 37228789f4feSZhao Lei btrfs_raid_array[raid_type]. 37238789f4feSZhao Lei tolerated_failures); 37248789f4feSZhao Lei } 3725943c6e99SZhao Lei 37268789f4feSZhao Lei if (min_tolerated == INT_MAX) { 3727ab8d0fc4SJeff Mahoney pr_warn("BTRFS: unknown raid flag: %llu", flags); 37288789f4feSZhao Lei min_tolerated = 0; 37298789f4feSZhao Lei } 37308789f4feSZhao Lei 37318789f4feSZhao Lei return min_tolerated; 3732943c6e99SZhao Lei } 3733943c6e99SZhao Lei 3734eece6a9cSDavid Sterba int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) 3735f2984462SChris Mason { 3736e5e9a520SChris Mason struct list_head *head; 3737f2984462SChris Mason struct btrfs_device *dev; 3738a061fc8dSChris Mason struct btrfs_super_block *sb; 3739f2984462SChris Mason struct btrfs_dev_item *dev_item; 3740f2984462SChris Mason int ret; 3741f2984462SChris Mason int do_barriers; 3742a236aed1SChris Mason int max_errors; 3743a236aed1SChris Mason int total_errors = 0; 3744a061fc8dSChris Mason u64 flags; 3745f2984462SChris Mason 37460b246afaSJeff Mahoney do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); 3747fed3b381SLiu Bo 3748fed3b381SLiu Bo /* 3749fed3b381SLiu Bo * max_mirrors == 0 indicates we're from commit_transaction, 3750fed3b381SLiu Bo * not from fsync where the tree roots in fs_info have not 3751fed3b381SLiu Bo * been consistent on disk. 3752fed3b381SLiu Bo */ 3753fed3b381SLiu Bo if (max_mirrors == 0) 37540b246afaSJeff Mahoney backup_super_roots(fs_info); 3755f2984462SChris Mason 37560b246afaSJeff Mahoney sb = fs_info->super_for_commit; 3757a061fc8dSChris Mason dev_item = &sb->dev_item; 3758e5e9a520SChris Mason 37590b246afaSJeff Mahoney mutex_lock(&fs_info->fs_devices->device_list_mutex); 37600b246afaSJeff Mahoney head = &fs_info->fs_devices->devices; 37610b246afaSJeff Mahoney max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; 3762387125fcSChris Mason 37635af3e8ccSStefan Behrens if (do_barriers) { 37640b246afaSJeff Mahoney ret = barrier_all_devices(fs_info); 37655af3e8ccSStefan Behrens if (ret) { 37665af3e8ccSStefan Behrens mutex_unlock( 37670b246afaSJeff Mahoney &fs_info->fs_devices->device_list_mutex); 37680b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 37695af3e8ccSStefan Behrens "errors while submitting device barriers."); 37705af3e8ccSStefan Behrens return ret; 37715af3e8ccSStefan Behrens } 37725af3e8ccSStefan Behrens } 3773387125fcSChris Mason 37741538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3775dfe25020SChris Mason if (!dev->bdev) { 3776dfe25020SChris Mason total_errors++; 3777dfe25020SChris Mason continue; 3778dfe25020SChris Mason } 3779e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3780ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3781dfe25020SChris Mason continue; 3782dfe25020SChris Mason 37832b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 3784a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 3785a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 37867df69d3eSMiao Xie btrfs_set_stack_device_total_bytes(dev_item, 3787935e5cc9SMiao Xie dev->commit_total_bytes); 3788ce7213c7SMiao Xie btrfs_set_stack_device_bytes_used(dev_item, 3789ce7213c7SMiao Xie dev->commit_bytes_used); 3790a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 3791a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 3792a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 3793a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 37947239ff4bSNikolay Borisov memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, 37957239ff4bSNikolay Borisov BTRFS_FSID_SIZE); 3796a512bbf8SYan Zheng 3797a061fc8dSChris Mason flags = btrfs_super_flags(sb); 3798a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 3799f2984462SChris Mason 380075cb857dSQu Wenruo ret = btrfs_validate_write_super(fs_info, sb); 380175cb857dSQu Wenruo if (ret < 0) { 380275cb857dSQu Wenruo mutex_unlock(&fs_info->fs_devices->device_list_mutex); 380375cb857dSQu Wenruo btrfs_handle_fs_error(fs_info, -EUCLEAN, 380475cb857dSQu Wenruo "unexpected superblock corruption detected"); 380575cb857dSQu Wenruo return -EUCLEAN; 380675cb857dSQu Wenruo } 380775cb857dSQu Wenruo 3808abbb3b8eSDavid Sterba ret = write_dev_supers(dev, sb, max_mirrors); 3809a236aed1SChris Mason if (ret) 3810a236aed1SChris Mason total_errors++; 3811f2984462SChris Mason } 3812a236aed1SChris Mason if (total_errors > max_errors) { 38130b246afaSJeff Mahoney btrfs_err(fs_info, "%d errors while writing supers", 3814d397712bSChris Mason total_errors); 38150b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 381679787eaaSJeff Mahoney 38179d565ba4SStefan Behrens /* FUA is masked off if unsupported and can't be the reason */ 38180b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 38190b246afaSJeff Mahoney "%d errors while writing supers", 38200b246afaSJeff Mahoney total_errors); 38219d565ba4SStefan Behrens return -EIO; 3822a236aed1SChris Mason } 3823f2984462SChris Mason 3824a512bbf8SYan Zheng total_errors = 0; 38251538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3826dfe25020SChris Mason if (!dev->bdev) 3827dfe25020SChris Mason continue; 3828e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3829ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3830dfe25020SChris Mason continue; 3831dfe25020SChris Mason 3832abbb3b8eSDavid Sterba ret = wait_dev_supers(dev, max_mirrors); 3833a512bbf8SYan Zheng if (ret) 38341259ab75SChris Mason total_errors++; 3835f2984462SChris Mason } 38360b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 3837a236aed1SChris Mason if (total_errors > max_errors) { 38380b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 38390b246afaSJeff Mahoney "%d errors while writing supers", 38400b246afaSJeff Mahoney total_errors); 384179787eaaSJeff Mahoney return -EIO; 3842a236aed1SChris Mason } 3843f2984462SChris Mason return 0; 3844f2984462SChris Mason } 3845f2984462SChris Mason 3846cb517eabSMiao Xie /* Drop a fs root from the radix tree and free it. */ 3847cb517eabSMiao Xie void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 3848cb517eabSMiao Xie struct btrfs_root *root) 38492619ba1fSChris Mason { 38504df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 38512619ba1fSChris Mason radix_tree_delete(&fs_info->fs_roots_radix, 38522619ba1fSChris Mason (unsigned long)root->root_key.objectid); 38534df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 385476dda93cSYan, Zheng 385576dda93cSYan, Zheng if (btrfs_root_refs(&root->root_item) == 0) 385676dda93cSYan, Zheng synchronize_srcu(&fs_info->subvol_srcu); 385776dda93cSYan, Zheng 38581c1ea4f7SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 38593321719eSLiu Bo btrfs_free_log(NULL, root); 38601c1ea4f7SLiu Bo if (root->reloc_root) { 38611c1ea4f7SLiu Bo free_extent_buffer(root->reloc_root->node); 38621c1ea4f7SLiu Bo free_extent_buffer(root->reloc_root->commit_root); 38631c1ea4f7SLiu Bo btrfs_put_fs_root(root->reloc_root); 38641c1ea4f7SLiu Bo root->reloc_root = NULL; 38651c1ea4f7SLiu Bo } 38661c1ea4f7SLiu Bo } 38673321719eSLiu Bo 3868faa2dbf0SJosef Bacik if (root->free_ino_pinned) 3869581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_pinned); 3870faa2dbf0SJosef Bacik if (root->free_ino_ctl) 3871581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_ctl); 387284db5ccfSDavid Sterba btrfs_free_fs_root(root); 38734df27c4dSYan, Zheng } 38744df27c4dSYan, Zheng 387584db5ccfSDavid Sterba void btrfs_free_fs_root(struct btrfs_root *root) 38764df27c4dSYan, Zheng { 387757cdc8dbSDavid Sterba iput(root->ino_cache_inode); 38784df27c4dSYan, Zheng WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 38790ee5dc67SAl Viro if (root->anon_dev) 38800ee5dc67SAl Viro free_anon_bdev(root->anon_dev); 38818257b2dcSMiao Xie if (root->subv_writers) 38828257b2dcSMiao Xie btrfs_free_subvolume_writers(root->subv_writers); 38835f39d397SChris Mason free_extent_buffer(root->node); 38845f39d397SChris Mason free_extent_buffer(root->commit_root); 3885581bb050SLi Zefan kfree(root->free_ino_ctl); 3886581bb050SLi Zefan kfree(root->free_ino_pinned); 3887b0feb9d9SMiao Xie btrfs_put_fs_root(root); 38882619ba1fSChris Mason } 38892619ba1fSChris Mason 3890c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 3891c146afadSYan Zheng { 3892c146afadSYan Zheng u64 root_objectid = 0; 3893c146afadSYan Zheng struct btrfs_root *gang[8]; 389465d33fd7SQu Wenruo int i = 0; 389565d33fd7SQu Wenruo int err = 0; 389665d33fd7SQu Wenruo unsigned int ret = 0; 389765d33fd7SQu Wenruo int index; 3898c146afadSYan Zheng 3899c146afadSYan Zheng while (1) { 390065d33fd7SQu Wenruo index = srcu_read_lock(&fs_info->subvol_srcu); 3901c146afadSYan Zheng ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 3902c146afadSYan Zheng (void **)gang, root_objectid, 3903c146afadSYan Zheng ARRAY_SIZE(gang)); 390465d33fd7SQu Wenruo if (!ret) { 390565d33fd7SQu Wenruo srcu_read_unlock(&fs_info->subvol_srcu, index); 3906c146afadSYan Zheng break; 390765d33fd7SQu Wenruo } 39085d4f98a2SYan Zheng root_objectid = gang[ret - 1]->root_key.objectid + 1; 390966b4ffd1SJosef Bacik 391065d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 391165d33fd7SQu Wenruo /* Avoid to grab roots in dead_roots */ 391265d33fd7SQu Wenruo if (btrfs_root_refs(&gang[i]->root_item) == 0) { 391365d33fd7SQu Wenruo gang[i] = NULL; 391465d33fd7SQu Wenruo continue; 391565d33fd7SQu Wenruo } 391665d33fd7SQu Wenruo /* grab all the search result for later use */ 391765d33fd7SQu Wenruo gang[i] = btrfs_grab_fs_root(gang[i]); 391865d33fd7SQu Wenruo } 391965d33fd7SQu Wenruo srcu_read_unlock(&fs_info->subvol_srcu, index); 392065d33fd7SQu Wenruo 392165d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 392265d33fd7SQu Wenruo if (!gang[i]) 392365d33fd7SQu Wenruo continue; 3924c146afadSYan Zheng root_objectid = gang[i]->root_key.objectid; 392566b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(gang[i]); 392666b4ffd1SJosef Bacik if (err) 392765d33fd7SQu Wenruo break; 392865d33fd7SQu Wenruo btrfs_put_fs_root(gang[i]); 3929c146afadSYan Zheng } 3930c146afadSYan Zheng root_objectid++; 3931c146afadSYan Zheng } 393265d33fd7SQu Wenruo 393365d33fd7SQu Wenruo /* release the uncleaned roots due to error */ 393465d33fd7SQu Wenruo for (; i < ret; i++) { 393565d33fd7SQu Wenruo if (gang[i]) 393665d33fd7SQu Wenruo btrfs_put_fs_root(gang[i]); 393765d33fd7SQu Wenruo } 393865d33fd7SQu Wenruo return err; 3939c146afadSYan Zheng } 3940c146afadSYan Zheng 39416bccf3abSJeff Mahoney int btrfs_commit_super(struct btrfs_fs_info *fs_info) 3942c146afadSYan Zheng { 39436bccf3abSJeff Mahoney struct btrfs_root *root = fs_info->tree_root; 3944c146afadSYan Zheng struct btrfs_trans_handle *trans; 3945c146afadSYan Zheng 39460b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 39472ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 39480b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 39490b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 3950c71bf099SYan, Zheng 3951c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 39520b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 39530b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 3954c71bf099SYan, Zheng 39557a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 39563612b495STsutomu Itoh if (IS_ERR(trans)) 39573612b495STsutomu Itoh return PTR_ERR(trans); 39583a45bb20SJeff Mahoney return btrfs_commit_transaction(trans); 3959c146afadSYan Zheng } 3960c146afadSYan Zheng 39616bccf3abSJeff Mahoney void close_ctree(struct btrfs_fs_info *fs_info) 3962eb60ceacSChris Mason { 3963c146afadSYan Zheng int ret; 3964e089f05cSChris Mason 3965afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); 3966d6fd0ae2SOmar Sandoval /* 3967d6fd0ae2SOmar Sandoval * We don't want the cleaner to start new transactions, add more delayed 3968d6fd0ae2SOmar Sandoval * iputs, etc. while we're closing. We can't use kthread_stop() yet 3969d6fd0ae2SOmar Sandoval * because that frees the task_struct, and the transaction kthread might 3970d6fd0ae2SOmar Sandoval * still try to wake up the cleaner. 3971d6fd0ae2SOmar Sandoval */ 3972d6fd0ae2SOmar Sandoval kthread_park(fs_info->cleaner_kthread); 3973a2135011SChris Mason 39747343dd61SJustin Maggard /* wait for the qgroup rescan worker to stop */ 3975d06f23d6SJeff Mahoney btrfs_qgroup_wait_for_completion(fs_info, false); 39767343dd61SJustin Maggard 3977803b2f54SStefan Behrens /* wait for the uuid_scan task to finish */ 3978803b2f54SStefan Behrens down(&fs_info->uuid_tree_rescan_sem); 3979803b2f54SStefan Behrens /* avoid complains from lockdep et al., set sem back to initial state */ 3980803b2f54SStefan Behrens up(&fs_info->uuid_tree_rescan_sem); 3981803b2f54SStefan Behrens 3982837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 3983aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 3984837d5b6eSIlya Dryomov 39858dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 39868dabb742SStefan Behrens 3987aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 39884cb5300bSChris Mason 39894cb5300bSChris Mason /* wait for any defraggers to finish */ 39904cb5300bSChris Mason wait_event(fs_info->transaction_wait, 39914cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 39924cb5300bSChris Mason 39934cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 399426176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 39954cb5300bSChris Mason 399621c7e756SMiao Xie cancel_work_sync(&fs_info->async_reclaim_work); 399721c7e756SMiao Xie 3998bc98a42cSDavid Howells if (!sb_rdonly(fs_info->sb)) { 3999e44163e1SJeff Mahoney /* 4000d6fd0ae2SOmar Sandoval * The cleaner kthread is stopped, so do one final pass over 4001d6fd0ae2SOmar Sandoval * unused block groups. 4002e44163e1SJeff Mahoney */ 40030b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 4004e44163e1SJeff Mahoney 40056bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 4006d397712bSChris Mason if (ret) 400704892340SEric Sandeen btrfs_err(fs_info, "commit super ret %d", ret); 4008c146afadSYan Zheng } 4009ed2ff2cbSChris Mason 4010af722733SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) || 4011af722733SLiu Bo test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) 40122ff7e61eSJeff Mahoney btrfs_error_commit_super(fs_info); 4013acce952bSliubo 4014e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 4015e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 40168929ecfaSYan, Zheng 4017e187831eSJosef Bacik ASSERT(list_empty(&fs_info->delayed_iputs)); 4018afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); 4019f25784b3SYan Zheng 402004892340SEric Sandeen btrfs_free_qgroup_config(fs_info); 4021fe816d0fSNikolay Borisov ASSERT(list_empty(&fs_info->delalloc_roots)); 4022bcef60f2SArne Jansen 4023963d678bSMiao Xie if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 402404892340SEric Sandeen btrfs_info(fs_info, "at unmount delalloc count %lld", 4025963d678bSMiao Xie percpu_counter_sum(&fs_info->delalloc_bytes)); 4026b0c68f8bSChris Mason } 402731153d81SYan Zheng 40284297ff84SJosef Bacik if (percpu_counter_sum(&fs_info->dio_bytes)) 40294297ff84SJosef Bacik btrfs_info(fs_info, "at unmount dio bytes count %lld", 40304297ff84SJosef Bacik percpu_counter_sum(&fs_info->dio_bytes)); 40314297ff84SJosef Bacik 40326618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 4033b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 40345ac1d209SJeff Mahoney 4035faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 4036d10c5f31SChris Mason 40371a4319ccSLiu Bo btrfs_put_block_group_cache(fs_info); 40381a4319ccSLiu Bo 4039de348ee0SWang Shilong /* 4040de348ee0SWang Shilong * we must make sure there is not any read request to 4041de348ee0SWang Shilong * submit after we stopping all workers. 4042de348ee0SWang Shilong */ 4043de348ee0SWang Shilong invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 404496192499SJosef Bacik btrfs_stop_all_workers(fs_info); 404596192499SJosef Bacik 40465cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 40475cdd7db6SFilipe Manana 4048afcdd129SJosef Bacik clear_bit(BTRFS_FS_OPEN, &fs_info->flags); 404913e6c37bSJosef Bacik free_root_pointers(fs_info, 1); 40509ad6b7bcSChris Mason 405113e6c37bSJosef Bacik iput(fs_info->btree_inode); 4052d6bfde87SChris Mason 405321adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 40540b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) 40552ff7e61eSJeff Mahoney btrfsic_unmount(fs_info->fs_devices); 405621adbd5cSStefan Behrens #endif 405721adbd5cSStefan Behrens 40580b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 405968c94e55SNikolay Borisov btrfs_close_devices(fs_info->fs_devices); 4060b248a415SChris Mason 4061e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 4062963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 40634297ff84SJosef Bacik percpu_counter_destroy(&fs_info->dio_bytes); 40647f8d236aSDavid Sterba percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 406576dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 40660b86a832SChris Mason 406753b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 4068fd708b81SJosef Bacik btrfs_free_ref_cache(fs_info); 4069eb60ceacSChris Mason } 4070eb60ceacSChris Mason 4071b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 4072b9fab919SChris Mason int atomic) 4073ccd467d6SChris Mason { 40741259ab75SChris Mason int ret; 4075727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 40761259ab75SChris Mason 40770b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 40781259ab75SChris Mason if (!ret) 40791259ab75SChris Mason return ret; 40801259ab75SChris Mason 40811259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 4082b9fab919SChris Mason parent_transid, atomic); 4083b9fab919SChris Mason if (ret == -EAGAIN) 4084b9fab919SChris Mason return ret; 40851259ab75SChris Mason return !ret; 40865f39d397SChris Mason } 40876702ed49SChris Mason 40885f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 40895f39d397SChris Mason { 40900b246afaSJeff Mahoney struct btrfs_fs_info *fs_info; 409106ea65a3SJosef Bacik struct btrfs_root *root; 40925f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 4093b9473439SChris Mason int was_dirty; 4094b4ce94deSChris Mason 409506ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 409606ea65a3SJosef Bacik /* 409706ea65a3SJosef Bacik * This is a fast path so only do this check if we have sanity tests 409852042d8eSAndrea Gelmini * enabled. Normal people shouldn't be using unmapped buffers as dirty 409906ea65a3SJosef Bacik * outside of the sanity tests. 410006ea65a3SJosef Bacik */ 4101b0132a3bSNikolay Borisov if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) 410206ea65a3SJosef Bacik return; 410306ea65a3SJosef Bacik #endif 410406ea65a3SJosef Bacik root = BTRFS_I(buf->pages[0]->mapping->host)->root; 41050b246afaSJeff Mahoney fs_info = root->fs_info; 4106b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 41070b246afaSJeff Mahoney if (transid != fs_info->generation) 41085d163e0eSJeff Mahoney WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", 41090b246afaSJeff Mahoney buf->start, transid, fs_info->generation); 41100b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 4111e2d84521SMiao Xie if (!was_dirty) 4112104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 4113e2d84521SMiao Xie buf->len, 41140b246afaSJeff Mahoney fs_info->dirty_metadata_batch); 41151f21ef0aSFilipe Manana #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 411669fc6cbbSQu Wenruo /* 411769fc6cbbSQu Wenruo * Since btrfs_mark_buffer_dirty() can be called with item pointer set 411869fc6cbbSQu Wenruo * but item data not updated. 411969fc6cbbSQu Wenruo * So here we should only check item pointers, not item data. 412069fc6cbbSQu Wenruo */ 412169fc6cbbSQu Wenruo if (btrfs_header_level(buf) == 0 && 4122cfdaad5eSDavid Sterba btrfs_check_leaf_relaxed(buf)) { 4123a4f78750SDavid Sterba btrfs_print_leaf(buf); 41241f21ef0aSFilipe Manana ASSERT(0); 41251f21ef0aSFilipe Manana } 41261f21ef0aSFilipe Manana #endif 4127eb60ceacSChris Mason } 4128eb60ceacSChris Mason 41292ff7e61eSJeff Mahoney static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, 4130b53d3f5dSLiu Bo int flush_delayed) 413135b7e476SChris Mason { 4132188de649SChris Mason /* 4133188de649SChris Mason * looks as though older kernels can get into trouble with 4134188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 4135188de649SChris Mason */ 4136e2d84521SMiao Xie int ret; 4137d6bfde87SChris Mason 41386933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 4139d6bfde87SChris Mason return; 4140d6bfde87SChris Mason 4141b53d3f5dSLiu Bo if (flush_delayed) 41422ff7e61eSJeff Mahoney btrfs_balance_delayed_items(fs_info); 414316cdcec7SMiao Xie 4144d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 4145d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 4146d814a491SEthan Lien fs_info->dirty_metadata_batch); 4147e2d84521SMiao Xie if (ret > 0) { 41480b246afaSJeff Mahoney balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); 414916cdcec7SMiao Xie } 415016cdcec7SMiao Xie } 415116cdcec7SMiao Xie 41522ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) 415316cdcec7SMiao Xie { 41542ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 1); 415535b7e476SChris Mason } 4156b53d3f5dSLiu Bo 41572ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) 4158b53d3f5dSLiu Bo { 41592ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 0); 4160d6bfde87SChris Mason } 41616b80053dSChris Mason 4162581c1760SQu Wenruo int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level, 4163581c1760SQu Wenruo struct btrfs_key *first_key) 41646b80053dSChris Mason { 41655ab12d1fSDavid Sterba return btree_read_extent_buffer_pages(buf, parent_transid, 4166581c1760SQu Wenruo level, first_key); 41676b80053dSChris Mason } 41680da5468fSChris Mason 41692ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) 4170acce952bSliubo { 4171fe816d0fSNikolay Borisov /* cleanup FS via transaction */ 4172fe816d0fSNikolay Borisov btrfs_cleanup_transaction(fs_info); 4173fe816d0fSNikolay Borisov 41740b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 41752ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 41760b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 4177acce952bSliubo 41780b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 41790b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4180acce952bSliubo } 4181acce952bSliubo 4182143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4183acce952bSliubo { 4184acce952bSliubo struct btrfs_ordered_extent *ordered; 4185acce952bSliubo 4186199c2a9cSMiao Xie spin_lock(&root->ordered_extent_lock); 4187779880efSJosef Bacik /* 4188779880efSJosef Bacik * This will just short circuit the ordered completion stuff which will 4189779880efSJosef Bacik * make sure the ordered extent gets properly cleaned up. 4190779880efSJosef Bacik */ 4191199c2a9cSMiao Xie list_for_each_entry(ordered, &root->ordered_extents, 4192779880efSJosef Bacik root_extent_list) 4193779880efSJosef Bacik set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4194199c2a9cSMiao Xie spin_unlock(&root->ordered_extent_lock); 4195199c2a9cSMiao Xie } 4196199c2a9cSMiao Xie 4197199c2a9cSMiao Xie static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4198199c2a9cSMiao Xie { 4199199c2a9cSMiao Xie struct btrfs_root *root; 4200199c2a9cSMiao Xie struct list_head splice; 4201199c2a9cSMiao Xie 4202199c2a9cSMiao Xie INIT_LIST_HEAD(&splice); 4203199c2a9cSMiao Xie 4204199c2a9cSMiao Xie spin_lock(&fs_info->ordered_root_lock); 4205199c2a9cSMiao Xie list_splice_init(&fs_info->ordered_roots, &splice); 4206199c2a9cSMiao Xie while (!list_empty(&splice)) { 4207199c2a9cSMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4208199c2a9cSMiao Xie ordered_root); 42091de2cfdeSJosef Bacik list_move_tail(&root->ordered_root, 42101de2cfdeSJosef Bacik &fs_info->ordered_roots); 4211199c2a9cSMiao Xie 42122a85d9caSLiu Bo spin_unlock(&fs_info->ordered_root_lock); 4213199c2a9cSMiao Xie btrfs_destroy_ordered_extents(root); 4214199c2a9cSMiao Xie 42152a85d9caSLiu Bo cond_resched(); 42162a85d9caSLiu Bo spin_lock(&fs_info->ordered_root_lock); 4217199c2a9cSMiao Xie } 4218199c2a9cSMiao Xie spin_unlock(&fs_info->ordered_root_lock); 421974d5d229SJosef Bacik 422074d5d229SJosef Bacik /* 422174d5d229SJosef Bacik * We need this here because if we've been flipped read-only we won't 422274d5d229SJosef Bacik * get sync() from the umount, so we need to make sure any ordered 422374d5d229SJosef Bacik * extents that haven't had their dirty pages IO start writeout yet 422474d5d229SJosef Bacik * actually get run and error out properly. 422574d5d229SJosef Bacik */ 422674d5d229SJosef Bacik btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 4227acce952bSliubo } 4228acce952bSliubo 422935a3621bSStefan Behrens static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 42302ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4231acce952bSliubo { 4232acce952bSliubo struct rb_node *node; 4233acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 4234acce952bSliubo struct btrfs_delayed_ref_node *ref; 4235acce952bSliubo int ret = 0; 4236acce952bSliubo 4237acce952bSliubo delayed_refs = &trans->delayed_refs; 4238acce952bSliubo 4239acce952bSliubo spin_lock(&delayed_refs->lock); 4240d7df2c79SJosef Bacik if (atomic_read(&delayed_refs->num_entries) == 0) { 4241cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 42420b246afaSJeff Mahoney btrfs_info(fs_info, "delayed_refs has NO entry"); 4243acce952bSliubo return ret; 4244acce952bSliubo } 4245acce952bSliubo 42465c9d028bSLiu Bo while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { 4247d7df2c79SJosef Bacik struct btrfs_delayed_ref_head *head; 42480e0adbcfSJosef Bacik struct rb_node *n; 4249e78417d1SJosef Bacik bool pin_bytes = false; 4250acce952bSliubo 4251d7df2c79SJosef Bacik head = rb_entry(node, struct btrfs_delayed_ref_head, 4252d7df2c79SJosef Bacik href_node); 42533069bd26SJosef Bacik if (btrfs_delayed_ref_lock(delayed_refs, head)) 4254b939d1abSJosef Bacik continue; 42553069bd26SJosef Bacik 4256d7df2c79SJosef Bacik spin_lock(&head->lock); 4257e3d03965SLiu Bo while ((n = rb_first_cached(&head->ref_tree)) != NULL) { 42580e0adbcfSJosef Bacik ref = rb_entry(n, struct btrfs_delayed_ref_node, 42590e0adbcfSJosef Bacik ref_node); 4260d7df2c79SJosef Bacik ref->in_tree = 0; 4261e3d03965SLiu Bo rb_erase_cached(&ref->ref_node, &head->ref_tree); 42620e0adbcfSJosef Bacik RB_CLEAR_NODE(&ref->ref_node); 42631d57ee94SWang Xiaoguang if (!list_empty(&ref->add_list)) 42641d57ee94SWang Xiaoguang list_del(&ref->add_list); 4265d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4266d7df2c79SJosef Bacik btrfs_put_delayed_ref(ref); 4267d7df2c79SJosef Bacik } 426854067ae9SJosef Bacik if (head->must_insert_reserved) 4269e78417d1SJosef Bacik pin_bytes = true; 427078a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 4271fa781ceaSJosef Bacik btrfs_delete_ref_head(delayed_refs, head); 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 { 4489bbbf7243SNikolay Borisov struct btrfs_device *dev, *tmp; 4490bbbf7243SNikolay Borisov 44912ff7e61eSJeff Mahoney btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 4492c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->dirty_bgs)); 4493c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->io_bgs)); 4494c79a1751SLiu Bo 4495bbbf7243SNikolay Borisov list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list, 4496bbbf7243SNikolay Borisov post_commit_list) { 4497bbbf7243SNikolay Borisov list_del_init(&dev->post_commit_list); 4498bbbf7243SNikolay Borisov } 4499bbbf7243SNikolay Borisov 45002ff7e61eSJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, fs_info); 450149b25e05SJeff Mahoney 45024a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 45030b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 450449b25e05SJeff Mahoney 45054a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 45060b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 450749b25e05SJeff Mahoney 4508ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 4509ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 451067cde344SMiao Xie 45112ff7e61eSJeff Mahoney btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, 451249b25e05SJeff Mahoney EXTENT_DIRTY); 45132ff7e61eSJeff Mahoney btrfs_destroy_pinned_extent(fs_info, 45140b246afaSJeff Mahoney fs_info->pinned_extents); 451549b25e05SJeff Mahoney 45164a9d8bdeSMiao Xie cur_trans->state =TRANS_STATE_COMPLETED; 45174a9d8bdeSMiao Xie wake_up(&cur_trans->commit_wait); 451849b25e05SJeff Mahoney } 451949b25e05SJeff Mahoney 45202ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) 4521acce952bSliubo { 4522acce952bSliubo struct btrfs_transaction *t; 4523acce952bSliubo 45240b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 4525acce952bSliubo 45260b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 45270b246afaSJeff Mahoney while (!list_empty(&fs_info->trans_list)) { 45280b246afaSJeff Mahoney t = list_first_entry(&fs_info->trans_list, 4529724e2315SJosef Bacik struct btrfs_transaction, list); 4530724e2315SJosef Bacik if (t->state >= TRANS_STATE_COMMIT_START) { 45319b64f57dSElena Reshetova refcount_inc(&t->use_count); 45320b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 45332ff7e61eSJeff Mahoney btrfs_wait_for_commit(fs_info, t->transid); 4534724e2315SJosef Bacik btrfs_put_transaction(t); 45350b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4536724e2315SJosef Bacik continue; 4537724e2315SJosef Bacik } 45380b246afaSJeff Mahoney if (t == fs_info->running_transaction) { 4539724e2315SJosef Bacik t->state = TRANS_STATE_COMMIT_DOING; 45400b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4541724e2315SJosef Bacik /* 4542724e2315SJosef Bacik * We wait for 0 num_writers since we don't hold a trans 4543724e2315SJosef Bacik * handle open currently for this transaction. 4544724e2315SJosef Bacik */ 4545724e2315SJosef Bacik wait_event(t->writer_wait, 4546724e2315SJosef Bacik atomic_read(&t->num_writers) == 0); 4547724e2315SJosef Bacik } else { 45480b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4549724e2315SJosef Bacik } 45502ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(t, fs_info); 4551724e2315SJosef Bacik 45520b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 45530b246afaSJeff Mahoney if (t == fs_info->running_transaction) 45540b246afaSJeff Mahoney fs_info->running_transaction = NULL; 4555724e2315SJosef Bacik list_del_init(&t->list); 45560b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4557a4abeea4SJosef Bacik 4558724e2315SJosef Bacik btrfs_put_transaction(t); 45592ff7e61eSJeff Mahoney trace_btrfs_transaction_commit(fs_info->tree_root); 45600b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4561724e2315SJosef Bacik } 45620b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 45630b246afaSJeff Mahoney btrfs_destroy_all_ordered_extents(fs_info); 4564ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 4565ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 45662ff7e61eSJeff Mahoney btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents); 45670b246afaSJeff Mahoney btrfs_destroy_all_delalloc_inodes(fs_info); 45680b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 4569acce952bSliubo 4570acce952bSliubo return 0; 4571acce952bSliubo } 4572acce952bSliubo 4573e8c9f186SDavid Sterba static const struct extent_io_ops btree_extent_io_ops = { 45744d53dddbSDavid Sterba /* mandatory callbacks */ 45750b86a832SChris Mason .submit_bio_hook = btree_submit_bio_hook, 45764d53dddbSDavid Sterba .readpage_end_io_hook = btree_readpage_end_io_hook, 45770da5468fSChris Mason }; 4578