1c1d7c514SDavid Sterba // SPDX-License-Identifier: GPL-2.0 26cbd5570SChris Mason /* 36cbd5570SChris Mason * Copyright (C) 2007 Oracle. All rights reserved. 46cbd5570SChris Mason */ 56cbd5570SChris Mason 6e20d96d6SChris Mason #include <linux/fs.h> 7d98237b3SChris Mason #include <linux/blkdev.h> 80f7d52f4SChris Mason #include <linux/radix-tree.h> 935b7e476SChris Mason #include <linux/writeback.h> 10d397712bSChris Mason #include <linux/buffer_head.h> 11ce9adaa5SChris Mason #include <linux/workqueue.h> 12a74a4b97SChris Mason #include <linux/kthread.h> 135a0e3ad6STejun Heo #include <linux/slab.h> 14784b4e29SChris Mason #include <linux/migrate.h> 157a36ddecSDavid Sterba #include <linux/ratelimit.h> 166463fe58SStefan Behrens #include <linux/uuid.h> 17803b2f54SStefan Behrens #include <linux/semaphore.h> 18540adea3SMasami Hiramatsu #include <linux/error-injection.h> 199678c543SNikolay Borisov #include <linux/crc32c.h> 20b89f6d1fSFilipe Manana #include <linux/sched/mm.h> 217e75bf3fSDavid Sterba #include <asm/unaligned.h> 226d97c6e3SJohannes Thumshirn #include <crypto/hash.h> 23eb60ceacSChris Mason #include "ctree.h" 24eb60ceacSChris Mason #include "disk-io.h" 25e089f05cSChris Mason #include "transaction.h" 260f7d52f4SChris Mason #include "btrfs_inode.h" 270b86a832SChris Mason #include "volumes.h" 28db94535dSChris Mason #include "print-tree.h" 29925baeddSChris Mason #include "locking.h" 30e02119d5SChris Mason #include "tree-log.h" 31fa9c0d79SChris Mason #include "free-space-cache.h" 3270f6d82eSOmar Sandoval #include "free-space-tree.h" 33581bb050SLi Zefan #include "inode-map.h" 3421adbd5cSStefan Behrens #include "check-integrity.h" 35606686eeSJosef Bacik #include "rcu-string.h" 368dabb742SStefan Behrens #include "dev-replace.h" 3753b381b3SDavid Woodhouse #include "raid56.h" 385ac1d209SJeff Mahoney #include "sysfs.h" 39fcebe456SJosef Bacik #include "qgroup.h" 40ebb8765bSAnand Jain #include "compression.h" 41557ea5ddSQu Wenruo #include "tree-checker.h" 42fd708b81SJosef Bacik #include "ref-verify.h" 43eb60ceacSChris Mason 44319e4d06SQu Wenruo #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\ 45319e4d06SQu Wenruo BTRFS_HEADER_FLAG_RELOC |\ 46319e4d06SQu Wenruo BTRFS_SUPER_FLAG_ERROR |\ 47319e4d06SQu Wenruo BTRFS_SUPER_FLAG_SEEDING |\ 48e2731e55SAnand Jain BTRFS_SUPER_FLAG_METADUMP |\ 49e2731e55SAnand Jain BTRFS_SUPER_FLAG_METADUMP_V2) 50319e4d06SQu Wenruo 51e8c9f186SDavid Sterba static const struct extent_io_ops btree_extent_io_ops; 528b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work); 53143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root); 54acce952bSliubo static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 552ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info); 56143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root); 572ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 58acce952bSliubo struct extent_io_tree *dirty_pages, 59acce952bSliubo int mark); 602ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 61acce952bSliubo struct extent_io_tree *pinned_extents); 622ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info); 632ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info); 64ce9adaa5SChris Mason 65d352ac68SChris Mason /* 6697eb6b69SDavid Sterba * btrfs_end_io_wq structs are used to do processing in task context when an IO 6797eb6b69SDavid Sterba * is complete. This is used during reads to verify checksums, and it is used 68d352ac68SChris Mason * by writes to insert metadata for new file extents after IO is complete. 69d352ac68SChris Mason */ 7097eb6b69SDavid Sterba struct btrfs_end_io_wq { 71ce9adaa5SChris Mason struct bio *bio; 72ce9adaa5SChris Mason bio_end_io_t *end_io; 73ce9adaa5SChris Mason void *private; 74ce9adaa5SChris Mason struct btrfs_fs_info *info; 754e4cbee9SChristoph Hellwig blk_status_t status; 76bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata; 778b712842SChris Mason struct btrfs_work work; 78ce9adaa5SChris Mason }; 790da5468fSChris Mason 8097eb6b69SDavid Sterba static struct kmem_cache *btrfs_end_io_wq_cache; 8197eb6b69SDavid Sterba 8297eb6b69SDavid Sterba int __init btrfs_end_io_wq_init(void) 8397eb6b69SDavid Sterba { 8497eb6b69SDavid Sterba btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq", 8597eb6b69SDavid Sterba sizeof(struct btrfs_end_io_wq), 8697eb6b69SDavid Sterba 0, 87fba4b697SNikolay Borisov SLAB_MEM_SPREAD, 8897eb6b69SDavid Sterba NULL); 8997eb6b69SDavid Sterba if (!btrfs_end_io_wq_cache) 9097eb6b69SDavid Sterba return -ENOMEM; 9197eb6b69SDavid Sterba return 0; 9297eb6b69SDavid Sterba } 9397eb6b69SDavid Sterba 94e67c718bSDavid Sterba void __cold btrfs_end_io_wq_exit(void) 9597eb6b69SDavid Sterba { 9697eb6b69SDavid Sterba kmem_cache_destroy(btrfs_end_io_wq_cache); 9797eb6b69SDavid Sterba } 9897eb6b69SDavid Sterba 99d352ac68SChris Mason /* 100d352ac68SChris Mason * async submit bios are used to offload expensive checksumming 101d352ac68SChris Mason * onto the worker threads. They checksum file and metadata bios 102d352ac68SChris Mason * just before they are sent down the IO stack. 103d352ac68SChris Mason */ 10444b8bd7eSChris Mason struct async_submit_bio { 105c6100a4bSJosef Bacik void *private_data; 10644b8bd7eSChris Mason struct bio *bio; 107a758781dSDavid Sterba extent_submit_bio_start_t *submit_bio_start; 10844b8bd7eSChris Mason int mirror_num; 109eaf25d93SChris Mason /* 110eaf25d93SChris Mason * bio_offset is optional, can be used if the pages in the bio 111eaf25d93SChris Mason * can't tell us where in the file the bio should go 112eaf25d93SChris Mason */ 113eaf25d93SChris Mason u64 bio_offset; 1148b712842SChris Mason struct btrfs_work work; 1154e4cbee9SChristoph Hellwig blk_status_t status; 11644b8bd7eSChris Mason }; 11744b8bd7eSChris Mason 11885d4e461SChris Mason /* 11985d4e461SChris Mason * Lockdep class keys for extent_buffer->lock's in this root. For a given 12085d4e461SChris Mason * eb, the lockdep key is determined by the btrfs_root it belongs to and 12185d4e461SChris Mason * the level the eb occupies in the tree. 1224008c04aSChris Mason * 12385d4e461SChris Mason * Different roots are used for different purposes and may nest inside each 12485d4e461SChris Mason * other and they require separate keysets. As lockdep keys should be 12585d4e461SChris Mason * static, assign keysets according to the purpose of the root as indicated 1264fd786e6SMisono Tomohiro * by btrfs_root->root_key.objectid. This ensures that all special purpose 1274fd786e6SMisono Tomohiro * roots have separate keysets. 1284008c04aSChris Mason * 12985d4e461SChris Mason * Lock-nesting across peer nodes is always done with the immediate parent 13085d4e461SChris Mason * node locked thus preventing deadlock. As lockdep doesn't know this, use 13185d4e461SChris Mason * subclass to avoid triggering lockdep warning in such cases. 1324008c04aSChris Mason * 13385d4e461SChris Mason * The key is set by the readpage_end_io_hook after the buffer has passed 13485d4e461SChris Mason * csum validation but before the pages are unlocked. It is also set by 13585d4e461SChris Mason * btrfs_init_new_buffer on freshly allocated blocks. 13685d4e461SChris Mason * 13785d4e461SChris Mason * We also add a check to make sure the highest level of the tree is the 13885d4e461SChris Mason * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code 13985d4e461SChris Mason * needs update as well. 1404008c04aSChris Mason */ 1414008c04aSChris Mason #ifdef CONFIG_DEBUG_LOCK_ALLOC 1424008c04aSChris Mason # if BTRFS_MAX_LEVEL != 8 1434008c04aSChris Mason # error 1444008c04aSChris Mason # endif 14585d4e461SChris Mason 14685d4e461SChris Mason static struct btrfs_lockdep_keyset { 14785d4e461SChris Mason u64 id; /* root objectid */ 14885d4e461SChris Mason const char *name_stem; /* lock name stem */ 14985d4e461SChris Mason char names[BTRFS_MAX_LEVEL + 1][20]; 15085d4e461SChris Mason struct lock_class_key keys[BTRFS_MAX_LEVEL + 1]; 15185d4e461SChris Mason } btrfs_lockdep_keysets[] = { 15285d4e461SChris Mason { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" }, 15385d4e461SChris Mason { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" }, 15485d4e461SChris Mason { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" }, 15585d4e461SChris Mason { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" }, 15685d4e461SChris Mason { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" }, 15785d4e461SChris Mason { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" }, 15860b62978SDavid Sterba { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" }, 15985d4e461SChris Mason { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" }, 16085d4e461SChris Mason { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" }, 16185d4e461SChris Mason { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" }, 16213fd8da9SDavid Sterba { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" }, 1636b20e0adSDavid Sterba { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" }, 16485d4e461SChris Mason { .id = 0, .name_stem = "tree" }, 1654008c04aSChris Mason }; 16685d4e461SChris Mason 16785d4e461SChris Mason void __init btrfs_init_lockdep(void) 16885d4e461SChris Mason { 16985d4e461SChris Mason int i, j; 17085d4e461SChris Mason 17185d4e461SChris Mason /* initialize lockdep class names */ 17285d4e461SChris Mason for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) { 17385d4e461SChris Mason struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i]; 17485d4e461SChris Mason 17585d4e461SChris Mason for (j = 0; j < ARRAY_SIZE(ks->names); j++) 17685d4e461SChris Mason snprintf(ks->names[j], sizeof(ks->names[j]), 17785d4e461SChris Mason "btrfs-%s-%02d", ks->name_stem, j); 17885d4e461SChris Mason } 17985d4e461SChris Mason } 18085d4e461SChris Mason 18185d4e461SChris Mason void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb, 18285d4e461SChris Mason int level) 18385d4e461SChris Mason { 18485d4e461SChris Mason struct btrfs_lockdep_keyset *ks; 18585d4e461SChris Mason 18685d4e461SChris Mason BUG_ON(level >= ARRAY_SIZE(ks->keys)); 18785d4e461SChris Mason 18885d4e461SChris Mason /* find the matching keyset, id 0 is the default entry */ 18985d4e461SChris Mason for (ks = btrfs_lockdep_keysets; ks->id; ks++) 19085d4e461SChris Mason if (ks->id == objectid) 19185d4e461SChris Mason break; 19285d4e461SChris Mason 19385d4e461SChris Mason lockdep_set_class_and_name(&eb->lock, 19485d4e461SChris Mason &ks->keys[level], ks->names[level]); 19585d4e461SChris Mason } 19685d4e461SChris Mason 1974008c04aSChris Mason #endif 1984008c04aSChris Mason 199d352ac68SChris Mason /* 200d352ac68SChris Mason * extents on the btree inode are pretty simple, there's one extent 201d352ac68SChris Mason * that covers the entire device 202d352ac68SChris Mason */ 2036af49dbdSDavid Sterba struct extent_map *btree_get_extent(struct btrfs_inode *inode, 204306e16ceSDavid Sterba struct page *page, size_t pg_offset, u64 start, u64 len, 2055f39d397SChris Mason int create) 2065f39d397SChris Mason { 2073ffbd68cSDavid Sterba struct btrfs_fs_info *fs_info = inode->root->fs_info; 208fc4f21b1SNikolay Borisov struct extent_map_tree *em_tree = &inode->extent_tree; 2095f39d397SChris Mason struct extent_map *em; 2105f39d397SChris Mason int ret; 2115f39d397SChris Mason 212890871beSChris Mason read_lock(&em_tree->lock); 213d1310b2eSChris Mason em = lookup_extent_mapping(em_tree, start, len); 214a061fc8dSChris Mason if (em) { 2150b246afaSJeff Mahoney em->bdev = fs_info->fs_devices->latest_bdev; 216890871beSChris Mason read_unlock(&em_tree->lock); 2175f39d397SChris Mason goto out; 218a061fc8dSChris Mason } 219890871beSChris Mason read_unlock(&em_tree->lock); 2207b13b7b1SChris Mason 221172ddd60SDavid Sterba em = alloc_extent_map(); 2225f39d397SChris Mason if (!em) { 2235f39d397SChris Mason em = ERR_PTR(-ENOMEM); 2245f39d397SChris Mason goto out; 2255f39d397SChris Mason } 2265f39d397SChris Mason em->start = 0; 2270afbaf8cSChris Mason em->len = (u64)-1; 228c8b97818SChris Mason em->block_len = (u64)-1; 2295f39d397SChris Mason em->block_start = 0; 2300b246afaSJeff Mahoney em->bdev = fs_info->fs_devices->latest_bdev; 231d1310b2eSChris Mason 232890871beSChris Mason write_lock(&em_tree->lock); 23309a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 2345f39d397SChris Mason if (ret == -EEXIST) { 2355f39d397SChris Mason free_extent_map(em); 2367b13b7b1SChris Mason em = lookup_extent_mapping(em_tree, start, len); 237b4f359abSTsutomu Itoh if (!em) 2380433f20dSTsutomu Itoh em = ERR_PTR(-EIO); 2395f39d397SChris Mason } else if (ret) { 2407b13b7b1SChris Mason free_extent_map(em); 2410433f20dSTsutomu Itoh em = ERR_PTR(ret); 2425f39d397SChris Mason } 243890871beSChris Mason write_unlock(&em_tree->lock); 2447b13b7b1SChris Mason 2455f39d397SChris Mason out: 2465f39d397SChris Mason return em; 2475f39d397SChris Mason } 2485f39d397SChris Mason 2499ed57367SDavid Sterba u32 btrfs_csum_data(const char *data, u32 seed, size_t len) 25019c00ddcSChris Mason { 2519678c543SNikolay Borisov return crc32c(seed, data, len); 25219c00ddcSChris Mason } 25319c00ddcSChris Mason 2540b5e3dafSDomagoj Tršan void btrfs_csum_final(u32 crc, u8 *result) 25519c00ddcSChris Mason { 2567e75bf3fSDavid Sterba put_unaligned_le32(~crc, result); 25719c00ddcSChris Mason } 25819c00ddcSChris Mason 259d352ac68SChris Mason /* 2602996e1f8SJohannes Thumshirn * Compute the csum of a btree block and store the result to provided buffer. 2612996e1f8SJohannes Thumshirn * 2622996e1f8SJohannes Thumshirn * Returns error if the extent buffer cannot be mapped. 263d352ac68SChris Mason */ 2642996e1f8SJohannes Thumshirn static int csum_tree_block(struct extent_buffer *buf, u8 *result) 26519c00ddcSChris Mason { 26619c00ddcSChris Mason unsigned long len; 26719c00ddcSChris Mason unsigned long cur_len; 26819c00ddcSChris Mason unsigned long offset = BTRFS_CSUM_SIZE; 26919c00ddcSChris Mason char *kaddr; 27019c00ddcSChris Mason unsigned long map_start; 27119c00ddcSChris Mason unsigned long map_len; 27219c00ddcSChris Mason int err; 27319c00ddcSChris Mason u32 crc = ~(u32)0; 27419c00ddcSChris Mason 27519c00ddcSChris Mason len = buf->len - offset; 27619c00ddcSChris Mason while (len > 0) { 277d2e174d5SJohannes Thumshirn /* 278d2e174d5SJohannes Thumshirn * Note: we don't need to check for the err == 1 case here, as 279d2e174d5SJohannes Thumshirn * with the given combination of 'start = BTRFS_CSUM_SIZE (32)' 280d2e174d5SJohannes Thumshirn * and 'min_len = 32' and the currently implemented mapping 281d2e174d5SJohannes Thumshirn * algorithm we cannot cross a page boundary. 282d2e174d5SJohannes Thumshirn */ 28319c00ddcSChris Mason err = map_private_extent_buffer(buf, offset, 32, 284a6591715SChris Mason &kaddr, &map_start, &map_len); 285c53839fcSJohannes Thumshirn if (WARN_ON(err)) 2868bd98f0eSAlex Lyakas return err; 28719c00ddcSChris Mason cur_len = min(len, map_len - (offset - map_start)); 288b0496686SLiu Bo crc = btrfs_csum_data(kaddr + offset - map_start, 28919c00ddcSChris Mason crc, cur_len); 29019c00ddcSChris Mason len -= cur_len; 29119c00ddcSChris Mason offset += cur_len; 29219c00ddcSChris Mason } 29371a63551SDavid Sterba memset(result, 0, BTRFS_CSUM_SIZE); 294607d432dSJosef Bacik 29519c00ddcSChris Mason btrfs_csum_final(crc, result); 29619c00ddcSChris Mason 29719c00ddcSChris Mason return 0; 29819c00ddcSChris Mason } 29919c00ddcSChris Mason 300d352ac68SChris Mason /* 301d352ac68SChris Mason * we can't consider a given block up to date unless the transid of the 302d352ac68SChris Mason * block matches the transid in the parent node's pointer. This is how we 303d352ac68SChris Mason * detect blocks that either didn't get written at all or got written 304d352ac68SChris Mason * in the wrong place. 305d352ac68SChris Mason */ 3061259ab75SChris Mason static int verify_parent_transid(struct extent_io_tree *io_tree, 307b9fab919SChris Mason struct extent_buffer *eb, u64 parent_transid, 308b9fab919SChris Mason int atomic) 3091259ab75SChris Mason { 3102ac55d41SJosef Bacik struct extent_state *cached_state = NULL; 3111259ab75SChris Mason int ret; 3122755a0deSDavid Sterba bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB); 3131259ab75SChris Mason 3141259ab75SChris Mason if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 3151259ab75SChris Mason return 0; 3161259ab75SChris Mason 317b9fab919SChris Mason if (atomic) 318b9fab919SChris Mason return -EAGAIN; 319b9fab919SChris Mason 320a26e8c9fSJosef Bacik if (need_lock) { 321a26e8c9fSJosef Bacik btrfs_tree_read_lock(eb); 322300aa896SDavid Sterba btrfs_set_lock_blocking_read(eb); 323a26e8c9fSJosef Bacik } 324a26e8c9fSJosef Bacik 3252ac55d41SJosef Bacik lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 326ff13db41SDavid Sterba &cached_state); 3270b32f4bbSJosef Bacik if (extent_buffer_uptodate(eb) && 3281259ab75SChris Mason btrfs_header_generation(eb) == parent_transid) { 3291259ab75SChris Mason ret = 0; 3301259ab75SChris Mason goto out; 3311259ab75SChris Mason } 33294647322SDavid Sterba btrfs_err_rl(eb->fs_info, 33394647322SDavid Sterba "parent transid verify failed on %llu wanted %llu found %llu", 33494647322SDavid Sterba eb->start, 33529549aecSWang Shilong parent_transid, btrfs_header_generation(eb)); 3361259ab75SChris Mason ret = 1; 337a26e8c9fSJosef Bacik 338a26e8c9fSJosef Bacik /* 339a26e8c9fSJosef Bacik * Things reading via commit roots that don't have normal protection, 340a26e8c9fSJosef Bacik * like send, can have a really old block in cache that may point at a 34101327610SNicholas D Steeves * block that has been freed and re-allocated. So don't clear uptodate 342a26e8c9fSJosef Bacik * if we find an eb that is under IO (dirty/writeback) because we could 343a26e8c9fSJosef Bacik * end up reading in the stale data and then writing it back out and 344a26e8c9fSJosef Bacik * making everybody very sad. 345a26e8c9fSJosef Bacik */ 346a26e8c9fSJosef Bacik if (!extent_buffer_under_io(eb)) 3470b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 34833958dc6SChris Mason out: 3492ac55d41SJosef Bacik unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 350e43bbe5eSDavid Sterba &cached_state); 351472b909fSJosef Bacik if (need_lock) 352a26e8c9fSJosef Bacik btrfs_tree_read_unlock_blocking(eb); 3531259ab75SChris Mason return ret; 3541259ab75SChris Mason } 3551259ab75SChris Mason 356e7e16f48SJohannes Thumshirn static bool btrfs_supported_super_csum(u16 csum_type) 357e7e16f48SJohannes Thumshirn { 358e7e16f48SJohannes Thumshirn switch (csum_type) { 359e7e16f48SJohannes Thumshirn case BTRFS_CSUM_TYPE_CRC32: 360e7e16f48SJohannes Thumshirn return true; 361e7e16f48SJohannes Thumshirn default: 362e7e16f48SJohannes Thumshirn return false; 363e7e16f48SJohannes Thumshirn } 364e7e16f48SJohannes Thumshirn } 365e7e16f48SJohannes Thumshirn 366d352ac68SChris Mason /* 3671104a885SDavid Sterba * Return 0 if the superblock checksum type matches the checksum value of that 3681104a885SDavid Sterba * algorithm. Pass the raw disk superblock data. 3691104a885SDavid Sterba */ 370ab8d0fc4SJeff Mahoney static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info, 371ab8d0fc4SJeff Mahoney char *raw_disk_sb) 3721104a885SDavid Sterba { 3731104a885SDavid Sterba struct btrfs_super_block *disk_sb = 3741104a885SDavid Sterba (struct btrfs_super_block *)raw_disk_sb; 3751104a885SDavid Sterba u32 crc = ~(u32)0; 37651bce6c9SJohannes Thumshirn char result[BTRFS_CSUM_SIZE]; 3771104a885SDavid Sterba 3781104a885SDavid Sterba /* 3791104a885SDavid Sterba * The super_block structure does not span the whole 38051bce6c9SJohannes Thumshirn * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is 38151bce6c9SJohannes Thumshirn * filled with zeros and is included in the checksum. 3821104a885SDavid Sterba */ 3831104a885SDavid Sterba crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE, 3841104a885SDavid Sterba crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE); 3851104a885SDavid Sterba btrfs_csum_final(crc, result); 3861104a885SDavid Sterba 38751bce6c9SJohannes Thumshirn if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb))) 388e7e16f48SJohannes Thumshirn return 1; 3891104a885SDavid Sterba 390e7e16f48SJohannes Thumshirn return 0; 3911104a885SDavid Sterba } 3921104a885SDavid Sterba 393e064d5e9SDavid Sterba int btrfs_verify_level_key(struct extent_buffer *eb, int level, 394ff76a864SLiu Bo struct btrfs_key *first_key, u64 parent_transid) 395581c1760SQu Wenruo { 396e064d5e9SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 397581c1760SQu Wenruo int found_level; 398581c1760SQu Wenruo struct btrfs_key found_key; 399581c1760SQu Wenruo int ret; 400581c1760SQu Wenruo 401581c1760SQu Wenruo found_level = btrfs_header_level(eb); 402581c1760SQu Wenruo if (found_level != level) { 40363489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 40463489055SQu Wenruo KERN_ERR "BTRFS: tree level check failed\n"); 405581c1760SQu Wenruo btrfs_err(fs_info, 406581c1760SQu Wenruo "tree level mismatch detected, bytenr=%llu level expected=%u has=%u", 407581c1760SQu Wenruo eb->start, level, found_level); 408581c1760SQu Wenruo return -EIO; 409581c1760SQu Wenruo } 410581c1760SQu Wenruo 411581c1760SQu Wenruo if (!first_key) 412581c1760SQu Wenruo return 0; 413581c1760SQu Wenruo 4145d41be6fSQu Wenruo /* 4155d41be6fSQu Wenruo * For live tree block (new tree blocks in current transaction), 4165d41be6fSQu Wenruo * we need proper lock context to avoid race, which is impossible here. 4175d41be6fSQu Wenruo * So we only checks tree blocks which is read from disk, whose 4185d41be6fSQu Wenruo * generation <= fs_info->last_trans_committed. 4195d41be6fSQu Wenruo */ 4205d41be6fSQu Wenruo if (btrfs_header_generation(eb) > fs_info->last_trans_committed) 4215d41be6fSQu Wenruo return 0; 422581c1760SQu Wenruo if (found_level) 423581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &found_key, 0); 424581c1760SQu Wenruo else 425581c1760SQu Wenruo btrfs_item_key_to_cpu(eb, &found_key, 0); 426581c1760SQu Wenruo ret = btrfs_comp_cpu_keys(first_key, &found_key); 427581c1760SQu Wenruo 428581c1760SQu Wenruo if (ret) { 42963489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 43063489055SQu Wenruo KERN_ERR "BTRFS: tree first key check failed\n"); 431581c1760SQu Wenruo btrfs_err(fs_info, 432ff76a864SLiu Bo "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", 433ff76a864SLiu Bo eb->start, parent_transid, first_key->objectid, 434ff76a864SLiu Bo first_key->type, first_key->offset, 435ff76a864SLiu Bo found_key.objectid, found_key.type, 436ff76a864SLiu Bo found_key.offset); 437581c1760SQu Wenruo } 438581c1760SQu Wenruo return ret; 439581c1760SQu Wenruo } 440581c1760SQu Wenruo 4411104a885SDavid Sterba /* 442d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 443d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 444581c1760SQu Wenruo * 445581c1760SQu Wenruo * @parent_transid: expected transid, skip check if 0 446581c1760SQu Wenruo * @level: expected level, mandatory check 447581c1760SQu Wenruo * @first_key: expected key of first slot, skip check if NULL 448d352ac68SChris Mason */ 4495ab12d1fSDavid Sterba static int btree_read_extent_buffer_pages(struct extent_buffer *eb, 450581c1760SQu Wenruo u64 parent_transid, int level, 451581c1760SQu Wenruo struct btrfs_key *first_key) 452f188591eSChris Mason { 4535ab12d1fSDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 454f188591eSChris Mason struct extent_io_tree *io_tree; 455ea466794SJosef Bacik int failed = 0; 456f188591eSChris Mason int ret; 457f188591eSChris Mason int num_copies = 0; 458f188591eSChris Mason int mirror_num = 0; 459ea466794SJosef Bacik int failed_mirror = 0; 460f188591eSChris Mason 4610b246afaSJeff Mahoney io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; 462f188591eSChris Mason while (1) { 463f8397d69SNikolay Borisov clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 464c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num); 465256dd1bbSStefan Behrens if (!ret) { 466581c1760SQu Wenruo if (verify_parent_transid(io_tree, eb, 467b9fab919SChris Mason parent_transid, 0)) 468256dd1bbSStefan Behrens ret = -EIO; 469e064d5e9SDavid Sterba else if (btrfs_verify_level_key(eb, level, 470ff76a864SLiu Bo first_key, parent_transid)) 471581c1760SQu Wenruo ret = -EUCLEAN; 472581c1760SQu Wenruo else 473581c1760SQu Wenruo break; 474256dd1bbSStefan Behrens } 475d397712bSChris Mason 4760b246afaSJeff Mahoney num_copies = btrfs_num_copies(fs_info, 477f188591eSChris Mason eb->start, eb->len); 4784235298eSChris Mason if (num_copies == 1) 479ea466794SJosef Bacik break; 4804235298eSChris Mason 4815cf1ab56SJosef Bacik if (!failed_mirror) { 4825cf1ab56SJosef Bacik failed = 1; 4835cf1ab56SJosef Bacik failed_mirror = eb->read_mirror; 4845cf1ab56SJosef Bacik } 4855cf1ab56SJosef Bacik 486f188591eSChris Mason mirror_num++; 487ea466794SJosef Bacik if (mirror_num == failed_mirror) 488ea466794SJosef Bacik mirror_num++; 489ea466794SJosef Bacik 4904235298eSChris Mason if (mirror_num > num_copies) 491ea466794SJosef Bacik break; 492f188591eSChris Mason } 493ea466794SJosef Bacik 494c0901581SStefan Behrens if (failed && !ret && failed_mirror) 49520a1fbf9SDavid Sterba btrfs_repair_eb_io_failure(eb, failed_mirror); 496ea466794SJosef Bacik 497ea466794SJosef Bacik return ret; 498f188591eSChris Mason } 49919c00ddcSChris Mason 500d352ac68SChris Mason /* 501d397712bSChris Mason * checksum a dirty tree block before IO. This has extra checks to make sure 502d397712bSChris Mason * we only fill in the checksum field in the first page of a multi-page block 503d352ac68SChris Mason */ 504d397712bSChris Mason 50501d58472SDaniel Dressler static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page) 50619c00ddcSChris Mason { 5074eee4fa4SMiao Xie u64 start = page_offset(page); 50819c00ddcSChris Mason u64 found_start; 5092996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 5102996e1f8SJohannes Thumshirn u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); 51119c00ddcSChris Mason struct extent_buffer *eb; 5128d47a0d8SQu Wenruo int ret; 513f188591eSChris Mason 5144f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 5154f2de97aSJosef Bacik if (page != eb->pages[0]) 5164f2de97aSJosef Bacik return 0; 5170f805531SAlex Lyakas 51819c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 5190f805531SAlex Lyakas /* 5200f805531SAlex Lyakas * Please do not consolidate these warnings into a single if. 5210f805531SAlex Lyakas * It is useful to know what went wrong. 5220f805531SAlex Lyakas */ 5230f805531SAlex Lyakas if (WARN_ON(found_start != start)) 5240f805531SAlex Lyakas return -EUCLEAN; 5250f805531SAlex Lyakas if (WARN_ON(!PageUptodate(page))) 5260f805531SAlex Lyakas return -EUCLEAN; 5270f805531SAlex Lyakas 528de37aa51SNikolay Borisov ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid, 5290f805531SAlex Lyakas btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0); 5300f805531SAlex Lyakas 5312996e1f8SJohannes Thumshirn if (csum_tree_block(eb, result)) 5322996e1f8SJohannes Thumshirn return -EINVAL; 5332996e1f8SJohannes Thumshirn 5348d47a0d8SQu Wenruo if (btrfs_header_level(eb)) 5358d47a0d8SQu Wenruo ret = btrfs_check_node(eb); 5368d47a0d8SQu Wenruo else 5378d47a0d8SQu Wenruo ret = btrfs_check_leaf_full(eb); 5388d47a0d8SQu Wenruo 5398d47a0d8SQu Wenruo if (ret < 0) { 5408d47a0d8SQu Wenruo btrfs_err(fs_info, 5418d47a0d8SQu Wenruo "block=%llu write time tree block corruption detected", 5428d47a0d8SQu Wenruo eb->start); 5438d47a0d8SQu Wenruo return ret; 5448d47a0d8SQu Wenruo } 5452996e1f8SJohannes Thumshirn write_extent_buffer(eb, result, 0, csum_size); 5468d47a0d8SQu Wenruo 5472996e1f8SJohannes Thumshirn return 0; 54819c00ddcSChris Mason } 54919c00ddcSChris Mason 550b0c9b3b0SDavid Sterba static int check_tree_block_fsid(struct extent_buffer *eb) 5512b82032cSYan Zheng { 552b0c9b3b0SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 55301d58472SDaniel Dressler struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 55444880fdcSAnand Jain u8 fsid[BTRFS_FSID_SIZE]; 5552b82032cSYan Zheng int ret = 1; 5562b82032cSYan Zheng 5570a4e5586SRoss Kirk read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE); 5582b82032cSYan Zheng while (fs_devices) { 5597239ff4bSNikolay Borisov u8 *metadata_uuid; 5607239ff4bSNikolay Borisov 5617239ff4bSNikolay Borisov /* 5627239ff4bSNikolay Borisov * Checking the incompat flag is only valid for the current 5637239ff4bSNikolay Borisov * fs. For seed devices it's forbidden to have their uuid 5647239ff4bSNikolay Borisov * changed so reading ->fsid in this case is fine 5657239ff4bSNikolay Borisov */ 5667239ff4bSNikolay Borisov if (fs_devices == fs_info->fs_devices && 5677239ff4bSNikolay Borisov btrfs_fs_incompat(fs_info, METADATA_UUID)) 5687239ff4bSNikolay Borisov metadata_uuid = fs_devices->metadata_uuid; 5697239ff4bSNikolay Borisov else 5707239ff4bSNikolay Borisov metadata_uuid = fs_devices->fsid; 5717239ff4bSNikolay Borisov 5727239ff4bSNikolay Borisov if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) { 5732b82032cSYan Zheng ret = 0; 5742b82032cSYan Zheng break; 5752b82032cSYan Zheng } 5762b82032cSYan Zheng fs_devices = fs_devices->seed; 5772b82032cSYan Zheng } 5782b82032cSYan Zheng return ret; 5792b82032cSYan Zheng } 5802b82032cSYan Zheng 581facc8a22SMiao Xie static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio, 582facc8a22SMiao Xie u64 phy_offset, struct page *page, 583facc8a22SMiao Xie u64 start, u64 end, int mirror) 584ce9adaa5SChris Mason { 585ce9adaa5SChris Mason u64 found_start; 586ce9adaa5SChris Mason int found_level; 587ce9adaa5SChris Mason struct extent_buffer *eb; 588ce9adaa5SChris Mason struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 58902873e43SZhao Lei struct btrfs_fs_info *fs_info = root->fs_info; 5902996e1f8SJohannes Thumshirn u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); 591f188591eSChris Mason int ret = 0; 5922996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 593727011e0SChris Mason int reads_done; 594ce9adaa5SChris Mason 595ce9adaa5SChris Mason if (!page->private) 596ce9adaa5SChris Mason goto out; 597d397712bSChris Mason 5984f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 599d397712bSChris Mason 6000b32f4bbSJosef Bacik /* the pending IO might have been the only thing that kept this buffer 6010b32f4bbSJosef Bacik * in memory. Make sure we have a ref for all this other checks 6020b32f4bbSJosef Bacik */ 6030b32f4bbSJosef Bacik extent_buffer_get(eb); 6040b32f4bbSJosef Bacik 6050b32f4bbSJosef Bacik reads_done = atomic_dec_and_test(&eb->io_pages); 606727011e0SChris Mason if (!reads_done) 607727011e0SChris Mason goto err; 608f188591eSChris Mason 6095cf1ab56SJosef Bacik eb->read_mirror = mirror; 610656f30dbSFilipe Manana if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 611ea466794SJosef Bacik ret = -EIO; 612ea466794SJosef Bacik goto err; 613ea466794SJosef Bacik } 614ea466794SJosef Bacik 615ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 616727011e0SChris Mason if (found_start != eb->start) { 617893bf4b1SSu Yue btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu", 618893bf4b1SSu Yue eb->start, found_start); 619f188591eSChris Mason ret = -EIO; 620ce9adaa5SChris Mason goto err; 621ce9adaa5SChris Mason } 622b0c9b3b0SDavid Sterba if (check_tree_block_fsid(eb)) { 62302873e43SZhao Lei btrfs_err_rl(fs_info, "bad fsid on block %llu", 62494647322SDavid Sterba eb->start); 6251259ab75SChris Mason ret = -EIO; 6261259ab75SChris Mason goto err; 6271259ab75SChris Mason } 628ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 6291c24c3ceSJosef Bacik if (found_level >= BTRFS_MAX_LEVEL) { 630893bf4b1SSu Yue btrfs_err(fs_info, "bad tree block level %d on %llu", 631893bf4b1SSu Yue (int)btrfs_header_level(eb), eb->start); 6321c24c3ceSJosef Bacik ret = -EIO; 6331c24c3ceSJosef Bacik goto err; 6341c24c3ceSJosef Bacik } 635ce9adaa5SChris Mason 63685d4e461SChris Mason btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), 63785d4e461SChris Mason eb, found_level); 6384008c04aSChris Mason 6392996e1f8SJohannes Thumshirn ret = csum_tree_block(eb, result); 6408bd98f0eSAlex Lyakas if (ret) 641a826d6dcSJosef Bacik goto err; 642a826d6dcSJosef Bacik 6432996e1f8SJohannes Thumshirn if (memcmp_extent_buffer(eb, result, 0, csum_size)) { 6442996e1f8SJohannes Thumshirn u32 val; 6452996e1f8SJohannes Thumshirn u32 found = 0; 6462996e1f8SJohannes Thumshirn 6472996e1f8SJohannes Thumshirn memcpy(&found, result, csum_size); 6482996e1f8SJohannes Thumshirn 6492996e1f8SJohannes Thumshirn read_extent_buffer(eb, &val, 0, csum_size); 6502996e1f8SJohannes Thumshirn btrfs_warn_rl(fs_info, 6512996e1f8SJohannes Thumshirn "%s checksum verify failed on %llu wanted %x found %x level %d", 6522996e1f8SJohannes Thumshirn fs_info->sb->s_id, eb->start, 6532996e1f8SJohannes Thumshirn val, found, btrfs_header_level(eb)); 6542996e1f8SJohannes Thumshirn ret = -EUCLEAN; 6552996e1f8SJohannes Thumshirn goto err; 6562996e1f8SJohannes Thumshirn } 6572996e1f8SJohannes Thumshirn 658a826d6dcSJosef Bacik /* 659a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 660a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 661a826d6dcSJosef Bacik * return -EIO. 662a826d6dcSJosef Bacik */ 6631c4360eeSDavid Sterba if (found_level == 0 && btrfs_check_leaf_full(eb)) { 664a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 665a826d6dcSJosef Bacik ret = -EIO; 666a826d6dcSJosef Bacik } 667ce9adaa5SChris Mason 668813fd1dcSDavid Sterba if (found_level > 0 && btrfs_check_node(eb)) 669053ab70fSLiu Bo ret = -EIO; 670053ab70fSLiu Bo 6710b32f4bbSJosef Bacik if (!ret) 6720b32f4bbSJosef Bacik set_extent_buffer_uptodate(eb); 67375391f0dSQu Wenruo else 67475391f0dSQu Wenruo btrfs_err(fs_info, 67575391f0dSQu Wenruo "block=%llu read time tree block corruption detected", 67675391f0dSQu Wenruo eb->start); 677ce9adaa5SChris Mason err: 67879fb65a1SJosef Bacik if (reads_done && 67979fb65a1SJosef Bacik test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 680d48d71aaSDavid Sterba btree_readahead_hook(eb, ret); 6814bb31e92SArne Jansen 68253b381b3SDavid Woodhouse if (ret) { 68353b381b3SDavid Woodhouse /* 68453b381b3SDavid Woodhouse * our io error hook is going to dec the io pages 68553b381b3SDavid Woodhouse * again, we have to make sure it has something 68653b381b3SDavid Woodhouse * to decrement 68753b381b3SDavid Woodhouse */ 68853b381b3SDavid Woodhouse atomic_inc(&eb->io_pages); 6890b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 69053b381b3SDavid Woodhouse } 6910b32f4bbSJosef Bacik free_extent_buffer(eb); 692ce9adaa5SChris Mason out: 693f188591eSChris Mason return ret; 694ce9adaa5SChris Mason } 695ce9adaa5SChris Mason 6964246a0b6SChristoph Hellwig static void end_workqueue_bio(struct bio *bio) 697ce9adaa5SChris Mason { 69897eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq = bio->bi_private; 699ce9adaa5SChris Mason struct btrfs_fs_info *fs_info; 7009e0af237SLiu Bo struct btrfs_workqueue *wq; 7019e0af237SLiu Bo btrfs_work_func_t func; 702ce9adaa5SChris Mason 703ce9adaa5SChris Mason fs_info = end_io_wq->info; 7044e4cbee9SChristoph Hellwig end_io_wq->status = bio->bi_status; 705d20f7043SChris Mason 70637226b21SMike Christie if (bio_op(bio) == REQ_OP_WRITE) { 7079e0af237SLiu Bo if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) { 7089e0af237SLiu Bo wq = fs_info->endio_meta_write_workers; 7099e0af237SLiu Bo func = btrfs_endio_meta_write_helper; 7109e0af237SLiu Bo } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) { 7119e0af237SLiu Bo wq = fs_info->endio_freespace_worker; 7129e0af237SLiu Bo func = btrfs_freespace_write_helper; 7139e0af237SLiu Bo } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) { 7149e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 7159e0af237SLiu Bo func = btrfs_endio_raid56_helper; 716d20f7043SChris Mason } else { 7179e0af237SLiu Bo wq = fs_info->endio_write_workers; 7189e0af237SLiu Bo func = btrfs_endio_write_helper; 719d20f7043SChris Mason } 7209e0af237SLiu Bo } else { 7218b110e39SMiao Xie if (unlikely(end_io_wq->metadata == 7228b110e39SMiao Xie BTRFS_WQ_ENDIO_DIO_REPAIR)) { 7238b110e39SMiao Xie wq = fs_info->endio_repair_workers; 7248b110e39SMiao Xie func = btrfs_endio_repair_helper; 7258b110e39SMiao Xie } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) { 7269e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 7279e0af237SLiu Bo func = btrfs_endio_raid56_helper; 7289e0af237SLiu Bo } else if (end_io_wq->metadata) { 7299e0af237SLiu Bo wq = fs_info->endio_meta_workers; 7309e0af237SLiu Bo func = btrfs_endio_meta_helper; 7319e0af237SLiu Bo } else { 7329e0af237SLiu Bo wq = fs_info->endio_workers; 7339e0af237SLiu Bo func = btrfs_endio_helper; 7349e0af237SLiu Bo } 7359e0af237SLiu Bo } 7369e0af237SLiu Bo 7379e0af237SLiu Bo btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL); 7389e0af237SLiu Bo btrfs_queue_work(wq, &end_io_wq->work); 739ce9adaa5SChris Mason } 740ce9adaa5SChris Mason 7414e4cbee9SChristoph Hellwig blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 742bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata) 7430b86a832SChris Mason { 74497eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 7458b110e39SMiao Xie 74697eb6b69SDavid Sterba end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS); 747ce9adaa5SChris Mason if (!end_io_wq) 7484e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 749ce9adaa5SChris Mason 750ce9adaa5SChris Mason end_io_wq->private = bio->bi_private; 751ce9adaa5SChris Mason end_io_wq->end_io = bio->bi_end_io; 75222c59948SChris Mason end_io_wq->info = info; 7534e4cbee9SChristoph Hellwig end_io_wq->status = 0; 754ce9adaa5SChris Mason end_io_wq->bio = bio; 75522c59948SChris Mason end_io_wq->metadata = metadata; 756ce9adaa5SChris Mason 757ce9adaa5SChris Mason bio->bi_private = end_io_wq; 758ce9adaa5SChris Mason bio->bi_end_io = end_workqueue_bio; 75922c59948SChris Mason return 0; 76022c59948SChris Mason } 76122c59948SChris Mason 7624a69a410SChris Mason static void run_one_async_start(struct btrfs_work *work) 7634a69a410SChris Mason { 7644a69a410SChris Mason struct async_submit_bio *async; 7654e4cbee9SChristoph Hellwig blk_status_t ret; 7664a69a410SChris Mason 7674a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 768c6100a4bSJosef Bacik ret = async->submit_bio_start(async->private_data, async->bio, 769eaf25d93SChris Mason async->bio_offset); 77079787eaaSJeff Mahoney if (ret) 7714e4cbee9SChristoph Hellwig async->status = ret; 7724a69a410SChris Mason } 7734a69a410SChris Mason 77406ea01b1SDavid Sterba /* 77506ea01b1SDavid Sterba * In order to insert checksums into the metadata in large chunks, we wait 77606ea01b1SDavid Sterba * until bio submission time. All the pages in the bio are checksummed and 77706ea01b1SDavid Sterba * sums are attached onto the ordered extent record. 77806ea01b1SDavid Sterba * 77906ea01b1SDavid Sterba * At IO completion time the csums attached on the ordered extent record are 78006ea01b1SDavid Sterba * inserted into the tree. 78106ea01b1SDavid Sterba */ 7824a69a410SChris Mason static void run_one_async_done(struct btrfs_work *work) 7838b712842SChris Mason { 7848b712842SChris Mason struct async_submit_bio *async; 78506ea01b1SDavid Sterba struct inode *inode; 78606ea01b1SDavid Sterba blk_status_t ret; 7878b712842SChris Mason 7888b712842SChris Mason async = container_of(work, struct async_submit_bio, work); 78906ea01b1SDavid Sterba inode = async->private_data; 7904854ddd0SChris Mason 791bb7ab3b9SAdam Buchbinder /* If an error occurred we just want to clean up the bio and move on */ 7924e4cbee9SChristoph Hellwig if (async->status) { 7934e4cbee9SChristoph Hellwig async->bio->bi_status = async->status; 7944246a0b6SChristoph Hellwig bio_endio(async->bio); 79579787eaaSJeff Mahoney return; 79679787eaaSJeff Mahoney } 79779787eaaSJeff Mahoney 79806ea01b1SDavid Sterba ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, 79906ea01b1SDavid Sterba async->mirror_num, 1); 80006ea01b1SDavid Sterba if (ret) { 80106ea01b1SDavid Sterba async->bio->bi_status = ret; 80206ea01b1SDavid Sterba bio_endio(async->bio); 80306ea01b1SDavid Sterba } 8044a69a410SChris Mason } 8054a69a410SChris Mason 8064a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 8074a69a410SChris Mason { 8084a69a410SChris Mason struct async_submit_bio *async; 8094a69a410SChris Mason 8104a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 8118b712842SChris Mason kfree(async); 8128b712842SChris Mason } 8138b712842SChris Mason 8148c27cb35SLinus Torvalds blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio, 815c6100a4bSJosef Bacik int mirror_num, unsigned long bio_flags, 816c6100a4bSJosef Bacik u64 bio_offset, void *private_data, 817e288c080SDavid Sterba extent_submit_bio_start_t *submit_bio_start) 81844b8bd7eSChris Mason { 81944b8bd7eSChris Mason struct async_submit_bio *async; 82044b8bd7eSChris Mason 82144b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 82244b8bd7eSChris Mason if (!async) 8234e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 82444b8bd7eSChris Mason 825c6100a4bSJosef Bacik async->private_data = private_data; 82644b8bd7eSChris Mason async->bio = bio; 82744b8bd7eSChris Mason async->mirror_num = mirror_num; 8284a69a410SChris Mason async->submit_bio_start = submit_bio_start; 8294a69a410SChris Mason 8309e0af237SLiu Bo btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start, 8315cdc7ad3SQu Wenruo run_one_async_done, run_one_async_free); 8324a69a410SChris Mason 833eaf25d93SChris Mason async->bio_offset = bio_offset; 8348c8bee1dSChris Mason 8354e4cbee9SChristoph Hellwig async->status = 0; 83679787eaaSJeff Mahoney 83767f055c7SChristoph Hellwig if (op_is_sync(bio->bi_opf)) 8385cdc7ad3SQu Wenruo btrfs_set_work_high_priority(&async->work); 839d313d7a3SChris Mason 8405cdc7ad3SQu Wenruo btrfs_queue_work(fs_info->workers, &async->work); 84144b8bd7eSChris Mason return 0; 84244b8bd7eSChris Mason } 84344b8bd7eSChris Mason 8444e4cbee9SChristoph Hellwig static blk_status_t btree_csum_one_bio(struct bio *bio) 845ce3ed71aSChris Mason { 8462c30c71bSKent Overstreet struct bio_vec *bvec; 847ce3ed71aSChris Mason struct btrfs_root *root; 8482b070cfeSChristoph Hellwig int ret = 0; 8496dc4f100SMing Lei struct bvec_iter_all iter_all; 850ce3ed71aSChris Mason 851c09abff8SDavid Sterba ASSERT(!bio_flagged(bio, BIO_CLONED)); 8522b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 853ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 85401d58472SDaniel Dressler ret = csum_dirty_buffer(root->fs_info, bvec->bv_page); 85579787eaaSJeff Mahoney if (ret) 85679787eaaSJeff Mahoney break; 857ce3ed71aSChris Mason } 8582c30c71bSKent Overstreet 8594e4cbee9SChristoph Hellwig return errno_to_blk_status(ret); 860ce3ed71aSChris Mason } 861ce3ed71aSChris Mason 862d0ee3934SDavid Sterba static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio, 863eaf25d93SChris Mason u64 bio_offset) 86422c59948SChris Mason { 8658b712842SChris Mason /* 8668b712842SChris Mason * when we're called for a write, we're already in the async 8675443be45SChris Mason * submission context. Just jump into btrfs_map_bio 8688b712842SChris Mason */ 86979787eaaSJeff Mahoney return btree_csum_one_bio(bio); 87022c59948SChris Mason } 87122c59948SChris Mason 8729b4e675aSDavid Sterba static int check_async_write(struct btrfs_fs_info *fs_info, 8739b4e675aSDavid Sterba struct btrfs_inode *bi) 874de0022b9SJosef Bacik { 8756300463bSLiu Bo if (atomic_read(&bi->sync_writers)) 8766300463bSLiu Bo return 0; 8779b4e675aSDavid Sterba if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags)) 878de0022b9SJosef Bacik return 0; 879de0022b9SJosef Bacik return 1; 880de0022b9SJosef Bacik } 881de0022b9SJosef Bacik 882a56b1c7bSNikolay Borisov static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio, 88350489a57SNikolay Borisov int mirror_num, 88450489a57SNikolay Borisov unsigned long bio_flags) 88544b8bd7eSChris Mason { 8860b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 8879b4e675aSDavid Sterba int async = check_async_write(fs_info, BTRFS_I(inode)); 8884e4cbee9SChristoph Hellwig blk_status_t ret; 889cad321adSChris Mason 89037226b21SMike Christie if (bio_op(bio) != REQ_OP_WRITE) { 891cad321adSChris Mason /* 892cad321adSChris Mason * called for a read, do the setup so that checksum validation 893cad321adSChris Mason * can happen in the async kernel threads 894cad321adSChris Mason */ 8950b246afaSJeff Mahoney ret = btrfs_bio_wq_end_io(fs_info, bio, 8960b246afaSJeff Mahoney BTRFS_WQ_ENDIO_METADATA); 8971d4284bdSChris Mason if (ret) 89861891923SStefan Behrens goto out_w_error; 8992ff7e61eSJeff Mahoney ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); 900de0022b9SJosef Bacik } else if (!async) { 901de0022b9SJosef Bacik ret = btree_csum_one_bio(bio); 902de0022b9SJosef Bacik if (ret) 90361891923SStefan Behrens goto out_w_error; 9042ff7e61eSJeff Mahoney ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); 90561891923SStefan Behrens } else { 906cad321adSChris Mason /* 90761891923SStefan Behrens * kthread helpers are used to submit writes so that 90861891923SStefan Behrens * checksumming can happen in parallel across all CPUs 909cad321adSChris Mason */ 910c6100a4bSJosef Bacik ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0, 911e68f2ee7SNikolay Borisov 0, inode, btree_submit_bio_start); 91244b8bd7eSChris Mason } 91344b8bd7eSChris Mason 9144246a0b6SChristoph Hellwig if (ret) 9154246a0b6SChristoph Hellwig goto out_w_error; 9164246a0b6SChristoph Hellwig return 0; 9174246a0b6SChristoph Hellwig 91861891923SStefan Behrens out_w_error: 9194e4cbee9SChristoph Hellwig bio->bi_status = ret; 9204246a0b6SChristoph Hellwig bio_endio(bio); 92161891923SStefan Behrens return ret; 92261891923SStefan Behrens } 92361891923SStefan Behrens 9243dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 925784b4e29SChris Mason static int btree_migratepage(struct address_space *mapping, 926a6bc32b8SMel Gorman struct page *newpage, struct page *page, 927a6bc32b8SMel Gorman enum migrate_mode mode) 928784b4e29SChris Mason { 929784b4e29SChris Mason /* 930784b4e29SChris Mason * we can't safely write a btree page from here, 931784b4e29SChris Mason * we haven't done the locking hook 932784b4e29SChris Mason */ 933784b4e29SChris Mason if (PageDirty(page)) 934784b4e29SChris Mason return -EAGAIN; 935784b4e29SChris Mason /* 936784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 937784b4e29SChris Mason * We must have no buffers or drop them. 938784b4e29SChris Mason */ 939784b4e29SChris Mason if (page_has_private(page) && 940784b4e29SChris Mason !try_to_release_page(page, GFP_KERNEL)) 941784b4e29SChris Mason return -EAGAIN; 942a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 943784b4e29SChris Mason } 9443dd1462eSJan Beulich #endif 945784b4e29SChris Mason 9460da5468fSChris Mason 9470da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 9480da5468fSChris Mason struct writeback_control *wbc) 9490da5468fSChris Mason { 950e2d84521SMiao Xie struct btrfs_fs_info *fs_info; 951e2d84521SMiao Xie int ret; 952e2d84521SMiao Xie 953d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 954448d640bSChris Mason 955448d640bSChris Mason if (wbc->for_kupdate) 956448d640bSChris Mason return 0; 957448d640bSChris Mason 958e2d84521SMiao Xie fs_info = BTRFS_I(mapping->host)->root->fs_info; 959b9473439SChris Mason /* this is a bit racy, but that's ok */ 960d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 961d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 962d814a491SEthan Lien fs_info->dirty_metadata_batch); 963e2d84521SMiao Xie if (ret < 0) 964793955bcSChris Mason return 0; 965793955bcSChris Mason } 9660b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 9670da5468fSChris Mason } 9680da5468fSChris Mason 969b2950863SChristoph Hellwig static int btree_readpage(struct file *file, struct page *page) 9705f39d397SChris Mason { 971d1310b2eSChris Mason struct extent_io_tree *tree; 972d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9738ddc7d9cSJan Schmidt return extent_read_full_page(tree, page, btree_get_extent, 0); 9745f39d397SChris Mason } 9755f39d397SChris Mason 97670dec807SChris Mason static int btree_releasepage(struct page *page, gfp_t gfp_flags) 9775f39d397SChris Mason { 97898509cfcSChris Mason if (PageWriteback(page) || PageDirty(page)) 97998509cfcSChris Mason return 0; 9800c4e538bSDavid Sterba 981f7a52a40SDavid Sterba return try_release_extent_buffer(page); 982d98237b3SChris Mason } 983d98237b3SChris Mason 984d47992f8SLukas Czerner static void btree_invalidatepage(struct page *page, unsigned int offset, 985d47992f8SLukas Czerner unsigned int length) 986d98237b3SChris Mason { 987d1310b2eSChris Mason struct extent_io_tree *tree; 988d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9895f39d397SChris Mason extent_invalidatepage(tree, page, offset); 9905f39d397SChris Mason btree_releasepage(page, GFP_NOFS); 9919ad6b7bcSChris Mason if (PagePrivate(page)) { 992efe120a0SFrank Holton btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info, 993efe120a0SFrank Holton "page private not zero on page %llu", 994efe120a0SFrank Holton (unsigned long long)page_offset(page)); 9959ad6b7bcSChris Mason ClearPagePrivate(page); 9969ad6b7bcSChris Mason set_page_private(page, 0); 99709cbfeafSKirill A. Shutemov put_page(page); 9989ad6b7bcSChris Mason } 999d98237b3SChris Mason } 1000d98237b3SChris Mason 10010b32f4bbSJosef Bacik static int btree_set_page_dirty(struct page *page) 10020b32f4bbSJosef Bacik { 1003bb146eb2SJosef Bacik #ifdef DEBUG 10040b32f4bbSJosef Bacik struct extent_buffer *eb; 10050b32f4bbSJosef Bacik 10060b32f4bbSJosef Bacik BUG_ON(!PagePrivate(page)); 10070b32f4bbSJosef Bacik eb = (struct extent_buffer *)page->private; 10080b32f4bbSJosef Bacik BUG_ON(!eb); 10090b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 10100b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 10110b32f4bbSJosef Bacik btrfs_assert_tree_locked(eb); 1012bb146eb2SJosef Bacik #endif 10130b32f4bbSJosef Bacik return __set_page_dirty_nobuffers(page); 10140b32f4bbSJosef Bacik } 10150b32f4bbSJosef Bacik 10167f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 1017d98237b3SChris Mason .readpage = btree_readpage, 10180da5468fSChris Mason .writepages = btree_writepages, 10195f39d397SChris Mason .releasepage = btree_releasepage, 10205f39d397SChris Mason .invalidatepage = btree_invalidatepage, 10215a92bc88SChris Mason #ifdef CONFIG_MIGRATION 1022784b4e29SChris Mason .migratepage = btree_migratepage, 10235a92bc88SChris Mason #endif 10240b32f4bbSJosef Bacik .set_page_dirty = btree_set_page_dirty, 1025d98237b3SChris Mason }; 1026123abc88SChris Mason 10272ff7e61eSJeff Mahoney void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr) 1028090d1875SChris Mason { 10295f39d397SChris Mason struct extent_buffer *buf = NULL; 1030537f38f0SNikolay Borisov int ret; 1031090d1875SChris Mason 10322ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 1033c871b0f2SLiu Bo if (IS_ERR(buf)) 10346197d86eSDavid Sterba return; 1035537f38f0SNikolay Borisov 1036c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(buf, WAIT_NONE, 0); 1037537f38f0SNikolay Borisov if (ret < 0) 1038537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 1039537f38f0SNikolay Borisov else 10405f39d397SChris Mason free_extent_buffer(buf); 1041090d1875SChris Mason } 1042090d1875SChris Mason 10432ff7e61eSJeff Mahoney int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr, 1044ab0fff03SArne Jansen int mirror_num, struct extent_buffer **eb) 1045ab0fff03SArne Jansen { 1046ab0fff03SArne Jansen struct extent_buffer *buf = NULL; 1047ab0fff03SArne Jansen int ret; 1048ab0fff03SArne Jansen 10492ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 1050c871b0f2SLiu Bo if (IS_ERR(buf)) 1051ab0fff03SArne Jansen return 0; 1052ab0fff03SArne Jansen 1053ab0fff03SArne Jansen set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags); 1054ab0fff03SArne Jansen 1055c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(buf, WAIT_PAGE_LOCK, mirror_num); 1056ab0fff03SArne Jansen if (ret) { 1057537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 1058ab0fff03SArne Jansen return ret; 1059ab0fff03SArne Jansen } 1060ab0fff03SArne Jansen 1061ab0fff03SArne Jansen if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) { 1062537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 1063ab0fff03SArne Jansen return -EIO; 10640b32f4bbSJosef Bacik } else if (extent_buffer_uptodate(buf)) { 1065ab0fff03SArne Jansen *eb = buf; 1066ab0fff03SArne Jansen } else { 1067ab0fff03SArne Jansen free_extent_buffer(buf); 1068ab0fff03SArne Jansen } 1069ab0fff03SArne Jansen return 0; 1070ab0fff03SArne Jansen } 1071ab0fff03SArne Jansen 10722ff7e61eSJeff Mahoney struct extent_buffer *btrfs_find_create_tree_block( 10732ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info, 10740308af44SDavid Sterba u64 bytenr) 10750999df54SChris Mason { 10760b246afaSJeff Mahoney if (btrfs_is_testing(fs_info)) 10770b246afaSJeff Mahoney return alloc_test_extent_buffer(fs_info, bytenr); 10780b246afaSJeff Mahoney return alloc_extent_buffer(fs_info, bytenr); 10790999df54SChris Mason } 10800999df54SChris Mason 1081581c1760SQu Wenruo /* 1082581c1760SQu Wenruo * Read tree block at logical address @bytenr and do variant basic but critical 1083581c1760SQu Wenruo * verification. 1084581c1760SQu Wenruo * 1085581c1760SQu Wenruo * @parent_transid: expected transid of this tree block, skip check if 0 1086581c1760SQu Wenruo * @level: expected level, mandatory check 1087581c1760SQu Wenruo * @first_key: expected key in slot 0, skip check if NULL 1088581c1760SQu Wenruo */ 10892ff7e61eSJeff Mahoney struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, 1090581c1760SQu Wenruo u64 parent_transid, int level, 1091581c1760SQu Wenruo struct btrfs_key *first_key) 1092e20d96d6SChris Mason { 10935f39d397SChris Mason struct extent_buffer *buf = NULL; 109419c00ddcSChris Mason int ret; 109519c00ddcSChris Mason 10962ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 1097c871b0f2SLiu Bo if (IS_ERR(buf)) 1098c871b0f2SLiu Bo return buf; 1099e4204dedSChris Mason 11005ab12d1fSDavid Sterba ret = btree_read_extent_buffer_pages(buf, parent_transid, 1101581c1760SQu Wenruo level, first_key); 11020f0fe8f7SFilipe David Borba Manana if (ret) { 1103537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 110464c043deSLiu Bo return ERR_PTR(ret); 11050f0fe8f7SFilipe David Borba Manana } 11065f39d397SChris Mason return buf; 1107ce9adaa5SChris Mason 1108eb60ceacSChris Mason } 1109eb60ceacSChris Mason 11106a884d7dSDavid Sterba void btrfs_clean_tree_block(struct extent_buffer *buf) 1111ed2ff2cbSChris Mason { 11126a884d7dSDavid Sterba struct btrfs_fs_info *fs_info = buf->fs_info; 111355c69072SChris Mason if (btrfs_header_generation(buf) == 1114e2d84521SMiao Xie fs_info->running_transaction->transid) { 1115b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1116b4ce94deSChris Mason 1117b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1118104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 1119e2d84521SMiao Xie -buf->len, 1120e2d84521SMiao Xie fs_info->dirty_metadata_batch); 1121b9473439SChris Mason /* ugh, clear_extent_buffer_dirty needs to lock the page */ 11228bead258SDavid Sterba btrfs_set_lock_blocking_write(buf); 11230b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 1124925baeddSChris Mason } 11255f39d397SChris Mason } 1126ed7b63ebSJosef Bacik } 11275f39d397SChris Mason 11288257b2dcSMiao Xie static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void) 11298257b2dcSMiao Xie { 11308257b2dcSMiao Xie struct btrfs_subvolume_writers *writers; 11318257b2dcSMiao Xie int ret; 11328257b2dcSMiao Xie 11338257b2dcSMiao Xie writers = kmalloc(sizeof(*writers), GFP_NOFS); 11348257b2dcSMiao Xie if (!writers) 11358257b2dcSMiao Xie return ERR_PTR(-ENOMEM); 11368257b2dcSMiao Xie 11378a5a916dSJeff Mahoney ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS); 11388257b2dcSMiao Xie if (ret < 0) { 11398257b2dcSMiao Xie kfree(writers); 11408257b2dcSMiao Xie return ERR_PTR(ret); 11418257b2dcSMiao Xie } 11428257b2dcSMiao Xie 11438257b2dcSMiao Xie init_waitqueue_head(&writers->wait); 11448257b2dcSMiao Xie return writers; 11458257b2dcSMiao Xie } 11468257b2dcSMiao Xie 11478257b2dcSMiao Xie static void 11488257b2dcSMiao Xie btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers) 11498257b2dcSMiao Xie { 11508257b2dcSMiao Xie percpu_counter_destroy(&writers->counter); 11518257b2dcSMiao Xie kfree(writers); 11528257b2dcSMiao Xie } 11538257b2dcSMiao Xie 1154da17066cSJeff Mahoney static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, 1155e20d96d6SChris Mason u64 objectid) 1156d97e63b6SChris Mason { 11577c0260eeSJeff Mahoney bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); 1158cfaa7295SChris Mason root->node = NULL; 1159a28ec197SChris Mason root->commit_root = NULL; 116027cdeb70SMiao Xie root->state = 0; 1161d68fc57bSYan, Zheng root->orphan_cleanup_state = 0; 11620b86a832SChris Mason 11630f7d52f4SChris Mason root->last_trans = 0; 116413a8a7c8SYan, Zheng root->highest_objectid = 0; 1165eb73c1b7SMiao Xie root->nr_delalloc_inodes = 0; 1166199c2a9cSMiao Xie root->nr_ordered_extents = 0; 11676bef4d31SEric Paris root->inode_tree = RB_ROOT; 116816cdcec7SMiao Xie INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1169f0486c68SYan, Zheng root->block_rsv = NULL; 11700b86a832SChris Mason 11710b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 11725d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 1173eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_inodes); 1174eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_root); 1175199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_extents); 1176199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_root); 1177d2311e69SQu Wenruo INIT_LIST_HEAD(&root->reloc_dirty_list); 11782ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 11792ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 11805d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1181eb73c1b7SMiao Xie spin_lock_init(&root->delalloc_lock); 1182199c2a9cSMiao Xie spin_lock_init(&root->ordered_extent_lock); 1183f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 11842ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 11852ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 11868287475aSQu Wenruo spin_lock_init(&root->qgroup_meta_rsv_lock); 1187a2135011SChris Mason mutex_init(&root->objectid_mutex); 1188e02119d5SChris Mason mutex_init(&root->log_mutex); 118931f3d255SMiao Xie mutex_init(&root->ordered_extent_mutex); 1190573bfb72SMiao Xie mutex_init(&root->delalloc_mutex); 11917237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 11927237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 11937237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 11948b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[0]); 11958b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[1]); 11967237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 11977237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 11987237f183SYan Zheng atomic_set(&root->log_writers, 0); 11992ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 12000700cea7SElena Reshetova refcount_set(&root->refs, 1); 1201ea14b57fSDavid Sterba atomic_set(&root->will_be_snapshotted, 0); 12028ecebf4dSRobbie Ko atomic_set(&root->snapshot_force_cow, 0); 1203eede2bf3SOmar Sandoval atomic_set(&root->nr_swapfiles, 0); 12047237f183SYan Zheng root->log_transid = 0; 1205d1433debSMiao Xie root->log_transid_committed = -1; 1206257c62e1SChris Mason root->last_log_commit = 0; 12077c0260eeSJeff Mahoney if (!dummy) 120843eb5f29SQu Wenruo extent_io_tree_init(fs_info, &root->dirty_log_pages, 120943eb5f29SQu Wenruo IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL); 1210017e5369SChris Mason 12113768f368SChris Mason memset(&root->root_key, 0, sizeof(root->root_key)); 12123768f368SChris Mason memset(&root->root_item, 0, sizeof(root->root_item)); 12136702ed49SChris Mason memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 12147c0260eeSJeff Mahoney if (!dummy) 12153f157a2fSChris Mason root->defrag_trans_start = fs_info->generation; 121606ea65a3SJosef Bacik else 121706ea65a3SJosef Bacik root->defrag_trans_start = 0; 12184d775673SChris Mason root->root_key.objectid = objectid; 12190ee5dc67SAl Viro root->anon_dev = 0; 12208ea05e3aSAlexander Block 12215f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 1222370a11b8SQu Wenruo btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks); 12233768f368SChris Mason } 12243768f368SChris Mason 122574e4d827SDavid Sterba static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, 122674e4d827SDavid Sterba gfp_t flags) 12276f07e42eSAl Viro { 122874e4d827SDavid Sterba struct btrfs_root *root = kzalloc(sizeof(*root), flags); 12296f07e42eSAl Viro if (root) 12306f07e42eSAl Viro root->fs_info = fs_info; 12316f07e42eSAl Viro return root; 12326f07e42eSAl Viro } 12336f07e42eSAl Viro 123406ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 123506ea65a3SJosef Bacik /* Should only be used by the testing infrastructure */ 1236da17066cSJeff Mahoney struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) 123706ea65a3SJosef Bacik { 123806ea65a3SJosef Bacik struct btrfs_root *root; 123906ea65a3SJosef Bacik 12407c0260eeSJeff Mahoney if (!fs_info) 12417c0260eeSJeff Mahoney return ERR_PTR(-EINVAL); 12427c0260eeSJeff Mahoney 12437c0260eeSJeff Mahoney root = btrfs_alloc_root(fs_info, GFP_KERNEL); 124406ea65a3SJosef Bacik if (!root) 124506ea65a3SJosef Bacik return ERR_PTR(-ENOMEM); 1246da17066cSJeff Mahoney 1247b9ef22deSFeifei Xu /* We don't use the stripesize in selftest, set it as sectorsize */ 1248da17066cSJeff Mahoney __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID); 1249faa2dbf0SJosef Bacik root->alloc_bytenr = 0; 125006ea65a3SJosef Bacik 125106ea65a3SJosef Bacik return root; 125206ea65a3SJosef Bacik } 125306ea65a3SJosef Bacik #endif 125406ea65a3SJosef Bacik 125520897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 125620897f5cSArne Jansen u64 objectid) 125720897f5cSArne Jansen { 12589b7a2440SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 125920897f5cSArne Jansen struct extent_buffer *leaf; 126020897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 126120897f5cSArne Jansen struct btrfs_root *root; 126220897f5cSArne Jansen struct btrfs_key key; 1263b89f6d1fSFilipe Manana unsigned int nofs_flag; 126420897f5cSArne Jansen int ret = 0; 126533d85fdaSQu Wenruo uuid_le uuid = NULL_UUID_LE; 126620897f5cSArne Jansen 1267b89f6d1fSFilipe Manana /* 1268b89f6d1fSFilipe Manana * We're holding a transaction handle, so use a NOFS memory allocation 1269b89f6d1fSFilipe Manana * context to avoid deadlock if reclaim happens. 1270b89f6d1fSFilipe Manana */ 1271b89f6d1fSFilipe Manana nofs_flag = memalloc_nofs_save(); 127274e4d827SDavid Sterba root = btrfs_alloc_root(fs_info, GFP_KERNEL); 1273b89f6d1fSFilipe Manana memalloc_nofs_restore(nofs_flag); 127420897f5cSArne Jansen if (!root) 127520897f5cSArne Jansen return ERR_PTR(-ENOMEM); 127620897f5cSArne Jansen 1277da17066cSJeff Mahoney __setup_root(root, fs_info, objectid); 127820897f5cSArne Jansen root->root_key.objectid = objectid; 127920897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 128020897f5cSArne Jansen root->root_key.offset = 0; 128120897f5cSArne Jansen 12824d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0); 128320897f5cSArne Jansen if (IS_ERR(leaf)) { 128420897f5cSArne Jansen ret = PTR_ERR(leaf); 12851dd05682STsutomu Itoh leaf = NULL; 128620897f5cSArne Jansen goto fail; 128720897f5cSArne Jansen } 128820897f5cSArne Jansen 128920897f5cSArne Jansen root->node = leaf; 129020897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 129120897f5cSArne Jansen 129220897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 129327cdeb70SMiao Xie set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 129420897f5cSArne Jansen 129520897f5cSArne Jansen root->root_item.flags = 0; 129620897f5cSArne Jansen root->root_item.byte_limit = 0; 129720897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 129820897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 129920897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 130020897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 130120897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 130220897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 130320897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 130433d85fdaSQu Wenruo if (is_fstree(objectid)) 13056463fe58SStefan Behrens uuid_le_gen(&uuid); 13066463fe58SStefan Behrens memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE); 130720897f5cSArne Jansen root->root_item.drop_level = 0; 130820897f5cSArne Jansen 130920897f5cSArne Jansen key.objectid = objectid; 131020897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 131120897f5cSArne Jansen key.offset = 0; 131220897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 131320897f5cSArne Jansen if (ret) 131420897f5cSArne Jansen goto fail; 131520897f5cSArne Jansen 131620897f5cSArne Jansen btrfs_tree_unlock(leaf); 131720897f5cSArne Jansen 131820897f5cSArne Jansen return root; 13191dd05682STsutomu Itoh 13201dd05682STsutomu Itoh fail: 13211dd05682STsutomu Itoh if (leaf) { 13221dd05682STsutomu Itoh btrfs_tree_unlock(leaf); 132359885b39STsutomu Itoh free_extent_buffer(root->commit_root); 13241dd05682STsutomu Itoh free_extent_buffer(leaf); 13251dd05682STsutomu Itoh } 13261dd05682STsutomu Itoh kfree(root); 13271dd05682STsutomu Itoh 13281dd05682STsutomu Itoh return ERR_PTR(ret); 132920897f5cSArne Jansen } 133020897f5cSArne Jansen 13317237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1332e02119d5SChris Mason struct btrfs_fs_info *fs_info) 13330f7d52f4SChris Mason { 13340f7d52f4SChris Mason struct btrfs_root *root; 13357237f183SYan Zheng struct extent_buffer *leaf; 1336e02119d5SChris Mason 133774e4d827SDavid Sterba root = btrfs_alloc_root(fs_info, GFP_NOFS); 1338e02119d5SChris Mason if (!root) 13397237f183SYan Zheng return ERR_PTR(-ENOMEM); 1340e02119d5SChris Mason 1341da17066cSJeff Mahoney __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID); 1342e02119d5SChris Mason 1343e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1344e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1345e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 134627cdeb70SMiao Xie 13477237f183SYan Zheng /* 134827cdeb70SMiao Xie * DON'T set REF_COWS for log trees 134927cdeb70SMiao Xie * 13507237f183SYan Zheng * log trees do not get reference counted because they go away 13517237f183SYan Zheng * before a real commit is actually done. They do store pointers 13527237f183SYan Zheng * to file data extents, and those reference counts still get 13537237f183SYan Zheng * updated (along with back refs to the log tree). 13547237f183SYan Zheng */ 1355e02119d5SChris Mason 13564d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 13574d75f8a9SDavid Sterba NULL, 0, 0, 0); 13587237f183SYan Zheng if (IS_ERR(leaf)) { 13597237f183SYan Zheng kfree(root); 13607237f183SYan Zheng return ERR_CAST(leaf); 13617237f183SYan Zheng } 1362e02119d5SChris Mason 13637237f183SYan Zheng root->node = leaf; 1364e02119d5SChris Mason 1365e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1366e02119d5SChris Mason btrfs_tree_unlock(root->node); 13677237f183SYan Zheng return root; 13687237f183SYan Zheng } 13697237f183SYan Zheng 13707237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 13717237f183SYan Zheng struct btrfs_fs_info *fs_info) 13727237f183SYan Zheng { 13737237f183SYan Zheng struct btrfs_root *log_root; 13747237f183SYan Zheng 13757237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 13767237f183SYan Zheng if (IS_ERR(log_root)) 13777237f183SYan Zheng return PTR_ERR(log_root); 13787237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 13797237f183SYan Zheng fs_info->log_root_tree = log_root; 13807237f183SYan Zheng return 0; 13817237f183SYan Zheng } 13827237f183SYan Zheng 13837237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 13847237f183SYan Zheng struct btrfs_root *root) 13857237f183SYan Zheng { 13860b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13877237f183SYan Zheng struct btrfs_root *log_root; 13887237f183SYan Zheng struct btrfs_inode_item *inode_item; 13897237f183SYan Zheng 13900b246afaSJeff Mahoney log_root = alloc_log_tree(trans, fs_info); 13917237f183SYan Zheng if (IS_ERR(log_root)) 13927237f183SYan Zheng return PTR_ERR(log_root); 13937237f183SYan Zheng 13947237f183SYan Zheng log_root->last_trans = trans->transid; 13957237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 13967237f183SYan Zheng 13977237f183SYan Zheng inode_item = &log_root->root_item.inode; 13983cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 13993cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 14003cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 1401da17066cSJeff Mahoney btrfs_set_stack_inode_nbytes(inode_item, 14020b246afaSJeff Mahoney fs_info->nodesize); 14033cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 14047237f183SYan Zheng 14055d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 14067237f183SYan Zheng 14077237f183SYan Zheng WARN_ON(root->log_root); 14087237f183SYan Zheng root->log_root = log_root; 14097237f183SYan Zheng root->log_transid = 0; 1410d1433debSMiao Xie root->log_transid_committed = -1; 1411257c62e1SChris Mason root->last_log_commit = 0; 1412e02119d5SChris Mason return 0; 1413e02119d5SChris Mason } 1414e02119d5SChris Mason 141535a3621bSStefan Behrens static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 1416cb517eabSMiao Xie struct btrfs_key *key) 1417e02119d5SChris Mason { 1418e02119d5SChris Mason struct btrfs_root *root; 1419e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 14200f7d52f4SChris Mason struct btrfs_path *path; 142184234f3aSYan Zheng u64 generation; 1422cb517eabSMiao Xie int ret; 1423581c1760SQu Wenruo int level; 1424cb517eabSMiao Xie 1425cb517eabSMiao Xie path = btrfs_alloc_path(); 1426cb517eabSMiao Xie if (!path) 1427cb517eabSMiao Xie return ERR_PTR(-ENOMEM); 14280f7d52f4SChris Mason 142974e4d827SDavid Sterba root = btrfs_alloc_root(fs_info, GFP_NOFS); 1430cb517eabSMiao Xie if (!root) { 1431cb517eabSMiao Xie ret = -ENOMEM; 1432cb517eabSMiao Xie goto alloc_fail; 14330f7d52f4SChris Mason } 14340f7d52f4SChris Mason 1435da17066cSJeff Mahoney __setup_root(root, fs_info, key->objectid); 14360f7d52f4SChris Mason 1437cb517eabSMiao Xie ret = btrfs_find_root(tree_root, key, path, 1438cb517eabSMiao Xie &root->root_item, &root->root_key); 14390f7d52f4SChris Mason if (ret) { 144013a8a7c8SYan, Zheng if (ret > 0) 144113a8a7c8SYan, Zheng ret = -ENOENT; 1442cb517eabSMiao Xie goto find_fail; 14430f7d52f4SChris Mason } 144413a8a7c8SYan, Zheng 144584234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1446581c1760SQu Wenruo level = btrfs_root_level(&root->root_item); 14472ff7e61eSJeff Mahoney root->node = read_tree_block(fs_info, 14482ff7e61eSJeff Mahoney btrfs_root_bytenr(&root->root_item), 1449581c1760SQu Wenruo generation, level, NULL); 145064c043deSLiu Bo if (IS_ERR(root->node)) { 145164c043deSLiu Bo ret = PTR_ERR(root->node); 1452cb517eabSMiao Xie goto find_fail; 1453cb517eabSMiao Xie } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1454cb517eabSMiao Xie ret = -EIO; 145564c043deSLiu Bo free_extent_buffer(root->node); 145664c043deSLiu Bo goto find_fail; 1457416bc658SJosef Bacik } 14585d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 145913a8a7c8SYan, Zheng out: 1460cb517eabSMiao Xie btrfs_free_path(path); 1461cb517eabSMiao Xie return root; 1462cb517eabSMiao Xie 1463cb517eabSMiao Xie find_fail: 1464cb517eabSMiao Xie kfree(root); 1465cb517eabSMiao Xie alloc_fail: 1466cb517eabSMiao Xie root = ERR_PTR(ret); 1467cb517eabSMiao Xie goto out; 1468cb517eabSMiao Xie } 1469cb517eabSMiao Xie 1470cb517eabSMiao Xie struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root, 1471cb517eabSMiao Xie struct btrfs_key *location) 1472cb517eabSMiao Xie { 1473cb517eabSMiao Xie struct btrfs_root *root; 1474cb517eabSMiao Xie 1475cb517eabSMiao Xie root = btrfs_read_tree_root(tree_root, location); 1476cb517eabSMiao Xie if (IS_ERR(root)) 1477cb517eabSMiao Xie return root; 1478cb517eabSMiao Xie 1479cb517eabSMiao Xie if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { 148027cdeb70SMiao Xie set_bit(BTRFS_ROOT_REF_COWS, &root->state); 148108fe4db1SLi Zefan btrfs_check_and_init_root_item(&root->root_item); 148208fe4db1SLi Zefan } 148313a8a7c8SYan, Zheng 14845eda7b5eSChris Mason return root; 14855eda7b5eSChris Mason } 14865eda7b5eSChris Mason 1487cb517eabSMiao Xie int btrfs_init_fs_root(struct btrfs_root *root) 1488cb517eabSMiao Xie { 1489cb517eabSMiao Xie int ret; 14908257b2dcSMiao Xie struct btrfs_subvolume_writers *writers; 1491cb517eabSMiao Xie 1492cb517eabSMiao Xie root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); 1493cb517eabSMiao Xie root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), 1494cb517eabSMiao Xie GFP_NOFS); 1495cb517eabSMiao Xie if (!root->free_ino_pinned || !root->free_ino_ctl) { 1496cb517eabSMiao Xie ret = -ENOMEM; 1497cb517eabSMiao Xie goto fail; 1498cb517eabSMiao Xie } 1499cb517eabSMiao Xie 15008257b2dcSMiao Xie writers = btrfs_alloc_subvolume_writers(); 15018257b2dcSMiao Xie if (IS_ERR(writers)) { 15028257b2dcSMiao Xie ret = PTR_ERR(writers); 15038257b2dcSMiao Xie goto fail; 15048257b2dcSMiao Xie } 15058257b2dcSMiao Xie root->subv_writers = writers; 15068257b2dcSMiao Xie 1507cb517eabSMiao Xie btrfs_init_free_ino_ctl(root); 150857cdc8dbSDavid Sterba spin_lock_init(&root->ino_cache_lock); 150957cdc8dbSDavid Sterba init_waitqueue_head(&root->ino_cache_wait); 1510cb517eabSMiao Xie 1511cb517eabSMiao Xie ret = get_anon_bdev(&root->anon_dev); 1512cb517eabSMiao Xie if (ret) 1513876d2cf1SLiu Bo goto fail; 1514f32e48e9SChandan Rajendra 1515f32e48e9SChandan Rajendra mutex_lock(&root->objectid_mutex); 1516f32e48e9SChandan Rajendra ret = btrfs_find_highest_objectid(root, 1517f32e48e9SChandan Rajendra &root->highest_objectid); 1518f32e48e9SChandan Rajendra if (ret) { 1519f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1520876d2cf1SLiu Bo goto fail; 1521f32e48e9SChandan Rajendra } 1522f32e48e9SChandan Rajendra 1523f32e48e9SChandan Rajendra ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 1524f32e48e9SChandan Rajendra 1525f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1526f32e48e9SChandan Rajendra 1527cb517eabSMiao Xie return 0; 1528cb517eabSMiao Xie fail: 152984db5ccfSDavid Sterba /* The caller is responsible to call btrfs_free_fs_root */ 1530cb517eabSMiao Xie return ret; 1531cb517eabSMiao Xie } 1532cb517eabSMiao Xie 153335bbb97fSJeff Mahoney struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1534cb517eabSMiao Xie u64 root_id) 1535cb517eabSMiao Xie { 1536cb517eabSMiao Xie struct btrfs_root *root; 1537cb517eabSMiao Xie 1538cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1539cb517eabSMiao Xie root = radix_tree_lookup(&fs_info->fs_roots_radix, 1540cb517eabSMiao Xie (unsigned long)root_id); 1541cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1542cb517eabSMiao Xie return root; 1543cb517eabSMiao Xie } 1544cb517eabSMiao Xie 1545cb517eabSMiao Xie int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1546cb517eabSMiao Xie struct btrfs_root *root) 1547cb517eabSMiao Xie { 1548cb517eabSMiao Xie int ret; 1549cb517eabSMiao Xie 1550e1860a77SDavid Sterba ret = radix_tree_preload(GFP_NOFS); 1551cb517eabSMiao Xie if (ret) 1552cb517eabSMiao Xie return ret; 1553cb517eabSMiao Xie 1554cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1555cb517eabSMiao Xie ret = radix_tree_insert(&fs_info->fs_roots_radix, 1556cb517eabSMiao Xie (unsigned long)root->root_key.objectid, 1557cb517eabSMiao Xie root); 1558cb517eabSMiao Xie if (ret == 0) 155927cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1560cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1561cb517eabSMiao Xie radix_tree_preload_end(); 1562cb517eabSMiao Xie 1563cb517eabSMiao Xie return ret; 1564cb517eabSMiao Xie } 1565cb517eabSMiao Xie 1566c00869f1SMiao Xie struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 1567c00869f1SMiao Xie struct btrfs_key *location, 1568c00869f1SMiao Xie bool check_ref) 15695eda7b5eSChris Mason { 15705eda7b5eSChris Mason struct btrfs_root *root; 1571381cf658SDavid Sterba struct btrfs_path *path; 15721d4c08e0SDavid Sterba struct btrfs_key key; 15735eda7b5eSChris Mason int ret; 15745eda7b5eSChris Mason 1575edbd8d4eSChris Mason if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 1576edbd8d4eSChris Mason return fs_info->tree_root; 1577edbd8d4eSChris Mason if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID) 1578edbd8d4eSChris Mason return fs_info->extent_root; 15798f18cf13SChris Mason if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID) 15808f18cf13SChris Mason return fs_info->chunk_root; 15818f18cf13SChris Mason if (location->objectid == BTRFS_DEV_TREE_OBJECTID) 15828f18cf13SChris Mason return fs_info->dev_root; 15830403e47eSYan Zheng if (location->objectid == BTRFS_CSUM_TREE_OBJECTID) 15840403e47eSYan Zheng return fs_info->csum_root; 1585bcef60f2SArne Jansen if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID) 1586bcef60f2SArne Jansen return fs_info->quota_root ? fs_info->quota_root : 1587bcef60f2SArne Jansen ERR_PTR(-ENOENT); 1588f7a81ea4SStefan Behrens if (location->objectid == BTRFS_UUID_TREE_OBJECTID) 1589f7a81ea4SStefan Behrens return fs_info->uuid_root ? fs_info->uuid_root : 1590f7a81ea4SStefan Behrens ERR_PTR(-ENOENT); 159170f6d82eSOmar Sandoval if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) 159270f6d82eSOmar Sandoval return fs_info->free_space_root ? fs_info->free_space_root : 159370f6d82eSOmar Sandoval ERR_PTR(-ENOENT); 15944df27c4dSYan, Zheng again: 1595cb517eabSMiao Xie root = btrfs_lookup_fs_root(fs_info, location->objectid); 159648475471SStefan Behrens if (root) { 1597c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) 159848475471SStefan Behrens return ERR_PTR(-ENOENT); 15995eda7b5eSChris Mason return root; 160048475471SStefan Behrens } 16015eda7b5eSChris Mason 1602cb517eabSMiao Xie root = btrfs_read_fs_root(fs_info->tree_root, location); 16035eda7b5eSChris Mason if (IS_ERR(root)) 16045eda7b5eSChris Mason return root; 16053394e160SChris Mason 1606c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1607d68fc57bSYan, Zheng ret = -ENOENT; 1608d68fc57bSYan, Zheng goto fail; 1609d68fc57bSYan, Zheng } 1610d68fc57bSYan, Zheng 1611cb517eabSMiao Xie ret = btrfs_init_fs_root(root); 1612cb517eabSMiao Xie if (ret) 1613cb517eabSMiao Xie goto fail; 1614cb517eabSMiao Xie 1615381cf658SDavid Sterba path = btrfs_alloc_path(); 1616381cf658SDavid Sterba if (!path) { 1617381cf658SDavid Sterba ret = -ENOMEM; 1618381cf658SDavid Sterba goto fail; 1619381cf658SDavid Sterba } 16201d4c08e0SDavid Sterba key.objectid = BTRFS_ORPHAN_OBJECTID; 16211d4c08e0SDavid Sterba key.type = BTRFS_ORPHAN_ITEM_KEY; 16221d4c08e0SDavid Sterba key.offset = location->objectid; 16231d4c08e0SDavid Sterba 16241d4c08e0SDavid Sterba ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1625381cf658SDavid Sterba btrfs_free_path(path); 1626d68fc57bSYan, Zheng if (ret < 0) 1627d68fc57bSYan, Zheng goto fail; 1628d68fc57bSYan, Zheng if (ret == 0) 162927cdeb70SMiao Xie set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1630d68fc57bSYan, Zheng 1631cb517eabSMiao Xie ret = btrfs_insert_fs_root(fs_info, root); 16320f7d52f4SChris Mason if (ret) { 16334df27c4dSYan, Zheng if (ret == -EEXIST) { 163484db5ccfSDavid Sterba btrfs_free_fs_root(root); 16354df27c4dSYan, Zheng goto again; 16360f7d52f4SChris Mason } 16374df27c4dSYan, Zheng goto fail; 16384df27c4dSYan, Zheng } 1639edbd8d4eSChris Mason return root; 16404df27c4dSYan, Zheng fail: 164184db5ccfSDavid Sterba btrfs_free_fs_root(root); 16424df27c4dSYan, Zheng return ERR_PTR(ret); 1643edbd8d4eSChris Mason } 1644edbd8d4eSChris Mason 164504160088SChris Mason static int btrfs_congested_fn(void *congested_data, int bdi_bits) 164604160088SChris Mason { 164704160088SChris Mason struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; 164804160088SChris Mason int ret = 0; 164904160088SChris Mason struct btrfs_device *device; 165004160088SChris Mason struct backing_dev_info *bdi; 1651b7967db7SChris Mason 16521f78160cSXiao Guangrong rcu_read_lock(); 16531f78160cSXiao Guangrong list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) { 1654dfe25020SChris Mason if (!device->bdev) 1655dfe25020SChris Mason continue; 1656efa7c9f9SJan Kara bdi = device->bdev->bd_bdi; 1657ff9ea323STejun Heo if (bdi_congested(bdi, bdi_bits)) { 165804160088SChris Mason ret = 1; 165904160088SChris Mason break; 166004160088SChris Mason } 166104160088SChris Mason } 16621f78160cSXiao Guangrong rcu_read_unlock(); 166304160088SChris Mason return ret; 166404160088SChris Mason } 166504160088SChris Mason 16668b712842SChris Mason /* 16678b712842SChris Mason * called by the kthread helper functions to finally call the bio end_io 16688b712842SChris Mason * functions. This is where read checksum verification actually happens 16698b712842SChris Mason */ 16708b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work) 1671ce9adaa5SChris Mason { 1672ce9adaa5SChris Mason struct bio *bio; 167397eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 1674ce9adaa5SChris Mason 167597eb6b69SDavid Sterba end_io_wq = container_of(work, struct btrfs_end_io_wq, work); 1676ce9adaa5SChris Mason bio = end_io_wq->bio; 16778b712842SChris Mason 16784e4cbee9SChristoph Hellwig bio->bi_status = end_io_wq->status; 1679ce9adaa5SChris Mason bio->bi_private = end_io_wq->private; 1680ce9adaa5SChris Mason bio->bi_end_io = end_io_wq->end_io; 168197eb6b69SDavid Sterba kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq); 16824246a0b6SChristoph Hellwig bio_endio(bio); 1683ce9adaa5SChris Mason } 168444b8bd7eSChris Mason 1685a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1686a74a4b97SChris Mason { 1687a74a4b97SChris Mason struct btrfs_root *root = arg; 16880b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1689d0278245SMiao Xie int again; 1690a74a4b97SChris Mason 1691d6fd0ae2SOmar Sandoval while (1) { 1692d0278245SMiao Xie again = 0; 16939d1a2a3aSDavid Sterba 1694fd340d0fSJosef Bacik set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1695fd340d0fSJosef Bacik 1696d0278245SMiao Xie /* Make the cleaner go to sleep early. */ 16972ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) 1698d0278245SMiao Xie goto sleep; 1699d0278245SMiao Xie 170090c711abSZygo Blaxell /* 170190c711abSZygo Blaxell * Do not do anything if we might cause open_ctree() to block 170290c711abSZygo Blaxell * before we have finished mounting the filesystem. 170390c711abSZygo Blaxell */ 17040b246afaSJeff Mahoney if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 170590c711abSZygo Blaxell goto sleep; 170690c711abSZygo Blaxell 17070b246afaSJeff Mahoney if (!mutex_trylock(&fs_info->cleaner_mutex)) 1708d0278245SMiao Xie goto sleep; 1709d0278245SMiao Xie 1710dc7f370cSMiao Xie /* 1711dc7f370cSMiao Xie * Avoid the problem that we change the status of the fs 1712dc7f370cSMiao Xie * during the above check and trylock. 1713dc7f370cSMiao Xie */ 17142ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) { 17150b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1716dc7f370cSMiao Xie goto sleep; 1717dc7f370cSMiao Xie } 1718dc7f370cSMiao Xie 17192ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 1720c2d6cb16SFilipe Manana 17219d1a2a3aSDavid Sterba again = btrfs_clean_one_deleted_snapshot(root); 17220b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1723a74a4b97SChris Mason 1724d0278245SMiao Xie /* 172505323cd1SMiao Xie * The defragger has dealt with the R/O remount and umount, 172605323cd1SMiao Xie * needn't do anything special here. 1727d0278245SMiao Xie */ 17280b246afaSJeff Mahoney btrfs_run_defrag_inodes(fs_info); 172967c5e7d4SFilipe Manana 173067c5e7d4SFilipe Manana /* 173167c5e7d4SFilipe Manana * Acquires fs_info->delete_unused_bgs_mutex to avoid racing 173267c5e7d4SFilipe Manana * with relocation (btrfs_relocate_chunk) and relocation 173367c5e7d4SFilipe Manana * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 173467c5e7d4SFilipe Manana * after acquiring fs_info->delete_unused_bgs_mutex. So we 173567c5e7d4SFilipe Manana * can't hold, nor need to, fs_info->cleaner_mutex when deleting 173667c5e7d4SFilipe Manana * unused block groups. 173767c5e7d4SFilipe Manana */ 17380b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 1739d0278245SMiao Xie sleep: 1740fd340d0fSJosef Bacik clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1741d6fd0ae2SOmar Sandoval if (kthread_should_park()) 1742d6fd0ae2SOmar Sandoval kthread_parkme(); 1743d6fd0ae2SOmar Sandoval if (kthread_should_stop()) 1744d6fd0ae2SOmar Sandoval return 0; 1745838fe188SJiri Kosina if (!again) { 1746a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 1747a74a4b97SChris Mason schedule(); 1748a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1749a74a4b97SChris Mason } 1750da288d28SFilipe Manana } 1751a74a4b97SChris Mason } 1752a74a4b97SChris Mason 1753a74a4b97SChris Mason static int transaction_kthread(void *arg) 1754a74a4b97SChris Mason { 1755a74a4b97SChris Mason struct btrfs_root *root = arg; 17560b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1757a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1758a74a4b97SChris Mason struct btrfs_transaction *cur; 17598929ecfaSYan, Zheng u64 transid; 1760a944442cSAllen Pais time64_t now; 1761a74a4b97SChris Mason unsigned long delay; 1762914b2007SJan Kara bool cannot_commit; 1763a74a4b97SChris Mason 1764a74a4b97SChris Mason do { 1765914b2007SJan Kara cannot_commit = false; 17660b246afaSJeff Mahoney delay = HZ * fs_info->commit_interval; 17670b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 1768a74a4b97SChris Mason 17690b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 17700b246afaSJeff Mahoney cur = fs_info->running_transaction; 1771a74a4b97SChris Mason if (!cur) { 17720b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1773a74a4b97SChris Mason goto sleep; 1774a74a4b97SChris Mason } 177531153d81SYan Zheng 1776afd48513SArnd Bergmann now = ktime_get_seconds(); 17774a9d8bdeSMiao Xie if (cur->state < TRANS_STATE_BLOCKED && 1778a514d638SQu Wenruo !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) && 17798b87dc17SDavid Sterba (now < cur->start_time || 17800b246afaSJeff Mahoney now - cur->start_time < fs_info->commit_interval)) { 17810b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1782a74a4b97SChris Mason delay = HZ * 5; 1783a74a4b97SChris Mason goto sleep; 1784a74a4b97SChris Mason } 17858929ecfaSYan, Zheng transid = cur->transid; 17860b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 178756bec294SChris Mason 178879787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 1789354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 1790914b2007SJan Kara if (IS_ERR(trans)) { 1791354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 1792914b2007SJan Kara cannot_commit = true; 179379787eaaSJeff Mahoney goto sleep; 1794914b2007SJan Kara } 17958929ecfaSYan, Zheng if (transid == trans->transid) { 17963a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 17978929ecfaSYan, Zheng } else { 17983a45bb20SJeff Mahoney btrfs_end_transaction(trans); 17998929ecfaSYan, Zheng } 1800a74a4b97SChris Mason sleep: 18010b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 18020b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 1803a74a4b97SChris Mason 18044e121c06SJosef Bacik if (unlikely(test_bit(BTRFS_FS_STATE_ERROR, 18050b246afaSJeff Mahoney &fs_info->fs_state))) 18062ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 18078929ecfaSYan, Zheng if (!kthread_should_stop() && 18080b246afaSJeff Mahoney (!btrfs_transaction_blocked(fs_info) || 1809914b2007SJan Kara cannot_commit)) 1810bc5511d0SNikolay Borisov schedule_timeout_interruptible(delay); 1811a74a4b97SChris Mason } while (!kthread_should_stop()); 1812a74a4b97SChris Mason return 0; 1813a74a4b97SChris Mason } 1814a74a4b97SChris Mason 1815af31f5e5SChris Mason /* 1816af31f5e5SChris Mason * this will find the highest generation in the array of 1817af31f5e5SChris Mason * root backups. The index of the highest array is returned, 1818af31f5e5SChris Mason * or -1 if we can't find anything. 1819af31f5e5SChris Mason * 1820af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1821af31f5e5SChris Mason * generation of the latest root in the array with the generation 1822af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1823af31f5e5SChris Mason */ 1824af31f5e5SChris Mason static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen) 1825af31f5e5SChris Mason { 1826af31f5e5SChris Mason u64 cur; 1827af31f5e5SChris Mason int newest_index = -1; 1828af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1829af31f5e5SChris Mason int i; 1830af31f5e5SChris Mason 1831af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1832af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1833af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1834af31f5e5SChris Mason if (cur == newest_gen) 1835af31f5e5SChris Mason newest_index = i; 1836af31f5e5SChris Mason } 1837af31f5e5SChris Mason 1838af31f5e5SChris Mason /* check to see if we actually wrapped around */ 1839af31f5e5SChris Mason if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) { 1840af31f5e5SChris Mason root_backup = info->super_copy->super_roots; 1841af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1842af31f5e5SChris Mason if (cur == newest_gen) 1843af31f5e5SChris Mason newest_index = 0; 1844af31f5e5SChris Mason } 1845af31f5e5SChris Mason return newest_index; 1846af31f5e5SChris Mason } 1847af31f5e5SChris Mason 1848af31f5e5SChris Mason 1849af31f5e5SChris Mason /* 1850af31f5e5SChris Mason * find the oldest backup so we know where to store new entries 1851af31f5e5SChris Mason * in the backup array. This will set the backup_root_index 1852af31f5e5SChris Mason * field in the fs_info struct 1853af31f5e5SChris Mason */ 1854af31f5e5SChris Mason static void find_oldest_super_backup(struct btrfs_fs_info *info, 1855af31f5e5SChris Mason u64 newest_gen) 1856af31f5e5SChris Mason { 1857af31f5e5SChris Mason int newest_index = -1; 1858af31f5e5SChris Mason 1859af31f5e5SChris Mason newest_index = find_newest_super_backup(info, newest_gen); 1860af31f5e5SChris Mason /* if there was garbage in there, just move along */ 1861af31f5e5SChris Mason if (newest_index == -1) { 1862af31f5e5SChris Mason info->backup_root_index = 0; 1863af31f5e5SChris Mason } else { 1864af31f5e5SChris Mason info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS; 1865af31f5e5SChris Mason } 1866af31f5e5SChris Mason } 1867af31f5e5SChris Mason 1868af31f5e5SChris Mason /* 1869af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1870af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1871af31f5e5SChris Mason * done 1872af31f5e5SChris Mason */ 1873af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1874af31f5e5SChris Mason { 1875af31f5e5SChris Mason int next_backup; 1876af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1877af31f5e5SChris Mason int last_backup; 1878af31f5e5SChris Mason 1879af31f5e5SChris Mason next_backup = info->backup_root_index; 1880af31f5e5SChris Mason last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) % 1881af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 1882af31f5e5SChris Mason 1883af31f5e5SChris Mason /* 1884af31f5e5SChris Mason * just overwrite the last backup if we're at the same generation 1885af31f5e5SChris Mason * this happens only at umount 1886af31f5e5SChris Mason */ 1887af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + last_backup; 1888af31f5e5SChris Mason if (btrfs_backup_tree_root_gen(root_backup) == 1889af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)) 1890af31f5e5SChris Mason next_backup = last_backup; 1891af31f5e5SChris Mason 1892af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1893af31f5e5SChris Mason 1894af31f5e5SChris Mason /* 1895af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1896af31f5e5SChris Mason * regardless of which ones we use today 1897af31f5e5SChris Mason */ 1898af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1899af31f5e5SChris Mason 1900af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1901af31f5e5SChris Mason 1902af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1903af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1904af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1905af31f5e5SChris Mason 1906af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1907af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1908af31f5e5SChris Mason 1909af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1910af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1911af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 1912af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 1913af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 1914af31f5e5SChris Mason 1915af31f5e5SChris Mason btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); 1916af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 1917af31f5e5SChris Mason btrfs_header_generation(info->extent_root->node)); 1918af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 1919af31f5e5SChris Mason btrfs_header_level(info->extent_root->node)); 1920af31f5e5SChris Mason 19217c7e82a7SChris Mason /* 19227c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 19237c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 19247c7e82a7SChris Mason */ 19257c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 19267c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 19277c7e82a7SChris Mason info->fs_root->node->start); 1928af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 1929af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 1930af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 1931af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 19327c7e82a7SChris Mason } 1933af31f5e5SChris Mason 1934af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 1935af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 1936af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 1937af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 1938af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 1939af31f5e5SChris Mason 1940af31f5e5SChris Mason btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); 1941af31f5e5SChris Mason btrfs_set_backup_csum_root_gen(root_backup, 1942af31f5e5SChris Mason btrfs_header_generation(info->csum_root->node)); 1943af31f5e5SChris Mason btrfs_set_backup_csum_root_level(root_backup, 1944af31f5e5SChris Mason btrfs_header_level(info->csum_root->node)); 1945af31f5e5SChris Mason 1946af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 1947af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 1948af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 1949af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 1950af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 1951af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 1952af31f5e5SChris Mason 1953af31f5e5SChris Mason /* 1954af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 1955af31f5e5SChris Mason * for the next commit 1956af31f5e5SChris Mason */ 1957af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 1958af31f5e5SChris Mason &info->super_for_commit->super_roots, 1959af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 1960af31f5e5SChris Mason } 1961af31f5e5SChris Mason 1962af31f5e5SChris Mason /* 1963af31f5e5SChris Mason * this copies info out of the root backup array and back into 1964af31f5e5SChris Mason * the in-memory super block. It is meant to help iterate through 1965af31f5e5SChris Mason * the array, so you send it the number of backups you've already 1966af31f5e5SChris Mason * tried and the last backup index you used. 1967af31f5e5SChris Mason * 1968af31f5e5SChris Mason * this returns -1 when it has tried all the backups 1969af31f5e5SChris Mason */ 1970af31f5e5SChris Mason static noinline int next_root_backup(struct btrfs_fs_info *info, 1971af31f5e5SChris Mason struct btrfs_super_block *super, 1972af31f5e5SChris Mason int *num_backups_tried, int *backup_index) 1973af31f5e5SChris Mason { 1974af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1975af31f5e5SChris Mason int newest = *backup_index; 1976af31f5e5SChris Mason 1977af31f5e5SChris Mason if (*num_backups_tried == 0) { 1978af31f5e5SChris Mason u64 gen = btrfs_super_generation(super); 1979af31f5e5SChris Mason 1980af31f5e5SChris Mason newest = find_newest_super_backup(info, gen); 1981af31f5e5SChris Mason if (newest == -1) 1982af31f5e5SChris Mason return -1; 1983af31f5e5SChris Mason 1984af31f5e5SChris Mason *backup_index = newest; 1985af31f5e5SChris Mason *num_backups_tried = 1; 1986af31f5e5SChris Mason } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) { 1987af31f5e5SChris Mason /* we've tried all the backups, all done */ 1988af31f5e5SChris Mason return -1; 1989af31f5e5SChris Mason } else { 1990af31f5e5SChris Mason /* jump to the next oldest backup */ 1991af31f5e5SChris Mason newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) % 1992af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 1993af31f5e5SChris Mason *backup_index = newest; 1994af31f5e5SChris Mason *num_backups_tried += 1; 1995af31f5e5SChris Mason } 1996af31f5e5SChris Mason root_backup = super->super_roots + newest; 1997af31f5e5SChris Mason 1998af31f5e5SChris Mason btrfs_set_super_generation(super, 1999af31f5e5SChris Mason btrfs_backup_tree_root_gen(root_backup)); 2000af31f5e5SChris Mason btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 2001af31f5e5SChris Mason btrfs_set_super_root_level(super, 2002af31f5e5SChris Mason btrfs_backup_tree_root_level(root_backup)); 2003af31f5e5SChris Mason btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 2004af31f5e5SChris Mason 2005af31f5e5SChris Mason /* 2006af31f5e5SChris Mason * fixme: the total bytes and num_devices need to match or we should 2007af31f5e5SChris Mason * need a fsck 2008af31f5e5SChris Mason */ 2009af31f5e5SChris Mason btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 2010af31f5e5SChris Mason btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 2011af31f5e5SChris Mason return 0; 2012af31f5e5SChris Mason } 2013af31f5e5SChris Mason 20147abadb64SLiu Bo /* helper to cleanup workers */ 20157abadb64SLiu Bo static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 20167abadb64SLiu Bo { 2017dc6e3209SQu Wenruo btrfs_destroy_workqueue(fs_info->fixup_workers); 2018afe3d242SQu Wenruo btrfs_destroy_workqueue(fs_info->delalloc_workers); 20195cdc7ad3SQu Wenruo btrfs_destroy_workqueue(fs_info->workers); 2020fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_workers); 2021fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_raid56_workers); 20228b110e39SMiao Xie btrfs_destroy_workqueue(fs_info->endio_repair_workers); 2023d05a33acSQu Wenruo btrfs_destroy_workqueue(fs_info->rmw_workers); 2024fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_write_workers); 2025fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 2026a8c93d4eSQu Wenruo btrfs_destroy_workqueue(fs_info->submit_workers); 20275b3bc44eSQu Wenruo btrfs_destroy_workqueue(fs_info->delayed_workers); 2028e66f0bb1SQu Wenruo btrfs_destroy_workqueue(fs_info->caching_workers); 2029736cfa15SQu Wenruo btrfs_destroy_workqueue(fs_info->readahead_workers); 2030a44903abSQu Wenruo btrfs_destroy_workqueue(fs_info->flush_workers); 2031fc97fab0SQu Wenruo btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2032a79b7d4bSChris Mason btrfs_destroy_workqueue(fs_info->extent_workers); 2033a9b9477dSFilipe Manana /* 2034a9b9477dSFilipe Manana * Now that all other work queues are destroyed, we can safely destroy 2035a9b9477dSFilipe Manana * the queues used for metadata I/O, since tasks from those other work 2036a9b9477dSFilipe Manana * queues can do metadata I/O operations. 2037a9b9477dSFilipe Manana */ 2038a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_workers); 2039a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_write_workers); 20407abadb64SLiu Bo } 20417abadb64SLiu Bo 20422e9f5954SRashika static void free_root_extent_buffers(struct btrfs_root *root) 20432e9f5954SRashika { 20442e9f5954SRashika if (root) { 20452e9f5954SRashika free_extent_buffer(root->node); 20462e9f5954SRashika free_extent_buffer(root->commit_root); 20472e9f5954SRashika root->node = NULL; 20482e9f5954SRashika root->commit_root = NULL; 20492e9f5954SRashika } 20502e9f5954SRashika } 20512e9f5954SRashika 2052af31f5e5SChris Mason /* helper to cleanup tree roots */ 2053af31f5e5SChris Mason static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root) 2054af31f5e5SChris Mason { 20552e9f5954SRashika free_root_extent_buffers(info->tree_root); 2056655b09feSJosef Bacik 20572e9f5954SRashika free_root_extent_buffers(info->dev_root); 20582e9f5954SRashika free_root_extent_buffers(info->extent_root); 20592e9f5954SRashika free_root_extent_buffers(info->csum_root); 20602e9f5954SRashika free_root_extent_buffers(info->quota_root); 20612e9f5954SRashika free_root_extent_buffers(info->uuid_root); 20622e9f5954SRashika if (chunk_root) 20632e9f5954SRashika free_root_extent_buffers(info->chunk_root); 206470f6d82eSOmar Sandoval free_root_extent_buffers(info->free_space_root); 2065af31f5e5SChris Mason } 2066af31f5e5SChris Mason 2067faa2dbf0SJosef Bacik void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2068171f6537SJosef Bacik { 2069171f6537SJosef Bacik int ret; 2070171f6537SJosef Bacik struct btrfs_root *gang[8]; 2071171f6537SJosef Bacik int i; 2072171f6537SJosef Bacik 2073171f6537SJosef Bacik while (!list_empty(&fs_info->dead_roots)) { 2074171f6537SJosef Bacik gang[0] = list_entry(fs_info->dead_roots.next, 2075171f6537SJosef Bacik struct btrfs_root, root_list); 2076171f6537SJosef Bacik list_del(&gang[0]->root_list); 2077171f6537SJosef Bacik 207827cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) { 2079cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[0]); 2080171f6537SJosef Bacik } else { 2081171f6537SJosef Bacik free_extent_buffer(gang[0]->node); 2082171f6537SJosef Bacik free_extent_buffer(gang[0]->commit_root); 2083b0feb9d9SMiao Xie btrfs_put_fs_root(gang[0]); 2084171f6537SJosef Bacik } 2085171f6537SJosef Bacik } 2086171f6537SJosef Bacik 2087171f6537SJosef Bacik while (1) { 2088171f6537SJosef Bacik ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2089171f6537SJosef Bacik (void **)gang, 0, 2090171f6537SJosef Bacik ARRAY_SIZE(gang)); 2091171f6537SJosef Bacik if (!ret) 2092171f6537SJosef Bacik break; 2093171f6537SJosef Bacik for (i = 0; i < ret; i++) 2094cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2095171f6537SJosef Bacik } 20961a4319ccSLiu Bo 20971a4319ccSLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 20981a4319ccSLiu Bo btrfs_free_log_root_tree(NULL, fs_info); 20992ff7e61eSJeff Mahoney btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents); 21001a4319ccSLiu Bo } 2101171f6537SJosef Bacik } 2102af31f5e5SChris Mason 2103638aa7edSEric Sandeen static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2104638aa7edSEric Sandeen { 2105638aa7edSEric Sandeen mutex_init(&fs_info->scrub_lock); 2106638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_running, 0); 2107638aa7edSEric Sandeen atomic_set(&fs_info->scrub_pause_req, 0); 2108638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_paused, 0); 2109638aa7edSEric Sandeen atomic_set(&fs_info->scrub_cancel_req, 0); 2110638aa7edSEric Sandeen init_waitqueue_head(&fs_info->scrub_pause_wait); 2111ff09c4caSAnand Jain refcount_set(&fs_info->scrub_workers_refcnt, 0); 2112638aa7edSEric Sandeen } 2113638aa7edSEric Sandeen 2114779a65a4SEric Sandeen static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2115779a65a4SEric Sandeen { 2116779a65a4SEric Sandeen spin_lock_init(&fs_info->balance_lock); 2117779a65a4SEric Sandeen mutex_init(&fs_info->balance_mutex); 2118779a65a4SEric Sandeen atomic_set(&fs_info->balance_pause_req, 0); 2119779a65a4SEric Sandeen atomic_set(&fs_info->balance_cancel_req, 0); 2120779a65a4SEric Sandeen fs_info->balance_ctl = NULL; 2121779a65a4SEric Sandeen init_waitqueue_head(&fs_info->balance_wait_q); 2122779a65a4SEric Sandeen } 2123779a65a4SEric Sandeen 21246bccf3abSJeff Mahoney static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info) 2125f37938e0SEric Sandeen { 21262ff7e61eSJeff Mahoney struct inode *inode = fs_info->btree_inode; 21272ff7e61eSJeff Mahoney 21282ff7e61eSJeff Mahoney inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 21292ff7e61eSJeff Mahoney set_nlink(inode, 1); 2130f37938e0SEric Sandeen /* 2131f37938e0SEric Sandeen * we set the i_size on the btree inode to the max possible int. 2132f37938e0SEric Sandeen * the real end of the address space is determined by all of 2133f37938e0SEric Sandeen * the devices in the system 2134f37938e0SEric Sandeen */ 21352ff7e61eSJeff Mahoney inode->i_size = OFFSET_MAX; 21362ff7e61eSJeff Mahoney inode->i_mapping->a_ops = &btree_aops; 2137f37938e0SEric Sandeen 21382ff7e61eSJeff Mahoney RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 213943eb5f29SQu Wenruo extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree, 214043eb5f29SQu Wenruo IO_TREE_INODE_IO, inode); 21417b439738SDavid Sterba BTRFS_I(inode)->io_tree.track_uptodate = false; 21422ff7e61eSJeff Mahoney extent_map_tree_init(&BTRFS_I(inode)->extent_tree); 2143f37938e0SEric Sandeen 21442ff7e61eSJeff Mahoney BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops; 2145f37938e0SEric Sandeen 21462ff7e61eSJeff Mahoney BTRFS_I(inode)->root = fs_info->tree_root; 21472ff7e61eSJeff Mahoney memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key)); 21482ff7e61eSJeff Mahoney set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); 21492ff7e61eSJeff Mahoney btrfs_insert_inode_hash(inode); 2150f37938e0SEric Sandeen } 2151f37938e0SEric Sandeen 2152ad618368SEric Sandeen static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2153ad618368SEric Sandeen { 2154ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2155129827e3SDavid Sterba init_rwsem(&fs_info->dev_replace.rwsem); 21567f8d236aSDavid Sterba init_waitqueue_head(&fs_info->dev_replace.replace_wait); 2157ad618368SEric Sandeen } 2158ad618368SEric Sandeen 2159f9e92e40SEric Sandeen static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2160f9e92e40SEric Sandeen { 2161f9e92e40SEric Sandeen spin_lock_init(&fs_info->qgroup_lock); 2162f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_ioctl_lock); 2163f9e92e40SEric Sandeen fs_info->qgroup_tree = RB_ROOT; 2164f9e92e40SEric Sandeen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2165f9e92e40SEric Sandeen fs_info->qgroup_seq = 1; 2166f9e92e40SEric Sandeen fs_info->qgroup_ulist = NULL; 2167d2c609b8SJeff Mahoney fs_info->qgroup_rescan_running = false; 2168f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_rescan_lock); 2169f9e92e40SEric Sandeen } 2170f9e92e40SEric Sandeen 21712a458198SEric Sandeen static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info, 21722a458198SEric Sandeen struct btrfs_fs_devices *fs_devices) 21732a458198SEric Sandeen { 2174f7b885beSAnand Jain u32 max_active = fs_info->thread_pool_size; 21756f011058SDavid Sterba unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 21762a458198SEric Sandeen 21772a458198SEric Sandeen fs_info->workers = 2178cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "worker", 2179cb001095SJeff Mahoney flags | WQ_HIGHPRI, max_active, 16); 21802a458198SEric Sandeen 21812a458198SEric Sandeen fs_info->delalloc_workers = 2182cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delalloc", 2183cb001095SJeff Mahoney flags, max_active, 2); 21842a458198SEric Sandeen 21852a458198SEric Sandeen fs_info->flush_workers = 2186cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "flush_delalloc", 2187cb001095SJeff Mahoney flags, max_active, 0); 21882a458198SEric Sandeen 21892a458198SEric Sandeen fs_info->caching_workers = 2190cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); 21912a458198SEric Sandeen 21922a458198SEric Sandeen /* 21932a458198SEric Sandeen * a higher idle thresh on the submit workers makes it much more 21942a458198SEric Sandeen * likely that bios will be send down in a sane order to the 21952a458198SEric Sandeen * devices 21962a458198SEric Sandeen */ 21972a458198SEric Sandeen fs_info->submit_workers = 2198cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "submit", flags, 21992a458198SEric Sandeen min_t(u64, fs_devices->num_devices, 22002a458198SEric Sandeen max_active), 64); 22012a458198SEric Sandeen 22022a458198SEric Sandeen fs_info->fixup_workers = 2203cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); 22042a458198SEric Sandeen 22052a458198SEric Sandeen /* 22062a458198SEric Sandeen * endios are largely parallel and should have a very 22072a458198SEric Sandeen * low idle thresh 22082a458198SEric Sandeen */ 22092a458198SEric Sandeen fs_info->endio_workers = 2210cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4); 22112a458198SEric Sandeen fs_info->endio_meta_workers = 2212cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta", flags, 2213cb001095SJeff Mahoney max_active, 4); 22142a458198SEric Sandeen fs_info->endio_meta_write_workers = 2215cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags, 2216cb001095SJeff Mahoney max_active, 2); 22172a458198SEric Sandeen fs_info->endio_raid56_workers = 2218cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-raid56", flags, 2219cb001095SJeff Mahoney max_active, 4); 22202a458198SEric Sandeen fs_info->endio_repair_workers = 2221cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0); 22222a458198SEric Sandeen fs_info->rmw_workers = 2223cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2); 22242a458198SEric Sandeen fs_info->endio_write_workers = 2225cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-write", flags, 2226cb001095SJeff Mahoney max_active, 2); 22272a458198SEric Sandeen fs_info->endio_freespace_worker = 2228cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "freespace-write", flags, 2229cb001095SJeff Mahoney max_active, 0); 22302a458198SEric Sandeen fs_info->delayed_workers = 2231cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, 2232cb001095SJeff Mahoney max_active, 0); 22332a458198SEric Sandeen fs_info->readahead_workers = 2234cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "readahead", flags, 2235cb001095SJeff Mahoney max_active, 2); 22362a458198SEric Sandeen fs_info->qgroup_rescan_workers = 2237cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); 22382a458198SEric Sandeen fs_info->extent_workers = 2239cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "extent-refs", flags, 22402a458198SEric Sandeen min_t(u64, fs_devices->num_devices, 22412a458198SEric Sandeen max_active), 8); 22422a458198SEric Sandeen 22432a458198SEric Sandeen if (!(fs_info->workers && fs_info->delalloc_workers && 22442a458198SEric Sandeen fs_info->submit_workers && fs_info->flush_workers && 22452a458198SEric Sandeen fs_info->endio_workers && fs_info->endio_meta_workers && 22462a458198SEric Sandeen fs_info->endio_meta_write_workers && 22472a458198SEric Sandeen fs_info->endio_repair_workers && 22482a458198SEric Sandeen fs_info->endio_write_workers && fs_info->endio_raid56_workers && 22492a458198SEric Sandeen fs_info->endio_freespace_worker && fs_info->rmw_workers && 22502a458198SEric Sandeen fs_info->caching_workers && fs_info->readahead_workers && 22512a458198SEric Sandeen fs_info->fixup_workers && fs_info->delayed_workers && 22522a458198SEric Sandeen fs_info->extent_workers && 22532a458198SEric Sandeen fs_info->qgroup_rescan_workers)) { 22542a458198SEric Sandeen return -ENOMEM; 22552a458198SEric Sandeen } 22562a458198SEric Sandeen 22572a458198SEric Sandeen return 0; 22582a458198SEric Sandeen } 22592a458198SEric Sandeen 22606d97c6e3SJohannes Thumshirn static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type) 22616d97c6e3SJohannes Thumshirn { 22626d97c6e3SJohannes Thumshirn struct crypto_shash *csum_shash; 22636d97c6e3SJohannes Thumshirn const char *csum_name = btrfs_super_csum_name(csum_type); 22646d97c6e3SJohannes Thumshirn 22656d97c6e3SJohannes Thumshirn csum_shash = crypto_alloc_shash(csum_name, 0, 0); 22666d97c6e3SJohannes Thumshirn 22676d97c6e3SJohannes Thumshirn if (IS_ERR(csum_shash)) { 22686d97c6e3SJohannes Thumshirn btrfs_err(fs_info, "error allocating %s hash for checksum", 22696d97c6e3SJohannes Thumshirn csum_name); 22706d97c6e3SJohannes Thumshirn return PTR_ERR(csum_shash); 22716d97c6e3SJohannes Thumshirn } 22726d97c6e3SJohannes Thumshirn 22736d97c6e3SJohannes Thumshirn fs_info->csum_shash = csum_shash; 22746d97c6e3SJohannes Thumshirn 22756d97c6e3SJohannes Thumshirn return 0; 22766d97c6e3SJohannes Thumshirn } 22776d97c6e3SJohannes Thumshirn 22786d97c6e3SJohannes Thumshirn static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info) 22796d97c6e3SJohannes Thumshirn { 22806d97c6e3SJohannes Thumshirn crypto_free_shash(fs_info->csum_shash); 22816d97c6e3SJohannes Thumshirn } 22826d97c6e3SJohannes Thumshirn 228363443bf5SEric Sandeen static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 228463443bf5SEric Sandeen struct btrfs_fs_devices *fs_devices) 228563443bf5SEric Sandeen { 228663443bf5SEric Sandeen int ret; 228763443bf5SEric Sandeen struct btrfs_root *log_tree_root; 228863443bf5SEric Sandeen struct btrfs_super_block *disk_super = fs_info->super_copy; 228963443bf5SEric Sandeen u64 bytenr = btrfs_super_log_root(disk_super); 2290581c1760SQu Wenruo int level = btrfs_super_log_root_level(disk_super); 229163443bf5SEric Sandeen 229263443bf5SEric Sandeen if (fs_devices->rw_devices == 0) { 2293f14d104dSDavid Sterba btrfs_warn(fs_info, "log replay required on RO media"); 229463443bf5SEric Sandeen return -EIO; 229563443bf5SEric Sandeen } 229663443bf5SEric Sandeen 229774e4d827SDavid Sterba log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL); 229863443bf5SEric Sandeen if (!log_tree_root) 229963443bf5SEric Sandeen return -ENOMEM; 230063443bf5SEric Sandeen 2301da17066cSJeff Mahoney __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID); 230263443bf5SEric Sandeen 23032ff7e61eSJeff Mahoney log_tree_root->node = read_tree_block(fs_info, bytenr, 2304581c1760SQu Wenruo fs_info->generation + 1, 2305581c1760SQu Wenruo level, NULL); 230664c043deSLiu Bo if (IS_ERR(log_tree_root->node)) { 2307f14d104dSDavid Sterba btrfs_warn(fs_info, "failed to read log tree"); 23080eeff236SLiu Bo ret = PTR_ERR(log_tree_root->node); 230964c043deSLiu Bo kfree(log_tree_root); 23100eeff236SLiu Bo return ret; 231164c043deSLiu Bo } else if (!extent_buffer_uptodate(log_tree_root->node)) { 2312f14d104dSDavid Sterba btrfs_err(fs_info, "failed to read log tree"); 231363443bf5SEric Sandeen free_extent_buffer(log_tree_root->node); 231463443bf5SEric Sandeen kfree(log_tree_root); 231563443bf5SEric Sandeen return -EIO; 231663443bf5SEric Sandeen } 231763443bf5SEric Sandeen /* returns with log_tree_root freed on success */ 231863443bf5SEric Sandeen ret = btrfs_recover_log_trees(log_tree_root); 231963443bf5SEric Sandeen if (ret) { 23200b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 232163443bf5SEric Sandeen "Failed to recover log tree"); 232263443bf5SEric Sandeen free_extent_buffer(log_tree_root->node); 232363443bf5SEric Sandeen kfree(log_tree_root); 232463443bf5SEric Sandeen return ret; 232563443bf5SEric Sandeen } 232663443bf5SEric Sandeen 2327bc98a42cSDavid Howells if (sb_rdonly(fs_info->sb)) { 23286bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 232963443bf5SEric Sandeen if (ret) 233063443bf5SEric Sandeen return ret; 233163443bf5SEric Sandeen } 233263443bf5SEric Sandeen 233363443bf5SEric Sandeen return 0; 233463443bf5SEric Sandeen } 233563443bf5SEric Sandeen 23366bccf3abSJeff Mahoney static int btrfs_read_roots(struct btrfs_fs_info *fs_info) 23374bbcaa64SEric Sandeen { 23386bccf3abSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 2339a4f3d2c4SDavid Sterba struct btrfs_root *root; 23404bbcaa64SEric Sandeen struct btrfs_key location; 23414bbcaa64SEric Sandeen int ret; 23424bbcaa64SEric Sandeen 23436bccf3abSJeff Mahoney BUG_ON(!fs_info->tree_root); 23446bccf3abSJeff Mahoney 23454bbcaa64SEric Sandeen location.objectid = BTRFS_EXTENT_TREE_OBJECTID; 23464bbcaa64SEric Sandeen location.type = BTRFS_ROOT_ITEM_KEY; 23474bbcaa64SEric Sandeen location.offset = 0; 23484bbcaa64SEric Sandeen 2349a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2350f50f4353SLiu Bo if (IS_ERR(root)) { 2351f50f4353SLiu Bo ret = PTR_ERR(root); 2352f50f4353SLiu Bo goto out; 2353f50f4353SLiu Bo } 2354a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2355a4f3d2c4SDavid Sterba fs_info->extent_root = root; 23564bbcaa64SEric Sandeen 23574bbcaa64SEric Sandeen location.objectid = BTRFS_DEV_TREE_OBJECTID; 2358a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2359f50f4353SLiu Bo if (IS_ERR(root)) { 2360f50f4353SLiu Bo ret = PTR_ERR(root); 2361f50f4353SLiu Bo goto out; 2362f50f4353SLiu Bo } 2363a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2364a4f3d2c4SDavid Sterba fs_info->dev_root = root; 23654bbcaa64SEric Sandeen btrfs_init_devices_late(fs_info); 23664bbcaa64SEric Sandeen 23674bbcaa64SEric Sandeen location.objectid = BTRFS_CSUM_TREE_OBJECTID; 2368a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2369f50f4353SLiu Bo if (IS_ERR(root)) { 2370f50f4353SLiu Bo ret = PTR_ERR(root); 2371f50f4353SLiu Bo goto out; 2372f50f4353SLiu Bo } 2373a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2374a4f3d2c4SDavid Sterba fs_info->csum_root = root; 23754bbcaa64SEric Sandeen 23764bbcaa64SEric Sandeen location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2377a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2378a4f3d2c4SDavid Sterba if (!IS_ERR(root)) { 2379a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2380afcdd129SJosef Bacik set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); 2381a4f3d2c4SDavid Sterba fs_info->quota_root = root; 23824bbcaa64SEric Sandeen } 23834bbcaa64SEric Sandeen 23844bbcaa64SEric Sandeen location.objectid = BTRFS_UUID_TREE_OBJECTID; 2385a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2386a4f3d2c4SDavid Sterba if (IS_ERR(root)) { 2387a4f3d2c4SDavid Sterba ret = PTR_ERR(root); 23884bbcaa64SEric Sandeen if (ret != -ENOENT) 2389f50f4353SLiu Bo goto out; 23904bbcaa64SEric Sandeen } else { 2391a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2392a4f3d2c4SDavid Sterba fs_info->uuid_root = root; 23934bbcaa64SEric Sandeen } 23944bbcaa64SEric Sandeen 239570f6d82eSOmar Sandoval if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 239670f6d82eSOmar Sandoval location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID; 239770f6d82eSOmar Sandoval root = btrfs_read_tree_root(tree_root, &location); 2398f50f4353SLiu Bo if (IS_ERR(root)) { 2399f50f4353SLiu Bo ret = PTR_ERR(root); 2400f50f4353SLiu Bo goto out; 2401f50f4353SLiu Bo } 240270f6d82eSOmar Sandoval set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 240370f6d82eSOmar Sandoval fs_info->free_space_root = root; 240470f6d82eSOmar Sandoval } 240570f6d82eSOmar Sandoval 24064bbcaa64SEric Sandeen return 0; 2407f50f4353SLiu Bo out: 2408f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", 2409f50f4353SLiu Bo location.objectid, ret); 2410f50f4353SLiu Bo return ret; 24114bbcaa64SEric Sandeen } 24124bbcaa64SEric Sandeen 2413069ec957SQu Wenruo /* 2414069ec957SQu Wenruo * Real super block validation 2415069ec957SQu Wenruo * NOTE: super csum type and incompat features will not be checked here. 2416069ec957SQu Wenruo * 2417069ec957SQu Wenruo * @sb: super block to check 2418069ec957SQu Wenruo * @mirror_num: the super block number to check its bytenr: 2419069ec957SQu Wenruo * 0 the primary (1st) sb 2420069ec957SQu Wenruo * 1, 2 2nd and 3rd backup copy 2421069ec957SQu Wenruo * -1 skip bytenr check 2422069ec957SQu Wenruo */ 2423069ec957SQu Wenruo static int validate_super(struct btrfs_fs_info *fs_info, 2424069ec957SQu Wenruo struct btrfs_super_block *sb, int mirror_num) 242521a852b0SQu Wenruo { 242621a852b0SQu Wenruo u64 nodesize = btrfs_super_nodesize(sb); 242721a852b0SQu Wenruo u64 sectorsize = btrfs_super_sectorsize(sb); 242821a852b0SQu Wenruo int ret = 0; 242921a852b0SQu Wenruo 243021a852b0SQu Wenruo if (btrfs_super_magic(sb) != BTRFS_MAGIC) { 243121a852b0SQu Wenruo btrfs_err(fs_info, "no valid FS found"); 243221a852b0SQu Wenruo ret = -EINVAL; 243321a852b0SQu Wenruo } 243421a852b0SQu Wenruo if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { 243521a852b0SQu Wenruo btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", 243621a852b0SQu Wenruo btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); 243721a852b0SQu Wenruo ret = -EINVAL; 243821a852b0SQu Wenruo } 243921a852b0SQu Wenruo if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 244021a852b0SQu Wenruo btrfs_err(fs_info, "tree_root level too big: %d >= %d", 244121a852b0SQu Wenruo btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 244221a852b0SQu Wenruo ret = -EINVAL; 244321a852b0SQu Wenruo } 244421a852b0SQu Wenruo if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 244521a852b0SQu Wenruo btrfs_err(fs_info, "chunk_root level too big: %d >= %d", 244621a852b0SQu Wenruo btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 244721a852b0SQu Wenruo ret = -EINVAL; 244821a852b0SQu Wenruo } 244921a852b0SQu Wenruo if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 245021a852b0SQu Wenruo btrfs_err(fs_info, "log_root level too big: %d >= %d", 245121a852b0SQu Wenruo btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 245221a852b0SQu Wenruo ret = -EINVAL; 245321a852b0SQu Wenruo } 245421a852b0SQu Wenruo 245521a852b0SQu Wenruo /* 245621a852b0SQu Wenruo * Check sectorsize and nodesize first, other check will need it. 245721a852b0SQu Wenruo * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. 245821a852b0SQu Wenruo */ 245921a852b0SQu Wenruo if (!is_power_of_2(sectorsize) || sectorsize < 4096 || 246021a852b0SQu Wenruo sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { 246121a852b0SQu Wenruo btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); 246221a852b0SQu Wenruo ret = -EINVAL; 246321a852b0SQu Wenruo } 246421a852b0SQu Wenruo /* Only PAGE SIZE is supported yet */ 246521a852b0SQu Wenruo if (sectorsize != PAGE_SIZE) { 246621a852b0SQu Wenruo btrfs_err(fs_info, 246721a852b0SQu Wenruo "sectorsize %llu not supported yet, only support %lu", 246821a852b0SQu Wenruo sectorsize, PAGE_SIZE); 246921a852b0SQu Wenruo ret = -EINVAL; 247021a852b0SQu Wenruo } 247121a852b0SQu Wenruo if (!is_power_of_2(nodesize) || nodesize < sectorsize || 247221a852b0SQu Wenruo nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { 247321a852b0SQu Wenruo btrfs_err(fs_info, "invalid nodesize %llu", nodesize); 247421a852b0SQu Wenruo ret = -EINVAL; 247521a852b0SQu Wenruo } 247621a852b0SQu Wenruo if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { 247721a852b0SQu Wenruo btrfs_err(fs_info, "invalid leafsize %u, should be %llu", 247821a852b0SQu Wenruo le32_to_cpu(sb->__unused_leafsize), nodesize); 247921a852b0SQu Wenruo ret = -EINVAL; 248021a852b0SQu Wenruo } 248121a852b0SQu Wenruo 248221a852b0SQu Wenruo /* Root alignment check */ 248321a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { 248421a852b0SQu Wenruo btrfs_warn(fs_info, "tree_root block unaligned: %llu", 248521a852b0SQu Wenruo btrfs_super_root(sb)); 248621a852b0SQu Wenruo ret = -EINVAL; 248721a852b0SQu Wenruo } 248821a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { 248921a852b0SQu Wenruo btrfs_warn(fs_info, "chunk_root block unaligned: %llu", 249021a852b0SQu Wenruo btrfs_super_chunk_root(sb)); 249121a852b0SQu Wenruo ret = -EINVAL; 249221a852b0SQu Wenruo } 249321a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { 249421a852b0SQu Wenruo btrfs_warn(fs_info, "log_root block unaligned: %llu", 249521a852b0SQu Wenruo btrfs_super_log_root(sb)); 249621a852b0SQu Wenruo ret = -EINVAL; 249721a852b0SQu Wenruo } 249821a852b0SQu Wenruo 2499de37aa51SNikolay Borisov if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, 25007239ff4bSNikolay Borisov BTRFS_FSID_SIZE) != 0) { 250121a852b0SQu Wenruo btrfs_err(fs_info, 25027239ff4bSNikolay Borisov "dev_item UUID does not match metadata fsid: %pU != %pU", 2503de37aa51SNikolay Borisov fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); 250421a852b0SQu Wenruo ret = -EINVAL; 250521a852b0SQu Wenruo } 250621a852b0SQu Wenruo 250721a852b0SQu Wenruo /* 250821a852b0SQu Wenruo * Hint to catch really bogus numbers, bitflips or so, more exact checks are 250921a852b0SQu Wenruo * done later 251021a852b0SQu Wenruo */ 251121a852b0SQu Wenruo if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { 251221a852b0SQu Wenruo btrfs_err(fs_info, "bytes_used is too small %llu", 251321a852b0SQu Wenruo btrfs_super_bytes_used(sb)); 251421a852b0SQu Wenruo ret = -EINVAL; 251521a852b0SQu Wenruo } 251621a852b0SQu Wenruo if (!is_power_of_2(btrfs_super_stripesize(sb))) { 251721a852b0SQu Wenruo btrfs_err(fs_info, "invalid stripesize %u", 251821a852b0SQu Wenruo btrfs_super_stripesize(sb)); 251921a852b0SQu Wenruo ret = -EINVAL; 252021a852b0SQu Wenruo } 252121a852b0SQu Wenruo if (btrfs_super_num_devices(sb) > (1UL << 31)) 252221a852b0SQu Wenruo btrfs_warn(fs_info, "suspicious number of devices: %llu", 252321a852b0SQu Wenruo btrfs_super_num_devices(sb)); 252421a852b0SQu Wenruo if (btrfs_super_num_devices(sb) == 0) { 252521a852b0SQu Wenruo btrfs_err(fs_info, "number of devices is 0"); 252621a852b0SQu Wenruo ret = -EINVAL; 252721a852b0SQu Wenruo } 252821a852b0SQu Wenruo 2529069ec957SQu Wenruo if (mirror_num >= 0 && 2530069ec957SQu Wenruo btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { 253121a852b0SQu Wenruo btrfs_err(fs_info, "super offset mismatch %llu != %u", 253221a852b0SQu Wenruo btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 253321a852b0SQu Wenruo ret = -EINVAL; 253421a852b0SQu Wenruo } 253521a852b0SQu Wenruo 253621a852b0SQu Wenruo /* 253721a852b0SQu Wenruo * Obvious sys_chunk_array corruptions, it must hold at least one key 253821a852b0SQu Wenruo * and one chunk 253921a852b0SQu Wenruo */ 254021a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 254121a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too big %u > %u", 254221a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 254321a852b0SQu Wenruo BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 254421a852b0SQu Wenruo ret = -EINVAL; 254521a852b0SQu Wenruo } 254621a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 254721a852b0SQu Wenruo + sizeof(struct btrfs_chunk)) { 254821a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too small %u < %zu", 254921a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 255021a852b0SQu Wenruo sizeof(struct btrfs_disk_key) 255121a852b0SQu Wenruo + sizeof(struct btrfs_chunk)); 255221a852b0SQu Wenruo ret = -EINVAL; 255321a852b0SQu Wenruo } 255421a852b0SQu Wenruo 255521a852b0SQu Wenruo /* 255621a852b0SQu Wenruo * The generation is a global counter, we'll trust it more than the others 255721a852b0SQu Wenruo * but it's still possible that it's the one that's wrong. 255821a852b0SQu Wenruo */ 255921a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 256021a852b0SQu Wenruo btrfs_warn(fs_info, 256121a852b0SQu Wenruo "suspicious: generation < chunk_root_generation: %llu < %llu", 256221a852b0SQu Wenruo btrfs_super_generation(sb), 256321a852b0SQu Wenruo btrfs_super_chunk_root_generation(sb)); 256421a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 256521a852b0SQu Wenruo && btrfs_super_cache_generation(sb) != (u64)-1) 256621a852b0SQu Wenruo btrfs_warn(fs_info, 256721a852b0SQu Wenruo "suspicious: generation < cache_generation: %llu < %llu", 256821a852b0SQu Wenruo btrfs_super_generation(sb), 256921a852b0SQu Wenruo btrfs_super_cache_generation(sb)); 257021a852b0SQu Wenruo 257121a852b0SQu Wenruo return ret; 257221a852b0SQu Wenruo } 257321a852b0SQu Wenruo 2574069ec957SQu Wenruo /* 2575069ec957SQu Wenruo * Validation of super block at mount time. 2576069ec957SQu Wenruo * Some checks already done early at mount time, like csum type and incompat 2577069ec957SQu Wenruo * flags will be skipped. 2578069ec957SQu Wenruo */ 2579069ec957SQu Wenruo static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) 2580069ec957SQu Wenruo { 2581069ec957SQu Wenruo return validate_super(fs_info, fs_info->super_copy, 0); 2582069ec957SQu Wenruo } 2583069ec957SQu Wenruo 258475cb857dSQu Wenruo /* 258575cb857dSQu Wenruo * Validation of super block at write time. 258675cb857dSQu Wenruo * Some checks like bytenr check will be skipped as their values will be 258775cb857dSQu Wenruo * overwritten soon. 258875cb857dSQu Wenruo * Extra checks like csum type and incompat flags will be done here. 258975cb857dSQu Wenruo */ 259075cb857dSQu Wenruo static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, 259175cb857dSQu Wenruo struct btrfs_super_block *sb) 259275cb857dSQu Wenruo { 259375cb857dSQu Wenruo int ret; 259475cb857dSQu Wenruo 259575cb857dSQu Wenruo ret = validate_super(fs_info, sb, -1); 259675cb857dSQu Wenruo if (ret < 0) 259775cb857dSQu Wenruo goto out; 2598e7e16f48SJohannes Thumshirn if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) { 259975cb857dSQu Wenruo ret = -EUCLEAN; 260075cb857dSQu Wenruo btrfs_err(fs_info, "invalid csum type, has %u want %u", 260175cb857dSQu Wenruo btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); 260275cb857dSQu Wenruo goto out; 260375cb857dSQu Wenruo } 260475cb857dSQu Wenruo if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 260575cb857dSQu Wenruo ret = -EUCLEAN; 260675cb857dSQu Wenruo btrfs_err(fs_info, 260775cb857dSQu Wenruo "invalid incompat flags, has 0x%llx valid mask 0x%llx", 260875cb857dSQu Wenruo btrfs_super_incompat_flags(sb), 260975cb857dSQu Wenruo (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); 261075cb857dSQu Wenruo goto out; 261175cb857dSQu Wenruo } 261275cb857dSQu Wenruo out: 261375cb857dSQu Wenruo if (ret < 0) 261475cb857dSQu Wenruo btrfs_err(fs_info, 261575cb857dSQu Wenruo "super block corruption detected before writing it to disk"); 261675cb857dSQu Wenruo return ret; 261775cb857dSQu Wenruo } 261875cb857dSQu Wenruo 2619ad2b2c80SAl Viro int open_ctree(struct super_block *sb, 2620dfe25020SChris Mason struct btrfs_fs_devices *fs_devices, 2621dfe25020SChris Mason char *options) 2622eb60ceacSChris Mason { 2623db94535dSChris Mason u32 sectorsize; 2624db94535dSChris Mason u32 nodesize; 262587ee04ebSChris Mason u32 stripesize; 262684234f3aSYan Zheng u64 generation; 2627f2b636e8SJosef Bacik u64 features; 262851bce6c9SJohannes Thumshirn u16 csum_type; 26293de4586cSChris Mason struct btrfs_key location; 2630a061fc8dSChris Mason struct buffer_head *bh; 26314d34b278SIlya Dryomov struct btrfs_super_block *disk_super; 2632815745cfSAl Viro struct btrfs_fs_info *fs_info = btrfs_sb(sb); 2633f84a8bd6SAl Viro struct btrfs_root *tree_root; 26344d34b278SIlya Dryomov struct btrfs_root *chunk_root; 2635eb60ceacSChris Mason int ret; 2636e58ca020SYan int err = -EINVAL; 2637af31f5e5SChris Mason int num_backups_tried = 0; 2638af31f5e5SChris Mason int backup_index = 0; 26396675df31SOmar Sandoval int clear_free_space_tree = 0; 2640581c1760SQu Wenruo int level; 26414543df7eSChris Mason 264274e4d827SDavid Sterba tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL); 264374e4d827SDavid Sterba chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL); 2644cb517eabSMiao Xie if (!tree_root || !chunk_root) { 264539279cc3SChris Mason err = -ENOMEM; 264639279cc3SChris Mason goto fail; 264739279cc3SChris Mason } 264876dda93cSYan, Zheng 264976dda93cSYan, Zheng ret = init_srcu_struct(&fs_info->subvol_srcu); 265076dda93cSYan, Zheng if (ret) { 265176dda93cSYan, Zheng err = ret; 265276dda93cSYan, Zheng goto fail; 265376dda93cSYan, Zheng } 265476dda93cSYan, Zheng 26554297ff84SJosef Bacik ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL); 2656e2d84521SMiao Xie if (ret) { 2657e2d84521SMiao Xie err = ret; 26589e11ceeeSJan Kara goto fail_srcu; 2659e2d84521SMiao Xie } 26604297ff84SJosef Bacik 26614297ff84SJosef Bacik ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 26624297ff84SJosef Bacik if (ret) { 26634297ff84SJosef Bacik err = ret; 26644297ff84SJosef Bacik goto fail_dio_bytes; 26654297ff84SJosef Bacik } 266609cbfeafSKirill A. Shutemov fs_info->dirty_metadata_batch = PAGE_SIZE * 2667e2d84521SMiao Xie (1 + ilog2(nr_cpu_ids)); 2668e2d84521SMiao Xie 2669908c7f19STejun Heo ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 2670963d678bSMiao Xie if (ret) { 2671963d678bSMiao Xie err = ret; 2672963d678bSMiao Xie goto fail_dirty_metadata_bytes; 2673963d678bSMiao Xie } 2674963d678bSMiao Xie 26757f8d236aSDavid Sterba ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, 26767f8d236aSDavid Sterba GFP_KERNEL); 2677c404e0dcSMiao Xie if (ret) { 2678c404e0dcSMiao Xie err = ret; 2679c404e0dcSMiao Xie goto fail_delalloc_bytes; 2680c404e0dcSMiao Xie } 2681c404e0dcSMiao Xie 268276dda93cSYan, Zheng INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 2683f28491e0SJosef Bacik INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 26848fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 2685facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 268624bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 2687eb73c1b7SMiao Xie INIT_LIST_HEAD(&fs_info->delalloc_roots); 268811833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 268975cb379dSJeff Mahoney INIT_LIST_HEAD(&fs_info->pending_raid_kobjs); 269075cb379dSJeff Mahoney spin_lock_init(&fs_info->pending_raid_kobjs_lock); 2691eb73c1b7SMiao Xie spin_lock_init(&fs_info->delalloc_root_lock); 2692a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 269376dda93cSYan, Zheng spin_lock_init(&fs_info->fs_roots_radix_lock); 269424bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 26954cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 2696f29021b2SJan Schmidt spin_lock_init(&fs_info->tree_mod_seq_lock); 2697ceda0864SMiao Xie spin_lock_init(&fs_info->super_lock); 2698f28491e0SJosef Bacik spin_lock_init(&fs_info->buffer_lock); 269947ab2a6cSJosef Bacik spin_lock_init(&fs_info->unused_bgs_lock); 2700f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 2701d7c15171SZhao Lei mutex_init(&fs_info->unused_bg_unpin_mutex); 270267c5e7d4SFilipe Manana mutex_init(&fs_info->delete_unused_bgs_mutex); 27037585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 2704573bfb72SMiao Xie mutex_init(&fs_info->delalloc_root_mutex); 2705de98ced9SMiao Xie seqlock_init(&fs_info->profiles_lock); 270619c00ddcSChris Mason 27070b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 27086324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 2709f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 271047ab2a6cSJosef Bacik INIT_LIST_HEAD(&fs_info->unused_bgs); 2711c8bf1b67SDavid Sterba extent_map_tree_init(&fs_info->mapping_tree); 271266d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 271366d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 271466d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 271566d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 271666d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 271766d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 271866d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 2719ba2c4d4eSJosef Bacik btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, 2720ba2c4d4eSJosef Bacik BTRFS_BLOCK_RSV_DELREFS); 2721ba2c4d4eSJosef Bacik 2722771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 27234cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 27242fefd558SZhao Lei atomic_set(&fs_info->reada_works_cnt, 0); 2725034f784dSJosef Bacik atomic_set(&fs_info->nr_delayed_iputs, 0); 2726fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 2727e20d96d6SChris Mason fs_info->sb = sb; 272895ac567aSFilipe David Borba Manana fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 27299ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 27304cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 2731a5ed45f8SNikolay Borisov atomic64_set(&fs_info->free_chunk_space, 0); 2732f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 27338b87dc17SDavid Sterba fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 2734f8c269d7SDavid Sterba fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 273590519d66SArne Jansen /* readahead state */ 2736d0164adcSMel Gorman INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); 273790519d66SArne Jansen spin_lock_init(&fs_info->reada_lock); 2738fd708b81SJosef Bacik btrfs_init_ref_verify(fs_info); 2739c8b97818SChris Mason 2740b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 2741b34b086cSChris Mason num_online_cpus() + 2, 8); 27420afbaf8cSChris Mason 2743199c2a9cSMiao Xie INIT_LIST_HEAD(&fs_info->ordered_roots); 2744199c2a9cSMiao Xie spin_lock_init(&fs_info->ordered_root_lock); 274569fe2d75SJosef Bacik 274669fe2d75SJosef Bacik fs_info->btree_inode = new_inode(sb); 274769fe2d75SJosef Bacik if (!fs_info->btree_inode) { 274869fe2d75SJosef Bacik err = -ENOMEM; 274969fe2d75SJosef Bacik goto fail_bio_counter; 275069fe2d75SJosef Bacik } 275169fe2d75SJosef Bacik mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 275269fe2d75SJosef Bacik 275316cdcec7SMiao Xie fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 275474e4d827SDavid Sterba GFP_KERNEL); 275516cdcec7SMiao Xie if (!fs_info->delayed_root) { 275616cdcec7SMiao Xie err = -ENOMEM; 275716cdcec7SMiao Xie goto fail_iput; 275816cdcec7SMiao Xie } 275916cdcec7SMiao Xie btrfs_init_delayed_root(fs_info->delayed_root); 27603eaa2885SChris Mason 2761638aa7edSEric Sandeen btrfs_init_scrub(fs_info); 276221adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 276321adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 276421adbd5cSStefan Behrens #endif 2765779a65a4SEric Sandeen btrfs_init_balance(fs_info); 276621c7e756SMiao Xie btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work); 2767a2de733cSArne Jansen 27689f6d2510SDavid Sterba sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; 27699f6d2510SDavid Sterba sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); 2770a061fc8dSChris Mason 27716bccf3abSJeff Mahoney btrfs_init_btree_inode(fs_info); 277239279cc3SChris Mason 2773e02119d5SChris Mason spin_lock_init(&fs_info->block_group_cache_lock); 27746bef4d31SEric Paris fs_info->block_group_cache_tree = RB_ROOT; 2775a1897fddSLiu Bo fs_info->first_logical_byte = (u64)-1; 2776925baeddSChris Mason 277743eb5f29SQu Wenruo extent_io_tree_init(fs_info, &fs_info->freed_extents[0], 277843eb5f29SQu Wenruo IO_TREE_FS_INFO_FREED_EXTENTS0, NULL); 277943eb5f29SQu Wenruo extent_io_tree_init(fs_info, &fs_info->freed_extents[1], 278043eb5f29SQu Wenruo IO_TREE_FS_INFO_FREED_EXTENTS1, NULL); 278111833d66SYan Zheng fs_info->pinned_extents = &fs_info->freed_extents[0]; 2782afcdd129SJosef Bacik set_bit(BTRFS_FS_BARRIER, &fs_info->flags); 2783509659cdSChris Mason 27845a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 278579154b1bSChris Mason mutex_init(&fs_info->tree_log_mutex); 2786925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 2787a74a4b97SChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 2788a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 27891bbc621eSChris Mason mutex_init(&fs_info->ro_block_group_mutex); 27909e351cc8SJosef Bacik init_rwsem(&fs_info->commit_root_sem); 2791c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 279276dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 2793803b2f54SStefan Behrens sema_init(&fs_info->uuid_tree_rescan_sem, 1); 2794fa9c0d79SChris Mason 2795ad618368SEric Sandeen btrfs_init_dev_replace_locks(fs_info); 2796f9e92e40SEric Sandeen btrfs_init_qgroup(fs_info); 2797416ac51dSArne Jansen 2798fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2799fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2800fa9c0d79SChris Mason 2801e6dcd2dcSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 2802f9295749SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 2803bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 28044854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 2805034f784dSJosef Bacik init_waitqueue_head(&fs_info->delayed_iputs_wait); 28063768f368SChris Mason 2807da17066cSJeff Mahoney /* Usable values until the real ones are cached from the superblock */ 2808da17066cSJeff Mahoney fs_info->nodesize = 4096; 2809da17066cSJeff Mahoney fs_info->sectorsize = 4096; 2810da17066cSJeff Mahoney fs_info->stripesize = 4096; 2811da17066cSJeff Mahoney 2812eede2bf3SOmar Sandoval spin_lock_init(&fs_info->swapfile_pins_lock); 2813eede2bf3SOmar Sandoval fs_info->swapfile_pins = RB_ROOT; 2814eede2bf3SOmar Sandoval 281553b381b3SDavid Woodhouse ret = btrfs_alloc_stripe_hash_table(fs_info); 281653b381b3SDavid Woodhouse if (ret) { 281783c8266aSDavid Sterba err = ret; 281853b381b3SDavid Woodhouse goto fail_alloc; 281953b381b3SDavid Woodhouse } 282053b381b3SDavid Woodhouse 2821da17066cSJeff Mahoney __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID); 28227eccb903SChris Mason 28233c4bb26bSChris Mason invalidate_bdev(fs_devices->latest_bdev); 28241104a885SDavid Sterba 28251104a885SDavid Sterba /* 28261104a885SDavid Sterba * Read super block and check the signature bytes only 28271104a885SDavid Sterba */ 2828a512bbf8SYan Zheng bh = btrfs_read_dev_super(fs_devices->latest_bdev); 282992fc03fbSAnand Jain if (IS_ERR(bh)) { 283092fc03fbSAnand Jain err = PTR_ERR(bh); 283116cdcec7SMiao Xie goto fail_alloc; 283220b45077SDave Young } 283339279cc3SChris Mason 28341104a885SDavid Sterba /* 28358dc3f22cSJohannes Thumshirn * Verify the type first, if that or the the checksum value are 28368dc3f22cSJohannes Thumshirn * corrupted, we'll find out 28378dc3f22cSJohannes Thumshirn */ 283851bce6c9SJohannes Thumshirn csum_type = btrfs_super_csum_type((struct btrfs_super_block *)bh->b_data); 283951bce6c9SJohannes Thumshirn if (!btrfs_supported_super_csum(csum_type)) { 28408dc3f22cSJohannes Thumshirn btrfs_err(fs_info, "unsupported checksum algorithm: %u", 284151bce6c9SJohannes Thumshirn csum_type); 28428dc3f22cSJohannes Thumshirn err = -EINVAL; 28438dc3f22cSJohannes Thumshirn brelse(bh); 28448dc3f22cSJohannes Thumshirn goto fail_alloc; 28458dc3f22cSJohannes Thumshirn } 28468dc3f22cSJohannes Thumshirn 28476d97c6e3SJohannes Thumshirn ret = btrfs_init_csum_hash(fs_info, csum_type); 28486d97c6e3SJohannes Thumshirn if (ret) { 28496d97c6e3SJohannes Thumshirn err = ret; 28506d97c6e3SJohannes Thumshirn goto fail_alloc; 28516d97c6e3SJohannes Thumshirn } 28526d97c6e3SJohannes Thumshirn 28538dc3f22cSJohannes Thumshirn /* 28541104a885SDavid Sterba * We want to check superblock checksum, the type is stored inside. 28551104a885SDavid Sterba * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 28561104a885SDavid Sterba */ 2857ab8d0fc4SJeff Mahoney if (btrfs_check_super_csum(fs_info, bh->b_data)) { 285805135f59SDavid Sterba btrfs_err(fs_info, "superblock checksum mismatch"); 28591104a885SDavid Sterba err = -EINVAL; 2860b2acdddfSAnand Jain brelse(bh); 28616d97c6e3SJohannes Thumshirn goto fail_csum; 28621104a885SDavid Sterba } 28631104a885SDavid Sterba 28641104a885SDavid Sterba /* 28651104a885SDavid Sterba * super_copy is zeroed at allocation time and we never touch the 28661104a885SDavid Sterba * following bytes up to INFO_SIZE, the checksum is calculated from 28671104a885SDavid Sterba * the whole block of INFO_SIZE 28681104a885SDavid Sterba */ 28696c41761fSDavid Sterba memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy)); 2870a061fc8dSChris Mason brelse(bh); 28715f39d397SChris Mason 2872fbc6feaeSNikolay Borisov disk_super = fs_info->super_copy; 2873fbc6feaeSNikolay Borisov 2874de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, 2875de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 2876de37aa51SNikolay Borisov 28777239ff4bSNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID)) { 2878de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid, 2879de37aa51SNikolay Borisov fs_info->super_copy->metadata_uuid, 2880de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 28817239ff4bSNikolay Borisov } 28820b86a832SChris Mason 2883fbc6feaeSNikolay Borisov features = btrfs_super_flags(disk_super); 2884fbc6feaeSNikolay Borisov if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { 2885fbc6feaeSNikolay Borisov features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; 2886fbc6feaeSNikolay Borisov btrfs_set_super_flags(disk_super, features); 2887fbc6feaeSNikolay Borisov btrfs_info(fs_info, 2888fbc6feaeSNikolay Borisov "found metadata UUID change in progress flag, clearing"); 2889fbc6feaeSNikolay Borisov } 2890fbc6feaeSNikolay Borisov 2891fbc6feaeSNikolay Borisov memcpy(fs_info->super_for_commit, fs_info->super_copy, 2892fbc6feaeSNikolay Borisov sizeof(*fs_info->super_for_commit)); 2893de37aa51SNikolay Borisov 2894069ec957SQu Wenruo ret = btrfs_validate_mount_super(fs_info); 28951104a885SDavid Sterba if (ret) { 289605135f59SDavid Sterba btrfs_err(fs_info, "superblock contains fatal errors"); 28971104a885SDavid Sterba err = -EINVAL; 28986d97c6e3SJohannes Thumshirn goto fail_csum; 28991104a885SDavid Sterba } 29001104a885SDavid Sterba 29010f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 29026d97c6e3SJohannes Thumshirn goto fail_csum; 29030f7d52f4SChris Mason 2904acce952bSliubo /* check FS state, whether FS is broken. */ 290587533c47SMiao Xie if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 290687533c47SMiao Xie set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 2907acce952bSliubo 290875e7cb7fSLiu Bo /* 2909af31f5e5SChris Mason * run through our array of backup supers and setup 2910af31f5e5SChris Mason * our ring pointer to the oldest one 2911af31f5e5SChris Mason */ 2912af31f5e5SChris Mason generation = btrfs_super_generation(disk_super); 2913af31f5e5SChris Mason find_oldest_super_backup(fs_info, generation); 2914af31f5e5SChris Mason 2915af31f5e5SChris Mason /* 291675e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 291775e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 291875e7cb7fSLiu Bo */ 291975e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 292075e7cb7fSLiu Bo 29212ff7e61eSJeff Mahoney ret = btrfs_parse_options(fs_info, options, sb->s_flags); 29222b82032cSYan Zheng if (ret) { 29232b82032cSYan Zheng err = ret; 29246d97c6e3SJohannes Thumshirn goto fail_csum; 29252b82032cSYan Zheng } 2926dfe25020SChris Mason 2927f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 2928f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 2929f2b636e8SJosef Bacik if (features) { 293005135f59SDavid Sterba btrfs_err(fs_info, 293105135f59SDavid Sterba "cannot mount because of unsupported optional features (%llx)", 2932c1c9ff7cSGeert Uytterhoeven features); 2933f2b636e8SJosef Bacik err = -EINVAL; 29346d97c6e3SJohannes Thumshirn goto fail_csum; 2935f2b636e8SJosef Bacik } 2936f2b636e8SJosef Bacik 29375d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 29385d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 29390b246afaSJeff Mahoney if (fs_info->compress_type == BTRFS_COMPRESS_LZO) 2940a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 29415c1aab1dSNick Terrell else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) 29425c1aab1dSNick Terrell features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; 2943727011e0SChris Mason 29443173a18fSJosef Bacik if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) 294505135f59SDavid Sterba btrfs_info(fs_info, "has skinny extents"); 29463173a18fSJosef Bacik 2947727011e0SChris Mason /* 2948727011e0SChris Mason * flag our filesystem as having big metadata blocks if 2949727011e0SChris Mason * they are bigger than the page size 2950727011e0SChris Mason */ 295109cbfeafSKirill A. Shutemov if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) { 2952727011e0SChris Mason if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 295305135f59SDavid Sterba btrfs_info(fs_info, 295405135f59SDavid Sterba "flagging fs with big metadata feature"); 2955727011e0SChris Mason features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 2956727011e0SChris Mason } 2957727011e0SChris Mason 2958bc3f116fSChris Mason nodesize = btrfs_super_nodesize(disk_super); 2959bc3f116fSChris Mason sectorsize = btrfs_super_sectorsize(disk_super); 2960b7f67055SChandan Rajendra stripesize = sectorsize; 2961707e8a07SDavid Sterba fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 2962963d678bSMiao Xie fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 2963bc3f116fSChris Mason 2964da17066cSJeff Mahoney /* Cache block sizes */ 2965da17066cSJeff Mahoney fs_info->nodesize = nodesize; 2966da17066cSJeff Mahoney fs_info->sectorsize = sectorsize; 2967da17066cSJeff Mahoney fs_info->stripesize = stripesize; 2968da17066cSJeff Mahoney 2969bc3f116fSChris Mason /* 2970bc3f116fSChris Mason * mixed block groups end up with duplicate but slightly offset 2971bc3f116fSChris Mason * extent buffers for the same range. It leads to corruptions 2972bc3f116fSChris Mason */ 2973bc3f116fSChris Mason if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 2974707e8a07SDavid Sterba (sectorsize != nodesize)) { 297505135f59SDavid Sterba btrfs_err(fs_info, 297605135f59SDavid Sterba "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", 297705135f59SDavid Sterba nodesize, sectorsize); 29786d97c6e3SJohannes Thumshirn goto fail_csum; 2979bc3f116fSChris Mason } 2980bc3f116fSChris Mason 2981ceda0864SMiao Xie /* 2982ceda0864SMiao Xie * Needn't use the lock because there is no other task which will 2983ceda0864SMiao Xie * update the flag. 2984ceda0864SMiao Xie */ 29855d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 29865d4f98a2SYan Zheng 2987f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 2988f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 2989bc98a42cSDavid Howells if (!sb_rdonly(sb) && features) { 299005135f59SDavid Sterba btrfs_err(fs_info, 299105135f59SDavid Sterba "cannot mount read-write because of unsupported optional features (%llx)", 2992c1c9ff7cSGeert Uytterhoeven features); 2993f2b636e8SJosef Bacik err = -EINVAL; 29946d97c6e3SJohannes Thumshirn goto fail_csum; 2995f2b636e8SJosef Bacik } 299661d92c32SChris Mason 29972a458198SEric Sandeen ret = btrfs_init_workqueues(fs_info, fs_devices); 29982a458198SEric Sandeen if (ret) { 29992a458198SEric Sandeen err = ret; 30000dc3b84aSJosef Bacik goto fail_sb_buffer; 30010dc3b84aSJosef Bacik } 30024543df7eSChris Mason 30039e11ceeeSJan Kara sb->s_bdi->congested_fn = btrfs_congested_fn; 30049e11ceeeSJan Kara sb->s_bdi->congested_data = fs_info; 30059e11ceeeSJan Kara sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK; 3006b5420237SNikolay Borisov sb->s_bdi->ra_pages = VM_READAHEAD_PAGES; 30079e11ceeeSJan Kara sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); 30089e11ceeeSJan Kara sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); 30094575c9ccSChris Mason 3010a061fc8dSChris Mason sb->s_blocksize = sectorsize; 3011a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 3012de37aa51SNikolay Borisov memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); 3013db94535dSChris Mason 3014925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 30156bccf3abSJeff Mahoney ret = btrfs_read_sys_array(fs_info); 3016925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 301784eed90fSChris Mason if (ret) { 301805135f59SDavid Sterba btrfs_err(fs_info, "failed to read the system array: %d", ret); 30195d4f98a2SYan Zheng goto fail_sb_buffer; 302084eed90fSChris Mason } 30210b86a832SChris Mason 302284234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 3023581c1760SQu Wenruo level = btrfs_super_chunk_root_level(disk_super); 30240b86a832SChris Mason 3025da17066cSJeff Mahoney __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID); 30260b86a832SChris Mason 30272ff7e61eSJeff Mahoney chunk_root->node = read_tree_block(fs_info, 30280b86a832SChris Mason btrfs_super_chunk_root(disk_super), 3029581c1760SQu Wenruo generation, level, NULL); 303064c043deSLiu Bo if (IS_ERR(chunk_root->node) || 303164c043deSLiu Bo !extent_buffer_uptodate(chunk_root->node)) { 303205135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk root"); 3033e5fffbacSchandan if (!IS_ERR(chunk_root->node)) 3034e5fffbacSchandan free_extent_buffer(chunk_root->node); 303595ab1f64SZhao Lei chunk_root->node = NULL; 3036af31f5e5SChris Mason goto fail_tree_roots; 303783121942SDavid Woodhouse } 30385d4f98a2SYan Zheng btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 30395d4f98a2SYan Zheng chunk_root->commit_root = btrfs_root_node(chunk_root); 30400b86a832SChris Mason 3041e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 3042b308bc2fSGeert Uytterhoeven btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE); 3043e17cade2SChris Mason 30445b4aacefSJeff Mahoney ret = btrfs_read_chunk_tree(fs_info); 30452b82032cSYan Zheng if (ret) { 304605135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk tree: %d", ret); 3047af31f5e5SChris Mason goto fail_tree_roots; 30482b82032cSYan Zheng } 30490b86a832SChris Mason 30508dabb742SStefan Behrens /* 30519b99b115SAnand Jain * Keep the devid that is marked to be the target device for the 30529b99b115SAnand Jain * device replace procedure 30538dabb742SStefan Behrens */ 30549b99b115SAnand Jain btrfs_free_extra_devids(fs_devices, 0); 3055dfe25020SChris Mason 3056a6b0d5c8SChris Mason if (!fs_devices->latest_bdev) { 305705135f59SDavid Sterba btrfs_err(fs_info, "failed to read devices"); 3058a6b0d5c8SChris Mason goto fail_tree_roots; 3059a6b0d5c8SChris Mason } 3060a6b0d5c8SChris Mason 3061af31f5e5SChris Mason retry_root_backup: 306284234f3aSYan Zheng generation = btrfs_super_generation(disk_super); 3063581c1760SQu Wenruo level = btrfs_super_root_level(disk_super); 30640b86a832SChris Mason 30652ff7e61eSJeff Mahoney tree_root->node = read_tree_block(fs_info, 3066db94535dSChris Mason btrfs_super_root(disk_super), 3067581c1760SQu Wenruo generation, level, NULL); 306864c043deSLiu Bo if (IS_ERR(tree_root->node) || 306964c043deSLiu Bo !extent_buffer_uptodate(tree_root->node)) { 307005135f59SDavid Sterba btrfs_warn(fs_info, "failed to read tree root"); 3071e5fffbacSchandan if (!IS_ERR(tree_root->node)) 3072e5fffbacSchandan free_extent_buffer(tree_root->node); 307395ab1f64SZhao Lei tree_root->node = NULL; 3074af31f5e5SChris Mason goto recovery_tree_root; 307583121942SDavid Woodhouse } 3076af31f5e5SChris Mason 30775d4f98a2SYan Zheng btrfs_set_root_node(&tree_root->root_item, tree_root->node); 30785d4f98a2SYan Zheng tree_root->commit_root = btrfs_root_node(tree_root); 307969e9c6c6SStefan Behrens btrfs_set_root_refs(&tree_root->root_item, 1); 3080db94535dSChris Mason 3081f32e48e9SChandan Rajendra mutex_lock(&tree_root->objectid_mutex); 3082f32e48e9SChandan Rajendra ret = btrfs_find_highest_objectid(tree_root, 3083f32e48e9SChandan Rajendra &tree_root->highest_objectid); 3084f32e48e9SChandan Rajendra if (ret) { 3085f32e48e9SChandan Rajendra mutex_unlock(&tree_root->objectid_mutex); 3086f32e48e9SChandan Rajendra goto recovery_tree_root; 3087f32e48e9SChandan Rajendra } 3088f32e48e9SChandan Rajendra 3089f32e48e9SChandan Rajendra ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 3090f32e48e9SChandan Rajendra 3091f32e48e9SChandan Rajendra mutex_unlock(&tree_root->objectid_mutex); 3092f32e48e9SChandan Rajendra 30936bccf3abSJeff Mahoney ret = btrfs_read_roots(fs_info); 30944bbcaa64SEric Sandeen if (ret) 3095af31f5e5SChris Mason goto recovery_tree_root; 3096f7a81ea4SStefan Behrens 30978929ecfaSYan, Zheng fs_info->generation = generation; 30988929ecfaSYan, Zheng fs_info->last_trans_committed = generation; 30998929ecfaSYan, Zheng 3100cf90d884SQu Wenruo ret = btrfs_verify_dev_extents(fs_info); 3101cf90d884SQu Wenruo if (ret) { 3102cf90d884SQu Wenruo btrfs_err(fs_info, 3103cf90d884SQu Wenruo "failed to verify dev extents against chunks: %d", 3104cf90d884SQu Wenruo ret); 3105cf90d884SQu Wenruo goto fail_block_groups; 3106cf90d884SQu Wenruo } 310768310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 310868310a5eSIlya Dryomov if (ret) { 310905135f59SDavid Sterba btrfs_err(fs_info, "failed to recover balance: %d", ret); 311068310a5eSIlya Dryomov goto fail_block_groups; 311168310a5eSIlya Dryomov } 311268310a5eSIlya Dryomov 3113733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 3114733f4fbbSStefan Behrens if (ret) { 311505135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_stats: %d", ret); 3116733f4fbbSStefan Behrens goto fail_block_groups; 3117733f4fbbSStefan Behrens } 3118733f4fbbSStefan Behrens 31198dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 31208dabb742SStefan Behrens if (ret) { 312105135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_replace: %d", ret); 31228dabb742SStefan Behrens goto fail_block_groups; 31238dabb742SStefan Behrens } 31248dabb742SStefan Behrens 31259b99b115SAnand Jain btrfs_free_extra_devids(fs_devices, 1); 31268dabb742SStefan Behrens 3127b7c35e81SAnand Jain ret = btrfs_sysfs_add_fsid(fs_devices, NULL); 3128b7c35e81SAnand Jain if (ret) { 312905135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", 313005135f59SDavid Sterba ret); 3131b7c35e81SAnand Jain goto fail_block_groups; 3132b7c35e81SAnand Jain } 3133b7c35e81SAnand Jain 3134b7c35e81SAnand Jain ret = btrfs_sysfs_add_device(fs_devices); 3135b7c35e81SAnand Jain if (ret) { 313605135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs device interface: %d", 313705135f59SDavid Sterba ret); 3138b7c35e81SAnand Jain goto fail_fsdev_sysfs; 3139b7c35e81SAnand Jain } 3140b7c35e81SAnand Jain 314196f3136eSAnand Jain ret = btrfs_sysfs_add_mounted(fs_info); 31425ac1d209SJeff Mahoney if (ret) { 314305135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); 3144b7c35e81SAnand Jain goto fail_fsdev_sysfs; 31455ac1d209SJeff Mahoney } 31465ac1d209SJeff Mahoney 3147c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 3148c59021f8Sliubo if (ret) { 314905135f59SDavid Sterba btrfs_err(fs_info, "failed to initialize space info: %d", ret); 31502365dd3cSAnand Jain goto fail_sysfs; 3151c59021f8Sliubo } 3152c59021f8Sliubo 31535b4aacefSJeff Mahoney ret = btrfs_read_block_groups(fs_info); 31541b1d1f66SJosef Bacik if (ret) { 315505135f59SDavid Sterba btrfs_err(fs_info, "failed to read block groups: %d", ret); 31562365dd3cSAnand Jain goto fail_sysfs; 31571b1d1f66SJosef Bacik } 31584330e183SQu Wenruo 31596528b99dSAnand Jain if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) { 316005135f59SDavid Sterba btrfs_warn(fs_info, 316152042d8eSAndrea Gelmini "writable mount is not allowed due to too many missing devices"); 31622365dd3cSAnand Jain goto fail_sysfs; 3163292fd7fcSStefan Behrens } 31649078a3e1SChris Mason 3165a74a4b97SChris Mason fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 3166a74a4b97SChris Mason "btrfs-cleaner"); 316757506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 31682365dd3cSAnand Jain goto fail_sysfs; 3169a74a4b97SChris Mason 3170a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 3171a74a4b97SChris Mason tree_root, 3172a74a4b97SChris Mason "btrfs-transaction"); 317357506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 31743f157a2fSChris Mason goto fail_cleaner; 3175a74a4b97SChris Mason 3176583b7231SHans van Kranenburg if (!btrfs_test_opt(fs_info, NOSSD) && 3177c289811cSChris Mason !fs_info->fs_devices->rotating) { 3178583b7231SHans van Kranenburg btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); 3179c289811cSChris Mason } 3180c289811cSChris Mason 3181572d9ab7SDavid Sterba /* 318201327610SNicholas D Steeves * Mount does not set all options immediately, we can do it now and do 3183572d9ab7SDavid Sterba * not have to wait for transaction commit 3184572d9ab7SDavid Sterba */ 3185572d9ab7SDavid Sterba btrfs_apply_pending_changes(fs_info); 31863818aea2SQu Wenruo 318721adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 31880b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { 31892ff7e61eSJeff Mahoney ret = btrfsic_mount(fs_info, fs_devices, 31900b246afaSJeff Mahoney btrfs_test_opt(fs_info, 319121adbd5cSStefan Behrens CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? 319221adbd5cSStefan Behrens 1 : 0, 319321adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 319421adbd5cSStefan Behrens if (ret) 319505135f59SDavid Sterba btrfs_warn(fs_info, 319605135f59SDavid Sterba "failed to initialize integrity check module: %d", 319705135f59SDavid Sterba ret); 319821adbd5cSStefan Behrens } 319921adbd5cSStefan Behrens #endif 3200bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 3201bcef60f2SArne Jansen if (ret) 3202bcef60f2SArne Jansen goto fail_trans_kthread; 320321adbd5cSStefan Behrens 3204fd708b81SJosef Bacik if (btrfs_build_ref_tree(fs_info)) 3205fd708b81SJosef Bacik btrfs_err(fs_info, "couldn't build ref tree"); 3206fd708b81SJosef Bacik 320796da0919SQu Wenruo /* do not make disk changes in broken FS or nologreplay is given */ 320896da0919SQu Wenruo if (btrfs_super_log_root(disk_super) != 0 && 32090b246afaSJeff Mahoney !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 321063443bf5SEric Sandeen ret = btrfs_replay_log(fs_info, fs_devices); 321179787eaaSJeff Mahoney if (ret) { 321263443bf5SEric Sandeen err = ret; 321328c16cbbSWang Shilong goto fail_qgroup; 3214e556ce2cSYan Zheng } 3215e02119d5SChris Mason } 32161a40e23bSZheng Yan 32176bccf3abSJeff Mahoney ret = btrfs_find_orphan_roots(fs_info); 321879787eaaSJeff Mahoney if (ret) 321928c16cbbSWang Shilong goto fail_qgroup; 322076dda93cSYan, Zheng 3221bc98a42cSDavid Howells if (!sb_rdonly(sb)) { 3222d68fc57bSYan, Zheng ret = btrfs_cleanup_fs_roots(fs_info); 322344c44af2SIlya Dryomov if (ret) 322428c16cbbSWang Shilong goto fail_qgroup; 322590c711abSZygo Blaxell 322690c711abSZygo Blaxell mutex_lock(&fs_info->cleaner_mutex); 32275d4f98a2SYan Zheng ret = btrfs_recover_relocation(tree_root); 322890c711abSZygo Blaxell mutex_unlock(&fs_info->cleaner_mutex); 3229d7ce5843SMiao Xie if (ret < 0) { 323005135f59SDavid Sterba btrfs_warn(fs_info, "failed to recover relocation: %d", 323105135f59SDavid Sterba ret); 3232d7ce5843SMiao Xie err = -EINVAL; 3233bcef60f2SArne Jansen goto fail_qgroup; 3234d7ce5843SMiao Xie } 32357c2ca468SChris Mason } 32361a40e23bSZheng Yan 32373de4586cSChris Mason location.objectid = BTRFS_FS_TREE_OBJECTID; 32383de4586cSChris Mason location.type = BTRFS_ROOT_ITEM_KEY; 3239cb517eabSMiao Xie location.offset = 0; 32403de4586cSChris Mason 32413de4586cSChris Mason fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location); 32423140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 32433140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 3244f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read fs tree: %d", err); 3245bcef60f2SArne Jansen goto fail_qgroup; 32463140c9a3SDan Carpenter } 3247c289811cSChris Mason 3248bc98a42cSDavid Howells if (sb_rdonly(sb)) 32492b6ba629SIlya Dryomov return 0; 32502b6ba629SIlya Dryomov 3251f8d468a1SOmar Sandoval if (btrfs_test_opt(fs_info, CLEAR_CACHE) && 3252f8d468a1SOmar Sandoval btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 32536675df31SOmar Sandoval clear_free_space_tree = 1; 32546675df31SOmar Sandoval } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 32556675df31SOmar Sandoval !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { 32566675df31SOmar Sandoval btrfs_warn(fs_info, "free space tree is invalid"); 32576675df31SOmar Sandoval clear_free_space_tree = 1; 32586675df31SOmar Sandoval } 32596675df31SOmar Sandoval 32606675df31SOmar Sandoval if (clear_free_space_tree) { 3261f8d468a1SOmar Sandoval btrfs_info(fs_info, "clearing free space tree"); 3262f8d468a1SOmar Sandoval ret = btrfs_clear_free_space_tree(fs_info); 3263f8d468a1SOmar Sandoval if (ret) { 3264f8d468a1SOmar Sandoval btrfs_warn(fs_info, 3265f8d468a1SOmar Sandoval "failed to clear free space tree: %d", ret); 32666bccf3abSJeff Mahoney close_ctree(fs_info); 3267f8d468a1SOmar Sandoval return ret; 3268f8d468a1SOmar Sandoval } 3269f8d468a1SOmar Sandoval } 3270f8d468a1SOmar Sandoval 32710b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && 3272511711afSChris Mason !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 327305135f59SDavid Sterba btrfs_info(fs_info, "creating free space tree"); 3274511711afSChris Mason ret = btrfs_create_free_space_tree(fs_info); 3275511711afSChris Mason if (ret) { 327605135f59SDavid Sterba btrfs_warn(fs_info, 327705135f59SDavid Sterba "failed to create free space tree: %d", ret); 32786bccf3abSJeff Mahoney close_ctree(fs_info); 3279511711afSChris Mason return ret; 3280511711afSChris Mason } 3281511711afSChris Mason } 3282511711afSChris Mason 3283e3acc2a6SJosef Bacik down_read(&fs_info->cleanup_work_sem); 32842b6ba629SIlya Dryomov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 32852b6ba629SIlya Dryomov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 32862b6ba629SIlya Dryomov up_read(&fs_info->cleanup_work_sem); 32876bccf3abSJeff Mahoney close_ctree(fs_info); 32882b6ba629SIlya Dryomov return ret; 32892b6ba629SIlya Dryomov } 3290e3acc2a6SJosef Bacik up_read(&fs_info->cleanup_work_sem); 329159641015SIlya Dryomov 32922b6ba629SIlya Dryomov ret = btrfs_resume_balance_async(fs_info); 32932b6ba629SIlya Dryomov if (ret) { 329405135f59SDavid Sterba btrfs_warn(fs_info, "failed to resume balance: %d", ret); 32956bccf3abSJeff Mahoney close_ctree(fs_info); 32962b6ba629SIlya Dryomov return ret; 3297e3acc2a6SJosef Bacik } 3298e3acc2a6SJosef Bacik 32998dabb742SStefan Behrens ret = btrfs_resume_dev_replace_async(fs_info); 33008dabb742SStefan Behrens if (ret) { 330105135f59SDavid Sterba btrfs_warn(fs_info, "failed to resume device replace: %d", ret); 33026bccf3abSJeff Mahoney close_ctree(fs_info); 33038dabb742SStefan Behrens return ret; 33048dabb742SStefan Behrens } 33058dabb742SStefan Behrens 3306b382a324SJan Schmidt btrfs_qgroup_rescan_resume(fs_info); 3307b382a324SJan Schmidt 33084bbcaa64SEric Sandeen if (!fs_info->uuid_root) { 330905135f59SDavid Sterba btrfs_info(fs_info, "creating UUID tree"); 3310f7a81ea4SStefan Behrens ret = btrfs_create_uuid_tree(fs_info); 3311f7a81ea4SStefan Behrens if (ret) { 331205135f59SDavid Sterba btrfs_warn(fs_info, 331305135f59SDavid Sterba "failed to create the UUID tree: %d", ret); 33146bccf3abSJeff Mahoney close_ctree(fs_info); 3315f7a81ea4SStefan Behrens return ret; 3316f7a81ea4SStefan Behrens } 33170b246afaSJeff Mahoney } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || 33184bbcaa64SEric Sandeen fs_info->generation != 33194bbcaa64SEric Sandeen btrfs_super_uuid_tree_generation(disk_super)) { 332005135f59SDavid Sterba btrfs_info(fs_info, "checking UUID tree"); 332170f80175SStefan Behrens ret = btrfs_check_uuid_tree(fs_info); 332270f80175SStefan Behrens if (ret) { 332305135f59SDavid Sterba btrfs_warn(fs_info, 332405135f59SDavid Sterba "failed to check the UUID tree: %d", ret); 33256bccf3abSJeff Mahoney close_ctree(fs_info); 332670f80175SStefan Behrens return ret; 332770f80175SStefan Behrens } 332870f80175SStefan Behrens } else { 3329afcdd129SJosef Bacik set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 3330f7a81ea4SStefan Behrens } 3331afcdd129SJosef Bacik set_bit(BTRFS_FS_OPEN, &fs_info->flags); 333247ab2a6cSJosef Bacik 33338dcddfa0SQu Wenruo /* 33348dcddfa0SQu Wenruo * backuproot only affect mount behavior, and if open_ctree succeeded, 33358dcddfa0SQu Wenruo * no need to keep the flag 33368dcddfa0SQu Wenruo */ 33378dcddfa0SQu Wenruo btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT); 33388dcddfa0SQu Wenruo 3339ad2b2c80SAl Viro return 0; 334039279cc3SChris Mason 3341bcef60f2SArne Jansen fail_qgroup: 3342bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 33437c2ca468SChris Mason fail_trans_kthread: 33447c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 33452ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 3346faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 33473f157a2fSChris Mason fail_cleaner: 3348a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 33497c2ca468SChris Mason 33507c2ca468SChris Mason /* 33517c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 33527c2ca468SChris Mason * kthreads 33537c2ca468SChris Mason */ 33547c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 33557c2ca468SChris Mason 33562365dd3cSAnand Jain fail_sysfs: 33576618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 33582365dd3cSAnand Jain 3359b7c35e81SAnand Jain fail_fsdev_sysfs: 3360b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3361b7c35e81SAnand Jain 33621b1d1f66SJosef Bacik fail_block_groups: 336354067ae9SJosef Bacik btrfs_put_block_group_cache(fs_info); 3364af31f5e5SChris Mason 3365af31f5e5SChris Mason fail_tree_roots: 3366af31f5e5SChris Mason free_root_pointers(fs_info, 1); 33672b8195bbSMiao Xie invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3368af31f5e5SChris Mason 336939279cc3SChris Mason fail_sb_buffer: 33707abadb64SLiu Bo btrfs_stop_all_workers(fs_info); 33715cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 33726d97c6e3SJohannes Thumshirn fail_csum: 33736d97c6e3SJohannes Thumshirn btrfs_free_csum_hash(fs_info); 337416cdcec7SMiao Xie fail_alloc: 33754543df7eSChris Mason fail_iput: 3376586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 3377586e46e2SIlya Dryomov 33784543df7eSChris Mason iput(fs_info->btree_inode); 3379c404e0dcSMiao Xie fail_bio_counter: 33807f8d236aSDavid Sterba percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 3381963d678bSMiao Xie fail_delalloc_bytes: 3382963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 3383e2d84521SMiao Xie fail_dirty_metadata_bytes: 3384e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 33854297ff84SJosef Bacik fail_dio_bytes: 33864297ff84SJosef Bacik percpu_counter_destroy(&fs_info->dio_bytes); 338776dda93cSYan, Zheng fail_srcu: 338876dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 33897e662854SQinghuang Feng fail: 339053b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 3391586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 3392ad2b2c80SAl Viro return err; 3393af31f5e5SChris Mason 3394af31f5e5SChris Mason recovery_tree_root: 33950b246afaSJeff Mahoney if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) 3396af31f5e5SChris Mason goto fail_tree_roots; 3397af31f5e5SChris Mason 3398af31f5e5SChris Mason free_root_pointers(fs_info, 0); 3399af31f5e5SChris Mason 3400af31f5e5SChris Mason /* don't use the log in recovery mode, it won't be valid */ 3401af31f5e5SChris Mason btrfs_set_super_log_root(disk_super, 0); 3402af31f5e5SChris Mason 3403af31f5e5SChris Mason /* we can't trust the free space cache either */ 3404af31f5e5SChris Mason btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 3405af31f5e5SChris Mason 3406af31f5e5SChris Mason ret = next_root_backup(fs_info, fs_info->super_copy, 3407af31f5e5SChris Mason &num_backups_tried, &backup_index); 3408af31f5e5SChris Mason if (ret == -1) 3409af31f5e5SChris Mason goto fail_block_groups; 3410af31f5e5SChris Mason goto retry_root_backup; 3411eb60ceacSChris Mason } 3412663faf9fSMasami Hiramatsu ALLOW_ERROR_INJECTION(open_ctree, ERRNO); 3413eb60ceacSChris Mason 3414f2984462SChris Mason static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate) 3415f2984462SChris Mason { 3416f2984462SChris Mason if (uptodate) { 3417f2984462SChris Mason set_buffer_uptodate(bh); 3418f2984462SChris Mason } else { 3419442a4f63SStefan Behrens struct btrfs_device *device = (struct btrfs_device *) 3420442a4f63SStefan Behrens bh->b_private; 3421442a4f63SStefan Behrens 3422fb456252SJeff Mahoney btrfs_warn_rl_in_rcu(device->fs_info, 3423b14af3b4SDavid Sterba "lost page write due to IO error on %s", 3424606686eeSJosef Bacik rcu_str_deref(device->name)); 342501327610SNicholas D Steeves /* note, we don't set_buffer_write_io_error because we have 34261259ab75SChris Mason * our own ways of dealing with the IO errors 34271259ab75SChris Mason */ 3428f2984462SChris Mason clear_buffer_uptodate(bh); 3429442a4f63SStefan Behrens btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS); 3430f2984462SChris Mason } 3431f2984462SChris Mason unlock_buffer(bh); 3432f2984462SChris Mason put_bh(bh); 3433f2984462SChris Mason } 3434f2984462SChris Mason 343529c36d72SAnand Jain int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num, 343629c36d72SAnand Jain struct buffer_head **bh_ret) 343729c36d72SAnand Jain { 343829c36d72SAnand Jain struct buffer_head *bh; 343929c36d72SAnand Jain struct btrfs_super_block *super; 344029c36d72SAnand Jain u64 bytenr; 344129c36d72SAnand Jain 344229c36d72SAnand Jain bytenr = btrfs_sb_offset(copy_num); 344329c36d72SAnand Jain if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode)) 344429c36d72SAnand Jain return -EINVAL; 344529c36d72SAnand Jain 34469f6d2510SDavid Sterba bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE); 344729c36d72SAnand Jain /* 344829c36d72SAnand Jain * If we fail to read from the underlying devices, as of now 344929c36d72SAnand Jain * the best option we have is to mark it EIO. 345029c36d72SAnand Jain */ 345129c36d72SAnand Jain if (!bh) 345229c36d72SAnand Jain return -EIO; 345329c36d72SAnand Jain 345429c36d72SAnand Jain super = (struct btrfs_super_block *)bh->b_data; 345529c36d72SAnand Jain if (btrfs_super_bytenr(super) != bytenr || 345629c36d72SAnand Jain btrfs_super_magic(super) != BTRFS_MAGIC) { 345729c36d72SAnand Jain brelse(bh); 345829c36d72SAnand Jain return -EINVAL; 345929c36d72SAnand Jain } 346029c36d72SAnand Jain 346129c36d72SAnand Jain *bh_ret = bh; 346229c36d72SAnand Jain return 0; 346329c36d72SAnand Jain } 346429c36d72SAnand Jain 346529c36d72SAnand Jain 3466a512bbf8SYan Zheng struct buffer_head *btrfs_read_dev_super(struct block_device *bdev) 3467a512bbf8SYan Zheng { 3468a512bbf8SYan Zheng struct buffer_head *bh; 3469a512bbf8SYan Zheng struct buffer_head *latest = NULL; 3470a512bbf8SYan Zheng struct btrfs_super_block *super; 3471a512bbf8SYan Zheng int i; 3472a512bbf8SYan Zheng u64 transid = 0; 347392fc03fbSAnand Jain int ret = -EINVAL; 3474a512bbf8SYan Zheng 3475a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 3476a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 3477a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 3478a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3479a512bbf8SYan Zheng */ 3480a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 348129c36d72SAnand Jain ret = btrfs_read_dev_one_super(bdev, i, &bh); 348229c36d72SAnand Jain if (ret) 3483a512bbf8SYan Zheng continue; 3484a512bbf8SYan Zheng 3485a512bbf8SYan Zheng super = (struct btrfs_super_block *)bh->b_data; 3486a512bbf8SYan Zheng 3487a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 3488a512bbf8SYan Zheng brelse(latest); 3489a512bbf8SYan Zheng latest = bh; 3490a512bbf8SYan Zheng transid = btrfs_super_generation(super); 3491a512bbf8SYan Zheng } else { 3492a512bbf8SYan Zheng brelse(bh); 3493a512bbf8SYan Zheng } 3494a512bbf8SYan Zheng } 349592fc03fbSAnand Jain 349692fc03fbSAnand Jain if (!latest) 349792fc03fbSAnand Jain return ERR_PTR(ret); 349892fc03fbSAnand Jain 3499a512bbf8SYan Zheng return latest; 3500a512bbf8SYan Zheng } 3501a512bbf8SYan Zheng 35024eedeb75SHisashi Hifumi /* 3503abbb3b8eSDavid Sterba * Write superblock @sb to the @device. Do not wait for completion, all the 3504abbb3b8eSDavid Sterba * buffer heads we write are pinned. 35054eedeb75SHisashi Hifumi * 3506abbb3b8eSDavid Sterba * Write @max_mirrors copies of the superblock, where 0 means default that fit 3507abbb3b8eSDavid Sterba * the expected device size at commit time. Note that max_mirrors must be 3508abbb3b8eSDavid Sterba * same for write and wait phases. 35094eedeb75SHisashi Hifumi * 3510abbb3b8eSDavid Sterba * Return number of errors when buffer head is not found or submission fails. 35114eedeb75SHisashi Hifumi */ 3512a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 3513abbb3b8eSDavid Sterba struct btrfs_super_block *sb, int max_mirrors) 3514a512bbf8SYan Zheng { 3515a512bbf8SYan Zheng struct buffer_head *bh; 3516a512bbf8SYan Zheng int i; 3517a512bbf8SYan Zheng int ret; 3518a512bbf8SYan Zheng int errors = 0; 3519a512bbf8SYan Zheng u32 crc; 3520a512bbf8SYan Zheng u64 bytenr; 35211b9e619cSOmar Sandoval int op_flags; 3522a512bbf8SYan Zheng 3523a512bbf8SYan Zheng if (max_mirrors == 0) 3524a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3525a512bbf8SYan Zheng 3526a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 3527a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 3528935e5cc9SMiao Xie if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3529935e5cc9SMiao Xie device->commit_total_bytes) 3530a512bbf8SYan Zheng break; 3531a512bbf8SYan Zheng 3532a512bbf8SYan Zheng btrfs_set_super_bytenr(sb, bytenr); 3533a512bbf8SYan Zheng 3534a512bbf8SYan Zheng crc = ~(u32)0; 3535abbb3b8eSDavid Sterba crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc, 3536abbb3b8eSDavid Sterba BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE); 3537a512bbf8SYan Zheng btrfs_csum_final(crc, sb->csum); 3538a512bbf8SYan Zheng 3539abbb3b8eSDavid Sterba /* One reference for us, and we leave it for the caller */ 35409f6d2510SDavid Sterba bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, 3541a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 3542634554dcSJosef Bacik if (!bh) { 3543fb456252SJeff Mahoney btrfs_err(device->fs_info, 3544f14d104dSDavid Sterba "couldn't get super buffer head for bytenr %llu", 3545f14d104dSDavid Sterba bytenr); 3546634554dcSJosef Bacik errors++; 3547634554dcSJosef Bacik continue; 3548634554dcSJosef Bacik } 3549634554dcSJosef Bacik 3550a512bbf8SYan Zheng memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE); 3551a512bbf8SYan Zheng 35524eedeb75SHisashi Hifumi /* one reference for submit_bh */ 3553a512bbf8SYan Zheng get_bh(bh); 35544eedeb75SHisashi Hifumi 35554eedeb75SHisashi Hifumi set_buffer_uptodate(bh); 3556a512bbf8SYan Zheng lock_buffer(bh); 3557a512bbf8SYan Zheng bh->b_end_io = btrfs_end_buffer_write_sync; 3558442a4f63SStefan Behrens bh->b_private = device; 3559a512bbf8SYan Zheng 3560387125fcSChris Mason /* 3561387125fcSChris Mason * we fua the first super. The others we allow 3562387125fcSChris Mason * to go down lazy. 3563387125fcSChris Mason */ 35641b9e619cSOmar Sandoval op_flags = REQ_SYNC | REQ_META | REQ_PRIO; 35651b9e619cSOmar Sandoval if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) 35661b9e619cSOmar Sandoval op_flags |= REQ_FUA; 35671b9e619cSOmar Sandoval ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh); 35684eedeb75SHisashi Hifumi if (ret) 3569a512bbf8SYan Zheng errors++; 3570a512bbf8SYan Zheng } 3571a512bbf8SYan Zheng return errors < i ? 0 : -1; 3572a512bbf8SYan Zheng } 3573a512bbf8SYan Zheng 3574387125fcSChris Mason /* 3575abbb3b8eSDavid Sterba * Wait for write completion of superblocks done by write_dev_supers, 3576abbb3b8eSDavid Sterba * @max_mirrors same for write and wait phases. 3577abbb3b8eSDavid Sterba * 3578abbb3b8eSDavid Sterba * Return number of errors when buffer head is not found or not marked up to 3579abbb3b8eSDavid Sterba * date. 3580abbb3b8eSDavid Sterba */ 3581abbb3b8eSDavid Sterba static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) 3582abbb3b8eSDavid Sterba { 3583abbb3b8eSDavid Sterba struct buffer_head *bh; 3584abbb3b8eSDavid Sterba int i; 3585abbb3b8eSDavid Sterba int errors = 0; 3586b6a535faSHoward McLauchlan bool primary_failed = false; 3587abbb3b8eSDavid Sterba u64 bytenr; 3588abbb3b8eSDavid Sterba 3589abbb3b8eSDavid Sterba if (max_mirrors == 0) 3590abbb3b8eSDavid Sterba max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3591abbb3b8eSDavid Sterba 3592abbb3b8eSDavid Sterba for (i = 0; i < max_mirrors; i++) { 3593abbb3b8eSDavid Sterba bytenr = btrfs_sb_offset(i); 3594abbb3b8eSDavid Sterba if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3595abbb3b8eSDavid Sterba device->commit_total_bytes) 3596abbb3b8eSDavid Sterba break; 3597abbb3b8eSDavid Sterba 35989f6d2510SDavid Sterba bh = __find_get_block(device->bdev, 35999f6d2510SDavid Sterba bytenr / BTRFS_BDEV_BLOCKSIZE, 3600abbb3b8eSDavid Sterba BTRFS_SUPER_INFO_SIZE); 3601abbb3b8eSDavid Sterba if (!bh) { 3602abbb3b8eSDavid Sterba errors++; 3603b6a535faSHoward McLauchlan if (i == 0) 3604b6a535faSHoward McLauchlan primary_failed = true; 3605abbb3b8eSDavid Sterba continue; 3606abbb3b8eSDavid Sterba } 3607abbb3b8eSDavid Sterba wait_on_buffer(bh); 3608b6a535faSHoward McLauchlan if (!buffer_uptodate(bh)) { 3609abbb3b8eSDavid Sterba errors++; 3610b6a535faSHoward McLauchlan if (i == 0) 3611b6a535faSHoward McLauchlan primary_failed = true; 3612b6a535faSHoward McLauchlan } 3613abbb3b8eSDavid Sterba 3614abbb3b8eSDavid Sterba /* drop our reference */ 3615abbb3b8eSDavid Sterba brelse(bh); 3616abbb3b8eSDavid Sterba 3617abbb3b8eSDavid Sterba /* drop the reference from the writing run */ 3618abbb3b8eSDavid Sterba brelse(bh); 3619abbb3b8eSDavid Sterba } 3620abbb3b8eSDavid Sterba 3621b6a535faSHoward McLauchlan /* log error, force error return */ 3622b6a535faSHoward McLauchlan if (primary_failed) { 3623b6a535faSHoward McLauchlan btrfs_err(device->fs_info, "error writing primary super block to device %llu", 3624b6a535faSHoward McLauchlan device->devid); 3625b6a535faSHoward McLauchlan return -1; 3626b6a535faSHoward McLauchlan } 3627b6a535faSHoward McLauchlan 3628abbb3b8eSDavid Sterba return errors < i ? 0 : -1; 3629abbb3b8eSDavid Sterba } 3630abbb3b8eSDavid Sterba 3631abbb3b8eSDavid Sterba /* 3632387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 3633387125fcSChris Mason * for the barrier when it is done 3634387125fcSChris Mason */ 36354246a0b6SChristoph Hellwig static void btrfs_end_empty_barrier(struct bio *bio) 3636387125fcSChris Mason { 3637387125fcSChris Mason complete(bio->bi_private); 3638387125fcSChris Mason } 3639387125fcSChris Mason 3640387125fcSChris Mason /* 36414fc6441aSAnand Jain * Submit a flush request to the device if it supports it. Error handling is 36424fc6441aSAnand Jain * done in the waiting counterpart. 3643387125fcSChris Mason */ 36444fc6441aSAnand Jain static void write_dev_flush(struct btrfs_device *device) 3645387125fcSChris Mason { 3646c2a9c7abSAnand Jain struct request_queue *q = bdev_get_queue(device->bdev); 3647e0ae9994SDavid Sterba struct bio *bio = device->flush_bio; 3648387125fcSChris Mason 3649c2a9c7abSAnand Jain if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) 36504fc6441aSAnand Jain return; 3651387125fcSChris Mason 3652e0ae9994SDavid Sterba bio_reset(bio); 3653387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 365474d46992SChristoph Hellwig bio_set_dev(bio, device->bdev); 36558d910125SJan Kara bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH; 3656387125fcSChris Mason init_completion(&device->flush_wait); 3657387125fcSChris Mason bio->bi_private = &device->flush_wait; 3658387125fcSChris Mason 365943a01111SLu Fengqi btrfsic_submit_bio(bio); 36601c3063b6SAnand Jain set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 36614fc6441aSAnand Jain } 3662387125fcSChris Mason 36634fc6441aSAnand Jain /* 36644fc6441aSAnand Jain * If the flush bio has been submitted by write_dev_flush, wait for it. 36654fc6441aSAnand Jain */ 36668c27cb35SLinus Torvalds static blk_status_t wait_dev_flush(struct btrfs_device *device) 36674fc6441aSAnand Jain { 36684fc6441aSAnand Jain struct bio *bio = device->flush_bio; 36694fc6441aSAnand Jain 36701c3063b6SAnand Jain if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) 367158efbc9fSOmar Sandoval return BLK_STS_OK; 36724fc6441aSAnand Jain 36731c3063b6SAnand Jain clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 36742980d574SDavid Sterba wait_for_completion_io(&device->flush_wait); 36754fc6441aSAnand Jain 36768c27cb35SLinus Torvalds return bio->bi_status; 3677387125fcSChris Mason } 3678387125fcSChris Mason 3679d10b82feSQu Wenruo static int check_barrier_error(struct btrfs_fs_info *fs_info) 3680401b41e5SAnand Jain { 36816528b99dSAnand Jain if (!btrfs_check_rw_degradable(fs_info, NULL)) 3682401b41e5SAnand Jain return -EIO; 3683387125fcSChris Mason return 0; 3684387125fcSChris Mason } 3685387125fcSChris Mason 3686387125fcSChris Mason /* 3687387125fcSChris Mason * send an empty flush down to each device in parallel, 3688387125fcSChris Mason * then wait for them 3689387125fcSChris Mason */ 3690387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 3691387125fcSChris Mason { 3692387125fcSChris Mason struct list_head *head; 3693387125fcSChris Mason struct btrfs_device *dev; 36945af3e8ccSStefan Behrens int errors_wait = 0; 36954e4cbee9SChristoph Hellwig blk_status_t ret; 3696387125fcSChris Mason 36971538e6c5SDavid Sterba lockdep_assert_held(&info->fs_devices->device_list_mutex); 3698387125fcSChris Mason /* send down all the barriers */ 3699387125fcSChris Mason head = &info->fs_devices->devices; 37001538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3701e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3702f88ba6a2SHidetoshi Seto continue; 3703cea7c8bfSAnand Jain if (!dev->bdev) 3704387125fcSChris Mason continue; 3705e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3706ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3707387125fcSChris Mason continue; 3708387125fcSChris Mason 37094fc6441aSAnand Jain write_dev_flush(dev); 371058efbc9fSOmar Sandoval dev->last_flush_error = BLK_STS_OK; 3711387125fcSChris Mason } 3712387125fcSChris Mason 3713387125fcSChris Mason /* wait for all the barriers */ 37141538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3715e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3716f88ba6a2SHidetoshi Seto continue; 3717387125fcSChris Mason if (!dev->bdev) { 37185af3e8ccSStefan Behrens errors_wait++; 3719387125fcSChris Mason continue; 3720387125fcSChris Mason } 3721e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3722ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3723387125fcSChris Mason continue; 3724387125fcSChris Mason 37254fc6441aSAnand Jain ret = wait_dev_flush(dev); 3726401b41e5SAnand Jain if (ret) { 3727401b41e5SAnand Jain dev->last_flush_error = ret; 372866b4993eSDavid Sterba btrfs_dev_stat_inc_and_print(dev, 372966b4993eSDavid Sterba BTRFS_DEV_STAT_FLUSH_ERRS); 37305af3e8ccSStefan Behrens errors_wait++; 3731387125fcSChris Mason } 3732401b41e5SAnand Jain } 3733401b41e5SAnand Jain 3734cea7c8bfSAnand Jain if (errors_wait) { 3735401b41e5SAnand Jain /* 3736401b41e5SAnand Jain * At some point we need the status of all disks 3737401b41e5SAnand Jain * to arrive at the volume status. So error checking 3738401b41e5SAnand Jain * is being pushed to a separate loop. 3739401b41e5SAnand Jain */ 3740d10b82feSQu Wenruo return check_barrier_error(info); 3741401b41e5SAnand Jain } 3742387125fcSChris Mason return 0; 3743387125fcSChris Mason } 3744387125fcSChris Mason 3745943c6e99SZhao Lei int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 3746943c6e99SZhao Lei { 37478789f4feSZhao Lei int raid_type; 37488789f4feSZhao Lei int min_tolerated = INT_MAX; 3749943c6e99SZhao Lei 37508789f4feSZhao Lei if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 37518789f4feSZhao Lei (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 37528c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 37538789f4feSZhao Lei btrfs_raid_array[BTRFS_RAID_SINGLE]. 37548789f4feSZhao Lei tolerated_failures); 3755943c6e99SZhao Lei 37568789f4feSZhao Lei for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 37578789f4feSZhao Lei if (raid_type == BTRFS_RAID_SINGLE) 37588789f4feSZhao Lei continue; 375941a6e891SAnand Jain if (!(flags & btrfs_raid_array[raid_type].bg_flag)) 37608789f4feSZhao Lei continue; 37618c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 37628789f4feSZhao Lei btrfs_raid_array[raid_type]. 37638789f4feSZhao Lei tolerated_failures); 37648789f4feSZhao Lei } 3765943c6e99SZhao Lei 37668789f4feSZhao Lei if (min_tolerated == INT_MAX) { 3767ab8d0fc4SJeff Mahoney pr_warn("BTRFS: unknown raid flag: %llu", flags); 37688789f4feSZhao Lei min_tolerated = 0; 37698789f4feSZhao Lei } 37708789f4feSZhao Lei 37718789f4feSZhao Lei return min_tolerated; 3772943c6e99SZhao Lei } 3773943c6e99SZhao Lei 3774eece6a9cSDavid Sterba int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) 3775f2984462SChris Mason { 3776e5e9a520SChris Mason struct list_head *head; 3777f2984462SChris Mason struct btrfs_device *dev; 3778a061fc8dSChris Mason struct btrfs_super_block *sb; 3779f2984462SChris Mason struct btrfs_dev_item *dev_item; 3780f2984462SChris Mason int ret; 3781f2984462SChris Mason int do_barriers; 3782a236aed1SChris Mason int max_errors; 3783a236aed1SChris Mason int total_errors = 0; 3784a061fc8dSChris Mason u64 flags; 3785f2984462SChris Mason 37860b246afaSJeff Mahoney do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); 3787fed3b381SLiu Bo 3788fed3b381SLiu Bo /* 3789fed3b381SLiu Bo * max_mirrors == 0 indicates we're from commit_transaction, 3790fed3b381SLiu Bo * not from fsync where the tree roots in fs_info have not 3791fed3b381SLiu Bo * been consistent on disk. 3792fed3b381SLiu Bo */ 3793fed3b381SLiu Bo if (max_mirrors == 0) 37940b246afaSJeff Mahoney backup_super_roots(fs_info); 3795f2984462SChris Mason 37960b246afaSJeff Mahoney sb = fs_info->super_for_commit; 3797a061fc8dSChris Mason dev_item = &sb->dev_item; 3798e5e9a520SChris Mason 37990b246afaSJeff Mahoney mutex_lock(&fs_info->fs_devices->device_list_mutex); 38000b246afaSJeff Mahoney head = &fs_info->fs_devices->devices; 38010b246afaSJeff Mahoney max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; 3802387125fcSChris Mason 38035af3e8ccSStefan Behrens if (do_barriers) { 38040b246afaSJeff Mahoney ret = barrier_all_devices(fs_info); 38055af3e8ccSStefan Behrens if (ret) { 38065af3e8ccSStefan Behrens mutex_unlock( 38070b246afaSJeff Mahoney &fs_info->fs_devices->device_list_mutex); 38080b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 38095af3e8ccSStefan Behrens "errors while submitting device barriers."); 38105af3e8ccSStefan Behrens return ret; 38115af3e8ccSStefan Behrens } 38125af3e8ccSStefan Behrens } 3813387125fcSChris Mason 38141538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3815dfe25020SChris Mason if (!dev->bdev) { 3816dfe25020SChris Mason total_errors++; 3817dfe25020SChris Mason continue; 3818dfe25020SChris Mason } 3819e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3820ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3821dfe25020SChris Mason continue; 3822dfe25020SChris Mason 38232b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 3824a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 3825a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 38267df69d3eSMiao Xie btrfs_set_stack_device_total_bytes(dev_item, 3827935e5cc9SMiao Xie dev->commit_total_bytes); 3828ce7213c7SMiao Xie btrfs_set_stack_device_bytes_used(dev_item, 3829ce7213c7SMiao Xie dev->commit_bytes_used); 3830a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 3831a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 3832a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 3833a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 38347239ff4bSNikolay Borisov memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, 38357239ff4bSNikolay Borisov BTRFS_FSID_SIZE); 3836a512bbf8SYan Zheng 3837a061fc8dSChris Mason flags = btrfs_super_flags(sb); 3838a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 3839f2984462SChris Mason 384075cb857dSQu Wenruo ret = btrfs_validate_write_super(fs_info, sb); 384175cb857dSQu Wenruo if (ret < 0) { 384275cb857dSQu Wenruo mutex_unlock(&fs_info->fs_devices->device_list_mutex); 384375cb857dSQu Wenruo btrfs_handle_fs_error(fs_info, -EUCLEAN, 384475cb857dSQu Wenruo "unexpected superblock corruption detected"); 384575cb857dSQu Wenruo return -EUCLEAN; 384675cb857dSQu Wenruo } 384775cb857dSQu Wenruo 3848abbb3b8eSDavid Sterba ret = write_dev_supers(dev, sb, max_mirrors); 3849a236aed1SChris Mason if (ret) 3850a236aed1SChris Mason total_errors++; 3851f2984462SChris Mason } 3852a236aed1SChris Mason if (total_errors > max_errors) { 38530b246afaSJeff Mahoney btrfs_err(fs_info, "%d errors while writing supers", 3854d397712bSChris Mason total_errors); 38550b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 385679787eaaSJeff Mahoney 38579d565ba4SStefan Behrens /* FUA is masked off if unsupported and can't be the reason */ 38580b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 38590b246afaSJeff Mahoney "%d errors while writing supers", 38600b246afaSJeff Mahoney total_errors); 38619d565ba4SStefan Behrens return -EIO; 3862a236aed1SChris Mason } 3863f2984462SChris Mason 3864a512bbf8SYan Zheng total_errors = 0; 38651538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3866dfe25020SChris Mason if (!dev->bdev) 3867dfe25020SChris Mason continue; 3868e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3869ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3870dfe25020SChris Mason continue; 3871dfe25020SChris Mason 3872abbb3b8eSDavid Sterba ret = wait_dev_supers(dev, max_mirrors); 3873a512bbf8SYan Zheng if (ret) 38741259ab75SChris Mason total_errors++; 3875f2984462SChris Mason } 38760b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 3877a236aed1SChris Mason if (total_errors > max_errors) { 38780b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 38790b246afaSJeff Mahoney "%d errors while writing supers", 38800b246afaSJeff Mahoney total_errors); 388179787eaaSJeff Mahoney return -EIO; 3882a236aed1SChris Mason } 3883f2984462SChris Mason return 0; 3884f2984462SChris Mason } 3885f2984462SChris Mason 3886cb517eabSMiao Xie /* Drop a fs root from the radix tree and free it. */ 3887cb517eabSMiao Xie void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 3888cb517eabSMiao Xie struct btrfs_root *root) 38892619ba1fSChris Mason { 38904df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 38912619ba1fSChris Mason radix_tree_delete(&fs_info->fs_roots_radix, 38922619ba1fSChris Mason (unsigned long)root->root_key.objectid); 38934df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 389476dda93cSYan, Zheng 389576dda93cSYan, Zheng if (btrfs_root_refs(&root->root_item) == 0) 389676dda93cSYan, Zheng synchronize_srcu(&fs_info->subvol_srcu); 389776dda93cSYan, Zheng 38981c1ea4f7SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 38993321719eSLiu Bo btrfs_free_log(NULL, root); 39001c1ea4f7SLiu Bo if (root->reloc_root) { 39011c1ea4f7SLiu Bo free_extent_buffer(root->reloc_root->node); 39021c1ea4f7SLiu Bo free_extent_buffer(root->reloc_root->commit_root); 39031c1ea4f7SLiu Bo btrfs_put_fs_root(root->reloc_root); 39041c1ea4f7SLiu Bo root->reloc_root = NULL; 39051c1ea4f7SLiu Bo } 39061c1ea4f7SLiu Bo } 39073321719eSLiu Bo 3908faa2dbf0SJosef Bacik if (root->free_ino_pinned) 3909581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_pinned); 3910faa2dbf0SJosef Bacik if (root->free_ino_ctl) 3911581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_ctl); 391284db5ccfSDavid Sterba btrfs_free_fs_root(root); 39134df27c4dSYan, Zheng } 39144df27c4dSYan, Zheng 391584db5ccfSDavid Sterba void btrfs_free_fs_root(struct btrfs_root *root) 39164df27c4dSYan, Zheng { 391757cdc8dbSDavid Sterba iput(root->ino_cache_inode); 39184df27c4dSYan, Zheng WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 39190ee5dc67SAl Viro if (root->anon_dev) 39200ee5dc67SAl Viro free_anon_bdev(root->anon_dev); 39218257b2dcSMiao Xie if (root->subv_writers) 39228257b2dcSMiao Xie btrfs_free_subvolume_writers(root->subv_writers); 39235f39d397SChris Mason free_extent_buffer(root->node); 39245f39d397SChris Mason free_extent_buffer(root->commit_root); 3925581bb050SLi Zefan kfree(root->free_ino_ctl); 3926581bb050SLi Zefan kfree(root->free_ino_pinned); 3927b0feb9d9SMiao Xie btrfs_put_fs_root(root); 39282619ba1fSChris Mason } 39292619ba1fSChris Mason 3930c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 3931c146afadSYan Zheng { 3932c146afadSYan Zheng u64 root_objectid = 0; 3933c146afadSYan Zheng struct btrfs_root *gang[8]; 393465d33fd7SQu Wenruo int i = 0; 393565d33fd7SQu Wenruo int err = 0; 393665d33fd7SQu Wenruo unsigned int ret = 0; 393765d33fd7SQu Wenruo int index; 3938c146afadSYan Zheng 3939c146afadSYan Zheng while (1) { 394065d33fd7SQu Wenruo index = srcu_read_lock(&fs_info->subvol_srcu); 3941c146afadSYan Zheng ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 3942c146afadSYan Zheng (void **)gang, root_objectid, 3943c146afadSYan Zheng ARRAY_SIZE(gang)); 394465d33fd7SQu Wenruo if (!ret) { 394565d33fd7SQu Wenruo srcu_read_unlock(&fs_info->subvol_srcu, index); 3946c146afadSYan Zheng break; 394765d33fd7SQu Wenruo } 39485d4f98a2SYan Zheng root_objectid = gang[ret - 1]->root_key.objectid + 1; 394966b4ffd1SJosef Bacik 395065d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 395165d33fd7SQu Wenruo /* Avoid to grab roots in dead_roots */ 395265d33fd7SQu Wenruo if (btrfs_root_refs(&gang[i]->root_item) == 0) { 395365d33fd7SQu Wenruo gang[i] = NULL; 395465d33fd7SQu Wenruo continue; 395565d33fd7SQu Wenruo } 395665d33fd7SQu Wenruo /* grab all the search result for later use */ 395765d33fd7SQu Wenruo gang[i] = btrfs_grab_fs_root(gang[i]); 395865d33fd7SQu Wenruo } 395965d33fd7SQu Wenruo srcu_read_unlock(&fs_info->subvol_srcu, index); 396065d33fd7SQu Wenruo 396165d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 396265d33fd7SQu Wenruo if (!gang[i]) 396365d33fd7SQu Wenruo continue; 3964c146afadSYan Zheng root_objectid = gang[i]->root_key.objectid; 396566b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(gang[i]); 396666b4ffd1SJosef Bacik if (err) 396765d33fd7SQu Wenruo break; 396865d33fd7SQu Wenruo btrfs_put_fs_root(gang[i]); 3969c146afadSYan Zheng } 3970c146afadSYan Zheng root_objectid++; 3971c146afadSYan Zheng } 397265d33fd7SQu Wenruo 397365d33fd7SQu Wenruo /* release the uncleaned roots due to error */ 397465d33fd7SQu Wenruo for (; i < ret; i++) { 397565d33fd7SQu Wenruo if (gang[i]) 397665d33fd7SQu Wenruo btrfs_put_fs_root(gang[i]); 397765d33fd7SQu Wenruo } 397865d33fd7SQu Wenruo return err; 3979c146afadSYan Zheng } 3980c146afadSYan Zheng 39816bccf3abSJeff Mahoney int btrfs_commit_super(struct btrfs_fs_info *fs_info) 3982c146afadSYan Zheng { 39836bccf3abSJeff Mahoney struct btrfs_root *root = fs_info->tree_root; 3984c146afadSYan Zheng struct btrfs_trans_handle *trans; 3985c146afadSYan Zheng 39860b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 39872ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 39880b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 39890b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 3990c71bf099SYan, Zheng 3991c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 39920b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 39930b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 3994c71bf099SYan, Zheng 39957a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 39963612b495STsutomu Itoh if (IS_ERR(trans)) 39973612b495STsutomu Itoh return PTR_ERR(trans); 39983a45bb20SJeff Mahoney return btrfs_commit_transaction(trans); 3999c146afadSYan Zheng } 4000c146afadSYan Zheng 40016bccf3abSJeff Mahoney void close_ctree(struct btrfs_fs_info *fs_info) 4002eb60ceacSChris Mason { 4003c146afadSYan Zheng int ret; 4004e089f05cSChris Mason 4005afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); 4006d6fd0ae2SOmar Sandoval /* 4007d6fd0ae2SOmar Sandoval * We don't want the cleaner to start new transactions, add more delayed 4008d6fd0ae2SOmar Sandoval * iputs, etc. while we're closing. We can't use kthread_stop() yet 4009d6fd0ae2SOmar Sandoval * because that frees the task_struct, and the transaction kthread might 4010d6fd0ae2SOmar Sandoval * still try to wake up the cleaner. 4011d6fd0ae2SOmar Sandoval */ 4012d6fd0ae2SOmar Sandoval kthread_park(fs_info->cleaner_kthread); 4013a2135011SChris Mason 40147343dd61SJustin Maggard /* wait for the qgroup rescan worker to stop */ 4015d06f23d6SJeff Mahoney btrfs_qgroup_wait_for_completion(fs_info, false); 40167343dd61SJustin Maggard 4017803b2f54SStefan Behrens /* wait for the uuid_scan task to finish */ 4018803b2f54SStefan Behrens down(&fs_info->uuid_tree_rescan_sem); 4019803b2f54SStefan Behrens /* avoid complains from lockdep et al., set sem back to initial state */ 4020803b2f54SStefan Behrens up(&fs_info->uuid_tree_rescan_sem); 4021803b2f54SStefan Behrens 4022837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 4023aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 4024837d5b6eSIlya Dryomov 40258dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 40268dabb742SStefan Behrens 4027aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 40284cb5300bSChris Mason 40294cb5300bSChris Mason /* wait for any defraggers to finish */ 40304cb5300bSChris Mason wait_event(fs_info->transaction_wait, 40314cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 40324cb5300bSChris Mason 40334cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 403426176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 40354cb5300bSChris Mason 403621c7e756SMiao Xie cancel_work_sync(&fs_info->async_reclaim_work); 403721c7e756SMiao Xie 4038bc98a42cSDavid Howells if (!sb_rdonly(fs_info->sb)) { 4039e44163e1SJeff Mahoney /* 4040d6fd0ae2SOmar Sandoval * The cleaner kthread is stopped, so do one final pass over 4041d6fd0ae2SOmar Sandoval * unused block groups. 4042e44163e1SJeff Mahoney */ 40430b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 4044e44163e1SJeff Mahoney 40456bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 4046d397712bSChris Mason if (ret) 404704892340SEric Sandeen btrfs_err(fs_info, "commit super ret %d", ret); 4048c146afadSYan Zheng } 4049ed2ff2cbSChris Mason 4050af722733SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) || 4051af722733SLiu Bo test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) 40522ff7e61eSJeff Mahoney btrfs_error_commit_super(fs_info); 4053acce952bSliubo 4054e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 4055e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 40568929ecfaSYan, Zheng 4057e187831eSJosef Bacik ASSERT(list_empty(&fs_info->delayed_iputs)); 4058afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); 4059f25784b3SYan Zheng 406004892340SEric Sandeen btrfs_free_qgroup_config(fs_info); 4061fe816d0fSNikolay Borisov ASSERT(list_empty(&fs_info->delalloc_roots)); 4062bcef60f2SArne Jansen 4063963d678bSMiao Xie if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 406404892340SEric Sandeen btrfs_info(fs_info, "at unmount delalloc count %lld", 4065963d678bSMiao Xie percpu_counter_sum(&fs_info->delalloc_bytes)); 4066b0c68f8bSChris Mason } 406731153d81SYan Zheng 40684297ff84SJosef Bacik if (percpu_counter_sum(&fs_info->dio_bytes)) 40694297ff84SJosef Bacik btrfs_info(fs_info, "at unmount dio bytes count %lld", 40704297ff84SJosef Bacik percpu_counter_sum(&fs_info->dio_bytes)); 40714297ff84SJosef Bacik 40726618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 4073b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 40745ac1d209SJeff Mahoney 4075faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 4076d10c5f31SChris Mason 40771a4319ccSLiu Bo btrfs_put_block_group_cache(fs_info); 40781a4319ccSLiu Bo 4079de348ee0SWang Shilong /* 4080de348ee0SWang Shilong * we must make sure there is not any read request to 4081de348ee0SWang Shilong * submit after we stopping all workers. 4082de348ee0SWang Shilong */ 4083de348ee0SWang Shilong invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 408496192499SJosef Bacik btrfs_stop_all_workers(fs_info); 408596192499SJosef Bacik 40865cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 40875cdd7db6SFilipe Manana 4088afcdd129SJosef Bacik clear_bit(BTRFS_FS_OPEN, &fs_info->flags); 408913e6c37bSJosef Bacik free_root_pointers(fs_info, 1); 40909ad6b7bcSChris Mason 409113e6c37bSJosef Bacik iput(fs_info->btree_inode); 4092d6bfde87SChris Mason 409321adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 40940b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) 40952ff7e61eSJeff Mahoney btrfsic_unmount(fs_info->fs_devices); 409621adbd5cSStefan Behrens #endif 409721adbd5cSStefan Behrens 40980b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 409968c94e55SNikolay Borisov btrfs_close_devices(fs_info->fs_devices); 4100b248a415SChris Mason 4101e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 4102963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 41034297ff84SJosef Bacik percpu_counter_destroy(&fs_info->dio_bytes); 41047f8d236aSDavid Sterba percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 410576dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 41060b86a832SChris Mason 410753b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 4108fd708b81SJosef Bacik btrfs_free_ref_cache(fs_info); 4109eb60ceacSChris Mason } 4110eb60ceacSChris Mason 4111b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 4112b9fab919SChris Mason int atomic) 4113ccd467d6SChris Mason { 41141259ab75SChris Mason int ret; 4115727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 41161259ab75SChris Mason 41170b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 41181259ab75SChris Mason if (!ret) 41191259ab75SChris Mason return ret; 41201259ab75SChris Mason 41211259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 4122b9fab919SChris Mason parent_transid, atomic); 4123b9fab919SChris Mason if (ret == -EAGAIN) 4124b9fab919SChris Mason return ret; 41251259ab75SChris Mason return !ret; 41265f39d397SChris Mason } 41276702ed49SChris Mason 41285f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 41295f39d397SChris Mason { 41300b246afaSJeff Mahoney struct btrfs_fs_info *fs_info; 413106ea65a3SJosef Bacik struct btrfs_root *root; 41325f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 4133b9473439SChris Mason int was_dirty; 4134b4ce94deSChris Mason 413506ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 413606ea65a3SJosef Bacik /* 413706ea65a3SJosef Bacik * This is a fast path so only do this check if we have sanity tests 413852042d8eSAndrea Gelmini * enabled. Normal people shouldn't be using unmapped buffers as dirty 413906ea65a3SJosef Bacik * outside of the sanity tests. 414006ea65a3SJosef Bacik */ 4141b0132a3bSNikolay Borisov if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) 414206ea65a3SJosef Bacik return; 414306ea65a3SJosef Bacik #endif 414406ea65a3SJosef Bacik root = BTRFS_I(buf->pages[0]->mapping->host)->root; 41450b246afaSJeff Mahoney fs_info = root->fs_info; 4146b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 41470b246afaSJeff Mahoney if (transid != fs_info->generation) 41485d163e0eSJeff Mahoney WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", 41490b246afaSJeff Mahoney buf->start, transid, fs_info->generation); 41500b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 4151e2d84521SMiao Xie if (!was_dirty) 4152104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 4153e2d84521SMiao Xie buf->len, 41540b246afaSJeff Mahoney fs_info->dirty_metadata_batch); 41551f21ef0aSFilipe Manana #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 415669fc6cbbSQu Wenruo /* 415769fc6cbbSQu Wenruo * Since btrfs_mark_buffer_dirty() can be called with item pointer set 415869fc6cbbSQu Wenruo * but item data not updated. 415969fc6cbbSQu Wenruo * So here we should only check item pointers, not item data. 416069fc6cbbSQu Wenruo */ 416169fc6cbbSQu Wenruo if (btrfs_header_level(buf) == 0 && 4162cfdaad5eSDavid Sterba btrfs_check_leaf_relaxed(buf)) { 4163a4f78750SDavid Sterba btrfs_print_leaf(buf); 41641f21ef0aSFilipe Manana ASSERT(0); 41651f21ef0aSFilipe Manana } 41661f21ef0aSFilipe Manana #endif 4167eb60ceacSChris Mason } 4168eb60ceacSChris Mason 41692ff7e61eSJeff Mahoney static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, 4170b53d3f5dSLiu Bo int flush_delayed) 417135b7e476SChris Mason { 4172188de649SChris Mason /* 4173188de649SChris Mason * looks as though older kernels can get into trouble with 4174188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 4175188de649SChris Mason */ 4176e2d84521SMiao Xie int ret; 4177d6bfde87SChris Mason 41786933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 4179d6bfde87SChris Mason return; 4180d6bfde87SChris Mason 4181b53d3f5dSLiu Bo if (flush_delayed) 41822ff7e61eSJeff Mahoney btrfs_balance_delayed_items(fs_info); 418316cdcec7SMiao Xie 4184d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 4185d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 4186d814a491SEthan Lien fs_info->dirty_metadata_batch); 4187e2d84521SMiao Xie if (ret > 0) { 41880b246afaSJeff Mahoney balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); 418916cdcec7SMiao Xie } 419016cdcec7SMiao Xie } 419116cdcec7SMiao Xie 41922ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) 419316cdcec7SMiao Xie { 41942ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 1); 419535b7e476SChris Mason } 4196b53d3f5dSLiu Bo 41972ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) 4198b53d3f5dSLiu Bo { 41992ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 0); 4200d6bfde87SChris Mason } 42016b80053dSChris Mason 4202581c1760SQu Wenruo int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level, 4203581c1760SQu Wenruo struct btrfs_key *first_key) 42046b80053dSChris Mason { 42055ab12d1fSDavid Sterba return btree_read_extent_buffer_pages(buf, parent_transid, 4206581c1760SQu Wenruo level, first_key); 42076b80053dSChris Mason } 42080da5468fSChris Mason 42092ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) 4210acce952bSliubo { 4211fe816d0fSNikolay Borisov /* cleanup FS via transaction */ 4212fe816d0fSNikolay Borisov btrfs_cleanup_transaction(fs_info); 4213fe816d0fSNikolay Borisov 42140b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 42152ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 42160b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 4217acce952bSliubo 42180b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 42190b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4220acce952bSliubo } 4221acce952bSliubo 4222143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4223acce952bSliubo { 4224acce952bSliubo struct btrfs_ordered_extent *ordered; 4225acce952bSliubo 4226199c2a9cSMiao Xie spin_lock(&root->ordered_extent_lock); 4227779880efSJosef Bacik /* 4228779880efSJosef Bacik * This will just short circuit the ordered completion stuff which will 4229779880efSJosef Bacik * make sure the ordered extent gets properly cleaned up. 4230779880efSJosef Bacik */ 4231199c2a9cSMiao Xie list_for_each_entry(ordered, &root->ordered_extents, 4232779880efSJosef Bacik root_extent_list) 4233779880efSJosef Bacik set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4234199c2a9cSMiao Xie spin_unlock(&root->ordered_extent_lock); 4235199c2a9cSMiao Xie } 4236199c2a9cSMiao Xie 4237199c2a9cSMiao Xie static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4238199c2a9cSMiao Xie { 4239199c2a9cSMiao Xie struct btrfs_root *root; 4240199c2a9cSMiao Xie struct list_head splice; 4241199c2a9cSMiao Xie 4242199c2a9cSMiao Xie INIT_LIST_HEAD(&splice); 4243199c2a9cSMiao Xie 4244199c2a9cSMiao Xie spin_lock(&fs_info->ordered_root_lock); 4245199c2a9cSMiao Xie list_splice_init(&fs_info->ordered_roots, &splice); 4246199c2a9cSMiao Xie while (!list_empty(&splice)) { 4247199c2a9cSMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4248199c2a9cSMiao Xie ordered_root); 42491de2cfdeSJosef Bacik list_move_tail(&root->ordered_root, 42501de2cfdeSJosef Bacik &fs_info->ordered_roots); 4251199c2a9cSMiao Xie 42522a85d9caSLiu Bo spin_unlock(&fs_info->ordered_root_lock); 4253199c2a9cSMiao Xie btrfs_destroy_ordered_extents(root); 4254199c2a9cSMiao Xie 42552a85d9caSLiu Bo cond_resched(); 42562a85d9caSLiu Bo spin_lock(&fs_info->ordered_root_lock); 4257199c2a9cSMiao Xie } 4258199c2a9cSMiao Xie spin_unlock(&fs_info->ordered_root_lock); 425974d5d229SJosef Bacik 426074d5d229SJosef Bacik /* 426174d5d229SJosef Bacik * We need this here because if we've been flipped read-only we won't 426274d5d229SJosef Bacik * get sync() from the umount, so we need to make sure any ordered 426374d5d229SJosef Bacik * extents that haven't had their dirty pages IO start writeout yet 426474d5d229SJosef Bacik * actually get run and error out properly. 426574d5d229SJosef Bacik */ 426674d5d229SJosef Bacik btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 4267acce952bSliubo } 4268acce952bSliubo 426935a3621bSStefan Behrens static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 42702ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4271acce952bSliubo { 4272acce952bSliubo struct rb_node *node; 4273acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 4274acce952bSliubo struct btrfs_delayed_ref_node *ref; 4275acce952bSliubo int ret = 0; 4276acce952bSliubo 4277acce952bSliubo delayed_refs = &trans->delayed_refs; 4278acce952bSliubo 4279acce952bSliubo spin_lock(&delayed_refs->lock); 4280d7df2c79SJosef Bacik if (atomic_read(&delayed_refs->num_entries) == 0) { 4281cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 42820b246afaSJeff Mahoney btrfs_info(fs_info, "delayed_refs has NO entry"); 4283acce952bSliubo return ret; 4284acce952bSliubo } 4285acce952bSliubo 42865c9d028bSLiu Bo while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { 4287d7df2c79SJosef Bacik struct btrfs_delayed_ref_head *head; 42880e0adbcfSJosef Bacik struct rb_node *n; 4289e78417d1SJosef Bacik bool pin_bytes = false; 4290acce952bSliubo 4291d7df2c79SJosef Bacik head = rb_entry(node, struct btrfs_delayed_ref_head, 4292d7df2c79SJosef Bacik href_node); 42933069bd26SJosef Bacik if (btrfs_delayed_ref_lock(delayed_refs, head)) 4294b939d1abSJosef Bacik continue; 42953069bd26SJosef Bacik 4296d7df2c79SJosef Bacik spin_lock(&head->lock); 4297e3d03965SLiu Bo while ((n = rb_first_cached(&head->ref_tree)) != NULL) { 42980e0adbcfSJosef Bacik ref = rb_entry(n, struct btrfs_delayed_ref_node, 42990e0adbcfSJosef Bacik ref_node); 4300d7df2c79SJosef Bacik ref->in_tree = 0; 4301e3d03965SLiu Bo rb_erase_cached(&ref->ref_node, &head->ref_tree); 43020e0adbcfSJosef Bacik RB_CLEAR_NODE(&ref->ref_node); 43031d57ee94SWang Xiaoguang if (!list_empty(&ref->add_list)) 43041d57ee94SWang Xiaoguang list_del(&ref->add_list); 4305d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4306d7df2c79SJosef Bacik btrfs_put_delayed_ref(ref); 4307d7df2c79SJosef Bacik } 430854067ae9SJosef Bacik if (head->must_insert_reserved) 4309e78417d1SJosef Bacik pin_bytes = true; 431078a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 4311fa781ceaSJosef Bacik btrfs_delete_ref_head(delayed_refs, head); 4312d7df2c79SJosef Bacik spin_unlock(&head->lock); 4313acce952bSliubo spin_unlock(&delayed_refs->lock); 4314e78417d1SJosef Bacik mutex_unlock(&head->mutex); 4315acce952bSliubo 4316d7df2c79SJosef Bacik if (pin_bytes) 4317d278850eSJosef Bacik btrfs_pin_extent(fs_info, head->bytenr, 4318d278850eSJosef Bacik head->num_bytes, 1); 431931890da0SJosef Bacik btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); 4320d278850eSJosef Bacik btrfs_put_delayed_ref_head(head); 4321acce952bSliubo cond_resched(); 4322acce952bSliubo spin_lock(&delayed_refs->lock); 4323acce952bSliubo } 4324acce952bSliubo 4325acce952bSliubo spin_unlock(&delayed_refs->lock); 4326acce952bSliubo 4327acce952bSliubo return ret; 4328acce952bSliubo } 4329acce952bSliubo 4330143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 4331acce952bSliubo { 4332acce952bSliubo struct btrfs_inode *btrfs_inode; 4333acce952bSliubo struct list_head splice; 4334acce952bSliubo 4335acce952bSliubo INIT_LIST_HEAD(&splice); 4336acce952bSliubo 4337eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4338eb73c1b7SMiao Xie list_splice_init(&root->delalloc_inodes, &splice); 4339acce952bSliubo 4340acce952bSliubo while (!list_empty(&splice)) { 4341fe816d0fSNikolay Borisov struct inode *inode = NULL; 4342eb73c1b7SMiao Xie btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 4343acce952bSliubo delalloc_inodes); 4344fe816d0fSNikolay Borisov __btrfs_del_delalloc_inode(root, btrfs_inode); 4345eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4346acce952bSliubo 4347fe816d0fSNikolay Borisov /* 4348fe816d0fSNikolay Borisov * Make sure we get a live inode and that it'll not disappear 4349fe816d0fSNikolay Borisov * meanwhile. 4350fe816d0fSNikolay Borisov */ 4351fe816d0fSNikolay Borisov inode = igrab(&btrfs_inode->vfs_inode); 4352fe816d0fSNikolay Borisov if (inode) { 4353fe816d0fSNikolay Borisov invalidate_inode_pages2(inode->i_mapping); 4354fe816d0fSNikolay Borisov iput(inode); 4355fe816d0fSNikolay Borisov } 4356eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4357acce952bSliubo } 4358eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4359eb73c1b7SMiao Xie } 4360eb73c1b7SMiao Xie 4361eb73c1b7SMiao Xie static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 4362eb73c1b7SMiao Xie { 4363eb73c1b7SMiao Xie struct btrfs_root *root; 4364eb73c1b7SMiao Xie struct list_head splice; 4365eb73c1b7SMiao Xie 4366eb73c1b7SMiao Xie INIT_LIST_HEAD(&splice); 4367eb73c1b7SMiao Xie 4368eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4369eb73c1b7SMiao Xie list_splice_init(&fs_info->delalloc_roots, &splice); 4370eb73c1b7SMiao Xie while (!list_empty(&splice)) { 4371eb73c1b7SMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4372eb73c1b7SMiao Xie delalloc_root); 4373eb73c1b7SMiao Xie root = btrfs_grab_fs_root(root); 4374eb73c1b7SMiao Xie BUG_ON(!root); 4375eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4376eb73c1b7SMiao Xie 4377eb73c1b7SMiao Xie btrfs_destroy_delalloc_inodes(root); 4378eb73c1b7SMiao Xie btrfs_put_fs_root(root); 4379eb73c1b7SMiao Xie 4380eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4381eb73c1b7SMiao Xie } 4382eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4383acce952bSliubo } 4384acce952bSliubo 43852ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 4386acce952bSliubo struct extent_io_tree *dirty_pages, 4387acce952bSliubo int mark) 4388acce952bSliubo { 4389acce952bSliubo int ret; 4390acce952bSliubo struct extent_buffer *eb; 4391acce952bSliubo u64 start = 0; 4392acce952bSliubo u64 end; 4393acce952bSliubo 4394acce952bSliubo while (1) { 4395acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 4396e6138876SJosef Bacik mark, NULL); 4397acce952bSliubo if (ret) 4398acce952bSliubo break; 4399acce952bSliubo 440091166212SDavid Sterba clear_extent_bits(dirty_pages, start, end, mark); 4401acce952bSliubo while (start <= end) { 44020b246afaSJeff Mahoney eb = find_extent_buffer(fs_info, start); 44030b246afaSJeff Mahoney start += fs_info->nodesize; 4404fd8b2b61SJosef Bacik if (!eb) 4405acce952bSliubo continue; 4406fd8b2b61SJosef Bacik wait_on_extent_buffer_writeback(eb); 4407acce952bSliubo 4408fd8b2b61SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 4409fd8b2b61SJosef Bacik &eb->bflags)) 4410fd8b2b61SJosef Bacik clear_extent_buffer_dirty(eb); 4411fd8b2b61SJosef Bacik free_extent_buffer_stale(eb); 4412acce952bSliubo } 4413acce952bSliubo } 4414acce952bSliubo 4415acce952bSliubo return ret; 4416acce952bSliubo } 4417acce952bSliubo 44182ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 4419acce952bSliubo struct extent_io_tree *pinned_extents) 4420acce952bSliubo { 4421acce952bSliubo struct extent_io_tree *unpin; 4422acce952bSliubo u64 start; 4423acce952bSliubo u64 end; 4424acce952bSliubo int ret; 4425ed0eaa14SLiu Bo bool loop = true; 4426acce952bSliubo 4427acce952bSliubo unpin = pinned_extents; 4428ed0eaa14SLiu Bo again: 4429acce952bSliubo while (1) { 44300e6ec385SFilipe Manana struct extent_state *cached_state = NULL; 44310e6ec385SFilipe Manana 4432fcd5e742SLu Fengqi /* 4433fcd5e742SLu Fengqi * The btrfs_finish_extent_commit() may get the same range as 4434fcd5e742SLu Fengqi * ours between find_first_extent_bit and clear_extent_dirty. 4435fcd5e742SLu Fengqi * Hence, hold the unused_bg_unpin_mutex to avoid double unpin 4436fcd5e742SLu Fengqi * the same extent range. 4437fcd5e742SLu Fengqi */ 4438fcd5e742SLu Fengqi mutex_lock(&fs_info->unused_bg_unpin_mutex); 4439acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 44400e6ec385SFilipe Manana EXTENT_DIRTY, &cached_state); 4441fcd5e742SLu Fengqi if (ret) { 4442fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4443acce952bSliubo break; 4444fcd5e742SLu Fengqi } 4445acce952bSliubo 44460e6ec385SFilipe Manana clear_extent_dirty(unpin, start, end, &cached_state); 44470e6ec385SFilipe Manana free_extent_state(cached_state); 44482ff7e61eSJeff Mahoney btrfs_error_unpin_extent_range(fs_info, start, end); 4449fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4450acce952bSliubo cond_resched(); 4451acce952bSliubo } 4452acce952bSliubo 4453ed0eaa14SLiu Bo if (loop) { 44540b246afaSJeff Mahoney if (unpin == &fs_info->freed_extents[0]) 44550b246afaSJeff Mahoney unpin = &fs_info->freed_extents[1]; 4456ed0eaa14SLiu Bo else 44570b246afaSJeff Mahoney unpin = &fs_info->freed_extents[0]; 4458ed0eaa14SLiu Bo loop = false; 4459ed0eaa14SLiu Bo goto again; 4460ed0eaa14SLiu Bo } 4461ed0eaa14SLiu Bo 4462acce952bSliubo return 0; 4463acce952bSliubo } 4464acce952bSliubo 4465c79a1751SLiu Bo static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache) 4466c79a1751SLiu Bo { 4467c79a1751SLiu Bo struct inode *inode; 4468c79a1751SLiu Bo 4469c79a1751SLiu Bo inode = cache->io_ctl.inode; 4470c79a1751SLiu Bo if (inode) { 4471c79a1751SLiu Bo invalidate_inode_pages2(inode->i_mapping); 4472c79a1751SLiu Bo BTRFS_I(inode)->generation = 0; 4473c79a1751SLiu Bo cache->io_ctl.inode = NULL; 4474c79a1751SLiu Bo iput(inode); 4475c79a1751SLiu Bo } 4476c79a1751SLiu Bo btrfs_put_block_group(cache); 4477c79a1751SLiu Bo } 4478c79a1751SLiu Bo 4479c79a1751SLiu Bo void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, 44802ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4481c79a1751SLiu Bo { 4482c79a1751SLiu Bo struct btrfs_block_group_cache *cache; 4483c79a1751SLiu Bo 4484c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4485c79a1751SLiu Bo while (!list_empty(&cur_trans->dirty_bgs)) { 4486c79a1751SLiu Bo cache = list_first_entry(&cur_trans->dirty_bgs, 4487c79a1751SLiu Bo struct btrfs_block_group_cache, 4488c79a1751SLiu Bo dirty_list); 4489c79a1751SLiu Bo 4490c79a1751SLiu Bo if (!list_empty(&cache->io_list)) { 4491c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4492c79a1751SLiu Bo list_del_init(&cache->io_list); 4493c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 4494c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4495c79a1751SLiu Bo } 4496c79a1751SLiu Bo 4497c79a1751SLiu Bo list_del_init(&cache->dirty_list); 4498c79a1751SLiu Bo spin_lock(&cache->lock); 4499c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 4500c79a1751SLiu Bo spin_unlock(&cache->lock); 4501c79a1751SLiu Bo 4502c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4503c79a1751SLiu Bo btrfs_put_block_group(cache); 4504ba2c4d4eSJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 4505c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4506c79a1751SLiu Bo } 4507c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4508c79a1751SLiu Bo 450945ae2c18SNikolay Borisov /* 451045ae2c18SNikolay Borisov * Refer to the definition of io_bgs member for details why it's safe 451145ae2c18SNikolay Borisov * to use it without any locking 451245ae2c18SNikolay Borisov */ 4513c79a1751SLiu Bo while (!list_empty(&cur_trans->io_bgs)) { 4514c79a1751SLiu Bo cache = list_first_entry(&cur_trans->io_bgs, 4515c79a1751SLiu Bo struct btrfs_block_group_cache, 4516c79a1751SLiu Bo io_list); 4517c79a1751SLiu Bo 4518c79a1751SLiu Bo list_del_init(&cache->io_list); 4519c79a1751SLiu Bo spin_lock(&cache->lock); 4520c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 4521c79a1751SLiu Bo spin_unlock(&cache->lock); 4522c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 4523c79a1751SLiu Bo } 4524c79a1751SLiu Bo } 4525c79a1751SLiu Bo 452649b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 45272ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 452849b25e05SJeff Mahoney { 4529bbbf7243SNikolay Borisov struct btrfs_device *dev, *tmp; 4530bbbf7243SNikolay Borisov 45312ff7e61eSJeff Mahoney btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 4532c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->dirty_bgs)); 4533c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->io_bgs)); 4534c79a1751SLiu Bo 4535bbbf7243SNikolay Borisov list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list, 4536bbbf7243SNikolay Borisov post_commit_list) { 4537bbbf7243SNikolay Borisov list_del_init(&dev->post_commit_list); 4538bbbf7243SNikolay Borisov } 4539bbbf7243SNikolay Borisov 45402ff7e61eSJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, fs_info); 454149b25e05SJeff Mahoney 45424a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 45430b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 454449b25e05SJeff Mahoney 45454a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 45460b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 454749b25e05SJeff Mahoney 4548ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 4549ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 455067cde344SMiao Xie 45512ff7e61eSJeff Mahoney btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, 455249b25e05SJeff Mahoney EXTENT_DIRTY); 45532ff7e61eSJeff Mahoney btrfs_destroy_pinned_extent(fs_info, 45540b246afaSJeff Mahoney fs_info->pinned_extents); 455549b25e05SJeff Mahoney 45564a9d8bdeSMiao Xie cur_trans->state =TRANS_STATE_COMPLETED; 45574a9d8bdeSMiao Xie wake_up(&cur_trans->commit_wait); 455849b25e05SJeff Mahoney } 455949b25e05SJeff Mahoney 45602ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) 4561acce952bSliubo { 4562acce952bSliubo struct btrfs_transaction *t; 4563acce952bSliubo 45640b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 4565acce952bSliubo 45660b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 45670b246afaSJeff Mahoney while (!list_empty(&fs_info->trans_list)) { 45680b246afaSJeff Mahoney t = list_first_entry(&fs_info->trans_list, 4569724e2315SJosef Bacik struct btrfs_transaction, list); 4570724e2315SJosef Bacik if (t->state >= TRANS_STATE_COMMIT_START) { 45719b64f57dSElena Reshetova refcount_inc(&t->use_count); 45720b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 45732ff7e61eSJeff Mahoney btrfs_wait_for_commit(fs_info, t->transid); 4574724e2315SJosef Bacik btrfs_put_transaction(t); 45750b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4576724e2315SJosef Bacik continue; 4577724e2315SJosef Bacik } 45780b246afaSJeff Mahoney if (t == fs_info->running_transaction) { 4579724e2315SJosef Bacik t->state = TRANS_STATE_COMMIT_DOING; 45800b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4581724e2315SJosef Bacik /* 4582724e2315SJosef Bacik * We wait for 0 num_writers since we don't hold a trans 4583724e2315SJosef Bacik * handle open currently for this transaction. 4584724e2315SJosef Bacik */ 4585724e2315SJosef Bacik wait_event(t->writer_wait, 4586724e2315SJosef Bacik atomic_read(&t->num_writers) == 0); 4587724e2315SJosef Bacik } else { 45880b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4589724e2315SJosef Bacik } 45902ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(t, fs_info); 4591724e2315SJosef Bacik 45920b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 45930b246afaSJeff Mahoney if (t == fs_info->running_transaction) 45940b246afaSJeff Mahoney fs_info->running_transaction = NULL; 4595724e2315SJosef Bacik list_del_init(&t->list); 45960b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4597a4abeea4SJosef Bacik 4598724e2315SJosef Bacik btrfs_put_transaction(t); 45992ff7e61eSJeff Mahoney trace_btrfs_transaction_commit(fs_info->tree_root); 46000b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4601724e2315SJosef Bacik } 46020b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 46030b246afaSJeff Mahoney btrfs_destroy_all_ordered_extents(fs_info); 4604ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 4605ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 46062ff7e61eSJeff Mahoney btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents); 46070b246afaSJeff Mahoney btrfs_destroy_all_delalloc_inodes(fs_info); 46080b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 4609acce952bSliubo 4610acce952bSliubo return 0; 4611acce952bSliubo } 4612acce952bSliubo 4613e8c9f186SDavid Sterba static const struct extent_io_ops btree_extent_io_ops = { 46144d53dddbSDavid Sterba /* mandatory callbacks */ 46150b86a832SChris Mason .submit_bio_hook = btree_submit_bio_hook, 46164d53dddbSDavid Sterba .readpage_end_io_hook = btree_readpage_end_io_hook, 46170da5468fSChris Mason }; 4618