1c1d7c514SDavid Sterba // SPDX-License-Identifier: GPL-2.0 26cbd5570SChris Mason /* 36cbd5570SChris Mason * Copyright (C) 2007 Oracle. All rights reserved. 46cbd5570SChris Mason */ 56cbd5570SChris Mason 6e20d96d6SChris Mason #include <linux/fs.h> 7d98237b3SChris Mason #include <linux/blkdev.h> 80f7d52f4SChris Mason #include <linux/radix-tree.h> 935b7e476SChris Mason #include <linux/writeback.h> 10d397712bSChris Mason #include <linux/buffer_head.h> 11ce9adaa5SChris Mason #include <linux/workqueue.h> 12a74a4b97SChris Mason #include <linux/kthread.h> 135a0e3ad6STejun Heo #include <linux/slab.h> 14784b4e29SChris Mason #include <linux/migrate.h> 157a36ddecSDavid Sterba #include <linux/ratelimit.h> 166463fe58SStefan Behrens #include <linux/uuid.h> 17803b2f54SStefan Behrens #include <linux/semaphore.h> 18540adea3SMasami Hiramatsu #include <linux/error-injection.h> 199678c543SNikolay Borisov #include <linux/crc32c.h> 20b89f6d1fSFilipe Manana #include <linux/sched/mm.h> 217e75bf3fSDavid Sterba #include <asm/unaligned.h> 226d97c6e3SJohannes Thumshirn #include <crypto/hash.h> 23eb60ceacSChris Mason #include "ctree.h" 24eb60ceacSChris Mason #include "disk-io.h" 25e089f05cSChris Mason #include "transaction.h" 260f7d52f4SChris Mason #include "btrfs_inode.h" 270b86a832SChris Mason #include "volumes.h" 28db94535dSChris Mason #include "print-tree.h" 29925baeddSChris Mason #include "locking.h" 30e02119d5SChris Mason #include "tree-log.h" 31fa9c0d79SChris Mason #include "free-space-cache.h" 3270f6d82eSOmar Sandoval #include "free-space-tree.h" 33581bb050SLi Zefan #include "inode-map.h" 3421adbd5cSStefan Behrens #include "check-integrity.h" 35606686eeSJosef Bacik #include "rcu-string.h" 368dabb742SStefan Behrens #include "dev-replace.h" 3753b381b3SDavid Woodhouse #include "raid56.h" 385ac1d209SJeff Mahoney #include "sysfs.h" 39fcebe456SJosef Bacik #include "qgroup.h" 40ebb8765bSAnand Jain #include "compression.h" 41557ea5ddSQu Wenruo #include "tree-checker.h" 42fd708b81SJosef Bacik #include "ref-verify.h" 43aac0023cSJosef Bacik #include "block-group.h" 44eb60ceacSChris Mason 45319e4d06SQu Wenruo #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\ 46319e4d06SQu Wenruo BTRFS_HEADER_FLAG_RELOC |\ 47319e4d06SQu Wenruo BTRFS_SUPER_FLAG_ERROR |\ 48319e4d06SQu Wenruo BTRFS_SUPER_FLAG_SEEDING |\ 49e2731e55SAnand Jain BTRFS_SUPER_FLAG_METADUMP |\ 50e2731e55SAnand Jain BTRFS_SUPER_FLAG_METADUMP_V2) 51319e4d06SQu Wenruo 52e8c9f186SDavid Sterba static const struct extent_io_ops btree_extent_io_ops; 538b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work); 54143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root); 55acce952bSliubo static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 562ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info); 57143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root); 582ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 59acce952bSliubo struct extent_io_tree *dirty_pages, 60acce952bSliubo int mark); 612ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 62acce952bSliubo struct extent_io_tree *pinned_extents); 632ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info); 642ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info); 65ce9adaa5SChris Mason 66d352ac68SChris Mason /* 6797eb6b69SDavid Sterba * btrfs_end_io_wq structs are used to do processing in task context when an IO 6897eb6b69SDavid Sterba * is complete. This is used during reads to verify checksums, and it is used 69d352ac68SChris Mason * by writes to insert metadata for new file extents after IO is complete. 70d352ac68SChris Mason */ 7197eb6b69SDavid Sterba struct btrfs_end_io_wq { 72ce9adaa5SChris Mason struct bio *bio; 73ce9adaa5SChris Mason bio_end_io_t *end_io; 74ce9adaa5SChris Mason void *private; 75ce9adaa5SChris Mason struct btrfs_fs_info *info; 764e4cbee9SChristoph Hellwig blk_status_t status; 77bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata; 788b712842SChris Mason struct btrfs_work work; 79ce9adaa5SChris Mason }; 800da5468fSChris Mason 8197eb6b69SDavid Sterba static struct kmem_cache *btrfs_end_io_wq_cache; 8297eb6b69SDavid Sterba 8397eb6b69SDavid Sterba int __init btrfs_end_io_wq_init(void) 8497eb6b69SDavid Sterba { 8597eb6b69SDavid Sterba btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq", 8697eb6b69SDavid Sterba sizeof(struct btrfs_end_io_wq), 8797eb6b69SDavid Sterba 0, 88fba4b697SNikolay Borisov SLAB_MEM_SPREAD, 8997eb6b69SDavid Sterba NULL); 9097eb6b69SDavid Sterba if (!btrfs_end_io_wq_cache) 9197eb6b69SDavid Sterba return -ENOMEM; 9297eb6b69SDavid Sterba return 0; 9397eb6b69SDavid Sterba } 9497eb6b69SDavid Sterba 95e67c718bSDavid Sterba void __cold btrfs_end_io_wq_exit(void) 9697eb6b69SDavid Sterba { 9797eb6b69SDavid Sterba kmem_cache_destroy(btrfs_end_io_wq_cache); 9897eb6b69SDavid Sterba } 9997eb6b69SDavid Sterba 100d352ac68SChris Mason /* 101d352ac68SChris Mason * async submit bios are used to offload expensive checksumming 102d352ac68SChris Mason * onto the worker threads. They checksum file and metadata bios 103d352ac68SChris Mason * just before they are sent down the IO stack. 104d352ac68SChris Mason */ 10544b8bd7eSChris Mason struct async_submit_bio { 106c6100a4bSJosef Bacik void *private_data; 10744b8bd7eSChris Mason struct bio *bio; 108a758781dSDavid Sterba extent_submit_bio_start_t *submit_bio_start; 10944b8bd7eSChris Mason int mirror_num; 110eaf25d93SChris Mason /* 111eaf25d93SChris Mason * bio_offset is optional, can be used if the pages in the bio 112eaf25d93SChris Mason * can't tell us where in the file the bio should go 113eaf25d93SChris Mason */ 114eaf25d93SChris Mason u64 bio_offset; 1158b712842SChris Mason struct btrfs_work work; 1164e4cbee9SChristoph Hellwig blk_status_t status; 11744b8bd7eSChris Mason }; 11844b8bd7eSChris Mason 11985d4e461SChris Mason /* 12085d4e461SChris Mason * Lockdep class keys for extent_buffer->lock's in this root. For a given 12185d4e461SChris Mason * eb, the lockdep key is determined by the btrfs_root it belongs to and 12285d4e461SChris Mason * the level the eb occupies in the tree. 1234008c04aSChris Mason * 12485d4e461SChris Mason * Different roots are used for different purposes and may nest inside each 12585d4e461SChris Mason * other and they require separate keysets. As lockdep keys should be 12685d4e461SChris Mason * static, assign keysets according to the purpose of the root as indicated 1274fd786e6SMisono Tomohiro * by btrfs_root->root_key.objectid. This ensures that all special purpose 1284fd786e6SMisono Tomohiro * roots have separate keysets. 1294008c04aSChris Mason * 13085d4e461SChris Mason * Lock-nesting across peer nodes is always done with the immediate parent 13185d4e461SChris Mason * node locked thus preventing deadlock. As lockdep doesn't know this, use 13285d4e461SChris Mason * subclass to avoid triggering lockdep warning in such cases. 1334008c04aSChris Mason * 13485d4e461SChris Mason * The key is set by the readpage_end_io_hook after the buffer has passed 13585d4e461SChris Mason * csum validation but before the pages are unlocked. It is also set by 13685d4e461SChris Mason * btrfs_init_new_buffer on freshly allocated blocks. 13785d4e461SChris Mason * 13885d4e461SChris Mason * We also add a check to make sure the highest level of the tree is the 13985d4e461SChris Mason * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code 14085d4e461SChris Mason * needs update as well. 1414008c04aSChris Mason */ 1424008c04aSChris Mason #ifdef CONFIG_DEBUG_LOCK_ALLOC 1434008c04aSChris Mason # if BTRFS_MAX_LEVEL != 8 1444008c04aSChris Mason # error 1454008c04aSChris Mason # endif 14685d4e461SChris Mason 14785d4e461SChris Mason static struct btrfs_lockdep_keyset { 14885d4e461SChris Mason u64 id; /* root objectid */ 14985d4e461SChris Mason const char *name_stem; /* lock name stem */ 15085d4e461SChris Mason char names[BTRFS_MAX_LEVEL + 1][20]; 15185d4e461SChris Mason struct lock_class_key keys[BTRFS_MAX_LEVEL + 1]; 15285d4e461SChris Mason } btrfs_lockdep_keysets[] = { 15385d4e461SChris Mason { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" }, 15485d4e461SChris Mason { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" }, 15585d4e461SChris Mason { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" }, 15685d4e461SChris Mason { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" }, 15785d4e461SChris Mason { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" }, 15885d4e461SChris Mason { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" }, 15960b62978SDavid Sterba { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" }, 16085d4e461SChris Mason { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" }, 16185d4e461SChris Mason { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" }, 16285d4e461SChris Mason { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" }, 16313fd8da9SDavid Sterba { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" }, 1646b20e0adSDavid Sterba { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" }, 16585d4e461SChris Mason { .id = 0, .name_stem = "tree" }, 1664008c04aSChris Mason }; 16785d4e461SChris Mason 16885d4e461SChris Mason void __init btrfs_init_lockdep(void) 16985d4e461SChris Mason { 17085d4e461SChris Mason int i, j; 17185d4e461SChris Mason 17285d4e461SChris Mason /* initialize lockdep class names */ 17385d4e461SChris Mason for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) { 17485d4e461SChris Mason struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i]; 17585d4e461SChris Mason 17685d4e461SChris Mason for (j = 0; j < ARRAY_SIZE(ks->names); j++) 17785d4e461SChris Mason snprintf(ks->names[j], sizeof(ks->names[j]), 17885d4e461SChris Mason "btrfs-%s-%02d", ks->name_stem, j); 17985d4e461SChris Mason } 18085d4e461SChris Mason } 18185d4e461SChris Mason 18285d4e461SChris Mason void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb, 18385d4e461SChris Mason int level) 18485d4e461SChris Mason { 18585d4e461SChris Mason struct btrfs_lockdep_keyset *ks; 18685d4e461SChris Mason 18785d4e461SChris Mason BUG_ON(level >= ARRAY_SIZE(ks->keys)); 18885d4e461SChris Mason 18985d4e461SChris Mason /* find the matching keyset, id 0 is the default entry */ 19085d4e461SChris Mason for (ks = btrfs_lockdep_keysets; ks->id; ks++) 19185d4e461SChris Mason if (ks->id == objectid) 19285d4e461SChris Mason break; 19385d4e461SChris Mason 19485d4e461SChris Mason lockdep_set_class_and_name(&eb->lock, 19585d4e461SChris Mason &ks->keys[level], ks->names[level]); 19685d4e461SChris Mason } 19785d4e461SChris Mason 1984008c04aSChris Mason #endif 1994008c04aSChris Mason 200d352ac68SChris Mason /* 201d352ac68SChris Mason * extents on the btree inode are pretty simple, there's one extent 202d352ac68SChris Mason * that covers the entire device 203d352ac68SChris Mason */ 2046af49dbdSDavid Sterba struct extent_map *btree_get_extent(struct btrfs_inode *inode, 205306e16ceSDavid Sterba struct page *page, size_t pg_offset, u64 start, u64 len, 2065f39d397SChris Mason int create) 2075f39d397SChris Mason { 2083ffbd68cSDavid Sterba struct btrfs_fs_info *fs_info = inode->root->fs_info; 209fc4f21b1SNikolay Borisov struct extent_map_tree *em_tree = &inode->extent_tree; 2105f39d397SChris Mason struct extent_map *em; 2115f39d397SChris Mason int ret; 2125f39d397SChris Mason 213890871beSChris Mason read_lock(&em_tree->lock); 214d1310b2eSChris Mason em = lookup_extent_mapping(em_tree, start, len); 215a061fc8dSChris Mason if (em) { 2160b246afaSJeff Mahoney em->bdev = fs_info->fs_devices->latest_bdev; 217890871beSChris Mason read_unlock(&em_tree->lock); 2185f39d397SChris Mason goto out; 219a061fc8dSChris Mason } 220890871beSChris Mason read_unlock(&em_tree->lock); 2217b13b7b1SChris Mason 222172ddd60SDavid Sterba em = alloc_extent_map(); 2235f39d397SChris Mason if (!em) { 2245f39d397SChris Mason em = ERR_PTR(-ENOMEM); 2255f39d397SChris Mason goto out; 2265f39d397SChris Mason } 2275f39d397SChris Mason em->start = 0; 2280afbaf8cSChris Mason em->len = (u64)-1; 229c8b97818SChris Mason em->block_len = (u64)-1; 2305f39d397SChris Mason em->block_start = 0; 2310b246afaSJeff Mahoney em->bdev = fs_info->fs_devices->latest_bdev; 232d1310b2eSChris Mason 233890871beSChris Mason write_lock(&em_tree->lock); 23409a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 2355f39d397SChris Mason if (ret == -EEXIST) { 2365f39d397SChris Mason free_extent_map(em); 2377b13b7b1SChris Mason em = lookup_extent_mapping(em_tree, start, len); 238b4f359abSTsutomu Itoh if (!em) 2390433f20dSTsutomu Itoh em = ERR_PTR(-EIO); 2405f39d397SChris Mason } else if (ret) { 2417b13b7b1SChris Mason free_extent_map(em); 2420433f20dSTsutomu Itoh em = ERR_PTR(ret); 2435f39d397SChris Mason } 244890871beSChris Mason write_unlock(&em_tree->lock); 2457b13b7b1SChris Mason 2465f39d397SChris Mason out: 2475f39d397SChris Mason return em; 2485f39d397SChris Mason } 2495f39d397SChris Mason 250d352ac68SChris Mason /* 2512996e1f8SJohannes Thumshirn * Compute the csum of a btree block and store the result to provided buffer. 2522996e1f8SJohannes Thumshirn * 2532996e1f8SJohannes Thumshirn * Returns error if the extent buffer cannot be mapped. 254d352ac68SChris Mason */ 2552996e1f8SJohannes Thumshirn static int csum_tree_block(struct extent_buffer *buf, u8 *result) 25619c00ddcSChris Mason { 257d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = buf->fs_info; 258d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 25919c00ddcSChris Mason unsigned long len; 26019c00ddcSChris Mason unsigned long cur_len; 26119c00ddcSChris Mason unsigned long offset = BTRFS_CSUM_SIZE; 26219c00ddcSChris Mason char *kaddr; 26319c00ddcSChris Mason unsigned long map_start; 26419c00ddcSChris Mason unsigned long map_len; 26519c00ddcSChris Mason int err; 266d5178578SJohannes Thumshirn 267d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 268d5178578SJohannes Thumshirn crypto_shash_init(shash); 26919c00ddcSChris Mason 27019c00ddcSChris Mason len = buf->len - offset; 271d5178578SJohannes Thumshirn 27219c00ddcSChris Mason while (len > 0) { 273d2e174d5SJohannes Thumshirn /* 274d2e174d5SJohannes Thumshirn * Note: we don't need to check for the err == 1 case here, as 275d2e174d5SJohannes Thumshirn * with the given combination of 'start = BTRFS_CSUM_SIZE (32)' 276d2e174d5SJohannes Thumshirn * and 'min_len = 32' and the currently implemented mapping 277d2e174d5SJohannes Thumshirn * algorithm we cannot cross a page boundary. 278d2e174d5SJohannes Thumshirn */ 27919c00ddcSChris Mason err = map_private_extent_buffer(buf, offset, 32, 280a6591715SChris Mason &kaddr, &map_start, &map_len); 281c53839fcSJohannes Thumshirn if (WARN_ON(err)) 2828bd98f0eSAlex Lyakas return err; 28319c00ddcSChris Mason cur_len = min(len, map_len - (offset - map_start)); 284d5178578SJohannes Thumshirn crypto_shash_update(shash, kaddr + offset - map_start, cur_len); 28519c00ddcSChris Mason len -= cur_len; 28619c00ddcSChris Mason offset += cur_len; 28719c00ddcSChris Mason } 28871a63551SDavid Sterba memset(result, 0, BTRFS_CSUM_SIZE); 289607d432dSJosef Bacik 290d5178578SJohannes Thumshirn crypto_shash_final(shash, result); 29119c00ddcSChris Mason 29219c00ddcSChris Mason return 0; 29319c00ddcSChris Mason } 29419c00ddcSChris Mason 295d352ac68SChris Mason /* 296d352ac68SChris Mason * we can't consider a given block up to date unless the transid of the 297d352ac68SChris Mason * block matches the transid in the parent node's pointer. This is how we 298d352ac68SChris Mason * detect blocks that either didn't get written at all or got written 299d352ac68SChris Mason * in the wrong place. 300d352ac68SChris Mason */ 3011259ab75SChris Mason static int verify_parent_transid(struct extent_io_tree *io_tree, 302b9fab919SChris Mason struct extent_buffer *eb, u64 parent_transid, 303b9fab919SChris Mason int atomic) 3041259ab75SChris Mason { 3052ac55d41SJosef Bacik struct extent_state *cached_state = NULL; 3061259ab75SChris Mason int ret; 3072755a0deSDavid Sterba bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB); 3081259ab75SChris Mason 3091259ab75SChris Mason if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 3101259ab75SChris Mason return 0; 3111259ab75SChris Mason 312b9fab919SChris Mason if (atomic) 313b9fab919SChris Mason return -EAGAIN; 314b9fab919SChris Mason 315a26e8c9fSJosef Bacik if (need_lock) { 316a26e8c9fSJosef Bacik btrfs_tree_read_lock(eb); 317300aa896SDavid Sterba btrfs_set_lock_blocking_read(eb); 318a26e8c9fSJosef Bacik } 319a26e8c9fSJosef Bacik 3202ac55d41SJosef Bacik lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 321ff13db41SDavid Sterba &cached_state); 3220b32f4bbSJosef Bacik if (extent_buffer_uptodate(eb) && 3231259ab75SChris Mason btrfs_header_generation(eb) == parent_transid) { 3241259ab75SChris Mason ret = 0; 3251259ab75SChris Mason goto out; 3261259ab75SChris Mason } 32794647322SDavid Sterba btrfs_err_rl(eb->fs_info, 32894647322SDavid Sterba "parent transid verify failed on %llu wanted %llu found %llu", 32994647322SDavid Sterba eb->start, 33029549aecSWang Shilong parent_transid, btrfs_header_generation(eb)); 3311259ab75SChris Mason ret = 1; 332a26e8c9fSJosef Bacik 333a26e8c9fSJosef Bacik /* 334a26e8c9fSJosef Bacik * Things reading via commit roots that don't have normal protection, 335a26e8c9fSJosef Bacik * like send, can have a really old block in cache that may point at a 33601327610SNicholas D Steeves * block that has been freed and re-allocated. So don't clear uptodate 337a26e8c9fSJosef Bacik * if we find an eb that is under IO (dirty/writeback) because we could 338a26e8c9fSJosef Bacik * end up reading in the stale data and then writing it back out and 339a26e8c9fSJosef Bacik * making everybody very sad. 340a26e8c9fSJosef Bacik */ 341a26e8c9fSJosef Bacik if (!extent_buffer_under_io(eb)) 3420b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 34333958dc6SChris Mason out: 3442ac55d41SJosef Bacik unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 345e43bbe5eSDavid Sterba &cached_state); 346472b909fSJosef Bacik if (need_lock) 347a26e8c9fSJosef Bacik btrfs_tree_read_unlock_blocking(eb); 3481259ab75SChris Mason return ret; 3491259ab75SChris Mason } 3501259ab75SChris Mason 351e7e16f48SJohannes Thumshirn static bool btrfs_supported_super_csum(u16 csum_type) 352e7e16f48SJohannes Thumshirn { 353e7e16f48SJohannes Thumshirn switch (csum_type) { 354e7e16f48SJohannes Thumshirn case BTRFS_CSUM_TYPE_CRC32: 3553951e7f0SJohannes Thumshirn case BTRFS_CSUM_TYPE_XXHASH: 3563831bf00SJohannes Thumshirn case BTRFS_CSUM_TYPE_SHA256: 357e7e16f48SJohannes Thumshirn return true; 358e7e16f48SJohannes Thumshirn default: 359e7e16f48SJohannes Thumshirn return false; 360e7e16f48SJohannes Thumshirn } 361e7e16f48SJohannes Thumshirn } 362e7e16f48SJohannes Thumshirn 363d352ac68SChris Mason /* 3641104a885SDavid Sterba * Return 0 if the superblock checksum type matches the checksum value of that 3651104a885SDavid Sterba * algorithm. Pass the raw disk superblock data. 3661104a885SDavid Sterba */ 367ab8d0fc4SJeff Mahoney static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info, 368ab8d0fc4SJeff Mahoney char *raw_disk_sb) 3691104a885SDavid Sterba { 3701104a885SDavid Sterba struct btrfs_super_block *disk_sb = 3711104a885SDavid Sterba (struct btrfs_super_block *)raw_disk_sb; 37251bce6c9SJohannes Thumshirn char result[BTRFS_CSUM_SIZE]; 373d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 374d5178578SJohannes Thumshirn 375d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 376d5178578SJohannes Thumshirn crypto_shash_init(shash); 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 */ 383d5178578SJohannes Thumshirn crypto_shash_update(shash, raw_disk_sb + BTRFS_CSUM_SIZE, 384d5178578SJohannes Thumshirn BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE); 385d5178578SJohannes Thumshirn crypto_shash_final(shash, 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; 42262fdaa52SQu Wenruo 42362fdaa52SQu Wenruo /* We have @first_key, so this @eb must have at least one item */ 42462fdaa52SQu Wenruo if (btrfs_header_nritems(eb) == 0) { 42562fdaa52SQu Wenruo btrfs_err(fs_info, 42662fdaa52SQu Wenruo "invalid tree nritems, bytenr=%llu nritems=0 expect >0", 42762fdaa52SQu Wenruo eb->start); 42862fdaa52SQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 42962fdaa52SQu Wenruo return -EUCLEAN; 43062fdaa52SQu Wenruo } 43162fdaa52SQu Wenruo 432581c1760SQu Wenruo if (found_level) 433581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &found_key, 0); 434581c1760SQu Wenruo else 435581c1760SQu Wenruo btrfs_item_key_to_cpu(eb, &found_key, 0); 436581c1760SQu Wenruo ret = btrfs_comp_cpu_keys(first_key, &found_key); 437581c1760SQu Wenruo 438581c1760SQu Wenruo if (ret) { 43963489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 44063489055SQu Wenruo KERN_ERR "BTRFS: tree first key check failed\n"); 441581c1760SQu Wenruo btrfs_err(fs_info, 442ff76a864SLiu Bo "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", 443ff76a864SLiu Bo eb->start, parent_transid, first_key->objectid, 444ff76a864SLiu Bo first_key->type, first_key->offset, 445ff76a864SLiu Bo found_key.objectid, found_key.type, 446ff76a864SLiu Bo found_key.offset); 447581c1760SQu Wenruo } 448581c1760SQu Wenruo return ret; 449581c1760SQu Wenruo } 450581c1760SQu Wenruo 4511104a885SDavid Sterba /* 452d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 453d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 454581c1760SQu Wenruo * 455581c1760SQu Wenruo * @parent_transid: expected transid, skip check if 0 456581c1760SQu Wenruo * @level: expected level, mandatory check 457581c1760SQu Wenruo * @first_key: expected key of first slot, skip check if NULL 458d352ac68SChris Mason */ 4595ab12d1fSDavid Sterba static int btree_read_extent_buffer_pages(struct extent_buffer *eb, 460581c1760SQu Wenruo u64 parent_transid, int level, 461581c1760SQu Wenruo struct btrfs_key *first_key) 462f188591eSChris Mason { 4635ab12d1fSDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 464f188591eSChris Mason struct extent_io_tree *io_tree; 465ea466794SJosef Bacik int failed = 0; 466f188591eSChris Mason int ret; 467f188591eSChris Mason int num_copies = 0; 468f188591eSChris Mason int mirror_num = 0; 469ea466794SJosef Bacik int failed_mirror = 0; 470f188591eSChris Mason 4710b246afaSJeff Mahoney io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; 472f188591eSChris Mason while (1) { 473f8397d69SNikolay Borisov clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 474c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num); 475256dd1bbSStefan Behrens if (!ret) { 476581c1760SQu Wenruo if (verify_parent_transid(io_tree, eb, 477b9fab919SChris Mason parent_transid, 0)) 478256dd1bbSStefan Behrens ret = -EIO; 479e064d5e9SDavid Sterba else if (btrfs_verify_level_key(eb, level, 480ff76a864SLiu Bo first_key, parent_transid)) 481581c1760SQu Wenruo ret = -EUCLEAN; 482581c1760SQu Wenruo else 483581c1760SQu Wenruo break; 484256dd1bbSStefan Behrens } 485d397712bSChris Mason 4860b246afaSJeff Mahoney num_copies = btrfs_num_copies(fs_info, 487f188591eSChris Mason eb->start, eb->len); 4884235298eSChris Mason if (num_copies == 1) 489ea466794SJosef Bacik break; 4904235298eSChris Mason 4915cf1ab56SJosef Bacik if (!failed_mirror) { 4925cf1ab56SJosef Bacik failed = 1; 4935cf1ab56SJosef Bacik failed_mirror = eb->read_mirror; 4945cf1ab56SJosef Bacik } 4955cf1ab56SJosef Bacik 496f188591eSChris Mason mirror_num++; 497ea466794SJosef Bacik if (mirror_num == failed_mirror) 498ea466794SJosef Bacik mirror_num++; 499ea466794SJosef Bacik 5004235298eSChris Mason if (mirror_num > num_copies) 501ea466794SJosef Bacik break; 502f188591eSChris Mason } 503ea466794SJosef Bacik 504c0901581SStefan Behrens if (failed && !ret && failed_mirror) 50520a1fbf9SDavid Sterba btrfs_repair_eb_io_failure(eb, failed_mirror); 506ea466794SJosef Bacik 507ea466794SJosef Bacik return ret; 508f188591eSChris Mason } 50919c00ddcSChris Mason 510d352ac68SChris Mason /* 511d397712bSChris Mason * checksum a dirty tree block before IO. This has extra checks to make sure 512d397712bSChris Mason * we only fill in the checksum field in the first page of a multi-page block 513d352ac68SChris Mason */ 514d397712bSChris Mason 51501d58472SDaniel Dressler static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page) 51619c00ddcSChris Mason { 5174eee4fa4SMiao Xie u64 start = page_offset(page); 51819c00ddcSChris Mason u64 found_start; 5192996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 5202996e1f8SJohannes Thumshirn u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); 52119c00ddcSChris Mason struct extent_buffer *eb; 5228d47a0d8SQu Wenruo int ret; 523f188591eSChris Mason 5244f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 5254f2de97aSJosef Bacik if (page != eb->pages[0]) 5264f2de97aSJosef Bacik return 0; 5270f805531SAlex Lyakas 52819c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 5290f805531SAlex Lyakas /* 5300f805531SAlex Lyakas * Please do not consolidate these warnings into a single if. 5310f805531SAlex Lyakas * It is useful to know what went wrong. 5320f805531SAlex Lyakas */ 5330f805531SAlex Lyakas if (WARN_ON(found_start != start)) 5340f805531SAlex Lyakas return -EUCLEAN; 5350f805531SAlex Lyakas if (WARN_ON(!PageUptodate(page))) 5360f805531SAlex Lyakas return -EUCLEAN; 5370f805531SAlex Lyakas 538de37aa51SNikolay Borisov ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid, 5390f805531SAlex Lyakas btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0); 5400f805531SAlex Lyakas 5412996e1f8SJohannes Thumshirn if (csum_tree_block(eb, result)) 5422996e1f8SJohannes Thumshirn return -EINVAL; 5432996e1f8SJohannes Thumshirn 5448d47a0d8SQu Wenruo if (btrfs_header_level(eb)) 5458d47a0d8SQu Wenruo ret = btrfs_check_node(eb); 5468d47a0d8SQu Wenruo else 5478d47a0d8SQu Wenruo ret = btrfs_check_leaf_full(eb); 5488d47a0d8SQu Wenruo 5498d47a0d8SQu Wenruo if (ret < 0) { 550c06631b0SQu Wenruo btrfs_print_tree(eb, 0); 5518d47a0d8SQu Wenruo btrfs_err(fs_info, 5528d47a0d8SQu Wenruo "block=%llu write time tree block corruption detected", 5538d47a0d8SQu Wenruo eb->start); 554c06631b0SQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 5558d47a0d8SQu Wenruo return ret; 5568d47a0d8SQu Wenruo } 5572996e1f8SJohannes Thumshirn write_extent_buffer(eb, result, 0, csum_size); 5588d47a0d8SQu Wenruo 5592996e1f8SJohannes Thumshirn return 0; 56019c00ddcSChris Mason } 56119c00ddcSChris Mason 562b0c9b3b0SDavid Sterba static int check_tree_block_fsid(struct extent_buffer *eb) 5632b82032cSYan Zheng { 564b0c9b3b0SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 56501d58472SDaniel Dressler struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 56644880fdcSAnand Jain u8 fsid[BTRFS_FSID_SIZE]; 5672b82032cSYan Zheng int ret = 1; 5682b82032cSYan Zheng 5690a4e5586SRoss Kirk read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE); 5702b82032cSYan Zheng while (fs_devices) { 5717239ff4bSNikolay Borisov u8 *metadata_uuid; 5727239ff4bSNikolay Borisov 5737239ff4bSNikolay Borisov /* 5747239ff4bSNikolay Borisov * Checking the incompat flag is only valid for the current 5757239ff4bSNikolay Borisov * fs. For seed devices it's forbidden to have their uuid 5767239ff4bSNikolay Borisov * changed so reading ->fsid in this case is fine 5777239ff4bSNikolay Borisov */ 5787239ff4bSNikolay Borisov if (fs_devices == fs_info->fs_devices && 5797239ff4bSNikolay Borisov btrfs_fs_incompat(fs_info, METADATA_UUID)) 5807239ff4bSNikolay Borisov metadata_uuid = fs_devices->metadata_uuid; 5817239ff4bSNikolay Borisov else 5827239ff4bSNikolay Borisov metadata_uuid = fs_devices->fsid; 5837239ff4bSNikolay Borisov 5847239ff4bSNikolay Borisov if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) { 5852b82032cSYan Zheng ret = 0; 5862b82032cSYan Zheng break; 5872b82032cSYan Zheng } 5882b82032cSYan Zheng fs_devices = fs_devices->seed; 5892b82032cSYan Zheng } 5902b82032cSYan Zheng return ret; 5912b82032cSYan Zheng } 5922b82032cSYan Zheng 593facc8a22SMiao Xie static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio, 594facc8a22SMiao Xie u64 phy_offset, struct page *page, 595facc8a22SMiao Xie u64 start, u64 end, int mirror) 596ce9adaa5SChris Mason { 597ce9adaa5SChris Mason u64 found_start; 598ce9adaa5SChris Mason int found_level; 599ce9adaa5SChris Mason struct extent_buffer *eb; 600ce9adaa5SChris Mason struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 60102873e43SZhao Lei struct btrfs_fs_info *fs_info = root->fs_info; 6022996e1f8SJohannes Thumshirn u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); 603f188591eSChris Mason int ret = 0; 6042996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 605727011e0SChris Mason int reads_done; 606ce9adaa5SChris Mason 607ce9adaa5SChris Mason if (!page->private) 608ce9adaa5SChris Mason goto out; 609d397712bSChris Mason 6104f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 611d397712bSChris Mason 6120b32f4bbSJosef Bacik /* the pending IO might have been the only thing that kept this buffer 6130b32f4bbSJosef Bacik * in memory. Make sure we have a ref for all this other checks 6140b32f4bbSJosef Bacik */ 61567439dadSDavid Sterba atomic_inc(&eb->refs); 6160b32f4bbSJosef Bacik 6170b32f4bbSJosef Bacik reads_done = atomic_dec_and_test(&eb->io_pages); 618727011e0SChris Mason if (!reads_done) 619727011e0SChris Mason goto err; 620f188591eSChris Mason 6215cf1ab56SJosef Bacik eb->read_mirror = mirror; 622656f30dbSFilipe Manana if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 623ea466794SJosef Bacik ret = -EIO; 624ea466794SJosef Bacik goto err; 625ea466794SJosef Bacik } 626ea466794SJosef Bacik 627ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 628727011e0SChris Mason if (found_start != eb->start) { 629893bf4b1SSu Yue btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu", 630893bf4b1SSu Yue eb->start, found_start); 631f188591eSChris Mason ret = -EIO; 632ce9adaa5SChris Mason goto err; 633ce9adaa5SChris Mason } 634b0c9b3b0SDavid Sterba if (check_tree_block_fsid(eb)) { 63502873e43SZhao Lei btrfs_err_rl(fs_info, "bad fsid on block %llu", 63694647322SDavid Sterba eb->start); 6371259ab75SChris Mason ret = -EIO; 6381259ab75SChris Mason goto err; 6391259ab75SChris Mason } 640ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 6411c24c3ceSJosef Bacik if (found_level >= BTRFS_MAX_LEVEL) { 642893bf4b1SSu Yue btrfs_err(fs_info, "bad tree block level %d on %llu", 643893bf4b1SSu Yue (int)btrfs_header_level(eb), eb->start); 6441c24c3ceSJosef Bacik ret = -EIO; 6451c24c3ceSJosef Bacik goto err; 6461c24c3ceSJosef Bacik } 647ce9adaa5SChris Mason 64885d4e461SChris Mason btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), 64985d4e461SChris Mason eb, found_level); 6504008c04aSChris Mason 6512996e1f8SJohannes Thumshirn ret = csum_tree_block(eb, result); 6528bd98f0eSAlex Lyakas if (ret) 653a826d6dcSJosef Bacik goto err; 654a826d6dcSJosef Bacik 6552996e1f8SJohannes Thumshirn if (memcmp_extent_buffer(eb, result, 0, csum_size)) { 6562996e1f8SJohannes Thumshirn u32 val; 6572996e1f8SJohannes Thumshirn u32 found = 0; 6582996e1f8SJohannes Thumshirn 6592996e1f8SJohannes Thumshirn memcpy(&found, result, csum_size); 6602996e1f8SJohannes Thumshirn 6612996e1f8SJohannes Thumshirn read_extent_buffer(eb, &val, 0, csum_size); 6622996e1f8SJohannes Thumshirn btrfs_warn_rl(fs_info, 6632996e1f8SJohannes Thumshirn "%s checksum verify failed on %llu wanted %x found %x level %d", 6642996e1f8SJohannes Thumshirn fs_info->sb->s_id, eb->start, 6652996e1f8SJohannes Thumshirn val, found, btrfs_header_level(eb)); 6662996e1f8SJohannes Thumshirn ret = -EUCLEAN; 6672996e1f8SJohannes Thumshirn goto err; 6682996e1f8SJohannes Thumshirn } 6692996e1f8SJohannes Thumshirn 670a826d6dcSJosef Bacik /* 671a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 672a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 673a826d6dcSJosef Bacik * return -EIO. 674a826d6dcSJosef Bacik */ 6751c4360eeSDavid Sterba if (found_level == 0 && btrfs_check_leaf_full(eb)) { 676a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 677a826d6dcSJosef Bacik ret = -EIO; 678a826d6dcSJosef Bacik } 679ce9adaa5SChris Mason 680813fd1dcSDavid Sterba if (found_level > 0 && btrfs_check_node(eb)) 681053ab70fSLiu Bo ret = -EIO; 682053ab70fSLiu Bo 6830b32f4bbSJosef Bacik if (!ret) 6840b32f4bbSJosef Bacik set_extent_buffer_uptodate(eb); 68575391f0dSQu Wenruo else 68675391f0dSQu Wenruo btrfs_err(fs_info, 68775391f0dSQu Wenruo "block=%llu read time tree block corruption detected", 68875391f0dSQu Wenruo eb->start); 689ce9adaa5SChris Mason err: 69079fb65a1SJosef Bacik if (reads_done && 69179fb65a1SJosef Bacik test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 692d48d71aaSDavid Sterba btree_readahead_hook(eb, ret); 6934bb31e92SArne Jansen 69453b381b3SDavid Woodhouse if (ret) { 69553b381b3SDavid Woodhouse /* 69653b381b3SDavid Woodhouse * our io error hook is going to dec the io pages 69753b381b3SDavid Woodhouse * again, we have to make sure it has something 69853b381b3SDavid Woodhouse * to decrement 69953b381b3SDavid Woodhouse */ 70053b381b3SDavid Woodhouse atomic_inc(&eb->io_pages); 7010b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 70253b381b3SDavid Woodhouse } 7030b32f4bbSJosef Bacik free_extent_buffer(eb); 704ce9adaa5SChris Mason out: 705f188591eSChris Mason return ret; 706ce9adaa5SChris Mason } 707ce9adaa5SChris Mason 7084246a0b6SChristoph Hellwig static void end_workqueue_bio(struct bio *bio) 709ce9adaa5SChris Mason { 71097eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq = bio->bi_private; 711ce9adaa5SChris Mason struct btrfs_fs_info *fs_info; 7129e0af237SLiu Bo struct btrfs_workqueue *wq; 713ce9adaa5SChris Mason 714ce9adaa5SChris Mason fs_info = end_io_wq->info; 7154e4cbee9SChristoph Hellwig end_io_wq->status = bio->bi_status; 716d20f7043SChris Mason 71737226b21SMike Christie if (bio_op(bio) == REQ_OP_WRITE) { 718a0cac0ecSOmar Sandoval if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) 7199e0af237SLiu Bo wq = fs_info->endio_meta_write_workers; 720a0cac0ecSOmar Sandoval else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) 7219e0af237SLiu Bo wq = fs_info->endio_freespace_worker; 722a0cac0ecSOmar Sandoval else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 7239e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 724a0cac0ecSOmar Sandoval else 7259e0af237SLiu Bo wq = fs_info->endio_write_workers; 7269e0af237SLiu Bo } else { 727a0cac0ecSOmar Sandoval if (unlikely(end_io_wq->metadata == BTRFS_WQ_ENDIO_DIO_REPAIR)) 7288b110e39SMiao Xie wq = fs_info->endio_repair_workers; 729a0cac0ecSOmar Sandoval else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 7309e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 731a0cac0ecSOmar Sandoval else if (end_io_wq->metadata) 7329e0af237SLiu Bo wq = fs_info->endio_meta_workers; 733a0cac0ecSOmar Sandoval else 7349e0af237SLiu Bo wq = fs_info->endio_workers; 7359e0af237SLiu Bo } 7369e0af237SLiu Bo 737a0cac0ecSOmar Sandoval btrfs_init_work(&end_io_wq->work, 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 798ec39f769SChris Mason /* 799ec39f769SChris Mason * All of the bios that pass through here are from async helpers. 800ec39f769SChris Mason * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context. 801ec39f769SChris Mason * This changes nothing when cgroups aren't in use. 802ec39f769SChris Mason */ 803ec39f769SChris Mason async->bio->bi_opf |= REQ_CGROUP_PUNT; 80408635baeSChris Mason ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num); 80506ea01b1SDavid Sterba if (ret) { 80606ea01b1SDavid Sterba async->bio->bi_status = ret; 80706ea01b1SDavid Sterba bio_endio(async->bio); 80806ea01b1SDavid Sterba } 8094a69a410SChris Mason } 8104a69a410SChris Mason 8114a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 8124a69a410SChris Mason { 8134a69a410SChris Mason struct async_submit_bio *async; 8144a69a410SChris Mason 8154a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 8168b712842SChris Mason kfree(async); 8178b712842SChris Mason } 8188b712842SChris Mason 8198c27cb35SLinus Torvalds blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio, 820c6100a4bSJosef Bacik int mirror_num, unsigned long bio_flags, 821c6100a4bSJosef Bacik u64 bio_offset, void *private_data, 822e288c080SDavid Sterba extent_submit_bio_start_t *submit_bio_start) 82344b8bd7eSChris Mason { 82444b8bd7eSChris Mason struct async_submit_bio *async; 82544b8bd7eSChris Mason 82644b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 82744b8bd7eSChris Mason if (!async) 8284e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 82944b8bd7eSChris Mason 830c6100a4bSJosef Bacik async->private_data = private_data; 83144b8bd7eSChris Mason async->bio = bio; 83244b8bd7eSChris Mason async->mirror_num = mirror_num; 8334a69a410SChris Mason async->submit_bio_start = submit_bio_start; 8344a69a410SChris Mason 835a0cac0ecSOmar Sandoval btrfs_init_work(&async->work, run_one_async_start, run_one_async_done, 836a0cac0ecSOmar Sandoval run_one_async_free); 8374a69a410SChris Mason 838eaf25d93SChris Mason async->bio_offset = bio_offset; 8398c8bee1dSChris Mason 8404e4cbee9SChristoph Hellwig async->status = 0; 84179787eaaSJeff Mahoney 84267f055c7SChristoph Hellwig if (op_is_sync(bio->bi_opf)) 8435cdc7ad3SQu Wenruo btrfs_set_work_high_priority(&async->work); 844d313d7a3SChris Mason 8455cdc7ad3SQu Wenruo btrfs_queue_work(fs_info->workers, &async->work); 84644b8bd7eSChris Mason return 0; 84744b8bd7eSChris Mason } 84844b8bd7eSChris Mason 8494e4cbee9SChristoph Hellwig static blk_status_t btree_csum_one_bio(struct bio *bio) 850ce3ed71aSChris Mason { 8512c30c71bSKent Overstreet struct bio_vec *bvec; 852ce3ed71aSChris Mason struct btrfs_root *root; 8532b070cfeSChristoph Hellwig int ret = 0; 8546dc4f100SMing Lei struct bvec_iter_all iter_all; 855ce3ed71aSChris Mason 856c09abff8SDavid Sterba ASSERT(!bio_flagged(bio, BIO_CLONED)); 8572b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 858ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 85901d58472SDaniel Dressler ret = csum_dirty_buffer(root->fs_info, bvec->bv_page); 86079787eaaSJeff Mahoney if (ret) 86179787eaaSJeff Mahoney break; 862ce3ed71aSChris Mason } 8632c30c71bSKent Overstreet 8644e4cbee9SChristoph Hellwig return errno_to_blk_status(ret); 865ce3ed71aSChris Mason } 866ce3ed71aSChris Mason 867d0ee3934SDavid Sterba static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio, 868eaf25d93SChris Mason u64 bio_offset) 86922c59948SChris Mason { 8708b712842SChris Mason /* 8718b712842SChris Mason * when we're called for a write, we're already in the async 8725443be45SChris Mason * submission context. Just jump into btrfs_map_bio 8738b712842SChris Mason */ 87479787eaaSJeff Mahoney return btree_csum_one_bio(bio); 87522c59948SChris Mason } 87622c59948SChris Mason 8779b4e675aSDavid Sterba static int check_async_write(struct btrfs_fs_info *fs_info, 8789b4e675aSDavid Sterba struct btrfs_inode *bi) 879de0022b9SJosef Bacik { 8806300463bSLiu Bo if (atomic_read(&bi->sync_writers)) 8816300463bSLiu Bo return 0; 8829b4e675aSDavid Sterba if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags)) 883de0022b9SJosef Bacik return 0; 884de0022b9SJosef Bacik return 1; 885de0022b9SJosef Bacik } 886de0022b9SJosef Bacik 887a56b1c7bSNikolay Borisov static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio, 88850489a57SNikolay Borisov int mirror_num, 88950489a57SNikolay Borisov unsigned long bio_flags) 89044b8bd7eSChris Mason { 8910b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 8929b4e675aSDavid Sterba int async = check_async_write(fs_info, BTRFS_I(inode)); 8934e4cbee9SChristoph Hellwig blk_status_t ret; 894cad321adSChris Mason 89537226b21SMike Christie if (bio_op(bio) != REQ_OP_WRITE) { 896cad321adSChris Mason /* 897cad321adSChris Mason * called for a read, do the setup so that checksum validation 898cad321adSChris Mason * can happen in the async kernel threads 899cad321adSChris Mason */ 9000b246afaSJeff Mahoney ret = btrfs_bio_wq_end_io(fs_info, bio, 9010b246afaSJeff Mahoney BTRFS_WQ_ENDIO_METADATA); 9021d4284bdSChris Mason if (ret) 90361891923SStefan Behrens goto out_w_error; 90408635baeSChris Mason ret = btrfs_map_bio(fs_info, bio, mirror_num); 905de0022b9SJosef Bacik } else if (!async) { 906de0022b9SJosef Bacik ret = btree_csum_one_bio(bio); 907de0022b9SJosef Bacik if (ret) 90861891923SStefan Behrens goto out_w_error; 90908635baeSChris Mason ret = btrfs_map_bio(fs_info, bio, mirror_num); 91061891923SStefan Behrens } else { 911cad321adSChris Mason /* 91261891923SStefan Behrens * kthread helpers are used to submit writes so that 91361891923SStefan Behrens * checksumming can happen in parallel across all CPUs 914cad321adSChris Mason */ 915c6100a4bSJosef Bacik ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0, 916e68f2ee7SNikolay Borisov 0, inode, btree_submit_bio_start); 91744b8bd7eSChris Mason } 91844b8bd7eSChris Mason 9194246a0b6SChristoph Hellwig if (ret) 9204246a0b6SChristoph Hellwig goto out_w_error; 9214246a0b6SChristoph Hellwig return 0; 9224246a0b6SChristoph Hellwig 92361891923SStefan Behrens out_w_error: 9244e4cbee9SChristoph Hellwig bio->bi_status = ret; 9254246a0b6SChristoph Hellwig bio_endio(bio); 92661891923SStefan Behrens return ret; 92761891923SStefan Behrens } 92861891923SStefan Behrens 9293dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 930784b4e29SChris Mason static int btree_migratepage(struct address_space *mapping, 931a6bc32b8SMel Gorman struct page *newpage, struct page *page, 932a6bc32b8SMel Gorman enum migrate_mode mode) 933784b4e29SChris Mason { 934784b4e29SChris Mason /* 935784b4e29SChris Mason * we can't safely write a btree page from here, 936784b4e29SChris Mason * we haven't done the locking hook 937784b4e29SChris Mason */ 938784b4e29SChris Mason if (PageDirty(page)) 939784b4e29SChris Mason return -EAGAIN; 940784b4e29SChris Mason /* 941784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 942784b4e29SChris Mason * We must have no buffers or drop them. 943784b4e29SChris Mason */ 944784b4e29SChris Mason if (page_has_private(page) && 945784b4e29SChris Mason !try_to_release_page(page, GFP_KERNEL)) 946784b4e29SChris Mason return -EAGAIN; 947a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 948784b4e29SChris Mason } 9493dd1462eSJan Beulich #endif 950784b4e29SChris Mason 9510da5468fSChris Mason 9520da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 9530da5468fSChris Mason struct writeback_control *wbc) 9540da5468fSChris Mason { 955e2d84521SMiao Xie struct btrfs_fs_info *fs_info; 956e2d84521SMiao Xie int ret; 957e2d84521SMiao Xie 958d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 959448d640bSChris Mason 960448d640bSChris Mason if (wbc->for_kupdate) 961448d640bSChris Mason return 0; 962448d640bSChris Mason 963e2d84521SMiao Xie fs_info = BTRFS_I(mapping->host)->root->fs_info; 964b9473439SChris Mason /* this is a bit racy, but that's ok */ 965d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 966d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 967d814a491SEthan Lien fs_info->dirty_metadata_batch); 968e2d84521SMiao Xie if (ret < 0) 969793955bcSChris Mason return 0; 970793955bcSChris Mason } 9710b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 9720da5468fSChris Mason } 9730da5468fSChris Mason 974b2950863SChristoph Hellwig static int btree_readpage(struct file *file, struct page *page) 9755f39d397SChris Mason { 976d1310b2eSChris Mason struct extent_io_tree *tree; 977d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9788ddc7d9cSJan Schmidt return extent_read_full_page(tree, page, btree_get_extent, 0); 9795f39d397SChris Mason } 9805f39d397SChris Mason 98170dec807SChris Mason static int btree_releasepage(struct page *page, gfp_t gfp_flags) 9825f39d397SChris Mason { 98398509cfcSChris Mason if (PageWriteback(page) || PageDirty(page)) 98498509cfcSChris Mason return 0; 9850c4e538bSDavid Sterba 986f7a52a40SDavid Sterba return try_release_extent_buffer(page); 987d98237b3SChris Mason } 988d98237b3SChris Mason 989d47992f8SLukas Czerner static void btree_invalidatepage(struct page *page, unsigned int offset, 990d47992f8SLukas Czerner unsigned int length) 991d98237b3SChris Mason { 992d1310b2eSChris Mason struct extent_io_tree *tree; 993d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9945f39d397SChris Mason extent_invalidatepage(tree, page, offset); 9955f39d397SChris Mason btree_releasepage(page, GFP_NOFS); 9969ad6b7bcSChris Mason if (PagePrivate(page)) { 997efe120a0SFrank Holton btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info, 998efe120a0SFrank Holton "page private not zero on page %llu", 999efe120a0SFrank Holton (unsigned long long)page_offset(page)); 10009ad6b7bcSChris Mason ClearPagePrivate(page); 10019ad6b7bcSChris Mason set_page_private(page, 0); 100209cbfeafSKirill A. Shutemov put_page(page); 10039ad6b7bcSChris Mason } 1004d98237b3SChris Mason } 1005d98237b3SChris Mason 10060b32f4bbSJosef Bacik static int btree_set_page_dirty(struct page *page) 10070b32f4bbSJosef Bacik { 1008bb146eb2SJosef Bacik #ifdef DEBUG 10090b32f4bbSJosef Bacik struct extent_buffer *eb; 10100b32f4bbSJosef Bacik 10110b32f4bbSJosef Bacik BUG_ON(!PagePrivate(page)); 10120b32f4bbSJosef Bacik eb = (struct extent_buffer *)page->private; 10130b32f4bbSJosef Bacik BUG_ON(!eb); 10140b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 10150b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 10160b32f4bbSJosef Bacik btrfs_assert_tree_locked(eb); 1017bb146eb2SJosef Bacik #endif 10180b32f4bbSJosef Bacik return __set_page_dirty_nobuffers(page); 10190b32f4bbSJosef Bacik } 10200b32f4bbSJosef Bacik 10217f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 1022d98237b3SChris Mason .readpage = btree_readpage, 10230da5468fSChris Mason .writepages = btree_writepages, 10245f39d397SChris Mason .releasepage = btree_releasepage, 10255f39d397SChris Mason .invalidatepage = btree_invalidatepage, 10265a92bc88SChris Mason #ifdef CONFIG_MIGRATION 1027784b4e29SChris Mason .migratepage = btree_migratepage, 10285a92bc88SChris Mason #endif 10290b32f4bbSJosef Bacik .set_page_dirty = btree_set_page_dirty, 1030d98237b3SChris Mason }; 1031123abc88SChris Mason 10322ff7e61eSJeff Mahoney void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr) 1033090d1875SChris Mason { 10345f39d397SChris Mason struct extent_buffer *buf = NULL; 1035537f38f0SNikolay Borisov int ret; 1036090d1875SChris Mason 10372ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 1038c871b0f2SLiu Bo if (IS_ERR(buf)) 10396197d86eSDavid Sterba return; 1040537f38f0SNikolay Borisov 1041c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(buf, WAIT_NONE, 0); 1042537f38f0SNikolay Borisov if (ret < 0) 1043537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 1044537f38f0SNikolay Borisov else 10455f39d397SChris Mason free_extent_buffer(buf); 1046090d1875SChris Mason } 1047090d1875SChris Mason 10482ff7e61eSJeff Mahoney struct extent_buffer *btrfs_find_create_tree_block( 10492ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info, 10500308af44SDavid Sterba u64 bytenr) 10510999df54SChris Mason { 10520b246afaSJeff Mahoney if (btrfs_is_testing(fs_info)) 10530b246afaSJeff Mahoney return alloc_test_extent_buffer(fs_info, bytenr); 10540b246afaSJeff Mahoney return alloc_extent_buffer(fs_info, bytenr); 10550999df54SChris Mason } 10560999df54SChris Mason 1057581c1760SQu Wenruo /* 1058581c1760SQu Wenruo * Read tree block at logical address @bytenr and do variant basic but critical 1059581c1760SQu Wenruo * verification. 1060581c1760SQu Wenruo * 1061581c1760SQu Wenruo * @parent_transid: expected transid of this tree block, skip check if 0 1062581c1760SQu Wenruo * @level: expected level, mandatory check 1063581c1760SQu Wenruo * @first_key: expected key in slot 0, skip check if NULL 1064581c1760SQu Wenruo */ 10652ff7e61eSJeff Mahoney struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, 1066581c1760SQu Wenruo u64 parent_transid, int level, 1067581c1760SQu Wenruo struct btrfs_key *first_key) 1068e20d96d6SChris Mason { 10695f39d397SChris Mason struct extent_buffer *buf = NULL; 107019c00ddcSChris Mason int ret; 107119c00ddcSChris Mason 10722ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 1073c871b0f2SLiu Bo if (IS_ERR(buf)) 1074c871b0f2SLiu Bo return buf; 1075e4204dedSChris Mason 10765ab12d1fSDavid Sterba ret = btree_read_extent_buffer_pages(buf, parent_transid, 1077581c1760SQu Wenruo level, first_key); 10780f0fe8f7SFilipe David Borba Manana if (ret) { 1079537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 108064c043deSLiu Bo return ERR_PTR(ret); 10810f0fe8f7SFilipe David Borba Manana } 10825f39d397SChris Mason return buf; 1083ce9adaa5SChris Mason 1084eb60ceacSChris Mason } 1085eb60ceacSChris Mason 10866a884d7dSDavid Sterba void btrfs_clean_tree_block(struct extent_buffer *buf) 1087ed2ff2cbSChris Mason { 10886a884d7dSDavid Sterba struct btrfs_fs_info *fs_info = buf->fs_info; 108955c69072SChris Mason if (btrfs_header_generation(buf) == 1090e2d84521SMiao Xie fs_info->running_transaction->transid) { 1091b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1092b4ce94deSChris Mason 1093b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1094104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 1095e2d84521SMiao Xie -buf->len, 1096e2d84521SMiao Xie fs_info->dirty_metadata_batch); 1097b9473439SChris Mason /* ugh, clear_extent_buffer_dirty needs to lock the page */ 10988bead258SDavid Sterba btrfs_set_lock_blocking_write(buf); 10990b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 1100925baeddSChris Mason } 11015f39d397SChris Mason } 1102ed7b63ebSJosef Bacik } 11035f39d397SChris Mason 11048257b2dcSMiao Xie static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void) 11058257b2dcSMiao Xie { 11068257b2dcSMiao Xie struct btrfs_subvolume_writers *writers; 11078257b2dcSMiao Xie int ret; 11088257b2dcSMiao Xie 11098257b2dcSMiao Xie writers = kmalloc(sizeof(*writers), GFP_NOFS); 11108257b2dcSMiao Xie if (!writers) 11118257b2dcSMiao Xie return ERR_PTR(-ENOMEM); 11128257b2dcSMiao Xie 11138a5a916dSJeff Mahoney ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS); 11148257b2dcSMiao Xie if (ret < 0) { 11158257b2dcSMiao Xie kfree(writers); 11168257b2dcSMiao Xie return ERR_PTR(ret); 11178257b2dcSMiao Xie } 11188257b2dcSMiao Xie 11198257b2dcSMiao Xie init_waitqueue_head(&writers->wait); 11208257b2dcSMiao Xie return writers; 11218257b2dcSMiao Xie } 11228257b2dcSMiao Xie 11238257b2dcSMiao Xie static void 11248257b2dcSMiao Xie btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers) 11258257b2dcSMiao Xie { 11268257b2dcSMiao Xie percpu_counter_destroy(&writers->counter); 11278257b2dcSMiao Xie kfree(writers); 11288257b2dcSMiao Xie } 11298257b2dcSMiao Xie 1130da17066cSJeff Mahoney static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, 1131e20d96d6SChris Mason u64 objectid) 1132d97e63b6SChris Mason { 11337c0260eeSJeff Mahoney bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); 1134cfaa7295SChris Mason root->node = NULL; 1135a28ec197SChris Mason root->commit_root = NULL; 113627cdeb70SMiao Xie root->state = 0; 1137d68fc57bSYan, Zheng root->orphan_cleanup_state = 0; 11380b86a832SChris Mason 11390f7d52f4SChris Mason root->last_trans = 0; 114013a8a7c8SYan, Zheng root->highest_objectid = 0; 1141eb73c1b7SMiao Xie root->nr_delalloc_inodes = 0; 1142199c2a9cSMiao Xie root->nr_ordered_extents = 0; 11436bef4d31SEric Paris root->inode_tree = RB_ROOT; 114416cdcec7SMiao Xie INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1145f0486c68SYan, Zheng root->block_rsv = NULL; 11460b86a832SChris Mason 11470b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 11485d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 1149eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_inodes); 1150eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_root); 1151199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_extents); 1152199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_root); 1153d2311e69SQu Wenruo INIT_LIST_HEAD(&root->reloc_dirty_list); 11542ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 11552ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 11565d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1157eb73c1b7SMiao Xie spin_lock_init(&root->delalloc_lock); 1158199c2a9cSMiao Xie spin_lock_init(&root->ordered_extent_lock); 1159f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 11602ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 11612ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 11628287475aSQu Wenruo spin_lock_init(&root->qgroup_meta_rsv_lock); 1163a2135011SChris Mason mutex_init(&root->objectid_mutex); 1164e02119d5SChris Mason mutex_init(&root->log_mutex); 116531f3d255SMiao Xie mutex_init(&root->ordered_extent_mutex); 1166573bfb72SMiao Xie mutex_init(&root->delalloc_mutex); 11677237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 11687237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 11697237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 11708b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[0]); 11718b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[1]); 11727237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 11737237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 11747237f183SYan Zheng atomic_set(&root->log_writers, 0); 11752ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 11760700cea7SElena Reshetova refcount_set(&root->refs, 1); 1177ea14b57fSDavid Sterba atomic_set(&root->will_be_snapshotted, 0); 11788ecebf4dSRobbie Ko atomic_set(&root->snapshot_force_cow, 0); 1179eede2bf3SOmar Sandoval atomic_set(&root->nr_swapfiles, 0); 11807237f183SYan Zheng root->log_transid = 0; 1181d1433debSMiao Xie root->log_transid_committed = -1; 1182257c62e1SChris Mason root->last_log_commit = 0; 11837c0260eeSJeff Mahoney if (!dummy) 118443eb5f29SQu Wenruo extent_io_tree_init(fs_info, &root->dirty_log_pages, 118543eb5f29SQu Wenruo IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL); 1186017e5369SChris Mason 11873768f368SChris Mason memset(&root->root_key, 0, sizeof(root->root_key)); 11883768f368SChris Mason memset(&root->root_item, 0, sizeof(root->root_item)); 11896702ed49SChris Mason memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 11907c0260eeSJeff Mahoney if (!dummy) 11913f157a2fSChris Mason root->defrag_trans_start = fs_info->generation; 119206ea65a3SJosef Bacik else 119306ea65a3SJosef Bacik root->defrag_trans_start = 0; 11944d775673SChris Mason root->root_key.objectid = objectid; 11950ee5dc67SAl Viro root->anon_dev = 0; 11968ea05e3aSAlexander Block 11975f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 1198370a11b8SQu Wenruo btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks); 11993768f368SChris Mason } 12003768f368SChris Mason 120174e4d827SDavid Sterba static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, 120274e4d827SDavid Sterba gfp_t flags) 12036f07e42eSAl Viro { 120474e4d827SDavid Sterba struct btrfs_root *root = kzalloc(sizeof(*root), flags); 12056f07e42eSAl Viro if (root) 12066f07e42eSAl Viro root->fs_info = fs_info; 12076f07e42eSAl Viro return root; 12086f07e42eSAl Viro } 12096f07e42eSAl Viro 121006ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 121106ea65a3SJosef Bacik /* Should only be used by the testing infrastructure */ 1212da17066cSJeff Mahoney struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) 121306ea65a3SJosef Bacik { 121406ea65a3SJosef Bacik struct btrfs_root *root; 121506ea65a3SJosef Bacik 12167c0260eeSJeff Mahoney if (!fs_info) 12177c0260eeSJeff Mahoney return ERR_PTR(-EINVAL); 12187c0260eeSJeff Mahoney 12197c0260eeSJeff Mahoney root = btrfs_alloc_root(fs_info, GFP_KERNEL); 122006ea65a3SJosef Bacik if (!root) 122106ea65a3SJosef Bacik return ERR_PTR(-ENOMEM); 1222da17066cSJeff Mahoney 1223b9ef22deSFeifei Xu /* We don't use the stripesize in selftest, set it as sectorsize */ 1224da17066cSJeff Mahoney __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID); 1225faa2dbf0SJosef Bacik root->alloc_bytenr = 0; 122606ea65a3SJosef Bacik 122706ea65a3SJosef Bacik return root; 122806ea65a3SJosef Bacik } 122906ea65a3SJosef Bacik #endif 123006ea65a3SJosef Bacik 123120897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 123220897f5cSArne Jansen u64 objectid) 123320897f5cSArne Jansen { 12349b7a2440SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 123520897f5cSArne Jansen struct extent_buffer *leaf; 123620897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 123720897f5cSArne Jansen struct btrfs_root *root; 123820897f5cSArne Jansen struct btrfs_key key; 1239b89f6d1fSFilipe Manana unsigned int nofs_flag; 124020897f5cSArne Jansen int ret = 0; 124133d85fdaSQu Wenruo uuid_le uuid = NULL_UUID_LE; 124220897f5cSArne Jansen 1243b89f6d1fSFilipe Manana /* 1244b89f6d1fSFilipe Manana * We're holding a transaction handle, so use a NOFS memory allocation 1245b89f6d1fSFilipe Manana * context to avoid deadlock if reclaim happens. 1246b89f6d1fSFilipe Manana */ 1247b89f6d1fSFilipe Manana nofs_flag = memalloc_nofs_save(); 124874e4d827SDavid Sterba root = btrfs_alloc_root(fs_info, GFP_KERNEL); 1249b89f6d1fSFilipe Manana memalloc_nofs_restore(nofs_flag); 125020897f5cSArne Jansen if (!root) 125120897f5cSArne Jansen return ERR_PTR(-ENOMEM); 125220897f5cSArne Jansen 1253da17066cSJeff Mahoney __setup_root(root, fs_info, objectid); 125420897f5cSArne Jansen root->root_key.objectid = objectid; 125520897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 125620897f5cSArne Jansen root->root_key.offset = 0; 125720897f5cSArne Jansen 12584d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0); 125920897f5cSArne Jansen if (IS_ERR(leaf)) { 126020897f5cSArne Jansen ret = PTR_ERR(leaf); 12611dd05682STsutomu Itoh leaf = NULL; 126220897f5cSArne Jansen goto fail; 126320897f5cSArne Jansen } 126420897f5cSArne Jansen 126520897f5cSArne Jansen root->node = leaf; 126620897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 126720897f5cSArne Jansen 126820897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 126927cdeb70SMiao Xie set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 127020897f5cSArne Jansen 127120897f5cSArne Jansen root->root_item.flags = 0; 127220897f5cSArne Jansen root->root_item.byte_limit = 0; 127320897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 127420897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 127520897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 127620897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 127720897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 127820897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 127920897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 128033d85fdaSQu Wenruo if (is_fstree(objectid)) 12816463fe58SStefan Behrens uuid_le_gen(&uuid); 12826463fe58SStefan Behrens memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE); 128320897f5cSArne Jansen root->root_item.drop_level = 0; 128420897f5cSArne Jansen 128520897f5cSArne Jansen key.objectid = objectid; 128620897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 128720897f5cSArne Jansen key.offset = 0; 128820897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 128920897f5cSArne Jansen if (ret) 129020897f5cSArne Jansen goto fail; 129120897f5cSArne Jansen 129220897f5cSArne Jansen btrfs_tree_unlock(leaf); 129320897f5cSArne Jansen 129420897f5cSArne Jansen return root; 12951dd05682STsutomu Itoh 12961dd05682STsutomu Itoh fail: 12971dd05682STsutomu Itoh if (leaf) { 12981dd05682STsutomu Itoh btrfs_tree_unlock(leaf); 129959885b39STsutomu Itoh free_extent_buffer(root->commit_root); 13001dd05682STsutomu Itoh free_extent_buffer(leaf); 13011dd05682STsutomu Itoh } 13021dd05682STsutomu Itoh kfree(root); 13031dd05682STsutomu Itoh 13041dd05682STsutomu Itoh return ERR_PTR(ret); 130520897f5cSArne Jansen } 130620897f5cSArne Jansen 13077237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1308e02119d5SChris Mason struct btrfs_fs_info *fs_info) 13090f7d52f4SChris Mason { 13100f7d52f4SChris Mason struct btrfs_root *root; 13117237f183SYan Zheng struct extent_buffer *leaf; 1312e02119d5SChris Mason 131374e4d827SDavid Sterba root = btrfs_alloc_root(fs_info, GFP_NOFS); 1314e02119d5SChris Mason if (!root) 13157237f183SYan Zheng return ERR_PTR(-ENOMEM); 1316e02119d5SChris Mason 1317da17066cSJeff Mahoney __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID); 1318e02119d5SChris Mason 1319e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1320e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1321e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 132227cdeb70SMiao Xie 13237237f183SYan Zheng /* 132427cdeb70SMiao Xie * DON'T set REF_COWS for log trees 132527cdeb70SMiao Xie * 13267237f183SYan Zheng * log trees do not get reference counted because they go away 13277237f183SYan Zheng * before a real commit is actually done. They do store pointers 13287237f183SYan Zheng * to file data extents, and those reference counts still get 13297237f183SYan Zheng * updated (along with back refs to the log tree). 13307237f183SYan Zheng */ 1331e02119d5SChris Mason 13324d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 13334d75f8a9SDavid Sterba NULL, 0, 0, 0); 13347237f183SYan Zheng if (IS_ERR(leaf)) { 13357237f183SYan Zheng kfree(root); 13367237f183SYan Zheng return ERR_CAST(leaf); 13377237f183SYan Zheng } 1338e02119d5SChris Mason 13397237f183SYan Zheng root->node = leaf; 1340e02119d5SChris Mason 1341e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1342e02119d5SChris Mason btrfs_tree_unlock(root->node); 13437237f183SYan Zheng return root; 13447237f183SYan Zheng } 13457237f183SYan Zheng 13467237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 13477237f183SYan Zheng struct btrfs_fs_info *fs_info) 13487237f183SYan Zheng { 13497237f183SYan Zheng struct btrfs_root *log_root; 13507237f183SYan Zheng 13517237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 13527237f183SYan Zheng if (IS_ERR(log_root)) 13537237f183SYan Zheng return PTR_ERR(log_root); 13547237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 13557237f183SYan Zheng fs_info->log_root_tree = log_root; 13567237f183SYan Zheng return 0; 13577237f183SYan Zheng } 13587237f183SYan Zheng 13597237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 13607237f183SYan Zheng struct btrfs_root *root) 13617237f183SYan Zheng { 13620b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13637237f183SYan Zheng struct btrfs_root *log_root; 13647237f183SYan Zheng struct btrfs_inode_item *inode_item; 13657237f183SYan Zheng 13660b246afaSJeff Mahoney log_root = alloc_log_tree(trans, fs_info); 13677237f183SYan Zheng if (IS_ERR(log_root)) 13687237f183SYan Zheng return PTR_ERR(log_root); 13697237f183SYan Zheng 13707237f183SYan Zheng log_root->last_trans = trans->transid; 13717237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 13727237f183SYan Zheng 13737237f183SYan Zheng inode_item = &log_root->root_item.inode; 13743cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 13753cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 13763cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 1377da17066cSJeff Mahoney btrfs_set_stack_inode_nbytes(inode_item, 13780b246afaSJeff Mahoney fs_info->nodesize); 13793cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 13807237f183SYan Zheng 13815d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 13827237f183SYan Zheng 13837237f183SYan Zheng WARN_ON(root->log_root); 13847237f183SYan Zheng root->log_root = log_root; 13857237f183SYan Zheng root->log_transid = 0; 1386d1433debSMiao Xie root->log_transid_committed = -1; 1387257c62e1SChris Mason root->last_log_commit = 0; 1388e02119d5SChris Mason return 0; 1389e02119d5SChris Mason } 1390e02119d5SChris Mason 139135a3621bSStefan Behrens static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 1392cb517eabSMiao Xie struct btrfs_key *key) 1393e02119d5SChris Mason { 1394e02119d5SChris Mason struct btrfs_root *root; 1395e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 13960f7d52f4SChris Mason struct btrfs_path *path; 139784234f3aSYan Zheng u64 generation; 1398cb517eabSMiao Xie int ret; 1399581c1760SQu Wenruo int level; 1400cb517eabSMiao Xie 1401cb517eabSMiao Xie path = btrfs_alloc_path(); 1402cb517eabSMiao Xie if (!path) 1403cb517eabSMiao Xie return ERR_PTR(-ENOMEM); 14040f7d52f4SChris Mason 140574e4d827SDavid Sterba root = btrfs_alloc_root(fs_info, GFP_NOFS); 1406cb517eabSMiao Xie if (!root) { 1407cb517eabSMiao Xie ret = -ENOMEM; 1408cb517eabSMiao Xie goto alloc_fail; 14090f7d52f4SChris Mason } 14100f7d52f4SChris Mason 1411da17066cSJeff Mahoney __setup_root(root, fs_info, key->objectid); 14120f7d52f4SChris Mason 1413cb517eabSMiao Xie ret = btrfs_find_root(tree_root, key, path, 1414cb517eabSMiao Xie &root->root_item, &root->root_key); 14150f7d52f4SChris Mason if (ret) { 141613a8a7c8SYan, Zheng if (ret > 0) 141713a8a7c8SYan, Zheng ret = -ENOENT; 1418cb517eabSMiao Xie goto find_fail; 14190f7d52f4SChris Mason } 142013a8a7c8SYan, Zheng 142184234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1422581c1760SQu Wenruo level = btrfs_root_level(&root->root_item); 14232ff7e61eSJeff Mahoney root->node = read_tree_block(fs_info, 14242ff7e61eSJeff Mahoney btrfs_root_bytenr(&root->root_item), 1425581c1760SQu Wenruo generation, level, NULL); 142664c043deSLiu Bo if (IS_ERR(root->node)) { 142764c043deSLiu Bo ret = PTR_ERR(root->node); 1428cb517eabSMiao Xie goto find_fail; 1429cb517eabSMiao Xie } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1430cb517eabSMiao Xie ret = -EIO; 143164c043deSLiu Bo free_extent_buffer(root->node); 143264c043deSLiu Bo goto find_fail; 1433416bc658SJosef Bacik } 14345d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 143513a8a7c8SYan, Zheng out: 1436cb517eabSMiao Xie btrfs_free_path(path); 1437cb517eabSMiao Xie return root; 1438cb517eabSMiao Xie 1439cb517eabSMiao Xie find_fail: 1440cb517eabSMiao Xie kfree(root); 1441cb517eabSMiao Xie alloc_fail: 1442cb517eabSMiao Xie root = ERR_PTR(ret); 1443cb517eabSMiao Xie goto out; 1444cb517eabSMiao Xie } 1445cb517eabSMiao Xie 1446cb517eabSMiao Xie struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root, 1447cb517eabSMiao Xie struct btrfs_key *location) 1448cb517eabSMiao Xie { 1449cb517eabSMiao Xie struct btrfs_root *root; 1450cb517eabSMiao Xie 1451cb517eabSMiao Xie root = btrfs_read_tree_root(tree_root, location); 1452cb517eabSMiao Xie if (IS_ERR(root)) 1453cb517eabSMiao Xie return root; 1454cb517eabSMiao Xie 1455cb517eabSMiao Xie if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { 145627cdeb70SMiao Xie set_bit(BTRFS_ROOT_REF_COWS, &root->state); 145708fe4db1SLi Zefan btrfs_check_and_init_root_item(&root->root_item); 145808fe4db1SLi Zefan } 145913a8a7c8SYan, Zheng 14605eda7b5eSChris Mason return root; 14615eda7b5eSChris Mason } 14625eda7b5eSChris Mason 1463cb517eabSMiao Xie int btrfs_init_fs_root(struct btrfs_root *root) 1464cb517eabSMiao Xie { 1465cb517eabSMiao Xie int ret; 14668257b2dcSMiao Xie struct btrfs_subvolume_writers *writers; 1467cb517eabSMiao Xie 1468cb517eabSMiao Xie root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); 1469cb517eabSMiao Xie root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), 1470cb517eabSMiao Xie GFP_NOFS); 1471cb517eabSMiao Xie if (!root->free_ino_pinned || !root->free_ino_ctl) { 1472cb517eabSMiao Xie ret = -ENOMEM; 1473cb517eabSMiao Xie goto fail; 1474cb517eabSMiao Xie } 1475cb517eabSMiao Xie 14768257b2dcSMiao Xie writers = btrfs_alloc_subvolume_writers(); 14778257b2dcSMiao Xie if (IS_ERR(writers)) { 14788257b2dcSMiao Xie ret = PTR_ERR(writers); 14798257b2dcSMiao Xie goto fail; 14808257b2dcSMiao Xie } 14818257b2dcSMiao Xie root->subv_writers = writers; 14828257b2dcSMiao Xie 1483cb517eabSMiao Xie btrfs_init_free_ino_ctl(root); 148457cdc8dbSDavid Sterba spin_lock_init(&root->ino_cache_lock); 148557cdc8dbSDavid Sterba init_waitqueue_head(&root->ino_cache_wait); 1486cb517eabSMiao Xie 1487cb517eabSMiao Xie ret = get_anon_bdev(&root->anon_dev); 1488cb517eabSMiao Xie if (ret) 1489876d2cf1SLiu Bo goto fail; 1490f32e48e9SChandan Rajendra 1491f32e48e9SChandan Rajendra mutex_lock(&root->objectid_mutex); 1492f32e48e9SChandan Rajendra ret = btrfs_find_highest_objectid(root, 1493f32e48e9SChandan Rajendra &root->highest_objectid); 1494f32e48e9SChandan Rajendra if (ret) { 1495f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1496876d2cf1SLiu Bo goto fail; 1497f32e48e9SChandan Rajendra } 1498f32e48e9SChandan Rajendra 1499f32e48e9SChandan Rajendra ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 1500f32e48e9SChandan Rajendra 1501f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1502f32e48e9SChandan Rajendra 1503cb517eabSMiao Xie return 0; 1504cb517eabSMiao Xie fail: 150584db5ccfSDavid Sterba /* The caller is responsible to call btrfs_free_fs_root */ 1506cb517eabSMiao Xie return ret; 1507cb517eabSMiao Xie } 1508cb517eabSMiao Xie 150935bbb97fSJeff Mahoney struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1510cb517eabSMiao Xie u64 root_id) 1511cb517eabSMiao Xie { 1512cb517eabSMiao Xie struct btrfs_root *root; 1513cb517eabSMiao Xie 1514cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1515cb517eabSMiao Xie root = radix_tree_lookup(&fs_info->fs_roots_radix, 1516cb517eabSMiao Xie (unsigned long)root_id); 1517cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1518cb517eabSMiao Xie return root; 1519cb517eabSMiao Xie } 1520cb517eabSMiao Xie 1521cb517eabSMiao Xie int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1522cb517eabSMiao Xie struct btrfs_root *root) 1523cb517eabSMiao Xie { 1524cb517eabSMiao Xie int ret; 1525cb517eabSMiao Xie 1526e1860a77SDavid Sterba ret = radix_tree_preload(GFP_NOFS); 1527cb517eabSMiao Xie if (ret) 1528cb517eabSMiao Xie return ret; 1529cb517eabSMiao Xie 1530cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1531cb517eabSMiao Xie ret = radix_tree_insert(&fs_info->fs_roots_radix, 1532cb517eabSMiao Xie (unsigned long)root->root_key.objectid, 1533cb517eabSMiao Xie root); 1534cb517eabSMiao Xie if (ret == 0) 153527cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1536cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1537cb517eabSMiao Xie radix_tree_preload_end(); 1538cb517eabSMiao Xie 1539cb517eabSMiao Xie return ret; 1540cb517eabSMiao Xie } 1541cb517eabSMiao Xie 1542c00869f1SMiao Xie struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 1543c00869f1SMiao Xie struct btrfs_key *location, 1544c00869f1SMiao Xie bool check_ref) 15455eda7b5eSChris Mason { 15465eda7b5eSChris Mason struct btrfs_root *root; 1547381cf658SDavid Sterba struct btrfs_path *path; 15481d4c08e0SDavid Sterba struct btrfs_key key; 15495eda7b5eSChris Mason int ret; 15505eda7b5eSChris Mason 1551edbd8d4eSChris Mason if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 1552edbd8d4eSChris Mason return fs_info->tree_root; 1553edbd8d4eSChris Mason if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID) 1554edbd8d4eSChris Mason return fs_info->extent_root; 15558f18cf13SChris Mason if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID) 15568f18cf13SChris Mason return fs_info->chunk_root; 15578f18cf13SChris Mason if (location->objectid == BTRFS_DEV_TREE_OBJECTID) 15588f18cf13SChris Mason return fs_info->dev_root; 15590403e47eSYan Zheng if (location->objectid == BTRFS_CSUM_TREE_OBJECTID) 15600403e47eSYan Zheng return fs_info->csum_root; 1561bcef60f2SArne Jansen if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID) 1562bcef60f2SArne Jansen return fs_info->quota_root ? fs_info->quota_root : 1563bcef60f2SArne Jansen ERR_PTR(-ENOENT); 1564f7a81ea4SStefan Behrens if (location->objectid == BTRFS_UUID_TREE_OBJECTID) 1565f7a81ea4SStefan Behrens return fs_info->uuid_root ? fs_info->uuid_root : 1566f7a81ea4SStefan Behrens ERR_PTR(-ENOENT); 156770f6d82eSOmar Sandoval if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) 156870f6d82eSOmar Sandoval return fs_info->free_space_root ? fs_info->free_space_root : 156970f6d82eSOmar Sandoval ERR_PTR(-ENOENT); 15704df27c4dSYan, Zheng again: 1571cb517eabSMiao Xie root = btrfs_lookup_fs_root(fs_info, location->objectid); 157248475471SStefan Behrens if (root) { 1573c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) 157448475471SStefan Behrens return ERR_PTR(-ENOENT); 15755eda7b5eSChris Mason return root; 157648475471SStefan Behrens } 15775eda7b5eSChris Mason 1578cb517eabSMiao Xie root = btrfs_read_fs_root(fs_info->tree_root, location); 15795eda7b5eSChris Mason if (IS_ERR(root)) 15805eda7b5eSChris Mason return root; 15813394e160SChris Mason 1582c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1583d68fc57bSYan, Zheng ret = -ENOENT; 1584d68fc57bSYan, Zheng goto fail; 1585d68fc57bSYan, Zheng } 1586d68fc57bSYan, Zheng 1587cb517eabSMiao Xie ret = btrfs_init_fs_root(root); 1588cb517eabSMiao Xie if (ret) 1589cb517eabSMiao Xie goto fail; 1590cb517eabSMiao Xie 1591381cf658SDavid Sterba path = btrfs_alloc_path(); 1592381cf658SDavid Sterba if (!path) { 1593381cf658SDavid Sterba ret = -ENOMEM; 1594381cf658SDavid Sterba goto fail; 1595381cf658SDavid Sterba } 15961d4c08e0SDavid Sterba key.objectid = BTRFS_ORPHAN_OBJECTID; 15971d4c08e0SDavid Sterba key.type = BTRFS_ORPHAN_ITEM_KEY; 15981d4c08e0SDavid Sterba key.offset = location->objectid; 15991d4c08e0SDavid Sterba 16001d4c08e0SDavid Sterba ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1601381cf658SDavid Sterba btrfs_free_path(path); 1602d68fc57bSYan, Zheng if (ret < 0) 1603d68fc57bSYan, Zheng goto fail; 1604d68fc57bSYan, Zheng if (ret == 0) 160527cdeb70SMiao Xie set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1606d68fc57bSYan, Zheng 1607cb517eabSMiao Xie ret = btrfs_insert_fs_root(fs_info, root); 16080f7d52f4SChris Mason if (ret) { 16094df27c4dSYan, Zheng if (ret == -EEXIST) { 161084db5ccfSDavid Sterba btrfs_free_fs_root(root); 16114df27c4dSYan, Zheng goto again; 16120f7d52f4SChris Mason } 16134df27c4dSYan, Zheng goto fail; 16144df27c4dSYan, Zheng } 1615edbd8d4eSChris Mason return root; 16164df27c4dSYan, Zheng fail: 161784db5ccfSDavid Sterba btrfs_free_fs_root(root); 16184df27c4dSYan, Zheng return ERR_PTR(ret); 1619edbd8d4eSChris Mason } 1620edbd8d4eSChris Mason 162104160088SChris Mason static int btrfs_congested_fn(void *congested_data, int bdi_bits) 162204160088SChris Mason { 162304160088SChris Mason struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; 162404160088SChris Mason int ret = 0; 162504160088SChris Mason struct btrfs_device *device; 162604160088SChris Mason struct backing_dev_info *bdi; 1627b7967db7SChris Mason 16281f78160cSXiao Guangrong rcu_read_lock(); 16291f78160cSXiao Guangrong list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) { 1630dfe25020SChris Mason if (!device->bdev) 1631dfe25020SChris Mason continue; 1632efa7c9f9SJan Kara bdi = device->bdev->bd_bdi; 1633ff9ea323STejun Heo if (bdi_congested(bdi, bdi_bits)) { 163404160088SChris Mason ret = 1; 163504160088SChris Mason break; 163604160088SChris Mason } 163704160088SChris Mason } 16381f78160cSXiao Guangrong rcu_read_unlock(); 163904160088SChris Mason return ret; 164004160088SChris Mason } 164104160088SChris Mason 16428b712842SChris Mason /* 16438b712842SChris Mason * called by the kthread helper functions to finally call the bio end_io 16448b712842SChris Mason * functions. This is where read checksum verification actually happens 16458b712842SChris Mason */ 16468b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work) 1647ce9adaa5SChris Mason { 1648ce9adaa5SChris Mason struct bio *bio; 164997eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 1650ce9adaa5SChris Mason 165197eb6b69SDavid Sterba end_io_wq = container_of(work, struct btrfs_end_io_wq, work); 1652ce9adaa5SChris Mason bio = end_io_wq->bio; 16538b712842SChris Mason 16544e4cbee9SChristoph Hellwig bio->bi_status = end_io_wq->status; 1655ce9adaa5SChris Mason bio->bi_private = end_io_wq->private; 1656ce9adaa5SChris Mason bio->bi_end_io = end_io_wq->end_io; 16574246a0b6SChristoph Hellwig bio_endio(bio); 16589be490f1SOmar Sandoval kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq); 1659ce9adaa5SChris Mason } 166044b8bd7eSChris Mason 1661a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1662a74a4b97SChris Mason { 1663a74a4b97SChris Mason struct btrfs_root *root = arg; 16640b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1665d0278245SMiao Xie int again; 1666a74a4b97SChris Mason 1667d6fd0ae2SOmar Sandoval while (1) { 1668d0278245SMiao Xie again = 0; 16699d1a2a3aSDavid Sterba 1670fd340d0fSJosef Bacik set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1671fd340d0fSJosef Bacik 1672d0278245SMiao Xie /* Make the cleaner go to sleep early. */ 16732ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) 1674d0278245SMiao Xie goto sleep; 1675d0278245SMiao Xie 167690c711abSZygo Blaxell /* 167790c711abSZygo Blaxell * Do not do anything if we might cause open_ctree() to block 167890c711abSZygo Blaxell * before we have finished mounting the filesystem. 167990c711abSZygo Blaxell */ 16800b246afaSJeff Mahoney if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 168190c711abSZygo Blaxell goto sleep; 168290c711abSZygo Blaxell 16830b246afaSJeff Mahoney if (!mutex_trylock(&fs_info->cleaner_mutex)) 1684d0278245SMiao Xie goto sleep; 1685d0278245SMiao Xie 1686dc7f370cSMiao Xie /* 1687dc7f370cSMiao Xie * Avoid the problem that we change the status of the fs 1688dc7f370cSMiao Xie * during the above check and trylock. 1689dc7f370cSMiao Xie */ 16902ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) { 16910b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1692dc7f370cSMiao Xie goto sleep; 1693dc7f370cSMiao Xie } 1694dc7f370cSMiao Xie 16952ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 1696c2d6cb16SFilipe Manana 16979d1a2a3aSDavid Sterba again = btrfs_clean_one_deleted_snapshot(root); 16980b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1699a74a4b97SChris Mason 1700d0278245SMiao Xie /* 170105323cd1SMiao Xie * The defragger has dealt with the R/O remount and umount, 170205323cd1SMiao Xie * needn't do anything special here. 1703d0278245SMiao Xie */ 17040b246afaSJeff Mahoney btrfs_run_defrag_inodes(fs_info); 170567c5e7d4SFilipe Manana 170667c5e7d4SFilipe Manana /* 170767c5e7d4SFilipe Manana * Acquires fs_info->delete_unused_bgs_mutex to avoid racing 170867c5e7d4SFilipe Manana * with relocation (btrfs_relocate_chunk) and relocation 170967c5e7d4SFilipe Manana * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 171067c5e7d4SFilipe Manana * after acquiring fs_info->delete_unused_bgs_mutex. So we 171167c5e7d4SFilipe Manana * can't hold, nor need to, fs_info->cleaner_mutex when deleting 171267c5e7d4SFilipe Manana * unused block groups. 171367c5e7d4SFilipe Manana */ 17140b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 1715d0278245SMiao Xie sleep: 1716fd340d0fSJosef Bacik clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1717d6fd0ae2SOmar Sandoval if (kthread_should_park()) 1718d6fd0ae2SOmar Sandoval kthread_parkme(); 1719d6fd0ae2SOmar Sandoval if (kthread_should_stop()) 1720d6fd0ae2SOmar Sandoval return 0; 1721838fe188SJiri Kosina if (!again) { 1722a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 1723a74a4b97SChris Mason schedule(); 1724a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1725a74a4b97SChris Mason } 1726da288d28SFilipe Manana } 1727a74a4b97SChris Mason } 1728a74a4b97SChris Mason 1729a74a4b97SChris Mason static int transaction_kthread(void *arg) 1730a74a4b97SChris Mason { 1731a74a4b97SChris Mason struct btrfs_root *root = arg; 17320b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1733a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1734a74a4b97SChris Mason struct btrfs_transaction *cur; 17358929ecfaSYan, Zheng u64 transid; 1736a944442cSAllen Pais time64_t now; 1737a74a4b97SChris Mason unsigned long delay; 1738914b2007SJan Kara bool cannot_commit; 1739a74a4b97SChris Mason 1740a74a4b97SChris Mason do { 1741914b2007SJan Kara cannot_commit = false; 17420b246afaSJeff Mahoney delay = HZ * fs_info->commit_interval; 17430b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 1744a74a4b97SChris Mason 17450b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 17460b246afaSJeff Mahoney cur = fs_info->running_transaction; 1747a74a4b97SChris Mason if (!cur) { 17480b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1749a74a4b97SChris Mason goto sleep; 1750a74a4b97SChris Mason } 175131153d81SYan Zheng 1752afd48513SArnd Bergmann now = ktime_get_seconds(); 17533296bf56SQu Wenruo if (cur->state < TRANS_STATE_COMMIT_START && 1754a514d638SQu Wenruo !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) && 17558b87dc17SDavid Sterba (now < cur->start_time || 17560b246afaSJeff Mahoney now - cur->start_time < fs_info->commit_interval)) { 17570b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1758a74a4b97SChris Mason delay = HZ * 5; 1759a74a4b97SChris Mason goto sleep; 1760a74a4b97SChris Mason } 17618929ecfaSYan, Zheng transid = cur->transid; 17620b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 176356bec294SChris Mason 176479787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 1765354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 1766914b2007SJan Kara if (IS_ERR(trans)) { 1767354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 1768914b2007SJan Kara cannot_commit = true; 176979787eaaSJeff Mahoney goto sleep; 1770914b2007SJan Kara } 17718929ecfaSYan, Zheng if (transid == trans->transid) { 17723a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 17738929ecfaSYan, Zheng } else { 17743a45bb20SJeff Mahoney btrfs_end_transaction(trans); 17758929ecfaSYan, Zheng } 1776a74a4b97SChris Mason sleep: 17770b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 17780b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 1779a74a4b97SChris Mason 17804e121c06SJosef Bacik if (unlikely(test_bit(BTRFS_FS_STATE_ERROR, 17810b246afaSJeff Mahoney &fs_info->fs_state))) 17822ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 17838929ecfaSYan, Zheng if (!kthread_should_stop() && 17840b246afaSJeff Mahoney (!btrfs_transaction_blocked(fs_info) || 1785914b2007SJan Kara cannot_commit)) 1786bc5511d0SNikolay Borisov schedule_timeout_interruptible(delay); 1787a74a4b97SChris Mason } while (!kthread_should_stop()); 1788a74a4b97SChris Mason return 0; 1789a74a4b97SChris Mason } 1790a74a4b97SChris Mason 1791af31f5e5SChris Mason /* 1792af31f5e5SChris Mason * this will find the highest generation in the array of 1793af31f5e5SChris Mason * root backups. The index of the highest array is returned, 1794af31f5e5SChris Mason * or -1 if we can't find anything. 1795af31f5e5SChris Mason * 1796af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1797af31f5e5SChris Mason * generation of the latest root in the array with the generation 1798af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1799af31f5e5SChris Mason */ 1800af31f5e5SChris Mason static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen) 1801af31f5e5SChris Mason { 1802af31f5e5SChris Mason u64 cur; 1803af31f5e5SChris Mason int newest_index = -1; 1804af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1805af31f5e5SChris Mason int i; 1806af31f5e5SChris Mason 1807af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1808af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1809af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1810af31f5e5SChris Mason if (cur == newest_gen) 1811af31f5e5SChris Mason newest_index = i; 1812af31f5e5SChris Mason } 1813af31f5e5SChris Mason 1814af31f5e5SChris Mason /* check to see if we actually wrapped around */ 1815af31f5e5SChris Mason if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) { 1816af31f5e5SChris Mason root_backup = info->super_copy->super_roots; 1817af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1818af31f5e5SChris Mason if (cur == newest_gen) 1819af31f5e5SChris Mason newest_index = 0; 1820af31f5e5SChris Mason } 1821af31f5e5SChris Mason return newest_index; 1822af31f5e5SChris Mason } 1823af31f5e5SChris Mason 1824af31f5e5SChris Mason 1825af31f5e5SChris Mason /* 1826af31f5e5SChris Mason * find the oldest backup so we know where to store new entries 1827af31f5e5SChris Mason * in the backup array. This will set the backup_root_index 1828af31f5e5SChris Mason * field in the fs_info struct 1829af31f5e5SChris Mason */ 1830af31f5e5SChris Mason static void find_oldest_super_backup(struct btrfs_fs_info *info, 1831af31f5e5SChris Mason u64 newest_gen) 1832af31f5e5SChris Mason { 1833af31f5e5SChris Mason int newest_index = -1; 1834af31f5e5SChris Mason 1835af31f5e5SChris Mason newest_index = find_newest_super_backup(info, newest_gen); 1836af31f5e5SChris Mason /* if there was garbage in there, just move along */ 1837af31f5e5SChris Mason if (newest_index == -1) { 1838af31f5e5SChris Mason info->backup_root_index = 0; 1839af31f5e5SChris Mason } else { 1840af31f5e5SChris Mason info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS; 1841af31f5e5SChris Mason } 1842af31f5e5SChris Mason } 1843af31f5e5SChris Mason 1844af31f5e5SChris Mason /* 1845af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1846af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1847af31f5e5SChris Mason * done 1848af31f5e5SChris Mason */ 1849af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1850af31f5e5SChris Mason { 1851af31f5e5SChris Mason int next_backup; 1852af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1853af31f5e5SChris Mason int last_backup; 1854af31f5e5SChris Mason 1855af31f5e5SChris Mason next_backup = info->backup_root_index; 1856af31f5e5SChris Mason last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) % 1857af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 1858af31f5e5SChris Mason 1859af31f5e5SChris Mason /* 1860af31f5e5SChris Mason * just overwrite the last backup if we're at the same generation 1861af31f5e5SChris Mason * this happens only at umount 1862af31f5e5SChris Mason */ 1863af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + last_backup; 1864af31f5e5SChris Mason if (btrfs_backup_tree_root_gen(root_backup) == 1865af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)) 1866af31f5e5SChris Mason next_backup = last_backup; 1867af31f5e5SChris Mason 1868af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1869af31f5e5SChris Mason 1870af31f5e5SChris Mason /* 1871af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1872af31f5e5SChris Mason * regardless of which ones we use today 1873af31f5e5SChris Mason */ 1874af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1875af31f5e5SChris Mason 1876af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1877af31f5e5SChris Mason 1878af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1879af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1880af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1881af31f5e5SChris Mason 1882af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1883af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1884af31f5e5SChris Mason 1885af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1886af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1887af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 1888af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 1889af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 1890af31f5e5SChris Mason 1891af31f5e5SChris Mason btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); 1892af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 1893af31f5e5SChris Mason btrfs_header_generation(info->extent_root->node)); 1894af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 1895af31f5e5SChris Mason btrfs_header_level(info->extent_root->node)); 1896af31f5e5SChris Mason 18977c7e82a7SChris Mason /* 18987c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 18997c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 19007c7e82a7SChris Mason */ 19017c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 19027c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 19037c7e82a7SChris Mason info->fs_root->node->start); 1904af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 1905af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 1906af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 1907af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 19087c7e82a7SChris Mason } 1909af31f5e5SChris Mason 1910af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 1911af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 1912af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 1913af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 1914af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 1915af31f5e5SChris Mason 1916af31f5e5SChris Mason btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); 1917af31f5e5SChris Mason btrfs_set_backup_csum_root_gen(root_backup, 1918af31f5e5SChris Mason btrfs_header_generation(info->csum_root->node)); 1919af31f5e5SChris Mason btrfs_set_backup_csum_root_level(root_backup, 1920af31f5e5SChris Mason btrfs_header_level(info->csum_root->node)); 1921af31f5e5SChris Mason 1922af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 1923af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 1924af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 1925af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 1926af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 1927af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 1928af31f5e5SChris Mason 1929af31f5e5SChris Mason /* 1930af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 1931af31f5e5SChris Mason * for the next commit 1932af31f5e5SChris Mason */ 1933af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 1934af31f5e5SChris Mason &info->super_for_commit->super_roots, 1935af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 1936af31f5e5SChris Mason } 1937af31f5e5SChris Mason 1938af31f5e5SChris Mason /* 1939af31f5e5SChris Mason * this copies info out of the root backup array and back into 1940af31f5e5SChris Mason * the in-memory super block. It is meant to help iterate through 1941af31f5e5SChris Mason * the array, so you send it the number of backups you've already 1942af31f5e5SChris Mason * tried and the last backup index you used. 1943af31f5e5SChris Mason * 1944af31f5e5SChris Mason * this returns -1 when it has tried all the backups 1945af31f5e5SChris Mason */ 1946af31f5e5SChris Mason static noinline int next_root_backup(struct btrfs_fs_info *info, 1947af31f5e5SChris Mason struct btrfs_super_block *super, 1948af31f5e5SChris Mason int *num_backups_tried, int *backup_index) 1949af31f5e5SChris Mason { 1950af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1951af31f5e5SChris Mason int newest = *backup_index; 1952af31f5e5SChris Mason 1953af31f5e5SChris Mason if (*num_backups_tried == 0) { 1954af31f5e5SChris Mason u64 gen = btrfs_super_generation(super); 1955af31f5e5SChris Mason 1956af31f5e5SChris Mason newest = find_newest_super_backup(info, gen); 1957af31f5e5SChris Mason if (newest == -1) 1958af31f5e5SChris Mason return -1; 1959af31f5e5SChris Mason 1960af31f5e5SChris Mason *backup_index = newest; 1961af31f5e5SChris Mason *num_backups_tried = 1; 1962af31f5e5SChris Mason } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) { 1963af31f5e5SChris Mason /* we've tried all the backups, all done */ 1964af31f5e5SChris Mason return -1; 1965af31f5e5SChris Mason } else { 1966af31f5e5SChris Mason /* jump to the next oldest backup */ 1967af31f5e5SChris Mason newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) % 1968af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 1969af31f5e5SChris Mason *backup_index = newest; 1970af31f5e5SChris Mason *num_backups_tried += 1; 1971af31f5e5SChris Mason } 1972af31f5e5SChris Mason root_backup = super->super_roots + newest; 1973af31f5e5SChris Mason 1974af31f5e5SChris Mason btrfs_set_super_generation(super, 1975af31f5e5SChris Mason btrfs_backup_tree_root_gen(root_backup)); 1976af31f5e5SChris Mason btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 1977af31f5e5SChris Mason btrfs_set_super_root_level(super, 1978af31f5e5SChris Mason btrfs_backup_tree_root_level(root_backup)); 1979af31f5e5SChris Mason btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 1980af31f5e5SChris Mason 1981af31f5e5SChris Mason /* 1982af31f5e5SChris Mason * fixme: the total bytes and num_devices need to match or we should 1983af31f5e5SChris Mason * need a fsck 1984af31f5e5SChris Mason */ 1985af31f5e5SChris Mason btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 1986af31f5e5SChris Mason btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 1987af31f5e5SChris Mason return 0; 1988af31f5e5SChris Mason } 1989af31f5e5SChris Mason 19907abadb64SLiu Bo /* helper to cleanup workers */ 19917abadb64SLiu Bo static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 19927abadb64SLiu Bo { 1993dc6e3209SQu Wenruo btrfs_destroy_workqueue(fs_info->fixup_workers); 1994afe3d242SQu Wenruo btrfs_destroy_workqueue(fs_info->delalloc_workers); 19955cdc7ad3SQu Wenruo btrfs_destroy_workqueue(fs_info->workers); 1996fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_workers); 1997fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_raid56_workers); 19988b110e39SMiao Xie btrfs_destroy_workqueue(fs_info->endio_repair_workers); 1999d05a33acSQu Wenruo btrfs_destroy_workqueue(fs_info->rmw_workers); 2000fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_write_workers); 2001fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 20025b3bc44eSQu Wenruo btrfs_destroy_workqueue(fs_info->delayed_workers); 2003e66f0bb1SQu Wenruo btrfs_destroy_workqueue(fs_info->caching_workers); 2004736cfa15SQu Wenruo btrfs_destroy_workqueue(fs_info->readahead_workers); 2005a44903abSQu Wenruo btrfs_destroy_workqueue(fs_info->flush_workers); 2006fc97fab0SQu Wenruo btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2007a9b9477dSFilipe Manana /* 2008a9b9477dSFilipe Manana * Now that all other work queues are destroyed, we can safely destroy 2009a9b9477dSFilipe Manana * the queues used for metadata I/O, since tasks from those other work 2010a9b9477dSFilipe Manana * queues can do metadata I/O operations. 2011a9b9477dSFilipe Manana */ 2012a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_workers); 2013a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_write_workers); 20147abadb64SLiu Bo } 20157abadb64SLiu Bo 20162e9f5954SRashika static void free_root_extent_buffers(struct btrfs_root *root) 20172e9f5954SRashika { 20182e9f5954SRashika if (root) { 20192e9f5954SRashika free_extent_buffer(root->node); 20202e9f5954SRashika free_extent_buffer(root->commit_root); 20212e9f5954SRashika root->node = NULL; 20222e9f5954SRashika root->commit_root = NULL; 20232e9f5954SRashika } 20242e9f5954SRashika } 20252e9f5954SRashika 2026af31f5e5SChris Mason /* helper to cleanup tree roots */ 20274273eaffSAnand Jain static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root) 2028af31f5e5SChris Mason { 20292e9f5954SRashika free_root_extent_buffers(info->tree_root); 2030655b09feSJosef Bacik 20312e9f5954SRashika free_root_extent_buffers(info->dev_root); 20322e9f5954SRashika free_root_extent_buffers(info->extent_root); 20332e9f5954SRashika free_root_extent_buffers(info->csum_root); 20342e9f5954SRashika free_root_extent_buffers(info->quota_root); 20352e9f5954SRashika free_root_extent_buffers(info->uuid_root); 20364273eaffSAnand Jain if (free_chunk_root) 20372e9f5954SRashika free_root_extent_buffers(info->chunk_root); 203870f6d82eSOmar Sandoval free_root_extent_buffers(info->free_space_root); 2039af31f5e5SChris Mason } 2040af31f5e5SChris Mason 2041faa2dbf0SJosef Bacik void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2042171f6537SJosef Bacik { 2043171f6537SJosef Bacik int ret; 2044171f6537SJosef Bacik struct btrfs_root *gang[8]; 2045171f6537SJosef Bacik int i; 2046171f6537SJosef Bacik 2047171f6537SJosef Bacik while (!list_empty(&fs_info->dead_roots)) { 2048171f6537SJosef Bacik gang[0] = list_entry(fs_info->dead_roots.next, 2049171f6537SJosef Bacik struct btrfs_root, root_list); 2050171f6537SJosef Bacik list_del(&gang[0]->root_list); 2051171f6537SJosef Bacik 205227cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) { 2053cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[0]); 2054171f6537SJosef Bacik } else { 2055171f6537SJosef Bacik free_extent_buffer(gang[0]->node); 2056171f6537SJosef Bacik free_extent_buffer(gang[0]->commit_root); 2057b0feb9d9SMiao Xie btrfs_put_fs_root(gang[0]); 2058171f6537SJosef Bacik } 2059171f6537SJosef Bacik } 2060171f6537SJosef Bacik 2061171f6537SJosef Bacik while (1) { 2062171f6537SJosef Bacik ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2063171f6537SJosef Bacik (void **)gang, 0, 2064171f6537SJosef Bacik ARRAY_SIZE(gang)); 2065171f6537SJosef Bacik if (!ret) 2066171f6537SJosef Bacik break; 2067171f6537SJosef Bacik for (i = 0; i < ret; i++) 2068cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2069171f6537SJosef Bacik } 20701a4319ccSLiu Bo 20711a4319ccSLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 20721a4319ccSLiu Bo btrfs_free_log_root_tree(NULL, fs_info); 20732ff7e61eSJeff Mahoney btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents); 20741a4319ccSLiu Bo } 2075171f6537SJosef Bacik } 2076af31f5e5SChris Mason 2077638aa7edSEric Sandeen static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2078638aa7edSEric Sandeen { 2079638aa7edSEric Sandeen mutex_init(&fs_info->scrub_lock); 2080638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_running, 0); 2081638aa7edSEric Sandeen atomic_set(&fs_info->scrub_pause_req, 0); 2082638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_paused, 0); 2083638aa7edSEric Sandeen atomic_set(&fs_info->scrub_cancel_req, 0); 2084638aa7edSEric Sandeen init_waitqueue_head(&fs_info->scrub_pause_wait); 2085ff09c4caSAnand Jain refcount_set(&fs_info->scrub_workers_refcnt, 0); 2086638aa7edSEric Sandeen } 2087638aa7edSEric Sandeen 2088779a65a4SEric Sandeen static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2089779a65a4SEric Sandeen { 2090779a65a4SEric Sandeen spin_lock_init(&fs_info->balance_lock); 2091779a65a4SEric Sandeen mutex_init(&fs_info->balance_mutex); 2092779a65a4SEric Sandeen atomic_set(&fs_info->balance_pause_req, 0); 2093779a65a4SEric Sandeen atomic_set(&fs_info->balance_cancel_req, 0); 2094779a65a4SEric Sandeen fs_info->balance_ctl = NULL; 2095779a65a4SEric Sandeen init_waitqueue_head(&fs_info->balance_wait_q); 2096779a65a4SEric Sandeen } 2097779a65a4SEric Sandeen 20986bccf3abSJeff Mahoney static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info) 2099f37938e0SEric Sandeen { 21002ff7e61eSJeff Mahoney struct inode *inode = fs_info->btree_inode; 21012ff7e61eSJeff Mahoney 21022ff7e61eSJeff Mahoney inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 21032ff7e61eSJeff Mahoney set_nlink(inode, 1); 2104f37938e0SEric Sandeen /* 2105f37938e0SEric Sandeen * we set the i_size on the btree inode to the max possible int. 2106f37938e0SEric Sandeen * the real end of the address space is determined by all of 2107f37938e0SEric Sandeen * the devices in the system 2108f37938e0SEric Sandeen */ 21092ff7e61eSJeff Mahoney inode->i_size = OFFSET_MAX; 21102ff7e61eSJeff Mahoney inode->i_mapping->a_ops = &btree_aops; 2111f37938e0SEric Sandeen 21122ff7e61eSJeff Mahoney RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 211343eb5f29SQu Wenruo extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree, 211443eb5f29SQu Wenruo IO_TREE_INODE_IO, inode); 21157b439738SDavid Sterba BTRFS_I(inode)->io_tree.track_uptodate = false; 21162ff7e61eSJeff Mahoney extent_map_tree_init(&BTRFS_I(inode)->extent_tree); 2117f37938e0SEric Sandeen 21182ff7e61eSJeff Mahoney BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops; 2119f37938e0SEric Sandeen 21202ff7e61eSJeff Mahoney BTRFS_I(inode)->root = fs_info->tree_root; 21212ff7e61eSJeff Mahoney memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key)); 21222ff7e61eSJeff Mahoney set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); 21232ff7e61eSJeff Mahoney btrfs_insert_inode_hash(inode); 2124f37938e0SEric Sandeen } 2125f37938e0SEric Sandeen 2126ad618368SEric Sandeen static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2127ad618368SEric Sandeen { 2128ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2129129827e3SDavid Sterba init_rwsem(&fs_info->dev_replace.rwsem); 21307f8d236aSDavid Sterba init_waitqueue_head(&fs_info->dev_replace.replace_wait); 2131ad618368SEric Sandeen } 2132ad618368SEric Sandeen 2133f9e92e40SEric Sandeen static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2134f9e92e40SEric Sandeen { 2135f9e92e40SEric Sandeen spin_lock_init(&fs_info->qgroup_lock); 2136f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_ioctl_lock); 2137f9e92e40SEric Sandeen fs_info->qgroup_tree = RB_ROOT; 2138f9e92e40SEric Sandeen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2139f9e92e40SEric Sandeen fs_info->qgroup_seq = 1; 2140f9e92e40SEric Sandeen fs_info->qgroup_ulist = NULL; 2141d2c609b8SJeff Mahoney fs_info->qgroup_rescan_running = false; 2142f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_rescan_lock); 2143f9e92e40SEric Sandeen } 2144f9e92e40SEric Sandeen 21452a458198SEric Sandeen static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info, 21462a458198SEric Sandeen struct btrfs_fs_devices *fs_devices) 21472a458198SEric Sandeen { 2148f7b885beSAnand Jain u32 max_active = fs_info->thread_pool_size; 21496f011058SDavid Sterba unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 21502a458198SEric Sandeen 21512a458198SEric Sandeen fs_info->workers = 2152cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "worker", 2153cb001095SJeff Mahoney flags | WQ_HIGHPRI, max_active, 16); 21542a458198SEric Sandeen 21552a458198SEric Sandeen fs_info->delalloc_workers = 2156cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delalloc", 2157cb001095SJeff Mahoney flags, max_active, 2); 21582a458198SEric Sandeen 21592a458198SEric Sandeen fs_info->flush_workers = 2160cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "flush_delalloc", 2161cb001095SJeff Mahoney flags, max_active, 0); 21622a458198SEric Sandeen 21632a458198SEric Sandeen fs_info->caching_workers = 2164cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); 21652a458198SEric Sandeen 21662a458198SEric Sandeen fs_info->fixup_workers = 2167cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); 21682a458198SEric Sandeen 21692a458198SEric Sandeen /* 21702a458198SEric Sandeen * endios are largely parallel and should have a very 21712a458198SEric Sandeen * low idle thresh 21722a458198SEric Sandeen */ 21732a458198SEric Sandeen fs_info->endio_workers = 2174cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4); 21752a458198SEric Sandeen fs_info->endio_meta_workers = 2176cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta", flags, 2177cb001095SJeff Mahoney max_active, 4); 21782a458198SEric Sandeen fs_info->endio_meta_write_workers = 2179cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags, 2180cb001095SJeff Mahoney max_active, 2); 21812a458198SEric Sandeen fs_info->endio_raid56_workers = 2182cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-raid56", flags, 2183cb001095SJeff Mahoney max_active, 4); 21842a458198SEric Sandeen fs_info->endio_repair_workers = 2185cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0); 21862a458198SEric Sandeen fs_info->rmw_workers = 2187cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2); 21882a458198SEric Sandeen fs_info->endio_write_workers = 2189cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-write", flags, 2190cb001095SJeff Mahoney max_active, 2); 21912a458198SEric Sandeen fs_info->endio_freespace_worker = 2192cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "freespace-write", flags, 2193cb001095SJeff Mahoney max_active, 0); 21942a458198SEric Sandeen fs_info->delayed_workers = 2195cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, 2196cb001095SJeff Mahoney max_active, 0); 21972a458198SEric Sandeen fs_info->readahead_workers = 2198cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "readahead", flags, 2199cb001095SJeff Mahoney max_active, 2); 22002a458198SEric Sandeen fs_info->qgroup_rescan_workers = 2201cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); 22022a458198SEric Sandeen 22032a458198SEric Sandeen if (!(fs_info->workers && fs_info->delalloc_workers && 2204ba8a9d07SChris Mason fs_info->flush_workers && 22052a458198SEric Sandeen fs_info->endio_workers && fs_info->endio_meta_workers && 22062a458198SEric Sandeen fs_info->endio_meta_write_workers && 22072a458198SEric Sandeen fs_info->endio_repair_workers && 22082a458198SEric Sandeen fs_info->endio_write_workers && fs_info->endio_raid56_workers && 22092a458198SEric Sandeen fs_info->endio_freespace_worker && fs_info->rmw_workers && 22102a458198SEric Sandeen fs_info->caching_workers && fs_info->readahead_workers && 22112a458198SEric Sandeen fs_info->fixup_workers && fs_info->delayed_workers && 22122a458198SEric Sandeen fs_info->qgroup_rescan_workers)) { 22132a458198SEric Sandeen return -ENOMEM; 22142a458198SEric Sandeen } 22152a458198SEric Sandeen 22162a458198SEric Sandeen return 0; 22172a458198SEric Sandeen } 22182a458198SEric Sandeen 22196d97c6e3SJohannes Thumshirn static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type) 22206d97c6e3SJohannes Thumshirn { 22216d97c6e3SJohannes Thumshirn struct crypto_shash *csum_shash; 22226d97c6e3SJohannes Thumshirn const char *csum_name = btrfs_super_csum_name(csum_type); 22236d97c6e3SJohannes Thumshirn 22246d97c6e3SJohannes Thumshirn csum_shash = crypto_alloc_shash(csum_name, 0, 0); 22256d97c6e3SJohannes Thumshirn 22266d97c6e3SJohannes Thumshirn if (IS_ERR(csum_shash)) { 22276d97c6e3SJohannes Thumshirn btrfs_err(fs_info, "error allocating %s hash for checksum", 22286d97c6e3SJohannes Thumshirn csum_name); 22296d97c6e3SJohannes Thumshirn return PTR_ERR(csum_shash); 22306d97c6e3SJohannes Thumshirn } 22316d97c6e3SJohannes Thumshirn 22326d97c6e3SJohannes Thumshirn fs_info->csum_shash = csum_shash; 22336d97c6e3SJohannes Thumshirn 22346d97c6e3SJohannes Thumshirn return 0; 22356d97c6e3SJohannes Thumshirn } 22366d97c6e3SJohannes Thumshirn 22376d97c6e3SJohannes Thumshirn static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info) 22386d97c6e3SJohannes Thumshirn { 22396d97c6e3SJohannes Thumshirn crypto_free_shash(fs_info->csum_shash); 22406d97c6e3SJohannes Thumshirn } 22416d97c6e3SJohannes Thumshirn 224263443bf5SEric Sandeen static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 224363443bf5SEric Sandeen struct btrfs_fs_devices *fs_devices) 224463443bf5SEric Sandeen { 224563443bf5SEric Sandeen int ret; 224663443bf5SEric Sandeen struct btrfs_root *log_tree_root; 224763443bf5SEric Sandeen struct btrfs_super_block *disk_super = fs_info->super_copy; 224863443bf5SEric Sandeen u64 bytenr = btrfs_super_log_root(disk_super); 2249581c1760SQu Wenruo int level = btrfs_super_log_root_level(disk_super); 225063443bf5SEric Sandeen 225163443bf5SEric Sandeen if (fs_devices->rw_devices == 0) { 2252f14d104dSDavid Sterba btrfs_warn(fs_info, "log replay required on RO media"); 225363443bf5SEric Sandeen return -EIO; 225463443bf5SEric Sandeen } 225563443bf5SEric Sandeen 225674e4d827SDavid Sterba log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL); 225763443bf5SEric Sandeen if (!log_tree_root) 225863443bf5SEric Sandeen return -ENOMEM; 225963443bf5SEric Sandeen 2260da17066cSJeff Mahoney __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID); 226163443bf5SEric Sandeen 22622ff7e61eSJeff Mahoney log_tree_root->node = read_tree_block(fs_info, bytenr, 2263581c1760SQu Wenruo fs_info->generation + 1, 2264581c1760SQu Wenruo level, NULL); 226564c043deSLiu Bo if (IS_ERR(log_tree_root->node)) { 2266f14d104dSDavid Sterba btrfs_warn(fs_info, "failed to read log tree"); 22670eeff236SLiu Bo ret = PTR_ERR(log_tree_root->node); 226864c043deSLiu Bo kfree(log_tree_root); 22690eeff236SLiu Bo return ret; 227064c043deSLiu Bo } else if (!extent_buffer_uptodate(log_tree_root->node)) { 2271f14d104dSDavid Sterba btrfs_err(fs_info, "failed to read log tree"); 227263443bf5SEric Sandeen free_extent_buffer(log_tree_root->node); 227363443bf5SEric Sandeen kfree(log_tree_root); 227463443bf5SEric Sandeen return -EIO; 227563443bf5SEric Sandeen } 227663443bf5SEric Sandeen /* returns with log_tree_root freed on success */ 227763443bf5SEric Sandeen ret = btrfs_recover_log_trees(log_tree_root); 227863443bf5SEric Sandeen if (ret) { 22790b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 228063443bf5SEric Sandeen "Failed to recover log tree"); 228163443bf5SEric Sandeen free_extent_buffer(log_tree_root->node); 228263443bf5SEric Sandeen kfree(log_tree_root); 228363443bf5SEric Sandeen return ret; 228463443bf5SEric Sandeen } 228563443bf5SEric Sandeen 2286bc98a42cSDavid Howells if (sb_rdonly(fs_info->sb)) { 22876bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 228863443bf5SEric Sandeen if (ret) 228963443bf5SEric Sandeen return ret; 229063443bf5SEric Sandeen } 229163443bf5SEric Sandeen 229263443bf5SEric Sandeen return 0; 229363443bf5SEric Sandeen } 229463443bf5SEric Sandeen 22956bccf3abSJeff Mahoney static int btrfs_read_roots(struct btrfs_fs_info *fs_info) 22964bbcaa64SEric Sandeen { 22976bccf3abSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 2298a4f3d2c4SDavid Sterba struct btrfs_root *root; 22994bbcaa64SEric Sandeen struct btrfs_key location; 23004bbcaa64SEric Sandeen int ret; 23014bbcaa64SEric Sandeen 23026bccf3abSJeff Mahoney BUG_ON(!fs_info->tree_root); 23036bccf3abSJeff Mahoney 23044bbcaa64SEric Sandeen location.objectid = BTRFS_EXTENT_TREE_OBJECTID; 23054bbcaa64SEric Sandeen location.type = BTRFS_ROOT_ITEM_KEY; 23064bbcaa64SEric Sandeen location.offset = 0; 23074bbcaa64SEric Sandeen 2308a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2309f50f4353SLiu Bo if (IS_ERR(root)) { 2310f50f4353SLiu Bo ret = PTR_ERR(root); 2311f50f4353SLiu Bo goto out; 2312f50f4353SLiu Bo } 2313a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2314a4f3d2c4SDavid Sterba fs_info->extent_root = root; 23154bbcaa64SEric Sandeen 23164bbcaa64SEric Sandeen location.objectid = BTRFS_DEV_TREE_OBJECTID; 2317a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2318f50f4353SLiu Bo if (IS_ERR(root)) { 2319f50f4353SLiu Bo ret = PTR_ERR(root); 2320f50f4353SLiu Bo goto out; 2321f50f4353SLiu Bo } 2322a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2323a4f3d2c4SDavid Sterba fs_info->dev_root = root; 23244bbcaa64SEric Sandeen btrfs_init_devices_late(fs_info); 23254bbcaa64SEric Sandeen 23264bbcaa64SEric Sandeen location.objectid = BTRFS_CSUM_TREE_OBJECTID; 2327a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2328f50f4353SLiu Bo if (IS_ERR(root)) { 2329f50f4353SLiu Bo ret = PTR_ERR(root); 2330f50f4353SLiu Bo goto out; 2331f50f4353SLiu Bo } 2332a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2333a4f3d2c4SDavid Sterba fs_info->csum_root = root; 23344bbcaa64SEric Sandeen 23354bbcaa64SEric Sandeen location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2336a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2337a4f3d2c4SDavid Sterba if (!IS_ERR(root)) { 2338a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2339afcdd129SJosef Bacik set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); 2340a4f3d2c4SDavid Sterba fs_info->quota_root = root; 23414bbcaa64SEric Sandeen } 23424bbcaa64SEric Sandeen 23434bbcaa64SEric Sandeen location.objectid = BTRFS_UUID_TREE_OBJECTID; 2344a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2345a4f3d2c4SDavid Sterba if (IS_ERR(root)) { 2346a4f3d2c4SDavid Sterba ret = PTR_ERR(root); 23474bbcaa64SEric Sandeen if (ret != -ENOENT) 2348f50f4353SLiu Bo goto out; 23494bbcaa64SEric Sandeen } else { 2350a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2351a4f3d2c4SDavid Sterba fs_info->uuid_root = root; 23524bbcaa64SEric Sandeen } 23534bbcaa64SEric Sandeen 235470f6d82eSOmar Sandoval if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 235570f6d82eSOmar Sandoval location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID; 235670f6d82eSOmar Sandoval root = btrfs_read_tree_root(tree_root, &location); 2357f50f4353SLiu Bo if (IS_ERR(root)) { 2358f50f4353SLiu Bo ret = PTR_ERR(root); 2359f50f4353SLiu Bo goto out; 2360f50f4353SLiu Bo } 236170f6d82eSOmar Sandoval set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 236270f6d82eSOmar Sandoval fs_info->free_space_root = root; 236370f6d82eSOmar Sandoval } 236470f6d82eSOmar Sandoval 23654bbcaa64SEric Sandeen return 0; 2366f50f4353SLiu Bo out: 2367f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", 2368f50f4353SLiu Bo location.objectid, ret); 2369f50f4353SLiu Bo return ret; 23704bbcaa64SEric Sandeen } 23714bbcaa64SEric Sandeen 2372069ec957SQu Wenruo /* 2373069ec957SQu Wenruo * Real super block validation 2374069ec957SQu Wenruo * NOTE: super csum type and incompat features will not be checked here. 2375069ec957SQu Wenruo * 2376069ec957SQu Wenruo * @sb: super block to check 2377069ec957SQu Wenruo * @mirror_num: the super block number to check its bytenr: 2378069ec957SQu Wenruo * 0 the primary (1st) sb 2379069ec957SQu Wenruo * 1, 2 2nd and 3rd backup copy 2380069ec957SQu Wenruo * -1 skip bytenr check 2381069ec957SQu Wenruo */ 2382069ec957SQu Wenruo static int validate_super(struct btrfs_fs_info *fs_info, 2383069ec957SQu Wenruo struct btrfs_super_block *sb, int mirror_num) 238421a852b0SQu Wenruo { 238521a852b0SQu Wenruo u64 nodesize = btrfs_super_nodesize(sb); 238621a852b0SQu Wenruo u64 sectorsize = btrfs_super_sectorsize(sb); 238721a852b0SQu Wenruo int ret = 0; 238821a852b0SQu Wenruo 238921a852b0SQu Wenruo if (btrfs_super_magic(sb) != BTRFS_MAGIC) { 239021a852b0SQu Wenruo btrfs_err(fs_info, "no valid FS found"); 239121a852b0SQu Wenruo ret = -EINVAL; 239221a852b0SQu Wenruo } 239321a852b0SQu Wenruo if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { 239421a852b0SQu Wenruo btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", 239521a852b0SQu Wenruo btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); 239621a852b0SQu Wenruo ret = -EINVAL; 239721a852b0SQu Wenruo } 239821a852b0SQu Wenruo if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 239921a852b0SQu Wenruo btrfs_err(fs_info, "tree_root level too big: %d >= %d", 240021a852b0SQu Wenruo btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 240121a852b0SQu Wenruo ret = -EINVAL; 240221a852b0SQu Wenruo } 240321a852b0SQu Wenruo if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 240421a852b0SQu Wenruo btrfs_err(fs_info, "chunk_root level too big: %d >= %d", 240521a852b0SQu Wenruo btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 240621a852b0SQu Wenruo ret = -EINVAL; 240721a852b0SQu Wenruo } 240821a852b0SQu Wenruo if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 240921a852b0SQu Wenruo btrfs_err(fs_info, "log_root level too big: %d >= %d", 241021a852b0SQu Wenruo btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 241121a852b0SQu Wenruo ret = -EINVAL; 241221a852b0SQu Wenruo } 241321a852b0SQu Wenruo 241421a852b0SQu Wenruo /* 241521a852b0SQu Wenruo * Check sectorsize and nodesize first, other check will need it. 241621a852b0SQu Wenruo * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. 241721a852b0SQu Wenruo */ 241821a852b0SQu Wenruo if (!is_power_of_2(sectorsize) || sectorsize < 4096 || 241921a852b0SQu Wenruo sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { 242021a852b0SQu Wenruo btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); 242121a852b0SQu Wenruo ret = -EINVAL; 242221a852b0SQu Wenruo } 242321a852b0SQu Wenruo /* Only PAGE SIZE is supported yet */ 242421a852b0SQu Wenruo if (sectorsize != PAGE_SIZE) { 242521a852b0SQu Wenruo btrfs_err(fs_info, 242621a852b0SQu Wenruo "sectorsize %llu not supported yet, only support %lu", 242721a852b0SQu Wenruo sectorsize, PAGE_SIZE); 242821a852b0SQu Wenruo ret = -EINVAL; 242921a852b0SQu Wenruo } 243021a852b0SQu Wenruo if (!is_power_of_2(nodesize) || nodesize < sectorsize || 243121a852b0SQu Wenruo nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { 243221a852b0SQu Wenruo btrfs_err(fs_info, "invalid nodesize %llu", nodesize); 243321a852b0SQu Wenruo ret = -EINVAL; 243421a852b0SQu Wenruo } 243521a852b0SQu Wenruo if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { 243621a852b0SQu Wenruo btrfs_err(fs_info, "invalid leafsize %u, should be %llu", 243721a852b0SQu Wenruo le32_to_cpu(sb->__unused_leafsize), nodesize); 243821a852b0SQu Wenruo ret = -EINVAL; 243921a852b0SQu Wenruo } 244021a852b0SQu Wenruo 244121a852b0SQu Wenruo /* Root alignment check */ 244221a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { 244321a852b0SQu Wenruo btrfs_warn(fs_info, "tree_root block unaligned: %llu", 244421a852b0SQu Wenruo btrfs_super_root(sb)); 244521a852b0SQu Wenruo ret = -EINVAL; 244621a852b0SQu Wenruo } 244721a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { 244821a852b0SQu Wenruo btrfs_warn(fs_info, "chunk_root block unaligned: %llu", 244921a852b0SQu Wenruo btrfs_super_chunk_root(sb)); 245021a852b0SQu Wenruo ret = -EINVAL; 245121a852b0SQu Wenruo } 245221a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { 245321a852b0SQu Wenruo btrfs_warn(fs_info, "log_root block unaligned: %llu", 245421a852b0SQu Wenruo btrfs_super_log_root(sb)); 245521a852b0SQu Wenruo ret = -EINVAL; 245621a852b0SQu Wenruo } 245721a852b0SQu Wenruo 2458de37aa51SNikolay Borisov if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, 24597239ff4bSNikolay Borisov BTRFS_FSID_SIZE) != 0) { 246021a852b0SQu Wenruo btrfs_err(fs_info, 24617239ff4bSNikolay Borisov "dev_item UUID does not match metadata fsid: %pU != %pU", 2462de37aa51SNikolay Borisov fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); 246321a852b0SQu Wenruo ret = -EINVAL; 246421a852b0SQu Wenruo } 246521a852b0SQu Wenruo 246621a852b0SQu Wenruo /* 246721a852b0SQu Wenruo * Hint to catch really bogus numbers, bitflips or so, more exact checks are 246821a852b0SQu Wenruo * done later 246921a852b0SQu Wenruo */ 247021a852b0SQu Wenruo if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { 247121a852b0SQu Wenruo btrfs_err(fs_info, "bytes_used is too small %llu", 247221a852b0SQu Wenruo btrfs_super_bytes_used(sb)); 247321a852b0SQu Wenruo ret = -EINVAL; 247421a852b0SQu Wenruo } 247521a852b0SQu Wenruo if (!is_power_of_2(btrfs_super_stripesize(sb))) { 247621a852b0SQu Wenruo btrfs_err(fs_info, "invalid stripesize %u", 247721a852b0SQu Wenruo btrfs_super_stripesize(sb)); 247821a852b0SQu Wenruo ret = -EINVAL; 247921a852b0SQu Wenruo } 248021a852b0SQu Wenruo if (btrfs_super_num_devices(sb) > (1UL << 31)) 248121a852b0SQu Wenruo btrfs_warn(fs_info, "suspicious number of devices: %llu", 248221a852b0SQu Wenruo btrfs_super_num_devices(sb)); 248321a852b0SQu Wenruo if (btrfs_super_num_devices(sb) == 0) { 248421a852b0SQu Wenruo btrfs_err(fs_info, "number of devices is 0"); 248521a852b0SQu Wenruo ret = -EINVAL; 248621a852b0SQu Wenruo } 248721a852b0SQu Wenruo 2488069ec957SQu Wenruo if (mirror_num >= 0 && 2489069ec957SQu Wenruo btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { 249021a852b0SQu Wenruo btrfs_err(fs_info, "super offset mismatch %llu != %u", 249121a852b0SQu Wenruo btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 249221a852b0SQu Wenruo ret = -EINVAL; 249321a852b0SQu Wenruo } 249421a852b0SQu Wenruo 249521a852b0SQu Wenruo /* 249621a852b0SQu Wenruo * Obvious sys_chunk_array corruptions, it must hold at least one key 249721a852b0SQu Wenruo * and one chunk 249821a852b0SQu Wenruo */ 249921a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 250021a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too big %u > %u", 250121a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 250221a852b0SQu Wenruo BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 250321a852b0SQu Wenruo ret = -EINVAL; 250421a852b0SQu Wenruo } 250521a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 250621a852b0SQu Wenruo + sizeof(struct btrfs_chunk)) { 250721a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too small %u < %zu", 250821a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 250921a852b0SQu Wenruo sizeof(struct btrfs_disk_key) 251021a852b0SQu Wenruo + sizeof(struct btrfs_chunk)); 251121a852b0SQu Wenruo ret = -EINVAL; 251221a852b0SQu Wenruo } 251321a852b0SQu Wenruo 251421a852b0SQu Wenruo /* 251521a852b0SQu Wenruo * The generation is a global counter, we'll trust it more than the others 251621a852b0SQu Wenruo * but it's still possible that it's the one that's wrong. 251721a852b0SQu Wenruo */ 251821a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 251921a852b0SQu Wenruo btrfs_warn(fs_info, 252021a852b0SQu Wenruo "suspicious: generation < chunk_root_generation: %llu < %llu", 252121a852b0SQu Wenruo btrfs_super_generation(sb), 252221a852b0SQu Wenruo btrfs_super_chunk_root_generation(sb)); 252321a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 252421a852b0SQu Wenruo && btrfs_super_cache_generation(sb) != (u64)-1) 252521a852b0SQu Wenruo btrfs_warn(fs_info, 252621a852b0SQu Wenruo "suspicious: generation < cache_generation: %llu < %llu", 252721a852b0SQu Wenruo btrfs_super_generation(sb), 252821a852b0SQu Wenruo btrfs_super_cache_generation(sb)); 252921a852b0SQu Wenruo 253021a852b0SQu Wenruo return ret; 253121a852b0SQu Wenruo } 253221a852b0SQu Wenruo 2533069ec957SQu Wenruo /* 2534069ec957SQu Wenruo * Validation of super block at mount time. 2535069ec957SQu Wenruo * Some checks already done early at mount time, like csum type and incompat 2536069ec957SQu Wenruo * flags will be skipped. 2537069ec957SQu Wenruo */ 2538069ec957SQu Wenruo static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) 2539069ec957SQu Wenruo { 2540069ec957SQu Wenruo return validate_super(fs_info, fs_info->super_copy, 0); 2541069ec957SQu Wenruo } 2542069ec957SQu Wenruo 254375cb857dSQu Wenruo /* 254475cb857dSQu Wenruo * Validation of super block at write time. 254575cb857dSQu Wenruo * Some checks like bytenr check will be skipped as their values will be 254675cb857dSQu Wenruo * overwritten soon. 254775cb857dSQu Wenruo * Extra checks like csum type and incompat flags will be done here. 254875cb857dSQu Wenruo */ 254975cb857dSQu Wenruo static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, 255075cb857dSQu Wenruo struct btrfs_super_block *sb) 255175cb857dSQu Wenruo { 255275cb857dSQu Wenruo int ret; 255375cb857dSQu Wenruo 255475cb857dSQu Wenruo ret = validate_super(fs_info, sb, -1); 255575cb857dSQu Wenruo if (ret < 0) 255675cb857dSQu Wenruo goto out; 2557e7e16f48SJohannes Thumshirn if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) { 255875cb857dSQu Wenruo ret = -EUCLEAN; 255975cb857dSQu Wenruo btrfs_err(fs_info, "invalid csum type, has %u want %u", 256075cb857dSQu Wenruo btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); 256175cb857dSQu Wenruo goto out; 256275cb857dSQu Wenruo } 256375cb857dSQu Wenruo if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 256475cb857dSQu Wenruo ret = -EUCLEAN; 256575cb857dSQu Wenruo btrfs_err(fs_info, 256675cb857dSQu Wenruo "invalid incompat flags, has 0x%llx valid mask 0x%llx", 256775cb857dSQu Wenruo btrfs_super_incompat_flags(sb), 256875cb857dSQu Wenruo (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); 256975cb857dSQu Wenruo goto out; 257075cb857dSQu Wenruo } 257175cb857dSQu Wenruo out: 257275cb857dSQu Wenruo if (ret < 0) 257375cb857dSQu Wenruo btrfs_err(fs_info, 257475cb857dSQu Wenruo "super block corruption detected before writing it to disk"); 257575cb857dSQu Wenruo return ret; 257675cb857dSQu Wenruo } 257775cb857dSQu Wenruo 2578b105e927SDavid Sterba int __cold open_ctree(struct super_block *sb, 2579dfe25020SChris Mason struct btrfs_fs_devices *fs_devices, 2580dfe25020SChris Mason char *options) 2581eb60ceacSChris Mason { 2582db94535dSChris Mason u32 sectorsize; 2583db94535dSChris Mason u32 nodesize; 258487ee04ebSChris Mason u32 stripesize; 258584234f3aSYan Zheng u64 generation; 2586f2b636e8SJosef Bacik u64 features; 258751bce6c9SJohannes Thumshirn u16 csum_type; 25883de4586cSChris Mason struct btrfs_key location; 2589a061fc8dSChris Mason struct buffer_head *bh; 25904d34b278SIlya Dryomov struct btrfs_super_block *disk_super; 2591815745cfSAl Viro struct btrfs_fs_info *fs_info = btrfs_sb(sb); 2592f84a8bd6SAl Viro struct btrfs_root *tree_root; 25934d34b278SIlya Dryomov struct btrfs_root *chunk_root; 2594eb60ceacSChris Mason int ret; 2595e58ca020SYan int err = -EINVAL; 2596af31f5e5SChris Mason int num_backups_tried = 0; 2597af31f5e5SChris Mason int backup_index = 0; 25986675df31SOmar Sandoval int clear_free_space_tree = 0; 2599581c1760SQu Wenruo int level; 26004543df7eSChris Mason 260174e4d827SDavid Sterba tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL); 260274e4d827SDavid Sterba chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL); 2603cb517eabSMiao Xie if (!tree_root || !chunk_root) { 260439279cc3SChris Mason err = -ENOMEM; 260539279cc3SChris Mason goto fail; 260639279cc3SChris Mason } 260776dda93cSYan, Zheng 260876dda93cSYan, Zheng ret = init_srcu_struct(&fs_info->subvol_srcu); 260976dda93cSYan, Zheng if (ret) { 261076dda93cSYan, Zheng err = ret; 261176dda93cSYan, Zheng goto fail; 261276dda93cSYan, Zheng } 261376dda93cSYan, Zheng 26144297ff84SJosef Bacik ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL); 2615e2d84521SMiao Xie if (ret) { 2616e2d84521SMiao Xie err = ret; 26179e11ceeeSJan Kara goto fail_srcu; 2618e2d84521SMiao Xie } 26194297ff84SJosef Bacik 26204297ff84SJosef Bacik ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 26214297ff84SJosef Bacik if (ret) { 26224297ff84SJosef Bacik err = ret; 26234297ff84SJosef Bacik goto fail_dio_bytes; 26244297ff84SJosef Bacik } 262509cbfeafSKirill A. Shutemov fs_info->dirty_metadata_batch = PAGE_SIZE * 2626e2d84521SMiao Xie (1 + ilog2(nr_cpu_ids)); 2627e2d84521SMiao Xie 2628908c7f19STejun Heo ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 2629963d678bSMiao Xie if (ret) { 2630963d678bSMiao Xie err = ret; 2631963d678bSMiao Xie goto fail_dirty_metadata_bytes; 2632963d678bSMiao Xie } 2633963d678bSMiao Xie 26347f8d236aSDavid Sterba ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, 26357f8d236aSDavid Sterba GFP_KERNEL); 2636c404e0dcSMiao Xie if (ret) { 2637c404e0dcSMiao Xie err = ret; 2638c404e0dcSMiao Xie goto fail_delalloc_bytes; 2639c404e0dcSMiao Xie } 2640c404e0dcSMiao Xie 264176dda93cSYan, Zheng INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 2642f28491e0SJosef Bacik INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 26438fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 2644facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 264524bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 2646eb73c1b7SMiao Xie INIT_LIST_HEAD(&fs_info->delalloc_roots); 264711833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 2648eb73c1b7SMiao Xie spin_lock_init(&fs_info->delalloc_root_lock); 2649a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 265076dda93cSYan, Zheng spin_lock_init(&fs_info->fs_roots_radix_lock); 265124bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 26524cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 2653f29021b2SJan Schmidt spin_lock_init(&fs_info->tree_mod_seq_lock); 2654ceda0864SMiao Xie spin_lock_init(&fs_info->super_lock); 2655f28491e0SJosef Bacik spin_lock_init(&fs_info->buffer_lock); 265647ab2a6cSJosef Bacik spin_lock_init(&fs_info->unused_bgs_lock); 2657f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 2658d7c15171SZhao Lei mutex_init(&fs_info->unused_bg_unpin_mutex); 265967c5e7d4SFilipe Manana mutex_init(&fs_info->delete_unused_bgs_mutex); 26607585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 2661573bfb72SMiao Xie mutex_init(&fs_info->delalloc_root_mutex); 2662de98ced9SMiao Xie seqlock_init(&fs_info->profiles_lock); 266319c00ddcSChris Mason 26640b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 26656324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 2666f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 266747ab2a6cSJosef Bacik INIT_LIST_HEAD(&fs_info->unused_bgs); 2668c8bf1b67SDavid Sterba extent_map_tree_init(&fs_info->mapping_tree); 266966d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 267066d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 267166d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 267266d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 267366d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 267466d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 267566d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 2676ba2c4d4eSJosef Bacik btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, 2677ba2c4d4eSJosef Bacik BTRFS_BLOCK_RSV_DELREFS); 2678ba2c4d4eSJosef Bacik 2679771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 26804cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 26812fefd558SZhao Lei atomic_set(&fs_info->reada_works_cnt, 0); 2682034f784dSJosef Bacik atomic_set(&fs_info->nr_delayed_iputs, 0); 2683fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 2684e20d96d6SChris Mason fs_info->sb = sb; 268595ac567aSFilipe David Borba Manana fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 26869ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 26874cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 2688a5ed45f8SNikolay Borisov atomic64_set(&fs_info->free_chunk_space, 0); 2689f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 26908b87dc17SDavid Sterba fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 2691f8c269d7SDavid Sterba fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 269290519d66SArne Jansen /* readahead state */ 2693d0164adcSMel Gorman INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); 269490519d66SArne Jansen spin_lock_init(&fs_info->reada_lock); 2695fd708b81SJosef Bacik btrfs_init_ref_verify(fs_info); 2696c8b97818SChris Mason 2697b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 2698b34b086cSChris Mason num_online_cpus() + 2, 8); 26990afbaf8cSChris Mason 2700199c2a9cSMiao Xie INIT_LIST_HEAD(&fs_info->ordered_roots); 2701199c2a9cSMiao Xie spin_lock_init(&fs_info->ordered_root_lock); 270269fe2d75SJosef Bacik 270369fe2d75SJosef Bacik fs_info->btree_inode = new_inode(sb); 270469fe2d75SJosef Bacik if (!fs_info->btree_inode) { 270569fe2d75SJosef Bacik err = -ENOMEM; 270669fe2d75SJosef Bacik goto fail_bio_counter; 270769fe2d75SJosef Bacik } 270869fe2d75SJosef Bacik mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 270969fe2d75SJosef Bacik 271016cdcec7SMiao Xie fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 271174e4d827SDavid Sterba GFP_KERNEL); 271216cdcec7SMiao Xie if (!fs_info->delayed_root) { 271316cdcec7SMiao Xie err = -ENOMEM; 271416cdcec7SMiao Xie goto fail_iput; 271516cdcec7SMiao Xie } 271616cdcec7SMiao Xie btrfs_init_delayed_root(fs_info->delayed_root); 27173eaa2885SChris Mason 2718638aa7edSEric Sandeen btrfs_init_scrub(fs_info); 271921adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 272021adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 272121adbd5cSStefan Behrens #endif 2722779a65a4SEric Sandeen btrfs_init_balance(fs_info); 272321c7e756SMiao Xie btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work); 2724a2de733cSArne Jansen 27259f6d2510SDavid Sterba sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; 27269f6d2510SDavid Sterba sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); 2727a061fc8dSChris Mason 27286bccf3abSJeff Mahoney btrfs_init_btree_inode(fs_info); 272939279cc3SChris Mason 2730e02119d5SChris Mason spin_lock_init(&fs_info->block_group_cache_lock); 27316bef4d31SEric Paris fs_info->block_group_cache_tree = RB_ROOT; 2732a1897fddSLiu Bo fs_info->first_logical_byte = (u64)-1; 2733925baeddSChris Mason 273443eb5f29SQu Wenruo extent_io_tree_init(fs_info, &fs_info->freed_extents[0], 273543eb5f29SQu Wenruo IO_TREE_FS_INFO_FREED_EXTENTS0, NULL); 273643eb5f29SQu Wenruo extent_io_tree_init(fs_info, &fs_info->freed_extents[1], 273743eb5f29SQu Wenruo IO_TREE_FS_INFO_FREED_EXTENTS1, NULL); 273811833d66SYan Zheng fs_info->pinned_extents = &fs_info->freed_extents[0]; 2739afcdd129SJosef Bacik set_bit(BTRFS_FS_BARRIER, &fs_info->flags); 2740509659cdSChris Mason 27415a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 274279154b1bSChris Mason mutex_init(&fs_info->tree_log_mutex); 2743925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 2744a74a4b97SChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 2745a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 27461bbc621eSChris Mason mutex_init(&fs_info->ro_block_group_mutex); 27479e351cc8SJosef Bacik init_rwsem(&fs_info->commit_root_sem); 2748c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 274976dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 2750803b2f54SStefan Behrens sema_init(&fs_info->uuid_tree_rescan_sem, 1); 2751fa9c0d79SChris Mason 2752ad618368SEric Sandeen btrfs_init_dev_replace_locks(fs_info); 2753f9e92e40SEric Sandeen btrfs_init_qgroup(fs_info); 2754416ac51dSArne Jansen 2755fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2756fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2757fa9c0d79SChris Mason 2758e6dcd2dcSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 2759f9295749SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 2760bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 27614854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 2762034f784dSJosef Bacik init_waitqueue_head(&fs_info->delayed_iputs_wait); 27633768f368SChris Mason 2764da17066cSJeff Mahoney /* Usable values until the real ones are cached from the superblock */ 2765da17066cSJeff Mahoney fs_info->nodesize = 4096; 2766da17066cSJeff Mahoney fs_info->sectorsize = 4096; 2767da17066cSJeff Mahoney fs_info->stripesize = 4096; 2768da17066cSJeff Mahoney 2769eede2bf3SOmar Sandoval spin_lock_init(&fs_info->swapfile_pins_lock); 2770eede2bf3SOmar Sandoval fs_info->swapfile_pins = RB_ROOT; 2771eede2bf3SOmar Sandoval 27729e967495SFilipe Manana fs_info->send_in_progress = 0; 27739e967495SFilipe Manana 277453b381b3SDavid Woodhouse ret = btrfs_alloc_stripe_hash_table(fs_info); 277553b381b3SDavid Woodhouse if (ret) { 277683c8266aSDavid Sterba err = ret; 277753b381b3SDavid Woodhouse goto fail_alloc; 277853b381b3SDavid Woodhouse } 277953b381b3SDavid Woodhouse 2780da17066cSJeff Mahoney __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID); 27817eccb903SChris Mason 27823c4bb26bSChris Mason invalidate_bdev(fs_devices->latest_bdev); 27831104a885SDavid Sterba 27841104a885SDavid Sterba /* 27851104a885SDavid Sterba * Read super block and check the signature bytes only 27861104a885SDavid Sterba */ 2787a512bbf8SYan Zheng bh = btrfs_read_dev_super(fs_devices->latest_bdev); 278892fc03fbSAnand Jain if (IS_ERR(bh)) { 278992fc03fbSAnand Jain err = PTR_ERR(bh); 279016cdcec7SMiao Xie goto fail_alloc; 279120b45077SDave Young } 279239279cc3SChris Mason 27931104a885SDavid Sterba /* 27948dc3f22cSJohannes Thumshirn * Verify the type first, if that or the the checksum value are 27958dc3f22cSJohannes Thumshirn * corrupted, we'll find out 27968dc3f22cSJohannes Thumshirn */ 279751bce6c9SJohannes Thumshirn csum_type = btrfs_super_csum_type((struct btrfs_super_block *)bh->b_data); 279851bce6c9SJohannes Thumshirn if (!btrfs_supported_super_csum(csum_type)) { 27998dc3f22cSJohannes Thumshirn btrfs_err(fs_info, "unsupported checksum algorithm: %u", 280051bce6c9SJohannes Thumshirn csum_type); 28018dc3f22cSJohannes Thumshirn err = -EINVAL; 28028dc3f22cSJohannes Thumshirn brelse(bh); 28038dc3f22cSJohannes Thumshirn goto fail_alloc; 28048dc3f22cSJohannes Thumshirn } 28058dc3f22cSJohannes Thumshirn 28066d97c6e3SJohannes Thumshirn ret = btrfs_init_csum_hash(fs_info, csum_type); 28076d97c6e3SJohannes Thumshirn if (ret) { 28086d97c6e3SJohannes Thumshirn err = ret; 28096d97c6e3SJohannes Thumshirn goto fail_alloc; 28106d97c6e3SJohannes Thumshirn } 28116d97c6e3SJohannes Thumshirn 28128dc3f22cSJohannes Thumshirn /* 28131104a885SDavid Sterba * We want to check superblock checksum, the type is stored inside. 28141104a885SDavid Sterba * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 28151104a885SDavid Sterba */ 2816ab8d0fc4SJeff Mahoney if (btrfs_check_super_csum(fs_info, bh->b_data)) { 281705135f59SDavid Sterba btrfs_err(fs_info, "superblock checksum mismatch"); 28181104a885SDavid Sterba err = -EINVAL; 2819b2acdddfSAnand Jain brelse(bh); 28206d97c6e3SJohannes Thumshirn goto fail_csum; 28211104a885SDavid Sterba } 28221104a885SDavid Sterba 28231104a885SDavid Sterba /* 28241104a885SDavid Sterba * super_copy is zeroed at allocation time and we never touch the 28251104a885SDavid Sterba * following bytes up to INFO_SIZE, the checksum is calculated from 28261104a885SDavid Sterba * the whole block of INFO_SIZE 28271104a885SDavid Sterba */ 28286c41761fSDavid Sterba memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy)); 2829a061fc8dSChris Mason brelse(bh); 28305f39d397SChris Mason 2831fbc6feaeSNikolay Borisov disk_super = fs_info->super_copy; 2832fbc6feaeSNikolay Borisov 2833de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, 2834de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 2835de37aa51SNikolay Borisov 28367239ff4bSNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID)) { 2837de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid, 2838de37aa51SNikolay Borisov fs_info->super_copy->metadata_uuid, 2839de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 28407239ff4bSNikolay Borisov } 28410b86a832SChris Mason 2842fbc6feaeSNikolay Borisov features = btrfs_super_flags(disk_super); 2843fbc6feaeSNikolay Borisov if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { 2844fbc6feaeSNikolay Borisov features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; 2845fbc6feaeSNikolay Borisov btrfs_set_super_flags(disk_super, features); 2846fbc6feaeSNikolay Borisov btrfs_info(fs_info, 2847fbc6feaeSNikolay Borisov "found metadata UUID change in progress flag, clearing"); 2848fbc6feaeSNikolay Borisov } 2849fbc6feaeSNikolay Borisov 2850fbc6feaeSNikolay Borisov memcpy(fs_info->super_for_commit, fs_info->super_copy, 2851fbc6feaeSNikolay Borisov sizeof(*fs_info->super_for_commit)); 2852de37aa51SNikolay Borisov 2853069ec957SQu Wenruo ret = btrfs_validate_mount_super(fs_info); 28541104a885SDavid Sterba if (ret) { 285505135f59SDavid Sterba btrfs_err(fs_info, "superblock contains fatal errors"); 28561104a885SDavid Sterba err = -EINVAL; 28576d97c6e3SJohannes Thumshirn goto fail_csum; 28581104a885SDavid Sterba } 28591104a885SDavid Sterba 28600f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 28616d97c6e3SJohannes Thumshirn goto fail_csum; 28620f7d52f4SChris Mason 2863acce952bSliubo /* check FS state, whether FS is broken. */ 286487533c47SMiao Xie if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 286587533c47SMiao Xie set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 2866acce952bSliubo 286775e7cb7fSLiu Bo /* 2868af31f5e5SChris Mason * run through our array of backup supers and setup 2869af31f5e5SChris Mason * our ring pointer to the oldest one 2870af31f5e5SChris Mason */ 2871af31f5e5SChris Mason generation = btrfs_super_generation(disk_super); 2872af31f5e5SChris Mason find_oldest_super_backup(fs_info, generation); 2873af31f5e5SChris Mason 2874af31f5e5SChris Mason /* 287575e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 287675e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 287775e7cb7fSLiu Bo */ 287875e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 287975e7cb7fSLiu Bo 28802ff7e61eSJeff Mahoney ret = btrfs_parse_options(fs_info, options, sb->s_flags); 28812b82032cSYan Zheng if (ret) { 28822b82032cSYan Zheng err = ret; 28836d97c6e3SJohannes Thumshirn goto fail_csum; 28842b82032cSYan Zheng } 2885dfe25020SChris Mason 2886f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 2887f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 2888f2b636e8SJosef Bacik if (features) { 288905135f59SDavid Sterba btrfs_err(fs_info, 289005135f59SDavid Sterba "cannot mount because of unsupported optional features (%llx)", 2891c1c9ff7cSGeert Uytterhoeven features); 2892f2b636e8SJosef Bacik err = -EINVAL; 28936d97c6e3SJohannes Thumshirn goto fail_csum; 2894f2b636e8SJosef Bacik } 2895f2b636e8SJosef Bacik 28965d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 28975d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 28980b246afaSJeff Mahoney if (fs_info->compress_type == BTRFS_COMPRESS_LZO) 2899a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 29005c1aab1dSNick Terrell else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) 29015c1aab1dSNick Terrell features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; 2902727011e0SChris Mason 29033173a18fSJosef Bacik if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) 290405135f59SDavid Sterba btrfs_info(fs_info, "has skinny extents"); 29053173a18fSJosef Bacik 2906727011e0SChris Mason /* 2907727011e0SChris Mason * flag our filesystem as having big metadata blocks if 2908727011e0SChris Mason * they are bigger than the page size 2909727011e0SChris Mason */ 291009cbfeafSKirill A. Shutemov if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) { 2911727011e0SChris Mason if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 291205135f59SDavid Sterba btrfs_info(fs_info, 291305135f59SDavid Sterba "flagging fs with big metadata feature"); 2914727011e0SChris Mason features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 2915727011e0SChris Mason } 2916727011e0SChris Mason 2917bc3f116fSChris Mason nodesize = btrfs_super_nodesize(disk_super); 2918bc3f116fSChris Mason sectorsize = btrfs_super_sectorsize(disk_super); 2919b7f67055SChandan Rajendra stripesize = sectorsize; 2920707e8a07SDavid Sterba fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 2921963d678bSMiao Xie fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 2922bc3f116fSChris Mason 2923da17066cSJeff Mahoney /* Cache block sizes */ 2924da17066cSJeff Mahoney fs_info->nodesize = nodesize; 2925da17066cSJeff Mahoney fs_info->sectorsize = sectorsize; 2926da17066cSJeff Mahoney fs_info->stripesize = stripesize; 2927da17066cSJeff Mahoney 2928bc3f116fSChris Mason /* 2929bc3f116fSChris Mason * mixed block groups end up with duplicate but slightly offset 2930bc3f116fSChris Mason * extent buffers for the same range. It leads to corruptions 2931bc3f116fSChris Mason */ 2932bc3f116fSChris Mason if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 2933707e8a07SDavid Sterba (sectorsize != nodesize)) { 293405135f59SDavid Sterba btrfs_err(fs_info, 293505135f59SDavid Sterba "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", 293605135f59SDavid Sterba nodesize, sectorsize); 29376d97c6e3SJohannes Thumshirn goto fail_csum; 2938bc3f116fSChris Mason } 2939bc3f116fSChris Mason 2940ceda0864SMiao Xie /* 2941ceda0864SMiao Xie * Needn't use the lock because there is no other task which will 2942ceda0864SMiao Xie * update the flag. 2943ceda0864SMiao Xie */ 29445d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 29455d4f98a2SYan Zheng 2946f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 2947f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 2948bc98a42cSDavid Howells if (!sb_rdonly(sb) && features) { 294905135f59SDavid Sterba btrfs_err(fs_info, 295005135f59SDavid Sterba "cannot mount read-write because of unsupported optional features (%llx)", 2951c1c9ff7cSGeert Uytterhoeven features); 2952f2b636e8SJosef Bacik err = -EINVAL; 29536d97c6e3SJohannes Thumshirn goto fail_csum; 2954f2b636e8SJosef Bacik } 295561d92c32SChris Mason 29562a458198SEric Sandeen ret = btrfs_init_workqueues(fs_info, fs_devices); 29572a458198SEric Sandeen if (ret) { 29582a458198SEric Sandeen err = ret; 29590dc3b84aSJosef Bacik goto fail_sb_buffer; 29600dc3b84aSJosef Bacik } 29614543df7eSChris Mason 29629e11ceeeSJan Kara sb->s_bdi->congested_fn = btrfs_congested_fn; 29639e11ceeeSJan Kara sb->s_bdi->congested_data = fs_info; 29649e11ceeeSJan Kara sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK; 2965b5420237SNikolay Borisov sb->s_bdi->ra_pages = VM_READAHEAD_PAGES; 29669e11ceeeSJan Kara sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); 29679e11ceeeSJan Kara sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); 29684575c9ccSChris Mason 2969a061fc8dSChris Mason sb->s_blocksize = sectorsize; 2970a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 2971de37aa51SNikolay Borisov memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); 2972db94535dSChris Mason 2973925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 29746bccf3abSJeff Mahoney ret = btrfs_read_sys_array(fs_info); 2975925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 297684eed90fSChris Mason if (ret) { 297705135f59SDavid Sterba btrfs_err(fs_info, "failed to read the system array: %d", ret); 29785d4f98a2SYan Zheng goto fail_sb_buffer; 297984eed90fSChris Mason } 29800b86a832SChris Mason 298184234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 2982581c1760SQu Wenruo level = btrfs_super_chunk_root_level(disk_super); 29830b86a832SChris Mason 2984da17066cSJeff Mahoney __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID); 29850b86a832SChris Mason 29862ff7e61eSJeff Mahoney chunk_root->node = read_tree_block(fs_info, 29870b86a832SChris Mason btrfs_super_chunk_root(disk_super), 2988581c1760SQu Wenruo generation, level, NULL); 298964c043deSLiu Bo if (IS_ERR(chunk_root->node) || 299064c043deSLiu Bo !extent_buffer_uptodate(chunk_root->node)) { 299105135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk root"); 2992e5fffbacSchandan if (!IS_ERR(chunk_root->node)) 2993e5fffbacSchandan free_extent_buffer(chunk_root->node); 299495ab1f64SZhao Lei chunk_root->node = NULL; 2995af31f5e5SChris Mason goto fail_tree_roots; 299683121942SDavid Woodhouse } 29975d4f98a2SYan Zheng btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 29985d4f98a2SYan Zheng chunk_root->commit_root = btrfs_root_node(chunk_root); 29990b86a832SChris Mason 3000e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 3001b308bc2fSGeert Uytterhoeven btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE); 3002e17cade2SChris Mason 30035b4aacefSJeff Mahoney ret = btrfs_read_chunk_tree(fs_info); 30042b82032cSYan Zheng if (ret) { 300505135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk tree: %d", ret); 3006af31f5e5SChris Mason goto fail_tree_roots; 30072b82032cSYan Zheng } 30080b86a832SChris Mason 30098dabb742SStefan Behrens /* 30109b99b115SAnand Jain * Keep the devid that is marked to be the target device for the 30119b99b115SAnand Jain * device replace procedure 30128dabb742SStefan Behrens */ 30139b99b115SAnand Jain btrfs_free_extra_devids(fs_devices, 0); 3014dfe25020SChris Mason 3015a6b0d5c8SChris Mason if (!fs_devices->latest_bdev) { 301605135f59SDavid Sterba btrfs_err(fs_info, "failed to read devices"); 3017a6b0d5c8SChris Mason goto fail_tree_roots; 3018a6b0d5c8SChris Mason } 3019a6b0d5c8SChris Mason 3020af31f5e5SChris Mason retry_root_backup: 302184234f3aSYan Zheng generation = btrfs_super_generation(disk_super); 3022581c1760SQu Wenruo level = btrfs_super_root_level(disk_super); 30230b86a832SChris Mason 30242ff7e61eSJeff Mahoney tree_root->node = read_tree_block(fs_info, 3025db94535dSChris Mason btrfs_super_root(disk_super), 3026581c1760SQu Wenruo generation, level, NULL); 302764c043deSLiu Bo if (IS_ERR(tree_root->node) || 302864c043deSLiu Bo !extent_buffer_uptodate(tree_root->node)) { 302905135f59SDavid Sterba btrfs_warn(fs_info, "failed to read tree root"); 3030e5fffbacSchandan if (!IS_ERR(tree_root->node)) 3031e5fffbacSchandan free_extent_buffer(tree_root->node); 303295ab1f64SZhao Lei tree_root->node = NULL; 3033af31f5e5SChris Mason goto recovery_tree_root; 303483121942SDavid Woodhouse } 3035af31f5e5SChris Mason 30365d4f98a2SYan Zheng btrfs_set_root_node(&tree_root->root_item, tree_root->node); 30375d4f98a2SYan Zheng tree_root->commit_root = btrfs_root_node(tree_root); 303869e9c6c6SStefan Behrens btrfs_set_root_refs(&tree_root->root_item, 1); 3039db94535dSChris Mason 3040f32e48e9SChandan Rajendra mutex_lock(&tree_root->objectid_mutex); 3041f32e48e9SChandan Rajendra ret = btrfs_find_highest_objectid(tree_root, 3042f32e48e9SChandan Rajendra &tree_root->highest_objectid); 3043f32e48e9SChandan Rajendra if (ret) { 3044f32e48e9SChandan Rajendra mutex_unlock(&tree_root->objectid_mutex); 3045f32e48e9SChandan Rajendra goto recovery_tree_root; 3046f32e48e9SChandan Rajendra } 3047f32e48e9SChandan Rajendra 3048f32e48e9SChandan Rajendra ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 3049f32e48e9SChandan Rajendra 3050f32e48e9SChandan Rajendra mutex_unlock(&tree_root->objectid_mutex); 3051f32e48e9SChandan Rajendra 30526bccf3abSJeff Mahoney ret = btrfs_read_roots(fs_info); 30534bbcaa64SEric Sandeen if (ret) 3054af31f5e5SChris Mason goto recovery_tree_root; 3055f7a81ea4SStefan Behrens 30568929ecfaSYan, Zheng fs_info->generation = generation; 30578929ecfaSYan, Zheng fs_info->last_trans_committed = generation; 30588929ecfaSYan, Zheng 3059cf90d884SQu Wenruo ret = btrfs_verify_dev_extents(fs_info); 3060cf90d884SQu Wenruo if (ret) { 3061cf90d884SQu Wenruo btrfs_err(fs_info, 3062cf90d884SQu Wenruo "failed to verify dev extents against chunks: %d", 3063cf90d884SQu Wenruo ret); 3064cf90d884SQu Wenruo goto fail_block_groups; 3065cf90d884SQu Wenruo } 306668310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 306768310a5eSIlya Dryomov if (ret) { 306805135f59SDavid Sterba btrfs_err(fs_info, "failed to recover balance: %d", ret); 306968310a5eSIlya Dryomov goto fail_block_groups; 307068310a5eSIlya Dryomov } 307168310a5eSIlya Dryomov 3072733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 3073733f4fbbSStefan Behrens if (ret) { 307405135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_stats: %d", ret); 3075733f4fbbSStefan Behrens goto fail_block_groups; 3076733f4fbbSStefan Behrens } 3077733f4fbbSStefan Behrens 30788dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 30798dabb742SStefan Behrens if (ret) { 308005135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_replace: %d", ret); 30818dabb742SStefan Behrens goto fail_block_groups; 30828dabb742SStefan Behrens } 30838dabb742SStefan Behrens 30849b99b115SAnand Jain btrfs_free_extra_devids(fs_devices, 1); 30858dabb742SStefan Behrens 3086b7c35e81SAnand Jain ret = btrfs_sysfs_add_fsid(fs_devices, NULL); 3087b7c35e81SAnand Jain if (ret) { 308805135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", 308905135f59SDavid Sterba ret); 3090b7c35e81SAnand Jain goto fail_block_groups; 3091b7c35e81SAnand Jain } 3092b7c35e81SAnand Jain 3093b7c35e81SAnand Jain ret = btrfs_sysfs_add_device(fs_devices); 3094b7c35e81SAnand Jain if (ret) { 309505135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs device interface: %d", 309605135f59SDavid Sterba ret); 3097b7c35e81SAnand Jain goto fail_fsdev_sysfs; 3098b7c35e81SAnand Jain } 3099b7c35e81SAnand Jain 310096f3136eSAnand Jain ret = btrfs_sysfs_add_mounted(fs_info); 31015ac1d209SJeff Mahoney if (ret) { 310205135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); 3103b7c35e81SAnand Jain goto fail_fsdev_sysfs; 31045ac1d209SJeff Mahoney } 31055ac1d209SJeff Mahoney 3106c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 3107c59021f8Sliubo if (ret) { 310805135f59SDavid Sterba btrfs_err(fs_info, "failed to initialize space info: %d", ret); 31092365dd3cSAnand Jain goto fail_sysfs; 3110c59021f8Sliubo } 3111c59021f8Sliubo 31125b4aacefSJeff Mahoney ret = btrfs_read_block_groups(fs_info); 31131b1d1f66SJosef Bacik if (ret) { 311405135f59SDavid Sterba btrfs_err(fs_info, "failed to read block groups: %d", ret); 31152365dd3cSAnand Jain goto fail_sysfs; 31161b1d1f66SJosef Bacik } 31174330e183SQu Wenruo 31186528b99dSAnand Jain if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) { 311905135f59SDavid Sterba btrfs_warn(fs_info, 312052042d8eSAndrea Gelmini "writable mount is not allowed due to too many missing devices"); 31212365dd3cSAnand Jain goto fail_sysfs; 3122292fd7fcSStefan Behrens } 31239078a3e1SChris Mason 3124a74a4b97SChris Mason fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 3125a74a4b97SChris Mason "btrfs-cleaner"); 312657506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 31272365dd3cSAnand Jain goto fail_sysfs; 3128a74a4b97SChris Mason 3129a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 3130a74a4b97SChris Mason tree_root, 3131a74a4b97SChris Mason "btrfs-transaction"); 313257506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 31333f157a2fSChris Mason goto fail_cleaner; 3134a74a4b97SChris Mason 3135583b7231SHans van Kranenburg if (!btrfs_test_opt(fs_info, NOSSD) && 3136c289811cSChris Mason !fs_info->fs_devices->rotating) { 3137583b7231SHans van Kranenburg btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); 3138c289811cSChris Mason } 3139c289811cSChris Mason 3140572d9ab7SDavid Sterba /* 314101327610SNicholas D Steeves * Mount does not set all options immediately, we can do it now and do 3142572d9ab7SDavid Sterba * not have to wait for transaction commit 3143572d9ab7SDavid Sterba */ 3144572d9ab7SDavid Sterba btrfs_apply_pending_changes(fs_info); 31453818aea2SQu Wenruo 314621adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 31470b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { 31482ff7e61eSJeff Mahoney ret = btrfsic_mount(fs_info, fs_devices, 31490b246afaSJeff Mahoney btrfs_test_opt(fs_info, 315021adbd5cSStefan Behrens CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? 315121adbd5cSStefan Behrens 1 : 0, 315221adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 315321adbd5cSStefan Behrens if (ret) 315405135f59SDavid Sterba btrfs_warn(fs_info, 315505135f59SDavid Sterba "failed to initialize integrity check module: %d", 315605135f59SDavid Sterba ret); 315721adbd5cSStefan Behrens } 315821adbd5cSStefan Behrens #endif 3159bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 3160bcef60f2SArne Jansen if (ret) 3161bcef60f2SArne Jansen goto fail_trans_kthread; 316221adbd5cSStefan Behrens 3163fd708b81SJosef Bacik if (btrfs_build_ref_tree(fs_info)) 3164fd708b81SJosef Bacik btrfs_err(fs_info, "couldn't build ref tree"); 3165fd708b81SJosef Bacik 316696da0919SQu Wenruo /* do not make disk changes in broken FS or nologreplay is given */ 316796da0919SQu Wenruo if (btrfs_super_log_root(disk_super) != 0 && 31680b246afaSJeff Mahoney !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 316963443bf5SEric Sandeen ret = btrfs_replay_log(fs_info, fs_devices); 317079787eaaSJeff Mahoney if (ret) { 317163443bf5SEric Sandeen err = ret; 317228c16cbbSWang Shilong goto fail_qgroup; 3173e556ce2cSYan Zheng } 3174e02119d5SChris Mason } 31751a40e23bSZheng Yan 31766bccf3abSJeff Mahoney ret = btrfs_find_orphan_roots(fs_info); 317779787eaaSJeff Mahoney if (ret) 317828c16cbbSWang Shilong goto fail_qgroup; 317976dda93cSYan, Zheng 3180bc98a42cSDavid Howells if (!sb_rdonly(sb)) { 3181d68fc57bSYan, Zheng ret = btrfs_cleanup_fs_roots(fs_info); 318244c44af2SIlya Dryomov if (ret) 318328c16cbbSWang Shilong goto fail_qgroup; 318490c711abSZygo Blaxell 318590c711abSZygo Blaxell mutex_lock(&fs_info->cleaner_mutex); 31865d4f98a2SYan Zheng ret = btrfs_recover_relocation(tree_root); 318790c711abSZygo Blaxell mutex_unlock(&fs_info->cleaner_mutex); 3188d7ce5843SMiao Xie if (ret < 0) { 318905135f59SDavid Sterba btrfs_warn(fs_info, "failed to recover relocation: %d", 319005135f59SDavid Sterba ret); 3191d7ce5843SMiao Xie err = -EINVAL; 3192bcef60f2SArne Jansen goto fail_qgroup; 3193d7ce5843SMiao Xie } 31947c2ca468SChris Mason } 31951a40e23bSZheng Yan 31963de4586cSChris Mason location.objectid = BTRFS_FS_TREE_OBJECTID; 31973de4586cSChris Mason location.type = BTRFS_ROOT_ITEM_KEY; 3198cb517eabSMiao Xie location.offset = 0; 31993de4586cSChris Mason 32003de4586cSChris Mason fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location); 32013140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 32023140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 3203f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read fs tree: %d", err); 3204bcef60f2SArne Jansen goto fail_qgroup; 32053140c9a3SDan Carpenter } 3206c289811cSChris Mason 3207bc98a42cSDavid Howells if (sb_rdonly(sb)) 32082b6ba629SIlya Dryomov return 0; 32092b6ba629SIlya Dryomov 3210f8d468a1SOmar Sandoval if (btrfs_test_opt(fs_info, CLEAR_CACHE) && 3211f8d468a1SOmar Sandoval btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 32126675df31SOmar Sandoval clear_free_space_tree = 1; 32136675df31SOmar Sandoval } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 32146675df31SOmar Sandoval !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { 32156675df31SOmar Sandoval btrfs_warn(fs_info, "free space tree is invalid"); 32166675df31SOmar Sandoval clear_free_space_tree = 1; 32176675df31SOmar Sandoval } 32186675df31SOmar Sandoval 32196675df31SOmar Sandoval if (clear_free_space_tree) { 3220f8d468a1SOmar Sandoval btrfs_info(fs_info, "clearing free space tree"); 3221f8d468a1SOmar Sandoval ret = btrfs_clear_free_space_tree(fs_info); 3222f8d468a1SOmar Sandoval if (ret) { 3223f8d468a1SOmar Sandoval btrfs_warn(fs_info, 3224f8d468a1SOmar Sandoval "failed to clear free space tree: %d", ret); 32256bccf3abSJeff Mahoney close_ctree(fs_info); 3226f8d468a1SOmar Sandoval return ret; 3227f8d468a1SOmar Sandoval } 3228f8d468a1SOmar Sandoval } 3229f8d468a1SOmar Sandoval 32300b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && 3231511711afSChris Mason !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 323205135f59SDavid Sterba btrfs_info(fs_info, "creating free space tree"); 3233511711afSChris Mason ret = btrfs_create_free_space_tree(fs_info); 3234511711afSChris Mason if (ret) { 323505135f59SDavid Sterba btrfs_warn(fs_info, 323605135f59SDavid Sterba "failed to create free space tree: %d", ret); 32376bccf3abSJeff Mahoney close_ctree(fs_info); 3238511711afSChris Mason return ret; 3239511711afSChris Mason } 3240511711afSChris Mason } 3241511711afSChris Mason 3242e3acc2a6SJosef Bacik down_read(&fs_info->cleanup_work_sem); 32432b6ba629SIlya Dryomov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 32442b6ba629SIlya Dryomov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 32452b6ba629SIlya Dryomov up_read(&fs_info->cleanup_work_sem); 32466bccf3abSJeff Mahoney close_ctree(fs_info); 32472b6ba629SIlya Dryomov return ret; 32482b6ba629SIlya Dryomov } 3249e3acc2a6SJosef Bacik up_read(&fs_info->cleanup_work_sem); 325059641015SIlya Dryomov 32512b6ba629SIlya Dryomov ret = btrfs_resume_balance_async(fs_info); 32522b6ba629SIlya Dryomov if (ret) { 325305135f59SDavid Sterba btrfs_warn(fs_info, "failed to resume balance: %d", ret); 32546bccf3abSJeff Mahoney close_ctree(fs_info); 32552b6ba629SIlya Dryomov return ret; 3256e3acc2a6SJosef Bacik } 3257e3acc2a6SJosef Bacik 32588dabb742SStefan Behrens ret = btrfs_resume_dev_replace_async(fs_info); 32598dabb742SStefan Behrens if (ret) { 326005135f59SDavid Sterba btrfs_warn(fs_info, "failed to resume device replace: %d", ret); 32616bccf3abSJeff Mahoney close_ctree(fs_info); 32628dabb742SStefan Behrens return ret; 32638dabb742SStefan Behrens } 32648dabb742SStefan Behrens 3265b382a324SJan Schmidt btrfs_qgroup_rescan_resume(fs_info); 3266b382a324SJan Schmidt 32674bbcaa64SEric Sandeen if (!fs_info->uuid_root) { 326805135f59SDavid Sterba btrfs_info(fs_info, "creating UUID tree"); 3269f7a81ea4SStefan Behrens ret = btrfs_create_uuid_tree(fs_info); 3270f7a81ea4SStefan Behrens if (ret) { 327105135f59SDavid Sterba btrfs_warn(fs_info, 327205135f59SDavid Sterba "failed to create the UUID tree: %d", ret); 32736bccf3abSJeff Mahoney close_ctree(fs_info); 3274f7a81ea4SStefan Behrens return ret; 3275f7a81ea4SStefan Behrens } 32760b246afaSJeff Mahoney } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || 32774bbcaa64SEric Sandeen fs_info->generation != 32784bbcaa64SEric Sandeen btrfs_super_uuid_tree_generation(disk_super)) { 327905135f59SDavid Sterba btrfs_info(fs_info, "checking UUID tree"); 328070f80175SStefan Behrens ret = btrfs_check_uuid_tree(fs_info); 328170f80175SStefan Behrens if (ret) { 328205135f59SDavid Sterba btrfs_warn(fs_info, 328305135f59SDavid Sterba "failed to check the UUID tree: %d", ret); 32846bccf3abSJeff Mahoney close_ctree(fs_info); 328570f80175SStefan Behrens return ret; 328670f80175SStefan Behrens } 328770f80175SStefan Behrens } else { 3288afcdd129SJosef Bacik set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 3289f7a81ea4SStefan Behrens } 3290afcdd129SJosef Bacik set_bit(BTRFS_FS_OPEN, &fs_info->flags); 329147ab2a6cSJosef Bacik 32928dcddfa0SQu Wenruo /* 32938dcddfa0SQu Wenruo * backuproot only affect mount behavior, and if open_ctree succeeded, 32948dcddfa0SQu Wenruo * no need to keep the flag 32958dcddfa0SQu Wenruo */ 32968dcddfa0SQu Wenruo btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT); 32978dcddfa0SQu Wenruo 3298ad2b2c80SAl Viro return 0; 329939279cc3SChris Mason 3300bcef60f2SArne Jansen fail_qgroup: 3301bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 33027c2ca468SChris Mason fail_trans_kthread: 33037c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 33042ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 3305faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 33063f157a2fSChris Mason fail_cleaner: 3307a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 33087c2ca468SChris Mason 33097c2ca468SChris Mason /* 33107c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 33117c2ca468SChris Mason * kthreads 33127c2ca468SChris Mason */ 33137c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 33147c2ca468SChris Mason 33152365dd3cSAnand Jain fail_sysfs: 33166618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 33172365dd3cSAnand Jain 3318b7c35e81SAnand Jain fail_fsdev_sysfs: 3319b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3320b7c35e81SAnand Jain 33211b1d1f66SJosef Bacik fail_block_groups: 332254067ae9SJosef Bacik btrfs_put_block_group_cache(fs_info); 3323af31f5e5SChris Mason 3324af31f5e5SChris Mason fail_tree_roots: 33254273eaffSAnand Jain free_root_pointers(fs_info, true); 33262b8195bbSMiao Xie invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3327af31f5e5SChris Mason 332839279cc3SChris Mason fail_sb_buffer: 33297abadb64SLiu Bo btrfs_stop_all_workers(fs_info); 33305cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 33316d97c6e3SJohannes Thumshirn fail_csum: 33326d97c6e3SJohannes Thumshirn btrfs_free_csum_hash(fs_info); 333316cdcec7SMiao Xie fail_alloc: 33344543df7eSChris Mason fail_iput: 3335586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 3336586e46e2SIlya Dryomov 33374543df7eSChris Mason iput(fs_info->btree_inode); 3338c404e0dcSMiao Xie fail_bio_counter: 33397f8d236aSDavid Sterba percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 3340963d678bSMiao Xie fail_delalloc_bytes: 3341963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 3342e2d84521SMiao Xie fail_dirty_metadata_bytes: 3343e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 33444297ff84SJosef Bacik fail_dio_bytes: 33454297ff84SJosef Bacik percpu_counter_destroy(&fs_info->dio_bytes); 334676dda93cSYan, Zheng fail_srcu: 334776dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 33487e662854SQinghuang Feng fail: 334953b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 3350586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 3351ad2b2c80SAl Viro return err; 3352af31f5e5SChris Mason 3353af31f5e5SChris Mason recovery_tree_root: 33540b246afaSJeff Mahoney if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) 3355af31f5e5SChris Mason goto fail_tree_roots; 3356af31f5e5SChris Mason 33574273eaffSAnand Jain free_root_pointers(fs_info, false); 3358af31f5e5SChris Mason 3359af31f5e5SChris Mason /* don't use the log in recovery mode, it won't be valid */ 3360af31f5e5SChris Mason btrfs_set_super_log_root(disk_super, 0); 3361af31f5e5SChris Mason 3362af31f5e5SChris Mason /* we can't trust the free space cache either */ 3363af31f5e5SChris Mason btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 3364af31f5e5SChris Mason 3365af31f5e5SChris Mason ret = next_root_backup(fs_info, fs_info->super_copy, 3366af31f5e5SChris Mason &num_backups_tried, &backup_index); 3367af31f5e5SChris Mason if (ret == -1) 3368af31f5e5SChris Mason goto fail_block_groups; 3369af31f5e5SChris Mason goto retry_root_backup; 3370eb60ceacSChris Mason } 3371663faf9fSMasami Hiramatsu ALLOW_ERROR_INJECTION(open_ctree, ERRNO); 3372eb60ceacSChris Mason 3373f2984462SChris Mason static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate) 3374f2984462SChris Mason { 3375f2984462SChris Mason if (uptodate) { 3376f2984462SChris Mason set_buffer_uptodate(bh); 3377f2984462SChris Mason } else { 3378442a4f63SStefan Behrens struct btrfs_device *device = (struct btrfs_device *) 3379442a4f63SStefan Behrens bh->b_private; 3380442a4f63SStefan Behrens 3381fb456252SJeff Mahoney btrfs_warn_rl_in_rcu(device->fs_info, 3382b14af3b4SDavid Sterba "lost page write due to IO error on %s", 3383606686eeSJosef Bacik rcu_str_deref(device->name)); 338401327610SNicholas D Steeves /* note, we don't set_buffer_write_io_error because we have 33851259ab75SChris Mason * our own ways of dealing with the IO errors 33861259ab75SChris Mason */ 3387f2984462SChris Mason clear_buffer_uptodate(bh); 3388442a4f63SStefan Behrens btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS); 3389f2984462SChris Mason } 3390f2984462SChris Mason unlock_buffer(bh); 3391f2984462SChris Mason put_bh(bh); 3392f2984462SChris Mason } 3393f2984462SChris Mason 339429c36d72SAnand Jain int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num, 339529c36d72SAnand Jain struct buffer_head **bh_ret) 339629c36d72SAnand Jain { 339729c36d72SAnand Jain struct buffer_head *bh; 339829c36d72SAnand Jain struct btrfs_super_block *super; 339929c36d72SAnand Jain u64 bytenr; 340029c36d72SAnand Jain 340129c36d72SAnand Jain bytenr = btrfs_sb_offset(copy_num); 340229c36d72SAnand Jain if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode)) 340329c36d72SAnand Jain return -EINVAL; 340429c36d72SAnand Jain 34059f6d2510SDavid Sterba bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE); 340629c36d72SAnand Jain /* 340729c36d72SAnand Jain * If we fail to read from the underlying devices, as of now 340829c36d72SAnand Jain * the best option we have is to mark it EIO. 340929c36d72SAnand Jain */ 341029c36d72SAnand Jain if (!bh) 341129c36d72SAnand Jain return -EIO; 341229c36d72SAnand Jain 341329c36d72SAnand Jain super = (struct btrfs_super_block *)bh->b_data; 341429c36d72SAnand Jain if (btrfs_super_bytenr(super) != bytenr || 341529c36d72SAnand Jain btrfs_super_magic(super) != BTRFS_MAGIC) { 341629c36d72SAnand Jain brelse(bh); 341729c36d72SAnand Jain return -EINVAL; 341829c36d72SAnand Jain } 341929c36d72SAnand Jain 342029c36d72SAnand Jain *bh_ret = bh; 342129c36d72SAnand Jain return 0; 342229c36d72SAnand Jain } 342329c36d72SAnand Jain 342429c36d72SAnand Jain 3425a512bbf8SYan Zheng struct buffer_head *btrfs_read_dev_super(struct block_device *bdev) 3426a512bbf8SYan Zheng { 3427a512bbf8SYan Zheng struct buffer_head *bh; 3428a512bbf8SYan Zheng struct buffer_head *latest = NULL; 3429a512bbf8SYan Zheng struct btrfs_super_block *super; 3430a512bbf8SYan Zheng int i; 3431a512bbf8SYan Zheng u64 transid = 0; 343292fc03fbSAnand Jain int ret = -EINVAL; 3433a512bbf8SYan Zheng 3434a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 3435a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 3436a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 3437a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3438a512bbf8SYan Zheng */ 3439a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 344029c36d72SAnand Jain ret = btrfs_read_dev_one_super(bdev, i, &bh); 344129c36d72SAnand Jain if (ret) 3442a512bbf8SYan Zheng continue; 3443a512bbf8SYan Zheng 3444a512bbf8SYan Zheng super = (struct btrfs_super_block *)bh->b_data; 3445a512bbf8SYan Zheng 3446a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 3447a512bbf8SYan Zheng brelse(latest); 3448a512bbf8SYan Zheng latest = bh; 3449a512bbf8SYan Zheng transid = btrfs_super_generation(super); 3450a512bbf8SYan Zheng } else { 3451a512bbf8SYan Zheng brelse(bh); 3452a512bbf8SYan Zheng } 3453a512bbf8SYan Zheng } 345492fc03fbSAnand Jain 345592fc03fbSAnand Jain if (!latest) 345692fc03fbSAnand Jain return ERR_PTR(ret); 345792fc03fbSAnand Jain 3458a512bbf8SYan Zheng return latest; 3459a512bbf8SYan Zheng } 3460a512bbf8SYan Zheng 34614eedeb75SHisashi Hifumi /* 3462abbb3b8eSDavid Sterba * Write superblock @sb to the @device. Do not wait for completion, all the 3463abbb3b8eSDavid Sterba * buffer heads we write are pinned. 34644eedeb75SHisashi Hifumi * 3465abbb3b8eSDavid Sterba * Write @max_mirrors copies of the superblock, where 0 means default that fit 3466abbb3b8eSDavid Sterba * the expected device size at commit time. Note that max_mirrors must be 3467abbb3b8eSDavid Sterba * same for write and wait phases. 34684eedeb75SHisashi Hifumi * 3469abbb3b8eSDavid Sterba * Return number of errors when buffer head is not found or submission fails. 34704eedeb75SHisashi Hifumi */ 3471a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 3472abbb3b8eSDavid Sterba struct btrfs_super_block *sb, int max_mirrors) 3473a512bbf8SYan Zheng { 3474d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = device->fs_info; 3475d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 3476a512bbf8SYan Zheng struct buffer_head *bh; 3477a512bbf8SYan Zheng int i; 3478a512bbf8SYan Zheng int ret; 3479a512bbf8SYan Zheng int errors = 0; 3480a512bbf8SYan Zheng u64 bytenr; 34811b9e619cSOmar Sandoval int op_flags; 3482a512bbf8SYan Zheng 3483a512bbf8SYan Zheng if (max_mirrors == 0) 3484a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3485a512bbf8SYan Zheng 3486d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 3487d5178578SJohannes Thumshirn 3488a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 3489a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 3490935e5cc9SMiao Xie if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3491935e5cc9SMiao Xie device->commit_total_bytes) 3492a512bbf8SYan Zheng break; 3493a512bbf8SYan Zheng 3494a512bbf8SYan Zheng btrfs_set_super_bytenr(sb, bytenr); 3495a512bbf8SYan Zheng 3496d5178578SJohannes Thumshirn crypto_shash_init(shash); 3497d5178578SJohannes Thumshirn crypto_shash_update(shash, (const char *)sb + BTRFS_CSUM_SIZE, 3498abbb3b8eSDavid Sterba BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE); 3499d5178578SJohannes Thumshirn crypto_shash_final(shash, sb->csum); 3500a512bbf8SYan Zheng 3501abbb3b8eSDavid Sterba /* One reference for us, and we leave it for the caller */ 35029f6d2510SDavid Sterba bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, 3503a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 3504634554dcSJosef Bacik if (!bh) { 3505fb456252SJeff Mahoney btrfs_err(device->fs_info, 3506f14d104dSDavid Sterba "couldn't get super buffer head for bytenr %llu", 3507f14d104dSDavid Sterba bytenr); 3508634554dcSJosef Bacik errors++; 3509634554dcSJosef Bacik continue; 3510634554dcSJosef Bacik } 3511634554dcSJosef Bacik 3512a512bbf8SYan Zheng memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE); 3513a512bbf8SYan Zheng 35144eedeb75SHisashi Hifumi /* one reference for submit_bh */ 3515a512bbf8SYan Zheng get_bh(bh); 35164eedeb75SHisashi Hifumi 35174eedeb75SHisashi Hifumi set_buffer_uptodate(bh); 3518a512bbf8SYan Zheng lock_buffer(bh); 3519a512bbf8SYan Zheng bh->b_end_io = btrfs_end_buffer_write_sync; 3520442a4f63SStefan Behrens bh->b_private = device; 3521a512bbf8SYan Zheng 3522387125fcSChris Mason /* 3523387125fcSChris Mason * we fua the first super. The others we allow 3524387125fcSChris Mason * to go down lazy. 3525387125fcSChris Mason */ 35261b9e619cSOmar Sandoval op_flags = REQ_SYNC | REQ_META | REQ_PRIO; 35271b9e619cSOmar Sandoval if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) 35281b9e619cSOmar Sandoval op_flags |= REQ_FUA; 35291b9e619cSOmar Sandoval ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh); 35304eedeb75SHisashi Hifumi if (ret) 3531a512bbf8SYan Zheng errors++; 3532a512bbf8SYan Zheng } 3533a512bbf8SYan Zheng return errors < i ? 0 : -1; 3534a512bbf8SYan Zheng } 3535a512bbf8SYan Zheng 3536387125fcSChris Mason /* 3537abbb3b8eSDavid Sterba * Wait for write completion of superblocks done by write_dev_supers, 3538abbb3b8eSDavid Sterba * @max_mirrors same for write and wait phases. 3539abbb3b8eSDavid Sterba * 3540abbb3b8eSDavid Sterba * Return number of errors when buffer head is not found or not marked up to 3541abbb3b8eSDavid Sterba * date. 3542abbb3b8eSDavid Sterba */ 3543abbb3b8eSDavid Sterba static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) 3544abbb3b8eSDavid Sterba { 3545abbb3b8eSDavid Sterba struct buffer_head *bh; 3546abbb3b8eSDavid Sterba int i; 3547abbb3b8eSDavid Sterba int errors = 0; 3548b6a535faSHoward McLauchlan bool primary_failed = false; 3549abbb3b8eSDavid Sterba u64 bytenr; 3550abbb3b8eSDavid Sterba 3551abbb3b8eSDavid Sterba if (max_mirrors == 0) 3552abbb3b8eSDavid Sterba max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3553abbb3b8eSDavid Sterba 3554abbb3b8eSDavid Sterba for (i = 0; i < max_mirrors; i++) { 3555abbb3b8eSDavid Sterba bytenr = btrfs_sb_offset(i); 3556abbb3b8eSDavid Sterba if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3557abbb3b8eSDavid Sterba device->commit_total_bytes) 3558abbb3b8eSDavid Sterba break; 3559abbb3b8eSDavid Sterba 35609f6d2510SDavid Sterba bh = __find_get_block(device->bdev, 35619f6d2510SDavid Sterba bytenr / BTRFS_BDEV_BLOCKSIZE, 3562abbb3b8eSDavid Sterba BTRFS_SUPER_INFO_SIZE); 3563abbb3b8eSDavid Sterba if (!bh) { 3564abbb3b8eSDavid Sterba errors++; 3565b6a535faSHoward McLauchlan if (i == 0) 3566b6a535faSHoward McLauchlan primary_failed = true; 3567abbb3b8eSDavid Sterba continue; 3568abbb3b8eSDavid Sterba } 3569abbb3b8eSDavid Sterba wait_on_buffer(bh); 3570b6a535faSHoward McLauchlan if (!buffer_uptodate(bh)) { 3571abbb3b8eSDavid Sterba errors++; 3572b6a535faSHoward McLauchlan if (i == 0) 3573b6a535faSHoward McLauchlan primary_failed = true; 3574b6a535faSHoward McLauchlan } 3575abbb3b8eSDavid Sterba 3576abbb3b8eSDavid Sterba /* drop our reference */ 3577abbb3b8eSDavid Sterba brelse(bh); 3578abbb3b8eSDavid Sterba 3579abbb3b8eSDavid Sterba /* drop the reference from the writing run */ 3580abbb3b8eSDavid Sterba brelse(bh); 3581abbb3b8eSDavid Sterba } 3582abbb3b8eSDavid Sterba 3583b6a535faSHoward McLauchlan /* log error, force error return */ 3584b6a535faSHoward McLauchlan if (primary_failed) { 3585b6a535faSHoward McLauchlan btrfs_err(device->fs_info, "error writing primary super block to device %llu", 3586b6a535faSHoward McLauchlan device->devid); 3587b6a535faSHoward McLauchlan return -1; 3588b6a535faSHoward McLauchlan } 3589b6a535faSHoward McLauchlan 3590abbb3b8eSDavid Sterba return errors < i ? 0 : -1; 3591abbb3b8eSDavid Sterba } 3592abbb3b8eSDavid Sterba 3593abbb3b8eSDavid Sterba /* 3594387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 3595387125fcSChris Mason * for the barrier when it is done 3596387125fcSChris Mason */ 35974246a0b6SChristoph Hellwig static void btrfs_end_empty_barrier(struct bio *bio) 3598387125fcSChris Mason { 3599387125fcSChris Mason complete(bio->bi_private); 3600387125fcSChris Mason } 3601387125fcSChris Mason 3602387125fcSChris Mason /* 36034fc6441aSAnand Jain * Submit a flush request to the device if it supports it. Error handling is 36044fc6441aSAnand Jain * done in the waiting counterpart. 3605387125fcSChris Mason */ 36064fc6441aSAnand Jain static void write_dev_flush(struct btrfs_device *device) 3607387125fcSChris Mason { 3608c2a9c7abSAnand Jain struct request_queue *q = bdev_get_queue(device->bdev); 3609e0ae9994SDavid Sterba struct bio *bio = device->flush_bio; 3610387125fcSChris Mason 3611c2a9c7abSAnand Jain if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) 36124fc6441aSAnand Jain return; 3613387125fcSChris Mason 3614e0ae9994SDavid Sterba bio_reset(bio); 3615387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 361674d46992SChristoph Hellwig bio_set_dev(bio, device->bdev); 36178d910125SJan Kara bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH; 3618387125fcSChris Mason init_completion(&device->flush_wait); 3619387125fcSChris Mason bio->bi_private = &device->flush_wait; 3620387125fcSChris Mason 362143a01111SLu Fengqi btrfsic_submit_bio(bio); 36221c3063b6SAnand Jain set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 36234fc6441aSAnand Jain } 3624387125fcSChris Mason 36254fc6441aSAnand Jain /* 36264fc6441aSAnand Jain * If the flush bio has been submitted by write_dev_flush, wait for it. 36274fc6441aSAnand Jain */ 36288c27cb35SLinus Torvalds static blk_status_t wait_dev_flush(struct btrfs_device *device) 36294fc6441aSAnand Jain { 36304fc6441aSAnand Jain struct bio *bio = device->flush_bio; 36314fc6441aSAnand Jain 36321c3063b6SAnand Jain if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) 363358efbc9fSOmar Sandoval return BLK_STS_OK; 36344fc6441aSAnand Jain 36351c3063b6SAnand Jain clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 36362980d574SDavid Sterba wait_for_completion_io(&device->flush_wait); 36374fc6441aSAnand Jain 36388c27cb35SLinus Torvalds return bio->bi_status; 3639387125fcSChris Mason } 3640387125fcSChris Mason 3641d10b82feSQu Wenruo static int check_barrier_error(struct btrfs_fs_info *fs_info) 3642401b41e5SAnand Jain { 36436528b99dSAnand Jain if (!btrfs_check_rw_degradable(fs_info, NULL)) 3644401b41e5SAnand Jain return -EIO; 3645387125fcSChris Mason return 0; 3646387125fcSChris Mason } 3647387125fcSChris Mason 3648387125fcSChris Mason /* 3649387125fcSChris Mason * send an empty flush down to each device in parallel, 3650387125fcSChris Mason * then wait for them 3651387125fcSChris Mason */ 3652387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 3653387125fcSChris Mason { 3654387125fcSChris Mason struct list_head *head; 3655387125fcSChris Mason struct btrfs_device *dev; 36565af3e8ccSStefan Behrens int errors_wait = 0; 36574e4cbee9SChristoph Hellwig blk_status_t ret; 3658387125fcSChris Mason 36591538e6c5SDavid Sterba lockdep_assert_held(&info->fs_devices->device_list_mutex); 3660387125fcSChris Mason /* send down all the barriers */ 3661387125fcSChris Mason head = &info->fs_devices->devices; 36621538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3663e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3664f88ba6a2SHidetoshi Seto continue; 3665cea7c8bfSAnand Jain if (!dev->bdev) 3666387125fcSChris Mason continue; 3667e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3668ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3669387125fcSChris Mason continue; 3670387125fcSChris Mason 36714fc6441aSAnand Jain write_dev_flush(dev); 367258efbc9fSOmar Sandoval dev->last_flush_error = BLK_STS_OK; 3673387125fcSChris Mason } 3674387125fcSChris Mason 3675387125fcSChris Mason /* wait for all the barriers */ 36761538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3677e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3678f88ba6a2SHidetoshi Seto continue; 3679387125fcSChris Mason if (!dev->bdev) { 36805af3e8ccSStefan Behrens errors_wait++; 3681387125fcSChris Mason continue; 3682387125fcSChris Mason } 3683e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3684ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3685387125fcSChris Mason continue; 3686387125fcSChris Mason 36874fc6441aSAnand Jain ret = wait_dev_flush(dev); 3688401b41e5SAnand Jain if (ret) { 3689401b41e5SAnand Jain dev->last_flush_error = ret; 369066b4993eSDavid Sterba btrfs_dev_stat_inc_and_print(dev, 369166b4993eSDavid Sterba BTRFS_DEV_STAT_FLUSH_ERRS); 36925af3e8ccSStefan Behrens errors_wait++; 3693387125fcSChris Mason } 3694401b41e5SAnand Jain } 3695401b41e5SAnand Jain 3696cea7c8bfSAnand Jain if (errors_wait) { 3697401b41e5SAnand Jain /* 3698401b41e5SAnand Jain * At some point we need the status of all disks 3699401b41e5SAnand Jain * to arrive at the volume status. So error checking 3700401b41e5SAnand Jain * is being pushed to a separate loop. 3701401b41e5SAnand Jain */ 3702d10b82feSQu Wenruo return check_barrier_error(info); 3703401b41e5SAnand Jain } 3704387125fcSChris Mason return 0; 3705387125fcSChris Mason } 3706387125fcSChris Mason 3707943c6e99SZhao Lei int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 3708943c6e99SZhao Lei { 37098789f4feSZhao Lei int raid_type; 37108789f4feSZhao Lei int min_tolerated = INT_MAX; 3711943c6e99SZhao Lei 37128789f4feSZhao Lei if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 37138789f4feSZhao Lei (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 37148c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 37158789f4feSZhao Lei btrfs_raid_array[BTRFS_RAID_SINGLE]. 37168789f4feSZhao Lei tolerated_failures); 3717943c6e99SZhao Lei 37188789f4feSZhao Lei for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 37198789f4feSZhao Lei if (raid_type == BTRFS_RAID_SINGLE) 37208789f4feSZhao Lei continue; 372141a6e891SAnand Jain if (!(flags & btrfs_raid_array[raid_type].bg_flag)) 37228789f4feSZhao Lei continue; 37238c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 37248789f4feSZhao Lei btrfs_raid_array[raid_type]. 37258789f4feSZhao Lei tolerated_failures); 37268789f4feSZhao Lei } 3727943c6e99SZhao Lei 37288789f4feSZhao Lei if (min_tolerated == INT_MAX) { 3729ab8d0fc4SJeff Mahoney pr_warn("BTRFS: unknown raid flag: %llu", flags); 37308789f4feSZhao Lei min_tolerated = 0; 37318789f4feSZhao Lei } 37328789f4feSZhao Lei 37338789f4feSZhao Lei return min_tolerated; 3734943c6e99SZhao Lei } 3735943c6e99SZhao Lei 3736eece6a9cSDavid Sterba int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) 3737f2984462SChris Mason { 3738e5e9a520SChris Mason struct list_head *head; 3739f2984462SChris Mason struct btrfs_device *dev; 3740a061fc8dSChris Mason struct btrfs_super_block *sb; 3741f2984462SChris Mason struct btrfs_dev_item *dev_item; 3742f2984462SChris Mason int ret; 3743f2984462SChris Mason int do_barriers; 3744a236aed1SChris Mason int max_errors; 3745a236aed1SChris Mason int total_errors = 0; 3746a061fc8dSChris Mason u64 flags; 3747f2984462SChris Mason 37480b246afaSJeff Mahoney do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); 3749fed3b381SLiu Bo 3750fed3b381SLiu Bo /* 3751fed3b381SLiu Bo * max_mirrors == 0 indicates we're from commit_transaction, 3752fed3b381SLiu Bo * not from fsync where the tree roots in fs_info have not 3753fed3b381SLiu Bo * been consistent on disk. 3754fed3b381SLiu Bo */ 3755fed3b381SLiu Bo if (max_mirrors == 0) 37560b246afaSJeff Mahoney backup_super_roots(fs_info); 3757f2984462SChris Mason 37580b246afaSJeff Mahoney sb = fs_info->super_for_commit; 3759a061fc8dSChris Mason dev_item = &sb->dev_item; 3760e5e9a520SChris Mason 37610b246afaSJeff Mahoney mutex_lock(&fs_info->fs_devices->device_list_mutex); 37620b246afaSJeff Mahoney head = &fs_info->fs_devices->devices; 37630b246afaSJeff Mahoney max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; 3764387125fcSChris Mason 37655af3e8ccSStefan Behrens if (do_barriers) { 37660b246afaSJeff Mahoney ret = barrier_all_devices(fs_info); 37675af3e8ccSStefan Behrens if (ret) { 37685af3e8ccSStefan Behrens mutex_unlock( 37690b246afaSJeff Mahoney &fs_info->fs_devices->device_list_mutex); 37700b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 37715af3e8ccSStefan Behrens "errors while submitting device barriers."); 37725af3e8ccSStefan Behrens return ret; 37735af3e8ccSStefan Behrens } 37745af3e8ccSStefan Behrens } 3775387125fcSChris Mason 37761538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3777dfe25020SChris Mason if (!dev->bdev) { 3778dfe25020SChris Mason total_errors++; 3779dfe25020SChris Mason continue; 3780dfe25020SChris Mason } 3781e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3782ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3783dfe25020SChris Mason continue; 3784dfe25020SChris Mason 37852b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 3786a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 3787a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 37887df69d3eSMiao Xie btrfs_set_stack_device_total_bytes(dev_item, 3789935e5cc9SMiao Xie dev->commit_total_bytes); 3790ce7213c7SMiao Xie btrfs_set_stack_device_bytes_used(dev_item, 3791ce7213c7SMiao Xie dev->commit_bytes_used); 3792a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 3793a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 3794a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 3795a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 37967239ff4bSNikolay Borisov memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, 37977239ff4bSNikolay Borisov BTRFS_FSID_SIZE); 3798a512bbf8SYan Zheng 3799a061fc8dSChris Mason flags = btrfs_super_flags(sb); 3800a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 3801f2984462SChris Mason 380275cb857dSQu Wenruo ret = btrfs_validate_write_super(fs_info, sb); 380375cb857dSQu Wenruo if (ret < 0) { 380475cb857dSQu Wenruo mutex_unlock(&fs_info->fs_devices->device_list_mutex); 380575cb857dSQu Wenruo btrfs_handle_fs_error(fs_info, -EUCLEAN, 380675cb857dSQu Wenruo "unexpected superblock corruption detected"); 380775cb857dSQu Wenruo return -EUCLEAN; 380875cb857dSQu Wenruo } 380975cb857dSQu Wenruo 3810abbb3b8eSDavid Sterba ret = write_dev_supers(dev, sb, max_mirrors); 3811a236aed1SChris Mason if (ret) 3812a236aed1SChris Mason total_errors++; 3813f2984462SChris Mason } 3814a236aed1SChris Mason if (total_errors > max_errors) { 38150b246afaSJeff Mahoney btrfs_err(fs_info, "%d errors while writing supers", 3816d397712bSChris Mason total_errors); 38170b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 381879787eaaSJeff Mahoney 38199d565ba4SStefan Behrens /* FUA is masked off if unsupported and can't be the reason */ 38200b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 38210b246afaSJeff Mahoney "%d errors while writing supers", 38220b246afaSJeff Mahoney total_errors); 38239d565ba4SStefan Behrens return -EIO; 3824a236aed1SChris Mason } 3825f2984462SChris Mason 3826a512bbf8SYan Zheng total_errors = 0; 38271538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3828dfe25020SChris Mason if (!dev->bdev) 3829dfe25020SChris Mason continue; 3830e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3831ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3832dfe25020SChris Mason continue; 3833dfe25020SChris Mason 3834abbb3b8eSDavid Sterba ret = wait_dev_supers(dev, max_mirrors); 3835a512bbf8SYan Zheng if (ret) 38361259ab75SChris Mason total_errors++; 3837f2984462SChris Mason } 38380b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 3839a236aed1SChris Mason if (total_errors > max_errors) { 38400b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 38410b246afaSJeff Mahoney "%d errors while writing supers", 38420b246afaSJeff Mahoney total_errors); 384379787eaaSJeff Mahoney return -EIO; 3844a236aed1SChris Mason } 3845f2984462SChris Mason return 0; 3846f2984462SChris Mason } 3847f2984462SChris Mason 3848cb517eabSMiao Xie /* Drop a fs root from the radix tree and free it. */ 3849cb517eabSMiao Xie void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 3850cb517eabSMiao Xie struct btrfs_root *root) 38512619ba1fSChris Mason { 38524df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 38532619ba1fSChris Mason radix_tree_delete(&fs_info->fs_roots_radix, 38542619ba1fSChris Mason (unsigned long)root->root_key.objectid); 38554df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 385676dda93cSYan, Zheng 385776dda93cSYan, Zheng if (btrfs_root_refs(&root->root_item) == 0) 385876dda93cSYan, Zheng synchronize_srcu(&fs_info->subvol_srcu); 385976dda93cSYan, Zheng 38601c1ea4f7SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 38613321719eSLiu Bo btrfs_free_log(NULL, root); 38621c1ea4f7SLiu Bo if (root->reloc_root) { 38631c1ea4f7SLiu Bo free_extent_buffer(root->reloc_root->node); 38641c1ea4f7SLiu Bo free_extent_buffer(root->reloc_root->commit_root); 38651c1ea4f7SLiu Bo btrfs_put_fs_root(root->reloc_root); 38661c1ea4f7SLiu Bo root->reloc_root = NULL; 38671c1ea4f7SLiu Bo } 38681c1ea4f7SLiu Bo } 38693321719eSLiu Bo 3870faa2dbf0SJosef Bacik if (root->free_ino_pinned) 3871581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_pinned); 3872faa2dbf0SJosef Bacik if (root->free_ino_ctl) 3873581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_ctl); 387484db5ccfSDavid Sterba btrfs_free_fs_root(root); 38754df27c4dSYan, Zheng } 38764df27c4dSYan, Zheng 387784db5ccfSDavid Sterba void btrfs_free_fs_root(struct btrfs_root *root) 38784df27c4dSYan, Zheng { 387957cdc8dbSDavid Sterba iput(root->ino_cache_inode); 38804df27c4dSYan, Zheng WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 38810ee5dc67SAl Viro if (root->anon_dev) 38820ee5dc67SAl Viro free_anon_bdev(root->anon_dev); 38838257b2dcSMiao Xie if (root->subv_writers) 38848257b2dcSMiao Xie btrfs_free_subvolume_writers(root->subv_writers); 38855f39d397SChris Mason free_extent_buffer(root->node); 38865f39d397SChris Mason free_extent_buffer(root->commit_root); 3887581bb050SLi Zefan kfree(root->free_ino_ctl); 3888581bb050SLi Zefan kfree(root->free_ino_pinned); 3889b0feb9d9SMiao Xie btrfs_put_fs_root(root); 38902619ba1fSChris Mason } 38912619ba1fSChris Mason 3892c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 3893c146afadSYan Zheng { 3894c146afadSYan Zheng u64 root_objectid = 0; 3895c146afadSYan Zheng struct btrfs_root *gang[8]; 389665d33fd7SQu Wenruo int i = 0; 389765d33fd7SQu Wenruo int err = 0; 389865d33fd7SQu Wenruo unsigned int ret = 0; 389965d33fd7SQu Wenruo int index; 3900c146afadSYan Zheng 3901c146afadSYan Zheng while (1) { 390265d33fd7SQu Wenruo index = srcu_read_lock(&fs_info->subvol_srcu); 3903c146afadSYan Zheng ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 3904c146afadSYan Zheng (void **)gang, root_objectid, 3905c146afadSYan Zheng ARRAY_SIZE(gang)); 390665d33fd7SQu Wenruo if (!ret) { 390765d33fd7SQu Wenruo srcu_read_unlock(&fs_info->subvol_srcu, index); 3908c146afadSYan Zheng break; 390965d33fd7SQu Wenruo } 39105d4f98a2SYan Zheng root_objectid = gang[ret - 1]->root_key.objectid + 1; 391166b4ffd1SJosef Bacik 391265d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 391365d33fd7SQu Wenruo /* Avoid to grab roots in dead_roots */ 391465d33fd7SQu Wenruo if (btrfs_root_refs(&gang[i]->root_item) == 0) { 391565d33fd7SQu Wenruo gang[i] = NULL; 391665d33fd7SQu Wenruo continue; 391765d33fd7SQu Wenruo } 391865d33fd7SQu Wenruo /* grab all the search result for later use */ 391965d33fd7SQu Wenruo gang[i] = btrfs_grab_fs_root(gang[i]); 392065d33fd7SQu Wenruo } 392165d33fd7SQu Wenruo srcu_read_unlock(&fs_info->subvol_srcu, index); 392265d33fd7SQu Wenruo 392365d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 392465d33fd7SQu Wenruo if (!gang[i]) 392565d33fd7SQu Wenruo continue; 3926c146afadSYan Zheng root_objectid = gang[i]->root_key.objectid; 392766b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(gang[i]); 392866b4ffd1SJosef Bacik if (err) 392965d33fd7SQu Wenruo break; 393065d33fd7SQu Wenruo btrfs_put_fs_root(gang[i]); 3931c146afadSYan Zheng } 3932c146afadSYan Zheng root_objectid++; 3933c146afadSYan Zheng } 393465d33fd7SQu Wenruo 393565d33fd7SQu Wenruo /* release the uncleaned roots due to error */ 393665d33fd7SQu Wenruo for (; i < ret; i++) { 393765d33fd7SQu Wenruo if (gang[i]) 393865d33fd7SQu Wenruo btrfs_put_fs_root(gang[i]); 393965d33fd7SQu Wenruo } 394065d33fd7SQu Wenruo return err; 3941c146afadSYan Zheng } 3942c146afadSYan Zheng 39436bccf3abSJeff Mahoney int btrfs_commit_super(struct btrfs_fs_info *fs_info) 3944c146afadSYan Zheng { 39456bccf3abSJeff Mahoney struct btrfs_root *root = fs_info->tree_root; 3946c146afadSYan Zheng struct btrfs_trans_handle *trans; 3947c146afadSYan Zheng 39480b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 39492ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 39500b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 39510b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 3952c71bf099SYan, Zheng 3953c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 39540b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 39550b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 3956c71bf099SYan, Zheng 39577a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 39583612b495STsutomu Itoh if (IS_ERR(trans)) 39593612b495STsutomu Itoh return PTR_ERR(trans); 39603a45bb20SJeff Mahoney return btrfs_commit_transaction(trans); 3961c146afadSYan Zheng } 3962c146afadSYan Zheng 3963b105e927SDavid Sterba void __cold close_ctree(struct btrfs_fs_info *fs_info) 3964eb60ceacSChris Mason { 3965c146afadSYan Zheng int ret; 3966e089f05cSChris Mason 3967afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); 3968d6fd0ae2SOmar Sandoval /* 3969d6fd0ae2SOmar Sandoval * We don't want the cleaner to start new transactions, add more delayed 3970d6fd0ae2SOmar Sandoval * iputs, etc. while we're closing. We can't use kthread_stop() yet 3971d6fd0ae2SOmar Sandoval * because that frees the task_struct, and the transaction kthread might 3972d6fd0ae2SOmar Sandoval * still try to wake up the cleaner. 3973d6fd0ae2SOmar Sandoval */ 3974d6fd0ae2SOmar Sandoval kthread_park(fs_info->cleaner_kthread); 3975a2135011SChris Mason 39767343dd61SJustin Maggard /* wait for the qgroup rescan worker to stop */ 3977d06f23d6SJeff Mahoney btrfs_qgroup_wait_for_completion(fs_info, false); 39787343dd61SJustin Maggard 3979803b2f54SStefan Behrens /* wait for the uuid_scan task to finish */ 3980803b2f54SStefan Behrens down(&fs_info->uuid_tree_rescan_sem); 3981803b2f54SStefan Behrens /* avoid complains from lockdep et al., set sem back to initial state */ 3982803b2f54SStefan Behrens up(&fs_info->uuid_tree_rescan_sem); 3983803b2f54SStefan Behrens 3984837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 3985aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 3986837d5b6eSIlya Dryomov 39878dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 39888dabb742SStefan Behrens 3989aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 39904cb5300bSChris Mason 39914cb5300bSChris Mason /* wait for any defraggers to finish */ 39924cb5300bSChris Mason wait_event(fs_info->transaction_wait, 39934cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 39944cb5300bSChris Mason 39954cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 399626176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 39974cb5300bSChris Mason 399821c7e756SMiao Xie cancel_work_sync(&fs_info->async_reclaim_work); 399921c7e756SMiao Xie 4000bc98a42cSDavid Howells if (!sb_rdonly(fs_info->sb)) { 4001e44163e1SJeff Mahoney /* 4002d6fd0ae2SOmar Sandoval * The cleaner kthread is stopped, so do one final pass over 4003d6fd0ae2SOmar Sandoval * unused block groups. 4004e44163e1SJeff Mahoney */ 40050b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 4006e44163e1SJeff Mahoney 40076bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 4008d397712bSChris Mason if (ret) 400904892340SEric Sandeen btrfs_err(fs_info, "commit super ret %d", ret); 4010c146afadSYan Zheng } 4011ed2ff2cbSChris Mason 4012af722733SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) || 4013af722733SLiu Bo test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) 40142ff7e61eSJeff Mahoney btrfs_error_commit_super(fs_info); 4015acce952bSliubo 4016e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 4017e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 40188929ecfaSYan, Zheng 4019e187831eSJosef Bacik ASSERT(list_empty(&fs_info->delayed_iputs)); 4020afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); 4021f25784b3SYan Zheng 402204892340SEric Sandeen btrfs_free_qgroup_config(fs_info); 4023fe816d0fSNikolay Borisov ASSERT(list_empty(&fs_info->delalloc_roots)); 4024bcef60f2SArne Jansen 4025963d678bSMiao Xie if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 402604892340SEric Sandeen btrfs_info(fs_info, "at unmount delalloc count %lld", 4027963d678bSMiao Xie percpu_counter_sum(&fs_info->delalloc_bytes)); 4028b0c68f8bSChris Mason } 402931153d81SYan Zheng 40304297ff84SJosef Bacik if (percpu_counter_sum(&fs_info->dio_bytes)) 40314297ff84SJosef Bacik btrfs_info(fs_info, "at unmount dio bytes count %lld", 40324297ff84SJosef Bacik percpu_counter_sum(&fs_info->dio_bytes)); 40334297ff84SJosef Bacik 40346618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 4035b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 40365ac1d209SJeff Mahoney 4037faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 4038d10c5f31SChris Mason 40391a4319ccSLiu Bo btrfs_put_block_group_cache(fs_info); 40401a4319ccSLiu Bo 4041de348ee0SWang Shilong /* 4042de348ee0SWang Shilong * we must make sure there is not any read request to 4043de348ee0SWang Shilong * submit after we stopping all workers. 4044de348ee0SWang Shilong */ 4045de348ee0SWang Shilong invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 404696192499SJosef Bacik btrfs_stop_all_workers(fs_info); 404796192499SJosef Bacik 40485cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 40495cdd7db6SFilipe Manana 4050afcdd129SJosef Bacik clear_bit(BTRFS_FS_OPEN, &fs_info->flags); 40514273eaffSAnand Jain free_root_pointers(fs_info, true); 40529ad6b7bcSChris Mason 405313e6c37bSJosef Bacik iput(fs_info->btree_inode); 4054d6bfde87SChris Mason 405521adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 40560b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) 40572ff7e61eSJeff Mahoney btrfsic_unmount(fs_info->fs_devices); 405821adbd5cSStefan Behrens #endif 405921adbd5cSStefan Behrens 40600b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 406168c94e55SNikolay Borisov btrfs_close_devices(fs_info->fs_devices); 4062b248a415SChris Mason 4063e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 4064963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 40654297ff84SJosef Bacik percpu_counter_destroy(&fs_info->dio_bytes); 40667f8d236aSDavid Sterba percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 406776dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 40680b86a832SChris Mason 4069bfcea1c6SJohannes Thumshirn btrfs_free_csum_hash(fs_info); 407053b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 4071fd708b81SJosef Bacik btrfs_free_ref_cache(fs_info); 4072eb60ceacSChris Mason } 4073eb60ceacSChris Mason 4074b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 4075b9fab919SChris Mason int atomic) 4076ccd467d6SChris Mason { 40771259ab75SChris Mason int ret; 4078727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 40791259ab75SChris Mason 40800b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 40811259ab75SChris Mason if (!ret) 40821259ab75SChris Mason return ret; 40831259ab75SChris Mason 40841259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 4085b9fab919SChris Mason parent_transid, atomic); 4086b9fab919SChris Mason if (ret == -EAGAIN) 4087b9fab919SChris Mason return ret; 40881259ab75SChris Mason return !ret; 40895f39d397SChris Mason } 40906702ed49SChris Mason 40915f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 40925f39d397SChris Mason { 40930b246afaSJeff Mahoney struct btrfs_fs_info *fs_info; 409406ea65a3SJosef Bacik struct btrfs_root *root; 40955f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 4096b9473439SChris Mason int was_dirty; 4097b4ce94deSChris Mason 409806ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 409906ea65a3SJosef Bacik /* 410006ea65a3SJosef Bacik * This is a fast path so only do this check if we have sanity tests 410152042d8eSAndrea Gelmini * enabled. Normal people shouldn't be using unmapped buffers as dirty 410206ea65a3SJosef Bacik * outside of the sanity tests. 410306ea65a3SJosef Bacik */ 4104b0132a3bSNikolay Borisov if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) 410506ea65a3SJosef Bacik return; 410606ea65a3SJosef Bacik #endif 410706ea65a3SJosef Bacik root = BTRFS_I(buf->pages[0]->mapping->host)->root; 41080b246afaSJeff Mahoney fs_info = root->fs_info; 4109b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 41100b246afaSJeff Mahoney if (transid != fs_info->generation) 41115d163e0eSJeff Mahoney WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", 41120b246afaSJeff Mahoney buf->start, transid, fs_info->generation); 41130b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 4114e2d84521SMiao Xie if (!was_dirty) 4115104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 4116e2d84521SMiao Xie buf->len, 41170b246afaSJeff Mahoney fs_info->dirty_metadata_batch); 41181f21ef0aSFilipe Manana #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 411969fc6cbbSQu Wenruo /* 412069fc6cbbSQu Wenruo * Since btrfs_mark_buffer_dirty() can be called with item pointer set 412169fc6cbbSQu Wenruo * but item data not updated. 412269fc6cbbSQu Wenruo * So here we should only check item pointers, not item data. 412369fc6cbbSQu Wenruo */ 412469fc6cbbSQu Wenruo if (btrfs_header_level(buf) == 0 && 4125cfdaad5eSDavid Sterba btrfs_check_leaf_relaxed(buf)) { 4126a4f78750SDavid Sterba btrfs_print_leaf(buf); 41271f21ef0aSFilipe Manana ASSERT(0); 41281f21ef0aSFilipe Manana } 41291f21ef0aSFilipe Manana #endif 4130eb60ceacSChris Mason } 4131eb60ceacSChris Mason 41322ff7e61eSJeff Mahoney static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, 4133b53d3f5dSLiu Bo int flush_delayed) 413435b7e476SChris Mason { 4135188de649SChris Mason /* 4136188de649SChris Mason * looks as though older kernels can get into trouble with 4137188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 4138188de649SChris Mason */ 4139e2d84521SMiao Xie int ret; 4140d6bfde87SChris Mason 41416933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 4142d6bfde87SChris Mason return; 4143d6bfde87SChris Mason 4144b53d3f5dSLiu Bo if (flush_delayed) 41452ff7e61eSJeff Mahoney btrfs_balance_delayed_items(fs_info); 414616cdcec7SMiao Xie 4147d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 4148d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 4149d814a491SEthan Lien fs_info->dirty_metadata_batch); 4150e2d84521SMiao Xie if (ret > 0) { 41510b246afaSJeff Mahoney balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); 415216cdcec7SMiao Xie } 415316cdcec7SMiao Xie } 415416cdcec7SMiao Xie 41552ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) 415616cdcec7SMiao Xie { 41572ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 1); 415835b7e476SChris Mason } 4159b53d3f5dSLiu Bo 41602ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) 4161b53d3f5dSLiu Bo { 41622ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 0); 4163d6bfde87SChris Mason } 41646b80053dSChris Mason 4165581c1760SQu Wenruo int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level, 4166581c1760SQu Wenruo struct btrfs_key *first_key) 41676b80053dSChris Mason { 41685ab12d1fSDavid Sterba return btree_read_extent_buffer_pages(buf, parent_transid, 4169581c1760SQu Wenruo level, first_key); 41706b80053dSChris Mason } 41710da5468fSChris Mason 41722ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) 4173acce952bSliubo { 4174fe816d0fSNikolay Borisov /* cleanup FS via transaction */ 4175fe816d0fSNikolay Borisov btrfs_cleanup_transaction(fs_info); 4176fe816d0fSNikolay Borisov 41770b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 41782ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 41790b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 4180acce952bSliubo 41810b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 41820b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4183acce952bSliubo } 4184acce952bSliubo 4185143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4186acce952bSliubo { 4187acce952bSliubo struct btrfs_ordered_extent *ordered; 4188acce952bSliubo 4189199c2a9cSMiao Xie spin_lock(&root->ordered_extent_lock); 4190779880efSJosef Bacik /* 4191779880efSJosef Bacik * This will just short circuit the ordered completion stuff which will 4192779880efSJosef Bacik * make sure the ordered extent gets properly cleaned up. 4193779880efSJosef Bacik */ 4194199c2a9cSMiao Xie list_for_each_entry(ordered, &root->ordered_extents, 4195779880efSJosef Bacik root_extent_list) 4196779880efSJosef Bacik set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4197199c2a9cSMiao Xie spin_unlock(&root->ordered_extent_lock); 4198199c2a9cSMiao Xie } 4199199c2a9cSMiao Xie 4200199c2a9cSMiao Xie static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4201199c2a9cSMiao Xie { 4202199c2a9cSMiao Xie struct btrfs_root *root; 4203199c2a9cSMiao Xie struct list_head splice; 4204199c2a9cSMiao Xie 4205199c2a9cSMiao Xie INIT_LIST_HEAD(&splice); 4206199c2a9cSMiao Xie 4207199c2a9cSMiao Xie spin_lock(&fs_info->ordered_root_lock); 4208199c2a9cSMiao Xie list_splice_init(&fs_info->ordered_roots, &splice); 4209199c2a9cSMiao Xie while (!list_empty(&splice)) { 4210199c2a9cSMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4211199c2a9cSMiao Xie ordered_root); 42121de2cfdeSJosef Bacik list_move_tail(&root->ordered_root, 42131de2cfdeSJosef Bacik &fs_info->ordered_roots); 4214199c2a9cSMiao Xie 42152a85d9caSLiu Bo spin_unlock(&fs_info->ordered_root_lock); 4216199c2a9cSMiao Xie btrfs_destroy_ordered_extents(root); 4217199c2a9cSMiao Xie 42182a85d9caSLiu Bo cond_resched(); 42192a85d9caSLiu Bo spin_lock(&fs_info->ordered_root_lock); 4220199c2a9cSMiao Xie } 4221199c2a9cSMiao Xie spin_unlock(&fs_info->ordered_root_lock); 422274d5d229SJosef Bacik 422374d5d229SJosef Bacik /* 422474d5d229SJosef Bacik * We need this here because if we've been flipped read-only we won't 422574d5d229SJosef Bacik * get sync() from the umount, so we need to make sure any ordered 422674d5d229SJosef Bacik * extents that haven't had their dirty pages IO start writeout yet 422774d5d229SJosef Bacik * actually get run and error out properly. 422874d5d229SJosef Bacik */ 422974d5d229SJosef Bacik btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 4230acce952bSliubo } 4231acce952bSliubo 423235a3621bSStefan Behrens static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 42332ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4234acce952bSliubo { 4235acce952bSliubo struct rb_node *node; 4236acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 4237acce952bSliubo struct btrfs_delayed_ref_node *ref; 4238acce952bSliubo int ret = 0; 4239acce952bSliubo 4240acce952bSliubo delayed_refs = &trans->delayed_refs; 4241acce952bSliubo 4242acce952bSliubo spin_lock(&delayed_refs->lock); 4243d7df2c79SJosef Bacik if (atomic_read(&delayed_refs->num_entries) == 0) { 4244cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 42450b246afaSJeff Mahoney btrfs_info(fs_info, "delayed_refs has NO entry"); 4246acce952bSliubo return ret; 4247acce952bSliubo } 4248acce952bSliubo 42495c9d028bSLiu Bo while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { 4250d7df2c79SJosef Bacik struct btrfs_delayed_ref_head *head; 42510e0adbcfSJosef Bacik struct rb_node *n; 4252e78417d1SJosef Bacik bool pin_bytes = false; 4253acce952bSliubo 4254d7df2c79SJosef Bacik head = rb_entry(node, struct btrfs_delayed_ref_head, 4255d7df2c79SJosef Bacik href_node); 42563069bd26SJosef Bacik if (btrfs_delayed_ref_lock(delayed_refs, head)) 4257b939d1abSJosef Bacik continue; 42583069bd26SJosef Bacik 4259d7df2c79SJosef Bacik spin_lock(&head->lock); 4260e3d03965SLiu Bo while ((n = rb_first_cached(&head->ref_tree)) != NULL) { 42610e0adbcfSJosef Bacik ref = rb_entry(n, struct btrfs_delayed_ref_node, 42620e0adbcfSJosef Bacik ref_node); 4263d7df2c79SJosef Bacik ref->in_tree = 0; 4264e3d03965SLiu Bo rb_erase_cached(&ref->ref_node, &head->ref_tree); 42650e0adbcfSJosef Bacik RB_CLEAR_NODE(&ref->ref_node); 42661d57ee94SWang Xiaoguang if (!list_empty(&ref->add_list)) 42671d57ee94SWang Xiaoguang list_del(&ref->add_list); 4268d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4269d7df2c79SJosef Bacik btrfs_put_delayed_ref(ref); 4270d7df2c79SJosef Bacik } 427154067ae9SJosef Bacik if (head->must_insert_reserved) 4272e78417d1SJosef Bacik pin_bytes = true; 427378a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 4274fa781ceaSJosef Bacik btrfs_delete_ref_head(delayed_refs, head); 4275d7df2c79SJosef Bacik spin_unlock(&head->lock); 4276acce952bSliubo spin_unlock(&delayed_refs->lock); 4277e78417d1SJosef Bacik mutex_unlock(&head->mutex); 4278acce952bSliubo 4279d7df2c79SJosef Bacik if (pin_bytes) 4280d278850eSJosef Bacik btrfs_pin_extent(fs_info, head->bytenr, 4281d278850eSJosef Bacik head->num_bytes, 1); 428231890da0SJosef Bacik btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); 4283d278850eSJosef Bacik btrfs_put_delayed_ref_head(head); 4284acce952bSliubo cond_resched(); 4285acce952bSliubo spin_lock(&delayed_refs->lock); 4286acce952bSliubo } 4287acce952bSliubo 4288acce952bSliubo spin_unlock(&delayed_refs->lock); 4289acce952bSliubo 4290acce952bSliubo return ret; 4291acce952bSliubo } 4292acce952bSliubo 4293143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 4294acce952bSliubo { 4295acce952bSliubo struct btrfs_inode *btrfs_inode; 4296acce952bSliubo struct list_head splice; 4297acce952bSliubo 4298acce952bSliubo INIT_LIST_HEAD(&splice); 4299acce952bSliubo 4300eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4301eb73c1b7SMiao Xie list_splice_init(&root->delalloc_inodes, &splice); 4302acce952bSliubo 4303acce952bSliubo while (!list_empty(&splice)) { 4304fe816d0fSNikolay Borisov struct inode *inode = NULL; 4305eb73c1b7SMiao Xie btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 4306acce952bSliubo delalloc_inodes); 4307fe816d0fSNikolay Borisov __btrfs_del_delalloc_inode(root, btrfs_inode); 4308eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4309acce952bSliubo 4310fe816d0fSNikolay Borisov /* 4311fe816d0fSNikolay Borisov * Make sure we get a live inode and that it'll not disappear 4312fe816d0fSNikolay Borisov * meanwhile. 4313fe816d0fSNikolay Borisov */ 4314fe816d0fSNikolay Borisov inode = igrab(&btrfs_inode->vfs_inode); 4315fe816d0fSNikolay Borisov if (inode) { 4316fe816d0fSNikolay Borisov invalidate_inode_pages2(inode->i_mapping); 4317fe816d0fSNikolay Borisov iput(inode); 4318fe816d0fSNikolay Borisov } 4319eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4320acce952bSliubo } 4321eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4322eb73c1b7SMiao Xie } 4323eb73c1b7SMiao Xie 4324eb73c1b7SMiao Xie static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 4325eb73c1b7SMiao Xie { 4326eb73c1b7SMiao Xie struct btrfs_root *root; 4327eb73c1b7SMiao Xie struct list_head splice; 4328eb73c1b7SMiao Xie 4329eb73c1b7SMiao Xie INIT_LIST_HEAD(&splice); 4330eb73c1b7SMiao Xie 4331eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4332eb73c1b7SMiao Xie list_splice_init(&fs_info->delalloc_roots, &splice); 4333eb73c1b7SMiao Xie while (!list_empty(&splice)) { 4334eb73c1b7SMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4335eb73c1b7SMiao Xie delalloc_root); 4336eb73c1b7SMiao Xie root = btrfs_grab_fs_root(root); 4337eb73c1b7SMiao Xie BUG_ON(!root); 4338eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4339eb73c1b7SMiao Xie 4340eb73c1b7SMiao Xie btrfs_destroy_delalloc_inodes(root); 4341eb73c1b7SMiao Xie btrfs_put_fs_root(root); 4342eb73c1b7SMiao Xie 4343eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4344eb73c1b7SMiao Xie } 4345eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4346acce952bSliubo } 4347acce952bSliubo 43482ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 4349acce952bSliubo struct extent_io_tree *dirty_pages, 4350acce952bSliubo int mark) 4351acce952bSliubo { 4352acce952bSliubo int ret; 4353acce952bSliubo struct extent_buffer *eb; 4354acce952bSliubo u64 start = 0; 4355acce952bSliubo u64 end; 4356acce952bSliubo 4357acce952bSliubo while (1) { 4358acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 4359e6138876SJosef Bacik mark, NULL); 4360acce952bSliubo if (ret) 4361acce952bSliubo break; 4362acce952bSliubo 436391166212SDavid Sterba clear_extent_bits(dirty_pages, start, end, mark); 4364acce952bSliubo while (start <= end) { 43650b246afaSJeff Mahoney eb = find_extent_buffer(fs_info, start); 43660b246afaSJeff Mahoney start += fs_info->nodesize; 4367fd8b2b61SJosef Bacik if (!eb) 4368acce952bSliubo continue; 4369fd8b2b61SJosef Bacik wait_on_extent_buffer_writeback(eb); 4370acce952bSliubo 4371fd8b2b61SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 4372fd8b2b61SJosef Bacik &eb->bflags)) 4373fd8b2b61SJosef Bacik clear_extent_buffer_dirty(eb); 4374fd8b2b61SJosef Bacik free_extent_buffer_stale(eb); 4375acce952bSliubo } 4376acce952bSliubo } 4377acce952bSliubo 4378acce952bSliubo return ret; 4379acce952bSliubo } 4380acce952bSliubo 43812ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 4382acce952bSliubo struct extent_io_tree *pinned_extents) 4383acce952bSliubo { 4384acce952bSliubo struct extent_io_tree *unpin; 4385acce952bSliubo u64 start; 4386acce952bSliubo u64 end; 4387acce952bSliubo int ret; 4388ed0eaa14SLiu Bo bool loop = true; 4389acce952bSliubo 4390acce952bSliubo unpin = pinned_extents; 4391ed0eaa14SLiu Bo again: 4392acce952bSliubo while (1) { 43930e6ec385SFilipe Manana struct extent_state *cached_state = NULL; 43940e6ec385SFilipe Manana 4395fcd5e742SLu Fengqi /* 4396fcd5e742SLu Fengqi * The btrfs_finish_extent_commit() may get the same range as 4397fcd5e742SLu Fengqi * ours between find_first_extent_bit and clear_extent_dirty. 4398fcd5e742SLu Fengqi * Hence, hold the unused_bg_unpin_mutex to avoid double unpin 4399fcd5e742SLu Fengqi * the same extent range. 4400fcd5e742SLu Fengqi */ 4401fcd5e742SLu Fengqi mutex_lock(&fs_info->unused_bg_unpin_mutex); 4402acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 44030e6ec385SFilipe Manana EXTENT_DIRTY, &cached_state); 4404fcd5e742SLu Fengqi if (ret) { 4405fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4406acce952bSliubo break; 4407fcd5e742SLu Fengqi } 4408acce952bSliubo 44090e6ec385SFilipe Manana clear_extent_dirty(unpin, start, end, &cached_state); 44100e6ec385SFilipe Manana free_extent_state(cached_state); 44112ff7e61eSJeff Mahoney btrfs_error_unpin_extent_range(fs_info, start, end); 4412fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4413acce952bSliubo cond_resched(); 4414acce952bSliubo } 4415acce952bSliubo 4416ed0eaa14SLiu Bo if (loop) { 44170b246afaSJeff Mahoney if (unpin == &fs_info->freed_extents[0]) 44180b246afaSJeff Mahoney unpin = &fs_info->freed_extents[1]; 4419ed0eaa14SLiu Bo else 44200b246afaSJeff Mahoney unpin = &fs_info->freed_extents[0]; 4421ed0eaa14SLiu Bo loop = false; 4422ed0eaa14SLiu Bo goto again; 4423ed0eaa14SLiu Bo } 4424ed0eaa14SLiu Bo 4425acce952bSliubo return 0; 4426acce952bSliubo } 4427acce952bSliubo 4428c79a1751SLiu Bo static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache) 4429c79a1751SLiu Bo { 4430c79a1751SLiu Bo struct inode *inode; 4431c79a1751SLiu Bo 4432c79a1751SLiu Bo inode = cache->io_ctl.inode; 4433c79a1751SLiu Bo if (inode) { 4434c79a1751SLiu Bo invalidate_inode_pages2(inode->i_mapping); 4435c79a1751SLiu Bo BTRFS_I(inode)->generation = 0; 4436c79a1751SLiu Bo cache->io_ctl.inode = NULL; 4437c79a1751SLiu Bo iput(inode); 4438c79a1751SLiu Bo } 4439c79a1751SLiu Bo btrfs_put_block_group(cache); 4440c79a1751SLiu Bo } 4441c79a1751SLiu Bo 4442c79a1751SLiu Bo void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, 44432ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4444c79a1751SLiu Bo { 4445c79a1751SLiu Bo struct btrfs_block_group_cache *cache; 4446c79a1751SLiu Bo 4447c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4448c79a1751SLiu Bo while (!list_empty(&cur_trans->dirty_bgs)) { 4449c79a1751SLiu Bo cache = list_first_entry(&cur_trans->dirty_bgs, 4450c79a1751SLiu Bo struct btrfs_block_group_cache, 4451c79a1751SLiu Bo dirty_list); 4452c79a1751SLiu Bo 4453c79a1751SLiu Bo if (!list_empty(&cache->io_list)) { 4454c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4455c79a1751SLiu Bo list_del_init(&cache->io_list); 4456c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 4457c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4458c79a1751SLiu Bo } 4459c79a1751SLiu Bo 4460c79a1751SLiu Bo list_del_init(&cache->dirty_list); 4461c79a1751SLiu Bo spin_lock(&cache->lock); 4462c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 4463c79a1751SLiu Bo spin_unlock(&cache->lock); 4464c79a1751SLiu Bo 4465c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4466c79a1751SLiu Bo btrfs_put_block_group(cache); 4467ba2c4d4eSJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 4468c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4469c79a1751SLiu Bo } 4470c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4471c79a1751SLiu Bo 447245ae2c18SNikolay Borisov /* 447345ae2c18SNikolay Borisov * Refer to the definition of io_bgs member for details why it's safe 447445ae2c18SNikolay Borisov * to use it without any locking 447545ae2c18SNikolay Borisov */ 4476c79a1751SLiu Bo while (!list_empty(&cur_trans->io_bgs)) { 4477c79a1751SLiu Bo cache = list_first_entry(&cur_trans->io_bgs, 4478c79a1751SLiu Bo struct btrfs_block_group_cache, 4479c79a1751SLiu Bo io_list); 4480c79a1751SLiu Bo 4481c79a1751SLiu Bo list_del_init(&cache->io_list); 4482c79a1751SLiu Bo spin_lock(&cache->lock); 4483c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 4484c79a1751SLiu Bo spin_unlock(&cache->lock); 4485c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 4486c79a1751SLiu Bo } 4487c79a1751SLiu Bo } 4488c79a1751SLiu Bo 448949b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 44902ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 449149b25e05SJeff Mahoney { 4492bbbf7243SNikolay Borisov struct btrfs_device *dev, *tmp; 4493bbbf7243SNikolay Borisov 44942ff7e61eSJeff Mahoney btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 4495c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->dirty_bgs)); 4496c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->io_bgs)); 4497c79a1751SLiu Bo 4498bbbf7243SNikolay Borisov list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list, 4499bbbf7243SNikolay Borisov post_commit_list) { 4500bbbf7243SNikolay Borisov list_del_init(&dev->post_commit_list); 4501bbbf7243SNikolay Borisov } 4502bbbf7243SNikolay Borisov 45032ff7e61eSJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, fs_info); 450449b25e05SJeff Mahoney 45054a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 45060b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 450749b25e05SJeff Mahoney 45084a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 45090b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 451049b25e05SJeff Mahoney 4511ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 4512ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 451367cde344SMiao Xie 45142ff7e61eSJeff Mahoney btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, 451549b25e05SJeff Mahoney EXTENT_DIRTY); 45162ff7e61eSJeff Mahoney btrfs_destroy_pinned_extent(fs_info, 45170b246afaSJeff Mahoney fs_info->pinned_extents); 451849b25e05SJeff Mahoney 45194a9d8bdeSMiao Xie cur_trans->state =TRANS_STATE_COMPLETED; 45204a9d8bdeSMiao Xie wake_up(&cur_trans->commit_wait); 452149b25e05SJeff Mahoney } 452249b25e05SJeff Mahoney 45232ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) 4524acce952bSliubo { 4525acce952bSliubo struct btrfs_transaction *t; 4526acce952bSliubo 45270b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 4528acce952bSliubo 45290b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 45300b246afaSJeff Mahoney while (!list_empty(&fs_info->trans_list)) { 45310b246afaSJeff Mahoney t = list_first_entry(&fs_info->trans_list, 4532724e2315SJosef Bacik struct btrfs_transaction, list); 4533724e2315SJosef Bacik if (t->state >= TRANS_STATE_COMMIT_START) { 45349b64f57dSElena Reshetova refcount_inc(&t->use_count); 45350b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 45362ff7e61eSJeff Mahoney btrfs_wait_for_commit(fs_info, t->transid); 4537724e2315SJosef Bacik btrfs_put_transaction(t); 45380b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4539724e2315SJosef Bacik continue; 4540724e2315SJosef Bacik } 45410b246afaSJeff Mahoney if (t == fs_info->running_transaction) { 4542724e2315SJosef Bacik t->state = TRANS_STATE_COMMIT_DOING; 45430b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4544724e2315SJosef Bacik /* 4545724e2315SJosef Bacik * We wait for 0 num_writers since we don't hold a trans 4546724e2315SJosef Bacik * handle open currently for this transaction. 4547724e2315SJosef Bacik */ 4548724e2315SJosef Bacik wait_event(t->writer_wait, 4549724e2315SJosef Bacik atomic_read(&t->num_writers) == 0); 4550724e2315SJosef Bacik } else { 45510b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4552724e2315SJosef Bacik } 45532ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(t, fs_info); 4554724e2315SJosef Bacik 45550b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 45560b246afaSJeff Mahoney if (t == fs_info->running_transaction) 45570b246afaSJeff Mahoney fs_info->running_transaction = NULL; 4558724e2315SJosef Bacik list_del_init(&t->list); 45590b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4560a4abeea4SJosef Bacik 4561724e2315SJosef Bacik btrfs_put_transaction(t); 45622ff7e61eSJeff Mahoney trace_btrfs_transaction_commit(fs_info->tree_root); 45630b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4564724e2315SJosef Bacik } 45650b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 45660b246afaSJeff Mahoney btrfs_destroy_all_ordered_extents(fs_info); 4567ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 4568ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 45692ff7e61eSJeff Mahoney btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents); 45700b246afaSJeff Mahoney btrfs_destroy_all_delalloc_inodes(fs_info); 45710b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 4572acce952bSliubo 4573acce952bSliubo return 0; 4574acce952bSliubo } 4575acce952bSliubo 4576e8c9f186SDavid Sterba static const struct extent_io_ops btree_extent_io_ops = { 45774d53dddbSDavid Sterba /* mandatory callbacks */ 45780b86a832SChris Mason .submit_bio_hook = btree_submit_bio_hook, 45794d53dddbSDavid Sterba .readpage_end_io_hook = btree_readpage_end_io_hook, 45800da5468fSChris Mason }; 4581