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: 357352ae07bSDavid Sterba case BTRFS_CSUM_TYPE_BLAKE2: 358e7e16f48SJohannes Thumshirn return true; 359e7e16f48SJohannes Thumshirn default: 360e7e16f48SJohannes Thumshirn return false; 361e7e16f48SJohannes Thumshirn } 362e7e16f48SJohannes Thumshirn } 363e7e16f48SJohannes Thumshirn 364d352ac68SChris Mason /* 3651104a885SDavid Sterba * Return 0 if the superblock checksum type matches the checksum value of that 3661104a885SDavid Sterba * algorithm. Pass the raw disk superblock data. 3671104a885SDavid Sterba */ 368ab8d0fc4SJeff Mahoney static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info, 369ab8d0fc4SJeff Mahoney char *raw_disk_sb) 3701104a885SDavid Sterba { 3711104a885SDavid Sterba struct btrfs_super_block *disk_sb = 3721104a885SDavid Sterba (struct btrfs_super_block *)raw_disk_sb; 37351bce6c9SJohannes Thumshirn char result[BTRFS_CSUM_SIZE]; 374d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 375d5178578SJohannes Thumshirn 376d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 377d5178578SJohannes Thumshirn crypto_shash_init(shash); 3781104a885SDavid Sterba 3791104a885SDavid Sterba /* 3801104a885SDavid Sterba * The super_block structure does not span the whole 38151bce6c9SJohannes Thumshirn * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is 38251bce6c9SJohannes Thumshirn * filled with zeros and is included in the checksum. 3831104a885SDavid Sterba */ 384d5178578SJohannes Thumshirn crypto_shash_update(shash, raw_disk_sb + BTRFS_CSUM_SIZE, 385d5178578SJohannes Thumshirn BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE); 386d5178578SJohannes Thumshirn crypto_shash_final(shash, result); 3871104a885SDavid Sterba 38851bce6c9SJohannes Thumshirn if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb))) 389e7e16f48SJohannes Thumshirn return 1; 3901104a885SDavid Sterba 391e7e16f48SJohannes Thumshirn return 0; 3921104a885SDavid Sterba } 3931104a885SDavid Sterba 394e064d5e9SDavid Sterba int btrfs_verify_level_key(struct extent_buffer *eb, int level, 395ff76a864SLiu Bo struct btrfs_key *first_key, u64 parent_transid) 396581c1760SQu Wenruo { 397e064d5e9SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 398581c1760SQu Wenruo int found_level; 399581c1760SQu Wenruo struct btrfs_key found_key; 400581c1760SQu Wenruo int ret; 401581c1760SQu Wenruo 402581c1760SQu Wenruo found_level = btrfs_header_level(eb); 403581c1760SQu Wenruo if (found_level != level) { 40463489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 40563489055SQu Wenruo KERN_ERR "BTRFS: tree level check failed\n"); 406581c1760SQu Wenruo btrfs_err(fs_info, 407581c1760SQu Wenruo "tree level mismatch detected, bytenr=%llu level expected=%u has=%u", 408581c1760SQu Wenruo eb->start, level, found_level); 409581c1760SQu Wenruo return -EIO; 410581c1760SQu Wenruo } 411581c1760SQu Wenruo 412581c1760SQu Wenruo if (!first_key) 413581c1760SQu Wenruo return 0; 414581c1760SQu Wenruo 4155d41be6fSQu Wenruo /* 4165d41be6fSQu Wenruo * For live tree block (new tree blocks in current transaction), 4175d41be6fSQu Wenruo * we need proper lock context to avoid race, which is impossible here. 4185d41be6fSQu Wenruo * So we only checks tree blocks which is read from disk, whose 4195d41be6fSQu Wenruo * generation <= fs_info->last_trans_committed. 4205d41be6fSQu Wenruo */ 4215d41be6fSQu Wenruo if (btrfs_header_generation(eb) > fs_info->last_trans_committed) 4225d41be6fSQu Wenruo return 0; 42362fdaa52SQu Wenruo 42462fdaa52SQu Wenruo /* We have @first_key, so this @eb must have at least one item */ 42562fdaa52SQu Wenruo if (btrfs_header_nritems(eb) == 0) { 42662fdaa52SQu Wenruo btrfs_err(fs_info, 42762fdaa52SQu Wenruo "invalid tree nritems, bytenr=%llu nritems=0 expect >0", 42862fdaa52SQu Wenruo eb->start); 42962fdaa52SQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 43062fdaa52SQu Wenruo return -EUCLEAN; 43162fdaa52SQu Wenruo } 43262fdaa52SQu Wenruo 433581c1760SQu Wenruo if (found_level) 434581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &found_key, 0); 435581c1760SQu Wenruo else 436581c1760SQu Wenruo btrfs_item_key_to_cpu(eb, &found_key, 0); 437581c1760SQu Wenruo ret = btrfs_comp_cpu_keys(first_key, &found_key); 438581c1760SQu Wenruo 439581c1760SQu Wenruo if (ret) { 44063489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 44163489055SQu Wenruo KERN_ERR "BTRFS: tree first key check failed\n"); 442581c1760SQu Wenruo btrfs_err(fs_info, 443ff76a864SLiu Bo "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", 444ff76a864SLiu Bo eb->start, parent_transid, first_key->objectid, 445ff76a864SLiu Bo first_key->type, first_key->offset, 446ff76a864SLiu Bo found_key.objectid, found_key.type, 447ff76a864SLiu Bo found_key.offset); 448581c1760SQu Wenruo } 449581c1760SQu Wenruo return ret; 450581c1760SQu Wenruo } 451581c1760SQu Wenruo 4521104a885SDavid Sterba /* 453d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 454d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 455581c1760SQu Wenruo * 456581c1760SQu Wenruo * @parent_transid: expected transid, skip check if 0 457581c1760SQu Wenruo * @level: expected level, mandatory check 458581c1760SQu Wenruo * @first_key: expected key of first slot, skip check if NULL 459d352ac68SChris Mason */ 4605ab12d1fSDavid Sterba static int btree_read_extent_buffer_pages(struct extent_buffer *eb, 461581c1760SQu Wenruo u64 parent_transid, int level, 462581c1760SQu Wenruo struct btrfs_key *first_key) 463f188591eSChris Mason { 4645ab12d1fSDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 465f188591eSChris Mason struct extent_io_tree *io_tree; 466ea466794SJosef Bacik int failed = 0; 467f188591eSChris Mason int ret; 468f188591eSChris Mason int num_copies = 0; 469f188591eSChris Mason int mirror_num = 0; 470ea466794SJosef Bacik int failed_mirror = 0; 471f188591eSChris Mason 4720b246afaSJeff Mahoney io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; 473f188591eSChris Mason while (1) { 474f8397d69SNikolay Borisov clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 475c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num); 476256dd1bbSStefan Behrens if (!ret) { 477581c1760SQu Wenruo if (verify_parent_transid(io_tree, eb, 478b9fab919SChris Mason parent_transid, 0)) 479256dd1bbSStefan Behrens ret = -EIO; 480e064d5e9SDavid Sterba else if (btrfs_verify_level_key(eb, level, 481ff76a864SLiu Bo first_key, parent_transid)) 482581c1760SQu Wenruo ret = -EUCLEAN; 483581c1760SQu Wenruo else 484581c1760SQu Wenruo break; 485256dd1bbSStefan Behrens } 486d397712bSChris Mason 4870b246afaSJeff Mahoney num_copies = btrfs_num_copies(fs_info, 488f188591eSChris Mason eb->start, eb->len); 4894235298eSChris Mason if (num_copies == 1) 490ea466794SJosef Bacik break; 4914235298eSChris Mason 4925cf1ab56SJosef Bacik if (!failed_mirror) { 4935cf1ab56SJosef Bacik failed = 1; 4945cf1ab56SJosef Bacik failed_mirror = eb->read_mirror; 4955cf1ab56SJosef Bacik } 4965cf1ab56SJosef Bacik 497f188591eSChris Mason mirror_num++; 498ea466794SJosef Bacik if (mirror_num == failed_mirror) 499ea466794SJosef Bacik mirror_num++; 500ea466794SJosef Bacik 5014235298eSChris Mason if (mirror_num > num_copies) 502ea466794SJosef Bacik break; 503f188591eSChris Mason } 504ea466794SJosef Bacik 505c0901581SStefan Behrens if (failed && !ret && failed_mirror) 50620a1fbf9SDavid Sterba btrfs_repair_eb_io_failure(eb, failed_mirror); 507ea466794SJosef Bacik 508ea466794SJosef Bacik return ret; 509f188591eSChris Mason } 51019c00ddcSChris Mason 511d352ac68SChris Mason /* 512d397712bSChris Mason * checksum a dirty tree block before IO. This has extra checks to make sure 513d397712bSChris Mason * we only fill in the checksum field in the first page of a multi-page block 514d352ac68SChris Mason */ 515d397712bSChris Mason 51601d58472SDaniel Dressler static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page) 51719c00ddcSChris Mason { 5184eee4fa4SMiao Xie u64 start = page_offset(page); 51919c00ddcSChris Mason u64 found_start; 5202996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 5212996e1f8SJohannes Thumshirn u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); 52219c00ddcSChris Mason struct extent_buffer *eb; 5238d47a0d8SQu Wenruo int ret; 524f188591eSChris Mason 5254f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 5264f2de97aSJosef Bacik if (page != eb->pages[0]) 5274f2de97aSJosef Bacik return 0; 5280f805531SAlex Lyakas 52919c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 5300f805531SAlex Lyakas /* 5310f805531SAlex Lyakas * Please do not consolidate these warnings into a single if. 5320f805531SAlex Lyakas * It is useful to know what went wrong. 5330f805531SAlex Lyakas */ 5340f805531SAlex Lyakas if (WARN_ON(found_start != start)) 5350f805531SAlex Lyakas return -EUCLEAN; 5360f805531SAlex Lyakas if (WARN_ON(!PageUptodate(page))) 5370f805531SAlex Lyakas return -EUCLEAN; 5380f805531SAlex Lyakas 539de37aa51SNikolay Borisov ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid, 5400f805531SAlex Lyakas btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0); 5410f805531SAlex Lyakas 5422996e1f8SJohannes Thumshirn if (csum_tree_block(eb, result)) 5432996e1f8SJohannes Thumshirn return -EINVAL; 5442996e1f8SJohannes Thumshirn 5458d47a0d8SQu Wenruo if (btrfs_header_level(eb)) 5468d47a0d8SQu Wenruo ret = btrfs_check_node(eb); 5478d47a0d8SQu Wenruo else 5488d47a0d8SQu Wenruo ret = btrfs_check_leaf_full(eb); 5498d47a0d8SQu Wenruo 5508d47a0d8SQu Wenruo if (ret < 0) { 551c06631b0SQu Wenruo btrfs_print_tree(eb, 0); 5528d47a0d8SQu Wenruo btrfs_err(fs_info, 5538d47a0d8SQu Wenruo "block=%llu write time tree block corruption detected", 5548d47a0d8SQu Wenruo eb->start); 555c06631b0SQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 5568d47a0d8SQu Wenruo return ret; 5578d47a0d8SQu Wenruo } 5582996e1f8SJohannes Thumshirn write_extent_buffer(eb, result, 0, csum_size); 5598d47a0d8SQu Wenruo 5602996e1f8SJohannes Thumshirn return 0; 56119c00ddcSChris Mason } 56219c00ddcSChris Mason 563b0c9b3b0SDavid Sterba static int check_tree_block_fsid(struct extent_buffer *eb) 5642b82032cSYan Zheng { 565b0c9b3b0SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 56601d58472SDaniel Dressler struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 56744880fdcSAnand Jain u8 fsid[BTRFS_FSID_SIZE]; 5682b82032cSYan Zheng int ret = 1; 5692b82032cSYan Zheng 5700a4e5586SRoss Kirk read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE); 5712b82032cSYan Zheng while (fs_devices) { 5727239ff4bSNikolay Borisov u8 *metadata_uuid; 5737239ff4bSNikolay Borisov 5747239ff4bSNikolay Borisov /* 5757239ff4bSNikolay Borisov * Checking the incompat flag is only valid for the current 5767239ff4bSNikolay Borisov * fs. For seed devices it's forbidden to have their uuid 5777239ff4bSNikolay Borisov * changed so reading ->fsid in this case is fine 5787239ff4bSNikolay Borisov */ 5797239ff4bSNikolay Borisov if (fs_devices == fs_info->fs_devices && 5807239ff4bSNikolay Borisov btrfs_fs_incompat(fs_info, METADATA_UUID)) 5817239ff4bSNikolay Borisov metadata_uuid = fs_devices->metadata_uuid; 5827239ff4bSNikolay Borisov else 5837239ff4bSNikolay Borisov metadata_uuid = fs_devices->fsid; 5847239ff4bSNikolay Borisov 5857239ff4bSNikolay Borisov if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) { 5862b82032cSYan Zheng ret = 0; 5872b82032cSYan Zheng break; 5882b82032cSYan Zheng } 5892b82032cSYan Zheng fs_devices = fs_devices->seed; 5902b82032cSYan Zheng } 5912b82032cSYan Zheng return ret; 5922b82032cSYan Zheng } 5932b82032cSYan Zheng 594facc8a22SMiao Xie static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio, 595facc8a22SMiao Xie u64 phy_offset, struct page *page, 596facc8a22SMiao Xie u64 start, u64 end, int mirror) 597ce9adaa5SChris Mason { 598ce9adaa5SChris Mason u64 found_start; 599ce9adaa5SChris Mason int found_level; 600ce9adaa5SChris Mason struct extent_buffer *eb; 601ce9adaa5SChris Mason struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 60202873e43SZhao Lei struct btrfs_fs_info *fs_info = root->fs_info; 6032996e1f8SJohannes Thumshirn u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); 604f188591eSChris Mason int ret = 0; 6052996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 606727011e0SChris Mason int reads_done; 607ce9adaa5SChris Mason 608ce9adaa5SChris Mason if (!page->private) 609ce9adaa5SChris Mason goto out; 610d397712bSChris Mason 6114f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 612d397712bSChris Mason 6130b32f4bbSJosef Bacik /* the pending IO might have been the only thing that kept this buffer 6140b32f4bbSJosef Bacik * in memory. Make sure we have a ref for all this other checks 6150b32f4bbSJosef Bacik */ 61667439dadSDavid Sterba atomic_inc(&eb->refs); 6170b32f4bbSJosef Bacik 6180b32f4bbSJosef Bacik reads_done = atomic_dec_and_test(&eb->io_pages); 619727011e0SChris Mason if (!reads_done) 620727011e0SChris Mason goto err; 621f188591eSChris Mason 6225cf1ab56SJosef Bacik eb->read_mirror = mirror; 623656f30dbSFilipe Manana if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 624ea466794SJosef Bacik ret = -EIO; 625ea466794SJosef Bacik goto err; 626ea466794SJosef Bacik } 627ea466794SJosef Bacik 628ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 629727011e0SChris Mason if (found_start != eb->start) { 630893bf4b1SSu Yue btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu", 631893bf4b1SSu Yue eb->start, found_start); 632f188591eSChris Mason ret = -EIO; 633ce9adaa5SChris Mason goto err; 634ce9adaa5SChris Mason } 635b0c9b3b0SDavid Sterba if (check_tree_block_fsid(eb)) { 63602873e43SZhao Lei btrfs_err_rl(fs_info, "bad fsid on block %llu", 63794647322SDavid Sterba eb->start); 6381259ab75SChris Mason ret = -EIO; 6391259ab75SChris Mason goto err; 6401259ab75SChris Mason } 641ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 6421c24c3ceSJosef Bacik if (found_level >= BTRFS_MAX_LEVEL) { 643893bf4b1SSu Yue btrfs_err(fs_info, "bad tree block level %d on %llu", 644893bf4b1SSu Yue (int)btrfs_header_level(eb), eb->start); 6451c24c3ceSJosef Bacik ret = -EIO; 6461c24c3ceSJosef Bacik goto err; 6471c24c3ceSJosef Bacik } 648ce9adaa5SChris Mason 64985d4e461SChris Mason btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), 65085d4e461SChris Mason eb, found_level); 6514008c04aSChris Mason 6522996e1f8SJohannes Thumshirn ret = csum_tree_block(eb, result); 6538bd98f0eSAlex Lyakas if (ret) 654a826d6dcSJosef Bacik goto err; 655a826d6dcSJosef Bacik 6562996e1f8SJohannes Thumshirn if (memcmp_extent_buffer(eb, result, 0, csum_size)) { 6572996e1f8SJohannes Thumshirn u32 val; 6582996e1f8SJohannes Thumshirn u32 found = 0; 6592996e1f8SJohannes Thumshirn 6602996e1f8SJohannes Thumshirn memcpy(&found, result, csum_size); 6612996e1f8SJohannes Thumshirn 6622996e1f8SJohannes Thumshirn read_extent_buffer(eb, &val, 0, csum_size); 6632996e1f8SJohannes Thumshirn btrfs_warn_rl(fs_info, 6642996e1f8SJohannes Thumshirn "%s checksum verify failed on %llu wanted %x found %x level %d", 6652996e1f8SJohannes Thumshirn fs_info->sb->s_id, eb->start, 6662996e1f8SJohannes Thumshirn val, found, btrfs_header_level(eb)); 6672996e1f8SJohannes Thumshirn ret = -EUCLEAN; 6682996e1f8SJohannes Thumshirn goto err; 6692996e1f8SJohannes Thumshirn } 6702996e1f8SJohannes Thumshirn 671a826d6dcSJosef Bacik /* 672a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 673a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 674a826d6dcSJosef Bacik * return -EIO. 675a826d6dcSJosef Bacik */ 6761c4360eeSDavid Sterba if (found_level == 0 && btrfs_check_leaf_full(eb)) { 677a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 678a826d6dcSJosef Bacik ret = -EIO; 679a826d6dcSJosef Bacik } 680ce9adaa5SChris Mason 681813fd1dcSDavid Sterba if (found_level > 0 && btrfs_check_node(eb)) 682053ab70fSLiu Bo ret = -EIO; 683053ab70fSLiu Bo 6840b32f4bbSJosef Bacik if (!ret) 6850b32f4bbSJosef Bacik set_extent_buffer_uptodate(eb); 68675391f0dSQu Wenruo else 68775391f0dSQu Wenruo btrfs_err(fs_info, 68875391f0dSQu Wenruo "block=%llu read time tree block corruption detected", 68975391f0dSQu Wenruo eb->start); 690ce9adaa5SChris Mason err: 69179fb65a1SJosef Bacik if (reads_done && 69279fb65a1SJosef Bacik test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 693d48d71aaSDavid Sterba btree_readahead_hook(eb, ret); 6944bb31e92SArne Jansen 69553b381b3SDavid Woodhouse if (ret) { 69653b381b3SDavid Woodhouse /* 69753b381b3SDavid Woodhouse * our io error hook is going to dec the io pages 69853b381b3SDavid Woodhouse * again, we have to make sure it has something 69953b381b3SDavid Woodhouse * to decrement 70053b381b3SDavid Woodhouse */ 70153b381b3SDavid Woodhouse atomic_inc(&eb->io_pages); 7020b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 70353b381b3SDavid Woodhouse } 7040b32f4bbSJosef Bacik free_extent_buffer(eb); 705ce9adaa5SChris Mason out: 706f188591eSChris Mason return ret; 707ce9adaa5SChris Mason } 708ce9adaa5SChris Mason 7094246a0b6SChristoph Hellwig static void end_workqueue_bio(struct bio *bio) 710ce9adaa5SChris Mason { 71197eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq = bio->bi_private; 712ce9adaa5SChris Mason struct btrfs_fs_info *fs_info; 7139e0af237SLiu Bo struct btrfs_workqueue *wq; 714ce9adaa5SChris Mason 715ce9adaa5SChris Mason fs_info = end_io_wq->info; 7164e4cbee9SChristoph Hellwig end_io_wq->status = bio->bi_status; 717d20f7043SChris Mason 71837226b21SMike Christie if (bio_op(bio) == REQ_OP_WRITE) { 719a0cac0ecSOmar Sandoval if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) 7209e0af237SLiu Bo wq = fs_info->endio_meta_write_workers; 721a0cac0ecSOmar Sandoval else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) 7229e0af237SLiu Bo wq = fs_info->endio_freespace_worker; 723a0cac0ecSOmar Sandoval else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 7249e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 725a0cac0ecSOmar Sandoval else 7269e0af237SLiu Bo wq = fs_info->endio_write_workers; 7279e0af237SLiu Bo } else { 728a0cac0ecSOmar Sandoval if (unlikely(end_io_wq->metadata == BTRFS_WQ_ENDIO_DIO_REPAIR)) 7298b110e39SMiao Xie wq = fs_info->endio_repair_workers; 730a0cac0ecSOmar Sandoval else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 7319e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 732a0cac0ecSOmar Sandoval else if (end_io_wq->metadata) 7339e0af237SLiu Bo wq = fs_info->endio_meta_workers; 734a0cac0ecSOmar Sandoval else 7359e0af237SLiu Bo wq = fs_info->endio_workers; 7369e0af237SLiu Bo } 7379e0af237SLiu Bo 738a0cac0ecSOmar Sandoval btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL); 7399e0af237SLiu Bo btrfs_queue_work(wq, &end_io_wq->work); 740ce9adaa5SChris Mason } 741ce9adaa5SChris Mason 7424e4cbee9SChristoph Hellwig blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 743bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata) 7440b86a832SChris Mason { 74597eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 7468b110e39SMiao Xie 74797eb6b69SDavid Sterba end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS); 748ce9adaa5SChris Mason if (!end_io_wq) 7494e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 750ce9adaa5SChris Mason 751ce9adaa5SChris Mason end_io_wq->private = bio->bi_private; 752ce9adaa5SChris Mason end_io_wq->end_io = bio->bi_end_io; 75322c59948SChris Mason end_io_wq->info = info; 7544e4cbee9SChristoph Hellwig end_io_wq->status = 0; 755ce9adaa5SChris Mason end_io_wq->bio = bio; 75622c59948SChris Mason end_io_wq->metadata = metadata; 757ce9adaa5SChris Mason 758ce9adaa5SChris Mason bio->bi_private = end_io_wq; 759ce9adaa5SChris Mason bio->bi_end_io = end_workqueue_bio; 76022c59948SChris Mason return 0; 76122c59948SChris Mason } 76222c59948SChris Mason 7634a69a410SChris Mason static void run_one_async_start(struct btrfs_work *work) 7644a69a410SChris Mason { 7654a69a410SChris Mason struct async_submit_bio *async; 7664e4cbee9SChristoph Hellwig blk_status_t ret; 7674a69a410SChris Mason 7684a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 769c6100a4bSJosef Bacik ret = async->submit_bio_start(async->private_data, async->bio, 770eaf25d93SChris Mason async->bio_offset); 77179787eaaSJeff Mahoney if (ret) 7724e4cbee9SChristoph Hellwig async->status = ret; 7734a69a410SChris Mason } 7744a69a410SChris Mason 77506ea01b1SDavid Sterba /* 77606ea01b1SDavid Sterba * In order to insert checksums into the metadata in large chunks, we wait 77706ea01b1SDavid Sterba * until bio submission time. All the pages in the bio are checksummed and 77806ea01b1SDavid Sterba * sums are attached onto the ordered extent record. 77906ea01b1SDavid Sterba * 78006ea01b1SDavid Sterba * At IO completion time the csums attached on the ordered extent record are 78106ea01b1SDavid Sterba * inserted into the tree. 78206ea01b1SDavid Sterba */ 7834a69a410SChris Mason static void run_one_async_done(struct btrfs_work *work) 7848b712842SChris Mason { 7858b712842SChris Mason struct async_submit_bio *async; 78606ea01b1SDavid Sterba struct inode *inode; 78706ea01b1SDavid Sterba blk_status_t ret; 7888b712842SChris Mason 7898b712842SChris Mason async = container_of(work, struct async_submit_bio, work); 79006ea01b1SDavid Sterba inode = async->private_data; 7914854ddd0SChris Mason 792bb7ab3b9SAdam Buchbinder /* If an error occurred we just want to clean up the bio and move on */ 7934e4cbee9SChristoph Hellwig if (async->status) { 7944e4cbee9SChristoph Hellwig async->bio->bi_status = async->status; 7954246a0b6SChristoph Hellwig bio_endio(async->bio); 79679787eaaSJeff Mahoney return; 79779787eaaSJeff Mahoney } 79879787eaaSJeff Mahoney 799ec39f769SChris Mason /* 800ec39f769SChris Mason * All of the bios that pass through here are from async helpers. 801ec39f769SChris Mason * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context. 802ec39f769SChris Mason * This changes nothing when cgroups aren't in use. 803ec39f769SChris Mason */ 804ec39f769SChris Mason async->bio->bi_opf |= REQ_CGROUP_PUNT; 80508635baeSChris Mason ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num); 80606ea01b1SDavid Sterba if (ret) { 80706ea01b1SDavid Sterba async->bio->bi_status = ret; 80806ea01b1SDavid Sterba bio_endio(async->bio); 80906ea01b1SDavid Sterba } 8104a69a410SChris Mason } 8114a69a410SChris Mason 8124a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 8134a69a410SChris Mason { 8144a69a410SChris Mason struct async_submit_bio *async; 8154a69a410SChris Mason 8164a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 8178b712842SChris Mason kfree(async); 8188b712842SChris Mason } 8198b712842SChris Mason 8208c27cb35SLinus Torvalds blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio, 821c6100a4bSJosef Bacik int mirror_num, unsigned long bio_flags, 822c6100a4bSJosef Bacik u64 bio_offset, void *private_data, 823e288c080SDavid Sterba extent_submit_bio_start_t *submit_bio_start) 82444b8bd7eSChris Mason { 82544b8bd7eSChris Mason struct async_submit_bio *async; 82644b8bd7eSChris Mason 82744b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 82844b8bd7eSChris Mason if (!async) 8294e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 83044b8bd7eSChris Mason 831c6100a4bSJosef Bacik async->private_data = private_data; 83244b8bd7eSChris Mason async->bio = bio; 83344b8bd7eSChris Mason async->mirror_num = mirror_num; 8344a69a410SChris Mason async->submit_bio_start = submit_bio_start; 8354a69a410SChris Mason 836a0cac0ecSOmar Sandoval btrfs_init_work(&async->work, run_one_async_start, run_one_async_done, 837a0cac0ecSOmar Sandoval run_one_async_free); 8384a69a410SChris Mason 839eaf25d93SChris Mason async->bio_offset = bio_offset; 8408c8bee1dSChris Mason 8414e4cbee9SChristoph Hellwig async->status = 0; 84279787eaaSJeff Mahoney 84367f055c7SChristoph Hellwig if (op_is_sync(bio->bi_opf)) 8445cdc7ad3SQu Wenruo btrfs_set_work_high_priority(&async->work); 845d313d7a3SChris Mason 8465cdc7ad3SQu Wenruo btrfs_queue_work(fs_info->workers, &async->work); 84744b8bd7eSChris Mason return 0; 84844b8bd7eSChris Mason } 84944b8bd7eSChris Mason 8504e4cbee9SChristoph Hellwig static blk_status_t btree_csum_one_bio(struct bio *bio) 851ce3ed71aSChris Mason { 8522c30c71bSKent Overstreet struct bio_vec *bvec; 853ce3ed71aSChris Mason struct btrfs_root *root; 8542b070cfeSChristoph Hellwig int ret = 0; 8556dc4f100SMing Lei struct bvec_iter_all iter_all; 856ce3ed71aSChris Mason 857c09abff8SDavid Sterba ASSERT(!bio_flagged(bio, BIO_CLONED)); 8582b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 859ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 86001d58472SDaniel Dressler ret = csum_dirty_buffer(root->fs_info, bvec->bv_page); 86179787eaaSJeff Mahoney if (ret) 86279787eaaSJeff Mahoney break; 863ce3ed71aSChris Mason } 8642c30c71bSKent Overstreet 8654e4cbee9SChristoph Hellwig return errno_to_blk_status(ret); 866ce3ed71aSChris Mason } 867ce3ed71aSChris Mason 868d0ee3934SDavid Sterba static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio, 869eaf25d93SChris Mason u64 bio_offset) 87022c59948SChris Mason { 8718b712842SChris Mason /* 8728b712842SChris Mason * when we're called for a write, we're already in the async 8735443be45SChris Mason * submission context. Just jump into btrfs_map_bio 8748b712842SChris Mason */ 87579787eaaSJeff Mahoney return btree_csum_one_bio(bio); 87622c59948SChris Mason } 87722c59948SChris Mason 8789b4e675aSDavid Sterba static int check_async_write(struct btrfs_fs_info *fs_info, 8799b4e675aSDavid Sterba struct btrfs_inode *bi) 880de0022b9SJosef Bacik { 8816300463bSLiu Bo if (atomic_read(&bi->sync_writers)) 8826300463bSLiu Bo return 0; 8839b4e675aSDavid Sterba if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags)) 884de0022b9SJosef Bacik return 0; 885de0022b9SJosef Bacik return 1; 886de0022b9SJosef Bacik } 887de0022b9SJosef Bacik 888a56b1c7bSNikolay Borisov static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio, 88950489a57SNikolay Borisov int mirror_num, 89050489a57SNikolay Borisov unsigned long bio_flags) 89144b8bd7eSChris Mason { 8920b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 8939b4e675aSDavid Sterba int async = check_async_write(fs_info, BTRFS_I(inode)); 8944e4cbee9SChristoph Hellwig blk_status_t ret; 895cad321adSChris Mason 89637226b21SMike Christie if (bio_op(bio) != REQ_OP_WRITE) { 897cad321adSChris Mason /* 898cad321adSChris Mason * called for a read, do the setup so that checksum validation 899cad321adSChris Mason * can happen in the async kernel threads 900cad321adSChris Mason */ 9010b246afaSJeff Mahoney ret = btrfs_bio_wq_end_io(fs_info, bio, 9020b246afaSJeff Mahoney BTRFS_WQ_ENDIO_METADATA); 9031d4284bdSChris Mason if (ret) 90461891923SStefan Behrens goto out_w_error; 90508635baeSChris Mason ret = btrfs_map_bio(fs_info, bio, mirror_num); 906de0022b9SJosef Bacik } else if (!async) { 907de0022b9SJosef Bacik ret = btree_csum_one_bio(bio); 908de0022b9SJosef Bacik if (ret) 90961891923SStefan Behrens goto out_w_error; 91008635baeSChris Mason ret = btrfs_map_bio(fs_info, bio, mirror_num); 91161891923SStefan Behrens } else { 912cad321adSChris Mason /* 91361891923SStefan Behrens * kthread helpers are used to submit writes so that 91461891923SStefan Behrens * checksumming can happen in parallel across all CPUs 915cad321adSChris Mason */ 916c6100a4bSJosef Bacik ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0, 917e68f2ee7SNikolay Borisov 0, inode, btree_submit_bio_start); 91844b8bd7eSChris Mason } 91944b8bd7eSChris Mason 9204246a0b6SChristoph Hellwig if (ret) 9214246a0b6SChristoph Hellwig goto out_w_error; 9224246a0b6SChristoph Hellwig return 0; 9234246a0b6SChristoph Hellwig 92461891923SStefan Behrens out_w_error: 9254e4cbee9SChristoph Hellwig bio->bi_status = ret; 9264246a0b6SChristoph Hellwig bio_endio(bio); 92761891923SStefan Behrens return ret; 92861891923SStefan Behrens } 92961891923SStefan Behrens 9303dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 931784b4e29SChris Mason static int btree_migratepage(struct address_space *mapping, 932a6bc32b8SMel Gorman struct page *newpage, struct page *page, 933a6bc32b8SMel Gorman enum migrate_mode mode) 934784b4e29SChris Mason { 935784b4e29SChris Mason /* 936784b4e29SChris Mason * we can't safely write a btree page from here, 937784b4e29SChris Mason * we haven't done the locking hook 938784b4e29SChris Mason */ 939784b4e29SChris Mason if (PageDirty(page)) 940784b4e29SChris Mason return -EAGAIN; 941784b4e29SChris Mason /* 942784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 943784b4e29SChris Mason * We must have no buffers or drop them. 944784b4e29SChris Mason */ 945784b4e29SChris Mason if (page_has_private(page) && 946784b4e29SChris Mason !try_to_release_page(page, GFP_KERNEL)) 947784b4e29SChris Mason return -EAGAIN; 948a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 949784b4e29SChris Mason } 9503dd1462eSJan Beulich #endif 951784b4e29SChris Mason 9520da5468fSChris Mason 9530da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 9540da5468fSChris Mason struct writeback_control *wbc) 9550da5468fSChris Mason { 956e2d84521SMiao Xie struct btrfs_fs_info *fs_info; 957e2d84521SMiao Xie int ret; 958e2d84521SMiao Xie 959d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 960448d640bSChris Mason 961448d640bSChris Mason if (wbc->for_kupdate) 962448d640bSChris Mason return 0; 963448d640bSChris Mason 964e2d84521SMiao Xie fs_info = BTRFS_I(mapping->host)->root->fs_info; 965b9473439SChris Mason /* this is a bit racy, but that's ok */ 966d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 967d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 968d814a491SEthan Lien fs_info->dirty_metadata_batch); 969e2d84521SMiao Xie if (ret < 0) 970793955bcSChris Mason return 0; 971793955bcSChris Mason } 9720b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 9730da5468fSChris Mason } 9740da5468fSChris Mason 975b2950863SChristoph Hellwig static int btree_readpage(struct file *file, struct page *page) 9765f39d397SChris Mason { 977d1310b2eSChris Mason struct extent_io_tree *tree; 978d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9798ddc7d9cSJan Schmidt return extent_read_full_page(tree, page, btree_get_extent, 0); 9805f39d397SChris Mason } 9815f39d397SChris Mason 98270dec807SChris Mason static int btree_releasepage(struct page *page, gfp_t gfp_flags) 9835f39d397SChris Mason { 98498509cfcSChris Mason if (PageWriteback(page) || PageDirty(page)) 98598509cfcSChris Mason return 0; 9860c4e538bSDavid Sterba 987f7a52a40SDavid Sterba return try_release_extent_buffer(page); 988d98237b3SChris Mason } 989d98237b3SChris Mason 990d47992f8SLukas Czerner static void btree_invalidatepage(struct page *page, unsigned int offset, 991d47992f8SLukas Czerner unsigned int length) 992d98237b3SChris Mason { 993d1310b2eSChris Mason struct extent_io_tree *tree; 994d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9955f39d397SChris Mason extent_invalidatepage(tree, page, offset); 9965f39d397SChris Mason btree_releasepage(page, GFP_NOFS); 9979ad6b7bcSChris Mason if (PagePrivate(page)) { 998efe120a0SFrank Holton btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info, 999efe120a0SFrank Holton "page private not zero on page %llu", 1000efe120a0SFrank Holton (unsigned long long)page_offset(page)); 10019ad6b7bcSChris Mason ClearPagePrivate(page); 10029ad6b7bcSChris Mason set_page_private(page, 0); 100309cbfeafSKirill A. Shutemov put_page(page); 10049ad6b7bcSChris Mason } 1005d98237b3SChris Mason } 1006d98237b3SChris Mason 10070b32f4bbSJosef Bacik static int btree_set_page_dirty(struct page *page) 10080b32f4bbSJosef Bacik { 1009bb146eb2SJosef Bacik #ifdef DEBUG 10100b32f4bbSJosef Bacik struct extent_buffer *eb; 10110b32f4bbSJosef Bacik 10120b32f4bbSJosef Bacik BUG_ON(!PagePrivate(page)); 10130b32f4bbSJosef Bacik eb = (struct extent_buffer *)page->private; 10140b32f4bbSJosef Bacik BUG_ON(!eb); 10150b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 10160b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 10170b32f4bbSJosef Bacik btrfs_assert_tree_locked(eb); 1018bb146eb2SJosef Bacik #endif 10190b32f4bbSJosef Bacik return __set_page_dirty_nobuffers(page); 10200b32f4bbSJosef Bacik } 10210b32f4bbSJosef Bacik 10227f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 1023d98237b3SChris Mason .readpage = btree_readpage, 10240da5468fSChris Mason .writepages = btree_writepages, 10255f39d397SChris Mason .releasepage = btree_releasepage, 10265f39d397SChris Mason .invalidatepage = btree_invalidatepage, 10275a92bc88SChris Mason #ifdef CONFIG_MIGRATION 1028784b4e29SChris Mason .migratepage = btree_migratepage, 10295a92bc88SChris Mason #endif 10300b32f4bbSJosef Bacik .set_page_dirty = btree_set_page_dirty, 1031d98237b3SChris Mason }; 1032123abc88SChris Mason 10332ff7e61eSJeff Mahoney void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr) 1034090d1875SChris Mason { 10355f39d397SChris Mason struct extent_buffer *buf = NULL; 1036537f38f0SNikolay Borisov int ret; 1037090d1875SChris Mason 10382ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 1039c871b0f2SLiu Bo if (IS_ERR(buf)) 10406197d86eSDavid Sterba return; 1041537f38f0SNikolay Borisov 1042c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(buf, WAIT_NONE, 0); 1043537f38f0SNikolay Borisov if (ret < 0) 1044537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 1045537f38f0SNikolay Borisov else 10465f39d397SChris Mason free_extent_buffer(buf); 1047090d1875SChris Mason } 1048090d1875SChris Mason 10492ff7e61eSJeff Mahoney struct extent_buffer *btrfs_find_create_tree_block( 10502ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info, 10510308af44SDavid Sterba u64 bytenr) 10520999df54SChris Mason { 10530b246afaSJeff Mahoney if (btrfs_is_testing(fs_info)) 10540b246afaSJeff Mahoney return alloc_test_extent_buffer(fs_info, bytenr); 10550b246afaSJeff Mahoney return alloc_extent_buffer(fs_info, bytenr); 10560999df54SChris Mason } 10570999df54SChris Mason 1058581c1760SQu Wenruo /* 1059581c1760SQu Wenruo * Read tree block at logical address @bytenr and do variant basic but critical 1060581c1760SQu Wenruo * verification. 1061581c1760SQu Wenruo * 1062581c1760SQu Wenruo * @parent_transid: expected transid of this tree block, skip check if 0 1063581c1760SQu Wenruo * @level: expected level, mandatory check 1064581c1760SQu Wenruo * @first_key: expected key in slot 0, skip check if NULL 1065581c1760SQu Wenruo */ 10662ff7e61eSJeff Mahoney struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, 1067581c1760SQu Wenruo u64 parent_transid, int level, 1068581c1760SQu Wenruo struct btrfs_key *first_key) 1069e20d96d6SChris Mason { 10705f39d397SChris Mason struct extent_buffer *buf = NULL; 107119c00ddcSChris Mason int ret; 107219c00ddcSChris Mason 10732ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 1074c871b0f2SLiu Bo if (IS_ERR(buf)) 1075c871b0f2SLiu Bo return buf; 1076e4204dedSChris Mason 10775ab12d1fSDavid Sterba ret = btree_read_extent_buffer_pages(buf, parent_transid, 1078581c1760SQu Wenruo level, first_key); 10790f0fe8f7SFilipe David Borba Manana if (ret) { 1080537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 108164c043deSLiu Bo return ERR_PTR(ret); 10820f0fe8f7SFilipe David Borba Manana } 10835f39d397SChris Mason return buf; 1084ce9adaa5SChris Mason 1085eb60ceacSChris Mason } 1086eb60ceacSChris Mason 10876a884d7dSDavid Sterba void btrfs_clean_tree_block(struct extent_buffer *buf) 1088ed2ff2cbSChris Mason { 10896a884d7dSDavid Sterba struct btrfs_fs_info *fs_info = buf->fs_info; 109055c69072SChris Mason if (btrfs_header_generation(buf) == 1091e2d84521SMiao Xie fs_info->running_transaction->transid) { 1092b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1093b4ce94deSChris Mason 1094b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1095104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 1096e2d84521SMiao Xie -buf->len, 1097e2d84521SMiao Xie fs_info->dirty_metadata_batch); 1098b9473439SChris Mason /* ugh, clear_extent_buffer_dirty needs to lock the page */ 10998bead258SDavid Sterba btrfs_set_lock_blocking_write(buf); 11000b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 1101925baeddSChris Mason } 11025f39d397SChris Mason } 1103ed7b63ebSJosef Bacik } 11045f39d397SChris Mason 11058257b2dcSMiao Xie static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void) 11068257b2dcSMiao Xie { 11078257b2dcSMiao Xie struct btrfs_subvolume_writers *writers; 11088257b2dcSMiao Xie int ret; 11098257b2dcSMiao Xie 11108257b2dcSMiao Xie writers = kmalloc(sizeof(*writers), GFP_NOFS); 11118257b2dcSMiao Xie if (!writers) 11128257b2dcSMiao Xie return ERR_PTR(-ENOMEM); 11138257b2dcSMiao Xie 11148a5a916dSJeff Mahoney ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS); 11158257b2dcSMiao Xie if (ret < 0) { 11168257b2dcSMiao Xie kfree(writers); 11178257b2dcSMiao Xie return ERR_PTR(ret); 11188257b2dcSMiao Xie } 11198257b2dcSMiao Xie 11208257b2dcSMiao Xie init_waitqueue_head(&writers->wait); 11218257b2dcSMiao Xie return writers; 11228257b2dcSMiao Xie } 11238257b2dcSMiao Xie 11248257b2dcSMiao Xie static void 11258257b2dcSMiao Xie btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers) 11268257b2dcSMiao Xie { 11278257b2dcSMiao Xie percpu_counter_destroy(&writers->counter); 11288257b2dcSMiao Xie kfree(writers); 11298257b2dcSMiao Xie } 11308257b2dcSMiao Xie 1131da17066cSJeff Mahoney static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, 1132e20d96d6SChris Mason u64 objectid) 1133d97e63b6SChris Mason { 11347c0260eeSJeff Mahoney bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); 1135cfaa7295SChris Mason root->node = NULL; 1136a28ec197SChris Mason root->commit_root = NULL; 113727cdeb70SMiao Xie root->state = 0; 1138d68fc57bSYan, Zheng root->orphan_cleanup_state = 0; 11390b86a832SChris Mason 11400f7d52f4SChris Mason root->last_trans = 0; 114113a8a7c8SYan, Zheng root->highest_objectid = 0; 1142eb73c1b7SMiao Xie root->nr_delalloc_inodes = 0; 1143199c2a9cSMiao Xie root->nr_ordered_extents = 0; 11446bef4d31SEric Paris root->inode_tree = RB_ROOT; 114516cdcec7SMiao Xie INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1146f0486c68SYan, Zheng root->block_rsv = NULL; 11470b86a832SChris Mason 11480b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 11495d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 1150eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_inodes); 1151eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_root); 1152199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_extents); 1153199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_root); 1154d2311e69SQu Wenruo INIT_LIST_HEAD(&root->reloc_dirty_list); 11552ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 11562ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 11575d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1158eb73c1b7SMiao Xie spin_lock_init(&root->delalloc_lock); 1159199c2a9cSMiao Xie spin_lock_init(&root->ordered_extent_lock); 1160f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 11612ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 11622ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 11638287475aSQu Wenruo spin_lock_init(&root->qgroup_meta_rsv_lock); 1164a2135011SChris Mason mutex_init(&root->objectid_mutex); 1165e02119d5SChris Mason mutex_init(&root->log_mutex); 116631f3d255SMiao Xie mutex_init(&root->ordered_extent_mutex); 1167573bfb72SMiao Xie mutex_init(&root->delalloc_mutex); 11687237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 11697237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 11707237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 11718b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[0]); 11728b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[1]); 11737237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 11747237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 11757237f183SYan Zheng atomic_set(&root->log_writers, 0); 11762ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 11770700cea7SElena Reshetova refcount_set(&root->refs, 1); 1178ea14b57fSDavid Sterba atomic_set(&root->will_be_snapshotted, 0); 11798ecebf4dSRobbie Ko atomic_set(&root->snapshot_force_cow, 0); 1180eede2bf3SOmar Sandoval atomic_set(&root->nr_swapfiles, 0); 11817237f183SYan Zheng root->log_transid = 0; 1182d1433debSMiao Xie root->log_transid_committed = -1; 1183257c62e1SChris Mason root->last_log_commit = 0; 11847c0260eeSJeff Mahoney if (!dummy) 118543eb5f29SQu Wenruo extent_io_tree_init(fs_info, &root->dirty_log_pages, 118643eb5f29SQu Wenruo IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL); 1187017e5369SChris Mason 11883768f368SChris Mason memset(&root->root_key, 0, sizeof(root->root_key)); 11893768f368SChris Mason memset(&root->root_item, 0, sizeof(root->root_item)); 11906702ed49SChris Mason memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 11917c0260eeSJeff Mahoney if (!dummy) 11923f157a2fSChris Mason root->defrag_trans_start = fs_info->generation; 119306ea65a3SJosef Bacik else 119406ea65a3SJosef Bacik root->defrag_trans_start = 0; 11954d775673SChris Mason root->root_key.objectid = objectid; 11960ee5dc67SAl Viro root->anon_dev = 0; 11978ea05e3aSAlexander Block 11985f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 1199370a11b8SQu Wenruo btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks); 12003768f368SChris Mason } 12013768f368SChris Mason 120274e4d827SDavid Sterba static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, 120374e4d827SDavid Sterba gfp_t flags) 12046f07e42eSAl Viro { 120574e4d827SDavid Sterba struct btrfs_root *root = kzalloc(sizeof(*root), flags); 12066f07e42eSAl Viro if (root) 12076f07e42eSAl Viro root->fs_info = fs_info; 12086f07e42eSAl Viro return root; 12096f07e42eSAl Viro } 12106f07e42eSAl Viro 121106ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 121206ea65a3SJosef Bacik /* Should only be used by the testing infrastructure */ 1213da17066cSJeff Mahoney struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) 121406ea65a3SJosef Bacik { 121506ea65a3SJosef Bacik struct btrfs_root *root; 121606ea65a3SJosef Bacik 12177c0260eeSJeff Mahoney if (!fs_info) 12187c0260eeSJeff Mahoney return ERR_PTR(-EINVAL); 12197c0260eeSJeff Mahoney 12207c0260eeSJeff Mahoney root = btrfs_alloc_root(fs_info, GFP_KERNEL); 122106ea65a3SJosef Bacik if (!root) 122206ea65a3SJosef Bacik return ERR_PTR(-ENOMEM); 1223da17066cSJeff Mahoney 1224b9ef22deSFeifei Xu /* We don't use the stripesize in selftest, set it as sectorsize */ 1225da17066cSJeff Mahoney __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID); 1226faa2dbf0SJosef Bacik root->alloc_bytenr = 0; 122706ea65a3SJosef Bacik 122806ea65a3SJosef Bacik return root; 122906ea65a3SJosef Bacik } 123006ea65a3SJosef Bacik #endif 123106ea65a3SJosef Bacik 123220897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 123320897f5cSArne Jansen u64 objectid) 123420897f5cSArne Jansen { 12359b7a2440SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 123620897f5cSArne Jansen struct extent_buffer *leaf; 123720897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 123820897f5cSArne Jansen struct btrfs_root *root; 123920897f5cSArne Jansen struct btrfs_key key; 1240b89f6d1fSFilipe Manana unsigned int nofs_flag; 124120897f5cSArne Jansen int ret = 0; 124233d85fdaSQu Wenruo uuid_le uuid = NULL_UUID_LE; 124320897f5cSArne Jansen 1244b89f6d1fSFilipe Manana /* 1245b89f6d1fSFilipe Manana * We're holding a transaction handle, so use a NOFS memory allocation 1246b89f6d1fSFilipe Manana * context to avoid deadlock if reclaim happens. 1247b89f6d1fSFilipe Manana */ 1248b89f6d1fSFilipe Manana nofs_flag = memalloc_nofs_save(); 124974e4d827SDavid Sterba root = btrfs_alloc_root(fs_info, GFP_KERNEL); 1250b89f6d1fSFilipe Manana memalloc_nofs_restore(nofs_flag); 125120897f5cSArne Jansen if (!root) 125220897f5cSArne Jansen return ERR_PTR(-ENOMEM); 125320897f5cSArne Jansen 1254da17066cSJeff Mahoney __setup_root(root, fs_info, objectid); 125520897f5cSArne Jansen root->root_key.objectid = objectid; 125620897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 125720897f5cSArne Jansen root->root_key.offset = 0; 125820897f5cSArne Jansen 12594d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0); 126020897f5cSArne Jansen if (IS_ERR(leaf)) { 126120897f5cSArne Jansen ret = PTR_ERR(leaf); 12621dd05682STsutomu Itoh leaf = NULL; 126320897f5cSArne Jansen goto fail; 126420897f5cSArne Jansen } 126520897f5cSArne Jansen 126620897f5cSArne Jansen root->node = leaf; 126720897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 126820897f5cSArne Jansen 126920897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 127027cdeb70SMiao Xie set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 127120897f5cSArne Jansen 127220897f5cSArne Jansen root->root_item.flags = 0; 127320897f5cSArne Jansen root->root_item.byte_limit = 0; 127420897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 127520897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 127620897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 127720897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 127820897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 127920897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 128020897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 128133d85fdaSQu Wenruo if (is_fstree(objectid)) 12826463fe58SStefan Behrens uuid_le_gen(&uuid); 12836463fe58SStefan Behrens memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE); 128420897f5cSArne Jansen root->root_item.drop_level = 0; 128520897f5cSArne Jansen 128620897f5cSArne Jansen key.objectid = objectid; 128720897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 128820897f5cSArne Jansen key.offset = 0; 128920897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 129020897f5cSArne Jansen if (ret) 129120897f5cSArne Jansen goto fail; 129220897f5cSArne Jansen 129320897f5cSArne Jansen btrfs_tree_unlock(leaf); 129420897f5cSArne Jansen 129520897f5cSArne Jansen return root; 12961dd05682STsutomu Itoh 12971dd05682STsutomu Itoh fail: 12981dd05682STsutomu Itoh if (leaf) { 12991dd05682STsutomu Itoh btrfs_tree_unlock(leaf); 130059885b39STsutomu Itoh free_extent_buffer(root->commit_root); 13011dd05682STsutomu Itoh free_extent_buffer(leaf); 13021dd05682STsutomu Itoh } 13031dd05682STsutomu Itoh kfree(root); 13041dd05682STsutomu Itoh 13051dd05682STsutomu Itoh return ERR_PTR(ret); 130620897f5cSArne Jansen } 130720897f5cSArne Jansen 13087237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1309e02119d5SChris Mason struct btrfs_fs_info *fs_info) 13100f7d52f4SChris Mason { 13110f7d52f4SChris Mason struct btrfs_root *root; 13127237f183SYan Zheng struct extent_buffer *leaf; 1313e02119d5SChris Mason 131474e4d827SDavid Sterba root = btrfs_alloc_root(fs_info, GFP_NOFS); 1315e02119d5SChris Mason if (!root) 13167237f183SYan Zheng return ERR_PTR(-ENOMEM); 1317e02119d5SChris Mason 1318da17066cSJeff Mahoney __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID); 1319e02119d5SChris Mason 1320e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1321e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1322e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 132327cdeb70SMiao Xie 13247237f183SYan Zheng /* 132527cdeb70SMiao Xie * DON'T set REF_COWS for log trees 132627cdeb70SMiao Xie * 13277237f183SYan Zheng * log trees do not get reference counted because they go away 13287237f183SYan Zheng * before a real commit is actually done. They do store pointers 13297237f183SYan Zheng * to file data extents, and those reference counts still get 13307237f183SYan Zheng * updated (along with back refs to the log tree). 13317237f183SYan Zheng */ 1332e02119d5SChris Mason 13334d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 13344d75f8a9SDavid Sterba NULL, 0, 0, 0); 13357237f183SYan Zheng if (IS_ERR(leaf)) { 13367237f183SYan Zheng kfree(root); 13377237f183SYan Zheng return ERR_CAST(leaf); 13387237f183SYan Zheng } 1339e02119d5SChris Mason 13407237f183SYan Zheng root->node = leaf; 1341e02119d5SChris Mason 1342e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1343e02119d5SChris Mason btrfs_tree_unlock(root->node); 13447237f183SYan Zheng return root; 13457237f183SYan Zheng } 13467237f183SYan Zheng 13477237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 13487237f183SYan Zheng struct btrfs_fs_info *fs_info) 13497237f183SYan Zheng { 13507237f183SYan Zheng struct btrfs_root *log_root; 13517237f183SYan Zheng 13527237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 13537237f183SYan Zheng if (IS_ERR(log_root)) 13547237f183SYan Zheng return PTR_ERR(log_root); 13557237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 13567237f183SYan Zheng fs_info->log_root_tree = log_root; 13577237f183SYan Zheng return 0; 13587237f183SYan Zheng } 13597237f183SYan Zheng 13607237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 13617237f183SYan Zheng struct btrfs_root *root) 13627237f183SYan Zheng { 13630b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13647237f183SYan Zheng struct btrfs_root *log_root; 13657237f183SYan Zheng struct btrfs_inode_item *inode_item; 13667237f183SYan Zheng 13670b246afaSJeff Mahoney log_root = alloc_log_tree(trans, fs_info); 13687237f183SYan Zheng if (IS_ERR(log_root)) 13697237f183SYan Zheng return PTR_ERR(log_root); 13707237f183SYan Zheng 13717237f183SYan Zheng log_root->last_trans = trans->transid; 13727237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 13737237f183SYan Zheng 13747237f183SYan Zheng inode_item = &log_root->root_item.inode; 13753cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 13763cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 13773cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 1378da17066cSJeff Mahoney btrfs_set_stack_inode_nbytes(inode_item, 13790b246afaSJeff Mahoney fs_info->nodesize); 13803cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 13817237f183SYan Zheng 13825d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 13837237f183SYan Zheng 13847237f183SYan Zheng WARN_ON(root->log_root); 13857237f183SYan Zheng root->log_root = log_root; 13867237f183SYan Zheng root->log_transid = 0; 1387d1433debSMiao Xie root->log_transid_committed = -1; 1388257c62e1SChris Mason root->last_log_commit = 0; 1389e02119d5SChris Mason return 0; 1390e02119d5SChris Mason } 1391e02119d5SChris Mason 139235a3621bSStefan Behrens static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 1393cb517eabSMiao Xie struct btrfs_key *key) 1394e02119d5SChris Mason { 1395e02119d5SChris Mason struct btrfs_root *root; 1396e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 13970f7d52f4SChris Mason struct btrfs_path *path; 139884234f3aSYan Zheng u64 generation; 1399cb517eabSMiao Xie int ret; 1400581c1760SQu Wenruo int level; 1401cb517eabSMiao Xie 1402cb517eabSMiao Xie path = btrfs_alloc_path(); 1403cb517eabSMiao Xie if (!path) 1404cb517eabSMiao Xie return ERR_PTR(-ENOMEM); 14050f7d52f4SChris Mason 140674e4d827SDavid Sterba root = btrfs_alloc_root(fs_info, GFP_NOFS); 1407cb517eabSMiao Xie if (!root) { 1408cb517eabSMiao Xie ret = -ENOMEM; 1409cb517eabSMiao Xie goto alloc_fail; 14100f7d52f4SChris Mason } 14110f7d52f4SChris Mason 1412da17066cSJeff Mahoney __setup_root(root, fs_info, key->objectid); 14130f7d52f4SChris Mason 1414cb517eabSMiao Xie ret = btrfs_find_root(tree_root, key, path, 1415cb517eabSMiao Xie &root->root_item, &root->root_key); 14160f7d52f4SChris Mason if (ret) { 141713a8a7c8SYan, Zheng if (ret > 0) 141813a8a7c8SYan, Zheng ret = -ENOENT; 1419cb517eabSMiao Xie goto find_fail; 14200f7d52f4SChris Mason } 142113a8a7c8SYan, Zheng 142284234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1423581c1760SQu Wenruo level = btrfs_root_level(&root->root_item); 14242ff7e61eSJeff Mahoney root->node = read_tree_block(fs_info, 14252ff7e61eSJeff Mahoney btrfs_root_bytenr(&root->root_item), 1426581c1760SQu Wenruo generation, level, NULL); 142764c043deSLiu Bo if (IS_ERR(root->node)) { 142864c043deSLiu Bo ret = PTR_ERR(root->node); 1429cb517eabSMiao Xie goto find_fail; 1430cb517eabSMiao Xie } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1431cb517eabSMiao Xie ret = -EIO; 143264c043deSLiu Bo free_extent_buffer(root->node); 143364c043deSLiu Bo goto find_fail; 1434416bc658SJosef Bacik } 14355d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 143613a8a7c8SYan, Zheng out: 1437cb517eabSMiao Xie btrfs_free_path(path); 1438cb517eabSMiao Xie return root; 1439cb517eabSMiao Xie 1440cb517eabSMiao Xie find_fail: 1441cb517eabSMiao Xie kfree(root); 1442cb517eabSMiao Xie alloc_fail: 1443cb517eabSMiao Xie root = ERR_PTR(ret); 1444cb517eabSMiao Xie goto out; 1445cb517eabSMiao Xie } 1446cb517eabSMiao Xie 1447cb517eabSMiao Xie struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root, 1448cb517eabSMiao Xie struct btrfs_key *location) 1449cb517eabSMiao Xie { 1450cb517eabSMiao Xie struct btrfs_root *root; 1451cb517eabSMiao Xie 1452cb517eabSMiao Xie root = btrfs_read_tree_root(tree_root, location); 1453cb517eabSMiao Xie if (IS_ERR(root)) 1454cb517eabSMiao Xie return root; 1455cb517eabSMiao Xie 1456cb517eabSMiao Xie if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { 145727cdeb70SMiao Xie set_bit(BTRFS_ROOT_REF_COWS, &root->state); 145808fe4db1SLi Zefan btrfs_check_and_init_root_item(&root->root_item); 145908fe4db1SLi Zefan } 146013a8a7c8SYan, Zheng 14615eda7b5eSChris Mason return root; 14625eda7b5eSChris Mason } 14635eda7b5eSChris Mason 1464cb517eabSMiao Xie int btrfs_init_fs_root(struct btrfs_root *root) 1465cb517eabSMiao Xie { 1466cb517eabSMiao Xie int ret; 14678257b2dcSMiao Xie struct btrfs_subvolume_writers *writers; 1468cb517eabSMiao Xie 1469cb517eabSMiao Xie root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); 1470cb517eabSMiao Xie root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), 1471cb517eabSMiao Xie GFP_NOFS); 1472cb517eabSMiao Xie if (!root->free_ino_pinned || !root->free_ino_ctl) { 1473cb517eabSMiao Xie ret = -ENOMEM; 1474cb517eabSMiao Xie goto fail; 1475cb517eabSMiao Xie } 1476cb517eabSMiao Xie 14778257b2dcSMiao Xie writers = btrfs_alloc_subvolume_writers(); 14788257b2dcSMiao Xie if (IS_ERR(writers)) { 14798257b2dcSMiao Xie ret = PTR_ERR(writers); 14808257b2dcSMiao Xie goto fail; 14818257b2dcSMiao Xie } 14828257b2dcSMiao Xie root->subv_writers = writers; 14838257b2dcSMiao Xie 1484cb517eabSMiao Xie btrfs_init_free_ino_ctl(root); 148557cdc8dbSDavid Sterba spin_lock_init(&root->ino_cache_lock); 148657cdc8dbSDavid Sterba init_waitqueue_head(&root->ino_cache_wait); 1487cb517eabSMiao Xie 1488cb517eabSMiao Xie ret = get_anon_bdev(&root->anon_dev); 1489cb517eabSMiao Xie if (ret) 1490876d2cf1SLiu Bo goto fail; 1491f32e48e9SChandan Rajendra 1492f32e48e9SChandan Rajendra mutex_lock(&root->objectid_mutex); 1493f32e48e9SChandan Rajendra ret = btrfs_find_highest_objectid(root, 1494f32e48e9SChandan Rajendra &root->highest_objectid); 1495f32e48e9SChandan Rajendra if (ret) { 1496f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1497876d2cf1SLiu Bo goto fail; 1498f32e48e9SChandan Rajendra } 1499f32e48e9SChandan Rajendra 1500f32e48e9SChandan Rajendra ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 1501f32e48e9SChandan Rajendra 1502f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1503f32e48e9SChandan Rajendra 1504cb517eabSMiao Xie return 0; 1505cb517eabSMiao Xie fail: 150684db5ccfSDavid Sterba /* The caller is responsible to call btrfs_free_fs_root */ 1507cb517eabSMiao Xie return ret; 1508cb517eabSMiao Xie } 1509cb517eabSMiao Xie 151035bbb97fSJeff Mahoney struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1511cb517eabSMiao Xie u64 root_id) 1512cb517eabSMiao Xie { 1513cb517eabSMiao Xie struct btrfs_root *root; 1514cb517eabSMiao Xie 1515cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1516cb517eabSMiao Xie root = radix_tree_lookup(&fs_info->fs_roots_radix, 1517cb517eabSMiao Xie (unsigned long)root_id); 1518cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1519cb517eabSMiao Xie return root; 1520cb517eabSMiao Xie } 1521cb517eabSMiao Xie 1522cb517eabSMiao Xie int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1523cb517eabSMiao Xie struct btrfs_root *root) 1524cb517eabSMiao Xie { 1525cb517eabSMiao Xie int ret; 1526cb517eabSMiao Xie 1527e1860a77SDavid Sterba ret = radix_tree_preload(GFP_NOFS); 1528cb517eabSMiao Xie if (ret) 1529cb517eabSMiao Xie return ret; 1530cb517eabSMiao Xie 1531cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1532cb517eabSMiao Xie ret = radix_tree_insert(&fs_info->fs_roots_radix, 1533cb517eabSMiao Xie (unsigned long)root->root_key.objectid, 1534cb517eabSMiao Xie root); 1535cb517eabSMiao Xie if (ret == 0) 153627cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1537cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1538cb517eabSMiao Xie radix_tree_preload_end(); 1539cb517eabSMiao Xie 1540cb517eabSMiao Xie return ret; 1541cb517eabSMiao Xie } 1542cb517eabSMiao Xie 1543c00869f1SMiao Xie struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 1544c00869f1SMiao Xie struct btrfs_key *location, 1545c00869f1SMiao Xie bool check_ref) 15465eda7b5eSChris Mason { 15475eda7b5eSChris Mason struct btrfs_root *root; 1548381cf658SDavid Sterba struct btrfs_path *path; 15491d4c08e0SDavid Sterba struct btrfs_key key; 15505eda7b5eSChris Mason int ret; 15515eda7b5eSChris Mason 1552edbd8d4eSChris Mason if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 1553edbd8d4eSChris Mason return fs_info->tree_root; 1554edbd8d4eSChris Mason if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID) 1555edbd8d4eSChris Mason return fs_info->extent_root; 15568f18cf13SChris Mason if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID) 15578f18cf13SChris Mason return fs_info->chunk_root; 15588f18cf13SChris Mason if (location->objectid == BTRFS_DEV_TREE_OBJECTID) 15598f18cf13SChris Mason return fs_info->dev_root; 15600403e47eSYan Zheng if (location->objectid == BTRFS_CSUM_TREE_OBJECTID) 15610403e47eSYan Zheng return fs_info->csum_root; 1562bcef60f2SArne Jansen if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID) 1563bcef60f2SArne Jansen return fs_info->quota_root ? fs_info->quota_root : 1564bcef60f2SArne Jansen ERR_PTR(-ENOENT); 1565f7a81ea4SStefan Behrens if (location->objectid == BTRFS_UUID_TREE_OBJECTID) 1566f7a81ea4SStefan Behrens return fs_info->uuid_root ? fs_info->uuid_root : 1567f7a81ea4SStefan Behrens ERR_PTR(-ENOENT); 156870f6d82eSOmar Sandoval if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) 156970f6d82eSOmar Sandoval return fs_info->free_space_root ? fs_info->free_space_root : 157070f6d82eSOmar Sandoval ERR_PTR(-ENOENT); 15714df27c4dSYan, Zheng again: 1572cb517eabSMiao Xie root = btrfs_lookup_fs_root(fs_info, location->objectid); 157348475471SStefan Behrens if (root) { 1574c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) 157548475471SStefan Behrens return ERR_PTR(-ENOENT); 15765eda7b5eSChris Mason return root; 157748475471SStefan Behrens } 15785eda7b5eSChris Mason 1579cb517eabSMiao Xie root = btrfs_read_fs_root(fs_info->tree_root, location); 15805eda7b5eSChris Mason if (IS_ERR(root)) 15815eda7b5eSChris Mason return root; 15823394e160SChris Mason 1583c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1584d68fc57bSYan, Zheng ret = -ENOENT; 1585d68fc57bSYan, Zheng goto fail; 1586d68fc57bSYan, Zheng } 1587d68fc57bSYan, Zheng 1588cb517eabSMiao Xie ret = btrfs_init_fs_root(root); 1589cb517eabSMiao Xie if (ret) 1590cb517eabSMiao Xie goto fail; 1591cb517eabSMiao Xie 1592381cf658SDavid Sterba path = btrfs_alloc_path(); 1593381cf658SDavid Sterba if (!path) { 1594381cf658SDavid Sterba ret = -ENOMEM; 1595381cf658SDavid Sterba goto fail; 1596381cf658SDavid Sterba } 15971d4c08e0SDavid Sterba key.objectid = BTRFS_ORPHAN_OBJECTID; 15981d4c08e0SDavid Sterba key.type = BTRFS_ORPHAN_ITEM_KEY; 15991d4c08e0SDavid Sterba key.offset = location->objectid; 16001d4c08e0SDavid Sterba 16011d4c08e0SDavid Sterba ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1602381cf658SDavid Sterba btrfs_free_path(path); 1603d68fc57bSYan, Zheng if (ret < 0) 1604d68fc57bSYan, Zheng goto fail; 1605d68fc57bSYan, Zheng if (ret == 0) 160627cdeb70SMiao Xie set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1607d68fc57bSYan, Zheng 1608cb517eabSMiao Xie ret = btrfs_insert_fs_root(fs_info, root); 16090f7d52f4SChris Mason if (ret) { 16104df27c4dSYan, Zheng if (ret == -EEXIST) { 161184db5ccfSDavid Sterba btrfs_free_fs_root(root); 16124df27c4dSYan, Zheng goto again; 16130f7d52f4SChris Mason } 16144df27c4dSYan, Zheng goto fail; 16154df27c4dSYan, Zheng } 1616edbd8d4eSChris Mason return root; 16174df27c4dSYan, Zheng fail: 161884db5ccfSDavid Sterba btrfs_free_fs_root(root); 16194df27c4dSYan, Zheng return ERR_PTR(ret); 1620edbd8d4eSChris Mason } 1621edbd8d4eSChris Mason 162204160088SChris Mason static int btrfs_congested_fn(void *congested_data, int bdi_bits) 162304160088SChris Mason { 162404160088SChris Mason struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; 162504160088SChris Mason int ret = 0; 162604160088SChris Mason struct btrfs_device *device; 162704160088SChris Mason struct backing_dev_info *bdi; 1628b7967db7SChris Mason 16291f78160cSXiao Guangrong rcu_read_lock(); 16301f78160cSXiao Guangrong list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) { 1631dfe25020SChris Mason if (!device->bdev) 1632dfe25020SChris Mason continue; 1633efa7c9f9SJan Kara bdi = device->bdev->bd_bdi; 1634ff9ea323STejun Heo if (bdi_congested(bdi, bdi_bits)) { 163504160088SChris Mason ret = 1; 163604160088SChris Mason break; 163704160088SChris Mason } 163804160088SChris Mason } 16391f78160cSXiao Guangrong rcu_read_unlock(); 164004160088SChris Mason return ret; 164104160088SChris Mason } 164204160088SChris Mason 16438b712842SChris Mason /* 16448b712842SChris Mason * called by the kthread helper functions to finally call the bio end_io 16458b712842SChris Mason * functions. This is where read checksum verification actually happens 16468b712842SChris Mason */ 16478b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work) 1648ce9adaa5SChris Mason { 1649ce9adaa5SChris Mason struct bio *bio; 165097eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 1651ce9adaa5SChris Mason 165297eb6b69SDavid Sterba end_io_wq = container_of(work, struct btrfs_end_io_wq, work); 1653ce9adaa5SChris Mason bio = end_io_wq->bio; 16548b712842SChris Mason 16554e4cbee9SChristoph Hellwig bio->bi_status = end_io_wq->status; 1656ce9adaa5SChris Mason bio->bi_private = end_io_wq->private; 1657ce9adaa5SChris Mason bio->bi_end_io = end_io_wq->end_io; 16584246a0b6SChristoph Hellwig bio_endio(bio); 16599be490f1SOmar Sandoval kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq); 1660ce9adaa5SChris Mason } 166144b8bd7eSChris Mason 1662a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1663a74a4b97SChris Mason { 1664a74a4b97SChris Mason struct btrfs_root *root = arg; 16650b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1666d0278245SMiao Xie int again; 1667a74a4b97SChris Mason 1668d6fd0ae2SOmar Sandoval while (1) { 1669d0278245SMiao Xie again = 0; 16709d1a2a3aSDavid Sterba 1671fd340d0fSJosef Bacik set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1672fd340d0fSJosef Bacik 1673d0278245SMiao Xie /* Make the cleaner go to sleep early. */ 16742ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) 1675d0278245SMiao Xie goto sleep; 1676d0278245SMiao Xie 167790c711abSZygo Blaxell /* 167890c711abSZygo Blaxell * Do not do anything if we might cause open_ctree() to block 167990c711abSZygo Blaxell * before we have finished mounting the filesystem. 168090c711abSZygo Blaxell */ 16810b246afaSJeff Mahoney if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 168290c711abSZygo Blaxell goto sleep; 168390c711abSZygo Blaxell 16840b246afaSJeff Mahoney if (!mutex_trylock(&fs_info->cleaner_mutex)) 1685d0278245SMiao Xie goto sleep; 1686d0278245SMiao Xie 1687dc7f370cSMiao Xie /* 1688dc7f370cSMiao Xie * Avoid the problem that we change the status of the fs 1689dc7f370cSMiao Xie * during the above check and trylock. 1690dc7f370cSMiao Xie */ 16912ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) { 16920b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1693dc7f370cSMiao Xie goto sleep; 1694dc7f370cSMiao Xie } 1695dc7f370cSMiao Xie 16962ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 1697c2d6cb16SFilipe Manana 16989d1a2a3aSDavid Sterba again = btrfs_clean_one_deleted_snapshot(root); 16990b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1700a74a4b97SChris Mason 1701d0278245SMiao Xie /* 170205323cd1SMiao Xie * The defragger has dealt with the R/O remount and umount, 170305323cd1SMiao Xie * needn't do anything special here. 1704d0278245SMiao Xie */ 17050b246afaSJeff Mahoney btrfs_run_defrag_inodes(fs_info); 170667c5e7d4SFilipe Manana 170767c5e7d4SFilipe Manana /* 170867c5e7d4SFilipe Manana * Acquires fs_info->delete_unused_bgs_mutex to avoid racing 170967c5e7d4SFilipe Manana * with relocation (btrfs_relocate_chunk) and relocation 171067c5e7d4SFilipe Manana * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 171167c5e7d4SFilipe Manana * after acquiring fs_info->delete_unused_bgs_mutex. So we 171267c5e7d4SFilipe Manana * can't hold, nor need to, fs_info->cleaner_mutex when deleting 171367c5e7d4SFilipe Manana * unused block groups. 171467c5e7d4SFilipe Manana */ 17150b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 1716d0278245SMiao Xie sleep: 1717fd340d0fSJosef Bacik clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1718d6fd0ae2SOmar Sandoval if (kthread_should_park()) 1719d6fd0ae2SOmar Sandoval kthread_parkme(); 1720d6fd0ae2SOmar Sandoval if (kthread_should_stop()) 1721d6fd0ae2SOmar Sandoval return 0; 1722838fe188SJiri Kosina if (!again) { 1723a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 1724a74a4b97SChris Mason schedule(); 1725a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1726a74a4b97SChris Mason } 1727da288d28SFilipe Manana } 1728a74a4b97SChris Mason } 1729a74a4b97SChris Mason 1730a74a4b97SChris Mason static int transaction_kthread(void *arg) 1731a74a4b97SChris Mason { 1732a74a4b97SChris Mason struct btrfs_root *root = arg; 17330b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1734a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1735a74a4b97SChris Mason struct btrfs_transaction *cur; 17368929ecfaSYan, Zheng u64 transid; 1737a944442cSAllen Pais time64_t now; 1738a74a4b97SChris Mason unsigned long delay; 1739914b2007SJan Kara bool cannot_commit; 1740a74a4b97SChris Mason 1741a74a4b97SChris Mason do { 1742914b2007SJan Kara cannot_commit = false; 17430b246afaSJeff Mahoney delay = HZ * fs_info->commit_interval; 17440b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 1745a74a4b97SChris Mason 17460b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 17470b246afaSJeff Mahoney cur = fs_info->running_transaction; 1748a74a4b97SChris Mason if (!cur) { 17490b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1750a74a4b97SChris Mason goto sleep; 1751a74a4b97SChris Mason } 175231153d81SYan Zheng 1753afd48513SArnd Bergmann now = ktime_get_seconds(); 17543296bf56SQu Wenruo if (cur->state < TRANS_STATE_COMMIT_START && 1755a514d638SQu Wenruo !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) && 17568b87dc17SDavid Sterba (now < cur->start_time || 17570b246afaSJeff Mahoney now - cur->start_time < fs_info->commit_interval)) { 17580b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1759a74a4b97SChris Mason delay = HZ * 5; 1760a74a4b97SChris Mason goto sleep; 1761a74a4b97SChris Mason } 17628929ecfaSYan, Zheng transid = cur->transid; 17630b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 176456bec294SChris Mason 176579787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 1766354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 1767914b2007SJan Kara if (IS_ERR(trans)) { 1768354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 1769914b2007SJan Kara cannot_commit = true; 177079787eaaSJeff Mahoney goto sleep; 1771914b2007SJan Kara } 17728929ecfaSYan, Zheng if (transid == trans->transid) { 17733a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 17748929ecfaSYan, Zheng } else { 17753a45bb20SJeff Mahoney btrfs_end_transaction(trans); 17768929ecfaSYan, Zheng } 1777a74a4b97SChris Mason sleep: 17780b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 17790b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 1780a74a4b97SChris Mason 17814e121c06SJosef Bacik if (unlikely(test_bit(BTRFS_FS_STATE_ERROR, 17820b246afaSJeff Mahoney &fs_info->fs_state))) 17832ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 17848929ecfaSYan, Zheng if (!kthread_should_stop() && 17850b246afaSJeff Mahoney (!btrfs_transaction_blocked(fs_info) || 1786914b2007SJan Kara cannot_commit)) 1787bc5511d0SNikolay Borisov schedule_timeout_interruptible(delay); 1788a74a4b97SChris Mason } while (!kthread_should_stop()); 1789a74a4b97SChris Mason return 0; 1790a74a4b97SChris Mason } 1791a74a4b97SChris Mason 1792af31f5e5SChris Mason /* 179301f0f9daSNikolay Borisov * This will find the highest generation in the array of root backups. The 179401f0f9daSNikolay Borisov * index of the highest array is returned, or -EINVAL if we can't find 179501f0f9daSNikolay Borisov * anything. 1796af31f5e5SChris Mason * 1797af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1798af31f5e5SChris Mason * generation of the latest root in the array with the generation 1799af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1800af31f5e5SChris Mason */ 180101f0f9daSNikolay Borisov static int find_newest_super_backup(struct btrfs_fs_info *info) 1802af31f5e5SChris Mason { 180301f0f9daSNikolay Borisov const u64 newest_gen = btrfs_super_generation(info->super_copy); 1804af31f5e5SChris Mason u64 cur; 1805af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1806af31f5e5SChris Mason int i; 1807af31f5e5SChris Mason 1808af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1809af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1810af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1811af31f5e5SChris Mason if (cur == newest_gen) 181201f0f9daSNikolay Borisov return i; 1813af31f5e5SChris Mason } 1814af31f5e5SChris Mason 181501f0f9daSNikolay Borisov return -EINVAL; 1816af31f5e5SChris Mason } 1817af31f5e5SChris Mason 1818af31f5e5SChris Mason 1819af31f5e5SChris Mason /* 1820af31f5e5SChris Mason * find the oldest backup so we know where to store new entries 1821af31f5e5SChris Mason * in the backup array. This will set the backup_root_index 1822af31f5e5SChris Mason * field in the fs_info struct 1823af31f5e5SChris Mason */ 1824fc2e4c5bSNikolay Borisov static void find_oldest_super_backup(struct btrfs_fs_info *info) 1825af31f5e5SChris Mason { 182601f0f9daSNikolay Borisov int newest_index; 1827af31f5e5SChris Mason 182801f0f9daSNikolay Borisov newest_index = find_newest_super_backup(info); 1829af31f5e5SChris Mason /* if there was garbage in there, just move along */ 183001f0f9daSNikolay Borisov if (newest_index == -EINVAL) { 1831af31f5e5SChris Mason info->backup_root_index = 0; 1832af31f5e5SChris Mason } else { 1833af31f5e5SChris Mason info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS; 1834af31f5e5SChris Mason } 1835af31f5e5SChris Mason } 1836af31f5e5SChris Mason 1837af31f5e5SChris Mason /* 1838af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1839af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1840af31f5e5SChris Mason * done 1841af31f5e5SChris Mason */ 1842af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1843af31f5e5SChris Mason { 1844af31f5e5SChris Mason int next_backup; 1845af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1846af31f5e5SChris Mason int last_backup; 1847af31f5e5SChris Mason 1848af31f5e5SChris Mason next_backup = info->backup_root_index; 1849af31f5e5SChris Mason last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) % 1850af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 1851af31f5e5SChris Mason 1852af31f5e5SChris Mason /* 1853af31f5e5SChris Mason * just overwrite the last backup if we're at the same generation 1854af31f5e5SChris Mason * this happens only at umount 1855af31f5e5SChris Mason */ 1856af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + last_backup; 1857af31f5e5SChris Mason if (btrfs_backup_tree_root_gen(root_backup) == 1858af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)) 1859af31f5e5SChris Mason next_backup = last_backup; 1860af31f5e5SChris Mason 1861af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1862af31f5e5SChris Mason 1863af31f5e5SChris Mason /* 1864af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1865af31f5e5SChris Mason * regardless of which ones we use today 1866af31f5e5SChris Mason */ 1867af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1868af31f5e5SChris Mason 1869af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1870af31f5e5SChris Mason 1871af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1872af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1873af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1874af31f5e5SChris Mason 1875af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1876af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1877af31f5e5SChris Mason 1878af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1879af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1880af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 1881af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 1882af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 1883af31f5e5SChris Mason 1884af31f5e5SChris Mason btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); 1885af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 1886af31f5e5SChris Mason btrfs_header_generation(info->extent_root->node)); 1887af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 1888af31f5e5SChris Mason btrfs_header_level(info->extent_root->node)); 1889af31f5e5SChris Mason 18907c7e82a7SChris Mason /* 18917c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 18927c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 18937c7e82a7SChris Mason */ 18947c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 18957c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 18967c7e82a7SChris Mason info->fs_root->node->start); 1897af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 1898af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 1899af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 1900af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 19017c7e82a7SChris Mason } 1902af31f5e5SChris Mason 1903af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 1904af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 1905af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 1906af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 1907af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 1908af31f5e5SChris Mason 1909af31f5e5SChris Mason btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); 1910af31f5e5SChris Mason btrfs_set_backup_csum_root_gen(root_backup, 1911af31f5e5SChris Mason btrfs_header_generation(info->csum_root->node)); 1912af31f5e5SChris Mason btrfs_set_backup_csum_root_level(root_backup, 1913af31f5e5SChris Mason btrfs_header_level(info->csum_root->node)); 1914af31f5e5SChris Mason 1915af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 1916af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 1917af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 1918af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 1919af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 1920af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 1921af31f5e5SChris Mason 1922af31f5e5SChris Mason /* 1923af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 1924af31f5e5SChris Mason * for the next commit 1925af31f5e5SChris Mason */ 1926af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 1927af31f5e5SChris Mason &info->super_for_commit->super_roots, 1928af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 1929af31f5e5SChris Mason } 1930af31f5e5SChris Mason 1931af31f5e5SChris Mason /* 1932bd2336b2SNikolay Borisov * read_backup_root - Reads a backup root based on the passed priority. Prio 0 1933bd2336b2SNikolay Borisov * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots 1934bd2336b2SNikolay Borisov * 1935bd2336b2SNikolay Borisov * fs_info - filesystem whose backup roots need to be read 1936bd2336b2SNikolay Borisov * priority - priority of backup root required 1937bd2336b2SNikolay Borisov * 1938bd2336b2SNikolay Borisov * Returns backup root index on success and -EINVAL otherwise. 1939bd2336b2SNikolay Borisov */ 1940bd2336b2SNikolay Borisov static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority) 1941bd2336b2SNikolay Borisov { 1942bd2336b2SNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 1943bd2336b2SNikolay Borisov struct btrfs_super_block *super = fs_info->super_copy; 1944bd2336b2SNikolay Borisov struct btrfs_root_backup *root_backup; 1945bd2336b2SNikolay Borisov 1946bd2336b2SNikolay Borisov if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) { 1947bd2336b2SNikolay Borisov if (priority == 0) 1948bd2336b2SNikolay Borisov return backup_index; 1949bd2336b2SNikolay Borisov 1950bd2336b2SNikolay Borisov backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority; 1951bd2336b2SNikolay Borisov backup_index %= BTRFS_NUM_BACKUP_ROOTS; 1952bd2336b2SNikolay Borisov } else { 1953bd2336b2SNikolay Borisov return -EINVAL; 1954bd2336b2SNikolay Borisov } 1955bd2336b2SNikolay Borisov 1956bd2336b2SNikolay Borisov root_backup = super->super_roots + backup_index; 1957bd2336b2SNikolay Borisov 1958bd2336b2SNikolay Borisov btrfs_set_super_generation(super, 1959bd2336b2SNikolay Borisov btrfs_backup_tree_root_gen(root_backup)); 1960bd2336b2SNikolay Borisov btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 1961bd2336b2SNikolay Borisov btrfs_set_super_root_level(super, 1962bd2336b2SNikolay Borisov btrfs_backup_tree_root_level(root_backup)); 1963bd2336b2SNikolay Borisov btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 1964bd2336b2SNikolay Borisov 1965bd2336b2SNikolay Borisov /* 1966bd2336b2SNikolay Borisov * Fixme: the total bytes and num_devices need to match or we should 1967bd2336b2SNikolay Borisov * need a fsck 1968bd2336b2SNikolay Borisov */ 1969bd2336b2SNikolay Borisov btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 1970bd2336b2SNikolay Borisov btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 1971bd2336b2SNikolay Borisov 1972bd2336b2SNikolay Borisov return backup_index; 1973bd2336b2SNikolay Borisov } 1974bd2336b2SNikolay Borisov 1975bd2336b2SNikolay Borisov /* 1976af31f5e5SChris Mason * this copies info out of the root backup array and back into 1977af31f5e5SChris Mason * the in-memory super block. It is meant to help iterate through 1978af31f5e5SChris Mason * the array, so you send it the number of backups you've already 1979af31f5e5SChris Mason * tried and the last backup index you used. 1980af31f5e5SChris Mason * 1981af31f5e5SChris Mason * this returns -1 when it has tried all the backups 1982af31f5e5SChris Mason */ 1983af31f5e5SChris Mason static noinline int next_root_backup(struct btrfs_fs_info *info, 1984af31f5e5SChris Mason struct btrfs_super_block *super, 1985af31f5e5SChris Mason int *num_backups_tried, int *backup_index) 1986af31f5e5SChris Mason { 1987af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1988af31f5e5SChris Mason int newest = *backup_index; 1989af31f5e5SChris Mason 1990af31f5e5SChris Mason if (*num_backups_tried == 0) { 199101f0f9daSNikolay Borisov newest = find_newest_super_backup(info); 1992af31f5e5SChris Mason if (newest == -1) 1993af31f5e5SChris Mason return -1; 1994af31f5e5SChris Mason 1995af31f5e5SChris Mason *backup_index = newest; 1996af31f5e5SChris Mason *num_backups_tried = 1; 1997af31f5e5SChris Mason } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) { 1998af31f5e5SChris Mason /* we've tried all the backups, all done */ 1999af31f5e5SChris Mason return -1; 2000af31f5e5SChris Mason } else { 2001af31f5e5SChris Mason /* jump to the next oldest backup */ 2002af31f5e5SChris Mason newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) % 2003af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 2004af31f5e5SChris Mason *backup_index = newest; 2005af31f5e5SChris Mason *num_backups_tried += 1; 2006af31f5e5SChris Mason } 2007af31f5e5SChris Mason root_backup = super->super_roots + newest; 2008af31f5e5SChris Mason 2009af31f5e5SChris Mason btrfs_set_super_generation(super, 2010af31f5e5SChris Mason btrfs_backup_tree_root_gen(root_backup)); 2011af31f5e5SChris Mason btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 2012af31f5e5SChris Mason btrfs_set_super_root_level(super, 2013af31f5e5SChris Mason btrfs_backup_tree_root_level(root_backup)); 2014af31f5e5SChris Mason btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 2015af31f5e5SChris Mason 2016af31f5e5SChris Mason /* 2017af31f5e5SChris Mason * fixme: the total bytes and num_devices need to match or we should 2018af31f5e5SChris Mason * need a fsck 2019af31f5e5SChris Mason */ 2020af31f5e5SChris Mason btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 2021af31f5e5SChris Mason btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 2022af31f5e5SChris Mason return 0; 2023af31f5e5SChris Mason } 2024af31f5e5SChris Mason 20257abadb64SLiu Bo /* helper to cleanup workers */ 20267abadb64SLiu Bo static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 20277abadb64SLiu Bo { 2028dc6e3209SQu Wenruo btrfs_destroy_workqueue(fs_info->fixup_workers); 2029afe3d242SQu Wenruo btrfs_destroy_workqueue(fs_info->delalloc_workers); 20305cdc7ad3SQu Wenruo btrfs_destroy_workqueue(fs_info->workers); 2031fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_workers); 2032fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_raid56_workers); 20338b110e39SMiao Xie btrfs_destroy_workqueue(fs_info->endio_repair_workers); 2034d05a33acSQu Wenruo btrfs_destroy_workqueue(fs_info->rmw_workers); 2035fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_write_workers); 2036fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 20375b3bc44eSQu Wenruo btrfs_destroy_workqueue(fs_info->delayed_workers); 2038e66f0bb1SQu Wenruo btrfs_destroy_workqueue(fs_info->caching_workers); 2039736cfa15SQu Wenruo btrfs_destroy_workqueue(fs_info->readahead_workers); 2040a44903abSQu Wenruo btrfs_destroy_workqueue(fs_info->flush_workers); 2041fc97fab0SQu Wenruo btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2042a9b9477dSFilipe Manana /* 2043a9b9477dSFilipe Manana * Now that all other work queues are destroyed, we can safely destroy 2044a9b9477dSFilipe Manana * the queues used for metadata I/O, since tasks from those other work 2045a9b9477dSFilipe Manana * queues can do metadata I/O operations. 2046a9b9477dSFilipe Manana */ 2047a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_workers); 2048a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_write_workers); 20497abadb64SLiu Bo } 20507abadb64SLiu Bo 20512e9f5954SRashika static void free_root_extent_buffers(struct btrfs_root *root) 20522e9f5954SRashika { 20532e9f5954SRashika if (root) { 20542e9f5954SRashika free_extent_buffer(root->node); 20552e9f5954SRashika free_extent_buffer(root->commit_root); 20562e9f5954SRashika root->node = NULL; 20572e9f5954SRashika root->commit_root = NULL; 20582e9f5954SRashika } 20592e9f5954SRashika } 20602e9f5954SRashika 2061af31f5e5SChris Mason /* helper to cleanup tree roots */ 20624273eaffSAnand Jain static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root) 2063af31f5e5SChris Mason { 20642e9f5954SRashika free_root_extent_buffers(info->tree_root); 2065655b09feSJosef Bacik 20662e9f5954SRashika free_root_extent_buffers(info->dev_root); 20672e9f5954SRashika free_root_extent_buffers(info->extent_root); 20682e9f5954SRashika free_root_extent_buffers(info->csum_root); 20692e9f5954SRashika free_root_extent_buffers(info->quota_root); 20702e9f5954SRashika free_root_extent_buffers(info->uuid_root); 20714273eaffSAnand Jain if (free_chunk_root) 20722e9f5954SRashika free_root_extent_buffers(info->chunk_root); 207370f6d82eSOmar Sandoval free_root_extent_buffers(info->free_space_root); 2074af31f5e5SChris Mason } 2075af31f5e5SChris Mason 2076faa2dbf0SJosef Bacik void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2077171f6537SJosef Bacik { 2078171f6537SJosef Bacik int ret; 2079171f6537SJosef Bacik struct btrfs_root *gang[8]; 2080171f6537SJosef Bacik int i; 2081171f6537SJosef Bacik 2082171f6537SJosef Bacik while (!list_empty(&fs_info->dead_roots)) { 2083171f6537SJosef Bacik gang[0] = list_entry(fs_info->dead_roots.next, 2084171f6537SJosef Bacik struct btrfs_root, root_list); 2085171f6537SJosef Bacik list_del(&gang[0]->root_list); 2086171f6537SJosef Bacik 208727cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) { 2088cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[0]); 2089171f6537SJosef Bacik } else { 2090171f6537SJosef Bacik free_extent_buffer(gang[0]->node); 2091171f6537SJosef Bacik free_extent_buffer(gang[0]->commit_root); 2092b0feb9d9SMiao Xie btrfs_put_fs_root(gang[0]); 2093171f6537SJosef Bacik } 2094171f6537SJosef Bacik } 2095171f6537SJosef Bacik 2096171f6537SJosef Bacik while (1) { 2097171f6537SJosef Bacik ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2098171f6537SJosef Bacik (void **)gang, 0, 2099171f6537SJosef Bacik ARRAY_SIZE(gang)); 2100171f6537SJosef Bacik if (!ret) 2101171f6537SJosef Bacik break; 2102171f6537SJosef Bacik for (i = 0; i < ret; i++) 2103cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2104171f6537SJosef Bacik } 21051a4319ccSLiu Bo 21061a4319ccSLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 21071a4319ccSLiu Bo btrfs_free_log_root_tree(NULL, fs_info); 21082ff7e61eSJeff Mahoney btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents); 21091a4319ccSLiu Bo } 2110171f6537SJosef Bacik } 2111af31f5e5SChris Mason 2112638aa7edSEric Sandeen static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2113638aa7edSEric Sandeen { 2114638aa7edSEric Sandeen mutex_init(&fs_info->scrub_lock); 2115638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_running, 0); 2116638aa7edSEric Sandeen atomic_set(&fs_info->scrub_pause_req, 0); 2117638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_paused, 0); 2118638aa7edSEric Sandeen atomic_set(&fs_info->scrub_cancel_req, 0); 2119638aa7edSEric Sandeen init_waitqueue_head(&fs_info->scrub_pause_wait); 2120ff09c4caSAnand Jain refcount_set(&fs_info->scrub_workers_refcnt, 0); 2121638aa7edSEric Sandeen } 2122638aa7edSEric Sandeen 2123779a65a4SEric Sandeen static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2124779a65a4SEric Sandeen { 2125779a65a4SEric Sandeen spin_lock_init(&fs_info->balance_lock); 2126779a65a4SEric Sandeen mutex_init(&fs_info->balance_mutex); 2127779a65a4SEric Sandeen atomic_set(&fs_info->balance_pause_req, 0); 2128779a65a4SEric Sandeen atomic_set(&fs_info->balance_cancel_req, 0); 2129779a65a4SEric Sandeen fs_info->balance_ctl = NULL; 2130779a65a4SEric Sandeen init_waitqueue_head(&fs_info->balance_wait_q); 2131779a65a4SEric Sandeen } 2132779a65a4SEric Sandeen 21336bccf3abSJeff Mahoney static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info) 2134f37938e0SEric Sandeen { 21352ff7e61eSJeff Mahoney struct inode *inode = fs_info->btree_inode; 21362ff7e61eSJeff Mahoney 21372ff7e61eSJeff Mahoney inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 21382ff7e61eSJeff Mahoney set_nlink(inode, 1); 2139f37938e0SEric Sandeen /* 2140f37938e0SEric Sandeen * we set the i_size on the btree inode to the max possible int. 2141f37938e0SEric Sandeen * the real end of the address space is determined by all of 2142f37938e0SEric Sandeen * the devices in the system 2143f37938e0SEric Sandeen */ 21442ff7e61eSJeff Mahoney inode->i_size = OFFSET_MAX; 21452ff7e61eSJeff Mahoney inode->i_mapping->a_ops = &btree_aops; 2146f37938e0SEric Sandeen 21472ff7e61eSJeff Mahoney RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 214843eb5f29SQu Wenruo extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree, 214943eb5f29SQu Wenruo IO_TREE_INODE_IO, inode); 21507b439738SDavid Sterba BTRFS_I(inode)->io_tree.track_uptodate = false; 21512ff7e61eSJeff Mahoney extent_map_tree_init(&BTRFS_I(inode)->extent_tree); 2152f37938e0SEric Sandeen 21532ff7e61eSJeff Mahoney BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops; 2154f37938e0SEric Sandeen 21552ff7e61eSJeff Mahoney BTRFS_I(inode)->root = fs_info->tree_root; 21562ff7e61eSJeff Mahoney memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key)); 21572ff7e61eSJeff Mahoney set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); 21582ff7e61eSJeff Mahoney btrfs_insert_inode_hash(inode); 2159f37938e0SEric Sandeen } 2160f37938e0SEric Sandeen 2161ad618368SEric Sandeen static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2162ad618368SEric Sandeen { 2163ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2164129827e3SDavid Sterba init_rwsem(&fs_info->dev_replace.rwsem); 21657f8d236aSDavid Sterba init_waitqueue_head(&fs_info->dev_replace.replace_wait); 2166ad618368SEric Sandeen } 2167ad618368SEric Sandeen 2168f9e92e40SEric Sandeen static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2169f9e92e40SEric Sandeen { 2170f9e92e40SEric Sandeen spin_lock_init(&fs_info->qgroup_lock); 2171f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_ioctl_lock); 2172f9e92e40SEric Sandeen fs_info->qgroup_tree = RB_ROOT; 2173f9e92e40SEric Sandeen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2174f9e92e40SEric Sandeen fs_info->qgroup_seq = 1; 2175f9e92e40SEric Sandeen fs_info->qgroup_ulist = NULL; 2176d2c609b8SJeff Mahoney fs_info->qgroup_rescan_running = false; 2177f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_rescan_lock); 2178f9e92e40SEric Sandeen } 2179f9e92e40SEric Sandeen 21802a458198SEric Sandeen static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info, 21812a458198SEric Sandeen struct btrfs_fs_devices *fs_devices) 21822a458198SEric Sandeen { 2183f7b885beSAnand Jain u32 max_active = fs_info->thread_pool_size; 21846f011058SDavid Sterba unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 21852a458198SEric Sandeen 21862a458198SEric Sandeen fs_info->workers = 2187cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "worker", 2188cb001095SJeff Mahoney flags | WQ_HIGHPRI, max_active, 16); 21892a458198SEric Sandeen 21902a458198SEric Sandeen fs_info->delalloc_workers = 2191cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delalloc", 2192cb001095SJeff Mahoney flags, max_active, 2); 21932a458198SEric Sandeen 21942a458198SEric Sandeen fs_info->flush_workers = 2195cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "flush_delalloc", 2196cb001095SJeff Mahoney flags, max_active, 0); 21972a458198SEric Sandeen 21982a458198SEric Sandeen fs_info->caching_workers = 2199cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); 22002a458198SEric Sandeen 22012a458198SEric Sandeen fs_info->fixup_workers = 2202cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); 22032a458198SEric Sandeen 22042a458198SEric Sandeen /* 22052a458198SEric Sandeen * endios are largely parallel and should have a very 22062a458198SEric Sandeen * low idle thresh 22072a458198SEric Sandeen */ 22082a458198SEric Sandeen fs_info->endio_workers = 2209cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4); 22102a458198SEric Sandeen fs_info->endio_meta_workers = 2211cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta", flags, 2212cb001095SJeff Mahoney max_active, 4); 22132a458198SEric Sandeen fs_info->endio_meta_write_workers = 2214cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags, 2215cb001095SJeff Mahoney max_active, 2); 22162a458198SEric Sandeen fs_info->endio_raid56_workers = 2217cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-raid56", flags, 2218cb001095SJeff Mahoney max_active, 4); 22192a458198SEric Sandeen fs_info->endio_repair_workers = 2220cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0); 22212a458198SEric Sandeen fs_info->rmw_workers = 2222cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2); 22232a458198SEric Sandeen fs_info->endio_write_workers = 2224cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-write", flags, 2225cb001095SJeff Mahoney max_active, 2); 22262a458198SEric Sandeen fs_info->endio_freespace_worker = 2227cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "freespace-write", flags, 2228cb001095SJeff Mahoney max_active, 0); 22292a458198SEric Sandeen fs_info->delayed_workers = 2230cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, 2231cb001095SJeff Mahoney max_active, 0); 22322a458198SEric Sandeen fs_info->readahead_workers = 2233cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "readahead", flags, 2234cb001095SJeff Mahoney max_active, 2); 22352a458198SEric Sandeen fs_info->qgroup_rescan_workers = 2236cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); 22372a458198SEric Sandeen 22382a458198SEric Sandeen if (!(fs_info->workers && fs_info->delalloc_workers && 2239ba8a9d07SChris Mason fs_info->flush_workers && 22402a458198SEric Sandeen fs_info->endio_workers && fs_info->endio_meta_workers && 22412a458198SEric Sandeen fs_info->endio_meta_write_workers && 22422a458198SEric Sandeen fs_info->endio_repair_workers && 22432a458198SEric Sandeen fs_info->endio_write_workers && fs_info->endio_raid56_workers && 22442a458198SEric Sandeen fs_info->endio_freespace_worker && fs_info->rmw_workers && 22452a458198SEric Sandeen fs_info->caching_workers && fs_info->readahead_workers && 22462a458198SEric Sandeen fs_info->fixup_workers && fs_info->delayed_workers && 22472a458198SEric Sandeen fs_info->qgroup_rescan_workers)) { 22482a458198SEric Sandeen return -ENOMEM; 22492a458198SEric Sandeen } 22502a458198SEric Sandeen 22512a458198SEric Sandeen return 0; 22522a458198SEric Sandeen } 22532a458198SEric Sandeen 22546d97c6e3SJohannes Thumshirn static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type) 22556d97c6e3SJohannes Thumshirn { 22566d97c6e3SJohannes Thumshirn struct crypto_shash *csum_shash; 2257b4e967beSDavid Sterba const char *csum_driver = btrfs_super_csum_driver(csum_type); 22586d97c6e3SJohannes Thumshirn 2259b4e967beSDavid Sterba csum_shash = crypto_alloc_shash(csum_driver, 0, 0); 22606d97c6e3SJohannes Thumshirn 22616d97c6e3SJohannes Thumshirn if (IS_ERR(csum_shash)) { 22626d97c6e3SJohannes Thumshirn btrfs_err(fs_info, "error allocating %s hash for checksum", 2263b4e967beSDavid Sterba csum_driver); 22646d97c6e3SJohannes Thumshirn return PTR_ERR(csum_shash); 22656d97c6e3SJohannes Thumshirn } 22666d97c6e3SJohannes Thumshirn 22676d97c6e3SJohannes Thumshirn fs_info->csum_shash = csum_shash; 22686d97c6e3SJohannes Thumshirn 22696d97c6e3SJohannes Thumshirn return 0; 22706d97c6e3SJohannes Thumshirn } 22716d97c6e3SJohannes Thumshirn 22726d97c6e3SJohannes Thumshirn static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info) 22736d97c6e3SJohannes Thumshirn { 22746d97c6e3SJohannes Thumshirn crypto_free_shash(fs_info->csum_shash); 22756d97c6e3SJohannes Thumshirn } 22766d97c6e3SJohannes Thumshirn 227763443bf5SEric Sandeen static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 227863443bf5SEric Sandeen struct btrfs_fs_devices *fs_devices) 227963443bf5SEric Sandeen { 228063443bf5SEric Sandeen int ret; 228163443bf5SEric Sandeen struct btrfs_root *log_tree_root; 228263443bf5SEric Sandeen struct btrfs_super_block *disk_super = fs_info->super_copy; 228363443bf5SEric Sandeen u64 bytenr = btrfs_super_log_root(disk_super); 2284581c1760SQu Wenruo int level = btrfs_super_log_root_level(disk_super); 228563443bf5SEric Sandeen 228663443bf5SEric Sandeen if (fs_devices->rw_devices == 0) { 2287f14d104dSDavid Sterba btrfs_warn(fs_info, "log replay required on RO media"); 228863443bf5SEric Sandeen return -EIO; 228963443bf5SEric Sandeen } 229063443bf5SEric Sandeen 229174e4d827SDavid Sterba log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL); 229263443bf5SEric Sandeen if (!log_tree_root) 229363443bf5SEric Sandeen return -ENOMEM; 229463443bf5SEric Sandeen 2295da17066cSJeff Mahoney __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID); 229663443bf5SEric Sandeen 22972ff7e61eSJeff Mahoney log_tree_root->node = read_tree_block(fs_info, bytenr, 2298581c1760SQu Wenruo fs_info->generation + 1, 2299581c1760SQu Wenruo level, NULL); 230064c043deSLiu Bo if (IS_ERR(log_tree_root->node)) { 2301f14d104dSDavid Sterba btrfs_warn(fs_info, "failed to read log tree"); 23020eeff236SLiu Bo ret = PTR_ERR(log_tree_root->node); 230364c043deSLiu Bo kfree(log_tree_root); 23040eeff236SLiu Bo return ret; 230564c043deSLiu Bo } else if (!extent_buffer_uptodate(log_tree_root->node)) { 2306f14d104dSDavid Sterba btrfs_err(fs_info, "failed to read log tree"); 230763443bf5SEric Sandeen free_extent_buffer(log_tree_root->node); 230863443bf5SEric Sandeen kfree(log_tree_root); 230963443bf5SEric Sandeen return -EIO; 231063443bf5SEric Sandeen } 231163443bf5SEric Sandeen /* returns with log_tree_root freed on success */ 231263443bf5SEric Sandeen ret = btrfs_recover_log_trees(log_tree_root); 231363443bf5SEric Sandeen if (ret) { 23140b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 231563443bf5SEric Sandeen "Failed to recover log tree"); 231663443bf5SEric Sandeen free_extent_buffer(log_tree_root->node); 231763443bf5SEric Sandeen kfree(log_tree_root); 231863443bf5SEric Sandeen return ret; 231963443bf5SEric Sandeen } 232063443bf5SEric Sandeen 2321bc98a42cSDavid Howells if (sb_rdonly(fs_info->sb)) { 23226bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 232363443bf5SEric Sandeen if (ret) 232463443bf5SEric Sandeen return ret; 232563443bf5SEric Sandeen } 232663443bf5SEric Sandeen 232763443bf5SEric Sandeen return 0; 232863443bf5SEric Sandeen } 232963443bf5SEric Sandeen 23306bccf3abSJeff Mahoney static int btrfs_read_roots(struct btrfs_fs_info *fs_info) 23314bbcaa64SEric Sandeen { 23326bccf3abSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 2333a4f3d2c4SDavid Sterba struct btrfs_root *root; 23344bbcaa64SEric Sandeen struct btrfs_key location; 23354bbcaa64SEric Sandeen int ret; 23364bbcaa64SEric Sandeen 23376bccf3abSJeff Mahoney BUG_ON(!fs_info->tree_root); 23386bccf3abSJeff Mahoney 23394bbcaa64SEric Sandeen location.objectid = BTRFS_EXTENT_TREE_OBJECTID; 23404bbcaa64SEric Sandeen location.type = BTRFS_ROOT_ITEM_KEY; 23414bbcaa64SEric Sandeen location.offset = 0; 23424bbcaa64SEric Sandeen 2343a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2344f50f4353SLiu Bo if (IS_ERR(root)) { 2345f50f4353SLiu Bo ret = PTR_ERR(root); 2346f50f4353SLiu Bo goto out; 2347f50f4353SLiu Bo } 2348a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2349a4f3d2c4SDavid Sterba fs_info->extent_root = root; 23504bbcaa64SEric Sandeen 23514bbcaa64SEric Sandeen location.objectid = BTRFS_DEV_TREE_OBJECTID; 2352a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2353f50f4353SLiu Bo if (IS_ERR(root)) { 2354f50f4353SLiu Bo ret = PTR_ERR(root); 2355f50f4353SLiu Bo goto out; 2356f50f4353SLiu Bo } 2357a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2358a4f3d2c4SDavid Sterba fs_info->dev_root = root; 23594bbcaa64SEric Sandeen btrfs_init_devices_late(fs_info); 23604bbcaa64SEric Sandeen 23614bbcaa64SEric Sandeen location.objectid = BTRFS_CSUM_TREE_OBJECTID; 2362a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2363f50f4353SLiu Bo if (IS_ERR(root)) { 2364f50f4353SLiu Bo ret = PTR_ERR(root); 2365f50f4353SLiu Bo goto out; 2366f50f4353SLiu Bo } 2367a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2368a4f3d2c4SDavid Sterba fs_info->csum_root = root; 23694bbcaa64SEric Sandeen 23704bbcaa64SEric Sandeen location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2371a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2372a4f3d2c4SDavid Sterba if (!IS_ERR(root)) { 2373a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2374afcdd129SJosef Bacik set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); 2375a4f3d2c4SDavid Sterba fs_info->quota_root = root; 23764bbcaa64SEric Sandeen } 23774bbcaa64SEric Sandeen 23784bbcaa64SEric Sandeen location.objectid = BTRFS_UUID_TREE_OBJECTID; 2379a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2380a4f3d2c4SDavid Sterba if (IS_ERR(root)) { 2381a4f3d2c4SDavid Sterba ret = PTR_ERR(root); 23824bbcaa64SEric Sandeen if (ret != -ENOENT) 2383f50f4353SLiu Bo goto out; 23844bbcaa64SEric Sandeen } else { 2385a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2386a4f3d2c4SDavid Sterba fs_info->uuid_root = root; 23874bbcaa64SEric Sandeen } 23884bbcaa64SEric Sandeen 238970f6d82eSOmar Sandoval if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 239070f6d82eSOmar Sandoval location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID; 239170f6d82eSOmar Sandoval root = btrfs_read_tree_root(tree_root, &location); 2392f50f4353SLiu Bo if (IS_ERR(root)) { 2393f50f4353SLiu Bo ret = PTR_ERR(root); 2394f50f4353SLiu Bo goto out; 2395f50f4353SLiu Bo } 239670f6d82eSOmar Sandoval set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 239770f6d82eSOmar Sandoval fs_info->free_space_root = root; 239870f6d82eSOmar Sandoval } 239970f6d82eSOmar Sandoval 24004bbcaa64SEric Sandeen return 0; 2401f50f4353SLiu Bo out: 2402f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", 2403f50f4353SLiu Bo location.objectid, ret); 2404f50f4353SLiu Bo return ret; 24054bbcaa64SEric Sandeen } 24064bbcaa64SEric Sandeen 2407069ec957SQu Wenruo /* 2408069ec957SQu Wenruo * Real super block validation 2409069ec957SQu Wenruo * NOTE: super csum type and incompat features will not be checked here. 2410069ec957SQu Wenruo * 2411069ec957SQu Wenruo * @sb: super block to check 2412069ec957SQu Wenruo * @mirror_num: the super block number to check its bytenr: 2413069ec957SQu Wenruo * 0 the primary (1st) sb 2414069ec957SQu Wenruo * 1, 2 2nd and 3rd backup copy 2415069ec957SQu Wenruo * -1 skip bytenr check 2416069ec957SQu Wenruo */ 2417069ec957SQu Wenruo static int validate_super(struct btrfs_fs_info *fs_info, 2418069ec957SQu Wenruo struct btrfs_super_block *sb, int mirror_num) 241921a852b0SQu Wenruo { 242021a852b0SQu Wenruo u64 nodesize = btrfs_super_nodesize(sb); 242121a852b0SQu Wenruo u64 sectorsize = btrfs_super_sectorsize(sb); 242221a852b0SQu Wenruo int ret = 0; 242321a852b0SQu Wenruo 242421a852b0SQu Wenruo if (btrfs_super_magic(sb) != BTRFS_MAGIC) { 242521a852b0SQu Wenruo btrfs_err(fs_info, "no valid FS found"); 242621a852b0SQu Wenruo ret = -EINVAL; 242721a852b0SQu Wenruo } 242821a852b0SQu Wenruo if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { 242921a852b0SQu Wenruo btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", 243021a852b0SQu Wenruo btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); 243121a852b0SQu Wenruo ret = -EINVAL; 243221a852b0SQu Wenruo } 243321a852b0SQu Wenruo if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 243421a852b0SQu Wenruo btrfs_err(fs_info, "tree_root level too big: %d >= %d", 243521a852b0SQu Wenruo btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 243621a852b0SQu Wenruo ret = -EINVAL; 243721a852b0SQu Wenruo } 243821a852b0SQu Wenruo if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 243921a852b0SQu Wenruo btrfs_err(fs_info, "chunk_root level too big: %d >= %d", 244021a852b0SQu Wenruo btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 244121a852b0SQu Wenruo ret = -EINVAL; 244221a852b0SQu Wenruo } 244321a852b0SQu Wenruo if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 244421a852b0SQu Wenruo btrfs_err(fs_info, "log_root level too big: %d >= %d", 244521a852b0SQu Wenruo btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 244621a852b0SQu Wenruo ret = -EINVAL; 244721a852b0SQu Wenruo } 244821a852b0SQu Wenruo 244921a852b0SQu Wenruo /* 245021a852b0SQu Wenruo * Check sectorsize and nodesize first, other check will need it. 245121a852b0SQu Wenruo * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. 245221a852b0SQu Wenruo */ 245321a852b0SQu Wenruo if (!is_power_of_2(sectorsize) || sectorsize < 4096 || 245421a852b0SQu Wenruo sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { 245521a852b0SQu Wenruo btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); 245621a852b0SQu Wenruo ret = -EINVAL; 245721a852b0SQu Wenruo } 245821a852b0SQu Wenruo /* Only PAGE SIZE is supported yet */ 245921a852b0SQu Wenruo if (sectorsize != PAGE_SIZE) { 246021a852b0SQu Wenruo btrfs_err(fs_info, 246121a852b0SQu Wenruo "sectorsize %llu not supported yet, only support %lu", 246221a852b0SQu Wenruo sectorsize, PAGE_SIZE); 246321a852b0SQu Wenruo ret = -EINVAL; 246421a852b0SQu Wenruo } 246521a852b0SQu Wenruo if (!is_power_of_2(nodesize) || nodesize < sectorsize || 246621a852b0SQu Wenruo nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { 246721a852b0SQu Wenruo btrfs_err(fs_info, "invalid nodesize %llu", nodesize); 246821a852b0SQu Wenruo ret = -EINVAL; 246921a852b0SQu Wenruo } 247021a852b0SQu Wenruo if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { 247121a852b0SQu Wenruo btrfs_err(fs_info, "invalid leafsize %u, should be %llu", 247221a852b0SQu Wenruo le32_to_cpu(sb->__unused_leafsize), nodesize); 247321a852b0SQu Wenruo ret = -EINVAL; 247421a852b0SQu Wenruo } 247521a852b0SQu Wenruo 247621a852b0SQu Wenruo /* Root alignment check */ 247721a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { 247821a852b0SQu Wenruo btrfs_warn(fs_info, "tree_root block unaligned: %llu", 247921a852b0SQu Wenruo btrfs_super_root(sb)); 248021a852b0SQu Wenruo ret = -EINVAL; 248121a852b0SQu Wenruo } 248221a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { 248321a852b0SQu Wenruo btrfs_warn(fs_info, "chunk_root block unaligned: %llu", 248421a852b0SQu Wenruo btrfs_super_chunk_root(sb)); 248521a852b0SQu Wenruo ret = -EINVAL; 248621a852b0SQu Wenruo } 248721a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { 248821a852b0SQu Wenruo btrfs_warn(fs_info, "log_root block unaligned: %llu", 248921a852b0SQu Wenruo btrfs_super_log_root(sb)); 249021a852b0SQu Wenruo ret = -EINVAL; 249121a852b0SQu Wenruo } 249221a852b0SQu Wenruo 2493de37aa51SNikolay Borisov if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, 24947239ff4bSNikolay Borisov BTRFS_FSID_SIZE) != 0) { 249521a852b0SQu Wenruo btrfs_err(fs_info, 24967239ff4bSNikolay Borisov "dev_item UUID does not match metadata fsid: %pU != %pU", 2497de37aa51SNikolay Borisov fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); 249821a852b0SQu Wenruo ret = -EINVAL; 249921a852b0SQu Wenruo } 250021a852b0SQu Wenruo 250121a852b0SQu Wenruo /* 250221a852b0SQu Wenruo * Hint to catch really bogus numbers, bitflips or so, more exact checks are 250321a852b0SQu Wenruo * done later 250421a852b0SQu Wenruo */ 250521a852b0SQu Wenruo if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { 250621a852b0SQu Wenruo btrfs_err(fs_info, "bytes_used is too small %llu", 250721a852b0SQu Wenruo btrfs_super_bytes_used(sb)); 250821a852b0SQu Wenruo ret = -EINVAL; 250921a852b0SQu Wenruo } 251021a852b0SQu Wenruo if (!is_power_of_2(btrfs_super_stripesize(sb))) { 251121a852b0SQu Wenruo btrfs_err(fs_info, "invalid stripesize %u", 251221a852b0SQu Wenruo btrfs_super_stripesize(sb)); 251321a852b0SQu Wenruo ret = -EINVAL; 251421a852b0SQu Wenruo } 251521a852b0SQu Wenruo if (btrfs_super_num_devices(sb) > (1UL << 31)) 251621a852b0SQu Wenruo btrfs_warn(fs_info, "suspicious number of devices: %llu", 251721a852b0SQu Wenruo btrfs_super_num_devices(sb)); 251821a852b0SQu Wenruo if (btrfs_super_num_devices(sb) == 0) { 251921a852b0SQu Wenruo btrfs_err(fs_info, "number of devices is 0"); 252021a852b0SQu Wenruo ret = -EINVAL; 252121a852b0SQu Wenruo } 252221a852b0SQu Wenruo 2523069ec957SQu Wenruo if (mirror_num >= 0 && 2524069ec957SQu Wenruo btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { 252521a852b0SQu Wenruo btrfs_err(fs_info, "super offset mismatch %llu != %u", 252621a852b0SQu Wenruo btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 252721a852b0SQu Wenruo ret = -EINVAL; 252821a852b0SQu Wenruo } 252921a852b0SQu Wenruo 253021a852b0SQu Wenruo /* 253121a852b0SQu Wenruo * Obvious sys_chunk_array corruptions, it must hold at least one key 253221a852b0SQu Wenruo * and one chunk 253321a852b0SQu Wenruo */ 253421a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 253521a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too big %u > %u", 253621a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 253721a852b0SQu Wenruo BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 253821a852b0SQu Wenruo ret = -EINVAL; 253921a852b0SQu Wenruo } 254021a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 254121a852b0SQu Wenruo + sizeof(struct btrfs_chunk)) { 254221a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too small %u < %zu", 254321a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 254421a852b0SQu Wenruo sizeof(struct btrfs_disk_key) 254521a852b0SQu Wenruo + sizeof(struct btrfs_chunk)); 254621a852b0SQu Wenruo ret = -EINVAL; 254721a852b0SQu Wenruo } 254821a852b0SQu Wenruo 254921a852b0SQu Wenruo /* 255021a852b0SQu Wenruo * The generation is a global counter, we'll trust it more than the others 255121a852b0SQu Wenruo * but it's still possible that it's the one that's wrong. 255221a852b0SQu Wenruo */ 255321a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 255421a852b0SQu Wenruo btrfs_warn(fs_info, 255521a852b0SQu Wenruo "suspicious: generation < chunk_root_generation: %llu < %llu", 255621a852b0SQu Wenruo btrfs_super_generation(sb), 255721a852b0SQu Wenruo btrfs_super_chunk_root_generation(sb)); 255821a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 255921a852b0SQu Wenruo && btrfs_super_cache_generation(sb) != (u64)-1) 256021a852b0SQu Wenruo btrfs_warn(fs_info, 256121a852b0SQu Wenruo "suspicious: generation < cache_generation: %llu < %llu", 256221a852b0SQu Wenruo btrfs_super_generation(sb), 256321a852b0SQu Wenruo btrfs_super_cache_generation(sb)); 256421a852b0SQu Wenruo 256521a852b0SQu Wenruo return ret; 256621a852b0SQu Wenruo } 256721a852b0SQu Wenruo 2568069ec957SQu Wenruo /* 2569069ec957SQu Wenruo * Validation of super block at mount time. 2570069ec957SQu Wenruo * Some checks already done early at mount time, like csum type and incompat 2571069ec957SQu Wenruo * flags will be skipped. 2572069ec957SQu Wenruo */ 2573069ec957SQu Wenruo static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) 2574069ec957SQu Wenruo { 2575069ec957SQu Wenruo return validate_super(fs_info, fs_info->super_copy, 0); 2576069ec957SQu Wenruo } 2577069ec957SQu Wenruo 257875cb857dSQu Wenruo /* 257975cb857dSQu Wenruo * Validation of super block at write time. 258075cb857dSQu Wenruo * Some checks like bytenr check will be skipped as their values will be 258175cb857dSQu Wenruo * overwritten soon. 258275cb857dSQu Wenruo * Extra checks like csum type and incompat flags will be done here. 258375cb857dSQu Wenruo */ 258475cb857dSQu Wenruo static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, 258575cb857dSQu Wenruo struct btrfs_super_block *sb) 258675cb857dSQu Wenruo { 258775cb857dSQu Wenruo int ret; 258875cb857dSQu Wenruo 258975cb857dSQu Wenruo ret = validate_super(fs_info, sb, -1); 259075cb857dSQu Wenruo if (ret < 0) 259175cb857dSQu Wenruo goto out; 2592e7e16f48SJohannes Thumshirn if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) { 259375cb857dSQu Wenruo ret = -EUCLEAN; 259475cb857dSQu Wenruo btrfs_err(fs_info, "invalid csum type, has %u want %u", 259575cb857dSQu Wenruo btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); 259675cb857dSQu Wenruo goto out; 259775cb857dSQu Wenruo } 259875cb857dSQu Wenruo if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 259975cb857dSQu Wenruo ret = -EUCLEAN; 260075cb857dSQu Wenruo btrfs_err(fs_info, 260175cb857dSQu Wenruo "invalid incompat flags, has 0x%llx valid mask 0x%llx", 260275cb857dSQu Wenruo btrfs_super_incompat_flags(sb), 260375cb857dSQu Wenruo (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); 260475cb857dSQu Wenruo goto out; 260575cb857dSQu Wenruo } 260675cb857dSQu Wenruo out: 260775cb857dSQu Wenruo if (ret < 0) 260875cb857dSQu Wenruo btrfs_err(fs_info, 260975cb857dSQu Wenruo "super block corruption detected before writing it to disk"); 261075cb857dSQu Wenruo return ret; 261175cb857dSQu Wenruo } 261275cb857dSQu Wenruo 2613b8522a1eSNikolay Borisov int __cold init_tree_roots(struct btrfs_fs_info *fs_info) 2614b8522a1eSNikolay Borisov { 2615b8522a1eSNikolay Borisov struct btrfs_super_block *sb = fs_info->super_copy; 2616b8522a1eSNikolay Borisov struct btrfs_root *tree_root = fs_info->tree_root; 2617b8522a1eSNikolay Borisov bool handle_error = false; 2618b8522a1eSNikolay Borisov int ret = 0; 2619b8522a1eSNikolay Borisov int i; 2620b8522a1eSNikolay Borisov 2621b8522a1eSNikolay Borisov for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 2622b8522a1eSNikolay Borisov u64 generation; 2623b8522a1eSNikolay Borisov int level; 2624b8522a1eSNikolay Borisov 2625b8522a1eSNikolay Borisov if (handle_error) { 2626b8522a1eSNikolay Borisov if (!IS_ERR(tree_root->node)) 2627b8522a1eSNikolay Borisov free_extent_buffer(tree_root->node); 2628b8522a1eSNikolay Borisov tree_root->node = NULL; 2629b8522a1eSNikolay Borisov 2630b8522a1eSNikolay Borisov if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) 2631b8522a1eSNikolay Borisov break; 2632b8522a1eSNikolay Borisov 2633b8522a1eSNikolay Borisov free_root_pointers(fs_info, 0); 2634b8522a1eSNikolay Borisov 2635b8522a1eSNikolay Borisov /* 2636b8522a1eSNikolay Borisov * Don't use the log in recovery mode, it won't be 2637b8522a1eSNikolay Borisov * valid 2638b8522a1eSNikolay Borisov */ 2639b8522a1eSNikolay Borisov btrfs_set_super_log_root(sb, 0); 2640b8522a1eSNikolay Borisov 2641b8522a1eSNikolay Borisov /* We can't trust the free space cache either */ 2642b8522a1eSNikolay Borisov btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 2643b8522a1eSNikolay Borisov 2644b8522a1eSNikolay Borisov ret = read_backup_root(fs_info, i); 2645b8522a1eSNikolay Borisov if (ret < 0) 2646b8522a1eSNikolay Borisov return ret; 2647b8522a1eSNikolay Borisov } 2648b8522a1eSNikolay Borisov generation = btrfs_super_generation(sb); 2649b8522a1eSNikolay Borisov level = btrfs_super_root_level(sb); 2650b8522a1eSNikolay Borisov tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb), 2651b8522a1eSNikolay Borisov generation, level, NULL); 2652b8522a1eSNikolay Borisov if (IS_ERR(tree_root->node) || 2653b8522a1eSNikolay Borisov !extent_buffer_uptodate(tree_root->node)) { 2654b8522a1eSNikolay Borisov handle_error = true; 2655b8522a1eSNikolay Borisov 2656b8522a1eSNikolay Borisov if (IS_ERR(tree_root->node)) 2657b8522a1eSNikolay Borisov ret = PTR_ERR(tree_root->node); 2658b8522a1eSNikolay Borisov else if (!extent_buffer_uptodate(tree_root->node)) 2659b8522a1eSNikolay Borisov ret = -EUCLEAN; 2660b8522a1eSNikolay Borisov 2661b8522a1eSNikolay Borisov btrfs_warn(fs_info, "failed to read tree root"); 2662b8522a1eSNikolay Borisov continue; 2663b8522a1eSNikolay Borisov } 2664b8522a1eSNikolay Borisov 2665b8522a1eSNikolay Borisov btrfs_set_root_node(&tree_root->root_item, tree_root->node); 2666b8522a1eSNikolay Borisov tree_root->commit_root = btrfs_root_node(tree_root); 2667b8522a1eSNikolay Borisov btrfs_set_root_refs(&tree_root->root_item, 1); 2668b8522a1eSNikolay Borisov 2669336a0d8dSNikolay Borisov /* 2670336a0d8dSNikolay Borisov * No need to hold btrfs_root::objectid_mutex since the fs 2671336a0d8dSNikolay Borisov * hasn't been fully initialised and we are the only user 2672336a0d8dSNikolay Borisov */ 2673b8522a1eSNikolay Borisov ret = btrfs_find_highest_objectid(tree_root, 2674b8522a1eSNikolay Borisov &tree_root->highest_objectid); 2675b8522a1eSNikolay Borisov if (ret < 0) { 2676b8522a1eSNikolay Borisov handle_error = true; 2677b8522a1eSNikolay Borisov continue; 2678b8522a1eSNikolay Borisov } 2679b8522a1eSNikolay Borisov 2680b8522a1eSNikolay Borisov ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 2681b8522a1eSNikolay Borisov 2682b8522a1eSNikolay Borisov ret = btrfs_read_roots(fs_info); 2683b8522a1eSNikolay Borisov if (ret < 0) { 2684b8522a1eSNikolay Borisov handle_error = true; 2685b8522a1eSNikolay Borisov continue; 2686b8522a1eSNikolay Borisov } 2687b8522a1eSNikolay Borisov 2688b8522a1eSNikolay Borisov /* All successful */ 2689b8522a1eSNikolay Borisov fs_info->generation = generation; 2690b8522a1eSNikolay Borisov fs_info->last_trans_committed = generation; 2691b8522a1eSNikolay Borisov break; 2692b8522a1eSNikolay Borisov } 2693b8522a1eSNikolay Borisov 2694b8522a1eSNikolay Borisov return ret; 2695b8522a1eSNikolay Borisov } 2696b8522a1eSNikolay Borisov 2697b105e927SDavid Sterba int __cold open_ctree(struct super_block *sb, 2698dfe25020SChris Mason struct btrfs_fs_devices *fs_devices, 2699dfe25020SChris Mason char *options) 2700eb60ceacSChris Mason { 2701db94535dSChris Mason u32 sectorsize; 2702db94535dSChris Mason u32 nodesize; 270387ee04ebSChris Mason u32 stripesize; 270484234f3aSYan Zheng u64 generation; 2705f2b636e8SJosef Bacik u64 features; 270651bce6c9SJohannes Thumshirn u16 csum_type; 27073de4586cSChris Mason struct btrfs_key location; 2708a061fc8dSChris Mason struct buffer_head *bh; 27094d34b278SIlya Dryomov struct btrfs_super_block *disk_super; 2710815745cfSAl Viro struct btrfs_fs_info *fs_info = btrfs_sb(sb); 2711f84a8bd6SAl Viro struct btrfs_root *tree_root; 27124d34b278SIlya Dryomov struct btrfs_root *chunk_root; 2713eb60ceacSChris Mason int ret; 2714e58ca020SYan int err = -EINVAL; 27156675df31SOmar Sandoval int clear_free_space_tree = 0; 2716581c1760SQu Wenruo int level; 27174543df7eSChris Mason 271874e4d827SDavid Sterba tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL); 271974e4d827SDavid Sterba chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL); 2720cb517eabSMiao Xie if (!tree_root || !chunk_root) { 272139279cc3SChris Mason err = -ENOMEM; 272239279cc3SChris Mason goto fail; 272339279cc3SChris Mason } 272476dda93cSYan, Zheng 272576dda93cSYan, Zheng ret = init_srcu_struct(&fs_info->subvol_srcu); 272676dda93cSYan, Zheng if (ret) { 272776dda93cSYan, Zheng err = ret; 272876dda93cSYan, Zheng goto fail; 272976dda93cSYan, Zheng } 273076dda93cSYan, Zheng 27314297ff84SJosef Bacik ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL); 2732e2d84521SMiao Xie if (ret) { 2733e2d84521SMiao Xie err = ret; 27349e11ceeeSJan Kara goto fail_srcu; 2735e2d84521SMiao Xie } 27364297ff84SJosef Bacik 27374297ff84SJosef Bacik ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 27384297ff84SJosef Bacik if (ret) { 27394297ff84SJosef Bacik err = ret; 27404297ff84SJosef Bacik goto fail_dio_bytes; 27414297ff84SJosef Bacik } 274209cbfeafSKirill A. Shutemov fs_info->dirty_metadata_batch = PAGE_SIZE * 2743e2d84521SMiao Xie (1 + ilog2(nr_cpu_ids)); 2744e2d84521SMiao Xie 2745908c7f19STejun Heo ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 2746963d678bSMiao Xie if (ret) { 2747963d678bSMiao Xie err = ret; 2748963d678bSMiao Xie goto fail_dirty_metadata_bytes; 2749963d678bSMiao Xie } 2750963d678bSMiao Xie 27517f8d236aSDavid Sterba ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, 27527f8d236aSDavid Sterba GFP_KERNEL); 2753c404e0dcSMiao Xie if (ret) { 2754c404e0dcSMiao Xie err = ret; 2755c404e0dcSMiao Xie goto fail_delalloc_bytes; 2756c404e0dcSMiao Xie } 2757c404e0dcSMiao Xie 275876dda93cSYan, Zheng INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 2759f28491e0SJosef Bacik INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 27608fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 2761facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 276224bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 2763eb73c1b7SMiao Xie INIT_LIST_HEAD(&fs_info->delalloc_roots); 276411833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 2765eb73c1b7SMiao Xie spin_lock_init(&fs_info->delalloc_root_lock); 2766a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 276776dda93cSYan, Zheng spin_lock_init(&fs_info->fs_roots_radix_lock); 276824bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 27694cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 2770f29021b2SJan Schmidt spin_lock_init(&fs_info->tree_mod_seq_lock); 2771ceda0864SMiao Xie spin_lock_init(&fs_info->super_lock); 2772f28491e0SJosef Bacik spin_lock_init(&fs_info->buffer_lock); 277347ab2a6cSJosef Bacik spin_lock_init(&fs_info->unused_bgs_lock); 2774f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 2775d7c15171SZhao Lei mutex_init(&fs_info->unused_bg_unpin_mutex); 277667c5e7d4SFilipe Manana mutex_init(&fs_info->delete_unused_bgs_mutex); 27777585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 2778573bfb72SMiao Xie mutex_init(&fs_info->delalloc_root_mutex); 2779de98ced9SMiao Xie seqlock_init(&fs_info->profiles_lock); 278019c00ddcSChris Mason 27810b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 27826324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 2783f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 278447ab2a6cSJosef Bacik INIT_LIST_HEAD(&fs_info->unused_bgs); 2785c8bf1b67SDavid Sterba extent_map_tree_init(&fs_info->mapping_tree); 278666d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 278766d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 278866d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 278966d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 279066d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 279166d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 279266d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 2793ba2c4d4eSJosef Bacik btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, 2794ba2c4d4eSJosef Bacik BTRFS_BLOCK_RSV_DELREFS); 2795ba2c4d4eSJosef Bacik 2796771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 27974cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 27982fefd558SZhao Lei atomic_set(&fs_info->reada_works_cnt, 0); 2799034f784dSJosef Bacik atomic_set(&fs_info->nr_delayed_iputs, 0); 2800fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 2801e20d96d6SChris Mason fs_info->sb = sb; 280295ac567aSFilipe David Borba Manana fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 28039ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 28044cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 2805a5ed45f8SNikolay Borisov atomic64_set(&fs_info->free_chunk_space, 0); 2806f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 28078b87dc17SDavid Sterba fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 2808f8c269d7SDavid Sterba fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 280990519d66SArne Jansen /* readahead state */ 2810d0164adcSMel Gorman INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); 281190519d66SArne Jansen spin_lock_init(&fs_info->reada_lock); 2812fd708b81SJosef Bacik btrfs_init_ref_verify(fs_info); 2813c8b97818SChris Mason 2814b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 2815b34b086cSChris Mason num_online_cpus() + 2, 8); 28160afbaf8cSChris Mason 2817199c2a9cSMiao Xie INIT_LIST_HEAD(&fs_info->ordered_roots); 2818199c2a9cSMiao Xie spin_lock_init(&fs_info->ordered_root_lock); 281969fe2d75SJosef Bacik 282069fe2d75SJosef Bacik fs_info->btree_inode = new_inode(sb); 282169fe2d75SJosef Bacik if (!fs_info->btree_inode) { 282269fe2d75SJosef Bacik err = -ENOMEM; 282369fe2d75SJosef Bacik goto fail_bio_counter; 282469fe2d75SJosef Bacik } 282569fe2d75SJosef Bacik mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 282669fe2d75SJosef Bacik 282716cdcec7SMiao Xie fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 282874e4d827SDavid Sterba GFP_KERNEL); 282916cdcec7SMiao Xie if (!fs_info->delayed_root) { 283016cdcec7SMiao Xie err = -ENOMEM; 283116cdcec7SMiao Xie goto fail_iput; 283216cdcec7SMiao Xie } 283316cdcec7SMiao Xie btrfs_init_delayed_root(fs_info->delayed_root); 28343eaa2885SChris Mason 2835638aa7edSEric Sandeen btrfs_init_scrub(fs_info); 283621adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 283721adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 283821adbd5cSStefan Behrens #endif 2839779a65a4SEric Sandeen btrfs_init_balance(fs_info); 284021c7e756SMiao Xie btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work); 2841a2de733cSArne Jansen 28429f6d2510SDavid Sterba sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; 28439f6d2510SDavid Sterba sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); 2844a061fc8dSChris Mason 28456bccf3abSJeff Mahoney btrfs_init_btree_inode(fs_info); 284639279cc3SChris Mason 2847e02119d5SChris Mason spin_lock_init(&fs_info->block_group_cache_lock); 28486bef4d31SEric Paris fs_info->block_group_cache_tree = RB_ROOT; 2849a1897fddSLiu Bo fs_info->first_logical_byte = (u64)-1; 2850925baeddSChris Mason 285143eb5f29SQu Wenruo extent_io_tree_init(fs_info, &fs_info->freed_extents[0], 285243eb5f29SQu Wenruo IO_TREE_FS_INFO_FREED_EXTENTS0, NULL); 285343eb5f29SQu Wenruo extent_io_tree_init(fs_info, &fs_info->freed_extents[1], 285443eb5f29SQu Wenruo IO_TREE_FS_INFO_FREED_EXTENTS1, NULL); 285511833d66SYan Zheng fs_info->pinned_extents = &fs_info->freed_extents[0]; 2856afcdd129SJosef Bacik set_bit(BTRFS_FS_BARRIER, &fs_info->flags); 2857509659cdSChris Mason 28585a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 285979154b1bSChris Mason mutex_init(&fs_info->tree_log_mutex); 2860925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 2861a74a4b97SChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 2862a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 28631bbc621eSChris Mason mutex_init(&fs_info->ro_block_group_mutex); 28649e351cc8SJosef Bacik init_rwsem(&fs_info->commit_root_sem); 2865c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 286676dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 2867803b2f54SStefan Behrens sema_init(&fs_info->uuid_tree_rescan_sem, 1); 2868fa9c0d79SChris Mason 2869ad618368SEric Sandeen btrfs_init_dev_replace_locks(fs_info); 2870f9e92e40SEric Sandeen btrfs_init_qgroup(fs_info); 2871416ac51dSArne Jansen 2872fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2873fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2874fa9c0d79SChris Mason 2875e6dcd2dcSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 2876f9295749SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 2877bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 28784854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 2879034f784dSJosef Bacik init_waitqueue_head(&fs_info->delayed_iputs_wait); 28803768f368SChris Mason 2881da17066cSJeff Mahoney /* Usable values until the real ones are cached from the superblock */ 2882da17066cSJeff Mahoney fs_info->nodesize = 4096; 2883da17066cSJeff Mahoney fs_info->sectorsize = 4096; 2884da17066cSJeff Mahoney fs_info->stripesize = 4096; 2885da17066cSJeff Mahoney 2886eede2bf3SOmar Sandoval spin_lock_init(&fs_info->swapfile_pins_lock); 2887eede2bf3SOmar Sandoval fs_info->swapfile_pins = RB_ROOT; 2888eede2bf3SOmar Sandoval 28899e967495SFilipe Manana fs_info->send_in_progress = 0; 28909e967495SFilipe Manana 289153b381b3SDavid Woodhouse ret = btrfs_alloc_stripe_hash_table(fs_info); 289253b381b3SDavid Woodhouse if (ret) { 289383c8266aSDavid Sterba err = ret; 289453b381b3SDavid Woodhouse goto fail_alloc; 289553b381b3SDavid Woodhouse } 289653b381b3SDavid Woodhouse 2897da17066cSJeff Mahoney __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID); 28987eccb903SChris Mason 28993c4bb26bSChris Mason invalidate_bdev(fs_devices->latest_bdev); 29001104a885SDavid Sterba 29011104a885SDavid Sterba /* 29021104a885SDavid Sterba * Read super block and check the signature bytes only 29031104a885SDavid Sterba */ 2904a512bbf8SYan Zheng bh = btrfs_read_dev_super(fs_devices->latest_bdev); 290592fc03fbSAnand Jain if (IS_ERR(bh)) { 290692fc03fbSAnand Jain err = PTR_ERR(bh); 290716cdcec7SMiao Xie goto fail_alloc; 290820b45077SDave Young } 290939279cc3SChris Mason 29101104a885SDavid Sterba /* 29118dc3f22cSJohannes Thumshirn * Verify the type first, if that or the the checksum value are 29128dc3f22cSJohannes Thumshirn * corrupted, we'll find out 29138dc3f22cSJohannes Thumshirn */ 291451bce6c9SJohannes Thumshirn csum_type = btrfs_super_csum_type((struct btrfs_super_block *)bh->b_data); 291551bce6c9SJohannes Thumshirn if (!btrfs_supported_super_csum(csum_type)) { 29168dc3f22cSJohannes Thumshirn btrfs_err(fs_info, "unsupported checksum algorithm: %u", 291751bce6c9SJohannes Thumshirn csum_type); 29188dc3f22cSJohannes Thumshirn err = -EINVAL; 29198dc3f22cSJohannes Thumshirn brelse(bh); 29208dc3f22cSJohannes Thumshirn goto fail_alloc; 29218dc3f22cSJohannes Thumshirn } 29228dc3f22cSJohannes Thumshirn 29236d97c6e3SJohannes Thumshirn ret = btrfs_init_csum_hash(fs_info, csum_type); 29246d97c6e3SJohannes Thumshirn if (ret) { 29256d97c6e3SJohannes Thumshirn err = ret; 29266d97c6e3SJohannes Thumshirn goto fail_alloc; 29276d97c6e3SJohannes Thumshirn } 29286d97c6e3SJohannes Thumshirn 29298dc3f22cSJohannes Thumshirn /* 29301104a885SDavid Sterba * We want to check superblock checksum, the type is stored inside. 29311104a885SDavid Sterba * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 29321104a885SDavid Sterba */ 2933ab8d0fc4SJeff Mahoney if (btrfs_check_super_csum(fs_info, bh->b_data)) { 293405135f59SDavid Sterba btrfs_err(fs_info, "superblock checksum mismatch"); 29351104a885SDavid Sterba err = -EINVAL; 2936b2acdddfSAnand Jain brelse(bh); 29376d97c6e3SJohannes Thumshirn goto fail_csum; 29381104a885SDavid Sterba } 29391104a885SDavid Sterba 29401104a885SDavid Sterba /* 29411104a885SDavid Sterba * super_copy is zeroed at allocation time and we never touch the 29421104a885SDavid Sterba * following bytes up to INFO_SIZE, the checksum is calculated from 29431104a885SDavid Sterba * the whole block of INFO_SIZE 29441104a885SDavid Sterba */ 29456c41761fSDavid Sterba memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy)); 2946a061fc8dSChris Mason brelse(bh); 29475f39d397SChris Mason 2948fbc6feaeSNikolay Borisov disk_super = fs_info->super_copy; 2949fbc6feaeSNikolay Borisov 2950de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, 2951de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 2952de37aa51SNikolay Borisov 29537239ff4bSNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID)) { 2954de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid, 2955de37aa51SNikolay Borisov fs_info->super_copy->metadata_uuid, 2956de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 29577239ff4bSNikolay Borisov } 29580b86a832SChris Mason 2959fbc6feaeSNikolay Borisov features = btrfs_super_flags(disk_super); 2960fbc6feaeSNikolay Borisov if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { 2961fbc6feaeSNikolay Borisov features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; 2962fbc6feaeSNikolay Borisov btrfs_set_super_flags(disk_super, features); 2963fbc6feaeSNikolay Borisov btrfs_info(fs_info, 2964fbc6feaeSNikolay Borisov "found metadata UUID change in progress flag, clearing"); 2965fbc6feaeSNikolay Borisov } 2966fbc6feaeSNikolay Borisov 2967fbc6feaeSNikolay Borisov memcpy(fs_info->super_for_commit, fs_info->super_copy, 2968fbc6feaeSNikolay Borisov sizeof(*fs_info->super_for_commit)); 2969de37aa51SNikolay Borisov 2970069ec957SQu Wenruo ret = btrfs_validate_mount_super(fs_info); 29711104a885SDavid Sterba if (ret) { 297205135f59SDavid Sterba btrfs_err(fs_info, "superblock contains fatal errors"); 29731104a885SDavid Sterba err = -EINVAL; 29746d97c6e3SJohannes Thumshirn goto fail_csum; 29751104a885SDavid Sterba } 29761104a885SDavid Sterba 29770f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 29786d97c6e3SJohannes Thumshirn goto fail_csum; 29790f7d52f4SChris Mason 2980acce952bSliubo /* check FS state, whether FS is broken. */ 298187533c47SMiao Xie if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 298287533c47SMiao Xie set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 2983acce952bSliubo 298475e7cb7fSLiu Bo /* 2985af31f5e5SChris Mason * run through our array of backup supers and setup 2986af31f5e5SChris Mason * our ring pointer to the oldest one 2987af31f5e5SChris Mason */ 2988fc2e4c5bSNikolay Borisov find_oldest_super_backup(fs_info); 2989af31f5e5SChris Mason 2990af31f5e5SChris Mason /* 299175e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 299275e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 299375e7cb7fSLiu Bo */ 299475e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 299575e7cb7fSLiu Bo 29962ff7e61eSJeff Mahoney ret = btrfs_parse_options(fs_info, options, sb->s_flags); 29972b82032cSYan Zheng if (ret) { 29982b82032cSYan Zheng err = ret; 29996d97c6e3SJohannes Thumshirn goto fail_csum; 30002b82032cSYan Zheng } 3001dfe25020SChris Mason 3002f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 3003f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 3004f2b636e8SJosef Bacik if (features) { 300505135f59SDavid Sterba btrfs_err(fs_info, 300605135f59SDavid Sterba "cannot mount because of unsupported optional features (%llx)", 3007c1c9ff7cSGeert Uytterhoeven features); 3008f2b636e8SJosef Bacik err = -EINVAL; 30096d97c6e3SJohannes Thumshirn goto fail_csum; 3010f2b636e8SJosef Bacik } 3011f2b636e8SJosef Bacik 30125d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 30135d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 30140b246afaSJeff Mahoney if (fs_info->compress_type == BTRFS_COMPRESS_LZO) 3015a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 30165c1aab1dSNick Terrell else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) 30175c1aab1dSNick Terrell features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; 3018727011e0SChris Mason 30193173a18fSJosef Bacik if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) 302005135f59SDavid Sterba btrfs_info(fs_info, "has skinny extents"); 30213173a18fSJosef Bacik 3022727011e0SChris Mason /* 3023727011e0SChris Mason * flag our filesystem as having big metadata blocks if 3024727011e0SChris Mason * they are bigger than the page size 3025727011e0SChris Mason */ 302609cbfeafSKirill A. Shutemov if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) { 3027727011e0SChris Mason if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 302805135f59SDavid Sterba btrfs_info(fs_info, 302905135f59SDavid Sterba "flagging fs with big metadata feature"); 3030727011e0SChris Mason features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 3031727011e0SChris Mason } 3032727011e0SChris Mason 3033bc3f116fSChris Mason nodesize = btrfs_super_nodesize(disk_super); 3034bc3f116fSChris Mason sectorsize = btrfs_super_sectorsize(disk_super); 3035b7f67055SChandan Rajendra stripesize = sectorsize; 3036707e8a07SDavid Sterba fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 3037963d678bSMiao Xie fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 3038bc3f116fSChris Mason 3039da17066cSJeff Mahoney /* Cache block sizes */ 3040da17066cSJeff Mahoney fs_info->nodesize = nodesize; 3041da17066cSJeff Mahoney fs_info->sectorsize = sectorsize; 3042da17066cSJeff Mahoney fs_info->stripesize = stripesize; 3043da17066cSJeff Mahoney 3044bc3f116fSChris Mason /* 3045bc3f116fSChris Mason * mixed block groups end up with duplicate but slightly offset 3046bc3f116fSChris Mason * extent buffers for the same range. It leads to corruptions 3047bc3f116fSChris Mason */ 3048bc3f116fSChris Mason if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 3049707e8a07SDavid Sterba (sectorsize != nodesize)) { 305005135f59SDavid Sterba btrfs_err(fs_info, 305105135f59SDavid Sterba "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", 305205135f59SDavid Sterba nodesize, sectorsize); 30536d97c6e3SJohannes Thumshirn goto fail_csum; 3054bc3f116fSChris Mason } 3055bc3f116fSChris Mason 3056ceda0864SMiao Xie /* 3057ceda0864SMiao Xie * Needn't use the lock because there is no other task which will 3058ceda0864SMiao Xie * update the flag. 3059ceda0864SMiao Xie */ 30605d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 30615d4f98a2SYan Zheng 3062f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 3063f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 3064bc98a42cSDavid Howells if (!sb_rdonly(sb) && features) { 306505135f59SDavid Sterba btrfs_err(fs_info, 306605135f59SDavid Sterba "cannot mount read-write because of unsupported optional features (%llx)", 3067c1c9ff7cSGeert Uytterhoeven features); 3068f2b636e8SJosef Bacik err = -EINVAL; 30696d97c6e3SJohannes Thumshirn goto fail_csum; 3070f2b636e8SJosef Bacik } 307161d92c32SChris Mason 30722a458198SEric Sandeen ret = btrfs_init_workqueues(fs_info, fs_devices); 30732a458198SEric Sandeen if (ret) { 30742a458198SEric Sandeen err = ret; 30750dc3b84aSJosef Bacik goto fail_sb_buffer; 30760dc3b84aSJosef Bacik } 30774543df7eSChris Mason 30789e11ceeeSJan Kara sb->s_bdi->congested_fn = btrfs_congested_fn; 30799e11ceeeSJan Kara sb->s_bdi->congested_data = fs_info; 30809e11ceeeSJan Kara sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK; 3081b5420237SNikolay Borisov sb->s_bdi->ra_pages = VM_READAHEAD_PAGES; 30829e11ceeeSJan Kara sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); 30839e11ceeeSJan Kara sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); 30844575c9ccSChris Mason 3085a061fc8dSChris Mason sb->s_blocksize = sectorsize; 3086a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 3087de37aa51SNikolay Borisov memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); 3088db94535dSChris Mason 3089925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 30906bccf3abSJeff Mahoney ret = btrfs_read_sys_array(fs_info); 3091925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 309284eed90fSChris Mason if (ret) { 309305135f59SDavid Sterba btrfs_err(fs_info, "failed to read the system array: %d", ret); 30945d4f98a2SYan Zheng goto fail_sb_buffer; 309584eed90fSChris Mason } 30960b86a832SChris Mason 309784234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 3098581c1760SQu Wenruo level = btrfs_super_chunk_root_level(disk_super); 30990b86a832SChris Mason 3100da17066cSJeff Mahoney __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID); 31010b86a832SChris Mason 31022ff7e61eSJeff Mahoney chunk_root->node = read_tree_block(fs_info, 31030b86a832SChris Mason btrfs_super_chunk_root(disk_super), 3104581c1760SQu Wenruo generation, level, NULL); 310564c043deSLiu Bo if (IS_ERR(chunk_root->node) || 310664c043deSLiu Bo !extent_buffer_uptodate(chunk_root->node)) { 310705135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk root"); 3108e5fffbacSchandan if (!IS_ERR(chunk_root->node)) 3109e5fffbacSchandan free_extent_buffer(chunk_root->node); 311095ab1f64SZhao Lei chunk_root->node = NULL; 3111af31f5e5SChris Mason goto fail_tree_roots; 311283121942SDavid Woodhouse } 31135d4f98a2SYan Zheng btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 31145d4f98a2SYan Zheng chunk_root->commit_root = btrfs_root_node(chunk_root); 31150b86a832SChris Mason 3116e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 3117b308bc2fSGeert Uytterhoeven btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE); 3118e17cade2SChris Mason 31195b4aacefSJeff Mahoney ret = btrfs_read_chunk_tree(fs_info); 31202b82032cSYan Zheng if (ret) { 312105135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk tree: %d", ret); 3122af31f5e5SChris Mason goto fail_tree_roots; 31232b82032cSYan Zheng } 31240b86a832SChris Mason 31258dabb742SStefan Behrens /* 31269b99b115SAnand Jain * Keep the devid that is marked to be the target device for the 31279b99b115SAnand Jain * device replace procedure 31288dabb742SStefan Behrens */ 31299b99b115SAnand Jain btrfs_free_extra_devids(fs_devices, 0); 3130dfe25020SChris Mason 3131a6b0d5c8SChris Mason if (!fs_devices->latest_bdev) { 313205135f59SDavid Sterba btrfs_err(fs_info, "failed to read devices"); 3133a6b0d5c8SChris Mason goto fail_tree_roots; 3134a6b0d5c8SChris Mason } 3135a6b0d5c8SChris Mason 3136b8522a1eSNikolay Borisov ret = init_tree_roots(fs_info); 31374bbcaa64SEric Sandeen if (ret) 3138b8522a1eSNikolay Borisov goto fail_tree_roots; 31398929ecfaSYan, Zheng 3140cf90d884SQu Wenruo ret = btrfs_verify_dev_extents(fs_info); 3141cf90d884SQu Wenruo if (ret) { 3142cf90d884SQu Wenruo btrfs_err(fs_info, 3143cf90d884SQu Wenruo "failed to verify dev extents against chunks: %d", 3144cf90d884SQu Wenruo ret); 3145cf90d884SQu Wenruo goto fail_block_groups; 3146cf90d884SQu Wenruo } 314768310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 314868310a5eSIlya Dryomov if (ret) { 314905135f59SDavid Sterba btrfs_err(fs_info, "failed to recover balance: %d", ret); 315068310a5eSIlya Dryomov goto fail_block_groups; 315168310a5eSIlya Dryomov } 315268310a5eSIlya Dryomov 3153733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 3154733f4fbbSStefan Behrens if (ret) { 315505135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_stats: %d", ret); 3156733f4fbbSStefan Behrens goto fail_block_groups; 3157733f4fbbSStefan Behrens } 3158733f4fbbSStefan Behrens 31598dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 31608dabb742SStefan Behrens if (ret) { 316105135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_replace: %d", ret); 31628dabb742SStefan Behrens goto fail_block_groups; 31638dabb742SStefan Behrens } 31648dabb742SStefan Behrens 31659b99b115SAnand Jain btrfs_free_extra_devids(fs_devices, 1); 31668dabb742SStefan Behrens 3167b7c35e81SAnand Jain ret = btrfs_sysfs_add_fsid(fs_devices, NULL); 3168b7c35e81SAnand Jain if (ret) { 316905135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", 317005135f59SDavid Sterba ret); 3171b7c35e81SAnand Jain goto fail_block_groups; 3172b7c35e81SAnand Jain } 3173b7c35e81SAnand Jain 3174b7c35e81SAnand Jain ret = btrfs_sysfs_add_device(fs_devices); 3175b7c35e81SAnand Jain if (ret) { 317605135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs device interface: %d", 317705135f59SDavid Sterba ret); 3178b7c35e81SAnand Jain goto fail_fsdev_sysfs; 3179b7c35e81SAnand Jain } 3180b7c35e81SAnand Jain 318196f3136eSAnand Jain ret = btrfs_sysfs_add_mounted(fs_info); 31825ac1d209SJeff Mahoney if (ret) { 318305135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); 3184b7c35e81SAnand Jain goto fail_fsdev_sysfs; 31855ac1d209SJeff Mahoney } 31865ac1d209SJeff Mahoney 3187c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 3188c59021f8Sliubo if (ret) { 318905135f59SDavid Sterba btrfs_err(fs_info, "failed to initialize space info: %d", ret); 31902365dd3cSAnand Jain goto fail_sysfs; 3191c59021f8Sliubo } 3192c59021f8Sliubo 31935b4aacefSJeff Mahoney ret = btrfs_read_block_groups(fs_info); 31941b1d1f66SJosef Bacik if (ret) { 319505135f59SDavid Sterba btrfs_err(fs_info, "failed to read block groups: %d", ret); 31962365dd3cSAnand Jain goto fail_sysfs; 31971b1d1f66SJosef Bacik } 31984330e183SQu Wenruo 31996528b99dSAnand Jain if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) { 320005135f59SDavid Sterba btrfs_warn(fs_info, 320152042d8eSAndrea Gelmini "writable mount is not allowed due to too many missing devices"); 32022365dd3cSAnand Jain goto fail_sysfs; 3203292fd7fcSStefan Behrens } 32049078a3e1SChris Mason 3205a74a4b97SChris Mason fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 3206a74a4b97SChris Mason "btrfs-cleaner"); 320757506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 32082365dd3cSAnand Jain goto fail_sysfs; 3209a74a4b97SChris Mason 3210a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 3211a74a4b97SChris Mason tree_root, 3212a74a4b97SChris Mason "btrfs-transaction"); 321357506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 32143f157a2fSChris Mason goto fail_cleaner; 3215a74a4b97SChris Mason 3216583b7231SHans van Kranenburg if (!btrfs_test_opt(fs_info, NOSSD) && 3217c289811cSChris Mason !fs_info->fs_devices->rotating) { 3218583b7231SHans van Kranenburg btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); 3219c289811cSChris Mason } 3220c289811cSChris Mason 3221572d9ab7SDavid Sterba /* 322201327610SNicholas D Steeves * Mount does not set all options immediately, we can do it now and do 3223572d9ab7SDavid Sterba * not have to wait for transaction commit 3224572d9ab7SDavid Sterba */ 3225572d9ab7SDavid Sterba btrfs_apply_pending_changes(fs_info); 32263818aea2SQu Wenruo 322721adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 32280b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { 32292ff7e61eSJeff Mahoney ret = btrfsic_mount(fs_info, fs_devices, 32300b246afaSJeff Mahoney btrfs_test_opt(fs_info, 323121adbd5cSStefan Behrens CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? 323221adbd5cSStefan Behrens 1 : 0, 323321adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 323421adbd5cSStefan Behrens if (ret) 323505135f59SDavid Sterba btrfs_warn(fs_info, 323605135f59SDavid Sterba "failed to initialize integrity check module: %d", 323705135f59SDavid Sterba ret); 323821adbd5cSStefan Behrens } 323921adbd5cSStefan Behrens #endif 3240bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 3241bcef60f2SArne Jansen if (ret) 3242bcef60f2SArne Jansen goto fail_trans_kthread; 324321adbd5cSStefan Behrens 3244fd708b81SJosef Bacik if (btrfs_build_ref_tree(fs_info)) 3245fd708b81SJosef Bacik btrfs_err(fs_info, "couldn't build ref tree"); 3246fd708b81SJosef Bacik 324796da0919SQu Wenruo /* do not make disk changes in broken FS or nologreplay is given */ 324896da0919SQu Wenruo if (btrfs_super_log_root(disk_super) != 0 && 32490b246afaSJeff Mahoney !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 325063443bf5SEric Sandeen ret = btrfs_replay_log(fs_info, fs_devices); 325179787eaaSJeff Mahoney if (ret) { 325263443bf5SEric Sandeen err = ret; 325328c16cbbSWang Shilong goto fail_qgroup; 3254e556ce2cSYan Zheng } 3255e02119d5SChris Mason } 32561a40e23bSZheng Yan 32576bccf3abSJeff Mahoney ret = btrfs_find_orphan_roots(fs_info); 325879787eaaSJeff Mahoney if (ret) 325928c16cbbSWang Shilong goto fail_qgroup; 326076dda93cSYan, Zheng 3261bc98a42cSDavid Howells if (!sb_rdonly(sb)) { 3262d68fc57bSYan, Zheng ret = btrfs_cleanup_fs_roots(fs_info); 326344c44af2SIlya Dryomov if (ret) 326428c16cbbSWang Shilong goto fail_qgroup; 326590c711abSZygo Blaxell 326690c711abSZygo Blaxell mutex_lock(&fs_info->cleaner_mutex); 32675d4f98a2SYan Zheng ret = btrfs_recover_relocation(tree_root); 326890c711abSZygo Blaxell mutex_unlock(&fs_info->cleaner_mutex); 3269d7ce5843SMiao Xie if (ret < 0) { 327005135f59SDavid Sterba btrfs_warn(fs_info, "failed to recover relocation: %d", 327105135f59SDavid Sterba ret); 3272d7ce5843SMiao Xie err = -EINVAL; 3273bcef60f2SArne Jansen goto fail_qgroup; 3274d7ce5843SMiao Xie } 32757c2ca468SChris Mason } 32761a40e23bSZheng Yan 32773de4586cSChris Mason location.objectid = BTRFS_FS_TREE_OBJECTID; 32783de4586cSChris Mason location.type = BTRFS_ROOT_ITEM_KEY; 3279cb517eabSMiao Xie location.offset = 0; 32803de4586cSChris Mason 32813de4586cSChris Mason fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location); 32823140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 32833140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 3284f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read fs tree: %d", err); 3285bcef60f2SArne Jansen goto fail_qgroup; 32863140c9a3SDan Carpenter } 3287c289811cSChris Mason 3288bc98a42cSDavid Howells if (sb_rdonly(sb)) 32892b6ba629SIlya Dryomov return 0; 32902b6ba629SIlya Dryomov 3291f8d468a1SOmar Sandoval if (btrfs_test_opt(fs_info, CLEAR_CACHE) && 3292f8d468a1SOmar Sandoval btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 32936675df31SOmar Sandoval clear_free_space_tree = 1; 32946675df31SOmar Sandoval } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 32956675df31SOmar Sandoval !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { 32966675df31SOmar Sandoval btrfs_warn(fs_info, "free space tree is invalid"); 32976675df31SOmar Sandoval clear_free_space_tree = 1; 32986675df31SOmar Sandoval } 32996675df31SOmar Sandoval 33006675df31SOmar Sandoval if (clear_free_space_tree) { 3301f8d468a1SOmar Sandoval btrfs_info(fs_info, "clearing free space tree"); 3302f8d468a1SOmar Sandoval ret = btrfs_clear_free_space_tree(fs_info); 3303f8d468a1SOmar Sandoval if (ret) { 3304f8d468a1SOmar Sandoval btrfs_warn(fs_info, 3305f8d468a1SOmar Sandoval "failed to clear free space tree: %d", ret); 33066bccf3abSJeff Mahoney close_ctree(fs_info); 3307f8d468a1SOmar Sandoval return ret; 3308f8d468a1SOmar Sandoval } 3309f8d468a1SOmar Sandoval } 3310f8d468a1SOmar Sandoval 33110b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && 3312511711afSChris Mason !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 331305135f59SDavid Sterba btrfs_info(fs_info, "creating free space tree"); 3314511711afSChris Mason ret = btrfs_create_free_space_tree(fs_info); 3315511711afSChris Mason if (ret) { 331605135f59SDavid Sterba btrfs_warn(fs_info, 331705135f59SDavid Sterba "failed to create free space tree: %d", ret); 33186bccf3abSJeff Mahoney close_ctree(fs_info); 3319511711afSChris Mason return ret; 3320511711afSChris Mason } 3321511711afSChris Mason } 3322511711afSChris Mason 3323e3acc2a6SJosef Bacik down_read(&fs_info->cleanup_work_sem); 33242b6ba629SIlya Dryomov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 33252b6ba629SIlya Dryomov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 33262b6ba629SIlya Dryomov up_read(&fs_info->cleanup_work_sem); 33276bccf3abSJeff Mahoney close_ctree(fs_info); 33282b6ba629SIlya Dryomov return ret; 33292b6ba629SIlya Dryomov } 3330e3acc2a6SJosef Bacik up_read(&fs_info->cleanup_work_sem); 333159641015SIlya Dryomov 33322b6ba629SIlya Dryomov ret = btrfs_resume_balance_async(fs_info); 33332b6ba629SIlya Dryomov if (ret) { 333405135f59SDavid Sterba btrfs_warn(fs_info, "failed to resume balance: %d", ret); 33356bccf3abSJeff Mahoney close_ctree(fs_info); 33362b6ba629SIlya Dryomov return ret; 3337e3acc2a6SJosef Bacik } 3338e3acc2a6SJosef Bacik 33398dabb742SStefan Behrens ret = btrfs_resume_dev_replace_async(fs_info); 33408dabb742SStefan Behrens if (ret) { 334105135f59SDavid Sterba btrfs_warn(fs_info, "failed to resume device replace: %d", ret); 33426bccf3abSJeff Mahoney close_ctree(fs_info); 33438dabb742SStefan Behrens return ret; 33448dabb742SStefan Behrens } 33458dabb742SStefan Behrens 3346b382a324SJan Schmidt btrfs_qgroup_rescan_resume(fs_info); 3347b382a324SJan Schmidt 33484bbcaa64SEric Sandeen if (!fs_info->uuid_root) { 334905135f59SDavid Sterba btrfs_info(fs_info, "creating UUID tree"); 3350f7a81ea4SStefan Behrens ret = btrfs_create_uuid_tree(fs_info); 3351f7a81ea4SStefan Behrens if (ret) { 335205135f59SDavid Sterba btrfs_warn(fs_info, 335305135f59SDavid Sterba "failed to create the UUID tree: %d", ret); 33546bccf3abSJeff Mahoney close_ctree(fs_info); 3355f7a81ea4SStefan Behrens return ret; 3356f7a81ea4SStefan Behrens } 33570b246afaSJeff Mahoney } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || 33584bbcaa64SEric Sandeen fs_info->generation != 33594bbcaa64SEric Sandeen btrfs_super_uuid_tree_generation(disk_super)) { 336005135f59SDavid Sterba btrfs_info(fs_info, "checking UUID tree"); 336170f80175SStefan Behrens ret = btrfs_check_uuid_tree(fs_info); 336270f80175SStefan Behrens if (ret) { 336305135f59SDavid Sterba btrfs_warn(fs_info, 336405135f59SDavid Sterba "failed to check the UUID tree: %d", ret); 33656bccf3abSJeff Mahoney close_ctree(fs_info); 336670f80175SStefan Behrens return ret; 336770f80175SStefan Behrens } 336870f80175SStefan Behrens } else { 3369afcdd129SJosef Bacik set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 3370f7a81ea4SStefan Behrens } 3371afcdd129SJosef Bacik set_bit(BTRFS_FS_OPEN, &fs_info->flags); 337247ab2a6cSJosef Bacik 33738dcddfa0SQu Wenruo /* 33748dcddfa0SQu Wenruo * backuproot only affect mount behavior, and if open_ctree succeeded, 33758dcddfa0SQu Wenruo * no need to keep the flag 33768dcddfa0SQu Wenruo */ 33778dcddfa0SQu Wenruo btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT); 33788dcddfa0SQu Wenruo 3379ad2b2c80SAl Viro return 0; 338039279cc3SChris Mason 3381bcef60f2SArne Jansen fail_qgroup: 3382bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 33837c2ca468SChris Mason fail_trans_kthread: 33847c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 33852ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 3386faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 33873f157a2fSChris Mason fail_cleaner: 3388a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 33897c2ca468SChris Mason 33907c2ca468SChris Mason /* 33917c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 33927c2ca468SChris Mason * kthreads 33937c2ca468SChris Mason */ 33947c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 33957c2ca468SChris Mason 33962365dd3cSAnand Jain fail_sysfs: 33976618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 33982365dd3cSAnand Jain 3399b7c35e81SAnand Jain fail_fsdev_sysfs: 3400b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3401b7c35e81SAnand Jain 34021b1d1f66SJosef Bacik fail_block_groups: 340354067ae9SJosef Bacik btrfs_put_block_group_cache(fs_info); 3404af31f5e5SChris Mason 3405af31f5e5SChris Mason fail_tree_roots: 34064273eaffSAnand Jain free_root_pointers(fs_info, true); 34072b8195bbSMiao Xie invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3408af31f5e5SChris Mason 340939279cc3SChris Mason fail_sb_buffer: 34107abadb64SLiu Bo btrfs_stop_all_workers(fs_info); 34115cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 34126d97c6e3SJohannes Thumshirn fail_csum: 34136d97c6e3SJohannes Thumshirn btrfs_free_csum_hash(fs_info); 341416cdcec7SMiao Xie fail_alloc: 34154543df7eSChris Mason fail_iput: 3416586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 3417586e46e2SIlya Dryomov 34184543df7eSChris Mason iput(fs_info->btree_inode); 3419c404e0dcSMiao Xie fail_bio_counter: 34207f8d236aSDavid Sterba percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 3421963d678bSMiao Xie fail_delalloc_bytes: 3422963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 3423e2d84521SMiao Xie fail_dirty_metadata_bytes: 3424e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 34254297ff84SJosef Bacik fail_dio_bytes: 34264297ff84SJosef Bacik percpu_counter_destroy(&fs_info->dio_bytes); 342776dda93cSYan, Zheng fail_srcu: 342876dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 34297e662854SQinghuang Feng fail: 343053b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 3431586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 3432ad2b2c80SAl Viro return err; 3433eb60ceacSChris Mason } 3434663faf9fSMasami Hiramatsu ALLOW_ERROR_INJECTION(open_ctree, ERRNO); 3435eb60ceacSChris Mason 3436f2984462SChris Mason static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate) 3437f2984462SChris Mason { 3438f2984462SChris Mason if (uptodate) { 3439f2984462SChris Mason set_buffer_uptodate(bh); 3440f2984462SChris Mason } else { 3441442a4f63SStefan Behrens struct btrfs_device *device = (struct btrfs_device *) 3442442a4f63SStefan Behrens bh->b_private; 3443442a4f63SStefan Behrens 3444fb456252SJeff Mahoney btrfs_warn_rl_in_rcu(device->fs_info, 3445b14af3b4SDavid Sterba "lost page write due to IO error on %s", 3446606686eeSJosef Bacik rcu_str_deref(device->name)); 344701327610SNicholas D Steeves /* note, we don't set_buffer_write_io_error because we have 34481259ab75SChris Mason * our own ways of dealing with the IO errors 34491259ab75SChris Mason */ 3450f2984462SChris Mason clear_buffer_uptodate(bh); 3451442a4f63SStefan Behrens btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS); 3452f2984462SChris Mason } 3453f2984462SChris Mason unlock_buffer(bh); 3454f2984462SChris Mason put_bh(bh); 3455f2984462SChris Mason } 3456f2984462SChris Mason 345729c36d72SAnand Jain int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num, 345829c36d72SAnand Jain struct buffer_head **bh_ret) 345929c36d72SAnand Jain { 346029c36d72SAnand Jain struct buffer_head *bh; 346129c36d72SAnand Jain struct btrfs_super_block *super; 346229c36d72SAnand Jain u64 bytenr; 346329c36d72SAnand Jain 346429c36d72SAnand Jain bytenr = btrfs_sb_offset(copy_num); 346529c36d72SAnand Jain if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode)) 346629c36d72SAnand Jain return -EINVAL; 346729c36d72SAnand Jain 34689f6d2510SDavid Sterba bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE); 346929c36d72SAnand Jain /* 347029c36d72SAnand Jain * If we fail to read from the underlying devices, as of now 347129c36d72SAnand Jain * the best option we have is to mark it EIO. 347229c36d72SAnand Jain */ 347329c36d72SAnand Jain if (!bh) 347429c36d72SAnand Jain return -EIO; 347529c36d72SAnand Jain 347629c36d72SAnand Jain super = (struct btrfs_super_block *)bh->b_data; 347729c36d72SAnand Jain if (btrfs_super_bytenr(super) != bytenr || 347829c36d72SAnand Jain btrfs_super_magic(super) != BTRFS_MAGIC) { 347929c36d72SAnand Jain brelse(bh); 348029c36d72SAnand Jain return -EINVAL; 348129c36d72SAnand Jain } 348229c36d72SAnand Jain 348329c36d72SAnand Jain *bh_ret = bh; 348429c36d72SAnand Jain return 0; 348529c36d72SAnand Jain } 348629c36d72SAnand Jain 348729c36d72SAnand Jain 3488a512bbf8SYan Zheng struct buffer_head *btrfs_read_dev_super(struct block_device *bdev) 3489a512bbf8SYan Zheng { 3490a512bbf8SYan Zheng struct buffer_head *bh; 3491a512bbf8SYan Zheng struct buffer_head *latest = NULL; 3492a512bbf8SYan Zheng struct btrfs_super_block *super; 3493a512bbf8SYan Zheng int i; 3494a512bbf8SYan Zheng u64 transid = 0; 349592fc03fbSAnand Jain int ret = -EINVAL; 3496a512bbf8SYan Zheng 3497a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 3498a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 3499a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 3500a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3501a512bbf8SYan Zheng */ 3502a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 350329c36d72SAnand Jain ret = btrfs_read_dev_one_super(bdev, i, &bh); 350429c36d72SAnand Jain if (ret) 3505a512bbf8SYan Zheng continue; 3506a512bbf8SYan Zheng 3507a512bbf8SYan Zheng super = (struct btrfs_super_block *)bh->b_data; 3508a512bbf8SYan Zheng 3509a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 3510a512bbf8SYan Zheng brelse(latest); 3511a512bbf8SYan Zheng latest = bh; 3512a512bbf8SYan Zheng transid = btrfs_super_generation(super); 3513a512bbf8SYan Zheng } else { 3514a512bbf8SYan Zheng brelse(bh); 3515a512bbf8SYan Zheng } 3516a512bbf8SYan Zheng } 351792fc03fbSAnand Jain 351892fc03fbSAnand Jain if (!latest) 351992fc03fbSAnand Jain return ERR_PTR(ret); 352092fc03fbSAnand Jain 3521a512bbf8SYan Zheng return latest; 3522a512bbf8SYan Zheng } 3523a512bbf8SYan Zheng 35244eedeb75SHisashi Hifumi /* 3525abbb3b8eSDavid Sterba * Write superblock @sb to the @device. Do not wait for completion, all the 3526abbb3b8eSDavid Sterba * buffer heads we write are pinned. 35274eedeb75SHisashi Hifumi * 3528abbb3b8eSDavid Sterba * Write @max_mirrors copies of the superblock, where 0 means default that fit 3529abbb3b8eSDavid Sterba * the expected device size at commit time. Note that max_mirrors must be 3530abbb3b8eSDavid Sterba * same for write and wait phases. 35314eedeb75SHisashi Hifumi * 3532abbb3b8eSDavid Sterba * Return number of errors when buffer head is not found or submission fails. 35334eedeb75SHisashi Hifumi */ 3534a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 3535abbb3b8eSDavid Sterba struct btrfs_super_block *sb, int max_mirrors) 3536a512bbf8SYan Zheng { 3537d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = device->fs_info; 3538d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 3539a512bbf8SYan Zheng struct buffer_head *bh; 3540a512bbf8SYan Zheng int i; 3541a512bbf8SYan Zheng int ret; 3542a512bbf8SYan Zheng int errors = 0; 3543a512bbf8SYan Zheng u64 bytenr; 35441b9e619cSOmar Sandoval int op_flags; 3545a512bbf8SYan Zheng 3546a512bbf8SYan Zheng if (max_mirrors == 0) 3547a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3548a512bbf8SYan Zheng 3549d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 3550d5178578SJohannes Thumshirn 3551a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 3552a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 3553935e5cc9SMiao Xie if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3554935e5cc9SMiao Xie device->commit_total_bytes) 3555a512bbf8SYan Zheng break; 3556a512bbf8SYan Zheng 3557a512bbf8SYan Zheng btrfs_set_super_bytenr(sb, bytenr); 3558a512bbf8SYan Zheng 3559d5178578SJohannes Thumshirn crypto_shash_init(shash); 3560d5178578SJohannes Thumshirn crypto_shash_update(shash, (const char *)sb + BTRFS_CSUM_SIZE, 3561abbb3b8eSDavid Sterba BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE); 3562d5178578SJohannes Thumshirn crypto_shash_final(shash, sb->csum); 3563a512bbf8SYan Zheng 3564abbb3b8eSDavid Sterba /* One reference for us, and we leave it for the caller */ 35659f6d2510SDavid Sterba bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, 3566a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 3567634554dcSJosef Bacik if (!bh) { 3568fb456252SJeff Mahoney btrfs_err(device->fs_info, 3569f14d104dSDavid Sterba "couldn't get super buffer head for bytenr %llu", 3570f14d104dSDavid Sterba bytenr); 3571634554dcSJosef Bacik errors++; 3572634554dcSJosef Bacik continue; 3573634554dcSJosef Bacik } 3574634554dcSJosef Bacik 3575a512bbf8SYan Zheng memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE); 3576a512bbf8SYan Zheng 35774eedeb75SHisashi Hifumi /* one reference for submit_bh */ 3578a512bbf8SYan Zheng get_bh(bh); 35794eedeb75SHisashi Hifumi 35804eedeb75SHisashi Hifumi set_buffer_uptodate(bh); 3581a512bbf8SYan Zheng lock_buffer(bh); 3582a512bbf8SYan Zheng bh->b_end_io = btrfs_end_buffer_write_sync; 3583442a4f63SStefan Behrens bh->b_private = device; 3584a512bbf8SYan Zheng 3585387125fcSChris Mason /* 3586387125fcSChris Mason * we fua the first super. The others we allow 3587387125fcSChris Mason * to go down lazy. 3588387125fcSChris Mason */ 35891b9e619cSOmar Sandoval op_flags = REQ_SYNC | REQ_META | REQ_PRIO; 35901b9e619cSOmar Sandoval if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) 35911b9e619cSOmar Sandoval op_flags |= REQ_FUA; 35921b9e619cSOmar Sandoval ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh); 35934eedeb75SHisashi Hifumi if (ret) 3594a512bbf8SYan Zheng errors++; 3595a512bbf8SYan Zheng } 3596a512bbf8SYan Zheng return errors < i ? 0 : -1; 3597a512bbf8SYan Zheng } 3598a512bbf8SYan Zheng 3599387125fcSChris Mason /* 3600abbb3b8eSDavid Sterba * Wait for write completion of superblocks done by write_dev_supers, 3601abbb3b8eSDavid Sterba * @max_mirrors same for write and wait phases. 3602abbb3b8eSDavid Sterba * 3603abbb3b8eSDavid Sterba * Return number of errors when buffer head is not found or not marked up to 3604abbb3b8eSDavid Sterba * date. 3605abbb3b8eSDavid Sterba */ 3606abbb3b8eSDavid Sterba static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) 3607abbb3b8eSDavid Sterba { 3608abbb3b8eSDavid Sterba struct buffer_head *bh; 3609abbb3b8eSDavid Sterba int i; 3610abbb3b8eSDavid Sterba int errors = 0; 3611b6a535faSHoward McLauchlan bool primary_failed = false; 3612abbb3b8eSDavid Sterba u64 bytenr; 3613abbb3b8eSDavid Sterba 3614abbb3b8eSDavid Sterba if (max_mirrors == 0) 3615abbb3b8eSDavid Sterba max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3616abbb3b8eSDavid Sterba 3617abbb3b8eSDavid Sterba for (i = 0; i < max_mirrors; i++) { 3618abbb3b8eSDavid Sterba bytenr = btrfs_sb_offset(i); 3619abbb3b8eSDavid Sterba if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3620abbb3b8eSDavid Sterba device->commit_total_bytes) 3621abbb3b8eSDavid Sterba break; 3622abbb3b8eSDavid Sterba 36239f6d2510SDavid Sterba bh = __find_get_block(device->bdev, 36249f6d2510SDavid Sterba bytenr / BTRFS_BDEV_BLOCKSIZE, 3625abbb3b8eSDavid Sterba BTRFS_SUPER_INFO_SIZE); 3626abbb3b8eSDavid Sterba if (!bh) { 3627abbb3b8eSDavid Sterba errors++; 3628b6a535faSHoward McLauchlan if (i == 0) 3629b6a535faSHoward McLauchlan primary_failed = true; 3630abbb3b8eSDavid Sterba continue; 3631abbb3b8eSDavid Sterba } 3632abbb3b8eSDavid Sterba wait_on_buffer(bh); 3633b6a535faSHoward McLauchlan if (!buffer_uptodate(bh)) { 3634abbb3b8eSDavid Sterba errors++; 3635b6a535faSHoward McLauchlan if (i == 0) 3636b6a535faSHoward McLauchlan primary_failed = true; 3637b6a535faSHoward McLauchlan } 3638abbb3b8eSDavid Sterba 3639abbb3b8eSDavid Sterba /* drop our reference */ 3640abbb3b8eSDavid Sterba brelse(bh); 3641abbb3b8eSDavid Sterba 3642abbb3b8eSDavid Sterba /* drop the reference from the writing run */ 3643abbb3b8eSDavid Sterba brelse(bh); 3644abbb3b8eSDavid Sterba } 3645abbb3b8eSDavid Sterba 3646b6a535faSHoward McLauchlan /* log error, force error return */ 3647b6a535faSHoward McLauchlan if (primary_failed) { 3648b6a535faSHoward McLauchlan btrfs_err(device->fs_info, "error writing primary super block to device %llu", 3649b6a535faSHoward McLauchlan device->devid); 3650b6a535faSHoward McLauchlan return -1; 3651b6a535faSHoward McLauchlan } 3652b6a535faSHoward McLauchlan 3653abbb3b8eSDavid Sterba return errors < i ? 0 : -1; 3654abbb3b8eSDavid Sterba } 3655abbb3b8eSDavid Sterba 3656abbb3b8eSDavid Sterba /* 3657387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 3658387125fcSChris Mason * for the barrier when it is done 3659387125fcSChris Mason */ 36604246a0b6SChristoph Hellwig static void btrfs_end_empty_barrier(struct bio *bio) 3661387125fcSChris Mason { 3662387125fcSChris Mason complete(bio->bi_private); 3663387125fcSChris Mason } 3664387125fcSChris Mason 3665387125fcSChris Mason /* 36664fc6441aSAnand Jain * Submit a flush request to the device if it supports it. Error handling is 36674fc6441aSAnand Jain * done in the waiting counterpart. 3668387125fcSChris Mason */ 36694fc6441aSAnand Jain static void write_dev_flush(struct btrfs_device *device) 3670387125fcSChris Mason { 3671c2a9c7abSAnand Jain struct request_queue *q = bdev_get_queue(device->bdev); 3672e0ae9994SDavid Sterba struct bio *bio = device->flush_bio; 3673387125fcSChris Mason 3674c2a9c7abSAnand Jain if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) 36754fc6441aSAnand Jain return; 3676387125fcSChris Mason 3677e0ae9994SDavid Sterba bio_reset(bio); 3678387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 367974d46992SChristoph Hellwig bio_set_dev(bio, device->bdev); 36808d910125SJan Kara bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH; 3681387125fcSChris Mason init_completion(&device->flush_wait); 3682387125fcSChris Mason bio->bi_private = &device->flush_wait; 3683387125fcSChris Mason 368443a01111SLu Fengqi btrfsic_submit_bio(bio); 36851c3063b6SAnand Jain set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 36864fc6441aSAnand Jain } 3687387125fcSChris Mason 36884fc6441aSAnand Jain /* 36894fc6441aSAnand Jain * If the flush bio has been submitted by write_dev_flush, wait for it. 36904fc6441aSAnand Jain */ 36918c27cb35SLinus Torvalds static blk_status_t wait_dev_flush(struct btrfs_device *device) 36924fc6441aSAnand Jain { 36934fc6441aSAnand Jain struct bio *bio = device->flush_bio; 36944fc6441aSAnand Jain 36951c3063b6SAnand Jain if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) 369658efbc9fSOmar Sandoval return BLK_STS_OK; 36974fc6441aSAnand Jain 36981c3063b6SAnand Jain clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 36992980d574SDavid Sterba wait_for_completion_io(&device->flush_wait); 37004fc6441aSAnand Jain 37018c27cb35SLinus Torvalds return bio->bi_status; 3702387125fcSChris Mason } 3703387125fcSChris Mason 3704d10b82feSQu Wenruo static int check_barrier_error(struct btrfs_fs_info *fs_info) 3705401b41e5SAnand Jain { 37066528b99dSAnand Jain if (!btrfs_check_rw_degradable(fs_info, NULL)) 3707401b41e5SAnand Jain return -EIO; 3708387125fcSChris Mason return 0; 3709387125fcSChris Mason } 3710387125fcSChris Mason 3711387125fcSChris Mason /* 3712387125fcSChris Mason * send an empty flush down to each device in parallel, 3713387125fcSChris Mason * then wait for them 3714387125fcSChris Mason */ 3715387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 3716387125fcSChris Mason { 3717387125fcSChris Mason struct list_head *head; 3718387125fcSChris Mason struct btrfs_device *dev; 37195af3e8ccSStefan Behrens int errors_wait = 0; 37204e4cbee9SChristoph Hellwig blk_status_t ret; 3721387125fcSChris Mason 37221538e6c5SDavid Sterba lockdep_assert_held(&info->fs_devices->device_list_mutex); 3723387125fcSChris Mason /* send down all the barriers */ 3724387125fcSChris Mason head = &info->fs_devices->devices; 37251538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3726e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3727f88ba6a2SHidetoshi Seto continue; 3728cea7c8bfSAnand Jain if (!dev->bdev) 3729387125fcSChris Mason continue; 3730e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3731ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3732387125fcSChris Mason continue; 3733387125fcSChris Mason 37344fc6441aSAnand Jain write_dev_flush(dev); 373558efbc9fSOmar Sandoval dev->last_flush_error = BLK_STS_OK; 3736387125fcSChris Mason } 3737387125fcSChris Mason 3738387125fcSChris Mason /* wait for all the barriers */ 37391538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3740e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3741f88ba6a2SHidetoshi Seto continue; 3742387125fcSChris Mason if (!dev->bdev) { 37435af3e8ccSStefan Behrens errors_wait++; 3744387125fcSChris Mason continue; 3745387125fcSChris Mason } 3746e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3747ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3748387125fcSChris Mason continue; 3749387125fcSChris Mason 37504fc6441aSAnand Jain ret = wait_dev_flush(dev); 3751401b41e5SAnand Jain if (ret) { 3752401b41e5SAnand Jain dev->last_flush_error = ret; 375366b4993eSDavid Sterba btrfs_dev_stat_inc_and_print(dev, 375466b4993eSDavid Sterba BTRFS_DEV_STAT_FLUSH_ERRS); 37555af3e8ccSStefan Behrens errors_wait++; 3756387125fcSChris Mason } 3757401b41e5SAnand Jain } 3758401b41e5SAnand Jain 3759cea7c8bfSAnand Jain if (errors_wait) { 3760401b41e5SAnand Jain /* 3761401b41e5SAnand Jain * At some point we need the status of all disks 3762401b41e5SAnand Jain * to arrive at the volume status. So error checking 3763401b41e5SAnand Jain * is being pushed to a separate loop. 3764401b41e5SAnand Jain */ 3765d10b82feSQu Wenruo return check_barrier_error(info); 3766401b41e5SAnand Jain } 3767387125fcSChris Mason return 0; 3768387125fcSChris Mason } 3769387125fcSChris Mason 3770943c6e99SZhao Lei int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 3771943c6e99SZhao Lei { 37728789f4feSZhao Lei int raid_type; 37738789f4feSZhao Lei int min_tolerated = INT_MAX; 3774943c6e99SZhao Lei 37758789f4feSZhao Lei if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 37768789f4feSZhao Lei (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 37778c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 37788789f4feSZhao Lei btrfs_raid_array[BTRFS_RAID_SINGLE]. 37798789f4feSZhao Lei tolerated_failures); 3780943c6e99SZhao Lei 37818789f4feSZhao Lei for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 37828789f4feSZhao Lei if (raid_type == BTRFS_RAID_SINGLE) 37838789f4feSZhao Lei continue; 378441a6e891SAnand Jain if (!(flags & btrfs_raid_array[raid_type].bg_flag)) 37858789f4feSZhao Lei continue; 37868c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 37878789f4feSZhao Lei btrfs_raid_array[raid_type]. 37888789f4feSZhao Lei tolerated_failures); 37898789f4feSZhao Lei } 3790943c6e99SZhao Lei 37918789f4feSZhao Lei if (min_tolerated == INT_MAX) { 3792ab8d0fc4SJeff Mahoney pr_warn("BTRFS: unknown raid flag: %llu", flags); 37938789f4feSZhao Lei min_tolerated = 0; 37948789f4feSZhao Lei } 37958789f4feSZhao Lei 37968789f4feSZhao Lei return min_tolerated; 3797943c6e99SZhao Lei } 3798943c6e99SZhao Lei 3799eece6a9cSDavid Sterba int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) 3800f2984462SChris Mason { 3801e5e9a520SChris Mason struct list_head *head; 3802f2984462SChris Mason struct btrfs_device *dev; 3803a061fc8dSChris Mason struct btrfs_super_block *sb; 3804f2984462SChris Mason struct btrfs_dev_item *dev_item; 3805f2984462SChris Mason int ret; 3806f2984462SChris Mason int do_barriers; 3807a236aed1SChris Mason int max_errors; 3808a236aed1SChris Mason int total_errors = 0; 3809a061fc8dSChris Mason u64 flags; 3810f2984462SChris Mason 38110b246afaSJeff Mahoney do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); 3812fed3b381SLiu Bo 3813fed3b381SLiu Bo /* 3814fed3b381SLiu Bo * max_mirrors == 0 indicates we're from commit_transaction, 3815fed3b381SLiu Bo * not from fsync where the tree roots in fs_info have not 3816fed3b381SLiu Bo * been consistent on disk. 3817fed3b381SLiu Bo */ 3818fed3b381SLiu Bo if (max_mirrors == 0) 38190b246afaSJeff Mahoney backup_super_roots(fs_info); 3820f2984462SChris Mason 38210b246afaSJeff Mahoney sb = fs_info->super_for_commit; 3822a061fc8dSChris Mason dev_item = &sb->dev_item; 3823e5e9a520SChris Mason 38240b246afaSJeff Mahoney mutex_lock(&fs_info->fs_devices->device_list_mutex); 38250b246afaSJeff Mahoney head = &fs_info->fs_devices->devices; 38260b246afaSJeff Mahoney max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; 3827387125fcSChris Mason 38285af3e8ccSStefan Behrens if (do_barriers) { 38290b246afaSJeff Mahoney ret = barrier_all_devices(fs_info); 38305af3e8ccSStefan Behrens if (ret) { 38315af3e8ccSStefan Behrens mutex_unlock( 38320b246afaSJeff Mahoney &fs_info->fs_devices->device_list_mutex); 38330b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 38345af3e8ccSStefan Behrens "errors while submitting device barriers."); 38355af3e8ccSStefan Behrens return ret; 38365af3e8ccSStefan Behrens } 38375af3e8ccSStefan Behrens } 3838387125fcSChris Mason 38391538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3840dfe25020SChris Mason if (!dev->bdev) { 3841dfe25020SChris Mason total_errors++; 3842dfe25020SChris Mason continue; 3843dfe25020SChris Mason } 3844e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3845ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3846dfe25020SChris Mason continue; 3847dfe25020SChris Mason 38482b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 3849a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 3850a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 38517df69d3eSMiao Xie btrfs_set_stack_device_total_bytes(dev_item, 3852935e5cc9SMiao Xie dev->commit_total_bytes); 3853ce7213c7SMiao Xie btrfs_set_stack_device_bytes_used(dev_item, 3854ce7213c7SMiao Xie dev->commit_bytes_used); 3855a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 3856a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 3857a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 3858a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 38597239ff4bSNikolay Borisov memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, 38607239ff4bSNikolay Borisov BTRFS_FSID_SIZE); 3861a512bbf8SYan Zheng 3862a061fc8dSChris Mason flags = btrfs_super_flags(sb); 3863a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 3864f2984462SChris Mason 386575cb857dSQu Wenruo ret = btrfs_validate_write_super(fs_info, sb); 386675cb857dSQu Wenruo if (ret < 0) { 386775cb857dSQu Wenruo mutex_unlock(&fs_info->fs_devices->device_list_mutex); 386875cb857dSQu Wenruo btrfs_handle_fs_error(fs_info, -EUCLEAN, 386975cb857dSQu Wenruo "unexpected superblock corruption detected"); 387075cb857dSQu Wenruo return -EUCLEAN; 387175cb857dSQu Wenruo } 387275cb857dSQu Wenruo 3873abbb3b8eSDavid Sterba ret = write_dev_supers(dev, sb, max_mirrors); 3874a236aed1SChris Mason if (ret) 3875a236aed1SChris Mason total_errors++; 3876f2984462SChris Mason } 3877a236aed1SChris Mason if (total_errors > max_errors) { 38780b246afaSJeff Mahoney btrfs_err(fs_info, "%d errors while writing supers", 3879d397712bSChris Mason total_errors); 38800b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 388179787eaaSJeff Mahoney 38829d565ba4SStefan Behrens /* FUA is masked off if unsupported and can't be the reason */ 38830b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 38840b246afaSJeff Mahoney "%d errors while writing supers", 38850b246afaSJeff Mahoney total_errors); 38869d565ba4SStefan Behrens return -EIO; 3887a236aed1SChris Mason } 3888f2984462SChris Mason 3889a512bbf8SYan Zheng total_errors = 0; 38901538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3891dfe25020SChris Mason if (!dev->bdev) 3892dfe25020SChris Mason continue; 3893e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3894ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3895dfe25020SChris Mason continue; 3896dfe25020SChris Mason 3897abbb3b8eSDavid Sterba ret = wait_dev_supers(dev, max_mirrors); 3898a512bbf8SYan Zheng if (ret) 38991259ab75SChris Mason total_errors++; 3900f2984462SChris Mason } 39010b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 3902a236aed1SChris Mason if (total_errors > max_errors) { 39030b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 39040b246afaSJeff Mahoney "%d errors while writing supers", 39050b246afaSJeff Mahoney total_errors); 390679787eaaSJeff Mahoney return -EIO; 3907a236aed1SChris Mason } 3908f2984462SChris Mason return 0; 3909f2984462SChris Mason } 3910f2984462SChris Mason 3911cb517eabSMiao Xie /* Drop a fs root from the radix tree and free it. */ 3912cb517eabSMiao Xie void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 3913cb517eabSMiao Xie struct btrfs_root *root) 39142619ba1fSChris Mason { 39154df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 39162619ba1fSChris Mason radix_tree_delete(&fs_info->fs_roots_radix, 39172619ba1fSChris Mason (unsigned long)root->root_key.objectid); 39184df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 391976dda93cSYan, Zheng 392076dda93cSYan, Zheng if (btrfs_root_refs(&root->root_item) == 0) 392176dda93cSYan, Zheng synchronize_srcu(&fs_info->subvol_srcu); 392276dda93cSYan, Zheng 39231c1ea4f7SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 39243321719eSLiu Bo btrfs_free_log(NULL, root); 39251c1ea4f7SLiu Bo if (root->reloc_root) { 39261c1ea4f7SLiu Bo free_extent_buffer(root->reloc_root->node); 39271c1ea4f7SLiu Bo free_extent_buffer(root->reloc_root->commit_root); 39281c1ea4f7SLiu Bo btrfs_put_fs_root(root->reloc_root); 39291c1ea4f7SLiu Bo root->reloc_root = NULL; 39301c1ea4f7SLiu Bo } 39311c1ea4f7SLiu Bo } 39323321719eSLiu Bo 3933faa2dbf0SJosef Bacik if (root->free_ino_pinned) 3934581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_pinned); 3935faa2dbf0SJosef Bacik if (root->free_ino_ctl) 3936581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_ctl); 393784db5ccfSDavid Sterba btrfs_free_fs_root(root); 39384df27c4dSYan, Zheng } 39394df27c4dSYan, Zheng 394084db5ccfSDavid Sterba void btrfs_free_fs_root(struct btrfs_root *root) 39414df27c4dSYan, Zheng { 394257cdc8dbSDavid Sterba iput(root->ino_cache_inode); 39434df27c4dSYan, Zheng WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 39440ee5dc67SAl Viro if (root->anon_dev) 39450ee5dc67SAl Viro free_anon_bdev(root->anon_dev); 39468257b2dcSMiao Xie if (root->subv_writers) 39478257b2dcSMiao Xie btrfs_free_subvolume_writers(root->subv_writers); 39485f39d397SChris Mason free_extent_buffer(root->node); 39495f39d397SChris Mason free_extent_buffer(root->commit_root); 3950581bb050SLi Zefan kfree(root->free_ino_ctl); 3951581bb050SLi Zefan kfree(root->free_ino_pinned); 3952b0feb9d9SMiao Xie btrfs_put_fs_root(root); 39532619ba1fSChris Mason } 39542619ba1fSChris Mason 3955c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 3956c146afadSYan Zheng { 3957c146afadSYan Zheng u64 root_objectid = 0; 3958c146afadSYan Zheng struct btrfs_root *gang[8]; 395965d33fd7SQu Wenruo int i = 0; 396065d33fd7SQu Wenruo int err = 0; 396165d33fd7SQu Wenruo unsigned int ret = 0; 396265d33fd7SQu Wenruo int index; 3963c146afadSYan Zheng 3964c146afadSYan Zheng while (1) { 396565d33fd7SQu Wenruo index = srcu_read_lock(&fs_info->subvol_srcu); 3966c146afadSYan Zheng ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 3967c146afadSYan Zheng (void **)gang, root_objectid, 3968c146afadSYan Zheng ARRAY_SIZE(gang)); 396965d33fd7SQu Wenruo if (!ret) { 397065d33fd7SQu Wenruo srcu_read_unlock(&fs_info->subvol_srcu, index); 3971c146afadSYan Zheng break; 397265d33fd7SQu Wenruo } 39735d4f98a2SYan Zheng root_objectid = gang[ret - 1]->root_key.objectid + 1; 397466b4ffd1SJosef Bacik 397565d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 397665d33fd7SQu Wenruo /* Avoid to grab roots in dead_roots */ 397765d33fd7SQu Wenruo if (btrfs_root_refs(&gang[i]->root_item) == 0) { 397865d33fd7SQu Wenruo gang[i] = NULL; 397965d33fd7SQu Wenruo continue; 398065d33fd7SQu Wenruo } 398165d33fd7SQu Wenruo /* grab all the search result for later use */ 398265d33fd7SQu Wenruo gang[i] = btrfs_grab_fs_root(gang[i]); 398365d33fd7SQu Wenruo } 398465d33fd7SQu Wenruo srcu_read_unlock(&fs_info->subvol_srcu, index); 398565d33fd7SQu Wenruo 398665d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 398765d33fd7SQu Wenruo if (!gang[i]) 398865d33fd7SQu Wenruo continue; 3989c146afadSYan Zheng root_objectid = gang[i]->root_key.objectid; 399066b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(gang[i]); 399166b4ffd1SJosef Bacik if (err) 399265d33fd7SQu Wenruo break; 399365d33fd7SQu Wenruo btrfs_put_fs_root(gang[i]); 3994c146afadSYan Zheng } 3995c146afadSYan Zheng root_objectid++; 3996c146afadSYan Zheng } 399765d33fd7SQu Wenruo 399865d33fd7SQu Wenruo /* release the uncleaned roots due to error */ 399965d33fd7SQu Wenruo for (; i < ret; i++) { 400065d33fd7SQu Wenruo if (gang[i]) 400165d33fd7SQu Wenruo btrfs_put_fs_root(gang[i]); 400265d33fd7SQu Wenruo } 400365d33fd7SQu Wenruo return err; 4004c146afadSYan Zheng } 4005c146afadSYan Zheng 40066bccf3abSJeff Mahoney int btrfs_commit_super(struct btrfs_fs_info *fs_info) 4007c146afadSYan Zheng { 40086bccf3abSJeff Mahoney struct btrfs_root *root = fs_info->tree_root; 4009c146afadSYan Zheng struct btrfs_trans_handle *trans; 4010c146afadSYan Zheng 40110b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 40122ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 40130b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 40140b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 4015c71bf099SYan, Zheng 4016c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 40170b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 40180b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4019c71bf099SYan, Zheng 40207a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 40213612b495STsutomu Itoh if (IS_ERR(trans)) 40223612b495STsutomu Itoh return PTR_ERR(trans); 40233a45bb20SJeff Mahoney return btrfs_commit_transaction(trans); 4024c146afadSYan Zheng } 4025c146afadSYan Zheng 4026b105e927SDavid Sterba void __cold close_ctree(struct btrfs_fs_info *fs_info) 4027eb60ceacSChris Mason { 4028c146afadSYan Zheng int ret; 4029e089f05cSChris Mason 4030afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); 4031d6fd0ae2SOmar Sandoval /* 4032d6fd0ae2SOmar Sandoval * We don't want the cleaner to start new transactions, add more delayed 4033d6fd0ae2SOmar Sandoval * iputs, etc. while we're closing. We can't use kthread_stop() yet 4034d6fd0ae2SOmar Sandoval * because that frees the task_struct, and the transaction kthread might 4035d6fd0ae2SOmar Sandoval * still try to wake up the cleaner. 4036d6fd0ae2SOmar Sandoval */ 4037d6fd0ae2SOmar Sandoval kthread_park(fs_info->cleaner_kthread); 4038a2135011SChris Mason 40397343dd61SJustin Maggard /* wait for the qgroup rescan worker to stop */ 4040d06f23d6SJeff Mahoney btrfs_qgroup_wait_for_completion(fs_info, false); 40417343dd61SJustin Maggard 4042803b2f54SStefan Behrens /* wait for the uuid_scan task to finish */ 4043803b2f54SStefan Behrens down(&fs_info->uuid_tree_rescan_sem); 4044803b2f54SStefan Behrens /* avoid complains from lockdep et al., set sem back to initial state */ 4045803b2f54SStefan Behrens up(&fs_info->uuid_tree_rescan_sem); 4046803b2f54SStefan Behrens 4047837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 4048aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 4049837d5b6eSIlya Dryomov 40508dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 40518dabb742SStefan Behrens 4052aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 40534cb5300bSChris Mason 40544cb5300bSChris Mason /* wait for any defraggers to finish */ 40554cb5300bSChris Mason wait_event(fs_info->transaction_wait, 40564cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 40574cb5300bSChris Mason 40584cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 405926176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 40604cb5300bSChris Mason 406121c7e756SMiao Xie cancel_work_sync(&fs_info->async_reclaim_work); 406221c7e756SMiao Xie 4063bc98a42cSDavid Howells if (!sb_rdonly(fs_info->sb)) { 4064e44163e1SJeff Mahoney /* 4065d6fd0ae2SOmar Sandoval * The cleaner kthread is stopped, so do one final pass over 4066d6fd0ae2SOmar Sandoval * unused block groups. 4067e44163e1SJeff Mahoney */ 40680b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 4069e44163e1SJeff Mahoney 40706bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 4071d397712bSChris Mason if (ret) 407204892340SEric Sandeen btrfs_err(fs_info, "commit super ret %d", ret); 4073c146afadSYan Zheng } 4074ed2ff2cbSChris Mason 4075af722733SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) || 4076af722733SLiu Bo test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) 40772ff7e61eSJeff Mahoney btrfs_error_commit_super(fs_info); 4078acce952bSliubo 4079e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 4080e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 40818929ecfaSYan, Zheng 4082e187831eSJosef Bacik ASSERT(list_empty(&fs_info->delayed_iputs)); 4083afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); 4084f25784b3SYan Zheng 408504892340SEric Sandeen btrfs_free_qgroup_config(fs_info); 4086fe816d0fSNikolay Borisov ASSERT(list_empty(&fs_info->delalloc_roots)); 4087bcef60f2SArne Jansen 4088963d678bSMiao Xie if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 408904892340SEric Sandeen btrfs_info(fs_info, "at unmount delalloc count %lld", 4090963d678bSMiao Xie percpu_counter_sum(&fs_info->delalloc_bytes)); 4091b0c68f8bSChris Mason } 409231153d81SYan Zheng 40934297ff84SJosef Bacik if (percpu_counter_sum(&fs_info->dio_bytes)) 40944297ff84SJosef Bacik btrfs_info(fs_info, "at unmount dio bytes count %lld", 40954297ff84SJosef Bacik percpu_counter_sum(&fs_info->dio_bytes)); 40964297ff84SJosef Bacik 40976618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 4098b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 40995ac1d209SJeff Mahoney 4100faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 4101d10c5f31SChris Mason 41021a4319ccSLiu Bo btrfs_put_block_group_cache(fs_info); 41031a4319ccSLiu Bo 4104de348ee0SWang Shilong /* 4105de348ee0SWang Shilong * we must make sure there is not any read request to 4106de348ee0SWang Shilong * submit after we stopping all workers. 4107de348ee0SWang Shilong */ 4108de348ee0SWang Shilong invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 410996192499SJosef Bacik btrfs_stop_all_workers(fs_info); 411096192499SJosef Bacik 41115cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 41125cdd7db6SFilipe Manana 4113afcdd129SJosef Bacik clear_bit(BTRFS_FS_OPEN, &fs_info->flags); 41144273eaffSAnand Jain free_root_pointers(fs_info, true); 41159ad6b7bcSChris Mason 411613e6c37bSJosef Bacik iput(fs_info->btree_inode); 4117d6bfde87SChris Mason 411821adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 41190b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) 41202ff7e61eSJeff Mahoney btrfsic_unmount(fs_info->fs_devices); 412121adbd5cSStefan Behrens #endif 412221adbd5cSStefan Behrens 41230b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 412468c94e55SNikolay Borisov btrfs_close_devices(fs_info->fs_devices); 4125b248a415SChris Mason 4126e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 4127963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 41284297ff84SJosef Bacik percpu_counter_destroy(&fs_info->dio_bytes); 41297f8d236aSDavid Sterba percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 413076dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 41310b86a832SChris Mason 4132bfcea1c6SJohannes Thumshirn btrfs_free_csum_hash(fs_info); 413353b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 4134fd708b81SJosef Bacik btrfs_free_ref_cache(fs_info); 4135eb60ceacSChris Mason } 4136eb60ceacSChris Mason 4137b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 4138b9fab919SChris Mason int atomic) 4139ccd467d6SChris Mason { 41401259ab75SChris Mason int ret; 4141727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 41421259ab75SChris Mason 41430b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 41441259ab75SChris Mason if (!ret) 41451259ab75SChris Mason return ret; 41461259ab75SChris Mason 41471259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 4148b9fab919SChris Mason parent_transid, atomic); 4149b9fab919SChris Mason if (ret == -EAGAIN) 4150b9fab919SChris Mason return ret; 41511259ab75SChris Mason return !ret; 41525f39d397SChris Mason } 41536702ed49SChris Mason 41545f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 41555f39d397SChris Mason { 41560b246afaSJeff Mahoney struct btrfs_fs_info *fs_info; 415706ea65a3SJosef Bacik struct btrfs_root *root; 41585f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 4159b9473439SChris Mason int was_dirty; 4160b4ce94deSChris Mason 416106ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 416206ea65a3SJosef Bacik /* 416306ea65a3SJosef Bacik * This is a fast path so only do this check if we have sanity tests 416452042d8eSAndrea Gelmini * enabled. Normal people shouldn't be using unmapped buffers as dirty 416506ea65a3SJosef Bacik * outside of the sanity tests. 416606ea65a3SJosef Bacik */ 4167b0132a3bSNikolay Borisov if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) 416806ea65a3SJosef Bacik return; 416906ea65a3SJosef Bacik #endif 417006ea65a3SJosef Bacik root = BTRFS_I(buf->pages[0]->mapping->host)->root; 41710b246afaSJeff Mahoney fs_info = root->fs_info; 4172b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 41730b246afaSJeff Mahoney if (transid != fs_info->generation) 41745d163e0eSJeff Mahoney WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", 41750b246afaSJeff Mahoney buf->start, transid, fs_info->generation); 41760b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 4177e2d84521SMiao Xie if (!was_dirty) 4178104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 4179e2d84521SMiao Xie buf->len, 41800b246afaSJeff Mahoney fs_info->dirty_metadata_batch); 41811f21ef0aSFilipe Manana #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 418269fc6cbbSQu Wenruo /* 418369fc6cbbSQu Wenruo * Since btrfs_mark_buffer_dirty() can be called with item pointer set 418469fc6cbbSQu Wenruo * but item data not updated. 418569fc6cbbSQu Wenruo * So here we should only check item pointers, not item data. 418669fc6cbbSQu Wenruo */ 418769fc6cbbSQu Wenruo if (btrfs_header_level(buf) == 0 && 4188cfdaad5eSDavid Sterba btrfs_check_leaf_relaxed(buf)) { 4189a4f78750SDavid Sterba btrfs_print_leaf(buf); 41901f21ef0aSFilipe Manana ASSERT(0); 41911f21ef0aSFilipe Manana } 41921f21ef0aSFilipe Manana #endif 4193eb60ceacSChris Mason } 4194eb60ceacSChris Mason 41952ff7e61eSJeff Mahoney static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, 4196b53d3f5dSLiu Bo int flush_delayed) 419735b7e476SChris Mason { 4198188de649SChris Mason /* 4199188de649SChris Mason * looks as though older kernels can get into trouble with 4200188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 4201188de649SChris Mason */ 4202e2d84521SMiao Xie int ret; 4203d6bfde87SChris Mason 42046933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 4205d6bfde87SChris Mason return; 4206d6bfde87SChris Mason 4207b53d3f5dSLiu Bo if (flush_delayed) 42082ff7e61eSJeff Mahoney btrfs_balance_delayed_items(fs_info); 420916cdcec7SMiao Xie 4210d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 4211d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 4212d814a491SEthan Lien fs_info->dirty_metadata_batch); 4213e2d84521SMiao Xie if (ret > 0) { 42140b246afaSJeff Mahoney balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); 421516cdcec7SMiao Xie } 421616cdcec7SMiao Xie } 421716cdcec7SMiao Xie 42182ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) 421916cdcec7SMiao Xie { 42202ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 1); 422135b7e476SChris Mason } 4222b53d3f5dSLiu Bo 42232ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) 4224b53d3f5dSLiu Bo { 42252ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 0); 4226d6bfde87SChris Mason } 42276b80053dSChris Mason 4228581c1760SQu Wenruo int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level, 4229581c1760SQu Wenruo struct btrfs_key *first_key) 42306b80053dSChris Mason { 42315ab12d1fSDavid Sterba return btree_read_extent_buffer_pages(buf, parent_transid, 4232581c1760SQu Wenruo level, first_key); 42336b80053dSChris Mason } 42340da5468fSChris Mason 42352ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) 4236acce952bSliubo { 4237fe816d0fSNikolay Borisov /* cleanup FS via transaction */ 4238fe816d0fSNikolay Borisov btrfs_cleanup_transaction(fs_info); 4239fe816d0fSNikolay Borisov 42400b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 42412ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 42420b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 4243acce952bSliubo 42440b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 42450b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4246acce952bSliubo } 4247acce952bSliubo 4248143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4249acce952bSliubo { 4250acce952bSliubo struct btrfs_ordered_extent *ordered; 4251acce952bSliubo 4252199c2a9cSMiao Xie spin_lock(&root->ordered_extent_lock); 4253779880efSJosef Bacik /* 4254779880efSJosef Bacik * This will just short circuit the ordered completion stuff which will 4255779880efSJosef Bacik * make sure the ordered extent gets properly cleaned up. 4256779880efSJosef Bacik */ 4257199c2a9cSMiao Xie list_for_each_entry(ordered, &root->ordered_extents, 4258779880efSJosef Bacik root_extent_list) 4259779880efSJosef Bacik set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4260199c2a9cSMiao Xie spin_unlock(&root->ordered_extent_lock); 4261199c2a9cSMiao Xie } 4262199c2a9cSMiao Xie 4263199c2a9cSMiao Xie static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4264199c2a9cSMiao Xie { 4265199c2a9cSMiao Xie struct btrfs_root *root; 4266199c2a9cSMiao Xie struct list_head splice; 4267199c2a9cSMiao Xie 4268199c2a9cSMiao Xie INIT_LIST_HEAD(&splice); 4269199c2a9cSMiao Xie 4270199c2a9cSMiao Xie spin_lock(&fs_info->ordered_root_lock); 4271199c2a9cSMiao Xie list_splice_init(&fs_info->ordered_roots, &splice); 4272199c2a9cSMiao Xie while (!list_empty(&splice)) { 4273199c2a9cSMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4274199c2a9cSMiao Xie ordered_root); 42751de2cfdeSJosef Bacik list_move_tail(&root->ordered_root, 42761de2cfdeSJosef Bacik &fs_info->ordered_roots); 4277199c2a9cSMiao Xie 42782a85d9caSLiu Bo spin_unlock(&fs_info->ordered_root_lock); 4279199c2a9cSMiao Xie btrfs_destroy_ordered_extents(root); 4280199c2a9cSMiao Xie 42812a85d9caSLiu Bo cond_resched(); 42822a85d9caSLiu Bo spin_lock(&fs_info->ordered_root_lock); 4283199c2a9cSMiao Xie } 4284199c2a9cSMiao Xie spin_unlock(&fs_info->ordered_root_lock); 428574d5d229SJosef Bacik 428674d5d229SJosef Bacik /* 428774d5d229SJosef Bacik * We need this here because if we've been flipped read-only we won't 428874d5d229SJosef Bacik * get sync() from the umount, so we need to make sure any ordered 428974d5d229SJosef Bacik * extents that haven't had their dirty pages IO start writeout yet 429074d5d229SJosef Bacik * actually get run and error out properly. 429174d5d229SJosef Bacik */ 429274d5d229SJosef Bacik btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 4293acce952bSliubo } 4294acce952bSliubo 429535a3621bSStefan Behrens static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 42962ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4297acce952bSliubo { 4298acce952bSliubo struct rb_node *node; 4299acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 4300acce952bSliubo struct btrfs_delayed_ref_node *ref; 4301acce952bSliubo int ret = 0; 4302acce952bSliubo 4303acce952bSliubo delayed_refs = &trans->delayed_refs; 4304acce952bSliubo 4305acce952bSliubo spin_lock(&delayed_refs->lock); 4306d7df2c79SJosef Bacik if (atomic_read(&delayed_refs->num_entries) == 0) { 4307cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 43080b246afaSJeff Mahoney btrfs_info(fs_info, "delayed_refs has NO entry"); 4309acce952bSliubo return ret; 4310acce952bSliubo } 4311acce952bSliubo 43125c9d028bSLiu Bo while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { 4313d7df2c79SJosef Bacik struct btrfs_delayed_ref_head *head; 43140e0adbcfSJosef Bacik struct rb_node *n; 4315e78417d1SJosef Bacik bool pin_bytes = false; 4316acce952bSliubo 4317d7df2c79SJosef Bacik head = rb_entry(node, struct btrfs_delayed_ref_head, 4318d7df2c79SJosef Bacik href_node); 43193069bd26SJosef Bacik if (btrfs_delayed_ref_lock(delayed_refs, head)) 4320b939d1abSJosef Bacik continue; 43213069bd26SJosef Bacik 4322d7df2c79SJosef Bacik spin_lock(&head->lock); 4323e3d03965SLiu Bo while ((n = rb_first_cached(&head->ref_tree)) != NULL) { 43240e0adbcfSJosef Bacik ref = rb_entry(n, struct btrfs_delayed_ref_node, 43250e0adbcfSJosef Bacik ref_node); 4326d7df2c79SJosef Bacik ref->in_tree = 0; 4327e3d03965SLiu Bo rb_erase_cached(&ref->ref_node, &head->ref_tree); 43280e0adbcfSJosef Bacik RB_CLEAR_NODE(&ref->ref_node); 43291d57ee94SWang Xiaoguang if (!list_empty(&ref->add_list)) 43301d57ee94SWang Xiaoguang list_del(&ref->add_list); 4331d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4332d7df2c79SJosef Bacik btrfs_put_delayed_ref(ref); 4333d7df2c79SJosef Bacik } 433454067ae9SJosef Bacik if (head->must_insert_reserved) 4335e78417d1SJosef Bacik pin_bytes = true; 433678a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 4337fa781ceaSJosef Bacik btrfs_delete_ref_head(delayed_refs, head); 4338d7df2c79SJosef Bacik spin_unlock(&head->lock); 4339acce952bSliubo spin_unlock(&delayed_refs->lock); 4340e78417d1SJosef Bacik mutex_unlock(&head->mutex); 4341acce952bSliubo 4342d7df2c79SJosef Bacik if (pin_bytes) 4343d278850eSJosef Bacik btrfs_pin_extent(fs_info, head->bytenr, 4344d278850eSJosef Bacik head->num_bytes, 1); 434531890da0SJosef Bacik btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); 4346d278850eSJosef Bacik btrfs_put_delayed_ref_head(head); 4347acce952bSliubo cond_resched(); 4348acce952bSliubo spin_lock(&delayed_refs->lock); 4349acce952bSliubo } 4350acce952bSliubo 4351acce952bSliubo spin_unlock(&delayed_refs->lock); 4352acce952bSliubo 4353acce952bSliubo return ret; 4354acce952bSliubo } 4355acce952bSliubo 4356143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 4357acce952bSliubo { 4358acce952bSliubo struct btrfs_inode *btrfs_inode; 4359acce952bSliubo struct list_head splice; 4360acce952bSliubo 4361acce952bSliubo INIT_LIST_HEAD(&splice); 4362acce952bSliubo 4363eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4364eb73c1b7SMiao Xie list_splice_init(&root->delalloc_inodes, &splice); 4365acce952bSliubo 4366acce952bSliubo while (!list_empty(&splice)) { 4367fe816d0fSNikolay Borisov struct inode *inode = NULL; 4368eb73c1b7SMiao Xie btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 4369acce952bSliubo delalloc_inodes); 4370fe816d0fSNikolay Borisov __btrfs_del_delalloc_inode(root, btrfs_inode); 4371eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4372acce952bSliubo 4373fe816d0fSNikolay Borisov /* 4374fe816d0fSNikolay Borisov * Make sure we get a live inode and that it'll not disappear 4375fe816d0fSNikolay Borisov * meanwhile. 4376fe816d0fSNikolay Borisov */ 4377fe816d0fSNikolay Borisov inode = igrab(&btrfs_inode->vfs_inode); 4378fe816d0fSNikolay Borisov if (inode) { 4379fe816d0fSNikolay Borisov invalidate_inode_pages2(inode->i_mapping); 4380fe816d0fSNikolay Borisov iput(inode); 4381fe816d0fSNikolay Borisov } 4382eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4383acce952bSliubo } 4384eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4385eb73c1b7SMiao Xie } 4386eb73c1b7SMiao Xie 4387eb73c1b7SMiao Xie static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 4388eb73c1b7SMiao Xie { 4389eb73c1b7SMiao Xie struct btrfs_root *root; 4390eb73c1b7SMiao Xie struct list_head splice; 4391eb73c1b7SMiao Xie 4392eb73c1b7SMiao Xie INIT_LIST_HEAD(&splice); 4393eb73c1b7SMiao Xie 4394eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4395eb73c1b7SMiao Xie list_splice_init(&fs_info->delalloc_roots, &splice); 4396eb73c1b7SMiao Xie while (!list_empty(&splice)) { 4397eb73c1b7SMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4398eb73c1b7SMiao Xie delalloc_root); 4399eb73c1b7SMiao Xie root = btrfs_grab_fs_root(root); 4400eb73c1b7SMiao Xie BUG_ON(!root); 4401eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4402eb73c1b7SMiao Xie 4403eb73c1b7SMiao Xie btrfs_destroy_delalloc_inodes(root); 4404eb73c1b7SMiao Xie btrfs_put_fs_root(root); 4405eb73c1b7SMiao Xie 4406eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4407eb73c1b7SMiao Xie } 4408eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4409acce952bSliubo } 4410acce952bSliubo 44112ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 4412acce952bSliubo struct extent_io_tree *dirty_pages, 4413acce952bSliubo int mark) 4414acce952bSliubo { 4415acce952bSliubo int ret; 4416acce952bSliubo struct extent_buffer *eb; 4417acce952bSliubo u64 start = 0; 4418acce952bSliubo u64 end; 4419acce952bSliubo 4420acce952bSliubo while (1) { 4421acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 4422e6138876SJosef Bacik mark, NULL); 4423acce952bSliubo if (ret) 4424acce952bSliubo break; 4425acce952bSliubo 442691166212SDavid Sterba clear_extent_bits(dirty_pages, start, end, mark); 4427acce952bSliubo while (start <= end) { 44280b246afaSJeff Mahoney eb = find_extent_buffer(fs_info, start); 44290b246afaSJeff Mahoney start += fs_info->nodesize; 4430fd8b2b61SJosef Bacik if (!eb) 4431acce952bSliubo continue; 4432fd8b2b61SJosef Bacik wait_on_extent_buffer_writeback(eb); 4433acce952bSliubo 4434fd8b2b61SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 4435fd8b2b61SJosef Bacik &eb->bflags)) 4436fd8b2b61SJosef Bacik clear_extent_buffer_dirty(eb); 4437fd8b2b61SJosef Bacik free_extent_buffer_stale(eb); 4438acce952bSliubo } 4439acce952bSliubo } 4440acce952bSliubo 4441acce952bSliubo return ret; 4442acce952bSliubo } 4443acce952bSliubo 44442ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 4445acce952bSliubo struct extent_io_tree *pinned_extents) 4446acce952bSliubo { 4447acce952bSliubo struct extent_io_tree *unpin; 4448acce952bSliubo u64 start; 4449acce952bSliubo u64 end; 4450acce952bSliubo int ret; 4451ed0eaa14SLiu Bo bool loop = true; 4452acce952bSliubo 4453acce952bSliubo unpin = pinned_extents; 4454ed0eaa14SLiu Bo again: 4455acce952bSliubo while (1) { 44560e6ec385SFilipe Manana struct extent_state *cached_state = NULL; 44570e6ec385SFilipe Manana 4458fcd5e742SLu Fengqi /* 4459fcd5e742SLu Fengqi * The btrfs_finish_extent_commit() may get the same range as 4460fcd5e742SLu Fengqi * ours between find_first_extent_bit and clear_extent_dirty. 4461fcd5e742SLu Fengqi * Hence, hold the unused_bg_unpin_mutex to avoid double unpin 4462fcd5e742SLu Fengqi * the same extent range. 4463fcd5e742SLu Fengqi */ 4464fcd5e742SLu Fengqi mutex_lock(&fs_info->unused_bg_unpin_mutex); 4465acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 44660e6ec385SFilipe Manana EXTENT_DIRTY, &cached_state); 4467fcd5e742SLu Fengqi if (ret) { 4468fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4469acce952bSliubo break; 4470fcd5e742SLu Fengqi } 4471acce952bSliubo 44720e6ec385SFilipe Manana clear_extent_dirty(unpin, start, end, &cached_state); 44730e6ec385SFilipe Manana free_extent_state(cached_state); 44742ff7e61eSJeff Mahoney btrfs_error_unpin_extent_range(fs_info, start, end); 4475fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4476acce952bSliubo cond_resched(); 4477acce952bSliubo } 4478acce952bSliubo 4479ed0eaa14SLiu Bo if (loop) { 44800b246afaSJeff Mahoney if (unpin == &fs_info->freed_extents[0]) 44810b246afaSJeff Mahoney unpin = &fs_info->freed_extents[1]; 4482ed0eaa14SLiu Bo else 44830b246afaSJeff Mahoney unpin = &fs_info->freed_extents[0]; 4484ed0eaa14SLiu Bo loop = false; 4485ed0eaa14SLiu Bo goto again; 4486ed0eaa14SLiu Bo } 4487ed0eaa14SLiu Bo 4488acce952bSliubo return 0; 4489acce952bSliubo } 4490acce952bSliubo 4491c79a1751SLiu Bo static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache) 4492c79a1751SLiu Bo { 4493c79a1751SLiu Bo struct inode *inode; 4494c79a1751SLiu Bo 4495c79a1751SLiu Bo inode = cache->io_ctl.inode; 4496c79a1751SLiu Bo if (inode) { 4497c79a1751SLiu Bo invalidate_inode_pages2(inode->i_mapping); 4498c79a1751SLiu Bo BTRFS_I(inode)->generation = 0; 4499c79a1751SLiu Bo cache->io_ctl.inode = NULL; 4500c79a1751SLiu Bo iput(inode); 4501c79a1751SLiu Bo } 4502c79a1751SLiu Bo btrfs_put_block_group(cache); 4503c79a1751SLiu Bo } 4504c79a1751SLiu Bo 4505c79a1751SLiu Bo void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, 45062ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4507c79a1751SLiu Bo { 4508c79a1751SLiu Bo struct btrfs_block_group_cache *cache; 4509c79a1751SLiu Bo 4510c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4511c79a1751SLiu Bo while (!list_empty(&cur_trans->dirty_bgs)) { 4512c79a1751SLiu Bo cache = list_first_entry(&cur_trans->dirty_bgs, 4513c79a1751SLiu Bo struct btrfs_block_group_cache, 4514c79a1751SLiu Bo dirty_list); 4515c79a1751SLiu Bo 4516c79a1751SLiu Bo if (!list_empty(&cache->io_list)) { 4517c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4518c79a1751SLiu Bo list_del_init(&cache->io_list); 4519c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 4520c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4521c79a1751SLiu Bo } 4522c79a1751SLiu Bo 4523c79a1751SLiu Bo list_del_init(&cache->dirty_list); 4524c79a1751SLiu Bo spin_lock(&cache->lock); 4525c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 4526c79a1751SLiu Bo spin_unlock(&cache->lock); 4527c79a1751SLiu Bo 4528c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4529c79a1751SLiu Bo btrfs_put_block_group(cache); 4530ba2c4d4eSJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 4531c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4532c79a1751SLiu Bo } 4533c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4534c79a1751SLiu Bo 453545ae2c18SNikolay Borisov /* 453645ae2c18SNikolay Borisov * Refer to the definition of io_bgs member for details why it's safe 453745ae2c18SNikolay Borisov * to use it without any locking 453845ae2c18SNikolay Borisov */ 4539c79a1751SLiu Bo while (!list_empty(&cur_trans->io_bgs)) { 4540c79a1751SLiu Bo cache = list_first_entry(&cur_trans->io_bgs, 4541c79a1751SLiu Bo struct btrfs_block_group_cache, 4542c79a1751SLiu Bo io_list); 4543c79a1751SLiu Bo 4544c79a1751SLiu Bo list_del_init(&cache->io_list); 4545c79a1751SLiu Bo spin_lock(&cache->lock); 4546c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 4547c79a1751SLiu Bo spin_unlock(&cache->lock); 4548c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 4549c79a1751SLiu Bo } 4550c79a1751SLiu Bo } 4551c79a1751SLiu Bo 455249b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 45532ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 455449b25e05SJeff Mahoney { 4555bbbf7243SNikolay Borisov struct btrfs_device *dev, *tmp; 4556bbbf7243SNikolay Borisov 45572ff7e61eSJeff Mahoney btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 4558c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->dirty_bgs)); 4559c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->io_bgs)); 4560c79a1751SLiu Bo 4561bbbf7243SNikolay Borisov list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list, 4562bbbf7243SNikolay Borisov post_commit_list) { 4563bbbf7243SNikolay Borisov list_del_init(&dev->post_commit_list); 4564bbbf7243SNikolay Borisov } 4565bbbf7243SNikolay Borisov 45662ff7e61eSJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, fs_info); 456749b25e05SJeff Mahoney 45684a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 45690b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 457049b25e05SJeff Mahoney 45714a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 45720b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 457349b25e05SJeff Mahoney 4574ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 4575ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 457667cde344SMiao Xie 45772ff7e61eSJeff Mahoney btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, 457849b25e05SJeff Mahoney EXTENT_DIRTY); 45792ff7e61eSJeff Mahoney btrfs_destroy_pinned_extent(fs_info, 45800b246afaSJeff Mahoney fs_info->pinned_extents); 458149b25e05SJeff Mahoney 45824a9d8bdeSMiao Xie cur_trans->state =TRANS_STATE_COMPLETED; 45834a9d8bdeSMiao Xie wake_up(&cur_trans->commit_wait); 458449b25e05SJeff Mahoney } 458549b25e05SJeff Mahoney 45862ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) 4587acce952bSliubo { 4588acce952bSliubo struct btrfs_transaction *t; 4589acce952bSliubo 45900b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 4591acce952bSliubo 45920b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 45930b246afaSJeff Mahoney while (!list_empty(&fs_info->trans_list)) { 45940b246afaSJeff Mahoney t = list_first_entry(&fs_info->trans_list, 4595724e2315SJosef Bacik struct btrfs_transaction, list); 4596724e2315SJosef Bacik if (t->state >= TRANS_STATE_COMMIT_START) { 45979b64f57dSElena Reshetova refcount_inc(&t->use_count); 45980b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 45992ff7e61eSJeff Mahoney btrfs_wait_for_commit(fs_info, t->transid); 4600724e2315SJosef Bacik btrfs_put_transaction(t); 46010b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4602724e2315SJosef Bacik continue; 4603724e2315SJosef Bacik } 46040b246afaSJeff Mahoney if (t == fs_info->running_transaction) { 4605724e2315SJosef Bacik t->state = TRANS_STATE_COMMIT_DOING; 46060b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4607724e2315SJosef Bacik /* 4608724e2315SJosef Bacik * We wait for 0 num_writers since we don't hold a trans 4609724e2315SJosef Bacik * handle open currently for this transaction. 4610724e2315SJosef Bacik */ 4611724e2315SJosef Bacik wait_event(t->writer_wait, 4612724e2315SJosef Bacik atomic_read(&t->num_writers) == 0); 4613724e2315SJosef Bacik } else { 46140b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4615724e2315SJosef Bacik } 46162ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(t, fs_info); 4617724e2315SJosef Bacik 46180b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 46190b246afaSJeff Mahoney if (t == fs_info->running_transaction) 46200b246afaSJeff Mahoney fs_info->running_transaction = NULL; 4621724e2315SJosef Bacik list_del_init(&t->list); 46220b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4623a4abeea4SJosef Bacik 4624724e2315SJosef Bacik btrfs_put_transaction(t); 46252ff7e61eSJeff Mahoney trace_btrfs_transaction_commit(fs_info->tree_root); 46260b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4627724e2315SJosef Bacik } 46280b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 46290b246afaSJeff Mahoney btrfs_destroy_all_ordered_extents(fs_info); 4630ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 4631ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 46322ff7e61eSJeff Mahoney btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents); 46330b246afaSJeff Mahoney btrfs_destroy_all_delalloc_inodes(fs_info); 46340b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 4635acce952bSliubo 4636acce952bSliubo return 0; 4637acce952bSliubo } 4638acce952bSliubo 4639e8c9f186SDavid Sterba static const struct extent_io_ops btree_extent_io_ops = { 46404d53dddbSDavid Sterba /* mandatory callbacks */ 46410b86a832SChris Mason .submit_bio_hook = btree_submit_bio_hook, 46424d53dddbSDavid Sterba .readpage_end_io_hook = btree_readpage_end_io_hook, 46430da5468fSChris Mason }; 4644