1c1d7c514SDavid Sterba // SPDX-License-Identifier: GPL-2.0 26cbd5570SChris Mason /* 36cbd5570SChris Mason * Copyright (C) 2007 Oracle. All rights reserved. 46cbd5570SChris Mason */ 56cbd5570SChris Mason 6e20d96d6SChris Mason #include <linux/fs.h> 7d98237b3SChris Mason #include <linux/blkdev.h> 80f7d52f4SChris Mason #include <linux/radix-tree.h> 935b7e476SChris Mason #include <linux/writeback.h> 10d397712bSChris Mason #include <linux/buffer_head.h> 11ce9adaa5SChris Mason #include <linux/workqueue.h> 12a74a4b97SChris Mason #include <linux/kthread.h> 135a0e3ad6STejun Heo #include <linux/slab.h> 14784b4e29SChris Mason #include <linux/migrate.h> 157a36ddecSDavid Sterba #include <linux/ratelimit.h> 166463fe58SStefan Behrens #include <linux/uuid.h> 17803b2f54SStefan Behrens #include <linux/semaphore.h> 18540adea3SMasami Hiramatsu #include <linux/error-injection.h> 199678c543SNikolay Borisov #include <linux/crc32c.h> 20b89f6d1fSFilipe Manana #include <linux/sched/mm.h> 217e75bf3fSDavid Sterba #include <asm/unaligned.h> 22eb60ceacSChris Mason #include "ctree.h" 23eb60ceacSChris Mason #include "disk-io.h" 24e089f05cSChris Mason #include "transaction.h" 250f7d52f4SChris Mason #include "btrfs_inode.h" 260b86a832SChris Mason #include "volumes.h" 27db94535dSChris Mason #include "print-tree.h" 28925baeddSChris Mason #include "locking.h" 29e02119d5SChris Mason #include "tree-log.h" 30fa9c0d79SChris Mason #include "free-space-cache.h" 3170f6d82eSOmar Sandoval #include "free-space-tree.h" 32581bb050SLi Zefan #include "inode-map.h" 3321adbd5cSStefan Behrens #include "check-integrity.h" 34606686eeSJosef Bacik #include "rcu-string.h" 358dabb742SStefan Behrens #include "dev-replace.h" 3653b381b3SDavid Woodhouse #include "raid56.h" 375ac1d209SJeff Mahoney #include "sysfs.h" 38fcebe456SJosef Bacik #include "qgroup.h" 39ebb8765bSAnand Jain #include "compression.h" 40557ea5ddSQu Wenruo #include "tree-checker.h" 41fd708b81SJosef Bacik #include "ref-verify.h" 42eb60ceacSChris Mason 43319e4d06SQu Wenruo #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\ 44319e4d06SQu Wenruo BTRFS_HEADER_FLAG_RELOC |\ 45319e4d06SQu Wenruo BTRFS_SUPER_FLAG_ERROR |\ 46319e4d06SQu Wenruo BTRFS_SUPER_FLAG_SEEDING |\ 47e2731e55SAnand Jain BTRFS_SUPER_FLAG_METADUMP |\ 48e2731e55SAnand Jain BTRFS_SUPER_FLAG_METADUMP_V2) 49319e4d06SQu Wenruo 50e8c9f186SDavid Sterba static const struct extent_io_ops btree_extent_io_ops; 518b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work); 52143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root); 53acce952bSliubo static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 542ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info); 55143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root); 562ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 57acce952bSliubo struct extent_io_tree *dirty_pages, 58acce952bSliubo int mark); 592ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 60acce952bSliubo struct extent_io_tree *pinned_extents); 612ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info); 622ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info); 63ce9adaa5SChris Mason 64d352ac68SChris Mason /* 6597eb6b69SDavid Sterba * btrfs_end_io_wq structs are used to do processing in task context when an IO 6697eb6b69SDavid Sterba * is complete. This is used during reads to verify checksums, and it is used 67d352ac68SChris Mason * by writes to insert metadata for new file extents after IO is complete. 68d352ac68SChris Mason */ 6997eb6b69SDavid Sterba struct btrfs_end_io_wq { 70ce9adaa5SChris Mason struct bio *bio; 71ce9adaa5SChris Mason bio_end_io_t *end_io; 72ce9adaa5SChris Mason void *private; 73ce9adaa5SChris Mason struct btrfs_fs_info *info; 744e4cbee9SChristoph Hellwig blk_status_t status; 75bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata; 768b712842SChris Mason struct btrfs_work work; 77ce9adaa5SChris Mason }; 780da5468fSChris Mason 7997eb6b69SDavid Sterba static struct kmem_cache *btrfs_end_io_wq_cache; 8097eb6b69SDavid Sterba 8197eb6b69SDavid Sterba int __init btrfs_end_io_wq_init(void) 8297eb6b69SDavid Sterba { 8397eb6b69SDavid Sterba btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq", 8497eb6b69SDavid Sterba sizeof(struct btrfs_end_io_wq), 8597eb6b69SDavid Sterba 0, 86fba4b697SNikolay Borisov SLAB_MEM_SPREAD, 8797eb6b69SDavid Sterba NULL); 8897eb6b69SDavid Sterba if (!btrfs_end_io_wq_cache) 8997eb6b69SDavid Sterba return -ENOMEM; 9097eb6b69SDavid Sterba return 0; 9197eb6b69SDavid Sterba } 9297eb6b69SDavid Sterba 93e67c718bSDavid Sterba void __cold btrfs_end_io_wq_exit(void) 9497eb6b69SDavid Sterba { 9597eb6b69SDavid Sterba kmem_cache_destroy(btrfs_end_io_wq_cache); 9697eb6b69SDavid Sterba } 9797eb6b69SDavid Sterba 98d352ac68SChris Mason /* 99d352ac68SChris Mason * async submit bios are used to offload expensive checksumming 100d352ac68SChris Mason * onto the worker threads. They checksum file and metadata bios 101d352ac68SChris Mason * just before they are sent down the IO stack. 102d352ac68SChris Mason */ 10344b8bd7eSChris Mason struct async_submit_bio { 104c6100a4bSJosef Bacik void *private_data; 10544b8bd7eSChris Mason struct bio *bio; 106a758781dSDavid Sterba extent_submit_bio_start_t *submit_bio_start; 10744b8bd7eSChris Mason int mirror_num; 108eaf25d93SChris Mason /* 109eaf25d93SChris Mason * bio_offset is optional, can be used if the pages in the bio 110eaf25d93SChris Mason * can't tell us where in the file the bio should go 111eaf25d93SChris Mason */ 112eaf25d93SChris Mason u64 bio_offset; 1138b712842SChris Mason struct btrfs_work work; 1144e4cbee9SChristoph Hellwig blk_status_t status; 11544b8bd7eSChris Mason }; 11644b8bd7eSChris Mason 11785d4e461SChris Mason /* 11885d4e461SChris Mason * Lockdep class keys for extent_buffer->lock's in this root. For a given 11985d4e461SChris Mason * eb, the lockdep key is determined by the btrfs_root it belongs to and 12085d4e461SChris Mason * the level the eb occupies in the tree. 1214008c04aSChris Mason * 12285d4e461SChris Mason * Different roots are used for different purposes and may nest inside each 12385d4e461SChris Mason * other and they require separate keysets. As lockdep keys should be 12485d4e461SChris Mason * static, assign keysets according to the purpose of the root as indicated 1254fd786e6SMisono Tomohiro * by btrfs_root->root_key.objectid. This ensures that all special purpose 1264fd786e6SMisono Tomohiro * roots have separate keysets. 1274008c04aSChris Mason * 12885d4e461SChris Mason * Lock-nesting across peer nodes is always done with the immediate parent 12985d4e461SChris Mason * node locked thus preventing deadlock. As lockdep doesn't know this, use 13085d4e461SChris Mason * subclass to avoid triggering lockdep warning in such cases. 1314008c04aSChris Mason * 13285d4e461SChris Mason * The key is set by the readpage_end_io_hook after the buffer has passed 13385d4e461SChris Mason * csum validation but before the pages are unlocked. It is also set by 13485d4e461SChris Mason * btrfs_init_new_buffer on freshly allocated blocks. 13585d4e461SChris Mason * 13685d4e461SChris Mason * We also add a check to make sure the highest level of the tree is the 13785d4e461SChris Mason * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code 13885d4e461SChris Mason * needs update as well. 1394008c04aSChris Mason */ 1404008c04aSChris Mason #ifdef CONFIG_DEBUG_LOCK_ALLOC 1414008c04aSChris Mason # if BTRFS_MAX_LEVEL != 8 1424008c04aSChris Mason # error 1434008c04aSChris Mason # endif 14485d4e461SChris Mason 14585d4e461SChris Mason static struct btrfs_lockdep_keyset { 14685d4e461SChris Mason u64 id; /* root objectid */ 14785d4e461SChris Mason const char *name_stem; /* lock name stem */ 14885d4e461SChris Mason char names[BTRFS_MAX_LEVEL + 1][20]; 14985d4e461SChris Mason struct lock_class_key keys[BTRFS_MAX_LEVEL + 1]; 15085d4e461SChris Mason } btrfs_lockdep_keysets[] = { 15185d4e461SChris Mason { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" }, 15285d4e461SChris Mason { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" }, 15385d4e461SChris Mason { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" }, 15485d4e461SChris Mason { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" }, 15585d4e461SChris Mason { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" }, 15685d4e461SChris Mason { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" }, 15760b62978SDavid Sterba { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" }, 15885d4e461SChris Mason { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" }, 15985d4e461SChris Mason { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" }, 16085d4e461SChris Mason { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" }, 16113fd8da9SDavid Sterba { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" }, 1626b20e0adSDavid Sterba { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" }, 16385d4e461SChris Mason { .id = 0, .name_stem = "tree" }, 1644008c04aSChris Mason }; 16585d4e461SChris Mason 16685d4e461SChris Mason void __init btrfs_init_lockdep(void) 16785d4e461SChris Mason { 16885d4e461SChris Mason int i, j; 16985d4e461SChris Mason 17085d4e461SChris Mason /* initialize lockdep class names */ 17185d4e461SChris Mason for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) { 17285d4e461SChris Mason struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i]; 17385d4e461SChris Mason 17485d4e461SChris Mason for (j = 0; j < ARRAY_SIZE(ks->names); j++) 17585d4e461SChris Mason snprintf(ks->names[j], sizeof(ks->names[j]), 17685d4e461SChris Mason "btrfs-%s-%02d", ks->name_stem, j); 17785d4e461SChris Mason } 17885d4e461SChris Mason } 17985d4e461SChris Mason 18085d4e461SChris Mason void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb, 18185d4e461SChris Mason int level) 18285d4e461SChris Mason { 18385d4e461SChris Mason struct btrfs_lockdep_keyset *ks; 18485d4e461SChris Mason 18585d4e461SChris Mason BUG_ON(level >= ARRAY_SIZE(ks->keys)); 18685d4e461SChris Mason 18785d4e461SChris Mason /* find the matching keyset, id 0 is the default entry */ 18885d4e461SChris Mason for (ks = btrfs_lockdep_keysets; ks->id; ks++) 18985d4e461SChris Mason if (ks->id == objectid) 19085d4e461SChris Mason break; 19185d4e461SChris Mason 19285d4e461SChris Mason lockdep_set_class_and_name(&eb->lock, 19385d4e461SChris Mason &ks->keys[level], ks->names[level]); 19485d4e461SChris Mason } 19585d4e461SChris Mason 1964008c04aSChris Mason #endif 1974008c04aSChris Mason 198d352ac68SChris Mason /* 199d352ac68SChris Mason * extents on the btree inode are pretty simple, there's one extent 200d352ac68SChris Mason * that covers the entire device 201d352ac68SChris Mason */ 2026af49dbdSDavid Sterba struct extent_map *btree_get_extent(struct btrfs_inode *inode, 203306e16ceSDavid Sterba struct page *page, size_t pg_offset, u64 start, u64 len, 2045f39d397SChris Mason int create) 2055f39d397SChris Mason { 2063ffbd68cSDavid Sterba struct btrfs_fs_info *fs_info = inode->root->fs_info; 207fc4f21b1SNikolay Borisov struct extent_map_tree *em_tree = &inode->extent_tree; 2085f39d397SChris Mason struct extent_map *em; 2095f39d397SChris Mason int ret; 2105f39d397SChris Mason 211890871beSChris Mason read_lock(&em_tree->lock); 212d1310b2eSChris Mason em = lookup_extent_mapping(em_tree, start, len); 213a061fc8dSChris Mason if (em) { 2140b246afaSJeff Mahoney em->bdev = fs_info->fs_devices->latest_bdev; 215890871beSChris Mason read_unlock(&em_tree->lock); 2165f39d397SChris Mason goto out; 217a061fc8dSChris Mason } 218890871beSChris Mason read_unlock(&em_tree->lock); 2197b13b7b1SChris Mason 220172ddd60SDavid Sterba em = alloc_extent_map(); 2215f39d397SChris Mason if (!em) { 2225f39d397SChris Mason em = ERR_PTR(-ENOMEM); 2235f39d397SChris Mason goto out; 2245f39d397SChris Mason } 2255f39d397SChris Mason em->start = 0; 2260afbaf8cSChris Mason em->len = (u64)-1; 227c8b97818SChris Mason em->block_len = (u64)-1; 2285f39d397SChris Mason em->block_start = 0; 2290b246afaSJeff Mahoney em->bdev = fs_info->fs_devices->latest_bdev; 230d1310b2eSChris Mason 231890871beSChris Mason write_lock(&em_tree->lock); 23209a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 2335f39d397SChris Mason if (ret == -EEXIST) { 2345f39d397SChris Mason free_extent_map(em); 2357b13b7b1SChris Mason em = lookup_extent_mapping(em_tree, start, len); 236b4f359abSTsutomu Itoh if (!em) 2370433f20dSTsutomu Itoh em = ERR_PTR(-EIO); 2385f39d397SChris Mason } else if (ret) { 2397b13b7b1SChris Mason free_extent_map(em); 2400433f20dSTsutomu Itoh em = ERR_PTR(ret); 2415f39d397SChris Mason } 242890871beSChris Mason write_unlock(&em_tree->lock); 2437b13b7b1SChris Mason 2445f39d397SChris Mason out: 2455f39d397SChris Mason return em; 2465f39d397SChris Mason } 2475f39d397SChris Mason 2489ed57367SDavid Sterba u32 btrfs_csum_data(const char *data, u32 seed, size_t len) 24919c00ddcSChris Mason { 2509678c543SNikolay Borisov return crc32c(seed, data, len); 25119c00ddcSChris Mason } 25219c00ddcSChris Mason 2530b5e3dafSDomagoj Tršan void btrfs_csum_final(u32 crc, u8 *result) 25419c00ddcSChris Mason { 2557e75bf3fSDavid Sterba put_unaligned_le32(~crc, result); 25619c00ddcSChris Mason } 25719c00ddcSChris Mason 258d352ac68SChris Mason /* 2592996e1f8SJohannes Thumshirn * Compute the csum of a btree block and store the result to provided buffer. 2602996e1f8SJohannes Thumshirn * 2612996e1f8SJohannes Thumshirn * Returns error if the extent buffer cannot be mapped. 262d352ac68SChris Mason */ 2632996e1f8SJohannes Thumshirn static int csum_tree_block(struct extent_buffer *buf, u8 *result) 26419c00ddcSChris Mason { 26519c00ddcSChris Mason unsigned long len; 26619c00ddcSChris Mason unsigned long cur_len; 26719c00ddcSChris Mason unsigned long offset = BTRFS_CSUM_SIZE; 26819c00ddcSChris Mason char *kaddr; 26919c00ddcSChris Mason unsigned long map_start; 27019c00ddcSChris Mason unsigned long map_len; 27119c00ddcSChris Mason int err; 27219c00ddcSChris Mason u32 crc = ~(u32)0; 27319c00ddcSChris Mason 27419c00ddcSChris Mason len = buf->len - offset; 27519c00ddcSChris Mason while (len > 0) { 276d2e174d5SJohannes Thumshirn /* 277d2e174d5SJohannes Thumshirn * Note: we don't need to check for the err == 1 case here, as 278d2e174d5SJohannes Thumshirn * with the given combination of 'start = BTRFS_CSUM_SIZE (32)' 279d2e174d5SJohannes Thumshirn * and 'min_len = 32' and the currently implemented mapping 280d2e174d5SJohannes Thumshirn * algorithm we cannot cross a page boundary. 281d2e174d5SJohannes Thumshirn */ 28219c00ddcSChris Mason err = map_private_extent_buffer(buf, offset, 32, 283a6591715SChris Mason &kaddr, &map_start, &map_len); 284c53839fcSJohannes Thumshirn if (WARN_ON(err)) 2858bd98f0eSAlex Lyakas return err; 28619c00ddcSChris Mason cur_len = min(len, map_len - (offset - map_start)); 287b0496686SLiu Bo crc = btrfs_csum_data(kaddr + offset - map_start, 28819c00ddcSChris Mason crc, cur_len); 28919c00ddcSChris Mason len -= cur_len; 29019c00ddcSChris Mason offset += cur_len; 29119c00ddcSChris Mason } 29271a63551SDavid Sterba memset(result, 0, BTRFS_CSUM_SIZE); 293607d432dSJosef Bacik 29419c00ddcSChris Mason btrfs_csum_final(crc, result); 29519c00ddcSChris Mason 29619c00ddcSChris Mason return 0; 29719c00ddcSChris Mason } 29819c00ddcSChris Mason 299d352ac68SChris Mason /* 300d352ac68SChris Mason * we can't consider a given block up to date unless the transid of the 301d352ac68SChris Mason * block matches the transid in the parent node's pointer. This is how we 302d352ac68SChris Mason * detect blocks that either didn't get written at all or got written 303d352ac68SChris Mason * in the wrong place. 304d352ac68SChris Mason */ 3051259ab75SChris Mason static int verify_parent_transid(struct extent_io_tree *io_tree, 306b9fab919SChris Mason struct extent_buffer *eb, u64 parent_transid, 307b9fab919SChris Mason int atomic) 3081259ab75SChris Mason { 3092ac55d41SJosef Bacik struct extent_state *cached_state = NULL; 3101259ab75SChris Mason int ret; 3112755a0deSDavid Sterba bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB); 3121259ab75SChris Mason 3131259ab75SChris Mason if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 3141259ab75SChris Mason return 0; 3151259ab75SChris Mason 316b9fab919SChris Mason if (atomic) 317b9fab919SChris Mason return -EAGAIN; 318b9fab919SChris Mason 319a26e8c9fSJosef Bacik if (need_lock) { 320a26e8c9fSJosef Bacik btrfs_tree_read_lock(eb); 321300aa896SDavid Sterba btrfs_set_lock_blocking_read(eb); 322a26e8c9fSJosef Bacik } 323a26e8c9fSJosef Bacik 3242ac55d41SJosef Bacik lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 325ff13db41SDavid Sterba &cached_state); 3260b32f4bbSJosef Bacik if (extent_buffer_uptodate(eb) && 3271259ab75SChris Mason btrfs_header_generation(eb) == parent_transid) { 3281259ab75SChris Mason ret = 0; 3291259ab75SChris Mason goto out; 3301259ab75SChris Mason } 33194647322SDavid Sterba btrfs_err_rl(eb->fs_info, 33294647322SDavid Sterba "parent transid verify failed on %llu wanted %llu found %llu", 33394647322SDavid Sterba eb->start, 33429549aecSWang Shilong parent_transid, btrfs_header_generation(eb)); 3351259ab75SChris Mason ret = 1; 336a26e8c9fSJosef Bacik 337a26e8c9fSJosef Bacik /* 338a26e8c9fSJosef Bacik * Things reading via commit roots that don't have normal protection, 339a26e8c9fSJosef Bacik * like send, can have a really old block in cache that may point at a 34001327610SNicholas D Steeves * block that has been freed and re-allocated. So don't clear uptodate 341a26e8c9fSJosef Bacik * if we find an eb that is under IO (dirty/writeback) because we could 342a26e8c9fSJosef Bacik * end up reading in the stale data and then writing it back out and 343a26e8c9fSJosef Bacik * making everybody very sad. 344a26e8c9fSJosef Bacik */ 345a26e8c9fSJosef Bacik if (!extent_buffer_under_io(eb)) 3460b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 34733958dc6SChris Mason out: 3482ac55d41SJosef Bacik unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 349e43bbe5eSDavid Sterba &cached_state); 350472b909fSJosef Bacik if (need_lock) 351a26e8c9fSJosef Bacik btrfs_tree_read_unlock_blocking(eb); 3521259ab75SChris Mason return ret; 3531259ab75SChris Mason } 3541259ab75SChris Mason 355e7e16f48SJohannes Thumshirn static bool btrfs_supported_super_csum(u16 csum_type) 356e7e16f48SJohannes Thumshirn { 357e7e16f48SJohannes Thumshirn switch (csum_type) { 358e7e16f48SJohannes Thumshirn case BTRFS_CSUM_TYPE_CRC32: 359e7e16f48SJohannes Thumshirn return true; 360e7e16f48SJohannes Thumshirn default: 361e7e16f48SJohannes Thumshirn return false; 362e7e16f48SJohannes Thumshirn } 363e7e16f48SJohannes Thumshirn } 364e7e16f48SJohannes Thumshirn 365d352ac68SChris Mason /* 3661104a885SDavid Sterba * Return 0 if the superblock checksum type matches the checksum value of that 3671104a885SDavid Sterba * algorithm. Pass the raw disk superblock data. 3681104a885SDavid Sterba */ 369ab8d0fc4SJeff Mahoney static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info, 370ab8d0fc4SJeff Mahoney char *raw_disk_sb) 3711104a885SDavid Sterba { 3721104a885SDavid Sterba struct btrfs_super_block *disk_sb = 3731104a885SDavid Sterba (struct btrfs_super_block *)raw_disk_sb; 3741104a885SDavid Sterba u16 csum_type = btrfs_super_csum_type(disk_sb); 375e7e16f48SJohannes Thumshirn 376e7e16f48SJohannes Thumshirn if (!btrfs_supported_super_csum(csum_type)) { 377e7e16f48SJohannes Thumshirn btrfs_err(fs_info, "unsupported checksum algorithm %u", 378e7e16f48SJohannes Thumshirn csum_type); 379e7e16f48SJohannes Thumshirn return 1; 380e7e16f48SJohannes Thumshirn } 3811104a885SDavid Sterba 3821104a885SDavid Sterba if (csum_type == BTRFS_CSUM_TYPE_CRC32) { 3831104a885SDavid Sterba u32 crc = ~(u32)0; 384776c4a7cSNikolay Borisov char result[sizeof(crc)]; 3851104a885SDavid Sterba 3861104a885SDavid Sterba /* 3871104a885SDavid Sterba * The super_block structure does not span the whole 3881104a885SDavid Sterba * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space 38901327610SNicholas D Steeves * is filled with zeros and is included in the checksum. 3901104a885SDavid Sterba */ 3911104a885SDavid Sterba crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE, 3921104a885SDavid Sterba crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE); 3931104a885SDavid Sterba btrfs_csum_final(crc, result); 3941104a885SDavid Sterba 395776c4a7cSNikolay Borisov if (memcmp(raw_disk_sb, result, sizeof(result))) 396e7e16f48SJohannes Thumshirn return 1; 3971104a885SDavid Sterba } 3981104a885SDavid Sterba 399e7e16f48SJohannes Thumshirn return 0; 4001104a885SDavid Sterba } 4011104a885SDavid Sterba 402e064d5e9SDavid Sterba int btrfs_verify_level_key(struct extent_buffer *eb, int level, 403ff76a864SLiu Bo struct btrfs_key *first_key, u64 parent_transid) 404581c1760SQu Wenruo { 405e064d5e9SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 406581c1760SQu Wenruo int found_level; 407581c1760SQu Wenruo struct btrfs_key found_key; 408581c1760SQu Wenruo int ret; 409581c1760SQu Wenruo 410581c1760SQu Wenruo found_level = btrfs_header_level(eb); 411581c1760SQu Wenruo if (found_level != level) { 41263489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 41363489055SQu Wenruo KERN_ERR "BTRFS: tree level check failed\n"); 414581c1760SQu Wenruo btrfs_err(fs_info, 415581c1760SQu Wenruo "tree level mismatch detected, bytenr=%llu level expected=%u has=%u", 416581c1760SQu Wenruo eb->start, level, found_level); 417581c1760SQu Wenruo return -EIO; 418581c1760SQu Wenruo } 419581c1760SQu Wenruo 420581c1760SQu Wenruo if (!first_key) 421581c1760SQu Wenruo return 0; 422581c1760SQu Wenruo 4235d41be6fSQu Wenruo /* 4245d41be6fSQu Wenruo * For live tree block (new tree blocks in current transaction), 4255d41be6fSQu Wenruo * we need proper lock context to avoid race, which is impossible here. 4265d41be6fSQu Wenruo * So we only checks tree blocks which is read from disk, whose 4275d41be6fSQu Wenruo * generation <= fs_info->last_trans_committed. 4285d41be6fSQu Wenruo */ 4295d41be6fSQu Wenruo if (btrfs_header_generation(eb) > fs_info->last_trans_committed) 4305d41be6fSQu Wenruo return 0; 431581c1760SQu Wenruo if (found_level) 432581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &found_key, 0); 433581c1760SQu Wenruo else 434581c1760SQu Wenruo btrfs_item_key_to_cpu(eb, &found_key, 0); 435581c1760SQu Wenruo ret = btrfs_comp_cpu_keys(first_key, &found_key); 436581c1760SQu Wenruo 437581c1760SQu Wenruo if (ret) { 43863489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 43963489055SQu Wenruo KERN_ERR "BTRFS: tree first key check failed\n"); 440581c1760SQu Wenruo btrfs_err(fs_info, 441ff76a864SLiu Bo "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", 442ff76a864SLiu Bo eb->start, parent_transid, first_key->objectid, 443ff76a864SLiu Bo first_key->type, first_key->offset, 444ff76a864SLiu Bo found_key.objectid, found_key.type, 445ff76a864SLiu Bo found_key.offset); 446581c1760SQu Wenruo } 447581c1760SQu Wenruo return ret; 448581c1760SQu Wenruo } 449581c1760SQu Wenruo 4501104a885SDavid Sterba /* 451d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 452d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 453581c1760SQu Wenruo * 454581c1760SQu Wenruo * @parent_transid: expected transid, skip check if 0 455581c1760SQu Wenruo * @level: expected level, mandatory check 456581c1760SQu Wenruo * @first_key: expected key of first slot, skip check if NULL 457d352ac68SChris Mason */ 4585ab12d1fSDavid Sterba static int btree_read_extent_buffer_pages(struct extent_buffer *eb, 459581c1760SQu Wenruo u64 parent_transid, int level, 460581c1760SQu Wenruo struct btrfs_key *first_key) 461f188591eSChris Mason { 4625ab12d1fSDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 463f188591eSChris Mason struct extent_io_tree *io_tree; 464ea466794SJosef Bacik int failed = 0; 465f188591eSChris Mason int ret; 466f188591eSChris Mason int num_copies = 0; 467f188591eSChris Mason int mirror_num = 0; 468ea466794SJosef Bacik int failed_mirror = 0; 469f188591eSChris Mason 4700b246afaSJeff Mahoney io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; 471f188591eSChris Mason while (1) { 472f8397d69SNikolay Borisov clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 473c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num); 474256dd1bbSStefan Behrens if (!ret) { 475581c1760SQu Wenruo if (verify_parent_transid(io_tree, eb, 476b9fab919SChris Mason parent_transid, 0)) 477256dd1bbSStefan Behrens ret = -EIO; 478e064d5e9SDavid Sterba else if (btrfs_verify_level_key(eb, level, 479ff76a864SLiu Bo first_key, parent_transid)) 480581c1760SQu Wenruo ret = -EUCLEAN; 481581c1760SQu Wenruo else 482581c1760SQu Wenruo break; 483256dd1bbSStefan Behrens } 484d397712bSChris Mason 4850b246afaSJeff Mahoney num_copies = btrfs_num_copies(fs_info, 486f188591eSChris Mason eb->start, eb->len); 4874235298eSChris Mason if (num_copies == 1) 488ea466794SJosef Bacik break; 4894235298eSChris Mason 4905cf1ab56SJosef Bacik if (!failed_mirror) { 4915cf1ab56SJosef Bacik failed = 1; 4925cf1ab56SJosef Bacik failed_mirror = eb->read_mirror; 4935cf1ab56SJosef Bacik } 4945cf1ab56SJosef Bacik 495f188591eSChris Mason mirror_num++; 496ea466794SJosef Bacik if (mirror_num == failed_mirror) 497ea466794SJosef Bacik mirror_num++; 498ea466794SJosef Bacik 4994235298eSChris Mason if (mirror_num > num_copies) 500ea466794SJosef Bacik break; 501f188591eSChris Mason } 502ea466794SJosef Bacik 503c0901581SStefan Behrens if (failed && !ret && failed_mirror) 50420a1fbf9SDavid Sterba btrfs_repair_eb_io_failure(eb, failed_mirror); 505ea466794SJosef Bacik 506ea466794SJosef Bacik return ret; 507f188591eSChris Mason } 50819c00ddcSChris Mason 509d352ac68SChris Mason /* 510d397712bSChris Mason * checksum a dirty tree block before IO. This has extra checks to make sure 511d397712bSChris Mason * we only fill in the checksum field in the first page of a multi-page block 512d352ac68SChris Mason */ 513d397712bSChris Mason 51401d58472SDaniel Dressler static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page) 51519c00ddcSChris Mason { 5164eee4fa4SMiao Xie u64 start = page_offset(page); 51719c00ddcSChris Mason u64 found_start; 5182996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 5192996e1f8SJohannes Thumshirn u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); 52019c00ddcSChris Mason struct extent_buffer *eb; 5218d47a0d8SQu Wenruo int ret; 522f188591eSChris Mason 5234f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 5244f2de97aSJosef Bacik if (page != eb->pages[0]) 5254f2de97aSJosef Bacik return 0; 5260f805531SAlex Lyakas 52719c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 5280f805531SAlex Lyakas /* 5290f805531SAlex Lyakas * Please do not consolidate these warnings into a single if. 5300f805531SAlex Lyakas * It is useful to know what went wrong. 5310f805531SAlex Lyakas */ 5320f805531SAlex Lyakas if (WARN_ON(found_start != start)) 5330f805531SAlex Lyakas return -EUCLEAN; 5340f805531SAlex Lyakas if (WARN_ON(!PageUptodate(page))) 5350f805531SAlex Lyakas return -EUCLEAN; 5360f805531SAlex Lyakas 537de37aa51SNikolay Borisov ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid, 5380f805531SAlex Lyakas btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0); 5390f805531SAlex Lyakas 5402996e1f8SJohannes Thumshirn if (csum_tree_block(eb, result)) 5412996e1f8SJohannes Thumshirn return -EINVAL; 5422996e1f8SJohannes Thumshirn 5438d47a0d8SQu Wenruo if (btrfs_header_level(eb)) 5448d47a0d8SQu Wenruo ret = btrfs_check_node(eb); 5458d47a0d8SQu Wenruo else 5468d47a0d8SQu Wenruo ret = btrfs_check_leaf_full(eb); 5478d47a0d8SQu Wenruo 5488d47a0d8SQu Wenruo if (ret < 0) { 5498d47a0d8SQu Wenruo btrfs_err(fs_info, 5508d47a0d8SQu Wenruo "block=%llu write time tree block corruption detected", 5518d47a0d8SQu Wenruo eb->start); 5528d47a0d8SQu Wenruo return ret; 5538d47a0d8SQu Wenruo } 5542996e1f8SJohannes Thumshirn write_extent_buffer(eb, result, 0, csum_size); 5558d47a0d8SQu Wenruo 5562996e1f8SJohannes Thumshirn return 0; 55719c00ddcSChris Mason } 55819c00ddcSChris Mason 559b0c9b3b0SDavid Sterba static int check_tree_block_fsid(struct extent_buffer *eb) 5602b82032cSYan Zheng { 561b0c9b3b0SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 56201d58472SDaniel Dressler struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 56344880fdcSAnand Jain u8 fsid[BTRFS_FSID_SIZE]; 5642b82032cSYan Zheng int ret = 1; 5652b82032cSYan Zheng 5660a4e5586SRoss Kirk read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE); 5672b82032cSYan Zheng while (fs_devices) { 5687239ff4bSNikolay Borisov u8 *metadata_uuid; 5697239ff4bSNikolay Borisov 5707239ff4bSNikolay Borisov /* 5717239ff4bSNikolay Borisov * Checking the incompat flag is only valid for the current 5727239ff4bSNikolay Borisov * fs. For seed devices it's forbidden to have their uuid 5737239ff4bSNikolay Borisov * changed so reading ->fsid in this case is fine 5747239ff4bSNikolay Borisov */ 5757239ff4bSNikolay Borisov if (fs_devices == fs_info->fs_devices && 5767239ff4bSNikolay Borisov btrfs_fs_incompat(fs_info, METADATA_UUID)) 5777239ff4bSNikolay Borisov metadata_uuid = fs_devices->metadata_uuid; 5787239ff4bSNikolay Borisov else 5797239ff4bSNikolay Borisov metadata_uuid = fs_devices->fsid; 5807239ff4bSNikolay Borisov 5817239ff4bSNikolay Borisov if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) { 5822b82032cSYan Zheng ret = 0; 5832b82032cSYan Zheng break; 5842b82032cSYan Zheng } 5852b82032cSYan Zheng fs_devices = fs_devices->seed; 5862b82032cSYan Zheng } 5872b82032cSYan Zheng return ret; 5882b82032cSYan Zheng } 5892b82032cSYan Zheng 590facc8a22SMiao Xie static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio, 591facc8a22SMiao Xie u64 phy_offset, struct page *page, 592facc8a22SMiao Xie u64 start, u64 end, int mirror) 593ce9adaa5SChris Mason { 594ce9adaa5SChris Mason u64 found_start; 595ce9adaa5SChris Mason int found_level; 596ce9adaa5SChris Mason struct extent_buffer *eb; 597ce9adaa5SChris Mason struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 59802873e43SZhao Lei struct btrfs_fs_info *fs_info = root->fs_info; 5992996e1f8SJohannes Thumshirn u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); 600f188591eSChris Mason int ret = 0; 6012996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 602727011e0SChris Mason int reads_done; 603ce9adaa5SChris Mason 604ce9adaa5SChris Mason if (!page->private) 605ce9adaa5SChris Mason goto out; 606d397712bSChris Mason 6074f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 608d397712bSChris Mason 6090b32f4bbSJosef Bacik /* the pending IO might have been the only thing that kept this buffer 6100b32f4bbSJosef Bacik * in memory. Make sure we have a ref for all this other checks 6110b32f4bbSJosef Bacik */ 6120b32f4bbSJosef Bacik extent_buffer_get(eb); 6130b32f4bbSJosef Bacik 6140b32f4bbSJosef Bacik reads_done = atomic_dec_and_test(&eb->io_pages); 615727011e0SChris Mason if (!reads_done) 616727011e0SChris Mason goto err; 617f188591eSChris Mason 6185cf1ab56SJosef Bacik eb->read_mirror = mirror; 619656f30dbSFilipe Manana if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 620ea466794SJosef Bacik ret = -EIO; 621ea466794SJosef Bacik goto err; 622ea466794SJosef Bacik } 623ea466794SJosef Bacik 624ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 625727011e0SChris Mason if (found_start != eb->start) { 626893bf4b1SSu Yue btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu", 627893bf4b1SSu Yue eb->start, found_start); 628f188591eSChris Mason ret = -EIO; 629ce9adaa5SChris Mason goto err; 630ce9adaa5SChris Mason } 631b0c9b3b0SDavid Sterba if (check_tree_block_fsid(eb)) { 63202873e43SZhao Lei btrfs_err_rl(fs_info, "bad fsid on block %llu", 63394647322SDavid Sterba eb->start); 6341259ab75SChris Mason ret = -EIO; 6351259ab75SChris Mason goto err; 6361259ab75SChris Mason } 637ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 6381c24c3ceSJosef Bacik if (found_level >= BTRFS_MAX_LEVEL) { 639893bf4b1SSu Yue btrfs_err(fs_info, "bad tree block level %d on %llu", 640893bf4b1SSu Yue (int)btrfs_header_level(eb), eb->start); 6411c24c3ceSJosef Bacik ret = -EIO; 6421c24c3ceSJosef Bacik goto err; 6431c24c3ceSJosef Bacik } 644ce9adaa5SChris Mason 64585d4e461SChris Mason btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), 64685d4e461SChris Mason eb, found_level); 6474008c04aSChris Mason 6482996e1f8SJohannes Thumshirn ret = csum_tree_block(eb, result); 6498bd98f0eSAlex Lyakas if (ret) 650a826d6dcSJosef Bacik goto err; 651a826d6dcSJosef Bacik 6522996e1f8SJohannes Thumshirn if (memcmp_extent_buffer(eb, result, 0, csum_size)) { 6532996e1f8SJohannes Thumshirn u32 val; 6542996e1f8SJohannes Thumshirn u32 found = 0; 6552996e1f8SJohannes Thumshirn 6562996e1f8SJohannes Thumshirn memcpy(&found, result, csum_size); 6572996e1f8SJohannes Thumshirn 6582996e1f8SJohannes Thumshirn read_extent_buffer(eb, &val, 0, csum_size); 6592996e1f8SJohannes Thumshirn btrfs_warn_rl(fs_info, 6602996e1f8SJohannes Thumshirn "%s checksum verify failed on %llu wanted %x found %x level %d", 6612996e1f8SJohannes Thumshirn fs_info->sb->s_id, eb->start, 6622996e1f8SJohannes Thumshirn val, found, btrfs_header_level(eb)); 6632996e1f8SJohannes Thumshirn ret = -EUCLEAN; 6642996e1f8SJohannes Thumshirn goto err; 6652996e1f8SJohannes Thumshirn } 6662996e1f8SJohannes Thumshirn 667a826d6dcSJosef Bacik /* 668a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 669a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 670a826d6dcSJosef Bacik * return -EIO. 671a826d6dcSJosef Bacik */ 6721c4360eeSDavid Sterba if (found_level == 0 && btrfs_check_leaf_full(eb)) { 673a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 674a826d6dcSJosef Bacik ret = -EIO; 675a826d6dcSJosef Bacik } 676ce9adaa5SChris Mason 677813fd1dcSDavid Sterba if (found_level > 0 && btrfs_check_node(eb)) 678053ab70fSLiu Bo ret = -EIO; 679053ab70fSLiu Bo 6800b32f4bbSJosef Bacik if (!ret) 6810b32f4bbSJosef Bacik set_extent_buffer_uptodate(eb); 68275391f0dSQu Wenruo else 68375391f0dSQu Wenruo btrfs_err(fs_info, 68475391f0dSQu Wenruo "block=%llu read time tree block corruption detected", 68575391f0dSQu Wenruo eb->start); 686ce9adaa5SChris Mason err: 68779fb65a1SJosef Bacik if (reads_done && 68879fb65a1SJosef Bacik test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 689d48d71aaSDavid Sterba btree_readahead_hook(eb, ret); 6904bb31e92SArne Jansen 69153b381b3SDavid Woodhouse if (ret) { 69253b381b3SDavid Woodhouse /* 69353b381b3SDavid Woodhouse * our io error hook is going to dec the io pages 69453b381b3SDavid Woodhouse * again, we have to make sure it has something 69553b381b3SDavid Woodhouse * to decrement 69653b381b3SDavid Woodhouse */ 69753b381b3SDavid Woodhouse atomic_inc(&eb->io_pages); 6980b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 69953b381b3SDavid Woodhouse } 7000b32f4bbSJosef Bacik free_extent_buffer(eb); 701ce9adaa5SChris Mason out: 702f188591eSChris Mason return ret; 703ce9adaa5SChris Mason } 704ce9adaa5SChris Mason 7054246a0b6SChristoph Hellwig static void end_workqueue_bio(struct bio *bio) 706ce9adaa5SChris Mason { 70797eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq = bio->bi_private; 708ce9adaa5SChris Mason struct btrfs_fs_info *fs_info; 7099e0af237SLiu Bo struct btrfs_workqueue *wq; 7109e0af237SLiu Bo btrfs_work_func_t func; 711ce9adaa5SChris Mason 712ce9adaa5SChris Mason fs_info = end_io_wq->info; 7134e4cbee9SChristoph Hellwig end_io_wq->status = bio->bi_status; 714d20f7043SChris Mason 71537226b21SMike Christie if (bio_op(bio) == REQ_OP_WRITE) { 7169e0af237SLiu Bo if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) { 7179e0af237SLiu Bo wq = fs_info->endio_meta_write_workers; 7189e0af237SLiu Bo func = btrfs_endio_meta_write_helper; 7199e0af237SLiu Bo } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) { 7209e0af237SLiu Bo wq = fs_info->endio_freespace_worker; 7219e0af237SLiu Bo func = btrfs_freespace_write_helper; 7229e0af237SLiu Bo } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) { 7239e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 7249e0af237SLiu Bo func = btrfs_endio_raid56_helper; 725d20f7043SChris Mason } else { 7269e0af237SLiu Bo wq = fs_info->endio_write_workers; 7279e0af237SLiu Bo func = btrfs_endio_write_helper; 728d20f7043SChris Mason } 7299e0af237SLiu Bo } else { 7308b110e39SMiao Xie if (unlikely(end_io_wq->metadata == 7318b110e39SMiao Xie BTRFS_WQ_ENDIO_DIO_REPAIR)) { 7328b110e39SMiao Xie wq = fs_info->endio_repair_workers; 7338b110e39SMiao Xie func = btrfs_endio_repair_helper; 7348b110e39SMiao Xie } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) { 7359e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 7369e0af237SLiu Bo func = btrfs_endio_raid56_helper; 7379e0af237SLiu Bo } else if (end_io_wq->metadata) { 7389e0af237SLiu Bo wq = fs_info->endio_meta_workers; 7399e0af237SLiu Bo func = btrfs_endio_meta_helper; 7409e0af237SLiu Bo } else { 7419e0af237SLiu Bo wq = fs_info->endio_workers; 7429e0af237SLiu Bo func = btrfs_endio_helper; 7439e0af237SLiu Bo } 7449e0af237SLiu Bo } 7459e0af237SLiu Bo 7469e0af237SLiu Bo btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL); 7479e0af237SLiu Bo btrfs_queue_work(wq, &end_io_wq->work); 748ce9adaa5SChris Mason } 749ce9adaa5SChris Mason 7504e4cbee9SChristoph Hellwig blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 751bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata) 7520b86a832SChris Mason { 75397eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 7548b110e39SMiao Xie 75597eb6b69SDavid Sterba end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS); 756ce9adaa5SChris Mason if (!end_io_wq) 7574e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 758ce9adaa5SChris Mason 759ce9adaa5SChris Mason end_io_wq->private = bio->bi_private; 760ce9adaa5SChris Mason end_io_wq->end_io = bio->bi_end_io; 76122c59948SChris Mason end_io_wq->info = info; 7624e4cbee9SChristoph Hellwig end_io_wq->status = 0; 763ce9adaa5SChris Mason end_io_wq->bio = bio; 76422c59948SChris Mason end_io_wq->metadata = metadata; 765ce9adaa5SChris Mason 766ce9adaa5SChris Mason bio->bi_private = end_io_wq; 767ce9adaa5SChris Mason bio->bi_end_io = end_workqueue_bio; 76822c59948SChris Mason return 0; 76922c59948SChris Mason } 77022c59948SChris Mason 7714a69a410SChris Mason static void run_one_async_start(struct btrfs_work *work) 7724a69a410SChris Mason { 7734a69a410SChris Mason struct async_submit_bio *async; 7744e4cbee9SChristoph Hellwig blk_status_t ret; 7754a69a410SChris Mason 7764a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 777c6100a4bSJosef Bacik ret = async->submit_bio_start(async->private_data, async->bio, 778eaf25d93SChris Mason async->bio_offset); 77979787eaaSJeff Mahoney if (ret) 7804e4cbee9SChristoph Hellwig async->status = ret; 7814a69a410SChris Mason } 7824a69a410SChris Mason 78306ea01b1SDavid Sterba /* 78406ea01b1SDavid Sterba * In order to insert checksums into the metadata in large chunks, we wait 78506ea01b1SDavid Sterba * until bio submission time. All the pages in the bio are checksummed and 78606ea01b1SDavid Sterba * sums are attached onto the ordered extent record. 78706ea01b1SDavid Sterba * 78806ea01b1SDavid Sterba * At IO completion time the csums attached on the ordered extent record are 78906ea01b1SDavid Sterba * inserted into the tree. 79006ea01b1SDavid Sterba */ 7914a69a410SChris Mason static void run_one_async_done(struct btrfs_work *work) 7928b712842SChris Mason { 7938b712842SChris Mason struct async_submit_bio *async; 79406ea01b1SDavid Sterba struct inode *inode; 79506ea01b1SDavid Sterba blk_status_t ret; 7968b712842SChris Mason 7978b712842SChris Mason async = container_of(work, struct async_submit_bio, work); 79806ea01b1SDavid Sterba inode = async->private_data; 7994854ddd0SChris Mason 800bb7ab3b9SAdam Buchbinder /* If an error occurred we just want to clean up the bio and move on */ 8014e4cbee9SChristoph Hellwig if (async->status) { 8024e4cbee9SChristoph Hellwig async->bio->bi_status = async->status; 8034246a0b6SChristoph Hellwig bio_endio(async->bio); 80479787eaaSJeff Mahoney return; 80579787eaaSJeff Mahoney } 80679787eaaSJeff Mahoney 80706ea01b1SDavid Sterba ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, 80806ea01b1SDavid Sterba async->mirror_num, 1); 80906ea01b1SDavid Sterba if (ret) { 81006ea01b1SDavid Sterba async->bio->bi_status = ret; 81106ea01b1SDavid Sterba bio_endio(async->bio); 81206ea01b1SDavid Sterba } 8134a69a410SChris Mason } 8144a69a410SChris Mason 8154a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 8164a69a410SChris Mason { 8174a69a410SChris Mason struct async_submit_bio *async; 8184a69a410SChris Mason 8194a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 8208b712842SChris Mason kfree(async); 8218b712842SChris Mason } 8228b712842SChris Mason 8238c27cb35SLinus Torvalds blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio, 824c6100a4bSJosef Bacik int mirror_num, unsigned long bio_flags, 825c6100a4bSJosef Bacik u64 bio_offset, void *private_data, 826e288c080SDavid Sterba extent_submit_bio_start_t *submit_bio_start) 82744b8bd7eSChris Mason { 82844b8bd7eSChris Mason struct async_submit_bio *async; 82944b8bd7eSChris Mason 83044b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 83144b8bd7eSChris Mason if (!async) 8324e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 83344b8bd7eSChris Mason 834c6100a4bSJosef Bacik async->private_data = private_data; 83544b8bd7eSChris Mason async->bio = bio; 83644b8bd7eSChris Mason async->mirror_num = mirror_num; 8374a69a410SChris Mason async->submit_bio_start = submit_bio_start; 8384a69a410SChris Mason 8399e0af237SLiu Bo btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start, 8405cdc7ad3SQu Wenruo run_one_async_done, run_one_async_free); 8414a69a410SChris Mason 842eaf25d93SChris Mason async->bio_offset = bio_offset; 8438c8bee1dSChris Mason 8444e4cbee9SChristoph Hellwig async->status = 0; 84579787eaaSJeff Mahoney 84667f055c7SChristoph Hellwig if (op_is_sync(bio->bi_opf)) 8475cdc7ad3SQu Wenruo btrfs_set_work_high_priority(&async->work); 848d313d7a3SChris Mason 8495cdc7ad3SQu Wenruo btrfs_queue_work(fs_info->workers, &async->work); 85044b8bd7eSChris Mason return 0; 85144b8bd7eSChris Mason } 85244b8bd7eSChris Mason 8534e4cbee9SChristoph Hellwig static blk_status_t btree_csum_one_bio(struct bio *bio) 854ce3ed71aSChris Mason { 8552c30c71bSKent Overstreet struct bio_vec *bvec; 856ce3ed71aSChris Mason struct btrfs_root *root; 8572b070cfeSChristoph Hellwig int ret = 0; 8586dc4f100SMing Lei struct bvec_iter_all iter_all; 859ce3ed71aSChris Mason 860c09abff8SDavid Sterba ASSERT(!bio_flagged(bio, BIO_CLONED)); 8612b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 862ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 86301d58472SDaniel Dressler ret = csum_dirty_buffer(root->fs_info, bvec->bv_page); 86479787eaaSJeff Mahoney if (ret) 86579787eaaSJeff Mahoney break; 866ce3ed71aSChris Mason } 8672c30c71bSKent Overstreet 8684e4cbee9SChristoph Hellwig return errno_to_blk_status(ret); 869ce3ed71aSChris Mason } 870ce3ed71aSChris Mason 871d0ee3934SDavid Sterba static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio, 872eaf25d93SChris Mason u64 bio_offset) 87322c59948SChris Mason { 8748b712842SChris Mason /* 8758b712842SChris Mason * when we're called for a write, we're already in the async 8765443be45SChris Mason * submission context. Just jump into btrfs_map_bio 8778b712842SChris Mason */ 87879787eaaSJeff Mahoney return btree_csum_one_bio(bio); 87922c59948SChris Mason } 88022c59948SChris Mason 8819b4e675aSDavid Sterba static int check_async_write(struct btrfs_fs_info *fs_info, 8829b4e675aSDavid Sterba struct btrfs_inode *bi) 883de0022b9SJosef Bacik { 8846300463bSLiu Bo if (atomic_read(&bi->sync_writers)) 8856300463bSLiu Bo return 0; 8869b4e675aSDavid Sterba if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags)) 887de0022b9SJosef Bacik return 0; 888de0022b9SJosef Bacik return 1; 889de0022b9SJosef Bacik } 890de0022b9SJosef Bacik 891a56b1c7bSNikolay Borisov static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio, 89250489a57SNikolay Borisov int mirror_num, 89350489a57SNikolay Borisov unsigned long bio_flags) 89444b8bd7eSChris Mason { 8950b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 8969b4e675aSDavid Sterba int async = check_async_write(fs_info, BTRFS_I(inode)); 8974e4cbee9SChristoph Hellwig blk_status_t ret; 898cad321adSChris Mason 89937226b21SMike Christie if (bio_op(bio) != REQ_OP_WRITE) { 900cad321adSChris Mason /* 901cad321adSChris Mason * called for a read, do the setup so that checksum validation 902cad321adSChris Mason * can happen in the async kernel threads 903cad321adSChris Mason */ 9040b246afaSJeff Mahoney ret = btrfs_bio_wq_end_io(fs_info, bio, 9050b246afaSJeff Mahoney BTRFS_WQ_ENDIO_METADATA); 9061d4284bdSChris Mason if (ret) 90761891923SStefan Behrens goto out_w_error; 9082ff7e61eSJeff Mahoney ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); 909de0022b9SJosef Bacik } else if (!async) { 910de0022b9SJosef Bacik ret = btree_csum_one_bio(bio); 911de0022b9SJosef Bacik if (ret) 91261891923SStefan Behrens goto out_w_error; 9132ff7e61eSJeff Mahoney ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); 91461891923SStefan Behrens } else { 915cad321adSChris Mason /* 91661891923SStefan Behrens * kthread helpers are used to submit writes so that 91761891923SStefan Behrens * checksumming can happen in parallel across all CPUs 918cad321adSChris Mason */ 919c6100a4bSJosef Bacik ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0, 920e68f2ee7SNikolay Borisov 0, inode, btree_submit_bio_start); 92144b8bd7eSChris Mason } 92244b8bd7eSChris Mason 9234246a0b6SChristoph Hellwig if (ret) 9244246a0b6SChristoph Hellwig goto out_w_error; 9254246a0b6SChristoph Hellwig return 0; 9264246a0b6SChristoph Hellwig 92761891923SStefan Behrens out_w_error: 9284e4cbee9SChristoph Hellwig bio->bi_status = ret; 9294246a0b6SChristoph Hellwig bio_endio(bio); 93061891923SStefan Behrens return ret; 93161891923SStefan Behrens } 93261891923SStefan Behrens 9333dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 934784b4e29SChris Mason static int btree_migratepage(struct address_space *mapping, 935a6bc32b8SMel Gorman struct page *newpage, struct page *page, 936a6bc32b8SMel Gorman enum migrate_mode mode) 937784b4e29SChris Mason { 938784b4e29SChris Mason /* 939784b4e29SChris Mason * we can't safely write a btree page from here, 940784b4e29SChris Mason * we haven't done the locking hook 941784b4e29SChris Mason */ 942784b4e29SChris Mason if (PageDirty(page)) 943784b4e29SChris Mason return -EAGAIN; 944784b4e29SChris Mason /* 945784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 946784b4e29SChris Mason * We must have no buffers or drop them. 947784b4e29SChris Mason */ 948784b4e29SChris Mason if (page_has_private(page) && 949784b4e29SChris Mason !try_to_release_page(page, GFP_KERNEL)) 950784b4e29SChris Mason return -EAGAIN; 951a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 952784b4e29SChris Mason } 9533dd1462eSJan Beulich #endif 954784b4e29SChris Mason 9550da5468fSChris Mason 9560da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 9570da5468fSChris Mason struct writeback_control *wbc) 9580da5468fSChris Mason { 959e2d84521SMiao Xie struct btrfs_fs_info *fs_info; 960e2d84521SMiao Xie int ret; 961e2d84521SMiao Xie 962d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 963448d640bSChris Mason 964448d640bSChris Mason if (wbc->for_kupdate) 965448d640bSChris Mason return 0; 966448d640bSChris Mason 967e2d84521SMiao Xie fs_info = BTRFS_I(mapping->host)->root->fs_info; 968b9473439SChris Mason /* this is a bit racy, but that's ok */ 969d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 970d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 971d814a491SEthan Lien fs_info->dirty_metadata_batch); 972e2d84521SMiao Xie if (ret < 0) 973793955bcSChris Mason return 0; 974793955bcSChris Mason } 9750b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 9760da5468fSChris Mason } 9770da5468fSChris Mason 978b2950863SChristoph Hellwig static int btree_readpage(struct file *file, struct page *page) 9795f39d397SChris Mason { 980d1310b2eSChris Mason struct extent_io_tree *tree; 981d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9828ddc7d9cSJan Schmidt return extent_read_full_page(tree, page, btree_get_extent, 0); 9835f39d397SChris Mason } 9845f39d397SChris Mason 98570dec807SChris Mason static int btree_releasepage(struct page *page, gfp_t gfp_flags) 9865f39d397SChris Mason { 98798509cfcSChris Mason if (PageWriteback(page) || PageDirty(page)) 98898509cfcSChris Mason return 0; 9890c4e538bSDavid Sterba 990f7a52a40SDavid Sterba return try_release_extent_buffer(page); 991d98237b3SChris Mason } 992d98237b3SChris Mason 993d47992f8SLukas Czerner static void btree_invalidatepage(struct page *page, unsigned int offset, 994d47992f8SLukas Czerner unsigned int length) 995d98237b3SChris Mason { 996d1310b2eSChris Mason struct extent_io_tree *tree; 997d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9985f39d397SChris Mason extent_invalidatepage(tree, page, offset); 9995f39d397SChris Mason btree_releasepage(page, GFP_NOFS); 10009ad6b7bcSChris Mason if (PagePrivate(page)) { 1001efe120a0SFrank Holton btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info, 1002efe120a0SFrank Holton "page private not zero on page %llu", 1003efe120a0SFrank Holton (unsigned long long)page_offset(page)); 10049ad6b7bcSChris Mason ClearPagePrivate(page); 10059ad6b7bcSChris Mason set_page_private(page, 0); 100609cbfeafSKirill A. Shutemov put_page(page); 10079ad6b7bcSChris Mason } 1008d98237b3SChris Mason } 1009d98237b3SChris Mason 10100b32f4bbSJosef Bacik static int btree_set_page_dirty(struct page *page) 10110b32f4bbSJosef Bacik { 1012bb146eb2SJosef Bacik #ifdef DEBUG 10130b32f4bbSJosef Bacik struct extent_buffer *eb; 10140b32f4bbSJosef Bacik 10150b32f4bbSJosef Bacik BUG_ON(!PagePrivate(page)); 10160b32f4bbSJosef Bacik eb = (struct extent_buffer *)page->private; 10170b32f4bbSJosef Bacik BUG_ON(!eb); 10180b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 10190b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 10200b32f4bbSJosef Bacik btrfs_assert_tree_locked(eb); 1021bb146eb2SJosef Bacik #endif 10220b32f4bbSJosef Bacik return __set_page_dirty_nobuffers(page); 10230b32f4bbSJosef Bacik } 10240b32f4bbSJosef Bacik 10257f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 1026d98237b3SChris Mason .readpage = btree_readpage, 10270da5468fSChris Mason .writepages = btree_writepages, 10285f39d397SChris Mason .releasepage = btree_releasepage, 10295f39d397SChris Mason .invalidatepage = btree_invalidatepage, 10305a92bc88SChris Mason #ifdef CONFIG_MIGRATION 1031784b4e29SChris Mason .migratepage = btree_migratepage, 10325a92bc88SChris Mason #endif 10330b32f4bbSJosef Bacik .set_page_dirty = btree_set_page_dirty, 1034d98237b3SChris Mason }; 1035123abc88SChris Mason 10362ff7e61eSJeff Mahoney void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr) 1037090d1875SChris Mason { 10385f39d397SChris Mason struct extent_buffer *buf = NULL; 1039537f38f0SNikolay Borisov int ret; 1040090d1875SChris Mason 10412ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 1042c871b0f2SLiu Bo if (IS_ERR(buf)) 10436197d86eSDavid Sterba return; 1044537f38f0SNikolay Borisov 1045c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(buf, WAIT_NONE, 0); 1046537f38f0SNikolay Borisov if (ret < 0) 1047537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 1048537f38f0SNikolay Borisov else 10495f39d397SChris Mason free_extent_buffer(buf); 1050090d1875SChris Mason } 1051090d1875SChris Mason 10522ff7e61eSJeff Mahoney int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr, 1053ab0fff03SArne Jansen int mirror_num, struct extent_buffer **eb) 1054ab0fff03SArne Jansen { 1055ab0fff03SArne Jansen struct extent_buffer *buf = NULL; 1056ab0fff03SArne Jansen int ret; 1057ab0fff03SArne Jansen 10582ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 1059c871b0f2SLiu Bo if (IS_ERR(buf)) 1060ab0fff03SArne Jansen return 0; 1061ab0fff03SArne Jansen 1062ab0fff03SArne Jansen set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags); 1063ab0fff03SArne Jansen 1064c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(buf, WAIT_PAGE_LOCK, mirror_num); 1065ab0fff03SArne Jansen if (ret) { 1066537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 1067ab0fff03SArne Jansen return ret; 1068ab0fff03SArne Jansen } 1069ab0fff03SArne Jansen 1070ab0fff03SArne Jansen if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) { 1071537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 1072ab0fff03SArne Jansen return -EIO; 10730b32f4bbSJosef Bacik } else if (extent_buffer_uptodate(buf)) { 1074ab0fff03SArne Jansen *eb = buf; 1075ab0fff03SArne Jansen } else { 1076ab0fff03SArne Jansen free_extent_buffer(buf); 1077ab0fff03SArne Jansen } 1078ab0fff03SArne Jansen return 0; 1079ab0fff03SArne Jansen } 1080ab0fff03SArne Jansen 10812ff7e61eSJeff Mahoney struct extent_buffer *btrfs_find_create_tree_block( 10822ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info, 10830308af44SDavid Sterba u64 bytenr) 10840999df54SChris Mason { 10850b246afaSJeff Mahoney if (btrfs_is_testing(fs_info)) 10860b246afaSJeff Mahoney return alloc_test_extent_buffer(fs_info, bytenr); 10870b246afaSJeff Mahoney return alloc_extent_buffer(fs_info, bytenr); 10880999df54SChris Mason } 10890999df54SChris Mason 1090581c1760SQu Wenruo /* 1091581c1760SQu Wenruo * Read tree block at logical address @bytenr and do variant basic but critical 1092581c1760SQu Wenruo * verification. 1093581c1760SQu Wenruo * 1094581c1760SQu Wenruo * @parent_transid: expected transid of this tree block, skip check if 0 1095581c1760SQu Wenruo * @level: expected level, mandatory check 1096581c1760SQu Wenruo * @first_key: expected key in slot 0, skip check if NULL 1097581c1760SQu Wenruo */ 10982ff7e61eSJeff Mahoney struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, 1099581c1760SQu Wenruo u64 parent_transid, int level, 1100581c1760SQu Wenruo struct btrfs_key *first_key) 1101e20d96d6SChris Mason { 11025f39d397SChris Mason struct extent_buffer *buf = NULL; 110319c00ddcSChris Mason int ret; 110419c00ddcSChris Mason 11052ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 1106c871b0f2SLiu Bo if (IS_ERR(buf)) 1107c871b0f2SLiu Bo return buf; 1108e4204dedSChris Mason 11095ab12d1fSDavid Sterba ret = btree_read_extent_buffer_pages(buf, parent_transid, 1110581c1760SQu Wenruo level, first_key); 11110f0fe8f7SFilipe David Borba Manana if (ret) { 1112537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 111364c043deSLiu Bo return ERR_PTR(ret); 11140f0fe8f7SFilipe David Borba Manana } 11155f39d397SChris Mason return buf; 1116ce9adaa5SChris Mason 1117eb60ceacSChris Mason } 1118eb60ceacSChris Mason 11196a884d7dSDavid Sterba void btrfs_clean_tree_block(struct extent_buffer *buf) 1120ed2ff2cbSChris Mason { 11216a884d7dSDavid Sterba struct btrfs_fs_info *fs_info = buf->fs_info; 112255c69072SChris Mason if (btrfs_header_generation(buf) == 1123e2d84521SMiao Xie fs_info->running_transaction->transid) { 1124b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1125b4ce94deSChris Mason 1126b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1127104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 1128e2d84521SMiao Xie -buf->len, 1129e2d84521SMiao Xie fs_info->dirty_metadata_batch); 1130b9473439SChris Mason /* ugh, clear_extent_buffer_dirty needs to lock the page */ 11318bead258SDavid Sterba btrfs_set_lock_blocking_write(buf); 11320b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 1133925baeddSChris Mason } 11345f39d397SChris Mason } 1135ed7b63ebSJosef Bacik } 11365f39d397SChris Mason 11378257b2dcSMiao Xie static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void) 11388257b2dcSMiao Xie { 11398257b2dcSMiao Xie struct btrfs_subvolume_writers *writers; 11408257b2dcSMiao Xie int ret; 11418257b2dcSMiao Xie 11428257b2dcSMiao Xie writers = kmalloc(sizeof(*writers), GFP_NOFS); 11438257b2dcSMiao Xie if (!writers) 11448257b2dcSMiao Xie return ERR_PTR(-ENOMEM); 11458257b2dcSMiao Xie 11468a5a916dSJeff Mahoney ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS); 11478257b2dcSMiao Xie if (ret < 0) { 11488257b2dcSMiao Xie kfree(writers); 11498257b2dcSMiao Xie return ERR_PTR(ret); 11508257b2dcSMiao Xie } 11518257b2dcSMiao Xie 11528257b2dcSMiao Xie init_waitqueue_head(&writers->wait); 11538257b2dcSMiao Xie return writers; 11548257b2dcSMiao Xie } 11558257b2dcSMiao Xie 11568257b2dcSMiao Xie static void 11578257b2dcSMiao Xie btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers) 11588257b2dcSMiao Xie { 11598257b2dcSMiao Xie percpu_counter_destroy(&writers->counter); 11608257b2dcSMiao Xie kfree(writers); 11618257b2dcSMiao Xie } 11628257b2dcSMiao Xie 1163da17066cSJeff Mahoney static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, 1164e20d96d6SChris Mason u64 objectid) 1165d97e63b6SChris Mason { 11667c0260eeSJeff Mahoney bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); 1167cfaa7295SChris Mason root->node = NULL; 1168a28ec197SChris Mason root->commit_root = NULL; 116927cdeb70SMiao Xie root->state = 0; 1170d68fc57bSYan, Zheng root->orphan_cleanup_state = 0; 11710b86a832SChris Mason 11720f7d52f4SChris Mason root->last_trans = 0; 117313a8a7c8SYan, Zheng root->highest_objectid = 0; 1174eb73c1b7SMiao Xie root->nr_delalloc_inodes = 0; 1175199c2a9cSMiao Xie root->nr_ordered_extents = 0; 11766bef4d31SEric Paris root->inode_tree = RB_ROOT; 117716cdcec7SMiao Xie INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1178f0486c68SYan, Zheng root->block_rsv = NULL; 11790b86a832SChris Mason 11800b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 11815d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 1182eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_inodes); 1183eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_root); 1184199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_extents); 1185199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_root); 1186d2311e69SQu Wenruo INIT_LIST_HEAD(&root->reloc_dirty_list); 11872ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 11882ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 11895d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1190eb73c1b7SMiao Xie spin_lock_init(&root->delalloc_lock); 1191199c2a9cSMiao Xie spin_lock_init(&root->ordered_extent_lock); 1192f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 11932ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 11942ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 11958287475aSQu Wenruo spin_lock_init(&root->qgroup_meta_rsv_lock); 1196a2135011SChris Mason mutex_init(&root->objectid_mutex); 1197e02119d5SChris Mason mutex_init(&root->log_mutex); 119831f3d255SMiao Xie mutex_init(&root->ordered_extent_mutex); 1199573bfb72SMiao Xie mutex_init(&root->delalloc_mutex); 12007237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 12017237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 12027237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 12038b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[0]); 12048b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[1]); 12057237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 12067237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 12077237f183SYan Zheng atomic_set(&root->log_writers, 0); 12082ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 12090700cea7SElena Reshetova refcount_set(&root->refs, 1); 1210ea14b57fSDavid Sterba atomic_set(&root->will_be_snapshotted, 0); 12118ecebf4dSRobbie Ko atomic_set(&root->snapshot_force_cow, 0); 1212eede2bf3SOmar Sandoval atomic_set(&root->nr_swapfiles, 0); 12137237f183SYan Zheng root->log_transid = 0; 1214d1433debSMiao Xie root->log_transid_committed = -1; 1215257c62e1SChris Mason root->last_log_commit = 0; 12167c0260eeSJeff Mahoney if (!dummy) 121743eb5f29SQu Wenruo extent_io_tree_init(fs_info, &root->dirty_log_pages, 121843eb5f29SQu Wenruo IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL); 1219017e5369SChris Mason 12203768f368SChris Mason memset(&root->root_key, 0, sizeof(root->root_key)); 12213768f368SChris Mason memset(&root->root_item, 0, sizeof(root->root_item)); 12226702ed49SChris Mason memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 12237c0260eeSJeff Mahoney if (!dummy) 12243f157a2fSChris Mason root->defrag_trans_start = fs_info->generation; 122506ea65a3SJosef Bacik else 122606ea65a3SJosef Bacik root->defrag_trans_start = 0; 12274d775673SChris Mason root->root_key.objectid = objectid; 12280ee5dc67SAl Viro root->anon_dev = 0; 12298ea05e3aSAlexander Block 12305f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 1231370a11b8SQu Wenruo btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks); 12323768f368SChris Mason } 12333768f368SChris Mason 123474e4d827SDavid Sterba static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, 123574e4d827SDavid Sterba gfp_t flags) 12366f07e42eSAl Viro { 123774e4d827SDavid Sterba struct btrfs_root *root = kzalloc(sizeof(*root), flags); 12386f07e42eSAl Viro if (root) 12396f07e42eSAl Viro root->fs_info = fs_info; 12406f07e42eSAl Viro return root; 12416f07e42eSAl Viro } 12426f07e42eSAl Viro 124306ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 124406ea65a3SJosef Bacik /* Should only be used by the testing infrastructure */ 1245da17066cSJeff Mahoney struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) 124606ea65a3SJosef Bacik { 124706ea65a3SJosef Bacik struct btrfs_root *root; 124806ea65a3SJosef Bacik 12497c0260eeSJeff Mahoney if (!fs_info) 12507c0260eeSJeff Mahoney return ERR_PTR(-EINVAL); 12517c0260eeSJeff Mahoney 12527c0260eeSJeff Mahoney root = btrfs_alloc_root(fs_info, GFP_KERNEL); 125306ea65a3SJosef Bacik if (!root) 125406ea65a3SJosef Bacik return ERR_PTR(-ENOMEM); 1255da17066cSJeff Mahoney 1256b9ef22deSFeifei Xu /* We don't use the stripesize in selftest, set it as sectorsize */ 1257da17066cSJeff Mahoney __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID); 1258faa2dbf0SJosef Bacik root->alloc_bytenr = 0; 125906ea65a3SJosef Bacik 126006ea65a3SJosef Bacik return root; 126106ea65a3SJosef Bacik } 126206ea65a3SJosef Bacik #endif 126306ea65a3SJosef Bacik 126420897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 126520897f5cSArne Jansen u64 objectid) 126620897f5cSArne Jansen { 12679b7a2440SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 126820897f5cSArne Jansen struct extent_buffer *leaf; 126920897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 127020897f5cSArne Jansen struct btrfs_root *root; 127120897f5cSArne Jansen struct btrfs_key key; 1272b89f6d1fSFilipe Manana unsigned int nofs_flag; 127320897f5cSArne Jansen int ret = 0; 127433d85fdaSQu Wenruo uuid_le uuid = NULL_UUID_LE; 127520897f5cSArne Jansen 1276b89f6d1fSFilipe Manana /* 1277b89f6d1fSFilipe Manana * We're holding a transaction handle, so use a NOFS memory allocation 1278b89f6d1fSFilipe Manana * context to avoid deadlock if reclaim happens. 1279b89f6d1fSFilipe Manana */ 1280b89f6d1fSFilipe Manana nofs_flag = memalloc_nofs_save(); 128174e4d827SDavid Sterba root = btrfs_alloc_root(fs_info, GFP_KERNEL); 1282b89f6d1fSFilipe Manana memalloc_nofs_restore(nofs_flag); 128320897f5cSArne Jansen if (!root) 128420897f5cSArne Jansen return ERR_PTR(-ENOMEM); 128520897f5cSArne Jansen 1286da17066cSJeff Mahoney __setup_root(root, fs_info, objectid); 128720897f5cSArne Jansen root->root_key.objectid = objectid; 128820897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 128920897f5cSArne Jansen root->root_key.offset = 0; 129020897f5cSArne Jansen 12914d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0); 129220897f5cSArne Jansen if (IS_ERR(leaf)) { 129320897f5cSArne Jansen ret = PTR_ERR(leaf); 12941dd05682STsutomu Itoh leaf = NULL; 129520897f5cSArne Jansen goto fail; 129620897f5cSArne Jansen } 129720897f5cSArne Jansen 129820897f5cSArne Jansen root->node = leaf; 129920897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 130020897f5cSArne Jansen 130120897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 130227cdeb70SMiao Xie set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 130320897f5cSArne Jansen 130420897f5cSArne Jansen root->root_item.flags = 0; 130520897f5cSArne Jansen root->root_item.byte_limit = 0; 130620897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 130720897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 130820897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 130920897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 131020897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 131120897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 131220897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 131333d85fdaSQu Wenruo if (is_fstree(objectid)) 13146463fe58SStefan Behrens uuid_le_gen(&uuid); 13156463fe58SStefan Behrens memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE); 131620897f5cSArne Jansen root->root_item.drop_level = 0; 131720897f5cSArne Jansen 131820897f5cSArne Jansen key.objectid = objectid; 131920897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 132020897f5cSArne Jansen key.offset = 0; 132120897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 132220897f5cSArne Jansen if (ret) 132320897f5cSArne Jansen goto fail; 132420897f5cSArne Jansen 132520897f5cSArne Jansen btrfs_tree_unlock(leaf); 132620897f5cSArne Jansen 132720897f5cSArne Jansen return root; 13281dd05682STsutomu Itoh 13291dd05682STsutomu Itoh fail: 13301dd05682STsutomu Itoh if (leaf) { 13311dd05682STsutomu Itoh btrfs_tree_unlock(leaf); 133259885b39STsutomu Itoh free_extent_buffer(root->commit_root); 13331dd05682STsutomu Itoh free_extent_buffer(leaf); 13341dd05682STsutomu Itoh } 13351dd05682STsutomu Itoh kfree(root); 13361dd05682STsutomu Itoh 13371dd05682STsutomu Itoh return ERR_PTR(ret); 133820897f5cSArne Jansen } 133920897f5cSArne Jansen 13407237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1341e02119d5SChris Mason struct btrfs_fs_info *fs_info) 13420f7d52f4SChris Mason { 13430f7d52f4SChris Mason struct btrfs_root *root; 13447237f183SYan Zheng struct extent_buffer *leaf; 1345e02119d5SChris Mason 134674e4d827SDavid Sterba root = btrfs_alloc_root(fs_info, GFP_NOFS); 1347e02119d5SChris Mason if (!root) 13487237f183SYan Zheng return ERR_PTR(-ENOMEM); 1349e02119d5SChris Mason 1350da17066cSJeff Mahoney __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID); 1351e02119d5SChris Mason 1352e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1353e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1354e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 135527cdeb70SMiao Xie 13567237f183SYan Zheng /* 135727cdeb70SMiao Xie * DON'T set REF_COWS for log trees 135827cdeb70SMiao Xie * 13597237f183SYan Zheng * log trees do not get reference counted because they go away 13607237f183SYan Zheng * before a real commit is actually done. They do store pointers 13617237f183SYan Zheng * to file data extents, and those reference counts still get 13627237f183SYan Zheng * updated (along with back refs to the log tree). 13637237f183SYan Zheng */ 1364e02119d5SChris Mason 13654d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 13664d75f8a9SDavid Sterba NULL, 0, 0, 0); 13677237f183SYan Zheng if (IS_ERR(leaf)) { 13687237f183SYan Zheng kfree(root); 13697237f183SYan Zheng return ERR_CAST(leaf); 13707237f183SYan Zheng } 1371e02119d5SChris Mason 13727237f183SYan Zheng root->node = leaf; 1373e02119d5SChris Mason 1374e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1375e02119d5SChris Mason btrfs_tree_unlock(root->node); 13767237f183SYan Zheng return root; 13777237f183SYan Zheng } 13787237f183SYan Zheng 13797237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 13807237f183SYan Zheng struct btrfs_fs_info *fs_info) 13817237f183SYan Zheng { 13827237f183SYan Zheng struct btrfs_root *log_root; 13837237f183SYan Zheng 13847237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 13857237f183SYan Zheng if (IS_ERR(log_root)) 13867237f183SYan Zheng return PTR_ERR(log_root); 13877237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 13887237f183SYan Zheng fs_info->log_root_tree = log_root; 13897237f183SYan Zheng return 0; 13907237f183SYan Zheng } 13917237f183SYan Zheng 13927237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 13937237f183SYan Zheng struct btrfs_root *root) 13947237f183SYan Zheng { 13950b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13967237f183SYan Zheng struct btrfs_root *log_root; 13977237f183SYan Zheng struct btrfs_inode_item *inode_item; 13987237f183SYan Zheng 13990b246afaSJeff Mahoney log_root = alloc_log_tree(trans, fs_info); 14007237f183SYan Zheng if (IS_ERR(log_root)) 14017237f183SYan Zheng return PTR_ERR(log_root); 14027237f183SYan Zheng 14037237f183SYan Zheng log_root->last_trans = trans->transid; 14047237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 14057237f183SYan Zheng 14067237f183SYan Zheng inode_item = &log_root->root_item.inode; 14073cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 14083cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 14093cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 1410da17066cSJeff Mahoney btrfs_set_stack_inode_nbytes(inode_item, 14110b246afaSJeff Mahoney fs_info->nodesize); 14123cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 14137237f183SYan Zheng 14145d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 14157237f183SYan Zheng 14167237f183SYan Zheng WARN_ON(root->log_root); 14177237f183SYan Zheng root->log_root = log_root; 14187237f183SYan Zheng root->log_transid = 0; 1419d1433debSMiao Xie root->log_transid_committed = -1; 1420257c62e1SChris Mason root->last_log_commit = 0; 1421e02119d5SChris Mason return 0; 1422e02119d5SChris Mason } 1423e02119d5SChris Mason 142435a3621bSStefan Behrens static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 1425cb517eabSMiao Xie struct btrfs_key *key) 1426e02119d5SChris Mason { 1427e02119d5SChris Mason struct btrfs_root *root; 1428e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 14290f7d52f4SChris Mason struct btrfs_path *path; 143084234f3aSYan Zheng u64 generation; 1431cb517eabSMiao Xie int ret; 1432581c1760SQu Wenruo int level; 1433cb517eabSMiao Xie 1434cb517eabSMiao Xie path = btrfs_alloc_path(); 1435cb517eabSMiao Xie if (!path) 1436cb517eabSMiao Xie return ERR_PTR(-ENOMEM); 14370f7d52f4SChris Mason 143874e4d827SDavid Sterba root = btrfs_alloc_root(fs_info, GFP_NOFS); 1439cb517eabSMiao Xie if (!root) { 1440cb517eabSMiao Xie ret = -ENOMEM; 1441cb517eabSMiao Xie goto alloc_fail; 14420f7d52f4SChris Mason } 14430f7d52f4SChris Mason 1444da17066cSJeff Mahoney __setup_root(root, fs_info, key->objectid); 14450f7d52f4SChris Mason 1446cb517eabSMiao Xie ret = btrfs_find_root(tree_root, key, path, 1447cb517eabSMiao Xie &root->root_item, &root->root_key); 14480f7d52f4SChris Mason if (ret) { 144913a8a7c8SYan, Zheng if (ret > 0) 145013a8a7c8SYan, Zheng ret = -ENOENT; 1451cb517eabSMiao Xie goto find_fail; 14520f7d52f4SChris Mason } 145313a8a7c8SYan, Zheng 145484234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1455581c1760SQu Wenruo level = btrfs_root_level(&root->root_item); 14562ff7e61eSJeff Mahoney root->node = read_tree_block(fs_info, 14572ff7e61eSJeff Mahoney btrfs_root_bytenr(&root->root_item), 1458581c1760SQu Wenruo generation, level, NULL); 145964c043deSLiu Bo if (IS_ERR(root->node)) { 146064c043deSLiu Bo ret = PTR_ERR(root->node); 1461cb517eabSMiao Xie goto find_fail; 1462cb517eabSMiao Xie } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1463cb517eabSMiao Xie ret = -EIO; 146464c043deSLiu Bo free_extent_buffer(root->node); 146564c043deSLiu Bo goto find_fail; 1466416bc658SJosef Bacik } 14675d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 146813a8a7c8SYan, Zheng out: 1469cb517eabSMiao Xie btrfs_free_path(path); 1470cb517eabSMiao Xie return root; 1471cb517eabSMiao Xie 1472cb517eabSMiao Xie find_fail: 1473cb517eabSMiao Xie kfree(root); 1474cb517eabSMiao Xie alloc_fail: 1475cb517eabSMiao Xie root = ERR_PTR(ret); 1476cb517eabSMiao Xie goto out; 1477cb517eabSMiao Xie } 1478cb517eabSMiao Xie 1479cb517eabSMiao Xie struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root, 1480cb517eabSMiao Xie struct btrfs_key *location) 1481cb517eabSMiao Xie { 1482cb517eabSMiao Xie struct btrfs_root *root; 1483cb517eabSMiao Xie 1484cb517eabSMiao Xie root = btrfs_read_tree_root(tree_root, location); 1485cb517eabSMiao Xie if (IS_ERR(root)) 1486cb517eabSMiao Xie return root; 1487cb517eabSMiao Xie 1488cb517eabSMiao Xie if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { 148927cdeb70SMiao Xie set_bit(BTRFS_ROOT_REF_COWS, &root->state); 149008fe4db1SLi Zefan btrfs_check_and_init_root_item(&root->root_item); 149108fe4db1SLi Zefan } 149213a8a7c8SYan, Zheng 14935eda7b5eSChris Mason return root; 14945eda7b5eSChris Mason } 14955eda7b5eSChris Mason 1496cb517eabSMiao Xie int btrfs_init_fs_root(struct btrfs_root *root) 1497cb517eabSMiao Xie { 1498cb517eabSMiao Xie int ret; 14998257b2dcSMiao Xie struct btrfs_subvolume_writers *writers; 1500cb517eabSMiao Xie 1501cb517eabSMiao Xie root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); 1502cb517eabSMiao Xie root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), 1503cb517eabSMiao Xie GFP_NOFS); 1504cb517eabSMiao Xie if (!root->free_ino_pinned || !root->free_ino_ctl) { 1505cb517eabSMiao Xie ret = -ENOMEM; 1506cb517eabSMiao Xie goto fail; 1507cb517eabSMiao Xie } 1508cb517eabSMiao Xie 15098257b2dcSMiao Xie writers = btrfs_alloc_subvolume_writers(); 15108257b2dcSMiao Xie if (IS_ERR(writers)) { 15118257b2dcSMiao Xie ret = PTR_ERR(writers); 15128257b2dcSMiao Xie goto fail; 15138257b2dcSMiao Xie } 15148257b2dcSMiao Xie root->subv_writers = writers; 15158257b2dcSMiao Xie 1516cb517eabSMiao Xie btrfs_init_free_ino_ctl(root); 151757cdc8dbSDavid Sterba spin_lock_init(&root->ino_cache_lock); 151857cdc8dbSDavid Sterba init_waitqueue_head(&root->ino_cache_wait); 1519cb517eabSMiao Xie 1520cb517eabSMiao Xie ret = get_anon_bdev(&root->anon_dev); 1521cb517eabSMiao Xie if (ret) 1522876d2cf1SLiu Bo goto fail; 1523f32e48e9SChandan Rajendra 1524f32e48e9SChandan Rajendra mutex_lock(&root->objectid_mutex); 1525f32e48e9SChandan Rajendra ret = btrfs_find_highest_objectid(root, 1526f32e48e9SChandan Rajendra &root->highest_objectid); 1527f32e48e9SChandan Rajendra if (ret) { 1528f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1529876d2cf1SLiu Bo goto fail; 1530f32e48e9SChandan Rajendra } 1531f32e48e9SChandan Rajendra 1532f32e48e9SChandan Rajendra ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 1533f32e48e9SChandan Rajendra 1534f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1535f32e48e9SChandan Rajendra 1536cb517eabSMiao Xie return 0; 1537cb517eabSMiao Xie fail: 153884db5ccfSDavid Sterba /* The caller is responsible to call btrfs_free_fs_root */ 1539cb517eabSMiao Xie return ret; 1540cb517eabSMiao Xie } 1541cb517eabSMiao Xie 154235bbb97fSJeff Mahoney struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1543cb517eabSMiao Xie u64 root_id) 1544cb517eabSMiao Xie { 1545cb517eabSMiao Xie struct btrfs_root *root; 1546cb517eabSMiao Xie 1547cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1548cb517eabSMiao Xie root = radix_tree_lookup(&fs_info->fs_roots_radix, 1549cb517eabSMiao Xie (unsigned long)root_id); 1550cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1551cb517eabSMiao Xie return root; 1552cb517eabSMiao Xie } 1553cb517eabSMiao Xie 1554cb517eabSMiao Xie int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1555cb517eabSMiao Xie struct btrfs_root *root) 1556cb517eabSMiao Xie { 1557cb517eabSMiao Xie int ret; 1558cb517eabSMiao Xie 1559e1860a77SDavid Sterba ret = radix_tree_preload(GFP_NOFS); 1560cb517eabSMiao Xie if (ret) 1561cb517eabSMiao Xie return ret; 1562cb517eabSMiao Xie 1563cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1564cb517eabSMiao Xie ret = radix_tree_insert(&fs_info->fs_roots_radix, 1565cb517eabSMiao Xie (unsigned long)root->root_key.objectid, 1566cb517eabSMiao Xie root); 1567cb517eabSMiao Xie if (ret == 0) 156827cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1569cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1570cb517eabSMiao Xie radix_tree_preload_end(); 1571cb517eabSMiao Xie 1572cb517eabSMiao Xie return ret; 1573cb517eabSMiao Xie } 1574cb517eabSMiao Xie 1575c00869f1SMiao Xie struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 1576c00869f1SMiao Xie struct btrfs_key *location, 1577c00869f1SMiao Xie bool check_ref) 15785eda7b5eSChris Mason { 15795eda7b5eSChris Mason struct btrfs_root *root; 1580381cf658SDavid Sterba struct btrfs_path *path; 15811d4c08e0SDavid Sterba struct btrfs_key key; 15825eda7b5eSChris Mason int ret; 15835eda7b5eSChris Mason 1584edbd8d4eSChris Mason if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 1585edbd8d4eSChris Mason return fs_info->tree_root; 1586edbd8d4eSChris Mason if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID) 1587edbd8d4eSChris Mason return fs_info->extent_root; 15888f18cf13SChris Mason if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID) 15898f18cf13SChris Mason return fs_info->chunk_root; 15908f18cf13SChris Mason if (location->objectid == BTRFS_DEV_TREE_OBJECTID) 15918f18cf13SChris Mason return fs_info->dev_root; 15920403e47eSYan Zheng if (location->objectid == BTRFS_CSUM_TREE_OBJECTID) 15930403e47eSYan Zheng return fs_info->csum_root; 1594bcef60f2SArne Jansen if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID) 1595bcef60f2SArne Jansen return fs_info->quota_root ? fs_info->quota_root : 1596bcef60f2SArne Jansen ERR_PTR(-ENOENT); 1597f7a81ea4SStefan Behrens if (location->objectid == BTRFS_UUID_TREE_OBJECTID) 1598f7a81ea4SStefan Behrens return fs_info->uuid_root ? fs_info->uuid_root : 1599f7a81ea4SStefan Behrens ERR_PTR(-ENOENT); 160070f6d82eSOmar Sandoval if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) 160170f6d82eSOmar Sandoval return fs_info->free_space_root ? fs_info->free_space_root : 160270f6d82eSOmar Sandoval ERR_PTR(-ENOENT); 16034df27c4dSYan, Zheng again: 1604cb517eabSMiao Xie root = btrfs_lookup_fs_root(fs_info, location->objectid); 160548475471SStefan Behrens if (root) { 1606c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) 160748475471SStefan Behrens return ERR_PTR(-ENOENT); 16085eda7b5eSChris Mason return root; 160948475471SStefan Behrens } 16105eda7b5eSChris Mason 1611cb517eabSMiao Xie root = btrfs_read_fs_root(fs_info->tree_root, location); 16125eda7b5eSChris Mason if (IS_ERR(root)) 16135eda7b5eSChris Mason return root; 16143394e160SChris Mason 1615c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1616d68fc57bSYan, Zheng ret = -ENOENT; 1617d68fc57bSYan, Zheng goto fail; 1618d68fc57bSYan, Zheng } 1619d68fc57bSYan, Zheng 1620cb517eabSMiao Xie ret = btrfs_init_fs_root(root); 1621cb517eabSMiao Xie if (ret) 1622cb517eabSMiao Xie goto fail; 1623cb517eabSMiao Xie 1624381cf658SDavid Sterba path = btrfs_alloc_path(); 1625381cf658SDavid Sterba if (!path) { 1626381cf658SDavid Sterba ret = -ENOMEM; 1627381cf658SDavid Sterba goto fail; 1628381cf658SDavid Sterba } 16291d4c08e0SDavid Sterba key.objectid = BTRFS_ORPHAN_OBJECTID; 16301d4c08e0SDavid Sterba key.type = BTRFS_ORPHAN_ITEM_KEY; 16311d4c08e0SDavid Sterba key.offset = location->objectid; 16321d4c08e0SDavid Sterba 16331d4c08e0SDavid Sterba ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1634381cf658SDavid Sterba btrfs_free_path(path); 1635d68fc57bSYan, Zheng if (ret < 0) 1636d68fc57bSYan, Zheng goto fail; 1637d68fc57bSYan, Zheng if (ret == 0) 163827cdeb70SMiao Xie set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1639d68fc57bSYan, Zheng 1640cb517eabSMiao Xie ret = btrfs_insert_fs_root(fs_info, root); 16410f7d52f4SChris Mason if (ret) { 16424df27c4dSYan, Zheng if (ret == -EEXIST) { 164384db5ccfSDavid Sterba btrfs_free_fs_root(root); 16444df27c4dSYan, Zheng goto again; 16450f7d52f4SChris Mason } 16464df27c4dSYan, Zheng goto fail; 16474df27c4dSYan, Zheng } 1648edbd8d4eSChris Mason return root; 16494df27c4dSYan, Zheng fail: 165084db5ccfSDavid Sterba btrfs_free_fs_root(root); 16514df27c4dSYan, Zheng return ERR_PTR(ret); 1652edbd8d4eSChris Mason } 1653edbd8d4eSChris Mason 165404160088SChris Mason static int btrfs_congested_fn(void *congested_data, int bdi_bits) 165504160088SChris Mason { 165604160088SChris Mason struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; 165704160088SChris Mason int ret = 0; 165804160088SChris Mason struct btrfs_device *device; 165904160088SChris Mason struct backing_dev_info *bdi; 1660b7967db7SChris Mason 16611f78160cSXiao Guangrong rcu_read_lock(); 16621f78160cSXiao Guangrong list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) { 1663dfe25020SChris Mason if (!device->bdev) 1664dfe25020SChris Mason continue; 1665efa7c9f9SJan Kara bdi = device->bdev->bd_bdi; 1666ff9ea323STejun Heo if (bdi_congested(bdi, bdi_bits)) { 166704160088SChris Mason ret = 1; 166804160088SChris Mason break; 166904160088SChris Mason } 167004160088SChris Mason } 16711f78160cSXiao Guangrong rcu_read_unlock(); 167204160088SChris Mason return ret; 167304160088SChris Mason } 167404160088SChris Mason 16758b712842SChris Mason /* 16768b712842SChris Mason * called by the kthread helper functions to finally call the bio end_io 16778b712842SChris Mason * functions. This is where read checksum verification actually happens 16788b712842SChris Mason */ 16798b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work) 1680ce9adaa5SChris Mason { 1681ce9adaa5SChris Mason struct bio *bio; 168297eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 1683ce9adaa5SChris Mason 168497eb6b69SDavid Sterba end_io_wq = container_of(work, struct btrfs_end_io_wq, work); 1685ce9adaa5SChris Mason bio = end_io_wq->bio; 16868b712842SChris Mason 16874e4cbee9SChristoph Hellwig bio->bi_status = end_io_wq->status; 1688ce9adaa5SChris Mason bio->bi_private = end_io_wq->private; 1689ce9adaa5SChris Mason bio->bi_end_io = end_io_wq->end_io; 169097eb6b69SDavid Sterba kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq); 16914246a0b6SChristoph Hellwig bio_endio(bio); 1692ce9adaa5SChris Mason } 169344b8bd7eSChris Mason 1694a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1695a74a4b97SChris Mason { 1696a74a4b97SChris Mason struct btrfs_root *root = arg; 16970b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1698d0278245SMiao Xie int again; 1699a74a4b97SChris Mason 1700d6fd0ae2SOmar Sandoval while (1) { 1701d0278245SMiao Xie again = 0; 17029d1a2a3aSDavid Sterba 1703fd340d0fSJosef Bacik set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1704fd340d0fSJosef Bacik 1705d0278245SMiao Xie /* Make the cleaner go to sleep early. */ 17062ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) 1707d0278245SMiao Xie goto sleep; 1708d0278245SMiao Xie 170990c711abSZygo Blaxell /* 171090c711abSZygo Blaxell * Do not do anything if we might cause open_ctree() to block 171190c711abSZygo Blaxell * before we have finished mounting the filesystem. 171290c711abSZygo Blaxell */ 17130b246afaSJeff Mahoney if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 171490c711abSZygo Blaxell goto sleep; 171590c711abSZygo Blaxell 17160b246afaSJeff Mahoney if (!mutex_trylock(&fs_info->cleaner_mutex)) 1717d0278245SMiao Xie goto sleep; 1718d0278245SMiao Xie 1719dc7f370cSMiao Xie /* 1720dc7f370cSMiao Xie * Avoid the problem that we change the status of the fs 1721dc7f370cSMiao Xie * during the above check and trylock. 1722dc7f370cSMiao Xie */ 17232ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) { 17240b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1725dc7f370cSMiao Xie goto sleep; 1726dc7f370cSMiao Xie } 1727dc7f370cSMiao Xie 17282ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 1729c2d6cb16SFilipe Manana 17309d1a2a3aSDavid Sterba again = btrfs_clean_one_deleted_snapshot(root); 17310b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1732a74a4b97SChris Mason 1733d0278245SMiao Xie /* 173405323cd1SMiao Xie * The defragger has dealt with the R/O remount and umount, 173505323cd1SMiao Xie * needn't do anything special here. 1736d0278245SMiao Xie */ 17370b246afaSJeff Mahoney btrfs_run_defrag_inodes(fs_info); 173867c5e7d4SFilipe Manana 173967c5e7d4SFilipe Manana /* 174067c5e7d4SFilipe Manana * Acquires fs_info->delete_unused_bgs_mutex to avoid racing 174167c5e7d4SFilipe Manana * with relocation (btrfs_relocate_chunk) and relocation 174267c5e7d4SFilipe Manana * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 174367c5e7d4SFilipe Manana * after acquiring fs_info->delete_unused_bgs_mutex. So we 174467c5e7d4SFilipe Manana * can't hold, nor need to, fs_info->cleaner_mutex when deleting 174567c5e7d4SFilipe Manana * unused block groups. 174667c5e7d4SFilipe Manana */ 17470b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 1748d0278245SMiao Xie sleep: 1749fd340d0fSJosef Bacik clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1750d6fd0ae2SOmar Sandoval if (kthread_should_park()) 1751d6fd0ae2SOmar Sandoval kthread_parkme(); 1752d6fd0ae2SOmar Sandoval if (kthread_should_stop()) 1753d6fd0ae2SOmar Sandoval return 0; 1754838fe188SJiri Kosina if (!again) { 1755a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 1756a74a4b97SChris Mason schedule(); 1757a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1758a74a4b97SChris Mason } 1759da288d28SFilipe Manana } 1760a74a4b97SChris Mason } 1761a74a4b97SChris Mason 1762a74a4b97SChris Mason static int transaction_kthread(void *arg) 1763a74a4b97SChris Mason { 1764a74a4b97SChris Mason struct btrfs_root *root = arg; 17650b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1766a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1767a74a4b97SChris Mason struct btrfs_transaction *cur; 17688929ecfaSYan, Zheng u64 transid; 1769a944442cSAllen Pais time64_t now; 1770a74a4b97SChris Mason unsigned long delay; 1771914b2007SJan Kara bool cannot_commit; 1772a74a4b97SChris Mason 1773a74a4b97SChris Mason do { 1774914b2007SJan Kara cannot_commit = false; 17750b246afaSJeff Mahoney delay = HZ * fs_info->commit_interval; 17760b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 1777a74a4b97SChris Mason 17780b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 17790b246afaSJeff Mahoney cur = fs_info->running_transaction; 1780a74a4b97SChris Mason if (!cur) { 17810b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1782a74a4b97SChris Mason goto sleep; 1783a74a4b97SChris Mason } 178431153d81SYan Zheng 1785afd48513SArnd Bergmann now = ktime_get_seconds(); 17864a9d8bdeSMiao Xie if (cur->state < TRANS_STATE_BLOCKED && 1787a514d638SQu Wenruo !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) && 17888b87dc17SDavid Sterba (now < cur->start_time || 17890b246afaSJeff Mahoney now - cur->start_time < fs_info->commit_interval)) { 17900b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1791a74a4b97SChris Mason delay = HZ * 5; 1792a74a4b97SChris Mason goto sleep; 1793a74a4b97SChris Mason } 17948929ecfaSYan, Zheng transid = cur->transid; 17950b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 179656bec294SChris Mason 179779787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 1798354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 1799914b2007SJan Kara if (IS_ERR(trans)) { 1800354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 1801914b2007SJan Kara cannot_commit = true; 180279787eaaSJeff Mahoney goto sleep; 1803914b2007SJan Kara } 18048929ecfaSYan, Zheng if (transid == trans->transid) { 18053a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 18068929ecfaSYan, Zheng } else { 18073a45bb20SJeff Mahoney btrfs_end_transaction(trans); 18088929ecfaSYan, Zheng } 1809a74a4b97SChris Mason sleep: 18100b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 18110b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 1812a74a4b97SChris Mason 18134e121c06SJosef Bacik if (unlikely(test_bit(BTRFS_FS_STATE_ERROR, 18140b246afaSJeff Mahoney &fs_info->fs_state))) 18152ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 18168929ecfaSYan, Zheng if (!kthread_should_stop() && 18170b246afaSJeff Mahoney (!btrfs_transaction_blocked(fs_info) || 1818914b2007SJan Kara cannot_commit)) 1819bc5511d0SNikolay Borisov schedule_timeout_interruptible(delay); 1820a74a4b97SChris Mason } while (!kthread_should_stop()); 1821a74a4b97SChris Mason return 0; 1822a74a4b97SChris Mason } 1823a74a4b97SChris Mason 1824af31f5e5SChris Mason /* 1825af31f5e5SChris Mason * this will find the highest generation in the array of 1826af31f5e5SChris Mason * root backups. The index of the highest array is returned, 1827af31f5e5SChris Mason * or -1 if we can't find anything. 1828af31f5e5SChris Mason * 1829af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1830af31f5e5SChris Mason * generation of the latest root in the array with the generation 1831af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1832af31f5e5SChris Mason */ 1833af31f5e5SChris Mason static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen) 1834af31f5e5SChris Mason { 1835af31f5e5SChris Mason u64 cur; 1836af31f5e5SChris Mason int newest_index = -1; 1837af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1838af31f5e5SChris Mason int i; 1839af31f5e5SChris Mason 1840af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1841af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1842af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1843af31f5e5SChris Mason if (cur == newest_gen) 1844af31f5e5SChris Mason newest_index = i; 1845af31f5e5SChris Mason } 1846af31f5e5SChris Mason 1847af31f5e5SChris Mason /* check to see if we actually wrapped around */ 1848af31f5e5SChris Mason if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) { 1849af31f5e5SChris Mason root_backup = info->super_copy->super_roots; 1850af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1851af31f5e5SChris Mason if (cur == newest_gen) 1852af31f5e5SChris Mason newest_index = 0; 1853af31f5e5SChris Mason } 1854af31f5e5SChris Mason return newest_index; 1855af31f5e5SChris Mason } 1856af31f5e5SChris Mason 1857af31f5e5SChris Mason 1858af31f5e5SChris Mason /* 1859af31f5e5SChris Mason * find the oldest backup so we know where to store new entries 1860af31f5e5SChris Mason * in the backup array. This will set the backup_root_index 1861af31f5e5SChris Mason * field in the fs_info struct 1862af31f5e5SChris Mason */ 1863af31f5e5SChris Mason static void find_oldest_super_backup(struct btrfs_fs_info *info, 1864af31f5e5SChris Mason u64 newest_gen) 1865af31f5e5SChris Mason { 1866af31f5e5SChris Mason int newest_index = -1; 1867af31f5e5SChris Mason 1868af31f5e5SChris Mason newest_index = find_newest_super_backup(info, newest_gen); 1869af31f5e5SChris Mason /* if there was garbage in there, just move along */ 1870af31f5e5SChris Mason if (newest_index == -1) { 1871af31f5e5SChris Mason info->backup_root_index = 0; 1872af31f5e5SChris Mason } else { 1873af31f5e5SChris Mason info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS; 1874af31f5e5SChris Mason } 1875af31f5e5SChris Mason } 1876af31f5e5SChris Mason 1877af31f5e5SChris Mason /* 1878af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1879af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1880af31f5e5SChris Mason * done 1881af31f5e5SChris Mason */ 1882af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1883af31f5e5SChris Mason { 1884af31f5e5SChris Mason int next_backup; 1885af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1886af31f5e5SChris Mason int last_backup; 1887af31f5e5SChris Mason 1888af31f5e5SChris Mason next_backup = info->backup_root_index; 1889af31f5e5SChris Mason last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) % 1890af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 1891af31f5e5SChris Mason 1892af31f5e5SChris Mason /* 1893af31f5e5SChris Mason * just overwrite the last backup if we're at the same generation 1894af31f5e5SChris Mason * this happens only at umount 1895af31f5e5SChris Mason */ 1896af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + last_backup; 1897af31f5e5SChris Mason if (btrfs_backup_tree_root_gen(root_backup) == 1898af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)) 1899af31f5e5SChris Mason next_backup = last_backup; 1900af31f5e5SChris Mason 1901af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1902af31f5e5SChris Mason 1903af31f5e5SChris Mason /* 1904af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1905af31f5e5SChris Mason * regardless of which ones we use today 1906af31f5e5SChris Mason */ 1907af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1908af31f5e5SChris Mason 1909af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1910af31f5e5SChris Mason 1911af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1912af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1913af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1914af31f5e5SChris Mason 1915af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1916af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1917af31f5e5SChris Mason 1918af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1919af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1920af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 1921af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 1922af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 1923af31f5e5SChris Mason 1924af31f5e5SChris Mason btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); 1925af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 1926af31f5e5SChris Mason btrfs_header_generation(info->extent_root->node)); 1927af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 1928af31f5e5SChris Mason btrfs_header_level(info->extent_root->node)); 1929af31f5e5SChris Mason 19307c7e82a7SChris Mason /* 19317c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 19327c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 19337c7e82a7SChris Mason */ 19347c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 19357c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 19367c7e82a7SChris Mason info->fs_root->node->start); 1937af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 1938af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 1939af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 1940af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 19417c7e82a7SChris Mason } 1942af31f5e5SChris Mason 1943af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 1944af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 1945af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 1946af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 1947af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 1948af31f5e5SChris Mason 1949af31f5e5SChris Mason btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); 1950af31f5e5SChris Mason btrfs_set_backup_csum_root_gen(root_backup, 1951af31f5e5SChris Mason btrfs_header_generation(info->csum_root->node)); 1952af31f5e5SChris Mason btrfs_set_backup_csum_root_level(root_backup, 1953af31f5e5SChris Mason btrfs_header_level(info->csum_root->node)); 1954af31f5e5SChris Mason 1955af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 1956af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 1957af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 1958af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 1959af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 1960af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 1961af31f5e5SChris Mason 1962af31f5e5SChris Mason /* 1963af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 1964af31f5e5SChris Mason * for the next commit 1965af31f5e5SChris Mason */ 1966af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 1967af31f5e5SChris Mason &info->super_for_commit->super_roots, 1968af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 1969af31f5e5SChris Mason } 1970af31f5e5SChris Mason 1971af31f5e5SChris Mason /* 1972af31f5e5SChris Mason * this copies info out of the root backup array and back into 1973af31f5e5SChris Mason * the in-memory super block. It is meant to help iterate through 1974af31f5e5SChris Mason * the array, so you send it the number of backups you've already 1975af31f5e5SChris Mason * tried and the last backup index you used. 1976af31f5e5SChris Mason * 1977af31f5e5SChris Mason * this returns -1 when it has tried all the backups 1978af31f5e5SChris Mason */ 1979af31f5e5SChris Mason static noinline int next_root_backup(struct btrfs_fs_info *info, 1980af31f5e5SChris Mason struct btrfs_super_block *super, 1981af31f5e5SChris Mason int *num_backups_tried, int *backup_index) 1982af31f5e5SChris Mason { 1983af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1984af31f5e5SChris Mason int newest = *backup_index; 1985af31f5e5SChris Mason 1986af31f5e5SChris Mason if (*num_backups_tried == 0) { 1987af31f5e5SChris Mason u64 gen = btrfs_super_generation(super); 1988af31f5e5SChris Mason 1989af31f5e5SChris Mason newest = find_newest_super_backup(info, gen); 1990af31f5e5SChris Mason if (newest == -1) 1991af31f5e5SChris Mason return -1; 1992af31f5e5SChris Mason 1993af31f5e5SChris Mason *backup_index = newest; 1994af31f5e5SChris Mason *num_backups_tried = 1; 1995af31f5e5SChris Mason } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) { 1996af31f5e5SChris Mason /* we've tried all the backups, all done */ 1997af31f5e5SChris Mason return -1; 1998af31f5e5SChris Mason } else { 1999af31f5e5SChris Mason /* jump to the next oldest backup */ 2000af31f5e5SChris Mason newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) % 2001af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 2002af31f5e5SChris Mason *backup_index = newest; 2003af31f5e5SChris Mason *num_backups_tried += 1; 2004af31f5e5SChris Mason } 2005af31f5e5SChris Mason root_backup = super->super_roots + newest; 2006af31f5e5SChris Mason 2007af31f5e5SChris Mason btrfs_set_super_generation(super, 2008af31f5e5SChris Mason btrfs_backup_tree_root_gen(root_backup)); 2009af31f5e5SChris Mason btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 2010af31f5e5SChris Mason btrfs_set_super_root_level(super, 2011af31f5e5SChris Mason btrfs_backup_tree_root_level(root_backup)); 2012af31f5e5SChris Mason btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 2013af31f5e5SChris Mason 2014af31f5e5SChris Mason /* 2015af31f5e5SChris Mason * fixme: the total bytes and num_devices need to match or we should 2016af31f5e5SChris Mason * need a fsck 2017af31f5e5SChris Mason */ 2018af31f5e5SChris Mason btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 2019af31f5e5SChris Mason btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 2020af31f5e5SChris Mason return 0; 2021af31f5e5SChris Mason } 2022af31f5e5SChris Mason 20237abadb64SLiu Bo /* helper to cleanup workers */ 20247abadb64SLiu Bo static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 20257abadb64SLiu Bo { 2026dc6e3209SQu Wenruo btrfs_destroy_workqueue(fs_info->fixup_workers); 2027afe3d242SQu Wenruo btrfs_destroy_workqueue(fs_info->delalloc_workers); 20285cdc7ad3SQu Wenruo btrfs_destroy_workqueue(fs_info->workers); 2029fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_workers); 2030fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_raid56_workers); 20318b110e39SMiao Xie btrfs_destroy_workqueue(fs_info->endio_repair_workers); 2032d05a33acSQu Wenruo btrfs_destroy_workqueue(fs_info->rmw_workers); 2033fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_write_workers); 2034fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 2035a8c93d4eSQu Wenruo btrfs_destroy_workqueue(fs_info->submit_workers); 20365b3bc44eSQu Wenruo btrfs_destroy_workqueue(fs_info->delayed_workers); 2037e66f0bb1SQu Wenruo btrfs_destroy_workqueue(fs_info->caching_workers); 2038736cfa15SQu Wenruo btrfs_destroy_workqueue(fs_info->readahead_workers); 2039a44903abSQu Wenruo btrfs_destroy_workqueue(fs_info->flush_workers); 2040fc97fab0SQu Wenruo btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2041a79b7d4bSChris Mason btrfs_destroy_workqueue(fs_info->extent_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 */ 2062af31f5e5SChris Mason static void free_root_pointers(struct btrfs_fs_info *info, int 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); 20712e9f5954SRashika if (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 /* 22022a458198SEric Sandeen * a higher idle thresh on the submit workers makes it much more 22032a458198SEric Sandeen * likely that bios will be send down in a sane order to the 22042a458198SEric Sandeen * devices 22052a458198SEric Sandeen */ 22062a458198SEric Sandeen fs_info->submit_workers = 2207cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "submit", flags, 22082a458198SEric Sandeen min_t(u64, fs_devices->num_devices, 22092a458198SEric Sandeen max_active), 64); 22102a458198SEric Sandeen 22112a458198SEric Sandeen fs_info->fixup_workers = 2212cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); 22132a458198SEric Sandeen 22142a458198SEric Sandeen /* 22152a458198SEric Sandeen * endios are largely parallel and should have a very 22162a458198SEric Sandeen * low idle thresh 22172a458198SEric Sandeen */ 22182a458198SEric Sandeen fs_info->endio_workers = 2219cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4); 22202a458198SEric Sandeen fs_info->endio_meta_workers = 2221cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta", flags, 2222cb001095SJeff Mahoney max_active, 4); 22232a458198SEric Sandeen fs_info->endio_meta_write_workers = 2224cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags, 2225cb001095SJeff Mahoney max_active, 2); 22262a458198SEric Sandeen fs_info->endio_raid56_workers = 2227cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-raid56", flags, 2228cb001095SJeff Mahoney max_active, 4); 22292a458198SEric Sandeen fs_info->endio_repair_workers = 2230cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0); 22312a458198SEric Sandeen fs_info->rmw_workers = 2232cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2); 22332a458198SEric Sandeen fs_info->endio_write_workers = 2234cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-write", flags, 2235cb001095SJeff Mahoney max_active, 2); 22362a458198SEric Sandeen fs_info->endio_freespace_worker = 2237cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "freespace-write", flags, 2238cb001095SJeff Mahoney max_active, 0); 22392a458198SEric Sandeen fs_info->delayed_workers = 2240cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, 2241cb001095SJeff Mahoney max_active, 0); 22422a458198SEric Sandeen fs_info->readahead_workers = 2243cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "readahead", flags, 2244cb001095SJeff Mahoney max_active, 2); 22452a458198SEric Sandeen fs_info->qgroup_rescan_workers = 2246cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); 22472a458198SEric Sandeen fs_info->extent_workers = 2248cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "extent-refs", flags, 22492a458198SEric Sandeen min_t(u64, fs_devices->num_devices, 22502a458198SEric Sandeen max_active), 8); 22512a458198SEric Sandeen 22522a458198SEric Sandeen if (!(fs_info->workers && fs_info->delalloc_workers && 22532a458198SEric Sandeen fs_info->submit_workers && fs_info->flush_workers && 22542a458198SEric Sandeen fs_info->endio_workers && fs_info->endio_meta_workers && 22552a458198SEric Sandeen fs_info->endio_meta_write_workers && 22562a458198SEric Sandeen fs_info->endio_repair_workers && 22572a458198SEric Sandeen fs_info->endio_write_workers && fs_info->endio_raid56_workers && 22582a458198SEric Sandeen fs_info->endio_freespace_worker && fs_info->rmw_workers && 22592a458198SEric Sandeen fs_info->caching_workers && fs_info->readahead_workers && 22602a458198SEric Sandeen fs_info->fixup_workers && fs_info->delayed_workers && 22612a458198SEric Sandeen fs_info->extent_workers && 22622a458198SEric Sandeen fs_info->qgroup_rescan_workers)) { 22632a458198SEric Sandeen return -ENOMEM; 22642a458198SEric Sandeen } 22652a458198SEric Sandeen 22662a458198SEric Sandeen return 0; 22672a458198SEric Sandeen } 22682a458198SEric Sandeen 226963443bf5SEric Sandeen static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 227063443bf5SEric Sandeen struct btrfs_fs_devices *fs_devices) 227163443bf5SEric Sandeen { 227263443bf5SEric Sandeen int ret; 227363443bf5SEric Sandeen struct btrfs_root *log_tree_root; 227463443bf5SEric Sandeen struct btrfs_super_block *disk_super = fs_info->super_copy; 227563443bf5SEric Sandeen u64 bytenr = btrfs_super_log_root(disk_super); 2276581c1760SQu Wenruo int level = btrfs_super_log_root_level(disk_super); 227763443bf5SEric Sandeen 227863443bf5SEric Sandeen if (fs_devices->rw_devices == 0) { 2279f14d104dSDavid Sterba btrfs_warn(fs_info, "log replay required on RO media"); 228063443bf5SEric Sandeen return -EIO; 228163443bf5SEric Sandeen } 228263443bf5SEric Sandeen 228374e4d827SDavid Sterba log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL); 228463443bf5SEric Sandeen if (!log_tree_root) 228563443bf5SEric Sandeen return -ENOMEM; 228663443bf5SEric Sandeen 2287da17066cSJeff Mahoney __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID); 228863443bf5SEric Sandeen 22892ff7e61eSJeff Mahoney log_tree_root->node = read_tree_block(fs_info, bytenr, 2290581c1760SQu Wenruo fs_info->generation + 1, 2291581c1760SQu Wenruo level, NULL); 229264c043deSLiu Bo if (IS_ERR(log_tree_root->node)) { 2293f14d104dSDavid Sterba btrfs_warn(fs_info, "failed to read log tree"); 22940eeff236SLiu Bo ret = PTR_ERR(log_tree_root->node); 229564c043deSLiu Bo kfree(log_tree_root); 22960eeff236SLiu Bo return ret; 229764c043deSLiu Bo } else if (!extent_buffer_uptodate(log_tree_root->node)) { 2298f14d104dSDavid Sterba btrfs_err(fs_info, "failed to read log tree"); 229963443bf5SEric Sandeen free_extent_buffer(log_tree_root->node); 230063443bf5SEric Sandeen kfree(log_tree_root); 230163443bf5SEric Sandeen return -EIO; 230263443bf5SEric Sandeen } 230363443bf5SEric Sandeen /* returns with log_tree_root freed on success */ 230463443bf5SEric Sandeen ret = btrfs_recover_log_trees(log_tree_root); 230563443bf5SEric Sandeen if (ret) { 23060b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 230763443bf5SEric Sandeen "Failed to recover log tree"); 230863443bf5SEric Sandeen free_extent_buffer(log_tree_root->node); 230963443bf5SEric Sandeen kfree(log_tree_root); 231063443bf5SEric Sandeen return ret; 231163443bf5SEric Sandeen } 231263443bf5SEric Sandeen 2313bc98a42cSDavid Howells if (sb_rdonly(fs_info->sb)) { 23146bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 231563443bf5SEric Sandeen if (ret) 231663443bf5SEric Sandeen return ret; 231763443bf5SEric Sandeen } 231863443bf5SEric Sandeen 231963443bf5SEric Sandeen return 0; 232063443bf5SEric Sandeen } 232163443bf5SEric Sandeen 23226bccf3abSJeff Mahoney static int btrfs_read_roots(struct btrfs_fs_info *fs_info) 23234bbcaa64SEric Sandeen { 23246bccf3abSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 2325a4f3d2c4SDavid Sterba struct btrfs_root *root; 23264bbcaa64SEric Sandeen struct btrfs_key location; 23274bbcaa64SEric Sandeen int ret; 23284bbcaa64SEric Sandeen 23296bccf3abSJeff Mahoney BUG_ON(!fs_info->tree_root); 23306bccf3abSJeff Mahoney 23314bbcaa64SEric Sandeen location.objectid = BTRFS_EXTENT_TREE_OBJECTID; 23324bbcaa64SEric Sandeen location.type = BTRFS_ROOT_ITEM_KEY; 23334bbcaa64SEric Sandeen location.offset = 0; 23344bbcaa64SEric Sandeen 2335a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2336f50f4353SLiu Bo if (IS_ERR(root)) { 2337f50f4353SLiu Bo ret = PTR_ERR(root); 2338f50f4353SLiu Bo goto out; 2339f50f4353SLiu Bo } 2340a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2341a4f3d2c4SDavid Sterba fs_info->extent_root = root; 23424bbcaa64SEric Sandeen 23434bbcaa64SEric Sandeen location.objectid = BTRFS_DEV_TREE_OBJECTID; 2344a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2345f50f4353SLiu Bo if (IS_ERR(root)) { 2346f50f4353SLiu Bo ret = PTR_ERR(root); 2347f50f4353SLiu Bo goto out; 2348f50f4353SLiu Bo } 2349a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2350a4f3d2c4SDavid Sterba fs_info->dev_root = root; 23514bbcaa64SEric Sandeen btrfs_init_devices_late(fs_info); 23524bbcaa64SEric Sandeen 23534bbcaa64SEric Sandeen location.objectid = BTRFS_CSUM_TREE_OBJECTID; 2354a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2355f50f4353SLiu Bo if (IS_ERR(root)) { 2356f50f4353SLiu Bo ret = PTR_ERR(root); 2357f50f4353SLiu Bo goto out; 2358f50f4353SLiu Bo } 2359a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2360a4f3d2c4SDavid Sterba fs_info->csum_root = root; 23614bbcaa64SEric Sandeen 23624bbcaa64SEric Sandeen location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2363a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2364a4f3d2c4SDavid Sterba if (!IS_ERR(root)) { 2365a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2366afcdd129SJosef Bacik set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); 2367a4f3d2c4SDavid Sterba fs_info->quota_root = root; 23684bbcaa64SEric Sandeen } 23694bbcaa64SEric Sandeen 23704bbcaa64SEric Sandeen location.objectid = BTRFS_UUID_TREE_OBJECTID; 2371a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2372a4f3d2c4SDavid Sterba if (IS_ERR(root)) { 2373a4f3d2c4SDavid Sterba ret = PTR_ERR(root); 23744bbcaa64SEric Sandeen if (ret != -ENOENT) 2375f50f4353SLiu Bo goto out; 23764bbcaa64SEric Sandeen } else { 2377a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2378a4f3d2c4SDavid Sterba fs_info->uuid_root = root; 23794bbcaa64SEric Sandeen } 23804bbcaa64SEric Sandeen 238170f6d82eSOmar Sandoval if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 238270f6d82eSOmar Sandoval location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID; 238370f6d82eSOmar Sandoval root = btrfs_read_tree_root(tree_root, &location); 2384f50f4353SLiu Bo if (IS_ERR(root)) { 2385f50f4353SLiu Bo ret = PTR_ERR(root); 2386f50f4353SLiu Bo goto out; 2387f50f4353SLiu Bo } 238870f6d82eSOmar Sandoval set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 238970f6d82eSOmar Sandoval fs_info->free_space_root = root; 239070f6d82eSOmar Sandoval } 239170f6d82eSOmar Sandoval 23924bbcaa64SEric Sandeen return 0; 2393f50f4353SLiu Bo out: 2394f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", 2395f50f4353SLiu Bo location.objectid, ret); 2396f50f4353SLiu Bo return ret; 23974bbcaa64SEric Sandeen } 23984bbcaa64SEric Sandeen 2399069ec957SQu Wenruo /* 2400069ec957SQu Wenruo * Real super block validation 2401069ec957SQu Wenruo * NOTE: super csum type and incompat features will not be checked here. 2402069ec957SQu Wenruo * 2403069ec957SQu Wenruo * @sb: super block to check 2404069ec957SQu Wenruo * @mirror_num: the super block number to check its bytenr: 2405069ec957SQu Wenruo * 0 the primary (1st) sb 2406069ec957SQu Wenruo * 1, 2 2nd and 3rd backup copy 2407069ec957SQu Wenruo * -1 skip bytenr check 2408069ec957SQu Wenruo */ 2409069ec957SQu Wenruo static int validate_super(struct btrfs_fs_info *fs_info, 2410069ec957SQu Wenruo struct btrfs_super_block *sb, int mirror_num) 241121a852b0SQu Wenruo { 241221a852b0SQu Wenruo u64 nodesize = btrfs_super_nodesize(sb); 241321a852b0SQu Wenruo u64 sectorsize = btrfs_super_sectorsize(sb); 241421a852b0SQu Wenruo int ret = 0; 241521a852b0SQu Wenruo 241621a852b0SQu Wenruo if (btrfs_super_magic(sb) != BTRFS_MAGIC) { 241721a852b0SQu Wenruo btrfs_err(fs_info, "no valid FS found"); 241821a852b0SQu Wenruo ret = -EINVAL; 241921a852b0SQu Wenruo } 242021a852b0SQu Wenruo if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { 242121a852b0SQu Wenruo btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", 242221a852b0SQu Wenruo btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); 242321a852b0SQu Wenruo ret = -EINVAL; 242421a852b0SQu Wenruo } 242521a852b0SQu Wenruo if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 242621a852b0SQu Wenruo btrfs_err(fs_info, "tree_root level too big: %d >= %d", 242721a852b0SQu Wenruo btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 242821a852b0SQu Wenruo ret = -EINVAL; 242921a852b0SQu Wenruo } 243021a852b0SQu Wenruo if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 243121a852b0SQu Wenruo btrfs_err(fs_info, "chunk_root level too big: %d >= %d", 243221a852b0SQu Wenruo btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 243321a852b0SQu Wenruo ret = -EINVAL; 243421a852b0SQu Wenruo } 243521a852b0SQu Wenruo if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 243621a852b0SQu Wenruo btrfs_err(fs_info, "log_root level too big: %d >= %d", 243721a852b0SQu Wenruo btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 243821a852b0SQu Wenruo ret = -EINVAL; 243921a852b0SQu Wenruo } 244021a852b0SQu Wenruo 244121a852b0SQu Wenruo /* 244221a852b0SQu Wenruo * Check sectorsize and nodesize first, other check will need it. 244321a852b0SQu Wenruo * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. 244421a852b0SQu Wenruo */ 244521a852b0SQu Wenruo if (!is_power_of_2(sectorsize) || sectorsize < 4096 || 244621a852b0SQu Wenruo sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { 244721a852b0SQu Wenruo btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); 244821a852b0SQu Wenruo ret = -EINVAL; 244921a852b0SQu Wenruo } 245021a852b0SQu Wenruo /* Only PAGE SIZE is supported yet */ 245121a852b0SQu Wenruo if (sectorsize != PAGE_SIZE) { 245221a852b0SQu Wenruo btrfs_err(fs_info, 245321a852b0SQu Wenruo "sectorsize %llu not supported yet, only support %lu", 245421a852b0SQu Wenruo sectorsize, PAGE_SIZE); 245521a852b0SQu Wenruo ret = -EINVAL; 245621a852b0SQu Wenruo } 245721a852b0SQu Wenruo if (!is_power_of_2(nodesize) || nodesize < sectorsize || 245821a852b0SQu Wenruo nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { 245921a852b0SQu Wenruo btrfs_err(fs_info, "invalid nodesize %llu", nodesize); 246021a852b0SQu Wenruo ret = -EINVAL; 246121a852b0SQu Wenruo } 246221a852b0SQu Wenruo if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { 246321a852b0SQu Wenruo btrfs_err(fs_info, "invalid leafsize %u, should be %llu", 246421a852b0SQu Wenruo le32_to_cpu(sb->__unused_leafsize), nodesize); 246521a852b0SQu Wenruo ret = -EINVAL; 246621a852b0SQu Wenruo } 246721a852b0SQu Wenruo 246821a852b0SQu Wenruo /* Root alignment check */ 246921a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { 247021a852b0SQu Wenruo btrfs_warn(fs_info, "tree_root block unaligned: %llu", 247121a852b0SQu Wenruo btrfs_super_root(sb)); 247221a852b0SQu Wenruo ret = -EINVAL; 247321a852b0SQu Wenruo } 247421a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { 247521a852b0SQu Wenruo btrfs_warn(fs_info, "chunk_root block unaligned: %llu", 247621a852b0SQu Wenruo btrfs_super_chunk_root(sb)); 247721a852b0SQu Wenruo ret = -EINVAL; 247821a852b0SQu Wenruo } 247921a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { 248021a852b0SQu Wenruo btrfs_warn(fs_info, "log_root block unaligned: %llu", 248121a852b0SQu Wenruo btrfs_super_log_root(sb)); 248221a852b0SQu Wenruo ret = -EINVAL; 248321a852b0SQu Wenruo } 248421a852b0SQu Wenruo 2485de37aa51SNikolay Borisov if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, 24867239ff4bSNikolay Borisov BTRFS_FSID_SIZE) != 0) { 248721a852b0SQu Wenruo btrfs_err(fs_info, 24887239ff4bSNikolay Borisov "dev_item UUID does not match metadata fsid: %pU != %pU", 2489de37aa51SNikolay Borisov fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); 249021a852b0SQu Wenruo ret = -EINVAL; 249121a852b0SQu Wenruo } 249221a852b0SQu Wenruo 249321a852b0SQu Wenruo /* 249421a852b0SQu Wenruo * Hint to catch really bogus numbers, bitflips or so, more exact checks are 249521a852b0SQu Wenruo * done later 249621a852b0SQu Wenruo */ 249721a852b0SQu Wenruo if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { 249821a852b0SQu Wenruo btrfs_err(fs_info, "bytes_used is too small %llu", 249921a852b0SQu Wenruo btrfs_super_bytes_used(sb)); 250021a852b0SQu Wenruo ret = -EINVAL; 250121a852b0SQu Wenruo } 250221a852b0SQu Wenruo if (!is_power_of_2(btrfs_super_stripesize(sb))) { 250321a852b0SQu Wenruo btrfs_err(fs_info, "invalid stripesize %u", 250421a852b0SQu Wenruo btrfs_super_stripesize(sb)); 250521a852b0SQu Wenruo ret = -EINVAL; 250621a852b0SQu Wenruo } 250721a852b0SQu Wenruo if (btrfs_super_num_devices(sb) > (1UL << 31)) 250821a852b0SQu Wenruo btrfs_warn(fs_info, "suspicious number of devices: %llu", 250921a852b0SQu Wenruo btrfs_super_num_devices(sb)); 251021a852b0SQu Wenruo if (btrfs_super_num_devices(sb) == 0) { 251121a852b0SQu Wenruo btrfs_err(fs_info, "number of devices is 0"); 251221a852b0SQu Wenruo ret = -EINVAL; 251321a852b0SQu Wenruo } 251421a852b0SQu Wenruo 2515069ec957SQu Wenruo if (mirror_num >= 0 && 2516069ec957SQu Wenruo btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { 251721a852b0SQu Wenruo btrfs_err(fs_info, "super offset mismatch %llu != %u", 251821a852b0SQu Wenruo btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 251921a852b0SQu Wenruo ret = -EINVAL; 252021a852b0SQu Wenruo } 252121a852b0SQu Wenruo 252221a852b0SQu Wenruo /* 252321a852b0SQu Wenruo * Obvious sys_chunk_array corruptions, it must hold at least one key 252421a852b0SQu Wenruo * and one chunk 252521a852b0SQu Wenruo */ 252621a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 252721a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too big %u > %u", 252821a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 252921a852b0SQu Wenruo BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 253021a852b0SQu Wenruo ret = -EINVAL; 253121a852b0SQu Wenruo } 253221a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 253321a852b0SQu Wenruo + sizeof(struct btrfs_chunk)) { 253421a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too small %u < %zu", 253521a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 253621a852b0SQu Wenruo sizeof(struct btrfs_disk_key) 253721a852b0SQu Wenruo + sizeof(struct btrfs_chunk)); 253821a852b0SQu Wenruo ret = -EINVAL; 253921a852b0SQu Wenruo } 254021a852b0SQu Wenruo 254121a852b0SQu Wenruo /* 254221a852b0SQu Wenruo * The generation is a global counter, we'll trust it more than the others 254321a852b0SQu Wenruo * but it's still possible that it's the one that's wrong. 254421a852b0SQu Wenruo */ 254521a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 254621a852b0SQu Wenruo btrfs_warn(fs_info, 254721a852b0SQu Wenruo "suspicious: generation < chunk_root_generation: %llu < %llu", 254821a852b0SQu Wenruo btrfs_super_generation(sb), 254921a852b0SQu Wenruo btrfs_super_chunk_root_generation(sb)); 255021a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 255121a852b0SQu Wenruo && btrfs_super_cache_generation(sb) != (u64)-1) 255221a852b0SQu Wenruo btrfs_warn(fs_info, 255321a852b0SQu Wenruo "suspicious: generation < cache_generation: %llu < %llu", 255421a852b0SQu Wenruo btrfs_super_generation(sb), 255521a852b0SQu Wenruo btrfs_super_cache_generation(sb)); 255621a852b0SQu Wenruo 255721a852b0SQu Wenruo return ret; 255821a852b0SQu Wenruo } 255921a852b0SQu Wenruo 2560069ec957SQu Wenruo /* 2561069ec957SQu Wenruo * Validation of super block at mount time. 2562069ec957SQu Wenruo * Some checks already done early at mount time, like csum type and incompat 2563069ec957SQu Wenruo * flags will be skipped. 2564069ec957SQu Wenruo */ 2565069ec957SQu Wenruo static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) 2566069ec957SQu Wenruo { 2567069ec957SQu Wenruo return validate_super(fs_info, fs_info->super_copy, 0); 2568069ec957SQu Wenruo } 2569069ec957SQu Wenruo 257075cb857dSQu Wenruo /* 257175cb857dSQu Wenruo * Validation of super block at write time. 257275cb857dSQu Wenruo * Some checks like bytenr check will be skipped as their values will be 257375cb857dSQu Wenruo * overwritten soon. 257475cb857dSQu Wenruo * Extra checks like csum type and incompat flags will be done here. 257575cb857dSQu Wenruo */ 257675cb857dSQu Wenruo static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, 257775cb857dSQu Wenruo struct btrfs_super_block *sb) 257875cb857dSQu Wenruo { 257975cb857dSQu Wenruo int ret; 258075cb857dSQu Wenruo 258175cb857dSQu Wenruo ret = validate_super(fs_info, sb, -1); 258275cb857dSQu Wenruo if (ret < 0) 258375cb857dSQu Wenruo goto out; 2584e7e16f48SJohannes Thumshirn if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) { 258575cb857dSQu Wenruo ret = -EUCLEAN; 258675cb857dSQu Wenruo btrfs_err(fs_info, "invalid csum type, has %u want %u", 258775cb857dSQu Wenruo btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); 258875cb857dSQu Wenruo goto out; 258975cb857dSQu Wenruo } 259075cb857dSQu Wenruo if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 259175cb857dSQu Wenruo ret = -EUCLEAN; 259275cb857dSQu Wenruo btrfs_err(fs_info, 259375cb857dSQu Wenruo "invalid incompat flags, has 0x%llx valid mask 0x%llx", 259475cb857dSQu Wenruo btrfs_super_incompat_flags(sb), 259575cb857dSQu Wenruo (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); 259675cb857dSQu Wenruo goto out; 259775cb857dSQu Wenruo } 259875cb857dSQu Wenruo out: 259975cb857dSQu Wenruo if (ret < 0) 260075cb857dSQu Wenruo btrfs_err(fs_info, 260175cb857dSQu Wenruo "super block corruption detected before writing it to disk"); 260275cb857dSQu Wenruo return ret; 260375cb857dSQu Wenruo } 260475cb857dSQu Wenruo 2605ad2b2c80SAl Viro int open_ctree(struct super_block *sb, 2606dfe25020SChris Mason struct btrfs_fs_devices *fs_devices, 2607dfe25020SChris Mason char *options) 2608eb60ceacSChris Mason { 2609db94535dSChris Mason u32 sectorsize; 2610db94535dSChris Mason u32 nodesize; 261187ee04ebSChris Mason u32 stripesize; 261284234f3aSYan Zheng u64 generation; 2613f2b636e8SJosef Bacik u64 features; 26143de4586cSChris Mason struct btrfs_key location; 2615a061fc8dSChris Mason struct buffer_head *bh; 26164d34b278SIlya Dryomov struct btrfs_super_block *disk_super; 2617815745cfSAl Viro struct btrfs_fs_info *fs_info = btrfs_sb(sb); 2618f84a8bd6SAl Viro struct btrfs_root *tree_root; 26194d34b278SIlya Dryomov struct btrfs_root *chunk_root; 2620eb60ceacSChris Mason int ret; 2621e58ca020SYan int err = -EINVAL; 2622af31f5e5SChris Mason int num_backups_tried = 0; 2623af31f5e5SChris Mason int backup_index = 0; 26246675df31SOmar Sandoval int clear_free_space_tree = 0; 2625581c1760SQu Wenruo int level; 26264543df7eSChris Mason 262774e4d827SDavid Sterba tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL); 262874e4d827SDavid Sterba chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL); 2629cb517eabSMiao Xie if (!tree_root || !chunk_root) { 263039279cc3SChris Mason err = -ENOMEM; 263139279cc3SChris Mason goto fail; 263239279cc3SChris Mason } 263376dda93cSYan, Zheng 263476dda93cSYan, Zheng ret = init_srcu_struct(&fs_info->subvol_srcu); 263576dda93cSYan, Zheng if (ret) { 263676dda93cSYan, Zheng err = ret; 263776dda93cSYan, Zheng goto fail; 263876dda93cSYan, Zheng } 263976dda93cSYan, Zheng 26404297ff84SJosef Bacik ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL); 2641e2d84521SMiao Xie if (ret) { 2642e2d84521SMiao Xie err = ret; 26439e11ceeeSJan Kara goto fail_srcu; 2644e2d84521SMiao Xie } 26454297ff84SJosef Bacik 26464297ff84SJosef Bacik ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 26474297ff84SJosef Bacik if (ret) { 26484297ff84SJosef Bacik err = ret; 26494297ff84SJosef Bacik goto fail_dio_bytes; 26504297ff84SJosef Bacik } 265109cbfeafSKirill A. Shutemov fs_info->dirty_metadata_batch = PAGE_SIZE * 2652e2d84521SMiao Xie (1 + ilog2(nr_cpu_ids)); 2653e2d84521SMiao Xie 2654908c7f19STejun Heo ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 2655963d678bSMiao Xie if (ret) { 2656963d678bSMiao Xie err = ret; 2657963d678bSMiao Xie goto fail_dirty_metadata_bytes; 2658963d678bSMiao Xie } 2659963d678bSMiao Xie 26607f8d236aSDavid Sterba ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, 26617f8d236aSDavid Sterba GFP_KERNEL); 2662c404e0dcSMiao Xie if (ret) { 2663c404e0dcSMiao Xie err = ret; 2664c404e0dcSMiao Xie goto fail_delalloc_bytes; 2665c404e0dcSMiao Xie } 2666c404e0dcSMiao Xie 266776dda93cSYan, Zheng INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 2668f28491e0SJosef Bacik INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 26698fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 2670facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 267124bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 2672eb73c1b7SMiao Xie INIT_LIST_HEAD(&fs_info->delalloc_roots); 267311833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 267475cb379dSJeff Mahoney INIT_LIST_HEAD(&fs_info->pending_raid_kobjs); 267575cb379dSJeff Mahoney spin_lock_init(&fs_info->pending_raid_kobjs_lock); 2676eb73c1b7SMiao Xie spin_lock_init(&fs_info->delalloc_root_lock); 2677a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 267876dda93cSYan, Zheng spin_lock_init(&fs_info->fs_roots_radix_lock); 267924bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 26804cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 2681f29021b2SJan Schmidt spin_lock_init(&fs_info->tree_mod_seq_lock); 2682ceda0864SMiao Xie spin_lock_init(&fs_info->super_lock); 2683f28491e0SJosef Bacik spin_lock_init(&fs_info->buffer_lock); 268447ab2a6cSJosef Bacik spin_lock_init(&fs_info->unused_bgs_lock); 2685f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 2686d7c15171SZhao Lei mutex_init(&fs_info->unused_bg_unpin_mutex); 268767c5e7d4SFilipe Manana mutex_init(&fs_info->delete_unused_bgs_mutex); 26887585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 2689573bfb72SMiao Xie mutex_init(&fs_info->delalloc_root_mutex); 2690de98ced9SMiao Xie seqlock_init(&fs_info->profiles_lock); 269119c00ddcSChris Mason 26920b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 26936324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 2694f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 269547ab2a6cSJosef Bacik INIT_LIST_HEAD(&fs_info->unused_bgs); 2696c8bf1b67SDavid Sterba extent_map_tree_init(&fs_info->mapping_tree); 269766d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 269866d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 269966d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 270066d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 270166d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 270266d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 270366d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 2704ba2c4d4eSJosef Bacik btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, 2705ba2c4d4eSJosef Bacik BTRFS_BLOCK_RSV_DELREFS); 2706ba2c4d4eSJosef Bacik 2707771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 27084cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 27092fefd558SZhao Lei atomic_set(&fs_info->reada_works_cnt, 0); 2710034f784dSJosef Bacik atomic_set(&fs_info->nr_delayed_iputs, 0); 2711fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 2712e20d96d6SChris Mason fs_info->sb = sb; 271395ac567aSFilipe David Borba Manana fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 27149ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 27154cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 2716a5ed45f8SNikolay Borisov atomic64_set(&fs_info->free_chunk_space, 0); 2717f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 27188b87dc17SDavid Sterba fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 2719f8c269d7SDavid Sterba fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 272090519d66SArne Jansen /* readahead state */ 2721d0164adcSMel Gorman INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); 272290519d66SArne Jansen spin_lock_init(&fs_info->reada_lock); 2723fd708b81SJosef Bacik btrfs_init_ref_verify(fs_info); 2724c8b97818SChris Mason 2725b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 2726b34b086cSChris Mason num_online_cpus() + 2, 8); 27270afbaf8cSChris Mason 2728199c2a9cSMiao Xie INIT_LIST_HEAD(&fs_info->ordered_roots); 2729199c2a9cSMiao Xie spin_lock_init(&fs_info->ordered_root_lock); 273069fe2d75SJosef Bacik 273169fe2d75SJosef Bacik fs_info->btree_inode = new_inode(sb); 273269fe2d75SJosef Bacik if (!fs_info->btree_inode) { 273369fe2d75SJosef Bacik err = -ENOMEM; 273469fe2d75SJosef Bacik goto fail_bio_counter; 273569fe2d75SJosef Bacik } 273669fe2d75SJosef Bacik mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 273769fe2d75SJosef Bacik 273816cdcec7SMiao Xie fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 273974e4d827SDavid Sterba GFP_KERNEL); 274016cdcec7SMiao Xie if (!fs_info->delayed_root) { 274116cdcec7SMiao Xie err = -ENOMEM; 274216cdcec7SMiao Xie goto fail_iput; 274316cdcec7SMiao Xie } 274416cdcec7SMiao Xie btrfs_init_delayed_root(fs_info->delayed_root); 27453eaa2885SChris Mason 2746638aa7edSEric Sandeen btrfs_init_scrub(fs_info); 274721adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 274821adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 274921adbd5cSStefan Behrens #endif 2750779a65a4SEric Sandeen btrfs_init_balance(fs_info); 275121c7e756SMiao Xie btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work); 2752a2de733cSArne Jansen 27539f6d2510SDavid Sterba sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; 27549f6d2510SDavid Sterba sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); 2755a061fc8dSChris Mason 27566bccf3abSJeff Mahoney btrfs_init_btree_inode(fs_info); 275739279cc3SChris Mason 2758e02119d5SChris Mason spin_lock_init(&fs_info->block_group_cache_lock); 27596bef4d31SEric Paris fs_info->block_group_cache_tree = RB_ROOT; 2760a1897fddSLiu Bo fs_info->first_logical_byte = (u64)-1; 2761925baeddSChris Mason 276243eb5f29SQu Wenruo extent_io_tree_init(fs_info, &fs_info->freed_extents[0], 276343eb5f29SQu Wenruo IO_TREE_FS_INFO_FREED_EXTENTS0, NULL); 276443eb5f29SQu Wenruo extent_io_tree_init(fs_info, &fs_info->freed_extents[1], 276543eb5f29SQu Wenruo IO_TREE_FS_INFO_FREED_EXTENTS1, NULL); 276611833d66SYan Zheng fs_info->pinned_extents = &fs_info->freed_extents[0]; 2767afcdd129SJosef Bacik set_bit(BTRFS_FS_BARRIER, &fs_info->flags); 2768509659cdSChris Mason 27695a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 277079154b1bSChris Mason mutex_init(&fs_info->tree_log_mutex); 2771925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 2772a74a4b97SChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 2773a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 27741bbc621eSChris Mason mutex_init(&fs_info->ro_block_group_mutex); 27759e351cc8SJosef Bacik init_rwsem(&fs_info->commit_root_sem); 2776c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 277776dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 2778803b2f54SStefan Behrens sema_init(&fs_info->uuid_tree_rescan_sem, 1); 2779fa9c0d79SChris Mason 2780ad618368SEric Sandeen btrfs_init_dev_replace_locks(fs_info); 2781f9e92e40SEric Sandeen btrfs_init_qgroup(fs_info); 2782416ac51dSArne Jansen 2783fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2784fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2785fa9c0d79SChris Mason 2786e6dcd2dcSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 2787f9295749SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 2788bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 27894854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 2790034f784dSJosef Bacik init_waitqueue_head(&fs_info->delayed_iputs_wait); 27913768f368SChris Mason 2792da17066cSJeff Mahoney /* Usable values until the real ones are cached from the superblock */ 2793da17066cSJeff Mahoney fs_info->nodesize = 4096; 2794da17066cSJeff Mahoney fs_info->sectorsize = 4096; 2795da17066cSJeff Mahoney fs_info->stripesize = 4096; 2796da17066cSJeff Mahoney 2797eede2bf3SOmar Sandoval spin_lock_init(&fs_info->swapfile_pins_lock); 2798eede2bf3SOmar Sandoval fs_info->swapfile_pins = RB_ROOT; 2799eede2bf3SOmar Sandoval 280053b381b3SDavid Woodhouse ret = btrfs_alloc_stripe_hash_table(fs_info); 280153b381b3SDavid Woodhouse if (ret) { 280283c8266aSDavid Sterba err = ret; 280353b381b3SDavid Woodhouse goto fail_alloc; 280453b381b3SDavid Woodhouse } 280553b381b3SDavid Woodhouse 2806da17066cSJeff Mahoney __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID); 28077eccb903SChris Mason 28083c4bb26bSChris Mason invalidate_bdev(fs_devices->latest_bdev); 28091104a885SDavid Sterba 28101104a885SDavid Sterba /* 28111104a885SDavid Sterba * Read super block and check the signature bytes only 28121104a885SDavid Sterba */ 2813a512bbf8SYan Zheng bh = btrfs_read_dev_super(fs_devices->latest_bdev); 281492fc03fbSAnand Jain if (IS_ERR(bh)) { 281592fc03fbSAnand Jain err = PTR_ERR(bh); 281616cdcec7SMiao Xie goto fail_alloc; 281720b45077SDave Young } 281839279cc3SChris Mason 28191104a885SDavid Sterba /* 28208dc3f22cSJohannes Thumshirn * Verify the type first, if that or the the checksum value are 28218dc3f22cSJohannes Thumshirn * corrupted, we'll find out 28228dc3f22cSJohannes Thumshirn */ 28238dc3f22cSJohannes Thumshirn if (!btrfs_supported_super_csum(btrfs_super_csum_type( 28248dc3f22cSJohannes Thumshirn (struct btrfs_super_block *) bh->b_data))) { 28258dc3f22cSJohannes Thumshirn btrfs_err(fs_info, "unsupported checksum algorithm: %u", 28268dc3f22cSJohannes Thumshirn btrfs_super_csum_type((struct btrfs_super_block *) 28278dc3f22cSJohannes Thumshirn bh->b_data)); 28288dc3f22cSJohannes Thumshirn err = -EINVAL; 28298dc3f22cSJohannes Thumshirn brelse(bh); 28308dc3f22cSJohannes Thumshirn goto fail_alloc; 28318dc3f22cSJohannes Thumshirn } 28328dc3f22cSJohannes Thumshirn 28338dc3f22cSJohannes Thumshirn /* 28341104a885SDavid Sterba * We want to check superblock checksum, the type is stored inside. 28351104a885SDavid Sterba * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 28361104a885SDavid Sterba */ 2837ab8d0fc4SJeff Mahoney if (btrfs_check_super_csum(fs_info, bh->b_data)) { 283805135f59SDavid Sterba btrfs_err(fs_info, "superblock checksum mismatch"); 28391104a885SDavid Sterba err = -EINVAL; 2840b2acdddfSAnand Jain brelse(bh); 28411104a885SDavid Sterba goto fail_alloc; 28421104a885SDavid Sterba } 28431104a885SDavid Sterba 28441104a885SDavid Sterba /* 28451104a885SDavid Sterba * super_copy is zeroed at allocation time and we never touch the 28461104a885SDavid Sterba * following bytes up to INFO_SIZE, the checksum is calculated from 28471104a885SDavid Sterba * the whole block of INFO_SIZE 28481104a885SDavid Sterba */ 28496c41761fSDavid Sterba memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy)); 2850a061fc8dSChris Mason brelse(bh); 28515f39d397SChris Mason 2852fbc6feaeSNikolay Borisov disk_super = fs_info->super_copy; 2853fbc6feaeSNikolay Borisov 2854de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, 2855de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 2856de37aa51SNikolay Borisov 28577239ff4bSNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID)) { 2858de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid, 2859de37aa51SNikolay Borisov fs_info->super_copy->metadata_uuid, 2860de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 28617239ff4bSNikolay Borisov } 28620b86a832SChris Mason 2863fbc6feaeSNikolay Borisov features = btrfs_super_flags(disk_super); 2864fbc6feaeSNikolay Borisov if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { 2865fbc6feaeSNikolay Borisov features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; 2866fbc6feaeSNikolay Borisov btrfs_set_super_flags(disk_super, features); 2867fbc6feaeSNikolay Borisov btrfs_info(fs_info, 2868fbc6feaeSNikolay Borisov "found metadata UUID change in progress flag, clearing"); 2869fbc6feaeSNikolay Borisov } 2870fbc6feaeSNikolay Borisov 2871fbc6feaeSNikolay Borisov memcpy(fs_info->super_for_commit, fs_info->super_copy, 2872fbc6feaeSNikolay Borisov sizeof(*fs_info->super_for_commit)); 2873de37aa51SNikolay Borisov 2874069ec957SQu Wenruo ret = btrfs_validate_mount_super(fs_info); 28751104a885SDavid Sterba if (ret) { 287605135f59SDavid Sterba btrfs_err(fs_info, "superblock contains fatal errors"); 28771104a885SDavid Sterba err = -EINVAL; 28781104a885SDavid Sterba goto fail_alloc; 28791104a885SDavid Sterba } 28801104a885SDavid Sterba 28810f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 288216cdcec7SMiao Xie goto fail_alloc; 28830f7d52f4SChris Mason 2884acce952bSliubo /* check FS state, whether FS is broken. */ 288587533c47SMiao Xie if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 288687533c47SMiao Xie set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 2887acce952bSliubo 288875e7cb7fSLiu Bo /* 2889af31f5e5SChris Mason * run through our array of backup supers and setup 2890af31f5e5SChris Mason * our ring pointer to the oldest one 2891af31f5e5SChris Mason */ 2892af31f5e5SChris Mason generation = btrfs_super_generation(disk_super); 2893af31f5e5SChris Mason find_oldest_super_backup(fs_info, generation); 2894af31f5e5SChris Mason 2895af31f5e5SChris Mason /* 289675e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 289775e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 289875e7cb7fSLiu Bo */ 289975e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 290075e7cb7fSLiu Bo 29012ff7e61eSJeff Mahoney ret = btrfs_parse_options(fs_info, options, sb->s_flags); 29022b82032cSYan Zheng if (ret) { 29032b82032cSYan Zheng err = ret; 290416cdcec7SMiao Xie goto fail_alloc; 29052b82032cSYan Zheng } 2906dfe25020SChris Mason 2907f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 2908f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 2909f2b636e8SJosef Bacik if (features) { 291005135f59SDavid Sterba btrfs_err(fs_info, 291105135f59SDavid Sterba "cannot mount because of unsupported optional features (%llx)", 2912c1c9ff7cSGeert Uytterhoeven features); 2913f2b636e8SJosef Bacik err = -EINVAL; 291416cdcec7SMiao Xie goto fail_alloc; 2915f2b636e8SJosef Bacik } 2916f2b636e8SJosef Bacik 29175d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 29185d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 29190b246afaSJeff Mahoney if (fs_info->compress_type == BTRFS_COMPRESS_LZO) 2920a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 29215c1aab1dSNick Terrell else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) 29225c1aab1dSNick Terrell features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; 2923727011e0SChris Mason 29243173a18fSJosef Bacik if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) 292505135f59SDavid Sterba btrfs_info(fs_info, "has skinny extents"); 29263173a18fSJosef Bacik 2927727011e0SChris Mason /* 2928727011e0SChris Mason * flag our filesystem as having big metadata blocks if 2929727011e0SChris Mason * they are bigger than the page size 2930727011e0SChris Mason */ 293109cbfeafSKirill A. Shutemov if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) { 2932727011e0SChris Mason if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 293305135f59SDavid Sterba btrfs_info(fs_info, 293405135f59SDavid Sterba "flagging fs with big metadata feature"); 2935727011e0SChris Mason features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 2936727011e0SChris Mason } 2937727011e0SChris Mason 2938bc3f116fSChris Mason nodesize = btrfs_super_nodesize(disk_super); 2939bc3f116fSChris Mason sectorsize = btrfs_super_sectorsize(disk_super); 2940b7f67055SChandan Rajendra stripesize = sectorsize; 2941707e8a07SDavid Sterba fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 2942963d678bSMiao Xie fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 2943bc3f116fSChris Mason 2944da17066cSJeff Mahoney /* Cache block sizes */ 2945da17066cSJeff Mahoney fs_info->nodesize = nodesize; 2946da17066cSJeff Mahoney fs_info->sectorsize = sectorsize; 2947da17066cSJeff Mahoney fs_info->stripesize = stripesize; 2948da17066cSJeff Mahoney 2949bc3f116fSChris Mason /* 2950bc3f116fSChris Mason * mixed block groups end up with duplicate but slightly offset 2951bc3f116fSChris Mason * extent buffers for the same range. It leads to corruptions 2952bc3f116fSChris Mason */ 2953bc3f116fSChris Mason if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 2954707e8a07SDavid Sterba (sectorsize != nodesize)) { 295505135f59SDavid Sterba btrfs_err(fs_info, 295605135f59SDavid Sterba "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", 295705135f59SDavid Sterba nodesize, sectorsize); 2958bc3f116fSChris Mason goto fail_alloc; 2959bc3f116fSChris Mason } 2960bc3f116fSChris Mason 2961ceda0864SMiao Xie /* 2962ceda0864SMiao Xie * Needn't use the lock because there is no other task which will 2963ceda0864SMiao Xie * update the flag. 2964ceda0864SMiao Xie */ 29655d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 29665d4f98a2SYan Zheng 2967f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 2968f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 2969bc98a42cSDavid Howells if (!sb_rdonly(sb) && features) { 297005135f59SDavid Sterba btrfs_err(fs_info, 297105135f59SDavid Sterba "cannot mount read-write because of unsupported optional features (%llx)", 2972c1c9ff7cSGeert Uytterhoeven features); 2973f2b636e8SJosef Bacik err = -EINVAL; 297416cdcec7SMiao Xie goto fail_alloc; 2975f2b636e8SJosef Bacik } 297661d92c32SChris Mason 29772a458198SEric Sandeen ret = btrfs_init_workqueues(fs_info, fs_devices); 29782a458198SEric Sandeen if (ret) { 29792a458198SEric Sandeen err = ret; 29800dc3b84aSJosef Bacik goto fail_sb_buffer; 29810dc3b84aSJosef Bacik } 29824543df7eSChris Mason 29839e11ceeeSJan Kara sb->s_bdi->congested_fn = btrfs_congested_fn; 29849e11ceeeSJan Kara sb->s_bdi->congested_data = fs_info; 29859e11ceeeSJan Kara sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK; 2986b5420237SNikolay Borisov sb->s_bdi->ra_pages = VM_READAHEAD_PAGES; 29879e11ceeeSJan Kara sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); 29889e11ceeeSJan Kara sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); 29894575c9ccSChris Mason 2990a061fc8dSChris Mason sb->s_blocksize = sectorsize; 2991a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 2992de37aa51SNikolay Borisov memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); 2993db94535dSChris Mason 2994925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 29956bccf3abSJeff Mahoney ret = btrfs_read_sys_array(fs_info); 2996925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 299784eed90fSChris Mason if (ret) { 299805135f59SDavid Sterba btrfs_err(fs_info, "failed to read the system array: %d", ret); 29995d4f98a2SYan Zheng goto fail_sb_buffer; 300084eed90fSChris Mason } 30010b86a832SChris Mason 300284234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 3003581c1760SQu Wenruo level = btrfs_super_chunk_root_level(disk_super); 30040b86a832SChris Mason 3005da17066cSJeff Mahoney __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID); 30060b86a832SChris Mason 30072ff7e61eSJeff Mahoney chunk_root->node = read_tree_block(fs_info, 30080b86a832SChris Mason btrfs_super_chunk_root(disk_super), 3009581c1760SQu Wenruo generation, level, NULL); 301064c043deSLiu Bo if (IS_ERR(chunk_root->node) || 301164c043deSLiu Bo !extent_buffer_uptodate(chunk_root->node)) { 301205135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk root"); 3013e5fffbacSchandan if (!IS_ERR(chunk_root->node)) 3014e5fffbacSchandan free_extent_buffer(chunk_root->node); 301595ab1f64SZhao Lei chunk_root->node = NULL; 3016af31f5e5SChris Mason goto fail_tree_roots; 301783121942SDavid Woodhouse } 30185d4f98a2SYan Zheng btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 30195d4f98a2SYan Zheng chunk_root->commit_root = btrfs_root_node(chunk_root); 30200b86a832SChris Mason 3021e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 3022b308bc2fSGeert Uytterhoeven btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE); 3023e17cade2SChris Mason 30245b4aacefSJeff Mahoney ret = btrfs_read_chunk_tree(fs_info); 30252b82032cSYan Zheng if (ret) { 302605135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk tree: %d", ret); 3027af31f5e5SChris Mason goto fail_tree_roots; 30282b82032cSYan Zheng } 30290b86a832SChris Mason 30308dabb742SStefan Behrens /* 30319b99b115SAnand Jain * Keep the devid that is marked to be the target device for the 30329b99b115SAnand Jain * device replace procedure 30338dabb742SStefan Behrens */ 30349b99b115SAnand Jain btrfs_free_extra_devids(fs_devices, 0); 3035dfe25020SChris Mason 3036a6b0d5c8SChris Mason if (!fs_devices->latest_bdev) { 303705135f59SDavid Sterba btrfs_err(fs_info, "failed to read devices"); 3038a6b0d5c8SChris Mason goto fail_tree_roots; 3039a6b0d5c8SChris Mason } 3040a6b0d5c8SChris Mason 3041af31f5e5SChris Mason retry_root_backup: 304284234f3aSYan Zheng generation = btrfs_super_generation(disk_super); 3043581c1760SQu Wenruo level = btrfs_super_root_level(disk_super); 30440b86a832SChris Mason 30452ff7e61eSJeff Mahoney tree_root->node = read_tree_block(fs_info, 3046db94535dSChris Mason btrfs_super_root(disk_super), 3047581c1760SQu Wenruo generation, level, NULL); 304864c043deSLiu Bo if (IS_ERR(tree_root->node) || 304964c043deSLiu Bo !extent_buffer_uptodate(tree_root->node)) { 305005135f59SDavid Sterba btrfs_warn(fs_info, "failed to read tree root"); 3051e5fffbacSchandan if (!IS_ERR(tree_root->node)) 3052e5fffbacSchandan free_extent_buffer(tree_root->node); 305395ab1f64SZhao Lei tree_root->node = NULL; 3054af31f5e5SChris Mason goto recovery_tree_root; 305583121942SDavid Woodhouse } 3056af31f5e5SChris Mason 30575d4f98a2SYan Zheng btrfs_set_root_node(&tree_root->root_item, tree_root->node); 30585d4f98a2SYan Zheng tree_root->commit_root = btrfs_root_node(tree_root); 305969e9c6c6SStefan Behrens btrfs_set_root_refs(&tree_root->root_item, 1); 3060db94535dSChris Mason 3061f32e48e9SChandan Rajendra mutex_lock(&tree_root->objectid_mutex); 3062f32e48e9SChandan Rajendra ret = btrfs_find_highest_objectid(tree_root, 3063f32e48e9SChandan Rajendra &tree_root->highest_objectid); 3064f32e48e9SChandan Rajendra if (ret) { 3065f32e48e9SChandan Rajendra mutex_unlock(&tree_root->objectid_mutex); 3066f32e48e9SChandan Rajendra goto recovery_tree_root; 3067f32e48e9SChandan Rajendra } 3068f32e48e9SChandan Rajendra 3069f32e48e9SChandan Rajendra ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 3070f32e48e9SChandan Rajendra 3071f32e48e9SChandan Rajendra mutex_unlock(&tree_root->objectid_mutex); 3072f32e48e9SChandan Rajendra 30736bccf3abSJeff Mahoney ret = btrfs_read_roots(fs_info); 30744bbcaa64SEric Sandeen if (ret) 3075af31f5e5SChris Mason goto recovery_tree_root; 3076f7a81ea4SStefan Behrens 30778929ecfaSYan, Zheng fs_info->generation = generation; 30788929ecfaSYan, Zheng fs_info->last_trans_committed = generation; 30798929ecfaSYan, Zheng 3080cf90d884SQu Wenruo ret = btrfs_verify_dev_extents(fs_info); 3081cf90d884SQu Wenruo if (ret) { 3082cf90d884SQu Wenruo btrfs_err(fs_info, 3083cf90d884SQu Wenruo "failed to verify dev extents against chunks: %d", 3084cf90d884SQu Wenruo ret); 3085cf90d884SQu Wenruo goto fail_block_groups; 3086cf90d884SQu Wenruo } 308768310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 308868310a5eSIlya Dryomov if (ret) { 308905135f59SDavid Sterba btrfs_err(fs_info, "failed to recover balance: %d", ret); 309068310a5eSIlya Dryomov goto fail_block_groups; 309168310a5eSIlya Dryomov } 309268310a5eSIlya Dryomov 3093733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 3094733f4fbbSStefan Behrens if (ret) { 309505135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_stats: %d", ret); 3096733f4fbbSStefan Behrens goto fail_block_groups; 3097733f4fbbSStefan Behrens } 3098733f4fbbSStefan Behrens 30998dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 31008dabb742SStefan Behrens if (ret) { 310105135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_replace: %d", ret); 31028dabb742SStefan Behrens goto fail_block_groups; 31038dabb742SStefan Behrens } 31048dabb742SStefan Behrens 31059b99b115SAnand Jain btrfs_free_extra_devids(fs_devices, 1); 31068dabb742SStefan Behrens 3107b7c35e81SAnand Jain ret = btrfs_sysfs_add_fsid(fs_devices, NULL); 3108b7c35e81SAnand Jain if (ret) { 310905135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", 311005135f59SDavid Sterba ret); 3111b7c35e81SAnand Jain goto fail_block_groups; 3112b7c35e81SAnand Jain } 3113b7c35e81SAnand Jain 3114b7c35e81SAnand Jain ret = btrfs_sysfs_add_device(fs_devices); 3115b7c35e81SAnand Jain if (ret) { 311605135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs device interface: %d", 311705135f59SDavid Sterba ret); 3118b7c35e81SAnand Jain goto fail_fsdev_sysfs; 3119b7c35e81SAnand Jain } 3120b7c35e81SAnand Jain 312196f3136eSAnand Jain ret = btrfs_sysfs_add_mounted(fs_info); 31225ac1d209SJeff Mahoney if (ret) { 312305135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); 3124b7c35e81SAnand Jain goto fail_fsdev_sysfs; 31255ac1d209SJeff Mahoney } 31265ac1d209SJeff Mahoney 3127c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 3128c59021f8Sliubo if (ret) { 312905135f59SDavid Sterba btrfs_err(fs_info, "failed to initialize space info: %d", ret); 31302365dd3cSAnand Jain goto fail_sysfs; 3131c59021f8Sliubo } 3132c59021f8Sliubo 31335b4aacefSJeff Mahoney ret = btrfs_read_block_groups(fs_info); 31341b1d1f66SJosef Bacik if (ret) { 313505135f59SDavid Sterba btrfs_err(fs_info, "failed to read block groups: %d", ret); 31362365dd3cSAnand Jain goto fail_sysfs; 31371b1d1f66SJosef Bacik } 31384330e183SQu Wenruo 31396528b99dSAnand Jain if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) { 314005135f59SDavid Sterba btrfs_warn(fs_info, 314152042d8eSAndrea Gelmini "writable mount is not allowed due to too many missing devices"); 31422365dd3cSAnand Jain goto fail_sysfs; 3143292fd7fcSStefan Behrens } 31449078a3e1SChris Mason 3145a74a4b97SChris Mason fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 3146a74a4b97SChris Mason "btrfs-cleaner"); 314757506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 31482365dd3cSAnand Jain goto fail_sysfs; 3149a74a4b97SChris Mason 3150a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 3151a74a4b97SChris Mason tree_root, 3152a74a4b97SChris Mason "btrfs-transaction"); 315357506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 31543f157a2fSChris Mason goto fail_cleaner; 3155a74a4b97SChris Mason 3156583b7231SHans van Kranenburg if (!btrfs_test_opt(fs_info, NOSSD) && 3157c289811cSChris Mason !fs_info->fs_devices->rotating) { 3158583b7231SHans van Kranenburg btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); 3159c289811cSChris Mason } 3160c289811cSChris Mason 3161572d9ab7SDavid Sterba /* 316201327610SNicholas D Steeves * Mount does not set all options immediately, we can do it now and do 3163572d9ab7SDavid Sterba * not have to wait for transaction commit 3164572d9ab7SDavid Sterba */ 3165572d9ab7SDavid Sterba btrfs_apply_pending_changes(fs_info); 31663818aea2SQu Wenruo 316721adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 31680b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { 31692ff7e61eSJeff Mahoney ret = btrfsic_mount(fs_info, fs_devices, 31700b246afaSJeff Mahoney btrfs_test_opt(fs_info, 317121adbd5cSStefan Behrens CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? 317221adbd5cSStefan Behrens 1 : 0, 317321adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 317421adbd5cSStefan Behrens if (ret) 317505135f59SDavid Sterba btrfs_warn(fs_info, 317605135f59SDavid Sterba "failed to initialize integrity check module: %d", 317705135f59SDavid Sterba ret); 317821adbd5cSStefan Behrens } 317921adbd5cSStefan Behrens #endif 3180bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 3181bcef60f2SArne Jansen if (ret) 3182bcef60f2SArne Jansen goto fail_trans_kthread; 318321adbd5cSStefan Behrens 3184fd708b81SJosef Bacik if (btrfs_build_ref_tree(fs_info)) 3185fd708b81SJosef Bacik btrfs_err(fs_info, "couldn't build ref tree"); 3186fd708b81SJosef Bacik 318796da0919SQu Wenruo /* do not make disk changes in broken FS or nologreplay is given */ 318896da0919SQu Wenruo if (btrfs_super_log_root(disk_super) != 0 && 31890b246afaSJeff Mahoney !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 319063443bf5SEric Sandeen ret = btrfs_replay_log(fs_info, fs_devices); 319179787eaaSJeff Mahoney if (ret) { 319263443bf5SEric Sandeen err = ret; 319328c16cbbSWang Shilong goto fail_qgroup; 3194e556ce2cSYan Zheng } 3195e02119d5SChris Mason } 31961a40e23bSZheng Yan 31976bccf3abSJeff Mahoney ret = btrfs_find_orphan_roots(fs_info); 319879787eaaSJeff Mahoney if (ret) 319928c16cbbSWang Shilong goto fail_qgroup; 320076dda93cSYan, Zheng 3201bc98a42cSDavid Howells if (!sb_rdonly(sb)) { 3202d68fc57bSYan, Zheng ret = btrfs_cleanup_fs_roots(fs_info); 320344c44af2SIlya Dryomov if (ret) 320428c16cbbSWang Shilong goto fail_qgroup; 320590c711abSZygo Blaxell 320690c711abSZygo Blaxell mutex_lock(&fs_info->cleaner_mutex); 32075d4f98a2SYan Zheng ret = btrfs_recover_relocation(tree_root); 320890c711abSZygo Blaxell mutex_unlock(&fs_info->cleaner_mutex); 3209d7ce5843SMiao Xie if (ret < 0) { 321005135f59SDavid Sterba btrfs_warn(fs_info, "failed to recover relocation: %d", 321105135f59SDavid Sterba ret); 3212d7ce5843SMiao Xie err = -EINVAL; 3213bcef60f2SArne Jansen goto fail_qgroup; 3214d7ce5843SMiao Xie } 32157c2ca468SChris Mason } 32161a40e23bSZheng Yan 32173de4586cSChris Mason location.objectid = BTRFS_FS_TREE_OBJECTID; 32183de4586cSChris Mason location.type = BTRFS_ROOT_ITEM_KEY; 3219cb517eabSMiao Xie location.offset = 0; 32203de4586cSChris Mason 32213de4586cSChris Mason fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location); 32223140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 32233140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 3224f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read fs tree: %d", err); 3225bcef60f2SArne Jansen goto fail_qgroup; 32263140c9a3SDan Carpenter } 3227c289811cSChris Mason 3228bc98a42cSDavid Howells if (sb_rdonly(sb)) 32292b6ba629SIlya Dryomov return 0; 32302b6ba629SIlya Dryomov 3231f8d468a1SOmar Sandoval if (btrfs_test_opt(fs_info, CLEAR_CACHE) && 3232f8d468a1SOmar Sandoval btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 32336675df31SOmar Sandoval clear_free_space_tree = 1; 32346675df31SOmar Sandoval } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 32356675df31SOmar Sandoval !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { 32366675df31SOmar Sandoval btrfs_warn(fs_info, "free space tree is invalid"); 32376675df31SOmar Sandoval clear_free_space_tree = 1; 32386675df31SOmar Sandoval } 32396675df31SOmar Sandoval 32406675df31SOmar Sandoval if (clear_free_space_tree) { 3241f8d468a1SOmar Sandoval btrfs_info(fs_info, "clearing free space tree"); 3242f8d468a1SOmar Sandoval ret = btrfs_clear_free_space_tree(fs_info); 3243f8d468a1SOmar Sandoval if (ret) { 3244f8d468a1SOmar Sandoval btrfs_warn(fs_info, 3245f8d468a1SOmar Sandoval "failed to clear free space tree: %d", ret); 32466bccf3abSJeff Mahoney close_ctree(fs_info); 3247f8d468a1SOmar Sandoval return ret; 3248f8d468a1SOmar Sandoval } 3249f8d468a1SOmar Sandoval } 3250f8d468a1SOmar Sandoval 32510b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && 3252511711afSChris Mason !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 325305135f59SDavid Sterba btrfs_info(fs_info, "creating free space tree"); 3254511711afSChris Mason ret = btrfs_create_free_space_tree(fs_info); 3255511711afSChris Mason if (ret) { 325605135f59SDavid Sterba btrfs_warn(fs_info, 325705135f59SDavid Sterba "failed to create free space tree: %d", ret); 32586bccf3abSJeff Mahoney close_ctree(fs_info); 3259511711afSChris Mason return ret; 3260511711afSChris Mason } 3261511711afSChris Mason } 3262511711afSChris Mason 3263e3acc2a6SJosef Bacik down_read(&fs_info->cleanup_work_sem); 32642b6ba629SIlya Dryomov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 32652b6ba629SIlya Dryomov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 32662b6ba629SIlya Dryomov up_read(&fs_info->cleanup_work_sem); 32676bccf3abSJeff Mahoney close_ctree(fs_info); 32682b6ba629SIlya Dryomov return ret; 32692b6ba629SIlya Dryomov } 3270e3acc2a6SJosef Bacik up_read(&fs_info->cleanup_work_sem); 327159641015SIlya Dryomov 32722b6ba629SIlya Dryomov ret = btrfs_resume_balance_async(fs_info); 32732b6ba629SIlya Dryomov if (ret) { 327405135f59SDavid Sterba btrfs_warn(fs_info, "failed to resume balance: %d", ret); 32756bccf3abSJeff Mahoney close_ctree(fs_info); 32762b6ba629SIlya Dryomov return ret; 3277e3acc2a6SJosef Bacik } 3278e3acc2a6SJosef Bacik 32798dabb742SStefan Behrens ret = btrfs_resume_dev_replace_async(fs_info); 32808dabb742SStefan Behrens if (ret) { 328105135f59SDavid Sterba btrfs_warn(fs_info, "failed to resume device replace: %d", ret); 32826bccf3abSJeff Mahoney close_ctree(fs_info); 32838dabb742SStefan Behrens return ret; 32848dabb742SStefan Behrens } 32858dabb742SStefan Behrens 3286b382a324SJan Schmidt btrfs_qgroup_rescan_resume(fs_info); 3287b382a324SJan Schmidt 32884bbcaa64SEric Sandeen if (!fs_info->uuid_root) { 328905135f59SDavid Sterba btrfs_info(fs_info, "creating UUID tree"); 3290f7a81ea4SStefan Behrens ret = btrfs_create_uuid_tree(fs_info); 3291f7a81ea4SStefan Behrens if (ret) { 329205135f59SDavid Sterba btrfs_warn(fs_info, 329305135f59SDavid Sterba "failed to create the UUID tree: %d", ret); 32946bccf3abSJeff Mahoney close_ctree(fs_info); 3295f7a81ea4SStefan Behrens return ret; 3296f7a81ea4SStefan Behrens } 32970b246afaSJeff Mahoney } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || 32984bbcaa64SEric Sandeen fs_info->generation != 32994bbcaa64SEric Sandeen btrfs_super_uuid_tree_generation(disk_super)) { 330005135f59SDavid Sterba btrfs_info(fs_info, "checking UUID tree"); 330170f80175SStefan Behrens ret = btrfs_check_uuid_tree(fs_info); 330270f80175SStefan Behrens if (ret) { 330305135f59SDavid Sterba btrfs_warn(fs_info, 330405135f59SDavid Sterba "failed to check the UUID tree: %d", ret); 33056bccf3abSJeff Mahoney close_ctree(fs_info); 330670f80175SStefan Behrens return ret; 330770f80175SStefan Behrens } 330870f80175SStefan Behrens } else { 3309afcdd129SJosef Bacik set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 3310f7a81ea4SStefan Behrens } 3311afcdd129SJosef Bacik set_bit(BTRFS_FS_OPEN, &fs_info->flags); 331247ab2a6cSJosef Bacik 33138dcddfa0SQu Wenruo /* 33148dcddfa0SQu Wenruo * backuproot only affect mount behavior, and if open_ctree succeeded, 33158dcddfa0SQu Wenruo * no need to keep the flag 33168dcddfa0SQu Wenruo */ 33178dcddfa0SQu Wenruo btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT); 33188dcddfa0SQu Wenruo 3319ad2b2c80SAl Viro return 0; 332039279cc3SChris Mason 3321bcef60f2SArne Jansen fail_qgroup: 3322bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 33237c2ca468SChris Mason fail_trans_kthread: 33247c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 33252ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 3326faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 33273f157a2fSChris Mason fail_cleaner: 3328a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 33297c2ca468SChris Mason 33307c2ca468SChris Mason /* 33317c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 33327c2ca468SChris Mason * kthreads 33337c2ca468SChris Mason */ 33347c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 33357c2ca468SChris Mason 33362365dd3cSAnand Jain fail_sysfs: 33376618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 33382365dd3cSAnand Jain 3339b7c35e81SAnand Jain fail_fsdev_sysfs: 3340b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3341b7c35e81SAnand Jain 33421b1d1f66SJosef Bacik fail_block_groups: 334354067ae9SJosef Bacik btrfs_put_block_group_cache(fs_info); 3344af31f5e5SChris Mason 3345af31f5e5SChris Mason fail_tree_roots: 3346af31f5e5SChris Mason free_root_pointers(fs_info, 1); 33472b8195bbSMiao Xie invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3348af31f5e5SChris Mason 334939279cc3SChris Mason fail_sb_buffer: 33507abadb64SLiu Bo btrfs_stop_all_workers(fs_info); 33515cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 335216cdcec7SMiao Xie fail_alloc: 33534543df7eSChris Mason fail_iput: 3354586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 3355586e46e2SIlya Dryomov 33564543df7eSChris Mason iput(fs_info->btree_inode); 3357c404e0dcSMiao Xie fail_bio_counter: 33587f8d236aSDavid Sterba percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 3359963d678bSMiao Xie fail_delalloc_bytes: 3360963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 3361e2d84521SMiao Xie fail_dirty_metadata_bytes: 3362e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 33634297ff84SJosef Bacik fail_dio_bytes: 33644297ff84SJosef Bacik percpu_counter_destroy(&fs_info->dio_bytes); 336576dda93cSYan, Zheng fail_srcu: 336676dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 33677e662854SQinghuang Feng fail: 336853b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 3369586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 3370ad2b2c80SAl Viro return err; 3371af31f5e5SChris Mason 3372af31f5e5SChris Mason recovery_tree_root: 33730b246afaSJeff Mahoney if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) 3374af31f5e5SChris Mason goto fail_tree_roots; 3375af31f5e5SChris Mason 3376af31f5e5SChris Mason free_root_pointers(fs_info, 0); 3377af31f5e5SChris Mason 3378af31f5e5SChris Mason /* don't use the log in recovery mode, it won't be valid */ 3379af31f5e5SChris Mason btrfs_set_super_log_root(disk_super, 0); 3380af31f5e5SChris Mason 3381af31f5e5SChris Mason /* we can't trust the free space cache either */ 3382af31f5e5SChris Mason btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 3383af31f5e5SChris Mason 3384af31f5e5SChris Mason ret = next_root_backup(fs_info, fs_info->super_copy, 3385af31f5e5SChris Mason &num_backups_tried, &backup_index); 3386af31f5e5SChris Mason if (ret == -1) 3387af31f5e5SChris Mason goto fail_block_groups; 3388af31f5e5SChris Mason goto retry_root_backup; 3389eb60ceacSChris Mason } 3390663faf9fSMasami Hiramatsu ALLOW_ERROR_INJECTION(open_ctree, ERRNO); 3391eb60ceacSChris Mason 3392f2984462SChris Mason static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate) 3393f2984462SChris Mason { 3394f2984462SChris Mason if (uptodate) { 3395f2984462SChris Mason set_buffer_uptodate(bh); 3396f2984462SChris Mason } else { 3397442a4f63SStefan Behrens struct btrfs_device *device = (struct btrfs_device *) 3398442a4f63SStefan Behrens bh->b_private; 3399442a4f63SStefan Behrens 3400fb456252SJeff Mahoney btrfs_warn_rl_in_rcu(device->fs_info, 3401b14af3b4SDavid Sterba "lost page write due to IO error on %s", 3402606686eeSJosef Bacik rcu_str_deref(device->name)); 340301327610SNicholas D Steeves /* note, we don't set_buffer_write_io_error because we have 34041259ab75SChris Mason * our own ways of dealing with the IO errors 34051259ab75SChris Mason */ 3406f2984462SChris Mason clear_buffer_uptodate(bh); 3407442a4f63SStefan Behrens btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS); 3408f2984462SChris Mason } 3409f2984462SChris Mason unlock_buffer(bh); 3410f2984462SChris Mason put_bh(bh); 3411f2984462SChris Mason } 3412f2984462SChris Mason 341329c36d72SAnand Jain int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num, 341429c36d72SAnand Jain struct buffer_head **bh_ret) 341529c36d72SAnand Jain { 341629c36d72SAnand Jain struct buffer_head *bh; 341729c36d72SAnand Jain struct btrfs_super_block *super; 341829c36d72SAnand Jain u64 bytenr; 341929c36d72SAnand Jain 342029c36d72SAnand Jain bytenr = btrfs_sb_offset(copy_num); 342129c36d72SAnand Jain if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode)) 342229c36d72SAnand Jain return -EINVAL; 342329c36d72SAnand Jain 34249f6d2510SDavid Sterba bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE); 342529c36d72SAnand Jain /* 342629c36d72SAnand Jain * If we fail to read from the underlying devices, as of now 342729c36d72SAnand Jain * the best option we have is to mark it EIO. 342829c36d72SAnand Jain */ 342929c36d72SAnand Jain if (!bh) 343029c36d72SAnand Jain return -EIO; 343129c36d72SAnand Jain 343229c36d72SAnand Jain super = (struct btrfs_super_block *)bh->b_data; 343329c36d72SAnand Jain if (btrfs_super_bytenr(super) != bytenr || 343429c36d72SAnand Jain btrfs_super_magic(super) != BTRFS_MAGIC) { 343529c36d72SAnand Jain brelse(bh); 343629c36d72SAnand Jain return -EINVAL; 343729c36d72SAnand Jain } 343829c36d72SAnand Jain 343929c36d72SAnand Jain *bh_ret = bh; 344029c36d72SAnand Jain return 0; 344129c36d72SAnand Jain } 344229c36d72SAnand Jain 344329c36d72SAnand Jain 3444a512bbf8SYan Zheng struct buffer_head *btrfs_read_dev_super(struct block_device *bdev) 3445a512bbf8SYan Zheng { 3446a512bbf8SYan Zheng struct buffer_head *bh; 3447a512bbf8SYan Zheng struct buffer_head *latest = NULL; 3448a512bbf8SYan Zheng struct btrfs_super_block *super; 3449a512bbf8SYan Zheng int i; 3450a512bbf8SYan Zheng u64 transid = 0; 345192fc03fbSAnand Jain int ret = -EINVAL; 3452a512bbf8SYan Zheng 3453a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 3454a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 3455a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 3456a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3457a512bbf8SYan Zheng */ 3458a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 345929c36d72SAnand Jain ret = btrfs_read_dev_one_super(bdev, i, &bh); 346029c36d72SAnand Jain if (ret) 3461a512bbf8SYan Zheng continue; 3462a512bbf8SYan Zheng 3463a512bbf8SYan Zheng super = (struct btrfs_super_block *)bh->b_data; 3464a512bbf8SYan Zheng 3465a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 3466a512bbf8SYan Zheng brelse(latest); 3467a512bbf8SYan Zheng latest = bh; 3468a512bbf8SYan Zheng transid = btrfs_super_generation(super); 3469a512bbf8SYan Zheng } else { 3470a512bbf8SYan Zheng brelse(bh); 3471a512bbf8SYan Zheng } 3472a512bbf8SYan Zheng } 347392fc03fbSAnand Jain 347492fc03fbSAnand Jain if (!latest) 347592fc03fbSAnand Jain return ERR_PTR(ret); 347692fc03fbSAnand Jain 3477a512bbf8SYan Zheng return latest; 3478a512bbf8SYan Zheng } 3479a512bbf8SYan Zheng 34804eedeb75SHisashi Hifumi /* 3481abbb3b8eSDavid Sterba * Write superblock @sb to the @device. Do not wait for completion, all the 3482abbb3b8eSDavid Sterba * buffer heads we write are pinned. 34834eedeb75SHisashi Hifumi * 3484abbb3b8eSDavid Sterba * Write @max_mirrors copies of the superblock, where 0 means default that fit 3485abbb3b8eSDavid Sterba * the expected device size at commit time. Note that max_mirrors must be 3486abbb3b8eSDavid Sterba * same for write and wait phases. 34874eedeb75SHisashi Hifumi * 3488abbb3b8eSDavid Sterba * Return number of errors when buffer head is not found or submission fails. 34894eedeb75SHisashi Hifumi */ 3490a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 3491abbb3b8eSDavid Sterba struct btrfs_super_block *sb, int max_mirrors) 3492a512bbf8SYan Zheng { 3493a512bbf8SYan Zheng struct buffer_head *bh; 3494a512bbf8SYan Zheng int i; 3495a512bbf8SYan Zheng int ret; 3496a512bbf8SYan Zheng int errors = 0; 3497a512bbf8SYan Zheng u32 crc; 3498a512bbf8SYan Zheng u64 bytenr; 34991b9e619cSOmar Sandoval int op_flags; 3500a512bbf8SYan Zheng 3501a512bbf8SYan Zheng if (max_mirrors == 0) 3502a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3503a512bbf8SYan Zheng 3504a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 3505a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 3506935e5cc9SMiao Xie if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3507935e5cc9SMiao Xie device->commit_total_bytes) 3508a512bbf8SYan Zheng break; 3509a512bbf8SYan Zheng 3510a512bbf8SYan Zheng btrfs_set_super_bytenr(sb, bytenr); 3511a512bbf8SYan Zheng 3512a512bbf8SYan Zheng crc = ~(u32)0; 3513abbb3b8eSDavid Sterba crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc, 3514abbb3b8eSDavid Sterba BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE); 3515a512bbf8SYan Zheng btrfs_csum_final(crc, sb->csum); 3516a512bbf8SYan Zheng 3517abbb3b8eSDavid Sterba /* One reference for us, and we leave it for the caller */ 35189f6d2510SDavid Sterba bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, 3519a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 3520634554dcSJosef Bacik if (!bh) { 3521fb456252SJeff Mahoney btrfs_err(device->fs_info, 3522f14d104dSDavid Sterba "couldn't get super buffer head for bytenr %llu", 3523f14d104dSDavid Sterba bytenr); 3524634554dcSJosef Bacik errors++; 3525634554dcSJosef Bacik continue; 3526634554dcSJosef Bacik } 3527634554dcSJosef Bacik 3528a512bbf8SYan Zheng memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE); 3529a512bbf8SYan Zheng 35304eedeb75SHisashi Hifumi /* one reference for submit_bh */ 3531a512bbf8SYan Zheng get_bh(bh); 35324eedeb75SHisashi Hifumi 35334eedeb75SHisashi Hifumi set_buffer_uptodate(bh); 3534a512bbf8SYan Zheng lock_buffer(bh); 3535a512bbf8SYan Zheng bh->b_end_io = btrfs_end_buffer_write_sync; 3536442a4f63SStefan Behrens bh->b_private = device; 3537a512bbf8SYan Zheng 3538387125fcSChris Mason /* 3539387125fcSChris Mason * we fua the first super. The others we allow 3540387125fcSChris Mason * to go down lazy. 3541387125fcSChris Mason */ 35421b9e619cSOmar Sandoval op_flags = REQ_SYNC | REQ_META | REQ_PRIO; 35431b9e619cSOmar Sandoval if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) 35441b9e619cSOmar Sandoval op_flags |= REQ_FUA; 35451b9e619cSOmar Sandoval ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh); 35464eedeb75SHisashi Hifumi if (ret) 3547a512bbf8SYan Zheng errors++; 3548a512bbf8SYan Zheng } 3549a512bbf8SYan Zheng return errors < i ? 0 : -1; 3550a512bbf8SYan Zheng } 3551a512bbf8SYan Zheng 3552387125fcSChris Mason /* 3553abbb3b8eSDavid Sterba * Wait for write completion of superblocks done by write_dev_supers, 3554abbb3b8eSDavid Sterba * @max_mirrors same for write and wait phases. 3555abbb3b8eSDavid Sterba * 3556abbb3b8eSDavid Sterba * Return number of errors when buffer head is not found or not marked up to 3557abbb3b8eSDavid Sterba * date. 3558abbb3b8eSDavid Sterba */ 3559abbb3b8eSDavid Sterba static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) 3560abbb3b8eSDavid Sterba { 3561abbb3b8eSDavid Sterba struct buffer_head *bh; 3562abbb3b8eSDavid Sterba int i; 3563abbb3b8eSDavid Sterba int errors = 0; 3564b6a535faSHoward McLauchlan bool primary_failed = false; 3565abbb3b8eSDavid Sterba u64 bytenr; 3566abbb3b8eSDavid Sterba 3567abbb3b8eSDavid Sterba if (max_mirrors == 0) 3568abbb3b8eSDavid Sterba max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3569abbb3b8eSDavid Sterba 3570abbb3b8eSDavid Sterba for (i = 0; i < max_mirrors; i++) { 3571abbb3b8eSDavid Sterba bytenr = btrfs_sb_offset(i); 3572abbb3b8eSDavid Sterba if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3573abbb3b8eSDavid Sterba device->commit_total_bytes) 3574abbb3b8eSDavid Sterba break; 3575abbb3b8eSDavid Sterba 35769f6d2510SDavid Sterba bh = __find_get_block(device->bdev, 35779f6d2510SDavid Sterba bytenr / BTRFS_BDEV_BLOCKSIZE, 3578abbb3b8eSDavid Sterba BTRFS_SUPER_INFO_SIZE); 3579abbb3b8eSDavid Sterba if (!bh) { 3580abbb3b8eSDavid Sterba errors++; 3581b6a535faSHoward McLauchlan if (i == 0) 3582b6a535faSHoward McLauchlan primary_failed = true; 3583abbb3b8eSDavid Sterba continue; 3584abbb3b8eSDavid Sterba } 3585abbb3b8eSDavid Sterba wait_on_buffer(bh); 3586b6a535faSHoward McLauchlan if (!buffer_uptodate(bh)) { 3587abbb3b8eSDavid Sterba errors++; 3588b6a535faSHoward McLauchlan if (i == 0) 3589b6a535faSHoward McLauchlan primary_failed = true; 3590b6a535faSHoward McLauchlan } 3591abbb3b8eSDavid Sterba 3592abbb3b8eSDavid Sterba /* drop our reference */ 3593abbb3b8eSDavid Sterba brelse(bh); 3594abbb3b8eSDavid Sterba 3595abbb3b8eSDavid Sterba /* drop the reference from the writing run */ 3596abbb3b8eSDavid Sterba brelse(bh); 3597abbb3b8eSDavid Sterba } 3598abbb3b8eSDavid Sterba 3599b6a535faSHoward McLauchlan /* log error, force error return */ 3600b6a535faSHoward McLauchlan if (primary_failed) { 3601b6a535faSHoward McLauchlan btrfs_err(device->fs_info, "error writing primary super block to device %llu", 3602b6a535faSHoward McLauchlan device->devid); 3603b6a535faSHoward McLauchlan return -1; 3604b6a535faSHoward McLauchlan } 3605b6a535faSHoward McLauchlan 3606abbb3b8eSDavid Sterba return errors < i ? 0 : -1; 3607abbb3b8eSDavid Sterba } 3608abbb3b8eSDavid Sterba 3609abbb3b8eSDavid Sterba /* 3610387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 3611387125fcSChris Mason * for the barrier when it is done 3612387125fcSChris Mason */ 36134246a0b6SChristoph Hellwig static void btrfs_end_empty_barrier(struct bio *bio) 3614387125fcSChris Mason { 3615387125fcSChris Mason complete(bio->bi_private); 3616387125fcSChris Mason } 3617387125fcSChris Mason 3618387125fcSChris Mason /* 36194fc6441aSAnand Jain * Submit a flush request to the device if it supports it. Error handling is 36204fc6441aSAnand Jain * done in the waiting counterpart. 3621387125fcSChris Mason */ 36224fc6441aSAnand Jain static void write_dev_flush(struct btrfs_device *device) 3623387125fcSChris Mason { 3624c2a9c7abSAnand Jain struct request_queue *q = bdev_get_queue(device->bdev); 3625e0ae9994SDavid Sterba struct bio *bio = device->flush_bio; 3626387125fcSChris Mason 3627c2a9c7abSAnand Jain if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) 36284fc6441aSAnand Jain return; 3629387125fcSChris Mason 3630e0ae9994SDavid Sterba bio_reset(bio); 3631387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 363274d46992SChristoph Hellwig bio_set_dev(bio, device->bdev); 36338d910125SJan Kara bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH; 3634387125fcSChris Mason init_completion(&device->flush_wait); 3635387125fcSChris Mason bio->bi_private = &device->flush_wait; 3636387125fcSChris Mason 363743a01111SLu Fengqi btrfsic_submit_bio(bio); 36381c3063b6SAnand Jain set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 36394fc6441aSAnand Jain } 3640387125fcSChris Mason 36414fc6441aSAnand Jain /* 36424fc6441aSAnand Jain * If the flush bio has been submitted by write_dev_flush, wait for it. 36434fc6441aSAnand Jain */ 36448c27cb35SLinus Torvalds static blk_status_t wait_dev_flush(struct btrfs_device *device) 36454fc6441aSAnand Jain { 36464fc6441aSAnand Jain struct bio *bio = device->flush_bio; 36474fc6441aSAnand Jain 36481c3063b6SAnand Jain if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) 364958efbc9fSOmar Sandoval return BLK_STS_OK; 36504fc6441aSAnand Jain 36511c3063b6SAnand Jain clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 36522980d574SDavid Sterba wait_for_completion_io(&device->flush_wait); 36534fc6441aSAnand Jain 36548c27cb35SLinus Torvalds return bio->bi_status; 3655387125fcSChris Mason } 3656387125fcSChris Mason 3657d10b82feSQu Wenruo static int check_barrier_error(struct btrfs_fs_info *fs_info) 3658401b41e5SAnand Jain { 36596528b99dSAnand Jain if (!btrfs_check_rw_degradable(fs_info, NULL)) 3660401b41e5SAnand Jain return -EIO; 3661387125fcSChris Mason return 0; 3662387125fcSChris Mason } 3663387125fcSChris Mason 3664387125fcSChris Mason /* 3665387125fcSChris Mason * send an empty flush down to each device in parallel, 3666387125fcSChris Mason * then wait for them 3667387125fcSChris Mason */ 3668387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 3669387125fcSChris Mason { 3670387125fcSChris Mason struct list_head *head; 3671387125fcSChris Mason struct btrfs_device *dev; 36725af3e8ccSStefan Behrens int errors_wait = 0; 36734e4cbee9SChristoph Hellwig blk_status_t ret; 3674387125fcSChris Mason 36751538e6c5SDavid Sterba lockdep_assert_held(&info->fs_devices->device_list_mutex); 3676387125fcSChris Mason /* send down all the barriers */ 3677387125fcSChris Mason head = &info->fs_devices->devices; 36781538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3679e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3680f88ba6a2SHidetoshi Seto continue; 3681cea7c8bfSAnand Jain if (!dev->bdev) 3682387125fcSChris Mason continue; 3683e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3684ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3685387125fcSChris Mason continue; 3686387125fcSChris Mason 36874fc6441aSAnand Jain write_dev_flush(dev); 368858efbc9fSOmar Sandoval dev->last_flush_error = BLK_STS_OK; 3689387125fcSChris Mason } 3690387125fcSChris Mason 3691387125fcSChris Mason /* wait for all the barriers */ 36921538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3693e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3694f88ba6a2SHidetoshi Seto continue; 3695387125fcSChris Mason if (!dev->bdev) { 36965af3e8ccSStefan Behrens errors_wait++; 3697387125fcSChris Mason continue; 3698387125fcSChris Mason } 3699e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3700ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3701387125fcSChris Mason continue; 3702387125fcSChris Mason 37034fc6441aSAnand Jain ret = wait_dev_flush(dev); 3704401b41e5SAnand Jain if (ret) { 3705401b41e5SAnand Jain dev->last_flush_error = ret; 370666b4993eSDavid Sterba btrfs_dev_stat_inc_and_print(dev, 370766b4993eSDavid Sterba BTRFS_DEV_STAT_FLUSH_ERRS); 37085af3e8ccSStefan Behrens errors_wait++; 3709387125fcSChris Mason } 3710401b41e5SAnand Jain } 3711401b41e5SAnand Jain 3712cea7c8bfSAnand Jain if (errors_wait) { 3713401b41e5SAnand Jain /* 3714401b41e5SAnand Jain * At some point we need the status of all disks 3715401b41e5SAnand Jain * to arrive at the volume status. So error checking 3716401b41e5SAnand Jain * is being pushed to a separate loop. 3717401b41e5SAnand Jain */ 3718d10b82feSQu Wenruo return check_barrier_error(info); 3719401b41e5SAnand Jain } 3720387125fcSChris Mason return 0; 3721387125fcSChris Mason } 3722387125fcSChris Mason 3723943c6e99SZhao Lei int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 3724943c6e99SZhao Lei { 37258789f4feSZhao Lei int raid_type; 37268789f4feSZhao Lei int min_tolerated = INT_MAX; 3727943c6e99SZhao Lei 37288789f4feSZhao Lei if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 37298789f4feSZhao Lei (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 37308c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 37318789f4feSZhao Lei btrfs_raid_array[BTRFS_RAID_SINGLE]. 37328789f4feSZhao Lei tolerated_failures); 3733943c6e99SZhao Lei 37348789f4feSZhao Lei for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 37358789f4feSZhao Lei if (raid_type == BTRFS_RAID_SINGLE) 37368789f4feSZhao Lei continue; 373741a6e891SAnand Jain if (!(flags & btrfs_raid_array[raid_type].bg_flag)) 37388789f4feSZhao Lei continue; 37398c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 37408789f4feSZhao Lei btrfs_raid_array[raid_type]. 37418789f4feSZhao Lei tolerated_failures); 37428789f4feSZhao Lei } 3743943c6e99SZhao Lei 37448789f4feSZhao Lei if (min_tolerated == INT_MAX) { 3745ab8d0fc4SJeff Mahoney pr_warn("BTRFS: unknown raid flag: %llu", flags); 37468789f4feSZhao Lei min_tolerated = 0; 37478789f4feSZhao Lei } 37488789f4feSZhao Lei 37498789f4feSZhao Lei return min_tolerated; 3750943c6e99SZhao Lei } 3751943c6e99SZhao Lei 3752eece6a9cSDavid Sterba int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) 3753f2984462SChris Mason { 3754e5e9a520SChris Mason struct list_head *head; 3755f2984462SChris Mason struct btrfs_device *dev; 3756a061fc8dSChris Mason struct btrfs_super_block *sb; 3757f2984462SChris Mason struct btrfs_dev_item *dev_item; 3758f2984462SChris Mason int ret; 3759f2984462SChris Mason int do_barriers; 3760a236aed1SChris Mason int max_errors; 3761a236aed1SChris Mason int total_errors = 0; 3762a061fc8dSChris Mason u64 flags; 3763f2984462SChris Mason 37640b246afaSJeff Mahoney do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); 3765fed3b381SLiu Bo 3766fed3b381SLiu Bo /* 3767fed3b381SLiu Bo * max_mirrors == 0 indicates we're from commit_transaction, 3768fed3b381SLiu Bo * not from fsync where the tree roots in fs_info have not 3769fed3b381SLiu Bo * been consistent on disk. 3770fed3b381SLiu Bo */ 3771fed3b381SLiu Bo if (max_mirrors == 0) 37720b246afaSJeff Mahoney backup_super_roots(fs_info); 3773f2984462SChris Mason 37740b246afaSJeff Mahoney sb = fs_info->super_for_commit; 3775a061fc8dSChris Mason dev_item = &sb->dev_item; 3776e5e9a520SChris Mason 37770b246afaSJeff Mahoney mutex_lock(&fs_info->fs_devices->device_list_mutex); 37780b246afaSJeff Mahoney head = &fs_info->fs_devices->devices; 37790b246afaSJeff Mahoney max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; 3780387125fcSChris Mason 37815af3e8ccSStefan Behrens if (do_barriers) { 37820b246afaSJeff Mahoney ret = barrier_all_devices(fs_info); 37835af3e8ccSStefan Behrens if (ret) { 37845af3e8ccSStefan Behrens mutex_unlock( 37850b246afaSJeff Mahoney &fs_info->fs_devices->device_list_mutex); 37860b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 37875af3e8ccSStefan Behrens "errors while submitting device barriers."); 37885af3e8ccSStefan Behrens return ret; 37895af3e8ccSStefan Behrens } 37905af3e8ccSStefan Behrens } 3791387125fcSChris Mason 37921538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3793dfe25020SChris Mason if (!dev->bdev) { 3794dfe25020SChris Mason total_errors++; 3795dfe25020SChris Mason continue; 3796dfe25020SChris Mason } 3797e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3798ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3799dfe25020SChris Mason continue; 3800dfe25020SChris Mason 38012b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 3802a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 3803a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 38047df69d3eSMiao Xie btrfs_set_stack_device_total_bytes(dev_item, 3805935e5cc9SMiao Xie dev->commit_total_bytes); 3806ce7213c7SMiao Xie btrfs_set_stack_device_bytes_used(dev_item, 3807ce7213c7SMiao Xie dev->commit_bytes_used); 3808a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 3809a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 3810a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 3811a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 38127239ff4bSNikolay Borisov memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, 38137239ff4bSNikolay Borisov BTRFS_FSID_SIZE); 3814a512bbf8SYan Zheng 3815a061fc8dSChris Mason flags = btrfs_super_flags(sb); 3816a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 3817f2984462SChris Mason 381875cb857dSQu Wenruo ret = btrfs_validate_write_super(fs_info, sb); 381975cb857dSQu Wenruo if (ret < 0) { 382075cb857dSQu Wenruo mutex_unlock(&fs_info->fs_devices->device_list_mutex); 382175cb857dSQu Wenruo btrfs_handle_fs_error(fs_info, -EUCLEAN, 382275cb857dSQu Wenruo "unexpected superblock corruption detected"); 382375cb857dSQu Wenruo return -EUCLEAN; 382475cb857dSQu Wenruo } 382575cb857dSQu Wenruo 3826abbb3b8eSDavid Sterba ret = write_dev_supers(dev, sb, max_mirrors); 3827a236aed1SChris Mason if (ret) 3828a236aed1SChris Mason total_errors++; 3829f2984462SChris Mason } 3830a236aed1SChris Mason if (total_errors > max_errors) { 38310b246afaSJeff Mahoney btrfs_err(fs_info, "%d errors while writing supers", 3832d397712bSChris Mason total_errors); 38330b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 383479787eaaSJeff Mahoney 38359d565ba4SStefan Behrens /* FUA is masked off if unsupported and can't be the reason */ 38360b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 38370b246afaSJeff Mahoney "%d errors while writing supers", 38380b246afaSJeff Mahoney total_errors); 38399d565ba4SStefan Behrens return -EIO; 3840a236aed1SChris Mason } 3841f2984462SChris Mason 3842a512bbf8SYan Zheng total_errors = 0; 38431538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3844dfe25020SChris Mason if (!dev->bdev) 3845dfe25020SChris Mason continue; 3846e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3847ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3848dfe25020SChris Mason continue; 3849dfe25020SChris Mason 3850abbb3b8eSDavid Sterba ret = wait_dev_supers(dev, max_mirrors); 3851a512bbf8SYan Zheng if (ret) 38521259ab75SChris Mason total_errors++; 3853f2984462SChris Mason } 38540b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 3855a236aed1SChris Mason if (total_errors > max_errors) { 38560b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 38570b246afaSJeff Mahoney "%d errors while writing supers", 38580b246afaSJeff Mahoney total_errors); 385979787eaaSJeff Mahoney return -EIO; 3860a236aed1SChris Mason } 3861f2984462SChris Mason return 0; 3862f2984462SChris Mason } 3863f2984462SChris Mason 3864cb517eabSMiao Xie /* Drop a fs root from the radix tree and free it. */ 3865cb517eabSMiao Xie void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 3866cb517eabSMiao Xie struct btrfs_root *root) 38672619ba1fSChris Mason { 38684df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 38692619ba1fSChris Mason radix_tree_delete(&fs_info->fs_roots_radix, 38702619ba1fSChris Mason (unsigned long)root->root_key.objectid); 38714df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 387276dda93cSYan, Zheng 387376dda93cSYan, Zheng if (btrfs_root_refs(&root->root_item) == 0) 387476dda93cSYan, Zheng synchronize_srcu(&fs_info->subvol_srcu); 387576dda93cSYan, Zheng 38761c1ea4f7SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 38773321719eSLiu Bo btrfs_free_log(NULL, root); 38781c1ea4f7SLiu Bo if (root->reloc_root) { 38791c1ea4f7SLiu Bo free_extent_buffer(root->reloc_root->node); 38801c1ea4f7SLiu Bo free_extent_buffer(root->reloc_root->commit_root); 38811c1ea4f7SLiu Bo btrfs_put_fs_root(root->reloc_root); 38821c1ea4f7SLiu Bo root->reloc_root = NULL; 38831c1ea4f7SLiu Bo } 38841c1ea4f7SLiu Bo } 38853321719eSLiu Bo 3886faa2dbf0SJosef Bacik if (root->free_ino_pinned) 3887581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_pinned); 3888faa2dbf0SJosef Bacik if (root->free_ino_ctl) 3889581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_ctl); 389084db5ccfSDavid Sterba btrfs_free_fs_root(root); 38914df27c4dSYan, Zheng } 38924df27c4dSYan, Zheng 389384db5ccfSDavid Sterba void btrfs_free_fs_root(struct btrfs_root *root) 38944df27c4dSYan, Zheng { 389557cdc8dbSDavid Sterba iput(root->ino_cache_inode); 38964df27c4dSYan, Zheng WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 38970ee5dc67SAl Viro if (root->anon_dev) 38980ee5dc67SAl Viro free_anon_bdev(root->anon_dev); 38998257b2dcSMiao Xie if (root->subv_writers) 39008257b2dcSMiao Xie btrfs_free_subvolume_writers(root->subv_writers); 39015f39d397SChris Mason free_extent_buffer(root->node); 39025f39d397SChris Mason free_extent_buffer(root->commit_root); 3903581bb050SLi Zefan kfree(root->free_ino_ctl); 3904581bb050SLi Zefan kfree(root->free_ino_pinned); 3905b0feb9d9SMiao Xie btrfs_put_fs_root(root); 39062619ba1fSChris Mason } 39072619ba1fSChris Mason 3908c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 3909c146afadSYan Zheng { 3910c146afadSYan Zheng u64 root_objectid = 0; 3911c146afadSYan Zheng struct btrfs_root *gang[8]; 391265d33fd7SQu Wenruo int i = 0; 391365d33fd7SQu Wenruo int err = 0; 391465d33fd7SQu Wenruo unsigned int ret = 0; 391565d33fd7SQu Wenruo int index; 3916c146afadSYan Zheng 3917c146afadSYan Zheng while (1) { 391865d33fd7SQu Wenruo index = srcu_read_lock(&fs_info->subvol_srcu); 3919c146afadSYan Zheng ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 3920c146afadSYan Zheng (void **)gang, root_objectid, 3921c146afadSYan Zheng ARRAY_SIZE(gang)); 392265d33fd7SQu Wenruo if (!ret) { 392365d33fd7SQu Wenruo srcu_read_unlock(&fs_info->subvol_srcu, index); 3924c146afadSYan Zheng break; 392565d33fd7SQu Wenruo } 39265d4f98a2SYan Zheng root_objectid = gang[ret - 1]->root_key.objectid + 1; 392766b4ffd1SJosef Bacik 392865d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 392965d33fd7SQu Wenruo /* Avoid to grab roots in dead_roots */ 393065d33fd7SQu Wenruo if (btrfs_root_refs(&gang[i]->root_item) == 0) { 393165d33fd7SQu Wenruo gang[i] = NULL; 393265d33fd7SQu Wenruo continue; 393365d33fd7SQu Wenruo } 393465d33fd7SQu Wenruo /* grab all the search result for later use */ 393565d33fd7SQu Wenruo gang[i] = btrfs_grab_fs_root(gang[i]); 393665d33fd7SQu Wenruo } 393765d33fd7SQu Wenruo srcu_read_unlock(&fs_info->subvol_srcu, index); 393865d33fd7SQu Wenruo 393965d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 394065d33fd7SQu Wenruo if (!gang[i]) 394165d33fd7SQu Wenruo continue; 3942c146afadSYan Zheng root_objectid = gang[i]->root_key.objectid; 394366b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(gang[i]); 394466b4ffd1SJosef Bacik if (err) 394565d33fd7SQu Wenruo break; 394665d33fd7SQu Wenruo btrfs_put_fs_root(gang[i]); 3947c146afadSYan Zheng } 3948c146afadSYan Zheng root_objectid++; 3949c146afadSYan Zheng } 395065d33fd7SQu Wenruo 395165d33fd7SQu Wenruo /* release the uncleaned roots due to error */ 395265d33fd7SQu Wenruo for (; i < ret; i++) { 395365d33fd7SQu Wenruo if (gang[i]) 395465d33fd7SQu Wenruo btrfs_put_fs_root(gang[i]); 395565d33fd7SQu Wenruo } 395665d33fd7SQu Wenruo return err; 3957c146afadSYan Zheng } 3958c146afadSYan Zheng 39596bccf3abSJeff Mahoney int btrfs_commit_super(struct btrfs_fs_info *fs_info) 3960c146afadSYan Zheng { 39616bccf3abSJeff Mahoney struct btrfs_root *root = fs_info->tree_root; 3962c146afadSYan Zheng struct btrfs_trans_handle *trans; 3963c146afadSYan Zheng 39640b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 39652ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 39660b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 39670b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 3968c71bf099SYan, Zheng 3969c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 39700b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 39710b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 3972c71bf099SYan, Zheng 39737a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 39743612b495STsutomu Itoh if (IS_ERR(trans)) 39753612b495STsutomu Itoh return PTR_ERR(trans); 39763a45bb20SJeff Mahoney return btrfs_commit_transaction(trans); 3977c146afadSYan Zheng } 3978c146afadSYan Zheng 39796bccf3abSJeff Mahoney void close_ctree(struct btrfs_fs_info *fs_info) 3980eb60ceacSChris Mason { 3981c146afadSYan Zheng int ret; 3982e089f05cSChris Mason 3983afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); 3984d6fd0ae2SOmar Sandoval /* 3985d6fd0ae2SOmar Sandoval * We don't want the cleaner to start new transactions, add more delayed 3986d6fd0ae2SOmar Sandoval * iputs, etc. while we're closing. We can't use kthread_stop() yet 3987d6fd0ae2SOmar Sandoval * because that frees the task_struct, and the transaction kthread might 3988d6fd0ae2SOmar Sandoval * still try to wake up the cleaner. 3989d6fd0ae2SOmar Sandoval */ 3990d6fd0ae2SOmar Sandoval kthread_park(fs_info->cleaner_kthread); 3991a2135011SChris Mason 39927343dd61SJustin Maggard /* wait for the qgroup rescan worker to stop */ 3993d06f23d6SJeff Mahoney btrfs_qgroup_wait_for_completion(fs_info, false); 39947343dd61SJustin Maggard 3995803b2f54SStefan Behrens /* wait for the uuid_scan task to finish */ 3996803b2f54SStefan Behrens down(&fs_info->uuid_tree_rescan_sem); 3997803b2f54SStefan Behrens /* avoid complains from lockdep et al., set sem back to initial state */ 3998803b2f54SStefan Behrens up(&fs_info->uuid_tree_rescan_sem); 3999803b2f54SStefan Behrens 4000837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 4001aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 4002837d5b6eSIlya Dryomov 40038dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 40048dabb742SStefan Behrens 4005aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 40064cb5300bSChris Mason 40074cb5300bSChris Mason /* wait for any defraggers to finish */ 40084cb5300bSChris Mason wait_event(fs_info->transaction_wait, 40094cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 40104cb5300bSChris Mason 40114cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 401226176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 40134cb5300bSChris Mason 401421c7e756SMiao Xie cancel_work_sync(&fs_info->async_reclaim_work); 401521c7e756SMiao Xie 4016bc98a42cSDavid Howells if (!sb_rdonly(fs_info->sb)) { 4017e44163e1SJeff Mahoney /* 4018d6fd0ae2SOmar Sandoval * The cleaner kthread is stopped, so do one final pass over 4019d6fd0ae2SOmar Sandoval * unused block groups. 4020e44163e1SJeff Mahoney */ 40210b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 4022e44163e1SJeff Mahoney 40236bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 4024d397712bSChris Mason if (ret) 402504892340SEric Sandeen btrfs_err(fs_info, "commit super ret %d", ret); 4026c146afadSYan Zheng } 4027ed2ff2cbSChris Mason 4028af722733SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) || 4029af722733SLiu Bo test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) 40302ff7e61eSJeff Mahoney btrfs_error_commit_super(fs_info); 4031acce952bSliubo 4032e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 4033e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 40348929ecfaSYan, Zheng 4035e187831eSJosef Bacik ASSERT(list_empty(&fs_info->delayed_iputs)); 4036afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); 4037f25784b3SYan Zheng 403804892340SEric Sandeen btrfs_free_qgroup_config(fs_info); 4039fe816d0fSNikolay Borisov ASSERT(list_empty(&fs_info->delalloc_roots)); 4040bcef60f2SArne Jansen 4041963d678bSMiao Xie if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 404204892340SEric Sandeen btrfs_info(fs_info, "at unmount delalloc count %lld", 4043963d678bSMiao Xie percpu_counter_sum(&fs_info->delalloc_bytes)); 4044b0c68f8bSChris Mason } 404531153d81SYan Zheng 40464297ff84SJosef Bacik if (percpu_counter_sum(&fs_info->dio_bytes)) 40474297ff84SJosef Bacik btrfs_info(fs_info, "at unmount dio bytes count %lld", 40484297ff84SJosef Bacik percpu_counter_sum(&fs_info->dio_bytes)); 40494297ff84SJosef Bacik 40506618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 4051b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 40525ac1d209SJeff Mahoney 4053faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 4054d10c5f31SChris Mason 40551a4319ccSLiu Bo btrfs_put_block_group_cache(fs_info); 40561a4319ccSLiu Bo 4057de348ee0SWang Shilong /* 4058de348ee0SWang Shilong * we must make sure there is not any read request to 4059de348ee0SWang Shilong * submit after we stopping all workers. 4060de348ee0SWang Shilong */ 4061de348ee0SWang Shilong invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 406296192499SJosef Bacik btrfs_stop_all_workers(fs_info); 406396192499SJosef Bacik 40645cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 40655cdd7db6SFilipe Manana 4066afcdd129SJosef Bacik clear_bit(BTRFS_FS_OPEN, &fs_info->flags); 406713e6c37bSJosef Bacik free_root_pointers(fs_info, 1); 40689ad6b7bcSChris Mason 406913e6c37bSJosef Bacik iput(fs_info->btree_inode); 4070d6bfde87SChris Mason 407121adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 40720b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) 40732ff7e61eSJeff Mahoney btrfsic_unmount(fs_info->fs_devices); 407421adbd5cSStefan Behrens #endif 407521adbd5cSStefan Behrens 40760b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 407768c94e55SNikolay Borisov btrfs_close_devices(fs_info->fs_devices); 4078b248a415SChris Mason 4079e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 4080963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 40814297ff84SJosef Bacik percpu_counter_destroy(&fs_info->dio_bytes); 40827f8d236aSDavid Sterba percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 408376dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 40840b86a832SChris Mason 408553b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 4086fd708b81SJosef Bacik btrfs_free_ref_cache(fs_info); 4087eb60ceacSChris Mason } 4088eb60ceacSChris Mason 4089b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 4090b9fab919SChris Mason int atomic) 4091ccd467d6SChris Mason { 40921259ab75SChris Mason int ret; 4093727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 40941259ab75SChris Mason 40950b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 40961259ab75SChris Mason if (!ret) 40971259ab75SChris Mason return ret; 40981259ab75SChris Mason 40991259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 4100b9fab919SChris Mason parent_transid, atomic); 4101b9fab919SChris Mason if (ret == -EAGAIN) 4102b9fab919SChris Mason return ret; 41031259ab75SChris Mason return !ret; 41045f39d397SChris Mason } 41056702ed49SChris Mason 41065f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 41075f39d397SChris Mason { 41080b246afaSJeff Mahoney struct btrfs_fs_info *fs_info; 410906ea65a3SJosef Bacik struct btrfs_root *root; 41105f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 4111b9473439SChris Mason int was_dirty; 4112b4ce94deSChris Mason 411306ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 411406ea65a3SJosef Bacik /* 411506ea65a3SJosef Bacik * This is a fast path so only do this check if we have sanity tests 411652042d8eSAndrea Gelmini * enabled. Normal people shouldn't be using unmapped buffers as dirty 411706ea65a3SJosef Bacik * outside of the sanity tests. 411806ea65a3SJosef Bacik */ 4119b0132a3bSNikolay Borisov if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) 412006ea65a3SJosef Bacik return; 412106ea65a3SJosef Bacik #endif 412206ea65a3SJosef Bacik root = BTRFS_I(buf->pages[0]->mapping->host)->root; 41230b246afaSJeff Mahoney fs_info = root->fs_info; 4124b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 41250b246afaSJeff Mahoney if (transid != fs_info->generation) 41265d163e0eSJeff Mahoney WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", 41270b246afaSJeff Mahoney buf->start, transid, fs_info->generation); 41280b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 4129e2d84521SMiao Xie if (!was_dirty) 4130104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 4131e2d84521SMiao Xie buf->len, 41320b246afaSJeff Mahoney fs_info->dirty_metadata_batch); 41331f21ef0aSFilipe Manana #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 413469fc6cbbSQu Wenruo /* 413569fc6cbbSQu Wenruo * Since btrfs_mark_buffer_dirty() can be called with item pointer set 413669fc6cbbSQu Wenruo * but item data not updated. 413769fc6cbbSQu Wenruo * So here we should only check item pointers, not item data. 413869fc6cbbSQu Wenruo */ 413969fc6cbbSQu Wenruo if (btrfs_header_level(buf) == 0 && 4140cfdaad5eSDavid Sterba btrfs_check_leaf_relaxed(buf)) { 4141a4f78750SDavid Sterba btrfs_print_leaf(buf); 41421f21ef0aSFilipe Manana ASSERT(0); 41431f21ef0aSFilipe Manana } 41441f21ef0aSFilipe Manana #endif 4145eb60ceacSChris Mason } 4146eb60ceacSChris Mason 41472ff7e61eSJeff Mahoney static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, 4148b53d3f5dSLiu Bo int flush_delayed) 414935b7e476SChris Mason { 4150188de649SChris Mason /* 4151188de649SChris Mason * looks as though older kernels can get into trouble with 4152188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 4153188de649SChris Mason */ 4154e2d84521SMiao Xie int ret; 4155d6bfde87SChris Mason 41566933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 4157d6bfde87SChris Mason return; 4158d6bfde87SChris Mason 4159b53d3f5dSLiu Bo if (flush_delayed) 41602ff7e61eSJeff Mahoney btrfs_balance_delayed_items(fs_info); 416116cdcec7SMiao Xie 4162d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 4163d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 4164d814a491SEthan Lien fs_info->dirty_metadata_batch); 4165e2d84521SMiao Xie if (ret > 0) { 41660b246afaSJeff Mahoney balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); 416716cdcec7SMiao Xie } 416816cdcec7SMiao Xie } 416916cdcec7SMiao Xie 41702ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) 417116cdcec7SMiao Xie { 41722ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 1); 417335b7e476SChris Mason } 4174b53d3f5dSLiu Bo 41752ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) 4176b53d3f5dSLiu Bo { 41772ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 0); 4178d6bfde87SChris Mason } 41796b80053dSChris Mason 4180581c1760SQu Wenruo int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level, 4181581c1760SQu Wenruo struct btrfs_key *first_key) 41826b80053dSChris Mason { 41835ab12d1fSDavid Sterba return btree_read_extent_buffer_pages(buf, parent_transid, 4184581c1760SQu Wenruo level, first_key); 41856b80053dSChris Mason } 41860da5468fSChris Mason 41872ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) 4188acce952bSliubo { 4189fe816d0fSNikolay Borisov /* cleanup FS via transaction */ 4190fe816d0fSNikolay Borisov btrfs_cleanup_transaction(fs_info); 4191fe816d0fSNikolay Borisov 41920b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 41932ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 41940b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 4195acce952bSliubo 41960b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 41970b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4198acce952bSliubo } 4199acce952bSliubo 4200143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4201acce952bSliubo { 4202acce952bSliubo struct btrfs_ordered_extent *ordered; 4203acce952bSliubo 4204199c2a9cSMiao Xie spin_lock(&root->ordered_extent_lock); 4205779880efSJosef Bacik /* 4206779880efSJosef Bacik * This will just short circuit the ordered completion stuff which will 4207779880efSJosef Bacik * make sure the ordered extent gets properly cleaned up. 4208779880efSJosef Bacik */ 4209199c2a9cSMiao Xie list_for_each_entry(ordered, &root->ordered_extents, 4210779880efSJosef Bacik root_extent_list) 4211779880efSJosef Bacik set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4212199c2a9cSMiao Xie spin_unlock(&root->ordered_extent_lock); 4213199c2a9cSMiao Xie } 4214199c2a9cSMiao Xie 4215199c2a9cSMiao Xie static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4216199c2a9cSMiao Xie { 4217199c2a9cSMiao Xie struct btrfs_root *root; 4218199c2a9cSMiao Xie struct list_head splice; 4219199c2a9cSMiao Xie 4220199c2a9cSMiao Xie INIT_LIST_HEAD(&splice); 4221199c2a9cSMiao Xie 4222199c2a9cSMiao Xie spin_lock(&fs_info->ordered_root_lock); 4223199c2a9cSMiao Xie list_splice_init(&fs_info->ordered_roots, &splice); 4224199c2a9cSMiao Xie while (!list_empty(&splice)) { 4225199c2a9cSMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4226199c2a9cSMiao Xie ordered_root); 42271de2cfdeSJosef Bacik list_move_tail(&root->ordered_root, 42281de2cfdeSJosef Bacik &fs_info->ordered_roots); 4229199c2a9cSMiao Xie 42302a85d9caSLiu Bo spin_unlock(&fs_info->ordered_root_lock); 4231199c2a9cSMiao Xie btrfs_destroy_ordered_extents(root); 4232199c2a9cSMiao Xie 42332a85d9caSLiu Bo cond_resched(); 42342a85d9caSLiu Bo spin_lock(&fs_info->ordered_root_lock); 4235199c2a9cSMiao Xie } 4236199c2a9cSMiao Xie spin_unlock(&fs_info->ordered_root_lock); 423774d5d229SJosef Bacik 423874d5d229SJosef Bacik /* 423974d5d229SJosef Bacik * We need this here because if we've been flipped read-only we won't 424074d5d229SJosef Bacik * get sync() from the umount, so we need to make sure any ordered 424174d5d229SJosef Bacik * extents that haven't had their dirty pages IO start writeout yet 424274d5d229SJosef Bacik * actually get run and error out properly. 424374d5d229SJosef Bacik */ 424474d5d229SJosef Bacik btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 4245acce952bSliubo } 4246acce952bSliubo 424735a3621bSStefan Behrens static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 42482ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4249acce952bSliubo { 4250acce952bSliubo struct rb_node *node; 4251acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 4252acce952bSliubo struct btrfs_delayed_ref_node *ref; 4253acce952bSliubo int ret = 0; 4254acce952bSliubo 4255acce952bSliubo delayed_refs = &trans->delayed_refs; 4256acce952bSliubo 4257acce952bSliubo spin_lock(&delayed_refs->lock); 4258d7df2c79SJosef Bacik if (atomic_read(&delayed_refs->num_entries) == 0) { 4259cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 42600b246afaSJeff Mahoney btrfs_info(fs_info, "delayed_refs has NO entry"); 4261acce952bSliubo return ret; 4262acce952bSliubo } 4263acce952bSliubo 42645c9d028bSLiu Bo while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { 4265d7df2c79SJosef Bacik struct btrfs_delayed_ref_head *head; 42660e0adbcfSJosef Bacik struct rb_node *n; 4267e78417d1SJosef Bacik bool pin_bytes = false; 4268acce952bSliubo 4269d7df2c79SJosef Bacik head = rb_entry(node, struct btrfs_delayed_ref_head, 4270d7df2c79SJosef Bacik href_node); 42713069bd26SJosef Bacik if (btrfs_delayed_ref_lock(delayed_refs, head)) 4272b939d1abSJosef Bacik continue; 42733069bd26SJosef Bacik 4274d7df2c79SJosef Bacik spin_lock(&head->lock); 4275e3d03965SLiu Bo while ((n = rb_first_cached(&head->ref_tree)) != NULL) { 42760e0adbcfSJosef Bacik ref = rb_entry(n, struct btrfs_delayed_ref_node, 42770e0adbcfSJosef Bacik ref_node); 4278d7df2c79SJosef Bacik ref->in_tree = 0; 4279e3d03965SLiu Bo rb_erase_cached(&ref->ref_node, &head->ref_tree); 42800e0adbcfSJosef Bacik RB_CLEAR_NODE(&ref->ref_node); 42811d57ee94SWang Xiaoguang if (!list_empty(&ref->add_list)) 42821d57ee94SWang Xiaoguang list_del(&ref->add_list); 4283d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4284d7df2c79SJosef Bacik btrfs_put_delayed_ref(ref); 4285d7df2c79SJosef Bacik } 428654067ae9SJosef Bacik if (head->must_insert_reserved) 4287e78417d1SJosef Bacik pin_bytes = true; 428878a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 4289fa781ceaSJosef Bacik btrfs_delete_ref_head(delayed_refs, head); 4290d7df2c79SJosef Bacik spin_unlock(&head->lock); 4291acce952bSliubo spin_unlock(&delayed_refs->lock); 4292e78417d1SJosef Bacik mutex_unlock(&head->mutex); 4293acce952bSliubo 4294d7df2c79SJosef Bacik if (pin_bytes) 4295d278850eSJosef Bacik btrfs_pin_extent(fs_info, head->bytenr, 4296d278850eSJosef Bacik head->num_bytes, 1); 429731890da0SJosef Bacik btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); 4298d278850eSJosef Bacik btrfs_put_delayed_ref_head(head); 4299acce952bSliubo cond_resched(); 4300acce952bSliubo spin_lock(&delayed_refs->lock); 4301acce952bSliubo } 4302acce952bSliubo 4303acce952bSliubo spin_unlock(&delayed_refs->lock); 4304acce952bSliubo 4305acce952bSliubo return ret; 4306acce952bSliubo } 4307acce952bSliubo 4308143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 4309acce952bSliubo { 4310acce952bSliubo struct btrfs_inode *btrfs_inode; 4311acce952bSliubo struct list_head splice; 4312acce952bSliubo 4313acce952bSliubo INIT_LIST_HEAD(&splice); 4314acce952bSliubo 4315eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4316eb73c1b7SMiao Xie list_splice_init(&root->delalloc_inodes, &splice); 4317acce952bSliubo 4318acce952bSliubo while (!list_empty(&splice)) { 4319fe816d0fSNikolay Borisov struct inode *inode = NULL; 4320eb73c1b7SMiao Xie btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 4321acce952bSliubo delalloc_inodes); 4322fe816d0fSNikolay Borisov __btrfs_del_delalloc_inode(root, btrfs_inode); 4323eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4324acce952bSliubo 4325fe816d0fSNikolay Borisov /* 4326fe816d0fSNikolay Borisov * Make sure we get a live inode and that it'll not disappear 4327fe816d0fSNikolay Borisov * meanwhile. 4328fe816d0fSNikolay Borisov */ 4329fe816d0fSNikolay Borisov inode = igrab(&btrfs_inode->vfs_inode); 4330fe816d0fSNikolay Borisov if (inode) { 4331fe816d0fSNikolay Borisov invalidate_inode_pages2(inode->i_mapping); 4332fe816d0fSNikolay Borisov iput(inode); 4333fe816d0fSNikolay Borisov } 4334eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4335acce952bSliubo } 4336eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4337eb73c1b7SMiao Xie } 4338eb73c1b7SMiao Xie 4339eb73c1b7SMiao Xie static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 4340eb73c1b7SMiao Xie { 4341eb73c1b7SMiao Xie struct btrfs_root *root; 4342eb73c1b7SMiao Xie struct list_head splice; 4343eb73c1b7SMiao Xie 4344eb73c1b7SMiao Xie INIT_LIST_HEAD(&splice); 4345eb73c1b7SMiao Xie 4346eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4347eb73c1b7SMiao Xie list_splice_init(&fs_info->delalloc_roots, &splice); 4348eb73c1b7SMiao Xie while (!list_empty(&splice)) { 4349eb73c1b7SMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4350eb73c1b7SMiao Xie delalloc_root); 4351eb73c1b7SMiao Xie root = btrfs_grab_fs_root(root); 4352eb73c1b7SMiao Xie BUG_ON(!root); 4353eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4354eb73c1b7SMiao Xie 4355eb73c1b7SMiao Xie btrfs_destroy_delalloc_inodes(root); 4356eb73c1b7SMiao Xie btrfs_put_fs_root(root); 4357eb73c1b7SMiao Xie 4358eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4359eb73c1b7SMiao Xie } 4360eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4361acce952bSliubo } 4362acce952bSliubo 43632ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 4364acce952bSliubo struct extent_io_tree *dirty_pages, 4365acce952bSliubo int mark) 4366acce952bSliubo { 4367acce952bSliubo int ret; 4368acce952bSliubo struct extent_buffer *eb; 4369acce952bSliubo u64 start = 0; 4370acce952bSliubo u64 end; 4371acce952bSliubo 4372acce952bSliubo while (1) { 4373acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 4374e6138876SJosef Bacik mark, NULL); 4375acce952bSliubo if (ret) 4376acce952bSliubo break; 4377acce952bSliubo 437891166212SDavid Sterba clear_extent_bits(dirty_pages, start, end, mark); 4379acce952bSliubo while (start <= end) { 43800b246afaSJeff Mahoney eb = find_extent_buffer(fs_info, start); 43810b246afaSJeff Mahoney start += fs_info->nodesize; 4382fd8b2b61SJosef Bacik if (!eb) 4383acce952bSliubo continue; 4384fd8b2b61SJosef Bacik wait_on_extent_buffer_writeback(eb); 4385acce952bSliubo 4386fd8b2b61SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 4387fd8b2b61SJosef Bacik &eb->bflags)) 4388fd8b2b61SJosef Bacik clear_extent_buffer_dirty(eb); 4389fd8b2b61SJosef Bacik free_extent_buffer_stale(eb); 4390acce952bSliubo } 4391acce952bSliubo } 4392acce952bSliubo 4393acce952bSliubo return ret; 4394acce952bSliubo } 4395acce952bSliubo 43962ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 4397acce952bSliubo struct extent_io_tree *pinned_extents) 4398acce952bSliubo { 4399acce952bSliubo struct extent_io_tree *unpin; 4400acce952bSliubo u64 start; 4401acce952bSliubo u64 end; 4402acce952bSliubo int ret; 4403ed0eaa14SLiu Bo bool loop = true; 4404acce952bSliubo 4405acce952bSliubo unpin = pinned_extents; 4406ed0eaa14SLiu Bo again: 4407acce952bSliubo while (1) { 44080e6ec385SFilipe Manana struct extent_state *cached_state = NULL; 44090e6ec385SFilipe Manana 4410fcd5e742SLu Fengqi /* 4411fcd5e742SLu Fengqi * The btrfs_finish_extent_commit() may get the same range as 4412fcd5e742SLu Fengqi * ours between find_first_extent_bit and clear_extent_dirty. 4413fcd5e742SLu Fengqi * Hence, hold the unused_bg_unpin_mutex to avoid double unpin 4414fcd5e742SLu Fengqi * the same extent range. 4415fcd5e742SLu Fengqi */ 4416fcd5e742SLu Fengqi mutex_lock(&fs_info->unused_bg_unpin_mutex); 4417acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 44180e6ec385SFilipe Manana EXTENT_DIRTY, &cached_state); 4419fcd5e742SLu Fengqi if (ret) { 4420fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4421acce952bSliubo break; 4422fcd5e742SLu Fengqi } 4423acce952bSliubo 44240e6ec385SFilipe Manana clear_extent_dirty(unpin, start, end, &cached_state); 44250e6ec385SFilipe Manana free_extent_state(cached_state); 44262ff7e61eSJeff Mahoney btrfs_error_unpin_extent_range(fs_info, start, end); 4427fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4428acce952bSliubo cond_resched(); 4429acce952bSliubo } 4430acce952bSliubo 4431ed0eaa14SLiu Bo if (loop) { 44320b246afaSJeff Mahoney if (unpin == &fs_info->freed_extents[0]) 44330b246afaSJeff Mahoney unpin = &fs_info->freed_extents[1]; 4434ed0eaa14SLiu Bo else 44350b246afaSJeff Mahoney unpin = &fs_info->freed_extents[0]; 4436ed0eaa14SLiu Bo loop = false; 4437ed0eaa14SLiu Bo goto again; 4438ed0eaa14SLiu Bo } 4439ed0eaa14SLiu Bo 4440acce952bSliubo return 0; 4441acce952bSliubo } 4442acce952bSliubo 4443c79a1751SLiu Bo static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache) 4444c79a1751SLiu Bo { 4445c79a1751SLiu Bo struct inode *inode; 4446c79a1751SLiu Bo 4447c79a1751SLiu Bo inode = cache->io_ctl.inode; 4448c79a1751SLiu Bo if (inode) { 4449c79a1751SLiu Bo invalidate_inode_pages2(inode->i_mapping); 4450c79a1751SLiu Bo BTRFS_I(inode)->generation = 0; 4451c79a1751SLiu Bo cache->io_ctl.inode = NULL; 4452c79a1751SLiu Bo iput(inode); 4453c79a1751SLiu Bo } 4454c79a1751SLiu Bo btrfs_put_block_group(cache); 4455c79a1751SLiu Bo } 4456c79a1751SLiu Bo 4457c79a1751SLiu Bo void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, 44582ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4459c79a1751SLiu Bo { 4460c79a1751SLiu Bo struct btrfs_block_group_cache *cache; 4461c79a1751SLiu Bo 4462c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4463c79a1751SLiu Bo while (!list_empty(&cur_trans->dirty_bgs)) { 4464c79a1751SLiu Bo cache = list_first_entry(&cur_trans->dirty_bgs, 4465c79a1751SLiu Bo struct btrfs_block_group_cache, 4466c79a1751SLiu Bo dirty_list); 4467c79a1751SLiu Bo 4468c79a1751SLiu Bo if (!list_empty(&cache->io_list)) { 4469c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4470c79a1751SLiu Bo list_del_init(&cache->io_list); 4471c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 4472c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4473c79a1751SLiu Bo } 4474c79a1751SLiu Bo 4475c79a1751SLiu Bo list_del_init(&cache->dirty_list); 4476c79a1751SLiu Bo spin_lock(&cache->lock); 4477c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 4478c79a1751SLiu Bo spin_unlock(&cache->lock); 4479c79a1751SLiu Bo 4480c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4481c79a1751SLiu Bo btrfs_put_block_group(cache); 4482ba2c4d4eSJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 4483c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4484c79a1751SLiu Bo } 4485c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4486c79a1751SLiu Bo 448745ae2c18SNikolay Borisov /* 448845ae2c18SNikolay Borisov * Refer to the definition of io_bgs member for details why it's safe 448945ae2c18SNikolay Borisov * to use it without any locking 449045ae2c18SNikolay Borisov */ 4491c79a1751SLiu Bo while (!list_empty(&cur_trans->io_bgs)) { 4492c79a1751SLiu Bo cache = list_first_entry(&cur_trans->io_bgs, 4493c79a1751SLiu Bo struct btrfs_block_group_cache, 4494c79a1751SLiu Bo io_list); 4495c79a1751SLiu Bo 4496c79a1751SLiu Bo list_del_init(&cache->io_list); 4497c79a1751SLiu Bo spin_lock(&cache->lock); 4498c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 4499c79a1751SLiu Bo spin_unlock(&cache->lock); 4500c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 4501c79a1751SLiu Bo } 4502c79a1751SLiu Bo } 4503c79a1751SLiu Bo 450449b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 45052ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 450649b25e05SJeff Mahoney { 4507bbbf7243SNikolay Borisov struct btrfs_device *dev, *tmp; 4508bbbf7243SNikolay Borisov 45092ff7e61eSJeff Mahoney btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 4510c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->dirty_bgs)); 4511c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->io_bgs)); 4512c79a1751SLiu Bo 4513bbbf7243SNikolay Borisov list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list, 4514bbbf7243SNikolay Borisov post_commit_list) { 4515bbbf7243SNikolay Borisov list_del_init(&dev->post_commit_list); 4516bbbf7243SNikolay Borisov } 4517bbbf7243SNikolay Borisov 45182ff7e61eSJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, fs_info); 451949b25e05SJeff Mahoney 45204a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 45210b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 452249b25e05SJeff Mahoney 45234a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 45240b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 452549b25e05SJeff Mahoney 4526ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 4527ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 452867cde344SMiao Xie 45292ff7e61eSJeff Mahoney btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, 453049b25e05SJeff Mahoney EXTENT_DIRTY); 45312ff7e61eSJeff Mahoney btrfs_destroy_pinned_extent(fs_info, 45320b246afaSJeff Mahoney fs_info->pinned_extents); 453349b25e05SJeff Mahoney 45344a9d8bdeSMiao Xie cur_trans->state =TRANS_STATE_COMPLETED; 45354a9d8bdeSMiao Xie wake_up(&cur_trans->commit_wait); 453649b25e05SJeff Mahoney } 453749b25e05SJeff Mahoney 45382ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) 4539acce952bSliubo { 4540acce952bSliubo struct btrfs_transaction *t; 4541acce952bSliubo 45420b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 4543acce952bSliubo 45440b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 45450b246afaSJeff Mahoney while (!list_empty(&fs_info->trans_list)) { 45460b246afaSJeff Mahoney t = list_first_entry(&fs_info->trans_list, 4547724e2315SJosef Bacik struct btrfs_transaction, list); 4548724e2315SJosef Bacik if (t->state >= TRANS_STATE_COMMIT_START) { 45499b64f57dSElena Reshetova refcount_inc(&t->use_count); 45500b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 45512ff7e61eSJeff Mahoney btrfs_wait_for_commit(fs_info, t->transid); 4552724e2315SJosef Bacik btrfs_put_transaction(t); 45530b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4554724e2315SJosef Bacik continue; 4555724e2315SJosef Bacik } 45560b246afaSJeff Mahoney if (t == fs_info->running_transaction) { 4557724e2315SJosef Bacik t->state = TRANS_STATE_COMMIT_DOING; 45580b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4559724e2315SJosef Bacik /* 4560724e2315SJosef Bacik * We wait for 0 num_writers since we don't hold a trans 4561724e2315SJosef Bacik * handle open currently for this transaction. 4562724e2315SJosef Bacik */ 4563724e2315SJosef Bacik wait_event(t->writer_wait, 4564724e2315SJosef Bacik atomic_read(&t->num_writers) == 0); 4565724e2315SJosef Bacik } else { 45660b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4567724e2315SJosef Bacik } 45682ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(t, fs_info); 4569724e2315SJosef Bacik 45700b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 45710b246afaSJeff Mahoney if (t == fs_info->running_transaction) 45720b246afaSJeff Mahoney fs_info->running_transaction = NULL; 4573724e2315SJosef Bacik list_del_init(&t->list); 45740b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4575a4abeea4SJosef Bacik 4576724e2315SJosef Bacik btrfs_put_transaction(t); 45772ff7e61eSJeff Mahoney trace_btrfs_transaction_commit(fs_info->tree_root); 45780b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4579724e2315SJosef Bacik } 45800b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 45810b246afaSJeff Mahoney btrfs_destroy_all_ordered_extents(fs_info); 4582ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 4583ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 45842ff7e61eSJeff Mahoney btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents); 45850b246afaSJeff Mahoney btrfs_destroy_all_delalloc_inodes(fs_info); 45860b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 4587acce952bSliubo 4588acce952bSliubo return 0; 4589acce952bSliubo } 4590acce952bSliubo 4591e8c9f186SDavid Sterba static const struct extent_io_ops btree_extent_io_ops = { 45924d53dddbSDavid Sterba /* mandatory callbacks */ 45930b86a832SChris Mason .submit_bio_hook = btree_submit_bio_hook, 45944d53dddbSDavid Sterba .readpage_end_io_hook = btree_readpage_end_io_hook, 45950da5468fSChris Mason }; 4596