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> 10ce9adaa5SChris Mason #include <linux/workqueue.h> 11a74a4b97SChris Mason #include <linux/kthread.h> 125a0e3ad6STejun Heo #include <linux/slab.h> 13784b4e29SChris Mason #include <linux/migrate.h> 147a36ddecSDavid Sterba #include <linux/ratelimit.h> 156463fe58SStefan Behrens #include <linux/uuid.h> 16803b2f54SStefan Behrens #include <linux/semaphore.h> 17540adea3SMasami Hiramatsu #include <linux/error-injection.h> 189678c543SNikolay Borisov #include <linux/crc32c.h> 19b89f6d1fSFilipe Manana #include <linux/sched/mm.h> 207e75bf3fSDavid Sterba #include <asm/unaligned.h> 216d97c6e3SJohannes Thumshirn #include <crypto/hash.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" 42aac0023cSJosef Bacik #include "block-group.h" 43b0643e59SDennis Zhou #include "discard.h" 44f603bb94SNikolay Borisov #include "space-info.h" 45eb60ceacSChris Mason 46319e4d06SQu Wenruo #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\ 47319e4d06SQu Wenruo BTRFS_HEADER_FLAG_RELOC |\ 48319e4d06SQu Wenruo BTRFS_SUPER_FLAG_ERROR |\ 49319e4d06SQu Wenruo BTRFS_SUPER_FLAG_SEEDING |\ 50e2731e55SAnand Jain BTRFS_SUPER_FLAG_METADUMP |\ 51e2731e55SAnand Jain BTRFS_SUPER_FLAG_METADUMP_V2) 52319e4d06SQu Wenruo 538b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work); 54143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root); 55acce952bSliubo static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 562ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info); 57143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root); 582ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 59acce952bSliubo struct extent_io_tree *dirty_pages, 60acce952bSliubo int mark); 612ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 62acce952bSliubo struct extent_io_tree *pinned_extents); 632ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info); 642ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info); 65ce9adaa5SChris Mason 66d352ac68SChris Mason /* 6797eb6b69SDavid Sterba * btrfs_end_io_wq structs are used to do processing in task context when an IO 6897eb6b69SDavid Sterba * is complete. This is used during reads to verify checksums, and it is used 69d352ac68SChris Mason * by writes to insert metadata for new file extents after IO is complete. 70d352ac68SChris Mason */ 7197eb6b69SDavid Sterba struct btrfs_end_io_wq { 72ce9adaa5SChris Mason struct bio *bio; 73ce9adaa5SChris Mason bio_end_io_t *end_io; 74ce9adaa5SChris Mason void *private; 75ce9adaa5SChris Mason struct btrfs_fs_info *info; 764e4cbee9SChristoph Hellwig blk_status_t status; 77bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata; 788b712842SChris Mason struct btrfs_work work; 79ce9adaa5SChris Mason }; 800da5468fSChris Mason 8197eb6b69SDavid Sterba static struct kmem_cache *btrfs_end_io_wq_cache; 8297eb6b69SDavid Sterba 8397eb6b69SDavid Sterba int __init btrfs_end_io_wq_init(void) 8497eb6b69SDavid Sterba { 8597eb6b69SDavid Sterba btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq", 8697eb6b69SDavid Sterba sizeof(struct btrfs_end_io_wq), 8797eb6b69SDavid Sterba 0, 88fba4b697SNikolay Borisov SLAB_MEM_SPREAD, 8997eb6b69SDavid Sterba NULL); 9097eb6b69SDavid Sterba if (!btrfs_end_io_wq_cache) 9197eb6b69SDavid Sterba return -ENOMEM; 9297eb6b69SDavid Sterba return 0; 9397eb6b69SDavid Sterba } 9497eb6b69SDavid Sterba 95e67c718bSDavid Sterba void __cold btrfs_end_io_wq_exit(void) 9697eb6b69SDavid Sterba { 9797eb6b69SDavid Sterba kmem_cache_destroy(btrfs_end_io_wq_cache); 9897eb6b69SDavid Sterba } 9997eb6b69SDavid Sterba 100141386e1SJosef Bacik static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info) 101141386e1SJosef Bacik { 102141386e1SJosef Bacik if (fs_info->csum_shash) 103141386e1SJosef Bacik crypto_free_shash(fs_info->csum_shash); 104141386e1SJosef Bacik } 105141386e1SJosef Bacik 106d352ac68SChris Mason /* 107d352ac68SChris Mason * async submit bios are used to offload expensive checksumming 108d352ac68SChris Mason * onto the worker threads. They checksum file and metadata bios 109d352ac68SChris Mason * just before they are sent down the IO stack. 110d352ac68SChris Mason */ 11144b8bd7eSChris Mason struct async_submit_bio { 1128896a08dSQu Wenruo struct inode *inode; 11344b8bd7eSChris Mason struct bio *bio; 114a758781dSDavid Sterba extent_submit_bio_start_t *submit_bio_start; 11544b8bd7eSChris Mason int mirror_num; 116eaf25d93SChris Mason /* 117eaf25d93SChris Mason * bio_offset is optional, can be used if the pages in the bio 118eaf25d93SChris Mason * can't tell us where in the file the bio should go 119eaf25d93SChris Mason */ 120eaf25d93SChris Mason u64 bio_offset; 1218b712842SChris Mason struct btrfs_work work; 1224e4cbee9SChristoph Hellwig blk_status_t status; 12344b8bd7eSChris Mason }; 12444b8bd7eSChris Mason 12585d4e461SChris Mason /* 12685d4e461SChris Mason * Lockdep class keys for extent_buffer->lock's in this root. For a given 12785d4e461SChris Mason * eb, the lockdep key is determined by the btrfs_root it belongs to and 12885d4e461SChris Mason * the level the eb occupies in the tree. 1294008c04aSChris Mason * 13085d4e461SChris Mason * Different roots are used for different purposes and may nest inside each 13185d4e461SChris Mason * other and they require separate keysets. As lockdep keys should be 13285d4e461SChris Mason * static, assign keysets according to the purpose of the root as indicated 1334fd786e6SMisono Tomohiro * by btrfs_root->root_key.objectid. This ensures that all special purpose 1344fd786e6SMisono Tomohiro * roots have separate keysets. 1354008c04aSChris Mason * 13685d4e461SChris Mason * Lock-nesting across peer nodes is always done with the immediate parent 13785d4e461SChris Mason * node locked thus preventing deadlock. As lockdep doesn't know this, use 13885d4e461SChris Mason * subclass to avoid triggering lockdep warning in such cases. 1394008c04aSChris Mason * 14085d4e461SChris Mason * The key is set by the readpage_end_io_hook after the buffer has passed 14185d4e461SChris Mason * csum validation but before the pages are unlocked. It is also set by 14285d4e461SChris Mason * btrfs_init_new_buffer on freshly allocated blocks. 14385d4e461SChris Mason * 14485d4e461SChris Mason * We also add a check to make sure the highest level of the tree is the 14585d4e461SChris Mason * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code 14685d4e461SChris Mason * needs update as well. 1474008c04aSChris Mason */ 1484008c04aSChris Mason #ifdef CONFIG_DEBUG_LOCK_ALLOC 1494008c04aSChris Mason # if BTRFS_MAX_LEVEL != 8 1504008c04aSChris Mason # error 1514008c04aSChris Mason # endif 15285d4e461SChris Mason 153ab1405aaSDavid Sterba #define DEFINE_LEVEL(stem, level) \ 154ab1405aaSDavid Sterba .names[level] = "btrfs-" stem "-0" #level, 155ab1405aaSDavid Sterba 156ab1405aaSDavid Sterba #define DEFINE_NAME(stem) \ 157ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 0) \ 158ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 1) \ 159ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 2) \ 160ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 3) \ 161ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 4) \ 162ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 5) \ 163ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 6) \ 164ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 7) 165ab1405aaSDavid Sterba 16685d4e461SChris Mason static struct btrfs_lockdep_keyset { 16785d4e461SChris Mason u64 id; /* root objectid */ 168ab1405aaSDavid Sterba /* Longest entry: btrfs-free-space-00 */ 169387824afSDavid Sterba char names[BTRFS_MAX_LEVEL][20]; 170387824afSDavid Sterba struct lock_class_key keys[BTRFS_MAX_LEVEL]; 17185d4e461SChris Mason } btrfs_lockdep_keysets[] = { 172ab1405aaSDavid Sterba { .id = BTRFS_ROOT_TREE_OBJECTID, DEFINE_NAME("root") }, 173ab1405aaSDavid Sterba { .id = BTRFS_EXTENT_TREE_OBJECTID, DEFINE_NAME("extent") }, 174ab1405aaSDavid Sterba { .id = BTRFS_CHUNK_TREE_OBJECTID, DEFINE_NAME("chunk") }, 175ab1405aaSDavid Sterba { .id = BTRFS_DEV_TREE_OBJECTID, DEFINE_NAME("dev") }, 176ab1405aaSDavid Sterba { .id = BTRFS_FS_TREE_OBJECTID, DEFINE_NAME("fs") }, 177ab1405aaSDavid Sterba { .id = BTRFS_CSUM_TREE_OBJECTID, DEFINE_NAME("csum") }, 178ab1405aaSDavid Sterba { .id = BTRFS_QUOTA_TREE_OBJECTID, DEFINE_NAME("quota") }, 179ab1405aaSDavid Sterba { .id = BTRFS_TREE_LOG_OBJECTID, DEFINE_NAME("log") }, 180ab1405aaSDavid Sterba { .id = BTRFS_TREE_RELOC_OBJECTID, DEFINE_NAME("treloc") }, 181ab1405aaSDavid Sterba { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, DEFINE_NAME("dreloc") }, 182ab1405aaSDavid Sterba { .id = BTRFS_UUID_TREE_OBJECTID, DEFINE_NAME("uuid") }, 183ab1405aaSDavid Sterba { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, DEFINE_NAME("free-space") }, 184ab1405aaSDavid Sterba { .id = 0, DEFINE_NAME("tree") }, 1854008c04aSChris Mason }; 18685d4e461SChris Mason 187ab1405aaSDavid Sterba #undef DEFINE_LEVEL 188ab1405aaSDavid Sterba #undef DEFINE_NAME 18985d4e461SChris Mason 19085d4e461SChris Mason void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb, 19185d4e461SChris Mason int level) 19285d4e461SChris Mason { 19385d4e461SChris Mason struct btrfs_lockdep_keyset *ks; 19485d4e461SChris Mason 19585d4e461SChris Mason BUG_ON(level >= ARRAY_SIZE(ks->keys)); 19685d4e461SChris Mason 19785d4e461SChris Mason /* find the matching keyset, id 0 is the default entry */ 19885d4e461SChris Mason for (ks = btrfs_lockdep_keysets; ks->id; ks++) 19985d4e461SChris Mason if (ks->id == objectid) 20085d4e461SChris Mason break; 20185d4e461SChris Mason 20285d4e461SChris Mason lockdep_set_class_and_name(&eb->lock, 20385d4e461SChris Mason &ks->keys[level], ks->names[level]); 20485d4e461SChris Mason } 20585d4e461SChris Mason 2064008c04aSChris Mason #endif 2074008c04aSChris Mason 208d352ac68SChris Mason /* 2092996e1f8SJohannes Thumshirn * Compute the csum of a btree block and store the result to provided buffer. 210d352ac68SChris Mason */ 211c67b3892SDavid Sterba static void csum_tree_block(struct extent_buffer *buf, u8 *result) 21219c00ddcSChris Mason { 213d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = buf->fs_info; 214e9be5a30SDavid Sterba const int num_pages = fs_info->nodesize >> PAGE_SHIFT; 215d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 21619c00ddcSChris Mason char *kaddr; 217e9be5a30SDavid Sterba int i; 218d5178578SJohannes Thumshirn 219d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 220d5178578SJohannes Thumshirn crypto_shash_init(shash); 221e9be5a30SDavid Sterba kaddr = page_address(buf->pages[0]); 222e9be5a30SDavid Sterba crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE, 223e9be5a30SDavid Sterba PAGE_SIZE - BTRFS_CSUM_SIZE); 22419c00ddcSChris Mason 225e9be5a30SDavid Sterba for (i = 1; i < num_pages; i++) { 226e9be5a30SDavid Sterba kaddr = page_address(buf->pages[i]); 227e9be5a30SDavid Sterba crypto_shash_update(shash, kaddr, PAGE_SIZE); 22819c00ddcSChris Mason } 22971a63551SDavid Sterba memset(result, 0, BTRFS_CSUM_SIZE); 230d5178578SJohannes Thumshirn crypto_shash_final(shash, result); 23119c00ddcSChris Mason } 23219c00ddcSChris Mason 233d352ac68SChris Mason /* 234d352ac68SChris Mason * we can't consider a given block up to date unless the transid of the 235d352ac68SChris Mason * block matches the transid in the parent node's pointer. This is how we 236d352ac68SChris Mason * detect blocks that either didn't get written at all or got written 237d352ac68SChris Mason * in the wrong place. 238d352ac68SChris Mason */ 2391259ab75SChris Mason static int verify_parent_transid(struct extent_io_tree *io_tree, 240b9fab919SChris Mason struct extent_buffer *eb, u64 parent_transid, 241b9fab919SChris Mason int atomic) 2421259ab75SChris Mason { 2432ac55d41SJosef Bacik struct extent_state *cached_state = NULL; 2441259ab75SChris Mason int ret; 2452755a0deSDavid Sterba bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB); 2461259ab75SChris Mason 2471259ab75SChris Mason if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 2481259ab75SChris Mason return 0; 2491259ab75SChris Mason 250b9fab919SChris Mason if (atomic) 251b9fab919SChris Mason return -EAGAIN; 252b9fab919SChris Mason 253ac5887c8SJosef Bacik if (need_lock) 254a26e8c9fSJosef Bacik btrfs_tree_read_lock(eb); 255a26e8c9fSJosef Bacik 2562ac55d41SJosef Bacik lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 257ff13db41SDavid Sterba &cached_state); 2580b32f4bbSJosef Bacik if (extent_buffer_uptodate(eb) && 2591259ab75SChris Mason btrfs_header_generation(eb) == parent_transid) { 2601259ab75SChris Mason ret = 0; 2611259ab75SChris Mason goto out; 2621259ab75SChris Mason } 26394647322SDavid Sterba btrfs_err_rl(eb->fs_info, 26494647322SDavid Sterba "parent transid verify failed on %llu wanted %llu found %llu", 26594647322SDavid Sterba eb->start, 26629549aecSWang Shilong parent_transid, btrfs_header_generation(eb)); 2671259ab75SChris Mason ret = 1; 268a26e8c9fSJosef Bacik 269a26e8c9fSJosef Bacik /* 270a26e8c9fSJosef Bacik * Things reading via commit roots that don't have normal protection, 271a26e8c9fSJosef Bacik * like send, can have a really old block in cache that may point at a 27201327610SNicholas D Steeves * block that has been freed and re-allocated. So don't clear uptodate 273a26e8c9fSJosef Bacik * if we find an eb that is under IO (dirty/writeback) because we could 274a26e8c9fSJosef Bacik * end up reading in the stale data and then writing it back out and 275a26e8c9fSJosef Bacik * making everybody very sad. 276a26e8c9fSJosef Bacik */ 277a26e8c9fSJosef Bacik if (!extent_buffer_under_io(eb)) 2780b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 27933958dc6SChris Mason out: 2802ac55d41SJosef Bacik unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 281e43bbe5eSDavid Sterba &cached_state); 282472b909fSJosef Bacik if (need_lock) 283ac5887c8SJosef Bacik btrfs_tree_read_unlock(eb); 2841259ab75SChris Mason return ret; 2851259ab75SChris Mason } 2861259ab75SChris Mason 287e7e16f48SJohannes Thumshirn static bool btrfs_supported_super_csum(u16 csum_type) 288e7e16f48SJohannes Thumshirn { 289e7e16f48SJohannes Thumshirn switch (csum_type) { 290e7e16f48SJohannes Thumshirn case BTRFS_CSUM_TYPE_CRC32: 2913951e7f0SJohannes Thumshirn case BTRFS_CSUM_TYPE_XXHASH: 2923831bf00SJohannes Thumshirn case BTRFS_CSUM_TYPE_SHA256: 293352ae07bSDavid Sterba case BTRFS_CSUM_TYPE_BLAKE2: 294e7e16f48SJohannes Thumshirn return true; 295e7e16f48SJohannes Thumshirn default: 296e7e16f48SJohannes Thumshirn return false; 297e7e16f48SJohannes Thumshirn } 298e7e16f48SJohannes Thumshirn } 299e7e16f48SJohannes Thumshirn 300d352ac68SChris Mason /* 3011104a885SDavid Sterba * Return 0 if the superblock checksum type matches the checksum value of that 3021104a885SDavid Sterba * algorithm. Pass the raw disk superblock data. 3031104a885SDavid Sterba */ 304ab8d0fc4SJeff Mahoney static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info, 305ab8d0fc4SJeff Mahoney char *raw_disk_sb) 3061104a885SDavid Sterba { 3071104a885SDavid Sterba struct btrfs_super_block *disk_sb = 3081104a885SDavid Sterba (struct btrfs_super_block *)raw_disk_sb; 30951bce6c9SJohannes Thumshirn char result[BTRFS_CSUM_SIZE]; 310d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 311d5178578SJohannes Thumshirn 312d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 3131104a885SDavid Sterba 3141104a885SDavid Sterba /* 3151104a885SDavid Sterba * The super_block structure does not span the whole 31651bce6c9SJohannes Thumshirn * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is 31751bce6c9SJohannes Thumshirn * filled with zeros and is included in the checksum. 3181104a885SDavid Sterba */ 319fd08001fSEric Biggers crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE, 320fd08001fSEric Biggers BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result); 3211104a885SDavid Sterba 32255fc29beSDavid Sterba if (memcmp(disk_sb->csum, result, fs_info->csum_size)) 323e7e16f48SJohannes Thumshirn return 1; 3241104a885SDavid Sterba 325e7e16f48SJohannes Thumshirn return 0; 3261104a885SDavid Sterba } 3271104a885SDavid Sterba 328e064d5e9SDavid Sterba int btrfs_verify_level_key(struct extent_buffer *eb, int level, 329ff76a864SLiu Bo struct btrfs_key *first_key, u64 parent_transid) 330581c1760SQu Wenruo { 331e064d5e9SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 332581c1760SQu Wenruo int found_level; 333581c1760SQu Wenruo struct btrfs_key found_key; 334581c1760SQu Wenruo int ret; 335581c1760SQu Wenruo 336581c1760SQu Wenruo found_level = btrfs_header_level(eb); 337581c1760SQu Wenruo if (found_level != level) { 33863489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 33963489055SQu Wenruo KERN_ERR "BTRFS: tree level check failed\n"); 340581c1760SQu Wenruo btrfs_err(fs_info, 341581c1760SQu Wenruo "tree level mismatch detected, bytenr=%llu level expected=%u has=%u", 342581c1760SQu Wenruo eb->start, level, found_level); 343581c1760SQu Wenruo return -EIO; 344581c1760SQu Wenruo } 345581c1760SQu Wenruo 346581c1760SQu Wenruo if (!first_key) 347581c1760SQu Wenruo return 0; 348581c1760SQu Wenruo 3495d41be6fSQu Wenruo /* 3505d41be6fSQu Wenruo * For live tree block (new tree blocks in current transaction), 3515d41be6fSQu Wenruo * we need proper lock context to avoid race, which is impossible here. 3525d41be6fSQu Wenruo * So we only checks tree blocks which is read from disk, whose 3535d41be6fSQu Wenruo * generation <= fs_info->last_trans_committed. 3545d41be6fSQu Wenruo */ 3555d41be6fSQu Wenruo if (btrfs_header_generation(eb) > fs_info->last_trans_committed) 3565d41be6fSQu Wenruo return 0; 35762fdaa52SQu Wenruo 35862fdaa52SQu Wenruo /* We have @first_key, so this @eb must have at least one item */ 35962fdaa52SQu Wenruo if (btrfs_header_nritems(eb) == 0) { 36062fdaa52SQu Wenruo btrfs_err(fs_info, 36162fdaa52SQu Wenruo "invalid tree nritems, bytenr=%llu nritems=0 expect >0", 36262fdaa52SQu Wenruo eb->start); 36362fdaa52SQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 36462fdaa52SQu Wenruo return -EUCLEAN; 36562fdaa52SQu Wenruo } 36662fdaa52SQu Wenruo 367581c1760SQu Wenruo if (found_level) 368581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &found_key, 0); 369581c1760SQu Wenruo else 370581c1760SQu Wenruo btrfs_item_key_to_cpu(eb, &found_key, 0); 371581c1760SQu Wenruo ret = btrfs_comp_cpu_keys(first_key, &found_key); 372581c1760SQu Wenruo 373581c1760SQu Wenruo if (ret) { 37463489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 37563489055SQu Wenruo KERN_ERR "BTRFS: tree first key check failed\n"); 376581c1760SQu Wenruo btrfs_err(fs_info, 377ff76a864SLiu Bo "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", 378ff76a864SLiu Bo eb->start, parent_transid, first_key->objectid, 379ff76a864SLiu Bo first_key->type, first_key->offset, 380ff76a864SLiu Bo found_key.objectid, found_key.type, 381ff76a864SLiu Bo found_key.offset); 382581c1760SQu Wenruo } 383581c1760SQu Wenruo return ret; 384581c1760SQu Wenruo } 385581c1760SQu Wenruo 3861104a885SDavid Sterba /* 387d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 388d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 389581c1760SQu Wenruo * 390581c1760SQu Wenruo * @parent_transid: expected transid, skip check if 0 391581c1760SQu Wenruo * @level: expected level, mandatory check 392581c1760SQu Wenruo * @first_key: expected key of first slot, skip check if NULL 393d352ac68SChris Mason */ 3945ab12d1fSDavid Sterba static int btree_read_extent_buffer_pages(struct extent_buffer *eb, 395581c1760SQu Wenruo u64 parent_transid, int level, 396581c1760SQu Wenruo struct btrfs_key *first_key) 397f188591eSChris Mason { 3985ab12d1fSDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 399f188591eSChris Mason struct extent_io_tree *io_tree; 400ea466794SJosef Bacik int failed = 0; 401f188591eSChris Mason int ret; 402f188591eSChris Mason int num_copies = 0; 403f188591eSChris Mason int mirror_num = 0; 404ea466794SJosef Bacik int failed_mirror = 0; 405f188591eSChris Mason 4060b246afaSJeff Mahoney io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; 407f188591eSChris Mason while (1) { 408f8397d69SNikolay Borisov clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 409c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num); 410256dd1bbSStefan Behrens if (!ret) { 411581c1760SQu Wenruo if (verify_parent_transid(io_tree, eb, 412b9fab919SChris Mason parent_transid, 0)) 413256dd1bbSStefan Behrens ret = -EIO; 414e064d5e9SDavid Sterba else if (btrfs_verify_level_key(eb, level, 415ff76a864SLiu Bo first_key, parent_transid)) 416581c1760SQu Wenruo ret = -EUCLEAN; 417581c1760SQu Wenruo else 418581c1760SQu Wenruo break; 419256dd1bbSStefan Behrens } 420d397712bSChris Mason 4210b246afaSJeff Mahoney num_copies = btrfs_num_copies(fs_info, 422f188591eSChris Mason eb->start, eb->len); 4234235298eSChris Mason if (num_copies == 1) 424ea466794SJosef Bacik break; 4254235298eSChris Mason 4265cf1ab56SJosef Bacik if (!failed_mirror) { 4275cf1ab56SJosef Bacik failed = 1; 4285cf1ab56SJosef Bacik failed_mirror = eb->read_mirror; 4295cf1ab56SJosef Bacik } 4305cf1ab56SJosef Bacik 431f188591eSChris Mason mirror_num++; 432ea466794SJosef Bacik if (mirror_num == failed_mirror) 433ea466794SJosef Bacik mirror_num++; 434ea466794SJosef Bacik 4354235298eSChris Mason if (mirror_num > num_copies) 436ea466794SJosef Bacik break; 437f188591eSChris Mason } 438ea466794SJosef Bacik 439c0901581SStefan Behrens if (failed && !ret && failed_mirror) 44020a1fbf9SDavid Sterba btrfs_repair_eb_io_failure(eb, failed_mirror); 441ea466794SJosef Bacik 442ea466794SJosef Bacik return ret; 443f188591eSChris Mason } 44419c00ddcSChris Mason 445d352ac68SChris Mason /* 446d397712bSChris Mason * checksum a dirty tree block before IO. This has extra checks to make sure 447d397712bSChris Mason * we only fill in the checksum field in the first page of a multi-page block 448d352ac68SChris Mason */ 449d397712bSChris Mason 45001d58472SDaniel Dressler static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page) 45119c00ddcSChris Mason { 4524eee4fa4SMiao Xie u64 start = page_offset(page); 45319c00ddcSChris Mason u64 found_start; 4542996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 45519c00ddcSChris Mason struct extent_buffer *eb; 4568d47a0d8SQu Wenruo int ret; 457f188591eSChris Mason 4584f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 4594f2de97aSJosef Bacik if (page != eb->pages[0]) 4604f2de97aSJosef Bacik return 0; 4610f805531SAlex Lyakas 46219c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 4630f805531SAlex Lyakas /* 4640f805531SAlex Lyakas * Please do not consolidate these warnings into a single if. 4650f805531SAlex Lyakas * It is useful to know what went wrong. 4660f805531SAlex Lyakas */ 4670f805531SAlex Lyakas if (WARN_ON(found_start != start)) 4680f805531SAlex Lyakas return -EUCLEAN; 4690f805531SAlex Lyakas if (WARN_ON(!PageUptodate(page))) 4700f805531SAlex Lyakas return -EUCLEAN; 4710f805531SAlex Lyakas 472de37aa51SNikolay Borisov ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid, 4739a8658e3SDavid Sterba offsetof(struct btrfs_header, fsid), 4749a8658e3SDavid Sterba BTRFS_FSID_SIZE) == 0); 4750f805531SAlex Lyakas 476c67b3892SDavid Sterba csum_tree_block(eb, result); 4772996e1f8SJohannes Thumshirn 4788d47a0d8SQu Wenruo if (btrfs_header_level(eb)) 4798d47a0d8SQu Wenruo ret = btrfs_check_node(eb); 4808d47a0d8SQu Wenruo else 4818d47a0d8SQu Wenruo ret = btrfs_check_leaf_full(eb); 4828d47a0d8SQu Wenruo 4838d47a0d8SQu Wenruo if (ret < 0) { 484c06631b0SQu Wenruo btrfs_print_tree(eb, 0); 4858d47a0d8SQu Wenruo btrfs_err(fs_info, 4868d47a0d8SQu Wenruo "block=%llu write time tree block corruption detected", 4878d47a0d8SQu Wenruo eb->start); 488c06631b0SQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 4898d47a0d8SQu Wenruo return ret; 4908d47a0d8SQu Wenruo } 491713cebfbSDavid Sterba write_extent_buffer(eb, result, 0, fs_info->csum_size); 4928d47a0d8SQu Wenruo 4932996e1f8SJohannes Thumshirn return 0; 49419c00ddcSChris Mason } 49519c00ddcSChris Mason 496b0c9b3b0SDavid Sterba static int check_tree_block_fsid(struct extent_buffer *eb) 4972b82032cSYan Zheng { 498b0c9b3b0SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 499944d3f9fSNikolay Borisov struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 50044880fdcSAnand Jain u8 fsid[BTRFS_FSID_SIZE]; 501944d3f9fSNikolay Borisov u8 *metadata_uuid; 5022b82032cSYan Zheng 5039a8658e3SDavid Sterba read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid), 5049a8658e3SDavid Sterba BTRFS_FSID_SIZE); 5057239ff4bSNikolay Borisov /* 506944d3f9fSNikolay Borisov * Checking the incompat flag is only valid for the current fs. For 507944d3f9fSNikolay Borisov * seed devices it's forbidden to have their uuid changed so reading 508944d3f9fSNikolay Borisov * ->fsid in this case is fine 5097239ff4bSNikolay Borisov */ 510944d3f9fSNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID)) 5117239ff4bSNikolay Borisov metadata_uuid = fs_devices->metadata_uuid; 5127239ff4bSNikolay Borisov else 5137239ff4bSNikolay Borisov metadata_uuid = fs_devices->fsid; 5147239ff4bSNikolay Borisov 515944d3f9fSNikolay Borisov if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) 516944d3f9fSNikolay Borisov return 0; 517944d3f9fSNikolay Borisov 518944d3f9fSNikolay Borisov list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) 519944d3f9fSNikolay Borisov if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE)) 520944d3f9fSNikolay Borisov return 0; 521944d3f9fSNikolay Borisov 522944d3f9fSNikolay Borisov return 1; 5232b82032cSYan Zheng } 5242b82032cSYan Zheng 5259a446d6aSNikolay Borisov int btrfs_validate_metadata_buffer(struct btrfs_io_bio *io_bio, u64 phy_offset, 5269a446d6aSNikolay Borisov struct page *page, u64 start, u64 end, 5279a446d6aSNikolay Borisov int mirror) 528ce9adaa5SChris Mason { 529ce9adaa5SChris Mason u64 found_start; 530ce9adaa5SChris Mason int found_level; 531ce9adaa5SChris Mason struct extent_buffer *eb; 53215b6e8a8SDavid Sterba struct btrfs_fs_info *fs_info; 53315b6e8a8SDavid Sterba u16 csum_size; 534f188591eSChris Mason int ret = 0; 5352996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 536727011e0SChris Mason int reads_done; 537ce9adaa5SChris Mason 538ce9adaa5SChris Mason if (!page->private) 539ce9adaa5SChris Mason goto out; 540d397712bSChris Mason 5414f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 54215b6e8a8SDavid Sterba fs_info = eb->fs_info; 54355fc29beSDavid Sterba csum_size = fs_info->csum_size; 544d397712bSChris Mason 5450b32f4bbSJosef Bacik /* the pending IO might have been the only thing that kept this buffer 5460b32f4bbSJosef Bacik * in memory. Make sure we have a ref for all this other checks 5470b32f4bbSJosef Bacik */ 54867439dadSDavid Sterba atomic_inc(&eb->refs); 5490b32f4bbSJosef Bacik 5500b32f4bbSJosef Bacik reads_done = atomic_dec_and_test(&eb->io_pages); 551727011e0SChris Mason if (!reads_done) 552727011e0SChris Mason goto err; 553f188591eSChris Mason 5545cf1ab56SJosef Bacik eb->read_mirror = mirror; 555656f30dbSFilipe Manana if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 556ea466794SJosef Bacik ret = -EIO; 557ea466794SJosef Bacik goto err; 558ea466794SJosef Bacik } 559ea466794SJosef Bacik 560ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 561727011e0SChris Mason if (found_start != eb->start) { 562893bf4b1SSu Yue btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu", 563893bf4b1SSu Yue eb->start, found_start); 564f188591eSChris Mason ret = -EIO; 565ce9adaa5SChris Mason goto err; 566ce9adaa5SChris Mason } 567b0c9b3b0SDavid Sterba if (check_tree_block_fsid(eb)) { 56802873e43SZhao Lei btrfs_err_rl(fs_info, "bad fsid on block %llu", 56994647322SDavid Sterba eb->start); 5701259ab75SChris Mason ret = -EIO; 5711259ab75SChris Mason goto err; 5721259ab75SChris Mason } 573ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 5741c24c3ceSJosef Bacik if (found_level >= BTRFS_MAX_LEVEL) { 575893bf4b1SSu Yue btrfs_err(fs_info, "bad tree block level %d on %llu", 576893bf4b1SSu Yue (int)btrfs_header_level(eb), eb->start); 5771c24c3ceSJosef Bacik ret = -EIO; 5781c24c3ceSJosef Bacik goto err; 5791c24c3ceSJosef Bacik } 580ce9adaa5SChris Mason 58185d4e461SChris Mason btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), 58285d4e461SChris Mason eb, found_level); 5834008c04aSChris Mason 584c67b3892SDavid Sterba csum_tree_block(eb, result); 585a826d6dcSJosef Bacik 5862996e1f8SJohannes Thumshirn if (memcmp_extent_buffer(eb, result, 0, csum_size)) { 58735be8851SJohannes Thumshirn u8 val[BTRFS_CSUM_SIZE] = { 0 }; 5882996e1f8SJohannes Thumshirn 5892996e1f8SJohannes Thumshirn read_extent_buffer(eb, &val, 0, csum_size); 5902996e1f8SJohannes Thumshirn btrfs_warn_rl(fs_info, 59135be8851SJohannes Thumshirn "%s checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d", 5922996e1f8SJohannes Thumshirn fs_info->sb->s_id, eb->start, 59335be8851SJohannes Thumshirn CSUM_FMT_VALUE(csum_size, val), 59435be8851SJohannes Thumshirn CSUM_FMT_VALUE(csum_size, result), 59535be8851SJohannes Thumshirn btrfs_header_level(eb)); 5962996e1f8SJohannes Thumshirn ret = -EUCLEAN; 5972996e1f8SJohannes Thumshirn goto err; 5982996e1f8SJohannes Thumshirn } 5992996e1f8SJohannes Thumshirn 600a826d6dcSJosef Bacik /* 601a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 602a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 603a826d6dcSJosef Bacik * return -EIO. 604a826d6dcSJosef Bacik */ 6051c4360eeSDavid Sterba if (found_level == 0 && btrfs_check_leaf_full(eb)) { 606a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 607a826d6dcSJosef Bacik ret = -EIO; 608a826d6dcSJosef Bacik } 609ce9adaa5SChris Mason 610813fd1dcSDavid Sterba if (found_level > 0 && btrfs_check_node(eb)) 611053ab70fSLiu Bo ret = -EIO; 612053ab70fSLiu Bo 6130b32f4bbSJosef Bacik if (!ret) 6140b32f4bbSJosef Bacik set_extent_buffer_uptodate(eb); 61575391f0dSQu Wenruo else 61675391f0dSQu Wenruo btrfs_err(fs_info, 61775391f0dSQu Wenruo "block=%llu read time tree block corruption detected", 61875391f0dSQu Wenruo eb->start); 619ce9adaa5SChris Mason err: 62079fb65a1SJosef Bacik if (reads_done && 62179fb65a1SJosef Bacik test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 622d48d71aaSDavid Sterba btree_readahead_hook(eb, ret); 6234bb31e92SArne Jansen 62453b381b3SDavid Woodhouse if (ret) { 62553b381b3SDavid Woodhouse /* 62653b381b3SDavid Woodhouse * our io error hook is going to dec the io pages 62753b381b3SDavid Woodhouse * again, we have to make sure it has something 62853b381b3SDavid Woodhouse * to decrement 62953b381b3SDavid Woodhouse */ 63053b381b3SDavid Woodhouse atomic_inc(&eb->io_pages); 6310b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 63253b381b3SDavid Woodhouse } 6330b32f4bbSJosef Bacik free_extent_buffer(eb); 634ce9adaa5SChris Mason out: 635f188591eSChris Mason return ret; 636ce9adaa5SChris Mason } 637ce9adaa5SChris Mason 6384246a0b6SChristoph Hellwig static void end_workqueue_bio(struct bio *bio) 639ce9adaa5SChris Mason { 64097eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq = bio->bi_private; 641ce9adaa5SChris Mason struct btrfs_fs_info *fs_info; 6429e0af237SLiu Bo struct btrfs_workqueue *wq; 643ce9adaa5SChris Mason 644ce9adaa5SChris Mason fs_info = end_io_wq->info; 6454e4cbee9SChristoph Hellwig end_io_wq->status = bio->bi_status; 646d20f7043SChris Mason 64737226b21SMike Christie if (bio_op(bio) == REQ_OP_WRITE) { 648a0cac0ecSOmar Sandoval if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) 6499e0af237SLiu Bo wq = fs_info->endio_meta_write_workers; 650a0cac0ecSOmar Sandoval else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) 6519e0af237SLiu Bo wq = fs_info->endio_freespace_worker; 652a0cac0ecSOmar Sandoval else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 6539e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 654a0cac0ecSOmar Sandoval else 6559e0af237SLiu Bo wq = fs_info->endio_write_workers; 6569e0af237SLiu Bo } else { 6575c047a69SOmar Sandoval if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 6589e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 659a0cac0ecSOmar Sandoval else if (end_io_wq->metadata) 6609e0af237SLiu Bo wq = fs_info->endio_meta_workers; 661a0cac0ecSOmar Sandoval else 6629e0af237SLiu Bo wq = fs_info->endio_workers; 6639e0af237SLiu Bo } 6649e0af237SLiu Bo 665a0cac0ecSOmar Sandoval btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL); 6669e0af237SLiu Bo btrfs_queue_work(wq, &end_io_wq->work); 667ce9adaa5SChris Mason } 668ce9adaa5SChris Mason 6694e4cbee9SChristoph Hellwig blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 670bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata) 6710b86a832SChris Mason { 67297eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 6738b110e39SMiao Xie 67497eb6b69SDavid Sterba end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS); 675ce9adaa5SChris Mason if (!end_io_wq) 6764e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 677ce9adaa5SChris Mason 678ce9adaa5SChris Mason end_io_wq->private = bio->bi_private; 679ce9adaa5SChris Mason end_io_wq->end_io = bio->bi_end_io; 68022c59948SChris Mason end_io_wq->info = info; 6814e4cbee9SChristoph Hellwig end_io_wq->status = 0; 682ce9adaa5SChris Mason end_io_wq->bio = bio; 68322c59948SChris Mason end_io_wq->metadata = metadata; 684ce9adaa5SChris Mason 685ce9adaa5SChris Mason bio->bi_private = end_io_wq; 686ce9adaa5SChris Mason bio->bi_end_io = end_workqueue_bio; 68722c59948SChris Mason return 0; 68822c59948SChris Mason } 68922c59948SChris Mason 6904a69a410SChris Mason static void run_one_async_start(struct btrfs_work *work) 6914a69a410SChris Mason { 6924a69a410SChris Mason struct async_submit_bio *async; 6934e4cbee9SChristoph Hellwig blk_status_t ret; 6944a69a410SChris Mason 6954a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 6968896a08dSQu Wenruo ret = async->submit_bio_start(async->inode, async->bio, async->bio_offset); 69779787eaaSJeff Mahoney if (ret) 6984e4cbee9SChristoph Hellwig async->status = ret; 6994a69a410SChris Mason } 7004a69a410SChris Mason 70106ea01b1SDavid Sterba /* 70206ea01b1SDavid Sterba * In order to insert checksums into the metadata in large chunks, we wait 70306ea01b1SDavid Sterba * until bio submission time. All the pages in the bio are checksummed and 70406ea01b1SDavid Sterba * sums are attached onto the ordered extent record. 70506ea01b1SDavid Sterba * 70606ea01b1SDavid Sterba * At IO completion time the csums attached on the ordered extent record are 70706ea01b1SDavid Sterba * inserted into the tree. 70806ea01b1SDavid Sterba */ 7094a69a410SChris Mason static void run_one_async_done(struct btrfs_work *work) 7108b712842SChris Mason { 7118b712842SChris Mason struct async_submit_bio *async; 71206ea01b1SDavid Sterba struct inode *inode; 71306ea01b1SDavid Sterba blk_status_t ret; 7148b712842SChris Mason 7158b712842SChris Mason async = container_of(work, struct async_submit_bio, work); 7168896a08dSQu Wenruo inode = async->inode; 7174854ddd0SChris Mason 718bb7ab3b9SAdam Buchbinder /* If an error occurred we just want to clean up the bio and move on */ 7194e4cbee9SChristoph Hellwig if (async->status) { 7204e4cbee9SChristoph Hellwig async->bio->bi_status = async->status; 7214246a0b6SChristoph Hellwig bio_endio(async->bio); 72279787eaaSJeff Mahoney return; 72379787eaaSJeff Mahoney } 72479787eaaSJeff Mahoney 725ec39f769SChris Mason /* 726ec39f769SChris Mason * All of the bios that pass through here are from async helpers. 727ec39f769SChris Mason * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context. 728ec39f769SChris Mason * This changes nothing when cgroups aren't in use. 729ec39f769SChris Mason */ 730ec39f769SChris Mason async->bio->bi_opf |= REQ_CGROUP_PUNT; 73108635baeSChris Mason ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num); 73206ea01b1SDavid Sterba if (ret) { 73306ea01b1SDavid Sterba async->bio->bi_status = ret; 73406ea01b1SDavid Sterba bio_endio(async->bio); 73506ea01b1SDavid Sterba } 7364a69a410SChris Mason } 7374a69a410SChris Mason 7384a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 7394a69a410SChris Mason { 7404a69a410SChris Mason struct async_submit_bio *async; 7414a69a410SChris Mason 7424a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 7438b712842SChris Mason kfree(async); 7448b712842SChris Mason } 7458b712842SChris Mason 7468896a08dSQu Wenruo blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio, 747c6100a4bSJosef Bacik int mirror_num, unsigned long bio_flags, 7488896a08dSQu Wenruo u64 bio_offset, 749e288c080SDavid Sterba extent_submit_bio_start_t *submit_bio_start) 75044b8bd7eSChris Mason { 7518896a08dSQu Wenruo struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; 75244b8bd7eSChris Mason struct async_submit_bio *async; 75344b8bd7eSChris Mason 75444b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 75544b8bd7eSChris Mason if (!async) 7564e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 75744b8bd7eSChris Mason 7588896a08dSQu Wenruo async->inode = inode; 75944b8bd7eSChris Mason async->bio = bio; 76044b8bd7eSChris Mason async->mirror_num = mirror_num; 7614a69a410SChris Mason async->submit_bio_start = submit_bio_start; 7624a69a410SChris Mason 763a0cac0ecSOmar Sandoval btrfs_init_work(&async->work, run_one_async_start, run_one_async_done, 764a0cac0ecSOmar Sandoval run_one_async_free); 7654a69a410SChris Mason 766eaf25d93SChris Mason async->bio_offset = bio_offset; 7678c8bee1dSChris Mason 7684e4cbee9SChristoph Hellwig async->status = 0; 76979787eaaSJeff Mahoney 77067f055c7SChristoph Hellwig if (op_is_sync(bio->bi_opf)) 7715cdc7ad3SQu Wenruo btrfs_set_work_high_priority(&async->work); 772d313d7a3SChris Mason 7735cdc7ad3SQu Wenruo btrfs_queue_work(fs_info->workers, &async->work); 77444b8bd7eSChris Mason return 0; 77544b8bd7eSChris Mason } 77644b8bd7eSChris Mason 7774e4cbee9SChristoph Hellwig static blk_status_t btree_csum_one_bio(struct bio *bio) 778ce3ed71aSChris Mason { 7792c30c71bSKent Overstreet struct bio_vec *bvec; 780ce3ed71aSChris Mason struct btrfs_root *root; 7812b070cfeSChristoph Hellwig int ret = 0; 7826dc4f100SMing Lei struct bvec_iter_all iter_all; 783ce3ed71aSChris Mason 784c09abff8SDavid Sterba ASSERT(!bio_flagged(bio, BIO_CLONED)); 7852b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 786ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 78701d58472SDaniel Dressler ret = csum_dirty_buffer(root->fs_info, bvec->bv_page); 78879787eaaSJeff Mahoney if (ret) 78979787eaaSJeff Mahoney break; 790ce3ed71aSChris Mason } 7912c30c71bSKent Overstreet 7924e4cbee9SChristoph Hellwig return errno_to_blk_status(ret); 793ce3ed71aSChris Mason } 794ce3ed71aSChris Mason 7958896a08dSQu Wenruo static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio, 796eaf25d93SChris Mason u64 bio_offset) 79722c59948SChris Mason { 7988b712842SChris Mason /* 7998b712842SChris Mason * when we're called for a write, we're already in the async 8005443be45SChris Mason * submission context. Just jump into btrfs_map_bio 8018b712842SChris Mason */ 80279787eaaSJeff Mahoney return btree_csum_one_bio(bio); 80322c59948SChris Mason } 80422c59948SChris Mason 8059b4e675aSDavid Sterba static int check_async_write(struct btrfs_fs_info *fs_info, 8069b4e675aSDavid Sterba struct btrfs_inode *bi) 807de0022b9SJosef Bacik { 8086300463bSLiu Bo if (atomic_read(&bi->sync_writers)) 8096300463bSLiu Bo return 0; 8109b4e675aSDavid Sterba if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags)) 811de0022b9SJosef Bacik return 0; 812de0022b9SJosef Bacik return 1; 813de0022b9SJosef Bacik } 814de0022b9SJosef Bacik 8151b36294aSNikolay Borisov blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio, 8161b36294aSNikolay Borisov int mirror_num, unsigned long bio_flags) 81744b8bd7eSChris Mason { 8180b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 8199b4e675aSDavid Sterba int async = check_async_write(fs_info, BTRFS_I(inode)); 8204e4cbee9SChristoph Hellwig blk_status_t ret; 821cad321adSChris Mason 82237226b21SMike Christie if (bio_op(bio) != REQ_OP_WRITE) { 823cad321adSChris Mason /* 824cad321adSChris Mason * called for a read, do the setup so that checksum validation 825cad321adSChris Mason * can happen in the async kernel threads 826cad321adSChris Mason */ 8270b246afaSJeff Mahoney ret = btrfs_bio_wq_end_io(fs_info, bio, 8280b246afaSJeff Mahoney BTRFS_WQ_ENDIO_METADATA); 8291d4284bdSChris Mason if (ret) 83061891923SStefan Behrens goto out_w_error; 83108635baeSChris Mason ret = btrfs_map_bio(fs_info, bio, mirror_num); 832de0022b9SJosef Bacik } else if (!async) { 833de0022b9SJosef Bacik ret = btree_csum_one_bio(bio); 834de0022b9SJosef Bacik if (ret) 83561891923SStefan Behrens goto out_w_error; 83608635baeSChris Mason ret = btrfs_map_bio(fs_info, bio, mirror_num); 83761891923SStefan Behrens } else { 838cad321adSChris Mason /* 83961891923SStefan Behrens * kthread helpers are used to submit writes so that 84061891923SStefan Behrens * checksumming can happen in parallel across all CPUs 841cad321adSChris Mason */ 8428896a08dSQu Wenruo ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0, 8438896a08dSQu Wenruo 0, btree_submit_bio_start); 84444b8bd7eSChris Mason } 84544b8bd7eSChris Mason 8464246a0b6SChristoph Hellwig if (ret) 8474246a0b6SChristoph Hellwig goto out_w_error; 8484246a0b6SChristoph Hellwig return 0; 8494246a0b6SChristoph Hellwig 85061891923SStefan Behrens out_w_error: 8514e4cbee9SChristoph Hellwig bio->bi_status = ret; 8524246a0b6SChristoph Hellwig bio_endio(bio); 85361891923SStefan Behrens return ret; 85461891923SStefan Behrens } 85561891923SStefan Behrens 8563dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 857784b4e29SChris Mason static int btree_migratepage(struct address_space *mapping, 858a6bc32b8SMel Gorman struct page *newpage, struct page *page, 859a6bc32b8SMel Gorman enum migrate_mode mode) 860784b4e29SChris Mason { 861784b4e29SChris Mason /* 862784b4e29SChris Mason * we can't safely write a btree page from here, 863784b4e29SChris Mason * we haven't done the locking hook 864784b4e29SChris Mason */ 865784b4e29SChris Mason if (PageDirty(page)) 866784b4e29SChris Mason return -EAGAIN; 867784b4e29SChris Mason /* 868784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 869784b4e29SChris Mason * We must have no buffers or drop them. 870784b4e29SChris Mason */ 871784b4e29SChris Mason if (page_has_private(page) && 872784b4e29SChris Mason !try_to_release_page(page, GFP_KERNEL)) 873784b4e29SChris Mason return -EAGAIN; 874a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 875784b4e29SChris Mason } 8763dd1462eSJan Beulich #endif 877784b4e29SChris Mason 8780da5468fSChris Mason 8790da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 8800da5468fSChris Mason struct writeback_control *wbc) 8810da5468fSChris Mason { 882e2d84521SMiao Xie struct btrfs_fs_info *fs_info; 883e2d84521SMiao Xie int ret; 884e2d84521SMiao Xie 885d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 886448d640bSChris Mason 887448d640bSChris Mason if (wbc->for_kupdate) 888448d640bSChris Mason return 0; 889448d640bSChris Mason 890e2d84521SMiao Xie fs_info = BTRFS_I(mapping->host)->root->fs_info; 891b9473439SChris Mason /* this is a bit racy, but that's ok */ 892d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 893d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 894d814a491SEthan Lien fs_info->dirty_metadata_batch); 895e2d84521SMiao Xie if (ret < 0) 896793955bcSChris Mason return 0; 897793955bcSChris Mason } 8980b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 8990da5468fSChris Mason } 9000da5468fSChris Mason 90170dec807SChris Mason static int btree_releasepage(struct page *page, gfp_t gfp_flags) 9025f39d397SChris Mason { 90398509cfcSChris Mason if (PageWriteback(page) || PageDirty(page)) 90498509cfcSChris Mason return 0; 9050c4e538bSDavid Sterba 906f7a52a40SDavid Sterba return try_release_extent_buffer(page); 907d98237b3SChris Mason } 908d98237b3SChris Mason 909d47992f8SLukas Czerner static void btree_invalidatepage(struct page *page, unsigned int offset, 910d47992f8SLukas Czerner unsigned int length) 911d98237b3SChris Mason { 912d1310b2eSChris Mason struct extent_io_tree *tree; 913d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9145f39d397SChris Mason extent_invalidatepage(tree, page, offset); 9155f39d397SChris Mason btree_releasepage(page, GFP_NOFS); 9169ad6b7bcSChris Mason if (PagePrivate(page)) { 917efe120a0SFrank Holton btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info, 918efe120a0SFrank Holton "page private not zero on page %llu", 919efe120a0SFrank Holton (unsigned long long)page_offset(page)); 920d1b89bc0SGuoqing Jiang detach_page_private(page); 9219ad6b7bcSChris Mason } 922d98237b3SChris Mason } 923d98237b3SChris Mason 9240b32f4bbSJosef Bacik static int btree_set_page_dirty(struct page *page) 9250b32f4bbSJosef Bacik { 926bb146eb2SJosef Bacik #ifdef DEBUG 9270b32f4bbSJosef Bacik struct extent_buffer *eb; 9280b32f4bbSJosef Bacik 9290b32f4bbSJosef Bacik BUG_ON(!PagePrivate(page)); 9300b32f4bbSJosef Bacik eb = (struct extent_buffer *)page->private; 9310b32f4bbSJosef Bacik BUG_ON(!eb); 9320b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 9330b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 9340b32f4bbSJosef Bacik btrfs_assert_tree_locked(eb); 935bb146eb2SJosef Bacik #endif 9360b32f4bbSJosef Bacik return __set_page_dirty_nobuffers(page); 9370b32f4bbSJosef Bacik } 9380b32f4bbSJosef Bacik 9397f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 9400da5468fSChris Mason .writepages = btree_writepages, 9415f39d397SChris Mason .releasepage = btree_releasepage, 9425f39d397SChris Mason .invalidatepage = btree_invalidatepage, 9435a92bc88SChris Mason #ifdef CONFIG_MIGRATION 944784b4e29SChris Mason .migratepage = btree_migratepage, 9455a92bc88SChris Mason #endif 9460b32f4bbSJosef Bacik .set_page_dirty = btree_set_page_dirty, 947d98237b3SChris Mason }; 948123abc88SChris Mason 9492ff7e61eSJeff Mahoney void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr) 950090d1875SChris Mason { 9515f39d397SChris Mason struct extent_buffer *buf = NULL; 952537f38f0SNikolay Borisov int ret; 953090d1875SChris Mason 9542ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 955c871b0f2SLiu Bo if (IS_ERR(buf)) 9566197d86eSDavid Sterba return; 957537f38f0SNikolay Borisov 958c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(buf, WAIT_NONE, 0); 959537f38f0SNikolay Borisov if (ret < 0) 960537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 961537f38f0SNikolay Borisov else 9625f39d397SChris Mason free_extent_buffer(buf); 963090d1875SChris Mason } 964090d1875SChris Mason 9652ff7e61eSJeff Mahoney struct extent_buffer *btrfs_find_create_tree_block( 9662ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info, 9670308af44SDavid Sterba u64 bytenr) 9680999df54SChris Mason { 9690b246afaSJeff Mahoney if (btrfs_is_testing(fs_info)) 9700b246afaSJeff Mahoney return alloc_test_extent_buffer(fs_info, bytenr); 9710b246afaSJeff Mahoney return alloc_extent_buffer(fs_info, bytenr); 9720999df54SChris Mason } 9730999df54SChris Mason 974581c1760SQu Wenruo /* 975581c1760SQu Wenruo * Read tree block at logical address @bytenr and do variant basic but critical 976581c1760SQu Wenruo * verification. 977581c1760SQu Wenruo * 978581c1760SQu Wenruo * @parent_transid: expected transid of this tree block, skip check if 0 979581c1760SQu Wenruo * @level: expected level, mandatory check 980581c1760SQu Wenruo * @first_key: expected key in slot 0, skip check if NULL 981581c1760SQu Wenruo */ 9822ff7e61eSJeff Mahoney struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, 983581c1760SQu Wenruo u64 parent_transid, int level, 984581c1760SQu Wenruo struct btrfs_key *first_key) 985e20d96d6SChris Mason { 9865f39d397SChris Mason struct extent_buffer *buf = NULL; 98719c00ddcSChris Mason int ret; 98819c00ddcSChris Mason 9892ff7e61eSJeff Mahoney buf = btrfs_find_create_tree_block(fs_info, bytenr); 990c871b0f2SLiu Bo if (IS_ERR(buf)) 991c871b0f2SLiu Bo return buf; 992e4204dedSChris Mason 9935ab12d1fSDavid Sterba ret = btree_read_extent_buffer_pages(buf, parent_transid, 994581c1760SQu Wenruo level, first_key); 9950f0fe8f7SFilipe David Borba Manana if (ret) { 996537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 99764c043deSLiu Bo return ERR_PTR(ret); 9980f0fe8f7SFilipe David Borba Manana } 9995f39d397SChris Mason return buf; 1000ce9adaa5SChris Mason 1001eb60ceacSChris Mason } 1002eb60ceacSChris Mason 10036a884d7dSDavid Sterba void btrfs_clean_tree_block(struct extent_buffer *buf) 1004ed2ff2cbSChris Mason { 10056a884d7dSDavid Sterba struct btrfs_fs_info *fs_info = buf->fs_info; 100655c69072SChris Mason if (btrfs_header_generation(buf) == 1007e2d84521SMiao Xie fs_info->running_transaction->transid) { 1008b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1009b4ce94deSChris Mason 1010b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1011104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 1012e2d84521SMiao Xie -buf->len, 1013e2d84521SMiao Xie fs_info->dirty_metadata_batch); 10140b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 1015925baeddSChris Mason } 10165f39d397SChris Mason } 1017ed7b63ebSJosef Bacik } 10185f39d397SChris Mason 1019da17066cSJeff Mahoney static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, 1020e20d96d6SChris Mason u64 objectid) 1021d97e63b6SChris Mason { 10227c0260eeSJeff Mahoney bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); 102396dfcb46SJosef Bacik root->fs_info = fs_info; 1024cfaa7295SChris Mason root->node = NULL; 1025a28ec197SChris Mason root->commit_root = NULL; 102627cdeb70SMiao Xie root->state = 0; 1027d68fc57bSYan, Zheng root->orphan_cleanup_state = 0; 10280b86a832SChris Mason 10290f7d52f4SChris Mason root->last_trans = 0; 103013a8a7c8SYan, Zheng root->highest_objectid = 0; 1031eb73c1b7SMiao Xie root->nr_delalloc_inodes = 0; 1032199c2a9cSMiao Xie root->nr_ordered_extents = 0; 10336bef4d31SEric Paris root->inode_tree = RB_ROOT; 103416cdcec7SMiao Xie INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1035f0486c68SYan, Zheng root->block_rsv = NULL; 10360b86a832SChris Mason 10370b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 10385d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 1039eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_inodes); 1040eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_root); 1041199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_extents); 1042199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_root); 1043d2311e69SQu Wenruo INIT_LIST_HEAD(&root->reloc_dirty_list); 10442ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 10452ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 10465d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1047eb73c1b7SMiao Xie spin_lock_init(&root->delalloc_lock); 1048199c2a9cSMiao Xie spin_lock_init(&root->ordered_extent_lock); 1049f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 10502ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 10512ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 10528287475aSQu Wenruo spin_lock_init(&root->qgroup_meta_rsv_lock); 1053a2135011SChris Mason mutex_init(&root->objectid_mutex); 1054e02119d5SChris Mason mutex_init(&root->log_mutex); 105531f3d255SMiao Xie mutex_init(&root->ordered_extent_mutex); 1056573bfb72SMiao Xie mutex_init(&root->delalloc_mutex); 1057c53e9653SQu Wenruo init_waitqueue_head(&root->qgroup_flush_wait); 10587237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 10597237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 10607237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 10618b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[0]); 10628b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[1]); 10637237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 10647237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 10657237f183SYan Zheng atomic_set(&root->log_writers, 0); 10662ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 10670700cea7SElena Reshetova refcount_set(&root->refs, 1); 10688ecebf4dSRobbie Ko atomic_set(&root->snapshot_force_cow, 0); 1069eede2bf3SOmar Sandoval atomic_set(&root->nr_swapfiles, 0); 10707237f183SYan Zheng root->log_transid = 0; 1071d1433debSMiao Xie root->log_transid_committed = -1; 1072257c62e1SChris Mason root->last_log_commit = 0; 1073e289f03eSFilipe Manana if (!dummy) { 107443eb5f29SQu Wenruo extent_io_tree_init(fs_info, &root->dirty_log_pages, 107543eb5f29SQu Wenruo IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL); 1076e289f03eSFilipe Manana extent_io_tree_init(fs_info, &root->log_csum_range, 1077e289f03eSFilipe Manana IO_TREE_LOG_CSUM_RANGE, NULL); 1078e289f03eSFilipe Manana } 1079017e5369SChris Mason 10803768f368SChris Mason memset(&root->root_key, 0, sizeof(root->root_key)); 10813768f368SChris Mason memset(&root->root_item, 0, sizeof(root->root_item)); 10826702ed49SChris Mason memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 10834d775673SChris Mason root->root_key.objectid = objectid; 10840ee5dc67SAl Viro root->anon_dev = 0; 10858ea05e3aSAlexander Block 10865f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 1087370a11b8SQu Wenruo btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks); 1088bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 1089bd647ce3SJosef Bacik INIT_LIST_HEAD(&root->leak_list); 1090bd647ce3SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 1091bd647ce3SJosef Bacik list_add_tail(&root->leak_list, &fs_info->allocated_roots); 1092bd647ce3SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 1093bd647ce3SJosef Bacik #endif 10943768f368SChris Mason } 10953768f368SChris Mason 109674e4d827SDavid Sterba static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, 109796dfcb46SJosef Bacik u64 objectid, gfp_t flags) 10986f07e42eSAl Viro { 109974e4d827SDavid Sterba struct btrfs_root *root = kzalloc(sizeof(*root), flags); 11006f07e42eSAl Viro if (root) 110196dfcb46SJosef Bacik __setup_root(root, fs_info, objectid); 11026f07e42eSAl Viro return root; 11036f07e42eSAl Viro } 11046f07e42eSAl Viro 110506ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 110606ea65a3SJosef Bacik /* Should only be used by the testing infrastructure */ 1107da17066cSJeff Mahoney struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) 110806ea65a3SJosef Bacik { 110906ea65a3SJosef Bacik struct btrfs_root *root; 111006ea65a3SJosef Bacik 11117c0260eeSJeff Mahoney if (!fs_info) 11127c0260eeSJeff Mahoney return ERR_PTR(-EINVAL); 11137c0260eeSJeff Mahoney 111496dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL); 111506ea65a3SJosef Bacik if (!root) 111606ea65a3SJosef Bacik return ERR_PTR(-ENOMEM); 1117da17066cSJeff Mahoney 1118b9ef22deSFeifei Xu /* We don't use the stripesize in selftest, set it as sectorsize */ 1119faa2dbf0SJosef Bacik root->alloc_bytenr = 0; 112006ea65a3SJosef Bacik 112106ea65a3SJosef Bacik return root; 112206ea65a3SJosef Bacik } 112306ea65a3SJosef Bacik #endif 112406ea65a3SJosef Bacik 112520897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 112620897f5cSArne Jansen u64 objectid) 112720897f5cSArne Jansen { 11289b7a2440SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 112920897f5cSArne Jansen struct extent_buffer *leaf; 113020897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 113120897f5cSArne Jansen struct btrfs_root *root; 113220897f5cSArne Jansen struct btrfs_key key; 1133b89f6d1fSFilipe Manana unsigned int nofs_flag; 113420897f5cSArne Jansen int ret = 0; 113520897f5cSArne Jansen 1136b89f6d1fSFilipe Manana /* 1137b89f6d1fSFilipe Manana * We're holding a transaction handle, so use a NOFS memory allocation 1138b89f6d1fSFilipe Manana * context to avoid deadlock if reclaim happens. 1139b89f6d1fSFilipe Manana */ 1140b89f6d1fSFilipe Manana nofs_flag = memalloc_nofs_save(); 114196dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL); 1142b89f6d1fSFilipe Manana memalloc_nofs_restore(nofs_flag); 114320897f5cSArne Jansen if (!root) 114420897f5cSArne Jansen return ERR_PTR(-ENOMEM); 114520897f5cSArne Jansen 114620897f5cSArne Jansen root->root_key.objectid = objectid; 114720897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 114820897f5cSArne Jansen root->root_key.offset = 0; 114920897f5cSArne Jansen 11509631e4ccSJosef Bacik leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0, 11519631e4ccSJosef Bacik BTRFS_NESTING_NORMAL); 115220897f5cSArne Jansen if (IS_ERR(leaf)) { 115320897f5cSArne Jansen ret = PTR_ERR(leaf); 11541dd05682STsutomu Itoh leaf = NULL; 115520897f5cSArne Jansen goto fail; 115620897f5cSArne Jansen } 115720897f5cSArne Jansen 115820897f5cSArne Jansen root->node = leaf; 115920897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 116020897f5cSArne Jansen 116120897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 116227cdeb70SMiao Xie set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 116320897f5cSArne Jansen 1164f944d2cbSDavid Sterba btrfs_set_root_flags(&root->root_item, 0); 1165f944d2cbSDavid Sterba btrfs_set_root_limit(&root->root_item, 0); 116620897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 116720897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 116820897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 116920897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 117020897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 117120897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 117220897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 117333d85fdaSQu Wenruo if (is_fstree(objectid)) 1174807fc790SAndy Shevchenko generate_random_guid(root->root_item.uuid); 1175807fc790SAndy Shevchenko else 1176807fc790SAndy Shevchenko export_guid(root->root_item.uuid, &guid_null); 1177c8422684SDavid Sterba btrfs_set_root_drop_level(&root->root_item, 0); 117820897f5cSArne Jansen 117920897f5cSArne Jansen key.objectid = objectid; 118020897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 118120897f5cSArne Jansen key.offset = 0; 118220897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 118320897f5cSArne Jansen if (ret) 118420897f5cSArne Jansen goto fail; 118520897f5cSArne Jansen 118620897f5cSArne Jansen btrfs_tree_unlock(leaf); 118720897f5cSArne Jansen 118820897f5cSArne Jansen return root; 11891dd05682STsutomu Itoh 11901dd05682STsutomu Itoh fail: 11918c38938cSJosef Bacik if (leaf) 11921dd05682STsutomu Itoh btrfs_tree_unlock(leaf); 119300246528SJosef Bacik btrfs_put_root(root); 11941dd05682STsutomu Itoh 11951dd05682STsutomu Itoh return ERR_PTR(ret); 119620897f5cSArne Jansen } 119720897f5cSArne Jansen 11987237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1199e02119d5SChris Mason struct btrfs_fs_info *fs_info) 12000f7d52f4SChris Mason { 12010f7d52f4SChris Mason struct btrfs_root *root; 12027237f183SYan Zheng struct extent_buffer *leaf; 1203e02119d5SChris Mason 120496dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS); 1205e02119d5SChris Mason if (!root) 12067237f183SYan Zheng return ERR_PTR(-ENOMEM); 1207e02119d5SChris Mason 1208e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1209e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1210e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 121127cdeb70SMiao Xie 12127237f183SYan Zheng /* 121392a7cc42SQu Wenruo * DON'T set SHAREABLE bit for log trees. 121427cdeb70SMiao Xie * 121592a7cc42SQu Wenruo * Log trees are not exposed to user space thus can't be snapshotted, 121692a7cc42SQu Wenruo * and they go away before a real commit is actually done. 121792a7cc42SQu Wenruo * 121892a7cc42SQu Wenruo * They do store pointers to file data extents, and those reference 121992a7cc42SQu Wenruo * counts still get updated (along with back refs to the log tree). 12207237f183SYan Zheng */ 1221e02119d5SChris Mason 12224d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 12239631e4ccSJosef Bacik NULL, 0, 0, 0, BTRFS_NESTING_NORMAL); 12247237f183SYan Zheng if (IS_ERR(leaf)) { 122500246528SJosef Bacik btrfs_put_root(root); 12267237f183SYan Zheng return ERR_CAST(leaf); 12277237f183SYan Zheng } 1228e02119d5SChris Mason 12297237f183SYan Zheng root->node = leaf; 1230e02119d5SChris Mason 1231e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1232e02119d5SChris Mason btrfs_tree_unlock(root->node); 12337237f183SYan Zheng return root; 12347237f183SYan Zheng } 12357237f183SYan Zheng 12367237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 12377237f183SYan Zheng struct btrfs_fs_info *fs_info) 12387237f183SYan Zheng { 12397237f183SYan Zheng struct btrfs_root *log_root; 12407237f183SYan Zheng 12417237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 12427237f183SYan Zheng if (IS_ERR(log_root)) 12437237f183SYan Zheng return PTR_ERR(log_root); 12447237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 12457237f183SYan Zheng fs_info->log_root_tree = log_root; 12467237f183SYan Zheng return 0; 12477237f183SYan Zheng } 12487237f183SYan Zheng 12497237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 12507237f183SYan Zheng struct btrfs_root *root) 12517237f183SYan Zheng { 12520b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12537237f183SYan Zheng struct btrfs_root *log_root; 12547237f183SYan Zheng struct btrfs_inode_item *inode_item; 12557237f183SYan Zheng 12560b246afaSJeff Mahoney log_root = alloc_log_tree(trans, fs_info); 12577237f183SYan Zheng if (IS_ERR(log_root)) 12587237f183SYan Zheng return PTR_ERR(log_root); 12597237f183SYan Zheng 12607237f183SYan Zheng log_root->last_trans = trans->transid; 12617237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 12627237f183SYan Zheng 12637237f183SYan Zheng inode_item = &log_root->root_item.inode; 12643cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 12653cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 12663cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 1267da17066cSJeff Mahoney btrfs_set_stack_inode_nbytes(inode_item, 12680b246afaSJeff Mahoney fs_info->nodesize); 12693cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 12707237f183SYan Zheng 12715d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 12727237f183SYan Zheng 12737237f183SYan Zheng WARN_ON(root->log_root); 12747237f183SYan Zheng root->log_root = log_root; 12757237f183SYan Zheng root->log_transid = 0; 1276d1433debSMiao Xie root->log_transid_committed = -1; 1277257c62e1SChris Mason root->last_log_commit = 0; 1278e02119d5SChris Mason return 0; 1279e02119d5SChris Mason } 1280e02119d5SChris Mason 128149d11beaSJosef Bacik static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root, 128249d11beaSJosef Bacik struct btrfs_path *path, 1283cb517eabSMiao Xie struct btrfs_key *key) 1284e02119d5SChris Mason { 1285e02119d5SChris Mason struct btrfs_root *root; 1286e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 128784234f3aSYan Zheng u64 generation; 1288cb517eabSMiao Xie int ret; 1289581c1760SQu Wenruo int level; 1290cb517eabSMiao Xie 129196dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS); 129249d11beaSJosef Bacik if (!root) 129349d11beaSJosef Bacik return ERR_PTR(-ENOMEM); 12940f7d52f4SChris Mason 1295cb517eabSMiao Xie ret = btrfs_find_root(tree_root, key, path, 1296cb517eabSMiao Xie &root->root_item, &root->root_key); 12970f7d52f4SChris Mason if (ret) { 129813a8a7c8SYan, Zheng if (ret > 0) 129913a8a7c8SYan, Zheng ret = -ENOENT; 130049d11beaSJosef Bacik goto fail; 13010f7d52f4SChris Mason } 130213a8a7c8SYan, Zheng 130384234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1304581c1760SQu Wenruo level = btrfs_root_level(&root->root_item); 13052ff7e61eSJeff Mahoney root->node = read_tree_block(fs_info, 13062ff7e61eSJeff Mahoney btrfs_root_bytenr(&root->root_item), 1307581c1760SQu Wenruo generation, level, NULL); 130864c043deSLiu Bo if (IS_ERR(root->node)) { 130964c043deSLiu Bo ret = PTR_ERR(root->node); 13108c38938cSJosef Bacik root->node = NULL; 131149d11beaSJosef Bacik goto fail; 1312cb517eabSMiao Xie } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1313cb517eabSMiao Xie ret = -EIO; 131449d11beaSJosef Bacik goto fail; 1315416bc658SJosef Bacik } 13165d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 1317cb517eabSMiao Xie return root; 131849d11beaSJosef Bacik fail: 131900246528SJosef Bacik btrfs_put_root(root); 132049d11beaSJosef Bacik return ERR_PTR(ret); 132149d11beaSJosef Bacik } 132249d11beaSJosef Bacik 132349d11beaSJosef Bacik struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 132449d11beaSJosef Bacik struct btrfs_key *key) 132549d11beaSJosef Bacik { 132649d11beaSJosef Bacik struct btrfs_root *root; 132749d11beaSJosef Bacik struct btrfs_path *path; 132849d11beaSJosef Bacik 132949d11beaSJosef Bacik path = btrfs_alloc_path(); 133049d11beaSJosef Bacik if (!path) 133149d11beaSJosef Bacik return ERR_PTR(-ENOMEM); 133249d11beaSJosef Bacik root = read_tree_root_path(tree_root, path, key); 133349d11beaSJosef Bacik btrfs_free_path(path); 133449d11beaSJosef Bacik 133549d11beaSJosef Bacik return root; 1336cb517eabSMiao Xie } 1337cb517eabSMiao Xie 13382dfb1e43SQu Wenruo /* 13392dfb1e43SQu Wenruo * Initialize subvolume root in-memory structure 13402dfb1e43SQu Wenruo * 13412dfb1e43SQu Wenruo * @anon_dev: anonymous device to attach to the root, if zero, allocate new 13422dfb1e43SQu Wenruo */ 13432dfb1e43SQu Wenruo static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev) 1344cb517eabSMiao Xie { 1345cb517eabSMiao Xie int ret; 1346dcc3eb96SNikolay Borisov unsigned int nofs_flag; 1347cb517eabSMiao Xie 1348cb517eabSMiao Xie root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); 1349cb517eabSMiao Xie root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), 1350cb517eabSMiao Xie GFP_NOFS); 1351cb517eabSMiao Xie if (!root->free_ino_pinned || !root->free_ino_ctl) { 1352cb517eabSMiao Xie ret = -ENOMEM; 1353cb517eabSMiao Xie goto fail; 1354cb517eabSMiao Xie } 1355cb517eabSMiao Xie 1356dcc3eb96SNikolay Borisov /* 1357dcc3eb96SNikolay Borisov * We might be called under a transaction (e.g. indirect backref 1358dcc3eb96SNikolay Borisov * resolution) which could deadlock if it triggers memory reclaim 1359dcc3eb96SNikolay Borisov */ 1360dcc3eb96SNikolay Borisov nofs_flag = memalloc_nofs_save(); 1361dcc3eb96SNikolay Borisov ret = btrfs_drew_lock_init(&root->snapshot_lock); 1362dcc3eb96SNikolay Borisov memalloc_nofs_restore(nofs_flag); 1363dcc3eb96SNikolay Borisov if (ret) 13648257b2dcSMiao Xie goto fail; 13658257b2dcSMiao Xie 1366aeb935a4SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID && 1367aeb935a4SQu Wenruo root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { 136892a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &root->state); 1369f39e4571SJosef Bacik btrfs_check_and_init_root_item(&root->root_item); 1370f39e4571SJosef Bacik } 1371f39e4571SJosef Bacik 1372cb517eabSMiao Xie btrfs_init_free_ino_ctl(root); 137357cdc8dbSDavid Sterba spin_lock_init(&root->ino_cache_lock); 137457cdc8dbSDavid Sterba init_waitqueue_head(&root->ino_cache_wait); 1375cb517eabSMiao Xie 1376851fd730SQu Wenruo /* 1377851fd730SQu Wenruo * Don't assign anonymous block device to roots that are not exposed to 1378851fd730SQu Wenruo * userspace, the id pool is limited to 1M 1379851fd730SQu Wenruo */ 1380851fd730SQu Wenruo if (is_fstree(root->root_key.objectid) && 1381851fd730SQu Wenruo btrfs_root_refs(&root->root_item) > 0) { 13822dfb1e43SQu Wenruo if (!anon_dev) { 1383cb517eabSMiao Xie ret = get_anon_bdev(&root->anon_dev); 1384cb517eabSMiao Xie if (ret) 1385876d2cf1SLiu Bo goto fail; 13862dfb1e43SQu Wenruo } else { 13872dfb1e43SQu Wenruo root->anon_dev = anon_dev; 13882dfb1e43SQu Wenruo } 1389851fd730SQu Wenruo } 1390f32e48e9SChandan Rajendra 1391f32e48e9SChandan Rajendra mutex_lock(&root->objectid_mutex); 1392f32e48e9SChandan Rajendra ret = btrfs_find_highest_objectid(root, 1393f32e48e9SChandan Rajendra &root->highest_objectid); 1394f32e48e9SChandan Rajendra if (ret) { 1395f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1396876d2cf1SLiu Bo goto fail; 1397f32e48e9SChandan Rajendra } 1398f32e48e9SChandan Rajendra 1399f32e48e9SChandan Rajendra ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 1400f32e48e9SChandan Rajendra 1401f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1402f32e48e9SChandan Rajendra 1403cb517eabSMiao Xie return 0; 1404cb517eabSMiao Xie fail: 140584db5ccfSDavid Sterba /* The caller is responsible to call btrfs_free_fs_root */ 1406cb517eabSMiao Xie return ret; 1407cb517eabSMiao Xie } 1408cb517eabSMiao Xie 1409a98db0f3SJosef Bacik static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1410cb517eabSMiao Xie u64 root_id) 1411cb517eabSMiao Xie { 1412cb517eabSMiao Xie struct btrfs_root *root; 1413cb517eabSMiao Xie 1414cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1415cb517eabSMiao Xie root = radix_tree_lookup(&fs_info->fs_roots_radix, 1416cb517eabSMiao Xie (unsigned long)root_id); 1417bc44d7c4SJosef Bacik if (root) 141800246528SJosef Bacik root = btrfs_grab_root(root); 1419cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1420cb517eabSMiao Xie return root; 1421cb517eabSMiao Xie } 1422cb517eabSMiao Xie 142349d11beaSJosef Bacik static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info, 142449d11beaSJosef Bacik u64 objectid) 142549d11beaSJosef Bacik { 142649d11beaSJosef Bacik if (objectid == BTRFS_ROOT_TREE_OBJECTID) 142749d11beaSJosef Bacik return btrfs_grab_root(fs_info->tree_root); 142849d11beaSJosef Bacik if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 142949d11beaSJosef Bacik return btrfs_grab_root(fs_info->extent_root); 143049d11beaSJosef Bacik if (objectid == BTRFS_CHUNK_TREE_OBJECTID) 143149d11beaSJosef Bacik return btrfs_grab_root(fs_info->chunk_root); 143249d11beaSJosef Bacik if (objectid == BTRFS_DEV_TREE_OBJECTID) 143349d11beaSJosef Bacik return btrfs_grab_root(fs_info->dev_root); 143449d11beaSJosef Bacik if (objectid == BTRFS_CSUM_TREE_OBJECTID) 143549d11beaSJosef Bacik return btrfs_grab_root(fs_info->csum_root); 143649d11beaSJosef Bacik if (objectid == BTRFS_QUOTA_TREE_OBJECTID) 143749d11beaSJosef Bacik return btrfs_grab_root(fs_info->quota_root) ? 143849d11beaSJosef Bacik fs_info->quota_root : ERR_PTR(-ENOENT); 143949d11beaSJosef Bacik if (objectid == BTRFS_UUID_TREE_OBJECTID) 144049d11beaSJosef Bacik return btrfs_grab_root(fs_info->uuid_root) ? 144149d11beaSJosef Bacik fs_info->uuid_root : ERR_PTR(-ENOENT); 144249d11beaSJosef Bacik if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) 144349d11beaSJosef Bacik return btrfs_grab_root(fs_info->free_space_root) ? 144449d11beaSJosef Bacik fs_info->free_space_root : ERR_PTR(-ENOENT); 144549d11beaSJosef Bacik return NULL; 144649d11beaSJosef Bacik } 144749d11beaSJosef Bacik 1448cb517eabSMiao Xie int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1449cb517eabSMiao Xie struct btrfs_root *root) 1450cb517eabSMiao Xie { 1451cb517eabSMiao Xie int ret; 1452cb517eabSMiao Xie 1453e1860a77SDavid Sterba ret = radix_tree_preload(GFP_NOFS); 1454cb517eabSMiao Xie if (ret) 1455cb517eabSMiao Xie return ret; 1456cb517eabSMiao Xie 1457cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1458cb517eabSMiao Xie ret = radix_tree_insert(&fs_info->fs_roots_radix, 1459cb517eabSMiao Xie (unsigned long)root->root_key.objectid, 1460cb517eabSMiao Xie root); 1461af01d2e5SJosef Bacik if (ret == 0) { 146200246528SJosef Bacik btrfs_grab_root(root); 146327cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1464af01d2e5SJosef Bacik } 1465cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1466cb517eabSMiao Xie radix_tree_preload_end(); 1467cb517eabSMiao Xie 1468cb517eabSMiao Xie return ret; 1469cb517eabSMiao Xie } 1470cb517eabSMiao Xie 1471bd647ce3SJosef Bacik void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info) 1472bd647ce3SJosef Bacik { 1473bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 1474bd647ce3SJosef Bacik struct btrfs_root *root; 1475bd647ce3SJosef Bacik 1476bd647ce3SJosef Bacik while (!list_empty(&fs_info->allocated_roots)) { 1477457f1864SJosef Bacik char buf[BTRFS_ROOT_NAME_BUF_LEN]; 1478457f1864SJosef Bacik 1479bd647ce3SJosef Bacik root = list_first_entry(&fs_info->allocated_roots, 1480bd647ce3SJosef Bacik struct btrfs_root, leak_list); 1481457f1864SJosef Bacik btrfs_err(fs_info, "leaked root %s refcount %d", 1482457f1864SJosef Bacik btrfs_root_name(root->root_key.objectid, buf), 1483bd647ce3SJosef Bacik refcount_read(&root->refs)); 1484bd647ce3SJosef Bacik while (refcount_read(&root->refs) > 1) 148500246528SJosef Bacik btrfs_put_root(root); 148600246528SJosef Bacik btrfs_put_root(root); 1487bd647ce3SJosef Bacik } 1488bd647ce3SJosef Bacik #endif 1489bd647ce3SJosef Bacik } 1490bd647ce3SJosef Bacik 14910d4b0463SJosef Bacik void btrfs_free_fs_info(struct btrfs_fs_info *fs_info) 14920d4b0463SJosef Bacik { 1493141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 1494141386e1SJosef Bacik percpu_counter_destroy(&fs_info->delalloc_bytes); 1495141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dio_bytes); 1496141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 1497141386e1SJosef Bacik btrfs_free_csum_hash(fs_info); 1498141386e1SJosef Bacik btrfs_free_stripe_hash_table(fs_info); 1499141386e1SJosef Bacik btrfs_free_ref_cache(fs_info); 15000d4b0463SJosef Bacik kfree(fs_info->balance_ctl); 15010d4b0463SJosef Bacik kfree(fs_info->delayed_root); 150200246528SJosef Bacik btrfs_put_root(fs_info->extent_root); 150300246528SJosef Bacik btrfs_put_root(fs_info->tree_root); 150400246528SJosef Bacik btrfs_put_root(fs_info->chunk_root); 150500246528SJosef Bacik btrfs_put_root(fs_info->dev_root); 150600246528SJosef Bacik btrfs_put_root(fs_info->csum_root); 150700246528SJosef Bacik btrfs_put_root(fs_info->quota_root); 150800246528SJosef Bacik btrfs_put_root(fs_info->uuid_root); 150900246528SJosef Bacik btrfs_put_root(fs_info->free_space_root); 151000246528SJosef Bacik btrfs_put_root(fs_info->fs_root); 1511aeb935a4SQu Wenruo btrfs_put_root(fs_info->data_reloc_root); 1512bd647ce3SJosef Bacik btrfs_check_leaked_roots(fs_info); 15133fd63727SJosef Bacik btrfs_extent_buffer_leak_debug_check(fs_info); 15140d4b0463SJosef Bacik kfree(fs_info->super_copy); 15150d4b0463SJosef Bacik kfree(fs_info->super_for_commit); 15160d4b0463SJosef Bacik kvfree(fs_info); 15170d4b0463SJosef Bacik } 15180d4b0463SJosef Bacik 15190d4b0463SJosef Bacik 15202dfb1e43SQu Wenruo /* 15212dfb1e43SQu Wenruo * Get an in-memory reference of a root structure. 15222dfb1e43SQu Wenruo * 15232dfb1e43SQu Wenruo * For essential trees like root/extent tree, we grab it from fs_info directly. 15242dfb1e43SQu Wenruo * For subvolume trees, we check the cached filesystem roots first. If not 15252dfb1e43SQu Wenruo * found, then read it from disk and add it to cached fs roots. 15262dfb1e43SQu Wenruo * 15272dfb1e43SQu Wenruo * Caller should release the root by calling btrfs_put_root() after the usage. 15282dfb1e43SQu Wenruo * 15292dfb1e43SQu Wenruo * NOTE: Reloc and log trees can't be read by this function as they share the 15302dfb1e43SQu Wenruo * same root objectid. 15312dfb1e43SQu Wenruo * 15322dfb1e43SQu Wenruo * @objectid: root id 15332dfb1e43SQu Wenruo * @anon_dev: preallocated anonymous block device number for new roots, 15342dfb1e43SQu Wenruo * pass 0 for new allocation. 15352dfb1e43SQu Wenruo * @check_ref: whether to check root item references, If true, return -ENOENT 15362dfb1e43SQu Wenruo * for orphan roots 15372dfb1e43SQu Wenruo */ 15382dfb1e43SQu Wenruo static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info, 15392dfb1e43SQu Wenruo u64 objectid, dev_t anon_dev, 15402dfb1e43SQu Wenruo bool check_ref) 15415eda7b5eSChris Mason { 15425eda7b5eSChris Mason struct btrfs_root *root; 1543381cf658SDavid Sterba struct btrfs_path *path; 15441d4c08e0SDavid Sterba struct btrfs_key key; 15455eda7b5eSChris Mason int ret; 15465eda7b5eSChris Mason 154749d11beaSJosef Bacik root = btrfs_get_global_root(fs_info, objectid); 154849d11beaSJosef Bacik if (root) 154949d11beaSJosef Bacik return root; 15504df27c4dSYan, Zheng again: 155156e9357aSDavid Sterba root = btrfs_lookup_fs_root(fs_info, objectid); 155248475471SStefan Behrens if (root) { 15532dfb1e43SQu Wenruo /* Shouldn't get preallocated anon_dev for cached roots */ 15542dfb1e43SQu Wenruo ASSERT(!anon_dev); 1555bc44d7c4SJosef Bacik if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 155600246528SJosef Bacik btrfs_put_root(root); 155748475471SStefan Behrens return ERR_PTR(-ENOENT); 1558bc44d7c4SJosef Bacik } 15595eda7b5eSChris Mason return root; 156048475471SStefan Behrens } 15615eda7b5eSChris Mason 156256e9357aSDavid Sterba key.objectid = objectid; 156356e9357aSDavid Sterba key.type = BTRFS_ROOT_ITEM_KEY; 156456e9357aSDavid Sterba key.offset = (u64)-1; 156556e9357aSDavid Sterba root = btrfs_read_tree_root(fs_info->tree_root, &key); 15665eda7b5eSChris Mason if (IS_ERR(root)) 15675eda7b5eSChris Mason return root; 15683394e160SChris Mason 1569c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1570d68fc57bSYan, Zheng ret = -ENOENT; 1571d68fc57bSYan, Zheng goto fail; 1572d68fc57bSYan, Zheng } 1573d68fc57bSYan, Zheng 15742dfb1e43SQu Wenruo ret = btrfs_init_fs_root(root, anon_dev); 1575cb517eabSMiao Xie if (ret) 1576cb517eabSMiao Xie goto fail; 1577cb517eabSMiao Xie 1578381cf658SDavid Sterba path = btrfs_alloc_path(); 1579381cf658SDavid Sterba if (!path) { 1580381cf658SDavid Sterba ret = -ENOMEM; 1581381cf658SDavid Sterba goto fail; 1582381cf658SDavid Sterba } 15831d4c08e0SDavid Sterba key.objectid = BTRFS_ORPHAN_OBJECTID; 15841d4c08e0SDavid Sterba key.type = BTRFS_ORPHAN_ITEM_KEY; 158556e9357aSDavid Sterba key.offset = objectid; 15861d4c08e0SDavid Sterba 15871d4c08e0SDavid Sterba ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1588381cf658SDavid Sterba btrfs_free_path(path); 1589d68fc57bSYan, Zheng if (ret < 0) 1590d68fc57bSYan, Zheng goto fail; 1591d68fc57bSYan, Zheng if (ret == 0) 159227cdeb70SMiao Xie set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1593d68fc57bSYan, Zheng 1594cb517eabSMiao Xie ret = btrfs_insert_fs_root(fs_info, root); 15950f7d52f4SChris Mason if (ret) { 159600246528SJosef Bacik btrfs_put_root(root); 15974785e24fSJosef Bacik if (ret == -EEXIST) 15984df27c4dSYan, Zheng goto again; 15994df27c4dSYan, Zheng goto fail; 16004df27c4dSYan, Zheng } 1601edbd8d4eSChris Mason return root; 16024df27c4dSYan, Zheng fail: 16038c38938cSJosef Bacik btrfs_put_root(root); 16044df27c4dSYan, Zheng return ERR_PTR(ret); 1605edbd8d4eSChris Mason } 1606edbd8d4eSChris Mason 16072dfb1e43SQu Wenruo /* 16082dfb1e43SQu Wenruo * Get in-memory reference of a root structure 16092dfb1e43SQu Wenruo * 16102dfb1e43SQu Wenruo * @objectid: tree objectid 16112dfb1e43SQu Wenruo * @check_ref: if set, verify that the tree exists and the item has at least 16122dfb1e43SQu Wenruo * one reference 16132dfb1e43SQu Wenruo */ 16142dfb1e43SQu Wenruo struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 16152dfb1e43SQu Wenruo u64 objectid, bool check_ref) 16162dfb1e43SQu Wenruo { 16172dfb1e43SQu Wenruo return btrfs_get_root_ref(fs_info, objectid, 0, check_ref); 16182dfb1e43SQu Wenruo } 16192dfb1e43SQu Wenruo 16202dfb1e43SQu Wenruo /* 16212dfb1e43SQu Wenruo * Get in-memory reference of a root structure, created as new, optionally pass 16222dfb1e43SQu Wenruo * the anonymous block device id 16232dfb1e43SQu Wenruo * 16242dfb1e43SQu Wenruo * @objectid: tree objectid 16252dfb1e43SQu Wenruo * @anon_dev: if zero, allocate a new anonymous block device or use the 16262dfb1e43SQu Wenruo * parameter value 16272dfb1e43SQu Wenruo */ 16282dfb1e43SQu Wenruo struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info, 16292dfb1e43SQu Wenruo u64 objectid, dev_t anon_dev) 16302dfb1e43SQu Wenruo { 16312dfb1e43SQu Wenruo return btrfs_get_root_ref(fs_info, objectid, anon_dev, true); 16322dfb1e43SQu Wenruo } 16332dfb1e43SQu Wenruo 16348b712842SChris Mason /* 163549d11beaSJosef Bacik * btrfs_get_fs_root_commit_root - return a root for the given objectid 163649d11beaSJosef Bacik * @fs_info: the fs_info 163749d11beaSJosef Bacik * @objectid: the objectid we need to lookup 163849d11beaSJosef Bacik * 163949d11beaSJosef Bacik * This is exclusively used for backref walking, and exists specifically because 164049d11beaSJosef Bacik * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref 164149d11beaSJosef Bacik * creation time, which means we may have to read the tree_root in order to look 164249d11beaSJosef Bacik * up a fs root that is not in memory. If the root is not in memory we will 164349d11beaSJosef Bacik * read the tree root commit root and look up the fs root from there. This is a 164449d11beaSJosef Bacik * temporary root, it will not be inserted into the radix tree as it doesn't 164549d11beaSJosef Bacik * have the most uptodate information, it'll simply be discarded once the 164649d11beaSJosef Bacik * backref code is finished using the root. 164749d11beaSJosef Bacik */ 164849d11beaSJosef Bacik struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info, 164949d11beaSJosef Bacik struct btrfs_path *path, 165049d11beaSJosef Bacik u64 objectid) 165149d11beaSJosef Bacik { 165249d11beaSJosef Bacik struct btrfs_root *root; 165349d11beaSJosef Bacik struct btrfs_key key; 165449d11beaSJosef Bacik 165549d11beaSJosef Bacik ASSERT(path->search_commit_root && path->skip_locking); 165649d11beaSJosef Bacik 165749d11beaSJosef Bacik /* 165849d11beaSJosef Bacik * This can return -ENOENT if we ask for a root that doesn't exist, but 165949d11beaSJosef Bacik * since this is called via the backref walking code we won't be looking 166049d11beaSJosef Bacik * up a root that doesn't exist, unless there's corruption. So if root 166149d11beaSJosef Bacik * != NULL just return it. 166249d11beaSJosef Bacik */ 166349d11beaSJosef Bacik root = btrfs_get_global_root(fs_info, objectid); 166449d11beaSJosef Bacik if (root) 166549d11beaSJosef Bacik return root; 166649d11beaSJosef Bacik 166749d11beaSJosef Bacik root = btrfs_lookup_fs_root(fs_info, objectid); 166849d11beaSJosef Bacik if (root) 166949d11beaSJosef Bacik return root; 167049d11beaSJosef Bacik 167149d11beaSJosef Bacik key.objectid = objectid; 167249d11beaSJosef Bacik key.type = BTRFS_ROOT_ITEM_KEY; 167349d11beaSJosef Bacik key.offset = (u64)-1; 167449d11beaSJosef Bacik root = read_tree_root_path(fs_info->tree_root, path, &key); 167549d11beaSJosef Bacik btrfs_release_path(path); 167649d11beaSJosef Bacik 167749d11beaSJosef Bacik return root; 167849d11beaSJosef Bacik } 167949d11beaSJosef Bacik 168049d11beaSJosef Bacik /* 16818b712842SChris Mason * called by the kthread helper functions to finally call the bio end_io 16828b712842SChris Mason * functions. This is where read checksum verification actually happens 16838b712842SChris Mason */ 16848b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work) 1685ce9adaa5SChris Mason { 1686ce9adaa5SChris Mason struct bio *bio; 168797eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 1688ce9adaa5SChris Mason 168997eb6b69SDavid Sterba end_io_wq = container_of(work, struct btrfs_end_io_wq, work); 1690ce9adaa5SChris Mason bio = end_io_wq->bio; 16918b712842SChris Mason 16924e4cbee9SChristoph Hellwig bio->bi_status = end_io_wq->status; 1693ce9adaa5SChris Mason bio->bi_private = end_io_wq->private; 1694ce9adaa5SChris Mason bio->bi_end_io = end_io_wq->end_io; 16954246a0b6SChristoph Hellwig bio_endio(bio); 16969be490f1SOmar Sandoval kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq); 1697ce9adaa5SChris Mason } 169844b8bd7eSChris Mason 1699a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1700a74a4b97SChris Mason { 1701a74a4b97SChris Mason struct btrfs_root *root = arg; 17020b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1703d0278245SMiao Xie int again; 1704a74a4b97SChris Mason 1705d6fd0ae2SOmar Sandoval while (1) { 1706d0278245SMiao Xie again = 0; 17079d1a2a3aSDavid Sterba 1708fd340d0fSJosef Bacik set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1709fd340d0fSJosef Bacik 1710d0278245SMiao Xie /* Make the cleaner go to sleep early. */ 17112ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) 1712d0278245SMiao Xie goto sleep; 1713d0278245SMiao Xie 171490c711abSZygo Blaxell /* 171590c711abSZygo Blaxell * Do not do anything if we might cause open_ctree() to block 171690c711abSZygo Blaxell * before we have finished mounting the filesystem. 171790c711abSZygo Blaxell */ 17180b246afaSJeff Mahoney if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 171990c711abSZygo Blaxell goto sleep; 172090c711abSZygo Blaxell 17210b246afaSJeff Mahoney if (!mutex_trylock(&fs_info->cleaner_mutex)) 1722d0278245SMiao Xie goto sleep; 1723d0278245SMiao Xie 1724dc7f370cSMiao Xie /* 1725dc7f370cSMiao Xie * Avoid the problem that we change the status of the fs 1726dc7f370cSMiao Xie * during the above check and trylock. 1727dc7f370cSMiao Xie */ 17282ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) { 17290b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1730dc7f370cSMiao Xie goto sleep; 1731dc7f370cSMiao Xie } 1732dc7f370cSMiao Xie 17332ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 1734c2d6cb16SFilipe Manana 17359d1a2a3aSDavid Sterba again = btrfs_clean_one_deleted_snapshot(root); 17360b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1737a74a4b97SChris Mason 1738d0278245SMiao Xie /* 173905323cd1SMiao Xie * The defragger has dealt with the R/O remount and umount, 174005323cd1SMiao Xie * needn't do anything special here. 1741d0278245SMiao Xie */ 17420b246afaSJeff Mahoney btrfs_run_defrag_inodes(fs_info); 174367c5e7d4SFilipe Manana 174467c5e7d4SFilipe Manana /* 174567c5e7d4SFilipe Manana * Acquires fs_info->delete_unused_bgs_mutex to avoid racing 174667c5e7d4SFilipe Manana * with relocation (btrfs_relocate_chunk) and relocation 174767c5e7d4SFilipe Manana * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 174867c5e7d4SFilipe Manana * after acquiring fs_info->delete_unused_bgs_mutex. So we 174967c5e7d4SFilipe Manana * can't hold, nor need to, fs_info->cleaner_mutex when deleting 175067c5e7d4SFilipe Manana * unused block groups. 175167c5e7d4SFilipe Manana */ 17520b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 1753d0278245SMiao Xie sleep: 1754fd340d0fSJosef Bacik clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1755d6fd0ae2SOmar Sandoval if (kthread_should_park()) 1756d6fd0ae2SOmar Sandoval kthread_parkme(); 1757d6fd0ae2SOmar Sandoval if (kthread_should_stop()) 1758d6fd0ae2SOmar Sandoval return 0; 1759838fe188SJiri Kosina if (!again) { 1760a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 1761a74a4b97SChris Mason schedule(); 1762a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1763a74a4b97SChris Mason } 1764da288d28SFilipe Manana } 1765a74a4b97SChris Mason } 1766a74a4b97SChris Mason 1767a74a4b97SChris Mason static int transaction_kthread(void *arg) 1768a74a4b97SChris Mason { 1769a74a4b97SChris Mason struct btrfs_root *root = arg; 17700b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1771a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1772a74a4b97SChris Mason struct btrfs_transaction *cur; 17738929ecfaSYan, Zheng u64 transid; 1774643900beSNikolay Borisov time64_t delta; 1775a74a4b97SChris Mason unsigned long delay; 1776914b2007SJan Kara bool cannot_commit; 1777a74a4b97SChris Mason 1778a74a4b97SChris Mason do { 1779914b2007SJan Kara cannot_commit = false; 1780ba1bc00fSNikolay Borisov delay = msecs_to_jiffies(fs_info->commit_interval * 1000); 17810b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 1782a74a4b97SChris Mason 17830b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 17840b246afaSJeff Mahoney cur = fs_info->running_transaction; 1785a74a4b97SChris Mason if (!cur) { 17860b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1787a74a4b97SChris Mason goto sleep; 1788a74a4b97SChris Mason } 178931153d81SYan Zheng 1790643900beSNikolay Borisov delta = ktime_get_seconds() - cur->start_time; 17913296bf56SQu Wenruo if (cur->state < TRANS_STATE_COMMIT_START && 1792643900beSNikolay Borisov delta < fs_info->commit_interval) { 17930b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1794fb8a7e94SNikolay Borisov delay -= msecs_to_jiffies((delta - 1) * 1000); 1795fb8a7e94SNikolay Borisov delay = min(delay, 1796fb8a7e94SNikolay Borisov msecs_to_jiffies(fs_info->commit_interval * 1000)); 1797a74a4b97SChris Mason goto sleep; 1798a74a4b97SChris Mason } 17998929ecfaSYan, Zheng transid = cur->transid; 18000b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 180156bec294SChris Mason 180279787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 1803354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 1804914b2007SJan Kara if (IS_ERR(trans)) { 1805354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 1806914b2007SJan Kara cannot_commit = true; 180779787eaaSJeff Mahoney goto sleep; 1808914b2007SJan Kara } 18098929ecfaSYan, Zheng if (transid == trans->transid) { 18103a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 18118929ecfaSYan, Zheng } else { 18123a45bb20SJeff Mahoney btrfs_end_transaction(trans); 18138929ecfaSYan, Zheng } 1814a74a4b97SChris Mason sleep: 18150b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 18160b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 1817a74a4b97SChris Mason 18184e121c06SJosef Bacik if (unlikely(test_bit(BTRFS_FS_STATE_ERROR, 18190b246afaSJeff Mahoney &fs_info->fs_state))) 18202ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 18218929ecfaSYan, Zheng if (!kthread_should_stop() && 18220b246afaSJeff Mahoney (!btrfs_transaction_blocked(fs_info) || 1823914b2007SJan Kara cannot_commit)) 1824bc5511d0SNikolay Borisov schedule_timeout_interruptible(delay); 1825a74a4b97SChris Mason } while (!kthread_should_stop()); 1826a74a4b97SChris Mason return 0; 1827a74a4b97SChris Mason } 1828a74a4b97SChris Mason 1829af31f5e5SChris Mason /* 183001f0f9daSNikolay Borisov * This will find the highest generation in the array of root backups. The 183101f0f9daSNikolay Borisov * index of the highest array is returned, or -EINVAL if we can't find 183201f0f9daSNikolay Borisov * anything. 1833af31f5e5SChris Mason * 1834af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1835af31f5e5SChris Mason * generation of the latest root in the array with the generation 1836af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1837af31f5e5SChris Mason */ 183801f0f9daSNikolay Borisov static int find_newest_super_backup(struct btrfs_fs_info *info) 1839af31f5e5SChris Mason { 184001f0f9daSNikolay Borisov const u64 newest_gen = btrfs_super_generation(info->super_copy); 1841af31f5e5SChris Mason u64 cur; 1842af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1843af31f5e5SChris Mason int i; 1844af31f5e5SChris Mason 1845af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1846af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1847af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1848af31f5e5SChris Mason if (cur == newest_gen) 184901f0f9daSNikolay Borisov return i; 1850af31f5e5SChris Mason } 1851af31f5e5SChris Mason 185201f0f9daSNikolay Borisov return -EINVAL; 1853af31f5e5SChris Mason } 1854af31f5e5SChris Mason 1855af31f5e5SChris Mason /* 1856af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1857af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1858af31f5e5SChris Mason * done 1859af31f5e5SChris Mason */ 1860af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1861af31f5e5SChris Mason { 18626ef108ddSNikolay Borisov const int next_backup = info->backup_root_index; 1863af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1864af31f5e5SChris Mason 1865af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1866af31f5e5SChris Mason 1867af31f5e5SChris Mason /* 1868af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1869af31f5e5SChris Mason * regardless of which ones we use today 1870af31f5e5SChris Mason */ 1871af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1872af31f5e5SChris Mason 1873af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1874af31f5e5SChris Mason 1875af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1876af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1877af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1878af31f5e5SChris Mason 1879af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1880af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1881af31f5e5SChris Mason 1882af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1883af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1884af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 1885af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 1886af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 1887af31f5e5SChris Mason 1888af31f5e5SChris Mason btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); 1889af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 1890af31f5e5SChris Mason btrfs_header_generation(info->extent_root->node)); 1891af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 1892af31f5e5SChris Mason btrfs_header_level(info->extent_root->node)); 1893af31f5e5SChris Mason 18947c7e82a7SChris Mason /* 18957c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 18967c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 18977c7e82a7SChris Mason */ 18987c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 18997c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 19007c7e82a7SChris Mason info->fs_root->node->start); 1901af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 1902af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 1903af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 1904af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 19057c7e82a7SChris Mason } 1906af31f5e5SChris Mason 1907af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 1908af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 1909af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 1910af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 1911af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 1912af31f5e5SChris Mason 1913af31f5e5SChris Mason btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); 1914af31f5e5SChris Mason btrfs_set_backup_csum_root_gen(root_backup, 1915af31f5e5SChris Mason btrfs_header_generation(info->csum_root->node)); 1916af31f5e5SChris Mason btrfs_set_backup_csum_root_level(root_backup, 1917af31f5e5SChris Mason btrfs_header_level(info->csum_root->node)); 1918af31f5e5SChris Mason 1919af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 1920af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 1921af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 1922af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 1923af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 1924af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 1925af31f5e5SChris Mason 1926af31f5e5SChris Mason /* 1927af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 1928af31f5e5SChris Mason * for the next commit 1929af31f5e5SChris Mason */ 1930af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 1931af31f5e5SChris Mason &info->super_for_commit->super_roots, 1932af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 1933af31f5e5SChris Mason } 1934af31f5e5SChris Mason 1935af31f5e5SChris Mason /* 1936bd2336b2SNikolay Borisov * read_backup_root - Reads a backup root based on the passed priority. Prio 0 1937bd2336b2SNikolay Borisov * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots 1938bd2336b2SNikolay Borisov * 1939bd2336b2SNikolay Borisov * fs_info - filesystem whose backup roots need to be read 1940bd2336b2SNikolay Borisov * priority - priority of backup root required 1941bd2336b2SNikolay Borisov * 1942bd2336b2SNikolay Borisov * Returns backup root index on success and -EINVAL otherwise. 1943bd2336b2SNikolay Borisov */ 1944bd2336b2SNikolay Borisov static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority) 1945bd2336b2SNikolay Borisov { 1946bd2336b2SNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 1947bd2336b2SNikolay Borisov struct btrfs_super_block *super = fs_info->super_copy; 1948bd2336b2SNikolay Borisov struct btrfs_root_backup *root_backup; 1949bd2336b2SNikolay Borisov 1950bd2336b2SNikolay Borisov if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) { 1951bd2336b2SNikolay Borisov if (priority == 0) 1952bd2336b2SNikolay Borisov return backup_index; 1953bd2336b2SNikolay Borisov 1954bd2336b2SNikolay Borisov backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority; 1955bd2336b2SNikolay Borisov backup_index %= BTRFS_NUM_BACKUP_ROOTS; 1956bd2336b2SNikolay Borisov } else { 1957bd2336b2SNikolay Borisov return -EINVAL; 1958bd2336b2SNikolay Borisov } 1959bd2336b2SNikolay Borisov 1960bd2336b2SNikolay Borisov root_backup = super->super_roots + backup_index; 1961bd2336b2SNikolay Borisov 1962bd2336b2SNikolay Borisov btrfs_set_super_generation(super, 1963bd2336b2SNikolay Borisov btrfs_backup_tree_root_gen(root_backup)); 1964bd2336b2SNikolay Borisov btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 1965bd2336b2SNikolay Borisov btrfs_set_super_root_level(super, 1966bd2336b2SNikolay Borisov btrfs_backup_tree_root_level(root_backup)); 1967bd2336b2SNikolay Borisov btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 1968bd2336b2SNikolay Borisov 1969bd2336b2SNikolay Borisov /* 1970bd2336b2SNikolay Borisov * Fixme: the total bytes and num_devices need to match or we should 1971bd2336b2SNikolay Borisov * need a fsck 1972bd2336b2SNikolay Borisov */ 1973bd2336b2SNikolay Borisov btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 1974bd2336b2SNikolay Borisov btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 1975bd2336b2SNikolay Borisov 1976bd2336b2SNikolay Borisov return backup_index; 1977bd2336b2SNikolay Borisov } 1978bd2336b2SNikolay Borisov 19797abadb64SLiu Bo /* helper to cleanup workers */ 19807abadb64SLiu Bo static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 19817abadb64SLiu Bo { 1982dc6e3209SQu Wenruo btrfs_destroy_workqueue(fs_info->fixup_workers); 1983afe3d242SQu Wenruo btrfs_destroy_workqueue(fs_info->delalloc_workers); 19845cdc7ad3SQu Wenruo btrfs_destroy_workqueue(fs_info->workers); 1985fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_workers); 1986fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_raid56_workers); 1987d05a33acSQu Wenruo btrfs_destroy_workqueue(fs_info->rmw_workers); 1988fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_write_workers); 1989fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 19905b3bc44eSQu Wenruo btrfs_destroy_workqueue(fs_info->delayed_workers); 1991e66f0bb1SQu Wenruo btrfs_destroy_workqueue(fs_info->caching_workers); 1992736cfa15SQu Wenruo btrfs_destroy_workqueue(fs_info->readahead_workers); 1993a44903abSQu Wenruo btrfs_destroy_workqueue(fs_info->flush_workers); 1994fc97fab0SQu Wenruo btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 1995b0643e59SDennis Zhou if (fs_info->discard_ctl.discard_workers) 1996b0643e59SDennis Zhou destroy_workqueue(fs_info->discard_ctl.discard_workers); 1997a9b9477dSFilipe Manana /* 1998a9b9477dSFilipe Manana * Now that all other work queues are destroyed, we can safely destroy 1999a9b9477dSFilipe Manana * the queues used for metadata I/O, since tasks from those other work 2000a9b9477dSFilipe Manana * queues can do metadata I/O operations. 2001a9b9477dSFilipe Manana */ 2002a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_workers); 2003a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_write_workers); 20047abadb64SLiu Bo } 20057abadb64SLiu Bo 20062e9f5954SRashika static void free_root_extent_buffers(struct btrfs_root *root) 20072e9f5954SRashika { 20082e9f5954SRashika if (root) { 20092e9f5954SRashika free_extent_buffer(root->node); 20102e9f5954SRashika free_extent_buffer(root->commit_root); 20112e9f5954SRashika root->node = NULL; 20122e9f5954SRashika root->commit_root = NULL; 20132e9f5954SRashika } 20142e9f5954SRashika } 20152e9f5954SRashika 2016af31f5e5SChris Mason /* helper to cleanup tree roots */ 20174273eaffSAnand Jain static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root) 2018af31f5e5SChris Mason { 20192e9f5954SRashika free_root_extent_buffers(info->tree_root); 2020655b09feSJosef Bacik 20212e9f5954SRashika free_root_extent_buffers(info->dev_root); 20222e9f5954SRashika free_root_extent_buffers(info->extent_root); 20232e9f5954SRashika free_root_extent_buffers(info->csum_root); 20242e9f5954SRashika free_root_extent_buffers(info->quota_root); 20252e9f5954SRashika free_root_extent_buffers(info->uuid_root); 20268c38938cSJosef Bacik free_root_extent_buffers(info->fs_root); 2027aeb935a4SQu Wenruo free_root_extent_buffers(info->data_reloc_root); 20284273eaffSAnand Jain if (free_chunk_root) 20292e9f5954SRashika free_root_extent_buffers(info->chunk_root); 203070f6d82eSOmar Sandoval free_root_extent_buffers(info->free_space_root); 2031af31f5e5SChris Mason } 2032af31f5e5SChris Mason 20338c38938cSJosef Bacik void btrfs_put_root(struct btrfs_root *root) 20348c38938cSJosef Bacik { 20358c38938cSJosef Bacik if (!root) 20368c38938cSJosef Bacik return; 20378c38938cSJosef Bacik 20388c38938cSJosef Bacik if (refcount_dec_and_test(&root->refs)) { 20398c38938cSJosef Bacik WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 20401dae7e0eSQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)); 20418c38938cSJosef Bacik if (root->anon_dev) 20428c38938cSJosef Bacik free_anon_bdev(root->anon_dev); 20438c38938cSJosef Bacik btrfs_drew_lock_destroy(&root->snapshot_lock); 2044923eb523SJohannes Thumshirn free_root_extent_buffers(root); 20458c38938cSJosef Bacik kfree(root->free_ino_ctl); 20468c38938cSJosef Bacik kfree(root->free_ino_pinned); 20478c38938cSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 20488c38938cSJosef Bacik spin_lock(&root->fs_info->fs_roots_radix_lock); 20498c38938cSJosef Bacik list_del_init(&root->leak_list); 20508c38938cSJosef Bacik spin_unlock(&root->fs_info->fs_roots_radix_lock); 20518c38938cSJosef Bacik #endif 20528c38938cSJosef Bacik kfree(root); 20538c38938cSJosef Bacik } 20548c38938cSJosef Bacik } 20558c38938cSJosef Bacik 2056faa2dbf0SJosef Bacik void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2057171f6537SJosef Bacik { 2058171f6537SJosef Bacik int ret; 2059171f6537SJosef Bacik struct btrfs_root *gang[8]; 2060171f6537SJosef Bacik int i; 2061171f6537SJosef Bacik 2062171f6537SJosef Bacik while (!list_empty(&fs_info->dead_roots)) { 2063171f6537SJosef Bacik gang[0] = list_entry(fs_info->dead_roots.next, 2064171f6537SJosef Bacik struct btrfs_root, root_list); 2065171f6537SJosef Bacik list_del(&gang[0]->root_list); 2066171f6537SJosef Bacik 20678c38938cSJosef Bacik if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) 2068cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[0]); 206900246528SJosef Bacik btrfs_put_root(gang[0]); 2070171f6537SJosef Bacik } 2071171f6537SJosef Bacik 2072171f6537SJosef Bacik while (1) { 2073171f6537SJosef Bacik ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2074171f6537SJosef Bacik (void **)gang, 0, 2075171f6537SJosef Bacik ARRAY_SIZE(gang)); 2076171f6537SJosef Bacik if (!ret) 2077171f6537SJosef Bacik break; 2078171f6537SJosef Bacik for (i = 0; i < ret; i++) 2079cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2080171f6537SJosef Bacik } 2081171f6537SJosef Bacik } 2082af31f5e5SChris Mason 2083638aa7edSEric Sandeen static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2084638aa7edSEric Sandeen { 2085638aa7edSEric Sandeen mutex_init(&fs_info->scrub_lock); 2086638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_running, 0); 2087638aa7edSEric Sandeen atomic_set(&fs_info->scrub_pause_req, 0); 2088638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_paused, 0); 2089638aa7edSEric Sandeen atomic_set(&fs_info->scrub_cancel_req, 0); 2090638aa7edSEric Sandeen init_waitqueue_head(&fs_info->scrub_pause_wait); 2091ff09c4caSAnand Jain refcount_set(&fs_info->scrub_workers_refcnt, 0); 2092638aa7edSEric Sandeen } 2093638aa7edSEric Sandeen 2094779a65a4SEric Sandeen static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2095779a65a4SEric Sandeen { 2096779a65a4SEric Sandeen spin_lock_init(&fs_info->balance_lock); 2097779a65a4SEric Sandeen mutex_init(&fs_info->balance_mutex); 2098779a65a4SEric Sandeen atomic_set(&fs_info->balance_pause_req, 0); 2099779a65a4SEric Sandeen atomic_set(&fs_info->balance_cancel_req, 0); 2100779a65a4SEric Sandeen fs_info->balance_ctl = NULL; 2101779a65a4SEric Sandeen init_waitqueue_head(&fs_info->balance_wait_q); 2102779a65a4SEric Sandeen } 2103779a65a4SEric Sandeen 21046bccf3abSJeff Mahoney static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info) 2105f37938e0SEric Sandeen { 21062ff7e61eSJeff Mahoney struct inode *inode = fs_info->btree_inode; 21072ff7e61eSJeff Mahoney 21082ff7e61eSJeff Mahoney inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 21092ff7e61eSJeff Mahoney set_nlink(inode, 1); 2110f37938e0SEric Sandeen /* 2111f37938e0SEric Sandeen * we set the i_size on the btree inode to the max possible int. 2112f37938e0SEric Sandeen * the real end of the address space is determined by all of 2113f37938e0SEric Sandeen * the devices in the system 2114f37938e0SEric Sandeen */ 21152ff7e61eSJeff Mahoney inode->i_size = OFFSET_MAX; 21162ff7e61eSJeff Mahoney inode->i_mapping->a_ops = &btree_aops; 2117f37938e0SEric Sandeen 21182ff7e61eSJeff Mahoney RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 211943eb5f29SQu Wenruo extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree, 21202c53a14dSQu Wenruo IO_TREE_BTREE_INODE_IO, inode); 21217b439738SDavid Sterba BTRFS_I(inode)->io_tree.track_uptodate = false; 21222ff7e61eSJeff Mahoney extent_map_tree_init(&BTRFS_I(inode)->extent_tree); 2123f37938e0SEric Sandeen 21245c8fd99fSJosef Bacik BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root); 21252ff7e61eSJeff Mahoney memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key)); 21262ff7e61eSJeff Mahoney set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); 21272ff7e61eSJeff Mahoney btrfs_insert_inode_hash(inode); 2128f37938e0SEric Sandeen } 2129f37938e0SEric Sandeen 2130ad618368SEric Sandeen static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2131ad618368SEric Sandeen { 2132ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2133129827e3SDavid Sterba init_rwsem(&fs_info->dev_replace.rwsem); 21347f8d236aSDavid Sterba init_waitqueue_head(&fs_info->dev_replace.replace_wait); 2135ad618368SEric Sandeen } 2136ad618368SEric Sandeen 2137f9e92e40SEric Sandeen static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2138f9e92e40SEric Sandeen { 2139f9e92e40SEric Sandeen spin_lock_init(&fs_info->qgroup_lock); 2140f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_ioctl_lock); 2141f9e92e40SEric Sandeen fs_info->qgroup_tree = RB_ROOT; 2142f9e92e40SEric Sandeen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2143f9e92e40SEric Sandeen fs_info->qgroup_seq = 1; 2144f9e92e40SEric Sandeen fs_info->qgroup_ulist = NULL; 2145d2c609b8SJeff Mahoney fs_info->qgroup_rescan_running = false; 2146f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_rescan_lock); 2147f9e92e40SEric Sandeen } 2148f9e92e40SEric Sandeen 21492a458198SEric Sandeen static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info, 21502a458198SEric Sandeen struct btrfs_fs_devices *fs_devices) 21512a458198SEric Sandeen { 2152f7b885beSAnand Jain u32 max_active = fs_info->thread_pool_size; 21536f011058SDavid Sterba unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 21542a458198SEric Sandeen 21552a458198SEric Sandeen fs_info->workers = 2156cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "worker", 2157cb001095SJeff Mahoney flags | WQ_HIGHPRI, max_active, 16); 21582a458198SEric Sandeen 21592a458198SEric Sandeen fs_info->delalloc_workers = 2160cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delalloc", 2161cb001095SJeff Mahoney flags, max_active, 2); 21622a458198SEric Sandeen 21632a458198SEric Sandeen fs_info->flush_workers = 2164cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "flush_delalloc", 2165cb001095SJeff Mahoney flags, max_active, 0); 21662a458198SEric Sandeen 21672a458198SEric Sandeen fs_info->caching_workers = 2168cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); 21692a458198SEric Sandeen 21702a458198SEric Sandeen fs_info->fixup_workers = 2171cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); 21722a458198SEric Sandeen 21732a458198SEric Sandeen /* 21742a458198SEric Sandeen * endios are largely parallel and should have a very 21752a458198SEric Sandeen * low idle thresh 21762a458198SEric Sandeen */ 21772a458198SEric Sandeen fs_info->endio_workers = 2178cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4); 21792a458198SEric Sandeen fs_info->endio_meta_workers = 2180cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta", flags, 2181cb001095SJeff Mahoney max_active, 4); 21822a458198SEric Sandeen fs_info->endio_meta_write_workers = 2183cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags, 2184cb001095SJeff Mahoney max_active, 2); 21852a458198SEric Sandeen fs_info->endio_raid56_workers = 2186cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-raid56", flags, 2187cb001095SJeff Mahoney max_active, 4); 21882a458198SEric Sandeen fs_info->rmw_workers = 2189cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2); 21902a458198SEric Sandeen fs_info->endio_write_workers = 2191cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-write", flags, 2192cb001095SJeff Mahoney max_active, 2); 21932a458198SEric Sandeen fs_info->endio_freespace_worker = 2194cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "freespace-write", flags, 2195cb001095SJeff Mahoney max_active, 0); 21962a458198SEric Sandeen fs_info->delayed_workers = 2197cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, 2198cb001095SJeff Mahoney max_active, 0); 21992a458198SEric Sandeen fs_info->readahead_workers = 2200cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "readahead", flags, 2201cb001095SJeff Mahoney max_active, 2); 22022a458198SEric Sandeen fs_info->qgroup_rescan_workers = 2203cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); 2204b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers = 2205b0643e59SDennis Zhou alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1); 22062a458198SEric Sandeen 22072a458198SEric Sandeen if (!(fs_info->workers && fs_info->delalloc_workers && 2208ba8a9d07SChris Mason fs_info->flush_workers && 22092a458198SEric Sandeen fs_info->endio_workers && fs_info->endio_meta_workers && 22102a458198SEric Sandeen fs_info->endio_meta_write_workers && 22112a458198SEric Sandeen fs_info->endio_write_workers && fs_info->endio_raid56_workers && 22122a458198SEric Sandeen fs_info->endio_freespace_worker && fs_info->rmw_workers && 22132a458198SEric Sandeen fs_info->caching_workers && fs_info->readahead_workers && 22142a458198SEric Sandeen fs_info->fixup_workers && fs_info->delayed_workers && 2215b0643e59SDennis Zhou fs_info->qgroup_rescan_workers && 2216b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers)) { 22172a458198SEric Sandeen return -ENOMEM; 22182a458198SEric Sandeen } 22192a458198SEric Sandeen 22202a458198SEric Sandeen return 0; 22212a458198SEric Sandeen } 22222a458198SEric Sandeen 22236d97c6e3SJohannes Thumshirn static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type) 22246d97c6e3SJohannes Thumshirn { 22256d97c6e3SJohannes Thumshirn struct crypto_shash *csum_shash; 2226b4e967beSDavid Sterba const char *csum_driver = btrfs_super_csum_driver(csum_type); 22276d97c6e3SJohannes Thumshirn 2228b4e967beSDavid Sterba csum_shash = crypto_alloc_shash(csum_driver, 0, 0); 22296d97c6e3SJohannes Thumshirn 22306d97c6e3SJohannes Thumshirn if (IS_ERR(csum_shash)) { 22316d97c6e3SJohannes Thumshirn btrfs_err(fs_info, "error allocating %s hash for checksum", 2232b4e967beSDavid Sterba csum_driver); 22336d97c6e3SJohannes Thumshirn return PTR_ERR(csum_shash); 22346d97c6e3SJohannes Thumshirn } 22356d97c6e3SJohannes Thumshirn 22366d97c6e3SJohannes Thumshirn fs_info->csum_shash = csum_shash; 22376d97c6e3SJohannes Thumshirn 22386d97c6e3SJohannes Thumshirn return 0; 22396d97c6e3SJohannes Thumshirn } 22406d97c6e3SJohannes Thumshirn 224163443bf5SEric Sandeen static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 224263443bf5SEric Sandeen struct btrfs_fs_devices *fs_devices) 224363443bf5SEric Sandeen { 224463443bf5SEric Sandeen int ret; 224563443bf5SEric Sandeen struct btrfs_root *log_tree_root; 224663443bf5SEric Sandeen struct btrfs_super_block *disk_super = fs_info->super_copy; 224763443bf5SEric Sandeen u64 bytenr = btrfs_super_log_root(disk_super); 2248581c1760SQu Wenruo int level = btrfs_super_log_root_level(disk_super); 224963443bf5SEric Sandeen 225063443bf5SEric Sandeen if (fs_devices->rw_devices == 0) { 2251f14d104dSDavid Sterba btrfs_warn(fs_info, "log replay required on RO media"); 225263443bf5SEric Sandeen return -EIO; 225363443bf5SEric Sandeen } 225463443bf5SEric Sandeen 225596dfcb46SJosef Bacik log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, 225696dfcb46SJosef Bacik GFP_KERNEL); 225763443bf5SEric Sandeen if (!log_tree_root) 225863443bf5SEric Sandeen return -ENOMEM; 225963443bf5SEric Sandeen 22602ff7e61eSJeff Mahoney log_tree_root->node = read_tree_block(fs_info, bytenr, 2261581c1760SQu Wenruo fs_info->generation + 1, 2262581c1760SQu Wenruo level, NULL); 226364c043deSLiu Bo if (IS_ERR(log_tree_root->node)) { 2264f14d104dSDavid Sterba btrfs_warn(fs_info, "failed to read log tree"); 22650eeff236SLiu Bo ret = PTR_ERR(log_tree_root->node); 22668c38938cSJosef Bacik log_tree_root->node = NULL; 226700246528SJosef Bacik btrfs_put_root(log_tree_root); 22680eeff236SLiu Bo return ret; 226964c043deSLiu Bo } else if (!extent_buffer_uptodate(log_tree_root->node)) { 2270f14d104dSDavid Sterba btrfs_err(fs_info, "failed to read log tree"); 227100246528SJosef Bacik btrfs_put_root(log_tree_root); 227263443bf5SEric Sandeen return -EIO; 227363443bf5SEric Sandeen } 227463443bf5SEric Sandeen /* returns with log_tree_root freed on success */ 227563443bf5SEric Sandeen ret = btrfs_recover_log_trees(log_tree_root); 227663443bf5SEric Sandeen if (ret) { 22770b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 227863443bf5SEric Sandeen "Failed to recover log tree"); 227900246528SJosef Bacik btrfs_put_root(log_tree_root); 228063443bf5SEric Sandeen return ret; 228163443bf5SEric Sandeen } 228263443bf5SEric Sandeen 2283bc98a42cSDavid Howells if (sb_rdonly(fs_info->sb)) { 22846bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 228563443bf5SEric Sandeen if (ret) 228663443bf5SEric Sandeen return ret; 228763443bf5SEric Sandeen } 228863443bf5SEric Sandeen 228963443bf5SEric Sandeen return 0; 229063443bf5SEric Sandeen } 229163443bf5SEric Sandeen 22926bccf3abSJeff Mahoney static int btrfs_read_roots(struct btrfs_fs_info *fs_info) 22934bbcaa64SEric Sandeen { 22946bccf3abSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 2295a4f3d2c4SDavid Sterba struct btrfs_root *root; 22964bbcaa64SEric Sandeen struct btrfs_key location; 22974bbcaa64SEric Sandeen int ret; 22984bbcaa64SEric Sandeen 22996bccf3abSJeff Mahoney BUG_ON(!fs_info->tree_root); 23006bccf3abSJeff Mahoney 23014bbcaa64SEric Sandeen location.objectid = BTRFS_EXTENT_TREE_OBJECTID; 23024bbcaa64SEric Sandeen location.type = BTRFS_ROOT_ITEM_KEY; 23034bbcaa64SEric Sandeen location.offset = 0; 23044bbcaa64SEric Sandeen 2305a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2306f50f4353SLiu Bo if (IS_ERR(root)) { 230742437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2308f50f4353SLiu Bo ret = PTR_ERR(root); 2309f50f4353SLiu Bo goto out; 2310f50f4353SLiu Bo } 231142437a63SJosef Bacik } else { 2312a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2313a4f3d2c4SDavid Sterba fs_info->extent_root = root; 231442437a63SJosef Bacik } 23154bbcaa64SEric Sandeen 23164bbcaa64SEric Sandeen location.objectid = BTRFS_DEV_TREE_OBJECTID; 2317a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2318f50f4353SLiu Bo if (IS_ERR(root)) { 231942437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2320f50f4353SLiu Bo ret = PTR_ERR(root); 2321f50f4353SLiu Bo goto out; 2322f50f4353SLiu Bo } 232342437a63SJosef Bacik } else { 2324a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2325a4f3d2c4SDavid Sterba fs_info->dev_root = root; 23264bbcaa64SEric Sandeen btrfs_init_devices_late(fs_info); 232742437a63SJosef Bacik } 23284bbcaa64SEric Sandeen 2329882dbe0cSJosef Bacik /* If IGNOREDATACSUMS is set don't bother reading the csum root. */ 2330882dbe0cSJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREDATACSUMS)) { 23314bbcaa64SEric Sandeen location.objectid = BTRFS_CSUM_TREE_OBJECTID; 2332a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2333f50f4353SLiu Bo if (IS_ERR(root)) { 233442437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2335f50f4353SLiu Bo ret = PTR_ERR(root); 2336f50f4353SLiu Bo goto out; 2337f50f4353SLiu Bo } 233842437a63SJosef Bacik } else { 2339a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2340a4f3d2c4SDavid Sterba fs_info->csum_root = root; 234142437a63SJosef Bacik } 2342882dbe0cSJosef Bacik } 23434bbcaa64SEric Sandeen 2344aeb935a4SQu Wenruo /* 2345aeb935a4SQu Wenruo * This tree can share blocks with some other fs tree during relocation 2346aeb935a4SQu Wenruo * and we need a proper setup by btrfs_get_fs_root 2347aeb935a4SQu Wenruo */ 234856e9357aSDavid Sterba root = btrfs_get_fs_root(tree_root->fs_info, 234956e9357aSDavid Sterba BTRFS_DATA_RELOC_TREE_OBJECTID, true); 2350aeb935a4SQu Wenruo if (IS_ERR(root)) { 235142437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2352aeb935a4SQu Wenruo ret = PTR_ERR(root); 2353aeb935a4SQu Wenruo goto out; 2354aeb935a4SQu Wenruo } 235542437a63SJosef Bacik } else { 2356aeb935a4SQu Wenruo set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2357aeb935a4SQu Wenruo fs_info->data_reloc_root = root; 235842437a63SJosef Bacik } 2359aeb935a4SQu Wenruo 23604bbcaa64SEric Sandeen location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2361a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2362a4f3d2c4SDavid Sterba if (!IS_ERR(root)) { 2363a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2364afcdd129SJosef Bacik set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); 2365a4f3d2c4SDavid Sterba fs_info->quota_root = root; 23664bbcaa64SEric Sandeen } 23674bbcaa64SEric Sandeen 23684bbcaa64SEric Sandeen location.objectid = BTRFS_UUID_TREE_OBJECTID; 2369a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2370a4f3d2c4SDavid Sterba if (IS_ERR(root)) { 237142437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2372a4f3d2c4SDavid Sterba ret = PTR_ERR(root); 23734bbcaa64SEric Sandeen if (ret != -ENOENT) 2374f50f4353SLiu Bo goto out; 237542437a63SJosef Bacik } 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)) { 238542437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2386f50f4353SLiu Bo ret = PTR_ERR(root); 2387f50f4353SLiu Bo goto out; 2388f50f4353SLiu Bo } 238942437a63SJosef Bacik } else { 239070f6d82eSOmar Sandoval set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 239170f6d82eSOmar Sandoval fs_info->free_space_root = root; 239270f6d82eSOmar Sandoval } 239342437a63SJosef Bacik } 239470f6d82eSOmar Sandoval 23954bbcaa64SEric Sandeen return 0; 2396f50f4353SLiu Bo out: 2397f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", 2398f50f4353SLiu Bo location.objectid, ret); 2399f50f4353SLiu Bo return ret; 24004bbcaa64SEric Sandeen } 24014bbcaa64SEric Sandeen 2402069ec957SQu Wenruo /* 2403069ec957SQu Wenruo * Real super block validation 2404069ec957SQu Wenruo * NOTE: super csum type and incompat features will not be checked here. 2405069ec957SQu Wenruo * 2406069ec957SQu Wenruo * @sb: super block to check 2407069ec957SQu Wenruo * @mirror_num: the super block number to check its bytenr: 2408069ec957SQu Wenruo * 0 the primary (1st) sb 2409069ec957SQu Wenruo * 1, 2 2nd and 3rd backup copy 2410069ec957SQu Wenruo * -1 skip bytenr check 2411069ec957SQu Wenruo */ 2412069ec957SQu Wenruo static int validate_super(struct btrfs_fs_info *fs_info, 2413069ec957SQu Wenruo struct btrfs_super_block *sb, int mirror_num) 241421a852b0SQu Wenruo { 241521a852b0SQu Wenruo u64 nodesize = btrfs_super_nodesize(sb); 241621a852b0SQu Wenruo u64 sectorsize = btrfs_super_sectorsize(sb); 241721a852b0SQu Wenruo int ret = 0; 241821a852b0SQu Wenruo 241921a852b0SQu Wenruo if (btrfs_super_magic(sb) != BTRFS_MAGIC) { 242021a852b0SQu Wenruo btrfs_err(fs_info, "no valid FS found"); 242121a852b0SQu Wenruo ret = -EINVAL; 242221a852b0SQu Wenruo } 242321a852b0SQu Wenruo if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { 242421a852b0SQu Wenruo btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", 242521a852b0SQu Wenruo btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); 242621a852b0SQu Wenruo ret = -EINVAL; 242721a852b0SQu Wenruo } 242821a852b0SQu Wenruo if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 242921a852b0SQu Wenruo btrfs_err(fs_info, "tree_root level too big: %d >= %d", 243021a852b0SQu Wenruo btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 243121a852b0SQu Wenruo ret = -EINVAL; 243221a852b0SQu Wenruo } 243321a852b0SQu Wenruo if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 243421a852b0SQu Wenruo btrfs_err(fs_info, "chunk_root level too big: %d >= %d", 243521a852b0SQu Wenruo btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 243621a852b0SQu Wenruo ret = -EINVAL; 243721a852b0SQu Wenruo } 243821a852b0SQu Wenruo if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 243921a852b0SQu Wenruo btrfs_err(fs_info, "log_root level too big: %d >= %d", 244021a852b0SQu Wenruo btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 244121a852b0SQu Wenruo ret = -EINVAL; 244221a852b0SQu Wenruo } 244321a852b0SQu Wenruo 244421a852b0SQu Wenruo /* 244521a852b0SQu Wenruo * Check sectorsize and nodesize first, other check will need it. 244621a852b0SQu Wenruo * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. 244721a852b0SQu Wenruo */ 244821a852b0SQu Wenruo if (!is_power_of_2(sectorsize) || sectorsize < 4096 || 244921a852b0SQu Wenruo sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { 245021a852b0SQu Wenruo btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); 245121a852b0SQu Wenruo ret = -EINVAL; 245221a852b0SQu Wenruo } 245321a852b0SQu Wenruo /* Only PAGE SIZE is supported yet */ 245421a852b0SQu Wenruo if (sectorsize != PAGE_SIZE) { 245521a852b0SQu Wenruo btrfs_err(fs_info, 245621a852b0SQu Wenruo "sectorsize %llu not supported yet, only support %lu", 245721a852b0SQu Wenruo sectorsize, PAGE_SIZE); 245821a852b0SQu Wenruo ret = -EINVAL; 245921a852b0SQu Wenruo } 246021a852b0SQu Wenruo if (!is_power_of_2(nodesize) || nodesize < sectorsize || 246121a852b0SQu Wenruo nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { 246221a852b0SQu Wenruo btrfs_err(fs_info, "invalid nodesize %llu", nodesize); 246321a852b0SQu Wenruo ret = -EINVAL; 246421a852b0SQu Wenruo } 246521a852b0SQu Wenruo if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { 246621a852b0SQu Wenruo btrfs_err(fs_info, "invalid leafsize %u, should be %llu", 246721a852b0SQu Wenruo le32_to_cpu(sb->__unused_leafsize), nodesize); 246821a852b0SQu Wenruo ret = -EINVAL; 246921a852b0SQu Wenruo } 247021a852b0SQu Wenruo 247121a852b0SQu Wenruo /* Root alignment check */ 247221a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { 247321a852b0SQu Wenruo btrfs_warn(fs_info, "tree_root block unaligned: %llu", 247421a852b0SQu Wenruo btrfs_super_root(sb)); 247521a852b0SQu Wenruo ret = -EINVAL; 247621a852b0SQu Wenruo } 247721a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { 247821a852b0SQu Wenruo btrfs_warn(fs_info, "chunk_root block unaligned: %llu", 247921a852b0SQu Wenruo btrfs_super_chunk_root(sb)); 248021a852b0SQu Wenruo ret = -EINVAL; 248121a852b0SQu Wenruo } 248221a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { 248321a852b0SQu Wenruo btrfs_warn(fs_info, "log_root block unaligned: %llu", 248421a852b0SQu Wenruo btrfs_super_log_root(sb)); 248521a852b0SQu Wenruo ret = -EINVAL; 248621a852b0SQu Wenruo } 248721a852b0SQu Wenruo 2488de37aa51SNikolay Borisov if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, 24897239ff4bSNikolay Borisov BTRFS_FSID_SIZE) != 0) { 249021a852b0SQu Wenruo btrfs_err(fs_info, 24917239ff4bSNikolay Borisov "dev_item UUID does not match metadata fsid: %pU != %pU", 2492de37aa51SNikolay Borisov fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); 249321a852b0SQu Wenruo ret = -EINVAL; 249421a852b0SQu Wenruo } 249521a852b0SQu Wenruo 249621a852b0SQu Wenruo /* 249721a852b0SQu Wenruo * Hint to catch really bogus numbers, bitflips or so, more exact checks are 249821a852b0SQu Wenruo * done later 249921a852b0SQu Wenruo */ 250021a852b0SQu Wenruo if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { 250121a852b0SQu Wenruo btrfs_err(fs_info, "bytes_used is too small %llu", 250221a852b0SQu Wenruo btrfs_super_bytes_used(sb)); 250321a852b0SQu Wenruo ret = -EINVAL; 250421a852b0SQu Wenruo } 250521a852b0SQu Wenruo if (!is_power_of_2(btrfs_super_stripesize(sb))) { 250621a852b0SQu Wenruo btrfs_err(fs_info, "invalid stripesize %u", 250721a852b0SQu Wenruo btrfs_super_stripesize(sb)); 250821a852b0SQu Wenruo ret = -EINVAL; 250921a852b0SQu Wenruo } 251021a852b0SQu Wenruo if (btrfs_super_num_devices(sb) > (1UL << 31)) 251121a852b0SQu Wenruo btrfs_warn(fs_info, "suspicious number of devices: %llu", 251221a852b0SQu Wenruo btrfs_super_num_devices(sb)); 251321a852b0SQu Wenruo if (btrfs_super_num_devices(sb) == 0) { 251421a852b0SQu Wenruo btrfs_err(fs_info, "number of devices is 0"); 251521a852b0SQu Wenruo ret = -EINVAL; 251621a852b0SQu Wenruo } 251721a852b0SQu Wenruo 2518069ec957SQu Wenruo if (mirror_num >= 0 && 2519069ec957SQu Wenruo btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { 252021a852b0SQu Wenruo btrfs_err(fs_info, "super offset mismatch %llu != %u", 252121a852b0SQu Wenruo btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 252221a852b0SQu Wenruo ret = -EINVAL; 252321a852b0SQu Wenruo } 252421a852b0SQu Wenruo 252521a852b0SQu Wenruo /* 252621a852b0SQu Wenruo * Obvious sys_chunk_array corruptions, it must hold at least one key 252721a852b0SQu Wenruo * and one chunk 252821a852b0SQu Wenruo */ 252921a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 253021a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too big %u > %u", 253121a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 253221a852b0SQu Wenruo BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 253321a852b0SQu Wenruo ret = -EINVAL; 253421a852b0SQu Wenruo } 253521a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 253621a852b0SQu Wenruo + sizeof(struct btrfs_chunk)) { 253721a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too small %u < %zu", 253821a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 253921a852b0SQu Wenruo sizeof(struct btrfs_disk_key) 254021a852b0SQu Wenruo + sizeof(struct btrfs_chunk)); 254121a852b0SQu Wenruo ret = -EINVAL; 254221a852b0SQu Wenruo } 254321a852b0SQu Wenruo 254421a852b0SQu Wenruo /* 254521a852b0SQu Wenruo * The generation is a global counter, we'll trust it more than the others 254621a852b0SQu Wenruo * but it's still possible that it's the one that's wrong. 254721a852b0SQu Wenruo */ 254821a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 254921a852b0SQu Wenruo btrfs_warn(fs_info, 255021a852b0SQu Wenruo "suspicious: generation < chunk_root_generation: %llu < %llu", 255121a852b0SQu Wenruo btrfs_super_generation(sb), 255221a852b0SQu Wenruo btrfs_super_chunk_root_generation(sb)); 255321a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 255421a852b0SQu Wenruo && btrfs_super_cache_generation(sb) != (u64)-1) 255521a852b0SQu Wenruo btrfs_warn(fs_info, 255621a852b0SQu Wenruo "suspicious: generation < cache_generation: %llu < %llu", 255721a852b0SQu Wenruo btrfs_super_generation(sb), 255821a852b0SQu Wenruo btrfs_super_cache_generation(sb)); 255921a852b0SQu Wenruo 256021a852b0SQu Wenruo return ret; 256121a852b0SQu Wenruo } 256221a852b0SQu Wenruo 2563069ec957SQu Wenruo /* 2564069ec957SQu Wenruo * Validation of super block at mount time. 2565069ec957SQu Wenruo * Some checks already done early at mount time, like csum type and incompat 2566069ec957SQu Wenruo * flags will be skipped. 2567069ec957SQu Wenruo */ 2568069ec957SQu Wenruo static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) 2569069ec957SQu Wenruo { 2570069ec957SQu Wenruo return validate_super(fs_info, fs_info->super_copy, 0); 2571069ec957SQu Wenruo } 2572069ec957SQu Wenruo 257375cb857dSQu Wenruo /* 257475cb857dSQu Wenruo * Validation of super block at write time. 257575cb857dSQu Wenruo * Some checks like bytenr check will be skipped as their values will be 257675cb857dSQu Wenruo * overwritten soon. 257775cb857dSQu Wenruo * Extra checks like csum type and incompat flags will be done here. 257875cb857dSQu Wenruo */ 257975cb857dSQu Wenruo static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, 258075cb857dSQu Wenruo struct btrfs_super_block *sb) 258175cb857dSQu Wenruo { 258275cb857dSQu Wenruo int ret; 258375cb857dSQu Wenruo 258475cb857dSQu Wenruo ret = validate_super(fs_info, sb, -1); 258575cb857dSQu Wenruo if (ret < 0) 258675cb857dSQu Wenruo goto out; 2587e7e16f48SJohannes Thumshirn if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) { 258875cb857dSQu Wenruo ret = -EUCLEAN; 258975cb857dSQu Wenruo btrfs_err(fs_info, "invalid csum type, has %u want %u", 259075cb857dSQu Wenruo btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); 259175cb857dSQu Wenruo goto out; 259275cb857dSQu Wenruo } 259375cb857dSQu Wenruo if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 259475cb857dSQu Wenruo ret = -EUCLEAN; 259575cb857dSQu Wenruo btrfs_err(fs_info, 259675cb857dSQu Wenruo "invalid incompat flags, has 0x%llx valid mask 0x%llx", 259775cb857dSQu Wenruo btrfs_super_incompat_flags(sb), 259875cb857dSQu Wenruo (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); 259975cb857dSQu Wenruo goto out; 260075cb857dSQu Wenruo } 260175cb857dSQu Wenruo out: 260275cb857dSQu Wenruo if (ret < 0) 260375cb857dSQu Wenruo btrfs_err(fs_info, 260475cb857dSQu Wenruo "super block corruption detected before writing it to disk"); 260575cb857dSQu Wenruo return ret; 260675cb857dSQu Wenruo } 260775cb857dSQu Wenruo 26086ef108ddSNikolay Borisov static int __cold init_tree_roots(struct btrfs_fs_info *fs_info) 2609b8522a1eSNikolay Borisov { 26106ef108ddSNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 2611b8522a1eSNikolay Borisov struct btrfs_super_block *sb = fs_info->super_copy; 2612b8522a1eSNikolay Borisov struct btrfs_root *tree_root = fs_info->tree_root; 2613b8522a1eSNikolay Borisov bool handle_error = false; 2614b8522a1eSNikolay Borisov int ret = 0; 2615b8522a1eSNikolay Borisov int i; 2616b8522a1eSNikolay Borisov 2617b8522a1eSNikolay Borisov for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 2618b8522a1eSNikolay Borisov u64 generation; 2619b8522a1eSNikolay Borisov int level; 2620b8522a1eSNikolay Borisov 2621b8522a1eSNikolay Borisov if (handle_error) { 2622b8522a1eSNikolay Borisov if (!IS_ERR(tree_root->node)) 2623b8522a1eSNikolay Borisov free_extent_buffer(tree_root->node); 2624b8522a1eSNikolay Borisov tree_root->node = NULL; 2625b8522a1eSNikolay Borisov 2626b8522a1eSNikolay Borisov if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) 2627b8522a1eSNikolay Borisov break; 2628b8522a1eSNikolay Borisov 2629b8522a1eSNikolay Borisov free_root_pointers(fs_info, 0); 2630b8522a1eSNikolay Borisov 2631b8522a1eSNikolay Borisov /* 2632b8522a1eSNikolay Borisov * Don't use the log in recovery mode, it won't be 2633b8522a1eSNikolay Borisov * valid 2634b8522a1eSNikolay Borisov */ 2635b8522a1eSNikolay Borisov btrfs_set_super_log_root(sb, 0); 2636b8522a1eSNikolay Borisov 2637b8522a1eSNikolay Borisov /* We can't trust the free space cache either */ 2638b8522a1eSNikolay Borisov btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 2639b8522a1eSNikolay Borisov 2640b8522a1eSNikolay Borisov ret = read_backup_root(fs_info, i); 26416ef108ddSNikolay Borisov backup_index = ret; 2642b8522a1eSNikolay Borisov if (ret < 0) 2643b8522a1eSNikolay Borisov return ret; 2644b8522a1eSNikolay Borisov } 2645b8522a1eSNikolay Borisov generation = btrfs_super_generation(sb); 2646b8522a1eSNikolay Borisov level = btrfs_super_root_level(sb); 2647b8522a1eSNikolay Borisov tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb), 2648b8522a1eSNikolay Borisov generation, level, NULL); 26490465337cSJosef Bacik if (IS_ERR(tree_root->node)) { 2650217f5004SNikolay Borisov handle_error = true; 2651b8522a1eSNikolay Borisov ret = PTR_ERR(tree_root->node); 26520465337cSJosef Bacik tree_root->node = NULL; 2653217f5004SNikolay Borisov btrfs_warn(fs_info, "couldn't read tree root"); 2654217f5004SNikolay Borisov continue; 2655b8522a1eSNikolay Borisov 2656217f5004SNikolay Borisov } else if (!extent_buffer_uptodate(tree_root->node)) { 2657217f5004SNikolay Borisov handle_error = true; 2658217f5004SNikolay Borisov ret = -EIO; 2659217f5004SNikolay Borisov btrfs_warn(fs_info, "error while reading tree root"); 2660b8522a1eSNikolay Borisov continue; 2661b8522a1eSNikolay Borisov } 2662b8522a1eSNikolay Borisov 2663b8522a1eSNikolay Borisov btrfs_set_root_node(&tree_root->root_item, tree_root->node); 2664b8522a1eSNikolay Borisov tree_root->commit_root = btrfs_root_node(tree_root); 2665b8522a1eSNikolay Borisov btrfs_set_root_refs(&tree_root->root_item, 1); 2666b8522a1eSNikolay Borisov 2667336a0d8dSNikolay Borisov /* 2668336a0d8dSNikolay Borisov * No need to hold btrfs_root::objectid_mutex since the fs 2669336a0d8dSNikolay Borisov * hasn't been fully initialised and we are the only user 2670336a0d8dSNikolay Borisov */ 2671b8522a1eSNikolay Borisov ret = btrfs_find_highest_objectid(tree_root, 2672b8522a1eSNikolay Borisov &tree_root->highest_objectid); 2673b8522a1eSNikolay Borisov if (ret < 0) { 2674b8522a1eSNikolay Borisov handle_error = true; 2675b8522a1eSNikolay Borisov continue; 2676b8522a1eSNikolay Borisov } 2677b8522a1eSNikolay Borisov 2678b8522a1eSNikolay Borisov ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 2679b8522a1eSNikolay Borisov 2680b8522a1eSNikolay Borisov ret = btrfs_read_roots(fs_info); 2681b8522a1eSNikolay Borisov if (ret < 0) { 2682b8522a1eSNikolay Borisov handle_error = true; 2683b8522a1eSNikolay Borisov continue; 2684b8522a1eSNikolay Borisov } 2685b8522a1eSNikolay Borisov 2686b8522a1eSNikolay Borisov /* All successful */ 2687b8522a1eSNikolay Borisov fs_info->generation = generation; 2688b8522a1eSNikolay Borisov fs_info->last_trans_committed = generation; 26896ef108ddSNikolay Borisov 26906ef108ddSNikolay Borisov /* Always begin writing backup roots after the one being used */ 26916ef108ddSNikolay Borisov if (backup_index < 0) { 26926ef108ddSNikolay Borisov fs_info->backup_root_index = 0; 26936ef108ddSNikolay Borisov } else { 26946ef108ddSNikolay Borisov fs_info->backup_root_index = backup_index + 1; 26956ef108ddSNikolay Borisov fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS; 26966ef108ddSNikolay Borisov } 2697b8522a1eSNikolay Borisov break; 2698b8522a1eSNikolay Borisov } 2699b8522a1eSNikolay Borisov 2700b8522a1eSNikolay Borisov return ret; 2701b8522a1eSNikolay Borisov } 2702b8522a1eSNikolay Borisov 27038260edbaSJosef Bacik void btrfs_init_fs_info(struct btrfs_fs_info *fs_info) 2704eb60ceacSChris Mason { 270576dda93cSYan, Zheng INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 2706f28491e0SJosef Bacik INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 27078fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 2708facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 270924bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 2710eb73c1b7SMiao Xie INIT_LIST_HEAD(&fs_info->delalloc_roots); 271111833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 2712eb73c1b7SMiao Xie spin_lock_init(&fs_info->delalloc_root_lock); 2713a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 271476dda93cSYan, Zheng spin_lock_init(&fs_info->fs_roots_radix_lock); 271524bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 27164cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 2717ceda0864SMiao Xie spin_lock_init(&fs_info->super_lock); 2718f28491e0SJosef Bacik spin_lock_init(&fs_info->buffer_lock); 271947ab2a6cSJosef Bacik spin_lock_init(&fs_info->unused_bgs_lock); 2720f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 2721d7c15171SZhao Lei mutex_init(&fs_info->unused_bg_unpin_mutex); 272267c5e7d4SFilipe Manana mutex_init(&fs_info->delete_unused_bgs_mutex); 27237585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 2724573bfb72SMiao Xie mutex_init(&fs_info->delalloc_root_mutex); 2725de98ced9SMiao Xie seqlock_init(&fs_info->profiles_lock); 272619c00ddcSChris Mason 27270b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 27286324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 2729f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 273047ab2a6cSJosef Bacik INIT_LIST_HEAD(&fs_info->unused_bgs); 2731bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 2732bd647ce3SJosef Bacik INIT_LIST_HEAD(&fs_info->allocated_roots); 27333fd63727SJosef Bacik INIT_LIST_HEAD(&fs_info->allocated_ebs); 27343fd63727SJosef Bacik spin_lock_init(&fs_info->eb_leak_lock); 2735bd647ce3SJosef Bacik #endif 2736c8bf1b67SDavid Sterba extent_map_tree_init(&fs_info->mapping_tree); 273766d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 273866d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 273966d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 274066d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 274166d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 274266d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 274366d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 2744ba2c4d4eSJosef Bacik btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, 2745ba2c4d4eSJosef Bacik BTRFS_BLOCK_RSV_DELREFS); 2746ba2c4d4eSJosef Bacik 2747771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 27484cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 27492fefd558SZhao Lei atomic_set(&fs_info->reada_works_cnt, 0); 2750034f784dSJosef Bacik atomic_set(&fs_info->nr_delayed_iputs, 0); 2751fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 275295ac567aSFilipe David Borba Manana fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 27539ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 27544cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 2755a5ed45f8SNikolay Borisov atomic64_set(&fs_info->free_chunk_space, 0); 2756f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 27578b87dc17SDavid Sterba fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 2758f8c269d7SDavid Sterba fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 275990519d66SArne Jansen /* readahead state */ 2760d0164adcSMel Gorman INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); 276190519d66SArne Jansen spin_lock_init(&fs_info->reada_lock); 2762fd708b81SJosef Bacik btrfs_init_ref_verify(fs_info); 2763c8b97818SChris Mason 2764b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 2765b34b086cSChris Mason num_online_cpus() + 2, 8); 27660afbaf8cSChris Mason 2767199c2a9cSMiao Xie INIT_LIST_HEAD(&fs_info->ordered_roots); 2768199c2a9cSMiao Xie spin_lock_init(&fs_info->ordered_root_lock); 276969fe2d75SJosef Bacik 2770638aa7edSEric Sandeen btrfs_init_scrub(fs_info); 277121adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 277221adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 277321adbd5cSStefan Behrens #endif 2774779a65a4SEric Sandeen btrfs_init_balance(fs_info); 277557056740SJosef Bacik btrfs_init_async_reclaim_work(fs_info); 2776a2de733cSArne Jansen 27770f9dd46cSJosef Bacik spin_lock_init(&fs_info->block_group_cache_lock); 27786bef4d31SEric Paris fs_info->block_group_cache_tree = RB_ROOT; 2779a1897fddSLiu Bo fs_info->first_logical_byte = (u64)-1; 27800f9dd46cSJosef Bacik 2781fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &fs_info->excluded_extents, 2782fe119a6eSNikolay Borisov IO_TREE_FS_EXCLUDED_EXTENTS, NULL); 2783afcdd129SJosef Bacik set_bit(BTRFS_FS_BARRIER, &fs_info->flags); 278439279cc3SChris Mason 27855a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 2786e02119d5SChris Mason mutex_init(&fs_info->tree_log_mutex); 2787925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 2788a74a4b97SChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 2789a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 27901bbc621eSChris Mason mutex_init(&fs_info->ro_block_group_mutex); 27919e351cc8SJosef Bacik init_rwsem(&fs_info->commit_root_sem); 2792c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 279376dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 2794803b2f54SStefan Behrens sema_init(&fs_info->uuid_tree_rescan_sem, 1); 2795fa9c0d79SChris Mason 2796ad618368SEric Sandeen btrfs_init_dev_replace_locks(fs_info); 2797f9e92e40SEric Sandeen btrfs_init_qgroup(fs_info); 2798b0643e59SDennis Zhou btrfs_discard_init(fs_info); 2799416ac51dSArne Jansen 2800fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2801fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2802fa9c0d79SChris Mason 2803e6dcd2dcSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 2804f9295749SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 2805bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 28064854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 2807034f784dSJosef Bacik init_waitqueue_head(&fs_info->delayed_iputs_wait); 28083768f368SChris Mason 2809da17066cSJeff Mahoney /* Usable values until the real ones are cached from the superblock */ 2810da17066cSJeff Mahoney fs_info->nodesize = 4096; 2811da17066cSJeff Mahoney fs_info->sectorsize = 4096; 2812ab108d99SDavid Sterba fs_info->sectorsize_bits = ilog2(4096); 2813da17066cSJeff Mahoney fs_info->stripesize = 4096; 2814da17066cSJeff Mahoney 2815eede2bf3SOmar Sandoval spin_lock_init(&fs_info->swapfile_pins_lock); 2816eede2bf3SOmar Sandoval fs_info->swapfile_pins = RB_ROOT; 2817eede2bf3SOmar Sandoval 28189e967495SFilipe Manana fs_info->send_in_progress = 0; 28198260edbaSJosef Bacik } 28208260edbaSJosef Bacik 28218260edbaSJosef Bacik static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb) 28228260edbaSJosef Bacik { 28238260edbaSJosef Bacik int ret; 28248260edbaSJosef Bacik 28258260edbaSJosef Bacik fs_info->sb = sb; 28268260edbaSJosef Bacik sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; 28278260edbaSJosef Bacik sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); 28289e967495SFilipe Manana 2829c75e8394SJosef Bacik ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL); 2830ae18c37aSJosef Bacik if (ret) 2831ae18c37aSJosef Bacik return ret; 2832ae18c37aSJosef Bacik 2833ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 2834ae18c37aSJosef Bacik if (ret) 2835c75e8394SJosef Bacik return ret; 2836ae18c37aSJosef Bacik 2837ae18c37aSJosef Bacik fs_info->dirty_metadata_batch = PAGE_SIZE * 2838ae18c37aSJosef Bacik (1 + ilog2(nr_cpu_ids)); 2839ae18c37aSJosef Bacik 2840ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 2841ae18c37aSJosef Bacik if (ret) 2842c75e8394SJosef Bacik return ret; 2843ae18c37aSJosef Bacik 2844ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, 2845ae18c37aSJosef Bacik GFP_KERNEL); 2846ae18c37aSJosef Bacik if (ret) 2847c75e8394SJosef Bacik return ret; 2848ae18c37aSJosef Bacik 2849ae18c37aSJosef Bacik fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 2850ae18c37aSJosef Bacik GFP_KERNEL); 2851c75e8394SJosef Bacik if (!fs_info->delayed_root) 2852c75e8394SJosef Bacik return -ENOMEM; 2853ae18c37aSJosef Bacik btrfs_init_delayed_root(fs_info->delayed_root); 2854ae18c37aSJosef Bacik 2855c75e8394SJosef Bacik return btrfs_alloc_stripe_hash_table(fs_info); 2856ae18c37aSJosef Bacik } 2857ae18c37aSJosef Bacik 285897f4dd09SNikolay Borisov static int btrfs_uuid_rescan_kthread(void *data) 285997f4dd09SNikolay Borisov { 286097f4dd09SNikolay Borisov struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; 286197f4dd09SNikolay Borisov int ret; 286297f4dd09SNikolay Borisov 286397f4dd09SNikolay Borisov /* 286497f4dd09SNikolay Borisov * 1st step is to iterate through the existing UUID tree and 286597f4dd09SNikolay Borisov * to delete all entries that contain outdated data. 286697f4dd09SNikolay Borisov * 2nd step is to add all missing entries to the UUID tree. 286797f4dd09SNikolay Borisov */ 286897f4dd09SNikolay Borisov ret = btrfs_uuid_tree_iterate(fs_info); 286997f4dd09SNikolay Borisov if (ret < 0) { 2870c94bec2cSJosef Bacik if (ret != -EINTR) 2871c94bec2cSJosef Bacik btrfs_warn(fs_info, "iterating uuid_tree failed %d", 2872c94bec2cSJosef Bacik ret); 287397f4dd09SNikolay Borisov up(&fs_info->uuid_tree_rescan_sem); 287497f4dd09SNikolay Borisov return ret; 287597f4dd09SNikolay Borisov } 287697f4dd09SNikolay Borisov return btrfs_uuid_scan_kthread(data); 287797f4dd09SNikolay Borisov } 287897f4dd09SNikolay Borisov 287997f4dd09SNikolay Borisov static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) 288097f4dd09SNikolay Borisov { 288197f4dd09SNikolay Borisov struct task_struct *task; 288297f4dd09SNikolay Borisov 288397f4dd09SNikolay Borisov down(&fs_info->uuid_tree_rescan_sem); 288497f4dd09SNikolay Borisov task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); 288597f4dd09SNikolay Borisov if (IS_ERR(task)) { 288697f4dd09SNikolay Borisov /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 288797f4dd09SNikolay Borisov btrfs_warn(fs_info, "failed to start uuid_rescan task"); 288897f4dd09SNikolay Borisov up(&fs_info->uuid_tree_rescan_sem); 288997f4dd09SNikolay Borisov return PTR_ERR(task); 289097f4dd09SNikolay Borisov } 289197f4dd09SNikolay Borisov 289297f4dd09SNikolay Borisov return 0; 289397f4dd09SNikolay Borisov } 289497f4dd09SNikolay Borisov 2895ae18c37aSJosef Bacik int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices, 2896ae18c37aSJosef Bacik char *options) 2897ae18c37aSJosef Bacik { 2898ae18c37aSJosef Bacik u32 sectorsize; 2899ae18c37aSJosef Bacik u32 nodesize; 2900ae18c37aSJosef Bacik u32 stripesize; 2901ae18c37aSJosef Bacik u64 generation; 2902ae18c37aSJosef Bacik u64 features; 2903ae18c37aSJosef Bacik u16 csum_type; 2904ae18c37aSJosef Bacik struct btrfs_super_block *disk_super; 2905ae18c37aSJosef Bacik struct btrfs_fs_info *fs_info = btrfs_sb(sb); 2906ae18c37aSJosef Bacik struct btrfs_root *tree_root; 2907ae18c37aSJosef Bacik struct btrfs_root *chunk_root; 2908ae18c37aSJosef Bacik int ret; 2909ae18c37aSJosef Bacik int err = -EINVAL; 2910ae18c37aSJosef Bacik int clear_free_space_tree = 0; 2911ae18c37aSJosef Bacik int level; 2912ae18c37aSJosef Bacik 29138260edbaSJosef Bacik ret = init_mount_fs_info(fs_info, sb); 291453b381b3SDavid Woodhouse if (ret) { 291583c8266aSDavid Sterba err = ret; 2916ae18c37aSJosef Bacik goto fail; 291753b381b3SDavid Woodhouse } 291853b381b3SDavid Woodhouse 2919ae18c37aSJosef Bacik /* These need to be init'ed before we start creating inodes and such. */ 2920ae18c37aSJosef Bacik tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, 2921ae18c37aSJosef Bacik GFP_KERNEL); 2922ae18c37aSJosef Bacik fs_info->tree_root = tree_root; 2923ae18c37aSJosef Bacik chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID, 2924ae18c37aSJosef Bacik GFP_KERNEL); 2925ae18c37aSJosef Bacik fs_info->chunk_root = chunk_root; 2926ae18c37aSJosef Bacik if (!tree_root || !chunk_root) { 2927ae18c37aSJosef Bacik err = -ENOMEM; 2928c75e8394SJosef Bacik goto fail; 2929ae18c37aSJosef Bacik } 2930ae18c37aSJosef Bacik 2931ae18c37aSJosef Bacik fs_info->btree_inode = new_inode(sb); 2932ae18c37aSJosef Bacik if (!fs_info->btree_inode) { 2933ae18c37aSJosef Bacik err = -ENOMEM; 2934c75e8394SJosef Bacik goto fail; 2935ae18c37aSJosef Bacik } 2936ae18c37aSJosef Bacik mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 2937ae18c37aSJosef Bacik btrfs_init_btree_inode(fs_info); 2938ae18c37aSJosef Bacik 29393c4bb26bSChris Mason invalidate_bdev(fs_devices->latest_bdev); 29401104a885SDavid Sterba 29411104a885SDavid Sterba /* 29421104a885SDavid Sterba * Read super block and check the signature bytes only 29431104a885SDavid Sterba */ 29448f32380dSJohannes Thumshirn disk_super = btrfs_read_dev_super(fs_devices->latest_bdev); 29458f32380dSJohannes Thumshirn if (IS_ERR(disk_super)) { 29468f32380dSJohannes Thumshirn err = PTR_ERR(disk_super); 294716cdcec7SMiao Xie goto fail_alloc; 294820b45077SDave Young } 294939279cc3SChris Mason 29501104a885SDavid Sterba /* 2951260db43cSRandy Dunlap * Verify the type first, if that or the checksum value are 29528dc3f22cSJohannes Thumshirn * corrupted, we'll find out 29538dc3f22cSJohannes Thumshirn */ 29548f32380dSJohannes Thumshirn csum_type = btrfs_super_csum_type(disk_super); 295551bce6c9SJohannes Thumshirn if (!btrfs_supported_super_csum(csum_type)) { 29568dc3f22cSJohannes Thumshirn btrfs_err(fs_info, "unsupported checksum algorithm: %u", 295751bce6c9SJohannes Thumshirn csum_type); 29588dc3f22cSJohannes Thumshirn err = -EINVAL; 29598f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 29608dc3f22cSJohannes Thumshirn goto fail_alloc; 29618dc3f22cSJohannes Thumshirn } 29628dc3f22cSJohannes Thumshirn 29636d97c6e3SJohannes Thumshirn ret = btrfs_init_csum_hash(fs_info, csum_type); 29646d97c6e3SJohannes Thumshirn if (ret) { 29656d97c6e3SJohannes Thumshirn err = ret; 29668f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 29676d97c6e3SJohannes Thumshirn goto fail_alloc; 29686d97c6e3SJohannes Thumshirn } 29696d97c6e3SJohannes Thumshirn 29708dc3f22cSJohannes Thumshirn /* 29711104a885SDavid Sterba * We want to check superblock checksum, the type is stored inside. 29721104a885SDavid Sterba * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 29731104a885SDavid Sterba */ 29748f32380dSJohannes Thumshirn if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) { 297505135f59SDavid Sterba btrfs_err(fs_info, "superblock checksum mismatch"); 29761104a885SDavid Sterba err = -EINVAL; 29778f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 2978141386e1SJosef Bacik goto fail_alloc; 29791104a885SDavid Sterba } 29801104a885SDavid Sterba 29811104a885SDavid Sterba /* 29821104a885SDavid Sterba * super_copy is zeroed at allocation time and we never touch the 29831104a885SDavid Sterba * following bytes up to INFO_SIZE, the checksum is calculated from 29841104a885SDavid Sterba * the whole block of INFO_SIZE 29851104a885SDavid Sterba */ 29868f32380dSJohannes Thumshirn memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy)); 29878f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 29885f39d397SChris Mason 2989fbc6feaeSNikolay Borisov disk_super = fs_info->super_copy; 2990fbc6feaeSNikolay Borisov 2991de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, 2992de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 2993de37aa51SNikolay Borisov 29947239ff4bSNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID)) { 2995de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid, 2996de37aa51SNikolay Borisov fs_info->super_copy->metadata_uuid, 2997de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 29987239ff4bSNikolay Borisov } 29990b86a832SChris Mason 3000fbc6feaeSNikolay Borisov features = btrfs_super_flags(disk_super); 3001fbc6feaeSNikolay Borisov if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { 3002fbc6feaeSNikolay Borisov features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; 3003fbc6feaeSNikolay Borisov btrfs_set_super_flags(disk_super, features); 3004fbc6feaeSNikolay Borisov btrfs_info(fs_info, 3005fbc6feaeSNikolay Borisov "found metadata UUID change in progress flag, clearing"); 3006fbc6feaeSNikolay Borisov } 3007fbc6feaeSNikolay Borisov 3008fbc6feaeSNikolay Borisov memcpy(fs_info->super_for_commit, fs_info->super_copy, 3009fbc6feaeSNikolay Borisov sizeof(*fs_info->super_for_commit)); 3010de37aa51SNikolay Borisov 3011069ec957SQu Wenruo ret = btrfs_validate_mount_super(fs_info); 30121104a885SDavid Sterba if (ret) { 301305135f59SDavid Sterba btrfs_err(fs_info, "superblock contains fatal errors"); 30141104a885SDavid Sterba err = -EINVAL; 3015141386e1SJosef Bacik goto fail_alloc; 30161104a885SDavid Sterba } 30171104a885SDavid Sterba 30180f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 3019141386e1SJosef Bacik goto fail_alloc; 30200f7d52f4SChris Mason 3021acce952bSliubo /* check FS state, whether FS is broken. */ 302287533c47SMiao Xie if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 302387533c47SMiao Xie set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 3024acce952bSliubo 302575e7cb7fSLiu Bo /* 302675e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 302775e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 302875e7cb7fSLiu Bo */ 302975e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 303075e7cb7fSLiu Bo 30312ff7e61eSJeff Mahoney ret = btrfs_parse_options(fs_info, options, sb->s_flags); 30322b82032cSYan Zheng if (ret) { 30332b82032cSYan Zheng err = ret; 3034141386e1SJosef Bacik goto fail_alloc; 30352b82032cSYan Zheng } 3036dfe25020SChris Mason 3037f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 3038f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 3039f2b636e8SJosef Bacik if (features) { 304005135f59SDavid Sterba btrfs_err(fs_info, 304105135f59SDavid Sterba "cannot mount because of unsupported optional features (%llx)", 3042c1c9ff7cSGeert Uytterhoeven features); 3043f2b636e8SJosef Bacik err = -EINVAL; 3044141386e1SJosef Bacik goto fail_alloc; 3045f2b636e8SJosef Bacik } 3046f2b636e8SJosef Bacik 30475d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 30485d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 30490b246afaSJeff Mahoney if (fs_info->compress_type == BTRFS_COMPRESS_LZO) 3050a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 30515c1aab1dSNick Terrell else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) 30525c1aab1dSNick Terrell features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; 3053727011e0SChris Mason 30543173a18fSJosef Bacik if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) 305505135f59SDavid Sterba btrfs_info(fs_info, "has skinny extents"); 30563173a18fSJosef Bacik 3057727011e0SChris Mason /* 3058727011e0SChris Mason * flag our filesystem as having big metadata blocks if 3059727011e0SChris Mason * they are bigger than the page size 3060727011e0SChris Mason */ 306109cbfeafSKirill A. Shutemov if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) { 3062727011e0SChris Mason if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 306305135f59SDavid Sterba btrfs_info(fs_info, 306405135f59SDavid Sterba "flagging fs with big metadata feature"); 3065727011e0SChris Mason features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 3066727011e0SChris Mason } 3067727011e0SChris Mason 3068bc3f116fSChris Mason nodesize = btrfs_super_nodesize(disk_super); 3069bc3f116fSChris Mason sectorsize = btrfs_super_sectorsize(disk_super); 3070b7f67055SChandan Rajendra stripesize = sectorsize; 3071707e8a07SDavid Sterba fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 3072963d678bSMiao Xie fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 3073bc3f116fSChris Mason 3074da17066cSJeff Mahoney /* Cache block sizes */ 3075da17066cSJeff Mahoney fs_info->nodesize = nodesize; 3076da17066cSJeff Mahoney fs_info->sectorsize = sectorsize; 3077ab108d99SDavid Sterba fs_info->sectorsize_bits = ilog2(sectorsize); 307822b6331dSDavid Sterba fs_info->csum_size = btrfs_super_csum_size(disk_super); 3079fe5ecbe8SDavid Sterba fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size; 3080da17066cSJeff Mahoney fs_info->stripesize = stripesize; 3081da17066cSJeff Mahoney 3082bc3f116fSChris Mason /* 3083bc3f116fSChris Mason * mixed block groups end up with duplicate but slightly offset 3084bc3f116fSChris Mason * extent buffers for the same range. It leads to corruptions 3085bc3f116fSChris Mason */ 3086bc3f116fSChris Mason if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 3087707e8a07SDavid Sterba (sectorsize != nodesize)) { 308805135f59SDavid Sterba btrfs_err(fs_info, 308905135f59SDavid Sterba "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", 309005135f59SDavid Sterba nodesize, sectorsize); 3091141386e1SJosef Bacik goto fail_alloc; 3092bc3f116fSChris Mason } 3093bc3f116fSChris Mason 3094ceda0864SMiao Xie /* 3095ceda0864SMiao Xie * Needn't use the lock because there is no other task which will 3096ceda0864SMiao Xie * update the flag. 3097ceda0864SMiao Xie */ 30985d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 30995d4f98a2SYan Zheng 3100f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 3101f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 3102bc98a42cSDavid Howells if (!sb_rdonly(sb) && features) { 310305135f59SDavid Sterba btrfs_err(fs_info, 310405135f59SDavid Sterba "cannot mount read-write because of unsupported optional features (%llx)", 3105c1c9ff7cSGeert Uytterhoeven features); 3106f2b636e8SJosef Bacik err = -EINVAL; 3107141386e1SJosef Bacik goto fail_alloc; 3108f2b636e8SJosef Bacik } 310961d92c32SChris Mason 31102a458198SEric Sandeen ret = btrfs_init_workqueues(fs_info, fs_devices); 31112a458198SEric Sandeen if (ret) { 31122a458198SEric Sandeen err = ret; 31130dc3b84aSJosef Bacik goto fail_sb_buffer; 31140dc3b84aSJosef Bacik } 31154543df7eSChris Mason 31169e11ceeeSJan Kara sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); 31179e11ceeeSJan Kara sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); 31184575c9ccSChris Mason 3119a061fc8dSChris Mason sb->s_blocksize = sectorsize; 3120a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 3121de37aa51SNikolay Borisov memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); 3122db94535dSChris Mason 3123925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 31246bccf3abSJeff Mahoney ret = btrfs_read_sys_array(fs_info); 3125925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 312684eed90fSChris Mason if (ret) { 312705135f59SDavid Sterba btrfs_err(fs_info, "failed to read the system array: %d", ret); 31285d4f98a2SYan Zheng goto fail_sb_buffer; 312984eed90fSChris Mason } 31300b86a832SChris Mason 313184234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 3132581c1760SQu Wenruo level = btrfs_super_chunk_root_level(disk_super); 31330b86a832SChris Mason 31342ff7e61eSJeff Mahoney chunk_root->node = read_tree_block(fs_info, 31350b86a832SChris Mason btrfs_super_chunk_root(disk_super), 3136581c1760SQu Wenruo generation, level, NULL); 313764c043deSLiu Bo if (IS_ERR(chunk_root->node) || 313864c043deSLiu Bo !extent_buffer_uptodate(chunk_root->node)) { 313905135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk root"); 3140e5fffbacSchandan if (!IS_ERR(chunk_root->node)) 3141e5fffbacSchandan free_extent_buffer(chunk_root->node); 314295ab1f64SZhao Lei chunk_root->node = NULL; 3143af31f5e5SChris Mason goto fail_tree_roots; 314483121942SDavid Woodhouse } 31455d4f98a2SYan Zheng btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 31465d4f98a2SYan Zheng chunk_root->commit_root = btrfs_root_node(chunk_root); 31470b86a832SChris Mason 3148e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 3149c4ac7541SDavid Sterba offsetof(struct btrfs_header, chunk_tree_uuid), 3150c4ac7541SDavid Sterba BTRFS_UUID_SIZE); 3151e17cade2SChris Mason 31525b4aacefSJeff Mahoney ret = btrfs_read_chunk_tree(fs_info); 31532b82032cSYan Zheng if (ret) { 315405135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk tree: %d", ret); 3155af31f5e5SChris Mason goto fail_tree_roots; 31562b82032cSYan Zheng } 31570b86a832SChris Mason 31588dabb742SStefan Behrens /* 31599b99b115SAnand Jain * Keep the devid that is marked to be the target device for the 31609b99b115SAnand Jain * device replace procedure 31618dabb742SStefan Behrens */ 31629b99b115SAnand Jain btrfs_free_extra_devids(fs_devices, 0); 3163dfe25020SChris Mason 3164a6b0d5c8SChris Mason if (!fs_devices->latest_bdev) { 316505135f59SDavid Sterba btrfs_err(fs_info, "failed to read devices"); 3166a6b0d5c8SChris Mason goto fail_tree_roots; 3167a6b0d5c8SChris Mason } 3168a6b0d5c8SChris Mason 3169b8522a1eSNikolay Borisov ret = init_tree_roots(fs_info); 31704bbcaa64SEric Sandeen if (ret) 3171b8522a1eSNikolay Borisov goto fail_tree_roots; 31728929ecfaSYan, Zheng 317375ec1db8SJosef Bacik /* 317475ec1db8SJosef Bacik * If we have a uuid root and we're not being told to rescan we need to 317575ec1db8SJosef Bacik * check the generation here so we can set the 317675ec1db8SJosef Bacik * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the 317775ec1db8SJosef Bacik * transaction during a balance or the log replay without updating the 317875ec1db8SJosef Bacik * uuid generation, and then if we crash we would rescan the uuid tree, 317975ec1db8SJosef Bacik * even though it was perfectly fine. 318075ec1db8SJosef Bacik */ 318175ec1db8SJosef Bacik if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) && 318275ec1db8SJosef Bacik fs_info->generation == btrfs_super_uuid_tree_generation(disk_super)) 318375ec1db8SJosef Bacik set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 318475ec1db8SJosef Bacik 3185cf90d884SQu Wenruo ret = btrfs_verify_dev_extents(fs_info); 3186cf90d884SQu Wenruo if (ret) { 3187cf90d884SQu Wenruo btrfs_err(fs_info, 3188cf90d884SQu Wenruo "failed to verify dev extents against chunks: %d", 3189cf90d884SQu Wenruo ret); 3190cf90d884SQu Wenruo goto fail_block_groups; 3191cf90d884SQu Wenruo } 319268310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 319368310a5eSIlya Dryomov if (ret) { 319405135f59SDavid Sterba btrfs_err(fs_info, "failed to recover balance: %d", ret); 319568310a5eSIlya Dryomov goto fail_block_groups; 319668310a5eSIlya Dryomov } 319768310a5eSIlya Dryomov 3198733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 3199733f4fbbSStefan Behrens if (ret) { 320005135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_stats: %d", ret); 3201733f4fbbSStefan Behrens goto fail_block_groups; 3202733f4fbbSStefan Behrens } 3203733f4fbbSStefan Behrens 32048dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 32058dabb742SStefan Behrens if (ret) { 320605135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_replace: %d", ret); 32078dabb742SStefan Behrens goto fail_block_groups; 32088dabb742SStefan Behrens } 32098dabb742SStefan Behrens 32109b99b115SAnand Jain btrfs_free_extra_devids(fs_devices, 1); 32118dabb742SStefan Behrens 3212c6761a9eSAnand Jain ret = btrfs_sysfs_add_fsid(fs_devices); 3213b7c35e81SAnand Jain if (ret) { 321405135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", 321505135f59SDavid Sterba ret); 3216b7c35e81SAnand Jain goto fail_block_groups; 3217b7c35e81SAnand Jain } 3218b7c35e81SAnand Jain 321996f3136eSAnand Jain ret = btrfs_sysfs_add_mounted(fs_info); 32205ac1d209SJeff Mahoney if (ret) { 322105135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); 3222b7c35e81SAnand Jain goto fail_fsdev_sysfs; 32235ac1d209SJeff Mahoney } 32245ac1d209SJeff Mahoney 3225c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 3226c59021f8Sliubo if (ret) { 322705135f59SDavid Sterba btrfs_err(fs_info, "failed to initialize space info: %d", ret); 32282365dd3cSAnand Jain goto fail_sysfs; 3229c59021f8Sliubo } 3230c59021f8Sliubo 32315b4aacefSJeff Mahoney ret = btrfs_read_block_groups(fs_info); 32321b1d1f66SJosef Bacik if (ret) { 323305135f59SDavid Sterba btrfs_err(fs_info, "failed to read block groups: %d", ret); 32342365dd3cSAnand Jain goto fail_sysfs; 32351b1d1f66SJosef Bacik } 32364330e183SQu Wenruo 32376528b99dSAnand Jain if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) { 323805135f59SDavid Sterba btrfs_warn(fs_info, 323952042d8eSAndrea Gelmini "writable mount is not allowed due to too many missing devices"); 32402365dd3cSAnand Jain goto fail_sysfs; 3241292fd7fcSStefan Behrens } 32429078a3e1SChris Mason 3243a74a4b97SChris Mason fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 3244a74a4b97SChris Mason "btrfs-cleaner"); 324557506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 32462365dd3cSAnand Jain goto fail_sysfs; 3247a74a4b97SChris Mason 3248a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 3249a74a4b97SChris Mason tree_root, 3250a74a4b97SChris Mason "btrfs-transaction"); 325157506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 32523f157a2fSChris Mason goto fail_cleaner; 3253a74a4b97SChris Mason 3254583b7231SHans van Kranenburg if (!btrfs_test_opt(fs_info, NOSSD) && 3255c289811cSChris Mason !fs_info->fs_devices->rotating) { 3256583b7231SHans van Kranenburg btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); 3257c289811cSChris Mason } 3258c289811cSChris Mason 3259572d9ab7SDavid Sterba /* 326001327610SNicholas D Steeves * Mount does not set all options immediately, we can do it now and do 3261572d9ab7SDavid Sterba * not have to wait for transaction commit 3262572d9ab7SDavid Sterba */ 3263572d9ab7SDavid Sterba btrfs_apply_pending_changes(fs_info); 32643818aea2SQu Wenruo 326521adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 32660b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { 32672ff7e61eSJeff Mahoney ret = btrfsic_mount(fs_info, fs_devices, 32680b246afaSJeff Mahoney btrfs_test_opt(fs_info, 326921adbd5cSStefan Behrens CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? 327021adbd5cSStefan Behrens 1 : 0, 327121adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 327221adbd5cSStefan Behrens if (ret) 327305135f59SDavid Sterba btrfs_warn(fs_info, 327405135f59SDavid Sterba "failed to initialize integrity check module: %d", 327505135f59SDavid Sterba ret); 327621adbd5cSStefan Behrens } 327721adbd5cSStefan Behrens #endif 3278bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 3279bcef60f2SArne Jansen if (ret) 3280bcef60f2SArne Jansen goto fail_trans_kthread; 328121adbd5cSStefan Behrens 3282fd708b81SJosef Bacik if (btrfs_build_ref_tree(fs_info)) 3283fd708b81SJosef Bacik btrfs_err(fs_info, "couldn't build ref tree"); 3284fd708b81SJosef Bacik 328596da0919SQu Wenruo /* do not make disk changes in broken FS or nologreplay is given */ 328696da0919SQu Wenruo if (btrfs_super_log_root(disk_super) != 0 && 32870b246afaSJeff Mahoney !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 3288e8294f2fSDavid Sterba btrfs_info(fs_info, "start tree-log replay"); 328963443bf5SEric Sandeen ret = btrfs_replay_log(fs_info, fs_devices); 329079787eaaSJeff Mahoney if (ret) { 329163443bf5SEric Sandeen err = ret; 329228c16cbbSWang Shilong goto fail_qgroup; 3293e556ce2cSYan Zheng } 3294e02119d5SChris Mason } 32951a40e23bSZheng Yan 32966bccf3abSJeff Mahoney ret = btrfs_find_orphan_roots(fs_info); 329779787eaaSJeff Mahoney if (ret) 329828c16cbbSWang Shilong goto fail_qgroup; 329976dda93cSYan, Zheng 3300bc98a42cSDavid Howells if (!sb_rdonly(sb)) { 3301d68fc57bSYan, Zheng ret = btrfs_cleanup_fs_roots(fs_info); 330244c44af2SIlya Dryomov if (ret) 330328c16cbbSWang Shilong goto fail_qgroup; 330490c711abSZygo Blaxell 330590c711abSZygo Blaxell mutex_lock(&fs_info->cleaner_mutex); 33065d4f98a2SYan Zheng ret = btrfs_recover_relocation(tree_root); 330790c711abSZygo Blaxell mutex_unlock(&fs_info->cleaner_mutex); 3308d7ce5843SMiao Xie if (ret < 0) { 330905135f59SDavid Sterba btrfs_warn(fs_info, "failed to recover relocation: %d", 331005135f59SDavid Sterba ret); 3311d7ce5843SMiao Xie err = -EINVAL; 3312bcef60f2SArne Jansen goto fail_qgroup; 3313d7ce5843SMiao Xie } 33147c2ca468SChris Mason } 33151a40e23bSZheng Yan 331656e9357aSDavid Sterba fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true); 33173140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 33183140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 3319f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read fs tree: %d", err); 3320315bf8efSJosef Bacik fs_info->fs_root = NULL; 3321bcef60f2SArne Jansen goto fail_qgroup; 33223140c9a3SDan Carpenter } 3323c289811cSChris Mason 3324bc98a42cSDavid Howells if (sb_rdonly(sb)) 33252b6ba629SIlya Dryomov return 0; 33262b6ba629SIlya Dryomov 3327f8d468a1SOmar Sandoval if (btrfs_test_opt(fs_info, CLEAR_CACHE) && 3328f8d468a1SOmar Sandoval btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 33296675df31SOmar Sandoval clear_free_space_tree = 1; 33306675df31SOmar Sandoval } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 33316675df31SOmar Sandoval !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { 33326675df31SOmar Sandoval btrfs_warn(fs_info, "free space tree is invalid"); 33336675df31SOmar Sandoval clear_free_space_tree = 1; 33346675df31SOmar Sandoval } 33356675df31SOmar Sandoval 33366675df31SOmar Sandoval if (clear_free_space_tree) { 3337f8d468a1SOmar Sandoval btrfs_info(fs_info, "clearing free space tree"); 3338f8d468a1SOmar Sandoval ret = btrfs_clear_free_space_tree(fs_info); 3339f8d468a1SOmar Sandoval if (ret) { 3340f8d468a1SOmar Sandoval btrfs_warn(fs_info, 3341f8d468a1SOmar Sandoval "failed to clear free space tree: %d", ret); 33426bccf3abSJeff Mahoney close_ctree(fs_info); 3343f8d468a1SOmar Sandoval return ret; 3344f8d468a1SOmar Sandoval } 3345f8d468a1SOmar Sandoval } 3346f8d468a1SOmar Sandoval 33470b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && 3348511711afSChris Mason !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 334905135f59SDavid Sterba btrfs_info(fs_info, "creating free space tree"); 3350511711afSChris Mason ret = btrfs_create_free_space_tree(fs_info); 3351511711afSChris Mason if (ret) { 335205135f59SDavid Sterba btrfs_warn(fs_info, 335305135f59SDavid Sterba "failed to create free space tree: %d", ret); 33546bccf3abSJeff Mahoney close_ctree(fs_info); 3355511711afSChris Mason return ret; 3356511711afSChris Mason } 3357511711afSChris Mason } 3358511711afSChris Mason 3359e3acc2a6SJosef Bacik down_read(&fs_info->cleanup_work_sem); 33602b6ba629SIlya Dryomov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 33612b6ba629SIlya Dryomov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 33622b6ba629SIlya Dryomov up_read(&fs_info->cleanup_work_sem); 33636bccf3abSJeff Mahoney close_ctree(fs_info); 33642b6ba629SIlya Dryomov return ret; 33652b6ba629SIlya Dryomov } 3366e3acc2a6SJosef Bacik up_read(&fs_info->cleanup_work_sem); 336759641015SIlya Dryomov 33682b6ba629SIlya Dryomov ret = btrfs_resume_balance_async(fs_info); 33692b6ba629SIlya Dryomov if (ret) { 337005135f59SDavid Sterba btrfs_warn(fs_info, "failed to resume balance: %d", ret); 33716bccf3abSJeff Mahoney close_ctree(fs_info); 33722b6ba629SIlya Dryomov return ret; 3373e3acc2a6SJosef Bacik } 3374e3acc2a6SJosef Bacik 33758dabb742SStefan Behrens ret = btrfs_resume_dev_replace_async(fs_info); 33768dabb742SStefan Behrens if (ret) { 337705135f59SDavid Sterba btrfs_warn(fs_info, "failed to resume device replace: %d", ret); 33786bccf3abSJeff Mahoney close_ctree(fs_info); 33798dabb742SStefan Behrens return ret; 33808dabb742SStefan Behrens } 33818dabb742SStefan Behrens 3382b382a324SJan Schmidt btrfs_qgroup_rescan_resume(fs_info); 3383b0643e59SDennis Zhou btrfs_discard_resume(fs_info); 3384b382a324SJan Schmidt 33854bbcaa64SEric Sandeen if (!fs_info->uuid_root) { 338605135f59SDavid Sterba btrfs_info(fs_info, "creating UUID tree"); 3387f7a81ea4SStefan Behrens ret = btrfs_create_uuid_tree(fs_info); 3388f7a81ea4SStefan Behrens if (ret) { 338905135f59SDavid Sterba btrfs_warn(fs_info, 339005135f59SDavid Sterba "failed to create the UUID tree: %d", ret); 33916bccf3abSJeff Mahoney close_ctree(fs_info); 3392f7a81ea4SStefan Behrens return ret; 3393f7a81ea4SStefan Behrens } 33940b246afaSJeff Mahoney } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || 33954bbcaa64SEric Sandeen fs_info->generation != 33964bbcaa64SEric Sandeen btrfs_super_uuid_tree_generation(disk_super)) { 339705135f59SDavid Sterba btrfs_info(fs_info, "checking UUID tree"); 339870f80175SStefan Behrens ret = btrfs_check_uuid_tree(fs_info); 339970f80175SStefan Behrens if (ret) { 340005135f59SDavid Sterba btrfs_warn(fs_info, 340105135f59SDavid Sterba "failed to check the UUID tree: %d", ret); 34026bccf3abSJeff Mahoney close_ctree(fs_info); 340370f80175SStefan Behrens return ret; 340470f80175SStefan Behrens } 3405f7a81ea4SStefan Behrens } 3406afcdd129SJosef Bacik set_bit(BTRFS_FS_OPEN, &fs_info->flags); 340747ab2a6cSJosef Bacik 3408ad2b2c80SAl Viro return 0; 340939279cc3SChris Mason 3410bcef60f2SArne Jansen fail_qgroup: 3411bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 34127c2ca468SChris Mason fail_trans_kthread: 34137c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 34142ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 3415faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 34163f157a2fSChris Mason fail_cleaner: 3417a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 34187c2ca468SChris Mason 34197c2ca468SChris Mason /* 34207c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 34217c2ca468SChris Mason * kthreads 34227c2ca468SChris Mason */ 34237c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 34247c2ca468SChris Mason 34252365dd3cSAnand Jain fail_sysfs: 34266618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 34272365dd3cSAnand Jain 3428b7c35e81SAnand Jain fail_fsdev_sysfs: 3429b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3430b7c35e81SAnand Jain 34311b1d1f66SJosef Bacik fail_block_groups: 343254067ae9SJosef Bacik btrfs_put_block_group_cache(fs_info); 3433af31f5e5SChris Mason 3434af31f5e5SChris Mason fail_tree_roots: 34359e3aa805SJosef Bacik if (fs_info->data_reloc_root) 34369e3aa805SJosef Bacik btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root); 34374273eaffSAnand Jain free_root_pointers(fs_info, true); 34382b8195bbSMiao Xie invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3439af31f5e5SChris Mason 344039279cc3SChris Mason fail_sb_buffer: 34417abadb64SLiu Bo btrfs_stop_all_workers(fs_info); 34425cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 344316cdcec7SMiao Xie fail_alloc: 3444586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 3445586e46e2SIlya Dryomov 34464543df7eSChris Mason iput(fs_info->btree_inode); 34477e662854SQinghuang Feng fail: 3448586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 3449ad2b2c80SAl Viro return err; 3450eb60ceacSChris Mason } 3451663faf9fSMasami Hiramatsu ALLOW_ERROR_INJECTION(open_ctree, ERRNO); 3452eb60ceacSChris Mason 3453314b6dd0SJohannes Thumshirn static void btrfs_end_super_write(struct bio *bio) 3454f2984462SChris Mason { 3455314b6dd0SJohannes Thumshirn struct btrfs_device *device = bio->bi_private; 3456314b6dd0SJohannes Thumshirn struct bio_vec *bvec; 3457314b6dd0SJohannes Thumshirn struct bvec_iter_all iter_all; 3458314b6dd0SJohannes Thumshirn struct page *page; 3459442a4f63SStefan Behrens 3460314b6dd0SJohannes Thumshirn bio_for_each_segment_all(bvec, bio, iter_all) { 3461314b6dd0SJohannes Thumshirn page = bvec->bv_page; 3462314b6dd0SJohannes Thumshirn 3463314b6dd0SJohannes Thumshirn if (bio->bi_status) { 3464fb456252SJeff Mahoney btrfs_warn_rl_in_rcu(device->fs_info, 3465314b6dd0SJohannes Thumshirn "lost page write due to IO error on %s (%d)", 3466314b6dd0SJohannes Thumshirn rcu_str_deref(device->name), 3467314b6dd0SJohannes Thumshirn blk_status_to_errno(bio->bi_status)); 3468314b6dd0SJohannes Thumshirn ClearPageUptodate(page); 3469314b6dd0SJohannes Thumshirn SetPageError(page); 3470314b6dd0SJohannes Thumshirn btrfs_dev_stat_inc_and_print(device, 3471314b6dd0SJohannes Thumshirn BTRFS_DEV_STAT_WRITE_ERRS); 3472314b6dd0SJohannes Thumshirn } else { 3473314b6dd0SJohannes Thumshirn SetPageUptodate(page); 3474f2984462SChris Mason } 3475314b6dd0SJohannes Thumshirn 3476314b6dd0SJohannes Thumshirn put_page(page); 3477314b6dd0SJohannes Thumshirn unlock_page(page); 3478314b6dd0SJohannes Thumshirn } 3479314b6dd0SJohannes Thumshirn 3480314b6dd0SJohannes Thumshirn bio_put(bio); 3481f2984462SChris Mason } 3482f2984462SChris Mason 34838f32380dSJohannes Thumshirn struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev, 34848f32380dSJohannes Thumshirn int copy_num) 348529c36d72SAnand Jain { 348629c36d72SAnand Jain struct btrfs_super_block *super; 34878f32380dSJohannes Thumshirn struct page *page; 348829c36d72SAnand Jain u64 bytenr; 34898f32380dSJohannes Thumshirn struct address_space *mapping = bdev->bd_inode->i_mapping; 349029c36d72SAnand Jain 349129c36d72SAnand Jain bytenr = btrfs_sb_offset(copy_num); 349229c36d72SAnand Jain if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode)) 34938f32380dSJohannes Thumshirn return ERR_PTR(-EINVAL); 349429c36d72SAnand Jain 34958f32380dSJohannes Thumshirn page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS); 34968f32380dSJohannes Thumshirn if (IS_ERR(page)) 34978f32380dSJohannes Thumshirn return ERR_CAST(page); 349829c36d72SAnand Jain 34998f32380dSJohannes Thumshirn super = page_address(page); 350096c2e067SAnand Jain if (btrfs_super_magic(super) != BTRFS_MAGIC) { 350196c2e067SAnand Jain btrfs_release_disk_super(super); 350296c2e067SAnand Jain return ERR_PTR(-ENODATA); 350396c2e067SAnand Jain } 350496c2e067SAnand Jain 350596c2e067SAnand Jain if (btrfs_super_bytenr(super) != bytenr) { 35068f32380dSJohannes Thumshirn btrfs_release_disk_super(super); 35078f32380dSJohannes Thumshirn return ERR_PTR(-EINVAL); 350829c36d72SAnand Jain } 350929c36d72SAnand Jain 35108f32380dSJohannes Thumshirn return super; 351129c36d72SAnand Jain } 351229c36d72SAnand Jain 351329c36d72SAnand Jain 35148f32380dSJohannes Thumshirn struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev) 3515a512bbf8SYan Zheng { 35168f32380dSJohannes Thumshirn struct btrfs_super_block *super, *latest = NULL; 3517a512bbf8SYan Zheng int i; 3518a512bbf8SYan Zheng u64 transid = 0; 3519a512bbf8SYan Zheng 3520a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 3521a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 3522a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 3523a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3524a512bbf8SYan Zheng */ 3525a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 35268f32380dSJohannes Thumshirn super = btrfs_read_dev_one_super(bdev, i); 35278f32380dSJohannes Thumshirn if (IS_ERR(super)) 3528a512bbf8SYan Zheng continue; 3529a512bbf8SYan Zheng 3530a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 35318f32380dSJohannes Thumshirn if (latest) 35328f32380dSJohannes Thumshirn btrfs_release_disk_super(super); 35338f32380dSJohannes Thumshirn 35348f32380dSJohannes Thumshirn latest = super; 3535a512bbf8SYan Zheng transid = btrfs_super_generation(super); 3536a512bbf8SYan Zheng } 3537a512bbf8SYan Zheng } 353892fc03fbSAnand Jain 35398f32380dSJohannes Thumshirn return super; 3540a512bbf8SYan Zheng } 3541a512bbf8SYan Zheng 35424eedeb75SHisashi Hifumi /* 3543abbb3b8eSDavid Sterba * Write superblock @sb to the @device. Do not wait for completion, all the 3544314b6dd0SJohannes Thumshirn * pages we use for writing are locked. 35454eedeb75SHisashi Hifumi * 3546abbb3b8eSDavid Sterba * Write @max_mirrors copies of the superblock, where 0 means default that fit 3547abbb3b8eSDavid Sterba * the expected device size at commit time. Note that max_mirrors must be 3548abbb3b8eSDavid Sterba * same for write and wait phases. 35494eedeb75SHisashi Hifumi * 3550314b6dd0SJohannes Thumshirn * Return number of errors when page is not found or submission fails. 35514eedeb75SHisashi Hifumi */ 3552a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 3553abbb3b8eSDavid Sterba struct btrfs_super_block *sb, int max_mirrors) 3554a512bbf8SYan Zheng { 3555d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = device->fs_info; 3556314b6dd0SJohannes Thumshirn struct address_space *mapping = device->bdev->bd_inode->i_mapping; 3557d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 3558a512bbf8SYan Zheng int i; 3559a512bbf8SYan Zheng int errors = 0; 3560a512bbf8SYan Zheng u64 bytenr; 3561a512bbf8SYan Zheng 3562a512bbf8SYan Zheng if (max_mirrors == 0) 3563a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3564a512bbf8SYan Zheng 3565d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 3566d5178578SJohannes Thumshirn 3567a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 3568314b6dd0SJohannes Thumshirn struct page *page; 3569314b6dd0SJohannes Thumshirn struct bio *bio; 3570314b6dd0SJohannes Thumshirn struct btrfs_super_block *disk_super; 3571314b6dd0SJohannes Thumshirn 3572a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 3573935e5cc9SMiao Xie if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3574935e5cc9SMiao Xie device->commit_total_bytes) 3575a512bbf8SYan Zheng break; 3576a512bbf8SYan Zheng 3577a512bbf8SYan Zheng btrfs_set_super_bytenr(sb, bytenr); 3578a512bbf8SYan Zheng 3579fd08001fSEric Biggers crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE, 3580fd08001fSEric Biggers BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, 3581fd08001fSEric Biggers sb->csum); 3582a512bbf8SYan Zheng 3583314b6dd0SJohannes Thumshirn page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT, 3584314b6dd0SJohannes Thumshirn GFP_NOFS); 3585314b6dd0SJohannes Thumshirn if (!page) { 3586fb456252SJeff Mahoney btrfs_err(device->fs_info, 3587314b6dd0SJohannes Thumshirn "couldn't get super block page for bytenr %llu", 3588f14d104dSDavid Sterba bytenr); 3589634554dcSJosef Bacik errors++; 3590634554dcSJosef Bacik continue; 3591634554dcSJosef Bacik } 3592634554dcSJosef Bacik 3593314b6dd0SJohannes Thumshirn /* Bump the refcount for wait_dev_supers() */ 3594314b6dd0SJohannes Thumshirn get_page(page); 3595a512bbf8SYan Zheng 3596314b6dd0SJohannes Thumshirn disk_super = page_address(page); 3597314b6dd0SJohannes Thumshirn memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE); 3598a512bbf8SYan Zheng 3599387125fcSChris Mason /* 3600314b6dd0SJohannes Thumshirn * Directly use bios here instead of relying on the page cache 3601314b6dd0SJohannes Thumshirn * to do I/O, so we don't lose the ability to do integrity 3602314b6dd0SJohannes Thumshirn * checking. 3603387125fcSChris Mason */ 3604314b6dd0SJohannes Thumshirn bio = bio_alloc(GFP_NOFS, 1); 3605314b6dd0SJohannes Thumshirn bio_set_dev(bio, device->bdev); 3606314b6dd0SJohannes Thumshirn bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT; 3607314b6dd0SJohannes Thumshirn bio->bi_private = device; 3608314b6dd0SJohannes Thumshirn bio->bi_end_io = btrfs_end_super_write; 3609314b6dd0SJohannes Thumshirn __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE, 3610314b6dd0SJohannes Thumshirn offset_in_page(bytenr)); 3611314b6dd0SJohannes Thumshirn 3612314b6dd0SJohannes Thumshirn /* 3613314b6dd0SJohannes Thumshirn * We FUA only the first super block. The others we allow to 3614314b6dd0SJohannes Thumshirn * go down lazy and there's a short window where the on-disk 3615314b6dd0SJohannes Thumshirn * copies might still contain the older version. 3616314b6dd0SJohannes Thumshirn */ 3617314b6dd0SJohannes Thumshirn bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO; 36181b9e619cSOmar Sandoval if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) 3619314b6dd0SJohannes Thumshirn bio->bi_opf |= REQ_FUA; 3620314b6dd0SJohannes Thumshirn 3621314b6dd0SJohannes Thumshirn btrfsic_submit_bio(bio); 3622a512bbf8SYan Zheng } 3623a512bbf8SYan Zheng return errors < i ? 0 : -1; 3624a512bbf8SYan Zheng } 3625a512bbf8SYan Zheng 3626387125fcSChris Mason /* 3627abbb3b8eSDavid Sterba * Wait for write completion of superblocks done by write_dev_supers, 3628abbb3b8eSDavid Sterba * @max_mirrors same for write and wait phases. 3629abbb3b8eSDavid Sterba * 3630314b6dd0SJohannes Thumshirn * Return number of errors when page is not found or not marked up to 3631abbb3b8eSDavid Sterba * date. 3632abbb3b8eSDavid Sterba */ 3633abbb3b8eSDavid Sterba static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) 3634abbb3b8eSDavid Sterba { 3635abbb3b8eSDavid Sterba int i; 3636abbb3b8eSDavid Sterba int errors = 0; 3637b6a535faSHoward McLauchlan bool primary_failed = false; 3638abbb3b8eSDavid Sterba u64 bytenr; 3639abbb3b8eSDavid Sterba 3640abbb3b8eSDavid Sterba if (max_mirrors == 0) 3641abbb3b8eSDavid Sterba max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3642abbb3b8eSDavid Sterba 3643abbb3b8eSDavid Sterba for (i = 0; i < max_mirrors; i++) { 3644314b6dd0SJohannes Thumshirn struct page *page; 3645314b6dd0SJohannes Thumshirn 3646abbb3b8eSDavid Sterba bytenr = btrfs_sb_offset(i); 3647abbb3b8eSDavid Sterba if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3648abbb3b8eSDavid Sterba device->commit_total_bytes) 3649abbb3b8eSDavid Sterba break; 3650abbb3b8eSDavid Sterba 3651314b6dd0SJohannes Thumshirn page = find_get_page(device->bdev->bd_inode->i_mapping, 3652314b6dd0SJohannes Thumshirn bytenr >> PAGE_SHIFT); 3653314b6dd0SJohannes Thumshirn if (!page) { 3654abbb3b8eSDavid Sterba errors++; 3655b6a535faSHoward McLauchlan if (i == 0) 3656b6a535faSHoward McLauchlan primary_failed = true; 3657abbb3b8eSDavid Sterba continue; 3658abbb3b8eSDavid Sterba } 3659314b6dd0SJohannes Thumshirn /* Page is submitted locked and unlocked once the IO completes */ 3660314b6dd0SJohannes Thumshirn wait_on_page_locked(page); 3661314b6dd0SJohannes Thumshirn if (PageError(page)) { 3662abbb3b8eSDavid Sterba errors++; 3663b6a535faSHoward McLauchlan if (i == 0) 3664b6a535faSHoward McLauchlan primary_failed = true; 3665b6a535faSHoward McLauchlan } 3666abbb3b8eSDavid Sterba 3667314b6dd0SJohannes Thumshirn /* Drop our reference */ 3668314b6dd0SJohannes Thumshirn put_page(page); 3669abbb3b8eSDavid Sterba 3670314b6dd0SJohannes Thumshirn /* Drop the reference from the writing run */ 3671314b6dd0SJohannes Thumshirn put_page(page); 3672abbb3b8eSDavid Sterba } 3673abbb3b8eSDavid Sterba 3674b6a535faSHoward McLauchlan /* log error, force error return */ 3675b6a535faSHoward McLauchlan if (primary_failed) { 3676b6a535faSHoward McLauchlan btrfs_err(device->fs_info, "error writing primary super block to device %llu", 3677b6a535faSHoward McLauchlan device->devid); 3678b6a535faSHoward McLauchlan return -1; 3679b6a535faSHoward McLauchlan } 3680b6a535faSHoward McLauchlan 3681abbb3b8eSDavid Sterba return errors < i ? 0 : -1; 3682abbb3b8eSDavid Sterba } 3683abbb3b8eSDavid Sterba 3684abbb3b8eSDavid Sterba /* 3685387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 3686387125fcSChris Mason * for the barrier when it is done 3687387125fcSChris Mason */ 36884246a0b6SChristoph Hellwig static void btrfs_end_empty_barrier(struct bio *bio) 3689387125fcSChris Mason { 3690387125fcSChris Mason complete(bio->bi_private); 3691387125fcSChris Mason } 3692387125fcSChris Mason 3693387125fcSChris Mason /* 36944fc6441aSAnand Jain * Submit a flush request to the device if it supports it. Error handling is 36954fc6441aSAnand Jain * done in the waiting counterpart. 3696387125fcSChris Mason */ 36974fc6441aSAnand Jain static void write_dev_flush(struct btrfs_device *device) 3698387125fcSChris Mason { 3699c2a9c7abSAnand Jain struct request_queue *q = bdev_get_queue(device->bdev); 3700e0ae9994SDavid Sterba struct bio *bio = device->flush_bio; 3701387125fcSChris Mason 3702c2a9c7abSAnand Jain if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) 37034fc6441aSAnand Jain return; 3704387125fcSChris Mason 3705e0ae9994SDavid Sterba bio_reset(bio); 3706387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 370774d46992SChristoph Hellwig bio_set_dev(bio, device->bdev); 37088d910125SJan Kara bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH; 3709387125fcSChris Mason init_completion(&device->flush_wait); 3710387125fcSChris Mason bio->bi_private = &device->flush_wait; 3711387125fcSChris Mason 371243a01111SLu Fengqi btrfsic_submit_bio(bio); 37131c3063b6SAnand Jain set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 37144fc6441aSAnand Jain } 3715387125fcSChris Mason 37164fc6441aSAnand Jain /* 37174fc6441aSAnand Jain * If the flush bio has been submitted by write_dev_flush, wait for it. 37184fc6441aSAnand Jain */ 37198c27cb35SLinus Torvalds static blk_status_t wait_dev_flush(struct btrfs_device *device) 37204fc6441aSAnand Jain { 37214fc6441aSAnand Jain struct bio *bio = device->flush_bio; 37224fc6441aSAnand Jain 37231c3063b6SAnand Jain if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) 372458efbc9fSOmar Sandoval return BLK_STS_OK; 37254fc6441aSAnand Jain 37261c3063b6SAnand Jain clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 37272980d574SDavid Sterba wait_for_completion_io(&device->flush_wait); 37284fc6441aSAnand Jain 37298c27cb35SLinus Torvalds return bio->bi_status; 3730387125fcSChris Mason } 3731387125fcSChris Mason 3732d10b82feSQu Wenruo static int check_barrier_error(struct btrfs_fs_info *fs_info) 3733401b41e5SAnand Jain { 37346528b99dSAnand Jain if (!btrfs_check_rw_degradable(fs_info, NULL)) 3735401b41e5SAnand Jain return -EIO; 3736387125fcSChris Mason return 0; 3737387125fcSChris Mason } 3738387125fcSChris Mason 3739387125fcSChris Mason /* 3740387125fcSChris Mason * send an empty flush down to each device in parallel, 3741387125fcSChris Mason * then wait for them 3742387125fcSChris Mason */ 3743387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 3744387125fcSChris Mason { 3745387125fcSChris Mason struct list_head *head; 3746387125fcSChris Mason struct btrfs_device *dev; 37475af3e8ccSStefan Behrens int errors_wait = 0; 37484e4cbee9SChristoph Hellwig blk_status_t ret; 3749387125fcSChris Mason 37501538e6c5SDavid Sterba lockdep_assert_held(&info->fs_devices->device_list_mutex); 3751387125fcSChris Mason /* send down all the barriers */ 3752387125fcSChris Mason head = &info->fs_devices->devices; 37531538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3754e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3755f88ba6a2SHidetoshi Seto continue; 3756cea7c8bfSAnand Jain if (!dev->bdev) 3757387125fcSChris Mason continue; 3758e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3759ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3760387125fcSChris Mason continue; 3761387125fcSChris Mason 37624fc6441aSAnand Jain write_dev_flush(dev); 376358efbc9fSOmar Sandoval dev->last_flush_error = BLK_STS_OK; 3764387125fcSChris Mason } 3765387125fcSChris Mason 3766387125fcSChris Mason /* wait for all the barriers */ 37671538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3768e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3769f88ba6a2SHidetoshi Seto continue; 3770387125fcSChris Mason if (!dev->bdev) { 37715af3e8ccSStefan Behrens errors_wait++; 3772387125fcSChris Mason continue; 3773387125fcSChris Mason } 3774e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3775ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3776387125fcSChris Mason continue; 3777387125fcSChris Mason 37784fc6441aSAnand Jain ret = wait_dev_flush(dev); 3779401b41e5SAnand Jain if (ret) { 3780401b41e5SAnand Jain dev->last_flush_error = ret; 378166b4993eSDavid Sterba btrfs_dev_stat_inc_and_print(dev, 378266b4993eSDavid Sterba BTRFS_DEV_STAT_FLUSH_ERRS); 37835af3e8ccSStefan Behrens errors_wait++; 3784387125fcSChris Mason } 3785401b41e5SAnand Jain } 3786401b41e5SAnand Jain 3787cea7c8bfSAnand Jain if (errors_wait) { 3788401b41e5SAnand Jain /* 3789401b41e5SAnand Jain * At some point we need the status of all disks 3790401b41e5SAnand Jain * to arrive at the volume status. So error checking 3791401b41e5SAnand Jain * is being pushed to a separate loop. 3792401b41e5SAnand Jain */ 3793d10b82feSQu Wenruo return check_barrier_error(info); 3794401b41e5SAnand Jain } 3795387125fcSChris Mason return 0; 3796387125fcSChris Mason } 3797387125fcSChris Mason 3798943c6e99SZhao Lei int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 3799943c6e99SZhao Lei { 38008789f4feSZhao Lei int raid_type; 38018789f4feSZhao Lei int min_tolerated = INT_MAX; 3802943c6e99SZhao Lei 38038789f4feSZhao Lei if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 38048789f4feSZhao Lei (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 38058c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 38068789f4feSZhao Lei btrfs_raid_array[BTRFS_RAID_SINGLE]. 38078789f4feSZhao Lei tolerated_failures); 3808943c6e99SZhao Lei 38098789f4feSZhao Lei for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 38108789f4feSZhao Lei if (raid_type == BTRFS_RAID_SINGLE) 38118789f4feSZhao Lei continue; 381241a6e891SAnand Jain if (!(flags & btrfs_raid_array[raid_type].bg_flag)) 38138789f4feSZhao Lei continue; 38148c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 38158789f4feSZhao Lei btrfs_raid_array[raid_type]. 38168789f4feSZhao Lei tolerated_failures); 38178789f4feSZhao Lei } 3818943c6e99SZhao Lei 38198789f4feSZhao Lei if (min_tolerated == INT_MAX) { 3820ab8d0fc4SJeff Mahoney pr_warn("BTRFS: unknown raid flag: %llu", flags); 38218789f4feSZhao Lei min_tolerated = 0; 38228789f4feSZhao Lei } 38238789f4feSZhao Lei 38248789f4feSZhao Lei return min_tolerated; 3825943c6e99SZhao Lei } 3826943c6e99SZhao Lei 3827eece6a9cSDavid Sterba int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) 3828f2984462SChris Mason { 3829e5e9a520SChris Mason struct list_head *head; 3830f2984462SChris Mason struct btrfs_device *dev; 3831a061fc8dSChris Mason struct btrfs_super_block *sb; 3832f2984462SChris Mason struct btrfs_dev_item *dev_item; 3833f2984462SChris Mason int ret; 3834f2984462SChris Mason int do_barriers; 3835a236aed1SChris Mason int max_errors; 3836a236aed1SChris Mason int total_errors = 0; 3837a061fc8dSChris Mason u64 flags; 3838f2984462SChris Mason 38390b246afaSJeff Mahoney do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); 3840fed3b381SLiu Bo 3841fed3b381SLiu Bo /* 3842fed3b381SLiu Bo * max_mirrors == 0 indicates we're from commit_transaction, 3843fed3b381SLiu Bo * not from fsync where the tree roots in fs_info have not 3844fed3b381SLiu Bo * been consistent on disk. 3845fed3b381SLiu Bo */ 3846fed3b381SLiu Bo if (max_mirrors == 0) 38470b246afaSJeff Mahoney backup_super_roots(fs_info); 3848f2984462SChris Mason 38490b246afaSJeff Mahoney sb = fs_info->super_for_commit; 3850a061fc8dSChris Mason dev_item = &sb->dev_item; 3851e5e9a520SChris Mason 38520b246afaSJeff Mahoney mutex_lock(&fs_info->fs_devices->device_list_mutex); 38530b246afaSJeff Mahoney head = &fs_info->fs_devices->devices; 38540b246afaSJeff Mahoney max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; 3855387125fcSChris Mason 38565af3e8ccSStefan Behrens if (do_barriers) { 38570b246afaSJeff Mahoney ret = barrier_all_devices(fs_info); 38585af3e8ccSStefan Behrens if (ret) { 38595af3e8ccSStefan Behrens mutex_unlock( 38600b246afaSJeff Mahoney &fs_info->fs_devices->device_list_mutex); 38610b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 38625af3e8ccSStefan Behrens "errors while submitting device barriers."); 38635af3e8ccSStefan Behrens return ret; 38645af3e8ccSStefan Behrens } 38655af3e8ccSStefan Behrens } 3866387125fcSChris Mason 38671538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3868dfe25020SChris Mason if (!dev->bdev) { 3869dfe25020SChris Mason total_errors++; 3870dfe25020SChris Mason continue; 3871dfe25020SChris Mason } 3872e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3873ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3874dfe25020SChris Mason continue; 3875dfe25020SChris Mason 38762b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 3877a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 3878a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 38797df69d3eSMiao Xie btrfs_set_stack_device_total_bytes(dev_item, 3880935e5cc9SMiao Xie dev->commit_total_bytes); 3881ce7213c7SMiao Xie btrfs_set_stack_device_bytes_used(dev_item, 3882ce7213c7SMiao Xie dev->commit_bytes_used); 3883a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 3884a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 3885a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 3886a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 38877239ff4bSNikolay Borisov memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, 38887239ff4bSNikolay Borisov BTRFS_FSID_SIZE); 3889a512bbf8SYan Zheng 3890a061fc8dSChris Mason flags = btrfs_super_flags(sb); 3891a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 3892f2984462SChris Mason 389375cb857dSQu Wenruo ret = btrfs_validate_write_super(fs_info, sb); 389475cb857dSQu Wenruo if (ret < 0) { 389575cb857dSQu Wenruo mutex_unlock(&fs_info->fs_devices->device_list_mutex); 389675cb857dSQu Wenruo btrfs_handle_fs_error(fs_info, -EUCLEAN, 389775cb857dSQu Wenruo "unexpected superblock corruption detected"); 389875cb857dSQu Wenruo return -EUCLEAN; 389975cb857dSQu Wenruo } 390075cb857dSQu Wenruo 3901abbb3b8eSDavid Sterba ret = write_dev_supers(dev, sb, max_mirrors); 3902a236aed1SChris Mason if (ret) 3903a236aed1SChris Mason total_errors++; 3904f2984462SChris Mason } 3905a236aed1SChris Mason if (total_errors > max_errors) { 39060b246afaSJeff Mahoney btrfs_err(fs_info, "%d errors while writing supers", 3907d397712bSChris Mason total_errors); 39080b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 390979787eaaSJeff Mahoney 39109d565ba4SStefan Behrens /* FUA is masked off if unsupported and can't be the reason */ 39110b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 39120b246afaSJeff Mahoney "%d errors while writing supers", 39130b246afaSJeff Mahoney total_errors); 39149d565ba4SStefan Behrens return -EIO; 3915a236aed1SChris Mason } 3916f2984462SChris Mason 3917a512bbf8SYan Zheng total_errors = 0; 39181538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3919dfe25020SChris Mason if (!dev->bdev) 3920dfe25020SChris Mason continue; 3921e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3922ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3923dfe25020SChris Mason continue; 3924dfe25020SChris Mason 3925abbb3b8eSDavid Sterba ret = wait_dev_supers(dev, max_mirrors); 3926a512bbf8SYan Zheng if (ret) 39271259ab75SChris Mason total_errors++; 3928f2984462SChris Mason } 39290b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 3930a236aed1SChris Mason if (total_errors > max_errors) { 39310b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 39320b246afaSJeff Mahoney "%d errors while writing supers", 39330b246afaSJeff Mahoney total_errors); 393479787eaaSJeff Mahoney return -EIO; 3935a236aed1SChris Mason } 3936f2984462SChris Mason return 0; 3937f2984462SChris Mason } 3938f2984462SChris Mason 3939cb517eabSMiao Xie /* Drop a fs root from the radix tree and free it. */ 3940cb517eabSMiao Xie void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 3941cb517eabSMiao Xie struct btrfs_root *root) 39422619ba1fSChris Mason { 39434785e24fSJosef Bacik bool drop_ref = false; 39444785e24fSJosef Bacik 39454df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 39462619ba1fSChris Mason radix_tree_delete(&fs_info->fs_roots_radix, 39472619ba1fSChris Mason (unsigned long)root->root_key.objectid); 3948af01d2e5SJosef Bacik if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state)) 39494785e24fSJosef Bacik drop_ref = true; 39504df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 395176dda93cSYan, Zheng 39521c1ea4f7SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 3953ef67963dSJosef Bacik ASSERT(root->log_root == NULL); 39541c1ea4f7SLiu Bo if (root->reloc_root) { 395500246528SJosef Bacik btrfs_put_root(root->reloc_root); 39561c1ea4f7SLiu Bo root->reloc_root = NULL; 39571c1ea4f7SLiu Bo } 39581c1ea4f7SLiu Bo } 39593321719eSLiu Bo 3960faa2dbf0SJosef Bacik if (root->free_ino_pinned) 3961581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_pinned); 3962faa2dbf0SJosef Bacik if (root->free_ino_ctl) 3963581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_ctl); 39640e996e7fSJosef Bacik if (root->ino_cache_inode) { 39650e996e7fSJosef Bacik iput(root->ino_cache_inode); 39660e996e7fSJosef Bacik root->ino_cache_inode = NULL; 39670e996e7fSJosef Bacik } 39684785e24fSJosef Bacik if (drop_ref) 396900246528SJosef Bacik btrfs_put_root(root); 39702619ba1fSChris Mason } 39712619ba1fSChris Mason 3972c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 3973c146afadSYan Zheng { 3974c146afadSYan Zheng u64 root_objectid = 0; 3975c146afadSYan Zheng struct btrfs_root *gang[8]; 397665d33fd7SQu Wenruo int i = 0; 397765d33fd7SQu Wenruo int err = 0; 397865d33fd7SQu Wenruo unsigned int ret = 0; 3979c146afadSYan Zheng 3980c146afadSYan Zheng while (1) { 3981efc34534SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 3982c146afadSYan Zheng ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 3983c146afadSYan Zheng (void **)gang, root_objectid, 3984c146afadSYan Zheng ARRAY_SIZE(gang)); 398565d33fd7SQu Wenruo if (!ret) { 3986efc34534SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 3987c146afadSYan Zheng break; 398865d33fd7SQu Wenruo } 39895d4f98a2SYan Zheng root_objectid = gang[ret - 1]->root_key.objectid + 1; 399066b4ffd1SJosef Bacik 399165d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 399265d33fd7SQu Wenruo /* Avoid to grab roots in dead_roots */ 399365d33fd7SQu Wenruo if (btrfs_root_refs(&gang[i]->root_item) == 0) { 399465d33fd7SQu Wenruo gang[i] = NULL; 399565d33fd7SQu Wenruo continue; 399665d33fd7SQu Wenruo } 399765d33fd7SQu Wenruo /* grab all the search result for later use */ 399800246528SJosef Bacik gang[i] = btrfs_grab_root(gang[i]); 399965d33fd7SQu Wenruo } 4000efc34534SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 400165d33fd7SQu Wenruo 400265d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 400365d33fd7SQu Wenruo if (!gang[i]) 400465d33fd7SQu Wenruo continue; 4005c146afadSYan Zheng root_objectid = gang[i]->root_key.objectid; 400666b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(gang[i]); 400766b4ffd1SJosef Bacik if (err) 400865d33fd7SQu Wenruo break; 400900246528SJosef Bacik btrfs_put_root(gang[i]); 4010c146afadSYan Zheng } 4011c146afadSYan Zheng root_objectid++; 4012c146afadSYan Zheng } 401365d33fd7SQu Wenruo 401465d33fd7SQu Wenruo /* release the uncleaned roots due to error */ 401565d33fd7SQu Wenruo for (; i < ret; i++) { 401665d33fd7SQu Wenruo if (gang[i]) 401700246528SJosef Bacik btrfs_put_root(gang[i]); 401865d33fd7SQu Wenruo } 401965d33fd7SQu Wenruo return err; 4020c146afadSYan Zheng } 4021c146afadSYan Zheng 40226bccf3abSJeff Mahoney int btrfs_commit_super(struct btrfs_fs_info *fs_info) 4023c146afadSYan Zheng { 40246bccf3abSJeff Mahoney struct btrfs_root *root = fs_info->tree_root; 4025c146afadSYan Zheng struct btrfs_trans_handle *trans; 4026c146afadSYan Zheng 40270b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 40282ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 40290b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 40300b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 4031c71bf099SYan, Zheng 4032c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 40330b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 40340b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4035c71bf099SYan, Zheng 40367a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 40373612b495STsutomu Itoh if (IS_ERR(trans)) 40383612b495STsutomu Itoh return PTR_ERR(trans); 40393a45bb20SJeff Mahoney return btrfs_commit_transaction(trans); 4040c146afadSYan Zheng } 4041c146afadSYan Zheng 4042b105e927SDavid Sterba void __cold close_ctree(struct btrfs_fs_info *fs_info) 4043eb60ceacSChris Mason { 4044c146afadSYan Zheng int ret; 4045e089f05cSChris Mason 4046afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); 4047d6fd0ae2SOmar Sandoval /* 4048d6fd0ae2SOmar Sandoval * We don't want the cleaner to start new transactions, add more delayed 4049d6fd0ae2SOmar Sandoval * iputs, etc. while we're closing. We can't use kthread_stop() yet 4050d6fd0ae2SOmar Sandoval * because that frees the task_struct, and the transaction kthread might 4051d6fd0ae2SOmar Sandoval * still try to wake up the cleaner. 4052d6fd0ae2SOmar Sandoval */ 4053d6fd0ae2SOmar Sandoval kthread_park(fs_info->cleaner_kthread); 4054a2135011SChris Mason 40557343dd61SJustin Maggard /* wait for the qgroup rescan worker to stop */ 4056d06f23d6SJeff Mahoney btrfs_qgroup_wait_for_completion(fs_info, false); 40577343dd61SJustin Maggard 4058803b2f54SStefan Behrens /* wait for the uuid_scan task to finish */ 4059803b2f54SStefan Behrens down(&fs_info->uuid_tree_rescan_sem); 4060803b2f54SStefan Behrens /* avoid complains from lockdep et al., set sem back to initial state */ 4061803b2f54SStefan Behrens up(&fs_info->uuid_tree_rescan_sem); 4062803b2f54SStefan Behrens 4063837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 4064aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 4065837d5b6eSIlya Dryomov 40668dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 40678dabb742SStefan Behrens 4068aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 40694cb5300bSChris Mason 40704cb5300bSChris Mason /* wait for any defraggers to finish */ 40714cb5300bSChris Mason wait_event(fs_info->transaction_wait, 40724cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 40734cb5300bSChris Mason 40744cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 407526176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 40764cb5300bSChris Mason 407721c7e756SMiao Xie cancel_work_sync(&fs_info->async_reclaim_work); 407857056740SJosef Bacik cancel_work_sync(&fs_info->async_data_reclaim_work); 407921c7e756SMiao Xie 4080b0643e59SDennis Zhou /* Cancel or finish ongoing discard work */ 4081b0643e59SDennis Zhou btrfs_discard_cleanup(fs_info); 4082b0643e59SDennis Zhou 4083bc98a42cSDavid Howells if (!sb_rdonly(fs_info->sb)) { 4084e44163e1SJeff Mahoney /* 4085d6fd0ae2SOmar Sandoval * The cleaner kthread is stopped, so do one final pass over 4086d6fd0ae2SOmar Sandoval * unused block groups. 4087e44163e1SJeff Mahoney */ 40880b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 4089e44163e1SJeff Mahoney 4090f0cc2cd7SFilipe Manana /* 4091f0cc2cd7SFilipe Manana * There might be existing delayed inode workers still running 4092f0cc2cd7SFilipe Manana * and holding an empty delayed inode item. We must wait for 4093f0cc2cd7SFilipe Manana * them to complete first because they can create a transaction. 4094f0cc2cd7SFilipe Manana * This happens when someone calls btrfs_balance_delayed_items() 4095f0cc2cd7SFilipe Manana * and then a transaction commit runs the same delayed nodes 4096f0cc2cd7SFilipe Manana * before any delayed worker has done something with the nodes. 4097f0cc2cd7SFilipe Manana * We must wait for any worker here and not at transaction 4098f0cc2cd7SFilipe Manana * commit time since that could cause a deadlock. 4099f0cc2cd7SFilipe Manana * This is a very rare case. 4100f0cc2cd7SFilipe Manana */ 4101f0cc2cd7SFilipe Manana btrfs_flush_workqueue(fs_info->delayed_workers); 4102f0cc2cd7SFilipe Manana 41036bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 4104d397712bSChris Mason if (ret) 410504892340SEric Sandeen btrfs_err(fs_info, "commit super ret %d", ret); 4106c146afadSYan Zheng } 4107ed2ff2cbSChris Mason 4108af722733SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) || 4109af722733SLiu Bo test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) 41102ff7e61eSJeff Mahoney btrfs_error_commit_super(fs_info); 4111acce952bSliubo 4112e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 4113e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 41148929ecfaSYan, Zheng 4115e187831eSJosef Bacik ASSERT(list_empty(&fs_info->delayed_iputs)); 4116afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); 4117f25784b3SYan Zheng 41185958253cSQu Wenruo if (btrfs_check_quota_leak(fs_info)) { 41195958253cSQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 41205958253cSQu Wenruo btrfs_err(fs_info, "qgroup reserved space leaked"); 41215958253cSQu Wenruo } 41225958253cSQu Wenruo 412304892340SEric Sandeen btrfs_free_qgroup_config(fs_info); 4124fe816d0fSNikolay Borisov ASSERT(list_empty(&fs_info->delalloc_roots)); 4125bcef60f2SArne Jansen 4126963d678bSMiao Xie if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 412704892340SEric Sandeen btrfs_info(fs_info, "at unmount delalloc count %lld", 4128963d678bSMiao Xie percpu_counter_sum(&fs_info->delalloc_bytes)); 4129b0c68f8bSChris Mason } 413031153d81SYan Zheng 41314297ff84SJosef Bacik if (percpu_counter_sum(&fs_info->dio_bytes)) 41324297ff84SJosef Bacik btrfs_info(fs_info, "at unmount dio bytes count %lld", 41334297ff84SJosef Bacik percpu_counter_sum(&fs_info->dio_bytes)); 41344297ff84SJosef Bacik 41356618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 4136b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 41375ac1d209SJeff Mahoney 41381a4319ccSLiu Bo btrfs_put_block_group_cache(fs_info); 41391a4319ccSLiu Bo 4140de348ee0SWang Shilong /* 4141de348ee0SWang Shilong * we must make sure there is not any read request to 4142de348ee0SWang Shilong * submit after we stopping all workers. 4143de348ee0SWang Shilong */ 4144de348ee0SWang Shilong invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 414596192499SJosef Bacik btrfs_stop_all_workers(fs_info); 414696192499SJosef Bacik 4147afcdd129SJosef Bacik clear_bit(BTRFS_FS_OPEN, &fs_info->flags); 41484273eaffSAnand Jain free_root_pointers(fs_info, true); 41498c38938cSJosef Bacik btrfs_free_fs_roots(fs_info); 41509ad6b7bcSChris Mason 41514e19443dSJosef Bacik /* 41524e19443dSJosef Bacik * We must free the block groups after dropping the fs_roots as we could 41534e19443dSJosef Bacik * have had an IO error and have left over tree log blocks that aren't 41544e19443dSJosef Bacik * cleaned up until the fs roots are freed. This makes the block group 41554e19443dSJosef Bacik * accounting appear to be wrong because there's pending reserved bytes, 41564e19443dSJosef Bacik * so make sure we do the block group cleanup afterwards. 41574e19443dSJosef Bacik */ 41584e19443dSJosef Bacik btrfs_free_block_groups(fs_info); 41594e19443dSJosef Bacik 416013e6c37bSJosef Bacik iput(fs_info->btree_inode); 4161d6bfde87SChris Mason 416221adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 41630b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) 41642ff7e61eSJeff Mahoney btrfsic_unmount(fs_info->fs_devices); 416521adbd5cSStefan Behrens #endif 416621adbd5cSStefan Behrens 41670b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 416868c94e55SNikolay Borisov btrfs_close_devices(fs_info->fs_devices); 4169eb60ceacSChris Mason } 4170eb60ceacSChris Mason 4171b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 4172b9fab919SChris Mason int atomic) 4173ccd467d6SChris Mason { 41741259ab75SChris Mason int ret; 4175727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 41761259ab75SChris Mason 41770b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 41781259ab75SChris Mason if (!ret) 41791259ab75SChris Mason return ret; 41801259ab75SChris Mason 41811259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 4182b9fab919SChris Mason parent_transid, atomic); 4183b9fab919SChris Mason if (ret == -EAGAIN) 4184b9fab919SChris Mason return ret; 41851259ab75SChris Mason return !ret; 41865f39d397SChris Mason } 41876702ed49SChris Mason 41885f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 41895f39d397SChris Mason { 4190*2f4d60dfSQu Wenruo struct btrfs_fs_info *fs_info = buf->fs_info; 41915f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 4192b9473439SChris Mason int was_dirty; 4193b4ce94deSChris Mason 419406ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 419506ea65a3SJosef Bacik /* 419606ea65a3SJosef Bacik * This is a fast path so only do this check if we have sanity tests 419752042d8eSAndrea Gelmini * enabled. Normal people shouldn't be using unmapped buffers as dirty 419806ea65a3SJosef Bacik * outside of the sanity tests. 419906ea65a3SJosef Bacik */ 4200b0132a3bSNikolay Borisov if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) 420106ea65a3SJosef Bacik return; 420206ea65a3SJosef Bacik #endif 4203b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 42040b246afaSJeff Mahoney if (transid != fs_info->generation) 42055d163e0eSJeff Mahoney WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", 42060b246afaSJeff Mahoney buf->start, transid, fs_info->generation); 42070b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 4208e2d84521SMiao Xie if (!was_dirty) 4209104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 4210e2d84521SMiao Xie buf->len, 42110b246afaSJeff Mahoney fs_info->dirty_metadata_batch); 42121f21ef0aSFilipe Manana #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 421369fc6cbbSQu Wenruo /* 421469fc6cbbSQu Wenruo * Since btrfs_mark_buffer_dirty() can be called with item pointer set 421569fc6cbbSQu Wenruo * but item data not updated. 421669fc6cbbSQu Wenruo * So here we should only check item pointers, not item data. 421769fc6cbbSQu Wenruo */ 421869fc6cbbSQu Wenruo if (btrfs_header_level(buf) == 0 && 4219cfdaad5eSDavid Sterba btrfs_check_leaf_relaxed(buf)) { 4220a4f78750SDavid Sterba btrfs_print_leaf(buf); 42211f21ef0aSFilipe Manana ASSERT(0); 42221f21ef0aSFilipe Manana } 42231f21ef0aSFilipe Manana #endif 4224eb60ceacSChris Mason } 4225eb60ceacSChris Mason 42262ff7e61eSJeff Mahoney static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, 4227b53d3f5dSLiu Bo int flush_delayed) 422835b7e476SChris Mason { 4229188de649SChris Mason /* 4230188de649SChris Mason * looks as though older kernels can get into trouble with 4231188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 4232188de649SChris Mason */ 4233e2d84521SMiao Xie int ret; 4234d6bfde87SChris Mason 42356933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 4236d6bfde87SChris Mason return; 4237d6bfde87SChris Mason 4238b53d3f5dSLiu Bo if (flush_delayed) 42392ff7e61eSJeff Mahoney btrfs_balance_delayed_items(fs_info); 424016cdcec7SMiao Xie 4241d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 4242d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 4243d814a491SEthan Lien fs_info->dirty_metadata_batch); 4244e2d84521SMiao Xie if (ret > 0) { 42450b246afaSJeff Mahoney balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); 424616cdcec7SMiao Xie } 424716cdcec7SMiao Xie } 424816cdcec7SMiao Xie 42492ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) 425016cdcec7SMiao Xie { 42512ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 1); 425235b7e476SChris Mason } 4253b53d3f5dSLiu Bo 42542ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) 4255b53d3f5dSLiu Bo { 42562ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 0); 4257d6bfde87SChris Mason } 42586b80053dSChris Mason 4259581c1760SQu Wenruo int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level, 4260581c1760SQu Wenruo struct btrfs_key *first_key) 42616b80053dSChris Mason { 42625ab12d1fSDavid Sterba return btree_read_extent_buffer_pages(buf, parent_transid, 4263581c1760SQu Wenruo level, first_key); 42646b80053dSChris Mason } 42650da5468fSChris Mason 42662ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) 4267acce952bSliubo { 4268fe816d0fSNikolay Borisov /* cleanup FS via transaction */ 4269fe816d0fSNikolay Borisov btrfs_cleanup_transaction(fs_info); 4270fe816d0fSNikolay Borisov 42710b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 42722ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 42730b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 4274acce952bSliubo 42750b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 42760b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4277acce952bSliubo } 4278acce952bSliubo 4279ef67963dSJosef Bacik static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info) 4280ef67963dSJosef Bacik { 4281ef67963dSJosef Bacik struct btrfs_root *gang[8]; 4282ef67963dSJosef Bacik u64 root_objectid = 0; 4283ef67963dSJosef Bacik int ret; 4284ef67963dSJosef Bacik 4285ef67963dSJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 4286ef67963dSJosef Bacik while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4287ef67963dSJosef Bacik (void **)gang, root_objectid, 4288ef67963dSJosef Bacik ARRAY_SIZE(gang))) != 0) { 4289ef67963dSJosef Bacik int i; 4290ef67963dSJosef Bacik 4291ef67963dSJosef Bacik for (i = 0; i < ret; i++) 4292ef67963dSJosef Bacik gang[i] = btrfs_grab_root(gang[i]); 4293ef67963dSJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 4294ef67963dSJosef Bacik 4295ef67963dSJosef Bacik for (i = 0; i < ret; i++) { 4296ef67963dSJosef Bacik if (!gang[i]) 4297ef67963dSJosef Bacik continue; 4298ef67963dSJosef Bacik root_objectid = gang[i]->root_key.objectid; 4299ef67963dSJosef Bacik btrfs_free_log(NULL, gang[i]); 4300ef67963dSJosef Bacik btrfs_put_root(gang[i]); 4301ef67963dSJosef Bacik } 4302ef67963dSJosef Bacik root_objectid++; 4303ef67963dSJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 4304ef67963dSJosef Bacik } 4305ef67963dSJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 4306ef67963dSJosef Bacik btrfs_free_log_root_tree(NULL, fs_info); 4307ef67963dSJosef Bacik } 4308ef67963dSJosef Bacik 4309143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4310acce952bSliubo { 4311acce952bSliubo struct btrfs_ordered_extent *ordered; 4312acce952bSliubo 4313199c2a9cSMiao Xie spin_lock(&root->ordered_extent_lock); 4314779880efSJosef Bacik /* 4315779880efSJosef Bacik * This will just short circuit the ordered completion stuff which will 4316779880efSJosef Bacik * make sure the ordered extent gets properly cleaned up. 4317779880efSJosef Bacik */ 4318199c2a9cSMiao Xie list_for_each_entry(ordered, &root->ordered_extents, 4319779880efSJosef Bacik root_extent_list) 4320779880efSJosef Bacik set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4321199c2a9cSMiao Xie spin_unlock(&root->ordered_extent_lock); 4322199c2a9cSMiao Xie } 4323199c2a9cSMiao Xie 4324199c2a9cSMiao Xie static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4325199c2a9cSMiao Xie { 4326199c2a9cSMiao Xie struct btrfs_root *root; 4327199c2a9cSMiao Xie struct list_head splice; 4328199c2a9cSMiao Xie 4329199c2a9cSMiao Xie INIT_LIST_HEAD(&splice); 4330199c2a9cSMiao Xie 4331199c2a9cSMiao Xie spin_lock(&fs_info->ordered_root_lock); 4332199c2a9cSMiao Xie list_splice_init(&fs_info->ordered_roots, &splice); 4333199c2a9cSMiao Xie while (!list_empty(&splice)) { 4334199c2a9cSMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4335199c2a9cSMiao Xie ordered_root); 43361de2cfdeSJosef Bacik list_move_tail(&root->ordered_root, 43371de2cfdeSJosef Bacik &fs_info->ordered_roots); 4338199c2a9cSMiao Xie 43392a85d9caSLiu Bo spin_unlock(&fs_info->ordered_root_lock); 4340199c2a9cSMiao Xie btrfs_destroy_ordered_extents(root); 4341199c2a9cSMiao Xie 43422a85d9caSLiu Bo cond_resched(); 43432a85d9caSLiu Bo spin_lock(&fs_info->ordered_root_lock); 4344199c2a9cSMiao Xie } 4345199c2a9cSMiao Xie spin_unlock(&fs_info->ordered_root_lock); 434674d5d229SJosef Bacik 434774d5d229SJosef Bacik /* 434874d5d229SJosef Bacik * We need this here because if we've been flipped read-only we won't 434974d5d229SJosef Bacik * get sync() from the umount, so we need to make sure any ordered 435074d5d229SJosef Bacik * extents that haven't had their dirty pages IO start writeout yet 435174d5d229SJosef Bacik * actually get run and error out properly. 435274d5d229SJosef Bacik */ 435374d5d229SJosef Bacik btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 4354acce952bSliubo } 4355acce952bSliubo 435635a3621bSStefan Behrens static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 43572ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4358acce952bSliubo { 4359acce952bSliubo struct rb_node *node; 4360acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 4361acce952bSliubo struct btrfs_delayed_ref_node *ref; 4362acce952bSliubo int ret = 0; 4363acce952bSliubo 4364acce952bSliubo delayed_refs = &trans->delayed_refs; 4365acce952bSliubo 4366acce952bSliubo spin_lock(&delayed_refs->lock); 4367d7df2c79SJosef Bacik if (atomic_read(&delayed_refs->num_entries) == 0) { 4368cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 4369b79ce3ddSDavid Sterba btrfs_debug(fs_info, "delayed_refs has NO entry"); 4370acce952bSliubo return ret; 4371acce952bSliubo } 4372acce952bSliubo 43735c9d028bSLiu Bo while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { 4374d7df2c79SJosef Bacik struct btrfs_delayed_ref_head *head; 43750e0adbcfSJosef Bacik struct rb_node *n; 4376e78417d1SJosef Bacik bool pin_bytes = false; 4377acce952bSliubo 4378d7df2c79SJosef Bacik head = rb_entry(node, struct btrfs_delayed_ref_head, 4379d7df2c79SJosef Bacik href_node); 43803069bd26SJosef Bacik if (btrfs_delayed_ref_lock(delayed_refs, head)) 4381b939d1abSJosef Bacik continue; 43823069bd26SJosef Bacik 4383d7df2c79SJosef Bacik spin_lock(&head->lock); 4384e3d03965SLiu Bo while ((n = rb_first_cached(&head->ref_tree)) != NULL) { 43850e0adbcfSJosef Bacik ref = rb_entry(n, struct btrfs_delayed_ref_node, 43860e0adbcfSJosef Bacik ref_node); 4387d7df2c79SJosef Bacik ref->in_tree = 0; 4388e3d03965SLiu Bo rb_erase_cached(&ref->ref_node, &head->ref_tree); 43890e0adbcfSJosef Bacik RB_CLEAR_NODE(&ref->ref_node); 43901d57ee94SWang Xiaoguang if (!list_empty(&ref->add_list)) 43911d57ee94SWang Xiaoguang list_del(&ref->add_list); 4392d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4393d7df2c79SJosef Bacik btrfs_put_delayed_ref(ref); 4394d7df2c79SJosef Bacik } 439554067ae9SJosef Bacik if (head->must_insert_reserved) 4396e78417d1SJosef Bacik pin_bytes = true; 439778a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 4398fa781ceaSJosef Bacik btrfs_delete_ref_head(delayed_refs, head); 4399d7df2c79SJosef Bacik spin_unlock(&head->lock); 4400acce952bSliubo spin_unlock(&delayed_refs->lock); 4401e78417d1SJosef Bacik mutex_unlock(&head->mutex); 4402acce952bSliubo 4403f603bb94SNikolay Borisov if (pin_bytes) { 4404f603bb94SNikolay Borisov struct btrfs_block_group *cache; 4405f603bb94SNikolay Borisov 4406f603bb94SNikolay Borisov cache = btrfs_lookup_block_group(fs_info, head->bytenr); 4407f603bb94SNikolay Borisov BUG_ON(!cache); 4408f603bb94SNikolay Borisov 4409f603bb94SNikolay Borisov spin_lock(&cache->space_info->lock); 4410f603bb94SNikolay Borisov spin_lock(&cache->lock); 4411f603bb94SNikolay Borisov cache->pinned += head->num_bytes; 4412f603bb94SNikolay Borisov btrfs_space_info_update_bytes_pinned(fs_info, 4413f603bb94SNikolay Borisov cache->space_info, head->num_bytes); 4414f603bb94SNikolay Borisov cache->reserved -= head->num_bytes; 4415f603bb94SNikolay Borisov cache->space_info->bytes_reserved -= head->num_bytes; 4416f603bb94SNikolay Borisov spin_unlock(&cache->lock); 4417f603bb94SNikolay Borisov spin_unlock(&cache->space_info->lock); 4418f603bb94SNikolay Borisov percpu_counter_add_batch( 4419f603bb94SNikolay Borisov &cache->space_info->total_bytes_pinned, 4420f603bb94SNikolay Borisov head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH); 4421f603bb94SNikolay Borisov 4422f603bb94SNikolay Borisov btrfs_put_block_group(cache); 4423f603bb94SNikolay Borisov 4424f603bb94SNikolay Borisov btrfs_error_unpin_extent_range(fs_info, head->bytenr, 4425f603bb94SNikolay Borisov head->bytenr + head->num_bytes - 1); 4426f603bb94SNikolay Borisov } 442731890da0SJosef Bacik btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); 4428d278850eSJosef Bacik btrfs_put_delayed_ref_head(head); 4429acce952bSliubo cond_resched(); 4430acce952bSliubo spin_lock(&delayed_refs->lock); 4431acce952bSliubo } 443281f7eb00SJeff Mahoney btrfs_qgroup_destroy_extent_records(trans); 4433acce952bSliubo 4434acce952bSliubo spin_unlock(&delayed_refs->lock); 4435acce952bSliubo 4436acce952bSliubo return ret; 4437acce952bSliubo } 4438acce952bSliubo 4439143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 4440acce952bSliubo { 4441acce952bSliubo struct btrfs_inode *btrfs_inode; 4442acce952bSliubo struct list_head splice; 4443acce952bSliubo 4444acce952bSliubo INIT_LIST_HEAD(&splice); 4445acce952bSliubo 4446eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4447eb73c1b7SMiao Xie list_splice_init(&root->delalloc_inodes, &splice); 4448acce952bSliubo 4449acce952bSliubo while (!list_empty(&splice)) { 4450fe816d0fSNikolay Borisov struct inode *inode = NULL; 4451eb73c1b7SMiao Xie btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 4452acce952bSliubo delalloc_inodes); 4453fe816d0fSNikolay Borisov __btrfs_del_delalloc_inode(root, btrfs_inode); 4454eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4455acce952bSliubo 4456fe816d0fSNikolay Borisov /* 4457fe816d0fSNikolay Borisov * Make sure we get a live inode and that it'll not disappear 4458fe816d0fSNikolay Borisov * meanwhile. 4459fe816d0fSNikolay Borisov */ 4460fe816d0fSNikolay Borisov inode = igrab(&btrfs_inode->vfs_inode); 4461fe816d0fSNikolay Borisov if (inode) { 4462fe816d0fSNikolay Borisov invalidate_inode_pages2(inode->i_mapping); 4463fe816d0fSNikolay Borisov iput(inode); 4464fe816d0fSNikolay Borisov } 4465eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4466acce952bSliubo } 4467eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4468eb73c1b7SMiao Xie } 4469eb73c1b7SMiao Xie 4470eb73c1b7SMiao Xie static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 4471eb73c1b7SMiao Xie { 4472eb73c1b7SMiao Xie struct btrfs_root *root; 4473eb73c1b7SMiao Xie struct list_head splice; 4474eb73c1b7SMiao Xie 4475eb73c1b7SMiao Xie INIT_LIST_HEAD(&splice); 4476eb73c1b7SMiao Xie 4477eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4478eb73c1b7SMiao Xie list_splice_init(&fs_info->delalloc_roots, &splice); 4479eb73c1b7SMiao Xie while (!list_empty(&splice)) { 4480eb73c1b7SMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4481eb73c1b7SMiao Xie delalloc_root); 448200246528SJosef Bacik root = btrfs_grab_root(root); 4483eb73c1b7SMiao Xie BUG_ON(!root); 4484eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4485eb73c1b7SMiao Xie 4486eb73c1b7SMiao Xie btrfs_destroy_delalloc_inodes(root); 448700246528SJosef Bacik btrfs_put_root(root); 4488eb73c1b7SMiao Xie 4489eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4490eb73c1b7SMiao Xie } 4491eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4492acce952bSliubo } 4493acce952bSliubo 44942ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 4495acce952bSliubo struct extent_io_tree *dirty_pages, 4496acce952bSliubo int mark) 4497acce952bSliubo { 4498acce952bSliubo int ret; 4499acce952bSliubo struct extent_buffer *eb; 4500acce952bSliubo u64 start = 0; 4501acce952bSliubo u64 end; 4502acce952bSliubo 4503acce952bSliubo while (1) { 4504acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 4505e6138876SJosef Bacik mark, NULL); 4506acce952bSliubo if (ret) 4507acce952bSliubo break; 4508acce952bSliubo 450991166212SDavid Sterba clear_extent_bits(dirty_pages, start, end, mark); 4510acce952bSliubo while (start <= end) { 45110b246afaSJeff Mahoney eb = find_extent_buffer(fs_info, start); 45120b246afaSJeff Mahoney start += fs_info->nodesize; 4513fd8b2b61SJosef Bacik if (!eb) 4514acce952bSliubo continue; 4515fd8b2b61SJosef Bacik wait_on_extent_buffer_writeback(eb); 4516acce952bSliubo 4517fd8b2b61SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 4518fd8b2b61SJosef Bacik &eb->bflags)) 4519fd8b2b61SJosef Bacik clear_extent_buffer_dirty(eb); 4520fd8b2b61SJosef Bacik free_extent_buffer_stale(eb); 4521acce952bSliubo } 4522acce952bSliubo } 4523acce952bSliubo 4524acce952bSliubo return ret; 4525acce952bSliubo } 4526acce952bSliubo 45272ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 4528fe119a6eSNikolay Borisov struct extent_io_tree *unpin) 4529acce952bSliubo { 4530acce952bSliubo u64 start; 4531acce952bSliubo u64 end; 4532acce952bSliubo int ret; 4533acce952bSliubo 4534acce952bSliubo while (1) { 45350e6ec385SFilipe Manana struct extent_state *cached_state = NULL; 45360e6ec385SFilipe Manana 4537fcd5e742SLu Fengqi /* 4538fcd5e742SLu Fengqi * The btrfs_finish_extent_commit() may get the same range as 4539fcd5e742SLu Fengqi * ours between find_first_extent_bit and clear_extent_dirty. 4540fcd5e742SLu Fengqi * Hence, hold the unused_bg_unpin_mutex to avoid double unpin 4541fcd5e742SLu Fengqi * the same extent range. 4542fcd5e742SLu Fengqi */ 4543fcd5e742SLu Fengqi mutex_lock(&fs_info->unused_bg_unpin_mutex); 4544acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 45450e6ec385SFilipe Manana EXTENT_DIRTY, &cached_state); 4546fcd5e742SLu Fengqi if (ret) { 4547fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4548acce952bSliubo break; 4549fcd5e742SLu Fengqi } 4550acce952bSliubo 45510e6ec385SFilipe Manana clear_extent_dirty(unpin, start, end, &cached_state); 45520e6ec385SFilipe Manana free_extent_state(cached_state); 45532ff7e61eSJeff Mahoney btrfs_error_unpin_extent_range(fs_info, start, end); 4554fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4555acce952bSliubo cond_resched(); 4556acce952bSliubo } 4557acce952bSliubo 4558acce952bSliubo return 0; 4559acce952bSliubo } 4560acce952bSliubo 456132da5386SDavid Sterba static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache) 4562c79a1751SLiu Bo { 4563c79a1751SLiu Bo struct inode *inode; 4564c79a1751SLiu Bo 4565c79a1751SLiu Bo inode = cache->io_ctl.inode; 4566c79a1751SLiu Bo if (inode) { 4567c79a1751SLiu Bo invalidate_inode_pages2(inode->i_mapping); 4568c79a1751SLiu Bo BTRFS_I(inode)->generation = 0; 4569c79a1751SLiu Bo cache->io_ctl.inode = NULL; 4570c79a1751SLiu Bo iput(inode); 4571c79a1751SLiu Bo } 4572bbc37d6eSFilipe Manana ASSERT(cache->io_ctl.pages == NULL); 4573c79a1751SLiu Bo btrfs_put_block_group(cache); 4574c79a1751SLiu Bo } 4575c79a1751SLiu Bo 4576c79a1751SLiu Bo void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, 45772ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4578c79a1751SLiu Bo { 457932da5386SDavid Sterba struct btrfs_block_group *cache; 4580c79a1751SLiu Bo 4581c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4582c79a1751SLiu Bo while (!list_empty(&cur_trans->dirty_bgs)) { 4583c79a1751SLiu Bo cache = list_first_entry(&cur_trans->dirty_bgs, 458432da5386SDavid Sterba struct btrfs_block_group, 4585c79a1751SLiu Bo dirty_list); 4586c79a1751SLiu Bo 4587c79a1751SLiu Bo if (!list_empty(&cache->io_list)) { 4588c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4589c79a1751SLiu Bo list_del_init(&cache->io_list); 4590c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 4591c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4592c79a1751SLiu Bo } 4593c79a1751SLiu Bo 4594c79a1751SLiu Bo list_del_init(&cache->dirty_list); 4595c79a1751SLiu Bo spin_lock(&cache->lock); 4596c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 4597c79a1751SLiu Bo spin_unlock(&cache->lock); 4598c79a1751SLiu Bo 4599c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4600c79a1751SLiu Bo btrfs_put_block_group(cache); 4601ba2c4d4eSJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 4602c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4603c79a1751SLiu Bo } 4604c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4605c79a1751SLiu Bo 460645ae2c18SNikolay Borisov /* 460745ae2c18SNikolay Borisov * Refer to the definition of io_bgs member for details why it's safe 460845ae2c18SNikolay Borisov * to use it without any locking 460945ae2c18SNikolay Borisov */ 4610c79a1751SLiu Bo while (!list_empty(&cur_trans->io_bgs)) { 4611c79a1751SLiu Bo cache = list_first_entry(&cur_trans->io_bgs, 461232da5386SDavid Sterba struct btrfs_block_group, 4613c79a1751SLiu Bo io_list); 4614c79a1751SLiu Bo 4615c79a1751SLiu Bo list_del_init(&cache->io_list); 4616c79a1751SLiu Bo spin_lock(&cache->lock); 4617c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 4618c79a1751SLiu Bo spin_unlock(&cache->lock); 4619c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 4620c79a1751SLiu Bo } 4621c79a1751SLiu Bo } 4622c79a1751SLiu Bo 462349b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 46242ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 462549b25e05SJeff Mahoney { 4626bbbf7243SNikolay Borisov struct btrfs_device *dev, *tmp; 4627bbbf7243SNikolay Borisov 46282ff7e61eSJeff Mahoney btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 4629c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->dirty_bgs)); 4630c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->io_bgs)); 4631c79a1751SLiu Bo 4632bbbf7243SNikolay Borisov list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list, 4633bbbf7243SNikolay Borisov post_commit_list) { 4634bbbf7243SNikolay Borisov list_del_init(&dev->post_commit_list); 4635bbbf7243SNikolay Borisov } 4636bbbf7243SNikolay Borisov 46372ff7e61eSJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, fs_info); 463849b25e05SJeff Mahoney 46394a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 46400b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 464149b25e05SJeff Mahoney 46424a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 46430b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 464449b25e05SJeff Mahoney 4645ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 464667cde344SMiao Xie 46472ff7e61eSJeff Mahoney btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, 464849b25e05SJeff Mahoney EXTENT_DIRTY); 4649fe119a6eSNikolay Borisov btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents); 465049b25e05SJeff Mahoney 46514a9d8bdeSMiao Xie cur_trans->state =TRANS_STATE_COMPLETED; 46524a9d8bdeSMiao Xie wake_up(&cur_trans->commit_wait); 465349b25e05SJeff Mahoney } 465449b25e05SJeff Mahoney 46552ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) 4656acce952bSliubo { 4657acce952bSliubo struct btrfs_transaction *t; 4658acce952bSliubo 46590b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 4660acce952bSliubo 46610b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 46620b246afaSJeff Mahoney while (!list_empty(&fs_info->trans_list)) { 46630b246afaSJeff Mahoney t = list_first_entry(&fs_info->trans_list, 4664724e2315SJosef Bacik struct btrfs_transaction, list); 4665724e2315SJosef Bacik if (t->state >= TRANS_STATE_COMMIT_START) { 46669b64f57dSElena Reshetova refcount_inc(&t->use_count); 46670b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 46682ff7e61eSJeff Mahoney btrfs_wait_for_commit(fs_info, t->transid); 4669724e2315SJosef Bacik btrfs_put_transaction(t); 46700b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4671724e2315SJosef Bacik continue; 4672724e2315SJosef Bacik } 46730b246afaSJeff Mahoney if (t == fs_info->running_transaction) { 4674724e2315SJosef Bacik t->state = TRANS_STATE_COMMIT_DOING; 46750b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4676724e2315SJosef Bacik /* 4677724e2315SJosef Bacik * We wait for 0 num_writers since we don't hold a trans 4678724e2315SJosef Bacik * handle open currently for this transaction. 4679724e2315SJosef Bacik */ 4680724e2315SJosef Bacik wait_event(t->writer_wait, 4681724e2315SJosef Bacik atomic_read(&t->num_writers) == 0); 4682724e2315SJosef Bacik } else { 46830b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4684724e2315SJosef Bacik } 46852ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(t, fs_info); 4686724e2315SJosef Bacik 46870b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 46880b246afaSJeff Mahoney if (t == fs_info->running_transaction) 46890b246afaSJeff Mahoney fs_info->running_transaction = NULL; 4690724e2315SJosef Bacik list_del_init(&t->list); 46910b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4692a4abeea4SJosef Bacik 4693724e2315SJosef Bacik btrfs_put_transaction(t); 46942ff7e61eSJeff Mahoney trace_btrfs_transaction_commit(fs_info->tree_root); 46950b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4696724e2315SJosef Bacik } 46970b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 46980b246afaSJeff Mahoney btrfs_destroy_all_ordered_extents(fs_info); 4699ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 4700ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 47010b246afaSJeff Mahoney btrfs_destroy_all_delalloc_inodes(fs_info); 4702ef67963dSJosef Bacik btrfs_drop_all_logs(fs_info); 47030b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 4704acce952bSliubo 4705acce952bSliubo return 0; 4706acce952bSliubo } 4707