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" 3221adbd5cSStefan Behrens #include "check-integrity.h" 33606686eeSJosef Bacik #include "rcu-string.h" 348dabb742SStefan Behrens #include "dev-replace.h" 3553b381b3SDavid Woodhouse #include "raid56.h" 365ac1d209SJeff Mahoney #include "sysfs.h" 37fcebe456SJosef Bacik #include "qgroup.h" 38ebb8765bSAnand Jain #include "compression.h" 39557ea5ddSQu Wenruo #include "tree-checker.h" 40fd708b81SJosef Bacik #include "ref-verify.h" 41aac0023cSJosef Bacik #include "block-group.h" 42b0643e59SDennis Zhou #include "discard.h" 43f603bb94SNikolay Borisov #include "space-info.h" 44b70f5097SNaohiro Aota #include "zoned.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; 1161941b64bSQu Wenruo 1171941b64bSQu Wenruo /* Optional parameter for submit_bio_start used by direct io */ 1181941b64bSQu Wenruo u64 dio_file_offset; 1198b712842SChris Mason struct btrfs_work work; 1204e4cbee9SChristoph Hellwig blk_status_t status; 12144b8bd7eSChris Mason }; 12244b8bd7eSChris Mason 12385d4e461SChris Mason /* 12485d4e461SChris Mason * Lockdep class keys for extent_buffer->lock's in this root. For a given 12585d4e461SChris Mason * eb, the lockdep key is determined by the btrfs_root it belongs to and 12685d4e461SChris Mason * the level the eb occupies in the tree. 1274008c04aSChris Mason * 12885d4e461SChris Mason * Different roots are used for different purposes and may nest inside each 12985d4e461SChris Mason * other and they require separate keysets. As lockdep keys should be 13085d4e461SChris Mason * static, assign keysets according to the purpose of the root as indicated 1314fd786e6SMisono Tomohiro * by btrfs_root->root_key.objectid. This ensures that all special purpose 1324fd786e6SMisono Tomohiro * roots have separate keysets. 1334008c04aSChris Mason * 13485d4e461SChris Mason * Lock-nesting across peer nodes is always done with the immediate parent 13585d4e461SChris Mason * node locked thus preventing deadlock. As lockdep doesn't know this, use 13685d4e461SChris Mason * subclass to avoid triggering lockdep warning in such cases. 1374008c04aSChris Mason * 13885d4e461SChris Mason * The key is set by the readpage_end_io_hook after the buffer has passed 13985d4e461SChris Mason * csum validation but before the pages are unlocked. It is also set by 14085d4e461SChris Mason * btrfs_init_new_buffer on freshly allocated blocks. 14185d4e461SChris Mason * 14285d4e461SChris Mason * We also add a check to make sure the highest level of the tree is the 14385d4e461SChris Mason * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code 14485d4e461SChris Mason * needs update as well. 1454008c04aSChris Mason */ 1464008c04aSChris Mason #ifdef CONFIG_DEBUG_LOCK_ALLOC 1474008c04aSChris Mason # if BTRFS_MAX_LEVEL != 8 1484008c04aSChris Mason # error 1494008c04aSChris Mason # endif 15085d4e461SChris Mason 151ab1405aaSDavid Sterba #define DEFINE_LEVEL(stem, level) \ 152ab1405aaSDavid Sterba .names[level] = "btrfs-" stem "-0" #level, 153ab1405aaSDavid Sterba 154ab1405aaSDavid Sterba #define DEFINE_NAME(stem) \ 155ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 0) \ 156ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 1) \ 157ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 2) \ 158ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 3) \ 159ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 4) \ 160ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 5) \ 161ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 6) \ 162ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 7) 163ab1405aaSDavid Sterba 16485d4e461SChris Mason static struct btrfs_lockdep_keyset { 16585d4e461SChris Mason u64 id; /* root objectid */ 166ab1405aaSDavid Sterba /* Longest entry: btrfs-free-space-00 */ 167387824afSDavid Sterba char names[BTRFS_MAX_LEVEL][20]; 168387824afSDavid Sterba struct lock_class_key keys[BTRFS_MAX_LEVEL]; 16985d4e461SChris Mason } btrfs_lockdep_keysets[] = { 170ab1405aaSDavid Sterba { .id = BTRFS_ROOT_TREE_OBJECTID, DEFINE_NAME("root") }, 171ab1405aaSDavid Sterba { .id = BTRFS_EXTENT_TREE_OBJECTID, DEFINE_NAME("extent") }, 172ab1405aaSDavid Sterba { .id = BTRFS_CHUNK_TREE_OBJECTID, DEFINE_NAME("chunk") }, 173ab1405aaSDavid Sterba { .id = BTRFS_DEV_TREE_OBJECTID, DEFINE_NAME("dev") }, 174ab1405aaSDavid Sterba { .id = BTRFS_CSUM_TREE_OBJECTID, DEFINE_NAME("csum") }, 175ab1405aaSDavid Sterba { .id = BTRFS_QUOTA_TREE_OBJECTID, DEFINE_NAME("quota") }, 176ab1405aaSDavid Sterba { .id = BTRFS_TREE_LOG_OBJECTID, DEFINE_NAME("log") }, 177ab1405aaSDavid Sterba { .id = BTRFS_TREE_RELOC_OBJECTID, DEFINE_NAME("treloc") }, 178ab1405aaSDavid Sterba { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, DEFINE_NAME("dreloc") }, 179ab1405aaSDavid Sterba { .id = BTRFS_UUID_TREE_OBJECTID, DEFINE_NAME("uuid") }, 180ab1405aaSDavid Sterba { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, DEFINE_NAME("free-space") }, 181ab1405aaSDavid Sterba { .id = 0, DEFINE_NAME("tree") }, 1824008c04aSChris Mason }; 18385d4e461SChris Mason 184ab1405aaSDavid Sterba #undef DEFINE_LEVEL 185ab1405aaSDavid Sterba #undef DEFINE_NAME 18685d4e461SChris Mason 18785d4e461SChris Mason void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb, 18885d4e461SChris Mason int level) 18985d4e461SChris Mason { 19085d4e461SChris Mason struct btrfs_lockdep_keyset *ks; 19185d4e461SChris Mason 19285d4e461SChris Mason BUG_ON(level >= ARRAY_SIZE(ks->keys)); 19385d4e461SChris Mason 19485d4e461SChris Mason /* find the matching keyset, id 0 is the default entry */ 19585d4e461SChris Mason for (ks = btrfs_lockdep_keysets; ks->id; ks++) 19685d4e461SChris Mason if (ks->id == objectid) 19785d4e461SChris Mason break; 19885d4e461SChris Mason 19985d4e461SChris Mason lockdep_set_class_and_name(&eb->lock, 20085d4e461SChris Mason &ks->keys[level], ks->names[level]); 20185d4e461SChris Mason } 20285d4e461SChris Mason 2034008c04aSChris Mason #endif 2044008c04aSChris Mason 205d352ac68SChris Mason /* 2062996e1f8SJohannes Thumshirn * Compute the csum of a btree block and store the result to provided buffer. 207d352ac68SChris Mason */ 208c67b3892SDavid Sterba static void csum_tree_block(struct extent_buffer *buf, u8 *result) 20919c00ddcSChris Mason { 210d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = buf->fs_info; 211e9be5a30SDavid Sterba const int num_pages = fs_info->nodesize >> PAGE_SHIFT; 212a26663e7SQu Wenruo const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize); 213d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 21419c00ddcSChris Mason char *kaddr; 215e9be5a30SDavid Sterba int i; 216d5178578SJohannes Thumshirn 217d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 218d5178578SJohannes Thumshirn crypto_shash_init(shash); 219a26663e7SQu Wenruo kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start); 220e9be5a30SDavid Sterba crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE, 221a26663e7SQu Wenruo first_page_part - BTRFS_CSUM_SIZE); 22219c00ddcSChris Mason 223e9be5a30SDavid Sterba for (i = 1; i < num_pages; i++) { 224e9be5a30SDavid Sterba kaddr = page_address(buf->pages[i]); 225e9be5a30SDavid Sterba crypto_shash_update(shash, kaddr, PAGE_SIZE); 22619c00ddcSChris Mason } 22771a63551SDavid Sterba memset(result, 0, BTRFS_CSUM_SIZE); 228d5178578SJohannes Thumshirn crypto_shash_final(shash, result); 22919c00ddcSChris Mason } 23019c00ddcSChris Mason 231d352ac68SChris Mason /* 232d352ac68SChris Mason * we can't consider a given block up to date unless the transid of the 233d352ac68SChris Mason * block matches the transid in the parent node's pointer. This is how we 234d352ac68SChris Mason * detect blocks that either didn't get written at all or got written 235d352ac68SChris Mason * in the wrong place. 236d352ac68SChris Mason */ 2371259ab75SChris Mason static int verify_parent_transid(struct extent_io_tree *io_tree, 238b9fab919SChris Mason struct extent_buffer *eb, u64 parent_transid, 239b9fab919SChris Mason int atomic) 2401259ab75SChris Mason { 2412ac55d41SJosef Bacik struct extent_state *cached_state = NULL; 2421259ab75SChris Mason int ret; 2432755a0deSDavid Sterba bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB); 2441259ab75SChris Mason 2451259ab75SChris Mason if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 2461259ab75SChris Mason return 0; 2471259ab75SChris Mason 248b9fab919SChris Mason if (atomic) 249b9fab919SChris Mason return -EAGAIN; 250b9fab919SChris Mason 251ac5887c8SJosef Bacik if (need_lock) 252a26e8c9fSJosef Bacik btrfs_tree_read_lock(eb); 253a26e8c9fSJosef Bacik 2542ac55d41SJosef Bacik lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 255ff13db41SDavid Sterba &cached_state); 2560b32f4bbSJosef Bacik if (extent_buffer_uptodate(eb) && 2571259ab75SChris Mason btrfs_header_generation(eb) == parent_transid) { 2581259ab75SChris Mason ret = 0; 2591259ab75SChris Mason goto out; 2601259ab75SChris Mason } 26194647322SDavid Sterba btrfs_err_rl(eb->fs_info, 26294647322SDavid Sterba "parent transid verify failed on %llu wanted %llu found %llu", 26394647322SDavid Sterba eb->start, 26429549aecSWang Shilong parent_transid, btrfs_header_generation(eb)); 2651259ab75SChris Mason ret = 1; 266a26e8c9fSJosef Bacik 267a26e8c9fSJosef Bacik /* 268a26e8c9fSJosef Bacik * Things reading via commit roots that don't have normal protection, 269a26e8c9fSJosef Bacik * like send, can have a really old block in cache that may point at a 27001327610SNicholas D Steeves * block that has been freed and re-allocated. So don't clear uptodate 271a26e8c9fSJosef Bacik * if we find an eb that is under IO (dirty/writeback) because we could 272a26e8c9fSJosef Bacik * end up reading in the stale data and then writing it back out and 273a26e8c9fSJosef Bacik * making everybody very sad. 274a26e8c9fSJosef Bacik */ 275a26e8c9fSJosef Bacik if (!extent_buffer_under_io(eb)) 2760b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 27733958dc6SChris Mason out: 2782ac55d41SJosef Bacik unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 279e43bbe5eSDavid Sterba &cached_state); 280472b909fSJosef Bacik if (need_lock) 281ac5887c8SJosef Bacik btrfs_tree_read_unlock(eb); 2821259ab75SChris Mason return ret; 2831259ab75SChris Mason } 2841259ab75SChris Mason 285e7e16f48SJohannes Thumshirn static bool btrfs_supported_super_csum(u16 csum_type) 286e7e16f48SJohannes Thumshirn { 287e7e16f48SJohannes Thumshirn switch (csum_type) { 288e7e16f48SJohannes Thumshirn case BTRFS_CSUM_TYPE_CRC32: 2893951e7f0SJohannes Thumshirn case BTRFS_CSUM_TYPE_XXHASH: 2903831bf00SJohannes Thumshirn case BTRFS_CSUM_TYPE_SHA256: 291352ae07bSDavid Sterba case BTRFS_CSUM_TYPE_BLAKE2: 292e7e16f48SJohannes Thumshirn return true; 293e7e16f48SJohannes Thumshirn default: 294e7e16f48SJohannes Thumshirn return false; 295e7e16f48SJohannes Thumshirn } 296e7e16f48SJohannes Thumshirn } 297e7e16f48SJohannes Thumshirn 298d352ac68SChris Mason /* 2991104a885SDavid Sterba * Return 0 if the superblock checksum type matches the checksum value of that 3001104a885SDavid Sterba * algorithm. Pass the raw disk superblock data. 3011104a885SDavid Sterba */ 302ab8d0fc4SJeff Mahoney static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info, 303ab8d0fc4SJeff Mahoney char *raw_disk_sb) 3041104a885SDavid Sterba { 3051104a885SDavid Sterba struct btrfs_super_block *disk_sb = 3061104a885SDavid Sterba (struct btrfs_super_block *)raw_disk_sb; 30751bce6c9SJohannes Thumshirn char result[BTRFS_CSUM_SIZE]; 308d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 309d5178578SJohannes Thumshirn 310d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 3111104a885SDavid Sterba 3121104a885SDavid Sterba /* 3131104a885SDavid Sterba * The super_block structure does not span the whole 31451bce6c9SJohannes Thumshirn * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is 31551bce6c9SJohannes Thumshirn * filled with zeros and is included in the checksum. 3161104a885SDavid Sterba */ 317fd08001fSEric Biggers crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE, 318fd08001fSEric Biggers BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result); 3191104a885SDavid Sterba 32055fc29beSDavid Sterba if (memcmp(disk_sb->csum, result, fs_info->csum_size)) 321e7e16f48SJohannes Thumshirn return 1; 3221104a885SDavid Sterba 323e7e16f48SJohannes Thumshirn return 0; 3241104a885SDavid Sterba } 3251104a885SDavid Sterba 326e064d5e9SDavid Sterba int btrfs_verify_level_key(struct extent_buffer *eb, int level, 327ff76a864SLiu Bo struct btrfs_key *first_key, u64 parent_transid) 328581c1760SQu Wenruo { 329e064d5e9SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 330581c1760SQu Wenruo int found_level; 331581c1760SQu Wenruo struct btrfs_key found_key; 332581c1760SQu Wenruo int ret; 333581c1760SQu Wenruo 334581c1760SQu Wenruo found_level = btrfs_header_level(eb); 335581c1760SQu Wenruo if (found_level != level) { 33663489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 33763489055SQu Wenruo KERN_ERR "BTRFS: tree level check failed\n"); 338581c1760SQu Wenruo btrfs_err(fs_info, 339581c1760SQu Wenruo "tree level mismatch detected, bytenr=%llu level expected=%u has=%u", 340581c1760SQu Wenruo eb->start, level, found_level); 341581c1760SQu Wenruo return -EIO; 342581c1760SQu Wenruo } 343581c1760SQu Wenruo 344581c1760SQu Wenruo if (!first_key) 345581c1760SQu Wenruo return 0; 346581c1760SQu Wenruo 3475d41be6fSQu Wenruo /* 3485d41be6fSQu Wenruo * For live tree block (new tree blocks in current transaction), 3495d41be6fSQu Wenruo * we need proper lock context to avoid race, which is impossible here. 3505d41be6fSQu Wenruo * So we only checks tree blocks which is read from disk, whose 3515d41be6fSQu Wenruo * generation <= fs_info->last_trans_committed. 3525d41be6fSQu Wenruo */ 3535d41be6fSQu Wenruo if (btrfs_header_generation(eb) > fs_info->last_trans_committed) 3545d41be6fSQu Wenruo return 0; 35562fdaa52SQu Wenruo 35662fdaa52SQu Wenruo /* We have @first_key, so this @eb must have at least one item */ 35762fdaa52SQu Wenruo if (btrfs_header_nritems(eb) == 0) { 35862fdaa52SQu Wenruo btrfs_err(fs_info, 35962fdaa52SQu Wenruo "invalid tree nritems, bytenr=%llu nritems=0 expect >0", 36062fdaa52SQu Wenruo eb->start); 36162fdaa52SQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 36262fdaa52SQu Wenruo return -EUCLEAN; 36362fdaa52SQu Wenruo } 36462fdaa52SQu Wenruo 365581c1760SQu Wenruo if (found_level) 366581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &found_key, 0); 367581c1760SQu Wenruo else 368581c1760SQu Wenruo btrfs_item_key_to_cpu(eb, &found_key, 0); 369581c1760SQu Wenruo ret = btrfs_comp_cpu_keys(first_key, &found_key); 370581c1760SQu Wenruo 371581c1760SQu Wenruo if (ret) { 37263489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 37363489055SQu Wenruo KERN_ERR "BTRFS: tree first key check failed\n"); 374581c1760SQu Wenruo btrfs_err(fs_info, 375ff76a864SLiu Bo "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", 376ff76a864SLiu Bo eb->start, parent_transid, first_key->objectid, 377ff76a864SLiu Bo first_key->type, first_key->offset, 378ff76a864SLiu Bo found_key.objectid, found_key.type, 379ff76a864SLiu Bo found_key.offset); 380581c1760SQu Wenruo } 381581c1760SQu Wenruo return ret; 382581c1760SQu Wenruo } 383581c1760SQu Wenruo 3841104a885SDavid Sterba /* 385d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 386d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 387581c1760SQu Wenruo * 388581c1760SQu Wenruo * @parent_transid: expected transid, skip check if 0 389581c1760SQu Wenruo * @level: expected level, mandatory check 390581c1760SQu Wenruo * @first_key: expected key of first slot, skip check if NULL 391d352ac68SChris Mason */ 3925ab12d1fSDavid Sterba static int btree_read_extent_buffer_pages(struct extent_buffer *eb, 393581c1760SQu Wenruo u64 parent_transid, int level, 394581c1760SQu Wenruo struct btrfs_key *first_key) 395f188591eSChris Mason { 3965ab12d1fSDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 397f188591eSChris Mason struct extent_io_tree *io_tree; 398ea466794SJosef Bacik int failed = 0; 399f188591eSChris Mason int ret; 400f188591eSChris Mason int num_copies = 0; 401f188591eSChris Mason int mirror_num = 0; 402ea466794SJosef Bacik int failed_mirror = 0; 403f188591eSChris Mason 4040b246afaSJeff Mahoney io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; 405f188591eSChris Mason while (1) { 406f8397d69SNikolay Borisov clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 407c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num); 408256dd1bbSStefan Behrens if (!ret) { 409581c1760SQu Wenruo if (verify_parent_transid(io_tree, eb, 410b9fab919SChris Mason parent_transid, 0)) 411256dd1bbSStefan Behrens ret = -EIO; 412e064d5e9SDavid Sterba else if (btrfs_verify_level_key(eb, level, 413ff76a864SLiu Bo first_key, parent_transid)) 414581c1760SQu Wenruo ret = -EUCLEAN; 415581c1760SQu Wenruo else 416581c1760SQu Wenruo break; 417256dd1bbSStefan Behrens } 418d397712bSChris Mason 4190b246afaSJeff Mahoney num_copies = btrfs_num_copies(fs_info, 420f188591eSChris Mason eb->start, eb->len); 4214235298eSChris Mason if (num_copies == 1) 422ea466794SJosef Bacik break; 4234235298eSChris Mason 4245cf1ab56SJosef Bacik if (!failed_mirror) { 4255cf1ab56SJosef Bacik failed = 1; 4265cf1ab56SJosef Bacik failed_mirror = eb->read_mirror; 4275cf1ab56SJosef Bacik } 4285cf1ab56SJosef Bacik 429f188591eSChris Mason mirror_num++; 430ea466794SJosef Bacik if (mirror_num == failed_mirror) 431ea466794SJosef Bacik mirror_num++; 432ea466794SJosef Bacik 4334235298eSChris Mason if (mirror_num > num_copies) 434ea466794SJosef Bacik break; 435f188591eSChris Mason } 436ea466794SJosef Bacik 437c0901581SStefan Behrens if (failed && !ret && failed_mirror) 43820a1fbf9SDavid Sterba btrfs_repair_eb_io_failure(eb, failed_mirror); 439ea466794SJosef Bacik 440ea466794SJosef Bacik return ret; 441f188591eSChris Mason } 44219c00ddcSChris Mason 443d352ac68SChris Mason /* 444ac303b69SQu Wenruo * Checksum a dirty tree block before IO. This has extra checks to make sure 445ac303b69SQu Wenruo * we only fill in the checksum field in the first page of a multi-page block. 446ac303b69SQu Wenruo * For subpage extent buffers we need bvec to also read the offset in the page. 447d352ac68SChris Mason */ 448ac303b69SQu Wenruo static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec) 44919c00ddcSChris Mason { 450ac303b69SQu Wenruo struct page *page = bvec->bv_page; 4514eee4fa4SMiao Xie u64 start = page_offset(page); 45219c00ddcSChris Mason u64 found_start; 4532996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 45419c00ddcSChris Mason struct extent_buffer *eb; 4558d47a0d8SQu Wenruo int ret; 456f188591eSChris Mason 4574f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 4584f2de97aSJosef Bacik if (page != eb->pages[0]) 4594f2de97aSJosef Bacik return 0; 4600f805531SAlex Lyakas 46119c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 462d3575156SNaohiro Aota 463d3575156SNaohiro Aota if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) { 464d3575156SNaohiro Aota WARN_ON(found_start != 0); 465d3575156SNaohiro Aota return 0; 466d3575156SNaohiro Aota } 467d3575156SNaohiro Aota 4680f805531SAlex Lyakas /* 4690f805531SAlex Lyakas * Please do not consolidate these warnings into a single if. 4700f805531SAlex Lyakas * It is useful to know what went wrong. 4710f805531SAlex Lyakas */ 4720f805531SAlex Lyakas if (WARN_ON(found_start != start)) 4730f805531SAlex Lyakas return -EUCLEAN; 4740f805531SAlex Lyakas if (WARN_ON(!PageUptodate(page))) 4750f805531SAlex Lyakas return -EUCLEAN; 4760f805531SAlex Lyakas 477de37aa51SNikolay Borisov ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid, 4789a8658e3SDavid Sterba offsetof(struct btrfs_header, fsid), 4799a8658e3SDavid Sterba BTRFS_FSID_SIZE) == 0); 4800f805531SAlex Lyakas 481c67b3892SDavid Sterba csum_tree_block(eb, result); 4822996e1f8SJohannes Thumshirn 4838d47a0d8SQu Wenruo if (btrfs_header_level(eb)) 4848d47a0d8SQu Wenruo ret = btrfs_check_node(eb); 4858d47a0d8SQu Wenruo else 4868d47a0d8SQu Wenruo ret = btrfs_check_leaf_full(eb); 4878d47a0d8SQu Wenruo 4888d47a0d8SQu Wenruo if (ret < 0) { 489c06631b0SQu Wenruo btrfs_print_tree(eb, 0); 4908d47a0d8SQu Wenruo btrfs_err(fs_info, 4918d47a0d8SQu Wenruo "block=%llu write time tree block corruption detected", 4928d47a0d8SQu Wenruo eb->start); 493c06631b0SQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 4948d47a0d8SQu Wenruo return ret; 4958d47a0d8SQu Wenruo } 496713cebfbSDavid Sterba write_extent_buffer(eb, result, 0, fs_info->csum_size); 4978d47a0d8SQu Wenruo 4982996e1f8SJohannes Thumshirn return 0; 49919c00ddcSChris Mason } 50019c00ddcSChris Mason 501b0c9b3b0SDavid Sterba static int check_tree_block_fsid(struct extent_buffer *eb) 5022b82032cSYan Zheng { 503b0c9b3b0SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 504944d3f9fSNikolay Borisov struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 50544880fdcSAnand Jain u8 fsid[BTRFS_FSID_SIZE]; 506944d3f9fSNikolay Borisov u8 *metadata_uuid; 5072b82032cSYan Zheng 5089a8658e3SDavid Sterba read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid), 5099a8658e3SDavid Sterba BTRFS_FSID_SIZE); 5107239ff4bSNikolay Borisov /* 511944d3f9fSNikolay Borisov * Checking the incompat flag is only valid for the current fs. For 512944d3f9fSNikolay Borisov * seed devices it's forbidden to have their uuid changed so reading 513944d3f9fSNikolay Borisov * ->fsid in this case is fine 5147239ff4bSNikolay Borisov */ 515944d3f9fSNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID)) 5167239ff4bSNikolay Borisov metadata_uuid = fs_devices->metadata_uuid; 5177239ff4bSNikolay Borisov else 5187239ff4bSNikolay Borisov metadata_uuid = fs_devices->fsid; 5197239ff4bSNikolay Borisov 520944d3f9fSNikolay Borisov if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) 521944d3f9fSNikolay Borisov return 0; 522944d3f9fSNikolay Borisov 523944d3f9fSNikolay Borisov list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) 524944d3f9fSNikolay Borisov if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE)) 525944d3f9fSNikolay Borisov return 0; 526944d3f9fSNikolay Borisov 527944d3f9fSNikolay Borisov return 1; 5282b82032cSYan Zheng } 5292b82032cSYan Zheng 53077bf40a2SQu Wenruo /* Do basic extent buffer checks at read time */ 53177bf40a2SQu Wenruo static int validate_extent_buffer(struct extent_buffer *eb) 532ce9adaa5SChris Mason { 53377bf40a2SQu Wenruo struct btrfs_fs_info *fs_info = eb->fs_info; 534ce9adaa5SChris Mason u64 found_start; 53577bf40a2SQu Wenruo const u32 csum_size = fs_info->csum_size; 53677bf40a2SQu Wenruo u8 found_level; 5372996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 53877bf40a2SQu Wenruo int ret = 0; 539ea466794SJosef Bacik 540ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 541727011e0SChris Mason if (found_start != eb->start) { 542893bf4b1SSu Yue btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu", 543893bf4b1SSu Yue eb->start, found_start); 544f188591eSChris Mason ret = -EIO; 54577bf40a2SQu Wenruo goto out; 546ce9adaa5SChris Mason } 547b0c9b3b0SDavid Sterba if (check_tree_block_fsid(eb)) { 54802873e43SZhao Lei btrfs_err_rl(fs_info, "bad fsid on block %llu", 54994647322SDavid Sterba eb->start); 5501259ab75SChris Mason ret = -EIO; 55177bf40a2SQu Wenruo goto out; 5521259ab75SChris Mason } 553ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 5541c24c3ceSJosef Bacik if (found_level >= BTRFS_MAX_LEVEL) { 555893bf4b1SSu Yue btrfs_err(fs_info, "bad tree block level %d on %llu", 556893bf4b1SSu Yue (int)btrfs_header_level(eb), eb->start); 5571c24c3ceSJosef Bacik ret = -EIO; 55877bf40a2SQu Wenruo goto out; 5591c24c3ceSJosef Bacik } 560ce9adaa5SChris Mason 561c67b3892SDavid Sterba csum_tree_block(eb, result); 562a826d6dcSJosef Bacik 5632996e1f8SJohannes Thumshirn if (memcmp_extent_buffer(eb, result, 0, csum_size)) { 56435be8851SJohannes Thumshirn u8 val[BTRFS_CSUM_SIZE] = { 0 }; 5652996e1f8SJohannes Thumshirn 5662996e1f8SJohannes Thumshirn read_extent_buffer(eb, &val, 0, csum_size); 5672996e1f8SJohannes Thumshirn btrfs_warn_rl(fs_info, 56835be8851SJohannes Thumshirn "%s checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d", 5692996e1f8SJohannes Thumshirn fs_info->sb->s_id, eb->start, 57035be8851SJohannes Thumshirn CSUM_FMT_VALUE(csum_size, val), 57135be8851SJohannes Thumshirn CSUM_FMT_VALUE(csum_size, result), 57235be8851SJohannes Thumshirn btrfs_header_level(eb)); 5732996e1f8SJohannes Thumshirn ret = -EUCLEAN; 57477bf40a2SQu Wenruo goto out; 5752996e1f8SJohannes Thumshirn } 5762996e1f8SJohannes Thumshirn 577a826d6dcSJosef Bacik /* 578a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 579a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 580a826d6dcSJosef Bacik * return -EIO. 581a826d6dcSJosef Bacik */ 5821c4360eeSDavid Sterba if (found_level == 0 && btrfs_check_leaf_full(eb)) { 583a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 584a826d6dcSJosef Bacik ret = -EIO; 585a826d6dcSJosef Bacik } 586ce9adaa5SChris Mason 587813fd1dcSDavid Sterba if (found_level > 0 && btrfs_check_node(eb)) 588053ab70fSLiu Bo ret = -EIO; 589053ab70fSLiu Bo 5900b32f4bbSJosef Bacik if (!ret) 5910b32f4bbSJosef Bacik set_extent_buffer_uptodate(eb); 59275391f0dSQu Wenruo else 59375391f0dSQu Wenruo btrfs_err(fs_info, 59475391f0dSQu Wenruo "block=%llu read time tree block corruption detected", 59575391f0dSQu Wenruo eb->start); 59677bf40a2SQu Wenruo out: 59777bf40a2SQu Wenruo return ret; 59877bf40a2SQu Wenruo } 59977bf40a2SQu Wenruo 600371cdc07SQu Wenruo static int validate_subpage_buffer(struct page *page, u64 start, u64 end, 601371cdc07SQu Wenruo int mirror) 602371cdc07SQu Wenruo { 603371cdc07SQu Wenruo struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); 604371cdc07SQu Wenruo struct extent_buffer *eb; 605371cdc07SQu Wenruo bool reads_done; 606371cdc07SQu Wenruo int ret = 0; 607371cdc07SQu Wenruo 608371cdc07SQu Wenruo /* 609371cdc07SQu Wenruo * We don't allow bio merge for subpage metadata read, so we should 610371cdc07SQu Wenruo * only get one eb for each endio hook. 611371cdc07SQu Wenruo */ 612371cdc07SQu Wenruo ASSERT(end == start + fs_info->nodesize - 1); 613371cdc07SQu Wenruo ASSERT(PagePrivate(page)); 614371cdc07SQu Wenruo 615371cdc07SQu Wenruo eb = find_extent_buffer(fs_info, start); 616371cdc07SQu Wenruo /* 617371cdc07SQu Wenruo * When we are reading one tree block, eb must have been inserted into 618371cdc07SQu Wenruo * the radix tree. If not, something is wrong. 619371cdc07SQu Wenruo */ 620371cdc07SQu Wenruo ASSERT(eb); 621371cdc07SQu Wenruo 622371cdc07SQu Wenruo reads_done = atomic_dec_and_test(&eb->io_pages); 623371cdc07SQu Wenruo /* Subpage read must finish in page read */ 624371cdc07SQu Wenruo ASSERT(reads_done); 625371cdc07SQu Wenruo 626371cdc07SQu Wenruo eb->read_mirror = mirror; 627371cdc07SQu Wenruo if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 628371cdc07SQu Wenruo ret = -EIO; 629371cdc07SQu Wenruo goto err; 630371cdc07SQu Wenruo } 631371cdc07SQu Wenruo ret = validate_extent_buffer(eb); 632371cdc07SQu Wenruo if (ret < 0) 633371cdc07SQu Wenruo goto err; 634371cdc07SQu Wenruo 635371cdc07SQu Wenruo if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 636371cdc07SQu Wenruo btree_readahead_hook(eb, ret); 637371cdc07SQu Wenruo 638371cdc07SQu Wenruo set_extent_buffer_uptodate(eb); 639371cdc07SQu Wenruo 640371cdc07SQu Wenruo free_extent_buffer(eb); 641371cdc07SQu Wenruo return ret; 642371cdc07SQu Wenruo err: 643371cdc07SQu Wenruo /* 644371cdc07SQu Wenruo * end_bio_extent_readpage decrements io_pages in case of error, 645371cdc07SQu Wenruo * make sure it has something to decrement. 646371cdc07SQu Wenruo */ 647371cdc07SQu Wenruo atomic_inc(&eb->io_pages); 648371cdc07SQu Wenruo clear_extent_buffer_uptodate(eb); 649371cdc07SQu Wenruo free_extent_buffer(eb); 650371cdc07SQu Wenruo return ret; 651371cdc07SQu Wenruo } 652371cdc07SQu Wenruo 6538e1dc982SQu Wenruo int btrfs_validate_metadata_buffer(struct btrfs_io_bio *io_bio, 65477bf40a2SQu Wenruo struct page *page, u64 start, u64 end, 65577bf40a2SQu Wenruo int mirror) 65677bf40a2SQu Wenruo { 65777bf40a2SQu Wenruo struct extent_buffer *eb; 65877bf40a2SQu Wenruo int ret = 0; 65977bf40a2SQu Wenruo int reads_done; 66077bf40a2SQu Wenruo 66177bf40a2SQu Wenruo ASSERT(page->private); 662371cdc07SQu Wenruo 663371cdc07SQu Wenruo if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE) 664371cdc07SQu Wenruo return validate_subpage_buffer(page, start, end, mirror); 665371cdc07SQu Wenruo 66677bf40a2SQu Wenruo eb = (struct extent_buffer *)page->private; 66777bf40a2SQu Wenruo 66877bf40a2SQu Wenruo /* 66977bf40a2SQu Wenruo * The pending IO might have been the only thing that kept this buffer 67077bf40a2SQu Wenruo * in memory. Make sure we have a ref for all this other checks 67177bf40a2SQu Wenruo */ 67277bf40a2SQu Wenruo atomic_inc(&eb->refs); 67377bf40a2SQu Wenruo 67477bf40a2SQu Wenruo reads_done = atomic_dec_and_test(&eb->io_pages); 67577bf40a2SQu Wenruo if (!reads_done) 67677bf40a2SQu Wenruo goto err; 67777bf40a2SQu Wenruo 67877bf40a2SQu Wenruo eb->read_mirror = mirror; 67977bf40a2SQu Wenruo if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 68077bf40a2SQu Wenruo ret = -EIO; 68177bf40a2SQu Wenruo goto err; 68277bf40a2SQu Wenruo } 68377bf40a2SQu Wenruo ret = validate_extent_buffer(eb); 684ce9adaa5SChris Mason err: 68579fb65a1SJosef Bacik if (reads_done && 68679fb65a1SJosef Bacik test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 687d48d71aaSDavid Sterba btree_readahead_hook(eb, ret); 6884bb31e92SArne Jansen 68953b381b3SDavid Woodhouse if (ret) { 69053b381b3SDavid Woodhouse /* 69153b381b3SDavid Woodhouse * our io error hook is going to dec the io pages 69253b381b3SDavid Woodhouse * again, we have to make sure it has something 69353b381b3SDavid Woodhouse * to decrement 69453b381b3SDavid Woodhouse */ 69553b381b3SDavid Woodhouse atomic_inc(&eb->io_pages); 6960b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 69753b381b3SDavid Woodhouse } 6980b32f4bbSJosef Bacik free_extent_buffer(eb); 69977bf40a2SQu Wenruo 700f188591eSChris Mason return ret; 701ce9adaa5SChris Mason } 702ce9adaa5SChris Mason 7034246a0b6SChristoph Hellwig static void end_workqueue_bio(struct bio *bio) 704ce9adaa5SChris Mason { 70597eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq = bio->bi_private; 706ce9adaa5SChris Mason struct btrfs_fs_info *fs_info; 7079e0af237SLiu Bo struct btrfs_workqueue *wq; 708ce9adaa5SChris Mason 709ce9adaa5SChris Mason fs_info = end_io_wq->info; 7104e4cbee9SChristoph Hellwig end_io_wq->status = bio->bi_status; 711d20f7043SChris Mason 712cfe94440SNaohiro Aota if (btrfs_op(bio) == BTRFS_MAP_WRITE) { 713a0cac0ecSOmar Sandoval if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) 7149e0af237SLiu Bo wq = fs_info->endio_meta_write_workers; 715a0cac0ecSOmar Sandoval else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) 7169e0af237SLiu Bo wq = fs_info->endio_freespace_worker; 717a0cac0ecSOmar Sandoval else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 7189e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 719a0cac0ecSOmar Sandoval else 7209e0af237SLiu Bo wq = fs_info->endio_write_workers; 7219e0af237SLiu Bo } else { 7225c047a69SOmar Sandoval if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 7239e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 724a0cac0ecSOmar Sandoval else if (end_io_wq->metadata) 7259e0af237SLiu Bo wq = fs_info->endio_meta_workers; 726a0cac0ecSOmar Sandoval else 7279e0af237SLiu Bo wq = fs_info->endio_workers; 7289e0af237SLiu Bo } 7299e0af237SLiu Bo 730a0cac0ecSOmar Sandoval btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL); 7319e0af237SLiu Bo btrfs_queue_work(wq, &end_io_wq->work); 732ce9adaa5SChris Mason } 733ce9adaa5SChris Mason 7344e4cbee9SChristoph Hellwig blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 735bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata) 7360b86a832SChris Mason { 73797eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 7388b110e39SMiao Xie 73997eb6b69SDavid Sterba end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS); 740ce9adaa5SChris Mason if (!end_io_wq) 7414e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 742ce9adaa5SChris Mason 743ce9adaa5SChris Mason end_io_wq->private = bio->bi_private; 744ce9adaa5SChris Mason end_io_wq->end_io = bio->bi_end_io; 74522c59948SChris Mason end_io_wq->info = info; 7464e4cbee9SChristoph Hellwig end_io_wq->status = 0; 747ce9adaa5SChris Mason end_io_wq->bio = bio; 74822c59948SChris Mason end_io_wq->metadata = metadata; 749ce9adaa5SChris Mason 750ce9adaa5SChris Mason bio->bi_private = end_io_wq; 751ce9adaa5SChris Mason bio->bi_end_io = end_workqueue_bio; 75222c59948SChris Mason return 0; 75322c59948SChris Mason } 75422c59948SChris Mason 7554a69a410SChris Mason static void run_one_async_start(struct btrfs_work *work) 7564a69a410SChris Mason { 7574a69a410SChris Mason struct async_submit_bio *async; 7584e4cbee9SChristoph Hellwig blk_status_t ret; 7594a69a410SChris Mason 7604a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 7611941b64bSQu Wenruo ret = async->submit_bio_start(async->inode, async->bio, 7621941b64bSQu Wenruo async->dio_file_offset); 76379787eaaSJeff Mahoney if (ret) 7644e4cbee9SChristoph Hellwig async->status = ret; 7654a69a410SChris Mason } 7664a69a410SChris Mason 76706ea01b1SDavid Sterba /* 76806ea01b1SDavid Sterba * In order to insert checksums into the metadata in large chunks, we wait 76906ea01b1SDavid Sterba * until bio submission time. All the pages in the bio are checksummed and 77006ea01b1SDavid Sterba * sums are attached onto the ordered extent record. 77106ea01b1SDavid Sterba * 77206ea01b1SDavid Sterba * At IO completion time the csums attached on the ordered extent record are 77306ea01b1SDavid Sterba * inserted into the tree. 77406ea01b1SDavid Sterba */ 7754a69a410SChris Mason static void run_one_async_done(struct btrfs_work *work) 7768b712842SChris Mason { 7778b712842SChris Mason struct async_submit_bio *async; 77806ea01b1SDavid Sterba struct inode *inode; 77906ea01b1SDavid Sterba blk_status_t ret; 7808b712842SChris Mason 7818b712842SChris Mason async = container_of(work, struct async_submit_bio, work); 7828896a08dSQu Wenruo inode = async->inode; 7834854ddd0SChris Mason 784bb7ab3b9SAdam Buchbinder /* If an error occurred we just want to clean up the bio and move on */ 7854e4cbee9SChristoph Hellwig if (async->status) { 7864e4cbee9SChristoph Hellwig async->bio->bi_status = async->status; 7874246a0b6SChristoph Hellwig bio_endio(async->bio); 78879787eaaSJeff Mahoney return; 78979787eaaSJeff Mahoney } 79079787eaaSJeff Mahoney 791ec39f769SChris Mason /* 792ec39f769SChris Mason * All of the bios that pass through here are from async helpers. 793ec39f769SChris Mason * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context. 794ec39f769SChris Mason * This changes nothing when cgroups aren't in use. 795ec39f769SChris Mason */ 796ec39f769SChris Mason async->bio->bi_opf |= REQ_CGROUP_PUNT; 79708635baeSChris Mason ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num); 79806ea01b1SDavid Sterba if (ret) { 79906ea01b1SDavid Sterba async->bio->bi_status = ret; 80006ea01b1SDavid Sterba bio_endio(async->bio); 80106ea01b1SDavid Sterba } 8024a69a410SChris Mason } 8034a69a410SChris Mason 8044a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 8054a69a410SChris Mason { 8064a69a410SChris Mason struct async_submit_bio *async; 8074a69a410SChris Mason 8084a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 8098b712842SChris Mason kfree(async); 8108b712842SChris Mason } 8118b712842SChris Mason 8128896a08dSQu Wenruo blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio, 813c6100a4bSJosef Bacik int mirror_num, unsigned long bio_flags, 8141941b64bSQu Wenruo u64 dio_file_offset, 815e288c080SDavid Sterba extent_submit_bio_start_t *submit_bio_start) 81644b8bd7eSChris Mason { 8178896a08dSQu Wenruo struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; 81844b8bd7eSChris Mason struct async_submit_bio *async; 81944b8bd7eSChris Mason 82044b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 82144b8bd7eSChris Mason if (!async) 8224e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 82344b8bd7eSChris Mason 8248896a08dSQu Wenruo async->inode = inode; 82544b8bd7eSChris Mason async->bio = bio; 82644b8bd7eSChris Mason async->mirror_num = mirror_num; 8274a69a410SChris Mason async->submit_bio_start = submit_bio_start; 8284a69a410SChris Mason 829a0cac0ecSOmar Sandoval btrfs_init_work(&async->work, run_one_async_start, run_one_async_done, 830a0cac0ecSOmar Sandoval run_one_async_free); 8314a69a410SChris Mason 8321941b64bSQu Wenruo async->dio_file_offset = dio_file_offset; 8338c8bee1dSChris Mason 8344e4cbee9SChristoph Hellwig async->status = 0; 83579787eaaSJeff Mahoney 83667f055c7SChristoph Hellwig if (op_is_sync(bio->bi_opf)) 8375cdc7ad3SQu Wenruo btrfs_set_work_high_priority(&async->work); 838d313d7a3SChris Mason 8395cdc7ad3SQu Wenruo btrfs_queue_work(fs_info->workers, &async->work); 84044b8bd7eSChris Mason return 0; 84144b8bd7eSChris Mason } 84244b8bd7eSChris Mason 8434e4cbee9SChristoph Hellwig static blk_status_t btree_csum_one_bio(struct bio *bio) 844ce3ed71aSChris Mason { 8452c30c71bSKent Overstreet struct bio_vec *bvec; 846ce3ed71aSChris Mason struct btrfs_root *root; 8472b070cfeSChristoph Hellwig int ret = 0; 8486dc4f100SMing Lei struct bvec_iter_all iter_all; 849ce3ed71aSChris Mason 850c09abff8SDavid Sterba ASSERT(!bio_flagged(bio, BIO_CLONED)); 8512b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 852ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 853ac303b69SQu Wenruo ret = csum_dirty_buffer(root->fs_info, bvec); 85479787eaaSJeff Mahoney if (ret) 85579787eaaSJeff Mahoney break; 856ce3ed71aSChris Mason } 8572c30c71bSKent Overstreet 8584e4cbee9SChristoph Hellwig return errno_to_blk_status(ret); 859ce3ed71aSChris Mason } 860ce3ed71aSChris Mason 8618896a08dSQu Wenruo static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio, 8621941b64bSQu Wenruo u64 dio_file_offset) 86322c59948SChris Mason { 8648b712842SChris Mason /* 8658b712842SChris Mason * when we're called for a write, we're already in the async 8665443be45SChris Mason * submission context. Just jump into btrfs_map_bio 8678b712842SChris Mason */ 86879787eaaSJeff Mahoney return btree_csum_one_bio(bio); 86922c59948SChris Mason } 87022c59948SChris Mason 8719b4e675aSDavid Sterba static int check_async_write(struct btrfs_fs_info *fs_info, 8729b4e675aSDavid Sterba struct btrfs_inode *bi) 873de0022b9SJosef Bacik { 8744eef29efSNaohiro Aota if (btrfs_is_zoned(fs_info)) 8754eef29efSNaohiro Aota return 0; 8766300463bSLiu Bo if (atomic_read(&bi->sync_writers)) 8776300463bSLiu Bo return 0; 8789b4e675aSDavid Sterba if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags)) 879de0022b9SJosef Bacik return 0; 880de0022b9SJosef Bacik return 1; 881de0022b9SJosef Bacik } 882de0022b9SJosef Bacik 8831b36294aSNikolay Borisov blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio, 8841b36294aSNikolay Borisov int mirror_num, unsigned long bio_flags) 88544b8bd7eSChris Mason { 8860b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 8879b4e675aSDavid Sterba int async = check_async_write(fs_info, BTRFS_I(inode)); 8884e4cbee9SChristoph Hellwig blk_status_t ret; 889cad321adSChris Mason 890cfe94440SNaohiro Aota if (btrfs_op(bio) != BTRFS_MAP_WRITE) { 891cad321adSChris Mason /* 892cad321adSChris Mason * called for a read, do the setup so that checksum validation 893cad321adSChris Mason * can happen in the async kernel threads 894cad321adSChris Mason */ 8950b246afaSJeff Mahoney ret = btrfs_bio_wq_end_io(fs_info, bio, 8960b246afaSJeff Mahoney BTRFS_WQ_ENDIO_METADATA); 8971d4284bdSChris Mason if (ret) 89861891923SStefan Behrens goto out_w_error; 89908635baeSChris Mason ret = btrfs_map_bio(fs_info, bio, mirror_num); 900de0022b9SJosef Bacik } else if (!async) { 901de0022b9SJosef Bacik ret = btree_csum_one_bio(bio); 902de0022b9SJosef Bacik if (ret) 90361891923SStefan Behrens goto out_w_error; 90408635baeSChris Mason ret = btrfs_map_bio(fs_info, bio, mirror_num); 90561891923SStefan Behrens } else { 906cad321adSChris Mason /* 90761891923SStefan Behrens * kthread helpers are used to submit writes so that 90861891923SStefan Behrens * checksumming can happen in parallel across all CPUs 909cad321adSChris Mason */ 9108896a08dSQu Wenruo ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0, 9118896a08dSQu Wenruo 0, btree_submit_bio_start); 91244b8bd7eSChris Mason } 91344b8bd7eSChris Mason 9144246a0b6SChristoph Hellwig if (ret) 9154246a0b6SChristoph Hellwig goto out_w_error; 9164246a0b6SChristoph Hellwig return 0; 9174246a0b6SChristoph Hellwig 91861891923SStefan Behrens out_w_error: 9194e4cbee9SChristoph Hellwig bio->bi_status = ret; 9204246a0b6SChristoph Hellwig bio_endio(bio); 92161891923SStefan Behrens return ret; 92261891923SStefan Behrens } 92361891923SStefan Behrens 9243dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 925784b4e29SChris Mason static int btree_migratepage(struct address_space *mapping, 926a6bc32b8SMel Gorman struct page *newpage, struct page *page, 927a6bc32b8SMel Gorman enum migrate_mode mode) 928784b4e29SChris Mason { 929784b4e29SChris Mason /* 930784b4e29SChris Mason * we can't safely write a btree page from here, 931784b4e29SChris Mason * we haven't done the locking hook 932784b4e29SChris Mason */ 933784b4e29SChris Mason if (PageDirty(page)) 934784b4e29SChris Mason return -EAGAIN; 935784b4e29SChris Mason /* 936784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 937784b4e29SChris Mason * We must have no buffers or drop them. 938784b4e29SChris Mason */ 939784b4e29SChris Mason if (page_has_private(page) && 940784b4e29SChris Mason !try_to_release_page(page, GFP_KERNEL)) 941784b4e29SChris Mason return -EAGAIN; 942a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 943784b4e29SChris Mason } 9443dd1462eSJan Beulich #endif 945784b4e29SChris Mason 9460da5468fSChris Mason 9470da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 9480da5468fSChris Mason struct writeback_control *wbc) 9490da5468fSChris Mason { 950e2d84521SMiao Xie struct btrfs_fs_info *fs_info; 951e2d84521SMiao Xie int ret; 952e2d84521SMiao Xie 953d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 954448d640bSChris Mason 955448d640bSChris Mason if (wbc->for_kupdate) 956448d640bSChris Mason return 0; 957448d640bSChris Mason 958e2d84521SMiao Xie fs_info = BTRFS_I(mapping->host)->root->fs_info; 959b9473439SChris Mason /* this is a bit racy, but that's ok */ 960d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 961d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 962d814a491SEthan Lien fs_info->dirty_metadata_batch); 963e2d84521SMiao Xie if (ret < 0) 964793955bcSChris Mason return 0; 965793955bcSChris Mason } 9660b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 9670da5468fSChris Mason } 9680da5468fSChris Mason 96970dec807SChris Mason static int btree_releasepage(struct page *page, gfp_t gfp_flags) 9705f39d397SChris Mason { 97198509cfcSChris Mason if (PageWriteback(page) || PageDirty(page)) 97298509cfcSChris Mason return 0; 9730c4e538bSDavid Sterba 974f7a52a40SDavid Sterba return try_release_extent_buffer(page); 975d98237b3SChris Mason } 976d98237b3SChris Mason 977d47992f8SLukas Czerner static void btree_invalidatepage(struct page *page, unsigned int offset, 978d47992f8SLukas Czerner unsigned int length) 979d98237b3SChris Mason { 980d1310b2eSChris Mason struct extent_io_tree *tree; 981d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9825f39d397SChris Mason extent_invalidatepage(tree, page, offset); 9835f39d397SChris Mason btree_releasepage(page, GFP_NOFS); 9849ad6b7bcSChris Mason if (PagePrivate(page)) { 985efe120a0SFrank Holton btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info, 986efe120a0SFrank Holton "page private not zero on page %llu", 987efe120a0SFrank Holton (unsigned long long)page_offset(page)); 988d1b89bc0SGuoqing Jiang detach_page_private(page); 9899ad6b7bcSChris Mason } 990d98237b3SChris Mason } 991d98237b3SChris Mason 9920b32f4bbSJosef Bacik static int btree_set_page_dirty(struct page *page) 9930b32f4bbSJosef Bacik { 994bb146eb2SJosef Bacik #ifdef DEBUG 9950b32f4bbSJosef Bacik struct extent_buffer *eb; 9960b32f4bbSJosef Bacik 9970b32f4bbSJosef Bacik BUG_ON(!PagePrivate(page)); 9980b32f4bbSJosef Bacik eb = (struct extent_buffer *)page->private; 9990b32f4bbSJosef Bacik BUG_ON(!eb); 10000b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 10010b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 10020b32f4bbSJosef Bacik btrfs_assert_tree_locked(eb); 1003bb146eb2SJosef Bacik #endif 10040b32f4bbSJosef Bacik return __set_page_dirty_nobuffers(page); 10050b32f4bbSJosef Bacik } 10060b32f4bbSJosef Bacik 10077f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 10080da5468fSChris Mason .writepages = btree_writepages, 10095f39d397SChris Mason .releasepage = btree_releasepage, 10105f39d397SChris Mason .invalidatepage = btree_invalidatepage, 10115a92bc88SChris Mason #ifdef CONFIG_MIGRATION 1012784b4e29SChris Mason .migratepage = btree_migratepage, 10135a92bc88SChris Mason #endif 10140b32f4bbSJosef Bacik .set_page_dirty = btree_set_page_dirty, 1015d98237b3SChris Mason }; 1016123abc88SChris Mason 10172ff7e61eSJeff Mahoney struct extent_buffer *btrfs_find_create_tree_block( 10182ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info, 10193fbaf258SJosef Bacik u64 bytenr, u64 owner_root, 10203fbaf258SJosef Bacik int level) 10210999df54SChris Mason { 10220b246afaSJeff Mahoney if (btrfs_is_testing(fs_info)) 10230b246afaSJeff Mahoney return alloc_test_extent_buffer(fs_info, bytenr); 10243fbaf258SJosef Bacik return alloc_extent_buffer(fs_info, bytenr, owner_root, level); 10250999df54SChris Mason } 10260999df54SChris Mason 1027581c1760SQu Wenruo /* 1028581c1760SQu Wenruo * Read tree block at logical address @bytenr and do variant basic but critical 1029581c1760SQu Wenruo * verification. 1030581c1760SQu Wenruo * 10311b7ec85eSJosef Bacik * @owner_root: the objectid of the root owner for this block. 1032581c1760SQu Wenruo * @parent_transid: expected transid of this tree block, skip check if 0 1033581c1760SQu Wenruo * @level: expected level, mandatory check 1034581c1760SQu Wenruo * @first_key: expected key in slot 0, skip check if NULL 1035581c1760SQu Wenruo */ 10362ff7e61eSJeff Mahoney struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, 10371b7ec85eSJosef Bacik u64 owner_root, u64 parent_transid, 10381b7ec85eSJosef Bacik int level, struct btrfs_key *first_key) 1039e20d96d6SChris Mason { 10405f39d397SChris Mason struct extent_buffer *buf = NULL; 104119c00ddcSChris Mason int ret; 104219c00ddcSChris Mason 10433fbaf258SJosef Bacik buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level); 1044c871b0f2SLiu Bo if (IS_ERR(buf)) 1045c871b0f2SLiu Bo return buf; 1046e4204dedSChris Mason 10475ab12d1fSDavid Sterba ret = btree_read_extent_buffer_pages(buf, parent_transid, 1048581c1760SQu Wenruo level, first_key); 10490f0fe8f7SFilipe David Borba Manana if (ret) { 1050537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 105164c043deSLiu Bo return ERR_PTR(ret); 10520f0fe8f7SFilipe David Borba Manana } 10535f39d397SChris Mason return buf; 1054ce9adaa5SChris Mason 1055eb60ceacSChris Mason } 1056eb60ceacSChris Mason 10576a884d7dSDavid Sterba void btrfs_clean_tree_block(struct extent_buffer *buf) 1058ed2ff2cbSChris Mason { 10596a884d7dSDavid Sterba struct btrfs_fs_info *fs_info = buf->fs_info; 106055c69072SChris Mason if (btrfs_header_generation(buf) == 1061e2d84521SMiao Xie fs_info->running_transaction->transid) { 1062b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1063b4ce94deSChris Mason 1064b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1065104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 1066e2d84521SMiao Xie -buf->len, 1067e2d84521SMiao Xie fs_info->dirty_metadata_batch); 10680b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 1069925baeddSChris Mason } 10705f39d397SChris Mason } 1071ed7b63ebSJosef Bacik } 10725f39d397SChris Mason 1073da17066cSJeff Mahoney static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, 1074e20d96d6SChris Mason u64 objectid) 1075d97e63b6SChris Mason { 10767c0260eeSJeff Mahoney bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); 107796dfcb46SJosef Bacik root->fs_info = fs_info; 1078cfaa7295SChris Mason root->node = NULL; 1079a28ec197SChris Mason root->commit_root = NULL; 108027cdeb70SMiao Xie root->state = 0; 1081d68fc57bSYan, Zheng root->orphan_cleanup_state = 0; 10820b86a832SChris Mason 10830f7d52f4SChris Mason root->last_trans = 0; 10846b8fad57SNikolay Borisov root->free_objectid = 0; 1085eb73c1b7SMiao Xie root->nr_delalloc_inodes = 0; 1086199c2a9cSMiao Xie root->nr_ordered_extents = 0; 10876bef4d31SEric Paris root->inode_tree = RB_ROOT; 108816cdcec7SMiao Xie INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1089f0486c68SYan, Zheng root->block_rsv = NULL; 10900b86a832SChris Mason 10910b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 10925d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 1093eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_inodes); 1094eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_root); 1095199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_extents); 1096199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_root); 1097d2311e69SQu Wenruo INIT_LIST_HEAD(&root->reloc_dirty_list); 10982ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 10992ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 11005d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1101eb73c1b7SMiao Xie spin_lock_init(&root->delalloc_lock); 1102199c2a9cSMiao Xie spin_lock_init(&root->ordered_extent_lock); 1103f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 11042ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 11052ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 11068287475aSQu Wenruo spin_lock_init(&root->qgroup_meta_rsv_lock); 1107a2135011SChris Mason mutex_init(&root->objectid_mutex); 1108e02119d5SChris Mason mutex_init(&root->log_mutex); 110931f3d255SMiao Xie mutex_init(&root->ordered_extent_mutex); 1110573bfb72SMiao Xie mutex_init(&root->delalloc_mutex); 1111c53e9653SQu Wenruo init_waitqueue_head(&root->qgroup_flush_wait); 11127237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 11137237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 11147237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 11158b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[0]); 11168b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[1]); 11177237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 11187237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 11197237f183SYan Zheng atomic_set(&root->log_writers, 0); 11202ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 11210700cea7SElena Reshetova refcount_set(&root->refs, 1); 11228ecebf4dSRobbie Ko atomic_set(&root->snapshot_force_cow, 0); 1123eede2bf3SOmar Sandoval atomic_set(&root->nr_swapfiles, 0); 11247237f183SYan Zheng root->log_transid = 0; 1125d1433debSMiao Xie root->log_transid_committed = -1; 1126257c62e1SChris Mason root->last_log_commit = 0; 1127e289f03eSFilipe Manana if (!dummy) { 112843eb5f29SQu Wenruo extent_io_tree_init(fs_info, &root->dirty_log_pages, 112943eb5f29SQu Wenruo IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL); 1130e289f03eSFilipe Manana extent_io_tree_init(fs_info, &root->log_csum_range, 1131e289f03eSFilipe Manana IO_TREE_LOG_CSUM_RANGE, NULL); 1132e289f03eSFilipe Manana } 1133017e5369SChris Mason 11343768f368SChris Mason memset(&root->root_key, 0, sizeof(root->root_key)); 11353768f368SChris Mason memset(&root->root_item, 0, sizeof(root->root_item)); 11366702ed49SChris Mason memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 11374d775673SChris Mason root->root_key.objectid = objectid; 11380ee5dc67SAl Viro root->anon_dev = 0; 11398ea05e3aSAlexander Block 11405f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 1141370a11b8SQu Wenruo btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks); 1142bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 1143bd647ce3SJosef Bacik INIT_LIST_HEAD(&root->leak_list); 1144bd647ce3SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 1145bd647ce3SJosef Bacik list_add_tail(&root->leak_list, &fs_info->allocated_roots); 1146bd647ce3SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 1147bd647ce3SJosef Bacik #endif 11483768f368SChris Mason } 11493768f368SChris Mason 115074e4d827SDavid Sterba static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, 115196dfcb46SJosef Bacik u64 objectid, gfp_t flags) 11526f07e42eSAl Viro { 115374e4d827SDavid Sterba struct btrfs_root *root = kzalloc(sizeof(*root), flags); 11546f07e42eSAl Viro if (root) 115596dfcb46SJosef Bacik __setup_root(root, fs_info, objectid); 11566f07e42eSAl Viro return root; 11576f07e42eSAl Viro } 11586f07e42eSAl Viro 115906ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 116006ea65a3SJosef Bacik /* Should only be used by the testing infrastructure */ 1161da17066cSJeff Mahoney struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) 116206ea65a3SJosef Bacik { 116306ea65a3SJosef Bacik struct btrfs_root *root; 116406ea65a3SJosef Bacik 11657c0260eeSJeff Mahoney if (!fs_info) 11667c0260eeSJeff Mahoney return ERR_PTR(-EINVAL); 11677c0260eeSJeff Mahoney 116896dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL); 116906ea65a3SJosef Bacik if (!root) 117006ea65a3SJosef Bacik return ERR_PTR(-ENOMEM); 1171da17066cSJeff Mahoney 1172b9ef22deSFeifei Xu /* We don't use the stripesize in selftest, set it as sectorsize */ 1173faa2dbf0SJosef Bacik root->alloc_bytenr = 0; 117406ea65a3SJosef Bacik 117506ea65a3SJosef Bacik return root; 117606ea65a3SJosef Bacik } 117706ea65a3SJosef Bacik #endif 117806ea65a3SJosef Bacik 117920897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 118020897f5cSArne Jansen u64 objectid) 118120897f5cSArne Jansen { 11829b7a2440SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 118320897f5cSArne Jansen struct extent_buffer *leaf; 118420897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 118520897f5cSArne Jansen struct btrfs_root *root; 118620897f5cSArne Jansen struct btrfs_key key; 1187b89f6d1fSFilipe Manana unsigned int nofs_flag; 118820897f5cSArne Jansen int ret = 0; 118920897f5cSArne Jansen 1190b89f6d1fSFilipe Manana /* 1191b89f6d1fSFilipe Manana * We're holding a transaction handle, so use a NOFS memory allocation 1192b89f6d1fSFilipe Manana * context to avoid deadlock if reclaim happens. 1193b89f6d1fSFilipe Manana */ 1194b89f6d1fSFilipe Manana nofs_flag = memalloc_nofs_save(); 119596dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL); 1196b89f6d1fSFilipe Manana memalloc_nofs_restore(nofs_flag); 119720897f5cSArne Jansen if (!root) 119820897f5cSArne Jansen return ERR_PTR(-ENOMEM); 119920897f5cSArne Jansen 120020897f5cSArne Jansen root->root_key.objectid = objectid; 120120897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 120220897f5cSArne Jansen root->root_key.offset = 0; 120320897f5cSArne Jansen 12049631e4ccSJosef Bacik leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0, 12059631e4ccSJosef Bacik BTRFS_NESTING_NORMAL); 120620897f5cSArne Jansen if (IS_ERR(leaf)) { 120720897f5cSArne Jansen ret = PTR_ERR(leaf); 12081dd05682STsutomu Itoh leaf = NULL; 12098a6a87cdSBoris Burkov goto fail_unlock; 121020897f5cSArne Jansen } 121120897f5cSArne Jansen 121220897f5cSArne Jansen root->node = leaf; 121320897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 121420897f5cSArne Jansen 121520897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 121627cdeb70SMiao Xie set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 121720897f5cSArne Jansen 1218f944d2cbSDavid Sterba btrfs_set_root_flags(&root->root_item, 0); 1219f944d2cbSDavid Sterba btrfs_set_root_limit(&root->root_item, 0); 122020897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 122120897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 122220897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 122320897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 122420897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 122520897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 122620897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 122733d85fdaSQu Wenruo if (is_fstree(objectid)) 1228807fc790SAndy Shevchenko generate_random_guid(root->root_item.uuid); 1229807fc790SAndy Shevchenko else 1230807fc790SAndy Shevchenko export_guid(root->root_item.uuid, &guid_null); 1231c8422684SDavid Sterba btrfs_set_root_drop_level(&root->root_item, 0); 123220897f5cSArne Jansen 12338a6a87cdSBoris Burkov btrfs_tree_unlock(leaf); 12348a6a87cdSBoris Burkov 123520897f5cSArne Jansen key.objectid = objectid; 123620897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 123720897f5cSArne Jansen key.offset = 0; 123820897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 123920897f5cSArne Jansen if (ret) 124020897f5cSArne Jansen goto fail; 124120897f5cSArne Jansen 124220897f5cSArne Jansen return root; 12431dd05682STsutomu Itoh 12448a6a87cdSBoris Burkov fail_unlock: 12458c38938cSJosef Bacik if (leaf) 12461dd05682STsutomu Itoh btrfs_tree_unlock(leaf); 12478a6a87cdSBoris Burkov fail: 124800246528SJosef Bacik btrfs_put_root(root); 12491dd05682STsutomu Itoh 12501dd05682STsutomu Itoh return ERR_PTR(ret); 125120897f5cSArne Jansen } 125220897f5cSArne Jansen 12537237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1254e02119d5SChris Mason struct btrfs_fs_info *fs_info) 12550f7d52f4SChris Mason { 12560f7d52f4SChris Mason struct btrfs_root *root; 1257e02119d5SChris Mason 125896dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS); 1259e02119d5SChris Mason if (!root) 12607237f183SYan Zheng return ERR_PTR(-ENOMEM); 1261e02119d5SChris Mason 1262e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1263e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1264e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 126527cdeb70SMiao Xie 12666ab6ebb7SNaohiro Aota return root; 12676ab6ebb7SNaohiro Aota } 12686ab6ebb7SNaohiro Aota 12696ab6ebb7SNaohiro Aota int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans, 12706ab6ebb7SNaohiro Aota struct btrfs_root *root) 12716ab6ebb7SNaohiro Aota { 12726ab6ebb7SNaohiro Aota struct extent_buffer *leaf; 12736ab6ebb7SNaohiro Aota 12747237f183SYan Zheng /* 127592a7cc42SQu Wenruo * DON'T set SHAREABLE bit for log trees. 127627cdeb70SMiao Xie * 127792a7cc42SQu Wenruo * Log trees are not exposed to user space thus can't be snapshotted, 127892a7cc42SQu Wenruo * and they go away before a real commit is actually done. 127992a7cc42SQu Wenruo * 128092a7cc42SQu Wenruo * They do store pointers to file data extents, and those reference 128192a7cc42SQu Wenruo * counts still get updated (along with back refs to the log tree). 12827237f183SYan Zheng */ 1283e02119d5SChris Mason 12844d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 12859631e4ccSJosef Bacik NULL, 0, 0, 0, BTRFS_NESTING_NORMAL); 12866ab6ebb7SNaohiro Aota if (IS_ERR(leaf)) 12876ab6ebb7SNaohiro Aota return PTR_ERR(leaf); 1288e02119d5SChris Mason 12897237f183SYan Zheng root->node = leaf; 1290e02119d5SChris Mason 1291e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1292e02119d5SChris Mason btrfs_tree_unlock(root->node); 12936ab6ebb7SNaohiro Aota 12946ab6ebb7SNaohiro Aota return 0; 12957237f183SYan Zheng } 12967237f183SYan Zheng 12977237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 12987237f183SYan Zheng struct btrfs_fs_info *fs_info) 12997237f183SYan Zheng { 13007237f183SYan Zheng struct btrfs_root *log_root; 13017237f183SYan Zheng 13027237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 13037237f183SYan Zheng if (IS_ERR(log_root)) 13047237f183SYan Zheng return PTR_ERR(log_root); 13056ab6ebb7SNaohiro Aota 13063ddebf27SNaohiro Aota if (!btrfs_is_zoned(fs_info)) { 13073ddebf27SNaohiro Aota int ret = btrfs_alloc_log_tree_node(trans, log_root); 13083ddebf27SNaohiro Aota 13096ab6ebb7SNaohiro Aota if (ret) { 13106ab6ebb7SNaohiro Aota btrfs_put_root(log_root); 13116ab6ebb7SNaohiro Aota return ret; 13126ab6ebb7SNaohiro Aota } 13133ddebf27SNaohiro Aota } 13146ab6ebb7SNaohiro Aota 13157237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 13167237f183SYan Zheng fs_info->log_root_tree = log_root; 13177237f183SYan Zheng return 0; 13187237f183SYan Zheng } 13197237f183SYan Zheng 13207237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 13217237f183SYan Zheng struct btrfs_root *root) 13227237f183SYan Zheng { 13230b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13247237f183SYan Zheng struct btrfs_root *log_root; 13257237f183SYan Zheng struct btrfs_inode_item *inode_item; 13266ab6ebb7SNaohiro Aota int ret; 13277237f183SYan Zheng 13280b246afaSJeff Mahoney log_root = alloc_log_tree(trans, fs_info); 13297237f183SYan Zheng if (IS_ERR(log_root)) 13307237f183SYan Zheng return PTR_ERR(log_root); 13317237f183SYan Zheng 13326ab6ebb7SNaohiro Aota ret = btrfs_alloc_log_tree_node(trans, log_root); 13336ab6ebb7SNaohiro Aota if (ret) { 13346ab6ebb7SNaohiro Aota btrfs_put_root(log_root); 13356ab6ebb7SNaohiro Aota return ret; 13366ab6ebb7SNaohiro Aota } 13376ab6ebb7SNaohiro Aota 13387237f183SYan Zheng log_root->last_trans = trans->transid; 13397237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 13407237f183SYan Zheng 13417237f183SYan Zheng inode_item = &log_root->root_item.inode; 13423cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 13433cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 13443cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 1345da17066cSJeff Mahoney btrfs_set_stack_inode_nbytes(inode_item, 13460b246afaSJeff Mahoney fs_info->nodesize); 13473cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 13487237f183SYan Zheng 13495d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 13507237f183SYan Zheng 13517237f183SYan Zheng WARN_ON(root->log_root); 13527237f183SYan Zheng root->log_root = log_root; 13537237f183SYan Zheng root->log_transid = 0; 1354d1433debSMiao Xie root->log_transid_committed = -1; 1355257c62e1SChris Mason root->last_log_commit = 0; 1356e02119d5SChris Mason return 0; 1357e02119d5SChris Mason } 1358e02119d5SChris Mason 135949d11beaSJosef Bacik static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root, 136049d11beaSJosef Bacik struct btrfs_path *path, 1361cb517eabSMiao Xie struct btrfs_key *key) 1362e02119d5SChris Mason { 1363e02119d5SChris Mason struct btrfs_root *root; 1364e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 136584234f3aSYan Zheng u64 generation; 1366cb517eabSMiao Xie int ret; 1367581c1760SQu Wenruo int level; 1368cb517eabSMiao Xie 136996dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS); 137049d11beaSJosef Bacik if (!root) 137149d11beaSJosef Bacik return ERR_PTR(-ENOMEM); 13720f7d52f4SChris Mason 1373cb517eabSMiao Xie ret = btrfs_find_root(tree_root, key, path, 1374cb517eabSMiao Xie &root->root_item, &root->root_key); 13750f7d52f4SChris Mason if (ret) { 137613a8a7c8SYan, Zheng if (ret > 0) 137713a8a7c8SYan, Zheng ret = -ENOENT; 137849d11beaSJosef Bacik goto fail; 13790f7d52f4SChris Mason } 138013a8a7c8SYan, Zheng 138184234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1382581c1760SQu Wenruo level = btrfs_root_level(&root->root_item); 13832ff7e61eSJeff Mahoney root->node = read_tree_block(fs_info, 13842ff7e61eSJeff Mahoney btrfs_root_bytenr(&root->root_item), 13851b7ec85eSJosef Bacik key->objectid, generation, level, NULL); 138664c043deSLiu Bo if (IS_ERR(root->node)) { 138764c043deSLiu Bo ret = PTR_ERR(root->node); 13888c38938cSJosef Bacik root->node = NULL; 138949d11beaSJosef Bacik goto fail; 1390cb517eabSMiao Xie } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1391cb517eabSMiao Xie ret = -EIO; 139249d11beaSJosef Bacik goto fail; 1393416bc658SJosef Bacik } 13945d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 1395cb517eabSMiao Xie return root; 139649d11beaSJosef Bacik fail: 139700246528SJosef Bacik btrfs_put_root(root); 139849d11beaSJosef Bacik return ERR_PTR(ret); 139949d11beaSJosef Bacik } 140049d11beaSJosef Bacik 140149d11beaSJosef Bacik struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 140249d11beaSJosef Bacik struct btrfs_key *key) 140349d11beaSJosef Bacik { 140449d11beaSJosef Bacik struct btrfs_root *root; 140549d11beaSJosef Bacik struct btrfs_path *path; 140649d11beaSJosef Bacik 140749d11beaSJosef Bacik path = btrfs_alloc_path(); 140849d11beaSJosef Bacik if (!path) 140949d11beaSJosef Bacik return ERR_PTR(-ENOMEM); 141049d11beaSJosef Bacik root = read_tree_root_path(tree_root, path, key); 141149d11beaSJosef Bacik btrfs_free_path(path); 141249d11beaSJosef Bacik 141349d11beaSJosef Bacik return root; 1414cb517eabSMiao Xie } 1415cb517eabSMiao Xie 14162dfb1e43SQu Wenruo /* 14172dfb1e43SQu Wenruo * Initialize subvolume root in-memory structure 14182dfb1e43SQu Wenruo * 14192dfb1e43SQu Wenruo * @anon_dev: anonymous device to attach to the root, if zero, allocate new 14202dfb1e43SQu Wenruo */ 14212dfb1e43SQu Wenruo static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev) 1422cb517eabSMiao Xie { 1423cb517eabSMiao Xie int ret; 1424dcc3eb96SNikolay Borisov unsigned int nofs_flag; 1425cb517eabSMiao Xie 1426dcc3eb96SNikolay Borisov /* 1427dcc3eb96SNikolay Borisov * We might be called under a transaction (e.g. indirect backref 1428dcc3eb96SNikolay Borisov * resolution) which could deadlock if it triggers memory reclaim 1429dcc3eb96SNikolay Borisov */ 1430dcc3eb96SNikolay Borisov nofs_flag = memalloc_nofs_save(); 1431dcc3eb96SNikolay Borisov ret = btrfs_drew_lock_init(&root->snapshot_lock); 1432dcc3eb96SNikolay Borisov memalloc_nofs_restore(nofs_flag); 1433dcc3eb96SNikolay Borisov if (ret) 14348257b2dcSMiao Xie goto fail; 14358257b2dcSMiao Xie 1436aeb935a4SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID && 1437aeb935a4SQu Wenruo root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { 143892a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &root->state); 1439f39e4571SJosef Bacik btrfs_check_and_init_root_item(&root->root_item); 1440f39e4571SJosef Bacik } 1441f39e4571SJosef Bacik 1442851fd730SQu Wenruo /* 1443851fd730SQu Wenruo * Don't assign anonymous block device to roots that are not exposed to 1444851fd730SQu Wenruo * userspace, the id pool is limited to 1M 1445851fd730SQu Wenruo */ 1446851fd730SQu Wenruo if (is_fstree(root->root_key.objectid) && 1447851fd730SQu Wenruo btrfs_root_refs(&root->root_item) > 0) { 14482dfb1e43SQu Wenruo if (!anon_dev) { 1449cb517eabSMiao Xie ret = get_anon_bdev(&root->anon_dev); 1450cb517eabSMiao Xie if (ret) 1451876d2cf1SLiu Bo goto fail; 14522dfb1e43SQu Wenruo } else { 14532dfb1e43SQu Wenruo root->anon_dev = anon_dev; 14542dfb1e43SQu Wenruo } 1455851fd730SQu Wenruo } 1456f32e48e9SChandan Rajendra 1457f32e48e9SChandan Rajendra mutex_lock(&root->objectid_mutex); 1458453e4873SNikolay Borisov ret = btrfs_init_root_free_objectid(root); 1459f32e48e9SChandan Rajendra if (ret) { 1460f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1461876d2cf1SLiu Bo goto fail; 1462f32e48e9SChandan Rajendra } 1463f32e48e9SChandan Rajendra 14646b8fad57SNikolay Borisov ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID); 1465f32e48e9SChandan Rajendra 1466f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1467f32e48e9SChandan Rajendra 1468cb517eabSMiao Xie return 0; 1469cb517eabSMiao Xie fail: 147084db5ccfSDavid Sterba /* The caller is responsible to call btrfs_free_fs_root */ 1471cb517eabSMiao Xie return ret; 1472cb517eabSMiao Xie } 1473cb517eabSMiao Xie 1474a98db0f3SJosef Bacik static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1475cb517eabSMiao Xie u64 root_id) 1476cb517eabSMiao Xie { 1477cb517eabSMiao Xie struct btrfs_root *root; 1478cb517eabSMiao Xie 1479cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1480cb517eabSMiao Xie root = radix_tree_lookup(&fs_info->fs_roots_radix, 1481cb517eabSMiao Xie (unsigned long)root_id); 1482bc44d7c4SJosef Bacik if (root) 148300246528SJosef Bacik root = btrfs_grab_root(root); 1484cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1485cb517eabSMiao Xie return root; 1486cb517eabSMiao Xie } 1487cb517eabSMiao Xie 148849d11beaSJosef Bacik static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info, 148949d11beaSJosef Bacik u64 objectid) 149049d11beaSJosef Bacik { 149149d11beaSJosef Bacik if (objectid == BTRFS_ROOT_TREE_OBJECTID) 149249d11beaSJosef Bacik return btrfs_grab_root(fs_info->tree_root); 149349d11beaSJosef Bacik if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 149449d11beaSJosef Bacik return btrfs_grab_root(fs_info->extent_root); 149549d11beaSJosef Bacik if (objectid == BTRFS_CHUNK_TREE_OBJECTID) 149649d11beaSJosef Bacik return btrfs_grab_root(fs_info->chunk_root); 149749d11beaSJosef Bacik if (objectid == BTRFS_DEV_TREE_OBJECTID) 149849d11beaSJosef Bacik return btrfs_grab_root(fs_info->dev_root); 149949d11beaSJosef Bacik if (objectid == BTRFS_CSUM_TREE_OBJECTID) 150049d11beaSJosef Bacik return btrfs_grab_root(fs_info->csum_root); 150149d11beaSJosef Bacik if (objectid == BTRFS_QUOTA_TREE_OBJECTID) 150249d11beaSJosef Bacik return btrfs_grab_root(fs_info->quota_root) ? 150349d11beaSJosef Bacik fs_info->quota_root : ERR_PTR(-ENOENT); 150449d11beaSJosef Bacik if (objectid == BTRFS_UUID_TREE_OBJECTID) 150549d11beaSJosef Bacik return btrfs_grab_root(fs_info->uuid_root) ? 150649d11beaSJosef Bacik fs_info->uuid_root : ERR_PTR(-ENOENT); 150749d11beaSJosef Bacik if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) 150849d11beaSJosef Bacik return btrfs_grab_root(fs_info->free_space_root) ? 150949d11beaSJosef Bacik fs_info->free_space_root : ERR_PTR(-ENOENT); 151049d11beaSJosef Bacik return NULL; 151149d11beaSJosef Bacik } 151249d11beaSJosef Bacik 1513cb517eabSMiao Xie int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1514cb517eabSMiao Xie struct btrfs_root *root) 1515cb517eabSMiao Xie { 1516cb517eabSMiao Xie int ret; 1517cb517eabSMiao Xie 1518e1860a77SDavid Sterba ret = radix_tree_preload(GFP_NOFS); 1519cb517eabSMiao Xie if (ret) 1520cb517eabSMiao Xie return ret; 1521cb517eabSMiao Xie 1522cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1523cb517eabSMiao Xie ret = radix_tree_insert(&fs_info->fs_roots_radix, 1524cb517eabSMiao Xie (unsigned long)root->root_key.objectid, 1525cb517eabSMiao Xie root); 1526af01d2e5SJosef Bacik if (ret == 0) { 152700246528SJosef Bacik btrfs_grab_root(root); 152827cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1529af01d2e5SJosef Bacik } 1530cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1531cb517eabSMiao Xie radix_tree_preload_end(); 1532cb517eabSMiao Xie 1533cb517eabSMiao Xie return ret; 1534cb517eabSMiao Xie } 1535cb517eabSMiao Xie 1536bd647ce3SJosef Bacik void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info) 1537bd647ce3SJosef Bacik { 1538bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 1539bd647ce3SJosef Bacik struct btrfs_root *root; 1540bd647ce3SJosef Bacik 1541bd647ce3SJosef Bacik while (!list_empty(&fs_info->allocated_roots)) { 1542457f1864SJosef Bacik char buf[BTRFS_ROOT_NAME_BUF_LEN]; 1543457f1864SJosef Bacik 1544bd647ce3SJosef Bacik root = list_first_entry(&fs_info->allocated_roots, 1545bd647ce3SJosef Bacik struct btrfs_root, leak_list); 1546457f1864SJosef Bacik btrfs_err(fs_info, "leaked root %s refcount %d", 154771008734SJosef Bacik btrfs_root_name(&root->root_key, buf), 1548bd647ce3SJosef Bacik refcount_read(&root->refs)); 1549bd647ce3SJosef Bacik while (refcount_read(&root->refs) > 1) 155000246528SJosef Bacik btrfs_put_root(root); 155100246528SJosef Bacik btrfs_put_root(root); 1552bd647ce3SJosef Bacik } 1553bd647ce3SJosef Bacik #endif 1554bd647ce3SJosef Bacik } 1555bd647ce3SJosef Bacik 15560d4b0463SJosef Bacik void btrfs_free_fs_info(struct btrfs_fs_info *fs_info) 15570d4b0463SJosef Bacik { 1558141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 1559141386e1SJosef Bacik percpu_counter_destroy(&fs_info->delalloc_bytes); 15605deb17e1SJosef Bacik percpu_counter_destroy(&fs_info->ordered_bytes); 1561141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 1562141386e1SJosef Bacik btrfs_free_csum_hash(fs_info); 1563141386e1SJosef Bacik btrfs_free_stripe_hash_table(fs_info); 1564141386e1SJosef Bacik btrfs_free_ref_cache(fs_info); 15650d4b0463SJosef Bacik kfree(fs_info->balance_ctl); 15660d4b0463SJosef Bacik kfree(fs_info->delayed_root); 156700246528SJosef Bacik btrfs_put_root(fs_info->extent_root); 156800246528SJosef Bacik btrfs_put_root(fs_info->tree_root); 156900246528SJosef Bacik btrfs_put_root(fs_info->chunk_root); 157000246528SJosef Bacik btrfs_put_root(fs_info->dev_root); 157100246528SJosef Bacik btrfs_put_root(fs_info->csum_root); 157200246528SJosef Bacik btrfs_put_root(fs_info->quota_root); 157300246528SJosef Bacik btrfs_put_root(fs_info->uuid_root); 157400246528SJosef Bacik btrfs_put_root(fs_info->free_space_root); 157500246528SJosef Bacik btrfs_put_root(fs_info->fs_root); 1576aeb935a4SQu Wenruo btrfs_put_root(fs_info->data_reloc_root); 1577bd647ce3SJosef Bacik btrfs_check_leaked_roots(fs_info); 15783fd63727SJosef Bacik btrfs_extent_buffer_leak_debug_check(fs_info); 15790d4b0463SJosef Bacik kfree(fs_info->super_copy); 15800d4b0463SJosef Bacik kfree(fs_info->super_for_commit); 15810d4b0463SJosef Bacik kvfree(fs_info); 15820d4b0463SJosef Bacik } 15830d4b0463SJosef Bacik 15840d4b0463SJosef Bacik 15852dfb1e43SQu Wenruo /* 15862dfb1e43SQu Wenruo * Get an in-memory reference of a root structure. 15872dfb1e43SQu Wenruo * 15882dfb1e43SQu Wenruo * For essential trees like root/extent tree, we grab it from fs_info directly. 15892dfb1e43SQu Wenruo * For subvolume trees, we check the cached filesystem roots first. If not 15902dfb1e43SQu Wenruo * found, then read it from disk and add it to cached fs roots. 15912dfb1e43SQu Wenruo * 15922dfb1e43SQu Wenruo * Caller should release the root by calling btrfs_put_root() after the usage. 15932dfb1e43SQu Wenruo * 15942dfb1e43SQu Wenruo * NOTE: Reloc and log trees can't be read by this function as they share the 15952dfb1e43SQu Wenruo * same root objectid. 15962dfb1e43SQu Wenruo * 15972dfb1e43SQu Wenruo * @objectid: root id 15982dfb1e43SQu Wenruo * @anon_dev: preallocated anonymous block device number for new roots, 15992dfb1e43SQu Wenruo * pass 0 for new allocation. 16002dfb1e43SQu Wenruo * @check_ref: whether to check root item references, If true, return -ENOENT 16012dfb1e43SQu Wenruo * for orphan roots 16022dfb1e43SQu Wenruo */ 16032dfb1e43SQu Wenruo static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info, 16042dfb1e43SQu Wenruo u64 objectid, dev_t anon_dev, 16052dfb1e43SQu Wenruo bool check_ref) 16065eda7b5eSChris Mason { 16075eda7b5eSChris Mason struct btrfs_root *root; 1608381cf658SDavid Sterba struct btrfs_path *path; 16091d4c08e0SDavid Sterba struct btrfs_key key; 16105eda7b5eSChris Mason int ret; 16115eda7b5eSChris Mason 161249d11beaSJosef Bacik root = btrfs_get_global_root(fs_info, objectid); 161349d11beaSJosef Bacik if (root) 161449d11beaSJosef Bacik return root; 16154df27c4dSYan, Zheng again: 161656e9357aSDavid Sterba root = btrfs_lookup_fs_root(fs_info, objectid); 161748475471SStefan Behrens if (root) { 16182dfb1e43SQu Wenruo /* Shouldn't get preallocated anon_dev for cached roots */ 16192dfb1e43SQu Wenruo ASSERT(!anon_dev); 1620bc44d7c4SJosef Bacik if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 162100246528SJosef Bacik btrfs_put_root(root); 162248475471SStefan Behrens return ERR_PTR(-ENOENT); 1623bc44d7c4SJosef Bacik } 16245eda7b5eSChris Mason return root; 162548475471SStefan Behrens } 16265eda7b5eSChris Mason 162756e9357aSDavid Sterba key.objectid = objectid; 162856e9357aSDavid Sterba key.type = BTRFS_ROOT_ITEM_KEY; 162956e9357aSDavid Sterba key.offset = (u64)-1; 163056e9357aSDavid Sterba root = btrfs_read_tree_root(fs_info->tree_root, &key); 16315eda7b5eSChris Mason if (IS_ERR(root)) 16325eda7b5eSChris Mason return root; 16333394e160SChris Mason 1634c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1635d68fc57bSYan, Zheng ret = -ENOENT; 1636d68fc57bSYan, Zheng goto fail; 1637d68fc57bSYan, Zheng } 1638d68fc57bSYan, Zheng 16392dfb1e43SQu Wenruo ret = btrfs_init_fs_root(root, anon_dev); 1640cb517eabSMiao Xie if (ret) 1641cb517eabSMiao Xie goto fail; 1642cb517eabSMiao Xie 1643381cf658SDavid Sterba path = btrfs_alloc_path(); 1644381cf658SDavid Sterba if (!path) { 1645381cf658SDavid Sterba ret = -ENOMEM; 1646381cf658SDavid Sterba goto fail; 1647381cf658SDavid Sterba } 16481d4c08e0SDavid Sterba key.objectid = BTRFS_ORPHAN_OBJECTID; 16491d4c08e0SDavid Sterba key.type = BTRFS_ORPHAN_ITEM_KEY; 165056e9357aSDavid Sterba key.offset = objectid; 16511d4c08e0SDavid Sterba 16521d4c08e0SDavid Sterba ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1653381cf658SDavid Sterba btrfs_free_path(path); 1654d68fc57bSYan, Zheng if (ret < 0) 1655d68fc57bSYan, Zheng goto fail; 1656d68fc57bSYan, Zheng if (ret == 0) 165727cdeb70SMiao Xie set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1658d68fc57bSYan, Zheng 1659cb517eabSMiao Xie ret = btrfs_insert_fs_root(fs_info, root); 16600f7d52f4SChris Mason if (ret) { 166100246528SJosef Bacik btrfs_put_root(root); 16624785e24fSJosef Bacik if (ret == -EEXIST) 16634df27c4dSYan, Zheng goto again; 16644df27c4dSYan, Zheng goto fail; 16654df27c4dSYan, Zheng } 1666edbd8d4eSChris Mason return root; 16674df27c4dSYan, Zheng fail: 16688c38938cSJosef Bacik btrfs_put_root(root); 16694df27c4dSYan, Zheng return ERR_PTR(ret); 1670edbd8d4eSChris Mason } 1671edbd8d4eSChris Mason 16722dfb1e43SQu Wenruo /* 16732dfb1e43SQu Wenruo * Get in-memory reference of a root structure 16742dfb1e43SQu Wenruo * 16752dfb1e43SQu Wenruo * @objectid: tree objectid 16762dfb1e43SQu Wenruo * @check_ref: if set, verify that the tree exists and the item has at least 16772dfb1e43SQu Wenruo * one reference 16782dfb1e43SQu Wenruo */ 16792dfb1e43SQu Wenruo struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 16802dfb1e43SQu Wenruo u64 objectid, bool check_ref) 16812dfb1e43SQu Wenruo { 16822dfb1e43SQu Wenruo return btrfs_get_root_ref(fs_info, objectid, 0, check_ref); 16832dfb1e43SQu Wenruo } 16842dfb1e43SQu Wenruo 16852dfb1e43SQu Wenruo /* 16862dfb1e43SQu Wenruo * Get in-memory reference of a root structure, created as new, optionally pass 16872dfb1e43SQu Wenruo * the anonymous block device id 16882dfb1e43SQu Wenruo * 16892dfb1e43SQu Wenruo * @objectid: tree objectid 16902dfb1e43SQu Wenruo * @anon_dev: if zero, allocate a new anonymous block device or use the 16912dfb1e43SQu Wenruo * parameter value 16922dfb1e43SQu Wenruo */ 16932dfb1e43SQu Wenruo struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info, 16942dfb1e43SQu Wenruo u64 objectid, dev_t anon_dev) 16952dfb1e43SQu Wenruo { 16962dfb1e43SQu Wenruo return btrfs_get_root_ref(fs_info, objectid, anon_dev, true); 16972dfb1e43SQu Wenruo } 16982dfb1e43SQu Wenruo 16998b712842SChris Mason /* 170049d11beaSJosef Bacik * btrfs_get_fs_root_commit_root - return a root for the given objectid 170149d11beaSJosef Bacik * @fs_info: the fs_info 170249d11beaSJosef Bacik * @objectid: the objectid we need to lookup 170349d11beaSJosef Bacik * 170449d11beaSJosef Bacik * This is exclusively used for backref walking, and exists specifically because 170549d11beaSJosef Bacik * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref 170649d11beaSJosef Bacik * creation time, which means we may have to read the tree_root in order to look 170749d11beaSJosef Bacik * up a fs root that is not in memory. If the root is not in memory we will 170849d11beaSJosef Bacik * read the tree root commit root and look up the fs root from there. This is a 170949d11beaSJosef Bacik * temporary root, it will not be inserted into the radix tree as it doesn't 171049d11beaSJosef Bacik * have the most uptodate information, it'll simply be discarded once the 171149d11beaSJosef Bacik * backref code is finished using the root. 171249d11beaSJosef Bacik */ 171349d11beaSJosef Bacik struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info, 171449d11beaSJosef Bacik struct btrfs_path *path, 171549d11beaSJosef Bacik u64 objectid) 171649d11beaSJosef Bacik { 171749d11beaSJosef Bacik struct btrfs_root *root; 171849d11beaSJosef Bacik struct btrfs_key key; 171949d11beaSJosef Bacik 172049d11beaSJosef Bacik ASSERT(path->search_commit_root && path->skip_locking); 172149d11beaSJosef Bacik 172249d11beaSJosef Bacik /* 172349d11beaSJosef Bacik * This can return -ENOENT if we ask for a root that doesn't exist, but 172449d11beaSJosef Bacik * since this is called via the backref walking code we won't be looking 172549d11beaSJosef Bacik * up a root that doesn't exist, unless there's corruption. So if root 172649d11beaSJosef Bacik * != NULL just return it. 172749d11beaSJosef Bacik */ 172849d11beaSJosef Bacik root = btrfs_get_global_root(fs_info, objectid); 172949d11beaSJosef Bacik if (root) 173049d11beaSJosef Bacik return root; 173149d11beaSJosef Bacik 173249d11beaSJosef Bacik root = btrfs_lookup_fs_root(fs_info, objectid); 173349d11beaSJosef Bacik if (root) 173449d11beaSJosef Bacik return root; 173549d11beaSJosef Bacik 173649d11beaSJosef Bacik key.objectid = objectid; 173749d11beaSJosef Bacik key.type = BTRFS_ROOT_ITEM_KEY; 173849d11beaSJosef Bacik key.offset = (u64)-1; 173949d11beaSJosef Bacik root = read_tree_root_path(fs_info->tree_root, path, &key); 174049d11beaSJosef Bacik btrfs_release_path(path); 174149d11beaSJosef Bacik 174249d11beaSJosef Bacik return root; 174349d11beaSJosef Bacik } 174449d11beaSJosef Bacik 174549d11beaSJosef Bacik /* 17468b712842SChris Mason * called by the kthread helper functions to finally call the bio end_io 17478b712842SChris Mason * functions. This is where read checksum verification actually happens 17488b712842SChris Mason */ 17498b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work) 1750ce9adaa5SChris Mason { 1751ce9adaa5SChris Mason struct bio *bio; 175297eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 1753ce9adaa5SChris Mason 175497eb6b69SDavid Sterba end_io_wq = container_of(work, struct btrfs_end_io_wq, work); 1755ce9adaa5SChris Mason bio = end_io_wq->bio; 17568b712842SChris Mason 17574e4cbee9SChristoph Hellwig bio->bi_status = end_io_wq->status; 1758ce9adaa5SChris Mason bio->bi_private = end_io_wq->private; 1759ce9adaa5SChris Mason bio->bi_end_io = end_io_wq->end_io; 17604246a0b6SChristoph Hellwig bio_endio(bio); 17619be490f1SOmar Sandoval kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq); 1762ce9adaa5SChris Mason } 176344b8bd7eSChris Mason 1764a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1765a74a4b97SChris Mason { 1766a74a4b97SChris Mason struct btrfs_root *root = arg; 17670b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1768d0278245SMiao Xie int again; 1769a74a4b97SChris Mason 1770d6fd0ae2SOmar Sandoval while (1) { 1771d0278245SMiao Xie again = 0; 17729d1a2a3aSDavid Sterba 1773fd340d0fSJosef Bacik set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1774fd340d0fSJosef Bacik 1775d0278245SMiao Xie /* Make the cleaner go to sleep early. */ 17762ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) 1777d0278245SMiao Xie goto sleep; 1778d0278245SMiao Xie 177990c711abSZygo Blaxell /* 178090c711abSZygo Blaxell * Do not do anything if we might cause open_ctree() to block 178190c711abSZygo Blaxell * before we have finished mounting the filesystem. 178290c711abSZygo Blaxell */ 17830b246afaSJeff Mahoney if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 178490c711abSZygo Blaxell goto sleep; 178590c711abSZygo Blaxell 17860b246afaSJeff Mahoney if (!mutex_trylock(&fs_info->cleaner_mutex)) 1787d0278245SMiao Xie goto sleep; 1788d0278245SMiao Xie 1789dc7f370cSMiao Xie /* 1790dc7f370cSMiao Xie * Avoid the problem that we change the status of the fs 1791dc7f370cSMiao Xie * during the above check and trylock. 1792dc7f370cSMiao Xie */ 17932ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) { 17940b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1795dc7f370cSMiao Xie goto sleep; 1796dc7f370cSMiao Xie } 1797dc7f370cSMiao Xie 17982ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 1799c2d6cb16SFilipe Manana 18009d1a2a3aSDavid Sterba again = btrfs_clean_one_deleted_snapshot(root); 18010b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1802a74a4b97SChris Mason 1803d0278245SMiao Xie /* 180405323cd1SMiao Xie * The defragger has dealt with the R/O remount and umount, 180505323cd1SMiao Xie * needn't do anything special here. 1806d0278245SMiao Xie */ 18070b246afaSJeff Mahoney btrfs_run_defrag_inodes(fs_info); 180867c5e7d4SFilipe Manana 180967c5e7d4SFilipe Manana /* 181067c5e7d4SFilipe Manana * Acquires fs_info->delete_unused_bgs_mutex to avoid racing 181167c5e7d4SFilipe Manana * with relocation (btrfs_relocate_chunk) and relocation 181267c5e7d4SFilipe Manana * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 181367c5e7d4SFilipe Manana * after acquiring fs_info->delete_unused_bgs_mutex. So we 181467c5e7d4SFilipe Manana * can't hold, nor need to, fs_info->cleaner_mutex when deleting 181567c5e7d4SFilipe Manana * unused block groups. 181667c5e7d4SFilipe Manana */ 18170b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 1818d0278245SMiao Xie sleep: 1819a0a1db70SFilipe Manana clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1820d6fd0ae2SOmar Sandoval if (kthread_should_park()) 1821d6fd0ae2SOmar Sandoval kthread_parkme(); 1822d6fd0ae2SOmar Sandoval if (kthread_should_stop()) 1823d6fd0ae2SOmar Sandoval return 0; 1824838fe188SJiri Kosina if (!again) { 1825a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 1826a74a4b97SChris Mason schedule(); 1827a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1828a74a4b97SChris Mason } 1829da288d28SFilipe Manana } 1830a74a4b97SChris Mason } 1831a74a4b97SChris Mason 1832a74a4b97SChris Mason static int transaction_kthread(void *arg) 1833a74a4b97SChris Mason { 1834a74a4b97SChris Mason struct btrfs_root *root = arg; 18350b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1836a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1837a74a4b97SChris Mason struct btrfs_transaction *cur; 18388929ecfaSYan, Zheng u64 transid; 1839643900beSNikolay Borisov time64_t delta; 1840a74a4b97SChris Mason unsigned long delay; 1841914b2007SJan Kara bool cannot_commit; 1842a74a4b97SChris Mason 1843a74a4b97SChris Mason do { 1844914b2007SJan Kara cannot_commit = false; 1845ba1bc00fSNikolay Borisov delay = msecs_to_jiffies(fs_info->commit_interval * 1000); 18460b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 1847a74a4b97SChris Mason 18480b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 18490b246afaSJeff Mahoney cur = fs_info->running_transaction; 1850a74a4b97SChris Mason if (!cur) { 18510b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1852a74a4b97SChris Mason goto sleep; 1853a74a4b97SChris Mason } 185431153d81SYan Zheng 1855643900beSNikolay Borisov delta = ktime_get_seconds() - cur->start_time; 18563296bf56SQu Wenruo if (cur->state < TRANS_STATE_COMMIT_START && 1857643900beSNikolay Borisov delta < fs_info->commit_interval) { 18580b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1859fb8a7e94SNikolay Borisov delay -= msecs_to_jiffies((delta - 1) * 1000); 1860fb8a7e94SNikolay Borisov delay = min(delay, 1861fb8a7e94SNikolay Borisov msecs_to_jiffies(fs_info->commit_interval * 1000)); 1862a74a4b97SChris Mason goto sleep; 1863a74a4b97SChris Mason } 18648929ecfaSYan, Zheng transid = cur->transid; 18650b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 186656bec294SChris Mason 186779787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 1868354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 1869914b2007SJan Kara if (IS_ERR(trans)) { 1870354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 1871914b2007SJan Kara cannot_commit = true; 187279787eaaSJeff Mahoney goto sleep; 1873914b2007SJan Kara } 18748929ecfaSYan, Zheng if (transid == trans->transid) { 18753a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 18768929ecfaSYan, Zheng } else { 18773a45bb20SJeff Mahoney btrfs_end_transaction(trans); 18788929ecfaSYan, Zheng } 1879a74a4b97SChris Mason sleep: 18800b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 18810b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 1882a74a4b97SChris Mason 18834e121c06SJosef Bacik if (unlikely(test_bit(BTRFS_FS_STATE_ERROR, 18840b246afaSJeff Mahoney &fs_info->fs_state))) 18852ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 18868929ecfaSYan, Zheng if (!kthread_should_stop() && 18870b246afaSJeff Mahoney (!btrfs_transaction_blocked(fs_info) || 1888914b2007SJan Kara cannot_commit)) 1889bc5511d0SNikolay Borisov schedule_timeout_interruptible(delay); 1890a74a4b97SChris Mason } while (!kthread_should_stop()); 1891a74a4b97SChris Mason return 0; 1892a74a4b97SChris Mason } 1893a74a4b97SChris Mason 1894af31f5e5SChris Mason /* 189501f0f9daSNikolay Borisov * This will find the highest generation in the array of root backups. The 189601f0f9daSNikolay Borisov * index of the highest array is returned, or -EINVAL if we can't find 189701f0f9daSNikolay Borisov * anything. 1898af31f5e5SChris Mason * 1899af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1900af31f5e5SChris Mason * generation of the latest root in the array with the generation 1901af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1902af31f5e5SChris Mason */ 190301f0f9daSNikolay Borisov static int find_newest_super_backup(struct btrfs_fs_info *info) 1904af31f5e5SChris Mason { 190501f0f9daSNikolay Borisov const u64 newest_gen = btrfs_super_generation(info->super_copy); 1906af31f5e5SChris Mason u64 cur; 1907af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1908af31f5e5SChris Mason int i; 1909af31f5e5SChris Mason 1910af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1911af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1912af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1913af31f5e5SChris Mason if (cur == newest_gen) 191401f0f9daSNikolay Borisov return i; 1915af31f5e5SChris Mason } 1916af31f5e5SChris Mason 191701f0f9daSNikolay Borisov return -EINVAL; 1918af31f5e5SChris Mason } 1919af31f5e5SChris Mason 1920af31f5e5SChris Mason /* 1921af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1922af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1923af31f5e5SChris Mason * done 1924af31f5e5SChris Mason */ 1925af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1926af31f5e5SChris Mason { 19276ef108ddSNikolay Borisov const int next_backup = info->backup_root_index; 1928af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1929af31f5e5SChris Mason 1930af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1931af31f5e5SChris Mason 1932af31f5e5SChris Mason /* 1933af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1934af31f5e5SChris Mason * regardless of which ones we use today 1935af31f5e5SChris Mason */ 1936af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1937af31f5e5SChris Mason 1938af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1939af31f5e5SChris Mason 1940af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1941af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1942af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1943af31f5e5SChris Mason 1944af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1945af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1946af31f5e5SChris Mason 1947af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1948af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1949af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 1950af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 1951af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 1952af31f5e5SChris Mason 1953af31f5e5SChris Mason btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); 1954af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 1955af31f5e5SChris Mason btrfs_header_generation(info->extent_root->node)); 1956af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 1957af31f5e5SChris Mason btrfs_header_level(info->extent_root->node)); 1958af31f5e5SChris Mason 19597c7e82a7SChris Mason /* 19607c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 19617c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 19627c7e82a7SChris Mason */ 19637c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 19647c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 19657c7e82a7SChris Mason info->fs_root->node->start); 1966af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 1967af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 1968af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 1969af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 19707c7e82a7SChris Mason } 1971af31f5e5SChris Mason 1972af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 1973af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 1974af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 1975af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 1976af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 1977af31f5e5SChris Mason 1978af31f5e5SChris Mason btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); 1979af31f5e5SChris Mason btrfs_set_backup_csum_root_gen(root_backup, 1980af31f5e5SChris Mason btrfs_header_generation(info->csum_root->node)); 1981af31f5e5SChris Mason btrfs_set_backup_csum_root_level(root_backup, 1982af31f5e5SChris Mason btrfs_header_level(info->csum_root->node)); 1983af31f5e5SChris Mason 1984af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 1985af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 1986af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 1987af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 1988af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 1989af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 1990af31f5e5SChris Mason 1991af31f5e5SChris Mason /* 1992af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 1993af31f5e5SChris Mason * for the next commit 1994af31f5e5SChris Mason */ 1995af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 1996af31f5e5SChris Mason &info->super_for_commit->super_roots, 1997af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 1998af31f5e5SChris Mason } 1999af31f5e5SChris Mason 2000af31f5e5SChris Mason /* 2001bd2336b2SNikolay Borisov * read_backup_root - Reads a backup root based on the passed priority. Prio 0 2002bd2336b2SNikolay Borisov * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots 2003bd2336b2SNikolay Borisov * 2004bd2336b2SNikolay Borisov * fs_info - filesystem whose backup roots need to be read 2005bd2336b2SNikolay Borisov * priority - priority of backup root required 2006bd2336b2SNikolay Borisov * 2007bd2336b2SNikolay Borisov * Returns backup root index on success and -EINVAL otherwise. 2008bd2336b2SNikolay Borisov */ 2009bd2336b2SNikolay Borisov static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority) 2010bd2336b2SNikolay Borisov { 2011bd2336b2SNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 2012bd2336b2SNikolay Borisov struct btrfs_super_block *super = fs_info->super_copy; 2013bd2336b2SNikolay Borisov struct btrfs_root_backup *root_backup; 2014bd2336b2SNikolay Borisov 2015bd2336b2SNikolay Borisov if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) { 2016bd2336b2SNikolay Borisov if (priority == 0) 2017bd2336b2SNikolay Borisov return backup_index; 2018bd2336b2SNikolay Borisov 2019bd2336b2SNikolay Borisov backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority; 2020bd2336b2SNikolay Borisov backup_index %= BTRFS_NUM_BACKUP_ROOTS; 2021bd2336b2SNikolay Borisov } else { 2022bd2336b2SNikolay Borisov return -EINVAL; 2023bd2336b2SNikolay Borisov } 2024bd2336b2SNikolay Borisov 2025bd2336b2SNikolay Borisov root_backup = super->super_roots + backup_index; 2026bd2336b2SNikolay Borisov 2027bd2336b2SNikolay Borisov btrfs_set_super_generation(super, 2028bd2336b2SNikolay Borisov btrfs_backup_tree_root_gen(root_backup)); 2029bd2336b2SNikolay Borisov btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 2030bd2336b2SNikolay Borisov btrfs_set_super_root_level(super, 2031bd2336b2SNikolay Borisov btrfs_backup_tree_root_level(root_backup)); 2032bd2336b2SNikolay Borisov btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 2033bd2336b2SNikolay Borisov 2034bd2336b2SNikolay Borisov /* 2035bd2336b2SNikolay Borisov * Fixme: the total bytes and num_devices need to match or we should 2036bd2336b2SNikolay Borisov * need a fsck 2037bd2336b2SNikolay Borisov */ 2038bd2336b2SNikolay Borisov btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 2039bd2336b2SNikolay Borisov btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 2040bd2336b2SNikolay Borisov 2041bd2336b2SNikolay Borisov return backup_index; 2042bd2336b2SNikolay Borisov } 2043bd2336b2SNikolay Borisov 20447abadb64SLiu Bo /* helper to cleanup workers */ 20457abadb64SLiu Bo static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 20467abadb64SLiu Bo { 2047dc6e3209SQu Wenruo btrfs_destroy_workqueue(fs_info->fixup_workers); 2048afe3d242SQu Wenruo btrfs_destroy_workqueue(fs_info->delalloc_workers); 20495cdc7ad3SQu Wenruo btrfs_destroy_workqueue(fs_info->workers); 2050fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_workers); 2051fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_raid56_workers); 2052d05a33acSQu Wenruo btrfs_destroy_workqueue(fs_info->rmw_workers); 2053fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_write_workers); 2054fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 20555b3bc44eSQu Wenruo btrfs_destroy_workqueue(fs_info->delayed_workers); 2056e66f0bb1SQu Wenruo btrfs_destroy_workqueue(fs_info->caching_workers); 2057736cfa15SQu Wenruo btrfs_destroy_workqueue(fs_info->readahead_workers); 2058a44903abSQu Wenruo btrfs_destroy_workqueue(fs_info->flush_workers); 2059fc97fab0SQu Wenruo btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2060b0643e59SDennis Zhou if (fs_info->discard_ctl.discard_workers) 2061b0643e59SDennis Zhou destroy_workqueue(fs_info->discard_ctl.discard_workers); 2062a9b9477dSFilipe Manana /* 2063a9b9477dSFilipe Manana * Now that all other work queues are destroyed, we can safely destroy 2064a9b9477dSFilipe Manana * the queues used for metadata I/O, since tasks from those other work 2065a9b9477dSFilipe Manana * queues can do metadata I/O operations. 2066a9b9477dSFilipe Manana */ 2067a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_workers); 2068a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_write_workers); 20697abadb64SLiu Bo } 20707abadb64SLiu Bo 20712e9f5954SRashika static void free_root_extent_buffers(struct btrfs_root *root) 20722e9f5954SRashika { 20732e9f5954SRashika if (root) { 20742e9f5954SRashika free_extent_buffer(root->node); 20752e9f5954SRashika free_extent_buffer(root->commit_root); 20762e9f5954SRashika root->node = NULL; 20772e9f5954SRashika root->commit_root = NULL; 20782e9f5954SRashika } 20792e9f5954SRashika } 20802e9f5954SRashika 2081af31f5e5SChris Mason /* helper to cleanup tree roots */ 20824273eaffSAnand Jain static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root) 2083af31f5e5SChris Mason { 20842e9f5954SRashika free_root_extent_buffers(info->tree_root); 2085655b09feSJosef Bacik 20862e9f5954SRashika free_root_extent_buffers(info->dev_root); 20872e9f5954SRashika free_root_extent_buffers(info->extent_root); 20882e9f5954SRashika free_root_extent_buffers(info->csum_root); 20892e9f5954SRashika free_root_extent_buffers(info->quota_root); 20902e9f5954SRashika free_root_extent_buffers(info->uuid_root); 20918c38938cSJosef Bacik free_root_extent_buffers(info->fs_root); 2092aeb935a4SQu Wenruo free_root_extent_buffers(info->data_reloc_root); 20934273eaffSAnand Jain if (free_chunk_root) 20942e9f5954SRashika free_root_extent_buffers(info->chunk_root); 209570f6d82eSOmar Sandoval free_root_extent_buffers(info->free_space_root); 2096af31f5e5SChris Mason } 2097af31f5e5SChris Mason 20988c38938cSJosef Bacik void btrfs_put_root(struct btrfs_root *root) 20998c38938cSJosef Bacik { 21008c38938cSJosef Bacik if (!root) 21018c38938cSJosef Bacik return; 21028c38938cSJosef Bacik 21038c38938cSJosef Bacik if (refcount_dec_and_test(&root->refs)) { 21048c38938cSJosef Bacik WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 21051dae7e0eSQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)); 21068c38938cSJosef Bacik if (root->anon_dev) 21078c38938cSJosef Bacik free_anon_bdev(root->anon_dev); 21088c38938cSJosef Bacik btrfs_drew_lock_destroy(&root->snapshot_lock); 2109923eb523SJohannes Thumshirn free_root_extent_buffers(root); 21108c38938cSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 21118c38938cSJosef Bacik spin_lock(&root->fs_info->fs_roots_radix_lock); 21128c38938cSJosef Bacik list_del_init(&root->leak_list); 21138c38938cSJosef Bacik spin_unlock(&root->fs_info->fs_roots_radix_lock); 21148c38938cSJosef Bacik #endif 21158c38938cSJosef Bacik kfree(root); 21168c38938cSJosef Bacik } 21178c38938cSJosef Bacik } 21188c38938cSJosef Bacik 2119faa2dbf0SJosef Bacik void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2120171f6537SJosef Bacik { 2121171f6537SJosef Bacik int ret; 2122171f6537SJosef Bacik struct btrfs_root *gang[8]; 2123171f6537SJosef Bacik int i; 2124171f6537SJosef Bacik 2125171f6537SJosef Bacik while (!list_empty(&fs_info->dead_roots)) { 2126171f6537SJosef Bacik gang[0] = list_entry(fs_info->dead_roots.next, 2127171f6537SJosef Bacik struct btrfs_root, root_list); 2128171f6537SJosef Bacik list_del(&gang[0]->root_list); 2129171f6537SJosef Bacik 21308c38938cSJosef Bacik if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) 2131cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[0]); 213200246528SJosef Bacik btrfs_put_root(gang[0]); 2133171f6537SJosef Bacik } 2134171f6537SJosef Bacik 2135171f6537SJosef Bacik while (1) { 2136171f6537SJosef Bacik ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2137171f6537SJosef Bacik (void **)gang, 0, 2138171f6537SJosef Bacik ARRAY_SIZE(gang)); 2139171f6537SJosef Bacik if (!ret) 2140171f6537SJosef Bacik break; 2141171f6537SJosef Bacik for (i = 0; i < ret; i++) 2142cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2143171f6537SJosef Bacik } 2144171f6537SJosef Bacik } 2145af31f5e5SChris Mason 2146638aa7edSEric Sandeen static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2147638aa7edSEric Sandeen { 2148638aa7edSEric Sandeen mutex_init(&fs_info->scrub_lock); 2149638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_running, 0); 2150638aa7edSEric Sandeen atomic_set(&fs_info->scrub_pause_req, 0); 2151638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_paused, 0); 2152638aa7edSEric Sandeen atomic_set(&fs_info->scrub_cancel_req, 0); 2153638aa7edSEric Sandeen init_waitqueue_head(&fs_info->scrub_pause_wait); 2154ff09c4caSAnand Jain refcount_set(&fs_info->scrub_workers_refcnt, 0); 2155638aa7edSEric Sandeen } 2156638aa7edSEric Sandeen 2157779a65a4SEric Sandeen static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2158779a65a4SEric Sandeen { 2159779a65a4SEric Sandeen spin_lock_init(&fs_info->balance_lock); 2160779a65a4SEric Sandeen mutex_init(&fs_info->balance_mutex); 2161779a65a4SEric Sandeen atomic_set(&fs_info->balance_pause_req, 0); 2162779a65a4SEric Sandeen atomic_set(&fs_info->balance_cancel_req, 0); 2163779a65a4SEric Sandeen fs_info->balance_ctl = NULL; 2164779a65a4SEric Sandeen init_waitqueue_head(&fs_info->balance_wait_q); 2165779a65a4SEric Sandeen } 2166779a65a4SEric Sandeen 21676bccf3abSJeff Mahoney static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info) 2168f37938e0SEric Sandeen { 21692ff7e61eSJeff Mahoney struct inode *inode = fs_info->btree_inode; 21702ff7e61eSJeff Mahoney 21712ff7e61eSJeff Mahoney inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 21722ff7e61eSJeff Mahoney set_nlink(inode, 1); 2173f37938e0SEric Sandeen /* 2174f37938e0SEric Sandeen * we set the i_size on the btree inode to the max possible int. 2175f37938e0SEric Sandeen * the real end of the address space is determined by all of 2176f37938e0SEric Sandeen * the devices in the system 2177f37938e0SEric Sandeen */ 21782ff7e61eSJeff Mahoney inode->i_size = OFFSET_MAX; 21792ff7e61eSJeff Mahoney inode->i_mapping->a_ops = &btree_aops; 2180f37938e0SEric Sandeen 21812ff7e61eSJeff Mahoney RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 218243eb5f29SQu Wenruo extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree, 21832c53a14dSQu Wenruo IO_TREE_BTREE_INODE_IO, inode); 21847b439738SDavid Sterba BTRFS_I(inode)->io_tree.track_uptodate = false; 21852ff7e61eSJeff Mahoney extent_map_tree_init(&BTRFS_I(inode)->extent_tree); 2186f37938e0SEric Sandeen 21875c8fd99fSJosef Bacik BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root); 21882ff7e61eSJeff Mahoney memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key)); 21892ff7e61eSJeff Mahoney set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); 21902ff7e61eSJeff Mahoney btrfs_insert_inode_hash(inode); 2191f37938e0SEric Sandeen } 2192f37938e0SEric Sandeen 2193ad618368SEric Sandeen static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2194ad618368SEric Sandeen { 2195ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2196129827e3SDavid Sterba init_rwsem(&fs_info->dev_replace.rwsem); 21977f8d236aSDavid Sterba init_waitqueue_head(&fs_info->dev_replace.replace_wait); 2198ad618368SEric Sandeen } 2199ad618368SEric Sandeen 2200f9e92e40SEric Sandeen static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2201f9e92e40SEric Sandeen { 2202f9e92e40SEric Sandeen spin_lock_init(&fs_info->qgroup_lock); 2203f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_ioctl_lock); 2204f9e92e40SEric Sandeen fs_info->qgroup_tree = RB_ROOT; 2205f9e92e40SEric Sandeen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2206f9e92e40SEric Sandeen fs_info->qgroup_seq = 1; 2207f9e92e40SEric Sandeen fs_info->qgroup_ulist = NULL; 2208d2c609b8SJeff Mahoney fs_info->qgroup_rescan_running = false; 2209f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_rescan_lock); 2210f9e92e40SEric Sandeen } 2211f9e92e40SEric Sandeen 22122a458198SEric Sandeen static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info, 22132a458198SEric Sandeen struct btrfs_fs_devices *fs_devices) 22142a458198SEric Sandeen { 2215f7b885beSAnand Jain u32 max_active = fs_info->thread_pool_size; 22166f011058SDavid Sterba unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 22172a458198SEric Sandeen 22182a458198SEric Sandeen fs_info->workers = 2219cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "worker", 2220cb001095SJeff Mahoney flags | WQ_HIGHPRI, max_active, 16); 22212a458198SEric Sandeen 22222a458198SEric Sandeen fs_info->delalloc_workers = 2223cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delalloc", 2224cb001095SJeff Mahoney flags, max_active, 2); 22252a458198SEric Sandeen 22262a458198SEric Sandeen fs_info->flush_workers = 2227cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "flush_delalloc", 2228cb001095SJeff Mahoney flags, max_active, 0); 22292a458198SEric Sandeen 22302a458198SEric Sandeen fs_info->caching_workers = 2231cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); 22322a458198SEric Sandeen 22332a458198SEric Sandeen fs_info->fixup_workers = 2234cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); 22352a458198SEric Sandeen 22362a458198SEric Sandeen /* 22372a458198SEric Sandeen * endios are largely parallel and should have a very 22382a458198SEric Sandeen * low idle thresh 22392a458198SEric Sandeen */ 22402a458198SEric Sandeen fs_info->endio_workers = 2241cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4); 22422a458198SEric Sandeen fs_info->endio_meta_workers = 2243cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta", flags, 2244cb001095SJeff Mahoney max_active, 4); 22452a458198SEric Sandeen fs_info->endio_meta_write_workers = 2246cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags, 2247cb001095SJeff Mahoney max_active, 2); 22482a458198SEric Sandeen fs_info->endio_raid56_workers = 2249cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-raid56", flags, 2250cb001095SJeff Mahoney max_active, 4); 22512a458198SEric Sandeen fs_info->rmw_workers = 2252cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2); 22532a458198SEric Sandeen fs_info->endio_write_workers = 2254cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-write", flags, 2255cb001095SJeff Mahoney max_active, 2); 22562a458198SEric Sandeen fs_info->endio_freespace_worker = 2257cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "freespace-write", flags, 2258cb001095SJeff Mahoney max_active, 0); 22592a458198SEric Sandeen fs_info->delayed_workers = 2260cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, 2261cb001095SJeff Mahoney max_active, 0); 22622a458198SEric Sandeen fs_info->readahead_workers = 2263cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "readahead", flags, 2264cb001095SJeff Mahoney max_active, 2); 22652a458198SEric Sandeen fs_info->qgroup_rescan_workers = 2266cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); 2267b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers = 2268b0643e59SDennis Zhou alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1); 22692a458198SEric Sandeen 22702a458198SEric Sandeen if (!(fs_info->workers && fs_info->delalloc_workers && 2271ba8a9d07SChris Mason fs_info->flush_workers && 22722a458198SEric Sandeen fs_info->endio_workers && fs_info->endio_meta_workers && 22732a458198SEric Sandeen fs_info->endio_meta_write_workers && 22742a458198SEric Sandeen fs_info->endio_write_workers && fs_info->endio_raid56_workers && 22752a458198SEric Sandeen fs_info->endio_freespace_worker && fs_info->rmw_workers && 22762a458198SEric Sandeen fs_info->caching_workers && fs_info->readahead_workers && 22772a458198SEric Sandeen fs_info->fixup_workers && fs_info->delayed_workers && 2278b0643e59SDennis Zhou fs_info->qgroup_rescan_workers && 2279b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers)) { 22802a458198SEric Sandeen return -ENOMEM; 22812a458198SEric Sandeen } 22822a458198SEric Sandeen 22832a458198SEric Sandeen return 0; 22842a458198SEric Sandeen } 22852a458198SEric Sandeen 22866d97c6e3SJohannes Thumshirn static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type) 22876d97c6e3SJohannes Thumshirn { 22886d97c6e3SJohannes Thumshirn struct crypto_shash *csum_shash; 2289b4e967beSDavid Sterba const char *csum_driver = btrfs_super_csum_driver(csum_type); 22906d97c6e3SJohannes Thumshirn 2291b4e967beSDavid Sterba csum_shash = crypto_alloc_shash(csum_driver, 0, 0); 22926d97c6e3SJohannes Thumshirn 22936d97c6e3SJohannes Thumshirn if (IS_ERR(csum_shash)) { 22946d97c6e3SJohannes Thumshirn btrfs_err(fs_info, "error allocating %s hash for checksum", 2295b4e967beSDavid Sterba csum_driver); 22966d97c6e3SJohannes Thumshirn return PTR_ERR(csum_shash); 22976d97c6e3SJohannes Thumshirn } 22986d97c6e3SJohannes Thumshirn 22996d97c6e3SJohannes Thumshirn fs_info->csum_shash = csum_shash; 23006d97c6e3SJohannes Thumshirn 23016d97c6e3SJohannes Thumshirn return 0; 23026d97c6e3SJohannes Thumshirn } 23036d97c6e3SJohannes Thumshirn 230463443bf5SEric Sandeen static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 230563443bf5SEric Sandeen struct btrfs_fs_devices *fs_devices) 230663443bf5SEric Sandeen { 230763443bf5SEric Sandeen int ret; 230863443bf5SEric Sandeen struct btrfs_root *log_tree_root; 230963443bf5SEric Sandeen struct btrfs_super_block *disk_super = fs_info->super_copy; 231063443bf5SEric Sandeen u64 bytenr = btrfs_super_log_root(disk_super); 2311581c1760SQu Wenruo int level = btrfs_super_log_root_level(disk_super); 231263443bf5SEric Sandeen 231363443bf5SEric Sandeen if (fs_devices->rw_devices == 0) { 2314f14d104dSDavid Sterba btrfs_warn(fs_info, "log replay required on RO media"); 231563443bf5SEric Sandeen return -EIO; 231663443bf5SEric Sandeen } 231763443bf5SEric Sandeen 231896dfcb46SJosef Bacik log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, 231996dfcb46SJosef Bacik GFP_KERNEL); 232063443bf5SEric Sandeen if (!log_tree_root) 232163443bf5SEric Sandeen return -ENOMEM; 232263443bf5SEric Sandeen 23232ff7e61eSJeff Mahoney log_tree_root->node = read_tree_block(fs_info, bytenr, 23241b7ec85eSJosef Bacik BTRFS_TREE_LOG_OBJECTID, 23251b7ec85eSJosef Bacik fs_info->generation + 1, level, 23261b7ec85eSJosef Bacik NULL); 232764c043deSLiu Bo if (IS_ERR(log_tree_root->node)) { 2328f14d104dSDavid Sterba btrfs_warn(fs_info, "failed to read log tree"); 23290eeff236SLiu Bo ret = PTR_ERR(log_tree_root->node); 23308c38938cSJosef Bacik log_tree_root->node = NULL; 233100246528SJosef Bacik btrfs_put_root(log_tree_root); 23320eeff236SLiu Bo return ret; 233364c043deSLiu Bo } else if (!extent_buffer_uptodate(log_tree_root->node)) { 2334f14d104dSDavid Sterba btrfs_err(fs_info, "failed to read log tree"); 233500246528SJosef Bacik btrfs_put_root(log_tree_root); 233663443bf5SEric Sandeen return -EIO; 233763443bf5SEric Sandeen } 233863443bf5SEric Sandeen /* returns with log_tree_root freed on success */ 233963443bf5SEric Sandeen ret = btrfs_recover_log_trees(log_tree_root); 234063443bf5SEric Sandeen if (ret) { 23410b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 234263443bf5SEric Sandeen "Failed to recover log tree"); 234300246528SJosef Bacik btrfs_put_root(log_tree_root); 234463443bf5SEric Sandeen return ret; 234563443bf5SEric Sandeen } 234663443bf5SEric Sandeen 2347bc98a42cSDavid Howells if (sb_rdonly(fs_info->sb)) { 23486bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 234963443bf5SEric Sandeen if (ret) 235063443bf5SEric Sandeen return ret; 235163443bf5SEric Sandeen } 235263443bf5SEric Sandeen 235363443bf5SEric Sandeen return 0; 235463443bf5SEric Sandeen } 235563443bf5SEric Sandeen 23566bccf3abSJeff Mahoney static int btrfs_read_roots(struct btrfs_fs_info *fs_info) 23574bbcaa64SEric Sandeen { 23586bccf3abSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 2359a4f3d2c4SDavid Sterba struct btrfs_root *root; 23604bbcaa64SEric Sandeen struct btrfs_key location; 23614bbcaa64SEric Sandeen int ret; 23624bbcaa64SEric Sandeen 23636bccf3abSJeff Mahoney BUG_ON(!fs_info->tree_root); 23646bccf3abSJeff Mahoney 23654bbcaa64SEric Sandeen location.objectid = BTRFS_EXTENT_TREE_OBJECTID; 23664bbcaa64SEric Sandeen location.type = BTRFS_ROOT_ITEM_KEY; 23674bbcaa64SEric Sandeen location.offset = 0; 23684bbcaa64SEric Sandeen 2369a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2370f50f4353SLiu Bo if (IS_ERR(root)) { 237142437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2372f50f4353SLiu Bo ret = PTR_ERR(root); 2373f50f4353SLiu Bo goto out; 2374f50f4353SLiu Bo } 237542437a63SJosef Bacik } else { 2376a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2377a4f3d2c4SDavid Sterba fs_info->extent_root = root; 237842437a63SJosef Bacik } 23794bbcaa64SEric Sandeen 23804bbcaa64SEric Sandeen location.objectid = BTRFS_DEV_TREE_OBJECTID; 2381a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2382f50f4353SLiu Bo if (IS_ERR(root)) { 238342437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2384f50f4353SLiu Bo ret = PTR_ERR(root); 2385f50f4353SLiu Bo goto out; 2386f50f4353SLiu Bo } 238742437a63SJosef Bacik } else { 2388a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2389a4f3d2c4SDavid Sterba fs_info->dev_root = root; 239042437a63SJosef Bacik } 2391*820a49daSJosef Bacik /* Initialize fs_info for all devices in any case */ 2392*820a49daSJosef Bacik btrfs_init_devices_late(fs_info); 23934bbcaa64SEric Sandeen 2394882dbe0cSJosef Bacik /* If IGNOREDATACSUMS is set don't bother reading the csum root. */ 2395882dbe0cSJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREDATACSUMS)) { 23964bbcaa64SEric Sandeen location.objectid = BTRFS_CSUM_TREE_OBJECTID; 2397a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2398f50f4353SLiu Bo if (IS_ERR(root)) { 239942437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2400f50f4353SLiu Bo ret = PTR_ERR(root); 2401f50f4353SLiu Bo goto out; 2402f50f4353SLiu Bo } 240342437a63SJosef Bacik } else { 2404a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2405a4f3d2c4SDavid Sterba fs_info->csum_root = root; 240642437a63SJosef Bacik } 2407882dbe0cSJosef Bacik } 24084bbcaa64SEric Sandeen 2409aeb935a4SQu Wenruo /* 2410aeb935a4SQu Wenruo * This tree can share blocks with some other fs tree during relocation 2411aeb935a4SQu Wenruo * and we need a proper setup by btrfs_get_fs_root 2412aeb935a4SQu Wenruo */ 241356e9357aSDavid Sterba root = btrfs_get_fs_root(tree_root->fs_info, 241456e9357aSDavid Sterba BTRFS_DATA_RELOC_TREE_OBJECTID, true); 2415aeb935a4SQu Wenruo if (IS_ERR(root)) { 241642437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2417aeb935a4SQu Wenruo ret = PTR_ERR(root); 2418aeb935a4SQu Wenruo goto out; 2419aeb935a4SQu Wenruo } 242042437a63SJosef Bacik } else { 2421aeb935a4SQu Wenruo set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2422aeb935a4SQu Wenruo fs_info->data_reloc_root = root; 242342437a63SJosef Bacik } 2424aeb935a4SQu Wenruo 24254bbcaa64SEric Sandeen location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2426a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2427a4f3d2c4SDavid Sterba if (!IS_ERR(root)) { 2428a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2429afcdd129SJosef Bacik set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); 2430a4f3d2c4SDavid Sterba fs_info->quota_root = root; 24314bbcaa64SEric Sandeen } 24324bbcaa64SEric Sandeen 24334bbcaa64SEric Sandeen location.objectid = BTRFS_UUID_TREE_OBJECTID; 2434a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2435a4f3d2c4SDavid Sterba if (IS_ERR(root)) { 243642437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2437a4f3d2c4SDavid Sterba ret = PTR_ERR(root); 24384bbcaa64SEric Sandeen if (ret != -ENOENT) 2439f50f4353SLiu Bo goto out; 244042437a63SJosef Bacik } 24414bbcaa64SEric Sandeen } else { 2442a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2443a4f3d2c4SDavid Sterba fs_info->uuid_root = root; 24444bbcaa64SEric Sandeen } 24454bbcaa64SEric Sandeen 244670f6d82eSOmar Sandoval if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 244770f6d82eSOmar Sandoval location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID; 244870f6d82eSOmar Sandoval root = btrfs_read_tree_root(tree_root, &location); 2449f50f4353SLiu Bo if (IS_ERR(root)) { 245042437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2451f50f4353SLiu Bo ret = PTR_ERR(root); 2452f50f4353SLiu Bo goto out; 2453f50f4353SLiu Bo } 245442437a63SJosef Bacik } else { 245570f6d82eSOmar Sandoval set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 245670f6d82eSOmar Sandoval fs_info->free_space_root = root; 245770f6d82eSOmar Sandoval } 245842437a63SJosef Bacik } 245970f6d82eSOmar Sandoval 24604bbcaa64SEric Sandeen return 0; 2461f50f4353SLiu Bo out: 2462f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", 2463f50f4353SLiu Bo location.objectid, ret); 2464f50f4353SLiu Bo return ret; 24654bbcaa64SEric Sandeen } 24664bbcaa64SEric Sandeen 2467069ec957SQu Wenruo /* 2468069ec957SQu Wenruo * Real super block validation 2469069ec957SQu Wenruo * NOTE: super csum type and incompat features will not be checked here. 2470069ec957SQu Wenruo * 2471069ec957SQu Wenruo * @sb: super block to check 2472069ec957SQu Wenruo * @mirror_num: the super block number to check its bytenr: 2473069ec957SQu Wenruo * 0 the primary (1st) sb 2474069ec957SQu Wenruo * 1, 2 2nd and 3rd backup copy 2475069ec957SQu Wenruo * -1 skip bytenr check 2476069ec957SQu Wenruo */ 2477069ec957SQu Wenruo static int validate_super(struct btrfs_fs_info *fs_info, 2478069ec957SQu Wenruo struct btrfs_super_block *sb, int mirror_num) 247921a852b0SQu Wenruo { 248021a852b0SQu Wenruo u64 nodesize = btrfs_super_nodesize(sb); 248121a852b0SQu Wenruo u64 sectorsize = btrfs_super_sectorsize(sb); 248221a852b0SQu Wenruo int ret = 0; 248321a852b0SQu Wenruo 248421a852b0SQu Wenruo if (btrfs_super_magic(sb) != BTRFS_MAGIC) { 248521a852b0SQu Wenruo btrfs_err(fs_info, "no valid FS found"); 248621a852b0SQu Wenruo ret = -EINVAL; 248721a852b0SQu Wenruo } 248821a852b0SQu Wenruo if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { 248921a852b0SQu Wenruo btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", 249021a852b0SQu Wenruo btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); 249121a852b0SQu Wenruo ret = -EINVAL; 249221a852b0SQu Wenruo } 249321a852b0SQu Wenruo if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 249421a852b0SQu Wenruo btrfs_err(fs_info, "tree_root level too big: %d >= %d", 249521a852b0SQu Wenruo btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 249621a852b0SQu Wenruo ret = -EINVAL; 249721a852b0SQu Wenruo } 249821a852b0SQu Wenruo if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 249921a852b0SQu Wenruo btrfs_err(fs_info, "chunk_root level too big: %d >= %d", 250021a852b0SQu Wenruo btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 250121a852b0SQu Wenruo ret = -EINVAL; 250221a852b0SQu Wenruo } 250321a852b0SQu Wenruo if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 250421a852b0SQu Wenruo btrfs_err(fs_info, "log_root level too big: %d >= %d", 250521a852b0SQu Wenruo btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 250621a852b0SQu Wenruo ret = -EINVAL; 250721a852b0SQu Wenruo } 250821a852b0SQu Wenruo 250921a852b0SQu Wenruo /* 251021a852b0SQu Wenruo * Check sectorsize and nodesize first, other check will need it. 251121a852b0SQu Wenruo * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. 251221a852b0SQu Wenruo */ 251321a852b0SQu Wenruo if (!is_power_of_2(sectorsize) || sectorsize < 4096 || 251421a852b0SQu Wenruo sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { 251521a852b0SQu Wenruo btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); 251621a852b0SQu Wenruo ret = -EINVAL; 251721a852b0SQu Wenruo } 25180bb3eb3eSQu Wenruo 25190bb3eb3eSQu Wenruo /* 25200bb3eb3eSQu Wenruo * For 4K page size, we only support 4K sector size. 25210bb3eb3eSQu Wenruo * For 64K page size, we support read-write for 64K sector size, and 25220bb3eb3eSQu Wenruo * read-only for 4K sector size. 25230bb3eb3eSQu Wenruo */ 25240bb3eb3eSQu Wenruo if ((PAGE_SIZE == SZ_4K && sectorsize != PAGE_SIZE) || 25250bb3eb3eSQu Wenruo (PAGE_SIZE == SZ_64K && (sectorsize != SZ_4K && 25260bb3eb3eSQu Wenruo sectorsize != SZ_64K))) { 252721a852b0SQu Wenruo btrfs_err(fs_info, 25280bb3eb3eSQu Wenruo "sectorsize %llu not yet supported for page size %lu", 252921a852b0SQu Wenruo sectorsize, PAGE_SIZE); 253021a852b0SQu Wenruo ret = -EINVAL; 253121a852b0SQu Wenruo } 25320bb3eb3eSQu Wenruo 253321a852b0SQu Wenruo if (!is_power_of_2(nodesize) || nodesize < sectorsize || 253421a852b0SQu Wenruo nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { 253521a852b0SQu Wenruo btrfs_err(fs_info, "invalid nodesize %llu", nodesize); 253621a852b0SQu Wenruo ret = -EINVAL; 253721a852b0SQu Wenruo } 253821a852b0SQu Wenruo if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { 253921a852b0SQu Wenruo btrfs_err(fs_info, "invalid leafsize %u, should be %llu", 254021a852b0SQu Wenruo le32_to_cpu(sb->__unused_leafsize), nodesize); 254121a852b0SQu Wenruo ret = -EINVAL; 254221a852b0SQu Wenruo } 254321a852b0SQu Wenruo 254421a852b0SQu Wenruo /* Root alignment check */ 254521a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { 254621a852b0SQu Wenruo btrfs_warn(fs_info, "tree_root block unaligned: %llu", 254721a852b0SQu Wenruo btrfs_super_root(sb)); 254821a852b0SQu Wenruo ret = -EINVAL; 254921a852b0SQu Wenruo } 255021a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { 255121a852b0SQu Wenruo btrfs_warn(fs_info, "chunk_root block unaligned: %llu", 255221a852b0SQu Wenruo btrfs_super_chunk_root(sb)); 255321a852b0SQu Wenruo ret = -EINVAL; 255421a852b0SQu Wenruo } 255521a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { 255621a852b0SQu Wenruo btrfs_warn(fs_info, "log_root block unaligned: %llu", 255721a852b0SQu Wenruo btrfs_super_log_root(sb)); 255821a852b0SQu Wenruo ret = -EINVAL; 255921a852b0SQu Wenruo } 256021a852b0SQu Wenruo 2561de37aa51SNikolay Borisov if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, 25627239ff4bSNikolay Borisov BTRFS_FSID_SIZE) != 0) { 256321a852b0SQu Wenruo btrfs_err(fs_info, 25647239ff4bSNikolay Borisov "dev_item UUID does not match metadata fsid: %pU != %pU", 2565de37aa51SNikolay Borisov fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); 256621a852b0SQu Wenruo ret = -EINVAL; 256721a852b0SQu Wenruo } 256821a852b0SQu Wenruo 256921a852b0SQu Wenruo /* 257021a852b0SQu Wenruo * Hint to catch really bogus numbers, bitflips or so, more exact checks are 257121a852b0SQu Wenruo * done later 257221a852b0SQu Wenruo */ 257321a852b0SQu Wenruo if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { 257421a852b0SQu Wenruo btrfs_err(fs_info, "bytes_used is too small %llu", 257521a852b0SQu Wenruo btrfs_super_bytes_used(sb)); 257621a852b0SQu Wenruo ret = -EINVAL; 257721a852b0SQu Wenruo } 257821a852b0SQu Wenruo if (!is_power_of_2(btrfs_super_stripesize(sb))) { 257921a852b0SQu Wenruo btrfs_err(fs_info, "invalid stripesize %u", 258021a852b0SQu Wenruo btrfs_super_stripesize(sb)); 258121a852b0SQu Wenruo ret = -EINVAL; 258221a852b0SQu Wenruo } 258321a852b0SQu Wenruo if (btrfs_super_num_devices(sb) > (1UL << 31)) 258421a852b0SQu Wenruo btrfs_warn(fs_info, "suspicious number of devices: %llu", 258521a852b0SQu Wenruo btrfs_super_num_devices(sb)); 258621a852b0SQu Wenruo if (btrfs_super_num_devices(sb) == 0) { 258721a852b0SQu Wenruo btrfs_err(fs_info, "number of devices is 0"); 258821a852b0SQu Wenruo ret = -EINVAL; 258921a852b0SQu Wenruo } 259021a852b0SQu Wenruo 2591069ec957SQu Wenruo if (mirror_num >= 0 && 2592069ec957SQu Wenruo btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { 259321a852b0SQu Wenruo btrfs_err(fs_info, "super offset mismatch %llu != %u", 259421a852b0SQu Wenruo btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 259521a852b0SQu Wenruo ret = -EINVAL; 259621a852b0SQu Wenruo } 259721a852b0SQu Wenruo 259821a852b0SQu Wenruo /* 259921a852b0SQu Wenruo * Obvious sys_chunk_array corruptions, it must hold at least one key 260021a852b0SQu Wenruo * and one chunk 260121a852b0SQu Wenruo */ 260221a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 260321a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too big %u > %u", 260421a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 260521a852b0SQu Wenruo BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 260621a852b0SQu Wenruo ret = -EINVAL; 260721a852b0SQu Wenruo } 260821a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 260921a852b0SQu Wenruo + sizeof(struct btrfs_chunk)) { 261021a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too small %u < %zu", 261121a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 261221a852b0SQu Wenruo sizeof(struct btrfs_disk_key) 261321a852b0SQu Wenruo + sizeof(struct btrfs_chunk)); 261421a852b0SQu Wenruo ret = -EINVAL; 261521a852b0SQu Wenruo } 261621a852b0SQu Wenruo 261721a852b0SQu Wenruo /* 261821a852b0SQu Wenruo * The generation is a global counter, we'll trust it more than the others 261921a852b0SQu Wenruo * but it's still possible that it's the one that's wrong. 262021a852b0SQu Wenruo */ 262121a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 262221a852b0SQu Wenruo btrfs_warn(fs_info, 262321a852b0SQu Wenruo "suspicious: generation < chunk_root_generation: %llu < %llu", 262421a852b0SQu Wenruo btrfs_super_generation(sb), 262521a852b0SQu Wenruo btrfs_super_chunk_root_generation(sb)); 262621a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 262721a852b0SQu Wenruo && btrfs_super_cache_generation(sb) != (u64)-1) 262821a852b0SQu Wenruo btrfs_warn(fs_info, 262921a852b0SQu Wenruo "suspicious: generation < cache_generation: %llu < %llu", 263021a852b0SQu Wenruo btrfs_super_generation(sb), 263121a852b0SQu Wenruo btrfs_super_cache_generation(sb)); 263221a852b0SQu Wenruo 263321a852b0SQu Wenruo return ret; 263421a852b0SQu Wenruo } 263521a852b0SQu Wenruo 2636069ec957SQu Wenruo /* 2637069ec957SQu Wenruo * Validation of super block at mount time. 2638069ec957SQu Wenruo * Some checks already done early at mount time, like csum type and incompat 2639069ec957SQu Wenruo * flags will be skipped. 2640069ec957SQu Wenruo */ 2641069ec957SQu Wenruo static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) 2642069ec957SQu Wenruo { 2643069ec957SQu Wenruo return validate_super(fs_info, fs_info->super_copy, 0); 2644069ec957SQu Wenruo } 2645069ec957SQu Wenruo 264675cb857dSQu Wenruo /* 264775cb857dSQu Wenruo * Validation of super block at write time. 264875cb857dSQu Wenruo * Some checks like bytenr check will be skipped as their values will be 264975cb857dSQu Wenruo * overwritten soon. 265075cb857dSQu Wenruo * Extra checks like csum type and incompat flags will be done here. 265175cb857dSQu Wenruo */ 265275cb857dSQu Wenruo static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, 265375cb857dSQu Wenruo struct btrfs_super_block *sb) 265475cb857dSQu Wenruo { 265575cb857dSQu Wenruo int ret; 265675cb857dSQu Wenruo 265775cb857dSQu Wenruo ret = validate_super(fs_info, sb, -1); 265875cb857dSQu Wenruo if (ret < 0) 265975cb857dSQu Wenruo goto out; 2660e7e16f48SJohannes Thumshirn if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) { 266175cb857dSQu Wenruo ret = -EUCLEAN; 266275cb857dSQu Wenruo btrfs_err(fs_info, "invalid csum type, has %u want %u", 266375cb857dSQu Wenruo btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); 266475cb857dSQu Wenruo goto out; 266575cb857dSQu Wenruo } 266675cb857dSQu Wenruo if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 266775cb857dSQu Wenruo ret = -EUCLEAN; 266875cb857dSQu Wenruo btrfs_err(fs_info, 266975cb857dSQu Wenruo "invalid incompat flags, has 0x%llx valid mask 0x%llx", 267075cb857dSQu Wenruo btrfs_super_incompat_flags(sb), 267175cb857dSQu Wenruo (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); 267275cb857dSQu Wenruo goto out; 267375cb857dSQu Wenruo } 267475cb857dSQu Wenruo out: 267575cb857dSQu Wenruo if (ret < 0) 267675cb857dSQu Wenruo btrfs_err(fs_info, 267775cb857dSQu Wenruo "super block corruption detected before writing it to disk"); 267875cb857dSQu Wenruo return ret; 267975cb857dSQu Wenruo } 268075cb857dSQu Wenruo 26816ef108ddSNikolay Borisov static int __cold init_tree_roots(struct btrfs_fs_info *fs_info) 2682b8522a1eSNikolay Borisov { 26836ef108ddSNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 2684b8522a1eSNikolay Borisov struct btrfs_super_block *sb = fs_info->super_copy; 2685b8522a1eSNikolay Borisov struct btrfs_root *tree_root = fs_info->tree_root; 2686b8522a1eSNikolay Borisov bool handle_error = false; 2687b8522a1eSNikolay Borisov int ret = 0; 2688b8522a1eSNikolay Borisov int i; 2689b8522a1eSNikolay Borisov 2690b8522a1eSNikolay Borisov for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 2691b8522a1eSNikolay Borisov u64 generation; 2692b8522a1eSNikolay Borisov int level; 2693b8522a1eSNikolay Borisov 2694b8522a1eSNikolay Borisov if (handle_error) { 2695b8522a1eSNikolay Borisov if (!IS_ERR(tree_root->node)) 2696b8522a1eSNikolay Borisov free_extent_buffer(tree_root->node); 2697b8522a1eSNikolay Borisov tree_root->node = NULL; 2698b8522a1eSNikolay Borisov 2699b8522a1eSNikolay Borisov if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) 2700b8522a1eSNikolay Borisov break; 2701b8522a1eSNikolay Borisov 2702b8522a1eSNikolay Borisov free_root_pointers(fs_info, 0); 2703b8522a1eSNikolay Borisov 2704b8522a1eSNikolay Borisov /* 2705b8522a1eSNikolay Borisov * Don't use the log in recovery mode, it won't be 2706b8522a1eSNikolay Borisov * valid 2707b8522a1eSNikolay Borisov */ 2708b8522a1eSNikolay Borisov btrfs_set_super_log_root(sb, 0); 2709b8522a1eSNikolay Borisov 2710b8522a1eSNikolay Borisov /* We can't trust the free space cache either */ 2711b8522a1eSNikolay Borisov btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 2712b8522a1eSNikolay Borisov 2713b8522a1eSNikolay Borisov ret = read_backup_root(fs_info, i); 27146ef108ddSNikolay Borisov backup_index = ret; 2715b8522a1eSNikolay Borisov if (ret < 0) 2716b8522a1eSNikolay Borisov return ret; 2717b8522a1eSNikolay Borisov } 2718b8522a1eSNikolay Borisov generation = btrfs_super_generation(sb); 2719b8522a1eSNikolay Borisov level = btrfs_super_root_level(sb); 2720b8522a1eSNikolay Borisov tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb), 27211b7ec85eSJosef Bacik BTRFS_ROOT_TREE_OBJECTID, 2722b8522a1eSNikolay Borisov generation, level, NULL); 27230465337cSJosef Bacik if (IS_ERR(tree_root->node)) { 2724217f5004SNikolay Borisov handle_error = true; 2725b8522a1eSNikolay Borisov ret = PTR_ERR(tree_root->node); 27260465337cSJosef Bacik tree_root->node = NULL; 2727217f5004SNikolay Borisov btrfs_warn(fs_info, "couldn't read tree root"); 2728217f5004SNikolay Borisov continue; 2729b8522a1eSNikolay Borisov 2730217f5004SNikolay Borisov } else if (!extent_buffer_uptodate(tree_root->node)) { 2731217f5004SNikolay Borisov handle_error = true; 2732217f5004SNikolay Borisov ret = -EIO; 2733217f5004SNikolay Borisov btrfs_warn(fs_info, "error while reading tree root"); 2734b8522a1eSNikolay Borisov continue; 2735b8522a1eSNikolay Borisov } 2736b8522a1eSNikolay Borisov 2737b8522a1eSNikolay Borisov btrfs_set_root_node(&tree_root->root_item, tree_root->node); 2738b8522a1eSNikolay Borisov tree_root->commit_root = btrfs_root_node(tree_root); 2739b8522a1eSNikolay Borisov btrfs_set_root_refs(&tree_root->root_item, 1); 2740b8522a1eSNikolay Borisov 2741336a0d8dSNikolay Borisov /* 2742336a0d8dSNikolay Borisov * No need to hold btrfs_root::objectid_mutex since the fs 2743336a0d8dSNikolay Borisov * hasn't been fully initialised and we are the only user 2744336a0d8dSNikolay Borisov */ 2745453e4873SNikolay Borisov ret = btrfs_init_root_free_objectid(tree_root); 2746b8522a1eSNikolay Borisov if (ret < 0) { 2747b8522a1eSNikolay Borisov handle_error = true; 2748b8522a1eSNikolay Borisov continue; 2749b8522a1eSNikolay Borisov } 2750b8522a1eSNikolay Borisov 27516b8fad57SNikolay Borisov ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID); 2752b8522a1eSNikolay Borisov 2753b8522a1eSNikolay Borisov ret = btrfs_read_roots(fs_info); 2754b8522a1eSNikolay Borisov if (ret < 0) { 2755b8522a1eSNikolay Borisov handle_error = true; 2756b8522a1eSNikolay Borisov continue; 2757b8522a1eSNikolay Borisov } 2758b8522a1eSNikolay Borisov 2759b8522a1eSNikolay Borisov /* All successful */ 2760b8522a1eSNikolay Borisov fs_info->generation = generation; 2761b8522a1eSNikolay Borisov fs_info->last_trans_committed = generation; 27626ef108ddSNikolay Borisov 27636ef108ddSNikolay Borisov /* Always begin writing backup roots after the one being used */ 27646ef108ddSNikolay Borisov if (backup_index < 0) { 27656ef108ddSNikolay Borisov fs_info->backup_root_index = 0; 27666ef108ddSNikolay Borisov } else { 27676ef108ddSNikolay Borisov fs_info->backup_root_index = backup_index + 1; 27686ef108ddSNikolay Borisov fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS; 27696ef108ddSNikolay Borisov } 2770b8522a1eSNikolay Borisov break; 2771b8522a1eSNikolay Borisov } 2772b8522a1eSNikolay Borisov 2773b8522a1eSNikolay Borisov return ret; 2774b8522a1eSNikolay Borisov } 2775b8522a1eSNikolay Borisov 27768260edbaSJosef Bacik void btrfs_init_fs_info(struct btrfs_fs_info *fs_info) 2777eb60ceacSChris Mason { 277876dda93cSYan, Zheng INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 2779f28491e0SJosef Bacik INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 27808fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 2781facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 278224bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 2783eb73c1b7SMiao Xie INIT_LIST_HEAD(&fs_info->delalloc_roots); 278411833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 2785eb73c1b7SMiao Xie spin_lock_init(&fs_info->delalloc_root_lock); 2786a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 278776dda93cSYan, Zheng spin_lock_init(&fs_info->fs_roots_radix_lock); 278824bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 27894cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 2790ceda0864SMiao Xie spin_lock_init(&fs_info->super_lock); 2791f28491e0SJosef Bacik spin_lock_init(&fs_info->buffer_lock); 279247ab2a6cSJosef Bacik spin_lock_init(&fs_info->unused_bgs_lock); 279340ab3be1SNaohiro Aota spin_lock_init(&fs_info->treelog_bg_lock); 2794f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 2795d7c15171SZhao Lei mutex_init(&fs_info->unused_bg_unpin_mutex); 279667c5e7d4SFilipe Manana mutex_init(&fs_info->delete_unused_bgs_mutex); 27977585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 2798573bfb72SMiao Xie mutex_init(&fs_info->delalloc_root_mutex); 27990bc09ca1SNaohiro Aota mutex_init(&fs_info->zoned_meta_io_lock); 2800de98ced9SMiao Xie seqlock_init(&fs_info->profiles_lock); 280119c00ddcSChris Mason 28020b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 28036324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 2804f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 280547ab2a6cSJosef Bacik INIT_LIST_HEAD(&fs_info->unused_bgs); 2806bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 2807bd647ce3SJosef Bacik INIT_LIST_HEAD(&fs_info->allocated_roots); 28083fd63727SJosef Bacik INIT_LIST_HEAD(&fs_info->allocated_ebs); 28093fd63727SJosef Bacik spin_lock_init(&fs_info->eb_leak_lock); 2810bd647ce3SJosef Bacik #endif 2811c8bf1b67SDavid Sterba extent_map_tree_init(&fs_info->mapping_tree); 281266d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 281366d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 281466d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 281566d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 281666d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 281766d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 281866d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 2819ba2c4d4eSJosef Bacik btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, 2820ba2c4d4eSJosef Bacik BTRFS_BLOCK_RSV_DELREFS); 2821ba2c4d4eSJosef Bacik 2822771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 28234cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 28242fefd558SZhao Lei atomic_set(&fs_info->reada_works_cnt, 0); 2825034f784dSJosef Bacik atomic_set(&fs_info->nr_delayed_iputs, 0); 2826fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 282795ac567aSFilipe David Borba Manana fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 28289ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 28294cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 2830a5ed45f8SNikolay Borisov atomic64_set(&fs_info->free_chunk_space, 0); 2831f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 28328b87dc17SDavid Sterba fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 2833f8c269d7SDavid Sterba fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 283490519d66SArne Jansen /* readahead state */ 2835d0164adcSMel Gorman INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); 283690519d66SArne Jansen spin_lock_init(&fs_info->reada_lock); 2837fd708b81SJosef Bacik btrfs_init_ref_verify(fs_info); 2838c8b97818SChris Mason 2839b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 2840b34b086cSChris Mason num_online_cpus() + 2, 8); 28410afbaf8cSChris Mason 2842199c2a9cSMiao Xie INIT_LIST_HEAD(&fs_info->ordered_roots); 2843199c2a9cSMiao Xie spin_lock_init(&fs_info->ordered_root_lock); 284469fe2d75SJosef Bacik 2845638aa7edSEric Sandeen btrfs_init_scrub(fs_info); 284621adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 284721adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 284821adbd5cSStefan Behrens #endif 2849779a65a4SEric Sandeen btrfs_init_balance(fs_info); 285057056740SJosef Bacik btrfs_init_async_reclaim_work(fs_info); 2851a2de733cSArne Jansen 28520f9dd46cSJosef Bacik spin_lock_init(&fs_info->block_group_cache_lock); 28536bef4d31SEric Paris fs_info->block_group_cache_tree = RB_ROOT; 2854a1897fddSLiu Bo fs_info->first_logical_byte = (u64)-1; 28550f9dd46cSJosef Bacik 2856fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &fs_info->excluded_extents, 2857fe119a6eSNikolay Borisov IO_TREE_FS_EXCLUDED_EXTENTS, NULL); 2858afcdd129SJosef Bacik set_bit(BTRFS_FS_BARRIER, &fs_info->flags); 285939279cc3SChris Mason 28605a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 2861e02119d5SChris Mason mutex_init(&fs_info->tree_log_mutex); 2862925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 2863a74a4b97SChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 2864a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 28651bbc621eSChris Mason mutex_init(&fs_info->ro_block_group_mutex); 28669e351cc8SJosef Bacik init_rwsem(&fs_info->commit_root_sem); 2867c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 286876dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 2869803b2f54SStefan Behrens sema_init(&fs_info->uuid_tree_rescan_sem, 1); 2870fa9c0d79SChris Mason 2871ad618368SEric Sandeen btrfs_init_dev_replace_locks(fs_info); 2872f9e92e40SEric Sandeen btrfs_init_qgroup(fs_info); 2873b0643e59SDennis Zhou btrfs_discard_init(fs_info); 2874416ac51dSArne Jansen 2875fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2876fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2877fa9c0d79SChris Mason 2878e6dcd2dcSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 2879f9295749SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 2880bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 28814854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 2882034f784dSJosef Bacik init_waitqueue_head(&fs_info->delayed_iputs_wait); 28833768f368SChris Mason 2884da17066cSJeff Mahoney /* Usable values until the real ones are cached from the superblock */ 2885da17066cSJeff Mahoney fs_info->nodesize = 4096; 2886da17066cSJeff Mahoney fs_info->sectorsize = 4096; 2887ab108d99SDavid Sterba fs_info->sectorsize_bits = ilog2(4096); 2888da17066cSJeff Mahoney fs_info->stripesize = 4096; 2889da17066cSJeff Mahoney 2890eede2bf3SOmar Sandoval spin_lock_init(&fs_info->swapfile_pins_lock); 2891eede2bf3SOmar Sandoval fs_info->swapfile_pins = RB_ROOT; 2892eede2bf3SOmar Sandoval 28939e967495SFilipe Manana fs_info->send_in_progress = 0; 28948260edbaSJosef Bacik } 28958260edbaSJosef Bacik 28968260edbaSJosef Bacik static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb) 28978260edbaSJosef Bacik { 28988260edbaSJosef Bacik int ret; 28998260edbaSJosef Bacik 29008260edbaSJosef Bacik fs_info->sb = sb; 29018260edbaSJosef Bacik sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; 29028260edbaSJosef Bacik sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); 29039e967495SFilipe Manana 29045deb17e1SJosef Bacik ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL); 2905ae18c37aSJosef Bacik if (ret) 2906ae18c37aSJosef Bacik return ret; 2907ae18c37aSJosef Bacik 2908ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 2909ae18c37aSJosef Bacik if (ret) 2910c75e8394SJosef Bacik return ret; 2911ae18c37aSJosef Bacik 2912ae18c37aSJosef Bacik fs_info->dirty_metadata_batch = PAGE_SIZE * 2913ae18c37aSJosef Bacik (1 + ilog2(nr_cpu_ids)); 2914ae18c37aSJosef Bacik 2915ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 2916ae18c37aSJosef Bacik if (ret) 2917c75e8394SJosef Bacik return ret; 2918ae18c37aSJosef Bacik 2919ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, 2920ae18c37aSJosef Bacik GFP_KERNEL); 2921ae18c37aSJosef Bacik if (ret) 2922c75e8394SJosef Bacik return ret; 2923ae18c37aSJosef Bacik 2924ae18c37aSJosef Bacik fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 2925ae18c37aSJosef Bacik GFP_KERNEL); 2926c75e8394SJosef Bacik if (!fs_info->delayed_root) 2927c75e8394SJosef Bacik return -ENOMEM; 2928ae18c37aSJosef Bacik btrfs_init_delayed_root(fs_info->delayed_root); 2929ae18c37aSJosef Bacik 2930a0a1db70SFilipe Manana if (sb_rdonly(sb)) 2931a0a1db70SFilipe Manana set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state); 2932a0a1db70SFilipe Manana 2933c75e8394SJosef Bacik return btrfs_alloc_stripe_hash_table(fs_info); 2934ae18c37aSJosef Bacik } 2935ae18c37aSJosef Bacik 293697f4dd09SNikolay Borisov static int btrfs_uuid_rescan_kthread(void *data) 293797f4dd09SNikolay Borisov { 293897f4dd09SNikolay Borisov struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; 293997f4dd09SNikolay Borisov int ret; 294097f4dd09SNikolay Borisov 294197f4dd09SNikolay Borisov /* 294297f4dd09SNikolay Borisov * 1st step is to iterate through the existing UUID tree and 294397f4dd09SNikolay Borisov * to delete all entries that contain outdated data. 294497f4dd09SNikolay Borisov * 2nd step is to add all missing entries to the UUID tree. 294597f4dd09SNikolay Borisov */ 294697f4dd09SNikolay Borisov ret = btrfs_uuid_tree_iterate(fs_info); 294797f4dd09SNikolay Borisov if (ret < 0) { 2948c94bec2cSJosef Bacik if (ret != -EINTR) 2949c94bec2cSJosef Bacik btrfs_warn(fs_info, "iterating uuid_tree failed %d", 2950c94bec2cSJosef Bacik ret); 295197f4dd09SNikolay Borisov up(&fs_info->uuid_tree_rescan_sem); 295297f4dd09SNikolay Borisov return ret; 295397f4dd09SNikolay Borisov } 295497f4dd09SNikolay Borisov return btrfs_uuid_scan_kthread(data); 295597f4dd09SNikolay Borisov } 295697f4dd09SNikolay Borisov 295797f4dd09SNikolay Borisov static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) 295897f4dd09SNikolay Borisov { 295997f4dd09SNikolay Borisov struct task_struct *task; 296097f4dd09SNikolay Borisov 296197f4dd09SNikolay Borisov down(&fs_info->uuid_tree_rescan_sem); 296297f4dd09SNikolay Borisov task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); 296397f4dd09SNikolay Borisov if (IS_ERR(task)) { 296497f4dd09SNikolay Borisov /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 296597f4dd09SNikolay Borisov btrfs_warn(fs_info, "failed to start uuid_rescan task"); 296697f4dd09SNikolay Borisov up(&fs_info->uuid_tree_rescan_sem); 296797f4dd09SNikolay Borisov return PTR_ERR(task); 296897f4dd09SNikolay Borisov } 296997f4dd09SNikolay Borisov 297097f4dd09SNikolay Borisov return 0; 297197f4dd09SNikolay Borisov } 297297f4dd09SNikolay Borisov 297344c0ca21SBoris Burkov /* 29748cd29088SBoris Burkov * Some options only have meaning at mount time and shouldn't persist across 29758cd29088SBoris Burkov * remounts, or be displayed. Clear these at the end of mount and remount 29768cd29088SBoris Burkov * code paths. 29778cd29088SBoris Burkov */ 29788cd29088SBoris Burkov void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info) 29798cd29088SBoris Burkov { 29808cd29088SBoris Burkov btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT); 29818b228324SBoris Burkov btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE); 29828cd29088SBoris Burkov } 29838cd29088SBoris Burkov 29848cd29088SBoris Burkov /* 298544c0ca21SBoris Burkov * Mounting logic specific to read-write file systems. Shared by open_ctree 298644c0ca21SBoris Burkov * and btrfs_remount when remounting from read-only to read-write. 298744c0ca21SBoris Burkov */ 298844c0ca21SBoris Burkov int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info) 298944c0ca21SBoris Burkov { 299044c0ca21SBoris Burkov int ret; 299194846229SBoris Burkov const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE); 29928b228324SBoris Burkov bool clear_free_space_tree = false; 29938b228324SBoris Burkov 29948b228324SBoris Burkov if (btrfs_test_opt(fs_info, CLEAR_CACHE) && 29958b228324SBoris Burkov btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 29968b228324SBoris Burkov clear_free_space_tree = true; 29978b228324SBoris Burkov } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 29988b228324SBoris Burkov !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { 29998b228324SBoris Burkov btrfs_warn(fs_info, "free space tree is invalid"); 30008b228324SBoris Burkov clear_free_space_tree = true; 30018b228324SBoris Burkov } 30028b228324SBoris Burkov 30038b228324SBoris Burkov if (clear_free_space_tree) { 30048b228324SBoris Burkov btrfs_info(fs_info, "clearing free space tree"); 30058b228324SBoris Burkov ret = btrfs_clear_free_space_tree(fs_info); 30068b228324SBoris Burkov if (ret) { 30078b228324SBoris Burkov btrfs_warn(fs_info, 30088b228324SBoris Burkov "failed to clear free space tree: %d", ret); 30098b228324SBoris Burkov goto out; 30108b228324SBoris Burkov } 30118b228324SBoris Burkov } 301244c0ca21SBoris Burkov 301344c0ca21SBoris Burkov ret = btrfs_cleanup_fs_roots(fs_info); 301444c0ca21SBoris Burkov if (ret) 301544c0ca21SBoris Burkov goto out; 301644c0ca21SBoris Burkov 30178f1c21d7SBoris Burkov down_read(&fs_info->cleanup_work_sem); 30188f1c21d7SBoris Burkov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 30198f1c21d7SBoris Burkov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 30208f1c21d7SBoris Burkov up_read(&fs_info->cleanup_work_sem); 30218f1c21d7SBoris Burkov goto out; 30228f1c21d7SBoris Burkov } 30238f1c21d7SBoris Burkov up_read(&fs_info->cleanup_work_sem); 30248f1c21d7SBoris Burkov 302544c0ca21SBoris Burkov mutex_lock(&fs_info->cleaner_mutex); 302644c0ca21SBoris Burkov ret = btrfs_recover_relocation(fs_info->tree_root); 302744c0ca21SBoris Burkov mutex_unlock(&fs_info->cleaner_mutex); 302844c0ca21SBoris Burkov if (ret < 0) { 302944c0ca21SBoris Burkov btrfs_warn(fs_info, "failed to recover relocation: %d", ret); 303044c0ca21SBoris Burkov goto out; 303144c0ca21SBoris Burkov } 303244c0ca21SBoris Burkov 30335011139aSBoris Burkov if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && 30345011139aSBoris Burkov !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 30355011139aSBoris Burkov btrfs_info(fs_info, "creating free space tree"); 30365011139aSBoris Burkov ret = btrfs_create_free_space_tree(fs_info); 30375011139aSBoris Burkov if (ret) { 30385011139aSBoris Burkov btrfs_warn(fs_info, 30395011139aSBoris Burkov "failed to create free space tree: %d", ret); 30405011139aSBoris Burkov goto out; 30415011139aSBoris Burkov } 30425011139aSBoris Burkov } 30435011139aSBoris Burkov 304494846229SBoris Burkov if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) { 304594846229SBoris Burkov ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt); 304694846229SBoris Burkov if (ret) 304794846229SBoris Burkov goto out; 304894846229SBoris Burkov } 304994846229SBoris Burkov 305044c0ca21SBoris Burkov ret = btrfs_resume_balance_async(fs_info); 305144c0ca21SBoris Burkov if (ret) 305244c0ca21SBoris Burkov goto out; 305344c0ca21SBoris Burkov 305444c0ca21SBoris Burkov ret = btrfs_resume_dev_replace_async(fs_info); 305544c0ca21SBoris Burkov if (ret) { 305644c0ca21SBoris Burkov btrfs_warn(fs_info, "failed to resume dev_replace"); 305744c0ca21SBoris Burkov goto out; 305844c0ca21SBoris Burkov } 305944c0ca21SBoris Burkov 306044c0ca21SBoris Burkov btrfs_qgroup_rescan_resume(fs_info); 306144c0ca21SBoris Burkov 306244c0ca21SBoris Burkov if (!fs_info->uuid_root) { 306344c0ca21SBoris Burkov btrfs_info(fs_info, "creating UUID tree"); 306444c0ca21SBoris Burkov ret = btrfs_create_uuid_tree(fs_info); 306544c0ca21SBoris Burkov if (ret) { 306644c0ca21SBoris Burkov btrfs_warn(fs_info, 306744c0ca21SBoris Burkov "failed to create the UUID tree %d", ret); 306844c0ca21SBoris Burkov goto out; 306944c0ca21SBoris Burkov } 307044c0ca21SBoris Burkov } 307144c0ca21SBoris Burkov 3072638331faSFilipe Manana ret = btrfs_find_orphan_roots(fs_info); 307344c0ca21SBoris Burkov out: 307444c0ca21SBoris Burkov return ret; 307544c0ca21SBoris Burkov } 307644c0ca21SBoris Burkov 3077ae18c37aSJosef Bacik int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices, 3078ae18c37aSJosef Bacik char *options) 3079ae18c37aSJosef Bacik { 3080ae18c37aSJosef Bacik u32 sectorsize; 3081ae18c37aSJosef Bacik u32 nodesize; 3082ae18c37aSJosef Bacik u32 stripesize; 3083ae18c37aSJosef Bacik u64 generation; 3084ae18c37aSJosef Bacik u64 features; 3085ae18c37aSJosef Bacik u16 csum_type; 3086ae18c37aSJosef Bacik struct btrfs_super_block *disk_super; 3087ae18c37aSJosef Bacik struct btrfs_fs_info *fs_info = btrfs_sb(sb); 3088ae18c37aSJosef Bacik struct btrfs_root *tree_root; 3089ae18c37aSJosef Bacik struct btrfs_root *chunk_root; 3090ae18c37aSJosef Bacik int ret; 3091ae18c37aSJosef Bacik int err = -EINVAL; 3092ae18c37aSJosef Bacik int level; 3093ae18c37aSJosef Bacik 30948260edbaSJosef Bacik ret = init_mount_fs_info(fs_info, sb); 309553b381b3SDavid Woodhouse if (ret) { 309683c8266aSDavid Sterba err = ret; 3097ae18c37aSJosef Bacik goto fail; 309853b381b3SDavid Woodhouse } 309953b381b3SDavid Woodhouse 3100ae18c37aSJosef Bacik /* These need to be init'ed before we start creating inodes and such. */ 3101ae18c37aSJosef Bacik tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, 3102ae18c37aSJosef Bacik GFP_KERNEL); 3103ae18c37aSJosef Bacik fs_info->tree_root = tree_root; 3104ae18c37aSJosef Bacik chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID, 3105ae18c37aSJosef Bacik GFP_KERNEL); 3106ae18c37aSJosef Bacik fs_info->chunk_root = chunk_root; 3107ae18c37aSJosef Bacik if (!tree_root || !chunk_root) { 3108ae18c37aSJosef Bacik err = -ENOMEM; 3109c75e8394SJosef Bacik goto fail; 3110ae18c37aSJosef Bacik } 3111ae18c37aSJosef Bacik 3112ae18c37aSJosef Bacik fs_info->btree_inode = new_inode(sb); 3113ae18c37aSJosef Bacik if (!fs_info->btree_inode) { 3114ae18c37aSJosef Bacik err = -ENOMEM; 3115c75e8394SJosef Bacik goto fail; 3116ae18c37aSJosef Bacik } 3117ae18c37aSJosef Bacik mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 3118ae18c37aSJosef Bacik btrfs_init_btree_inode(fs_info); 3119ae18c37aSJosef Bacik 31203c4bb26bSChris Mason invalidate_bdev(fs_devices->latest_bdev); 31211104a885SDavid Sterba 31221104a885SDavid Sterba /* 31231104a885SDavid Sterba * Read super block and check the signature bytes only 31241104a885SDavid Sterba */ 31258f32380dSJohannes Thumshirn disk_super = btrfs_read_dev_super(fs_devices->latest_bdev); 31268f32380dSJohannes Thumshirn if (IS_ERR(disk_super)) { 31278f32380dSJohannes Thumshirn err = PTR_ERR(disk_super); 312816cdcec7SMiao Xie goto fail_alloc; 312920b45077SDave Young } 313039279cc3SChris Mason 31311104a885SDavid Sterba /* 3132260db43cSRandy Dunlap * Verify the type first, if that or the checksum value are 31338dc3f22cSJohannes Thumshirn * corrupted, we'll find out 31348dc3f22cSJohannes Thumshirn */ 31358f32380dSJohannes Thumshirn csum_type = btrfs_super_csum_type(disk_super); 313651bce6c9SJohannes Thumshirn if (!btrfs_supported_super_csum(csum_type)) { 31378dc3f22cSJohannes Thumshirn btrfs_err(fs_info, "unsupported checksum algorithm: %u", 313851bce6c9SJohannes Thumshirn csum_type); 31398dc3f22cSJohannes Thumshirn err = -EINVAL; 31408f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 31418dc3f22cSJohannes Thumshirn goto fail_alloc; 31428dc3f22cSJohannes Thumshirn } 31438dc3f22cSJohannes Thumshirn 31446d97c6e3SJohannes Thumshirn ret = btrfs_init_csum_hash(fs_info, csum_type); 31456d97c6e3SJohannes Thumshirn if (ret) { 31466d97c6e3SJohannes Thumshirn err = ret; 31478f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 31486d97c6e3SJohannes Thumshirn goto fail_alloc; 31496d97c6e3SJohannes Thumshirn } 31506d97c6e3SJohannes Thumshirn 31518dc3f22cSJohannes Thumshirn /* 31521104a885SDavid Sterba * We want to check superblock checksum, the type is stored inside. 31531104a885SDavid Sterba * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 31541104a885SDavid Sterba */ 31558f32380dSJohannes Thumshirn if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) { 315605135f59SDavid Sterba btrfs_err(fs_info, "superblock checksum mismatch"); 31571104a885SDavid Sterba err = -EINVAL; 31588f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 3159141386e1SJosef Bacik goto fail_alloc; 31601104a885SDavid Sterba } 31611104a885SDavid Sterba 31621104a885SDavid Sterba /* 31631104a885SDavid Sterba * super_copy is zeroed at allocation time and we never touch the 31641104a885SDavid Sterba * following bytes up to INFO_SIZE, the checksum is calculated from 31651104a885SDavid Sterba * the whole block of INFO_SIZE 31661104a885SDavid Sterba */ 31678f32380dSJohannes Thumshirn memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy)); 31688f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 31695f39d397SChris Mason 3170fbc6feaeSNikolay Borisov disk_super = fs_info->super_copy; 3171fbc6feaeSNikolay Borisov 3172de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, 3173de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 3174de37aa51SNikolay Borisov 31757239ff4bSNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID)) { 3176de37aa51SNikolay Borisov ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid, 3177de37aa51SNikolay Borisov fs_info->super_copy->metadata_uuid, 3178de37aa51SNikolay Borisov BTRFS_FSID_SIZE)); 31797239ff4bSNikolay Borisov } 31800b86a832SChris Mason 3181fbc6feaeSNikolay Borisov features = btrfs_super_flags(disk_super); 3182fbc6feaeSNikolay Borisov if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { 3183fbc6feaeSNikolay Borisov features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; 3184fbc6feaeSNikolay Borisov btrfs_set_super_flags(disk_super, features); 3185fbc6feaeSNikolay Borisov btrfs_info(fs_info, 3186fbc6feaeSNikolay Borisov "found metadata UUID change in progress flag, clearing"); 3187fbc6feaeSNikolay Borisov } 3188fbc6feaeSNikolay Borisov 3189fbc6feaeSNikolay Borisov memcpy(fs_info->super_for_commit, fs_info->super_copy, 3190fbc6feaeSNikolay Borisov sizeof(*fs_info->super_for_commit)); 3191de37aa51SNikolay Borisov 3192069ec957SQu Wenruo ret = btrfs_validate_mount_super(fs_info); 31931104a885SDavid Sterba if (ret) { 319405135f59SDavid Sterba btrfs_err(fs_info, "superblock contains fatal errors"); 31951104a885SDavid Sterba err = -EINVAL; 3196141386e1SJosef Bacik goto fail_alloc; 31971104a885SDavid Sterba } 31981104a885SDavid Sterba 31990f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 3200141386e1SJosef Bacik goto fail_alloc; 32010f7d52f4SChris Mason 3202acce952bSliubo /* check FS state, whether FS is broken. */ 320387533c47SMiao Xie if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 320487533c47SMiao Xie set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 3205acce952bSliubo 320675e7cb7fSLiu Bo /* 320775e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 320875e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 320975e7cb7fSLiu Bo */ 321075e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 321175e7cb7fSLiu Bo 32122ff7e61eSJeff Mahoney ret = btrfs_parse_options(fs_info, options, sb->s_flags); 32132b82032cSYan Zheng if (ret) { 32142b82032cSYan Zheng err = ret; 3215141386e1SJosef Bacik goto fail_alloc; 32162b82032cSYan Zheng } 3217dfe25020SChris Mason 3218f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 3219f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 3220f2b636e8SJosef Bacik if (features) { 322105135f59SDavid Sterba btrfs_err(fs_info, 322205135f59SDavid Sterba "cannot mount because of unsupported optional features (%llx)", 3223c1c9ff7cSGeert Uytterhoeven features); 3224f2b636e8SJosef Bacik err = -EINVAL; 3225141386e1SJosef Bacik goto fail_alloc; 3226f2b636e8SJosef Bacik } 3227f2b636e8SJosef Bacik 32285d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 32295d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 32300b246afaSJeff Mahoney if (fs_info->compress_type == BTRFS_COMPRESS_LZO) 3231a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 32325c1aab1dSNick Terrell else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) 32335c1aab1dSNick Terrell features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; 3234727011e0SChris Mason 32353173a18fSJosef Bacik if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) 323605135f59SDavid Sterba btrfs_info(fs_info, "has skinny extents"); 32373173a18fSJosef Bacik 3238727011e0SChris Mason /* 3239727011e0SChris Mason * flag our filesystem as having big metadata blocks if 3240727011e0SChris Mason * they are bigger than the page size 3241727011e0SChris Mason */ 324209cbfeafSKirill A. Shutemov if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) { 3243727011e0SChris Mason if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 324405135f59SDavid Sterba btrfs_info(fs_info, 324505135f59SDavid Sterba "flagging fs with big metadata feature"); 3246727011e0SChris Mason features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 3247727011e0SChris Mason } 3248727011e0SChris Mason 3249bc3f116fSChris Mason nodesize = btrfs_super_nodesize(disk_super); 3250bc3f116fSChris Mason sectorsize = btrfs_super_sectorsize(disk_super); 3251b7f67055SChandan Rajendra stripesize = sectorsize; 3252707e8a07SDavid Sterba fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 3253963d678bSMiao Xie fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 3254bc3f116fSChris Mason 3255da17066cSJeff Mahoney /* Cache block sizes */ 3256da17066cSJeff Mahoney fs_info->nodesize = nodesize; 3257da17066cSJeff Mahoney fs_info->sectorsize = sectorsize; 3258ab108d99SDavid Sterba fs_info->sectorsize_bits = ilog2(sectorsize); 325922b6331dSDavid Sterba fs_info->csum_size = btrfs_super_csum_size(disk_super); 3260fe5ecbe8SDavid Sterba fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size; 3261da17066cSJeff Mahoney fs_info->stripesize = stripesize; 3262da17066cSJeff Mahoney 3263bc3f116fSChris Mason /* 3264bc3f116fSChris Mason * mixed block groups end up with duplicate but slightly offset 3265bc3f116fSChris Mason * extent buffers for the same range. It leads to corruptions 3266bc3f116fSChris Mason */ 3267bc3f116fSChris Mason if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 3268707e8a07SDavid Sterba (sectorsize != nodesize)) { 326905135f59SDavid Sterba btrfs_err(fs_info, 327005135f59SDavid Sterba "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", 327105135f59SDavid Sterba nodesize, sectorsize); 3272141386e1SJosef Bacik goto fail_alloc; 3273bc3f116fSChris Mason } 3274bc3f116fSChris Mason 3275ceda0864SMiao Xie /* 3276ceda0864SMiao Xie * Needn't use the lock because there is no other task which will 3277ceda0864SMiao Xie * update the flag. 3278ceda0864SMiao Xie */ 32795d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 32805d4f98a2SYan Zheng 3281f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 3282f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 3283bc98a42cSDavid Howells if (!sb_rdonly(sb) && features) { 328405135f59SDavid Sterba btrfs_err(fs_info, 328505135f59SDavid Sterba "cannot mount read-write because of unsupported optional features (%llx)", 3286c1c9ff7cSGeert Uytterhoeven features); 3287f2b636e8SJosef Bacik err = -EINVAL; 3288141386e1SJosef Bacik goto fail_alloc; 3289f2b636e8SJosef Bacik } 329061d92c32SChris Mason 32910bb3eb3eSQu Wenruo /* For 4K sector size support, it's only read-only */ 32920bb3eb3eSQu Wenruo if (PAGE_SIZE == SZ_64K && sectorsize == SZ_4K) { 32930bb3eb3eSQu Wenruo if (!sb_rdonly(sb) || btrfs_super_log_root(disk_super)) { 32940bb3eb3eSQu Wenruo btrfs_err(fs_info, 32950bb3eb3eSQu Wenruo "subpage sectorsize %u only supported read-only for page size %lu", 32960bb3eb3eSQu Wenruo sectorsize, PAGE_SIZE); 32970bb3eb3eSQu Wenruo err = -EINVAL; 32980bb3eb3eSQu Wenruo goto fail_alloc; 32990bb3eb3eSQu Wenruo } 33000bb3eb3eSQu Wenruo } 33010bb3eb3eSQu Wenruo 33022a458198SEric Sandeen ret = btrfs_init_workqueues(fs_info, fs_devices); 33032a458198SEric Sandeen if (ret) { 33042a458198SEric Sandeen err = ret; 33050dc3b84aSJosef Bacik goto fail_sb_buffer; 33060dc3b84aSJosef Bacik } 33074543df7eSChris Mason 33089e11ceeeSJan Kara sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); 33099e11ceeeSJan Kara sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); 33104575c9ccSChris Mason 3311a061fc8dSChris Mason sb->s_blocksize = sectorsize; 3312a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 3313de37aa51SNikolay Borisov memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); 3314db94535dSChris Mason 3315925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 33166bccf3abSJeff Mahoney ret = btrfs_read_sys_array(fs_info); 3317925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 331884eed90fSChris Mason if (ret) { 331905135f59SDavid Sterba btrfs_err(fs_info, "failed to read the system array: %d", ret); 33205d4f98a2SYan Zheng goto fail_sb_buffer; 332184eed90fSChris Mason } 33220b86a832SChris Mason 332384234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 3324581c1760SQu Wenruo level = btrfs_super_chunk_root_level(disk_super); 33250b86a832SChris Mason 33262ff7e61eSJeff Mahoney chunk_root->node = read_tree_block(fs_info, 33270b86a832SChris Mason btrfs_super_chunk_root(disk_super), 33281b7ec85eSJosef Bacik BTRFS_CHUNK_TREE_OBJECTID, 3329581c1760SQu Wenruo generation, level, NULL); 333064c043deSLiu Bo if (IS_ERR(chunk_root->node) || 333164c043deSLiu Bo !extent_buffer_uptodate(chunk_root->node)) { 333205135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk root"); 3333e5fffbacSchandan if (!IS_ERR(chunk_root->node)) 3334e5fffbacSchandan free_extent_buffer(chunk_root->node); 333595ab1f64SZhao Lei chunk_root->node = NULL; 3336af31f5e5SChris Mason goto fail_tree_roots; 333783121942SDavid Woodhouse } 33385d4f98a2SYan Zheng btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 33395d4f98a2SYan Zheng chunk_root->commit_root = btrfs_root_node(chunk_root); 33400b86a832SChris Mason 3341e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 3342c4ac7541SDavid Sterba offsetof(struct btrfs_header, chunk_tree_uuid), 3343c4ac7541SDavid Sterba BTRFS_UUID_SIZE); 3344e17cade2SChris Mason 33455b4aacefSJeff Mahoney ret = btrfs_read_chunk_tree(fs_info); 33462b82032cSYan Zheng if (ret) { 334705135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk tree: %d", ret); 3348af31f5e5SChris Mason goto fail_tree_roots; 33492b82032cSYan Zheng } 33500b86a832SChris Mason 33518dabb742SStefan Behrens /* 3352bacce86aSAnand Jain * At this point we know all the devices that make this filesystem, 3353bacce86aSAnand Jain * including the seed devices but we don't know yet if the replace 3354bacce86aSAnand Jain * target is required. So free devices that are not part of this 3355bacce86aSAnand Jain * filesystem but skip the replace traget device which is checked 3356bacce86aSAnand Jain * below in btrfs_init_dev_replace(). 33578dabb742SStefan Behrens */ 3358bacce86aSAnand Jain btrfs_free_extra_devids(fs_devices); 3359a6b0d5c8SChris Mason if (!fs_devices->latest_bdev) { 336005135f59SDavid Sterba btrfs_err(fs_info, "failed to read devices"); 3361a6b0d5c8SChris Mason goto fail_tree_roots; 3362a6b0d5c8SChris Mason } 3363a6b0d5c8SChris Mason 3364b8522a1eSNikolay Borisov ret = init_tree_roots(fs_info); 33654bbcaa64SEric Sandeen if (ret) 3366b8522a1eSNikolay Borisov goto fail_tree_roots; 33678929ecfaSYan, Zheng 336875ec1db8SJosef Bacik /* 336973651042SNaohiro Aota * Get zone type information of zoned block devices. This will also 337073651042SNaohiro Aota * handle emulation of a zoned filesystem if a regular device has the 337173651042SNaohiro Aota * zoned incompat feature flag set. 337273651042SNaohiro Aota */ 337373651042SNaohiro Aota ret = btrfs_get_dev_zone_info_all_devices(fs_info); 337473651042SNaohiro Aota if (ret) { 337573651042SNaohiro Aota btrfs_err(fs_info, 337673651042SNaohiro Aota "zoned: failed to read device zone info: %d", 337773651042SNaohiro Aota ret); 337873651042SNaohiro Aota goto fail_block_groups; 337973651042SNaohiro Aota } 338073651042SNaohiro Aota 338173651042SNaohiro Aota /* 338275ec1db8SJosef Bacik * If we have a uuid root and we're not being told to rescan we need to 338375ec1db8SJosef Bacik * check the generation here so we can set the 338475ec1db8SJosef Bacik * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the 338575ec1db8SJosef Bacik * transaction during a balance or the log replay without updating the 338675ec1db8SJosef Bacik * uuid generation, and then if we crash we would rescan the uuid tree, 338775ec1db8SJosef Bacik * even though it was perfectly fine. 338875ec1db8SJosef Bacik */ 338975ec1db8SJosef Bacik if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) && 339075ec1db8SJosef Bacik fs_info->generation == btrfs_super_uuid_tree_generation(disk_super)) 339175ec1db8SJosef Bacik set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 339275ec1db8SJosef Bacik 3393cf90d884SQu Wenruo ret = btrfs_verify_dev_extents(fs_info); 3394cf90d884SQu Wenruo if (ret) { 3395cf90d884SQu Wenruo btrfs_err(fs_info, 3396cf90d884SQu Wenruo "failed to verify dev extents against chunks: %d", 3397cf90d884SQu Wenruo ret); 3398cf90d884SQu Wenruo goto fail_block_groups; 3399cf90d884SQu Wenruo } 340068310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 340168310a5eSIlya Dryomov if (ret) { 340205135f59SDavid Sterba btrfs_err(fs_info, "failed to recover balance: %d", ret); 340368310a5eSIlya Dryomov goto fail_block_groups; 340468310a5eSIlya Dryomov } 340568310a5eSIlya Dryomov 3406733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 3407733f4fbbSStefan Behrens if (ret) { 340805135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_stats: %d", ret); 3409733f4fbbSStefan Behrens goto fail_block_groups; 3410733f4fbbSStefan Behrens } 3411733f4fbbSStefan Behrens 34128dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 34138dabb742SStefan Behrens if (ret) { 341405135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_replace: %d", ret); 34158dabb742SStefan Behrens goto fail_block_groups; 34168dabb742SStefan Behrens } 34178dabb742SStefan Behrens 3418b70f5097SNaohiro Aota ret = btrfs_check_zoned_mode(fs_info); 3419b70f5097SNaohiro Aota if (ret) { 3420b70f5097SNaohiro Aota btrfs_err(fs_info, "failed to initialize zoned mode: %d", 3421b70f5097SNaohiro Aota ret); 3422b70f5097SNaohiro Aota goto fail_block_groups; 3423b70f5097SNaohiro Aota } 3424b70f5097SNaohiro Aota 3425c6761a9eSAnand Jain ret = btrfs_sysfs_add_fsid(fs_devices); 3426b7c35e81SAnand Jain if (ret) { 342705135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", 342805135f59SDavid Sterba ret); 3429b7c35e81SAnand Jain goto fail_block_groups; 3430b7c35e81SAnand Jain } 3431b7c35e81SAnand Jain 343296f3136eSAnand Jain ret = btrfs_sysfs_add_mounted(fs_info); 34335ac1d209SJeff Mahoney if (ret) { 343405135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); 3435b7c35e81SAnand Jain goto fail_fsdev_sysfs; 34365ac1d209SJeff Mahoney } 34375ac1d209SJeff Mahoney 3438c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 3439c59021f8Sliubo if (ret) { 344005135f59SDavid Sterba btrfs_err(fs_info, "failed to initialize space info: %d", ret); 34412365dd3cSAnand Jain goto fail_sysfs; 3442c59021f8Sliubo } 3443c59021f8Sliubo 34445b4aacefSJeff Mahoney ret = btrfs_read_block_groups(fs_info); 34451b1d1f66SJosef Bacik if (ret) { 344605135f59SDavid Sterba btrfs_err(fs_info, "failed to read block groups: %d", ret); 34472365dd3cSAnand Jain goto fail_sysfs; 34481b1d1f66SJosef Bacik } 34494330e183SQu Wenruo 34506528b99dSAnand Jain if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) { 345105135f59SDavid Sterba btrfs_warn(fs_info, 345252042d8eSAndrea Gelmini "writable mount is not allowed due to too many missing devices"); 34532365dd3cSAnand Jain goto fail_sysfs; 3454292fd7fcSStefan Behrens } 34559078a3e1SChris Mason 3456a74a4b97SChris Mason fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 3457a74a4b97SChris Mason "btrfs-cleaner"); 345857506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 34592365dd3cSAnand Jain goto fail_sysfs; 3460a74a4b97SChris Mason 3461a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 3462a74a4b97SChris Mason tree_root, 3463a74a4b97SChris Mason "btrfs-transaction"); 346457506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 34653f157a2fSChris Mason goto fail_cleaner; 3466a74a4b97SChris Mason 3467583b7231SHans van Kranenburg if (!btrfs_test_opt(fs_info, NOSSD) && 3468c289811cSChris Mason !fs_info->fs_devices->rotating) { 3469583b7231SHans van Kranenburg btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); 3470c289811cSChris Mason } 3471c289811cSChris Mason 3472572d9ab7SDavid Sterba /* 347301327610SNicholas D Steeves * Mount does not set all options immediately, we can do it now and do 3474572d9ab7SDavid Sterba * not have to wait for transaction commit 3475572d9ab7SDavid Sterba */ 3476572d9ab7SDavid Sterba btrfs_apply_pending_changes(fs_info); 34773818aea2SQu Wenruo 347821adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 34790b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { 34802ff7e61eSJeff Mahoney ret = btrfsic_mount(fs_info, fs_devices, 34810b246afaSJeff Mahoney btrfs_test_opt(fs_info, 348221adbd5cSStefan Behrens CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? 348321adbd5cSStefan Behrens 1 : 0, 348421adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 348521adbd5cSStefan Behrens if (ret) 348605135f59SDavid Sterba btrfs_warn(fs_info, 348705135f59SDavid Sterba "failed to initialize integrity check module: %d", 348805135f59SDavid Sterba ret); 348921adbd5cSStefan Behrens } 349021adbd5cSStefan Behrens #endif 3491bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 3492bcef60f2SArne Jansen if (ret) 3493bcef60f2SArne Jansen goto fail_trans_kthread; 349421adbd5cSStefan Behrens 3495fd708b81SJosef Bacik if (btrfs_build_ref_tree(fs_info)) 3496fd708b81SJosef Bacik btrfs_err(fs_info, "couldn't build ref tree"); 3497fd708b81SJosef Bacik 349896da0919SQu Wenruo /* do not make disk changes in broken FS or nologreplay is given */ 349996da0919SQu Wenruo if (btrfs_super_log_root(disk_super) != 0 && 35000b246afaSJeff Mahoney !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 3501e8294f2fSDavid Sterba btrfs_info(fs_info, "start tree-log replay"); 350263443bf5SEric Sandeen ret = btrfs_replay_log(fs_info, fs_devices); 350379787eaaSJeff Mahoney if (ret) { 350463443bf5SEric Sandeen err = ret; 350528c16cbbSWang Shilong goto fail_qgroup; 3506e556ce2cSYan Zheng } 3507e02119d5SChris Mason } 35081a40e23bSZheng Yan 350956e9357aSDavid Sterba fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true); 35103140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 35113140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 3512f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read fs tree: %d", err); 3513315bf8efSJosef Bacik fs_info->fs_root = NULL; 3514bcef60f2SArne Jansen goto fail_qgroup; 35153140c9a3SDan Carpenter } 3516c289811cSChris Mason 3517bc98a42cSDavid Howells if (sb_rdonly(sb)) 35188cd29088SBoris Burkov goto clear_oneshot; 35192b6ba629SIlya Dryomov 352044c0ca21SBoris Burkov ret = btrfs_start_pre_rw_mount(fs_info); 35212b6ba629SIlya Dryomov if (ret) { 35226bccf3abSJeff Mahoney close_ctree(fs_info); 35232b6ba629SIlya Dryomov return ret; 3524e3acc2a6SJosef Bacik } 3525b0643e59SDennis Zhou btrfs_discard_resume(fs_info); 3526b382a324SJan Schmidt 352744c0ca21SBoris Burkov if (fs_info->uuid_root && 352844c0ca21SBoris Burkov (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || 352944c0ca21SBoris Burkov fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) { 353005135f59SDavid Sterba btrfs_info(fs_info, "checking UUID tree"); 353170f80175SStefan Behrens ret = btrfs_check_uuid_tree(fs_info); 353270f80175SStefan Behrens if (ret) { 353305135f59SDavid Sterba btrfs_warn(fs_info, 353405135f59SDavid Sterba "failed to check the UUID tree: %d", ret); 35356bccf3abSJeff Mahoney close_ctree(fs_info); 353670f80175SStefan Behrens return ret; 353770f80175SStefan Behrens } 3538f7a81ea4SStefan Behrens } 353994846229SBoris Burkov 3540afcdd129SJosef Bacik set_bit(BTRFS_FS_OPEN, &fs_info->flags); 354147ab2a6cSJosef Bacik 35428cd29088SBoris Burkov clear_oneshot: 35438cd29088SBoris Burkov btrfs_clear_oneshot_options(fs_info); 3544ad2b2c80SAl Viro return 0; 354539279cc3SChris Mason 3546bcef60f2SArne Jansen fail_qgroup: 3547bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 35487c2ca468SChris Mason fail_trans_kthread: 35497c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 35502ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 3551faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 35523f157a2fSChris Mason fail_cleaner: 3553a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 35547c2ca468SChris Mason 35557c2ca468SChris Mason /* 35567c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 35577c2ca468SChris Mason * kthreads 35587c2ca468SChris Mason */ 35597c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 35607c2ca468SChris Mason 35612365dd3cSAnand Jain fail_sysfs: 35626618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 35632365dd3cSAnand Jain 3564b7c35e81SAnand Jain fail_fsdev_sysfs: 3565b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3566b7c35e81SAnand Jain 35671b1d1f66SJosef Bacik fail_block_groups: 356854067ae9SJosef Bacik btrfs_put_block_group_cache(fs_info); 3569af31f5e5SChris Mason 3570af31f5e5SChris Mason fail_tree_roots: 35719e3aa805SJosef Bacik if (fs_info->data_reloc_root) 35729e3aa805SJosef Bacik btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root); 35734273eaffSAnand Jain free_root_pointers(fs_info, true); 35742b8195bbSMiao Xie invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3575af31f5e5SChris Mason 357639279cc3SChris Mason fail_sb_buffer: 35777abadb64SLiu Bo btrfs_stop_all_workers(fs_info); 35785cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 357916cdcec7SMiao Xie fail_alloc: 3580586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 3581586e46e2SIlya Dryomov 35824543df7eSChris Mason iput(fs_info->btree_inode); 35837e662854SQinghuang Feng fail: 3584586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 3585ad2b2c80SAl Viro return err; 3586eb60ceacSChris Mason } 3587663faf9fSMasami Hiramatsu ALLOW_ERROR_INJECTION(open_ctree, ERRNO); 3588eb60ceacSChris Mason 3589314b6dd0SJohannes Thumshirn static void btrfs_end_super_write(struct bio *bio) 3590f2984462SChris Mason { 3591314b6dd0SJohannes Thumshirn struct btrfs_device *device = bio->bi_private; 3592314b6dd0SJohannes Thumshirn struct bio_vec *bvec; 3593314b6dd0SJohannes Thumshirn struct bvec_iter_all iter_all; 3594314b6dd0SJohannes Thumshirn struct page *page; 3595442a4f63SStefan Behrens 3596314b6dd0SJohannes Thumshirn bio_for_each_segment_all(bvec, bio, iter_all) { 3597314b6dd0SJohannes Thumshirn page = bvec->bv_page; 3598314b6dd0SJohannes Thumshirn 3599314b6dd0SJohannes Thumshirn if (bio->bi_status) { 3600fb456252SJeff Mahoney btrfs_warn_rl_in_rcu(device->fs_info, 3601314b6dd0SJohannes Thumshirn "lost page write due to IO error on %s (%d)", 3602314b6dd0SJohannes Thumshirn rcu_str_deref(device->name), 3603314b6dd0SJohannes Thumshirn blk_status_to_errno(bio->bi_status)); 3604314b6dd0SJohannes Thumshirn ClearPageUptodate(page); 3605314b6dd0SJohannes Thumshirn SetPageError(page); 3606314b6dd0SJohannes Thumshirn btrfs_dev_stat_inc_and_print(device, 3607314b6dd0SJohannes Thumshirn BTRFS_DEV_STAT_WRITE_ERRS); 3608314b6dd0SJohannes Thumshirn } else { 3609314b6dd0SJohannes Thumshirn SetPageUptodate(page); 3610f2984462SChris Mason } 3611314b6dd0SJohannes Thumshirn 3612314b6dd0SJohannes Thumshirn put_page(page); 3613314b6dd0SJohannes Thumshirn unlock_page(page); 3614314b6dd0SJohannes Thumshirn } 3615314b6dd0SJohannes Thumshirn 3616314b6dd0SJohannes Thumshirn bio_put(bio); 3617f2984462SChris Mason } 3618f2984462SChris Mason 36198f32380dSJohannes Thumshirn struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev, 36208f32380dSJohannes Thumshirn int copy_num) 362129c36d72SAnand Jain { 362229c36d72SAnand Jain struct btrfs_super_block *super; 36238f32380dSJohannes Thumshirn struct page *page; 362412659251SNaohiro Aota u64 bytenr, bytenr_orig; 36258f32380dSJohannes Thumshirn struct address_space *mapping = bdev->bd_inode->i_mapping; 362612659251SNaohiro Aota int ret; 362729c36d72SAnand Jain 362812659251SNaohiro Aota bytenr_orig = btrfs_sb_offset(copy_num); 362912659251SNaohiro Aota ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr); 363012659251SNaohiro Aota if (ret == -ENOENT) 363112659251SNaohiro Aota return ERR_PTR(-EINVAL); 363212659251SNaohiro Aota else if (ret) 363312659251SNaohiro Aota return ERR_PTR(ret); 363412659251SNaohiro Aota 363529c36d72SAnand Jain if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode)) 36368f32380dSJohannes Thumshirn return ERR_PTR(-EINVAL); 363729c36d72SAnand Jain 36388f32380dSJohannes Thumshirn page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS); 36398f32380dSJohannes Thumshirn if (IS_ERR(page)) 36408f32380dSJohannes Thumshirn return ERR_CAST(page); 364129c36d72SAnand Jain 36428f32380dSJohannes Thumshirn super = page_address(page); 364396c2e067SAnand Jain if (btrfs_super_magic(super) != BTRFS_MAGIC) { 364496c2e067SAnand Jain btrfs_release_disk_super(super); 364596c2e067SAnand Jain return ERR_PTR(-ENODATA); 364696c2e067SAnand Jain } 364796c2e067SAnand Jain 364812659251SNaohiro Aota if (btrfs_super_bytenr(super) != bytenr_orig) { 36498f32380dSJohannes Thumshirn btrfs_release_disk_super(super); 36508f32380dSJohannes Thumshirn return ERR_PTR(-EINVAL); 365129c36d72SAnand Jain } 365229c36d72SAnand Jain 36538f32380dSJohannes Thumshirn return super; 365429c36d72SAnand Jain } 365529c36d72SAnand Jain 365629c36d72SAnand Jain 36578f32380dSJohannes Thumshirn struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev) 3658a512bbf8SYan Zheng { 36598f32380dSJohannes Thumshirn struct btrfs_super_block *super, *latest = NULL; 3660a512bbf8SYan Zheng int i; 3661a512bbf8SYan Zheng u64 transid = 0; 3662a512bbf8SYan Zheng 3663a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 3664a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 3665a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 3666a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3667a512bbf8SYan Zheng */ 3668a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 36698f32380dSJohannes Thumshirn super = btrfs_read_dev_one_super(bdev, i); 36708f32380dSJohannes Thumshirn if (IS_ERR(super)) 3671a512bbf8SYan Zheng continue; 3672a512bbf8SYan Zheng 3673a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 36748f32380dSJohannes Thumshirn if (latest) 36758f32380dSJohannes Thumshirn btrfs_release_disk_super(super); 36768f32380dSJohannes Thumshirn 36778f32380dSJohannes Thumshirn latest = super; 3678a512bbf8SYan Zheng transid = btrfs_super_generation(super); 3679a512bbf8SYan Zheng } 3680a512bbf8SYan Zheng } 368192fc03fbSAnand Jain 36828f32380dSJohannes Thumshirn return super; 3683a512bbf8SYan Zheng } 3684a512bbf8SYan Zheng 36854eedeb75SHisashi Hifumi /* 3686abbb3b8eSDavid Sterba * Write superblock @sb to the @device. Do not wait for completion, all the 3687314b6dd0SJohannes Thumshirn * pages we use for writing are locked. 36884eedeb75SHisashi Hifumi * 3689abbb3b8eSDavid Sterba * Write @max_mirrors copies of the superblock, where 0 means default that fit 3690abbb3b8eSDavid Sterba * the expected device size at commit time. Note that max_mirrors must be 3691abbb3b8eSDavid Sterba * same for write and wait phases. 36924eedeb75SHisashi Hifumi * 3693314b6dd0SJohannes Thumshirn * Return number of errors when page is not found or submission fails. 36944eedeb75SHisashi Hifumi */ 3695a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 3696abbb3b8eSDavid Sterba struct btrfs_super_block *sb, int max_mirrors) 3697a512bbf8SYan Zheng { 3698d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = device->fs_info; 3699314b6dd0SJohannes Thumshirn struct address_space *mapping = device->bdev->bd_inode->i_mapping; 3700d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 3701a512bbf8SYan Zheng int i; 3702a512bbf8SYan Zheng int errors = 0; 370312659251SNaohiro Aota int ret; 370412659251SNaohiro Aota u64 bytenr, bytenr_orig; 3705a512bbf8SYan Zheng 3706a512bbf8SYan Zheng if (max_mirrors == 0) 3707a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3708a512bbf8SYan Zheng 3709d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 3710d5178578SJohannes Thumshirn 3711a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 3712314b6dd0SJohannes Thumshirn struct page *page; 3713314b6dd0SJohannes Thumshirn struct bio *bio; 3714314b6dd0SJohannes Thumshirn struct btrfs_super_block *disk_super; 3715314b6dd0SJohannes Thumshirn 371612659251SNaohiro Aota bytenr_orig = btrfs_sb_offset(i); 371712659251SNaohiro Aota ret = btrfs_sb_log_location(device, i, WRITE, &bytenr); 371812659251SNaohiro Aota if (ret == -ENOENT) { 371912659251SNaohiro Aota continue; 372012659251SNaohiro Aota } else if (ret < 0) { 372112659251SNaohiro Aota btrfs_err(device->fs_info, 372212659251SNaohiro Aota "couldn't get super block location for mirror %d", 372312659251SNaohiro Aota i); 372412659251SNaohiro Aota errors++; 372512659251SNaohiro Aota continue; 372612659251SNaohiro Aota } 3727935e5cc9SMiao Xie if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3728935e5cc9SMiao Xie device->commit_total_bytes) 3729a512bbf8SYan Zheng break; 3730a512bbf8SYan Zheng 373112659251SNaohiro Aota btrfs_set_super_bytenr(sb, bytenr_orig); 3732a512bbf8SYan Zheng 3733fd08001fSEric Biggers crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE, 3734fd08001fSEric Biggers BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, 3735fd08001fSEric Biggers sb->csum); 3736a512bbf8SYan Zheng 3737314b6dd0SJohannes Thumshirn page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT, 3738314b6dd0SJohannes Thumshirn GFP_NOFS); 3739314b6dd0SJohannes Thumshirn if (!page) { 3740fb456252SJeff Mahoney btrfs_err(device->fs_info, 3741314b6dd0SJohannes Thumshirn "couldn't get super block page for bytenr %llu", 3742f14d104dSDavid Sterba bytenr); 3743634554dcSJosef Bacik errors++; 3744634554dcSJosef Bacik continue; 3745634554dcSJosef Bacik } 3746634554dcSJosef Bacik 3747314b6dd0SJohannes Thumshirn /* Bump the refcount for wait_dev_supers() */ 3748314b6dd0SJohannes Thumshirn get_page(page); 3749a512bbf8SYan Zheng 3750314b6dd0SJohannes Thumshirn disk_super = page_address(page); 3751314b6dd0SJohannes Thumshirn memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE); 3752a512bbf8SYan Zheng 3753387125fcSChris Mason /* 3754314b6dd0SJohannes Thumshirn * Directly use bios here instead of relying on the page cache 3755314b6dd0SJohannes Thumshirn * to do I/O, so we don't lose the ability to do integrity 3756314b6dd0SJohannes Thumshirn * checking. 3757387125fcSChris Mason */ 3758314b6dd0SJohannes Thumshirn bio = bio_alloc(GFP_NOFS, 1); 3759314b6dd0SJohannes Thumshirn bio_set_dev(bio, device->bdev); 3760314b6dd0SJohannes Thumshirn bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT; 3761314b6dd0SJohannes Thumshirn bio->bi_private = device; 3762314b6dd0SJohannes Thumshirn bio->bi_end_io = btrfs_end_super_write; 3763314b6dd0SJohannes Thumshirn __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE, 3764314b6dd0SJohannes Thumshirn offset_in_page(bytenr)); 3765314b6dd0SJohannes Thumshirn 3766314b6dd0SJohannes Thumshirn /* 3767314b6dd0SJohannes Thumshirn * We FUA only the first super block. The others we allow to 3768314b6dd0SJohannes Thumshirn * go down lazy and there's a short window where the on-disk 3769314b6dd0SJohannes Thumshirn * copies might still contain the older version. 3770314b6dd0SJohannes Thumshirn */ 3771314b6dd0SJohannes Thumshirn bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO; 37721b9e619cSOmar Sandoval if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) 3773314b6dd0SJohannes Thumshirn bio->bi_opf |= REQ_FUA; 3774314b6dd0SJohannes Thumshirn 3775314b6dd0SJohannes Thumshirn btrfsic_submit_bio(bio); 377612659251SNaohiro Aota btrfs_advance_sb_log(device, i); 3777a512bbf8SYan Zheng } 3778a512bbf8SYan Zheng return errors < i ? 0 : -1; 3779a512bbf8SYan Zheng } 3780a512bbf8SYan Zheng 3781387125fcSChris Mason /* 3782abbb3b8eSDavid Sterba * Wait for write completion of superblocks done by write_dev_supers, 3783abbb3b8eSDavid Sterba * @max_mirrors same for write and wait phases. 3784abbb3b8eSDavid Sterba * 3785314b6dd0SJohannes Thumshirn * Return number of errors when page is not found or not marked up to 3786abbb3b8eSDavid Sterba * date. 3787abbb3b8eSDavid Sterba */ 3788abbb3b8eSDavid Sterba static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) 3789abbb3b8eSDavid Sterba { 3790abbb3b8eSDavid Sterba int i; 3791abbb3b8eSDavid Sterba int errors = 0; 3792b6a535faSHoward McLauchlan bool primary_failed = false; 379312659251SNaohiro Aota int ret; 3794abbb3b8eSDavid Sterba u64 bytenr; 3795abbb3b8eSDavid Sterba 3796abbb3b8eSDavid Sterba if (max_mirrors == 0) 3797abbb3b8eSDavid Sterba max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3798abbb3b8eSDavid Sterba 3799abbb3b8eSDavid Sterba for (i = 0; i < max_mirrors; i++) { 3800314b6dd0SJohannes Thumshirn struct page *page; 3801314b6dd0SJohannes Thumshirn 380212659251SNaohiro Aota ret = btrfs_sb_log_location(device, i, READ, &bytenr); 380312659251SNaohiro Aota if (ret == -ENOENT) { 380412659251SNaohiro Aota break; 380512659251SNaohiro Aota } else if (ret < 0) { 380612659251SNaohiro Aota errors++; 380712659251SNaohiro Aota if (i == 0) 380812659251SNaohiro Aota primary_failed = true; 380912659251SNaohiro Aota continue; 381012659251SNaohiro Aota } 3811abbb3b8eSDavid Sterba if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3812abbb3b8eSDavid Sterba device->commit_total_bytes) 3813abbb3b8eSDavid Sterba break; 3814abbb3b8eSDavid Sterba 3815314b6dd0SJohannes Thumshirn page = find_get_page(device->bdev->bd_inode->i_mapping, 3816314b6dd0SJohannes Thumshirn bytenr >> PAGE_SHIFT); 3817314b6dd0SJohannes Thumshirn if (!page) { 3818abbb3b8eSDavid Sterba errors++; 3819b6a535faSHoward McLauchlan if (i == 0) 3820b6a535faSHoward McLauchlan primary_failed = true; 3821abbb3b8eSDavid Sterba continue; 3822abbb3b8eSDavid Sterba } 3823314b6dd0SJohannes Thumshirn /* Page is submitted locked and unlocked once the IO completes */ 3824314b6dd0SJohannes Thumshirn wait_on_page_locked(page); 3825314b6dd0SJohannes Thumshirn if (PageError(page)) { 3826abbb3b8eSDavid Sterba errors++; 3827b6a535faSHoward McLauchlan if (i == 0) 3828b6a535faSHoward McLauchlan primary_failed = true; 3829b6a535faSHoward McLauchlan } 3830abbb3b8eSDavid Sterba 3831314b6dd0SJohannes Thumshirn /* Drop our reference */ 3832314b6dd0SJohannes Thumshirn put_page(page); 3833abbb3b8eSDavid Sterba 3834314b6dd0SJohannes Thumshirn /* Drop the reference from the writing run */ 3835314b6dd0SJohannes Thumshirn put_page(page); 3836abbb3b8eSDavid Sterba } 3837abbb3b8eSDavid Sterba 3838b6a535faSHoward McLauchlan /* log error, force error return */ 3839b6a535faSHoward McLauchlan if (primary_failed) { 3840b6a535faSHoward McLauchlan btrfs_err(device->fs_info, "error writing primary super block to device %llu", 3841b6a535faSHoward McLauchlan device->devid); 3842b6a535faSHoward McLauchlan return -1; 3843b6a535faSHoward McLauchlan } 3844b6a535faSHoward McLauchlan 3845abbb3b8eSDavid Sterba return errors < i ? 0 : -1; 3846abbb3b8eSDavid Sterba } 3847abbb3b8eSDavid Sterba 3848abbb3b8eSDavid Sterba /* 3849387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 3850387125fcSChris Mason * for the barrier when it is done 3851387125fcSChris Mason */ 38524246a0b6SChristoph Hellwig static void btrfs_end_empty_barrier(struct bio *bio) 3853387125fcSChris Mason { 3854387125fcSChris Mason complete(bio->bi_private); 3855387125fcSChris Mason } 3856387125fcSChris Mason 3857387125fcSChris Mason /* 38584fc6441aSAnand Jain * Submit a flush request to the device if it supports it. Error handling is 38594fc6441aSAnand Jain * done in the waiting counterpart. 3860387125fcSChris Mason */ 38614fc6441aSAnand Jain static void write_dev_flush(struct btrfs_device *device) 3862387125fcSChris Mason { 3863c2a9c7abSAnand Jain struct request_queue *q = bdev_get_queue(device->bdev); 3864e0ae9994SDavid Sterba struct bio *bio = device->flush_bio; 3865387125fcSChris Mason 3866c2a9c7abSAnand Jain if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) 38674fc6441aSAnand Jain return; 3868387125fcSChris Mason 3869e0ae9994SDavid Sterba bio_reset(bio); 3870387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 387174d46992SChristoph Hellwig bio_set_dev(bio, device->bdev); 38728d910125SJan Kara bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH; 3873387125fcSChris Mason init_completion(&device->flush_wait); 3874387125fcSChris Mason bio->bi_private = &device->flush_wait; 3875387125fcSChris Mason 387643a01111SLu Fengqi btrfsic_submit_bio(bio); 38771c3063b6SAnand Jain set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 38784fc6441aSAnand Jain } 3879387125fcSChris Mason 38804fc6441aSAnand Jain /* 38814fc6441aSAnand Jain * If the flush bio has been submitted by write_dev_flush, wait for it. 38824fc6441aSAnand Jain */ 38838c27cb35SLinus Torvalds static blk_status_t wait_dev_flush(struct btrfs_device *device) 38844fc6441aSAnand Jain { 38854fc6441aSAnand Jain struct bio *bio = device->flush_bio; 38864fc6441aSAnand Jain 38871c3063b6SAnand Jain if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) 388858efbc9fSOmar Sandoval return BLK_STS_OK; 38894fc6441aSAnand Jain 38901c3063b6SAnand Jain clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 38912980d574SDavid Sterba wait_for_completion_io(&device->flush_wait); 38924fc6441aSAnand Jain 38938c27cb35SLinus Torvalds return bio->bi_status; 3894387125fcSChris Mason } 3895387125fcSChris Mason 3896d10b82feSQu Wenruo static int check_barrier_error(struct btrfs_fs_info *fs_info) 3897401b41e5SAnand Jain { 38986528b99dSAnand Jain if (!btrfs_check_rw_degradable(fs_info, NULL)) 3899401b41e5SAnand Jain return -EIO; 3900387125fcSChris Mason return 0; 3901387125fcSChris Mason } 3902387125fcSChris Mason 3903387125fcSChris Mason /* 3904387125fcSChris Mason * send an empty flush down to each device in parallel, 3905387125fcSChris Mason * then wait for them 3906387125fcSChris Mason */ 3907387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 3908387125fcSChris Mason { 3909387125fcSChris Mason struct list_head *head; 3910387125fcSChris Mason struct btrfs_device *dev; 39115af3e8ccSStefan Behrens int errors_wait = 0; 39124e4cbee9SChristoph Hellwig blk_status_t ret; 3913387125fcSChris Mason 39141538e6c5SDavid Sterba lockdep_assert_held(&info->fs_devices->device_list_mutex); 3915387125fcSChris Mason /* send down all the barriers */ 3916387125fcSChris Mason head = &info->fs_devices->devices; 39171538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3918e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3919f88ba6a2SHidetoshi Seto continue; 3920cea7c8bfSAnand Jain if (!dev->bdev) 3921387125fcSChris Mason continue; 3922e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3923ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3924387125fcSChris Mason continue; 3925387125fcSChris Mason 39264fc6441aSAnand Jain write_dev_flush(dev); 392758efbc9fSOmar Sandoval dev->last_flush_error = BLK_STS_OK; 3928387125fcSChris Mason } 3929387125fcSChris Mason 3930387125fcSChris Mason /* wait for all the barriers */ 39311538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 3932e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3933f88ba6a2SHidetoshi Seto continue; 3934387125fcSChris Mason if (!dev->bdev) { 39355af3e8ccSStefan Behrens errors_wait++; 3936387125fcSChris Mason continue; 3937387125fcSChris Mason } 3938e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3939ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3940387125fcSChris Mason continue; 3941387125fcSChris Mason 39424fc6441aSAnand Jain ret = wait_dev_flush(dev); 3943401b41e5SAnand Jain if (ret) { 3944401b41e5SAnand Jain dev->last_flush_error = ret; 394566b4993eSDavid Sterba btrfs_dev_stat_inc_and_print(dev, 394666b4993eSDavid Sterba BTRFS_DEV_STAT_FLUSH_ERRS); 39475af3e8ccSStefan Behrens errors_wait++; 3948387125fcSChris Mason } 3949401b41e5SAnand Jain } 3950401b41e5SAnand Jain 3951cea7c8bfSAnand Jain if (errors_wait) { 3952401b41e5SAnand Jain /* 3953401b41e5SAnand Jain * At some point we need the status of all disks 3954401b41e5SAnand Jain * to arrive at the volume status. So error checking 3955401b41e5SAnand Jain * is being pushed to a separate loop. 3956401b41e5SAnand Jain */ 3957d10b82feSQu Wenruo return check_barrier_error(info); 3958401b41e5SAnand Jain } 3959387125fcSChris Mason return 0; 3960387125fcSChris Mason } 3961387125fcSChris Mason 3962943c6e99SZhao Lei int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 3963943c6e99SZhao Lei { 39648789f4feSZhao Lei int raid_type; 39658789f4feSZhao Lei int min_tolerated = INT_MAX; 3966943c6e99SZhao Lei 39678789f4feSZhao Lei if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 39688789f4feSZhao Lei (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 39698c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 39708789f4feSZhao Lei btrfs_raid_array[BTRFS_RAID_SINGLE]. 39718789f4feSZhao Lei tolerated_failures); 3972943c6e99SZhao Lei 39738789f4feSZhao Lei for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 39748789f4feSZhao Lei if (raid_type == BTRFS_RAID_SINGLE) 39758789f4feSZhao Lei continue; 397641a6e891SAnand Jain if (!(flags & btrfs_raid_array[raid_type].bg_flag)) 39778789f4feSZhao Lei continue; 39788c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 39798789f4feSZhao Lei btrfs_raid_array[raid_type]. 39808789f4feSZhao Lei tolerated_failures); 39818789f4feSZhao Lei } 3982943c6e99SZhao Lei 39838789f4feSZhao Lei if (min_tolerated == INT_MAX) { 3984ab8d0fc4SJeff Mahoney pr_warn("BTRFS: unknown raid flag: %llu", flags); 39858789f4feSZhao Lei min_tolerated = 0; 39868789f4feSZhao Lei } 39878789f4feSZhao Lei 39888789f4feSZhao Lei return min_tolerated; 3989943c6e99SZhao Lei } 3990943c6e99SZhao Lei 3991eece6a9cSDavid Sterba int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) 3992f2984462SChris Mason { 3993e5e9a520SChris Mason struct list_head *head; 3994f2984462SChris Mason struct btrfs_device *dev; 3995a061fc8dSChris Mason struct btrfs_super_block *sb; 3996f2984462SChris Mason struct btrfs_dev_item *dev_item; 3997f2984462SChris Mason int ret; 3998f2984462SChris Mason int do_barriers; 3999a236aed1SChris Mason int max_errors; 4000a236aed1SChris Mason int total_errors = 0; 4001a061fc8dSChris Mason u64 flags; 4002f2984462SChris Mason 40030b246afaSJeff Mahoney do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); 4004fed3b381SLiu Bo 4005fed3b381SLiu Bo /* 4006fed3b381SLiu Bo * max_mirrors == 0 indicates we're from commit_transaction, 4007fed3b381SLiu Bo * not from fsync where the tree roots in fs_info have not 4008fed3b381SLiu Bo * been consistent on disk. 4009fed3b381SLiu Bo */ 4010fed3b381SLiu Bo if (max_mirrors == 0) 40110b246afaSJeff Mahoney backup_super_roots(fs_info); 4012f2984462SChris Mason 40130b246afaSJeff Mahoney sb = fs_info->super_for_commit; 4014a061fc8dSChris Mason dev_item = &sb->dev_item; 4015e5e9a520SChris Mason 40160b246afaSJeff Mahoney mutex_lock(&fs_info->fs_devices->device_list_mutex); 40170b246afaSJeff Mahoney head = &fs_info->fs_devices->devices; 40180b246afaSJeff Mahoney max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; 4019387125fcSChris Mason 40205af3e8ccSStefan Behrens if (do_barriers) { 40210b246afaSJeff Mahoney ret = barrier_all_devices(fs_info); 40225af3e8ccSStefan Behrens if (ret) { 40235af3e8ccSStefan Behrens mutex_unlock( 40240b246afaSJeff Mahoney &fs_info->fs_devices->device_list_mutex); 40250b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 40265af3e8ccSStefan Behrens "errors while submitting device barriers."); 40275af3e8ccSStefan Behrens return ret; 40285af3e8ccSStefan Behrens } 40295af3e8ccSStefan Behrens } 4030387125fcSChris Mason 40311538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4032dfe25020SChris Mason if (!dev->bdev) { 4033dfe25020SChris Mason total_errors++; 4034dfe25020SChris Mason continue; 4035dfe25020SChris Mason } 4036e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4037ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4038dfe25020SChris Mason continue; 4039dfe25020SChris Mason 40402b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 4041a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 4042a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 40437df69d3eSMiao Xie btrfs_set_stack_device_total_bytes(dev_item, 4044935e5cc9SMiao Xie dev->commit_total_bytes); 4045ce7213c7SMiao Xie btrfs_set_stack_device_bytes_used(dev_item, 4046ce7213c7SMiao Xie dev->commit_bytes_used); 4047a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 4048a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 4049a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 4050a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 40517239ff4bSNikolay Borisov memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, 40527239ff4bSNikolay Borisov BTRFS_FSID_SIZE); 4053a512bbf8SYan Zheng 4054a061fc8dSChris Mason flags = btrfs_super_flags(sb); 4055a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 4056f2984462SChris Mason 405775cb857dSQu Wenruo ret = btrfs_validate_write_super(fs_info, sb); 405875cb857dSQu Wenruo if (ret < 0) { 405975cb857dSQu Wenruo mutex_unlock(&fs_info->fs_devices->device_list_mutex); 406075cb857dSQu Wenruo btrfs_handle_fs_error(fs_info, -EUCLEAN, 406175cb857dSQu Wenruo "unexpected superblock corruption detected"); 406275cb857dSQu Wenruo return -EUCLEAN; 406375cb857dSQu Wenruo } 406475cb857dSQu Wenruo 4065abbb3b8eSDavid Sterba ret = write_dev_supers(dev, sb, max_mirrors); 4066a236aed1SChris Mason if (ret) 4067a236aed1SChris Mason total_errors++; 4068f2984462SChris Mason } 4069a236aed1SChris Mason if (total_errors > max_errors) { 40700b246afaSJeff Mahoney btrfs_err(fs_info, "%d errors while writing supers", 4071d397712bSChris Mason total_errors); 40720b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 407379787eaaSJeff Mahoney 40749d565ba4SStefan Behrens /* FUA is masked off if unsupported and can't be the reason */ 40750b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 40760b246afaSJeff Mahoney "%d errors while writing supers", 40770b246afaSJeff Mahoney total_errors); 40789d565ba4SStefan Behrens return -EIO; 4079a236aed1SChris Mason } 4080f2984462SChris Mason 4081a512bbf8SYan Zheng total_errors = 0; 40821538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4083dfe25020SChris Mason if (!dev->bdev) 4084dfe25020SChris Mason continue; 4085e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4086ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4087dfe25020SChris Mason continue; 4088dfe25020SChris Mason 4089abbb3b8eSDavid Sterba ret = wait_dev_supers(dev, max_mirrors); 4090a512bbf8SYan Zheng if (ret) 40911259ab75SChris Mason total_errors++; 4092f2984462SChris Mason } 40930b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 4094a236aed1SChris Mason if (total_errors > max_errors) { 40950b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 40960b246afaSJeff Mahoney "%d errors while writing supers", 40970b246afaSJeff Mahoney total_errors); 409879787eaaSJeff Mahoney return -EIO; 4099a236aed1SChris Mason } 4100f2984462SChris Mason return 0; 4101f2984462SChris Mason } 4102f2984462SChris Mason 4103cb517eabSMiao Xie /* Drop a fs root from the radix tree and free it. */ 4104cb517eabSMiao Xie void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 4105cb517eabSMiao Xie struct btrfs_root *root) 41062619ba1fSChris Mason { 41074785e24fSJosef Bacik bool drop_ref = false; 41084785e24fSJosef Bacik 41094df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 41102619ba1fSChris Mason radix_tree_delete(&fs_info->fs_roots_radix, 41112619ba1fSChris Mason (unsigned long)root->root_key.objectid); 4112af01d2e5SJosef Bacik if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state)) 41134785e24fSJosef Bacik drop_ref = true; 41144df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 411576dda93cSYan, Zheng 41161c1ea4f7SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 4117ef67963dSJosef Bacik ASSERT(root->log_root == NULL); 41181c1ea4f7SLiu Bo if (root->reloc_root) { 411900246528SJosef Bacik btrfs_put_root(root->reloc_root); 41201c1ea4f7SLiu Bo root->reloc_root = NULL; 41211c1ea4f7SLiu Bo } 41221c1ea4f7SLiu Bo } 41233321719eSLiu Bo 41244785e24fSJosef Bacik if (drop_ref) 412500246528SJosef Bacik btrfs_put_root(root); 41262619ba1fSChris Mason } 41272619ba1fSChris Mason 4128c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 4129c146afadSYan Zheng { 4130c146afadSYan Zheng u64 root_objectid = 0; 4131c146afadSYan Zheng struct btrfs_root *gang[8]; 413265d33fd7SQu Wenruo int i = 0; 413365d33fd7SQu Wenruo int err = 0; 413465d33fd7SQu Wenruo unsigned int ret = 0; 4135c146afadSYan Zheng 4136c146afadSYan Zheng while (1) { 4137efc34534SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 4138c146afadSYan Zheng ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4139c146afadSYan Zheng (void **)gang, root_objectid, 4140c146afadSYan Zheng ARRAY_SIZE(gang)); 414165d33fd7SQu Wenruo if (!ret) { 4142efc34534SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 4143c146afadSYan Zheng break; 414465d33fd7SQu Wenruo } 41455d4f98a2SYan Zheng root_objectid = gang[ret - 1]->root_key.objectid + 1; 414666b4ffd1SJosef Bacik 414765d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 414865d33fd7SQu Wenruo /* Avoid to grab roots in dead_roots */ 414965d33fd7SQu Wenruo if (btrfs_root_refs(&gang[i]->root_item) == 0) { 415065d33fd7SQu Wenruo gang[i] = NULL; 415165d33fd7SQu Wenruo continue; 415265d33fd7SQu Wenruo } 415365d33fd7SQu Wenruo /* grab all the search result for later use */ 415400246528SJosef Bacik gang[i] = btrfs_grab_root(gang[i]); 415565d33fd7SQu Wenruo } 4156efc34534SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 415765d33fd7SQu Wenruo 415865d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 415965d33fd7SQu Wenruo if (!gang[i]) 416065d33fd7SQu Wenruo continue; 4161c146afadSYan Zheng root_objectid = gang[i]->root_key.objectid; 416266b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(gang[i]); 416366b4ffd1SJosef Bacik if (err) 416465d33fd7SQu Wenruo break; 416500246528SJosef Bacik btrfs_put_root(gang[i]); 4166c146afadSYan Zheng } 4167c146afadSYan Zheng root_objectid++; 4168c146afadSYan Zheng } 416965d33fd7SQu Wenruo 417065d33fd7SQu Wenruo /* release the uncleaned roots due to error */ 417165d33fd7SQu Wenruo for (; i < ret; i++) { 417265d33fd7SQu Wenruo if (gang[i]) 417300246528SJosef Bacik btrfs_put_root(gang[i]); 417465d33fd7SQu Wenruo } 417565d33fd7SQu Wenruo return err; 4176c146afadSYan Zheng } 4177c146afadSYan Zheng 41786bccf3abSJeff Mahoney int btrfs_commit_super(struct btrfs_fs_info *fs_info) 4179c146afadSYan Zheng { 41806bccf3abSJeff Mahoney struct btrfs_root *root = fs_info->tree_root; 4181c146afadSYan Zheng struct btrfs_trans_handle *trans; 4182c146afadSYan Zheng 41830b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 41842ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 41850b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 41860b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 4187c71bf099SYan, Zheng 4188c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 41890b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 41900b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4191c71bf099SYan, Zheng 41927a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 41933612b495STsutomu Itoh if (IS_ERR(trans)) 41943612b495STsutomu Itoh return PTR_ERR(trans); 41953a45bb20SJeff Mahoney return btrfs_commit_transaction(trans); 4196c146afadSYan Zheng } 4197c146afadSYan Zheng 4198b105e927SDavid Sterba void __cold close_ctree(struct btrfs_fs_info *fs_info) 4199eb60ceacSChris Mason { 4200c146afadSYan Zheng int ret; 4201e089f05cSChris Mason 4202afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); 4203d6fd0ae2SOmar Sandoval /* 4204d6fd0ae2SOmar Sandoval * We don't want the cleaner to start new transactions, add more delayed 4205d6fd0ae2SOmar Sandoval * iputs, etc. while we're closing. We can't use kthread_stop() yet 4206d6fd0ae2SOmar Sandoval * because that frees the task_struct, and the transaction kthread might 4207d6fd0ae2SOmar Sandoval * still try to wake up the cleaner. 4208d6fd0ae2SOmar Sandoval */ 4209d6fd0ae2SOmar Sandoval kthread_park(fs_info->cleaner_kthread); 4210a2135011SChris Mason 42117343dd61SJustin Maggard /* wait for the qgroup rescan worker to stop */ 4212d06f23d6SJeff Mahoney btrfs_qgroup_wait_for_completion(fs_info, false); 42137343dd61SJustin Maggard 4214803b2f54SStefan Behrens /* wait for the uuid_scan task to finish */ 4215803b2f54SStefan Behrens down(&fs_info->uuid_tree_rescan_sem); 4216803b2f54SStefan Behrens /* avoid complains from lockdep et al., set sem back to initial state */ 4217803b2f54SStefan Behrens up(&fs_info->uuid_tree_rescan_sem); 4218803b2f54SStefan Behrens 4219837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 4220aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 4221837d5b6eSIlya Dryomov 42228dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 42238dabb742SStefan Behrens 4224aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 42254cb5300bSChris Mason 42264cb5300bSChris Mason /* wait for any defraggers to finish */ 42274cb5300bSChris Mason wait_event(fs_info->transaction_wait, 42284cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 42294cb5300bSChris Mason 42304cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 423126176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 42324cb5300bSChris Mason 423321c7e756SMiao Xie cancel_work_sync(&fs_info->async_reclaim_work); 423457056740SJosef Bacik cancel_work_sync(&fs_info->async_data_reclaim_work); 4235576fa348SJosef Bacik cancel_work_sync(&fs_info->preempt_reclaim_work); 423621c7e756SMiao Xie 4237b0643e59SDennis Zhou /* Cancel or finish ongoing discard work */ 4238b0643e59SDennis Zhou btrfs_discard_cleanup(fs_info); 4239b0643e59SDennis Zhou 4240bc98a42cSDavid Howells if (!sb_rdonly(fs_info->sb)) { 4241e44163e1SJeff Mahoney /* 4242d6fd0ae2SOmar Sandoval * The cleaner kthread is stopped, so do one final pass over 4243d6fd0ae2SOmar Sandoval * unused block groups. 4244e44163e1SJeff Mahoney */ 42450b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 4246e44163e1SJeff Mahoney 4247f0cc2cd7SFilipe Manana /* 4248f0cc2cd7SFilipe Manana * There might be existing delayed inode workers still running 4249f0cc2cd7SFilipe Manana * and holding an empty delayed inode item. We must wait for 4250f0cc2cd7SFilipe Manana * them to complete first because they can create a transaction. 4251f0cc2cd7SFilipe Manana * This happens when someone calls btrfs_balance_delayed_items() 4252f0cc2cd7SFilipe Manana * and then a transaction commit runs the same delayed nodes 4253f0cc2cd7SFilipe Manana * before any delayed worker has done something with the nodes. 4254f0cc2cd7SFilipe Manana * We must wait for any worker here and not at transaction 4255f0cc2cd7SFilipe Manana * commit time since that could cause a deadlock. 4256f0cc2cd7SFilipe Manana * This is a very rare case. 4257f0cc2cd7SFilipe Manana */ 4258f0cc2cd7SFilipe Manana btrfs_flush_workqueue(fs_info->delayed_workers); 4259f0cc2cd7SFilipe Manana 42606bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 4261d397712bSChris Mason if (ret) 426204892340SEric Sandeen btrfs_err(fs_info, "commit super ret %d", ret); 4263c146afadSYan Zheng } 4264ed2ff2cbSChris Mason 4265af722733SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) || 4266af722733SLiu Bo test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) 42672ff7e61eSJeff Mahoney btrfs_error_commit_super(fs_info); 4268acce952bSliubo 4269e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 4270e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 42718929ecfaSYan, Zheng 4272e187831eSJosef Bacik ASSERT(list_empty(&fs_info->delayed_iputs)); 4273afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); 4274f25784b3SYan Zheng 42755958253cSQu Wenruo if (btrfs_check_quota_leak(fs_info)) { 42765958253cSQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 42775958253cSQu Wenruo btrfs_err(fs_info, "qgroup reserved space leaked"); 42785958253cSQu Wenruo } 42795958253cSQu Wenruo 428004892340SEric Sandeen btrfs_free_qgroup_config(fs_info); 4281fe816d0fSNikolay Borisov ASSERT(list_empty(&fs_info->delalloc_roots)); 4282bcef60f2SArne Jansen 4283963d678bSMiao Xie if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 428404892340SEric Sandeen btrfs_info(fs_info, "at unmount delalloc count %lld", 4285963d678bSMiao Xie percpu_counter_sum(&fs_info->delalloc_bytes)); 4286b0c68f8bSChris Mason } 428731153d81SYan Zheng 42885deb17e1SJosef Bacik if (percpu_counter_sum(&fs_info->ordered_bytes)) 42894297ff84SJosef Bacik btrfs_info(fs_info, "at unmount dio bytes count %lld", 42905deb17e1SJosef Bacik percpu_counter_sum(&fs_info->ordered_bytes)); 42914297ff84SJosef Bacik 42926618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 4293b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 42945ac1d209SJeff Mahoney 42951a4319ccSLiu Bo btrfs_put_block_group_cache(fs_info); 42961a4319ccSLiu Bo 4297de348ee0SWang Shilong /* 4298de348ee0SWang Shilong * we must make sure there is not any read request to 4299de348ee0SWang Shilong * submit after we stopping all workers. 4300de348ee0SWang Shilong */ 4301de348ee0SWang Shilong invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 430296192499SJosef Bacik btrfs_stop_all_workers(fs_info); 430396192499SJosef Bacik 43040a31daa4SFilipe Manana /* We shouldn't have any transaction open at this point */ 43050a31daa4SFilipe Manana ASSERT(list_empty(&fs_info->trans_list)); 43060a31daa4SFilipe Manana 4307afcdd129SJosef Bacik clear_bit(BTRFS_FS_OPEN, &fs_info->flags); 43084273eaffSAnand Jain free_root_pointers(fs_info, true); 43098c38938cSJosef Bacik btrfs_free_fs_roots(fs_info); 43109ad6b7bcSChris Mason 43114e19443dSJosef Bacik /* 43124e19443dSJosef Bacik * We must free the block groups after dropping the fs_roots as we could 43134e19443dSJosef Bacik * have had an IO error and have left over tree log blocks that aren't 43144e19443dSJosef Bacik * cleaned up until the fs roots are freed. This makes the block group 43154e19443dSJosef Bacik * accounting appear to be wrong because there's pending reserved bytes, 43164e19443dSJosef Bacik * so make sure we do the block group cleanup afterwards. 43174e19443dSJosef Bacik */ 43184e19443dSJosef Bacik btrfs_free_block_groups(fs_info); 43194e19443dSJosef Bacik 432013e6c37bSJosef Bacik iput(fs_info->btree_inode); 4321d6bfde87SChris Mason 432221adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 43230b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) 43242ff7e61eSJeff Mahoney btrfsic_unmount(fs_info->fs_devices); 432521adbd5cSStefan Behrens #endif 432621adbd5cSStefan Behrens 43270b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 432868c94e55SNikolay Borisov btrfs_close_devices(fs_info->fs_devices); 4329eb60ceacSChris Mason } 4330eb60ceacSChris Mason 4331b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 4332b9fab919SChris Mason int atomic) 4333ccd467d6SChris Mason { 43341259ab75SChris Mason int ret; 4335727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 43361259ab75SChris Mason 43370b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 43381259ab75SChris Mason if (!ret) 43391259ab75SChris Mason return ret; 43401259ab75SChris Mason 43411259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 4342b9fab919SChris Mason parent_transid, atomic); 4343b9fab919SChris Mason if (ret == -EAGAIN) 4344b9fab919SChris Mason return ret; 43451259ab75SChris Mason return !ret; 43465f39d397SChris Mason } 43476702ed49SChris Mason 43485f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 43495f39d397SChris Mason { 43502f4d60dfSQu Wenruo struct btrfs_fs_info *fs_info = buf->fs_info; 43515f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 4352b9473439SChris Mason int was_dirty; 4353b4ce94deSChris Mason 435406ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 435506ea65a3SJosef Bacik /* 435606ea65a3SJosef Bacik * This is a fast path so only do this check if we have sanity tests 435752042d8eSAndrea Gelmini * enabled. Normal people shouldn't be using unmapped buffers as dirty 435806ea65a3SJosef Bacik * outside of the sanity tests. 435906ea65a3SJosef Bacik */ 4360b0132a3bSNikolay Borisov if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) 436106ea65a3SJosef Bacik return; 436206ea65a3SJosef Bacik #endif 4363b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 43640b246afaSJeff Mahoney if (transid != fs_info->generation) 43655d163e0eSJeff Mahoney WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", 43660b246afaSJeff Mahoney buf->start, transid, fs_info->generation); 43670b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 4368e2d84521SMiao Xie if (!was_dirty) 4369104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 4370e2d84521SMiao Xie buf->len, 43710b246afaSJeff Mahoney fs_info->dirty_metadata_batch); 43721f21ef0aSFilipe Manana #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 437369fc6cbbSQu Wenruo /* 437469fc6cbbSQu Wenruo * Since btrfs_mark_buffer_dirty() can be called with item pointer set 437569fc6cbbSQu Wenruo * but item data not updated. 437669fc6cbbSQu Wenruo * So here we should only check item pointers, not item data. 437769fc6cbbSQu Wenruo */ 437869fc6cbbSQu Wenruo if (btrfs_header_level(buf) == 0 && 4379cfdaad5eSDavid Sterba btrfs_check_leaf_relaxed(buf)) { 4380a4f78750SDavid Sterba btrfs_print_leaf(buf); 43811f21ef0aSFilipe Manana ASSERT(0); 43821f21ef0aSFilipe Manana } 43831f21ef0aSFilipe Manana #endif 4384eb60ceacSChris Mason } 4385eb60ceacSChris Mason 43862ff7e61eSJeff Mahoney static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, 4387b53d3f5dSLiu Bo int flush_delayed) 438835b7e476SChris Mason { 4389188de649SChris Mason /* 4390188de649SChris Mason * looks as though older kernels can get into trouble with 4391188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 4392188de649SChris Mason */ 4393e2d84521SMiao Xie int ret; 4394d6bfde87SChris Mason 43956933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 4396d6bfde87SChris Mason return; 4397d6bfde87SChris Mason 4398b53d3f5dSLiu Bo if (flush_delayed) 43992ff7e61eSJeff Mahoney btrfs_balance_delayed_items(fs_info); 440016cdcec7SMiao Xie 4401d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 4402d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 4403d814a491SEthan Lien fs_info->dirty_metadata_batch); 4404e2d84521SMiao Xie if (ret > 0) { 44050b246afaSJeff Mahoney balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); 440616cdcec7SMiao Xie } 440716cdcec7SMiao Xie } 440816cdcec7SMiao Xie 44092ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) 441016cdcec7SMiao Xie { 44112ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 1); 441235b7e476SChris Mason } 4413b53d3f5dSLiu Bo 44142ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) 4415b53d3f5dSLiu Bo { 44162ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 0); 4417d6bfde87SChris Mason } 44186b80053dSChris Mason 4419581c1760SQu Wenruo int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level, 4420581c1760SQu Wenruo struct btrfs_key *first_key) 44216b80053dSChris Mason { 44225ab12d1fSDavid Sterba return btree_read_extent_buffer_pages(buf, parent_transid, 4423581c1760SQu Wenruo level, first_key); 44246b80053dSChris Mason } 44250da5468fSChris Mason 44262ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) 4427acce952bSliubo { 4428fe816d0fSNikolay Borisov /* cleanup FS via transaction */ 4429fe816d0fSNikolay Borisov btrfs_cleanup_transaction(fs_info); 4430fe816d0fSNikolay Borisov 44310b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 44322ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 44330b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 4434acce952bSliubo 44350b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 44360b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4437acce952bSliubo } 4438acce952bSliubo 4439ef67963dSJosef Bacik static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info) 4440ef67963dSJosef Bacik { 4441ef67963dSJosef Bacik struct btrfs_root *gang[8]; 4442ef67963dSJosef Bacik u64 root_objectid = 0; 4443ef67963dSJosef Bacik int ret; 4444ef67963dSJosef Bacik 4445ef67963dSJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 4446ef67963dSJosef Bacik while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4447ef67963dSJosef Bacik (void **)gang, root_objectid, 4448ef67963dSJosef Bacik ARRAY_SIZE(gang))) != 0) { 4449ef67963dSJosef Bacik int i; 4450ef67963dSJosef Bacik 4451ef67963dSJosef Bacik for (i = 0; i < ret; i++) 4452ef67963dSJosef Bacik gang[i] = btrfs_grab_root(gang[i]); 4453ef67963dSJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 4454ef67963dSJosef Bacik 4455ef67963dSJosef Bacik for (i = 0; i < ret; i++) { 4456ef67963dSJosef Bacik if (!gang[i]) 4457ef67963dSJosef Bacik continue; 4458ef67963dSJosef Bacik root_objectid = gang[i]->root_key.objectid; 4459ef67963dSJosef Bacik btrfs_free_log(NULL, gang[i]); 4460ef67963dSJosef Bacik btrfs_put_root(gang[i]); 4461ef67963dSJosef Bacik } 4462ef67963dSJosef Bacik root_objectid++; 4463ef67963dSJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 4464ef67963dSJosef Bacik } 4465ef67963dSJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 4466ef67963dSJosef Bacik btrfs_free_log_root_tree(NULL, fs_info); 4467ef67963dSJosef Bacik } 4468ef67963dSJosef Bacik 4469143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4470acce952bSliubo { 4471acce952bSliubo struct btrfs_ordered_extent *ordered; 4472acce952bSliubo 4473199c2a9cSMiao Xie spin_lock(&root->ordered_extent_lock); 4474779880efSJosef Bacik /* 4475779880efSJosef Bacik * This will just short circuit the ordered completion stuff which will 4476779880efSJosef Bacik * make sure the ordered extent gets properly cleaned up. 4477779880efSJosef Bacik */ 4478199c2a9cSMiao Xie list_for_each_entry(ordered, &root->ordered_extents, 4479779880efSJosef Bacik root_extent_list) 4480779880efSJosef Bacik set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4481199c2a9cSMiao Xie spin_unlock(&root->ordered_extent_lock); 4482199c2a9cSMiao Xie } 4483199c2a9cSMiao Xie 4484199c2a9cSMiao Xie static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4485199c2a9cSMiao Xie { 4486199c2a9cSMiao Xie struct btrfs_root *root; 4487199c2a9cSMiao Xie struct list_head splice; 4488199c2a9cSMiao Xie 4489199c2a9cSMiao Xie INIT_LIST_HEAD(&splice); 4490199c2a9cSMiao Xie 4491199c2a9cSMiao Xie spin_lock(&fs_info->ordered_root_lock); 4492199c2a9cSMiao Xie list_splice_init(&fs_info->ordered_roots, &splice); 4493199c2a9cSMiao Xie while (!list_empty(&splice)) { 4494199c2a9cSMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4495199c2a9cSMiao Xie ordered_root); 44961de2cfdeSJosef Bacik list_move_tail(&root->ordered_root, 44971de2cfdeSJosef Bacik &fs_info->ordered_roots); 4498199c2a9cSMiao Xie 44992a85d9caSLiu Bo spin_unlock(&fs_info->ordered_root_lock); 4500199c2a9cSMiao Xie btrfs_destroy_ordered_extents(root); 4501199c2a9cSMiao Xie 45022a85d9caSLiu Bo cond_resched(); 45032a85d9caSLiu Bo spin_lock(&fs_info->ordered_root_lock); 4504199c2a9cSMiao Xie } 4505199c2a9cSMiao Xie spin_unlock(&fs_info->ordered_root_lock); 450674d5d229SJosef Bacik 450774d5d229SJosef Bacik /* 450874d5d229SJosef Bacik * We need this here because if we've been flipped read-only we won't 450974d5d229SJosef Bacik * get sync() from the umount, so we need to make sure any ordered 451074d5d229SJosef Bacik * extents that haven't had their dirty pages IO start writeout yet 451174d5d229SJosef Bacik * actually get run and error out properly. 451274d5d229SJosef Bacik */ 451374d5d229SJosef Bacik btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 4514acce952bSliubo } 4515acce952bSliubo 451635a3621bSStefan Behrens static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 45172ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4518acce952bSliubo { 4519acce952bSliubo struct rb_node *node; 4520acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 4521acce952bSliubo struct btrfs_delayed_ref_node *ref; 4522acce952bSliubo int ret = 0; 4523acce952bSliubo 4524acce952bSliubo delayed_refs = &trans->delayed_refs; 4525acce952bSliubo 4526acce952bSliubo spin_lock(&delayed_refs->lock); 4527d7df2c79SJosef Bacik if (atomic_read(&delayed_refs->num_entries) == 0) { 4528cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 4529b79ce3ddSDavid Sterba btrfs_debug(fs_info, "delayed_refs has NO entry"); 4530acce952bSliubo return ret; 4531acce952bSliubo } 4532acce952bSliubo 45335c9d028bSLiu Bo while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { 4534d7df2c79SJosef Bacik struct btrfs_delayed_ref_head *head; 45350e0adbcfSJosef Bacik struct rb_node *n; 4536e78417d1SJosef Bacik bool pin_bytes = false; 4537acce952bSliubo 4538d7df2c79SJosef Bacik head = rb_entry(node, struct btrfs_delayed_ref_head, 4539d7df2c79SJosef Bacik href_node); 45403069bd26SJosef Bacik if (btrfs_delayed_ref_lock(delayed_refs, head)) 4541b939d1abSJosef Bacik continue; 45423069bd26SJosef Bacik 4543d7df2c79SJosef Bacik spin_lock(&head->lock); 4544e3d03965SLiu Bo while ((n = rb_first_cached(&head->ref_tree)) != NULL) { 45450e0adbcfSJosef Bacik ref = rb_entry(n, struct btrfs_delayed_ref_node, 45460e0adbcfSJosef Bacik ref_node); 4547d7df2c79SJosef Bacik ref->in_tree = 0; 4548e3d03965SLiu Bo rb_erase_cached(&ref->ref_node, &head->ref_tree); 45490e0adbcfSJosef Bacik RB_CLEAR_NODE(&ref->ref_node); 45501d57ee94SWang Xiaoguang if (!list_empty(&ref->add_list)) 45511d57ee94SWang Xiaoguang list_del(&ref->add_list); 4552d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4553d7df2c79SJosef Bacik btrfs_put_delayed_ref(ref); 4554d7df2c79SJosef Bacik } 455554067ae9SJosef Bacik if (head->must_insert_reserved) 4556e78417d1SJosef Bacik pin_bytes = true; 455778a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 4558fa781ceaSJosef Bacik btrfs_delete_ref_head(delayed_refs, head); 4559d7df2c79SJosef Bacik spin_unlock(&head->lock); 4560acce952bSliubo spin_unlock(&delayed_refs->lock); 4561e78417d1SJosef Bacik mutex_unlock(&head->mutex); 4562acce952bSliubo 4563f603bb94SNikolay Borisov if (pin_bytes) { 4564f603bb94SNikolay Borisov struct btrfs_block_group *cache; 4565f603bb94SNikolay Borisov 4566f603bb94SNikolay Borisov cache = btrfs_lookup_block_group(fs_info, head->bytenr); 4567f603bb94SNikolay Borisov BUG_ON(!cache); 4568f603bb94SNikolay Borisov 4569f603bb94SNikolay Borisov spin_lock(&cache->space_info->lock); 4570f603bb94SNikolay Borisov spin_lock(&cache->lock); 4571f603bb94SNikolay Borisov cache->pinned += head->num_bytes; 4572f603bb94SNikolay Borisov btrfs_space_info_update_bytes_pinned(fs_info, 4573f603bb94SNikolay Borisov cache->space_info, head->num_bytes); 4574f603bb94SNikolay Borisov cache->reserved -= head->num_bytes; 4575f603bb94SNikolay Borisov cache->space_info->bytes_reserved -= head->num_bytes; 4576f603bb94SNikolay Borisov spin_unlock(&cache->lock); 4577f603bb94SNikolay Borisov spin_unlock(&cache->space_info->lock); 4578f603bb94SNikolay Borisov percpu_counter_add_batch( 4579f603bb94SNikolay Borisov &cache->space_info->total_bytes_pinned, 4580f603bb94SNikolay Borisov head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH); 4581f603bb94SNikolay Borisov 4582f603bb94SNikolay Borisov btrfs_put_block_group(cache); 4583f603bb94SNikolay Borisov 4584f603bb94SNikolay Borisov btrfs_error_unpin_extent_range(fs_info, head->bytenr, 4585f603bb94SNikolay Borisov head->bytenr + head->num_bytes - 1); 4586f603bb94SNikolay Borisov } 458731890da0SJosef Bacik btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); 4588d278850eSJosef Bacik btrfs_put_delayed_ref_head(head); 4589acce952bSliubo cond_resched(); 4590acce952bSliubo spin_lock(&delayed_refs->lock); 4591acce952bSliubo } 459281f7eb00SJeff Mahoney btrfs_qgroup_destroy_extent_records(trans); 4593acce952bSliubo 4594acce952bSliubo spin_unlock(&delayed_refs->lock); 4595acce952bSliubo 4596acce952bSliubo return ret; 4597acce952bSliubo } 4598acce952bSliubo 4599143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 4600acce952bSliubo { 4601acce952bSliubo struct btrfs_inode *btrfs_inode; 4602acce952bSliubo struct list_head splice; 4603acce952bSliubo 4604acce952bSliubo INIT_LIST_HEAD(&splice); 4605acce952bSliubo 4606eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4607eb73c1b7SMiao Xie list_splice_init(&root->delalloc_inodes, &splice); 4608acce952bSliubo 4609acce952bSliubo while (!list_empty(&splice)) { 4610fe816d0fSNikolay Borisov struct inode *inode = NULL; 4611eb73c1b7SMiao Xie btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 4612acce952bSliubo delalloc_inodes); 4613fe816d0fSNikolay Borisov __btrfs_del_delalloc_inode(root, btrfs_inode); 4614eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4615acce952bSliubo 4616fe816d0fSNikolay Borisov /* 4617fe816d0fSNikolay Borisov * Make sure we get a live inode and that it'll not disappear 4618fe816d0fSNikolay Borisov * meanwhile. 4619fe816d0fSNikolay Borisov */ 4620fe816d0fSNikolay Borisov inode = igrab(&btrfs_inode->vfs_inode); 4621fe816d0fSNikolay Borisov if (inode) { 4622fe816d0fSNikolay Borisov invalidate_inode_pages2(inode->i_mapping); 4623fe816d0fSNikolay Borisov iput(inode); 4624fe816d0fSNikolay Borisov } 4625eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4626acce952bSliubo } 4627eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4628eb73c1b7SMiao Xie } 4629eb73c1b7SMiao Xie 4630eb73c1b7SMiao Xie static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 4631eb73c1b7SMiao Xie { 4632eb73c1b7SMiao Xie struct btrfs_root *root; 4633eb73c1b7SMiao Xie struct list_head splice; 4634eb73c1b7SMiao Xie 4635eb73c1b7SMiao Xie INIT_LIST_HEAD(&splice); 4636eb73c1b7SMiao Xie 4637eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4638eb73c1b7SMiao Xie list_splice_init(&fs_info->delalloc_roots, &splice); 4639eb73c1b7SMiao Xie while (!list_empty(&splice)) { 4640eb73c1b7SMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4641eb73c1b7SMiao Xie delalloc_root); 464200246528SJosef Bacik root = btrfs_grab_root(root); 4643eb73c1b7SMiao Xie BUG_ON(!root); 4644eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4645eb73c1b7SMiao Xie 4646eb73c1b7SMiao Xie btrfs_destroy_delalloc_inodes(root); 464700246528SJosef Bacik btrfs_put_root(root); 4648eb73c1b7SMiao Xie 4649eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4650eb73c1b7SMiao Xie } 4651eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4652acce952bSliubo } 4653acce952bSliubo 46542ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 4655acce952bSliubo struct extent_io_tree *dirty_pages, 4656acce952bSliubo int mark) 4657acce952bSliubo { 4658acce952bSliubo int ret; 4659acce952bSliubo struct extent_buffer *eb; 4660acce952bSliubo u64 start = 0; 4661acce952bSliubo u64 end; 4662acce952bSliubo 4663acce952bSliubo while (1) { 4664acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 4665e6138876SJosef Bacik mark, NULL); 4666acce952bSliubo if (ret) 4667acce952bSliubo break; 4668acce952bSliubo 466991166212SDavid Sterba clear_extent_bits(dirty_pages, start, end, mark); 4670acce952bSliubo while (start <= end) { 46710b246afaSJeff Mahoney eb = find_extent_buffer(fs_info, start); 46720b246afaSJeff Mahoney start += fs_info->nodesize; 4673fd8b2b61SJosef Bacik if (!eb) 4674acce952bSliubo continue; 4675fd8b2b61SJosef Bacik wait_on_extent_buffer_writeback(eb); 4676acce952bSliubo 4677fd8b2b61SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 4678fd8b2b61SJosef Bacik &eb->bflags)) 4679fd8b2b61SJosef Bacik clear_extent_buffer_dirty(eb); 4680fd8b2b61SJosef Bacik free_extent_buffer_stale(eb); 4681acce952bSliubo } 4682acce952bSliubo } 4683acce952bSliubo 4684acce952bSliubo return ret; 4685acce952bSliubo } 4686acce952bSliubo 46872ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 4688fe119a6eSNikolay Borisov struct extent_io_tree *unpin) 4689acce952bSliubo { 4690acce952bSliubo u64 start; 4691acce952bSliubo u64 end; 4692acce952bSliubo int ret; 4693acce952bSliubo 4694acce952bSliubo while (1) { 46950e6ec385SFilipe Manana struct extent_state *cached_state = NULL; 46960e6ec385SFilipe Manana 4697fcd5e742SLu Fengqi /* 4698fcd5e742SLu Fengqi * The btrfs_finish_extent_commit() may get the same range as 4699fcd5e742SLu Fengqi * ours between find_first_extent_bit and clear_extent_dirty. 4700fcd5e742SLu Fengqi * Hence, hold the unused_bg_unpin_mutex to avoid double unpin 4701fcd5e742SLu Fengqi * the same extent range. 4702fcd5e742SLu Fengqi */ 4703fcd5e742SLu Fengqi mutex_lock(&fs_info->unused_bg_unpin_mutex); 4704acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 47050e6ec385SFilipe Manana EXTENT_DIRTY, &cached_state); 4706fcd5e742SLu Fengqi if (ret) { 4707fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4708acce952bSliubo break; 4709fcd5e742SLu Fengqi } 4710acce952bSliubo 47110e6ec385SFilipe Manana clear_extent_dirty(unpin, start, end, &cached_state); 47120e6ec385SFilipe Manana free_extent_state(cached_state); 47132ff7e61eSJeff Mahoney btrfs_error_unpin_extent_range(fs_info, start, end); 4714fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4715acce952bSliubo cond_resched(); 4716acce952bSliubo } 4717acce952bSliubo 4718acce952bSliubo return 0; 4719acce952bSliubo } 4720acce952bSliubo 472132da5386SDavid Sterba static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache) 4722c79a1751SLiu Bo { 4723c79a1751SLiu Bo struct inode *inode; 4724c79a1751SLiu Bo 4725c79a1751SLiu Bo inode = cache->io_ctl.inode; 4726c79a1751SLiu Bo if (inode) { 4727c79a1751SLiu Bo invalidate_inode_pages2(inode->i_mapping); 4728c79a1751SLiu Bo BTRFS_I(inode)->generation = 0; 4729c79a1751SLiu Bo cache->io_ctl.inode = NULL; 4730c79a1751SLiu Bo iput(inode); 4731c79a1751SLiu Bo } 4732bbc37d6eSFilipe Manana ASSERT(cache->io_ctl.pages == NULL); 4733c79a1751SLiu Bo btrfs_put_block_group(cache); 4734c79a1751SLiu Bo } 4735c79a1751SLiu Bo 4736c79a1751SLiu Bo void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, 47372ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4738c79a1751SLiu Bo { 473932da5386SDavid Sterba struct btrfs_block_group *cache; 4740c79a1751SLiu Bo 4741c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4742c79a1751SLiu Bo while (!list_empty(&cur_trans->dirty_bgs)) { 4743c79a1751SLiu Bo cache = list_first_entry(&cur_trans->dirty_bgs, 474432da5386SDavid Sterba struct btrfs_block_group, 4745c79a1751SLiu Bo dirty_list); 4746c79a1751SLiu Bo 4747c79a1751SLiu Bo if (!list_empty(&cache->io_list)) { 4748c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4749c79a1751SLiu Bo list_del_init(&cache->io_list); 4750c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 4751c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4752c79a1751SLiu Bo } 4753c79a1751SLiu Bo 4754c79a1751SLiu Bo list_del_init(&cache->dirty_list); 4755c79a1751SLiu Bo spin_lock(&cache->lock); 4756c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 4757c79a1751SLiu Bo spin_unlock(&cache->lock); 4758c79a1751SLiu Bo 4759c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4760c79a1751SLiu Bo btrfs_put_block_group(cache); 4761ba2c4d4eSJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 4762c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 4763c79a1751SLiu Bo } 4764c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 4765c79a1751SLiu Bo 476645ae2c18SNikolay Borisov /* 476745ae2c18SNikolay Borisov * Refer to the definition of io_bgs member for details why it's safe 476845ae2c18SNikolay Borisov * to use it without any locking 476945ae2c18SNikolay Borisov */ 4770c79a1751SLiu Bo while (!list_empty(&cur_trans->io_bgs)) { 4771c79a1751SLiu Bo cache = list_first_entry(&cur_trans->io_bgs, 477232da5386SDavid Sterba struct btrfs_block_group, 4773c79a1751SLiu Bo io_list); 4774c79a1751SLiu Bo 4775c79a1751SLiu Bo list_del_init(&cache->io_list); 4776c79a1751SLiu Bo spin_lock(&cache->lock); 4777c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 4778c79a1751SLiu Bo spin_unlock(&cache->lock); 4779c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 4780c79a1751SLiu Bo } 4781c79a1751SLiu Bo } 4782c79a1751SLiu Bo 478349b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 47842ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 478549b25e05SJeff Mahoney { 4786bbbf7243SNikolay Borisov struct btrfs_device *dev, *tmp; 4787bbbf7243SNikolay Borisov 47882ff7e61eSJeff Mahoney btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 4789c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->dirty_bgs)); 4790c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->io_bgs)); 4791c79a1751SLiu Bo 4792bbbf7243SNikolay Borisov list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list, 4793bbbf7243SNikolay Borisov post_commit_list) { 4794bbbf7243SNikolay Borisov list_del_init(&dev->post_commit_list); 4795bbbf7243SNikolay Borisov } 4796bbbf7243SNikolay Borisov 47972ff7e61eSJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, fs_info); 479849b25e05SJeff Mahoney 47994a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 48000b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 480149b25e05SJeff Mahoney 48024a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 48030b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 480449b25e05SJeff Mahoney 4805ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 480667cde344SMiao Xie 48072ff7e61eSJeff Mahoney btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, 480849b25e05SJeff Mahoney EXTENT_DIRTY); 4809fe119a6eSNikolay Borisov btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents); 481049b25e05SJeff Mahoney 4811d3575156SNaohiro Aota btrfs_free_redirty_list(cur_trans); 4812d3575156SNaohiro Aota 48134a9d8bdeSMiao Xie cur_trans->state =TRANS_STATE_COMPLETED; 48144a9d8bdeSMiao Xie wake_up(&cur_trans->commit_wait); 481549b25e05SJeff Mahoney } 481649b25e05SJeff Mahoney 48172ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) 4818acce952bSliubo { 4819acce952bSliubo struct btrfs_transaction *t; 4820acce952bSliubo 48210b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 4822acce952bSliubo 48230b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 48240b246afaSJeff Mahoney while (!list_empty(&fs_info->trans_list)) { 48250b246afaSJeff Mahoney t = list_first_entry(&fs_info->trans_list, 4826724e2315SJosef Bacik struct btrfs_transaction, list); 4827724e2315SJosef Bacik if (t->state >= TRANS_STATE_COMMIT_START) { 48289b64f57dSElena Reshetova refcount_inc(&t->use_count); 48290b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 48302ff7e61eSJeff Mahoney btrfs_wait_for_commit(fs_info, t->transid); 4831724e2315SJosef Bacik btrfs_put_transaction(t); 48320b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4833724e2315SJosef Bacik continue; 4834724e2315SJosef Bacik } 48350b246afaSJeff Mahoney if (t == fs_info->running_transaction) { 4836724e2315SJosef Bacik t->state = TRANS_STATE_COMMIT_DOING; 48370b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4838724e2315SJosef Bacik /* 4839724e2315SJosef Bacik * We wait for 0 num_writers since we don't hold a trans 4840724e2315SJosef Bacik * handle open currently for this transaction. 4841724e2315SJosef Bacik */ 4842724e2315SJosef Bacik wait_event(t->writer_wait, 4843724e2315SJosef Bacik atomic_read(&t->num_writers) == 0); 4844724e2315SJosef Bacik } else { 48450b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4846724e2315SJosef Bacik } 48472ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(t, fs_info); 4848724e2315SJosef Bacik 48490b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 48500b246afaSJeff Mahoney if (t == fs_info->running_transaction) 48510b246afaSJeff Mahoney fs_info->running_transaction = NULL; 4852724e2315SJosef Bacik list_del_init(&t->list); 48530b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 4854a4abeea4SJosef Bacik 4855724e2315SJosef Bacik btrfs_put_transaction(t); 48562ff7e61eSJeff Mahoney trace_btrfs_transaction_commit(fs_info->tree_root); 48570b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 4858724e2315SJosef Bacik } 48590b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 48600b246afaSJeff Mahoney btrfs_destroy_all_ordered_extents(fs_info); 4861ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 4862ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 48630b246afaSJeff Mahoney btrfs_destroy_all_delalloc_inodes(fs_info); 4864ef67963dSJosef Bacik btrfs_drop_all_logs(fs_info); 48650b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 4866acce952bSliubo 4867acce952bSliubo return 0; 4868acce952bSliubo } 4869ec7d6dfdSNikolay Borisov 4870453e4873SNikolay Borisov int btrfs_init_root_free_objectid(struct btrfs_root *root) 4871ec7d6dfdSNikolay Borisov { 4872ec7d6dfdSNikolay Borisov struct btrfs_path *path; 4873ec7d6dfdSNikolay Borisov int ret; 4874ec7d6dfdSNikolay Borisov struct extent_buffer *l; 4875ec7d6dfdSNikolay Borisov struct btrfs_key search_key; 4876ec7d6dfdSNikolay Borisov struct btrfs_key found_key; 4877ec7d6dfdSNikolay Borisov int slot; 4878ec7d6dfdSNikolay Borisov 4879ec7d6dfdSNikolay Borisov path = btrfs_alloc_path(); 4880ec7d6dfdSNikolay Borisov if (!path) 4881ec7d6dfdSNikolay Borisov return -ENOMEM; 4882ec7d6dfdSNikolay Borisov 4883ec7d6dfdSNikolay Borisov search_key.objectid = BTRFS_LAST_FREE_OBJECTID; 4884ec7d6dfdSNikolay Borisov search_key.type = -1; 4885ec7d6dfdSNikolay Borisov search_key.offset = (u64)-1; 4886ec7d6dfdSNikolay Borisov ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); 4887ec7d6dfdSNikolay Borisov if (ret < 0) 4888ec7d6dfdSNikolay Borisov goto error; 4889ec7d6dfdSNikolay Borisov BUG_ON(ret == 0); /* Corruption */ 4890ec7d6dfdSNikolay Borisov if (path->slots[0] > 0) { 4891ec7d6dfdSNikolay Borisov slot = path->slots[0] - 1; 4892ec7d6dfdSNikolay Borisov l = path->nodes[0]; 4893ec7d6dfdSNikolay Borisov btrfs_item_key_to_cpu(l, &found_key, slot); 489423125104SNikolay Borisov root->free_objectid = max_t(u64, found_key.objectid + 1, 489523125104SNikolay Borisov BTRFS_FIRST_FREE_OBJECTID); 4896ec7d6dfdSNikolay Borisov } else { 489723125104SNikolay Borisov root->free_objectid = BTRFS_FIRST_FREE_OBJECTID; 4898ec7d6dfdSNikolay Borisov } 4899ec7d6dfdSNikolay Borisov ret = 0; 4900ec7d6dfdSNikolay Borisov error: 4901ec7d6dfdSNikolay Borisov btrfs_free_path(path); 4902ec7d6dfdSNikolay Borisov return ret; 4903ec7d6dfdSNikolay Borisov } 4904ec7d6dfdSNikolay Borisov 4905543068a2SNikolay Borisov int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid) 4906ec7d6dfdSNikolay Borisov { 4907ec7d6dfdSNikolay Borisov int ret; 4908ec7d6dfdSNikolay Borisov mutex_lock(&root->objectid_mutex); 4909ec7d6dfdSNikolay Borisov 49106b8fad57SNikolay Borisov if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) { 4911ec7d6dfdSNikolay Borisov btrfs_warn(root->fs_info, 4912ec7d6dfdSNikolay Borisov "the objectid of root %llu reaches its highest value", 4913ec7d6dfdSNikolay Borisov root->root_key.objectid); 4914ec7d6dfdSNikolay Borisov ret = -ENOSPC; 4915ec7d6dfdSNikolay Borisov goto out; 4916ec7d6dfdSNikolay Borisov } 4917ec7d6dfdSNikolay Borisov 491823125104SNikolay Borisov *objectid = root->free_objectid++; 4919ec7d6dfdSNikolay Borisov ret = 0; 4920ec7d6dfdSNikolay Borisov out: 4921ec7d6dfdSNikolay Borisov mutex_unlock(&root->objectid_mutex); 4922ec7d6dfdSNikolay Borisov return ret; 4923ec7d6dfdSNikolay Borisov } 4924