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" 45139e8cd3SQu Wenruo #include "subpage.h" 46eb60ceacSChris Mason 47319e4d06SQu Wenruo #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\ 48319e4d06SQu Wenruo BTRFS_HEADER_FLAG_RELOC |\ 49319e4d06SQu Wenruo BTRFS_SUPER_FLAG_ERROR |\ 50319e4d06SQu Wenruo BTRFS_SUPER_FLAG_SEEDING |\ 51e2731e55SAnand Jain BTRFS_SUPER_FLAG_METADUMP |\ 52e2731e55SAnand Jain BTRFS_SUPER_FLAG_METADUMP_V2) 53319e4d06SQu Wenruo 548b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work); 55143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root); 56acce952bSliubo static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 572ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info); 58143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root); 592ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 60acce952bSliubo struct extent_io_tree *dirty_pages, 61acce952bSliubo int mark); 622ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 63acce952bSliubo struct extent_io_tree *pinned_extents); 642ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info); 652ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info); 66ce9adaa5SChris Mason 67d352ac68SChris Mason /* 6897eb6b69SDavid Sterba * btrfs_end_io_wq structs are used to do processing in task context when an IO 6997eb6b69SDavid Sterba * is complete. This is used during reads to verify checksums, and it is used 70d352ac68SChris Mason * by writes to insert metadata for new file extents after IO is complete. 71d352ac68SChris Mason */ 7297eb6b69SDavid Sterba struct btrfs_end_io_wq { 73ce9adaa5SChris Mason struct bio *bio; 74ce9adaa5SChris Mason bio_end_io_t *end_io; 75ce9adaa5SChris Mason void *private; 76ce9adaa5SChris Mason struct btrfs_fs_info *info; 774e4cbee9SChristoph Hellwig blk_status_t status; 78bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata; 798b712842SChris Mason struct btrfs_work work; 80ce9adaa5SChris Mason }; 810da5468fSChris Mason 8297eb6b69SDavid Sterba static struct kmem_cache *btrfs_end_io_wq_cache; 8397eb6b69SDavid Sterba 8497eb6b69SDavid Sterba int __init btrfs_end_io_wq_init(void) 8597eb6b69SDavid Sterba { 8697eb6b69SDavid Sterba btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq", 8797eb6b69SDavid Sterba sizeof(struct btrfs_end_io_wq), 8897eb6b69SDavid Sterba 0, 89fba4b697SNikolay Borisov SLAB_MEM_SPREAD, 9097eb6b69SDavid Sterba NULL); 9197eb6b69SDavid Sterba if (!btrfs_end_io_wq_cache) 9297eb6b69SDavid Sterba return -ENOMEM; 9397eb6b69SDavid Sterba return 0; 9497eb6b69SDavid Sterba } 9597eb6b69SDavid Sterba 96e67c718bSDavid Sterba void __cold btrfs_end_io_wq_exit(void) 9797eb6b69SDavid Sterba { 9897eb6b69SDavid Sterba kmem_cache_destroy(btrfs_end_io_wq_cache); 9997eb6b69SDavid Sterba } 10097eb6b69SDavid Sterba 101141386e1SJosef Bacik static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info) 102141386e1SJosef Bacik { 103141386e1SJosef Bacik if (fs_info->csum_shash) 104141386e1SJosef Bacik crypto_free_shash(fs_info->csum_shash); 105141386e1SJosef Bacik } 106141386e1SJosef Bacik 107d352ac68SChris Mason /* 108d352ac68SChris Mason * async submit bios are used to offload expensive checksumming 109d352ac68SChris Mason * onto the worker threads. They checksum file and metadata bios 110d352ac68SChris Mason * just before they are sent down the IO stack. 111d352ac68SChris Mason */ 11244b8bd7eSChris Mason struct async_submit_bio { 1138896a08dSQu Wenruo struct inode *inode; 11444b8bd7eSChris Mason struct bio *bio; 115a758781dSDavid Sterba extent_submit_bio_start_t *submit_bio_start; 11644b8bd7eSChris Mason int mirror_num; 1171941b64bSQu Wenruo 1181941b64bSQu Wenruo /* Optional parameter for submit_bio_start used by direct io */ 1191941b64bSQu Wenruo u64 dio_file_offset; 1208b712842SChris Mason struct btrfs_work work; 1214e4cbee9SChristoph Hellwig blk_status_t status; 12244b8bd7eSChris Mason }; 12344b8bd7eSChris Mason 12485d4e461SChris Mason /* 12585d4e461SChris Mason * Lockdep class keys for extent_buffer->lock's in this root. For a given 12685d4e461SChris Mason * eb, the lockdep key is determined by the btrfs_root it belongs to and 12785d4e461SChris Mason * the level the eb occupies in the tree. 1284008c04aSChris Mason * 12985d4e461SChris Mason * Different roots are used for different purposes and may nest inside each 13085d4e461SChris Mason * other and they require separate keysets. As lockdep keys should be 13185d4e461SChris Mason * static, assign keysets according to the purpose of the root as indicated 1324fd786e6SMisono Tomohiro * by btrfs_root->root_key.objectid. This ensures that all special purpose 1334fd786e6SMisono Tomohiro * roots have separate keysets. 1344008c04aSChris Mason * 13585d4e461SChris Mason * Lock-nesting across peer nodes is always done with the immediate parent 13685d4e461SChris Mason * node locked thus preventing deadlock. As lockdep doesn't know this, use 13785d4e461SChris Mason * subclass to avoid triggering lockdep warning in such cases. 1384008c04aSChris Mason * 13985d4e461SChris Mason * The key is set by the readpage_end_io_hook after the buffer has passed 14085d4e461SChris Mason * csum validation but before the pages are unlocked. It is also set by 14185d4e461SChris Mason * btrfs_init_new_buffer on freshly allocated blocks. 14285d4e461SChris Mason * 14385d4e461SChris Mason * We also add a check to make sure the highest level of the tree is the 14485d4e461SChris Mason * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code 14585d4e461SChris Mason * needs update as well. 1464008c04aSChris Mason */ 1474008c04aSChris Mason #ifdef CONFIG_DEBUG_LOCK_ALLOC 1484008c04aSChris Mason # if BTRFS_MAX_LEVEL != 8 1494008c04aSChris Mason # error 1504008c04aSChris Mason # endif 15185d4e461SChris Mason 152ab1405aaSDavid Sterba #define DEFINE_LEVEL(stem, level) \ 153ab1405aaSDavid Sterba .names[level] = "btrfs-" stem "-0" #level, 154ab1405aaSDavid Sterba 155ab1405aaSDavid Sterba #define DEFINE_NAME(stem) \ 156ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 0) \ 157ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 1) \ 158ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 2) \ 159ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 3) \ 160ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 4) \ 161ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 5) \ 162ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 6) \ 163ab1405aaSDavid Sterba DEFINE_LEVEL(stem, 7) 164ab1405aaSDavid Sterba 16585d4e461SChris Mason static struct btrfs_lockdep_keyset { 16685d4e461SChris Mason u64 id; /* root objectid */ 167ab1405aaSDavid Sterba /* Longest entry: btrfs-free-space-00 */ 168387824afSDavid Sterba char names[BTRFS_MAX_LEVEL][20]; 169387824afSDavid Sterba struct lock_class_key keys[BTRFS_MAX_LEVEL]; 17085d4e461SChris Mason } btrfs_lockdep_keysets[] = { 171ab1405aaSDavid Sterba { .id = BTRFS_ROOT_TREE_OBJECTID, DEFINE_NAME("root") }, 172ab1405aaSDavid Sterba { .id = BTRFS_EXTENT_TREE_OBJECTID, DEFINE_NAME("extent") }, 173ab1405aaSDavid Sterba { .id = BTRFS_CHUNK_TREE_OBJECTID, DEFINE_NAME("chunk") }, 174ab1405aaSDavid Sterba { .id = BTRFS_DEV_TREE_OBJECTID, DEFINE_NAME("dev") }, 175ab1405aaSDavid Sterba { .id = BTRFS_CSUM_TREE_OBJECTID, DEFINE_NAME("csum") }, 176ab1405aaSDavid Sterba { .id = BTRFS_QUOTA_TREE_OBJECTID, DEFINE_NAME("quota") }, 177ab1405aaSDavid Sterba { .id = BTRFS_TREE_LOG_OBJECTID, DEFINE_NAME("log") }, 178ab1405aaSDavid Sterba { .id = BTRFS_TREE_RELOC_OBJECTID, DEFINE_NAME("treloc") }, 179ab1405aaSDavid Sterba { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, DEFINE_NAME("dreloc") }, 180ab1405aaSDavid Sterba { .id = BTRFS_UUID_TREE_OBJECTID, DEFINE_NAME("uuid") }, 181ab1405aaSDavid Sterba { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, DEFINE_NAME("free-space") }, 182ab1405aaSDavid Sterba { .id = 0, DEFINE_NAME("tree") }, 1834008c04aSChris Mason }; 18485d4e461SChris Mason 185ab1405aaSDavid Sterba #undef DEFINE_LEVEL 186ab1405aaSDavid Sterba #undef DEFINE_NAME 18785d4e461SChris Mason 18885d4e461SChris Mason void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb, 18985d4e461SChris Mason int level) 19085d4e461SChris Mason { 19185d4e461SChris Mason struct btrfs_lockdep_keyset *ks; 19285d4e461SChris Mason 19385d4e461SChris Mason BUG_ON(level >= ARRAY_SIZE(ks->keys)); 19485d4e461SChris Mason 19585d4e461SChris Mason /* find the matching keyset, id 0 is the default entry */ 19685d4e461SChris Mason for (ks = btrfs_lockdep_keysets; ks->id; ks++) 19785d4e461SChris Mason if (ks->id == objectid) 19885d4e461SChris Mason break; 19985d4e461SChris Mason 20085d4e461SChris Mason lockdep_set_class_and_name(&eb->lock, 20185d4e461SChris Mason &ks->keys[level], ks->names[level]); 20285d4e461SChris Mason } 20385d4e461SChris Mason 2044008c04aSChris Mason #endif 2054008c04aSChris Mason 206d352ac68SChris Mason /* 2072996e1f8SJohannes Thumshirn * Compute the csum of a btree block and store the result to provided buffer. 208d352ac68SChris Mason */ 209c67b3892SDavid Sterba static void csum_tree_block(struct extent_buffer *buf, u8 *result) 21019c00ddcSChris Mason { 211d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = buf->fs_info; 2127280305eSDavid Sterba const int num_pages = num_extent_pages(buf); 213a26663e7SQu Wenruo const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize); 214d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 21519c00ddcSChris Mason char *kaddr; 216e9be5a30SDavid Sterba int i; 217d5178578SJohannes Thumshirn 218d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 219d5178578SJohannes Thumshirn crypto_shash_init(shash); 220a26663e7SQu Wenruo kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start); 221e9be5a30SDavid Sterba crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE, 222a26663e7SQu Wenruo first_page_part - BTRFS_CSUM_SIZE); 22319c00ddcSChris Mason 224e9be5a30SDavid Sterba for (i = 1; i < num_pages; i++) { 225e9be5a30SDavid Sterba kaddr = page_address(buf->pages[i]); 226e9be5a30SDavid Sterba crypto_shash_update(shash, kaddr, PAGE_SIZE); 22719c00ddcSChris Mason } 22871a63551SDavid Sterba memset(result, 0, BTRFS_CSUM_SIZE); 229d5178578SJohannes Thumshirn crypto_shash_final(shash, result); 23019c00ddcSChris Mason } 23119c00ddcSChris Mason 232d352ac68SChris Mason /* 233d352ac68SChris Mason * we can't consider a given block up to date unless the transid of the 234d352ac68SChris Mason * block matches the transid in the parent node's pointer. This is how we 235d352ac68SChris Mason * detect blocks that either didn't get written at all or got written 236d352ac68SChris Mason * in the wrong place. 237d352ac68SChris Mason */ 2381259ab75SChris Mason static int verify_parent_transid(struct extent_io_tree *io_tree, 239b9fab919SChris Mason struct extent_buffer *eb, u64 parent_transid, 240b9fab919SChris Mason int atomic) 2411259ab75SChris Mason { 2422ac55d41SJosef Bacik struct extent_state *cached_state = NULL; 2431259ab75SChris Mason int ret; 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 2512ac55d41SJosef Bacik lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 252ff13db41SDavid Sterba &cached_state); 2530b32f4bbSJosef Bacik if (extent_buffer_uptodate(eb) && 2541259ab75SChris Mason btrfs_header_generation(eb) == parent_transid) { 2551259ab75SChris Mason ret = 0; 2561259ab75SChris Mason goto out; 2571259ab75SChris Mason } 25894647322SDavid Sterba btrfs_err_rl(eb->fs_info, 25994647322SDavid Sterba "parent transid verify failed on %llu wanted %llu found %llu", 26094647322SDavid Sterba eb->start, 26129549aecSWang Shilong parent_transid, btrfs_header_generation(eb)); 2621259ab75SChris Mason ret = 1; 2630b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 26433958dc6SChris Mason out: 2652ac55d41SJosef Bacik unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 266e43bbe5eSDavid Sterba &cached_state); 2671259ab75SChris Mason return ret; 2681259ab75SChris Mason } 2691259ab75SChris Mason 270e7e16f48SJohannes Thumshirn static bool btrfs_supported_super_csum(u16 csum_type) 271e7e16f48SJohannes Thumshirn { 272e7e16f48SJohannes Thumshirn switch (csum_type) { 273e7e16f48SJohannes Thumshirn case BTRFS_CSUM_TYPE_CRC32: 2743951e7f0SJohannes Thumshirn case BTRFS_CSUM_TYPE_XXHASH: 2753831bf00SJohannes Thumshirn case BTRFS_CSUM_TYPE_SHA256: 276352ae07bSDavid Sterba case BTRFS_CSUM_TYPE_BLAKE2: 277e7e16f48SJohannes Thumshirn return true; 278e7e16f48SJohannes Thumshirn default: 279e7e16f48SJohannes Thumshirn return false; 280e7e16f48SJohannes Thumshirn } 281e7e16f48SJohannes Thumshirn } 282e7e16f48SJohannes Thumshirn 283d352ac68SChris Mason /* 2841104a885SDavid Sterba * Return 0 if the superblock checksum type matches the checksum value of that 2851104a885SDavid Sterba * algorithm. Pass the raw disk superblock data. 2861104a885SDavid Sterba */ 287ab8d0fc4SJeff Mahoney static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info, 288ab8d0fc4SJeff Mahoney char *raw_disk_sb) 2891104a885SDavid Sterba { 2901104a885SDavid Sterba struct btrfs_super_block *disk_sb = 2911104a885SDavid Sterba (struct btrfs_super_block *)raw_disk_sb; 29251bce6c9SJohannes Thumshirn char result[BTRFS_CSUM_SIZE]; 293d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 294d5178578SJohannes Thumshirn 295d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 2961104a885SDavid Sterba 2971104a885SDavid Sterba /* 2981104a885SDavid Sterba * The super_block structure does not span the whole 29951bce6c9SJohannes Thumshirn * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is 30051bce6c9SJohannes Thumshirn * filled with zeros and is included in the checksum. 3011104a885SDavid Sterba */ 302fd08001fSEric Biggers crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE, 303fd08001fSEric Biggers BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result); 3041104a885SDavid Sterba 30555fc29beSDavid Sterba if (memcmp(disk_sb->csum, result, fs_info->csum_size)) 306e7e16f48SJohannes Thumshirn return 1; 3071104a885SDavid Sterba 308e7e16f48SJohannes Thumshirn return 0; 3091104a885SDavid Sterba } 3101104a885SDavid Sterba 311e064d5e9SDavid Sterba int btrfs_verify_level_key(struct extent_buffer *eb, int level, 312ff76a864SLiu Bo struct btrfs_key *first_key, u64 parent_transid) 313581c1760SQu Wenruo { 314e064d5e9SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 315581c1760SQu Wenruo int found_level; 316581c1760SQu Wenruo struct btrfs_key found_key; 317581c1760SQu Wenruo int ret; 318581c1760SQu Wenruo 319581c1760SQu Wenruo found_level = btrfs_header_level(eb); 320581c1760SQu Wenruo if (found_level != level) { 32163489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 32263489055SQu Wenruo KERN_ERR "BTRFS: tree level check failed\n"); 323581c1760SQu Wenruo btrfs_err(fs_info, 324581c1760SQu Wenruo "tree level mismatch detected, bytenr=%llu level expected=%u has=%u", 325581c1760SQu Wenruo eb->start, level, found_level); 326581c1760SQu Wenruo return -EIO; 327581c1760SQu Wenruo } 328581c1760SQu Wenruo 329581c1760SQu Wenruo if (!first_key) 330581c1760SQu Wenruo return 0; 331581c1760SQu Wenruo 3325d41be6fSQu Wenruo /* 3335d41be6fSQu Wenruo * For live tree block (new tree blocks in current transaction), 3345d41be6fSQu Wenruo * we need proper lock context to avoid race, which is impossible here. 3355d41be6fSQu Wenruo * So we only checks tree blocks which is read from disk, whose 3365d41be6fSQu Wenruo * generation <= fs_info->last_trans_committed. 3375d41be6fSQu Wenruo */ 3385d41be6fSQu Wenruo if (btrfs_header_generation(eb) > fs_info->last_trans_committed) 3395d41be6fSQu Wenruo return 0; 34062fdaa52SQu Wenruo 34162fdaa52SQu Wenruo /* We have @first_key, so this @eb must have at least one item */ 34262fdaa52SQu Wenruo if (btrfs_header_nritems(eb) == 0) { 34362fdaa52SQu Wenruo btrfs_err(fs_info, 34462fdaa52SQu Wenruo "invalid tree nritems, bytenr=%llu nritems=0 expect >0", 34562fdaa52SQu Wenruo eb->start); 34662fdaa52SQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 34762fdaa52SQu Wenruo return -EUCLEAN; 34862fdaa52SQu Wenruo } 34962fdaa52SQu Wenruo 350581c1760SQu Wenruo if (found_level) 351581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &found_key, 0); 352581c1760SQu Wenruo else 353581c1760SQu Wenruo btrfs_item_key_to_cpu(eb, &found_key, 0); 354581c1760SQu Wenruo ret = btrfs_comp_cpu_keys(first_key, &found_key); 355581c1760SQu Wenruo 356581c1760SQu Wenruo if (ret) { 35763489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 35863489055SQu Wenruo KERN_ERR "BTRFS: tree first key check failed\n"); 359581c1760SQu Wenruo btrfs_err(fs_info, 360ff76a864SLiu Bo "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", 361ff76a864SLiu Bo eb->start, parent_transid, first_key->objectid, 362ff76a864SLiu Bo first_key->type, first_key->offset, 363ff76a864SLiu Bo found_key.objectid, found_key.type, 364ff76a864SLiu Bo found_key.offset); 365581c1760SQu Wenruo } 366581c1760SQu Wenruo return ret; 367581c1760SQu Wenruo } 368581c1760SQu Wenruo 3691104a885SDavid Sterba /* 370d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 371d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 372581c1760SQu Wenruo * 373581c1760SQu Wenruo * @parent_transid: expected transid, skip check if 0 374581c1760SQu Wenruo * @level: expected level, mandatory check 375581c1760SQu Wenruo * @first_key: expected key of first slot, skip check if NULL 376d352ac68SChris Mason */ 377*6a2e9dc4SFilipe Manana int btrfs_read_extent_buffer(struct extent_buffer *eb, 378581c1760SQu Wenruo u64 parent_transid, int level, 379581c1760SQu Wenruo struct btrfs_key *first_key) 380f188591eSChris Mason { 3815ab12d1fSDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 382f188591eSChris Mason struct extent_io_tree *io_tree; 383ea466794SJosef Bacik int failed = 0; 384f188591eSChris Mason int ret; 385f188591eSChris Mason int num_copies = 0; 386f188591eSChris Mason int mirror_num = 0; 387ea466794SJosef Bacik int failed_mirror = 0; 388f188591eSChris Mason 3890b246afaSJeff Mahoney io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; 390f188591eSChris Mason while (1) { 391f8397d69SNikolay Borisov clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 392c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num); 393256dd1bbSStefan Behrens if (!ret) { 394581c1760SQu Wenruo if (verify_parent_transid(io_tree, eb, 395b9fab919SChris Mason parent_transid, 0)) 396256dd1bbSStefan Behrens ret = -EIO; 397e064d5e9SDavid Sterba else if (btrfs_verify_level_key(eb, level, 398ff76a864SLiu Bo first_key, parent_transid)) 399581c1760SQu Wenruo ret = -EUCLEAN; 400581c1760SQu Wenruo else 401581c1760SQu Wenruo break; 402256dd1bbSStefan Behrens } 403d397712bSChris Mason 4040b246afaSJeff Mahoney num_copies = btrfs_num_copies(fs_info, 405f188591eSChris Mason eb->start, eb->len); 4064235298eSChris Mason if (num_copies == 1) 407ea466794SJosef Bacik break; 4084235298eSChris Mason 4095cf1ab56SJosef Bacik if (!failed_mirror) { 4105cf1ab56SJosef Bacik failed = 1; 4115cf1ab56SJosef Bacik failed_mirror = eb->read_mirror; 4125cf1ab56SJosef Bacik } 4135cf1ab56SJosef Bacik 414f188591eSChris Mason mirror_num++; 415ea466794SJosef Bacik if (mirror_num == failed_mirror) 416ea466794SJosef Bacik mirror_num++; 417ea466794SJosef Bacik 4184235298eSChris Mason if (mirror_num > num_copies) 419ea466794SJosef Bacik break; 420f188591eSChris Mason } 421ea466794SJosef Bacik 422c0901581SStefan Behrens if (failed && !ret && failed_mirror) 42320a1fbf9SDavid Sterba btrfs_repair_eb_io_failure(eb, failed_mirror); 424ea466794SJosef Bacik 425ea466794SJosef Bacik return ret; 426f188591eSChris Mason } 42719c00ddcSChris Mason 428eca0f6f6SQu Wenruo static int csum_one_extent_buffer(struct extent_buffer *eb) 429eca0f6f6SQu Wenruo { 430eca0f6f6SQu Wenruo struct btrfs_fs_info *fs_info = eb->fs_info; 431eca0f6f6SQu Wenruo u8 result[BTRFS_CSUM_SIZE]; 432eca0f6f6SQu Wenruo int ret; 433eca0f6f6SQu Wenruo 434eca0f6f6SQu Wenruo ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid, 435eca0f6f6SQu Wenruo offsetof(struct btrfs_header, fsid), 436eca0f6f6SQu Wenruo BTRFS_FSID_SIZE) == 0); 437eca0f6f6SQu Wenruo csum_tree_block(eb, result); 438eca0f6f6SQu Wenruo 439eca0f6f6SQu Wenruo if (btrfs_header_level(eb)) 440eca0f6f6SQu Wenruo ret = btrfs_check_node(eb); 441eca0f6f6SQu Wenruo else 442eca0f6f6SQu Wenruo ret = btrfs_check_leaf_full(eb); 443eca0f6f6SQu Wenruo 4443777369fSQu Wenruo if (ret < 0) 4453777369fSQu Wenruo goto error; 4463777369fSQu Wenruo 4473777369fSQu Wenruo /* 4483777369fSQu Wenruo * Also check the generation, the eb reached here must be newer than 4493777369fSQu Wenruo * last committed. Or something seriously wrong happened. 4503777369fSQu Wenruo */ 4513777369fSQu Wenruo if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) { 4523777369fSQu Wenruo ret = -EUCLEAN; 453eca0f6f6SQu Wenruo btrfs_err(fs_info, 4543777369fSQu Wenruo "block=%llu bad generation, have %llu expect > %llu", 4553777369fSQu Wenruo eb->start, btrfs_header_generation(eb), 4563777369fSQu Wenruo fs_info->last_trans_committed); 4573777369fSQu Wenruo goto error; 458eca0f6f6SQu Wenruo } 459eca0f6f6SQu Wenruo write_extent_buffer(eb, result, 0, fs_info->csum_size); 460eca0f6f6SQu Wenruo 461eca0f6f6SQu Wenruo return 0; 4623777369fSQu Wenruo 4633777369fSQu Wenruo error: 4643777369fSQu Wenruo btrfs_print_tree(eb, 0); 4653777369fSQu Wenruo btrfs_err(fs_info, "block=%llu write time tree block corruption detected", 4663777369fSQu Wenruo eb->start); 4673777369fSQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 4683777369fSQu Wenruo return ret; 469eca0f6f6SQu Wenruo } 470eca0f6f6SQu Wenruo 471eca0f6f6SQu Wenruo /* Checksum all dirty extent buffers in one bio_vec */ 472eca0f6f6SQu Wenruo static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info, 473eca0f6f6SQu Wenruo struct bio_vec *bvec) 474eca0f6f6SQu Wenruo { 475eca0f6f6SQu Wenruo struct page *page = bvec->bv_page; 476eca0f6f6SQu Wenruo u64 bvec_start = page_offset(page) + bvec->bv_offset; 477eca0f6f6SQu Wenruo u64 cur; 478eca0f6f6SQu Wenruo int ret = 0; 479eca0f6f6SQu Wenruo 480eca0f6f6SQu Wenruo for (cur = bvec_start; cur < bvec_start + bvec->bv_len; 481eca0f6f6SQu Wenruo cur += fs_info->nodesize) { 482eca0f6f6SQu Wenruo struct extent_buffer *eb; 483eca0f6f6SQu Wenruo bool uptodate; 484eca0f6f6SQu Wenruo 485eca0f6f6SQu Wenruo eb = find_extent_buffer(fs_info, cur); 486eca0f6f6SQu Wenruo uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur, 487eca0f6f6SQu Wenruo fs_info->nodesize); 488eca0f6f6SQu Wenruo 489eca0f6f6SQu Wenruo /* A dirty eb shouldn't disappear from buffer_radix */ 490eca0f6f6SQu Wenruo if (WARN_ON(!eb)) 491eca0f6f6SQu Wenruo return -EUCLEAN; 492eca0f6f6SQu Wenruo 493eca0f6f6SQu Wenruo if (WARN_ON(cur != btrfs_header_bytenr(eb))) { 494eca0f6f6SQu Wenruo free_extent_buffer(eb); 495eca0f6f6SQu Wenruo return -EUCLEAN; 496eca0f6f6SQu Wenruo } 497eca0f6f6SQu Wenruo if (WARN_ON(!uptodate)) { 498eca0f6f6SQu Wenruo free_extent_buffer(eb); 499eca0f6f6SQu Wenruo return -EUCLEAN; 500eca0f6f6SQu Wenruo } 501eca0f6f6SQu Wenruo 502eca0f6f6SQu Wenruo ret = csum_one_extent_buffer(eb); 503eca0f6f6SQu Wenruo free_extent_buffer(eb); 504eca0f6f6SQu Wenruo if (ret < 0) 505eca0f6f6SQu Wenruo return ret; 506eca0f6f6SQu Wenruo } 507eca0f6f6SQu Wenruo return ret; 508eca0f6f6SQu Wenruo } 509eca0f6f6SQu Wenruo 510d352ac68SChris Mason /* 511ac303b69SQu Wenruo * Checksum a dirty tree block before IO. This has extra checks to make sure 512ac303b69SQu Wenruo * we only fill in the checksum field in the first page of a multi-page block. 513ac303b69SQu Wenruo * For subpage extent buffers we need bvec to also read the offset in the page. 514d352ac68SChris Mason */ 515ac303b69SQu Wenruo static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec) 51619c00ddcSChris Mason { 517ac303b69SQu Wenruo struct page *page = bvec->bv_page; 5184eee4fa4SMiao Xie u64 start = page_offset(page); 51919c00ddcSChris Mason u64 found_start; 52019c00ddcSChris Mason struct extent_buffer *eb; 521eca0f6f6SQu Wenruo 522eca0f6f6SQu Wenruo if (fs_info->sectorsize < PAGE_SIZE) 523eca0f6f6SQu Wenruo return csum_dirty_subpage_buffers(fs_info, bvec); 524f188591eSChris Mason 5254f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 5264f2de97aSJosef Bacik if (page != eb->pages[0]) 5274f2de97aSJosef Bacik return 0; 5280f805531SAlex Lyakas 52919c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 530d3575156SNaohiro Aota 531d3575156SNaohiro Aota if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) { 532d3575156SNaohiro Aota WARN_ON(found_start != 0); 533d3575156SNaohiro Aota return 0; 534d3575156SNaohiro Aota } 535d3575156SNaohiro Aota 5360f805531SAlex Lyakas /* 5370f805531SAlex Lyakas * Please do not consolidate these warnings into a single if. 5380f805531SAlex Lyakas * It is useful to know what went wrong. 5390f805531SAlex Lyakas */ 5400f805531SAlex Lyakas if (WARN_ON(found_start != start)) 5410f805531SAlex Lyakas return -EUCLEAN; 5420f805531SAlex Lyakas if (WARN_ON(!PageUptodate(page))) 5430f805531SAlex Lyakas return -EUCLEAN; 5440f805531SAlex Lyakas 545eca0f6f6SQu Wenruo return csum_one_extent_buffer(eb); 54619c00ddcSChris Mason } 54719c00ddcSChris Mason 548b0c9b3b0SDavid Sterba static int check_tree_block_fsid(struct extent_buffer *eb) 5492b82032cSYan Zheng { 550b0c9b3b0SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 551944d3f9fSNikolay Borisov struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 55244880fdcSAnand Jain u8 fsid[BTRFS_FSID_SIZE]; 553944d3f9fSNikolay Borisov u8 *metadata_uuid; 5542b82032cSYan Zheng 5559a8658e3SDavid Sterba read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid), 5569a8658e3SDavid Sterba BTRFS_FSID_SIZE); 5577239ff4bSNikolay Borisov /* 558944d3f9fSNikolay Borisov * Checking the incompat flag is only valid for the current fs. For 559944d3f9fSNikolay Borisov * seed devices it's forbidden to have their uuid changed so reading 560944d3f9fSNikolay Borisov * ->fsid in this case is fine 5617239ff4bSNikolay Borisov */ 562944d3f9fSNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID)) 5637239ff4bSNikolay Borisov metadata_uuid = fs_devices->metadata_uuid; 5647239ff4bSNikolay Borisov else 5657239ff4bSNikolay Borisov metadata_uuid = fs_devices->fsid; 5667239ff4bSNikolay Borisov 567944d3f9fSNikolay Borisov if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) 568944d3f9fSNikolay Borisov return 0; 569944d3f9fSNikolay Borisov 570944d3f9fSNikolay Borisov list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) 571944d3f9fSNikolay Borisov if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE)) 572944d3f9fSNikolay Borisov return 0; 573944d3f9fSNikolay Borisov 574944d3f9fSNikolay Borisov return 1; 5752b82032cSYan Zheng } 5762b82032cSYan Zheng 57777bf40a2SQu Wenruo /* Do basic extent buffer checks at read time */ 57877bf40a2SQu Wenruo static int validate_extent_buffer(struct extent_buffer *eb) 579ce9adaa5SChris Mason { 58077bf40a2SQu Wenruo struct btrfs_fs_info *fs_info = eb->fs_info; 581ce9adaa5SChris Mason u64 found_start; 58277bf40a2SQu Wenruo const u32 csum_size = fs_info->csum_size; 58377bf40a2SQu Wenruo u8 found_level; 5842996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 585dfd29eedSDavid Sterba const u8 *header_csum; 58677bf40a2SQu Wenruo int ret = 0; 587ea466794SJosef Bacik 588ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 589727011e0SChris Mason if (found_start != eb->start) { 590893bf4b1SSu Yue btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu", 591893bf4b1SSu Yue eb->start, found_start); 592f188591eSChris Mason ret = -EIO; 59377bf40a2SQu Wenruo goto out; 594ce9adaa5SChris Mason } 595b0c9b3b0SDavid Sterba if (check_tree_block_fsid(eb)) { 59602873e43SZhao Lei btrfs_err_rl(fs_info, "bad fsid on block %llu", 59794647322SDavid Sterba eb->start); 5981259ab75SChris Mason ret = -EIO; 59977bf40a2SQu Wenruo goto out; 6001259ab75SChris Mason } 601ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 6021c24c3ceSJosef Bacik if (found_level >= BTRFS_MAX_LEVEL) { 603893bf4b1SSu Yue btrfs_err(fs_info, "bad tree block level %d on %llu", 604893bf4b1SSu Yue (int)btrfs_header_level(eb), eb->start); 6051c24c3ceSJosef Bacik ret = -EIO; 60677bf40a2SQu Wenruo goto out; 6071c24c3ceSJosef Bacik } 608ce9adaa5SChris Mason 609c67b3892SDavid Sterba csum_tree_block(eb, result); 610dfd29eedSDavid Sterba header_csum = page_address(eb->pages[0]) + 611dfd29eedSDavid Sterba get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum)); 612a826d6dcSJosef Bacik 613dfd29eedSDavid Sterba if (memcmp(result, header_csum, csum_size) != 0) { 6142996e1f8SJohannes Thumshirn btrfs_warn_rl(fs_info, 615ff14aa79SDavid Sterba "checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d", 616ff14aa79SDavid Sterba eb->start, 617dfd29eedSDavid Sterba CSUM_FMT_VALUE(csum_size, header_csum), 61835be8851SJohannes Thumshirn CSUM_FMT_VALUE(csum_size, result), 61935be8851SJohannes Thumshirn btrfs_header_level(eb)); 6202996e1f8SJohannes Thumshirn ret = -EUCLEAN; 62177bf40a2SQu Wenruo goto out; 6222996e1f8SJohannes Thumshirn } 6232996e1f8SJohannes Thumshirn 624a826d6dcSJosef Bacik /* 625a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 626a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 627a826d6dcSJosef Bacik * return -EIO. 628a826d6dcSJosef Bacik */ 6291c4360eeSDavid Sterba if (found_level == 0 && btrfs_check_leaf_full(eb)) { 630a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 631a826d6dcSJosef Bacik ret = -EIO; 632a826d6dcSJosef Bacik } 633ce9adaa5SChris Mason 634813fd1dcSDavid Sterba if (found_level > 0 && btrfs_check_node(eb)) 635053ab70fSLiu Bo ret = -EIO; 636053ab70fSLiu Bo 6370b32f4bbSJosef Bacik if (!ret) 6380b32f4bbSJosef Bacik set_extent_buffer_uptodate(eb); 63975391f0dSQu Wenruo else 64075391f0dSQu Wenruo btrfs_err(fs_info, 64175391f0dSQu Wenruo "block=%llu read time tree block corruption detected", 64275391f0dSQu Wenruo eb->start); 64377bf40a2SQu Wenruo out: 64477bf40a2SQu Wenruo return ret; 64577bf40a2SQu Wenruo } 64677bf40a2SQu Wenruo 647371cdc07SQu Wenruo static int validate_subpage_buffer(struct page *page, u64 start, u64 end, 648371cdc07SQu Wenruo int mirror) 649371cdc07SQu Wenruo { 650371cdc07SQu Wenruo struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); 651371cdc07SQu Wenruo struct extent_buffer *eb; 652371cdc07SQu Wenruo bool reads_done; 653371cdc07SQu Wenruo int ret = 0; 654371cdc07SQu Wenruo 655371cdc07SQu Wenruo /* 656371cdc07SQu Wenruo * We don't allow bio merge for subpage metadata read, so we should 657371cdc07SQu Wenruo * only get one eb for each endio hook. 658371cdc07SQu Wenruo */ 659371cdc07SQu Wenruo ASSERT(end == start + fs_info->nodesize - 1); 660371cdc07SQu Wenruo ASSERT(PagePrivate(page)); 661371cdc07SQu Wenruo 662371cdc07SQu Wenruo eb = find_extent_buffer(fs_info, start); 663371cdc07SQu Wenruo /* 664371cdc07SQu Wenruo * When we are reading one tree block, eb must have been inserted into 665371cdc07SQu Wenruo * the radix tree. If not, something is wrong. 666371cdc07SQu Wenruo */ 667371cdc07SQu Wenruo ASSERT(eb); 668371cdc07SQu Wenruo 669371cdc07SQu Wenruo reads_done = atomic_dec_and_test(&eb->io_pages); 670371cdc07SQu Wenruo /* Subpage read must finish in page read */ 671371cdc07SQu Wenruo ASSERT(reads_done); 672371cdc07SQu Wenruo 673371cdc07SQu Wenruo eb->read_mirror = mirror; 674371cdc07SQu Wenruo if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 675371cdc07SQu Wenruo ret = -EIO; 676371cdc07SQu Wenruo goto err; 677371cdc07SQu Wenruo } 678371cdc07SQu Wenruo ret = validate_extent_buffer(eb); 679371cdc07SQu Wenruo if (ret < 0) 680371cdc07SQu Wenruo goto err; 681371cdc07SQu Wenruo 682371cdc07SQu Wenruo set_extent_buffer_uptodate(eb); 683371cdc07SQu Wenruo 684371cdc07SQu Wenruo free_extent_buffer(eb); 685371cdc07SQu Wenruo return ret; 686371cdc07SQu Wenruo err: 687371cdc07SQu Wenruo /* 688371cdc07SQu Wenruo * end_bio_extent_readpage decrements io_pages in case of error, 689371cdc07SQu Wenruo * make sure it has something to decrement. 690371cdc07SQu Wenruo */ 691371cdc07SQu Wenruo atomic_inc(&eb->io_pages); 692371cdc07SQu Wenruo clear_extent_buffer_uptodate(eb); 693371cdc07SQu Wenruo free_extent_buffer(eb); 694371cdc07SQu Wenruo return ret; 695371cdc07SQu Wenruo } 696371cdc07SQu Wenruo 697c3a3b19bSQu Wenruo int btrfs_validate_metadata_buffer(struct btrfs_bio *bbio, 69877bf40a2SQu Wenruo struct page *page, u64 start, u64 end, 69977bf40a2SQu Wenruo int mirror) 70077bf40a2SQu Wenruo { 70177bf40a2SQu Wenruo struct extent_buffer *eb; 70277bf40a2SQu Wenruo int ret = 0; 70377bf40a2SQu Wenruo int reads_done; 70477bf40a2SQu Wenruo 70577bf40a2SQu Wenruo ASSERT(page->private); 706371cdc07SQu Wenruo 707371cdc07SQu Wenruo if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE) 708371cdc07SQu Wenruo return validate_subpage_buffer(page, start, end, mirror); 709371cdc07SQu Wenruo 71077bf40a2SQu Wenruo eb = (struct extent_buffer *)page->private; 71177bf40a2SQu Wenruo 71277bf40a2SQu Wenruo /* 71377bf40a2SQu Wenruo * The pending IO might have been the only thing that kept this buffer 71477bf40a2SQu Wenruo * in memory. Make sure we have a ref for all this other checks 71577bf40a2SQu Wenruo */ 71677bf40a2SQu Wenruo atomic_inc(&eb->refs); 71777bf40a2SQu Wenruo 71877bf40a2SQu Wenruo reads_done = atomic_dec_and_test(&eb->io_pages); 71977bf40a2SQu Wenruo if (!reads_done) 72077bf40a2SQu Wenruo goto err; 72177bf40a2SQu Wenruo 72277bf40a2SQu Wenruo eb->read_mirror = mirror; 72377bf40a2SQu Wenruo if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 72477bf40a2SQu Wenruo ret = -EIO; 72577bf40a2SQu Wenruo goto err; 72677bf40a2SQu Wenruo } 72777bf40a2SQu Wenruo ret = validate_extent_buffer(eb); 728ce9adaa5SChris Mason err: 72953b381b3SDavid Woodhouse if (ret) { 73053b381b3SDavid Woodhouse /* 73153b381b3SDavid Woodhouse * our io error hook is going to dec the io pages 73253b381b3SDavid Woodhouse * again, we have to make sure it has something 73353b381b3SDavid Woodhouse * to decrement 73453b381b3SDavid Woodhouse */ 73553b381b3SDavid Woodhouse atomic_inc(&eb->io_pages); 7360b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 73753b381b3SDavid Woodhouse } 7380b32f4bbSJosef Bacik free_extent_buffer(eb); 73977bf40a2SQu Wenruo 740f188591eSChris Mason return ret; 741ce9adaa5SChris Mason } 742ce9adaa5SChris Mason 7434246a0b6SChristoph Hellwig static void end_workqueue_bio(struct bio *bio) 744ce9adaa5SChris Mason { 74597eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq = bio->bi_private; 746ce9adaa5SChris Mason struct btrfs_fs_info *fs_info; 7479e0af237SLiu Bo struct btrfs_workqueue *wq; 748ce9adaa5SChris Mason 749ce9adaa5SChris Mason fs_info = end_io_wq->info; 7504e4cbee9SChristoph Hellwig end_io_wq->status = bio->bi_status; 751d20f7043SChris Mason 752cfe94440SNaohiro Aota if (btrfs_op(bio) == BTRFS_MAP_WRITE) { 753a0cac0ecSOmar Sandoval if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) 7549e0af237SLiu Bo wq = fs_info->endio_meta_write_workers; 755a0cac0ecSOmar Sandoval else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) 7569e0af237SLiu Bo wq = fs_info->endio_freespace_worker; 757a0cac0ecSOmar Sandoval else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 7589e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 759a0cac0ecSOmar Sandoval else 7609e0af237SLiu Bo wq = fs_info->endio_write_workers; 7619e0af237SLiu Bo } else { 7625c047a69SOmar Sandoval if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 7639e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 764a0cac0ecSOmar Sandoval else if (end_io_wq->metadata) 7659e0af237SLiu Bo wq = fs_info->endio_meta_workers; 766a0cac0ecSOmar Sandoval else 7679e0af237SLiu Bo wq = fs_info->endio_workers; 7689e0af237SLiu Bo } 7699e0af237SLiu Bo 770a0cac0ecSOmar Sandoval btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL); 7719e0af237SLiu Bo btrfs_queue_work(wq, &end_io_wq->work); 772ce9adaa5SChris Mason } 773ce9adaa5SChris Mason 7744e4cbee9SChristoph Hellwig blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 775bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata) 7760b86a832SChris Mason { 77797eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 7788b110e39SMiao Xie 77997eb6b69SDavid Sterba end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS); 780ce9adaa5SChris Mason if (!end_io_wq) 7814e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 782ce9adaa5SChris Mason 783ce9adaa5SChris Mason end_io_wq->private = bio->bi_private; 784ce9adaa5SChris Mason end_io_wq->end_io = bio->bi_end_io; 78522c59948SChris Mason end_io_wq->info = info; 7864e4cbee9SChristoph Hellwig end_io_wq->status = 0; 787ce9adaa5SChris Mason end_io_wq->bio = bio; 78822c59948SChris Mason end_io_wq->metadata = metadata; 789ce9adaa5SChris Mason 790ce9adaa5SChris Mason bio->bi_private = end_io_wq; 791ce9adaa5SChris Mason bio->bi_end_io = end_workqueue_bio; 79222c59948SChris Mason return 0; 79322c59948SChris Mason } 79422c59948SChris Mason 7954a69a410SChris Mason static void run_one_async_start(struct btrfs_work *work) 7964a69a410SChris Mason { 7974a69a410SChris Mason struct async_submit_bio *async; 7984e4cbee9SChristoph Hellwig blk_status_t ret; 7994a69a410SChris Mason 8004a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 8011941b64bSQu Wenruo ret = async->submit_bio_start(async->inode, async->bio, 8021941b64bSQu Wenruo async->dio_file_offset); 80379787eaaSJeff Mahoney if (ret) 8044e4cbee9SChristoph Hellwig async->status = ret; 8054a69a410SChris Mason } 8064a69a410SChris Mason 80706ea01b1SDavid Sterba /* 80806ea01b1SDavid Sterba * In order to insert checksums into the metadata in large chunks, we wait 80906ea01b1SDavid Sterba * until bio submission time. All the pages in the bio are checksummed and 81006ea01b1SDavid Sterba * sums are attached onto the ordered extent record. 81106ea01b1SDavid Sterba * 81206ea01b1SDavid Sterba * At IO completion time the csums attached on the ordered extent record are 81306ea01b1SDavid Sterba * inserted into the tree. 81406ea01b1SDavid Sterba */ 8154a69a410SChris Mason static void run_one_async_done(struct btrfs_work *work) 8168b712842SChris Mason { 8178b712842SChris Mason struct async_submit_bio *async; 81806ea01b1SDavid Sterba struct inode *inode; 81906ea01b1SDavid Sterba blk_status_t ret; 8208b712842SChris Mason 8218b712842SChris Mason async = container_of(work, struct async_submit_bio, work); 8228896a08dSQu Wenruo inode = async->inode; 8234854ddd0SChris Mason 824bb7ab3b9SAdam Buchbinder /* If an error occurred we just want to clean up the bio and move on */ 8254e4cbee9SChristoph Hellwig if (async->status) { 8264e4cbee9SChristoph Hellwig async->bio->bi_status = async->status; 8274246a0b6SChristoph Hellwig bio_endio(async->bio); 82879787eaaSJeff Mahoney return; 82979787eaaSJeff Mahoney } 83079787eaaSJeff Mahoney 831ec39f769SChris Mason /* 832ec39f769SChris Mason * All of the bios that pass through here are from async helpers. 833ec39f769SChris Mason * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context. 834ec39f769SChris Mason * This changes nothing when cgroups aren't in use. 835ec39f769SChris Mason */ 836ec39f769SChris Mason async->bio->bi_opf |= REQ_CGROUP_PUNT; 83708635baeSChris Mason ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num); 83806ea01b1SDavid Sterba if (ret) { 83906ea01b1SDavid Sterba async->bio->bi_status = ret; 84006ea01b1SDavid Sterba bio_endio(async->bio); 84106ea01b1SDavid Sterba } 8424a69a410SChris Mason } 8434a69a410SChris Mason 8444a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 8454a69a410SChris Mason { 8464a69a410SChris Mason struct async_submit_bio *async; 8474a69a410SChris Mason 8484a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 8498b712842SChris Mason kfree(async); 8508b712842SChris Mason } 8518b712842SChris Mason 8528896a08dSQu Wenruo blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio, 853c6100a4bSJosef Bacik int mirror_num, unsigned long bio_flags, 8541941b64bSQu Wenruo u64 dio_file_offset, 855e288c080SDavid Sterba extent_submit_bio_start_t *submit_bio_start) 85644b8bd7eSChris Mason { 8578896a08dSQu Wenruo struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; 85844b8bd7eSChris Mason struct async_submit_bio *async; 85944b8bd7eSChris Mason 86044b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 86144b8bd7eSChris Mason if (!async) 8624e4cbee9SChristoph Hellwig return BLK_STS_RESOURCE; 86344b8bd7eSChris Mason 8648896a08dSQu Wenruo async->inode = inode; 86544b8bd7eSChris Mason async->bio = bio; 86644b8bd7eSChris Mason async->mirror_num = mirror_num; 8674a69a410SChris Mason async->submit_bio_start = submit_bio_start; 8684a69a410SChris Mason 869a0cac0ecSOmar Sandoval btrfs_init_work(&async->work, run_one_async_start, run_one_async_done, 870a0cac0ecSOmar Sandoval run_one_async_free); 8714a69a410SChris Mason 8721941b64bSQu Wenruo async->dio_file_offset = dio_file_offset; 8738c8bee1dSChris Mason 8744e4cbee9SChristoph Hellwig async->status = 0; 87579787eaaSJeff Mahoney 87667f055c7SChristoph Hellwig if (op_is_sync(bio->bi_opf)) 8775cdc7ad3SQu Wenruo btrfs_set_work_high_priority(&async->work); 878d313d7a3SChris Mason 8795cdc7ad3SQu Wenruo btrfs_queue_work(fs_info->workers, &async->work); 88044b8bd7eSChris Mason return 0; 88144b8bd7eSChris Mason } 88244b8bd7eSChris Mason 8834e4cbee9SChristoph Hellwig static blk_status_t btree_csum_one_bio(struct bio *bio) 884ce3ed71aSChris Mason { 8852c30c71bSKent Overstreet struct bio_vec *bvec; 886ce3ed71aSChris Mason struct btrfs_root *root; 8872b070cfeSChristoph Hellwig int ret = 0; 8886dc4f100SMing Lei struct bvec_iter_all iter_all; 889ce3ed71aSChris Mason 890c09abff8SDavid Sterba ASSERT(!bio_flagged(bio, BIO_CLONED)); 8912b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 892ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 893ac303b69SQu Wenruo ret = csum_dirty_buffer(root->fs_info, bvec); 89479787eaaSJeff Mahoney if (ret) 89579787eaaSJeff Mahoney break; 896ce3ed71aSChris Mason } 8972c30c71bSKent Overstreet 8984e4cbee9SChristoph Hellwig return errno_to_blk_status(ret); 899ce3ed71aSChris Mason } 900ce3ed71aSChris Mason 9018896a08dSQu Wenruo static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio, 9021941b64bSQu Wenruo u64 dio_file_offset) 90322c59948SChris Mason { 9048b712842SChris Mason /* 9058b712842SChris Mason * when we're called for a write, we're already in the async 9065443be45SChris Mason * submission context. Just jump into btrfs_map_bio 9078b712842SChris Mason */ 90879787eaaSJeff Mahoney return btree_csum_one_bio(bio); 90922c59948SChris Mason } 91022c59948SChris Mason 911f4dcfb30SJohannes Thumshirn static bool should_async_write(struct btrfs_fs_info *fs_info, 9129b4e675aSDavid Sterba struct btrfs_inode *bi) 913de0022b9SJosef Bacik { 9144eef29efSNaohiro Aota if (btrfs_is_zoned(fs_info)) 915f4dcfb30SJohannes Thumshirn return false; 9166300463bSLiu Bo if (atomic_read(&bi->sync_writers)) 917f4dcfb30SJohannes Thumshirn return false; 9189b4e675aSDavid Sterba if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags)) 919f4dcfb30SJohannes Thumshirn return false; 920f4dcfb30SJohannes Thumshirn return true; 921de0022b9SJosef Bacik } 922de0022b9SJosef Bacik 9231b36294aSNikolay Borisov blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio, 9241b36294aSNikolay Borisov int mirror_num, unsigned long bio_flags) 92544b8bd7eSChris Mason { 9260b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 9274e4cbee9SChristoph Hellwig blk_status_t ret; 928cad321adSChris Mason 929cfe94440SNaohiro Aota if (btrfs_op(bio) != BTRFS_MAP_WRITE) { 930cad321adSChris Mason /* 931cad321adSChris Mason * called for a read, do the setup so that checksum validation 932cad321adSChris Mason * can happen in the async kernel threads 933cad321adSChris Mason */ 9340b246afaSJeff Mahoney ret = btrfs_bio_wq_end_io(fs_info, bio, 9350b246afaSJeff Mahoney BTRFS_WQ_ENDIO_METADATA); 9361d4284bdSChris Mason if (ret) 93761891923SStefan Behrens goto out_w_error; 93808635baeSChris Mason ret = btrfs_map_bio(fs_info, bio, mirror_num); 939f4dcfb30SJohannes Thumshirn } else if (!should_async_write(fs_info, BTRFS_I(inode))) { 940de0022b9SJosef Bacik ret = btree_csum_one_bio(bio); 941de0022b9SJosef Bacik if (ret) 94261891923SStefan Behrens goto out_w_error; 94308635baeSChris Mason ret = btrfs_map_bio(fs_info, bio, mirror_num); 94461891923SStefan Behrens } else { 945cad321adSChris Mason /* 94661891923SStefan Behrens * kthread helpers are used to submit writes so that 94761891923SStefan Behrens * checksumming can happen in parallel across all CPUs 948cad321adSChris Mason */ 9498896a08dSQu Wenruo ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0, 9508896a08dSQu Wenruo 0, btree_submit_bio_start); 95144b8bd7eSChris Mason } 95244b8bd7eSChris Mason 9534246a0b6SChristoph Hellwig if (ret) 9544246a0b6SChristoph Hellwig goto out_w_error; 9554246a0b6SChristoph Hellwig return 0; 9564246a0b6SChristoph Hellwig 95761891923SStefan Behrens out_w_error: 9584e4cbee9SChristoph Hellwig bio->bi_status = ret; 9594246a0b6SChristoph Hellwig bio_endio(bio); 96061891923SStefan Behrens return ret; 96161891923SStefan Behrens } 96261891923SStefan Behrens 9633dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 964784b4e29SChris Mason static int btree_migratepage(struct address_space *mapping, 965a6bc32b8SMel Gorman struct page *newpage, struct page *page, 966a6bc32b8SMel Gorman enum migrate_mode mode) 967784b4e29SChris Mason { 968784b4e29SChris Mason /* 969784b4e29SChris Mason * we can't safely write a btree page from here, 970784b4e29SChris Mason * we haven't done the locking hook 971784b4e29SChris Mason */ 972784b4e29SChris Mason if (PageDirty(page)) 973784b4e29SChris Mason return -EAGAIN; 974784b4e29SChris Mason /* 975784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 976784b4e29SChris Mason * We must have no buffers or drop them. 977784b4e29SChris Mason */ 978784b4e29SChris Mason if (page_has_private(page) && 979784b4e29SChris Mason !try_to_release_page(page, GFP_KERNEL)) 980784b4e29SChris Mason return -EAGAIN; 981a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 982784b4e29SChris Mason } 9833dd1462eSJan Beulich #endif 984784b4e29SChris Mason 9850da5468fSChris Mason 9860da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 9870da5468fSChris Mason struct writeback_control *wbc) 9880da5468fSChris Mason { 989e2d84521SMiao Xie struct btrfs_fs_info *fs_info; 990e2d84521SMiao Xie int ret; 991e2d84521SMiao Xie 992d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 993448d640bSChris Mason 994448d640bSChris Mason if (wbc->for_kupdate) 995448d640bSChris Mason return 0; 996448d640bSChris Mason 997e2d84521SMiao Xie fs_info = BTRFS_I(mapping->host)->root->fs_info; 998b9473439SChris Mason /* this is a bit racy, but that's ok */ 999d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 1000d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 1001d814a491SEthan Lien fs_info->dirty_metadata_batch); 1002e2d84521SMiao Xie if (ret < 0) 1003793955bcSChris Mason return 0; 1004793955bcSChris Mason } 10050b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 10060da5468fSChris Mason } 10070da5468fSChris Mason 100870dec807SChris Mason static int btree_releasepage(struct page *page, gfp_t gfp_flags) 10095f39d397SChris Mason { 101098509cfcSChris Mason if (PageWriteback(page) || PageDirty(page)) 101198509cfcSChris Mason return 0; 10120c4e538bSDavid Sterba 1013f7a52a40SDavid Sterba return try_release_extent_buffer(page); 1014d98237b3SChris Mason } 1015d98237b3SChris Mason 1016895586ebSMatthew Wilcox (Oracle) static void btree_invalidate_folio(struct folio *folio, size_t offset, 1017895586ebSMatthew Wilcox (Oracle) size_t length) 1018d98237b3SChris Mason { 1019d1310b2eSChris Mason struct extent_io_tree *tree; 1020895586ebSMatthew Wilcox (Oracle) tree = &BTRFS_I(folio->mapping->host)->io_tree; 1021895586ebSMatthew Wilcox (Oracle) extent_invalidate_folio(tree, folio, offset); 1022895586ebSMatthew Wilcox (Oracle) btree_releasepage(&folio->page, GFP_NOFS); 1023895586ebSMatthew Wilcox (Oracle) if (folio_get_private(folio)) { 1024895586ebSMatthew Wilcox (Oracle) btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info, 1025895586ebSMatthew Wilcox (Oracle) "folio private not zero on folio %llu", 1026895586ebSMatthew Wilcox (Oracle) (unsigned long long)folio_pos(folio)); 1027895586ebSMatthew Wilcox (Oracle) folio_detach_private(folio); 10289ad6b7bcSChris Mason } 1029d98237b3SChris Mason } 1030d98237b3SChris Mason 1031bb146eb2SJosef Bacik #ifdef DEBUG 10320079c3b1SMatthew Wilcox (Oracle) static bool btree_dirty_folio(struct address_space *mapping, 10330079c3b1SMatthew Wilcox (Oracle) struct folio *folio) 10340079c3b1SMatthew Wilcox (Oracle) { 10350079c3b1SMatthew Wilcox (Oracle) struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb); 1036139e8cd3SQu Wenruo struct btrfs_subpage *subpage; 10370b32f4bbSJosef Bacik struct extent_buffer *eb; 1038139e8cd3SQu Wenruo int cur_bit = 0; 10390079c3b1SMatthew Wilcox (Oracle) u64 page_start = folio_pos(folio); 10400b32f4bbSJosef Bacik 1041139e8cd3SQu Wenruo if (fs_info->sectorsize == PAGE_SIZE) { 10420079c3b1SMatthew Wilcox (Oracle) eb = folio_get_private(folio); 10430b32f4bbSJosef Bacik BUG_ON(!eb); 10440b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 10450b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 104649d0c642SFilipe Manana btrfs_assert_tree_write_locked(eb); 10470079c3b1SMatthew Wilcox (Oracle) return filemap_dirty_folio(mapping, folio); 1048139e8cd3SQu Wenruo } 10490079c3b1SMatthew Wilcox (Oracle) subpage = folio_get_private(folio); 1050139e8cd3SQu Wenruo 1051139e8cd3SQu Wenruo ASSERT(subpage->dirty_bitmap); 1052139e8cd3SQu Wenruo while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) { 1053139e8cd3SQu Wenruo unsigned long flags; 1054139e8cd3SQu Wenruo u64 cur; 1055139e8cd3SQu Wenruo u16 tmp = (1 << cur_bit); 1056139e8cd3SQu Wenruo 1057139e8cd3SQu Wenruo spin_lock_irqsave(&subpage->lock, flags); 1058139e8cd3SQu Wenruo if (!(tmp & subpage->dirty_bitmap)) { 1059139e8cd3SQu Wenruo spin_unlock_irqrestore(&subpage->lock, flags); 1060139e8cd3SQu Wenruo cur_bit++; 1061139e8cd3SQu Wenruo continue; 1062139e8cd3SQu Wenruo } 1063139e8cd3SQu Wenruo spin_unlock_irqrestore(&subpage->lock, flags); 1064139e8cd3SQu Wenruo cur = page_start + cur_bit * fs_info->sectorsize; 1065139e8cd3SQu Wenruo 1066139e8cd3SQu Wenruo eb = find_extent_buffer(fs_info, cur); 1067139e8cd3SQu Wenruo ASSERT(eb); 1068139e8cd3SQu Wenruo ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 1069139e8cd3SQu Wenruo ASSERT(atomic_read(&eb->refs)); 107049d0c642SFilipe Manana btrfs_assert_tree_write_locked(eb); 1071139e8cd3SQu Wenruo free_extent_buffer(eb); 1072139e8cd3SQu Wenruo 1073139e8cd3SQu Wenruo cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits); 1074139e8cd3SQu Wenruo } 10750079c3b1SMatthew Wilcox (Oracle) return filemap_dirty_folio(mapping, folio); 10760b32f4bbSJosef Bacik } 10770079c3b1SMatthew Wilcox (Oracle) #else 10780079c3b1SMatthew Wilcox (Oracle) #define btree_dirty_folio filemap_dirty_folio 10790079c3b1SMatthew Wilcox (Oracle) #endif 10800b32f4bbSJosef Bacik 10817f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 10820da5468fSChris Mason .writepages = btree_writepages, 10835f39d397SChris Mason .releasepage = btree_releasepage, 1084895586ebSMatthew Wilcox (Oracle) .invalidate_folio = btree_invalidate_folio, 10855a92bc88SChris Mason #ifdef CONFIG_MIGRATION 1086784b4e29SChris Mason .migratepage = btree_migratepage, 10875a92bc88SChris Mason #endif 10880079c3b1SMatthew Wilcox (Oracle) .dirty_folio = btree_dirty_folio, 1089d98237b3SChris Mason }; 1090123abc88SChris Mason 10912ff7e61eSJeff Mahoney struct extent_buffer *btrfs_find_create_tree_block( 10922ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info, 10933fbaf258SJosef Bacik u64 bytenr, u64 owner_root, 10943fbaf258SJosef Bacik int level) 10950999df54SChris Mason { 10960b246afaSJeff Mahoney if (btrfs_is_testing(fs_info)) 10970b246afaSJeff Mahoney return alloc_test_extent_buffer(fs_info, bytenr); 10983fbaf258SJosef Bacik return alloc_extent_buffer(fs_info, bytenr, owner_root, level); 10990999df54SChris Mason } 11000999df54SChris Mason 1101581c1760SQu Wenruo /* 1102581c1760SQu Wenruo * Read tree block at logical address @bytenr and do variant basic but critical 1103581c1760SQu Wenruo * verification. 1104581c1760SQu Wenruo * 11051b7ec85eSJosef Bacik * @owner_root: the objectid of the root owner for this block. 1106581c1760SQu Wenruo * @parent_transid: expected transid of this tree block, skip check if 0 1107581c1760SQu Wenruo * @level: expected level, mandatory check 1108581c1760SQu Wenruo * @first_key: expected key in slot 0, skip check if NULL 1109581c1760SQu Wenruo */ 11102ff7e61eSJeff Mahoney struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, 11111b7ec85eSJosef Bacik u64 owner_root, u64 parent_transid, 11121b7ec85eSJosef Bacik int level, struct btrfs_key *first_key) 1113e20d96d6SChris Mason { 11145f39d397SChris Mason struct extent_buffer *buf = NULL; 111519c00ddcSChris Mason int ret; 111619c00ddcSChris Mason 11173fbaf258SJosef Bacik buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level); 1118c871b0f2SLiu Bo if (IS_ERR(buf)) 1119c871b0f2SLiu Bo return buf; 1120e4204dedSChris Mason 1121*6a2e9dc4SFilipe Manana ret = btrfs_read_extent_buffer(buf, parent_transid, level, first_key); 11220f0fe8f7SFilipe David Borba Manana if (ret) { 1123537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 112464c043deSLiu Bo return ERR_PTR(ret); 11250f0fe8f7SFilipe David Borba Manana } 11265f39d397SChris Mason return buf; 1127ce9adaa5SChris Mason 1128eb60ceacSChris Mason } 1129eb60ceacSChris Mason 11306a884d7dSDavid Sterba void btrfs_clean_tree_block(struct extent_buffer *buf) 1131ed2ff2cbSChris Mason { 11326a884d7dSDavid Sterba struct btrfs_fs_info *fs_info = buf->fs_info; 113355c69072SChris Mason if (btrfs_header_generation(buf) == 1134e2d84521SMiao Xie fs_info->running_transaction->transid) { 113549d0c642SFilipe Manana btrfs_assert_tree_write_locked(buf); 1136b4ce94deSChris Mason 1137b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1138104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 1139e2d84521SMiao Xie -buf->len, 1140e2d84521SMiao Xie fs_info->dirty_metadata_batch); 11410b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 1142925baeddSChris Mason } 11435f39d397SChris Mason } 1144ed7b63ebSJosef Bacik } 11455f39d397SChris Mason 1146da17066cSJeff Mahoney static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, 1147e20d96d6SChris Mason u64 objectid) 1148d97e63b6SChris Mason { 11497c0260eeSJeff Mahoney bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); 11502e608bd1SJosef Bacik 11512e608bd1SJosef Bacik memset(&root->root_key, 0, sizeof(root->root_key)); 11522e608bd1SJosef Bacik memset(&root->root_item, 0, sizeof(root->root_item)); 11532e608bd1SJosef Bacik memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 115496dfcb46SJosef Bacik root->fs_info = fs_info; 11552e608bd1SJosef Bacik root->root_key.objectid = objectid; 1156cfaa7295SChris Mason root->node = NULL; 1157a28ec197SChris Mason root->commit_root = NULL; 115827cdeb70SMiao Xie root->state = 0; 1159abed4aaaSJosef Bacik RB_CLEAR_NODE(&root->rb_node); 11600b86a832SChris Mason 11610f7d52f4SChris Mason root->last_trans = 0; 11626b8fad57SNikolay Borisov root->free_objectid = 0; 1163eb73c1b7SMiao Xie root->nr_delalloc_inodes = 0; 1164199c2a9cSMiao Xie root->nr_ordered_extents = 0; 11656bef4d31SEric Paris root->inode_tree = RB_ROOT; 116616cdcec7SMiao Xie INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 11672e608bd1SJosef Bacik 11682e608bd1SJosef Bacik btrfs_init_root_block_rsv(root); 11690b86a832SChris Mason 11700b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 11715d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 1172eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_inodes); 1173eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_root); 1174199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_extents); 1175199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_root); 1176d2311e69SQu Wenruo INIT_LIST_HEAD(&root->reloc_dirty_list); 11772ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 11782ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 11795d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1180eb73c1b7SMiao Xie spin_lock_init(&root->delalloc_lock); 1181199c2a9cSMiao Xie spin_lock_init(&root->ordered_extent_lock); 1182f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 11832ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 11842ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 11858287475aSQu Wenruo spin_lock_init(&root->qgroup_meta_rsv_lock); 1186a2135011SChris Mason mutex_init(&root->objectid_mutex); 1187e02119d5SChris Mason mutex_init(&root->log_mutex); 118831f3d255SMiao Xie mutex_init(&root->ordered_extent_mutex); 1189573bfb72SMiao Xie mutex_init(&root->delalloc_mutex); 1190c53e9653SQu Wenruo init_waitqueue_head(&root->qgroup_flush_wait); 11917237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 11927237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 11937237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 11948b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[0]); 11958b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[1]); 11967237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 11977237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 11987237f183SYan Zheng atomic_set(&root->log_writers, 0); 11992ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 12000700cea7SElena Reshetova refcount_set(&root->refs, 1); 12018ecebf4dSRobbie Ko atomic_set(&root->snapshot_force_cow, 0); 1202eede2bf3SOmar Sandoval atomic_set(&root->nr_swapfiles, 0); 12037237f183SYan Zheng root->log_transid = 0; 1204d1433debSMiao Xie root->log_transid_committed = -1; 1205257c62e1SChris Mason root->last_log_commit = 0; 12062e608bd1SJosef Bacik root->anon_dev = 0; 1207e289f03eSFilipe Manana if (!dummy) { 120843eb5f29SQu Wenruo extent_io_tree_init(fs_info, &root->dirty_log_pages, 120943eb5f29SQu Wenruo IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL); 1210e289f03eSFilipe Manana extent_io_tree_init(fs_info, &root->log_csum_range, 1211e289f03eSFilipe Manana IO_TREE_LOG_CSUM_RANGE, NULL); 1212e289f03eSFilipe Manana } 1213017e5369SChris Mason 12145f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 1215370a11b8SQu Wenruo btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks); 1216bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 1217bd647ce3SJosef Bacik INIT_LIST_HEAD(&root->leak_list); 1218bd647ce3SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 1219bd647ce3SJosef Bacik list_add_tail(&root->leak_list, &fs_info->allocated_roots); 1220bd647ce3SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 1221bd647ce3SJosef Bacik #endif 12223768f368SChris Mason } 12233768f368SChris Mason 122474e4d827SDavid Sterba static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, 122596dfcb46SJosef Bacik u64 objectid, gfp_t flags) 12266f07e42eSAl Viro { 122774e4d827SDavid Sterba struct btrfs_root *root = kzalloc(sizeof(*root), flags); 12286f07e42eSAl Viro if (root) 122996dfcb46SJosef Bacik __setup_root(root, fs_info, objectid); 12306f07e42eSAl Viro return root; 12316f07e42eSAl Viro } 12326f07e42eSAl Viro 123306ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 123406ea65a3SJosef Bacik /* Should only be used by the testing infrastructure */ 1235da17066cSJeff Mahoney struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) 123606ea65a3SJosef Bacik { 123706ea65a3SJosef Bacik struct btrfs_root *root; 123806ea65a3SJosef Bacik 12397c0260eeSJeff Mahoney if (!fs_info) 12407c0260eeSJeff Mahoney return ERR_PTR(-EINVAL); 12417c0260eeSJeff Mahoney 124296dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL); 124306ea65a3SJosef Bacik if (!root) 124406ea65a3SJosef Bacik return ERR_PTR(-ENOMEM); 1245da17066cSJeff Mahoney 1246b9ef22deSFeifei Xu /* We don't use the stripesize in selftest, set it as sectorsize */ 1247faa2dbf0SJosef Bacik root->alloc_bytenr = 0; 124806ea65a3SJosef Bacik 124906ea65a3SJosef Bacik return root; 125006ea65a3SJosef Bacik } 125106ea65a3SJosef Bacik #endif 125206ea65a3SJosef Bacik 1253abed4aaaSJosef Bacik static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node) 1254abed4aaaSJosef Bacik { 1255abed4aaaSJosef Bacik const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node); 1256abed4aaaSJosef Bacik const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node); 1257abed4aaaSJosef Bacik 1258abed4aaaSJosef Bacik return btrfs_comp_cpu_keys(&a->root_key, &b->root_key); 1259abed4aaaSJosef Bacik } 1260abed4aaaSJosef Bacik 1261abed4aaaSJosef Bacik static int global_root_key_cmp(const void *k, const struct rb_node *node) 1262abed4aaaSJosef Bacik { 1263abed4aaaSJosef Bacik const struct btrfs_key *key = k; 1264abed4aaaSJosef Bacik const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node); 1265abed4aaaSJosef Bacik 1266abed4aaaSJosef Bacik return btrfs_comp_cpu_keys(key, &root->root_key); 1267abed4aaaSJosef Bacik } 1268abed4aaaSJosef Bacik 1269abed4aaaSJosef Bacik int btrfs_global_root_insert(struct btrfs_root *root) 1270abed4aaaSJosef Bacik { 1271abed4aaaSJosef Bacik struct btrfs_fs_info *fs_info = root->fs_info; 1272abed4aaaSJosef Bacik struct rb_node *tmp; 1273abed4aaaSJosef Bacik 1274abed4aaaSJosef Bacik write_lock(&fs_info->global_root_lock); 1275abed4aaaSJosef Bacik tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp); 1276abed4aaaSJosef Bacik write_unlock(&fs_info->global_root_lock); 1277abed4aaaSJosef Bacik ASSERT(!tmp); 1278abed4aaaSJosef Bacik 1279abed4aaaSJosef Bacik return tmp ? -EEXIST : 0; 1280abed4aaaSJosef Bacik } 1281abed4aaaSJosef Bacik 1282abed4aaaSJosef Bacik void btrfs_global_root_delete(struct btrfs_root *root) 1283abed4aaaSJosef Bacik { 1284abed4aaaSJosef Bacik struct btrfs_fs_info *fs_info = root->fs_info; 1285abed4aaaSJosef Bacik 1286abed4aaaSJosef Bacik write_lock(&fs_info->global_root_lock); 1287abed4aaaSJosef Bacik rb_erase(&root->rb_node, &fs_info->global_root_tree); 1288abed4aaaSJosef Bacik write_unlock(&fs_info->global_root_lock); 1289abed4aaaSJosef Bacik } 1290abed4aaaSJosef Bacik 1291abed4aaaSJosef Bacik struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info, 1292abed4aaaSJosef Bacik struct btrfs_key *key) 1293abed4aaaSJosef Bacik { 1294abed4aaaSJosef Bacik struct rb_node *node; 1295abed4aaaSJosef Bacik struct btrfs_root *root = NULL; 1296abed4aaaSJosef Bacik 1297abed4aaaSJosef Bacik read_lock(&fs_info->global_root_lock); 1298abed4aaaSJosef Bacik node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp); 1299abed4aaaSJosef Bacik if (node) 1300abed4aaaSJosef Bacik root = container_of(node, struct btrfs_root, rb_node); 1301abed4aaaSJosef Bacik read_unlock(&fs_info->global_root_lock); 1302abed4aaaSJosef Bacik 1303abed4aaaSJosef Bacik return root; 1304abed4aaaSJosef Bacik } 1305abed4aaaSJosef Bacik 1306f7238e50SJosef Bacik static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr) 1307f7238e50SJosef Bacik { 1308f7238e50SJosef Bacik struct btrfs_block_group *block_group; 1309f7238e50SJosef Bacik u64 ret; 1310f7238e50SJosef Bacik 1311f7238e50SJosef Bacik if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) 1312f7238e50SJosef Bacik return 0; 1313f7238e50SJosef Bacik 1314f7238e50SJosef Bacik if (bytenr) 1315f7238e50SJosef Bacik block_group = btrfs_lookup_block_group(fs_info, bytenr); 1316f7238e50SJosef Bacik else 1317f7238e50SJosef Bacik block_group = btrfs_lookup_first_block_group(fs_info, bytenr); 1318f7238e50SJosef Bacik ASSERT(block_group); 1319f7238e50SJosef Bacik if (!block_group) 1320f7238e50SJosef Bacik return 0; 1321f7238e50SJosef Bacik ret = block_group->global_root_id; 1322f7238e50SJosef Bacik btrfs_put_block_group(block_group); 1323f7238e50SJosef Bacik 1324f7238e50SJosef Bacik return ret; 1325f7238e50SJosef Bacik } 1326f7238e50SJosef Bacik 1327abed4aaaSJosef Bacik struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr) 1328abed4aaaSJosef Bacik { 1329abed4aaaSJosef Bacik struct btrfs_key key = { 1330abed4aaaSJosef Bacik .objectid = BTRFS_CSUM_TREE_OBJECTID, 1331abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 1332f7238e50SJosef Bacik .offset = btrfs_global_root_id(fs_info, bytenr), 1333abed4aaaSJosef Bacik }; 1334abed4aaaSJosef Bacik 1335abed4aaaSJosef Bacik return btrfs_global_root(fs_info, &key); 1336abed4aaaSJosef Bacik } 1337abed4aaaSJosef Bacik 1338abed4aaaSJosef Bacik struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr) 1339abed4aaaSJosef Bacik { 1340abed4aaaSJosef Bacik struct btrfs_key key = { 1341abed4aaaSJosef Bacik .objectid = BTRFS_EXTENT_TREE_OBJECTID, 1342abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 1343f7238e50SJosef Bacik .offset = btrfs_global_root_id(fs_info, bytenr), 1344abed4aaaSJosef Bacik }; 1345abed4aaaSJosef Bacik 1346abed4aaaSJosef Bacik return btrfs_global_root(fs_info, &key); 1347abed4aaaSJosef Bacik } 1348abed4aaaSJosef Bacik 134920897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 135020897f5cSArne Jansen u64 objectid) 135120897f5cSArne Jansen { 13529b7a2440SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 135320897f5cSArne Jansen struct extent_buffer *leaf; 135420897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 135520897f5cSArne Jansen struct btrfs_root *root; 135620897f5cSArne Jansen struct btrfs_key key; 1357b89f6d1fSFilipe Manana unsigned int nofs_flag; 135820897f5cSArne Jansen int ret = 0; 135920897f5cSArne Jansen 1360b89f6d1fSFilipe Manana /* 1361b89f6d1fSFilipe Manana * We're holding a transaction handle, so use a NOFS memory allocation 1362b89f6d1fSFilipe Manana * context to avoid deadlock if reclaim happens. 1363b89f6d1fSFilipe Manana */ 1364b89f6d1fSFilipe Manana nofs_flag = memalloc_nofs_save(); 136596dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL); 1366b89f6d1fSFilipe Manana memalloc_nofs_restore(nofs_flag); 136720897f5cSArne Jansen if (!root) 136820897f5cSArne Jansen return ERR_PTR(-ENOMEM); 136920897f5cSArne Jansen 137020897f5cSArne Jansen root->root_key.objectid = objectid; 137120897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 137220897f5cSArne Jansen root->root_key.offset = 0; 137320897f5cSArne Jansen 13749631e4ccSJosef Bacik leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0, 13759631e4ccSJosef Bacik BTRFS_NESTING_NORMAL); 137620897f5cSArne Jansen if (IS_ERR(leaf)) { 137720897f5cSArne Jansen ret = PTR_ERR(leaf); 13781dd05682STsutomu Itoh leaf = NULL; 13798a6a87cdSBoris Burkov goto fail_unlock; 138020897f5cSArne Jansen } 138120897f5cSArne Jansen 138220897f5cSArne Jansen root->node = leaf; 138320897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 138420897f5cSArne Jansen 138520897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 138627cdeb70SMiao Xie set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 138720897f5cSArne Jansen 1388f944d2cbSDavid Sterba btrfs_set_root_flags(&root->root_item, 0); 1389f944d2cbSDavid Sterba btrfs_set_root_limit(&root->root_item, 0); 139020897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 139120897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 139220897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 139320897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 139420897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 139520897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 139620897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 139733d85fdaSQu Wenruo if (is_fstree(objectid)) 1398807fc790SAndy Shevchenko generate_random_guid(root->root_item.uuid); 1399807fc790SAndy Shevchenko else 1400807fc790SAndy Shevchenko export_guid(root->root_item.uuid, &guid_null); 1401c8422684SDavid Sterba btrfs_set_root_drop_level(&root->root_item, 0); 140220897f5cSArne Jansen 14038a6a87cdSBoris Burkov btrfs_tree_unlock(leaf); 14048a6a87cdSBoris Burkov 140520897f5cSArne Jansen key.objectid = objectid; 140620897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 140720897f5cSArne Jansen key.offset = 0; 140820897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 140920897f5cSArne Jansen if (ret) 141020897f5cSArne Jansen goto fail; 141120897f5cSArne Jansen 141220897f5cSArne Jansen return root; 14131dd05682STsutomu Itoh 14148a6a87cdSBoris Burkov fail_unlock: 14158c38938cSJosef Bacik if (leaf) 14161dd05682STsutomu Itoh btrfs_tree_unlock(leaf); 14178a6a87cdSBoris Burkov fail: 141800246528SJosef Bacik btrfs_put_root(root); 14191dd05682STsutomu Itoh 14201dd05682STsutomu Itoh return ERR_PTR(ret); 142120897f5cSArne Jansen } 142220897f5cSArne Jansen 14237237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1424e02119d5SChris Mason struct btrfs_fs_info *fs_info) 14250f7d52f4SChris Mason { 14260f7d52f4SChris Mason struct btrfs_root *root; 1427e02119d5SChris Mason 142896dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS); 1429e02119d5SChris Mason if (!root) 14307237f183SYan Zheng return ERR_PTR(-ENOMEM); 1431e02119d5SChris Mason 1432e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1433e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1434e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 143527cdeb70SMiao Xie 14366ab6ebb7SNaohiro Aota return root; 14376ab6ebb7SNaohiro Aota } 14386ab6ebb7SNaohiro Aota 14396ab6ebb7SNaohiro Aota int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans, 14406ab6ebb7SNaohiro Aota struct btrfs_root *root) 14416ab6ebb7SNaohiro Aota { 14426ab6ebb7SNaohiro Aota struct extent_buffer *leaf; 14436ab6ebb7SNaohiro Aota 14447237f183SYan Zheng /* 144592a7cc42SQu Wenruo * DON'T set SHAREABLE bit for log trees. 144627cdeb70SMiao Xie * 144792a7cc42SQu Wenruo * Log trees are not exposed to user space thus can't be snapshotted, 144892a7cc42SQu Wenruo * and they go away before a real commit is actually done. 144992a7cc42SQu Wenruo * 145092a7cc42SQu Wenruo * They do store pointers to file data extents, and those reference 145192a7cc42SQu Wenruo * counts still get updated (along with back refs to the log tree). 14527237f183SYan Zheng */ 1453e02119d5SChris Mason 14544d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 14559631e4ccSJosef Bacik NULL, 0, 0, 0, BTRFS_NESTING_NORMAL); 14566ab6ebb7SNaohiro Aota if (IS_ERR(leaf)) 14576ab6ebb7SNaohiro Aota return PTR_ERR(leaf); 1458e02119d5SChris Mason 14597237f183SYan Zheng root->node = leaf; 1460e02119d5SChris Mason 1461e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1462e02119d5SChris Mason btrfs_tree_unlock(root->node); 14636ab6ebb7SNaohiro Aota 14646ab6ebb7SNaohiro Aota return 0; 14657237f183SYan Zheng } 14667237f183SYan Zheng 14677237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 14687237f183SYan Zheng struct btrfs_fs_info *fs_info) 14697237f183SYan Zheng { 14707237f183SYan Zheng struct btrfs_root *log_root; 14717237f183SYan Zheng 14727237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 14737237f183SYan Zheng if (IS_ERR(log_root)) 14747237f183SYan Zheng return PTR_ERR(log_root); 14756ab6ebb7SNaohiro Aota 14763ddebf27SNaohiro Aota if (!btrfs_is_zoned(fs_info)) { 14773ddebf27SNaohiro Aota int ret = btrfs_alloc_log_tree_node(trans, log_root); 14783ddebf27SNaohiro Aota 14796ab6ebb7SNaohiro Aota if (ret) { 14806ab6ebb7SNaohiro Aota btrfs_put_root(log_root); 14816ab6ebb7SNaohiro Aota return ret; 14826ab6ebb7SNaohiro Aota } 14833ddebf27SNaohiro Aota } 14846ab6ebb7SNaohiro Aota 14857237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 14867237f183SYan Zheng fs_info->log_root_tree = log_root; 14877237f183SYan Zheng return 0; 14887237f183SYan Zheng } 14897237f183SYan Zheng 14907237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 14917237f183SYan Zheng struct btrfs_root *root) 14927237f183SYan Zheng { 14930b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 14947237f183SYan Zheng struct btrfs_root *log_root; 14957237f183SYan Zheng struct btrfs_inode_item *inode_item; 14966ab6ebb7SNaohiro Aota int ret; 14977237f183SYan Zheng 14980b246afaSJeff Mahoney log_root = alloc_log_tree(trans, fs_info); 14997237f183SYan Zheng if (IS_ERR(log_root)) 15007237f183SYan Zheng return PTR_ERR(log_root); 15017237f183SYan Zheng 15026ab6ebb7SNaohiro Aota ret = btrfs_alloc_log_tree_node(trans, log_root); 15036ab6ebb7SNaohiro Aota if (ret) { 15046ab6ebb7SNaohiro Aota btrfs_put_root(log_root); 15056ab6ebb7SNaohiro Aota return ret; 15066ab6ebb7SNaohiro Aota } 15076ab6ebb7SNaohiro Aota 15087237f183SYan Zheng log_root->last_trans = trans->transid; 15097237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 15107237f183SYan Zheng 15117237f183SYan Zheng inode_item = &log_root->root_item.inode; 15123cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 15133cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 15143cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 1515da17066cSJeff Mahoney btrfs_set_stack_inode_nbytes(inode_item, 15160b246afaSJeff Mahoney fs_info->nodesize); 15173cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 15187237f183SYan Zheng 15195d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 15207237f183SYan Zheng 15217237f183SYan Zheng WARN_ON(root->log_root); 15227237f183SYan Zheng root->log_root = log_root; 15237237f183SYan Zheng root->log_transid = 0; 1524d1433debSMiao Xie root->log_transid_committed = -1; 1525257c62e1SChris Mason root->last_log_commit = 0; 1526e02119d5SChris Mason return 0; 1527e02119d5SChris Mason } 1528e02119d5SChris Mason 152949d11beaSJosef Bacik static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root, 153049d11beaSJosef Bacik struct btrfs_path *path, 1531cb517eabSMiao Xie struct btrfs_key *key) 1532e02119d5SChris Mason { 1533e02119d5SChris Mason struct btrfs_root *root; 1534e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 153584234f3aSYan Zheng u64 generation; 1536cb517eabSMiao Xie int ret; 1537581c1760SQu Wenruo int level; 1538cb517eabSMiao Xie 153996dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS); 154049d11beaSJosef Bacik if (!root) 154149d11beaSJosef Bacik return ERR_PTR(-ENOMEM); 15420f7d52f4SChris Mason 1543cb517eabSMiao Xie ret = btrfs_find_root(tree_root, key, path, 1544cb517eabSMiao Xie &root->root_item, &root->root_key); 15450f7d52f4SChris Mason if (ret) { 154613a8a7c8SYan, Zheng if (ret > 0) 154713a8a7c8SYan, Zheng ret = -ENOENT; 154849d11beaSJosef Bacik goto fail; 15490f7d52f4SChris Mason } 155013a8a7c8SYan, Zheng 155184234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1552581c1760SQu Wenruo level = btrfs_root_level(&root->root_item); 15532ff7e61eSJeff Mahoney root->node = read_tree_block(fs_info, 15542ff7e61eSJeff Mahoney btrfs_root_bytenr(&root->root_item), 15551b7ec85eSJosef Bacik key->objectid, generation, level, NULL); 155664c043deSLiu Bo if (IS_ERR(root->node)) { 155764c043deSLiu Bo ret = PTR_ERR(root->node); 15588c38938cSJosef Bacik root->node = NULL; 155949d11beaSJosef Bacik goto fail; 15604eb150d6SQu Wenruo } 15614eb150d6SQu Wenruo if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1562cb517eabSMiao Xie ret = -EIO; 156349d11beaSJosef Bacik goto fail; 1564416bc658SJosef Bacik } 15655d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 1566cb517eabSMiao Xie return root; 156749d11beaSJosef Bacik fail: 156800246528SJosef Bacik btrfs_put_root(root); 156949d11beaSJosef Bacik return ERR_PTR(ret); 157049d11beaSJosef Bacik } 157149d11beaSJosef Bacik 157249d11beaSJosef Bacik struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 157349d11beaSJosef Bacik struct btrfs_key *key) 157449d11beaSJosef Bacik { 157549d11beaSJosef Bacik struct btrfs_root *root; 157649d11beaSJosef Bacik struct btrfs_path *path; 157749d11beaSJosef Bacik 157849d11beaSJosef Bacik path = btrfs_alloc_path(); 157949d11beaSJosef Bacik if (!path) 158049d11beaSJosef Bacik return ERR_PTR(-ENOMEM); 158149d11beaSJosef Bacik root = read_tree_root_path(tree_root, path, key); 158249d11beaSJosef Bacik btrfs_free_path(path); 158349d11beaSJosef Bacik 158449d11beaSJosef Bacik return root; 1585cb517eabSMiao Xie } 1586cb517eabSMiao Xie 15872dfb1e43SQu Wenruo /* 15882dfb1e43SQu Wenruo * Initialize subvolume root in-memory structure 15892dfb1e43SQu Wenruo * 15902dfb1e43SQu Wenruo * @anon_dev: anonymous device to attach to the root, if zero, allocate new 15912dfb1e43SQu Wenruo */ 15922dfb1e43SQu Wenruo static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev) 1593cb517eabSMiao Xie { 1594cb517eabSMiao Xie int ret; 1595dcc3eb96SNikolay Borisov unsigned int nofs_flag; 1596cb517eabSMiao Xie 1597dcc3eb96SNikolay Borisov /* 1598dcc3eb96SNikolay Borisov * We might be called under a transaction (e.g. indirect backref 1599dcc3eb96SNikolay Borisov * resolution) which could deadlock if it triggers memory reclaim 1600dcc3eb96SNikolay Borisov */ 1601dcc3eb96SNikolay Borisov nofs_flag = memalloc_nofs_save(); 1602dcc3eb96SNikolay Borisov ret = btrfs_drew_lock_init(&root->snapshot_lock); 1603dcc3eb96SNikolay Borisov memalloc_nofs_restore(nofs_flag); 1604dcc3eb96SNikolay Borisov if (ret) 16058257b2dcSMiao Xie goto fail; 16068257b2dcSMiao Xie 1607aeb935a4SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID && 160837f00a6dSJohannes Thumshirn !btrfs_is_data_reloc_root(root)) { 160992a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &root->state); 1610f39e4571SJosef Bacik btrfs_check_and_init_root_item(&root->root_item); 1611f39e4571SJosef Bacik } 1612f39e4571SJosef Bacik 1613851fd730SQu Wenruo /* 1614851fd730SQu Wenruo * Don't assign anonymous block device to roots that are not exposed to 1615851fd730SQu Wenruo * userspace, the id pool is limited to 1M 1616851fd730SQu Wenruo */ 1617851fd730SQu Wenruo if (is_fstree(root->root_key.objectid) && 1618851fd730SQu Wenruo btrfs_root_refs(&root->root_item) > 0) { 16192dfb1e43SQu Wenruo if (!anon_dev) { 1620cb517eabSMiao Xie ret = get_anon_bdev(&root->anon_dev); 1621cb517eabSMiao Xie if (ret) 1622876d2cf1SLiu Bo goto fail; 16232dfb1e43SQu Wenruo } else { 16242dfb1e43SQu Wenruo root->anon_dev = anon_dev; 16252dfb1e43SQu Wenruo } 1626851fd730SQu Wenruo } 1627f32e48e9SChandan Rajendra 1628f32e48e9SChandan Rajendra mutex_lock(&root->objectid_mutex); 1629453e4873SNikolay Borisov ret = btrfs_init_root_free_objectid(root); 1630f32e48e9SChandan Rajendra if (ret) { 1631f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1632876d2cf1SLiu Bo goto fail; 1633f32e48e9SChandan Rajendra } 1634f32e48e9SChandan Rajendra 16356b8fad57SNikolay Borisov ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID); 1636f32e48e9SChandan Rajendra 1637f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1638f32e48e9SChandan Rajendra 1639cb517eabSMiao Xie return 0; 1640cb517eabSMiao Xie fail: 164184db5ccfSDavid Sterba /* The caller is responsible to call btrfs_free_fs_root */ 1642cb517eabSMiao Xie return ret; 1643cb517eabSMiao Xie } 1644cb517eabSMiao Xie 1645a98db0f3SJosef Bacik static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1646cb517eabSMiao Xie u64 root_id) 1647cb517eabSMiao Xie { 1648cb517eabSMiao Xie struct btrfs_root *root; 1649cb517eabSMiao Xie 1650cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1651cb517eabSMiao Xie root = radix_tree_lookup(&fs_info->fs_roots_radix, 1652cb517eabSMiao Xie (unsigned long)root_id); 1653bc44d7c4SJosef Bacik if (root) 165400246528SJosef Bacik root = btrfs_grab_root(root); 1655cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1656cb517eabSMiao Xie return root; 1657cb517eabSMiao Xie } 1658cb517eabSMiao Xie 165949d11beaSJosef Bacik static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info, 166049d11beaSJosef Bacik u64 objectid) 166149d11beaSJosef Bacik { 1662abed4aaaSJosef Bacik struct btrfs_key key = { 1663abed4aaaSJosef Bacik .objectid = objectid, 1664abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 1665abed4aaaSJosef Bacik .offset = 0, 1666abed4aaaSJosef Bacik }; 1667abed4aaaSJosef Bacik 166849d11beaSJosef Bacik if (objectid == BTRFS_ROOT_TREE_OBJECTID) 166949d11beaSJosef Bacik return btrfs_grab_root(fs_info->tree_root); 167049d11beaSJosef Bacik if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 1671abed4aaaSJosef Bacik return btrfs_grab_root(btrfs_global_root(fs_info, &key)); 167249d11beaSJosef Bacik if (objectid == BTRFS_CHUNK_TREE_OBJECTID) 167349d11beaSJosef Bacik return btrfs_grab_root(fs_info->chunk_root); 167449d11beaSJosef Bacik if (objectid == BTRFS_DEV_TREE_OBJECTID) 167549d11beaSJosef Bacik return btrfs_grab_root(fs_info->dev_root); 167649d11beaSJosef Bacik if (objectid == BTRFS_CSUM_TREE_OBJECTID) 1677abed4aaaSJosef Bacik return btrfs_grab_root(btrfs_global_root(fs_info, &key)); 167849d11beaSJosef Bacik if (objectid == BTRFS_QUOTA_TREE_OBJECTID) 167949d11beaSJosef Bacik return btrfs_grab_root(fs_info->quota_root) ? 168049d11beaSJosef Bacik fs_info->quota_root : ERR_PTR(-ENOENT); 168149d11beaSJosef Bacik if (objectid == BTRFS_UUID_TREE_OBJECTID) 168249d11beaSJosef Bacik return btrfs_grab_root(fs_info->uuid_root) ? 168349d11beaSJosef Bacik fs_info->uuid_root : ERR_PTR(-ENOENT); 1684abed4aaaSJosef Bacik if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) { 1685abed4aaaSJosef Bacik struct btrfs_root *root = btrfs_global_root(fs_info, &key); 1686abed4aaaSJosef Bacik 1687abed4aaaSJosef Bacik return btrfs_grab_root(root) ? root : ERR_PTR(-ENOENT); 1688abed4aaaSJosef Bacik } 168949d11beaSJosef Bacik return NULL; 169049d11beaSJosef Bacik } 169149d11beaSJosef Bacik 1692cb517eabSMiao Xie int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1693cb517eabSMiao Xie struct btrfs_root *root) 1694cb517eabSMiao Xie { 1695cb517eabSMiao Xie int ret; 1696cb517eabSMiao Xie 1697e1860a77SDavid Sterba ret = radix_tree_preload(GFP_NOFS); 1698cb517eabSMiao Xie if (ret) 1699cb517eabSMiao Xie return ret; 1700cb517eabSMiao Xie 1701cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1702cb517eabSMiao Xie ret = radix_tree_insert(&fs_info->fs_roots_radix, 1703cb517eabSMiao Xie (unsigned long)root->root_key.objectid, 1704cb517eabSMiao Xie root); 1705af01d2e5SJosef Bacik if (ret == 0) { 170600246528SJosef Bacik btrfs_grab_root(root); 170727cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1708af01d2e5SJosef Bacik } 1709cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1710cb517eabSMiao Xie radix_tree_preload_end(); 1711cb517eabSMiao Xie 1712cb517eabSMiao Xie return ret; 1713cb517eabSMiao Xie } 1714cb517eabSMiao Xie 1715bd647ce3SJosef Bacik void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info) 1716bd647ce3SJosef Bacik { 1717bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 1718bd647ce3SJosef Bacik struct btrfs_root *root; 1719bd647ce3SJosef Bacik 1720bd647ce3SJosef Bacik while (!list_empty(&fs_info->allocated_roots)) { 1721457f1864SJosef Bacik char buf[BTRFS_ROOT_NAME_BUF_LEN]; 1722457f1864SJosef Bacik 1723bd647ce3SJosef Bacik root = list_first_entry(&fs_info->allocated_roots, 1724bd647ce3SJosef Bacik struct btrfs_root, leak_list); 1725457f1864SJosef Bacik btrfs_err(fs_info, "leaked root %s refcount %d", 172671008734SJosef Bacik btrfs_root_name(&root->root_key, buf), 1727bd647ce3SJosef Bacik refcount_read(&root->refs)); 1728bd647ce3SJosef Bacik while (refcount_read(&root->refs) > 1) 172900246528SJosef Bacik btrfs_put_root(root); 173000246528SJosef Bacik btrfs_put_root(root); 1731bd647ce3SJosef Bacik } 1732bd647ce3SJosef Bacik #endif 1733bd647ce3SJosef Bacik } 1734bd647ce3SJosef Bacik 1735abed4aaaSJosef Bacik static void free_global_roots(struct btrfs_fs_info *fs_info) 1736abed4aaaSJosef Bacik { 1737abed4aaaSJosef Bacik struct btrfs_root *root; 1738abed4aaaSJosef Bacik struct rb_node *node; 1739abed4aaaSJosef Bacik 1740abed4aaaSJosef Bacik while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) { 1741abed4aaaSJosef Bacik root = rb_entry(node, struct btrfs_root, rb_node); 1742abed4aaaSJosef Bacik rb_erase(&root->rb_node, &fs_info->global_root_tree); 1743abed4aaaSJosef Bacik btrfs_put_root(root); 1744abed4aaaSJosef Bacik } 1745abed4aaaSJosef Bacik } 1746abed4aaaSJosef Bacik 17470d4b0463SJosef Bacik void btrfs_free_fs_info(struct btrfs_fs_info *fs_info) 17480d4b0463SJosef Bacik { 1749141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 1750141386e1SJosef Bacik percpu_counter_destroy(&fs_info->delalloc_bytes); 17515deb17e1SJosef Bacik percpu_counter_destroy(&fs_info->ordered_bytes); 1752141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 1753141386e1SJosef Bacik btrfs_free_csum_hash(fs_info); 1754141386e1SJosef Bacik btrfs_free_stripe_hash_table(fs_info); 1755141386e1SJosef Bacik btrfs_free_ref_cache(fs_info); 17560d4b0463SJosef Bacik kfree(fs_info->balance_ctl); 17570d4b0463SJosef Bacik kfree(fs_info->delayed_root); 1758abed4aaaSJosef Bacik free_global_roots(fs_info); 175900246528SJosef Bacik btrfs_put_root(fs_info->tree_root); 176000246528SJosef Bacik btrfs_put_root(fs_info->chunk_root); 176100246528SJosef Bacik btrfs_put_root(fs_info->dev_root); 176200246528SJosef Bacik btrfs_put_root(fs_info->quota_root); 176300246528SJosef Bacik btrfs_put_root(fs_info->uuid_root); 176400246528SJosef Bacik btrfs_put_root(fs_info->fs_root); 1765aeb935a4SQu Wenruo btrfs_put_root(fs_info->data_reloc_root); 17669c54e80dSJosef Bacik btrfs_put_root(fs_info->block_group_root); 1767bd647ce3SJosef Bacik btrfs_check_leaked_roots(fs_info); 17683fd63727SJosef Bacik btrfs_extent_buffer_leak_debug_check(fs_info); 17690d4b0463SJosef Bacik kfree(fs_info->super_copy); 17700d4b0463SJosef Bacik kfree(fs_info->super_for_commit); 17718481dd80SQu Wenruo kfree(fs_info->subpage_info); 17720d4b0463SJosef Bacik kvfree(fs_info); 17730d4b0463SJosef Bacik } 17740d4b0463SJosef Bacik 17750d4b0463SJosef Bacik 17762dfb1e43SQu Wenruo /* 17772dfb1e43SQu Wenruo * Get an in-memory reference of a root structure. 17782dfb1e43SQu Wenruo * 17792dfb1e43SQu Wenruo * For essential trees like root/extent tree, we grab it from fs_info directly. 17802dfb1e43SQu Wenruo * For subvolume trees, we check the cached filesystem roots first. If not 17812dfb1e43SQu Wenruo * found, then read it from disk and add it to cached fs roots. 17822dfb1e43SQu Wenruo * 17832dfb1e43SQu Wenruo * Caller should release the root by calling btrfs_put_root() after the usage. 17842dfb1e43SQu Wenruo * 17852dfb1e43SQu Wenruo * NOTE: Reloc and log trees can't be read by this function as they share the 17862dfb1e43SQu Wenruo * same root objectid. 17872dfb1e43SQu Wenruo * 17882dfb1e43SQu Wenruo * @objectid: root id 17892dfb1e43SQu Wenruo * @anon_dev: preallocated anonymous block device number for new roots, 17902dfb1e43SQu Wenruo * pass 0 for new allocation. 17912dfb1e43SQu Wenruo * @check_ref: whether to check root item references, If true, return -ENOENT 17922dfb1e43SQu Wenruo * for orphan roots 17932dfb1e43SQu Wenruo */ 17942dfb1e43SQu Wenruo static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info, 17952dfb1e43SQu Wenruo u64 objectid, dev_t anon_dev, 17962dfb1e43SQu Wenruo bool check_ref) 17975eda7b5eSChris Mason { 17985eda7b5eSChris Mason struct btrfs_root *root; 1799381cf658SDavid Sterba struct btrfs_path *path; 18001d4c08e0SDavid Sterba struct btrfs_key key; 18015eda7b5eSChris Mason int ret; 18025eda7b5eSChris Mason 180349d11beaSJosef Bacik root = btrfs_get_global_root(fs_info, objectid); 180449d11beaSJosef Bacik if (root) 180549d11beaSJosef Bacik return root; 18064df27c4dSYan, Zheng again: 180756e9357aSDavid Sterba root = btrfs_lookup_fs_root(fs_info, objectid); 180848475471SStefan Behrens if (root) { 18092dfb1e43SQu Wenruo /* Shouldn't get preallocated anon_dev for cached roots */ 18102dfb1e43SQu Wenruo ASSERT(!anon_dev); 1811bc44d7c4SJosef Bacik if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 181200246528SJosef Bacik btrfs_put_root(root); 181348475471SStefan Behrens return ERR_PTR(-ENOENT); 1814bc44d7c4SJosef Bacik } 18155eda7b5eSChris Mason return root; 181648475471SStefan Behrens } 18175eda7b5eSChris Mason 181856e9357aSDavid Sterba key.objectid = objectid; 181956e9357aSDavid Sterba key.type = BTRFS_ROOT_ITEM_KEY; 182056e9357aSDavid Sterba key.offset = (u64)-1; 182156e9357aSDavid Sterba root = btrfs_read_tree_root(fs_info->tree_root, &key); 18225eda7b5eSChris Mason if (IS_ERR(root)) 18235eda7b5eSChris Mason return root; 18243394e160SChris Mason 1825c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1826d68fc57bSYan, Zheng ret = -ENOENT; 1827d68fc57bSYan, Zheng goto fail; 1828d68fc57bSYan, Zheng } 1829d68fc57bSYan, Zheng 18302dfb1e43SQu Wenruo ret = btrfs_init_fs_root(root, anon_dev); 1831cb517eabSMiao Xie if (ret) 1832cb517eabSMiao Xie goto fail; 1833cb517eabSMiao Xie 1834381cf658SDavid Sterba path = btrfs_alloc_path(); 1835381cf658SDavid Sterba if (!path) { 1836381cf658SDavid Sterba ret = -ENOMEM; 1837381cf658SDavid Sterba goto fail; 1838381cf658SDavid Sterba } 18391d4c08e0SDavid Sterba key.objectid = BTRFS_ORPHAN_OBJECTID; 18401d4c08e0SDavid Sterba key.type = BTRFS_ORPHAN_ITEM_KEY; 184156e9357aSDavid Sterba key.offset = objectid; 18421d4c08e0SDavid Sterba 18431d4c08e0SDavid Sterba ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1844381cf658SDavid Sterba btrfs_free_path(path); 1845d68fc57bSYan, Zheng if (ret < 0) 1846d68fc57bSYan, Zheng goto fail; 1847d68fc57bSYan, Zheng if (ret == 0) 184827cdeb70SMiao Xie set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1849d68fc57bSYan, Zheng 1850cb517eabSMiao Xie ret = btrfs_insert_fs_root(fs_info, root); 18510f7d52f4SChris Mason if (ret) { 1852168a2f77SJia-Ju Bai if (ret == -EEXIST) { 185300246528SJosef Bacik btrfs_put_root(root); 18544df27c4dSYan, Zheng goto again; 1855168a2f77SJia-Ju Bai } 18564df27c4dSYan, Zheng goto fail; 18574df27c4dSYan, Zheng } 1858edbd8d4eSChris Mason return root; 18594df27c4dSYan, Zheng fail: 186033fab972SFilipe Manana /* 186133fab972SFilipe Manana * If our caller provided us an anonymous device, then it's his 186233fab972SFilipe Manana * responsability to free it in case we fail. So we have to set our 186333fab972SFilipe Manana * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root() 186433fab972SFilipe Manana * and once again by our caller. 186533fab972SFilipe Manana */ 186633fab972SFilipe Manana if (anon_dev) 186733fab972SFilipe Manana root->anon_dev = 0; 18688c38938cSJosef Bacik btrfs_put_root(root); 18694df27c4dSYan, Zheng return ERR_PTR(ret); 1870edbd8d4eSChris Mason } 1871edbd8d4eSChris Mason 18722dfb1e43SQu Wenruo /* 18732dfb1e43SQu Wenruo * Get in-memory reference of a root structure 18742dfb1e43SQu Wenruo * 18752dfb1e43SQu Wenruo * @objectid: tree objectid 18762dfb1e43SQu Wenruo * @check_ref: if set, verify that the tree exists and the item has at least 18772dfb1e43SQu Wenruo * one reference 18782dfb1e43SQu Wenruo */ 18792dfb1e43SQu Wenruo struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 18802dfb1e43SQu Wenruo u64 objectid, bool check_ref) 18812dfb1e43SQu Wenruo { 18822dfb1e43SQu Wenruo return btrfs_get_root_ref(fs_info, objectid, 0, check_ref); 18832dfb1e43SQu Wenruo } 18842dfb1e43SQu Wenruo 18852dfb1e43SQu Wenruo /* 18862dfb1e43SQu Wenruo * Get in-memory reference of a root structure, created as new, optionally pass 18872dfb1e43SQu Wenruo * the anonymous block device id 18882dfb1e43SQu Wenruo * 18892dfb1e43SQu Wenruo * @objectid: tree objectid 18902dfb1e43SQu Wenruo * @anon_dev: if zero, allocate a new anonymous block device or use the 18912dfb1e43SQu Wenruo * parameter value 18922dfb1e43SQu Wenruo */ 18932dfb1e43SQu Wenruo struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info, 18942dfb1e43SQu Wenruo u64 objectid, dev_t anon_dev) 18952dfb1e43SQu Wenruo { 18962dfb1e43SQu Wenruo return btrfs_get_root_ref(fs_info, objectid, anon_dev, true); 18972dfb1e43SQu Wenruo } 18982dfb1e43SQu Wenruo 18998b712842SChris Mason /* 190049d11beaSJosef Bacik * btrfs_get_fs_root_commit_root - return a root for the given objectid 190149d11beaSJosef Bacik * @fs_info: the fs_info 190249d11beaSJosef Bacik * @objectid: the objectid we need to lookup 190349d11beaSJosef Bacik * 190449d11beaSJosef Bacik * This is exclusively used for backref walking, and exists specifically because 190549d11beaSJosef Bacik * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref 190649d11beaSJosef Bacik * creation time, which means we may have to read the tree_root in order to look 190749d11beaSJosef Bacik * up a fs root that is not in memory. If the root is not in memory we will 190849d11beaSJosef Bacik * read the tree root commit root and look up the fs root from there. This is a 190949d11beaSJosef Bacik * temporary root, it will not be inserted into the radix tree as it doesn't 191049d11beaSJosef Bacik * have the most uptodate information, it'll simply be discarded once the 191149d11beaSJosef Bacik * backref code is finished using the root. 191249d11beaSJosef Bacik */ 191349d11beaSJosef Bacik struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info, 191449d11beaSJosef Bacik struct btrfs_path *path, 191549d11beaSJosef Bacik u64 objectid) 191649d11beaSJosef Bacik { 191749d11beaSJosef Bacik struct btrfs_root *root; 191849d11beaSJosef Bacik struct btrfs_key key; 191949d11beaSJosef Bacik 192049d11beaSJosef Bacik ASSERT(path->search_commit_root && path->skip_locking); 192149d11beaSJosef Bacik 192249d11beaSJosef Bacik /* 192349d11beaSJosef Bacik * This can return -ENOENT if we ask for a root that doesn't exist, but 192449d11beaSJosef Bacik * since this is called via the backref walking code we won't be looking 192549d11beaSJosef Bacik * up a root that doesn't exist, unless there's corruption. So if root 192649d11beaSJosef Bacik * != NULL just return it. 192749d11beaSJosef Bacik */ 192849d11beaSJosef Bacik root = btrfs_get_global_root(fs_info, objectid); 192949d11beaSJosef Bacik if (root) 193049d11beaSJosef Bacik return root; 193149d11beaSJosef Bacik 193249d11beaSJosef Bacik root = btrfs_lookup_fs_root(fs_info, objectid); 193349d11beaSJosef Bacik if (root) 193449d11beaSJosef Bacik return root; 193549d11beaSJosef Bacik 193649d11beaSJosef Bacik key.objectid = objectid; 193749d11beaSJosef Bacik key.type = BTRFS_ROOT_ITEM_KEY; 193849d11beaSJosef Bacik key.offset = (u64)-1; 193949d11beaSJosef Bacik root = read_tree_root_path(fs_info->tree_root, path, &key); 194049d11beaSJosef Bacik btrfs_release_path(path); 194149d11beaSJosef Bacik 194249d11beaSJosef Bacik return root; 194349d11beaSJosef Bacik } 194449d11beaSJosef Bacik 194549d11beaSJosef Bacik /* 19468b712842SChris Mason * called by the kthread helper functions to finally call the bio end_io 19478b712842SChris Mason * functions. This is where read checksum verification actually happens 19488b712842SChris Mason */ 19498b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work) 1950ce9adaa5SChris Mason { 1951ce9adaa5SChris Mason struct bio *bio; 195297eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 1953ce9adaa5SChris Mason 195497eb6b69SDavid Sterba end_io_wq = container_of(work, struct btrfs_end_io_wq, work); 1955ce9adaa5SChris Mason bio = end_io_wq->bio; 19568b712842SChris Mason 19574e4cbee9SChristoph Hellwig bio->bi_status = end_io_wq->status; 1958ce9adaa5SChris Mason bio->bi_private = end_io_wq->private; 1959ce9adaa5SChris Mason bio->bi_end_io = end_io_wq->end_io; 19604246a0b6SChristoph Hellwig bio_endio(bio); 19619be490f1SOmar Sandoval kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq); 1962ce9adaa5SChris Mason } 196344b8bd7eSChris Mason 1964a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1965a74a4b97SChris Mason { 196633c44184SJosef Bacik struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)arg; 1967d0278245SMiao Xie int again; 1968a74a4b97SChris Mason 1969d6fd0ae2SOmar Sandoval while (1) { 1970d0278245SMiao Xie again = 0; 19719d1a2a3aSDavid Sterba 1972fd340d0fSJosef Bacik set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1973fd340d0fSJosef Bacik 1974d0278245SMiao Xie /* Make the cleaner go to sleep early. */ 19752ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) 1976d0278245SMiao Xie goto sleep; 1977d0278245SMiao Xie 197890c711abSZygo Blaxell /* 197990c711abSZygo Blaxell * Do not do anything if we might cause open_ctree() to block 198090c711abSZygo Blaxell * before we have finished mounting the filesystem. 198190c711abSZygo Blaxell */ 19820b246afaSJeff Mahoney if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 198390c711abSZygo Blaxell goto sleep; 198490c711abSZygo Blaxell 19850b246afaSJeff Mahoney if (!mutex_trylock(&fs_info->cleaner_mutex)) 1986d0278245SMiao Xie goto sleep; 1987d0278245SMiao Xie 1988dc7f370cSMiao Xie /* 1989dc7f370cSMiao Xie * Avoid the problem that we change the status of the fs 1990dc7f370cSMiao Xie * during the above check and trylock. 1991dc7f370cSMiao Xie */ 19922ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) { 19930b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1994dc7f370cSMiao Xie goto sleep; 1995dc7f370cSMiao Xie } 1996dc7f370cSMiao Xie 19972ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 1998c2d6cb16SFilipe Manana 199933c44184SJosef Bacik again = btrfs_clean_one_deleted_snapshot(fs_info); 20000b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 2001a74a4b97SChris Mason 2002d0278245SMiao Xie /* 200305323cd1SMiao Xie * The defragger has dealt with the R/O remount and umount, 200405323cd1SMiao Xie * needn't do anything special here. 2005d0278245SMiao Xie */ 20060b246afaSJeff Mahoney btrfs_run_defrag_inodes(fs_info); 200767c5e7d4SFilipe Manana 200867c5e7d4SFilipe Manana /* 2009f3372065SJohannes Thumshirn * Acquires fs_info->reclaim_bgs_lock to avoid racing 201067c5e7d4SFilipe Manana * with relocation (btrfs_relocate_chunk) and relocation 201167c5e7d4SFilipe Manana * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 2012f3372065SJohannes Thumshirn * after acquiring fs_info->reclaim_bgs_lock. So we 201367c5e7d4SFilipe Manana * can't hold, nor need to, fs_info->cleaner_mutex when deleting 201467c5e7d4SFilipe Manana * unused block groups. 201567c5e7d4SFilipe Manana */ 20160b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 201718bb8bbfSJohannes Thumshirn 201818bb8bbfSJohannes Thumshirn /* 201918bb8bbfSJohannes Thumshirn * Reclaim block groups in the reclaim_bgs list after we deleted 202018bb8bbfSJohannes Thumshirn * all unused block_groups. This possibly gives us some more free 202118bb8bbfSJohannes Thumshirn * space. 202218bb8bbfSJohannes Thumshirn */ 202318bb8bbfSJohannes Thumshirn btrfs_reclaim_bgs(fs_info); 2024d0278245SMiao Xie sleep: 2025a0a1db70SFilipe Manana clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 2026d6fd0ae2SOmar Sandoval if (kthread_should_park()) 2027d6fd0ae2SOmar Sandoval kthread_parkme(); 2028d6fd0ae2SOmar Sandoval if (kthread_should_stop()) 2029d6fd0ae2SOmar Sandoval return 0; 2030838fe188SJiri Kosina if (!again) { 2031a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 2032a74a4b97SChris Mason schedule(); 2033a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 2034a74a4b97SChris Mason } 2035da288d28SFilipe Manana } 2036a74a4b97SChris Mason } 2037a74a4b97SChris Mason 2038a74a4b97SChris Mason static int transaction_kthread(void *arg) 2039a74a4b97SChris Mason { 2040a74a4b97SChris Mason struct btrfs_root *root = arg; 20410b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 2042a74a4b97SChris Mason struct btrfs_trans_handle *trans; 2043a74a4b97SChris Mason struct btrfs_transaction *cur; 20448929ecfaSYan, Zheng u64 transid; 2045643900beSNikolay Borisov time64_t delta; 2046a74a4b97SChris Mason unsigned long delay; 2047914b2007SJan Kara bool cannot_commit; 2048a74a4b97SChris Mason 2049a74a4b97SChris Mason do { 2050914b2007SJan Kara cannot_commit = false; 2051ba1bc00fSNikolay Borisov delay = msecs_to_jiffies(fs_info->commit_interval * 1000); 20520b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 2053a74a4b97SChris Mason 20540b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 20550b246afaSJeff Mahoney cur = fs_info->running_transaction; 2056a74a4b97SChris Mason if (!cur) { 20570b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2058a74a4b97SChris Mason goto sleep; 2059a74a4b97SChris Mason } 206031153d81SYan Zheng 2061643900beSNikolay Borisov delta = ktime_get_seconds() - cur->start_time; 2062fdfbf020SJosef Bacik if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) && 2063fdfbf020SJosef Bacik cur->state < TRANS_STATE_COMMIT_START && 2064643900beSNikolay Borisov delta < fs_info->commit_interval) { 20650b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2066fb8a7e94SNikolay Borisov delay -= msecs_to_jiffies((delta - 1) * 1000); 2067fb8a7e94SNikolay Borisov delay = min(delay, 2068fb8a7e94SNikolay Borisov msecs_to_jiffies(fs_info->commit_interval * 1000)); 2069a74a4b97SChris Mason goto sleep; 2070a74a4b97SChris Mason } 20718929ecfaSYan, Zheng transid = cur->transid; 20720b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 207356bec294SChris Mason 207479787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 2075354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 2076914b2007SJan Kara if (IS_ERR(trans)) { 2077354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 2078914b2007SJan Kara cannot_commit = true; 207979787eaaSJeff Mahoney goto sleep; 2080914b2007SJan Kara } 20818929ecfaSYan, Zheng if (transid == trans->transid) { 20823a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 20838929ecfaSYan, Zheng } else { 20843a45bb20SJeff Mahoney btrfs_end_transaction(trans); 20858929ecfaSYan, Zheng } 2086a74a4b97SChris Mason sleep: 20870b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 20880b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 2089a74a4b97SChris Mason 209084961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) 20912ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 20928929ecfaSYan, Zheng if (!kthread_should_stop() && 20930b246afaSJeff Mahoney (!btrfs_transaction_blocked(fs_info) || 2094914b2007SJan Kara cannot_commit)) 2095bc5511d0SNikolay Borisov schedule_timeout_interruptible(delay); 2096a74a4b97SChris Mason } while (!kthread_should_stop()); 2097a74a4b97SChris Mason return 0; 2098a74a4b97SChris Mason } 2099a74a4b97SChris Mason 2100af31f5e5SChris Mason /* 210101f0f9daSNikolay Borisov * This will find the highest generation in the array of root backups. The 210201f0f9daSNikolay Borisov * index of the highest array is returned, or -EINVAL if we can't find 210301f0f9daSNikolay Borisov * anything. 2104af31f5e5SChris Mason * 2105af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 2106af31f5e5SChris Mason * generation of the latest root in the array with the generation 2107af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 2108af31f5e5SChris Mason */ 210901f0f9daSNikolay Borisov static int find_newest_super_backup(struct btrfs_fs_info *info) 2110af31f5e5SChris Mason { 211101f0f9daSNikolay Borisov const u64 newest_gen = btrfs_super_generation(info->super_copy); 2112af31f5e5SChris Mason u64 cur; 2113af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 2114af31f5e5SChris Mason int i; 2115af31f5e5SChris Mason 2116af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 2117af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 2118af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 2119af31f5e5SChris Mason if (cur == newest_gen) 212001f0f9daSNikolay Borisov return i; 2121af31f5e5SChris Mason } 2122af31f5e5SChris Mason 212301f0f9daSNikolay Borisov return -EINVAL; 2124af31f5e5SChris Mason } 2125af31f5e5SChris Mason 2126af31f5e5SChris Mason /* 2127af31f5e5SChris Mason * copy all the root pointers into the super backup array. 2128af31f5e5SChris Mason * this will bump the backup pointer by one when it is 2129af31f5e5SChris Mason * done 2130af31f5e5SChris Mason */ 2131af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 2132af31f5e5SChris Mason { 21336ef108ddSNikolay Borisov const int next_backup = info->backup_root_index; 2134af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 2135af31f5e5SChris Mason 2136af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 2137af31f5e5SChris Mason 2138af31f5e5SChris Mason /* 2139af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 2140af31f5e5SChris Mason * regardless of which ones we use today 2141af31f5e5SChris Mason */ 2142af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 2143af31f5e5SChris Mason 2144af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 2145af31f5e5SChris Mason 2146af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 2147af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 2148af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 2149af31f5e5SChris Mason 2150af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 2151af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 2152af31f5e5SChris Mason 2153af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 2154af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 2155af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 2156af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 2157af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 2158af31f5e5SChris Mason 21599c54e80dSJosef Bacik if (btrfs_fs_incompat(info, EXTENT_TREE_V2)) { 21609c54e80dSJosef Bacik btrfs_set_backup_block_group_root(root_backup, 21619c54e80dSJosef Bacik info->block_group_root->node->start); 21629c54e80dSJosef Bacik btrfs_set_backup_block_group_root_gen(root_backup, 21639c54e80dSJosef Bacik btrfs_header_generation(info->block_group_root->node)); 21649c54e80dSJosef Bacik btrfs_set_backup_block_group_root_level(root_backup, 21659c54e80dSJosef Bacik btrfs_header_level(info->block_group_root->node)); 21669c54e80dSJosef Bacik } else { 21679c54e80dSJosef Bacik struct btrfs_root *extent_root = btrfs_extent_root(info, 0); 2168f7238e50SJosef Bacik struct btrfs_root *csum_root = btrfs_csum_root(info, 0); 21699c54e80dSJosef Bacik 21709c54e80dSJosef Bacik btrfs_set_backup_extent_root(root_backup, 21719c54e80dSJosef Bacik extent_root->node->start); 2172af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 217329cbcf40SJosef Bacik btrfs_header_generation(extent_root->node)); 2174af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 217529cbcf40SJosef Bacik btrfs_header_level(extent_root->node)); 2176af31f5e5SChris Mason 2177f7238e50SJosef Bacik btrfs_set_backup_csum_root(root_backup, csum_root->node->start); 2178f7238e50SJosef Bacik btrfs_set_backup_csum_root_gen(root_backup, 2179f7238e50SJosef Bacik btrfs_header_generation(csum_root->node)); 2180f7238e50SJosef Bacik btrfs_set_backup_csum_root_level(root_backup, 2181f7238e50SJosef Bacik btrfs_header_level(csum_root->node)); 21829c54e80dSJosef Bacik } 2183af31f5e5SChris Mason 21847c7e82a7SChris Mason /* 21857c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 21867c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 21877c7e82a7SChris Mason */ 21887c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 21897c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 21907c7e82a7SChris Mason info->fs_root->node->start); 2191af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 2192af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 2193af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 2194af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 21957c7e82a7SChris Mason } 2196af31f5e5SChris Mason 2197af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 2198af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 2199af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 2200af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 2201af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 2202af31f5e5SChris Mason 2203af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 2204af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 2205af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 2206af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 2207af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 2208af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 2209af31f5e5SChris Mason 2210af31f5e5SChris Mason /* 2211af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 2212af31f5e5SChris Mason * for the next commit 2213af31f5e5SChris Mason */ 2214af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 2215af31f5e5SChris Mason &info->super_for_commit->super_roots, 2216af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 2217af31f5e5SChris Mason } 2218af31f5e5SChris Mason 2219af31f5e5SChris Mason /* 2220bd2336b2SNikolay Borisov * read_backup_root - Reads a backup root based on the passed priority. Prio 0 2221bd2336b2SNikolay Borisov * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots 2222bd2336b2SNikolay Borisov * 2223bd2336b2SNikolay Borisov * fs_info - filesystem whose backup roots need to be read 2224bd2336b2SNikolay Borisov * priority - priority of backup root required 2225bd2336b2SNikolay Borisov * 2226bd2336b2SNikolay Borisov * Returns backup root index on success and -EINVAL otherwise. 2227bd2336b2SNikolay Borisov */ 2228bd2336b2SNikolay Borisov static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority) 2229bd2336b2SNikolay Borisov { 2230bd2336b2SNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 2231bd2336b2SNikolay Borisov struct btrfs_super_block *super = fs_info->super_copy; 2232bd2336b2SNikolay Borisov struct btrfs_root_backup *root_backup; 2233bd2336b2SNikolay Borisov 2234bd2336b2SNikolay Borisov if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) { 2235bd2336b2SNikolay Borisov if (priority == 0) 2236bd2336b2SNikolay Borisov return backup_index; 2237bd2336b2SNikolay Borisov 2238bd2336b2SNikolay Borisov backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority; 2239bd2336b2SNikolay Borisov backup_index %= BTRFS_NUM_BACKUP_ROOTS; 2240bd2336b2SNikolay Borisov } else { 2241bd2336b2SNikolay Borisov return -EINVAL; 2242bd2336b2SNikolay Borisov } 2243bd2336b2SNikolay Borisov 2244bd2336b2SNikolay Borisov root_backup = super->super_roots + backup_index; 2245bd2336b2SNikolay Borisov 2246bd2336b2SNikolay Borisov btrfs_set_super_generation(super, 2247bd2336b2SNikolay Borisov btrfs_backup_tree_root_gen(root_backup)); 2248bd2336b2SNikolay Borisov btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 2249bd2336b2SNikolay Borisov btrfs_set_super_root_level(super, 2250bd2336b2SNikolay Borisov btrfs_backup_tree_root_level(root_backup)); 2251bd2336b2SNikolay Borisov btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 2252bd2336b2SNikolay Borisov 2253bd2336b2SNikolay Borisov /* 2254bd2336b2SNikolay Borisov * Fixme: the total bytes and num_devices need to match or we should 2255bd2336b2SNikolay Borisov * need a fsck 2256bd2336b2SNikolay Borisov */ 2257bd2336b2SNikolay Borisov btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 2258bd2336b2SNikolay Borisov btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 2259bd2336b2SNikolay Borisov 2260bd2336b2SNikolay Borisov return backup_index; 2261bd2336b2SNikolay Borisov } 2262bd2336b2SNikolay Borisov 22637abadb64SLiu Bo /* helper to cleanup workers */ 22647abadb64SLiu Bo static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 22657abadb64SLiu Bo { 2266dc6e3209SQu Wenruo btrfs_destroy_workqueue(fs_info->fixup_workers); 2267afe3d242SQu Wenruo btrfs_destroy_workqueue(fs_info->delalloc_workers); 22685cdc7ad3SQu Wenruo btrfs_destroy_workqueue(fs_info->workers); 2269fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_workers); 2270fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_raid56_workers); 2271d05a33acSQu Wenruo btrfs_destroy_workqueue(fs_info->rmw_workers); 2272fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_write_workers); 2273fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 22745b3bc44eSQu Wenruo btrfs_destroy_workqueue(fs_info->delayed_workers); 2275e66f0bb1SQu Wenruo btrfs_destroy_workqueue(fs_info->caching_workers); 2276a44903abSQu Wenruo btrfs_destroy_workqueue(fs_info->flush_workers); 2277fc97fab0SQu Wenruo btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2278b0643e59SDennis Zhou if (fs_info->discard_ctl.discard_workers) 2279b0643e59SDennis Zhou destroy_workqueue(fs_info->discard_ctl.discard_workers); 2280a9b9477dSFilipe Manana /* 2281a9b9477dSFilipe Manana * Now that all other work queues are destroyed, we can safely destroy 2282a9b9477dSFilipe Manana * the queues used for metadata I/O, since tasks from those other work 2283a9b9477dSFilipe Manana * queues can do metadata I/O operations. 2284a9b9477dSFilipe Manana */ 2285a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_workers); 2286a9b9477dSFilipe Manana btrfs_destroy_workqueue(fs_info->endio_meta_write_workers); 22877abadb64SLiu Bo } 22887abadb64SLiu Bo 22892e9f5954SRashika static void free_root_extent_buffers(struct btrfs_root *root) 22902e9f5954SRashika { 22912e9f5954SRashika if (root) { 22922e9f5954SRashika free_extent_buffer(root->node); 22932e9f5954SRashika free_extent_buffer(root->commit_root); 22942e9f5954SRashika root->node = NULL; 22952e9f5954SRashika root->commit_root = NULL; 22962e9f5954SRashika } 22972e9f5954SRashika } 22982e9f5954SRashika 2299abed4aaaSJosef Bacik static void free_global_root_pointers(struct btrfs_fs_info *fs_info) 2300abed4aaaSJosef Bacik { 2301abed4aaaSJosef Bacik struct btrfs_root *root, *tmp; 2302abed4aaaSJosef Bacik 2303abed4aaaSJosef Bacik rbtree_postorder_for_each_entry_safe(root, tmp, 2304abed4aaaSJosef Bacik &fs_info->global_root_tree, 2305abed4aaaSJosef Bacik rb_node) 2306abed4aaaSJosef Bacik free_root_extent_buffers(root); 2307abed4aaaSJosef Bacik } 2308abed4aaaSJosef Bacik 2309af31f5e5SChris Mason /* helper to cleanup tree roots */ 23104273eaffSAnand Jain static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root) 2311af31f5e5SChris Mason { 23122e9f5954SRashika free_root_extent_buffers(info->tree_root); 2313655b09feSJosef Bacik 2314abed4aaaSJosef Bacik free_global_root_pointers(info); 23152e9f5954SRashika free_root_extent_buffers(info->dev_root); 23162e9f5954SRashika free_root_extent_buffers(info->quota_root); 23172e9f5954SRashika free_root_extent_buffers(info->uuid_root); 23188c38938cSJosef Bacik free_root_extent_buffers(info->fs_root); 2319aeb935a4SQu Wenruo free_root_extent_buffers(info->data_reloc_root); 23209c54e80dSJosef Bacik free_root_extent_buffers(info->block_group_root); 23214273eaffSAnand Jain if (free_chunk_root) 23222e9f5954SRashika free_root_extent_buffers(info->chunk_root); 2323af31f5e5SChris Mason } 2324af31f5e5SChris Mason 23258c38938cSJosef Bacik void btrfs_put_root(struct btrfs_root *root) 23268c38938cSJosef Bacik { 23278c38938cSJosef Bacik if (!root) 23288c38938cSJosef Bacik return; 23298c38938cSJosef Bacik 23308c38938cSJosef Bacik if (refcount_dec_and_test(&root->refs)) { 23318c38938cSJosef Bacik WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 23321dae7e0eSQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)); 23338c38938cSJosef Bacik if (root->anon_dev) 23348c38938cSJosef Bacik free_anon_bdev(root->anon_dev); 23358c38938cSJosef Bacik btrfs_drew_lock_destroy(&root->snapshot_lock); 2336923eb523SJohannes Thumshirn free_root_extent_buffers(root); 23378c38938cSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 23388c38938cSJosef Bacik spin_lock(&root->fs_info->fs_roots_radix_lock); 23398c38938cSJosef Bacik list_del_init(&root->leak_list); 23408c38938cSJosef Bacik spin_unlock(&root->fs_info->fs_roots_radix_lock); 23418c38938cSJosef Bacik #endif 23428c38938cSJosef Bacik kfree(root); 23438c38938cSJosef Bacik } 23448c38938cSJosef Bacik } 23458c38938cSJosef Bacik 2346faa2dbf0SJosef Bacik void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2347171f6537SJosef Bacik { 2348171f6537SJosef Bacik int ret; 2349171f6537SJosef Bacik struct btrfs_root *gang[8]; 2350171f6537SJosef Bacik int i; 2351171f6537SJosef Bacik 2352171f6537SJosef Bacik while (!list_empty(&fs_info->dead_roots)) { 2353171f6537SJosef Bacik gang[0] = list_entry(fs_info->dead_roots.next, 2354171f6537SJosef Bacik struct btrfs_root, root_list); 2355171f6537SJosef Bacik list_del(&gang[0]->root_list); 2356171f6537SJosef Bacik 23578c38938cSJosef Bacik if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) 2358cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[0]); 235900246528SJosef Bacik btrfs_put_root(gang[0]); 2360171f6537SJosef Bacik } 2361171f6537SJosef Bacik 2362171f6537SJosef Bacik while (1) { 2363171f6537SJosef Bacik ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2364171f6537SJosef Bacik (void **)gang, 0, 2365171f6537SJosef Bacik ARRAY_SIZE(gang)); 2366171f6537SJosef Bacik if (!ret) 2367171f6537SJosef Bacik break; 2368171f6537SJosef Bacik for (i = 0; i < ret; i++) 2369cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2370171f6537SJosef Bacik } 2371171f6537SJosef Bacik } 2372af31f5e5SChris Mason 2373638aa7edSEric Sandeen static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2374638aa7edSEric Sandeen { 2375638aa7edSEric Sandeen mutex_init(&fs_info->scrub_lock); 2376638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_running, 0); 2377638aa7edSEric Sandeen atomic_set(&fs_info->scrub_pause_req, 0); 2378638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_paused, 0); 2379638aa7edSEric Sandeen atomic_set(&fs_info->scrub_cancel_req, 0); 2380638aa7edSEric Sandeen init_waitqueue_head(&fs_info->scrub_pause_wait); 2381ff09c4caSAnand Jain refcount_set(&fs_info->scrub_workers_refcnt, 0); 2382638aa7edSEric Sandeen } 2383638aa7edSEric Sandeen 2384779a65a4SEric Sandeen static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2385779a65a4SEric Sandeen { 2386779a65a4SEric Sandeen spin_lock_init(&fs_info->balance_lock); 2387779a65a4SEric Sandeen mutex_init(&fs_info->balance_mutex); 2388779a65a4SEric Sandeen atomic_set(&fs_info->balance_pause_req, 0); 2389779a65a4SEric Sandeen atomic_set(&fs_info->balance_cancel_req, 0); 2390779a65a4SEric Sandeen fs_info->balance_ctl = NULL; 2391779a65a4SEric Sandeen init_waitqueue_head(&fs_info->balance_wait_q); 2392907d2710SDavid Sterba atomic_set(&fs_info->reloc_cancel_req, 0); 2393779a65a4SEric Sandeen } 2394779a65a4SEric Sandeen 23956bccf3abSJeff Mahoney static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info) 2396f37938e0SEric Sandeen { 23972ff7e61eSJeff Mahoney struct inode *inode = fs_info->btree_inode; 23982ff7e61eSJeff Mahoney 23992ff7e61eSJeff Mahoney inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 24002ff7e61eSJeff Mahoney set_nlink(inode, 1); 2401f37938e0SEric Sandeen /* 2402f37938e0SEric Sandeen * we set the i_size on the btree inode to the max possible int. 2403f37938e0SEric Sandeen * the real end of the address space is determined by all of 2404f37938e0SEric Sandeen * the devices in the system 2405f37938e0SEric Sandeen */ 24062ff7e61eSJeff Mahoney inode->i_size = OFFSET_MAX; 24072ff7e61eSJeff Mahoney inode->i_mapping->a_ops = &btree_aops; 2408f37938e0SEric Sandeen 24092ff7e61eSJeff Mahoney RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 241043eb5f29SQu Wenruo extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree, 24112c53a14dSQu Wenruo IO_TREE_BTREE_INODE_IO, inode); 24127b439738SDavid Sterba BTRFS_I(inode)->io_tree.track_uptodate = false; 24132ff7e61eSJeff Mahoney extent_map_tree_init(&BTRFS_I(inode)->extent_tree); 2414f37938e0SEric Sandeen 24155c8fd99fSJosef Bacik BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root); 24162ff7e61eSJeff Mahoney memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key)); 24172ff7e61eSJeff Mahoney set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); 24182ff7e61eSJeff Mahoney btrfs_insert_inode_hash(inode); 2419f37938e0SEric Sandeen } 2420f37938e0SEric Sandeen 2421ad618368SEric Sandeen static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2422ad618368SEric Sandeen { 2423ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2424129827e3SDavid Sterba init_rwsem(&fs_info->dev_replace.rwsem); 24257f8d236aSDavid Sterba init_waitqueue_head(&fs_info->dev_replace.replace_wait); 2426ad618368SEric Sandeen } 2427ad618368SEric Sandeen 2428f9e92e40SEric Sandeen static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2429f9e92e40SEric Sandeen { 2430f9e92e40SEric Sandeen spin_lock_init(&fs_info->qgroup_lock); 2431f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_ioctl_lock); 2432f9e92e40SEric Sandeen fs_info->qgroup_tree = RB_ROOT; 2433f9e92e40SEric Sandeen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2434f9e92e40SEric Sandeen fs_info->qgroup_seq = 1; 2435f9e92e40SEric Sandeen fs_info->qgroup_ulist = NULL; 2436d2c609b8SJeff Mahoney fs_info->qgroup_rescan_running = false; 2437f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_rescan_lock); 2438f9e92e40SEric Sandeen } 2439f9e92e40SEric Sandeen 2440d21deec5SSu Yue static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info) 24412a458198SEric Sandeen { 2442f7b885beSAnand Jain u32 max_active = fs_info->thread_pool_size; 24436f011058SDavid Sterba unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 24442a458198SEric Sandeen 24452a458198SEric Sandeen fs_info->workers = 2446cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "worker", 2447cb001095SJeff Mahoney flags | WQ_HIGHPRI, max_active, 16); 24482a458198SEric Sandeen 24492a458198SEric Sandeen fs_info->delalloc_workers = 2450cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delalloc", 2451cb001095SJeff Mahoney flags, max_active, 2); 24522a458198SEric Sandeen 24532a458198SEric Sandeen fs_info->flush_workers = 2454cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "flush_delalloc", 2455cb001095SJeff Mahoney flags, max_active, 0); 24562a458198SEric Sandeen 24572a458198SEric Sandeen fs_info->caching_workers = 2458cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); 24592a458198SEric Sandeen 24602a458198SEric Sandeen fs_info->fixup_workers = 2461cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); 24622a458198SEric Sandeen 24632a458198SEric Sandeen /* 24642a458198SEric Sandeen * endios are largely parallel and should have a very 24652a458198SEric Sandeen * low idle thresh 24662a458198SEric Sandeen */ 24672a458198SEric Sandeen fs_info->endio_workers = 2468cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4); 24692a458198SEric Sandeen fs_info->endio_meta_workers = 2470cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta", flags, 2471cb001095SJeff Mahoney max_active, 4); 24722a458198SEric Sandeen fs_info->endio_meta_write_workers = 2473cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags, 2474cb001095SJeff Mahoney max_active, 2); 24752a458198SEric Sandeen fs_info->endio_raid56_workers = 2476cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-raid56", flags, 2477cb001095SJeff Mahoney max_active, 4); 24782a458198SEric Sandeen fs_info->rmw_workers = 2479cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2); 24802a458198SEric Sandeen fs_info->endio_write_workers = 2481cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-write", flags, 2482cb001095SJeff Mahoney max_active, 2); 24832a458198SEric Sandeen fs_info->endio_freespace_worker = 2484cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "freespace-write", flags, 2485cb001095SJeff Mahoney max_active, 0); 24862a458198SEric Sandeen fs_info->delayed_workers = 2487cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, 2488cb001095SJeff Mahoney max_active, 0); 24892a458198SEric Sandeen fs_info->qgroup_rescan_workers = 2490cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); 2491b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers = 2492b0643e59SDennis Zhou alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1); 24932a458198SEric Sandeen 24942a458198SEric Sandeen if (!(fs_info->workers && fs_info->delalloc_workers && 2495ba8a9d07SChris Mason fs_info->flush_workers && 24962a458198SEric Sandeen fs_info->endio_workers && fs_info->endio_meta_workers && 24972a458198SEric Sandeen fs_info->endio_meta_write_workers && 24982a458198SEric Sandeen fs_info->endio_write_workers && fs_info->endio_raid56_workers && 24992a458198SEric Sandeen fs_info->endio_freespace_worker && fs_info->rmw_workers && 2500f26c9238SQu Wenruo fs_info->caching_workers && fs_info->fixup_workers && 2501f26c9238SQu Wenruo fs_info->delayed_workers && fs_info->qgroup_rescan_workers && 2502b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers)) { 25032a458198SEric Sandeen return -ENOMEM; 25042a458198SEric Sandeen } 25052a458198SEric Sandeen 25062a458198SEric Sandeen return 0; 25072a458198SEric Sandeen } 25082a458198SEric Sandeen 25096d97c6e3SJohannes Thumshirn static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type) 25106d97c6e3SJohannes Thumshirn { 25116d97c6e3SJohannes Thumshirn struct crypto_shash *csum_shash; 2512b4e967beSDavid Sterba const char *csum_driver = btrfs_super_csum_driver(csum_type); 25136d97c6e3SJohannes Thumshirn 2514b4e967beSDavid Sterba csum_shash = crypto_alloc_shash(csum_driver, 0, 0); 25156d97c6e3SJohannes Thumshirn 25166d97c6e3SJohannes Thumshirn if (IS_ERR(csum_shash)) { 25176d97c6e3SJohannes Thumshirn btrfs_err(fs_info, "error allocating %s hash for checksum", 2518b4e967beSDavid Sterba csum_driver); 25196d97c6e3SJohannes Thumshirn return PTR_ERR(csum_shash); 25206d97c6e3SJohannes Thumshirn } 25216d97c6e3SJohannes Thumshirn 25226d97c6e3SJohannes Thumshirn fs_info->csum_shash = csum_shash; 25236d97c6e3SJohannes Thumshirn 25246d97c6e3SJohannes Thumshirn return 0; 25256d97c6e3SJohannes Thumshirn } 25266d97c6e3SJohannes Thumshirn 252763443bf5SEric Sandeen static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 252863443bf5SEric Sandeen struct btrfs_fs_devices *fs_devices) 252963443bf5SEric Sandeen { 253063443bf5SEric Sandeen int ret; 253163443bf5SEric Sandeen struct btrfs_root *log_tree_root; 253263443bf5SEric Sandeen struct btrfs_super_block *disk_super = fs_info->super_copy; 253363443bf5SEric Sandeen u64 bytenr = btrfs_super_log_root(disk_super); 2534581c1760SQu Wenruo int level = btrfs_super_log_root_level(disk_super); 253563443bf5SEric Sandeen 253663443bf5SEric Sandeen if (fs_devices->rw_devices == 0) { 2537f14d104dSDavid Sterba btrfs_warn(fs_info, "log replay required on RO media"); 253863443bf5SEric Sandeen return -EIO; 253963443bf5SEric Sandeen } 254063443bf5SEric Sandeen 254196dfcb46SJosef Bacik log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, 254296dfcb46SJosef Bacik GFP_KERNEL); 254363443bf5SEric Sandeen if (!log_tree_root) 254463443bf5SEric Sandeen return -ENOMEM; 254563443bf5SEric Sandeen 25462ff7e61eSJeff Mahoney log_tree_root->node = read_tree_block(fs_info, bytenr, 25471b7ec85eSJosef Bacik BTRFS_TREE_LOG_OBJECTID, 25481b7ec85eSJosef Bacik fs_info->generation + 1, level, 25491b7ec85eSJosef Bacik NULL); 255064c043deSLiu Bo if (IS_ERR(log_tree_root->node)) { 2551f14d104dSDavid Sterba btrfs_warn(fs_info, "failed to read log tree"); 25520eeff236SLiu Bo ret = PTR_ERR(log_tree_root->node); 25538c38938cSJosef Bacik log_tree_root->node = NULL; 255400246528SJosef Bacik btrfs_put_root(log_tree_root); 25550eeff236SLiu Bo return ret; 25564eb150d6SQu Wenruo } 25574eb150d6SQu Wenruo if (!extent_buffer_uptodate(log_tree_root->node)) { 2558f14d104dSDavid Sterba btrfs_err(fs_info, "failed to read log tree"); 255900246528SJosef Bacik btrfs_put_root(log_tree_root); 256063443bf5SEric Sandeen return -EIO; 256163443bf5SEric Sandeen } 25624eb150d6SQu Wenruo 256363443bf5SEric Sandeen /* returns with log_tree_root freed on success */ 256463443bf5SEric Sandeen ret = btrfs_recover_log_trees(log_tree_root); 256563443bf5SEric Sandeen if (ret) { 25660b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 256763443bf5SEric Sandeen "Failed to recover log tree"); 256800246528SJosef Bacik btrfs_put_root(log_tree_root); 256963443bf5SEric Sandeen return ret; 257063443bf5SEric Sandeen } 257163443bf5SEric Sandeen 2572bc98a42cSDavid Howells if (sb_rdonly(fs_info->sb)) { 25736bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 257463443bf5SEric Sandeen if (ret) 257563443bf5SEric Sandeen return ret; 257663443bf5SEric Sandeen } 257763443bf5SEric Sandeen 257863443bf5SEric Sandeen return 0; 257963443bf5SEric Sandeen } 258063443bf5SEric Sandeen 2581abed4aaaSJosef Bacik static int load_global_roots_objectid(struct btrfs_root *tree_root, 2582abed4aaaSJosef Bacik struct btrfs_path *path, u64 objectid, 2583abed4aaaSJosef Bacik const char *name) 2584abed4aaaSJosef Bacik { 2585abed4aaaSJosef Bacik struct btrfs_fs_info *fs_info = tree_root->fs_info; 2586abed4aaaSJosef Bacik struct btrfs_root *root; 2587f7238e50SJosef Bacik u64 max_global_id = 0; 2588abed4aaaSJosef Bacik int ret; 2589abed4aaaSJosef Bacik struct btrfs_key key = { 2590abed4aaaSJosef Bacik .objectid = objectid, 2591abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 2592abed4aaaSJosef Bacik .offset = 0, 2593abed4aaaSJosef Bacik }; 2594abed4aaaSJosef Bacik bool found = false; 2595abed4aaaSJosef Bacik 2596abed4aaaSJosef Bacik /* If we have IGNOREDATACSUMS skip loading these roots. */ 2597abed4aaaSJosef Bacik if (objectid == BTRFS_CSUM_TREE_OBJECTID && 2598abed4aaaSJosef Bacik btrfs_test_opt(fs_info, IGNOREDATACSUMS)) { 2599abed4aaaSJosef Bacik set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state); 2600abed4aaaSJosef Bacik return 0; 2601abed4aaaSJosef Bacik } 2602abed4aaaSJosef Bacik 2603abed4aaaSJosef Bacik while (1) { 2604abed4aaaSJosef Bacik ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0); 2605abed4aaaSJosef Bacik if (ret < 0) 2606abed4aaaSJosef Bacik break; 2607abed4aaaSJosef Bacik 2608abed4aaaSJosef Bacik if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 2609abed4aaaSJosef Bacik ret = btrfs_next_leaf(tree_root, path); 2610abed4aaaSJosef Bacik if (ret) { 2611abed4aaaSJosef Bacik if (ret > 0) 2612abed4aaaSJosef Bacik ret = 0; 2613abed4aaaSJosef Bacik break; 2614abed4aaaSJosef Bacik } 2615abed4aaaSJosef Bacik } 2616abed4aaaSJosef Bacik ret = 0; 2617abed4aaaSJosef Bacik 2618abed4aaaSJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 2619abed4aaaSJosef Bacik if (key.objectid != objectid) 2620abed4aaaSJosef Bacik break; 2621abed4aaaSJosef Bacik btrfs_release_path(path); 2622abed4aaaSJosef Bacik 2623f7238e50SJosef Bacik /* 2624f7238e50SJosef Bacik * Just worry about this for extent tree, it'll be the same for 2625f7238e50SJosef Bacik * everybody. 2626f7238e50SJosef Bacik */ 2627f7238e50SJosef Bacik if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 2628f7238e50SJosef Bacik max_global_id = max(max_global_id, key.offset); 2629f7238e50SJosef Bacik 2630abed4aaaSJosef Bacik found = true; 2631abed4aaaSJosef Bacik root = read_tree_root_path(tree_root, path, &key); 2632abed4aaaSJosef Bacik if (IS_ERR(root)) { 2633abed4aaaSJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) 2634abed4aaaSJosef Bacik ret = PTR_ERR(root); 2635abed4aaaSJosef Bacik break; 2636abed4aaaSJosef Bacik } 2637abed4aaaSJosef Bacik set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2638abed4aaaSJosef Bacik ret = btrfs_global_root_insert(root); 2639abed4aaaSJosef Bacik if (ret) { 2640abed4aaaSJosef Bacik btrfs_put_root(root); 2641abed4aaaSJosef Bacik break; 2642abed4aaaSJosef Bacik } 2643abed4aaaSJosef Bacik key.offset++; 2644abed4aaaSJosef Bacik } 2645abed4aaaSJosef Bacik btrfs_release_path(path); 2646abed4aaaSJosef Bacik 2647f7238e50SJosef Bacik if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 2648f7238e50SJosef Bacik fs_info->nr_global_roots = max_global_id + 1; 2649f7238e50SJosef Bacik 2650abed4aaaSJosef Bacik if (!found || ret) { 2651abed4aaaSJosef Bacik if (objectid == BTRFS_CSUM_TREE_OBJECTID) 2652abed4aaaSJosef Bacik set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state); 2653abed4aaaSJosef Bacik 2654abed4aaaSJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) 2655abed4aaaSJosef Bacik ret = ret ? ret : -ENOENT; 2656abed4aaaSJosef Bacik else 2657abed4aaaSJosef Bacik ret = 0; 2658abed4aaaSJosef Bacik btrfs_err(fs_info, "failed to load root %s", name); 2659abed4aaaSJosef Bacik } 2660abed4aaaSJosef Bacik return ret; 2661abed4aaaSJosef Bacik } 2662abed4aaaSJosef Bacik 2663abed4aaaSJosef Bacik static int load_global_roots(struct btrfs_root *tree_root) 2664abed4aaaSJosef Bacik { 2665abed4aaaSJosef Bacik struct btrfs_path *path; 2666abed4aaaSJosef Bacik int ret = 0; 2667abed4aaaSJosef Bacik 2668abed4aaaSJosef Bacik path = btrfs_alloc_path(); 2669abed4aaaSJosef Bacik if (!path) 2670abed4aaaSJosef Bacik return -ENOMEM; 2671abed4aaaSJosef Bacik 2672abed4aaaSJosef Bacik ret = load_global_roots_objectid(tree_root, path, 2673abed4aaaSJosef Bacik BTRFS_EXTENT_TREE_OBJECTID, "extent"); 2674abed4aaaSJosef Bacik if (ret) 2675abed4aaaSJosef Bacik goto out; 2676abed4aaaSJosef Bacik ret = load_global_roots_objectid(tree_root, path, 2677abed4aaaSJosef Bacik BTRFS_CSUM_TREE_OBJECTID, "csum"); 2678abed4aaaSJosef Bacik if (ret) 2679abed4aaaSJosef Bacik goto out; 2680abed4aaaSJosef Bacik if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE)) 2681abed4aaaSJosef Bacik goto out; 2682abed4aaaSJosef Bacik ret = load_global_roots_objectid(tree_root, path, 2683abed4aaaSJosef Bacik BTRFS_FREE_SPACE_TREE_OBJECTID, 2684abed4aaaSJosef Bacik "free space"); 2685abed4aaaSJosef Bacik out: 2686abed4aaaSJosef Bacik btrfs_free_path(path); 2687abed4aaaSJosef Bacik return ret; 2688abed4aaaSJosef Bacik } 2689abed4aaaSJosef Bacik 26906bccf3abSJeff Mahoney static int btrfs_read_roots(struct btrfs_fs_info *fs_info) 26914bbcaa64SEric Sandeen { 26926bccf3abSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 2693a4f3d2c4SDavid Sterba struct btrfs_root *root; 26944bbcaa64SEric Sandeen struct btrfs_key location; 26954bbcaa64SEric Sandeen int ret; 26964bbcaa64SEric Sandeen 26976bccf3abSJeff Mahoney BUG_ON(!fs_info->tree_root); 26986bccf3abSJeff Mahoney 2699abed4aaaSJosef Bacik ret = load_global_roots(tree_root); 2700abed4aaaSJosef Bacik if (ret) 2701abed4aaaSJosef Bacik return ret; 2702abed4aaaSJosef Bacik 2703abed4aaaSJosef Bacik location.objectid = BTRFS_DEV_TREE_OBJECTID; 27044bbcaa64SEric Sandeen location.type = BTRFS_ROOT_ITEM_KEY; 27054bbcaa64SEric Sandeen location.offset = 0; 27064bbcaa64SEric Sandeen 2707a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2708f50f4353SLiu Bo if (IS_ERR(root)) { 270942437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2710f50f4353SLiu Bo ret = PTR_ERR(root); 2711f50f4353SLiu Bo goto out; 2712f50f4353SLiu Bo } 271342437a63SJosef Bacik } else { 2714a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2715a4f3d2c4SDavid Sterba fs_info->dev_root = root; 271642437a63SJosef Bacik } 2717820a49daSJosef Bacik /* Initialize fs_info for all devices in any case */ 2718820a49daSJosef Bacik btrfs_init_devices_late(fs_info); 27194bbcaa64SEric Sandeen 2720aeb935a4SQu Wenruo /* 2721aeb935a4SQu Wenruo * This tree can share blocks with some other fs tree during relocation 2722aeb935a4SQu Wenruo * and we need a proper setup by btrfs_get_fs_root 2723aeb935a4SQu Wenruo */ 272456e9357aSDavid Sterba root = btrfs_get_fs_root(tree_root->fs_info, 272556e9357aSDavid Sterba BTRFS_DATA_RELOC_TREE_OBJECTID, true); 2726aeb935a4SQu Wenruo if (IS_ERR(root)) { 272742437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2728aeb935a4SQu Wenruo ret = PTR_ERR(root); 2729aeb935a4SQu Wenruo goto out; 2730aeb935a4SQu Wenruo } 273142437a63SJosef Bacik } else { 2732aeb935a4SQu Wenruo set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2733aeb935a4SQu Wenruo fs_info->data_reloc_root = root; 273442437a63SJosef Bacik } 2735aeb935a4SQu Wenruo 27364bbcaa64SEric Sandeen location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2737a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2738a4f3d2c4SDavid Sterba if (!IS_ERR(root)) { 2739a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2740afcdd129SJosef Bacik set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); 2741a4f3d2c4SDavid Sterba fs_info->quota_root = root; 27424bbcaa64SEric Sandeen } 27434bbcaa64SEric Sandeen 27444bbcaa64SEric Sandeen location.objectid = BTRFS_UUID_TREE_OBJECTID; 2745a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2746a4f3d2c4SDavid Sterba if (IS_ERR(root)) { 274742437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2748a4f3d2c4SDavid Sterba ret = PTR_ERR(root); 27494bbcaa64SEric Sandeen if (ret != -ENOENT) 2750f50f4353SLiu Bo goto out; 275142437a63SJosef Bacik } 27524bbcaa64SEric Sandeen } else { 2753a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2754a4f3d2c4SDavid Sterba fs_info->uuid_root = root; 27554bbcaa64SEric Sandeen } 27564bbcaa64SEric Sandeen 27574bbcaa64SEric Sandeen return 0; 2758f50f4353SLiu Bo out: 2759f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", 2760f50f4353SLiu Bo location.objectid, ret); 2761f50f4353SLiu Bo return ret; 27624bbcaa64SEric Sandeen } 27634bbcaa64SEric Sandeen 2764069ec957SQu Wenruo /* 2765069ec957SQu Wenruo * Real super block validation 2766069ec957SQu Wenruo * NOTE: super csum type and incompat features will not be checked here. 2767069ec957SQu Wenruo * 2768069ec957SQu Wenruo * @sb: super block to check 2769069ec957SQu Wenruo * @mirror_num: the super block number to check its bytenr: 2770069ec957SQu Wenruo * 0 the primary (1st) sb 2771069ec957SQu Wenruo * 1, 2 2nd and 3rd backup copy 2772069ec957SQu Wenruo * -1 skip bytenr check 2773069ec957SQu Wenruo */ 2774069ec957SQu Wenruo static int validate_super(struct btrfs_fs_info *fs_info, 2775069ec957SQu Wenruo struct btrfs_super_block *sb, int mirror_num) 277621a852b0SQu Wenruo { 277721a852b0SQu Wenruo u64 nodesize = btrfs_super_nodesize(sb); 277821a852b0SQu Wenruo u64 sectorsize = btrfs_super_sectorsize(sb); 277921a852b0SQu Wenruo int ret = 0; 278021a852b0SQu Wenruo 278121a852b0SQu Wenruo if (btrfs_super_magic(sb) != BTRFS_MAGIC) { 278221a852b0SQu Wenruo btrfs_err(fs_info, "no valid FS found"); 278321a852b0SQu Wenruo ret = -EINVAL; 278421a852b0SQu Wenruo } 278521a852b0SQu Wenruo if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { 278621a852b0SQu Wenruo btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", 278721a852b0SQu Wenruo btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); 278821a852b0SQu Wenruo ret = -EINVAL; 278921a852b0SQu Wenruo } 279021a852b0SQu Wenruo if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 279121a852b0SQu Wenruo btrfs_err(fs_info, "tree_root level too big: %d >= %d", 279221a852b0SQu Wenruo btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 279321a852b0SQu Wenruo ret = -EINVAL; 279421a852b0SQu Wenruo } 279521a852b0SQu Wenruo if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 279621a852b0SQu Wenruo btrfs_err(fs_info, "chunk_root level too big: %d >= %d", 279721a852b0SQu Wenruo btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 279821a852b0SQu Wenruo ret = -EINVAL; 279921a852b0SQu Wenruo } 280021a852b0SQu Wenruo if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 280121a852b0SQu Wenruo btrfs_err(fs_info, "log_root level too big: %d >= %d", 280221a852b0SQu Wenruo btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 280321a852b0SQu Wenruo ret = -EINVAL; 280421a852b0SQu Wenruo } 280521a852b0SQu Wenruo 280621a852b0SQu Wenruo /* 280721a852b0SQu Wenruo * Check sectorsize and nodesize first, other check will need it. 280821a852b0SQu Wenruo * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. 280921a852b0SQu Wenruo */ 281021a852b0SQu Wenruo if (!is_power_of_2(sectorsize) || sectorsize < 4096 || 281121a852b0SQu Wenruo sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { 281221a852b0SQu Wenruo btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); 281321a852b0SQu Wenruo ret = -EINVAL; 281421a852b0SQu Wenruo } 28150bb3eb3eSQu Wenruo 28160bb3eb3eSQu Wenruo /* 28170bb3eb3eSQu Wenruo * For 4K page size, we only support 4K sector size. 281850780d9bSAnand Jain * For 64K page size, we support 64K and 4K sector sizes. 28190bb3eb3eSQu Wenruo */ 28200bb3eb3eSQu Wenruo if ((PAGE_SIZE == SZ_4K && sectorsize != PAGE_SIZE) || 28210bb3eb3eSQu Wenruo (PAGE_SIZE == SZ_64K && (sectorsize != SZ_4K && 28220bb3eb3eSQu Wenruo sectorsize != SZ_64K))) { 282321a852b0SQu Wenruo btrfs_err(fs_info, 28240bb3eb3eSQu Wenruo "sectorsize %llu not yet supported for page size %lu", 282521a852b0SQu Wenruo sectorsize, PAGE_SIZE); 282621a852b0SQu Wenruo ret = -EINVAL; 282721a852b0SQu Wenruo } 28280bb3eb3eSQu Wenruo 282921a852b0SQu Wenruo if (!is_power_of_2(nodesize) || nodesize < sectorsize || 283021a852b0SQu Wenruo nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { 283121a852b0SQu Wenruo btrfs_err(fs_info, "invalid nodesize %llu", nodesize); 283221a852b0SQu Wenruo ret = -EINVAL; 283321a852b0SQu Wenruo } 283421a852b0SQu Wenruo if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { 283521a852b0SQu Wenruo btrfs_err(fs_info, "invalid leafsize %u, should be %llu", 283621a852b0SQu Wenruo le32_to_cpu(sb->__unused_leafsize), nodesize); 283721a852b0SQu Wenruo ret = -EINVAL; 283821a852b0SQu Wenruo } 283921a852b0SQu Wenruo 284021a852b0SQu Wenruo /* Root alignment check */ 284121a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { 284221a852b0SQu Wenruo btrfs_warn(fs_info, "tree_root block unaligned: %llu", 284321a852b0SQu Wenruo btrfs_super_root(sb)); 284421a852b0SQu Wenruo ret = -EINVAL; 284521a852b0SQu Wenruo } 284621a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { 284721a852b0SQu Wenruo btrfs_warn(fs_info, "chunk_root block unaligned: %llu", 284821a852b0SQu Wenruo btrfs_super_chunk_root(sb)); 284921a852b0SQu Wenruo ret = -EINVAL; 285021a852b0SQu Wenruo } 285121a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { 285221a852b0SQu Wenruo btrfs_warn(fs_info, "log_root block unaligned: %llu", 285321a852b0SQu Wenruo btrfs_super_log_root(sb)); 285421a852b0SQu Wenruo ret = -EINVAL; 285521a852b0SQu Wenruo } 285621a852b0SQu Wenruo 2857aefd7f70SNikolay Borisov if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, 2858aefd7f70SNikolay Borisov BTRFS_FSID_SIZE)) { 2859aefd7f70SNikolay Borisov btrfs_err(fs_info, 2860aefd7f70SNikolay Borisov "superblock fsid doesn't match fsid of fs_devices: %pU != %pU", 2861aefd7f70SNikolay Borisov fs_info->super_copy->fsid, fs_info->fs_devices->fsid); 2862aefd7f70SNikolay Borisov ret = -EINVAL; 2863aefd7f70SNikolay Borisov } 2864aefd7f70SNikolay Borisov 2865aefd7f70SNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID) && 2866aefd7f70SNikolay Borisov memcmp(fs_info->fs_devices->metadata_uuid, 2867aefd7f70SNikolay Borisov fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) { 2868aefd7f70SNikolay Borisov btrfs_err(fs_info, 2869aefd7f70SNikolay Borisov "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU", 2870aefd7f70SNikolay Borisov fs_info->super_copy->metadata_uuid, 2871aefd7f70SNikolay Borisov fs_info->fs_devices->metadata_uuid); 2872aefd7f70SNikolay Borisov ret = -EINVAL; 2873aefd7f70SNikolay Borisov } 2874aefd7f70SNikolay Borisov 2875de37aa51SNikolay Borisov if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, 28767239ff4bSNikolay Borisov BTRFS_FSID_SIZE) != 0) { 287721a852b0SQu Wenruo btrfs_err(fs_info, 28787239ff4bSNikolay Borisov "dev_item UUID does not match metadata fsid: %pU != %pU", 2879de37aa51SNikolay Borisov fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); 288021a852b0SQu Wenruo ret = -EINVAL; 288121a852b0SQu Wenruo } 288221a852b0SQu Wenruo 288321a852b0SQu Wenruo /* 288421a852b0SQu Wenruo * Hint to catch really bogus numbers, bitflips or so, more exact checks are 288521a852b0SQu Wenruo * done later 288621a852b0SQu Wenruo */ 288721a852b0SQu Wenruo if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { 288821a852b0SQu Wenruo btrfs_err(fs_info, "bytes_used is too small %llu", 288921a852b0SQu Wenruo btrfs_super_bytes_used(sb)); 289021a852b0SQu Wenruo ret = -EINVAL; 289121a852b0SQu Wenruo } 289221a852b0SQu Wenruo if (!is_power_of_2(btrfs_super_stripesize(sb))) { 289321a852b0SQu Wenruo btrfs_err(fs_info, "invalid stripesize %u", 289421a852b0SQu Wenruo btrfs_super_stripesize(sb)); 289521a852b0SQu Wenruo ret = -EINVAL; 289621a852b0SQu Wenruo } 289721a852b0SQu Wenruo if (btrfs_super_num_devices(sb) > (1UL << 31)) 289821a852b0SQu Wenruo btrfs_warn(fs_info, "suspicious number of devices: %llu", 289921a852b0SQu Wenruo btrfs_super_num_devices(sb)); 290021a852b0SQu Wenruo if (btrfs_super_num_devices(sb) == 0) { 290121a852b0SQu Wenruo btrfs_err(fs_info, "number of devices is 0"); 290221a852b0SQu Wenruo ret = -EINVAL; 290321a852b0SQu Wenruo } 290421a852b0SQu Wenruo 2905069ec957SQu Wenruo if (mirror_num >= 0 && 2906069ec957SQu Wenruo btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { 290721a852b0SQu Wenruo btrfs_err(fs_info, "super offset mismatch %llu != %u", 290821a852b0SQu Wenruo btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 290921a852b0SQu Wenruo ret = -EINVAL; 291021a852b0SQu Wenruo } 291121a852b0SQu Wenruo 291221a852b0SQu Wenruo /* 291321a852b0SQu Wenruo * Obvious sys_chunk_array corruptions, it must hold at least one key 291421a852b0SQu Wenruo * and one chunk 291521a852b0SQu Wenruo */ 291621a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 291721a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too big %u > %u", 291821a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 291921a852b0SQu Wenruo BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 292021a852b0SQu Wenruo ret = -EINVAL; 292121a852b0SQu Wenruo } 292221a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 292321a852b0SQu Wenruo + sizeof(struct btrfs_chunk)) { 292421a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too small %u < %zu", 292521a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 292621a852b0SQu Wenruo sizeof(struct btrfs_disk_key) 292721a852b0SQu Wenruo + sizeof(struct btrfs_chunk)); 292821a852b0SQu Wenruo ret = -EINVAL; 292921a852b0SQu Wenruo } 293021a852b0SQu Wenruo 293121a852b0SQu Wenruo /* 293221a852b0SQu Wenruo * The generation is a global counter, we'll trust it more than the others 293321a852b0SQu Wenruo * but it's still possible that it's the one that's wrong. 293421a852b0SQu Wenruo */ 293521a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 293621a852b0SQu Wenruo btrfs_warn(fs_info, 293721a852b0SQu Wenruo "suspicious: generation < chunk_root_generation: %llu < %llu", 293821a852b0SQu Wenruo btrfs_super_generation(sb), 293921a852b0SQu Wenruo btrfs_super_chunk_root_generation(sb)); 294021a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 294121a852b0SQu Wenruo && btrfs_super_cache_generation(sb) != (u64)-1) 294221a852b0SQu Wenruo btrfs_warn(fs_info, 294321a852b0SQu Wenruo "suspicious: generation < cache_generation: %llu < %llu", 294421a852b0SQu Wenruo btrfs_super_generation(sb), 294521a852b0SQu Wenruo btrfs_super_cache_generation(sb)); 294621a852b0SQu Wenruo 294721a852b0SQu Wenruo return ret; 294821a852b0SQu Wenruo } 294921a852b0SQu Wenruo 2950069ec957SQu Wenruo /* 2951069ec957SQu Wenruo * Validation of super block at mount time. 2952069ec957SQu Wenruo * Some checks already done early at mount time, like csum type and incompat 2953069ec957SQu Wenruo * flags will be skipped. 2954069ec957SQu Wenruo */ 2955069ec957SQu Wenruo static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) 2956069ec957SQu Wenruo { 2957069ec957SQu Wenruo return validate_super(fs_info, fs_info->super_copy, 0); 2958069ec957SQu Wenruo } 2959069ec957SQu Wenruo 296075cb857dSQu Wenruo /* 296175cb857dSQu Wenruo * Validation of super block at write time. 296275cb857dSQu Wenruo * Some checks like bytenr check will be skipped as their values will be 296375cb857dSQu Wenruo * overwritten soon. 296475cb857dSQu Wenruo * Extra checks like csum type and incompat flags will be done here. 296575cb857dSQu Wenruo */ 296675cb857dSQu Wenruo static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, 296775cb857dSQu Wenruo struct btrfs_super_block *sb) 296875cb857dSQu Wenruo { 296975cb857dSQu Wenruo int ret; 297075cb857dSQu Wenruo 297175cb857dSQu Wenruo ret = validate_super(fs_info, sb, -1); 297275cb857dSQu Wenruo if (ret < 0) 297375cb857dSQu Wenruo goto out; 2974e7e16f48SJohannes Thumshirn if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) { 297575cb857dSQu Wenruo ret = -EUCLEAN; 297675cb857dSQu Wenruo btrfs_err(fs_info, "invalid csum type, has %u want %u", 297775cb857dSQu Wenruo btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); 297875cb857dSQu Wenruo goto out; 297975cb857dSQu Wenruo } 298075cb857dSQu Wenruo if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 298175cb857dSQu Wenruo ret = -EUCLEAN; 298275cb857dSQu Wenruo btrfs_err(fs_info, 298375cb857dSQu Wenruo "invalid incompat flags, has 0x%llx valid mask 0x%llx", 298475cb857dSQu Wenruo btrfs_super_incompat_flags(sb), 298575cb857dSQu Wenruo (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); 298675cb857dSQu Wenruo goto out; 298775cb857dSQu Wenruo } 298875cb857dSQu Wenruo out: 298975cb857dSQu Wenruo if (ret < 0) 299075cb857dSQu Wenruo btrfs_err(fs_info, 299175cb857dSQu Wenruo "super block corruption detected before writing it to disk"); 299275cb857dSQu Wenruo return ret; 299375cb857dSQu Wenruo } 299475cb857dSQu Wenruo 2995bd676446SJosef Bacik static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level) 2996bd676446SJosef Bacik { 2997bd676446SJosef Bacik int ret = 0; 2998bd676446SJosef Bacik 2999bd676446SJosef Bacik root->node = read_tree_block(root->fs_info, bytenr, 3000bd676446SJosef Bacik root->root_key.objectid, gen, level, NULL); 3001bd676446SJosef Bacik if (IS_ERR(root->node)) { 3002bd676446SJosef Bacik ret = PTR_ERR(root->node); 3003bd676446SJosef Bacik root->node = NULL; 30044eb150d6SQu Wenruo return ret; 30054eb150d6SQu Wenruo } 30064eb150d6SQu Wenruo if (!extent_buffer_uptodate(root->node)) { 3007bd676446SJosef Bacik free_extent_buffer(root->node); 3008bd676446SJosef Bacik root->node = NULL; 30094eb150d6SQu Wenruo return -EIO; 3010bd676446SJosef Bacik } 3011bd676446SJosef Bacik 3012bd676446SJosef Bacik btrfs_set_root_node(&root->root_item, root->node); 3013bd676446SJosef Bacik root->commit_root = btrfs_root_node(root); 3014bd676446SJosef Bacik btrfs_set_root_refs(&root->root_item, 1); 3015bd676446SJosef Bacik return ret; 3016bd676446SJosef Bacik } 3017bd676446SJosef Bacik 3018bd676446SJosef Bacik static int load_important_roots(struct btrfs_fs_info *fs_info) 3019bd676446SJosef Bacik { 3020bd676446SJosef Bacik struct btrfs_super_block *sb = fs_info->super_copy; 3021bd676446SJosef Bacik u64 gen, bytenr; 3022bd676446SJosef Bacik int level, ret; 3023bd676446SJosef Bacik 3024bd676446SJosef Bacik bytenr = btrfs_super_root(sb); 3025bd676446SJosef Bacik gen = btrfs_super_generation(sb); 3026bd676446SJosef Bacik level = btrfs_super_root_level(sb); 3027bd676446SJosef Bacik ret = load_super_root(fs_info->tree_root, bytenr, gen, level); 30289c54e80dSJosef Bacik if (ret) { 3029bd676446SJosef Bacik btrfs_warn(fs_info, "couldn't read tree root"); 3030bd676446SJosef Bacik return ret; 3031bd676446SJosef Bacik } 3032bd676446SJosef Bacik 30339c54e80dSJosef Bacik if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) 30349c54e80dSJosef Bacik return 0; 30359c54e80dSJosef Bacik 30369c54e80dSJosef Bacik bytenr = btrfs_super_block_group_root(sb); 30379c54e80dSJosef Bacik gen = btrfs_super_block_group_root_generation(sb); 30389c54e80dSJosef Bacik level = btrfs_super_block_group_root_level(sb); 30399c54e80dSJosef Bacik ret = load_super_root(fs_info->block_group_root, bytenr, gen, level); 30409c54e80dSJosef Bacik if (ret) 30419c54e80dSJosef Bacik btrfs_warn(fs_info, "couldn't read block group root"); 30429c54e80dSJosef Bacik return ret; 30439c54e80dSJosef Bacik } 30449c54e80dSJosef Bacik 30456ef108ddSNikolay Borisov static int __cold init_tree_roots(struct btrfs_fs_info *fs_info) 3046b8522a1eSNikolay Borisov { 30476ef108ddSNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 3048b8522a1eSNikolay Borisov struct btrfs_super_block *sb = fs_info->super_copy; 3049b8522a1eSNikolay Borisov struct btrfs_root *tree_root = fs_info->tree_root; 3050b8522a1eSNikolay Borisov bool handle_error = false; 3051b8522a1eSNikolay Borisov int ret = 0; 3052b8522a1eSNikolay Borisov int i; 3053b8522a1eSNikolay Borisov 30549c54e80dSJosef Bacik if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 30559c54e80dSJosef Bacik struct btrfs_root *root; 3056b8522a1eSNikolay Borisov 30579c54e80dSJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_BLOCK_GROUP_TREE_OBJECTID, 30589c54e80dSJosef Bacik GFP_KERNEL); 30599c54e80dSJosef Bacik if (!root) 30609c54e80dSJosef Bacik return -ENOMEM; 30619c54e80dSJosef Bacik fs_info->block_group_root = root; 30629c54e80dSJosef Bacik } 30639c54e80dSJosef Bacik 3064b8522a1eSNikolay Borisov for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 3065b8522a1eSNikolay Borisov if (handle_error) { 3066b8522a1eSNikolay Borisov if (!IS_ERR(tree_root->node)) 3067b8522a1eSNikolay Borisov free_extent_buffer(tree_root->node); 3068b8522a1eSNikolay Borisov tree_root->node = NULL; 3069b8522a1eSNikolay Borisov 3070b8522a1eSNikolay Borisov if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) 3071b8522a1eSNikolay Borisov break; 3072b8522a1eSNikolay Borisov 3073b8522a1eSNikolay Borisov free_root_pointers(fs_info, 0); 3074b8522a1eSNikolay Borisov 3075b8522a1eSNikolay Borisov /* 3076b8522a1eSNikolay Borisov * Don't use the log in recovery mode, it won't be 3077b8522a1eSNikolay Borisov * valid 3078b8522a1eSNikolay Borisov */ 3079b8522a1eSNikolay Borisov btrfs_set_super_log_root(sb, 0); 3080b8522a1eSNikolay Borisov 3081b8522a1eSNikolay Borisov /* We can't trust the free space cache either */ 3082b8522a1eSNikolay Borisov btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 3083b8522a1eSNikolay Borisov 3084b8522a1eSNikolay Borisov ret = read_backup_root(fs_info, i); 30856ef108ddSNikolay Borisov backup_index = ret; 3086b8522a1eSNikolay Borisov if (ret < 0) 3087b8522a1eSNikolay Borisov return ret; 3088b8522a1eSNikolay Borisov } 3089b8522a1eSNikolay Borisov 3090bd676446SJosef Bacik ret = load_important_roots(fs_info); 3091bd676446SJosef Bacik if (ret) { 3092217f5004SNikolay Borisov handle_error = true; 3093b8522a1eSNikolay Borisov continue; 3094b8522a1eSNikolay Borisov } 3095b8522a1eSNikolay Borisov 3096336a0d8dSNikolay Borisov /* 3097336a0d8dSNikolay Borisov * No need to hold btrfs_root::objectid_mutex since the fs 3098336a0d8dSNikolay Borisov * hasn't been fully initialised and we are the only user 3099336a0d8dSNikolay Borisov */ 3100453e4873SNikolay Borisov ret = btrfs_init_root_free_objectid(tree_root); 3101b8522a1eSNikolay Borisov if (ret < 0) { 3102b8522a1eSNikolay Borisov handle_error = true; 3103b8522a1eSNikolay Borisov continue; 3104b8522a1eSNikolay Borisov } 3105b8522a1eSNikolay Borisov 31066b8fad57SNikolay Borisov ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID); 3107b8522a1eSNikolay Borisov 3108b8522a1eSNikolay Borisov ret = btrfs_read_roots(fs_info); 3109b8522a1eSNikolay Borisov if (ret < 0) { 3110b8522a1eSNikolay Borisov handle_error = true; 3111b8522a1eSNikolay Borisov continue; 3112b8522a1eSNikolay Borisov } 3113b8522a1eSNikolay Borisov 3114b8522a1eSNikolay Borisov /* All successful */ 3115bd676446SJosef Bacik fs_info->generation = btrfs_header_generation(tree_root->node); 3116bd676446SJosef Bacik fs_info->last_trans_committed = fs_info->generation; 3117d96b3424SFilipe Manana fs_info->last_reloc_trans = 0; 31186ef108ddSNikolay Borisov 31196ef108ddSNikolay Borisov /* Always begin writing backup roots after the one being used */ 31206ef108ddSNikolay Borisov if (backup_index < 0) { 31216ef108ddSNikolay Borisov fs_info->backup_root_index = 0; 31226ef108ddSNikolay Borisov } else { 31236ef108ddSNikolay Borisov fs_info->backup_root_index = backup_index + 1; 31246ef108ddSNikolay Borisov fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS; 31256ef108ddSNikolay Borisov } 3126b8522a1eSNikolay Borisov break; 3127b8522a1eSNikolay Borisov } 3128b8522a1eSNikolay Borisov 3129b8522a1eSNikolay Borisov return ret; 3130b8522a1eSNikolay Borisov } 3131b8522a1eSNikolay Borisov 31328260edbaSJosef Bacik void btrfs_init_fs_info(struct btrfs_fs_info *fs_info) 3133eb60ceacSChris Mason { 313476dda93cSYan, Zheng INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 3135f28491e0SJosef Bacik INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 31368fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 3137facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 313824bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 3139eb73c1b7SMiao Xie INIT_LIST_HEAD(&fs_info->delalloc_roots); 314011833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 3141eb73c1b7SMiao Xie spin_lock_init(&fs_info->delalloc_root_lock); 3142a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 314376dda93cSYan, Zheng spin_lock_init(&fs_info->fs_roots_radix_lock); 314424bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 31454cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 3146ceda0864SMiao Xie spin_lock_init(&fs_info->super_lock); 3147f28491e0SJosef Bacik spin_lock_init(&fs_info->buffer_lock); 314847ab2a6cSJosef Bacik spin_lock_init(&fs_info->unused_bgs_lock); 314940ab3be1SNaohiro Aota spin_lock_init(&fs_info->treelog_bg_lock); 3150afba2bc0SNaohiro Aota spin_lock_init(&fs_info->zone_active_bgs_lock); 3151c2707a25SJohannes Thumshirn spin_lock_init(&fs_info->relocation_bg_lock); 3152f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 3153abed4aaaSJosef Bacik rwlock_init(&fs_info->global_root_lock); 3154d7c15171SZhao Lei mutex_init(&fs_info->unused_bg_unpin_mutex); 3155f3372065SJohannes Thumshirn mutex_init(&fs_info->reclaim_bgs_lock); 31567585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 3157573bfb72SMiao Xie mutex_init(&fs_info->delalloc_root_mutex); 31580bc09ca1SNaohiro Aota mutex_init(&fs_info->zoned_meta_io_lock); 31595f0addf7SNaohiro Aota mutex_init(&fs_info->zoned_data_reloc_io_lock); 3160de98ced9SMiao Xie seqlock_init(&fs_info->profiles_lock); 316119c00ddcSChris Mason 31620b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 31636324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 3164f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 316547ab2a6cSJosef Bacik INIT_LIST_HEAD(&fs_info->unused_bgs); 316618bb8bbfSJohannes Thumshirn INIT_LIST_HEAD(&fs_info->reclaim_bgs); 3167afba2bc0SNaohiro Aota INIT_LIST_HEAD(&fs_info->zone_active_bgs); 3168bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 3169bd647ce3SJosef Bacik INIT_LIST_HEAD(&fs_info->allocated_roots); 31703fd63727SJosef Bacik INIT_LIST_HEAD(&fs_info->allocated_ebs); 31713fd63727SJosef Bacik spin_lock_init(&fs_info->eb_leak_lock); 3172bd647ce3SJosef Bacik #endif 3173c8bf1b67SDavid Sterba extent_map_tree_init(&fs_info->mapping_tree); 317466d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 317566d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 317666d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 317766d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 317866d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 317966d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 318066d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 3181ba2c4d4eSJosef Bacik btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, 3182ba2c4d4eSJosef Bacik BTRFS_BLOCK_RSV_DELREFS); 3183ba2c4d4eSJosef Bacik 3184771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 31854cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 3186034f784dSJosef Bacik atomic_set(&fs_info->nr_delayed_iputs, 0); 3187fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 3188abed4aaaSJosef Bacik fs_info->global_root_tree = RB_ROOT; 318995ac567aSFilipe David Borba Manana fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 31909ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 31914cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 3192a5ed45f8SNikolay Borisov atomic64_set(&fs_info->free_chunk_space, 0); 3193f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 31948b87dc17SDavid Sterba fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 3195f8c269d7SDavid Sterba fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 3196fd708b81SJosef Bacik btrfs_init_ref_verify(fs_info); 3197c8b97818SChris Mason 3198b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 3199b34b086cSChris Mason num_online_cpus() + 2, 8); 32000afbaf8cSChris Mason 3201199c2a9cSMiao Xie INIT_LIST_HEAD(&fs_info->ordered_roots); 3202199c2a9cSMiao Xie spin_lock_init(&fs_info->ordered_root_lock); 320369fe2d75SJosef Bacik 3204638aa7edSEric Sandeen btrfs_init_scrub(fs_info); 320521adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 320621adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 320721adbd5cSStefan Behrens #endif 3208779a65a4SEric Sandeen btrfs_init_balance(fs_info); 320957056740SJosef Bacik btrfs_init_async_reclaim_work(fs_info); 3210a2de733cSArne Jansen 32110f9dd46cSJosef Bacik spin_lock_init(&fs_info->block_group_cache_lock); 32126bef4d31SEric Paris fs_info->block_group_cache_tree = RB_ROOT; 3213a1897fddSLiu Bo fs_info->first_logical_byte = (u64)-1; 32140f9dd46cSJosef Bacik 3215fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &fs_info->excluded_extents, 3216fe119a6eSNikolay Borisov IO_TREE_FS_EXCLUDED_EXTENTS, NULL); 321739279cc3SChris Mason 32185a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 3219e02119d5SChris Mason mutex_init(&fs_info->tree_log_mutex); 3220925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 3221a74a4b97SChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 3222a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 32231bbc621eSChris Mason mutex_init(&fs_info->ro_block_group_mutex); 32249e351cc8SJosef Bacik init_rwsem(&fs_info->commit_root_sem); 3225c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 322676dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 3227803b2f54SStefan Behrens sema_init(&fs_info->uuid_tree_rescan_sem, 1); 3228fa9c0d79SChris Mason 3229ad618368SEric Sandeen btrfs_init_dev_replace_locks(fs_info); 3230f9e92e40SEric Sandeen btrfs_init_qgroup(fs_info); 3231b0643e59SDennis Zhou btrfs_discard_init(fs_info); 3232416ac51dSArne Jansen 3233fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 3234fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 3235fa9c0d79SChris Mason 3236e6dcd2dcSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 3237f9295749SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 3238bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 32394854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 3240034f784dSJosef Bacik init_waitqueue_head(&fs_info->delayed_iputs_wait); 32413768f368SChris Mason 3242da17066cSJeff Mahoney /* Usable values until the real ones are cached from the superblock */ 3243da17066cSJeff Mahoney fs_info->nodesize = 4096; 3244da17066cSJeff Mahoney fs_info->sectorsize = 4096; 3245ab108d99SDavid Sterba fs_info->sectorsize_bits = ilog2(4096); 3246da17066cSJeff Mahoney fs_info->stripesize = 4096; 3247da17066cSJeff Mahoney 3248eede2bf3SOmar Sandoval spin_lock_init(&fs_info->swapfile_pins_lock); 3249eede2bf3SOmar Sandoval fs_info->swapfile_pins = RB_ROOT; 3250eede2bf3SOmar Sandoval 325118bb8bbfSJohannes Thumshirn fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH; 325218bb8bbfSJohannes Thumshirn INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work); 32538260edbaSJosef Bacik } 32548260edbaSJosef Bacik 32558260edbaSJosef Bacik static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb) 32568260edbaSJosef Bacik { 32578260edbaSJosef Bacik int ret; 32588260edbaSJosef Bacik 32598260edbaSJosef Bacik fs_info->sb = sb; 32608260edbaSJosef Bacik sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; 32618260edbaSJosef Bacik sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); 32629e967495SFilipe Manana 32635deb17e1SJosef Bacik ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL); 3264ae18c37aSJosef Bacik if (ret) 3265ae18c37aSJosef Bacik return ret; 3266ae18c37aSJosef Bacik 3267ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 3268ae18c37aSJosef Bacik if (ret) 3269c75e8394SJosef Bacik return ret; 3270ae18c37aSJosef Bacik 3271ae18c37aSJosef Bacik fs_info->dirty_metadata_batch = PAGE_SIZE * 3272ae18c37aSJosef Bacik (1 + ilog2(nr_cpu_ids)); 3273ae18c37aSJosef Bacik 3274ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 3275ae18c37aSJosef Bacik if (ret) 3276c75e8394SJosef Bacik return ret; 3277ae18c37aSJosef Bacik 3278ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, 3279ae18c37aSJosef Bacik GFP_KERNEL); 3280ae18c37aSJosef Bacik if (ret) 3281c75e8394SJosef Bacik return ret; 3282ae18c37aSJosef Bacik 3283ae18c37aSJosef Bacik fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 3284ae18c37aSJosef Bacik GFP_KERNEL); 3285c75e8394SJosef Bacik if (!fs_info->delayed_root) 3286c75e8394SJosef Bacik return -ENOMEM; 3287ae18c37aSJosef Bacik btrfs_init_delayed_root(fs_info->delayed_root); 3288ae18c37aSJosef Bacik 3289a0a1db70SFilipe Manana if (sb_rdonly(sb)) 3290a0a1db70SFilipe Manana set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state); 3291a0a1db70SFilipe Manana 3292c75e8394SJosef Bacik return btrfs_alloc_stripe_hash_table(fs_info); 3293ae18c37aSJosef Bacik } 3294ae18c37aSJosef Bacik 329597f4dd09SNikolay Borisov static int btrfs_uuid_rescan_kthread(void *data) 329697f4dd09SNikolay Borisov { 329797f4dd09SNikolay Borisov struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; 329897f4dd09SNikolay Borisov int ret; 329997f4dd09SNikolay Borisov 330097f4dd09SNikolay Borisov /* 330197f4dd09SNikolay Borisov * 1st step is to iterate through the existing UUID tree and 330297f4dd09SNikolay Borisov * to delete all entries that contain outdated data. 330397f4dd09SNikolay Borisov * 2nd step is to add all missing entries to the UUID tree. 330497f4dd09SNikolay Borisov */ 330597f4dd09SNikolay Borisov ret = btrfs_uuid_tree_iterate(fs_info); 330697f4dd09SNikolay Borisov if (ret < 0) { 3307c94bec2cSJosef Bacik if (ret != -EINTR) 3308c94bec2cSJosef Bacik btrfs_warn(fs_info, "iterating uuid_tree failed %d", 3309c94bec2cSJosef Bacik ret); 331097f4dd09SNikolay Borisov up(&fs_info->uuid_tree_rescan_sem); 331197f4dd09SNikolay Borisov return ret; 331297f4dd09SNikolay Borisov } 331397f4dd09SNikolay Borisov return btrfs_uuid_scan_kthread(data); 331497f4dd09SNikolay Borisov } 331597f4dd09SNikolay Borisov 331697f4dd09SNikolay Borisov static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) 331797f4dd09SNikolay Borisov { 331897f4dd09SNikolay Borisov struct task_struct *task; 331997f4dd09SNikolay Borisov 332097f4dd09SNikolay Borisov down(&fs_info->uuid_tree_rescan_sem); 332197f4dd09SNikolay Borisov task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); 332297f4dd09SNikolay Borisov if (IS_ERR(task)) { 332397f4dd09SNikolay Borisov /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 332497f4dd09SNikolay Borisov btrfs_warn(fs_info, "failed to start uuid_rescan task"); 332597f4dd09SNikolay Borisov up(&fs_info->uuid_tree_rescan_sem); 332697f4dd09SNikolay Borisov return PTR_ERR(task); 332797f4dd09SNikolay Borisov } 332897f4dd09SNikolay Borisov 332997f4dd09SNikolay Borisov return 0; 333097f4dd09SNikolay Borisov } 333197f4dd09SNikolay Borisov 333244c0ca21SBoris Burkov /* 33338cd29088SBoris Burkov * Some options only have meaning at mount time and shouldn't persist across 33348cd29088SBoris Burkov * remounts, or be displayed. Clear these at the end of mount and remount 33358cd29088SBoris Burkov * code paths. 33368cd29088SBoris Burkov */ 33378cd29088SBoris Burkov void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info) 33388cd29088SBoris Burkov { 33398cd29088SBoris Burkov btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT); 33408b228324SBoris Burkov btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE); 33418cd29088SBoris Burkov } 33428cd29088SBoris Burkov 33438cd29088SBoris Burkov /* 334444c0ca21SBoris Burkov * Mounting logic specific to read-write file systems. Shared by open_ctree 334544c0ca21SBoris Burkov * and btrfs_remount when remounting from read-only to read-write. 334644c0ca21SBoris Burkov */ 334744c0ca21SBoris Burkov int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info) 334844c0ca21SBoris Burkov { 334944c0ca21SBoris Burkov int ret; 335094846229SBoris Burkov const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE); 33518b228324SBoris Burkov bool clear_free_space_tree = false; 33528b228324SBoris Burkov 33538b228324SBoris Burkov if (btrfs_test_opt(fs_info, CLEAR_CACHE) && 33548b228324SBoris Burkov btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 33558b228324SBoris Burkov clear_free_space_tree = true; 33568b228324SBoris Burkov } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 33578b228324SBoris Burkov !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { 33588b228324SBoris Burkov btrfs_warn(fs_info, "free space tree is invalid"); 33598b228324SBoris Burkov clear_free_space_tree = true; 33608b228324SBoris Burkov } 33618b228324SBoris Burkov 33628b228324SBoris Burkov if (clear_free_space_tree) { 33638b228324SBoris Burkov btrfs_info(fs_info, "clearing free space tree"); 33648b228324SBoris Burkov ret = btrfs_clear_free_space_tree(fs_info); 33658b228324SBoris Burkov if (ret) { 33668b228324SBoris Burkov btrfs_warn(fs_info, 33678b228324SBoris Burkov "failed to clear free space tree: %d", ret); 33688b228324SBoris Burkov goto out; 33698b228324SBoris Burkov } 33708b228324SBoris Burkov } 337144c0ca21SBoris Burkov 33728d488a8cSFilipe Manana /* 33738d488a8cSFilipe Manana * btrfs_find_orphan_roots() is responsible for finding all the dead 33748d488a8cSFilipe Manana * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load 33758d488a8cSFilipe Manana * them into the fs_info->fs_roots_radix tree. This must be done before 33768d488a8cSFilipe Manana * calling btrfs_orphan_cleanup() on the tree root. If we don't do it 33778d488a8cSFilipe Manana * first, then btrfs_orphan_cleanup() will delete a dead root's orphan 33788d488a8cSFilipe Manana * item before the root's tree is deleted - this means that if we unmount 33798d488a8cSFilipe Manana * or crash before the deletion completes, on the next mount we will not 33808d488a8cSFilipe Manana * delete what remains of the tree because the orphan item does not 33818d488a8cSFilipe Manana * exists anymore, which is what tells us we have a pending deletion. 33828d488a8cSFilipe Manana */ 33838d488a8cSFilipe Manana ret = btrfs_find_orphan_roots(fs_info); 33848d488a8cSFilipe Manana if (ret) 33858d488a8cSFilipe Manana goto out; 33868d488a8cSFilipe Manana 338744c0ca21SBoris Burkov ret = btrfs_cleanup_fs_roots(fs_info); 338844c0ca21SBoris Burkov if (ret) 338944c0ca21SBoris Burkov goto out; 339044c0ca21SBoris Burkov 33918f1c21d7SBoris Burkov down_read(&fs_info->cleanup_work_sem); 33928f1c21d7SBoris Burkov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 33938f1c21d7SBoris Burkov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 33948f1c21d7SBoris Burkov up_read(&fs_info->cleanup_work_sem); 33958f1c21d7SBoris Burkov goto out; 33968f1c21d7SBoris Burkov } 33978f1c21d7SBoris Burkov up_read(&fs_info->cleanup_work_sem); 33988f1c21d7SBoris Burkov 339944c0ca21SBoris Burkov mutex_lock(&fs_info->cleaner_mutex); 34007eefae6bSJosef Bacik ret = btrfs_recover_relocation(fs_info); 340144c0ca21SBoris Burkov mutex_unlock(&fs_info->cleaner_mutex); 340244c0ca21SBoris Burkov if (ret < 0) { 340344c0ca21SBoris Burkov btrfs_warn(fs_info, "failed to recover relocation: %d", ret); 340444c0ca21SBoris Burkov goto out; 340544c0ca21SBoris Burkov } 340644c0ca21SBoris Burkov 34075011139aSBoris Burkov if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && 34085011139aSBoris Burkov !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 34095011139aSBoris Burkov btrfs_info(fs_info, "creating free space tree"); 34105011139aSBoris Burkov ret = btrfs_create_free_space_tree(fs_info); 34115011139aSBoris Burkov if (ret) { 34125011139aSBoris Burkov btrfs_warn(fs_info, 34135011139aSBoris Burkov "failed to create free space tree: %d", ret); 34145011139aSBoris Burkov goto out; 34155011139aSBoris Burkov } 34165011139aSBoris Burkov } 34175011139aSBoris Burkov 341894846229SBoris Burkov if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) { 341994846229SBoris Burkov ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt); 342094846229SBoris Burkov if (ret) 342194846229SBoris Burkov goto out; 342294846229SBoris Burkov } 342394846229SBoris Burkov 342444c0ca21SBoris Burkov ret = btrfs_resume_balance_async(fs_info); 342544c0ca21SBoris Burkov if (ret) 342644c0ca21SBoris Burkov goto out; 342744c0ca21SBoris Burkov 342844c0ca21SBoris Burkov ret = btrfs_resume_dev_replace_async(fs_info); 342944c0ca21SBoris Burkov if (ret) { 343044c0ca21SBoris Burkov btrfs_warn(fs_info, "failed to resume dev_replace"); 343144c0ca21SBoris Burkov goto out; 343244c0ca21SBoris Burkov } 343344c0ca21SBoris Burkov 343444c0ca21SBoris Burkov btrfs_qgroup_rescan_resume(fs_info); 343544c0ca21SBoris Burkov 343644c0ca21SBoris Burkov if (!fs_info->uuid_root) { 343744c0ca21SBoris Burkov btrfs_info(fs_info, "creating UUID tree"); 343844c0ca21SBoris Burkov ret = btrfs_create_uuid_tree(fs_info); 343944c0ca21SBoris Burkov if (ret) { 344044c0ca21SBoris Burkov btrfs_warn(fs_info, 344144c0ca21SBoris Burkov "failed to create the UUID tree %d", ret); 344244c0ca21SBoris Burkov goto out; 344344c0ca21SBoris Burkov } 344444c0ca21SBoris Burkov } 344544c0ca21SBoris Burkov 344644c0ca21SBoris Burkov out: 344744c0ca21SBoris Burkov return ret; 344844c0ca21SBoris Burkov } 344944c0ca21SBoris Burkov 3450ae18c37aSJosef Bacik int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices, 3451ae18c37aSJosef Bacik char *options) 3452ae18c37aSJosef Bacik { 3453ae18c37aSJosef Bacik u32 sectorsize; 3454ae18c37aSJosef Bacik u32 nodesize; 3455ae18c37aSJosef Bacik u32 stripesize; 3456ae18c37aSJosef Bacik u64 generation; 3457ae18c37aSJosef Bacik u64 features; 3458ae18c37aSJosef Bacik u16 csum_type; 3459ae18c37aSJosef Bacik struct btrfs_super_block *disk_super; 3460ae18c37aSJosef Bacik struct btrfs_fs_info *fs_info = btrfs_sb(sb); 3461ae18c37aSJosef Bacik struct btrfs_root *tree_root; 3462ae18c37aSJosef Bacik struct btrfs_root *chunk_root; 3463ae18c37aSJosef Bacik int ret; 3464ae18c37aSJosef Bacik int err = -EINVAL; 3465ae18c37aSJosef Bacik int level; 3466ae18c37aSJosef Bacik 34678260edbaSJosef Bacik ret = init_mount_fs_info(fs_info, sb); 346853b381b3SDavid Woodhouse if (ret) { 346983c8266aSDavid Sterba err = ret; 3470ae18c37aSJosef Bacik goto fail; 347153b381b3SDavid Woodhouse } 347253b381b3SDavid Woodhouse 3473ae18c37aSJosef Bacik /* These need to be init'ed before we start creating inodes and such. */ 3474ae18c37aSJosef Bacik tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, 3475ae18c37aSJosef Bacik GFP_KERNEL); 3476ae18c37aSJosef Bacik fs_info->tree_root = tree_root; 3477ae18c37aSJosef Bacik chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID, 3478ae18c37aSJosef Bacik GFP_KERNEL); 3479ae18c37aSJosef Bacik fs_info->chunk_root = chunk_root; 3480ae18c37aSJosef Bacik if (!tree_root || !chunk_root) { 3481ae18c37aSJosef Bacik err = -ENOMEM; 3482c75e8394SJosef Bacik goto fail; 3483ae18c37aSJosef Bacik } 3484ae18c37aSJosef Bacik 3485ae18c37aSJosef Bacik fs_info->btree_inode = new_inode(sb); 3486ae18c37aSJosef Bacik if (!fs_info->btree_inode) { 3487ae18c37aSJosef Bacik err = -ENOMEM; 3488c75e8394SJosef Bacik goto fail; 3489ae18c37aSJosef Bacik } 3490ae18c37aSJosef Bacik mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 3491ae18c37aSJosef Bacik btrfs_init_btree_inode(fs_info); 3492ae18c37aSJosef Bacik 3493d24fa5c1SAnand Jain invalidate_bdev(fs_devices->latest_dev->bdev); 34941104a885SDavid Sterba 34951104a885SDavid Sterba /* 34961104a885SDavid Sterba * Read super block and check the signature bytes only 34971104a885SDavid Sterba */ 3498d24fa5c1SAnand Jain disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev); 34998f32380dSJohannes Thumshirn if (IS_ERR(disk_super)) { 35008f32380dSJohannes Thumshirn err = PTR_ERR(disk_super); 350116cdcec7SMiao Xie goto fail_alloc; 350220b45077SDave Young } 350339279cc3SChris Mason 35041104a885SDavid Sterba /* 3505260db43cSRandy Dunlap * Verify the type first, if that or the checksum value are 35068dc3f22cSJohannes Thumshirn * corrupted, we'll find out 35078dc3f22cSJohannes Thumshirn */ 35088f32380dSJohannes Thumshirn csum_type = btrfs_super_csum_type(disk_super); 350951bce6c9SJohannes Thumshirn if (!btrfs_supported_super_csum(csum_type)) { 35108dc3f22cSJohannes Thumshirn btrfs_err(fs_info, "unsupported checksum algorithm: %u", 351151bce6c9SJohannes Thumshirn csum_type); 35128dc3f22cSJohannes Thumshirn err = -EINVAL; 35138f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 35148dc3f22cSJohannes Thumshirn goto fail_alloc; 35158dc3f22cSJohannes Thumshirn } 35168dc3f22cSJohannes Thumshirn 351783c68bbcSSu Yue fs_info->csum_size = btrfs_super_csum_size(disk_super); 351883c68bbcSSu Yue 35196d97c6e3SJohannes Thumshirn ret = btrfs_init_csum_hash(fs_info, csum_type); 35206d97c6e3SJohannes Thumshirn if (ret) { 35216d97c6e3SJohannes Thumshirn err = ret; 35228f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 35236d97c6e3SJohannes Thumshirn goto fail_alloc; 35246d97c6e3SJohannes Thumshirn } 35256d97c6e3SJohannes Thumshirn 35268dc3f22cSJohannes Thumshirn /* 35271104a885SDavid Sterba * We want to check superblock checksum, the type is stored inside. 35281104a885SDavid Sterba * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 35291104a885SDavid Sterba */ 35308f32380dSJohannes Thumshirn if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) { 353105135f59SDavid Sterba btrfs_err(fs_info, "superblock checksum mismatch"); 35321104a885SDavid Sterba err = -EINVAL; 35338f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 3534141386e1SJosef Bacik goto fail_alloc; 35351104a885SDavid Sterba } 35361104a885SDavid Sterba 35371104a885SDavid Sterba /* 35381104a885SDavid Sterba * super_copy is zeroed at allocation time and we never touch the 35391104a885SDavid Sterba * following bytes up to INFO_SIZE, the checksum is calculated from 35401104a885SDavid Sterba * the whole block of INFO_SIZE 35411104a885SDavid Sterba */ 35428f32380dSJohannes Thumshirn memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy)); 35438f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 35445f39d397SChris Mason 3545fbc6feaeSNikolay Borisov disk_super = fs_info->super_copy; 3546fbc6feaeSNikolay Borisov 35470b86a832SChris Mason 3548fbc6feaeSNikolay Borisov features = btrfs_super_flags(disk_super); 3549fbc6feaeSNikolay Borisov if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { 3550fbc6feaeSNikolay Borisov features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; 3551fbc6feaeSNikolay Borisov btrfs_set_super_flags(disk_super, features); 3552fbc6feaeSNikolay Borisov btrfs_info(fs_info, 3553fbc6feaeSNikolay Borisov "found metadata UUID change in progress flag, clearing"); 3554fbc6feaeSNikolay Borisov } 3555fbc6feaeSNikolay Borisov 3556fbc6feaeSNikolay Borisov memcpy(fs_info->super_for_commit, fs_info->super_copy, 3557fbc6feaeSNikolay Borisov sizeof(*fs_info->super_for_commit)); 3558de37aa51SNikolay Borisov 3559069ec957SQu Wenruo ret = btrfs_validate_mount_super(fs_info); 35601104a885SDavid Sterba if (ret) { 356105135f59SDavid Sterba btrfs_err(fs_info, "superblock contains fatal errors"); 35621104a885SDavid Sterba err = -EINVAL; 3563141386e1SJosef Bacik goto fail_alloc; 35641104a885SDavid Sterba } 35651104a885SDavid Sterba 35660f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 3567141386e1SJosef Bacik goto fail_alloc; 35680f7d52f4SChris Mason 3569acce952bSliubo /* check FS state, whether FS is broken. */ 357087533c47SMiao Xie if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 357187533c47SMiao Xie set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 3572acce952bSliubo 357375e7cb7fSLiu Bo /* 357475e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 357575e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 357675e7cb7fSLiu Bo */ 357775e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 357875e7cb7fSLiu Bo 35796f93e834SAnand Jain /* 35806f93e834SAnand Jain * Flag our filesystem as having big metadata blocks if they are bigger 35816f93e834SAnand Jain * than the page size. 35826f93e834SAnand Jain */ 35836f93e834SAnand Jain if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) { 35846f93e834SAnand Jain if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 35856f93e834SAnand Jain btrfs_info(fs_info, 35866f93e834SAnand Jain "flagging fs with big metadata feature"); 35876f93e834SAnand Jain features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 35886f93e834SAnand Jain } 35896f93e834SAnand Jain 35906f93e834SAnand Jain /* Set up fs_info before parsing mount options */ 35916f93e834SAnand Jain nodesize = btrfs_super_nodesize(disk_super); 35926f93e834SAnand Jain sectorsize = btrfs_super_sectorsize(disk_super); 35936f93e834SAnand Jain stripesize = sectorsize; 35946f93e834SAnand Jain fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 35956f93e834SAnand Jain fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 35966f93e834SAnand Jain 35976f93e834SAnand Jain fs_info->nodesize = nodesize; 35986f93e834SAnand Jain fs_info->sectorsize = sectorsize; 35996f93e834SAnand Jain fs_info->sectorsize_bits = ilog2(sectorsize); 36006f93e834SAnand Jain fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size; 36016f93e834SAnand Jain fs_info->stripesize = stripesize; 36026f93e834SAnand Jain 36032ff7e61eSJeff Mahoney ret = btrfs_parse_options(fs_info, options, sb->s_flags); 36042b82032cSYan Zheng if (ret) { 36052b82032cSYan Zheng err = ret; 3606141386e1SJosef Bacik goto fail_alloc; 36072b82032cSYan Zheng } 3608dfe25020SChris Mason 3609f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 3610f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 3611f2b636e8SJosef Bacik if (features) { 361205135f59SDavid Sterba btrfs_err(fs_info, 361305135f59SDavid Sterba "cannot mount because of unsupported optional features (%llx)", 3614c1c9ff7cSGeert Uytterhoeven features); 3615f2b636e8SJosef Bacik err = -EINVAL; 3616141386e1SJosef Bacik goto fail_alloc; 3617f2b636e8SJosef Bacik } 3618f2b636e8SJosef Bacik 36195d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 36205d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 36210b246afaSJeff Mahoney if (fs_info->compress_type == BTRFS_COMPRESS_LZO) 3622a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 36235c1aab1dSNick Terrell else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) 36245c1aab1dSNick Terrell features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; 3625727011e0SChris Mason 36263173a18fSJosef Bacik if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) 362705135f59SDavid Sterba btrfs_info(fs_info, "has skinny extents"); 36283173a18fSJosef Bacik 3629727011e0SChris Mason /* 3630bc3f116fSChris Mason * mixed block groups end up with duplicate but slightly offset 3631bc3f116fSChris Mason * extent buffers for the same range. It leads to corruptions 3632bc3f116fSChris Mason */ 3633bc3f116fSChris Mason if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 3634707e8a07SDavid Sterba (sectorsize != nodesize)) { 363505135f59SDavid Sterba btrfs_err(fs_info, 363605135f59SDavid Sterba "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", 363705135f59SDavid Sterba nodesize, sectorsize); 3638141386e1SJosef Bacik goto fail_alloc; 3639bc3f116fSChris Mason } 3640bc3f116fSChris Mason 3641ceda0864SMiao Xie /* 3642ceda0864SMiao Xie * Needn't use the lock because there is no other task which will 3643ceda0864SMiao Xie * update the flag. 3644ceda0864SMiao Xie */ 36455d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 36465d4f98a2SYan Zheng 3647f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 3648f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 3649bc98a42cSDavid Howells if (!sb_rdonly(sb) && features) { 365005135f59SDavid Sterba btrfs_err(fs_info, 365105135f59SDavid Sterba "cannot mount read-write because of unsupported optional features (%llx)", 3652c1c9ff7cSGeert Uytterhoeven features); 3653f2b636e8SJosef Bacik err = -EINVAL; 3654141386e1SJosef Bacik goto fail_alloc; 3655f2b636e8SJosef Bacik } 365661d92c32SChris Mason 36578481dd80SQu Wenruo if (sectorsize < PAGE_SIZE) { 36588481dd80SQu Wenruo struct btrfs_subpage_info *subpage_info; 36598481dd80SQu Wenruo 36609f73f1aeSQu Wenruo /* 36619f73f1aeSQu Wenruo * V1 space cache has some hardcoded PAGE_SIZE usage, and is 36629f73f1aeSQu Wenruo * going to be deprecated. 36639f73f1aeSQu Wenruo * 36649f73f1aeSQu Wenruo * Force to use v2 cache for subpage case. 36659f73f1aeSQu Wenruo */ 36669f73f1aeSQu Wenruo btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE); 36679f73f1aeSQu Wenruo btrfs_set_and_info(fs_info, FREE_SPACE_TREE, 36689f73f1aeSQu Wenruo "forcing free space tree for sector size %u with page size %lu", 36699f73f1aeSQu Wenruo sectorsize, PAGE_SIZE); 36709f73f1aeSQu Wenruo 367195ea0486SQu Wenruo btrfs_warn(fs_info, 367295ea0486SQu Wenruo "read-write for sector size %u with page size %lu is experimental", 36730bb3eb3eSQu Wenruo sectorsize, PAGE_SIZE); 3674c8050b3bSQu Wenruo if (btrfs_super_incompat_flags(fs_info->super_copy) & 3675c8050b3bSQu Wenruo BTRFS_FEATURE_INCOMPAT_RAID56) { 3676c8050b3bSQu Wenruo btrfs_err(fs_info, 3677c8050b3bSQu Wenruo "RAID56 is not yet supported for sector size %u with page size %lu", 3678c8050b3bSQu Wenruo sectorsize, PAGE_SIZE); 3679c8050b3bSQu Wenruo err = -EINVAL; 3680c8050b3bSQu Wenruo goto fail_alloc; 3681c8050b3bSQu Wenruo } 36828481dd80SQu Wenruo subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL); 36838481dd80SQu Wenruo if (!subpage_info) 36848481dd80SQu Wenruo goto fail_alloc; 36858481dd80SQu Wenruo btrfs_init_subpage_info(subpage_info, sectorsize); 36868481dd80SQu Wenruo fs_info->subpage_info = subpage_info; 3687c8050b3bSQu Wenruo } 36880bb3eb3eSQu Wenruo 3689d21deec5SSu Yue ret = btrfs_init_workqueues(fs_info); 36902a458198SEric Sandeen if (ret) { 36912a458198SEric Sandeen err = ret; 36920dc3b84aSJosef Bacik goto fail_sb_buffer; 36930dc3b84aSJosef Bacik } 36944543df7eSChris Mason 36959e11ceeeSJan Kara sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); 36969e11ceeeSJan Kara sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); 36974575c9ccSChris Mason 3698a061fc8dSChris Mason sb->s_blocksize = sectorsize; 3699a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 3700de37aa51SNikolay Borisov memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); 3701db94535dSChris Mason 3702925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 37036bccf3abSJeff Mahoney ret = btrfs_read_sys_array(fs_info); 3704925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 370584eed90fSChris Mason if (ret) { 370605135f59SDavid Sterba btrfs_err(fs_info, "failed to read the system array: %d", ret); 37075d4f98a2SYan Zheng goto fail_sb_buffer; 370884eed90fSChris Mason } 37090b86a832SChris Mason 371084234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 3711581c1760SQu Wenruo level = btrfs_super_chunk_root_level(disk_super); 3712bd676446SJosef Bacik ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super), 3713bd676446SJosef Bacik generation, level); 3714bd676446SJosef Bacik if (ret) { 371505135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk root"); 3716af31f5e5SChris Mason goto fail_tree_roots; 371783121942SDavid Woodhouse } 37180b86a832SChris Mason 3719e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 3720c4ac7541SDavid Sterba offsetof(struct btrfs_header, chunk_tree_uuid), 3721c4ac7541SDavid Sterba BTRFS_UUID_SIZE); 3722e17cade2SChris Mason 37235b4aacefSJeff Mahoney ret = btrfs_read_chunk_tree(fs_info); 37242b82032cSYan Zheng if (ret) { 372505135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk tree: %d", ret); 3726af31f5e5SChris Mason goto fail_tree_roots; 37272b82032cSYan Zheng } 37280b86a832SChris Mason 37298dabb742SStefan Behrens /* 3730bacce86aSAnand Jain * At this point we know all the devices that make this filesystem, 3731bacce86aSAnand Jain * including the seed devices but we don't know yet if the replace 3732bacce86aSAnand Jain * target is required. So free devices that are not part of this 37331a9fd417SDavid Sterba * filesystem but skip the replace target device which is checked 3734bacce86aSAnand Jain * below in btrfs_init_dev_replace(). 37358dabb742SStefan Behrens */ 3736bacce86aSAnand Jain btrfs_free_extra_devids(fs_devices); 3737d24fa5c1SAnand Jain if (!fs_devices->latest_dev->bdev) { 373805135f59SDavid Sterba btrfs_err(fs_info, "failed to read devices"); 3739a6b0d5c8SChris Mason goto fail_tree_roots; 3740a6b0d5c8SChris Mason } 3741a6b0d5c8SChris Mason 3742b8522a1eSNikolay Borisov ret = init_tree_roots(fs_info); 37434bbcaa64SEric Sandeen if (ret) 3744b8522a1eSNikolay Borisov goto fail_tree_roots; 37458929ecfaSYan, Zheng 374675ec1db8SJosef Bacik /* 374773651042SNaohiro Aota * Get zone type information of zoned block devices. This will also 374873651042SNaohiro Aota * handle emulation of a zoned filesystem if a regular device has the 374973651042SNaohiro Aota * zoned incompat feature flag set. 375073651042SNaohiro Aota */ 375173651042SNaohiro Aota ret = btrfs_get_dev_zone_info_all_devices(fs_info); 375273651042SNaohiro Aota if (ret) { 375373651042SNaohiro Aota btrfs_err(fs_info, 375473651042SNaohiro Aota "zoned: failed to read device zone info: %d", 375573651042SNaohiro Aota ret); 375673651042SNaohiro Aota goto fail_block_groups; 375773651042SNaohiro Aota } 375873651042SNaohiro Aota 375973651042SNaohiro Aota /* 376075ec1db8SJosef Bacik * If we have a uuid root and we're not being told to rescan we need to 376175ec1db8SJosef Bacik * check the generation here so we can set the 376275ec1db8SJosef Bacik * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the 376375ec1db8SJosef Bacik * transaction during a balance or the log replay without updating the 376475ec1db8SJosef Bacik * uuid generation, and then if we crash we would rescan the uuid tree, 376575ec1db8SJosef Bacik * even though it was perfectly fine. 376675ec1db8SJosef Bacik */ 376775ec1db8SJosef Bacik if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) && 376875ec1db8SJosef Bacik fs_info->generation == btrfs_super_uuid_tree_generation(disk_super)) 376975ec1db8SJosef Bacik set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 377075ec1db8SJosef Bacik 3771cf90d884SQu Wenruo ret = btrfs_verify_dev_extents(fs_info); 3772cf90d884SQu Wenruo if (ret) { 3773cf90d884SQu Wenruo btrfs_err(fs_info, 3774cf90d884SQu Wenruo "failed to verify dev extents against chunks: %d", 3775cf90d884SQu Wenruo ret); 3776cf90d884SQu Wenruo goto fail_block_groups; 3777cf90d884SQu Wenruo } 377868310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 377968310a5eSIlya Dryomov if (ret) { 378005135f59SDavid Sterba btrfs_err(fs_info, "failed to recover balance: %d", ret); 378168310a5eSIlya Dryomov goto fail_block_groups; 378268310a5eSIlya Dryomov } 378368310a5eSIlya Dryomov 3784733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 3785733f4fbbSStefan Behrens if (ret) { 378605135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_stats: %d", ret); 3787733f4fbbSStefan Behrens goto fail_block_groups; 3788733f4fbbSStefan Behrens } 3789733f4fbbSStefan Behrens 37908dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 37918dabb742SStefan Behrens if (ret) { 379205135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_replace: %d", ret); 37938dabb742SStefan Behrens goto fail_block_groups; 37948dabb742SStefan Behrens } 37958dabb742SStefan Behrens 3796b70f5097SNaohiro Aota ret = btrfs_check_zoned_mode(fs_info); 3797b70f5097SNaohiro Aota if (ret) { 3798b70f5097SNaohiro Aota btrfs_err(fs_info, "failed to initialize zoned mode: %d", 3799b70f5097SNaohiro Aota ret); 3800b70f5097SNaohiro Aota goto fail_block_groups; 3801b70f5097SNaohiro Aota } 3802b70f5097SNaohiro Aota 3803c6761a9eSAnand Jain ret = btrfs_sysfs_add_fsid(fs_devices); 3804b7c35e81SAnand Jain if (ret) { 380505135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", 380605135f59SDavid Sterba ret); 3807b7c35e81SAnand Jain goto fail_block_groups; 3808b7c35e81SAnand Jain } 3809b7c35e81SAnand Jain 381096f3136eSAnand Jain ret = btrfs_sysfs_add_mounted(fs_info); 38115ac1d209SJeff Mahoney if (ret) { 381205135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); 3813b7c35e81SAnand Jain goto fail_fsdev_sysfs; 38145ac1d209SJeff Mahoney } 38155ac1d209SJeff Mahoney 3816c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 3817c59021f8Sliubo if (ret) { 381805135f59SDavid Sterba btrfs_err(fs_info, "failed to initialize space info: %d", ret); 38192365dd3cSAnand Jain goto fail_sysfs; 3820c59021f8Sliubo } 3821c59021f8Sliubo 38225b4aacefSJeff Mahoney ret = btrfs_read_block_groups(fs_info); 38231b1d1f66SJosef Bacik if (ret) { 382405135f59SDavid Sterba btrfs_err(fs_info, "failed to read block groups: %d", ret); 38252365dd3cSAnand Jain goto fail_sysfs; 38261b1d1f66SJosef Bacik } 38274330e183SQu Wenruo 382816beac87SNaohiro Aota btrfs_free_zone_cache(fs_info); 382916beac87SNaohiro Aota 38305c78a5e7SAnand Jain if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices && 38315c78a5e7SAnand Jain !btrfs_check_rw_degradable(fs_info, NULL)) { 383205135f59SDavid Sterba btrfs_warn(fs_info, 383352042d8eSAndrea Gelmini "writable mount is not allowed due to too many missing devices"); 38342365dd3cSAnand Jain goto fail_sysfs; 3835292fd7fcSStefan Behrens } 38369078a3e1SChris Mason 383733c44184SJosef Bacik fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info, 3838a74a4b97SChris Mason "btrfs-cleaner"); 383957506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 38402365dd3cSAnand Jain goto fail_sysfs; 3841a74a4b97SChris Mason 3842a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 3843a74a4b97SChris Mason tree_root, 3844a74a4b97SChris Mason "btrfs-transaction"); 384557506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 38463f157a2fSChris Mason goto fail_cleaner; 3847a74a4b97SChris Mason 3848583b7231SHans van Kranenburg if (!btrfs_test_opt(fs_info, NOSSD) && 3849c289811cSChris Mason !fs_info->fs_devices->rotating) { 3850583b7231SHans van Kranenburg btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); 3851c289811cSChris Mason } 3852c289811cSChris Mason 3853572d9ab7SDavid Sterba /* 385401327610SNicholas D Steeves * Mount does not set all options immediately, we can do it now and do 3855572d9ab7SDavid Sterba * not have to wait for transaction commit 3856572d9ab7SDavid Sterba */ 3857572d9ab7SDavid Sterba btrfs_apply_pending_changes(fs_info); 38583818aea2SQu Wenruo 385921adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 38600b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { 38612ff7e61eSJeff Mahoney ret = btrfsic_mount(fs_info, fs_devices, 38620b246afaSJeff Mahoney btrfs_test_opt(fs_info, 3863cbeaae4fSDavid Sterba CHECK_INTEGRITY_DATA) ? 1 : 0, 386421adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 386521adbd5cSStefan Behrens if (ret) 386605135f59SDavid Sterba btrfs_warn(fs_info, 386705135f59SDavid Sterba "failed to initialize integrity check module: %d", 386805135f59SDavid Sterba ret); 386921adbd5cSStefan Behrens } 387021adbd5cSStefan Behrens #endif 3871bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 3872bcef60f2SArne Jansen if (ret) 3873bcef60f2SArne Jansen goto fail_trans_kthread; 387421adbd5cSStefan Behrens 3875fd708b81SJosef Bacik if (btrfs_build_ref_tree(fs_info)) 3876fd708b81SJosef Bacik btrfs_err(fs_info, "couldn't build ref tree"); 3877fd708b81SJosef Bacik 387896da0919SQu Wenruo /* do not make disk changes in broken FS or nologreplay is given */ 387996da0919SQu Wenruo if (btrfs_super_log_root(disk_super) != 0 && 38800b246afaSJeff Mahoney !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 3881e8294f2fSDavid Sterba btrfs_info(fs_info, "start tree-log replay"); 388263443bf5SEric Sandeen ret = btrfs_replay_log(fs_info, fs_devices); 388379787eaaSJeff Mahoney if (ret) { 388463443bf5SEric Sandeen err = ret; 388528c16cbbSWang Shilong goto fail_qgroup; 3886e556ce2cSYan Zheng } 3887e02119d5SChris Mason } 38881a40e23bSZheng Yan 388956e9357aSDavid Sterba fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true); 38903140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 38913140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 3892f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read fs tree: %d", err); 3893315bf8efSJosef Bacik fs_info->fs_root = NULL; 3894bcef60f2SArne Jansen goto fail_qgroup; 38953140c9a3SDan Carpenter } 3896c289811cSChris Mason 3897bc98a42cSDavid Howells if (sb_rdonly(sb)) 38988cd29088SBoris Burkov goto clear_oneshot; 38992b6ba629SIlya Dryomov 390044c0ca21SBoris Burkov ret = btrfs_start_pre_rw_mount(fs_info); 39012b6ba629SIlya Dryomov if (ret) { 39026bccf3abSJeff Mahoney close_ctree(fs_info); 39032b6ba629SIlya Dryomov return ret; 3904e3acc2a6SJosef Bacik } 3905b0643e59SDennis Zhou btrfs_discard_resume(fs_info); 3906b382a324SJan Schmidt 390744c0ca21SBoris Burkov if (fs_info->uuid_root && 390844c0ca21SBoris Burkov (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || 390944c0ca21SBoris Burkov fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) { 391005135f59SDavid Sterba btrfs_info(fs_info, "checking UUID tree"); 391170f80175SStefan Behrens ret = btrfs_check_uuid_tree(fs_info); 391270f80175SStefan Behrens if (ret) { 391305135f59SDavid Sterba btrfs_warn(fs_info, 391405135f59SDavid Sterba "failed to check the UUID tree: %d", ret); 39156bccf3abSJeff Mahoney close_ctree(fs_info); 391670f80175SStefan Behrens return ret; 391770f80175SStefan Behrens } 3918f7a81ea4SStefan Behrens } 391994846229SBoris Burkov 3920afcdd129SJosef Bacik set_bit(BTRFS_FS_OPEN, &fs_info->flags); 392147ab2a6cSJosef Bacik 3922b4be6aefSJosef Bacik /* Kick the cleaner thread so it'll start deleting snapshots. */ 3923b4be6aefSJosef Bacik if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags)) 3924b4be6aefSJosef Bacik wake_up_process(fs_info->cleaner_kthread); 3925b4be6aefSJosef Bacik 39268cd29088SBoris Burkov clear_oneshot: 39278cd29088SBoris Burkov btrfs_clear_oneshot_options(fs_info); 3928ad2b2c80SAl Viro return 0; 392939279cc3SChris Mason 3930bcef60f2SArne Jansen fail_qgroup: 3931bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 39327c2ca468SChris Mason fail_trans_kthread: 39337c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 39342ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 3935faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 39363f157a2fSChris Mason fail_cleaner: 3937a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 39387c2ca468SChris Mason 39397c2ca468SChris Mason /* 39407c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 39417c2ca468SChris Mason * kthreads 39427c2ca468SChris Mason */ 39437c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 39447c2ca468SChris Mason 39452365dd3cSAnand Jain fail_sysfs: 39466618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 39472365dd3cSAnand Jain 3948b7c35e81SAnand Jain fail_fsdev_sysfs: 3949b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3950b7c35e81SAnand Jain 39511b1d1f66SJosef Bacik fail_block_groups: 395254067ae9SJosef Bacik btrfs_put_block_group_cache(fs_info); 3953af31f5e5SChris Mason 3954af31f5e5SChris Mason fail_tree_roots: 39559e3aa805SJosef Bacik if (fs_info->data_reloc_root) 39569e3aa805SJosef Bacik btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root); 39574273eaffSAnand Jain free_root_pointers(fs_info, true); 39582b8195bbSMiao Xie invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3959af31f5e5SChris Mason 396039279cc3SChris Mason fail_sb_buffer: 39617abadb64SLiu Bo btrfs_stop_all_workers(fs_info); 39625cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 396316cdcec7SMiao Xie fail_alloc: 3964586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 3965586e46e2SIlya Dryomov 39664543df7eSChris Mason iput(fs_info->btree_inode); 39677e662854SQinghuang Feng fail: 3968586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 3969ad2b2c80SAl Viro return err; 3970eb60ceacSChris Mason } 3971663faf9fSMasami Hiramatsu ALLOW_ERROR_INJECTION(open_ctree, ERRNO); 3972eb60ceacSChris Mason 3973314b6dd0SJohannes Thumshirn static void btrfs_end_super_write(struct bio *bio) 3974f2984462SChris Mason { 3975314b6dd0SJohannes Thumshirn struct btrfs_device *device = bio->bi_private; 3976314b6dd0SJohannes Thumshirn struct bio_vec *bvec; 3977314b6dd0SJohannes Thumshirn struct bvec_iter_all iter_all; 3978314b6dd0SJohannes Thumshirn struct page *page; 3979442a4f63SStefan Behrens 3980314b6dd0SJohannes Thumshirn bio_for_each_segment_all(bvec, bio, iter_all) { 3981314b6dd0SJohannes Thumshirn page = bvec->bv_page; 3982314b6dd0SJohannes Thumshirn 3983314b6dd0SJohannes Thumshirn if (bio->bi_status) { 3984fb456252SJeff Mahoney btrfs_warn_rl_in_rcu(device->fs_info, 3985314b6dd0SJohannes Thumshirn "lost page write due to IO error on %s (%d)", 3986314b6dd0SJohannes Thumshirn rcu_str_deref(device->name), 3987314b6dd0SJohannes Thumshirn blk_status_to_errno(bio->bi_status)); 3988314b6dd0SJohannes Thumshirn ClearPageUptodate(page); 3989314b6dd0SJohannes Thumshirn SetPageError(page); 3990314b6dd0SJohannes Thumshirn btrfs_dev_stat_inc_and_print(device, 3991314b6dd0SJohannes Thumshirn BTRFS_DEV_STAT_WRITE_ERRS); 3992314b6dd0SJohannes Thumshirn } else { 3993314b6dd0SJohannes Thumshirn SetPageUptodate(page); 3994f2984462SChris Mason } 3995314b6dd0SJohannes Thumshirn 3996314b6dd0SJohannes Thumshirn put_page(page); 3997314b6dd0SJohannes Thumshirn unlock_page(page); 3998314b6dd0SJohannes Thumshirn } 3999314b6dd0SJohannes Thumshirn 4000314b6dd0SJohannes Thumshirn bio_put(bio); 4001f2984462SChris Mason } 4002f2984462SChris Mason 40038f32380dSJohannes Thumshirn struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev, 40048f32380dSJohannes Thumshirn int copy_num) 400529c36d72SAnand Jain { 400629c36d72SAnand Jain struct btrfs_super_block *super; 40078f32380dSJohannes Thumshirn struct page *page; 400812659251SNaohiro Aota u64 bytenr, bytenr_orig; 40098f32380dSJohannes Thumshirn struct address_space *mapping = bdev->bd_inode->i_mapping; 401012659251SNaohiro Aota int ret; 401129c36d72SAnand Jain 401212659251SNaohiro Aota bytenr_orig = btrfs_sb_offset(copy_num); 401312659251SNaohiro Aota ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr); 401412659251SNaohiro Aota if (ret == -ENOENT) 401512659251SNaohiro Aota return ERR_PTR(-EINVAL); 401612659251SNaohiro Aota else if (ret) 401712659251SNaohiro Aota return ERR_PTR(ret); 401812659251SNaohiro Aota 4019cda00ebaSChristoph Hellwig if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev)) 40208f32380dSJohannes Thumshirn return ERR_PTR(-EINVAL); 402129c36d72SAnand Jain 40228f32380dSJohannes Thumshirn page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS); 40238f32380dSJohannes Thumshirn if (IS_ERR(page)) 40248f32380dSJohannes Thumshirn return ERR_CAST(page); 402529c36d72SAnand Jain 40268f32380dSJohannes Thumshirn super = page_address(page); 402796c2e067SAnand Jain if (btrfs_super_magic(super) != BTRFS_MAGIC) { 402896c2e067SAnand Jain btrfs_release_disk_super(super); 402996c2e067SAnand Jain return ERR_PTR(-ENODATA); 403096c2e067SAnand Jain } 403196c2e067SAnand Jain 403212659251SNaohiro Aota if (btrfs_super_bytenr(super) != bytenr_orig) { 40338f32380dSJohannes Thumshirn btrfs_release_disk_super(super); 40348f32380dSJohannes Thumshirn return ERR_PTR(-EINVAL); 403529c36d72SAnand Jain } 403629c36d72SAnand Jain 40378f32380dSJohannes Thumshirn return super; 403829c36d72SAnand Jain } 403929c36d72SAnand Jain 404029c36d72SAnand Jain 40418f32380dSJohannes Thumshirn struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev) 4042a512bbf8SYan Zheng { 40438f32380dSJohannes Thumshirn struct btrfs_super_block *super, *latest = NULL; 4044a512bbf8SYan Zheng int i; 4045a512bbf8SYan Zheng u64 transid = 0; 4046a512bbf8SYan Zheng 4047a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 4048a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 4049a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 4050a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 4051a512bbf8SYan Zheng */ 4052a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 40538f32380dSJohannes Thumshirn super = btrfs_read_dev_one_super(bdev, i); 40548f32380dSJohannes Thumshirn if (IS_ERR(super)) 4055a512bbf8SYan Zheng continue; 4056a512bbf8SYan Zheng 4057a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 40588f32380dSJohannes Thumshirn if (latest) 40598f32380dSJohannes Thumshirn btrfs_release_disk_super(super); 40608f32380dSJohannes Thumshirn 40618f32380dSJohannes Thumshirn latest = super; 4062a512bbf8SYan Zheng transid = btrfs_super_generation(super); 4063a512bbf8SYan Zheng } 4064a512bbf8SYan Zheng } 406592fc03fbSAnand Jain 40668f32380dSJohannes Thumshirn return super; 4067a512bbf8SYan Zheng } 4068a512bbf8SYan Zheng 40694eedeb75SHisashi Hifumi /* 4070abbb3b8eSDavid Sterba * Write superblock @sb to the @device. Do not wait for completion, all the 4071314b6dd0SJohannes Thumshirn * pages we use for writing are locked. 40724eedeb75SHisashi Hifumi * 4073abbb3b8eSDavid Sterba * Write @max_mirrors copies of the superblock, where 0 means default that fit 4074abbb3b8eSDavid Sterba * the expected device size at commit time. Note that max_mirrors must be 4075abbb3b8eSDavid Sterba * same for write and wait phases. 40764eedeb75SHisashi Hifumi * 4077314b6dd0SJohannes Thumshirn * Return number of errors when page is not found or submission fails. 40784eedeb75SHisashi Hifumi */ 4079a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 4080abbb3b8eSDavid Sterba struct btrfs_super_block *sb, int max_mirrors) 4081a512bbf8SYan Zheng { 4082d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = device->fs_info; 4083314b6dd0SJohannes Thumshirn struct address_space *mapping = device->bdev->bd_inode->i_mapping; 4084d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 4085a512bbf8SYan Zheng int i; 4086a512bbf8SYan Zheng int errors = 0; 408712659251SNaohiro Aota int ret; 408812659251SNaohiro Aota u64 bytenr, bytenr_orig; 4089a512bbf8SYan Zheng 4090a512bbf8SYan Zheng if (max_mirrors == 0) 4091a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 4092a512bbf8SYan Zheng 4093d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 4094d5178578SJohannes Thumshirn 4095a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 4096314b6dd0SJohannes Thumshirn struct page *page; 4097314b6dd0SJohannes Thumshirn struct bio *bio; 4098314b6dd0SJohannes Thumshirn struct btrfs_super_block *disk_super; 4099314b6dd0SJohannes Thumshirn 410012659251SNaohiro Aota bytenr_orig = btrfs_sb_offset(i); 410112659251SNaohiro Aota ret = btrfs_sb_log_location(device, i, WRITE, &bytenr); 410212659251SNaohiro Aota if (ret == -ENOENT) { 410312659251SNaohiro Aota continue; 410412659251SNaohiro Aota } else if (ret < 0) { 410512659251SNaohiro Aota btrfs_err(device->fs_info, 410612659251SNaohiro Aota "couldn't get super block location for mirror %d", 410712659251SNaohiro Aota i); 410812659251SNaohiro Aota errors++; 410912659251SNaohiro Aota continue; 411012659251SNaohiro Aota } 4111935e5cc9SMiao Xie if (bytenr + BTRFS_SUPER_INFO_SIZE >= 4112935e5cc9SMiao Xie device->commit_total_bytes) 4113a512bbf8SYan Zheng break; 4114a512bbf8SYan Zheng 411512659251SNaohiro Aota btrfs_set_super_bytenr(sb, bytenr_orig); 4116a512bbf8SYan Zheng 4117fd08001fSEric Biggers crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE, 4118fd08001fSEric Biggers BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, 4119fd08001fSEric Biggers sb->csum); 4120a512bbf8SYan Zheng 4121314b6dd0SJohannes Thumshirn page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT, 4122314b6dd0SJohannes Thumshirn GFP_NOFS); 4123314b6dd0SJohannes Thumshirn if (!page) { 4124fb456252SJeff Mahoney btrfs_err(device->fs_info, 4125314b6dd0SJohannes Thumshirn "couldn't get super block page for bytenr %llu", 4126f14d104dSDavid Sterba bytenr); 4127634554dcSJosef Bacik errors++; 4128634554dcSJosef Bacik continue; 4129634554dcSJosef Bacik } 4130634554dcSJosef Bacik 4131314b6dd0SJohannes Thumshirn /* Bump the refcount for wait_dev_supers() */ 4132314b6dd0SJohannes Thumshirn get_page(page); 4133a512bbf8SYan Zheng 4134314b6dd0SJohannes Thumshirn disk_super = page_address(page); 4135314b6dd0SJohannes Thumshirn memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE); 4136a512bbf8SYan Zheng 4137387125fcSChris Mason /* 4138314b6dd0SJohannes Thumshirn * Directly use bios here instead of relying on the page cache 4139314b6dd0SJohannes Thumshirn * to do I/O, so we don't lose the ability to do integrity 4140314b6dd0SJohannes Thumshirn * checking. 4141387125fcSChris Mason */ 414207888c66SChristoph Hellwig bio = bio_alloc(device->bdev, 1, 414307888c66SChristoph Hellwig REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO, 414407888c66SChristoph Hellwig GFP_NOFS); 4145314b6dd0SJohannes Thumshirn bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT; 4146314b6dd0SJohannes Thumshirn bio->bi_private = device; 4147314b6dd0SJohannes Thumshirn bio->bi_end_io = btrfs_end_super_write; 4148314b6dd0SJohannes Thumshirn __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE, 4149314b6dd0SJohannes Thumshirn offset_in_page(bytenr)); 4150314b6dd0SJohannes Thumshirn 4151314b6dd0SJohannes Thumshirn /* 4152314b6dd0SJohannes Thumshirn * We FUA only the first super block. The others we allow to 4153314b6dd0SJohannes Thumshirn * go down lazy and there's a short window where the on-disk 4154314b6dd0SJohannes Thumshirn * copies might still contain the older version. 4155314b6dd0SJohannes Thumshirn */ 41561b9e619cSOmar Sandoval if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) 4157314b6dd0SJohannes Thumshirn bio->bi_opf |= REQ_FUA; 4158314b6dd0SJohannes Thumshirn 4159314b6dd0SJohannes Thumshirn btrfsic_submit_bio(bio); 41608376d9e1SNaohiro Aota 41618376d9e1SNaohiro Aota if (btrfs_advance_sb_log(device, i)) 41628376d9e1SNaohiro Aota errors++; 4163a512bbf8SYan Zheng } 4164a512bbf8SYan Zheng return errors < i ? 0 : -1; 4165a512bbf8SYan Zheng } 4166a512bbf8SYan Zheng 4167387125fcSChris Mason /* 4168abbb3b8eSDavid Sterba * Wait for write completion of superblocks done by write_dev_supers, 4169abbb3b8eSDavid Sterba * @max_mirrors same for write and wait phases. 4170abbb3b8eSDavid Sterba * 4171314b6dd0SJohannes Thumshirn * Return number of errors when page is not found or not marked up to 4172abbb3b8eSDavid Sterba * date. 4173abbb3b8eSDavid Sterba */ 4174abbb3b8eSDavid Sterba static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) 4175abbb3b8eSDavid Sterba { 4176abbb3b8eSDavid Sterba int i; 4177abbb3b8eSDavid Sterba int errors = 0; 4178b6a535faSHoward McLauchlan bool primary_failed = false; 417912659251SNaohiro Aota int ret; 4180abbb3b8eSDavid Sterba u64 bytenr; 4181abbb3b8eSDavid Sterba 4182abbb3b8eSDavid Sterba if (max_mirrors == 0) 4183abbb3b8eSDavid Sterba max_mirrors = BTRFS_SUPER_MIRROR_MAX; 4184abbb3b8eSDavid Sterba 4185abbb3b8eSDavid Sterba for (i = 0; i < max_mirrors; i++) { 4186314b6dd0SJohannes Thumshirn struct page *page; 4187314b6dd0SJohannes Thumshirn 418812659251SNaohiro Aota ret = btrfs_sb_log_location(device, i, READ, &bytenr); 418912659251SNaohiro Aota if (ret == -ENOENT) { 419012659251SNaohiro Aota break; 419112659251SNaohiro Aota } else if (ret < 0) { 419212659251SNaohiro Aota errors++; 419312659251SNaohiro Aota if (i == 0) 419412659251SNaohiro Aota primary_failed = true; 419512659251SNaohiro Aota continue; 419612659251SNaohiro Aota } 4197abbb3b8eSDavid Sterba if (bytenr + BTRFS_SUPER_INFO_SIZE >= 4198abbb3b8eSDavid Sterba device->commit_total_bytes) 4199abbb3b8eSDavid Sterba break; 4200abbb3b8eSDavid Sterba 4201314b6dd0SJohannes Thumshirn page = find_get_page(device->bdev->bd_inode->i_mapping, 4202314b6dd0SJohannes Thumshirn bytenr >> PAGE_SHIFT); 4203314b6dd0SJohannes Thumshirn if (!page) { 4204abbb3b8eSDavid Sterba errors++; 4205b6a535faSHoward McLauchlan if (i == 0) 4206b6a535faSHoward McLauchlan primary_failed = true; 4207abbb3b8eSDavid Sterba continue; 4208abbb3b8eSDavid Sterba } 4209314b6dd0SJohannes Thumshirn /* Page is submitted locked and unlocked once the IO completes */ 4210314b6dd0SJohannes Thumshirn wait_on_page_locked(page); 4211314b6dd0SJohannes Thumshirn if (PageError(page)) { 4212abbb3b8eSDavid Sterba errors++; 4213b6a535faSHoward McLauchlan if (i == 0) 4214b6a535faSHoward McLauchlan primary_failed = true; 4215b6a535faSHoward McLauchlan } 4216abbb3b8eSDavid Sterba 4217314b6dd0SJohannes Thumshirn /* Drop our reference */ 4218314b6dd0SJohannes Thumshirn put_page(page); 4219abbb3b8eSDavid Sterba 4220314b6dd0SJohannes Thumshirn /* Drop the reference from the writing run */ 4221314b6dd0SJohannes Thumshirn put_page(page); 4222abbb3b8eSDavid Sterba } 4223abbb3b8eSDavid Sterba 4224b6a535faSHoward McLauchlan /* log error, force error return */ 4225b6a535faSHoward McLauchlan if (primary_failed) { 4226b6a535faSHoward McLauchlan btrfs_err(device->fs_info, "error writing primary super block to device %llu", 4227b6a535faSHoward McLauchlan device->devid); 4228b6a535faSHoward McLauchlan return -1; 4229b6a535faSHoward McLauchlan } 4230b6a535faSHoward McLauchlan 4231abbb3b8eSDavid Sterba return errors < i ? 0 : -1; 4232abbb3b8eSDavid Sterba } 4233abbb3b8eSDavid Sterba 4234abbb3b8eSDavid Sterba /* 4235387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 4236387125fcSChris Mason * for the barrier when it is done 4237387125fcSChris Mason */ 42384246a0b6SChristoph Hellwig static void btrfs_end_empty_barrier(struct bio *bio) 4239387125fcSChris Mason { 4240387125fcSChris Mason complete(bio->bi_private); 4241387125fcSChris Mason } 4242387125fcSChris Mason 4243387125fcSChris Mason /* 42444fc6441aSAnand Jain * Submit a flush request to the device if it supports it. Error handling is 42454fc6441aSAnand Jain * done in the waiting counterpart. 4246387125fcSChris Mason */ 42474fc6441aSAnand Jain static void write_dev_flush(struct btrfs_device *device) 4248387125fcSChris Mason { 4249e0ae9994SDavid Sterba struct bio *bio = device->flush_bio; 4250387125fcSChris Mason 4251a91cf0ffSWang Yugui #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY 4252a91cf0ffSWang Yugui /* 4253a91cf0ffSWang Yugui * When a disk has write caching disabled, we skip submission of a bio 4254a91cf0ffSWang Yugui * with flush and sync requests before writing the superblock, since 4255a91cf0ffSWang Yugui * it's not needed. However when the integrity checker is enabled, this 4256a91cf0ffSWang Yugui * results in reports that there are metadata blocks referred by a 4257a91cf0ffSWang Yugui * superblock that were not properly flushed. So don't skip the bio 4258a91cf0ffSWang Yugui * submission only when the integrity checker is enabled for the sake 4259a91cf0ffSWang Yugui * of simplicity, since this is a debug tool and not meant for use in 4260a91cf0ffSWang Yugui * non-debug builds. 4261a91cf0ffSWang Yugui */ 4262a91cf0ffSWang Yugui struct request_queue *q = bdev_get_queue(device->bdev); 4263c2a9c7abSAnand Jain if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) 42644fc6441aSAnand Jain return; 4265a91cf0ffSWang Yugui #endif 4266387125fcSChris Mason 4267a7c50c94SChristoph Hellwig bio_reset(bio, device->bdev, REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH); 4268387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 4269387125fcSChris Mason init_completion(&device->flush_wait); 4270387125fcSChris Mason bio->bi_private = &device->flush_wait; 4271387125fcSChris Mason 427243a01111SLu Fengqi btrfsic_submit_bio(bio); 42731c3063b6SAnand Jain set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 42744fc6441aSAnand Jain } 4275387125fcSChris Mason 42764fc6441aSAnand Jain /* 42774fc6441aSAnand Jain * If the flush bio has been submitted by write_dev_flush, wait for it. 42784fc6441aSAnand Jain */ 42798c27cb35SLinus Torvalds static blk_status_t wait_dev_flush(struct btrfs_device *device) 42804fc6441aSAnand Jain { 42814fc6441aSAnand Jain struct bio *bio = device->flush_bio; 42824fc6441aSAnand Jain 42831c3063b6SAnand Jain if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) 428458efbc9fSOmar Sandoval return BLK_STS_OK; 42854fc6441aSAnand Jain 42861c3063b6SAnand Jain clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 42872980d574SDavid Sterba wait_for_completion_io(&device->flush_wait); 42884fc6441aSAnand Jain 42898c27cb35SLinus Torvalds return bio->bi_status; 4290387125fcSChris Mason } 4291387125fcSChris Mason 4292d10b82feSQu Wenruo static int check_barrier_error(struct btrfs_fs_info *fs_info) 4293401b41e5SAnand Jain { 42946528b99dSAnand Jain if (!btrfs_check_rw_degradable(fs_info, NULL)) 4295401b41e5SAnand Jain return -EIO; 4296387125fcSChris Mason return 0; 4297387125fcSChris Mason } 4298387125fcSChris Mason 4299387125fcSChris Mason /* 4300387125fcSChris Mason * send an empty flush down to each device in parallel, 4301387125fcSChris Mason * then wait for them 4302387125fcSChris Mason */ 4303387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 4304387125fcSChris Mason { 4305387125fcSChris Mason struct list_head *head; 4306387125fcSChris Mason struct btrfs_device *dev; 43075af3e8ccSStefan Behrens int errors_wait = 0; 43084e4cbee9SChristoph Hellwig blk_status_t ret; 4309387125fcSChris Mason 43101538e6c5SDavid Sterba lockdep_assert_held(&info->fs_devices->device_list_mutex); 4311387125fcSChris Mason /* send down all the barriers */ 4312387125fcSChris Mason head = &info->fs_devices->devices; 43131538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4314e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 4315f88ba6a2SHidetoshi Seto continue; 4316cea7c8bfSAnand Jain if (!dev->bdev) 4317387125fcSChris Mason continue; 4318e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4319ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4320387125fcSChris Mason continue; 4321387125fcSChris Mason 43224fc6441aSAnand Jain write_dev_flush(dev); 432358efbc9fSOmar Sandoval dev->last_flush_error = BLK_STS_OK; 4324387125fcSChris Mason } 4325387125fcSChris Mason 4326387125fcSChris Mason /* wait for all the barriers */ 43271538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4328e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 4329f88ba6a2SHidetoshi Seto continue; 4330387125fcSChris Mason if (!dev->bdev) { 43315af3e8ccSStefan Behrens errors_wait++; 4332387125fcSChris Mason continue; 4333387125fcSChris Mason } 4334e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4335ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4336387125fcSChris Mason continue; 4337387125fcSChris Mason 43384fc6441aSAnand Jain ret = wait_dev_flush(dev); 4339401b41e5SAnand Jain if (ret) { 4340401b41e5SAnand Jain dev->last_flush_error = ret; 434166b4993eSDavid Sterba btrfs_dev_stat_inc_and_print(dev, 434266b4993eSDavid Sterba BTRFS_DEV_STAT_FLUSH_ERRS); 43435af3e8ccSStefan Behrens errors_wait++; 4344387125fcSChris Mason } 4345401b41e5SAnand Jain } 4346401b41e5SAnand Jain 4347cea7c8bfSAnand Jain if (errors_wait) { 4348401b41e5SAnand Jain /* 4349401b41e5SAnand Jain * At some point we need the status of all disks 4350401b41e5SAnand Jain * to arrive at the volume status. So error checking 4351401b41e5SAnand Jain * is being pushed to a separate loop. 4352401b41e5SAnand Jain */ 4353d10b82feSQu Wenruo return check_barrier_error(info); 4354401b41e5SAnand Jain } 4355387125fcSChris Mason return 0; 4356387125fcSChris Mason } 4357387125fcSChris Mason 4358943c6e99SZhao Lei int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 4359943c6e99SZhao Lei { 43608789f4feSZhao Lei int raid_type; 43618789f4feSZhao Lei int min_tolerated = INT_MAX; 4362943c6e99SZhao Lei 43638789f4feSZhao Lei if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 43648789f4feSZhao Lei (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 43658c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 43668789f4feSZhao Lei btrfs_raid_array[BTRFS_RAID_SINGLE]. 43678789f4feSZhao Lei tolerated_failures); 4368943c6e99SZhao Lei 43698789f4feSZhao Lei for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 43708789f4feSZhao Lei if (raid_type == BTRFS_RAID_SINGLE) 43718789f4feSZhao Lei continue; 437241a6e891SAnand Jain if (!(flags & btrfs_raid_array[raid_type].bg_flag)) 43738789f4feSZhao Lei continue; 43748c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 43758789f4feSZhao Lei btrfs_raid_array[raid_type]. 43768789f4feSZhao Lei tolerated_failures); 43778789f4feSZhao Lei } 4378943c6e99SZhao Lei 43798789f4feSZhao Lei if (min_tolerated == INT_MAX) { 4380ab8d0fc4SJeff Mahoney pr_warn("BTRFS: unknown raid flag: %llu", flags); 43818789f4feSZhao Lei min_tolerated = 0; 43828789f4feSZhao Lei } 43838789f4feSZhao Lei 43848789f4feSZhao Lei return min_tolerated; 4385943c6e99SZhao Lei } 4386943c6e99SZhao Lei 4387eece6a9cSDavid Sterba int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) 4388f2984462SChris Mason { 4389e5e9a520SChris Mason struct list_head *head; 4390f2984462SChris Mason struct btrfs_device *dev; 4391a061fc8dSChris Mason struct btrfs_super_block *sb; 4392f2984462SChris Mason struct btrfs_dev_item *dev_item; 4393f2984462SChris Mason int ret; 4394f2984462SChris Mason int do_barriers; 4395a236aed1SChris Mason int max_errors; 4396a236aed1SChris Mason int total_errors = 0; 4397a061fc8dSChris Mason u64 flags; 4398f2984462SChris Mason 43990b246afaSJeff Mahoney do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); 4400fed3b381SLiu Bo 4401fed3b381SLiu Bo /* 4402fed3b381SLiu Bo * max_mirrors == 0 indicates we're from commit_transaction, 4403fed3b381SLiu Bo * not from fsync where the tree roots in fs_info have not 4404fed3b381SLiu Bo * been consistent on disk. 4405fed3b381SLiu Bo */ 4406fed3b381SLiu Bo if (max_mirrors == 0) 44070b246afaSJeff Mahoney backup_super_roots(fs_info); 4408f2984462SChris Mason 44090b246afaSJeff Mahoney sb = fs_info->super_for_commit; 4410a061fc8dSChris Mason dev_item = &sb->dev_item; 4411e5e9a520SChris Mason 44120b246afaSJeff Mahoney mutex_lock(&fs_info->fs_devices->device_list_mutex); 44130b246afaSJeff Mahoney head = &fs_info->fs_devices->devices; 44140b246afaSJeff Mahoney max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; 4415387125fcSChris Mason 44165af3e8ccSStefan Behrens if (do_barriers) { 44170b246afaSJeff Mahoney ret = barrier_all_devices(fs_info); 44185af3e8ccSStefan Behrens if (ret) { 44195af3e8ccSStefan Behrens mutex_unlock( 44200b246afaSJeff Mahoney &fs_info->fs_devices->device_list_mutex); 44210b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 44225af3e8ccSStefan Behrens "errors while submitting device barriers."); 44235af3e8ccSStefan Behrens return ret; 44245af3e8ccSStefan Behrens } 44255af3e8ccSStefan Behrens } 4426387125fcSChris Mason 44271538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4428dfe25020SChris Mason if (!dev->bdev) { 4429dfe25020SChris Mason total_errors++; 4430dfe25020SChris Mason continue; 4431dfe25020SChris Mason } 4432e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4433ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4434dfe25020SChris Mason continue; 4435dfe25020SChris Mason 44362b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 4437a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 4438a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 44397df69d3eSMiao Xie btrfs_set_stack_device_total_bytes(dev_item, 4440935e5cc9SMiao Xie dev->commit_total_bytes); 4441ce7213c7SMiao Xie btrfs_set_stack_device_bytes_used(dev_item, 4442ce7213c7SMiao Xie dev->commit_bytes_used); 4443a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 4444a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 4445a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 4446a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 44477239ff4bSNikolay Borisov memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, 44487239ff4bSNikolay Borisov BTRFS_FSID_SIZE); 4449a512bbf8SYan Zheng 4450a061fc8dSChris Mason flags = btrfs_super_flags(sb); 4451a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 4452f2984462SChris Mason 445375cb857dSQu Wenruo ret = btrfs_validate_write_super(fs_info, sb); 445475cb857dSQu Wenruo if (ret < 0) { 445575cb857dSQu Wenruo mutex_unlock(&fs_info->fs_devices->device_list_mutex); 445675cb857dSQu Wenruo btrfs_handle_fs_error(fs_info, -EUCLEAN, 445775cb857dSQu Wenruo "unexpected superblock corruption detected"); 445875cb857dSQu Wenruo return -EUCLEAN; 445975cb857dSQu Wenruo } 446075cb857dSQu Wenruo 4461abbb3b8eSDavid Sterba ret = write_dev_supers(dev, sb, max_mirrors); 4462a236aed1SChris Mason if (ret) 4463a236aed1SChris Mason total_errors++; 4464f2984462SChris Mason } 4465a236aed1SChris Mason if (total_errors > max_errors) { 44660b246afaSJeff Mahoney btrfs_err(fs_info, "%d errors while writing supers", 4467d397712bSChris Mason total_errors); 44680b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 446979787eaaSJeff Mahoney 44709d565ba4SStefan Behrens /* FUA is masked off if unsupported and can't be the reason */ 44710b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 44720b246afaSJeff Mahoney "%d errors while writing supers", 44730b246afaSJeff Mahoney total_errors); 44749d565ba4SStefan Behrens return -EIO; 4475a236aed1SChris Mason } 4476f2984462SChris Mason 4477a512bbf8SYan Zheng total_errors = 0; 44781538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4479dfe25020SChris Mason if (!dev->bdev) 4480dfe25020SChris Mason continue; 4481e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4482ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4483dfe25020SChris Mason continue; 4484dfe25020SChris Mason 4485abbb3b8eSDavid Sterba ret = wait_dev_supers(dev, max_mirrors); 4486a512bbf8SYan Zheng if (ret) 44871259ab75SChris Mason total_errors++; 4488f2984462SChris Mason } 44890b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 4490a236aed1SChris Mason if (total_errors > max_errors) { 44910b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 44920b246afaSJeff Mahoney "%d errors while writing supers", 44930b246afaSJeff Mahoney total_errors); 449479787eaaSJeff Mahoney return -EIO; 4495a236aed1SChris Mason } 4496f2984462SChris Mason return 0; 4497f2984462SChris Mason } 4498f2984462SChris Mason 4499cb517eabSMiao Xie /* Drop a fs root from the radix tree and free it. */ 4500cb517eabSMiao Xie void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 4501cb517eabSMiao Xie struct btrfs_root *root) 45022619ba1fSChris Mason { 45034785e24fSJosef Bacik bool drop_ref = false; 45044785e24fSJosef Bacik 45054df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 45062619ba1fSChris Mason radix_tree_delete(&fs_info->fs_roots_radix, 45072619ba1fSChris Mason (unsigned long)root->root_key.objectid); 4508af01d2e5SJosef Bacik if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state)) 45094785e24fSJosef Bacik drop_ref = true; 45104df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 451176dda93cSYan, Zheng 451284961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 4513ef67963dSJosef Bacik ASSERT(root->log_root == NULL); 45141c1ea4f7SLiu Bo if (root->reloc_root) { 451500246528SJosef Bacik btrfs_put_root(root->reloc_root); 45161c1ea4f7SLiu Bo root->reloc_root = NULL; 45171c1ea4f7SLiu Bo } 45181c1ea4f7SLiu Bo } 45193321719eSLiu Bo 45204785e24fSJosef Bacik if (drop_ref) 452100246528SJosef Bacik btrfs_put_root(root); 45222619ba1fSChris Mason } 45232619ba1fSChris Mason 4524c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 4525c146afadSYan Zheng { 4526c146afadSYan Zheng u64 root_objectid = 0; 4527c146afadSYan Zheng struct btrfs_root *gang[8]; 452865d33fd7SQu Wenruo int i = 0; 452965d33fd7SQu Wenruo int err = 0; 453065d33fd7SQu Wenruo unsigned int ret = 0; 4531c146afadSYan Zheng 4532c146afadSYan Zheng while (1) { 4533efc34534SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 4534c146afadSYan Zheng ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4535c146afadSYan Zheng (void **)gang, root_objectid, 4536c146afadSYan Zheng ARRAY_SIZE(gang)); 453765d33fd7SQu Wenruo if (!ret) { 4538efc34534SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 4539c146afadSYan Zheng break; 454065d33fd7SQu Wenruo } 45415d4f98a2SYan Zheng root_objectid = gang[ret - 1]->root_key.objectid + 1; 454266b4ffd1SJosef Bacik 454365d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 454465d33fd7SQu Wenruo /* Avoid to grab roots in dead_roots */ 454565d33fd7SQu Wenruo if (btrfs_root_refs(&gang[i]->root_item) == 0) { 454665d33fd7SQu Wenruo gang[i] = NULL; 454765d33fd7SQu Wenruo continue; 454865d33fd7SQu Wenruo } 454965d33fd7SQu Wenruo /* grab all the search result for later use */ 455000246528SJosef Bacik gang[i] = btrfs_grab_root(gang[i]); 455165d33fd7SQu Wenruo } 4552efc34534SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 455365d33fd7SQu Wenruo 455465d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 455565d33fd7SQu Wenruo if (!gang[i]) 455665d33fd7SQu Wenruo continue; 4557c146afadSYan Zheng root_objectid = gang[i]->root_key.objectid; 455866b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(gang[i]); 455966b4ffd1SJosef Bacik if (err) 456065d33fd7SQu Wenruo break; 456100246528SJosef Bacik btrfs_put_root(gang[i]); 4562c146afadSYan Zheng } 4563c146afadSYan Zheng root_objectid++; 4564c146afadSYan Zheng } 456565d33fd7SQu Wenruo 456665d33fd7SQu Wenruo /* release the uncleaned roots due to error */ 456765d33fd7SQu Wenruo for (; i < ret; i++) { 456865d33fd7SQu Wenruo if (gang[i]) 456900246528SJosef Bacik btrfs_put_root(gang[i]); 457065d33fd7SQu Wenruo } 457165d33fd7SQu Wenruo return err; 4572c146afadSYan Zheng } 4573c146afadSYan Zheng 45746bccf3abSJeff Mahoney int btrfs_commit_super(struct btrfs_fs_info *fs_info) 4575c146afadSYan Zheng { 45766bccf3abSJeff Mahoney struct btrfs_root *root = fs_info->tree_root; 4577c146afadSYan Zheng struct btrfs_trans_handle *trans; 4578c146afadSYan Zheng 45790b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 45802ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 45810b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 45820b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 4583c71bf099SYan, Zheng 4584c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 45850b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 45860b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4587c71bf099SYan, Zheng 45887a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 45893612b495STsutomu Itoh if (IS_ERR(trans)) 45903612b495STsutomu Itoh return PTR_ERR(trans); 45913a45bb20SJeff Mahoney return btrfs_commit_transaction(trans); 4592c146afadSYan Zheng } 4593c146afadSYan Zheng 459436c86a9eSQu Wenruo static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info) 459536c86a9eSQu Wenruo { 459636c86a9eSQu Wenruo struct btrfs_transaction *trans; 459736c86a9eSQu Wenruo struct btrfs_transaction *tmp; 459836c86a9eSQu Wenruo bool found = false; 459936c86a9eSQu Wenruo 460036c86a9eSQu Wenruo if (list_empty(&fs_info->trans_list)) 460136c86a9eSQu Wenruo return; 460236c86a9eSQu Wenruo 460336c86a9eSQu Wenruo /* 460436c86a9eSQu Wenruo * This function is only called at the very end of close_ctree(), 460536c86a9eSQu Wenruo * thus no other running transaction, no need to take trans_lock. 460636c86a9eSQu Wenruo */ 460736c86a9eSQu Wenruo ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags)); 460836c86a9eSQu Wenruo list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) { 460936c86a9eSQu Wenruo struct extent_state *cached = NULL; 461036c86a9eSQu Wenruo u64 dirty_bytes = 0; 461136c86a9eSQu Wenruo u64 cur = 0; 461236c86a9eSQu Wenruo u64 found_start; 461336c86a9eSQu Wenruo u64 found_end; 461436c86a9eSQu Wenruo 461536c86a9eSQu Wenruo found = true; 461636c86a9eSQu Wenruo while (!find_first_extent_bit(&trans->dirty_pages, cur, 461736c86a9eSQu Wenruo &found_start, &found_end, EXTENT_DIRTY, &cached)) { 461836c86a9eSQu Wenruo dirty_bytes += found_end + 1 - found_start; 461936c86a9eSQu Wenruo cur = found_end + 1; 462036c86a9eSQu Wenruo } 462136c86a9eSQu Wenruo btrfs_warn(fs_info, 462236c86a9eSQu Wenruo "transaction %llu (with %llu dirty metadata bytes) is not committed", 462336c86a9eSQu Wenruo trans->transid, dirty_bytes); 462436c86a9eSQu Wenruo btrfs_cleanup_one_transaction(trans, fs_info); 462536c86a9eSQu Wenruo 462636c86a9eSQu Wenruo if (trans == fs_info->running_transaction) 462736c86a9eSQu Wenruo fs_info->running_transaction = NULL; 462836c86a9eSQu Wenruo list_del_init(&trans->list); 462936c86a9eSQu Wenruo 463036c86a9eSQu Wenruo btrfs_put_transaction(trans); 463136c86a9eSQu Wenruo trace_btrfs_transaction_commit(fs_info); 463236c86a9eSQu Wenruo } 463336c86a9eSQu Wenruo ASSERT(!found); 463436c86a9eSQu Wenruo } 463536c86a9eSQu Wenruo 4636b105e927SDavid Sterba void __cold close_ctree(struct btrfs_fs_info *fs_info) 4637eb60ceacSChris Mason { 4638c146afadSYan Zheng int ret; 4639e089f05cSChris Mason 4640afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); 4641d6fd0ae2SOmar Sandoval /* 4642d6fd0ae2SOmar Sandoval * We don't want the cleaner to start new transactions, add more delayed 4643d6fd0ae2SOmar Sandoval * iputs, etc. while we're closing. We can't use kthread_stop() yet 4644d6fd0ae2SOmar Sandoval * because that frees the task_struct, and the transaction kthread might 4645d6fd0ae2SOmar Sandoval * still try to wake up the cleaner. 4646d6fd0ae2SOmar Sandoval */ 4647d6fd0ae2SOmar Sandoval kthread_park(fs_info->cleaner_kthread); 4648a2135011SChris Mason 4649b4be6aefSJosef Bacik /* 4650b4be6aefSJosef Bacik * If we had UNFINISHED_DROPS we could still be processing them, so 4651b4be6aefSJosef Bacik * clear that bit and wake up relocation so it can stop. 4652b4be6aefSJosef Bacik */ 4653b4be6aefSJosef Bacik btrfs_wake_unfinished_drop(fs_info); 4654b4be6aefSJosef Bacik 46557343dd61SJustin Maggard /* wait for the qgroup rescan worker to stop */ 4656d06f23d6SJeff Mahoney btrfs_qgroup_wait_for_completion(fs_info, false); 46577343dd61SJustin Maggard 4658803b2f54SStefan Behrens /* wait for the uuid_scan task to finish */ 4659803b2f54SStefan Behrens down(&fs_info->uuid_tree_rescan_sem); 4660803b2f54SStefan Behrens /* avoid complains from lockdep et al., set sem back to initial state */ 4661803b2f54SStefan Behrens up(&fs_info->uuid_tree_rescan_sem); 4662803b2f54SStefan Behrens 4663837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 4664aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 4665837d5b6eSIlya Dryomov 46668dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 46678dabb742SStefan Behrens 4668aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 46694cb5300bSChris Mason 46704cb5300bSChris Mason /* wait for any defraggers to finish */ 46714cb5300bSChris Mason wait_event(fs_info->transaction_wait, 46724cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 46734cb5300bSChris Mason 46744cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 467526176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 46764cb5300bSChris Mason 467721c7e756SMiao Xie cancel_work_sync(&fs_info->async_reclaim_work); 467857056740SJosef Bacik cancel_work_sync(&fs_info->async_data_reclaim_work); 4679576fa348SJosef Bacik cancel_work_sync(&fs_info->preempt_reclaim_work); 468021c7e756SMiao Xie 468118bb8bbfSJohannes Thumshirn cancel_work_sync(&fs_info->reclaim_bgs_work); 468218bb8bbfSJohannes Thumshirn 4683b0643e59SDennis Zhou /* Cancel or finish ongoing discard work */ 4684b0643e59SDennis Zhou btrfs_discard_cleanup(fs_info); 4685b0643e59SDennis Zhou 4686bc98a42cSDavid Howells if (!sb_rdonly(fs_info->sb)) { 4687e44163e1SJeff Mahoney /* 4688d6fd0ae2SOmar Sandoval * The cleaner kthread is stopped, so do one final pass over 4689d6fd0ae2SOmar Sandoval * unused block groups. 4690e44163e1SJeff Mahoney */ 46910b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 4692e44163e1SJeff Mahoney 4693f0cc2cd7SFilipe Manana /* 4694f0cc2cd7SFilipe Manana * There might be existing delayed inode workers still running 4695f0cc2cd7SFilipe Manana * and holding an empty delayed inode item. We must wait for 4696f0cc2cd7SFilipe Manana * them to complete first because they can create a transaction. 4697f0cc2cd7SFilipe Manana * This happens when someone calls btrfs_balance_delayed_items() 4698f0cc2cd7SFilipe Manana * and then a transaction commit runs the same delayed nodes 4699f0cc2cd7SFilipe Manana * before any delayed worker has done something with the nodes. 4700f0cc2cd7SFilipe Manana * We must wait for any worker here and not at transaction 4701f0cc2cd7SFilipe Manana * commit time since that could cause a deadlock. 4702f0cc2cd7SFilipe Manana * This is a very rare case. 4703f0cc2cd7SFilipe Manana */ 4704f0cc2cd7SFilipe Manana btrfs_flush_workqueue(fs_info->delayed_workers); 4705f0cc2cd7SFilipe Manana 47066bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 4707d397712bSChris Mason if (ret) 470804892340SEric Sandeen btrfs_err(fs_info, "commit super ret %d", ret); 4709c146afadSYan Zheng } 4710ed2ff2cbSChris Mason 471184961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) 47122ff7e61eSJeff Mahoney btrfs_error_commit_super(fs_info); 4713acce952bSliubo 4714e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 4715e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 47168929ecfaSYan, Zheng 4717e187831eSJosef Bacik ASSERT(list_empty(&fs_info->delayed_iputs)); 4718afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); 4719f25784b3SYan Zheng 47205958253cSQu Wenruo if (btrfs_check_quota_leak(fs_info)) { 47215958253cSQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 47225958253cSQu Wenruo btrfs_err(fs_info, "qgroup reserved space leaked"); 47235958253cSQu Wenruo } 47245958253cSQu Wenruo 472504892340SEric Sandeen btrfs_free_qgroup_config(fs_info); 4726fe816d0fSNikolay Borisov ASSERT(list_empty(&fs_info->delalloc_roots)); 4727bcef60f2SArne Jansen 4728963d678bSMiao Xie if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 472904892340SEric Sandeen btrfs_info(fs_info, "at unmount delalloc count %lld", 4730963d678bSMiao Xie percpu_counter_sum(&fs_info->delalloc_bytes)); 4731b0c68f8bSChris Mason } 473231153d81SYan Zheng 47335deb17e1SJosef Bacik if (percpu_counter_sum(&fs_info->ordered_bytes)) 47344297ff84SJosef Bacik btrfs_info(fs_info, "at unmount dio bytes count %lld", 47355deb17e1SJosef Bacik percpu_counter_sum(&fs_info->ordered_bytes)); 47364297ff84SJosef Bacik 47376618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 4738b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 47395ac1d209SJeff Mahoney 47401a4319ccSLiu Bo btrfs_put_block_group_cache(fs_info); 47411a4319ccSLiu Bo 4742de348ee0SWang Shilong /* 4743de348ee0SWang Shilong * we must make sure there is not any read request to 4744de348ee0SWang Shilong * submit after we stopping all workers. 4745de348ee0SWang Shilong */ 4746de348ee0SWang Shilong invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 474796192499SJosef Bacik btrfs_stop_all_workers(fs_info); 474896192499SJosef Bacik 47490a31daa4SFilipe Manana /* We shouldn't have any transaction open at this point */ 475036c86a9eSQu Wenruo warn_about_uncommitted_trans(fs_info); 47510a31daa4SFilipe Manana 4752afcdd129SJosef Bacik clear_bit(BTRFS_FS_OPEN, &fs_info->flags); 47534273eaffSAnand Jain free_root_pointers(fs_info, true); 47548c38938cSJosef Bacik btrfs_free_fs_roots(fs_info); 47559ad6b7bcSChris Mason 47564e19443dSJosef Bacik /* 47574e19443dSJosef Bacik * We must free the block groups after dropping the fs_roots as we could 47584e19443dSJosef Bacik * have had an IO error and have left over tree log blocks that aren't 47594e19443dSJosef Bacik * cleaned up until the fs roots are freed. This makes the block group 47604e19443dSJosef Bacik * accounting appear to be wrong because there's pending reserved bytes, 47614e19443dSJosef Bacik * so make sure we do the block group cleanup afterwards. 47624e19443dSJosef Bacik */ 47634e19443dSJosef Bacik btrfs_free_block_groups(fs_info); 47644e19443dSJosef Bacik 476513e6c37bSJosef Bacik iput(fs_info->btree_inode); 4766d6bfde87SChris Mason 476721adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 47680b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) 47692ff7e61eSJeff Mahoney btrfsic_unmount(fs_info->fs_devices); 477021adbd5cSStefan Behrens #endif 477121adbd5cSStefan Behrens 47720b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 477368c94e55SNikolay Borisov btrfs_close_devices(fs_info->fs_devices); 4774eb60ceacSChris Mason } 4775eb60ceacSChris Mason 4776b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 4777b9fab919SChris Mason int atomic) 4778ccd467d6SChris Mason { 47791259ab75SChris Mason int ret; 4780727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 47811259ab75SChris Mason 47820b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 47831259ab75SChris Mason if (!ret) 47841259ab75SChris Mason return ret; 47851259ab75SChris Mason 47861259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 4787b9fab919SChris Mason parent_transid, atomic); 4788b9fab919SChris Mason if (ret == -EAGAIN) 4789b9fab919SChris Mason return ret; 47901259ab75SChris Mason return !ret; 47915f39d397SChris Mason } 47926702ed49SChris Mason 47935f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 47945f39d397SChris Mason { 47952f4d60dfSQu Wenruo struct btrfs_fs_info *fs_info = buf->fs_info; 47965f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 4797b9473439SChris Mason int was_dirty; 4798b4ce94deSChris Mason 479906ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 480006ea65a3SJosef Bacik /* 480106ea65a3SJosef Bacik * This is a fast path so only do this check if we have sanity tests 480252042d8eSAndrea Gelmini * enabled. Normal people shouldn't be using unmapped buffers as dirty 480306ea65a3SJosef Bacik * outside of the sanity tests. 480406ea65a3SJosef Bacik */ 4805b0132a3bSNikolay Borisov if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) 480606ea65a3SJosef Bacik return; 480706ea65a3SJosef Bacik #endif 480849d0c642SFilipe Manana btrfs_assert_tree_write_locked(buf); 48090b246afaSJeff Mahoney if (transid != fs_info->generation) 48105d163e0eSJeff Mahoney WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", 48110b246afaSJeff Mahoney buf->start, transid, fs_info->generation); 48120b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 4813e2d84521SMiao Xie if (!was_dirty) 4814104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 4815e2d84521SMiao Xie buf->len, 48160b246afaSJeff Mahoney fs_info->dirty_metadata_batch); 48171f21ef0aSFilipe Manana #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 481869fc6cbbSQu Wenruo /* 481969fc6cbbSQu Wenruo * Since btrfs_mark_buffer_dirty() can be called with item pointer set 482069fc6cbbSQu Wenruo * but item data not updated. 482169fc6cbbSQu Wenruo * So here we should only check item pointers, not item data. 482269fc6cbbSQu Wenruo */ 482369fc6cbbSQu Wenruo if (btrfs_header_level(buf) == 0 && 4824cfdaad5eSDavid Sterba btrfs_check_leaf_relaxed(buf)) { 4825a4f78750SDavid Sterba btrfs_print_leaf(buf); 48261f21ef0aSFilipe Manana ASSERT(0); 48271f21ef0aSFilipe Manana } 48281f21ef0aSFilipe Manana #endif 4829eb60ceacSChris Mason } 4830eb60ceacSChris Mason 48312ff7e61eSJeff Mahoney static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, 4832b53d3f5dSLiu Bo int flush_delayed) 483335b7e476SChris Mason { 4834188de649SChris Mason /* 4835188de649SChris Mason * looks as though older kernels can get into trouble with 4836188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 4837188de649SChris Mason */ 4838e2d84521SMiao Xie int ret; 4839d6bfde87SChris Mason 48406933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 4841d6bfde87SChris Mason return; 4842d6bfde87SChris Mason 4843b53d3f5dSLiu Bo if (flush_delayed) 48442ff7e61eSJeff Mahoney btrfs_balance_delayed_items(fs_info); 484516cdcec7SMiao Xie 4846d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 4847d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 4848d814a491SEthan Lien fs_info->dirty_metadata_batch); 4849e2d84521SMiao Xie if (ret > 0) { 48500b246afaSJeff Mahoney balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); 485116cdcec7SMiao Xie } 485216cdcec7SMiao Xie } 485316cdcec7SMiao Xie 48542ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) 485516cdcec7SMiao Xie { 48562ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 1); 485735b7e476SChris Mason } 4858b53d3f5dSLiu Bo 48592ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) 4860b53d3f5dSLiu Bo { 48612ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 0); 4862d6bfde87SChris Mason } 48636b80053dSChris Mason 48642ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) 4865acce952bSliubo { 4866fe816d0fSNikolay Borisov /* cleanup FS via transaction */ 4867fe816d0fSNikolay Borisov btrfs_cleanup_transaction(fs_info); 4868fe816d0fSNikolay Borisov 48690b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 48702ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 48710b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 4872acce952bSliubo 48730b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 48740b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4875acce952bSliubo } 4876acce952bSliubo 4877ef67963dSJosef Bacik static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info) 4878ef67963dSJosef Bacik { 4879ef67963dSJosef Bacik struct btrfs_root *gang[8]; 4880ef67963dSJosef Bacik u64 root_objectid = 0; 4881ef67963dSJosef Bacik int ret; 4882ef67963dSJosef Bacik 4883ef67963dSJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 4884ef67963dSJosef Bacik while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4885ef67963dSJosef Bacik (void **)gang, root_objectid, 4886ef67963dSJosef Bacik ARRAY_SIZE(gang))) != 0) { 4887ef67963dSJosef Bacik int i; 4888ef67963dSJosef Bacik 4889ef67963dSJosef Bacik for (i = 0; i < ret; i++) 4890ef67963dSJosef Bacik gang[i] = btrfs_grab_root(gang[i]); 4891ef67963dSJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 4892ef67963dSJosef Bacik 4893ef67963dSJosef Bacik for (i = 0; i < ret; i++) { 4894ef67963dSJosef Bacik if (!gang[i]) 4895ef67963dSJosef Bacik continue; 4896ef67963dSJosef Bacik root_objectid = gang[i]->root_key.objectid; 4897ef67963dSJosef Bacik btrfs_free_log(NULL, gang[i]); 4898ef67963dSJosef Bacik btrfs_put_root(gang[i]); 4899ef67963dSJosef Bacik } 4900ef67963dSJosef Bacik root_objectid++; 4901ef67963dSJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 4902ef67963dSJosef Bacik } 4903ef67963dSJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 4904ef67963dSJosef Bacik btrfs_free_log_root_tree(NULL, fs_info); 4905ef67963dSJosef Bacik } 4906ef67963dSJosef Bacik 4907143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4908acce952bSliubo { 4909acce952bSliubo struct btrfs_ordered_extent *ordered; 4910acce952bSliubo 4911199c2a9cSMiao Xie spin_lock(&root->ordered_extent_lock); 4912779880efSJosef Bacik /* 4913779880efSJosef Bacik * This will just short circuit the ordered completion stuff which will 4914779880efSJosef Bacik * make sure the ordered extent gets properly cleaned up. 4915779880efSJosef Bacik */ 4916199c2a9cSMiao Xie list_for_each_entry(ordered, &root->ordered_extents, 4917779880efSJosef Bacik root_extent_list) 4918779880efSJosef Bacik set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4919199c2a9cSMiao Xie spin_unlock(&root->ordered_extent_lock); 4920199c2a9cSMiao Xie } 4921199c2a9cSMiao Xie 4922199c2a9cSMiao Xie static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4923199c2a9cSMiao Xie { 4924199c2a9cSMiao Xie struct btrfs_root *root; 4925199c2a9cSMiao Xie struct list_head splice; 4926199c2a9cSMiao Xie 4927199c2a9cSMiao Xie INIT_LIST_HEAD(&splice); 4928199c2a9cSMiao Xie 4929199c2a9cSMiao Xie spin_lock(&fs_info->ordered_root_lock); 4930199c2a9cSMiao Xie list_splice_init(&fs_info->ordered_roots, &splice); 4931199c2a9cSMiao Xie while (!list_empty(&splice)) { 4932199c2a9cSMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4933199c2a9cSMiao Xie ordered_root); 49341de2cfdeSJosef Bacik list_move_tail(&root->ordered_root, 49351de2cfdeSJosef Bacik &fs_info->ordered_roots); 4936199c2a9cSMiao Xie 49372a85d9caSLiu Bo spin_unlock(&fs_info->ordered_root_lock); 4938199c2a9cSMiao Xie btrfs_destroy_ordered_extents(root); 4939199c2a9cSMiao Xie 49402a85d9caSLiu Bo cond_resched(); 49412a85d9caSLiu Bo spin_lock(&fs_info->ordered_root_lock); 4942199c2a9cSMiao Xie } 4943199c2a9cSMiao Xie spin_unlock(&fs_info->ordered_root_lock); 494474d5d229SJosef Bacik 494574d5d229SJosef Bacik /* 494674d5d229SJosef Bacik * We need this here because if we've been flipped read-only we won't 494774d5d229SJosef Bacik * get sync() from the umount, so we need to make sure any ordered 494874d5d229SJosef Bacik * extents that haven't had their dirty pages IO start writeout yet 494974d5d229SJosef Bacik * actually get run and error out properly. 495074d5d229SJosef Bacik */ 495174d5d229SJosef Bacik btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 4952acce952bSliubo } 4953acce952bSliubo 495435a3621bSStefan Behrens static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 49552ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4956acce952bSliubo { 4957acce952bSliubo struct rb_node *node; 4958acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 4959acce952bSliubo struct btrfs_delayed_ref_node *ref; 4960acce952bSliubo int ret = 0; 4961acce952bSliubo 4962acce952bSliubo delayed_refs = &trans->delayed_refs; 4963acce952bSliubo 4964acce952bSliubo spin_lock(&delayed_refs->lock); 4965d7df2c79SJosef Bacik if (atomic_read(&delayed_refs->num_entries) == 0) { 4966cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 4967b79ce3ddSDavid Sterba btrfs_debug(fs_info, "delayed_refs has NO entry"); 4968acce952bSliubo return ret; 4969acce952bSliubo } 4970acce952bSliubo 49715c9d028bSLiu Bo while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { 4972d7df2c79SJosef Bacik struct btrfs_delayed_ref_head *head; 49730e0adbcfSJosef Bacik struct rb_node *n; 4974e78417d1SJosef Bacik bool pin_bytes = false; 4975acce952bSliubo 4976d7df2c79SJosef Bacik head = rb_entry(node, struct btrfs_delayed_ref_head, 4977d7df2c79SJosef Bacik href_node); 49783069bd26SJosef Bacik if (btrfs_delayed_ref_lock(delayed_refs, head)) 4979b939d1abSJosef Bacik continue; 49803069bd26SJosef Bacik 4981d7df2c79SJosef Bacik spin_lock(&head->lock); 4982e3d03965SLiu Bo while ((n = rb_first_cached(&head->ref_tree)) != NULL) { 49830e0adbcfSJosef Bacik ref = rb_entry(n, struct btrfs_delayed_ref_node, 49840e0adbcfSJosef Bacik ref_node); 4985d7df2c79SJosef Bacik ref->in_tree = 0; 4986e3d03965SLiu Bo rb_erase_cached(&ref->ref_node, &head->ref_tree); 49870e0adbcfSJosef Bacik RB_CLEAR_NODE(&ref->ref_node); 49881d57ee94SWang Xiaoguang if (!list_empty(&ref->add_list)) 49891d57ee94SWang Xiaoguang list_del(&ref->add_list); 4990d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4991d7df2c79SJosef Bacik btrfs_put_delayed_ref(ref); 4992d7df2c79SJosef Bacik } 499354067ae9SJosef Bacik if (head->must_insert_reserved) 4994e78417d1SJosef Bacik pin_bytes = true; 499578a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 4996fa781ceaSJosef Bacik btrfs_delete_ref_head(delayed_refs, head); 4997d7df2c79SJosef Bacik spin_unlock(&head->lock); 4998acce952bSliubo spin_unlock(&delayed_refs->lock); 4999e78417d1SJosef Bacik mutex_unlock(&head->mutex); 5000acce952bSliubo 5001f603bb94SNikolay Borisov if (pin_bytes) { 5002f603bb94SNikolay Borisov struct btrfs_block_group *cache; 5003f603bb94SNikolay Borisov 5004f603bb94SNikolay Borisov cache = btrfs_lookup_block_group(fs_info, head->bytenr); 5005f603bb94SNikolay Borisov BUG_ON(!cache); 5006f603bb94SNikolay Borisov 5007f603bb94SNikolay Borisov spin_lock(&cache->space_info->lock); 5008f603bb94SNikolay Borisov spin_lock(&cache->lock); 5009f603bb94SNikolay Borisov cache->pinned += head->num_bytes; 5010f603bb94SNikolay Borisov btrfs_space_info_update_bytes_pinned(fs_info, 5011f603bb94SNikolay Borisov cache->space_info, head->num_bytes); 5012f603bb94SNikolay Borisov cache->reserved -= head->num_bytes; 5013f603bb94SNikolay Borisov cache->space_info->bytes_reserved -= head->num_bytes; 5014f603bb94SNikolay Borisov spin_unlock(&cache->lock); 5015f603bb94SNikolay Borisov spin_unlock(&cache->space_info->lock); 5016f603bb94SNikolay Borisov 5017f603bb94SNikolay Borisov btrfs_put_block_group(cache); 5018f603bb94SNikolay Borisov 5019f603bb94SNikolay Borisov btrfs_error_unpin_extent_range(fs_info, head->bytenr, 5020f603bb94SNikolay Borisov head->bytenr + head->num_bytes - 1); 5021f603bb94SNikolay Borisov } 502231890da0SJosef Bacik btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); 5023d278850eSJosef Bacik btrfs_put_delayed_ref_head(head); 5024acce952bSliubo cond_resched(); 5025acce952bSliubo spin_lock(&delayed_refs->lock); 5026acce952bSliubo } 502781f7eb00SJeff Mahoney btrfs_qgroup_destroy_extent_records(trans); 5028acce952bSliubo 5029acce952bSliubo spin_unlock(&delayed_refs->lock); 5030acce952bSliubo 5031acce952bSliubo return ret; 5032acce952bSliubo } 5033acce952bSliubo 5034143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 5035acce952bSliubo { 5036acce952bSliubo struct btrfs_inode *btrfs_inode; 5037acce952bSliubo struct list_head splice; 5038acce952bSliubo 5039acce952bSliubo INIT_LIST_HEAD(&splice); 5040acce952bSliubo 5041eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 5042eb73c1b7SMiao Xie list_splice_init(&root->delalloc_inodes, &splice); 5043acce952bSliubo 5044acce952bSliubo while (!list_empty(&splice)) { 5045fe816d0fSNikolay Borisov struct inode *inode = NULL; 5046eb73c1b7SMiao Xie btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 5047acce952bSliubo delalloc_inodes); 5048fe816d0fSNikolay Borisov __btrfs_del_delalloc_inode(root, btrfs_inode); 5049eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 5050acce952bSliubo 5051fe816d0fSNikolay Borisov /* 5052fe816d0fSNikolay Borisov * Make sure we get a live inode and that it'll not disappear 5053fe816d0fSNikolay Borisov * meanwhile. 5054fe816d0fSNikolay Borisov */ 5055fe816d0fSNikolay Borisov inode = igrab(&btrfs_inode->vfs_inode); 5056fe816d0fSNikolay Borisov if (inode) { 5057fe816d0fSNikolay Borisov invalidate_inode_pages2(inode->i_mapping); 5058fe816d0fSNikolay Borisov iput(inode); 5059fe816d0fSNikolay Borisov } 5060eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 5061acce952bSliubo } 5062eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 5063eb73c1b7SMiao Xie } 5064eb73c1b7SMiao Xie 5065eb73c1b7SMiao Xie static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 5066eb73c1b7SMiao Xie { 5067eb73c1b7SMiao Xie struct btrfs_root *root; 5068eb73c1b7SMiao Xie struct list_head splice; 5069eb73c1b7SMiao Xie 5070eb73c1b7SMiao Xie INIT_LIST_HEAD(&splice); 5071eb73c1b7SMiao Xie 5072eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 5073eb73c1b7SMiao Xie list_splice_init(&fs_info->delalloc_roots, &splice); 5074eb73c1b7SMiao Xie while (!list_empty(&splice)) { 5075eb73c1b7SMiao Xie root = list_first_entry(&splice, struct btrfs_root, 5076eb73c1b7SMiao Xie delalloc_root); 507700246528SJosef Bacik root = btrfs_grab_root(root); 5078eb73c1b7SMiao Xie BUG_ON(!root); 5079eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 5080eb73c1b7SMiao Xie 5081eb73c1b7SMiao Xie btrfs_destroy_delalloc_inodes(root); 508200246528SJosef Bacik btrfs_put_root(root); 5083eb73c1b7SMiao Xie 5084eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 5085eb73c1b7SMiao Xie } 5086eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 5087acce952bSliubo } 5088acce952bSliubo 50892ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 5090acce952bSliubo struct extent_io_tree *dirty_pages, 5091acce952bSliubo int mark) 5092acce952bSliubo { 5093acce952bSliubo int ret; 5094acce952bSliubo struct extent_buffer *eb; 5095acce952bSliubo u64 start = 0; 5096acce952bSliubo u64 end; 5097acce952bSliubo 5098acce952bSliubo while (1) { 5099acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 5100e6138876SJosef Bacik mark, NULL); 5101acce952bSliubo if (ret) 5102acce952bSliubo break; 5103acce952bSliubo 510491166212SDavid Sterba clear_extent_bits(dirty_pages, start, end, mark); 5105acce952bSliubo while (start <= end) { 51060b246afaSJeff Mahoney eb = find_extent_buffer(fs_info, start); 51070b246afaSJeff Mahoney start += fs_info->nodesize; 5108fd8b2b61SJosef Bacik if (!eb) 5109acce952bSliubo continue; 5110fd8b2b61SJosef Bacik wait_on_extent_buffer_writeback(eb); 5111acce952bSliubo 5112fd8b2b61SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 5113fd8b2b61SJosef Bacik &eb->bflags)) 5114fd8b2b61SJosef Bacik clear_extent_buffer_dirty(eb); 5115fd8b2b61SJosef Bacik free_extent_buffer_stale(eb); 5116acce952bSliubo } 5117acce952bSliubo } 5118acce952bSliubo 5119acce952bSliubo return ret; 5120acce952bSliubo } 5121acce952bSliubo 51222ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 5123fe119a6eSNikolay Borisov struct extent_io_tree *unpin) 5124acce952bSliubo { 5125acce952bSliubo u64 start; 5126acce952bSliubo u64 end; 5127acce952bSliubo int ret; 5128acce952bSliubo 5129acce952bSliubo while (1) { 51300e6ec385SFilipe Manana struct extent_state *cached_state = NULL; 51310e6ec385SFilipe Manana 5132fcd5e742SLu Fengqi /* 5133fcd5e742SLu Fengqi * The btrfs_finish_extent_commit() may get the same range as 5134fcd5e742SLu Fengqi * ours between find_first_extent_bit and clear_extent_dirty. 5135fcd5e742SLu Fengqi * Hence, hold the unused_bg_unpin_mutex to avoid double unpin 5136fcd5e742SLu Fengqi * the same extent range. 5137fcd5e742SLu Fengqi */ 5138fcd5e742SLu Fengqi mutex_lock(&fs_info->unused_bg_unpin_mutex); 5139acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 51400e6ec385SFilipe Manana EXTENT_DIRTY, &cached_state); 5141fcd5e742SLu Fengqi if (ret) { 5142fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 5143acce952bSliubo break; 5144fcd5e742SLu Fengqi } 5145acce952bSliubo 51460e6ec385SFilipe Manana clear_extent_dirty(unpin, start, end, &cached_state); 51470e6ec385SFilipe Manana free_extent_state(cached_state); 51482ff7e61eSJeff Mahoney btrfs_error_unpin_extent_range(fs_info, start, end); 5149fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 5150acce952bSliubo cond_resched(); 5151acce952bSliubo } 5152acce952bSliubo 5153acce952bSliubo return 0; 5154acce952bSliubo } 5155acce952bSliubo 515632da5386SDavid Sterba static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache) 5157c79a1751SLiu Bo { 5158c79a1751SLiu Bo struct inode *inode; 5159c79a1751SLiu Bo 5160c79a1751SLiu Bo inode = cache->io_ctl.inode; 5161c79a1751SLiu Bo if (inode) { 5162c79a1751SLiu Bo invalidate_inode_pages2(inode->i_mapping); 5163c79a1751SLiu Bo BTRFS_I(inode)->generation = 0; 5164c79a1751SLiu Bo cache->io_ctl.inode = NULL; 5165c79a1751SLiu Bo iput(inode); 5166c79a1751SLiu Bo } 5167bbc37d6eSFilipe Manana ASSERT(cache->io_ctl.pages == NULL); 5168c79a1751SLiu Bo btrfs_put_block_group(cache); 5169c79a1751SLiu Bo } 5170c79a1751SLiu Bo 5171c79a1751SLiu Bo void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, 51722ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 5173c79a1751SLiu Bo { 517432da5386SDavid Sterba struct btrfs_block_group *cache; 5175c79a1751SLiu Bo 5176c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 5177c79a1751SLiu Bo while (!list_empty(&cur_trans->dirty_bgs)) { 5178c79a1751SLiu Bo cache = list_first_entry(&cur_trans->dirty_bgs, 517932da5386SDavid Sterba struct btrfs_block_group, 5180c79a1751SLiu Bo dirty_list); 5181c79a1751SLiu Bo 5182c79a1751SLiu Bo if (!list_empty(&cache->io_list)) { 5183c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 5184c79a1751SLiu Bo list_del_init(&cache->io_list); 5185c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 5186c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 5187c79a1751SLiu Bo } 5188c79a1751SLiu Bo 5189c79a1751SLiu Bo list_del_init(&cache->dirty_list); 5190c79a1751SLiu Bo spin_lock(&cache->lock); 5191c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 5192c79a1751SLiu Bo spin_unlock(&cache->lock); 5193c79a1751SLiu Bo 5194c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 5195c79a1751SLiu Bo btrfs_put_block_group(cache); 5196ba2c4d4eSJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 5197c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 5198c79a1751SLiu Bo } 5199c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 5200c79a1751SLiu Bo 520145ae2c18SNikolay Borisov /* 520245ae2c18SNikolay Borisov * Refer to the definition of io_bgs member for details why it's safe 520345ae2c18SNikolay Borisov * to use it without any locking 520445ae2c18SNikolay Borisov */ 5205c79a1751SLiu Bo while (!list_empty(&cur_trans->io_bgs)) { 5206c79a1751SLiu Bo cache = list_first_entry(&cur_trans->io_bgs, 520732da5386SDavid Sterba struct btrfs_block_group, 5208c79a1751SLiu Bo io_list); 5209c79a1751SLiu Bo 5210c79a1751SLiu Bo list_del_init(&cache->io_list); 5211c79a1751SLiu Bo spin_lock(&cache->lock); 5212c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 5213c79a1751SLiu Bo spin_unlock(&cache->lock); 5214c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 5215c79a1751SLiu Bo } 5216c79a1751SLiu Bo } 5217c79a1751SLiu Bo 521849b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 52192ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 522049b25e05SJeff Mahoney { 5221bbbf7243SNikolay Borisov struct btrfs_device *dev, *tmp; 5222bbbf7243SNikolay Borisov 52232ff7e61eSJeff Mahoney btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 5224c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->dirty_bgs)); 5225c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->io_bgs)); 5226c79a1751SLiu Bo 5227bbbf7243SNikolay Borisov list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list, 5228bbbf7243SNikolay Borisov post_commit_list) { 5229bbbf7243SNikolay Borisov list_del_init(&dev->post_commit_list); 5230bbbf7243SNikolay Borisov } 5231bbbf7243SNikolay Borisov 52322ff7e61eSJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, fs_info); 523349b25e05SJeff Mahoney 52344a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 52350b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 523649b25e05SJeff Mahoney 52374a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 52380b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 523949b25e05SJeff Mahoney 5240ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 524167cde344SMiao Xie 52422ff7e61eSJeff Mahoney btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, 524349b25e05SJeff Mahoney EXTENT_DIRTY); 5244fe119a6eSNikolay Borisov btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents); 524549b25e05SJeff Mahoney 5246d3575156SNaohiro Aota btrfs_free_redirty_list(cur_trans); 5247d3575156SNaohiro Aota 52484a9d8bdeSMiao Xie cur_trans->state =TRANS_STATE_COMPLETED; 52494a9d8bdeSMiao Xie wake_up(&cur_trans->commit_wait); 525049b25e05SJeff Mahoney } 525149b25e05SJeff Mahoney 52522ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) 5253acce952bSliubo { 5254acce952bSliubo struct btrfs_transaction *t; 5255acce952bSliubo 52560b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 5257acce952bSliubo 52580b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 52590b246afaSJeff Mahoney while (!list_empty(&fs_info->trans_list)) { 52600b246afaSJeff Mahoney t = list_first_entry(&fs_info->trans_list, 5261724e2315SJosef Bacik struct btrfs_transaction, list); 5262724e2315SJosef Bacik if (t->state >= TRANS_STATE_COMMIT_START) { 52639b64f57dSElena Reshetova refcount_inc(&t->use_count); 52640b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 52652ff7e61eSJeff Mahoney btrfs_wait_for_commit(fs_info, t->transid); 5266724e2315SJosef Bacik btrfs_put_transaction(t); 52670b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5268724e2315SJosef Bacik continue; 5269724e2315SJosef Bacik } 52700b246afaSJeff Mahoney if (t == fs_info->running_transaction) { 5271724e2315SJosef Bacik t->state = TRANS_STATE_COMMIT_DOING; 52720b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 5273724e2315SJosef Bacik /* 5274724e2315SJosef Bacik * We wait for 0 num_writers since we don't hold a trans 5275724e2315SJosef Bacik * handle open currently for this transaction. 5276724e2315SJosef Bacik */ 5277724e2315SJosef Bacik wait_event(t->writer_wait, 5278724e2315SJosef Bacik atomic_read(&t->num_writers) == 0); 5279724e2315SJosef Bacik } else { 52800b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 5281724e2315SJosef Bacik } 52822ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(t, fs_info); 5283724e2315SJosef Bacik 52840b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 52850b246afaSJeff Mahoney if (t == fs_info->running_transaction) 52860b246afaSJeff Mahoney fs_info->running_transaction = NULL; 5287724e2315SJosef Bacik list_del_init(&t->list); 52880b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 5289a4abeea4SJosef Bacik 5290724e2315SJosef Bacik btrfs_put_transaction(t); 52912e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 52920b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5293724e2315SJosef Bacik } 52940b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 52950b246afaSJeff Mahoney btrfs_destroy_all_ordered_extents(fs_info); 5296ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 5297ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 52980b246afaSJeff Mahoney btrfs_destroy_all_delalloc_inodes(fs_info); 5299ef67963dSJosef Bacik btrfs_drop_all_logs(fs_info); 53000b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 5301acce952bSliubo 5302acce952bSliubo return 0; 5303acce952bSliubo } 5304ec7d6dfdSNikolay Borisov 5305453e4873SNikolay Borisov int btrfs_init_root_free_objectid(struct btrfs_root *root) 5306ec7d6dfdSNikolay Borisov { 5307ec7d6dfdSNikolay Borisov struct btrfs_path *path; 5308ec7d6dfdSNikolay Borisov int ret; 5309ec7d6dfdSNikolay Borisov struct extent_buffer *l; 5310ec7d6dfdSNikolay Borisov struct btrfs_key search_key; 5311ec7d6dfdSNikolay Borisov struct btrfs_key found_key; 5312ec7d6dfdSNikolay Borisov int slot; 5313ec7d6dfdSNikolay Borisov 5314ec7d6dfdSNikolay Borisov path = btrfs_alloc_path(); 5315ec7d6dfdSNikolay Borisov if (!path) 5316ec7d6dfdSNikolay Borisov return -ENOMEM; 5317ec7d6dfdSNikolay Borisov 5318ec7d6dfdSNikolay Borisov search_key.objectid = BTRFS_LAST_FREE_OBJECTID; 5319ec7d6dfdSNikolay Borisov search_key.type = -1; 5320ec7d6dfdSNikolay Borisov search_key.offset = (u64)-1; 5321ec7d6dfdSNikolay Borisov ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); 5322ec7d6dfdSNikolay Borisov if (ret < 0) 5323ec7d6dfdSNikolay Borisov goto error; 5324ec7d6dfdSNikolay Borisov BUG_ON(ret == 0); /* Corruption */ 5325ec7d6dfdSNikolay Borisov if (path->slots[0] > 0) { 5326ec7d6dfdSNikolay Borisov slot = path->slots[0] - 1; 5327ec7d6dfdSNikolay Borisov l = path->nodes[0]; 5328ec7d6dfdSNikolay Borisov btrfs_item_key_to_cpu(l, &found_key, slot); 532923125104SNikolay Borisov root->free_objectid = max_t(u64, found_key.objectid + 1, 533023125104SNikolay Borisov BTRFS_FIRST_FREE_OBJECTID); 5331ec7d6dfdSNikolay Borisov } else { 533223125104SNikolay Borisov root->free_objectid = BTRFS_FIRST_FREE_OBJECTID; 5333ec7d6dfdSNikolay Borisov } 5334ec7d6dfdSNikolay Borisov ret = 0; 5335ec7d6dfdSNikolay Borisov error: 5336ec7d6dfdSNikolay Borisov btrfs_free_path(path); 5337ec7d6dfdSNikolay Borisov return ret; 5338ec7d6dfdSNikolay Borisov } 5339ec7d6dfdSNikolay Borisov 5340543068a2SNikolay Borisov int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid) 5341ec7d6dfdSNikolay Borisov { 5342ec7d6dfdSNikolay Borisov int ret; 5343ec7d6dfdSNikolay Borisov mutex_lock(&root->objectid_mutex); 5344ec7d6dfdSNikolay Borisov 53456b8fad57SNikolay Borisov if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) { 5346ec7d6dfdSNikolay Borisov btrfs_warn(root->fs_info, 5347ec7d6dfdSNikolay Borisov "the objectid of root %llu reaches its highest value", 5348ec7d6dfdSNikolay Borisov root->root_key.objectid); 5349ec7d6dfdSNikolay Borisov ret = -ENOSPC; 5350ec7d6dfdSNikolay Borisov goto out; 5351ec7d6dfdSNikolay Borisov } 5352ec7d6dfdSNikolay Borisov 535323125104SNikolay Borisov *objectid = root->free_objectid++; 5354ec7d6dfdSNikolay Borisov ret = 0; 5355ec7d6dfdSNikolay Borisov out: 5356ec7d6dfdSNikolay Borisov mutex_unlock(&root->objectid_mutex); 5357ec7d6dfdSNikolay Borisov return ret; 5358ec7d6dfdSNikolay Borisov } 5359