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> 8fc7cbcd4SDavid Sterba #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 54143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root); 55acce952bSliubo static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 562ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info); 57143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root); 582ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 59acce952bSliubo struct extent_io_tree *dirty_pages, 60acce952bSliubo int mark); 612ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 62acce952bSliubo struct extent_io_tree *pinned_extents); 632ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info); 642ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info); 65ce9adaa5SChris Mason 66141386e1SJosef Bacik static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info) 67141386e1SJosef Bacik { 68141386e1SJosef Bacik if (fs_info->csum_shash) 69141386e1SJosef Bacik crypto_free_shash(fs_info->csum_shash); 70141386e1SJosef Bacik } 71141386e1SJosef Bacik 72d352ac68SChris Mason /* 73d352ac68SChris Mason * async submit bios are used to offload expensive checksumming 74d352ac68SChris Mason * onto the worker threads. They checksum file and metadata bios 75d352ac68SChris Mason * just before they are sent down the IO stack. 76d352ac68SChris Mason */ 7744b8bd7eSChris Mason struct async_submit_bio { 788896a08dSQu Wenruo struct inode *inode; 7944b8bd7eSChris Mason struct bio *bio; 80a758781dSDavid Sterba extent_submit_bio_start_t *submit_bio_start; 8144b8bd7eSChris Mason int mirror_num; 821941b64bSQu Wenruo 831941b64bSQu Wenruo /* Optional parameter for submit_bio_start used by direct io */ 841941b64bSQu Wenruo u64 dio_file_offset; 858b712842SChris Mason struct btrfs_work work; 864e4cbee9SChristoph Hellwig blk_status_t status; 8744b8bd7eSChris Mason }; 8844b8bd7eSChris Mason 8985d4e461SChris Mason /* 902996e1f8SJohannes Thumshirn * Compute the csum of a btree block and store the result to provided buffer. 91d352ac68SChris Mason */ 92c67b3892SDavid Sterba static void csum_tree_block(struct extent_buffer *buf, u8 *result) 9319c00ddcSChris Mason { 94d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = buf->fs_info; 957280305eSDavid Sterba const int num_pages = num_extent_pages(buf); 96a26663e7SQu Wenruo const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize); 97d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 9819c00ddcSChris Mason char *kaddr; 99e9be5a30SDavid Sterba int i; 100d5178578SJohannes Thumshirn 101d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 102d5178578SJohannes Thumshirn crypto_shash_init(shash); 103a26663e7SQu Wenruo kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start); 104e9be5a30SDavid Sterba crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE, 105a26663e7SQu Wenruo first_page_part - BTRFS_CSUM_SIZE); 10619c00ddcSChris Mason 107e9be5a30SDavid Sterba for (i = 1; i < num_pages; i++) { 108e9be5a30SDavid Sterba kaddr = page_address(buf->pages[i]); 109e9be5a30SDavid Sterba crypto_shash_update(shash, kaddr, PAGE_SIZE); 11019c00ddcSChris Mason } 11171a63551SDavid Sterba memset(result, 0, BTRFS_CSUM_SIZE); 112d5178578SJohannes Thumshirn crypto_shash_final(shash, result); 11319c00ddcSChris Mason } 11419c00ddcSChris Mason 115d352ac68SChris Mason /* 116d352ac68SChris Mason * we can't consider a given block up to date unless the transid of the 117d352ac68SChris Mason * block matches the transid in the parent node's pointer. This is how we 118d352ac68SChris Mason * detect blocks that either didn't get written at all or got written 119d352ac68SChris Mason * in the wrong place. 120d352ac68SChris Mason */ 1211259ab75SChris Mason static int verify_parent_transid(struct extent_io_tree *io_tree, 122b9fab919SChris Mason struct extent_buffer *eb, u64 parent_transid, 123b9fab919SChris Mason int atomic) 1241259ab75SChris Mason { 1252ac55d41SJosef Bacik struct extent_state *cached_state = NULL; 1261259ab75SChris Mason int ret; 1271259ab75SChris Mason 1281259ab75SChris Mason if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 1291259ab75SChris Mason return 0; 1301259ab75SChris Mason 131b9fab919SChris Mason if (atomic) 132b9fab919SChris Mason return -EAGAIN; 133b9fab919SChris Mason 1342ac55d41SJosef Bacik lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 135ff13db41SDavid Sterba &cached_state); 1360b32f4bbSJosef Bacik if (extent_buffer_uptodate(eb) && 1371259ab75SChris Mason btrfs_header_generation(eb) == parent_transid) { 1381259ab75SChris Mason ret = 0; 1391259ab75SChris Mason goto out; 1401259ab75SChris Mason } 14194647322SDavid Sterba btrfs_err_rl(eb->fs_info, 1428f0ed7d4SQu Wenruo "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu", 1438f0ed7d4SQu Wenruo eb->start, eb->read_mirror, 14429549aecSWang Shilong parent_transid, btrfs_header_generation(eb)); 1451259ab75SChris Mason ret = 1; 1460b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 14733958dc6SChris Mason out: 1482ac55d41SJosef Bacik unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 149e43bbe5eSDavid Sterba &cached_state); 1501259ab75SChris Mason return ret; 1511259ab75SChris Mason } 1521259ab75SChris Mason 153e7e16f48SJohannes Thumshirn static bool btrfs_supported_super_csum(u16 csum_type) 154e7e16f48SJohannes Thumshirn { 155e7e16f48SJohannes Thumshirn switch (csum_type) { 156e7e16f48SJohannes Thumshirn case BTRFS_CSUM_TYPE_CRC32: 1573951e7f0SJohannes Thumshirn case BTRFS_CSUM_TYPE_XXHASH: 1583831bf00SJohannes Thumshirn case BTRFS_CSUM_TYPE_SHA256: 159352ae07bSDavid Sterba case BTRFS_CSUM_TYPE_BLAKE2: 160e7e16f48SJohannes Thumshirn return true; 161e7e16f48SJohannes Thumshirn default: 162e7e16f48SJohannes Thumshirn return false; 163e7e16f48SJohannes Thumshirn } 164e7e16f48SJohannes Thumshirn } 165e7e16f48SJohannes Thumshirn 166d352ac68SChris Mason /* 1671104a885SDavid Sterba * Return 0 if the superblock checksum type matches the checksum value of that 1681104a885SDavid Sterba * algorithm. Pass the raw disk superblock data. 1691104a885SDavid Sterba */ 170ab8d0fc4SJeff Mahoney static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info, 171ab8d0fc4SJeff Mahoney char *raw_disk_sb) 1721104a885SDavid Sterba { 1731104a885SDavid Sterba struct btrfs_super_block *disk_sb = 1741104a885SDavid Sterba (struct btrfs_super_block *)raw_disk_sb; 17551bce6c9SJohannes Thumshirn char result[BTRFS_CSUM_SIZE]; 176d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 177d5178578SJohannes Thumshirn 178d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 1791104a885SDavid Sterba 1801104a885SDavid Sterba /* 1811104a885SDavid Sterba * The super_block structure does not span the whole 18251bce6c9SJohannes Thumshirn * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is 18351bce6c9SJohannes Thumshirn * filled with zeros and is included in the checksum. 1841104a885SDavid Sterba */ 185fd08001fSEric Biggers crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE, 186fd08001fSEric Biggers BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result); 1871104a885SDavid Sterba 18855fc29beSDavid Sterba if (memcmp(disk_sb->csum, result, fs_info->csum_size)) 189e7e16f48SJohannes Thumshirn return 1; 1901104a885SDavid Sterba 191e7e16f48SJohannes Thumshirn return 0; 1921104a885SDavid Sterba } 1931104a885SDavid Sterba 194e064d5e9SDavid Sterba int btrfs_verify_level_key(struct extent_buffer *eb, int level, 195ff76a864SLiu Bo struct btrfs_key *first_key, u64 parent_transid) 196581c1760SQu Wenruo { 197e064d5e9SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 198581c1760SQu Wenruo int found_level; 199581c1760SQu Wenruo struct btrfs_key found_key; 200581c1760SQu Wenruo int ret; 201581c1760SQu Wenruo 202581c1760SQu Wenruo found_level = btrfs_header_level(eb); 203581c1760SQu Wenruo if (found_level != level) { 20463489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 20563489055SQu Wenruo KERN_ERR "BTRFS: tree level check failed\n"); 206581c1760SQu Wenruo btrfs_err(fs_info, 207581c1760SQu Wenruo "tree level mismatch detected, bytenr=%llu level expected=%u has=%u", 208581c1760SQu Wenruo eb->start, level, found_level); 209581c1760SQu Wenruo return -EIO; 210581c1760SQu Wenruo } 211581c1760SQu Wenruo 212581c1760SQu Wenruo if (!first_key) 213581c1760SQu Wenruo return 0; 214581c1760SQu Wenruo 2155d41be6fSQu Wenruo /* 2165d41be6fSQu Wenruo * For live tree block (new tree blocks in current transaction), 2175d41be6fSQu Wenruo * we need proper lock context to avoid race, which is impossible here. 2185d41be6fSQu Wenruo * So we only checks tree blocks which is read from disk, whose 2195d41be6fSQu Wenruo * generation <= fs_info->last_trans_committed. 2205d41be6fSQu Wenruo */ 2215d41be6fSQu Wenruo if (btrfs_header_generation(eb) > fs_info->last_trans_committed) 2225d41be6fSQu Wenruo return 0; 22362fdaa52SQu Wenruo 22462fdaa52SQu Wenruo /* We have @first_key, so this @eb must have at least one item */ 22562fdaa52SQu Wenruo if (btrfs_header_nritems(eb) == 0) { 22662fdaa52SQu Wenruo btrfs_err(fs_info, 22762fdaa52SQu Wenruo "invalid tree nritems, bytenr=%llu nritems=0 expect >0", 22862fdaa52SQu Wenruo eb->start); 22962fdaa52SQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 23062fdaa52SQu Wenruo return -EUCLEAN; 23162fdaa52SQu Wenruo } 23262fdaa52SQu Wenruo 233581c1760SQu Wenruo if (found_level) 234581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &found_key, 0); 235581c1760SQu Wenruo else 236581c1760SQu Wenruo btrfs_item_key_to_cpu(eb, &found_key, 0); 237581c1760SQu Wenruo ret = btrfs_comp_cpu_keys(first_key, &found_key); 238581c1760SQu Wenruo 239581c1760SQu Wenruo if (ret) { 24063489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 24163489055SQu Wenruo KERN_ERR "BTRFS: tree first key check failed\n"); 242581c1760SQu Wenruo btrfs_err(fs_info, 243ff76a864SLiu Bo "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", 244ff76a864SLiu Bo eb->start, parent_transid, first_key->objectid, 245ff76a864SLiu Bo first_key->type, first_key->offset, 246ff76a864SLiu Bo found_key.objectid, found_key.type, 247ff76a864SLiu Bo found_key.offset); 248581c1760SQu Wenruo } 249581c1760SQu Wenruo return ret; 250581c1760SQu Wenruo } 251581c1760SQu Wenruo 2521104a885SDavid Sterba /* 253d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 254d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 255581c1760SQu Wenruo * 256581c1760SQu Wenruo * @parent_transid: expected transid, skip check if 0 257581c1760SQu Wenruo * @level: expected level, mandatory check 258581c1760SQu Wenruo * @first_key: expected key of first slot, skip check if NULL 259d352ac68SChris Mason */ 2606a2e9dc4SFilipe Manana int btrfs_read_extent_buffer(struct extent_buffer *eb, 261581c1760SQu Wenruo u64 parent_transid, int level, 262581c1760SQu Wenruo struct btrfs_key *first_key) 263f188591eSChris Mason { 2645ab12d1fSDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 265f188591eSChris Mason struct extent_io_tree *io_tree; 266ea466794SJosef Bacik int failed = 0; 267f188591eSChris Mason int ret; 268f188591eSChris Mason int num_copies = 0; 269f188591eSChris Mason int mirror_num = 0; 270ea466794SJosef Bacik int failed_mirror = 0; 271f188591eSChris Mason 2720b246afaSJeff Mahoney io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; 273f188591eSChris Mason while (1) { 274f8397d69SNikolay Borisov clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 275c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num); 276256dd1bbSStefan Behrens if (!ret) { 277581c1760SQu Wenruo if (verify_parent_transid(io_tree, eb, 278b9fab919SChris Mason parent_transid, 0)) 279256dd1bbSStefan Behrens ret = -EIO; 280e064d5e9SDavid Sterba else if (btrfs_verify_level_key(eb, level, 281ff76a864SLiu Bo first_key, parent_transid)) 282581c1760SQu Wenruo ret = -EUCLEAN; 283581c1760SQu Wenruo else 284581c1760SQu Wenruo break; 285256dd1bbSStefan Behrens } 286d397712bSChris Mason 2870b246afaSJeff Mahoney num_copies = btrfs_num_copies(fs_info, 288f188591eSChris Mason eb->start, eb->len); 2894235298eSChris Mason if (num_copies == 1) 290ea466794SJosef Bacik break; 2914235298eSChris Mason 2925cf1ab56SJosef Bacik if (!failed_mirror) { 2935cf1ab56SJosef Bacik failed = 1; 2945cf1ab56SJosef Bacik failed_mirror = eb->read_mirror; 2955cf1ab56SJosef Bacik } 2965cf1ab56SJosef Bacik 297f188591eSChris Mason mirror_num++; 298ea466794SJosef Bacik if (mirror_num == failed_mirror) 299ea466794SJosef Bacik mirror_num++; 300ea466794SJosef Bacik 3014235298eSChris Mason if (mirror_num > num_copies) 302ea466794SJosef Bacik break; 303f188591eSChris Mason } 304ea466794SJosef Bacik 305c0901581SStefan Behrens if (failed && !ret && failed_mirror) 30620a1fbf9SDavid Sterba btrfs_repair_eb_io_failure(eb, failed_mirror); 307ea466794SJosef Bacik 308ea466794SJosef Bacik return ret; 309f188591eSChris Mason } 31019c00ddcSChris Mason 311eca0f6f6SQu Wenruo static int csum_one_extent_buffer(struct extent_buffer *eb) 312eca0f6f6SQu Wenruo { 313eca0f6f6SQu Wenruo struct btrfs_fs_info *fs_info = eb->fs_info; 314eca0f6f6SQu Wenruo u8 result[BTRFS_CSUM_SIZE]; 315eca0f6f6SQu Wenruo int ret; 316eca0f6f6SQu Wenruo 317eca0f6f6SQu Wenruo ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid, 318eca0f6f6SQu Wenruo offsetof(struct btrfs_header, fsid), 319eca0f6f6SQu Wenruo BTRFS_FSID_SIZE) == 0); 320eca0f6f6SQu Wenruo csum_tree_block(eb, result); 321eca0f6f6SQu Wenruo 322eca0f6f6SQu Wenruo if (btrfs_header_level(eb)) 323eca0f6f6SQu Wenruo ret = btrfs_check_node(eb); 324eca0f6f6SQu Wenruo else 325eca0f6f6SQu Wenruo ret = btrfs_check_leaf_full(eb); 326eca0f6f6SQu Wenruo 3273777369fSQu Wenruo if (ret < 0) 3283777369fSQu Wenruo goto error; 3293777369fSQu Wenruo 3303777369fSQu Wenruo /* 3313777369fSQu Wenruo * Also check the generation, the eb reached here must be newer than 3323777369fSQu Wenruo * last committed. Or something seriously wrong happened. 3333777369fSQu Wenruo */ 3343777369fSQu Wenruo if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) { 3353777369fSQu Wenruo ret = -EUCLEAN; 336eca0f6f6SQu Wenruo btrfs_err(fs_info, 3373777369fSQu Wenruo "block=%llu bad generation, have %llu expect > %llu", 3383777369fSQu Wenruo eb->start, btrfs_header_generation(eb), 3393777369fSQu Wenruo fs_info->last_trans_committed); 3403777369fSQu Wenruo goto error; 341eca0f6f6SQu Wenruo } 342eca0f6f6SQu Wenruo write_extent_buffer(eb, result, 0, fs_info->csum_size); 343eca0f6f6SQu Wenruo 344eca0f6f6SQu Wenruo return 0; 3453777369fSQu Wenruo 3463777369fSQu Wenruo error: 3473777369fSQu Wenruo btrfs_print_tree(eb, 0); 3483777369fSQu Wenruo btrfs_err(fs_info, "block=%llu write time tree block corruption detected", 3493777369fSQu Wenruo eb->start); 3503777369fSQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 3513777369fSQu Wenruo return ret; 352eca0f6f6SQu Wenruo } 353eca0f6f6SQu Wenruo 354eca0f6f6SQu Wenruo /* Checksum all dirty extent buffers in one bio_vec */ 355eca0f6f6SQu Wenruo static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info, 356eca0f6f6SQu Wenruo struct bio_vec *bvec) 357eca0f6f6SQu Wenruo { 358eca0f6f6SQu Wenruo struct page *page = bvec->bv_page; 359eca0f6f6SQu Wenruo u64 bvec_start = page_offset(page) + bvec->bv_offset; 360eca0f6f6SQu Wenruo u64 cur; 361eca0f6f6SQu Wenruo int ret = 0; 362eca0f6f6SQu Wenruo 363eca0f6f6SQu Wenruo for (cur = bvec_start; cur < bvec_start + bvec->bv_len; 364eca0f6f6SQu Wenruo cur += fs_info->nodesize) { 365eca0f6f6SQu Wenruo struct extent_buffer *eb; 366eca0f6f6SQu Wenruo bool uptodate; 367eca0f6f6SQu Wenruo 368eca0f6f6SQu Wenruo eb = find_extent_buffer(fs_info, cur); 369eca0f6f6SQu Wenruo uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur, 370eca0f6f6SQu Wenruo fs_info->nodesize); 371eca0f6f6SQu Wenruo 37201cd3909SDavid Sterba /* A dirty eb shouldn't disappear from buffer_radix */ 373eca0f6f6SQu Wenruo if (WARN_ON(!eb)) 374eca0f6f6SQu Wenruo return -EUCLEAN; 375eca0f6f6SQu Wenruo 376eca0f6f6SQu Wenruo if (WARN_ON(cur != btrfs_header_bytenr(eb))) { 377eca0f6f6SQu Wenruo free_extent_buffer(eb); 378eca0f6f6SQu Wenruo return -EUCLEAN; 379eca0f6f6SQu Wenruo } 380eca0f6f6SQu Wenruo if (WARN_ON(!uptodate)) { 381eca0f6f6SQu Wenruo free_extent_buffer(eb); 382eca0f6f6SQu Wenruo return -EUCLEAN; 383eca0f6f6SQu Wenruo } 384eca0f6f6SQu Wenruo 385eca0f6f6SQu Wenruo ret = csum_one_extent_buffer(eb); 386eca0f6f6SQu Wenruo free_extent_buffer(eb); 387eca0f6f6SQu Wenruo if (ret < 0) 388eca0f6f6SQu Wenruo return ret; 389eca0f6f6SQu Wenruo } 390eca0f6f6SQu Wenruo return ret; 391eca0f6f6SQu Wenruo } 392eca0f6f6SQu Wenruo 393d352ac68SChris Mason /* 394ac303b69SQu Wenruo * Checksum a dirty tree block before IO. This has extra checks to make sure 395ac303b69SQu Wenruo * we only fill in the checksum field in the first page of a multi-page block. 396ac303b69SQu Wenruo * For subpage extent buffers we need bvec to also read the offset in the page. 397d352ac68SChris Mason */ 398ac303b69SQu Wenruo static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec) 39919c00ddcSChris Mason { 400ac303b69SQu Wenruo struct page *page = bvec->bv_page; 4014eee4fa4SMiao Xie u64 start = page_offset(page); 40219c00ddcSChris Mason u64 found_start; 40319c00ddcSChris Mason struct extent_buffer *eb; 404eca0f6f6SQu Wenruo 405fbca46ebSQu Wenruo if (fs_info->nodesize < PAGE_SIZE) 406eca0f6f6SQu Wenruo return csum_dirty_subpage_buffers(fs_info, bvec); 407f188591eSChris Mason 4084f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 4094f2de97aSJosef Bacik if (page != eb->pages[0]) 4104f2de97aSJosef Bacik return 0; 4110f805531SAlex Lyakas 41219c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 413d3575156SNaohiro Aota 414d3575156SNaohiro Aota if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) { 415d3575156SNaohiro Aota WARN_ON(found_start != 0); 416d3575156SNaohiro Aota return 0; 417d3575156SNaohiro Aota } 418d3575156SNaohiro Aota 4190f805531SAlex Lyakas /* 4200f805531SAlex Lyakas * Please do not consolidate these warnings into a single if. 4210f805531SAlex Lyakas * It is useful to know what went wrong. 4220f805531SAlex Lyakas */ 4230f805531SAlex Lyakas if (WARN_ON(found_start != start)) 4240f805531SAlex Lyakas return -EUCLEAN; 4250f805531SAlex Lyakas if (WARN_ON(!PageUptodate(page))) 4260f805531SAlex Lyakas return -EUCLEAN; 4270f805531SAlex Lyakas 428eca0f6f6SQu Wenruo return csum_one_extent_buffer(eb); 42919c00ddcSChris Mason } 43019c00ddcSChris Mason 431b0c9b3b0SDavid Sterba static int check_tree_block_fsid(struct extent_buffer *eb) 4322b82032cSYan Zheng { 433b0c9b3b0SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 434944d3f9fSNikolay Borisov struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 43544880fdcSAnand Jain u8 fsid[BTRFS_FSID_SIZE]; 436944d3f9fSNikolay Borisov u8 *metadata_uuid; 4372b82032cSYan Zheng 4389a8658e3SDavid Sterba read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid), 4399a8658e3SDavid Sterba BTRFS_FSID_SIZE); 4407239ff4bSNikolay Borisov /* 441944d3f9fSNikolay Borisov * Checking the incompat flag is only valid for the current fs. For 442944d3f9fSNikolay Borisov * seed devices it's forbidden to have their uuid changed so reading 443944d3f9fSNikolay Borisov * ->fsid in this case is fine 4447239ff4bSNikolay Borisov */ 445944d3f9fSNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID)) 4467239ff4bSNikolay Borisov metadata_uuid = fs_devices->metadata_uuid; 4477239ff4bSNikolay Borisov else 4487239ff4bSNikolay Borisov metadata_uuid = fs_devices->fsid; 4497239ff4bSNikolay Borisov 450944d3f9fSNikolay Borisov if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) 451944d3f9fSNikolay Borisov return 0; 452944d3f9fSNikolay Borisov 453944d3f9fSNikolay Borisov list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) 454944d3f9fSNikolay Borisov if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE)) 455944d3f9fSNikolay Borisov return 0; 456944d3f9fSNikolay Borisov 457944d3f9fSNikolay Borisov return 1; 4582b82032cSYan Zheng } 4592b82032cSYan Zheng 46077bf40a2SQu Wenruo /* Do basic extent buffer checks at read time */ 46177bf40a2SQu Wenruo static int validate_extent_buffer(struct extent_buffer *eb) 462ce9adaa5SChris Mason { 46377bf40a2SQu Wenruo struct btrfs_fs_info *fs_info = eb->fs_info; 464ce9adaa5SChris Mason u64 found_start; 46577bf40a2SQu Wenruo const u32 csum_size = fs_info->csum_size; 46677bf40a2SQu Wenruo u8 found_level; 4672996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 468dfd29eedSDavid Sterba const u8 *header_csum; 46977bf40a2SQu Wenruo int ret = 0; 470ea466794SJosef Bacik 471ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 472727011e0SChris Mason if (found_start != eb->start) { 4738f0ed7d4SQu Wenruo btrfs_err_rl(fs_info, 4748f0ed7d4SQu Wenruo "bad tree block start, mirror %u want %llu have %llu", 4758f0ed7d4SQu Wenruo eb->read_mirror, eb->start, found_start); 476f188591eSChris Mason ret = -EIO; 47777bf40a2SQu Wenruo goto out; 478ce9adaa5SChris Mason } 479b0c9b3b0SDavid Sterba if (check_tree_block_fsid(eb)) { 4808f0ed7d4SQu Wenruo btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u", 4818f0ed7d4SQu Wenruo eb->start, eb->read_mirror); 4821259ab75SChris Mason ret = -EIO; 48377bf40a2SQu Wenruo goto out; 4841259ab75SChris Mason } 485ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 4861c24c3ceSJosef Bacik if (found_level >= BTRFS_MAX_LEVEL) { 4878f0ed7d4SQu Wenruo btrfs_err(fs_info, 4888f0ed7d4SQu Wenruo "bad tree block level, mirror %u level %d on logical %llu", 4898f0ed7d4SQu Wenruo eb->read_mirror, btrfs_header_level(eb), eb->start); 4901c24c3ceSJosef Bacik ret = -EIO; 49177bf40a2SQu Wenruo goto out; 4921c24c3ceSJosef Bacik } 493ce9adaa5SChris Mason 494c67b3892SDavid Sterba csum_tree_block(eb, result); 495dfd29eedSDavid Sterba header_csum = page_address(eb->pages[0]) + 496dfd29eedSDavid Sterba get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum)); 497a826d6dcSJosef Bacik 498dfd29eedSDavid Sterba if (memcmp(result, header_csum, csum_size) != 0) { 4992996e1f8SJohannes Thumshirn btrfs_warn_rl(fs_info, 5008f0ed7d4SQu Wenruo "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d", 5018f0ed7d4SQu Wenruo eb->start, eb->read_mirror, 502dfd29eedSDavid Sterba CSUM_FMT_VALUE(csum_size, header_csum), 50335be8851SJohannes Thumshirn CSUM_FMT_VALUE(csum_size, result), 50435be8851SJohannes Thumshirn btrfs_header_level(eb)); 5052996e1f8SJohannes Thumshirn ret = -EUCLEAN; 50677bf40a2SQu Wenruo goto out; 5072996e1f8SJohannes Thumshirn } 5082996e1f8SJohannes Thumshirn 509a826d6dcSJosef Bacik /* 510a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 511a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 512a826d6dcSJosef Bacik * return -EIO. 513a826d6dcSJosef Bacik */ 5141c4360eeSDavid Sterba if (found_level == 0 && btrfs_check_leaf_full(eb)) { 515a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 516a826d6dcSJosef Bacik ret = -EIO; 517a826d6dcSJosef Bacik } 518ce9adaa5SChris Mason 519813fd1dcSDavid Sterba if (found_level > 0 && btrfs_check_node(eb)) 520053ab70fSLiu Bo ret = -EIO; 521053ab70fSLiu Bo 5220b32f4bbSJosef Bacik if (!ret) 5230b32f4bbSJosef Bacik set_extent_buffer_uptodate(eb); 52475391f0dSQu Wenruo else 52575391f0dSQu Wenruo btrfs_err(fs_info, 5268f0ed7d4SQu Wenruo "read time tree block corruption detected on logical %llu mirror %u", 5278f0ed7d4SQu Wenruo eb->start, eb->read_mirror); 52877bf40a2SQu Wenruo out: 52977bf40a2SQu Wenruo return ret; 53077bf40a2SQu Wenruo } 53177bf40a2SQu Wenruo 532371cdc07SQu Wenruo static int validate_subpage_buffer(struct page *page, u64 start, u64 end, 533371cdc07SQu Wenruo int mirror) 534371cdc07SQu Wenruo { 535371cdc07SQu Wenruo struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); 536371cdc07SQu Wenruo struct extent_buffer *eb; 537371cdc07SQu Wenruo bool reads_done; 538371cdc07SQu Wenruo int ret = 0; 539371cdc07SQu Wenruo 540371cdc07SQu Wenruo /* 541371cdc07SQu Wenruo * We don't allow bio merge for subpage metadata read, so we should 542371cdc07SQu Wenruo * only get one eb for each endio hook. 543371cdc07SQu Wenruo */ 544371cdc07SQu Wenruo ASSERT(end == start + fs_info->nodesize - 1); 545371cdc07SQu Wenruo ASSERT(PagePrivate(page)); 546371cdc07SQu Wenruo 547371cdc07SQu Wenruo eb = find_extent_buffer(fs_info, start); 548371cdc07SQu Wenruo /* 549371cdc07SQu Wenruo * When we are reading one tree block, eb must have been inserted into 550371cdc07SQu Wenruo * the radix tree. If not, something is wrong. 551371cdc07SQu Wenruo */ 552371cdc07SQu Wenruo ASSERT(eb); 553371cdc07SQu Wenruo 554371cdc07SQu Wenruo reads_done = atomic_dec_and_test(&eb->io_pages); 555371cdc07SQu Wenruo /* Subpage read must finish in page read */ 556371cdc07SQu Wenruo ASSERT(reads_done); 557371cdc07SQu Wenruo 558371cdc07SQu Wenruo eb->read_mirror = mirror; 559371cdc07SQu Wenruo if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 560371cdc07SQu Wenruo ret = -EIO; 561371cdc07SQu Wenruo goto err; 562371cdc07SQu Wenruo } 563371cdc07SQu Wenruo ret = validate_extent_buffer(eb); 564371cdc07SQu Wenruo if (ret < 0) 565371cdc07SQu Wenruo goto err; 566371cdc07SQu Wenruo 567371cdc07SQu Wenruo set_extent_buffer_uptodate(eb); 568371cdc07SQu Wenruo 569371cdc07SQu Wenruo free_extent_buffer(eb); 570371cdc07SQu Wenruo return ret; 571371cdc07SQu Wenruo err: 572371cdc07SQu Wenruo /* 573371cdc07SQu Wenruo * end_bio_extent_readpage decrements io_pages in case of error, 574371cdc07SQu Wenruo * make sure it has something to decrement. 575371cdc07SQu Wenruo */ 576371cdc07SQu Wenruo atomic_inc(&eb->io_pages); 577371cdc07SQu Wenruo clear_extent_buffer_uptodate(eb); 578371cdc07SQu Wenruo free_extent_buffer(eb); 579371cdc07SQu Wenruo return ret; 580371cdc07SQu Wenruo } 581371cdc07SQu Wenruo 582c3a3b19bSQu Wenruo int btrfs_validate_metadata_buffer(struct btrfs_bio *bbio, 58377bf40a2SQu Wenruo struct page *page, u64 start, u64 end, 58477bf40a2SQu Wenruo int mirror) 58577bf40a2SQu Wenruo { 58677bf40a2SQu Wenruo struct extent_buffer *eb; 58777bf40a2SQu Wenruo int ret = 0; 58877bf40a2SQu Wenruo int reads_done; 58977bf40a2SQu Wenruo 59077bf40a2SQu Wenruo ASSERT(page->private); 591371cdc07SQu Wenruo 592fbca46ebSQu Wenruo if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE) 593371cdc07SQu Wenruo return validate_subpage_buffer(page, start, end, mirror); 594371cdc07SQu Wenruo 59577bf40a2SQu Wenruo eb = (struct extent_buffer *)page->private; 59677bf40a2SQu Wenruo 59777bf40a2SQu Wenruo /* 59877bf40a2SQu Wenruo * The pending IO might have been the only thing that kept this buffer 59977bf40a2SQu Wenruo * in memory. Make sure we have a ref for all this other checks 60077bf40a2SQu Wenruo */ 60177bf40a2SQu Wenruo atomic_inc(&eb->refs); 60277bf40a2SQu Wenruo 60377bf40a2SQu Wenruo reads_done = atomic_dec_and_test(&eb->io_pages); 60477bf40a2SQu Wenruo if (!reads_done) 60577bf40a2SQu Wenruo goto err; 60677bf40a2SQu Wenruo 60777bf40a2SQu Wenruo eb->read_mirror = mirror; 60877bf40a2SQu Wenruo if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 60977bf40a2SQu Wenruo ret = -EIO; 61077bf40a2SQu Wenruo goto err; 61177bf40a2SQu Wenruo } 61277bf40a2SQu Wenruo ret = validate_extent_buffer(eb); 613ce9adaa5SChris Mason err: 61453b381b3SDavid Woodhouse if (ret) { 61553b381b3SDavid Woodhouse /* 61653b381b3SDavid Woodhouse * our io error hook is going to dec the io pages 61753b381b3SDavid Woodhouse * again, we have to make sure it has something 61853b381b3SDavid Woodhouse * to decrement 61953b381b3SDavid Woodhouse */ 62053b381b3SDavid Woodhouse atomic_inc(&eb->io_pages); 6210b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 62253b381b3SDavid Woodhouse } 6230b32f4bbSJosef Bacik free_extent_buffer(eb); 62477bf40a2SQu Wenruo 625f188591eSChris Mason return ret; 626ce9adaa5SChris Mason } 627ce9adaa5SChris Mason 6284a69a410SChris Mason static void run_one_async_start(struct btrfs_work *work) 6294a69a410SChris Mason { 6304a69a410SChris Mason struct async_submit_bio *async; 6314e4cbee9SChristoph Hellwig blk_status_t ret; 6324a69a410SChris Mason 6334a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 6341941b64bSQu Wenruo ret = async->submit_bio_start(async->inode, async->bio, 6351941b64bSQu Wenruo async->dio_file_offset); 63679787eaaSJeff Mahoney if (ret) 6374e4cbee9SChristoph Hellwig async->status = ret; 6384a69a410SChris Mason } 6394a69a410SChris Mason 64006ea01b1SDavid Sterba /* 64106ea01b1SDavid Sterba * In order to insert checksums into the metadata in large chunks, we wait 64206ea01b1SDavid Sterba * until bio submission time. All the pages in the bio are checksummed and 64306ea01b1SDavid Sterba * sums are attached onto the ordered extent record. 64406ea01b1SDavid Sterba * 64506ea01b1SDavid Sterba * At IO completion time the csums attached on the ordered extent record are 64606ea01b1SDavid Sterba * inserted into the tree. 64706ea01b1SDavid Sterba */ 6484a69a410SChris Mason static void run_one_async_done(struct btrfs_work *work) 6498b712842SChris Mason { 650917f32a2SChristoph Hellwig struct async_submit_bio *async = 651917f32a2SChristoph Hellwig container_of(work, struct async_submit_bio, work); 652917f32a2SChristoph Hellwig struct inode *inode = async->inode; 653917f32a2SChristoph Hellwig struct btrfs_bio *bbio = btrfs_bio(async->bio); 6544854ddd0SChris Mason 655bb7ab3b9SAdam Buchbinder /* If an error occurred we just want to clean up the bio and move on */ 6564e4cbee9SChristoph Hellwig if (async->status) { 657917f32a2SChristoph Hellwig btrfs_bio_end_io(bbio, async->status); 65879787eaaSJeff Mahoney return; 65979787eaaSJeff Mahoney } 66079787eaaSJeff Mahoney 661ec39f769SChris Mason /* 662ec39f769SChris Mason * All of the bios that pass through here are from async helpers. 663ec39f769SChris Mason * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context. 664ec39f769SChris Mason * This changes nothing when cgroups aren't in use. 665ec39f769SChris Mason */ 666ec39f769SChris Mason async->bio->bi_opf |= REQ_CGROUP_PUNT; 6671a722d8fSChristoph Hellwig btrfs_submit_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num); 6684a69a410SChris Mason } 6694a69a410SChris Mason 6704a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 6714a69a410SChris Mason { 6724a69a410SChris Mason struct async_submit_bio *async; 6734a69a410SChris Mason 6744a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 6758b712842SChris Mason kfree(async); 6768b712842SChris Mason } 6778b712842SChris Mason 678ea1f0cedSChristoph Hellwig /* 679ea1f0cedSChristoph Hellwig * Submit bio to an async queue. 680ea1f0cedSChristoph Hellwig * 681ea1f0cedSChristoph Hellwig * Retrun: 682ea1f0cedSChristoph Hellwig * - true if the work has been succesfuly submitted 683ea1f0cedSChristoph Hellwig * - false in case of error 684ea1f0cedSChristoph Hellwig */ 685ea1f0cedSChristoph Hellwig bool btrfs_wq_submit_bio(struct inode *inode, struct bio *bio, int mirror_num, 686ea1f0cedSChristoph Hellwig u64 dio_file_offset, 687e288c080SDavid Sterba extent_submit_bio_start_t *submit_bio_start) 68844b8bd7eSChris Mason { 6898896a08dSQu Wenruo struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; 69044b8bd7eSChris Mason struct async_submit_bio *async; 69144b8bd7eSChris Mason 69244b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 69344b8bd7eSChris Mason if (!async) 694ea1f0cedSChristoph Hellwig return false; 69544b8bd7eSChris Mason 6968896a08dSQu Wenruo async->inode = inode; 69744b8bd7eSChris Mason async->bio = bio; 69844b8bd7eSChris Mason async->mirror_num = mirror_num; 6994a69a410SChris Mason async->submit_bio_start = submit_bio_start; 7004a69a410SChris Mason 701a0cac0ecSOmar Sandoval btrfs_init_work(&async->work, run_one_async_start, run_one_async_done, 702a0cac0ecSOmar Sandoval run_one_async_free); 7034a69a410SChris Mason 7041941b64bSQu Wenruo async->dio_file_offset = dio_file_offset; 7058c8bee1dSChris Mason 7064e4cbee9SChristoph Hellwig async->status = 0; 70779787eaaSJeff Mahoney 70867f055c7SChristoph Hellwig if (op_is_sync(bio->bi_opf)) 709a31b4a43SChristoph Hellwig btrfs_queue_work(fs_info->hipri_workers, &async->work); 710a31b4a43SChristoph Hellwig else 7115cdc7ad3SQu Wenruo btrfs_queue_work(fs_info->workers, &async->work); 712ea1f0cedSChristoph Hellwig return true; 71344b8bd7eSChris Mason } 71444b8bd7eSChris Mason 7154e4cbee9SChristoph Hellwig static blk_status_t btree_csum_one_bio(struct bio *bio) 716ce3ed71aSChris Mason { 7172c30c71bSKent Overstreet struct bio_vec *bvec; 718ce3ed71aSChris Mason struct btrfs_root *root; 7192b070cfeSChristoph Hellwig int ret = 0; 7206dc4f100SMing Lei struct bvec_iter_all iter_all; 721ce3ed71aSChris Mason 722c09abff8SDavid Sterba ASSERT(!bio_flagged(bio, BIO_CLONED)); 7232b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 724ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 725ac303b69SQu Wenruo ret = csum_dirty_buffer(root->fs_info, bvec); 72679787eaaSJeff Mahoney if (ret) 72779787eaaSJeff Mahoney break; 728ce3ed71aSChris Mason } 7292c30c71bSKent Overstreet 7304e4cbee9SChristoph Hellwig return errno_to_blk_status(ret); 731ce3ed71aSChris Mason } 732ce3ed71aSChris Mason 7338896a08dSQu Wenruo static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio, 7341941b64bSQu Wenruo u64 dio_file_offset) 73522c59948SChris Mason { 7368b712842SChris Mason /* 7378b712842SChris Mason * when we're called for a write, we're already in the async 7381a722d8fSChristoph Hellwig * submission context. Just jump into btrfs_submit_bio. 7398b712842SChris Mason */ 74079787eaaSJeff Mahoney return btree_csum_one_bio(bio); 74122c59948SChris Mason } 74222c59948SChris Mason 743f4dcfb30SJohannes Thumshirn static bool should_async_write(struct btrfs_fs_info *fs_info, 7449b4e675aSDavid Sterba struct btrfs_inode *bi) 745de0022b9SJosef Bacik { 7464eef29efSNaohiro Aota if (btrfs_is_zoned(fs_info)) 747f4dcfb30SJohannes Thumshirn return false; 7486300463bSLiu Bo if (atomic_read(&bi->sync_writers)) 749f4dcfb30SJohannes Thumshirn return false; 7509b4e675aSDavid Sterba if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags)) 751f4dcfb30SJohannes Thumshirn return false; 752f4dcfb30SJohannes Thumshirn return true; 753de0022b9SJosef Bacik } 754de0022b9SJosef Bacik 75594d9e11bSChristoph Hellwig void btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio, int mirror_num) 75644b8bd7eSChris Mason { 7570b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 758917f32a2SChristoph Hellwig struct btrfs_bio *bbio = btrfs_bio(bio); 7594e4cbee9SChristoph Hellwig blk_status_t ret; 760cad321adSChris Mason 76108a6f464SChristoph Hellwig bio->bi_opf |= REQ_META; 76208a6f464SChristoph Hellwig 763cfe94440SNaohiro Aota if (btrfs_op(bio) != BTRFS_MAP_WRITE) { 7641a722d8fSChristoph Hellwig btrfs_submit_bio(fs_info, bio, mirror_num); 7651a722d8fSChristoph Hellwig return; 76644b8bd7eSChris Mason } 76744b8bd7eSChris Mason 768ea1f0cedSChristoph Hellwig /* 769ea1f0cedSChristoph Hellwig * Kthread helpers are used to submit writes so that checksumming can 770ea1f0cedSChristoph Hellwig * happen in parallel across all CPUs. 771ea1f0cedSChristoph Hellwig */ 772ea1f0cedSChristoph Hellwig if (should_async_write(fs_info, BTRFS_I(inode)) && 773ea1f0cedSChristoph Hellwig btrfs_wq_submit_bio(inode, bio, mirror_num, 0, btree_submit_bio_start)) 774ea1f0cedSChristoph Hellwig return; 775ea1f0cedSChristoph Hellwig 776ea1f0cedSChristoph Hellwig ret = btree_csum_one_bio(bio); 77794d9e11bSChristoph Hellwig if (ret) { 778917f32a2SChristoph Hellwig btrfs_bio_end_io(bbio, ret); 779ea1f0cedSChristoph Hellwig return; 78094d9e11bSChristoph Hellwig } 781ea1f0cedSChristoph Hellwig 782ea1f0cedSChristoph Hellwig btrfs_submit_bio(fs_info, bio, mirror_num); 78361891923SStefan Behrens } 78461891923SStefan Behrens 7853dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 7868958b551SMatthew Wilcox (Oracle) static int btree_migrate_folio(struct address_space *mapping, 7878958b551SMatthew Wilcox (Oracle) struct folio *dst, struct folio *src, enum migrate_mode mode) 788784b4e29SChris Mason { 789784b4e29SChris Mason /* 790784b4e29SChris Mason * we can't safely write a btree page from here, 791784b4e29SChris Mason * we haven't done the locking hook 792784b4e29SChris Mason */ 7938958b551SMatthew Wilcox (Oracle) if (folio_test_dirty(src)) 794784b4e29SChris Mason return -EAGAIN; 795784b4e29SChris Mason /* 796784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 797784b4e29SChris Mason * We must have no buffers or drop them. 798784b4e29SChris Mason */ 7998958b551SMatthew Wilcox (Oracle) if (folio_get_private(src) && 8008958b551SMatthew Wilcox (Oracle) !filemap_release_folio(src, GFP_KERNEL)) 801784b4e29SChris Mason return -EAGAIN; 80254184650SMatthew Wilcox (Oracle) return migrate_folio(mapping, dst, src, mode); 803784b4e29SChris Mason } 8048958b551SMatthew Wilcox (Oracle) #else 8058958b551SMatthew Wilcox (Oracle) #define btree_migrate_folio NULL 8063dd1462eSJan Beulich #endif 807784b4e29SChris Mason 8080da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 8090da5468fSChris Mason struct writeback_control *wbc) 8100da5468fSChris Mason { 811e2d84521SMiao Xie struct btrfs_fs_info *fs_info; 812e2d84521SMiao Xie int ret; 813e2d84521SMiao Xie 814d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 815448d640bSChris Mason 816448d640bSChris Mason if (wbc->for_kupdate) 817448d640bSChris Mason return 0; 818448d640bSChris Mason 819e2d84521SMiao Xie fs_info = BTRFS_I(mapping->host)->root->fs_info; 820b9473439SChris Mason /* this is a bit racy, but that's ok */ 821d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 822d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 823d814a491SEthan Lien fs_info->dirty_metadata_batch); 824e2d84521SMiao Xie if (ret < 0) 825793955bcSChris Mason return 0; 826793955bcSChris Mason } 8270b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 8280da5468fSChris Mason } 8290da5468fSChris Mason 830f913cff3SMatthew Wilcox (Oracle) static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags) 8315f39d397SChris Mason { 832f913cff3SMatthew Wilcox (Oracle) if (folio_test_writeback(folio) || folio_test_dirty(folio)) 833f913cff3SMatthew Wilcox (Oracle) return false; 8340c4e538bSDavid Sterba 835f913cff3SMatthew Wilcox (Oracle) return try_release_extent_buffer(&folio->page); 836d98237b3SChris Mason } 837d98237b3SChris Mason 838895586ebSMatthew Wilcox (Oracle) static void btree_invalidate_folio(struct folio *folio, size_t offset, 839895586ebSMatthew Wilcox (Oracle) size_t length) 840d98237b3SChris Mason { 841d1310b2eSChris Mason struct extent_io_tree *tree; 842895586ebSMatthew Wilcox (Oracle) tree = &BTRFS_I(folio->mapping->host)->io_tree; 843895586ebSMatthew Wilcox (Oracle) extent_invalidate_folio(tree, folio, offset); 844f913cff3SMatthew Wilcox (Oracle) btree_release_folio(folio, GFP_NOFS); 845895586ebSMatthew Wilcox (Oracle) if (folio_get_private(folio)) { 846895586ebSMatthew Wilcox (Oracle) btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info, 847895586ebSMatthew Wilcox (Oracle) "folio private not zero on folio %llu", 848895586ebSMatthew Wilcox (Oracle) (unsigned long long)folio_pos(folio)); 849895586ebSMatthew Wilcox (Oracle) folio_detach_private(folio); 8509ad6b7bcSChris Mason } 851d98237b3SChris Mason } 852d98237b3SChris Mason 853bb146eb2SJosef Bacik #ifdef DEBUG 8540079c3b1SMatthew Wilcox (Oracle) static bool btree_dirty_folio(struct address_space *mapping, 8550079c3b1SMatthew Wilcox (Oracle) struct folio *folio) 8560079c3b1SMatthew Wilcox (Oracle) { 8570079c3b1SMatthew Wilcox (Oracle) struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb); 858139e8cd3SQu Wenruo struct btrfs_subpage *subpage; 8590b32f4bbSJosef Bacik struct extent_buffer *eb; 860139e8cd3SQu Wenruo int cur_bit = 0; 8610079c3b1SMatthew Wilcox (Oracle) u64 page_start = folio_pos(folio); 8620b32f4bbSJosef Bacik 863139e8cd3SQu Wenruo if (fs_info->sectorsize == PAGE_SIZE) { 8640079c3b1SMatthew Wilcox (Oracle) eb = folio_get_private(folio); 8650b32f4bbSJosef Bacik BUG_ON(!eb); 8660b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 8670b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 86849d0c642SFilipe Manana btrfs_assert_tree_write_locked(eb); 8690079c3b1SMatthew Wilcox (Oracle) return filemap_dirty_folio(mapping, folio); 870139e8cd3SQu Wenruo } 8710079c3b1SMatthew Wilcox (Oracle) subpage = folio_get_private(folio); 872139e8cd3SQu Wenruo 873139e8cd3SQu Wenruo ASSERT(subpage->dirty_bitmap); 874139e8cd3SQu Wenruo while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) { 875139e8cd3SQu Wenruo unsigned long flags; 876139e8cd3SQu Wenruo u64 cur; 877139e8cd3SQu Wenruo u16 tmp = (1 << cur_bit); 878139e8cd3SQu Wenruo 879139e8cd3SQu Wenruo spin_lock_irqsave(&subpage->lock, flags); 880139e8cd3SQu Wenruo if (!(tmp & subpage->dirty_bitmap)) { 881139e8cd3SQu Wenruo spin_unlock_irqrestore(&subpage->lock, flags); 882139e8cd3SQu Wenruo cur_bit++; 883139e8cd3SQu Wenruo continue; 884139e8cd3SQu Wenruo } 885139e8cd3SQu Wenruo spin_unlock_irqrestore(&subpage->lock, flags); 886139e8cd3SQu Wenruo cur = page_start + cur_bit * fs_info->sectorsize; 887139e8cd3SQu Wenruo 888139e8cd3SQu Wenruo eb = find_extent_buffer(fs_info, cur); 889139e8cd3SQu Wenruo ASSERT(eb); 890139e8cd3SQu Wenruo ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 891139e8cd3SQu Wenruo ASSERT(atomic_read(&eb->refs)); 89249d0c642SFilipe Manana btrfs_assert_tree_write_locked(eb); 893139e8cd3SQu Wenruo free_extent_buffer(eb); 894139e8cd3SQu Wenruo 895139e8cd3SQu Wenruo cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits); 896139e8cd3SQu Wenruo } 8970079c3b1SMatthew Wilcox (Oracle) return filemap_dirty_folio(mapping, folio); 8980b32f4bbSJosef Bacik } 8990079c3b1SMatthew Wilcox (Oracle) #else 9000079c3b1SMatthew Wilcox (Oracle) #define btree_dirty_folio filemap_dirty_folio 9010079c3b1SMatthew Wilcox (Oracle) #endif 9020b32f4bbSJosef Bacik 9037f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 9040da5468fSChris Mason .writepages = btree_writepages, 905f913cff3SMatthew Wilcox (Oracle) .release_folio = btree_release_folio, 906895586ebSMatthew Wilcox (Oracle) .invalidate_folio = btree_invalidate_folio, 9078958b551SMatthew Wilcox (Oracle) .migrate_folio = btree_migrate_folio, 9080079c3b1SMatthew Wilcox (Oracle) .dirty_folio = btree_dirty_folio, 909d98237b3SChris Mason }; 910123abc88SChris Mason 9112ff7e61eSJeff Mahoney struct extent_buffer *btrfs_find_create_tree_block( 9122ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info, 9133fbaf258SJosef Bacik u64 bytenr, u64 owner_root, 9143fbaf258SJosef Bacik int level) 9150999df54SChris Mason { 9160b246afaSJeff Mahoney if (btrfs_is_testing(fs_info)) 9170b246afaSJeff Mahoney return alloc_test_extent_buffer(fs_info, bytenr); 9183fbaf258SJosef Bacik return alloc_extent_buffer(fs_info, bytenr, owner_root, level); 9190999df54SChris Mason } 9200999df54SChris Mason 921581c1760SQu Wenruo /* 922581c1760SQu Wenruo * Read tree block at logical address @bytenr and do variant basic but critical 923581c1760SQu Wenruo * verification. 924581c1760SQu Wenruo * 9251b7ec85eSJosef Bacik * @owner_root: the objectid of the root owner for this block. 926581c1760SQu Wenruo * @parent_transid: expected transid of this tree block, skip check if 0 927581c1760SQu Wenruo * @level: expected level, mandatory check 928581c1760SQu Wenruo * @first_key: expected key in slot 0, skip check if NULL 929581c1760SQu Wenruo */ 9302ff7e61eSJeff Mahoney struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, 9311b7ec85eSJosef Bacik u64 owner_root, u64 parent_transid, 9321b7ec85eSJosef Bacik int level, struct btrfs_key *first_key) 933e20d96d6SChris Mason { 9345f39d397SChris Mason struct extent_buffer *buf = NULL; 93519c00ddcSChris Mason int ret; 93619c00ddcSChris Mason 9373fbaf258SJosef Bacik buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level); 938c871b0f2SLiu Bo if (IS_ERR(buf)) 939c871b0f2SLiu Bo return buf; 940e4204dedSChris Mason 9416a2e9dc4SFilipe Manana ret = btrfs_read_extent_buffer(buf, parent_transid, level, first_key); 9420f0fe8f7SFilipe David Borba Manana if (ret) { 943537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 94464c043deSLiu Bo return ERR_PTR(ret); 9450f0fe8f7SFilipe David Borba Manana } 94688c602abSQu Wenruo if (btrfs_check_eb_owner(buf, owner_root)) { 94788c602abSQu Wenruo free_extent_buffer_stale(buf); 94888c602abSQu Wenruo return ERR_PTR(-EUCLEAN); 94988c602abSQu Wenruo } 9505f39d397SChris Mason return buf; 951ce9adaa5SChris Mason 952eb60ceacSChris Mason } 953eb60ceacSChris Mason 9546a884d7dSDavid Sterba void btrfs_clean_tree_block(struct extent_buffer *buf) 955ed2ff2cbSChris Mason { 9566a884d7dSDavid Sterba struct btrfs_fs_info *fs_info = buf->fs_info; 95755c69072SChris Mason if (btrfs_header_generation(buf) == 958e2d84521SMiao Xie fs_info->running_transaction->transid) { 95949d0c642SFilipe Manana btrfs_assert_tree_write_locked(buf); 960b4ce94deSChris Mason 961b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 962104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 963e2d84521SMiao Xie -buf->len, 964e2d84521SMiao Xie fs_info->dirty_metadata_batch); 9650b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 966925baeddSChris Mason } 9675f39d397SChris Mason } 968ed7b63ebSJosef Bacik } 9695f39d397SChris Mason 970da17066cSJeff Mahoney static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, 971e20d96d6SChris Mason u64 objectid) 972d97e63b6SChris Mason { 9737c0260eeSJeff Mahoney bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); 9742e608bd1SJosef Bacik 9752e608bd1SJosef Bacik memset(&root->root_key, 0, sizeof(root->root_key)); 9762e608bd1SJosef Bacik memset(&root->root_item, 0, sizeof(root->root_item)); 9772e608bd1SJosef Bacik memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 97896dfcb46SJosef Bacik root->fs_info = fs_info; 9792e608bd1SJosef Bacik root->root_key.objectid = objectid; 980cfaa7295SChris Mason root->node = NULL; 981a28ec197SChris Mason root->commit_root = NULL; 98227cdeb70SMiao Xie root->state = 0; 983abed4aaaSJosef Bacik RB_CLEAR_NODE(&root->rb_node); 9840b86a832SChris Mason 9850f7d52f4SChris Mason root->last_trans = 0; 9866b8fad57SNikolay Borisov root->free_objectid = 0; 987eb73c1b7SMiao Xie root->nr_delalloc_inodes = 0; 988199c2a9cSMiao Xie root->nr_ordered_extents = 0; 9896bef4d31SEric Paris root->inode_tree = RB_ROOT; 990088aea3bSDavid Sterba INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 9912e608bd1SJosef Bacik 9922e608bd1SJosef Bacik btrfs_init_root_block_rsv(root); 9930b86a832SChris Mason 9940b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 9955d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 996eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_inodes); 997eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_root); 998199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_extents); 999199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_root); 1000d2311e69SQu Wenruo INIT_LIST_HEAD(&root->reloc_dirty_list); 10012ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 10022ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 10035d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1004eb73c1b7SMiao Xie spin_lock_init(&root->delalloc_lock); 1005199c2a9cSMiao Xie spin_lock_init(&root->ordered_extent_lock); 1006f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 10072ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 10082ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 10098287475aSQu Wenruo spin_lock_init(&root->qgroup_meta_rsv_lock); 1010a2135011SChris Mason mutex_init(&root->objectid_mutex); 1011e02119d5SChris Mason mutex_init(&root->log_mutex); 101231f3d255SMiao Xie mutex_init(&root->ordered_extent_mutex); 1013573bfb72SMiao Xie mutex_init(&root->delalloc_mutex); 1014c53e9653SQu Wenruo init_waitqueue_head(&root->qgroup_flush_wait); 10157237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 10167237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 10177237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 10188b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[0]); 10198b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[1]); 10207237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 10217237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 10227237f183SYan Zheng atomic_set(&root->log_writers, 0); 10232ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 10240700cea7SElena Reshetova refcount_set(&root->refs, 1); 10258ecebf4dSRobbie Ko atomic_set(&root->snapshot_force_cow, 0); 1026eede2bf3SOmar Sandoval atomic_set(&root->nr_swapfiles, 0); 10277237f183SYan Zheng root->log_transid = 0; 1028d1433debSMiao Xie root->log_transid_committed = -1; 1029257c62e1SChris Mason root->last_log_commit = 0; 10302e608bd1SJosef Bacik root->anon_dev = 0; 1031e289f03eSFilipe Manana if (!dummy) { 103243eb5f29SQu Wenruo extent_io_tree_init(fs_info, &root->dirty_log_pages, 103343eb5f29SQu Wenruo IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL); 1034e289f03eSFilipe Manana extent_io_tree_init(fs_info, &root->log_csum_range, 1035e289f03eSFilipe Manana IO_TREE_LOG_CSUM_RANGE, NULL); 1036e289f03eSFilipe Manana } 1037017e5369SChris Mason 10385f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 1039370a11b8SQu Wenruo btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks); 1040bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 1041bd647ce3SJosef Bacik INIT_LIST_HEAD(&root->leak_list); 1042fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1043bd647ce3SJosef Bacik list_add_tail(&root->leak_list, &fs_info->allocated_roots); 1044fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 1045bd647ce3SJosef Bacik #endif 10463768f368SChris Mason } 10473768f368SChris Mason 104874e4d827SDavid Sterba static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, 104996dfcb46SJosef Bacik u64 objectid, gfp_t flags) 10506f07e42eSAl Viro { 105174e4d827SDavid Sterba struct btrfs_root *root = kzalloc(sizeof(*root), flags); 10526f07e42eSAl Viro if (root) 105396dfcb46SJosef Bacik __setup_root(root, fs_info, objectid); 10546f07e42eSAl Viro return root; 10556f07e42eSAl Viro } 10566f07e42eSAl Viro 105706ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 105806ea65a3SJosef Bacik /* Should only be used by the testing infrastructure */ 1059da17066cSJeff Mahoney struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) 106006ea65a3SJosef Bacik { 106106ea65a3SJosef Bacik struct btrfs_root *root; 106206ea65a3SJosef Bacik 10637c0260eeSJeff Mahoney if (!fs_info) 10647c0260eeSJeff Mahoney return ERR_PTR(-EINVAL); 10657c0260eeSJeff Mahoney 106696dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL); 106706ea65a3SJosef Bacik if (!root) 106806ea65a3SJosef Bacik return ERR_PTR(-ENOMEM); 1069da17066cSJeff Mahoney 1070b9ef22deSFeifei Xu /* We don't use the stripesize in selftest, set it as sectorsize */ 1071faa2dbf0SJosef Bacik root->alloc_bytenr = 0; 107206ea65a3SJosef Bacik 107306ea65a3SJosef Bacik return root; 107406ea65a3SJosef Bacik } 107506ea65a3SJosef Bacik #endif 107606ea65a3SJosef Bacik 1077abed4aaaSJosef Bacik static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node) 1078abed4aaaSJosef Bacik { 1079abed4aaaSJosef Bacik const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node); 1080abed4aaaSJosef Bacik const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node); 1081abed4aaaSJosef Bacik 1082abed4aaaSJosef Bacik return btrfs_comp_cpu_keys(&a->root_key, &b->root_key); 1083abed4aaaSJosef Bacik } 1084abed4aaaSJosef Bacik 1085abed4aaaSJosef Bacik static int global_root_key_cmp(const void *k, const struct rb_node *node) 1086abed4aaaSJosef Bacik { 1087abed4aaaSJosef Bacik const struct btrfs_key *key = k; 1088abed4aaaSJosef Bacik const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node); 1089abed4aaaSJosef Bacik 1090abed4aaaSJosef Bacik return btrfs_comp_cpu_keys(key, &root->root_key); 1091abed4aaaSJosef Bacik } 1092abed4aaaSJosef Bacik 1093abed4aaaSJosef Bacik int btrfs_global_root_insert(struct btrfs_root *root) 1094abed4aaaSJosef Bacik { 1095abed4aaaSJosef Bacik struct btrfs_fs_info *fs_info = root->fs_info; 1096abed4aaaSJosef Bacik struct rb_node *tmp; 1097abed4aaaSJosef Bacik 1098abed4aaaSJosef Bacik write_lock(&fs_info->global_root_lock); 1099abed4aaaSJosef Bacik tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp); 1100abed4aaaSJosef Bacik write_unlock(&fs_info->global_root_lock); 1101abed4aaaSJosef Bacik ASSERT(!tmp); 1102abed4aaaSJosef Bacik 1103abed4aaaSJosef Bacik return tmp ? -EEXIST : 0; 1104abed4aaaSJosef Bacik } 1105abed4aaaSJosef Bacik 1106abed4aaaSJosef Bacik void btrfs_global_root_delete(struct btrfs_root *root) 1107abed4aaaSJosef Bacik { 1108abed4aaaSJosef Bacik struct btrfs_fs_info *fs_info = root->fs_info; 1109abed4aaaSJosef Bacik 1110abed4aaaSJosef Bacik write_lock(&fs_info->global_root_lock); 1111abed4aaaSJosef Bacik rb_erase(&root->rb_node, &fs_info->global_root_tree); 1112abed4aaaSJosef Bacik write_unlock(&fs_info->global_root_lock); 1113abed4aaaSJosef Bacik } 1114abed4aaaSJosef Bacik 1115abed4aaaSJosef Bacik struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info, 1116abed4aaaSJosef Bacik struct btrfs_key *key) 1117abed4aaaSJosef Bacik { 1118abed4aaaSJosef Bacik struct rb_node *node; 1119abed4aaaSJosef Bacik struct btrfs_root *root = NULL; 1120abed4aaaSJosef Bacik 1121abed4aaaSJosef Bacik read_lock(&fs_info->global_root_lock); 1122abed4aaaSJosef Bacik node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp); 1123abed4aaaSJosef Bacik if (node) 1124abed4aaaSJosef Bacik root = container_of(node, struct btrfs_root, rb_node); 1125abed4aaaSJosef Bacik read_unlock(&fs_info->global_root_lock); 1126abed4aaaSJosef Bacik 1127abed4aaaSJosef Bacik return root; 1128abed4aaaSJosef Bacik } 1129abed4aaaSJosef Bacik 1130f7238e50SJosef Bacik static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr) 1131f7238e50SJosef Bacik { 1132f7238e50SJosef Bacik struct btrfs_block_group *block_group; 1133f7238e50SJosef Bacik u64 ret; 1134f7238e50SJosef Bacik 1135f7238e50SJosef Bacik if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) 1136f7238e50SJosef Bacik return 0; 1137f7238e50SJosef Bacik 1138f7238e50SJosef Bacik if (bytenr) 1139f7238e50SJosef Bacik block_group = btrfs_lookup_block_group(fs_info, bytenr); 1140f7238e50SJosef Bacik else 1141f7238e50SJosef Bacik block_group = btrfs_lookup_first_block_group(fs_info, bytenr); 1142f7238e50SJosef Bacik ASSERT(block_group); 1143f7238e50SJosef Bacik if (!block_group) 1144f7238e50SJosef Bacik return 0; 1145f7238e50SJosef Bacik ret = block_group->global_root_id; 1146f7238e50SJosef Bacik btrfs_put_block_group(block_group); 1147f7238e50SJosef Bacik 1148f7238e50SJosef Bacik return ret; 1149f7238e50SJosef Bacik } 1150f7238e50SJosef Bacik 1151abed4aaaSJosef Bacik struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr) 1152abed4aaaSJosef Bacik { 1153abed4aaaSJosef Bacik struct btrfs_key key = { 1154abed4aaaSJosef Bacik .objectid = BTRFS_CSUM_TREE_OBJECTID, 1155abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 1156f7238e50SJosef Bacik .offset = btrfs_global_root_id(fs_info, bytenr), 1157abed4aaaSJosef Bacik }; 1158abed4aaaSJosef Bacik 1159abed4aaaSJosef Bacik return btrfs_global_root(fs_info, &key); 1160abed4aaaSJosef Bacik } 1161abed4aaaSJosef Bacik 1162abed4aaaSJosef Bacik struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr) 1163abed4aaaSJosef Bacik { 1164abed4aaaSJosef Bacik struct btrfs_key key = { 1165abed4aaaSJosef Bacik .objectid = BTRFS_EXTENT_TREE_OBJECTID, 1166abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 1167f7238e50SJosef Bacik .offset = btrfs_global_root_id(fs_info, bytenr), 1168abed4aaaSJosef Bacik }; 1169abed4aaaSJosef Bacik 1170abed4aaaSJosef Bacik return btrfs_global_root(fs_info, &key); 1171abed4aaaSJosef Bacik } 1172abed4aaaSJosef Bacik 117320897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 117420897f5cSArne Jansen u64 objectid) 117520897f5cSArne Jansen { 11769b7a2440SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 117720897f5cSArne Jansen struct extent_buffer *leaf; 117820897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 117920897f5cSArne Jansen struct btrfs_root *root; 118020897f5cSArne Jansen struct btrfs_key key; 1181b89f6d1fSFilipe Manana unsigned int nofs_flag; 118220897f5cSArne Jansen int ret = 0; 118320897f5cSArne Jansen 1184b89f6d1fSFilipe Manana /* 1185b89f6d1fSFilipe Manana * We're holding a transaction handle, so use a NOFS memory allocation 1186b89f6d1fSFilipe Manana * context to avoid deadlock if reclaim happens. 1187b89f6d1fSFilipe Manana */ 1188b89f6d1fSFilipe Manana nofs_flag = memalloc_nofs_save(); 118996dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL); 1190b89f6d1fSFilipe Manana memalloc_nofs_restore(nofs_flag); 119120897f5cSArne Jansen if (!root) 119220897f5cSArne Jansen return ERR_PTR(-ENOMEM); 119320897f5cSArne Jansen 119420897f5cSArne Jansen root->root_key.objectid = objectid; 119520897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 119620897f5cSArne Jansen root->root_key.offset = 0; 119720897f5cSArne Jansen 11989631e4ccSJosef Bacik leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0, 11999631e4ccSJosef Bacik BTRFS_NESTING_NORMAL); 120020897f5cSArne Jansen if (IS_ERR(leaf)) { 120120897f5cSArne Jansen ret = PTR_ERR(leaf); 12021dd05682STsutomu Itoh leaf = NULL; 12038a6a87cdSBoris Burkov goto fail_unlock; 120420897f5cSArne Jansen } 120520897f5cSArne Jansen 120620897f5cSArne Jansen root->node = leaf; 120720897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 120820897f5cSArne Jansen 120920897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 121027cdeb70SMiao Xie set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 121120897f5cSArne Jansen 1212f944d2cbSDavid Sterba btrfs_set_root_flags(&root->root_item, 0); 1213f944d2cbSDavid Sterba btrfs_set_root_limit(&root->root_item, 0); 121420897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 121520897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 121620897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 121720897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 121820897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 121920897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 122020897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 122133d85fdaSQu Wenruo if (is_fstree(objectid)) 1222807fc790SAndy Shevchenko generate_random_guid(root->root_item.uuid); 1223807fc790SAndy Shevchenko else 1224807fc790SAndy Shevchenko export_guid(root->root_item.uuid, &guid_null); 1225c8422684SDavid Sterba btrfs_set_root_drop_level(&root->root_item, 0); 122620897f5cSArne Jansen 12278a6a87cdSBoris Burkov btrfs_tree_unlock(leaf); 12288a6a87cdSBoris Burkov 122920897f5cSArne Jansen key.objectid = objectid; 123020897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 123120897f5cSArne Jansen key.offset = 0; 123220897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 123320897f5cSArne Jansen if (ret) 123420897f5cSArne Jansen goto fail; 123520897f5cSArne Jansen 123620897f5cSArne Jansen return root; 12371dd05682STsutomu Itoh 12388a6a87cdSBoris Burkov fail_unlock: 12398c38938cSJosef Bacik if (leaf) 12401dd05682STsutomu Itoh btrfs_tree_unlock(leaf); 12418a6a87cdSBoris Burkov fail: 124200246528SJosef Bacik btrfs_put_root(root); 12431dd05682STsutomu Itoh 12441dd05682STsutomu Itoh return ERR_PTR(ret); 124520897f5cSArne Jansen } 124620897f5cSArne Jansen 12477237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1248e02119d5SChris Mason struct btrfs_fs_info *fs_info) 12490f7d52f4SChris Mason { 12500f7d52f4SChris Mason struct btrfs_root *root; 1251e02119d5SChris Mason 125296dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS); 1253e02119d5SChris Mason if (!root) 12547237f183SYan Zheng return ERR_PTR(-ENOMEM); 1255e02119d5SChris Mason 1256e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1257e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1258e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 125927cdeb70SMiao Xie 12606ab6ebb7SNaohiro Aota return root; 12616ab6ebb7SNaohiro Aota } 12626ab6ebb7SNaohiro Aota 12636ab6ebb7SNaohiro Aota int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans, 12646ab6ebb7SNaohiro Aota struct btrfs_root *root) 12656ab6ebb7SNaohiro Aota { 12666ab6ebb7SNaohiro Aota struct extent_buffer *leaf; 12676ab6ebb7SNaohiro Aota 12687237f183SYan Zheng /* 126992a7cc42SQu Wenruo * DON'T set SHAREABLE bit for log trees. 127027cdeb70SMiao Xie * 127192a7cc42SQu Wenruo * Log trees are not exposed to user space thus can't be snapshotted, 127292a7cc42SQu Wenruo * and they go away before a real commit is actually done. 127392a7cc42SQu Wenruo * 127492a7cc42SQu Wenruo * They do store pointers to file data extents, and those reference 127592a7cc42SQu Wenruo * counts still get updated (along with back refs to the log tree). 12767237f183SYan Zheng */ 1277e02119d5SChris Mason 12784d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 12799631e4ccSJosef Bacik NULL, 0, 0, 0, BTRFS_NESTING_NORMAL); 12806ab6ebb7SNaohiro Aota if (IS_ERR(leaf)) 12816ab6ebb7SNaohiro Aota return PTR_ERR(leaf); 1282e02119d5SChris Mason 12837237f183SYan Zheng root->node = leaf; 1284e02119d5SChris Mason 1285e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1286e02119d5SChris Mason btrfs_tree_unlock(root->node); 12876ab6ebb7SNaohiro Aota 12886ab6ebb7SNaohiro Aota return 0; 12897237f183SYan Zheng } 12907237f183SYan Zheng 12917237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 12927237f183SYan Zheng struct btrfs_fs_info *fs_info) 12937237f183SYan Zheng { 12947237f183SYan Zheng struct btrfs_root *log_root; 12957237f183SYan Zheng 12967237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 12977237f183SYan Zheng if (IS_ERR(log_root)) 12987237f183SYan Zheng return PTR_ERR(log_root); 12996ab6ebb7SNaohiro Aota 13003ddebf27SNaohiro Aota if (!btrfs_is_zoned(fs_info)) { 13013ddebf27SNaohiro Aota int ret = btrfs_alloc_log_tree_node(trans, log_root); 13023ddebf27SNaohiro Aota 13036ab6ebb7SNaohiro Aota if (ret) { 13046ab6ebb7SNaohiro Aota btrfs_put_root(log_root); 13056ab6ebb7SNaohiro Aota return ret; 13066ab6ebb7SNaohiro Aota } 13073ddebf27SNaohiro Aota } 13086ab6ebb7SNaohiro Aota 13097237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 13107237f183SYan Zheng fs_info->log_root_tree = log_root; 13117237f183SYan Zheng return 0; 13127237f183SYan Zheng } 13137237f183SYan Zheng 13147237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 13157237f183SYan Zheng struct btrfs_root *root) 13167237f183SYan Zheng { 13170b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13187237f183SYan Zheng struct btrfs_root *log_root; 13197237f183SYan Zheng struct btrfs_inode_item *inode_item; 13206ab6ebb7SNaohiro Aota int ret; 13217237f183SYan Zheng 13220b246afaSJeff Mahoney log_root = alloc_log_tree(trans, fs_info); 13237237f183SYan Zheng if (IS_ERR(log_root)) 13247237f183SYan Zheng return PTR_ERR(log_root); 13257237f183SYan Zheng 13266ab6ebb7SNaohiro Aota ret = btrfs_alloc_log_tree_node(trans, log_root); 13276ab6ebb7SNaohiro Aota if (ret) { 13286ab6ebb7SNaohiro Aota btrfs_put_root(log_root); 13296ab6ebb7SNaohiro Aota return ret; 13306ab6ebb7SNaohiro Aota } 13316ab6ebb7SNaohiro Aota 13327237f183SYan Zheng log_root->last_trans = trans->transid; 13337237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 13347237f183SYan Zheng 13357237f183SYan Zheng inode_item = &log_root->root_item.inode; 13363cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 13373cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 13383cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 1339da17066cSJeff Mahoney btrfs_set_stack_inode_nbytes(inode_item, 13400b246afaSJeff Mahoney fs_info->nodesize); 13413cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 13427237f183SYan Zheng 13435d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 13447237f183SYan Zheng 13457237f183SYan Zheng WARN_ON(root->log_root); 13467237f183SYan Zheng root->log_root = log_root; 13477237f183SYan Zheng root->log_transid = 0; 1348d1433debSMiao Xie root->log_transid_committed = -1; 1349257c62e1SChris Mason root->last_log_commit = 0; 1350e02119d5SChris Mason return 0; 1351e02119d5SChris Mason } 1352e02119d5SChris Mason 135349d11beaSJosef Bacik static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root, 135449d11beaSJosef Bacik struct btrfs_path *path, 1355cb517eabSMiao Xie struct btrfs_key *key) 1356e02119d5SChris Mason { 1357e02119d5SChris Mason struct btrfs_root *root; 1358e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 135984234f3aSYan Zheng u64 generation; 1360cb517eabSMiao Xie int ret; 1361581c1760SQu Wenruo int level; 1362cb517eabSMiao Xie 136396dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS); 136449d11beaSJosef Bacik if (!root) 136549d11beaSJosef Bacik return ERR_PTR(-ENOMEM); 13660f7d52f4SChris Mason 1367cb517eabSMiao Xie ret = btrfs_find_root(tree_root, key, path, 1368cb517eabSMiao Xie &root->root_item, &root->root_key); 13690f7d52f4SChris Mason if (ret) { 137013a8a7c8SYan, Zheng if (ret > 0) 137113a8a7c8SYan, Zheng ret = -ENOENT; 137249d11beaSJosef Bacik goto fail; 13730f7d52f4SChris Mason } 137413a8a7c8SYan, Zheng 137584234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1376581c1760SQu Wenruo level = btrfs_root_level(&root->root_item); 13772ff7e61eSJeff Mahoney root->node = read_tree_block(fs_info, 13782ff7e61eSJeff Mahoney btrfs_root_bytenr(&root->root_item), 13791b7ec85eSJosef Bacik key->objectid, generation, level, NULL); 138064c043deSLiu Bo if (IS_ERR(root->node)) { 138164c043deSLiu Bo ret = PTR_ERR(root->node); 13828c38938cSJosef Bacik root->node = NULL; 138349d11beaSJosef Bacik goto fail; 13844eb150d6SQu Wenruo } 13854eb150d6SQu Wenruo if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1386cb517eabSMiao Xie ret = -EIO; 138749d11beaSJosef Bacik goto fail; 1388416bc658SJosef Bacik } 138988c602abSQu Wenruo 139088c602abSQu Wenruo /* 139188c602abSQu Wenruo * For real fs, and not log/reloc trees, root owner must 139288c602abSQu Wenruo * match its root node owner 139388c602abSQu Wenruo */ 139488c602abSQu Wenruo if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) && 139588c602abSQu Wenruo root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID && 139688c602abSQu Wenruo root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && 139788c602abSQu Wenruo root->root_key.objectid != btrfs_header_owner(root->node)) { 139888c602abSQu Wenruo btrfs_crit(fs_info, 139988c602abSQu Wenruo "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu", 140088c602abSQu Wenruo root->root_key.objectid, root->node->start, 140188c602abSQu Wenruo btrfs_header_owner(root->node), 140288c602abSQu Wenruo root->root_key.objectid); 140388c602abSQu Wenruo ret = -EUCLEAN; 140488c602abSQu Wenruo goto fail; 140588c602abSQu Wenruo } 14065d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 1407cb517eabSMiao Xie return root; 140849d11beaSJosef Bacik fail: 140900246528SJosef Bacik btrfs_put_root(root); 141049d11beaSJosef Bacik return ERR_PTR(ret); 141149d11beaSJosef Bacik } 141249d11beaSJosef Bacik 141349d11beaSJosef Bacik struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 141449d11beaSJosef Bacik struct btrfs_key *key) 141549d11beaSJosef Bacik { 141649d11beaSJosef Bacik struct btrfs_root *root; 141749d11beaSJosef Bacik struct btrfs_path *path; 141849d11beaSJosef Bacik 141949d11beaSJosef Bacik path = btrfs_alloc_path(); 142049d11beaSJosef Bacik if (!path) 142149d11beaSJosef Bacik return ERR_PTR(-ENOMEM); 142249d11beaSJosef Bacik root = read_tree_root_path(tree_root, path, key); 142349d11beaSJosef Bacik btrfs_free_path(path); 142449d11beaSJosef Bacik 142549d11beaSJosef Bacik return root; 1426cb517eabSMiao Xie } 1427cb517eabSMiao Xie 14282dfb1e43SQu Wenruo /* 14292dfb1e43SQu Wenruo * Initialize subvolume root in-memory structure 14302dfb1e43SQu Wenruo * 14312dfb1e43SQu Wenruo * @anon_dev: anonymous device to attach to the root, if zero, allocate new 14322dfb1e43SQu Wenruo */ 14332dfb1e43SQu Wenruo static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev) 1434cb517eabSMiao Xie { 1435cb517eabSMiao Xie int ret; 1436dcc3eb96SNikolay Borisov unsigned int nofs_flag; 1437cb517eabSMiao Xie 1438dcc3eb96SNikolay Borisov /* 1439dcc3eb96SNikolay Borisov * We might be called under a transaction (e.g. indirect backref 1440dcc3eb96SNikolay Borisov * resolution) which could deadlock if it triggers memory reclaim 1441dcc3eb96SNikolay Borisov */ 1442dcc3eb96SNikolay Borisov nofs_flag = memalloc_nofs_save(); 1443dcc3eb96SNikolay Borisov ret = btrfs_drew_lock_init(&root->snapshot_lock); 1444dcc3eb96SNikolay Borisov memalloc_nofs_restore(nofs_flag); 1445dcc3eb96SNikolay Borisov if (ret) 14468257b2dcSMiao Xie goto fail; 14478257b2dcSMiao Xie 1448aeb935a4SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID && 144937f00a6dSJohannes Thumshirn !btrfs_is_data_reloc_root(root)) { 145092a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &root->state); 1451f39e4571SJosef Bacik btrfs_check_and_init_root_item(&root->root_item); 1452f39e4571SJosef Bacik } 1453f39e4571SJosef Bacik 1454851fd730SQu Wenruo /* 1455851fd730SQu Wenruo * Don't assign anonymous block device to roots that are not exposed to 1456851fd730SQu Wenruo * userspace, the id pool is limited to 1M 1457851fd730SQu Wenruo */ 1458851fd730SQu Wenruo if (is_fstree(root->root_key.objectid) && 1459851fd730SQu Wenruo btrfs_root_refs(&root->root_item) > 0) { 14602dfb1e43SQu Wenruo if (!anon_dev) { 1461cb517eabSMiao Xie ret = get_anon_bdev(&root->anon_dev); 1462cb517eabSMiao Xie if (ret) 1463876d2cf1SLiu Bo goto fail; 14642dfb1e43SQu Wenruo } else { 14652dfb1e43SQu Wenruo root->anon_dev = anon_dev; 14662dfb1e43SQu Wenruo } 1467851fd730SQu Wenruo } 1468f32e48e9SChandan Rajendra 1469f32e48e9SChandan Rajendra mutex_lock(&root->objectid_mutex); 1470453e4873SNikolay Borisov ret = btrfs_init_root_free_objectid(root); 1471f32e48e9SChandan Rajendra if (ret) { 1472f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1473876d2cf1SLiu Bo goto fail; 1474f32e48e9SChandan Rajendra } 1475f32e48e9SChandan Rajendra 14766b8fad57SNikolay Borisov ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID); 1477f32e48e9SChandan Rajendra 1478f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1479f32e48e9SChandan Rajendra 1480cb517eabSMiao Xie return 0; 1481cb517eabSMiao Xie fail: 148284db5ccfSDavid Sterba /* The caller is responsible to call btrfs_free_fs_root */ 1483cb517eabSMiao Xie return ret; 1484cb517eabSMiao Xie } 1485cb517eabSMiao Xie 1486a98db0f3SJosef Bacik static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1487cb517eabSMiao Xie u64 root_id) 1488cb517eabSMiao Xie { 1489cb517eabSMiao Xie struct btrfs_root *root; 1490cb517eabSMiao Xie 1491fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1492fc7cbcd4SDavid Sterba root = radix_tree_lookup(&fs_info->fs_roots_radix, 1493fc7cbcd4SDavid Sterba (unsigned long)root_id); 1494bc44d7c4SJosef Bacik if (root) 149500246528SJosef Bacik root = btrfs_grab_root(root); 1496fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 1497cb517eabSMiao Xie return root; 1498cb517eabSMiao Xie } 1499cb517eabSMiao Xie 150049d11beaSJosef Bacik static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info, 150149d11beaSJosef Bacik u64 objectid) 150249d11beaSJosef Bacik { 1503abed4aaaSJosef Bacik struct btrfs_key key = { 1504abed4aaaSJosef Bacik .objectid = objectid, 1505abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 1506abed4aaaSJosef Bacik .offset = 0, 1507abed4aaaSJosef Bacik }; 1508abed4aaaSJosef Bacik 150949d11beaSJosef Bacik if (objectid == BTRFS_ROOT_TREE_OBJECTID) 151049d11beaSJosef Bacik return btrfs_grab_root(fs_info->tree_root); 151149d11beaSJosef Bacik if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 1512abed4aaaSJosef Bacik return btrfs_grab_root(btrfs_global_root(fs_info, &key)); 151349d11beaSJosef Bacik if (objectid == BTRFS_CHUNK_TREE_OBJECTID) 151449d11beaSJosef Bacik return btrfs_grab_root(fs_info->chunk_root); 151549d11beaSJosef Bacik if (objectid == BTRFS_DEV_TREE_OBJECTID) 151649d11beaSJosef Bacik return btrfs_grab_root(fs_info->dev_root); 151749d11beaSJosef Bacik if (objectid == BTRFS_CSUM_TREE_OBJECTID) 1518abed4aaaSJosef Bacik return btrfs_grab_root(btrfs_global_root(fs_info, &key)); 151949d11beaSJosef Bacik if (objectid == BTRFS_QUOTA_TREE_OBJECTID) 152049d11beaSJosef Bacik return btrfs_grab_root(fs_info->quota_root) ? 152149d11beaSJosef Bacik fs_info->quota_root : ERR_PTR(-ENOENT); 152249d11beaSJosef Bacik if (objectid == BTRFS_UUID_TREE_OBJECTID) 152349d11beaSJosef Bacik return btrfs_grab_root(fs_info->uuid_root) ? 152449d11beaSJosef Bacik fs_info->uuid_root : ERR_PTR(-ENOENT); 152514033b08SQu Wenruo if (objectid == BTRFS_BLOCK_GROUP_TREE_OBJECTID) 152614033b08SQu Wenruo return btrfs_grab_root(fs_info->block_group_root) ? 152714033b08SQu Wenruo fs_info->block_group_root : ERR_PTR(-ENOENT); 1528abed4aaaSJosef Bacik if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) { 1529abed4aaaSJosef Bacik struct btrfs_root *root = btrfs_global_root(fs_info, &key); 1530abed4aaaSJosef Bacik 1531abed4aaaSJosef Bacik return btrfs_grab_root(root) ? root : ERR_PTR(-ENOENT); 1532abed4aaaSJosef Bacik } 153349d11beaSJosef Bacik return NULL; 153449d11beaSJosef Bacik } 153549d11beaSJosef Bacik 1536cb517eabSMiao Xie int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1537cb517eabSMiao Xie struct btrfs_root *root) 1538cb517eabSMiao Xie { 1539cb517eabSMiao Xie int ret; 1540cb517eabSMiao Xie 1541fc7cbcd4SDavid Sterba ret = radix_tree_preload(GFP_NOFS); 1542fc7cbcd4SDavid Sterba if (ret) 1543fc7cbcd4SDavid Sterba return ret; 1544fc7cbcd4SDavid Sterba 1545fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1546fc7cbcd4SDavid Sterba ret = radix_tree_insert(&fs_info->fs_roots_radix, 1547fc7cbcd4SDavid Sterba (unsigned long)root->root_key.objectid, 1548fc7cbcd4SDavid Sterba root); 1549af01d2e5SJosef Bacik if (ret == 0) { 155000246528SJosef Bacik btrfs_grab_root(root); 1551fc7cbcd4SDavid Sterba set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1552af01d2e5SJosef Bacik } 1553fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 1554fc7cbcd4SDavid Sterba radix_tree_preload_end(); 1555cb517eabSMiao Xie 1556cb517eabSMiao Xie return ret; 1557cb517eabSMiao Xie } 1558cb517eabSMiao Xie 1559bd647ce3SJosef Bacik void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info) 1560bd647ce3SJosef Bacik { 1561bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 1562bd647ce3SJosef Bacik struct btrfs_root *root; 1563bd647ce3SJosef Bacik 1564bd647ce3SJosef Bacik while (!list_empty(&fs_info->allocated_roots)) { 1565457f1864SJosef Bacik char buf[BTRFS_ROOT_NAME_BUF_LEN]; 1566457f1864SJosef Bacik 1567bd647ce3SJosef Bacik root = list_first_entry(&fs_info->allocated_roots, 1568bd647ce3SJosef Bacik struct btrfs_root, leak_list); 1569457f1864SJosef Bacik btrfs_err(fs_info, "leaked root %s refcount %d", 157071008734SJosef Bacik btrfs_root_name(&root->root_key, buf), 1571bd647ce3SJosef Bacik refcount_read(&root->refs)); 1572bd647ce3SJosef Bacik while (refcount_read(&root->refs) > 1) 157300246528SJosef Bacik btrfs_put_root(root); 157400246528SJosef Bacik btrfs_put_root(root); 1575bd647ce3SJosef Bacik } 1576bd647ce3SJosef Bacik #endif 1577bd647ce3SJosef Bacik } 1578bd647ce3SJosef Bacik 1579abed4aaaSJosef Bacik static void free_global_roots(struct btrfs_fs_info *fs_info) 1580abed4aaaSJosef Bacik { 1581abed4aaaSJosef Bacik struct btrfs_root *root; 1582abed4aaaSJosef Bacik struct rb_node *node; 1583abed4aaaSJosef Bacik 1584abed4aaaSJosef Bacik while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) { 1585abed4aaaSJosef Bacik root = rb_entry(node, struct btrfs_root, rb_node); 1586abed4aaaSJosef Bacik rb_erase(&root->rb_node, &fs_info->global_root_tree); 1587abed4aaaSJosef Bacik btrfs_put_root(root); 1588abed4aaaSJosef Bacik } 1589abed4aaaSJosef Bacik } 1590abed4aaaSJosef Bacik 15910d4b0463SJosef Bacik void btrfs_free_fs_info(struct btrfs_fs_info *fs_info) 15920d4b0463SJosef Bacik { 1593141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 1594141386e1SJosef Bacik percpu_counter_destroy(&fs_info->delalloc_bytes); 15955deb17e1SJosef Bacik percpu_counter_destroy(&fs_info->ordered_bytes); 1596141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 1597141386e1SJosef Bacik btrfs_free_csum_hash(fs_info); 1598141386e1SJosef Bacik btrfs_free_stripe_hash_table(fs_info); 1599141386e1SJosef Bacik btrfs_free_ref_cache(fs_info); 16000d4b0463SJosef Bacik kfree(fs_info->balance_ctl); 16010d4b0463SJosef Bacik kfree(fs_info->delayed_root); 1602abed4aaaSJosef Bacik free_global_roots(fs_info); 160300246528SJosef Bacik btrfs_put_root(fs_info->tree_root); 160400246528SJosef Bacik btrfs_put_root(fs_info->chunk_root); 160500246528SJosef Bacik btrfs_put_root(fs_info->dev_root); 160600246528SJosef Bacik btrfs_put_root(fs_info->quota_root); 160700246528SJosef Bacik btrfs_put_root(fs_info->uuid_root); 160800246528SJosef Bacik btrfs_put_root(fs_info->fs_root); 1609aeb935a4SQu Wenruo btrfs_put_root(fs_info->data_reloc_root); 16109c54e80dSJosef Bacik btrfs_put_root(fs_info->block_group_root); 1611bd647ce3SJosef Bacik btrfs_check_leaked_roots(fs_info); 16123fd63727SJosef Bacik btrfs_extent_buffer_leak_debug_check(fs_info); 16130d4b0463SJosef Bacik kfree(fs_info->super_copy); 16140d4b0463SJosef Bacik kfree(fs_info->super_for_commit); 16158481dd80SQu Wenruo kfree(fs_info->subpage_info); 16160d4b0463SJosef Bacik kvfree(fs_info); 16170d4b0463SJosef Bacik } 16180d4b0463SJosef Bacik 16190d4b0463SJosef Bacik 16202dfb1e43SQu Wenruo /* 16212dfb1e43SQu Wenruo * Get an in-memory reference of a root structure. 16222dfb1e43SQu Wenruo * 16232dfb1e43SQu Wenruo * For essential trees like root/extent tree, we grab it from fs_info directly. 16242dfb1e43SQu Wenruo * For subvolume trees, we check the cached filesystem roots first. If not 16252dfb1e43SQu Wenruo * found, then read it from disk and add it to cached fs roots. 16262dfb1e43SQu Wenruo * 16272dfb1e43SQu Wenruo * Caller should release the root by calling btrfs_put_root() after the usage. 16282dfb1e43SQu Wenruo * 16292dfb1e43SQu Wenruo * NOTE: Reloc and log trees can't be read by this function as they share the 16302dfb1e43SQu Wenruo * same root objectid. 16312dfb1e43SQu Wenruo * 16322dfb1e43SQu Wenruo * @objectid: root id 16332dfb1e43SQu Wenruo * @anon_dev: preallocated anonymous block device number for new roots, 16342dfb1e43SQu Wenruo * pass 0 for new allocation. 16352dfb1e43SQu Wenruo * @check_ref: whether to check root item references, If true, return -ENOENT 16362dfb1e43SQu Wenruo * for orphan roots 16372dfb1e43SQu Wenruo */ 16382dfb1e43SQu Wenruo static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info, 16392dfb1e43SQu Wenruo u64 objectid, dev_t anon_dev, 16402dfb1e43SQu Wenruo bool check_ref) 16415eda7b5eSChris Mason { 16425eda7b5eSChris Mason struct btrfs_root *root; 1643381cf658SDavid Sterba struct btrfs_path *path; 16441d4c08e0SDavid Sterba struct btrfs_key key; 16455eda7b5eSChris Mason int ret; 16465eda7b5eSChris Mason 164749d11beaSJosef Bacik root = btrfs_get_global_root(fs_info, objectid); 164849d11beaSJosef Bacik if (root) 164949d11beaSJosef Bacik return root; 16504df27c4dSYan, Zheng again: 165156e9357aSDavid Sterba root = btrfs_lookup_fs_root(fs_info, objectid); 165248475471SStefan Behrens if (root) { 16532dfb1e43SQu Wenruo /* Shouldn't get preallocated anon_dev for cached roots */ 16542dfb1e43SQu Wenruo ASSERT(!anon_dev); 1655bc44d7c4SJosef Bacik if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 165600246528SJosef Bacik btrfs_put_root(root); 165748475471SStefan Behrens return ERR_PTR(-ENOENT); 1658bc44d7c4SJosef Bacik } 16595eda7b5eSChris Mason return root; 166048475471SStefan Behrens } 16615eda7b5eSChris Mason 166256e9357aSDavid Sterba key.objectid = objectid; 166356e9357aSDavid Sterba key.type = BTRFS_ROOT_ITEM_KEY; 166456e9357aSDavid Sterba key.offset = (u64)-1; 166556e9357aSDavid Sterba root = btrfs_read_tree_root(fs_info->tree_root, &key); 16665eda7b5eSChris Mason if (IS_ERR(root)) 16675eda7b5eSChris Mason return root; 16683394e160SChris Mason 1669c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1670d68fc57bSYan, Zheng ret = -ENOENT; 1671d68fc57bSYan, Zheng goto fail; 1672d68fc57bSYan, Zheng } 1673d68fc57bSYan, Zheng 16742dfb1e43SQu Wenruo ret = btrfs_init_fs_root(root, anon_dev); 1675cb517eabSMiao Xie if (ret) 1676cb517eabSMiao Xie goto fail; 1677cb517eabSMiao Xie 1678381cf658SDavid Sterba path = btrfs_alloc_path(); 1679381cf658SDavid Sterba if (!path) { 1680381cf658SDavid Sterba ret = -ENOMEM; 1681381cf658SDavid Sterba goto fail; 1682381cf658SDavid Sterba } 16831d4c08e0SDavid Sterba key.objectid = BTRFS_ORPHAN_OBJECTID; 16841d4c08e0SDavid Sterba key.type = BTRFS_ORPHAN_ITEM_KEY; 168556e9357aSDavid Sterba key.offset = objectid; 16861d4c08e0SDavid Sterba 16871d4c08e0SDavid Sterba ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1688381cf658SDavid Sterba btrfs_free_path(path); 1689d68fc57bSYan, Zheng if (ret < 0) 1690d68fc57bSYan, Zheng goto fail; 1691d68fc57bSYan, Zheng if (ret == 0) 169227cdeb70SMiao Xie set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1693d68fc57bSYan, Zheng 1694cb517eabSMiao Xie ret = btrfs_insert_fs_root(fs_info, root); 16950f7d52f4SChris Mason if (ret) { 1696168a2f77SJia-Ju Bai if (ret == -EEXIST) { 169700246528SJosef Bacik btrfs_put_root(root); 16984df27c4dSYan, Zheng goto again; 1699168a2f77SJia-Ju Bai } 17004df27c4dSYan, Zheng goto fail; 17014df27c4dSYan, Zheng } 1702edbd8d4eSChris Mason return root; 17034df27c4dSYan, Zheng fail: 170433fab972SFilipe Manana /* 170533fab972SFilipe Manana * If our caller provided us an anonymous device, then it's his 1706143823cfSDavid Sterba * responsibility to free it in case we fail. So we have to set our 170733fab972SFilipe Manana * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root() 170833fab972SFilipe Manana * and once again by our caller. 170933fab972SFilipe Manana */ 171033fab972SFilipe Manana if (anon_dev) 171133fab972SFilipe Manana root->anon_dev = 0; 17128c38938cSJosef Bacik btrfs_put_root(root); 17134df27c4dSYan, Zheng return ERR_PTR(ret); 1714edbd8d4eSChris Mason } 1715edbd8d4eSChris Mason 17162dfb1e43SQu Wenruo /* 17172dfb1e43SQu Wenruo * Get in-memory reference of a root structure 17182dfb1e43SQu Wenruo * 17192dfb1e43SQu Wenruo * @objectid: tree objectid 17202dfb1e43SQu Wenruo * @check_ref: if set, verify that the tree exists and the item has at least 17212dfb1e43SQu Wenruo * one reference 17222dfb1e43SQu Wenruo */ 17232dfb1e43SQu Wenruo struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 17242dfb1e43SQu Wenruo u64 objectid, bool check_ref) 17252dfb1e43SQu Wenruo { 17262dfb1e43SQu Wenruo return btrfs_get_root_ref(fs_info, objectid, 0, check_ref); 17272dfb1e43SQu Wenruo } 17282dfb1e43SQu Wenruo 17292dfb1e43SQu Wenruo /* 17302dfb1e43SQu Wenruo * Get in-memory reference of a root structure, created as new, optionally pass 17312dfb1e43SQu Wenruo * the anonymous block device id 17322dfb1e43SQu Wenruo * 17332dfb1e43SQu Wenruo * @objectid: tree objectid 17342dfb1e43SQu Wenruo * @anon_dev: if zero, allocate a new anonymous block device or use the 17352dfb1e43SQu Wenruo * parameter value 17362dfb1e43SQu Wenruo */ 17372dfb1e43SQu Wenruo struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info, 17382dfb1e43SQu Wenruo u64 objectid, dev_t anon_dev) 17392dfb1e43SQu Wenruo { 17402dfb1e43SQu Wenruo return btrfs_get_root_ref(fs_info, objectid, anon_dev, true); 17412dfb1e43SQu Wenruo } 17422dfb1e43SQu Wenruo 17438b712842SChris Mason /* 174449d11beaSJosef Bacik * btrfs_get_fs_root_commit_root - return a root for the given objectid 174549d11beaSJosef Bacik * @fs_info: the fs_info 174649d11beaSJosef Bacik * @objectid: the objectid we need to lookup 174749d11beaSJosef Bacik * 174849d11beaSJosef Bacik * This is exclusively used for backref walking, and exists specifically because 174949d11beaSJosef Bacik * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref 175049d11beaSJosef Bacik * creation time, which means we may have to read the tree_root in order to look 175149d11beaSJosef Bacik * up a fs root that is not in memory. If the root is not in memory we will 175249d11beaSJosef Bacik * read the tree root commit root and look up the fs root from there. This is a 175349d11beaSJosef Bacik * temporary root, it will not be inserted into the radix tree as it doesn't 175449d11beaSJosef Bacik * have the most uptodate information, it'll simply be discarded once the 175549d11beaSJosef Bacik * backref code is finished using the root. 175649d11beaSJosef Bacik */ 175749d11beaSJosef Bacik struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info, 175849d11beaSJosef Bacik struct btrfs_path *path, 175949d11beaSJosef Bacik u64 objectid) 176049d11beaSJosef Bacik { 176149d11beaSJosef Bacik struct btrfs_root *root; 176249d11beaSJosef Bacik struct btrfs_key key; 176349d11beaSJosef Bacik 176449d11beaSJosef Bacik ASSERT(path->search_commit_root && path->skip_locking); 176549d11beaSJosef Bacik 176649d11beaSJosef Bacik /* 176749d11beaSJosef Bacik * This can return -ENOENT if we ask for a root that doesn't exist, but 176849d11beaSJosef Bacik * since this is called via the backref walking code we won't be looking 176949d11beaSJosef Bacik * up a root that doesn't exist, unless there's corruption. So if root 177049d11beaSJosef Bacik * != NULL just return it. 177149d11beaSJosef Bacik */ 177249d11beaSJosef Bacik root = btrfs_get_global_root(fs_info, objectid); 177349d11beaSJosef Bacik if (root) 177449d11beaSJosef Bacik return root; 177549d11beaSJosef Bacik 177649d11beaSJosef Bacik root = btrfs_lookup_fs_root(fs_info, objectid); 177749d11beaSJosef Bacik if (root) 177849d11beaSJosef Bacik return root; 177949d11beaSJosef Bacik 178049d11beaSJosef Bacik key.objectid = objectid; 178149d11beaSJosef Bacik key.type = BTRFS_ROOT_ITEM_KEY; 178249d11beaSJosef Bacik key.offset = (u64)-1; 178349d11beaSJosef Bacik root = read_tree_root_path(fs_info->tree_root, path, &key); 178449d11beaSJosef Bacik btrfs_release_path(path); 178549d11beaSJosef Bacik 178649d11beaSJosef Bacik return root; 178749d11beaSJosef Bacik } 178849d11beaSJosef Bacik 1789a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1790a74a4b97SChris Mason { 17910d031dc4SYu Zhe struct btrfs_fs_info *fs_info = arg; 1792d0278245SMiao Xie int again; 1793a74a4b97SChris Mason 1794d6fd0ae2SOmar Sandoval while (1) { 1795d0278245SMiao Xie again = 0; 17969d1a2a3aSDavid Sterba 1797fd340d0fSJosef Bacik set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1798fd340d0fSJosef Bacik 1799d0278245SMiao Xie /* Make the cleaner go to sleep early. */ 18002ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) 1801d0278245SMiao Xie goto sleep; 1802d0278245SMiao Xie 180390c711abSZygo Blaxell /* 180490c711abSZygo Blaxell * Do not do anything if we might cause open_ctree() to block 180590c711abSZygo Blaxell * before we have finished mounting the filesystem. 180690c711abSZygo Blaxell */ 18070b246afaSJeff Mahoney if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 180890c711abSZygo Blaxell goto sleep; 180990c711abSZygo Blaxell 18100b246afaSJeff Mahoney if (!mutex_trylock(&fs_info->cleaner_mutex)) 1811d0278245SMiao Xie goto sleep; 1812d0278245SMiao Xie 1813dc7f370cSMiao Xie /* 1814dc7f370cSMiao Xie * Avoid the problem that we change the status of the fs 1815dc7f370cSMiao Xie * during the above check and trylock. 1816dc7f370cSMiao Xie */ 18172ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) { 18180b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1819dc7f370cSMiao Xie goto sleep; 1820dc7f370cSMiao Xie } 1821dc7f370cSMiao Xie 18222ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 1823c2d6cb16SFilipe Manana 182433c44184SJosef Bacik again = btrfs_clean_one_deleted_snapshot(fs_info); 18250b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1826a74a4b97SChris Mason 1827d0278245SMiao Xie /* 182805323cd1SMiao Xie * The defragger has dealt with the R/O remount and umount, 182905323cd1SMiao Xie * needn't do anything special here. 1830d0278245SMiao Xie */ 18310b246afaSJeff Mahoney btrfs_run_defrag_inodes(fs_info); 183267c5e7d4SFilipe Manana 183367c5e7d4SFilipe Manana /* 1834f3372065SJohannes Thumshirn * Acquires fs_info->reclaim_bgs_lock to avoid racing 183567c5e7d4SFilipe Manana * with relocation (btrfs_relocate_chunk) and relocation 183667c5e7d4SFilipe Manana * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 1837f3372065SJohannes Thumshirn * after acquiring fs_info->reclaim_bgs_lock. So we 183867c5e7d4SFilipe Manana * can't hold, nor need to, fs_info->cleaner_mutex when deleting 183967c5e7d4SFilipe Manana * unused block groups. 184067c5e7d4SFilipe Manana */ 18410b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 184218bb8bbfSJohannes Thumshirn 184318bb8bbfSJohannes Thumshirn /* 184418bb8bbfSJohannes Thumshirn * Reclaim block groups in the reclaim_bgs list after we deleted 184518bb8bbfSJohannes Thumshirn * all unused block_groups. This possibly gives us some more free 184618bb8bbfSJohannes Thumshirn * space. 184718bb8bbfSJohannes Thumshirn */ 184818bb8bbfSJohannes Thumshirn btrfs_reclaim_bgs(fs_info); 1849d0278245SMiao Xie sleep: 1850a0a1db70SFilipe Manana clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1851d6fd0ae2SOmar Sandoval if (kthread_should_park()) 1852d6fd0ae2SOmar Sandoval kthread_parkme(); 1853d6fd0ae2SOmar Sandoval if (kthread_should_stop()) 1854d6fd0ae2SOmar Sandoval return 0; 1855838fe188SJiri Kosina if (!again) { 1856a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 1857a74a4b97SChris Mason schedule(); 1858a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1859a74a4b97SChris Mason } 1860da288d28SFilipe Manana } 1861a74a4b97SChris Mason } 1862a74a4b97SChris Mason 1863a74a4b97SChris Mason static int transaction_kthread(void *arg) 1864a74a4b97SChris Mason { 1865a74a4b97SChris Mason struct btrfs_root *root = arg; 18660b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1867a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1868a74a4b97SChris Mason struct btrfs_transaction *cur; 18698929ecfaSYan, Zheng u64 transid; 1870643900beSNikolay Borisov time64_t delta; 1871a74a4b97SChris Mason unsigned long delay; 1872914b2007SJan Kara bool cannot_commit; 1873a74a4b97SChris Mason 1874a74a4b97SChris Mason do { 1875914b2007SJan Kara cannot_commit = false; 1876ba1bc00fSNikolay Borisov delay = msecs_to_jiffies(fs_info->commit_interval * 1000); 18770b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 1878a74a4b97SChris Mason 18790b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 18800b246afaSJeff Mahoney cur = fs_info->running_transaction; 1881a74a4b97SChris Mason if (!cur) { 18820b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1883a74a4b97SChris Mason goto sleep; 1884a74a4b97SChris Mason } 188531153d81SYan Zheng 1886643900beSNikolay Borisov delta = ktime_get_seconds() - cur->start_time; 1887fdfbf020SJosef Bacik if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) && 1888fdfbf020SJosef Bacik cur->state < TRANS_STATE_COMMIT_START && 1889643900beSNikolay Borisov delta < fs_info->commit_interval) { 18900b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1891fb8a7e94SNikolay Borisov delay -= msecs_to_jiffies((delta - 1) * 1000); 1892fb8a7e94SNikolay Borisov delay = min(delay, 1893fb8a7e94SNikolay Borisov msecs_to_jiffies(fs_info->commit_interval * 1000)); 1894a74a4b97SChris Mason goto sleep; 1895a74a4b97SChris Mason } 18968929ecfaSYan, Zheng transid = cur->transid; 18970b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 189856bec294SChris Mason 189979787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 1900354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 1901914b2007SJan Kara if (IS_ERR(trans)) { 1902354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 1903914b2007SJan Kara cannot_commit = true; 190479787eaaSJeff Mahoney goto sleep; 1905914b2007SJan Kara } 19068929ecfaSYan, Zheng if (transid == trans->transid) { 19073a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 19088929ecfaSYan, Zheng } else { 19093a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19108929ecfaSYan, Zheng } 1911a74a4b97SChris Mason sleep: 19120b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 19130b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 1914a74a4b97SChris Mason 191584961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) 19162ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 19178929ecfaSYan, Zheng if (!kthread_should_stop() && 19180b246afaSJeff Mahoney (!btrfs_transaction_blocked(fs_info) || 1919914b2007SJan Kara cannot_commit)) 1920bc5511d0SNikolay Borisov schedule_timeout_interruptible(delay); 1921a74a4b97SChris Mason } while (!kthread_should_stop()); 1922a74a4b97SChris Mason return 0; 1923a74a4b97SChris Mason } 1924a74a4b97SChris Mason 1925af31f5e5SChris Mason /* 192601f0f9daSNikolay Borisov * This will find the highest generation in the array of root backups. The 192701f0f9daSNikolay Borisov * index of the highest array is returned, or -EINVAL if we can't find 192801f0f9daSNikolay Borisov * anything. 1929af31f5e5SChris Mason * 1930af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1931af31f5e5SChris Mason * generation of the latest root in the array with the generation 1932af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1933af31f5e5SChris Mason */ 193401f0f9daSNikolay Borisov static int find_newest_super_backup(struct btrfs_fs_info *info) 1935af31f5e5SChris Mason { 193601f0f9daSNikolay Borisov const u64 newest_gen = btrfs_super_generation(info->super_copy); 1937af31f5e5SChris Mason u64 cur; 1938af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1939af31f5e5SChris Mason int i; 1940af31f5e5SChris Mason 1941af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1942af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1943af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1944af31f5e5SChris Mason if (cur == newest_gen) 194501f0f9daSNikolay Borisov return i; 1946af31f5e5SChris Mason } 1947af31f5e5SChris Mason 194801f0f9daSNikolay Borisov return -EINVAL; 1949af31f5e5SChris Mason } 1950af31f5e5SChris Mason 1951af31f5e5SChris Mason /* 1952af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1953af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1954af31f5e5SChris Mason * done 1955af31f5e5SChris Mason */ 1956af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1957af31f5e5SChris Mason { 19586ef108ddSNikolay Borisov const int next_backup = info->backup_root_index; 1959af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1960af31f5e5SChris Mason 1961af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1962af31f5e5SChris Mason 1963af31f5e5SChris Mason /* 1964af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1965af31f5e5SChris Mason * regardless of which ones we use today 1966af31f5e5SChris Mason */ 1967af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1968af31f5e5SChris Mason 1969af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1970af31f5e5SChris Mason 1971af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1972af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1973af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1974af31f5e5SChris Mason 1975af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1976af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1977af31f5e5SChris Mason 1978af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1979af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1980af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 1981af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 1982af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 1983af31f5e5SChris Mason 19841c56ab99SQu Wenruo if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) { 19859c54e80dSJosef Bacik struct btrfs_root *extent_root = btrfs_extent_root(info, 0); 1986f7238e50SJosef Bacik struct btrfs_root *csum_root = btrfs_csum_root(info, 0); 19879c54e80dSJosef Bacik 19889c54e80dSJosef Bacik btrfs_set_backup_extent_root(root_backup, 19899c54e80dSJosef Bacik extent_root->node->start); 1990af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 199129cbcf40SJosef Bacik btrfs_header_generation(extent_root->node)); 1992af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 199329cbcf40SJosef Bacik btrfs_header_level(extent_root->node)); 1994af31f5e5SChris Mason 1995f7238e50SJosef Bacik btrfs_set_backup_csum_root(root_backup, csum_root->node->start); 1996f7238e50SJosef Bacik btrfs_set_backup_csum_root_gen(root_backup, 1997f7238e50SJosef Bacik btrfs_header_generation(csum_root->node)); 1998f7238e50SJosef Bacik btrfs_set_backup_csum_root_level(root_backup, 1999f7238e50SJosef Bacik btrfs_header_level(csum_root->node)); 20009c54e80dSJosef Bacik } 2001af31f5e5SChris Mason 20027c7e82a7SChris Mason /* 20037c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 20047c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 20057c7e82a7SChris Mason */ 20067c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 20077c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 20087c7e82a7SChris Mason info->fs_root->node->start); 2009af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 2010af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 2011af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 2012af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 20137c7e82a7SChris Mason } 2014af31f5e5SChris Mason 2015af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 2016af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 2017af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 2018af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 2019af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 2020af31f5e5SChris Mason 2021af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 2022af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 2023af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 2024af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 2025af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 2026af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 2027af31f5e5SChris Mason 2028af31f5e5SChris Mason /* 2029af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 2030af31f5e5SChris Mason * for the next commit 2031af31f5e5SChris Mason */ 2032af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 2033af31f5e5SChris Mason &info->super_for_commit->super_roots, 2034af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 2035af31f5e5SChris Mason } 2036af31f5e5SChris Mason 2037af31f5e5SChris Mason /* 2038bd2336b2SNikolay Borisov * read_backup_root - Reads a backup root based on the passed priority. Prio 0 2039bd2336b2SNikolay Borisov * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots 2040bd2336b2SNikolay Borisov * 2041bd2336b2SNikolay Borisov * fs_info - filesystem whose backup roots need to be read 2042bd2336b2SNikolay Borisov * priority - priority of backup root required 2043bd2336b2SNikolay Borisov * 2044bd2336b2SNikolay Borisov * Returns backup root index on success and -EINVAL otherwise. 2045bd2336b2SNikolay Borisov */ 2046bd2336b2SNikolay Borisov static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority) 2047bd2336b2SNikolay Borisov { 2048bd2336b2SNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 2049bd2336b2SNikolay Borisov struct btrfs_super_block *super = fs_info->super_copy; 2050bd2336b2SNikolay Borisov struct btrfs_root_backup *root_backup; 2051bd2336b2SNikolay Borisov 2052bd2336b2SNikolay Borisov if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) { 2053bd2336b2SNikolay Borisov if (priority == 0) 2054bd2336b2SNikolay Borisov return backup_index; 2055bd2336b2SNikolay Borisov 2056bd2336b2SNikolay Borisov backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority; 2057bd2336b2SNikolay Borisov backup_index %= BTRFS_NUM_BACKUP_ROOTS; 2058bd2336b2SNikolay Borisov } else { 2059bd2336b2SNikolay Borisov return -EINVAL; 2060bd2336b2SNikolay Borisov } 2061bd2336b2SNikolay Borisov 2062bd2336b2SNikolay Borisov root_backup = super->super_roots + backup_index; 2063bd2336b2SNikolay Borisov 2064bd2336b2SNikolay Borisov btrfs_set_super_generation(super, 2065bd2336b2SNikolay Borisov btrfs_backup_tree_root_gen(root_backup)); 2066bd2336b2SNikolay Borisov btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 2067bd2336b2SNikolay Borisov btrfs_set_super_root_level(super, 2068bd2336b2SNikolay Borisov btrfs_backup_tree_root_level(root_backup)); 2069bd2336b2SNikolay Borisov btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 2070bd2336b2SNikolay Borisov 2071bd2336b2SNikolay Borisov /* 2072bd2336b2SNikolay Borisov * Fixme: the total bytes and num_devices need to match or we should 2073bd2336b2SNikolay Borisov * need a fsck 2074bd2336b2SNikolay Borisov */ 2075bd2336b2SNikolay Borisov btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 2076bd2336b2SNikolay Borisov btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 2077bd2336b2SNikolay Borisov 2078bd2336b2SNikolay Borisov return backup_index; 2079bd2336b2SNikolay Borisov } 2080bd2336b2SNikolay Borisov 20817abadb64SLiu Bo /* helper to cleanup workers */ 20827abadb64SLiu Bo static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 20837abadb64SLiu Bo { 2084dc6e3209SQu Wenruo btrfs_destroy_workqueue(fs_info->fixup_workers); 2085afe3d242SQu Wenruo btrfs_destroy_workqueue(fs_info->delalloc_workers); 2086a31b4a43SChristoph Hellwig btrfs_destroy_workqueue(fs_info->hipri_workers); 20875cdc7ad3SQu Wenruo btrfs_destroy_workqueue(fs_info->workers); 2088d7b9416fSChristoph Hellwig if (fs_info->endio_workers) 2089d7b9416fSChristoph Hellwig destroy_workqueue(fs_info->endio_workers); 2090d34e123dSChristoph Hellwig if (fs_info->endio_raid56_workers) 2091d34e123dSChristoph Hellwig destroy_workqueue(fs_info->endio_raid56_workers); 2092385de0efSChristoph Hellwig if (fs_info->rmw_workers) 2093385de0efSChristoph Hellwig destroy_workqueue(fs_info->rmw_workers); 2094fed8a72dSChristoph Hellwig if (fs_info->compressed_write_workers) 2095fed8a72dSChristoph Hellwig destroy_workqueue(fs_info->compressed_write_workers); 2096fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_write_workers); 2097fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 20985b3bc44eSQu Wenruo btrfs_destroy_workqueue(fs_info->delayed_workers); 2099e66f0bb1SQu Wenruo btrfs_destroy_workqueue(fs_info->caching_workers); 2100a44903abSQu Wenruo btrfs_destroy_workqueue(fs_info->flush_workers); 2101fc97fab0SQu Wenruo btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2102b0643e59SDennis Zhou if (fs_info->discard_ctl.discard_workers) 2103b0643e59SDennis Zhou destroy_workqueue(fs_info->discard_ctl.discard_workers); 2104a9b9477dSFilipe Manana /* 2105a9b9477dSFilipe Manana * Now that all other work queues are destroyed, we can safely destroy 2106a9b9477dSFilipe Manana * the queues used for metadata I/O, since tasks from those other work 2107a9b9477dSFilipe Manana * queues can do metadata I/O operations. 2108a9b9477dSFilipe Manana */ 2109d7b9416fSChristoph Hellwig if (fs_info->endio_meta_workers) 2110d7b9416fSChristoph Hellwig destroy_workqueue(fs_info->endio_meta_workers); 21117abadb64SLiu Bo } 21127abadb64SLiu Bo 21132e9f5954SRashika static void free_root_extent_buffers(struct btrfs_root *root) 21142e9f5954SRashika { 21152e9f5954SRashika if (root) { 21162e9f5954SRashika free_extent_buffer(root->node); 21172e9f5954SRashika free_extent_buffer(root->commit_root); 21182e9f5954SRashika root->node = NULL; 21192e9f5954SRashika root->commit_root = NULL; 21202e9f5954SRashika } 21212e9f5954SRashika } 21222e9f5954SRashika 2123abed4aaaSJosef Bacik static void free_global_root_pointers(struct btrfs_fs_info *fs_info) 2124abed4aaaSJosef Bacik { 2125abed4aaaSJosef Bacik struct btrfs_root *root, *tmp; 2126abed4aaaSJosef Bacik 2127abed4aaaSJosef Bacik rbtree_postorder_for_each_entry_safe(root, tmp, 2128abed4aaaSJosef Bacik &fs_info->global_root_tree, 2129abed4aaaSJosef Bacik rb_node) 2130abed4aaaSJosef Bacik free_root_extent_buffers(root); 2131abed4aaaSJosef Bacik } 2132abed4aaaSJosef Bacik 2133af31f5e5SChris Mason /* helper to cleanup tree roots */ 21344273eaffSAnand Jain static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root) 2135af31f5e5SChris Mason { 21362e9f5954SRashika free_root_extent_buffers(info->tree_root); 2137655b09feSJosef Bacik 2138abed4aaaSJosef Bacik free_global_root_pointers(info); 21392e9f5954SRashika free_root_extent_buffers(info->dev_root); 21402e9f5954SRashika free_root_extent_buffers(info->quota_root); 21412e9f5954SRashika free_root_extent_buffers(info->uuid_root); 21428c38938cSJosef Bacik free_root_extent_buffers(info->fs_root); 2143aeb935a4SQu Wenruo free_root_extent_buffers(info->data_reloc_root); 21449c54e80dSJosef Bacik free_root_extent_buffers(info->block_group_root); 21454273eaffSAnand Jain if (free_chunk_root) 21462e9f5954SRashika free_root_extent_buffers(info->chunk_root); 2147af31f5e5SChris Mason } 2148af31f5e5SChris Mason 21498c38938cSJosef Bacik void btrfs_put_root(struct btrfs_root *root) 21508c38938cSJosef Bacik { 21518c38938cSJosef Bacik if (!root) 21528c38938cSJosef Bacik return; 21538c38938cSJosef Bacik 21548c38938cSJosef Bacik if (refcount_dec_and_test(&root->refs)) { 21558c38938cSJosef Bacik WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 21561dae7e0eSQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)); 21578c38938cSJosef Bacik if (root->anon_dev) 21588c38938cSJosef Bacik free_anon_bdev(root->anon_dev); 21598c38938cSJosef Bacik btrfs_drew_lock_destroy(&root->snapshot_lock); 2160923eb523SJohannes Thumshirn free_root_extent_buffers(root); 21618c38938cSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 2162fc7cbcd4SDavid Sterba spin_lock(&root->fs_info->fs_roots_radix_lock); 21638c38938cSJosef Bacik list_del_init(&root->leak_list); 2164fc7cbcd4SDavid Sterba spin_unlock(&root->fs_info->fs_roots_radix_lock); 21658c38938cSJosef Bacik #endif 21668c38938cSJosef Bacik kfree(root); 21678c38938cSJosef Bacik } 21688c38938cSJosef Bacik } 21698c38938cSJosef Bacik 2170faa2dbf0SJosef Bacik void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2171171f6537SJosef Bacik { 2172fc7cbcd4SDavid Sterba int ret; 2173fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 2174fc7cbcd4SDavid Sterba int i; 2175171f6537SJosef Bacik 2176171f6537SJosef Bacik while (!list_empty(&fs_info->dead_roots)) { 2177fc7cbcd4SDavid Sterba gang[0] = list_entry(fs_info->dead_roots.next, 2178171f6537SJosef Bacik struct btrfs_root, root_list); 2179fc7cbcd4SDavid Sterba list_del(&gang[0]->root_list); 2180171f6537SJosef Bacik 2181fc7cbcd4SDavid Sterba if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) 2182fc7cbcd4SDavid Sterba btrfs_drop_and_free_fs_root(fs_info, gang[0]); 2183fc7cbcd4SDavid Sterba btrfs_put_root(gang[0]); 2184171f6537SJosef Bacik } 2185171f6537SJosef Bacik 2186fc7cbcd4SDavid Sterba while (1) { 2187fc7cbcd4SDavid Sterba ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2188fc7cbcd4SDavid Sterba (void **)gang, 0, 2189fc7cbcd4SDavid Sterba ARRAY_SIZE(gang)); 2190fc7cbcd4SDavid Sterba if (!ret) 2191fc7cbcd4SDavid Sterba break; 2192fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) 2193fc7cbcd4SDavid Sterba btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2194171f6537SJosef Bacik } 2195171f6537SJosef Bacik } 2196af31f5e5SChris Mason 2197638aa7edSEric Sandeen static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2198638aa7edSEric Sandeen { 2199638aa7edSEric Sandeen mutex_init(&fs_info->scrub_lock); 2200638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_running, 0); 2201638aa7edSEric Sandeen atomic_set(&fs_info->scrub_pause_req, 0); 2202638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_paused, 0); 2203638aa7edSEric Sandeen atomic_set(&fs_info->scrub_cancel_req, 0); 2204638aa7edSEric Sandeen init_waitqueue_head(&fs_info->scrub_pause_wait); 2205ff09c4caSAnand Jain refcount_set(&fs_info->scrub_workers_refcnt, 0); 2206638aa7edSEric Sandeen } 2207638aa7edSEric Sandeen 2208779a65a4SEric Sandeen static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2209779a65a4SEric Sandeen { 2210779a65a4SEric Sandeen spin_lock_init(&fs_info->balance_lock); 2211779a65a4SEric Sandeen mutex_init(&fs_info->balance_mutex); 2212779a65a4SEric Sandeen atomic_set(&fs_info->balance_pause_req, 0); 2213779a65a4SEric Sandeen atomic_set(&fs_info->balance_cancel_req, 0); 2214779a65a4SEric Sandeen fs_info->balance_ctl = NULL; 2215779a65a4SEric Sandeen init_waitqueue_head(&fs_info->balance_wait_q); 2216907d2710SDavid Sterba atomic_set(&fs_info->reloc_cancel_req, 0); 2217779a65a4SEric Sandeen } 2218779a65a4SEric Sandeen 22196bccf3abSJeff Mahoney static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info) 2220f37938e0SEric Sandeen { 22212ff7e61eSJeff Mahoney struct inode *inode = fs_info->btree_inode; 22222ff7e61eSJeff Mahoney 22232ff7e61eSJeff Mahoney inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 22242ff7e61eSJeff Mahoney set_nlink(inode, 1); 2225f37938e0SEric Sandeen /* 2226f37938e0SEric Sandeen * we set the i_size on the btree inode to the max possible int. 2227f37938e0SEric Sandeen * the real end of the address space is determined by all of 2228f37938e0SEric Sandeen * the devices in the system 2229f37938e0SEric Sandeen */ 22302ff7e61eSJeff Mahoney inode->i_size = OFFSET_MAX; 22312ff7e61eSJeff Mahoney inode->i_mapping->a_ops = &btree_aops; 2232f37938e0SEric Sandeen 22332ff7e61eSJeff Mahoney RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 223443eb5f29SQu Wenruo extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree, 22352c53a14dSQu Wenruo IO_TREE_BTREE_INODE_IO, inode); 22367b439738SDavid Sterba BTRFS_I(inode)->io_tree.track_uptodate = false; 22372ff7e61eSJeff Mahoney extent_map_tree_init(&BTRFS_I(inode)->extent_tree); 2238f37938e0SEric Sandeen 22395c8fd99fSJosef Bacik BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root); 2240adac5584SFilipe Manana BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID; 2241adac5584SFilipe Manana BTRFS_I(inode)->location.type = 0; 2242adac5584SFilipe Manana BTRFS_I(inode)->location.offset = 0; 22432ff7e61eSJeff Mahoney set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); 22442ff7e61eSJeff Mahoney btrfs_insert_inode_hash(inode); 2245f37938e0SEric Sandeen } 2246f37938e0SEric Sandeen 2247ad618368SEric Sandeen static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2248ad618368SEric Sandeen { 2249ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2250129827e3SDavid Sterba init_rwsem(&fs_info->dev_replace.rwsem); 22517f8d236aSDavid Sterba init_waitqueue_head(&fs_info->dev_replace.replace_wait); 2252ad618368SEric Sandeen } 2253ad618368SEric Sandeen 2254f9e92e40SEric Sandeen static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2255f9e92e40SEric Sandeen { 2256f9e92e40SEric Sandeen spin_lock_init(&fs_info->qgroup_lock); 2257f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_ioctl_lock); 2258f9e92e40SEric Sandeen fs_info->qgroup_tree = RB_ROOT; 2259f9e92e40SEric Sandeen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2260f9e92e40SEric Sandeen fs_info->qgroup_seq = 1; 2261f9e92e40SEric Sandeen fs_info->qgroup_ulist = NULL; 2262d2c609b8SJeff Mahoney fs_info->qgroup_rescan_running = false; 2263*011b46c3SQu Wenruo fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL; 2264f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_rescan_lock); 2265f9e92e40SEric Sandeen } 2266f9e92e40SEric Sandeen 2267d21deec5SSu Yue static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info) 22682a458198SEric Sandeen { 2269f7b885beSAnand Jain u32 max_active = fs_info->thread_pool_size; 22706f011058SDavid Sterba unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 22712a458198SEric Sandeen 22722a458198SEric Sandeen fs_info->workers = 2273a31b4a43SChristoph Hellwig btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16); 2274a31b4a43SChristoph Hellwig fs_info->hipri_workers = 2275a31b4a43SChristoph Hellwig btrfs_alloc_workqueue(fs_info, "worker-high", 2276cb001095SJeff Mahoney flags | WQ_HIGHPRI, max_active, 16); 22772a458198SEric Sandeen 22782a458198SEric Sandeen fs_info->delalloc_workers = 2279cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delalloc", 2280cb001095SJeff Mahoney flags, max_active, 2); 22812a458198SEric Sandeen 22822a458198SEric Sandeen fs_info->flush_workers = 2283cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "flush_delalloc", 2284cb001095SJeff Mahoney flags, max_active, 0); 22852a458198SEric Sandeen 22862a458198SEric Sandeen fs_info->caching_workers = 2287cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); 22882a458198SEric Sandeen 22892a458198SEric Sandeen fs_info->fixup_workers = 2290cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); 22912a458198SEric Sandeen 22922a458198SEric Sandeen fs_info->endio_workers = 2293d7b9416fSChristoph Hellwig alloc_workqueue("btrfs-endio", flags, max_active); 22942a458198SEric Sandeen fs_info->endio_meta_workers = 2295d7b9416fSChristoph Hellwig alloc_workqueue("btrfs-endio-meta", flags, max_active); 22962a458198SEric Sandeen fs_info->endio_raid56_workers = 2297d34e123dSChristoph Hellwig alloc_workqueue("btrfs-endio-raid56", flags, max_active); 2298385de0efSChristoph Hellwig fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active); 22992a458198SEric Sandeen fs_info->endio_write_workers = 2300cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-write", flags, 2301cb001095SJeff Mahoney max_active, 2); 2302fed8a72dSChristoph Hellwig fs_info->compressed_write_workers = 2303fed8a72dSChristoph Hellwig alloc_workqueue("btrfs-compressed-write", flags, max_active); 23042a458198SEric Sandeen fs_info->endio_freespace_worker = 2305cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "freespace-write", flags, 2306cb001095SJeff Mahoney max_active, 0); 23072a458198SEric Sandeen fs_info->delayed_workers = 2308cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, 2309cb001095SJeff Mahoney max_active, 0); 23102a458198SEric Sandeen fs_info->qgroup_rescan_workers = 2311cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); 2312b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers = 2313b0643e59SDennis Zhou alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1); 23142a458198SEric Sandeen 2315a31b4a43SChristoph Hellwig if (!(fs_info->workers && fs_info->hipri_workers && 2316a31b4a43SChristoph Hellwig fs_info->delalloc_workers && fs_info->flush_workers && 23172a458198SEric Sandeen fs_info->endio_workers && fs_info->endio_meta_workers && 2318fed8a72dSChristoph Hellwig fs_info->compressed_write_workers && 23192a458198SEric Sandeen fs_info->endio_write_workers && fs_info->endio_raid56_workers && 23202a458198SEric Sandeen fs_info->endio_freespace_worker && fs_info->rmw_workers && 2321f26c9238SQu Wenruo fs_info->caching_workers && fs_info->fixup_workers && 2322f26c9238SQu Wenruo fs_info->delayed_workers && fs_info->qgroup_rescan_workers && 2323b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers)) { 23242a458198SEric Sandeen return -ENOMEM; 23252a458198SEric Sandeen } 23262a458198SEric Sandeen 23272a458198SEric Sandeen return 0; 23282a458198SEric Sandeen } 23292a458198SEric Sandeen 23306d97c6e3SJohannes Thumshirn static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type) 23316d97c6e3SJohannes Thumshirn { 23326d97c6e3SJohannes Thumshirn struct crypto_shash *csum_shash; 2333b4e967beSDavid Sterba const char *csum_driver = btrfs_super_csum_driver(csum_type); 23346d97c6e3SJohannes Thumshirn 2335b4e967beSDavid Sterba csum_shash = crypto_alloc_shash(csum_driver, 0, 0); 23366d97c6e3SJohannes Thumshirn 23376d97c6e3SJohannes Thumshirn if (IS_ERR(csum_shash)) { 23386d97c6e3SJohannes Thumshirn btrfs_err(fs_info, "error allocating %s hash for checksum", 2339b4e967beSDavid Sterba csum_driver); 23406d97c6e3SJohannes Thumshirn return PTR_ERR(csum_shash); 23416d97c6e3SJohannes Thumshirn } 23426d97c6e3SJohannes Thumshirn 23436d97c6e3SJohannes Thumshirn fs_info->csum_shash = csum_shash; 23446d97c6e3SJohannes Thumshirn 2345c8a5f8caSDavid Sterba btrfs_info(fs_info, "using %s (%s) checksum algorithm", 2346c8a5f8caSDavid Sterba btrfs_super_csum_name(csum_type), 2347c8a5f8caSDavid Sterba crypto_shash_driver_name(csum_shash)); 23486d97c6e3SJohannes Thumshirn return 0; 23496d97c6e3SJohannes Thumshirn } 23506d97c6e3SJohannes Thumshirn 235163443bf5SEric Sandeen static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 235263443bf5SEric Sandeen struct btrfs_fs_devices *fs_devices) 235363443bf5SEric Sandeen { 235463443bf5SEric Sandeen int ret; 235563443bf5SEric Sandeen struct btrfs_root *log_tree_root; 235663443bf5SEric Sandeen struct btrfs_super_block *disk_super = fs_info->super_copy; 235763443bf5SEric Sandeen u64 bytenr = btrfs_super_log_root(disk_super); 2358581c1760SQu Wenruo int level = btrfs_super_log_root_level(disk_super); 235963443bf5SEric Sandeen 236063443bf5SEric Sandeen if (fs_devices->rw_devices == 0) { 2361f14d104dSDavid Sterba btrfs_warn(fs_info, "log replay required on RO media"); 236263443bf5SEric Sandeen return -EIO; 236363443bf5SEric Sandeen } 236463443bf5SEric Sandeen 236596dfcb46SJosef Bacik log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, 236696dfcb46SJosef Bacik GFP_KERNEL); 236763443bf5SEric Sandeen if (!log_tree_root) 236863443bf5SEric Sandeen return -ENOMEM; 236963443bf5SEric Sandeen 23702ff7e61eSJeff Mahoney log_tree_root->node = read_tree_block(fs_info, bytenr, 23711b7ec85eSJosef Bacik BTRFS_TREE_LOG_OBJECTID, 23721b7ec85eSJosef Bacik fs_info->generation + 1, level, 23731b7ec85eSJosef Bacik NULL); 237464c043deSLiu Bo if (IS_ERR(log_tree_root->node)) { 2375f14d104dSDavid Sterba btrfs_warn(fs_info, "failed to read log tree"); 23760eeff236SLiu Bo ret = PTR_ERR(log_tree_root->node); 23778c38938cSJosef Bacik log_tree_root->node = NULL; 237800246528SJosef Bacik btrfs_put_root(log_tree_root); 23790eeff236SLiu Bo return ret; 23804eb150d6SQu Wenruo } 23814eb150d6SQu Wenruo if (!extent_buffer_uptodate(log_tree_root->node)) { 2382f14d104dSDavid Sterba btrfs_err(fs_info, "failed to read log tree"); 238300246528SJosef Bacik btrfs_put_root(log_tree_root); 238463443bf5SEric Sandeen return -EIO; 238563443bf5SEric Sandeen } 23864eb150d6SQu Wenruo 238763443bf5SEric Sandeen /* returns with log_tree_root freed on success */ 238863443bf5SEric Sandeen ret = btrfs_recover_log_trees(log_tree_root); 238963443bf5SEric Sandeen if (ret) { 23900b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 239163443bf5SEric Sandeen "Failed to recover log tree"); 239200246528SJosef Bacik btrfs_put_root(log_tree_root); 239363443bf5SEric Sandeen return ret; 239463443bf5SEric Sandeen } 239563443bf5SEric Sandeen 2396bc98a42cSDavid Howells if (sb_rdonly(fs_info->sb)) { 23976bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 239863443bf5SEric Sandeen if (ret) 239963443bf5SEric Sandeen return ret; 240063443bf5SEric Sandeen } 240163443bf5SEric Sandeen 240263443bf5SEric Sandeen return 0; 240363443bf5SEric Sandeen } 240463443bf5SEric Sandeen 2405abed4aaaSJosef Bacik static int load_global_roots_objectid(struct btrfs_root *tree_root, 2406abed4aaaSJosef Bacik struct btrfs_path *path, u64 objectid, 2407abed4aaaSJosef Bacik const char *name) 2408abed4aaaSJosef Bacik { 2409abed4aaaSJosef Bacik struct btrfs_fs_info *fs_info = tree_root->fs_info; 2410abed4aaaSJosef Bacik struct btrfs_root *root; 2411f7238e50SJosef Bacik u64 max_global_id = 0; 2412abed4aaaSJosef Bacik int ret; 2413abed4aaaSJosef Bacik struct btrfs_key key = { 2414abed4aaaSJosef Bacik .objectid = objectid, 2415abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 2416abed4aaaSJosef Bacik .offset = 0, 2417abed4aaaSJosef Bacik }; 2418abed4aaaSJosef Bacik bool found = false; 2419abed4aaaSJosef Bacik 2420abed4aaaSJosef Bacik /* If we have IGNOREDATACSUMS skip loading these roots. */ 2421abed4aaaSJosef Bacik if (objectid == BTRFS_CSUM_TREE_OBJECTID && 2422abed4aaaSJosef Bacik btrfs_test_opt(fs_info, IGNOREDATACSUMS)) { 2423abed4aaaSJosef Bacik set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state); 2424abed4aaaSJosef Bacik return 0; 2425abed4aaaSJosef Bacik } 2426abed4aaaSJosef Bacik 2427abed4aaaSJosef Bacik while (1) { 2428abed4aaaSJosef Bacik ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0); 2429abed4aaaSJosef Bacik if (ret < 0) 2430abed4aaaSJosef Bacik break; 2431abed4aaaSJosef Bacik 2432abed4aaaSJosef Bacik if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 2433abed4aaaSJosef Bacik ret = btrfs_next_leaf(tree_root, path); 2434abed4aaaSJosef Bacik if (ret) { 2435abed4aaaSJosef Bacik if (ret > 0) 2436abed4aaaSJosef Bacik ret = 0; 2437abed4aaaSJosef Bacik break; 2438abed4aaaSJosef Bacik } 2439abed4aaaSJosef Bacik } 2440abed4aaaSJosef Bacik ret = 0; 2441abed4aaaSJosef Bacik 2442abed4aaaSJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 2443abed4aaaSJosef Bacik if (key.objectid != objectid) 2444abed4aaaSJosef Bacik break; 2445abed4aaaSJosef Bacik btrfs_release_path(path); 2446abed4aaaSJosef Bacik 2447f7238e50SJosef Bacik /* 2448f7238e50SJosef Bacik * Just worry about this for extent tree, it'll be the same for 2449f7238e50SJosef Bacik * everybody. 2450f7238e50SJosef Bacik */ 2451f7238e50SJosef Bacik if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 2452f7238e50SJosef Bacik max_global_id = max(max_global_id, key.offset); 2453f7238e50SJosef Bacik 2454abed4aaaSJosef Bacik found = true; 2455abed4aaaSJosef Bacik root = read_tree_root_path(tree_root, path, &key); 2456abed4aaaSJosef Bacik if (IS_ERR(root)) { 2457abed4aaaSJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) 2458abed4aaaSJosef Bacik ret = PTR_ERR(root); 2459abed4aaaSJosef Bacik break; 2460abed4aaaSJosef Bacik } 2461abed4aaaSJosef Bacik set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2462abed4aaaSJosef Bacik ret = btrfs_global_root_insert(root); 2463abed4aaaSJosef Bacik if (ret) { 2464abed4aaaSJosef Bacik btrfs_put_root(root); 2465abed4aaaSJosef Bacik break; 2466abed4aaaSJosef Bacik } 2467abed4aaaSJosef Bacik key.offset++; 2468abed4aaaSJosef Bacik } 2469abed4aaaSJosef Bacik btrfs_release_path(path); 2470abed4aaaSJosef Bacik 2471f7238e50SJosef Bacik if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 2472f7238e50SJosef Bacik fs_info->nr_global_roots = max_global_id + 1; 2473f7238e50SJosef Bacik 2474abed4aaaSJosef Bacik if (!found || ret) { 2475abed4aaaSJosef Bacik if (objectid == BTRFS_CSUM_TREE_OBJECTID) 2476abed4aaaSJosef Bacik set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state); 2477abed4aaaSJosef Bacik 2478abed4aaaSJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) 2479abed4aaaSJosef Bacik ret = ret ? ret : -ENOENT; 2480abed4aaaSJosef Bacik else 2481abed4aaaSJosef Bacik ret = 0; 2482abed4aaaSJosef Bacik btrfs_err(fs_info, "failed to load root %s", name); 2483abed4aaaSJosef Bacik } 2484abed4aaaSJosef Bacik return ret; 2485abed4aaaSJosef Bacik } 2486abed4aaaSJosef Bacik 2487abed4aaaSJosef Bacik static int load_global_roots(struct btrfs_root *tree_root) 2488abed4aaaSJosef Bacik { 2489abed4aaaSJosef Bacik struct btrfs_path *path; 2490abed4aaaSJosef Bacik int ret = 0; 2491abed4aaaSJosef Bacik 2492abed4aaaSJosef Bacik path = btrfs_alloc_path(); 2493abed4aaaSJosef Bacik if (!path) 2494abed4aaaSJosef Bacik return -ENOMEM; 2495abed4aaaSJosef Bacik 2496abed4aaaSJosef Bacik ret = load_global_roots_objectid(tree_root, path, 2497abed4aaaSJosef Bacik BTRFS_EXTENT_TREE_OBJECTID, "extent"); 2498abed4aaaSJosef Bacik if (ret) 2499abed4aaaSJosef Bacik goto out; 2500abed4aaaSJosef Bacik ret = load_global_roots_objectid(tree_root, path, 2501abed4aaaSJosef Bacik BTRFS_CSUM_TREE_OBJECTID, "csum"); 2502abed4aaaSJosef Bacik if (ret) 2503abed4aaaSJosef Bacik goto out; 2504abed4aaaSJosef Bacik if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE)) 2505abed4aaaSJosef Bacik goto out; 2506abed4aaaSJosef Bacik ret = load_global_roots_objectid(tree_root, path, 2507abed4aaaSJosef Bacik BTRFS_FREE_SPACE_TREE_OBJECTID, 2508abed4aaaSJosef Bacik "free space"); 2509abed4aaaSJosef Bacik out: 2510abed4aaaSJosef Bacik btrfs_free_path(path); 2511abed4aaaSJosef Bacik return ret; 2512abed4aaaSJosef Bacik } 2513abed4aaaSJosef Bacik 25146bccf3abSJeff Mahoney static int btrfs_read_roots(struct btrfs_fs_info *fs_info) 25154bbcaa64SEric Sandeen { 25166bccf3abSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 2517a4f3d2c4SDavid Sterba struct btrfs_root *root; 25184bbcaa64SEric Sandeen struct btrfs_key location; 25194bbcaa64SEric Sandeen int ret; 25204bbcaa64SEric Sandeen 25216bccf3abSJeff Mahoney BUG_ON(!fs_info->tree_root); 25226bccf3abSJeff Mahoney 2523abed4aaaSJosef Bacik ret = load_global_roots(tree_root); 2524abed4aaaSJosef Bacik if (ret) 2525abed4aaaSJosef Bacik return ret; 2526abed4aaaSJosef Bacik 25274bbcaa64SEric Sandeen location.type = BTRFS_ROOT_ITEM_KEY; 25284bbcaa64SEric Sandeen location.offset = 0; 25294bbcaa64SEric Sandeen 25301c56ab99SQu Wenruo if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) { 253114033b08SQu Wenruo location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID; 253214033b08SQu Wenruo root = btrfs_read_tree_root(tree_root, &location); 253314033b08SQu Wenruo if (IS_ERR(root)) { 253414033b08SQu Wenruo if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 253514033b08SQu Wenruo ret = PTR_ERR(root); 253614033b08SQu Wenruo goto out; 253714033b08SQu Wenruo } 253814033b08SQu Wenruo } else { 253914033b08SQu Wenruo set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 254014033b08SQu Wenruo fs_info->block_group_root = root; 254114033b08SQu Wenruo } 254214033b08SQu Wenruo } 254314033b08SQu Wenruo 254414033b08SQu Wenruo location.objectid = BTRFS_DEV_TREE_OBJECTID; 2545a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2546f50f4353SLiu Bo if (IS_ERR(root)) { 254742437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2548f50f4353SLiu Bo ret = PTR_ERR(root); 2549f50f4353SLiu Bo goto out; 2550f50f4353SLiu Bo } 255142437a63SJosef Bacik } else { 2552a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2553a4f3d2c4SDavid Sterba fs_info->dev_root = root; 255442437a63SJosef Bacik } 2555820a49daSJosef Bacik /* Initialize fs_info for all devices in any case */ 2556820a49daSJosef Bacik btrfs_init_devices_late(fs_info); 25574bbcaa64SEric Sandeen 2558aeb935a4SQu Wenruo /* 2559aeb935a4SQu Wenruo * This tree can share blocks with some other fs tree during relocation 2560aeb935a4SQu Wenruo * and we need a proper setup by btrfs_get_fs_root 2561aeb935a4SQu Wenruo */ 256256e9357aSDavid Sterba root = btrfs_get_fs_root(tree_root->fs_info, 256356e9357aSDavid Sterba BTRFS_DATA_RELOC_TREE_OBJECTID, true); 2564aeb935a4SQu Wenruo if (IS_ERR(root)) { 256542437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2566aeb935a4SQu Wenruo ret = PTR_ERR(root); 2567aeb935a4SQu Wenruo goto out; 2568aeb935a4SQu Wenruo } 256942437a63SJosef Bacik } else { 2570aeb935a4SQu Wenruo set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2571aeb935a4SQu Wenruo fs_info->data_reloc_root = root; 257242437a63SJosef Bacik } 2573aeb935a4SQu Wenruo 25744bbcaa64SEric Sandeen location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2575a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2576a4f3d2c4SDavid Sterba if (!IS_ERR(root)) { 2577a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2578afcdd129SJosef Bacik set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); 2579a4f3d2c4SDavid Sterba fs_info->quota_root = root; 25804bbcaa64SEric Sandeen } 25814bbcaa64SEric Sandeen 25824bbcaa64SEric Sandeen location.objectid = BTRFS_UUID_TREE_OBJECTID; 2583a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2584a4f3d2c4SDavid Sterba if (IS_ERR(root)) { 258542437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2586a4f3d2c4SDavid Sterba ret = PTR_ERR(root); 25874bbcaa64SEric Sandeen if (ret != -ENOENT) 2588f50f4353SLiu Bo goto out; 258942437a63SJosef Bacik } 25904bbcaa64SEric Sandeen } else { 2591a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2592a4f3d2c4SDavid Sterba fs_info->uuid_root = root; 25934bbcaa64SEric Sandeen } 25944bbcaa64SEric Sandeen 25954bbcaa64SEric Sandeen return 0; 2596f50f4353SLiu Bo out: 2597f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", 2598f50f4353SLiu Bo location.objectid, ret); 2599f50f4353SLiu Bo return ret; 26004bbcaa64SEric Sandeen } 26014bbcaa64SEric Sandeen 2602069ec957SQu Wenruo /* 2603069ec957SQu Wenruo * Real super block validation 2604069ec957SQu Wenruo * NOTE: super csum type and incompat features will not be checked here. 2605069ec957SQu Wenruo * 2606069ec957SQu Wenruo * @sb: super block to check 2607069ec957SQu Wenruo * @mirror_num: the super block number to check its bytenr: 2608069ec957SQu Wenruo * 0 the primary (1st) sb 2609069ec957SQu Wenruo * 1, 2 2nd and 3rd backup copy 2610069ec957SQu Wenruo * -1 skip bytenr check 2611069ec957SQu Wenruo */ 2612a05d3c91SQu Wenruo int btrfs_validate_super(struct btrfs_fs_info *fs_info, 2613069ec957SQu Wenruo struct btrfs_super_block *sb, int mirror_num) 261421a852b0SQu Wenruo { 261521a852b0SQu Wenruo u64 nodesize = btrfs_super_nodesize(sb); 261621a852b0SQu Wenruo u64 sectorsize = btrfs_super_sectorsize(sb); 261721a852b0SQu Wenruo int ret = 0; 261821a852b0SQu Wenruo 261921a852b0SQu Wenruo if (btrfs_super_magic(sb) != BTRFS_MAGIC) { 262021a852b0SQu Wenruo btrfs_err(fs_info, "no valid FS found"); 262121a852b0SQu Wenruo ret = -EINVAL; 262221a852b0SQu Wenruo } 262321a852b0SQu Wenruo if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { 262421a852b0SQu Wenruo btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", 262521a852b0SQu Wenruo btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); 262621a852b0SQu Wenruo ret = -EINVAL; 262721a852b0SQu Wenruo } 262821a852b0SQu Wenruo if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 262921a852b0SQu Wenruo btrfs_err(fs_info, "tree_root level too big: %d >= %d", 263021a852b0SQu Wenruo btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 263121a852b0SQu Wenruo ret = -EINVAL; 263221a852b0SQu Wenruo } 263321a852b0SQu Wenruo if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 263421a852b0SQu Wenruo btrfs_err(fs_info, "chunk_root level too big: %d >= %d", 263521a852b0SQu Wenruo btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 263621a852b0SQu Wenruo ret = -EINVAL; 263721a852b0SQu Wenruo } 263821a852b0SQu Wenruo if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 263921a852b0SQu Wenruo btrfs_err(fs_info, "log_root level too big: %d >= %d", 264021a852b0SQu Wenruo btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 264121a852b0SQu Wenruo ret = -EINVAL; 264221a852b0SQu Wenruo } 264321a852b0SQu Wenruo 264421a852b0SQu Wenruo /* 264521a852b0SQu Wenruo * Check sectorsize and nodesize first, other check will need it. 264621a852b0SQu Wenruo * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. 264721a852b0SQu Wenruo */ 264821a852b0SQu Wenruo if (!is_power_of_2(sectorsize) || sectorsize < 4096 || 264921a852b0SQu Wenruo sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { 265021a852b0SQu Wenruo btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); 265121a852b0SQu Wenruo ret = -EINVAL; 265221a852b0SQu Wenruo } 26530bb3eb3eSQu Wenruo 26540bb3eb3eSQu Wenruo /* 26551a42daabSQu Wenruo * We only support at most two sectorsizes: 4K and PAGE_SIZE. 26561a42daabSQu Wenruo * 26571a42daabSQu Wenruo * We can support 16K sectorsize with 64K page size without problem, 26581a42daabSQu Wenruo * but such sectorsize/pagesize combination doesn't make much sense. 26591a42daabSQu Wenruo * 4K will be our future standard, PAGE_SIZE is supported from the very 26601a42daabSQu Wenruo * beginning. 26610bb3eb3eSQu Wenruo */ 26621a42daabSQu Wenruo if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) { 266321a852b0SQu Wenruo btrfs_err(fs_info, 26640bb3eb3eSQu Wenruo "sectorsize %llu not yet supported for page size %lu", 266521a852b0SQu Wenruo sectorsize, PAGE_SIZE); 266621a852b0SQu Wenruo ret = -EINVAL; 266721a852b0SQu Wenruo } 26680bb3eb3eSQu Wenruo 266921a852b0SQu Wenruo if (!is_power_of_2(nodesize) || nodesize < sectorsize || 267021a852b0SQu Wenruo nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { 267121a852b0SQu Wenruo btrfs_err(fs_info, "invalid nodesize %llu", nodesize); 267221a852b0SQu Wenruo ret = -EINVAL; 267321a852b0SQu Wenruo } 267421a852b0SQu Wenruo if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { 267521a852b0SQu Wenruo btrfs_err(fs_info, "invalid leafsize %u, should be %llu", 267621a852b0SQu Wenruo le32_to_cpu(sb->__unused_leafsize), nodesize); 267721a852b0SQu Wenruo ret = -EINVAL; 267821a852b0SQu Wenruo } 267921a852b0SQu Wenruo 268021a852b0SQu Wenruo /* Root alignment check */ 268121a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { 268221a852b0SQu Wenruo btrfs_warn(fs_info, "tree_root block unaligned: %llu", 268321a852b0SQu Wenruo btrfs_super_root(sb)); 268421a852b0SQu Wenruo ret = -EINVAL; 268521a852b0SQu Wenruo } 268621a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { 268721a852b0SQu Wenruo btrfs_warn(fs_info, "chunk_root block unaligned: %llu", 268821a852b0SQu Wenruo btrfs_super_chunk_root(sb)); 268921a852b0SQu Wenruo ret = -EINVAL; 269021a852b0SQu Wenruo } 269121a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { 269221a852b0SQu Wenruo btrfs_warn(fs_info, "log_root block unaligned: %llu", 269321a852b0SQu Wenruo btrfs_super_log_root(sb)); 269421a852b0SQu Wenruo ret = -EINVAL; 269521a852b0SQu Wenruo } 269621a852b0SQu Wenruo 2697aefd7f70SNikolay Borisov if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, 2698aefd7f70SNikolay Borisov BTRFS_FSID_SIZE)) { 2699aefd7f70SNikolay Borisov btrfs_err(fs_info, 2700aefd7f70SNikolay Borisov "superblock fsid doesn't match fsid of fs_devices: %pU != %pU", 2701aefd7f70SNikolay Borisov fs_info->super_copy->fsid, fs_info->fs_devices->fsid); 2702aefd7f70SNikolay Borisov ret = -EINVAL; 2703aefd7f70SNikolay Borisov } 2704aefd7f70SNikolay Borisov 2705aefd7f70SNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID) && 2706aefd7f70SNikolay Borisov memcmp(fs_info->fs_devices->metadata_uuid, 2707aefd7f70SNikolay Borisov fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) { 2708aefd7f70SNikolay Borisov btrfs_err(fs_info, 2709aefd7f70SNikolay Borisov "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU", 2710aefd7f70SNikolay Borisov fs_info->super_copy->metadata_uuid, 2711aefd7f70SNikolay Borisov fs_info->fs_devices->metadata_uuid); 2712aefd7f70SNikolay Borisov ret = -EINVAL; 2713aefd7f70SNikolay Borisov } 2714aefd7f70SNikolay Borisov 27151c56ab99SQu Wenruo /* 27161c56ab99SQu Wenruo * Artificial requirement for block-group-tree to force newer features 27171c56ab99SQu Wenruo * (free-space-tree, no-holes) so the test matrix is smaller. 27181c56ab99SQu Wenruo */ 27191c56ab99SQu Wenruo if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) && 27201c56ab99SQu Wenruo (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) || 27211c56ab99SQu Wenruo !btrfs_fs_incompat(fs_info, NO_HOLES))) { 27221c56ab99SQu Wenruo btrfs_err(fs_info, 27231c56ab99SQu Wenruo "block-group-tree feature requires fres-space-tree and no-holes"); 27241c56ab99SQu Wenruo ret = -EINVAL; 27251c56ab99SQu Wenruo } 27261c56ab99SQu Wenruo 2727de37aa51SNikolay Borisov if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, 27287239ff4bSNikolay Borisov BTRFS_FSID_SIZE) != 0) { 272921a852b0SQu Wenruo btrfs_err(fs_info, 27307239ff4bSNikolay Borisov "dev_item UUID does not match metadata fsid: %pU != %pU", 2731de37aa51SNikolay Borisov fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); 273221a852b0SQu Wenruo ret = -EINVAL; 273321a852b0SQu Wenruo } 273421a852b0SQu Wenruo 273521a852b0SQu Wenruo /* 273621a852b0SQu Wenruo * Hint to catch really bogus numbers, bitflips or so, more exact checks are 273721a852b0SQu Wenruo * done later 273821a852b0SQu Wenruo */ 273921a852b0SQu Wenruo if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { 274021a852b0SQu Wenruo btrfs_err(fs_info, "bytes_used is too small %llu", 274121a852b0SQu Wenruo btrfs_super_bytes_used(sb)); 274221a852b0SQu Wenruo ret = -EINVAL; 274321a852b0SQu Wenruo } 274421a852b0SQu Wenruo if (!is_power_of_2(btrfs_super_stripesize(sb))) { 274521a852b0SQu Wenruo btrfs_err(fs_info, "invalid stripesize %u", 274621a852b0SQu Wenruo btrfs_super_stripesize(sb)); 274721a852b0SQu Wenruo ret = -EINVAL; 274821a852b0SQu Wenruo } 274921a852b0SQu Wenruo if (btrfs_super_num_devices(sb) > (1UL << 31)) 275021a852b0SQu Wenruo btrfs_warn(fs_info, "suspicious number of devices: %llu", 275121a852b0SQu Wenruo btrfs_super_num_devices(sb)); 275221a852b0SQu Wenruo if (btrfs_super_num_devices(sb) == 0) { 275321a852b0SQu Wenruo btrfs_err(fs_info, "number of devices is 0"); 275421a852b0SQu Wenruo ret = -EINVAL; 275521a852b0SQu Wenruo } 275621a852b0SQu Wenruo 2757069ec957SQu Wenruo if (mirror_num >= 0 && 2758069ec957SQu Wenruo btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { 275921a852b0SQu Wenruo btrfs_err(fs_info, "super offset mismatch %llu != %u", 276021a852b0SQu Wenruo btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 276121a852b0SQu Wenruo ret = -EINVAL; 276221a852b0SQu Wenruo } 276321a852b0SQu Wenruo 276421a852b0SQu Wenruo /* 276521a852b0SQu Wenruo * Obvious sys_chunk_array corruptions, it must hold at least one key 276621a852b0SQu Wenruo * and one chunk 276721a852b0SQu Wenruo */ 276821a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 276921a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too big %u > %u", 277021a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 277121a852b0SQu Wenruo BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 277221a852b0SQu Wenruo ret = -EINVAL; 277321a852b0SQu Wenruo } 277421a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 277521a852b0SQu Wenruo + sizeof(struct btrfs_chunk)) { 277621a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too small %u < %zu", 277721a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 277821a852b0SQu Wenruo sizeof(struct btrfs_disk_key) 277921a852b0SQu Wenruo + sizeof(struct btrfs_chunk)); 278021a852b0SQu Wenruo ret = -EINVAL; 278121a852b0SQu Wenruo } 278221a852b0SQu Wenruo 278321a852b0SQu Wenruo /* 278421a852b0SQu Wenruo * The generation is a global counter, we'll trust it more than the others 278521a852b0SQu Wenruo * but it's still possible that it's the one that's wrong. 278621a852b0SQu Wenruo */ 278721a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 278821a852b0SQu Wenruo btrfs_warn(fs_info, 278921a852b0SQu Wenruo "suspicious: generation < chunk_root_generation: %llu < %llu", 279021a852b0SQu Wenruo btrfs_super_generation(sb), 279121a852b0SQu Wenruo btrfs_super_chunk_root_generation(sb)); 279221a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 279321a852b0SQu Wenruo && btrfs_super_cache_generation(sb) != (u64)-1) 279421a852b0SQu Wenruo btrfs_warn(fs_info, 279521a852b0SQu Wenruo "suspicious: generation < cache_generation: %llu < %llu", 279621a852b0SQu Wenruo btrfs_super_generation(sb), 279721a852b0SQu Wenruo btrfs_super_cache_generation(sb)); 279821a852b0SQu Wenruo 279921a852b0SQu Wenruo return ret; 280021a852b0SQu Wenruo } 280121a852b0SQu Wenruo 2802069ec957SQu Wenruo /* 2803069ec957SQu Wenruo * Validation of super block at mount time. 2804069ec957SQu Wenruo * Some checks already done early at mount time, like csum type and incompat 2805069ec957SQu Wenruo * flags will be skipped. 2806069ec957SQu Wenruo */ 2807069ec957SQu Wenruo static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) 2808069ec957SQu Wenruo { 2809a05d3c91SQu Wenruo return btrfs_validate_super(fs_info, fs_info->super_copy, 0); 2810069ec957SQu Wenruo } 2811069ec957SQu Wenruo 281275cb857dSQu Wenruo /* 281375cb857dSQu Wenruo * Validation of super block at write time. 281475cb857dSQu Wenruo * Some checks like bytenr check will be skipped as their values will be 281575cb857dSQu Wenruo * overwritten soon. 281675cb857dSQu Wenruo * Extra checks like csum type and incompat flags will be done here. 281775cb857dSQu Wenruo */ 281875cb857dSQu Wenruo static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, 281975cb857dSQu Wenruo struct btrfs_super_block *sb) 282075cb857dSQu Wenruo { 282175cb857dSQu Wenruo int ret; 282275cb857dSQu Wenruo 2823a05d3c91SQu Wenruo ret = btrfs_validate_super(fs_info, sb, -1); 282475cb857dSQu Wenruo if (ret < 0) 282575cb857dSQu Wenruo goto out; 2826e7e16f48SJohannes Thumshirn if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) { 282775cb857dSQu Wenruo ret = -EUCLEAN; 282875cb857dSQu Wenruo btrfs_err(fs_info, "invalid csum type, has %u want %u", 282975cb857dSQu Wenruo btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); 283075cb857dSQu Wenruo goto out; 283175cb857dSQu Wenruo } 283275cb857dSQu Wenruo if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 283375cb857dSQu Wenruo ret = -EUCLEAN; 283475cb857dSQu Wenruo btrfs_err(fs_info, 283575cb857dSQu Wenruo "invalid incompat flags, has 0x%llx valid mask 0x%llx", 283675cb857dSQu Wenruo btrfs_super_incompat_flags(sb), 283775cb857dSQu Wenruo (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); 283875cb857dSQu Wenruo goto out; 283975cb857dSQu Wenruo } 284075cb857dSQu Wenruo out: 284175cb857dSQu Wenruo if (ret < 0) 284275cb857dSQu Wenruo btrfs_err(fs_info, 284375cb857dSQu Wenruo "super block corruption detected before writing it to disk"); 284475cb857dSQu Wenruo return ret; 284575cb857dSQu Wenruo } 284675cb857dSQu Wenruo 2847bd676446SJosef Bacik static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level) 2848bd676446SJosef Bacik { 2849bd676446SJosef Bacik int ret = 0; 2850bd676446SJosef Bacik 2851bd676446SJosef Bacik root->node = read_tree_block(root->fs_info, bytenr, 2852bd676446SJosef Bacik root->root_key.objectid, gen, level, NULL); 2853bd676446SJosef Bacik if (IS_ERR(root->node)) { 2854bd676446SJosef Bacik ret = PTR_ERR(root->node); 2855bd676446SJosef Bacik root->node = NULL; 28564eb150d6SQu Wenruo return ret; 28574eb150d6SQu Wenruo } 28584eb150d6SQu Wenruo if (!extent_buffer_uptodate(root->node)) { 2859bd676446SJosef Bacik free_extent_buffer(root->node); 2860bd676446SJosef Bacik root->node = NULL; 28614eb150d6SQu Wenruo return -EIO; 2862bd676446SJosef Bacik } 2863bd676446SJosef Bacik 2864bd676446SJosef Bacik btrfs_set_root_node(&root->root_item, root->node); 2865bd676446SJosef Bacik root->commit_root = btrfs_root_node(root); 2866bd676446SJosef Bacik btrfs_set_root_refs(&root->root_item, 1); 2867bd676446SJosef Bacik return ret; 2868bd676446SJosef Bacik } 2869bd676446SJosef Bacik 2870bd676446SJosef Bacik static int load_important_roots(struct btrfs_fs_info *fs_info) 2871bd676446SJosef Bacik { 2872bd676446SJosef Bacik struct btrfs_super_block *sb = fs_info->super_copy; 2873bd676446SJosef Bacik u64 gen, bytenr; 2874bd676446SJosef Bacik int level, ret; 2875bd676446SJosef Bacik 2876bd676446SJosef Bacik bytenr = btrfs_super_root(sb); 2877bd676446SJosef Bacik gen = btrfs_super_generation(sb); 2878bd676446SJosef Bacik level = btrfs_super_root_level(sb); 2879bd676446SJosef Bacik ret = load_super_root(fs_info->tree_root, bytenr, gen, level); 28809c54e80dSJosef Bacik if (ret) { 2881bd676446SJosef Bacik btrfs_warn(fs_info, "couldn't read tree root"); 2882bd676446SJosef Bacik return ret; 2883bd676446SJosef Bacik } 28849c54e80dSJosef Bacik return 0; 28859c54e80dSJosef Bacik } 28869c54e80dSJosef Bacik 28876ef108ddSNikolay Borisov static int __cold init_tree_roots(struct btrfs_fs_info *fs_info) 2888b8522a1eSNikolay Borisov { 28896ef108ddSNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 2890b8522a1eSNikolay Borisov struct btrfs_super_block *sb = fs_info->super_copy; 2891b8522a1eSNikolay Borisov struct btrfs_root *tree_root = fs_info->tree_root; 2892b8522a1eSNikolay Borisov bool handle_error = false; 2893b8522a1eSNikolay Borisov int ret = 0; 2894b8522a1eSNikolay Borisov int i; 2895b8522a1eSNikolay Borisov 2896b8522a1eSNikolay Borisov for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 2897b8522a1eSNikolay Borisov if (handle_error) { 2898b8522a1eSNikolay Borisov if (!IS_ERR(tree_root->node)) 2899b8522a1eSNikolay Borisov free_extent_buffer(tree_root->node); 2900b8522a1eSNikolay Borisov tree_root->node = NULL; 2901b8522a1eSNikolay Borisov 2902b8522a1eSNikolay Borisov if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) 2903b8522a1eSNikolay Borisov break; 2904b8522a1eSNikolay Borisov 2905b8522a1eSNikolay Borisov free_root_pointers(fs_info, 0); 2906b8522a1eSNikolay Borisov 2907b8522a1eSNikolay Borisov /* 2908b8522a1eSNikolay Borisov * Don't use the log in recovery mode, it won't be 2909b8522a1eSNikolay Borisov * valid 2910b8522a1eSNikolay Borisov */ 2911b8522a1eSNikolay Borisov btrfs_set_super_log_root(sb, 0); 2912b8522a1eSNikolay Borisov 2913b8522a1eSNikolay Borisov /* We can't trust the free space cache either */ 2914b8522a1eSNikolay Borisov btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 2915b8522a1eSNikolay Borisov 2916b8522a1eSNikolay Borisov ret = read_backup_root(fs_info, i); 29176ef108ddSNikolay Borisov backup_index = ret; 2918b8522a1eSNikolay Borisov if (ret < 0) 2919b8522a1eSNikolay Borisov return ret; 2920b8522a1eSNikolay Borisov } 2921b8522a1eSNikolay Borisov 2922bd676446SJosef Bacik ret = load_important_roots(fs_info); 2923bd676446SJosef Bacik if (ret) { 2924217f5004SNikolay Borisov handle_error = true; 2925b8522a1eSNikolay Borisov continue; 2926b8522a1eSNikolay Borisov } 2927b8522a1eSNikolay Borisov 2928336a0d8dSNikolay Borisov /* 2929336a0d8dSNikolay Borisov * No need to hold btrfs_root::objectid_mutex since the fs 2930336a0d8dSNikolay Borisov * hasn't been fully initialised and we are the only user 2931336a0d8dSNikolay Borisov */ 2932453e4873SNikolay Borisov ret = btrfs_init_root_free_objectid(tree_root); 2933b8522a1eSNikolay Borisov if (ret < 0) { 2934b8522a1eSNikolay Borisov handle_error = true; 2935b8522a1eSNikolay Borisov continue; 2936b8522a1eSNikolay Borisov } 2937b8522a1eSNikolay Borisov 29386b8fad57SNikolay Borisov ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID); 2939b8522a1eSNikolay Borisov 2940b8522a1eSNikolay Borisov ret = btrfs_read_roots(fs_info); 2941b8522a1eSNikolay Borisov if (ret < 0) { 2942b8522a1eSNikolay Borisov handle_error = true; 2943b8522a1eSNikolay Borisov continue; 2944b8522a1eSNikolay Borisov } 2945b8522a1eSNikolay Borisov 2946b8522a1eSNikolay Borisov /* All successful */ 2947bd676446SJosef Bacik fs_info->generation = btrfs_header_generation(tree_root->node); 2948bd676446SJosef Bacik fs_info->last_trans_committed = fs_info->generation; 2949d96b3424SFilipe Manana fs_info->last_reloc_trans = 0; 29506ef108ddSNikolay Borisov 29516ef108ddSNikolay Borisov /* Always begin writing backup roots after the one being used */ 29526ef108ddSNikolay Borisov if (backup_index < 0) { 29536ef108ddSNikolay Borisov fs_info->backup_root_index = 0; 29546ef108ddSNikolay Borisov } else { 29556ef108ddSNikolay Borisov fs_info->backup_root_index = backup_index + 1; 29566ef108ddSNikolay Borisov fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS; 29576ef108ddSNikolay Borisov } 2958b8522a1eSNikolay Borisov break; 2959b8522a1eSNikolay Borisov } 2960b8522a1eSNikolay Borisov 2961b8522a1eSNikolay Borisov return ret; 2962b8522a1eSNikolay Borisov } 2963b8522a1eSNikolay Borisov 29648260edbaSJosef Bacik void btrfs_init_fs_info(struct btrfs_fs_info *fs_info) 2965eb60ceacSChris Mason { 2966fc7cbcd4SDavid Sterba INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 296701cd3909SDavid Sterba INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 29688fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 2969facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 297024bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 2971eb73c1b7SMiao Xie INIT_LIST_HEAD(&fs_info->delalloc_roots); 297211833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 2973eb73c1b7SMiao Xie spin_lock_init(&fs_info->delalloc_root_lock); 2974a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 2975fc7cbcd4SDavid Sterba spin_lock_init(&fs_info->fs_roots_radix_lock); 297624bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 29774cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 2978ceda0864SMiao Xie spin_lock_init(&fs_info->super_lock); 2979f28491e0SJosef Bacik spin_lock_init(&fs_info->buffer_lock); 298047ab2a6cSJosef Bacik spin_lock_init(&fs_info->unused_bgs_lock); 298140ab3be1SNaohiro Aota spin_lock_init(&fs_info->treelog_bg_lock); 2982afba2bc0SNaohiro Aota spin_lock_init(&fs_info->zone_active_bgs_lock); 2983c2707a25SJohannes Thumshirn spin_lock_init(&fs_info->relocation_bg_lock); 2984f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 2985abed4aaaSJosef Bacik rwlock_init(&fs_info->global_root_lock); 2986d7c15171SZhao Lei mutex_init(&fs_info->unused_bg_unpin_mutex); 2987f3372065SJohannes Thumshirn mutex_init(&fs_info->reclaim_bgs_lock); 29887585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 2989573bfb72SMiao Xie mutex_init(&fs_info->delalloc_root_mutex); 29900bc09ca1SNaohiro Aota mutex_init(&fs_info->zoned_meta_io_lock); 29915f0addf7SNaohiro Aota mutex_init(&fs_info->zoned_data_reloc_io_lock); 2992de98ced9SMiao Xie seqlock_init(&fs_info->profiles_lock); 299319c00ddcSChris Mason 2994e1489b4fSIoannis Angelakopoulos btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers); 29955a9ba670SIoannis Angelakopoulos btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters); 29968b53779eSIoannis Angelakopoulos btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered); 29975f4403e1SIoannis Angelakopoulos btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent); 29983e738c53SIoannis Angelakopoulos btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_start, 29993e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_COMMIT_START); 30003e738c53SIoannis Angelakopoulos btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked, 30013e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_UNBLOCKED); 30023e738c53SIoannis Angelakopoulos btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed, 30033e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 30043e738c53SIoannis Angelakopoulos btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed, 30053e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_COMPLETED); 3006e1489b4fSIoannis Angelakopoulos 30070b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 30086324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 3009f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 301047ab2a6cSJosef Bacik INIT_LIST_HEAD(&fs_info->unused_bgs); 301118bb8bbfSJohannes Thumshirn INIT_LIST_HEAD(&fs_info->reclaim_bgs); 3012afba2bc0SNaohiro Aota INIT_LIST_HEAD(&fs_info->zone_active_bgs); 3013bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 3014bd647ce3SJosef Bacik INIT_LIST_HEAD(&fs_info->allocated_roots); 30153fd63727SJosef Bacik INIT_LIST_HEAD(&fs_info->allocated_ebs); 30163fd63727SJosef Bacik spin_lock_init(&fs_info->eb_leak_lock); 3017bd647ce3SJosef Bacik #endif 3018c8bf1b67SDavid Sterba extent_map_tree_init(&fs_info->mapping_tree); 301966d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 302066d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 302166d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 302266d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 302366d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 302466d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 302566d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 3026ba2c4d4eSJosef Bacik btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, 3027ba2c4d4eSJosef Bacik BTRFS_BLOCK_RSV_DELREFS); 3028ba2c4d4eSJosef Bacik 3029771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 30304cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 3031034f784dSJosef Bacik atomic_set(&fs_info->nr_delayed_iputs, 0); 3032fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 3033abed4aaaSJosef Bacik fs_info->global_root_tree = RB_ROOT; 303495ac567aSFilipe David Borba Manana fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 30359ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 30364cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 3037a5ed45f8SNikolay Borisov atomic64_set(&fs_info->free_chunk_space, 0); 3038f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 30398b87dc17SDavid Sterba fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 3040f8c269d7SDavid Sterba fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 3041fd708b81SJosef Bacik btrfs_init_ref_verify(fs_info); 3042c8b97818SChris Mason 3043b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 3044b34b086cSChris Mason num_online_cpus() + 2, 8); 30450afbaf8cSChris Mason 3046199c2a9cSMiao Xie INIT_LIST_HEAD(&fs_info->ordered_roots); 3047199c2a9cSMiao Xie spin_lock_init(&fs_info->ordered_root_lock); 304869fe2d75SJosef Bacik 3049638aa7edSEric Sandeen btrfs_init_scrub(fs_info); 305021adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 305121adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 305221adbd5cSStefan Behrens #endif 3053779a65a4SEric Sandeen btrfs_init_balance(fs_info); 305457056740SJosef Bacik btrfs_init_async_reclaim_work(fs_info); 3055a2de733cSArne Jansen 305616b0c258SFilipe Manana rwlock_init(&fs_info->block_group_cache_lock); 305708dddb29SFilipe Manana fs_info->block_group_cache_tree = RB_ROOT_CACHED; 30580f9dd46cSJosef Bacik 3059fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &fs_info->excluded_extents, 3060fe119a6eSNikolay Borisov IO_TREE_FS_EXCLUDED_EXTENTS, NULL); 306139279cc3SChris Mason 30625a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 3063e02119d5SChris Mason mutex_init(&fs_info->tree_log_mutex); 3064925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 3065a74a4b97SChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 3066a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 30671bbc621eSChris Mason mutex_init(&fs_info->ro_block_group_mutex); 30689e351cc8SJosef Bacik init_rwsem(&fs_info->commit_root_sem); 3069c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 307076dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 3071803b2f54SStefan Behrens sema_init(&fs_info->uuid_tree_rescan_sem, 1); 3072fa9c0d79SChris Mason 3073ad618368SEric Sandeen btrfs_init_dev_replace_locks(fs_info); 3074f9e92e40SEric Sandeen btrfs_init_qgroup(fs_info); 3075b0643e59SDennis Zhou btrfs_discard_init(fs_info); 3076416ac51dSArne Jansen 3077fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 3078fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 3079fa9c0d79SChris Mason 3080e6dcd2dcSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 3081f9295749SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 3082bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 30834854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 3084034f784dSJosef Bacik init_waitqueue_head(&fs_info->delayed_iputs_wait); 30853768f368SChris Mason 3086da17066cSJeff Mahoney /* Usable values until the real ones are cached from the superblock */ 3087da17066cSJeff Mahoney fs_info->nodesize = 4096; 3088da17066cSJeff Mahoney fs_info->sectorsize = 4096; 3089ab108d99SDavid Sterba fs_info->sectorsize_bits = ilog2(4096); 3090da17066cSJeff Mahoney fs_info->stripesize = 4096; 3091da17066cSJeff Mahoney 3092f7b12a62SNaohiro Aota fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE; 3093f7b12a62SNaohiro Aota 3094eede2bf3SOmar Sandoval spin_lock_init(&fs_info->swapfile_pins_lock); 3095eede2bf3SOmar Sandoval fs_info->swapfile_pins = RB_ROOT; 3096eede2bf3SOmar Sandoval 309718bb8bbfSJohannes Thumshirn fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH; 309818bb8bbfSJohannes Thumshirn INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work); 30998260edbaSJosef Bacik } 31008260edbaSJosef Bacik 31018260edbaSJosef Bacik static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb) 31028260edbaSJosef Bacik { 31038260edbaSJosef Bacik int ret; 31048260edbaSJosef Bacik 31058260edbaSJosef Bacik fs_info->sb = sb; 31068260edbaSJosef Bacik sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; 31078260edbaSJosef Bacik sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); 31089e967495SFilipe Manana 31095deb17e1SJosef Bacik ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL); 3110ae18c37aSJosef Bacik if (ret) 3111ae18c37aSJosef Bacik return ret; 3112ae18c37aSJosef Bacik 3113ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 3114ae18c37aSJosef Bacik if (ret) 3115c75e8394SJosef Bacik return ret; 3116ae18c37aSJosef Bacik 3117ae18c37aSJosef Bacik fs_info->dirty_metadata_batch = PAGE_SIZE * 3118ae18c37aSJosef Bacik (1 + ilog2(nr_cpu_ids)); 3119ae18c37aSJosef Bacik 3120ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 3121ae18c37aSJosef Bacik if (ret) 3122c75e8394SJosef Bacik return ret; 3123ae18c37aSJosef Bacik 3124ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, 3125ae18c37aSJosef Bacik GFP_KERNEL); 3126ae18c37aSJosef Bacik if (ret) 3127c75e8394SJosef Bacik return ret; 3128ae18c37aSJosef Bacik 3129ae18c37aSJosef Bacik fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 3130ae18c37aSJosef Bacik GFP_KERNEL); 3131c75e8394SJosef Bacik if (!fs_info->delayed_root) 3132c75e8394SJosef Bacik return -ENOMEM; 3133ae18c37aSJosef Bacik btrfs_init_delayed_root(fs_info->delayed_root); 3134ae18c37aSJosef Bacik 3135a0a1db70SFilipe Manana if (sb_rdonly(sb)) 3136a0a1db70SFilipe Manana set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state); 3137a0a1db70SFilipe Manana 3138c75e8394SJosef Bacik return btrfs_alloc_stripe_hash_table(fs_info); 3139ae18c37aSJosef Bacik } 3140ae18c37aSJosef Bacik 314197f4dd09SNikolay Borisov static int btrfs_uuid_rescan_kthread(void *data) 314297f4dd09SNikolay Borisov { 31430d031dc4SYu Zhe struct btrfs_fs_info *fs_info = data; 314497f4dd09SNikolay Borisov int ret; 314597f4dd09SNikolay Borisov 314697f4dd09SNikolay Borisov /* 314797f4dd09SNikolay Borisov * 1st step is to iterate through the existing UUID tree and 314897f4dd09SNikolay Borisov * to delete all entries that contain outdated data. 314997f4dd09SNikolay Borisov * 2nd step is to add all missing entries to the UUID tree. 315097f4dd09SNikolay Borisov */ 315197f4dd09SNikolay Borisov ret = btrfs_uuid_tree_iterate(fs_info); 315297f4dd09SNikolay Borisov if (ret < 0) { 3153c94bec2cSJosef Bacik if (ret != -EINTR) 3154c94bec2cSJosef Bacik btrfs_warn(fs_info, "iterating uuid_tree failed %d", 3155c94bec2cSJosef Bacik ret); 315697f4dd09SNikolay Borisov up(&fs_info->uuid_tree_rescan_sem); 315797f4dd09SNikolay Borisov return ret; 315897f4dd09SNikolay Borisov } 315997f4dd09SNikolay Borisov return btrfs_uuid_scan_kthread(data); 316097f4dd09SNikolay Borisov } 316197f4dd09SNikolay Borisov 316297f4dd09SNikolay Borisov static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) 316397f4dd09SNikolay Borisov { 316497f4dd09SNikolay Borisov struct task_struct *task; 316597f4dd09SNikolay Borisov 316697f4dd09SNikolay Borisov down(&fs_info->uuid_tree_rescan_sem); 316797f4dd09SNikolay Borisov task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); 316897f4dd09SNikolay Borisov if (IS_ERR(task)) { 316997f4dd09SNikolay Borisov /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 317097f4dd09SNikolay Borisov btrfs_warn(fs_info, "failed to start uuid_rescan task"); 317197f4dd09SNikolay Borisov up(&fs_info->uuid_tree_rescan_sem); 317297f4dd09SNikolay Borisov return PTR_ERR(task); 317397f4dd09SNikolay Borisov } 317497f4dd09SNikolay Borisov 317597f4dd09SNikolay Borisov return 0; 317697f4dd09SNikolay Borisov } 317797f4dd09SNikolay Borisov 317844c0ca21SBoris Burkov /* 31798cd29088SBoris Burkov * Some options only have meaning at mount time and shouldn't persist across 31808cd29088SBoris Burkov * remounts, or be displayed. Clear these at the end of mount and remount 31818cd29088SBoris Burkov * code paths. 31828cd29088SBoris Burkov */ 31838cd29088SBoris Burkov void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info) 31848cd29088SBoris Burkov { 31858cd29088SBoris Burkov btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT); 31868b228324SBoris Burkov btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE); 31878cd29088SBoris Burkov } 31888cd29088SBoris Burkov 31898cd29088SBoris Burkov /* 319044c0ca21SBoris Burkov * Mounting logic specific to read-write file systems. Shared by open_ctree 319144c0ca21SBoris Burkov * and btrfs_remount when remounting from read-only to read-write. 319244c0ca21SBoris Burkov */ 319344c0ca21SBoris Burkov int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info) 319444c0ca21SBoris Burkov { 319544c0ca21SBoris Burkov int ret; 319694846229SBoris Burkov const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE); 31978b228324SBoris Burkov bool clear_free_space_tree = false; 31988b228324SBoris Burkov 31998b228324SBoris Burkov if (btrfs_test_opt(fs_info, CLEAR_CACHE) && 32008b228324SBoris Burkov btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 32018b228324SBoris Burkov clear_free_space_tree = true; 32028b228324SBoris Burkov } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 32038b228324SBoris Burkov !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { 32048b228324SBoris Burkov btrfs_warn(fs_info, "free space tree is invalid"); 32058b228324SBoris Burkov clear_free_space_tree = true; 32068b228324SBoris Burkov } 32078b228324SBoris Burkov 32088b228324SBoris Burkov if (clear_free_space_tree) { 32098b228324SBoris Burkov btrfs_info(fs_info, "clearing free space tree"); 32108b228324SBoris Burkov ret = btrfs_clear_free_space_tree(fs_info); 32118b228324SBoris Burkov if (ret) { 32128b228324SBoris Burkov btrfs_warn(fs_info, 32138b228324SBoris Burkov "failed to clear free space tree: %d", ret); 32148b228324SBoris Burkov goto out; 32158b228324SBoris Burkov } 32168b228324SBoris Burkov } 321744c0ca21SBoris Burkov 32188d488a8cSFilipe Manana /* 32198d488a8cSFilipe Manana * btrfs_find_orphan_roots() is responsible for finding all the dead 32208d488a8cSFilipe Manana * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load 3221fc7cbcd4SDavid Sterba * them into the fs_info->fs_roots_radix tree. This must be done before 32228d488a8cSFilipe Manana * calling btrfs_orphan_cleanup() on the tree root. If we don't do it 32238d488a8cSFilipe Manana * first, then btrfs_orphan_cleanup() will delete a dead root's orphan 32248d488a8cSFilipe Manana * item before the root's tree is deleted - this means that if we unmount 32258d488a8cSFilipe Manana * or crash before the deletion completes, on the next mount we will not 32268d488a8cSFilipe Manana * delete what remains of the tree because the orphan item does not 32278d488a8cSFilipe Manana * exists anymore, which is what tells us we have a pending deletion. 32288d488a8cSFilipe Manana */ 32298d488a8cSFilipe Manana ret = btrfs_find_orphan_roots(fs_info); 32308d488a8cSFilipe Manana if (ret) 32318d488a8cSFilipe Manana goto out; 32328d488a8cSFilipe Manana 323344c0ca21SBoris Burkov ret = btrfs_cleanup_fs_roots(fs_info); 323444c0ca21SBoris Burkov if (ret) 323544c0ca21SBoris Burkov goto out; 323644c0ca21SBoris Burkov 32378f1c21d7SBoris Burkov down_read(&fs_info->cleanup_work_sem); 32388f1c21d7SBoris Burkov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 32398f1c21d7SBoris Burkov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 32408f1c21d7SBoris Burkov up_read(&fs_info->cleanup_work_sem); 32418f1c21d7SBoris Burkov goto out; 32428f1c21d7SBoris Burkov } 32438f1c21d7SBoris Burkov up_read(&fs_info->cleanup_work_sem); 32448f1c21d7SBoris Burkov 324544c0ca21SBoris Burkov mutex_lock(&fs_info->cleaner_mutex); 32467eefae6bSJosef Bacik ret = btrfs_recover_relocation(fs_info); 324744c0ca21SBoris Burkov mutex_unlock(&fs_info->cleaner_mutex); 324844c0ca21SBoris Burkov if (ret < 0) { 324944c0ca21SBoris Burkov btrfs_warn(fs_info, "failed to recover relocation: %d", ret); 325044c0ca21SBoris Burkov goto out; 325144c0ca21SBoris Burkov } 325244c0ca21SBoris Burkov 32535011139aSBoris Burkov if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && 32545011139aSBoris Burkov !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 32555011139aSBoris Burkov btrfs_info(fs_info, "creating free space tree"); 32565011139aSBoris Burkov ret = btrfs_create_free_space_tree(fs_info); 32575011139aSBoris Burkov if (ret) { 32585011139aSBoris Burkov btrfs_warn(fs_info, 32595011139aSBoris Burkov "failed to create free space tree: %d", ret); 32605011139aSBoris Burkov goto out; 32615011139aSBoris Burkov } 32625011139aSBoris Burkov } 32635011139aSBoris Burkov 326494846229SBoris Burkov if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) { 326594846229SBoris Burkov ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt); 326694846229SBoris Burkov if (ret) 326794846229SBoris Burkov goto out; 326894846229SBoris Burkov } 326994846229SBoris Burkov 327044c0ca21SBoris Burkov ret = btrfs_resume_balance_async(fs_info); 327144c0ca21SBoris Burkov if (ret) 327244c0ca21SBoris Burkov goto out; 327344c0ca21SBoris Burkov 327444c0ca21SBoris Burkov ret = btrfs_resume_dev_replace_async(fs_info); 327544c0ca21SBoris Burkov if (ret) { 327644c0ca21SBoris Burkov btrfs_warn(fs_info, "failed to resume dev_replace"); 327744c0ca21SBoris Burkov goto out; 327844c0ca21SBoris Burkov } 327944c0ca21SBoris Burkov 328044c0ca21SBoris Burkov btrfs_qgroup_rescan_resume(fs_info); 328144c0ca21SBoris Burkov 328244c0ca21SBoris Burkov if (!fs_info->uuid_root) { 328344c0ca21SBoris Burkov btrfs_info(fs_info, "creating UUID tree"); 328444c0ca21SBoris Burkov ret = btrfs_create_uuid_tree(fs_info); 328544c0ca21SBoris Burkov if (ret) { 328644c0ca21SBoris Burkov btrfs_warn(fs_info, 328744c0ca21SBoris Burkov "failed to create the UUID tree %d", ret); 328844c0ca21SBoris Burkov goto out; 328944c0ca21SBoris Burkov } 329044c0ca21SBoris Burkov } 329144c0ca21SBoris Burkov 329244c0ca21SBoris Burkov out: 329344c0ca21SBoris Burkov return ret; 329444c0ca21SBoris Burkov } 329544c0ca21SBoris Burkov 3296ae18c37aSJosef Bacik int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices, 3297ae18c37aSJosef Bacik char *options) 3298ae18c37aSJosef Bacik { 3299ae18c37aSJosef Bacik u32 sectorsize; 3300ae18c37aSJosef Bacik u32 nodesize; 3301ae18c37aSJosef Bacik u32 stripesize; 3302ae18c37aSJosef Bacik u64 generation; 3303ae18c37aSJosef Bacik u64 features; 3304ae18c37aSJosef Bacik u16 csum_type; 3305ae18c37aSJosef Bacik struct btrfs_super_block *disk_super; 3306ae18c37aSJosef Bacik struct btrfs_fs_info *fs_info = btrfs_sb(sb); 3307ae18c37aSJosef Bacik struct btrfs_root *tree_root; 3308ae18c37aSJosef Bacik struct btrfs_root *chunk_root; 3309ae18c37aSJosef Bacik int ret; 3310ae18c37aSJosef Bacik int err = -EINVAL; 3311ae18c37aSJosef Bacik int level; 3312ae18c37aSJosef Bacik 33138260edbaSJosef Bacik ret = init_mount_fs_info(fs_info, sb); 331453b381b3SDavid Woodhouse if (ret) { 331583c8266aSDavid Sterba err = ret; 3316ae18c37aSJosef Bacik goto fail; 331753b381b3SDavid Woodhouse } 331853b381b3SDavid Woodhouse 3319ae18c37aSJosef Bacik /* These need to be init'ed before we start creating inodes and such. */ 3320ae18c37aSJosef Bacik tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, 3321ae18c37aSJosef Bacik GFP_KERNEL); 3322ae18c37aSJosef Bacik fs_info->tree_root = tree_root; 3323ae18c37aSJosef Bacik chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID, 3324ae18c37aSJosef Bacik GFP_KERNEL); 3325ae18c37aSJosef Bacik fs_info->chunk_root = chunk_root; 3326ae18c37aSJosef Bacik if (!tree_root || !chunk_root) { 3327ae18c37aSJosef Bacik err = -ENOMEM; 3328c75e8394SJosef Bacik goto fail; 3329ae18c37aSJosef Bacik } 3330ae18c37aSJosef Bacik 3331ae18c37aSJosef Bacik fs_info->btree_inode = new_inode(sb); 3332ae18c37aSJosef Bacik if (!fs_info->btree_inode) { 3333ae18c37aSJosef Bacik err = -ENOMEM; 3334c75e8394SJosef Bacik goto fail; 3335ae18c37aSJosef Bacik } 3336ae18c37aSJosef Bacik mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 3337ae18c37aSJosef Bacik btrfs_init_btree_inode(fs_info); 3338ae18c37aSJosef Bacik 3339d24fa5c1SAnand Jain invalidate_bdev(fs_devices->latest_dev->bdev); 33401104a885SDavid Sterba 33411104a885SDavid Sterba /* 33421104a885SDavid Sterba * Read super block and check the signature bytes only 33431104a885SDavid Sterba */ 3344d24fa5c1SAnand Jain disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev); 33458f32380dSJohannes Thumshirn if (IS_ERR(disk_super)) { 33468f32380dSJohannes Thumshirn err = PTR_ERR(disk_super); 334716cdcec7SMiao Xie goto fail_alloc; 334820b45077SDave Young } 334939279cc3SChris Mason 33501104a885SDavid Sterba /* 3351260db43cSRandy Dunlap * Verify the type first, if that or the checksum value are 33528dc3f22cSJohannes Thumshirn * corrupted, we'll find out 33538dc3f22cSJohannes Thumshirn */ 33548f32380dSJohannes Thumshirn csum_type = btrfs_super_csum_type(disk_super); 335551bce6c9SJohannes Thumshirn if (!btrfs_supported_super_csum(csum_type)) { 33568dc3f22cSJohannes Thumshirn btrfs_err(fs_info, "unsupported checksum algorithm: %u", 335751bce6c9SJohannes Thumshirn csum_type); 33588dc3f22cSJohannes Thumshirn err = -EINVAL; 33598f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 33608dc3f22cSJohannes Thumshirn goto fail_alloc; 33618dc3f22cSJohannes Thumshirn } 33628dc3f22cSJohannes Thumshirn 336383c68bbcSSu Yue fs_info->csum_size = btrfs_super_csum_size(disk_super); 336483c68bbcSSu Yue 33656d97c6e3SJohannes Thumshirn ret = btrfs_init_csum_hash(fs_info, csum_type); 33666d97c6e3SJohannes Thumshirn if (ret) { 33676d97c6e3SJohannes Thumshirn err = ret; 33688f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 33696d97c6e3SJohannes Thumshirn goto fail_alloc; 33706d97c6e3SJohannes Thumshirn } 33716d97c6e3SJohannes Thumshirn 33728dc3f22cSJohannes Thumshirn /* 33731104a885SDavid Sterba * We want to check superblock checksum, the type is stored inside. 33741104a885SDavid Sterba * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 33751104a885SDavid Sterba */ 33768f32380dSJohannes Thumshirn if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) { 337705135f59SDavid Sterba btrfs_err(fs_info, "superblock checksum mismatch"); 33781104a885SDavid Sterba err = -EINVAL; 33798f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 3380141386e1SJosef Bacik goto fail_alloc; 33811104a885SDavid Sterba } 33821104a885SDavid Sterba 33831104a885SDavid Sterba /* 33841104a885SDavid Sterba * super_copy is zeroed at allocation time and we never touch the 33851104a885SDavid Sterba * following bytes up to INFO_SIZE, the checksum is calculated from 33861104a885SDavid Sterba * the whole block of INFO_SIZE 33871104a885SDavid Sterba */ 33888f32380dSJohannes Thumshirn memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy)); 33898f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 33905f39d397SChris Mason 3391fbc6feaeSNikolay Borisov disk_super = fs_info->super_copy; 3392fbc6feaeSNikolay Borisov 33930b86a832SChris Mason 3394fbc6feaeSNikolay Borisov features = btrfs_super_flags(disk_super); 3395fbc6feaeSNikolay Borisov if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { 3396fbc6feaeSNikolay Borisov features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; 3397fbc6feaeSNikolay Borisov btrfs_set_super_flags(disk_super, features); 3398fbc6feaeSNikolay Borisov btrfs_info(fs_info, 3399fbc6feaeSNikolay Borisov "found metadata UUID change in progress flag, clearing"); 3400fbc6feaeSNikolay Borisov } 3401fbc6feaeSNikolay Borisov 3402fbc6feaeSNikolay Borisov memcpy(fs_info->super_for_commit, fs_info->super_copy, 3403fbc6feaeSNikolay Borisov sizeof(*fs_info->super_for_commit)); 3404de37aa51SNikolay Borisov 3405069ec957SQu Wenruo ret = btrfs_validate_mount_super(fs_info); 34061104a885SDavid Sterba if (ret) { 340705135f59SDavid Sterba btrfs_err(fs_info, "superblock contains fatal errors"); 34081104a885SDavid Sterba err = -EINVAL; 3409141386e1SJosef Bacik goto fail_alloc; 34101104a885SDavid Sterba } 34111104a885SDavid Sterba 34120f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 3413141386e1SJosef Bacik goto fail_alloc; 34140f7d52f4SChris Mason 3415acce952bSliubo /* check FS state, whether FS is broken. */ 341687533c47SMiao Xie if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 341787533c47SMiao Xie set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 3418acce952bSliubo 341975e7cb7fSLiu Bo /* 342075e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 342175e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 342275e7cb7fSLiu Bo */ 342375e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 342475e7cb7fSLiu Bo 34256f93e834SAnand Jain 34266f93e834SAnand Jain /* Set up fs_info before parsing mount options */ 34276f93e834SAnand Jain nodesize = btrfs_super_nodesize(disk_super); 34286f93e834SAnand Jain sectorsize = btrfs_super_sectorsize(disk_super); 34296f93e834SAnand Jain stripesize = sectorsize; 34306f93e834SAnand Jain fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 34316f93e834SAnand Jain fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 34326f93e834SAnand Jain 34336f93e834SAnand Jain fs_info->nodesize = nodesize; 34346f93e834SAnand Jain fs_info->sectorsize = sectorsize; 34356f93e834SAnand Jain fs_info->sectorsize_bits = ilog2(sectorsize); 34366f93e834SAnand Jain fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size; 34376f93e834SAnand Jain fs_info->stripesize = stripesize; 34386f93e834SAnand Jain 34392ff7e61eSJeff Mahoney ret = btrfs_parse_options(fs_info, options, sb->s_flags); 34402b82032cSYan Zheng if (ret) { 34412b82032cSYan Zheng err = ret; 3442141386e1SJosef Bacik goto fail_alloc; 34432b82032cSYan Zheng } 3444dfe25020SChris Mason 3445f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 3446f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 3447f2b636e8SJosef Bacik if (features) { 344805135f59SDavid Sterba btrfs_err(fs_info, 3449d5321a0fSQu Wenruo "cannot mount because of unsupported optional features (0x%llx)", 3450c1c9ff7cSGeert Uytterhoeven features); 3451f2b636e8SJosef Bacik err = -EINVAL; 3452141386e1SJosef Bacik goto fail_alloc; 3453f2b636e8SJosef Bacik } 3454f2b636e8SJosef Bacik 34555d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 34565d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 34570b246afaSJeff Mahoney if (fs_info->compress_type == BTRFS_COMPRESS_LZO) 3458a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 34595c1aab1dSNick Terrell else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) 34605c1aab1dSNick Terrell features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; 3461727011e0SChris Mason 3462727011e0SChris Mason /* 3463e26b04c4SNikolay Borisov * Flag our filesystem as having big metadata blocks if they are bigger 3464e26b04c4SNikolay Borisov * than the page size. 3465e26b04c4SNikolay Borisov */ 34666b769dacSNikolay Borisov if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) 3467e26b04c4SNikolay Borisov features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 3468727011e0SChris Mason 3469727011e0SChris Mason /* 3470bc3f116fSChris Mason * mixed block groups end up with duplicate but slightly offset 3471bc3f116fSChris Mason * extent buffers for the same range. It leads to corruptions 3472bc3f116fSChris Mason */ 3473bc3f116fSChris Mason if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 3474707e8a07SDavid Sterba (sectorsize != nodesize)) { 347505135f59SDavid Sterba btrfs_err(fs_info, 347605135f59SDavid Sterba "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", 347705135f59SDavid Sterba nodesize, sectorsize); 3478141386e1SJosef Bacik goto fail_alloc; 3479bc3f116fSChris Mason } 3480bc3f116fSChris Mason 3481ceda0864SMiao Xie /* 3482ceda0864SMiao Xie * Needn't use the lock because there is no other task which will 3483ceda0864SMiao Xie * update the flag. 3484ceda0864SMiao Xie */ 34855d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 34865d4f98a2SYan Zheng 3487f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 3488f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 3489bc98a42cSDavid Howells if (!sb_rdonly(sb) && features) { 349005135f59SDavid Sterba btrfs_err(fs_info, 3491d5321a0fSQu Wenruo "cannot mount read-write because of unsupported optional features (0x%llx)", 3492c1c9ff7cSGeert Uytterhoeven features); 3493f2b636e8SJosef Bacik err = -EINVAL; 3494141386e1SJosef Bacik goto fail_alloc; 3495f2b636e8SJosef Bacik } 3496dc4d3168SQu Wenruo /* 3497dc4d3168SQu Wenruo * We have unsupported RO compat features, although RO mounted, we 3498dc4d3168SQu Wenruo * should not cause any metadata write, including log replay. 3499dc4d3168SQu Wenruo * Or we could screw up whatever the new feature requires. 3500dc4d3168SQu Wenruo */ 3501dc4d3168SQu Wenruo if (unlikely(features && btrfs_super_log_root(disk_super) && 3502dc4d3168SQu Wenruo !btrfs_test_opt(fs_info, NOLOGREPLAY))) { 3503dc4d3168SQu Wenruo btrfs_err(fs_info, 3504dc4d3168SQu Wenruo "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay", 3505dc4d3168SQu Wenruo features); 3506dc4d3168SQu Wenruo err = -EINVAL; 3507dc4d3168SQu Wenruo goto fail_alloc; 3508dc4d3168SQu Wenruo } 3509dc4d3168SQu Wenruo 351061d92c32SChris Mason 35118481dd80SQu Wenruo if (sectorsize < PAGE_SIZE) { 35128481dd80SQu Wenruo struct btrfs_subpage_info *subpage_info; 35138481dd80SQu Wenruo 35149f73f1aeSQu Wenruo /* 35159f73f1aeSQu Wenruo * V1 space cache has some hardcoded PAGE_SIZE usage, and is 35169f73f1aeSQu Wenruo * going to be deprecated. 35179f73f1aeSQu Wenruo * 35189f73f1aeSQu Wenruo * Force to use v2 cache for subpage case. 35199f73f1aeSQu Wenruo */ 35209f73f1aeSQu Wenruo btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE); 35219f73f1aeSQu Wenruo btrfs_set_and_info(fs_info, FREE_SPACE_TREE, 35229f73f1aeSQu Wenruo "forcing free space tree for sector size %u with page size %lu", 35239f73f1aeSQu Wenruo sectorsize, PAGE_SIZE); 35249f73f1aeSQu Wenruo 352595ea0486SQu Wenruo btrfs_warn(fs_info, 352695ea0486SQu Wenruo "read-write for sector size %u with page size %lu is experimental", 35270bb3eb3eSQu Wenruo sectorsize, PAGE_SIZE); 35288481dd80SQu Wenruo subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL); 35298481dd80SQu Wenruo if (!subpage_info) 35308481dd80SQu Wenruo goto fail_alloc; 35318481dd80SQu Wenruo btrfs_init_subpage_info(subpage_info, sectorsize); 35328481dd80SQu Wenruo fs_info->subpage_info = subpage_info; 3533c8050b3bSQu Wenruo } 35340bb3eb3eSQu Wenruo 3535d21deec5SSu Yue ret = btrfs_init_workqueues(fs_info); 35362a458198SEric Sandeen if (ret) { 35372a458198SEric Sandeen err = ret; 35380dc3b84aSJosef Bacik goto fail_sb_buffer; 35390dc3b84aSJosef Bacik } 35404543df7eSChris Mason 35419e11ceeeSJan Kara sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); 35429e11ceeeSJan Kara sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); 35434575c9ccSChris Mason 3544a061fc8dSChris Mason sb->s_blocksize = sectorsize; 3545a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 3546de37aa51SNikolay Borisov memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); 3547db94535dSChris Mason 3548925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 35496bccf3abSJeff Mahoney ret = btrfs_read_sys_array(fs_info); 3550925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 355184eed90fSChris Mason if (ret) { 355205135f59SDavid Sterba btrfs_err(fs_info, "failed to read the system array: %d", ret); 35535d4f98a2SYan Zheng goto fail_sb_buffer; 355484eed90fSChris Mason } 35550b86a832SChris Mason 355684234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 3557581c1760SQu Wenruo level = btrfs_super_chunk_root_level(disk_super); 3558bd676446SJosef Bacik ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super), 3559bd676446SJosef Bacik generation, level); 3560bd676446SJosef Bacik if (ret) { 356105135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk root"); 3562af31f5e5SChris Mason goto fail_tree_roots; 356383121942SDavid Woodhouse } 35640b86a832SChris Mason 3565e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 3566c4ac7541SDavid Sterba offsetof(struct btrfs_header, chunk_tree_uuid), 3567c4ac7541SDavid Sterba BTRFS_UUID_SIZE); 3568e17cade2SChris Mason 35695b4aacefSJeff Mahoney ret = btrfs_read_chunk_tree(fs_info); 35702b82032cSYan Zheng if (ret) { 357105135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk tree: %d", ret); 3572af31f5e5SChris Mason goto fail_tree_roots; 35732b82032cSYan Zheng } 35740b86a832SChris Mason 35758dabb742SStefan Behrens /* 3576bacce86aSAnand Jain * At this point we know all the devices that make this filesystem, 3577bacce86aSAnand Jain * including the seed devices but we don't know yet if the replace 3578bacce86aSAnand Jain * target is required. So free devices that are not part of this 35791a9fd417SDavid Sterba * filesystem but skip the replace target device which is checked 3580bacce86aSAnand Jain * below in btrfs_init_dev_replace(). 35818dabb742SStefan Behrens */ 3582bacce86aSAnand Jain btrfs_free_extra_devids(fs_devices); 3583d24fa5c1SAnand Jain if (!fs_devices->latest_dev->bdev) { 358405135f59SDavid Sterba btrfs_err(fs_info, "failed to read devices"); 3585a6b0d5c8SChris Mason goto fail_tree_roots; 3586a6b0d5c8SChris Mason } 3587a6b0d5c8SChris Mason 3588b8522a1eSNikolay Borisov ret = init_tree_roots(fs_info); 35894bbcaa64SEric Sandeen if (ret) 3590b8522a1eSNikolay Borisov goto fail_tree_roots; 35918929ecfaSYan, Zheng 359275ec1db8SJosef Bacik /* 359373651042SNaohiro Aota * Get zone type information of zoned block devices. This will also 359473651042SNaohiro Aota * handle emulation of a zoned filesystem if a regular device has the 359573651042SNaohiro Aota * zoned incompat feature flag set. 359673651042SNaohiro Aota */ 359773651042SNaohiro Aota ret = btrfs_get_dev_zone_info_all_devices(fs_info); 359873651042SNaohiro Aota if (ret) { 359973651042SNaohiro Aota btrfs_err(fs_info, 360073651042SNaohiro Aota "zoned: failed to read device zone info: %d", 360173651042SNaohiro Aota ret); 360273651042SNaohiro Aota goto fail_block_groups; 360373651042SNaohiro Aota } 360473651042SNaohiro Aota 360573651042SNaohiro Aota /* 360675ec1db8SJosef Bacik * If we have a uuid root and we're not being told to rescan we need to 360775ec1db8SJosef Bacik * check the generation here so we can set the 360875ec1db8SJosef Bacik * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the 360975ec1db8SJosef Bacik * transaction during a balance or the log replay without updating the 361075ec1db8SJosef Bacik * uuid generation, and then if we crash we would rescan the uuid tree, 361175ec1db8SJosef Bacik * even though it was perfectly fine. 361275ec1db8SJosef Bacik */ 361375ec1db8SJosef Bacik if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) && 361475ec1db8SJosef Bacik fs_info->generation == btrfs_super_uuid_tree_generation(disk_super)) 361575ec1db8SJosef Bacik set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 361675ec1db8SJosef Bacik 3617cf90d884SQu Wenruo ret = btrfs_verify_dev_extents(fs_info); 3618cf90d884SQu Wenruo if (ret) { 3619cf90d884SQu Wenruo btrfs_err(fs_info, 3620cf90d884SQu Wenruo "failed to verify dev extents against chunks: %d", 3621cf90d884SQu Wenruo ret); 3622cf90d884SQu Wenruo goto fail_block_groups; 3623cf90d884SQu Wenruo } 362468310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 362568310a5eSIlya Dryomov if (ret) { 362605135f59SDavid Sterba btrfs_err(fs_info, "failed to recover balance: %d", ret); 362768310a5eSIlya Dryomov goto fail_block_groups; 362868310a5eSIlya Dryomov } 362968310a5eSIlya Dryomov 3630733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 3631733f4fbbSStefan Behrens if (ret) { 363205135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_stats: %d", ret); 3633733f4fbbSStefan Behrens goto fail_block_groups; 3634733f4fbbSStefan Behrens } 3635733f4fbbSStefan Behrens 36368dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 36378dabb742SStefan Behrens if (ret) { 363805135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_replace: %d", ret); 36398dabb742SStefan Behrens goto fail_block_groups; 36408dabb742SStefan Behrens } 36418dabb742SStefan Behrens 3642b70f5097SNaohiro Aota ret = btrfs_check_zoned_mode(fs_info); 3643b70f5097SNaohiro Aota if (ret) { 3644b70f5097SNaohiro Aota btrfs_err(fs_info, "failed to initialize zoned mode: %d", 3645b70f5097SNaohiro Aota ret); 3646b70f5097SNaohiro Aota goto fail_block_groups; 3647b70f5097SNaohiro Aota } 3648b70f5097SNaohiro Aota 3649c6761a9eSAnand Jain ret = btrfs_sysfs_add_fsid(fs_devices); 3650b7c35e81SAnand Jain if (ret) { 365105135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", 365205135f59SDavid Sterba ret); 3653b7c35e81SAnand Jain goto fail_block_groups; 3654b7c35e81SAnand Jain } 3655b7c35e81SAnand Jain 365696f3136eSAnand Jain ret = btrfs_sysfs_add_mounted(fs_info); 36575ac1d209SJeff Mahoney if (ret) { 365805135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); 3659b7c35e81SAnand Jain goto fail_fsdev_sysfs; 36605ac1d209SJeff Mahoney } 36615ac1d209SJeff Mahoney 3662c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 3663c59021f8Sliubo if (ret) { 366405135f59SDavid Sterba btrfs_err(fs_info, "failed to initialize space info: %d", ret); 36652365dd3cSAnand Jain goto fail_sysfs; 3666c59021f8Sliubo } 3667c59021f8Sliubo 36685b4aacefSJeff Mahoney ret = btrfs_read_block_groups(fs_info); 36691b1d1f66SJosef Bacik if (ret) { 367005135f59SDavid Sterba btrfs_err(fs_info, "failed to read block groups: %d", ret); 36712365dd3cSAnand Jain goto fail_sysfs; 36721b1d1f66SJosef Bacik } 36734330e183SQu Wenruo 367416beac87SNaohiro Aota btrfs_free_zone_cache(fs_info); 367516beac87SNaohiro Aota 36765c78a5e7SAnand Jain if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices && 36775c78a5e7SAnand Jain !btrfs_check_rw_degradable(fs_info, NULL)) { 367805135f59SDavid Sterba btrfs_warn(fs_info, 367952042d8eSAndrea Gelmini "writable mount is not allowed due to too many missing devices"); 36802365dd3cSAnand Jain goto fail_sysfs; 3681292fd7fcSStefan Behrens } 36829078a3e1SChris Mason 368333c44184SJosef Bacik fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info, 3684a74a4b97SChris Mason "btrfs-cleaner"); 368557506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 36862365dd3cSAnand Jain goto fail_sysfs; 3687a74a4b97SChris Mason 3688a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 3689a74a4b97SChris Mason tree_root, 3690a74a4b97SChris Mason "btrfs-transaction"); 369157506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 36923f157a2fSChris Mason goto fail_cleaner; 3693a74a4b97SChris Mason 3694583b7231SHans van Kranenburg if (!btrfs_test_opt(fs_info, NOSSD) && 3695c289811cSChris Mason !fs_info->fs_devices->rotating) { 3696583b7231SHans van Kranenburg btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); 3697c289811cSChris Mason } 3698c289811cSChris Mason 3699572d9ab7SDavid Sterba /* 370001327610SNicholas D Steeves * Mount does not set all options immediately, we can do it now and do 3701572d9ab7SDavid Sterba * not have to wait for transaction commit 3702572d9ab7SDavid Sterba */ 3703572d9ab7SDavid Sterba btrfs_apply_pending_changes(fs_info); 37043818aea2SQu Wenruo 370521adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 37060b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { 37072ff7e61eSJeff Mahoney ret = btrfsic_mount(fs_info, fs_devices, 37080b246afaSJeff Mahoney btrfs_test_opt(fs_info, 3709cbeaae4fSDavid Sterba CHECK_INTEGRITY_DATA) ? 1 : 0, 371021adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 371121adbd5cSStefan Behrens if (ret) 371205135f59SDavid Sterba btrfs_warn(fs_info, 371305135f59SDavid Sterba "failed to initialize integrity check module: %d", 371405135f59SDavid Sterba ret); 371521adbd5cSStefan Behrens } 371621adbd5cSStefan Behrens #endif 3717bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 3718bcef60f2SArne Jansen if (ret) 3719bcef60f2SArne Jansen goto fail_trans_kthread; 372021adbd5cSStefan Behrens 3721fd708b81SJosef Bacik if (btrfs_build_ref_tree(fs_info)) 3722fd708b81SJosef Bacik btrfs_err(fs_info, "couldn't build ref tree"); 3723fd708b81SJosef Bacik 372496da0919SQu Wenruo /* do not make disk changes in broken FS or nologreplay is given */ 372596da0919SQu Wenruo if (btrfs_super_log_root(disk_super) != 0 && 37260b246afaSJeff Mahoney !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 3727e8294f2fSDavid Sterba btrfs_info(fs_info, "start tree-log replay"); 372863443bf5SEric Sandeen ret = btrfs_replay_log(fs_info, fs_devices); 372979787eaaSJeff Mahoney if (ret) { 373063443bf5SEric Sandeen err = ret; 373128c16cbbSWang Shilong goto fail_qgroup; 3732e556ce2cSYan Zheng } 3733e02119d5SChris Mason } 37341a40e23bSZheng Yan 373556e9357aSDavid Sterba fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true); 37363140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 37373140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 3738f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read fs tree: %d", err); 3739315bf8efSJosef Bacik fs_info->fs_root = NULL; 3740bcef60f2SArne Jansen goto fail_qgroup; 37413140c9a3SDan Carpenter } 3742c289811cSChris Mason 3743bc98a42cSDavid Howells if (sb_rdonly(sb)) 37448cd29088SBoris Burkov goto clear_oneshot; 37452b6ba629SIlya Dryomov 374644c0ca21SBoris Burkov ret = btrfs_start_pre_rw_mount(fs_info); 37472b6ba629SIlya Dryomov if (ret) { 37486bccf3abSJeff Mahoney close_ctree(fs_info); 37492b6ba629SIlya Dryomov return ret; 3750e3acc2a6SJosef Bacik } 3751b0643e59SDennis Zhou btrfs_discard_resume(fs_info); 3752b382a324SJan Schmidt 375344c0ca21SBoris Burkov if (fs_info->uuid_root && 375444c0ca21SBoris Burkov (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || 375544c0ca21SBoris Burkov fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) { 375605135f59SDavid Sterba btrfs_info(fs_info, "checking UUID tree"); 375770f80175SStefan Behrens ret = btrfs_check_uuid_tree(fs_info); 375870f80175SStefan Behrens if (ret) { 375905135f59SDavid Sterba btrfs_warn(fs_info, 376005135f59SDavid Sterba "failed to check the UUID tree: %d", ret); 37616bccf3abSJeff Mahoney close_ctree(fs_info); 376270f80175SStefan Behrens return ret; 376370f80175SStefan Behrens } 3764f7a81ea4SStefan Behrens } 376594846229SBoris Burkov 3766afcdd129SJosef Bacik set_bit(BTRFS_FS_OPEN, &fs_info->flags); 376747ab2a6cSJosef Bacik 3768b4be6aefSJosef Bacik /* Kick the cleaner thread so it'll start deleting snapshots. */ 3769b4be6aefSJosef Bacik if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags)) 3770b4be6aefSJosef Bacik wake_up_process(fs_info->cleaner_kthread); 3771b4be6aefSJosef Bacik 37728cd29088SBoris Burkov clear_oneshot: 37738cd29088SBoris Burkov btrfs_clear_oneshot_options(fs_info); 3774ad2b2c80SAl Viro return 0; 377539279cc3SChris Mason 3776bcef60f2SArne Jansen fail_qgroup: 3777bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 37787c2ca468SChris Mason fail_trans_kthread: 37797c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 37802ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 3781faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 37823f157a2fSChris Mason fail_cleaner: 3783a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 37847c2ca468SChris Mason 37857c2ca468SChris Mason /* 37867c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 37877c2ca468SChris Mason * kthreads 37887c2ca468SChris Mason */ 37897c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 37907c2ca468SChris Mason 37912365dd3cSAnand Jain fail_sysfs: 37926618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 37932365dd3cSAnand Jain 3794b7c35e81SAnand Jain fail_fsdev_sysfs: 3795b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3796b7c35e81SAnand Jain 37971b1d1f66SJosef Bacik fail_block_groups: 379854067ae9SJosef Bacik btrfs_put_block_group_cache(fs_info); 3799af31f5e5SChris Mason 3800af31f5e5SChris Mason fail_tree_roots: 38019e3aa805SJosef Bacik if (fs_info->data_reloc_root) 38029e3aa805SJosef Bacik btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root); 38034273eaffSAnand Jain free_root_pointers(fs_info, true); 38042b8195bbSMiao Xie invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3805af31f5e5SChris Mason 380639279cc3SChris Mason fail_sb_buffer: 38077abadb64SLiu Bo btrfs_stop_all_workers(fs_info); 38085cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 380916cdcec7SMiao Xie fail_alloc: 3810586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 3811586e46e2SIlya Dryomov 38124543df7eSChris Mason iput(fs_info->btree_inode); 38137e662854SQinghuang Feng fail: 3814586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 3815ad2b2c80SAl Viro return err; 3816eb60ceacSChris Mason } 3817663faf9fSMasami Hiramatsu ALLOW_ERROR_INJECTION(open_ctree, ERRNO); 3818eb60ceacSChris Mason 3819314b6dd0SJohannes Thumshirn static void btrfs_end_super_write(struct bio *bio) 3820f2984462SChris Mason { 3821314b6dd0SJohannes Thumshirn struct btrfs_device *device = bio->bi_private; 3822314b6dd0SJohannes Thumshirn struct bio_vec *bvec; 3823314b6dd0SJohannes Thumshirn struct bvec_iter_all iter_all; 3824314b6dd0SJohannes Thumshirn struct page *page; 3825442a4f63SStefan Behrens 3826314b6dd0SJohannes Thumshirn bio_for_each_segment_all(bvec, bio, iter_all) { 3827314b6dd0SJohannes Thumshirn page = bvec->bv_page; 3828314b6dd0SJohannes Thumshirn 3829314b6dd0SJohannes Thumshirn if (bio->bi_status) { 3830fb456252SJeff Mahoney btrfs_warn_rl_in_rcu(device->fs_info, 3831314b6dd0SJohannes Thumshirn "lost page write due to IO error on %s (%d)", 3832314b6dd0SJohannes Thumshirn rcu_str_deref(device->name), 3833314b6dd0SJohannes Thumshirn blk_status_to_errno(bio->bi_status)); 3834314b6dd0SJohannes Thumshirn ClearPageUptodate(page); 3835314b6dd0SJohannes Thumshirn SetPageError(page); 3836314b6dd0SJohannes Thumshirn btrfs_dev_stat_inc_and_print(device, 3837314b6dd0SJohannes Thumshirn BTRFS_DEV_STAT_WRITE_ERRS); 3838314b6dd0SJohannes Thumshirn } else { 3839314b6dd0SJohannes Thumshirn SetPageUptodate(page); 3840f2984462SChris Mason } 3841314b6dd0SJohannes Thumshirn 3842314b6dd0SJohannes Thumshirn put_page(page); 3843314b6dd0SJohannes Thumshirn unlock_page(page); 3844314b6dd0SJohannes Thumshirn } 3845314b6dd0SJohannes Thumshirn 3846314b6dd0SJohannes Thumshirn bio_put(bio); 3847f2984462SChris Mason } 3848f2984462SChris Mason 38498f32380dSJohannes Thumshirn struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev, 3850a05d3c91SQu Wenruo int copy_num, bool drop_cache) 385129c36d72SAnand Jain { 385229c36d72SAnand Jain struct btrfs_super_block *super; 38538f32380dSJohannes Thumshirn struct page *page; 385412659251SNaohiro Aota u64 bytenr, bytenr_orig; 38558f32380dSJohannes Thumshirn struct address_space *mapping = bdev->bd_inode->i_mapping; 385612659251SNaohiro Aota int ret; 385729c36d72SAnand Jain 385812659251SNaohiro Aota bytenr_orig = btrfs_sb_offset(copy_num); 385912659251SNaohiro Aota ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr); 386012659251SNaohiro Aota if (ret == -ENOENT) 386112659251SNaohiro Aota return ERR_PTR(-EINVAL); 386212659251SNaohiro Aota else if (ret) 386312659251SNaohiro Aota return ERR_PTR(ret); 386412659251SNaohiro Aota 3865cda00ebaSChristoph Hellwig if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev)) 38668f32380dSJohannes Thumshirn return ERR_PTR(-EINVAL); 386729c36d72SAnand Jain 3868a05d3c91SQu Wenruo if (drop_cache) { 3869a05d3c91SQu Wenruo /* This should only be called with the primary sb. */ 3870a05d3c91SQu Wenruo ASSERT(copy_num == 0); 3871a05d3c91SQu Wenruo 3872a05d3c91SQu Wenruo /* 3873a05d3c91SQu Wenruo * Drop the page of the primary superblock, so later read will 3874a05d3c91SQu Wenruo * always read from the device. 3875a05d3c91SQu Wenruo */ 3876a05d3c91SQu Wenruo invalidate_inode_pages2_range(mapping, 3877a05d3c91SQu Wenruo bytenr >> PAGE_SHIFT, 3878a05d3c91SQu Wenruo (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT); 3879a05d3c91SQu Wenruo } 3880a05d3c91SQu Wenruo 38818f32380dSJohannes Thumshirn page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS); 38828f32380dSJohannes Thumshirn if (IS_ERR(page)) 38838f32380dSJohannes Thumshirn return ERR_CAST(page); 388429c36d72SAnand Jain 38858f32380dSJohannes Thumshirn super = page_address(page); 388696c2e067SAnand Jain if (btrfs_super_magic(super) != BTRFS_MAGIC) { 388796c2e067SAnand Jain btrfs_release_disk_super(super); 388896c2e067SAnand Jain return ERR_PTR(-ENODATA); 388996c2e067SAnand Jain } 389096c2e067SAnand Jain 389112659251SNaohiro Aota if (btrfs_super_bytenr(super) != bytenr_orig) { 38928f32380dSJohannes Thumshirn btrfs_release_disk_super(super); 38938f32380dSJohannes Thumshirn return ERR_PTR(-EINVAL); 389429c36d72SAnand Jain } 389529c36d72SAnand Jain 38968f32380dSJohannes Thumshirn return super; 389729c36d72SAnand Jain } 389829c36d72SAnand Jain 389929c36d72SAnand Jain 39008f32380dSJohannes Thumshirn struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev) 3901a512bbf8SYan Zheng { 39028f32380dSJohannes Thumshirn struct btrfs_super_block *super, *latest = NULL; 3903a512bbf8SYan Zheng int i; 3904a512bbf8SYan Zheng u64 transid = 0; 3905a512bbf8SYan Zheng 3906a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 3907a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 3908a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 3909a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3910a512bbf8SYan Zheng */ 3911a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 3912a05d3c91SQu Wenruo super = btrfs_read_dev_one_super(bdev, i, false); 39138f32380dSJohannes Thumshirn if (IS_ERR(super)) 3914a512bbf8SYan Zheng continue; 3915a512bbf8SYan Zheng 3916a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 39178f32380dSJohannes Thumshirn if (latest) 39188f32380dSJohannes Thumshirn btrfs_release_disk_super(super); 39198f32380dSJohannes Thumshirn 39208f32380dSJohannes Thumshirn latest = super; 3921a512bbf8SYan Zheng transid = btrfs_super_generation(super); 3922a512bbf8SYan Zheng } 3923a512bbf8SYan Zheng } 392492fc03fbSAnand Jain 39258f32380dSJohannes Thumshirn return super; 3926a512bbf8SYan Zheng } 3927a512bbf8SYan Zheng 39284eedeb75SHisashi Hifumi /* 3929abbb3b8eSDavid Sterba * Write superblock @sb to the @device. Do not wait for completion, all the 3930314b6dd0SJohannes Thumshirn * pages we use for writing are locked. 39314eedeb75SHisashi Hifumi * 3932abbb3b8eSDavid Sterba * Write @max_mirrors copies of the superblock, where 0 means default that fit 3933abbb3b8eSDavid Sterba * the expected device size at commit time. Note that max_mirrors must be 3934abbb3b8eSDavid Sterba * same for write and wait phases. 39354eedeb75SHisashi Hifumi * 3936314b6dd0SJohannes Thumshirn * Return number of errors when page is not found or submission fails. 39374eedeb75SHisashi Hifumi */ 3938a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 3939abbb3b8eSDavid Sterba struct btrfs_super_block *sb, int max_mirrors) 3940a512bbf8SYan Zheng { 3941d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = device->fs_info; 3942314b6dd0SJohannes Thumshirn struct address_space *mapping = device->bdev->bd_inode->i_mapping; 3943d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 3944a512bbf8SYan Zheng int i; 3945a512bbf8SYan Zheng int errors = 0; 394612659251SNaohiro Aota int ret; 394712659251SNaohiro Aota u64 bytenr, bytenr_orig; 3948a512bbf8SYan Zheng 3949a512bbf8SYan Zheng if (max_mirrors == 0) 3950a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3951a512bbf8SYan Zheng 3952d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 3953d5178578SJohannes Thumshirn 3954a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 3955314b6dd0SJohannes Thumshirn struct page *page; 3956314b6dd0SJohannes Thumshirn struct bio *bio; 3957314b6dd0SJohannes Thumshirn struct btrfs_super_block *disk_super; 3958314b6dd0SJohannes Thumshirn 395912659251SNaohiro Aota bytenr_orig = btrfs_sb_offset(i); 396012659251SNaohiro Aota ret = btrfs_sb_log_location(device, i, WRITE, &bytenr); 396112659251SNaohiro Aota if (ret == -ENOENT) { 396212659251SNaohiro Aota continue; 396312659251SNaohiro Aota } else if (ret < 0) { 396412659251SNaohiro Aota btrfs_err(device->fs_info, 396512659251SNaohiro Aota "couldn't get super block location for mirror %d", 396612659251SNaohiro Aota i); 396712659251SNaohiro Aota errors++; 396812659251SNaohiro Aota continue; 396912659251SNaohiro Aota } 3970935e5cc9SMiao Xie if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3971935e5cc9SMiao Xie device->commit_total_bytes) 3972a512bbf8SYan Zheng break; 3973a512bbf8SYan Zheng 397412659251SNaohiro Aota btrfs_set_super_bytenr(sb, bytenr_orig); 3975a512bbf8SYan Zheng 3976fd08001fSEric Biggers crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE, 3977fd08001fSEric Biggers BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, 3978fd08001fSEric Biggers sb->csum); 3979a512bbf8SYan Zheng 3980314b6dd0SJohannes Thumshirn page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT, 3981314b6dd0SJohannes Thumshirn GFP_NOFS); 3982314b6dd0SJohannes Thumshirn if (!page) { 3983fb456252SJeff Mahoney btrfs_err(device->fs_info, 3984314b6dd0SJohannes Thumshirn "couldn't get super block page for bytenr %llu", 3985f14d104dSDavid Sterba bytenr); 3986634554dcSJosef Bacik errors++; 3987634554dcSJosef Bacik continue; 3988634554dcSJosef Bacik } 3989634554dcSJosef Bacik 3990314b6dd0SJohannes Thumshirn /* Bump the refcount for wait_dev_supers() */ 3991314b6dd0SJohannes Thumshirn get_page(page); 3992a512bbf8SYan Zheng 3993314b6dd0SJohannes Thumshirn disk_super = page_address(page); 3994314b6dd0SJohannes Thumshirn memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE); 3995a512bbf8SYan Zheng 3996387125fcSChris Mason /* 3997314b6dd0SJohannes Thumshirn * Directly use bios here instead of relying on the page cache 3998314b6dd0SJohannes Thumshirn * to do I/O, so we don't lose the ability to do integrity 3999314b6dd0SJohannes Thumshirn * checking. 4000387125fcSChris Mason */ 400107888c66SChristoph Hellwig bio = bio_alloc(device->bdev, 1, 400207888c66SChristoph Hellwig REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO, 400307888c66SChristoph Hellwig GFP_NOFS); 4004314b6dd0SJohannes Thumshirn bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT; 4005314b6dd0SJohannes Thumshirn bio->bi_private = device; 4006314b6dd0SJohannes Thumshirn bio->bi_end_io = btrfs_end_super_write; 4007314b6dd0SJohannes Thumshirn __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE, 4008314b6dd0SJohannes Thumshirn offset_in_page(bytenr)); 4009314b6dd0SJohannes Thumshirn 4010314b6dd0SJohannes Thumshirn /* 4011314b6dd0SJohannes Thumshirn * We FUA only the first super block. The others we allow to 4012314b6dd0SJohannes Thumshirn * go down lazy and there's a short window where the on-disk 4013314b6dd0SJohannes Thumshirn * copies might still contain the older version. 4014314b6dd0SJohannes Thumshirn */ 40151b9e619cSOmar Sandoval if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) 4016314b6dd0SJohannes Thumshirn bio->bi_opf |= REQ_FUA; 4017314b6dd0SJohannes Thumshirn 401858ff51f1SChristoph Hellwig btrfsic_check_bio(bio); 401958ff51f1SChristoph Hellwig submit_bio(bio); 40208376d9e1SNaohiro Aota 40218376d9e1SNaohiro Aota if (btrfs_advance_sb_log(device, i)) 40228376d9e1SNaohiro Aota errors++; 4023a512bbf8SYan Zheng } 4024a512bbf8SYan Zheng return errors < i ? 0 : -1; 4025a512bbf8SYan Zheng } 4026a512bbf8SYan Zheng 4027387125fcSChris Mason /* 4028abbb3b8eSDavid Sterba * Wait for write completion of superblocks done by write_dev_supers, 4029abbb3b8eSDavid Sterba * @max_mirrors same for write and wait phases. 4030abbb3b8eSDavid Sterba * 4031314b6dd0SJohannes Thumshirn * Return number of errors when page is not found or not marked up to 4032abbb3b8eSDavid Sterba * date. 4033abbb3b8eSDavid Sterba */ 4034abbb3b8eSDavid Sterba static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) 4035abbb3b8eSDavid Sterba { 4036abbb3b8eSDavid Sterba int i; 4037abbb3b8eSDavid Sterba int errors = 0; 4038b6a535faSHoward McLauchlan bool primary_failed = false; 403912659251SNaohiro Aota int ret; 4040abbb3b8eSDavid Sterba u64 bytenr; 4041abbb3b8eSDavid Sterba 4042abbb3b8eSDavid Sterba if (max_mirrors == 0) 4043abbb3b8eSDavid Sterba max_mirrors = BTRFS_SUPER_MIRROR_MAX; 4044abbb3b8eSDavid Sterba 4045abbb3b8eSDavid Sterba for (i = 0; i < max_mirrors; i++) { 4046314b6dd0SJohannes Thumshirn struct page *page; 4047314b6dd0SJohannes Thumshirn 404812659251SNaohiro Aota ret = btrfs_sb_log_location(device, i, READ, &bytenr); 404912659251SNaohiro Aota if (ret == -ENOENT) { 405012659251SNaohiro Aota break; 405112659251SNaohiro Aota } else if (ret < 0) { 405212659251SNaohiro Aota errors++; 405312659251SNaohiro Aota if (i == 0) 405412659251SNaohiro Aota primary_failed = true; 405512659251SNaohiro Aota continue; 405612659251SNaohiro Aota } 4057abbb3b8eSDavid Sterba if (bytenr + BTRFS_SUPER_INFO_SIZE >= 4058abbb3b8eSDavid Sterba device->commit_total_bytes) 4059abbb3b8eSDavid Sterba break; 4060abbb3b8eSDavid Sterba 4061314b6dd0SJohannes Thumshirn page = find_get_page(device->bdev->bd_inode->i_mapping, 4062314b6dd0SJohannes Thumshirn bytenr >> PAGE_SHIFT); 4063314b6dd0SJohannes Thumshirn if (!page) { 4064abbb3b8eSDavid Sterba errors++; 4065b6a535faSHoward McLauchlan if (i == 0) 4066b6a535faSHoward McLauchlan primary_failed = true; 4067abbb3b8eSDavid Sterba continue; 4068abbb3b8eSDavid Sterba } 4069314b6dd0SJohannes Thumshirn /* Page is submitted locked and unlocked once the IO completes */ 4070314b6dd0SJohannes Thumshirn wait_on_page_locked(page); 4071314b6dd0SJohannes Thumshirn if (PageError(page)) { 4072abbb3b8eSDavid Sterba errors++; 4073b6a535faSHoward McLauchlan if (i == 0) 4074b6a535faSHoward McLauchlan primary_failed = true; 4075b6a535faSHoward McLauchlan } 4076abbb3b8eSDavid Sterba 4077314b6dd0SJohannes Thumshirn /* Drop our reference */ 4078314b6dd0SJohannes Thumshirn put_page(page); 4079abbb3b8eSDavid Sterba 4080314b6dd0SJohannes Thumshirn /* Drop the reference from the writing run */ 4081314b6dd0SJohannes Thumshirn put_page(page); 4082abbb3b8eSDavid Sterba } 4083abbb3b8eSDavid Sterba 4084b6a535faSHoward McLauchlan /* log error, force error return */ 4085b6a535faSHoward McLauchlan if (primary_failed) { 4086b6a535faSHoward McLauchlan btrfs_err(device->fs_info, "error writing primary super block to device %llu", 4087b6a535faSHoward McLauchlan device->devid); 4088b6a535faSHoward McLauchlan return -1; 4089b6a535faSHoward McLauchlan } 4090b6a535faSHoward McLauchlan 4091abbb3b8eSDavid Sterba return errors < i ? 0 : -1; 4092abbb3b8eSDavid Sterba } 4093abbb3b8eSDavid Sterba 4094abbb3b8eSDavid Sterba /* 4095387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 4096387125fcSChris Mason * for the barrier when it is done 4097387125fcSChris Mason */ 40984246a0b6SChristoph Hellwig static void btrfs_end_empty_barrier(struct bio *bio) 4099387125fcSChris Mason { 4100f9e69aa9SChristoph Hellwig bio_uninit(bio); 4101387125fcSChris Mason complete(bio->bi_private); 4102387125fcSChris Mason } 4103387125fcSChris Mason 4104387125fcSChris Mason /* 41054fc6441aSAnand Jain * Submit a flush request to the device if it supports it. Error handling is 41064fc6441aSAnand Jain * done in the waiting counterpart. 4107387125fcSChris Mason */ 41084fc6441aSAnand Jain static void write_dev_flush(struct btrfs_device *device) 4109387125fcSChris Mason { 4110f9e69aa9SChristoph Hellwig struct bio *bio = &device->flush_bio; 4111387125fcSChris Mason 4112a91cf0ffSWang Yugui #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY 4113a91cf0ffSWang Yugui /* 4114a91cf0ffSWang Yugui * When a disk has write caching disabled, we skip submission of a bio 4115a91cf0ffSWang Yugui * with flush and sync requests before writing the superblock, since 4116a91cf0ffSWang Yugui * it's not needed. However when the integrity checker is enabled, this 4117a91cf0ffSWang Yugui * results in reports that there are metadata blocks referred by a 4118a91cf0ffSWang Yugui * superblock that were not properly flushed. So don't skip the bio 4119a91cf0ffSWang Yugui * submission only when the integrity checker is enabled for the sake 4120a91cf0ffSWang Yugui * of simplicity, since this is a debug tool and not meant for use in 4121a91cf0ffSWang Yugui * non-debug builds. 4122a91cf0ffSWang Yugui */ 412308e688fdSChristoph Hellwig if (!bdev_write_cache(device->bdev)) 41244fc6441aSAnand Jain return; 4125a91cf0ffSWang Yugui #endif 4126387125fcSChris Mason 4127f9e69aa9SChristoph Hellwig bio_init(bio, device->bdev, NULL, 0, 4128f9e69aa9SChristoph Hellwig REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH); 4129387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 4130387125fcSChris Mason init_completion(&device->flush_wait); 4131387125fcSChris Mason bio->bi_private = &device->flush_wait; 4132387125fcSChris Mason 413358ff51f1SChristoph Hellwig btrfsic_check_bio(bio); 413458ff51f1SChristoph Hellwig submit_bio(bio); 41351c3063b6SAnand Jain set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 41364fc6441aSAnand Jain } 4137387125fcSChris Mason 41384fc6441aSAnand Jain /* 41394fc6441aSAnand Jain * If the flush bio has been submitted by write_dev_flush, wait for it. 41404fc6441aSAnand Jain */ 41418c27cb35SLinus Torvalds static blk_status_t wait_dev_flush(struct btrfs_device *device) 41424fc6441aSAnand Jain { 4143f9e69aa9SChristoph Hellwig struct bio *bio = &device->flush_bio; 41444fc6441aSAnand Jain 41451c3063b6SAnand Jain if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) 414658efbc9fSOmar Sandoval return BLK_STS_OK; 41474fc6441aSAnand Jain 41481c3063b6SAnand Jain clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 41492980d574SDavid Sterba wait_for_completion_io(&device->flush_wait); 41504fc6441aSAnand Jain 41518c27cb35SLinus Torvalds return bio->bi_status; 4152387125fcSChris Mason } 4153387125fcSChris Mason 4154d10b82feSQu Wenruo static int check_barrier_error(struct btrfs_fs_info *fs_info) 4155401b41e5SAnand Jain { 41566528b99dSAnand Jain if (!btrfs_check_rw_degradable(fs_info, NULL)) 4157401b41e5SAnand Jain return -EIO; 4158387125fcSChris Mason return 0; 4159387125fcSChris Mason } 4160387125fcSChris Mason 4161387125fcSChris Mason /* 4162387125fcSChris Mason * send an empty flush down to each device in parallel, 4163387125fcSChris Mason * then wait for them 4164387125fcSChris Mason */ 4165387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 4166387125fcSChris Mason { 4167387125fcSChris Mason struct list_head *head; 4168387125fcSChris Mason struct btrfs_device *dev; 41695af3e8ccSStefan Behrens int errors_wait = 0; 41704e4cbee9SChristoph Hellwig blk_status_t ret; 4171387125fcSChris Mason 41721538e6c5SDavid Sterba lockdep_assert_held(&info->fs_devices->device_list_mutex); 4173387125fcSChris Mason /* send down all the barriers */ 4174387125fcSChris Mason head = &info->fs_devices->devices; 41751538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4176e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 4177f88ba6a2SHidetoshi Seto continue; 4178cea7c8bfSAnand Jain if (!dev->bdev) 4179387125fcSChris Mason continue; 4180e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4181ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4182387125fcSChris Mason continue; 4183387125fcSChris Mason 41844fc6441aSAnand Jain write_dev_flush(dev); 418558efbc9fSOmar Sandoval dev->last_flush_error = BLK_STS_OK; 4186387125fcSChris Mason } 4187387125fcSChris Mason 4188387125fcSChris Mason /* wait for all the barriers */ 41891538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4190e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 4191f88ba6a2SHidetoshi Seto continue; 4192387125fcSChris Mason if (!dev->bdev) { 41935af3e8ccSStefan Behrens errors_wait++; 4194387125fcSChris Mason continue; 4195387125fcSChris Mason } 4196e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4197ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4198387125fcSChris Mason continue; 4199387125fcSChris Mason 42004fc6441aSAnand Jain ret = wait_dev_flush(dev); 4201401b41e5SAnand Jain if (ret) { 4202401b41e5SAnand Jain dev->last_flush_error = ret; 420366b4993eSDavid Sterba btrfs_dev_stat_inc_and_print(dev, 420466b4993eSDavid Sterba BTRFS_DEV_STAT_FLUSH_ERRS); 42055af3e8ccSStefan Behrens errors_wait++; 4206387125fcSChris Mason } 4207401b41e5SAnand Jain } 4208401b41e5SAnand Jain 4209cea7c8bfSAnand Jain if (errors_wait) { 4210401b41e5SAnand Jain /* 4211401b41e5SAnand Jain * At some point we need the status of all disks 4212401b41e5SAnand Jain * to arrive at the volume status. So error checking 4213401b41e5SAnand Jain * is being pushed to a separate loop. 4214401b41e5SAnand Jain */ 4215d10b82feSQu Wenruo return check_barrier_error(info); 4216401b41e5SAnand Jain } 4217387125fcSChris Mason return 0; 4218387125fcSChris Mason } 4219387125fcSChris Mason 4220943c6e99SZhao Lei int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 4221943c6e99SZhao Lei { 42228789f4feSZhao Lei int raid_type; 42238789f4feSZhao Lei int min_tolerated = INT_MAX; 4224943c6e99SZhao Lei 42258789f4feSZhao Lei if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 42268789f4feSZhao Lei (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 42278c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 42288789f4feSZhao Lei btrfs_raid_array[BTRFS_RAID_SINGLE]. 42298789f4feSZhao Lei tolerated_failures); 4230943c6e99SZhao Lei 42318789f4feSZhao Lei for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 42328789f4feSZhao Lei if (raid_type == BTRFS_RAID_SINGLE) 42338789f4feSZhao Lei continue; 423441a6e891SAnand Jain if (!(flags & btrfs_raid_array[raid_type].bg_flag)) 42358789f4feSZhao Lei continue; 42368c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 42378789f4feSZhao Lei btrfs_raid_array[raid_type]. 42388789f4feSZhao Lei tolerated_failures); 42398789f4feSZhao Lei } 4240943c6e99SZhao Lei 42418789f4feSZhao Lei if (min_tolerated == INT_MAX) { 4242ab8d0fc4SJeff Mahoney pr_warn("BTRFS: unknown raid flag: %llu", flags); 42438789f4feSZhao Lei min_tolerated = 0; 42448789f4feSZhao Lei } 42458789f4feSZhao Lei 42468789f4feSZhao Lei return min_tolerated; 4247943c6e99SZhao Lei } 4248943c6e99SZhao Lei 4249eece6a9cSDavid Sterba int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) 4250f2984462SChris Mason { 4251e5e9a520SChris Mason struct list_head *head; 4252f2984462SChris Mason struct btrfs_device *dev; 4253a061fc8dSChris Mason struct btrfs_super_block *sb; 4254f2984462SChris Mason struct btrfs_dev_item *dev_item; 4255f2984462SChris Mason int ret; 4256f2984462SChris Mason int do_barriers; 4257a236aed1SChris Mason int max_errors; 4258a236aed1SChris Mason int total_errors = 0; 4259a061fc8dSChris Mason u64 flags; 4260f2984462SChris Mason 42610b246afaSJeff Mahoney do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); 4262fed3b381SLiu Bo 4263fed3b381SLiu Bo /* 4264fed3b381SLiu Bo * max_mirrors == 0 indicates we're from commit_transaction, 4265fed3b381SLiu Bo * not from fsync where the tree roots in fs_info have not 4266fed3b381SLiu Bo * been consistent on disk. 4267fed3b381SLiu Bo */ 4268fed3b381SLiu Bo if (max_mirrors == 0) 42690b246afaSJeff Mahoney backup_super_roots(fs_info); 4270f2984462SChris Mason 42710b246afaSJeff Mahoney sb = fs_info->super_for_commit; 4272a061fc8dSChris Mason dev_item = &sb->dev_item; 4273e5e9a520SChris Mason 42740b246afaSJeff Mahoney mutex_lock(&fs_info->fs_devices->device_list_mutex); 42750b246afaSJeff Mahoney head = &fs_info->fs_devices->devices; 42760b246afaSJeff Mahoney max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; 4277387125fcSChris Mason 42785af3e8ccSStefan Behrens if (do_barriers) { 42790b246afaSJeff Mahoney ret = barrier_all_devices(fs_info); 42805af3e8ccSStefan Behrens if (ret) { 42815af3e8ccSStefan Behrens mutex_unlock( 42820b246afaSJeff Mahoney &fs_info->fs_devices->device_list_mutex); 42830b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 42845af3e8ccSStefan Behrens "errors while submitting device barriers."); 42855af3e8ccSStefan Behrens return ret; 42865af3e8ccSStefan Behrens } 42875af3e8ccSStefan Behrens } 4288387125fcSChris Mason 42891538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4290dfe25020SChris Mason if (!dev->bdev) { 4291dfe25020SChris Mason total_errors++; 4292dfe25020SChris Mason continue; 4293dfe25020SChris Mason } 4294e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4295ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4296dfe25020SChris Mason continue; 4297dfe25020SChris Mason 42982b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 4299a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 4300a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 43017df69d3eSMiao Xie btrfs_set_stack_device_total_bytes(dev_item, 4302935e5cc9SMiao Xie dev->commit_total_bytes); 4303ce7213c7SMiao Xie btrfs_set_stack_device_bytes_used(dev_item, 4304ce7213c7SMiao Xie dev->commit_bytes_used); 4305a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 4306a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 4307a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 4308a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 43097239ff4bSNikolay Borisov memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, 43107239ff4bSNikolay Borisov BTRFS_FSID_SIZE); 4311a512bbf8SYan Zheng 4312a061fc8dSChris Mason flags = btrfs_super_flags(sb); 4313a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 4314f2984462SChris Mason 431575cb857dSQu Wenruo ret = btrfs_validate_write_super(fs_info, sb); 431675cb857dSQu Wenruo if (ret < 0) { 431775cb857dSQu Wenruo mutex_unlock(&fs_info->fs_devices->device_list_mutex); 431875cb857dSQu Wenruo btrfs_handle_fs_error(fs_info, -EUCLEAN, 431975cb857dSQu Wenruo "unexpected superblock corruption detected"); 432075cb857dSQu Wenruo return -EUCLEAN; 432175cb857dSQu Wenruo } 432275cb857dSQu Wenruo 4323abbb3b8eSDavid Sterba ret = write_dev_supers(dev, sb, max_mirrors); 4324a236aed1SChris Mason if (ret) 4325a236aed1SChris Mason total_errors++; 4326f2984462SChris Mason } 4327a236aed1SChris Mason if (total_errors > max_errors) { 43280b246afaSJeff Mahoney btrfs_err(fs_info, "%d errors while writing supers", 4329d397712bSChris Mason total_errors); 43300b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 433179787eaaSJeff Mahoney 43329d565ba4SStefan Behrens /* FUA is masked off if unsupported and can't be the reason */ 43330b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 43340b246afaSJeff Mahoney "%d errors while writing supers", 43350b246afaSJeff Mahoney total_errors); 43369d565ba4SStefan Behrens return -EIO; 4337a236aed1SChris Mason } 4338f2984462SChris Mason 4339a512bbf8SYan Zheng total_errors = 0; 43401538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4341dfe25020SChris Mason if (!dev->bdev) 4342dfe25020SChris Mason continue; 4343e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4344ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4345dfe25020SChris Mason continue; 4346dfe25020SChris Mason 4347abbb3b8eSDavid Sterba ret = wait_dev_supers(dev, max_mirrors); 4348a512bbf8SYan Zheng if (ret) 43491259ab75SChris Mason total_errors++; 4350f2984462SChris Mason } 43510b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 4352a236aed1SChris Mason if (total_errors > max_errors) { 43530b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 43540b246afaSJeff Mahoney "%d errors while writing supers", 43550b246afaSJeff Mahoney total_errors); 435679787eaaSJeff Mahoney return -EIO; 4357a236aed1SChris Mason } 4358f2984462SChris Mason return 0; 4359f2984462SChris Mason } 4360f2984462SChris Mason 4361cb517eabSMiao Xie /* Drop a fs root from the radix tree and free it. */ 4362cb517eabSMiao Xie void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 4363cb517eabSMiao Xie struct btrfs_root *root) 43642619ba1fSChris Mason { 43654785e24fSJosef Bacik bool drop_ref = false; 43664785e24fSJosef Bacik 4367fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4368fc7cbcd4SDavid Sterba radix_tree_delete(&fs_info->fs_roots_radix, 4369fc7cbcd4SDavid Sterba (unsigned long)root->root_key.objectid); 4370fc7cbcd4SDavid Sterba if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state)) 43714785e24fSJosef Bacik drop_ref = true; 4372fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 437376dda93cSYan, Zheng 437484961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 4375ef67963dSJosef Bacik ASSERT(root->log_root == NULL); 43761c1ea4f7SLiu Bo if (root->reloc_root) { 437700246528SJosef Bacik btrfs_put_root(root->reloc_root); 43781c1ea4f7SLiu Bo root->reloc_root = NULL; 43791c1ea4f7SLiu Bo } 43801c1ea4f7SLiu Bo } 43813321719eSLiu Bo 43824785e24fSJosef Bacik if (drop_ref) 438300246528SJosef Bacik btrfs_put_root(root); 43842619ba1fSChris Mason } 43852619ba1fSChris Mason 4386c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 4387c146afadSYan Zheng { 4388fc7cbcd4SDavid Sterba u64 root_objectid = 0; 4389fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 4390fc7cbcd4SDavid Sterba int i = 0; 439165d33fd7SQu Wenruo int err = 0; 4392fc7cbcd4SDavid Sterba unsigned int ret = 0; 4393c146afadSYan Zheng 4394c146afadSYan Zheng while (1) { 4395fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4396fc7cbcd4SDavid Sterba ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4397fc7cbcd4SDavid Sterba (void **)gang, root_objectid, 4398fc7cbcd4SDavid Sterba ARRAY_SIZE(gang)); 4399fc7cbcd4SDavid Sterba if (!ret) { 4400fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4401c146afadSYan Zheng break; 440265d33fd7SQu Wenruo } 4403fc7cbcd4SDavid Sterba root_objectid = gang[ret - 1]->root_key.objectid + 1; 440466b4ffd1SJosef Bacik 4405fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 4406fc7cbcd4SDavid Sterba /* Avoid to grab roots in dead_roots */ 4407fc7cbcd4SDavid Sterba if (btrfs_root_refs(&gang[i]->root_item) == 0) { 4408fc7cbcd4SDavid Sterba gang[i] = NULL; 440965d33fd7SQu Wenruo continue; 4410c146afadSYan Zheng } 4411fc7cbcd4SDavid Sterba /* grab all the search result for later use */ 4412fc7cbcd4SDavid Sterba gang[i] = btrfs_grab_root(gang[i]); 4413fc7cbcd4SDavid Sterba } 4414fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4415fc7cbcd4SDavid Sterba 4416fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 4417fc7cbcd4SDavid Sterba if (!gang[i]) 4418fc7cbcd4SDavid Sterba continue; 4419fc7cbcd4SDavid Sterba root_objectid = gang[i]->root_key.objectid; 4420fc7cbcd4SDavid Sterba err = btrfs_orphan_cleanup(gang[i]); 4421fc7cbcd4SDavid Sterba if (err) 4422fc7cbcd4SDavid Sterba break; 4423fc7cbcd4SDavid Sterba btrfs_put_root(gang[i]); 4424fc7cbcd4SDavid Sterba } 4425fc7cbcd4SDavid Sterba root_objectid++; 4426c146afadSYan Zheng } 442765d33fd7SQu Wenruo 4428fc7cbcd4SDavid Sterba /* release the uncleaned roots due to error */ 4429fc7cbcd4SDavid Sterba for (; i < ret; i++) { 4430fc7cbcd4SDavid Sterba if (gang[i]) 4431fc7cbcd4SDavid Sterba btrfs_put_root(gang[i]); 443265d33fd7SQu Wenruo } 443365d33fd7SQu Wenruo return err; 4434c146afadSYan Zheng } 4435c146afadSYan Zheng 44366bccf3abSJeff Mahoney int btrfs_commit_super(struct btrfs_fs_info *fs_info) 4437c146afadSYan Zheng { 44386bccf3abSJeff Mahoney struct btrfs_root *root = fs_info->tree_root; 4439c146afadSYan Zheng struct btrfs_trans_handle *trans; 4440c146afadSYan Zheng 44410b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 44422ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 44430b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 44440b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 4445c71bf099SYan, Zheng 4446c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 44470b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 44480b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4449c71bf099SYan, Zheng 44507a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 44513612b495STsutomu Itoh if (IS_ERR(trans)) 44523612b495STsutomu Itoh return PTR_ERR(trans); 44533a45bb20SJeff Mahoney return btrfs_commit_transaction(trans); 4454c146afadSYan Zheng } 4455c146afadSYan Zheng 445636c86a9eSQu Wenruo static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info) 445736c86a9eSQu Wenruo { 445836c86a9eSQu Wenruo struct btrfs_transaction *trans; 445936c86a9eSQu Wenruo struct btrfs_transaction *tmp; 446036c86a9eSQu Wenruo bool found = false; 446136c86a9eSQu Wenruo 446236c86a9eSQu Wenruo if (list_empty(&fs_info->trans_list)) 446336c86a9eSQu Wenruo return; 446436c86a9eSQu Wenruo 446536c86a9eSQu Wenruo /* 446636c86a9eSQu Wenruo * This function is only called at the very end of close_ctree(), 446736c86a9eSQu Wenruo * thus no other running transaction, no need to take trans_lock. 446836c86a9eSQu Wenruo */ 446936c86a9eSQu Wenruo ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags)); 447036c86a9eSQu Wenruo list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) { 447136c86a9eSQu Wenruo struct extent_state *cached = NULL; 447236c86a9eSQu Wenruo u64 dirty_bytes = 0; 447336c86a9eSQu Wenruo u64 cur = 0; 447436c86a9eSQu Wenruo u64 found_start; 447536c86a9eSQu Wenruo u64 found_end; 447636c86a9eSQu Wenruo 447736c86a9eSQu Wenruo found = true; 447836c86a9eSQu Wenruo while (!find_first_extent_bit(&trans->dirty_pages, cur, 447936c86a9eSQu Wenruo &found_start, &found_end, EXTENT_DIRTY, &cached)) { 448036c86a9eSQu Wenruo dirty_bytes += found_end + 1 - found_start; 448136c86a9eSQu Wenruo cur = found_end + 1; 448236c86a9eSQu Wenruo } 448336c86a9eSQu Wenruo btrfs_warn(fs_info, 448436c86a9eSQu Wenruo "transaction %llu (with %llu dirty metadata bytes) is not committed", 448536c86a9eSQu Wenruo trans->transid, dirty_bytes); 448636c86a9eSQu Wenruo btrfs_cleanup_one_transaction(trans, fs_info); 448736c86a9eSQu Wenruo 448836c86a9eSQu Wenruo if (trans == fs_info->running_transaction) 448936c86a9eSQu Wenruo fs_info->running_transaction = NULL; 449036c86a9eSQu Wenruo list_del_init(&trans->list); 449136c86a9eSQu Wenruo 449236c86a9eSQu Wenruo btrfs_put_transaction(trans); 449336c86a9eSQu Wenruo trace_btrfs_transaction_commit(fs_info); 449436c86a9eSQu Wenruo } 449536c86a9eSQu Wenruo ASSERT(!found); 449636c86a9eSQu Wenruo } 449736c86a9eSQu Wenruo 4498b105e927SDavid Sterba void __cold close_ctree(struct btrfs_fs_info *fs_info) 4499eb60ceacSChris Mason { 4500c146afadSYan Zheng int ret; 4501e089f05cSChris Mason 4502afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); 450331e70e52SFilipe Manana 450431e70e52SFilipe Manana /* 45058a1f1e3dSFilipe Manana * If we had UNFINISHED_DROPS we could still be processing them, so 45068a1f1e3dSFilipe Manana * clear that bit and wake up relocation so it can stop. 45078a1f1e3dSFilipe Manana * We must do this before stopping the block group reclaim task, because 45088a1f1e3dSFilipe Manana * at btrfs_relocate_block_group() we wait for this bit, and after the 45098a1f1e3dSFilipe Manana * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we 45108a1f1e3dSFilipe Manana * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will 45118a1f1e3dSFilipe Manana * return 1. 45128a1f1e3dSFilipe Manana */ 45138a1f1e3dSFilipe Manana btrfs_wake_unfinished_drop(fs_info); 45148a1f1e3dSFilipe Manana 45158a1f1e3dSFilipe Manana /* 451631e70e52SFilipe Manana * We may have the reclaim task running and relocating a data block group, 451731e70e52SFilipe Manana * in which case it may create delayed iputs. So stop it before we park 451831e70e52SFilipe Manana * the cleaner kthread otherwise we can get new delayed iputs after 451931e70e52SFilipe Manana * parking the cleaner, and that can make the async reclaim task to hang 452031e70e52SFilipe Manana * if it's waiting for delayed iputs to complete, since the cleaner is 452131e70e52SFilipe Manana * parked and can not run delayed iputs - this will make us hang when 452231e70e52SFilipe Manana * trying to stop the async reclaim task. 452331e70e52SFilipe Manana */ 452431e70e52SFilipe Manana cancel_work_sync(&fs_info->reclaim_bgs_work); 4525d6fd0ae2SOmar Sandoval /* 4526d6fd0ae2SOmar Sandoval * We don't want the cleaner to start new transactions, add more delayed 4527d6fd0ae2SOmar Sandoval * iputs, etc. while we're closing. We can't use kthread_stop() yet 4528d6fd0ae2SOmar Sandoval * because that frees the task_struct, and the transaction kthread might 4529d6fd0ae2SOmar Sandoval * still try to wake up the cleaner. 4530d6fd0ae2SOmar Sandoval */ 4531d6fd0ae2SOmar Sandoval kthread_park(fs_info->cleaner_kthread); 4532a2135011SChris Mason 45337343dd61SJustin Maggard /* wait for the qgroup rescan worker to stop */ 4534d06f23d6SJeff Mahoney btrfs_qgroup_wait_for_completion(fs_info, false); 45357343dd61SJustin Maggard 4536803b2f54SStefan Behrens /* wait for the uuid_scan task to finish */ 4537803b2f54SStefan Behrens down(&fs_info->uuid_tree_rescan_sem); 4538803b2f54SStefan Behrens /* avoid complains from lockdep et al., set sem back to initial state */ 4539803b2f54SStefan Behrens up(&fs_info->uuid_tree_rescan_sem); 4540803b2f54SStefan Behrens 4541837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 4542aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 4543837d5b6eSIlya Dryomov 45448dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 45458dabb742SStefan Behrens 4546aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 45474cb5300bSChris Mason 45484cb5300bSChris Mason /* wait for any defraggers to finish */ 45494cb5300bSChris Mason wait_event(fs_info->transaction_wait, 45504cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 45514cb5300bSChris Mason 45524cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 455326176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 45544cb5300bSChris Mason 4555a362bb86SFilipe Manana /* 4556a362bb86SFilipe Manana * After we parked the cleaner kthread, ordered extents may have 4557a362bb86SFilipe Manana * completed and created new delayed iputs. If one of the async reclaim 4558a362bb86SFilipe Manana * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we 4559a362bb86SFilipe Manana * can hang forever trying to stop it, because if a delayed iput is 4560a362bb86SFilipe Manana * added after it ran btrfs_run_delayed_iputs() and before it called 4561a362bb86SFilipe Manana * btrfs_wait_on_delayed_iputs(), it will hang forever since there is 4562a362bb86SFilipe Manana * no one else to run iputs. 4563a362bb86SFilipe Manana * 4564a362bb86SFilipe Manana * So wait for all ongoing ordered extents to complete and then run 4565a362bb86SFilipe Manana * delayed iputs. This works because once we reach this point no one 4566a362bb86SFilipe Manana * can either create new ordered extents nor create delayed iputs 4567a362bb86SFilipe Manana * through some other means. 4568a362bb86SFilipe Manana * 4569a362bb86SFilipe Manana * Also note that btrfs_wait_ordered_roots() is not safe here, because 4570a362bb86SFilipe Manana * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent, 4571a362bb86SFilipe Manana * but the delayed iput for the respective inode is made only when doing 4572a362bb86SFilipe Manana * the final btrfs_put_ordered_extent() (which must happen at 4573a362bb86SFilipe Manana * btrfs_finish_ordered_io() when we are unmounting). 4574a362bb86SFilipe Manana */ 4575a362bb86SFilipe Manana btrfs_flush_workqueue(fs_info->endio_write_workers); 4576a362bb86SFilipe Manana /* Ordered extents for free space inodes. */ 4577a362bb86SFilipe Manana btrfs_flush_workqueue(fs_info->endio_freespace_worker); 4578a362bb86SFilipe Manana btrfs_run_delayed_iputs(fs_info); 4579a362bb86SFilipe Manana 458021c7e756SMiao Xie cancel_work_sync(&fs_info->async_reclaim_work); 458157056740SJosef Bacik cancel_work_sync(&fs_info->async_data_reclaim_work); 4582576fa348SJosef Bacik cancel_work_sync(&fs_info->preempt_reclaim_work); 458321c7e756SMiao Xie 4584b0643e59SDennis Zhou /* Cancel or finish ongoing discard work */ 4585b0643e59SDennis Zhou btrfs_discard_cleanup(fs_info); 4586b0643e59SDennis Zhou 4587bc98a42cSDavid Howells if (!sb_rdonly(fs_info->sb)) { 4588e44163e1SJeff Mahoney /* 4589d6fd0ae2SOmar Sandoval * The cleaner kthread is stopped, so do one final pass over 4590d6fd0ae2SOmar Sandoval * unused block groups. 4591e44163e1SJeff Mahoney */ 45920b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 4593e44163e1SJeff Mahoney 4594f0cc2cd7SFilipe Manana /* 4595f0cc2cd7SFilipe Manana * There might be existing delayed inode workers still running 4596f0cc2cd7SFilipe Manana * and holding an empty delayed inode item. We must wait for 4597f0cc2cd7SFilipe Manana * them to complete first because they can create a transaction. 4598f0cc2cd7SFilipe Manana * This happens when someone calls btrfs_balance_delayed_items() 4599f0cc2cd7SFilipe Manana * and then a transaction commit runs the same delayed nodes 4600f0cc2cd7SFilipe Manana * before any delayed worker has done something with the nodes. 4601f0cc2cd7SFilipe Manana * We must wait for any worker here and not at transaction 4602f0cc2cd7SFilipe Manana * commit time since that could cause a deadlock. 4603f0cc2cd7SFilipe Manana * This is a very rare case. 4604f0cc2cd7SFilipe Manana */ 4605f0cc2cd7SFilipe Manana btrfs_flush_workqueue(fs_info->delayed_workers); 4606f0cc2cd7SFilipe Manana 46076bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 4608d397712bSChris Mason if (ret) 460904892340SEric Sandeen btrfs_err(fs_info, "commit super ret %d", ret); 4610c146afadSYan Zheng } 4611ed2ff2cbSChris Mason 461284961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) 46132ff7e61eSJeff Mahoney btrfs_error_commit_super(fs_info); 4614acce952bSliubo 4615e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 4616e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 46178929ecfaSYan, Zheng 4618e187831eSJosef Bacik ASSERT(list_empty(&fs_info->delayed_iputs)); 4619afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); 4620f25784b3SYan Zheng 46215958253cSQu Wenruo if (btrfs_check_quota_leak(fs_info)) { 46225958253cSQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 46235958253cSQu Wenruo btrfs_err(fs_info, "qgroup reserved space leaked"); 46245958253cSQu Wenruo } 46255958253cSQu Wenruo 462604892340SEric Sandeen btrfs_free_qgroup_config(fs_info); 4627fe816d0fSNikolay Borisov ASSERT(list_empty(&fs_info->delalloc_roots)); 4628bcef60f2SArne Jansen 4629963d678bSMiao Xie if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 463004892340SEric Sandeen btrfs_info(fs_info, "at unmount delalloc count %lld", 4631963d678bSMiao Xie percpu_counter_sum(&fs_info->delalloc_bytes)); 4632b0c68f8bSChris Mason } 463331153d81SYan Zheng 46345deb17e1SJosef Bacik if (percpu_counter_sum(&fs_info->ordered_bytes)) 46354297ff84SJosef Bacik btrfs_info(fs_info, "at unmount dio bytes count %lld", 46365deb17e1SJosef Bacik percpu_counter_sum(&fs_info->ordered_bytes)); 46374297ff84SJosef Bacik 46386618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 4639b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 46405ac1d209SJeff Mahoney 46411a4319ccSLiu Bo btrfs_put_block_group_cache(fs_info); 46421a4319ccSLiu Bo 4643de348ee0SWang Shilong /* 4644de348ee0SWang Shilong * we must make sure there is not any read request to 4645de348ee0SWang Shilong * submit after we stopping all workers. 4646de348ee0SWang Shilong */ 4647de348ee0SWang Shilong invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 464896192499SJosef Bacik btrfs_stop_all_workers(fs_info); 464996192499SJosef Bacik 46500a31daa4SFilipe Manana /* We shouldn't have any transaction open at this point */ 465136c86a9eSQu Wenruo warn_about_uncommitted_trans(fs_info); 46520a31daa4SFilipe Manana 4653afcdd129SJosef Bacik clear_bit(BTRFS_FS_OPEN, &fs_info->flags); 46544273eaffSAnand Jain free_root_pointers(fs_info, true); 46558c38938cSJosef Bacik btrfs_free_fs_roots(fs_info); 46569ad6b7bcSChris Mason 46574e19443dSJosef Bacik /* 46584e19443dSJosef Bacik * We must free the block groups after dropping the fs_roots as we could 46594e19443dSJosef Bacik * have had an IO error and have left over tree log blocks that aren't 46604e19443dSJosef Bacik * cleaned up until the fs roots are freed. This makes the block group 46614e19443dSJosef Bacik * accounting appear to be wrong because there's pending reserved bytes, 46624e19443dSJosef Bacik * so make sure we do the block group cleanup afterwards. 46634e19443dSJosef Bacik */ 46644e19443dSJosef Bacik btrfs_free_block_groups(fs_info); 46654e19443dSJosef Bacik 466613e6c37bSJosef Bacik iput(fs_info->btree_inode); 4667d6bfde87SChris Mason 466821adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 46690b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) 46702ff7e61eSJeff Mahoney btrfsic_unmount(fs_info->fs_devices); 467121adbd5cSStefan Behrens #endif 467221adbd5cSStefan Behrens 46730b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 467468c94e55SNikolay Borisov btrfs_close_devices(fs_info->fs_devices); 4675eb60ceacSChris Mason } 4676eb60ceacSChris Mason 4677b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 4678b9fab919SChris Mason int atomic) 4679ccd467d6SChris Mason { 46801259ab75SChris Mason int ret; 4681727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 46821259ab75SChris Mason 46830b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 46841259ab75SChris Mason if (!ret) 46851259ab75SChris Mason return ret; 46861259ab75SChris Mason 46871259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 4688b9fab919SChris Mason parent_transid, atomic); 4689b9fab919SChris Mason if (ret == -EAGAIN) 4690b9fab919SChris Mason return ret; 46911259ab75SChris Mason return !ret; 46925f39d397SChris Mason } 46936702ed49SChris Mason 46945f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 46955f39d397SChris Mason { 46962f4d60dfSQu Wenruo struct btrfs_fs_info *fs_info = buf->fs_info; 46975f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 4698b9473439SChris Mason int was_dirty; 4699b4ce94deSChris Mason 470006ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 470106ea65a3SJosef Bacik /* 470206ea65a3SJosef Bacik * This is a fast path so only do this check if we have sanity tests 470352042d8eSAndrea Gelmini * enabled. Normal people shouldn't be using unmapped buffers as dirty 470406ea65a3SJosef Bacik * outside of the sanity tests. 470506ea65a3SJosef Bacik */ 4706b0132a3bSNikolay Borisov if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) 470706ea65a3SJosef Bacik return; 470806ea65a3SJosef Bacik #endif 470949d0c642SFilipe Manana btrfs_assert_tree_write_locked(buf); 47100b246afaSJeff Mahoney if (transid != fs_info->generation) 47115d163e0eSJeff Mahoney WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", 47120b246afaSJeff Mahoney buf->start, transid, fs_info->generation); 47130b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 4714e2d84521SMiao Xie if (!was_dirty) 4715104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 4716e2d84521SMiao Xie buf->len, 47170b246afaSJeff Mahoney fs_info->dirty_metadata_batch); 47181f21ef0aSFilipe Manana #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 471969fc6cbbSQu Wenruo /* 472069fc6cbbSQu Wenruo * Since btrfs_mark_buffer_dirty() can be called with item pointer set 472169fc6cbbSQu Wenruo * but item data not updated. 472269fc6cbbSQu Wenruo * So here we should only check item pointers, not item data. 472369fc6cbbSQu Wenruo */ 472469fc6cbbSQu Wenruo if (btrfs_header_level(buf) == 0 && 4725cfdaad5eSDavid Sterba btrfs_check_leaf_relaxed(buf)) { 4726a4f78750SDavid Sterba btrfs_print_leaf(buf); 47271f21ef0aSFilipe Manana ASSERT(0); 47281f21ef0aSFilipe Manana } 47291f21ef0aSFilipe Manana #endif 4730eb60ceacSChris Mason } 4731eb60ceacSChris Mason 47322ff7e61eSJeff Mahoney static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, 4733b53d3f5dSLiu Bo int flush_delayed) 473435b7e476SChris Mason { 4735188de649SChris Mason /* 4736188de649SChris Mason * looks as though older kernels can get into trouble with 4737188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 4738188de649SChris Mason */ 4739e2d84521SMiao Xie int ret; 4740d6bfde87SChris Mason 47416933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 4742d6bfde87SChris Mason return; 4743d6bfde87SChris Mason 4744b53d3f5dSLiu Bo if (flush_delayed) 47452ff7e61eSJeff Mahoney btrfs_balance_delayed_items(fs_info); 474616cdcec7SMiao Xie 4747d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 4748d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 4749d814a491SEthan Lien fs_info->dirty_metadata_batch); 4750e2d84521SMiao Xie if (ret > 0) { 47510b246afaSJeff Mahoney balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); 475216cdcec7SMiao Xie } 475316cdcec7SMiao Xie } 475416cdcec7SMiao Xie 47552ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) 475616cdcec7SMiao Xie { 47572ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 1); 475835b7e476SChris Mason } 4759b53d3f5dSLiu Bo 47602ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) 4761b53d3f5dSLiu Bo { 47622ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 0); 4763d6bfde87SChris Mason } 47646b80053dSChris Mason 47652ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) 4766acce952bSliubo { 4767fe816d0fSNikolay Borisov /* cleanup FS via transaction */ 4768fe816d0fSNikolay Borisov btrfs_cleanup_transaction(fs_info); 4769fe816d0fSNikolay Borisov 47700b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 47712ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 47720b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 4773acce952bSliubo 47740b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 47750b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4776acce952bSliubo } 4777acce952bSliubo 4778ef67963dSJosef Bacik static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info) 4779ef67963dSJosef Bacik { 4780fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 4781fc7cbcd4SDavid Sterba u64 root_objectid = 0; 4782fc7cbcd4SDavid Sterba int ret; 4783ef67963dSJosef Bacik 4784fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4785fc7cbcd4SDavid Sterba while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4786fc7cbcd4SDavid Sterba (void **)gang, root_objectid, 4787fc7cbcd4SDavid Sterba ARRAY_SIZE(gang))) != 0) { 4788fc7cbcd4SDavid Sterba int i; 4789ef67963dSJosef Bacik 4790fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) 4791fc7cbcd4SDavid Sterba gang[i] = btrfs_grab_root(gang[i]); 4792fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4793fc7cbcd4SDavid Sterba 4794fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 4795fc7cbcd4SDavid Sterba if (!gang[i]) 4796ef67963dSJosef Bacik continue; 4797fc7cbcd4SDavid Sterba root_objectid = gang[i]->root_key.objectid; 4798fc7cbcd4SDavid Sterba btrfs_free_log(NULL, gang[i]); 4799fc7cbcd4SDavid Sterba btrfs_put_root(gang[i]); 4800ef67963dSJosef Bacik } 4801fc7cbcd4SDavid Sterba root_objectid++; 4802fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4803ef67963dSJosef Bacik } 4804fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4805ef67963dSJosef Bacik btrfs_free_log_root_tree(NULL, fs_info); 4806ef67963dSJosef Bacik } 4807ef67963dSJosef Bacik 4808143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4809acce952bSliubo { 4810acce952bSliubo struct btrfs_ordered_extent *ordered; 4811acce952bSliubo 4812199c2a9cSMiao Xie spin_lock(&root->ordered_extent_lock); 4813779880efSJosef Bacik /* 4814779880efSJosef Bacik * This will just short circuit the ordered completion stuff which will 4815779880efSJosef Bacik * make sure the ordered extent gets properly cleaned up. 4816779880efSJosef Bacik */ 4817199c2a9cSMiao Xie list_for_each_entry(ordered, &root->ordered_extents, 4818779880efSJosef Bacik root_extent_list) 4819779880efSJosef Bacik set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4820199c2a9cSMiao Xie spin_unlock(&root->ordered_extent_lock); 4821199c2a9cSMiao Xie } 4822199c2a9cSMiao Xie 4823199c2a9cSMiao Xie static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4824199c2a9cSMiao Xie { 4825199c2a9cSMiao Xie struct btrfs_root *root; 4826199c2a9cSMiao Xie struct list_head splice; 4827199c2a9cSMiao Xie 4828199c2a9cSMiao Xie INIT_LIST_HEAD(&splice); 4829199c2a9cSMiao Xie 4830199c2a9cSMiao Xie spin_lock(&fs_info->ordered_root_lock); 4831199c2a9cSMiao Xie list_splice_init(&fs_info->ordered_roots, &splice); 4832199c2a9cSMiao Xie while (!list_empty(&splice)) { 4833199c2a9cSMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4834199c2a9cSMiao Xie ordered_root); 48351de2cfdeSJosef Bacik list_move_tail(&root->ordered_root, 48361de2cfdeSJosef Bacik &fs_info->ordered_roots); 4837199c2a9cSMiao Xie 48382a85d9caSLiu Bo spin_unlock(&fs_info->ordered_root_lock); 4839199c2a9cSMiao Xie btrfs_destroy_ordered_extents(root); 4840199c2a9cSMiao Xie 48412a85d9caSLiu Bo cond_resched(); 48422a85d9caSLiu Bo spin_lock(&fs_info->ordered_root_lock); 4843199c2a9cSMiao Xie } 4844199c2a9cSMiao Xie spin_unlock(&fs_info->ordered_root_lock); 484574d5d229SJosef Bacik 484674d5d229SJosef Bacik /* 484774d5d229SJosef Bacik * We need this here because if we've been flipped read-only we won't 484874d5d229SJosef Bacik * get sync() from the umount, so we need to make sure any ordered 484974d5d229SJosef Bacik * extents that haven't had their dirty pages IO start writeout yet 485074d5d229SJosef Bacik * actually get run and error out properly. 485174d5d229SJosef Bacik */ 485274d5d229SJosef Bacik btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 4853acce952bSliubo } 4854acce952bSliubo 485535a3621bSStefan Behrens static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 48562ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4857acce952bSliubo { 4858acce952bSliubo struct rb_node *node; 4859acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 4860acce952bSliubo struct btrfs_delayed_ref_node *ref; 4861acce952bSliubo int ret = 0; 4862acce952bSliubo 4863acce952bSliubo delayed_refs = &trans->delayed_refs; 4864acce952bSliubo 4865acce952bSliubo spin_lock(&delayed_refs->lock); 4866d7df2c79SJosef Bacik if (atomic_read(&delayed_refs->num_entries) == 0) { 4867cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 4868b79ce3ddSDavid Sterba btrfs_debug(fs_info, "delayed_refs has NO entry"); 4869acce952bSliubo return ret; 4870acce952bSliubo } 4871acce952bSliubo 48725c9d028bSLiu Bo while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { 4873d7df2c79SJosef Bacik struct btrfs_delayed_ref_head *head; 48740e0adbcfSJosef Bacik struct rb_node *n; 4875e78417d1SJosef Bacik bool pin_bytes = false; 4876acce952bSliubo 4877d7df2c79SJosef Bacik head = rb_entry(node, struct btrfs_delayed_ref_head, 4878d7df2c79SJosef Bacik href_node); 48793069bd26SJosef Bacik if (btrfs_delayed_ref_lock(delayed_refs, head)) 4880b939d1abSJosef Bacik continue; 48813069bd26SJosef Bacik 4882d7df2c79SJosef Bacik spin_lock(&head->lock); 4883e3d03965SLiu Bo while ((n = rb_first_cached(&head->ref_tree)) != NULL) { 48840e0adbcfSJosef Bacik ref = rb_entry(n, struct btrfs_delayed_ref_node, 48850e0adbcfSJosef Bacik ref_node); 4886d7df2c79SJosef Bacik ref->in_tree = 0; 4887e3d03965SLiu Bo rb_erase_cached(&ref->ref_node, &head->ref_tree); 48880e0adbcfSJosef Bacik RB_CLEAR_NODE(&ref->ref_node); 48891d57ee94SWang Xiaoguang if (!list_empty(&ref->add_list)) 48901d57ee94SWang Xiaoguang list_del(&ref->add_list); 4891d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4892d7df2c79SJosef Bacik btrfs_put_delayed_ref(ref); 4893d7df2c79SJosef Bacik } 489454067ae9SJosef Bacik if (head->must_insert_reserved) 4895e78417d1SJosef Bacik pin_bytes = true; 489678a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 4897fa781ceaSJosef Bacik btrfs_delete_ref_head(delayed_refs, head); 4898d7df2c79SJosef Bacik spin_unlock(&head->lock); 4899acce952bSliubo spin_unlock(&delayed_refs->lock); 4900e78417d1SJosef Bacik mutex_unlock(&head->mutex); 4901acce952bSliubo 4902f603bb94SNikolay Borisov if (pin_bytes) { 4903f603bb94SNikolay Borisov struct btrfs_block_group *cache; 4904f603bb94SNikolay Borisov 4905f603bb94SNikolay Borisov cache = btrfs_lookup_block_group(fs_info, head->bytenr); 4906f603bb94SNikolay Borisov BUG_ON(!cache); 4907f603bb94SNikolay Borisov 4908f603bb94SNikolay Borisov spin_lock(&cache->space_info->lock); 4909f603bb94SNikolay Borisov spin_lock(&cache->lock); 4910f603bb94SNikolay Borisov cache->pinned += head->num_bytes; 4911f603bb94SNikolay Borisov btrfs_space_info_update_bytes_pinned(fs_info, 4912f603bb94SNikolay Borisov cache->space_info, head->num_bytes); 4913f603bb94SNikolay Borisov cache->reserved -= head->num_bytes; 4914f603bb94SNikolay Borisov cache->space_info->bytes_reserved -= head->num_bytes; 4915f603bb94SNikolay Borisov spin_unlock(&cache->lock); 4916f603bb94SNikolay Borisov spin_unlock(&cache->space_info->lock); 4917f603bb94SNikolay Borisov 4918f603bb94SNikolay Borisov btrfs_put_block_group(cache); 4919f603bb94SNikolay Borisov 4920f603bb94SNikolay Borisov btrfs_error_unpin_extent_range(fs_info, head->bytenr, 4921f603bb94SNikolay Borisov head->bytenr + head->num_bytes - 1); 4922f603bb94SNikolay Borisov } 492331890da0SJosef Bacik btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); 4924d278850eSJosef Bacik btrfs_put_delayed_ref_head(head); 4925acce952bSliubo cond_resched(); 4926acce952bSliubo spin_lock(&delayed_refs->lock); 4927acce952bSliubo } 492881f7eb00SJeff Mahoney btrfs_qgroup_destroy_extent_records(trans); 4929acce952bSliubo 4930acce952bSliubo spin_unlock(&delayed_refs->lock); 4931acce952bSliubo 4932acce952bSliubo return ret; 4933acce952bSliubo } 4934acce952bSliubo 4935143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 4936acce952bSliubo { 4937acce952bSliubo struct btrfs_inode *btrfs_inode; 4938acce952bSliubo struct list_head splice; 4939acce952bSliubo 4940acce952bSliubo INIT_LIST_HEAD(&splice); 4941acce952bSliubo 4942eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4943eb73c1b7SMiao Xie list_splice_init(&root->delalloc_inodes, &splice); 4944acce952bSliubo 4945acce952bSliubo while (!list_empty(&splice)) { 4946fe816d0fSNikolay Borisov struct inode *inode = NULL; 4947eb73c1b7SMiao Xie btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 4948acce952bSliubo delalloc_inodes); 4949fe816d0fSNikolay Borisov __btrfs_del_delalloc_inode(root, btrfs_inode); 4950eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4951acce952bSliubo 4952fe816d0fSNikolay Borisov /* 4953fe816d0fSNikolay Borisov * Make sure we get a live inode and that it'll not disappear 4954fe816d0fSNikolay Borisov * meanwhile. 4955fe816d0fSNikolay Borisov */ 4956fe816d0fSNikolay Borisov inode = igrab(&btrfs_inode->vfs_inode); 4957fe816d0fSNikolay Borisov if (inode) { 4958fe816d0fSNikolay Borisov invalidate_inode_pages2(inode->i_mapping); 4959fe816d0fSNikolay Borisov iput(inode); 4960fe816d0fSNikolay Borisov } 4961eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4962acce952bSliubo } 4963eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4964eb73c1b7SMiao Xie } 4965eb73c1b7SMiao Xie 4966eb73c1b7SMiao Xie static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 4967eb73c1b7SMiao Xie { 4968eb73c1b7SMiao Xie struct btrfs_root *root; 4969eb73c1b7SMiao Xie struct list_head splice; 4970eb73c1b7SMiao Xie 4971eb73c1b7SMiao Xie INIT_LIST_HEAD(&splice); 4972eb73c1b7SMiao Xie 4973eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4974eb73c1b7SMiao Xie list_splice_init(&fs_info->delalloc_roots, &splice); 4975eb73c1b7SMiao Xie while (!list_empty(&splice)) { 4976eb73c1b7SMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4977eb73c1b7SMiao Xie delalloc_root); 497800246528SJosef Bacik root = btrfs_grab_root(root); 4979eb73c1b7SMiao Xie BUG_ON(!root); 4980eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4981eb73c1b7SMiao Xie 4982eb73c1b7SMiao Xie btrfs_destroy_delalloc_inodes(root); 498300246528SJosef Bacik btrfs_put_root(root); 4984eb73c1b7SMiao Xie 4985eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4986eb73c1b7SMiao Xie } 4987eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4988acce952bSliubo } 4989acce952bSliubo 49902ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 4991acce952bSliubo struct extent_io_tree *dirty_pages, 4992acce952bSliubo int mark) 4993acce952bSliubo { 4994acce952bSliubo int ret; 4995acce952bSliubo struct extent_buffer *eb; 4996acce952bSliubo u64 start = 0; 4997acce952bSliubo u64 end; 4998acce952bSliubo 4999acce952bSliubo while (1) { 5000acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 5001e6138876SJosef Bacik mark, NULL); 5002acce952bSliubo if (ret) 5003acce952bSliubo break; 5004acce952bSliubo 500591166212SDavid Sterba clear_extent_bits(dirty_pages, start, end, mark); 5006acce952bSliubo while (start <= end) { 50070b246afaSJeff Mahoney eb = find_extent_buffer(fs_info, start); 50080b246afaSJeff Mahoney start += fs_info->nodesize; 5009fd8b2b61SJosef Bacik if (!eb) 5010acce952bSliubo continue; 5011fd8b2b61SJosef Bacik wait_on_extent_buffer_writeback(eb); 5012acce952bSliubo 5013fd8b2b61SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 5014fd8b2b61SJosef Bacik &eb->bflags)) 5015fd8b2b61SJosef Bacik clear_extent_buffer_dirty(eb); 5016fd8b2b61SJosef Bacik free_extent_buffer_stale(eb); 5017acce952bSliubo } 5018acce952bSliubo } 5019acce952bSliubo 5020acce952bSliubo return ret; 5021acce952bSliubo } 5022acce952bSliubo 50232ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 5024fe119a6eSNikolay Borisov struct extent_io_tree *unpin) 5025acce952bSliubo { 5026acce952bSliubo u64 start; 5027acce952bSliubo u64 end; 5028acce952bSliubo int ret; 5029acce952bSliubo 5030acce952bSliubo while (1) { 50310e6ec385SFilipe Manana struct extent_state *cached_state = NULL; 50320e6ec385SFilipe Manana 5033fcd5e742SLu Fengqi /* 5034fcd5e742SLu Fengqi * The btrfs_finish_extent_commit() may get the same range as 5035fcd5e742SLu Fengqi * ours between find_first_extent_bit and clear_extent_dirty. 5036fcd5e742SLu Fengqi * Hence, hold the unused_bg_unpin_mutex to avoid double unpin 5037fcd5e742SLu Fengqi * the same extent range. 5038fcd5e742SLu Fengqi */ 5039fcd5e742SLu Fengqi mutex_lock(&fs_info->unused_bg_unpin_mutex); 5040acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 50410e6ec385SFilipe Manana EXTENT_DIRTY, &cached_state); 5042fcd5e742SLu Fengqi if (ret) { 5043fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 5044acce952bSliubo break; 5045fcd5e742SLu Fengqi } 5046acce952bSliubo 50470e6ec385SFilipe Manana clear_extent_dirty(unpin, start, end, &cached_state); 50480e6ec385SFilipe Manana free_extent_state(cached_state); 50492ff7e61eSJeff Mahoney btrfs_error_unpin_extent_range(fs_info, start, end); 5050fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 5051acce952bSliubo cond_resched(); 5052acce952bSliubo } 5053acce952bSliubo 5054acce952bSliubo return 0; 5055acce952bSliubo } 5056acce952bSliubo 505732da5386SDavid Sterba static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache) 5058c79a1751SLiu Bo { 5059c79a1751SLiu Bo struct inode *inode; 5060c79a1751SLiu Bo 5061c79a1751SLiu Bo inode = cache->io_ctl.inode; 5062c79a1751SLiu Bo if (inode) { 5063c79a1751SLiu Bo invalidate_inode_pages2(inode->i_mapping); 5064c79a1751SLiu Bo BTRFS_I(inode)->generation = 0; 5065c79a1751SLiu Bo cache->io_ctl.inode = NULL; 5066c79a1751SLiu Bo iput(inode); 5067c79a1751SLiu Bo } 5068bbc37d6eSFilipe Manana ASSERT(cache->io_ctl.pages == NULL); 5069c79a1751SLiu Bo btrfs_put_block_group(cache); 5070c79a1751SLiu Bo } 5071c79a1751SLiu Bo 5072c79a1751SLiu Bo void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, 50732ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 5074c79a1751SLiu Bo { 507532da5386SDavid Sterba struct btrfs_block_group *cache; 5076c79a1751SLiu Bo 5077c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 5078c79a1751SLiu Bo while (!list_empty(&cur_trans->dirty_bgs)) { 5079c79a1751SLiu Bo cache = list_first_entry(&cur_trans->dirty_bgs, 508032da5386SDavid Sterba struct btrfs_block_group, 5081c79a1751SLiu Bo dirty_list); 5082c79a1751SLiu Bo 5083c79a1751SLiu Bo if (!list_empty(&cache->io_list)) { 5084c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 5085c79a1751SLiu Bo list_del_init(&cache->io_list); 5086c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 5087c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 5088c79a1751SLiu Bo } 5089c79a1751SLiu Bo 5090c79a1751SLiu Bo list_del_init(&cache->dirty_list); 5091c79a1751SLiu Bo spin_lock(&cache->lock); 5092c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 5093c79a1751SLiu Bo spin_unlock(&cache->lock); 5094c79a1751SLiu Bo 5095c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 5096c79a1751SLiu Bo btrfs_put_block_group(cache); 5097ba2c4d4eSJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 5098c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 5099c79a1751SLiu Bo } 5100c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 5101c79a1751SLiu Bo 510245ae2c18SNikolay Borisov /* 510345ae2c18SNikolay Borisov * Refer to the definition of io_bgs member for details why it's safe 510445ae2c18SNikolay Borisov * to use it without any locking 510545ae2c18SNikolay Borisov */ 5106c79a1751SLiu Bo while (!list_empty(&cur_trans->io_bgs)) { 5107c79a1751SLiu Bo cache = list_first_entry(&cur_trans->io_bgs, 510832da5386SDavid Sterba struct btrfs_block_group, 5109c79a1751SLiu Bo io_list); 5110c79a1751SLiu Bo 5111c79a1751SLiu Bo list_del_init(&cache->io_list); 5112c79a1751SLiu Bo spin_lock(&cache->lock); 5113c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 5114c79a1751SLiu Bo spin_unlock(&cache->lock); 5115c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 5116c79a1751SLiu Bo } 5117c79a1751SLiu Bo } 5118c79a1751SLiu Bo 511949b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 51202ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 512149b25e05SJeff Mahoney { 5122bbbf7243SNikolay Borisov struct btrfs_device *dev, *tmp; 5123bbbf7243SNikolay Borisov 51242ff7e61eSJeff Mahoney btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 5125c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->dirty_bgs)); 5126c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->io_bgs)); 5127c79a1751SLiu Bo 5128bbbf7243SNikolay Borisov list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list, 5129bbbf7243SNikolay Borisov post_commit_list) { 5130bbbf7243SNikolay Borisov list_del_init(&dev->post_commit_list); 5131bbbf7243SNikolay Borisov } 5132bbbf7243SNikolay Borisov 51332ff7e61eSJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, fs_info); 513449b25e05SJeff Mahoney 51354a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 51360b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 513749b25e05SJeff Mahoney 51384a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 51390b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 514049b25e05SJeff Mahoney 5141ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 514267cde344SMiao Xie 51432ff7e61eSJeff Mahoney btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, 514449b25e05SJeff Mahoney EXTENT_DIRTY); 5145fe119a6eSNikolay Borisov btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents); 514649b25e05SJeff Mahoney 5147d3575156SNaohiro Aota btrfs_free_redirty_list(cur_trans); 5148d3575156SNaohiro Aota 51494a9d8bdeSMiao Xie cur_trans->state =TRANS_STATE_COMPLETED; 51504a9d8bdeSMiao Xie wake_up(&cur_trans->commit_wait); 515149b25e05SJeff Mahoney } 515249b25e05SJeff Mahoney 51532ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) 5154acce952bSliubo { 5155acce952bSliubo struct btrfs_transaction *t; 5156acce952bSliubo 51570b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 5158acce952bSliubo 51590b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 51600b246afaSJeff Mahoney while (!list_empty(&fs_info->trans_list)) { 51610b246afaSJeff Mahoney t = list_first_entry(&fs_info->trans_list, 5162724e2315SJosef Bacik struct btrfs_transaction, list); 5163724e2315SJosef Bacik if (t->state >= TRANS_STATE_COMMIT_START) { 51649b64f57dSElena Reshetova refcount_inc(&t->use_count); 51650b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 51662ff7e61eSJeff Mahoney btrfs_wait_for_commit(fs_info, t->transid); 5167724e2315SJosef Bacik btrfs_put_transaction(t); 51680b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5169724e2315SJosef Bacik continue; 5170724e2315SJosef Bacik } 51710b246afaSJeff Mahoney if (t == fs_info->running_transaction) { 5172724e2315SJosef Bacik t->state = TRANS_STATE_COMMIT_DOING; 51730b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 5174724e2315SJosef Bacik /* 5175724e2315SJosef Bacik * We wait for 0 num_writers since we don't hold a trans 5176724e2315SJosef Bacik * handle open currently for this transaction. 5177724e2315SJosef Bacik */ 5178724e2315SJosef Bacik wait_event(t->writer_wait, 5179724e2315SJosef Bacik atomic_read(&t->num_writers) == 0); 5180724e2315SJosef Bacik } else { 51810b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 5182724e2315SJosef Bacik } 51832ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(t, fs_info); 5184724e2315SJosef Bacik 51850b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 51860b246afaSJeff Mahoney if (t == fs_info->running_transaction) 51870b246afaSJeff Mahoney fs_info->running_transaction = NULL; 5188724e2315SJosef Bacik list_del_init(&t->list); 51890b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 5190a4abeea4SJosef Bacik 5191724e2315SJosef Bacik btrfs_put_transaction(t); 51922e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 51930b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5194724e2315SJosef Bacik } 51950b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 51960b246afaSJeff Mahoney btrfs_destroy_all_ordered_extents(fs_info); 5197ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 5198ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 51990b246afaSJeff Mahoney btrfs_destroy_all_delalloc_inodes(fs_info); 5200ef67963dSJosef Bacik btrfs_drop_all_logs(fs_info); 52010b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 5202acce952bSliubo 5203acce952bSliubo return 0; 5204acce952bSliubo } 5205ec7d6dfdSNikolay Borisov 5206453e4873SNikolay Borisov int btrfs_init_root_free_objectid(struct btrfs_root *root) 5207ec7d6dfdSNikolay Borisov { 5208ec7d6dfdSNikolay Borisov struct btrfs_path *path; 5209ec7d6dfdSNikolay Borisov int ret; 5210ec7d6dfdSNikolay Borisov struct extent_buffer *l; 5211ec7d6dfdSNikolay Borisov struct btrfs_key search_key; 5212ec7d6dfdSNikolay Borisov struct btrfs_key found_key; 5213ec7d6dfdSNikolay Borisov int slot; 5214ec7d6dfdSNikolay Borisov 5215ec7d6dfdSNikolay Borisov path = btrfs_alloc_path(); 5216ec7d6dfdSNikolay Borisov if (!path) 5217ec7d6dfdSNikolay Borisov return -ENOMEM; 5218ec7d6dfdSNikolay Borisov 5219ec7d6dfdSNikolay Borisov search_key.objectid = BTRFS_LAST_FREE_OBJECTID; 5220ec7d6dfdSNikolay Borisov search_key.type = -1; 5221ec7d6dfdSNikolay Borisov search_key.offset = (u64)-1; 5222ec7d6dfdSNikolay Borisov ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); 5223ec7d6dfdSNikolay Borisov if (ret < 0) 5224ec7d6dfdSNikolay Borisov goto error; 5225ec7d6dfdSNikolay Borisov BUG_ON(ret == 0); /* Corruption */ 5226ec7d6dfdSNikolay Borisov if (path->slots[0] > 0) { 5227ec7d6dfdSNikolay Borisov slot = path->slots[0] - 1; 5228ec7d6dfdSNikolay Borisov l = path->nodes[0]; 5229ec7d6dfdSNikolay Borisov btrfs_item_key_to_cpu(l, &found_key, slot); 523023125104SNikolay Borisov root->free_objectid = max_t(u64, found_key.objectid + 1, 523123125104SNikolay Borisov BTRFS_FIRST_FREE_OBJECTID); 5232ec7d6dfdSNikolay Borisov } else { 523323125104SNikolay Borisov root->free_objectid = BTRFS_FIRST_FREE_OBJECTID; 5234ec7d6dfdSNikolay Borisov } 5235ec7d6dfdSNikolay Borisov ret = 0; 5236ec7d6dfdSNikolay Borisov error: 5237ec7d6dfdSNikolay Borisov btrfs_free_path(path); 5238ec7d6dfdSNikolay Borisov return ret; 5239ec7d6dfdSNikolay Borisov } 5240ec7d6dfdSNikolay Borisov 5241543068a2SNikolay Borisov int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid) 5242ec7d6dfdSNikolay Borisov { 5243ec7d6dfdSNikolay Borisov int ret; 5244ec7d6dfdSNikolay Borisov mutex_lock(&root->objectid_mutex); 5245ec7d6dfdSNikolay Borisov 52466b8fad57SNikolay Borisov if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) { 5247ec7d6dfdSNikolay Borisov btrfs_warn(root->fs_info, 5248ec7d6dfdSNikolay Borisov "the objectid of root %llu reaches its highest value", 5249ec7d6dfdSNikolay Borisov root->root_key.objectid); 5250ec7d6dfdSNikolay Borisov ret = -ENOSPC; 5251ec7d6dfdSNikolay Borisov goto out; 5252ec7d6dfdSNikolay Borisov } 5253ec7d6dfdSNikolay Borisov 525423125104SNikolay Borisov *objectid = root->free_objectid++; 5255ec7d6dfdSNikolay Borisov ret = 0; 5256ec7d6dfdSNikolay Borisov out: 5257ec7d6dfdSNikolay Borisov mutex_unlock(&root->objectid_mutex); 5258ec7d6dfdSNikolay Borisov return ret; 5259ec7d6dfdSNikolay Borisov } 5260