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 { 6508b712842SChris Mason struct async_submit_bio *async; 65106ea01b1SDavid Sterba struct inode *inode; 6528b712842SChris Mason 6538b712842SChris Mason async = container_of(work, struct async_submit_bio, work); 6548896a08dSQu Wenruo inode = async->inode; 6554854ddd0SChris Mason 656bb7ab3b9SAdam Buchbinder /* If an error occurred we just want to clean up the bio and move on */ 6574e4cbee9SChristoph Hellwig if (async->status) { 6584e4cbee9SChristoph Hellwig async->bio->bi_status = async->status; 6594246a0b6SChristoph Hellwig bio_endio(async->bio); 66079787eaaSJeff Mahoney return; 66179787eaaSJeff Mahoney } 66279787eaaSJeff Mahoney 663ec39f769SChris Mason /* 664ec39f769SChris Mason * All of the bios that pass through here are from async helpers. 665ec39f769SChris Mason * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context. 666ec39f769SChris Mason * This changes nothing when cgroups aren't in use. 667ec39f769SChris Mason */ 668ec39f769SChris Mason async->bio->bi_opf |= REQ_CGROUP_PUNT; 6691a722d8fSChristoph Hellwig btrfs_submit_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num); 6704a69a410SChris Mason } 6714a69a410SChris Mason 6724a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 6734a69a410SChris Mason { 6744a69a410SChris Mason struct async_submit_bio *async; 6754a69a410SChris Mason 6764a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 6778b712842SChris Mason kfree(async); 6788b712842SChris Mason } 6798b712842SChris Mason 680ea1f0cedSChristoph Hellwig /* 681ea1f0cedSChristoph Hellwig * Submit bio to an async queue. 682ea1f0cedSChristoph Hellwig * 683ea1f0cedSChristoph Hellwig * Retrun: 684ea1f0cedSChristoph Hellwig * - true if the work has been succesfuly submitted 685ea1f0cedSChristoph Hellwig * - false in case of error 686ea1f0cedSChristoph Hellwig */ 687ea1f0cedSChristoph Hellwig bool btrfs_wq_submit_bio(struct inode *inode, struct bio *bio, int mirror_num, 688ea1f0cedSChristoph Hellwig u64 dio_file_offset, 689e288c080SDavid Sterba extent_submit_bio_start_t *submit_bio_start) 69044b8bd7eSChris Mason { 6918896a08dSQu Wenruo struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; 69244b8bd7eSChris Mason struct async_submit_bio *async; 69344b8bd7eSChris Mason 69444b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 69544b8bd7eSChris Mason if (!async) 696ea1f0cedSChristoph Hellwig return false; 69744b8bd7eSChris Mason 6988896a08dSQu Wenruo async->inode = inode; 69944b8bd7eSChris Mason async->bio = bio; 70044b8bd7eSChris Mason async->mirror_num = mirror_num; 7014a69a410SChris Mason async->submit_bio_start = submit_bio_start; 7024a69a410SChris Mason 703a0cac0ecSOmar Sandoval btrfs_init_work(&async->work, run_one_async_start, run_one_async_done, 704a0cac0ecSOmar Sandoval run_one_async_free); 7054a69a410SChris Mason 7061941b64bSQu Wenruo async->dio_file_offset = dio_file_offset; 7078c8bee1dSChris Mason 7084e4cbee9SChristoph Hellwig async->status = 0; 70979787eaaSJeff Mahoney 71067f055c7SChristoph Hellwig if (op_is_sync(bio->bi_opf)) 711a31b4a43SChristoph Hellwig btrfs_queue_work(fs_info->hipri_workers, &async->work); 712a31b4a43SChristoph Hellwig else 7135cdc7ad3SQu Wenruo btrfs_queue_work(fs_info->workers, &async->work); 714ea1f0cedSChristoph Hellwig return true; 71544b8bd7eSChris Mason } 71644b8bd7eSChris Mason 7174e4cbee9SChristoph Hellwig static blk_status_t btree_csum_one_bio(struct bio *bio) 718ce3ed71aSChris Mason { 7192c30c71bSKent Overstreet struct bio_vec *bvec; 720ce3ed71aSChris Mason struct btrfs_root *root; 7212b070cfeSChristoph Hellwig int ret = 0; 7226dc4f100SMing Lei struct bvec_iter_all iter_all; 723ce3ed71aSChris Mason 724c09abff8SDavid Sterba ASSERT(!bio_flagged(bio, BIO_CLONED)); 7252b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 726ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 727ac303b69SQu Wenruo ret = csum_dirty_buffer(root->fs_info, bvec); 72879787eaaSJeff Mahoney if (ret) 72979787eaaSJeff Mahoney break; 730ce3ed71aSChris Mason } 7312c30c71bSKent Overstreet 7324e4cbee9SChristoph Hellwig return errno_to_blk_status(ret); 733ce3ed71aSChris Mason } 734ce3ed71aSChris Mason 7358896a08dSQu Wenruo static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio, 7361941b64bSQu Wenruo u64 dio_file_offset) 73722c59948SChris Mason { 7388b712842SChris Mason /* 7398b712842SChris Mason * when we're called for a write, we're already in the async 7401a722d8fSChristoph Hellwig * submission context. Just jump into btrfs_submit_bio. 7418b712842SChris Mason */ 74279787eaaSJeff Mahoney return btree_csum_one_bio(bio); 74322c59948SChris Mason } 74422c59948SChris Mason 745f4dcfb30SJohannes Thumshirn static bool should_async_write(struct btrfs_fs_info *fs_info, 7469b4e675aSDavid Sterba struct btrfs_inode *bi) 747de0022b9SJosef Bacik { 7484eef29efSNaohiro Aota if (btrfs_is_zoned(fs_info)) 749f4dcfb30SJohannes Thumshirn return false; 7506300463bSLiu Bo if (atomic_read(&bi->sync_writers)) 751f4dcfb30SJohannes Thumshirn return false; 7529b4e675aSDavid Sterba if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags)) 753f4dcfb30SJohannes Thumshirn return false; 754f4dcfb30SJohannes Thumshirn return true; 755de0022b9SJosef Bacik } 756de0022b9SJosef Bacik 75794d9e11bSChristoph Hellwig void btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio, int mirror_num) 75844b8bd7eSChris Mason { 7590b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 7604e4cbee9SChristoph Hellwig blk_status_t ret; 761cad321adSChris Mason 76208a6f464SChristoph Hellwig bio->bi_opf |= REQ_META; 76308a6f464SChristoph Hellwig 764cfe94440SNaohiro Aota if (btrfs_op(bio) != BTRFS_MAP_WRITE) { 7651a722d8fSChristoph Hellwig btrfs_submit_bio(fs_info, bio, mirror_num); 7661a722d8fSChristoph Hellwig return; 76744b8bd7eSChris Mason } 76844b8bd7eSChris Mason 769ea1f0cedSChristoph Hellwig /* 770ea1f0cedSChristoph Hellwig * Kthread helpers are used to submit writes so that checksumming can 771ea1f0cedSChristoph Hellwig * happen in parallel across all CPUs. 772ea1f0cedSChristoph Hellwig */ 773ea1f0cedSChristoph Hellwig if (should_async_write(fs_info, BTRFS_I(inode)) && 774ea1f0cedSChristoph Hellwig btrfs_wq_submit_bio(inode, bio, mirror_num, 0, btree_submit_bio_start)) 775ea1f0cedSChristoph Hellwig return; 776ea1f0cedSChristoph Hellwig 777ea1f0cedSChristoph Hellwig ret = btree_csum_one_bio(bio); 77894d9e11bSChristoph Hellwig if (ret) { 7794e4cbee9SChristoph Hellwig bio->bi_status = ret; 7804246a0b6SChristoph Hellwig bio_endio(bio); 781ea1f0cedSChristoph Hellwig return; 78294d9e11bSChristoph Hellwig } 783ea1f0cedSChristoph Hellwig 784ea1f0cedSChristoph Hellwig btrfs_submit_bio(fs_info, bio, mirror_num); 78561891923SStefan Behrens } 78661891923SStefan Behrens 7873dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 7888958b551SMatthew Wilcox (Oracle) static int btree_migrate_folio(struct address_space *mapping, 7898958b551SMatthew Wilcox (Oracle) struct folio *dst, struct folio *src, enum migrate_mode mode) 790784b4e29SChris Mason { 791784b4e29SChris Mason /* 792784b4e29SChris Mason * we can't safely write a btree page from here, 793784b4e29SChris Mason * we haven't done the locking hook 794784b4e29SChris Mason */ 7958958b551SMatthew Wilcox (Oracle) if (folio_test_dirty(src)) 796784b4e29SChris Mason return -EAGAIN; 797784b4e29SChris Mason /* 798784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 799784b4e29SChris Mason * We must have no buffers or drop them. 800784b4e29SChris Mason */ 8018958b551SMatthew Wilcox (Oracle) if (folio_get_private(src) && 8028958b551SMatthew Wilcox (Oracle) !filemap_release_folio(src, GFP_KERNEL)) 803784b4e29SChris Mason return -EAGAIN; 80454184650SMatthew Wilcox (Oracle) return migrate_folio(mapping, dst, src, mode); 805784b4e29SChris Mason } 8068958b551SMatthew Wilcox (Oracle) #else 8078958b551SMatthew Wilcox (Oracle) #define btree_migrate_folio NULL 8083dd1462eSJan Beulich #endif 809784b4e29SChris Mason 8100da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 8110da5468fSChris Mason struct writeback_control *wbc) 8120da5468fSChris Mason { 813e2d84521SMiao Xie struct btrfs_fs_info *fs_info; 814e2d84521SMiao Xie int ret; 815e2d84521SMiao Xie 816d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 817448d640bSChris Mason 818448d640bSChris Mason if (wbc->for_kupdate) 819448d640bSChris Mason return 0; 820448d640bSChris Mason 821e2d84521SMiao Xie fs_info = BTRFS_I(mapping->host)->root->fs_info; 822b9473439SChris Mason /* this is a bit racy, but that's ok */ 823d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 824d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 825d814a491SEthan Lien fs_info->dirty_metadata_batch); 826e2d84521SMiao Xie if (ret < 0) 827793955bcSChris Mason return 0; 828793955bcSChris Mason } 8290b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 8300da5468fSChris Mason } 8310da5468fSChris Mason 832f913cff3SMatthew Wilcox (Oracle) static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags) 8335f39d397SChris Mason { 834f913cff3SMatthew Wilcox (Oracle) if (folio_test_writeback(folio) || folio_test_dirty(folio)) 835f913cff3SMatthew Wilcox (Oracle) return false; 8360c4e538bSDavid Sterba 837f913cff3SMatthew Wilcox (Oracle) return try_release_extent_buffer(&folio->page); 838d98237b3SChris Mason } 839d98237b3SChris Mason 840895586ebSMatthew Wilcox (Oracle) static void btree_invalidate_folio(struct folio *folio, size_t offset, 841895586ebSMatthew Wilcox (Oracle) size_t length) 842d98237b3SChris Mason { 843d1310b2eSChris Mason struct extent_io_tree *tree; 844895586ebSMatthew Wilcox (Oracle) tree = &BTRFS_I(folio->mapping->host)->io_tree; 845895586ebSMatthew Wilcox (Oracle) extent_invalidate_folio(tree, folio, offset); 846f913cff3SMatthew Wilcox (Oracle) btree_release_folio(folio, GFP_NOFS); 847895586ebSMatthew Wilcox (Oracle) if (folio_get_private(folio)) { 848895586ebSMatthew Wilcox (Oracle) btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info, 849895586ebSMatthew Wilcox (Oracle) "folio private not zero on folio %llu", 850895586ebSMatthew Wilcox (Oracle) (unsigned long long)folio_pos(folio)); 851895586ebSMatthew Wilcox (Oracle) folio_detach_private(folio); 8529ad6b7bcSChris Mason } 853d98237b3SChris Mason } 854d98237b3SChris Mason 855bb146eb2SJosef Bacik #ifdef DEBUG 8560079c3b1SMatthew Wilcox (Oracle) static bool btree_dirty_folio(struct address_space *mapping, 8570079c3b1SMatthew Wilcox (Oracle) struct folio *folio) 8580079c3b1SMatthew Wilcox (Oracle) { 8590079c3b1SMatthew Wilcox (Oracle) struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb); 860139e8cd3SQu Wenruo struct btrfs_subpage *subpage; 8610b32f4bbSJosef Bacik struct extent_buffer *eb; 862139e8cd3SQu Wenruo int cur_bit = 0; 8630079c3b1SMatthew Wilcox (Oracle) u64 page_start = folio_pos(folio); 8640b32f4bbSJosef Bacik 865139e8cd3SQu Wenruo if (fs_info->sectorsize == PAGE_SIZE) { 8660079c3b1SMatthew Wilcox (Oracle) eb = folio_get_private(folio); 8670b32f4bbSJosef Bacik BUG_ON(!eb); 8680b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 8690b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 87049d0c642SFilipe Manana btrfs_assert_tree_write_locked(eb); 8710079c3b1SMatthew Wilcox (Oracle) return filemap_dirty_folio(mapping, folio); 872139e8cd3SQu Wenruo } 8730079c3b1SMatthew Wilcox (Oracle) subpage = folio_get_private(folio); 874139e8cd3SQu Wenruo 875139e8cd3SQu Wenruo ASSERT(subpage->dirty_bitmap); 876139e8cd3SQu Wenruo while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) { 877139e8cd3SQu Wenruo unsigned long flags; 878139e8cd3SQu Wenruo u64 cur; 879139e8cd3SQu Wenruo u16 tmp = (1 << cur_bit); 880139e8cd3SQu Wenruo 881139e8cd3SQu Wenruo spin_lock_irqsave(&subpage->lock, flags); 882139e8cd3SQu Wenruo if (!(tmp & subpage->dirty_bitmap)) { 883139e8cd3SQu Wenruo spin_unlock_irqrestore(&subpage->lock, flags); 884139e8cd3SQu Wenruo cur_bit++; 885139e8cd3SQu Wenruo continue; 886139e8cd3SQu Wenruo } 887139e8cd3SQu Wenruo spin_unlock_irqrestore(&subpage->lock, flags); 888139e8cd3SQu Wenruo cur = page_start + cur_bit * fs_info->sectorsize; 889139e8cd3SQu Wenruo 890139e8cd3SQu Wenruo eb = find_extent_buffer(fs_info, cur); 891139e8cd3SQu Wenruo ASSERT(eb); 892139e8cd3SQu Wenruo ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 893139e8cd3SQu Wenruo ASSERT(atomic_read(&eb->refs)); 89449d0c642SFilipe Manana btrfs_assert_tree_write_locked(eb); 895139e8cd3SQu Wenruo free_extent_buffer(eb); 896139e8cd3SQu Wenruo 897139e8cd3SQu Wenruo cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits); 898139e8cd3SQu Wenruo } 8990079c3b1SMatthew Wilcox (Oracle) return filemap_dirty_folio(mapping, folio); 9000b32f4bbSJosef Bacik } 9010079c3b1SMatthew Wilcox (Oracle) #else 9020079c3b1SMatthew Wilcox (Oracle) #define btree_dirty_folio filemap_dirty_folio 9030079c3b1SMatthew Wilcox (Oracle) #endif 9040b32f4bbSJosef Bacik 9057f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 9060da5468fSChris Mason .writepages = btree_writepages, 907f913cff3SMatthew Wilcox (Oracle) .release_folio = btree_release_folio, 908895586ebSMatthew Wilcox (Oracle) .invalidate_folio = btree_invalidate_folio, 9098958b551SMatthew Wilcox (Oracle) .migrate_folio = btree_migrate_folio, 9100079c3b1SMatthew Wilcox (Oracle) .dirty_folio = btree_dirty_folio, 911d98237b3SChris Mason }; 912123abc88SChris Mason 9132ff7e61eSJeff Mahoney struct extent_buffer *btrfs_find_create_tree_block( 9142ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info, 9153fbaf258SJosef Bacik u64 bytenr, u64 owner_root, 9163fbaf258SJosef Bacik int level) 9170999df54SChris Mason { 9180b246afaSJeff Mahoney if (btrfs_is_testing(fs_info)) 9190b246afaSJeff Mahoney return alloc_test_extent_buffer(fs_info, bytenr); 9203fbaf258SJosef Bacik return alloc_extent_buffer(fs_info, bytenr, owner_root, level); 9210999df54SChris Mason } 9220999df54SChris Mason 923581c1760SQu Wenruo /* 924581c1760SQu Wenruo * Read tree block at logical address @bytenr and do variant basic but critical 925581c1760SQu Wenruo * verification. 926581c1760SQu Wenruo * 9271b7ec85eSJosef Bacik * @owner_root: the objectid of the root owner for this block. 928581c1760SQu Wenruo * @parent_transid: expected transid of this tree block, skip check if 0 929581c1760SQu Wenruo * @level: expected level, mandatory check 930581c1760SQu Wenruo * @first_key: expected key in slot 0, skip check if NULL 931581c1760SQu Wenruo */ 9322ff7e61eSJeff Mahoney struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, 9331b7ec85eSJosef Bacik u64 owner_root, u64 parent_transid, 9341b7ec85eSJosef Bacik int level, struct btrfs_key *first_key) 935e20d96d6SChris Mason { 9365f39d397SChris Mason struct extent_buffer *buf = NULL; 93719c00ddcSChris Mason int ret; 93819c00ddcSChris Mason 9393fbaf258SJosef Bacik buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level); 940c871b0f2SLiu Bo if (IS_ERR(buf)) 941c871b0f2SLiu Bo return buf; 942e4204dedSChris Mason 9436a2e9dc4SFilipe Manana ret = btrfs_read_extent_buffer(buf, parent_transid, level, first_key); 9440f0fe8f7SFilipe David Borba Manana if (ret) { 945537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 94664c043deSLiu Bo return ERR_PTR(ret); 9470f0fe8f7SFilipe David Borba Manana } 94888c602abSQu Wenruo if (btrfs_check_eb_owner(buf, owner_root)) { 94988c602abSQu Wenruo free_extent_buffer_stale(buf); 95088c602abSQu Wenruo return ERR_PTR(-EUCLEAN); 95188c602abSQu Wenruo } 9525f39d397SChris Mason return buf; 953ce9adaa5SChris Mason 954eb60ceacSChris Mason } 955eb60ceacSChris Mason 9566a884d7dSDavid Sterba void btrfs_clean_tree_block(struct extent_buffer *buf) 957ed2ff2cbSChris Mason { 9586a884d7dSDavid Sterba struct btrfs_fs_info *fs_info = buf->fs_info; 95955c69072SChris Mason if (btrfs_header_generation(buf) == 960e2d84521SMiao Xie fs_info->running_transaction->transid) { 96149d0c642SFilipe Manana btrfs_assert_tree_write_locked(buf); 962b4ce94deSChris Mason 963b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 964104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 965e2d84521SMiao Xie -buf->len, 966e2d84521SMiao Xie fs_info->dirty_metadata_batch); 9670b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 968925baeddSChris Mason } 9695f39d397SChris Mason } 970ed7b63ebSJosef Bacik } 9715f39d397SChris Mason 972da17066cSJeff Mahoney static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, 973e20d96d6SChris Mason u64 objectid) 974d97e63b6SChris Mason { 9757c0260eeSJeff Mahoney bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); 9762e608bd1SJosef Bacik 9772e608bd1SJosef Bacik memset(&root->root_key, 0, sizeof(root->root_key)); 9782e608bd1SJosef Bacik memset(&root->root_item, 0, sizeof(root->root_item)); 9792e608bd1SJosef Bacik memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 98096dfcb46SJosef Bacik root->fs_info = fs_info; 9812e608bd1SJosef Bacik root->root_key.objectid = objectid; 982cfaa7295SChris Mason root->node = NULL; 983a28ec197SChris Mason root->commit_root = NULL; 98427cdeb70SMiao Xie root->state = 0; 985abed4aaaSJosef Bacik RB_CLEAR_NODE(&root->rb_node); 9860b86a832SChris Mason 9870f7d52f4SChris Mason root->last_trans = 0; 9886b8fad57SNikolay Borisov root->free_objectid = 0; 989eb73c1b7SMiao Xie root->nr_delalloc_inodes = 0; 990199c2a9cSMiao Xie root->nr_ordered_extents = 0; 9916bef4d31SEric Paris root->inode_tree = RB_ROOT; 992088aea3bSDavid Sterba INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 9932e608bd1SJosef Bacik 9942e608bd1SJosef Bacik btrfs_init_root_block_rsv(root); 9950b86a832SChris Mason 9960b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 9975d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 998eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_inodes); 999eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_root); 1000199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_extents); 1001199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_root); 1002d2311e69SQu Wenruo INIT_LIST_HEAD(&root->reloc_dirty_list); 10032ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 10042ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 10055d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1006eb73c1b7SMiao Xie spin_lock_init(&root->delalloc_lock); 1007199c2a9cSMiao Xie spin_lock_init(&root->ordered_extent_lock); 1008f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 10092ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 10102ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 10118287475aSQu Wenruo spin_lock_init(&root->qgroup_meta_rsv_lock); 1012a2135011SChris Mason mutex_init(&root->objectid_mutex); 1013e02119d5SChris Mason mutex_init(&root->log_mutex); 101431f3d255SMiao Xie mutex_init(&root->ordered_extent_mutex); 1015573bfb72SMiao Xie mutex_init(&root->delalloc_mutex); 1016c53e9653SQu Wenruo init_waitqueue_head(&root->qgroup_flush_wait); 10177237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 10187237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 10197237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 10208b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[0]); 10218b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[1]); 10227237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 10237237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 10247237f183SYan Zheng atomic_set(&root->log_writers, 0); 10252ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 10260700cea7SElena Reshetova refcount_set(&root->refs, 1); 10278ecebf4dSRobbie Ko atomic_set(&root->snapshot_force_cow, 0); 1028eede2bf3SOmar Sandoval atomic_set(&root->nr_swapfiles, 0); 10297237f183SYan Zheng root->log_transid = 0; 1030d1433debSMiao Xie root->log_transid_committed = -1; 1031257c62e1SChris Mason root->last_log_commit = 0; 10322e608bd1SJosef Bacik root->anon_dev = 0; 1033e289f03eSFilipe Manana if (!dummy) { 103443eb5f29SQu Wenruo extent_io_tree_init(fs_info, &root->dirty_log_pages, 103543eb5f29SQu Wenruo IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL); 1036e289f03eSFilipe Manana extent_io_tree_init(fs_info, &root->log_csum_range, 1037e289f03eSFilipe Manana IO_TREE_LOG_CSUM_RANGE, NULL); 1038e289f03eSFilipe Manana } 1039017e5369SChris Mason 10405f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 1041370a11b8SQu Wenruo btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks); 1042bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 1043bd647ce3SJosef Bacik INIT_LIST_HEAD(&root->leak_list); 1044fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1045bd647ce3SJosef Bacik list_add_tail(&root->leak_list, &fs_info->allocated_roots); 1046fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 1047bd647ce3SJosef Bacik #endif 10483768f368SChris Mason } 10493768f368SChris Mason 105074e4d827SDavid Sterba static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, 105196dfcb46SJosef Bacik u64 objectid, gfp_t flags) 10526f07e42eSAl Viro { 105374e4d827SDavid Sterba struct btrfs_root *root = kzalloc(sizeof(*root), flags); 10546f07e42eSAl Viro if (root) 105596dfcb46SJosef Bacik __setup_root(root, fs_info, objectid); 10566f07e42eSAl Viro return root; 10576f07e42eSAl Viro } 10586f07e42eSAl Viro 105906ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 106006ea65a3SJosef Bacik /* Should only be used by the testing infrastructure */ 1061da17066cSJeff Mahoney struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) 106206ea65a3SJosef Bacik { 106306ea65a3SJosef Bacik struct btrfs_root *root; 106406ea65a3SJosef Bacik 10657c0260eeSJeff Mahoney if (!fs_info) 10667c0260eeSJeff Mahoney return ERR_PTR(-EINVAL); 10677c0260eeSJeff Mahoney 106896dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL); 106906ea65a3SJosef Bacik if (!root) 107006ea65a3SJosef Bacik return ERR_PTR(-ENOMEM); 1071da17066cSJeff Mahoney 1072b9ef22deSFeifei Xu /* We don't use the stripesize in selftest, set it as sectorsize */ 1073faa2dbf0SJosef Bacik root->alloc_bytenr = 0; 107406ea65a3SJosef Bacik 107506ea65a3SJosef Bacik return root; 107606ea65a3SJosef Bacik } 107706ea65a3SJosef Bacik #endif 107806ea65a3SJosef Bacik 1079abed4aaaSJosef Bacik static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node) 1080abed4aaaSJosef Bacik { 1081abed4aaaSJosef Bacik const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node); 1082abed4aaaSJosef Bacik const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node); 1083abed4aaaSJosef Bacik 1084abed4aaaSJosef Bacik return btrfs_comp_cpu_keys(&a->root_key, &b->root_key); 1085abed4aaaSJosef Bacik } 1086abed4aaaSJosef Bacik 1087abed4aaaSJosef Bacik static int global_root_key_cmp(const void *k, const struct rb_node *node) 1088abed4aaaSJosef Bacik { 1089abed4aaaSJosef Bacik const struct btrfs_key *key = k; 1090abed4aaaSJosef Bacik const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node); 1091abed4aaaSJosef Bacik 1092abed4aaaSJosef Bacik return btrfs_comp_cpu_keys(key, &root->root_key); 1093abed4aaaSJosef Bacik } 1094abed4aaaSJosef Bacik 1095abed4aaaSJosef Bacik int btrfs_global_root_insert(struct btrfs_root *root) 1096abed4aaaSJosef Bacik { 1097abed4aaaSJosef Bacik struct btrfs_fs_info *fs_info = root->fs_info; 1098abed4aaaSJosef Bacik struct rb_node *tmp; 1099abed4aaaSJosef Bacik 1100abed4aaaSJosef Bacik write_lock(&fs_info->global_root_lock); 1101abed4aaaSJosef Bacik tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp); 1102abed4aaaSJosef Bacik write_unlock(&fs_info->global_root_lock); 1103abed4aaaSJosef Bacik ASSERT(!tmp); 1104abed4aaaSJosef Bacik 1105abed4aaaSJosef Bacik return tmp ? -EEXIST : 0; 1106abed4aaaSJosef Bacik } 1107abed4aaaSJosef Bacik 1108abed4aaaSJosef Bacik void btrfs_global_root_delete(struct btrfs_root *root) 1109abed4aaaSJosef Bacik { 1110abed4aaaSJosef Bacik struct btrfs_fs_info *fs_info = root->fs_info; 1111abed4aaaSJosef Bacik 1112abed4aaaSJosef Bacik write_lock(&fs_info->global_root_lock); 1113abed4aaaSJosef Bacik rb_erase(&root->rb_node, &fs_info->global_root_tree); 1114abed4aaaSJosef Bacik write_unlock(&fs_info->global_root_lock); 1115abed4aaaSJosef Bacik } 1116abed4aaaSJosef Bacik 1117abed4aaaSJosef Bacik struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info, 1118abed4aaaSJosef Bacik struct btrfs_key *key) 1119abed4aaaSJosef Bacik { 1120abed4aaaSJosef Bacik struct rb_node *node; 1121abed4aaaSJosef Bacik struct btrfs_root *root = NULL; 1122abed4aaaSJosef Bacik 1123abed4aaaSJosef Bacik read_lock(&fs_info->global_root_lock); 1124abed4aaaSJosef Bacik node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp); 1125abed4aaaSJosef Bacik if (node) 1126abed4aaaSJosef Bacik root = container_of(node, struct btrfs_root, rb_node); 1127abed4aaaSJosef Bacik read_unlock(&fs_info->global_root_lock); 1128abed4aaaSJosef Bacik 1129abed4aaaSJosef Bacik return root; 1130abed4aaaSJosef Bacik } 1131abed4aaaSJosef Bacik 1132f7238e50SJosef Bacik static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr) 1133f7238e50SJosef Bacik { 1134f7238e50SJosef Bacik struct btrfs_block_group *block_group; 1135f7238e50SJosef Bacik u64 ret; 1136f7238e50SJosef Bacik 1137f7238e50SJosef Bacik if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) 1138f7238e50SJosef Bacik return 0; 1139f7238e50SJosef Bacik 1140f7238e50SJosef Bacik if (bytenr) 1141f7238e50SJosef Bacik block_group = btrfs_lookup_block_group(fs_info, bytenr); 1142f7238e50SJosef Bacik else 1143f7238e50SJosef Bacik block_group = btrfs_lookup_first_block_group(fs_info, bytenr); 1144f7238e50SJosef Bacik ASSERT(block_group); 1145f7238e50SJosef Bacik if (!block_group) 1146f7238e50SJosef Bacik return 0; 1147f7238e50SJosef Bacik ret = block_group->global_root_id; 1148f7238e50SJosef Bacik btrfs_put_block_group(block_group); 1149f7238e50SJosef Bacik 1150f7238e50SJosef Bacik return ret; 1151f7238e50SJosef Bacik } 1152f7238e50SJosef Bacik 1153abed4aaaSJosef Bacik struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr) 1154abed4aaaSJosef Bacik { 1155abed4aaaSJosef Bacik struct btrfs_key key = { 1156abed4aaaSJosef Bacik .objectid = BTRFS_CSUM_TREE_OBJECTID, 1157abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 1158f7238e50SJosef Bacik .offset = btrfs_global_root_id(fs_info, bytenr), 1159abed4aaaSJosef Bacik }; 1160abed4aaaSJosef Bacik 1161abed4aaaSJosef Bacik return btrfs_global_root(fs_info, &key); 1162abed4aaaSJosef Bacik } 1163abed4aaaSJosef Bacik 1164abed4aaaSJosef Bacik struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr) 1165abed4aaaSJosef Bacik { 1166abed4aaaSJosef Bacik struct btrfs_key key = { 1167abed4aaaSJosef Bacik .objectid = BTRFS_EXTENT_TREE_OBJECTID, 1168abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 1169f7238e50SJosef Bacik .offset = btrfs_global_root_id(fs_info, bytenr), 1170abed4aaaSJosef Bacik }; 1171abed4aaaSJosef Bacik 1172abed4aaaSJosef Bacik return btrfs_global_root(fs_info, &key); 1173abed4aaaSJosef Bacik } 1174abed4aaaSJosef Bacik 117520897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 117620897f5cSArne Jansen u64 objectid) 117720897f5cSArne Jansen { 11789b7a2440SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 117920897f5cSArne Jansen struct extent_buffer *leaf; 118020897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 118120897f5cSArne Jansen struct btrfs_root *root; 118220897f5cSArne Jansen struct btrfs_key key; 1183b89f6d1fSFilipe Manana unsigned int nofs_flag; 118420897f5cSArne Jansen int ret = 0; 118520897f5cSArne Jansen 1186b89f6d1fSFilipe Manana /* 1187b89f6d1fSFilipe Manana * We're holding a transaction handle, so use a NOFS memory allocation 1188b89f6d1fSFilipe Manana * context to avoid deadlock if reclaim happens. 1189b89f6d1fSFilipe Manana */ 1190b89f6d1fSFilipe Manana nofs_flag = memalloc_nofs_save(); 119196dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL); 1192b89f6d1fSFilipe Manana memalloc_nofs_restore(nofs_flag); 119320897f5cSArne Jansen if (!root) 119420897f5cSArne Jansen return ERR_PTR(-ENOMEM); 119520897f5cSArne Jansen 119620897f5cSArne Jansen root->root_key.objectid = objectid; 119720897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 119820897f5cSArne Jansen root->root_key.offset = 0; 119920897f5cSArne Jansen 12009631e4ccSJosef Bacik leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0, 12019631e4ccSJosef Bacik BTRFS_NESTING_NORMAL); 120220897f5cSArne Jansen if (IS_ERR(leaf)) { 120320897f5cSArne Jansen ret = PTR_ERR(leaf); 12041dd05682STsutomu Itoh leaf = NULL; 12058a6a87cdSBoris Burkov goto fail_unlock; 120620897f5cSArne Jansen } 120720897f5cSArne Jansen 120820897f5cSArne Jansen root->node = leaf; 120920897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 121020897f5cSArne Jansen 121120897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 121227cdeb70SMiao Xie set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 121320897f5cSArne Jansen 1214f944d2cbSDavid Sterba btrfs_set_root_flags(&root->root_item, 0); 1215f944d2cbSDavid Sterba btrfs_set_root_limit(&root->root_item, 0); 121620897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 121720897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 121820897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 121920897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 122020897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 122120897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 122220897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 122333d85fdaSQu Wenruo if (is_fstree(objectid)) 1224807fc790SAndy Shevchenko generate_random_guid(root->root_item.uuid); 1225807fc790SAndy Shevchenko else 1226807fc790SAndy Shevchenko export_guid(root->root_item.uuid, &guid_null); 1227c8422684SDavid Sterba btrfs_set_root_drop_level(&root->root_item, 0); 122820897f5cSArne Jansen 12298a6a87cdSBoris Burkov btrfs_tree_unlock(leaf); 12308a6a87cdSBoris Burkov 123120897f5cSArne Jansen key.objectid = objectid; 123220897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 123320897f5cSArne Jansen key.offset = 0; 123420897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 123520897f5cSArne Jansen if (ret) 123620897f5cSArne Jansen goto fail; 123720897f5cSArne Jansen 123820897f5cSArne Jansen return root; 12391dd05682STsutomu Itoh 12408a6a87cdSBoris Burkov fail_unlock: 12418c38938cSJosef Bacik if (leaf) 12421dd05682STsutomu Itoh btrfs_tree_unlock(leaf); 12438a6a87cdSBoris Burkov fail: 124400246528SJosef Bacik btrfs_put_root(root); 12451dd05682STsutomu Itoh 12461dd05682STsutomu Itoh return ERR_PTR(ret); 124720897f5cSArne Jansen } 124820897f5cSArne Jansen 12497237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1250e02119d5SChris Mason struct btrfs_fs_info *fs_info) 12510f7d52f4SChris Mason { 12520f7d52f4SChris Mason struct btrfs_root *root; 1253e02119d5SChris Mason 125496dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS); 1255e02119d5SChris Mason if (!root) 12567237f183SYan Zheng return ERR_PTR(-ENOMEM); 1257e02119d5SChris Mason 1258e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1259e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1260e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 126127cdeb70SMiao Xie 12626ab6ebb7SNaohiro Aota return root; 12636ab6ebb7SNaohiro Aota } 12646ab6ebb7SNaohiro Aota 12656ab6ebb7SNaohiro Aota int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans, 12666ab6ebb7SNaohiro Aota struct btrfs_root *root) 12676ab6ebb7SNaohiro Aota { 12686ab6ebb7SNaohiro Aota struct extent_buffer *leaf; 12696ab6ebb7SNaohiro Aota 12707237f183SYan Zheng /* 127192a7cc42SQu Wenruo * DON'T set SHAREABLE bit for log trees. 127227cdeb70SMiao Xie * 127392a7cc42SQu Wenruo * Log trees are not exposed to user space thus can't be snapshotted, 127492a7cc42SQu Wenruo * and they go away before a real commit is actually done. 127592a7cc42SQu Wenruo * 127692a7cc42SQu Wenruo * They do store pointers to file data extents, and those reference 127792a7cc42SQu Wenruo * counts still get updated (along with back refs to the log tree). 12787237f183SYan Zheng */ 1279e02119d5SChris Mason 12804d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 12819631e4ccSJosef Bacik NULL, 0, 0, 0, BTRFS_NESTING_NORMAL); 12826ab6ebb7SNaohiro Aota if (IS_ERR(leaf)) 12836ab6ebb7SNaohiro Aota return PTR_ERR(leaf); 1284e02119d5SChris Mason 12857237f183SYan Zheng root->node = leaf; 1286e02119d5SChris Mason 1287e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1288e02119d5SChris Mason btrfs_tree_unlock(root->node); 12896ab6ebb7SNaohiro Aota 12906ab6ebb7SNaohiro Aota return 0; 12917237f183SYan Zheng } 12927237f183SYan Zheng 12937237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 12947237f183SYan Zheng struct btrfs_fs_info *fs_info) 12957237f183SYan Zheng { 12967237f183SYan Zheng struct btrfs_root *log_root; 12977237f183SYan Zheng 12987237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 12997237f183SYan Zheng if (IS_ERR(log_root)) 13007237f183SYan Zheng return PTR_ERR(log_root); 13016ab6ebb7SNaohiro Aota 13023ddebf27SNaohiro Aota if (!btrfs_is_zoned(fs_info)) { 13033ddebf27SNaohiro Aota int ret = btrfs_alloc_log_tree_node(trans, log_root); 13043ddebf27SNaohiro Aota 13056ab6ebb7SNaohiro Aota if (ret) { 13066ab6ebb7SNaohiro Aota btrfs_put_root(log_root); 13076ab6ebb7SNaohiro Aota return ret; 13086ab6ebb7SNaohiro Aota } 13093ddebf27SNaohiro Aota } 13106ab6ebb7SNaohiro Aota 13117237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 13127237f183SYan Zheng fs_info->log_root_tree = log_root; 13137237f183SYan Zheng return 0; 13147237f183SYan Zheng } 13157237f183SYan Zheng 13167237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 13177237f183SYan Zheng struct btrfs_root *root) 13187237f183SYan Zheng { 13190b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13207237f183SYan Zheng struct btrfs_root *log_root; 13217237f183SYan Zheng struct btrfs_inode_item *inode_item; 13226ab6ebb7SNaohiro Aota int ret; 13237237f183SYan Zheng 13240b246afaSJeff Mahoney log_root = alloc_log_tree(trans, fs_info); 13257237f183SYan Zheng if (IS_ERR(log_root)) 13267237f183SYan Zheng return PTR_ERR(log_root); 13277237f183SYan Zheng 13286ab6ebb7SNaohiro Aota ret = btrfs_alloc_log_tree_node(trans, log_root); 13296ab6ebb7SNaohiro Aota if (ret) { 13306ab6ebb7SNaohiro Aota btrfs_put_root(log_root); 13316ab6ebb7SNaohiro Aota return ret; 13326ab6ebb7SNaohiro Aota } 13336ab6ebb7SNaohiro Aota 13347237f183SYan Zheng log_root->last_trans = trans->transid; 13357237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 13367237f183SYan Zheng 13377237f183SYan Zheng inode_item = &log_root->root_item.inode; 13383cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 13393cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 13403cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 1341da17066cSJeff Mahoney btrfs_set_stack_inode_nbytes(inode_item, 13420b246afaSJeff Mahoney fs_info->nodesize); 13433cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 13447237f183SYan Zheng 13455d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 13467237f183SYan Zheng 13477237f183SYan Zheng WARN_ON(root->log_root); 13487237f183SYan Zheng root->log_root = log_root; 13497237f183SYan Zheng root->log_transid = 0; 1350d1433debSMiao Xie root->log_transid_committed = -1; 1351257c62e1SChris Mason root->last_log_commit = 0; 1352e02119d5SChris Mason return 0; 1353e02119d5SChris Mason } 1354e02119d5SChris Mason 135549d11beaSJosef Bacik static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root, 135649d11beaSJosef Bacik struct btrfs_path *path, 1357cb517eabSMiao Xie struct btrfs_key *key) 1358e02119d5SChris Mason { 1359e02119d5SChris Mason struct btrfs_root *root; 1360e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 136184234f3aSYan Zheng u64 generation; 1362cb517eabSMiao Xie int ret; 1363581c1760SQu Wenruo int level; 1364cb517eabSMiao Xie 136596dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS); 136649d11beaSJosef Bacik if (!root) 136749d11beaSJosef Bacik return ERR_PTR(-ENOMEM); 13680f7d52f4SChris Mason 1369cb517eabSMiao Xie ret = btrfs_find_root(tree_root, key, path, 1370cb517eabSMiao Xie &root->root_item, &root->root_key); 13710f7d52f4SChris Mason if (ret) { 137213a8a7c8SYan, Zheng if (ret > 0) 137313a8a7c8SYan, Zheng ret = -ENOENT; 137449d11beaSJosef Bacik goto fail; 13750f7d52f4SChris Mason } 137613a8a7c8SYan, Zheng 137784234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1378581c1760SQu Wenruo level = btrfs_root_level(&root->root_item); 13792ff7e61eSJeff Mahoney root->node = read_tree_block(fs_info, 13802ff7e61eSJeff Mahoney btrfs_root_bytenr(&root->root_item), 13811b7ec85eSJosef Bacik key->objectid, generation, level, NULL); 138264c043deSLiu Bo if (IS_ERR(root->node)) { 138364c043deSLiu Bo ret = PTR_ERR(root->node); 13848c38938cSJosef Bacik root->node = NULL; 138549d11beaSJosef Bacik goto fail; 13864eb150d6SQu Wenruo } 13874eb150d6SQu Wenruo if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1388cb517eabSMiao Xie ret = -EIO; 138949d11beaSJosef Bacik goto fail; 1390416bc658SJosef Bacik } 139188c602abSQu Wenruo 139288c602abSQu Wenruo /* 139388c602abSQu Wenruo * For real fs, and not log/reloc trees, root owner must 139488c602abSQu Wenruo * match its root node owner 139588c602abSQu Wenruo */ 139688c602abSQu Wenruo if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) && 139788c602abSQu Wenruo root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID && 139888c602abSQu Wenruo root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && 139988c602abSQu Wenruo root->root_key.objectid != btrfs_header_owner(root->node)) { 140088c602abSQu Wenruo btrfs_crit(fs_info, 140188c602abSQu Wenruo "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu", 140288c602abSQu Wenruo root->root_key.objectid, root->node->start, 140388c602abSQu Wenruo btrfs_header_owner(root->node), 140488c602abSQu Wenruo root->root_key.objectid); 140588c602abSQu Wenruo ret = -EUCLEAN; 140688c602abSQu Wenruo goto fail; 140788c602abSQu Wenruo } 14085d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 1409cb517eabSMiao Xie return root; 141049d11beaSJosef Bacik fail: 141100246528SJosef Bacik btrfs_put_root(root); 141249d11beaSJosef Bacik return ERR_PTR(ret); 141349d11beaSJosef Bacik } 141449d11beaSJosef Bacik 141549d11beaSJosef Bacik struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 141649d11beaSJosef Bacik struct btrfs_key *key) 141749d11beaSJosef Bacik { 141849d11beaSJosef Bacik struct btrfs_root *root; 141949d11beaSJosef Bacik struct btrfs_path *path; 142049d11beaSJosef Bacik 142149d11beaSJosef Bacik path = btrfs_alloc_path(); 142249d11beaSJosef Bacik if (!path) 142349d11beaSJosef Bacik return ERR_PTR(-ENOMEM); 142449d11beaSJosef Bacik root = read_tree_root_path(tree_root, path, key); 142549d11beaSJosef Bacik btrfs_free_path(path); 142649d11beaSJosef Bacik 142749d11beaSJosef Bacik return root; 1428cb517eabSMiao Xie } 1429cb517eabSMiao Xie 14302dfb1e43SQu Wenruo /* 14312dfb1e43SQu Wenruo * Initialize subvolume root in-memory structure 14322dfb1e43SQu Wenruo * 14332dfb1e43SQu Wenruo * @anon_dev: anonymous device to attach to the root, if zero, allocate new 14342dfb1e43SQu Wenruo */ 14352dfb1e43SQu Wenruo static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev) 1436cb517eabSMiao Xie { 1437cb517eabSMiao Xie int ret; 1438dcc3eb96SNikolay Borisov unsigned int nofs_flag; 1439cb517eabSMiao Xie 1440dcc3eb96SNikolay Borisov /* 1441dcc3eb96SNikolay Borisov * We might be called under a transaction (e.g. indirect backref 1442dcc3eb96SNikolay Borisov * resolution) which could deadlock if it triggers memory reclaim 1443dcc3eb96SNikolay Borisov */ 1444dcc3eb96SNikolay Borisov nofs_flag = memalloc_nofs_save(); 1445dcc3eb96SNikolay Borisov ret = btrfs_drew_lock_init(&root->snapshot_lock); 1446dcc3eb96SNikolay Borisov memalloc_nofs_restore(nofs_flag); 1447dcc3eb96SNikolay Borisov if (ret) 14488257b2dcSMiao Xie goto fail; 14498257b2dcSMiao Xie 1450aeb935a4SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID && 145137f00a6dSJohannes Thumshirn !btrfs_is_data_reloc_root(root)) { 145292a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &root->state); 1453f39e4571SJosef Bacik btrfs_check_and_init_root_item(&root->root_item); 1454f39e4571SJosef Bacik } 1455f39e4571SJosef Bacik 1456851fd730SQu Wenruo /* 1457851fd730SQu Wenruo * Don't assign anonymous block device to roots that are not exposed to 1458851fd730SQu Wenruo * userspace, the id pool is limited to 1M 1459851fd730SQu Wenruo */ 1460851fd730SQu Wenruo if (is_fstree(root->root_key.objectid) && 1461851fd730SQu Wenruo btrfs_root_refs(&root->root_item) > 0) { 14622dfb1e43SQu Wenruo if (!anon_dev) { 1463cb517eabSMiao Xie ret = get_anon_bdev(&root->anon_dev); 1464cb517eabSMiao Xie if (ret) 1465876d2cf1SLiu Bo goto fail; 14662dfb1e43SQu Wenruo } else { 14672dfb1e43SQu Wenruo root->anon_dev = anon_dev; 14682dfb1e43SQu Wenruo } 1469851fd730SQu Wenruo } 1470f32e48e9SChandan Rajendra 1471f32e48e9SChandan Rajendra mutex_lock(&root->objectid_mutex); 1472453e4873SNikolay Borisov ret = btrfs_init_root_free_objectid(root); 1473f32e48e9SChandan Rajendra if (ret) { 1474f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1475876d2cf1SLiu Bo goto fail; 1476f32e48e9SChandan Rajendra } 1477f32e48e9SChandan Rajendra 14786b8fad57SNikolay Borisov ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID); 1479f32e48e9SChandan Rajendra 1480f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1481f32e48e9SChandan Rajendra 1482cb517eabSMiao Xie return 0; 1483cb517eabSMiao Xie fail: 148484db5ccfSDavid Sterba /* The caller is responsible to call btrfs_free_fs_root */ 1485cb517eabSMiao Xie return ret; 1486cb517eabSMiao Xie } 1487cb517eabSMiao Xie 1488a98db0f3SJosef Bacik static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1489cb517eabSMiao Xie u64 root_id) 1490cb517eabSMiao Xie { 1491cb517eabSMiao Xie struct btrfs_root *root; 1492cb517eabSMiao Xie 1493fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1494fc7cbcd4SDavid Sterba root = radix_tree_lookup(&fs_info->fs_roots_radix, 1495fc7cbcd4SDavid Sterba (unsigned long)root_id); 1496bc44d7c4SJosef Bacik if (root) 149700246528SJosef Bacik root = btrfs_grab_root(root); 1498fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 1499cb517eabSMiao Xie return root; 1500cb517eabSMiao Xie } 1501cb517eabSMiao Xie 150249d11beaSJosef Bacik static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info, 150349d11beaSJosef Bacik u64 objectid) 150449d11beaSJosef Bacik { 1505abed4aaaSJosef Bacik struct btrfs_key key = { 1506abed4aaaSJosef Bacik .objectid = objectid, 1507abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 1508abed4aaaSJosef Bacik .offset = 0, 1509abed4aaaSJosef Bacik }; 1510abed4aaaSJosef Bacik 151149d11beaSJosef Bacik if (objectid == BTRFS_ROOT_TREE_OBJECTID) 151249d11beaSJosef Bacik return btrfs_grab_root(fs_info->tree_root); 151349d11beaSJosef Bacik if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 1514abed4aaaSJosef Bacik return btrfs_grab_root(btrfs_global_root(fs_info, &key)); 151549d11beaSJosef Bacik if (objectid == BTRFS_CHUNK_TREE_OBJECTID) 151649d11beaSJosef Bacik return btrfs_grab_root(fs_info->chunk_root); 151749d11beaSJosef Bacik if (objectid == BTRFS_DEV_TREE_OBJECTID) 151849d11beaSJosef Bacik return btrfs_grab_root(fs_info->dev_root); 151949d11beaSJosef Bacik if (objectid == BTRFS_CSUM_TREE_OBJECTID) 1520abed4aaaSJosef Bacik return btrfs_grab_root(btrfs_global_root(fs_info, &key)); 152149d11beaSJosef Bacik if (objectid == BTRFS_QUOTA_TREE_OBJECTID) 152249d11beaSJosef Bacik return btrfs_grab_root(fs_info->quota_root) ? 152349d11beaSJosef Bacik fs_info->quota_root : ERR_PTR(-ENOENT); 152449d11beaSJosef Bacik if (objectid == BTRFS_UUID_TREE_OBJECTID) 152549d11beaSJosef Bacik return btrfs_grab_root(fs_info->uuid_root) ? 152649d11beaSJosef Bacik fs_info->uuid_root : ERR_PTR(-ENOENT); 1527abed4aaaSJosef Bacik if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) { 1528abed4aaaSJosef Bacik struct btrfs_root *root = btrfs_global_root(fs_info, &key); 1529abed4aaaSJosef Bacik 1530abed4aaaSJosef Bacik return btrfs_grab_root(root) ? root : ERR_PTR(-ENOENT); 1531abed4aaaSJosef Bacik } 153249d11beaSJosef Bacik return NULL; 153349d11beaSJosef Bacik } 153449d11beaSJosef Bacik 1535cb517eabSMiao Xie int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1536cb517eabSMiao Xie struct btrfs_root *root) 1537cb517eabSMiao Xie { 1538cb517eabSMiao Xie int ret; 1539cb517eabSMiao Xie 1540fc7cbcd4SDavid Sterba ret = radix_tree_preload(GFP_NOFS); 1541fc7cbcd4SDavid Sterba if (ret) 1542fc7cbcd4SDavid Sterba return ret; 1543fc7cbcd4SDavid Sterba 1544fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1545fc7cbcd4SDavid Sterba ret = radix_tree_insert(&fs_info->fs_roots_radix, 1546fc7cbcd4SDavid Sterba (unsigned long)root->root_key.objectid, 1547fc7cbcd4SDavid Sterba root); 1548af01d2e5SJosef Bacik if (ret == 0) { 154900246528SJosef Bacik btrfs_grab_root(root); 1550fc7cbcd4SDavid Sterba set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1551af01d2e5SJosef Bacik } 1552fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 1553fc7cbcd4SDavid Sterba radix_tree_preload_end(); 1554cb517eabSMiao Xie 1555cb517eabSMiao Xie return ret; 1556cb517eabSMiao Xie } 1557cb517eabSMiao Xie 1558bd647ce3SJosef Bacik void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info) 1559bd647ce3SJosef Bacik { 1560bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 1561bd647ce3SJosef Bacik struct btrfs_root *root; 1562bd647ce3SJosef Bacik 1563bd647ce3SJosef Bacik while (!list_empty(&fs_info->allocated_roots)) { 1564457f1864SJosef Bacik char buf[BTRFS_ROOT_NAME_BUF_LEN]; 1565457f1864SJosef Bacik 1566bd647ce3SJosef Bacik root = list_first_entry(&fs_info->allocated_roots, 1567bd647ce3SJosef Bacik struct btrfs_root, leak_list); 1568457f1864SJosef Bacik btrfs_err(fs_info, "leaked root %s refcount %d", 156971008734SJosef Bacik btrfs_root_name(&root->root_key, buf), 1570bd647ce3SJosef Bacik refcount_read(&root->refs)); 1571bd647ce3SJosef Bacik while (refcount_read(&root->refs) > 1) 157200246528SJosef Bacik btrfs_put_root(root); 157300246528SJosef Bacik btrfs_put_root(root); 1574bd647ce3SJosef Bacik } 1575bd647ce3SJosef Bacik #endif 1576bd647ce3SJosef Bacik } 1577bd647ce3SJosef Bacik 1578abed4aaaSJosef Bacik static void free_global_roots(struct btrfs_fs_info *fs_info) 1579abed4aaaSJosef Bacik { 1580abed4aaaSJosef Bacik struct btrfs_root *root; 1581abed4aaaSJosef Bacik struct rb_node *node; 1582abed4aaaSJosef Bacik 1583abed4aaaSJosef Bacik while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) { 1584abed4aaaSJosef Bacik root = rb_entry(node, struct btrfs_root, rb_node); 1585abed4aaaSJosef Bacik rb_erase(&root->rb_node, &fs_info->global_root_tree); 1586abed4aaaSJosef Bacik btrfs_put_root(root); 1587abed4aaaSJosef Bacik } 1588abed4aaaSJosef Bacik } 1589abed4aaaSJosef Bacik 15900d4b0463SJosef Bacik void btrfs_free_fs_info(struct btrfs_fs_info *fs_info) 15910d4b0463SJosef Bacik { 1592141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 1593141386e1SJosef Bacik percpu_counter_destroy(&fs_info->delalloc_bytes); 15945deb17e1SJosef Bacik percpu_counter_destroy(&fs_info->ordered_bytes); 1595141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 1596141386e1SJosef Bacik btrfs_free_csum_hash(fs_info); 1597141386e1SJosef Bacik btrfs_free_stripe_hash_table(fs_info); 1598141386e1SJosef Bacik btrfs_free_ref_cache(fs_info); 15990d4b0463SJosef Bacik kfree(fs_info->balance_ctl); 16000d4b0463SJosef Bacik kfree(fs_info->delayed_root); 1601abed4aaaSJosef Bacik free_global_roots(fs_info); 160200246528SJosef Bacik btrfs_put_root(fs_info->tree_root); 160300246528SJosef Bacik btrfs_put_root(fs_info->chunk_root); 160400246528SJosef Bacik btrfs_put_root(fs_info->dev_root); 160500246528SJosef Bacik btrfs_put_root(fs_info->quota_root); 160600246528SJosef Bacik btrfs_put_root(fs_info->uuid_root); 160700246528SJosef Bacik btrfs_put_root(fs_info->fs_root); 1608aeb935a4SQu Wenruo btrfs_put_root(fs_info->data_reloc_root); 16099c54e80dSJosef Bacik btrfs_put_root(fs_info->block_group_root); 1610bd647ce3SJosef Bacik btrfs_check_leaked_roots(fs_info); 16113fd63727SJosef Bacik btrfs_extent_buffer_leak_debug_check(fs_info); 16120d4b0463SJosef Bacik kfree(fs_info->super_copy); 16130d4b0463SJosef Bacik kfree(fs_info->super_for_commit); 16148481dd80SQu Wenruo kfree(fs_info->subpage_info); 16150d4b0463SJosef Bacik kvfree(fs_info); 16160d4b0463SJosef Bacik } 16170d4b0463SJosef Bacik 16180d4b0463SJosef Bacik 16192dfb1e43SQu Wenruo /* 16202dfb1e43SQu Wenruo * Get an in-memory reference of a root structure. 16212dfb1e43SQu Wenruo * 16222dfb1e43SQu Wenruo * For essential trees like root/extent tree, we grab it from fs_info directly. 16232dfb1e43SQu Wenruo * For subvolume trees, we check the cached filesystem roots first. If not 16242dfb1e43SQu Wenruo * found, then read it from disk and add it to cached fs roots. 16252dfb1e43SQu Wenruo * 16262dfb1e43SQu Wenruo * Caller should release the root by calling btrfs_put_root() after the usage. 16272dfb1e43SQu Wenruo * 16282dfb1e43SQu Wenruo * NOTE: Reloc and log trees can't be read by this function as they share the 16292dfb1e43SQu Wenruo * same root objectid. 16302dfb1e43SQu Wenruo * 16312dfb1e43SQu Wenruo * @objectid: root id 16322dfb1e43SQu Wenruo * @anon_dev: preallocated anonymous block device number for new roots, 16332dfb1e43SQu Wenruo * pass 0 for new allocation. 16342dfb1e43SQu Wenruo * @check_ref: whether to check root item references, If true, return -ENOENT 16352dfb1e43SQu Wenruo * for orphan roots 16362dfb1e43SQu Wenruo */ 16372dfb1e43SQu Wenruo static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info, 16382dfb1e43SQu Wenruo u64 objectid, dev_t anon_dev, 16392dfb1e43SQu Wenruo bool check_ref) 16405eda7b5eSChris Mason { 16415eda7b5eSChris Mason struct btrfs_root *root; 1642381cf658SDavid Sterba struct btrfs_path *path; 16431d4c08e0SDavid Sterba struct btrfs_key key; 16445eda7b5eSChris Mason int ret; 16455eda7b5eSChris Mason 164649d11beaSJosef Bacik root = btrfs_get_global_root(fs_info, objectid); 164749d11beaSJosef Bacik if (root) 164849d11beaSJosef Bacik return root; 16494df27c4dSYan, Zheng again: 165056e9357aSDavid Sterba root = btrfs_lookup_fs_root(fs_info, objectid); 165148475471SStefan Behrens if (root) { 16522dfb1e43SQu Wenruo /* Shouldn't get preallocated anon_dev for cached roots */ 16532dfb1e43SQu Wenruo ASSERT(!anon_dev); 1654bc44d7c4SJosef Bacik if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 165500246528SJosef Bacik btrfs_put_root(root); 165648475471SStefan Behrens return ERR_PTR(-ENOENT); 1657bc44d7c4SJosef Bacik } 16585eda7b5eSChris Mason return root; 165948475471SStefan Behrens } 16605eda7b5eSChris Mason 166156e9357aSDavid Sterba key.objectid = objectid; 166256e9357aSDavid Sterba key.type = BTRFS_ROOT_ITEM_KEY; 166356e9357aSDavid Sterba key.offset = (u64)-1; 166456e9357aSDavid Sterba root = btrfs_read_tree_root(fs_info->tree_root, &key); 16655eda7b5eSChris Mason if (IS_ERR(root)) 16665eda7b5eSChris Mason return root; 16673394e160SChris Mason 1668c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1669d68fc57bSYan, Zheng ret = -ENOENT; 1670d68fc57bSYan, Zheng goto fail; 1671d68fc57bSYan, Zheng } 1672d68fc57bSYan, Zheng 16732dfb1e43SQu Wenruo ret = btrfs_init_fs_root(root, anon_dev); 1674cb517eabSMiao Xie if (ret) 1675cb517eabSMiao Xie goto fail; 1676cb517eabSMiao Xie 1677381cf658SDavid Sterba path = btrfs_alloc_path(); 1678381cf658SDavid Sterba if (!path) { 1679381cf658SDavid Sterba ret = -ENOMEM; 1680381cf658SDavid Sterba goto fail; 1681381cf658SDavid Sterba } 16821d4c08e0SDavid Sterba key.objectid = BTRFS_ORPHAN_OBJECTID; 16831d4c08e0SDavid Sterba key.type = BTRFS_ORPHAN_ITEM_KEY; 168456e9357aSDavid Sterba key.offset = objectid; 16851d4c08e0SDavid Sterba 16861d4c08e0SDavid Sterba ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1687381cf658SDavid Sterba btrfs_free_path(path); 1688d68fc57bSYan, Zheng if (ret < 0) 1689d68fc57bSYan, Zheng goto fail; 1690d68fc57bSYan, Zheng if (ret == 0) 169127cdeb70SMiao Xie set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1692d68fc57bSYan, Zheng 1693cb517eabSMiao Xie ret = btrfs_insert_fs_root(fs_info, root); 16940f7d52f4SChris Mason if (ret) { 1695168a2f77SJia-Ju Bai if (ret == -EEXIST) { 169600246528SJosef Bacik btrfs_put_root(root); 16974df27c4dSYan, Zheng goto again; 1698168a2f77SJia-Ju Bai } 16994df27c4dSYan, Zheng goto fail; 17004df27c4dSYan, Zheng } 1701edbd8d4eSChris Mason return root; 17024df27c4dSYan, Zheng fail: 170333fab972SFilipe Manana /* 170433fab972SFilipe Manana * If our caller provided us an anonymous device, then it's his 1705143823cfSDavid Sterba * responsibility to free it in case we fail. So we have to set our 170633fab972SFilipe Manana * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root() 170733fab972SFilipe Manana * and once again by our caller. 170833fab972SFilipe Manana */ 170933fab972SFilipe Manana if (anon_dev) 171033fab972SFilipe Manana root->anon_dev = 0; 17118c38938cSJosef Bacik btrfs_put_root(root); 17124df27c4dSYan, Zheng return ERR_PTR(ret); 1713edbd8d4eSChris Mason } 1714edbd8d4eSChris Mason 17152dfb1e43SQu Wenruo /* 17162dfb1e43SQu Wenruo * Get in-memory reference of a root structure 17172dfb1e43SQu Wenruo * 17182dfb1e43SQu Wenruo * @objectid: tree objectid 17192dfb1e43SQu Wenruo * @check_ref: if set, verify that the tree exists and the item has at least 17202dfb1e43SQu Wenruo * one reference 17212dfb1e43SQu Wenruo */ 17222dfb1e43SQu Wenruo struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 17232dfb1e43SQu Wenruo u64 objectid, bool check_ref) 17242dfb1e43SQu Wenruo { 17252dfb1e43SQu Wenruo return btrfs_get_root_ref(fs_info, objectid, 0, check_ref); 17262dfb1e43SQu Wenruo } 17272dfb1e43SQu Wenruo 17282dfb1e43SQu Wenruo /* 17292dfb1e43SQu Wenruo * Get in-memory reference of a root structure, created as new, optionally pass 17302dfb1e43SQu Wenruo * the anonymous block device id 17312dfb1e43SQu Wenruo * 17322dfb1e43SQu Wenruo * @objectid: tree objectid 17332dfb1e43SQu Wenruo * @anon_dev: if zero, allocate a new anonymous block device or use the 17342dfb1e43SQu Wenruo * parameter value 17352dfb1e43SQu Wenruo */ 17362dfb1e43SQu Wenruo struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info, 17372dfb1e43SQu Wenruo u64 objectid, dev_t anon_dev) 17382dfb1e43SQu Wenruo { 17392dfb1e43SQu Wenruo return btrfs_get_root_ref(fs_info, objectid, anon_dev, true); 17402dfb1e43SQu Wenruo } 17412dfb1e43SQu Wenruo 17428b712842SChris Mason /* 174349d11beaSJosef Bacik * btrfs_get_fs_root_commit_root - return a root for the given objectid 174449d11beaSJosef Bacik * @fs_info: the fs_info 174549d11beaSJosef Bacik * @objectid: the objectid we need to lookup 174649d11beaSJosef Bacik * 174749d11beaSJosef Bacik * This is exclusively used for backref walking, and exists specifically because 174849d11beaSJosef Bacik * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref 174949d11beaSJosef Bacik * creation time, which means we may have to read the tree_root in order to look 175049d11beaSJosef Bacik * up a fs root that is not in memory. If the root is not in memory we will 175149d11beaSJosef Bacik * read the tree root commit root and look up the fs root from there. This is a 175249d11beaSJosef Bacik * temporary root, it will not be inserted into the radix tree as it doesn't 175349d11beaSJosef Bacik * have the most uptodate information, it'll simply be discarded once the 175449d11beaSJosef Bacik * backref code is finished using the root. 175549d11beaSJosef Bacik */ 175649d11beaSJosef Bacik struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info, 175749d11beaSJosef Bacik struct btrfs_path *path, 175849d11beaSJosef Bacik u64 objectid) 175949d11beaSJosef Bacik { 176049d11beaSJosef Bacik struct btrfs_root *root; 176149d11beaSJosef Bacik struct btrfs_key key; 176249d11beaSJosef Bacik 176349d11beaSJosef Bacik ASSERT(path->search_commit_root && path->skip_locking); 176449d11beaSJosef Bacik 176549d11beaSJosef Bacik /* 176649d11beaSJosef Bacik * This can return -ENOENT if we ask for a root that doesn't exist, but 176749d11beaSJosef Bacik * since this is called via the backref walking code we won't be looking 176849d11beaSJosef Bacik * up a root that doesn't exist, unless there's corruption. So if root 176949d11beaSJosef Bacik * != NULL just return it. 177049d11beaSJosef Bacik */ 177149d11beaSJosef Bacik root = btrfs_get_global_root(fs_info, objectid); 177249d11beaSJosef Bacik if (root) 177349d11beaSJosef Bacik return root; 177449d11beaSJosef Bacik 177549d11beaSJosef Bacik root = btrfs_lookup_fs_root(fs_info, objectid); 177649d11beaSJosef Bacik if (root) 177749d11beaSJosef Bacik return root; 177849d11beaSJosef Bacik 177949d11beaSJosef Bacik key.objectid = objectid; 178049d11beaSJosef Bacik key.type = BTRFS_ROOT_ITEM_KEY; 178149d11beaSJosef Bacik key.offset = (u64)-1; 178249d11beaSJosef Bacik root = read_tree_root_path(fs_info->tree_root, path, &key); 178349d11beaSJosef Bacik btrfs_release_path(path); 178449d11beaSJosef Bacik 178549d11beaSJosef Bacik return root; 178649d11beaSJosef Bacik } 178749d11beaSJosef Bacik 1788a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1789a74a4b97SChris Mason { 17900d031dc4SYu Zhe struct btrfs_fs_info *fs_info = arg; 1791d0278245SMiao Xie int again; 1792a74a4b97SChris Mason 1793d6fd0ae2SOmar Sandoval while (1) { 1794d0278245SMiao Xie again = 0; 17959d1a2a3aSDavid Sterba 1796fd340d0fSJosef Bacik set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1797fd340d0fSJosef Bacik 1798d0278245SMiao Xie /* Make the cleaner go to sleep early. */ 17992ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) 1800d0278245SMiao Xie goto sleep; 1801d0278245SMiao Xie 180290c711abSZygo Blaxell /* 180390c711abSZygo Blaxell * Do not do anything if we might cause open_ctree() to block 180490c711abSZygo Blaxell * before we have finished mounting the filesystem. 180590c711abSZygo Blaxell */ 18060b246afaSJeff Mahoney if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 180790c711abSZygo Blaxell goto sleep; 180890c711abSZygo Blaxell 18090b246afaSJeff Mahoney if (!mutex_trylock(&fs_info->cleaner_mutex)) 1810d0278245SMiao Xie goto sleep; 1811d0278245SMiao Xie 1812dc7f370cSMiao Xie /* 1813dc7f370cSMiao Xie * Avoid the problem that we change the status of the fs 1814dc7f370cSMiao Xie * during the above check and trylock. 1815dc7f370cSMiao Xie */ 18162ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) { 18170b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1818dc7f370cSMiao Xie goto sleep; 1819dc7f370cSMiao Xie } 1820dc7f370cSMiao Xie 18212ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 1822c2d6cb16SFilipe Manana 182333c44184SJosef Bacik again = btrfs_clean_one_deleted_snapshot(fs_info); 18240b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1825a74a4b97SChris Mason 1826d0278245SMiao Xie /* 182705323cd1SMiao Xie * The defragger has dealt with the R/O remount and umount, 182805323cd1SMiao Xie * needn't do anything special here. 1829d0278245SMiao Xie */ 18300b246afaSJeff Mahoney btrfs_run_defrag_inodes(fs_info); 183167c5e7d4SFilipe Manana 183267c5e7d4SFilipe Manana /* 1833f3372065SJohannes Thumshirn * Acquires fs_info->reclaim_bgs_lock to avoid racing 183467c5e7d4SFilipe Manana * with relocation (btrfs_relocate_chunk) and relocation 183567c5e7d4SFilipe Manana * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 1836f3372065SJohannes Thumshirn * after acquiring fs_info->reclaim_bgs_lock. So we 183767c5e7d4SFilipe Manana * can't hold, nor need to, fs_info->cleaner_mutex when deleting 183867c5e7d4SFilipe Manana * unused block groups. 183967c5e7d4SFilipe Manana */ 18400b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 184118bb8bbfSJohannes Thumshirn 184218bb8bbfSJohannes Thumshirn /* 184318bb8bbfSJohannes Thumshirn * Reclaim block groups in the reclaim_bgs list after we deleted 184418bb8bbfSJohannes Thumshirn * all unused block_groups. This possibly gives us some more free 184518bb8bbfSJohannes Thumshirn * space. 184618bb8bbfSJohannes Thumshirn */ 184718bb8bbfSJohannes Thumshirn btrfs_reclaim_bgs(fs_info); 1848d0278245SMiao Xie sleep: 1849a0a1db70SFilipe Manana clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1850d6fd0ae2SOmar Sandoval if (kthread_should_park()) 1851d6fd0ae2SOmar Sandoval kthread_parkme(); 1852d6fd0ae2SOmar Sandoval if (kthread_should_stop()) 1853d6fd0ae2SOmar Sandoval return 0; 1854838fe188SJiri Kosina if (!again) { 1855a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 1856a74a4b97SChris Mason schedule(); 1857a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1858a74a4b97SChris Mason } 1859da288d28SFilipe Manana } 1860a74a4b97SChris Mason } 1861a74a4b97SChris Mason 1862a74a4b97SChris Mason static int transaction_kthread(void *arg) 1863a74a4b97SChris Mason { 1864a74a4b97SChris Mason struct btrfs_root *root = arg; 18650b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1866a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1867a74a4b97SChris Mason struct btrfs_transaction *cur; 18688929ecfaSYan, Zheng u64 transid; 1869643900beSNikolay Borisov time64_t delta; 1870a74a4b97SChris Mason unsigned long delay; 1871914b2007SJan Kara bool cannot_commit; 1872a74a4b97SChris Mason 1873a74a4b97SChris Mason do { 1874914b2007SJan Kara cannot_commit = false; 1875ba1bc00fSNikolay Borisov delay = msecs_to_jiffies(fs_info->commit_interval * 1000); 18760b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 1877a74a4b97SChris Mason 18780b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 18790b246afaSJeff Mahoney cur = fs_info->running_transaction; 1880a74a4b97SChris Mason if (!cur) { 18810b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1882a74a4b97SChris Mason goto sleep; 1883a74a4b97SChris Mason } 188431153d81SYan Zheng 1885643900beSNikolay Borisov delta = ktime_get_seconds() - cur->start_time; 1886fdfbf020SJosef Bacik if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) && 1887fdfbf020SJosef Bacik cur->state < TRANS_STATE_COMMIT_START && 1888643900beSNikolay Borisov delta < fs_info->commit_interval) { 18890b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1890fb8a7e94SNikolay Borisov delay -= msecs_to_jiffies((delta - 1) * 1000); 1891fb8a7e94SNikolay Borisov delay = min(delay, 1892fb8a7e94SNikolay Borisov msecs_to_jiffies(fs_info->commit_interval * 1000)); 1893a74a4b97SChris Mason goto sleep; 1894a74a4b97SChris Mason } 18958929ecfaSYan, Zheng transid = cur->transid; 18960b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 189756bec294SChris Mason 189879787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 1899354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 1900914b2007SJan Kara if (IS_ERR(trans)) { 1901354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 1902914b2007SJan Kara cannot_commit = true; 190379787eaaSJeff Mahoney goto sleep; 1904914b2007SJan Kara } 19058929ecfaSYan, Zheng if (transid == trans->transid) { 19063a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 19078929ecfaSYan, Zheng } else { 19083a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19098929ecfaSYan, Zheng } 1910a74a4b97SChris Mason sleep: 19110b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 19120b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 1913a74a4b97SChris Mason 191484961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) 19152ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 19168929ecfaSYan, Zheng if (!kthread_should_stop() && 19170b246afaSJeff Mahoney (!btrfs_transaction_blocked(fs_info) || 1918914b2007SJan Kara cannot_commit)) 1919bc5511d0SNikolay Borisov schedule_timeout_interruptible(delay); 1920a74a4b97SChris Mason } while (!kthread_should_stop()); 1921a74a4b97SChris Mason return 0; 1922a74a4b97SChris Mason } 1923a74a4b97SChris Mason 1924af31f5e5SChris Mason /* 192501f0f9daSNikolay Borisov * This will find the highest generation in the array of root backups. The 192601f0f9daSNikolay Borisov * index of the highest array is returned, or -EINVAL if we can't find 192701f0f9daSNikolay Borisov * anything. 1928af31f5e5SChris Mason * 1929af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1930af31f5e5SChris Mason * generation of the latest root in the array with the generation 1931af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1932af31f5e5SChris Mason */ 193301f0f9daSNikolay Borisov static int find_newest_super_backup(struct btrfs_fs_info *info) 1934af31f5e5SChris Mason { 193501f0f9daSNikolay Borisov const u64 newest_gen = btrfs_super_generation(info->super_copy); 1936af31f5e5SChris Mason u64 cur; 1937af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1938af31f5e5SChris Mason int i; 1939af31f5e5SChris Mason 1940af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1941af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1942af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1943af31f5e5SChris Mason if (cur == newest_gen) 194401f0f9daSNikolay Borisov return i; 1945af31f5e5SChris Mason } 1946af31f5e5SChris Mason 194701f0f9daSNikolay Borisov return -EINVAL; 1948af31f5e5SChris Mason } 1949af31f5e5SChris Mason 1950af31f5e5SChris Mason /* 1951af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1952af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1953af31f5e5SChris Mason * done 1954af31f5e5SChris Mason */ 1955af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1956af31f5e5SChris Mason { 19576ef108ddSNikolay Borisov const int next_backup = info->backup_root_index; 1958af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1959af31f5e5SChris Mason 1960af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1961af31f5e5SChris Mason 1962af31f5e5SChris Mason /* 1963af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1964af31f5e5SChris Mason * regardless of which ones we use today 1965af31f5e5SChris Mason */ 1966af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1967af31f5e5SChris Mason 1968af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1969af31f5e5SChris Mason 1970af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1971af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1972af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1973af31f5e5SChris Mason 1974af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1975af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1976af31f5e5SChris Mason 1977af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1978af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1979af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 1980af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 1981af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 1982af31f5e5SChris Mason 19839c54e80dSJosef Bacik if (btrfs_fs_incompat(info, EXTENT_TREE_V2)) { 19849c54e80dSJosef Bacik btrfs_set_backup_block_group_root(root_backup, 19859c54e80dSJosef Bacik info->block_group_root->node->start); 19869c54e80dSJosef Bacik btrfs_set_backup_block_group_root_gen(root_backup, 19879c54e80dSJosef Bacik btrfs_header_generation(info->block_group_root->node)); 19889c54e80dSJosef Bacik btrfs_set_backup_block_group_root_level(root_backup, 19899c54e80dSJosef Bacik btrfs_header_level(info->block_group_root->node)); 19909c54e80dSJosef Bacik } else { 19919c54e80dSJosef Bacik struct btrfs_root *extent_root = btrfs_extent_root(info, 0); 1992f7238e50SJosef Bacik struct btrfs_root *csum_root = btrfs_csum_root(info, 0); 19939c54e80dSJosef Bacik 19949c54e80dSJosef Bacik btrfs_set_backup_extent_root(root_backup, 19959c54e80dSJosef Bacik extent_root->node->start); 1996af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 199729cbcf40SJosef Bacik btrfs_header_generation(extent_root->node)); 1998af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 199929cbcf40SJosef Bacik btrfs_header_level(extent_root->node)); 2000af31f5e5SChris Mason 2001f7238e50SJosef Bacik btrfs_set_backup_csum_root(root_backup, csum_root->node->start); 2002f7238e50SJosef Bacik btrfs_set_backup_csum_root_gen(root_backup, 2003f7238e50SJosef Bacik btrfs_header_generation(csum_root->node)); 2004f7238e50SJosef Bacik btrfs_set_backup_csum_root_level(root_backup, 2005f7238e50SJosef Bacik btrfs_header_level(csum_root->node)); 20069c54e80dSJosef Bacik } 2007af31f5e5SChris Mason 20087c7e82a7SChris Mason /* 20097c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 20107c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 20117c7e82a7SChris Mason */ 20127c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 20137c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 20147c7e82a7SChris Mason info->fs_root->node->start); 2015af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 2016af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 2017af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 2018af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 20197c7e82a7SChris Mason } 2020af31f5e5SChris Mason 2021af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 2022af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 2023af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 2024af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 2025af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 2026af31f5e5SChris Mason 2027af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 2028af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 2029af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 2030af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 2031af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 2032af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 2033af31f5e5SChris Mason 2034af31f5e5SChris Mason /* 2035af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 2036af31f5e5SChris Mason * for the next commit 2037af31f5e5SChris Mason */ 2038af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 2039af31f5e5SChris Mason &info->super_for_commit->super_roots, 2040af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 2041af31f5e5SChris Mason } 2042af31f5e5SChris Mason 2043af31f5e5SChris Mason /* 2044bd2336b2SNikolay Borisov * read_backup_root - Reads a backup root based on the passed priority. Prio 0 2045bd2336b2SNikolay Borisov * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots 2046bd2336b2SNikolay Borisov * 2047bd2336b2SNikolay Borisov * fs_info - filesystem whose backup roots need to be read 2048bd2336b2SNikolay Borisov * priority - priority of backup root required 2049bd2336b2SNikolay Borisov * 2050bd2336b2SNikolay Borisov * Returns backup root index on success and -EINVAL otherwise. 2051bd2336b2SNikolay Borisov */ 2052bd2336b2SNikolay Borisov static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority) 2053bd2336b2SNikolay Borisov { 2054bd2336b2SNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 2055bd2336b2SNikolay Borisov struct btrfs_super_block *super = fs_info->super_copy; 2056bd2336b2SNikolay Borisov struct btrfs_root_backup *root_backup; 2057bd2336b2SNikolay Borisov 2058bd2336b2SNikolay Borisov if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) { 2059bd2336b2SNikolay Borisov if (priority == 0) 2060bd2336b2SNikolay Borisov return backup_index; 2061bd2336b2SNikolay Borisov 2062bd2336b2SNikolay Borisov backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority; 2063bd2336b2SNikolay Borisov backup_index %= BTRFS_NUM_BACKUP_ROOTS; 2064bd2336b2SNikolay Borisov } else { 2065bd2336b2SNikolay Borisov return -EINVAL; 2066bd2336b2SNikolay Borisov } 2067bd2336b2SNikolay Borisov 2068bd2336b2SNikolay Borisov root_backup = super->super_roots + backup_index; 2069bd2336b2SNikolay Borisov 2070bd2336b2SNikolay Borisov btrfs_set_super_generation(super, 2071bd2336b2SNikolay Borisov btrfs_backup_tree_root_gen(root_backup)); 2072bd2336b2SNikolay Borisov btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 2073bd2336b2SNikolay Borisov btrfs_set_super_root_level(super, 2074bd2336b2SNikolay Borisov btrfs_backup_tree_root_level(root_backup)); 2075bd2336b2SNikolay Borisov btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 2076bd2336b2SNikolay Borisov 2077bd2336b2SNikolay Borisov /* 2078bd2336b2SNikolay Borisov * Fixme: the total bytes and num_devices need to match or we should 2079bd2336b2SNikolay Borisov * need a fsck 2080bd2336b2SNikolay Borisov */ 2081bd2336b2SNikolay Borisov btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 2082bd2336b2SNikolay Borisov btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 2083bd2336b2SNikolay Borisov 2084bd2336b2SNikolay Borisov return backup_index; 2085bd2336b2SNikolay Borisov } 2086bd2336b2SNikolay Borisov 20877abadb64SLiu Bo /* helper to cleanup workers */ 20887abadb64SLiu Bo static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 20897abadb64SLiu Bo { 2090dc6e3209SQu Wenruo btrfs_destroy_workqueue(fs_info->fixup_workers); 2091afe3d242SQu Wenruo btrfs_destroy_workqueue(fs_info->delalloc_workers); 2092a31b4a43SChristoph Hellwig btrfs_destroy_workqueue(fs_info->hipri_workers); 20935cdc7ad3SQu Wenruo btrfs_destroy_workqueue(fs_info->workers); 2094d7b9416fSChristoph Hellwig if (fs_info->endio_workers) 2095d7b9416fSChristoph Hellwig destroy_workqueue(fs_info->endio_workers); 2096d34e123dSChristoph Hellwig if (fs_info->endio_raid56_workers) 2097d34e123dSChristoph Hellwig destroy_workqueue(fs_info->endio_raid56_workers); 2098385de0efSChristoph Hellwig if (fs_info->rmw_workers) 2099385de0efSChristoph Hellwig destroy_workqueue(fs_info->rmw_workers); 2100fed8a72dSChristoph Hellwig if (fs_info->compressed_write_workers) 2101fed8a72dSChristoph Hellwig destroy_workqueue(fs_info->compressed_write_workers); 2102fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_write_workers); 2103fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 21045b3bc44eSQu Wenruo btrfs_destroy_workqueue(fs_info->delayed_workers); 2105e66f0bb1SQu Wenruo btrfs_destroy_workqueue(fs_info->caching_workers); 2106a44903abSQu Wenruo btrfs_destroy_workqueue(fs_info->flush_workers); 2107fc97fab0SQu Wenruo btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2108b0643e59SDennis Zhou if (fs_info->discard_ctl.discard_workers) 2109b0643e59SDennis Zhou destroy_workqueue(fs_info->discard_ctl.discard_workers); 2110a9b9477dSFilipe Manana /* 2111a9b9477dSFilipe Manana * Now that all other work queues are destroyed, we can safely destroy 2112a9b9477dSFilipe Manana * the queues used for metadata I/O, since tasks from those other work 2113a9b9477dSFilipe Manana * queues can do metadata I/O operations. 2114a9b9477dSFilipe Manana */ 2115d7b9416fSChristoph Hellwig if (fs_info->endio_meta_workers) 2116d7b9416fSChristoph Hellwig destroy_workqueue(fs_info->endio_meta_workers); 21177abadb64SLiu Bo } 21187abadb64SLiu Bo 21192e9f5954SRashika static void free_root_extent_buffers(struct btrfs_root *root) 21202e9f5954SRashika { 21212e9f5954SRashika if (root) { 21222e9f5954SRashika free_extent_buffer(root->node); 21232e9f5954SRashika free_extent_buffer(root->commit_root); 21242e9f5954SRashika root->node = NULL; 21252e9f5954SRashika root->commit_root = NULL; 21262e9f5954SRashika } 21272e9f5954SRashika } 21282e9f5954SRashika 2129abed4aaaSJosef Bacik static void free_global_root_pointers(struct btrfs_fs_info *fs_info) 2130abed4aaaSJosef Bacik { 2131abed4aaaSJosef Bacik struct btrfs_root *root, *tmp; 2132abed4aaaSJosef Bacik 2133abed4aaaSJosef Bacik rbtree_postorder_for_each_entry_safe(root, tmp, 2134abed4aaaSJosef Bacik &fs_info->global_root_tree, 2135abed4aaaSJosef Bacik rb_node) 2136abed4aaaSJosef Bacik free_root_extent_buffers(root); 2137abed4aaaSJosef Bacik } 2138abed4aaaSJosef Bacik 2139af31f5e5SChris Mason /* helper to cleanup tree roots */ 21404273eaffSAnand Jain static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root) 2141af31f5e5SChris Mason { 21422e9f5954SRashika free_root_extent_buffers(info->tree_root); 2143655b09feSJosef Bacik 2144abed4aaaSJosef Bacik free_global_root_pointers(info); 21452e9f5954SRashika free_root_extent_buffers(info->dev_root); 21462e9f5954SRashika free_root_extent_buffers(info->quota_root); 21472e9f5954SRashika free_root_extent_buffers(info->uuid_root); 21488c38938cSJosef Bacik free_root_extent_buffers(info->fs_root); 2149aeb935a4SQu Wenruo free_root_extent_buffers(info->data_reloc_root); 21509c54e80dSJosef Bacik free_root_extent_buffers(info->block_group_root); 21514273eaffSAnand Jain if (free_chunk_root) 21522e9f5954SRashika free_root_extent_buffers(info->chunk_root); 2153af31f5e5SChris Mason } 2154af31f5e5SChris Mason 21558c38938cSJosef Bacik void btrfs_put_root(struct btrfs_root *root) 21568c38938cSJosef Bacik { 21578c38938cSJosef Bacik if (!root) 21588c38938cSJosef Bacik return; 21598c38938cSJosef Bacik 21608c38938cSJosef Bacik if (refcount_dec_and_test(&root->refs)) { 21618c38938cSJosef Bacik WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 21621dae7e0eSQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)); 21638c38938cSJosef Bacik if (root->anon_dev) 21648c38938cSJosef Bacik free_anon_bdev(root->anon_dev); 21658c38938cSJosef Bacik btrfs_drew_lock_destroy(&root->snapshot_lock); 2166923eb523SJohannes Thumshirn free_root_extent_buffers(root); 21678c38938cSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 2168fc7cbcd4SDavid Sterba spin_lock(&root->fs_info->fs_roots_radix_lock); 21698c38938cSJosef Bacik list_del_init(&root->leak_list); 2170fc7cbcd4SDavid Sterba spin_unlock(&root->fs_info->fs_roots_radix_lock); 21718c38938cSJosef Bacik #endif 21728c38938cSJosef Bacik kfree(root); 21738c38938cSJosef Bacik } 21748c38938cSJosef Bacik } 21758c38938cSJosef Bacik 2176faa2dbf0SJosef Bacik void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2177171f6537SJosef Bacik { 2178fc7cbcd4SDavid Sterba int ret; 2179fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 2180fc7cbcd4SDavid Sterba int i; 2181171f6537SJosef Bacik 2182171f6537SJosef Bacik while (!list_empty(&fs_info->dead_roots)) { 2183fc7cbcd4SDavid Sterba gang[0] = list_entry(fs_info->dead_roots.next, 2184171f6537SJosef Bacik struct btrfs_root, root_list); 2185fc7cbcd4SDavid Sterba list_del(&gang[0]->root_list); 2186171f6537SJosef Bacik 2187fc7cbcd4SDavid Sterba if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) 2188fc7cbcd4SDavid Sterba btrfs_drop_and_free_fs_root(fs_info, gang[0]); 2189fc7cbcd4SDavid Sterba btrfs_put_root(gang[0]); 2190171f6537SJosef Bacik } 2191171f6537SJosef Bacik 2192fc7cbcd4SDavid Sterba while (1) { 2193fc7cbcd4SDavid Sterba ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2194fc7cbcd4SDavid Sterba (void **)gang, 0, 2195fc7cbcd4SDavid Sterba ARRAY_SIZE(gang)); 2196fc7cbcd4SDavid Sterba if (!ret) 2197fc7cbcd4SDavid Sterba break; 2198fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) 2199fc7cbcd4SDavid Sterba btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2200171f6537SJosef Bacik } 2201171f6537SJosef Bacik } 2202af31f5e5SChris Mason 2203638aa7edSEric Sandeen static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2204638aa7edSEric Sandeen { 2205638aa7edSEric Sandeen mutex_init(&fs_info->scrub_lock); 2206638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_running, 0); 2207638aa7edSEric Sandeen atomic_set(&fs_info->scrub_pause_req, 0); 2208638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_paused, 0); 2209638aa7edSEric Sandeen atomic_set(&fs_info->scrub_cancel_req, 0); 2210638aa7edSEric Sandeen init_waitqueue_head(&fs_info->scrub_pause_wait); 2211ff09c4caSAnand Jain refcount_set(&fs_info->scrub_workers_refcnt, 0); 2212638aa7edSEric Sandeen } 2213638aa7edSEric Sandeen 2214779a65a4SEric Sandeen static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2215779a65a4SEric Sandeen { 2216779a65a4SEric Sandeen spin_lock_init(&fs_info->balance_lock); 2217779a65a4SEric Sandeen mutex_init(&fs_info->balance_mutex); 2218779a65a4SEric Sandeen atomic_set(&fs_info->balance_pause_req, 0); 2219779a65a4SEric Sandeen atomic_set(&fs_info->balance_cancel_req, 0); 2220779a65a4SEric Sandeen fs_info->balance_ctl = NULL; 2221779a65a4SEric Sandeen init_waitqueue_head(&fs_info->balance_wait_q); 2222907d2710SDavid Sterba atomic_set(&fs_info->reloc_cancel_req, 0); 2223779a65a4SEric Sandeen } 2224779a65a4SEric Sandeen 22256bccf3abSJeff Mahoney static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info) 2226f37938e0SEric Sandeen { 22272ff7e61eSJeff Mahoney struct inode *inode = fs_info->btree_inode; 22282ff7e61eSJeff Mahoney 22292ff7e61eSJeff Mahoney inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 22302ff7e61eSJeff Mahoney set_nlink(inode, 1); 2231f37938e0SEric Sandeen /* 2232f37938e0SEric Sandeen * we set the i_size on the btree inode to the max possible int. 2233f37938e0SEric Sandeen * the real end of the address space is determined by all of 2234f37938e0SEric Sandeen * the devices in the system 2235f37938e0SEric Sandeen */ 22362ff7e61eSJeff Mahoney inode->i_size = OFFSET_MAX; 22372ff7e61eSJeff Mahoney inode->i_mapping->a_ops = &btree_aops; 2238f37938e0SEric Sandeen 22392ff7e61eSJeff Mahoney RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 224043eb5f29SQu Wenruo extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree, 22412c53a14dSQu Wenruo IO_TREE_BTREE_INODE_IO, inode); 22427b439738SDavid Sterba BTRFS_I(inode)->io_tree.track_uptodate = false; 22432ff7e61eSJeff Mahoney extent_map_tree_init(&BTRFS_I(inode)->extent_tree); 2244f37938e0SEric Sandeen 22455c8fd99fSJosef Bacik BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root); 2246adac5584SFilipe Manana BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID; 2247adac5584SFilipe Manana BTRFS_I(inode)->location.type = 0; 2248adac5584SFilipe Manana BTRFS_I(inode)->location.offset = 0; 22492ff7e61eSJeff Mahoney set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); 22502ff7e61eSJeff Mahoney btrfs_insert_inode_hash(inode); 2251f37938e0SEric Sandeen } 2252f37938e0SEric Sandeen 2253ad618368SEric Sandeen static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2254ad618368SEric Sandeen { 2255ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2256129827e3SDavid Sterba init_rwsem(&fs_info->dev_replace.rwsem); 22577f8d236aSDavid Sterba init_waitqueue_head(&fs_info->dev_replace.replace_wait); 2258ad618368SEric Sandeen } 2259ad618368SEric Sandeen 2260f9e92e40SEric Sandeen static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2261f9e92e40SEric Sandeen { 2262f9e92e40SEric Sandeen spin_lock_init(&fs_info->qgroup_lock); 2263f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_ioctl_lock); 2264f9e92e40SEric Sandeen fs_info->qgroup_tree = RB_ROOT; 2265f9e92e40SEric Sandeen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2266f9e92e40SEric Sandeen fs_info->qgroup_seq = 1; 2267f9e92e40SEric Sandeen fs_info->qgroup_ulist = NULL; 2268d2c609b8SJeff Mahoney fs_info->qgroup_rescan_running = false; 2269f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_rescan_lock); 2270f9e92e40SEric Sandeen } 2271f9e92e40SEric Sandeen 2272d21deec5SSu Yue static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info) 22732a458198SEric Sandeen { 2274f7b885beSAnand Jain u32 max_active = fs_info->thread_pool_size; 22756f011058SDavid Sterba unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 22762a458198SEric Sandeen 22772a458198SEric Sandeen fs_info->workers = 2278a31b4a43SChristoph Hellwig btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16); 2279a31b4a43SChristoph Hellwig fs_info->hipri_workers = 2280a31b4a43SChristoph Hellwig btrfs_alloc_workqueue(fs_info, "worker-high", 2281cb001095SJeff Mahoney flags | WQ_HIGHPRI, max_active, 16); 22822a458198SEric Sandeen 22832a458198SEric Sandeen fs_info->delalloc_workers = 2284cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delalloc", 2285cb001095SJeff Mahoney flags, max_active, 2); 22862a458198SEric Sandeen 22872a458198SEric Sandeen fs_info->flush_workers = 2288cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "flush_delalloc", 2289cb001095SJeff Mahoney flags, max_active, 0); 22902a458198SEric Sandeen 22912a458198SEric Sandeen fs_info->caching_workers = 2292cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); 22932a458198SEric Sandeen 22942a458198SEric Sandeen fs_info->fixup_workers = 2295cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); 22962a458198SEric Sandeen 22972a458198SEric Sandeen fs_info->endio_workers = 2298d7b9416fSChristoph Hellwig alloc_workqueue("btrfs-endio", flags, max_active); 22992a458198SEric Sandeen fs_info->endio_meta_workers = 2300d7b9416fSChristoph Hellwig alloc_workqueue("btrfs-endio-meta", flags, max_active); 23012a458198SEric Sandeen fs_info->endio_raid56_workers = 2302d34e123dSChristoph Hellwig alloc_workqueue("btrfs-endio-raid56", flags, max_active); 2303385de0efSChristoph Hellwig fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active); 23042a458198SEric Sandeen fs_info->endio_write_workers = 2305cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-write", flags, 2306cb001095SJeff Mahoney max_active, 2); 2307fed8a72dSChristoph Hellwig fs_info->compressed_write_workers = 2308fed8a72dSChristoph Hellwig alloc_workqueue("btrfs-compressed-write", flags, max_active); 23092a458198SEric Sandeen fs_info->endio_freespace_worker = 2310cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "freespace-write", flags, 2311cb001095SJeff Mahoney max_active, 0); 23122a458198SEric Sandeen fs_info->delayed_workers = 2313cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, 2314cb001095SJeff Mahoney max_active, 0); 23152a458198SEric Sandeen fs_info->qgroup_rescan_workers = 2316cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); 2317b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers = 2318b0643e59SDennis Zhou alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1); 23192a458198SEric Sandeen 2320a31b4a43SChristoph Hellwig if (!(fs_info->workers && fs_info->hipri_workers && 2321a31b4a43SChristoph Hellwig fs_info->delalloc_workers && fs_info->flush_workers && 23222a458198SEric Sandeen fs_info->endio_workers && fs_info->endio_meta_workers && 2323fed8a72dSChristoph Hellwig fs_info->compressed_write_workers && 23242a458198SEric Sandeen fs_info->endio_write_workers && fs_info->endio_raid56_workers && 23252a458198SEric Sandeen fs_info->endio_freespace_worker && fs_info->rmw_workers && 2326f26c9238SQu Wenruo fs_info->caching_workers && fs_info->fixup_workers && 2327f26c9238SQu Wenruo fs_info->delayed_workers && fs_info->qgroup_rescan_workers && 2328b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers)) { 23292a458198SEric Sandeen return -ENOMEM; 23302a458198SEric Sandeen } 23312a458198SEric Sandeen 23322a458198SEric Sandeen return 0; 23332a458198SEric Sandeen } 23342a458198SEric Sandeen 23356d97c6e3SJohannes Thumshirn static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type) 23366d97c6e3SJohannes Thumshirn { 23376d97c6e3SJohannes Thumshirn struct crypto_shash *csum_shash; 2338b4e967beSDavid Sterba const char *csum_driver = btrfs_super_csum_driver(csum_type); 23396d97c6e3SJohannes Thumshirn 2340b4e967beSDavid Sterba csum_shash = crypto_alloc_shash(csum_driver, 0, 0); 23416d97c6e3SJohannes Thumshirn 23426d97c6e3SJohannes Thumshirn if (IS_ERR(csum_shash)) { 23436d97c6e3SJohannes Thumshirn btrfs_err(fs_info, "error allocating %s hash for checksum", 2344b4e967beSDavid Sterba csum_driver); 23456d97c6e3SJohannes Thumshirn return PTR_ERR(csum_shash); 23466d97c6e3SJohannes Thumshirn } 23476d97c6e3SJohannes Thumshirn 23486d97c6e3SJohannes Thumshirn fs_info->csum_shash = csum_shash; 23496d97c6e3SJohannes Thumshirn 2350c8a5f8caSDavid Sterba btrfs_info(fs_info, "using %s (%s) checksum algorithm", 2351c8a5f8caSDavid Sterba btrfs_super_csum_name(csum_type), 2352c8a5f8caSDavid Sterba crypto_shash_driver_name(csum_shash)); 23536d97c6e3SJohannes Thumshirn return 0; 23546d97c6e3SJohannes Thumshirn } 23556d97c6e3SJohannes Thumshirn 235663443bf5SEric Sandeen static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 235763443bf5SEric Sandeen struct btrfs_fs_devices *fs_devices) 235863443bf5SEric Sandeen { 235963443bf5SEric Sandeen int ret; 236063443bf5SEric Sandeen struct btrfs_root *log_tree_root; 236163443bf5SEric Sandeen struct btrfs_super_block *disk_super = fs_info->super_copy; 236263443bf5SEric Sandeen u64 bytenr = btrfs_super_log_root(disk_super); 2363581c1760SQu Wenruo int level = btrfs_super_log_root_level(disk_super); 236463443bf5SEric Sandeen 236563443bf5SEric Sandeen if (fs_devices->rw_devices == 0) { 2366f14d104dSDavid Sterba btrfs_warn(fs_info, "log replay required on RO media"); 236763443bf5SEric Sandeen return -EIO; 236863443bf5SEric Sandeen } 236963443bf5SEric Sandeen 237096dfcb46SJosef Bacik log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, 237196dfcb46SJosef Bacik GFP_KERNEL); 237263443bf5SEric Sandeen if (!log_tree_root) 237363443bf5SEric Sandeen return -ENOMEM; 237463443bf5SEric Sandeen 23752ff7e61eSJeff Mahoney log_tree_root->node = read_tree_block(fs_info, bytenr, 23761b7ec85eSJosef Bacik BTRFS_TREE_LOG_OBJECTID, 23771b7ec85eSJosef Bacik fs_info->generation + 1, level, 23781b7ec85eSJosef Bacik NULL); 237964c043deSLiu Bo if (IS_ERR(log_tree_root->node)) { 2380f14d104dSDavid Sterba btrfs_warn(fs_info, "failed to read log tree"); 23810eeff236SLiu Bo ret = PTR_ERR(log_tree_root->node); 23828c38938cSJosef Bacik log_tree_root->node = NULL; 238300246528SJosef Bacik btrfs_put_root(log_tree_root); 23840eeff236SLiu Bo return ret; 23854eb150d6SQu Wenruo } 23864eb150d6SQu Wenruo if (!extent_buffer_uptodate(log_tree_root->node)) { 2387f14d104dSDavid Sterba btrfs_err(fs_info, "failed to read log tree"); 238800246528SJosef Bacik btrfs_put_root(log_tree_root); 238963443bf5SEric Sandeen return -EIO; 239063443bf5SEric Sandeen } 23914eb150d6SQu Wenruo 239263443bf5SEric Sandeen /* returns with log_tree_root freed on success */ 239363443bf5SEric Sandeen ret = btrfs_recover_log_trees(log_tree_root); 239463443bf5SEric Sandeen if (ret) { 23950b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 239663443bf5SEric Sandeen "Failed to recover log tree"); 239700246528SJosef Bacik btrfs_put_root(log_tree_root); 239863443bf5SEric Sandeen return ret; 239963443bf5SEric Sandeen } 240063443bf5SEric Sandeen 2401bc98a42cSDavid Howells if (sb_rdonly(fs_info->sb)) { 24026bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 240363443bf5SEric Sandeen if (ret) 240463443bf5SEric Sandeen return ret; 240563443bf5SEric Sandeen } 240663443bf5SEric Sandeen 240763443bf5SEric Sandeen return 0; 240863443bf5SEric Sandeen } 240963443bf5SEric Sandeen 2410abed4aaaSJosef Bacik static int load_global_roots_objectid(struct btrfs_root *tree_root, 2411abed4aaaSJosef Bacik struct btrfs_path *path, u64 objectid, 2412abed4aaaSJosef Bacik const char *name) 2413abed4aaaSJosef Bacik { 2414abed4aaaSJosef Bacik struct btrfs_fs_info *fs_info = tree_root->fs_info; 2415abed4aaaSJosef Bacik struct btrfs_root *root; 2416f7238e50SJosef Bacik u64 max_global_id = 0; 2417abed4aaaSJosef Bacik int ret; 2418abed4aaaSJosef Bacik struct btrfs_key key = { 2419abed4aaaSJosef Bacik .objectid = objectid, 2420abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 2421abed4aaaSJosef Bacik .offset = 0, 2422abed4aaaSJosef Bacik }; 2423abed4aaaSJosef Bacik bool found = false; 2424abed4aaaSJosef Bacik 2425abed4aaaSJosef Bacik /* If we have IGNOREDATACSUMS skip loading these roots. */ 2426abed4aaaSJosef Bacik if (objectid == BTRFS_CSUM_TREE_OBJECTID && 2427abed4aaaSJosef Bacik btrfs_test_opt(fs_info, IGNOREDATACSUMS)) { 2428abed4aaaSJosef Bacik set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state); 2429abed4aaaSJosef Bacik return 0; 2430abed4aaaSJosef Bacik } 2431abed4aaaSJosef Bacik 2432abed4aaaSJosef Bacik while (1) { 2433abed4aaaSJosef Bacik ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0); 2434abed4aaaSJosef Bacik if (ret < 0) 2435abed4aaaSJosef Bacik break; 2436abed4aaaSJosef Bacik 2437abed4aaaSJosef Bacik if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 2438abed4aaaSJosef Bacik ret = btrfs_next_leaf(tree_root, path); 2439abed4aaaSJosef Bacik if (ret) { 2440abed4aaaSJosef Bacik if (ret > 0) 2441abed4aaaSJosef Bacik ret = 0; 2442abed4aaaSJosef Bacik break; 2443abed4aaaSJosef Bacik } 2444abed4aaaSJosef Bacik } 2445abed4aaaSJosef Bacik ret = 0; 2446abed4aaaSJosef Bacik 2447abed4aaaSJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 2448abed4aaaSJosef Bacik if (key.objectid != objectid) 2449abed4aaaSJosef Bacik break; 2450abed4aaaSJosef Bacik btrfs_release_path(path); 2451abed4aaaSJosef Bacik 2452f7238e50SJosef Bacik /* 2453f7238e50SJosef Bacik * Just worry about this for extent tree, it'll be the same for 2454f7238e50SJosef Bacik * everybody. 2455f7238e50SJosef Bacik */ 2456f7238e50SJosef Bacik if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 2457f7238e50SJosef Bacik max_global_id = max(max_global_id, key.offset); 2458f7238e50SJosef Bacik 2459abed4aaaSJosef Bacik found = true; 2460abed4aaaSJosef Bacik root = read_tree_root_path(tree_root, path, &key); 2461abed4aaaSJosef Bacik if (IS_ERR(root)) { 2462abed4aaaSJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) 2463abed4aaaSJosef Bacik ret = PTR_ERR(root); 2464abed4aaaSJosef Bacik break; 2465abed4aaaSJosef Bacik } 2466abed4aaaSJosef Bacik set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2467abed4aaaSJosef Bacik ret = btrfs_global_root_insert(root); 2468abed4aaaSJosef Bacik if (ret) { 2469abed4aaaSJosef Bacik btrfs_put_root(root); 2470abed4aaaSJosef Bacik break; 2471abed4aaaSJosef Bacik } 2472abed4aaaSJosef Bacik key.offset++; 2473abed4aaaSJosef Bacik } 2474abed4aaaSJosef Bacik btrfs_release_path(path); 2475abed4aaaSJosef Bacik 2476f7238e50SJosef Bacik if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 2477f7238e50SJosef Bacik fs_info->nr_global_roots = max_global_id + 1; 2478f7238e50SJosef Bacik 2479abed4aaaSJosef Bacik if (!found || ret) { 2480abed4aaaSJosef Bacik if (objectid == BTRFS_CSUM_TREE_OBJECTID) 2481abed4aaaSJosef Bacik set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state); 2482abed4aaaSJosef Bacik 2483abed4aaaSJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) 2484abed4aaaSJosef Bacik ret = ret ? ret : -ENOENT; 2485abed4aaaSJosef Bacik else 2486abed4aaaSJosef Bacik ret = 0; 2487abed4aaaSJosef Bacik btrfs_err(fs_info, "failed to load root %s", name); 2488abed4aaaSJosef Bacik } 2489abed4aaaSJosef Bacik return ret; 2490abed4aaaSJosef Bacik } 2491abed4aaaSJosef Bacik 2492abed4aaaSJosef Bacik static int load_global_roots(struct btrfs_root *tree_root) 2493abed4aaaSJosef Bacik { 2494abed4aaaSJosef Bacik struct btrfs_path *path; 2495abed4aaaSJosef Bacik int ret = 0; 2496abed4aaaSJosef Bacik 2497abed4aaaSJosef Bacik path = btrfs_alloc_path(); 2498abed4aaaSJosef Bacik if (!path) 2499abed4aaaSJosef Bacik return -ENOMEM; 2500abed4aaaSJosef Bacik 2501abed4aaaSJosef Bacik ret = load_global_roots_objectid(tree_root, path, 2502abed4aaaSJosef Bacik BTRFS_EXTENT_TREE_OBJECTID, "extent"); 2503abed4aaaSJosef Bacik if (ret) 2504abed4aaaSJosef Bacik goto out; 2505abed4aaaSJosef Bacik ret = load_global_roots_objectid(tree_root, path, 2506abed4aaaSJosef Bacik BTRFS_CSUM_TREE_OBJECTID, "csum"); 2507abed4aaaSJosef Bacik if (ret) 2508abed4aaaSJosef Bacik goto out; 2509abed4aaaSJosef Bacik if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE)) 2510abed4aaaSJosef Bacik goto out; 2511abed4aaaSJosef Bacik ret = load_global_roots_objectid(tree_root, path, 2512abed4aaaSJosef Bacik BTRFS_FREE_SPACE_TREE_OBJECTID, 2513abed4aaaSJosef Bacik "free space"); 2514abed4aaaSJosef Bacik out: 2515abed4aaaSJosef Bacik btrfs_free_path(path); 2516abed4aaaSJosef Bacik return ret; 2517abed4aaaSJosef Bacik } 2518abed4aaaSJosef Bacik 25196bccf3abSJeff Mahoney static int btrfs_read_roots(struct btrfs_fs_info *fs_info) 25204bbcaa64SEric Sandeen { 25216bccf3abSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 2522a4f3d2c4SDavid Sterba struct btrfs_root *root; 25234bbcaa64SEric Sandeen struct btrfs_key location; 25244bbcaa64SEric Sandeen int ret; 25254bbcaa64SEric Sandeen 25266bccf3abSJeff Mahoney BUG_ON(!fs_info->tree_root); 25276bccf3abSJeff Mahoney 2528abed4aaaSJosef Bacik ret = load_global_roots(tree_root); 2529abed4aaaSJosef Bacik if (ret) 2530abed4aaaSJosef Bacik return ret; 2531abed4aaaSJosef Bacik 2532abed4aaaSJosef Bacik location.objectid = BTRFS_DEV_TREE_OBJECTID; 25334bbcaa64SEric Sandeen location.type = BTRFS_ROOT_ITEM_KEY; 25344bbcaa64SEric Sandeen location.offset = 0; 25354bbcaa64SEric Sandeen 2536a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2537f50f4353SLiu Bo if (IS_ERR(root)) { 253842437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2539f50f4353SLiu Bo ret = PTR_ERR(root); 2540f50f4353SLiu Bo goto out; 2541f50f4353SLiu Bo } 254242437a63SJosef Bacik } else { 2543a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2544a4f3d2c4SDavid Sterba fs_info->dev_root = root; 254542437a63SJosef Bacik } 2546820a49daSJosef Bacik /* Initialize fs_info for all devices in any case */ 2547820a49daSJosef Bacik btrfs_init_devices_late(fs_info); 25484bbcaa64SEric Sandeen 2549aeb935a4SQu Wenruo /* 2550aeb935a4SQu Wenruo * This tree can share blocks with some other fs tree during relocation 2551aeb935a4SQu Wenruo * and we need a proper setup by btrfs_get_fs_root 2552aeb935a4SQu Wenruo */ 255356e9357aSDavid Sterba root = btrfs_get_fs_root(tree_root->fs_info, 255456e9357aSDavid Sterba BTRFS_DATA_RELOC_TREE_OBJECTID, true); 2555aeb935a4SQu Wenruo if (IS_ERR(root)) { 255642437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2557aeb935a4SQu Wenruo ret = PTR_ERR(root); 2558aeb935a4SQu Wenruo goto out; 2559aeb935a4SQu Wenruo } 256042437a63SJosef Bacik } else { 2561aeb935a4SQu Wenruo set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2562aeb935a4SQu Wenruo fs_info->data_reloc_root = root; 256342437a63SJosef Bacik } 2564aeb935a4SQu Wenruo 25654bbcaa64SEric Sandeen location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2566a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2567a4f3d2c4SDavid Sterba if (!IS_ERR(root)) { 2568a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2569afcdd129SJosef Bacik set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); 2570a4f3d2c4SDavid Sterba fs_info->quota_root = root; 25714bbcaa64SEric Sandeen } 25724bbcaa64SEric Sandeen 25734bbcaa64SEric Sandeen location.objectid = BTRFS_UUID_TREE_OBJECTID; 2574a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2575a4f3d2c4SDavid Sterba if (IS_ERR(root)) { 257642437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2577a4f3d2c4SDavid Sterba ret = PTR_ERR(root); 25784bbcaa64SEric Sandeen if (ret != -ENOENT) 2579f50f4353SLiu Bo goto out; 258042437a63SJosef Bacik } 25814bbcaa64SEric Sandeen } else { 2582a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2583a4f3d2c4SDavid Sterba fs_info->uuid_root = root; 25844bbcaa64SEric Sandeen } 25854bbcaa64SEric Sandeen 25864bbcaa64SEric Sandeen return 0; 2587f50f4353SLiu Bo out: 2588f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", 2589f50f4353SLiu Bo location.objectid, ret); 2590f50f4353SLiu Bo return ret; 25914bbcaa64SEric Sandeen } 25924bbcaa64SEric Sandeen 2593069ec957SQu Wenruo /* 2594069ec957SQu Wenruo * Real super block validation 2595069ec957SQu Wenruo * NOTE: super csum type and incompat features will not be checked here. 2596069ec957SQu Wenruo * 2597069ec957SQu Wenruo * @sb: super block to check 2598069ec957SQu Wenruo * @mirror_num: the super block number to check its bytenr: 2599069ec957SQu Wenruo * 0 the primary (1st) sb 2600069ec957SQu Wenruo * 1, 2 2nd and 3rd backup copy 2601069ec957SQu Wenruo * -1 skip bytenr check 2602069ec957SQu Wenruo */ 2603069ec957SQu Wenruo static int validate_super(struct btrfs_fs_info *fs_info, 2604069ec957SQu Wenruo struct btrfs_super_block *sb, int mirror_num) 260521a852b0SQu Wenruo { 260621a852b0SQu Wenruo u64 nodesize = btrfs_super_nodesize(sb); 260721a852b0SQu Wenruo u64 sectorsize = btrfs_super_sectorsize(sb); 260821a852b0SQu Wenruo int ret = 0; 260921a852b0SQu Wenruo 261021a852b0SQu Wenruo if (btrfs_super_magic(sb) != BTRFS_MAGIC) { 261121a852b0SQu Wenruo btrfs_err(fs_info, "no valid FS found"); 261221a852b0SQu Wenruo ret = -EINVAL; 261321a852b0SQu Wenruo } 261421a852b0SQu Wenruo if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { 261521a852b0SQu Wenruo btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", 261621a852b0SQu Wenruo btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); 261721a852b0SQu Wenruo ret = -EINVAL; 261821a852b0SQu Wenruo } 261921a852b0SQu Wenruo if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 262021a852b0SQu Wenruo btrfs_err(fs_info, "tree_root level too big: %d >= %d", 262121a852b0SQu Wenruo btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 262221a852b0SQu Wenruo ret = -EINVAL; 262321a852b0SQu Wenruo } 262421a852b0SQu Wenruo if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 262521a852b0SQu Wenruo btrfs_err(fs_info, "chunk_root level too big: %d >= %d", 262621a852b0SQu Wenruo btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 262721a852b0SQu Wenruo ret = -EINVAL; 262821a852b0SQu Wenruo } 262921a852b0SQu Wenruo if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 263021a852b0SQu Wenruo btrfs_err(fs_info, "log_root level too big: %d >= %d", 263121a852b0SQu Wenruo btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 263221a852b0SQu Wenruo ret = -EINVAL; 263321a852b0SQu Wenruo } 263421a852b0SQu Wenruo 263521a852b0SQu Wenruo /* 263621a852b0SQu Wenruo * Check sectorsize and nodesize first, other check will need it. 263721a852b0SQu Wenruo * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. 263821a852b0SQu Wenruo */ 263921a852b0SQu Wenruo if (!is_power_of_2(sectorsize) || sectorsize < 4096 || 264021a852b0SQu Wenruo sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { 264121a852b0SQu Wenruo btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); 264221a852b0SQu Wenruo ret = -EINVAL; 264321a852b0SQu Wenruo } 26440bb3eb3eSQu Wenruo 26450bb3eb3eSQu Wenruo /* 26461a42daabSQu Wenruo * We only support at most two sectorsizes: 4K and PAGE_SIZE. 26471a42daabSQu Wenruo * 26481a42daabSQu Wenruo * We can support 16K sectorsize with 64K page size without problem, 26491a42daabSQu Wenruo * but such sectorsize/pagesize combination doesn't make much sense. 26501a42daabSQu Wenruo * 4K will be our future standard, PAGE_SIZE is supported from the very 26511a42daabSQu Wenruo * beginning. 26520bb3eb3eSQu Wenruo */ 26531a42daabSQu Wenruo if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) { 265421a852b0SQu Wenruo btrfs_err(fs_info, 26550bb3eb3eSQu Wenruo "sectorsize %llu not yet supported for page size %lu", 265621a852b0SQu Wenruo sectorsize, PAGE_SIZE); 265721a852b0SQu Wenruo ret = -EINVAL; 265821a852b0SQu Wenruo } 26590bb3eb3eSQu Wenruo 266021a852b0SQu Wenruo if (!is_power_of_2(nodesize) || nodesize < sectorsize || 266121a852b0SQu Wenruo nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { 266221a852b0SQu Wenruo btrfs_err(fs_info, "invalid nodesize %llu", nodesize); 266321a852b0SQu Wenruo ret = -EINVAL; 266421a852b0SQu Wenruo } 266521a852b0SQu Wenruo if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { 266621a852b0SQu Wenruo btrfs_err(fs_info, "invalid leafsize %u, should be %llu", 266721a852b0SQu Wenruo le32_to_cpu(sb->__unused_leafsize), nodesize); 266821a852b0SQu Wenruo ret = -EINVAL; 266921a852b0SQu Wenruo } 267021a852b0SQu Wenruo 267121a852b0SQu Wenruo /* Root alignment check */ 267221a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { 267321a852b0SQu Wenruo btrfs_warn(fs_info, "tree_root block unaligned: %llu", 267421a852b0SQu Wenruo btrfs_super_root(sb)); 267521a852b0SQu Wenruo ret = -EINVAL; 267621a852b0SQu Wenruo } 267721a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { 267821a852b0SQu Wenruo btrfs_warn(fs_info, "chunk_root block unaligned: %llu", 267921a852b0SQu Wenruo btrfs_super_chunk_root(sb)); 268021a852b0SQu Wenruo ret = -EINVAL; 268121a852b0SQu Wenruo } 268221a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { 268321a852b0SQu Wenruo btrfs_warn(fs_info, "log_root block unaligned: %llu", 268421a852b0SQu Wenruo btrfs_super_log_root(sb)); 268521a852b0SQu Wenruo ret = -EINVAL; 268621a852b0SQu Wenruo } 268721a852b0SQu Wenruo 2688aefd7f70SNikolay Borisov if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, 2689aefd7f70SNikolay Borisov BTRFS_FSID_SIZE)) { 2690aefd7f70SNikolay Borisov btrfs_err(fs_info, 2691aefd7f70SNikolay Borisov "superblock fsid doesn't match fsid of fs_devices: %pU != %pU", 2692aefd7f70SNikolay Borisov fs_info->super_copy->fsid, fs_info->fs_devices->fsid); 2693aefd7f70SNikolay Borisov ret = -EINVAL; 2694aefd7f70SNikolay Borisov } 2695aefd7f70SNikolay Borisov 2696aefd7f70SNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID) && 2697aefd7f70SNikolay Borisov memcmp(fs_info->fs_devices->metadata_uuid, 2698aefd7f70SNikolay Borisov fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) { 2699aefd7f70SNikolay Borisov btrfs_err(fs_info, 2700aefd7f70SNikolay Borisov "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU", 2701aefd7f70SNikolay Borisov fs_info->super_copy->metadata_uuid, 2702aefd7f70SNikolay Borisov fs_info->fs_devices->metadata_uuid); 2703aefd7f70SNikolay Borisov ret = -EINVAL; 2704aefd7f70SNikolay Borisov } 2705aefd7f70SNikolay Borisov 2706de37aa51SNikolay Borisov if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, 27077239ff4bSNikolay Borisov BTRFS_FSID_SIZE) != 0) { 270821a852b0SQu Wenruo btrfs_err(fs_info, 27097239ff4bSNikolay Borisov "dev_item UUID does not match metadata fsid: %pU != %pU", 2710de37aa51SNikolay Borisov fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); 271121a852b0SQu Wenruo ret = -EINVAL; 271221a852b0SQu Wenruo } 271321a852b0SQu Wenruo 271421a852b0SQu Wenruo /* 271521a852b0SQu Wenruo * Hint to catch really bogus numbers, bitflips or so, more exact checks are 271621a852b0SQu Wenruo * done later 271721a852b0SQu Wenruo */ 271821a852b0SQu Wenruo if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { 271921a852b0SQu Wenruo btrfs_err(fs_info, "bytes_used is too small %llu", 272021a852b0SQu Wenruo btrfs_super_bytes_used(sb)); 272121a852b0SQu Wenruo ret = -EINVAL; 272221a852b0SQu Wenruo } 272321a852b0SQu Wenruo if (!is_power_of_2(btrfs_super_stripesize(sb))) { 272421a852b0SQu Wenruo btrfs_err(fs_info, "invalid stripesize %u", 272521a852b0SQu Wenruo btrfs_super_stripesize(sb)); 272621a852b0SQu Wenruo ret = -EINVAL; 272721a852b0SQu Wenruo } 272821a852b0SQu Wenruo if (btrfs_super_num_devices(sb) > (1UL << 31)) 272921a852b0SQu Wenruo btrfs_warn(fs_info, "suspicious number of devices: %llu", 273021a852b0SQu Wenruo btrfs_super_num_devices(sb)); 273121a852b0SQu Wenruo if (btrfs_super_num_devices(sb) == 0) { 273221a852b0SQu Wenruo btrfs_err(fs_info, "number of devices is 0"); 273321a852b0SQu Wenruo ret = -EINVAL; 273421a852b0SQu Wenruo } 273521a852b0SQu Wenruo 2736069ec957SQu Wenruo if (mirror_num >= 0 && 2737069ec957SQu Wenruo btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { 273821a852b0SQu Wenruo btrfs_err(fs_info, "super offset mismatch %llu != %u", 273921a852b0SQu Wenruo btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 274021a852b0SQu Wenruo ret = -EINVAL; 274121a852b0SQu Wenruo } 274221a852b0SQu Wenruo 274321a852b0SQu Wenruo /* 274421a852b0SQu Wenruo * Obvious sys_chunk_array corruptions, it must hold at least one key 274521a852b0SQu Wenruo * and one chunk 274621a852b0SQu Wenruo */ 274721a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 274821a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too big %u > %u", 274921a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 275021a852b0SQu Wenruo BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 275121a852b0SQu Wenruo ret = -EINVAL; 275221a852b0SQu Wenruo } 275321a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 275421a852b0SQu Wenruo + sizeof(struct btrfs_chunk)) { 275521a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too small %u < %zu", 275621a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 275721a852b0SQu Wenruo sizeof(struct btrfs_disk_key) 275821a852b0SQu Wenruo + sizeof(struct btrfs_chunk)); 275921a852b0SQu Wenruo ret = -EINVAL; 276021a852b0SQu Wenruo } 276121a852b0SQu Wenruo 276221a852b0SQu Wenruo /* 276321a852b0SQu Wenruo * The generation is a global counter, we'll trust it more than the others 276421a852b0SQu Wenruo * but it's still possible that it's the one that's wrong. 276521a852b0SQu Wenruo */ 276621a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 276721a852b0SQu Wenruo btrfs_warn(fs_info, 276821a852b0SQu Wenruo "suspicious: generation < chunk_root_generation: %llu < %llu", 276921a852b0SQu Wenruo btrfs_super_generation(sb), 277021a852b0SQu Wenruo btrfs_super_chunk_root_generation(sb)); 277121a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 277221a852b0SQu Wenruo && btrfs_super_cache_generation(sb) != (u64)-1) 277321a852b0SQu Wenruo btrfs_warn(fs_info, 277421a852b0SQu Wenruo "suspicious: generation < cache_generation: %llu < %llu", 277521a852b0SQu Wenruo btrfs_super_generation(sb), 277621a852b0SQu Wenruo btrfs_super_cache_generation(sb)); 277721a852b0SQu Wenruo 277821a852b0SQu Wenruo return ret; 277921a852b0SQu Wenruo } 278021a852b0SQu Wenruo 2781069ec957SQu Wenruo /* 2782069ec957SQu Wenruo * Validation of super block at mount time. 2783069ec957SQu Wenruo * Some checks already done early at mount time, like csum type and incompat 2784069ec957SQu Wenruo * flags will be skipped. 2785069ec957SQu Wenruo */ 2786069ec957SQu Wenruo static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) 2787069ec957SQu Wenruo { 2788069ec957SQu Wenruo return validate_super(fs_info, fs_info->super_copy, 0); 2789069ec957SQu Wenruo } 2790069ec957SQu Wenruo 279175cb857dSQu Wenruo /* 279275cb857dSQu Wenruo * Validation of super block at write time. 279375cb857dSQu Wenruo * Some checks like bytenr check will be skipped as their values will be 279475cb857dSQu Wenruo * overwritten soon. 279575cb857dSQu Wenruo * Extra checks like csum type and incompat flags will be done here. 279675cb857dSQu Wenruo */ 279775cb857dSQu Wenruo static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, 279875cb857dSQu Wenruo struct btrfs_super_block *sb) 279975cb857dSQu Wenruo { 280075cb857dSQu Wenruo int ret; 280175cb857dSQu Wenruo 280275cb857dSQu Wenruo ret = validate_super(fs_info, sb, -1); 280375cb857dSQu Wenruo if (ret < 0) 280475cb857dSQu Wenruo goto out; 2805e7e16f48SJohannes Thumshirn if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) { 280675cb857dSQu Wenruo ret = -EUCLEAN; 280775cb857dSQu Wenruo btrfs_err(fs_info, "invalid csum type, has %u want %u", 280875cb857dSQu Wenruo btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); 280975cb857dSQu Wenruo goto out; 281075cb857dSQu Wenruo } 281175cb857dSQu Wenruo if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 281275cb857dSQu Wenruo ret = -EUCLEAN; 281375cb857dSQu Wenruo btrfs_err(fs_info, 281475cb857dSQu Wenruo "invalid incompat flags, has 0x%llx valid mask 0x%llx", 281575cb857dSQu Wenruo btrfs_super_incompat_flags(sb), 281675cb857dSQu Wenruo (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); 281775cb857dSQu Wenruo goto out; 281875cb857dSQu Wenruo } 281975cb857dSQu Wenruo out: 282075cb857dSQu Wenruo if (ret < 0) 282175cb857dSQu Wenruo btrfs_err(fs_info, 282275cb857dSQu Wenruo "super block corruption detected before writing it to disk"); 282375cb857dSQu Wenruo return ret; 282475cb857dSQu Wenruo } 282575cb857dSQu Wenruo 2826bd676446SJosef Bacik static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level) 2827bd676446SJosef Bacik { 2828bd676446SJosef Bacik int ret = 0; 2829bd676446SJosef Bacik 2830bd676446SJosef Bacik root->node = read_tree_block(root->fs_info, bytenr, 2831bd676446SJosef Bacik root->root_key.objectid, gen, level, NULL); 2832bd676446SJosef Bacik if (IS_ERR(root->node)) { 2833bd676446SJosef Bacik ret = PTR_ERR(root->node); 2834bd676446SJosef Bacik root->node = NULL; 28354eb150d6SQu Wenruo return ret; 28364eb150d6SQu Wenruo } 28374eb150d6SQu Wenruo if (!extent_buffer_uptodate(root->node)) { 2838bd676446SJosef Bacik free_extent_buffer(root->node); 2839bd676446SJosef Bacik root->node = NULL; 28404eb150d6SQu Wenruo return -EIO; 2841bd676446SJosef Bacik } 2842bd676446SJosef Bacik 2843bd676446SJosef Bacik btrfs_set_root_node(&root->root_item, root->node); 2844bd676446SJosef Bacik root->commit_root = btrfs_root_node(root); 2845bd676446SJosef Bacik btrfs_set_root_refs(&root->root_item, 1); 2846bd676446SJosef Bacik return ret; 2847bd676446SJosef Bacik } 2848bd676446SJosef Bacik 2849bd676446SJosef Bacik static int load_important_roots(struct btrfs_fs_info *fs_info) 2850bd676446SJosef Bacik { 2851bd676446SJosef Bacik struct btrfs_super_block *sb = fs_info->super_copy; 2852bd676446SJosef Bacik u64 gen, bytenr; 2853bd676446SJosef Bacik int level, ret; 2854bd676446SJosef Bacik 2855bd676446SJosef Bacik bytenr = btrfs_super_root(sb); 2856bd676446SJosef Bacik gen = btrfs_super_generation(sb); 2857bd676446SJosef Bacik level = btrfs_super_root_level(sb); 2858bd676446SJosef Bacik ret = load_super_root(fs_info->tree_root, bytenr, gen, level); 28599c54e80dSJosef Bacik if (ret) { 2860bd676446SJosef Bacik btrfs_warn(fs_info, "couldn't read tree root"); 2861bd676446SJosef Bacik return ret; 2862bd676446SJosef Bacik } 2863bd676446SJosef Bacik 28649c54e80dSJosef Bacik if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) 28659c54e80dSJosef Bacik return 0; 28669c54e80dSJosef Bacik 28679c54e80dSJosef Bacik bytenr = btrfs_super_block_group_root(sb); 28689c54e80dSJosef Bacik gen = btrfs_super_block_group_root_generation(sb); 28699c54e80dSJosef Bacik level = btrfs_super_block_group_root_level(sb); 28709c54e80dSJosef Bacik ret = load_super_root(fs_info->block_group_root, bytenr, gen, level); 28719c54e80dSJosef Bacik if (ret) 28729c54e80dSJosef Bacik btrfs_warn(fs_info, "couldn't read block group root"); 28739c54e80dSJosef Bacik return ret; 28749c54e80dSJosef Bacik } 28759c54e80dSJosef Bacik 28766ef108ddSNikolay Borisov static int __cold init_tree_roots(struct btrfs_fs_info *fs_info) 2877b8522a1eSNikolay Borisov { 28786ef108ddSNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 2879b8522a1eSNikolay Borisov struct btrfs_super_block *sb = fs_info->super_copy; 2880b8522a1eSNikolay Borisov struct btrfs_root *tree_root = fs_info->tree_root; 2881b8522a1eSNikolay Borisov bool handle_error = false; 2882b8522a1eSNikolay Borisov int ret = 0; 2883b8522a1eSNikolay Borisov int i; 2884b8522a1eSNikolay Borisov 28859c54e80dSJosef Bacik if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 28869c54e80dSJosef Bacik struct btrfs_root *root; 2887b8522a1eSNikolay Borisov 28889c54e80dSJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_BLOCK_GROUP_TREE_OBJECTID, 28899c54e80dSJosef Bacik GFP_KERNEL); 28909c54e80dSJosef Bacik if (!root) 28919c54e80dSJosef Bacik return -ENOMEM; 28929c54e80dSJosef Bacik fs_info->block_group_root = root; 28939c54e80dSJosef Bacik } 28949c54e80dSJosef Bacik 2895b8522a1eSNikolay Borisov for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 2896b8522a1eSNikolay Borisov if (handle_error) { 2897b8522a1eSNikolay Borisov if (!IS_ERR(tree_root->node)) 2898b8522a1eSNikolay Borisov free_extent_buffer(tree_root->node); 2899b8522a1eSNikolay Borisov tree_root->node = NULL; 2900b8522a1eSNikolay Borisov 2901b8522a1eSNikolay Borisov if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) 2902b8522a1eSNikolay Borisov break; 2903b8522a1eSNikolay Borisov 2904b8522a1eSNikolay Borisov free_root_pointers(fs_info, 0); 2905b8522a1eSNikolay Borisov 2906b8522a1eSNikolay Borisov /* 2907b8522a1eSNikolay Borisov * Don't use the log in recovery mode, it won't be 2908b8522a1eSNikolay Borisov * valid 2909b8522a1eSNikolay Borisov */ 2910b8522a1eSNikolay Borisov btrfs_set_super_log_root(sb, 0); 2911b8522a1eSNikolay Borisov 2912b8522a1eSNikolay Borisov /* We can't trust the free space cache either */ 2913b8522a1eSNikolay Borisov btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 2914b8522a1eSNikolay Borisov 2915b8522a1eSNikolay Borisov ret = read_backup_root(fs_info, i); 29166ef108ddSNikolay Borisov backup_index = ret; 2917b8522a1eSNikolay Borisov if (ret < 0) 2918b8522a1eSNikolay Borisov return ret; 2919b8522a1eSNikolay Borisov } 2920b8522a1eSNikolay Borisov 2921bd676446SJosef Bacik ret = load_important_roots(fs_info); 2922bd676446SJosef Bacik if (ret) { 2923217f5004SNikolay Borisov handle_error = true; 2924b8522a1eSNikolay Borisov continue; 2925b8522a1eSNikolay Borisov } 2926b8522a1eSNikolay Borisov 2927336a0d8dSNikolay Borisov /* 2928336a0d8dSNikolay Borisov * No need to hold btrfs_root::objectid_mutex since the fs 2929336a0d8dSNikolay Borisov * hasn't been fully initialised and we are the only user 2930336a0d8dSNikolay Borisov */ 2931453e4873SNikolay Borisov ret = btrfs_init_root_free_objectid(tree_root); 2932b8522a1eSNikolay Borisov if (ret < 0) { 2933b8522a1eSNikolay Borisov handle_error = true; 2934b8522a1eSNikolay Borisov continue; 2935b8522a1eSNikolay Borisov } 2936b8522a1eSNikolay Borisov 29376b8fad57SNikolay Borisov ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID); 2938b8522a1eSNikolay Borisov 2939b8522a1eSNikolay Borisov ret = btrfs_read_roots(fs_info); 2940b8522a1eSNikolay Borisov if (ret < 0) { 2941b8522a1eSNikolay Borisov handle_error = true; 2942b8522a1eSNikolay Borisov continue; 2943b8522a1eSNikolay Borisov } 2944b8522a1eSNikolay Borisov 2945b8522a1eSNikolay Borisov /* All successful */ 2946bd676446SJosef Bacik fs_info->generation = btrfs_header_generation(tree_root->node); 2947bd676446SJosef Bacik fs_info->last_trans_committed = fs_info->generation; 2948d96b3424SFilipe Manana fs_info->last_reloc_trans = 0; 29496ef108ddSNikolay Borisov 29506ef108ddSNikolay Borisov /* Always begin writing backup roots after the one being used */ 29516ef108ddSNikolay Borisov if (backup_index < 0) { 29526ef108ddSNikolay Borisov fs_info->backup_root_index = 0; 29536ef108ddSNikolay Borisov } else { 29546ef108ddSNikolay Borisov fs_info->backup_root_index = backup_index + 1; 29556ef108ddSNikolay Borisov fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS; 29566ef108ddSNikolay Borisov } 2957b8522a1eSNikolay Borisov break; 2958b8522a1eSNikolay Borisov } 2959b8522a1eSNikolay Borisov 2960b8522a1eSNikolay Borisov return ret; 2961b8522a1eSNikolay Borisov } 2962b8522a1eSNikolay Borisov 29638260edbaSJosef Bacik void btrfs_init_fs_info(struct btrfs_fs_info *fs_info) 2964eb60ceacSChris Mason { 2965fc7cbcd4SDavid Sterba INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 296601cd3909SDavid Sterba INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 29678fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 2968facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 296924bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 2970eb73c1b7SMiao Xie INIT_LIST_HEAD(&fs_info->delalloc_roots); 297111833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 2972eb73c1b7SMiao Xie spin_lock_init(&fs_info->delalloc_root_lock); 2973a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 2974fc7cbcd4SDavid Sterba spin_lock_init(&fs_info->fs_roots_radix_lock); 297524bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 29764cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 2977ceda0864SMiao Xie spin_lock_init(&fs_info->super_lock); 2978f28491e0SJosef Bacik spin_lock_init(&fs_info->buffer_lock); 297947ab2a6cSJosef Bacik spin_lock_init(&fs_info->unused_bgs_lock); 298040ab3be1SNaohiro Aota spin_lock_init(&fs_info->treelog_bg_lock); 2981afba2bc0SNaohiro Aota spin_lock_init(&fs_info->zone_active_bgs_lock); 2982c2707a25SJohannes Thumshirn spin_lock_init(&fs_info->relocation_bg_lock); 2983f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 2984abed4aaaSJosef Bacik rwlock_init(&fs_info->global_root_lock); 2985d7c15171SZhao Lei mutex_init(&fs_info->unused_bg_unpin_mutex); 2986f3372065SJohannes Thumshirn mutex_init(&fs_info->reclaim_bgs_lock); 29877585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 2988573bfb72SMiao Xie mutex_init(&fs_info->delalloc_root_mutex); 29890bc09ca1SNaohiro Aota mutex_init(&fs_info->zoned_meta_io_lock); 29905f0addf7SNaohiro Aota mutex_init(&fs_info->zoned_data_reloc_io_lock); 2991de98ced9SMiao Xie seqlock_init(&fs_info->profiles_lock); 299219c00ddcSChris Mason 2993*e1489b4fSIoannis Angelakopoulos btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers); 2994*e1489b4fSIoannis Angelakopoulos 29950b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 29966324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 2997f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 299847ab2a6cSJosef Bacik INIT_LIST_HEAD(&fs_info->unused_bgs); 299918bb8bbfSJohannes Thumshirn INIT_LIST_HEAD(&fs_info->reclaim_bgs); 3000afba2bc0SNaohiro Aota INIT_LIST_HEAD(&fs_info->zone_active_bgs); 3001bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 3002bd647ce3SJosef Bacik INIT_LIST_HEAD(&fs_info->allocated_roots); 30033fd63727SJosef Bacik INIT_LIST_HEAD(&fs_info->allocated_ebs); 30043fd63727SJosef Bacik spin_lock_init(&fs_info->eb_leak_lock); 3005bd647ce3SJosef Bacik #endif 3006c8bf1b67SDavid Sterba extent_map_tree_init(&fs_info->mapping_tree); 300766d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 300866d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 300966d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 301066d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 301166d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 301266d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 301366d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 3014ba2c4d4eSJosef Bacik btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, 3015ba2c4d4eSJosef Bacik BTRFS_BLOCK_RSV_DELREFS); 3016ba2c4d4eSJosef Bacik 3017771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 30184cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 3019034f784dSJosef Bacik atomic_set(&fs_info->nr_delayed_iputs, 0); 3020fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 3021abed4aaaSJosef Bacik fs_info->global_root_tree = RB_ROOT; 302295ac567aSFilipe David Borba Manana fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 30239ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 30244cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 3025a5ed45f8SNikolay Borisov atomic64_set(&fs_info->free_chunk_space, 0); 3026f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 30278b87dc17SDavid Sterba fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 3028f8c269d7SDavid Sterba fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 3029fd708b81SJosef Bacik btrfs_init_ref_verify(fs_info); 3030c8b97818SChris Mason 3031b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 3032b34b086cSChris Mason num_online_cpus() + 2, 8); 30330afbaf8cSChris Mason 3034199c2a9cSMiao Xie INIT_LIST_HEAD(&fs_info->ordered_roots); 3035199c2a9cSMiao Xie spin_lock_init(&fs_info->ordered_root_lock); 303669fe2d75SJosef Bacik 3037638aa7edSEric Sandeen btrfs_init_scrub(fs_info); 303821adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 303921adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 304021adbd5cSStefan Behrens #endif 3041779a65a4SEric Sandeen btrfs_init_balance(fs_info); 304257056740SJosef Bacik btrfs_init_async_reclaim_work(fs_info); 3043a2de733cSArne Jansen 304416b0c258SFilipe Manana rwlock_init(&fs_info->block_group_cache_lock); 304508dddb29SFilipe Manana fs_info->block_group_cache_tree = RB_ROOT_CACHED; 30460f9dd46cSJosef Bacik 3047fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &fs_info->excluded_extents, 3048fe119a6eSNikolay Borisov IO_TREE_FS_EXCLUDED_EXTENTS, NULL); 304939279cc3SChris Mason 30505a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 3051e02119d5SChris Mason mutex_init(&fs_info->tree_log_mutex); 3052925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 3053a74a4b97SChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 3054a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 30551bbc621eSChris Mason mutex_init(&fs_info->ro_block_group_mutex); 30569e351cc8SJosef Bacik init_rwsem(&fs_info->commit_root_sem); 3057c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 305876dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 3059803b2f54SStefan Behrens sema_init(&fs_info->uuid_tree_rescan_sem, 1); 3060fa9c0d79SChris Mason 3061ad618368SEric Sandeen btrfs_init_dev_replace_locks(fs_info); 3062f9e92e40SEric Sandeen btrfs_init_qgroup(fs_info); 3063b0643e59SDennis Zhou btrfs_discard_init(fs_info); 3064416ac51dSArne Jansen 3065fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 3066fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 3067fa9c0d79SChris Mason 3068e6dcd2dcSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 3069f9295749SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 3070bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 30714854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 3072034f784dSJosef Bacik init_waitqueue_head(&fs_info->delayed_iputs_wait); 30733768f368SChris Mason 3074da17066cSJeff Mahoney /* Usable values until the real ones are cached from the superblock */ 3075da17066cSJeff Mahoney fs_info->nodesize = 4096; 3076da17066cSJeff Mahoney fs_info->sectorsize = 4096; 3077ab108d99SDavid Sterba fs_info->sectorsize_bits = ilog2(4096); 3078da17066cSJeff Mahoney fs_info->stripesize = 4096; 3079da17066cSJeff Mahoney 3080f7b12a62SNaohiro Aota fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE; 3081f7b12a62SNaohiro Aota 3082eede2bf3SOmar Sandoval spin_lock_init(&fs_info->swapfile_pins_lock); 3083eede2bf3SOmar Sandoval fs_info->swapfile_pins = RB_ROOT; 3084eede2bf3SOmar Sandoval 308518bb8bbfSJohannes Thumshirn fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH; 308618bb8bbfSJohannes Thumshirn INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work); 30878260edbaSJosef Bacik } 30888260edbaSJosef Bacik 30898260edbaSJosef Bacik static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb) 30908260edbaSJosef Bacik { 30918260edbaSJosef Bacik int ret; 30928260edbaSJosef Bacik 30938260edbaSJosef Bacik fs_info->sb = sb; 30948260edbaSJosef Bacik sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; 30958260edbaSJosef Bacik sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); 30969e967495SFilipe Manana 30975deb17e1SJosef Bacik ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL); 3098ae18c37aSJosef Bacik if (ret) 3099ae18c37aSJosef Bacik return ret; 3100ae18c37aSJosef Bacik 3101ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 3102ae18c37aSJosef Bacik if (ret) 3103c75e8394SJosef Bacik return ret; 3104ae18c37aSJosef Bacik 3105ae18c37aSJosef Bacik fs_info->dirty_metadata_batch = PAGE_SIZE * 3106ae18c37aSJosef Bacik (1 + ilog2(nr_cpu_ids)); 3107ae18c37aSJosef Bacik 3108ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 3109ae18c37aSJosef Bacik if (ret) 3110c75e8394SJosef Bacik return ret; 3111ae18c37aSJosef Bacik 3112ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, 3113ae18c37aSJosef Bacik GFP_KERNEL); 3114ae18c37aSJosef Bacik if (ret) 3115c75e8394SJosef Bacik return ret; 3116ae18c37aSJosef Bacik 3117ae18c37aSJosef Bacik fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 3118ae18c37aSJosef Bacik GFP_KERNEL); 3119c75e8394SJosef Bacik if (!fs_info->delayed_root) 3120c75e8394SJosef Bacik return -ENOMEM; 3121ae18c37aSJosef Bacik btrfs_init_delayed_root(fs_info->delayed_root); 3122ae18c37aSJosef Bacik 3123a0a1db70SFilipe Manana if (sb_rdonly(sb)) 3124a0a1db70SFilipe Manana set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state); 3125a0a1db70SFilipe Manana 3126c75e8394SJosef Bacik return btrfs_alloc_stripe_hash_table(fs_info); 3127ae18c37aSJosef Bacik } 3128ae18c37aSJosef Bacik 312997f4dd09SNikolay Borisov static int btrfs_uuid_rescan_kthread(void *data) 313097f4dd09SNikolay Borisov { 31310d031dc4SYu Zhe struct btrfs_fs_info *fs_info = data; 313297f4dd09SNikolay Borisov int ret; 313397f4dd09SNikolay Borisov 313497f4dd09SNikolay Borisov /* 313597f4dd09SNikolay Borisov * 1st step is to iterate through the existing UUID tree and 313697f4dd09SNikolay Borisov * to delete all entries that contain outdated data. 313797f4dd09SNikolay Borisov * 2nd step is to add all missing entries to the UUID tree. 313897f4dd09SNikolay Borisov */ 313997f4dd09SNikolay Borisov ret = btrfs_uuid_tree_iterate(fs_info); 314097f4dd09SNikolay Borisov if (ret < 0) { 3141c94bec2cSJosef Bacik if (ret != -EINTR) 3142c94bec2cSJosef Bacik btrfs_warn(fs_info, "iterating uuid_tree failed %d", 3143c94bec2cSJosef Bacik ret); 314497f4dd09SNikolay Borisov up(&fs_info->uuid_tree_rescan_sem); 314597f4dd09SNikolay Borisov return ret; 314697f4dd09SNikolay Borisov } 314797f4dd09SNikolay Borisov return btrfs_uuid_scan_kthread(data); 314897f4dd09SNikolay Borisov } 314997f4dd09SNikolay Borisov 315097f4dd09SNikolay Borisov static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) 315197f4dd09SNikolay Borisov { 315297f4dd09SNikolay Borisov struct task_struct *task; 315397f4dd09SNikolay Borisov 315497f4dd09SNikolay Borisov down(&fs_info->uuid_tree_rescan_sem); 315597f4dd09SNikolay Borisov task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); 315697f4dd09SNikolay Borisov if (IS_ERR(task)) { 315797f4dd09SNikolay Borisov /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 315897f4dd09SNikolay Borisov btrfs_warn(fs_info, "failed to start uuid_rescan task"); 315997f4dd09SNikolay Borisov up(&fs_info->uuid_tree_rescan_sem); 316097f4dd09SNikolay Borisov return PTR_ERR(task); 316197f4dd09SNikolay Borisov } 316297f4dd09SNikolay Borisov 316397f4dd09SNikolay Borisov return 0; 316497f4dd09SNikolay Borisov } 316597f4dd09SNikolay Borisov 316644c0ca21SBoris Burkov /* 31678cd29088SBoris Burkov * Some options only have meaning at mount time and shouldn't persist across 31688cd29088SBoris Burkov * remounts, or be displayed. Clear these at the end of mount and remount 31698cd29088SBoris Burkov * code paths. 31708cd29088SBoris Burkov */ 31718cd29088SBoris Burkov void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info) 31728cd29088SBoris Burkov { 31738cd29088SBoris Burkov btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT); 31748b228324SBoris Burkov btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE); 31758cd29088SBoris Burkov } 31768cd29088SBoris Burkov 31778cd29088SBoris Burkov /* 317844c0ca21SBoris Burkov * Mounting logic specific to read-write file systems. Shared by open_ctree 317944c0ca21SBoris Burkov * and btrfs_remount when remounting from read-only to read-write. 318044c0ca21SBoris Burkov */ 318144c0ca21SBoris Burkov int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info) 318244c0ca21SBoris Burkov { 318344c0ca21SBoris Burkov int ret; 318494846229SBoris Burkov const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE); 31858b228324SBoris Burkov bool clear_free_space_tree = false; 31868b228324SBoris Burkov 31878b228324SBoris Burkov if (btrfs_test_opt(fs_info, CLEAR_CACHE) && 31888b228324SBoris Burkov btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 31898b228324SBoris Burkov clear_free_space_tree = true; 31908b228324SBoris Burkov } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 31918b228324SBoris Burkov !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { 31928b228324SBoris Burkov btrfs_warn(fs_info, "free space tree is invalid"); 31938b228324SBoris Burkov clear_free_space_tree = true; 31948b228324SBoris Burkov } 31958b228324SBoris Burkov 31968b228324SBoris Burkov if (clear_free_space_tree) { 31978b228324SBoris Burkov btrfs_info(fs_info, "clearing free space tree"); 31988b228324SBoris Burkov ret = btrfs_clear_free_space_tree(fs_info); 31998b228324SBoris Burkov if (ret) { 32008b228324SBoris Burkov btrfs_warn(fs_info, 32018b228324SBoris Burkov "failed to clear free space tree: %d", ret); 32028b228324SBoris Burkov goto out; 32038b228324SBoris Burkov } 32048b228324SBoris Burkov } 320544c0ca21SBoris Burkov 32068d488a8cSFilipe Manana /* 32078d488a8cSFilipe Manana * btrfs_find_orphan_roots() is responsible for finding all the dead 32088d488a8cSFilipe Manana * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load 3209fc7cbcd4SDavid Sterba * them into the fs_info->fs_roots_radix tree. This must be done before 32108d488a8cSFilipe Manana * calling btrfs_orphan_cleanup() on the tree root. If we don't do it 32118d488a8cSFilipe Manana * first, then btrfs_orphan_cleanup() will delete a dead root's orphan 32128d488a8cSFilipe Manana * item before the root's tree is deleted - this means that if we unmount 32138d488a8cSFilipe Manana * or crash before the deletion completes, on the next mount we will not 32148d488a8cSFilipe Manana * delete what remains of the tree because the orphan item does not 32158d488a8cSFilipe Manana * exists anymore, which is what tells us we have a pending deletion. 32168d488a8cSFilipe Manana */ 32178d488a8cSFilipe Manana ret = btrfs_find_orphan_roots(fs_info); 32188d488a8cSFilipe Manana if (ret) 32198d488a8cSFilipe Manana goto out; 32208d488a8cSFilipe Manana 322144c0ca21SBoris Burkov ret = btrfs_cleanup_fs_roots(fs_info); 322244c0ca21SBoris Burkov if (ret) 322344c0ca21SBoris Burkov goto out; 322444c0ca21SBoris Burkov 32258f1c21d7SBoris Burkov down_read(&fs_info->cleanup_work_sem); 32268f1c21d7SBoris Burkov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 32278f1c21d7SBoris Burkov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 32288f1c21d7SBoris Burkov up_read(&fs_info->cleanup_work_sem); 32298f1c21d7SBoris Burkov goto out; 32308f1c21d7SBoris Burkov } 32318f1c21d7SBoris Burkov up_read(&fs_info->cleanup_work_sem); 32328f1c21d7SBoris Burkov 323344c0ca21SBoris Burkov mutex_lock(&fs_info->cleaner_mutex); 32347eefae6bSJosef Bacik ret = btrfs_recover_relocation(fs_info); 323544c0ca21SBoris Burkov mutex_unlock(&fs_info->cleaner_mutex); 323644c0ca21SBoris Burkov if (ret < 0) { 323744c0ca21SBoris Burkov btrfs_warn(fs_info, "failed to recover relocation: %d", ret); 323844c0ca21SBoris Burkov goto out; 323944c0ca21SBoris Burkov } 324044c0ca21SBoris Burkov 32415011139aSBoris Burkov if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && 32425011139aSBoris Burkov !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 32435011139aSBoris Burkov btrfs_info(fs_info, "creating free space tree"); 32445011139aSBoris Burkov ret = btrfs_create_free_space_tree(fs_info); 32455011139aSBoris Burkov if (ret) { 32465011139aSBoris Burkov btrfs_warn(fs_info, 32475011139aSBoris Burkov "failed to create free space tree: %d", ret); 32485011139aSBoris Burkov goto out; 32495011139aSBoris Burkov } 32505011139aSBoris Burkov } 32515011139aSBoris Burkov 325294846229SBoris Burkov if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) { 325394846229SBoris Burkov ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt); 325494846229SBoris Burkov if (ret) 325594846229SBoris Burkov goto out; 325694846229SBoris Burkov } 325794846229SBoris Burkov 325844c0ca21SBoris Burkov ret = btrfs_resume_balance_async(fs_info); 325944c0ca21SBoris Burkov if (ret) 326044c0ca21SBoris Burkov goto out; 326144c0ca21SBoris Burkov 326244c0ca21SBoris Burkov ret = btrfs_resume_dev_replace_async(fs_info); 326344c0ca21SBoris Burkov if (ret) { 326444c0ca21SBoris Burkov btrfs_warn(fs_info, "failed to resume dev_replace"); 326544c0ca21SBoris Burkov goto out; 326644c0ca21SBoris Burkov } 326744c0ca21SBoris Burkov 326844c0ca21SBoris Burkov btrfs_qgroup_rescan_resume(fs_info); 326944c0ca21SBoris Burkov 327044c0ca21SBoris Burkov if (!fs_info->uuid_root) { 327144c0ca21SBoris Burkov btrfs_info(fs_info, "creating UUID tree"); 327244c0ca21SBoris Burkov ret = btrfs_create_uuid_tree(fs_info); 327344c0ca21SBoris Burkov if (ret) { 327444c0ca21SBoris Burkov btrfs_warn(fs_info, 327544c0ca21SBoris Burkov "failed to create the UUID tree %d", ret); 327644c0ca21SBoris Burkov goto out; 327744c0ca21SBoris Burkov } 327844c0ca21SBoris Burkov } 327944c0ca21SBoris Burkov 328044c0ca21SBoris Burkov out: 328144c0ca21SBoris Burkov return ret; 328244c0ca21SBoris Burkov } 328344c0ca21SBoris Burkov 3284ae18c37aSJosef Bacik int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices, 3285ae18c37aSJosef Bacik char *options) 3286ae18c37aSJosef Bacik { 3287ae18c37aSJosef Bacik u32 sectorsize; 3288ae18c37aSJosef Bacik u32 nodesize; 3289ae18c37aSJosef Bacik u32 stripesize; 3290ae18c37aSJosef Bacik u64 generation; 3291ae18c37aSJosef Bacik u64 features; 3292ae18c37aSJosef Bacik u16 csum_type; 3293ae18c37aSJosef Bacik struct btrfs_super_block *disk_super; 3294ae18c37aSJosef Bacik struct btrfs_fs_info *fs_info = btrfs_sb(sb); 3295ae18c37aSJosef Bacik struct btrfs_root *tree_root; 3296ae18c37aSJosef Bacik struct btrfs_root *chunk_root; 3297ae18c37aSJosef Bacik int ret; 3298ae18c37aSJosef Bacik int err = -EINVAL; 3299ae18c37aSJosef Bacik int level; 3300ae18c37aSJosef Bacik 33018260edbaSJosef Bacik ret = init_mount_fs_info(fs_info, sb); 330253b381b3SDavid Woodhouse if (ret) { 330383c8266aSDavid Sterba err = ret; 3304ae18c37aSJosef Bacik goto fail; 330553b381b3SDavid Woodhouse } 330653b381b3SDavid Woodhouse 3307ae18c37aSJosef Bacik /* These need to be init'ed before we start creating inodes and such. */ 3308ae18c37aSJosef Bacik tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, 3309ae18c37aSJosef Bacik GFP_KERNEL); 3310ae18c37aSJosef Bacik fs_info->tree_root = tree_root; 3311ae18c37aSJosef Bacik chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID, 3312ae18c37aSJosef Bacik GFP_KERNEL); 3313ae18c37aSJosef Bacik fs_info->chunk_root = chunk_root; 3314ae18c37aSJosef Bacik if (!tree_root || !chunk_root) { 3315ae18c37aSJosef Bacik err = -ENOMEM; 3316c75e8394SJosef Bacik goto fail; 3317ae18c37aSJosef Bacik } 3318ae18c37aSJosef Bacik 3319ae18c37aSJosef Bacik fs_info->btree_inode = new_inode(sb); 3320ae18c37aSJosef Bacik if (!fs_info->btree_inode) { 3321ae18c37aSJosef Bacik err = -ENOMEM; 3322c75e8394SJosef Bacik goto fail; 3323ae18c37aSJosef Bacik } 3324ae18c37aSJosef Bacik mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 3325ae18c37aSJosef Bacik btrfs_init_btree_inode(fs_info); 3326ae18c37aSJosef Bacik 3327d24fa5c1SAnand Jain invalidate_bdev(fs_devices->latest_dev->bdev); 33281104a885SDavid Sterba 33291104a885SDavid Sterba /* 33301104a885SDavid Sterba * Read super block and check the signature bytes only 33311104a885SDavid Sterba */ 3332d24fa5c1SAnand Jain disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev); 33338f32380dSJohannes Thumshirn if (IS_ERR(disk_super)) { 33348f32380dSJohannes Thumshirn err = PTR_ERR(disk_super); 333516cdcec7SMiao Xie goto fail_alloc; 333620b45077SDave Young } 333739279cc3SChris Mason 33381104a885SDavid Sterba /* 3339260db43cSRandy Dunlap * Verify the type first, if that or the checksum value are 33408dc3f22cSJohannes Thumshirn * corrupted, we'll find out 33418dc3f22cSJohannes Thumshirn */ 33428f32380dSJohannes Thumshirn csum_type = btrfs_super_csum_type(disk_super); 334351bce6c9SJohannes Thumshirn if (!btrfs_supported_super_csum(csum_type)) { 33448dc3f22cSJohannes Thumshirn btrfs_err(fs_info, "unsupported checksum algorithm: %u", 334551bce6c9SJohannes Thumshirn csum_type); 33468dc3f22cSJohannes Thumshirn err = -EINVAL; 33478f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 33488dc3f22cSJohannes Thumshirn goto fail_alloc; 33498dc3f22cSJohannes Thumshirn } 33508dc3f22cSJohannes Thumshirn 335183c68bbcSSu Yue fs_info->csum_size = btrfs_super_csum_size(disk_super); 335283c68bbcSSu Yue 33536d97c6e3SJohannes Thumshirn ret = btrfs_init_csum_hash(fs_info, csum_type); 33546d97c6e3SJohannes Thumshirn if (ret) { 33556d97c6e3SJohannes Thumshirn err = ret; 33568f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 33576d97c6e3SJohannes Thumshirn goto fail_alloc; 33586d97c6e3SJohannes Thumshirn } 33596d97c6e3SJohannes Thumshirn 33608dc3f22cSJohannes Thumshirn /* 33611104a885SDavid Sterba * We want to check superblock checksum, the type is stored inside. 33621104a885SDavid Sterba * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 33631104a885SDavid Sterba */ 33648f32380dSJohannes Thumshirn if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) { 336505135f59SDavid Sterba btrfs_err(fs_info, "superblock checksum mismatch"); 33661104a885SDavid Sterba err = -EINVAL; 33678f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 3368141386e1SJosef Bacik goto fail_alloc; 33691104a885SDavid Sterba } 33701104a885SDavid Sterba 33711104a885SDavid Sterba /* 33721104a885SDavid Sterba * super_copy is zeroed at allocation time and we never touch the 33731104a885SDavid Sterba * following bytes up to INFO_SIZE, the checksum is calculated from 33741104a885SDavid Sterba * the whole block of INFO_SIZE 33751104a885SDavid Sterba */ 33768f32380dSJohannes Thumshirn memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy)); 33778f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 33785f39d397SChris Mason 3379fbc6feaeSNikolay Borisov disk_super = fs_info->super_copy; 3380fbc6feaeSNikolay Borisov 33810b86a832SChris Mason 3382fbc6feaeSNikolay Borisov features = btrfs_super_flags(disk_super); 3383fbc6feaeSNikolay Borisov if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { 3384fbc6feaeSNikolay Borisov features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; 3385fbc6feaeSNikolay Borisov btrfs_set_super_flags(disk_super, features); 3386fbc6feaeSNikolay Borisov btrfs_info(fs_info, 3387fbc6feaeSNikolay Borisov "found metadata UUID change in progress flag, clearing"); 3388fbc6feaeSNikolay Borisov } 3389fbc6feaeSNikolay Borisov 3390fbc6feaeSNikolay Borisov memcpy(fs_info->super_for_commit, fs_info->super_copy, 3391fbc6feaeSNikolay Borisov sizeof(*fs_info->super_for_commit)); 3392de37aa51SNikolay Borisov 3393069ec957SQu Wenruo ret = btrfs_validate_mount_super(fs_info); 33941104a885SDavid Sterba if (ret) { 339505135f59SDavid Sterba btrfs_err(fs_info, "superblock contains fatal errors"); 33961104a885SDavid Sterba err = -EINVAL; 3397141386e1SJosef Bacik goto fail_alloc; 33981104a885SDavid Sterba } 33991104a885SDavid Sterba 34000f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 3401141386e1SJosef Bacik goto fail_alloc; 34020f7d52f4SChris Mason 3403acce952bSliubo /* check FS state, whether FS is broken. */ 340487533c47SMiao Xie if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 340587533c47SMiao Xie set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 3406acce952bSliubo 340775e7cb7fSLiu Bo /* 340875e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 340975e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 341075e7cb7fSLiu Bo */ 341175e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 341275e7cb7fSLiu Bo 34136f93e834SAnand Jain 34146f93e834SAnand Jain /* Set up fs_info before parsing mount options */ 34156f93e834SAnand Jain nodesize = btrfs_super_nodesize(disk_super); 34166f93e834SAnand Jain sectorsize = btrfs_super_sectorsize(disk_super); 34176f93e834SAnand Jain stripesize = sectorsize; 34186f93e834SAnand Jain fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 34196f93e834SAnand Jain fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 34206f93e834SAnand Jain 34216f93e834SAnand Jain fs_info->nodesize = nodesize; 34226f93e834SAnand Jain fs_info->sectorsize = sectorsize; 34236f93e834SAnand Jain fs_info->sectorsize_bits = ilog2(sectorsize); 34246f93e834SAnand Jain fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size; 34256f93e834SAnand Jain fs_info->stripesize = stripesize; 34266f93e834SAnand Jain 34272ff7e61eSJeff Mahoney ret = btrfs_parse_options(fs_info, options, sb->s_flags); 34282b82032cSYan Zheng if (ret) { 34292b82032cSYan Zheng err = ret; 3430141386e1SJosef Bacik goto fail_alloc; 34312b82032cSYan Zheng } 3432dfe25020SChris Mason 3433f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 3434f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 3435f2b636e8SJosef Bacik if (features) { 343605135f59SDavid Sterba btrfs_err(fs_info, 3437d5321a0fSQu Wenruo "cannot mount because of unsupported optional features (0x%llx)", 3438c1c9ff7cSGeert Uytterhoeven features); 3439f2b636e8SJosef Bacik err = -EINVAL; 3440141386e1SJosef Bacik goto fail_alloc; 3441f2b636e8SJosef Bacik } 3442f2b636e8SJosef Bacik 34435d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 34445d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 34450b246afaSJeff Mahoney if (fs_info->compress_type == BTRFS_COMPRESS_LZO) 3446a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 34475c1aab1dSNick Terrell else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) 34485c1aab1dSNick Terrell features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; 3449727011e0SChris Mason 3450727011e0SChris Mason /* 3451e26b04c4SNikolay Borisov * Flag our filesystem as having big metadata blocks if they are bigger 3452e26b04c4SNikolay Borisov * than the page size. 3453e26b04c4SNikolay Borisov */ 34546b769dacSNikolay Borisov if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) 3455e26b04c4SNikolay Borisov features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 3456727011e0SChris Mason 3457727011e0SChris Mason /* 3458bc3f116fSChris Mason * mixed block groups end up with duplicate but slightly offset 3459bc3f116fSChris Mason * extent buffers for the same range. It leads to corruptions 3460bc3f116fSChris Mason */ 3461bc3f116fSChris Mason if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 3462707e8a07SDavid Sterba (sectorsize != nodesize)) { 346305135f59SDavid Sterba btrfs_err(fs_info, 346405135f59SDavid Sterba "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", 346505135f59SDavid Sterba nodesize, sectorsize); 3466141386e1SJosef Bacik goto fail_alloc; 3467bc3f116fSChris Mason } 3468bc3f116fSChris Mason 3469ceda0864SMiao Xie /* 3470ceda0864SMiao Xie * Needn't use the lock because there is no other task which will 3471ceda0864SMiao Xie * update the flag. 3472ceda0864SMiao Xie */ 34735d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 34745d4f98a2SYan Zheng 3475f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 3476f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 3477bc98a42cSDavid Howells if (!sb_rdonly(sb) && features) { 347805135f59SDavid Sterba btrfs_err(fs_info, 3479d5321a0fSQu Wenruo "cannot mount read-write because of unsupported optional features (0x%llx)", 3480c1c9ff7cSGeert Uytterhoeven features); 3481f2b636e8SJosef Bacik err = -EINVAL; 3482141386e1SJosef Bacik goto fail_alloc; 3483f2b636e8SJosef Bacik } 3484dc4d3168SQu Wenruo /* 3485dc4d3168SQu Wenruo * We have unsupported RO compat features, although RO mounted, we 3486dc4d3168SQu Wenruo * should not cause any metadata write, including log replay. 3487dc4d3168SQu Wenruo * Or we could screw up whatever the new feature requires. 3488dc4d3168SQu Wenruo */ 3489dc4d3168SQu Wenruo if (unlikely(features && btrfs_super_log_root(disk_super) && 3490dc4d3168SQu Wenruo !btrfs_test_opt(fs_info, NOLOGREPLAY))) { 3491dc4d3168SQu Wenruo btrfs_err(fs_info, 3492dc4d3168SQu Wenruo "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay", 3493dc4d3168SQu Wenruo features); 3494dc4d3168SQu Wenruo err = -EINVAL; 3495dc4d3168SQu Wenruo goto fail_alloc; 3496dc4d3168SQu Wenruo } 3497dc4d3168SQu Wenruo 349861d92c32SChris Mason 34998481dd80SQu Wenruo if (sectorsize < PAGE_SIZE) { 35008481dd80SQu Wenruo struct btrfs_subpage_info *subpage_info; 35018481dd80SQu Wenruo 35029f73f1aeSQu Wenruo /* 35039f73f1aeSQu Wenruo * V1 space cache has some hardcoded PAGE_SIZE usage, and is 35049f73f1aeSQu Wenruo * going to be deprecated. 35059f73f1aeSQu Wenruo * 35069f73f1aeSQu Wenruo * Force to use v2 cache for subpage case. 35079f73f1aeSQu Wenruo */ 35089f73f1aeSQu Wenruo btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE); 35099f73f1aeSQu Wenruo btrfs_set_and_info(fs_info, FREE_SPACE_TREE, 35109f73f1aeSQu Wenruo "forcing free space tree for sector size %u with page size %lu", 35119f73f1aeSQu Wenruo sectorsize, PAGE_SIZE); 35129f73f1aeSQu Wenruo 351395ea0486SQu Wenruo btrfs_warn(fs_info, 351495ea0486SQu Wenruo "read-write for sector size %u with page size %lu is experimental", 35150bb3eb3eSQu Wenruo sectorsize, PAGE_SIZE); 35168481dd80SQu Wenruo subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL); 35178481dd80SQu Wenruo if (!subpage_info) 35188481dd80SQu Wenruo goto fail_alloc; 35198481dd80SQu Wenruo btrfs_init_subpage_info(subpage_info, sectorsize); 35208481dd80SQu Wenruo fs_info->subpage_info = subpage_info; 3521c8050b3bSQu Wenruo } 35220bb3eb3eSQu Wenruo 3523d21deec5SSu Yue ret = btrfs_init_workqueues(fs_info); 35242a458198SEric Sandeen if (ret) { 35252a458198SEric Sandeen err = ret; 35260dc3b84aSJosef Bacik goto fail_sb_buffer; 35270dc3b84aSJosef Bacik } 35284543df7eSChris Mason 35299e11ceeeSJan Kara sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); 35309e11ceeeSJan Kara sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); 35314575c9ccSChris Mason 3532a061fc8dSChris Mason sb->s_blocksize = sectorsize; 3533a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 3534de37aa51SNikolay Borisov memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); 3535db94535dSChris Mason 3536925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 35376bccf3abSJeff Mahoney ret = btrfs_read_sys_array(fs_info); 3538925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 353984eed90fSChris Mason if (ret) { 354005135f59SDavid Sterba btrfs_err(fs_info, "failed to read the system array: %d", ret); 35415d4f98a2SYan Zheng goto fail_sb_buffer; 354284eed90fSChris Mason } 35430b86a832SChris Mason 354484234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 3545581c1760SQu Wenruo level = btrfs_super_chunk_root_level(disk_super); 3546bd676446SJosef Bacik ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super), 3547bd676446SJosef Bacik generation, level); 3548bd676446SJosef Bacik if (ret) { 354905135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk root"); 3550af31f5e5SChris Mason goto fail_tree_roots; 355183121942SDavid Woodhouse } 35520b86a832SChris Mason 3553e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 3554c4ac7541SDavid Sterba offsetof(struct btrfs_header, chunk_tree_uuid), 3555c4ac7541SDavid Sterba BTRFS_UUID_SIZE); 3556e17cade2SChris Mason 35575b4aacefSJeff Mahoney ret = btrfs_read_chunk_tree(fs_info); 35582b82032cSYan Zheng if (ret) { 355905135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk tree: %d", ret); 3560af31f5e5SChris Mason goto fail_tree_roots; 35612b82032cSYan Zheng } 35620b86a832SChris Mason 35638dabb742SStefan Behrens /* 3564bacce86aSAnand Jain * At this point we know all the devices that make this filesystem, 3565bacce86aSAnand Jain * including the seed devices but we don't know yet if the replace 3566bacce86aSAnand Jain * target is required. So free devices that are not part of this 35671a9fd417SDavid Sterba * filesystem but skip the replace target device which is checked 3568bacce86aSAnand Jain * below in btrfs_init_dev_replace(). 35698dabb742SStefan Behrens */ 3570bacce86aSAnand Jain btrfs_free_extra_devids(fs_devices); 3571d24fa5c1SAnand Jain if (!fs_devices->latest_dev->bdev) { 357205135f59SDavid Sterba btrfs_err(fs_info, "failed to read devices"); 3573a6b0d5c8SChris Mason goto fail_tree_roots; 3574a6b0d5c8SChris Mason } 3575a6b0d5c8SChris Mason 3576b8522a1eSNikolay Borisov ret = init_tree_roots(fs_info); 35774bbcaa64SEric Sandeen if (ret) 3578b8522a1eSNikolay Borisov goto fail_tree_roots; 35798929ecfaSYan, Zheng 358075ec1db8SJosef Bacik /* 358173651042SNaohiro Aota * Get zone type information of zoned block devices. This will also 358273651042SNaohiro Aota * handle emulation of a zoned filesystem if a regular device has the 358373651042SNaohiro Aota * zoned incompat feature flag set. 358473651042SNaohiro Aota */ 358573651042SNaohiro Aota ret = btrfs_get_dev_zone_info_all_devices(fs_info); 358673651042SNaohiro Aota if (ret) { 358773651042SNaohiro Aota btrfs_err(fs_info, 358873651042SNaohiro Aota "zoned: failed to read device zone info: %d", 358973651042SNaohiro Aota ret); 359073651042SNaohiro Aota goto fail_block_groups; 359173651042SNaohiro Aota } 359273651042SNaohiro Aota 359373651042SNaohiro Aota /* 359475ec1db8SJosef Bacik * If we have a uuid root and we're not being told to rescan we need to 359575ec1db8SJosef Bacik * check the generation here so we can set the 359675ec1db8SJosef Bacik * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the 359775ec1db8SJosef Bacik * transaction during a balance or the log replay without updating the 359875ec1db8SJosef Bacik * uuid generation, and then if we crash we would rescan the uuid tree, 359975ec1db8SJosef Bacik * even though it was perfectly fine. 360075ec1db8SJosef Bacik */ 360175ec1db8SJosef Bacik if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) && 360275ec1db8SJosef Bacik fs_info->generation == btrfs_super_uuid_tree_generation(disk_super)) 360375ec1db8SJosef Bacik set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 360475ec1db8SJosef Bacik 3605cf90d884SQu Wenruo ret = btrfs_verify_dev_extents(fs_info); 3606cf90d884SQu Wenruo if (ret) { 3607cf90d884SQu Wenruo btrfs_err(fs_info, 3608cf90d884SQu Wenruo "failed to verify dev extents against chunks: %d", 3609cf90d884SQu Wenruo ret); 3610cf90d884SQu Wenruo goto fail_block_groups; 3611cf90d884SQu Wenruo } 361268310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 361368310a5eSIlya Dryomov if (ret) { 361405135f59SDavid Sterba btrfs_err(fs_info, "failed to recover balance: %d", ret); 361568310a5eSIlya Dryomov goto fail_block_groups; 361668310a5eSIlya Dryomov } 361768310a5eSIlya Dryomov 3618733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 3619733f4fbbSStefan Behrens if (ret) { 362005135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_stats: %d", ret); 3621733f4fbbSStefan Behrens goto fail_block_groups; 3622733f4fbbSStefan Behrens } 3623733f4fbbSStefan Behrens 36248dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 36258dabb742SStefan Behrens if (ret) { 362605135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_replace: %d", ret); 36278dabb742SStefan Behrens goto fail_block_groups; 36288dabb742SStefan Behrens } 36298dabb742SStefan Behrens 3630b70f5097SNaohiro Aota ret = btrfs_check_zoned_mode(fs_info); 3631b70f5097SNaohiro Aota if (ret) { 3632b70f5097SNaohiro Aota btrfs_err(fs_info, "failed to initialize zoned mode: %d", 3633b70f5097SNaohiro Aota ret); 3634b70f5097SNaohiro Aota goto fail_block_groups; 3635b70f5097SNaohiro Aota } 3636b70f5097SNaohiro Aota 3637c6761a9eSAnand Jain ret = btrfs_sysfs_add_fsid(fs_devices); 3638b7c35e81SAnand Jain if (ret) { 363905135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", 364005135f59SDavid Sterba ret); 3641b7c35e81SAnand Jain goto fail_block_groups; 3642b7c35e81SAnand Jain } 3643b7c35e81SAnand Jain 364496f3136eSAnand Jain ret = btrfs_sysfs_add_mounted(fs_info); 36455ac1d209SJeff Mahoney if (ret) { 364605135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); 3647b7c35e81SAnand Jain goto fail_fsdev_sysfs; 36485ac1d209SJeff Mahoney } 36495ac1d209SJeff Mahoney 3650c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 3651c59021f8Sliubo if (ret) { 365205135f59SDavid Sterba btrfs_err(fs_info, "failed to initialize space info: %d", ret); 36532365dd3cSAnand Jain goto fail_sysfs; 3654c59021f8Sliubo } 3655c59021f8Sliubo 36565b4aacefSJeff Mahoney ret = btrfs_read_block_groups(fs_info); 36571b1d1f66SJosef Bacik if (ret) { 365805135f59SDavid Sterba btrfs_err(fs_info, "failed to read block groups: %d", ret); 36592365dd3cSAnand Jain goto fail_sysfs; 36601b1d1f66SJosef Bacik } 36614330e183SQu Wenruo 366216beac87SNaohiro Aota btrfs_free_zone_cache(fs_info); 366316beac87SNaohiro Aota 36645c78a5e7SAnand Jain if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices && 36655c78a5e7SAnand Jain !btrfs_check_rw_degradable(fs_info, NULL)) { 366605135f59SDavid Sterba btrfs_warn(fs_info, 366752042d8eSAndrea Gelmini "writable mount is not allowed due to too many missing devices"); 36682365dd3cSAnand Jain goto fail_sysfs; 3669292fd7fcSStefan Behrens } 36709078a3e1SChris Mason 367133c44184SJosef Bacik fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info, 3672a74a4b97SChris Mason "btrfs-cleaner"); 367357506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 36742365dd3cSAnand Jain goto fail_sysfs; 3675a74a4b97SChris Mason 3676a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 3677a74a4b97SChris Mason tree_root, 3678a74a4b97SChris Mason "btrfs-transaction"); 367957506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 36803f157a2fSChris Mason goto fail_cleaner; 3681a74a4b97SChris Mason 3682583b7231SHans van Kranenburg if (!btrfs_test_opt(fs_info, NOSSD) && 3683c289811cSChris Mason !fs_info->fs_devices->rotating) { 3684583b7231SHans van Kranenburg btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); 3685c289811cSChris Mason } 3686c289811cSChris Mason 3687572d9ab7SDavid Sterba /* 368801327610SNicholas D Steeves * Mount does not set all options immediately, we can do it now and do 3689572d9ab7SDavid Sterba * not have to wait for transaction commit 3690572d9ab7SDavid Sterba */ 3691572d9ab7SDavid Sterba btrfs_apply_pending_changes(fs_info); 36923818aea2SQu Wenruo 369321adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 36940b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { 36952ff7e61eSJeff Mahoney ret = btrfsic_mount(fs_info, fs_devices, 36960b246afaSJeff Mahoney btrfs_test_opt(fs_info, 3697cbeaae4fSDavid Sterba CHECK_INTEGRITY_DATA) ? 1 : 0, 369821adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 369921adbd5cSStefan Behrens if (ret) 370005135f59SDavid Sterba btrfs_warn(fs_info, 370105135f59SDavid Sterba "failed to initialize integrity check module: %d", 370205135f59SDavid Sterba ret); 370321adbd5cSStefan Behrens } 370421adbd5cSStefan Behrens #endif 3705bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 3706bcef60f2SArne Jansen if (ret) 3707bcef60f2SArne Jansen goto fail_trans_kthread; 370821adbd5cSStefan Behrens 3709fd708b81SJosef Bacik if (btrfs_build_ref_tree(fs_info)) 3710fd708b81SJosef Bacik btrfs_err(fs_info, "couldn't build ref tree"); 3711fd708b81SJosef Bacik 371296da0919SQu Wenruo /* do not make disk changes in broken FS or nologreplay is given */ 371396da0919SQu Wenruo if (btrfs_super_log_root(disk_super) != 0 && 37140b246afaSJeff Mahoney !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 3715e8294f2fSDavid Sterba btrfs_info(fs_info, "start tree-log replay"); 371663443bf5SEric Sandeen ret = btrfs_replay_log(fs_info, fs_devices); 371779787eaaSJeff Mahoney if (ret) { 371863443bf5SEric Sandeen err = ret; 371928c16cbbSWang Shilong goto fail_qgroup; 3720e556ce2cSYan Zheng } 3721e02119d5SChris Mason } 37221a40e23bSZheng Yan 372356e9357aSDavid Sterba fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true); 37243140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 37253140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 3726f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read fs tree: %d", err); 3727315bf8efSJosef Bacik fs_info->fs_root = NULL; 3728bcef60f2SArne Jansen goto fail_qgroup; 37293140c9a3SDan Carpenter } 3730c289811cSChris Mason 3731bc98a42cSDavid Howells if (sb_rdonly(sb)) 37328cd29088SBoris Burkov goto clear_oneshot; 37332b6ba629SIlya Dryomov 373444c0ca21SBoris Burkov ret = btrfs_start_pre_rw_mount(fs_info); 37352b6ba629SIlya Dryomov if (ret) { 37366bccf3abSJeff Mahoney close_ctree(fs_info); 37372b6ba629SIlya Dryomov return ret; 3738e3acc2a6SJosef Bacik } 3739b0643e59SDennis Zhou btrfs_discard_resume(fs_info); 3740b382a324SJan Schmidt 374144c0ca21SBoris Burkov if (fs_info->uuid_root && 374244c0ca21SBoris Burkov (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || 374344c0ca21SBoris Burkov fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) { 374405135f59SDavid Sterba btrfs_info(fs_info, "checking UUID tree"); 374570f80175SStefan Behrens ret = btrfs_check_uuid_tree(fs_info); 374670f80175SStefan Behrens if (ret) { 374705135f59SDavid Sterba btrfs_warn(fs_info, 374805135f59SDavid Sterba "failed to check the UUID tree: %d", ret); 37496bccf3abSJeff Mahoney close_ctree(fs_info); 375070f80175SStefan Behrens return ret; 375170f80175SStefan Behrens } 3752f7a81ea4SStefan Behrens } 375394846229SBoris Burkov 3754afcdd129SJosef Bacik set_bit(BTRFS_FS_OPEN, &fs_info->flags); 375547ab2a6cSJosef Bacik 3756b4be6aefSJosef Bacik /* Kick the cleaner thread so it'll start deleting snapshots. */ 3757b4be6aefSJosef Bacik if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags)) 3758b4be6aefSJosef Bacik wake_up_process(fs_info->cleaner_kthread); 3759b4be6aefSJosef Bacik 37608cd29088SBoris Burkov clear_oneshot: 37618cd29088SBoris Burkov btrfs_clear_oneshot_options(fs_info); 3762ad2b2c80SAl Viro return 0; 376339279cc3SChris Mason 3764bcef60f2SArne Jansen fail_qgroup: 3765bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 37667c2ca468SChris Mason fail_trans_kthread: 37677c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 37682ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 3769faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 37703f157a2fSChris Mason fail_cleaner: 3771a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 37727c2ca468SChris Mason 37737c2ca468SChris Mason /* 37747c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 37757c2ca468SChris Mason * kthreads 37767c2ca468SChris Mason */ 37777c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 37787c2ca468SChris Mason 37792365dd3cSAnand Jain fail_sysfs: 37806618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 37812365dd3cSAnand Jain 3782b7c35e81SAnand Jain fail_fsdev_sysfs: 3783b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3784b7c35e81SAnand Jain 37851b1d1f66SJosef Bacik fail_block_groups: 378654067ae9SJosef Bacik btrfs_put_block_group_cache(fs_info); 3787af31f5e5SChris Mason 3788af31f5e5SChris Mason fail_tree_roots: 37899e3aa805SJosef Bacik if (fs_info->data_reloc_root) 37909e3aa805SJosef Bacik btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root); 37914273eaffSAnand Jain free_root_pointers(fs_info, true); 37922b8195bbSMiao Xie invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3793af31f5e5SChris Mason 379439279cc3SChris Mason fail_sb_buffer: 37957abadb64SLiu Bo btrfs_stop_all_workers(fs_info); 37965cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 379716cdcec7SMiao Xie fail_alloc: 3798586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 3799586e46e2SIlya Dryomov 38004543df7eSChris Mason iput(fs_info->btree_inode); 38017e662854SQinghuang Feng fail: 3802586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 3803ad2b2c80SAl Viro return err; 3804eb60ceacSChris Mason } 3805663faf9fSMasami Hiramatsu ALLOW_ERROR_INJECTION(open_ctree, ERRNO); 3806eb60ceacSChris Mason 3807314b6dd0SJohannes Thumshirn static void btrfs_end_super_write(struct bio *bio) 3808f2984462SChris Mason { 3809314b6dd0SJohannes Thumshirn struct btrfs_device *device = bio->bi_private; 3810314b6dd0SJohannes Thumshirn struct bio_vec *bvec; 3811314b6dd0SJohannes Thumshirn struct bvec_iter_all iter_all; 3812314b6dd0SJohannes Thumshirn struct page *page; 3813442a4f63SStefan Behrens 3814314b6dd0SJohannes Thumshirn bio_for_each_segment_all(bvec, bio, iter_all) { 3815314b6dd0SJohannes Thumshirn page = bvec->bv_page; 3816314b6dd0SJohannes Thumshirn 3817314b6dd0SJohannes Thumshirn if (bio->bi_status) { 3818fb456252SJeff Mahoney btrfs_warn_rl_in_rcu(device->fs_info, 3819314b6dd0SJohannes Thumshirn "lost page write due to IO error on %s (%d)", 3820314b6dd0SJohannes Thumshirn rcu_str_deref(device->name), 3821314b6dd0SJohannes Thumshirn blk_status_to_errno(bio->bi_status)); 3822314b6dd0SJohannes Thumshirn ClearPageUptodate(page); 3823314b6dd0SJohannes Thumshirn SetPageError(page); 3824314b6dd0SJohannes Thumshirn btrfs_dev_stat_inc_and_print(device, 3825314b6dd0SJohannes Thumshirn BTRFS_DEV_STAT_WRITE_ERRS); 3826314b6dd0SJohannes Thumshirn } else { 3827314b6dd0SJohannes Thumshirn SetPageUptodate(page); 3828f2984462SChris Mason } 3829314b6dd0SJohannes Thumshirn 3830314b6dd0SJohannes Thumshirn put_page(page); 3831314b6dd0SJohannes Thumshirn unlock_page(page); 3832314b6dd0SJohannes Thumshirn } 3833314b6dd0SJohannes Thumshirn 3834314b6dd0SJohannes Thumshirn bio_put(bio); 3835f2984462SChris Mason } 3836f2984462SChris Mason 38378f32380dSJohannes Thumshirn struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev, 38388f32380dSJohannes Thumshirn int copy_num) 383929c36d72SAnand Jain { 384029c36d72SAnand Jain struct btrfs_super_block *super; 38418f32380dSJohannes Thumshirn struct page *page; 384212659251SNaohiro Aota u64 bytenr, bytenr_orig; 38438f32380dSJohannes Thumshirn struct address_space *mapping = bdev->bd_inode->i_mapping; 384412659251SNaohiro Aota int ret; 384529c36d72SAnand Jain 384612659251SNaohiro Aota bytenr_orig = btrfs_sb_offset(copy_num); 384712659251SNaohiro Aota ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr); 384812659251SNaohiro Aota if (ret == -ENOENT) 384912659251SNaohiro Aota return ERR_PTR(-EINVAL); 385012659251SNaohiro Aota else if (ret) 385112659251SNaohiro Aota return ERR_PTR(ret); 385212659251SNaohiro Aota 3853cda00ebaSChristoph Hellwig if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev)) 38548f32380dSJohannes Thumshirn return ERR_PTR(-EINVAL); 385529c36d72SAnand Jain 38568f32380dSJohannes Thumshirn page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS); 38578f32380dSJohannes Thumshirn if (IS_ERR(page)) 38588f32380dSJohannes Thumshirn return ERR_CAST(page); 385929c36d72SAnand Jain 38608f32380dSJohannes Thumshirn super = page_address(page); 386196c2e067SAnand Jain if (btrfs_super_magic(super) != BTRFS_MAGIC) { 386296c2e067SAnand Jain btrfs_release_disk_super(super); 386396c2e067SAnand Jain return ERR_PTR(-ENODATA); 386496c2e067SAnand Jain } 386596c2e067SAnand Jain 386612659251SNaohiro Aota if (btrfs_super_bytenr(super) != bytenr_orig) { 38678f32380dSJohannes Thumshirn btrfs_release_disk_super(super); 38688f32380dSJohannes Thumshirn return ERR_PTR(-EINVAL); 386929c36d72SAnand Jain } 387029c36d72SAnand Jain 38718f32380dSJohannes Thumshirn return super; 387229c36d72SAnand Jain } 387329c36d72SAnand Jain 387429c36d72SAnand Jain 38758f32380dSJohannes Thumshirn struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev) 3876a512bbf8SYan Zheng { 38778f32380dSJohannes Thumshirn struct btrfs_super_block *super, *latest = NULL; 3878a512bbf8SYan Zheng int i; 3879a512bbf8SYan Zheng u64 transid = 0; 3880a512bbf8SYan Zheng 3881a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 3882a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 3883a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 3884a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3885a512bbf8SYan Zheng */ 3886a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 38878f32380dSJohannes Thumshirn super = btrfs_read_dev_one_super(bdev, i); 38888f32380dSJohannes Thumshirn if (IS_ERR(super)) 3889a512bbf8SYan Zheng continue; 3890a512bbf8SYan Zheng 3891a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 38928f32380dSJohannes Thumshirn if (latest) 38938f32380dSJohannes Thumshirn btrfs_release_disk_super(super); 38948f32380dSJohannes Thumshirn 38958f32380dSJohannes Thumshirn latest = super; 3896a512bbf8SYan Zheng transid = btrfs_super_generation(super); 3897a512bbf8SYan Zheng } 3898a512bbf8SYan Zheng } 389992fc03fbSAnand Jain 39008f32380dSJohannes Thumshirn return super; 3901a512bbf8SYan Zheng } 3902a512bbf8SYan Zheng 39034eedeb75SHisashi Hifumi /* 3904abbb3b8eSDavid Sterba * Write superblock @sb to the @device. Do not wait for completion, all the 3905314b6dd0SJohannes Thumshirn * pages we use for writing are locked. 39064eedeb75SHisashi Hifumi * 3907abbb3b8eSDavid Sterba * Write @max_mirrors copies of the superblock, where 0 means default that fit 3908abbb3b8eSDavid Sterba * the expected device size at commit time. Note that max_mirrors must be 3909abbb3b8eSDavid Sterba * same for write and wait phases. 39104eedeb75SHisashi Hifumi * 3911314b6dd0SJohannes Thumshirn * Return number of errors when page is not found or submission fails. 39124eedeb75SHisashi Hifumi */ 3913a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 3914abbb3b8eSDavid Sterba struct btrfs_super_block *sb, int max_mirrors) 3915a512bbf8SYan Zheng { 3916d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = device->fs_info; 3917314b6dd0SJohannes Thumshirn struct address_space *mapping = device->bdev->bd_inode->i_mapping; 3918d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 3919a512bbf8SYan Zheng int i; 3920a512bbf8SYan Zheng int errors = 0; 392112659251SNaohiro Aota int ret; 392212659251SNaohiro Aota u64 bytenr, bytenr_orig; 3923a512bbf8SYan Zheng 3924a512bbf8SYan Zheng if (max_mirrors == 0) 3925a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3926a512bbf8SYan Zheng 3927d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 3928d5178578SJohannes Thumshirn 3929a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 3930314b6dd0SJohannes Thumshirn struct page *page; 3931314b6dd0SJohannes Thumshirn struct bio *bio; 3932314b6dd0SJohannes Thumshirn struct btrfs_super_block *disk_super; 3933314b6dd0SJohannes Thumshirn 393412659251SNaohiro Aota bytenr_orig = btrfs_sb_offset(i); 393512659251SNaohiro Aota ret = btrfs_sb_log_location(device, i, WRITE, &bytenr); 393612659251SNaohiro Aota if (ret == -ENOENT) { 393712659251SNaohiro Aota continue; 393812659251SNaohiro Aota } else if (ret < 0) { 393912659251SNaohiro Aota btrfs_err(device->fs_info, 394012659251SNaohiro Aota "couldn't get super block location for mirror %d", 394112659251SNaohiro Aota i); 394212659251SNaohiro Aota errors++; 394312659251SNaohiro Aota continue; 394412659251SNaohiro Aota } 3945935e5cc9SMiao Xie if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3946935e5cc9SMiao Xie device->commit_total_bytes) 3947a512bbf8SYan Zheng break; 3948a512bbf8SYan Zheng 394912659251SNaohiro Aota btrfs_set_super_bytenr(sb, bytenr_orig); 3950a512bbf8SYan Zheng 3951fd08001fSEric Biggers crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE, 3952fd08001fSEric Biggers BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, 3953fd08001fSEric Biggers sb->csum); 3954a512bbf8SYan Zheng 3955314b6dd0SJohannes Thumshirn page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT, 3956314b6dd0SJohannes Thumshirn GFP_NOFS); 3957314b6dd0SJohannes Thumshirn if (!page) { 3958fb456252SJeff Mahoney btrfs_err(device->fs_info, 3959314b6dd0SJohannes Thumshirn "couldn't get super block page for bytenr %llu", 3960f14d104dSDavid Sterba bytenr); 3961634554dcSJosef Bacik errors++; 3962634554dcSJosef Bacik continue; 3963634554dcSJosef Bacik } 3964634554dcSJosef Bacik 3965314b6dd0SJohannes Thumshirn /* Bump the refcount for wait_dev_supers() */ 3966314b6dd0SJohannes Thumshirn get_page(page); 3967a512bbf8SYan Zheng 3968314b6dd0SJohannes Thumshirn disk_super = page_address(page); 3969314b6dd0SJohannes Thumshirn memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE); 3970a512bbf8SYan Zheng 3971387125fcSChris Mason /* 3972314b6dd0SJohannes Thumshirn * Directly use bios here instead of relying on the page cache 3973314b6dd0SJohannes Thumshirn * to do I/O, so we don't lose the ability to do integrity 3974314b6dd0SJohannes Thumshirn * checking. 3975387125fcSChris Mason */ 397607888c66SChristoph Hellwig bio = bio_alloc(device->bdev, 1, 397707888c66SChristoph Hellwig REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO, 397807888c66SChristoph Hellwig GFP_NOFS); 3979314b6dd0SJohannes Thumshirn bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT; 3980314b6dd0SJohannes Thumshirn bio->bi_private = device; 3981314b6dd0SJohannes Thumshirn bio->bi_end_io = btrfs_end_super_write; 3982314b6dd0SJohannes Thumshirn __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE, 3983314b6dd0SJohannes Thumshirn offset_in_page(bytenr)); 3984314b6dd0SJohannes Thumshirn 3985314b6dd0SJohannes Thumshirn /* 3986314b6dd0SJohannes Thumshirn * We FUA only the first super block. The others we allow to 3987314b6dd0SJohannes Thumshirn * go down lazy and there's a short window where the on-disk 3988314b6dd0SJohannes Thumshirn * copies might still contain the older version. 3989314b6dd0SJohannes Thumshirn */ 39901b9e619cSOmar Sandoval if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) 3991314b6dd0SJohannes Thumshirn bio->bi_opf |= REQ_FUA; 3992314b6dd0SJohannes Thumshirn 399358ff51f1SChristoph Hellwig btrfsic_check_bio(bio); 399458ff51f1SChristoph Hellwig submit_bio(bio); 39958376d9e1SNaohiro Aota 39968376d9e1SNaohiro Aota if (btrfs_advance_sb_log(device, i)) 39978376d9e1SNaohiro Aota errors++; 3998a512bbf8SYan Zheng } 3999a512bbf8SYan Zheng return errors < i ? 0 : -1; 4000a512bbf8SYan Zheng } 4001a512bbf8SYan Zheng 4002387125fcSChris Mason /* 4003abbb3b8eSDavid Sterba * Wait for write completion of superblocks done by write_dev_supers, 4004abbb3b8eSDavid Sterba * @max_mirrors same for write and wait phases. 4005abbb3b8eSDavid Sterba * 4006314b6dd0SJohannes Thumshirn * Return number of errors when page is not found or not marked up to 4007abbb3b8eSDavid Sterba * date. 4008abbb3b8eSDavid Sterba */ 4009abbb3b8eSDavid Sterba static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) 4010abbb3b8eSDavid Sterba { 4011abbb3b8eSDavid Sterba int i; 4012abbb3b8eSDavid Sterba int errors = 0; 4013b6a535faSHoward McLauchlan bool primary_failed = false; 401412659251SNaohiro Aota int ret; 4015abbb3b8eSDavid Sterba u64 bytenr; 4016abbb3b8eSDavid Sterba 4017abbb3b8eSDavid Sterba if (max_mirrors == 0) 4018abbb3b8eSDavid Sterba max_mirrors = BTRFS_SUPER_MIRROR_MAX; 4019abbb3b8eSDavid Sterba 4020abbb3b8eSDavid Sterba for (i = 0; i < max_mirrors; i++) { 4021314b6dd0SJohannes Thumshirn struct page *page; 4022314b6dd0SJohannes Thumshirn 402312659251SNaohiro Aota ret = btrfs_sb_log_location(device, i, READ, &bytenr); 402412659251SNaohiro Aota if (ret == -ENOENT) { 402512659251SNaohiro Aota break; 402612659251SNaohiro Aota } else if (ret < 0) { 402712659251SNaohiro Aota errors++; 402812659251SNaohiro Aota if (i == 0) 402912659251SNaohiro Aota primary_failed = true; 403012659251SNaohiro Aota continue; 403112659251SNaohiro Aota } 4032abbb3b8eSDavid Sterba if (bytenr + BTRFS_SUPER_INFO_SIZE >= 4033abbb3b8eSDavid Sterba device->commit_total_bytes) 4034abbb3b8eSDavid Sterba break; 4035abbb3b8eSDavid Sterba 4036314b6dd0SJohannes Thumshirn page = find_get_page(device->bdev->bd_inode->i_mapping, 4037314b6dd0SJohannes Thumshirn bytenr >> PAGE_SHIFT); 4038314b6dd0SJohannes Thumshirn if (!page) { 4039abbb3b8eSDavid Sterba errors++; 4040b6a535faSHoward McLauchlan if (i == 0) 4041b6a535faSHoward McLauchlan primary_failed = true; 4042abbb3b8eSDavid Sterba continue; 4043abbb3b8eSDavid Sterba } 4044314b6dd0SJohannes Thumshirn /* Page is submitted locked and unlocked once the IO completes */ 4045314b6dd0SJohannes Thumshirn wait_on_page_locked(page); 4046314b6dd0SJohannes Thumshirn if (PageError(page)) { 4047abbb3b8eSDavid Sterba errors++; 4048b6a535faSHoward McLauchlan if (i == 0) 4049b6a535faSHoward McLauchlan primary_failed = true; 4050b6a535faSHoward McLauchlan } 4051abbb3b8eSDavid Sterba 4052314b6dd0SJohannes Thumshirn /* Drop our reference */ 4053314b6dd0SJohannes Thumshirn put_page(page); 4054abbb3b8eSDavid Sterba 4055314b6dd0SJohannes Thumshirn /* Drop the reference from the writing run */ 4056314b6dd0SJohannes Thumshirn put_page(page); 4057abbb3b8eSDavid Sterba } 4058abbb3b8eSDavid Sterba 4059b6a535faSHoward McLauchlan /* log error, force error return */ 4060b6a535faSHoward McLauchlan if (primary_failed) { 4061b6a535faSHoward McLauchlan btrfs_err(device->fs_info, "error writing primary super block to device %llu", 4062b6a535faSHoward McLauchlan device->devid); 4063b6a535faSHoward McLauchlan return -1; 4064b6a535faSHoward McLauchlan } 4065b6a535faSHoward McLauchlan 4066abbb3b8eSDavid Sterba return errors < i ? 0 : -1; 4067abbb3b8eSDavid Sterba } 4068abbb3b8eSDavid Sterba 4069abbb3b8eSDavid Sterba /* 4070387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 4071387125fcSChris Mason * for the barrier when it is done 4072387125fcSChris Mason */ 40734246a0b6SChristoph Hellwig static void btrfs_end_empty_barrier(struct bio *bio) 4074387125fcSChris Mason { 4075f9e69aa9SChristoph Hellwig bio_uninit(bio); 4076387125fcSChris Mason complete(bio->bi_private); 4077387125fcSChris Mason } 4078387125fcSChris Mason 4079387125fcSChris Mason /* 40804fc6441aSAnand Jain * Submit a flush request to the device if it supports it. Error handling is 40814fc6441aSAnand Jain * done in the waiting counterpart. 4082387125fcSChris Mason */ 40834fc6441aSAnand Jain static void write_dev_flush(struct btrfs_device *device) 4084387125fcSChris Mason { 4085f9e69aa9SChristoph Hellwig struct bio *bio = &device->flush_bio; 4086387125fcSChris Mason 4087a91cf0ffSWang Yugui #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY 4088a91cf0ffSWang Yugui /* 4089a91cf0ffSWang Yugui * When a disk has write caching disabled, we skip submission of a bio 4090a91cf0ffSWang Yugui * with flush and sync requests before writing the superblock, since 4091a91cf0ffSWang Yugui * it's not needed. However when the integrity checker is enabled, this 4092a91cf0ffSWang Yugui * results in reports that there are metadata blocks referred by a 4093a91cf0ffSWang Yugui * superblock that were not properly flushed. So don't skip the bio 4094a91cf0ffSWang Yugui * submission only when the integrity checker is enabled for the sake 4095a91cf0ffSWang Yugui * of simplicity, since this is a debug tool and not meant for use in 4096a91cf0ffSWang Yugui * non-debug builds. 4097a91cf0ffSWang Yugui */ 409808e688fdSChristoph Hellwig if (!bdev_write_cache(device->bdev)) 40994fc6441aSAnand Jain return; 4100a91cf0ffSWang Yugui #endif 4101387125fcSChris Mason 4102f9e69aa9SChristoph Hellwig bio_init(bio, device->bdev, NULL, 0, 4103f9e69aa9SChristoph Hellwig REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH); 4104387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 4105387125fcSChris Mason init_completion(&device->flush_wait); 4106387125fcSChris Mason bio->bi_private = &device->flush_wait; 4107387125fcSChris Mason 410858ff51f1SChristoph Hellwig btrfsic_check_bio(bio); 410958ff51f1SChristoph Hellwig submit_bio(bio); 41101c3063b6SAnand Jain set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 41114fc6441aSAnand Jain } 4112387125fcSChris Mason 41134fc6441aSAnand Jain /* 41144fc6441aSAnand Jain * If the flush bio has been submitted by write_dev_flush, wait for it. 41154fc6441aSAnand Jain */ 41168c27cb35SLinus Torvalds static blk_status_t wait_dev_flush(struct btrfs_device *device) 41174fc6441aSAnand Jain { 4118f9e69aa9SChristoph Hellwig struct bio *bio = &device->flush_bio; 41194fc6441aSAnand Jain 41201c3063b6SAnand Jain if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) 412158efbc9fSOmar Sandoval return BLK_STS_OK; 41224fc6441aSAnand Jain 41231c3063b6SAnand Jain clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 41242980d574SDavid Sterba wait_for_completion_io(&device->flush_wait); 41254fc6441aSAnand Jain 41268c27cb35SLinus Torvalds return bio->bi_status; 4127387125fcSChris Mason } 4128387125fcSChris Mason 4129d10b82feSQu Wenruo static int check_barrier_error(struct btrfs_fs_info *fs_info) 4130401b41e5SAnand Jain { 41316528b99dSAnand Jain if (!btrfs_check_rw_degradable(fs_info, NULL)) 4132401b41e5SAnand Jain return -EIO; 4133387125fcSChris Mason return 0; 4134387125fcSChris Mason } 4135387125fcSChris Mason 4136387125fcSChris Mason /* 4137387125fcSChris Mason * send an empty flush down to each device in parallel, 4138387125fcSChris Mason * then wait for them 4139387125fcSChris Mason */ 4140387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 4141387125fcSChris Mason { 4142387125fcSChris Mason struct list_head *head; 4143387125fcSChris Mason struct btrfs_device *dev; 41445af3e8ccSStefan Behrens int errors_wait = 0; 41454e4cbee9SChristoph Hellwig blk_status_t ret; 4146387125fcSChris Mason 41471538e6c5SDavid Sterba lockdep_assert_held(&info->fs_devices->device_list_mutex); 4148387125fcSChris Mason /* send down all the barriers */ 4149387125fcSChris Mason head = &info->fs_devices->devices; 41501538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4151e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 4152f88ba6a2SHidetoshi Seto continue; 4153cea7c8bfSAnand Jain if (!dev->bdev) 4154387125fcSChris Mason continue; 4155e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4156ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4157387125fcSChris Mason continue; 4158387125fcSChris Mason 41594fc6441aSAnand Jain write_dev_flush(dev); 416058efbc9fSOmar Sandoval dev->last_flush_error = BLK_STS_OK; 4161387125fcSChris Mason } 4162387125fcSChris Mason 4163387125fcSChris Mason /* wait for all the barriers */ 41641538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4165e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 4166f88ba6a2SHidetoshi Seto continue; 4167387125fcSChris Mason if (!dev->bdev) { 41685af3e8ccSStefan Behrens errors_wait++; 4169387125fcSChris Mason continue; 4170387125fcSChris Mason } 4171e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4172ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4173387125fcSChris Mason continue; 4174387125fcSChris Mason 41754fc6441aSAnand Jain ret = wait_dev_flush(dev); 4176401b41e5SAnand Jain if (ret) { 4177401b41e5SAnand Jain dev->last_flush_error = ret; 417866b4993eSDavid Sterba btrfs_dev_stat_inc_and_print(dev, 417966b4993eSDavid Sterba BTRFS_DEV_STAT_FLUSH_ERRS); 41805af3e8ccSStefan Behrens errors_wait++; 4181387125fcSChris Mason } 4182401b41e5SAnand Jain } 4183401b41e5SAnand Jain 4184cea7c8bfSAnand Jain if (errors_wait) { 4185401b41e5SAnand Jain /* 4186401b41e5SAnand Jain * At some point we need the status of all disks 4187401b41e5SAnand Jain * to arrive at the volume status. So error checking 4188401b41e5SAnand Jain * is being pushed to a separate loop. 4189401b41e5SAnand Jain */ 4190d10b82feSQu Wenruo return check_barrier_error(info); 4191401b41e5SAnand Jain } 4192387125fcSChris Mason return 0; 4193387125fcSChris Mason } 4194387125fcSChris Mason 4195943c6e99SZhao Lei int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 4196943c6e99SZhao Lei { 41978789f4feSZhao Lei int raid_type; 41988789f4feSZhao Lei int min_tolerated = INT_MAX; 4199943c6e99SZhao Lei 42008789f4feSZhao Lei if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 42018789f4feSZhao Lei (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 42028c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 42038789f4feSZhao Lei btrfs_raid_array[BTRFS_RAID_SINGLE]. 42048789f4feSZhao Lei tolerated_failures); 4205943c6e99SZhao Lei 42068789f4feSZhao Lei for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 42078789f4feSZhao Lei if (raid_type == BTRFS_RAID_SINGLE) 42088789f4feSZhao Lei continue; 420941a6e891SAnand Jain if (!(flags & btrfs_raid_array[raid_type].bg_flag)) 42108789f4feSZhao Lei continue; 42118c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 42128789f4feSZhao Lei btrfs_raid_array[raid_type]. 42138789f4feSZhao Lei tolerated_failures); 42148789f4feSZhao Lei } 4215943c6e99SZhao Lei 42168789f4feSZhao Lei if (min_tolerated == INT_MAX) { 4217ab8d0fc4SJeff Mahoney pr_warn("BTRFS: unknown raid flag: %llu", flags); 42188789f4feSZhao Lei min_tolerated = 0; 42198789f4feSZhao Lei } 42208789f4feSZhao Lei 42218789f4feSZhao Lei return min_tolerated; 4222943c6e99SZhao Lei } 4223943c6e99SZhao Lei 4224eece6a9cSDavid Sterba int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) 4225f2984462SChris Mason { 4226e5e9a520SChris Mason struct list_head *head; 4227f2984462SChris Mason struct btrfs_device *dev; 4228a061fc8dSChris Mason struct btrfs_super_block *sb; 4229f2984462SChris Mason struct btrfs_dev_item *dev_item; 4230f2984462SChris Mason int ret; 4231f2984462SChris Mason int do_barriers; 4232a236aed1SChris Mason int max_errors; 4233a236aed1SChris Mason int total_errors = 0; 4234a061fc8dSChris Mason u64 flags; 4235f2984462SChris Mason 42360b246afaSJeff Mahoney do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); 4237fed3b381SLiu Bo 4238fed3b381SLiu Bo /* 4239fed3b381SLiu Bo * max_mirrors == 0 indicates we're from commit_transaction, 4240fed3b381SLiu Bo * not from fsync where the tree roots in fs_info have not 4241fed3b381SLiu Bo * been consistent on disk. 4242fed3b381SLiu Bo */ 4243fed3b381SLiu Bo if (max_mirrors == 0) 42440b246afaSJeff Mahoney backup_super_roots(fs_info); 4245f2984462SChris Mason 42460b246afaSJeff Mahoney sb = fs_info->super_for_commit; 4247a061fc8dSChris Mason dev_item = &sb->dev_item; 4248e5e9a520SChris Mason 42490b246afaSJeff Mahoney mutex_lock(&fs_info->fs_devices->device_list_mutex); 42500b246afaSJeff Mahoney head = &fs_info->fs_devices->devices; 42510b246afaSJeff Mahoney max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; 4252387125fcSChris Mason 42535af3e8ccSStefan Behrens if (do_barriers) { 42540b246afaSJeff Mahoney ret = barrier_all_devices(fs_info); 42555af3e8ccSStefan Behrens if (ret) { 42565af3e8ccSStefan Behrens mutex_unlock( 42570b246afaSJeff Mahoney &fs_info->fs_devices->device_list_mutex); 42580b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 42595af3e8ccSStefan Behrens "errors while submitting device barriers."); 42605af3e8ccSStefan Behrens return ret; 42615af3e8ccSStefan Behrens } 42625af3e8ccSStefan Behrens } 4263387125fcSChris Mason 42641538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4265dfe25020SChris Mason if (!dev->bdev) { 4266dfe25020SChris Mason total_errors++; 4267dfe25020SChris Mason continue; 4268dfe25020SChris Mason } 4269e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4270ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4271dfe25020SChris Mason continue; 4272dfe25020SChris Mason 42732b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 4274a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 4275a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 42767df69d3eSMiao Xie btrfs_set_stack_device_total_bytes(dev_item, 4277935e5cc9SMiao Xie dev->commit_total_bytes); 4278ce7213c7SMiao Xie btrfs_set_stack_device_bytes_used(dev_item, 4279ce7213c7SMiao Xie dev->commit_bytes_used); 4280a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 4281a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 4282a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 4283a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 42847239ff4bSNikolay Borisov memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, 42857239ff4bSNikolay Borisov BTRFS_FSID_SIZE); 4286a512bbf8SYan Zheng 4287a061fc8dSChris Mason flags = btrfs_super_flags(sb); 4288a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 4289f2984462SChris Mason 429075cb857dSQu Wenruo ret = btrfs_validate_write_super(fs_info, sb); 429175cb857dSQu Wenruo if (ret < 0) { 429275cb857dSQu Wenruo mutex_unlock(&fs_info->fs_devices->device_list_mutex); 429375cb857dSQu Wenruo btrfs_handle_fs_error(fs_info, -EUCLEAN, 429475cb857dSQu Wenruo "unexpected superblock corruption detected"); 429575cb857dSQu Wenruo return -EUCLEAN; 429675cb857dSQu Wenruo } 429775cb857dSQu Wenruo 4298abbb3b8eSDavid Sterba ret = write_dev_supers(dev, sb, max_mirrors); 4299a236aed1SChris Mason if (ret) 4300a236aed1SChris Mason total_errors++; 4301f2984462SChris Mason } 4302a236aed1SChris Mason if (total_errors > max_errors) { 43030b246afaSJeff Mahoney btrfs_err(fs_info, "%d errors while writing supers", 4304d397712bSChris Mason total_errors); 43050b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 430679787eaaSJeff Mahoney 43079d565ba4SStefan Behrens /* FUA is masked off if unsupported and can't be the reason */ 43080b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 43090b246afaSJeff Mahoney "%d errors while writing supers", 43100b246afaSJeff Mahoney total_errors); 43119d565ba4SStefan Behrens return -EIO; 4312a236aed1SChris Mason } 4313f2984462SChris Mason 4314a512bbf8SYan Zheng total_errors = 0; 43151538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4316dfe25020SChris Mason if (!dev->bdev) 4317dfe25020SChris Mason continue; 4318e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4319ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4320dfe25020SChris Mason continue; 4321dfe25020SChris Mason 4322abbb3b8eSDavid Sterba ret = wait_dev_supers(dev, max_mirrors); 4323a512bbf8SYan Zheng if (ret) 43241259ab75SChris Mason total_errors++; 4325f2984462SChris Mason } 43260b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 4327a236aed1SChris Mason if (total_errors > max_errors) { 43280b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 43290b246afaSJeff Mahoney "%d errors while writing supers", 43300b246afaSJeff Mahoney total_errors); 433179787eaaSJeff Mahoney return -EIO; 4332a236aed1SChris Mason } 4333f2984462SChris Mason return 0; 4334f2984462SChris Mason } 4335f2984462SChris Mason 4336cb517eabSMiao Xie /* Drop a fs root from the radix tree and free it. */ 4337cb517eabSMiao Xie void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 4338cb517eabSMiao Xie struct btrfs_root *root) 43392619ba1fSChris Mason { 43404785e24fSJosef Bacik bool drop_ref = false; 43414785e24fSJosef Bacik 4342fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4343fc7cbcd4SDavid Sterba radix_tree_delete(&fs_info->fs_roots_radix, 4344fc7cbcd4SDavid Sterba (unsigned long)root->root_key.objectid); 4345fc7cbcd4SDavid Sterba if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state)) 43464785e24fSJosef Bacik drop_ref = true; 4347fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 434876dda93cSYan, Zheng 434984961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 4350ef67963dSJosef Bacik ASSERT(root->log_root == NULL); 43511c1ea4f7SLiu Bo if (root->reloc_root) { 435200246528SJosef Bacik btrfs_put_root(root->reloc_root); 43531c1ea4f7SLiu Bo root->reloc_root = NULL; 43541c1ea4f7SLiu Bo } 43551c1ea4f7SLiu Bo } 43563321719eSLiu Bo 43574785e24fSJosef Bacik if (drop_ref) 435800246528SJosef Bacik btrfs_put_root(root); 43592619ba1fSChris Mason } 43602619ba1fSChris Mason 4361c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 4362c146afadSYan Zheng { 4363fc7cbcd4SDavid Sterba u64 root_objectid = 0; 4364fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 4365fc7cbcd4SDavid Sterba int i = 0; 436665d33fd7SQu Wenruo int err = 0; 4367fc7cbcd4SDavid Sterba unsigned int ret = 0; 4368c146afadSYan Zheng 4369c146afadSYan Zheng while (1) { 4370fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4371fc7cbcd4SDavid Sterba ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4372fc7cbcd4SDavid Sterba (void **)gang, root_objectid, 4373fc7cbcd4SDavid Sterba ARRAY_SIZE(gang)); 4374fc7cbcd4SDavid Sterba if (!ret) { 4375fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4376c146afadSYan Zheng break; 437765d33fd7SQu Wenruo } 4378fc7cbcd4SDavid Sterba root_objectid = gang[ret - 1]->root_key.objectid + 1; 437966b4ffd1SJosef Bacik 4380fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 4381fc7cbcd4SDavid Sterba /* Avoid to grab roots in dead_roots */ 4382fc7cbcd4SDavid Sterba if (btrfs_root_refs(&gang[i]->root_item) == 0) { 4383fc7cbcd4SDavid Sterba gang[i] = NULL; 438465d33fd7SQu Wenruo continue; 4385c146afadSYan Zheng } 4386fc7cbcd4SDavid Sterba /* grab all the search result for later use */ 4387fc7cbcd4SDavid Sterba gang[i] = btrfs_grab_root(gang[i]); 4388fc7cbcd4SDavid Sterba } 4389fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4390fc7cbcd4SDavid Sterba 4391fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 4392fc7cbcd4SDavid Sterba if (!gang[i]) 4393fc7cbcd4SDavid Sterba continue; 4394fc7cbcd4SDavid Sterba root_objectid = gang[i]->root_key.objectid; 4395fc7cbcd4SDavid Sterba err = btrfs_orphan_cleanup(gang[i]); 4396fc7cbcd4SDavid Sterba if (err) 4397fc7cbcd4SDavid Sterba break; 4398fc7cbcd4SDavid Sterba btrfs_put_root(gang[i]); 4399fc7cbcd4SDavid Sterba } 4400fc7cbcd4SDavid Sterba root_objectid++; 4401c146afadSYan Zheng } 440265d33fd7SQu Wenruo 4403fc7cbcd4SDavid Sterba /* release the uncleaned roots due to error */ 4404fc7cbcd4SDavid Sterba for (; i < ret; i++) { 4405fc7cbcd4SDavid Sterba if (gang[i]) 4406fc7cbcd4SDavid Sterba btrfs_put_root(gang[i]); 440765d33fd7SQu Wenruo } 440865d33fd7SQu Wenruo return err; 4409c146afadSYan Zheng } 4410c146afadSYan Zheng 44116bccf3abSJeff Mahoney int btrfs_commit_super(struct btrfs_fs_info *fs_info) 4412c146afadSYan Zheng { 44136bccf3abSJeff Mahoney struct btrfs_root *root = fs_info->tree_root; 4414c146afadSYan Zheng struct btrfs_trans_handle *trans; 4415c146afadSYan Zheng 44160b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 44172ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 44180b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 44190b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 4420c71bf099SYan, Zheng 4421c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 44220b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 44230b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4424c71bf099SYan, Zheng 44257a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 44263612b495STsutomu Itoh if (IS_ERR(trans)) 44273612b495STsutomu Itoh return PTR_ERR(trans); 44283a45bb20SJeff Mahoney return btrfs_commit_transaction(trans); 4429c146afadSYan Zheng } 4430c146afadSYan Zheng 443136c86a9eSQu Wenruo static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info) 443236c86a9eSQu Wenruo { 443336c86a9eSQu Wenruo struct btrfs_transaction *trans; 443436c86a9eSQu Wenruo struct btrfs_transaction *tmp; 443536c86a9eSQu Wenruo bool found = false; 443636c86a9eSQu Wenruo 443736c86a9eSQu Wenruo if (list_empty(&fs_info->trans_list)) 443836c86a9eSQu Wenruo return; 443936c86a9eSQu Wenruo 444036c86a9eSQu Wenruo /* 444136c86a9eSQu Wenruo * This function is only called at the very end of close_ctree(), 444236c86a9eSQu Wenruo * thus no other running transaction, no need to take trans_lock. 444336c86a9eSQu Wenruo */ 444436c86a9eSQu Wenruo ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags)); 444536c86a9eSQu Wenruo list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) { 444636c86a9eSQu Wenruo struct extent_state *cached = NULL; 444736c86a9eSQu Wenruo u64 dirty_bytes = 0; 444836c86a9eSQu Wenruo u64 cur = 0; 444936c86a9eSQu Wenruo u64 found_start; 445036c86a9eSQu Wenruo u64 found_end; 445136c86a9eSQu Wenruo 445236c86a9eSQu Wenruo found = true; 445336c86a9eSQu Wenruo while (!find_first_extent_bit(&trans->dirty_pages, cur, 445436c86a9eSQu Wenruo &found_start, &found_end, EXTENT_DIRTY, &cached)) { 445536c86a9eSQu Wenruo dirty_bytes += found_end + 1 - found_start; 445636c86a9eSQu Wenruo cur = found_end + 1; 445736c86a9eSQu Wenruo } 445836c86a9eSQu Wenruo btrfs_warn(fs_info, 445936c86a9eSQu Wenruo "transaction %llu (with %llu dirty metadata bytes) is not committed", 446036c86a9eSQu Wenruo trans->transid, dirty_bytes); 446136c86a9eSQu Wenruo btrfs_cleanup_one_transaction(trans, fs_info); 446236c86a9eSQu Wenruo 446336c86a9eSQu Wenruo if (trans == fs_info->running_transaction) 446436c86a9eSQu Wenruo fs_info->running_transaction = NULL; 446536c86a9eSQu Wenruo list_del_init(&trans->list); 446636c86a9eSQu Wenruo 446736c86a9eSQu Wenruo btrfs_put_transaction(trans); 446836c86a9eSQu Wenruo trace_btrfs_transaction_commit(fs_info); 446936c86a9eSQu Wenruo } 447036c86a9eSQu Wenruo ASSERT(!found); 447136c86a9eSQu Wenruo } 447236c86a9eSQu Wenruo 4473b105e927SDavid Sterba void __cold close_ctree(struct btrfs_fs_info *fs_info) 4474eb60ceacSChris Mason { 4475c146afadSYan Zheng int ret; 4476e089f05cSChris Mason 4477afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); 447831e70e52SFilipe Manana 447931e70e52SFilipe Manana /* 44808a1f1e3dSFilipe Manana * If we had UNFINISHED_DROPS we could still be processing them, so 44818a1f1e3dSFilipe Manana * clear that bit and wake up relocation so it can stop. 44828a1f1e3dSFilipe Manana * We must do this before stopping the block group reclaim task, because 44838a1f1e3dSFilipe Manana * at btrfs_relocate_block_group() we wait for this bit, and after the 44848a1f1e3dSFilipe Manana * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we 44858a1f1e3dSFilipe Manana * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will 44868a1f1e3dSFilipe Manana * return 1. 44878a1f1e3dSFilipe Manana */ 44888a1f1e3dSFilipe Manana btrfs_wake_unfinished_drop(fs_info); 44898a1f1e3dSFilipe Manana 44908a1f1e3dSFilipe Manana /* 449131e70e52SFilipe Manana * We may have the reclaim task running and relocating a data block group, 449231e70e52SFilipe Manana * in which case it may create delayed iputs. So stop it before we park 449331e70e52SFilipe Manana * the cleaner kthread otherwise we can get new delayed iputs after 449431e70e52SFilipe Manana * parking the cleaner, and that can make the async reclaim task to hang 449531e70e52SFilipe Manana * if it's waiting for delayed iputs to complete, since the cleaner is 449631e70e52SFilipe Manana * parked and can not run delayed iputs - this will make us hang when 449731e70e52SFilipe Manana * trying to stop the async reclaim task. 449831e70e52SFilipe Manana */ 449931e70e52SFilipe Manana cancel_work_sync(&fs_info->reclaim_bgs_work); 4500d6fd0ae2SOmar Sandoval /* 4501d6fd0ae2SOmar Sandoval * We don't want the cleaner to start new transactions, add more delayed 4502d6fd0ae2SOmar Sandoval * iputs, etc. while we're closing. We can't use kthread_stop() yet 4503d6fd0ae2SOmar Sandoval * because that frees the task_struct, and the transaction kthread might 4504d6fd0ae2SOmar Sandoval * still try to wake up the cleaner. 4505d6fd0ae2SOmar Sandoval */ 4506d6fd0ae2SOmar Sandoval kthread_park(fs_info->cleaner_kthread); 4507a2135011SChris Mason 45087343dd61SJustin Maggard /* wait for the qgroup rescan worker to stop */ 4509d06f23d6SJeff Mahoney btrfs_qgroup_wait_for_completion(fs_info, false); 45107343dd61SJustin Maggard 4511803b2f54SStefan Behrens /* wait for the uuid_scan task to finish */ 4512803b2f54SStefan Behrens down(&fs_info->uuid_tree_rescan_sem); 4513803b2f54SStefan Behrens /* avoid complains from lockdep et al., set sem back to initial state */ 4514803b2f54SStefan Behrens up(&fs_info->uuid_tree_rescan_sem); 4515803b2f54SStefan Behrens 4516837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 4517aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 4518837d5b6eSIlya Dryomov 45198dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 45208dabb742SStefan Behrens 4521aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 45224cb5300bSChris Mason 45234cb5300bSChris Mason /* wait for any defraggers to finish */ 45244cb5300bSChris Mason wait_event(fs_info->transaction_wait, 45254cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 45264cb5300bSChris Mason 45274cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 452826176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 45294cb5300bSChris Mason 4530a362bb86SFilipe Manana /* 4531a362bb86SFilipe Manana * After we parked the cleaner kthread, ordered extents may have 4532a362bb86SFilipe Manana * completed and created new delayed iputs. If one of the async reclaim 4533a362bb86SFilipe Manana * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we 4534a362bb86SFilipe Manana * can hang forever trying to stop it, because if a delayed iput is 4535a362bb86SFilipe Manana * added after it ran btrfs_run_delayed_iputs() and before it called 4536a362bb86SFilipe Manana * btrfs_wait_on_delayed_iputs(), it will hang forever since there is 4537a362bb86SFilipe Manana * no one else to run iputs. 4538a362bb86SFilipe Manana * 4539a362bb86SFilipe Manana * So wait for all ongoing ordered extents to complete and then run 4540a362bb86SFilipe Manana * delayed iputs. This works because once we reach this point no one 4541a362bb86SFilipe Manana * can either create new ordered extents nor create delayed iputs 4542a362bb86SFilipe Manana * through some other means. 4543a362bb86SFilipe Manana * 4544a362bb86SFilipe Manana * Also note that btrfs_wait_ordered_roots() is not safe here, because 4545a362bb86SFilipe Manana * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent, 4546a362bb86SFilipe Manana * but the delayed iput for the respective inode is made only when doing 4547a362bb86SFilipe Manana * the final btrfs_put_ordered_extent() (which must happen at 4548a362bb86SFilipe Manana * btrfs_finish_ordered_io() when we are unmounting). 4549a362bb86SFilipe Manana */ 4550a362bb86SFilipe Manana btrfs_flush_workqueue(fs_info->endio_write_workers); 4551a362bb86SFilipe Manana /* Ordered extents for free space inodes. */ 4552a362bb86SFilipe Manana btrfs_flush_workqueue(fs_info->endio_freespace_worker); 4553a362bb86SFilipe Manana btrfs_run_delayed_iputs(fs_info); 4554a362bb86SFilipe Manana 455521c7e756SMiao Xie cancel_work_sync(&fs_info->async_reclaim_work); 455657056740SJosef Bacik cancel_work_sync(&fs_info->async_data_reclaim_work); 4557576fa348SJosef Bacik cancel_work_sync(&fs_info->preempt_reclaim_work); 455821c7e756SMiao Xie 4559b0643e59SDennis Zhou /* Cancel or finish ongoing discard work */ 4560b0643e59SDennis Zhou btrfs_discard_cleanup(fs_info); 4561b0643e59SDennis Zhou 4562bc98a42cSDavid Howells if (!sb_rdonly(fs_info->sb)) { 4563e44163e1SJeff Mahoney /* 4564d6fd0ae2SOmar Sandoval * The cleaner kthread is stopped, so do one final pass over 4565d6fd0ae2SOmar Sandoval * unused block groups. 4566e44163e1SJeff Mahoney */ 45670b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 4568e44163e1SJeff Mahoney 4569f0cc2cd7SFilipe Manana /* 4570f0cc2cd7SFilipe Manana * There might be existing delayed inode workers still running 4571f0cc2cd7SFilipe Manana * and holding an empty delayed inode item. We must wait for 4572f0cc2cd7SFilipe Manana * them to complete first because they can create a transaction. 4573f0cc2cd7SFilipe Manana * This happens when someone calls btrfs_balance_delayed_items() 4574f0cc2cd7SFilipe Manana * and then a transaction commit runs the same delayed nodes 4575f0cc2cd7SFilipe Manana * before any delayed worker has done something with the nodes. 4576f0cc2cd7SFilipe Manana * We must wait for any worker here and not at transaction 4577f0cc2cd7SFilipe Manana * commit time since that could cause a deadlock. 4578f0cc2cd7SFilipe Manana * This is a very rare case. 4579f0cc2cd7SFilipe Manana */ 4580f0cc2cd7SFilipe Manana btrfs_flush_workqueue(fs_info->delayed_workers); 4581f0cc2cd7SFilipe Manana 45826bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 4583d397712bSChris Mason if (ret) 458404892340SEric Sandeen btrfs_err(fs_info, "commit super ret %d", ret); 4585c146afadSYan Zheng } 4586ed2ff2cbSChris Mason 458784961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) 45882ff7e61eSJeff Mahoney btrfs_error_commit_super(fs_info); 4589acce952bSliubo 4590e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 4591e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 45928929ecfaSYan, Zheng 4593e187831eSJosef Bacik ASSERT(list_empty(&fs_info->delayed_iputs)); 4594afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); 4595f25784b3SYan Zheng 45965958253cSQu Wenruo if (btrfs_check_quota_leak(fs_info)) { 45975958253cSQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 45985958253cSQu Wenruo btrfs_err(fs_info, "qgroup reserved space leaked"); 45995958253cSQu Wenruo } 46005958253cSQu Wenruo 460104892340SEric Sandeen btrfs_free_qgroup_config(fs_info); 4602fe816d0fSNikolay Borisov ASSERT(list_empty(&fs_info->delalloc_roots)); 4603bcef60f2SArne Jansen 4604963d678bSMiao Xie if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 460504892340SEric Sandeen btrfs_info(fs_info, "at unmount delalloc count %lld", 4606963d678bSMiao Xie percpu_counter_sum(&fs_info->delalloc_bytes)); 4607b0c68f8bSChris Mason } 460831153d81SYan Zheng 46095deb17e1SJosef Bacik if (percpu_counter_sum(&fs_info->ordered_bytes)) 46104297ff84SJosef Bacik btrfs_info(fs_info, "at unmount dio bytes count %lld", 46115deb17e1SJosef Bacik percpu_counter_sum(&fs_info->ordered_bytes)); 46124297ff84SJosef Bacik 46136618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 4614b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 46155ac1d209SJeff Mahoney 46161a4319ccSLiu Bo btrfs_put_block_group_cache(fs_info); 46171a4319ccSLiu Bo 4618de348ee0SWang Shilong /* 4619de348ee0SWang Shilong * we must make sure there is not any read request to 4620de348ee0SWang Shilong * submit after we stopping all workers. 4621de348ee0SWang Shilong */ 4622de348ee0SWang Shilong invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 462396192499SJosef Bacik btrfs_stop_all_workers(fs_info); 462496192499SJosef Bacik 46250a31daa4SFilipe Manana /* We shouldn't have any transaction open at this point */ 462636c86a9eSQu Wenruo warn_about_uncommitted_trans(fs_info); 46270a31daa4SFilipe Manana 4628afcdd129SJosef Bacik clear_bit(BTRFS_FS_OPEN, &fs_info->flags); 46294273eaffSAnand Jain free_root_pointers(fs_info, true); 46308c38938cSJosef Bacik btrfs_free_fs_roots(fs_info); 46319ad6b7bcSChris Mason 46324e19443dSJosef Bacik /* 46334e19443dSJosef Bacik * We must free the block groups after dropping the fs_roots as we could 46344e19443dSJosef Bacik * have had an IO error and have left over tree log blocks that aren't 46354e19443dSJosef Bacik * cleaned up until the fs roots are freed. This makes the block group 46364e19443dSJosef Bacik * accounting appear to be wrong because there's pending reserved bytes, 46374e19443dSJosef Bacik * so make sure we do the block group cleanup afterwards. 46384e19443dSJosef Bacik */ 46394e19443dSJosef Bacik btrfs_free_block_groups(fs_info); 46404e19443dSJosef Bacik 464113e6c37bSJosef Bacik iput(fs_info->btree_inode); 4642d6bfde87SChris Mason 464321adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 46440b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) 46452ff7e61eSJeff Mahoney btrfsic_unmount(fs_info->fs_devices); 464621adbd5cSStefan Behrens #endif 464721adbd5cSStefan Behrens 46480b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 464968c94e55SNikolay Borisov btrfs_close_devices(fs_info->fs_devices); 4650eb60ceacSChris Mason } 4651eb60ceacSChris Mason 4652b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 4653b9fab919SChris Mason int atomic) 4654ccd467d6SChris Mason { 46551259ab75SChris Mason int ret; 4656727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 46571259ab75SChris Mason 46580b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 46591259ab75SChris Mason if (!ret) 46601259ab75SChris Mason return ret; 46611259ab75SChris Mason 46621259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 4663b9fab919SChris Mason parent_transid, atomic); 4664b9fab919SChris Mason if (ret == -EAGAIN) 4665b9fab919SChris Mason return ret; 46661259ab75SChris Mason return !ret; 46675f39d397SChris Mason } 46686702ed49SChris Mason 46695f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 46705f39d397SChris Mason { 46712f4d60dfSQu Wenruo struct btrfs_fs_info *fs_info = buf->fs_info; 46725f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 4673b9473439SChris Mason int was_dirty; 4674b4ce94deSChris Mason 467506ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 467606ea65a3SJosef Bacik /* 467706ea65a3SJosef Bacik * This is a fast path so only do this check if we have sanity tests 467852042d8eSAndrea Gelmini * enabled. Normal people shouldn't be using unmapped buffers as dirty 467906ea65a3SJosef Bacik * outside of the sanity tests. 468006ea65a3SJosef Bacik */ 4681b0132a3bSNikolay Borisov if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) 468206ea65a3SJosef Bacik return; 468306ea65a3SJosef Bacik #endif 468449d0c642SFilipe Manana btrfs_assert_tree_write_locked(buf); 46850b246afaSJeff Mahoney if (transid != fs_info->generation) 46865d163e0eSJeff Mahoney WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", 46870b246afaSJeff Mahoney buf->start, transid, fs_info->generation); 46880b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 4689e2d84521SMiao Xie if (!was_dirty) 4690104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 4691e2d84521SMiao Xie buf->len, 46920b246afaSJeff Mahoney fs_info->dirty_metadata_batch); 46931f21ef0aSFilipe Manana #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 469469fc6cbbSQu Wenruo /* 469569fc6cbbSQu Wenruo * Since btrfs_mark_buffer_dirty() can be called with item pointer set 469669fc6cbbSQu Wenruo * but item data not updated. 469769fc6cbbSQu Wenruo * So here we should only check item pointers, not item data. 469869fc6cbbSQu Wenruo */ 469969fc6cbbSQu Wenruo if (btrfs_header_level(buf) == 0 && 4700cfdaad5eSDavid Sterba btrfs_check_leaf_relaxed(buf)) { 4701a4f78750SDavid Sterba btrfs_print_leaf(buf); 47021f21ef0aSFilipe Manana ASSERT(0); 47031f21ef0aSFilipe Manana } 47041f21ef0aSFilipe Manana #endif 4705eb60ceacSChris Mason } 4706eb60ceacSChris Mason 47072ff7e61eSJeff Mahoney static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, 4708b53d3f5dSLiu Bo int flush_delayed) 470935b7e476SChris Mason { 4710188de649SChris Mason /* 4711188de649SChris Mason * looks as though older kernels can get into trouble with 4712188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 4713188de649SChris Mason */ 4714e2d84521SMiao Xie int ret; 4715d6bfde87SChris Mason 47166933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 4717d6bfde87SChris Mason return; 4718d6bfde87SChris Mason 4719b53d3f5dSLiu Bo if (flush_delayed) 47202ff7e61eSJeff Mahoney btrfs_balance_delayed_items(fs_info); 472116cdcec7SMiao Xie 4722d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 4723d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 4724d814a491SEthan Lien fs_info->dirty_metadata_batch); 4725e2d84521SMiao Xie if (ret > 0) { 47260b246afaSJeff Mahoney balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); 472716cdcec7SMiao Xie } 472816cdcec7SMiao Xie } 472916cdcec7SMiao Xie 47302ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) 473116cdcec7SMiao Xie { 47322ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 1); 473335b7e476SChris Mason } 4734b53d3f5dSLiu Bo 47352ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) 4736b53d3f5dSLiu Bo { 47372ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 0); 4738d6bfde87SChris Mason } 47396b80053dSChris Mason 47402ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) 4741acce952bSliubo { 4742fe816d0fSNikolay Borisov /* cleanup FS via transaction */ 4743fe816d0fSNikolay Borisov btrfs_cleanup_transaction(fs_info); 4744fe816d0fSNikolay Borisov 47450b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 47462ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 47470b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 4748acce952bSliubo 47490b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 47500b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4751acce952bSliubo } 4752acce952bSliubo 4753ef67963dSJosef Bacik static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info) 4754ef67963dSJosef Bacik { 4755fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 4756fc7cbcd4SDavid Sterba u64 root_objectid = 0; 4757fc7cbcd4SDavid Sterba int ret; 4758ef67963dSJosef Bacik 4759fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4760fc7cbcd4SDavid Sterba while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4761fc7cbcd4SDavid Sterba (void **)gang, root_objectid, 4762fc7cbcd4SDavid Sterba ARRAY_SIZE(gang))) != 0) { 4763fc7cbcd4SDavid Sterba int i; 4764ef67963dSJosef Bacik 4765fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) 4766fc7cbcd4SDavid Sterba gang[i] = btrfs_grab_root(gang[i]); 4767fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4768fc7cbcd4SDavid Sterba 4769fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 4770fc7cbcd4SDavid Sterba if (!gang[i]) 4771ef67963dSJosef Bacik continue; 4772fc7cbcd4SDavid Sterba root_objectid = gang[i]->root_key.objectid; 4773fc7cbcd4SDavid Sterba btrfs_free_log(NULL, gang[i]); 4774fc7cbcd4SDavid Sterba btrfs_put_root(gang[i]); 4775ef67963dSJosef Bacik } 4776fc7cbcd4SDavid Sterba root_objectid++; 4777fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4778ef67963dSJosef Bacik } 4779fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4780ef67963dSJosef Bacik btrfs_free_log_root_tree(NULL, fs_info); 4781ef67963dSJosef Bacik } 4782ef67963dSJosef Bacik 4783143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4784acce952bSliubo { 4785acce952bSliubo struct btrfs_ordered_extent *ordered; 4786acce952bSliubo 4787199c2a9cSMiao Xie spin_lock(&root->ordered_extent_lock); 4788779880efSJosef Bacik /* 4789779880efSJosef Bacik * This will just short circuit the ordered completion stuff which will 4790779880efSJosef Bacik * make sure the ordered extent gets properly cleaned up. 4791779880efSJosef Bacik */ 4792199c2a9cSMiao Xie list_for_each_entry(ordered, &root->ordered_extents, 4793779880efSJosef Bacik root_extent_list) 4794779880efSJosef Bacik set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4795199c2a9cSMiao Xie spin_unlock(&root->ordered_extent_lock); 4796199c2a9cSMiao Xie } 4797199c2a9cSMiao Xie 4798199c2a9cSMiao Xie static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4799199c2a9cSMiao Xie { 4800199c2a9cSMiao Xie struct btrfs_root *root; 4801199c2a9cSMiao Xie struct list_head splice; 4802199c2a9cSMiao Xie 4803199c2a9cSMiao Xie INIT_LIST_HEAD(&splice); 4804199c2a9cSMiao Xie 4805199c2a9cSMiao Xie spin_lock(&fs_info->ordered_root_lock); 4806199c2a9cSMiao Xie list_splice_init(&fs_info->ordered_roots, &splice); 4807199c2a9cSMiao Xie while (!list_empty(&splice)) { 4808199c2a9cSMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4809199c2a9cSMiao Xie ordered_root); 48101de2cfdeSJosef Bacik list_move_tail(&root->ordered_root, 48111de2cfdeSJosef Bacik &fs_info->ordered_roots); 4812199c2a9cSMiao Xie 48132a85d9caSLiu Bo spin_unlock(&fs_info->ordered_root_lock); 4814199c2a9cSMiao Xie btrfs_destroy_ordered_extents(root); 4815199c2a9cSMiao Xie 48162a85d9caSLiu Bo cond_resched(); 48172a85d9caSLiu Bo spin_lock(&fs_info->ordered_root_lock); 4818199c2a9cSMiao Xie } 4819199c2a9cSMiao Xie spin_unlock(&fs_info->ordered_root_lock); 482074d5d229SJosef Bacik 482174d5d229SJosef Bacik /* 482274d5d229SJosef Bacik * We need this here because if we've been flipped read-only we won't 482374d5d229SJosef Bacik * get sync() from the umount, so we need to make sure any ordered 482474d5d229SJosef Bacik * extents that haven't had their dirty pages IO start writeout yet 482574d5d229SJosef Bacik * actually get run and error out properly. 482674d5d229SJosef Bacik */ 482774d5d229SJosef Bacik btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 4828acce952bSliubo } 4829acce952bSliubo 483035a3621bSStefan Behrens static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 48312ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4832acce952bSliubo { 4833acce952bSliubo struct rb_node *node; 4834acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 4835acce952bSliubo struct btrfs_delayed_ref_node *ref; 4836acce952bSliubo int ret = 0; 4837acce952bSliubo 4838acce952bSliubo delayed_refs = &trans->delayed_refs; 4839acce952bSliubo 4840acce952bSliubo spin_lock(&delayed_refs->lock); 4841d7df2c79SJosef Bacik if (atomic_read(&delayed_refs->num_entries) == 0) { 4842cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 4843b79ce3ddSDavid Sterba btrfs_debug(fs_info, "delayed_refs has NO entry"); 4844acce952bSliubo return ret; 4845acce952bSliubo } 4846acce952bSliubo 48475c9d028bSLiu Bo while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { 4848d7df2c79SJosef Bacik struct btrfs_delayed_ref_head *head; 48490e0adbcfSJosef Bacik struct rb_node *n; 4850e78417d1SJosef Bacik bool pin_bytes = false; 4851acce952bSliubo 4852d7df2c79SJosef Bacik head = rb_entry(node, struct btrfs_delayed_ref_head, 4853d7df2c79SJosef Bacik href_node); 48543069bd26SJosef Bacik if (btrfs_delayed_ref_lock(delayed_refs, head)) 4855b939d1abSJosef Bacik continue; 48563069bd26SJosef Bacik 4857d7df2c79SJosef Bacik spin_lock(&head->lock); 4858e3d03965SLiu Bo while ((n = rb_first_cached(&head->ref_tree)) != NULL) { 48590e0adbcfSJosef Bacik ref = rb_entry(n, struct btrfs_delayed_ref_node, 48600e0adbcfSJosef Bacik ref_node); 4861d7df2c79SJosef Bacik ref->in_tree = 0; 4862e3d03965SLiu Bo rb_erase_cached(&ref->ref_node, &head->ref_tree); 48630e0adbcfSJosef Bacik RB_CLEAR_NODE(&ref->ref_node); 48641d57ee94SWang Xiaoguang if (!list_empty(&ref->add_list)) 48651d57ee94SWang Xiaoguang list_del(&ref->add_list); 4866d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4867d7df2c79SJosef Bacik btrfs_put_delayed_ref(ref); 4868d7df2c79SJosef Bacik } 486954067ae9SJosef Bacik if (head->must_insert_reserved) 4870e78417d1SJosef Bacik pin_bytes = true; 487178a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 4872fa781ceaSJosef Bacik btrfs_delete_ref_head(delayed_refs, head); 4873d7df2c79SJosef Bacik spin_unlock(&head->lock); 4874acce952bSliubo spin_unlock(&delayed_refs->lock); 4875e78417d1SJosef Bacik mutex_unlock(&head->mutex); 4876acce952bSliubo 4877f603bb94SNikolay Borisov if (pin_bytes) { 4878f603bb94SNikolay Borisov struct btrfs_block_group *cache; 4879f603bb94SNikolay Borisov 4880f603bb94SNikolay Borisov cache = btrfs_lookup_block_group(fs_info, head->bytenr); 4881f603bb94SNikolay Borisov BUG_ON(!cache); 4882f603bb94SNikolay Borisov 4883f603bb94SNikolay Borisov spin_lock(&cache->space_info->lock); 4884f603bb94SNikolay Borisov spin_lock(&cache->lock); 4885f603bb94SNikolay Borisov cache->pinned += head->num_bytes; 4886f603bb94SNikolay Borisov btrfs_space_info_update_bytes_pinned(fs_info, 4887f603bb94SNikolay Borisov cache->space_info, head->num_bytes); 4888f603bb94SNikolay Borisov cache->reserved -= head->num_bytes; 4889f603bb94SNikolay Borisov cache->space_info->bytes_reserved -= head->num_bytes; 4890f603bb94SNikolay Borisov spin_unlock(&cache->lock); 4891f603bb94SNikolay Borisov spin_unlock(&cache->space_info->lock); 4892f603bb94SNikolay Borisov 4893f603bb94SNikolay Borisov btrfs_put_block_group(cache); 4894f603bb94SNikolay Borisov 4895f603bb94SNikolay Borisov btrfs_error_unpin_extent_range(fs_info, head->bytenr, 4896f603bb94SNikolay Borisov head->bytenr + head->num_bytes - 1); 4897f603bb94SNikolay Borisov } 489831890da0SJosef Bacik btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); 4899d278850eSJosef Bacik btrfs_put_delayed_ref_head(head); 4900acce952bSliubo cond_resched(); 4901acce952bSliubo spin_lock(&delayed_refs->lock); 4902acce952bSliubo } 490381f7eb00SJeff Mahoney btrfs_qgroup_destroy_extent_records(trans); 4904acce952bSliubo 4905acce952bSliubo spin_unlock(&delayed_refs->lock); 4906acce952bSliubo 4907acce952bSliubo return ret; 4908acce952bSliubo } 4909acce952bSliubo 4910143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 4911acce952bSliubo { 4912acce952bSliubo struct btrfs_inode *btrfs_inode; 4913acce952bSliubo struct list_head splice; 4914acce952bSliubo 4915acce952bSliubo INIT_LIST_HEAD(&splice); 4916acce952bSliubo 4917eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4918eb73c1b7SMiao Xie list_splice_init(&root->delalloc_inodes, &splice); 4919acce952bSliubo 4920acce952bSliubo while (!list_empty(&splice)) { 4921fe816d0fSNikolay Borisov struct inode *inode = NULL; 4922eb73c1b7SMiao Xie btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 4923acce952bSliubo delalloc_inodes); 4924fe816d0fSNikolay Borisov __btrfs_del_delalloc_inode(root, btrfs_inode); 4925eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4926acce952bSliubo 4927fe816d0fSNikolay Borisov /* 4928fe816d0fSNikolay Borisov * Make sure we get a live inode and that it'll not disappear 4929fe816d0fSNikolay Borisov * meanwhile. 4930fe816d0fSNikolay Borisov */ 4931fe816d0fSNikolay Borisov inode = igrab(&btrfs_inode->vfs_inode); 4932fe816d0fSNikolay Borisov if (inode) { 4933fe816d0fSNikolay Borisov invalidate_inode_pages2(inode->i_mapping); 4934fe816d0fSNikolay Borisov iput(inode); 4935fe816d0fSNikolay Borisov } 4936eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4937acce952bSliubo } 4938eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4939eb73c1b7SMiao Xie } 4940eb73c1b7SMiao Xie 4941eb73c1b7SMiao Xie static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 4942eb73c1b7SMiao Xie { 4943eb73c1b7SMiao Xie struct btrfs_root *root; 4944eb73c1b7SMiao Xie struct list_head splice; 4945eb73c1b7SMiao Xie 4946eb73c1b7SMiao Xie INIT_LIST_HEAD(&splice); 4947eb73c1b7SMiao Xie 4948eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4949eb73c1b7SMiao Xie list_splice_init(&fs_info->delalloc_roots, &splice); 4950eb73c1b7SMiao Xie while (!list_empty(&splice)) { 4951eb73c1b7SMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4952eb73c1b7SMiao Xie delalloc_root); 495300246528SJosef Bacik root = btrfs_grab_root(root); 4954eb73c1b7SMiao Xie BUG_ON(!root); 4955eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4956eb73c1b7SMiao Xie 4957eb73c1b7SMiao Xie btrfs_destroy_delalloc_inodes(root); 495800246528SJosef Bacik btrfs_put_root(root); 4959eb73c1b7SMiao Xie 4960eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4961eb73c1b7SMiao Xie } 4962eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4963acce952bSliubo } 4964acce952bSliubo 49652ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 4966acce952bSliubo struct extent_io_tree *dirty_pages, 4967acce952bSliubo int mark) 4968acce952bSliubo { 4969acce952bSliubo int ret; 4970acce952bSliubo struct extent_buffer *eb; 4971acce952bSliubo u64 start = 0; 4972acce952bSliubo u64 end; 4973acce952bSliubo 4974acce952bSliubo while (1) { 4975acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 4976e6138876SJosef Bacik mark, NULL); 4977acce952bSliubo if (ret) 4978acce952bSliubo break; 4979acce952bSliubo 498091166212SDavid Sterba clear_extent_bits(dirty_pages, start, end, mark); 4981acce952bSliubo while (start <= end) { 49820b246afaSJeff Mahoney eb = find_extent_buffer(fs_info, start); 49830b246afaSJeff Mahoney start += fs_info->nodesize; 4984fd8b2b61SJosef Bacik if (!eb) 4985acce952bSliubo continue; 4986fd8b2b61SJosef Bacik wait_on_extent_buffer_writeback(eb); 4987acce952bSliubo 4988fd8b2b61SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 4989fd8b2b61SJosef Bacik &eb->bflags)) 4990fd8b2b61SJosef Bacik clear_extent_buffer_dirty(eb); 4991fd8b2b61SJosef Bacik free_extent_buffer_stale(eb); 4992acce952bSliubo } 4993acce952bSliubo } 4994acce952bSliubo 4995acce952bSliubo return ret; 4996acce952bSliubo } 4997acce952bSliubo 49982ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 4999fe119a6eSNikolay Borisov struct extent_io_tree *unpin) 5000acce952bSliubo { 5001acce952bSliubo u64 start; 5002acce952bSliubo u64 end; 5003acce952bSliubo int ret; 5004acce952bSliubo 5005acce952bSliubo while (1) { 50060e6ec385SFilipe Manana struct extent_state *cached_state = NULL; 50070e6ec385SFilipe Manana 5008fcd5e742SLu Fengqi /* 5009fcd5e742SLu Fengqi * The btrfs_finish_extent_commit() may get the same range as 5010fcd5e742SLu Fengqi * ours between find_first_extent_bit and clear_extent_dirty. 5011fcd5e742SLu Fengqi * Hence, hold the unused_bg_unpin_mutex to avoid double unpin 5012fcd5e742SLu Fengqi * the same extent range. 5013fcd5e742SLu Fengqi */ 5014fcd5e742SLu Fengqi mutex_lock(&fs_info->unused_bg_unpin_mutex); 5015acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 50160e6ec385SFilipe Manana EXTENT_DIRTY, &cached_state); 5017fcd5e742SLu Fengqi if (ret) { 5018fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 5019acce952bSliubo break; 5020fcd5e742SLu Fengqi } 5021acce952bSliubo 50220e6ec385SFilipe Manana clear_extent_dirty(unpin, start, end, &cached_state); 50230e6ec385SFilipe Manana free_extent_state(cached_state); 50242ff7e61eSJeff Mahoney btrfs_error_unpin_extent_range(fs_info, start, end); 5025fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 5026acce952bSliubo cond_resched(); 5027acce952bSliubo } 5028acce952bSliubo 5029acce952bSliubo return 0; 5030acce952bSliubo } 5031acce952bSliubo 503232da5386SDavid Sterba static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache) 5033c79a1751SLiu Bo { 5034c79a1751SLiu Bo struct inode *inode; 5035c79a1751SLiu Bo 5036c79a1751SLiu Bo inode = cache->io_ctl.inode; 5037c79a1751SLiu Bo if (inode) { 5038c79a1751SLiu Bo invalidate_inode_pages2(inode->i_mapping); 5039c79a1751SLiu Bo BTRFS_I(inode)->generation = 0; 5040c79a1751SLiu Bo cache->io_ctl.inode = NULL; 5041c79a1751SLiu Bo iput(inode); 5042c79a1751SLiu Bo } 5043bbc37d6eSFilipe Manana ASSERT(cache->io_ctl.pages == NULL); 5044c79a1751SLiu Bo btrfs_put_block_group(cache); 5045c79a1751SLiu Bo } 5046c79a1751SLiu Bo 5047c79a1751SLiu Bo void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, 50482ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 5049c79a1751SLiu Bo { 505032da5386SDavid Sterba struct btrfs_block_group *cache; 5051c79a1751SLiu Bo 5052c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 5053c79a1751SLiu Bo while (!list_empty(&cur_trans->dirty_bgs)) { 5054c79a1751SLiu Bo cache = list_first_entry(&cur_trans->dirty_bgs, 505532da5386SDavid Sterba struct btrfs_block_group, 5056c79a1751SLiu Bo dirty_list); 5057c79a1751SLiu Bo 5058c79a1751SLiu Bo if (!list_empty(&cache->io_list)) { 5059c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 5060c79a1751SLiu Bo list_del_init(&cache->io_list); 5061c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 5062c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 5063c79a1751SLiu Bo } 5064c79a1751SLiu Bo 5065c79a1751SLiu Bo list_del_init(&cache->dirty_list); 5066c79a1751SLiu Bo spin_lock(&cache->lock); 5067c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 5068c79a1751SLiu Bo spin_unlock(&cache->lock); 5069c79a1751SLiu Bo 5070c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 5071c79a1751SLiu Bo btrfs_put_block_group(cache); 5072ba2c4d4eSJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 5073c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 5074c79a1751SLiu Bo } 5075c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 5076c79a1751SLiu Bo 507745ae2c18SNikolay Borisov /* 507845ae2c18SNikolay Borisov * Refer to the definition of io_bgs member for details why it's safe 507945ae2c18SNikolay Borisov * to use it without any locking 508045ae2c18SNikolay Borisov */ 5081c79a1751SLiu Bo while (!list_empty(&cur_trans->io_bgs)) { 5082c79a1751SLiu Bo cache = list_first_entry(&cur_trans->io_bgs, 508332da5386SDavid Sterba struct btrfs_block_group, 5084c79a1751SLiu Bo io_list); 5085c79a1751SLiu Bo 5086c79a1751SLiu Bo list_del_init(&cache->io_list); 5087c79a1751SLiu Bo spin_lock(&cache->lock); 5088c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 5089c79a1751SLiu Bo spin_unlock(&cache->lock); 5090c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 5091c79a1751SLiu Bo } 5092c79a1751SLiu Bo } 5093c79a1751SLiu Bo 509449b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 50952ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 509649b25e05SJeff Mahoney { 5097bbbf7243SNikolay Borisov struct btrfs_device *dev, *tmp; 5098bbbf7243SNikolay Borisov 50992ff7e61eSJeff Mahoney btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 5100c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->dirty_bgs)); 5101c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->io_bgs)); 5102c79a1751SLiu Bo 5103bbbf7243SNikolay Borisov list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list, 5104bbbf7243SNikolay Borisov post_commit_list) { 5105bbbf7243SNikolay Borisov list_del_init(&dev->post_commit_list); 5106bbbf7243SNikolay Borisov } 5107bbbf7243SNikolay Borisov 51082ff7e61eSJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, fs_info); 510949b25e05SJeff Mahoney 51104a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 51110b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 511249b25e05SJeff Mahoney 51134a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 51140b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 511549b25e05SJeff Mahoney 5116ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 511767cde344SMiao Xie 51182ff7e61eSJeff Mahoney btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, 511949b25e05SJeff Mahoney EXTENT_DIRTY); 5120fe119a6eSNikolay Borisov btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents); 512149b25e05SJeff Mahoney 5122d3575156SNaohiro Aota btrfs_free_redirty_list(cur_trans); 5123d3575156SNaohiro Aota 51244a9d8bdeSMiao Xie cur_trans->state =TRANS_STATE_COMPLETED; 51254a9d8bdeSMiao Xie wake_up(&cur_trans->commit_wait); 512649b25e05SJeff Mahoney } 512749b25e05SJeff Mahoney 51282ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) 5129acce952bSliubo { 5130acce952bSliubo struct btrfs_transaction *t; 5131acce952bSliubo 51320b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 5133acce952bSliubo 51340b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 51350b246afaSJeff Mahoney while (!list_empty(&fs_info->trans_list)) { 51360b246afaSJeff Mahoney t = list_first_entry(&fs_info->trans_list, 5137724e2315SJosef Bacik struct btrfs_transaction, list); 5138724e2315SJosef Bacik if (t->state >= TRANS_STATE_COMMIT_START) { 51399b64f57dSElena Reshetova refcount_inc(&t->use_count); 51400b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 51412ff7e61eSJeff Mahoney btrfs_wait_for_commit(fs_info, t->transid); 5142724e2315SJosef Bacik btrfs_put_transaction(t); 51430b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5144724e2315SJosef Bacik continue; 5145724e2315SJosef Bacik } 51460b246afaSJeff Mahoney if (t == fs_info->running_transaction) { 5147724e2315SJosef Bacik t->state = TRANS_STATE_COMMIT_DOING; 51480b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 5149724e2315SJosef Bacik /* 5150724e2315SJosef Bacik * We wait for 0 num_writers since we don't hold a trans 5151724e2315SJosef Bacik * handle open currently for this transaction. 5152724e2315SJosef Bacik */ 5153724e2315SJosef Bacik wait_event(t->writer_wait, 5154724e2315SJosef Bacik atomic_read(&t->num_writers) == 0); 5155724e2315SJosef Bacik } else { 51560b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 5157724e2315SJosef Bacik } 51582ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(t, fs_info); 5159724e2315SJosef Bacik 51600b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 51610b246afaSJeff Mahoney if (t == fs_info->running_transaction) 51620b246afaSJeff Mahoney fs_info->running_transaction = NULL; 5163724e2315SJosef Bacik list_del_init(&t->list); 51640b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 5165a4abeea4SJosef Bacik 5166724e2315SJosef Bacik btrfs_put_transaction(t); 51672e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 51680b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5169724e2315SJosef Bacik } 51700b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 51710b246afaSJeff Mahoney btrfs_destroy_all_ordered_extents(fs_info); 5172ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 5173ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 51740b246afaSJeff Mahoney btrfs_destroy_all_delalloc_inodes(fs_info); 5175ef67963dSJosef Bacik btrfs_drop_all_logs(fs_info); 51760b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 5177acce952bSliubo 5178acce952bSliubo return 0; 5179acce952bSliubo } 5180ec7d6dfdSNikolay Borisov 5181453e4873SNikolay Borisov int btrfs_init_root_free_objectid(struct btrfs_root *root) 5182ec7d6dfdSNikolay Borisov { 5183ec7d6dfdSNikolay Borisov struct btrfs_path *path; 5184ec7d6dfdSNikolay Borisov int ret; 5185ec7d6dfdSNikolay Borisov struct extent_buffer *l; 5186ec7d6dfdSNikolay Borisov struct btrfs_key search_key; 5187ec7d6dfdSNikolay Borisov struct btrfs_key found_key; 5188ec7d6dfdSNikolay Borisov int slot; 5189ec7d6dfdSNikolay Borisov 5190ec7d6dfdSNikolay Borisov path = btrfs_alloc_path(); 5191ec7d6dfdSNikolay Borisov if (!path) 5192ec7d6dfdSNikolay Borisov return -ENOMEM; 5193ec7d6dfdSNikolay Borisov 5194ec7d6dfdSNikolay Borisov search_key.objectid = BTRFS_LAST_FREE_OBJECTID; 5195ec7d6dfdSNikolay Borisov search_key.type = -1; 5196ec7d6dfdSNikolay Borisov search_key.offset = (u64)-1; 5197ec7d6dfdSNikolay Borisov ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); 5198ec7d6dfdSNikolay Borisov if (ret < 0) 5199ec7d6dfdSNikolay Borisov goto error; 5200ec7d6dfdSNikolay Borisov BUG_ON(ret == 0); /* Corruption */ 5201ec7d6dfdSNikolay Borisov if (path->slots[0] > 0) { 5202ec7d6dfdSNikolay Borisov slot = path->slots[0] - 1; 5203ec7d6dfdSNikolay Borisov l = path->nodes[0]; 5204ec7d6dfdSNikolay Borisov btrfs_item_key_to_cpu(l, &found_key, slot); 520523125104SNikolay Borisov root->free_objectid = max_t(u64, found_key.objectid + 1, 520623125104SNikolay Borisov BTRFS_FIRST_FREE_OBJECTID); 5207ec7d6dfdSNikolay Borisov } else { 520823125104SNikolay Borisov root->free_objectid = BTRFS_FIRST_FREE_OBJECTID; 5209ec7d6dfdSNikolay Borisov } 5210ec7d6dfdSNikolay Borisov ret = 0; 5211ec7d6dfdSNikolay Borisov error: 5212ec7d6dfdSNikolay Borisov btrfs_free_path(path); 5213ec7d6dfdSNikolay Borisov return ret; 5214ec7d6dfdSNikolay Borisov } 5215ec7d6dfdSNikolay Borisov 5216543068a2SNikolay Borisov int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid) 5217ec7d6dfdSNikolay Borisov { 5218ec7d6dfdSNikolay Borisov int ret; 5219ec7d6dfdSNikolay Borisov mutex_lock(&root->objectid_mutex); 5220ec7d6dfdSNikolay Borisov 52216b8fad57SNikolay Borisov if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) { 5222ec7d6dfdSNikolay Borisov btrfs_warn(root->fs_info, 5223ec7d6dfdSNikolay Borisov "the objectid of root %llu reaches its highest value", 5224ec7d6dfdSNikolay Borisov root->root_key.objectid); 5225ec7d6dfdSNikolay Borisov ret = -ENOSPC; 5226ec7d6dfdSNikolay Borisov goto out; 5227ec7d6dfdSNikolay Borisov } 5228ec7d6dfdSNikolay Borisov 522923125104SNikolay Borisov *objectid = root->free_objectid++; 5230ec7d6dfdSNikolay Borisov ret = 0; 5231ec7d6dfdSNikolay Borisov out: 5232ec7d6dfdSNikolay Borisov mutex_unlock(&root->objectid_mutex); 5233ec7d6dfdSNikolay Borisov return ret; 5234ec7d6dfdSNikolay Borisov } 5235