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 134570eb97bSJosef Bacik lock_extent(io_tree, eb->start, eb->start + eb->len - 1, &cached_state); 1350b32f4bbSJosef Bacik if (extent_buffer_uptodate(eb) && 1361259ab75SChris Mason btrfs_header_generation(eb) == parent_transid) { 1371259ab75SChris Mason ret = 0; 1381259ab75SChris Mason goto out; 1391259ab75SChris Mason } 14094647322SDavid Sterba btrfs_err_rl(eb->fs_info, 1418f0ed7d4SQu Wenruo "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu", 1428f0ed7d4SQu Wenruo eb->start, eb->read_mirror, 14329549aecSWang Shilong parent_transid, btrfs_header_generation(eb)); 1441259ab75SChris Mason ret = 1; 1450b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 14633958dc6SChris Mason out: 147570eb97bSJosef Bacik unlock_extent(io_tree, eb->start, eb->start + eb->len - 1, 148e43bbe5eSDavid Sterba &cached_state); 1491259ab75SChris Mason return ret; 1501259ab75SChris Mason } 1511259ab75SChris Mason 152e7e16f48SJohannes Thumshirn static bool btrfs_supported_super_csum(u16 csum_type) 153e7e16f48SJohannes Thumshirn { 154e7e16f48SJohannes Thumshirn switch (csum_type) { 155e7e16f48SJohannes Thumshirn case BTRFS_CSUM_TYPE_CRC32: 1563951e7f0SJohannes Thumshirn case BTRFS_CSUM_TYPE_XXHASH: 1573831bf00SJohannes Thumshirn case BTRFS_CSUM_TYPE_SHA256: 158352ae07bSDavid Sterba case BTRFS_CSUM_TYPE_BLAKE2: 159e7e16f48SJohannes Thumshirn return true; 160e7e16f48SJohannes Thumshirn default: 161e7e16f48SJohannes Thumshirn return false; 162e7e16f48SJohannes Thumshirn } 163e7e16f48SJohannes Thumshirn } 164e7e16f48SJohannes Thumshirn 165d352ac68SChris Mason /* 1661104a885SDavid Sterba * Return 0 if the superblock checksum type matches the checksum value of that 1671104a885SDavid Sterba * algorithm. Pass the raw disk superblock data. 1681104a885SDavid Sterba */ 1693d17adeaSQu Wenruo int btrfs_check_super_csum(struct btrfs_fs_info *fs_info, 1703d17adeaSQu Wenruo const struct btrfs_super_block *disk_sb) 1711104a885SDavid Sterba { 17251bce6c9SJohannes Thumshirn char result[BTRFS_CSUM_SIZE]; 173d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 174d5178578SJohannes Thumshirn 175d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 1761104a885SDavid Sterba 1771104a885SDavid Sterba /* 1781104a885SDavid Sterba * The super_block structure does not span the whole 17951bce6c9SJohannes Thumshirn * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is 18051bce6c9SJohannes Thumshirn * filled with zeros and is included in the checksum. 1811104a885SDavid Sterba */ 1823d17adeaSQu Wenruo crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE, 183fd08001fSEric Biggers BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result); 1841104a885SDavid Sterba 18555fc29beSDavid Sterba if (memcmp(disk_sb->csum, result, fs_info->csum_size)) 186e7e16f48SJohannes Thumshirn return 1; 1871104a885SDavid Sterba 188e7e16f48SJohannes Thumshirn return 0; 1891104a885SDavid Sterba } 1901104a885SDavid Sterba 191e064d5e9SDavid Sterba int btrfs_verify_level_key(struct extent_buffer *eb, int level, 192ff76a864SLiu Bo struct btrfs_key *first_key, u64 parent_transid) 193581c1760SQu Wenruo { 194e064d5e9SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 195581c1760SQu Wenruo int found_level; 196581c1760SQu Wenruo struct btrfs_key found_key; 197581c1760SQu Wenruo int ret; 198581c1760SQu Wenruo 199581c1760SQu Wenruo found_level = btrfs_header_level(eb); 200581c1760SQu Wenruo if (found_level != level) { 20163489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 20263489055SQu Wenruo KERN_ERR "BTRFS: tree level check failed\n"); 203581c1760SQu Wenruo btrfs_err(fs_info, 204581c1760SQu Wenruo "tree level mismatch detected, bytenr=%llu level expected=%u has=%u", 205581c1760SQu Wenruo eb->start, level, found_level); 206581c1760SQu Wenruo return -EIO; 207581c1760SQu Wenruo } 208581c1760SQu Wenruo 209581c1760SQu Wenruo if (!first_key) 210581c1760SQu Wenruo return 0; 211581c1760SQu Wenruo 2125d41be6fSQu Wenruo /* 2135d41be6fSQu Wenruo * For live tree block (new tree blocks in current transaction), 2145d41be6fSQu Wenruo * we need proper lock context to avoid race, which is impossible here. 2155d41be6fSQu Wenruo * So we only checks tree blocks which is read from disk, whose 2165d41be6fSQu Wenruo * generation <= fs_info->last_trans_committed. 2175d41be6fSQu Wenruo */ 2185d41be6fSQu Wenruo if (btrfs_header_generation(eb) > fs_info->last_trans_committed) 2195d41be6fSQu Wenruo return 0; 22062fdaa52SQu Wenruo 22162fdaa52SQu Wenruo /* We have @first_key, so this @eb must have at least one item */ 22262fdaa52SQu Wenruo if (btrfs_header_nritems(eb) == 0) { 22362fdaa52SQu Wenruo btrfs_err(fs_info, 22462fdaa52SQu Wenruo "invalid tree nritems, bytenr=%llu nritems=0 expect >0", 22562fdaa52SQu Wenruo eb->start); 22662fdaa52SQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 22762fdaa52SQu Wenruo return -EUCLEAN; 22862fdaa52SQu Wenruo } 22962fdaa52SQu Wenruo 230581c1760SQu Wenruo if (found_level) 231581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &found_key, 0); 232581c1760SQu Wenruo else 233581c1760SQu Wenruo btrfs_item_key_to_cpu(eb, &found_key, 0); 234581c1760SQu Wenruo ret = btrfs_comp_cpu_keys(first_key, &found_key); 235581c1760SQu Wenruo 236581c1760SQu Wenruo if (ret) { 23763489055SQu Wenruo WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 23863489055SQu Wenruo KERN_ERR "BTRFS: tree first key check failed\n"); 239581c1760SQu Wenruo btrfs_err(fs_info, 240ff76a864SLiu Bo "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", 241ff76a864SLiu Bo eb->start, parent_transid, first_key->objectid, 242ff76a864SLiu Bo first_key->type, first_key->offset, 243ff76a864SLiu Bo found_key.objectid, found_key.type, 244ff76a864SLiu Bo found_key.offset); 245581c1760SQu Wenruo } 246581c1760SQu Wenruo return ret; 247581c1760SQu Wenruo } 248581c1760SQu Wenruo 2491104a885SDavid Sterba /* 250d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 251d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 252581c1760SQu Wenruo * 253581c1760SQu Wenruo * @parent_transid: expected transid, skip check if 0 254581c1760SQu Wenruo * @level: expected level, mandatory check 255581c1760SQu Wenruo * @first_key: expected key of first slot, skip check if NULL 256d352ac68SChris Mason */ 2576a2e9dc4SFilipe Manana int btrfs_read_extent_buffer(struct extent_buffer *eb, 258581c1760SQu Wenruo u64 parent_transid, int level, 259581c1760SQu Wenruo struct btrfs_key *first_key) 260f188591eSChris Mason { 2615ab12d1fSDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 262f188591eSChris Mason struct extent_io_tree *io_tree; 263ea466794SJosef Bacik int failed = 0; 264f188591eSChris Mason int ret; 265f188591eSChris Mason int num_copies = 0; 266f188591eSChris Mason int mirror_num = 0; 267ea466794SJosef Bacik int failed_mirror = 0; 268f188591eSChris Mason 2690b246afaSJeff Mahoney io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; 270f188591eSChris Mason while (1) { 271f8397d69SNikolay Borisov clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 272c2ccfbc6SNikolay Borisov ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num); 273256dd1bbSStefan Behrens if (!ret) { 274581c1760SQu Wenruo if (verify_parent_transid(io_tree, eb, 275b9fab919SChris Mason parent_transid, 0)) 276256dd1bbSStefan Behrens ret = -EIO; 277e064d5e9SDavid Sterba else if (btrfs_verify_level_key(eb, level, 278ff76a864SLiu Bo first_key, parent_transid)) 279581c1760SQu Wenruo ret = -EUCLEAN; 280581c1760SQu Wenruo else 281581c1760SQu Wenruo break; 282256dd1bbSStefan Behrens } 283d397712bSChris Mason 2840b246afaSJeff Mahoney num_copies = btrfs_num_copies(fs_info, 285f188591eSChris Mason eb->start, eb->len); 2864235298eSChris Mason if (num_copies == 1) 287ea466794SJosef Bacik break; 2884235298eSChris Mason 2895cf1ab56SJosef Bacik if (!failed_mirror) { 2905cf1ab56SJosef Bacik failed = 1; 2915cf1ab56SJosef Bacik failed_mirror = eb->read_mirror; 2925cf1ab56SJosef Bacik } 2935cf1ab56SJosef Bacik 294f188591eSChris Mason mirror_num++; 295ea466794SJosef Bacik if (mirror_num == failed_mirror) 296ea466794SJosef Bacik mirror_num++; 297ea466794SJosef Bacik 2984235298eSChris Mason if (mirror_num > num_copies) 299ea466794SJosef Bacik break; 300f188591eSChris Mason } 301ea466794SJosef Bacik 302c0901581SStefan Behrens if (failed && !ret && failed_mirror) 30320a1fbf9SDavid Sterba btrfs_repair_eb_io_failure(eb, failed_mirror); 304ea466794SJosef Bacik 305ea466794SJosef Bacik return ret; 306f188591eSChris Mason } 30719c00ddcSChris Mason 308eca0f6f6SQu Wenruo static int csum_one_extent_buffer(struct extent_buffer *eb) 309eca0f6f6SQu Wenruo { 310eca0f6f6SQu Wenruo struct btrfs_fs_info *fs_info = eb->fs_info; 311eca0f6f6SQu Wenruo u8 result[BTRFS_CSUM_SIZE]; 312eca0f6f6SQu Wenruo int ret; 313eca0f6f6SQu Wenruo 314eca0f6f6SQu Wenruo ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid, 315eca0f6f6SQu Wenruo offsetof(struct btrfs_header, fsid), 316eca0f6f6SQu Wenruo BTRFS_FSID_SIZE) == 0); 317eca0f6f6SQu Wenruo csum_tree_block(eb, result); 318eca0f6f6SQu Wenruo 319eca0f6f6SQu Wenruo if (btrfs_header_level(eb)) 320eca0f6f6SQu Wenruo ret = btrfs_check_node(eb); 321eca0f6f6SQu Wenruo else 322eca0f6f6SQu Wenruo ret = btrfs_check_leaf_full(eb); 323eca0f6f6SQu Wenruo 3243777369fSQu Wenruo if (ret < 0) 3253777369fSQu Wenruo goto error; 3263777369fSQu Wenruo 3273777369fSQu Wenruo /* 3283777369fSQu Wenruo * Also check the generation, the eb reached here must be newer than 3293777369fSQu Wenruo * last committed. Or something seriously wrong happened. 3303777369fSQu Wenruo */ 3313777369fSQu Wenruo if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) { 3323777369fSQu Wenruo ret = -EUCLEAN; 333eca0f6f6SQu Wenruo btrfs_err(fs_info, 3343777369fSQu Wenruo "block=%llu bad generation, have %llu expect > %llu", 3353777369fSQu Wenruo eb->start, btrfs_header_generation(eb), 3363777369fSQu Wenruo fs_info->last_trans_committed); 3373777369fSQu Wenruo goto error; 338eca0f6f6SQu Wenruo } 339eca0f6f6SQu Wenruo write_extent_buffer(eb, result, 0, fs_info->csum_size); 340eca0f6f6SQu Wenruo 341eca0f6f6SQu Wenruo return 0; 3423777369fSQu Wenruo 3433777369fSQu Wenruo error: 3443777369fSQu Wenruo btrfs_print_tree(eb, 0); 3453777369fSQu Wenruo btrfs_err(fs_info, "block=%llu write time tree block corruption detected", 3463777369fSQu Wenruo eb->start); 3473777369fSQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 3483777369fSQu Wenruo return ret; 349eca0f6f6SQu Wenruo } 350eca0f6f6SQu Wenruo 351eca0f6f6SQu Wenruo /* Checksum all dirty extent buffers in one bio_vec */ 352eca0f6f6SQu Wenruo static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info, 353eca0f6f6SQu Wenruo struct bio_vec *bvec) 354eca0f6f6SQu Wenruo { 355eca0f6f6SQu Wenruo struct page *page = bvec->bv_page; 356eca0f6f6SQu Wenruo u64 bvec_start = page_offset(page) + bvec->bv_offset; 357eca0f6f6SQu Wenruo u64 cur; 358eca0f6f6SQu Wenruo int ret = 0; 359eca0f6f6SQu Wenruo 360eca0f6f6SQu Wenruo for (cur = bvec_start; cur < bvec_start + bvec->bv_len; 361eca0f6f6SQu Wenruo cur += fs_info->nodesize) { 362eca0f6f6SQu Wenruo struct extent_buffer *eb; 363eca0f6f6SQu Wenruo bool uptodate; 364eca0f6f6SQu Wenruo 365eca0f6f6SQu Wenruo eb = find_extent_buffer(fs_info, cur); 366eca0f6f6SQu Wenruo uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur, 367eca0f6f6SQu Wenruo fs_info->nodesize); 368eca0f6f6SQu Wenruo 36901cd3909SDavid Sterba /* A dirty eb shouldn't disappear from buffer_radix */ 370eca0f6f6SQu Wenruo if (WARN_ON(!eb)) 371eca0f6f6SQu Wenruo return -EUCLEAN; 372eca0f6f6SQu Wenruo 373eca0f6f6SQu Wenruo if (WARN_ON(cur != btrfs_header_bytenr(eb))) { 374eca0f6f6SQu Wenruo free_extent_buffer(eb); 375eca0f6f6SQu Wenruo return -EUCLEAN; 376eca0f6f6SQu Wenruo } 377eca0f6f6SQu Wenruo if (WARN_ON(!uptodate)) { 378eca0f6f6SQu Wenruo free_extent_buffer(eb); 379eca0f6f6SQu Wenruo return -EUCLEAN; 380eca0f6f6SQu Wenruo } 381eca0f6f6SQu Wenruo 382eca0f6f6SQu Wenruo ret = csum_one_extent_buffer(eb); 383eca0f6f6SQu Wenruo free_extent_buffer(eb); 384eca0f6f6SQu Wenruo if (ret < 0) 385eca0f6f6SQu Wenruo return ret; 386eca0f6f6SQu Wenruo } 387eca0f6f6SQu Wenruo return ret; 388eca0f6f6SQu Wenruo } 389eca0f6f6SQu Wenruo 390d352ac68SChris Mason /* 391ac303b69SQu Wenruo * Checksum a dirty tree block before IO. This has extra checks to make sure 392ac303b69SQu Wenruo * we only fill in the checksum field in the first page of a multi-page block. 393ac303b69SQu Wenruo * For subpage extent buffers we need bvec to also read the offset in the page. 394d352ac68SChris Mason */ 395ac303b69SQu Wenruo static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec) 39619c00ddcSChris Mason { 397ac303b69SQu Wenruo struct page *page = bvec->bv_page; 3984eee4fa4SMiao Xie u64 start = page_offset(page); 39919c00ddcSChris Mason u64 found_start; 40019c00ddcSChris Mason struct extent_buffer *eb; 401eca0f6f6SQu Wenruo 402fbca46ebSQu Wenruo if (fs_info->nodesize < PAGE_SIZE) 403eca0f6f6SQu Wenruo return csum_dirty_subpage_buffers(fs_info, bvec); 404f188591eSChris Mason 4054f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 4064f2de97aSJosef Bacik if (page != eb->pages[0]) 4074f2de97aSJosef Bacik return 0; 4080f805531SAlex Lyakas 40919c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 410d3575156SNaohiro Aota 411d3575156SNaohiro Aota if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) { 412d3575156SNaohiro Aota WARN_ON(found_start != 0); 413d3575156SNaohiro Aota return 0; 414d3575156SNaohiro Aota } 415d3575156SNaohiro Aota 4160f805531SAlex Lyakas /* 4170f805531SAlex Lyakas * Please do not consolidate these warnings into a single if. 4180f805531SAlex Lyakas * It is useful to know what went wrong. 4190f805531SAlex Lyakas */ 4200f805531SAlex Lyakas if (WARN_ON(found_start != start)) 4210f805531SAlex Lyakas return -EUCLEAN; 4220f805531SAlex Lyakas if (WARN_ON(!PageUptodate(page))) 4230f805531SAlex Lyakas return -EUCLEAN; 4240f805531SAlex Lyakas 425eca0f6f6SQu Wenruo return csum_one_extent_buffer(eb); 42619c00ddcSChris Mason } 42719c00ddcSChris Mason 428b0c9b3b0SDavid Sterba static int check_tree_block_fsid(struct extent_buffer *eb) 4292b82032cSYan Zheng { 430b0c9b3b0SDavid Sterba struct btrfs_fs_info *fs_info = eb->fs_info; 431944d3f9fSNikolay Borisov struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 43244880fdcSAnand Jain u8 fsid[BTRFS_FSID_SIZE]; 433944d3f9fSNikolay Borisov u8 *metadata_uuid; 4342b82032cSYan Zheng 4359a8658e3SDavid Sterba read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid), 4369a8658e3SDavid Sterba BTRFS_FSID_SIZE); 4377239ff4bSNikolay Borisov /* 438944d3f9fSNikolay Borisov * Checking the incompat flag is only valid for the current fs. For 439944d3f9fSNikolay Borisov * seed devices it's forbidden to have their uuid changed so reading 440944d3f9fSNikolay Borisov * ->fsid in this case is fine 4417239ff4bSNikolay Borisov */ 442944d3f9fSNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID)) 4437239ff4bSNikolay Borisov metadata_uuid = fs_devices->metadata_uuid; 4447239ff4bSNikolay Borisov else 4457239ff4bSNikolay Borisov metadata_uuid = fs_devices->fsid; 4467239ff4bSNikolay Borisov 447944d3f9fSNikolay Borisov if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) 448944d3f9fSNikolay Borisov return 0; 449944d3f9fSNikolay Borisov 450944d3f9fSNikolay Borisov list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) 451944d3f9fSNikolay Borisov if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE)) 452944d3f9fSNikolay Borisov return 0; 453944d3f9fSNikolay Borisov 454944d3f9fSNikolay Borisov return 1; 4552b82032cSYan Zheng } 4562b82032cSYan Zheng 45777bf40a2SQu Wenruo /* Do basic extent buffer checks at read time */ 45877bf40a2SQu Wenruo static int validate_extent_buffer(struct extent_buffer *eb) 459ce9adaa5SChris Mason { 46077bf40a2SQu Wenruo struct btrfs_fs_info *fs_info = eb->fs_info; 461ce9adaa5SChris Mason u64 found_start; 46277bf40a2SQu Wenruo const u32 csum_size = fs_info->csum_size; 46377bf40a2SQu Wenruo u8 found_level; 4642996e1f8SJohannes Thumshirn u8 result[BTRFS_CSUM_SIZE]; 465dfd29eedSDavid Sterba const u8 *header_csum; 46677bf40a2SQu Wenruo int ret = 0; 467ea466794SJosef Bacik 468ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 469727011e0SChris Mason if (found_start != eb->start) { 4708f0ed7d4SQu Wenruo btrfs_err_rl(fs_info, 4718f0ed7d4SQu Wenruo "bad tree block start, mirror %u want %llu have %llu", 4728f0ed7d4SQu Wenruo eb->read_mirror, eb->start, found_start); 473f188591eSChris Mason ret = -EIO; 47477bf40a2SQu Wenruo goto out; 475ce9adaa5SChris Mason } 476b0c9b3b0SDavid Sterba if (check_tree_block_fsid(eb)) { 4778f0ed7d4SQu Wenruo btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u", 4788f0ed7d4SQu Wenruo eb->start, eb->read_mirror); 4791259ab75SChris Mason ret = -EIO; 48077bf40a2SQu Wenruo goto out; 4811259ab75SChris Mason } 482ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 4831c24c3ceSJosef Bacik if (found_level >= BTRFS_MAX_LEVEL) { 4848f0ed7d4SQu Wenruo btrfs_err(fs_info, 4858f0ed7d4SQu Wenruo "bad tree block level, mirror %u level %d on logical %llu", 4868f0ed7d4SQu Wenruo eb->read_mirror, btrfs_header_level(eb), eb->start); 4871c24c3ceSJosef Bacik ret = -EIO; 48877bf40a2SQu Wenruo goto out; 4891c24c3ceSJosef Bacik } 490ce9adaa5SChris Mason 491c67b3892SDavid Sterba csum_tree_block(eb, result); 492dfd29eedSDavid Sterba header_csum = page_address(eb->pages[0]) + 493dfd29eedSDavid Sterba get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum)); 494a826d6dcSJosef Bacik 495dfd29eedSDavid Sterba if (memcmp(result, header_csum, csum_size) != 0) { 4962996e1f8SJohannes Thumshirn btrfs_warn_rl(fs_info, 4978f0ed7d4SQu Wenruo "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d", 4988f0ed7d4SQu Wenruo eb->start, eb->read_mirror, 499dfd29eedSDavid Sterba CSUM_FMT_VALUE(csum_size, header_csum), 50035be8851SJohannes Thumshirn CSUM_FMT_VALUE(csum_size, result), 50135be8851SJohannes Thumshirn btrfs_header_level(eb)); 5022996e1f8SJohannes Thumshirn ret = -EUCLEAN; 50377bf40a2SQu Wenruo goto out; 5042996e1f8SJohannes Thumshirn } 5052996e1f8SJohannes Thumshirn 506a826d6dcSJosef Bacik /* 507a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 508a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 509a826d6dcSJosef Bacik * return -EIO. 510a826d6dcSJosef Bacik */ 5111c4360eeSDavid Sterba if (found_level == 0 && btrfs_check_leaf_full(eb)) { 512a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 513a826d6dcSJosef Bacik ret = -EIO; 514a826d6dcSJosef Bacik } 515ce9adaa5SChris Mason 516813fd1dcSDavid Sterba if (found_level > 0 && btrfs_check_node(eb)) 517053ab70fSLiu Bo ret = -EIO; 518053ab70fSLiu Bo 5190b32f4bbSJosef Bacik if (!ret) 5200b32f4bbSJosef Bacik set_extent_buffer_uptodate(eb); 52175391f0dSQu Wenruo else 52275391f0dSQu Wenruo btrfs_err(fs_info, 5238f0ed7d4SQu Wenruo "read time tree block corruption detected on logical %llu mirror %u", 5248f0ed7d4SQu Wenruo eb->start, eb->read_mirror); 52577bf40a2SQu Wenruo out: 52677bf40a2SQu Wenruo return ret; 52777bf40a2SQu Wenruo } 52877bf40a2SQu Wenruo 529371cdc07SQu Wenruo static int validate_subpage_buffer(struct page *page, u64 start, u64 end, 530371cdc07SQu Wenruo int mirror) 531371cdc07SQu Wenruo { 532371cdc07SQu Wenruo struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); 533371cdc07SQu Wenruo struct extent_buffer *eb; 534371cdc07SQu Wenruo bool reads_done; 535371cdc07SQu Wenruo int ret = 0; 536371cdc07SQu Wenruo 537371cdc07SQu Wenruo /* 538371cdc07SQu Wenruo * We don't allow bio merge for subpage metadata read, so we should 539371cdc07SQu Wenruo * only get one eb for each endio hook. 540371cdc07SQu Wenruo */ 541371cdc07SQu Wenruo ASSERT(end == start + fs_info->nodesize - 1); 542371cdc07SQu Wenruo ASSERT(PagePrivate(page)); 543371cdc07SQu Wenruo 544371cdc07SQu Wenruo eb = find_extent_buffer(fs_info, start); 545371cdc07SQu Wenruo /* 546371cdc07SQu Wenruo * When we are reading one tree block, eb must have been inserted into 547371cdc07SQu Wenruo * the radix tree. If not, something is wrong. 548371cdc07SQu Wenruo */ 549371cdc07SQu Wenruo ASSERT(eb); 550371cdc07SQu Wenruo 551371cdc07SQu Wenruo reads_done = atomic_dec_and_test(&eb->io_pages); 552371cdc07SQu Wenruo /* Subpage read must finish in page read */ 553371cdc07SQu Wenruo ASSERT(reads_done); 554371cdc07SQu Wenruo 555371cdc07SQu Wenruo eb->read_mirror = mirror; 556371cdc07SQu Wenruo if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 557371cdc07SQu Wenruo ret = -EIO; 558371cdc07SQu Wenruo goto err; 559371cdc07SQu Wenruo } 560371cdc07SQu Wenruo ret = validate_extent_buffer(eb); 561371cdc07SQu Wenruo if (ret < 0) 562371cdc07SQu Wenruo goto err; 563371cdc07SQu Wenruo 564371cdc07SQu Wenruo set_extent_buffer_uptodate(eb); 565371cdc07SQu Wenruo 566371cdc07SQu Wenruo free_extent_buffer(eb); 567371cdc07SQu Wenruo return ret; 568371cdc07SQu Wenruo err: 569371cdc07SQu Wenruo /* 570371cdc07SQu Wenruo * end_bio_extent_readpage decrements io_pages in case of error, 571371cdc07SQu Wenruo * make sure it has something to decrement. 572371cdc07SQu Wenruo */ 573371cdc07SQu Wenruo atomic_inc(&eb->io_pages); 574371cdc07SQu Wenruo clear_extent_buffer_uptodate(eb); 575371cdc07SQu Wenruo free_extent_buffer(eb); 576371cdc07SQu Wenruo return ret; 577371cdc07SQu Wenruo } 578371cdc07SQu Wenruo 579c3a3b19bSQu Wenruo int btrfs_validate_metadata_buffer(struct btrfs_bio *bbio, 58077bf40a2SQu Wenruo struct page *page, u64 start, u64 end, 58177bf40a2SQu Wenruo int mirror) 58277bf40a2SQu Wenruo { 58377bf40a2SQu Wenruo struct extent_buffer *eb; 58477bf40a2SQu Wenruo int ret = 0; 58577bf40a2SQu Wenruo int reads_done; 58677bf40a2SQu Wenruo 58777bf40a2SQu Wenruo ASSERT(page->private); 588371cdc07SQu Wenruo 589fbca46ebSQu Wenruo if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE) 590371cdc07SQu Wenruo return validate_subpage_buffer(page, start, end, mirror); 591371cdc07SQu Wenruo 59277bf40a2SQu Wenruo eb = (struct extent_buffer *)page->private; 59377bf40a2SQu Wenruo 59477bf40a2SQu Wenruo /* 59577bf40a2SQu Wenruo * The pending IO might have been the only thing that kept this buffer 59677bf40a2SQu Wenruo * in memory. Make sure we have a ref for all this other checks 59777bf40a2SQu Wenruo */ 59877bf40a2SQu Wenruo atomic_inc(&eb->refs); 59977bf40a2SQu Wenruo 60077bf40a2SQu Wenruo reads_done = atomic_dec_and_test(&eb->io_pages); 60177bf40a2SQu Wenruo if (!reads_done) 60277bf40a2SQu Wenruo goto err; 60377bf40a2SQu Wenruo 60477bf40a2SQu Wenruo eb->read_mirror = mirror; 60577bf40a2SQu Wenruo if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 60677bf40a2SQu Wenruo ret = -EIO; 60777bf40a2SQu Wenruo goto err; 60877bf40a2SQu Wenruo } 60977bf40a2SQu Wenruo ret = validate_extent_buffer(eb); 610ce9adaa5SChris Mason err: 61153b381b3SDavid Woodhouse if (ret) { 61253b381b3SDavid Woodhouse /* 61353b381b3SDavid Woodhouse * our io error hook is going to dec the io pages 61453b381b3SDavid Woodhouse * again, we have to make sure it has something 61553b381b3SDavid Woodhouse * to decrement 61653b381b3SDavid Woodhouse */ 61753b381b3SDavid Woodhouse atomic_inc(&eb->io_pages); 6180b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 61953b381b3SDavid Woodhouse } 6200b32f4bbSJosef Bacik free_extent_buffer(eb); 62177bf40a2SQu Wenruo 622f188591eSChris Mason return ret; 623ce9adaa5SChris Mason } 624ce9adaa5SChris Mason 6254a69a410SChris Mason static void run_one_async_start(struct btrfs_work *work) 6264a69a410SChris Mason { 6274a69a410SChris Mason struct async_submit_bio *async; 6284e4cbee9SChristoph Hellwig blk_status_t ret; 6294a69a410SChris Mason 6304a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 6311941b64bSQu Wenruo ret = async->submit_bio_start(async->inode, async->bio, 6321941b64bSQu Wenruo async->dio_file_offset); 63379787eaaSJeff Mahoney if (ret) 6344e4cbee9SChristoph Hellwig async->status = ret; 6354a69a410SChris Mason } 6364a69a410SChris Mason 63706ea01b1SDavid Sterba /* 63806ea01b1SDavid Sterba * In order to insert checksums into the metadata in large chunks, we wait 63906ea01b1SDavid Sterba * until bio submission time. All the pages in the bio are checksummed and 64006ea01b1SDavid Sterba * sums are attached onto the ordered extent record. 64106ea01b1SDavid Sterba * 64206ea01b1SDavid Sterba * At IO completion time the csums attached on the ordered extent record are 64306ea01b1SDavid Sterba * inserted into the tree. 64406ea01b1SDavid Sterba */ 6454a69a410SChris Mason static void run_one_async_done(struct btrfs_work *work) 6468b712842SChris Mason { 647917f32a2SChristoph Hellwig struct async_submit_bio *async = 648917f32a2SChristoph Hellwig container_of(work, struct async_submit_bio, work); 649917f32a2SChristoph Hellwig struct inode *inode = async->inode; 650917f32a2SChristoph Hellwig struct btrfs_bio *bbio = btrfs_bio(async->bio); 6514854ddd0SChris Mason 652bb7ab3b9SAdam Buchbinder /* If an error occurred we just want to clean up the bio and move on */ 6534e4cbee9SChristoph Hellwig if (async->status) { 654917f32a2SChristoph Hellwig btrfs_bio_end_io(bbio, async->status); 65579787eaaSJeff Mahoney return; 65679787eaaSJeff Mahoney } 65779787eaaSJeff Mahoney 658ec39f769SChris Mason /* 659ec39f769SChris Mason * All of the bios that pass through here are from async helpers. 660ec39f769SChris Mason * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context. 661ec39f769SChris Mason * This changes nothing when cgroups aren't in use. 662ec39f769SChris Mason */ 663ec39f769SChris Mason async->bio->bi_opf |= REQ_CGROUP_PUNT; 6641a722d8fSChristoph Hellwig btrfs_submit_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num); 6654a69a410SChris Mason } 6664a69a410SChris Mason 6674a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 6684a69a410SChris Mason { 6694a69a410SChris Mason struct async_submit_bio *async; 6704a69a410SChris Mason 6714a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 6728b712842SChris Mason kfree(async); 6738b712842SChris Mason } 6748b712842SChris Mason 675ea1f0cedSChristoph Hellwig /* 676ea1f0cedSChristoph Hellwig * Submit bio to an async queue. 677ea1f0cedSChristoph Hellwig * 678ea1f0cedSChristoph Hellwig * Retrun: 679ea1f0cedSChristoph Hellwig * - true if the work has been succesfuly submitted 680ea1f0cedSChristoph Hellwig * - false in case of error 681ea1f0cedSChristoph Hellwig */ 682ea1f0cedSChristoph Hellwig bool btrfs_wq_submit_bio(struct inode *inode, struct bio *bio, int mirror_num, 683ea1f0cedSChristoph Hellwig u64 dio_file_offset, 684e288c080SDavid Sterba extent_submit_bio_start_t *submit_bio_start) 68544b8bd7eSChris Mason { 6868896a08dSQu Wenruo struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; 68744b8bd7eSChris Mason struct async_submit_bio *async; 68844b8bd7eSChris Mason 68944b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 69044b8bd7eSChris Mason if (!async) 691ea1f0cedSChristoph Hellwig return false; 69244b8bd7eSChris Mason 6938896a08dSQu Wenruo async->inode = inode; 69444b8bd7eSChris Mason async->bio = bio; 69544b8bd7eSChris Mason async->mirror_num = mirror_num; 6964a69a410SChris Mason async->submit_bio_start = submit_bio_start; 6974a69a410SChris Mason 698a0cac0ecSOmar Sandoval btrfs_init_work(&async->work, run_one_async_start, run_one_async_done, 699a0cac0ecSOmar Sandoval run_one_async_free); 7004a69a410SChris Mason 7011941b64bSQu Wenruo async->dio_file_offset = dio_file_offset; 7028c8bee1dSChris Mason 7034e4cbee9SChristoph Hellwig async->status = 0; 70479787eaaSJeff Mahoney 70567f055c7SChristoph Hellwig if (op_is_sync(bio->bi_opf)) 706a31b4a43SChristoph Hellwig btrfs_queue_work(fs_info->hipri_workers, &async->work); 707a31b4a43SChristoph Hellwig else 7085cdc7ad3SQu Wenruo btrfs_queue_work(fs_info->workers, &async->work); 709ea1f0cedSChristoph Hellwig return true; 71044b8bd7eSChris Mason } 71144b8bd7eSChris Mason 7124e4cbee9SChristoph Hellwig static blk_status_t btree_csum_one_bio(struct bio *bio) 713ce3ed71aSChris Mason { 7142c30c71bSKent Overstreet struct bio_vec *bvec; 715ce3ed71aSChris Mason struct btrfs_root *root; 7162b070cfeSChristoph Hellwig int ret = 0; 7176dc4f100SMing Lei struct bvec_iter_all iter_all; 718ce3ed71aSChris Mason 719c09abff8SDavid Sterba ASSERT(!bio_flagged(bio, BIO_CLONED)); 7202b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 721ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 722ac303b69SQu Wenruo ret = csum_dirty_buffer(root->fs_info, bvec); 72379787eaaSJeff Mahoney if (ret) 72479787eaaSJeff Mahoney break; 725ce3ed71aSChris Mason } 7262c30c71bSKent Overstreet 7274e4cbee9SChristoph Hellwig return errno_to_blk_status(ret); 728ce3ed71aSChris Mason } 729ce3ed71aSChris Mason 7308896a08dSQu Wenruo static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio, 7311941b64bSQu Wenruo u64 dio_file_offset) 73222c59948SChris Mason { 7338b712842SChris Mason /* 7348b712842SChris Mason * when we're called for a write, we're already in the async 7351a722d8fSChristoph Hellwig * submission context. Just jump into btrfs_submit_bio. 7368b712842SChris Mason */ 73779787eaaSJeff Mahoney return btree_csum_one_bio(bio); 73822c59948SChris Mason } 73922c59948SChris Mason 740f4dcfb30SJohannes Thumshirn static bool should_async_write(struct btrfs_fs_info *fs_info, 7419b4e675aSDavid Sterba struct btrfs_inode *bi) 742de0022b9SJosef Bacik { 7434eef29efSNaohiro Aota if (btrfs_is_zoned(fs_info)) 744f4dcfb30SJohannes Thumshirn return false; 7456300463bSLiu Bo if (atomic_read(&bi->sync_writers)) 746f4dcfb30SJohannes Thumshirn return false; 7479b4e675aSDavid Sterba if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags)) 748f4dcfb30SJohannes Thumshirn return false; 749f4dcfb30SJohannes Thumshirn return true; 750de0022b9SJosef Bacik } 751de0022b9SJosef Bacik 75294d9e11bSChristoph Hellwig void btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio, int mirror_num) 75344b8bd7eSChris Mason { 7540b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 755917f32a2SChristoph Hellwig struct btrfs_bio *bbio = btrfs_bio(bio); 7564e4cbee9SChristoph Hellwig blk_status_t ret; 757cad321adSChris Mason 75808a6f464SChristoph Hellwig bio->bi_opf |= REQ_META; 75908a6f464SChristoph Hellwig 760cfe94440SNaohiro Aota if (btrfs_op(bio) != BTRFS_MAP_WRITE) { 7611a722d8fSChristoph Hellwig btrfs_submit_bio(fs_info, bio, mirror_num); 7621a722d8fSChristoph Hellwig return; 76344b8bd7eSChris Mason } 76444b8bd7eSChris Mason 765ea1f0cedSChristoph Hellwig /* 766ea1f0cedSChristoph Hellwig * Kthread helpers are used to submit writes so that checksumming can 767ea1f0cedSChristoph Hellwig * happen in parallel across all CPUs. 768ea1f0cedSChristoph Hellwig */ 769ea1f0cedSChristoph Hellwig if (should_async_write(fs_info, BTRFS_I(inode)) && 770ea1f0cedSChristoph Hellwig btrfs_wq_submit_bio(inode, bio, mirror_num, 0, btree_submit_bio_start)) 771ea1f0cedSChristoph Hellwig return; 772ea1f0cedSChristoph Hellwig 773ea1f0cedSChristoph Hellwig ret = btree_csum_one_bio(bio); 77494d9e11bSChristoph Hellwig if (ret) { 775917f32a2SChristoph Hellwig btrfs_bio_end_io(bbio, ret); 776ea1f0cedSChristoph Hellwig return; 77794d9e11bSChristoph Hellwig } 778ea1f0cedSChristoph Hellwig 779ea1f0cedSChristoph Hellwig btrfs_submit_bio(fs_info, bio, mirror_num); 78061891923SStefan Behrens } 78161891923SStefan Behrens 7823dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 7838958b551SMatthew Wilcox (Oracle) static int btree_migrate_folio(struct address_space *mapping, 7848958b551SMatthew Wilcox (Oracle) struct folio *dst, struct folio *src, enum migrate_mode mode) 785784b4e29SChris Mason { 786784b4e29SChris Mason /* 787784b4e29SChris Mason * we can't safely write a btree page from here, 788784b4e29SChris Mason * we haven't done the locking hook 789784b4e29SChris Mason */ 7908958b551SMatthew Wilcox (Oracle) if (folio_test_dirty(src)) 791784b4e29SChris Mason return -EAGAIN; 792784b4e29SChris Mason /* 793784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 794784b4e29SChris Mason * We must have no buffers or drop them. 795784b4e29SChris Mason */ 7968958b551SMatthew Wilcox (Oracle) if (folio_get_private(src) && 7978958b551SMatthew Wilcox (Oracle) !filemap_release_folio(src, GFP_KERNEL)) 798784b4e29SChris Mason return -EAGAIN; 79954184650SMatthew Wilcox (Oracle) return migrate_folio(mapping, dst, src, mode); 800784b4e29SChris Mason } 8018958b551SMatthew Wilcox (Oracle) #else 8028958b551SMatthew Wilcox (Oracle) #define btree_migrate_folio NULL 8033dd1462eSJan Beulich #endif 804784b4e29SChris Mason 8050da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 8060da5468fSChris Mason struct writeback_control *wbc) 8070da5468fSChris Mason { 808e2d84521SMiao Xie struct btrfs_fs_info *fs_info; 809e2d84521SMiao Xie int ret; 810e2d84521SMiao Xie 811d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 812448d640bSChris Mason 813448d640bSChris Mason if (wbc->for_kupdate) 814448d640bSChris Mason return 0; 815448d640bSChris Mason 816e2d84521SMiao Xie fs_info = BTRFS_I(mapping->host)->root->fs_info; 817b9473439SChris Mason /* this is a bit racy, but that's ok */ 818d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 819d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 820d814a491SEthan Lien fs_info->dirty_metadata_batch); 821e2d84521SMiao Xie if (ret < 0) 822793955bcSChris Mason return 0; 823793955bcSChris Mason } 8240b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 8250da5468fSChris Mason } 8260da5468fSChris Mason 827f913cff3SMatthew Wilcox (Oracle) static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags) 8285f39d397SChris Mason { 829f913cff3SMatthew Wilcox (Oracle) if (folio_test_writeback(folio) || folio_test_dirty(folio)) 830f913cff3SMatthew Wilcox (Oracle) return false; 8310c4e538bSDavid Sterba 832f913cff3SMatthew Wilcox (Oracle) return try_release_extent_buffer(&folio->page); 833d98237b3SChris Mason } 834d98237b3SChris Mason 835895586ebSMatthew Wilcox (Oracle) static void btree_invalidate_folio(struct folio *folio, size_t offset, 836895586ebSMatthew Wilcox (Oracle) size_t length) 837d98237b3SChris Mason { 838d1310b2eSChris Mason struct extent_io_tree *tree; 839895586ebSMatthew Wilcox (Oracle) tree = &BTRFS_I(folio->mapping->host)->io_tree; 840895586ebSMatthew Wilcox (Oracle) extent_invalidate_folio(tree, folio, offset); 841f913cff3SMatthew Wilcox (Oracle) btree_release_folio(folio, GFP_NOFS); 842895586ebSMatthew Wilcox (Oracle) if (folio_get_private(folio)) { 843895586ebSMatthew Wilcox (Oracle) btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info, 844895586ebSMatthew Wilcox (Oracle) "folio private not zero on folio %llu", 845895586ebSMatthew Wilcox (Oracle) (unsigned long long)folio_pos(folio)); 846895586ebSMatthew Wilcox (Oracle) folio_detach_private(folio); 8479ad6b7bcSChris Mason } 848d98237b3SChris Mason } 849d98237b3SChris Mason 850bb146eb2SJosef Bacik #ifdef DEBUG 8510079c3b1SMatthew Wilcox (Oracle) static bool btree_dirty_folio(struct address_space *mapping, 8520079c3b1SMatthew Wilcox (Oracle) struct folio *folio) 8530079c3b1SMatthew Wilcox (Oracle) { 8540079c3b1SMatthew Wilcox (Oracle) struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb); 855139e8cd3SQu Wenruo struct btrfs_subpage *subpage; 8560b32f4bbSJosef Bacik struct extent_buffer *eb; 857139e8cd3SQu Wenruo int cur_bit = 0; 8580079c3b1SMatthew Wilcox (Oracle) u64 page_start = folio_pos(folio); 8590b32f4bbSJosef Bacik 860139e8cd3SQu Wenruo if (fs_info->sectorsize == PAGE_SIZE) { 8610079c3b1SMatthew Wilcox (Oracle) eb = folio_get_private(folio); 8620b32f4bbSJosef Bacik BUG_ON(!eb); 8630b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 8640b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 86549d0c642SFilipe Manana btrfs_assert_tree_write_locked(eb); 8660079c3b1SMatthew Wilcox (Oracle) return filemap_dirty_folio(mapping, folio); 867139e8cd3SQu Wenruo } 8680079c3b1SMatthew Wilcox (Oracle) subpage = folio_get_private(folio); 869139e8cd3SQu Wenruo 870139e8cd3SQu Wenruo ASSERT(subpage->dirty_bitmap); 871139e8cd3SQu Wenruo while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) { 872139e8cd3SQu Wenruo unsigned long flags; 873139e8cd3SQu Wenruo u64 cur; 874139e8cd3SQu Wenruo u16 tmp = (1 << cur_bit); 875139e8cd3SQu Wenruo 876139e8cd3SQu Wenruo spin_lock_irqsave(&subpage->lock, flags); 877139e8cd3SQu Wenruo if (!(tmp & subpage->dirty_bitmap)) { 878139e8cd3SQu Wenruo spin_unlock_irqrestore(&subpage->lock, flags); 879139e8cd3SQu Wenruo cur_bit++; 880139e8cd3SQu Wenruo continue; 881139e8cd3SQu Wenruo } 882139e8cd3SQu Wenruo spin_unlock_irqrestore(&subpage->lock, flags); 883139e8cd3SQu Wenruo cur = page_start + cur_bit * fs_info->sectorsize; 884139e8cd3SQu Wenruo 885139e8cd3SQu Wenruo eb = find_extent_buffer(fs_info, cur); 886139e8cd3SQu Wenruo ASSERT(eb); 887139e8cd3SQu Wenruo ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 888139e8cd3SQu Wenruo ASSERT(atomic_read(&eb->refs)); 88949d0c642SFilipe Manana btrfs_assert_tree_write_locked(eb); 890139e8cd3SQu Wenruo free_extent_buffer(eb); 891139e8cd3SQu Wenruo 892139e8cd3SQu Wenruo cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits); 893139e8cd3SQu Wenruo } 8940079c3b1SMatthew Wilcox (Oracle) return filemap_dirty_folio(mapping, folio); 8950b32f4bbSJosef Bacik } 8960079c3b1SMatthew Wilcox (Oracle) #else 8970079c3b1SMatthew Wilcox (Oracle) #define btree_dirty_folio filemap_dirty_folio 8980079c3b1SMatthew Wilcox (Oracle) #endif 8990b32f4bbSJosef Bacik 9007f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 9010da5468fSChris Mason .writepages = btree_writepages, 902f913cff3SMatthew Wilcox (Oracle) .release_folio = btree_release_folio, 903895586ebSMatthew Wilcox (Oracle) .invalidate_folio = btree_invalidate_folio, 9048958b551SMatthew Wilcox (Oracle) .migrate_folio = btree_migrate_folio, 9050079c3b1SMatthew Wilcox (Oracle) .dirty_folio = btree_dirty_folio, 906d98237b3SChris Mason }; 907123abc88SChris Mason 9082ff7e61eSJeff Mahoney struct extent_buffer *btrfs_find_create_tree_block( 9092ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info, 9103fbaf258SJosef Bacik u64 bytenr, u64 owner_root, 9113fbaf258SJosef Bacik int level) 9120999df54SChris Mason { 9130b246afaSJeff Mahoney if (btrfs_is_testing(fs_info)) 9140b246afaSJeff Mahoney return alloc_test_extent_buffer(fs_info, bytenr); 9153fbaf258SJosef Bacik return alloc_extent_buffer(fs_info, bytenr, owner_root, level); 9160999df54SChris Mason } 9170999df54SChris Mason 918581c1760SQu Wenruo /* 919581c1760SQu Wenruo * Read tree block at logical address @bytenr and do variant basic but critical 920581c1760SQu Wenruo * verification. 921581c1760SQu Wenruo * 9221b7ec85eSJosef Bacik * @owner_root: the objectid of the root owner for this block. 923581c1760SQu Wenruo * @parent_transid: expected transid of this tree block, skip check if 0 924581c1760SQu Wenruo * @level: expected level, mandatory check 925581c1760SQu Wenruo * @first_key: expected key in slot 0, skip check if NULL 926581c1760SQu Wenruo */ 9272ff7e61eSJeff Mahoney struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, 9281b7ec85eSJosef Bacik u64 owner_root, u64 parent_transid, 9291b7ec85eSJosef Bacik int level, struct btrfs_key *first_key) 930e20d96d6SChris Mason { 9315f39d397SChris Mason struct extent_buffer *buf = NULL; 93219c00ddcSChris Mason int ret; 93319c00ddcSChris Mason 9343fbaf258SJosef Bacik buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level); 935c871b0f2SLiu Bo if (IS_ERR(buf)) 936c871b0f2SLiu Bo return buf; 937e4204dedSChris Mason 9386a2e9dc4SFilipe Manana ret = btrfs_read_extent_buffer(buf, parent_transid, level, first_key); 9390f0fe8f7SFilipe David Borba Manana if (ret) { 940537f38f0SNikolay Borisov free_extent_buffer_stale(buf); 94164c043deSLiu Bo return ERR_PTR(ret); 9420f0fe8f7SFilipe David Borba Manana } 94388c602abSQu Wenruo if (btrfs_check_eb_owner(buf, owner_root)) { 94488c602abSQu Wenruo free_extent_buffer_stale(buf); 94588c602abSQu Wenruo return ERR_PTR(-EUCLEAN); 94688c602abSQu Wenruo } 9475f39d397SChris Mason return buf; 948ce9adaa5SChris Mason 949eb60ceacSChris Mason } 950eb60ceacSChris Mason 9516a884d7dSDavid Sterba void btrfs_clean_tree_block(struct extent_buffer *buf) 952ed2ff2cbSChris Mason { 9536a884d7dSDavid Sterba struct btrfs_fs_info *fs_info = buf->fs_info; 95455c69072SChris Mason if (btrfs_header_generation(buf) == 955e2d84521SMiao Xie fs_info->running_transaction->transid) { 95649d0c642SFilipe Manana btrfs_assert_tree_write_locked(buf); 957b4ce94deSChris Mason 958b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 959104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 960e2d84521SMiao Xie -buf->len, 961e2d84521SMiao Xie fs_info->dirty_metadata_batch); 9620b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 963925baeddSChris Mason } 9645f39d397SChris Mason } 965ed7b63ebSJosef Bacik } 9665f39d397SChris Mason 967da17066cSJeff Mahoney static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, 968e20d96d6SChris Mason u64 objectid) 969d97e63b6SChris Mason { 9707c0260eeSJeff Mahoney bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); 9712e608bd1SJosef Bacik 9722e608bd1SJosef Bacik memset(&root->root_key, 0, sizeof(root->root_key)); 9732e608bd1SJosef Bacik memset(&root->root_item, 0, sizeof(root->root_item)); 9742e608bd1SJosef Bacik memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 97596dfcb46SJosef Bacik root->fs_info = fs_info; 9762e608bd1SJosef Bacik root->root_key.objectid = objectid; 977cfaa7295SChris Mason root->node = NULL; 978a28ec197SChris Mason root->commit_root = NULL; 97927cdeb70SMiao Xie root->state = 0; 980abed4aaaSJosef Bacik RB_CLEAR_NODE(&root->rb_node); 9810b86a832SChris Mason 9820f7d52f4SChris Mason root->last_trans = 0; 9836b8fad57SNikolay Borisov root->free_objectid = 0; 984eb73c1b7SMiao Xie root->nr_delalloc_inodes = 0; 985199c2a9cSMiao Xie root->nr_ordered_extents = 0; 9866bef4d31SEric Paris root->inode_tree = RB_ROOT; 987088aea3bSDavid Sterba INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 9882e608bd1SJosef Bacik 9892e608bd1SJosef Bacik btrfs_init_root_block_rsv(root); 9900b86a832SChris Mason 9910b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 9925d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 993eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_inodes); 994eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_root); 995199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_extents); 996199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_root); 997d2311e69SQu Wenruo INIT_LIST_HEAD(&root->reloc_dirty_list); 9982ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 9992ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 10005d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1001eb73c1b7SMiao Xie spin_lock_init(&root->delalloc_lock); 1002199c2a9cSMiao Xie spin_lock_init(&root->ordered_extent_lock); 1003f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 10042ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 10052ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 10068287475aSQu Wenruo spin_lock_init(&root->qgroup_meta_rsv_lock); 1007a2135011SChris Mason mutex_init(&root->objectid_mutex); 1008e02119d5SChris Mason mutex_init(&root->log_mutex); 100931f3d255SMiao Xie mutex_init(&root->ordered_extent_mutex); 1010573bfb72SMiao Xie mutex_init(&root->delalloc_mutex); 1011c53e9653SQu Wenruo init_waitqueue_head(&root->qgroup_flush_wait); 10127237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 10137237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 10147237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 10158b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[0]); 10168b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[1]); 10177237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 10187237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 10197237f183SYan Zheng atomic_set(&root->log_writers, 0); 10202ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 10210700cea7SElena Reshetova refcount_set(&root->refs, 1); 10228ecebf4dSRobbie Ko atomic_set(&root->snapshot_force_cow, 0); 1023eede2bf3SOmar Sandoval atomic_set(&root->nr_swapfiles, 0); 10247237f183SYan Zheng root->log_transid = 0; 1025d1433debSMiao Xie root->log_transid_committed = -1; 1026257c62e1SChris Mason root->last_log_commit = 0; 10272e608bd1SJosef Bacik root->anon_dev = 0; 1028e289f03eSFilipe Manana if (!dummy) { 102943eb5f29SQu Wenruo extent_io_tree_init(fs_info, &root->dirty_log_pages, 103043eb5f29SQu Wenruo IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL); 1031e289f03eSFilipe Manana extent_io_tree_init(fs_info, &root->log_csum_range, 1032e289f03eSFilipe Manana IO_TREE_LOG_CSUM_RANGE, NULL); 1033e289f03eSFilipe Manana } 1034017e5369SChris Mason 10355f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 1036370a11b8SQu Wenruo btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks); 1037bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 1038bd647ce3SJosef Bacik INIT_LIST_HEAD(&root->leak_list); 1039fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1040bd647ce3SJosef Bacik list_add_tail(&root->leak_list, &fs_info->allocated_roots); 1041fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 1042bd647ce3SJosef Bacik #endif 10433768f368SChris Mason } 10443768f368SChris Mason 104574e4d827SDavid Sterba static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, 104696dfcb46SJosef Bacik u64 objectid, gfp_t flags) 10476f07e42eSAl Viro { 104874e4d827SDavid Sterba struct btrfs_root *root = kzalloc(sizeof(*root), flags); 10496f07e42eSAl Viro if (root) 105096dfcb46SJosef Bacik __setup_root(root, fs_info, objectid); 10516f07e42eSAl Viro return root; 10526f07e42eSAl Viro } 10536f07e42eSAl Viro 105406ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 105506ea65a3SJosef Bacik /* Should only be used by the testing infrastructure */ 1056da17066cSJeff Mahoney struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) 105706ea65a3SJosef Bacik { 105806ea65a3SJosef Bacik struct btrfs_root *root; 105906ea65a3SJosef Bacik 10607c0260eeSJeff Mahoney if (!fs_info) 10617c0260eeSJeff Mahoney return ERR_PTR(-EINVAL); 10627c0260eeSJeff Mahoney 106396dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL); 106406ea65a3SJosef Bacik if (!root) 106506ea65a3SJosef Bacik return ERR_PTR(-ENOMEM); 1066da17066cSJeff Mahoney 1067b9ef22deSFeifei Xu /* We don't use the stripesize in selftest, set it as sectorsize */ 1068faa2dbf0SJosef Bacik root->alloc_bytenr = 0; 106906ea65a3SJosef Bacik 107006ea65a3SJosef Bacik return root; 107106ea65a3SJosef Bacik } 107206ea65a3SJosef Bacik #endif 107306ea65a3SJosef Bacik 1074abed4aaaSJosef Bacik static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node) 1075abed4aaaSJosef Bacik { 1076abed4aaaSJosef Bacik const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node); 1077abed4aaaSJosef Bacik const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node); 1078abed4aaaSJosef Bacik 1079abed4aaaSJosef Bacik return btrfs_comp_cpu_keys(&a->root_key, &b->root_key); 1080abed4aaaSJosef Bacik } 1081abed4aaaSJosef Bacik 1082abed4aaaSJosef Bacik static int global_root_key_cmp(const void *k, const struct rb_node *node) 1083abed4aaaSJosef Bacik { 1084abed4aaaSJosef Bacik const struct btrfs_key *key = k; 1085abed4aaaSJosef Bacik const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node); 1086abed4aaaSJosef Bacik 1087abed4aaaSJosef Bacik return btrfs_comp_cpu_keys(key, &root->root_key); 1088abed4aaaSJosef Bacik } 1089abed4aaaSJosef Bacik 1090abed4aaaSJosef Bacik int btrfs_global_root_insert(struct btrfs_root *root) 1091abed4aaaSJosef Bacik { 1092abed4aaaSJosef Bacik struct btrfs_fs_info *fs_info = root->fs_info; 1093abed4aaaSJosef Bacik struct rb_node *tmp; 1094abed4aaaSJosef Bacik 1095abed4aaaSJosef Bacik write_lock(&fs_info->global_root_lock); 1096abed4aaaSJosef Bacik tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp); 1097abed4aaaSJosef Bacik write_unlock(&fs_info->global_root_lock); 1098abed4aaaSJosef Bacik ASSERT(!tmp); 1099abed4aaaSJosef Bacik 1100abed4aaaSJosef Bacik return tmp ? -EEXIST : 0; 1101abed4aaaSJosef Bacik } 1102abed4aaaSJosef Bacik 1103abed4aaaSJosef Bacik void btrfs_global_root_delete(struct btrfs_root *root) 1104abed4aaaSJosef Bacik { 1105abed4aaaSJosef Bacik struct btrfs_fs_info *fs_info = root->fs_info; 1106abed4aaaSJosef Bacik 1107abed4aaaSJosef Bacik write_lock(&fs_info->global_root_lock); 1108abed4aaaSJosef Bacik rb_erase(&root->rb_node, &fs_info->global_root_tree); 1109abed4aaaSJosef Bacik write_unlock(&fs_info->global_root_lock); 1110abed4aaaSJosef Bacik } 1111abed4aaaSJosef Bacik 1112abed4aaaSJosef Bacik struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info, 1113abed4aaaSJosef Bacik struct btrfs_key *key) 1114abed4aaaSJosef Bacik { 1115abed4aaaSJosef Bacik struct rb_node *node; 1116abed4aaaSJosef Bacik struct btrfs_root *root = NULL; 1117abed4aaaSJosef Bacik 1118abed4aaaSJosef Bacik read_lock(&fs_info->global_root_lock); 1119abed4aaaSJosef Bacik node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp); 1120abed4aaaSJosef Bacik if (node) 1121abed4aaaSJosef Bacik root = container_of(node, struct btrfs_root, rb_node); 1122abed4aaaSJosef Bacik read_unlock(&fs_info->global_root_lock); 1123abed4aaaSJosef Bacik 1124abed4aaaSJosef Bacik return root; 1125abed4aaaSJosef Bacik } 1126abed4aaaSJosef Bacik 1127f7238e50SJosef Bacik static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr) 1128f7238e50SJosef Bacik { 1129f7238e50SJosef Bacik struct btrfs_block_group *block_group; 1130f7238e50SJosef Bacik u64 ret; 1131f7238e50SJosef Bacik 1132f7238e50SJosef Bacik if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) 1133f7238e50SJosef Bacik return 0; 1134f7238e50SJosef Bacik 1135f7238e50SJosef Bacik if (bytenr) 1136f7238e50SJosef Bacik block_group = btrfs_lookup_block_group(fs_info, bytenr); 1137f7238e50SJosef Bacik else 1138f7238e50SJosef Bacik block_group = btrfs_lookup_first_block_group(fs_info, bytenr); 1139f7238e50SJosef Bacik ASSERT(block_group); 1140f7238e50SJosef Bacik if (!block_group) 1141f7238e50SJosef Bacik return 0; 1142f7238e50SJosef Bacik ret = block_group->global_root_id; 1143f7238e50SJosef Bacik btrfs_put_block_group(block_group); 1144f7238e50SJosef Bacik 1145f7238e50SJosef Bacik return ret; 1146f7238e50SJosef Bacik } 1147f7238e50SJosef Bacik 1148abed4aaaSJosef Bacik struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr) 1149abed4aaaSJosef Bacik { 1150abed4aaaSJosef Bacik struct btrfs_key key = { 1151abed4aaaSJosef Bacik .objectid = BTRFS_CSUM_TREE_OBJECTID, 1152abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 1153f7238e50SJosef Bacik .offset = btrfs_global_root_id(fs_info, bytenr), 1154abed4aaaSJosef Bacik }; 1155abed4aaaSJosef Bacik 1156abed4aaaSJosef Bacik return btrfs_global_root(fs_info, &key); 1157abed4aaaSJosef Bacik } 1158abed4aaaSJosef Bacik 1159abed4aaaSJosef Bacik struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr) 1160abed4aaaSJosef Bacik { 1161abed4aaaSJosef Bacik struct btrfs_key key = { 1162abed4aaaSJosef Bacik .objectid = BTRFS_EXTENT_TREE_OBJECTID, 1163abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 1164f7238e50SJosef Bacik .offset = btrfs_global_root_id(fs_info, bytenr), 1165abed4aaaSJosef Bacik }; 1166abed4aaaSJosef Bacik 1167abed4aaaSJosef Bacik return btrfs_global_root(fs_info, &key); 1168abed4aaaSJosef Bacik } 1169abed4aaaSJosef Bacik 117020897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 117120897f5cSArne Jansen u64 objectid) 117220897f5cSArne Jansen { 11739b7a2440SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 117420897f5cSArne Jansen struct extent_buffer *leaf; 117520897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 117620897f5cSArne Jansen struct btrfs_root *root; 117720897f5cSArne Jansen struct btrfs_key key; 1178b89f6d1fSFilipe Manana unsigned int nofs_flag; 117920897f5cSArne Jansen int ret = 0; 118020897f5cSArne Jansen 1181b89f6d1fSFilipe Manana /* 1182b89f6d1fSFilipe Manana * We're holding a transaction handle, so use a NOFS memory allocation 1183b89f6d1fSFilipe Manana * context to avoid deadlock if reclaim happens. 1184b89f6d1fSFilipe Manana */ 1185b89f6d1fSFilipe Manana nofs_flag = memalloc_nofs_save(); 118696dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL); 1187b89f6d1fSFilipe Manana memalloc_nofs_restore(nofs_flag); 118820897f5cSArne Jansen if (!root) 118920897f5cSArne Jansen return ERR_PTR(-ENOMEM); 119020897f5cSArne Jansen 119120897f5cSArne Jansen root->root_key.objectid = objectid; 119220897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 119320897f5cSArne Jansen root->root_key.offset = 0; 119420897f5cSArne Jansen 11959631e4ccSJosef Bacik leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0, 11969631e4ccSJosef Bacik BTRFS_NESTING_NORMAL); 119720897f5cSArne Jansen if (IS_ERR(leaf)) { 119820897f5cSArne Jansen ret = PTR_ERR(leaf); 11991dd05682STsutomu Itoh leaf = NULL; 12008a6a87cdSBoris Burkov goto fail_unlock; 120120897f5cSArne Jansen } 120220897f5cSArne Jansen 120320897f5cSArne Jansen root->node = leaf; 120420897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 120520897f5cSArne Jansen 120620897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 120727cdeb70SMiao Xie set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 120820897f5cSArne Jansen 1209f944d2cbSDavid Sterba btrfs_set_root_flags(&root->root_item, 0); 1210f944d2cbSDavid Sterba btrfs_set_root_limit(&root->root_item, 0); 121120897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 121220897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 121320897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 121420897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 121520897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 121620897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 121720897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 121833d85fdaSQu Wenruo if (is_fstree(objectid)) 1219807fc790SAndy Shevchenko generate_random_guid(root->root_item.uuid); 1220807fc790SAndy Shevchenko else 1221807fc790SAndy Shevchenko export_guid(root->root_item.uuid, &guid_null); 1222c8422684SDavid Sterba btrfs_set_root_drop_level(&root->root_item, 0); 122320897f5cSArne Jansen 12248a6a87cdSBoris Burkov btrfs_tree_unlock(leaf); 12258a6a87cdSBoris Burkov 122620897f5cSArne Jansen key.objectid = objectid; 122720897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 122820897f5cSArne Jansen key.offset = 0; 122920897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 123020897f5cSArne Jansen if (ret) 123120897f5cSArne Jansen goto fail; 123220897f5cSArne Jansen 123320897f5cSArne Jansen return root; 12341dd05682STsutomu Itoh 12358a6a87cdSBoris Burkov fail_unlock: 12368c38938cSJosef Bacik if (leaf) 12371dd05682STsutomu Itoh btrfs_tree_unlock(leaf); 12388a6a87cdSBoris Burkov fail: 123900246528SJosef Bacik btrfs_put_root(root); 12401dd05682STsutomu Itoh 12411dd05682STsutomu Itoh return ERR_PTR(ret); 124220897f5cSArne Jansen } 124320897f5cSArne Jansen 12447237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1245e02119d5SChris Mason struct btrfs_fs_info *fs_info) 12460f7d52f4SChris Mason { 12470f7d52f4SChris Mason struct btrfs_root *root; 1248e02119d5SChris Mason 124996dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS); 1250e02119d5SChris Mason if (!root) 12517237f183SYan Zheng return ERR_PTR(-ENOMEM); 1252e02119d5SChris Mason 1253e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1254e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1255e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 125627cdeb70SMiao Xie 12576ab6ebb7SNaohiro Aota return root; 12586ab6ebb7SNaohiro Aota } 12596ab6ebb7SNaohiro Aota 12606ab6ebb7SNaohiro Aota int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans, 12616ab6ebb7SNaohiro Aota struct btrfs_root *root) 12626ab6ebb7SNaohiro Aota { 12636ab6ebb7SNaohiro Aota struct extent_buffer *leaf; 12646ab6ebb7SNaohiro Aota 12657237f183SYan Zheng /* 126692a7cc42SQu Wenruo * DON'T set SHAREABLE bit for log trees. 126727cdeb70SMiao Xie * 126892a7cc42SQu Wenruo * Log trees are not exposed to user space thus can't be snapshotted, 126992a7cc42SQu Wenruo * and they go away before a real commit is actually done. 127092a7cc42SQu Wenruo * 127192a7cc42SQu Wenruo * They do store pointers to file data extents, and those reference 127292a7cc42SQu Wenruo * counts still get updated (along with back refs to the log tree). 12737237f183SYan Zheng */ 1274e02119d5SChris Mason 12754d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 12769631e4ccSJosef Bacik NULL, 0, 0, 0, BTRFS_NESTING_NORMAL); 12776ab6ebb7SNaohiro Aota if (IS_ERR(leaf)) 12786ab6ebb7SNaohiro Aota return PTR_ERR(leaf); 1279e02119d5SChris Mason 12807237f183SYan Zheng root->node = leaf; 1281e02119d5SChris Mason 1282e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1283e02119d5SChris Mason btrfs_tree_unlock(root->node); 12846ab6ebb7SNaohiro Aota 12856ab6ebb7SNaohiro Aota return 0; 12867237f183SYan Zheng } 12877237f183SYan Zheng 12887237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 12897237f183SYan Zheng struct btrfs_fs_info *fs_info) 12907237f183SYan Zheng { 12917237f183SYan Zheng struct btrfs_root *log_root; 12927237f183SYan Zheng 12937237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 12947237f183SYan Zheng if (IS_ERR(log_root)) 12957237f183SYan Zheng return PTR_ERR(log_root); 12966ab6ebb7SNaohiro Aota 12973ddebf27SNaohiro Aota if (!btrfs_is_zoned(fs_info)) { 12983ddebf27SNaohiro Aota int ret = btrfs_alloc_log_tree_node(trans, log_root); 12993ddebf27SNaohiro Aota 13006ab6ebb7SNaohiro Aota if (ret) { 13016ab6ebb7SNaohiro Aota btrfs_put_root(log_root); 13026ab6ebb7SNaohiro Aota return ret; 13036ab6ebb7SNaohiro Aota } 13043ddebf27SNaohiro Aota } 13056ab6ebb7SNaohiro Aota 13067237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 13077237f183SYan Zheng fs_info->log_root_tree = log_root; 13087237f183SYan Zheng return 0; 13097237f183SYan Zheng } 13107237f183SYan Zheng 13117237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 13127237f183SYan Zheng struct btrfs_root *root) 13137237f183SYan Zheng { 13140b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13157237f183SYan Zheng struct btrfs_root *log_root; 13167237f183SYan Zheng struct btrfs_inode_item *inode_item; 13176ab6ebb7SNaohiro Aota int ret; 13187237f183SYan Zheng 13190b246afaSJeff Mahoney log_root = alloc_log_tree(trans, fs_info); 13207237f183SYan Zheng if (IS_ERR(log_root)) 13217237f183SYan Zheng return PTR_ERR(log_root); 13227237f183SYan Zheng 13236ab6ebb7SNaohiro Aota ret = btrfs_alloc_log_tree_node(trans, log_root); 13246ab6ebb7SNaohiro Aota if (ret) { 13256ab6ebb7SNaohiro Aota btrfs_put_root(log_root); 13266ab6ebb7SNaohiro Aota return ret; 13276ab6ebb7SNaohiro Aota } 13286ab6ebb7SNaohiro Aota 13297237f183SYan Zheng log_root->last_trans = trans->transid; 13307237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 13317237f183SYan Zheng 13327237f183SYan Zheng inode_item = &log_root->root_item.inode; 13333cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 13343cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 13353cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 1336da17066cSJeff Mahoney btrfs_set_stack_inode_nbytes(inode_item, 13370b246afaSJeff Mahoney fs_info->nodesize); 13383cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 13397237f183SYan Zheng 13405d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 13417237f183SYan Zheng 13427237f183SYan Zheng WARN_ON(root->log_root); 13437237f183SYan Zheng root->log_root = log_root; 13447237f183SYan Zheng root->log_transid = 0; 1345d1433debSMiao Xie root->log_transid_committed = -1; 1346257c62e1SChris Mason root->last_log_commit = 0; 1347e02119d5SChris Mason return 0; 1348e02119d5SChris Mason } 1349e02119d5SChris Mason 135049d11beaSJosef Bacik static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root, 135149d11beaSJosef Bacik struct btrfs_path *path, 1352cb517eabSMiao Xie struct btrfs_key *key) 1353e02119d5SChris Mason { 1354e02119d5SChris Mason struct btrfs_root *root; 1355e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 135684234f3aSYan Zheng u64 generation; 1357cb517eabSMiao Xie int ret; 1358581c1760SQu Wenruo int level; 1359cb517eabSMiao Xie 136096dfcb46SJosef Bacik root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS); 136149d11beaSJosef Bacik if (!root) 136249d11beaSJosef Bacik return ERR_PTR(-ENOMEM); 13630f7d52f4SChris Mason 1364cb517eabSMiao Xie ret = btrfs_find_root(tree_root, key, path, 1365cb517eabSMiao Xie &root->root_item, &root->root_key); 13660f7d52f4SChris Mason if (ret) { 136713a8a7c8SYan, Zheng if (ret > 0) 136813a8a7c8SYan, Zheng ret = -ENOENT; 136949d11beaSJosef Bacik goto fail; 13700f7d52f4SChris Mason } 137113a8a7c8SYan, Zheng 137284234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1373581c1760SQu Wenruo level = btrfs_root_level(&root->root_item); 13742ff7e61eSJeff Mahoney root->node = read_tree_block(fs_info, 13752ff7e61eSJeff Mahoney btrfs_root_bytenr(&root->root_item), 13761b7ec85eSJosef Bacik key->objectid, generation, level, NULL); 137764c043deSLiu Bo if (IS_ERR(root->node)) { 137864c043deSLiu Bo ret = PTR_ERR(root->node); 13798c38938cSJosef Bacik root->node = NULL; 138049d11beaSJosef Bacik goto fail; 13814eb150d6SQu Wenruo } 13824eb150d6SQu Wenruo if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1383cb517eabSMiao Xie ret = -EIO; 138449d11beaSJosef Bacik goto fail; 1385416bc658SJosef Bacik } 138688c602abSQu Wenruo 138788c602abSQu Wenruo /* 138888c602abSQu Wenruo * For real fs, and not log/reloc trees, root owner must 138988c602abSQu Wenruo * match its root node owner 139088c602abSQu Wenruo */ 139188c602abSQu Wenruo if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) && 139288c602abSQu Wenruo root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID && 139388c602abSQu Wenruo root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && 139488c602abSQu Wenruo root->root_key.objectid != btrfs_header_owner(root->node)) { 139588c602abSQu Wenruo btrfs_crit(fs_info, 139688c602abSQu Wenruo "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu", 139788c602abSQu Wenruo root->root_key.objectid, root->node->start, 139888c602abSQu Wenruo btrfs_header_owner(root->node), 139988c602abSQu Wenruo root->root_key.objectid); 140088c602abSQu Wenruo ret = -EUCLEAN; 140188c602abSQu Wenruo goto fail; 140288c602abSQu Wenruo } 14035d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 1404cb517eabSMiao Xie return root; 140549d11beaSJosef Bacik fail: 140600246528SJosef Bacik btrfs_put_root(root); 140749d11beaSJosef Bacik return ERR_PTR(ret); 140849d11beaSJosef Bacik } 140949d11beaSJosef Bacik 141049d11beaSJosef Bacik struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 141149d11beaSJosef Bacik struct btrfs_key *key) 141249d11beaSJosef Bacik { 141349d11beaSJosef Bacik struct btrfs_root *root; 141449d11beaSJosef Bacik struct btrfs_path *path; 141549d11beaSJosef Bacik 141649d11beaSJosef Bacik path = btrfs_alloc_path(); 141749d11beaSJosef Bacik if (!path) 141849d11beaSJosef Bacik return ERR_PTR(-ENOMEM); 141949d11beaSJosef Bacik root = read_tree_root_path(tree_root, path, key); 142049d11beaSJosef Bacik btrfs_free_path(path); 142149d11beaSJosef Bacik 142249d11beaSJosef Bacik return root; 1423cb517eabSMiao Xie } 1424cb517eabSMiao Xie 14252dfb1e43SQu Wenruo /* 14262dfb1e43SQu Wenruo * Initialize subvolume root in-memory structure 14272dfb1e43SQu Wenruo * 14282dfb1e43SQu Wenruo * @anon_dev: anonymous device to attach to the root, if zero, allocate new 14292dfb1e43SQu Wenruo */ 14302dfb1e43SQu Wenruo static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev) 1431cb517eabSMiao Xie { 1432cb517eabSMiao Xie int ret; 1433dcc3eb96SNikolay Borisov unsigned int nofs_flag; 1434cb517eabSMiao Xie 1435dcc3eb96SNikolay Borisov /* 1436dcc3eb96SNikolay Borisov * We might be called under a transaction (e.g. indirect backref 1437dcc3eb96SNikolay Borisov * resolution) which could deadlock if it triggers memory reclaim 1438dcc3eb96SNikolay Borisov */ 1439dcc3eb96SNikolay Borisov nofs_flag = memalloc_nofs_save(); 1440dcc3eb96SNikolay Borisov ret = btrfs_drew_lock_init(&root->snapshot_lock); 1441dcc3eb96SNikolay Borisov memalloc_nofs_restore(nofs_flag); 1442dcc3eb96SNikolay Borisov if (ret) 14438257b2dcSMiao Xie goto fail; 14448257b2dcSMiao Xie 1445aeb935a4SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID && 144637f00a6dSJohannes Thumshirn !btrfs_is_data_reloc_root(root)) { 144792a7cc42SQu Wenruo set_bit(BTRFS_ROOT_SHAREABLE, &root->state); 1448f39e4571SJosef Bacik btrfs_check_and_init_root_item(&root->root_item); 1449f39e4571SJosef Bacik } 1450f39e4571SJosef Bacik 1451851fd730SQu Wenruo /* 1452851fd730SQu Wenruo * Don't assign anonymous block device to roots that are not exposed to 1453851fd730SQu Wenruo * userspace, the id pool is limited to 1M 1454851fd730SQu Wenruo */ 1455851fd730SQu Wenruo if (is_fstree(root->root_key.objectid) && 1456851fd730SQu Wenruo btrfs_root_refs(&root->root_item) > 0) { 14572dfb1e43SQu Wenruo if (!anon_dev) { 1458cb517eabSMiao Xie ret = get_anon_bdev(&root->anon_dev); 1459cb517eabSMiao Xie if (ret) 1460876d2cf1SLiu Bo goto fail; 14612dfb1e43SQu Wenruo } else { 14622dfb1e43SQu Wenruo root->anon_dev = anon_dev; 14632dfb1e43SQu Wenruo } 1464851fd730SQu Wenruo } 1465f32e48e9SChandan Rajendra 1466f32e48e9SChandan Rajendra mutex_lock(&root->objectid_mutex); 1467453e4873SNikolay Borisov ret = btrfs_init_root_free_objectid(root); 1468f32e48e9SChandan Rajendra if (ret) { 1469f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1470876d2cf1SLiu Bo goto fail; 1471f32e48e9SChandan Rajendra } 1472f32e48e9SChandan Rajendra 14736b8fad57SNikolay Borisov ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID); 1474f32e48e9SChandan Rajendra 1475f32e48e9SChandan Rajendra mutex_unlock(&root->objectid_mutex); 1476f32e48e9SChandan Rajendra 1477cb517eabSMiao Xie return 0; 1478cb517eabSMiao Xie fail: 147984db5ccfSDavid Sterba /* The caller is responsible to call btrfs_free_fs_root */ 1480cb517eabSMiao Xie return ret; 1481cb517eabSMiao Xie } 1482cb517eabSMiao Xie 1483a98db0f3SJosef Bacik static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1484cb517eabSMiao Xie u64 root_id) 1485cb517eabSMiao Xie { 1486cb517eabSMiao Xie struct btrfs_root *root; 1487cb517eabSMiao Xie 1488fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1489fc7cbcd4SDavid Sterba root = radix_tree_lookup(&fs_info->fs_roots_radix, 1490fc7cbcd4SDavid Sterba (unsigned long)root_id); 1491bc44d7c4SJosef Bacik if (root) 149200246528SJosef Bacik root = btrfs_grab_root(root); 1493fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 1494cb517eabSMiao Xie return root; 1495cb517eabSMiao Xie } 1496cb517eabSMiao Xie 149749d11beaSJosef Bacik static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info, 149849d11beaSJosef Bacik u64 objectid) 149949d11beaSJosef Bacik { 1500abed4aaaSJosef Bacik struct btrfs_key key = { 1501abed4aaaSJosef Bacik .objectid = objectid, 1502abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 1503abed4aaaSJosef Bacik .offset = 0, 1504abed4aaaSJosef Bacik }; 1505abed4aaaSJosef Bacik 150649d11beaSJosef Bacik if (objectid == BTRFS_ROOT_TREE_OBJECTID) 150749d11beaSJosef Bacik return btrfs_grab_root(fs_info->tree_root); 150849d11beaSJosef Bacik if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 1509abed4aaaSJosef Bacik return btrfs_grab_root(btrfs_global_root(fs_info, &key)); 151049d11beaSJosef Bacik if (objectid == BTRFS_CHUNK_TREE_OBJECTID) 151149d11beaSJosef Bacik return btrfs_grab_root(fs_info->chunk_root); 151249d11beaSJosef Bacik if (objectid == BTRFS_DEV_TREE_OBJECTID) 151349d11beaSJosef Bacik return btrfs_grab_root(fs_info->dev_root); 151449d11beaSJosef Bacik if (objectid == BTRFS_CSUM_TREE_OBJECTID) 1515abed4aaaSJosef Bacik return btrfs_grab_root(btrfs_global_root(fs_info, &key)); 151649d11beaSJosef Bacik if (objectid == BTRFS_QUOTA_TREE_OBJECTID) 151749d11beaSJosef Bacik return btrfs_grab_root(fs_info->quota_root) ? 151849d11beaSJosef Bacik fs_info->quota_root : ERR_PTR(-ENOENT); 151949d11beaSJosef Bacik if (objectid == BTRFS_UUID_TREE_OBJECTID) 152049d11beaSJosef Bacik return btrfs_grab_root(fs_info->uuid_root) ? 152149d11beaSJosef Bacik fs_info->uuid_root : ERR_PTR(-ENOENT); 152214033b08SQu Wenruo if (objectid == BTRFS_BLOCK_GROUP_TREE_OBJECTID) 152314033b08SQu Wenruo return btrfs_grab_root(fs_info->block_group_root) ? 152414033b08SQu Wenruo fs_info->block_group_root : ERR_PTR(-ENOENT); 1525abed4aaaSJosef Bacik if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) { 1526abed4aaaSJosef Bacik struct btrfs_root *root = btrfs_global_root(fs_info, &key); 1527abed4aaaSJosef Bacik 1528abed4aaaSJosef Bacik return btrfs_grab_root(root) ? root : ERR_PTR(-ENOENT); 1529abed4aaaSJosef Bacik } 153049d11beaSJosef Bacik return NULL; 153149d11beaSJosef Bacik } 153249d11beaSJosef Bacik 1533cb517eabSMiao Xie int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1534cb517eabSMiao Xie struct btrfs_root *root) 1535cb517eabSMiao Xie { 1536cb517eabSMiao Xie int ret; 1537cb517eabSMiao Xie 1538fc7cbcd4SDavid Sterba ret = radix_tree_preload(GFP_NOFS); 1539fc7cbcd4SDavid Sterba if (ret) 1540fc7cbcd4SDavid Sterba return ret; 1541fc7cbcd4SDavid Sterba 1542fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 1543fc7cbcd4SDavid Sterba ret = radix_tree_insert(&fs_info->fs_roots_radix, 1544fc7cbcd4SDavid Sterba (unsigned long)root->root_key.objectid, 1545fc7cbcd4SDavid Sterba root); 1546af01d2e5SJosef Bacik if (ret == 0) { 154700246528SJosef Bacik btrfs_grab_root(root); 1548fc7cbcd4SDavid Sterba set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1549af01d2e5SJosef Bacik } 1550fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 1551fc7cbcd4SDavid Sterba radix_tree_preload_end(); 1552cb517eabSMiao Xie 1553cb517eabSMiao Xie return ret; 1554cb517eabSMiao Xie } 1555cb517eabSMiao Xie 1556bd647ce3SJosef Bacik void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info) 1557bd647ce3SJosef Bacik { 1558bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 1559bd647ce3SJosef Bacik struct btrfs_root *root; 1560bd647ce3SJosef Bacik 1561bd647ce3SJosef Bacik while (!list_empty(&fs_info->allocated_roots)) { 1562457f1864SJosef Bacik char buf[BTRFS_ROOT_NAME_BUF_LEN]; 1563457f1864SJosef Bacik 1564bd647ce3SJosef Bacik root = list_first_entry(&fs_info->allocated_roots, 1565bd647ce3SJosef Bacik struct btrfs_root, leak_list); 1566457f1864SJosef Bacik btrfs_err(fs_info, "leaked root %s refcount %d", 156771008734SJosef Bacik btrfs_root_name(&root->root_key, buf), 1568bd647ce3SJosef Bacik refcount_read(&root->refs)); 1569bd647ce3SJosef Bacik while (refcount_read(&root->refs) > 1) 157000246528SJosef Bacik btrfs_put_root(root); 157100246528SJosef Bacik btrfs_put_root(root); 1572bd647ce3SJosef Bacik } 1573bd647ce3SJosef Bacik #endif 1574bd647ce3SJosef Bacik } 1575bd647ce3SJosef Bacik 1576abed4aaaSJosef Bacik static void free_global_roots(struct btrfs_fs_info *fs_info) 1577abed4aaaSJosef Bacik { 1578abed4aaaSJosef Bacik struct btrfs_root *root; 1579abed4aaaSJosef Bacik struct rb_node *node; 1580abed4aaaSJosef Bacik 1581abed4aaaSJosef Bacik while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) { 1582abed4aaaSJosef Bacik root = rb_entry(node, struct btrfs_root, rb_node); 1583abed4aaaSJosef Bacik rb_erase(&root->rb_node, &fs_info->global_root_tree); 1584abed4aaaSJosef Bacik btrfs_put_root(root); 1585abed4aaaSJosef Bacik } 1586abed4aaaSJosef Bacik } 1587abed4aaaSJosef Bacik 15880d4b0463SJosef Bacik void btrfs_free_fs_info(struct btrfs_fs_info *fs_info) 15890d4b0463SJosef Bacik { 1590141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 1591141386e1SJosef Bacik percpu_counter_destroy(&fs_info->delalloc_bytes); 15925deb17e1SJosef Bacik percpu_counter_destroy(&fs_info->ordered_bytes); 1593141386e1SJosef Bacik percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 1594141386e1SJosef Bacik btrfs_free_csum_hash(fs_info); 1595141386e1SJosef Bacik btrfs_free_stripe_hash_table(fs_info); 1596141386e1SJosef Bacik btrfs_free_ref_cache(fs_info); 15970d4b0463SJosef Bacik kfree(fs_info->balance_ctl); 15980d4b0463SJosef Bacik kfree(fs_info->delayed_root); 1599abed4aaaSJosef Bacik free_global_roots(fs_info); 160000246528SJosef Bacik btrfs_put_root(fs_info->tree_root); 160100246528SJosef Bacik btrfs_put_root(fs_info->chunk_root); 160200246528SJosef Bacik btrfs_put_root(fs_info->dev_root); 160300246528SJosef Bacik btrfs_put_root(fs_info->quota_root); 160400246528SJosef Bacik btrfs_put_root(fs_info->uuid_root); 160500246528SJosef Bacik btrfs_put_root(fs_info->fs_root); 1606aeb935a4SQu Wenruo btrfs_put_root(fs_info->data_reloc_root); 16079c54e80dSJosef Bacik btrfs_put_root(fs_info->block_group_root); 1608bd647ce3SJosef Bacik btrfs_check_leaked_roots(fs_info); 16093fd63727SJosef Bacik btrfs_extent_buffer_leak_debug_check(fs_info); 16100d4b0463SJosef Bacik kfree(fs_info->super_copy); 16110d4b0463SJosef Bacik kfree(fs_info->super_for_commit); 16128481dd80SQu Wenruo kfree(fs_info->subpage_info); 16130d4b0463SJosef Bacik kvfree(fs_info); 16140d4b0463SJosef Bacik } 16150d4b0463SJosef Bacik 16160d4b0463SJosef Bacik 16172dfb1e43SQu Wenruo /* 16182dfb1e43SQu Wenruo * Get an in-memory reference of a root structure. 16192dfb1e43SQu Wenruo * 16202dfb1e43SQu Wenruo * For essential trees like root/extent tree, we grab it from fs_info directly. 16212dfb1e43SQu Wenruo * For subvolume trees, we check the cached filesystem roots first. If not 16222dfb1e43SQu Wenruo * found, then read it from disk and add it to cached fs roots. 16232dfb1e43SQu Wenruo * 16242dfb1e43SQu Wenruo * Caller should release the root by calling btrfs_put_root() after the usage. 16252dfb1e43SQu Wenruo * 16262dfb1e43SQu Wenruo * NOTE: Reloc and log trees can't be read by this function as they share the 16272dfb1e43SQu Wenruo * same root objectid. 16282dfb1e43SQu Wenruo * 16292dfb1e43SQu Wenruo * @objectid: root id 16302dfb1e43SQu Wenruo * @anon_dev: preallocated anonymous block device number for new roots, 16312dfb1e43SQu Wenruo * pass 0 for new allocation. 16322dfb1e43SQu Wenruo * @check_ref: whether to check root item references, If true, return -ENOENT 16332dfb1e43SQu Wenruo * for orphan roots 16342dfb1e43SQu Wenruo */ 16352dfb1e43SQu Wenruo static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info, 16362dfb1e43SQu Wenruo u64 objectid, dev_t anon_dev, 16372dfb1e43SQu Wenruo bool check_ref) 16385eda7b5eSChris Mason { 16395eda7b5eSChris Mason struct btrfs_root *root; 1640381cf658SDavid Sterba struct btrfs_path *path; 16411d4c08e0SDavid Sterba struct btrfs_key key; 16425eda7b5eSChris Mason int ret; 16435eda7b5eSChris Mason 164449d11beaSJosef Bacik root = btrfs_get_global_root(fs_info, objectid); 164549d11beaSJosef Bacik if (root) 164649d11beaSJosef Bacik return root; 16474df27c4dSYan, Zheng again: 164856e9357aSDavid Sterba root = btrfs_lookup_fs_root(fs_info, objectid); 164948475471SStefan Behrens if (root) { 16502dfb1e43SQu Wenruo /* Shouldn't get preallocated anon_dev for cached roots */ 16512dfb1e43SQu Wenruo ASSERT(!anon_dev); 1652bc44d7c4SJosef Bacik if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 165300246528SJosef Bacik btrfs_put_root(root); 165448475471SStefan Behrens return ERR_PTR(-ENOENT); 1655bc44d7c4SJosef Bacik } 16565eda7b5eSChris Mason return root; 165748475471SStefan Behrens } 16585eda7b5eSChris Mason 165956e9357aSDavid Sterba key.objectid = objectid; 166056e9357aSDavid Sterba key.type = BTRFS_ROOT_ITEM_KEY; 166156e9357aSDavid Sterba key.offset = (u64)-1; 166256e9357aSDavid Sterba root = btrfs_read_tree_root(fs_info->tree_root, &key); 16635eda7b5eSChris Mason if (IS_ERR(root)) 16645eda7b5eSChris Mason return root; 16653394e160SChris Mason 1666c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1667d68fc57bSYan, Zheng ret = -ENOENT; 1668d68fc57bSYan, Zheng goto fail; 1669d68fc57bSYan, Zheng } 1670d68fc57bSYan, Zheng 16712dfb1e43SQu Wenruo ret = btrfs_init_fs_root(root, anon_dev); 1672cb517eabSMiao Xie if (ret) 1673cb517eabSMiao Xie goto fail; 1674cb517eabSMiao Xie 1675381cf658SDavid Sterba path = btrfs_alloc_path(); 1676381cf658SDavid Sterba if (!path) { 1677381cf658SDavid Sterba ret = -ENOMEM; 1678381cf658SDavid Sterba goto fail; 1679381cf658SDavid Sterba } 16801d4c08e0SDavid Sterba key.objectid = BTRFS_ORPHAN_OBJECTID; 16811d4c08e0SDavid Sterba key.type = BTRFS_ORPHAN_ITEM_KEY; 168256e9357aSDavid Sterba key.offset = objectid; 16831d4c08e0SDavid Sterba 16841d4c08e0SDavid Sterba ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1685381cf658SDavid Sterba btrfs_free_path(path); 1686d68fc57bSYan, Zheng if (ret < 0) 1687d68fc57bSYan, Zheng goto fail; 1688d68fc57bSYan, Zheng if (ret == 0) 168927cdeb70SMiao Xie set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1690d68fc57bSYan, Zheng 1691cb517eabSMiao Xie ret = btrfs_insert_fs_root(fs_info, root); 16920f7d52f4SChris Mason if (ret) { 1693168a2f77SJia-Ju Bai if (ret == -EEXIST) { 169400246528SJosef Bacik btrfs_put_root(root); 16954df27c4dSYan, Zheng goto again; 1696168a2f77SJia-Ju Bai } 16974df27c4dSYan, Zheng goto fail; 16984df27c4dSYan, Zheng } 1699edbd8d4eSChris Mason return root; 17004df27c4dSYan, Zheng fail: 170133fab972SFilipe Manana /* 170233fab972SFilipe Manana * If our caller provided us an anonymous device, then it's his 1703143823cfSDavid Sterba * responsibility to free it in case we fail. So we have to set our 170433fab972SFilipe Manana * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root() 170533fab972SFilipe Manana * and once again by our caller. 170633fab972SFilipe Manana */ 170733fab972SFilipe Manana if (anon_dev) 170833fab972SFilipe Manana root->anon_dev = 0; 17098c38938cSJosef Bacik btrfs_put_root(root); 17104df27c4dSYan, Zheng return ERR_PTR(ret); 1711edbd8d4eSChris Mason } 1712edbd8d4eSChris Mason 17132dfb1e43SQu Wenruo /* 17142dfb1e43SQu Wenruo * Get in-memory reference of a root structure 17152dfb1e43SQu Wenruo * 17162dfb1e43SQu Wenruo * @objectid: tree objectid 17172dfb1e43SQu Wenruo * @check_ref: if set, verify that the tree exists and the item has at least 17182dfb1e43SQu Wenruo * one reference 17192dfb1e43SQu Wenruo */ 17202dfb1e43SQu Wenruo struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 17212dfb1e43SQu Wenruo u64 objectid, bool check_ref) 17222dfb1e43SQu Wenruo { 17232dfb1e43SQu Wenruo return btrfs_get_root_ref(fs_info, objectid, 0, check_ref); 17242dfb1e43SQu Wenruo } 17252dfb1e43SQu Wenruo 17262dfb1e43SQu Wenruo /* 17272dfb1e43SQu Wenruo * Get in-memory reference of a root structure, created as new, optionally pass 17282dfb1e43SQu Wenruo * the anonymous block device id 17292dfb1e43SQu Wenruo * 17302dfb1e43SQu Wenruo * @objectid: tree objectid 17312dfb1e43SQu Wenruo * @anon_dev: if zero, allocate a new anonymous block device or use the 17322dfb1e43SQu Wenruo * parameter value 17332dfb1e43SQu Wenruo */ 17342dfb1e43SQu Wenruo struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info, 17352dfb1e43SQu Wenruo u64 objectid, dev_t anon_dev) 17362dfb1e43SQu Wenruo { 17372dfb1e43SQu Wenruo return btrfs_get_root_ref(fs_info, objectid, anon_dev, true); 17382dfb1e43SQu Wenruo } 17392dfb1e43SQu Wenruo 17408b712842SChris Mason /* 174149d11beaSJosef Bacik * btrfs_get_fs_root_commit_root - return a root for the given objectid 174249d11beaSJosef Bacik * @fs_info: the fs_info 174349d11beaSJosef Bacik * @objectid: the objectid we need to lookup 174449d11beaSJosef Bacik * 174549d11beaSJosef Bacik * This is exclusively used for backref walking, and exists specifically because 174649d11beaSJosef Bacik * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref 174749d11beaSJosef Bacik * creation time, which means we may have to read the tree_root in order to look 174849d11beaSJosef Bacik * up a fs root that is not in memory. If the root is not in memory we will 174949d11beaSJosef Bacik * read the tree root commit root and look up the fs root from there. This is a 175049d11beaSJosef Bacik * temporary root, it will not be inserted into the radix tree as it doesn't 175149d11beaSJosef Bacik * have the most uptodate information, it'll simply be discarded once the 175249d11beaSJosef Bacik * backref code is finished using the root. 175349d11beaSJosef Bacik */ 175449d11beaSJosef Bacik struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info, 175549d11beaSJosef Bacik struct btrfs_path *path, 175649d11beaSJosef Bacik u64 objectid) 175749d11beaSJosef Bacik { 175849d11beaSJosef Bacik struct btrfs_root *root; 175949d11beaSJosef Bacik struct btrfs_key key; 176049d11beaSJosef Bacik 176149d11beaSJosef Bacik ASSERT(path->search_commit_root && path->skip_locking); 176249d11beaSJosef Bacik 176349d11beaSJosef Bacik /* 176449d11beaSJosef Bacik * This can return -ENOENT if we ask for a root that doesn't exist, but 176549d11beaSJosef Bacik * since this is called via the backref walking code we won't be looking 176649d11beaSJosef Bacik * up a root that doesn't exist, unless there's corruption. So if root 176749d11beaSJosef Bacik * != NULL just return it. 176849d11beaSJosef Bacik */ 176949d11beaSJosef Bacik root = btrfs_get_global_root(fs_info, objectid); 177049d11beaSJosef Bacik if (root) 177149d11beaSJosef Bacik return root; 177249d11beaSJosef Bacik 177349d11beaSJosef Bacik root = btrfs_lookup_fs_root(fs_info, objectid); 177449d11beaSJosef Bacik if (root) 177549d11beaSJosef Bacik return root; 177649d11beaSJosef Bacik 177749d11beaSJosef Bacik key.objectid = objectid; 177849d11beaSJosef Bacik key.type = BTRFS_ROOT_ITEM_KEY; 177949d11beaSJosef Bacik key.offset = (u64)-1; 178049d11beaSJosef Bacik root = read_tree_root_path(fs_info->tree_root, path, &key); 178149d11beaSJosef Bacik btrfs_release_path(path); 178249d11beaSJosef Bacik 178349d11beaSJosef Bacik return root; 178449d11beaSJosef Bacik } 178549d11beaSJosef Bacik 1786a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1787a74a4b97SChris Mason { 17880d031dc4SYu Zhe struct btrfs_fs_info *fs_info = arg; 1789d0278245SMiao Xie int again; 1790a74a4b97SChris Mason 1791d6fd0ae2SOmar Sandoval while (1) { 1792d0278245SMiao Xie again = 0; 17939d1a2a3aSDavid Sterba 1794fd340d0fSJosef Bacik set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1795fd340d0fSJosef Bacik 1796d0278245SMiao Xie /* Make the cleaner go to sleep early. */ 17972ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) 1798d0278245SMiao Xie goto sleep; 1799d0278245SMiao Xie 180090c711abSZygo Blaxell /* 180190c711abSZygo Blaxell * Do not do anything if we might cause open_ctree() to block 180290c711abSZygo Blaxell * before we have finished mounting the filesystem. 180390c711abSZygo Blaxell */ 18040b246afaSJeff Mahoney if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 180590c711abSZygo Blaxell goto sleep; 180690c711abSZygo Blaxell 18070b246afaSJeff Mahoney if (!mutex_trylock(&fs_info->cleaner_mutex)) 1808d0278245SMiao Xie goto sleep; 1809d0278245SMiao Xie 1810dc7f370cSMiao Xie /* 1811dc7f370cSMiao Xie * Avoid the problem that we change the status of the fs 1812dc7f370cSMiao Xie * during the above check and trylock. 1813dc7f370cSMiao Xie */ 18142ff7e61eSJeff Mahoney if (btrfs_need_cleaner_sleep(fs_info)) { 18150b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1816dc7f370cSMiao Xie goto sleep; 1817dc7f370cSMiao Xie } 1818dc7f370cSMiao Xie 18192ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 1820c2d6cb16SFilipe Manana 182133c44184SJosef Bacik again = btrfs_clean_one_deleted_snapshot(fs_info); 18220b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 1823a74a4b97SChris Mason 1824d0278245SMiao Xie /* 182505323cd1SMiao Xie * The defragger has dealt with the R/O remount and umount, 182605323cd1SMiao Xie * needn't do anything special here. 1827d0278245SMiao Xie */ 18280b246afaSJeff Mahoney btrfs_run_defrag_inodes(fs_info); 182967c5e7d4SFilipe Manana 183067c5e7d4SFilipe Manana /* 1831f3372065SJohannes Thumshirn * Acquires fs_info->reclaim_bgs_lock to avoid racing 183267c5e7d4SFilipe Manana * with relocation (btrfs_relocate_chunk) and relocation 183367c5e7d4SFilipe Manana * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 1834f3372065SJohannes Thumshirn * after acquiring fs_info->reclaim_bgs_lock. So we 183567c5e7d4SFilipe Manana * can't hold, nor need to, fs_info->cleaner_mutex when deleting 183667c5e7d4SFilipe Manana * unused block groups. 183767c5e7d4SFilipe Manana */ 18380b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 183918bb8bbfSJohannes Thumshirn 184018bb8bbfSJohannes Thumshirn /* 184118bb8bbfSJohannes Thumshirn * Reclaim block groups in the reclaim_bgs list after we deleted 184218bb8bbfSJohannes Thumshirn * all unused block_groups. This possibly gives us some more free 184318bb8bbfSJohannes Thumshirn * space. 184418bb8bbfSJohannes Thumshirn */ 184518bb8bbfSJohannes Thumshirn btrfs_reclaim_bgs(fs_info); 1846d0278245SMiao Xie sleep: 1847a0a1db70SFilipe Manana clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1848d6fd0ae2SOmar Sandoval if (kthread_should_park()) 1849d6fd0ae2SOmar Sandoval kthread_parkme(); 1850d6fd0ae2SOmar Sandoval if (kthread_should_stop()) 1851d6fd0ae2SOmar Sandoval return 0; 1852838fe188SJiri Kosina if (!again) { 1853a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 1854a74a4b97SChris Mason schedule(); 1855a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1856a74a4b97SChris Mason } 1857da288d28SFilipe Manana } 1858a74a4b97SChris Mason } 1859a74a4b97SChris Mason 1860a74a4b97SChris Mason static int transaction_kthread(void *arg) 1861a74a4b97SChris Mason { 1862a74a4b97SChris Mason struct btrfs_root *root = arg; 18630b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1864a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1865a74a4b97SChris Mason struct btrfs_transaction *cur; 18668929ecfaSYan, Zheng u64 transid; 1867643900beSNikolay Borisov time64_t delta; 1868a74a4b97SChris Mason unsigned long delay; 1869914b2007SJan Kara bool cannot_commit; 1870a74a4b97SChris Mason 1871a74a4b97SChris Mason do { 1872914b2007SJan Kara cannot_commit = false; 1873ba1bc00fSNikolay Borisov delay = msecs_to_jiffies(fs_info->commit_interval * 1000); 18740b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 1875a74a4b97SChris Mason 18760b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 18770b246afaSJeff Mahoney cur = fs_info->running_transaction; 1878a74a4b97SChris Mason if (!cur) { 18790b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1880a74a4b97SChris Mason goto sleep; 1881a74a4b97SChris Mason } 188231153d81SYan Zheng 1883643900beSNikolay Borisov delta = ktime_get_seconds() - cur->start_time; 1884fdfbf020SJosef Bacik if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) && 1885fdfbf020SJosef Bacik cur->state < TRANS_STATE_COMMIT_START && 1886643900beSNikolay Borisov delta < fs_info->commit_interval) { 18870b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1888fb8a7e94SNikolay Borisov delay -= msecs_to_jiffies((delta - 1) * 1000); 1889fb8a7e94SNikolay Borisov delay = min(delay, 1890fb8a7e94SNikolay Borisov msecs_to_jiffies(fs_info->commit_interval * 1000)); 1891a74a4b97SChris Mason goto sleep; 1892a74a4b97SChris Mason } 18938929ecfaSYan, Zheng transid = cur->transid; 18940b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 189556bec294SChris Mason 189679787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 1897354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 1898914b2007SJan Kara if (IS_ERR(trans)) { 1899354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 1900914b2007SJan Kara cannot_commit = true; 190179787eaaSJeff Mahoney goto sleep; 1902914b2007SJan Kara } 19038929ecfaSYan, Zheng if (transid == trans->transid) { 19043a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 19058929ecfaSYan, Zheng } else { 19063a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19078929ecfaSYan, Zheng } 1908a74a4b97SChris Mason sleep: 19090b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 19100b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 1911a74a4b97SChris Mason 191284961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) 19132ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 19148929ecfaSYan, Zheng if (!kthread_should_stop() && 19150b246afaSJeff Mahoney (!btrfs_transaction_blocked(fs_info) || 1916914b2007SJan Kara cannot_commit)) 1917bc5511d0SNikolay Borisov schedule_timeout_interruptible(delay); 1918a74a4b97SChris Mason } while (!kthread_should_stop()); 1919a74a4b97SChris Mason return 0; 1920a74a4b97SChris Mason } 1921a74a4b97SChris Mason 1922af31f5e5SChris Mason /* 192301f0f9daSNikolay Borisov * This will find the highest generation in the array of root backups. The 192401f0f9daSNikolay Borisov * index of the highest array is returned, or -EINVAL if we can't find 192501f0f9daSNikolay Borisov * anything. 1926af31f5e5SChris Mason * 1927af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1928af31f5e5SChris Mason * generation of the latest root in the array with the generation 1929af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1930af31f5e5SChris Mason */ 193101f0f9daSNikolay Borisov static int find_newest_super_backup(struct btrfs_fs_info *info) 1932af31f5e5SChris Mason { 193301f0f9daSNikolay Borisov const u64 newest_gen = btrfs_super_generation(info->super_copy); 1934af31f5e5SChris Mason u64 cur; 1935af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1936af31f5e5SChris Mason int i; 1937af31f5e5SChris Mason 1938af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1939af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1940af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1941af31f5e5SChris Mason if (cur == newest_gen) 194201f0f9daSNikolay Borisov return i; 1943af31f5e5SChris Mason } 1944af31f5e5SChris Mason 194501f0f9daSNikolay Borisov return -EINVAL; 1946af31f5e5SChris Mason } 1947af31f5e5SChris Mason 1948af31f5e5SChris Mason /* 1949af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1950af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1951af31f5e5SChris Mason * done 1952af31f5e5SChris Mason */ 1953af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1954af31f5e5SChris Mason { 19556ef108ddSNikolay Borisov const int next_backup = info->backup_root_index; 1956af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1957af31f5e5SChris Mason 1958af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1959af31f5e5SChris Mason 1960af31f5e5SChris Mason /* 1961af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1962af31f5e5SChris Mason * regardless of which ones we use today 1963af31f5e5SChris Mason */ 1964af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1965af31f5e5SChris Mason 1966af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1967af31f5e5SChris Mason 1968af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1969af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1970af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1971af31f5e5SChris Mason 1972af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1973af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1974af31f5e5SChris Mason 1975af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1976af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1977af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 1978af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 1979af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 1980af31f5e5SChris Mason 19811c56ab99SQu Wenruo if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) { 19829c54e80dSJosef Bacik struct btrfs_root *extent_root = btrfs_extent_root(info, 0); 1983f7238e50SJosef Bacik struct btrfs_root *csum_root = btrfs_csum_root(info, 0); 19849c54e80dSJosef Bacik 19859c54e80dSJosef Bacik btrfs_set_backup_extent_root(root_backup, 19869c54e80dSJosef Bacik extent_root->node->start); 1987af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 198829cbcf40SJosef Bacik btrfs_header_generation(extent_root->node)); 1989af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 199029cbcf40SJosef Bacik btrfs_header_level(extent_root->node)); 1991af31f5e5SChris Mason 1992f7238e50SJosef Bacik btrfs_set_backup_csum_root(root_backup, csum_root->node->start); 1993f7238e50SJosef Bacik btrfs_set_backup_csum_root_gen(root_backup, 1994f7238e50SJosef Bacik btrfs_header_generation(csum_root->node)); 1995f7238e50SJosef Bacik btrfs_set_backup_csum_root_level(root_backup, 1996f7238e50SJosef Bacik btrfs_header_level(csum_root->node)); 19979c54e80dSJosef Bacik } 1998af31f5e5SChris Mason 19997c7e82a7SChris Mason /* 20007c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 20017c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 20027c7e82a7SChris Mason */ 20037c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 20047c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 20057c7e82a7SChris Mason info->fs_root->node->start); 2006af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 2007af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 2008af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 2009af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 20107c7e82a7SChris Mason } 2011af31f5e5SChris Mason 2012af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 2013af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 2014af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 2015af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 2016af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 2017af31f5e5SChris Mason 2018af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 2019af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 2020af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 2021af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 2022af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 2023af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 2024af31f5e5SChris Mason 2025af31f5e5SChris Mason /* 2026af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 2027af31f5e5SChris Mason * for the next commit 2028af31f5e5SChris Mason */ 2029af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 2030af31f5e5SChris Mason &info->super_for_commit->super_roots, 2031af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 2032af31f5e5SChris Mason } 2033af31f5e5SChris Mason 2034af31f5e5SChris Mason /* 2035bd2336b2SNikolay Borisov * read_backup_root - Reads a backup root based on the passed priority. Prio 0 2036bd2336b2SNikolay Borisov * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots 2037bd2336b2SNikolay Borisov * 2038bd2336b2SNikolay Borisov * fs_info - filesystem whose backup roots need to be read 2039bd2336b2SNikolay Borisov * priority - priority of backup root required 2040bd2336b2SNikolay Borisov * 2041bd2336b2SNikolay Borisov * Returns backup root index on success and -EINVAL otherwise. 2042bd2336b2SNikolay Borisov */ 2043bd2336b2SNikolay Borisov static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority) 2044bd2336b2SNikolay Borisov { 2045bd2336b2SNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 2046bd2336b2SNikolay Borisov struct btrfs_super_block *super = fs_info->super_copy; 2047bd2336b2SNikolay Borisov struct btrfs_root_backup *root_backup; 2048bd2336b2SNikolay Borisov 2049bd2336b2SNikolay Borisov if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) { 2050bd2336b2SNikolay Borisov if (priority == 0) 2051bd2336b2SNikolay Borisov return backup_index; 2052bd2336b2SNikolay Borisov 2053bd2336b2SNikolay Borisov backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority; 2054bd2336b2SNikolay Borisov backup_index %= BTRFS_NUM_BACKUP_ROOTS; 2055bd2336b2SNikolay Borisov } else { 2056bd2336b2SNikolay Borisov return -EINVAL; 2057bd2336b2SNikolay Borisov } 2058bd2336b2SNikolay Borisov 2059bd2336b2SNikolay Borisov root_backup = super->super_roots + backup_index; 2060bd2336b2SNikolay Borisov 2061bd2336b2SNikolay Borisov btrfs_set_super_generation(super, 2062bd2336b2SNikolay Borisov btrfs_backup_tree_root_gen(root_backup)); 2063bd2336b2SNikolay Borisov btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 2064bd2336b2SNikolay Borisov btrfs_set_super_root_level(super, 2065bd2336b2SNikolay Borisov btrfs_backup_tree_root_level(root_backup)); 2066bd2336b2SNikolay Borisov btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 2067bd2336b2SNikolay Borisov 2068bd2336b2SNikolay Borisov /* 2069bd2336b2SNikolay Borisov * Fixme: the total bytes and num_devices need to match or we should 2070bd2336b2SNikolay Borisov * need a fsck 2071bd2336b2SNikolay Borisov */ 2072bd2336b2SNikolay Borisov btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 2073bd2336b2SNikolay Borisov btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 2074bd2336b2SNikolay Borisov 2075bd2336b2SNikolay Borisov return backup_index; 2076bd2336b2SNikolay Borisov } 2077bd2336b2SNikolay Borisov 20787abadb64SLiu Bo /* helper to cleanup workers */ 20797abadb64SLiu Bo static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 20807abadb64SLiu Bo { 2081dc6e3209SQu Wenruo btrfs_destroy_workqueue(fs_info->fixup_workers); 2082afe3d242SQu Wenruo btrfs_destroy_workqueue(fs_info->delalloc_workers); 2083a31b4a43SChristoph Hellwig btrfs_destroy_workqueue(fs_info->hipri_workers); 20845cdc7ad3SQu Wenruo btrfs_destroy_workqueue(fs_info->workers); 2085d7b9416fSChristoph Hellwig if (fs_info->endio_workers) 2086d7b9416fSChristoph Hellwig destroy_workqueue(fs_info->endio_workers); 2087d34e123dSChristoph Hellwig if (fs_info->endio_raid56_workers) 2088d34e123dSChristoph Hellwig destroy_workqueue(fs_info->endio_raid56_workers); 2089385de0efSChristoph Hellwig if (fs_info->rmw_workers) 2090385de0efSChristoph Hellwig destroy_workqueue(fs_info->rmw_workers); 2091fed8a72dSChristoph Hellwig if (fs_info->compressed_write_workers) 2092fed8a72dSChristoph Hellwig destroy_workqueue(fs_info->compressed_write_workers); 2093fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_write_workers); 2094fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 20955b3bc44eSQu Wenruo btrfs_destroy_workqueue(fs_info->delayed_workers); 2096e66f0bb1SQu Wenruo btrfs_destroy_workqueue(fs_info->caching_workers); 2097a44903abSQu Wenruo btrfs_destroy_workqueue(fs_info->flush_workers); 2098fc97fab0SQu Wenruo btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2099b0643e59SDennis Zhou if (fs_info->discard_ctl.discard_workers) 2100b0643e59SDennis Zhou destroy_workqueue(fs_info->discard_ctl.discard_workers); 2101a9b9477dSFilipe Manana /* 2102a9b9477dSFilipe Manana * Now that all other work queues are destroyed, we can safely destroy 2103a9b9477dSFilipe Manana * the queues used for metadata I/O, since tasks from those other work 2104a9b9477dSFilipe Manana * queues can do metadata I/O operations. 2105a9b9477dSFilipe Manana */ 2106d7b9416fSChristoph Hellwig if (fs_info->endio_meta_workers) 2107d7b9416fSChristoph Hellwig destroy_workqueue(fs_info->endio_meta_workers); 21087abadb64SLiu Bo } 21097abadb64SLiu Bo 21102e9f5954SRashika static void free_root_extent_buffers(struct btrfs_root *root) 21112e9f5954SRashika { 21122e9f5954SRashika if (root) { 21132e9f5954SRashika free_extent_buffer(root->node); 21142e9f5954SRashika free_extent_buffer(root->commit_root); 21152e9f5954SRashika root->node = NULL; 21162e9f5954SRashika root->commit_root = NULL; 21172e9f5954SRashika } 21182e9f5954SRashika } 21192e9f5954SRashika 2120abed4aaaSJosef Bacik static void free_global_root_pointers(struct btrfs_fs_info *fs_info) 2121abed4aaaSJosef Bacik { 2122abed4aaaSJosef Bacik struct btrfs_root *root, *tmp; 2123abed4aaaSJosef Bacik 2124abed4aaaSJosef Bacik rbtree_postorder_for_each_entry_safe(root, tmp, 2125abed4aaaSJosef Bacik &fs_info->global_root_tree, 2126abed4aaaSJosef Bacik rb_node) 2127abed4aaaSJosef Bacik free_root_extent_buffers(root); 2128abed4aaaSJosef Bacik } 2129abed4aaaSJosef Bacik 2130af31f5e5SChris Mason /* helper to cleanup tree roots */ 21314273eaffSAnand Jain static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root) 2132af31f5e5SChris Mason { 21332e9f5954SRashika free_root_extent_buffers(info->tree_root); 2134655b09feSJosef Bacik 2135abed4aaaSJosef Bacik free_global_root_pointers(info); 21362e9f5954SRashika free_root_extent_buffers(info->dev_root); 21372e9f5954SRashika free_root_extent_buffers(info->quota_root); 21382e9f5954SRashika free_root_extent_buffers(info->uuid_root); 21398c38938cSJosef Bacik free_root_extent_buffers(info->fs_root); 2140aeb935a4SQu Wenruo free_root_extent_buffers(info->data_reloc_root); 21419c54e80dSJosef Bacik free_root_extent_buffers(info->block_group_root); 21424273eaffSAnand Jain if (free_chunk_root) 21432e9f5954SRashika free_root_extent_buffers(info->chunk_root); 2144af31f5e5SChris Mason } 2145af31f5e5SChris Mason 21468c38938cSJosef Bacik void btrfs_put_root(struct btrfs_root *root) 21478c38938cSJosef Bacik { 21488c38938cSJosef Bacik if (!root) 21498c38938cSJosef Bacik return; 21508c38938cSJosef Bacik 21518c38938cSJosef Bacik if (refcount_dec_and_test(&root->refs)) { 21528c38938cSJosef Bacik WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 21531dae7e0eSQu Wenruo WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)); 21548c38938cSJosef Bacik if (root->anon_dev) 21558c38938cSJosef Bacik free_anon_bdev(root->anon_dev); 21568c38938cSJosef Bacik btrfs_drew_lock_destroy(&root->snapshot_lock); 2157923eb523SJohannes Thumshirn free_root_extent_buffers(root); 21588c38938cSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 2159fc7cbcd4SDavid Sterba spin_lock(&root->fs_info->fs_roots_radix_lock); 21608c38938cSJosef Bacik list_del_init(&root->leak_list); 2161fc7cbcd4SDavid Sterba spin_unlock(&root->fs_info->fs_roots_radix_lock); 21628c38938cSJosef Bacik #endif 21638c38938cSJosef Bacik kfree(root); 21648c38938cSJosef Bacik } 21658c38938cSJosef Bacik } 21668c38938cSJosef Bacik 2167faa2dbf0SJosef Bacik void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2168171f6537SJosef Bacik { 2169fc7cbcd4SDavid Sterba int ret; 2170fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 2171fc7cbcd4SDavid Sterba int i; 2172171f6537SJosef Bacik 2173171f6537SJosef Bacik while (!list_empty(&fs_info->dead_roots)) { 2174fc7cbcd4SDavid Sterba gang[0] = list_entry(fs_info->dead_roots.next, 2175171f6537SJosef Bacik struct btrfs_root, root_list); 2176fc7cbcd4SDavid Sterba list_del(&gang[0]->root_list); 2177171f6537SJosef Bacik 2178fc7cbcd4SDavid Sterba if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) 2179fc7cbcd4SDavid Sterba btrfs_drop_and_free_fs_root(fs_info, gang[0]); 2180fc7cbcd4SDavid Sterba btrfs_put_root(gang[0]); 2181171f6537SJosef Bacik } 2182171f6537SJosef Bacik 2183fc7cbcd4SDavid Sterba while (1) { 2184fc7cbcd4SDavid Sterba ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2185fc7cbcd4SDavid Sterba (void **)gang, 0, 2186fc7cbcd4SDavid Sterba ARRAY_SIZE(gang)); 2187fc7cbcd4SDavid Sterba if (!ret) 2188fc7cbcd4SDavid Sterba break; 2189fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) 2190fc7cbcd4SDavid Sterba btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2191171f6537SJosef Bacik } 2192171f6537SJosef Bacik } 2193af31f5e5SChris Mason 2194638aa7edSEric Sandeen static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2195638aa7edSEric Sandeen { 2196638aa7edSEric Sandeen mutex_init(&fs_info->scrub_lock); 2197638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_running, 0); 2198638aa7edSEric Sandeen atomic_set(&fs_info->scrub_pause_req, 0); 2199638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_paused, 0); 2200638aa7edSEric Sandeen atomic_set(&fs_info->scrub_cancel_req, 0); 2201638aa7edSEric Sandeen init_waitqueue_head(&fs_info->scrub_pause_wait); 2202ff09c4caSAnand Jain refcount_set(&fs_info->scrub_workers_refcnt, 0); 2203638aa7edSEric Sandeen } 2204638aa7edSEric Sandeen 2205779a65a4SEric Sandeen static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2206779a65a4SEric Sandeen { 2207779a65a4SEric Sandeen spin_lock_init(&fs_info->balance_lock); 2208779a65a4SEric Sandeen mutex_init(&fs_info->balance_mutex); 2209779a65a4SEric Sandeen atomic_set(&fs_info->balance_pause_req, 0); 2210779a65a4SEric Sandeen atomic_set(&fs_info->balance_cancel_req, 0); 2211779a65a4SEric Sandeen fs_info->balance_ctl = NULL; 2212779a65a4SEric Sandeen init_waitqueue_head(&fs_info->balance_wait_q); 2213907d2710SDavid Sterba atomic_set(&fs_info->reloc_cancel_req, 0); 2214779a65a4SEric Sandeen } 2215779a65a4SEric Sandeen 22166bccf3abSJeff Mahoney static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info) 2217f37938e0SEric Sandeen { 22182ff7e61eSJeff Mahoney struct inode *inode = fs_info->btree_inode; 2219e256927bSJosef Bacik unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID, 2220e256927bSJosef Bacik fs_info->tree_root); 22212ff7e61eSJeff Mahoney 22222ff7e61eSJeff Mahoney inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 22232ff7e61eSJeff Mahoney set_nlink(inode, 1); 2224f37938e0SEric Sandeen /* 2225f37938e0SEric Sandeen * we set the i_size on the btree inode to the max possible int. 2226f37938e0SEric Sandeen * the real end of the address space is determined by all of 2227f37938e0SEric Sandeen * the devices in the system 2228f37938e0SEric Sandeen */ 22292ff7e61eSJeff Mahoney inode->i_size = OFFSET_MAX; 22302ff7e61eSJeff Mahoney inode->i_mapping->a_ops = &btree_aops; 2231f37938e0SEric Sandeen 22322ff7e61eSJeff Mahoney RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 223343eb5f29SQu Wenruo extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree, 2234efb0645bSJosef Bacik IO_TREE_BTREE_INODE_IO, NULL); 22352ff7e61eSJeff Mahoney extent_map_tree_init(&BTRFS_I(inode)->extent_tree); 2236f37938e0SEric Sandeen 22375c8fd99fSJosef Bacik BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root); 2238adac5584SFilipe Manana BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID; 2239adac5584SFilipe Manana BTRFS_I(inode)->location.type = 0; 2240adac5584SFilipe Manana BTRFS_I(inode)->location.offset = 0; 22412ff7e61eSJeff Mahoney set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); 2242e256927bSJosef Bacik __insert_inode_hash(inode, hash); 2243f37938e0SEric Sandeen } 2244f37938e0SEric Sandeen 2245ad618368SEric Sandeen static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2246ad618368SEric Sandeen { 2247ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2248129827e3SDavid Sterba init_rwsem(&fs_info->dev_replace.rwsem); 22497f8d236aSDavid Sterba init_waitqueue_head(&fs_info->dev_replace.replace_wait); 2250ad618368SEric Sandeen } 2251ad618368SEric Sandeen 2252f9e92e40SEric Sandeen static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2253f9e92e40SEric Sandeen { 2254f9e92e40SEric Sandeen spin_lock_init(&fs_info->qgroup_lock); 2255f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_ioctl_lock); 2256f9e92e40SEric Sandeen fs_info->qgroup_tree = RB_ROOT; 2257f9e92e40SEric Sandeen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2258f9e92e40SEric Sandeen fs_info->qgroup_seq = 1; 2259f9e92e40SEric Sandeen fs_info->qgroup_ulist = NULL; 2260d2c609b8SJeff Mahoney fs_info->qgroup_rescan_running = false; 2261011b46c3SQu Wenruo fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL; 2262f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_rescan_lock); 2263f9e92e40SEric Sandeen } 2264f9e92e40SEric Sandeen 2265d21deec5SSu Yue static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info) 22662a458198SEric Sandeen { 2267f7b885beSAnand Jain u32 max_active = fs_info->thread_pool_size; 22686f011058SDavid Sterba unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 22692a458198SEric Sandeen 22702a458198SEric Sandeen fs_info->workers = 2271a31b4a43SChristoph Hellwig btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16); 2272a31b4a43SChristoph Hellwig fs_info->hipri_workers = 2273a31b4a43SChristoph Hellwig btrfs_alloc_workqueue(fs_info, "worker-high", 2274cb001095SJeff Mahoney flags | WQ_HIGHPRI, max_active, 16); 22752a458198SEric Sandeen 22762a458198SEric Sandeen fs_info->delalloc_workers = 2277cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delalloc", 2278cb001095SJeff Mahoney flags, max_active, 2); 22792a458198SEric Sandeen 22802a458198SEric Sandeen fs_info->flush_workers = 2281cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "flush_delalloc", 2282cb001095SJeff Mahoney flags, max_active, 0); 22832a458198SEric Sandeen 22842a458198SEric Sandeen fs_info->caching_workers = 2285cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); 22862a458198SEric Sandeen 22872a458198SEric Sandeen fs_info->fixup_workers = 2288cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); 22892a458198SEric Sandeen 22902a458198SEric Sandeen fs_info->endio_workers = 2291d7b9416fSChristoph Hellwig alloc_workqueue("btrfs-endio", flags, max_active); 22922a458198SEric Sandeen fs_info->endio_meta_workers = 2293d7b9416fSChristoph Hellwig alloc_workqueue("btrfs-endio-meta", flags, max_active); 22942a458198SEric Sandeen fs_info->endio_raid56_workers = 2295d34e123dSChristoph Hellwig alloc_workqueue("btrfs-endio-raid56", flags, max_active); 2296385de0efSChristoph Hellwig fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active); 22972a458198SEric Sandeen fs_info->endio_write_workers = 2298cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "endio-write", flags, 2299cb001095SJeff Mahoney max_active, 2); 2300fed8a72dSChristoph Hellwig fs_info->compressed_write_workers = 2301fed8a72dSChristoph Hellwig alloc_workqueue("btrfs-compressed-write", flags, max_active); 23022a458198SEric Sandeen fs_info->endio_freespace_worker = 2303cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "freespace-write", flags, 2304cb001095SJeff Mahoney max_active, 0); 23052a458198SEric Sandeen fs_info->delayed_workers = 2306cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, 2307cb001095SJeff Mahoney max_active, 0); 23082a458198SEric Sandeen fs_info->qgroup_rescan_workers = 2309cb001095SJeff Mahoney btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); 2310b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers = 2311b0643e59SDennis Zhou alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1); 23122a458198SEric Sandeen 2313a31b4a43SChristoph Hellwig if (!(fs_info->workers && fs_info->hipri_workers && 2314a31b4a43SChristoph Hellwig fs_info->delalloc_workers && fs_info->flush_workers && 23152a458198SEric Sandeen fs_info->endio_workers && fs_info->endio_meta_workers && 2316fed8a72dSChristoph Hellwig fs_info->compressed_write_workers && 23172a458198SEric Sandeen fs_info->endio_write_workers && fs_info->endio_raid56_workers && 23182a458198SEric Sandeen fs_info->endio_freespace_worker && fs_info->rmw_workers && 2319f26c9238SQu Wenruo fs_info->caching_workers && fs_info->fixup_workers && 2320f26c9238SQu Wenruo fs_info->delayed_workers && fs_info->qgroup_rescan_workers && 2321b0643e59SDennis Zhou fs_info->discard_ctl.discard_workers)) { 23222a458198SEric Sandeen return -ENOMEM; 23232a458198SEric Sandeen } 23242a458198SEric Sandeen 23252a458198SEric Sandeen return 0; 23262a458198SEric Sandeen } 23272a458198SEric Sandeen 23286d97c6e3SJohannes Thumshirn static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type) 23296d97c6e3SJohannes Thumshirn { 23306d97c6e3SJohannes Thumshirn struct crypto_shash *csum_shash; 2331b4e967beSDavid Sterba const char *csum_driver = btrfs_super_csum_driver(csum_type); 23326d97c6e3SJohannes Thumshirn 2333b4e967beSDavid Sterba csum_shash = crypto_alloc_shash(csum_driver, 0, 0); 23346d97c6e3SJohannes Thumshirn 23356d97c6e3SJohannes Thumshirn if (IS_ERR(csum_shash)) { 23366d97c6e3SJohannes Thumshirn btrfs_err(fs_info, "error allocating %s hash for checksum", 2337b4e967beSDavid Sterba csum_driver); 23386d97c6e3SJohannes Thumshirn return PTR_ERR(csum_shash); 23396d97c6e3SJohannes Thumshirn } 23406d97c6e3SJohannes Thumshirn 23416d97c6e3SJohannes Thumshirn fs_info->csum_shash = csum_shash; 23426d97c6e3SJohannes Thumshirn 2343c8a5f8caSDavid Sterba btrfs_info(fs_info, "using %s (%s) checksum algorithm", 2344c8a5f8caSDavid Sterba btrfs_super_csum_name(csum_type), 2345c8a5f8caSDavid Sterba crypto_shash_driver_name(csum_shash)); 23466d97c6e3SJohannes Thumshirn return 0; 23476d97c6e3SJohannes Thumshirn } 23486d97c6e3SJohannes Thumshirn 234963443bf5SEric Sandeen static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 235063443bf5SEric Sandeen struct btrfs_fs_devices *fs_devices) 235163443bf5SEric Sandeen { 235263443bf5SEric Sandeen int ret; 235363443bf5SEric Sandeen struct btrfs_root *log_tree_root; 235463443bf5SEric Sandeen struct btrfs_super_block *disk_super = fs_info->super_copy; 235563443bf5SEric Sandeen u64 bytenr = btrfs_super_log_root(disk_super); 2356581c1760SQu Wenruo int level = btrfs_super_log_root_level(disk_super); 235763443bf5SEric Sandeen 235863443bf5SEric Sandeen if (fs_devices->rw_devices == 0) { 2359f14d104dSDavid Sterba btrfs_warn(fs_info, "log replay required on RO media"); 236063443bf5SEric Sandeen return -EIO; 236163443bf5SEric Sandeen } 236263443bf5SEric Sandeen 236396dfcb46SJosef Bacik log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, 236496dfcb46SJosef Bacik GFP_KERNEL); 236563443bf5SEric Sandeen if (!log_tree_root) 236663443bf5SEric Sandeen return -ENOMEM; 236763443bf5SEric Sandeen 23682ff7e61eSJeff Mahoney log_tree_root->node = read_tree_block(fs_info, bytenr, 23691b7ec85eSJosef Bacik BTRFS_TREE_LOG_OBJECTID, 23701b7ec85eSJosef Bacik fs_info->generation + 1, level, 23711b7ec85eSJosef Bacik NULL); 237264c043deSLiu Bo if (IS_ERR(log_tree_root->node)) { 2373f14d104dSDavid Sterba btrfs_warn(fs_info, "failed to read log tree"); 23740eeff236SLiu Bo ret = PTR_ERR(log_tree_root->node); 23758c38938cSJosef Bacik log_tree_root->node = NULL; 237600246528SJosef Bacik btrfs_put_root(log_tree_root); 23770eeff236SLiu Bo return ret; 23784eb150d6SQu Wenruo } 23794eb150d6SQu Wenruo if (!extent_buffer_uptodate(log_tree_root->node)) { 2380f14d104dSDavid Sterba btrfs_err(fs_info, "failed to read log tree"); 238100246528SJosef Bacik btrfs_put_root(log_tree_root); 238263443bf5SEric Sandeen return -EIO; 238363443bf5SEric Sandeen } 23844eb150d6SQu Wenruo 238563443bf5SEric Sandeen /* returns with log_tree_root freed on success */ 238663443bf5SEric Sandeen ret = btrfs_recover_log_trees(log_tree_root); 238763443bf5SEric Sandeen if (ret) { 23880b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 238963443bf5SEric Sandeen "Failed to recover log tree"); 239000246528SJosef Bacik btrfs_put_root(log_tree_root); 239163443bf5SEric Sandeen return ret; 239263443bf5SEric Sandeen } 239363443bf5SEric Sandeen 2394bc98a42cSDavid Howells if (sb_rdonly(fs_info->sb)) { 23956bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 239663443bf5SEric Sandeen if (ret) 239763443bf5SEric Sandeen return ret; 239863443bf5SEric Sandeen } 239963443bf5SEric Sandeen 240063443bf5SEric Sandeen return 0; 240163443bf5SEric Sandeen } 240263443bf5SEric Sandeen 2403abed4aaaSJosef Bacik static int load_global_roots_objectid(struct btrfs_root *tree_root, 2404abed4aaaSJosef Bacik struct btrfs_path *path, u64 objectid, 2405abed4aaaSJosef Bacik const char *name) 2406abed4aaaSJosef Bacik { 2407abed4aaaSJosef Bacik struct btrfs_fs_info *fs_info = tree_root->fs_info; 2408abed4aaaSJosef Bacik struct btrfs_root *root; 2409f7238e50SJosef Bacik u64 max_global_id = 0; 2410abed4aaaSJosef Bacik int ret; 2411abed4aaaSJosef Bacik struct btrfs_key key = { 2412abed4aaaSJosef Bacik .objectid = objectid, 2413abed4aaaSJosef Bacik .type = BTRFS_ROOT_ITEM_KEY, 2414abed4aaaSJosef Bacik .offset = 0, 2415abed4aaaSJosef Bacik }; 2416abed4aaaSJosef Bacik bool found = false; 2417abed4aaaSJosef Bacik 2418abed4aaaSJosef Bacik /* If we have IGNOREDATACSUMS skip loading these roots. */ 2419abed4aaaSJosef Bacik if (objectid == BTRFS_CSUM_TREE_OBJECTID && 2420abed4aaaSJosef Bacik btrfs_test_opt(fs_info, IGNOREDATACSUMS)) { 2421abed4aaaSJosef Bacik set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state); 2422abed4aaaSJosef Bacik return 0; 2423abed4aaaSJosef Bacik } 2424abed4aaaSJosef Bacik 2425abed4aaaSJosef Bacik while (1) { 2426abed4aaaSJosef Bacik ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0); 2427abed4aaaSJosef Bacik if (ret < 0) 2428abed4aaaSJosef Bacik break; 2429abed4aaaSJosef Bacik 2430abed4aaaSJosef Bacik if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 2431abed4aaaSJosef Bacik ret = btrfs_next_leaf(tree_root, path); 2432abed4aaaSJosef Bacik if (ret) { 2433abed4aaaSJosef Bacik if (ret > 0) 2434abed4aaaSJosef Bacik ret = 0; 2435abed4aaaSJosef Bacik break; 2436abed4aaaSJosef Bacik } 2437abed4aaaSJosef Bacik } 2438abed4aaaSJosef Bacik ret = 0; 2439abed4aaaSJosef Bacik 2440abed4aaaSJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 2441abed4aaaSJosef Bacik if (key.objectid != objectid) 2442abed4aaaSJosef Bacik break; 2443abed4aaaSJosef Bacik btrfs_release_path(path); 2444abed4aaaSJosef Bacik 2445f7238e50SJosef Bacik /* 2446f7238e50SJosef Bacik * Just worry about this for extent tree, it'll be the same for 2447f7238e50SJosef Bacik * everybody. 2448f7238e50SJosef Bacik */ 2449f7238e50SJosef Bacik if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 2450f7238e50SJosef Bacik max_global_id = max(max_global_id, key.offset); 2451f7238e50SJosef Bacik 2452abed4aaaSJosef Bacik found = true; 2453abed4aaaSJosef Bacik root = read_tree_root_path(tree_root, path, &key); 2454abed4aaaSJosef Bacik if (IS_ERR(root)) { 2455abed4aaaSJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) 2456abed4aaaSJosef Bacik ret = PTR_ERR(root); 2457abed4aaaSJosef Bacik break; 2458abed4aaaSJosef Bacik } 2459abed4aaaSJosef Bacik set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2460abed4aaaSJosef Bacik ret = btrfs_global_root_insert(root); 2461abed4aaaSJosef Bacik if (ret) { 2462abed4aaaSJosef Bacik btrfs_put_root(root); 2463abed4aaaSJosef Bacik break; 2464abed4aaaSJosef Bacik } 2465abed4aaaSJosef Bacik key.offset++; 2466abed4aaaSJosef Bacik } 2467abed4aaaSJosef Bacik btrfs_release_path(path); 2468abed4aaaSJosef Bacik 2469f7238e50SJosef Bacik if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 2470f7238e50SJosef Bacik fs_info->nr_global_roots = max_global_id + 1; 2471f7238e50SJosef Bacik 2472abed4aaaSJosef Bacik if (!found || ret) { 2473abed4aaaSJosef Bacik if (objectid == BTRFS_CSUM_TREE_OBJECTID) 2474abed4aaaSJosef Bacik set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state); 2475abed4aaaSJosef Bacik 2476abed4aaaSJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) 2477abed4aaaSJosef Bacik ret = ret ? ret : -ENOENT; 2478abed4aaaSJosef Bacik else 2479abed4aaaSJosef Bacik ret = 0; 2480abed4aaaSJosef Bacik btrfs_err(fs_info, "failed to load root %s", name); 2481abed4aaaSJosef Bacik } 2482abed4aaaSJosef Bacik return ret; 2483abed4aaaSJosef Bacik } 2484abed4aaaSJosef Bacik 2485abed4aaaSJosef Bacik static int load_global_roots(struct btrfs_root *tree_root) 2486abed4aaaSJosef Bacik { 2487abed4aaaSJosef Bacik struct btrfs_path *path; 2488abed4aaaSJosef Bacik int ret = 0; 2489abed4aaaSJosef Bacik 2490abed4aaaSJosef Bacik path = btrfs_alloc_path(); 2491abed4aaaSJosef Bacik if (!path) 2492abed4aaaSJosef Bacik return -ENOMEM; 2493abed4aaaSJosef Bacik 2494abed4aaaSJosef Bacik ret = load_global_roots_objectid(tree_root, path, 2495abed4aaaSJosef Bacik BTRFS_EXTENT_TREE_OBJECTID, "extent"); 2496abed4aaaSJosef Bacik if (ret) 2497abed4aaaSJosef Bacik goto out; 2498abed4aaaSJosef Bacik ret = load_global_roots_objectid(tree_root, path, 2499abed4aaaSJosef Bacik BTRFS_CSUM_TREE_OBJECTID, "csum"); 2500abed4aaaSJosef Bacik if (ret) 2501abed4aaaSJosef Bacik goto out; 2502abed4aaaSJosef Bacik if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE)) 2503abed4aaaSJosef Bacik goto out; 2504abed4aaaSJosef Bacik ret = load_global_roots_objectid(tree_root, path, 2505abed4aaaSJosef Bacik BTRFS_FREE_SPACE_TREE_OBJECTID, 2506abed4aaaSJosef Bacik "free space"); 2507abed4aaaSJosef Bacik out: 2508abed4aaaSJosef Bacik btrfs_free_path(path); 2509abed4aaaSJosef Bacik return ret; 2510abed4aaaSJosef Bacik } 2511abed4aaaSJosef Bacik 25126bccf3abSJeff Mahoney static int btrfs_read_roots(struct btrfs_fs_info *fs_info) 25134bbcaa64SEric Sandeen { 25146bccf3abSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 2515a4f3d2c4SDavid Sterba struct btrfs_root *root; 25164bbcaa64SEric Sandeen struct btrfs_key location; 25174bbcaa64SEric Sandeen int ret; 25184bbcaa64SEric Sandeen 25196bccf3abSJeff Mahoney BUG_ON(!fs_info->tree_root); 25206bccf3abSJeff Mahoney 2521abed4aaaSJosef Bacik ret = load_global_roots(tree_root); 2522abed4aaaSJosef Bacik if (ret) 2523abed4aaaSJosef Bacik return ret; 2524abed4aaaSJosef Bacik 25254bbcaa64SEric Sandeen location.type = BTRFS_ROOT_ITEM_KEY; 25264bbcaa64SEric Sandeen location.offset = 0; 25274bbcaa64SEric Sandeen 25281c56ab99SQu Wenruo if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) { 252914033b08SQu Wenruo location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID; 253014033b08SQu Wenruo root = btrfs_read_tree_root(tree_root, &location); 253114033b08SQu Wenruo if (IS_ERR(root)) { 253214033b08SQu Wenruo if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 253314033b08SQu Wenruo ret = PTR_ERR(root); 253414033b08SQu Wenruo goto out; 253514033b08SQu Wenruo } 253614033b08SQu Wenruo } else { 253714033b08SQu Wenruo set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 253814033b08SQu Wenruo fs_info->block_group_root = root; 253914033b08SQu Wenruo } 254014033b08SQu Wenruo } 254114033b08SQu Wenruo 254214033b08SQu Wenruo location.objectid = BTRFS_DEV_TREE_OBJECTID; 2543a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2544f50f4353SLiu Bo if (IS_ERR(root)) { 254542437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2546f50f4353SLiu Bo ret = PTR_ERR(root); 2547f50f4353SLiu Bo goto out; 2548f50f4353SLiu Bo } 254942437a63SJosef Bacik } else { 2550a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2551a4f3d2c4SDavid Sterba fs_info->dev_root = root; 255242437a63SJosef Bacik } 2553820a49daSJosef Bacik /* Initialize fs_info for all devices in any case */ 2554*a8d1b164SJohannes Thumshirn ret = btrfs_init_devices_late(fs_info); 2555*a8d1b164SJohannes Thumshirn if (ret) 2556*a8d1b164SJohannes Thumshirn goto out; 25574bbcaa64SEric Sandeen 2558aeb935a4SQu Wenruo /* 2559aeb935a4SQu Wenruo * This tree can share blocks with some other fs tree during relocation 2560aeb935a4SQu Wenruo * and we need a proper setup by btrfs_get_fs_root 2561aeb935a4SQu Wenruo */ 256256e9357aSDavid Sterba root = btrfs_get_fs_root(tree_root->fs_info, 256356e9357aSDavid Sterba BTRFS_DATA_RELOC_TREE_OBJECTID, true); 2564aeb935a4SQu Wenruo if (IS_ERR(root)) { 256542437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2566aeb935a4SQu Wenruo ret = PTR_ERR(root); 2567aeb935a4SQu Wenruo goto out; 2568aeb935a4SQu Wenruo } 256942437a63SJosef Bacik } else { 2570aeb935a4SQu Wenruo set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2571aeb935a4SQu Wenruo fs_info->data_reloc_root = root; 257242437a63SJosef Bacik } 2573aeb935a4SQu Wenruo 25744bbcaa64SEric Sandeen location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2575a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2576a4f3d2c4SDavid Sterba if (!IS_ERR(root)) { 2577a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2578afcdd129SJosef Bacik set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); 2579a4f3d2c4SDavid Sterba fs_info->quota_root = root; 25804bbcaa64SEric Sandeen } 25814bbcaa64SEric Sandeen 25824bbcaa64SEric Sandeen location.objectid = BTRFS_UUID_TREE_OBJECTID; 2583a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2584a4f3d2c4SDavid Sterba if (IS_ERR(root)) { 258542437a63SJosef Bacik if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2586a4f3d2c4SDavid Sterba ret = PTR_ERR(root); 25874bbcaa64SEric Sandeen if (ret != -ENOENT) 2588f50f4353SLiu Bo goto out; 258942437a63SJosef Bacik } 25904bbcaa64SEric Sandeen } else { 2591a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2592a4f3d2c4SDavid Sterba fs_info->uuid_root = root; 25934bbcaa64SEric Sandeen } 25944bbcaa64SEric Sandeen 25954bbcaa64SEric Sandeen return 0; 2596f50f4353SLiu Bo out: 2597f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", 2598f50f4353SLiu Bo location.objectid, ret); 2599f50f4353SLiu Bo return ret; 26004bbcaa64SEric Sandeen } 26014bbcaa64SEric Sandeen 2602069ec957SQu Wenruo /* 2603069ec957SQu Wenruo * Real super block validation 2604069ec957SQu Wenruo * NOTE: super csum type and incompat features will not be checked here. 2605069ec957SQu Wenruo * 2606069ec957SQu Wenruo * @sb: super block to check 2607069ec957SQu Wenruo * @mirror_num: the super block number to check its bytenr: 2608069ec957SQu Wenruo * 0 the primary (1st) sb 2609069ec957SQu Wenruo * 1, 2 2nd and 3rd backup copy 2610069ec957SQu Wenruo * -1 skip bytenr check 2611069ec957SQu Wenruo */ 2612a05d3c91SQu Wenruo int btrfs_validate_super(struct btrfs_fs_info *fs_info, 2613069ec957SQu Wenruo struct btrfs_super_block *sb, int mirror_num) 261421a852b0SQu Wenruo { 261521a852b0SQu Wenruo u64 nodesize = btrfs_super_nodesize(sb); 261621a852b0SQu Wenruo u64 sectorsize = btrfs_super_sectorsize(sb); 261721a852b0SQu Wenruo int ret = 0; 261821a852b0SQu Wenruo 261921a852b0SQu Wenruo if (btrfs_super_magic(sb) != BTRFS_MAGIC) { 262021a852b0SQu Wenruo btrfs_err(fs_info, "no valid FS found"); 262121a852b0SQu Wenruo ret = -EINVAL; 262221a852b0SQu Wenruo } 262321a852b0SQu Wenruo if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { 262421a852b0SQu Wenruo btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", 262521a852b0SQu Wenruo btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); 262621a852b0SQu Wenruo ret = -EINVAL; 262721a852b0SQu Wenruo } 262821a852b0SQu Wenruo if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 262921a852b0SQu Wenruo btrfs_err(fs_info, "tree_root level too big: %d >= %d", 263021a852b0SQu Wenruo btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 263121a852b0SQu Wenruo ret = -EINVAL; 263221a852b0SQu Wenruo } 263321a852b0SQu Wenruo if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 263421a852b0SQu Wenruo btrfs_err(fs_info, "chunk_root level too big: %d >= %d", 263521a852b0SQu Wenruo btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 263621a852b0SQu Wenruo ret = -EINVAL; 263721a852b0SQu Wenruo } 263821a852b0SQu Wenruo if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 263921a852b0SQu Wenruo btrfs_err(fs_info, "log_root level too big: %d >= %d", 264021a852b0SQu Wenruo btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 264121a852b0SQu Wenruo ret = -EINVAL; 264221a852b0SQu Wenruo } 264321a852b0SQu Wenruo 264421a852b0SQu Wenruo /* 264521a852b0SQu Wenruo * Check sectorsize and nodesize first, other check will need it. 264621a852b0SQu Wenruo * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. 264721a852b0SQu Wenruo */ 264821a852b0SQu Wenruo if (!is_power_of_2(sectorsize) || sectorsize < 4096 || 264921a852b0SQu Wenruo sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { 265021a852b0SQu Wenruo btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); 265121a852b0SQu Wenruo ret = -EINVAL; 265221a852b0SQu Wenruo } 26530bb3eb3eSQu Wenruo 26540bb3eb3eSQu Wenruo /* 26551a42daabSQu Wenruo * We only support at most two sectorsizes: 4K and PAGE_SIZE. 26561a42daabSQu Wenruo * 26571a42daabSQu Wenruo * We can support 16K sectorsize with 64K page size without problem, 26581a42daabSQu Wenruo * but such sectorsize/pagesize combination doesn't make much sense. 26591a42daabSQu Wenruo * 4K will be our future standard, PAGE_SIZE is supported from the very 26601a42daabSQu Wenruo * beginning. 26610bb3eb3eSQu Wenruo */ 26621a42daabSQu Wenruo if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) { 266321a852b0SQu Wenruo btrfs_err(fs_info, 26640bb3eb3eSQu Wenruo "sectorsize %llu not yet supported for page size %lu", 266521a852b0SQu Wenruo sectorsize, PAGE_SIZE); 266621a852b0SQu Wenruo ret = -EINVAL; 266721a852b0SQu Wenruo } 26680bb3eb3eSQu Wenruo 266921a852b0SQu Wenruo if (!is_power_of_2(nodesize) || nodesize < sectorsize || 267021a852b0SQu Wenruo nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { 267121a852b0SQu Wenruo btrfs_err(fs_info, "invalid nodesize %llu", nodesize); 267221a852b0SQu Wenruo ret = -EINVAL; 267321a852b0SQu Wenruo } 267421a852b0SQu Wenruo if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { 267521a852b0SQu Wenruo btrfs_err(fs_info, "invalid leafsize %u, should be %llu", 267621a852b0SQu Wenruo le32_to_cpu(sb->__unused_leafsize), nodesize); 267721a852b0SQu Wenruo ret = -EINVAL; 267821a852b0SQu Wenruo } 267921a852b0SQu Wenruo 268021a852b0SQu Wenruo /* Root alignment check */ 268121a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { 268221a852b0SQu Wenruo btrfs_warn(fs_info, "tree_root block unaligned: %llu", 268321a852b0SQu Wenruo btrfs_super_root(sb)); 268421a852b0SQu Wenruo ret = -EINVAL; 268521a852b0SQu Wenruo } 268621a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { 268721a852b0SQu Wenruo btrfs_warn(fs_info, "chunk_root block unaligned: %llu", 268821a852b0SQu Wenruo btrfs_super_chunk_root(sb)); 268921a852b0SQu Wenruo ret = -EINVAL; 269021a852b0SQu Wenruo } 269121a852b0SQu Wenruo if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { 269221a852b0SQu Wenruo btrfs_warn(fs_info, "log_root block unaligned: %llu", 269321a852b0SQu Wenruo btrfs_super_log_root(sb)); 269421a852b0SQu Wenruo ret = -EINVAL; 269521a852b0SQu Wenruo } 269621a852b0SQu Wenruo 2697aefd7f70SNikolay Borisov if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, 2698aefd7f70SNikolay Borisov BTRFS_FSID_SIZE)) { 2699aefd7f70SNikolay Borisov btrfs_err(fs_info, 2700aefd7f70SNikolay Borisov "superblock fsid doesn't match fsid of fs_devices: %pU != %pU", 2701aefd7f70SNikolay Borisov fs_info->super_copy->fsid, fs_info->fs_devices->fsid); 2702aefd7f70SNikolay Borisov ret = -EINVAL; 2703aefd7f70SNikolay Borisov } 2704aefd7f70SNikolay Borisov 2705aefd7f70SNikolay Borisov if (btrfs_fs_incompat(fs_info, METADATA_UUID) && 2706aefd7f70SNikolay Borisov memcmp(fs_info->fs_devices->metadata_uuid, 2707aefd7f70SNikolay Borisov fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) { 2708aefd7f70SNikolay Borisov btrfs_err(fs_info, 2709aefd7f70SNikolay Borisov "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU", 2710aefd7f70SNikolay Borisov fs_info->super_copy->metadata_uuid, 2711aefd7f70SNikolay Borisov fs_info->fs_devices->metadata_uuid); 2712aefd7f70SNikolay Borisov ret = -EINVAL; 2713aefd7f70SNikolay Borisov } 2714aefd7f70SNikolay Borisov 27151c56ab99SQu Wenruo /* 27161c56ab99SQu Wenruo * Artificial requirement for block-group-tree to force newer features 27171c56ab99SQu Wenruo * (free-space-tree, no-holes) so the test matrix is smaller. 27181c56ab99SQu Wenruo */ 27191c56ab99SQu Wenruo if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) && 27201c56ab99SQu Wenruo (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) || 27211c56ab99SQu Wenruo !btrfs_fs_incompat(fs_info, NO_HOLES))) { 27221c56ab99SQu Wenruo btrfs_err(fs_info, 27231c56ab99SQu Wenruo "block-group-tree feature requires fres-space-tree and no-holes"); 27241c56ab99SQu Wenruo ret = -EINVAL; 27251c56ab99SQu Wenruo } 27261c56ab99SQu Wenruo 2727de37aa51SNikolay Borisov if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, 27287239ff4bSNikolay Borisov BTRFS_FSID_SIZE) != 0) { 272921a852b0SQu Wenruo btrfs_err(fs_info, 27307239ff4bSNikolay Borisov "dev_item UUID does not match metadata fsid: %pU != %pU", 2731de37aa51SNikolay Borisov fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); 273221a852b0SQu Wenruo ret = -EINVAL; 273321a852b0SQu Wenruo } 273421a852b0SQu Wenruo 273521a852b0SQu Wenruo /* 273621a852b0SQu Wenruo * Hint to catch really bogus numbers, bitflips or so, more exact checks are 273721a852b0SQu Wenruo * done later 273821a852b0SQu Wenruo */ 273921a852b0SQu Wenruo if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { 274021a852b0SQu Wenruo btrfs_err(fs_info, "bytes_used is too small %llu", 274121a852b0SQu Wenruo btrfs_super_bytes_used(sb)); 274221a852b0SQu Wenruo ret = -EINVAL; 274321a852b0SQu Wenruo } 274421a852b0SQu Wenruo if (!is_power_of_2(btrfs_super_stripesize(sb))) { 274521a852b0SQu Wenruo btrfs_err(fs_info, "invalid stripesize %u", 274621a852b0SQu Wenruo btrfs_super_stripesize(sb)); 274721a852b0SQu Wenruo ret = -EINVAL; 274821a852b0SQu Wenruo } 274921a852b0SQu Wenruo if (btrfs_super_num_devices(sb) > (1UL << 31)) 275021a852b0SQu Wenruo btrfs_warn(fs_info, "suspicious number of devices: %llu", 275121a852b0SQu Wenruo btrfs_super_num_devices(sb)); 275221a852b0SQu Wenruo if (btrfs_super_num_devices(sb) == 0) { 275321a852b0SQu Wenruo btrfs_err(fs_info, "number of devices is 0"); 275421a852b0SQu Wenruo ret = -EINVAL; 275521a852b0SQu Wenruo } 275621a852b0SQu Wenruo 2757069ec957SQu Wenruo if (mirror_num >= 0 && 2758069ec957SQu Wenruo btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { 275921a852b0SQu Wenruo btrfs_err(fs_info, "super offset mismatch %llu != %u", 276021a852b0SQu Wenruo btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 276121a852b0SQu Wenruo ret = -EINVAL; 276221a852b0SQu Wenruo } 276321a852b0SQu Wenruo 276421a852b0SQu Wenruo /* 276521a852b0SQu Wenruo * Obvious sys_chunk_array corruptions, it must hold at least one key 276621a852b0SQu Wenruo * and one chunk 276721a852b0SQu Wenruo */ 276821a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 276921a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too big %u > %u", 277021a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 277121a852b0SQu Wenruo BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 277221a852b0SQu Wenruo ret = -EINVAL; 277321a852b0SQu Wenruo } 277421a852b0SQu Wenruo if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 277521a852b0SQu Wenruo + sizeof(struct btrfs_chunk)) { 277621a852b0SQu Wenruo btrfs_err(fs_info, "system chunk array too small %u < %zu", 277721a852b0SQu Wenruo btrfs_super_sys_array_size(sb), 277821a852b0SQu Wenruo sizeof(struct btrfs_disk_key) 277921a852b0SQu Wenruo + sizeof(struct btrfs_chunk)); 278021a852b0SQu Wenruo ret = -EINVAL; 278121a852b0SQu Wenruo } 278221a852b0SQu Wenruo 278321a852b0SQu Wenruo /* 278421a852b0SQu Wenruo * The generation is a global counter, we'll trust it more than the others 278521a852b0SQu Wenruo * but it's still possible that it's the one that's wrong. 278621a852b0SQu Wenruo */ 278721a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 278821a852b0SQu Wenruo btrfs_warn(fs_info, 278921a852b0SQu Wenruo "suspicious: generation < chunk_root_generation: %llu < %llu", 279021a852b0SQu Wenruo btrfs_super_generation(sb), 279121a852b0SQu Wenruo btrfs_super_chunk_root_generation(sb)); 279221a852b0SQu Wenruo if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 279321a852b0SQu Wenruo && btrfs_super_cache_generation(sb) != (u64)-1) 279421a852b0SQu Wenruo btrfs_warn(fs_info, 279521a852b0SQu Wenruo "suspicious: generation < cache_generation: %llu < %llu", 279621a852b0SQu Wenruo btrfs_super_generation(sb), 279721a852b0SQu Wenruo btrfs_super_cache_generation(sb)); 279821a852b0SQu Wenruo 279921a852b0SQu Wenruo return ret; 280021a852b0SQu Wenruo } 280121a852b0SQu Wenruo 2802069ec957SQu Wenruo /* 2803069ec957SQu Wenruo * Validation of super block at mount time. 2804069ec957SQu Wenruo * Some checks already done early at mount time, like csum type and incompat 2805069ec957SQu Wenruo * flags will be skipped. 2806069ec957SQu Wenruo */ 2807069ec957SQu Wenruo static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) 2808069ec957SQu Wenruo { 2809a05d3c91SQu Wenruo return btrfs_validate_super(fs_info, fs_info->super_copy, 0); 2810069ec957SQu Wenruo } 2811069ec957SQu Wenruo 281275cb857dSQu Wenruo /* 281375cb857dSQu Wenruo * Validation of super block at write time. 281475cb857dSQu Wenruo * Some checks like bytenr check will be skipped as their values will be 281575cb857dSQu Wenruo * overwritten soon. 281675cb857dSQu Wenruo * Extra checks like csum type and incompat flags will be done here. 281775cb857dSQu Wenruo */ 281875cb857dSQu Wenruo static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, 281975cb857dSQu Wenruo struct btrfs_super_block *sb) 282075cb857dSQu Wenruo { 282175cb857dSQu Wenruo int ret; 282275cb857dSQu Wenruo 2823a05d3c91SQu Wenruo ret = btrfs_validate_super(fs_info, sb, -1); 282475cb857dSQu Wenruo if (ret < 0) 282575cb857dSQu Wenruo goto out; 2826e7e16f48SJohannes Thumshirn if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) { 282775cb857dSQu Wenruo ret = -EUCLEAN; 282875cb857dSQu Wenruo btrfs_err(fs_info, "invalid csum type, has %u want %u", 282975cb857dSQu Wenruo btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); 283075cb857dSQu Wenruo goto out; 283175cb857dSQu Wenruo } 283275cb857dSQu Wenruo if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 283375cb857dSQu Wenruo ret = -EUCLEAN; 283475cb857dSQu Wenruo btrfs_err(fs_info, 283575cb857dSQu Wenruo "invalid incompat flags, has 0x%llx valid mask 0x%llx", 283675cb857dSQu Wenruo btrfs_super_incompat_flags(sb), 283775cb857dSQu Wenruo (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); 283875cb857dSQu Wenruo goto out; 283975cb857dSQu Wenruo } 284075cb857dSQu Wenruo out: 284175cb857dSQu Wenruo if (ret < 0) 284275cb857dSQu Wenruo btrfs_err(fs_info, 284375cb857dSQu Wenruo "super block corruption detected before writing it to disk"); 284475cb857dSQu Wenruo return ret; 284575cb857dSQu Wenruo } 284675cb857dSQu Wenruo 2847bd676446SJosef Bacik static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level) 2848bd676446SJosef Bacik { 2849bd676446SJosef Bacik int ret = 0; 2850bd676446SJosef Bacik 2851bd676446SJosef Bacik root->node = read_tree_block(root->fs_info, bytenr, 2852bd676446SJosef Bacik root->root_key.objectid, gen, level, NULL); 2853bd676446SJosef Bacik if (IS_ERR(root->node)) { 2854bd676446SJosef Bacik ret = PTR_ERR(root->node); 2855bd676446SJosef Bacik root->node = NULL; 28564eb150d6SQu Wenruo return ret; 28574eb150d6SQu Wenruo } 28584eb150d6SQu Wenruo if (!extent_buffer_uptodate(root->node)) { 2859bd676446SJosef Bacik free_extent_buffer(root->node); 2860bd676446SJosef Bacik root->node = NULL; 28614eb150d6SQu Wenruo return -EIO; 2862bd676446SJosef Bacik } 2863bd676446SJosef Bacik 2864bd676446SJosef Bacik btrfs_set_root_node(&root->root_item, root->node); 2865bd676446SJosef Bacik root->commit_root = btrfs_root_node(root); 2866bd676446SJosef Bacik btrfs_set_root_refs(&root->root_item, 1); 2867bd676446SJosef Bacik return ret; 2868bd676446SJosef Bacik } 2869bd676446SJosef Bacik 2870bd676446SJosef Bacik static int load_important_roots(struct btrfs_fs_info *fs_info) 2871bd676446SJosef Bacik { 2872bd676446SJosef Bacik struct btrfs_super_block *sb = fs_info->super_copy; 2873bd676446SJosef Bacik u64 gen, bytenr; 2874bd676446SJosef Bacik int level, ret; 2875bd676446SJosef Bacik 2876bd676446SJosef Bacik bytenr = btrfs_super_root(sb); 2877bd676446SJosef Bacik gen = btrfs_super_generation(sb); 2878bd676446SJosef Bacik level = btrfs_super_root_level(sb); 2879bd676446SJosef Bacik ret = load_super_root(fs_info->tree_root, bytenr, gen, level); 28809c54e80dSJosef Bacik if (ret) { 2881bd676446SJosef Bacik btrfs_warn(fs_info, "couldn't read tree root"); 2882bd676446SJosef Bacik return ret; 2883bd676446SJosef Bacik } 28849c54e80dSJosef Bacik return 0; 28859c54e80dSJosef Bacik } 28869c54e80dSJosef Bacik 28876ef108ddSNikolay Borisov static int __cold init_tree_roots(struct btrfs_fs_info *fs_info) 2888b8522a1eSNikolay Borisov { 28896ef108ddSNikolay Borisov int backup_index = find_newest_super_backup(fs_info); 2890b8522a1eSNikolay Borisov struct btrfs_super_block *sb = fs_info->super_copy; 2891b8522a1eSNikolay Borisov struct btrfs_root *tree_root = fs_info->tree_root; 2892b8522a1eSNikolay Borisov bool handle_error = false; 2893b8522a1eSNikolay Borisov int ret = 0; 2894b8522a1eSNikolay Borisov int i; 2895b8522a1eSNikolay Borisov 2896b8522a1eSNikolay Borisov for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 2897b8522a1eSNikolay Borisov if (handle_error) { 2898b8522a1eSNikolay Borisov if (!IS_ERR(tree_root->node)) 2899b8522a1eSNikolay Borisov free_extent_buffer(tree_root->node); 2900b8522a1eSNikolay Borisov tree_root->node = NULL; 2901b8522a1eSNikolay Borisov 2902b8522a1eSNikolay Borisov if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) 2903b8522a1eSNikolay Borisov break; 2904b8522a1eSNikolay Borisov 2905b8522a1eSNikolay Borisov free_root_pointers(fs_info, 0); 2906b8522a1eSNikolay Borisov 2907b8522a1eSNikolay Borisov /* 2908b8522a1eSNikolay Borisov * Don't use the log in recovery mode, it won't be 2909b8522a1eSNikolay Borisov * valid 2910b8522a1eSNikolay Borisov */ 2911b8522a1eSNikolay Borisov btrfs_set_super_log_root(sb, 0); 2912b8522a1eSNikolay Borisov 2913b8522a1eSNikolay Borisov /* We can't trust the free space cache either */ 2914b8522a1eSNikolay Borisov btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 2915b8522a1eSNikolay Borisov 2916b8522a1eSNikolay Borisov ret = read_backup_root(fs_info, i); 29176ef108ddSNikolay Borisov backup_index = ret; 2918b8522a1eSNikolay Borisov if (ret < 0) 2919b8522a1eSNikolay Borisov return ret; 2920b8522a1eSNikolay Borisov } 2921b8522a1eSNikolay Borisov 2922bd676446SJosef Bacik ret = load_important_roots(fs_info); 2923bd676446SJosef Bacik if (ret) { 2924217f5004SNikolay Borisov handle_error = true; 2925b8522a1eSNikolay Borisov continue; 2926b8522a1eSNikolay Borisov } 2927b8522a1eSNikolay Borisov 2928336a0d8dSNikolay Borisov /* 2929336a0d8dSNikolay Borisov * No need to hold btrfs_root::objectid_mutex since the fs 2930336a0d8dSNikolay Borisov * hasn't been fully initialised and we are the only user 2931336a0d8dSNikolay Borisov */ 2932453e4873SNikolay Borisov ret = btrfs_init_root_free_objectid(tree_root); 2933b8522a1eSNikolay Borisov if (ret < 0) { 2934b8522a1eSNikolay Borisov handle_error = true; 2935b8522a1eSNikolay Borisov continue; 2936b8522a1eSNikolay Borisov } 2937b8522a1eSNikolay Borisov 29386b8fad57SNikolay Borisov ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID); 2939b8522a1eSNikolay Borisov 2940b8522a1eSNikolay Borisov ret = btrfs_read_roots(fs_info); 2941b8522a1eSNikolay Borisov if (ret < 0) { 2942b8522a1eSNikolay Borisov handle_error = true; 2943b8522a1eSNikolay Borisov continue; 2944b8522a1eSNikolay Borisov } 2945b8522a1eSNikolay Borisov 2946b8522a1eSNikolay Borisov /* All successful */ 2947bd676446SJosef Bacik fs_info->generation = btrfs_header_generation(tree_root->node); 2948bd676446SJosef Bacik fs_info->last_trans_committed = fs_info->generation; 2949d96b3424SFilipe Manana fs_info->last_reloc_trans = 0; 29506ef108ddSNikolay Borisov 29516ef108ddSNikolay Borisov /* Always begin writing backup roots after the one being used */ 29526ef108ddSNikolay Borisov if (backup_index < 0) { 29536ef108ddSNikolay Borisov fs_info->backup_root_index = 0; 29546ef108ddSNikolay Borisov } else { 29556ef108ddSNikolay Borisov fs_info->backup_root_index = backup_index + 1; 29566ef108ddSNikolay Borisov fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS; 29576ef108ddSNikolay Borisov } 2958b8522a1eSNikolay Borisov break; 2959b8522a1eSNikolay Borisov } 2960b8522a1eSNikolay Borisov 2961b8522a1eSNikolay Borisov return ret; 2962b8522a1eSNikolay Borisov } 2963b8522a1eSNikolay Borisov 29648260edbaSJosef Bacik void btrfs_init_fs_info(struct btrfs_fs_info *fs_info) 2965eb60ceacSChris Mason { 2966fc7cbcd4SDavid Sterba INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 296701cd3909SDavid Sterba INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 29688fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 2969facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 297024bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 2971eb73c1b7SMiao Xie INIT_LIST_HEAD(&fs_info->delalloc_roots); 297211833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 2973eb73c1b7SMiao Xie spin_lock_init(&fs_info->delalloc_root_lock); 2974a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 2975fc7cbcd4SDavid Sterba spin_lock_init(&fs_info->fs_roots_radix_lock); 297624bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 29774cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 2978ceda0864SMiao Xie spin_lock_init(&fs_info->super_lock); 2979f28491e0SJosef Bacik spin_lock_init(&fs_info->buffer_lock); 298047ab2a6cSJosef Bacik spin_lock_init(&fs_info->unused_bgs_lock); 298140ab3be1SNaohiro Aota spin_lock_init(&fs_info->treelog_bg_lock); 2982afba2bc0SNaohiro Aota spin_lock_init(&fs_info->zone_active_bgs_lock); 2983c2707a25SJohannes Thumshirn spin_lock_init(&fs_info->relocation_bg_lock); 2984f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 2985abed4aaaSJosef Bacik rwlock_init(&fs_info->global_root_lock); 2986d7c15171SZhao Lei mutex_init(&fs_info->unused_bg_unpin_mutex); 2987f3372065SJohannes Thumshirn mutex_init(&fs_info->reclaim_bgs_lock); 29887585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 2989573bfb72SMiao Xie mutex_init(&fs_info->delalloc_root_mutex); 29900bc09ca1SNaohiro Aota mutex_init(&fs_info->zoned_meta_io_lock); 29915f0addf7SNaohiro Aota mutex_init(&fs_info->zoned_data_reloc_io_lock); 2992de98ced9SMiao Xie seqlock_init(&fs_info->profiles_lock); 299319c00ddcSChris Mason 2994e1489b4fSIoannis Angelakopoulos btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers); 29955a9ba670SIoannis Angelakopoulos btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters); 29968b53779eSIoannis Angelakopoulos btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered); 29975f4403e1SIoannis Angelakopoulos btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent); 29983e738c53SIoannis Angelakopoulos btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_start, 29993e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_COMMIT_START); 30003e738c53SIoannis Angelakopoulos btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked, 30013e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_UNBLOCKED); 30023e738c53SIoannis Angelakopoulos btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed, 30033e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 30043e738c53SIoannis Angelakopoulos btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed, 30053e738c53SIoannis Angelakopoulos BTRFS_LOCKDEP_TRANS_COMPLETED); 3006e1489b4fSIoannis Angelakopoulos 30070b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 30086324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 3009f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 301047ab2a6cSJosef Bacik INIT_LIST_HEAD(&fs_info->unused_bgs); 301118bb8bbfSJohannes Thumshirn INIT_LIST_HEAD(&fs_info->reclaim_bgs); 3012afba2bc0SNaohiro Aota INIT_LIST_HEAD(&fs_info->zone_active_bgs); 3013bd647ce3SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 3014bd647ce3SJosef Bacik INIT_LIST_HEAD(&fs_info->allocated_roots); 30153fd63727SJosef Bacik INIT_LIST_HEAD(&fs_info->allocated_ebs); 30163fd63727SJosef Bacik spin_lock_init(&fs_info->eb_leak_lock); 3017bd647ce3SJosef Bacik #endif 3018c8bf1b67SDavid Sterba extent_map_tree_init(&fs_info->mapping_tree); 301966d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 302066d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 302166d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 302266d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 302366d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 302466d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 302566d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 3026ba2c4d4eSJosef Bacik btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, 3027ba2c4d4eSJosef Bacik BTRFS_BLOCK_RSV_DELREFS); 3028ba2c4d4eSJosef Bacik 3029771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 30304cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 3031034f784dSJosef Bacik atomic_set(&fs_info->nr_delayed_iputs, 0); 3032fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 3033abed4aaaSJosef Bacik fs_info->global_root_tree = RB_ROOT; 303495ac567aSFilipe David Borba Manana fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 30359ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 30364cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 3037a5ed45f8SNikolay Borisov atomic64_set(&fs_info->free_chunk_space, 0); 3038f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 30398b87dc17SDavid Sterba fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 3040f8c269d7SDavid Sterba fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 3041fd708b81SJosef Bacik btrfs_init_ref_verify(fs_info); 3042c8b97818SChris Mason 3043b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 3044b34b086cSChris Mason num_online_cpus() + 2, 8); 30450afbaf8cSChris Mason 3046199c2a9cSMiao Xie INIT_LIST_HEAD(&fs_info->ordered_roots); 3047199c2a9cSMiao Xie spin_lock_init(&fs_info->ordered_root_lock); 304869fe2d75SJosef Bacik 3049638aa7edSEric Sandeen btrfs_init_scrub(fs_info); 305021adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 305121adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 305221adbd5cSStefan Behrens #endif 3053779a65a4SEric Sandeen btrfs_init_balance(fs_info); 305457056740SJosef Bacik btrfs_init_async_reclaim_work(fs_info); 3055a2de733cSArne Jansen 305616b0c258SFilipe Manana rwlock_init(&fs_info->block_group_cache_lock); 305708dddb29SFilipe Manana fs_info->block_group_cache_tree = RB_ROOT_CACHED; 30580f9dd46cSJosef Bacik 3059fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &fs_info->excluded_extents, 3060fe119a6eSNikolay Borisov IO_TREE_FS_EXCLUDED_EXTENTS, NULL); 306139279cc3SChris Mason 30625a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 3063e02119d5SChris Mason mutex_init(&fs_info->tree_log_mutex); 3064925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 3065a74a4b97SChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 3066a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 30671bbc621eSChris Mason mutex_init(&fs_info->ro_block_group_mutex); 30689e351cc8SJosef Bacik init_rwsem(&fs_info->commit_root_sem); 3069c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 307076dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 3071803b2f54SStefan Behrens sema_init(&fs_info->uuid_tree_rescan_sem, 1); 3072fa9c0d79SChris Mason 3073ad618368SEric Sandeen btrfs_init_dev_replace_locks(fs_info); 3074f9e92e40SEric Sandeen btrfs_init_qgroup(fs_info); 3075b0643e59SDennis Zhou btrfs_discard_init(fs_info); 3076416ac51dSArne Jansen 3077fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 3078fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 3079fa9c0d79SChris Mason 3080e6dcd2dcSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 3081f9295749SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 3082bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 30834854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 3084034f784dSJosef Bacik init_waitqueue_head(&fs_info->delayed_iputs_wait); 30853768f368SChris Mason 3086da17066cSJeff Mahoney /* Usable values until the real ones are cached from the superblock */ 3087da17066cSJeff Mahoney fs_info->nodesize = 4096; 3088da17066cSJeff Mahoney fs_info->sectorsize = 4096; 3089ab108d99SDavid Sterba fs_info->sectorsize_bits = ilog2(4096); 3090da17066cSJeff Mahoney fs_info->stripesize = 4096; 3091da17066cSJeff Mahoney 3092f7b12a62SNaohiro Aota fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE; 3093f7b12a62SNaohiro Aota 3094eede2bf3SOmar Sandoval spin_lock_init(&fs_info->swapfile_pins_lock); 3095eede2bf3SOmar Sandoval fs_info->swapfile_pins = RB_ROOT; 3096eede2bf3SOmar Sandoval 309718bb8bbfSJohannes Thumshirn fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH; 309818bb8bbfSJohannes Thumshirn INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work); 30998260edbaSJosef Bacik } 31008260edbaSJosef Bacik 31018260edbaSJosef Bacik static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb) 31028260edbaSJosef Bacik { 31038260edbaSJosef Bacik int ret; 31048260edbaSJosef Bacik 31058260edbaSJosef Bacik fs_info->sb = sb; 31068260edbaSJosef Bacik sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; 31078260edbaSJosef Bacik sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); 31089e967495SFilipe Manana 31095deb17e1SJosef Bacik ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL); 3110ae18c37aSJosef Bacik if (ret) 3111ae18c37aSJosef Bacik return ret; 3112ae18c37aSJosef Bacik 3113ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 3114ae18c37aSJosef Bacik if (ret) 3115c75e8394SJosef Bacik return ret; 3116ae18c37aSJosef Bacik 3117ae18c37aSJosef Bacik fs_info->dirty_metadata_batch = PAGE_SIZE * 3118ae18c37aSJosef Bacik (1 + ilog2(nr_cpu_ids)); 3119ae18c37aSJosef Bacik 3120ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 3121ae18c37aSJosef Bacik if (ret) 3122c75e8394SJosef Bacik return ret; 3123ae18c37aSJosef Bacik 3124ae18c37aSJosef Bacik ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, 3125ae18c37aSJosef Bacik GFP_KERNEL); 3126ae18c37aSJosef Bacik if (ret) 3127c75e8394SJosef Bacik return ret; 3128ae18c37aSJosef Bacik 3129ae18c37aSJosef Bacik fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 3130ae18c37aSJosef Bacik GFP_KERNEL); 3131c75e8394SJosef Bacik if (!fs_info->delayed_root) 3132c75e8394SJosef Bacik return -ENOMEM; 3133ae18c37aSJosef Bacik btrfs_init_delayed_root(fs_info->delayed_root); 3134ae18c37aSJosef Bacik 3135a0a1db70SFilipe Manana if (sb_rdonly(sb)) 3136a0a1db70SFilipe Manana set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state); 3137a0a1db70SFilipe Manana 3138c75e8394SJosef Bacik return btrfs_alloc_stripe_hash_table(fs_info); 3139ae18c37aSJosef Bacik } 3140ae18c37aSJosef Bacik 314197f4dd09SNikolay Borisov static int btrfs_uuid_rescan_kthread(void *data) 314297f4dd09SNikolay Borisov { 31430d031dc4SYu Zhe struct btrfs_fs_info *fs_info = data; 314497f4dd09SNikolay Borisov int ret; 314597f4dd09SNikolay Borisov 314697f4dd09SNikolay Borisov /* 314797f4dd09SNikolay Borisov * 1st step is to iterate through the existing UUID tree and 314897f4dd09SNikolay Borisov * to delete all entries that contain outdated data. 314997f4dd09SNikolay Borisov * 2nd step is to add all missing entries to the UUID tree. 315097f4dd09SNikolay Borisov */ 315197f4dd09SNikolay Borisov ret = btrfs_uuid_tree_iterate(fs_info); 315297f4dd09SNikolay Borisov if (ret < 0) { 3153c94bec2cSJosef Bacik if (ret != -EINTR) 3154c94bec2cSJosef Bacik btrfs_warn(fs_info, "iterating uuid_tree failed %d", 3155c94bec2cSJosef Bacik ret); 315697f4dd09SNikolay Borisov up(&fs_info->uuid_tree_rescan_sem); 315797f4dd09SNikolay Borisov return ret; 315897f4dd09SNikolay Borisov } 315997f4dd09SNikolay Borisov return btrfs_uuid_scan_kthread(data); 316097f4dd09SNikolay Borisov } 316197f4dd09SNikolay Borisov 316297f4dd09SNikolay Borisov static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) 316397f4dd09SNikolay Borisov { 316497f4dd09SNikolay Borisov struct task_struct *task; 316597f4dd09SNikolay Borisov 316697f4dd09SNikolay Borisov down(&fs_info->uuid_tree_rescan_sem); 316797f4dd09SNikolay Borisov task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); 316897f4dd09SNikolay Borisov if (IS_ERR(task)) { 316997f4dd09SNikolay Borisov /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 317097f4dd09SNikolay Borisov btrfs_warn(fs_info, "failed to start uuid_rescan task"); 317197f4dd09SNikolay Borisov up(&fs_info->uuid_tree_rescan_sem); 317297f4dd09SNikolay Borisov return PTR_ERR(task); 317397f4dd09SNikolay Borisov } 317497f4dd09SNikolay Borisov 317597f4dd09SNikolay Borisov return 0; 317697f4dd09SNikolay Borisov } 317797f4dd09SNikolay Borisov 317844c0ca21SBoris Burkov /* 31798cd29088SBoris Burkov * Some options only have meaning at mount time and shouldn't persist across 31808cd29088SBoris Burkov * remounts, or be displayed. Clear these at the end of mount and remount 31818cd29088SBoris Burkov * code paths. 31828cd29088SBoris Burkov */ 31838cd29088SBoris Burkov void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info) 31848cd29088SBoris Burkov { 31858cd29088SBoris Burkov btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT); 31868b228324SBoris Burkov btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE); 31878cd29088SBoris Burkov } 31888cd29088SBoris Burkov 31898cd29088SBoris Burkov /* 319044c0ca21SBoris Burkov * Mounting logic specific to read-write file systems. Shared by open_ctree 319144c0ca21SBoris Burkov * and btrfs_remount when remounting from read-only to read-write. 319244c0ca21SBoris Burkov */ 319344c0ca21SBoris Burkov int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info) 319444c0ca21SBoris Burkov { 319544c0ca21SBoris Burkov int ret; 319694846229SBoris Burkov const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE); 31978b228324SBoris Burkov bool clear_free_space_tree = false; 31988b228324SBoris Burkov 31998b228324SBoris Burkov if (btrfs_test_opt(fs_info, CLEAR_CACHE) && 32008b228324SBoris Burkov btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 32018b228324SBoris Burkov clear_free_space_tree = true; 32028b228324SBoris Burkov } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 32038b228324SBoris Burkov !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { 32048b228324SBoris Burkov btrfs_warn(fs_info, "free space tree is invalid"); 32058b228324SBoris Burkov clear_free_space_tree = true; 32068b228324SBoris Burkov } 32078b228324SBoris Burkov 32088b228324SBoris Burkov if (clear_free_space_tree) { 32098b228324SBoris Burkov btrfs_info(fs_info, "clearing free space tree"); 32108b228324SBoris Burkov ret = btrfs_clear_free_space_tree(fs_info); 32118b228324SBoris Burkov if (ret) { 32128b228324SBoris Burkov btrfs_warn(fs_info, 32138b228324SBoris Burkov "failed to clear free space tree: %d", ret); 32148b228324SBoris Burkov goto out; 32158b228324SBoris Burkov } 32168b228324SBoris Burkov } 321744c0ca21SBoris Burkov 32188d488a8cSFilipe Manana /* 32198d488a8cSFilipe Manana * btrfs_find_orphan_roots() is responsible for finding all the dead 32208d488a8cSFilipe Manana * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load 3221fc7cbcd4SDavid Sterba * them into the fs_info->fs_roots_radix tree. This must be done before 32228d488a8cSFilipe Manana * calling btrfs_orphan_cleanup() on the tree root. If we don't do it 32238d488a8cSFilipe Manana * first, then btrfs_orphan_cleanup() will delete a dead root's orphan 32248d488a8cSFilipe Manana * item before the root's tree is deleted - this means that if we unmount 32258d488a8cSFilipe Manana * or crash before the deletion completes, on the next mount we will not 32268d488a8cSFilipe Manana * delete what remains of the tree because the orphan item does not 32278d488a8cSFilipe Manana * exists anymore, which is what tells us we have a pending deletion. 32288d488a8cSFilipe Manana */ 32298d488a8cSFilipe Manana ret = btrfs_find_orphan_roots(fs_info); 32308d488a8cSFilipe Manana if (ret) 32318d488a8cSFilipe Manana goto out; 32328d488a8cSFilipe Manana 323344c0ca21SBoris Burkov ret = btrfs_cleanup_fs_roots(fs_info); 323444c0ca21SBoris Burkov if (ret) 323544c0ca21SBoris Burkov goto out; 323644c0ca21SBoris Burkov 32378f1c21d7SBoris Burkov down_read(&fs_info->cleanup_work_sem); 32388f1c21d7SBoris Burkov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 32398f1c21d7SBoris Burkov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 32408f1c21d7SBoris Burkov up_read(&fs_info->cleanup_work_sem); 32418f1c21d7SBoris Burkov goto out; 32428f1c21d7SBoris Burkov } 32438f1c21d7SBoris Burkov up_read(&fs_info->cleanup_work_sem); 32448f1c21d7SBoris Burkov 324544c0ca21SBoris Burkov mutex_lock(&fs_info->cleaner_mutex); 32467eefae6bSJosef Bacik ret = btrfs_recover_relocation(fs_info); 324744c0ca21SBoris Burkov mutex_unlock(&fs_info->cleaner_mutex); 324844c0ca21SBoris Burkov if (ret < 0) { 324944c0ca21SBoris Burkov btrfs_warn(fs_info, "failed to recover relocation: %d", ret); 325044c0ca21SBoris Burkov goto out; 325144c0ca21SBoris Burkov } 325244c0ca21SBoris Burkov 32535011139aSBoris Burkov if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && 32545011139aSBoris Burkov !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 32555011139aSBoris Burkov btrfs_info(fs_info, "creating free space tree"); 32565011139aSBoris Burkov ret = btrfs_create_free_space_tree(fs_info); 32575011139aSBoris Burkov if (ret) { 32585011139aSBoris Burkov btrfs_warn(fs_info, 32595011139aSBoris Burkov "failed to create free space tree: %d", ret); 32605011139aSBoris Burkov goto out; 32615011139aSBoris Burkov } 32625011139aSBoris Burkov } 32635011139aSBoris Burkov 326494846229SBoris Burkov if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) { 326594846229SBoris Burkov ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt); 326694846229SBoris Burkov if (ret) 326794846229SBoris Burkov goto out; 326894846229SBoris Burkov } 326994846229SBoris Burkov 327044c0ca21SBoris Burkov ret = btrfs_resume_balance_async(fs_info); 327144c0ca21SBoris Burkov if (ret) 327244c0ca21SBoris Burkov goto out; 327344c0ca21SBoris Burkov 327444c0ca21SBoris Burkov ret = btrfs_resume_dev_replace_async(fs_info); 327544c0ca21SBoris Burkov if (ret) { 327644c0ca21SBoris Burkov btrfs_warn(fs_info, "failed to resume dev_replace"); 327744c0ca21SBoris Burkov goto out; 327844c0ca21SBoris Burkov } 327944c0ca21SBoris Burkov 328044c0ca21SBoris Burkov btrfs_qgroup_rescan_resume(fs_info); 328144c0ca21SBoris Burkov 328244c0ca21SBoris Burkov if (!fs_info->uuid_root) { 328344c0ca21SBoris Burkov btrfs_info(fs_info, "creating UUID tree"); 328444c0ca21SBoris Burkov ret = btrfs_create_uuid_tree(fs_info); 328544c0ca21SBoris Burkov if (ret) { 328644c0ca21SBoris Burkov btrfs_warn(fs_info, 328744c0ca21SBoris Burkov "failed to create the UUID tree %d", ret); 328844c0ca21SBoris Burkov goto out; 328944c0ca21SBoris Burkov } 329044c0ca21SBoris Burkov } 329144c0ca21SBoris Burkov 329244c0ca21SBoris Burkov out: 329344c0ca21SBoris Burkov return ret; 329444c0ca21SBoris Burkov } 329544c0ca21SBoris Burkov 3296d7f67ac9SQu Wenruo /* 3297d7f67ac9SQu Wenruo * Do various sanity and dependency checks of different features. 3298d7f67ac9SQu Wenruo * 3299d7f67ac9SQu Wenruo * This is the place for less strict checks (like for subpage or artificial 3300d7f67ac9SQu Wenruo * feature dependencies). 3301d7f67ac9SQu Wenruo * 3302d7f67ac9SQu Wenruo * For strict checks or possible corruption detection, see 3303d7f67ac9SQu Wenruo * btrfs_validate_super(). 3304d7f67ac9SQu Wenruo * 3305d7f67ac9SQu Wenruo * This should be called after btrfs_parse_options(), as some mount options 3306d7f67ac9SQu Wenruo * (space cache related) can modify on-disk format like free space tree and 3307d7f67ac9SQu Wenruo * screw up certain feature dependencies. 3308d7f67ac9SQu Wenruo */ 3309d7f67ac9SQu Wenruo int btrfs_check_features(struct btrfs_fs_info *fs_info, struct super_block *sb) 3310d7f67ac9SQu Wenruo { 3311d7f67ac9SQu Wenruo struct btrfs_super_block *disk_super = fs_info->super_copy; 3312d7f67ac9SQu Wenruo u64 incompat = btrfs_super_incompat_flags(disk_super); 3313d7f67ac9SQu Wenruo const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super); 3314d7f67ac9SQu Wenruo const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP); 3315d7f67ac9SQu Wenruo 3316d7f67ac9SQu Wenruo if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 3317d7f67ac9SQu Wenruo btrfs_err(fs_info, 3318d7f67ac9SQu Wenruo "cannot mount because of unknown incompat features (0x%llx)", 3319d7f67ac9SQu Wenruo incompat); 3320d7f67ac9SQu Wenruo return -EINVAL; 3321d7f67ac9SQu Wenruo } 3322d7f67ac9SQu Wenruo 3323d7f67ac9SQu Wenruo /* Runtime limitation for mixed block groups. */ 3324d7f67ac9SQu Wenruo if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 3325d7f67ac9SQu Wenruo (fs_info->sectorsize != fs_info->nodesize)) { 3326d7f67ac9SQu Wenruo btrfs_err(fs_info, 3327d7f67ac9SQu Wenruo "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", 3328d7f67ac9SQu Wenruo fs_info->nodesize, fs_info->sectorsize); 3329d7f67ac9SQu Wenruo return -EINVAL; 3330d7f67ac9SQu Wenruo } 3331d7f67ac9SQu Wenruo 3332d7f67ac9SQu Wenruo /* Mixed backref is an always-enabled feature. */ 3333d7f67ac9SQu Wenruo incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 3334d7f67ac9SQu Wenruo 3335d7f67ac9SQu Wenruo /* Set compression related flags just in case. */ 3336d7f67ac9SQu Wenruo if (fs_info->compress_type == BTRFS_COMPRESS_LZO) 3337d7f67ac9SQu Wenruo incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 3338d7f67ac9SQu Wenruo else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) 3339d7f67ac9SQu Wenruo incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; 3340d7f67ac9SQu Wenruo 3341d7f67ac9SQu Wenruo /* 3342d7f67ac9SQu Wenruo * An ancient flag, which should really be marked deprecated. 3343d7f67ac9SQu Wenruo * Such runtime limitation doesn't really need a incompat flag. 3344d7f67ac9SQu Wenruo */ 3345d7f67ac9SQu Wenruo if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) 3346d7f67ac9SQu Wenruo incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 3347d7f67ac9SQu Wenruo 3348d7f67ac9SQu Wenruo if (compat_ro_unsupp && !sb_rdonly(sb)) { 3349d7f67ac9SQu Wenruo btrfs_err(fs_info, 3350d7f67ac9SQu Wenruo "cannot mount read-write because of unknown compat_ro features (0x%llx)", 3351d7f67ac9SQu Wenruo compat_ro); 3352d7f67ac9SQu Wenruo return -EINVAL; 3353d7f67ac9SQu Wenruo } 3354d7f67ac9SQu Wenruo 3355d7f67ac9SQu Wenruo /* 3356d7f67ac9SQu Wenruo * We have unsupported RO compat features, although RO mounted, we 3357d7f67ac9SQu Wenruo * should not cause any metadata writes, including log replay. 3358d7f67ac9SQu Wenruo * Or we could screw up whatever the new feature requires. 3359d7f67ac9SQu Wenruo */ 3360d7f67ac9SQu Wenruo if (compat_ro_unsupp && btrfs_super_log_root(disk_super) && 3361d7f67ac9SQu Wenruo !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 3362d7f67ac9SQu Wenruo btrfs_err(fs_info, 3363d7f67ac9SQu Wenruo "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay", 3364d7f67ac9SQu Wenruo compat_ro); 3365d7f67ac9SQu Wenruo return -EINVAL; 3366d7f67ac9SQu Wenruo } 3367d7f67ac9SQu Wenruo 3368d7f67ac9SQu Wenruo /* 3369d7f67ac9SQu Wenruo * Artificial limitations for block group tree, to force 3370d7f67ac9SQu Wenruo * block-group-tree to rely on no-holes and free-space-tree. 3371d7f67ac9SQu Wenruo */ 3372d7f67ac9SQu Wenruo if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) && 3373d7f67ac9SQu Wenruo (!btrfs_fs_incompat(fs_info, NO_HOLES) || 3374d7f67ac9SQu Wenruo !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) { 3375d7f67ac9SQu Wenruo btrfs_err(fs_info, 3376d7f67ac9SQu Wenruo "block-group-tree feature requires no-holes and free-space-tree features"); 3377d7f67ac9SQu Wenruo return -EINVAL; 3378d7f67ac9SQu Wenruo } 3379d7f67ac9SQu Wenruo 3380d7f67ac9SQu Wenruo /* 3381d7f67ac9SQu Wenruo * Subpage runtime limitation on v1 cache. 3382d7f67ac9SQu Wenruo * 3383d7f67ac9SQu Wenruo * V1 space cache still has some hard codeed PAGE_SIZE usage, while 3384d7f67ac9SQu Wenruo * we're already defaulting to v2 cache, no need to bother v1 as it's 3385d7f67ac9SQu Wenruo * going to be deprecated anyway. 3386d7f67ac9SQu Wenruo */ 3387d7f67ac9SQu Wenruo if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) { 3388d7f67ac9SQu Wenruo btrfs_warn(fs_info, 3389d7f67ac9SQu Wenruo "v1 space cache is not supported for page size %lu with sectorsize %u", 3390d7f67ac9SQu Wenruo PAGE_SIZE, fs_info->sectorsize); 3391d7f67ac9SQu Wenruo return -EINVAL; 3392d7f67ac9SQu Wenruo } 3393d7f67ac9SQu Wenruo 3394d7f67ac9SQu Wenruo /* This can be called by remount, we need to protect the super block. */ 3395d7f67ac9SQu Wenruo spin_lock(&fs_info->super_lock); 3396d7f67ac9SQu Wenruo btrfs_set_super_incompat_flags(disk_super, incompat); 3397d7f67ac9SQu Wenruo spin_unlock(&fs_info->super_lock); 3398d7f67ac9SQu Wenruo 3399d7f67ac9SQu Wenruo return 0; 3400d7f67ac9SQu Wenruo } 3401d7f67ac9SQu Wenruo 3402ae18c37aSJosef Bacik int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices, 3403ae18c37aSJosef Bacik char *options) 3404ae18c37aSJosef Bacik { 3405ae18c37aSJosef Bacik u32 sectorsize; 3406ae18c37aSJosef Bacik u32 nodesize; 3407ae18c37aSJosef Bacik u32 stripesize; 3408ae18c37aSJosef Bacik u64 generation; 3409ae18c37aSJosef Bacik u64 features; 3410ae18c37aSJosef Bacik u16 csum_type; 3411ae18c37aSJosef Bacik struct btrfs_super_block *disk_super; 3412ae18c37aSJosef Bacik struct btrfs_fs_info *fs_info = btrfs_sb(sb); 3413ae18c37aSJosef Bacik struct btrfs_root *tree_root; 3414ae18c37aSJosef Bacik struct btrfs_root *chunk_root; 3415ae18c37aSJosef Bacik int ret; 3416ae18c37aSJosef Bacik int err = -EINVAL; 3417ae18c37aSJosef Bacik int level; 3418ae18c37aSJosef Bacik 34198260edbaSJosef Bacik ret = init_mount_fs_info(fs_info, sb); 342053b381b3SDavid Woodhouse if (ret) { 342183c8266aSDavid Sterba err = ret; 3422ae18c37aSJosef Bacik goto fail; 342353b381b3SDavid Woodhouse } 342453b381b3SDavid Woodhouse 3425ae18c37aSJosef Bacik /* These need to be init'ed before we start creating inodes and such. */ 3426ae18c37aSJosef Bacik tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, 3427ae18c37aSJosef Bacik GFP_KERNEL); 3428ae18c37aSJosef Bacik fs_info->tree_root = tree_root; 3429ae18c37aSJosef Bacik chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID, 3430ae18c37aSJosef Bacik GFP_KERNEL); 3431ae18c37aSJosef Bacik fs_info->chunk_root = chunk_root; 3432ae18c37aSJosef Bacik if (!tree_root || !chunk_root) { 3433ae18c37aSJosef Bacik err = -ENOMEM; 3434c75e8394SJosef Bacik goto fail; 3435ae18c37aSJosef Bacik } 3436ae18c37aSJosef Bacik 3437ae18c37aSJosef Bacik fs_info->btree_inode = new_inode(sb); 3438ae18c37aSJosef Bacik if (!fs_info->btree_inode) { 3439ae18c37aSJosef Bacik err = -ENOMEM; 3440c75e8394SJosef Bacik goto fail; 3441ae18c37aSJosef Bacik } 3442ae18c37aSJosef Bacik mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 3443ae18c37aSJosef Bacik btrfs_init_btree_inode(fs_info); 3444ae18c37aSJosef Bacik 3445d24fa5c1SAnand Jain invalidate_bdev(fs_devices->latest_dev->bdev); 34461104a885SDavid Sterba 34471104a885SDavid Sterba /* 34481104a885SDavid Sterba * Read super block and check the signature bytes only 34491104a885SDavid Sterba */ 3450d24fa5c1SAnand Jain disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev); 34518f32380dSJohannes Thumshirn if (IS_ERR(disk_super)) { 34528f32380dSJohannes Thumshirn err = PTR_ERR(disk_super); 345316cdcec7SMiao Xie goto fail_alloc; 345420b45077SDave Young } 345539279cc3SChris Mason 34561104a885SDavid Sterba /* 3457260db43cSRandy Dunlap * Verify the type first, if that or the checksum value are 34588dc3f22cSJohannes Thumshirn * corrupted, we'll find out 34598dc3f22cSJohannes Thumshirn */ 34608f32380dSJohannes Thumshirn csum_type = btrfs_super_csum_type(disk_super); 346151bce6c9SJohannes Thumshirn if (!btrfs_supported_super_csum(csum_type)) { 34628dc3f22cSJohannes Thumshirn btrfs_err(fs_info, "unsupported checksum algorithm: %u", 346351bce6c9SJohannes Thumshirn csum_type); 34648dc3f22cSJohannes Thumshirn err = -EINVAL; 34658f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 34668dc3f22cSJohannes Thumshirn goto fail_alloc; 34678dc3f22cSJohannes Thumshirn } 34688dc3f22cSJohannes Thumshirn 346983c68bbcSSu Yue fs_info->csum_size = btrfs_super_csum_size(disk_super); 347083c68bbcSSu Yue 34716d97c6e3SJohannes Thumshirn ret = btrfs_init_csum_hash(fs_info, csum_type); 34726d97c6e3SJohannes Thumshirn if (ret) { 34736d97c6e3SJohannes Thumshirn err = ret; 34748f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 34756d97c6e3SJohannes Thumshirn goto fail_alloc; 34766d97c6e3SJohannes Thumshirn } 34776d97c6e3SJohannes Thumshirn 34788dc3f22cSJohannes Thumshirn /* 34791104a885SDavid Sterba * We want to check superblock checksum, the type is stored inside. 34801104a885SDavid Sterba * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 34811104a885SDavid Sterba */ 34823d17adeaSQu Wenruo if (btrfs_check_super_csum(fs_info, disk_super)) { 348305135f59SDavid Sterba btrfs_err(fs_info, "superblock checksum mismatch"); 34841104a885SDavid Sterba err = -EINVAL; 34858f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 3486141386e1SJosef Bacik goto fail_alloc; 34871104a885SDavid Sterba } 34881104a885SDavid Sterba 34891104a885SDavid Sterba /* 34901104a885SDavid Sterba * super_copy is zeroed at allocation time and we never touch the 34911104a885SDavid Sterba * following bytes up to INFO_SIZE, the checksum is calculated from 34921104a885SDavid Sterba * the whole block of INFO_SIZE 34931104a885SDavid Sterba */ 34948f32380dSJohannes Thumshirn memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy)); 34958f32380dSJohannes Thumshirn btrfs_release_disk_super(disk_super); 34965f39d397SChris Mason 3497fbc6feaeSNikolay Borisov disk_super = fs_info->super_copy; 3498fbc6feaeSNikolay Borisov 34990b86a832SChris Mason 3500fbc6feaeSNikolay Borisov features = btrfs_super_flags(disk_super); 3501fbc6feaeSNikolay Borisov if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { 3502fbc6feaeSNikolay Borisov features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; 3503fbc6feaeSNikolay Borisov btrfs_set_super_flags(disk_super, features); 3504fbc6feaeSNikolay Borisov btrfs_info(fs_info, 3505fbc6feaeSNikolay Borisov "found metadata UUID change in progress flag, clearing"); 3506fbc6feaeSNikolay Borisov } 3507fbc6feaeSNikolay Borisov 3508fbc6feaeSNikolay Borisov memcpy(fs_info->super_for_commit, fs_info->super_copy, 3509fbc6feaeSNikolay Borisov sizeof(*fs_info->super_for_commit)); 3510de37aa51SNikolay Borisov 3511069ec957SQu Wenruo ret = btrfs_validate_mount_super(fs_info); 35121104a885SDavid Sterba if (ret) { 351305135f59SDavid Sterba btrfs_err(fs_info, "superblock contains fatal errors"); 35141104a885SDavid Sterba err = -EINVAL; 3515141386e1SJosef Bacik goto fail_alloc; 35161104a885SDavid Sterba } 35171104a885SDavid Sterba 35180f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 3519141386e1SJosef Bacik goto fail_alloc; 35200f7d52f4SChris Mason 3521acce952bSliubo /* check FS state, whether FS is broken. */ 352287533c47SMiao Xie if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 352387533c47SMiao Xie set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 3524acce952bSliubo 352575e7cb7fSLiu Bo /* 352675e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 352775e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 352875e7cb7fSLiu Bo */ 352975e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 353075e7cb7fSLiu Bo 35316f93e834SAnand Jain 35326f93e834SAnand Jain /* Set up fs_info before parsing mount options */ 35336f93e834SAnand Jain nodesize = btrfs_super_nodesize(disk_super); 35346f93e834SAnand Jain sectorsize = btrfs_super_sectorsize(disk_super); 35356f93e834SAnand Jain stripesize = sectorsize; 35366f93e834SAnand Jain fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 35376f93e834SAnand Jain fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 35386f93e834SAnand Jain 35396f93e834SAnand Jain fs_info->nodesize = nodesize; 35406f93e834SAnand Jain fs_info->sectorsize = sectorsize; 35416f93e834SAnand Jain fs_info->sectorsize_bits = ilog2(sectorsize); 35426f93e834SAnand Jain fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size; 35436f93e834SAnand Jain fs_info->stripesize = stripesize; 35446f93e834SAnand Jain 35452ff7e61eSJeff Mahoney ret = btrfs_parse_options(fs_info, options, sb->s_flags); 35462b82032cSYan Zheng if (ret) { 35472b82032cSYan Zheng err = ret; 3548141386e1SJosef Bacik goto fail_alloc; 35492b82032cSYan Zheng } 3550dfe25020SChris Mason 3551d7f67ac9SQu Wenruo ret = btrfs_check_features(fs_info, sb); 3552d7f67ac9SQu Wenruo if (ret < 0) { 3553d7f67ac9SQu Wenruo err = ret; 3554141386e1SJosef Bacik goto fail_alloc; 3555f2b636e8SJosef Bacik } 3556f2b636e8SJosef Bacik 35578481dd80SQu Wenruo if (sectorsize < PAGE_SIZE) { 35588481dd80SQu Wenruo struct btrfs_subpage_info *subpage_info; 35598481dd80SQu Wenruo 35609f73f1aeSQu Wenruo /* 35619f73f1aeSQu Wenruo * V1 space cache has some hardcoded PAGE_SIZE usage, and is 35629f73f1aeSQu Wenruo * going to be deprecated. 35639f73f1aeSQu Wenruo * 35649f73f1aeSQu Wenruo * Force to use v2 cache for subpage case. 35659f73f1aeSQu Wenruo */ 35669f73f1aeSQu Wenruo btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE); 35679f73f1aeSQu Wenruo btrfs_set_and_info(fs_info, FREE_SPACE_TREE, 35689f73f1aeSQu Wenruo "forcing free space tree for sector size %u with page size %lu", 35699f73f1aeSQu Wenruo sectorsize, PAGE_SIZE); 35709f73f1aeSQu Wenruo 357195ea0486SQu Wenruo btrfs_warn(fs_info, 357295ea0486SQu Wenruo "read-write for sector size %u with page size %lu is experimental", 35730bb3eb3eSQu Wenruo sectorsize, PAGE_SIZE); 35748481dd80SQu Wenruo subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL); 35758481dd80SQu Wenruo if (!subpage_info) 35768481dd80SQu Wenruo goto fail_alloc; 35778481dd80SQu Wenruo btrfs_init_subpage_info(subpage_info, sectorsize); 35788481dd80SQu Wenruo fs_info->subpage_info = subpage_info; 3579c8050b3bSQu Wenruo } 35800bb3eb3eSQu Wenruo 3581d21deec5SSu Yue ret = btrfs_init_workqueues(fs_info); 35822a458198SEric Sandeen if (ret) { 35832a458198SEric Sandeen err = ret; 35840dc3b84aSJosef Bacik goto fail_sb_buffer; 35850dc3b84aSJosef Bacik } 35864543df7eSChris Mason 35879e11ceeeSJan Kara sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); 35889e11ceeeSJan Kara sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); 35894575c9ccSChris Mason 3590a061fc8dSChris Mason sb->s_blocksize = sectorsize; 3591a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 3592de37aa51SNikolay Borisov memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); 3593db94535dSChris Mason 3594925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 35956bccf3abSJeff Mahoney ret = btrfs_read_sys_array(fs_info); 3596925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 359784eed90fSChris Mason if (ret) { 359805135f59SDavid Sterba btrfs_err(fs_info, "failed to read the system array: %d", ret); 35995d4f98a2SYan Zheng goto fail_sb_buffer; 360084eed90fSChris Mason } 36010b86a832SChris Mason 360284234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 3603581c1760SQu Wenruo level = btrfs_super_chunk_root_level(disk_super); 3604bd676446SJosef Bacik ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super), 3605bd676446SJosef Bacik generation, level); 3606bd676446SJosef Bacik if (ret) { 360705135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk root"); 3608af31f5e5SChris Mason goto fail_tree_roots; 360983121942SDavid Woodhouse } 36100b86a832SChris Mason 3611e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 3612c4ac7541SDavid Sterba offsetof(struct btrfs_header, chunk_tree_uuid), 3613c4ac7541SDavid Sterba BTRFS_UUID_SIZE); 3614e17cade2SChris Mason 36155b4aacefSJeff Mahoney ret = btrfs_read_chunk_tree(fs_info); 36162b82032cSYan Zheng if (ret) { 361705135f59SDavid Sterba btrfs_err(fs_info, "failed to read chunk tree: %d", ret); 3618af31f5e5SChris Mason goto fail_tree_roots; 36192b82032cSYan Zheng } 36200b86a832SChris Mason 36218dabb742SStefan Behrens /* 3622bacce86aSAnand Jain * At this point we know all the devices that make this filesystem, 3623bacce86aSAnand Jain * including the seed devices but we don't know yet if the replace 3624bacce86aSAnand Jain * target is required. So free devices that are not part of this 36251a9fd417SDavid Sterba * filesystem but skip the replace target device which is checked 3626bacce86aSAnand Jain * below in btrfs_init_dev_replace(). 36278dabb742SStefan Behrens */ 3628bacce86aSAnand Jain btrfs_free_extra_devids(fs_devices); 3629d24fa5c1SAnand Jain if (!fs_devices->latest_dev->bdev) { 363005135f59SDavid Sterba btrfs_err(fs_info, "failed to read devices"); 3631a6b0d5c8SChris Mason goto fail_tree_roots; 3632a6b0d5c8SChris Mason } 3633a6b0d5c8SChris Mason 3634b8522a1eSNikolay Borisov ret = init_tree_roots(fs_info); 36354bbcaa64SEric Sandeen if (ret) 3636b8522a1eSNikolay Borisov goto fail_tree_roots; 36378929ecfaSYan, Zheng 363875ec1db8SJosef Bacik /* 363973651042SNaohiro Aota * Get zone type information of zoned block devices. This will also 364073651042SNaohiro Aota * handle emulation of a zoned filesystem if a regular device has the 364173651042SNaohiro Aota * zoned incompat feature flag set. 364273651042SNaohiro Aota */ 364373651042SNaohiro Aota ret = btrfs_get_dev_zone_info_all_devices(fs_info); 364473651042SNaohiro Aota if (ret) { 364573651042SNaohiro Aota btrfs_err(fs_info, 364673651042SNaohiro Aota "zoned: failed to read device zone info: %d", 364773651042SNaohiro Aota ret); 364873651042SNaohiro Aota goto fail_block_groups; 364973651042SNaohiro Aota } 365073651042SNaohiro Aota 365173651042SNaohiro Aota /* 365275ec1db8SJosef Bacik * If we have a uuid root and we're not being told to rescan we need to 365375ec1db8SJosef Bacik * check the generation here so we can set the 365475ec1db8SJosef Bacik * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the 365575ec1db8SJosef Bacik * transaction during a balance or the log replay without updating the 365675ec1db8SJosef Bacik * uuid generation, and then if we crash we would rescan the uuid tree, 365775ec1db8SJosef Bacik * even though it was perfectly fine. 365875ec1db8SJosef Bacik */ 365975ec1db8SJosef Bacik if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) && 366075ec1db8SJosef Bacik fs_info->generation == btrfs_super_uuid_tree_generation(disk_super)) 366175ec1db8SJosef Bacik set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 366275ec1db8SJosef Bacik 3663cf90d884SQu Wenruo ret = btrfs_verify_dev_extents(fs_info); 3664cf90d884SQu Wenruo if (ret) { 3665cf90d884SQu Wenruo btrfs_err(fs_info, 3666cf90d884SQu Wenruo "failed to verify dev extents against chunks: %d", 3667cf90d884SQu Wenruo ret); 3668cf90d884SQu Wenruo goto fail_block_groups; 3669cf90d884SQu Wenruo } 367068310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 367168310a5eSIlya Dryomov if (ret) { 367205135f59SDavid Sterba btrfs_err(fs_info, "failed to recover balance: %d", ret); 367368310a5eSIlya Dryomov goto fail_block_groups; 367468310a5eSIlya Dryomov } 367568310a5eSIlya Dryomov 3676733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 3677733f4fbbSStefan Behrens if (ret) { 367805135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_stats: %d", ret); 3679733f4fbbSStefan Behrens goto fail_block_groups; 3680733f4fbbSStefan Behrens } 3681733f4fbbSStefan Behrens 36828dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 36838dabb742SStefan Behrens if (ret) { 368405135f59SDavid Sterba btrfs_err(fs_info, "failed to init dev_replace: %d", ret); 36858dabb742SStefan Behrens goto fail_block_groups; 36868dabb742SStefan Behrens } 36878dabb742SStefan Behrens 3688b70f5097SNaohiro Aota ret = btrfs_check_zoned_mode(fs_info); 3689b70f5097SNaohiro Aota if (ret) { 3690b70f5097SNaohiro Aota btrfs_err(fs_info, "failed to initialize zoned mode: %d", 3691b70f5097SNaohiro Aota ret); 3692b70f5097SNaohiro Aota goto fail_block_groups; 3693b70f5097SNaohiro Aota } 3694b70f5097SNaohiro Aota 3695c6761a9eSAnand Jain ret = btrfs_sysfs_add_fsid(fs_devices); 3696b7c35e81SAnand Jain if (ret) { 369705135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", 369805135f59SDavid Sterba ret); 3699b7c35e81SAnand Jain goto fail_block_groups; 3700b7c35e81SAnand Jain } 3701b7c35e81SAnand Jain 370296f3136eSAnand Jain ret = btrfs_sysfs_add_mounted(fs_info); 37035ac1d209SJeff Mahoney if (ret) { 370405135f59SDavid Sterba btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); 3705b7c35e81SAnand Jain goto fail_fsdev_sysfs; 37065ac1d209SJeff Mahoney } 37075ac1d209SJeff Mahoney 3708c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 3709c59021f8Sliubo if (ret) { 371005135f59SDavid Sterba btrfs_err(fs_info, "failed to initialize space info: %d", ret); 37112365dd3cSAnand Jain goto fail_sysfs; 3712c59021f8Sliubo } 3713c59021f8Sliubo 37145b4aacefSJeff Mahoney ret = btrfs_read_block_groups(fs_info); 37151b1d1f66SJosef Bacik if (ret) { 371605135f59SDavid Sterba btrfs_err(fs_info, "failed to read block groups: %d", ret); 37172365dd3cSAnand Jain goto fail_sysfs; 37181b1d1f66SJosef Bacik } 37194330e183SQu Wenruo 372016beac87SNaohiro Aota btrfs_free_zone_cache(fs_info); 372116beac87SNaohiro Aota 37225c78a5e7SAnand Jain if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices && 37235c78a5e7SAnand Jain !btrfs_check_rw_degradable(fs_info, NULL)) { 372405135f59SDavid Sterba btrfs_warn(fs_info, 372552042d8eSAndrea Gelmini "writable mount is not allowed due to too many missing devices"); 37262365dd3cSAnand Jain goto fail_sysfs; 3727292fd7fcSStefan Behrens } 37289078a3e1SChris Mason 372933c44184SJosef Bacik fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info, 3730a74a4b97SChris Mason "btrfs-cleaner"); 373157506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 37322365dd3cSAnand Jain goto fail_sysfs; 3733a74a4b97SChris Mason 3734a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 3735a74a4b97SChris Mason tree_root, 3736a74a4b97SChris Mason "btrfs-transaction"); 373757506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 37383f157a2fSChris Mason goto fail_cleaner; 3739a74a4b97SChris Mason 3740583b7231SHans van Kranenburg if (!btrfs_test_opt(fs_info, NOSSD) && 3741c289811cSChris Mason !fs_info->fs_devices->rotating) { 3742583b7231SHans van Kranenburg btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); 3743c289811cSChris Mason } 3744c289811cSChris Mason 3745572d9ab7SDavid Sterba /* 374601327610SNicholas D Steeves * Mount does not set all options immediately, we can do it now and do 3747572d9ab7SDavid Sterba * not have to wait for transaction commit 3748572d9ab7SDavid Sterba */ 3749572d9ab7SDavid Sterba btrfs_apply_pending_changes(fs_info); 37503818aea2SQu Wenruo 375121adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 37520b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { 37532ff7e61eSJeff Mahoney ret = btrfsic_mount(fs_info, fs_devices, 37540b246afaSJeff Mahoney btrfs_test_opt(fs_info, 3755cbeaae4fSDavid Sterba CHECK_INTEGRITY_DATA) ? 1 : 0, 375621adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 375721adbd5cSStefan Behrens if (ret) 375805135f59SDavid Sterba btrfs_warn(fs_info, 375905135f59SDavid Sterba "failed to initialize integrity check module: %d", 376005135f59SDavid Sterba ret); 376121adbd5cSStefan Behrens } 376221adbd5cSStefan Behrens #endif 3763bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 3764bcef60f2SArne Jansen if (ret) 3765bcef60f2SArne Jansen goto fail_trans_kthread; 376621adbd5cSStefan Behrens 3767fd708b81SJosef Bacik if (btrfs_build_ref_tree(fs_info)) 3768fd708b81SJosef Bacik btrfs_err(fs_info, "couldn't build ref tree"); 3769fd708b81SJosef Bacik 377096da0919SQu Wenruo /* do not make disk changes in broken FS or nologreplay is given */ 377196da0919SQu Wenruo if (btrfs_super_log_root(disk_super) != 0 && 37720b246afaSJeff Mahoney !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 3773e8294f2fSDavid Sterba btrfs_info(fs_info, "start tree-log replay"); 377463443bf5SEric Sandeen ret = btrfs_replay_log(fs_info, fs_devices); 377579787eaaSJeff Mahoney if (ret) { 377663443bf5SEric Sandeen err = ret; 377728c16cbbSWang Shilong goto fail_qgroup; 3778e556ce2cSYan Zheng } 3779e02119d5SChris Mason } 37801a40e23bSZheng Yan 378156e9357aSDavid Sterba fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true); 37823140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 37833140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 3784f50f4353SLiu Bo btrfs_warn(fs_info, "failed to read fs tree: %d", err); 3785315bf8efSJosef Bacik fs_info->fs_root = NULL; 3786bcef60f2SArne Jansen goto fail_qgroup; 37873140c9a3SDan Carpenter } 3788c289811cSChris Mason 3789bc98a42cSDavid Howells if (sb_rdonly(sb)) 37908cd29088SBoris Burkov goto clear_oneshot; 37912b6ba629SIlya Dryomov 379244c0ca21SBoris Burkov ret = btrfs_start_pre_rw_mount(fs_info); 37932b6ba629SIlya Dryomov if (ret) { 37946bccf3abSJeff Mahoney close_ctree(fs_info); 37952b6ba629SIlya Dryomov return ret; 3796e3acc2a6SJosef Bacik } 3797b0643e59SDennis Zhou btrfs_discard_resume(fs_info); 3798b382a324SJan Schmidt 379944c0ca21SBoris Burkov if (fs_info->uuid_root && 380044c0ca21SBoris Burkov (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || 380144c0ca21SBoris Burkov fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) { 380205135f59SDavid Sterba btrfs_info(fs_info, "checking UUID tree"); 380370f80175SStefan Behrens ret = btrfs_check_uuid_tree(fs_info); 380470f80175SStefan Behrens if (ret) { 380505135f59SDavid Sterba btrfs_warn(fs_info, 380605135f59SDavid Sterba "failed to check the UUID tree: %d", ret); 38076bccf3abSJeff Mahoney close_ctree(fs_info); 380870f80175SStefan Behrens return ret; 380970f80175SStefan Behrens } 3810f7a81ea4SStefan Behrens } 381194846229SBoris Burkov 3812afcdd129SJosef Bacik set_bit(BTRFS_FS_OPEN, &fs_info->flags); 381347ab2a6cSJosef Bacik 3814b4be6aefSJosef Bacik /* Kick the cleaner thread so it'll start deleting snapshots. */ 3815b4be6aefSJosef Bacik if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags)) 3816b4be6aefSJosef Bacik wake_up_process(fs_info->cleaner_kthread); 3817b4be6aefSJosef Bacik 38188cd29088SBoris Burkov clear_oneshot: 38198cd29088SBoris Burkov btrfs_clear_oneshot_options(fs_info); 3820ad2b2c80SAl Viro return 0; 382139279cc3SChris Mason 3822bcef60f2SArne Jansen fail_qgroup: 3823bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 38247c2ca468SChris Mason fail_trans_kthread: 38257c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 38262ff7e61eSJeff Mahoney btrfs_cleanup_transaction(fs_info); 3827faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 38283f157a2fSChris Mason fail_cleaner: 3829a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 38307c2ca468SChris Mason 38317c2ca468SChris Mason /* 38327c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 38337c2ca468SChris Mason * kthreads 38347c2ca468SChris Mason */ 38357c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 38367c2ca468SChris Mason 38372365dd3cSAnand Jain fail_sysfs: 38386618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 38392365dd3cSAnand Jain 3840b7c35e81SAnand Jain fail_fsdev_sysfs: 3841b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3842b7c35e81SAnand Jain 38431b1d1f66SJosef Bacik fail_block_groups: 384454067ae9SJosef Bacik btrfs_put_block_group_cache(fs_info); 3845af31f5e5SChris Mason 3846af31f5e5SChris Mason fail_tree_roots: 38479e3aa805SJosef Bacik if (fs_info->data_reloc_root) 38489e3aa805SJosef Bacik btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root); 38494273eaffSAnand Jain free_root_pointers(fs_info, true); 38502b8195bbSMiao Xie invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3851af31f5e5SChris Mason 385239279cc3SChris Mason fail_sb_buffer: 38537abadb64SLiu Bo btrfs_stop_all_workers(fs_info); 38545cdd7db6SFilipe Manana btrfs_free_block_groups(fs_info); 385516cdcec7SMiao Xie fail_alloc: 3856586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 3857586e46e2SIlya Dryomov 38584543df7eSChris Mason iput(fs_info->btree_inode); 38597e662854SQinghuang Feng fail: 3860586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 3861ad2b2c80SAl Viro return err; 3862eb60ceacSChris Mason } 3863663faf9fSMasami Hiramatsu ALLOW_ERROR_INJECTION(open_ctree, ERRNO); 3864eb60ceacSChris Mason 3865314b6dd0SJohannes Thumshirn static void btrfs_end_super_write(struct bio *bio) 3866f2984462SChris Mason { 3867314b6dd0SJohannes Thumshirn struct btrfs_device *device = bio->bi_private; 3868314b6dd0SJohannes Thumshirn struct bio_vec *bvec; 3869314b6dd0SJohannes Thumshirn struct bvec_iter_all iter_all; 3870314b6dd0SJohannes Thumshirn struct page *page; 3871442a4f63SStefan Behrens 3872314b6dd0SJohannes Thumshirn bio_for_each_segment_all(bvec, bio, iter_all) { 3873314b6dd0SJohannes Thumshirn page = bvec->bv_page; 3874314b6dd0SJohannes Thumshirn 3875314b6dd0SJohannes Thumshirn if (bio->bi_status) { 3876fb456252SJeff Mahoney btrfs_warn_rl_in_rcu(device->fs_info, 3877314b6dd0SJohannes Thumshirn "lost page write due to IO error on %s (%d)", 3878314b6dd0SJohannes Thumshirn rcu_str_deref(device->name), 3879314b6dd0SJohannes Thumshirn blk_status_to_errno(bio->bi_status)); 3880314b6dd0SJohannes Thumshirn ClearPageUptodate(page); 3881314b6dd0SJohannes Thumshirn SetPageError(page); 3882314b6dd0SJohannes Thumshirn btrfs_dev_stat_inc_and_print(device, 3883314b6dd0SJohannes Thumshirn BTRFS_DEV_STAT_WRITE_ERRS); 3884314b6dd0SJohannes Thumshirn } else { 3885314b6dd0SJohannes Thumshirn SetPageUptodate(page); 3886f2984462SChris Mason } 3887314b6dd0SJohannes Thumshirn 3888314b6dd0SJohannes Thumshirn put_page(page); 3889314b6dd0SJohannes Thumshirn unlock_page(page); 3890314b6dd0SJohannes Thumshirn } 3891314b6dd0SJohannes Thumshirn 3892314b6dd0SJohannes Thumshirn bio_put(bio); 3893f2984462SChris Mason } 3894f2984462SChris Mason 38958f32380dSJohannes Thumshirn struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev, 3896a05d3c91SQu Wenruo int copy_num, bool drop_cache) 389729c36d72SAnand Jain { 389829c36d72SAnand Jain struct btrfs_super_block *super; 38998f32380dSJohannes Thumshirn struct page *page; 390012659251SNaohiro Aota u64 bytenr, bytenr_orig; 39018f32380dSJohannes Thumshirn struct address_space *mapping = bdev->bd_inode->i_mapping; 390212659251SNaohiro Aota int ret; 390329c36d72SAnand Jain 390412659251SNaohiro Aota bytenr_orig = btrfs_sb_offset(copy_num); 390512659251SNaohiro Aota ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr); 390612659251SNaohiro Aota if (ret == -ENOENT) 390712659251SNaohiro Aota return ERR_PTR(-EINVAL); 390812659251SNaohiro Aota else if (ret) 390912659251SNaohiro Aota return ERR_PTR(ret); 391012659251SNaohiro Aota 3911cda00ebaSChristoph Hellwig if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev)) 39128f32380dSJohannes Thumshirn return ERR_PTR(-EINVAL); 391329c36d72SAnand Jain 3914a05d3c91SQu Wenruo if (drop_cache) { 3915a05d3c91SQu Wenruo /* This should only be called with the primary sb. */ 3916a05d3c91SQu Wenruo ASSERT(copy_num == 0); 3917a05d3c91SQu Wenruo 3918a05d3c91SQu Wenruo /* 3919a05d3c91SQu Wenruo * Drop the page of the primary superblock, so later read will 3920a05d3c91SQu Wenruo * always read from the device. 3921a05d3c91SQu Wenruo */ 3922a05d3c91SQu Wenruo invalidate_inode_pages2_range(mapping, 3923a05d3c91SQu Wenruo bytenr >> PAGE_SHIFT, 3924a05d3c91SQu Wenruo (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT); 3925a05d3c91SQu Wenruo } 3926a05d3c91SQu Wenruo 39278f32380dSJohannes Thumshirn page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS); 39288f32380dSJohannes Thumshirn if (IS_ERR(page)) 39298f32380dSJohannes Thumshirn return ERR_CAST(page); 393029c36d72SAnand Jain 39318f32380dSJohannes Thumshirn super = page_address(page); 393296c2e067SAnand Jain if (btrfs_super_magic(super) != BTRFS_MAGIC) { 393396c2e067SAnand Jain btrfs_release_disk_super(super); 393496c2e067SAnand Jain return ERR_PTR(-ENODATA); 393596c2e067SAnand Jain } 393696c2e067SAnand Jain 393712659251SNaohiro Aota if (btrfs_super_bytenr(super) != bytenr_orig) { 39388f32380dSJohannes Thumshirn btrfs_release_disk_super(super); 39398f32380dSJohannes Thumshirn return ERR_PTR(-EINVAL); 394029c36d72SAnand Jain } 394129c36d72SAnand Jain 39428f32380dSJohannes Thumshirn return super; 394329c36d72SAnand Jain } 394429c36d72SAnand Jain 394529c36d72SAnand Jain 39468f32380dSJohannes Thumshirn struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev) 3947a512bbf8SYan Zheng { 39488f32380dSJohannes Thumshirn struct btrfs_super_block *super, *latest = NULL; 3949a512bbf8SYan Zheng int i; 3950a512bbf8SYan Zheng u64 transid = 0; 3951a512bbf8SYan Zheng 3952a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 3953a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 3954a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 3955a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3956a512bbf8SYan Zheng */ 3957a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 3958a05d3c91SQu Wenruo super = btrfs_read_dev_one_super(bdev, i, false); 39598f32380dSJohannes Thumshirn if (IS_ERR(super)) 3960a512bbf8SYan Zheng continue; 3961a512bbf8SYan Zheng 3962a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 39638f32380dSJohannes Thumshirn if (latest) 39648f32380dSJohannes Thumshirn btrfs_release_disk_super(super); 39658f32380dSJohannes Thumshirn 39668f32380dSJohannes Thumshirn latest = super; 3967a512bbf8SYan Zheng transid = btrfs_super_generation(super); 3968a512bbf8SYan Zheng } 3969a512bbf8SYan Zheng } 397092fc03fbSAnand Jain 39718f32380dSJohannes Thumshirn return super; 3972a512bbf8SYan Zheng } 3973a512bbf8SYan Zheng 39744eedeb75SHisashi Hifumi /* 3975abbb3b8eSDavid Sterba * Write superblock @sb to the @device. Do not wait for completion, all the 3976314b6dd0SJohannes Thumshirn * pages we use for writing are locked. 39774eedeb75SHisashi Hifumi * 3978abbb3b8eSDavid Sterba * Write @max_mirrors copies of the superblock, where 0 means default that fit 3979abbb3b8eSDavid Sterba * the expected device size at commit time. Note that max_mirrors must be 3980abbb3b8eSDavid Sterba * same for write and wait phases. 39814eedeb75SHisashi Hifumi * 3982314b6dd0SJohannes Thumshirn * Return number of errors when page is not found or submission fails. 39834eedeb75SHisashi Hifumi */ 3984a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 3985abbb3b8eSDavid Sterba struct btrfs_super_block *sb, int max_mirrors) 3986a512bbf8SYan Zheng { 3987d5178578SJohannes Thumshirn struct btrfs_fs_info *fs_info = device->fs_info; 3988314b6dd0SJohannes Thumshirn struct address_space *mapping = device->bdev->bd_inode->i_mapping; 3989d5178578SJohannes Thumshirn SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 3990a512bbf8SYan Zheng int i; 3991a512bbf8SYan Zheng int errors = 0; 399212659251SNaohiro Aota int ret; 399312659251SNaohiro Aota u64 bytenr, bytenr_orig; 3994a512bbf8SYan Zheng 3995a512bbf8SYan Zheng if (max_mirrors == 0) 3996a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3997a512bbf8SYan Zheng 3998d5178578SJohannes Thumshirn shash->tfm = fs_info->csum_shash; 3999d5178578SJohannes Thumshirn 4000a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 4001314b6dd0SJohannes Thumshirn struct page *page; 4002314b6dd0SJohannes Thumshirn struct bio *bio; 4003314b6dd0SJohannes Thumshirn struct btrfs_super_block *disk_super; 4004314b6dd0SJohannes Thumshirn 400512659251SNaohiro Aota bytenr_orig = btrfs_sb_offset(i); 400612659251SNaohiro Aota ret = btrfs_sb_log_location(device, i, WRITE, &bytenr); 400712659251SNaohiro Aota if (ret == -ENOENT) { 400812659251SNaohiro Aota continue; 400912659251SNaohiro Aota } else if (ret < 0) { 401012659251SNaohiro Aota btrfs_err(device->fs_info, 401112659251SNaohiro Aota "couldn't get super block location for mirror %d", 401212659251SNaohiro Aota i); 401312659251SNaohiro Aota errors++; 401412659251SNaohiro Aota continue; 401512659251SNaohiro Aota } 4016935e5cc9SMiao Xie if (bytenr + BTRFS_SUPER_INFO_SIZE >= 4017935e5cc9SMiao Xie device->commit_total_bytes) 4018a512bbf8SYan Zheng break; 4019a512bbf8SYan Zheng 402012659251SNaohiro Aota btrfs_set_super_bytenr(sb, bytenr_orig); 4021a512bbf8SYan Zheng 4022fd08001fSEric Biggers crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE, 4023fd08001fSEric Biggers BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, 4024fd08001fSEric Biggers sb->csum); 4025a512bbf8SYan Zheng 4026314b6dd0SJohannes Thumshirn page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT, 4027314b6dd0SJohannes Thumshirn GFP_NOFS); 4028314b6dd0SJohannes Thumshirn if (!page) { 4029fb456252SJeff Mahoney btrfs_err(device->fs_info, 4030314b6dd0SJohannes Thumshirn "couldn't get super block page for bytenr %llu", 4031f14d104dSDavid Sterba bytenr); 4032634554dcSJosef Bacik errors++; 4033634554dcSJosef Bacik continue; 4034634554dcSJosef Bacik } 4035634554dcSJosef Bacik 4036314b6dd0SJohannes Thumshirn /* Bump the refcount for wait_dev_supers() */ 4037314b6dd0SJohannes Thumshirn get_page(page); 4038a512bbf8SYan Zheng 4039314b6dd0SJohannes Thumshirn disk_super = page_address(page); 4040314b6dd0SJohannes Thumshirn memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE); 4041a512bbf8SYan Zheng 4042387125fcSChris Mason /* 4043314b6dd0SJohannes Thumshirn * Directly use bios here instead of relying on the page cache 4044314b6dd0SJohannes Thumshirn * to do I/O, so we don't lose the ability to do integrity 4045314b6dd0SJohannes Thumshirn * checking. 4046387125fcSChris Mason */ 404707888c66SChristoph Hellwig bio = bio_alloc(device->bdev, 1, 404807888c66SChristoph Hellwig REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO, 404907888c66SChristoph Hellwig GFP_NOFS); 4050314b6dd0SJohannes Thumshirn bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT; 4051314b6dd0SJohannes Thumshirn bio->bi_private = device; 4052314b6dd0SJohannes Thumshirn bio->bi_end_io = btrfs_end_super_write; 4053314b6dd0SJohannes Thumshirn __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE, 4054314b6dd0SJohannes Thumshirn offset_in_page(bytenr)); 4055314b6dd0SJohannes Thumshirn 4056314b6dd0SJohannes Thumshirn /* 4057314b6dd0SJohannes Thumshirn * We FUA only the first super block. The others we allow to 4058314b6dd0SJohannes Thumshirn * go down lazy and there's a short window where the on-disk 4059314b6dd0SJohannes Thumshirn * copies might still contain the older version. 4060314b6dd0SJohannes Thumshirn */ 40611b9e619cSOmar Sandoval if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) 4062314b6dd0SJohannes Thumshirn bio->bi_opf |= REQ_FUA; 4063314b6dd0SJohannes Thumshirn 406458ff51f1SChristoph Hellwig btrfsic_check_bio(bio); 406558ff51f1SChristoph Hellwig submit_bio(bio); 40668376d9e1SNaohiro Aota 40678376d9e1SNaohiro Aota if (btrfs_advance_sb_log(device, i)) 40688376d9e1SNaohiro Aota errors++; 4069a512bbf8SYan Zheng } 4070a512bbf8SYan Zheng return errors < i ? 0 : -1; 4071a512bbf8SYan Zheng } 4072a512bbf8SYan Zheng 4073387125fcSChris Mason /* 4074abbb3b8eSDavid Sterba * Wait for write completion of superblocks done by write_dev_supers, 4075abbb3b8eSDavid Sterba * @max_mirrors same for write and wait phases. 4076abbb3b8eSDavid Sterba * 4077314b6dd0SJohannes Thumshirn * Return number of errors when page is not found or not marked up to 4078abbb3b8eSDavid Sterba * date. 4079abbb3b8eSDavid Sterba */ 4080abbb3b8eSDavid Sterba static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) 4081abbb3b8eSDavid Sterba { 4082abbb3b8eSDavid Sterba int i; 4083abbb3b8eSDavid Sterba int errors = 0; 4084b6a535faSHoward McLauchlan bool primary_failed = false; 408512659251SNaohiro Aota int ret; 4086abbb3b8eSDavid Sterba u64 bytenr; 4087abbb3b8eSDavid Sterba 4088abbb3b8eSDavid Sterba if (max_mirrors == 0) 4089abbb3b8eSDavid Sterba max_mirrors = BTRFS_SUPER_MIRROR_MAX; 4090abbb3b8eSDavid Sterba 4091abbb3b8eSDavid Sterba for (i = 0; i < max_mirrors; i++) { 4092314b6dd0SJohannes Thumshirn struct page *page; 4093314b6dd0SJohannes Thumshirn 409412659251SNaohiro Aota ret = btrfs_sb_log_location(device, i, READ, &bytenr); 409512659251SNaohiro Aota if (ret == -ENOENT) { 409612659251SNaohiro Aota break; 409712659251SNaohiro Aota } else if (ret < 0) { 409812659251SNaohiro Aota errors++; 409912659251SNaohiro Aota if (i == 0) 410012659251SNaohiro Aota primary_failed = true; 410112659251SNaohiro Aota continue; 410212659251SNaohiro Aota } 4103abbb3b8eSDavid Sterba if (bytenr + BTRFS_SUPER_INFO_SIZE >= 4104abbb3b8eSDavid Sterba device->commit_total_bytes) 4105abbb3b8eSDavid Sterba break; 4106abbb3b8eSDavid Sterba 4107314b6dd0SJohannes Thumshirn page = find_get_page(device->bdev->bd_inode->i_mapping, 4108314b6dd0SJohannes Thumshirn bytenr >> PAGE_SHIFT); 4109314b6dd0SJohannes Thumshirn if (!page) { 4110abbb3b8eSDavid Sterba errors++; 4111b6a535faSHoward McLauchlan if (i == 0) 4112b6a535faSHoward McLauchlan primary_failed = true; 4113abbb3b8eSDavid Sterba continue; 4114abbb3b8eSDavid Sterba } 4115314b6dd0SJohannes Thumshirn /* Page is submitted locked and unlocked once the IO completes */ 4116314b6dd0SJohannes Thumshirn wait_on_page_locked(page); 4117314b6dd0SJohannes Thumshirn if (PageError(page)) { 4118abbb3b8eSDavid Sterba errors++; 4119b6a535faSHoward McLauchlan if (i == 0) 4120b6a535faSHoward McLauchlan primary_failed = true; 4121b6a535faSHoward McLauchlan } 4122abbb3b8eSDavid Sterba 4123314b6dd0SJohannes Thumshirn /* Drop our reference */ 4124314b6dd0SJohannes Thumshirn put_page(page); 4125abbb3b8eSDavid Sterba 4126314b6dd0SJohannes Thumshirn /* Drop the reference from the writing run */ 4127314b6dd0SJohannes Thumshirn put_page(page); 4128abbb3b8eSDavid Sterba } 4129abbb3b8eSDavid Sterba 4130b6a535faSHoward McLauchlan /* log error, force error return */ 4131b6a535faSHoward McLauchlan if (primary_failed) { 4132b6a535faSHoward McLauchlan btrfs_err(device->fs_info, "error writing primary super block to device %llu", 4133b6a535faSHoward McLauchlan device->devid); 4134b6a535faSHoward McLauchlan return -1; 4135b6a535faSHoward McLauchlan } 4136b6a535faSHoward McLauchlan 4137abbb3b8eSDavid Sterba return errors < i ? 0 : -1; 4138abbb3b8eSDavid Sterba } 4139abbb3b8eSDavid Sterba 4140abbb3b8eSDavid Sterba /* 4141387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 4142387125fcSChris Mason * for the barrier when it is done 4143387125fcSChris Mason */ 41444246a0b6SChristoph Hellwig static void btrfs_end_empty_barrier(struct bio *bio) 4145387125fcSChris Mason { 4146f9e69aa9SChristoph Hellwig bio_uninit(bio); 4147387125fcSChris Mason complete(bio->bi_private); 4148387125fcSChris Mason } 4149387125fcSChris Mason 4150387125fcSChris Mason /* 41514fc6441aSAnand Jain * Submit a flush request to the device if it supports it. Error handling is 41524fc6441aSAnand Jain * done in the waiting counterpart. 4153387125fcSChris Mason */ 41544fc6441aSAnand Jain static void write_dev_flush(struct btrfs_device *device) 4155387125fcSChris Mason { 4156f9e69aa9SChristoph Hellwig struct bio *bio = &device->flush_bio; 4157387125fcSChris Mason 4158a91cf0ffSWang Yugui #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY 4159a91cf0ffSWang Yugui /* 4160a91cf0ffSWang Yugui * When a disk has write caching disabled, we skip submission of a bio 4161a91cf0ffSWang Yugui * with flush and sync requests before writing the superblock, since 4162a91cf0ffSWang Yugui * it's not needed. However when the integrity checker is enabled, this 4163a91cf0ffSWang Yugui * results in reports that there are metadata blocks referred by a 4164a91cf0ffSWang Yugui * superblock that were not properly flushed. So don't skip the bio 4165a91cf0ffSWang Yugui * submission only when the integrity checker is enabled for the sake 4166a91cf0ffSWang Yugui * of simplicity, since this is a debug tool and not meant for use in 4167a91cf0ffSWang Yugui * non-debug builds. 4168a91cf0ffSWang Yugui */ 416908e688fdSChristoph Hellwig if (!bdev_write_cache(device->bdev)) 41704fc6441aSAnand Jain return; 4171a91cf0ffSWang Yugui #endif 4172387125fcSChris Mason 4173f9e69aa9SChristoph Hellwig bio_init(bio, device->bdev, NULL, 0, 4174f9e69aa9SChristoph Hellwig REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH); 4175387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 4176387125fcSChris Mason init_completion(&device->flush_wait); 4177387125fcSChris Mason bio->bi_private = &device->flush_wait; 4178387125fcSChris Mason 417958ff51f1SChristoph Hellwig btrfsic_check_bio(bio); 418058ff51f1SChristoph Hellwig submit_bio(bio); 41811c3063b6SAnand Jain set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 41824fc6441aSAnand Jain } 4183387125fcSChris Mason 41844fc6441aSAnand Jain /* 41854fc6441aSAnand Jain * If the flush bio has been submitted by write_dev_flush, wait for it. 41864fc6441aSAnand Jain */ 41878c27cb35SLinus Torvalds static blk_status_t wait_dev_flush(struct btrfs_device *device) 41884fc6441aSAnand Jain { 4189f9e69aa9SChristoph Hellwig struct bio *bio = &device->flush_bio; 41904fc6441aSAnand Jain 41911c3063b6SAnand Jain if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) 419258efbc9fSOmar Sandoval return BLK_STS_OK; 41934fc6441aSAnand Jain 41941c3063b6SAnand Jain clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 41952980d574SDavid Sterba wait_for_completion_io(&device->flush_wait); 41964fc6441aSAnand Jain 41978c27cb35SLinus Torvalds return bio->bi_status; 4198387125fcSChris Mason } 4199387125fcSChris Mason 4200d10b82feSQu Wenruo static int check_barrier_error(struct btrfs_fs_info *fs_info) 4201401b41e5SAnand Jain { 42026528b99dSAnand Jain if (!btrfs_check_rw_degradable(fs_info, NULL)) 4203401b41e5SAnand Jain return -EIO; 4204387125fcSChris Mason return 0; 4205387125fcSChris Mason } 4206387125fcSChris Mason 4207387125fcSChris Mason /* 4208387125fcSChris Mason * send an empty flush down to each device in parallel, 4209387125fcSChris Mason * then wait for them 4210387125fcSChris Mason */ 4211387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 4212387125fcSChris Mason { 4213387125fcSChris Mason struct list_head *head; 4214387125fcSChris Mason struct btrfs_device *dev; 42155af3e8ccSStefan Behrens int errors_wait = 0; 42164e4cbee9SChristoph Hellwig blk_status_t ret; 4217387125fcSChris Mason 42181538e6c5SDavid Sterba lockdep_assert_held(&info->fs_devices->device_list_mutex); 4219387125fcSChris Mason /* send down all the barriers */ 4220387125fcSChris Mason head = &info->fs_devices->devices; 42211538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4222e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 4223f88ba6a2SHidetoshi Seto continue; 4224cea7c8bfSAnand Jain if (!dev->bdev) 4225387125fcSChris Mason continue; 4226e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4227ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4228387125fcSChris Mason continue; 4229387125fcSChris Mason 42304fc6441aSAnand Jain write_dev_flush(dev); 423158efbc9fSOmar Sandoval dev->last_flush_error = BLK_STS_OK; 4232387125fcSChris Mason } 4233387125fcSChris Mason 4234387125fcSChris Mason /* wait for all the barriers */ 42351538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4236e6e674bdSAnand Jain if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 4237f88ba6a2SHidetoshi Seto continue; 4238387125fcSChris Mason if (!dev->bdev) { 42395af3e8ccSStefan Behrens errors_wait++; 4240387125fcSChris Mason continue; 4241387125fcSChris Mason } 4242e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4243ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4244387125fcSChris Mason continue; 4245387125fcSChris Mason 42464fc6441aSAnand Jain ret = wait_dev_flush(dev); 4247401b41e5SAnand Jain if (ret) { 4248401b41e5SAnand Jain dev->last_flush_error = ret; 424966b4993eSDavid Sterba btrfs_dev_stat_inc_and_print(dev, 425066b4993eSDavid Sterba BTRFS_DEV_STAT_FLUSH_ERRS); 42515af3e8ccSStefan Behrens errors_wait++; 4252387125fcSChris Mason } 4253401b41e5SAnand Jain } 4254401b41e5SAnand Jain 4255cea7c8bfSAnand Jain if (errors_wait) { 4256401b41e5SAnand Jain /* 4257401b41e5SAnand Jain * At some point we need the status of all disks 4258401b41e5SAnand Jain * to arrive at the volume status. So error checking 4259401b41e5SAnand Jain * is being pushed to a separate loop. 4260401b41e5SAnand Jain */ 4261d10b82feSQu Wenruo return check_barrier_error(info); 4262401b41e5SAnand Jain } 4263387125fcSChris Mason return 0; 4264387125fcSChris Mason } 4265387125fcSChris Mason 4266943c6e99SZhao Lei int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 4267943c6e99SZhao Lei { 42688789f4feSZhao Lei int raid_type; 42698789f4feSZhao Lei int min_tolerated = INT_MAX; 4270943c6e99SZhao Lei 42718789f4feSZhao Lei if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 42728789f4feSZhao Lei (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 42738c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 42748789f4feSZhao Lei btrfs_raid_array[BTRFS_RAID_SINGLE]. 42758789f4feSZhao Lei tolerated_failures); 4276943c6e99SZhao Lei 42778789f4feSZhao Lei for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 42788789f4feSZhao Lei if (raid_type == BTRFS_RAID_SINGLE) 42798789f4feSZhao Lei continue; 428041a6e891SAnand Jain if (!(flags & btrfs_raid_array[raid_type].bg_flag)) 42818789f4feSZhao Lei continue; 42828c3e3582SDavid Sterba min_tolerated = min_t(int, min_tolerated, 42838789f4feSZhao Lei btrfs_raid_array[raid_type]. 42848789f4feSZhao Lei tolerated_failures); 42858789f4feSZhao Lei } 4286943c6e99SZhao Lei 42878789f4feSZhao Lei if (min_tolerated == INT_MAX) { 4288ab8d0fc4SJeff Mahoney pr_warn("BTRFS: unknown raid flag: %llu", flags); 42898789f4feSZhao Lei min_tolerated = 0; 42908789f4feSZhao Lei } 42918789f4feSZhao Lei 42928789f4feSZhao Lei return min_tolerated; 4293943c6e99SZhao Lei } 4294943c6e99SZhao Lei 4295eece6a9cSDavid Sterba int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) 4296f2984462SChris Mason { 4297e5e9a520SChris Mason struct list_head *head; 4298f2984462SChris Mason struct btrfs_device *dev; 4299a061fc8dSChris Mason struct btrfs_super_block *sb; 4300f2984462SChris Mason struct btrfs_dev_item *dev_item; 4301f2984462SChris Mason int ret; 4302f2984462SChris Mason int do_barriers; 4303a236aed1SChris Mason int max_errors; 4304a236aed1SChris Mason int total_errors = 0; 4305a061fc8dSChris Mason u64 flags; 4306f2984462SChris Mason 43070b246afaSJeff Mahoney do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); 4308fed3b381SLiu Bo 4309fed3b381SLiu Bo /* 4310fed3b381SLiu Bo * max_mirrors == 0 indicates we're from commit_transaction, 4311fed3b381SLiu Bo * not from fsync where the tree roots in fs_info have not 4312fed3b381SLiu Bo * been consistent on disk. 4313fed3b381SLiu Bo */ 4314fed3b381SLiu Bo if (max_mirrors == 0) 43150b246afaSJeff Mahoney backup_super_roots(fs_info); 4316f2984462SChris Mason 43170b246afaSJeff Mahoney sb = fs_info->super_for_commit; 4318a061fc8dSChris Mason dev_item = &sb->dev_item; 4319e5e9a520SChris Mason 43200b246afaSJeff Mahoney mutex_lock(&fs_info->fs_devices->device_list_mutex); 43210b246afaSJeff Mahoney head = &fs_info->fs_devices->devices; 43220b246afaSJeff Mahoney max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; 4323387125fcSChris Mason 43245af3e8ccSStefan Behrens if (do_barriers) { 43250b246afaSJeff Mahoney ret = barrier_all_devices(fs_info); 43265af3e8ccSStefan Behrens if (ret) { 43275af3e8ccSStefan Behrens mutex_unlock( 43280b246afaSJeff Mahoney &fs_info->fs_devices->device_list_mutex); 43290b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 43305af3e8ccSStefan Behrens "errors while submitting device barriers."); 43315af3e8ccSStefan Behrens return ret; 43325af3e8ccSStefan Behrens } 43335af3e8ccSStefan Behrens } 4334387125fcSChris Mason 43351538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4336dfe25020SChris Mason if (!dev->bdev) { 4337dfe25020SChris Mason total_errors++; 4338dfe25020SChris Mason continue; 4339dfe25020SChris Mason } 4340e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4341ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4342dfe25020SChris Mason continue; 4343dfe25020SChris Mason 43442b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 4345a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 4346a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 43477df69d3eSMiao Xie btrfs_set_stack_device_total_bytes(dev_item, 4348935e5cc9SMiao Xie dev->commit_total_bytes); 4349ce7213c7SMiao Xie btrfs_set_stack_device_bytes_used(dev_item, 4350ce7213c7SMiao Xie dev->commit_bytes_used); 4351a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 4352a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 4353a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 4354a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 43557239ff4bSNikolay Borisov memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, 43567239ff4bSNikolay Borisov BTRFS_FSID_SIZE); 4357a512bbf8SYan Zheng 4358a061fc8dSChris Mason flags = btrfs_super_flags(sb); 4359a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 4360f2984462SChris Mason 436175cb857dSQu Wenruo ret = btrfs_validate_write_super(fs_info, sb); 436275cb857dSQu Wenruo if (ret < 0) { 436375cb857dSQu Wenruo mutex_unlock(&fs_info->fs_devices->device_list_mutex); 436475cb857dSQu Wenruo btrfs_handle_fs_error(fs_info, -EUCLEAN, 436575cb857dSQu Wenruo "unexpected superblock corruption detected"); 436675cb857dSQu Wenruo return -EUCLEAN; 436775cb857dSQu Wenruo } 436875cb857dSQu Wenruo 4369abbb3b8eSDavid Sterba ret = write_dev_supers(dev, sb, max_mirrors); 4370a236aed1SChris Mason if (ret) 4371a236aed1SChris Mason total_errors++; 4372f2984462SChris Mason } 4373a236aed1SChris Mason if (total_errors > max_errors) { 43740b246afaSJeff Mahoney btrfs_err(fs_info, "%d errors while writing supers", 4375d397712bSChris Mason total_errors); 43760b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 437779787eaaSJeff Mahoney 43789d565ba4SStefan Behrens /* FUA is masked off if unsupported and can't be the reason */ 43790b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 43800b246afaSJeff Mahoney "%d errors while writing supers", 43810b246afaSJeff Mahoney total_errors); 43829d565ba4SStefan Behrens return -EIO; 4383a236aed1SChris Mason } 4384f2984462SChris Mason 4385a512bbf8SYan Zheng total_errors = 0; 43861538e6c5SDavid Sterba list_for_each_entry(dev, head, dev_list) { 4387dfe25020SChris Mason if (!dev->bdev) 4388dfe25020SChris Mason continue; 4389e12c9621SAnand Jain if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4390ebbede42SAnand Jain !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4391dfe25020SChris Mason continue; 4392dfe25020SChris Mason 4393abbb3b8eSDavid Sterba ret = wait_dev_supers(dev, max_mirrors); 4394a512bbf8SYan Zheng if (ret) 43951259ab75SChris Mason total_errors++; 4396f2984462SChris Mason } 43970b246afaSJeff Mahoney mutex_unlock(&fs_info->fs_devices->device_list_mutex); 4398a236aed1SChris Mason if (total_errors > max_errors) { 43990b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, -EIO, 44000b246afaSJeff Mahoney "%d errors while writing supers", 44010b246afaSJeff Mahoney total_errors); 440279787eaaSJeff Mahoney return -EIO; 4403a236aed1SChris Mason } 4404f2984462SChris Mason return 0; 4405f2984462SChris Mason } 4406f2984462SChris Mason 4407cb517eabSMiao Xie /* Drop a fs root from the radix tree and free it. */ 4408cb517eabSMiao Xie void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 4409cb517eabSMiao Xie struct btrfs_root *root) 44102619ba1fSChris Mason { 44114785e24fSJosef Bacik bool drop_ref = false; 44124785e24fSJosef Bacik 4413fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4414fc7cbcd4SDavid Sterba radix_tree_delete(&fs_info->fs_roots_radix, 4415fc7cbcd4SDavid Sterba (unsigned long)root->root_key.objectid); 4416fc7cbcd4SDavid Sterba if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state)) 44174785e24fSJosef Bacik drop_ref = true; 4418fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 441976dda93cSYan, Zheng 442084961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 4421ef67963dSJosef Bacik ASSERT(root->log_root == NULL); 44221c1ea4f7SLiu Bo if (root->reloc_root) { 442300246528SJosef Bacik btrfs_put_root(root->reloc_root); 44241c1ea4f7SLiu Bo root->reloc_root = NULL; 44251c1ea4f7SLiu Bo } 44261c1ea4f7SLiu Bo } 44273321719eSLiu Bo 44284785e24fSJosef Bacik if (drop_ref) 442900246528SJosef Bacik btrfs_put_root(root); 44302619ba1fSChris Mason } 44312619ba1fSChris Mason 4432c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 4433c146afadSYan Zheng { 4434fc7cbcd4SDavid Sterba u64 root_objectid = 0; 4435fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 4436fc7cbcd4SDavid Sterba int i = 0; 443765d33fd7SQu Wenruo int err = 0; 4438fc7cbcd4SDavid Sterba unsigned int ret = 0; 4439c146afadSYan Zheng 4440c146afadSYan Zheng while (1) { 4441fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4442fc7cbcd4SDavid Sterba ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4443fc7cbcd4SDavid Sterba (void **)gang, root_objectid, 4444fc7cbcd4SDavid Sterba ARRAY_SIZE(gang)); 4445fc7cbcd4SDavid Sterba if (!ret) { 4446fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4447c146afadSYan Zheng break; 444865d33fd7SQu Wenruo } 4449fc7cbcd4SDavid Sterba root_objectid = gang[ret - 1]->root_key.objectid + 1; 445066b4ffd1SJosef Bacik 4451fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 4452fc7cbcd4SDavid Sterba /* Avoid to grab roots in dead_roots */ 4453fc7cbcd4SDavid Sterba if (btrfs_root_refs(&gang[i]->root_item) == 0) { 4454fc7cbcd4SDavid Sterba gang[i] = NULL; 445565d33fd7SQu Wenruo continue; 4456c146afadSYan Zheng } 4457fc7cbcd4SDavid Sterba /* grab all the search result for later use */ 4458fc7cbcd4SDavid Sterba gang[i] = btrfs_grab_root(gang[i]); 4459fc7cbcd4SDavid Sterba } 4460fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4461fc7cbcd4SDavid Sterba 4462fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 4463fc7cbcd4SDavid Sterba if (!gang[i]) 4464fc7cbcd4SDavid Sterba continue; 4465fc7cbcd4SDavid Sterba root_objectid = gang[i]->root_key.objectid; 4466fc7cbcd4SDavid Sterba err = btrfs_orphan_cleanup(gang[i]); 4467fc7cbcd4SDavid Sterba if (err) 4468fc7cbcd4SDavid Sterba break; 4469fc7cbcd4SDavid Sterba btrfs_put_root(gang[i]); 4470fc7cbcd4SDavid Sterba } 4471fc7cbcd4SDavid Sterba root_objectid++; 4472c146afadSYan Zheng } 447365d33fd7SQu Wenruo 4474fc7cbcd4SDavid Sterba /* release the uncleaned roots due to error */ 4475fc7cbcd4SDavid Sterba for (; i < ret; i++) { 4476fc7cbcd4SDavid Sterba if (gang[i]) 4477fc7cbcd4SDavid Sterba btrfs_put_root(gang[i]); 447865d33fd7SQu Wenruo } 447965d33fd7SQu Wenruo return err; 4480c146afadSYan Zheng } 4481c146afadSYan Zheng 44826bccf3abSJeff Mahoney int btrfs_commit_super(struct btrfs_fs_info *fs_info) 4483c146afadSYan Zheng { 44846bccf3abSJeff Mahoney struct btrfs_root *root = fs_info->tree_root; 4485c146afadSYan Zheng struct btrfs_trans_handle *trans; 4486c146afadSYan Zheng 44870b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 44882ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 44890b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 44900b246afaSJeff Mahoney wake_up_process(fs_info->cleaner_kthread); 4491c71bf099SYan, Zheng 4492c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 44930b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 44940b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4495c71bf099SYan, Zheng 44967a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 44973612b495STsutomu Itoh if (IS_ERR(trans)) 44983612b495STsutomu Itoh return PTR_ERR(trans); 44993a45bb20SJeff Mahoney return btrfs_commit_transaction(trans); 4500c146afadSYan Zheng } 4501c146afadSYan Zheng 450236c86a9eSQu Wenruo static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info) 450336c86a9eSQu Wenruo { 450436c86a9eSQu Wenruo struct btrfs_transaction *trans; 450536c86a9eSQu Wenruo struct btrfs_transaction *tmp; 450636c86a9eSQu Wenruo bool found = false; 450736c86a9eSQu Wenruo 450836c86a9eSQu Wenruo if (list_empty(&fs_info->trans_list)) 450936c86a9eSQu Wenruo return; 451036c86a9eSQu Wenruo 451136c86a9eSQu Wenruo /* 451236c86a9eSQu Wenruo * This function is only called at the very end of close_ctree(), 451336c86a9eSQu Wenruo * thus no other running transaction, no need to take trans_lock. 451436c86a9eSQu Wenruo */ 451536c86a9eSQu Wenruo ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags)); 451636c86a9eSQu Wenruo list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) { 451736c86a9eSQu Wenruo struct extent_state *cached = NULL; 451836c86a9eSQu Wenruo u64 dirty_bytes = 0; 451936c86a9eSQu Wenruo u64 cur = 0; 452036c86a9eSQu Wenruo u64 found_start; 452136c86a9eSQu Wenruo u64 found_end; 452236c86a9eSQu Wenruo 452336c86a9eSQu Wenruo found = true; 452436c86a9eSQu Wenruo while (!find_first_extent_bit(&trans->dirty_pages, cur, 452536c86a9eSQu Wenruo &found_start, &found_end, EXTENT_DIRTY, &cached)) { 452636c86a9eSQu Wenruo dirty_bytes += found_end + 1 - found_start; 452736c86a9eSQu Wenruo cur = found_end + 1; 452836c86a9eSQu Wenruo } 452936c86a9eSQu Wenruo btrfs_warn(fs_info, 453036c86a9eSQu Wenruo "transaction %llu (with %llu dirty metadata bytes) is not committed", 453136c86a9eSQu Wenruo trans->transid, dirty_bytes); 453236c86a9eSQu Wenruo btrfs_cleanup_one_transaction(trans, fs_info); 453336c86a9eSQu Wenruo 453436c86a9eSQu Wenruo if (trans == fs_info->running_transaction) 453536c86a9eSQu Wenruo fs_info->running_transaction = NULL; 453636c86a9eSQu Wenruo list_del_init(&trans->list); 453736c86a9eSQu Wenruo 453836c86a9eSQu Wenruo btrfs_put_transaction(trans); 453936c86a9eSQu Wenruo trace_btrfs_transaction_commit(fs_info); 454036c86a9eSQu Wenruo } 454136c86a9eSQu Wenruo ASSERT(!found); 454236c86a9eSQu Wenruo } 454336c86a9eSQu Wenruo 4544b105e927SDavid Sterba void __cold close_ctree(struct btrfs_fs_info *fs_info) 4545eb60ceacSChris Mason { 4546c146afadSYan Zheng int ret; 4547e089f05cSChris Mason 4548afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); 454931e70e52SFilipe Manana 455031e70e52SFilipe Manana /* 45518a1f1e3dSFilipe Manana * If we had UNFINISHED_DROPS we could still be processing them, so 45528a1f1e3dSFilipe Manana * clear that bit and wake up relocation so it can stop. 45538a1f1e3dSFilipe Manana * We must do this before stopping the block group reclaim task, because 45548a1f1e3dSFilipe Manana * at btrfs_relocate_block_group() we wait for this bit, and after the 45558a1f1e3dSFilipe Manana * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we 45568a1f1e3dSFilipe Manana * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will 45578a1f1e3dSFilipe Manana * return 1. 45588a1f1e3dSFilipe Manana */ 45598a1f1e3dSFilipe Manana btrfs_wake_unfinished_drop(fs_info); 45608a1f1e3dSFilipe Manana 45618a1f1e3dSFilipe Manana /* 456231e70e52SFilipe Manana * We may have the reclaim task running and relocating a data block group, 456331e70e52SFilipe Manana * in which case it may create delayed iputs. So stop it before we park 456431e70e52SFilipe Manana * the cleaner kthread otherwise we can get new delayed iputs after 456531e70e52SFilipe Manana * parking the cleaner, and that can make the async reclaim task to hang 456631e70e52SFilipe Manana * if it's waiting for delayed iputs to complete, since the cleaner is 456731e70e52SFilipe Manana * parked and can not run delayed iputs - this will make us hang when 456831e70e52SFilipe Manana * trying to stop the async reclaim task. 456931e70e52SFilipe Manana */ 457031e70e52SFilipe Manana cancel_work_sync(&fs_info->reclaim_bgs_work); 4571d6fd0ae2SOmar Sandoval /* 4572d6fd0ae2SOmar Sandoval * We don't want the cleaner to start new transactions, add more delayed 4573d6fd0ae2SOmar Sandoval * iputs, etc. while we're closing. We can't use kthread_stop() yet 4574d6fd0ae2SOmar Sandoval * because that frees the task_struct, and the transaction kthread might 4575d6fd0ae2SOmar Sandoval * still try to wake up the cleaner. 4576d6fd0ae2SOmar Sandoval */ 4577d6fd0ae2SOmar Sandoval kthread_park(fs_info->cleaner_kthread); 4578a2135011SChris Mason 45797343dd61SJustin Maggard /* wait for the qgroup rescan worker to stop */ 4580d06f23d6SJeff Mahoney btrfs_qgroup_wait_for_completion(fs_info, false); 45817343dd61SJustin Maggard 4582803b2f54SStefan Behrens /* wait for the uuid_scan task to finish */ 4583803b2f54SStefan Behrens down(&fs_info->uuid_tree_rescan_sem); 4584803b2f54SStefan Behrens /* avoid complains from lockdep et al., set sem back to initial state */ 4585803b2f54SStefan Behrens up(&fs_info->uuid_tree_rescan_sem); 4586803b2f54SStefan Behrens 4587837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 4588aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 4589837d5b6eSIlya Dryomov 45908dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 45918dabb742SStefan Behrens 4592aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 45934cb5300bSChris Mason 45944cb5300bSChris Mason /* wait for any defraggers to finish */ 45954cb5300bSChris Mason wait_event(fs_info->transaction_wait, 45964cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 45974cb5300bSChris Mason 45984cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 459926176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 46004cb5300bSChris Mason 4601a362bb86SFilipe Manana /* 4602a362bb86SFilipe Manana * After we parked the cleaner kthread, ordered extents may have 4603a362bb86SFilipe Manana * completed and created new delayed iputs. If one of the async reclaim 4604a362bb86SFilipe Manana * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we 4605a362bb86SFilipe Manana * can hang forever trying to stop it, because if a delayed iput is 4606a362bb86SFilipe Manana * added after it ran btrfs_run_delayed_iputs() and before it called 4607a362bb86SFilipe Manana * btrfs_wait_on_delayed_iputs(), it will hang forever since there is 4608a362bb86SFilipe Manana * no one else to run iputs. 4609a362bb86SFilipe Manana * 4610a362bb86SFilipe Manana * So wait for all ongoing ordered extents to complete and then run 4611a362bb86SFilipe Manana * delayed iputs. This works because once we reach this point no one 4612a362bb86SFilipe Manana * can either create new ordered extents nor create delayed iputs 4613a362bb86SFilipe Manana * through some other means. 4614a362bb86SFilipe Manana * 4615a362bb86SFilipe Manana * Also note that btrfs_wait_ordered_roots() is not safe here, because 4616a362bb86SFilipe Manana * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent, 4617a362bb86SFilipe Manana * but the delayed iput for the respective inode is made only when doing 4618a362bb86SFilipe Manana * the final btrfs_put_ordered_extent() (which must happen at 4619a362bb86SFilipe Manana * btrfs_finish_ordered_io() when we are unmounting). 4620a362bb86SFilipe Manana */ 4621a362bb86SFilipe Manana btrfs_flush_workqueue(fs_info->endio_write_workers); 4622a362bb86SFilipe Manana /* Ordered extents for free space inodes. */ 4623a362bb86SFilipe Manana btrfs_flush_workqueue(fs_info->endio_freespace_worker); 4624a362bb86SFilipe Manana btrfs_run_delayed_iputs(fs_info); 4625a362bb86SFilipe Manana 462621c7e756SMiao Xie cancel_work_sync(&fs_info->async_reclaim_work); 462757056740SJosef Bacik cancel_work_sync(&fs_info->async_data_reclaim_work); 4628576fa348SJosef Bacik cancel_work_sync(&fs_info->preempt_reclaim_work); 462921c7e756SMiao Xie 4630b0643e59SDennis Zhou /* Cancel or finish ongoing discard work */ 4631b0643e59SDennis Zhou btrfs_discard_cleanup(fs_info); 4632b0643e59SDennis Zhou 4633bc98a42cSDavid Howells if (!sb_rdonly(fs_info->sb)) { 4634e44163e1SJeff Mahoney /* 4635d6fd0ae2SOmar Sandoval * The cleaner kthread is stopped, so do one final pass over 4636d6fd0ae2SOmar Sandoval * unused block groups. 4637e44163e1SJeff Mahoney */ 46380b246afaSJeff Mahoney btrfs_delete_unused_bgs(fs_info); 4639e44163e1SJeff Mahoney 4640f0cc2cd7SFilipe Manana /* 4641f0cc2cd7SFilipe Manana * There might be existing delayed inode workers still running 4642f0cc2cd7SFilipe Manana * and holding an empty delayed inode item. We must wait for 4643f0cc2cd7SFilipe Manana * them to complete first because they can create a transaction. 4644f0cc2cd7SFilipe Manana * This happens when someone calls btrfs_balance_delayed_items() 4645f0cc2cd7SFilipe Manana * and then a transaction commit runs the same delayed nodes 4646f0cc2cd7SFilipe Manana * before any delayed worker has done something with the nodes. 4647f0cc2cd7SFilipe Manana * We must wait for any worker here and not at transaction 4648f0cc2cd7SFilipe Manana * commit time since that could cause a deadlock. 4649f0cc2cd7SFilipe Manana * This is a very rare case. 4650f0cc2cd7SFilipe Manana */ 4651f0cc2cd7SFilipe Manana btrfs_flush_workqueue(fs_info->delayed_workers); 4652f0cc2cd7SFilipe Manana 46536bccf3abSJeff Mahoney ret = btrfs_commit_super(fs_info); 4654d397712bSChris Mason if (ret) 465504892340SEric Sandeen btrfs_err(fs_info, "commit super ret %d", ret); 4656c146afadSYan Zheng } 4657ed2ff2cbSChris Mason 465884961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) 46592ff7e61eSJeff Mahoney btrfs_error_commit_super(fs_info); 4660acce952bSliubo 4661e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 4662e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 46638929ecfaSYan, Zheng 4664e187831eSJosef Bacik ASSERT(list_empty(&fs_info->delayed_iputs)); 4665afcdd129SJosef Bacik set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); 4666f25784b3SYan Zheng 46675958253cSQu Wenruo if (btrfs_check_quota_leak(fs_info)) { 46685958253cSQu Wenruo WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 46695958253cSQu Wenruo btrfs_err(fs_info, "qgroup reserved space leaked"); 46705958253cSQu Wenruo } 46715958253cSQu Wenruo 467204892340SEric Sandeen btrfs_free_qgroup_config(fs_info); 4673fe816d0fSNikolay Borisov ASSERT(list_empty(&fs_info->delalloc_roots)); 4674bcef60f2SArne Jansen 4675963d678bSMiao Xie if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 467604892340SEric Sandeen btrfs_info(fs_info, "at unmount delalloc count %lld", 4677963d678bSMiao Xie percpu_counter_sum(&fs_info->delalloc_bytes)); 4678b0c68f8bSChris Mason } 467931153d81SYan Zheng 46805deb17e1SJosef Bacik if (percpu_counter_sum(&fs_info->ordered_bytes)) 46814297ff84SJosef Bacik btrfs_info(fs_info, "at unmount dio bytes count %lld", 46825deb17e1SJosef Bacik percpu_counter_sum(&fs_info->ordered_bytes)); 46834297ff84SJosef Bacik 46846618a59bSAnand Jain btrfs_sysfs_remove_mounted(fs_info); 4685b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 46865ac1d209SJeff Mahoney 46871a4319ccSLiu Bo btrfs_put_block_group_cache(fs_info); 46881a4319ccSLiu Bo 4689de348ee0SWang Shilong /* 4690de348ee0SWang Shilong * we must make sure there is not any read request to 4691de348ee0SWang Shilong * submit after we stopping all workers. 4692de348ee0SWang Shilong */ 4693de348ee0SWang Shilong invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 469496192499SJosef Bacik btrfs_stop_all_workers(fs_info); 469596192499SJosef Bacik 46960a31daa4SFilipe Manana /* We shouldn't have any transaction open at this point */ 469736c86a9eSQu Wenruo warn_about_uncommitted_trans(fs_info); 46980a31daa4SFilipe Manana 4699afcdd129SJosef Bacik clear_bit(BTRFS_FS_OPEN, &fs_info->flags); 47004273eaffSAnand Jain free_root_pointers(fs_info, true); 47018c38938cSJosef Bacik btrfs_free_fs_roots(fs_info); 47029ad6b7bcSChris Mason 47034e19443dSJosef Bacik /* 47044e19443dSJosef Bacik * We must free the block groups after dropping the fs_roots as we could 47054e19443dSJosef Bacik * have had an IO error and have left over tree log blocks that aren't 47064e19443dSJosef Bacik * cleaned up until the fs roots are freed. This makes the block group 47074e19443dSJosef Bacik * accounting appear to be wrong because there's pending reserved bytes, 47084e19443dSJosef Bacik * so make sure we do the block group cleanup afterwards. 47094e19443dSJosef Bacik */ 47104e19443dSJosef Bacik btrfs_free_block_groups(fs_info); 47114e19443dSJosef Bacik 471213e6c37bSJosef Bacik iput(fs_info->btree_inode); 4713d6bfde87SChris Mason 471421adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 47150b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) 47162ff7e61eSJeff Mahoney btrfsic_unmount(fs_info->fs_devices); 471721adbd5cSStefan Behrens #endif 471821adbd5cSStefan Behrens 47190b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 472068c94e55SNikolay Borisov btrfs_close_devices(fs_info->fs_devices); 4721eb60ceacSChris Mason } 4722eb60ceacSChris Mason 4723b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 4724b9fab919SChris Mason int atomic) 4725ccd467d6SChris Mason { 47261259ab75SChris Mason int ret; 4727727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 47281259ab75SChris Mason 47290b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 47301259ab75SChris Mason if (!ret) 47311259ab75SChris Mason return ret; 47321259ab75SChris Mason 47331259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 4734b9fab919SChris Mason parent_transid, atomic); 4735b9fab919SChris Mason if (ret == -EAGAIN) 4736b9fab919SChris Mason return ret; 47371259ab75SChris Mason return !ret; 47385f39d397SChris Mason } 47396702ed49SChris Mason 47405f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 47415f39d397SChris Mason { 47422f4d60dfSQu Wenruo struct btrfs_fs_info *fs_info = buf->fs_info; 47435f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 4744b9473439SChris Mason int was_dirty; 4745b4ce94deSChris Mason 474606ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 474706ea65a3SJosef Bacik /* 474806ea65a3SJosef Bacik * This is a fast path so only do this check if we have sanity tests 474952042d8eSAndrea Gelmini * enabled. Normal people shouldn't be using unmapped buffers as dirty 475006ea65a3SJosef Bacik * outside of the sanity tests. 475106ea65a3SJosef Bacik */ 4752b0132a3bSNikolay Borisov if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) 475306ea65a3SJosef Bacik return; 475406ea65a3SJosef Bacik #endif 475549d0c642SFilipe Manana btrfs_assert_tree_write_locked(buf); 47560b246afaSJeff Mahoney if (transid != fs_info->generation) 47575d163e0eSJeff Mahoney WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", 47580b246afaSJeff Mahoney buf->start, transid, fs_info->generation); 47590b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 4760e2d84521SMiao Xie if (!was_dirty) 4761104b4e51SNikolay Borisov percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 4762e2d84521SMiao Xie buf->len, 47630b246afaSJeff Mahoney fs_info->dirty_metadata_batch); 47641f21ef0aSFilipe Manana #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 476569fc6cbbSQu Wenruo /* 476669fc6cbbSQu Wenruo * Since btrfs_mark_buffer_dirty() can be called with item pointer set 476769fc6cbbSQu Wenruo * but item data not updated. 476869fc6cbbSQu Wenruo * So here we should only check item pointers, not item data. 476969fc6cbbSQu Wenruo */ 477069fc6cbbSQu Wenruo if (btrfs_header_level(buf) == 0 && 4771cfdaad5eSDavid Sterba btrfs_check_leaf_relaxed(buf)) { 4772a4f78750SDavid Sterba btrfs_print_leaf(buf); 47731f21ef0aSFilipe Manana ASSERT(0); 47741f21ef0aSFilipe Manana } 47751f21ef0aSFilipe Manana #endif 4776eb60ceacSChris Mason } 4777eb60ceacSChris Mason 47782ff7e61eSJeff Mahoney static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, 4779b53d3f5dSLiu Bo int flush_delayed) 478035b7e476SChris Mason { 4781188de649SChris Mason /* 4782188de649SChris Mason * looks as though older kernels can get into trouble with 4783188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 4784188de649SChris Mason */ 4785e2d84521SMiao Xie int ret; 4786d6bfde87SChris Mason 47876933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 4788d6bfde87SChris Mason return; 4789d6bfde87SChris Mason 4790b53d3f5dSLiu Bo if (flush_delayed) 47912ff7e61eSJeff Mahoney btrfs_balance_delayed_items(fs_info); 479216cdcec7SMiao Xie 4793d814a491SEthan Lien ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 4794d814a491SEthan Lien BTRFS_DIRTY_METADATA_THRESH, 4795d814a491SEthan Lien fs_info->dirty_metadata_batch); 4796e2d84521SMiao Xie if (ret > 0) { 47970b246afaSJeff Mahoney balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); 479816cdcec7SMiao Xie } 479916cdcec7SMiao Xie } 480016cdcec7SMiao Xie 48012ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) 480216cdcec7SMiao Xie { 48032ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 1); 480435b7e476SChris Mason } 4805b53d3f5dSLiu Bo 48062ff7e61eSJeff Mahoney void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) 4807b53d3f5dSLiu Bo { 48082ff7e61eSJeff Mahoney __btrfs_btree_balance_dirty(fs_info, 0); 4809d6bfde87SChris Mason } 48106b80053dSChris Mason 48112ff7e61eSJeff Mahoney static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) 4812acce952bSliubo { 4813fe816d0fSNikolay Borisov /* cleanup FS via transaction */ 4814fe816d0fSNikolay Borisov btrfs_cleanup_transaction(fs_info); 4815fe816d0fSNikolay Borisov 48160b246afaSJeff Mahoney mutex_lock(&fs_info->cleaner_mutex); 48172ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(fs_info); 48180b246afaSJeff Mahoney mutex_unlock(&fs_info->cleaner_mutex); 4819acce952bSliubo 48200b246afaSJeff Mahoney down_write(&fs_info->cleanup_work_sem); 48210b246afaSJeff Mahoney up_write(&fs_info->cleanup_work_sem); 4822acce952bSliubo } 4823acce952bSliubo 4824ef67963dSJosef Bacik static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info) 4825ef67963dSJosef Bacik { 4826fc7cbcd4SDavid Sterba struct btrfs_root *gang[8]; 4827fc7cbcd4SDavid Sterba u64 root_objectid = 0; 4828fc7cbcd4SDavid Sterba int ret; 4829ef67963dSJosef Bacik 4830fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4831fc7cbcd4SDavid Sterba while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4832fc7cbcd4SDavid Sterba (void **)gang, root_objectid, 4833fc7cbcd4SDavid Sterba ARRAY_SIZE(gang))) != 0) { 4834fc7cbcd4SDavid Sterba int i; 4835ef67963dSJosef Bacik 4836fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) 4837fc7cbcd4SDavid Sterba gang[i] = btrfs_grab_root(gang[i]); 4838fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4839fc7cbcd4SDavid Sterba 4840fc7cbcd4SDavid Sterba for (i = 0; i < ret; i++) { 4841fc7cbcd4SDavid Sterba if (!gang[i]) 4842ef67963dSJosef Bacik continue; 4843fc7cbcd4SDavid Sterba root_objectid = gang[i]->root_key.objectid; 4844fc7cbcd4SDavid Sterba btrfs_free_log(NULL, gang[i]); 4845fc7cbcd4SDavid Sterba btrfs_put_root(gang[i]); 4846ef67963dSJosef Bacik } 4847fc7cbcd4SDavid Sterba root_objectid++; 4848fc7cbcd4SDavid Sterba spin_lock(&fs_info->fs_roots_radix_lock); 4849ef67963dSJosef Bacik } 4850fc7cbcd4SDavid Sterba spin_unlock(&fs_info->fs_roots_radix_lock); 4851ef67963dSJosef Bacik btrfs_free_log_root_tree(NULL, fs_info); 4852ef67963dSJosef Bacik } 4853ef67963dSJosef Bacik 4854143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4855acce952bSliubo { 4856acce952bSliubo struct btrfs_ordered_extent *ordered; 4857acce952bSliubo 4858199c2a9cSMiao Xie spin_lock(&root->ordered_extent_lock); 4859779880efSJosef Bacik /* 4860779880efSJosef Bacik * This will just short circuit the ordered completion stuff which will 4861779880efSJosef Bacik * make sure the ordered extent gets properly cleaned up. 4862779880efSJosef Bacik */ 4863199c2a9cSMiao Xie list_for_each_entry(ordered, &root->ordered_extents, 4864779880efSJosef Bacik root_extent_list) 4865779880efSJosef Bacik set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4866199c2a9cSMiao Xie spin_unlock(&root->ordered_extent_lock); 4867199c2a9cSMiao Xie } 4868199c2a9cSMiao Xie 4869199c2a9cSMiao Xie static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4870199c2a9cSMiao Xie { 4871199c2a9cSMiao Xie struct btrfs_root *root; 4872199c2a9cSMiao Xie struct list_head splice; 4873199c2a9cSMiao Xie 4874199c2a9cSMiao Xie INIT_LIST_HEAD(&splice); 4875199c2a9cSMiao Xie 4876199c2a9cSMiao Xie spin_lock(&fs_info->ordered_root_lock); 4877199c2a9cSMiao Xie list_splice_init(&fs_info->ordered_roots, &splice); 4878199c2a9cSMiao Xie while (!list_empty(&splice)) { 4879199c2a9cSMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4880199c2a9cSMiao Xie ordered_root); 48811de2cfdeSJosef Bacik list_move_tail(&root->ordered_root, 48821de2cfdeSJosef Bacik &fs_info->ordered_roots); 4883199c2a9cSMiao Xie 48842a85d9caSLiu Bo spin_unlock(&fs_info->ordered_root_lock); 4885199c2a9cSMiao Xie btrfs_destroy_ordered_extents(root); 4886199c2a9cSMiao Xie 48872a85d9caSLiu Bo cond_resched(); 48882a85d9caSLiu Bo spin_lock(&fs_info->ordered_root_lock); 4889199c2a9cSMiao Xie } 4890199c2a9cSMiao Xie spin_unlock(&fs_info->ordered_root_lock); 489174d5d229SJosef Bacik 489274d5d229SJosef Bacik /* 489374d5d229SJosef Bacik * We need this here because if we've been flipped read-only we won't 489474d5d229SJosef Bacik * get sync() from the umount, so we need to make sure any ordered 489574d5d229SJosef Bacik * extents that haven't had their dirty pages IO start writeout yet 489674d5d229SJosef Bacik * actually get run and error out properly. 489774d5d229SJosef Bacik */ 489874d5d229SJosef Bacik btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 4899acce952bSliubo } 4900acce952bSliubo 490135a3621bSStefan Behrens static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 49022ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 4903acce952bSliubo { 4904acce952bSliubo struct rb_node *node; 4905acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 4906acce952bSliubo struct btrfs_delayed_ref_node *ref; 4907acce952bSliubo int ret = 0; 4908acce952bSliubo 4909acce952bSliubo delayed_refs = &trans->delayed_refs; 4910acce952bSliubo 4911acce952bSliubo spin_lock(&delayed_refs->lock); 4912d7df2c79SJosef Bacik if (atomic_read(&delayed_refs->num_entries) == 0) { 4913cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 4914b79ce3ddSDavid Sterba btrfs_debug(fs_info, "delayed_refs has NO entry"); 4915acce952bSliubo return ret; 4916acce952bSliubo } 4917acce952bSliubo 49185c9d028bSLiu Bo while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { 4919d7df2c79SJosef Bacik struct btrfs_delayed_ref_head *head; 49200e0adbcfSJosef Bacik struct rb_node *n; 4921e78417d1SJosef Bacik bool pin_bytes = false; 4922acce952bSliubo 4923d7df2c79SJosef Bacik head = rb_entry(node, struct btrfs_delayed_ref_head, 4924d7df2c79SJosef Bacik href_node); 49253069bd26SJosef Bacik if (btrfs_delayed_ref_lock(delayed_refs, head)) 4926b939d1abSJosef Bacik continue; 49273069bd26SJosef Bacik 4928d7df2c79SJosef Bacik spin_lock(&head->lock); 4929e3d03965SLiu Bo while ((n = rb_first_cached(&head->ref_tree)) != NULL) { 49300e0adbcfSJosef Bacik ref = rb_entry(n, struct btrfs_delayed_ref_node, 49310e0adbcfSJosef Bacik ref_node); 4932d7df2c79SJosef Bacik ref->in_tree = 0; 4933e3d03965SLiu Bo rb_erase_cached(&ref->ref_node, &head->ref_tree); 49340e0adbcfSJosef Bacik RB_CLEAR_NODE(&ref->ref_node); 49351d57ee94SWang Xiaoguang if (!list_empty(&ref->add_list)) 49361d57ee94SWang Xiaoguang list_del(&ref->add_list); 4937d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4938d7df2c79SJosef Bacik btrfs_put_delayed_ref(ref); 4939d7df2c79SJosef Bacik } 494054067ae9SJosef Bacik if (head->must_insert_reserved) 4941e78417d1SJosef Bacik pin_bytes = true; 494278a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 4943fa781ceaSJosef Bacik btrfs_delete_ref_head(delayed_refs, head); 4944d7df2c79SJosef Bacik spin_unlock(&head->lock); 4945acce952bSliubo spin_unlock(&delayed_refs->lock); 4946e78417d1SJosef Bacik mutex_unlock(&head->mutex); 4947acce952bSliubo 4948f603bb94SNikolay Borisov if (pin_bytes) { 4949f603bb94SNikolay Borisov struct btrfs_block_group *cache; 4950f603bb94SNikolay Borisov 4951f603bb94SNikolay Borisov cache = btrfs_lookup_block_group(fs_info, head->bytenr); 4952f603bb94SNikolay Borisov BUG_ON(!cache); 4953f603bb94SNikolay Borisov 4954f603bb94SNikolay Borisov spin_lock(&cache->space_info->lock); 4955f603bb94SNikolay Borisov spin_lock(&cache->lock); 4956f603bb94SNikolay Borisov cache->pinned += head->num_bytes; 4957f603bb94SNikolay Borisov btrfs_space_info_update_bytes_pinned(fs_info, 4958f603bb94SNikolay Borisov cache->space_info, head->num_bytes); 4959f603bb94SNikolay Borisov cache->reserved -= head->num_bytes; 4960f603bb94SNikolay Borisov cache->space_info->bytes_reserved -= head->num_bytes; 4961f603bb94SNikolay Borisov spin_unlock(&cache->lock); 4962f603bb94SNikolay Borisov spin_unlock(&cache->space_info->lock); 4963f603bb94SNikolay Borisov 4964f603bb94SNikolay Borisov btrfs_put_block_group(cache); 4965f603bb94SNikolay Borisov 4966f603bb94SNikolay Borisov btrfs_error_unpin_extent_range(fs_info, head->bytenr, 4967f603bb94SNikolay Borisov head->bytenr + head->num_bytes - 1); 4968f603bb94SNikolay Borisov } 496931890da0SJosef Bacik btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); 4970d278850eSJosef Bacik btrfs_put_delayed_ref_head(head); 4971acce952bSliubo cond_resched(); 4972acce952bSliubo spin_lock(&delayed_refs->lock); 4973acce952bSliubo } 497481f7eb00SJeff Mahoney btrfs_qgroup_destroy_extent_records(trans); 4975acce952bSliubo 4976acce952bSliubo spin_unlock(&delayed_refs->lock); 4977acce952bSliubo 4978acce952bSliubo return ret; 4979acce952bSliubo } 4980acce952bSliubo 4981143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 4982acce952bSliubo { 4983acce952bSliubo struct btrfs_inode *btrfs_inode; 4984acce952bSliubo struct list_head splice; 4985acce952bSliubo 4986acce952bSliubo INIT_LIST_HEAD(&splice); 4987acce952bSliubo 4988eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4989eb73c1b7SMiao Xie list_splice_init(&root->delalloc_inodes, &splice); 4990acce952bSliubo 4991acce952bSliubo while (!list_empty(&splice)) { 4992fe816d0fSNikolay Borisov struct inode *inode = NULL; 4993eb73c1b7SMiao Xie btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 4994acce952bSliubo delalloc_inodes); 4995fe816d0fSNikolay Borisov __btrfs_del_delalloc_inode(root, btrfs_inode); 4996eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4997acce952bSliubo 4998fe816d0fSNikolay Borisov /* 4999fe816d0fSNikolay Borisov * Make sure we get a live inode and that it'll not disappear 5000fe816d0fSNikolay Borisov * meanwhile. 5001fe816d0fSNikolay Borisov */ 5002fe816d0fSNikolay Borisov inode = igrab(&btrfs_inode->vfs_inode); 5003fe816d0fSNikolay Borisov if (inode) { 5004fe816d0fSNikolay Borisov invalidate_inode_pages2(inode->i_mapping); 5005fe816d0fSNikolay Borisov iput(inode); 5006fe816d0fSNikolay Borisov } 5007eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 5008acce952bSliubo } 5009eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 5010eb73c1b7SMiao Xie } 5011eb73c1b7SMiao Xie 5012eb73c1b7SMiao Xie static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 5013eb73c1b7SMiao Xie { 5014eb73c1b7SMiao Xie struct btrfs_root *root; 5015eb73c1b7SMiao Xie struct list_head splice; 5016eb73c1b7SMiao Xie 5017eb73c1b7SMiao Xie INIT_LIST_HEAD(&splice); 5018eb73c1b7SMiao Xie 5019eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 5020eb73c1b7SMiao Xie list_splice_init(&fs_info->delalloc_roots, &splice); 5021eb73c1b7SMiao Xie while (!list_empty(&splice)) { 5022eb73c1b7SMiao Xie root = list_first_entry(&splice, struct btrfs_root, 5023eb73c1b7SMiao Xie delalloc_root); 502400246528SJosef Bacik root = btrfs_grab_root(root); 5025eb73c1b7SMiao Xie BUG_ON(!root); 5026eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 5027eb73c1b7SMiao Xie 5028eb73c1b7SMiao Xie btrfs_destroy_delalloc_inodes(root); 502900246528SJosef Bacik btrfs_put_root(root); 5030eb73c1b7SMiao Xie 5031eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 5032eb73c1b7SMiao Xie } 5033eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 5034acce952bSliubo } 5035acce952bSliubo 50362ff7e61eSJeff Mahoney static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 5037acce952bSliubo struct extent_io_tree *dirty_pages, 5038acce952bSliubo int mark) 5039acce952bSliubo { 5040acce952bSliubo int ret; 5041acce952bSliubo struct extent_buffer *eb; 5042acce952bSliubo u64 start = 0; 5043acce952bSliubo u64 end; 5044acce952bSliubo 5045acce952bSliubo while (1) { 5046acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 5047e6138876SJosef Bacik mark, NULL); 5048acce952bSliubo if (ret) 5049acce952bSliubo break; 5050acce952bSliubo 505191166212SDavid Sterba clear_extent_bits(dirty_pages, start, end, mark); 5052acce952bSliubo while (start <= end) { 50530b246afaSJeff Mahoney eb = find_extent_buffer(fs_info, start); 50540b246afaSJeff Mahoney start += fs_info->nodesize; 5055fd8b2b61SJosef Bacik if (!eb) 5056acce952bSliubo continue; 5057fd8b2b61SJosef Bacik wait_on_extent_buffer_writeback(eb); 5058acce952bSliubo 5059fd8b2b61SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 5060fd8b2b61SJosef Bacik &eb->bflags)) 5061fd8b2b61SJosef Bacik clear_extent_buffer_dirty(eb); 5062fd8b2b61SJosef Bacik free_extent_buffer_stale(eb); 5063acce952bSliubo } 5064acce952bSliubo } 5065acce952bSliubo 5066acce952bSliubo return ret; 5067acce952bSliubo } 5068acce952bSliubo 50692ff7e61eSJeff Mahoney static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 5070fe119a6eSNikolay Borisov struct extent_io_tree *unpin) 5071acce952bSliubo { 5072acce952bSliubo u64 start; 5073acce952bSliubo u64 end; 5074acce952bSliubo int ret; 5075acce952bSliubo 5076acce952bSliubo while (1) { 50770e6ec385SFilipe Manana struct extent_state *cached_state = NULL; 50780e6ec385SFilipe Manana 5079fcd5e742SLu Fengqi /* 5080fcd5e742SLu Fengqi * The btrfs_finish_extent_commit() may get the same range as 5081fcd5e742SLu Fengqi * ours between find_first_extent_bit and clear_extent_dirty. 5082fcd5e742SLu Fengqi * Hence, hold the unused_bg_unpin_mutex to avoid double unpin 5083fcd5e742SLu Fengqi * the same extent range. 5084fcd5e742SLu Fengqi */ 5085fcd5e742SLu Fengqi mutex_lock(&fs_info->unused_bg_unpin_mutex); 5086acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 50870e6ec385SFilipe Manana EXTENT_DIRTY, &cached_state); 5088fcd5e742SLu Fengqi if (ret) { 5089fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 5090acce952bSliubo break; 5091fcd5e742SLu Fengqi } 5092acce952bSliubo 50930e6ec385SFilipe Manana clear_extent_dirty(unpin, start, end, &cached_state); 50940e6ec385SFilipe Manana free_extent_state(cached_state); 50952ff7e61eSJeff Mahoney btrfs_error_unpin_extent_range(fs_info, start, end); 5096fcd5e742SLu Fengqi mutex_unlock(&fs_info->unused_bg_unpin_mutex); 5097acce952bSliubo cond_resched(); 5098acce952bSliubo } 5099acce952bSliubo 5100acce952bSliubo return 0; 5101acce952bSliubo } 5102acce952bSliubo 510332da5386SDavid Sterba static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache) 5104c79a1751SLiu Bo { 5105c79a1751SLiu Bo struct inode *inode; 5106c79a1751SLiu Bo 5107c79a1751SLiu Bo inode = cache->io_ctl.inode; 5108c79a1751SLiu Bo if (inode) { 5109c79a1751SLiu Bo invalidate_inode_pages2(inode->i_mapping); 5110c79a1751SLiu Bo BTRFS_I(inode)->generation = 0; 5111c79a1751SLiu Bo cache->io_ctl.inode = NULL; 5112c79a1751SLiu Bo iput(inode); 5113c79a1751SLiu Bo } 5114bbc37d6eSFilipe Manana ASSERT(cache->io_ctl.pages == NULL); 5115c79a1751SLiu Bo btrfs_put_block_group(cache); 5116c79a1751SLiu Bo } 5117c79a1751SLiu Bo 5118c79a1751SLiu Bo void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, 51192ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 5120c79a1751SLiu Bo { 512132da5386SDavid Sterba struct btrfs_block_group *cache; 5122c79a1751SLiu Bo 5123c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 5124c79a1751SLiu Bo while (!list_empty(&cur_trans->dirty_bgs)) { 5125c79a1751SLiu Bo cache = list_first_entry(&cur_trans->dirty_bgs, 512632da5386SDavid Sterba struct btrfs_block_group, 5127c79a1751SLiu Bo dirty_list); 5128c79a1751SLiu Bo 5129c79a1751SLiu Bo if (!list_empty(&cache->io_list)) { 5130c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 5131c79a1751SLiu Bo list_del_init(&cache->io_list); 5132c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 5133c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 5134c79a1751SLiu Bo } 5135c79a1751SLiu Bo 5136c79a1751SLiu Bo list_del_init(&cache->dirty_list); 5137c79a1751SLiu Bo spin_lock(&cache->lock); 5138c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 5139c79a1751SLiu Bo spin_unlock(&cache->lock); 5140c79a1751SLiu Bo 5141c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 5142c79a1751SLiu Bo btrfs_put_block_group(cache); 5143ba2c4d4eSJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 5144c79a1751SLiu Bo spin_lock(&cur_trans->dirty_bgs_lock); 5145c79a1751SLiu Bo } 5146c79a1751SLiu Bo spin_unlock(&cur_trans->dirty_bgs_lock); 5147c79a1751SLiu Bo 514845ae2c18SNikolay Borisov /* 514945ae2c18SNikolay Borisov * Refer to the definition of io_bgs member for details why it's safe 515045ae2c18SNikolay Borisov * to use it without any locking 515145ae2c18SNikolay Borisov */ 5152c79a1751SLiu Bo while (!list_empty(&cur_trans->io_bgs)) { 5153c79a1751SLiu Bo cache = list_first_entry(&cur_trans->io_bgs, 515432da5386SDavid Sterba struct btrfs_block_group, 5155c79a1751SLiu Bo io_list); 5156c79a1751SLiu Bo 5157c79a1751SLiu Bo list_del_init(&cache->io_list); 5158c79a1751SLiu Bo spin_lock(&cache->lock); 5159c79a1751SLiu Bo cache->disk_cache_state = BTRFS_DC_ERROR; 5160c79a1751SLiu Bo spin_unlock(&cache->lock); 5161c79a1751SLiu Bo btrfs_cleanup_bg_io(cache); 5162c79a1751SLiu Bo } 5163c79a1751SLiu Bo } 5164c79a1751SLiu Bo 516549b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 51662ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info) 516749b25e05SJeff Mahoney { 5168bbbf7243SNikolay Borisov struct btrfs_device *dev, *tmp; 5169bbbf7243SNikolay Borisov 51702ff7e61eSJeff Mahoney btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 5171c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->dirty_bgs)); 5172c79a1751SLiu Bo ASSERT(list_empty(&cur_trans->io_bgs)); 5173c79a1751SLiu Bo 5174bbbf7243SNikolay Borisov list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list, 5175bbbf7243SNikolay Borisov post_commit_list) { 5176bbbf7243SNikolay Borisov list_del_init(&dev->post_commit_list); 5177bbbf7243SNikolay Borisov } 5178bbbf7243SNikolay Borisov 51792ff7e61eSJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, fs_info); 518049b25e05SJeff Mahoney 51814a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 51820b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 518349b25e05SJeff Mahoney 51844a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 51850b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 518649b25e05SJeff Mahoney 5187ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 518867cde344SMiao Xie 51892ff7e61eSJeff Mahoney btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, 519049b25e05SJeff Mahoney EXTENT_DIRTY); 5191fe119a6eSNikolay Borisov btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents); 519249b25e05SJeff Mahoney 5193d3575156SNaohiro Aota btrfs_free_redirty_list(cur_trans); 5194d3575156SNaohiro Aota 51954a9d8bdeSMiao Xie cur_trans->state =TRANS_STATE_COMPLETED; 51964a9d8bdeSMiao Xie wake_up(&cur_trans->commit_wait); 519749b25e05SJeff Mahoney } 519849b25e05SJeff Mahoney 51992ff7e61eSJeff Mahoney static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) 5200acce952bSliubo { 5201acce952bSliubo struct btrfs_transaction *t; 5202acce952bSliubo 52030b246afaSJeff Mahoney mutex_lock(&fs_info->transaction_kthread_mutex); 5204acce952bSliubo 52050b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 52060b246afaSJeff Mahoney while (!list_empty(&fs_info->trans_list)) { 52070b246afaSJeff Mahoney t = list_first_entry(&fs_info->trans_list, 5208724e2315SJosef Bacik struct btrfs_transaction, list); 5209724e2315SJosef Bacik if (t->state >= TRANS_STATE_COMMIT_START) { 52109b64f57dSElena Reshetova refcount_inc(&t->use_count); 52110b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 52122ff7e61eSJeff Mahoney btrfs_wait_for_commit(fs_info, t->transid); 5213724e2315SJosef Bacik btrfs_put_transaction(t); 52140b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5215724e2315SJosef Bacik continue; 5216724e2315SJosef Bacik } 52170b246afaSJeff Mahoney if (t == fs_info->running_transaction) { 5218724e2315SJosef Bacik t->state = TRANS_STATE_COMMIT_DOING; 52190b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 5220724e2315SJosef Bacik /* 5221724e2315SJosef Bacik * We wait for 0 num_writers since we don't hold a trans 5222724e2315SJosef Bacik * handle open currently for this transaction. 5223724e2315SJosef Bacik */ 5224724e2315SJosef Bacik wait_event(t->writer_wait, 5225724e2315SJosef Bacik atomic_read(&t->num_writers) == 0); 5226724e2315SJosef Bacik } else { 52270b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 5228724e2315SJosef Bacik } 52292ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(t, fs_info); 5230724e2315SJosef Bacik 52310b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 52320b246afaSJeff Mahoney if (t == fs_info->running_transaction) 52330b246afaSJeff Mahoney fs_info->running_transaction = NULL; 5234724e2315SJosef Bacik list_del_init(&t->list); 52350b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 5236a4abeea4SJosef Bacik 5237724e2315SJosef Bacik btrfs_put_transaction(t); 52382e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 52390b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5240724e2315SJosef Bacik } 52410b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 52420b246afaSJeff Mahoney btrfs_destroy_all_ordered_extents(fs_info); 5243ccdf9b30SJeff Mahoney btrfs_destroy_delayed_inodes(fs_info); 5244ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 52450b246afaSJeff Mahoney btrfs_destroy_all_delalloc_inodes(fs_info); 5246ef67963dSJosef Bacik btrfs_drop_all_logs(fs_info); 52470b246afaSJeff Mahoney mutex_unlock(&fs_info->transaction_kthread_mutex); 5248acce952bSliubo 5249acce952bSliubo return 0; 5250acce952bSliubo } 5251ec7d6dfdSNikolay Borisov 5252453e4873SNikolay Borisov int btrfs_init_root_free_objectid(struct btrfs_root *root) 5253ec7d6dfdSNikolay Borisov { 5254ec7d6dfdSNikolay Borisov struct btrfs_path *path; 5255ec7d6dfdSNikolay Borisov int ret; 5256ec7d6dfdSNikolay Borisov struct extent_buffer *l; 5257ec7d6dfdSNikolay Borisov struct btrfs_key search_key; 5258ec7d6dfdSNikolay Borisov struct btrfs_key found_key; 5259ec7d6dfdSNikolay Borisov int slot; 5260ec7d6dfdSNikolay Borisov 5261ec7d6dfdSNikolay Borisov path = btrfs_alloc_path(); 5262ec7d6dfdSNikolay Borisov if (!path) 5263ec7d6dfdSNikolay Borisov return -ENOMEM; 5264ec7d6dfdSNikolay Borisov 5265ec7d6dfdSNikolay Borisov search_key.objectid = BTRFS_LAST_FREE_OBJECTID; 5266ec7d6dfdSNikolay Borisov search_key.type = -1; 5267ec7d6dfdSNikolay Borisov search_key.offset = (u64)-1; 5268ec7d6dfdSNikolay Borisov ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); 5269ec7d6dfdSNikolay Borisov if (ret < 0) 5270ec7d6dfdSNikolay Borisov goto error; 5271ec7d6dfdSNikolay Borisov BUG_ON(ret == 0); /* Corruption */ 5272ec7d6dfdSNikolay Borisov if (path->slots[0] > 0) { 5273ec7d6dfdSNikolay Borisov slot = path->slots[0] - 1; 5274ec7d6dfdSNikolay Borisov l = path->nodes[0]; 5275ec7d6dfdSNikolay Borisov btrfs_item_key_to_cpu(l, &found_key, slot); 527623125104SNikolay Borisov root->free_objectid = max_t(u64, found_key.objectid + 1, 527723125104SNikolay Borisov BTRFS_FIRST_FREE_OBJECTID); 5278ec7d6dfdSNikolay Borisov } else { 527923125104SNikolay Borisov root->free_objectid = BTRFS_FIRST_FREE_OBJECTID; 5280ec7d6dfdSNikolay Borisov } 5281ec7d6dfdSNikolay Borisov ret = 0; 5282ec7d6dfdSNikolay Borisov error: 5283ec7d6dfdSNikolay Borisov btrfs_free_path(path); 5284ec7d6dfdSNikolay Borisov return ret; 5285ec7d6dfdSNikolay Borisov } 5286ec7d6dfdSNikolay Borisov 5287543068a2SNikolay Borisov int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid) 5288ec7d6dfdSNikolay Borisov { 5289ec7d6dfdSNikolay Borisov int ret; 5290ec7d6dfdSNikolay Borisov mutex_lock(&root->objectid_mutex); 5291ec7d6dfdSNikolay Borisov 52926b8fad57SNikolay Borisov if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) { 5293ec7d6dfdSNikolay Borisov btrfs_warn(root->fs_info, 5294ec7d6dfdSNikolay Borisov "the objectid of root %llu reaches its highest value", 5295ec7d6dfdSNikolay Borisov root->root_key.objectid); 5296ec7d6dfdSNikolay Borisov ret = -ENOSPC; 5297ec7d6dfdSNikolay Borisov goto out; 5298ec7d6dfdSNikolay Borisov } 5299ec7d6dfdSNikolay Borisov 530023125104SNikolay Borisov *objectid = root->free_objectid++; 5301ec7d6dfdSNikolay Borisov ret = 0; 5302ec7d6dfdSNikolay Borisov out: 5303ec7d6dfdSNikolay Borisov mutex_unlock(&root->objectid_mutex); 5304ec7d6dfdSNikolay Borisov return ret; 5305ec7d6dfdSNikolay Borisov } 5306