16cbd5570SChris Mason /* 26cbd5570SChris Mason * Copyright (C) 2007 Oracle. All rights reserved. 36cbd5570SChris Mason * 46cbd5570SChris Mason * This program is free software; you can redistribute it and/or 56cbd5570SChris Mason * modify it under the terms of the GNU General Public 66cbd5570SChris Mason * License v2 as published by the Free Software Foundation. 76cbd5570SChris Mason * 86cbd5570SChris Mason * This program is distributed in the hope that it will be useful, 96cbd5570SChris Mason * but WITHOUT ANY WARRANTY; without even the implied warranty of 106cbd5570SChris Mason * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 116cbd5570SChris Mason * General Public License for more details. 126cbd5570SChris Mason * 136cbd5570SChris Mason * You should have received a copy of the GNU General Public 146cbd5570SChris Mason * License along with this program; if not, write to the 156cbd5570SChris Mason * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 166cbd5570SChris Mason * Boston, MA 021110-1307, USA. 176cbd5570SChris Mason */ 186cbd5570SChris Mason 19e20d96d6SChris Mason #include <linux/fs.h> 20d98237b3SChris Mason #include <linux/blkdev.h> 2187cbda5cSChris Mason #include <linux/scatterlist.h> 2222b0ebdaSChris Mason #include <linux/swap.h> 230f7d52f4SChris Mason #include <linux/radix-tree.h> 2435b7e476SChris Mason #include <linux/writeback.h> 25d397712bSChris Mason #include <linux/buffer_head.h> 26ce9adaa5SChris Mason #include <linux/workqueue.h> 27a74a4b97SChris Mason #include <linux/kthread.h> 28a74a4b97SChris Mason #include <linux/freezer.h> 29163e783eSDavid Woodhouse #include <linux/crc32c.h> 305a0e3ad6STejun Heo #include <linux/slab.h> 31784b4e29SChris Mason #include <linux/migrate.h> 327a36ddecSDavid Sterba #include <linux/ratelimit.h> 337e75bf3fSDavid Sterba #include <asm/unaligned.h> 344b4e25f2SChris Mason #include "compat.h" 35eb60ceacSChris Mason #include "ctree.h" 36eb60ceacSChris Mason #include "disk-io.h" 37e089f05cSChris Mason #include "transaction.h" 380f7d52f4SChris Mason #include "btrfs_inode.h" 390b86a832SChris Mason #include "volumes.h" 40db94535dSChris Mason #include "print-tree.h" 418b712842SChris Mason #include "async-thread.h" 42925baeddSChris Mason #include "locking.h" 43e02119d5SChris Mason #include "tree-log.h" 44fa9c0d79SChris Mason #include "free-space-cache.h" 45581bb050SLi Zefan #include "inode-map.h" 4621adbd5cSStefan Behrens #include "check-integrity.h" 47606686eeSJosef Bacik #include "rcu-string.h" 48eb60ceacSChris Mason 49d1310b2eSChris Mason static struct extent_io_ops btree_extent_io_ops; 508b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work); 514df27c4dSYan, Zheng static void free_fs_root(struct btrfs_root *root); 52fcd1f065SDavid Sterba static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info, 53acce952bSliubo int read_only); 54143bede5SJeff Mahoney static void btrfs_destroy_ordered_operations(struct btrfs_root *root); 55143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root); 56acce952bSliubo static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 57acce952bSliubo struct btrfs_root *root); 58143bede5SJeff Mahoney static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t); 59143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root); 60acce952bSliubo static int btrfs_destroy_marked_extents(struct btrfs_root *root, 61acce952bSliubo struct extent_io_tree *dirty_pages, 62acce952bSliubo int mark); 63acce952bSliubo static int btrfs_destroy_pinned_extent(struct btrfs_root *root, 64acce952bSliubo struct extent_io_tree *pinned_extents); 65ce9adaa5SChris Mason 66d352ac68SChris Mason /* 67d352ac68SChris Mason * end_io_wq structs are used to do processing in task context when an IO is 68d352ac68SChris Mason * complete. This is used during reads to verify checksums, and it is used 69d352ac68SChris Mason * by writes to insert metadata for new file extents after IO is complete. 70d352ac68SChris Mason */ 71ce9adaa5SChris Mason struct end_io_wq { 72ce9adaa5SChris Mason struct bio *bio; 73ce9adaa5SChris Mason bio_end_io_t *end_io; 74ce9adaa5SChris Mason void *private; 75ce9adaa5SChris Mason struct btrfs_fs_info *info; 76ce9adaa5SChris Mason int error; 7722c59948SChris Mason int metadata; 78ce9adaa5SChris Mason struct list_head list; 798b712842SChris Mason struct btrfs_work work; 80ce9adaa5SChris Mason }; 810da5468fSChris Mason 82d352ac68SChris Mason /* 83d352ac68SChris Mason * async submit bios are used to offload expensive checksumming 84d352ac68SChris Mason * onto the worker threads. They checksum file and metadata bios 85d352ac68SChris Mason * just before they are sent down the IO stack. 86d352ac68SChris Mason */ 8744b8bd7eSChris Mason struct async_submit_bio { 8844b8bd7eSChris Mason struct inode *inode; 8944b8bd7eSChris Mason struct bio *bio; 9044b8bd7eSChris Mason struct list_head list; 914a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_start; 924a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_done; 9344b8bd7eSChris Mason int rw; 9444b8bd7eSChris Mason int mirror_num; 95c8b97818SChris Mason unsigned long bio_flags; 96eaf25d93SChris Mason /* 97eaf25d93SChris Mason * bio_offset is optional, can be used if the pages in the bio 98eaf25d93SChris Mason * can't tell us where in the file the bio should go 99eaf25d93SChris Mason */ 100eaf25d93SChris Mason u64 bio_offset; 1018b712842SChris Mason struct btrfs_work work; 10279787eaaSJeff Mahoney int error; 10344b8bd7eSChris Mason }; 10444b8bd7eSChris Mason 10585d4e461SChris Mason /* 10685d4e461SChris Mason * Lockdep class keys for extent_buffer->lock's in this root. For a given 10785d4e461SChris Mason * eb, the lockdep key is determined by the btrfs_root it belongs to and 10885d4e461SChris Mason * the level the eb occupies in the tree. 1094008c04aSChris Mason * 11085d4e461SChris Mason * Different roots are used for different purposes and may nest inside each 11185d4e461SChris Mason * other and they require separate keysets. As lockdep keys should be 11285d4e461SChris Mason * static, assign keysets according to the purpose of the root as indicated 11385d4e461SChris Mason * by btrfs_root->objectid. This ensures that all special purpose roots 11485d4e461SChris Mason * have separate keysets. 1154008c04aSChris Mason * 11685d4e461SChris Mason * Lock-nesting across peer nodes is always done with the immediate parent 11785d4e461SChris Mason * node locked thus preventing deadlock. As lockdep doesn't know this, use 11885d4e461SChris Mason * subclass to avoid triggering lockdep warning in such cases. 1194008c04aSChris Mason * 12085d4e461SChris Mason * The key is set by the readpage_end_io_hook after the buffer has passed 12185d4e461SChris Mason * csum validation but before the pages are unlocked. It is also set by 12285d4e461SChris Mason * btrfs_init_new_buffer on freshly allocated blocks. 12385d4e461SChris Mason * 12485d4e461SChris Mason * We also add a check to make sure the highest level of the tree is the 12585d4e461SChris Mason * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code 12685d4e461SChris Mason * needs update as well. 1274008c04aSChris Mason */ 1284008c04aSChris Mason #ifdef CONFIG_DEBUG_LOCK_ALLOC 1294008c04aSChris Mason # if BTRFS_MAX_LEVEL != 8 1304008c04aSChris Mason # error 1314008c04aSChris Mason # endif 13285d4e461SChris Mason 13385d4e461SChris Mason static struct btrfs_lockdep_keyset { 13485d4e461SChris Mason u64 id; /* root objectid */ 13585d4e461SChris Mason const char *name_stem; /* lock name stem */ 13685d4e461SChris Mason char names[BTRFS_MAX_LEVEL + 1][20]; 13785d4e461SChris Mason struct lock_class_key keys[BTRFS_MAX_LEVEL + 1]; 13885d4e461SChris Mason } btrfs_lockdep_keysets[] = { 13985d4e461SChris Mason { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" }, 14085d4e461SChris Mason { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" }, 14185d4e461SChris Mason { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" }, 14285d4e461SChris Mason { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" }, 14385d4e461SChris Mason { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" }, 14485d4e461SChris Mason { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" }, 14585d4e461SChris Mason { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" }, 14685d4e461SChris Mason { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" }, 14785d4e461SChris Mason { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" }, 14885d4e461SChris Mason { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" }, 14985d4e461SChris Mason { .id = 0, .name_stem = "tree" }, 1504008c04aSChris Mason }; 15185d4e461SChris Mason 15285d4e461SChris Mason void __init btrfs_init_lockdep(void) 15385d4e461SChris Mason { 15485d4e461SChris Mason int i, j; 15585d4e461SChris Mason 15685d4e461SChris Mason /* initialize lockdep class names */ 15785d4e461SChris Mason for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) { 15885d4e461SChris Mason struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i]; 15985d4e461SChris Mason 16085d4e461SChris Mason for (j = 0; j < ARRAY_SIZE(ks->names); j++) 16185d4e461SChris Mason snprintf(ks->names[j], sizeof(ks->names[j]), 16285d4e461SChris Mason "btrfs-%s-%02d", ks->name_stem, j); 16385d4e461SChris Mason } 16485d4e461SChris Mason } 16585d4e461SChris Mason 16685d4e461SChris Mason void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb, 16785d4e461SChris Mason int level) 16885d4e461SChris Mason { 16985d4e461SChris Mason struct btrfs_lockdep_keyset *ks; 17085d4e461SChris Mason 17185d4e461SChris Mason BUG_ON(level >= ARRAY_SIZE(ks->keys)); 17285d4e461SChris Mason 17385d4e461SChris Mason /* find the matching keyset, id 0 is the default entry */ 17485d4e461SChris Mason for (ks = btrfs_lockdep_keysets; ks->id; ks++) 17585d4e461SChris Mason if (ks->id == objectid) 17685d4e461SChris Mason break; 17785d4e461SChris Mason 17885d4e461SChris Mason lockdep_set_class_and_name(&eb->lock, 17985d4e461SChris Mason &ks->keys[level], ks->names[level]); 18085d4e461SChris Mason } 18185d4e461SChris Mason 1824008c04aSChris Mason #endif 1834008c04aSChris Mason 184d352ac68SChris Mason /* 185d352ac68SChris Mason * extents on the btree inode are pretty simple, there's one extent 186d352ac68SChris Mason * that covers the entire device 187d352ac68SChris Mason */ 188b2950863SChristoph Hellwig static struct extent_map *btree_get_extent(struct inode *inode, 189306e16ceSDavid Sterba struct page *page, size_t pg_offset, u64 start, u64 len, 1905f39d397SChris Mason int create) 1915f39d397SChris Mason { 1925f39d397SChris Mason struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 1935f39d397SChris Mason struct extent_map *em; 1945f39d397SChris Mason int ret; 1955f39d397SChris Mason 196890871beSChris Mason read_lock(&em_tree->lock); 197d1310b2eSChris Mason em = lookup_extent_mapping(em_tree, start, len); 198a061fc8dSChris Mason if (em) { 199a061fc8dSChris Mason em->bdev = 200a061fc8dSChris Mason BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 201890871beSChris Mason read_unlock(&em_tree->lock); 2025f39d397SChris Mason goto out; 203a061fc8dSChris Mason } 204890871beSChris Mason read_unlock(&em_tree->lock); 2057b13b7b1SChris Mason 206172ddd60SDavid Sterba em = alloc_extent_map(); 2075f39d397SChris Mason if (!em) { 2085f39d397SChris Mason em = ERR_PTR(-ENOMEM); 2095f39d397SChris Mason goto out; 2105f39d397SChris Mason } 2115f39d397SChris Mason em->start = 0; 2120afbaf8cSChris Mason em->len = (u64)-1; 213c8b97818SChris Mason em->block_len = (u64)-1; 2145f39d397SChris Mason em->block_start = 0; 215a061fc8dSChris Mason em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 216d1310b2eSChris Mason 217890871beSChris Mason write_lock(&em_tree->lock); 2185f39d397SChris Mason ret = add_extent_mapping(em_tree, em); 2195f39d397SChris Mason if (ret == -EEXIST) { 2200afbaf8cSChris Mason u64 failed_start = em->start; 2210afbaf8cSChris Mason u64 failed_len = em->len; 2220afbaf8cSChris Mason 2235f39d397SChris Mason free_extent_map(em); 2247b13b7b1SChris Mason em = lookup_extent_mapping(em_tree, start, len); 2250afbaf8cSChris Mason if (em) { 2267b13b7b1SChris Mason ret = 0; 2270afbaf8cSChris Mason } else { 2280afbaf8cSChris Mason em = lookup_extent_mapping(em_tree, failed_start, 2290afbaf8cSChris Mason failed_len); 2307b13b7b1SChris Mason ret = -EIO; 2310afbaf8cSChris Mason } 2325f39d397SChris Mason } else if (ret) { 2337b13b7b1SChris Mason free_extent_map(em); 2347b13b7b1SChris Mason em = NULL; 2355f39d397SChris Mason } 236890871beSChris Mason write_unlock(&em_tree->lock); 2377b13b7b1SChris Mason 2387b13b7b1SChris Mason if (ret) 2397b13b7b1SChris Mason em = ERR_PTR(ret); 2405f39d397SChris Mason out: 2415f39d397SChris Mason return em; 2425f39d397SChris Mason } 2435f39d397SChris Mason 24419c00ddcSChris Mason u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len) 24519c00ddcSChris Mason { 246163e783eSDavid Woodhouse return crc32c(seed, data, len); 24719c00ddcSChris Mason } 24819c00ddcSChris Mason 24919c00ddcSChris Mason void btrfs_csum_final(u32 crc, char *result) 25019c00ddcSChris Mason { 2517e75bf3fSDavid Sterba put_unaligned_le32(~crc, result); 25219c00ddcSChris Mason } 25319c00ddcSChris Mason 254d352ac68SChris Mason /* 255d352ac68SChris Mason * compute the csum for a btree block, and either verify it or write it 256d352ac68SChris Mason * into the csum field of the block. 257d352ac68SChris Mason */ 25819c00ddcSChris Mason static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf, 25919c00ddcSChris Mason int verify) 26019c00ddcSChris Mason { 2616c41761fSDavid Sterba u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy); 262607d432dSJosef Bacik char *result = NULL; 26319c00ddcSChris Mason unsigned long len; 26419c00ddcSChris Mason unsigned long cur_len; 26519c00ddcSChris Mason unsigned long offset = BTRFS_CSUM_SIZE; 26619c00ddcSChris Mason char *kaddr; 26719c00ddcSChris Mason unsigned long map_start; 26819c00ddcSChris Mason unsigned long map_len; 26919c00ddcSChris Mason int err; 27019c00ddcSChris Mason u32 crc = ~(u32)0; 271607d432dSJosef Bacik unsigned long inline_result; 27219c00ddcSChris Mason 27319c00ddcSChris Mason len = buf->len - offset; 27419c00ddcSChris Mason while (len > 0) { 27519c00ddcSChris Mason err = map_private_extent_buffer(buf, offset, 32, 276a6591715SChris Mason &kaddr, &map_start, &map_len); 277d397712bSChris Mason if (err) 27819c00ddcSChris Mason return 1; 27919c00ddcSChris Mason cur_len = min(len, map_len - (offset - map_start)); 28019c00ddcSChris Mason crc = btrfs_csum_data(root, kaddr + offset - map_start, 28119c00ddcSChris Mason crc, cur_len); 28219c00ddcSChris Mason len -= cur_len; 28319c00ddcSChris Mason offset += cur_len; 28419c00ddcSChris Mason } 285607d432dSJosef Bacik if (csum_size > sizeof(inline_result)) { 286607d432dSJosef Bacik result = kzalloc(csum_size * sizeof(char), GFP_NOFS); 287607d432dSJosef Bacik if (!result) 288607d432dSJosef Bacik return 1; 289607d432dSJosef Bacik } else { 290607d432dSJosef Bacik result = (char *)&inline_result; 291607d432dSJosef Bacik } 292607d432dSJosef Bacik 29319c00ddcSChris Mason btrfs_csum_final(crc, result); 29419c00ddcSChris Mason 29519c00ddcSChris Mason if (verify) { 296607d432dSJosef Bacik if (memcmp_extent_buffer(buf, result, 0, csum_size)) { 297e4204dedSChris Mason u32 val; 298e4204dedSChris Mason u32 found = 0; 299607d432dSJosef Bacik memcpy(&found, result, csum_size); 300e4204dedSChris Mason 301607d432dSJosef Bacik read_extent_buffer(buf, &val, 0, csum_size); 3027a36ddecSDavid Sterba printk_ratelimited(KERN_INFO "btrfs: %s checksum verify " 303193f284dSChris Mason "failed on %llu wanted %X found %X " 304193f284dSChris Mason "level %d\n", 30519c00ddcSChris Mason root->fs_info->sb->s_id, 306193f284dSChris Mason (unsigned long long)buf->start, val, found, 307193f284dSChris Mason btrfs_header_level(buf)); 308607d432dSJosef Bacik if (result != (char *)&inline_result) 309607d432dSJosef Bacik kfree(result); 31019c00ddcSChris Mason return 1; 31119c00ddcSChris Mason } 31219c00ddcSChris Mason } else { 313607d432dSJosef Bacik write_extent_buffer(buf, result, 0, csum_size); 31419c00ddcSChris Mason } 315607d432dSJosef Bacik if (result != (char *)&inline_result) 316607d432dSJosef Bacik kfree(result); 31719c00ddcSChris Mason return 0; 31819c00ddcSChris Mason } 31919c00ddcSChris Mason 320d352ac68SChris Mason /* 321d352ac68SChris Mason * we can't consider a given block up to date unless the transid of the 322d352ac68SChris Mason * block matches the transid in the parent node's pointer. This is how we 323d352ac68SChris Mason * detect blocks that either didn't get written at all or got written 324d352ac68SChris Mason * in the wrong place. 325d352ac68SChris Mason */ 3261259ab75SChris Mason static int verify_parent_transid(struct extent_io_tree *io_tree, 327b9fab919SChris Mason struct extent_buffer *eb, u64 parent_transid, 328b9fab919SChris Mason int atomic) 3291259ab75SChris Mason { 3302ac55d41SJosef Bacik struct extent_state *cached_state = NULL; 3311259ab75SChris Mason int ret; 3321259ab75SChris Mason 3331259ab75SChris Mason if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 3341259ab75SChris Mason return 0; 3351259ab75SChris Mason 336b9fab919SChris Mason if (atomic) 337b9fab919SChris Mason return -EAGAIN; 338b9fab919SChris Mason 3392ac55d41SJosef Bacik lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 340d0082371SJeff Mahoney 0, &cached_state); 3410b32f4bbSJosef Bacik if (extent_buffer_uptodate(eb) && 3421259ab75SChris Mason btrfs_header_generation(eb) == parent_transid) { 3431259ab75SChris Mason ret = 0; 3441259ab75SChris Mason goto out; 3451259ab75SChris Mason } 3467a36ddecSDavid Sterba printk_ratelimited("parent transid verify failed on %llu wanted %llu " 347193f284dSChris Mason "found %llu\n", 3481259ab75SChris Mason (unsigned long long)eb->start, 3491259ab75SChris Mason (unsigned long long)parent_transid, 3501259ab75SChris Mason (unsigned long long)btrfs_header_generation(eb)); 3511259ab75SChris Mason ret = 1; 3520b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 35333958dc6SChris Mason out: 3542ac55d41SJosef Bacik unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 3552ac55d41SJosef Bacik &cached_state, GFP_NOFS); 3561259ab75SChris Mason return ret; 3571259ab75SChris Mason } 3581259ab75SChris Mason 359d352ac68SChris Mason /* 360d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 361d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 362d352ac68SChris Mason */ 363f188591eSChris Mason static int btree_read_extent_buffer_pages(struct btrfs_root *root, 364f188591eSChris Mason struct extent_buffer *eb, 365ca7a79adSChris Mason u64 start, u64 parent_transid) 366f188591eSChris Mason { 367f188591eSChris Mason struct extent_io_tree *io_tree; 368ea466794SJosef Bacik int failed = 0; 369f188591eSChris Mason int ret; 370f188591eSChris Mason int num_copies = 0; 371f188591eSChris Mason int mirror_num = 0; 372ea466794SJosef Bacik int failed_mirror = 0; 373f188591eSChris Mason 374a826d6dcSJosef Bacik clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 375f188591eSChris Mason io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree; 376f188591eSChris Mason while (1) { 377bb82ab88SArne Jansen ret = read_extent_buffer_pages(io_tree, eb, start, 378bb82ab88SArne Jansen WAIT_COMPLETE, 379f188591eSChris Mason btree_get_extent, mirror_num); 380b9fab919SChris Mason if (!ret && !verify_parent_transid(io_tree, eb, 381b9fab919SChris Mason parent_transid, 0)) 382ea466794SJosef Bacik break; 383d397712bSChris Mason 384a826d6dcSJosef Bacik /* 385a826d6dcSJosef Bacik * This buffer's crc is fine, but its contents are corrupted, so 386a826d6dcSJosef Bacik * there is no reason to read the other copies, they won't be 387a826d6dcSJosef Bacik * any less wrong. 388a826d6dcSJosef Bacik */ 389a826d6dcSJosef Bacik if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags)) 390ea466794SJosef Bacik break; 391ea466794SJosef Bacik 392f188591eSChris Mason num_copies = btrfs_num_copies(&root->fs_info->mapping_tree, 393f188591eSChris Mason eb->start, eb->len); 3944235298eSChris Mason if (num_copies == 1) 395ea466794SJosef Bacik break; 3964235298eSChris Mason 3975cf1ab56SJosef Bacik if (!failed_mirror) { 3985cf1ab56SJosef Bacik failed = 1; 3995cf1ab56SJosef Bacik failed_mirror = eb->read_mirror; 4005cf1ab56SJosef Bacik } 4015cf1ab56SJosef Bacik 402f188591eSChris Mason mirror_num++; 403ea466794SJosef Bacik if (mirror_num == failed_mirror) 404ea466794SJosef Bacik mirror_num++; 405ea466794SJosef Bacik 4064235298eSChris Mason if (mirror_num > num_copies) 407ea466794SJosef Bacik break; 408f188591eSChris Mason } 409ea466794SJosef Bacik 410ea466794SJosef Bacik if (failed && !ret) 411ea466794SJosef Bacik repair_eb_io_failure(root, eb, failed_mirror); 412ea466794SJosef Bacik 413ea466794SJosef Bacik return ret; 414f188591eSChris Mason } 41519c00ddcSChris Mason 416d352ac68SChris Mason /* 417d397712bSChris Mason * checksum a dirty tree block before IO. This has extra checks to make sure 418d397712bSChris Mason * we only fill in the checksum field in the first page of a multi-page block 419d352ac68SChris Mason */ 420d397712bSChris Mason 421b2950863SChristoph Hellwig static int csum_dirty_buffer(struct btrfs_root *root, struct page *page) 42219c00ddcSChris Mason { 423d1310b2eSChris Mason struct extent_io_tree *tree; 42435ebb934SChris Mason u64 start = (u64)page->index << PAGE_CACHE_SHIFT; 42519c00ddcSChris Mason u64 found_start; 42619c00ddcSChris Mason struct extent_buffer *eb; 427f188591eSChris Mason 428d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 42919c00ddcSChris Mason 4304f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 4314f2de97aSJosef Bacik if (page != eb->pages[0]) 4324f2de97aSJosef Bacik return 0; 43319c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 43419c00ddcSChris Mason if (found_start != start) { 43555c69072SChris Mason WARN_ON(1); 4364f2de97aSJosef Bacik return 0; 43755c69072SChris Mason } 438727011e0SChris Mason if (eb->pages[0] != page) { 43955c69072SChris Mason WARN_ON(1); 4404f2de97aSJosef Bacik return 0; 44155c69072SChris Mason } 44255c69072SChris Mason if (!PageUptodate(page)) { 44355c69072SChris Mason WARN_ON(1); 4444f2de97aSJosef Bacik return 0; 44519c00ddcSChris Mason } 44619c00ddcSChris Mason csum_tree_block(root, eb, 0); 44719c00ddcSChris Mason return 0; 44819c00ddcSChris Mason } 44919c00ddcSChris Mason 4502b82032cSYan Zheng static int check_tree_block_fsid(struct btrfs_root *root, 4512b82032cSYan Zheng struct extent_buffer *eb) 4522b82032cSYan Zheng { 4532b82032cSYan Zheng struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; 4542b82032cSYan Zheng u8 fsid[BTRFS_UUID_SIZE]; 4552b82032cSYan Zheng int ret = 1; 4562b82032cSYan Zheng 4572b82032cSYan Zheng read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb), 4582b82032cSYan Zheng BTRFS_FSID_SIZE); 4592b82032cSYan Zheng while (fs_devices) { 4602b82032cSYan Zheng if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) { 4612b82032cSYan Zheng ret = 0; 4622b82032cSYan Zheng break; 4632b82032cSYan Zheng } 4642b82032cSYan Zheng fs_devices = fs_devices->seed; 4652b82032cSYan Zheng } 4662b82032cSYan Zheng return ret; 4672b82032cSYan Zheng } 4682b82032cSYan Zheng 469a826d6dcSJosef Bacik #define CORRUPT(reason, eb, root, slot) \ 470a826d6dcSJosef Bacik printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \ 471a826d6dcSJosef Bacik "root=%llu, slot=%d\n", reason, \ 472a826d6dcSJosef Bacik (unsigned long long)btrfs_header_bytenr(eb), \ 473a826d6dcSJosef Bacik (unsigned long long)root->objectid, slot) 474a826d6dcSJosef Bacik 475a826d6dcSJosef Bacik static noinline int check_leaf(struct btrfs_root *root, 476a826d6dcSJosef Bacik struct extent_buffer *leaf) 477a826d6dcSJosef Bacik { 478a826d6dcSJosef Bacik struct btrfs_key key; 479a826d6dcSJosef Bacik struct btrfs_key leaf_key; 480a826d6dcSJosef Bacik u32 nritems = btrfs_header_nritems(leaf); 481a826d6dcSJosef Bacik int slot; 482a826d6dcSJosef Bacik 483a826d6dcSJosef Bacik if (nritems == 0) 484a826d6dcSJosef Bacik return 0; 485a826d6dcSJosef Bacik 486a826d6dcSJosef Bacik /* Check the 0 item */ 487a826d6dcSJosef Bacik if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) != 488a826d6dcSJosef Bacik BTRFS_LEAF_DATA_SIZE(root)) { 489a826d6dcSJosef Bacik CORRUPT("invalid item offset size pair", leaf, root, 0); 490a826d6dcSJosef Bacik return -EIO; 491a826d6dcSJosef Bacik } 492a826d6dcSJosef Bacik 493a826d6dcSJosef Bacik /* 494a826d6dcSJosef Bacik * Check to make sure each items keys are in the correct order and their 495a826d6dcSJosef Bacik * offsets make sense. We only have to loop through nritems-1 because 496a826d6dcSJosef Bacik * we check the current slot against the next slot, which verifies the 497a826d6dcSJosef Bacik * next slot's offset+size makes sense and that the current's slot 498a826d6dcSJosef Bacik * offset is correct. 499a826d6dcSJosef Bacik */ 500a826d6dcSJosef Bacik for (slot = 0; slot < nritems - 1; slot++) { 501a826d6dcSJosef Bacik btrfs_item_key_to_cpu(leaf, &leaf_key, slot); 502a826d6dcSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, slot + 1); 503a826d6dcSJosef Bacik 504a826d6dcSJosef Bacik /* Make sure the keys are in the right order */ 505a826d6dcSJosef Bacik if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) { 506a826d6dcSJosef Bacik CORRUPT("bad key order", leaf, root, slot); 507a826d6dcSJosef Bacik return -EIO; 508a826d6dcSJosef Bacik } 509a826d6dcSJosef Bacik 510a826d6dcSJosef Bacik /* 511a826d6dcSJosef Bacik * Make sure the offset and ends are right, remember that the 512a826d6dcSJosef Bacik * item data starts at the end of the leaf and grows towards the 513a826d6dcSJosef Bacik * front. 514a826d6dcSJosef Bacik */ 515a826d6dcSJosef Bacik if (btrfs_item_offset_nr(leaf, slot) != 516a826d6dcSJosef Bacik btrfs_item_end_nr(leaf, slot + 1)) { 517a826d6dcSJosef Bacik CORRUPT("slot offset bad", leaf, root, slot); 518a826d6dcSJosef Bacik return -EIO; 519a826d6dcSJosef Bacik } 520a826d6dcSJosef Bacik 521a826d6dcSJosef Bacik /* 522a826d6dcSJosef Bacik * Check to make sure that we don't point outside of the leaf, 523a826d6dcSJosef Bacik * just incase all the items are consistent to eachother, but 524a826d6dcSJosef Bacik * all point outside of the leaf. 525a826d6dcSJosef Bacik */ 526a826d6dcSJosef Bacik if (btrfs_item_end_nr(leaf, slot) > 527a826d6dcSJosef Bacik BTRFS_LEAF_DATA_SIZE(root)) { 528a826d6dcSJosef Bacik CORRUPT("slot end outside of leaf", leaf, root, slot); 529a826d6dcSJosef Bacik return -EIO; 530a826d6dcSJosef Bacik } 531a826d6dcSJosef Bacik } 532a826d6dcSJosef Bacik 533a826d6dcSJosef Bacik return 0; 534a826d6dcSJosef Bacik } 535a826d6dcSJosef Bacik 536727011e0SChris Mason struct extent_buffer *find_eb_for_page(struct extent_io_tree *tree, 537727011e0SChris Mason struct page *page, int max_walk) 538727011e0SChris Mason { 539727011e0SChris Mason struct extent_buffer *eb; 540727011e0SChris Mason u64 start = page_offset(page); 541727011e0SChris Mason u64 target = start; 542727011e0SChris Mason u64 min_start; 543727011e0SChris Mason 544727011e0SChris Mason if (start < max_walk) 545727011e0SChris Mason min_start = 0; 546727011e0SChris Mason else 547727011e0SChris Mason min_start = start - max_walk; 548727011e0SChris Mason 549727011e0SChris Mason while (start >= min_start) { 550727011e0SChris Mason eb = find_extent_buffer(tree, start, 0); 551727011e0SChris Mason if (eb) { 552727011e0SChris Mason /* 553727011e0SChris Mason * we found an extent buffer and it contains our page 554727011e0SChris Mason * horray! 555727011e0SChris Mason */ 556727011e0SChris Mason if (eb->start <= target && 557727011e0SChris Mason eb->start + eb->len > target) 558727011e0SChris Mason return eb; 559727011e0SChris Mason 560727011e0SChris Mason /* we found an extent buffer that wasn't for us */ 561727011e0SChris Mason free_extent_buffer(eb); 562727011e0SChris Mason return NULL; 563727011e0SChris Mason } 564727011e0SChris Mason if (start == 0) 565727011e0SChris Mason break; 566727011e0SChris Mason start -= PAGE_CACHE_SIZE; 567727011e0SChris Mason } 568727011e0SChris Mason return NULL; 569727011e0SChris Mason } 570727011e0SChris Mason 571b2950863SChristoph Hellwig static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end, 5725cf1ab56SJosef Bacik struct extent_state *state, int mirror) 573ce9adaa5SChris Mason { 574ce9adaa5SChris Mason struct extent_io_tree *tree; 575ce9adaa5SChris Mason u64 found_start; 576ce9adaa5SChris Mason int found_level; 577ce9adaa5SChris Mason struct extent_buffer *eb; 578ce9adaa5SChris Mason struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 579f188591eSChris Mason int ret = 0; 580727011e0SChris Mason int reads_done; 581ce9adaa5SChris Mason 582ce9adaa5SChris Mason if (!page->private) 583ce9adaa5SChris Mason goto out; 584d397712bSChris Mason 585727011e0SChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 5864f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 587d397712bSChris Mason 5880b32f4bbSJosef Bacik /* the pending IO might have been the only thing that kept this buffer 5890b32f4bbSJosef Bacik * in memory. Make sure we have a ref for all this other checks 5900b32f4bbSJosef Bacik */ 5910b32f4bbSJosef Bacik extent_buffer_get(eb); 5920b32f4bbSJosef Bacik 5930b32f4bbSJosef Bacik reads_done = atomic_dec_and_test(&eb->io_pages); 594727011e0SChris Mason if (!reads_done) 595727011e0SChris Mason goto err; 596f188591eSChris Mason 5975cf1ab56SJosef Bacik eb->read_mirror = mirror; 598ea466794SJosef Bacik if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) { 599ea466794SJosef Bacik ret = -EIO; 600ea466794SJosef Bacik goto err; 601ea466794SJosef Bacik } 602ea466794SJosef Bacik 603ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 604727011e0SChris Mason if (found_start != eb->start) { 6057a36ddecSDavid Sterba printk_ratelimited(KERN_INFO "btrfs bad tree block start " 606193f284dSChris Mason "%llu %llu\n", 607a1b32a59SChris Mason (unsigned long long)found_start, 608a1b32a59SChris Mason (unsigned long long)eb->start); 609f188591eSChris Mason ret = -EIO; 610ce9adaa5SChris Mason goto err; 611ce9adaa5SChris Mason } 6122b82032cSYan Zheng if (check_tree_block_fsid(root, eb)) { 6137a36ddecSDavid Sterba printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n", 614d397712bSChris Mason (unsigned long long)eb->start); 6151259ab75SChris Mason ret = -EIO; 6161259ab75SChris Mason goto err; 6171259ab75SChris Mason } 618ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 619ce9adaa5SChris Mason 62085d4e461SChris Mason btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), 62185d4e461SChris Mason eb, found_level); 6224008c04aSChris Mason 623ce9adaa5SChris Mason ret = csum_tree_block(root, eb, 1); 624a826d6dcSJosef Bacik if (ret) { 625f188591eSChris Mason ret = -EIO; 626a826d6dcSJosef Bacik goto err; 627a826d6dcSJosef Bacik } 628a826d6dcSJosef Bacik 629a826d6dcSJosef Bacik /* 630a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 631a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 632a826d6dcSJosef Bacik * return -EIO. 633a826d6dcSJosef Bacik */ 634a826d6dcSJosef Bacik if (found_level == 0 && check_leaf(root, eb)) { 635a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 636a826d6dcSJosef Bacik ret = -EIO; 637a826d6dcSJosef Bacik } 638ce9adaa5SChris Mason 6390b32f4bbSJosef Bacik if (!ret) 6400b32f4bbSJosef Bacik set_extent_buffer_uptodate(eb); 641ce9adaa5SChris Mason err: 6424bb31e92SArne Jansen if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) { 6434bb31e92SArne Jansen clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags); 6444bb31e92SArne Jansen btree_readahead_hook(root, eb, eb->start, ret); 6454bb31e92SArne Jansen } 6464bb31e92SArne Jansen 6470b32f4bbSJosef Bacik if (ret) 6480b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 6490b32f4bbSJosef Bacik free_extent_buffer(eb); 650ce9adaa5SChris Mason out: 651f188591eSChris Mason return ret; 652ce9adaa5SChris Mason } 653ce9adaa5SChris Mason 654ea466794SJosef Bacik static int btree_io_failed_hook(struct page *page, int failed_mirror) 6554bb31e92SArne Jansen { 6564bb31e92SArne Jansen struct extent_buffer *eb; 6574bb31e92SArne Jansen struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 6584bb31e92SArne Jansen 6594f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 660ea466794SJosef Bacik set_bit(EXTENT_BUFFER_IOERR, &eb->bflags); 6615cf1ab56SJosef Bacik eb->read_mirror = failed_mirror; 662ea466794SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 6634bb31e92SArne Jansen btree_readahead_hook(root, eb, eb->start, -EIO); 6644bb31e92SArne Jansen return -EIO; /* we fixed nothing */ 6654bb31e92SArne Jansen } 6664bb31e92SArne Jansen 667ce9adaa5SChris Mason static void end_workqueue_bio(struct bio *bio, int err) 668ce9adaa5SChris Mason { 669ce9adaa5SChris Mason struct end_io_wq *end_io_wq = bio->bi_private; 670ce9adaa5SChris Mason struct btrfs_fs_info *fs_info; 671ce9adaa5SChris Mason 672ce9adaa5SChris Mason fs_info = end_io_wq->info; 673ce9adaa5SChris Mason end_io_wq->error = err; 6748b712842SChris Mason end_io_wq->work.func = end_workqueue_fn; 6758b712842SChris Mason end_io_wq->work.flags = 0; 676d20f7043SChris Mason 6777b6d91daSChristoph Hellwig if (bio->bi_rw & REQ_WRITE) { 6780cb59c99SJosef Bacik if (end_io_wq->metadata == 1) 679cad321adSChris Mason btrfs_queue_worker(&fs_info->endio_meta_write_workers, 680cad321adSChris Mason &end_io_wq->work); 6810cb59c99SJosef Bacik else if (end_io_wq->metadata == 2) 6820cb59c99SJosef Bacik btrfs_queue_worker(&fs_info->endio_freespace_worker, 6830cb59c99SJosef Bacik &end_io_wq->work); 684cad321adSChris Mason else 685e6dcd2dcSChris Mason btrfs_queue_worker(&fs_info->endio_write_workers, 686e6dcd2dcSChris Mason &end_io_wq->work); 687d20f7043SChris Mason } else { 688d20f7043SChris Mason if (end_io_wq->metadata) 689d20f7043SChris Mason btrfs_queue_worker(&fs_info->endio_meta_workers, 690d20f7043SChris Mason &end_io_wq->work); 691e6dcd2dcSChris Mason else 692d20f7043SChris Mason btrfs_queue_worker(&fs_info->endio_workers, 693d20f7043SChris Mason &end_io_wq->work); 694d20f7043SChris Mason } 695ce9adaa5SChris Mason } 696ce9adaa5SChris Mason 6970cb59c99SJosef Bacik /* 6980cb59c99SJosef Bacik * For the metadata arg you want 6990cb59c99SJosef Bacik * 7000cb59c99SJosef Bacik * 0 - if data 7010cb59c99SJosef Bacik * 1 - if normal metadta 7020cb59c99SJosef Bacik * 2 - if writing to the free space cache area 7030cb59c99SJosef Bacik */ 70422c59948SChris Mason int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 70522c59948SChris Mason int metadata) 7060b86a832SChris Mason { 707ce9adaa5SChris Mason struct end_io_wq *end_io_wq; 708ce9adaa5SChris Mason end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS); 709ce9adaa5SChris Mason if (!end_io_wq) 710ce9adaa5SChris Mason return -ENOMEM; 711ce9adaa5SChris Mason 712ce9adaa5SChris Mason end_io_wq->private = bio->bi_private; 713ce9adaa5SChris Mason end_io_wq->end_io = bio->bi_end_io; 71422c59948SChris Mason end_io_wq->info = info; 715ce9adaa5SChris Mason end_io_wq->error = 0; 716ce9adaa5SChris Mason end_io_wq->bio = bio; 71722c59948SChris Mason end_io_wq->metadata = metadata; 718ce9adaa5SChris Mason 719ce9adaa5SChris Mason bio->bi_private = end_io_wq; 720ce9adaa5SChris Mason bio->bi_end_io = end_workqueue_bio; 72122c59948SChris Mason return 0; 72222c59948SChris Mason } 72322c59948SChris Mason 724b64a2851SChris Mason unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info) 7254854ddd0SChris Mason { 7264854ddd0SChris Mason unsigned long limit = min_t(unsigned long, 7274854ddd0SChris Mason info->workers.max_workers, 7284854ddd0SChris Mason info->fs_devices->open_devices); 7294854ddd0SChris Mason return 256 * limit; 7304854ddd0SChris Mason } 7314854ddd0SChris Mason 7324a69a410SChris Mason static void run_one_async_start(struct btrfs_work *work) 7334a69a410SChris Mason { 7344a69a410SChris Mason struct async_submit_bio *async; 73579787eaaSJeff Mahoney int ret; 7364a69a410SChris Mason 7374a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 73879787eaaSJeff Mahoney ret = async->submit_bio_start(async->inode, async->rw, async->bio, 739eaf25d93SChris Mason async->mirror_num, async->bio_flags, 740eaf25d93SChris Mason async->bio_offset); 74179787eaaSJeff Mahoney if (ret) 74279787eaaSJeff Mahoney async->error = ret; 7434a69a410SChris Mason } 7444a69a410SChris Mason 7454a69a410SChris Mason static void run_one_async_done(struct btrfs_work *work) 7468b712842SChris Mason { 7478b712842SChris Mason struct btrfs_fs_info *fs_info; 7488b712842SChris Mason struct async_submit_bio *async; 7494854ddd0SChris Mason int limit; 7508b712842SChris Mason 7518b712842SChris Mason async = container_of(work, struct async_submit_bio, work); 7528b712842SChris Mason fs_info = BTRFS_I(async->inode)->root->fs_info; 7534854ddd0SChris Mason 754b64a2851SChris Mason limit = btrfs_async_submit_limit(fs_info); 7554854ddd0SChris Mason limit = limit * 2 / 3; 7564854ddd0SChris Mason 7578b712842SChris Mason atomic_dec(&fs_info->nr_async_submits); 7580986fe9eSChris Mason 759b64a2851SChris Mason if (atomic_read(&fs_info->nr_async_submits) < limit && 760b64a2851SChris Mason waitqueue_active(&fs_info->async_submit_wait)) 7614854ddd0SChris Mason wake_up(&fs_info->async_submit_wait); 7624854ddd0SChris Mason 76379787eaaSJeff Mahoney /* If an error occured we just want to clean up the bio and move on */ 76479787eaaSJeff Mahoney if (async->error) { 76579787eaaSJeff Mahoney bio_endio(async->bio, async->error); 76679787eaaSJeff Mahoney return; 76779787eaaSJeff Mahoney } 76879787eaaSJeff Mahoney 7694a69a410SChris Mason async->submit_bio_done(async->inode, async->rw, async->bio, 770eaf25d93SChris Mason async->mirror_num, async->bio_flags, 771eaf25d93SChris Mason async->bio_offset); 7724a69a410SChris Mason } 7734a69a410SChris Mason 7744a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 7754a69a410SChris Mason { 7764a69a410SChris Mason struct async_submit_bio *async; 7774a69a410SChris Mason 7784a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 7798b712842SChris Mason kfree(async); 7808b712842SChris Mason } 7818b712842SChris Mason 78244b8bd7eSChris Mason int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode, 78344b8bd7eSChris Mason int rw, struct bio *bio, int mirror_num, 784c8b97818SChris Mason unsigned long bio_flags, 785eaf25d93SChris Mason u64 bio_offset, 7864a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_start, 7874a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_done) 78844b8bd7eSChris Mason { 78944b8bd7eSChris Mason struct async_submit_bio *async; 79044b8bd7eSChris Mason 79144b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 79244b8bd7eSChris Mason if (!async) 79344b8bd7eSChris Mason return -ENOMEM; 79444b8bd7eSChris Mason 79544b8bd7eSChris Mason async->inode = inode; 79644b8bd7eSChris Mason async->rw = rw; 79744b8bd7eSChris Mason async->bio = bio; 79844b8bd7eSChris Mason async->mirror_num = mirror_num; 7994a69a410SChris Mason async->submit_bio_start = submit_bio_start; 8004a69a410SChris Mason async->submit_bio_done = submit_bio_done; 8014a69a410SChris Mason 8024a69a410SChris Mason async->work.func = run_one_async_start; 8034a69a410SChris Mason async->work.ordered_func = run_one_async_done; 8044a69a410SChris Mason async->work.ordered_free = run_one_async_free; 8054a69a410SChris Mason 8068b712842SChris Mason async->work.flags = 0; 807c8b97818SChris Mason async->bio_flags = bio_flags; 808eaf25d93SChris Mason async->bio_offset = bio_offset; 8098c8bee1dSChris Mason 81079787eaaSJeff Mahoney async->error = 0; 81179787eaaSJeff Mahoney 812cb03c743SChris Mason atomic_inc(&fs_info->nr_async_submits); 813d313d7a3SChris Mason 8147b6d91daSChristoph Hellwig if (rw & REQ_SYNC) 815d313d7a3SChris Mason btrfs_set_work_high_prio(&async->work); 816d313d7a3SChris Mason 8178b712842SChris Mason btrfs_queue_worker(&fs_info->workers, &async->work); 8189473f16cSChris Mason 819771ed689SChris Mason while (atomic_read(&fs_info->async_submit_draining) && 820771ed689SChris Mason atomic_read(&fs_info->nr_async_submits)) { 821771ed689SChris Mason wait_event(fs_info->async_submit_wait, 822771ed689SChris Mason (atomic_read(&fs_info->nr_async_submits) == 0)); 823771ed689SChris Mason } 824771ed689SChris Mason 82544b8bd7eSChris Mason return 0; 82644b8bd7eSChris Mason } 82744b8bd7eSChris Mason 828ce3ed71aSChris Mason static int btree_csum_one_bio(struct bio *bio) 829ce3ed71aSChris Mason { 830ce3ed71aSChris Mason struct bio_vec *bvec = bio->bi_io_vec; 831ce3ed71aSChris Mason int bio_index = 0; 832ce3ed71aSChris Mason struct btrfs_root *root; 83379787eaaSJeff Mahoney int ret = 0; 834ce3ed71aSChris Mason 835ce3ed71aSChris Mason WARN_ON(bio->bi_vcnt <= 0); 836ce3ed71aSChris Mason while (bio_index < bio->bi_vcnt) { 837ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 83879787eaaSJeff Mahoney ret = csum_dirty_buffer(root, bvec->bv_page); 83979787eaaSJeff Mahoney if (ret) 84079787eaaSJeff Mahoney break; 841ce3ed71aSChris Mason bio_index++; 842ce3ed71aSChris Mason bvec++; 843ce3ed71aSChris Mason } 84479787eaaSJeff Mahoney return ret; 845ce3ed71aSChris Mason } 846ce3ed71aSChris Mason 8474a69a410SChris Mason static int __btree_submit_bio_start(struct inode *inode, int rw, 8484a69a410SChris Mason struct bio *bio, int mirror_num, 849eaf25d93SChris Mason unsigned long bio_flags, 850eaf25d93SChris Mason u64 bio_offset) 85122c59948SChris Mason { 8528b712842SChris Mason /* 8538b712842SChris Mason * when we're called for a write, we're already in the async 8545443be45SChris Mason * submission context. Just jump into btrfs_map_bio 8558b712842SChris Mason */ 85679787eaaSJeff Mahoney return btree_csum_one_bio(bio); 85722c59948SChris Mason } 85822c59948SChris Mason 8594a69a410SChris Mason static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio, 860eaf25d93SChris Mason int mirror_num, unsigned long bio_flags, 861eaf25d93SChris Mason u64 bio_offset) 8624a69a410SChris Mason { 8638b712842SChris Mason /* 8644a69a410SChris Mason * when we're called for a write, we're already in the async 8654a69a410SChris Mason * submission context. Just jump into btrfs_map_bio 8668b712842SChris Mason */ 8678b712842SChris Mason return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1); 8680b86a832SChris Mason } 8690b86a832SChris Mason 87044b8bd7eSChris Mason static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, 871eaf25d93SChris Mason int mirror_num, unsigned long bio_flags, 872eaf25d93SChris Mason u64 bio_offset) 87344b8bd7eSChris Mason { 8744a69a410SChris Mason int ret; 875cad321adSChris Mason 8767b6d91daSChristoph Hellwig if (!(rw & REQ_WRITE)) { 877f3f266abSChris Mason 878cad321adSChris Mason /* 879cad321adSChris Mason * called for a read, do the setup so that checksum validation 880cad321adSChris Mason * can happen in the async kernel threads 881cad321adSChris Mason */ 882f3f266abSChris Mason ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info, 883f3f266abSChris Mason bio, 1); 8841d4284bdSChris Mason if (ret) 8851d4284bdSChris Mason return ret; 8864a69a410SChris Mason return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, 8876f3577bdSChris Mason mirror_num, 0); 88844b8bd7eSChris Mason } 889d313d7a3SChris Mason 890cad321adSChris Mason /* 891cad321adSChris Mason * kthread helpers are used to submit writes so that checksumming 892cad321adSChris Mason * can happen in parallel across all CPUs 893cad321adSChris Mason */ 89444b8bd7eSChris Mason return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, 895c8b97818SChris Mason inode, rw, bio, mirror_num, 0, 896eaf25d93SChris Mason bio_offset, 8974a69a410SChris Mason __btree_submit_bio_start, 8984a69a410SChris Mason __btree_submit_bio_done); 89944b8bd7eSChris Mason } 90044b8bd7eSChris Mason 9013dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 902784b4e29SChris Mason static int btree_migratepage(struct address_space *mapping, 903a6bc32b8SMel Gorman struct page *newpage, struct page *page, 904a6bc32b8SMel Gorman enum migrate_mode mode) 905784b4e29SChris Mason { 906784b4e29SChris Mason /* 907784b4e29SChris Mason * we can't safely write a btree page from here, 908784b4e29SChris Mason * we haven't done the locking hook 909784b4e29SChris Mason */ 910784b4e29SChris Mason if (PageDirty(page)) 911784b4e29SChris Mason return -EAGAIN; 912784b4e29SChris Mason /* 913784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 914784b4e29SChris Mason * We must have no buffers or drop them. 915784b4e29SChris Mason */ 916784b4e29SChris Mason if (page_has_private(page) && 917784b4e29SChris Mason !try_to_release_page(page, GFP_KERNEL)) 918784b4e29SChris Mason return -EAGAIN; 919a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 920784b4e29SChris Mason } 9213dd1462eSJan Beulich #endif 922784b4e29SChris Mason 9230da5468fSChris Mason 9240da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 9250da5468fSChris Mason struct writeback_control *wbc) 9260da5468fSChris Mason { 927d1310b2eSChris Mason struct extent_io_tree *tree; 928d1310b2eSChris Mason tree = &BTRFS_I(mapping->host)->io_tree; 929d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 930b9473439SChris Mason struct btrfs_root *root = BTRFS_I(mapping->host)->root; 931793955bcSChris Mason u64 num_dirty; 93224ab9cd8SChris Mason unsigned long thresh = 32 * 1024 * 1024; 933448d640bSChris Mason 934448d640bSChris Mason if (wbc->for_kupdate) 935448d640bSChris Mason return 0; 936448d640bSChris Mason 937b9473439SChris Mason /* this is a bit racy, but that's ok */ 938b9473439SChris Mason num_dirty = root->fs_info->dirty_metadata_bytes; 939d397712bSChris Mason if (num_dirty < thresh) 940793955bcSChris Mason return 0; 941793955bcSChris Mason } 9420b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 9430da5468fSChris Mason } 9440da5468fSChris Mason 945b2950863SChristoph Hellwig static int btree_readpage(struct file *file, struct page *page) 9465f39d397SChris Mason { 947d1310b2eSChris Mason struct extent_io_tree *tree; 948d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9498ddc7d9cSJan Schmidt return extent_read_full_page(tree, page, btree_get_extent, 0); 9505f39d397SChris Mason } 9515f39d397SChris Mason 95270dec807SChris Mason static int btree_releasepage(struct page *page, gfp_t gfp_flags) 9535f39d397SChris Mason { 95498509cfcSChris Mason if (PageWriteback(page) || PageDirty(page)) 95598509cfcSChris Mason return 0; 9560c4e538bSDavid Sterba /* 9570c4e538bSDavid Sterba * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing 9580c4e538bSDavid Sterba * slab allocation from alloc_extent_state down the callchain where 9590c4e538bSDavid Sterba * it'd hit a BUG_ON as those flags are not allowed. 9600c4e538bSDavid Sterba */ 9610c4e538bSDavid Sterba gfp_flags &= ~GFP_SLAB_BUG_MASK; 9620c4e538bSDavid Sterba 9633083ee2eSJosef Bacik return try_release_extent_buffer(page, gfp_flags); 964d98237b3SChris Mason } 965d98237b3SChris Mason 9665f39d397SChris Mason static void btree_invalidatepage(struct page *page, unsigned long offset) 967d98237b3SChris Mason { 968d1310b2eSChris Mason struct extent_io_tree *tree; 969d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9705f39d397SChris Mason extent_invalidatepage(tree, page, offset); 9715f39d397SChris Mason btree_releasepage(page, GFP_NOFS); 9729ad6b7bcSChris Mason if (PagePrivate(page)) { 973d397712bSChris Mason printk(KERN_WARNING "btrfs warning page private not zero " 974d397712bSChris Mason "on page %llu\n", (unsigned long long)page_offset(page)); 9759ad6b7bcSChris Mason ClearPagePrivate(page); 9769ad6b7bcSChris Mason set_page_private(page, 0); 9779ad6b7bcSChris Mason page_cache_release(page); 9789ad6b7bcSChris Mason } 979d98237b3SChris Mason } 980d98237b3SChris Mason 9810b32f4bbSJosef Bacik static int btree_set_page_dirty(struct page *page) 9820b32f4bbSJosef Bacik { 9830b32f4bbSJosef Bacik struct extent_buffer *eb; 9840b32f4bbSJosef Bacik 9850b32f4bbSJosef Bacik BUG_ON(!PagePrivate(page)); 9860b32f4bbSJosef Bacik eb = (struct extent_buffer *)page->private; 9870b32f4bbSJosef Bacik BUG_ON(!eb); 9880b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 9890b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 9900b32f4bbSJosef Bacik btrfs_assert_tree_locked(eb); 9910b32f4bbSJosef Bacik return __set_page_dirty_nobuffers(page); 9920b32f4bbSJosef Bacik } 9930b32f4bbSJosef Bacik 9947f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 995d98237b3SChris Mason .readpage = btree_readpage, 9960da5468fSChris Mason .writepages = btree_writepages, 9975f39d397SChris Mason .releasepage = btree_releasepage, 9985f39d397SChris Mason .invalidatepage = btree_invalidatepage, 9995a92bc88SChris Mason #ifdef CONFIG_MIGRATION 1000784b4e29SChris Mason .migratepage = btree_migratepage, 10015a92bc88SChris Mason #endif 10020b32f4bbSJosef Bacik .set_page_dirty = btree_set_page_dirty, 1003d98237b3SChris Mason }; 1004123abc88SChris Mason 1005ca7a79adSChris Mason int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize, 1006ca7a79adSChris Mason u64 parent_transid) 1007090d1875SChris Mason { 10085f39d397SChris Mason struct extent_buffer *buf = NULL; 10095f39d397SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 1010de428b63SChris Mason int ret = 0; 1011090d1875SChris Mason 1012db94535dSChris Mason buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 10135f39d397SChris Mason if (!buf) 1014090d1875SChris Mason return 0; 1015d1310b2eSChris Mason read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, 1016bb82ab88SArne Jansen buf, 0, WAIT_NONE, btree_get_extent, 0); 10175f39d397SChris Mason free_extent_buffer(buf); 1018de428b63SChris Mason return ret; 1019090d1875SChris Mason } 1020090d1875SChris Mason 1021ab0fff03SArne Jansen int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize, 1022ab0fff03SArne Jansen int mirror_num, struct extent_buffer **eb) 1023ab0fff03SArne Jansen { 1024ab0fff03SArne Jansen struct extent_buffer *buf = NULL; 1025ab0fff03SArne Jansen struct inode *btree_inode = root->fs_info->btree_inode; 1026ab0fff03SArne Jansen struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree; 1027ab0fff03SArne Jansen int ret; 1028ab0fff03SArne Jansen 1029ab0fff03SArne Jansen buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 1030ab0fff03SArne Jansen if (!buf) 1031ab0fff03SArne Jansen return 0; 1032ab0fff03SArne Jansen 1033ab0fff03SArne Jansen set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags); 1034ab0fff03SArne Jansen 1035ab0fff03SArne Jansen ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK, 1036ab0fff03SArne Jansen btree_get_extent, mirror_num); 1037ab0fff03SArne Jansen if (ret) { 1038ab0fff03SArne Jansen free_extent_buffer(buf); 1039ab0fff03SArne Jansen return ret; 1040ab0fff03SArne Jansen } 1041ab0fff03SArne Jansen 1042ab0fff03SArne Jansen if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) { 1043ab0fff03SArne Jansen free_extent_buffer(buf); 1044ab0fff03SArne Jansen return -EIO; 10450b32f4bbSJosef Bacik } else if (extent_buffer_uptodate(buf)) { 1046ab0fff03SArne Jansen *eb = buf; 1047ab0fff03SArne Jansen } else { 1048ab0fff03SArne Jansen free_extent_buffer(buf); 1049ab0fff03SArne Jansen } 1050ab0fff03SArne Jansen return 0; 1051ab0fff03SArne Jansen } 1052ab0fff03SArne Jansen 10530999df54SChris Mason struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root, 10540999df54SChris Mason u64 bytenr, u32 blocksize) 10550999df54SChris Mason { 10560999df54SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 10570999df54SChris Mason struct extent_buffer *eb; 10580999df54SChris Mason eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree, 1059f09d1f60SDavid Sterba bytenr, blocksize); 10600999df54SChris Mason return eb; 10610999df54SChris Mason } 10620999df54SChris Mason 10630999df54SChris Mason struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root, 10640999df54SChris Mason u64 bytenr, u32 blocksize) 10650999df54SChris Mason { 10660999df54SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 10670999df54SChris Mason struct extent_buffer *eb; 10680999df54SChris Mason 10690999df54SChris Mason eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree, 1070727011e0SChris Mason bytenr, blocksize); 10710999df54SChris Mason return eb; 10720999df54SChris Mason } 10730999df54SChris Mason 10740999df54SChris Mason 1075e02119d5SChris Mason int btrfs_write_tree_block(struct extent_buffer *buf) 1076e02119d5SChris Mason { 1077727011e0SChris Mason return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start, 10788aa38c31SChristoph Hellwig buf->start + buf->len - 1); 1079e02119d5SChris Mason } 1080e02119d5SChris Mason 1081e02119d5SChris Mason int btrfs_wait_tree_block_writeback(struct extent_buffer *buf) 1082e02119d5SChris Mason { 1083727011e0SChris Mason return filemap_fdatawait_range(buf->pages[0]->mapping, 10848aa38c31SChristoph Hellwig buf->start, buf->start + buf->len - 1); 1085e02119d5SChris Mason } 1086e02119d5SChris Mason 1087db94535dSChris Mason struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr, 1088ca7a79adSChris Mason u32 blocksize, u64 parent_transid) 1089e20d96d6SChris Mason { 10905f39d397SChris Mason struct extent_buffer *buf = NULL; 109119c00ddcSChris Mason int ret; 109219c00ddcSChris Mason 1093db94535dSChris Mason buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 10945f39d397SChris Mason if (!buf) 1095d98237b3SChris Mason return NULL; 1096e4204dedSChris Mason 1097ca7a79adSChris Mason ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 10985f39d397SChris Mason return buf; 1099ce9adaa5SChris Mason 1100eb60ceacSChris Mason } 1101eb60ceacSChris Mason 1102d5c13f92SJeff Mahoney void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root, 11035f39d397SChris Mason struct extent_buffer *buf) 1104ed2ff2cbSChris Mason { 110555c69072SChris Mason if (btrfs_header_generation(buf) == 1106925baeddSChris Mason root->fs_info->running_transaction->transid) { 1107b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1108b4ce94deSChris Mason 1109b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1110b9473439SChris Mason spin_lock(&root->fs_info->delalloc_lock); 1111b9473439SChris Mason if (root->fs_info->dirty_metadata_bytes >= buf->len) 1112b9473439SChris Mason root->fs_info->dirty_metadata_bytes -= buf->len; 1113d5c13f92SJeff Mahoney else { 1114d5c13f92SJeff Mahoney spin_unlock(&root->fs_info->delalloc_lock); 1115d5c13f92SJeff Mahoney btrfs_panic(root->fs_info, -EOVERFLOW, 1116d5c13f92SJeff Mahoney "Can't clear %lu bytes from " 1117d5c13f92SJeff Mahoney " dirty_mdatadata_bytes (%lu)", 1118d5c13f92SJeff Mahoney buf->len, 1119d5c13f92SJeff Mahoney root->fs_info->dirty_metadata_bytes); 1120d5c13f92SJeff Mahoney } 1121b9473439SChris Mason spin_unlock(&root->fs_info->delalloc_lock); 1122b9473439SChris Mason } 1123b4ce94deSChris Mason 1124b9473439SChris Mason /* ugh, clear_extent_buffer_dirty needs to lock the page */ 1125b9473439SChris Mason btrfs_set_lock_blocking(buf); 11260b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 1127925baeddSChris Mason } 11285f39d397SChris Mason } 11295f39d397SChris Mason 1130143bede5SJeff Mahoney static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize, 113187ee04ebSChris Mason u32 stripesize, struct btrfs_root *root, 11329f5fae2fSChris Mason struct btrfs_fs_info *fs_info, 1133e20d96d6SChris Mason u64 objectid) 1134d97e63b6SChris Mason { 1135cfaa7295SChris Mason root->node = NULL; 1136a28ec197SChris Mason root->commit_root = NULL; 1137db94535dSChris Mason root->sectorsize = sectorsize; 1138db94535dSChris Mason root->nodesize = nodesize; 1139db94535dSChris Mason root->leafsize = leafsize; 114087ee04ebSChris Mason root->stripesize = stripesize; 1141123abc88SChris Mason root->ref_cows = 0; 11420b86a832SChris Mason root->track_dirty = 0; 1143c71bf099SYan, Zheng root->in_radix = 0; 1144d68fc57bSYan, Zheng root->orphan_item_inserted = 0; 1145d68fc57bSYan, Zheng root->orphan_cleanup_state = 0; 11460b86a832SChris Mason 11470f7d52f4SChris Mason root->objectid = objectid; 11480f7d52f4SChris Mason root->last_trans = 0; 114913a8a7c8SYan, Zheng root->highest_objectid = 0; 115058176a96SJosef Bacik root->name = NULL; 11516bef4d31SEric Paris root->inode_tree = RB_ROOT; 115216cdcec7SMiao Xie INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1153f0486c68SYan, Zheng root->block_rsv = NULL; 1154d68fc57bSYan, Zheng root->orphan_block_rsv = NULL; 11550b86a832SChris Mason 11560b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 11575d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 1158d68fc57bSYan, Zheng spin_lock_init(&root->orphan_lock); 11595d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1160f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 1161a2135011SChris Mason mutex_init(&root->objectid_mutex); 1162e02119d5SChris Mason mutex_init(&root->log_mutex); 11637237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 11647237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 11657237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 11667237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 11677237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 11687237f183SYan Zheng atomic_set(&root->log_writers, 0); 11698a35d95fSJosef Bacik atomic_set(&root->orphan_inodes, 0); 11707237f183SYan Zheng root->log_batch = 0; 11717237f183SYan Zheng root->log_transid = 0; 1172257c62e1SChris Mason root->last_log_commit = 0; 1173d0c803c4SChris Mason extent_io_tree_init(&root->dirty_log_pages, 1174f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 1175017e5369SChris Mason 11763768f368SChris Mason memset(&root->root_key, 0, sizeof(root->root_key)); 11773768f368SChris Mason memset(&root->root_item, 0, sizeof(root->root_item)); 11786702ed49SChris Mason memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 117958176a96SJosef Bacik memset(&root->root_kobj, 0, sizeof(root->root_kobj)); 11803f157a2fSChris Mason root->defrag_trans_start = fs_info->generation; 118158176a96SJosef Bacik init_completion(&root->kobj_unregister); 11826702ed49SChris Mason root->defrag_running = 0; 11834d775673SChris Mason root->root_key.objectid = objectid; 11840ee5dc67SAl Viro root->anon_dev = 0; 11853768f368SChris Mason } 11863768f368SChris Mason 1187200a5c17SJeff Mahoney static int __must_check find_and_setup_root(struct btrfs_root *tree_root, 11889f5fae2fSChris Mason struct btrfs_fs_info *fs_info, 11899f5fae2fSChris Mason u64 objectid, 1190e20d96d6SChris Mason struct btrfs_root *root) 11913768f368SChris Mason { 11923768f368SChris Mason int ret; 1193db94535dSChris Mason u32 blocksize; 119484234f3aSYan Zheng u64 generation; 11953768f368SChris Mason 1196db94535dSChris Mason __setup_root(tree_root->nodesize, tree_root->leafsize, 119787ee04ebSChris Mason tree_root->sectorsize, tree_root->stripesize, 119887ee04ebSChris Mason root, fs_info, objectid); 11993768f368SChris Mason ret = btrfs_find_last_root(tree_root, objectid, 12003768f368SChris Mason &root->root_item, &root->root_key); 12014df27c4dSYan, Zheng if (ret > 0) 12024df27c4dSYan, Zheng return -ENOENT; 1203200a5c17SJeff Mahoney else if (ret < 0) 1204200a5c17SJeff Mahoney return ret; 12053768f368SChris Mason 120684234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1207db94535dSChris Mason blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item)); 1208af31f5e5SChris Mason root->commit_root = NULL; 1209db94535dSChris Mason root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item), 121084234f3aSYan Zheng blocksize, generation); 1211b9fab919SChris Mason if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) { 121268433b73SChris Mason free_extent_buffer(root->node); 1213af31f5e5SChris Mason root->node = NULL; 121468433b73SChris Mason return -EIO; 121568433b73SChris Mason } 12164df27c4dSYan, Zheng root->commit_root = btrfs_root_node(root); 1217d97e63b6SChris Mason return 0; 1218d97e63b6SChris Mason } 1219d97e63b6SChris Mason 1220f84a8bd6SAl Viro static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info) 12216f07e42eSAl Viro { 12226f07e42eSAl Viro struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS); 12236f07e42eSAl Viro if (root) 12246f07e42eSAl Viro root->fs_info = fs_info; 12256f07e42eSAl Viro return root; 12266f07e42eSAl Viro } 12276f07e42eSAl Viro 122820897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 122920897f5cSArne Jansen struct btrfs_fs_info *fs_info, 123020897f5cSArne Jansen u64 objectid) 123120897f5cSArne Jansen { 123220897f5cSArne Jansen struct extent_buffer *leaf; 123320897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 123420897f5cSArne Jansen struct btrfs_root *root; 123520897f5cSArne Jansen struct btrfs_key key; 123620897f5cSArne Jansen int ret = 0; 123720897f5cSArne Jansen u64 bytenr; 123820897f5cSArne Jansen 123920897f5cSArne Jansen root = btrfs_alloc_root(fs_info); 124020897f5cSArne Jansen if (!root) 124120897f5cSArne Jansen return ERR_PTR(-ENOMEM); 124220897f5cSArne Jansen 124320897f5cSArne Jansen __setup_root(tree_root->nodesize, tree_root->leafsize, 124420897f5cSArne Jansen tree_root->sectorsize, tree_root->stripesize, 124520897f5cSArne Jansen root, fs_info, objectid); 124620897f5cSArne Jansen root->root_key.objectid = objectid; 124720897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 124820897f5cSArne Jansen root->root_key.offset = 0; 124920897f5cSArne Jansen 125020897f5cSArne Jansen leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 125120897f5cSArne Jansen 0, objectid, NULL, 0, 0, 0); 125220897f5cSArne Jansen if (IS_ERR(leaf)) { 125320897f5cSArne Jansen ret = PTR_ERR(leaf); 125420897f5cSArne Jansen goto fail; 125520897f5cSArne Jansen } 125620897f5cSArne Jansen 125720897f5cSArne Jansen bytenr = leaf->start; 125820897f5cSArne Jansen memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 125920897f5cSArne Jansen btrfs_set_header_bytenr(leaf, leaf->start); 126020897f5cSArne Jansen btrfs_set_header_generation(leaf, trans->transid); 126120897f5cSArne Jansen btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 126220897f5cSArne Jansen btrfs_set_header_owner(leaf, objectid); 126320897f5cSArne Jansen root->node = leaf; 126420897f5cSArne Jansen 126520897f5cSArne Jansen write_extent_buffer(leaf, fs_info->fsid, 126620897f5cSArne Jansen (unsigned long)btrfs_header_fsid(leaf), 126720897f5cSArne Jansen BTRFS_FSID_SIZE); 126820897f5cSArne Jansen write_extent_buffer(leaf, fs_info->chunk_tree_uuid, 126920897f5cSArne Jansen (unsigned long)btrfs_header_chunk_tree_uuid(leaf), 127020897f5cSArne Jansen BTRFS_UUID_SIZE); 127120897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 127220897f5cSArne Jansen 127320897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 127420897f5cSArne Jansen root->track_dirty = 1; 127520897f5cSArne Jansen 127620897f5cSArne Jansen 127720897f5cSArne Jansen root->root_item.flags = 0; 127820897f5cSArne Jansen root->root_item.byte_limit = 0; 127920897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 128020897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 128120897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 128220897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 128320897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 128420897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 128520897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 128620897f5cSArne Jansen root->root_item.drop_level = 0; 128720897f5cSArne Jansen 128820897f5cSArne Jansen key.objectid = objectid; 128920897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 129020897f5cSArne Jansen key.offset = 0; 129120897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 129220897f5cSArne Jansen if (ret) 129320897f5cSArne Jansen goto fail; 129420897f5cSArne Jansen 129520897f5cSArne Jansen btrfs_tree_unlock(leaf); 129620897f5cSArne Jansen 129720897f5cSArne Jansen fail: 129820897f5cSArne Jansen if (ret) 129920897f5cSArne Jansen return ERR_PTR(ret); 130020897f5cSArne Jansen 130120897f5cSArne Jansen return root; 130220897f5cSArne Jansen } 130320897f5cSArne Jansen 13047237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1305e02119d5SChris Mason struct btrfs_fs_info *fs_info) 13060f7d52f4SChris Mason { 13070f7d52f4SChris Mason struct btrfs_root *root; 13080f7d52f4SChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 13097237f183SYan Zheng struct extent_buffer *leaf; 1310e02119d5SChris Mason 13116f07e42eSAl Viro root = btrfs_alloc_root(fs_info); 1312e02119d5SChris Mason if (!root) 13137237f183SYan Zheng return ERR_PTR(-ENOMEM); 1314e02119d5SChris Mason 1315e02119d5SChris Mason __setup_root(tree_root->nodesize, tree_root->leafsize, 1316e02119d5SChris Mason tree_root->sectorsize, tree_root->stripesize, 1317e02119d5SChris Mason root, fs_info, BTRFS_TREE_LOG_OBJECTID); 1318e02119d5SChris Mason 1319e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1320e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1321e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 13227237f183SYan Zheng /* 13237237f183SYan Zheng * log trees do not get reference counted because they go away 13247237f183SYan Zheng * before a real commit is actually done. They do store pointers 13257237f183SYan Zheng * to file data extents, and those reference counts still get 13267237f183SYan Zheng * updated (along with back refs to the log tree). 13277237f183SYan Zheng */ 1328e02119d5SChris Mason root->ref_cows = 0; 1329e02119d5SChris Mason 13305d4f98a2SYan Zheng leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0, 133166d7e7f0SArne Jansen BTRFS_TREE_LOG_OBJECTID, NULL, 13325581a51aSJan Schmidt 0, 0, 0); 13337237f183SYan Zheng if (IS_ERR(leaf)) { 13347237f183SYan Zheng kfree(root); 13357237f183SYan Zheng return ERR_CAST(leaf); 13367237f183SYan Zheng } 1337e02119d5SChris Mason 13385d4f98a2SYan Zheng memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 13395d4f98a2SYan Zheng btrfs_set_header_bytenr(leaf, leaf->start); 13405d4f98a2SYan Zheng btrfs_set_header_generation(leaf, trans->transid); 13415d4f98a2SYan Zheng btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 13425d4f98a2SYan Zheng btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID); 13437237f183SYan Zheng root->node = leaf; 1344e02119d5SChris Mason 1345e02119d5SChris Mason write_extent_buffer(root->node, root->fs_info->fsid, 1346e02119d5SChris Mason (unsigned long)btrfs_header_fsid(root->node), 1347e02119d5SChris Mason BTRFS_FSID_SIZE); 1348e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1349e02119d5SChris Mason btrfs_tree_unlock(root->node); 13507237f183SYan Zheng return root; 13517237f183SYan Zheng } 13527237f183SYan Zheng 13537237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 13547237f183SYan Zheng struct btrfs_fs_info *fs_info) 13557237f183SYan Zheng { 13567237f183SYan Zheng struct btrfs_root *log_root; 13577237f183SYan Zheng 13587237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 13597237f183SYan Zheng if (IS_ERR(log_root)) 13607237f183SYan Zheng return PTR_ERR(log_root); 13617237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 13627237f183SYan Zheng fs_info->log_root_tree = log_root; 13637237f183SYan Zheng return 0; 13647237f183SYan Zheng } 13657237f183SYan Zheng 13667237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 13677237f183SYan Zheng struct btrfs_root *root) 13687237f183SYan Zheng { 13697237f183SYan Zheng struct btrfs_root *log_root; 13707237f183SYan Zheng struct btrfs_inode_item *inode_item; 13717237f183SYan Zheng 13727237f183SYan Zheng log_root = alloc_log_tree(trans, root->fs_info); 13737237f183SYan Zheng if (IS_ERR(log_root)) 13747237f183SYan Zheng return PTR_ERR(log_root); 13757237f183SYan Zheng 13767237f183SYan Zheng log_root->last_trans = trans->transid; 13777237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 13787237f183SYan Zheng 13797237f183SYan Zheng inode_item = &log_root->root_item.inode; 13807237f183SYan Zheng inode_item->generation = cpu_to_le64(1); 13817237f183SYan Zheng inode_item->size = cpu_to_le64(3); 13827237f183SYan Zheng inode_item->nlink = cpu_to_le32(1); 13837237f183SYan Zheng inode_item->nbytes = cpu_to_le64(root->leafsize); 13847237f183SYan Zheng inode_item->mode = cpu_to_le32(S_IFDIR | 0755); 13857237f183SYan Zheng 13865d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 13877237f183SYan Zheng 13887237f183SYan Zheng WARN_ON(root->log_root); 13897237f183SYan Zheng root->log_root = log_root; 13907237f183SYan Zheng root->log_transid = 0; 1391257c62e1SChris Mason root->last_log_commit = 0; 1392e02119d5SChris Mason return 0; 1393e02119d5SChris Mason } 1394e02119d5SChris Mason 1395e02119d5SChris Mason struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root, 1396e02119d5SChris Mason struct btrfs_key *location) 1397e02119d5SChris Mason { 1398e02119d5SChris Mason struct btrfs_root *root; 1399e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 14000f7d52f4SChris Mason struct btrfs_path *path; 14015f39d397SChris Mason struct extent_buffer *l; 140284234f3aSYan Zheng u64 generation; 1403db94535dSChris Mason u32 blocksize; 14040f7d52f4SChris Mason int ret = 0; 14050f7d52f4SChris Mason 14066f07e42eSAl Viro root = btrfs_alloc_root(fs_info); 14070cf6c620SChris Mason if (!root) 14080f7d52f4SChris Mason return ERR_PTR(-ENOMEM); 14090f7d52f4SChris Mason if (location->offset == (u64)-1) { 1410db94535dSChris Mason ret = find_and_setup_root(tree_root, fs_info, 14110f7d52f4SChris Mason location->objectid, root); 14120f7d52f4SChris Mason if (ret) { 14130f7d52f4SChris Mason kfree(root); 14140f7d52f4SChris Mason return ERR_PTR(ret); 14150f7d52f4SChris Mason } 141613a8a7c8SYan, Zheng goto out; 14170f7d52f4SChris Mason } 14180f7d52f4SChris Mason 1419db94535dSChris Mason __setup_root(tree_root->nodesize, tree_root->leafsize, 142087ee04ebSChris Mason tree_root->sectorsize, tree_root->stripesize, 142187ee04ebSChris Mason root, fs_info, location->objectid); 14220f7d52f4SChris Mason 14230f7d52f4SChris Mason path = btrfs_alloc_path(); 1424db5b493aSTsutomu Itoh if (!path) { 1425db5b493aSTsutomu Itoh kfree(root); 1426db5b493aSTsutomu Itoh return ERR_PTR(-ENOMEM); 1427db5b493aSTsutomu Itoh } 14280f7d52f4SChris Mason ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0); 142913a8a7c8SYan, Zheng if (ret == 0) { 14305f39d397SChris Mason l = path->nodes[0]; 14315f39d397SChris Mason read_extent_buffer(l, &root->root_item, 14325f39d397SChris Mason btrfs_item_ptr_offset(l, path->slots[0]), 14330f7d52f4SChris Mason sizeof(root->root_item)); 143444b36eb2SYan Zheng memcpy(&root->root_key, location, sizeof(*location)); 143513a8a7c8SYan, Zheng } 14360f7d52f4SChris Mason btrfs_free_path(path); 14370f7d52f4SChris Mason if (ret) { 14385e540f77STsutomu Itoh kfree(root); 143913a8a7c8SYan, Zheng if (ret > 0) 144013a8a7c8SYan, Zheng ret = -ENOENT; 14410f7d52f4SChris Mason return ERR_PTR(ret); 14420f7d52f4SChris Mason } 144313a8a7c8SYan, Zheng 144484234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1445db94535dSChris Mason blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item)); 1446db94535dSChris Mason root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item), 144784234f3aSYan Zheng blocksize, generation); 14485d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 144979787eaaSJeff Mahoney BUG_ON(!root->node); /* -ENOMEM */ 145013a8a7c8SYan, Zheng out: 145108fe4db1SLi Zefan if (location->objectid != BTRFS_TREE_LOG_OBJECTID) { 14520f7d52f4SChris Mason root->ref_cows = 1; 145308fe4db1SLi Zefan btrfs_check_and_init_root_item(&root->root_item); 145408fe4db1SLi Zefan } 145513a8a7c8SYan, Zheng 14565eda7b5eSChris Mason return root; 14575eda7b5eSChris Mason } 14585eda7b5eSChris Mason 1459edbd8d4eSChris Mason struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info, 1460edbd8d4eSChris Mason struct btrfs_key *location) 14615eda7b5eSChris Mason { 14625eda7b5eSChris Mason struct btrfs_root *root; 14635eda7b5eSChris Mason int ret; 14645eda7b5eSChris Mason 1465edbd8d4eSChris Mason if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 1466edbd8d4eSChris Mason return fs_info->tree_root; 1467edbd8d4eSChris Mason if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID) 1468edbd8d4eSChris Mason return fs_info->extent_root; 14698f18cf13SChris Mason if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID) 14708f18cf13SChris Mason return fs_info->chunk_root; 14718f18cf13SChris Mason if (location->objectid == BTRFS_DEV_TREE_OBJECTID) 14728f18cf13SChris Mason return fs_info->dev_root; 14730403e47eSYan Zheng if (location->objectid == BTRFS_CSUM_TREE_OBJECTID) 14740403e47eSYan Zheng return fs_info->csum_root; 1475bcef60f2SArne Jansen if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID) 1476bcef60f2SArne Jansen return fs_info->quota_root ? fs_info->quota_root : 1477bcef60f2SArne Jansen ERR_PTR(-ENOENT); 14784df27c4dSYan, Zheng again: 14794df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 14805eda7b5eSChris Mason root = radix_tree_lookup(&fs_info->fs_roots_radix, 14815eda7b5eSChris Mason (unsigned long)location->objectid); 14824df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 14835eda7b5eSChris Mason if (root) 14845eda7b5eSChris Mason return root; 14855eda7b5eSChris Mason 1486e02119d5SChris Mason root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location); 14875eda7b5eSChris Mason if (IS_ERR(root)) 14885eda7b5eSChris Mason return root; 14893394e160SChris Mason 1490581bb050SLi Zefan root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); 1491581bb050SLi Zefan root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), 1492581bb050SLi Zefan GFP_NOFS); 149335a30d7cSDavid Sterba if (!root->free_ino_pinned || !root->free_ino_ctl) { 149435a30d7cSDavid Sterba ret = -ENOMEM; 1495581bb050SLi Zefan goto fail; 149635a30d7cSDavid Sterba } 1497581bb050SLi Zefan 1498581bb050SLi Zefan btrfs_init_free_ino_ctl(root); 1499581bb050SLi Zefan mutex_init(&root->fs_commit_mutex); 1500581bb050SLi Zefan spin_lock_init(&root->cache_lock); 1501581bb050SLi Zefan init_waitqueue_head(&root->cache_wait); 1502581bb050SLi Zefan 15030ee5dc67SAl Viro ret = get_anon_bdev(&root->anon_dev); 1504ac08aedfSChris Mason if (ret) 1505ac08aedfSChris Mason goto fail; 15063394e160SChris Mason 1507d68fc57bSYan, Zheng if (btrfs_root_refs(&root->root_item) == 0) { 1508d68fc57bSYan, Zheng ret = -ENOENT; 1509d68fc57bSYan, Zheng goto fail; 1510d68fc57bSYan, Zheng } 1511d68fc57bSYan, Zheng 1512d68fc57bSYan, Zheng ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid); 1513d68fc57bSYan, Zheng if (ret < 0) 1514d68fc57bSYan, Zheng goto fail; 1515d68fc57bSYan, Zheng if (ret == 0) 1516d68fc57bSYan, Zheng root->orphan_item_inserted = 1; 1517d68fc57bSYan, Zheng 15184df27c4dSYan, Zheng ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); 15194df27c4dSYan, Zheng if (ret) 15204df27c4dSYan, Zheng goto fail; 15214df27c4dSYan, Zheng 15224df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 15232619ba1fSChris Mason ret = radix_tree_insert(&fs_info->fs_roots_radix, 15242619ba1fSChris Mason (unsigned long)root->root_key.objectid, 15250f7d52f4SChris Mason root); 1526d68fc57bSYan, Zheng if (ret == 0) 15274df27c4dSYan, Zheng root->in_radix = 1; 1528d68fc57bSYan, Zheng 15294df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 15304df27c4dSYan, Zheng radix_tree_preload_end(); 15310f7d52f4SChris Mason if (ret) { 15324df27c4dSYan, Zheng if (ret == -EEXIST) { 15334df27c4dSYan, Zheng free_fs_root(root); 15344df27c4dSYan, Zheng goto again; 15350f7d52f4SChris Mason } 15364df27c4dSYan, Zheng goto fail; 15374df27c4dSYan, Zheng } 15384df27c4dSYan, Zheng 1539edbd8d4eSChris Mason ret = btrfs_find_dead_roots(fs_info->tree_root, 15405d4f98a2SYan Zheng root->root_key.objectid); 15414df27c4dSYan, Zheng WARN_ON(ret); 1542edbd8d4eSChris Mason return root; 15434df27c4dSYan, Zheng fail: 15444df27c4dSYan, Zheng free_fs_root(root); 15454df27c4dSYan, Zheng return ERR_PTR(ret); 1546edbd8d4eSChris Mason } 1547edbd8d4eSChris Mason 154804160088SChris Mason static int btrfs_congested_fn(void *congested_data, int bdi_bits) 154904160088SChris Mason { 155004160088SChris Mason struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; 155104160088SChris Mason int ret = 0; 155204160088SChris Mason struct btrfs_device *device; 155304160088SChris Mason struct backing_dev_info *bdi; 1554b7967db7SChris Mason 15551f78160cSXiao Guangrong rcu_read_lock(); 15561f78160cSXiao Guangrong list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) { 1557dfe25020SChris Mason if (!device->bdev) 1558dfe25020SChris Mason continue; 155904160088SChris Mason bdi = blk_get_backing_dev_info(device->bdev); 156004160088SChris Mason if (bdi && bdi_congested(bdi, bdi_bits)) { 156104160088SChris Mason ret = 1; 156204160088SChris Mason break; 156304160088SChris Mason } 156404160088SChris Mason } 15651f78160cSXiao Guangrong rcu_read_unlock(); 156604160088SChris Mason return ret; 156704160088SChris Mason } 156804160088SChris Mason 156938b66988SChris Mason /* 1570ad081f14SJens Axboe * If this fails, caller must call bdi_destroy() to get rid of the 1571ad081f14SJens Axboe * bdi again. 1572ad081f14SJens Axboe */ 157304160088SChris Mason static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi) 157404160088SChris Mason { 1575ad081f14SJens Axboe int err; 1576ad081f14SJens Axboe 1577ad081f14SJens Axboe bdi->capabilities = BDI_CAP_MAP_COPY; 1578e6d086d8SJens Axboe err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY); 1579ad081f14SJens Axboe if (err) 1580ad081f14SJens Axboe return err; 1581ad081f14SJens Axboe 15824575c9ccSChris Mason bdi->ra_pages = default_backing_dev_info.ra_pages; 158304160088SChris Mason bdi->congested_fn = btrfs_congested_fn; 158404160088SChris Mason bdi->congested_data = info; 158504160088SChris Mason return 0; 158604160088SChris Mason } 158704160088SChris Mason 15888b712842SChris Mason /* 15898b712842SChris Mason * called by the kthread helper functions to finally call the bio end_io 15908b712842SChris Mason * functions. This is where read checksum verification actually happens 15918b712842SChris Mason */ 15928b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work) 1593ce9adaa5SChris Mason { 1594ce9adaa5SChris Mason struct bio *bio; 15958b712842SChris Mason struct end_io_wq *end_io_wq; 15968b712842SChris Mason struct btrfs_fs_info *fs_info; 1597ce9adaa5SChris Mason int error; 1598ce9adaa5SChris Mason 15998b712842SChris Mason end_io_wq = container_of(work, struct end_io_wq, work); 1600ce9adaa5SChris Mason bio = end_io_wq->bio; 16018b712842SChris Mason fs_info = end_io_wq->info; 16028b712842SChris Mason 1603ce9adaa5SChris Mason error = end_io_wq->error; 1604ce9adaa5SChris Mason bio->bi_private = end_io_wq->private; 1605ce9adaa5SChris Mason bio->bi_end_io = end_io_wq->end_io; 1606ce9adaa5SChris Mason kfree(end_io_wq); 1607ce9adaa5SChris Mason bio_endio(bio, error); 1608ce9adaa5SChris Mason } 160944b8bd7eSChris Mason 1610a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1611a74a4b97SChris Mason { 1612a74a4b97SChris Mason struct btrfs_root *root = arg; 1613a74a4b97SChris Mason 1614a74a4b97SChris Mason do { 1615a74a4b97SChris Mason vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE); 161676dda93cSYan, Zheng 161776dda93cSYan, Zheng if (!(root->fs_info->sb->s_flags & MS_RDONLY) && 161876dda93cSYan, Zheng mutex_trylock(&root->fs_info->cleaner_mutex)) { 161924bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 1620a74a4b97SChris Mason btrfs_clean_old_snapshots(root); 1621a74a4b97SChris Mason mutex_unlock(&root->fs_info->cleaner_mutex); 16224cb5300bSChris Mason btrfs_run_defrag_inodes(root->fs_info); 162376dda93cSYan, Zheng } 1624a74a4b97SChris Mason 1625a0acae0eSTejun Heo if (!try_to_freeze()) { 1626a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 16278929ecfaSYan, Zheng if (!kthread_should_stop()) 1628a74a4b97SChris Mason schedule(); 1629a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1630a74a4b97SChris Mason } 1631a74a4b97SChris Mason } while (!kthread_should_stop()); 1632a74a4b97SChris Mason return 0; 1633a74a4b97SChris Mason } 1634a74a4b97SChris Mason 1635a74a4b97SChris Mason static int transaction_kthread(void *arg) 1636a74a4b97SChris Mason { 1637a74a4b97SChris Mason struct btrfs_root *root = arg; 1638a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1639a74a4b97SChris Mason struct btrfs_transaction *cur; 16408929ecfaSYan, Zheng u64 transid; 1641a74a4b97SChris Mason unsigned long now; 1642a74a4b97SChris Mason unsigned long delay; 1643914b2007SJan Kara bool cannot_commit; 1644a74a4b97SChris Mason 1645a74a4b97SChris Mason do { 1646914b2007SJan Kara cannot_commit = false; 1647a74a4b97SChris Mason delay = HZ * 30; 1648a74a4b97SChris Mason vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE); 1649a74a4b97SChris Mason mutex_lock(&root->fs_info->transaction_kthread_mutex); 1650a74a4b97SChris Mason 1651a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 1652a74a4b97SChris Mason cur = root->fs_info->running_transaction; 1653a74a4b97SChris Mason if (!cur) { 1654a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1655a74a4b97SChris Mason goto sleep; 1656a74a4b97SChris Mason } 165731153d81SYan Zheng 1658a74a4b97SChris Mason now = get_seconds(); 16598929ecfaSYan, Zheng if (!cur->blocked && 16608929ecfaSYan, Zheng (now < cur->start_time || now - cur->start_time < 30)) { 1661a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1662a74a4b97SChris Mason delay = HZ * 5; 1663a74a4b97SChris Mason goto sleep; 1664a74a4b97SChris Mason } 16658929ecfaSYan, Zheng transid = cur->transid; 1666a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 166756bec294SChris Mason 166879787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 16697a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 1670914b2007SJan Kara if (IS_ERR(trans)) { 1671914b2007SJan Kara cannot_commit = true; 167279787eaaSJeff Mahoney goto sleep; 1673914b2007SJan Kara } 16748929ecfaSYan, Zheng if (transid == trans->transid) { 167579787eaaSJeff Mahoney btrfs_commit_transaction(trans, root); 16768929ecfaSYan, Zheng } else { 16778929ecfaSYan, Zheng btrfs_end_transaction(trans, root); 16788929ecfaSYan, Zheng } 1679a74a4b97SChris Mason sleep: 1680a74a4b97SChris Mason wake_up_process(root->fs_info->cleaner_kthread); 1681a74a4b97SChris Mason mutex_unlock(&root->fs_info->transaction_kthread_mutex); 1682a74a4b97SChris Mason 1683a0acae0eSTejun Heo if (!try_to_freeze()) { 1684a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 16858929ecfaSYan, Zheng if (!kthread_should_stop() && 1686914b2007SJan Kara (!btrfs_transaction_blocked(root->fs_info) || 1687914b2007SJan Kara cannot_commit)) 1688a74a4b97SChris Mason schedule_timeout(delay); 1689a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1690a74a4b97SChris Mason } 1691a74a4b97SChris Mason } while (!kthread_should_stop()); 1692a74a4b97SChris Mason return 0; 1693a74a4b97SChris Mason } 1694a74a4b97SChris Mason 1695af31f5e5SChris Mason /* 1696af31f5e5SChris Mason * this will find the highest generation in the array of 1697af31f5e5SChris Mason * root backups. The index of the highest array is returned, 1698af31f5e5SChris Mason * or -1 if we can't find anything. 1699af31f5e5SChris Mason * 1700af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1701af31f5e5SChris Mason * generation of the latest root in the array with the generation 1702af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1703af31f5e5SChris Mason */ 1704af31f5e5SChris Mason static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen) 1705af31f5e5SChris Mason { 1706af31f5e5SChris Mason u64 cur; 1707af31f5e5SChris Mason int newest_index = -1; 1708af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1709af31f5e5SChris Mason int i; 1710af31f5e5SChris Mason 1711af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1712af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1713af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1714af31f5e5SChris Mason if (cur == newest_gen) 1715af31f5e5SChris Mason newest_index = i; 1716af31f5e5SChris Mason } 1717af31f5e5SChris Mason 1718af31f5e5SChris Mason /* check to see if we actually wrapped around */ 1719af31f5e5SChris Mason if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) { 1720af31f5e5SChris Mason root_backup = info->super_copy->super_roots; 1721af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1722af31f5e5SChris Mason if (cur == newest_gen) 1723af31f5e5SChris Mason newest_index = 0; 1724af31f5e5SChris Mason } 1725af31f5e5SChris Mason return newest_index; 1726af31f5e5SChris Mason } 1727af31f5e5SChris Mason 1728af31f5e5SChris Mason 1729af31f5e5SChris Mason /* 1730af31f5e5SChris Mason * find the oldest backup so we know where to store new entries 1731af31f5e5SChris Mason * in the backup array. This will set the backup_root_index 1732af31f5e5SChris Mason * field in the fs_info struct 1733af31f5e5SChris Mason */ 1734af31f5e5SChris Mason static void find_oldest_super_backup(struct btrfs_fs_info *info, 1735af31f5e5SChris Mason u64 newest_gen) 1736af31f5e5SChris Mason { 1737af31f5e5SChris Mason int newest_index = -1; 1738af31f5e5SChris Mason 1739af31f5e5SChris Mason newest_index = find_newest_super_backup(info, newest_gen); 1740af31f5e5SChris Mason /* if there was garbage in there, just move along */ 1741af31f5e5SChris Mason if (newest_index == -1) { 1742af31f5e5SChris Mason info->backup_root_index = 0; 1743af31f5e5SChris Mason } else { 1744af31f5e5SChris Mason info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS; 1745af31f5e5SChris Mason } 1746af31f5e5SChris Mason } 1747af31f5e5SChris Mason 1748af31f5e5SChris Mason /* 1749af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1750af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1751af31f5e5SChris Mason * done 1752af31f5e5SChris Mason */ 1753af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1754af31f5e5SChris Mason { 1755af31f5e5SChris Mason int next_backup; 1756af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1757af31f5e5SChris Mason int last_backup; 1758af31f5e5SChris Mason 1759af31f5e5SChris Mason next_backup = info->backup_root_index; 1760af31f5e5SChris Mason last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) % 1761af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 1762af31f5e5SChris Mason 1763af31f5e5SChris Mason /* 1764af31f5e5SChris Mason * just overwrite the last backup if we're at the same generation 1765af31f5e5SChris Mason * this happens only at umount 1766af31f5e5SChris Mason */ 1767af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + last_backup; 1768af31f5e5SChris Mason if (btrfs_backup_tree_root_gen(root_backup) == 1769af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)) 1770af31f5e5SChris Mason next_backup = last_backup; 1771af31f5e5SChris Mason 1772af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1773af31f5e5SChris Mason 1774af31f5e5SChris Mason /* 1775af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1776af31f5e5SChris Mason * regardless of which ones we use today 1777af31f5e5SChris Mason */ 1778af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1779af31f5e5SChris Mason 1780af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1781af31f5e5SChris Mason 1782af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1783af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1784af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1785af31f5e5SChris Mason 1786af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1787af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1788af31f5e5SChris Mason 1789af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1790af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1791af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 1792af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 1793af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 1794af31f5e5SChris Mason 1795af31f5e5SChris Mason btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); 1796af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 1797af31f5e5SChris Mason btrfs_header_generation(info->extent_root->node)); 1798af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 1799af31f5e5SChris Mason btrfs_header_level(info->extent_root->node)); 1800af31f5e5SChris Mason 18017c7e82a7SChris Mason /* 18027c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 18037c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 18047c7e82a7SChris Mason */ 18057c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 18067c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 18077c7e82a7SChris Mason info->fs_root->node->start); 1808af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 1809af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 1810af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 1811af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 18127c7e82a7SChris Mason } 1813af31f5e5SChris Mason 1814af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 1815af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 1816af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 1817af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 1818af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 1819af31f5e5SChris Mason 1820af31f5e5SChris Mason btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); 1821af31f5e5SChris Mason btrfs_set_backup_csum_root_gen(root_backup, 1822af31f5e5SChris Mason btrfs_header_generation(info->csum_root->node)); 1823af31f5e5SChris Mason btrfs_set_backup_csum_root_level(root_backup, 1824af31f5e5SChris Mason btrfs_header_level(info->csum_root->node)); 1825af31f5e5SChris Mason 1826af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 1827af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 1828af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 1829af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 1830af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 1831af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 1832af31f5e5SChris Mason 1833af31f5e5SChris Mason /* 1834af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 1835af31f5e5SChris Mason * for the next commit 1836af31f5e5SChris Mason */ 1837af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 1838af31f5e5SChris Mason &info->super_for_commit->super_roots, 1839af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 1840af31f5e5SChris Mason } 1841af31f5e5SChris Mason 1842af31f5e5SChris Mason /* 1843af31f5e5SChris Mason * this copies info out of the root backup array and back into 1844af31f5e5SChris Mason * the in-memory super block. It is meant to help iterate through 1845af31f5e5SChris Mason * the array, so you send it the number of backups you've already 1846af31f5e5SChris Mason * tried and the last backup index you used. 1847af31f5e5SChris Mason * 1848af31f5e5SChris Mason * this returns -1 when it has tried all the backups 1849af31f5e5SChris Mason */ 1850af31f5e5SChris Mason static noinline int next_root_backup(struct btrfs_fs_info *info, 1851af31f5e5SChris Mason struct btrfs_super_block *super, 1852af31f5e5SChris Mason int *num_backups_tried, int *backup_index) 1853af31f5e5SChris Mason { 1854af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1855af31f5e5SChris Mason int newest = *backup_index; 1856af31f5e5SChris Mason 1857af31f5e5SChris Mason if (*num_backups_tried == 0) { 1858af31f5e5SChris Mason u64 gen = btrfs_super_generation(super); 1859af31f5e5SChris Mason 1860af31f5e5SChris Mason newest = find_newest_super_backup(info, gen); 1861af31f5e5SChris Mason if (newest == -1) 1862af31f5e5SChris Mason return -1; 1863af31f5e5SChris Mason 1864af31f5e5SChris Mason *backup_index = newest; 1865af31f5e5SChris Mason *num_backups_tried = 1; 1866af31f5e5SChris Mason } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) { 1867af31f5e5SChris Mason /* we've tried all the backups, all done */ 1868af31f5e5SChris Mason return -1; 1869af31f5e5SChris Mason } else { 1870af31f5e5SChris Mason /* jump to the next oldest backup */ 1871af31f5e5SChris Mason newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) % 1872af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 1873af31f5e5SChris Mason *backup_index = newest; 1874af31f5e5SChris Mason *num_backups_tried += 1; 1875af31f5e5SChris Mason } 1876af31f5e5SChris Mason root_backup = super->super_roots + newest; 1877af31f5e5SChris Mason 1878af31f5e5SChris Mason btrfs_set_super_generation(super, 1879af31f5e5SChris Mason btrfs_backup_tree_root_gen(root_backup)); 1880af31f5e5SChris Mason btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 1881af31f5e5SChris Mason btrfs_set_super_root_level(super, 1882af31f5e5SChris Mason btrfs_backup_tree_root_level(root_backup)); 1883af31f5e5SChris Mason btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 1884af31f5e5SChris Mason 1885af31f5e5SChris Mason /* 1886af31f5e5SChris Mason * fixme: the total bytes and num_devices need to match or we should 1887af31f5e5SChris Mason * need a fsck 1888af31f5e5SChris Mason */ 1889af31f5e5SChris Mason btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 1890af31f5e5SChris Mason btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 1891af31f5e5SChris Mason return 0; 1892af31f5e5SChris Mason } 1893af31f5e5SChris Mason 1894af31f5e5SChris Mason /* helper to cleanup tree roots */ 1895af31f5e5SChris Mason static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root) 1896af31f5e5SChris Mason { 1897af31f5e5SChris Mason free_extent_buffer(info->tree_root->node); 1898af31f5e5SChris Mason free_extent_buffer(info->tree_root->commit_root); 1899af31f5e5SChris Mason free_extent_buffer(info->dev_root->node); 1900af31f5e5SChris Mason free_extent_buffer(info->dev_root->commit_root); 1901af31f5e5SChris Mason free_extent_buffer(info->extent_root->node); 1902af31f5e5SChris Mason free_extent_buffer(info->extent_root->commit_root); 1903af31f5e5SChris Mason free_extent_buffer(info->csum_root->node); 1904af31f5e5SChris Mason free_extent_buffer(info->csum_root->commit_root); 1905bcef60f2SArne Jansen if (info->quota_root) { 1906bcef60f2SArne Jansen free_extent_buffer(info->quota_root->node); 1907bcef60f2SArne Jansen free_extent_buffer(info->quota_root->commit_root); 1908bcef60f2SArne Jansen } 1909af31f5e5SChris Mason 1910af31f5e5SChris Mason info->tree_root->node = NULL; 1911af31f5e5SChris Mason info->tree_root->commit_root = NULL; 1912af31f5e5SChris Mason info->dev_root->node = NULL; 1913af31f5e5SChris Mason info->dev_root->commit_root = NULL; 1914af31f5e5SChris Mason info->extent_root->node = NULL; 1915af31f5e5SChris Mason info->extent_root->commit_root = NULL; 1916af31f5e5SChris Mason info->csum_root->node = NULL; 1917af31f5e5SChris Mason info->csum_root->commit_root = NULL; 1918bcef60f2SArne Jansen if (info->quota_root) { 1919bcef60f2SArne Jansen info->quota_root->node = NULL; 1920bcef60f2SArne Jansen info->quota_root->commit_root = NULL; 1921bcef60f2SArne Jansen } 1922af31f5e5SChris Mason 1923af31f5e5SChris Mason if (chunk_root) { 1924af31f5e5SChris Mason free_extent_buffer(info->chunk_root->node); 1925af31f5e5SChris Mason free_extent_buffer(info->chunk_root->commit_root); 1926af31f5e5SChris Mason info->chunk_root->node = NULL; 1927af31f5e5SChris Mason info->chunk_root->commit_root = NULL; 1928af31f5e5SChris Mason } 1929af31f5e5SChris Mason } 1930af31f5e5SChris Mason 1931af31f5e5SChris Mason 1932ad2b2c80SAl Viro int open_ctree(struct super_block *sb, 1933dfe25020SChris Mason struct btrfs_fs_devices *fs_devices, 1934dfe25020SChris Mason char *options) 1935eb60ceacSChris Mason { 1936db94535dSChris Mason u32 sectorsize; 1937db94535dSChris Mason u32 nodesize; 1938db94535dSChris Mason u32 leafsize; 1939db94535dSChris Mason u32 blocksize; 194087ee04ebSChris Mason u32 stripesize; 194184234f3aSYan Zheng u64 generation; 1942f2b636e8SJosef Bacik u64 features; 19433de4586cSChris Mason struct btrfs_key location; 1944a061fc8dSChris Mason struct buffer_head *bh; 19454d34b278SIlya Dryomov struct btrfs_super_block *disk_super; 1946815745cfSAl Viro struct btrfs_fs_info *fs_info = btrfs_sb(sb); 1947f84a8bd6SAl Viro struct btrfs_root *tree_root; 19484d34b278SIlya Dryomov struct btrfs_root *extent_root; 19494d34b278SIlya Dryomov struct btrfs_root *csum_root; 19504d34b278SIlya Dryomov struct btrfs_root *chunk_root; 19514d34b278SIlya Dryomov struct btrfs_root *dev_root; 1952bcef60f2SArne Jansen struct btrfs_root *quota_root; 1953e02119d5SChris Mason struct btrfs_root *log_tree_root; 1954eb60ceacSChris Mason int ret; 1955e58ca020SYan int err = -EINVAL; 1956af31f5e5SChris Mason int num_backups_tried = 0; 1957af31f5e5SChris Mason int backup_index = 0; 19584543df7eSChris Mason 1959f84a8bd6SAl Viro tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info); 19606f07e42eSAl Viro extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info); 19616f07e42eSAl Viro csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info); 19626f07e42eSAl Viro chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info); 19636f07e42eSAl Viro dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info); 1964bcef60f2SArne Jansen quota_root = fs_info->quota_root = btrfs_alloc_root(fs_info); 19658790d502SChris Mason 1966f84a8bd6SAl Viro if (!tree_root || !extent_root || !csum_root || 1967bcef60f2SArne Jansen !chunk_root || !dev_root || !quota_root) { 196839279cc3SChris Mason err = -ENOMEM; 196939279cc3SChris Mason goto fail; 197039279cc3SChris Mason } 197176dda93cSYan, Zheng 197276dda93cSYan, Zheng ret = init_srcu_struct(&fs_info->subvol_srcu); 197376dda93cSYan, Zheng if (ret) { 197476dda93cSYan, Zheng err = ret; 197576dda93cSYan, Zheng goto fail; 197676dda93cSYan, Zheng } 197776dda93cSYan, Zheng 197876dda93cSYan, Zheng ret = setup_bdi(fs_info, &fs_info->bdi); 197976dda93cSYan, Zheng if (ret) { 198076dda93cSYan, Zheng err = ret; 198176dda93cSYan, Zheng goto fail_srcu; 198276dda93cSYan, Zheng } 198376dda93cSYan, Zheng 198476dda93cSYan, Zheng fs_info->btree_inode = new_inode(sb); 198576dda93cSYan, Zheng if (!fs_info->btree_inode) { 198676dda93cSYan, Zheng err = -ENOMEM; 198776dda93cSYan, Zheng goto fail_bdi; 198876dda93cSYan, Zheng } 198976dda93cSYan, Zheng 1990a6591715SChris Mason mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 19911561dedaSMiao Xie 199276dda93cSYan, Zheng INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 19938fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 1994facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 199524bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 199619c00ddcSChris Mason INIT_LIST_HEAD(&fs_info->hashers); 1997ea8c2819SChris Mason INIT_LIST_HEAD(&fs_info->delalloc_inodes); 19985a3f23d5SChris Mason INIT_LIST_HEAD(&fs_info->ordered_operations); 199911833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 20001832a6d5SChris Mason spin_lock_init(&fs_info->delalloc_lock); 2001a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 200231153d81SYan Zheng spin_lock_init(&fs_info->ref_cache_lock); 200376dda93cSYan, Zheng spin_lock_init(&fs_info->fs_roots_radix_lock); 200424bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 20054cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 20062bf64758SJosef Bacik spin_lock_init(&fs_info->free_chunk_lock); 2007f29021b2SJan Schmidt spin_lock_init(&fs_info->tree_mod_seq_lock); 2008f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 20097585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 201019c00ddcSChris Mason 201158176a96SJosef Bacik init_completion(&fs_info->kobj_unregister); 20120b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 20136324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 2014f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 20150b86a832SChris Mason btrfs_mapping_init(&fs_info->mapping_tree); 2016f0486c68SYan, Zheng btrfs_init_block_rsv(&fs_info->global_block_rsv); 2017f0486c68SYan, Zheng btrfs_init_block_rsv(&fs_info->delalloc_block_rsv); 2018f0486c68SYan, Zheng btrfs_init_block_rsv(&fs_info->trans_block_rsv); 2019f0486c68SYan, Zheng btrfs_init_block_rsv(&fs_info->chunk_block_rsv); 2020f0486c68SYan, Zheng btrfs_init_block_rsv(&fs_info->empty_block_rsv); 20216d668ddaSJosef Bacik btrfs_init_block_rsv(&fs_info->delayed_block_rsv); 2022cb03c743SChris Mason atomic_set(&fs_info->nr_async_submits, 0); 2023771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 20248c8bee1dSChris Mason atomic_set(&fs_info->async_submit_draining, 0); 20250986fe9eSChris Mason atomic_set(&fs_info->nr_async_bios, 0); 20264cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 2027f29021b2SJan Schmidt atomic_set(&fs_info->tree_mod_seq, 0); 2028e20d96d6SChris Mason fs_info->sb = sb; 20296f568d35SChris Mason fs_info->max_inline = 8192 * 1024; 20309ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 20314cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 2032a4abeea4SJosef Bacik fs_info->trans_no_join = 0; 20332bf64758SJosef Bacik fs_info->free_chunk_space = 0; 2034f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 2035c8b97818SChris Mason 2036097b8a7cSJan Schmidt init_waitqueue_head(&fs_info->tree_mod_seq_wait); 2037097b8a7cSJan Schmidt 203890519d66SArne Jansen /* readahead state */ 203990519d66SArne Jansen INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT); 204090519d66SArne Jansen spin_lock_init(&fs_info->reada_lock); 20412c90e5d6SChris Mason 2042b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 2043b34b086cSChris Mason num_online_cpus() + 2, 8); 20440afbaf8cSChris Mason 20453eaa2885SChris Mason INIT_LIST_HEAD(&fs_info->ordered_extents); 20463eaa2885SChris Mason spin_lock_init(&fs_info->ordered_extent_lock); 204716cdcec7SMiao Xie fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 204816cdcec7SMiao Xie GFP_NOFS); 204916cdcec7SMiao Xie if (!fs_info->delayed_root) { 205016cdcec7SMiao Xie err = -ENOMEM; 205116cdcec7SMiao Xie goto fail_iput; 205216cdcec7SMiao Xie } 205316cdcec7SMiao Xie btrfs_init_delayed_root(fs_info->delayed_root); 20543eaa2885SChris Mason 2055a2de733cSArne Jansen mutex_init(&fs_info->scrub_lock); 2056a2de733cSArne Jansen atomic_set(&fs_info->scrubs_running, 0); 2057a2de733cSArne Jansen atomic_set(&fs_info->scrub_pause_req, 0); 2058a2de733cSArne Jansen atomic_set(&fs_info->scrubs_paused, 0); 2059a2de733cSArne Jansen atomic_set(&fs_info->scrub_cancel_req, 0); 2060a2de733cSArne Jansen init_waitqueue_head(&fs_info->scrub_pause_wait); 2061a2de733cSArne Jansen init_rwsem(&fs_info->scrub_super_lock); 2062a2de733cSArne Jansen fs_info->scrub_workers_refcnt = 0; 206321adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 206421adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 206521adbd5cSStefan Behrens #endif 2066a2de733cSArne Jansen 2067c9e9f97bSIlya Dryomov spin_lock_init(&fs_info->balance_lock); 2068c9e9f97bSIlya Dryomov mutex_init(&fs_info->balance_mutex); 2069837d5b6eSIlya Dryomov atomic_set(&fs_info->balance_running, 0); 2070837d5b6eSIlya Dryomov atomic_set(&fs_info->balance_pause_req, 0); 2071a7e99c69SIlya Dryomov atomic_set(&fs_info->balance_cancel_req, 0); 2072c9e9f97bSIlya Dryomov fs_info->balance_ctl = NULL; 2073837d5b6eSIlya Dryomov init_waitqueue_head(&fs_info->balance_wait_q); 2074a061fc8dSChris Mason 20750afbaf8cSChris Mason sb->s_blocksize = 4096; 20760afbaf8cSChris Mason sb->s_blocksize_bits = blksize_bits(4096); 207732a88aa1SJens Axboe sb->s_bdi = &fs_info->bdi; 20780afbaf8cSChris Mason 207976dda93cSYan, Zheng fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 2080bfe86848SMiklos Szeredi set_nlink(fs_info->btree_inode, 1); 20810afbaf8cSChris Mason /* 20820afbaf8cSChris Mason * we set the i_size on the btree inode to the max possible int. 20830afbaf8cSChris Mason * the real end of the address space is determined by all of 20840afbaf8cSChris Mason * the devices in the system 20850afbaf8cSChris Mason */ 20860afbaf8cSChris Mason fs_info->btree_inode->i_size = OFFSET_MAX; 2087d98237b3SChris Mason fs_info->btree_inode->i_mapping->a_ops = &btree_aops; 208804160088SChris Mason fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi; 208904160088SChris Mason 20905d4f98a2SYan Zheng RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node); 2091d1310b2eSChris Mason extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree, 2092f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 20930b32f4bbSJosef Bacik BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0; 2094a8067e02SDavid Sterba extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree); 20950da5468fSChris Mason 2096d1310b2eSChris Mason BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops; 2097d1310b2eSChris Mason 20980f7d52f4SChris Mason BTRFS_I(fs_info->btree_inode)->root = tree_root; 20990f7d52f4SChris Mason memset(&BTRFS_I(fs_info->btree_inode)->location, 0, 21000f7d52f4SChris Mason sizeof(struct btrfs_key)); 210172ac3c0dSJosef Bacik set_bit(BTRFS_INODE_DUMMY, 210272ac3c0dSJosef Bacik &BTRFS_I(fs_info->btree_inode)->runtime_flags); 210322b0ebdaSChris Mason insert_inode_hash(fs_info->btree_inode); 210439279cc3SChris Mason 2105e02119d5SChris Mason spin_lock_init(&fs_info->block_group_cache_lock); 21066bef4d31SEric Paris fs_info->block_group_cache_tree = RB_ROOT; 2107925baeddSChris Mason 210811833d66SYan Zheng extent_io_tree_init(&fs_info->freed_extents[0], 2109f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 211011833d66SYan Zheng extent_io_tree_init(&fs_info->freed_extents[1], 2111f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 211211833d66SYan Zheng fs_info->pinned_extents = &fs_info->freed_extents[0]; 2113e6dcd2dcSChris Mason fs_info->do_barriers = 1; 2114f9295749SChris Mason 2115d98237b3SChris Mason 21165a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 211730ae8467SChris Mason mutex_init(&fs_info->tree_log_mutex); 2118925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 2119925baeddSChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 2120a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 2121a74a4b97SChris Mason mutex_init(&fs_info->volume_mutex); 2122276e680dSYan Zheng init_rwsem(&fs_info->extent_commit_sem); 2123c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 212476dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 2125fa9c0d79SChris Mason 2126416ac51dSArne Jansen spin_lock_init(&fs_info->qgroup_lock); 2127416ac51dSArne Jansen fs_info->qgroup_tree = RB_ROOT; 2128416ac51dSArne Jansen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2129416ac51dSArne Jansen fs_info->qgroup_seq = 1; 2130416ac51dSArne Jansen fs_info->quota_enabled = 0; 2131416ac51dSArne Jansen fs_info->pending_quota_state = 0; 2132416ac51dSArne Jansen 2133fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2134fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2135fa9c0d79SChris Mason 21367d9eb12cSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 21373768f368SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 2138bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 21394854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 21409a8dd150SChris Mason 21410b86a832SChris Mason __setup_root(4096, 4096, 4096, 4096, tree_root, 21422c90e5d6SChris Mason fs_info, BTRFS_ROOT_TREE_OBJECTID); 21437eccb903SChris Mason 21443c4bb26bSChris Mason invalidate_bdev(fs_devices->latest_bdev); 2145a512bbf8SYan Zheng bh = btrfs_read_dev_super(fs_devices->latest_bdev); 214620b45077SDave Young if (!bh) { 214720b45077SDave Young err = -EINVAL; 214816cdcec7SMiao Xie goto fail_alloc; 214920b45077SDave Young } 215039279cc3SChris Mason 21516c41761fSDavid Sterba memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy)); 21526c41761fSDavid Sterba memcpy(fs_info->super_for_commit, fs_info->super_copy, 21536c41761fSDavid Sterba sizeof(*fs_info->super_for_commit)); 2154a061fc8dSChris Mason brelse(bh); 21555f39d397SChris Mason 21566c41761fSDavid Sterba memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE); 21570b86a832SChris Mason 21586c41761fSDavid Sterba disk_super = fs_info->super_copy; 21590f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 216016cdcec7SMiao Xie goto fail_alloc; 21610f7d52f4SChris Mason 2162acce952bSliubo /* check FS state, whether FS is broken. */ 2163acce952bSliubo fs_info->fs_state |= btrfs_super_flags(disk_super); 2164acce952bSliubo 2165fcd1f065SDavid Sterba ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY); 2166fcd1f065SDavid Sterba if (ret) { 2167fcd1f065SDavid Sterba printk(KERN_ERR "btrfs: superblock contains fatal errors\n"); 2168fcd1f065SDavid Sterba err = ret; 2169fcd1f065SDavid Sterba goto fail_alloc; 2170fcd1f065SDavid Sterba } 2171acce952bSliubo 217275e7cb7fSLiu Bo /* 2173af31f5e5SChris Mason * run through our array of backup supers and setup 2174af31f5e5SChris Mason * our ring pointer to the oldest one 2175af31f5e5SChris Mason */ 2176af31f5e5SChris Mason generation = btrfs_super_generation(disk_super); 2177af31f5e5SChris Mason find_oldest_super_backup(fs_info, generation); 2178af31f5e5SChris Mason 2179af31f5e5SChris Mason /* 218075e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 218175e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 218275e7cb7fSLiu Bo */ 218375e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 218475e7cb7fSLiu Bo 21852b82032cSYan Zheng ret = btrfs_parse_options(tree_root, options); 21862b82032cSYan Zheng if (ret) { 21872b82032cSYan Zheng err = ret; 218816cdcec7SMiao Xie goto fail_alloc; 21892b82032cSYan Zheng } 2190dfe25020SChris Mason 2191f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 2192f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 2193f2b636e8SJosef Bacik if (features) { 2194f2b636e8SJosef Bacik printk(KERN_ERR "BTRFS: couldn't mount because of " 2195f2b636e8SJosef Bacik "unsupported optional features (%Lx).\n", 219621380931SJoel Becker (unsigned long long)features); 2197f2b636e8SJosef Bacik err = -EINVAL; 219816cdcec7SMiao Xie goto fail_alloc; 2199f2b636e8SJosef Bacik } 2200f2b636e8SJosef Bacik 2201727011e0SChris Mason if (btrfs_super_leafsize(disk_super) != 2202727011e0SChris Mason btrfs_super_nodesize(disk_super)) { 2203727011e0SChris Mason printk(KERN_ERR "BTRFS: couldn't mount because metadata " 2204727011e0SChris Mason "blocksizes don't match. node %d leaf %d\n", 2205727011e0SChris Mason btrfs_super_nodesize(disk_super), 2206727011e0SChris Mason btrfs_super_leafsize(disk_super)); 2207727011e0SChris Mason err = -EINVAL; 2208727011e0SChris Mason goto fail_alloc; 2209727011e0SChris Mason } 2210727011e0SChris Mason if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) { 2211727011e0SChris Mason printk(KERN_ERR "BTRFS: couldn't mount because metadata " 2212727011e0SChris Mason "blocksize (%d) was too large\n", 2213727011e0SChris Mason btrfs_super_leafsize(disk_super)); 2214727011e0SChris Mason err = -EINVAL; 2215727011e0SChris Mason goto fail_alloc; 2216727011e0SChris Mason } 2217727011e0SChris Mason 22185d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 22195d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 222069e380d1SLi Zefan if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO) 2221a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 2222727011e0SChris Mason 2223727011e0SChris Mason /* 2224727011e0SChris Mason * flag our filesystem as having big metadata blocks if 2225727011e0SChris Mason * they are bigger than the page size 2226727011e0SChris Mason */ 2227727011e0SChris Mason if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) { 2228727011e0SChris Mason if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 2229727011e0SChris Mason printk(KERN_INFO "btrfs flagging fs with big metadata feature\n"); 2230727011e0SChris Mason features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 2231727011e0SChris Mason } 2232727011e0SChris Mason 2233bc3f116fSChris Mason nodesize = btrfs_super_nodesize(disk_super); 2234bc3f116fSChris Mason leafsize = btrfs_super_leafsize(disk_super); 2235bc3f116fSChris Mason sectorsize = btrfs_super_sectorsize(disk_super); 2236bc3f116fSChris Mason stripesize = btrfs_super_stripesize(disk_super); 2237bc3f116fSChris Mason 2238bc3f116fSChris Mason /* 2239bc3f116fSChris Mason * mixed block groups end up with duplicate but slightly offset 2240bc3f116fSChris Mason * extent buffers for the same range. It leads to corruptions 2241bc3f116fSChris Mason */ 2242bc3f116fSChris Mason if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 2243bc3f116fSChris Mason (sectorsize != leafsize)) { 2244bc3f116fSChris Mason printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes " 2245bc3f116fSChris Mason "are not allowed for mixed block groups on %s\n", 2246bc3f116fSChris Mason sb->s_id); 2247bc3f116fSChris Mason goto fail_alloc; 2248bc3f116fSChris Mason } 2249bc3f116fSChris Mason 22505d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 22515d4f98a2SYan Zheng 2252f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 2253f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 2254f2b636e8SJosef Bacik if (!(sb->s_flags & MS_RDONLY) && features) { 2255f2b636e8SJosef Bacik printk(KERN_ERR "BTRFS: couldn't mount RDWR because of " 2256f2b636e8SJosef Bacik "unsupported option features (%Lx).\n", 225721380931SJoel Becker (unsigned long long)features); 2258f2b636e8SJosef Bacik err = -EINVAL; 225916cdcec7SMiao Xie goto fail_alloc; 2260f2b636e8SJosef Bacik } 226161d92c32SChris Mason 226261d92c32SChris Mason btrfs_init_workers(&fs_info->generic_worker, 226361d92c32SChris Mason "genwork", 1, NULL); 226461d92c32SChris Mason 22655443be45SChris Mason btrfs_init_workers(&fs_info->workers, "worker", 226661d92c32SChris Mason fs_info->thread_pool_size, 226761d92c32SChris Mason &fs_info->generic_worker); 2268c8b97818SChris Mason 2269771ed689SChris Mason btrfs_init_workers(&fs_info->delalloc_workers, "delalloc", 227061d92c32SChris Mason fs_info->thread_pool_size, 227161d92c32SChris Mason &fs_info->generic_worker); 2272771ed689SChris Mason 22735443be45SChris Mason btrfs_init_workers(&fs_info->submit_workers, "submit", 2274b720d209SChris Mason min_t(u64, fs_devices->num_devices, 227561d92c32SChris Mason fs_info->thread_pool_size), 227661d92c32SChris Mason &fs_info->generic_worker); 227761b49440SChris Mason 2278bab39bf9SJosef Bacik btrfs_init_workers(&fs_info->caching_workers, "cache", 2279bab39bf9SJosef Bacik 2, &fs_info->generic_worker); 2280bab39bf9SJosef Bacik 228161b49440SChris Mason /* a higher idle thresh on the submit workers makes it much more 228261b49440SChris Mason * likely that bios will be send down in a sane order to the 228361b49440SChris Mason * devices 228461b49440SChris Mason */ 228561b49440SChris Mason fs_info->submit_workers.idle_thresh = 64; 228653863232SChris Mason 2287771ed689SChris Mason fs_info->workers.idle_thresh = 16; 22884a69a410SChris Mason fs_info->workers.ordered = 1; 228961b49440SChris Mason 2290771ed689SChris Mason fs_info->delalloc_workers.idle_thresh = 2; 2291771ed689SChris Mason fs_info->delalloc_workers.ordered = 1; 2292771ed689SChris Mason 229361d92c32SChris Mason btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1, 229461d92c32SChris Mason &fs_info->generic_worker); 22955443be45SChris Mason btrfs_init_workers(&fs_info->endio_workers, "endio", 229661d92c32SChris Mason fs_info->thread_pool_size, 229761d92c32SChris Mason &fs_info->generic_worker); 2298d20f7043SChris Mason btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta", 229961d92c32SChris Mason fs_info->thread_pool_size, 230061d92c32SChris Mason &fs_info->generic_worker); 2301cad321adSChris Mason btrfs_init_workers(&fs_info->endio_meta_write_workers, 230261d92c32SChris Mason "endio-meta-write", fs_info->thread_pool_size, 230361d92c32SChris Mason &fs_info->generic_worker); 23045443be45SChris Mason btrfs_init_workers(&fs_info->endio_write_workers, "endio-write", 230561d92c32SChris Mason fs_info->thread_pool_size, 230661d92c32SChris Mason &fs_info->generic_worker); 23070cb59c99SJosef Bacik btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write", 23080cb59c99SJosef Bacik 1, &fs_info->generic_worker); 230916cdcec7SMiao Xie btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta", 231016cdcec7SMiao Xie fs_info->thread_pool_size, 231116cdcec7SMiao Xie &fs_info->generic_worker); 231290519d66SArne Jansen btrfs_init_workers(&fs_info->readahead_workers, "readahead", 231390519d66SArne Jansen fs_info->thread_pool_size, 231490519d66SArne Jansen &fs_info->generic_worker); 231561b49440SChris Mason 231661b49440SChris Mason /* 231761b49440SChris Mason * endios are largely parallel and should have a very 231861b49440SChris Mason * low idle thresh 231961b49440SChris Mason */ 232061b49440SChris Mason fs_info->endio_workers.idle_thresh = 4; 2321b51912c9SChris Mason fs_info->endio_meta_workers.idle_thresh = 4; 2322b51912c9SChris Mason 23239042846bSChris Mason fs_info->endio_write_workers.idle_thresh = 2; 23249042846bSChris Mason fs_info->endio_meta_write_workers.idle_thresh = 2; 232590519d66SArne Jansen fs_info->readahead_workers.idle_thresh = 2; 23269042846bSChris Mason 23270dc3b84aSJosef Bacik /* 23280dc3b84aSJosef Bacik * btrfs_start_workers can really only fail because of ENOMEM so just 23290dc3b84aSJosef Bacik * return -ENOMEM if any of these fail. 23300dc3b84aSJosef Bacik */ 23310dc3b84aSJosef Bacik ret = btrfs_start_workers(&fs_info->workers); 23320dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->generic_worker); 23330dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->submit_workers); 23340dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->delalloc_workers); 23350dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->fixup_workers); 23360dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->endio_workers); 23370dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->endio_meta_workers); 23380dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers); 23390dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->endio_write_workers); 23400dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->endio_freespace_worker); 23410dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->delayed_workers); 23420dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->caching_workers); 23430dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->readahead_workers); 23440dc3b84aSJosef Bacik if (ret) { 23450dc3b84aSJosef Bacik ret = -ENOMEM; 23460dc3b84aSJosef Bacik goto fail_sb_buffer; 23470dc3b84aSJosef Bacik } 23484543df7eSChris Mason 23494575c9ccSChris Mason fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super); 2350c8b97818SChris Mason fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages, 2351c8b97818SChris Mason 4 * 1024 * 1024 / PAGE_CACHE_SIZE); 23524575c9ccSChris Mason 2353db94535dSChris Mason tree_root->nodesize = nodesize; 2354db94535dSChris Mason tree_root->leafsize = leafsize; 2355db94535dSChris Mason tree_root->sectorsize = sectorsize; 235687ee04ebSChris Mason tree_root->stripesize = stripesize; 2357a061fc8dSChris Mason 2358a061fc8dSChris Mason sb->s_blocksize = sectorsize; 2359a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 2360db94535dSChris Mason 236139279cc3SChris Mason if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC, 236239279cc3SChris Mason sizeof(disk_super->magic))) { 2363d397712bSChris Mason printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id); 236439279cc3SChris Mason goto fail_sb_buffer; 236539279cc3SChris Mason } 236619c00ddcSChris Mason 23678d082fb7SLiu Bo if (sectorsize != PAGE_SIZE) { 23688d082fb7SLiu Bo printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) " 23698d082fb7SLiu Bo "found on %s\n", (unsigned long)sectorsize, sb->s_id); 2370941b2ddfSKeith Mannthey goto fail_sb_buffer; 2371941b2ddfSKeith Mannthey } 2372941b2ddfSKeith Mannthey 2373925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 2374e4404d6eSYan Zheng ret = btrfs_read_sys_array(tree_root); 2375925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 237684eed90fSChris Mason if (ret) { 2377d397712bSChris Mason printk(KERN_WARNING "btrfs: failed to read the system " 2378d397712bSChris Mason "array on %s\n", sb->s_id); 23795d4f98a2SYan Zheng goto fail_sb_buffer; 238084eed90fSChris Mason } 23810b86a832SChris Mason 23820b86a832SChris Mason blocksize = btrfs_level_size(tree_root, 23830b86a832SChris Mason btrfs_super_chunk_root_level(disk_super)); 238484234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 23850b86a832SChris Mason 23860b86a832SChris Mason __setup_root(nodesize, leafsize, sectorsize, stripesize, 23870b86a832SChris Mason chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID); 23880b86a832SChris Mason 23890b86a832SChris Mason chunk_root->node = read_tree_block(chunk_root, 23900b86a832SChris Mason btrfs_super_chunk_root(disk_super), 239184234f3aSYan Zheng blocksize, generation); 239279787eaaSJeff Mahoney BUG_ON(!chunk_root->node); /* -ENOMEM */ 239383121942SDavid Woodhouse if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) { 239483121942SDavid Woodhouse printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n", 239583121942SDavid Woodhouse sb->s_id); 2396af31f5e5SChris Mason goto fail_tree_roots; 239783121942SDavid Woodhouse } 23985d4f98a2SYan Zheng btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 23995d4f98a2SYan Zheng chunk_root->commit_root = btrfs_root_node(chunk_root); 24000b86a832SChris Mason 2401e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 2402e17cade2SChris Mason (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node), 2403e17cade2SChris Mason BTRFS_UUID_SIZE); 2404e17cade2SChris Mason 24050b86a832SChris Mason ret = btrfs_read_chunk_tree(chunk_root); 24062b82032cSYan Zheng if (ret) { 2407d397712bSChris Mason printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n", 2408d397712bSChris Mason sb->s_id); 2409af31f5e5SChris Mason goto fail_tree_roots; 24102b82032cSYan Zheng } 24110b86a832SChris Mason 2412dfe25020SChris Mason btrfs_close_extra_devices(fs_devices); 2413dfe25020SChris Mason 2414a6b0d5c8SChris Mason if (!fs_devices->latest_bdev) { 2415a6b0d5c8SChris Mason printk(KERN_CRIT "btrfs: failed to read devices on %s\n", 2416a6b0d5c8SChris Mason sb->s_id); 2417a6b0d5c8SChris Mason goto fail_tree_roots; 2418a6b0d5c8SChris Mason } 2419a6b0d5c8SChris Mason 2420af31f5e5SChris Mason retry_root_backup: 2421db94535dSChris Mason blocksize = btrfs_level_size(tree_root, 2422db94535dSChris Mason btrfs_super_root_level(disk_super)); 242384234f3aSYan Zheng generation = btrfs_super_generation(disk_super); 24240b86a832SChris Mason 2425e20d96d6SChris Mason tree_root->node = read_tree_block(tree_root, 2426db94535dSChris Mason btrfs_super_root(disk_super), 242784234f3aSYan Zheng blocksize, generation); 2428af31f5e5SChris Mason if (!tree_root->node || 2429af31f5e5SChris Mason !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) { 243083121942SDavid Woodhouse printk(KERN_WARNING "btrfs: failed to read tree root on %s\n", 243183121942SDavid Woodhouse sb->s_id); 2432af31f5e5SChris Mason 2433af31f5e5SChris Mason goto recovery_tree_root; 243483121942SDavid Woodhouse } 2435af31f5e5SChris Mason 24365d4f98a2SYan Zheng btrfs_set_root_node(&tree_root->root_item, tree_root->node); 24375d4f98a2SYan Zheng tree_root->commit_root = btrfs_root_node(tree_root); 2438db94535dSChris Mason 2439db94535dSChris Mason ret = find_and_setup_root(tree_root, fs_info, 2440e20d96d6SChris Mason BTRFS_EXTENT_TREE_OBJECTID, extent_root); 24410b86a832SChris Mason if (ret) 2442af31f5e5SChris Mason goto recovery_tree_root; 24430b86a832SChris Mason extent_root->track_dirty = 1; 24440b86a832SChris Mason 24450b86a832SChris Mason ret = find_and_setup_root(tree_root, fs_info, 24460b86a832SChris Mason BTRFS_DEV_TREE_OBJECTID, dev_root); 24470b86a832SChris Mason if (ret) 2448af31f5e5SChris Mason goto recovery_tree_root; 24495d4f98a2SYan Zheng dev_root->track_dirty = 1; 24503768f368SChris Mason 2451d20f7043SChris Mason ret = find_and_setup_root(tree_root, fs_info, 2452d20f7043SChris Mason BTRFS_CSUM_TREE_OBJECTID, csum_root); 2453d20f7043SChris Mason if (ret) 2454af31f5e5SChris Mason goto recovery_tree_root; 2455d20f7043SChris Mason csum_root->track_dirty = 1; 2456d20f7043SChris Mason 2457bcef60f2SArne Jansen ret = find_and_setup_root(tree_root, fs_info, 2458bcef60f2SArne Jansen BTRFS_QUOTA_TREE_OBJECTID, quota_root); 2459bcef60f2SArne Jansen if (ret) { 2460bcef60f2SArne Jansen kfree(quota_root); 2461bcef60f2SArne Jansen quota_root = fs_info->quota_root = NULL; 2462bcef60f2SArne Jansen } else { 2463bcef60f2SArne Jansen quota_root->track_dirty = 1; 2464bcef60f2SArne Jansen fs_info->quota_enabled = 1; 2465bcef60f2SArne Jansen fs_info->pending_quota_state = 1; 2466bcef60f2SArne Jansen } 2467bcef60f2SArne Jansen 24688929ecfaSYan, Zheng fs_info->generation = generation; 24698929ecfaSYan, Zheng fs_info->last_trans_committed = generation; 24708929ecfaSYan, Zheng 247168310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 247268310a5eSIlya Dryomov if (ret) { 247368310a5eSIlya Dryomov printk(KERN_WARNING "btrfs: failed to recover balance\n"); 247468310a5eSIlya Dryomov goto fail_block_groups; 247568310a5eSIlya Dryomov } 247668310a5eSIlya Dryomov 2477733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 2478733f4fbbSStefan Behrens if (ret) { 2479733f4fbbSStefan Behrens printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n", 2480733f4fbbSStefan Behrens ret); 2481733f4fbbSStefan Behrens goto fail_block_groups; 2482733f4fbbSStefan Behrens } 2483733f4fbbSStefan Behrens 2484c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 2485c59021f8Sliubo if (ret) { 2486c59021f8Sliubo printk(KERN_ERR "Failed to initial space info: %d\n", ret); 2487c59021f8Sliubo goto fail_block_groups; 2488c59021f8Sliubo } 2489c59021f8Sliubo 24901b1d1f66SJosef Bacik ret = btrfs_read_block_groups(extent_root); 24911b1d1f66SJosef Bacik if (ret) { 24921b1d1f66SJosef Bacik printk(KERN_ERR "Failed to read block groups: %d\n", ret); 24931b1d1f66SJosef Bacik goto fail_block_groups; 24941b1d1f66SJosef Bacik } 24959078a3e1SChris Mason 2496a74a4b97SChris Mason fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 2497a74a4b97SChris Mason "btrfs-cleaner"); 249857506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 24991b1d1f66SJosef Bacik goto fail_block_groups; 2500a74a4b97SChris Mason 2501a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 2502a74a4b97SChris Mason tree_root, 2503a74a4b97SChris Mason "btrfs-transaction"); 250457506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 25053f157a2fSChris Mason goto fail_cleaner; 2506a74a4b97SChris Mason 2507c289811cSChris Mason if (!btrfs_test_opt(tree_root, SSD) && 2508c289811cSChris Mason !btrfs_test_opt(tree_root, NOSSD) && 2509c289811cSChris Mason !fs_info->fs_devices->rotating) { 2510c289811cSChris Mason printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD " 2511c289811cSChris Mason "mode\n"); 2512c289811cSChris Mason btrfs_set_opt(fs_info->mount_opt, SSD); 2513c289811cSChris Mason } 2514c289811cSChris Mason 251521adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 251621adbd5cSStefan Behrens if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) { 251721adbd5cSStefan Behrens ret = btrfsic_mount(tree_root, fs_devices, 251821adbd5cSStefan Behrens btrfs_test_opt(tree_root, 251921adbd5cSStefan Behrens CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? 252021adbd5cSStefan Behrens 1 : 0, 252121adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 252221adbd5cSStefan Behrens if (ret) 252321adbd5cSStefan Behrens printk(KERN_WARNING "btrfs: failed to initialize" 252421adbd5cSStefan Behrens " integrity check module %s\n", sb->s_id); 252521adbd5cSStefan Behrens } 252621adbd5cSStefan Behrens #endif 2527bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 2528bcef60f2SArne Jansen if (ret) 2529bcef60f2SArne Jansen goto fail_trans_kthread; 253021adbd5cSStefan Behrens 2531acce952bSliubo /* do not make disk changes in broken FS */ 2532acce952bSliubo if (btrfs_super_log_root(disk_super) != 0 && 2533acce952bSliubo !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) { 2534e02119d5SChris Mason u64 bytenr = btrfs_super_log_root(disk_super); 2535d18a2c44SChris Mason 25367c2ca468SChris Mason if (fs_devices->rw_devices == 0) { 2537d397712bSChris Mason printk(KERN_WARNING "Btrfs log replay required " 2538d397712bSChris Mason "on RO media\n"); 25397c2ca468SChris Mason err = -EIO; 2540bcef60f2SArne Jansen goto fail_qgroup; 25417c2ca468SChris Mason } 2542e02119d5SChris Mason blocksize = 2543e02119d5SChris Mason btrfs_level_size(tree_root, 2544e02119d5SChris Mason btrfs_super_log_root_level(disk_super)); 2545e02119d5SChris Mason 25466f07e42eSAl Viro log_tree_root = btrfs_alloc_root(fs_info); 2547676e4c86SDan Carpenter if (!log_tree_root) { 2548676e4c86SDan Carpenter err = -ENOMEM; 2549bcef60f2SArne Jansen goto fail_qgroup; 2550676e4c86SDan Carpenter } 2551e02119d5SChris Mason 2552e02119d5SChris Mason __setup_root(nodesize, leafsize, sectorsize, stripesize, 2553e02119d5SChris Mason log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID); 2554e02119d5SChris Mason 2555e02119d5SChris Mason log_tree_root->node = read_tree_block(tree_root, bytenr, 255684234f3aSYan Zheng blocksize, 255784234f3aSYan Zheng generation + 1); 255879787eaaSJeff Mahoney /* returns with log_tree_root freed on success */ 2559e02119d5SChris Mason ret = btrfs_recover_log_trees(log_tree_root); 256079787eaaSJeff Mahoney if (ret) { 256179787eaaSJeff Mahoney btrfs_error(tree_root->fs_info, ret, 256279787eaaSJeff Mahoney "Failed to recover log tree"); 256379787eaaSJeff Mahoney free_extent_buffer(log_tree_root->node); 256479787eaaSJeff Mahoney kfree(log_tree_root); 256579787eaaSJeff Mahoney goto fail_trans_kthread; 256679787eaaSJeff Mahoney } 2567e556ce2cSYan Zheng 2568e556ce2cSYan Zheng if (sb->s_flags & MS_RDONLY) { 2569e556ce2cSYan Zheng ret = btrfs_commit_super(tree_root); 257079787eaaSJeff Mahoney if (ret) 257179787eaaSJeff Mahoney goto fail_trans_kthread; 2572e556ce2cSYan Zheng } 2573e02119d5SChris Mason } 25741a40e23bSZheng Yan 257576dda93cSYan, Zheng ret = btrfs_find_orphan_roots(tree_root); 257679787eaaSJeff Mahoney if (ret) 257779787eaaSJeff Mahoney goto fail_trans_kthread; 257876dda93cSYan, Zheng 25797c2ca468SChris Mason if (!(sb->s_flags & MS_RDONLY)) { 2580d68fc57bSYan, Zheng ret = btrfs_cleanup_fs_roots(fs_info); 258179787eaaSJeff Mahoney if (ret) { 258279787eaaSJeff Mahoney } 2583d68fc57bSYan, Zheng 25845d4f98a2SYan Zheng ret = btrfs_recover_relocation(tree_root); 2585d7ce5843SMiao Xie if (ret < 0) { 2586d7ce5843SMiao Xie printk(KERN_WARNING 2587d7ce5843SMiao Xie "btrfs: failed to recover relocation\n"); 2588d7ce5843SMiao Xie err = -EINVAL; 2589bcef60f2SArne Jansen goto fail_qgroup; 2590d7ce5843SMiao Xie } 25917c2ca468SChris Mason } 25921a40e23bSZheng Yan 25933de4586cSChris Mason location.objectid = BTRFS_FS_TREE_OBJECTID; 25943de4586cSChris Mason location.type = BTRFS_ROOT_ITEM_KEY; 25953de4586cSChris Mason location.offset = (u64)-1; 25963de4586cSChris Mason 25973de4586cSChris Mason fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location); 25983de4586cSChris Mason if (!fs_info->fs_root) 2599bcef60f2SArne Jansen goto fail_qgroup; 26003140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 26013140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 2602bcef60f2SArne Jansen goto fail_qgroup; 26033140c9a3SDan Carpenter } 2604c289811cSChris Mason 26052b6ba629SIlya Dryomov if (sb->s_flags & MS_RDONLY) 26062b6ba629SIlya Dryomov return 0; 26072b6ba629SIlya Dryomov 2608e3acc2a6SJosef Bacik down_read(&fs_info->cleanup_work_sem); 26092b6ba629SIlya Dryomov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 26102b6ba629SIlya Dryomov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 26112b6ba629SIlya Dryomov up_read(&fs_info->cleanup_work_sem); 26122b6ba629SIlya Dryomov close_ctree(tree_root); 26132b6ba629SIlya Dryomov return ret; 26142b6ba629SIlya Dryomov } 2615e3acc2a6SJosef Bacik up_read(&fs_info->cleanup_work_sem); 261659641015SIlya Dryomov 26172b6ba629SIlya Dryomov ret = btrfs_resume_balance_async(fs_info); 26182b6ba629SIlya Dryomov if (ret) { 26192b6ba629SIlya Dryomov printk(KERN_WARNING "btrfs: failed to resume balance\n"); 262066b4ffd1SJosef Bacik close_ctree(tree_root); 26212b6ba629SIlya Dryomov return ret; 2622e3acc2a6SJosef Bacik } 2623e3acc2a6SJosef Bacik 2624ad2b2c80SAl Viro return 0; 262539279cc3SChris Mason 2626bcef60f2SArne Jansen fail_qgroup: 2627bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 26287c2ca468SChris Mason fail_trans_kthread: 26297c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 26303f157a2fSChris Mason fail_cleaner: 2631a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 26327c2ca468SChris Mason 26337c2ca468SChris Mason /* 26347c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 26357c2ca468SChris Mason * kthreads 26367c2ca468SChris Mason */ 26377c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 26387c2ca468SChris Mason invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 26397c2ca468SChris Mason 26401b1d1f66SJosef Bacik fail_block_groups: 26411b1d1f66SJosef Bacik btrfs_free_block_groups(fs_info); 2642af31f5e5SChris Mason 2643af31f5e5SChris Mason fail_tree_roots: 2644af31f5e5SChris Mason free_root_pointers(fs_info, 1); 2645af31f5e5SChris Mason 264639279cc3SChris Mason fail_sb_buffer: 264761d92c32SChris Mason btrfs_stop_workers(&fs_info->generic_worker); 2648306c8b68SChris Mason btrfs_stop_workers(&fs_info->readahead_workers); 2649247e743cSChris Mason btrfs_stop_workers(&fs_info->fixup_workers); 2650771ed689SChris Mason btrfs_stop_workers(&fs_info->delalloc_workers); 26518b712842SChris Mason btrfs_stop_workers(&fs_info->workers); 26528b712842SChris Mason btrfs_stop_workers(&fs_info->endio_workers); 2653d20f7043SChris Mason btrfs_stop_workers(&fs_info->endio_meta_workers); 2654cad321adSChris Mason btrfs_stop_workers(&fs_info->endio_meta_write_workers); 2655e6dcd2dcSChris Mason btrfs_stop_workers(&fs_info->endio_write_workers); 26560cb59c99SJosef Bacik btrfs_stop_workers(&fs_info->endio_freespace_worker); 26571cc127b5SChris Mason btrfs_stop_workers(&fs_info->submit_workers); 265816cdcec7SMiao Xie btrfs_stop_workers(&fs_info->delayed_workers); 2659bab39bf9SJosef Bacik btrfs_stop_workers(&fs_info->caching_workers); 266016cdcec7SMiao Xie fail_alloc: 26614543df7eSChris Mason fail_iput: 2662586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 2663586e46e2SIlya Dryomov 26647c2ca468SChris Mason invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 26654543df7eSChris Mason iput(fs_info->btree_inode); 2666ad081f14SJens Axboe fail_bdi: 26677e662854SQinghuang Feng bdi_destroy(&fs_info->bdi); 266876dda93cSYan, Zheng fail_srcu: 266976dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 26707e662854SQinghuang Feng fail: 2671586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 2672ad2b2c80SAl Viro return err; 2673af31f5e5SChris Mason 2674af31f5e5SChris Mason recovery_tree_root: 2675af31f5e5SChris Mason if (!btrfs_test_opt(tree_root, RECOVERY)) 2676af31f5e5SChris Mason goto fail_tree_roots; 2677af31f5e5SChris Mason 2678af31f5e5SChris Mason free_root_pointers(fs_info, 0); 2679af31f5e5SChris Mason 2680af31f5e5SChris Mason /* don't use the log in recovery mode, it won't be valid */ 2681af31f5e5SChris Mason btrfs_set_super_log_root(disk_super, 0); 2682af31f5e5SChris Mason 2683af31f5e5SChris Mason /* we can't trust the free space cache either */ 2684af31f5e5SChris Mason btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 2685af31f5e5SChris Mason 2686af31f5e5SChris Mason ret = next_root_backup(fs_info, fs_info->super_copy, 2687af31f5e5SChris Mason &num_backups_tried, &backup_index); 2688af31f5e5SChris Mason if (ret == -1) 2689af31f5e5SChris Mason goto fail_block_groups; 2690af31f5e5SChris Mason goto retry_root_backup; 2691eb60ceacSChris Mason } 2692eb60ceacSChris Mason 2693f2984462SChris Mason static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate) 2694f2984462SChris Mason { 2695f2984462SChris Mason if (uptodate) { 2696f2984462SChris Mason set_buffer_uptodate(bh); 2697f2984462SChris Mason } else { 2698442a4f63SStefan Behrens struct btrfs_device *device = (struct btrfs_device *) 2699442a4f63SStefan Behrens bh->b_private; 2700442a4f63SStefan Behrens 2701606686eeSJosef Bacik printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to " 2702606686eeSJosef Bacik "I/O error on %s\n", 2703606686eeSJosef Bacik rcu_str_deref(device->name)); 27041259ab75SChris Mason /* note, we dont' set_buffer_write_io_error because we have 27051259ab75SChris Mason * our own ways of dealing with the IO errors 27061259ab75SChris Mason */ 2707f2984462SChris Mason clear_buffer_uptodate(bh); 2708442a4f63SStefan Behrens btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS); 2709f2984462SChris Mason } 2710f2984462SChris Mason unlock_buffer(bh); 2711f2984462SChris Mason put_bh(bh); 2712f2984462SChris Mason } 2713f2984462SChris Mason 2714a512bbf8SYan Zheng struct buffer_head *btrfs_read_dev_super(struct block_device *bdev) 2715a512bbf8SYan Zheng { 2716a512bbf8SYan Zheng struct buffer_head *bh; 2717a512bbf8SYan Zheng struct buffer_head *latest = NULL; 2718a512bbf8SYan Zheng struct btrfs_super_block *super; 2719a512bbf8SYan Zheng int i; 2720a512bbf8SYan Zheng u64 transid = 0; 2721a512bbf8SYan Zheng u64 bytenr; 2722a512bbf8SYan Zheng 2723a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 2724a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 2725a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 2726a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 2727a512bbf8SYan Zheng */ 2728a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 2729a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 2730a512bbf8SYan Zheng if (bytenr + 4096 >= i_size_read(bdev->bd_inode)) 2731a512bbf8SYan Zheng break; 2732a512bbf8SYan Zheng bh = __bread(bdev, bytenr / 4096, 4096); 2733a512bbf8SYan Zheng if (!bh) 2734a512bbf8SYan Zheng continue; 2735a512bbf8SYan Zheng 2736a512bbf8SYan Zheng super = (struct btrfs_super_block *)bh->b_data; 2737a512bbf8SYan Zheng if (btrfs_super_bytenr(super) != bytenr || 2738a512bbf8SYan Zheng strncmp((char *)(&super->magic), BTRFS_MAGIC, 2739a512bbf8SYan Zheng sizeof(super->magic))) { 2740a512bbf8SYan Zheng brelse(bh); 2741a512bbf8SYan Zheng continue; 2742a512bbf8SYan Zheng } 2743a512bbf8SYan Zheng 2744a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 2745a512bbf8SYan Zheng brelse(latest); 2746a512bbf8SYan Zheng latest = bh; 2747a512bbf8SYan Zheng transid = btrfs_super_generation(super); 2748a512bbf8SYan Zheng } else { 2749a512bbf8SYan Zheng brelse(bh); 2750a512bbf8SYan Zheng } 2751a512bbf8SYan Zheng } 2752a512bbf8SYan Zheng return latest; 2753a512bbf8SYan Zheng } 2754a512bbf8SYan Zheng 27554eedeb75SHisashi Hifumi /* 27564eedeb75SHisashi Hifumi * this should be called twice, once with wait == 0 and 27574eedeb75SHisashi Hifumi * once with wait == 1. When wait == 0 is done, all the buffer heads 27584eedeb75SHisashi Hifumi * we write are pinned. 27594eedeb75SHisashi Hifumi * 27604eedeb75SHisashi Hifumi * They are released when wait == 1 is done. 27614eedeb75SHisashi Hifumi * max_mirrors must be the same for both runs, and it indicates how 27624eedeb75SHisashi Hifumi * many supers on this one device should be written. 27634eedeb75SHisashi Hifumi * 27644eedeb75SHisashi Hifumi * max_mirrors == 0 means to write them all. 27654eedeb75SHisashi Hifumi */ 2766a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 2767a512bbf8SYan Zheng struct btrfs_super_block *sb, 2768a512bbf8SYan Zheng int do_barriers, int wait, int max_mirrors) 2769a512bbf8SYan Zheng { 2770a512bbf8SYan Zheng struct buffer_head *bh; 2771a512bbf8SYan Zheng int i; 2772a512bbf8SYan Zheng int ret; 2773a512bbf8SYan Zheng int errors = 0; 2774a512bbf8SYan Zheng u32 crc; 2775a512bbf8SYan Zheng u64 bytenr; 2776a512bbf8SYan Zheng 2777a512bbf8SYan Zheng if (max_mirrors == 0) 2778a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 2779a512bbf8SYan Zheng 2780a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 2781a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 2782a512bbf8SYan Zheng if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes) 2783a512bbf8SYan Zheng break; 2784a512bbf8SYan Zheng 2785a512bbf8SYan Zheng if (wait) { 2786a512bbf8SYan Zheng bh = __find_get_block(device->bdev, bytenr / 4096, 2787a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 2788a512bbf8SYan Zheng BUG_ON(!bh); 2789a512bbf8SYan Zheng wait_on_buffer(bh); 27904eedeb75SHisashi Hifumi if (!buffer_uptodate(bh)) 27914eedeb75SHisashi Hifumi errors++; 27924eedeb75SHisashi Hifumi 27934eedeb75SHisashi Hifumi /* drop our reference */ 27944eedeb75SHisashi Hifumi brelse(bh); 27954eedeb75SHisashi Hifumi 27964eedeb75SHisashi Hifumi /* drop the reference from the wait == 0 run */ 2797a512bbf8SYan Zheng brelse(bh); 2798a512bbf8SYan Zheng continue; 2799a512bbf8SYan Zheng } else { 2800a512bbf8SYan Zheng btrfs_set_super_bytenr(sb, bytenr); 2801a512bbf8SYan Zheng 2802a512bbf8SYan Zheng crc = ~(u32)0; 2803a512bbf8SYan Zheng crc = btrfs_csum_data(NULL, (char *)sb + 2804a512bbf8SYan Zheng BTRFS_CSUM_SIZE, crc, 2805a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE - 2806a512bbf8SYan Zheng BTRFS_CSUM_SIZE); 2807a512bbf8SYan Zheng btrfs_csum_final(crc, sb->csum); 2808a512bbf8SYan Zheng 28094eedeb75SHisashi Hifumi /* 28104eedeb75SHisashi Hifumi * one reference for us, and we leave it for the 28114eedeb75SHisashi Hifumi * caller 28124eedeb75SHisashi Hifumi */ 2813a512bbf8SYan Zheng bh = __getblk(device->bdev, bytenr / 4096, 2814a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 2815a512bbf8SYan Zheng memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE); 2816a512bbf8SYan Zheng 28174eedeb75SHisashi Hifumi /* one reference for submit_bh */ 2818a512bbf8SYan Zheng get_bh(bh); 28194eedeb75SHisashi Hifumi 28204eedeb75SHisashi Hifumi set_buffer_uptodate(bh); 2821a512bbf8SYan Zheng lock_buffer(bh); 2822a512bbf8SYan Zheng bh->b_end_io = btrfs_end_buffer_write_sync; 2823442a4f63SStefan Behrens bh->b_private = device; 2824a512bbf8SYan Zheng } 2825a512bbf8SYan Zheng 2826387125fcSChris Mason /* 2827387125fcSChris Mason * we fua the first super. The others we allow 2828387125fcSChris Mason * to go down lazy. 2829387125fcSChris Mason */ 283021adbd5cSStefan Behrens ret = btrfsic_submit_bh(WRITE_FUA, bh); 28314eedeb75SHisashi Hifumi if (ret) 2832a512bbf8SYan Zheng errors++; 2833a512bbf8SYan Zheng } 2834a512bbf8SYan Zheng return errors < i ? 0 : -1; 2835a512bbf8SYan Zheng } 2836a512bbf8SYan Zheng 2837387125fcSChris Mason /* 2838387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 2839387125fcSChris Mason * for the barrier when it is done 2840387125fcSChris Mason */ 2841387125fcSChris Mason static void btrfs_end_empty_barrier(struct bio *bio, int err) 2842387125fcSChris Mason { 2843387125fcSChris Mason if (err) { 2844387125fcSChris Mason if (err == -EOPNOTSUPP) 2845387125fcSChris Mason set_bit(BIO_EOPNOTSUPP, &bio->bi_flags); 2846387125fcSChris Mason clear_bit(BIO_UPTODATE, &bio->bi_flags); 2847387125fcSChris Mason } 2848387125fcSChris Mason if (bio->bi_private) 2849387125fcSChris Mason complete(bio->bi_private); 2850387125fcSChris Mason bio_put(bio); 2851387125fcSChris Mason } 2852387125fcSChris Mason 2853387125fcSChris Mason /* 2854387125fcSChris Mason * trigger flushes for one the devices. If you pass wait == 0, the flushes are 2855387125fcSChris Mason * sent down. With wait == 1, it waits for the previous flush. 2856387125fcSChris Mason * 2857387125fcSChris Mason * any device where the flush fails with eopnotsupp are flagged as not-barrier 2858387125fcSChris Mason * capable 2859387125fcSChris Mason */ 2860387125fcSChris Mason static int write_dev_flush(struct btrfs_device *device, int wait) 2861387125fcSChris Mason { 2862387125fcSChris Mason struct bio *bio; 2863387125fcSChris Mason int ret = 0; 2864387125fcSChris Mason 2865387125fcSChris Mason if (device->nobarriers) 2866387125fcSChris Mason return 0; 2867387125fcSChris Mason 2868387125fcSChris Mason if (wait) { 2869387125fcSChris Mason bio = device->flush_bio; 2870387125fcSChris Mason if (!bio) 2871387125fcSChris Mason return 0; 2872387125fcSChris Mason 2873387125fcSChris Mason wait_for_completion(&device->flush_wait); 2874387125fcSChris Mason 2875387125fcSChris Mason if (bio_flagged(bio, BIO_EOPNOTSUPP)) { 2876606686eeSJosef Bacik printk_in_rcu("btrfs: disabling barriers on dev %s\n", 2877606686eeSJosef Bacik rcu_str_deref(device->name)); 2878387125fcSChris Mason device->nobarriers = 1; 2879387125fcSChris Mason } 2880387125fcSChris Mason if (!bio_flagged(bio, BIO_UPTODATE)) { 2881387125fcSChris Mason ret = -EIO; 2882442a4f63SStefan Behrens if (!bio_flagged(bio, BIO_EOPNOTSUPP)) 2883442a4f63SStefan Behrens btrfs_dev_stat_inc_and_print(device, 2884442a4f63SStefan Behrens BTRFS_DEV_STAT_FLUSH_ERRS); 2885387125fcSChris Mason } 2886387125fcSChris Mason 2887387125fcSChris Mason /* drop the reference from the wait == 0 run */ 2888387125fcSChris Mason bio_put(bio); 2889387125fcSChris Mason device->flush_bio = NULL; 2890387125fcSChris Mason 2891387125fcSChris Mason return ret; 2892387125fcSChris Mason } 2893387125fcSChris Mason 2894387125fcSChris Mason /* 2895387125fcSChris Mason * one reference for us, and we leave it for the 2896387125fcSChris Mason * caller 2897387125fcSChris Mason */ 2898387125fcSChris Mason device->flush_bio = NULL;; 2899387125fcSChris Mason bio = bio_alloc(GFP_NOFS, 0); 2900387125fcSChris Mason if (!bio) 2901387125fcSChris Mason return -ENOMEM; 2902387125fcSChris Mason 2903387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 2904387125fcSChris Mason bio->bi_bdev = device->bdev; 2905387125fcSChris Mason init_completion(&device->flush_wait); 2906387125fcSChris Mason bio->bi_private = &device->flush_wait; 2907387125fcSChris Mason device->flush_bio = bio; 2908387125fcSChris Mason 2909387125fcSChris Mason bio_get(bio); 291021adbd5cSStefan Behrens btrfsic_submit_bio(WRITE_FLUSH, bio); 2911387125fcSChris Mason 2912387125fcSChris Mason return 0; 2913387125fcSChris Mason } 2914387125fcSChris Mason 2915387125fcSChris Mason /* 2916387125fcSChris Mason * send an empty flush down to each device in parallel, 2917387125fcSChris Mason * then wait for them 2918387125fcSChris Mason */ 2919387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 2920387125fcSChris Mason { 2921387125fcSChris Mason struct list_head *head; 2922387125fcSChris Mason struct btrfs_device *dev; 2923387125fcSChris Mason int errors = 0; 2924387125fcSChris Mason int ret; 2925387125fcSChris Mason 2926387125fcSChris Mason /* send down all the barriers */ 2927387125fcSChris Mason head = &info->fs_devices->devices; 2928387125fcSChris Mason list_for_each_entry_rcu(dev, head, dev_list) { 2929387125fcSChris Mason if (!dev->bdev) { 2930387125fcSChris Mason errors++; 2931387125fcSChris Mason continue; 2932387125fcSChris Mason } 2933387125fcSChris Mason if (!dev->in_fs_metadata || !dev->writeable) 2934387125fcSChris Mason continue; 2935387125fcSChris Mason 2936387125fcSChris Mason ret = write_dev_flush(dev, 0); 2937387125fcSChris Mason if (ret) 2938387125fcSChris Mason errors++; 2939387125fcSChris Mason } 2940387125fcSChris Mason 2941387125fcSChris Mason /* wait for all the barriers */ 2942387125fcSChris Mason list_for_each_entry_rcu(dev, head, dev_list) { 2943387125fcSChris Mason if (!dev->bdev) { 2944387125fcSChris Mason errors++; 2945387125fcSChris Mason continue; 2946387125fcSChris Mason } 2947387125fcSChris Mason if (!dev->in_fs_metadata || !dev->writeable) 2948387125fcSChris Mason continue; 2949387125fcSChris Mason 2950387125fcSChris Mason ret = write_dev_flush(dev, 1); 2951387125fcSChris Mason if (ret) 2952387125fcSChris Mason errors++; 2953387125fcSChris Mason } 2954387125fcSChris Mason if (errors) 2955387125fcSChris Mason return -EIO; 2956387125fcSChris Mason return 0; 2957387125fcSChris Mason } 2958387125fcSChris Mason 2959a512bbf8SYan Zheng int write_all_supers(struct btrfs_root *root, int max_mirrors) 2960f2984462SChris Mason { 2961e5e9a520SChris Mason struct list_head *head; 2962f2984462SChris Mason struct btrfs_device *dev; 2963a061fc8dSChris Mason struct btrfs_super_block *sb; 2964f2984462SChris Mason struct btrfs_dev_item *dev_item; 2965f2984462SChris Mason int ret; 2966f2984462SChris Mason int do_barriers; 2967a236aed1SChris Mason int max_errors; 2968a236aed1SChris Mason int total_errors = 0; 2969a061fc8dSChris Mason u64 flags; 2970f2984462SChris Mason 29716c41761fSDavid Sterba max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1; 2972f2984462SChris Mason do_barriers = !btrfs_test_opt(root, NOBARRIER); 2973af31f5e5SChris Mason backup_super_roots(root->fs_info); 2974f2984462SChris Mason 29756c41761fSDavid Sterba sb = root->fs_info->super_for_commit; 2976a061fc8dSChris Mason dev_item = &sb->dev_item; 2977e5e9a520SChris Mason 2978174ba509SChris Mason mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 2979e5e9a520SChris Mason head = &root->fs_info->fs_devices->devices; 2980387125fcSChris Mason 2981387125fcSChris Mason if (do_barriers) 2982387125fcSChris Mason barrier_all_devices(root->fs_info); 2983387125fcSChris Mason 29841f78160cSXiao Guangrong list_for_each_entry_rcu(dev, head, dev_list) { 2985dfe25020SChris Mason if (!dev->bdev) { 2986dfe25020SChris Mason total_errors++; 2987dfe25020SChris Mason continue; 2988dfe25020SChris Mason } 29892b82032cSYan Zheng if (!dev->in_fs_metadata || !dev->writeable) 2990dfe25020SChris Mason continue; 2991dfe25020SChris Mason 29922b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 2993a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 2994a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 2995a061fc8dSChris Mason btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes); 2996a061fc8dSChris Mason btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used); 2997a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 2998a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 2999a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 3000a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 30012b82032cSYan Zheng memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE); 3002a512bbf8SYan Zheng 3003a061fc8dSChris Mason flags = btrfs_super_flags(sb); 3004a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 3005f2984462SChris Mason 3006a512bbf8SYan Zheng ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors); 3007a236aed1SChris Mason if (ret) 3008a236aed1SChris Mason total_errors++; 3009f2984462SChris Mason } 3010a236aed1SChris Mason if (total_errors > max_errors) { 3011d397712bSChris Mason printk(KERN_ERR "btrfs: %d errors while writing supers\n", 3012d397712bSChris Mason total_errors); 301379787eaaSJeff Mahoney 301479787eaaSJeff Mahoney /* This shouldn't happen. FUA is masked off if unsupported */ 3015a236aed1SChris Mason BUG(); 3016a236aed1SChris Mason } 3017f2984462SChris Mason 3018a512bbf8SYan Zheng total_errors = 0; 30191f78160cSXiao Guangrong list_for_each_entry_rcu(dev, head, dev_list) { 3020dfe25020SChris Mason if (!dev->bdev) 3021dfe25020SChris Mason continue; 30222b82032cSYan Zheng if (!dev->in_fs_metadata || !dev->writeable) 3023dfe25020SChris Mason continue; 3024dfe25020SChris Mason 3025a512bbf8SYan Zheng ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors); 3026a512bbf8SYan Zheng if (ret) 30271259ab75SChris Mason total_errors++; 3028f2984462SChris Mason } 3029174ba509SChris Mason mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 3030a236aed1SChris Mason if (total_errors > max_errors) { 303179787eaaSJeff Mahoney btrfs_error(root->fs_info, -EIO, 303279787eaaSJeff Mahoney "%d errors while writing supers", total_errors); 303379787eaaSJeff Mahoney return -EIO; 3034a236aed1SChris Mason } 3035f2984462SChris Mason return 0; 3036f2984462SChris Mason } 3037f2984462SChris Mason 3038a512bbf8SYan Zheng int write_ctree_super(struct btrfs_trans_handle *trans, 3039a512bbf8SYan Zheng struct btrfs_root *root, int max_mirrors) 3040cfaa7295SChris Mason { 3041e66f709bSChris Mason int ret; 30422c90e5d6SChris Mason 3043a512bbf8SYan Zheng ret = write_all_supers(root, max_mirrors); 30445f39d397SChris Mason return ret; 3045cfaa7295SChris Mason } 3046cfaa7295SChris Mason 3047143bede5SJeff Mahoney void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root) 30482619ba1fSChris Mason { 30494df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 30502619ba1fSChris Mason radix_tree_delete(&fs_info->fs_roots_radix, 30512619ba1fSChris Mason (unsigned long)root->root_key.objectid); 30524df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 305376dda93cSYan, Zheng 305476dda93cSYan, Zheng if (btrfs_root_refs(&root->root_item) == 0) 305576dda93cSYan, Zheng synchronize_srcu(&fs_info->subvol_srcu); 305676dda93cSYan, Zheng 3057581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_pinned); 3058581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_ctl); 30594df27c4dSYan, Zheng free_fs_root(root); 30604df27c4dSYan, Zheng } 30614df27c4dSYan, Zheng 30624df27c4dSYan, Zheng static void free_fs_root(struct btrfs_root *root) 30634df27c4dSYan, Zheng { 306482d5902dSLi Zefan iput(root->cache_inode); 30654df27c4dSYan, Zheng WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 30660ee5dc67SAl Viro if (root->anon_dev) 30670ee5dc67SAl Viro free_anon_bdev(root->anon_dev); 30685f39d397SChris Mason free_extent_buffer(root->node); 30695f39d397SChris Mason free_extent_buffer(root->commit_root); 3070581bb050SLi Zefan kfree(root->free_ino_ctl); 3071581bb050SLi Zefan kfree(root->free_ino_pinned); 307258176a96SJosef Bacik kfree(root->name); 30732619ba1fSChris Mason kfree(root); 30742619ba1fSChris Mason } 30752619ba1fSChris Mason 3076143bede5SJeff Mahoney static void del_fs_roots(struct btrfs_fs_info *fs_info) 30770f7d52f4SChris Mason { 30780f7d52f4SChris Mason int ret; 30790f7d52f4SChris Mason struct btrfs_root *gang[8]; 30800f7d52f4SChris Mason int i; 30810f7d52f4SChris Mason 308276dda93cSYan, Zheng while (!list_empty(&fs_info->dead_roots)) { 308376dda93cSYan, Zheng gang[0] = list_entry(fs_info->dead_roots.next, 308476dda93cSYan, Zheng struct btrfs_root, root_list); 308576dda93cSYan, Zheng list_del(&gang[0]->root_list); 308676dda93cSYan, Zheng 308776dda93cSYan, Zheng if (gang[0]->in_radix) { 308876dda93cSYan, Zheng btrfs_free_fs_root(fs_info, gang[0]); 308976dda93cSYan, Zheng } else { 309076dda93cSYan, Zheng free_extent_buffer(gang[0]->node); 309176dda93cSYan, Zheng free_extent_buffer(gang[0]->commit_root); 309276dda93cSYan, Zheng kfree(gang[0]); 309376dda93cSYan, Zheng } 309476dda93cSYan, Zheng } 309576dda93cSYan, Zheng 30960f7d52f4SChris Mason while (1) { 30970f7d52f4SChris Mason ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 30980f7d52f4SChris Mason (void **)gang, 0, 30990f7d52f4SChris Mason ARRAY_SIZE(gang)); 31000f7d52f4SChris Mason if (!ret) 31010f7d52f4SChris Mason break; 31022619ba1fSChris Mason for (i = 0; i < ret; i++) 31035eda7b5eSChris Mason btrfs_free_fs_root(fs_info, gang[i]); 31040f7d52f4SChris Mason } 31050f7d52f4SChris Mason } 3106b4100d64SChris Mason 3107c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 3108c146afadSYan Zheng { 3109c146afadSYan Zheng u64 root_objectid = 0; 3110c146afadSYan Zheng struct btrfs_root *gang[8]; 3111c146afadSYan Zheng int i; 3112c146afadSYan Zheng int ret; 3113c146afadSYan Zheng 3114c146afadSYan Zheng while (1) { 3115c146afadSYan Zheng ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 3116c146afadSYan Zheng (void **)gang, root_objectid, 3117c146afadSYan Zheng ARRAY_SIZE(gang)); 3118c146afadSYan Zheng if (!ret) 3119c146afadSYan Zheng break; 31205d4f98a2SYan Zheng 31215d4f98a2SYan Zheng root_objectid = gang[ret - 1]->root_key.objectid + 1; 3122c146afadSYan Zheng for (i = 0; i < ret; i++) { 312366b4ffd1SJosef Bacik int err; 312466b4ffd1SJosef Bacik 3125c146afadSYan Zheng root_objectid = gang[i]->root_key.objectid; 312666b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(gang[i]); 312766b4ffd1SJosef Bacik if (err) 312866b4ffd1SJosef Bacik return err; 3129c146afadSYan Zheng } 3130c146afadSYan Zheng root_objectid++; 3131c146afadSYan Zheng } 3132c146afadSYan Zheng return 0; 3133c146afadSYan Zheng } 3134c146afadSYan Zheng 3135c146afadSYan Zheng int btrfs_commit_super(struct btrfs_root *root) 3136c146afadSYan Zheng { 3137c146afadSYan Zheng struct btrfs_trans_handle *trans; 3138c146afadSYan Zheng int ret; 3139c146afadSYan Zheng 3140c146afadSYan Zheng mutex_lock(&root->fs_info->cleaner_mutex); 314124bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 3142c146afadSYan Zheng btrfs_clean_old_snapshots(root); 3143c146afadSYan Zheng mutex_unlock(&root->fs_info->cleaner_mutex); 3144c71bf099SYan, Zheng 3145c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 3146c71bf099SYan, Zheng down_write(&root->fs_info->cleanup_work_sem); 3147c71bf099SYan, Zheng up_write(&root->fs_info->cleanup_work_sem); 3148c71bf099SYan, Zheng 31497a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 31503612b495STsutomu Itoh if (IS_ERR(trans)) 31513612b495STsutomu Itoh return PTR_ERR(trans); 3152c146afadSYan Zheng ret = btrfs_commit_transaction(trans, root); 315379787eaaSJeff Mahoney if (ret) 315479787eaaSJeff Mahoney return ret; 3155c146afadSYan Zheng /* run commit again to drop the original snapshot */ 31567a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 31573612b495STsutomu Itoh if (IS_ERR(trans)) 31583612b495STsutomu Itoh return PTR_ERR(trans); 315979787eaaSJeff Mahoney ret = btrfs_commit_transaction(trans, root); 316079787eaaSJeff Mahoney if (ret) 316179787eaaSJeff Mahoney return ret; 3162c146afadSYan Zheng ret = btrfs_write_and_wait_transaction(NULL, root); 316379787eaaSJeff Mahoney if (ret) { 316479787eaaSJeff Mahoney btrfs_error(root->fs_info, ret, 316579787eaaSJeff Mahoney "Failed to sync btree inode to disk."); 316679787eaaSJeff Mahoney return ret; 316779787eaaSJeff Mahoney } 3168c146afadSYan Zheng 3169a512bbf8SYan Zheng ret = write_ctree_super(NULL, root, 0); 3170c146afadSYan Zheng return ret; 3171c146afadSYan Zheng } 3172c146afadSYan Zheng 3173e20d96d6SChris Mason int close_ctree(struct btrfs_root *root) 3174eb60ceacSChris Mason { 31750f7d52f4SChris Mason struct btrfs_fs_info *fs_info = root->fs_info; 3176c146afadSYan Zheng int ret; 3177e089f05cSChris Mason 3178facda1e7SChris Mason fs_info->closing = 1; 3179a2135011SChris Mason smp_mb(); 3180a2135011SChris Mason 3181837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 3182837d5b6eSIlya Dryomov btrfs_pause_balance(root->fs_info); 3183837d5b6eSIlya Dryomov 3184a2de733cSArne Jansen btrfs_scrub_cancel(root); 31854cb5300bSChris Mason 31864cb5300bSChris Mason /* wait for any defraggers to finish */ 31874cb5300bSChris Mason wait_event(fs_info->transaction_wait, 31884cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 31894cb5300bSChris Mason 31904cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 3191e3029d9fSAl Viro btrfs_run_defrag_inodes(fs_info); 31924cb5300bSChris Mason 3193acce952bSliubo /* 3194acce952bSliubo * Here come 2 situations when btrfs is broken to flip readonly: 3195acce952bSliubo * 3196acce952bSliubo * 1. when btrfs flips readonly somewhere else before 3197acce952bSliubo * btrfs_commit_super, sb->s_flags has MS_RDONLY flag, 3198acce952bSliubo * and btrfs will skip to write sb directly to keep 3199acce952bSliubo * ERROR state on disk. 3200acce952bSliubo * 3201acce952bSliubo * 2. when btrfs flips readonly just in btrfs_commit_super, 3202ae0e47f0SJustin P. Mattock * and in such case, btrfs cannot write sb via btrfs_commit_super, 3203acce952bSliubo * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag, 3204acce952bSliubo * btrfs will cleanup all FS resources first and write sb then. 3205acce952bSliubo */ 3206c146afadSYan Zheng if (!(fs_info->sb->s_flags & MS_RDONLY)) { 3207c146afadSYan Zheng ret = btrfs_commit_super(root); 3208d397712bSChris Mason if (ret) 3209d397712bSChris Mason printk(KERN_ERR "btrfs: commit super ret %d\n", ret); 3210c146afadSYan Zheng } 3211ed2ff2cbSChris Mason 3212acce952bSliubo if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { 3213acce952bSliubo ret = btrfs_error_commit_super(root); 3214acce952bSliubo if (ret) 3215acce952bSliubo printk(KERN_ERR "btrfs: commit super ret %d\n", ret); 3216acce952bSliubo } 3217acce952bSliubo 3218300e4f8aSJosef Bacik btrfs_put_block_group_cache(fs_info); 3219300e4f8aSJosef Bacik 3220e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 3221e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 32228929ecfaSYan, Zheng 3223f25784b3SYan Zheng fs_info->closing = 2; 3224f25784b3SYan Zheng smp_mb(); 3225f25784b3SYan Zheng 3226bcef60f2SArne Jansen btrfs_free_qgroup_config(root->fs_info); 3227bcef60f2SArne Jansen 3228b0c68f8bSChris Mason if (fs_info->delalloc_bytes) { 3229d397712bSChris Mason printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n", 323021380931SJoel Becker (unsigned long long)fs_info->delalloc_bytes); 3231b0c68f8bSChris Mason } 323231153d81SYan Zheng if (fs_info->total_ref_cache_size) { 3233d397712bSChris Mason printk(KERN_INFO "btrfs: at umount reference cache size %llu\n", 3234d397712bSChris Mason (unsigned long long)fs_info->total_ref_cache_size); 323531153d81SYan Zheng } 323631153d81SYan Zheng 32375f39d397SChris Mason free_extent_buffer(fs_info->extent_root->node); 32385d4f98a2SYan Zheng free_extent_buffer(fs_info->extent_root->commit_root); 32395f39d397SChris Mason free_extent_buffer(fs_info->tree_root->node); 32405d4f98a2SYan Zheng free_extent_buffer(fs_info->tree_root->commit_root); 3241e3029d9fSAl Viro free_extent_buffer(fs_info->chunk_root->node); 3242e3029d9fSAl Viro free_extent_buffer(fs_info->chunk_root->commit_root); 3243e3029d9fSAl Viro free_extent_buffer(fs_info->dev_root->node); 3244e3029d9fSAl Viro free_extent_buffer(fs_info->dev_root->commit_root); 3245e3029d9fSAl Viro free_extent_buffer(fs_info->csum_root->node); 3246e3029d9fSAl Viro free_extent_buffer(fs_info->csum_root->commit_root); 3247bcef60f2SArne Jansen if (fs_info->quota_root) { 3248bcef60f2SArne Jansen free_extent_buffer(fs_info->quota_root->node); 3249bcef60f2SArne Jansen free_extent_buffer(fs_info->quota_root->commit_root); 3250bcef60f2SArne Jansen } 3251d20f7043SChris Mason 3252e3029d9fSAl Viro btrfs_free_block_groups(fs_info); 3253c146afadSYan Zheng 32540f7d52f4SChris Mason del_fs_roots(fs_info); 3255d10c5f31SChris Mason 3256c146afadSYan Zheng iput(fs_info->btree_inode); 32579ad6b7bcSChris Mason 325861d92c32SChris Mason btrfs_stop_workers(&fs_info->generic_worker); 3259247e743cSChris Mason btrfs_stop_workers(&fs_info->fixup_workers); 3260771ed689SChris Mason btrfs_stop_workers(&fs_info->delalloc_workers); 32618b712842SChris Mason btrfs_stop_workers(&fs_info->workers); 32628b712842SChris Mason btrfs_stop_workers(&fs_info->endio_workers); 3263d20f7043SChris Mason btrfs_stop_workers(&fs_info->endio_meta_workers); 3264cad321adSChris Mason btrfs_stop_workers(&fs_info->endio_meta_write_workers); 3265e6dcd2dcSChris Mason btrfs_stop_workers(&fs_info->endio_write_workers); 32660cb59c99SJosef Bacik btrfs_stop_workers(&fs_info->endio_freespace_worker); 32671cc127b5SChris Mason btrfs_stop_workers(&fs_info->submit_workers); 326816cdcec7SMiao Xie btrfs_stop_workers(&fs_info->delayed_workers); 3269bab39bf9SJosef Bacik btrfs_stop_workers(&fs_info->caching_workers); 327090519d66SArne Jansen btrfs_stop_workers(&fs_info->readahead_workers); 3271d6bfde87SChris Mason 327221adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 327321adbd5cSStefan Behrens if (btrfs_test_opt(root, CHECK_INTEGRITY)) 327421adbd5cSStefan Behrens btrfsic_unmount(root, fs_info->fs_devices); 327521adbd5cSStefan Behrens #endif 327621adbd5cSStefan Behrens 3277dfe25020SChris Mason btrfs_close_devices(fs_info->fs_devices); 32780b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 3279b248a415SChris Mason 328004160088SChris Mason bdi_destroy(&fs_info->bdi); 328176dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 32820b86a832SChris Mason 3283eb60ceacSChris Mason return 0; 3284eb60ceacSChris Mason } 3285eb60ceacSChris Mason 3286b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 3287b9fab919SChris Mason int atomic) 3288ccd467d6SChris Mason { 32891259ab75SChris Mason int ret; 3290727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 32911259ab75SChris Mason 32920b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 32931259ab75SChris Mason if (!ret) 32941259ab75SChris Mason return ret; 32951259ab75SChris Mason 32961259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 3297b9fab919SChris Mason parent_transid, atomic); 3298b9fab919SChris Mason if (ret == -EAGAIN) 3299b9fab919SChris Mason return ret; 33001259ab75SChris Mason return !ret; 33015f39d397SChris Mason } 33026702ed49SChris Mason 33035f39d397SChris Mason int btrfs_set_buffer_uptodate(struct extent_buffer *buf) 33045f39d397SChris Mason { 33050b32f4bbSJosef Bacik return set_extent_buffer_uptodate(buf); 33065f39d397SChris Mason } 33075f39d397SChris Mason 33085f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 33095f39d397SChris Mason { 3310727011e0SChris Mason struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root; 33115f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 3312b9473439SChris Mason int was_dirty; 3313b4ce94deSChris Mason 3314b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 3315ccd467d6SChris Mason if (transid != root->fs_info->generation) { 3316d397712bSChris Mason printk(KERN_CRIT "btrfs transid mismatch buffer %llu, " 3317d397712bSChris Mason "found %llu running %llu\n", 3318db94535dSChris Mason (unsigned long long)buf->start, 3319d397712bSChris Mason (unsigned long long)transid, 3320d397712bSChris Mason (unsigned long long)root->fs_info->generation); 3321ccd467d6SChris Mason WARN_ON(1); 3322ccd467d6SChris Mason } 33230b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 3324b9473439SChris Mason if (!was_dirty) { 3325b9473439SChris Mason spin_lock(&root->fs_info->delalloc_lock); 3326b9473439SChris Mason root->fs_info->dirty_metadata_bytes += buf->len; 3327b9473439SChris Mason spin_unlock(&root->fs_info->delalloc_lock); 3328b9473439SChris Mason } 3329eb60ceacSChris Mason } 3330eb60ceacSChris Mason 3331d3c2fdcfSChris Mason void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr) 333235b7e476SChris Mason { 3333188de649SChris Mason /* 3334188de649SChris Mason * looks as though older kernels can get into trouble with 3335188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 3336188de649SChris Mason */ 3337d6bfde87SChris Mason u64 num_dirty; 3338771ed689SChris Mason unsigned long thresh = 32 * 1024 * 1024; 3339d6bfde87SChris Mason 33406933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 3341d6bfde87SChris Mason return; 3342d6bfde87SChris Mason 334316cdcec7SMiao Xie btrfs_balance_delayed_items(root); 334416cdcec7SMiao Xie 334516cdcec7SMiao Xie num_dirty = root->fs_info->dirty_metadata_bytes; 334616cdcec7SMiao Xie 334716cdcec7SMiao Xie if (num_dirty > thresh) { 334816cdcec7SMiao Xie balance_dirty_pages_ratelimited_nr( 334916cdcec7SMiao Xie root->fs_info->btree_inode->i_mapping, 1); 335016cdcec7SMiao Xie } 335116cdcec7SMiao Xie return; 335216cdcec7SMiao Xie } 335316cdcec7SMiao Xie 335416cdcec7SMiao Xie void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr) 335516cdcec7SMiao Xie { 335616cdcec7SMiao Xie /* 335716cdcec7SMiao Xie * looks as though older kernels can get into trouble with 335816cdcec7SMiao Xie * this code, they end up stuck in balance_dirty_pages forever 335916cdcec7SMiao Xie */ 336016cdcec7SMiao Xie u64 num_dirty; 336116cdcec7SMiao Xie unsigned long thresh = 32 * 1024 * 1024; 336216cdcec7SMiao Xie 336316cdcec7SMiao Xie if (current->flags & PF_MEMALLOC) 336416cdcec7SMiao Xie return; 336516cdcec7SMiao Xie 3366585ad2c3SChris Mason num_dirty = root->fs_info->dirty_metadata_bytes; 3367585ad2c3SChris Mason 3368d6bfde87SChris Mason if (num_dirty > thresh) { 3369d3c2fdcfSChris Mason balance_dirty_pages_ratelimited_nr( 3370304fced6SChris Mason root->fs_info->btree_inode->i_mapping, 1); 337135b7e476SChris Mason } 3372188de649SChris Mason return; 3373d6bfde87SChris Mason } 33746b80053dSChris Mason 3375ca7a79adSChris Mason int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid) 33766b80053dSChris Mason { 3377727011e0SChris Mason struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root; 33780b32f4bbSJosef Bacik return btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 33796b80053dSChris Mason } 33800da5468fSChris Mason 338120897f5cSArne Jansen int btree_lock_page_hook(struct page *page, void *data, 338201d658f2SChris Mason void (*flush_fn)(void *)) 33834bef0848SChris Mason { 33844bef0848SChris Mason struct inode *inode = page->mapping->host; 3385b9473439SChris Mason struct btrfs_root *root = BTRFS_I(inode)->root; 33864bef0848SChris Mason struct extent_buffer *eb; 33874bef0848SChris Mason 33884f2de97aSJosef Bacik /* 33894f2de97aSJosef Bacik * We culled this eb but the page is still hanging out on the mapping, 33904f2de97aSJosef Bacik * carry on. 33914f2de97aSJosef Bacik */ 33924f2de97aSJosef Bacik if (!PagePrivate(page)) 33934bef0848SChris Mason goto out; 33944bef0848SChris Mason 33954f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 33964f2de97aSJosef Bacik if (!eb) { 33974f2de97aSJosef Bacik WARN_ON(1); 33984f2de97aSJosef Bacik goto out; 33994f2de97aSJosef Bacik } 34004f2de97aSJosef Bacik if (page != eb->pages[0]) 34014bef0848SChris Mason goto out; 34024bef0848SChris Mason 340301d658f2SChris Mason if (!btrfs_try_tree_write_lock(eb)) { 340401d658f2SChris Mason flush_fn(data); 34054bef0848SChris Mason btrfs_tree_lock(eb); 340601d658f2SChris Mason } 34074bef0848SChris Mason btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); 3408b9473439SChris Mason 3409b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { 3410b9473439SChris Mason spin_lock(&root->fs_info->delalloc_lock); 3411b9473439SChris Mason if (root->fs_info->dirty_metadata_bytes >= eb->len) 3412b9473439SChris Mason root->fs_info->dirty_metadata_bytes -= eb->len; 3413b9473439SChris Mason else 3414b9473439SChris Mason WARN_ON(1); 3415b9473439SChris Mason spin_unlock(&root->fs_info->delalloc_lock); 3416b9473439SChris Mason } 3417b9473439SChris Mason 34184bef0848SChris Mason btrfs_tree_unlock(eb); 34194bef0848SChris Mason out: 342001d658f2SChris Mason if (!trylock_page(page)) { 342101d658f2SChris Mason flush_fn(data); 34224bef0848SChris Mason lock_page(page); 342301d658f2SChris Mason } 34244bef0848SChris Mason return 0; 34254bef0848SChris Mason } 34264bef0848SChris Mason 3427fcd1f065SDavid Sterba static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info, 3428acce952bSliubo int read_only) 3429acce952bSliubo { 3430fcd1f065SDavid Sterba if (btrfs_super_csum_type(fs_info->super_copy) >= ARRAY_SIZE(btrfs_csum_sizes)) { 3431fcd1f065SDavid Sterba printk(KERN_ERR "btrfs: unsupported checksum algorithm\n"); 3432fcd1f065SDavid Sterba return -EINVAL; 3433fcd1f065SDavid Sterba } 3434acce952bSliubo 3435fcd1f065SDavid Sterba if (read_only) 3436fcd1f065SDavid Sterba return 0; 3437fcd1f065SDavid Sterba 3438fcd1f065SDavid Sterba if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { 3439acce952bSliubo printk(KERN_WARNING "warning: mount fs with errors, " 3440acce952bSliubo "running btrfsck is recommended\n"); 3441acce952bSliubo } 3442acce952bSliubo 3443fcd1f065SDavid Sterba return 0; 3444fcd1f065SDavid Sterba } 3445fcd1f065SDavid Sterba 3446acce952bSliubo int btrfs_error_commit_super(struct btrfs_root *root) 3447acce952bSliubo { 3448acce952bSliubo int ret; 3449acce952bSliubo 3450acce952bSliubo mutex_lock(&root->fs_info->cleaner_mutex); 3451acce952bSliubo btrfs_run_delayed_iputs(root); 3452acce952bSliubo mutex_unlock(&root->fs_info->cleaner_mutex); 3453acce952bSliubo 3454acce952bSliubo down_write(&root->fs_info->cleanup_work_sem); 3455acce952bSliubo up_write(&root->fs_info->cleanup_work_sem); 3456acce952bSliubo 3457acce952bSliubo /* cleanup FS via transaction */ 3458acce952bSliubo btrfs_cleanup_transaction(root); 3459acce952bSliubo 3460acce952bSliubo ret = write_ctree_super(NULL, root, 0); 3461acce952bSliubo 3462acce952bSliubo return ret; 3463acce952bSliubo } 3464acce952bSliubo 3465143bede5SJeff Mahoney static void btrfs_destroy_ordered_operations(struct btrfs_root *root) 3466acce952bSliubo { 3467acce952bSliubo struct btrfs_inode *btrfs_inode; 3468acce952bSliubo struct list_head splice; 3469acce952bSliubo 3470acce952bSliubo INIT_LIST_HEAD(&splice); 3471acce952bSliubo 3472acce952bSliubo mutex_lock(&root->fs_info->ordered_operations_mutex); 3473acce952bSliubo spin_lock(&root->fs_info->ordered_extent_lock); 3474acce952bSliubo 3475acce952bSliubo list_splice_init(&root->fs_info->ordered_operations, &splice); 3476acce952bSliubo while (!list_empty(&splice)) { 3477acce952bSliubo btrfs_inode = list_entry(splice.next, struct btrfs_inode, 3478acce952bSliubo ordered_operations); 3479acce952bSliubo 3480acce952bSliubo list_del_init(&btrfs_inode->ordered_operations); 3481acce952bSliubo 3482acce952bSliubo btrfs_invalidate_inodes(btrfs_inode->root); 3483acce952bSliubo } 3484acce952bSliubo 3485acce952bSliubo spin_unlock(&root->fs_info->ordered_extent_lock); 3486acce952bSliubo mutex_unlock(&root->fs_info->ordered_operations_mutex); 3487acce952bSliubo } 3488acce952bSliubo 3489143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 3490acce952bSliubo { 3491acce952bSliubo struct list_head splice; 3492acce952bSliubo struct btrfs_ordered_extent *ordered; 3493acce952bSliubo struct inode *inode; 3494acce952bSliubo 3495acce952bSliubo INIT_LIST_HEAD(&splice); 3496acce952bSliubo 3497acce952bSliubo spin_lock(&root->fs_info->ordered_extent_lock); 3498acce952bSliubo 3499acce952bSliubo list_splice_init(&root->fs_info->ordered_extents, &splice); 3500acce952bSliubo while (!list_empty(&splice)) { 3501acce952bSliubo ordered = list_entry(splice.next, struct btrfs_ordered_extent, 3502acce952bSliubo root_extent_list); 3503acce952bSliubo 3504acce952bSliubo list_del_init(&ordered->root_extent_list); 3505acce952bSliubo atomic_inc(&ordered->refs); 3506acce952bSliubo 3507acce952bSliubo /* the inode may be getting freed (in sys_unlink path). */ 3508acce952bSliubo inode = igrab(ordered->inode); 3509acce952bSliubo 3510acce952bSliubo spin_unlock(&root->fs_info->ordered_extent_lock); 3511acce952bSliubo if (inode) 3512acce952bSliubo iput(inode); 3513acce952bSliubo 3514acce952bSliubo atomic_set(&ordered->refs, 1); 3515acce952bSliubo btrfs_put_ordered_extent(ordered); 3516acce952bSliubo 3517acce952bSliubo spin_lock(&root->fs_info->ordered_extent_lock); 3518acce952bSliubo } 3519acce952bSliubo 3520acce952bSliubo spin_unlock(&root->fs_info->ordered_extent_lock); 3521acce952bSliubo } 3522acce952bSliubo 352379787eaaSJeff Mahoney int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 3524acce952bSliubo struct btrfs_root *root) 3525acce952bSliubo { 3526acce952bSliubo struct rb_node *node; 3527acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 3528acce952bSliubo struct btrfs_delayed_ref_node *ref; 3529acce952bSliubo int ret = 0; 3530acce952bSliubo 3531acce952bSliubo delayed_refs = &trans->delayed_refs; 3532acce952bSliubo 3533acce952bSliubo spin_lock(&delayed_refs->lock); 3534acce952bSliubo if (delayed_refs->num_entries == 0) { 3535cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 3536acce952bSliubo printk(KERN_INFO "delayed_refs has NO entry\n"); 3537acce952bSliubo return ret; 3538acce952bSliubo } 3539acce952bSliubo 3540b939d1abSJosef Bacik while ((node = rb_first(&delayed_refs->root)) != NULL) { 3541acce952bSliubo ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); 3542acce952bSliubo 3543acce952bSliubo atomic_set(&ref->refs, 1); 3544acce952bSliubo if (btrfs_delayed_ref_is_head(ref)) { 3545acce952bSliubo struct btrfs_delayed_ref_head *head; 3546acce952bSliubo 3547acce952bSliubo head = btrfs_delayed_node_to_head(ref); 3548b939d1abSJosef Bacik if (!mutex_trylock(&head->mutex)) { 3549b939d1abSJosef Bacik atomic_inc(&ref->refs); 355079787eaaSJeff Mahoney spin_unlock(&delayed_refs->lock); 3551b939d1abSJosef Bacik 3552b939d1abSJosef Bacik /* Need to wait for the delayed ref to run */ 3553acce952bSliubo mutex_lock(&head->mutex); 3554b939d1abSJosef Bacik mutex_unlock(&head->mutex); 3555b939d1abSJosef Bacik btrfs_put_delayed_ref(ref); 3556b939d1abSJosef Bacik 3557e18fca73SJosef Bacik spin_lock(&delayed_refs->lock); 3558b939d1abSJosef Bacik continue; 3559b939d1abSJosef Bacik } 3560b939d1abSJosef Bacik 3561acce952bSliubo kfree(head->extent_op); 3562acce952bSliubo delayed_refs->num_heads--; 3563acce952bSliubo if (list_empty(&head->cluster)) 3564acce952bSliubo delayed_refs->num_heads_ready--; 3565acce952bSliubo list_del_init(&head->cluster); 3566acce952bSliubo } 3567b939d1abSJosef Bacik ref->in_tree = 0; 3568b939d1abSJosef Bacik rb_erase(&ref->rb_node, &delayed_refs->root); 3569b939d1abSJosef Bacik delayed_refs->num_entries--; 3570b939d1abSJosef Bacik 3571acce952bSliubo spin_unlock(&delayed_refs->lock); 3572acce952bSliubo btrfs_put_delayed_ref(ref); 3573acce952bSliubo 3574acce952bSliubo cond_resched(); 3575acce952bSliubo spin_lock(&delayed_refs->lock); 3576acce952bSliubo } 3577acce952bSliubo 3578acce952bSliubo spin_unlock(&delayed_refs->lock); 3579acce952bSliubo 3580acce952bSliubo return ret; 3581acce952bSliubo } 3582acce952bSliubo 3583143bede5SJeff Mahoney static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t) 3584acce952bSliubo { 3585acce952bSliubo struct btrfs_pending_snapshot *snapshot; 3586acce952bSliubo struct list_head splice; 3587acce952bSliubo 3588acce952bSliubo INIT_LIST_HEAD(&splice); 3589acce952bSliubo 3590acce952bSliubo list_splice_init(&t->pending_snapshots, &splice); 3591acce952bSliubo 3592acce952bSliubo while (!list_empty(&splice)) { 3593acce952bSliubo snapshot = list_entry(splice.next, 3594acce952bSliubo struct btrfs_pending_snapshot, 3595acce952bSliubo list); 3596acce952bSliubo 3597acce952bSliubo list_del_init(&snapshot->list); 3598acce952bSliubo 3599acce952bSliubo kfree(snapshot); 3600acce952bSliubo } 3601acce952bSliubo } 3602acce952bSliubo 3603143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 3604acce952bSliubo { 3605acce952bSliubo struct btrfs_inode *btrfs_inode; 3606acce952bSliubo struct list_head splice; 3607acce952bSliubo 3608acce952bSliubo INIT_LIST_HEAD(&splice); 3609acce952bSliubo 3610acce952bSliubo spin_lock(&root->fs_info->delalloc_lock); 36115be76758SDavid Sterba list_splice_init(&root->fs_info->delalloc_inodes, &splice); 3612acce952bSliubo 3613acce952bSliubo while (!list_empty(&splice)) { 3614acce952bSliubo btrfs_inode = list_entry(splice.next, struct btrfs_inode, 3615acce952bSliubo delalloc_inodes); 3616acce952bSliubo 3617acce952bSliubo list_del_init(&btrfs_inode->delalloc_inodes); 3618acce952bSliubo 3619acce952bSliubo btrfs_invalidate_inodes(btrfs_inode->root); 3620acce952bSliubo } 3621acce952bSliubo 3622acce952bSliubo spin_unlock(&root->fs_info->delalloc_lock); 3623acce952bSliubo } 3624acce952bSliubo 3625acce952bSliubo static int btrfs_destroy_marked_extents(struct btrfs_root *root, 3626acce952bSliubo struct extent_io_tree *dirty_pages, 3627acce952bSliubo int mark) 3628acce952bSliubo { 3629acce952bSliubo int ret; 3630acce952bSliubo struct page *page; 3631acce952bSliubo struct inode *btree_inode = root->fs_info->btree_inode; 3632acce952bSliubo struct extent_buffer *eb; 3633acce952bSliubo u64 start = 0; 3634acce952bSliubo u64 end; 3635acce952bSliubo u64 offset; 3636acce952bSliubo unsigned long index; 3637acce952bSliubo 3638acce952bSliubo while (1) { 3639acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 3640acce952bSliubo mark); 3641acce952bSliubo if (ret) 3642acce952bSliubo break; 3643acce952bSliubo 3644acce952bSliubo clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); 3645acce952bSliubo while (start <= end) { 3646acce952bSliubo index = start >> PAGE_CACHE_SHIFT; 3647acce952bSliubo start = (u64)(index + 1) << PAGE_CACHE_SHIFT; 3648acce952bSliubo page = find_get_page(btree_inode->i_mapping, index); 3649acce952bSliubo if (!page) 3650acce952bSliubo continue; 3651acce952bSliubo offset = page_offset(page); 3652acce952bSliubo 3653acce952bSliubo spin_lock(&dirty_pages->buffer_lock); 3654acce952bSliubo eb = radix_tree_lookup( 3655acce952bSliubo &(&BTRFS_I(page->mapping->host)->io_tree)->buffer, 3656acce952bSliubo offset >> PAGE_CACHE_SHIFT); 3657acce952bSliubo spin_unlock(&dirty_pages->buffer_lock); 3658ee670f0aSJosef Bacik if (eb) 3659acce952bSliubo ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY, 3660acce952bSliubo &eb->bflags); 3661acce952bSliubo if (PageWriteback(page)) 3662acce952bSliubo end_page_writeback(page); 3663acce952bSliubo 3664acce952bSliubo lock_page(page); 3665acce952bSliubo if (PageDirty(page)) { 3666acce952bSliubo clear_page_dirty_for_io(page); 3667acce952bSliubo spin_lock_irq(&page->mapping->tree_lock); 3668acce952bSliubo radix_tree_tag_clear(&page->mapping->page_tree, 3669acce952bSliubo page_index(page), 3670acce952bSliubo PAGECACHE_TAG_DIRTY); 3671acce952bSliubo spin_unlock_irq(&page->mapping->tree_lock); 3672acce952bSliubo } 3673acce952bSliubo 3674acce952bSliubo unlock_page(page); 3675ee670f0aSJosef Bacik page_cache_release(page); 3676acce952bSliubo } 3677acce952bSliubo } 3678acce952bSliubo 3679acce952bSliubo return ret; 3680acce952bSliubo } 3681acce952bSliubo 3682acce952bSliubo static int btrfs_destroy_pinned_extent(struct btrfs_root *root, 3683acce952bSliubo struct extent_io_tree *pinned_extents) 3684acce952bSliubo { 3685acce952bSliubo struct extent_io_tree *unpin; 3686acce952bSliubo u64 start; 3687acce952bSliubo u64 end; 3688acce952bSliubo int ret; 3689ed0eaa14SLiu Bo bool loop = true; 3690acce952bSliubo 3691acce952bSliubo unpin = pinned_extents; 3692ed0eaa14SLiu Bo again: 3693acce952bSliubo while (1) { 3694acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 3695acce952bSliubo EXTENT_DIRTY); 3696acce952bSliubo if (ret) 3697acce952bSliubo break; 3698acce952bSliubo 3699acce952bSliubo /* opt_discard */ 37005378e607SLi Dongyang if (btrfs_test_opt(root, DISCARD)) 37015378e607SLi Dongyang ret = btrfs_error_discard_extent(root, start, 37025378e607SLi Dongyang end + 1 - start, 37035378e607SLi Dongyang NULL); 3704acce952bSliubo 3705acce952bSliubo clear_extent_dirty(unpin, start, end, GFP_NOFS); 3706acce952bSliubo btrfs_error_unpin_extent_range(root, start, end); 3707acce952bSliubo cond_resched(); 3708acce952bSliubo } 3709acce952bSliubo 3710ed0eaa14SLiu Bo if (loop) { 3711ed0eaa14SLiu Bo if (unpin == &root->fs_info->freed_extents[0]) 3712ed0eaa14SLiu Bo unpin = &root->fs_info->freed_extents[1]; 3713ed0eaa14SLiu Bo else 3714ed0eaa14SLiu Bo unpin = &root->fs_info->freed_extents[0]; 3715ed0eaa14SLiu Bo loop = false; 3716ed0eaa14SLiu Bo goto again; 3717ed0eaa14SLiu Bo } 3718ed0eaa14SLiu Bo 3719acce952bSliubo return 0; 3720acce952bSliubo } 3721acce952bSliubo 372249b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 372349b25e05SJeff Mahoney struct btrfs_root *root) 372449b25e05SJeff Mahoney { 372549b25e05SJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, root); 372649b25e05SJeff Mahoney btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv, 372749b25e05SJeff Mahoney cur_trans->dirty_pages.dirty_bytes); 372849b25e05SJeff Mahoney 372949b25e05SJeff Mahoney /* FIXME: cleanup wait for commit */ 373049b25e05SJeff Mahoney cur_trans->in_commit = 1; 373149b25e05SJeff Mahoney cur_trans->blocked = 1; 373249b25e05SJeff Mahoney wake_up(&root->fs_info->transaction_blocked_wait); 373349b25e05SJeff Mahoney 373449b25e05SJeff Mahoney cur_trans->blocked = 0; 373549b25e05SJeff Mahoney wake_up(&root->fs_info->transaction_wait); 373649b25e05SJeff Mahoney 373749b25e05SJeff Mahoney cur_trans->commit_done = 1; 373849b25e05SJeff Mahoney wake_up(&cur_trans->commit_wait); 373949b25e05SJeff Mahoney 374067cde344SMiao Xie btrfs_destroy_delayed_inodes(root); 374167cde344SMiao Xie btrfs_assert_delayed_root_empty(root); 374267cde344SMiao Xie 374349b25e05SJeff Mahoney btrfs_destroy_pending_snapshots(cur_trans); 374449b25e05SJeff Mahoney 374549b25e05SJeff Mahoney btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages, 374649b25e05SJeff Mahoney EXTENT_DIRTY); 37476e841e32SLiu Bo btrfs_destroy_pinned_extent(root, 37486e841e32SLiu Bo root->fs_info->pinned_extents); 374949b25e05SJeff Mahoney 375049b25e05SJeff Mahoney /* 375149b25e05SJeff Mahoney memset(cur_trans, 0, sizeof(*cur_trans)); 375249b25e05SJeff Mahoney kmem_cache_free(btrfs_transaction_cachep, cur_trans); 375349b25e05SJeff Mahoney */ 375449b25e05SJeff Mahoney } 375549b25e05SJeff Mahoney 375649b25e05SJeff Mahoney int btrfs_cleanup_transaction(struct btrfs_root *root) 3757acce952bSliubo { 3758acce952bSliubo struct btrfs_transaction *t; 3759acce952bSliubo LIST_HEAD(list); 3760acce952bSliubo 3761acce952bSliubo mutex_lock(&root->fs_info->transaction_kthread_mutex); 3762acce952bSliubo 3763a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 3764acce952bSliubo list_splice_init(&root->fs_info->trans_list, &list); 3765a4abeea4SJosef Bacik root->fs_info->trans_no_join = 1; 3766a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 3767a4abeea4SJosef Bacik 3768acce952bSliubo while (!list_empty(&list)) { 3769acce952bSliubo t = list_entry(list.next, struct btrfs_transaction, list); 3770acce952bSliubo if (!t) 3771acce952bSliubo break; 3772acce952bSliubo 3773acce952bSliubo btrfs_destroy_ordered_operations(root); 3774acce952bSliubo 3775acce952bSliubo btrfs_destroy_ordered_extents(root); 3776acce952bSliubo 3777acce952bSliubo btrfs_destroy_delayed_refs(t, root); 3778acce952bSliubo 3779acce952bSliubo btrfs_block_rsv_release(root, 3780acce952bSliubo &root->fs_info->trans_block_rsv, 3781acce952bSliubo t->dirty_pages.dirty_bytes); 3782acce952bSliubo 3783acce952bSliubo /* FIXME: cleanup wait for commit */ 3784acce952bSliubo t->in_commit = 1; 3785acce952bSliubo t->blocked = 1; 3786acce952bSliubo if (waitqueue_active(&root->fs_info->transaction_blocked_wait)) 3787acce952bSliubo wake_up(&root->fs_info->transaction_blocked_wait); 3788acce952bSliubo 3789acce952bSliubo t->blocked = 0; 3790acce952bSliubo if (waitqueue_active(&root->fs_info->transaction_wait)) 3791acce952bSliubo wake_up(&root->fs_info->transaction_wait); 3792acce952bSliubo 3793acce952bSliubo t->commit_done = 1; 3794acce952bSliubo if (waitqueue_active(&t->commit_wait)) 3795acce952bSliubo wake_up(&t->commit_wait); 3796acce952bSliubo 379767cde344SMiao Xie btrfs_destroy_delayed_inodes(root); 379867cde344SMiao Xie btrfs_assert_delayed_root_empty(root); 379967cde344SMiao Xie 3800acce952bSliubo btrfs_destroy_pending_snapshots(t); 3801acce952bSliubo 3802acce952bSliubo btrfs_destroy_delalloc_inodes(root); 3803acce952bSliubo 3804a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 3805acce952bSliubo root->fs_info->running_transaction = NULL; 3806a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 3807acce952bSliubo 3808acce952bSliubo btrfs_destroy_marked_extents(root, &t->dirty_pages, 3809acce952bSliubo EXTENT_DIRTY); 3810acce952bSliubo 3811acce952bSliubo btrfs_destroy_pinned_extent(root, 3812acce952bSliubo root->fs_info->pinned_extents); 3813acce952bSliubo 381413c5a93eSJosef Bacik atomic_set(&t->use_count, 0); 3815acce952bSliubo list_del_init(&t->list); 3816acce952bSliubo memset(t, 0, sizeof(*t)); 3817acce952bSliubo kmem_cache_free(btrfs_transaction_cachep, t); 3818acce952bSliubo } 3819acce952bSliubo 3820a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 3821a4abeea4SJosef Bacik root->fs_info->trans_no_join = 0; 3822a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 3823acce952bSliubo mutex_unlock(&root->fs_info->transaction_kthread_mutex); 3824acce952bSliubo 3825acce952bSliubo return 0; 3826acce952bSliubo } 3827acce952bSliubo 3828d1310b2eSChris Mason static struct extent_io_ops btree_extent_io_ops = { 38294bef0848SChris Mason .write_cache_pages_lock_hook = btree_lock_page_hook, 3830ce9adaa5SChris Mason .readpage_end_io_hook = btree_readpage_end_io_hook, 38314bb31e92SArne Jansen .readpage_io_failed_hook = btree_io_failed_hook, 38320b86a832SChris Mason .submit_bio_hook = btree_submit_bio_hook, 3833239b14b3SChris Mason /* note we're sharing with inode.c for the merge bio hook */ 3834239b14b3SChris Mason .merge_bio_hook = btrfs_merge_bio_hook, 38350da5468fSChris Mason }; 3836