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" 488dabb742SStefan Behrens #include "dev-replace.h" 49eb60ceacSChris Mason 50de0022b9SJosef Bacik #ifdef CONFIG_X86 51de0022b9SJosef Bacik #include <asm/cpufeature.h> 52de0022b9SJosef Bacik #endif 53de0022b9SJosef Bacik 54d1310b2eSChris Mason static struct extent_io_ops btree_extent_io_ops; 558b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work); 564df27c4dSYan, Zheng static void free_fs_root(struct btrfs_root *root); 57fcd1f065SDavid Sterba static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info, 58acce952bSliubo int read_only); 59143bede5SJeff Mahoney static void btrfs_destroy_ordered_operations(struct btrfs_root *root); 60143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root); 61acce952bSliubo static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 62acce952bSliubo struct btrfs_root *root); 63143bede5SJeff Mahoney static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t); 64143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root); 65acce952bSliubo static int btrfs_destroy_marked_extents(struct btrfs_root *root, 66acce952bSliubo struct extent_io_tree *dirty_pages, 67acce952bSliubo int mark); 68acce952bSliubo static int btrfs_destroy_pinned_extent(struct btrfs_root *root, 69acce952bSliubo struct extent_io_tree *pinned_extents); 70ce9adaa5SChris Mason 71d352ac68SChris Mason /* 72d352ac68SChris Mason * end_io_wq structs are used to do processing in task context when an IO is 73d352ac68SChris Mason * complete. This is used during reads to verify checksums, and it is used 74d352ac68SChris Mason * by writes to insert metadata for new file extents after IO is complete. 75d352ac68SChris Mason */ 76ce9adaa5SChris Mason struct end_io_wq { 77ce9adaa5SChris Mason struct bio *bio; 78ce9adaa5SChris Mason bio_end_io_t *end_io; 79ce9adaa5SChris Mason void *private; 80ce9adaa5SChris Mason struct btrfs_fs_info *info; 81ce9adaa5SChris Mason int error; 8222c59948SChris Mason int metadata; 83ce9adaa5SChris Mason struct list_head list; 848b712842SChris Mason struct btrfs_work work; 85ce9adaa5SChris Mason }; 860da5468fSChris Mason 87d352ac68SChris Mason /* 88d352ac68SChris Mason * async submit bios are used to offload expensive checksumming 89d352ac68SChris Mason * onto the worker threads. They checksum file and metadata bios 90d352ac68SChris Mason * just before they are sent down the IO stack. 91d352ac68SChris Mason */ 9244b8bd7eSChris Mason struct async_submit_bio { 9344b8bd7eSChris Mason struct inode *inode; 9444b8bd7eSChris Mason struct bio *bio; 9544b8bd7eSChris Mason struct list_head list; 964a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_start; 974a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_done; 9844b8bd7eSChris Mason int rw; 9944b8bd7eSChris Mason int mirror_num; 100c8b97818SChris Mason unsigned long bio_flags; 101eaf25d93SChris Mason /* 102eaf25d93SChris Mason * bio_offset is optional, can be used if the pages in the bio 103eaf25d93SChris Mason * can't tell us where in the file the bio should go 104eaf25d93SChris Mason */ 105eaf25d93SChris Mason u64 bio_offset; 1068b712842SChris Mason struct btrfs_work work; 10779787eaaSJeff Mahoney int error; 10844b8bd7eSChris Mason }; 10944b8bd7eSChris Mason 11085d4e461SChris Mason /* 11185d4e461SChris Mason * Lockdep class keys for extent_buffer->lock's in this root. For a given 11285d4e461SChris Mason * eb, the lockdep key is determined by the btrfs_root it belongs to and 11385d4e461SChris Mason * the level the eb occupies in the tree. 1144008c04aSChris Mason * 11585d4e461SChris Mason * Different roots are used for different purposes and may nest inside each 11685d4e461SChris Mason * other and they require separate keysets. As lockdep keys should be 11785d4e461SChris Mason * static, assign keysets according to the purpose of the root as indicated 11885d4e461SChris Mason * by btrfs_root->objectid. This ensures that all special purpose roots 11985d4e461SChris Mason * have separate keysets. 1204008c04aSChris Mason * 12185d4e461SChris Mason * Lock-nesting across peer nodes is always done with the immediate parent 12285d4e461SChris Mason * node locked thus preventing deadlock. As lockdep doesn't know this, use 12385d4e461SChris Mason * subclass to avoid triggering lockdep warning in such cases. 1244008c04aSChris Mason * 12585d4e461SChris Mason * The key is set by the readpage_end_io_hook after the buffer has passed 12685d4e461SChris Mason * csum validation but before the pages are unlocked. It is also set by 12785d4e461SChris Mason * btrfs_init_new_buffer on freshly allocated blocks. 12885d4e461SChris Mason * 12985d4e461SChris Mason * We also add a check to make sure the highest level of the tree is the 13085d4e461SChris Mason * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code 13185d4e461SChris Mason * needs update as well. 1324008c04aSChris Mason */ 1334008c04aSChris Mason #ifdef CONFIG_DEBUG_LOCK_ALLOC 1344008c04aSChris Mason # if BTRFS_MAX_LEVEL != 8 1354008c04aSChris Mason # error 1364008c04aSChris Mason # endif 13785d4e461SChris Mason 13885d4e461SChris Mason static struct btrfs_lockdep_keyset { 13985d4e461SChris Mason u64 id; /* root objectid */ 14085d4e461SChris Mason const char *name_stem; /* lock name stem */ 14185d4e461SChris Mason char names[BTRFS_MAX_LEVEL + 1][20]; 14285d4e461SChris Mason struct lock_class_key keys[BTRFS_MAX_LEVEL + 1]; 14385d4e461SChris Mason } btrfs_lockdep_keysets[] = { 14485d4e461SChris Mason { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" }, 14585d4e461SChris Mason { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" }, 14685d4e461SChris Mason { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" }, 14785d4e461SChris Mason { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" }, 14885d4e461SChris Mason { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" }, 14985d4e461SChris Mason { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" }, 15085d4e461SChris Mason { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" }, 15185d4e461SChris Mason { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" }, 15285d4e461SChris Mason { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" }, 15385d4e461SChris Mason { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" }, 15485d4e461SChris Mason { .id = 0, .name_stem = "tree" }, 1554008c04aSChris Mason }; 15685d4e461SChris Mason 15785d4e461SChris Mason void __init btrfs_init_lockdep(void) 15885d4e461SChris Mason { 15985d4e461SChris Mason int i, j; 16085d4e461SChris Mason 16185d4e461SChris Mason /* initialize lockdep class names */ 16285d4e461SChris Mason for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) { 16385d4e461SChris Mason struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i]; 16485d4e461SChris Mason 16585d4e461SChris Mason for (j = 0; j < ARRAY_SIZE(ks->names); j++) 16685d4e461SChris Mason snprintf(ks->names[j], sizeof(ks->names[j]), 16785d4e461SChris Mason "btrfs-%s-%02d", ks->name_stem, j); 16885d4e461SChris Mason } 16985d4e461SChris Mason } 17085d4e461SChris Mason 17185d4e461SChris Mason void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb, 17285d4e461SChris Mason int level) 17385d4e461SChris Mason { 17485d4e461SChris Mason struct btrfs_lockdep_keyset *ks; 17585d4e461SChris Mason 17685d4e461SChris Mason BUG_ON(level >= ARRAY_SIZE(ks->keys)); 17785d4e461SChris Mason 17885d4e461SChris Mason /* find the matching keyset, id 0 is the default entry */ 17985d4e461SChris Mason for (ks = btrfs_lockdep_keysets; ks->id; ks++) 18085d4e461SChris Mason if (ks->id == objectid) 18185d4e461SChris Mason break; 18285d4e461SChris Mason 18385d4e461SChris Mason lockdep_set_class_and_name(&eb->lock, 18485d4e461SChris Mason &ks->keys[level], ks->names[level]); 18585d4e461SChris Mason } 18685d4e461SChris Mason 1874008c04aSChris Mason #endif 1884008c04aSChris Mason 189d352ac68SChris Mason /* 190d352ac68SChris Mason * extents on the btree inode are pretty simple, there's one extent 191d352ac68SChris Mason * that covers the entire device 192d352ac68SChris Mason */ 193b2950863SChristoph Hellwig static struct extent_map *btree_get_extent(struct inode *inode, 194306e16ceSDavid Sterba struct page *page, size_t pg_offset, u64 start, u64 len, 1955f39d397SChris Mason int create) 1965f39d397SChris Mason { 1975f39d397SChris Mason struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 1985f39d397SChris Mason struct extent_map *em; 1995f39d397SChris Mason int ret; 2005f39d397SChris Mason 201890871beSChris Mason read_lock(&em_tree->lock); 202d1310b2eSChris Mason em = lookup_extent_mapping(em_tree, start, len); 203a061fc8dSChris Mason if (em) { 204a061fc8dSChris Mason em->bdev = 205a061fc8dSChris Mason BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 206890871beSChris Mason read_unlock(&em_tree->lock); 2075f39d397SChris Mason goto out; 208a061fc8dSChris Mason } 209890871beSChris Mason read_unlock(&em_tree->lock); 2107b13b7b1SChris Mason 211172ddd60SDavid Sterba em = alloc_extent_map(); 2125f39d397SChris Mason if (!em) { 2135f39d397SChris Mason em = ERR_PTR(-ENOMEM); 2145f39d397SChris Mason goto out; 2155f39d397SChris Mason } 2165f39d397SChris Mason em->start = 0; 2170afbaf8cSChris Mason em->len = (u64)-1; 218c8b97818SChris Mason em->block_len = (u64)-1; 2195f39d397SChris Mason em->block_start = 0; 220a061fc8dSChris Mason em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 221d1310b2eSChris Mason 222890871beSChris Mason write_lock(&em_tree->lock); 2235f39d397SChris Mason ret = add_extent_mapping(em_tree, em); 2245f39d397SChris Mason if (ret == -EEXIST) { 2255f39d397SChris Mason free_extent_map(em); 2267b13b7b1SChris Mason em = lookup_extent_mapping(em_tree, start, len); 227b4f359abSTsutomu Itoh if (!em) 2280433f20dSTsutomu Itoh em = ERR_PTR(-EIO); 2295f39d397SChris Mason } else if (ret) { 2307b13b7b1SChris Mason free_extent_map(em); 2310433f20dSTsutomu Itoh em = ERR_PTR(ret); 2325f39d397SChris Mason } 233890871beSChris Mason write_unlock(&em_tree->lock); 2347b13b7b1SChris Mason 2355f39d397SChris Mason out: 2365f39d397SChris Mason return em; 2375f39d397SChris Mason } 2385f39d397SChris Mason 23919c00ddcSChris Mason u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len) 24019c00ddcSChris Mason { 241163e783eSDavid Woodhouse return crc32c(seed, data, len); 24219c00ddcSChris Mason } 24319c00ddcSChris Mason 24419c00ddcSChris Mason void btrfs_csum_final(u32 crc, char *result) 24519c00ddcSChris Mason { 2467e75bf3fSDavid Sterba put_unaligned_le32(~crc, result); 24719c00ddcSChris Mason } 24819c00ddcSChris Mason 249d352ac68SChris Mason /* 250d352ac68SChris Mason * compute the csum for a btree block, and either verify it or write it 251d352ac68SChris Mason * into the csum field of the block. 252d352ac68SChris Mason */ 25319c00ddcSChris Mason static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf, 25419c00ddcSChris Mason int verify) 25519c00ddcSChris Mason { 2566c41761fSDavid Sterba u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy); 257607d432dSJosef Bacik char *result = NULL; 25819c00ddcSChris Mason unsigned long len; 25919c00ddcSChris Mason unsigned long cur_len; 26019c00ddcSChris Mason unsigned long offset = BTRFS_CSUM_SIZE; 26119c00ddcSChris Mason char *kaddr; 26219c00ddcSChris Mason unsigned long map_start; 26319c00ddcSChris Mason unsigned long map_len; 26419c00ddcSChris Mason int err; 26519c00ddcSChris Mason u32 crc = ~(u32)0; 266607d432dSJosef Bacik unsigned long inline_result; 26719c00ddcSChris Mason 26819c00ddcSChris Mason len = buf->len - offset; 26919c00ddcSChris Mason while (len > 0) { 27019c00ddcSChris Mason err = map_private_extent_buffer(buf, offset, 32, 271a6591715SChris Mason &kaddr, &map_start, &map_len); 272d397712bSChris Mason if (err) 27319c00ddcSChris Mason return 1; 27419c00ddcSChris Mason cur_len = min(len, map_len - (offset - map_start)); 27519c00ddcSChris Mason crc = btrfs_csum_data(root, kaddr + offset - map_start, 27619c00ddcSChris Mason crc, cur_len); 27719c00ddcSChris Mason len -= cur_len; 27819c00ddcSChris Mason offset += cur_len; 27919c00ddcSChris Mason } 280607d432dSJosef Bacik if (csum_size > sizeof(inline_result)) { 281607d432dSJosef Bacik result = kzalloc(csum_size * sizeof(char), GFP_NOFS); 282607d432dSJosef Bacik if (!result) 283607d432dSJosef Bacik return 1; 284607d432dSJosef Bacik } else { 285607d432dSJosef Bacik result = (char *)&inline_result; 286607d432dSJosef Bacik } 287607d432dSJosef Bacik 28819c00ddcSChris Mason btrfs_csum_final(crc, result); 28919c00ddcSChris Mason 29019c00ddcSChris Mason if (verify) { 291607d432dSJosef Bacik if (memcmp_extent_buffer(buf, result, 0, csum_size)) { 292e4204dedSChris Mason u32 val; 293e4204dedSChris Mason u32 found = 0; 294607d432dSJosef Bacik memcpy(&found, result, csum_size); 295e4204dedSChris Mason 296607d432dSJosef Bacik read_extent_buffer(buf, &val, 0, csum_size); 2977a36ddecSDavid Sterba printk_ratelimited(KERN_INFO "btrfs: %s checksum verify " 298193f284dSChris Mason "failed on %llu wanted %X found %X " 299193f284dSChris Mason "level %d\n", 30019c00ddcSChris Mason root->fs_info->sb->s_id, 301193f284dSChris Mason (unsigned long long)buf->start, val, found, 302193f284dSChris Mason btrfs_header_level(buf)); 303607d432dSJosef Bacik if (result != (char *)&inline_result) 304607d432dSJosef Bacik kfree(result); 30519c00ddcSChris Mason return 1; 30619c00ddcSChris Mason } 30719c00ddcSChris Mason } else { 308607d432dSJosef Bacik write_extent_buffer(buf, result, 0, csum_size); 30919c00ddcSChris Mason } 310607d432dSJosef Bacik if (result != (char *)&inline_result) 311607d432dSJosef Bacik kfree(result); 31219c00ddcSChris Mason return 0; 31319c00ddcSChris Mason } 31419c00ddcSChris Mason 315d352ac68SChris Mason /* 316d352ac68SChris Mason * we can't consider a given block up to date unless the transid of the 317d352ac68SChris Mason * block matches the transid in the parent node's pointer. This is how we 318d352ac68SChris Mason * detect blocks that either didn't get written at all or got written 319d352ac68SChris Mason * in the wrong place. 320d352ac68SChris Mason */ 3211259ab75SChris Mason static int verify_parent_transid(struct extent_io_tree *io_tree, 322b9fab919SChris Mason struct extent_buffer *eb, u64 parent_transid, 323b9fab919SChris Mason int atomic) 3241259ab75SChris Mason { 3252ac55d41SJosef Bacik struct extent_state *cached_state = NULL; 3261259ab75SChris Mason int ret; 3271259ab75SChris Mason 3281259ab75SChris Mason if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 3291259ab75SChris Mason return 0; 3301259ab75SChris Mason 331b9fab919SChris Mason if (atomic) 332b9fab919SChris Mason return -EAGAIN; 333b9fab919SChris Mason 3342ac55d41SJosef Bacik lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 335d0082371SJeff Mahoney 0, &cached_state); 3360b32f4bbSJosef Bacik if (extent_buffer_uptodate(eb) && 3371259ab75SChris Mason btrfs_header_generation(eb) == parent_transid) { 3381259ab75SChris Mason ret = 0; 3391259ab75SChris Mason goto out; 3401259ab75SChris Mason } 3417a36ddecSDavid Sterba printk_ratelimited("parent transid verify failed on %llu wanted %llu " 342193f284dSChris Mason "found %llu\n", 3431259ab75SChris Mason (unsigned long long)eb->start, 3441259ab75SChris Mason (unsigned long long)parent_transid, 3451259ab75SChris Mason (unsigned long long)btrfs_header_generation(eb)); 3461259ab75SChris Mason ret = 1; 3470b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 34833958dc6SChris Mason out: 3492ac55d41SJosef Bacik unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 3502ac55d41SJosef Bacik &cached_state, GFP_NOFS); 3511259ab75SChris Mason return ret; 3521259ab75SChris Mason } 3531259ab75SChris Mason 354d352ac68SChris Mason /* 355d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 356d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 357d352ac68SChris Mason */ 358f188591eSChris Mason static int btree_read_extent_buffer_pages(struct btrfs_root *root, 359f188591eSChris Mason struct extent_buffer *eb, 360ca7a79adSChris Mason u64 start, u64 parent_transid) 361f188591eSChris Mason { 362f188591eSChris Mason struct extent_io_tree *io_tree; 363ea466794SJosef Bacik int failed = 0; 364f188591eSChris Mason int ret; 365f188591eSChris Mason int num_copies = 0; 366f188591eSChris Mason int mirror_num = 0; 367ea466794SJosef Bacik int failed_mirror = 0; 368f188591eSChris Mason 369a826d6dcSJosef Bacik clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 370f188591eSChris Mason io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree; 371f188591eSChris Mason while (1) { 372bb82ab88SArne Jansen ret = read_extent_buffer_pages(io_tree, eb, start, 373bb82ab88SArne Jansen WAIT_COMPLETE, 374f188591eSChris Mason btree_get_extent, mirror_num); 375256dd1bbSStefan Behrens if (!ret) { 376256dd1bbSStefan Behrens if (!verify_parent_transid(io_tree, eb, 377b9fab919SChris Mason parent_transid, 0)) 378ea466794SJosef Bacik break; 379256dd1bbSStefan Behrens else 380256dd1bbSStefan Behrens ret = -EIO; 381256dd1bbSStefan Behrens } 382d397712bSChris Mason 383a826d6dcSJosef Bacik /* 384a826d6dcSJosef Bacik * This buffer's crc is fine, but its contents are corrupted, so 385a826d6dcSJosef Bacik * there is no reason to read the other copies, they won't be 386a826d6dcSJosef Bacik * any less wrong. 387a826d6dcSJosef Bacik */ 388a826d6dcSJosef Bacik if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags)) 389ea466794SJosef Bacik break; 390ea466794SJosef Bacik 3915d964051SStefan Behrens num_copies = btrfs_num_copies(root->fs_info, 392f188591eSChris Mason eb->start, eb->len); 3934235298eSChris Mason if (num_copies == 1) 394ea466794SJosef Bacik break; 3954235298eSChris Mason 3965cf1ab56SJosef Bacik if (!failed_mirror) { 3975cf1ab56SJosef Bacik failed = 1; 3985cf1ab56SJosef Bacik failed_mirror = eb->read_mirror; 3995cf1ab56SJosef Bacik } 4005cf1ab56SJosef Bacik 401f188591eSChris Mason mirror_num++; 402ea466794SJosef Bacik if (mirror_num == failed_mirror) 403ea466794SJosef Bacik mirror_num++; 404ea466794SJosef Bacik 4054235298eSChris Mason if (mirror_num > num_copies) 406ea466794SJosef Bacik break; 407f188591eSChris Mason } 408ea466794SJosef Bacik 409c0901581SStefan Behrens if (failed && !ret && failed_mirror) 410ea466794SJosef Bacik repair_eb_io_failure(root, eb, failed_mirror); 411ea466794SJosef Bacik 412ea466794SJosef Bacik return ret; 413f188591eSChris Mason } 41419c00ddcSChris Mason 415d352ac68SChris Mason /* 416d397712bSChris Mason * checksum a dirty tree block before IO. This has extra checks to make sure 417d397712bSChris Mason * we only fill in the checksum field in the first page of a multi-page block 418d352ac68SChris Mason */ 419d397712bSChris Mason 420b2950863SChristoph Hellwig static int csum_dirty_buffer(struct btrfs_root *root, struct page *page) 42119c00ddcSChris Mason { 422d1310b2eSChris Mason struct extent_io_tree *tree; 4234eee4fa4SMiao Xie u64 start = page_offset(page); 42419c00ddcSChris Mason u64 found_start; 42519c00ddcSChris Mason struct extent_buffer *eb; 426f188591eSChris Mason 427d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 42819c00ddcSChris Mason 4294f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 4304f2de97aSJosef Bacik if (page != eb->pages[0]) 4314f2de97aSJosef Bacik return 0; 43219c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 43319c00ddcSChris Mason if (found_start != start) { 43455c69072SChris Mason WARN_ON(1); 4354f2de97aSJosef Bacik return 0; 43655c69072SChris Mason } 43755c69072SChris Mason if (!PageUptodate(page)) { 43855c69072SChris Mason WARN_ON(1); 4394f2de97aSJosef Bacik return 0; 44019c00ddcSChris Mason } 44119c00ddcSChris Mason csum_tree_block(root, eb, 0); 44219c00ddcSChris Mason return 0; 44319c00ddcSChris Mason } 44419c00ddcSChris Mason 4452b82032cSYan Zheng static int check_tree_block_fsid(struct btrfs_root *root, 4462b82032cSYan Zheng struct extent_buffer *eb) 4472b82032cSYan Zheng { 4482b82032cSYan Zheng struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; 4492b82032cSYan Zheng u8 fsid[BTRFS_UUID_SIZE]; 4502b82032cSYan Zheng int ret = 1; 4512b82032cSYan Zheng 4522b82032cSYan Zheng read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb), 4532b82032cSYan Zheng BTRFS_FSID_SIZE); 4542b82032cSYan Zheng while (fs_devices) { 4552b82032cSYan Zheng if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) { 4562b82032cSYan Zheng ret = 0; 4572b82032cSYan Zheng break; 4582b82032cSYan Zheng } 4592b82032cSYan Zheng fs_devices = fs_devices->seed; 4602b82032cSYan Zheng } 4612b82032cSYan Zheng return ret; 4622b82032cSYan Zheng } 4632b82032cSYan Zheng 464a826d6dcSJosef Bacik #define CORRUPT(reason, eb, root, slot) \ 465a826d6dcSJosef Bacik printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \ 466a826d6dcSJosef Bacik "root=%llu, slot=%d\n", reason, \ 467a826d6dcSJosef Bacik (unsigned long long)btrfs_header_bytenr(eb), \ 468a826d6dcSJosef Bacik (unsigned long long)root->objectid, slot) 469a826d6dcSJosef Bacik 470a826d6dcSJosef Bacik static noinline int check_leaf(struct btrfs_root *root, 471a826d6dcSJosef Bacik struct extent_buffer *leaf) 472a826d6dcSJosef Bacik { 473a826d6dcSJosef Bacik struct btrfs_key key; 474a826d6dcSJosef Bacik struct btrfs_key leaf_key; 475a826d6dcSJosef Bacik u32 nritems = btrfs_header_nritems(leaf); 476a826d6dcSJosef Bacik int slot; 477a826d6dcSJosef Bacik 478a826d6dcSJosef Bacik if (nritems == 0) 479a826d6dcSJosef Bacik return 0; 480a826d6dcSJosef Bacik 481a826d6dcSJosef Bacik /* Check the 0 item */ 482a826d6dcSJosef Bacik if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) != 483a826d6dcSJosef Bacik BTRFS_LEAF_DATA_SIZE(root)) { 484a826d6dcSJosef Bacik CORRUPT("invalid item offset size pair", leaf, root, 0); 485a826d6dcSJosef Bacik return -EIO; 486a826d6dcSJosef Bacik } 487a826d6dcSJosef Bacik 488a826d6dcSJosef Bacik /* 489a826d6dcSJosef Bacik * Check to make sure each items keys are in the correct order and their 490a826d6dcSJosef Bacik * offsets make sense. We only have to loop through nritems-1 because 491a826d6dcSJosef Bacik * we check the current slot against the next slot, which verifies the 492a826d6dcSJosef Bacik * next slot's offset+size makes sense and that the current's slot 493a826d6dcSJosef Bacik * offset is correct. 494a826d6dcSJosef Bacik */ 495a826d6dcSJosef Bacik for (slot = 0; slot < nritems - 1; slot++) { 496a826d6dcSJosef Bacik btrfs_item_key_to_cpu(leaf, &leaf_key, slot); 497a826d6dcSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, slot + 1); 498a826d6dcSJosef Bacik 499a826d6dcSJosef Bacik /* Make sure the keys are in the right order */ 500a826d6dcSJosef Bacik if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) { 501a826d6dcSJosef Bacik CORRUPT("bad key order", leaf, root, slot); 502a826d6dcSJosef Bacik return -EIO; 503a826d6dcSJosef Bacik } 504a826d6dcSJosef Bacik 505a826d6dcSJosef Bacik /* 506a826d6dcSJosef Bacik * Make sure the offset and ends are right, remember that the 507a826d6dcSJosef Bacik * item data starts at the end of the leaf and grows towards the 508a826d6dcSJosef Bacik * front. 509a826d6dcSJosef Bacik */ 510a826d6dcSJosef Bacik if (btrfs_item_offset_nr(leaf, slot) != 511a826d6dcSJosef Bacik btrfs_item_end_nr(leaf, slot + 1)) { 512a826d6dcSJosef Bacik CORRUPT("slot offset bad", leaf, root, slot); 513a826d6dcSJosef Bacik return -EIO; 514a826d6dcSJosef Bacik } 515a826d6dcSJosef Bacik 516a826d6dcSJosef Bacik /* 517a826d6dcSJosef Bacik * Check to make sure that we don't point outside of the leaf, 518a826d6dcSJosef Bacik * just incase all the items are consistent to eachother, but 519a826d6dcSJosef Bacik * all point outside of the leaf. 520a826d6dcSJosef Bacik */ 521a826d6dcSJosef Bacik if (btrfs_item_end_nr(leaf, slot) > 522a826d6dcSJosef Bacik BTRFS_LEAF_DATA_SIZE(root)) { 523a826d6dcSJosef Bacik CORRUPT("slot end outside of leaf", leaf, root, slot); 524a826d6dcSJosef Bacik return -EIO; 525a826d6dcSJosef Bacik } 526a826d6dcSJosef Bacik } 527a826d6dcSJosef Bacik 528a826d6dcSJosef Bacik return 0; 529a826d6dcSJosef Bacik } 530a826d6dcSJosef Bacik 531727011e0SChris Mason struct extent_buffer *find_eb_for_page(struct extent_io_tree *tree, 532727011e0SChris Mason struct page *page, int max_walk) 533727011e0SChris Mason { 534727011e0SChris Mason struct extent_buffer *eb; 535727011e0SChris Mason u64 start = page_offset(page); 536727011e0SChris Mason u64 target = start; 537727011e0SChris Mason u64 min_start; 538727011e0SChris Mason 539727011e0SChris Mason if (start < max_walk) 540727011e0SChris Mason min_start = 0; 541727011e0SChris Mason else 542727011e0SChris Mason min_start = start - max_walk; 543727011e0SChris Mason 544727011e0SChris Mason while (start >= min_start) { 545727011e0SChris Mason eb = find_extent_buffer(tree, start, 0); 546727011e0SChris Mason if (eb) { 547727011e0SChris Mason /* 548727011e0SChris Mason * we found an extent buffer and it contains our page 549727011e0SChris Mason * horray! 550727011e0SChris Mason */ 551727011e0SChris Mason if (eb->start <= target && 552727011e0SChris Mason eb->start + eb->len > target) 553727011e0SChris Mason return eb; 554727011e0SChris Mason 555727011e0SChris Mason /* we found an extent buffer that wasn't for us */ 556727011e0SChris Mason free_extent_buffer(eb); 557727011e0SChris Mason return NULL; 558727011e0SChris Mason } 559727011e0SChris Mason if (start == 0) 560727011e0SChris Mason break; 561727011e0SChris Mason start -= PAGE_CACHE_SIZE; 562727011e0SChris Mason } 563727011e0SChris Mason return NULL; 564727011e0SChris Mason } 565727011e0SChris Mason 566b2950863SChristoph Hellwig static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end, 5675cf1ab56SJosef Bacik struct extent_state *state, int mirror) 568ce9adaa5SChris Mason { 569ce9adaa5SChris Mason struct extent_io_tree *tree; 570ce9adaa5SChris Mason u64 found_start; 571ce9adaa5SChris Mason int found_level; 572ce9adaa5SChris Mason struct extent_buffer *eb; 573ce9adaa5SChris Mason struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 574f188591eSChris Mason int ret = 0; 575727011e0SChris Mason int reads_done; 576ce9adaa5SChris Mason 577ce9adaa5SChris Mason if (!page->private) 578ce9adaa5SChris Mason goto out; 579d397712bSChris Mason 580727011e0SChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 5814f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 582d397712bSChris Mason 5830b32f4bbSJosef Bacik /* the pending IO might have been the only thing that kept this buffer 5840b32f4bbSJosef Bacik * in memory. Make sure we have a ref for all this other checks 5850b32f4bbSJosef Bacik */ 5860b32f4bbSJosef Bacik extent_buffer_get(eb); 5870b32f4bbSJosef Bacik 5880b32f4bbSJosef Bacik reads_done = atomic_dec_and_test(&eb->io_pages); 589727011e0SChris Mason if (!reads_done) 590727011e0SChris Mason goto err; 591f188591eSChris Mason 5925cf1ab56SJosef Bacik eb->read_mirror = mirror; 593ea466794SJosef Bacik if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) { 594ea466794SJosef Bacik ret = -EIO; 595ea466794SJosef Bacik goto err; 596ea466794SJosef Bacik } 597ea466794SJosef Bacik 598ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 599727011e0SChris Mason if (found_start != eb->start) { 6007a36ddecSDavid Sterba printk_ratelimited(KERN_INFO "btrfs bad tree block start " 601193f284dSChris Mason "%llu %llu\n", 602a1b32a59SChris Mason (unsigned long long)found_start, 603a1b32a59SChris Mason (unsigned long long)eb->start); 604f188591eSChris Mason ret = -EIO; 605ce9adaa5SChris Mason goto err; 606ce9adaa5SChris Mason } 6072b82032cSYan Zheng if (check_tree_block_fsid(root, eb)) { 6087a36ddecSDavid Sterba printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n", 609d397712bSChris Mason (unsigned long long)eb->start); 6101259ab75SChris Mason ret = -EIO; 6111259ab75SChris Mason goto err; 6121259ab75SChris Mason } 613ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 614ce9adaa5SChris Mason 61585d4e461SChris Mason btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), 61685d4e461SChris Mason eb, found_level); 6174008c04aSChris Mason 618ce9adaa5SChris Mason ret = csum_tree_block(root, eb, 1); 619a826d6dcSJosef Bacik if (ret) { 620f188591eSChris Mason ret = -EIO; 621a826d6dcSJosef Bacik goto err; 622a826d6dcSJosef Bacik } 623a826d6dcSJosef Bacik 624a826d6dcSJosef Bacik /* 625a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 626a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 627a826d6dcSJosef Bacik * return -EIO. 628a826d6dcSJosef Bacik */ 629a826d6dcSJosef Bacik if (found_level == 0 && check_leaf(root, eb)) { 630a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 631a826d6dcSJosef Bacik ret = -EIO; 632a826d6dcSJosef Bacik } 633ce9adaa5SChris Mason 6340b32f4bbSJosef Bacik if (!ret) 6350b32f4bbSJosef Bacik set_extent_buffer_uptodate(eb); 636ce9adaa5SChris Mason err: 6374bb31e92SArne Jansen if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) { 6384bb31e92SArne Jansen clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags); 6394bb31e92SArne Jansen btree_readahead_hook(root, eb, eb->start, ret); 6404bb31e92SArne Jansen } 6414bb31e92SArne Jansen 6420b32f4bbSJosef Bacik if (ret) 6430b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 6440b32f4bbSJosef Bacik free_extent_buffer(eb); 645ce9adaa5SChris Mason out: 646f188591eSChris Mason return ret; 647ce9adaa5SChris Mason } 648ce9adaa5SChris Mason 649ea466794SJosef Bacik static int btree_io_failed_hook(struct page *page, int failed_mirror) 6504bb31e92SArne Jansen { 6514bb31e92SArne Jansen struct extent_buffer *eb; 6524bb31e92SArne Jansen struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 6534bb31e92SArne Jansen 6544f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 655ea466794SJosef Bacik set_bit(EXTENT_BUFFER_IOERR, &eb->bflags); 6565cf1ab56SJosef Bacik eb->read_mirror = failed_mirror; 657ea466794SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 6584bb31e92SArne Jansen btree_readahead_hook(root, eb, eb->start, -EIO); 6594bb31e92SArne Jansen return -EIO; /* we fixed nothing */ 6604bb31e92SArne Jansen } 6614bb31e92SArne Jansen 662ce9adaa5SChris Mason static void end_workqueue_bio(struct bio *bio, int err) 663ce9adaa5SChris Mason { 664ce9adaa5SChris Mason struct end_io_wq *end_io_wq = bio->bi_private; 665ce9adaa5SChris Mason struct btrfs_fs_info *fs_info; 666ce9adaa5SChris Mason 667ce9adaa5SChris Mason fs_info = end_io_wq->info; 668ce9adaa5SChris Mason end_io_wq->error = err; 6698b712842SChris Mason end_io_wq->work.func = end_workqueue_fn; 6708b712842SChris Mason end_io_wq->work.flags = 0; 671d20f7043SChris Mason 6727b6d91daSChristoph Hellwig if (bio->bi_rw & REQ_WRITE) { 6730cb59c99SJosef Bacik if (end_io_wq->metadata == 1) 674cad321adSChris Mason btrfs_queue_worker(&fs_info->endio_meta_write_workers, 675cad321adSChris Mason &end_io_wq->work); 6760cb59c99SJosef Bacik else if (end_io_wq->metadata == 2) 6770cb59c99SJosef Bacik btrfs_queue_worker(&fs_info->endio_freespace_worker, 6780cb59c99SJosef Bacik &end_io_wq->work); 679cad321adSChris Mason else 680e6dcd2dcSChris Mason btrfs_queue_worker(&fs_info->endio_write_workers, 681e6dcd2dcSChris Mason &end_io_wq->work); 682d20f7043SChris Mason } else { 683d20f7043SChris Mason if (end_io_wq->metadata) 684d20f7043SChris Mason btrfs_queue_worker(&fs_info->endio_meta_workers, 685d20f7043SChris Mason &end_io_wq->work); 686e6dcd2dcSChris Mason else 687d20f7043SChris Mason btrfs_queue_worker(&fs_info->endio_workers, 688d20f7043SChris Mason &end_io_wq->work); 689d20f7043SChris Mason } 690ce9adaa5SChris Mason } 691ce9adaa5SChris Mason 6920cb59c99SJosef Bacik /* 6930cb59c99SJosef Bacik * For the metadata arg you want 6940cb59c99SJosef Bacik * 6950cb59c99SJosef Bacik * 0 - if data 6960cb59c99SJosef Bacik * 1 - if normal metadta 6970cb59c99SJosef Bacik * 2 - if writing to the free space cache area 6980cb59c99SJosef Bacik */ 69922c59948SChris Mason int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 70022c59948SChris Mason int metadata) 7010b86a832SChris Mason { 702ce9adaa5SChris Mason struct end_io_wq *end_io_wq; 703ce9adaa5SChris Mason end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS); 704ce9adaa5SChris Mason if (!end_io_wq) 705ce9adaa5SChris Mason return -ENOMEM; 706ce9adaa5SChris Mason 707ce9adaa5SChris Mason end_io_wq->private = bio->bi_private; 708ce9adaa5SChris Mason end_io_wq->end_io = bio->bi_end_io; 70922c59948SChris Mason end_io_wq->info = info; 710ce9adaa5SChris Mason end_io_wq->error = 0; 711ce9adaa5SChris Mason end_io_wq->bio = bio; 71222c59948SChris Mason end_io_wq->metadata = metadata; 713ce9adaa5SChris Mason 714ce9adaa5SChris Mason bio->bi_private = end_io_wq; 715ce9adaa5SChris Mason bio->bi_end_io = end_workqueue_bio; 71622c59948SChris Mason return 0; 71722c59948SChris Mason } 71822c59948SChris Mason 719b64a2851SChris Mason unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info) 7204854ddd0SChris Mason { 7214854ddd0SChris Mason unsigned long limit = min_t(unsigned long, 7224854ddd0SChris Mason info->workers.max_workers, 7234854ddd0SChris Mason info->fs_devices->open_devices); 7244854ddd0SChris Mason return 256 * limit; 7254854ddd0SChris Mason } 7264854ddd0SChris Mason 7274a69a410SChris Mason static void run_one_async_start(struct btrfs_work *work) 7284a69a410SChris Mason { 7294a69a410SChris Mason struct async_submit_bio *async; 73079787eaaSJeff Mahoney int ret; 7314a69a410SChris Mason 7324a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 73379787eaaSJeff Mahoney ret = async->submit_bio_start(async->inode, async->rw, async->bio, 734eaf25d93SChris Mason async->mirror_num, async->bio_flags, 735eaf25d93SChris Mason async->bio_offset); 73679787eaaSJeff Mahoney if (ret) 73779787eaaSJeff Mahoney async->error = ret; 7384a69a410SChris Mason } 7394a69a410SChris Mason 7404a69a410SChris Mason static void run_one_async_done(struct btrfs_work *work) 7418b712842SChris Mason { 7428b712842SChris Mason struct btrfs_fs_info *fs_info; 7438b712842SChris Mason struct async_submit_bio *async; 7444854ddd0SChris Mason int limit; 7458b712842SChris Mason 7468b712842SChris Mason async = container_of(work, struct async_submit_bio, work); 7478b712842SChris Mason fs_info = BTRFS_I(async->inode)->root->fs_info; 7484854ddd0SChris Mason 749b64a2851SChris Mason limit = btrfs_async_submit_limit(fs_info); 7504854ddd0SChris Mason limit = limit * 2 / 3; 7514854ddd0SChris Mason 75266657b31SJosef Bacik if (atomic_dec_return(&fs_info->nr_async_submits) < limit && 753b64a2851SChris Mason waitqueue_active(&fs_info->async_submit_wait)) 7544854ddd0SChris Mason wake_up(&fs_info->async_submit_wait); 7554854ddd0SChris Mason 75679787eaaSJeff Mahoney /* If an error occured we just want to clean up the bio and move on */ 75779787eaaSJeff Mahoney if (async->error) { 75879787eaaSJeff Mahoney bio_endio(async->bio, async->error); 75979787eaaSJeff Mahoney return; 76079787eaaSJeff Mahoney } 76179787eaaSJeff Mahoney 7624a69a410SChris Mason async->submit_bio_done(async->inode, async->rw, async->bio, 763eaf25d93SChris Mason async->mirror_num, async->bio_flags, 764eaf25d93SChris Mason async->bio_offset); 7654a69a410SChris Mason } 7664a69a410SChris Mason 7674a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 7684a69a410SChris Mason { 7694a69a410SChris Mason struct async_submit_bio *async; 7704a69a410SChris Mason 7714a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 7728b712842SChris Mason kfree(async); 7738b712842SChris Mason } 7748b712842SChris Mason 77544b8bd7eSChris Mason int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode, 77644b8bd7eSChris Mason int rw, struct bio *bio, int mirror_num, 777c8b97818SChris Mason unsigned long bio_flags, 778eaf25d93SChris Mason u64 bio_offset, 7794a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_start, 7804a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_done) 78144b8bd7eSChris Mason { 78244b8bd7eSChris Mason struct async_submit_bio *async; 78344b8bd7eSChris Mason 78444b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 78544b8bd7eSChris Mason if (!async) 78644b8bd7eSChris Mason return -ENOMEM; 78744b8bd7eSChris Mason 78844b8bd7eSChris Mason async->inode = inode; 78944b8bd7eSChris Mason async->rw = rw; 79044b8bd7eSChris Mason async->bio = bio; 79144b8bd7eSChris Mason async->mirror_num = mirror_num; 7924a69a410SChris Mason async->submit_bio_start = submit_bio_start; 7934a69a410SChris Mason async->submit_bio_done = submit_bio_done; 7944a69a410SChris Mason 7954a69a410SChris Mason async->work.func = run_one_async_start; 7964a69a410SChris Mason async->work.ordered_func = run_one_async_done; 7974a69a410SChris Mason async->work.ordered_free = run_one_async_free; 7984a69a410SChris Mason 7998b712842SChris Mason async->work.flags = 0; 800c8b97818SChris Mason async->bio_flags = bio_flags; 801eaf25d93SChris Mason async->bio_offset = bio_offset; 8028c8bee1dSChris Mason 80379787eaaSJeff Mahoney async->error = 0; 80479787eaaSJeff Mahoney 805cb03c743SChris Mason atomic_inc(&fs_info->nr_async_submits); 806d313d7a3SChris Mason 8077b6d91daSChristoph Hellwig if (rw & REQ_SYNC) 808d313d7a3SChris Mason btrfs_set_work_high_prio(&async->work); 809d313d7a3SChris Mason 8108b712842SChris Mason btrfs_queue_worker(&fs_info->workers, &async->work); 8119473f16cSChris Mason 812771ed689SChris Mason while (atomic_read(&fs_info->async_submit_draining) && 813771ed689SChris Mason atomic_read(&fs_info->nr_async_submits)) { 814771ed689SChris Mason wait_event(fs_info->async_submit_wait, 815771ed689SChris Mason (atomic_read(&fs_info->nr_async_submits) == 0)); 816771ed689SChris Mason } 817771ed689SChris Mason 81844b8bd7eSChris Mason return 0; 81944b8bd7eSChris Mason } 82044b8bd7eSChris Mason 821ce3ed71aSChris Mason static int btree_csum_one_bio(struct bio *bio) 822ce3ed71aSChris Mason { 823ce3ed71aSChris Mason struct bio_vec *bvec = bio->bi_io_vec; 824ce3ed71aSChris Mason int bio_index = 0; 825ce3ed71aSChris Mason struct btrfs_root *root; 82679787eaaSJeff Mahoney int ret = 0; 827ce3ed71aSChris Mason 828ce3ed71aSChris Mason WARN_ON(bio->bi_vcnt <= 0); 829ce3ed71aSChris Mason while (bio_index < bio->bi_vcnt) { 830ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 83179787eaaSJeff Mahoney ret = csum_dirty_buffer(root, bvec->bv_page); 83279787eaaSJeff Mahoney if (ret) 83379787eaaSJeff Mahoney break; 834ce3ed71aSChris Mason bio_index++; 835ce3ed71aSChris Mason bvec++; 836ce3ed71aSChris Mason } 83779787eaaSJeff Mahoney return ret; 838ce3ed71aSChris Mason } 839ce3ed71aSChris Mason 8404a69a410SChris Mason static int __btree_submit_bio_start(struct inode *inode, int rw, 8414a69a410SChris Mason struct bio *bio, int mirror_num, 842eaf25d93SChris Mason unsigned long bio_flags, 843eaf25d93SChris Mason u64 bio_offset) 84422c59948SChris Mason { 8458b712842SChris Mason /* 8468b712842SChris Mason * when we're called for a write, we're already in the async 8475443be45SChris Mason * submission context. Just jump into btrfs_map_bio 8488b712842SChris Mason */ 84979787eaaSJeff Mahoney return btree_csum_one_bio(bio); 85022c59948SChris Mason } 85122c59948SChris Mason 8524a69a410SChris Mason static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio, 853eaf25d93SChris Mason int mirror_num, unsigned long bio_flags, 854eaf25d93SChris Mason u64 bio_offset) 8554a69a410SChris Mason { 85661891923SStefan Behrens int ret; 85761891923SStefan Behrens 8588b712842SChris Mason /* 8594a69a410SChris Mason * when we're called for a write, we're already in the async 8604a69a410SChris Mason * submission context. Just jump into btrfs_map_bio 8618b712842SChris Mason */ 86261891923SStefan Behrens ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1); 86361891923SStefan Behrens if (ret) 86461891923SStefan Behrens bio_endio(bio, ret); 86561891923SStefan Behrens return ret; 8660b86a832SChris Mason } 8670b86a832SChris Mason 868de0022b9SJosef Bacik static int check_async_write(struct inode *inode, unsigned long bio_flags) 869de0022b9SJosef Bacik { 870de0022b9SJosef Bacik if (bio_flags & EXTENT_BIO_TREE_LOG) 871de0022b9SJosef Bacik return 0; 872de0022b9SJosef Bacik #ifdef CONFIG_X86 873de0022b9SJosef Bacik if (cpu_has_xmm4_2) 874de0022b9SJosef Bacik return 0; 875de0022b9SJosef Bacik #endif 876de0022b9SJosef Bacik return 1; 877de0022b9SJosef Bacik } 878de0022b9SJosef Bacik 87944b8bd7eSChris Mason static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, 880eaf25d93SChris Mason int mirror_num, unsigned long bio_flags, 881eaf25d93SChris Mason u64 bio_offset) 88244b8bd7eSChris Mason { 883de0022b9SJosef Bacik int async = check_async_write(inode, bio_flags); 8844a69a410SChris Mason int ret; 885cad321adSChris Mason 8867b6d91daSChristoph Hellwig if (!(rw & REQ_WRITE)) { 887cad321adSChris Mason /* 888cad321adSChris Mason * called for a read, do the setup so that checksum validation 889cad321adSChris Mason * can happen in the async kernel threads 890cad321adSChris Mason */ 891f3f266abSChris Mason ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info, 892f3f266abSChris Mason bio, 1); 8931d4284bdSChris Mason if (ret) 89461891923SStefan Behrens goto out_w_error; 89561891923SStefan Behrens ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, 8966f3577bdSChris Mason mirror_num, 0); 897de0022b9SJosef Bacik } else if (!async) { 898de0022b9SJosef Bacik ret = btree_csum_one_bio(bio); 899de0022b9SJosef Bacik if (ret) 90061891923SStefan Behrens goto out_w_error; 90161891923SStefan Behrens ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, 902de0022b9SJosef Bacik mirror_num, 0); 90361891923SStefan Behrens } else { 904cad321adSChris Mason /* 90561891923SStefan Behrens * kthread helpers are used to submit writes so that 90661891923SStefan Behrens * checksumming can happen in parallel across all CPUs 907cad321adSChris Mason */ 90861891923SStefan Behrens ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, 909c8b97818SChris Mason inode, rw, bio, mirror_num, 0, 910eaf25d93SChris Mason bio_offset, 9114a69a410SChris Mason __btree_submit_bio_start, 9124a69a410SChris Mason __btree_submit_bio_done); 91344b8bd7eSChris Mason } 91444b8bd7eSChris Mason 91561891923SStefan Behrens if (ret) { 91661891923SStefan Behrens out_w_error: 91761891923SStefan Behrens bio_endio(bio, ret); 91861891923SStefan Behrens } 91961891923SStefan Behrens return ret; 92061891923SStefan Behrens } 92161891923SStefan Behrens 9223dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 923784b4e29SChris Mason static int btree_migratepage(struct address_space *mapping, 924a6bc32b8SMel Gorman struct page *newpage, struct page *page, 925a6bc32b8SMel Gorman enum migrate_mode mode) 926784b4e29SChris Mason { 927784b4e29SChris Mason /* 928784b4e29SChris Mason * we can't safely write a btree page from here, 929784b4e29SChris Mason * we haven't done the locking hook 930784b4e29SChris Mason */ 931784b4e29SChris Mason if (PageDirty(page)) 932784b4e29SChris Mason return -EAGAIN; 933784b4e29SChris Mason /* 934784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 935784b4e29SChris Mason * We must have no buffers or drop them. 936784b4e29SChris Mason */ 937784b4e29SChris Mason if (page_has_private(page) && 938784b4e29SChris Mason !try_to_release_page(page, GFP_KERNEL)) 939784b4e29SChris Mason return -EAGAIN; 940a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 941784b4e29SChris Mason } 9423dd1462eSJan Beulich #endif 943784b4e29SChris Mason 9440da5468fSChris Mason 9450da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 9460da5468fSChris Mason struct writeback_control *wbc) 9470da5468fSChris Mason { 948d1310b2eSChris Mason struct extent_io_tree *tree; 949d1310b2eSChris Mason tree = &BTRFS_I(mapping->host)->io_tree; 950d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 951b9473439SChris Mason struct btrfs_root *root = BTRFS_I(mapping->host)->root; 952793955bcSChris Mason u64 num_dirty; 95324ab9cd8SChris Mason unsigned long thresh = 32 * 1024 * 1024; 954448d640bSChris Mason 955448d640bSChris Mason if (wbc->for_kupdate) 956448d640bSChris Mason return 0; 957448d640bSChris Mason 958b9473439SChris Mason /* this is a bit racy, but that's ok */ 959b9473439SChris Mason num_dirty = root->fs_info->dirty_metadata_bytes; 960d397712bSChris Mason if (num_dirty < thresh) 961793955bcSChris Mason return 0; 962793955bcSChris Mason } 9630b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 9640da5468fSChris Mason } 9650da5468fSChris Mason 966b2950863SChristoph Hellwig static int btree_readpage(struct file *file, struct page *page) 9675f39d397SChris Mason { 968d1310b2eSChris Mason struct extent_io_tree *tree; 969d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9708ddc7d9cSJan Schmidt return extent_read_full_page(tree, page, btree_get_extent, 0); 9715f39d397SChris Mason } 9725f39d397SChris Mason 97370dec807SChris Mason static int btree_releasepage(struct page *page, gfp_t gfp_flags) 9745f39d397SChris Mason { 97598509cfcSChris Mason if (PageWriteback(page) || PageDirty(page)) 97698509cfcSChris Mason return 0; 9770c4e538bSDavid Sterba /* 9780c4e538bSDavid Sterba * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing 9790c4e538bSDavid Sterba * slab allocation from alloc_extent_state down the callchain where 9800c4e538bSDavid Sterba * it'd hit a BUG_ON as those flags are not allowed. 9810c4e538bSDavid Sterba */ 9820c4e538bSDavid Sterba gfp_flags &= ~GFP_SLAB_BUG_MASK; 9830c4e538bSDavid Sterba 9843083ee2eSJosef Bacik return try_release_extent_buffer(page, gfp_flags); 985d98237b3SChris Mason } 986d98237b3SChris Mason 9875f39d397SChris Mason static void btree_invalidatepage(struct page *page, unsigned long offset) 988d98237b3SChris Mason { 989d1310b2eSChris Mason struct extent_io_tree *tree; 990d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9915f39d397SChris Mason extent_invalidatepage(tree, page, offset); 9925f39d397SChris Mason btree_releasepage(page, GFP_NOFS); 9939ad6b7bcSChris Mason if (PagePrivate(page)) { 994d397712bSChris Mason printk(KERN_WARNING "btrfs warning page private not zero " 995d397712bSChris Mason "on page %llu\n", (unsigned long long)page_offset(page)); 9969ad6b7bcSChris Mason ClearPagePrivate(page); 9979ad6b7bcSChris Mason set_page_private(page, 0); 9989ad6b7bcSChris Mason page_cache_release(page); 9999ad6b7bcSChris Mason } 1000d98237b3SChris Mason } 1001d98237b3SChris Mason 10020b32f4bbSJosef Bacik static int btree_set_page_dirty(struct page *page) 10030b32f4bbSJosef Bacik { 1004bb146eb2SJosef Bacik #ifdef DEBUG 10050b32f4bbSJosef Bacik struct extent_buffer *eb; 10060b32f4bbSJosef Bacik 10070b32f4bbSJosef Bacik BUG_ON(!PagePrivate(page)); 10080b32f4bbSJosef Bacik eb = (struct extent_buffer *)page->private; 10090b32f4bbSJosef Bacik BUG_ON(!eb); 10100b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 10110b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 10120b32f4bbSJosef Bacik btrfs_assert_tree_locked(eb); 1013bb146eb2SJosef Bacik #endif 10140b32f4bbSJosef Bacik return __set_page_dirty_nobuffers(page); 10150b32f4bbSJosef Bacik } 10160b32f4bbSJosef Bacik 10177f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 1018d98237b3SChris Mason .readpage = btree_readpage, 10190da5468fSChris Mason .writepages = btree_writepages, 10205f39d397SChris Mason .releasepage = btree_releasepage, 10215f39d397SChris Mason .invalidatepage = btree_invalidatepage, 10225a92bc88SChris Mason #ifdef CONFIG_MIGRATION 1023784b4e29SChris Mason .migratepage = btree_migratepage, 10245a92bc88SChris Mason #endif 10250b32f4bbSJosef Bacik .set_page_dirty = btree_set_page_dirty, 1026d98237b3SChris Mason }; 1027123abc88SChris Mason 1028ca7a79adSChris Mason int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize, 1029ca7a79adSChris Mason u64 parent_transid) 1030090d1875SChris Mason { 10315f39d397SChris Mason struct extent_buffer *buf = NULL; 10325f39d397SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 1033de428b63SChris Mason int ret = 0; 1034090d1875SChris Mason 1035db94535dSChris Mason buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 10365f39d397SChris Mason if (!buf) 1037090d1875SChris Mason return 0; 1038d1310b2eSChris Mason read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, 1039bb82ab88SArne Jansen buf, 0, WAIT_NONE, btree_get_extent, 0); 10405f39d397SChris Mason free_extent_buffer(buf); 1041de428b63SChris Mason return ret; 1042090d1875SChris Mason } 1043090d1875SChris Mason 1044ab0fff03SArne Jansen int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize, 1045ab0fff03SArne Jansen int mirror_num, struct extent_buffer **eb) 1046ab0fff03SArne Jansen { 1047ab0fff03SArne Jansen struct extent_buffer *buf = NULL; 1048ab0fff03SArne Jansen struct inode *btree_inode = root->fs_info->btree_inode; 1049ab0fff03SArne Jansen struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree; 1050ab0fff03SArne Jansen int ret; 1051ab0fff03SArne Jansen 1052ab0fff03SArne Jansen buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 1053ab0fff03SArne Jansen if (!buf) 1054ab0fff03SArne Jansen return 0; 1055ab0fff03SArne Jansen 1056ab0fff03SArne Jansen set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags); 1057ab0fff03SArne Jansen 1058ab0fff03SArne Jansen ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK, 1059ab0fff03SArne Jansen btree_get_extent, mirror_num); 1060ab0fff03SArne Jansen if (ret) { 1061ab0fff03SArne Jansen free_extent_buffer(buf); 1062ab0fff03SArne Jansen return ret; 1063ab0fff03SArne Jansen } 1064ab0fff03SArne Jansen 1065ab0fff03SArne Jansen if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) { 1066ab0fff03SArne Jansen free_extent_buffer(buf); 1067ab0fff03SArne Jansen return -EIO; 10680b32f4bbSJosef Bacik } else if (extent_buffer_uptodate(buf)) { 1069ab0fff03SArne Jansen *eb = buf; 1070ab0fff03SArne Jansen } else { 1071ab0fff03SArne Jansen free_extent_buffer(buf); 1072ab0fff03SArne Jansen } 1073ab0fff03SArne Jansen return 0; 1074ab0fff03SArne Jansen } 1075ab0fff03SArne Jansen 10760999df54SChris Mason struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root, 10770999df54SChris Mason u64 bytenr, u32 blocksize) 10780999df54SChris Mason { 10790999df54SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 10800999df54SChris Mason struct extent_buffer *eb; 10810999df54SChris Mason eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree, 1082f09d1f60SDavid Sterba bytenr, blocksize); 10830999df54SChris Mason return eb; 10840999df54SChris Mason } 10850999df54SChris Mason 10860999df54SChris Mason struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root, 10870999df54SChris Mason u64 bytenr, u32 blocksize) 10880999df54SChris Mason { 10890999df54SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 10900999df54SChris Mason struct extent_buffer *eb; 10910999df54SChris Mason 10920999df54SChris Mason eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree, 1093727011e0SChris Mason bytenr, blocksize); 10940999df54SChris Mason return eb; 10950999df54SChris Mason } 10960999df54SChris Mason 10970999df54SChris Mason 1098e02119d5SChris Mason int btrfs_write_tree_block(struct extent_buffer *buf) 1099e02119d5SChris Mason { 1100727011e0SChris Mason return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start, 11018aa38c31SChristoph Hellwig buf->start + buf->len - 1); 1102e02119d5SChris Mason } 1103e02119d5SChris Mason 1104e02119d5SChris Mason int btrfs_wait_tree_block_writeback(struct extent_buffer *buf) 1105e02119d5SChris Mason { 1106727011e0SChris Mason return filemap_fdatawait_range(buf->pages[0]->mapping, 11078aa38c31SChristoph Hellwig buf->start, buf->start + buf->len - 1); 1108e02119d5SChris Mason } 1109e02119d5SChris Mason 1110db94535dSChris Mason struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr, 1111ca7a79adSChris Mason u32 blocksize, u64 parent_transid) 1112e20d96d6SChris Mason { 11135f39d397SChris Mason struct extent_buffer *buf = NULL; 111419c00ddcSChris Mason int ret; 111519c00ddcSChris Mason 1116db94535dSChris Mason buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 11175f39d397SChris Mason if (!buf) 1118d98237b3SChris Mason return NULL; 1119e4204dedSChris Mason 1120ca7a79adSChris Mason ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 11215f39d397SChris Mason return buf; 1122ce9adaa5SChris Mason 1123eb60ceacSChris Mason } 1124eb60ceacSChris Mason 1125d5c13f92SJeff Mahoney void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root, 11265f39d397SChris Mason struct extent_buffer *buf) 1127ed2ff2cbSChris Mason { 112855c69072SChris Mason if (btrfs_header_generation(buf) == 1129925baeddSChris Mason root->fs_info->running_transaction->transid) { 1130b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1131b4ce94deSChris Mason 1132b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1133b9473439SChris Mason spin_lock(&root->fs_info->delalloc_lock); 1134b9473439SChris Mason if (root->fs_info->dirty_metadata_bytes >= buf->len) 1135b9473439SChris Mason root->fs_info->dirty_metadata_bytes -= buf->len; 1136d5c13f92SJeff Mahoney else { 1137d5c13f92SJeff Mahoney spin_unlock(&root->fs_info->delalloc_lock); 1138d5c13f92SJeff Mahoney btrfs_panic(root->fs_info, -EOVERFLOW, 1139d5c13f92SJeff Mahoney "Can't clear %lu bytes from " 1140533574c6SJoe Perches " dirty_mdatadata_bytes (%llu)", 1141d5c13f92SJeff Mahoney buf->len, 1142d5c13f92SJeff Mahoney root->fs_info->dirty_metadata_bytes); 1143d5c13f92SJeff Mahoney } 1144b9473439SChris Mason spin_unlock(&root->fs_info->delalloc_lock); 1145b4ce94deSChris Mason 1146b9473439SChris Mason /* ugh, clear_extent_buffer_dirty needs to lock the page */ 1147b9473439SChris Mason btrfs_set_lock_blocking(buf); 11480b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 1149925baeddSChris Mason } 11505f39d397SChris Mason } 1151ed7b63ebSJosef Bacik } 11525f39d397SChris Mason 1153143bede5SJeff Mahoney static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize, 115487ee04ebSChris Mason u32 stripesize, struct btrfs_root *root, 11559f5fae2fSChris Mason struct btrfs_fs_info *fs_info, 1156e20d96d6SChris Mason u64 objectid) 1157d97e63b6SChris Mason { 1158cfaa7295SChris Mason root->node = NULL; 1159a28ec197SChris Mason root->commit_root = NULL; 1160db94535dSChris Mason root->sectorsize = sectorsize; 1161db94535dSChris Mason root->nodesize = nodesize; 1162db94535dSChris Mason root->leafsize = leafsize; 116387ee04ebSChris Mason root->stripesize = stripesize; 1164123abc88SChris Mason root->ref_cows = 0; 11650b86a832SChris Mason root->track_dirty = 0; 1166c71bf099SYan, Zheng root->in_radix = 0; 1167d68fc57bSYan, Zheng root->orphan_item_inserted = 0; 1168d68fc57bSYan, Zheng root->orphan_cleanup_state = 0; 11690b86a832SChris Mason 11700f7d52f4SChris Mason root->objectid = objectid; 11710f7d52f4SChris Mason root->last_trans = 0; 117213a8a7c8SYan, Zheng root->highest_objectid = 0; 117358176a96SJosef Bacik root->name = NULL; 11746bef4d31SEric Paris root->inode_tree = RB_ROOT; 117516cdcec7SMiao Xie INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1176f0486c68SYan, Zheng root->block_rsv = NULL; 1177d68fc57bSYan, Zheng root->orphan_block_rsv = NULL; 11780b86a832SChris Mason 11790b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 11805d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 11812ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 11822ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 1183d68fc57bSYan, Zheng spin_lock_init(&root->orphan_lock); 11845d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1185f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 11862ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 11872ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 1188a2135011SChris Mason mutex_init(&root->objectid_mutex); 1189e02119d5SChris Mason mutex_init(&root->log_mutex); 11907237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 11917237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 11927237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 11937237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 11947237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 11957237f183SYan Zheng atomic_set(&root->log_writers, 0); 11962ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 11978a35d95fSJosef Bacik atomic_set(&root->orphan_inodes, 0); 11987237f183SYan Zheng root->log_transid = 0; 1199257c62e1SChris Mason root->last_log_commit = 0; 1200d0c803c4SChris Mason extent_io_tree_init(&root->dirty_log_pages, 1201f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 1202017e5369SChris Mason 12033768f368SChris Mason memset(&root->root_key, 0, sizeof(root->root_key)); 12043768f368SChris Mason memset(&root->root_item, 0, sizeof(root->root_item)); 12056702ed49SChris Mason memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 120658176a96SJosef Bacik memset(&root->root_kobj, 0, sizeof(root->root_kobj)); 12073f157a2fSChris Mason root->defrag_trans_start = fs_info->generation; 120858176a96SJosef Bacik init_completion(&root->kobj_unregister); 12096702ed49SChris Mason root->defrag_running = 0; 12104d775673SChris Mason root->root_key.objectid = objectid; 12110ee5dc67SAl Viro root->anon_dev = 0; 12128ea05e3aSAlexander Block 12135f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 12143768f368SChris Mason } 12153768f368SChris Mason 1216200a5c17SJeff Mahoney static int __must_check find_and_setup_root(struct btrfs_root *tree_root, 12179f5fae2fSChris Mason struct btrfs_fs_info *fs_info, 12189f5fae2fSChris Mason u64 objectid, 1219e20d96d6SChris Mason struct btrfs_root *root) 12203768f368SChris Mason { 12213768f368SChris Mason int ret; 1222db94535dSChris Mason u32 blocksize; 122384234f3aSYan Zheng u64 generation; 12243768f368SChris Mason 1225db94535dSChris Mason __setup_root(tree_root->nodesize, tree_root->leafsize, 122687ee04ebSChris Mason tree_root->sectorsize, tree_root->stripesize, 122787ee04ebSChris Mason root, fs_info, objectid); 12283768f368SChris Mason ret = btrfs_find_last_root(tree_root, objectid, 12293768f368SChris Mason &root->root_item, &root->root_key); 12304df27c4dSYan, Zheng if (ret > 0) 12314df27c4dSYan, Zheng return -ENOENT; 1232200a5c17SJeff Mahoney else if (ret < 0) 1233200a5c17SJeff Mahoney return ret; 12343768f368SChris Mason 123584234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1236db94535dSChris Mason blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item)); 1237af31f5e5SChris Mason root->commit_root = NULL; 1238db94535dSChris Mason root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item), 123984234f3aSYan Zheng blocksize, generation); 1240b9fab919SChris Mason if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) { 124168433b73SChris Mason free_extent_buffer(root->node); 1242af31f5e5SChris Mason root->node = NULL; 124368433b73SChris Mason return -EIO; 124468433b73SChris Mason } 12454df27c4dSYan, Zheng root->commit_root = btrfs_root_node(root); 1246d97e63b6SChris Mason return 0; 1247d97e63b6SChris Mason } 1248d97e63b6SChris Mason 1249f84a8bd6SAl Viro static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info) 12506f07e42eSAl Viro { 12516f07e42eSAl Viro struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS); 12526f07e42eSAl Viro if (root) 12536f07e42eSAl Viro root->fs_info = fs_info; 12546f07e42eSAl Viro return root; 12556f07e42eSAl Viro } 12566f07e42eSAl Viro 125720897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 125820897f5cSArne Jansen struct btrfs_fs_info *fs_info, 125920897f5cSArne Jansen u64 objectid) 126020897f5cSArne Jansen { 126120897f5cSArne Jansen struct extent_buffer *leaf; 126220897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 126320897f5cSArne Jansen struct btrfs_root *root; 126420897f5cSArne Jansen struct btrfs_key key; 126520897f5cSArne Jansen int ret = 0; 126620897f5cSArne Jansen u64 bytenr; 126720897f5cSArne Jansen 126820897f5cSArne Jansen root = btrfs_alloc_root(fs_info); 126920897f5cSArne Jansen if (!root) 127020897f5cSArne Jansen return ERR_PTR(-ENOMEM); 127120897f5cSArne Jansen 127220897f5cSArne Jansen __setup_root(tree_root->nodesize, tree_root->leafsize, 127320897f5cSArne Jansen tree_root->sectorsize, tree_root->stripesize, 127420897f5cSArne Jansen root, fs_info, objectid); 127520897f5cSArne Jansen root->root_key.objectid = objectid; 127620897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 127720897f5cSArne Jansen root->root_key.offset = 0; 127820897f5cSArne Jansen 127920897f5cSArne Jansen leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 128020897f5cSArne Jansen 0, objectid, NULL, 0, 0, 0); 128120897f5cSArne Jansen if (IS_ERR(leaf)) { 128220897f5cSArne Jansen ret = PTR_ERR(leaf); 128320897f5cSArne Jansen goto fail; 128420897f5cSArne Jansen } 128520897f5cSArne Jansen 128620897f5cSArne Jansen bytenr = leaf->start; 128720897f5cSArne Jansen memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 128820897f5cSArne Jansen btrfs_set_header_bytenr(leaf, leaf->start); 128920897f5cSArne Jansen btrfs_set_header_generation(leaf, trans->transid); 129020897f5cSArne Jansen btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 129120897f5cSArne Jansen btrfs_set_header_owner(leaf, objectid); 129220897f5cSArne Jansen root->node = leaf; 129320897f5cSArne Jansen 129420897f5cSArne Jansen write_extent_buffer(leaf, fs_info->fsid, 129520897f5cSArne Jansen (unsigned long)btrfs_header_fsid(leaf), 129620897f5cSArne Jansen BTRFS_FSID_SIZE); 129720897f5cSArne Jansen write_extent_buffer(leaf, fs_info->chunk_tree_uuid, 129820897f5cSArne Jansen (unsigned long)btrfs_header_chunk_tree_uuid(leaf), 129920897f5cSArne Jansen BTRFS_UUID_SIZE); 130020897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 130120897f5cSArne Jansen 130220897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 130320897f5cSArne Jansen root->track_dirty = 1; 130420897f5cSArne Jansen 130520897f5cSArne Jansen 130620897f5cSArne Jansen root->root_item.flags = 0; 130720897f5cSArne Jansen root->root_item.byte_limit = 0; 130820897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 130920897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 131020897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 131120897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 131220897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 131320897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 131420897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 131520897f5cSArne Jansen root->root_item.drop_level = 0; 131620897f5cSArne Jansen 131720897f5cSArne Jansen key.objectid = objectid; 131820897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 131920897f5cSArne Jansen key.offset = 0; 132020897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 132120897f5cSArne Jansen if (ret) 132220897f5cSArne Jansen goto fail; 132320897f5cSArne Jansen 132420897f5cSArne Jansen btrfs_tree_unlock(leaf); 132520897f5cSArne Jansen 132620897f5cSArne Jansen fail: 132720897f5cSArne Jansen if (ret) 132820897f5cSArne Jansen return ERR_PTR(ret); 132920897f5cSArne Jansen 133020897f5cSArne Jansen return root; 133120897f5cSArne Jansen } 133220897f5cSArne Jansen 13337237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1334e02119d5SChris Mason struct btrfs_fs_info *fs_info) 13350f7d52f4SChris Mason { 13360f7d52f4SChris Mason struct btrfs_root *root; 13370f7d52f4SChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 13387237f183SYan Zheng struct extent_buffer *leaf; 1339e02119d5SChris Mason 13406f07e42eSAl Viro root = btrfs_alloc_root(fs_info); 1341e02119d5SChris Mason if (!root) 13427237f183SYan Zheng return ERR_PTR(-ENOMEM); 1343e02119d5SChris Mason 1344e02119d5SChris Mason __setup_root(tree_root->nodesize, tree_root->leafsize, 1345e02119d5SChris Mason tree_root->sectorsize, tree_root->stripesize, 1346e02119d5SChris Mason root, fs_info, BTRFS_TREE_LOG_OBJECTID); 1347e02119d5SChris Mason 1348e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1349e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1350e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 13517237f183SYan Zheng /* 13527237f183SYan Zheng * log trees do not get reference counted because they go away 13537237f183SYan Zheng * before a real commit is actually done. They do store pointers 13547237f183SYan Zheng * to file data extents, and those reference counts still get 13557237f183SYan Zheng * updated (along with back refs to the log tree). 13567237f183SYan Zheng */ 1357e02119d5SChris Mason root->ref_cows = 0; 1358e02119d5SChris Mason 13595d4f98a2SYan Zheng leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0, 136066d7e7f0SArne Jansen BTRFS_TREE_LOG_OBJECTID, NULL, 13615581a51aSJan Schmidt 0, 0, 0); 13627237f183SYan Zheng if (IS_ERR(leaf)) { 13637237f183SYan Zheng kfree(root); 13647237f183SYan Zheng return ERR_CAST(leaf); 13657237f183SYan Zheng } 1366e02119d5SChris Mason 13675d4f98a2SYan Zheng memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 13685d4f98a2SYan Zheng btrfs_set_header_bytenr(leaf, leaf->start); 13695d4f98a2SYan Zheng btrfs_set_header_generation(leaf, trans->transid); 13705d4f98a2SYan Zheng btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 13715d4f98a2SYan Zheng btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID); 13727237f183SYan Zheng root->node = leaf; 1373e02119d5SChris Mason 1374e02119d5SChris Mason write_extent_buffer(root->node, root->fs_info->fsid, 1375e02119d5SChris Mason (unsigned long)btrfs_header_fsid(root->node), 1376e02119d5SChris Mason BTRFS_FSID_SIZE); 1377e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1378e02119d5SChris Mason btrfs_tree_unlock(root->node); 13797237f183SYan Zheng return root; 13807237f183SYan Zheng } 13817237f183SYan Zheng 13827237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 13837237f183SYan Zheng struct btrfs_fs_info *fs_info) 13847237f183SYan Zheng { 13857237f183SYan Zheng struct btrfs_root *log_root; 13867237f183SYan Zheng 13877237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 13887237f183SYan Zheng if (IS_ERR(log_root)) 13897237f183SYan Zheng return PTR_ERR(log_root); 13907237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 13917237f183SYan Zheng fs_info->log_root_tree = log_root; 13927237f183SYan Zheng return 0; 13937237f183SYan Zheng } 13947237f183SYan Zheng 13957237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 13967237f183SYan Zheng struct btrfs_root *root) 13977237f183SYan Zheng { 13987237f183SYan Zheng struct btrfs_root *log_root; 13997237f183SYan Zheng struct btrfs_inode_item *inode_item; 14007237f183SYan Zheng 14017237f183SYan Zheng log_root = alloc_log_tree(trans, root->fs_info); 14027237f183SYan Zheng if (IS_ERR(log_root)) 14037237f183SYan Zheng return PTR_ERR(log_root); 14047237f183SYan Zheng 14057237f183SYan Zheng log_root->last_trans = trans->transid; 14067237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 14077237f183SYan Zheng 14087237f183SYan Zheng inode_item = &log_root->root_item.inode; 14097237f183SYan Zheng inode_item->generation = cpu_to_le64(1); 14107237f183SYan Zheng inode_item->size = cpu_to_le64(3); 14117237f183SYan Zheng inode_item->nlink = cpu_to_le32(1); 14127237f183SYan Zheng inode_item->nbytes = cpu_to_le64(root->leafsize); 14137237f183SYan Zheng inode_item->mode = cpu_to_le32(S_IFDIR | 0755); 14147237f183SYan Zheng 14155d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 14167237f183SYan Zheng 14177237f183SYan Zheng WARN_ON(root->log_root); 14187237f183SYan Zheng root->log_root = log_root; 14197237f183SYan Zheng root->log_transid = 0; 1420257c62e1SChris Mason root->last_log_commit = 0; 1421e02119d5SChris Mason return 0; 1422e02119d5SChris Mason } 1423e02119d5SChris Mason 1424e02119d5SChris Mason struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root, 1425e02119d5SChris Mason struct btrfs_key *location) 1426e02119d5SChris Mason { 1427e02119d5SChris Mason struct btrfs_root *root; 1428e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 14290f7d52f4SChris Mason struct btrfs_path *path; 14305f39d397SChris Mason struct extent_buffer *l; 143184234f3aSYan Zheng u64 generation; 1432db94535dSChris Mason u32 blocksize; 14330f7d52f4SChris Mason int ret = 0; 14348ea05e3aSAlexander Block int slot; 14350f7d52f4SChris Mason 14366f07e42eSAl Viro root = btrfs_alloc_root(fs_info); 14370cf6c620SChris Mason if (!root) 14380f7d52f4SChris Mason return ERR_PTR(-ENOMEM); 14390f7d52f4SChris Mason if (location->offset == (u64)-1) { 1440db94535dSChris Mason ret = find_and_setup_root(tree_root, fs_info, 14410f7d52f4SChris Mason location->objectid, root); 14420f7d52f4SChris Mason if (ret) { 14430f7d52f4SChris Mason kfree(root); 14440f7d52f4SChris Mason return ERR_PTR(ret); 14450f7d52f4SChris Mason } 144613a8a7c8SYan, Zheng goto out; 14470f7d52f4SChris Mason } 14480f7d52f4SChris Mason 1449db94535dSChris Mason __setup_root(tree_root->nodesize, tree_root->leafsize, 145087ee04ebSChris Mason tree_root->sectorsize, tree_root->stripesize, 145187ee04ebSChris Mason root, fs_info, location->objectid); 14520f7d52f4SChris Mason 14530f7d52f4SChris Mason path = btrfs_alloc_path(); 1454db5b493aSTsutomu Itoh if (!path) { 1455db5b493aSTsutomu Itoh kfree(root); 1456db5b493aSTsutomu Itoh return ERR_PTR(-ENOMEM); 1457db5b493aSTsutomu Itoh } 14580f7d52f4SChris Mason ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0); 145913a8a7c8SYan, Zheng if (ret == 0) { 14605f39d397SChris Mason l = path->nodes[0]; 14618ea05e3aSAlexander Block slot = path->slots[0]; 14628ea05e3aSAlexander Block btrfs_read_root_item(tree_root, l, slot, &root->root_item); 146344b36eb2SYan Zheng memcpy(&root->root_key, location, sizeof(*location)); 146413a8a7c8SYan, Zheng } 14650f7d52f4SChris Mason btrfs_free_path(path); 14660f7d52f4SChris Mason if (ret) { 14675e540f77STsutomu Itoh kfree(root); 146813a8a7c8SYan, Zheng if (ret > 0) 146913a8a7c8SYan, Zheng ret = -ENOENT; 14700f7d52f4SChris Mason return ERR_PTR(ret); 14710f7d52f4SChris Mason } 147213a8a7c8SYan, Zheng 147384234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1474db94535dSChris Mason blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item)); 1475db94535dSChris Mason root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item), 147684234f3aSYan Zheng blocksize, generation); 14775d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 147879787eaaSJeff Mahoney BUG_ON(!root->node); /* -ENOMEM */ 147913a8a7c8SYan, Zheng out: 148008fe4db1SLi Zefan if (location->objectid != BTRFS_TREE_LOG_OBJECTID) { 14810f7d52f4SChris Mason root->ref_cows = 1; 148208fe4db1SLi Zefan btrfs_check_and_init_root_item(&root->root_item); 148308fe4db1SLi Zefan } 148413a8a7c8SYan, Zheng 14855eda7b5eSChris Mason return root; 14865eda7b5eSChris Mason } 14875eda7b5eSChris Mason 1488edbd8d4eSChris Mason struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info, 1489edbd8d4eSChris Mason struct btrfs_key *location) 14905eda7b5eSChris Mason { 14915eda7b5eSChris Mason struct btrfs_root *root; 14925eda7b5eSChris Mason int ret; 14935eda7b5eSChris Mason 1494edbd8d4eSChris Mason if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 1495edbd8d4eSChris Mason return fs_info->tree_root; 1496edbd8d4eSChris Mason if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID) 1497edbd8d4eSChris Mason return fs_info->extent_root; 14988f18cf13SChris Mason if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID) 14998f18cf13SChris Mason return fs_info->chunk_root; 15008f18cf13SChris Mason if (location->objectid == BTRFS_DEV_TREE_OBJECTID) 15018f18cf13SChris Mason return fs_info->dev_root; 15020403e47eSYan Zheng if (location->objectid == BTRFS_CSUM_TREE_OBJECTID) 15030403e47eSYan Zheng return fs_info->csum_root; 1504bcef60f2SArne Jansen if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID) 1505bcef60f2SArne Jansen return fs_info->quota_root ? fs_info->quota_root : 1506bcef60f2SArne Jansen ERR_PTR(-ENOENT); 15074df27c4dSYan, Zheng again: 15084df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 15095eda7b5eSChris Mason root = radix_tree_lookup(&fs_info->fs_roots_radix, 15105eda7b5eSChris Mason (unsigned long)location->objectid); 15114df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 15125eda7b5eSChris Mason if (root) 15135eda7b5eSChris Mason return root; 15145eda7b5eSChris Mason 1515e02119d5SChris Mason root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location); 15165eda7b5eSChris Mason if (IS_ERR(root)) 15175eda7b5eSChris Mason return root; 15183394e160SChris Mason 1519581bb050SLi Zefan root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); 1520581bb050SLi Zefan root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), 1521581bb050SLi Zefan GFP_NOFS); 152235a30d7cSDavid Sterba if (!root->free_ino_pinned || !root->free_ino_ctl) { 152335a30d7cSDavid Sterba ret = -ENOMEM; 1524581bb050SLi Zefan goto fail; 152535a30d7cSDavid Sterba } 1526581bb050SLi Zefan 1527581bb050SLi Zefan btrfs_init_free_ino_ctl(root); 1528581bb050SLi Zefan mutex_init(&root->fs_commit_mutex); 1529581bb050SLi Zefan spin_lock_init(&root->cache_lock); 1530581bb050SLi Zefan init_waitqueue_head(&root->cache_wait); 1531581bb050SLi Zefan 15320ee5dc67SAl Viro ret = get_anon_bdev(&root->anon_dev); 1533ac08aedfSChris Mason if (ret) 1534ac08aedfSChris Mason goto fail; 15353394e160SChris Mason 1536d68fc57bSYan, Zheng if (btrfs_root_refs(&root->root_item) == 0) { 1537d68fc57bSYan, Zheng ret = -ENOENT; 1538d68fc57bSYan, Zheng goto fail; 1539d68fc57bSYan, Zheng } 1540d68fc57bSYan, Zheng 1541d68fc57bSYan, Zheng ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid); 1542d68fc57bSYan, Zheng if (ret < 0) 1543d68fc57bSYan, Zheng goto fail; 1544d68fc57bSYan, Zheng if (ret == 0) 1545d68fc57bSYan, Zheng root->orphan_item_inserted = 1; 1546d68fc57bSYan, Zheng 15474df27c4dSYan, Zheng ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); 15484df27c4dSYan, Zheng if (ret) 15494df27c4dSYan, Zheng goto fail; 15504df27c4dSYan, Zheng 15514df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 15522619ba1fSChris Mason ret = radix_tree_insert(&fs_info->fs_roots_radix, 15532619ba1fSChris Mason (unsigned long)root->root_key.objectid, 15540f7d52f4SChris Mason root); 1555d68fc57bSYan, Zheng if (ret == 0) 15564df27c4dSYan, Zheng root->in_radix = 1; 1557d68fc57bSYan, Zheng 15584df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 15594df27c4dSYan, Zheng radix_tree_preload_end(); 15600f7d52f4SChris Mason if (ret) { 15614df27c4dSYan, Zheng if (ret == -EEXIST) { 15624df27c4dSYan, Zheng free_fs_root(root); 15634df27c4dSYan, Zheng goto again; 15640f7d52f4SChris Mason } 15654df27c4dSYan, Zheng goto fail; 15664df27c4dSYan, Zheng } 15674df27c4dSYan, Zheng 1568edbd8d4eSChris Mason ret = btrfs_find_dead_roots(fs_info->tree_root, 15695d4f98a2SYan Zheng root->root_key.objectid); 15704df27c4dSYan, Zheng WARN_ON(ret); 1571edbd8d4eSChris Mason return root; 15724df27c4dSYan, Zheng fail: 15734df27c4dSYan, Zheng free_fs_root(root); 15744df27c4dSYan, Zheng return ERR_PTR(ret); 1575edbd8d4eSChris Mason } 1576edbd8d4eSChris Mason 157704160088SChris Mason static int btrfs_congested_fn(void *congested_data, int bdi_bits) 157804160088SChris Mason { 157904160088SChris Mason struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; 158004160088SChris Mason int ret = 0; 158104160088SChris Mason struct btrfs_device *device; 158204160088SChris Mason struct backing_dev_info *bdi; 1583b7967db7SChris Mason 15841f78160cSXiao Guangrong rcu_read_lock(); 15851f78160cSXiao Guangrong list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) { 1586dfe25020SChris Mason if (!device->bdev) 1587dfe25020SChris Mason continue; 158804160088SChris Mason bdi = blk_get_backing_dev_info(device->bdev); 158904160088SChris Mason if (bdi && bdi_congested(bdi, bdi_bits)) { 159004160088SChris Mason ret = 1; 159104160088SChris Mason break; 159204160088SChris Mason } 159304160088SChris Mason } 15941f78160cSXiao Guangrong rcu_read_unlock(); 159504160088SChris Mason return ret; 159604160088SChris Mason } 159704160088SChris Mason 159838b66988SChris Mason /* 1599ad081f14SJens Axboe * If this fails, caller must call bdi_destroy() to get rid of the 1600ad081f14SJens Axboe * bdi again. 1601ad081f14SJens Axboe */ 160204160088SChris Mason static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi) 160304160088SChris Mason { 1604ad081f14SJens Axboe int err; 1605ad081f14SJens Axboe 1606ad081f14SJens Axboe bdi->capabilities = BDI_CAP_MAP_COPY; 1607e6d086d8SJens Axboe err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY); 1608ad081f14SJens Axboe if (err) 1609ad081f14SJens Axboe return err; 1610ad081f14SJens Axboe 16114575c9ccSChris Mason bdi->ra_pages = default_backing_dev_info.ra_pages; 161204160088SChris Mason bdi->congested_fn = btrfs_congested_fn; 161304160088SChris Mason bdi->congested_data = info; 161404160088SChris Mason return 0; 161504160088SChris Mason } 161604160088SChris Mason 16178b712842SChris Mason /* 16188b712842SChris Mason * called by the kthread helper functions to finally call the bio end_io 16198b712842SChris Mason * functions. This is where read checksum verification actually happens 16208b712842SChris Mason */ 16218b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work) 1622ce9adaa5SChris Mason { 1623ce9adaa5SChris Mason struct bio *bio; 16248b712842SChris Mason struct end_io_wq *end_io_wq; 16258b712842SChris Mason struct btrfs_fs_info *fs_info; 1626ce9adaa5SChris Mason int error; 1627ce9adaa5SChris Mason 16288b712842SChris Mason end_io_wq = container_of(work, struct end_io_wq, work); 1629ce9adaa5SChris Mason bio = end_io_wq->bio; 16308b712842SChris Mason fs_info = end_io_wq->info; 16318b712842SChris Mason 1632ce9adaa5SChris Mason error = end_io_wq->error; 1633ce9adaa5SChris Mason bio->bi_private = end_io_wq->private; 1634ce9adaa5SChris Mason bio->bi_end_io = end_io_wq->end_io; 1635ce9adaa5SChris Mason kfree(end_io_wq); 1636ce9adaa5SChris Mason bio_endio(bio, error); 1637ce9adaa5SChris Mason } 163844b8bd7eSChris Mason 1639a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1640a74a4b97SChris Mason { 1641a74a4b97SChris Mason struct btrfs_root *root = arg; 1642a74a4b97SChris Mason 1643a74a4b97SChris Mason do { 164476dda93cSYan, Zheng if (!(root->fs_info->sb->s_flags & MS_RDONLY) && 164576dda93cSYan, Zheng mutex_trylock(&root->fs_info->cleaner_mutex)) { 164624bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 1647a74a4b97SChris Mason btrfs_clean_old_snapshots(root); 1648a74a4b97SChris Mason mutex_unlock(&root->fs_info->cleaner_mutex); 16494cb5300bSChris Mason btrfs_run_defrag_inodes(root->fs_info); 165076dda93cSYan, Zheng } 1651a74a4b97SChris Mason 1652a0acae0eSTejun Heo if (!try_to_freeze()) { 1653a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 16548929ecfaSYan, Zheng if (!kthread_should_stop()) 1655a74a4b97SChris Mason schedule(); 1656a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1657a74a4b97SChris Mason } 1658a74a4b97SChris Mason } while (!kthread_should_stop()); 1659a74a4b97SChris Mason return 0; 1660a74a4b97SChris Mason } 1661a74a4b97SChris Mason 1662a74a4b97SChris Mason static int transaction_kthread(void *arg) 1663a74a4b97SChris Mason { 1664a74a4b97SChris Mason struct btrfs_root *root = arg; 1665a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1666a74a4b97SChris Mason struct btrfs_transaction *cur; 16678929ecfaSYan, Zheng u64 transid; 1668a74a4b97SChris Mason unsigned long now; 1669a74a4b97SChris Mason unsigned long delay; 1670914b2007SJan Kara bool cannot_commit; 1671a74a4b97SChris Mason 1672a74a4b97SChris Mason do { 1673914b2007SJan Kara cannot_commit = false; 1674a74a4b97SChris Mason delay = HZ * 30; 1675a74a4b97SChris Mason mutex_lock(&root->fs_info->transaction_kthread_mutex); 1676a74a4b97SChris Mason 1677a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 1678a74a4b97SChris Mason cur = root->fs_info->running_transaction; 1679a74a4b97SChris Mason if (!cur) { 1680a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1681a74a4b97SChris Mason goto sleep; 1682a74a4b97SChris Mason } 168331153d81SYan Zheng 1684a74a4b97SChris Mason now = get_seconds(); 16858929ecfaSYan, Zheng if (!cur->blocked && 16868929ecfaSYan, Zheng (now < cur->start_time || now - cur->start_time < 30)) { 1687a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1688a74a4b97SChris Mason delay = HZ * 5; 1689a74a4b97SChris Mason goto sleep; 1690a74a4b97SChris Mason } 16918929ecfaSYan, Zheng transid = cur->transid; 1692a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 169356bec294SChris Mason 169479787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 1695354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 1696914b2007SJan Kara if (IS_ERR(trans)) { 1697354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 1698914b2007SJan Kara cannot_commit = true; 169979787eaaSJeff Mahoney goto sleep; 1700914b2007SJan Kara } 17018929ecfaSYan, Zheng if (transid == trans->transid) { 170279787eaaSJeff Mahoney btrfs_commit_transaction(trans, root); 17038929ecfaSYan, Zheng } else { 17048929ecfaSYan, Zheng btrfs_end_transaction(trans, root); 17058929ecfaSYan, Zheng } 1706a74a4b97SChris Mason sleep: 1707a74a4b97SChris Mason wake_up_process(root->fs_info->cleaner_kthread); 1708a74a4b97SChris Mason mutex_unlock(&root->fs_info->transaction_kthread_mutex); 1709a74a4b97SChris Mason 1710a0acae0eSTejun Heo if (!try_to_freeze()) { 1711a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 17128929ecfaSYan, Zheng if (!kthread_should_stop() && 1713914b2007SJan Kara (!btrfs_transaction_blocked(root->fs_info) || 1714914b2007SJan Kara cannot_commit)) 1715a74a4b97SChris Mason schedule_timeout(delay); 1716a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1717a74a4b97SChris Mason } 1718a74a4b97SChris Mason } while (!kthread_should_stop()); 1719a74a4b97SChris Mason return 0; 1720a74a4b97SChris Mason } 1721a74a4b97SChris Mason 1722af31f5e5SChris Mason /* 1723af31f5e5SChris Mason * this will find the highest generation in the array of 1724af31f5e5SChris Mason * root backups. The index of the highest array is returned, 1725af31f5e5SChris Mason * or -1 if we can't find anything. 1726af31f5e5SChris Mason * 1727af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1728af31f5e5SChris Mason * generation of the latest root in the array with the generation 1729af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1730af31f5e5SChris Mason */ 1731af31f5e5SChris Mason static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen) 1732af31f5e5SChris Mason { 1733af31f5e5SChris Mason u64 cur; 1734af31f5e5SChris Mason int newest_index = -1; 1735af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1736af31f5e5SChris Mason int i; 1737af31f5e5SChris Mason 1738af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1739af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1740af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1741af31f5e5SChris Mason if (cur == newest_gen) 1742af31f5e5SChris Mason newest_index = i; 1743af31f5e5SChris Mason } 1744af31f5e5SChris Mason 1745af31f5e5SChris Mason /* check to see if we actually wrapped around */ 1746af31f5e5SChris Mason if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) { 1747af31f5e5SChris Mason root_backup = info->super_copy->super_roots; 1748af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1749af31f5e5SChris Mason if (cur == newest_gen) 1750af31f5e5SChris Mason newest_index = 0; 1751af31f5e5SChris Mason } 1752af31f5e5SChris Mason return newest_index; 1753af31f5e5SChris Mason } 1754af31f5e5SChris Mason 1755af31f5e5SChris Mason 1756af31f5e5SChris Mason /* 1757af31f5e5SChris Mason * find the oldest backup so we know where to store new entries 1758af31f5e5SChris Mason * in the backup array. This will set the backup_root_index 1759af31f5e5SChris Mason * field in the fs_info struct 1760af31f5e5SChris Mason */ 1761af31f5e5SChris Mason static void find_oldest_super_backup(struct btrfs_fs_info *info, 1762af31f5e5SChris Mason u64 newest_gen) 1763af31f5e5SChris Mason { 1764af31f5e5SChris Mason int newest_index = -1; 1765af31f5e5SChris Mason 1766af31f5e5SChris Mason newest_index = find_newest_super_backup(info, newest_gen); 1767af31f5e5SChris Mason /* if there was garbage in there, just move along */ 1768af31f5e5SChris Mason if (newest_index == -1) { 1769af31f5e5SChris Mason info->backup_root_index = 0; 1770af31f5e5SChris Mason } else { 1771af31f5e5SChris Mason info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS; 1772af31f5e5SChris Mason } 1773af31f5e5SChris Mason } 1774af31f5e5SChris Mason 1775af31f5e5SChris Mason /* 1776af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1777af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1778af31f5e5SChris Mason * done 1779af31f5e5SChris Mason */ 1780af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1781af31f5e5SChris Mason { 1782af31f5e5SChris Mason int next_backup; 1783af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1784af31f5e5SChris Mason int last_backup; 1785af31f5e5SChris Mason 1786af31f5e5SChris Mason next_backup = info->backup_root_index; 1787af31f5e5SChris Mason last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) % 1788af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 1789af31f5e5SChris Mason 1790af31f5e5SChris Mason /* 1791af31f5e5SChris Mason * just overwrite the last backup if we're at the same generation 1792af31f5e5SChris Mason * this happens only at umount 1793af31f5e5SChris Mason */ 1794af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + last_backup; 1795af31f5e5SChris Mason if (btrfs_backup_tree_root_gen(root_backup) == 1796af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)) 1797af31f5e5SChris Mason next_backup = last_backup; 1798af31f5e5SChris Mason 1799af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1800af31f5e5SChris Mason 1801af31f5e5SChris Mason /* 1802af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1803af31f5e5SChris Mason * regardless of which ones we use today 1804af31f5e5SChris Mason */ 1805af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1806af31f5e5SChris Mason 1807af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1808af31f5e5SChris Mason 1809af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1810af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1811af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1812af31f5e5SChris Mason 1813af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1814af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1815af31f5e5SChris Mason 1816af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1817af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1818af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 1819af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 1820af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 1821af31f5e5SChris Mason 1822af31f5e5SChris Mason btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); 1823af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 1824af31f5e5SChris Mason btrfs_header_generation(info->extent_root->node)); 1825af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 1826af31f5e5SChris Mason btrfs_header_level(info->extent_root->node)); 1827af31f5e5SChris Mason 18287c7e82a7SChris Mason /* 18297c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 18307c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 18317c7e82a7SChris Mason */ 18327c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 18337c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 18347c7e82a7SChris Mason info->fs_root->node->start); 1835af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 1836af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 1837af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 1838af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 18397c7e82a7SChris Mason } 1840af31f5e5SChris Mason 1841af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 1842af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 1843af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 1844af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 1845af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 1846af31f5e5SChris Mason 1847af31f5e5SChris Mason btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); 1848af31f5e5SChris Mason btrfs_set_backup_csum_root_gen(root_backup, 1849af31f5e5SChris Mason btrfs_header_generation(info->csum_root->node)); 1850af31f5e5SChris Mason btrfs_set_backup_csum_root_level(root_backup, 1851af31f5e5SChris Mason btrfs_header_level(info->csum_root->node)); 1852af31f5e5SChris Mason 1853af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 1854af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 1855af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 1856af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 1857af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 1858af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 1859af31f5e5SChris Mason 1860af31f5e5SChris Mason /* 1861af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 1862af31f5e5SChris Mason * for the next commit 1863af31f5e5SChris Mason */ 1864af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 1865af31f5e5SChris Mason &info->super_for_commit->super_roots, 1866af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 1867af31f5e5SChris Mason } 1868af31f5e5SChris Mason 1869af31f5e5SChris Mason /* 1870af31f5e5SChris Mason * this copies info out of the root backup array and back into 1871af31f5e5SChris Mason * the in-memory super block. It is meant to help iterate through 1872af31f5e5SChris Mason * the array, so you send it the number of backups you've already 1873af31f5e5SChris Mason * tried and the last backup index you used. 1874af31f5e5SChris Mason * 1875af31f5e5SChris Mason * this returns -1 when it has tried all the backups 1876af31f5e5SChris Mason */ 1877af31f5e5SChris Mason static noinline int next_root_backup(struct btrfs_fs_info *info, 1878af31f5e5SChris Mason struct btrfs_super_block *super, 1879af31f5e5SChris Mason int *num_backups_tried, int *backup_index) 1880af31f5e5SChris Mason { 1881af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1882af31f5e5SChris Mason int newest = *backup_index; 1883af31f5e5SChris Mason 1884af31f5e5SChris Mason if (*num_backups_tried == 0) { 1885af31f5e5SChris Mason u64 gen = btrfs_super_generation(super); 1886af31f5e5SChris Mason 1887af31f5e5SChris Mason newest = find_newest_super_backup(info, gen); 1888af31f5e5SChris Mason if (newest == -1) 1889af31f5e5SChris Mason return -1; 1890af31f5e5SChris Mason 1891af31f5e5SChris Mason *backup_index = newest; 1892af31f5e5SChris Mason *num_backups_tried = 1; 1893af31f5e5SChris Mason } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) { 1894af31f5e5SChris Mason /* we've tried all the backups, all done */ 1895af31f5e5SChris Mason return -1; 1896af31f5e5SChris Mason } else { 1897af31f5e5SChris Mason /* jump to the next oldest backup */ 1898af31f5e5SChris Mason newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) % 1899af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 1900af31f5e5SChris Mason *backup_index = newest; 1901af31f5e5SChris Mason *num_backups_tried += 1; 1902af31f5e5SChris Mason } 1903af31f5e5SChris Mason root_backup = super->super_roots + newest; 1904af31f5e5SChris Mason 1905af31f5e5SChris Mason btrfs_set_super_generation(super, 1906af31f5e5SChris Mason btrfs_backup_tree_root_gen(root_backup)); 1907af31f5e5SChris Mason btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 1908af31f5e5SChris Mason btrfs_set_super_root_level(super, 1909af31f5e5SChris Mason btrfs_backup_tree_root_level(root_backup)); 1910af31f5e5SChris Mason btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 1911af31f5e5SChris Mason 1912af31f5e5SChris Mason /* 1913af31f5e5SChris Mason * fixme: the total bytes and num_devices need to match or we should 1914af31f5e5SChris Mason * need a fsck 1915af31f5e5SChris Mason */ 1916af31f5e5SChris Mason btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 1917af31f5e5SChris Mason btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 1918af31f5e5SChris Mason return 0; 1919af31f5e5SChris Mason } 1920af31f5e5SChris Mason 1921af31f5e5SChris Mason /* helper to cleanup tree roots */ 1922af31f5e5SChris Mason static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root) 1923af31f5e5SChris Mason { 1924af31f5e5SChris Mason free_extent_buffer(info->tree_root->node); 1925af31f5e5SChris Mason free_extent_buffer(info->tree_root->commit_root); 1926af31f5e5SChris Mason free_extent_buffer(info->dev_root->node); 1927af31f5e5SChris Mason free_extent_buffer(info->dev_root->commit_root); 1928af31f5e5SChris Mason free_extent_buffer(info->extent_root->node); 1929af31f5e5SChris Mason free_extent_buffer(info->extent_root->commit_root); 1930af31f5e5SChris Mason free_extent_buffer(info->csum_root->node); 1931af31f5e5SChris Mason free_extent_buffer(info->csum_root->commit_root); 1932bcef60f2SArne Jansen if (info->quota_root) { 1933bcef60f2SArne Jansen free_extent_buffer(info->quota_root->node); 1934bcef60f2SArne Jansen free_extent_buffer(info->quota_root->commit_root); 1935bcef60f2SArne Jansen } 1936af31f5e5SChris Mason 1937af31f5e5SChris Mason info->tree_root->node = NULL; 1938af31f5e5SChris Mason info->tree_root->commit_root = NULL; 1939af31f5e5SChris Mason info->dev_root->node = NULL; 1940af31f5e5SChris Mason info->dev_root->commit_root = NULL; 1941af31f5e5SChris Mason info->extent_root->node = NULL; 1942af31f5e5SChris Mason info->extent_root->commit_root = NULL; 1943af31f5e5SChris Mason info->csum_root->node = NULL; 1944af31f5e5SChris Mason info->csum_root->commit_root = NULL; 1945bcef60f2SArne Jansen if (info->quota_root) { 1946bcef60f2SArne Jansen info->quota_root->node = NULL; 1947bcef60f2SArne Jansen info->quota_root->commit_root = NULL; 1948bcef60f2SArne Jansen } 1949af31f5e5SChris Mason 1950af31f5e5SChris Mason if (chunk_root) { 1951af31f5e5SChris Mason free_extent_buffer(info->chunk_root->node); 1952af31f5e5SChris Mason free_extent_buffer(info->chunk_root->commit_root); 1953af31f5e5SChris Mason info->chunk_root->node = NULL; 1954af31f5e5SChris Mason info->chunk_root->commit_root = NULL; 1955af31f5e5SChris Mason } 1956af31f5e5SChris Mason } 1957af31f5e5SChris Mason 1958af31f5e5SChris Mason 1959ad2b2c80SAl Viro int open_ctree(struct super_block *sb, 1960dfe25020SChris Mason struct btrfs_fs_devices *fs_devices, 1961dfe25020SChris Mason char *options) 1962eb60ceacSChris Mason { 1963db94535dSChris Mason u32 sectorsize; 1964db94535dSChris Mason u32 nodesize; 1965db94535dSChris Mason u32 leafsize; 1966db94535dSChris Mason u32 blocksize; 196787ee04ebSChris Mason u32 stripesize; 196884234f3aSYan Zheng u64 generation; 1969f2b636e8SJosef Bacik u64 features; 19703de4586cSChris Mason struct btrfs_key location; 1971a061fc8dSChris Mason struct buffer_head *bh; 19724d34b278SIlya Dryomov struct btrfs_super_block *disk_super; 1973815745cfSAl Viro struct btrfs_fs_info *fs_info = btrfs_sb(sb); 1974f84a8bd6SAl Viro struct btrfs_root *tree_root; 19754d34b278SIlya Dryomov struct btrfs_root *extent_root; 19764d34b278SIlya Dryomov struct btrfs_root *csum_root; 19774d34b278SIlya Dryomov struct btrfs_root *chunk_root; 19784d34b278SIlya Dryomov struct btrfs_root *dev_root; 1979bcef60f2SArne Jansen struct btrfs_root *quota_root; 1980e02119d5SChris Mason struct btrfs_root *log_tree_root; 1981eb60ceacSChris Mason int ret; 1982e58ca020SYan int err = -EINVAL; 1983af31f5e5SChris Mason int num_backups_tried = 0; 1984af31f5e5SChris Mason int backup_index = 0; 19854543df7eSChris Mason 1986f84a8bd6SAl Viro tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info); 19876f07e42eSAl Viro extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info); 19886f07e42eSAl Viro csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info); 19896f07e42eSAl Viro chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info); 19906f07e42eSAl Viro dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info); 1991bcef60f2SArne Jansen quota_root = fs_info->quota_root = btrfs_alloc_root(fs_info); 19928790d502SChris Mason 1993f84a8bd6SAl Viro if (!tree_root || !extent_root || !csum_root || 1994bcef60f2SArne Jansen !chunk_root || !dev_root || !quota_root) { 199539279cc3SChris Mason err = -ENOMEM; 199639279cc3SChris Mason goto fail; 199739279cc3SChris Mason } 199876dda93cSYan, Zheng 199976dda93cSYan, Zheng ret = init_srcu_struct(&fs_info->subvol_srcu); 200076dda93cSYan, Zheng if (ret) { 200176dda93cSYan, Zheng err = ret; 200276dda93cSYan, Zheng goto fail; 200376dda93cSYan, Zheng } 200476dda93cSYan, Zheng 200576dda93cSYan, Zheng ret = setup_bdi(fs_info, &fs_info->bdi); 200676dda93cSYan, Zheng if (ret) { 200776dda93cSYan, Zheng err = ret; 200876dda93cSYan, Zheng goto fail_srcu; 200976dda93cSYan, Zheng } 201076dda93cSYan, Zheng 201176dda93cSYan, Zheng fs_info->btree_inode = new_inode(sb); 201276dda93cSYan, Zheng if (!fs_info->btree_inode) { 201376dda93cSYan, Zheng err = -ENOMEM; 201476dda93cSYan, Zheng goto fail_bdi; 201576dda93cSYan, Zheng } 201676dda93cSYan, Zheng 2017a6591715SChris Mason mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 20181561dedaSMiao Xie 201976dda93cSYan, Zheng INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 20208fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 2021facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 202224bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 2023ea8c2819SChris Mason INIT_LIST_HEAD(&fs_info->delalloc_inodes); 20245a3f23d5SChris Mason INIT_LIST_HEAD(&fs_info->ordered_operations); 202511833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 20261832a6d5SChris Mason spin_lock_init(&fs_info->delalloc_lock); 2027a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 202876dda93cSYan, Zheng spin_lock_init(&fs_info->fs_roots_radix_lock); 202924bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 20304cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 20312bf64758SJosef Bacik spin_lock_init(&fs_info->free_chunk_lock); 2032f29021b2SJan Schmidt spin_lock_init(&fs_info->tree_mod_seq_lock); 2033f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 20347585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 203519c00ddcSChris Mason 203658176a96SJosef Bacik init_completion(&fs_info->kobj_unregister); 20370b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 20386324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 2039f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 20400b86a832SChris Mason btrfs_mapping_init(&fs_info->mapping_tree); 204166d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 204266d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 204366d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delalloc_block_rsv, 204466d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELALLOC); 204566d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 204666d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 204766d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 204866d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 204966d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 2050cb03c743SChris Mason atomic_set(&fs_info->nr_async_submits, 0); 2051771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 20528c8bee1dSChris Mason atomic_set(&fs_info->async_submit_draining, 0); 20530986fe9eSChris Mason atomic_set(&fs_info->nr_async_bios, 0); 20544cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 2055f29021b2SJan Schmidt atomic_set(&fs_info->tree_mod_seq, 0); 2056e20d96d6SChris Mason fs_info->sb = sb; 20576f568d35SChris Mason fs_info->max_inline = 8192 * 1024; 20589ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 20594cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 2060a4abeea4SJosef Bacik fs_info->trans_no_join = 0; 20612bf64758SJosef Bacik fs_info->free_chunk_space = 0; 2062f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 2063c8b97818SChris Mason 206490519d66SArne Jansen /* readahead state */ 206590519d66SArne Jansen INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT); 206690519d66SArne Jansen spin_lock_init(&fs_info->reada_lock); 20672c90e5d6SChris Mason 2068b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 2069b34b086cSChris Mason num_online_cpus() + 2, 8); 20700afbaf8cSChris Mason 20713eaa2885SChris Mason INIT_LIST_HEAD(&fs_info->ordered_extents); 20723eaa2885SChris Mason spin_lock_init(&fs_info->ordered_extent_lock); 207316cdcec7SMiao Xie fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 207416cdcec7SMiao Xie GFP_NOFS); 207516cdcec7SMiao Xie if (!fs_info->delayed_root) { 207616cdcec7SMiao Xie err = -ENOMEM; 207716cdcec7SMiao Xie goto fail_iput; 207816cdcec7SMiao Xie } 207916cdcec7SMiao Xie btrfs_init_delayed_root(fs_info->delayed_root); 20803eaa2885SChris Mason 2081a2de733cSArne Jansen mutex_init(&fs_info->scrub_lock); 2082a2de733cSArne Jansen atomic_set(&fs_info->scrubs_running, 0); 2083a2de733cSArne Jansen atomic_set(&fs_info->scrub_pause_req, 0); 2084a2de733cSArne Jansen atomic_set(&fs_info->scrubs_paused, 0); 2085a2de733cSArne Jansen atomic_set(&fs_info->scrub_cancel_req, 0); 2086a2de733cSArne Jansen init_waitqueue_head(&fs_info->scrub_pause_wait); 2087a2de733cSArne Jansen init_rwsem(&fs_info->scrub_super_lock); 2088a2de733cSArne Jansen fs_info->scrub_workers_refcnt = 0; 208921adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 209021adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 209121adbd5cSStefan Behrens #endif 2092a2de733cSArne Jansen 2093c9e9f97bSIlya Dryomov spin_lock_init(&fs_info->balance_lock); 2094c9e9f97bSIlya Dryomov mutex_init(&fs_info->balance_mutex); 2095837d5b6eSIlya Dryomov atomic_set(&fs_info->balance_running, 0); 2096837d5b6eSIlya Dryomov atomic_set(&fs_info->balance_pause_req, 0); 2097a7e99c69SIlya Dryomov atomic_set(&fs_info->balance_cancel_req, 0); 2098c9e9f97bSIlya Dryomov fs_info->balance_ctl = NULL; 2099837d5b6eSIlya Dryomov init_waitqueue_head(&fs_info->balance_wait_q); 2100a061fc8dSChris Mason 21010afbaf8cSChris Mason sb->s_blocksize = 4096; 21020afbaf8cSChris Mason sb->s_blocksize_bits = blksize_bits(4096); 210332a88aa1SJens Axboe sb->s_bdi = &fs_info->bdi; 21040afbaf8cSChris Mason 210576dda93cSYan, Zheng fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 2106bfe86848SMiklos Szeredi set_nlink(fs_info->btree_inode, 1); 21070afbaf8cSChris Mason /* 21080afbaf8cSChris Mason * we set the i_size on the btree inode to the max possible int. 21090afbaf8cSChris Mason * the real end of the address space is determined by all of 21100afbaf8cSChris Mason * the devices in the system 21110afbaf8cSChris Mason */ 21120afbaf8cSChris Mason fs_info->btree_inode->i_size = OFFSET_MAX; 2113d98237b3SChris Mason fs_info->btree_inode->i_mapping->a_ops = &btree_aops; 211404160088SChris Mason fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi; 211504160088SChris Mason 21165d4f98a2SYan Zheng RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node); 2117d1310b2eSChris Mason extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree, 2118f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 21190b32f4bbSJosef Bacik BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0; 2120a8067e02SDavid Sterba extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree); 21210da5468fSChris Mason 2122d1310b2eSChris Mason BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops; 2123d1310b2eSChris Mason 21240f7d52f4SChris Mason BTRFS_I(fs_info->btree_inode)->root = tree_root; 21250f7d52f4SChris Mason memset(&BTRFS_I(fs_info->btree_inode)->location, 0, 21260f7d52f4SChris Mason sizeof(struct btrfs_key)); 212772ac3c0dSJosef Bacik set_bit(BTRFS_INODE_DUMMY, 212872ac3c0dSJosef Bacik &BTRFS_I(fs_info->btree_inode)->runtime_flags); 212922b0ebdaSChris Mason insert_inode_hash(fs_info->btree_inode); 213039279cc3SChris Mason 2131e02119d5SChris Mason spin_lock_init(&fs_info->block_group_cache_lock); 21326bef4d31SEric Paris fs_info->block_group_cache_tree = RB_ROOT; 2133925baeddSChris Mason 213411833d66SYan Zheng extent_io_tree_init(&fs_info->freed_extents[0], 2135f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 213611833d66SYan Zheng extent_io_tree_init(&fs_info->freed_extents[1], 2137f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 213811833d66SYan Zheng fs_info->pinned_extents = &fs_info->freed_extents[0]; 2139e6dcd2dcSChris Mason fs_info->do_barriers = 1; 2140f9295749SChris Mason 2141d98237b3SChris Mason 21425a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 214330ae8467SChris Mason mutex_init(&fs_info->tree_log_mutex); 2144925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 2145925baeddSChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 2146a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 2147a74a4b97SChris Mason mutex_init(&fs_info->volume_mutex); 2148276e680dSYan Zheng init_rwsem(&fs_info->extent_commit_sem); 2149c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 215076dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 2151e922e087SStefan Behrens fs_info->dev_replace.lock_owner = 0; 2152e922e087SStefan Behrens atomic_set(&fs_info->dev_replace.nesting_level, 0); 2153e922e087SStefan Behrens mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2154e922e087SStefan Behrens mutex_init(&fs_info->dev_replace.lock_management_lock); 2155e922e087SStefan Behrens mutex_init(&fs_info->dev_replace.lock); 2156fa9c0d79SChris Mason 2157416ac51dSArne Jansen spin_lock_init(&fs_info->qgroup_lock); 2158416ac51dSArne Jansen fs_info->qgroup_tree = RB_ROOT; 2159416ac51dSArne Jansen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2160416ac51dSArne Jansen fs_info->qgroup_seq = 1; 2161416ac51dSArne Jansen fs_info->quota_enabled = 0; 2162416ac51dSArne Jansen fs_info->pending_quota_state = 0; 2163416ac51dSArne Jansen 2164fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2165fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2166fa9c0d79SChris Mason 21677d9eb12cSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 21683768f368SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 2169bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 21704854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 21719a8dd150SChris Mason 21720b86a832SChris Mason __setup_root(4096, 4096, 4096, 4096, tree_root, 21732c90e5d6SChris Mason fs_info, BTRFS_ROOT_TREE_OBJECTID); 21747eccb903SChris Mason 21753c4bb26bSChris Mason invalidate_bdev(fs_devices->latest_bdev); 2176a512bbf8SYan Zheng bh = btrfs_read_dev_super(fs_devices->latest_bdev); 217720b45077SDave Young if (!bh) { 217820b45077SDave Young err = -EINVAL; 217916cdcec7SMiao Xie goto fail_alloc; 218020b45077SDave Young } 218139279cc3SChris Mason 21826c41761fSDavid Sterba memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy)); 21836c41761fSDavid Sterba memcpy(fs_info->super_for_commit, fs_info->super_copy, 21846c41761fSDavid Sterba sizeof(*fs_info->super_for_commit)); 2185a061fc8dSChris Mason brelse(bh); 21865f39d397SChris Mason 21876c41761fSDavid Sterba memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE); 21880b86a832SChris Mason 21896c41761fSDavid Sterba disk_super = fs_info->super_copy; 21900f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 219116cdcec7SMiao Xie goto fail_alloc; 21920f7d52f4SChris Mason 2193acce952bSliubo /* check FS state, whether FS is broken. */ 2194acce952bSliubo fs_info->fs_state |= btrfs_super_flags(disk_super); 2195acce952bSliubo 2196fcd1f065SDavid Sterba ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY); 2197fcd1f065SDavid Sterba if (ret) { 2198fcd1f065SDavid Sterba printk(KERN_ERR "btrfs: superblock contains fatal errors\n"); 2199fcd1f065SDavid Sterba err = ret; 2200fcd1f065SDavid Sterba goto fail_alloc; 2201fcd1f065SDavid Sterba } 2202acce952bSliubo 220375e7cb7fSLiu Bo /* 2204af31f5e5SChris Mason * run through our array of backup supers and setup 2205af31f5e5SChris Mason * our ring pointer to the oldest one 2206af31f5e5SChris Mason */ 2207af31f5e5SChris Mason generation = btrfs_super_generation(disk_super); 2208af31f5e5SChris Mason find_oldest_super_backup(fs_info, generation); 2209af31f5e5SChris Mason 2210af31f5e5SChris Mason /* 221175e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 221275e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 221375e7cb7fSLiu Bo */ 221475e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 221575e7cb7fSLiu Bo 22162b82032cSYan Zheng ret = btrfs_parse_options(tree_root, options); 22172b82032cSYan Zheng if (ret) { 22182b82032cSYan Zheng err = ret; 221916cdcec7SMiao Xie goto fail_alloc; 22202b82032cSYan Zheng } 2221dfe25020SChris Mason 2222f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 2223f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 2224f2b636e8SJosef Bacik if (features) { 2225f2b636e8SJosef Bacik printk(KERN_ERR "BTRFS: couldn't mount because of " 2226f2b636e8SJosef Bacik "unsupported optional features (%Lx).\n", 222721380931SJoel Becker (unsigned long long)features); 2228f2b636e8SJosef Bacik err = -EINVAL; 222916cdcec7SMiao Xie goto fail_alloc; 2230f2b636e8SJosef Bacik } 2231f2b636e8SJosef Bacik 2232727011e0SChris Mason if (btrfs_super_leafsize(disk_super) != 2233727011e0SChris Mason btrfs_super_nodesize(disk_super)) { 2234727011e0SChris Mason printk(KERN_ERR "BTRFS: couldn't mount because metadata " 2235727011e0SChris Mason "blocksizes don't match. node %d leaf %d\n", 2236727011e0SChris Mason btrfs_super_nodesize(disk_super), 2237727011e0SChris Mason btrfs_super_leafsize(disk_super)); 2238727011e0SChris Mason err = -EINVAL; 2239727011e0SChris Mason goto fail_alloc; 2240727011e0SChris Mason } 2241727011e0SChris Mason if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) { 2242727011e0SChris Mason printk(KERN_ERR "BTRFS: couldn't mount because metadata " 2243727011e0SChris Mason "blocksize (%d) was too large\n", 2244727011e0SChris Mason btrfs_super_leafsize(disk_super)); 2245727011e0SChris Mason err = -EINVAL; 2246727011e0SChris Mason goto fail_alloc; 2247727011e0SChris Mason } 2248727011e0SChris Mason 22495d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 22505d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 225169e380d1SLi Zefan if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO) 2252a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 2253727011e0SChris Mason 2254727011e0SChris Mason /* 2255727011e0SChris Mason * flag our filesystem as having big metadata blocks if 2256727011e0SChris Mason * they are bigger than the page size 2257727011e0SChris Mason */ 2258727011e0SChris Mason if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) { 2259727011e0SChris Mason if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 2260727011e0SChris Mason printk(KERN_INFO "btrfs flagging fs with big metadata feature\n"); 2261727011e0SChris Mason features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 2262727011e0SChris Mason } 2263727011e0SChris Mason 2264bc3f116fSChris Mason nodesize = btrfs_super_nodesize(disk_super); 2265bc3f116fSChris Mason leafsize = btrfs_super_leafsize(disk_super); 2266bc3f116fSChris Mason sectorsize = btrfs_super_sectorsize(disk_super); 2267bc3f116fSChris Mason stripesize = btrfs_super_stripesize(disk_super); 2268bc3f116fSChris Mason 2269bc3f116fSChris Mason /* 2270bc3f116fSChris Mason * mixed block groups end up with duplicate but slightly offset 2271bc3f116fSChris Mason * extent buffers for the same range. It leads to corruptions 2272bc3f116fSChris Mason */ 2273bc3f116fSChris Mason if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 2274bc3f116fSChris Mason (sectorsize != leafsize)) { 2275bc3f116fSChris Mason printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes " 2276bc3f116fSChris Mason "are not allowed for mixed block groups on %s\n", 2277bc3f116fSChris Mason sb->s_id); 2278bc3f116fSChris Mason goto fail_alloc; 2279bc3f116fSChris Mason } 2280bc3f116fSChris Mason 22815d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 22825d4f98a2SYan Zheng 2283f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 2284f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 2285f2b636e8SJosef Bacik if (!(sb->s_flags & MS_RDONLY) && features) { 2286f2b636e8SJosef Bacik printk(KERN_ERR "BTRFS: couldn't mount RDWR because of " 2287f2b636e8SJosef Bacik "unsupported option features (%Lx).\n", 228821380931SJoel Becker (unsigned long long)features); 2289f2b636e8SJosef Bacik err = -EINVAL; 229016cdcec7SMiao Xie goto fail_alloc; 2291f2b636e8SJosef Bacik } 229261d92c32SChris Mason 229361d92c32SChris Mason btrfs_init_workers(&fs_info->generic_worker, 229461d92c32SChris Mason "genwork", 1, NULL); 229561d92c32SChris Mason 22965443be45SChris Mason btrfs_init_workers(&fs_info->workers, "worker", 229761d92c32SChris Mason fs_info->thread_pool_size, 229861d92c32SChris Mason &fs_info->generic_worker); 2299c8b97818SChris Mason 2300771ed689SChris Mason btrfs_init_workers(&fs_info->delalloc_workers, "delalloc", 230161d92c32SChris Mason fs_info->thread_pool_size, 230261d92c32SChris Mason &fs_info->generic_worker); 2303771ed689SChris Mason 23048ccf6f19SMiao Xie btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc", 23058ccf6f19SMiao Xie fs_info->thread_pool_size, 23068ccf6f19SMiao Xie &fs_info->generic_worker); 23078ccf6f19SMiao Xie 23085443be45SChris Mason btrfs_init_workers(&fs_info->submit_workers, "submit", 2309b720d209SChris Mason min_t(u64, fs_devices->num_devices, 231061d92c32SChris Mason fs_info->thread_pool_size), 231161d92c32SChris Mason &fs_info->generic_worker); 231261b49440SChris Mason 2313bab39bf9SJosef Bacik btrfs_init_workers(&fs_info->caching_workers, "cache", 2314bab39bf9SJosef Bacik 2, &fs_info->generic_worker); 2315bab39bf9SJosef Bacik 231661b49440SChris Mason /* a higher idle thresh on the submit workers makes it much more 231761b49440SChris Mason * likely that bios will be send down in a sane order to the 231861b49440SChris Mason * devices 231961b49440SChris Mason */ 232061b49440SChris Mason fs_info->submit_workers.idle_thresh = 64; 232153863232SChris Mason 2322771ed689SChris Mason fs_info->workers.idle_thresh = 16; 23234a69a410SChris Mason fs_info->workers.ordered = 1; 232461b49440SChris Mason 2325771ed689SChris Mason fs_info->delalloc_workers.idle_thresh = 2; 2326771ed689SChris Mason fs_info->delalloc_workers.ordered = 1; 2327771ed689SChris Mason 232861d92c32SChris Mason btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1, 232961d92c32SChris Mason &fs_info->generic_worker); 23305443be45SChris Mason btrfs_init_workers(&fs_info->endio_workers, "endio", 233161d92c32SChris Mason fs_info->thread_pool_size, 233261d92c32SChris Mason &fs_info->generic_worker); 2333d20f7043SChris Mason btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta", 233461d92c32SChris Mason fs_info->thread_pool_size, 233561d92c32SChris Mason &fs_info->generic_worker); 2336cad321adSChris Mason btrfs_init_workers(&fs_info->endio_meta_write_workers, 233761d92c32SChris Mason "endio-meta-write", fs_info->thread_pool_size, 233861d92c32SChris Mason &fs_info->generic_worker); 23395443be45SChris Mason btrfs_init_workers(&fs_info->endio_write_workers, "endio-write", 234061d92c32SChris Mason fs_info->thread_pool_size, 234161d92c32SChris Mason &fs_info->generic_worker); 23420cb59c99SJosef Bacik btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write", 23430cb59c99SJosef Bacik 1, &fs_info->generic_worker); 234416cdcec7SMiao Xie btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta", 234516cdcec7SMiao Xie fs_info->thread_pool_size, 234616cdcec7SMiao Xie &fs_info->generic_worker); 234790519d66SArne Jansen btrfs_init_workers(&fs_info->readahead_workers, "readahead", 234890519d66SArne Jansen fs_info->thread_pool_size, 234990519d66SArne Jansen &fs_info->generic_worker); 235061b49440SChris Mason 235161b49440SChris Mason /* 235261b49440SChris Mason * endios are largely parallel and should have a very 235361b49440SChris Mason * low idle thresh 235461b49440SChris Mason */ 235561b49440SChris Mason fs_info->endio_workers.idle_thresh = 4; 2356b51912c9SChris Mason fs_info->endio_meta_workers.idle_thresh = 4; 2357b51912c9SChris Mason 23589042846bSChris Mason fs_info->endio_write_workers.idle_thresh = 2; 23599042846bSChris Mason fs_info->endio_meta_write_workers.idle_thresh = 2; 236090519d66SArne Jansen fs_info->readahead_workers.idle_thresh = 2; 23619042846bSChris Mason 23620dc3b84aSJosef Bacik /* 23630dc3b84aSJosef Bacik * btrfs_start_workers can really only fail because of ENOMEM so just 23640dc3b84aSJosef Bacik * return -ENOMEM if any of these fail. 23650dc3b84aSJosef Bacik */ 23660dc3b84aSJosef Bacik ret = btrfs_start_workers(&fs_info->workers); 23670dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->generic_worker); 23680dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->submit_workers); 23690dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->delalloc_workers); 23700dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->fixup_workers); 23710dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->endio_workers); 23720dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->endio_meta_workers); 23730dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers); 23740dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->endio_write_workers); 23750dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->endio_freespace_worker); 23760dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->delayed_workers); 23770dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->caching_workers); 23780dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->readahead_workers); 23798ccf6f19SMiao Xie ret |= btrfs_start_workers(&fs_info->flush_workers); 23800dc3b84aSJosef Bacik if (ret) { 2381fed425c7SIlya Dryomov err = -ENOMEM; 23820dc3b84aSJosef Bacik goto fail_sb_buffer; 23830dc3b84aSJosef Bacik } 23844543df7eSChris Mason 23854575c9ccSChris Mason fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super); 2386c8b97818SChris Mason fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages, 2387c8b97818SChris Mason 4 * 1024 * 1024 / PAGE_CACHE_SIZE); 23884575c9ccSChris Mason 2389db94535dSChris Mason tree_root->nodesize = nodesize; 2390db94535dSChris Mason tree_root->leafsize = leafsize; 2391db94535dSChris Mason tree_root->sectorsize = sectorsize; 239287ee04ebSChris Mason tree_root->stripesize = stripesize; 2393a061fc8dSChris Mason 2394a061fc8dSChris Mason sb->s_blocksize = sectorsize; 2395a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 2396db94535dSChris Mason 239739279cc3SChris Mason if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC, 239839279cc3SChris Mason sizeof(disk_super->magic))) { 2399d397712bSChris Mason printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id); 240039279cc3SChris Mason goto fail_sb_buffer; 240139279cc3SChris Mason } 240219c00ddcSChris Mason 24038d082fb7SLiu Bo if (sectorsize != PAGE_SIZE) { 24048d082fb7SLiu Bo printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) " 24058d082fb7SLiu Bo "found on %s\n", (unsigned long)sectorsize, sb->s_id); 2406941b2ddfSKeith Mannthey goto fail_sb_buffer; 2407941b2ddfSKeith Mannthey } 2408941b2ddfSKeith Mannthey 2409925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 2410e4404d6eSYan Zheng ret = btrfs_read_sys_array(tree_root); 2411925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 241284eed90fSChris Mason if (ret) { 2413d397712bSChris Mason printk(KERN_WARNING "btrfs: failed to read the system " 2414d397712bSChris Mason "array on %s\n", sb->s_id); 24155d4f98a2SYan Zheng goto fail_sb_buffer; 241684eed90fSChris Mason } 24170b86a832SChris Mason 24180b86a832SChris Mason blocksize = btrfs_level_size(tree_root, 24190b86a832SChris Mason btrfs_super_chunk_root_level(disk_super)); 242084234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 24210b86a832SChris Mason 24220b86a832SChris Mason __setup_root(nodesize, leafsize, sectorsize, stripesize, 24230b86a832SChris Mason chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID); 24240b86a832SChris Mason 24250b86a832SChris Mason chunk_root->node = read_tree_block(chunk_root, 24260b86a832SChris Mason btrfs_super_chunk_root(disk_super), 242784234f3aSYan Zheng blocksize, generation); 242879787eaaSJeff Mahoney BUG_ON(!chunk_root->node); /* -ENOMEM */ 242983121942SDavid Woodhouse if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) { 243083121942SDavid Woodhouse printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n", 243183121942SDavid Woodhouse sb->s_id); 2432af31f5e5SChris Mason goto fail_tree_roots; 243383121942SDavid Woodhouse } 24345d4f98a2SYan Zheng btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 24355d4f98a2SYan Zheng chunk_root->commit_root = btrfs_root_node(chunk_root); 24360b86a832SChris Mason 2437e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 2438e17cade2SChris Mason (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node), 2439e17cade2SChris Mason BTRFS_UUID_SIZE); 2440e17cade2SChris Mason 24410b86a832SChris Mason ret = btrfs_read_chunk_tree(chunk_root); 24422b82032cSYan Zheng if (ret) { 2443d397712bSChris Mason printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n", 2444d397712bSChris Mason sb->s_id); 2445af31f5e5SChris Mason goto fail_tree_roots; 24462b82032cSYan Zheng } 24470b86a832SChris Mason 24488dabb742SStefan Behrens /* 24498dabb742SStefan Behrens * keep the device that is marked to be the target device for the 24508dabb742SStefan Behrens * dev_replace procedure 24518dabb742SStefan Behrens */ 24528dabb742SStefan Behrens btrfs_close_extra_devices(fs_info, fs_devices, 0); 2453dfe25020SChris Mason 2454a6b0d5c8SChris Mason if (!fs_devices->latest_bdev) { 2455a6b0d5c8SChris Mason printk(KERN_CRIT "btrfs: failed to read devices on %s\n", 2456a6b0d5c8SChris Mason sb->s_id); 2457a6b0d5c8SChris Mason goto fail_tree_roots; 2458a6b0d5c8SChris Mason } 2459a6b0d5c8SChris Mason 2460af31f5e5SChris Mason retry_root_backup: 2461db94535dSChris Mason blocksize = btrfs_level_size(tree_root, 2462db94535dSChris Mason btrfs_super_root_level(disk_super)); 246384234f3aSYan Zheng generation = btrfs_super_generation(disk_super); 24640b86a832SChris Mason 2465e20d96d6SChris Mason tree_root->node = read_tree_block(tree_root, 2466db94535dSChris Mason btrfs_super_root(disk_super), 246784234f3aSYan Zheng blocksize, generation); 2468af31f5e5SChris Mason if (!tree_root->node || 2469af31f5e5SChris Mason !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) { 247083121942SDavid Woodhouse printk(KERN_WARNING "btrfs: failed to read tree root on %s\n", 247183121942SDavid Woodhouse sb->s_id); 2472af31f5e5SChris Mason 2473af31f5e5SChris Mason goto recovery_tree_root; 247483121942SDavid Woodhouse } 2475af31f5e5SChris Mason 24765d4f98a2SYan Zheng btrfs_set_root_node(&tree_root->root_item, tree_root->node); 24775d4f98a2SYan Zheng tree_root->commit_root = btrfs_root_node(tree_root); 2478db94535dSChris Mason 2479db94535dSChris Mason ret = find_and_setup_root(tree_root, fs_info, 2480e20d96d6SChris Mason BTRFS_EXTENT_TREE_OBJECTID, extent_root); 24810b86a832SChris Mason if (ret) 2482af31f5e5SChris Mason goto recovery_tree_root; 24830b86a832SChris Mason extent_root->track_dirty = 1; 24840b86a832SChris Mason 24850b86a832SChris Mason ret = find_and_setup_root(tree_root, fs_info, 24860b86a832SChris Mason BTRFS_DEV_TREE_OBJECTID, dev_root); 24870b86a832SChris Mason if (ret) 2488af31f5e5SChris Mason goto recovery_tree_root; 24895d4f98a2SYan Zheng dev_root->track_dirty = 1; 24903768f368SChris Mason 2491d20f7043SChris Mason ret = find_and_setup_root(tree_root, fs_info, 2492d20f7043SChris Mason BTRFS_CSUM_TREE_OBJECTID, csum_root); 2493d20f7043SChris Mason if (ret) 2494af31f5e5SChris Mason goto recovery_tree_root; 2495d20f7043SChris Mason csum_root->track_dirty = 1; 2496d20f7043SChris Mason 2497bcef60f2SArne Jansen ret = find_and_setup_root(tree_root, fs_info, 2498bcef60f2SArne Jansen BTRFS_QUOTA_TREE_OBJECTID, quota_root); 2499bcef60f2SArne Jansen if (ret) { 2500bcef60f2SArne Jansen kfree(quota_root); 2501bcef60f2SArne Jansen quota_root = fs_info->quota_root = NULL; 2502bcef60f2SArne Jansen } else { 2503bcef60f2SArne Jansen quota_root->track_dirty = 1; 2504bcef60f2SArne Jansen fs_info->quota_enabled = 1; 2505bcef60f2SArne Jansen fs_info->pending_quota_state = 1; 2506bcef60f2SArne Jansen } 2507bcef60f2SArne Jansen 25088929ecfaSYan, Zheng fs_info->generation = generation; 25098929ecfaSYan, Zheng fs_info->last_trans_committed = generation; 25108929ecfaSYan, Zheng 251168310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 251268310a5eSIlya Dryomov if (ret) { 251368310a5eSIlya Dryomov printk(KERN_WARNING "btrfs: failed to recover balance\n"); 251468310a5eSIlya Dryomov goto fail_block_groups; 251568310a5eSIlya Dryomov } 251668310a5eSIlya Dryomov 2517733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 2518733f4fbbSStefan Behrens if (ret) { 2519733f4fbbSStefan Behrens printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n", 2520733f4fbbSStefan Behrens ret); 2521733f4fbbSStefan Behrens goto fail_block_groups; 2522733f4fbbSStefan Behrens } 2523733f4fbbSStefan Behrens 25248dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 25258dabb742SStefan Behrens if (ret) { 25268dabb742SStefan Behrens pr_err("btrfs: failed to init dev_replace: %d\n", ret); 25278dabb742SStefan Behrens goto fail_block_groups; 25288dabb742SStefan Behrens } 25298dabb742SStefan Behrens 25308dabb742SStefan Behrens btrfs_close_extra_devices(fs_info, fs_devices, 1); 25318dabb742SStefan Behrens 2532c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 2533c59021f8Sliubo if (ret) { 2534c59021f8Sliubo printk(KERN_ERR "Failed to initial space info: %d\n", ret); 2535c59021f8Sliubo goto fail_block_groups; 2536c59021f8Sliubo } 2537c59021f8Sliubo 25381b1d1f66SJosef Bacik ret = btrfs_read_block_groups(extent_root); 25391b1d1f66SJosef Bacik if (ret) { 25401b1d1f66SJosef Bacik printk(KERN_ERR "Failed to read block groups: %d\n", ret); 25411b1d1f66SJosef Bacik goto fail_block_groups; 25421b1d1f66SJosef Bacik } 25435af3e8ccSStefan Behrens fs_info->num_tolerated_disk_barrier_failures = 25445af3e8ccSStefan Behrens btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); 2545292fd7fcSStefan Behrens if (fs_info->fs_devices->missing_devices > 2546292fd7fcSStefan Behrens fs_info->num_tolerated_disk_barrier_failures && 2547292fd7fcSStefan Behrens !(sb->s_flags & MS_RDONLY)) { 2548292fd7fcSStefan Behrens printk(KERN_WARNING 2549292fd7fcSStefan Behrens "Btrfs: too many missing devices, writeable mount is not allowed\n"); 2550292fd7fcSStefan Behrens goto fail_block_groups; 2551292fd7fcSStefan Behrens } 25529078a3e1SChris Mason 2553a74a4b97SChris Mason fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 2554a74a4b97SChris Mason "btrfs-cleaner"); 255557506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 25561b1d1f66SJosef Bacik goto fail_block_groups; 2557a74a4b97SChris Mason 2558a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 2559a74a4b97SChris Mason tree_root, 2560a74a4b97SChris Mason "btrfs-transaction"); 256157506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 25623f157a2fSChris Mason goto fail_cleaner; 2563a74a4b97SChris Mason 2564c289811cSChris Mason if (!btrfs_test_opt(tree_root, SSD) && 2565c289811cSChris Mason !btrfs_test_opt(tree_root, NOSSD) && 2566c289811cSChris Mason !fs_info->fs_devices->rotating) { 2567c289811cSChris Mason printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD " 2568c289811cSChris Mason "mode\n"); 2569c289811cSChris Mason btrfs_set_opt(fs_info->mount_opt, SSD); 2570c289811cSChris Mason } 2571c289811cSChris Mason 257221adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 257321adbd5cSStefan Behrens if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) { 257421adbd5cSStefan Behrens ret = btrfsic_mount(tree_root, fs_devices, 257521adbd5cSStefan Behrens btrfs_test_opt(tree_root, 257621adbd5cSStefan Behrens CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? 257721adbd5cSStefan Behrens 1 : 0, 257821adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 257921adbd5cSStefan Behrens if (ret) 258021adbd5cSStefan Behrens printk(KERN_WARNING "btrfs: failed to initialize" 258121adbd5cSStefan Behrens " integrity check module %s\n", sb->s_id); 258221adbd5cSStefan Behrens } 258321adbd5cSStefan Behrens #endif 2584bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 2585bcef60f2SArne Jansen if (ret) 2586bcef60f2SArne Jansen goto fail_trans_kthread; 258721adbd5cSStefan Behrens 2588acce952bSliubo /* do not make disk changes in broken FS */ 258968ce9682SStefan Behrens if (btrfs_super_log_root(disk_super) != 0) { 2590e02119d5SChris Mason u64 bytenr = btrfs_super_log_root(disk_super); 2591d18a2c44SChris Mason 25927c2ca468SChris Mason if (fs_devices->rw_devices == 0) { 2593d397712bSChris Mason printk(KERN_WARNING "Btrfs log replay required " 2594d397712bSChris Mason "on RO media\n"); 25957c2ca468SChris Mason err = -EIO; 2596bcef60f2SArne Jansen goto fail_qgroup; 25977c2ca468SChris Mason } 2598e02119d5SChris Mason blocksize = 2599e02119d5SChris Mason btrfs_level_size(tree_root, 2600e02119d5SChris Mason btrfs_super_log_root_level(disk_super)); 2601e02119d5SChris Mason 26026f07e42eSAl Viro log_tree_root = btrfs_alloc_root(fs_info); 2603676e4c86SDan Carpenter if (!log_tree_root) { 2604676e4c86SDan Carpenter err = -ENOMEM; 2605bcef60f2SArne Jansen goto fail_qgroup; 2606676e4c86SDan Carpenter } 2607e02119d5SChris Mason 2608e02119d5SChris Mason __setup_root(nodesize, leafsize, sectorsize, stripesize, 2609e02119d5SChris Mason log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID); 2610e02119d5SChris Mason 2611e02119d5SChris Mason log_tree_root->node = read_tree_block(tree_root, bytenr, 261284234f3aSYan Zheng blocksize, 261384234f3aSYan Zheng generation + 1); 261479787eaaSJeff Mahoney /* returns with log_tree_root freed on success */ 2615e02119d5SChris Mason ret = btrfs_recover_log_trees(log_tree_root); 261679787eaaSJeff Mahoney if (ret) { 261779787eaaSJeff Mahoney btrfs_error(tree_root->fs_info, ret, 261879787eaaSJeff Mahoney "Failed to recover log tree"); 261979787eaaSJeff Mahoney free_extent_buffer(log_tree_root->node); 262079787eaaSJeff Mahoney kfree(log_tree_root); 262179787eaaSJeff Mahoney goto fail_trans_kthread; 262279787eaaSJeff Mahoney } 2623e556ce2cSYan Zheng 2624e556ce2cSYan Zheng if (sb->s_flags & MS_RDONLY) { 2625e556ce2cSYan Zheng ret = btrfs_commit_super(tree_root); 262679787eaaSJeff Mahoney if (ret) 262779787eaaSJeff Mahoney goto fail_trans_kthread; 2628e556ce2cSYan Zheng } 2629e02119d5SChris Mason } 26301a40e23bSZheng Yan 263176dda93cSYan, Zheng ret = btrfs_find_orphan_roots(tree_root); 263279787eaaSJeff Mahoney if (ret) 263379787eaaSJeff Mahoney goto fail_trans_kthread; 263476dda93cSYan, Zheng 26357c2ca468SChris Mason if (!(sb->s_flags & MS_RDONLY)) { 2636d68fc57bSYan, Zheng ret = btrfs_cleanup_fs_roots(fs_info); 263744c44af2SIlya Dryomov if (ret) 263844c44af2SIlya Dryomov goto fail_trans_kthread; 2639d68fc57bSYan, Zheng 26405d4f98a2SYan Zheng ret = btrfs_recover_relocation(tree_root); 2641d7ce5843SMiao Xie if (ret < 0) { 2642d7ce5843SMiao Xie printk(KERN_WARNING 2643d7ce5843SMiao Xie "btrfs: failed to recover relocation\n"); 2644d7ce5843SMiao Xie err = -EINVAL; 2645bcef60f2SArne Jansen goto fail_qgroup; 2646d7ce5843SMiao Xie } 26477c2ca468SChris Mason } 26481a40e23bSZheng Yan 26493de4586cSChris Mason location.objectid = BTRFS_FS_TREE_OBJECTID; 26503de4586cSChris Mason location.type = BTRFS_ROOT_ITEM_KEY; 26513de4586cSChris Mason location.offset = (u64)-1; 26523de4586cSChris Mason 26533de4586cSChris Mason fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location); 26543de4586cSChris Mason if (!fs_info->fs_root) 2655bcef60f2SArne Jansen goto fail_qgroup; 26563140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 26573140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 2658bcef60f2SArne Jansen goto fail_qgroup; 26593140c9a3SDan Carpenter } 2660c289811cSChris Mason 26612b6ba629SIlya Dryomov if (sb->s_flags & MS_RDONLY) 26622b6ba629SIlya Dryomov return 0; 26632b6ba629SIlya Dryomov 2664e3acc2a6SJosef Bacik down_read(&fs_info->cleanup_work_sem); 26652b6ba629SIlya Dryomov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 26662b6ba629SIlya Dryomov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 26672b6ba629SIlya Dryomov up_read(&fs_info->cleanup_work_sem); 26682b6ba629SIlya Dryomov close_ctree(tree_root); 26692b6ba629SIlya Dryomov return ret; 26702b6ba629SIlya Dryomov } 2671e3acc2a6SJosef Bacik up_read(&fs_info->cleanup_work_sem); 267259641015SIlya Dryomov 26732b6ba629SIlya Dryomov ret = btrfs_resume_balance_async(fs_info); 26742b6ba629SIlya Dryomov if (ret) { 26752b6ba629SIlya Dryomov printk(KERN_WARNING "btrfs: failed to resume balance\n"); 267666b4ffd1SJosef Bacik close_ctree(tree_root); 26772b6ba629SIlya Dryomov return ret; 2678e3acc2a6SJosef Bacik } 2679e3acc2a6SJosef Bacik 26808dabb742SStefan Behrens ret = btrfs_resume_dev_replace_async(fs_info); 26818dabb742SStefan Behrens if (ret) { 26828dabb742SStefan Behrens pr_warn("btrfs: failed to resume dev_replace\n"); 26838dabb742SStefan Behrens close_ctree(tree_root); 26848dabb742SStefan Behrens return ret; 26858dabb742SStefan Behrens } 26868dabb742SStefan Behrens 2687ad2b2c80SAl Viro return 0; 268839279cc3SChris Mason 2689bcef60f2SArne Jansen fail_qgroup: 2690bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 26917c2ca468SChris Mason fail_trans_kthread: 26927c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 26933f157a2fSChris Mason fail_cleaner: 2694a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 26957c2ca468SChris Mason 26967c2ca468SChris Mason /* 26977c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 26987c2ca468SChris Mason * kthreads 26997c2ca468SChris Mason */ 27007c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 27017c2ca468SChris Mason invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 27027c2ca468SChris Mason 27031b1d1f66SJosef Bacik fail_block_groups: 27041b1d1f66SJosef Bacik btrfs_free_block_groups(fs_info); 2705af31f5e5SChris Mason 2706af31f5e5SChris Mason fail_tree_roots: 2707af31f5e5SChris Mason free_root_pointers(fs_info, 1); 2708af31f5e5SChris Mason 270939279cc3SChris Mason fail_sb_buffer: 271061d92c32SChris Mason btrfs_stop_workers(&fs_info->generic_worker); 2711306c8b68SChris Mason btrfs_stop_workers(&fs_info->readahead_workers); 2712247e743cSChris Mason btrfs_stop_workers(&fs_info->fixup_workers); 2713771ed689SChris Mason btrfs_stop_workers(&fs_info->delalloc_workers); 27148b712842SChris Mason btrfs_stop_workers(&fs_info->workers); 27158b712842SChris Mason btrfs_stop_workers(&fs_info->endio_workers); 2716d20f7043SChris Mason btrfs_stop_workers(&fs_info->endio_meta_workers); 2717cad321adSChris Mason btrfs_stop_workers(&fs_info->endio_meta_write_workers); 2718e6dcd2dcSChris Mason btrfs_stop_workers(&fs_info->endio_write_workers); 27190cb59c99SJosef Bacik btrfs_stop_workers(&fs_info->endio_freespace_worker); 27201cc127b5SChris Mason btrfs_stop_workers(&fs_info->submit_workers); 272116cdcec7SMiao Xie btrfs_stop_workers(&fs_info->delayed_workers); 2722bab39bf9SJosef Bacik btrfs_stop_workers(&fs_info->caching_workers); 27238ccf6f19SMiao Xie btrfs_stop_workers(&fs_info->flush_workers); 272416cdcec7SMiao Xie fail_alloc: 27254543df7eSChris Mason fail_iput: 2726586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 2727586e46e2SIlya Dryomov 27287c2ca468SChris Mason invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 27294543df7eSChris Mason iput(fs_info->btree_inode); 2730ad081f14SJens Axboe fail_bdi: 27317e662854SQinghuang Feng bdi_destroy(&fs_info->bdi); 273276dda93cSYan, Zheng fail_srcu: 273376dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 27347e662854SQinghuang Feng fail: 2735586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 2736ad2b2c80SAl Viro return err; 2737af31f5e5SChris Mason 2738af31f5e5SChris Mason recovery_tree_root: 2739af31f5e5SChris Mason if (!btrfs_test_opt(tree_root, RECOVERY)) 2740af31f5e5SChris Mason goto fail_tree_roots; 2741af31f5e5SChris Mason 2742af31f5e5SChris Mason free_root_pointers(fs_info, 0); 2743af31f5e5SChris Mason 2744af31f5e5SChris Mason /* don't use the log in recovery mode, it won't be valid */ 2745af31f5e5SChris Mason btrfs_set_super_log_root(disk_super, 0); 2746af31f5e5SChris Mason 2747af31f5e5SChris Mason /* we can't trust the free space cache either */ 2748af31f5e5SChris Mason btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 2749af31f5e5SChris Mason 2750af31f5e5SChris Mason ret = next_root_backup(fs_info, fs_info->super_copy, 2751af31f5e5SChris Mason &num_backups_tried, &backup_index); 2752af31f5e5SChris Mason if (ret == -1) 2753af31f5e5SChris Mason goto fail_block_groups; 2754af31f5e5SChris Mason goto retry_root_backup; 2755eb60ceacSChris Mason } 2756eb60ceacSChris Mason 2757f2984462SChris Mason static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate) 2758f2984462SChris Mason { 2759f2984462SChris Mason if (uptodate) { 2760f2984462SChris Mason set_buffer_uptodate(bh); 2761f2984462SChris Mason } else { 2762442a4f63SStefan Behrens struct btrfs_device *device = (struct btrfs_device *) 2763442a4f63SStefan Behrens bh->b_private; 2764442a4f63SStefan Behrens 2765606686eeSJosef Bacik printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to " 2766606686eeSJosef Bacik "I/O error on %s\n", 2767606686eeSJosef Bacik rcu_str_deref(device->name)); 27681259ab75SChris Mason /* note, we dont' set_buffer_write_io_error because we have 27691259ab75SChris Mason * our own ways of dealing with the IO errors 27701259ab75SChris Mason */ 2771f2984462SChris Mason clear_buffer_uptodate(bh); 2772442a4f63SStefan Behrens btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS); 2773f2984462SChris Mason } 2774f2984462SChris Mason unlock_buffer(bh); 2775f2984462SChris Mason put_bh(bh); 2776f2984462SChris Mason } 2777f2984462SChris Mason 2778a512bbf8SYan Zheng struct buffer_head *btrfs_read_dev_super(struct block_device *bdev) 2779a512bbf8SYan Zheng { 2780a512bbf8SYan Zheng struct buffer_head *bh; 2781a512bbf8SYan Zheng struct buffer_head *latest = NULL; 2782a512bbf8SYan Zheng struct btrfs_super_block *super; 2783a512bbf8SYan Zheng int i; 2784a512bbf8SYan Zheng u64 transid = 0; 2785a512bbf8SYan Zheng u64 bytenr; 2786a512bbf8SYan Zheng 2787a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 2788a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 2789a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 2790a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 2791a512bbf8SYan Zheng */ 2792a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 2793a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 2794a512bbf8SYan Zheng if (bytenr + 4096 >= i_size_read(bdev->bd_inode)) 2795a512bbf8SYan Zheng break; 2796a512bbf8SYan Zheng bh = __bread(bdev, bytenr / 4096, 4096); 2797a512bbf8SYan Zheng if (!bh) 2798a512bbf8SYan Zheng continue; 2799a512bbf8SYan Zheng 2800a512bbf8SYan Zheng super = (struct btrfs_super_block *)bh->b_data; 2801a512bbf8SYan Zheng if (btrfs_super_bytenr(super) != bytenr || 2802a512bbf8SYan Zheng strncmp((char *)(&super->magic), BTRFS_MAGIC, 2803a512bbf8SYan Zheng sizeof(super->magic))) { 2804a512bbf8SYan Zheng brelse(bh); 2805a512bbf8SYan Zheng continue; 2806a512bbf8SYan Zheng } 2807a512bbf8SYan Zheng 2808a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 2809a512bbf8SYan Zheng brelse(latest); 2810a512bbf8SYan Zheng latest = bh; 2811a512bbf8SYan Zheng transid = btrfs_super_generation(super); 2812a512bbf8SYan Zheng } else { 2813a512bbf8SYan Zheng brelse(bh); 2814a512bbf8SYan Zheng } 2815a512bbf8SYan Zheng } 2816a512bbf8SYan Zheng return latest; 2817a512bbf8SYan Zheng } 2818a512bbf8SYan Zheng 28194eedeb75SHisashi Hifumi /* 28204eedeb75SHisashi Hifumi * this should be called twice, once with wait == 0 and 28214eedeb75SHisashi Hifumi * once with wait == 1. When wait == 0 is done, all the buffer heads 28224eedeb75SHisashi Hifumi * we write are pinned. 28234eedeb75SHisashi Hifumi * 28244eedeb75SHisashi Hifumi * They are released when wait == 1 is done. 28254eedeb75SHisashi Hifumi * max_mirrors must be the same for both runs, and it indicates how 28264eedeb75SHisashi Hifumi * many supers on this one device should be written. 28274eedeb75SHisashi Hifumi * 28284eedeb75SHisashi Hifumi * max_mirrors == 0 means to write them all. 28294eedeb75SHisashi Hifumi */ 2830a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 2831a512bbf8SYan Zheng struct btrfs_super_block *sb, 2832a512bbf8SYan Zheng int do_barriers, int wait, int max_mirrors) 2833a512bbf8SYan Zheng { 2834a512bbf8SYan Zheng struct buffer_head *bh; 2835a512bbf8SYan Zheng int i; 2836a512bbf8SYan Zheng int ret; 2837a512bbf8SYan Zheng int errors = 0; 2838a512bbf8SYan Zheng u32 crc; 2839a512bbf8SYan Zheng u64 bytenr; 2840a512bbf8SYan Zheng 2841a512bbf8SYan Zheng if (max_mirrors == 0) 2842a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 2843a512bbf8SYan Zheng 2844a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 2845a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 2846a512bbf8SYan Zheng if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes) 2847a512bbf8SYan Zheng break; 2848a512bbf8SYan Zheng 2849a512bbf8SYan Zheng if (wait) { 2850a512bbf8SYan Zheng bh = __find_get_block(device->bdev, bytenr / 4096, 2851a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 2852a512bbf8SYan Zheng BUG_ON(!bh); 2853a512bbf8SYan Zheng wait_on_buffer(bh); 28544eedeb75SHisashi Hifumi if (!buffer_uptodate(bh)) 28554eedeb75SHisashi Hifumi errors++; 28564eedeb75SHisashi Hifumi 28574eedeb75SHisashi Hifumi /* drop our reference */ 28584eedeb75SHisashi Hifumi brelse(bh); 28594eedeb75SHisashi Hifumi 28604eedeb75SHisashi Hifumi /* drop the reference from the wait == 0 run */ 2861a512bbf8SYan Zheng brelse(bh); 2862a512bbf8SYan Zheng continue; 2863a512bbf8SYan Zheng } else { 2864a512bbf8SYan Zheng btrfs_set_super_bytenr(sb, bytenr); 2865a512bbf8SYan Zheng 2866a512bbf8SYan Zheng crc = ~(u32)0; 2867a512bbf8SYan Zheng crc = btrfs_csum_data(NULL, (char *)sb + 2868a512bbf8SYan Zheng BTRFS_CSUM_SIZE, crc, 2869a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE - 2870a512bbf8SYan Zheng BTRFS_CSUM_SIZE); 2871a512bbf8SYan Zheng btrfs_csum_final(crc, sb->csum); 2872a512bbf8SYan Zheng 28734eedeb75SHisashi Hifumi /* 28744eedeb75SHisashi Hifumi * one reference for us, and we leave it for the 28754eedeb75SHisashi Hifumi * caller 28764eedeb75SHisashi Hifumi */ 2877a512bbf8SYan Zheng bh = __getblk(device->bdev, bytenr / 4096, 2878a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 2879a512bbf8SYan Zheng memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE); 2880a512bbf8SYan Zheng 28814eedeb75SHisashi Hifumi /* one reference for submit_bh */ 2882a512bbf8SYan Zheng get_bh(bh); 28834eedeb75SHisashi Hifumi 28844eedeb75SHisashi Hifumi set_buffer_uptodate(bh); 2885a512bbf8SYan Zheng lock_buffer(bh); 2886a512bbf8SYan Zheng bh->b_end_io = btrfs_end_buffer_write_sync; 2887442a4f63SStefan Behrens bh->b_private = device; 2888a512bbf8SYan Zheng } 2889a512bbf8SYan Zheng 2890387125fcSChris Mason /* 2891387125fcSChris Mason * we fua the first super. The others we allow 2892387125fcSChris Mason * to go down lazy. 2893387125fcSChris Mason */ 289421adbd5cSStefan Behrens ret = btrfsic_submit_bh(WRITE_FUA, bh); 28954eedeb75SHisashi Hifumi if (ret) 2896a512bbf8SYan Zheng errors++; 2897a512bbf8SYan Zheng } 2898a512bbf8SYan Zheng return errors < i ? 0 : -1; 2899a512bbf8SYan Zheng } 2900a512bbf8SYan Zheng 2901387125fcSChris Mason /* 2902387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 2903387125fcSChris Mason * for the barrier when it is done 2904387125fcSChris Mason */ 2905387125fcSChris Mason static void btrfs_end_empty_barrier(struct bio *bio, int err) 2906387125fcSChris Mason { 2907387125fcSChris Mason if (err) { 2908387125fcSChris Mason if (err == -EOPNOTSUPP) 2909387125fcSChris Mason set_bit(BIO_EOPNOTSUPP, &bio->bi_flags); 2910387125fcSChris Mason clear_bit(BIO_UPTODATE, &bio->bi_flags); 2911387125fcSChris Mason } 2912387125fcSChris Mason if (bio->bi_private) 2913387125fcSChris Mason complete(bio->bi_private); 2914387125fcSChris Mason bio_put(bio); 2915387125fcSChris Mason } 2916387125fcSChris Mason 2917387125fcSChris Mason /* 2918387125fcSChris Mason * trigger flushes for one the devices. If you pass wait == 0, the flushes are 2919387125fcSChris Mason * sent down. With wait == 1, it waits for the previous flush. 2920387125fcSChris Mason * 2921387125fcSChris Mason * any device where the flush fails with eopnotsupp are flagged as not-barrier 2922387125fcSChris Mason * capable 2923387125fcSChris Mason */ 2924387125fcSChris Mason static int write_dev_flush(struct btrfs_device *device, int wait) 2925387125fcSChris Mason { 2926387125fcSChris Mason struct bio *bio; 2927387125fcSChris Mason int ret = 0; 2928387125fcSChris Mason 2929387125fcSChris Mason if (device->nobarriers) 2930387125fcSChris Mason return 0; 2931387125fcSChris Mason 2932387125fcSChris Mason if (wait) { 2933387125fcSChris Mason bio = device->flush_bio; 2934387125fcSChris Mason if (!bio) 2935387125fcSChris Mason return 0; 2936387125fcSChris Mason 2937387125fcSChris Mason wait_for_completion(&device->flush_wait); 2938387125fcSChris Mason 2939387125fcSChris Mason if (bio_flagged(bio, BIO_EOPNOTSUPP)) { 2940606686eeSJosef Bacik printk_in_rcu("btrfs: disabling barriers on dev %s\n", 2941606686eeSJosef Bacik rcu_str_deref(device->name)); 2942387125fcSChris Mason device->nobarriers = 1; 29435af3e8ccSStefan Behrens } else if (!bio_flagged(bio, BIO_UPTODATE)) { 2944387125fcSChris Mason ret = -EIO; 2945442a4f63SStefan Behrens btrfs_dev_stat_inc_and_print(device, 2946442a4f63SStefan Behrens BTRFS_DEV_STAT_FLUSH_ERRS); 2947387125fcSChris Mason } 2948387125fcSChris Mason 2949387125fcSChris Mason /* drop the reference from the wait == 0 run */ 2950387125fcSChris Mason bio_put(bio); 2951387125fcSChris Mason device->flush_bio = NULL; 2952387125fcSChris Mason 2953387125fcSChris Mason return ret; 2954387125fcSChris Mason } 2955387125fcSChris Mason 2956387125fcSChris Mason /* 2957387125fcSChris Mason * one reference for us, and we leave it for the 2958387125fcSChris Mason * caller 2959387125fcSChris Mason */ 29609c017abcSJesper Juhl device->flush_bio = NULL; 2961387125fcSChris Mason bio = bio_alloc(GFP_NOFS, 0); 2962387125fcSChris Mason if (!bio) 2963387125fcSChris Mason return -ENOMEM; 2964387125fcSChris Mason 2965387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 2966387125fcSChris Mason bio->bi_bdev = device->bdev; 2967387125fcSChris Mason init_completion(&device->flush_wait); 2968387125fcSChris Mason bio->bi_private = &device->flush_wait; 2969387125fcSChris Mason device->flush_bio = bio; 2970387125fcSChris Mason 2971387125fcSChris Mason bio_get(bio); 297221adbd5cSStefan Behrens btrfsic_submit_bio(WRITE_FLUSH, bio); 2973387125fcSChris Mason 2974387125fcSChris Mason return 0; 2975387125fcSChris Mason } 2976387125fcSChris Mason 2977387125fcSChris Mason /* 2978387125fcSChris Mason * send an empty flush down to each device in parallel, 2979387125fcSChris Mason * then wait for them 2980387125fcSChris Mason */ 2981387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 2982387125fcSChris Mason { 2983387125fcSChris Mason struct list_head *head; 2984387125fcSChris Mason struct btrfs_device *dev; 29855af3e8ccSStefan Behrens int errors_send = 0; 29865af3e8ccSStefan Behrens int errors_wait = 0; 2987387125fcSChris Mason int ret; 2988387125fcSChris Mason 2989387125fcSChris Mason /* send down all the barriers */ 2990387125fcSChris Mason head = &info->fs_devices->devices; 2991387125fcSChris Mason list_for_each_entry_rcu(dev, head, dev_list) { 2992387125fcSChris Mason if (!dev->bdev) { 29935af3e8ccSStefan Behrens errors_send++; 2994387125fcSChris Mason continue; 2995387125fcSChris Mason } 2996387125fcSChris Mason if (!dev->in_fs_metadata || !dev->writeable) 2997387125fcSChris Mason continue; 2998387125fcSChris Mason 2999387125fcSChris Mason ret = write_dev_flush(dev, 0); 3000387125fcSChris Mason if (ret) 30015af3e8ccSStefan Behrens errors_send++; 3002387125fcSChris Mason } 3003387125fcSChris Mason 3004387125fcSChris Mason /* wait for all the barriers */ 3005387125fcSChris Mason list_for_each_entry_rcu(dev, head, dev_list) { 3006387125fcSChris Mason if (!dev->bdev) { 30075af3e8ccSStefan Behrens errors_wait++; 3008387125fcSChris Mason continue; 3009387125fcSChris Mason } 3010387125fcSChris Mason if (!dev->in_fs_metadata || !dev->writeable) 3011387125fcSChris Mason continue; 3012387125fcSChris Mason 3013387125fcSChris Mason ret = write_dev_flush(dev, 1); 3014387125fcSChris Mason if (ret) 30155af3e8ccSStefan Behrens errors_wait++; 3016387125fcSChris Mason } 30175af3e8ccSStefan Behrens if (errors_send > info->num_tolerated_disk_barrier_failures || 30185af3e8ccSStefan Behrens errors_wait > info->num_tolerated_disk_barrier_failures) 3019387125fcSChris Mason return -EIO; 3020387125fcSChris Mason return 0; 3021387125fcSChris Mason } 3022387125fcSChris Mason 30235af3e8ccSStefan Behrens int btrfs_calc_num_tolerated_disk_barrier_failures( 30245af3e8ccSStefan Behrens struct btrfs_fs_info *fs_info) 30255af3e8ccSStefan Behrens { 30265af3e8ccSStefan Behrens struct btrfs_ioctl_space_info space; 30275af3e8ccSStefan Behrens struct btrfs_space_info *sinfo; 30285af3e8ccSStefan Behrens u64 types[] = {BTRFS_BLOCK_GROUP_DATA, 30295af3e8ccSStefan Behrens BTRFS_BLOCK_GROUP_SYSTEM, 30305af3e8ccSStefan Behrens BTRFS_BLOCK_GROUP_METADATA, 30315af3e8ccSStefan Behrens BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA}; 30325af3e8ccSStefan Behrens int num_types = 4; 30335af3e8ccSStefan Behrens int i; 30345af3e8ccSStefan Behrens int c; 30355af3e8ccSStefan Behrens int num_tolerated_disk_barrier_failures = 30365af3e8ccSStefan Behrens (int)fs_info->fs_devices->num_devices; 30375af3e8ccSStefan Behrens 30385af3e8ccSStefan Behrens for (i = 0; i < num_types; i++) { 30395af3e8ccSStefan Behrens struct btrfs_space_info *tmp; 30405af3e8ccSStefan Behrens 30415af3e8ccSStefan Behrens sinfo = NULL; 30425af3e8ccSStefan Behrens rcu_read_lock(); 30435af3e8ccSStefan Behrens list_for_each_entry_rcu(tmp, &fs_info->space_info, list) { 30445af3e8ccSStefan Behrens if (tmp->flags == types[i]) { 30455af3e8ccSStefan Behrens sinfo = tmp; 30465af3e8ccSStefan Behrens break; 30475af3e8ccSStefan Behrens } 30485af3e8ccSStefan Behrens } 30495af3e8ccSStefan Behrens rcu_read_unlock(); 30505af3e8ccSStefan Behrens 30515af3e8ccSStefan Behrens if (!sinfo) 30525af3e8ccSStefan Behrens continue; 30535af3e8ccSStefan Behrens 30545af3e8ccSStefan Behrens down_read(&sinfo->groups_sem); 30555af3e8ccSStefan Behrens for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 30565af3e8ccSStefan Behrens if (!list_empty(&sinfo->block_groups[c])) { 30575af3e8ccSStefan Behrens u64 flags; 30585af3e8ccSStefan Behrens 30595af3e8ccSStefan Behrens btrfs_get_block_group_info( 30605af3e8ccSStefan Behrens &sinfo->block_groups[c], &space); 30615af3e8ccSStefan Behrens if (space.total_bytes == 0 || 30625af3e8ccSStefan Behrens space.used_bytes == 0) 30635af3e8ccSStefan Behrens continue; 30645af3e8ccSStefan Behrens flags = space.flags; 30655af3e8ccSStefan Behrens /* 30665af3e8ccSStefan Behrens * return 30675af3e8ccSStefan Behrens * 0: if dup, single or RAID0 is configured for 30685af3e8ccSStefan Behrens * any of metadata, system or data, else 30695af3e8ccSStefan Behrens * 1: if RAID5 is configured, or if RAID1 or 30705af3e8ccSStefan Behrens * RAID10 is configured and only two mirrors 30715af3e8ccSStefan Behrens * are used, else 30725af3e8ccSStefan Behrens * 2: if RAID6 is configured, else 30735af3e8ccSStefan Behrens * num_mirrors - 1: if RAID1 or RAID10 is 30745af3e8ccSStefan Behrens * configured and more than 30755af3e8ccSStefan Behrens * 2 mirrors are used. 30765af3e8ccSStefan Behrens */ 30775af3e8ccSStefan Behrens if (num_tolerated_disk_barrier_failures > 0 && 30785af3e8ccSStefan Behrens ((flags & (BTRFS_BLOCK_GROUP_DUP | 30795af3e8ccSStefan Behrens BTRFS_BLOCK_GROUP_RAID0)) || 30805af3e8ccSStefan Behrens ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) 30815af3e8ccSStefan Behrens == 0))) 30825af3e8ccSStefan Behrens num_tolerated_disk_barrier_failures = 0; 30835af3e8ccSStefan Behrens else if (num_tolerated_disk_barrier_failures > 1 30845af3e8ccSStefan Behrens && 30855af3e8ccSStefan Behrens (flags & (BTRFS_BLOCK_GROUP_RAID1 | 30865af3e8ccSStefan Behrens BTRFS_BLOCK_GROUP_RAID10))) 30875af3e8ccSStefan Behrens num_tolerated_disk_barrier_failures = 1; 30885af3e8ccSStefan Behrens } 30895af3e8ccSStefan Behrens } 30905af3e8ccSStefan Behrens up_read(&sinfo->groups_sem); 30915af3e8ccSStefan Behrens } 30925af3e8ccSStefan Behrens 30935af3e8ccSStefan Behrens return num_tolerated_disk_barrier_failures; 30945af3e8ccSStefan Behrens } 30955af3e8ccSStefan Behrens 3096a512bbf8SYan Zheng int write_all_supers(struct btrfs_root *root, int max_mirrors) 3097f2984462SChris Mason { 3098e5e9a520SChris Mason struct list_head *head; 3099f2984462SChris Mason struct btrfs_device *dev; 3100a061fc8dSChris Mason struct btrfs_super_block *sb; 3101f2984462SChris Mason struct btrfs_dev_item *dev_item; 3102f2984462SChris Mason int ret; 3103f2984462SChris Mason int do_barriers; 3104a236aed1SChris Mason int max_errors; 3105a236aed1SChris Mason int total_errors = 0; 3106a061fc8dSChris Mason u64 flags; 3107f2984462SChris Mason 31086c41761fSDavid Sterba max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1; 3109f2984462SChris Mason do_barriers = !btrfs_test_opt(root, NOBARRIER); 3110af31f5e5SChris Mason backup_super_roots(root->fs_info); 3111f2984462SChris Mason 31126c41761fSDavid Sterba sb = root->fs_info->super_for_commit; 3113a061fc8dSChris Mason dev_item = &sb->dev_item; 3114e5e9a520SChris Mason 3115174ba509SChris Mason mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 3116e5e9a520SChris Mason head = &root->fs_info->fs_devices->devices; 3117387125fcSChris Mason 31185af3e8ccSStefan Behrens if (do_barriers) { 31195af3e8ccSStefan Behrens ret = barrier_all_devices(root->fs_info); 31205af3e8ccSStefan Behrens if (ret) { 31215af3e8ccSStefan Behrens mutex_unlock( 31225af3e8ccSStefan Behrens &root->fs_info->fs_devices->device_list_mutex); 31235af3e8ccSStefan Behrens btrfs_error(root->fs_info, ret, 31245af3e8ccSStefan Behrens "errors while submitting device barriers."); 31255af3e8ccSStefan Behrens return ret; 31265af3e8ccSStefan Behrens } 31275af3e8ccSStefan Behrens } 3128387125fcSChris Mason 31291f78160cSXiao Guangrong list_for_each_entry_rcu(dev, head, dev_list) { 3130dfe25020SChris Mason if (!dev->bdev) { 3131dfe25020SChris Mason total_errors++; 3132dfe25020SChris Mason continue; 3133dfe25020SChris Mason } 31342b82032cSYan Zheng if (!dev->in_fs_metadata || !dev->writeable) 3135dfe25020SChris Mason continue; 3136dfe25020SChris Mason 31372b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 3138a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 3139a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 3140a061fc8dSChris Mason btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes); 3141a061fc8dSChris Mason btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used); 3142a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 3143a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 3144a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 3145a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 31462b82032cSYan Zheng memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE); 3147a512bbf8SYan Zheng 3148a061fc8dSChris Mason flags = btrfs_super_flags(sb); 3149a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 3150f2984462SChris Mason 3151a512bbf8SYan Zheng ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors); 3152a236aed1SChris Mason if (ret) 3153a236aed1SChris Mason total_errors++; 3154f2984462SChris Mason } 3155a236aed1SChris Mason if (total_errors > max_errors) { 3156d397712bSChris Mason printk(KERN_ERR "btrfs: %d errors while writing supers\n", 3157d397712bSChris Mason total_errors); 315879787eaaSJeff Mahoney 315979787eaaSJeff Mahoney /* This shouldn't happen. FUA is masked off if unsupported */ 3160a236aed1SChris Mason BUG(); 3161a236aed1SChris Mason } 3162f2984462SChris Mason 3163a512bbf8SYan Zheng total_errors = 0; 31641f78160cSXiao Guangrong list_for_each_entry_rcu(dev, head, dev_list) { 3165dfe25020SChris Mason if (!dev->bdev) 3166dfe25020SChris Mason continue; 31672b82032cSYan Zheng if (!dev->in_fs_metadata || !dev->writeable) 3168dfe25020SChris Mason continue; 3169dfe25020SChris Mason 3170a512bbf8SYan Zheng ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors); 3171a512bbf8SYan Zheng if (ret) 31721259ab75SChris Mason total_errors++; 3173f2984462SChris Mason } 3174174ba509SChris Mason mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 3175a236aed1SChris Mason if (total_errors > max_errors) { 317679787eaaSJeff Mahoney btrfs_error(root->fs_info, -EIO, 317779787eaaSJeff Mahoney "%d errors while writing supers", total_errors); 317879787eaaSJeff Mahoney return -EIO; 3179a236aed1SChris Mason } 3180f2984462SChris Mason return 0; 3181f2984462SChris Mason } 3182f2984462SChris Mason 3183a512bbf8SYan Zheng int write_ctree_super(struct btrfs_trans_handle *trans, 3184a512bbf8SYan Zheng struct btrfs_root *root, int max_mirrors) 3185cfaa7295SChris Mason { 3186e66f709bSChris Mason int ret; 31872c90e5d6SChris Mason 3188a512bbf8SYan Zheng ret = write_all_supers(root, max_mirrors); 31895f39d397SChris Mason return ret; 3190cfaa7295SChris Mason } 3191cfaa7295SChris Mason 3192143bede5SJeff Mahoney void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root) 31932619ba1fSChris Mason { 31944df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 31952619ba1fSChris Mason radix_tree_delete(&fs_info->fs_roots_radix, 31962619ba1fSChris Mason (unsigned long)root->root_key.objectid); 31974df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 319876dda93cSYan, Zheng 319976dda93cSYan, Zheng if (btrfs_root_refs(&root->root_item) == 0) 320076dda93cSYan, Zheng synchronize_srcu(&fs_info->subvol_srcu); 320176dda93cSYan, Zheng 3202581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_pinned); 3203581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_ctl); 32044df27c4dSYan, Zheng free_fs_root(root); 32054df27c4dSYan, Zheng } 32064df27c4dSYan, Zheng 32074df27c4dSYan, Zheng static void free_fs_root(struct btrfs_root *root) 32084df27c4dSYan, Zheng { 320982d5902dSLi Zefan iput(root->cache_inode); 32104df27c4dSYan, Zheng WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 32110ee5dc67SAl Viro if (root->anon_dev) 32120ee5dc67SAl Viro free_anon_bdev(root->anon_dev); 32135f39d397SChris Mason free_extent_buffer(root->node); 32145f39d397SChris Mason free_extent_buffer(root->commit_root); 3215581bb050SLi Zefan kfree(root->free_ino_ctl); 3216581bb050SLi Zefan kfree(root->free_ino_pinned); 321758176a96SJosef Bacik kfree(root->name); 32182619ba1fSChris Mason kfree(root); 32192619ba1fSChris Mason } 32202619ba1fSChris Mason 3221143bede5SJeff Mahoney static void del_fs_roots(struct btrfs_fs_info *fs_info) 32220f7d52f4SChris Mason { 32230f7d52f4SChris Mason int ret; 32240f7d52f4SChris Mason struct btrfs_root *gang[8]; 32250f7d52f4SChris Mason int i; 32260f7d52f4SChris Mason 322776dda93cSYan, Zheng while (!list_empty(&fs_info->dead_roots)) { 322876dda93cSYan, Zheng gang[0] = list_entry(fs_info->dead_roots.next, 322976dda93cSYan, Zheng struct btrfs_root, root_list); 323076dda93cSYan, Zheng list_del(&gang[0]->root_list); 323176dda93cSYan, Zheng 323276dda93cSYan, Zheng if (gang[0]->in_radix) { 323376dda93cSYan, Zheng btrfs_free_fs_root(fs_info, gang[0]); 323476dda93cSYan, Zheng } else { 323576dda93cSYan, Zheng free_extent_buffer(gang[0]->node); 323676dda93cSYan, Zheng free_extent_buffer(gang[0]->commit_root); 323776dda93cSYan, Zheng kfree(gang[0]); 323876dda93cSYan, Zheng } 323976dda93cSYan, Zheng } 324076dda93cSYan, Zheng 32410f7d52f4SChris Mason while (1) { 32420f7d52f4SChris Mason ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 32430f7d52f4SChris Mason (void **)gang, 0, 32440f7d52f4SChris Mason ARRAY_SIZE(gang)); 32450f7d52f4SChris Mason if (!ret) 32460f7d52f4SChris Mason break; 32472619ba1fSChris Mason for (i = 0; i < ret; i++) 32485eda7b5eSChris Mason btrfs_free_fs_root(fs_info, gang[i]); 32490f7d52f4SChris Mason } 32500f7d52f4SChris Mason } 3251b4100d64SChris Mason 3252c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 3253c146afadSYan Zheng { 3254c146afadSYan Zheng u64 root_objectid = 0; 3255c146afadSYan Zheng struct btrfs_root *gang[8]; 3256c146afadSYan Zheng int i; 3257c146afadSYan Zheng int ret; 3258c146afadSYan Zheng 3259c146afadSYan Zheng while (1) { 3260c146afadSYan Zheng ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 3261c146afadSYan Zheng (void **)gang, root_objectid, 3262c146afadSYan Zheng ARRAY_SIZE(gang)); 3263c146afadSYan Zheng if (!ret) 3264c146afadSYan Zheng break; 32655d4f98a2SYan Zheng 32665d4f98a2SYan Zheng root_objectid = gang[ret - 1]->root_key.objectid + 1; 3267c146afadSYan Zheng for (i = 0; i < ret; i++) { 326866b4ffd1SJosef Bacik int err; 326966b4ffd1SJosef Bacik 3270c146afadSYan Zheng root_objectid = gang[i]->root_key.objectid; 327166b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(gang[i]); 327266b4ffd1SJosef Bacik if (err) 327366b4ffd1SJosef Bacik return err; 3274c146afadSYan Zheng } 3275c146afadSYan Zheng root_objectid++; 3276c146afadSYan Zheng } 3277c146afadSYan Zheng return 0; 3278c146afadSYan Zheng } 3279c146afadSYan Zheng 3280c146afadSYan Zheng int btrfs_commit_super(struct btrfs_root *root) 3281c146afadSYan Zheng { 3282c146afadSYan Zheng struct btrfs_trans_handle *trans; 3283c146afadSYan Zheng int ret; 3284c146afadSYan Zheng 3285c146afadSYan Zheng mutex_lock(&root->fs_info->cleaner_mutex); 328624bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 3287c146afadSYan Zheng btrfs_clean_old_snapshots(root); 3288c146afadSYan Zheng mutex_unlock(&root->fs_info->cleaner_mutex); 3289c71bf099SYan, Zheng 3290c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 3291c71bf099SYan, Zheng down_write(&root->fs_info->cleanup_work_sem); 3292c71bf099SYan, Zheng up_write(&root->fs_info->cleanup_work_sem); 3293c71bf099SYan, Zheng 32947a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 32953612b495STsutomu Itoh if (IS_ERR(trans)) 32963612b495STsutomu Itoh return PTR_ERR(trans); 3297c146afadSYan Zheng ret = btrfs_commit_transaction(trans, root); 329879787eaaSJeff Mahoney if (ret) 329979787eaaSJeff Mahoney return ret; 3300c146afadSYan Zheng /* run commit again to drop the original snapshot */ 33017a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 33023612b495STsutomu Itoh if (IS_ERR(trans)) 33033612b495STsutomu Itoh return PTR_ERR(trans); 330479787eaaSJeff Mahoney ret = btrfs_commit_transaction(trans, root); 330579787eaaSJeff Mahoney if (ret) 330679787eaaSJeff Mahoney return ret; 3307c146afadSYan Zheng ret = btrfs_write_and_wait_transaction(NULL, root); 330879787eaaSJeff Mahoney if (ret) { 330979787eaaSJeff Mahoney btrfs_error(root->fs_info, ret, 331079787eaaSJeff Mahoney "Failed to sync btree inode to disk."); 331179787eaaSJeff Mahoney return ret; 331279787eaaSJeff Mahoney } 3313c146afadSYan Zheng 3314a512bbf8SYan Zheng ret = write_ctree_super(NULL, root, 0); 3315c146afadSYan Zheng return ret; 3316c146afadSYan Zheng } 3317c146afadSYan Zheng 3318e20d96d6SChris Mason int close_ctree(struct btrfs_root *root) 3319eb60ceacSChris Mason { 33200f7d52f4SChris Mason struct btrfs_fs_info *fs_info = root->fs_info; 3321c146afadSYan Zheng int ret; 3322e089f05cSChris Mason 3323facda1e7SChris Mason fs_info->closing = 1; 3324a2135011SChris Mason smp_mb(); 3325a2135011SChris Mason 3326837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 3327aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 3328837d5b6eSIlya Dryomov 33298dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 33308dabb742SStefan Behrens 3331aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 33324cb5300bSChris Mason 33334cb5300bSChris Mason /* wait for any defraggers to finish */ 33344cb5300bSChris Mason wait_event(fs_info->transaction_wait, 33354cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 33364cb5300bSChris Mason 33374cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 333826176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 33394cb5300bSChris Mason 3340c146afadSYan Zheng if (!(fs_info->sb->s_flags & MS_RDONLY)) { 3341c146afadSYan Zheng ret = btrfs_commit_super(root); 3342d397712bSChris Mason if (ret) 3343d397712bSChris Mason printk(KERN_ERR "btrfs: commit super ret %d\n", ret); 3344c146afadSYan Zheng } 3345ed2ff2cbSChris Mason 334668ce9682SStefan Behrens if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) 334768ce9682SStefan Behrens btrfs_error_commit_super(root); 3348acce952bSliubo 3349300e4f8aSJosef Bacik btrfs_put_block_group_cache(fs_info); 3350300e4f8aSJosef Bacik 3351e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 3352e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 33538929ecfaSYan, Zheng 3354f25784b3SYan Zheng fs_info->closing = 2; 3355f25784b3SYan Zheng smp_mb(); 3356f25784b3SYan Zheng 3357bcef60f2SArne Jansen btrfs_free_qgroup_config(root->fs_info); 3358bcef60f2SArne Jansen 3359b0c68f8bSChris Mason if (fs_info->delalloc_bytes) { 3360d397712bSChris Mason printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n", 336121380931SJoel Becker (unsigned long long)fs_info->delalloc_bytes); 3362b0c68f8bSChris Mason } 336331153d81SYan Zheng 33645f39d397SChris Mason free_extent_buffer(fs_info->extent_root->node); 33655d4f98a2SYan Zheng free_extent_buffer(fs_info->extent_root->commit_root); 33665f39d397SChris Mason free_extent_buffer(fs_info->tree_root->node); 33675d4f98a2SYan Zheng free_extent_buffer(fs_info->tree_root->commit_root); 3368e3029d9fSAl Viro free_extent_buffer(fs_info->chunk_root->node); 3369e3029d9fSAl Viro free_extent_buffer(fs_info->chunk_root->commit_root); 3370e3029d9fSAl Viro free_extent_buffer(fs_info->dev_root->node); 3371e3029d9fSAl Viro free_extent_buffer(fs_info->dev_root->commit_root); 3372e3029d9fSAl Viro free_extent_buffer(fs_info->csum_root->node); 3373e3029d9fSAl Viro free_extent_buffer(fs_info->csum_root->commit_root); 3374bcef60f2SArne Jansen if (fs_info->quota_root) { 3375bcef60f2SArne Jansen free_extent_buffer(fs_info->quota_root->node); 3376bcef60f2SArne Jansen free_extent_buffer(fs_info->quota_root->commit_root); 3377bcef60f2SArne Jansen } 3378d20f7043SChris Mason 3379e3029d9fSAl Viro btrfs_free_block_groups(fs_info); 3380c146afadSYan Zheng 33810f7d52f4SChris Mason del_fs_roots(fs_info); 3382d10c5f31SChris Mason 3383c146afadSYan Zheng iput(fs_info->btree_inode); 33849ad6b7bcSChris Mason 338561d92c32SChris Mason btrfs_stop_workers(&fs_info->generic_worker); 3386247e743cSChris Mason btrfs_stop_workers(&fs_info->fixup_workers); 3387771ed689SChris Mason btrfs_stop_workers(&fs_info->delalloc_workers); 33888b712842SChris Mason btrfs_stop_workers(&fs_info->workers); 33898b712842SChris Mason btrfs_stop_workers(&fs_info->endio_workers); 3390d20f7043SChris Mason btrfs_stop_workers(&fs_info->endio_meta_workers); 3391cad321adSChris Mason btrfs_stop_workers(&fs_info->endio_meta_write_workers); 3392e6dcd2dcSChris Mason btrfs_stop_workers(&fs_info->endio_write_workers); 33930cb59c99SJosef Bacik btrfs_stop_workers(&fs_info->endio_freespace_worker); 33941cc127b5SChris Mason btrfs_stop_workers(&fs_info->submit_workers); 339516cdcec7SMiao Xie btrfs_stop_workers(&fs_info->delayed_workers); 3396bab39bf9SJosef Bacik btrfs_stop_workers(&fs_info->caching_workers); 339790519d66SArne Jansen btrfs_stop_workers(&fs_info->readahead_workers); 33988ccf6f19SMiao Xie btrfs_stop_workers(&fs_info->flush_workers); 3399d6bfde87SChris Mason 340021adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 340121adbd5cSStefan Behrens if (btrfs_test_opt(root, CHECK_INTEGRITY)) 340221adbd5cSStefan Behrens btrfsic_unmount(root, fs_info->fs_devices); 340321adbd5cSStefan Behrens #endif 340421adbd5cSStefan Behrens 3405dfe25020SChris Mason btrfs_close_devices(fs_info->fs_devices); 34060b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 3407b248a415SChris Mason 340804160088SChris Mason bdi_destroy(&fs_info->bdi); 340976dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 34100b86a832SChris Mason 3411eb60ceacSChris Mason return 0; 3412eb60ceacSChris Mason } 3413eb60ceacSChris Mason 3414b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 3415b9fab919SChris Mason int atomic) 3416ccd467d6SChris Mason { 34171259ab75SChris Mason int ret; 3418727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 34191259ab75SChris Mason 34200b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 34211259ab75SChris Mason if (!ret) 34221259ab75SChris Mason return ret; 34231259ab75SChris Mason 34241259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 3425b9fab919SChris Mason parent_transid, atomic); 3426b9fab919SChris Mason if (ret == -EAGAIN) 3427b9fab919SChris Mason return ret; 34281259ab75SChris Mason return !ret; 34295f39d397SChris Mason } 34306702ed49SChris Mason 34315f39d397SChris Mason int btrfs_set_buffer_uptodate(struct extent_buffer *buf) 34325f39d397SChris Mason { 34330b32f4bbSJosef Bacik return set_extent_buffer_uptodate(buf); 34345f39d397SChris Mason } 34355f39d397SChris Mason 34365f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 34375f39d397SChris Mason { 3438727011e0SChris Mason struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root; 34395f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 3440b9473439SChris Mason int was_dirty; 3441b4ce94deSChris Mason 3442b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 344331b1a2bdSJulia Lawall if (transid != root->fs_info->generation) 344431b1a2bdSJulia Lawall WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, " 3445d397712bSChris Mason "found %llu running %llu\n", 3446db94535dSChris Mason (unsigned long long)buf->start, 3447d397712bSChris Mason (unsigned long long)transid, 3448d397712bSChris Mason (unsigned long long)root->fs_info->generation); 34490b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 3450b9473439SChris Mason if (!was_dirty) { 3451b9473439SChris Mason spin_lock(&root->fs_info->delalloc_lock); 3452b9473439SChris Mason root->fs_info->dirty_metadata_bytes += buf->len; 3453b9473439SChris Mason spin_unlock(&root->fs_info->delalloc_lock); 3454b9473439SChris Mason } 3455eb60ceacSChris Mason } 3456eb60ceacSChris Mason 3457b53d3f5dSLiu Bo static void __btrfs_btree_balance_dirty(struct btrfs_root *root, 3458b53d3f5dSLiu Bo int flush_delayed) 345935b7e476SChris Mason { 3460188de649SChris Mason /* 3461188de649SChris Mason * looks as though older kernels can get into trouble with 3462188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 3463188de649SChris Mason */ 3464d6bfde87SChris Mason u64 num_dirty; 3465771ed689SChris Mason unsigned long thresh = 32 * 1024 * 1024; 3466d6bfde87SChris Mason 34676933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 3468d6bfde87SChris Mason return; 3469d6bfde87SChris Mason 3470b53d3f5dSLiu Bo if (flush_delayed) 347116cdcec7SMiao Xie btrfs_balance_delayed_items(root); 347216cdcec7SMiao Xie 347316cdcec7SMiao Xie num_dirty = root->fs_info->dirty_metadata_bytes; 347416cdcec7SMiao Xie 347516cdcec7SMiao Xie if (num_dirty > thresh) { 347616cdcec7SMiao Xie balance_dirty_pages_ratelimited_nr( 347716cdcec7SMiao Xie root->fs_info->btree_inode->i_mapping, 1); 347816cdcec7SMiao Xie } 347916cdcec7SMiao Xie return; 348016cdcec7SMiao Xie } 348116cdcec7SMiao Xie 3482b53d3f5dSLiu Bo void btrfs_btree_balance_dirty(struct btrfs_root *root) 348316cdcec7SMiao Xie { 3484b53d3f5dSLiu Bo __btrfs_btree_balance_dirty(root, 1); 348535b7e476SChris Mason } 3486b53d3f5dSLiu Bo 3487b53d3f5dSLiu Bo void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root) 3488b53d3f5dSLiu Bo { 3489b53d3f5dSLiu Bo __btrfs_btree_balance_dirty(root, 0); 3490d6bfde87SChris Mason } 34916b80053dSChris Mason 3492ca7a79adSChris Mason int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid) 34936b80053dSChris Mason { 3494727011e0SChris Mason struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root; 34950b32f4bbSJosef Bacik return btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 34966b80053dSChris Mason } 34970da5468fSChris Mason 3498fcd1f065SDavid Sterba static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info, 3499acce952bSliubo int read_only) 3500acce952bSliubo { 3501fcd1f065SDavid Sterba if (btrfs_super_csum_type(fs_info->super_copy) >= ARRAY_SIZE(btrfs_csum_sizes)) { 3502fcd1f065SDavid Sterba printk(KERN_ERR "btrfs: unsupported checksum algorithm\n"); 3503fcd1f065SDavid Sterba return -EINVAL; 3504fcd1f065SDavid Sterba } 3505acce952bSliubo 3506fcd1f065SDavid Sterba if (read_only) 3507fcd1f065SDavid Sterba return 0; 3508fcd1f065SDavid Sterba 3509fcd1f065SDavid Sterba return 0; 3510fcd1f065SDavid Sterba } 3511fcd1f065SDavid Sterba 351268ce9682SStefan Behrens void btrfs_error_commit_super(struct btrfs_root *root) 3513acce952bSliubo { 3514acce952bSliubo mutex_lock(&root->fs_info->cleaner_mutex); 3515acce952bSliubo btrfs_run_delayed_iputs(root); 3516acce952bSliubo mutex_unlock(&root->fs_info->cleaner_mutex); 3517acce952bSliubo 3518acce952bSliubo down_write(&root->fs_info->cleanup_work_sem); 3519acce952bSliubo up_write(&root->fs_info->cleanup_work_sem); 3520acce952bSliubo 3521acce952bSliubo /* cleanup FS via transaction */ 3522acce952bSliubo btrfs_cleanup_transaction(root); 3523acce952bSliubo } 3524acce952bSliubo 3525143bede5SJeff Mahoney static void btrfs_destroy_ordered_operations(struct btrfs_root *root) 3526acce952bSliubo { 3527acce952bSliubo struct btrfs_inode *btrfs_inode; 3528acce952bSliubo struct list_head splice; 3529acce952bSliubo 3530acce952bSliubo INIT_LIST_HEAD(&splice); 3531acce952bSliubo 3532acce952bSliubo mutex_lock(&root->fs_info->ordered_operations_mutex); 3533acce952bSliubo spin_lock(&root->fs_info->ordered_extent_lock); 3534acce952bSliubo 3535acce952bSliubo list_splice_init(&root->fs_info->ordered_operations, &splice); 3536acce952bSliubo while (!list_empty(&splice)) { 3537acce952bSliubo btrfs_inode = list_entry(splice.next, struct btrfs_inode, 3538acce952bSliubo ordered_operations); 3539acce952bSliubo 3540acce952bSliubo list_del_init(&btrfs_inode->ordered_operations); 3541acce952bSliubo 3542acce952bSliubo btrfs_invalidate_inodes(btrfs_inode->root); 3543acce952bSliubo } 3544acce952bSliubo 3545acce952bSliubo spin_unlock(&root->fs_info->ordered_extent_lock); 3546acce952bSliubo mutex_unlock(&root->fs_info->ordered_operations_mutex); 3547acce952bSliubo } 3548acce952bSliubo 3549143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 3550acce952bSliubo { 3551acce952bSliubo struct list_head splice; 3552acce952bSliubo struct btrfs_ordered_extent *ordered; 3553acce952bSliubo struct inode *inode; 3554acce952bSliubo 3555acce952bSliubo INIT_LIST_HEAD(&splice); 3556acce952bSliubo 3557acce952bSliubo spin_lock(&root->fs_info->ordered_extent_lock); 3558acce952bSliubo 3559acce952bSliubo list_splice_init(&root->fs_info->ordered_extents, &splice); 3560acce952bSliubo while (!list_empty(&splice)) { 3561acce952bSliubo ordered = list_entry(splice.next, struct btrfs_ordered_extent, 3562acce952bSliubo root_extent_list); 3563acce952bSliubo 3564acce952bSliubo list_del_init(&ordered->root_extent_list); 3565acce952bSliubo atomic_inc(&ordered->refs); 3566acce952bSliubo 3567acce952bSliubo /* the inode may be getting freed (in sys_unlink path). */ 3568acce952bSliubo inode = igrab(ordered->inode); 3569acce952bSliubo 3570acce952bSliubo spin_unlock(&root->fs_info->ordered_extent_lock); 3571acce952bSliubo if (inode) 3572acce952bSliubo iput(inode); 3573acce952bSliubo 3574acce952bSliubo atomic_set(&ordered->refs, 1); 3575acce952bSliubo btrfs_put_ordered_extent(ordered); 3576acce952bSliubo 3577acce952bSliubo spin_lock(&root->fs_info->ordered_extent_lock); 3578acce952bSliubo } 3579acce952bSliubo 3580acce952bSliubo spin_unlock(&root->fs_info->ordered_extent_lock); 3581acce952bSliubo } 3582acce952bSliubo 358379787eaaSJeff Mahoney int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 3584acce952bSliubo struct btrfs_root *root) 3585acce952bSliubo { 3586acce952bSliubo struct rb_node *node; 3587acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 3588acce952bSliubo struct btrfs_delayed_ref_node *ref; 3589acce952bSliubo int ret = 0; 3590acce952bSliubo 3591acce952bSliubo delayed_refs = &trans->delayed_refs; 3592acce952bSliubo 3593acce952bSliubo spin_lock(&delayed_refs->lock); 3594acce952bSliubo if (delayed_refs->num_entries == 0) { 3595cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 3596acce952bSliubo printk(KERN_INFO "delayed_refs has NO entry\n"); 3597acce952bSliubo return ret; 3598acce952bSliubo } 3599acce952bSliubo 3600b939d1abSJosef Bacik while ((node = rb_first(&delayed_refs->root)) != NULL) { 3601acce952bSliubo ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); 3602acce952bSliubo 3603acce952bSliubo atomic_set(&ref->refs, 1); 3604acce952bSliubo if (btrfs_delayed_ref_is_head(ref)) { 3605acce952bSliubo struct btrfs_delayed_ref_head *head; 3606acce952bSliubo 3607acce952bSliubo head = btrfs_delayed_node_to_head(ref); 3608b939d1abSJosef Bacik if (!mutex_trylock(&head->mutex)) { 3609b939d1abSJosef Bacik atomic_inc(&ref->refs); 361079787eaaSJeff Mahoney spin_unlock(&delayed_refs->lock); 3611b939d1abSJosef Bacik 3612b939d1abSJosef Bacik /* Need to wait for the delayed ref to run */ 3613acce952bSliubo mutex_lock(&head->mutex); 3614b939d1abSJosef Bacik mutex_unlock(&head->mutex); 3615b939d1abSJosef Bacik btrfs_put_delayed_ref(ref); 3616b939d1abSJosef Bacik 3617e18fca73SJosef Bacik spin_lock(&delayed_refs->lock); 3618b939d1abSJosef Bacik continue; 3619b939d1abSJosef Bacik } 3620b939d1abSJosef Bacik 362178a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 3622acce952bSliubo delayed_refs->num_heads--; 3623acce952bSliubo if (list_empty(&head->cluster)) 3624acce952bSliubo delayed_refs->num_heads_ready--; 3625acce952bSliubo list_del_init(&head->cluster); 3626acce952bSliubo } 3627b939d1abSJosef Bacik ref->in_tree = 0; 3628b939d1abSJosef Bacik rb_erase(&ref->rb_node, &delayed_refs->root); 3629b939d1abSJosef Bacik delayed_refs->num_entries--; 3630b939d1abSJosef Bacik 3631acce952bSliubo spin_unlock(&delayed_refs->lock); 3632acce952bSliubo btrfs_put_delayed_ref(ref); 3633acce952bSliubo 3634acce952bSliubo cond_resched(); 3635acce952bSliubo spin_lock(&delayed_refs->lock); 3636acce952bSliubo } 3637acce952bSliubo 3638acce952bSliubo spin_unlock(&delayed_refs->lock); 3639acce952bSliubo 3640acce952bSliubo return ret; 3641acce952bSliubo } 3642acce952bSliubo 3643143bede5SJeff Mahoney static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t) 3644acce952bSliubo { 3645acce952bSliubo struct btrfs_pending_snapshot *snapshot; 3646acce952bSliubo struct list_head splice; 3647acce952bSliubo 3648acce952bSliubo INIT_LIST_HEAD(&splice); 3649acce952bSliubo 3650acce952bSliubo list_splice_init(&t->pending_snapshots, &splice); 3651acce952bSliubo 3652acce952bSliubo while (!list_empty(&splice)) { 3653acce952bSliubo snapshot = list_entry(splice.next, 3654acce952bSliubo struct btrfs_pending_snapshot, 3655acce952bSliubo list); 3656acce952bSliubo 3657acce952bSliubo list_del_init(&snapshot->list); 3658acce952bSliubo 3659acce952bSliubo kfree(snapshot); 3660acce952bSliubo } 3661acce952bSliubo } 3662acce952bSliubo 3663143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 3664acce952bSliubo { 3665acce952bSliubo struct btrfs_inode *btrfs_inode; 3666acce952bSliubo struct list_head splice; 3667acce952bSliubo 3668acce952bSliubo INIT_LIST_HEAD(&splice); 3669acce952bSliubo 3670acce952bSliubo spin_lock(&root->fs_info->delalloc_lock); 36715be76758SDavid Sterba list_splice_init(&root->fs_info->delalloc_inodes, &splice); 3672acce952bSliubo 3673acce952bSliubo while (!list_empty(&splice)) { 3674acce952bSliubo btrfs_inode = list_entry(splice.next, struct btrfs_inode, 3675acce952bSliubo delalloc_inodes); 3676acce952bSliubo 3677acce952bSliubo list_del_init(&btrfs_inode->delalloc_inodes); 3678acce952bSliubo 3679acce952bSliubo btrfs_invalidate_inodes(btrfs_inode->root); 3680acce952bSliubo } 3681acce952bSliubo 3682acce952bSliubo spin_unlock(&root->fs_info->delalloc_lock); 3683acce952bSliubo } 3684acce952bSliubo 3685acce952bSliubo static int btrfs_destroy_marked_extents(struct btrfs_root *root, 3686acce952bSliubo struct extent_io_tree *dirty_pages, 3687acce952bSliubo int mark) 3688acce952bSliubo { 3689acce952bSliubo int ret; 3690acce952bSliubo struct page *page; 3691acce952bSliubo struct inode *btree_inode = root->fs_info->btree_inode; 3692acce952bSliubo struct extent_buffer *eb; 3693acce952bSliubo u64 start = 0; 3694acce952bSliubo u64 end; 3695acce952bSliubo u64 offset; 3696acce952bSliubo unsigned long index; 3697acce952bSliubo 3698acce952bSliubo while (1) { 3699acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 3700e6138876SJosef Bacik mark, NULL); 3701acce952bSliubo if (ret) 3702acce952bSliubo break; 3703acce952bSliubo 3704acce952bSliubo clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); 3705acce952bSliubo while (start <= end) { 3706acce952bSliubo index = start >> PAGE_CACHE_SHIFT; 3707acce952bSliubo start = (u64)(index + 1) << PAGE_CACHE_SHIFT; 3708acce952bSliubo page = find_get_page(btree_inode->i_mapping, index); 3709acce952bSliubo if (!page) 3710acce952bSliubo continue; 3711acce952bSliubo offset = page_offset(page); 3712acce952bSliubo 3713acce952bSliubo spin_lock(&dirty_pages->buffer_lock); 3714acce952bSliubo eb = radix_tree_lookup( 3715acce952bSliubo &(&BTRFS_I(page->mapping->host)->io_tree)->buffer, 3716acce952bSliubo offset >> PAGE_CACHE_SHIFT); 3717acce952bSliubo spin_unlock(&dirty_pages->buffer_lock); 3718ee670f0aSJosef Bacik if (eb) 3719acce952bSliubo ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY, 3720acce952bSliubo &eb->bflags); 3721acce952bSliubo if (PageWriteback(page)) 3722acce952bSliubo end_page_writeback(page); 3723acce952bSliubo 3724acce952bSliubo lock_page(page); 3725acce952bSliubo if (PageDirty(page)) { 3726acce952bSliubo clear_page_dirty_for_io(page); 3727acce952bSliubo spin_lock_irq(&page->mapping->tree_lock); 3728acce952bSliubo radix_tree_tag_clear(&page->mapping->page_tree, 3729acce952bSliubo page_index(page), 3730acce952bSliubo PAGECACHE_TAG_DIRTY); 3731acce952bSliubo spin_unlock_irq(&page->mapping->tree_lock); 3732acce952bSliubo } 3733acce952bSliubo 3734acce952bSliubo unlock_page(page); 3735ee670f0aSJosef Bacik page_cache_release(page); 3736acce952bSliubo } 3737acce952bSliubo } 3738acce952bSliubo 3739acce952bSliubo return ret; 3740acce952bSliubo } 3741acce952bSliubo 3742acce952bSliubo static int btrfs_destroy_pinned_extent(struct btrfs_root *root, 3743acce952bSliubo struct extent_io_tree *pinned_extents) 3744acce952bSliubo { 3745acce952bSliubo struct extent_io_tree *unpin; 3746acce952bSliubo u64 start; 3747acce952bSliubo u64 end; 3748acce952bSliubo int ret; 3749ed0eaa14SLiu Bo bool loop = true; 3750acce952bSliubo 3751acce952bSliubo unpin = pinned_extents; 3752ed0eaa14SLiu Bo again: 3753acce952bSliubo while (1) { 3754acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 3755e6138876SJosef Bacik EXTENT_DIRTY, NULL); 3756acce952bSliubo if (ret) 3757acce952bSliubo break; 3758acce952bSliubo 3759acce952bSliubo /* opt_discard */ 37605378e607SLi Dongyang if (btrfs_test_opt(root, DISCARD)) 37615378e607SLi Dongyang ret = btrfs_error_discard_extent(root, start, 37625378e607SLi Dongyang end + 1 - start, 37635378e607SLi Dongyang NULL); 3764acce952bSliubo 3765acce952bSliubo clear_extent_dirty(unpin, start, end, GFP_NOFS); 3766acce952bSliubo btrfs_error_unpin_extent_range(root, start, end); 3767acce952bSliubo cond_resched(); 3768acce952bSliubo } 3769acce952bSliubo 3770ed0eaa14SLiu Bo if (loop) { 3771ed0eaa14SLiu Bo if (unpin == &root->fs_info->freed_extents[0]) 3772ed0eaa14SLiu Bo unpin = &root->fs_info->freed_extents[1]; 3773ed0eaa14SLiu Bo else 3774ed0eaa14SLiu Bo unpin = &root->fs_info->freed_extents[0]; 3775ed0eaa14SLiu Bo loop = false; 3776ed0eaa14SLiu Bo goto again; 3777ed0eaa14SLiu Bo } 3778ed0eaa14SLiu Bo 3779acce952bSliubo return 0; 3780acce952bSliubo } 3781acce952bSliubo 378249b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 378349b25e05SJeff Mahoney struct btrfs_root *root) 378449b25e05SJeff Mahoney { 378549b25e05SJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, root); 378649b25e05SJeff Mahoney btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv, 378749b25e05SJeff Mahoney cur_trans->dirty_pages.dirty_bytes); 378849b25e05SJeff Mahoney 378949b25e05SJeff Mahoney /* FIXME: cleanup wait for commit */ 379049b25e05SJeff Mahoney cur_trans->in_commit = 1; 379149b25e05SJeff Mahoney cur_trans->blocked = 1; 379249b25e05SJeff Mahoney wake_up(&root->fs_info->transaction_blocked_wait); 379349b25e05SJeff Mahoney 379449b25e05SJeff Mahoney cur_trans->blocked = 0; 379549b25e05SJeff Mahoney wake_up(&root->fs_info->transaction_wait); 379649b25e05SJeff Mahoney 379749b25e05SJeff Mahoney cur_trans->commit_done = 1; 379849b25e05SJeff Mahoney wake_up(&cur_trans->commit_wait); 379949b25e05SJeff Mahoney 380067cde344SMiao Xie btrfs_destroy_delayed_inodes(root); 380167cde344SMiao Xie btrfs_assert_delayed_root_empty(root); 380267cde344SMiao Xie 380349b25e05SJeff Mahoney btrfs_destroy_pending_snapshots(cur_trans); 380449b25e05SJeff Mahoney 380549b25e05SJeff Mahoney btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages, 380649b25e05SJeff Mahoney EXTENT_DIRTY); 38076e841e32SLiu Bo btrfs_destroy_pinned_extent(root, 38086e841e32SLiu Bo root->fs_info->pinned_extents); 380949b25e05SJeff Mahoney 381049b25e05SJeff Mahoney /* 381149b25e05SJeff Mahoney memset(cur_trans, 0, sizeof(*cur_trans)); 381249b25e05SJeff Mahoney kmem_cache_free(btrfs_transaction_cachep, cur_trans); 381349b25e05SJeff Mahoney */ 381449b25e05SJeff Mahoney } 381549b25e05SJeff Mahoney 381649b25e05SJeff Mahoney int btrfs_cleanup_transaction(struct btrfs_root *root) 3817acce952bSliubo { 3818acce952bSliubo struct btrfs_transaction *t; 3819acce952bSliubo LIST_HEAD(list); 3820acce952bSliubo 3821acce952bSliubo mutex_lock(&root->fs_info->transaction_kthread_mutex); 3822acce952bSliubo 3823a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 3824acce952bSliubo list_splice_init(&root->fs_info->trans_list, &list); 3825a4abeea4SJosef Bacik root->fs_info->trans_no_join = 1; 3826a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 3827a4abeea4SJosef Bacik 3828acce952bSliubo while (!list_empty(&list)) { 3829acce952bSliubo t = list_entry(list.next, struct btrfs_transaction, list); 3830acce952bSliubo if (!t) 3831acce952bSliubo break; 3832acce952bSliubo 3833acce952bSliubo btrfs_destroy_ordered_operations(root); 3834acce952bSliubo 3835acce952bSliubo btrfs_destroy_ordered_extents(root); 3836acce952bSliubo 3837acce952bSliubo btrfs_destroy_delayed_refs(t, root); 3838acce952bSliubo 3839acce952bSliubo btrfs_block_rsv_release(root, 3840acce952bSliubo &root->fs_info->trans_block_rsv, 3841acce952bSliubo t->dirty_pages.dirty_bytes); 3842acce952bSliubo 3843acce952bSliubo /* FIXME: cleanup wait for commit */ 3844acce952bSliubo t->in_commit = 1; 3845acce952bSliubo t->blocked = 1; 384666657b31SJosef Bacik smp_mb(); 3847acce952bSliubo if (waitqueue_active(&root->fs_info->transaction_blocked_wait)) 3848acce952bSliubo wake_up(&root->fs_info->transaction_blocked_wait); 3849acce952bSliubo 3850acce952bSliubo t->blocked = 0; 385166657b31SJosef Bacik smp_mb(); 3852acce952bSliubo if (waitqueue_active(&root->fs_info->transaction_wait)) 3853acce952bSliubo wake_up(&root->fs_info->transaction_wait); 3854acce952bSliubo 3855acce952bSliubo t->commit_done = 1; 385666657b31SJosef Bacik smp_mb(); 3857acce952bSliubo if (waitqueue_active(&t->commit_wait)) 3858acce952bSliubo wake_up(&t->commit_wait); 3859acce952bSliubo 386067cde344SMiao Xie btrfs_destroy_delayed_inodes(root); 386167cde344SMiao Xie btrfs_assert_delayed_root_empty(root); 386267cde344SMiao Xie 3863acce952bSliubo btrfs_destroy_pending_snapshots(t); 3864acce952bSliubo 3865acce952bSliubo btrfs_destroy_delalloc_inodes(root); 3866acce952bSliubo 3867a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 3868acce952bSliubo root->fs_info->running_transaction = NULL; 3869a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 3870acce952bSliubo 3871acce952bSliubo btrfs_destroy_marked_extents(root, &t->dirty_pages, 3872acce952bSliubo EXTENT_DIRTY); 3873acce952bSliubo 3874acce952bSliubo btrfs_destroy_pinned_extent(root, 3875acce952bSliubo root->fs_info->pinned_extents); 3876acce952bSliubo 387713c5a93eSJosef Bacik atomic_set(&t->use_count, 0); 3878acce952bSliubo list_del_init(&t->list); 3879acce952bSliubo memset(t, 0, sizeof(*t)); 3880acce952bSliubo kmem_cache_free(btrfs_transaction_cachep, t); 3881acce952bSliubo } 3882acce952bSliubo 3883a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 3884a4abeea4SJosef Bacik root->fs_info->trans_no_join = 0; 3885a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 3886acce952bSliubo mutex_unlock(&root->fs_info->transaction_kthread_mutex); 3887acce952bSliubo 3888acce952bSliubo return 0; 3889acce952bSliubo } 3890acce952bSliubo 3891d1310b2eSChris Mason static struct extent_io_ops btree_extent_io_ops = { 3892ce9adaa5SChris Mason .readpage_end_io_hook = btree_readpage_end_io_hook, 38934bb31e92SArne Jansen .readpage_io_failed_hook = btree_io_failed_hook, 38940b86a832SChris Mason .submit_bio_hook = btree_submit_bio_hook, 3895239b14b3SChris Mason /* note we're sharing with inode.c for the merge bio hook */ 3896239b14b3SChris Mason .merge_bio_hook = btrfs_merge_bio_hook, 38970da5468fSChris Mason }; 3898