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> 295a0e3ad6STejun Heo #include <linux/slab.h> 30784b4e29SChris Mason #include <linux/migrate.h> 317a36ddecSDavid Sterba #include <linux/ratelimit.h> 326463fe58SStefan Behrens #include <linux/uuid.h> 33803b2f54SStefan Behrens #include <linux/semaphore.h> 347e75bf3fSDavid Sterba #include <asm/unaligned.h> 35eb60ceacSChris Mason #include "ctree.h" 36eb60ceacSChris Mason #include "disk-io.h" 370b947affSFilipe David Borba Manana #include "hash.h" 38e089f05cSChris Mason #include "transaction.h" 390f7d52f4SChris Mason #include "btrfs_inode.h" 400b86a832SChris Mason #include "volumes.h" 41db94535dSChris Mason #include "print-tree.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" 4953b381b3SDavid Woodhouse #include "raid56.h" 505ac1d209SJeff Mahoney #include "sysfs.h" 51fcebe456SJosef Bacik #include "qgroup.h" 52eb60ceacSChris Mason 53de0022b9SJosef Bacik #ifdef CONFIG_X86 54de0022b9SJosef Bacik #include <asm/cpufeature.h> 55de0022b9SJosef Bacik #endif 56de0022b9SJosef Bacik 57e8c9f186SDavid Sterba static const struct extent_io_ops btree_extent_io_ops; 588b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work); 594df27c4dSYan, Zheng static void free_fs_root(struct btrfs_root *root); 60fcd1f065SDavid Sterba static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info, 61acce952bSliubo int read_only); 62143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root); 63acce952bSliubo static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 64acce952bSliubo struct btrfs_root *root); 65143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root); 66acce952bSliubo static int btrfs_destroy_marked_extents(struct btrfs_root *root, 67acce952bSliubo struct extent_io_tree *dirty_pages, 68acce952bSliubo int mark); 69acce952bSliubo static int btrfs_destroy_pinned_extent(struct btrfs_root *root, 70acce952bSliubo struct extent_io_tree *pinned_extents); 7148a3b636SEric Sandeen static int btrfs_cleanup_transaction(struct btrfs_root *root); 7248a3b636SEric Sandeen static void btrfs_error_commit_super(struct btrfs_root *root); 73ce9adaa5SChris Mason 74d352ac68SChris Mason /* 7597eb6b69SDavid Sterba * btrfs_end_io_wq structs are used to do processing in task context when an IO 7697eb6b69SDavid Sterba * is complete. This is used during reads to verify checksums, and it is used 77d352ac68SChris Mason * by writes to insert metadata for new file extents after IO is complete. 78d352ac68SChris Mason */ 7997eb6b69SDavid Sterba struct btrfs_end_io_wq { 80ce9adaa5SChris Mason struct bio *bio; 81ce9adaa5SChris Mason bio_end_io_t *end_io; 82ce9adaa5SChris Mason void *private; 83ce9adaa5SChris Mason struct btrfs_fs_info *info; 84ce9adaa5SChris Mason int error; 85bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata; 86ce9adaa5SChris Mason struct list_head list; 878b712842SChris Mason struct btrfs_work work; 88ce9adaa5SChris Mason }; 890da5468fSChris Mason 9097eb6b69SDavid Sterba static struct kmem_cache *btrfs_end_io_wq_cache; 9197eb6b69SDavid Sterba 9297eb6b69SDavid Sterba int __init btrfs_end_io_wq_init(void) 9397eb6b69SDavid Sterba { 9497eb6b69SDavid Sterba btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq", 9597eb6b69SDavid Sterba sizeof(struct btrfs_end_io_wq), 9697eb6b69SDavid Sterba 0, 9797eb6b69SDavid Sterba SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, 9897eb6b69SDavid Sterba NULL); 9997eb6b69SDavid Sterba if (!btrfs_end_io_wq_cache) 10097eb6b69SDavid Sterba return -ENOMEM; 10197eb6b69SDavid Sterba return 0; 10297eb6b69SDavid Sterba } 10397eb6b69SDavid Sterba 10497eb6b69SDavid Sterba void btrfs_end_io_wq_exit(void) 10597eb6b69SDavid Sterba { 10697eb6b69SDavid Sterba if (btrfs_end_io_wq_cache) 10797eb6b69SDavid Sterba kmem_cache_destroy(btrfs_end_io_wq_cache); 10897eb6b69SDavid Sterba } 10997eb6b69SDavid Sterba 110d352ac68SChris Mason /* 111d352ac68SChris Mason * async submit bios are used to offload expensive checksumming 112d352ac68SChris Mason * onto the worker threads. They checksum file and metadata bios 113d352ac68SChris Mason * just before they are sent down the IO stack. 114d352ac68SChris Mason */ 11544b8bd7eSChris Mason struct async_submit_bio { 11644b8bd7eSChris Mason struct inode *inode; 11744b8bd7eSChris Mason struct bio *bio; 11844b8bd7eSChris Mason struct list_head list; 1194a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_start; 1204a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_done; 12144b8bd7eSChris Mason int rw; 12244b8bd7eSChris Mason int mirror_num; 123c8b97818SChris Mason unsigned long bio_flags; 124eaf25d93SChris Mason /* 125eaf25d93SChris Mason * bio_offset is optional, can be used if the pages in the bio 126eaf25d93SChris Mason * can't tell us where in the file the bio should go 127eaf25d93SChris Mason */ 128eaf25d93SChris Mason u64 bio_offset; 1298b712842SChris Mason struct btrfs_work work; 13079787eaaSJeff Mahoney int error; 13144b8bd7eSChris Mason }; 13244b8bd7eSChris Mason 13385d4e461SChris Mason /* 13485d4e461SChris Mason * Lockdep class keys for extent_buffer->lock's in this root. For a given 13585d4e461SChris Mason * eb, the lockdep key is determined by the btrfs_root it belongs to and 13685d4e461SChris Mason * the level the eb occupies in the tree. 1374008c04aSChris Mason * 13885d4e461SChris Mason * Different roots are used for different purposes and may nest inside each 13985d4e461SChris Mason * other and they require separate keysets. As lockdep keys should be 14085d4e461SChris Mason * static, assign keysets according to the purpose of the root as indicated 14185d4e461SChris Mason * by btrfs_root->objectid. This ensures that all special purpose roots 14285d4e461SChris Mason * have separate keysets. 1434008c04aSChris Mason * 14485d4e461SChris Mason * Lock-nesting across peer nodes is always done with the immediate parent 14585d4e461SChris Mason * node locked thus preventing deadlock. As lockdep doesn't know this, use 14685d4e461SChris Mason * subclass to avoid triggering lockdep warning in such cases. 1474008c04aSChris Mason * 14885d4e461SChris Mason * The key is set by the readpage_end_io_hook after the buffer has passed 14985d4e461SChris Mason * csum validation but before the pages are unlocked. It is also set by 15085d4e461SChris Mason * btrfs_init_new_buffer on freshly allocated blocks. 15185d4e461SChris Mason * 15285d4e461SChris Mason * We also add a check to make sure the highest level of the tree is the 15385d4e461SChris Mason * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code 15485d4e461SChris Mason * needs update as well. 1554008c04aSChris Mason */ 1564008c04aSChris Mason #ifdef CONFIG_DEBUG_LOCK_ALLOC 1574008c04aSChris Mason # if BTRFS_MAX_LEVEL != 8 1584008c04aSChris Mason # error 1594008c04aSChris Mason # endif 16085d4e461SChris Mason 16185d4e461SChris Mason static struct btrfs_lockdep_keyset { 16285d4e461SChris Mason u64 id; /* root objectid */ 16385d4e461SChris Mason const char *name_stem; /* lock name stem */ 16485d4e461SChris Mason char names[BTRFS_MAX_LEVEL + 1][20]; 16585d4e461SChris Mason struct lock_class_key keys[BTRFS_MAX_LEVEL + 1]; 16685d4e461SChris Mason } btrfs_lockdep_keysets[] = { 16785d4e461SChris Mason { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" }, 16885d4e461SChris Mason { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" }, 16985d4e461SChris Mason { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" }, 17085d4e461SChris Mason { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" }, 17185d4e461SChris Mason { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" }, 17285d4e461SChris Mason { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" }, 17360b62978SDavid Sterba { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" }, 17485d4e461SChris Mason { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" }, 17585d4e461SChris Mason { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" }, 17685d4e461SChris Mason { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" }, 17713fd8da9SDavid Sterba { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" }, 17885d4e461SChris Mason { .id = 0, .name_stem = "tree" }, 1794008c04aSChris Mason }; 18085d4e461SChris Mason 18185d4e461SChris Mason void __init btrfs_init_lockdep(void) 18285d4e461SChris Mason { 18385d4e461SChris Mason int i, j; 18485d4e461SChris Mason 18585d4e461SChris Mason /* initialize lockdep class names */ 18685d4e461SChris Mason for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) { 18785d4e461SChris Mason struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i]; 18885d4e461SChris Mason 18985d4e461SChris Mason for (j = 0; j < ARRAY_SIZE(ks->names); j++) 19085d4e461SChris Mason snprintf(ks->names[j], sizeof(ks->names[j]), 19185d4e461SChris Mason "btrfs-%s-%02d", ks->name_stem, j); 19285d4e461SChris Mason } 19385d4e461SChris Mason } 19485d4e461SChris Mason 19585d4e461SChris Mason void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb, 19685d4e461SChris Mason int level) 19785d4e461SChris Mason { 19885d4e461SChris Mason struct btrfs_lockdep_keyset *ks; 19985d4e461SChris Mason 20085d4e461SChris Mason BUG_ON(level >= ARRAY_SIZE(ks->keys)); 20185d4e461SChris Mason 20285d4e461SChris Mason /* find the matching keyset, id 0 is the default entry */ 20385d4e461SChris Mason for (ks = btrfs_lockdep_keysets; ks->id; ks++) 20485d4e461SChris Mason if (ks->id == objectid) 20585d4e461SChris Mason break; 20685d4e461SChris Mason 20785d4e461SChris Mason lockdep_set_class_and_name(&eb->lock, 20885d4e461SChris Mason &ks->keys[level], ks->names[level]); 20985d4e461SChris Mason } 21085d4e461SChris Mason 2114008c04aSChris Mason #endif 2124008c04aSChris Mason 213d352ac68SChris Mason /* 214d352ac68SChris Mason * extents on the btree inode are pretty simple, there's one extent 215d352ac68SChris Mason * that covers the entire device 216d352ac68SChris Mason */ 217b2950863SChristoph Hellwig static struct extent_map *btree_get_extent(struct inode *inode, 218306e16ceSDavid Sterba struct page *page, size_t pg_offset, u64 start, u64 len, 2195f39d397SChris Mason int create) 2205f39d397SChris Mason { 2215f39d397SChris Mason struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 2225f39d397SChris Mason struct extent_map *em; 2235f39d397SChris Mason int ret; 2245f39d397SChris Mason 225890871beSChris Mason read_lock(&em_tree->lock); 226d1310b2eSChris Mason em = lookup_extent_mapping(em_tree, start, len); 227a061fc8dSChris Mason if (em) { 228a061fc8dSChris Mason em->bdev = 229a061fc8dSChris Mason BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 230890871beSChris Mason read_unlock(&em_tree->lock); 2315f39d397SChris Mason goto out; 232a061fc8dSChris Mason } 233890871beSChris Mason read_unlock(&em_tree->lock); 2347b13b7b1SChris Mason 235172ddd60SDavid Sterba em = alloc_extent_map(); 2365f39d397SChris Mason if (!em) { 2375f39d397SChris Mason em = ERR_PTR(-ENOMEM); 2385f39d397SChris Mason goto out; 2395f39d397SChris Mason } 2405f39d397SChris Mason em->start = 0; 2410afbaf8cSChris Mason em->len = (u64)-1; 242c8b97818SChris Mason em->block_len = (u64)-1; 2435f39d397SChris Mason em->block_start = 0; 244a061fc8dSChris Mason em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 245d1310b2eSChris Mason 246890871beSChris Mason write_lock(&em_tree->lock); 24709a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 2485f39d397SChris Mason if (ret == -EEXIST) { 2495f39d397SChris Mason free_extent_map(em); 2507b13b7b1SChris Mason em = lookup_extent_mapping(em_tree, start, len); 251b4f359abSTsutomu Itoh if (!em) 2520433f20dSTsutomu Itoh em = ERR_PTR(-EIO); 2535f39d397SChris Mason } else if (ret) { 2547b13b7b1SChris Mason free_extent_map(em); 2550433f20dSTsutomu Itoh em = ERR_PTR(ret); 2565f39d397SChris Mason } 257890871beSChris Mason write_unlock(&em_tree->lock); 2587b13b7b1SChris Mason 2595f39d397SChris Mason out: 2605f39d397SChris Mason return em; 2615f39d397SChris Mason } 2625f39d397SChris Mason 263b0496686SLiu Bo u32 btrfs_csum_data(char *data, u32 seed, size_t len) 26419c00ddcSChris Mason { 2650b947affSFilipe David Borba Manana return btrfs_crc32c(seed, data, len); 26619c00ddcSChris Mason } 26719c00ddcSChris Mason 26819c00ddcSChris Mason void btrfs_csum_final(u32 crc, char *result) 26919c00ddcSChris Mason { 2707e75bf3fSDavid Sterba put_unaligned_le32(~crc, result); 27119c00ddcSChris Mason } 27219c00ddcSChris Mason 273d352ac68SChris Mason /* 274d352ac68SChris Mason * compute the csum for a btree block, and either verify it or write it 275d352ac68SChris Mason * into the csum field of the block. 276d352ac68SChris Mason */ 27701d58472SDaniel Dressler static int csum_tree_block(struct btrfs_fs_info *fs_info, 27801d58472SDaniel Dressler struct extent_buffer *buf, 27919c00ddcSChris Mason int verify) 28019c00ddcSChris Mason { 28101d58472SDaniel Dressler u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); 282607d432dSJosef Bacik char *result = NULL; 28319c00ddcSChris Mason unsigned long len; 28419c00ddcSChris Mason unsigned long cur_len; 28519c00ddcSChris Mason unsigned long offset = BTRFS_CSUM_SIZE; 28619c00ddcSChris Mason char *kaddr; 28719c00ddcSChris Mason unsigned long map_start; 28819c00ddcSChris Mason unsigned long map_len; 28919c00ddcSChris Mason int err; 29019c00ddcSChris Mason u32 crc = ~(u32)0; 291607d432dSJosef Bacik unsigned long inline_result; 29219c00ddcSChris Mason 29319c00ddcSChris Mason len = buf->len - offset; 29419c00ddcSChris Mason while (len > 0) { 29519c00ddcSChris Mason err = map_private_extent_buffer(buf, offset, 32, 296a6591715SChris Mason &kaddr, &map_start, &map_len); 297d397712bSChris Mason if (err) 29819c00ddcSChris Mason return 1; 29919c00ddcSChris Mason cur_len = min(len, map_len - (offset - map_start)); 300b0496686SLiu Bo crc = btrfs_csum_data(kaddr + offset - map_start, 30119c00ddcSChris Mason crc, cur_len); 30219c00ddcSChris Mason len -= cur_len; 30319c00ddcSChris Mason offset += cur_len; 30419c00ddcSChris Mason } 305607d432dSJosef Bacik if (csum_size > sizeof(inline_result)) { 306607d432dSJosef Bacik result = kzalloc(csum_size * sizeof(char), GFP_NOFS); 307607d432dSJosef Bacik if (!result) 308607d432dSJosef Bacik return 1; 309607d432dSJosef Bacik } else { 310607d432dSJosef Bacik result = (char *)&inline_result; 311607d432dSJosef Bacik } 312607d432dSJosef Bacik 31319c00ddcSChris Mason btrfs_csum_final(crc, result); 31419c00ddcSChris Mason 31519c00ddcSChris Mason if (verify) { 316607d432dSJosef Bacik if (memcmp_extent_buffer(buf, result, 0, csum_size)) { 317e4204dedSChris Mason u32 val; 318e4204dedSChris Mason u32 found = 0; 319607d432dSJosef Bacik memcpy(&found, result, csum_size); 320e4204dedSChris Mason 321607d432dSJosef Bacik read_extent_buffer(buf, &val, 0, csum_size); 322f0954c66SDavid Sterba printk_ratelimited(KERN_WARNING 323efe120a0SFrank Holton "BTRFS: %s checksum verify failed on %llu wanted %X found %X " 324193f284dSChris Mason "level %d\n", 32501d58472SDaniel Dressler fs_info->sb->s_id, buf->start, 326c1c9ff7cSGeert Uytterhoeven val, found, btrfs_header_level(buf)); 327607d432dSJosef Bacik if (result != (char *)&inline_result) 328607d432dSJosef Bacik kfree(result); 32919c00ddcSChris Mason return 1; 33019c00ddcSChris Mason } 33119c00ddcSChris Mason } else { 332607d432dSJosef Bacik write_extent_buffer(buf, result, 0, csum_size); 33319c00ddcSChris Mason } 334607d432dSJosef Bacik if (result != (char *)&inline_result) 335607d432dSJosef Bacik kfree(result); 33619c00ddcSChris Mason return 0; 33719c00ddcSChris Mason } 33819c00ddcSChris Mason 339d352ac68SChris Mason /* 340d352ac68SChris Mason * we can't consider a given block up to date unless the transid of the 341d352ac68SChris Mason * block matches the transid in the parent node's pointer. This is how we 342d352ac68SChris Mason * detect blocks that either didn't get written at all or got written 343d352ac68SChris Mason * in the wrong place. 344d352ac68SChris Mason */ 3451259ab75SChris Mason static int verify_parent_transid(struct extent_io_tree *io_tree, 346b9fab919SChris Mason struct extent_buffer *eb, u64 parent_transid, 347b9fab919SChris Mason int atomic) 3481259ab75SChris Mason { 3492ac55d41SJosef Bacik struct extent_state *cached_state = NULL; 3501259ab75SChris Mason int ret; 3512755a0deSDavid Sterba bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB); 3521259ab75SChris Mason 3531259ab75SChris Mason if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 3541259ab75SChris Mason return 0; 3551259ab75SChris Mason 356b9fab919SChris Mason if (atomic) 357b9fab919SChris Mason return -EAGAIN; 358b9fab919SChris Mason 359a26e8c9fSJosef Bacik if (need_lock) { 360a26e8c9fSJosef Bacik btrfs_tree_read_lock(eb); 361a26e8c9fSJosef Bacik btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); 362a26e8c9fSJosef Bacik } 363a26e8c9fSJosef Bacik 3642ac55d41SJosef Bacik lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 365d0082371SJeff Mahoney 0, &cached_state); 3660b32f4bbSJosef Bacik if (extent_buffer_uptodate(eb) && 3671259ab75SChris Mason btrfs_header_generation(eb) == parent_transid) { 3681259ab75SChris Mason ret = 0; 3691259ab75SChris Mason goto out; 3701259ab75SChris Mason } 37168b663d1SDavid Sterba printk_ratelimited(KERN_ERR 37268b663d1SDavid Sterba "BTRFS (device %s): parent transid verify failed on %llu wanted %llu found %llu\n", 37329549aecSWang Shilong eb->fs_info->sb->s_id, eb->start, 37429549aecSWang Shilong parent_transid, btrfs_header_generation(eb)); 3751259ab75SChris Mason ret = 1; 376a26e8c9fSJosef Bacik 377a26e8c9fSJosef Bacik /* 378a26e8c9fSJosef Bacik * Things reading via commit roots that don't have normal protection, 379a26e8c9fSJosef Bacik * like send, can have a really old block in cache that may point at a 380a26e8c9fSJosef Bacik * block that has been free'd and re-allocated. So don't clear uptodate 381a26e8c9fSJosef Bacik * if we find an eb that is under IO (dirty/writeback) because we could 382a26e8c9fSJosef Bacik * end up reading in the stale data and then writing it back out and 383a26e8c9fSJosef Bacik * making everybody very sad. 384a26e8c9fSJosef Bacik */ 385a26e8c9fSJosef Bacik if (!extent_buffer_under_io(eb)) 3860b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 38733958dc6SChris Mason out: 3882ac55d41SJosef Bacik unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 3892ac55d41SJosef Bacik &cached_state, GFP_NOFS); 390472b909fSJosef Bacik if (need_lock) 391a26e8c9fSJosef Bacik btrfs_tree_read_unlock_blocking(eb); 3921259ab75SChris Mason return ret; 3931259ab75SChris Mason } 3941259ab75SChris Mason 395d352ac68SChris Mason /* 3961104a885SDavid Sterba * Return 0 if the superblock checksum type matches the checksum value of that 3971104a885SDavid Sterba * algorithm. Pass the raw disk superblock data. 3981104a885SDavid Sterba */ 3991104a885SDavid Sterba static int btrfs_check_super_csum(char *raw_disk_sb) 4001104a885SDavid Sterba { 4011104a885SDavid Sterba struct btrfs_super_block *disk_sb = 4021104a885SDavid Sterba (struct btrfs_super_block *)raw_disk_sb; 4031104a885SDavid Sterba u16 csum_type = btrfs_super_csum_type(disk_sb); 4041104a885SDavid Sterba int ret = 0; 4051104a885SDavid Sterba 4061104a885SDavid Sterba if (csum_type == BTRFS_CSUM_TYPE_CRC32) { 4071104a885SDavid Sterba u32 crc = ~(u32)0; 4081104a885SDavid Sterba const int csum_size = sizeof(crc); 4091104a885SDavid Sterba char result[csum_size]; 4101104a885SDavid Sterba 4111104a885SDavid Sterba /* 4121104a885SDavid Sterba * The super_block structure does not span the whole 4131104a885SDavid Sterba * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space 4141104a885SDavid Sterba * is filled with zeros and is included in the checkum. 4151104a885SDavid Sterba */ 4161104a885SDavid Sterba crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE, 4171104a885SDavid Sterba crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE); 4181104a885SDavid Sterba btrfs_csum_final(crc, result); 4191104a885SDavid Sterba 4201104a885SDavid Sterba if (memcmp(raw_disk_sb, result, csum_size)) 4211104a885SDavid Sterba ret = 1; 422667e7d94SChris Mason 423667e7d94SChris Mason if (ret && btrfs_super_generation(disk_sb) < 10) { 424efe120a0SFrank Holton printk(KERN_WARNING 425efe120a0SFrank Holton "BTRFS: super block crcs don't match, older mkfs detected\n"); 426667e7d94SChris Mason ret = 0; 427667e7d94SChris Mason } 4281104a885SDavid Sterba } 4291104a885SDavid Sterba 4301104a885SDavid Sterba if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) { 431efe120a0SFrank Holton printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n", 4321104a885SDavid Sterba csum_type); 4331104a885SDavid Sterba ret = 1; 4341104a885SDavid Sterba } 4351104a885SDavid Sterba 4361104a885SDavid Sterba return ret; 4371104a885SDavid Sterba } 4381104a885SDavid Sterba 4391104a885SDavid Sterba /* 440d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 441d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 442d352ac68SChris Mason */ 443f188591eSChris Mason static int btree_read_extent_buffer_pages(struct btrfs_root *root, 444f188591eSChris Mason struct extent_buffer *eb, 445ca7a79adSChris Mason u64 start, u64 parent_transid) 446f188591eSChris Mason { 447f188591eSChris Mason struct extent_io_tree *io_tree; 448ea466794SJosef Bacik int failed = 0; 449f188591eSChris Mason int ret; 450f188591eSChris Mason int num_copies = 0; 451f188591eSChris Mason int mirror_num = 0; 452ea466794SJosef Bacik int failed_mirror = 0; 453f188591eSChris Mason 454a826d6dcSJosef Bacik clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 455f188591eSChris Mason io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree; 456f188591eSChris Mason while (1) { 457bb82ab88SArne Jansen ret = read_extent_buffer_pages(io_tree, eb, start, 458bb82ab88SArne Jansen WAIT_COMPLETE, 459f188591eSChris Mason btree_get_extent, mirror_num); 460256dd1bbSStefan Behrens if (!ret) { 461256dd1bbSStefan Behrens if (!verify_parent_transid(io_tree, eb, 462b9fab919SChris Mason parent_transid, 0)) 463ea466794SJosef Bacik break; 464256dd1bbSStefan Behrens else 465256dd1bbSStefan Behrens ret = -EIO; 466256dd1bbSStefan Behrens } 467d397712bSChris Mason 468a826d6dcSJosef Bacik /* 469a826d6dcSJosef Bacik * This buffer's crc is fine, but its contents are corrupted, so 470a826d6dcSJosef Bacik * there is no reason to read the other copies, they won't be 471a826d6dcSJosef Bacik * any less wrong. 472a826d6dcSJosef Bacik */ 473a826d6dcSJosef Bacik if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags)) 474ea466794SJosef Bacik break; 475ea466794SJosef Bacik 4765d964051SStefan Behrens num_copies = btrfs_num_copies(root->fs_info, 477f188591eSChris Mason eb->start, eb->len); 4784235298eSChris Mason if (num_copies == 1) 479ea466794SJosef Bacik break; 4804235298eSChris Mason 4815cf1ab56SJosef Bacik if (!failed_mirror) { 4825cf1ab56SJosef Bacik failed = 1; 4835cf1ab56SJosef Bacik failed_mirror = eb->read_mirror; 4845cf1ab56SJosef Bacik } 4855cf1ab56SJosef Bacik 486f188591eSChris Mason mirror_num++; 487ea466794SJosef Bacik if (mirror_num == failed_mirror) 488ea466794SJosef Bacik mirror_num++; 489ea466794SJosef Bacik 4904235298eSChris Mason if (mirror_num > num_copies) 491ea466794SJosef Bacik break; 492f188591eSChris Mason } 493ea466794SJosef Bacik 494c0901581SStefan Behrens if (failed && !ret && failed_mirror) 495ea466794SJosef Bacik repair_eb_io_failure(root, eb, failed_mirror); 496ea466794SJosef Bacik 497ea466794SJosef Bacik return ret; 498f188591eSChris Mason } 49919c00ddcSChris Mason 500d352ac68SChris Mason /* 501d397712bSChris Mason * checksum a dirty tree block before IO. This has extra checks to make sure 502d397712bSChris Mason * we only fill in the checksum field in the first page of a multi-page block 503d352ac68SChris Mason */ 504d397712bSChris Mason 50501d58472SDaniel Dressler static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page) 50619c00ddcSChris Mason { 5074eee4fa4SMiao Xie u64 start = page_offset(page); 50819c00ddcSChris Mason u64 found_start; 50919c00ddcSChris Mason struct extent_buffer *eb; 510f188591eSChris Mason 5114f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 5124f2de97aSJosef Bacik if (page != eb->pages[0]) 5134f2de97aSJosef Bacik return 0; 51419c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 515fae7f21cSDulshani Gunawardhana if (WARN_ON(found_start != start || !PageUptodate(page))) 5164f2de97aSJosef Bacik return 0; 51701d58472SDaniel Dressler csum_tree_block(fs_info, eb, 0); 51819c00ddcSChris Mason return 0; 51919c00ddcSChris Mason } 52019c00ddcSChris Mason 52101d58472SDaniel Dressler static int check_tree_block_fsid(struct btrfs_fs_info *fs_info, 5222b82032cSYan Zheng struct extent_buffer *eb) 5232b82032cSYan Zheng { 52401d58472SDaniel Dressler struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 5252b82032cSYan Zheng u8 fsid[BTRFS_UUID_SIZE]; 5262b82032cSYan Zheng int ret = 1; 5272b82032cSYan Zheng 5280a4e5586SRoss Kirk read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE); 5292b82032cSYan Zheng while (fs_devices) { 5302b82032cSYan Zheng if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) { 5312b82032cSYan Zheng ret = 0; 5322b82032cSYan Zheng break; 5332b82032cSYan Zheng } 5342b82032cSYan Zheng fs_devices = fs_devices->seed; 5352b82032cSYan Zheng } 5362b82032cSYan Zheng return ret; 5372b82032cSYan Zheng } 5382b82032cSYan Zheng 539a826d6dcSJosef Bacik #define CORRUPT(reason, eb, root, slot) \ 540efe120a0SFrank Holton btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \ 541efe120a0SFrank Holton "root=%llu, slot=%d", reason, \ 542c1c9ff7cSGeert Uytterhoeven btrfs_header_bytenr(eb), root->objectid, slot) 543a826d6dcSJosef Bacik 544a826d6dcSJosef Bacik static noinline int check_leaf(struct btrfs_root *root, 545a826d6dcSJosef Bacik struct extent_buffer *leaf) 546a826d6dcSJosef Bacik { 547a826d6dcSJosef Bacik struct btrfs_key key; 548a826d6dcSJosef Bacik struct btrfs_key leaf_key; 549a826d6dcSJosef Bacik u32 nritems = btrfs_header_nritems(leaf); 550a826d6dcSJosef Bacik int slot; 551a826d6dcSJosef Bacik 552a826d6dcSJosef Bacik if (nritems == 0) 553a826d6dcSJosef Bacik return 0; 554a826d6dcSJosef Bacik 555a826d6dcSJosef Bacik /* Check the 0 item */ 556a826d6dcSJosef Bacik if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) != 557a826d6dcSJosef Bacik BTRFS_LEAF_DATA_SIZE(root)) { 558a826d6dcSJosef Bacik CORRUPT("invalid item offset size pair", leaf, root, 0); 559a826d6dcSJosef Bacik return -EIO; 560a826d6dcSJosef Bacik } 561a826d6dcSJosef Bacik 562a826d6dcSJosef Bacik /* 563a826d6dcSJosef Bacik * Check to make sure each items keys are in the correct order and their 564a826d6dcSJosef Bacik * offsets make sense. We only have to loop through nritems-1 because 565a826d6dcSJosef Bacik * we check the current slot against the next slot, which verifies the 566a826d6dcSJosef Bacik * next slot's offset+size makes sense and that the current's slot 567a826d6dcSJosef Bacik * offset is correct. 568a826d6dcSJosef Bacik */ 569a826d6dcSJosef Bacik for (slot = 0; slot < nritems - 1; slot++) { 570a826d6dcSJosef Bacik btrfs_item_key_to_cpu(leaf, &leaf_key, slot); 571a826d6dcSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, slot + 1); 572a826d6dcSJosef Bacik 573a826d6dcSJosef Bacik /* Make sure the keys are in the right order */ 574a826d6dcSJosef Bacik if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) { 575a826d6dcSJosef Bacik CORRUPT("bad key order", leaf, root, slot); 576a826d6dcSJosef Bacik return -EIO; 577a826d6dcSJosef Bacik } 578a826d6dcSJosef Bacik 579a826d6dcSJosef Bacik /* 580a826d6dcSJosef Bacik * Make sure the offset and ends are right, remember that the 581a826d6dcSJosef Bacik * item data starts at the end of the leaf and grows towards the 582a826d6dcSJosef Bacik * front. 583a826d6dcSJosef Bacik */ 584a826d6dcSJosef Bacik if (btrfs_item_offset_nr(leaf, slot) != 585a826d6dcSJosef Bacik btrfs_item_end_nr(leaf, slot + 1)) { 586a826d6dcSJosef Bacik CORRUPT("slot offset bad", leaf, root, slot); 587a826d6dcSJosef Bacik return -EIO; 588a826d6dcSJosef Bacik } 589a826d6dcSJosef Bacik 590a826d6dcSJosef Bacik /* 591a826d6dcSJosef Bacik * Check to make sure that we don't point outside of the leaf, 592a826d6dcSJosef Bacik * just incase all the items are consistent to eachother, but 593a826d6dcSJosef Bacik * all point outside of the leaf. 594a826d6dcSJosef Bacik */ 595a826d6dcSJosef Bacik if (btrfs_item_end_nr(leaf, slot) > 596a826d6dcSJosef Bacik BTRFS_LEAF_DATA_SIZE(root)) { 597a826d6dcSJosef Bacik CORRUPT("slot end outside of leaf", leaf, root, slot); 598a826d6dcSJosef Bacik return -EIO; 599a826d6dcSJosef Bacik } 600a826d6dcSJosef Bacik } 601a826d6dcSJosef Bacik 602a826d6dcSJosef Bacik return 0; 603a826d6dcSJosef Bacik } 604a826d6dcSJosef Bacik 605facc8a22SMiao Xie static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio, 606facc8a22SMiao Xie u64 phy_offset, struct page *page, 607facc8a22SMiao Xie u64 start, u64 end, int mirror) 608ce9adaa5SChris Mason { 609ce9adaa5SChris Mason u64 found_start; 610ce9adaa5SChris Mason int found_level; 611ce9adaa5SChris Mason struct extent_buffer *eb; 612ce9adaa5SChris Mason struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 613f188591eSChris Mason int ret = 0; 614727011e0SChris Mason int reads_done; 615ce9adaa5SChris Mason 616ce9adaa5SChris Mason if (!page->private) 617ce9adaa5SChris Mason goto out; 618d397712bSChris Mason 6194f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 620d397712bSChris Mason 6210b32f4bbSJosef Bacik /* the pending IO might have been the only thing that kept this buffer 6220b32f4bbSJosef Bacik * in memory. Make sure we have a ref for all this other checks 6230b32f4bbSJosef Bacik */ 6240b32f4bbSJosef Bacik extent_buffer_get(eb); 6250b32f4bbSJosef Bacik 6260b32f4bbSJosef Bacik reads_done = atomic_dec_and_test(&eb->io_pages); 627727011e0SChris Mason if (!reads_done) 628727011e0SChris Mason goto err; 629f188591eSChris Mason 6305cf1ab56SJosef Bacik eb->read_mirror = mirror; 631656f30dbSFilipe Manana if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 632ea466794SJosef Bacik ret = -EIO; 633ea466794SJosef Bacik goto err; 634ea466794SJosef Bacik } 635ea466794SJosef Bacik 636ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 637727011e0SChris Mason if (found_start != eb->start) { 63868b663d1SDavid Sterba printk_ratelimited(KERN_ERR "BTRFS (device %s): bad tree block start " 639193f284dSChris Mason "%llu %llu\n", 64029549aecSWang Shilong eb->fs_info->sb->s_id, found_start, eb->start); 641f188591eSChris Mason ret = -EIO; 642ce9adaa5SChris Mason goto err; 643ce9adaa5SChris Mason } 64401d58472SDaniel Dressler if (check_tree_block_fsid(root->fs_info, eb)) { 64568b663d1SDavid Sterba printk_ratelimited(KERN_ERR "BTRFS (device %s): bad fsid on block %llu\n", 64629549aecSWang Shilong eb->fs_info->sb->s_id, eb->start); 6471259ab75SChris Mason ret = -EIO; 6481259ab75SChris Mason goto err; 6491259ab75SChris Mason } 650ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 6511c24c3ceSJosef Bacik if (found_level >= BTRFS_MAX_LEVEL) { 65268b663d1SDavid Sterba btrfs_err(root->fs_info, "bad tree block level %d", 6531c24c3ceSJosef Bacik (int)btrfs_header_level(eb)); 6541c24c3ceSJosef Bacik ret = -EIO; 6551c24c3ceSJosef Bacik goto err; 6561c24c3ceSJosef Bacik } 657ce9adaa5SChris Mason 65885d4e461SChris Mason btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), 65985d4e461SChris Mason eb, found_level); 6604008c04aSChris Mason 66101d58472SDaniel Dressler ret = csum_tree_block(root->fs_info, eb, 1); 662a826d6dcSJosef Bacik if (ret) { 663f188591eSChris Mason ret = -EIO; 664a826d6dcSJosef Bacik goto err; 665a826d6dcSJosef Bacik } 666a826d6dcSJosef Bacik 667a826d6dcSJosef Bacik /* 668a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 669a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 670a826d6dcSJosef Bacik * return -EIO. 671a826d6dcSJosef Bacik */ 672a826d6dcSJosef Bacik if (found_level == 0 && check_leaf(root, eb)) { 673a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 674a826d6dcSJosef Bacik ret = -EIO; 675a826d6dcSJosef Bacik } 676ce9adaa5SChris Mason 6770b32f4bbSJosef Bacik if (!ret) 6780b32f4bbSJosef Bacik set_extent_buffer_uptodate(eb); 679ce9adaa5SChris Mason err: 68079fb65a1SJosef Bacik if (reads_done && 68179fb65a1SJosef Bacik test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 6824bb31e92SArne Jansen btree_readahead_hook(root, eb, eb->start, ret); 6834bb31e92SArne Jansen 68453b381b3SDavid Woodhouse if (ret) { 68553b381b3SDavid Woodhouse /* 68653b381b3SDavid Woodhouse * our io error hook is going to dec the io pages 68753b381b3SDavid Woodhouse * again, we have to make sure it has something 68853b381b3SDavid Woodhouse * to decrement 68953b381b3SDavid Woodhouse */ 69053b381b3SDavid Woodhouse atomic_inc(&eb->io_pages); 6910b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 69253b381b3SDavid Woodhouse } 6930b32f4bbSJosef Bacik free_extent_buffer(eb); 694ce9adaa5SChris Mason out: 695f188591eSChris Mason return ret; 696ce9adaa5SChris Mason } 697ce9adaa5SChris Mason 698ea466794SJosef Bacik static int btree_io_failed_hook(struct page *page, int failed_mirror) 6994bb31e92SArne Jansen { 7004bb31e92SArne Jansen struct extent_buffer *eb; 7014bb31e92SArne Jansen struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 7024bb31e92SArne Jansen 7034f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 704656f30dbSFilipe Manana set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); 7055cf1ab56SJosef Bacik eb->read_mirror = failed_mirror; 70653b381b3SDavid Woodhouse atomic_dec(&eb->io_pages); 707ea466794SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 7084bb31e92SArne Jansen btree_readahead_hook(root, eb, eb->start, -EIO); 7094bb31e92SArne Jansen return -EIO; /* we fixed nothing */ 7104bb31e92SArne Jansen } 7114bb31e92SArne Jansen 712ce9adaa5SChris Mason static void end_workqueue_bio(struct bio *bio, int err) 713ce9adaa5SChris Mason { 71497eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq = bio->bi_private; 715ce9adaa5SChris Mason struct btrfs_fs_info *fs_info; 7169e0af237SLiu Bo struct btrfs_workqueue *wq; 7179e0af237SLiu Bo btrfs_work_func_t func; 718ce9adaa5SChris Mason 719ce9adaa5SChris Mason fs_info = end_io_wq->info; 720ce9adaa5SChris Mason end_io_wq->error = err; 721d20f7043SChris Mason 7227b6d91daSChristoph Hellwig if (bio->bi_rw & REQ_WRITE) { 7239e0af237SLiu Bo if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) { 7249e0af237SLiu Bo wq = fs_info->endio_meta_write_workers; 7259e0af237SLiu Bo func = btrfs_endio_meta_write_helper; 7269e0af237SLiu Bo } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) { 7279e0af237SLiu Bo wq = fs_info->endio_freespace_worker; 7289e0af237SLiu Bo func = btrfs_freespace_write_helper; 7299e0af237SLiu Bo } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) { 7309e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 7319e0af237SLiu Bo func = btrfs_endio_raid56_helper; 732d20f7043SChris Mason } else { 7339e0af237SLiu Bo wq = fs_info->endio_write_workers; 7349e0af237SLiu Bo func = btrfs_endio_write_helper; 735d20f7043SChris Mason } 7369e0af237SLiu Bo } else { 7378b110e39SMiao Xie if (unlikely(end_io_wq->metadata == 7388b110e39SMiao Xie BTRFS_WQ_ENDIO_DIO_REPAIR)) { 7398b110e39SMiao Xie wq = fs_info->endio_repair_workers; 7408b110e39SMiao Xie func = btrfs_endio_repair_helper; 7418b110e39SMiao Xie } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) { 7429e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 7439e0af237SLiu Bo func = btrfs_endio_raid56_helper; 7449e0af237SLiu Bo } else if (end_io_wq->metadata) { 7459e0af237SLiu Bo wq = fs_info->endio_meta_workers; 7469e0af237SLiu Bo func = btrfs_endio_meta_helper; 7479e0af237SLiu Bo } else { 7489e0af237SLiu Bo wq = fs_info->endio_workers; 7499e0af237SLiu Bo func = btrfs_endio_helper; 7509e0af237SLiu Bo } 7519e0af237SLiu Bo } 7529e0af237SLiu Bo 7539e0af237SLiu Bo btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL); 7549e0af237SLiu Bo btrfs_queue_work(wq, &end_io_wq->work); 755ce9adaa5SChris Mason } 756ce9adaa5SChris Mason 75722c59948SChris Mason int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 758bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata) 7590b86a832SChris Mason { 76097eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 7618b110e39SMiao Xie 76297eb6b69SDavid Sterba end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS); 763ce9adaa5SChris Mason if (!end_io_wq) 764ce9adaa5SChris Mason return -ENOMEM; 765ce9adaa5SChris Mason 766ce9adaa5SChris Mason end_io_wq->private = bio->bi_private; 767ce9adaa5SChris Mason end_io_wq->end_io = bio->bi_end_io; 76822c59948SChris Mason end_io_wq->info = info; 769ce9adaa5SChris Mason end_io_wq->error = 0; 770ce9adaa5SChris Mason end_io_wq->bio = bio; 77122c59948SChris Mason end_io_wq->metadata = metadata; 772ce9adaa5SChris Mason 773ce9adaa5SChris Mason bio->bi_private = end_io_wq; 774ce9adaa5SChris Mason bio->bi_end_io = end_workqueue_bio; 77522c59948SChris Mason return 0; 77622c59948SChris Mason } 77722c59948SChris Mason 778b64a2851SChris Mason unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info) 7794854ddd0SChris Mason { 7804854ddd0SChris Mason unsigned long limit = min_t(unsigned long, 7815cdc7ad3SQu Wenruo info->thread_pool_size, 7824854ddd0SChris Mason info->fs_devices->open_devices); 7834854ddd0SChris Mason return 256 * limit; 7844854ddd0SChris Mason } 7854854ddd0SChris Mason 7864a69a410SChris Mason static void run_one_async_start(struct btrfs_work *work) 7874a69a410SChris Mason { 7884a69a410SChris Mason struct async_submit_bio *async; 78979787eaaSJeff Mahoney int ret; 7904a69a410SChris Mason 7914a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 79279787eaaSJeff Mahoney ret = async->submit_bio_start(async->inode, async->rw, async->bio, 793eaf25d93SChris Mason async->mirror_num, async->bio_flags, 794eaf25d93SChris Mason async->bio_offset); 79579787eaaSJeff Mahoney if (ret) 79679787eaaSJeff Mahoney async->error = ret; 7974a69a410SChris Mason } 7984a69a410SChris Mason 7994a69a410SChris Mason static void run_one_async_done(struct btrfs_work *work) 8008b712842SChris Mason { 8018b712842SChris Mason struct btrfs_fs_info *fs_info; 8028b712842SChris Mason struct async_submit_bio *async; 8034854ddd0SChris Mason int limit; 8048b712842SChris Mason 8058b712842SChris Mason async = container_of(work, struct async_submit_bio, work); 8068b712842SChris Mason fs_info = BTRFS_I(async->inode)->root->fs_info; 8074854ddd0SChris Mason 808b64a2851SChris Mason limit = btrfs_async_submit_limit(fs_info); 8094854ddd0SChris Mason limit = limit * 2 / 3; 8104854ddd0SChris Mason 81166657b31SJosef Bacik if (atomic_dec_return(&fs_info->nr_async_submits) < limit && 812b64a2851SChris Mason waitqueue_active(&fs_info->async_submit_wait)) 8134854ddd0SChris Mason wake_up(&fs_info->async_submit_wait); 8144854ddd0SChris Mason 81579787eaaSJeff Mahoney /* If an error occured we just want to clean up the bio and move on */ 81679787eaaSJeff Mahoney if (async->error) { 81779787eaaSJeff Mahoney bio_endio(async->bio, async->error); 81879787eaaSJeff Mahoney return; 81979787eaaSJeff Mahoney } 82079787eaaSJeff Mahoney 8214a69a410SChris Mason async->submit_bio_done(async->inode, async->rw, async->bio, 822eaf25d93SChris Mason async->mirror_num, async->bio_flags, 823eaf25d93SChris Mason async->bio_offset); 8244a69a410SChris Mason } 8254a69a410SChris Mason 8264a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 8274a69a410SChris Mason { 8284a69a410SChris Mason struct async_submit_bio *async; 8294a69a410SChris Mason 8304a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 8318b712842SChris Mason kfree(async); 8328b712842SChris Mason } 8338b712842SChris Mason 83444b8bd7eSChris Mason int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode, 83544b8bd7eSChris Mason int rw, struct bio *bio, int mirror_num, 836c8b97818SChris Mason unsigned long bio_flags, 837eaf25d93SChris Mason u64 bio_offset, 8384a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_start, 8394a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_done) 84044b8bd7eSChris Mason { 84144b8bd7eSChris Mason struct async_submit_bio *async; 84244b8bd7eSChris Mason 84344b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 84444b8bd7eSChris Mason if (!async) 84544b8bd7eSChris Mason return -ENOMEM; 84644b8bd7eSChris Mason 84744b8bd7eSChris Mason async->inode = inode; 84844b8bd7eSChris Mason async->rw = rw; 84944b8bd7eSChris Mason async->bio = bio; 85044b8bd7eSChris Mason async->mirror_num = mirror_num; 8514a69a410SChris Mason async->submit_bio_start = submit_bio_start; 8524a69a410SChris Mason async->submit_bio_done = submit_bio_done; 8534a69a410SChris Mason 8549e0af237SLiu Bo btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start, 8555cdc7ad3SQu Wenruo run_one_async_done, run_one_async_free); 8564a69a410SChris Mason 857c8b97818SChris Mason async->bio_flags = bio_flags; 858eaf25d93SChris Mason async->bio_offset = bio_offset; 8598c8bee1dSChris Mason 86079787eaaSJeff Mahoney async->error = 0; 86179787eaaSJeff Mahoney 862cb03c743SChris Mason atomic_inc(&fs_info->nr_async_submits); 863d313d7a3SChris Mason 8647b6d91daSChristoph Hellwig if (rw & REQ_SYNC) 8655cdc7ad3SQu Wenruo btrfs_set_work_high_priority(&async->work); 866d313d7a3SChris Mason 8675cdc7ad3SQu Wenruo btrfs_queue_work(fs_info->workers, &async->work); 8689473f16cSChris Mason 869771ed689SChris Mason while (atomic_read(&fs_info->async_submit_draining) && 870771ed689SChris Mason atomic_read(&fs_info->nr_async_submits)) { 871771ed689SChris Mason wait_event(fs_info->async_submit_wait, 872771ed689SChris Mason (atomic_read(&fs_info->nr_async_submits) == 0)); 873771ed689SChris Mason } 874771ed689SChris Mason 87544b8bd7eSChris Mason return 0; 87644b8bd7eSChris Mason } 87744b8bd7eSChris Mason 878ce3ed71aSChris Mason static int btree_csum_one_bio(struct bio *bio) 879ce3ed71aSChris Mason { 8802c30c71bSKent Overstreet struct bio_vec *bvec; 881ce3ed71aSChris Mason struct btrfs_root *root; 8822c30c71bSKent Overstreet int i, ret = 0; 883ce3ed71aSChris Mason 8842c30c71bSKent Overstreet bio_for_each_segment_all(bvec, bio, i) { 885ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 88601d58472SDaniel Dressler ret = csum_dirty_buffer(root->fs_info, bvec->bv_page); 88779787eaaSJeff Mahoney if (ret) 88879787eaaSJeff Mahoney break; 889ce3ed71aSChris Mason } 8902c30c71bSKent Overstreet 89179787eaaSJeff Mahoney return ret; 892ce3ed71aSChris Mason } 893ce3ed71aSChris Mason 8944a69a410SChris Mason static int __btree_submit_bio_start(struct inode *inode, int rw, 8954a69a410SChris Mason struct bio *bio, int mirror_num, 896eaf25d93SChris Mason unsigned long bio_flags, 897eaf25d93SChris Mason u64 bio_offset) 89822c59948SChris Mason { 8998b712842SChris Mason /* 9008b712842SChris Mason * when we're called for a write, we're already in the async 9015443be45SChris Mason * submission context. Just jump into btrfs_map_bio 9028b712842SChris Mason */ 90379787eaaSJeff Mahoney return btree_csum_one_bio(bio); 90422c59948SChris Mason } 90522c59948SChris Mason 9064a69a410SChris Mason static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio, 907eaf25d93SChris Mason int mirror_num, unsigned long bio_flags, 908eaf25d93SChris Mason u64 bio_offset) 9094a69a410SChris Mason { 91061891923SStefan Behrens int ret; 91161891923SStefan Behrens 9128b712842SChris Mason /* 9134a69a410SChris Mason * when we're called for a write, we're already in the async 9144a69a410SChris Mason * submission context. Just jump into btrfs_map_bio 9158b712842SChris Mason */ 91661891923SStefan Behrens ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1); 91761891923SStefan Behrens if (ret) 91861891923SStefan Behrens bio_endio(bio, ret); 91961891923SStefan Behrens return ret; 9200b86a832SChris Mason } 9210b86a832SChris Mason 922de0022b9SJosef Bacik static int check_async_write(struct inode *inode, unsigned long bio_flags) 923de0022b9SJosef Bacik { 924de0022b9SJosef Bacik if (bio_flags & EXTENT_BIO_TREE_LOG) 925de0022b9SJosef Bacik return 0; 926de0022b9SJosef Bacik #ifdef CONFIG_X86 927de0022b9SJosef Bacik if (cpu_has_xmm4_2) 928de0022b9SJosef Bacik return 0; 929de0022b9SJosef Bacik #endif 930de0022b9SJosef Bacik return 1; 931de0022b9SJosef Bacik } 932de0022b9SJosef Bacik 93344b8bd7eSChris Mason static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, 934eaf25d93SChris Mason int mirror_num, unsigned long bio_flags, 935eaf25d93SChris Mason u64 bio_offset) 93644b8bd7eSChris Mason { 937de0022b9SJosef Bacik int async = check_async_write(inode, bio_flags); 9384a69a410SChris Mason int ret; 939cad321adSChris Mason 9407b6d91daSChristoph Hellwig if (!(rw & REQ_WRITE)) { 941cad321adSChris Mason /* 942cad321adSChris Mason * called for a read, do the setup so that checksum validation 943cad321adSChris Mason * can happen in the async kernel threads 944cad321adSChris Mason */ 945f3f266abSChris Mason ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info, 946bfebd8b5SDavid Sterba bio, BTRFS_WQ_ENDIO_METADATA); 9471d4284bdSChris Mason if (ret) 94861891923SStefan Behrens goto out_w_error; 94961891923SStefan Behrens ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, 9506f3577bdSChris Mason mirror_num, 0); 951de0022b9SJosef Bacik } else if (!async) { 952de0022b9SJosef Bacik ret = btree_csum_one_bio(bio); 953de0022b9SJosef Bacik if (ret) 95461891923SStefan Behrens goto out_w_error; 95561891923SStefan Behrens ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, 956de0022b9SJosef Bacik mirror_num, 0); 95761891923SStefan Behrens } else { 958cad321adSChris Mason /* 95961891923SStefan Behrens * kthread helpers are used to submit writes so that 96061891923SStefan Behrens * checksumming can happen in parallel across all CPUs 961cad321adSChris Mason */ 96261891923SStefan Behrens ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, 963c8b97818SChris Mason inode, rw, bio, mirror_num, 0, 964eaf25d93SChris Mason bio_offset, 9654a69a410SChris Mason __btree_submit_bio_start, 9664a69a410SChris Mason __btree_submit_bio_done); 96744b8bd7eSChris Mason } 96844b8bd7eSChris Mason 96961891923SStefan Behrens if (ret) { 97061891923SStefan Behrens out_w_error: 97161891923SStefan Behrens bio_endio(bio, ret); 97261891923SStefan Behrens } 97361891923SStefan Behrens return ret; 97461891923SStefan Behrens } 97561891923SStefan Behrens 9763dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 977784b4e29SChris Mason static int btree_migratepage(struct address_space *mapping, 978a6bc32b8SMel Gorman struct page *newpage, struct page *page, 979a6bc32b8SMel Gorman enum migrate_mode mode) 980784b4e29SChris Mason { 981784b4e29SChris Mason /* 982784b4e29SChris Mason * we can't safely write a btree page from here, 983784b4e29SChris Mason * we haven't done the locking hook 984784b4e29SChris Mason */ 985784b4e29SChris Mason if (PageDirty(page)) 986784b4e29SChris Mason return -EAGAIN; 987784b4e29SChris Mason /* 988784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 989784b4e29SChris Mason * We must have no buffers or drop them. 990784b4e29SChris Mason */ 991784b4e29SChris Mason if (page_has_private(page) && 992784b4e29SChris Mason !try_to_release_page(page, GFP_KERNEL)) 993784b4e29SChris Mason return -EAGAIN; 994a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 995784b4e29SChris Mason } 9963dd1462eSJan Beulich #endif 997784b4e29SChris Mason 9980da5468fSChris Mason 9990da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 10000da5468fSChris Mason struct writeback_control *wbc) 10010da5468fSChris Mason { 1002e2d84521SMiao Xie struct btrfs_fs_info *fs_info; 1003e2d84521SMiao Xie int ret; 1004e2d84521SMiao Xie 1005d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 1006448d640bSChris Mason 1007448d640bSChris Mason if (wbc->for_kupdate) 1008448d640bSChris Mason return 0; 1009448d640bSChris Mason 1010e2d84521SMiao Xie fs_info = BTRFS_I(mapping->host)->root->fs_info; 1011b9473439SChris Mason /* this is a bit racy, but that's ok */ 1012e2d84521SMiao Xie ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes, 1013e2d84521SMiao Xie BTRFS_DIRTY_METADATA_THRESH); 1014e2d84521SMiao Xie if (ret < 0) 1015793955bcSChris Mason return 0; 1016793955bcSChris Mason } 10170b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 10180da5468fSChris Mason } 10190da5468fSChris Mason 1020b2950863SChristoph Hellwig static int btree_readpage(struct file *file, struct page *page) 10215f39d397SChris Mason { 1022d1310b2eSChris Mason struct extent_io_tree *tree; 1023d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 10248ddc7d9cSJan Schmidt return extent_read_full_page(tree, page, btree_get_extent, 0); 10255f39d397SChris Mason } 10265f39d397SChris Mason 102770dec807SChris Mason static int btree_releasepage(struct page *page, gfp_t gfp_flags) 10285f39d397SChris Mason { 102998509cfcSChris Mason if (PageWriteback(page) || PageDirty(page)) 103098509cfcSChris Mason return 0; 10310c4e538bSDavid Sterba 1032f7a52a40SDavid Sterba return try_release_extent_buffer(page); 1033d98237b3SChris Mason } 1034d98237b3SChris Mason 1035d47992f8SLukas Czerner static void btree_invalidatepage(struct page *page, unsigned int offset, 1036d47992f8SLukas Czerner unsigned int length) 1037d98237b3SChris Mason { 1038d1310b2eSChris Mason struct extent_io_tree *tree; 1039d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 10405f39d397SChris Mason extent_invalidatepage(tree, page, offset); 10415f39d397SChris Mason btree_releasepage(page, GFP_NOFS); 10429ad6b7bcSChris Mason if (PagePrivate(page)) { 1043efe120a0SFrank Holton btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info, 1044efe120a0SFrank Holton "page private not zero on page %llu", 1045efe120a0SFrank Holton (unsigned long long)page_offset(page)); 10469ad6b7bcSChris Mason ClearPagePrivate(page); 10479ad6b7bcSChris Mason set_page_private(page, 0); 10489ad6b7bcSChris Mason page_cache_release(page); 10499ad6b7bcSChris Mason } 1050d98237b3SChris Mason } 1051d98237b3SChris Mason 10520b32f4bbSJosef Bacik static int btree_set_page_dirty(struct page *page) 10530b32f4bbSJosef Bacik { 1054bb146eb2SJosef Bacik #ifdef DEBUG 10550b32f4bbSJosef Bacik struct extent_buffer *eb; 10560b32f4bbSJosef Bacik 10570b32f4bbSJosef Bacik BUG_ON(!PagePrivate(page)); 10580b32f4bbSJosef Bacik eb = (struct extent_buffer *)page->private; 10590b32f4bbSJosef Bacik BUG_ON(!eb); 10600b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 10610b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 10620b32f4bbSJosef Bacik btrfs_assert_tree_locked(eb); 1063bb146eb2SJosef Bacik #endif 10640b32f4bbSJosef Bacik return __set_page_dirty_nobuffers(page); 10650b32f4bbSJosef Bacik } 10660b32f4bbSJosef Bacik 10677f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 1068d98237b3SChris Mason .readpage = btree_readpage, 10690da5468fSChris Mason .writepages = btree_writepages, 10705f39d397SChris Mason .releasepage = btree_releasepage, 10715f39d397SChris Mason .invalidatepage = btree_invalidatepage, 10725a92bc88SChris Mason #ifdef CONFIG_MIGRATION 1073784b4e29SChris Mason .migratepage = btree_migratepage, 10745a92bc88SChris Mason #endif 10750b32f4bbSJosef Bacik .set_page_dirty = btree_set_page_dirty, 1076d98237b3SChris Mason }; 1077123abc88SChris Mason 1078d3e46feaSDavid Sterba void readahead_tree_block(struct btrfs_root *root, u64 bytenr) 1079090d1875SChris Mason { 10805f39d397SChris Mason struct extent_buffer *buf = NULL; 10815f39d397SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 1082090d1875SChris Mason 1083a83fffb7SDavid Sterba buf = btrfs_find_create_tree_block(root, bytenr); 10845f39d397SChris Mason if (!buf) 10856197d86eSDavid Sterba return; 1086d1310b2eSChris Mason read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, 1087bb82ab88SArne Jansen buf, 0, WAIT_NONE, btree_get_extent, 0); 10885f39d397SChris Mason free_extent_buffer(buf); 1089090d1875SChris Mason } 1090090d1875SChris Mason 1091c0dcaa4dSDavid Sterba int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, 1092ab0fff03SArne Jansen int mirror_num, struct extent_buffer **eb) 1093ab0fff03SArne Jansen { 1094ab0fff03SArne Jansen struct extent_buffer *buf = NULL; 1095ab0fff03SArne Jansen struct inode *btree_inode = root->fs_info->btree_inode; 1096ab0fff03SArne Jansen struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree; 1097ab0fff03SArne Jansen int ret; 1098ab0fff03SArne Jansen 1099a83fffb7SDavid Sterba buf = btrfs_find_create_tree_block(root, bytenr); 1100ab0fff03SArne Jansen if (!buf) 1101ab0fff03SArne Jansen return 0; 1102ab0fff03SArne Jansen 1103ab0fff03SArne Jansen set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags); 1104ab0fff03SArne Jansen 1105ab0fff03SArne Jansen ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK, 1106ab0fff03SArne Jansen btree_get_extent, mirror_num); 1107ab0fff03SArne Jansen if (ret) { 1108ab0fff03SArne Jansen free_extent_buffer(buf); 1109ab0fff03SArne Jansen return ret; 1110ab0fff03SArne Jansen } 1111ab0fff03SArne Jansen 1112ab0fff03SArne Jansen if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) { 1113ab0fff03SArne Jansen free_extent_buffer(buf); 1114ab0fff03SArne Jansen return -EIO; 11150b32f4bbSJosef Bacik } else if (extent_buffer_uptodate(buf)) { 1116ab0fff03SArne Jansen *eb = buf; 1117ab0fff03SArne Jansen } else { 1118ab0fff03SArne Jansen free_extent_buffer(buf); 1119ab0fff03SArne Jansen } 1120ab0fff03SArne Jansen return 0; 1121ab0fff03SArne Jansen } 1122ab0fff03SArne Jansen 112301d58472SDaniel Dressler struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info, 11240308af44SDavid Sterba u64 bytenr) 11250999df54SChris Mason { 112601d58472SDaniel Dressler return find_extent_buffer(fs_info, bytenr); 11270999df54SChris Mason } 11280999df54SChris Mason 11290999df54SChris Mason struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root, 1130a83fffb7SDavid Sterba u64 bytenr) 11310999df54SChris Mason { 1132fccb84c9SDavid Sterba if (btrfs_test_is_dummy_root(root)) 1133ce3e6984SDavid Sterba return alloc_test_extent_buffer(root->fs_info, bytenr); 1134ce3e6984SDavid Sterba return alloc_extent_buffer(root->fs_info, bytenr); 11350999df54SChris Mason } 11360999df54SChris Mason 11370999df54SChris Mason 1138e02119d5SChris Mason int btrfs_write_tree_block(struct extent_buffer *buf) 1139e02119d5SChris Mason { 1140727011e0SChris Mason return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start, 11418aa38c31SChristoph Hellwig buf->start + buf->len - 1); 1142e02119d5SChris Mason } 1143e02119d5SChris Mason 1144e02119d5SChris Mason int btrfs_wait_tree_block_writeback(struct extent_buffer *buf) 1145e02119d5SChris Mason { 1146727011e0SChris Mason return filemap_fdatawait_range(buf->pages[0]->mapping, 11478aa38c31SChristoph Hellwig buf->start, buf->start + buf->len - 1); 1148e02119d5SChris Mason } 1149e02119d5SChris Mason 1150db94535dSChris Mason struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr, 1151ce86cd59SDavid Sterba u64 parent_transid) 1152e20d96d6SChris Mason { 11535f39d397SChris Mason struct extent_buffer *buf = NULL; 115419c00ddcSChris Mason int ret; 115519c00ddcSChris Mason 1156a83fffb7SDavid Sterba buf = btrfs_find_create_tree_block(root, bytenr); 11575f39d397SChris Mason if (!buf) 1158d98237b3SChris Mason return NULL; 1159e4204dedSChris Mason 1160ca7a79adSChris Mason ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 11610f0fe8f7SFilipe David Borba Manana if (ret) { 11620f0fe8f7SFilipe David Borba Manana free_extent_buffer(buf); 11630f0fe8f7SFilipe David Borba Manana return NULL; 11640f0fe8f7SFilipe David Borba Manana } 11655f39d397SChris Mason return buf; 1166ce9adaa5SChris Mason 1167eb60ceacSChris Mason } 1168eb60ceacSChris Mason 116901d58472SDaniel Dressler void clean_tree_block(struct btrfs_trans_handle *trans, 117001d58472SDaniel Dressler struct btrfs_fs_info *fs_info, 11715f39d397SChris Mason struct extent_buffer *buf) 1172ed2ff2cbSChris Mason { 117355c69072SChris Mason if (btrfs_header_generation(buf) == 1174e2d84521SMiao Xie fs_info->running_transaction->transid) { 1175b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1176b4ce94deSChris Mason 1177b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1178e2d84521SMiao Xie __percpu_counter_add(&fs_info->dirty_metadata_bytes, 1179e2d84521SMiao Xie -buf->len, 1180e2d84521SMiao Xie fs_info->dirty_metadata_batch); 1181b9473439SChris Mason /* ugh, clear_extent_buffer_dirty needs to lock the page */ 1182b9473439SChris Mason btrfs_set_lock_blocking(buf); 11830b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 1184925baeddSChris Mason } 11855f39d397SChris Mason } 1186ed7b63ebSJosef Bacik } 11875f39d397SChris Mason 11888257b2dcSMiao Xie static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void) 11898257b2dcSMiao Xie { 11908257b2dcSMiao Xie struct btrfs_subvolume_writers *writers; 11918257b2dcSMiao Xie int ret; 11928257b2dcSMiao Xie 11938257b2dcSMiao Xie writers = kmalloc(sizeof(*writers), GFP_NOFS); 11948257b2dcSMiao Xie if (!writers) 11958257b2dcSMiao Xie return ERR_PTR(-ENOMEM); 11968257b2dcSMiao Xie 1197908c7f19STejun Heo ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL); 11988257b2dcSMiao Xie if (ret < 0) { 11998257b2dcSMiao Xie kfree(writers); 12008257b2dcSMiao Xie return ERR_PTR(ret); 12018257b2dcSMiao Xie } 12028257b2dcSMiao Xie 12038257b2dcSMiao Xie init_waitqueue_head(&writers->wait); 12048257b2dcSMiao Xie return writers; 12058257b2dcSMiao Xie } 12068257b2dcSMiao Xie 12078257b2dcSMiao Xie static void 12088257b2dcSMiao Xie btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers) 12098257b2dcSMiao Xie { 12108257b2dcSMiao Xie percpu_counter_destroy(&writers->counter); 12118257b2dcSMiao Xie kfree(writers); 12128257b2dcSMiao Xie } 12138257b2dcSMiao Xie 1214707e8a07SDavid Sterba static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize, 1215707e8a07SDavid Sterba struct btrfs_root *root, struct btrfs_fs_info *fs_info, 1216e20d96d6SChris Mason u64 objectid) 1217d97e63b6SChris Mason { 1218cfaa7295SChris Mason root->node = NULL; 1219a28ec197SChris Mason root->commit_root = NULL; 1220db94535dSChris Mason root->sectorsize = sectorsize; 1221db94535dSChris Mason root->nodesize = nodesize; 122287ee04ebSChris Mason root->stripesize = stripesize; 122327cdeb70SMiao Xie root->state = 0; 1224d68fc57bSYan, Zheng root->orphan_cleanup_state = 0; 12250b86a832SChris Mason 12260f7d52f4SChris Mason root->objectid = objectid; 12270f7d52f4SChris Mason root->last_trans = 0; 122813a8a7c8SYan, Zheng root->highest_objectid = 0; 1229eb73c1b7SMiao Xie root->nr_delalloc_inodes = 0; 1230199c2a9cSMiao Xie root->nr_ordered_extents = 0; 123158176a96SJosef Bacik root->name = NULL; 12326bef4d31SEric Paris root->inode_tree = RB_ROOT; 123316cdcec7SMiao Xie INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1234f0486c68SYan, Zheng root->block_rsv = NULL; 1235d68fc57bSYan, Zheng root->orphan_block_rsv = NULL; 12360b86a832SChris Mason 12370b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 12385d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 1239eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_inodes); 1240eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_root); 1241199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_extents); 1242199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_root); 12432ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 12442ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 1245d68fc57bSYan, Zheng spin_lock_init(&root->orphan_lock); 12465d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1247eb73c1b7SMiao Xie spin_lock_init(&root->delalloc_lock); 1248199c2a9cSMiao Xie spin_lock_init(&root->ordered_extent_lock); 1249f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 12502ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 12512ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 1252a2135011SChris Mason mutex_init(&root->objectid_mutex); 1253e02119d5SChris Mason mutex_init(&root->log_mutex); 125431f3d255SMiao Xie mutex_init(&root->ordered_extent_mutex); 1255573bfb72SMiao Xie mutex_init(&root->delalloc_mutex); 12567237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 12577237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 12587237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 12598b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[0]); 12608b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[1]); 12617237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 12627237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 12637237f183SYan Zheng atomic_set(&root->log_writers, 0); 12642ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 12658a35d95fSJosef Bacik atomic_set(&root->orphan_inodes, 0); 1266b0feb9d9SMiao Xie atomic_set(&root->refs, 1); 12678257b2dcSMiao Xie atomic_set(&root->will_be_snapshoted, 0); 12687237f183SYan Zheng root->log_transid = 0; 1269d1433debSMiao Xie root->log_transid_committed = -1; 1270257c62e1SChris Mason root->last_log_commit = 0; 127106ea65a3SJosef Bacik if (fs_info) 1272d0c803c4SChris Mason extent_io_tree_init(&root->dirty_log_pages, 1273f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 1274017e5369SChris Mason 12753768f368SChris Mason memset(&root->root_key, 0, sizeof(root->root_key)); 12763768f368SChris Mason memset(&root->root_item, 0, sizeof(root->root_item)); 12776702ed49SChris Mason memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 127806ea65a3SJosef Bacik if (fs_info) 12793f157a2fSChris Mason root->defrag_trans_start = fs_info->generation; 128006ea65a3SJosef Bacik else 128106ea65a3SJosef Bacik root->defrag_trans_start = 0; 12824d775673SChris Mason root->root_key.objectid = objectid; 12830ee5dc67SAl Viro root->anon_dev = 0; 12848ea05e3aSAlexander Block 12855f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 12863768f368SChris Mason } 12873768f368SChris Mason 1288f84a8bd6SAl Viro static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info) 12896f07e42eSAl Viro { 12906f07e42eSAl Viro struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS); 12916f07e42eSAl Viro if (root) 12926f07e42eSAl Viro root->fs_info = fs_info; 12936f07e42eSAl Viro return root; 12946f07e42eSAl Viro } 12956f07e42eSAl Viro 129606ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 129706ea65a3SJosef Bacik /* Should only be used by the testing infrastructure */ 129806ea65a3SJosef Bacik struct btrfs_root *btrfs_alloc_dummy_root(void) 129906ea65a3SJosef Bacik { 130006ea65a3SJosef Bacik struct btrfs_root *root; 130106ea65a3SJosef Bacik 130206ea65a3SJosef Bacik root = btrfs_alloc_root(NULL); 130306ea65a3SJosef Bacik if (!root) 130406ea65a3SJosef Bacik return ERR_PTR(-ENOMEM); 1305707e8a07SDavid Sterba __setup_root(4096, 4096, 4096, root, NULL, 1); 130627cdeb70SMiao Xie set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state); 1307faa2dbf0SJosef Bacik root->alloc_bytenr = 0; 130806ea65a3SJosef Bacik 130906ea65a3SJosef Bacik return root; 131006ea65a3SJosef Bacik } 131106ea65a3SJosef Bacik #endif 131206ea65a3SJosef Bacik 131320897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 131420897f5cSArne Jansen struct btrfs_fs_info *fs_info, 131520897f5cSArne Jansen u64 objectid) 131620897f5cSArne Jansen { 131720897f5cSArne Jansen struct extent_buffer *leaf; 131820897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 131920897f5cSArne Jansen struct btrfs_root *root; 132020897f5cSArne Jansen struct btrfs_key key; 132120897f5cSArne Jansen int ret = 0; 13226463fe58SStefan Behrens uuid_le uuid; 132320897f5cSArne Jansen 132420897f5cSArne Jansen root = btrfs_alloc_root(fs_info); 132520897f5cSArne Jansen if (!root) 132620897f5cSArne Jansen return ERR_PTR(-ENOMEM); 132720897f5cSArne Jansen 1328707e8a07SDavid Sterba __setup_root(tree_root->nodesize, tree_root->sectorsize, 1329707e8a07SDavid Sterba tree_root->stripesize, root, fs_info, objectid); 133020897f5cSArne Jansen root->root_key.objectid = objectid; 133120897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 133220897f5cSArne Jansen root->root_key.offset = 0; 133320897f5cSArne Jansen 13344d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0); 133520897f5cSArne Jansen if (IS_ERR(leaf)) { 133620897f5cSArne Jansen ret = PTR_ERR(leaf); 13371dd05682STsutomu Itoh leaf = NULL; 133820897f5cSArne Jansen goto fail; 133920897f5cSArne Jansen } 134020897f5cSArne Jansen 134120897f5cSArne Jansen memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 134220897f5cSArne Jansen btrfs_set_header_bytenr(leaf, leaf->start); 134320897f5cSArne Jansen btrfs_set_header_generation(leaf, trans->transid); 134420897f5cSArne Jansen btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 134520897f5cSArne Jansen btrfs_set_header_owner(leaf, objectid); 134620897f5cSArne Jansen root->node = leaf; 134720897f5cSArne Jansen 13480a4e5586SRoss Kirk write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(), 134920897f5cSArne Jansen BTRFS_FSID_SIZE); 135020897f5cSArne Jansen write_extent_buffer(leaf, fs_info->chunk_tree_uuid, 1351b308bc2fSGeert Uytterhoeven btrfs_header_chunk_tree_uuid(leaf), 135220897f5cSArne Jansen BTRFS_UUID_SIZE); 135320897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 135420897f5cSArne Jansen 135520897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 135627cdeb70SMiao Xie set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 135720897f5cSArne Jansen 135820897f5cSArne Jansen root->root_item.flags = 0; 135920897f5cSArne Jansen root->root_item.byte_limit = 0; 136020897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 136120897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 136220897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 136320897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 136420897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 136520897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 136620897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 13676463fe58SStefan Behrens uuid_le_gen(&uuid); 13686463fe58SStefan Behrens memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE); 136920897f5cSArne Jansen root->root_item.drop_level = 0; 137020897f5cSArne Jansen 137120897f5cSArne Jansen key.objectid = objectid; 137220897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 137320897f5cSArne Jansen key.offset = 0; 137420897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 137520897f5cSArne Jansen if (ret) 137620897f5cSArne Jansen goto fail; 137720897f5cSArne Jansen 137820897f5cSArne Jansen btrfs_tree_unlock(leaf); 137920897f5cSArne Jansen 138020897f5cSArne Jansen return root; 13811dd05682STsutomu Itoh 13821dd05682STsutomu Itoh fail: 13831dd05682STsutomu Itoh if (leaf) { 13841dd05682STsutomu Itoh btrfs_tree_unlock(leaf); 138559885b39STsutomu Itoh free_extent_buffer(root->commit_root); 13861dd05682STsutomu Itoh free_extent_buffer(leaf); 13871dd05682STsutomu Itoh } 13881dd05682STsutomu Itoh kfree(root); 13891dd05682STsutomu Itoh 13901dd05682STsutomu Itoh return ERR_PTR(ret); 139120897f5cSArne Jansen } 139220897f5cSArne Jansen 13937237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1394e02119d5SChris Mason struct btrfs_fs_info *fs_info) 13950f7d52f4SChris Mason { 13960f7d52f4SChris Mason struct btrfs_root *root; 13970f7d52f4SChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 13987237f183SYan Zheng struct extent_buffer *leaf; 1399e02119d5SChris Mason 14006f07e42eSAl Viro root = btrfs_alloc_root(fs_info); 1401e02119d5SChris Mason if (!root) 14027237f183SYan Zheng return ERR_PTR(-ENOMEM); 1403e02119d5SChris Mason 1404707e8a07SDavid Sterba __setup_root(tree_root->nodesize, tree_root->sectorsize, 1405707e8a07SDavid Sterba tree_root->stripesize, root, fs_info, 1406707e8a07SDavid Sterba BTRFS_TREE_LOG_OBJECTID); 1407e02119d5SChris Mason 1408e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1409e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1410e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 141127cdeb70SMiao Xie 14127237f183SYan Zheng /* 141327cdeb70SMiao Xie * DON'T set REF_COWS for log trees 141427cdeb70SMiao Xie * 14157237f183SYan Zheng * log trees do not get reference counted because they go away 14167237f183SYan Zheng * before a real commit is actually done. They do store pointers 14177237f183SYan Zheng * to file data extents, and those reference counts still get 14187237f183SYan Zheng * updated (along with back refs to the log tree). 14197237f183SYan Zheng */ 1420e02119d5SChris Mason 14214d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 14224d75f8a9SDavid Sterba NULL, 0, 0, 0); 14237237f183SYan Zheng if (IS_ERR(leaf)) { 14247237f183SYan Zheng kfree(root); 14257237f183SYan Zheng return ERR_CAST(leaf); 14267237f183SYan Zheng } 1427e02119d5SChris Mason 14285d4f98a2SYan Zheng memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 14295d4f98a2SYan Zheng btrfs_set_header_bytenr(leaf, leaf->start); 14305d4f98a2SYan Zheng btrfs_set_header_generation(leaf, trans->transid); 14315d4f98a2SYan Zheng btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 14325d4f98a2SYan Zheng btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID); 14337237f183SYan Zheng root->node = leaf; 1434e02119d5SChris Mason 1435e02119d5SChris Mason write_extent_buffer(root->node, root->fs_info->fsid, 14360a4e5586SRoss Kirk btrfs_header_fsid(), BTRFS_FSID_SIZE); 1437e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1438e02119d5SChris Mason btrfs_tree_unlock(root->node); 14397237f183SYan Zheng return root; 14407237f183SYan Zheng } 14417237f183SYan Zheng 14427237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 14437237f183SYan Zheng struct btrfs_fs_info *fs_info) 14447237f183SYan Zheng { 14457237f183SYan Zheng struct btrfs_root *log_root; 14467237f183SYan Zheng 14477237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 14487237f183SYan Zheng if (IS_ERR(log_root)) 14497237f183SYan Zheng return PTR_ERR(log_root); 14507237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 14517237f183SYan Zheng fs_info->log_root_tree = log_root; 14527237f183SYan Zheng return 0; 14537237f183SYan Zheng } 14547237f183SYan Zheng 14557237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 14567237f183SYan Zheng struct btrfs_root *root) 14577237f183SYan Zheng { 14587237f183SYan Zheng struct btrfs_root *log_root; 14597237f183SYan Zheng struct btrfs_inode_item *inode_item; 14607237f183SYan Zheng 14617237f183SYan Zheng log_root = alloc_log_tree(trans, root->fs_info); 14627237f183SYan Zheng if (IS_ERR(log_root)) 14637237f183SYan Zheng return PTR_ERR(log_root); 14647237f183SYan Zheng 14657237f183SYan Zheng log_root->last_trans = trans->transid; 14667237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 14677237f183SYan Zheng 14687237f183SYan Zheng inode_item = &log_root->root_item.inode; 14693cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 14703cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 14713cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 1472707e8a07SDavid Sterba btrfs_set_stack_inode_nbytes(inode_item, root->nodesize); 14733cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 14747237f183SYan Zheng 14755d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 14767237f183SYan Zheng 14777237f183SYan Zheng WARN_ON(root->log_root); 14787237f183SYan Zheng root->log_root = log_root; 14797237f183SYan Zheng root->log_transid = 0; 1480d1433debSMiao Xie root->log_transid_committed = -1; 1481257c62e1SChris Mason root->last_log_commit = 0; 1482e02119d5SChris Mason return 0; 1483e02119d5SChris Mason } 1484e02119d5SChris Mason 148535a3621bSStefan Behrens static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 1486cb517eabSMiao Xie struct btrfs_key *key) 1487e02119d5SChris Mason { 1488e02119d5SChris Mason struct btrfs_root *root; 1489e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 14900f7d52f4SChris Mason struct btrfs_path *path; 149184234f3aSYan Zheng u64 generation; 1492cb517eabSMiao Xie int ret; 1493cb517eabSMiao Xie 1494cb517eabSMiao Xie path = btrfs_alloc_path(); 1495cb517eabSMiao Xie if (!path) 1496cb517eabSMiao Xie return ERR_PTR(-ENOMEM); 14970f7d52f4SChris Mason 14986f07e42eSAl Viro root = btrfs_alloc_root(fs_info); 1499cb517eabSMiao Xie if (!root) { 1500cb517eabSMiao Xie ret = -ENOMEM; 1501cb517eabSMiao Xie goto alloc_fail; 15020f7d52f4SChris Mason } 15030f7d52f4SChris Mason 1504707e8a07SDavid Sterba __setup_root(tree_root->nodesize, tree_root->sectorsize, 1505707e8a07SDavid Sterba tree_root->stripesize, root, fs_info, key->objectid); 15060f7d52f4SChris Mason 1507cb517eabSMiao Xie ret = btrfs_find_root(tree_root, key, path, 1508cb517eabSMiao Xie &root->root_item, &root->root_key); 15090f7d52f4SChris Mason if (ret) { 151013a8a7c8SYan, Zheng if (ret > 0) 151113a8a7c8SYan, Zheng ret = -ENOENT; 1512cb517eabSMiao Xie goto find_fail; 15130f7d52f4SChris Mason } 151413a8a7c8SYan, Zheng 151584234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1516db94535dSChris Mason root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item), 1517ce86cd59SDavid Sterba generation); 1518cb517eabSMiao Xie if (!root->node) { 1519cb517eabSMiao Xie ret = -ENOMEM; 1520cb517eabSMiao Xie goto find_fail; 1521cb517eabSMiao Xie } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1522cb517eabSMiao Xie ret = -EIO; 1523cb517eabSMiao Xie goto read_fail; 1524416bc658SJosef Bacik } 15255d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 152613a8a7c8SYan, Zheng out: 1527cb517eabSMiao Xie btrfs_free_path(path); 1528cb517eabSMiao Xie return root; 1529cb517eabSMiao Xie 1530cb517eabSMiao Xie read_fail: 1531cb517eabSMiao Xie free_extent_buffer(root->node); 1532cb517eabSMiao Xie find_fail: 1533cb517eabSMiao Xie kfree(root); 1534cb517eabSMiao Xie alloc_fail: 1535cb517eabSMiao Xie root = ERR_PTR(ret); 1536cb517eabSMiao Xie goto out; 1537cb517eabSMiao Xie } 1538cb517eabSMiao Xie 1539cb517eabSMiao Xie struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root, 1540cb517eabSMiao Xie struct btrfs_key *location) 1541cb517eabSMiao Xie { 1542cb517eabSMiao Xie struct btrfs_root *root; 1543cb517eabSMiao Xie 1544cb517eabSMiao Xie root = btrfs_read_tree_root(tree_root, location); 1545cb517eabSMiao Xie if (IS_ERR(root)) 1546cb517eabSMiao Xie return root; 1547cb517eabSMiao Xie 1548cb517eabSMiao Xie if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { 154927cdeb70SMiao Xie set_bit(BTRFS_ROOT_REF_COWS, &root->state); 155008fe4db1SLi Zefan btrfs_check_and_init_root_item(&root->root_item); 155108fe4db1SLi Zefan } 155213a8a7c8SYan, Zheng 15535eda7b5eSChris Mason return root; 15545eda7b5eSChris Mason } 15555eda7b5eSChris Mason 1556cb517eabSMiao Xie int btrfs_init_fs_root(struct btrfs_root *root) 1557cb517eabSMiao Xie { 1558cb517eabSMiao Xie int ret; 15598257b2dcSMiao Xie struct btrfs_subvolume_writers *writers; 1560cb517eabSMiao Xie 1561cb517eabSMiao Xie root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); 1562cb517eabSMiao Xie root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), 1563cb517eabSMiao Xie GFP_NOFS); 1564cb517eabSMiao Xie if (!root->free_ino_pinned || !root->free_ino_ctl) { 1565cb517eabSMiao Xie ret = -ENOMEM; 1566cb517eabSMiao Xie goto fail; 1567cb517eabSMiao Xie } 1568cb517eabSMiao Xie 15698257b2dcSMiao Xie writers = btrfs_alloc_subvolume_writers(); 15708257b2dcSMiao Xie if (IS_ERR(writers)) { 15718257b2dcSMiao Xie ret = PTR_ERR(writers); 15728257b2dcSMiao Xie goto fail; 15738257b2dcSMiao Xie } 15748257b2dcSMiao Xie root->subv_writers = writers; 15758257b2dcSMiao Xie 1576cb517eabSMiao Xie btrfs_init_free_ino_ctl(root); 157757cdc8dbSDavid Sterba spin_lock_init(&root->ino_cache_lock); 157857cdc8dbSDavid Sterba init_waitqueue_head(&root->ino_cache_wait); 1579cb517eabSMiao Xie 1580cb517eabSMiao Xie ret = get_anon_bdev(&root->anon_dev); 1581cb517eabSMiao Xie if (ret) 15828257b2dcSMiao Xie goto free_writers; 1583cb517eabSMiao Xie return 0; 15848257b2dcSMiao Xie 15858257b2dcSMiao Xie free_writers: 15868257b2dcSMiao Xie btrfs_free_subvolume_writers(root->subv_writers); 1587cb517eabSMiao Xie fail: 1588cb517eabSMiao Xie kfree(root->free_ino_ctl); 1589cb517eabSMiao Xie kfree(root->free_ino_pinned); 1590cb517eabSMiao Xie return ret; 1591cb517eabSMiao Xie } 1592cb517eabSMiao Xie 1593171170c1SSergei Trofimovich static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1594cb517eabSMiao Xie u64 root_id) 1595cb517eabSMiao Xie { 1596cb517eabSMiao Xie struct btrfs_root *root; 1597cb517eabSMiao Xie 1598cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1599cb517eabSMiao Xie root = radix_tree_lookup(&fs_info->fs_roots_radix, 1600cb517eabSMiao Xie (unsigned long)root_id); 1601cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1602cb517eabSMiao Xie return root; 1603cb517eabSMiao Xie } 1604cb517eabSMiao Xie 1605cb517eabSMiao Xie int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1606cb517eabSMiao Xie struct btrfs_root *root) 1607cb517eabSMiao Xie { 1608cb517eabSMiao Xie int ret; 1609cb517eabSMiao Xie 1610cb517eabSMiao Xie ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); 1611cb517eabSMiao Xie if (ret) 1612cb517eabSMiao Xie return ret; 1613cb517eabSMiao Xie 1614cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1615cb517eabSMiao Xie ret = radix_tree_insert(&fs_info->fs_roots_radix, 1616cb517eabSMiao Xie (unsigned long)root->root_key.objectid, 1617cb517eabSMiao Xie root); 1618cb517eabSMiao Xie if (ret == 0) 161927cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1620cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1621cb517eabSMiao Xie radix_tree_preload_end(); 1622cb517eabSMiao Xie 1623cb517eabSMiao Xie return ret; 1624cb517eabSMiao Xie } 1625cb517eabSMiao Xie 1626c00869f1SMiao Xie struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 1627c00869f1SMiao Xie struct btrfs_key *location, 1628c00869f1SMiao Xie bool check_ref) 16295eda7b5eSChris Mason { 16305eda7b5eSChris Mason struct btrfs_root *root; 1631381cf658SDavid Sterba struct btrfs_path *path; 16321d4c08e0SDavid Sterba struct btrfs_key key; 16335eda7b5eSChris Mason int ret; 16345eda7b5eSChris Mason 1635edbd8d4eSChris Mason if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 1636edbd8d4eSChris Mason return fs_info->tree_root; 1637edbd8d4eSChris Mason if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID) 1638edbd8d4eSChris Mason return fs_info->extent_root; 16398f18cf13SChris Mason if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID) 16408f18cf13SChris Mason return fs_info->chunk_root; 16418f18cf13SChris Mason if (location->objectid == BTRFS_DEV_TREE_OBJECTID) 16428f18cf13SChris Mason return fs_info->dev_root; 16430403e47eSYan Zheng if (location->objectid == BTRFS_CSUM_TREE_OBJECTID) 16440403e47eSYan Zheng return fs_info->csum_root; 1645bcef60f2SArne Jansen if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID) 1646bcef60f2SArne Jansen return fs_info->quota_root ? fs_info->quota_root : 1647bcef60f2SArne Jansen ERR_PTR(-ENOENT); 1648f7a81ea4SStefan Behrens if (location->objectid == BTRFS_UUID_TREE_OBJECTID) 1649f7a81ea4SStefan Behrens return fs_info->uuid_root ? fs_info->uuid_root : 1650f7a81ea4SStefan Behrens ERR_PTR(-ENOENT); 16514df27c4dSYan, Zheng again: 1652cb517eabSMiao Xie root = btrfs_lookup_fs_root(fs_info, location->objectid); 165348475471SStefan Behrens if (root) { 1654c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) 165548475471SStefan Behrens return ERR_PTR(-ENOENT); 16565eda7b5eSChris Mason return root; 165748475471SStefan Behrens } 16585eda7b5eSChris Mason 1659cb517eabSMiao Xie root = btrfs_read_fs_root(fs_info->tree_root, location); 16605eda7b5eSChris Mason if (IS_ERR(root)) 16615eda7b5eSChris Mason return root; 16623394e160SChris Mason 1663c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1664d68fc57bSYan, Zheng ret = -ENOENT; 1665d68fc57bSYan, Zheng goto fail; 1666d68fc57bSYan, Zheng } 1667d68fc57bSYan, Zheng 1668cb517eabSMiao Xie ret = btrfs_init_fs_root(root); 1669cb517eabSMiao Xie if (ret) 1670cb517eabSMiao Xie goto fail; 1671cb517eabSMiao Xie 1672381cf658SDavid Sterba path = btrfs_alloc_path(); 1673381cf658SDavid Sterba if (!path) { 1674381cf658SDavid Sterba ret = -ENOMEM; 1675381cf658SDavid Sterba goto fail; 1676381cf658SDavid Sterba } 16771d4c08e0SDavid Sterba key.objectid = BTRFS_ORPHAN_OBJECTID; 16781d4c08e0SDavid Sterba key.type = BTRFS_ORPHAN_ITEM_KEY; 16791d4c08e0SDavid Sterba key.offset = location->objectid; 16801d4c08e0SDavid Sterba 16811d4c08e0SDavid Sterba ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1682381cf658SDavid Sterba btrfs_free_path(path); 1683d68fc57bSYan, Zheng if (ret < 0) 1684d68fc57bSYan, Zheng goto fail; 1685d68fc57bSYan, Zheng if (ret == 0) 168627cdeb70SMiao Xie set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1687d68fc57bSYan, Zheng 1688cb517eabSMiao Xie ret = btrfs_insert_fs_root(fs_info, root); 16890f7d52f4SChris Mason if (ret) { 16904df27c4dSYan, Zheng if (ret == -EEXIST) { 16914df27c4dSYan, Zheng free_fs_root(root); 16924df27c4dSYan, Zheng goto again; 16930f7d52f4SChris Mason } 16944df27c4dSYan, Zheng goto fail; 16954df27c4dSYan, Zheng } 1696edbd8d4eSChris Mason return root; 16974df27c4dSYan, Zheng fail: 16984df27c4dSYan, Zheng free_fs_root(root); 16994df27c4dSYan, Zheng return ERR_PTR(ret); 1700edbd8d4eSChris Mason } 1701edbd8d4eSChris Mason 170204160088SChris Mason static int btrfs_congested_fn(void *congested_data, int bdi_bits) 170304160088SChris Mason { 170404160088SChris Mason struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; 170504160088SChris Mason int ret = 0; 170604160088SChris Mason struct btrfs_device *device; 170704160088SChris Mason struct backing_dev_info *bdi; 1708b7967db7SChris Mason 17091f78160cSXiao Guangrong rcu_read_lock(); 17101f78160cSXiao Guangrong list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) { 1711dfe25020SChris Mason if (!device->bdev) 1712dfe25020SChris Mason continue; 171304160088SChris Mason bdi = blk_get_backing_dev_info(device->bdev); 1714ff9ea323STejun Heo if (bdi_congested(bdi, bdi_bits)) { 171504160088SChris Mason ret = 1; 171604160088SChris Mason break; 171704160088SChris Mason } 171804160088SChris Mason } 17191f78160cSXiao Guangrong rcu_read_unlock(); 172004160088SChris Mason return ret; 172104160088SChris Mason } 172204160088SChris Mason 172304160088SChris Mason static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi) 172404160088SChris Mason { 1725ad081f14SJens Axboe int err; 1726ad081f14SJens Axboe 1727ad081f14SJens Axboe bdi->capabilities = BDI_CAP_MAP_COPY; 1728e6d086d8SJens Axboe err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY); 1729ad081f14SJens Axboe if (err) 1730ad081f14SJens Axboe return err; 1731ad081f14SJens Axboe 17324575c9ccSChris Mason bdi->ra_pages = default_backing_dev_info.ra_pages; 173304160088SChris Mason bdi->congested_fn = btrfs_congested_fn; 173404160088SChris Mason bdi->congested_data = info; 173504160088SChris Mason return 0; 173604160088SChris Mason } 173704160088SChris Mason 17388b712842SChris Mason /* 17398b712842SChris Mason * called by the kthread helper functions to finally call the bio end_io 17408b712842SChris Mason * functions. This is where read checksum verification actually happens 17418b712842SChris Mason */ 17428b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work) 1743ce9adaa5SChris Mason { 1744ce9adaa5SChris Mason struct bio *bio; 174597eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 1746ce9adaa5SChris Mason int error; 1747ce9adaa5SChris Mason 174897eb6b69SDavid Sterba end_io_wq = container_of(work, struct btrfs_end_io_wq, work); 1749ce9adaa5SChris Mason bio = end_io_wq->bio; 17508b712842SChris Mason 1751ce9adaa5SChris Mason error = end_io_wq->error; 1752ce9adaa5SChris Mason bio->bi_private = end_io_wq->private; 1753ce9adaa5SChris Mason bio->bi_end_io = end_io_wq->end_io; 175497eb6b69SDavid Sterba kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq); 1755bc1e79acSKent Overstreet bio_endio_nodec(bio, error); 1756ce9adaa5SChris Mason } 175744b8bd7eSChris Mason 1758a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1759a74a4b97SChris Mason { 1760a74a4b97SChris Mason struct btrfs_root *root = arg; 1761d0278245SMiao Xie int again; 1762a74a4b97SChris Mason 1763a74a4b97SChris Mason do { 1764d0278245SMiao Xie again = 0; 17659d1a2a3aSDavid Sterba 1766d0278245SMiao Xie /* Make the cleaner go to sleep early. */ 1767babbf170SMiao Xie if (btrfs_need_cleaner_sleep(root)) 1768d0278245SMiao Xie goto sleep; 1769d0278245SMiao Xie 1770d0278245SMiao Xie if (!mutex_trylock(&root->fs_info->cleaner_mutex)) 1771d0278245SMiao Xie goto sleep; 1772d0278245SMiao Xie 1773dc7f370cSMiao Xie /* 1774dc7f370cSMiao Xie * Avoid the problem that we change the status of the fs 1775dc7f370cSMiao Xie * during the above check and trylock. 1776dc7f370cSMiao Xie */ 1777babbf170SMiao Xie if (btrfs_need_cleaner_sleep(root)) { 1778dc7f370cSMiao Xie mutex_unlock(&root->fs_info->cleaner_mutex); 1779dc7f370cSMiao Xie goto sleep; 1780dc7f370cSMiao Xie } 1781dc7f370cSMiao Xie 178224bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 178347ab2a6cSJosef Bacik btrfs_delete_unused_bgs(root->fs_info); 17849d1a2a3aSDavid Sterba again = btrfs_clean_one_deleted_snapshot(root); 1785a74a4b97SChris Mason mutex_unlock(&root->fs_info->cleaner_mutex); 1786a74a4b97SChris Mason 1787d0278245SMiao Xie /* 178805323cd1SMiao Xie * The defragger has dealt with the R/O remount and umount, 178905323cd1SMiao Xie * needn't do anything special here. 1790d0278245SMiao Xie */ 1791d0278245SMiao Xie btrfs_run_defrag_inodes(root->fs_info); 1792d0278245SMiao Xie sleep: 17939d1a2a3aSDavid Sterba if (!try_to_freeze() && !again) { 1794a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 17958929ecfaSYan, Zheng if (!kthread_should_stop()) 1796a74a4b97SChris Mason schedule(); 1797a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1798a74a4b97SChris Mason } 1799a74a4b97SChris Mason } while (!kthread_should_stop()); 1800a74a4b97SChris Mason return 0; 1801a74a4b97SChris Mason } 1802a74a4b97SChris Mason 1803a74a4b97SChris Mason static int transaction_kthread(void *arg) 1804a74a4b97SChris Mason { 1805a74a4b97SChris Mason struct btrfs_root *root = arg; 1806a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1807a74a4b97SChris Mason struct btrfs_transaction *cur; 18088929ecfaSYan, Zheng u64 transid; 1809a74a4b97SChris Mason unsigned long now; 1810a74a4b97SChris Mason unsigned long delay; 1811914b2007SJan Kara bool cannot_commit; 1812a74a4b97SChris Mason 1813a74a4b97SChris Mason do { 1814914b2007SJan Kara cannot_commit = false; 18158b87dc17SDavid Sterba delay = HZ * root->fs_info->commit_interval; 1816a74a4b97SChris Mason mutex_lock(&root->fs_info->transaction_kthread_mutex); 1817a74a4b97SChris Mason 1818a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 1819a74a4b97SChris Mason cur = root->fs_info->running_transaction; 1820a74a4b97SChris Mason if (!cur) { 1821a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1822a74a4b97SChris Mason goto sleep; 1823a74a4b97SChris Mason } 182431153d81SYan Zheng 1825a74a4b97SChris Mason now = get_seconds(); 18264a9d8bdeSMiao Xie if (cur->state < TRANS_STATE_BLOCKED && 18278b87dc17SDavid Sterba (now < cur->start_time || 18288b87dc17SDavid Sterba now - cur->start_time < root->fs_info->commit_interval)) { 1829a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1830a74a4b97SChris Mason delay = HZ * 5; 1831a74a4b97SChris Mason goto sleep; 1832a74a4b97SChris Mason } 18338929ecfaSYan, Zheng transid = cur->transid; 1834a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 183556bec294SChris Mason 183679787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 1837354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 1838914b2007SJan Kara if (IS_ERR(trans)) { 1839354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 1840914b2007SJan Kara cannot_commit = true; 184179787eaaSJeff Mahoney goto sleep; 1842914b2007SJan Kara } 18438929ecfaSYan, Zheng if (transid == trans->transid) { 184479787eaaSJeff Mahoney btrfs_commit_transaction(trans, root); 18458929ecfaSYan, Zheng } else { 18468929ecfaSYan, Zheng btrfs_end_transaction(trans, root); 18478929ecfaSYan, Zheng } 1848a74a4b97SChris Mason sleep: 1849a74a4b97SChris Mason wake_up_process(root->fs_info->cleaner_kthread); 1850a74a4b97SChris Mason mutex_unlock(&root->fs_info->transaction_kthread_mutex); 1851a74a4b97SChris Mason 18524e121c06SJosef Bacik if (unlikely(test_bit(BTRFS_FS_STATE_ERROR, 18534e121c06SJosef Bacik &root->fs_info->fs_state))) 18544e121c06SJosef Bacik btrfs_cleanup_transaction(root); 1855a0acae0eSTejun Heo if (!try_to_freeze()) { 1856a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 18578929ecfaSYan, Zheng if (!kthread_should_stop() && 1858914b2007SJan Kara (!btrfs_transaction_blocked(root->fs_info) || 1859914b2007SJan Kara cannot_commit)) 1860a74a4b97SChris Mason schedule_timeout(delay); 1861a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1862a74a4b97SChris Mason } 1863a74a4b97SChris Mason } while (!kthread_should_stop()); 1864a74a4b97SChris Mason return 0; 1865a74a4b97SChris Mason } 1866a74a4b97SChris Mason 1867af31f5e5SChris Mason /* 1868af31f5e5SChris Mason * this will find the highest generation in the array of 1869af31f5e5SChris Mason * root backups. The index of the highest array is returned, 1870af31f5e5SChris Mason * or -1 if we can't find anything. 1871af31f5e5SChris Mason * 1872af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1873af31f5e5SChris Mason * generation of the latest root in the array with the generation 1874af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1875af31f5e5SChris Mason */ 1876af31f5e5SChris Mason static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen) 1877af31f5e5SChris Mason { 1878af31f5e5SChris Mason u64 cur; 1879af31f5e5SChris Mason int newest_index = -1; 1880af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1881af31f5e5SChris Mason int i; 1882af31f5e5SChris Mason 1883af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1884af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1885af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1886af31f5e5SChris Mason if (cur == newest_gen) 1887af31f5e5SChris Mason newest_index = i; 1888af31f5e5SChris Mason } 1889af31f5e5SChris Mason 1890af31f5e5SChris Mason /* check to see if we actually wrapped around */ 1891af31f5e5SChris Mason if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) { 1892af31f5e5SChris Mason root_backup = info->super_copy->super_roots; 1893af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1894af31f5e5SChris Mason if (cur == newest_gen) 1895af31f5e5SChris Mason newest_index = 0; 1896af31f5e5SChris Mason } 1897af31f5e5SChris Mason return newest_index; 1898af31f5e5SChris Mason } 1899af31f5e5SChris Mason 1900af31f5e5SChris Mason 1901af31f5e5SChris Mason /* 1902af31f5e5SChris Mason * find the oldest backup so we know where to store new entries 1903af31f5e5SChris Mason * in the backup array. This will set the backup_root_index 1904af31f5e5SChris Mason * field in the fs_info struct 1905af31f5e5SChris Mason */ 1906af31f5e5SChris Mason static void find_oldest_super_backup(struct btrfs_fs_info *info, 1907af31f5e5SChris Mason u64 newest_gen) 1908af31f5e5SChris Mason { 1909af31f5e5SChris Mason int newest_index = -1; 1910af31f5e5SChris Mason 1911af31f5e5SChris Mason newest_index = find_newest_super_backup(info, newest_gen); 1912af31f5e5SChris Mason /* if there was garbage in there, just move along */ 1913af31f5e5SChris Mason if (newest_index == -1) { 1914af31f5e5SChris Mason info->backup_root_index = 0; 1915af31f5e5SChris Mason } else { 1916af31f5e5SChris Mason info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS; 1917af31f5e5SChris Mason } 1918af31f5e5SChris Mason } 1919af31f5e5SChris Mason 1920af31f5e5SChris Mason /* 1921af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1922af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1923af31f5e5SChris Mason * done 1924af31f5e5SChris Mason */ 1925af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1926af31f5e5SChris Mason { 1927af31f5e5SChris Mason int next_backup; 1928af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1929af31f5e5SChris Mason int last_backup; 1930af31f5e5SChris Mason 1931af31f5e5SChris Mason next_backup = info->backup_root_index; 1932af31f5e5SChris Mason last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) % 1933af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 1934af31f5e5SChris Mason 1935af31f5e5SChris Mason /* 1936af31f5e5SChris Mason * just overwrite the last backup if we're at the same generation 1937af31f5e5SChris Mason * this happens only at umount 1938af31f5e5SChris Mason */ 1939af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + last_backup; 1940af31f5e5SChris Mason if (btrfs_backup_tree_root_gen(root_backup) == 1941af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)) 1942af31f5e5SChris Mason next_backup = last_backup; 1943af31f5e5SChris Mason 1944af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1945af31f5e5SChris Mason 1946af31f5e5SChris Mason /* 1947af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1948af31f5e5SChris Mason * regardless of which ones we use today 1949af31f5e5SChris Mason */ 1950af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1951af31f5e5SChris Mason 1952af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1953af31f5e5SChris Mason 1954af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1955af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1956af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1957af31f5e5SChris Mason 1958af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1959af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1960af31f5e5SChris Mason 1961af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1962af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1963af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 1964af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 1965af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 1966af31f5e5SChris Mason 1967af31f5e5SChris Mason btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); 1968af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 1969af31f5e5SChris Mason btrfs_header_generation(info->extent_root->node)); 1970af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 1971af31f5e5SChris Mason btrfs_header_level(info->extent_root->node)); 1972af31f5e5SChris Mason 19737c7e82a7SChris Mason /* 19747c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 19757c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 19767c7e82a7SChris Mason */ 19777c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 19787c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 19797c7e82a7SChris Mason info->fs_root->node->start); 1980af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 1981af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 1982af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 1983af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 19847c7e82a7SChris Mason } 1985af31f5e5SChris Mason 1986af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 1987af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 1988af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 1989af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 1990af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 1991af31f5e5SChris Mason 1992af31f5e5SChris Mason btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); 1993af31f5e5SChris Mason btrfs_set_backup_csum_root_gen(root_backup, 1994af31f5e5SChris Mason btrfs_header_generation(info->csum_root->node)); 1995af31f5e5SChris Mason btrfs_set_backup_csum_root_level(root_backup, 1996af31f5e5SChris Mason btrfs_header_level(info->csum_root->node)); 1997af31f5e5SChris Mason 1998af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 1999af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 2000af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 2001af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 2002af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 2003af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 2004af31f5e5SChris Mason 2005af31f5e5SChris Mason /* 2006af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 2007af31f5e5SChris Mason * for the next commit 2008af31f5e5SChris Mason */ 2009af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 2010af31f5e5SChris Mason &info->super_for_commit->super_roots, 2011af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 2012af31f5e5SChris Mason } 2013af31f5e5SChris Mason 2014af31f5e5SChris Mason /* 2015af31f5e5SChris Mason * this copies info out of the root backup array and back into 2016af31f5e5SChris Mason * the in-memory super block. It is meant to help iterate through 2017af31f5e5SChris Mason * the array, so you send it the number of backups you've already 2018af31f5e5SChris Mason * tried and the last backup index you used. 2019af31f5e5SChris Mason * 2020af31f5e5SChris Mason * this returns -1 when it has tried all the backups 2021af31f5e5SChris Mason */ 2022af31f5e5SChris Mason static noinline int next_root_backup(struct btrfs_fs_info *info, 2023af31f5e5SChris Mason struct btrfs_super_block *super, 2024af31f5e5SChris Mason int *num_backups_tried, int *backup_index) 2025af31f5e5SChris Mason { 2026af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 2027af31f5e5SChris Mason int newest = *backup_index; 2028af31f5e5SChris Mason 2029af31f5e5SChris Mason if (*num_backups_tried == 0) { 2030af31f5e5SChris Mason u64 gen = btrfs_super_generation(super); 2031af31f5e5SChris Mason 2032af31f5e5SChris Mason newest = find_newest_super_backup(info, gen); 2033af31f5e5SChris Mason if (newest == -1) 2034af31f5e5SChris Mason return -1; 2035af31f5e5SChris Mason 2036af31f5e5SChris Mason *backup_index = newest; 2037af31f5e5SChris Mason *num_backups_tried = 1; 2038af31f5e5SChris Mason } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) { 2039af31f5e5SChris Mason /* we've tried all the backups, all done */ 2040af31f5e5SChris Mason return -1; 2041af31f5e5SChris Mason } else { 2042af31f5e5SChris Mason /* jump to the next oldest backup */ 2043af31f5e5SChris Mason newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) % 2044af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 2045af31f5e5SChris Mason *backup_index = newest; 2046af31f5e5SChris Mason *num_backups_tried += 1; 2047af31f5e5SChris Mason } 2048af31f5e5SChris Mason root_backup = super->super_roots + newest; 2049af31f5e5SChris Mason 2050af31f5e5SChris Mason btrfs_set_super_generation(super, 2051af31f5e5SChris Mason btrfs_backup_tree_root_gen(root_backup)); 2052af31f5e5SChris Mason btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 2053af31f5e5SChris Mason btrfs_set_super_root_level(super, 2054af31f5e5SChris Mason btrfs_backup_tree_root_level(root_backup)); 2055af31f5e5SChris Mason btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 2056af31f5e5SChris Mason 2057af31f5e5SChris Mason /* 2058af31f5e5SChris Mason * fixme: the total bytes and num_devices need to match or we should 2059af31f5e5SChris Mason * need a fsck 2060af31f5e5SChris Mason */ 2061af31f5e5SChris Mason btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 2062af31f5e5SChris Mason btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 2063af31f5e5SChris Mason return 0; 2064af31f5e5SChris Mason } 2065af31f5e5SChris Mason 20667abadb64SLiu Bo /* helper to cleanup workers */ 20677abadb64SLiu Bo static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 20687abadb64SLiu Bo { 2069dc6e3209SQu Wenruo btrfs_destroy_workqueue(fs_info->fixup_workers); 2070afe3d242SQu Wenruo btrfs_destroy_workqueue(fs_info->delalloc_workers); 20715cdc7ad3SQu Wenruo btrfs_destroy_workqueue(fs_info->workers); 2072fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_workers); 2073fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_meta_workers); 2074fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_raid56_workers); 20758b110e39SMiao Xie btrfs_destroy_workqueue(fs_info->endio_repair_workers); 2076d05a33acSQu Wenruo btrfs_destroy_workqueue(fs_info->rmw_workers); 2077fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_meta_write_workers); 2078fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_write_workers); 2079fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 2080a8c93d4eSQu Wenruo btrfs_destroy_workqueue(fs_info->submit_workers); 20815b3bc44eSQu Wenruo btrfs_destroy_workqueue(fs_info->delayed_workers); 2082e66f0bb1SQu Wenruo btrfs_destroy_workqueue(fs_info->caching_workers); 2083736cfa15SQu Wenruo btrfs_destroy_workqueue(fs_info->readahead_workers); 2084a44903abSQu Wenruo btrfs_destroy_workqueue(fs_info->flush_workers); 2085fc97fab0SQu Wenruo btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2086a79b7d4bSChris Mason btrfs_destroy_workqueue(fs_info->extent_workers); 20877abadb64SLiu Bo } 20887abadb64SLiu Bo 20892e9f5954SRashika static void free_root_extent_buffers(struct btrfs_root *root) 20902e9f5954SRashika { 20912e9f5954SRashika if (root) { 20922e9f5954SRashika free_extent_buffer(root->node); 20932e9f5954SRashika free_extent_buffer(root->commit_root); 20942e9f5954SRashika root->node = NULL; 20952e9f5954SRashika root->commit_root = NULL; 20962e9f5954SRashika } 20972e9f5954SRashika } 20982e9f5954SRashika 2099af31f5e5SChris Mason /* helper to cleanup tree roots */ 2100af31f5e5SChris Mason static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root) 2101af31f5e5SChris Mason { 21022e9f5954SRashika free_root_extent_buffers(info->tree_root); 2103655b09feSJosef Bacik 21042e9f5954SRashika free_root_extent_buffers(info->dev_root); 21052e9f5954SRashika free_root_extent_buffers(info->extent_root); 21062e9f5954SRashika free_root_extent_buffers(info->csum_root); 21072e9f5954SRashika free_root_extent_buffers(info->quota_root); 21082e9f5954SRashika free_root_extent_buffers(info->uuid_root); 21092e9f5954SRashika if (chunk_root) 21102e9f5954SRashika free_root_extent_buffers(info->chunk_root); 2111af31f5e5SChris Mason } 2112af31f5e5SChris Mason 2113faa2dbf0SJosef Bacik void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2114171f6537SJosef Bacik { 2115171f6537SJosef Bacik int ret; 2116171f6537SJosef Bacik struct btrfs_root *gang[8]; 2117171f6537SJosef Bacik int i; 2118171f6537SJosef Bacik 2119171f6537SJosef Bacik while (!list_empty(&fs_info->dead_roots)) { 2120171f6537SJosef Bacik gang[0] = list_entry(fs_info->dead_roots.next, 2121171f6537SJosef Bacik struct btrfs_root, root_list); 2122171f6537SJosef Bacik list_del(&gang[0]->root_list); 2123171f6537SJosef Bacik 212427cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) { 2125cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[0]); 2126171f6537SJosef Bacik } else { 2127171f6537SJosef Bacik free_extent_buffer(gang[0]->node); 2128171f6537SJosef Bacik free_extent_buffer(gang[0]->commit_root); 2129b0feb9d9SMiao Xie btrfs_put_fs_root(gang[0]); 2130171f6537SJosef Bacik } 2131171f6537SJosef Bacik } 2132171f6537SJosef Bacik 2133171f6537SJosef Bacik while (1) { 2134171f6537SJosef Bacik ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2135171f6537SJosef Bacik (void **)gang, 0, 2136171f6537SJosef Bacik ARRAY_SIZE(gang)); 2137171f6537SJosef Bacik if (!ret) 2138171f6537SJosef Bacik break; 2139171f6537SJosef Bacik for (i = 0; i < ret; i++) 2140cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2141171f6537SJosef Bacik } 21421a4319ccSLiu Bo 21431a4319ccSLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 21441a4319ccSLiu Bo btrfs_free_log_root_tree(NULL, fs_info); 21451a4319ccSLiu Bo btrfs_destroy_pinned_extent(fs_info->tree_root, 21461a4319ccSLiu Bo fs_info->pinned_extents); 21471a4319ccSLiu Bo } 2148171f6537SJosef Bacik } 2149af31f5e5SChris Mason 2150638aa7edSEric Sandeen static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2151638aa7edSEric Sandeen { 2152638aa7edSEric Sandeen mutex_init(&fs_info->scrub_lock); 2153638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_running, 0); 2154638aa7edSEric Sandeen atomic_set(&fs_info->scrub_pause_req, 0); 2155638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_paused, 0); 2156638aa7edSEric Sandeen atomic_set(&fs_info->scrub_cancel_req, 0); 2157638aa7edSEric Sandeen init_waitqueue_head(&fs_info->scrub_pause_wait); 2158638aa7edSEric Sandeen fs_info->scrub_workers_refcnt = 0; 2159638aa7edSEric Sandeen } 2160638aa7edSEric Sandeen 2161779a65a4SEric Sandeen static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2162779a65a4SEric Sandeen { 2163779a65a4SEric Sandeen spin_lock_init(&fs_info->balance_lock); 2164779a65a4SEric Sandeen mutex_init(&fs_info->balance_mutex); 2165779a65a4SEric Sandeen atomic_set(&fs_info->balance_running, 0); 2166779a65a4SEric Sandeen atomic_set(&fs_info->balance_pause_req, 0); 2167779a65a4SEric Sandeen atomic_set(&fs_info->balance_cancel_req, 0); 2168779a65a4SEric Sandeen fs_info->balance_ctl = NULL; 2169779a65a4SEric Sandeen init_waitqueue_head(&fs_info->balance_wait_q); 2170779a65a4SEric Sandeen } 2171779a65a4SEric Sandeen 2172f37938e0SEric Sandeen static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info, 2173f37938e0SEric Sandeen struct btrfs_root *tree_root) 2174f37938e0SEric Sandeen { 2175f37938e0SEric Sandeen fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 2176f37938e0SEric Sandeen set_nlink(fs_info->btree_inode, 1); 2177f37938e0SEric Sandeen /* 2178f37938e0SEric Sandeen * we set the i_size on the btree inode to the max possible int. 2179f37938e0SEric Sandeen * the real end of the address space is determined by all of 2180f37938e0SEric Sandeen * the devices in the system 2181f37938e0SEric Sandeen */ 2182f37938e0SEric Sandeen fs_info->btree_inode->i_size = OFFSET_MAX; 2183f37938e0SEric Sandeen fs_info->btree_inode->i_mapping->a_ops = &btree_aops; 2184f37938e0SEric Sandeen fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi; 2185f37938e0SEric Sandeen 2186f37938e0SEric Sandeen RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node); 2187f37938e0SEric Sandeen extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree, 2188f37938e0SEric Sandeen fs_info->btree_inode->i_mapping); 2189f37938e0SEric Sandeen BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0; 2190f37938e0SEric Sandeen extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree); 2191f37938e0SEric Sandeen 2192f37938e0SEric Sandeen BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops; 2193f37938e0SEric Sandeen 2194f37938e0SEric Sandeen BTRFS_I(fs_info->btree_inode)->root = tree_root; 2195f37938e0SEric Sandeen memset(&BTRFS_I(fs_info->btree_inode)->location, 0, 2196f37938e0SEric Sandeen sizeof(struct btrfs_key)); 2197f37938e0SEric Sandeen set_bit(BTRFS_INODE_DUMMY, 2198f37938e0SEric Sandeen &BTRFS_I(fs_info->btree_inode)->runtime_flags); 2199f37938e0SEric Sandeen btrfs_insert_inode_hash(fs_info->btree_inode); 2200f37938e0SEric Sandeen } 2201f37938e0SEric Sandeen 2202ad618368SEric Sandeen static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2203ad618368SEric Sandeen { 2204ad618368SEric Sandeen fs_info->dev_replace.lock_owner = 0; 2205ad618368SEric Sandeen atomic_set(&fs_info->dev_replace.nesting_level, 0); 2206ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2207ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_management_lock); 2208ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock); 2209ad618368SEric Sandeen init_waitqueue_head(&fs_info->replace_wait); 2210ad618368SEric Sandeen } 2211ad618368SEric Sandeen 2212f9e92e40SEric Sandeen static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2213f9e92e40SEric Sandeen { 2214f9e92e40SEric Sandeen spin_lock_init(&fs_info->qgroup_lock); 2215f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_ioctl_lock); 2216f9e92e40SEric Sandeen fs_info->qgroup_tree = RB_ROOT; 2217f9e92e40SEric Sandeen fs_info->qgroup_op_tree = RB_ROOT; 2218f9e92e40SEric Sandeen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2219f9e92e40SEric Sandeen fs_info->qgroup_seq = 1; 2220f9e92e40SEric Sandeen fs_info->quota_enabled = 0; 2221f9e92e40SEric Sandeen fs_info->pending_quota_state = 0; 2222f9e92e40SEric Sandeen fs_info->qgroup_ulist = NULL; 2223f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_rescan_lock); 2224f9e92e40SEric Sandeen } 2225f9e92e40SEric Sandeen 22262a458198SEric Sandeen static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info, 22272a458198SEric Sandeen struct btrfs_fs_devices *fs_devices) 22282a458198SEric Sandeen { 22292a458198SEric Sandeen int max_active = fs_info->thread_pool_size; 22306f011058SDavid Sterba unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 22312a458198SEric Sandeen 22322a458198SEric Sandeen fs_info->workers = 22332a458198SEric Sandeen btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI, 22342a458198SEric Sandeen max_active, 16); 22352a458198SEric Sandeen 22362a458198SEric Sandeen fs_info->delalloc_workers = 22372a458198SEric Sandeen btrfs_alloc_workqueue("delalloc", flags, max_active, 2); 22382a458198SEric Sandeen 22392a458198SEric Sandeen fs_info->flush_workers = 22402a458198SEric Sandeen btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0); 22412a458198SEric Sandeen 22422a458198SEric Sandeen fs_info->caching_workers = 22432a458198SEric Sandeen btrfs_alloc_workqueue("cache", flags, max_active, 0); 22442a458198SEric Sandeen 22452a458198SEric Sandeen /* 22462a458198SEric Sandeen * a higher idle thresh on the submit workers makes it much more 22472a458198SEric Sandeen * likely that bios will be send down in a sane order to the 22482a458198SEric Sandeen * devices 22492a458198SEric Sandeen */ 22502a458198SEric Sandeen fs_info->submit_workers = 22512a458198SEric Sandeen btrfs_alloc_workqueue("submit", flags, 22522a458198SEric Sandeen min_t(u64, fs_devices->num_devices, 22532a458198SEric Sandeen max_active), 64); 22542a458198SEric Sandeen 22552a458198SEric Sandeen fs_info->fixup_workers = 22562a458198SEric Sandeen btrfs_alloc_workqueue("fixup", flags, 1, 0); 22572a458198SEric Sandeen 22582a458198SEric Sandeen /* 22592a458198SEric Sandeen * endios are largely parallel and should have a very 22602a458198SEric Sandeen * low idle thresh 22612a458198SEric Sandeen */ 22622a458198SEric Sandeen fs_info->endio_workers = 22632a458198SEric Sandeen btrfs_alloc_workqueue("endio", flags, max_active, 4); 22642a458198SEric Sandeen fs_info->endio_meta_workers = 22652a458198SEric Sandeen btrfs_alloc_workqueue("endio-meta", flags, max_active, 4); 22662a458198SEric Sandeen fs_info->endio_meta_write_workers = 22672a458198SEric Sandeen btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2); 22682a458198SEric Sandeen fs_info->endio_raid56_workers = 22692a458198SEric Sandeen btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4); 22702a458198SEric Sandeen fs_info->endio_repair_workers = 22712a458198SEric Sandeen btrfs_alloc_workqueue("endio-repair", flags, 1, 0); 22722a458198SEric Sandeen fs_info->rmw_workers = 22732a458198SEric Sandeen btrfs_alloc_workqueue("rmw", flags, max_active, 2); 22742a458198SEric Sandeen fs_info->endio_write_workers = 22752a458198SEric Sandeen btrfs_alloc_workqueue("endio-write", flags, max_active, 2); 22762a458198SEric Sandeen fs_info->endio_freespace_worker = 22772a458198SEric Sandeen btrfs_alloc_workqueue("freespace-write", flags, max_active, 0); 22782a458198SEric Sandeen fs_info->delayed_workers = 22792a458198SEric Sandeen btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0); 22802a458198SEric Sandeen fs_info->readahead_workers = 22812a458198SEric Sandeen btrfs_alloc_workqueue("readahead", flags, max_active, 2); 22822a458198SEric Sandeen fs_info->qgroup_rescan_workers = 22832a458198SEric Sandeen btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0); 22842a458198SEric Sandeen fs_info->extent_workers = 22852a458198SEric Sandeen btrfs_alloc_workqueue("extent-refs", flags, 22862a458198SEric Sandeen min_t(u64, fs_devices->num_devices, 22872a458198SEric Sandeen max_active), 8); 22882a458198SEric Sandeen 22892a458198SEric Sandeen if (!(fs_info->workers && fs_info->delalloc_workers && 22902a458198SEric Sandeen fs_info->submit_workers && fs_info->flush_workers && 22912a458198SEric Sandeen fs_info->endio_workers && fs_info->endio_meta_workers && 22922a458198SEric Sandeen fs_info->endio_meta_write_workers && 22932a458198SEric Sandeen fs_info->endio_repair_workers && 22942a458198SEric Sandeen fs_info->endio_write_workers && fs_info->endio_raid56_workers && 22952a458198SEric Sandeen fs_info->endio_freespace_worker && fs_info->rmw_workers && 22962a458198SEric Sandeen fs_info->caching_workers && fs_info->readahead_workers && 22972a458198SEric Sandeen fs_info->fixup_workers && fs_info->delayed_workers && 22982a458198SEric Sandeen fs_info->extent_workers && 22992a458198SEric Sandeen fs_info->qgroup_rescan_workers)) { 23002a458198SEric Sandeen return -ENOMEM; 23012a458198SEric Sandeen } 23022a458198SEric Sandeen 23032a458198SEric Sandeen return 0; 23042a458198SEric Sandeen } 23052a458198SEric Sandeen 230663443bf5SEric Sandeen static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 230763443bf5SEric Sandeen struct btrfs_fs_devices *fs_devices) 230863443bf5SEric Sandeen { 230963443bf5SEric Sandeen int ret; 231063443bf5SEric Sandeen struct btrfs_root *tree_root = fs_info->tree_root; 231163443bf5SEric Sandeen struct btrfs_root *log_tree_root; 231263443bf5SEric Sandeen struct btrfs_super_block *disk_super = fs_info->super_copy; 231363443bf5SEric Sandeen u64 bytenr = btrfs_super_log_root(disk_super); 231463443bf5SEric Sandeen 231563443bf5SEric Sandeen if (fs_devices->rw_devices == 0) { 231663443bf5SEric Sandeen printk(KERN_WARNING "BTRFS: log replay required " 231763443bf5SEric Sandeen "on RO media\n"); 231863443bf5SEric Sandeen return -EIO; 231963443bf5SEric Sandeen } 232063443bf5SEric Sandeen 232163443bf5SEric Sandeen log_tree_root = btrfs_alloc_root(fs_info); 232263443bf5SEric Sandeen if (!log_tree_root) 232363443bf5SEric Sandeen return -ENOMEM; 232463443bf5SEric Sandeen 232563443bf5SEric Sandeen __setup_root(tree_root->nodesize, tree_root->sectorsize, 232663443bf5SEric Sandeen tree_root->stripesize, log_tree_root, fs_info, 232763443bf5SEric Sandeen BTRFS_TREE_LOG_OBJECTID); 232863443bf5SEric Sandeen 232963443bf5SEric Sandeen log_tree_root->node = read_tree_block(tree_root, bytenr, 233063443bf5SEric Sandeen fs_info->generation + 1); 233163443bf5SEric Sandeen if (!log_tree_root->node || 233263443bf5SEric Sandeen !extent_buffer_uptodate(log_tree_root->node)) { 233363443bf5SEric Sandeen printk(KERN_ERR "BTRFS: failed to read log tree\n"); 233463443bf5SEric Sandeen free_extent_buffer(log_tree_root->node); 233563443bf5SEric Sandeen kfree(log_tree_root); 233663443bf5SEric Sandeen return -EIO; 233763443bf5SEric Sandeen } 233863443bf5SEric Sandeen /* returns with log_tree_root freed on success */ 233963443bf5SEric Sandeen ret = btrfs_recover_log_trees(log_tree_root); 234063443bf5SEric Sandeen if (ret) { 234163443bf5SEric Sandeen btrfs_error(tree_root->fs_info, ret, 234263443bf5SEric Sandeen "Failed to recover log tree"); 234363443bf5SEric Sandeen free_extent_buffer(log_tree_root->node); 234463443bf5SEric Sandeen kfree(log_tree_root); 234563443bf5SEric Sandeen return ret; 234663443bf5SEric Sandeen } 234763443bf5SEric Sandeen 234863443bf5SEric Sandeen if (fs_info->sb->s_flags & MS_RDONLY) { 234963443bf5SEric Sandeen ret = btrfs_commit_super(tree_root); 235063443bf5SEric Sandeen if (ret) 235163443bf5SEric Sandeen return ret; 235263443bf5SEric Sandeen } 235363443bf5SEric Sandeen 235463443bf5SEric Sandeen return 0; 235563443bf5SEric Sandeen } 235663443bf5SEric Sandeen 23574bbcaa64SEric Sandeen static int btrfs_read_roots(struct btrfs_fs_info *fs_info, 23584bbcaa64SEric Sandeen struct btrfs_root *tree_root) 23594bbcaa64SEric Sandeen { 2360a4f3d2c4SDavid Sterba struct btrfs_root *root; 23614bbcaa64SEric Sandeen struct btrfs_key location; 23624bbcaa64SEric Sandeen int ret; 23634bbcaa64SEric Sandeen 23644bbcaa64SEric Sandeen location.objectid = BTRFS_EXTENT_TREE_OBJECTID; 23654bbcaa64SEric Sandeen location.type = BTRFS_ROOT_ITEM_KEY; 23664bbcaa64SEric Sandeen location.offset = 0; 23674bbcaa64SEric Sandeen 2368a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2369a4f3d2c4SDavid Sterba if (IS_ERR(root)) 2370a4f3d2c4SDavid Sterba return PTR_ERR(root); 2371a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2372a4f3d2c4SDavid Sterba fs_info->extent_root = root; 23734bbcaa64SEric Sandeen 23744bbcaa64SEric Sandeen location.objectid = BTRFS_DEV_TREE_OBJECTID; 2375a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2376a4f3d2c4SDavid Sterba if (IS_ERR(root)) 2377a4f3d2c4SDavid Sterba return PTR_ERR(root); 2378a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2379a4f3d2c4SDavid Sterba fs_info->dev_root = root; 23804bbcaa64SEric Sandeen btrfs_init_devices_late(fs_info); 23814bbcaa64SEric Sandeen 23824bbcaa64SEric Sandeen location.objectid = BTRFS_CSUM_TREE_OBJECTID; 2383a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2384a4f3d2c4SDavid Sterba if (IS_ERR(root)) 2385a4f3d2c4SDavid Sterba return PTR_ERR(root); 2386a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2387a4f3d2c4SDavid Sterba fs_info->csum_root = root; 23884bbcaa64SEric Sandeen 23894bbcaa64SEric Sandeen location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2390a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2391a4f3d2c4SDavid Sterba if (!IS_ERR(root)) { 2392a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 23934bbcaa64SEric Sandeen fs_info->quota_enabled = 1; 23944bbcaa64SEric Sandeen fs_info->pending_quota_state = 1; 2395a4f3d2c4SDavid Sterba fs_info->quota_root = root; 23964bbcaa64SEric Sandeen } 23974bbcaa64SEric Sandeen 23984bbcaa64SEric Sandeen location.objectid = BTRFS_UUID_TREE_OBJECTID; 2399a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2400a4f3d2c4SDavid Sterba if (IS_ERR(root)) { 2401a4f3d2c4SDavid Sterba ret = PTR_ERR(root); 24024bbcaa64SEric Sandeen if (ret != -ENOENT) 24034bbcaa64SEric Sandeen return ret; 24044bbcaa64SEric Sandeen } else { 2405a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2406a4f3d2c4SDavid Sterba fs_info->uuid_root = root; 24074bbcaa64SEric Sandeen } 24084bbcaa64SEric Sandeen 24094bbcaa64SEric Sandeen return 0; 24104bbcaa64SEric Sandeen } 24114bbcaa64SEric Sandeen 2412ad2b2c80SAl Viro int open_ctree(struct super_block *sb, 2413dfe25020SChris Mason struct btrfs_fs_devices *fs_devices, 2414dfe25020SChris Mason char *options) 2415eb60ceacSChris Mason { 2416db94535dSChris Mason u32 sectorsize; 2417db94535dSChris Mason u32 nodesize; 241887ee04ebSChris Mason u32 stripesize; 241984234f3aSYan Zheng u64 generation; 2420f2b636e8SJosef Bacik u64 features; 24213de4586cSChris Mason struct btrfs_key location; 2422a061fc8dSChris Mason struct buffer_head *bh; 24234d34b278SIlya Dryomov struct btrfs_super_block *disk_super; 2424815745cfSAl Viro struct btrfs_fs_info *fs_info = btrfs_sb(sb); 2425f84a8bd6SAl Viro struct btrfs_root *tree_root; 24264d34b278SIlya Dryomov struct btrfs_root *chunk_root; 2427eb60ceacSChris Mason int ret; 2428e58ca020SYan int err = -EINVAL; 2429af31f5e5SChris Mason int num_backups_tried = 0; 2430af31f5e5SChris Mason int backup_index = 0; 24315cdc7ad3SQu Wenruo int max_active; 24324543df7eSChris Mason 2433f84a8bd6SAl Viro tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info); 24346f07e42eSAl Viro chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info); 2435cb517eabSMiao Xie if (!tree_root || !chunk_root) { 243639279cc3SChris Mason err = -ENOMEM; 243739279cc3SChris Mason goto fail; 243839279cc3SChris Mason } 243976dda93cSYan, Zheng 244076dda93cSYan, Zheng ret = init_srcu_struct(&fs_info->subvol_srcu); 244176dda93cSYan, Zheng if (ret) { 244276dda93cSYan, Zheng err = ret; 244376dda93cSYan, Zheng goto fail; 244476dda93cSYan, Zheng } 244576dda93cSYan, Zheng 244676dda93cSYan, Zheng ret = setup_bdi(fs_info, &fs_info->bdi); 244776dda93cSYan, Zheng if (ret) { 244876dda93cSYan, Zheng err = ret; 244976dda93cSYan, Zheng goto fail_srcu; 245076dda93cSYan, Zheng } 245176dda93cSYan, Zheng 2452908c7f19STejun Heo ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 2453e2d84521SMiao Xie if (ret) { 2454e2d84521SMiao Xie err = ret; 2455e2d84521SMiao Xie goto fail_bdi; 2456e2d84521SMiao Xie } 2457e2d84521SMiao Xie fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE * 2458e2d84521SMiao Xie (1 + ilog2(nr_cpu_ids)); 2459e2d84521SMiao Xie 2460908c7f19STejun Heo ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 2461963d678bSMiao Xie if (ret) { 2462963d678bSMiao Xie err = ret; 2463963d678bSMiao Xie goto fail_dirty_metadata_bytes; 2464963d678bSMiao Xie } 2465963d678bSMiao Xie 2466908c7f19STejun Heo ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL); 2467c404e0dcSMiao Xie if (ret) { 2468c404e0dcSMiao Xie err = ret; 2469c404e0dcSMiao Xie goto fail_delalloc_bytes; 2470c404e0dcSMiao Xie } 2471c404e0dcSMiao Xie 247276dda93cSYan, Zheng fs_info->btree_inode = new_inode(sb); 247376dda93cSYan, Zheng if (!fs_info->btree_inode) { 247476dda93cSYan, Zheng err = -ENOMEM; 2475c404e0dcSMiao Xie goto fail_bio_counter; 247676dda93cSYan, Zheng } 247776dda93cSYan, Zheng 2478a6591715SChris Mason mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 24791561dedaSMiao Xie 248076dda93cSYan, Zheng INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 2481f28491e0SJosef Bacik INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 24828fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 2483facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 248424bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 2485eb73c1b7SMiao Xie INIT_LIST_HEAD(&fs_info->delalloc_roots); 248611833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 2487eb73c1b7SMiao Xie spin_lock_init(&fs_info->delalloc_root_lock); 2488a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 248976dda93cSYan, Zheng spin_lock_init(&fs_info->fs_roots_radix_lock); 249024bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 24914cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 24922bf64758SJosef Bacik spin_lock_init(&fs_info->free_chunk_lock); 2493f29021b2SJan Schmidt spin_lock_init(&fs_info->tree_mod_seq_lock); 2494ceda0864SMiao Xie spin_lock_init(&fs_info->super_lock); 2495fcebe456SJosef Bacik spin_lock_init(&fs_info->qgroup_op_lock); 2496f28491e0SJosef Bacik spin_lock_init(&fs_info->buffer_lock); 249747ab2a6cSJosef Bacik spin_lock_init(&fs_info->unused_bgs_lock); 2498d4b450cdSFilipe Manana mutex_init(&fs_info->unused_bg_unpin_mutex); 2499f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 25007585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 2501573bfb72SMiao Xie mutex_init(&fs_info->delalloc_root_mutex); 2502de98ced9SMiao Xie seqlock_init(&fs_info->profiles_lock); 250319c00ddcSChris Mason 250458176a96SJosef Bacik init_completion(&fs_info->kobj_unregister); 25050b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 25066324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 2507f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 250847ab2a6cSJosef Bacik INIT_LIST_HEAD(&fs_info->unused_bgs); 25090b86a832SChris Mason btrfs_mapping_init(&fs_info->mapping_tree); 251066d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 251166d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 251266d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delalloc_block_rsv, 251366d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELALLOC); 251466d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 251566d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 251666d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 251766d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 251866d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 2519cb03c743SChris Mason atomic_set(&fs_info->nr_async_submits, 0); 2520771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 25218c8bee1dSChris Mason atomic_set(&fs_info->async_submit_draining, 0); 25220986fe9eSChris Mason atomic_set(&fs_info->nr_async_bios, 0); 25234cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 2524fcebe456SJosef Bacik atomic_set(&fs_info->qgroup_op_seq, 0); 2525fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 2526e20d96d6SChris Mason fs_info->sb = sb; 252795ac567aSFilipe David Borba Manana fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 25289ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 25294cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 25302bf64758SJosef Bacik fs_info->free_chunk_space = 0; 2531f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 25328b87dc17SDavid Sterba fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 25330a2b2a84SJosef Bacik fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64); 253490519d66SArne Jansen /* readahead state */ 253590519d66SArne Jansen INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT); 253690519d66SArne Jansen spin_lock_init(&fs_info->reada_lock); 2537c8b97818SChris Mason 2538b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 2539b34b086cSChris Mason num_online_cpus() + 2, 8); 25400afbaf8cSChris Mason 2541199c2a9cSMiao Xie INIT_LIST_HEAD(&fs_info->ordered_roots); 2542199c2a9cSMiao Xie spin_lock_init(&fs_info->ordered_root_lock); 254316cdcec7SMiao Xie fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 254416cdcec7SMiao Xie GFP_NOFS); 254516cdcec7SMiao Xie if (!fs_info->delayed_root) { 254616cdcec7SMiao Xie err = -ENOMEM; 254716cdcec7SMiao Xie goto fail_iput; 254816cdcec7SMiao Xie } 254916cdcec7SMiao Xie btrfs_init_delayed_root(fs_info->delayed_root); 25503eaa2885SChris Mason 2551638aa7edSEric Sandeen btrfs_init_scrub(fs_info); 255221adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 255321adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 255421adbd5cSStefan Behrens #endif 2555779a65a4SEric Sandeen btrfs_init_balance(fs_info); 255621c7e756SMiao Xie btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work); 2557a2de733cSArne Jansen 2558a061fc8dSChris Mason sb->s_blocksize = 4096; 2559a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(4096); 256032a88aa1SJens Axboe sb->s_bdi = &fs_info->bdi; 2561a061fc8dSChris Mason 2562f37938e0SEric Sandeen btrfs_init_btree_inode(fs_info, tree_root); 256339279cc3SChris Mason 2564e02119d5SChris Mason spin_lock_init(&fs_info->block_group_cache_lock); 25656bef4d31SEric Paris fs_info->block_group_cache_tree = RB_ROOT; 2566a1897fddSLiu Bo fs_info->first_logical_byte = (u64)-1; 2567925baeddSChris Mason 256811833d66SYan Zheng extent_io_tree_init(&fs_info->freed_extents[0], 2569f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 257011833d66SYan Zheng extent_io_tree_init(&fs_info->freed_extents[1], 2571f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 257211833d66SYan Zheng fs_info->pinned_extents = &fs_info->freed_extents[0]; 2573e6dcd2dcSChris Mason fs_info->do_barriers = 1; 2574f9295749SChris Mason 2575d98237b3SChris Mason 25765a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 25779ffba8cdSJosef Bacik mutex_init(&fs_info->ordered_extent_flush_mutex); 257830ae8467SChris Mason mutex_init(&fs_info->tree_log_mutex); 2579925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 2580925baeddSChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 2581a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 2582a74a4b97SChris Mason mutex_init(&fs_info->volume_mutex); 25839e351cc8SJosef Bacik init_rwsem(&fs_info->commit_root_sem); 2584c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 258576dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 2586803b2f54SStefan Behrens sema_init(&fs_info->uuid_tree_rescan_sem, 1); 2587ad618368SEric Sandeen 2588ad618368SEric Sandeen btrfs_init_dev_replace_locks(fs_info); 2589f9e92e40SEric Sandeen btrfs_init_qgroup(fs_info); 2590416ac51dSArne Jansen 2591fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2592fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2593fa9c0d79SChris Mason 25947d9eb12cSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 25953768f368SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 2596bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 25974854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 25989a8dd150SChris Mason 259904216820SFilipe Manana INIT_LIST_HEAD(&fs_info->pinned_chunks); 260004216820SFilipe Manana 260153b381b3SDavid Woodhouse ret = btrfs_alloc_stripe_hash_table(fs_info); 260253b381b3SDavid Woodhouse if (ret) { 260383c8266aSDavid Sterba err = ret; 260453b381b3SDavid Woodhouse goto fail_alloc; 260553b381b3SDavid Woodhouse } 260653b381b3SDavid Woodhouse 2607707e8a07SDavid Sterba __setup_root(4096, 4096, 4096, tree_root, 26082c90e5d6SChris Mason fs_info, BTRFS_ROOT_TREE_OBJECTID); 26097eccb903SChris Mason 26103c4bb26bSChris Mason invalidate_bdev(fs_devices->latest_bdev); 26111104a885SDavid Sterba 26121104a885SDavid Sterba /* 26131104a885SDavid Sterba * Read super block and check the signature bytes only 26141104a885SDavid Sterba */ 2615a512bbf8SYan Zheng bh = btrfs_read_dev_super(fs_devices->latest_bdev); 261620b45077SDave Young if (!bh) { 261720b45077SDave Young err = -EINVAL; 261816cdcec7SMiao Xie goto fail_alloc; 261920b45077SDave Young } 262039279cc3SChris Mason 26211104a885SDavid Sterba /* 26221104a885SDavid Sterba * We want to check superblock checksum, the type is stored inside. 26231104a885SDavid Sterba * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 26241104a885SDavid Sterba */ 26251104a885SDavid Sterba if (btrfs_check_super_csum(bh->b_data)) { 2626efe120a0SFrank Holton printk(KERN_ERR "BTRFS: superblock checksum mismatch\n"); 26271104a885SDavid Sterba err = -EINVAL; 26281104a885SDavid Sterba goto fail_alloc; 26291104a885SDavid Sterba } 26301104a885SDavid Sterba 26311104a885SDavid Sterba /* 26321104a885SDavid Sterba * super_copy is zeroed at allocation time and we never touch the 26331104a885SDavid Sterba * following bytes up to INFO_SIZE, the checksum is calculated from 26341104a885SDavid Sterba * the whole block of INFO_SIZE 26351104a885SDavid Sterba */ 26366c41761fSDavid Sterba memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy)); 26376c41761fSDavid Sterba memcpy(fs_info->super_for_commit, fs_info->super_copy, 26386c41761fSDavid Sterba sizeof(*fs_info->super_for_commit)); 2639a061fc8dSChris Mason brelse(bh); 26405f39d397SChris Mason 26416c41761fSDavid Sterba memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE); 26420b86a832SChris Mason 26431104a885SDavid Sterba ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY); 26441104a885SDavid Sterba if (ret) { 2645efe120a0SFrank Holton printk(KERN_ERR "BTRFS: superblock contains fatal errors\n"); 26461104a885SDavid Sterba err = -EINVAL; 26471104a885SDavid Sterba goto fail_alloc; 26481104a885SDavid Sterba } 26491104a885SDavid Sterba 26506c41761fSDavid Sterba disk_super = fs_info->super_copy; 26510f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 265216cdcec7SMiao Xie goto fail_alloc; 26530f7d52f4SChris Mason 2654acce952bSliubo /* check FS state, whether FS is broken. */ 265587533c47SMiao Xie if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 265687533c47SMiao Xie set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 2657acce952bSliubo 265875e7cb7fSLiu Bo /* 2659af31f5e5SChris Mason * run through our array of backup supers and setup 2660af31f5e5SChris Mason * our ring pointer to the oldest one 2661af31f5e5SChris Mason */ 2662af31f5e5SChris Mason generation = btrfs_super_generation(disk_super); 2663af31f5e5SChris Mason find_oldest_super_backup(fs_info, generation); 2664af31f5e5SChris Mason 2665af31f5e5SChris Mason /* 266675e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 266775e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 266875e7cb7fSLiu Bo */ 266975e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 267075e7cb7fSLiu Bo 26712b82032cSYan Zheng ret = btrfs_parse_options(tree_root, options); 26722b82032cSYan Zheng if (ret) { 26732b82032cSYan Zheng err = ret; 267416cdcec7SMiao Xie goto fail_alloc; 26752b82032cSYan Zheng } 2676dfe25020SChris Mason 2677f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 2678f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 2679f2b636e8SJosef Bacik if (features) { 2680f2b636e8SJosef Bacik printk(KERN_ERR "BTRFS: couldn't mount because of " 2681f2b636e8SJosef Bacik "unsupported optional features (%Lx).\n", 2682c1c9ff7cSGeert Uytterhoeven features); 2683f2b636e8SJosef Bacik err = -EINVAL; 268416cdcec7SMiao Xie goto fail_alloc; 2685f2b636e8SJosef Bacik } 2686f2b636e8SJosef Bacik 2687707e8a07SDavid Sterba /* 2688707e8a07SDavid Sterba * Leafsize and nodesize were always equal, this is only a sanity check. 2689707e8a07SDavid Sterba */ 2690707e8a07SDavid Sterba if (le32_to_cpu(disk_super->__unused_leafsize) != 2691727011e0SChris Mason btrfs_super_nodesize(disk_super)) { 2692727011e0SChris Mason printk(KERN_ERR "BTRFS: couldn't mount because metadata " 2693727011e0SChris Mason "blocksizes don't match. node %d leaf %d\n", 2694727011e0SChris Mason btrfs_super_nodesize(disk_super), 2695707e8a07SDavid Sterba le32_to_cpu(disk_super->__unused_leafsize)); 2696727011e0SChris Mason err = -EINVAL; 2697727011e0SChris Mason goto fail_alloc; 2698727011e0SChris Mason } 2699707e8a07SDavid Sterba if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) { 2700727011e0SChris Mason printk(KERN_ERR "BTRFS: couldn't mount because metadata " 2701727011e0SChris Mason "blocksize (%d) was too large\n", 2702707e8a07SDavid Sterba btrfs_super_nodesize(disk_super)); 2703727011e0SChris Mason err = -EINVAL; 2704727011e0SChris Mason goto fail_alloc; 2705727011e0SChris Mason } 2706727011e0SChris Mason 27075d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 27085d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 270969e380d1SLi Zefan if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO) 2710a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 2711727011e0SChris Mason 27123173a18fSJosef Bacik if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) 27135efa0490SDavid Sterba printk(KERN_INFO "BTRFS: has skinny extents\n"); 27143173a18fSJosef Bacik 2715727011e0SChris Mason /* 2716727011e0SChris Mason * flag our filesystem as having big metadata blocks if 2717727011e0SChris Mason * they are bigger than the page size 2718727011e0SChris Mason */ 2719707e8a07SDavid Sterba if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) { 2720727011e0SChris Mason if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 2721efe120a0SFrank Holton printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n"); 2722727011e0SChris Mason features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 2723727011e0SChris Mason } 2724727011e0SChris Mason 2725bc3f116fSChris Mason nodesize = btrfs_super_nodesize(disk_super); 2726bc3f116fSChris Mason sectorsize = btrfs_super_sectorsize(disk_super); 2727bc3f116fSChris Mason stripesize = btrfs_super_stripesize(disk_super); 2728707e8a07SDavid Sterba fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 2729963d678bSMiao Xie fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 2730bc3f116fSChris Mason 2731bc3f116fSChris Mason /* 2732bc3f116fSChris Mason * mixed block groups end up with duplicate but slightly offset 2733bc3f116fSChris Mason * extent buffers for the same range. It leads to corruptions 2734bc3f116fSChris Mason */ 2735bc3f116fSChris Mason if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 2736707e8a07SDavid Sterba (sectorsize != nodesize)) { 2737aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: unequal leaf/node/sector sizes " 2738bc3f116fSChris Mason "are not allowed for mixed block groups on %s\n", 2739bc3f116fSChris Mason sb->s_id); 2740bc3f116fSChris Mason goto fail_alloc; 2741bc3f116fSChris Mason } 2742bc3f116fSChris Mason 2743ceda0864SMiao Xie /* 2744ceda0864SMiao Xie * Needn't use the lock because there is no other task which will 2745ceda0864SMiao Xie * update the flag. 2746ceda0864SMiao Xie */ 27475d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 27485d4f98a2SYan Zheng 2749f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 2750f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 2751f2b636e8SJosef Bacik if (!(sb->s_flags & MS_RDONLY) && features) { 2752f2b636e8SJosef Bacik printk(KERN_ERR "BTRFS: couldn't mount RDWR because of " 2753f2b636e8SJosef Bacik "unsupported option features (%Lx).\n", 2754c1c9ff7cSGeert Uytterhoeven features); 2755f2b636e8SJosef Bacik err = -EINVAL; 275616cdcec7SMiao Xie goto fail_alloc; 2757f2b636e8SJosef Bacik } 275861d92c32SChris Mason 27595cdc7ad3SQu Wenruo max_active = fs_info->thread_pool_size; 276061d92c32SChris Mason 27612a458198SEric Sandeen ret = btrfs_init_workqueues(fs_info, fs_devices); 27622a458198SEric Sandeen if (ret) { 27632a458198SEric Sandeen err = ret; 27640dc3b84aSJosef Bacik goto fail_sb_buffer; 27650dc3b84aSJosef Bacik } 27664543df7eSChris Mason 27674575c9ccSChris Mason fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super); 2768c8b97818SChris Mason fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages, 2769c8b97818SChris Mason 4 * 1024 * 1024 / PAGE_CACHE_SIZE); 27704575c9ccSChris Mason 2771db94535dSChris Mason tree_root->nodesize = nodesize; 2772db94535dSChris Mason tree_root->sectorsize = sectorsize; 277387ee04ebSChris Mason tree_root->stripesize = stripesize; 2774a061fc8dSChris Mason 2775a061fc8dSChris Mason sb->s_blocksize = sectorsize; 2776a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 2777db94535dSChris Mason 27783cae210fSQu Wenruo if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) { 2779aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: valid FS not found on %s\n", sb->s_id); 278039279cc3SChris Mason goto fail_sb_buffer; 278139279cc3SChris Mason } 278219c00ddcSChris Mason 27838d082fb7SLiu Bo if (sectorsize != PAGE_SIZE) { 2784aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: incompatible sector size (%lu) " 27858d082fb7SLiu Bo "found on %s\n", (unsigned long)sectorsize, sb->s_id); 2786941b2ddfSKeith Mannthey goto fail_sb_buffer; 2787941b2ddfSKeith Mannthey } 2788941b2ddfSKeith Mannthey 2789925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 2790e4404d6eSYan Zheng ret = btrfs_read_sys_array(tree_root); 2791925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 279284eed90fSChris Mason if (ret) { 2793aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: failed to read the system " 2794d397712bSChris Mason "array on %s\n", sb->s_id); 27955d4f98a2SYan Zheng goto fail_sb_buffer; 279684eed90fSChris Mason } 27970b86a832SChris Mason 279884234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 27990b86a832SChris Mason 2800707e8a07SDavid Sterba __setup_root(nodesize, sectorsize, stripesize, chunk_root, 2801707e8a07SDavid Sterba fs_info, BTRFS_CHUNK_TREE_OBJECTID); 28020b86a832SChris Mason 28030b86a832SChris Mason chunk_root->node = read_tree_block(chunk_root, 28040b86a832SChris Mason btrfs_super_chunk_root(disk_super), 2805ce86cd59SDavid Sterba generation); 2806416bc658SJosef Bacik if (!chunk_root->node || 2807416bc658SJosef Bacik !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) { 2808aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: failed to read chunk root on %s\n", 280983121942SDavid Woodhouse sb->s_id); 2810af31f5e5SChris Mason goto fail_tree_roots; 281183121942SDavid Woodhouse } 28125d4f98a2SYan Zheng btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 28135d4f98a2SYan Zheng chunk_root->commit_root = btrfs_root_node(chunk_root); 28140b86a832SChris Mason 2815e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 2816b308bc2fSGeert Uytterhoeven btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE); 2817e17cade2SChris Mason 28180b86a832SChris Mason ret = btrfs_read_chunk_tree(chunk_root); 28192b82032cSYan Zheng if (ret) { 2820aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: failed to read chunk tree on %s\n", 2821d397712bSChris Mason sb->s_id); 2822af31f5e5SChris Mason goto fail_tree_roots; 28232b82032cSYan Zheng } 28240b86a832SChris Mason 28258dabb742SStefan Behrens /* 28268dabb742SStefan Behrens * keep the device that is marked to be the target device for the 28278dabb742SStefan Behrens * dev_replace procedure 28288dabb742SStefan Behrens */ 28299eaed21eSEric Sandeen btrfs_close_extra_devices(fs_devices, 0); 2830dfe25020SChris Mason 2831a6b0d5c8SChris Mason if (!fs_devices->latest_bdev) { 2832aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: failed to read devices on %s\n", 2833a6b0d5c8SChris Mason sb->s_id); 2834a6b0d5c8SChris Mason goto fail_tree_roots; 2835a6b0d5c8SChris Mason } 2836a6b0d5c8SChris Mason 2837af31f5e5SChris Mason retry_root_backup: 283884234f3aSYan Zheng generation = btrfs_super_generation(disk_super); 28390b86a832SChris Mason 2840e20d96d6SChris Mason tree_root->node = read_tree_block(tree_root, 2841db94535dSChris Mason btrfs_super_root(disk_super), 2842ce86cd59SDavid Sterba generation); 2843af31f5e5SChris Mason if (!tree_root->node || 2844af31f5e5SChris Mason !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) { 2845efe120a0SFrank Holton printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n", 284683121942SDavid Woodhouse sb->s_id); 2847af31f5e5SChris Mason 2848af31f5e5SChris Mason goto recovery_tree_root; 284983121942SDavid Woodhouse } 2850af31f5e5SChris Mason 28515d4f98a2SYan Zheng btrfs_set_root_node(&tree_root->root_item, tree_root->node); 28525d4f98a2SYan Zheng tree_root->commit_root = btrfs_root_node(tree_root); 285369e9c6c6SStefan Behrens btrfs_set_root_refs(&tree_root->root_item, 1); 2854db94535dSChris Mason 28554bbcaa64SEric Sandeen ret = btrfs_read_roots(fs_info, tree_root); 28564bbcaa64SEric Sandeen if (ret) 2857af31f5e5SChris Mason goto recovery_tree_root; 2858f7a81ea4SStefan Behrens 28598929ecfaSYan, Zheng fs_info->generation = generation; 28608929ecfaSYan, Zheng fs_info->last_trans_committed = generation; 28618929ecfaSYan, Zheng 286268310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 286368310a5eSIlya Dryomov if (ret) { 2864aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: failed to recover balance\n"); 286568310a5eSIlya Dryomov goto fail_block_groups; 286668310a5eSIlya Dryomov } 286768310a5eSIlya Dryomov 2868733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 2869733f4fbbSStefan Behrens if (ret) { 2870efe120a0SFrank Holton printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n", 2871733f4fbbSStefan Behrens ret); 2872733f4fbbSStefan Behrens goto fail_block_groups; 2873733f4fbbSStefan Behrens } 2874733f4fbbSStefan Behrens 28758dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 28768dabb742SStefan Behrens if (ret) { 2877efe120a0SFrank Holton pr_err("BTRFS: failed to init dev_replace: %d\n", ret); 28788dabb742SStefan Behrens goto fail_block_groups; 28798dabb742SStefan Behrens } 28808dabb742SStefan Behrens 28819eaed21eSEric Sandeen btrfs_close_extra_devices(fs_devices, 1); 28828dabb742SStefan Behrens 28835ac1d209SJeff Mahoney ret = btrfs_sysfs_add_one(fs_info); 28845ac1d209SJeff Mahoney if (ret) { 2885efe120a0SFrank Holton pr_err("BTRFS: failed to init sysfs interface: %d\n", ret); 28865ac1d209SJeff Mahoney goto fail_block_groups; 28875ac1d209SJeff Mahoney } 28885ac1d209SJeff Mahoney 2889c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 2890c59021f8Sliubo if (ret) { 2891efe120a0SFrank Holton printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret); 28922365dd3cSAnand Jain goto fail_sysfs; 2893c59021f8Sliubo } 2894c59021f8Sliubo 28954bbcaa64SEric Sandeen ret = btrfs_read_block_groups(fs_info->extent_root); 28961b1d1f66SJosef Bacik if (ret) { 2897efe120a0SFrank Holton printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret); 28982365dd3cSAnand Jain goto fail_sysfs; 28991b1d1f66SJosef Bacik } 29005af3e8ccSStefan Behrens fs_info->num_tolerated_disk_barrier_failures = 29015af3e8ccSStefan Behrens btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); 2902292fd7fcSStefan Behrens if (fs_info->fs_devices->missing_devices > 2903292fd7fcSStefan Behrens fs_info->num_tolerated_disk_barrier_failures && 2904292fd7fcSStefan Behrens !(sb->s_flags & MS_RDONLY)) { 2905efe120a0SFrank Holton printk(KERN_WARNING "BTRFS: " 2906efe120a0SFrank Holton "too many missing devices, writeable mount is not allowed\n"); 29072365dd3cSAnand Jain goto fail_sysfs; 2908292fd7fcSStefan Behrens } 29099078a3e1SChris Mason 2910a74a4b97SChris Mason fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 2911a74a4b97SChris Mason "btrfs-cleaner"); 291257506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 29132365dd3cSAnand Jain goto fail_sysfs; 2914a74a4b97SChris Mason 2915a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 2916a74a4b97SChris Mason tree_root, 2917a74a4b97SChris Mason "btrfs-transaction"); 291857506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 29193f157a2fSChris Mason goto fail_cleaner; 2920a74a4b97SChris Mason 2921c289811cSChris Mason if (!btrfs_test_opt(tree_root, SSD) && 2922c289811cSChris Mason !btrfs_test_opt(tree_root, NOSSD) && 2923c289811cSChris Mason !fs_info->fs_devices->rotating) { 2924efe120a0SFrank Holton printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD " 2925c289811cSChris Mason "mode\n"); 2926c289811cSChris Mason btrfs_set_opt(fs_info->mount_opt, SSD); 2927c289811cSChris Mason } 2928c289811cSChris Mason 2929572d9ab7SDavid Sterba /* 2930572d9ab7SDavid Sterba * Mount does not set all options immediatelly, we can do it now and do 2931572d9ab7SDavid Sterba * not have to wait for transaction commit 2932572d9ab7SDavid Sterba */ 2933572d9ab7SDavid Sterba btrfs_apply_pending_changes(fs_info); 29343818aea2SQu Wenruo 293521adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 293621adbd5cSStefan Behrens if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) { 293721adbd5cSStefan Behrens ret = btrfsic_mount(tree_root, fs_devices, 293821adbd5cSStefan Behrens btrfs_test_opt(tree_root, 293921adbd5cSStefan Behrens CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? 294021adbd5cSStefan Behrens 1 : 0, 294121adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 294221adbd5cSStefan Behrens if (ret) 2943efe120a0SFrank Holton printk(KERN_WARNING "BTRFS: failed to initialize" 294421adbd5cSStefan Behrens " integrity check module %s\n", sb->s_id); 294521adbd5cSStefan Behrens } 294621adbd5cSStefan Behrens #endif 2947bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 2948bcef60f2SArne Jansen if (ret) 2949bcef60f2SArne Jansen goto fail_trans_kthread; 295021adbd5cSStefan Behrens 2951acce952bSliubo /* do not make disk changes in broken FS */ 295268ce9682SStefan Behrens if (btrfs_super_log_root(disk_super) != 0) { 295363443bf5SEric Sandeen ret = btrfs_replay_log(fs_info, fs_devices); 295479787eaaSJeff Mahoney if (ret) { 295563443bf5SEric Sandeen err = ret; 295628c16cbbSWang Shilong goto fail_qgroup; 2957e556ce2cSYan Zheng } 2958e02119d5SChris Mason } 29591a40e23bSZheng Yan 296076dda93cSYan, Zheng ret = btrfs_find_orphan_roots(tree_root); 296179787eaaSJeff Mahoney if (ret) 296228c16cbbSWang Shilong goto fail_qgroup; 296376dda93cSYan, Zheng 29647c2ca468SChris Mason if (!(sb->s_flags & MS_RDONLY)) { 2965d68fc57bSYan, Zheng ret = btrfs_cleanup_fs_roots(fs_info); 296644c44af2SIlya Dryomov if (ret) 296728c16cbbSWang Shilong goto fail_qgroup; 2968d68fc57bSYan, Zheng 29695f316481SWang Shilong mutex_lock(&fs_info->cleaner_mutex); 29705d4f98a2SYan Zheng ret = btrfs_recover_relocation(tree_root); 29715f316481SWang Shilong mutex_unlock(&fs_info->cleaner_mutex); 2972d7ce5843SMiao Xie if (ret < 0) { 2973d7ce5843SMiao Xie printk(KERN_WARNING 2974efe120a0SFrank Holton "BTRFS: failed to recover relocation\n"); 2975d7ce5843SMiao Xie err = -EINVAL; 2976bcef60f2SArne Jansen goto fail_qgroup; 2977d7ce5843SMiao Xie } 29787c2ca468SChris Mason } 29791a40e23bSZheng Yan 29803de4586cSChris Mason location.objectid = BTRFS_FS_TREE_OBJECTID; 29813de4586cSChris Mason location.type = BTRFS_ROOT_ITEM_KEY; 2982cb517eabSMiao Xie location.offset = 0; 29833de4586cSChris Mason 29843de4586cSChris Mason fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location); 29853140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 29863140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 2987bcef60f2SArne Jansen goto fail_qgroup; 29883140c9a3SDan Carpenter } 2989c289811cSChris Mason 29902b6ba629SIlya Dryomov if (sb->s_flags & MS_RDONLY) 29912b6ba629SIlya Dryomov return 0; 29922b6ba629SIlya Dryomov 2993e3acc2a6SJosef Bacik down_read(&fs_info->cleanup_work_sem); 29942b6ba629SIlya Dryomov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 29952b6ba629SIlya Dryomov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 29962b6ba629SIlya Dryomov up_read(&fs_info->cleanup_work_sem); 29972b6ba629SIlya Dryomov close_ctree(tree_root); 29982b6ba629SIlya Dryomov return ret; 29992b6ba629SIlya Dryomov } 3000e3acc2a6SJosef Bacik up_read(&fs_info->cleanup_work_sem); 300159641015SIlya Dryomov 30022b6ba629SIlya Dryomov ret = btrfs_resume_balance_async(fs_info); 30032b6ba629SIlya Dryomov if (ret) { 3004efe120a0SFrank Holton printk(KERN_WARNING "BTRFS: failed to resume balance\n"); 300566b4ffd1SJosef Bacik close_ctree(tree_root); 30062b6ba629SIlya Dryomov return ret; 3007e3acc2a6SJosef Bacik } 3008e3acc2a6SJosef Bacik 30098dabb742SStefan Behrens ret = btrfs_resume_dev_replace_async(fs_info); 30108dabb742SStefan Behrens if (ret) { 3011efe120a0SFrank Holton pr_warn("BTRFS: failed to resume dev_replace\n"); 30128dabb742SStefan Behrens close_ctree(tree_root); 30138dabb742SStefan Behrens return ret; 30148dabb742SStefan Behrens } 30158dabb742SStefan Behrens 3016b382a324SJan Schmidt btrfs_qgroup_rescan_resume(fs_info); 3017b382a324SJan Schmidt 30184bbcaa64SEric Sandeen if (!fs_info->uuid_root) { 3019efe120a0SFrank Holton pr_info("BTRFS: creating UUID tree\n"); 3020f7a81ea4SStefan Behrens ret = btrfs_create_uuid_tree(fs_info); 3021f7a81ea4SStefan Behrens if (ret) { 3022efe120a0SFrank Holton pr_warn("BTRFS: failed to create the UUID tree %d\n", 3023f7a81ea4SStefan Behrens ret); 3024f7a81ea4SStefan Behrens close_ctree(tree_root); 3025f7a81ea4SStefan Behrens return ret; 3026f7a81ea4SStefan Behrens } 30274bbcaa64SEric Sandeen } else if (btrfs_test_opt(tree_root, RESCAN_UUID_TREE) || 30284bbcaa64SEric Sandeen fs_info->generation != 30294bbcaa64SEric Sandeen btrfs_super_uuid_tree_generation(disk_super)) { 3030efe120a0SFrank Holton pr_info("BTRFS: checking UUID tree\n"); 303170f80175SStefan Behrens ret = btrfs_check_uuid_tree(fs_info); 303270f80175SStefan Behrens if (ret) { 3033efe120a0SFrank Holton pr_warn("BTRFS: failed to check the UUID tree %d\n", 303470f80175SStefan Behrens ret); 303570f80175SStefan Behrens close_ctree(tree_root); 303670f80175SStefan Behrens return ret; 303770f80175SStefan Behrens } 303870f80175SStefan Behrens } else { 303970f80175SStefan Behrens fs_info->update_uuid_tree_gen = 1; 3040f7a81ea4SStefan Behrens } 3041f7a81ea4SStefan Behrens 304247ab2a6cSJosef Bacik fs_info->open = 1; 304347ab2a6cSJosef Bacik 3044ad2b2c80SAl Viro return 0; 304539279cc3SChris Mason 3046bcef60f2SArne Jansen fail_qgroup: 3047bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 30487c2ca468SChris Mason fail_trans_kthread: 30497c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 305054067ae9SJosef Bacik btrfs_cleanup_transaction(fs_info->tree_root); 3051faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 30523f157a2fSChris Mason fail_cleaner: 3053a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 30547c2ca468SChris Mason 30557c2ca468SChris Mason /* 30567c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 30577c2ca468SChris Mason * kthreads 30587c2ca468SChris Mason */ 30597c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 30607c2ca468SChris Mason 30612365dd3cSAnand Jain fail_sysfs: 30622365dd3cSAnand Jain btrfs_sysfs_remove_one(fs_info); 30632365dd3cSAnand Jain 30641b1d1f66SJosef Bacik fail_block_groups: 306554067ae9SJosef Bacik btrfs_put_block_group_cache(fs_info); 30661b1d1f66SJosef Bacik btrfs_free_block_groups(fs_info); 3067af31f5e5SChris Mason 3068af31f5e5SChris Mason fail_tree_roots: 3069af31f5e5SChris Mason free_root_pointers(fs_info, 1); 30702b8195bbSMiao Xie invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3071af31f5e5SChris Mason 307239279cc3SChris Mason fail_sb_buffer: 30737abadb64SLiu Bo btrfs_stop_all_workers(fs_info); 307416cdcec7SMiao Xie fail_alloc: 30754543df7eSChris Mason fail_iput: 3076586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 3077586e46e2SIlya Dryomov 30784543df7eSChris Mason iput(fs_info->btree_inode); 3079c404e0dcSMiao Xie fail_bio_counter: 3080c404e0dcSMiao Xie percpu_counter_destroy(&fs_info->bio_counter); 3081963d678bSMiao Xie fail_delalloc_bytes: 3082963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 3083e2d84521SMiao Xie fail_dirty_metadata_bytes: 3084e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 3085ad081f14SJens Axboe fail_bdi: 30867e662854SQinghuang Feng bdi_destroy(&fs_info->bdi); 308776dda93cSYan, Zheng fail_srcu: 308876dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 30897e662854SQinghuang Feng fail: 309053b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 3091586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 3092ad2b2c80SAl Viro return err; 3093af31f5e5SChris Mason 3094af31f5e5SChris Mason recovery_tree_root: 3095af31f5e5SChris Mason if (!btrfs_test_opt(tree_root, RECOVERY)) 3096af31f5e5SChris Mason goto fail_tree_roots; 3097af31f5e5SChris Mason 3098af31f5e5SChris Mason free_root_pointers(fs_info, 0); 3099af31f5e5SChris Mason 3100af31f5e5SChris Mason /* don't use the log in recovery mode, it won't be valid */ 3101af31f5e5SChris Mason btrfs_set_super_log_root(disk_super, 0); 3102af31f5e5SChris Mason 3103af31f5e5SChris Mason /* we can't trust the free space cache either */ 3104af31f5e5SChris Mason btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 3105af31f5e5SChris Mason 3106af31f5e5SChris Mason ret = next_root_backup(fs_info, fs_info->super_copy, 3107af31f5e5SChris Mason &num_backups_tried, &backup_index); 3108af31f5e5SChris Mason if (ret == -1) 3109af31f5e5SChris Mason goto fail_block_groups; 3110af31f5e5SChris Mason goto retry_root_backup; 3111eb60ceacSChris Mason } 3112eb60ceacSChris Mason 3113f2984462SChris Mason static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate) 3114f2984462SChris Mason { 3115f2984462SChris Mason if (uptodate) { 3116f2984462SChris Mason set_buffer_uptodate(bh); 3117f2984462SChris Mason } else { 3118442a4f63SStefan Behrens struct btrfs_device *device = (struct btrfs_device *) 3119442a4f63SStefan Behrens bh->b_private; 3120442a4f63SStefan Behrens 3121efe120a0SFrank Holton printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to " 3122606686eeSJosef Bacik "I/O error on %s\n", 3123606686eeSJosef Bacik rcu_str_deref(device->name)); 31241259ab75SChris Mason /* note, we dont' set_buffer_write_io_error because we have 31251259ab75SChris Mason * our own ways of dealing with the IO errors 31261259ab75SChris Mason */ 3127f2984462SChris Mason clear_buffer_uptodate(bh); 3128442a4f63SStefan Behrens btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS); 3129f2984462SChris Mason } 3130f2984462SChris Mason unlock_buffer(bh); 3131f2984462SChris Mason put_bh(bh); 3132f2984462SChris Mason } 3133f2984462SChris Mason 3134a512bbf8SYan Zheng struct buffer_head *btrfs_read_dev_super(struct block_device *bdev) 3135a512bbf8SYan Zheng { 3136a512bbf8SYan Zheng struct buffer_head *bh; 3137a512bbf8SYan Zheng struct buffer_head *latest = NULL; 3138a512bbf8SYan Zheng struct btrfs_super_block *super; 3139a512bbf8SYan Zheng int i; 3140a512bbf8SYan Zheng u64 transid = 0; 3141a512bbf8SYan Zheng u64 bytenr; 3142a512bbf8SYan Zheng 3143a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 3144a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 3145a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 3146a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3147a512bbf8SYan Zheng */ 3148a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 3149a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 31508068a47eSAnand Jain if (bytenr + BTRFS_SUPER_INFO_SIZE >= 31518068a47eSAnand Jain i_size_read(bdev->bd_inode)) 3152a512bbf8SYan Zheng break; 31538068a47eSAnand Jain bh = __bread(bdev, bytenr / 4096, 31548068a47eSAnand Jain BTRFS_SUPER_INFO_SIZE); 3155a512bbf8SYan Zheng if (!bh) 3156a512bbf8SYan Zheng continue; 3157a512bbf8SYan Zheng 3158a512bbf8SYan Zheng super = (struct btrfs_super_block *)bh->b_data; 3159a512bbf8SYan Zheng if (btrfs_super_bytenr(super) != bytenr || 31603cae210fSQu Wenruo btrfs_super_magic(super) != BTRFS_MAGIC) { 3161a512bbf8SYan Zheng brelse(bh); 3162a512bbf8SYan Zheng continue; 3163a512bbf8SYan Zheng } 3164a512bbf8SYan Zheng 3165a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 3166a512bbf8SYan Zheng brelse(latest); 3167a512bbf8SYan Zheng latest = bh; 3168a512bbf8SYan Zheng transid = btrfs_super_generation(super); 3169a512bbf8SYan Zheng } else { 3170a512bbf8SYan Zheng brelse(bh); 3171a512bbf8SYan Zheng } 3172a512bbf8SYan Zheng } 3173a512bbf8SYan Zheng return latest; 3174a512bbf8SYan Zheng } 3175a512bbf8SYan Zheng 31764eedeb75SHisashi Hifumi /* 31774eedeb75SHisashi Hifumi * this should be called twice, once with wait == 0 and 31784eedeb75SHisashi Hifumi * once with wait == 1. When wait == 0 is done, all the buffer heads 31794eedeb75SHisashi Hifumi * we write are pinned. 31804eedeb75SHisashi Hifumi * 31814eedeb75SHisashi Hifumi * They are released when wait == 1 is done. 31824eedeb75SHisashi Hifumi * max_mirrors must be the same for both runs, and it indicates how 31834eedeb75SHisashi Hifumi * many supers on this one device should be written. 31844eedeb75SHisashi Hifumi * 31854eedeb75SHisashi Hifumi * max_mirrors == 0 means to write them all. 31864eedeb75SHisashi Hifumi */ 3187a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 3188a512bbf8SYan Zheng struct btrfs_super_block *sb, 3189a512bbf8SYan Zheng int do_barriers, int wait, int max_mirrors) 3190a512bbf8SYan Zheng { 3191a512bbf8SYan Zheng struct buffer_head *bh; 3192a512bbf8SYan Zheng int i; 3193a512bbf8SYan Zheng int ret; 3194a512bbf8SYan Zheng int errors = 0; 3195a512bbf8SYan Zheng u32 crc; 3196a512bbf8SYan Zheng u64 bytenr; 3197a512bbf8SYan Zheng 3198a512bbf8SYan Zheng if (max_mirrors == 0) 3199a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3200a512bbf8SYan Zheng 3201a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 3202a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 3203935e5cc9SMiao Xie if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3204935e5cc9SMiao Xie device->commit_total_bytes) 3205a512bbf8SYan Zheng break; 3206a512bbf8SYan Zheng 3207a512bbf8SYan Zheng if (wait) { 3208a512bbf8SYan Zheng bh = __find_get_block(device->bdev, bytenr / 4096, 3209a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 3210634554dcSJosef Bacik if (!bh) { 3211634554dcSJosef Bacik errors++; 3212634554dcSJosef Bacik continue; 3213634554dcSJosef Bacik } 3214a512bbf8SYan Zheng wait_on_buffer(bh); 32154eedeb75SHisashi Hifumi if (!buffer_uptodate(bh)) 32164eedeb75SHisashi Hifumi errors++; 32174eedeb75SHisashi Hifumi 32184eedeb75SHisashi Hifumi /* drop our reference */ 32194eedeb75SHisashi Hifumi brelse(bh); 32204eedeb75SHisashi Hifumi 32214eedeb75SHisashi Hifumi /* drop the reference from the wait == 0 run */ 3222a512bbf8SYan Zheng brelse(bh); 3223a512bbf8SYan Zheng continue; 3224a512bbf8SYan Zheng } else { 3225a512bbf8SYan Zheng btrfs_set_super_bytenr(sb, bytenr); 3226a512bbf8SYan Zheng 3227a512bbf8SYan Zheng crc = ~(u32)0; 3228b0496686SLiu Bo crc = btrfs_csum_data((char *)sb + 3229a512bbf8SYan Zheng BTRFS_CSUM_SIZE, crc, 3230a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE - 3231a512bbf8SYan Zheng BTRFS_CSUM_SIZE); 3232a512bbf8SYan Zheng btrfs_csum_final(crc, sb->csum); 3233a512bbf8SYan Zheng 32344eedeb75SHisashi Hifumi /* 32354eedeb75SHisashi Hifumi * one reference for us, and we leave it for the 32364eedeb75SHisashi Hifumi * caller 32374eedeb75SHisashi Hifumi */ 3238a512bbf8SYan Zheng bh = __getblk(device->bdev, bytenr / 4096, 3239a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 3240634554dcSJosef Bacik if (!bh) { 3241efe120a0SFrank Holton printk(KERN_ERR "BTRFS: couldn't get super " 3242634554dcSJosef Bacik "buffer head for bytenr %Lu\n", bytenr); 3243634554dcSJosef Bacik errors++; 3244634554dcSJosef Bacik continue; 3245634554dcSJosef Bacik } 3246634554dcSJosef Bacik 3247a512bbf8SYan Zheng memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE); 3248a512bbf8SYan Zheng 32494eedeb75SHisashi Hifumi /* one reference for submit_bh */ 3250a512bbf8SYan Zheng get_bh(bh); 32514eedeb75SHisashi Hifumi 32524eedeb75SHisashi Hifumi set_buffer_uptodate(bh); 3253a512bbf8SYan Zheng lock_buffer(bh); 3254a512bbf8SYan Zheng bh->b_end_io = btrfs_end_buffer_write_sync; 3255442a4f63SStefan Behrens bh->b_private = device; 3256a512bbf8SYan Zheng } 3257a512bbf8SYan Zheng 3258387125fcSChris Mason /* 3259387125fcSChris Mason * we fua the first super. The others we allow 3260387125fcSChris Mason * to go down lazy. 3261387125fcSChris Mason */ 3262e8117c26SWang Shilong if (i == 0) 326321adbd5cSStefan Behrens ret = btrfsic_submit_bh(WRITE_FUA, bh); 3264e8117c26SWang Shilong else 3265e8117c26SWang Shilong ret = btrfsic_submit_bh(WRITE_SYNC, bh); 32664eedeb75SHisashi Hifumi if (ret) 3267a512bbf8SYan Zheng errors++; 3268a512bbf8SYan Zheng } 3269a512bbf8SYan Zheng return errors < i ? 0 : -1; 3270a512bbf8SYan Zheng } 3271a512bbf8SYan Zheng 3272387125fcSChris Mason /* 3273387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 3274387125fcSChris Mason * for the barrier when it is done 3275387125fcSChris Mason */ 3276387125fcSChris Mason static void btrfs_end_empty_barrier(struct bio *bio, int err) 3277387125fcSChris Mason { 3278387125fcSChris Mason if (err) { 3279387125fcSChris Mason if (err == -EOPNOTSUPP) 3280387125fcSChris Mason set_bit(BIO_EOPNOTSUPP, &bio->bi_flags); 3281387125fcSChris Mason clear_bit(BIO_UPTODATE, &bio->bi_flags); 3282387125fcSChris Mason } 3283387125fcSChris Mason if (bio->bi_private) 3284387125fcSChris Mason complete(bio->bi_private); 3285387125fcSChris Mason bio_put(bio); 3286387125fcSChris Mason } 3287387125fcSChris Mason 3288387125fcSChris Mason /* 3289387125fcSChris Mason * trigger flushes for one the devices. If you pass wait == 0, the flushes are 3290387125fcSChris Mason * sent down. With wait == 1, it waits for the previous flush. 3291387125fcSChris Mason * 3292387125fcSChris Mason * any device where the flush fails with eopnotsupp are flagged as not-barrier 3293387125fcSChris Mason * capable 3294387125fcSChris Mason */ 3295387125fcSChris Mason static int write_dev_flush(struct btrfs_device *device, int wait) 3296387125fcSChris Mason { 3297387125fcSChris Mason struct bio *bio; 3298387125fcSChris Mason int ret = 0; 3299387125fcSChris Mason 3300387125fcSChris Mason if (device->nobarriers) 3301387125fcSChris Mason return 0; 3302387125fcSChris Mason 3303387125fcSChris Mason if (wait) { 3304387125fcSChris Mason bio = device->flush_bio; 3305387125fcSChris Mason if (!bio) 3306387125fcSChris Mason return 0; 3307387125fcSChris Mason 3308387125fcSChris Mason wait_for_completion(&device->flush_wait); 3309387125fcSChris Mason 3310387125fcSChris Mason if (bio_flagged(bio, BIO_EOPNOTSUPP)) { 3311efe120a0SFrank Holton printk_in_rcu("BTRFS: disabling barriers on dev %s\n", 3312606686eeSJosef Bacik rcu_str_deref(device->name)); 3313387125fcSChris Mason device->nobarriers = 1; 33145af3e8ccSStefan Behrens } else if (!bio_flagged(bio, BIO_UPTODATE)) { 3315387125fcSChris Mason ret = -EIO; 3316442a4f63SStefan Behrens btrfs_dev_stat_inc_and_print(device, 3317442a4f63SStefan Behrens BTRFS_DEV_STAT_FLUSH_ERRS); 3318387125fcSChris Mason } 3319387125fcSChris Mason 3320387125fcSChris Mason /* drop the reference from the wait == 0 run */ 3321387125fcSChris Mason bio_put(bio); 3322387125fcSChris Mason device->flush_bio = NULL; 3323387125fcSChris Mason 3324387125fcSChris Mason return ret; 3325387125fcSChris Mason } 3326387125fcSChris Mason 3327387125fcSChris Mason /* 3328387125fcSChris Mason * one reference for us, and we leave it for the 3329387125fcSChris Mason * caller 3330387125fcSChris Mason */ 33319c017abcSJesper Juhl device->flush_bio = NULL; 33329be3395bSChris Mason bio = btrfs_io_bio_alloc(GFP_NOFS, 0); 3333387125fcSChris Mason if (!bio) 3334387125fcSChris Mason return -ENOMEM; 3335387125fcSChris Mason 3336387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 3337387125fcSChris Mason bio->bi_bdev = device->bdev; 3338387125fcSChris Mason init_completion(&device->flush_wait); 3339387125fcSChris Mason bio->bi_private = &device->flush_wait; 3340387125fcSChris Mason device->flush_bio = bio; 3341387125fcSChris Mason 3342387125fcSChris Mason bio_get(bio); 334321adbd5cSStefan Behrens btrfsic_submit_bio(WRITE_FLUSH, bio); 3344387125fcSChris Mason 3345387125fcSChris Mason return 0; 3346387125fcSChris Mason } 3347387125fcSChris Mason 3348387125fcSChris Mason /* 3349387125fcSChris Mason * send an empty flush down to each device in parallel, 3350387125fcSChris Mason * then wait for them 3351387125fcSChris Mason */ 3352387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 3353387125fcSChris Mason { 3354387125fcSChris Mason struct list_head *head; 3355387125fcSChris Mason struct btrfs_device *dev; 33565af3e8ccSStefan Behrens int errors_send = 0; 33575af3e8ccSStefan Behrens int errors_wait = 0; 3358387125fcSChris Mason int ret; 3359387125fcSChris Mason 3360387125fcSChris Mason /* send down all the barriers */ 3361387125fcSChris Mason head = &info->fs_devices->devices; 3362387125fcSChris Mason list_for_each_entry_rcu(dev, head, dev_list) { 3363f88ba6a2SHidetoshi Seto if (dev->missing) 3364f88ba6a2SHidetoshi Seto continue; 3365387125fcSChris Mason if (!dev->bdev) { 33665af3e8ccSStefan Behrens errors_send++; 3367387125fcSChris Mason continue; 3368387125fcSChris Mason } 3369387125fcSChris Mason if (!dev->in_fs_metadata || !dev->writeable) 3370387125fcSChris Mason continue; 3371387125fcSChris Mason 3372387125fcSChris Mason ret = write_dev_flush(dev, 0); 3373387125fcSChris Mason if (ret) 33745af3e8ccSStefan Behrens errors_send++; 3375387125fcSChris Mason } 3376387125fcSChris Mason 3377387125fcSChris Mason /* wait for all the barriers */ 3378387125fcSChris Mason list_for_each_entry_rcu(dev, head, dev_list) { 3379f88ba6a2SHidetoshi Seto if (dev->missing) 3380f88ba6a2SHidetoshi Seto continue; 3381387125fcSChris Mason if (!dev->bdev) { 33825af3e8ccSStefan Behrens errors_wait++; 3383387125fcSChris Mason continue; 3384387125fcSChris Mason } 3385387125fcSChris Mason if (!dev->in_fs_metadata || !dev->writeable) 3386387125fcSChris Mason continue; 3387387125fcSChris Mason 3388387125fcSChris Mason ret = write_dev_flush(dev, 1); 3389387125fcSChris Mason if (ret) 33905af3e8ccSStefan Behrens errors_wait++; 3391387125fcSChris Mason } 33925af3e8ccSStefan Behrens if (errors_send > info->num_tolerated_disk_barrier_failures || 33935af3e8ccSStefan Behrens errors_wait > info->num_tolerated_disk_barrier_failures) 3394387125fcSChris Mason return -EIO; 3395387125fcSChris Mason return 0; 3396387125fcSChris Mason } 3397387125fcSChris Mason 33985af3e8ccSStefan Behrens int btrfs_calc_num_tolerated_disk_barrier_failures( 33995af3e8ccSStefan Behrens struct btrfs_fs_info *fs_info) 34005af3e8ccSStefan Behrens { 34015af3e8ccSStefan Behrens struct btrfs_ioctl_space_info space; 34025af3e8ccSStefan Behrens struct btrfs_space_info *sinfo; 34035af3e8ccSStefan Behrens u64 types[] = {BTRFS_BLOCK_GROUP_DATA, 34045af3e8ccSStefan Behrens BTRFS_BLOCK_GROUP_SYSTEM, 34055af3e8ccSStefan Behrens BTRFS_BLOCK_GROUP_METADATA, 34065af3e8ccSStefan Behrens BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA}; 34075af3e8ccSStefan Behrens int num_types = 4; 34085af3e8ccSStefan Behrens int i; 34095af3e8ccSStefan Behrens int c; 34105af3e8ccSStefan Behrens int num_tolerated_disk_barrier_failures = 34115af3e8ccSStefan Behrens (int)fs_info->fs_devices->num_devices; 34125af3e8ccSStefan Behrens 34135af3e8ccSStefan Behrens for (i = 0; i < num_types; i++) { 34145af3e8ccSStefan Behrens struct btrfs_space_info *tmp; 34155af3e8ccSStefan Behrens 34165af3e8ccSStefan Behrens sinfo = NULL; 34175af3e8ccSStefan Behrens rcu_read_lock(); 34185af3e8ccSStefan Behrens list_for_each_entry_rcu(tmp, &fs_info->space_info, list) { 34195af3e8ccSStefan Behrens if (tmp->flags == types[i]) { 34205af3e8ccSStefan Behrens sinfo = tmp; 34215af3e8ccSStefan Behrens break; 34225af3e8ccSStefan Behrens } 34235af3e8ccSStefan Behrens } 34245af3e8ccSStefan Behrens rcu_read_unlock(); 34255af3e8ccSStefan Behrens 34265af3e8ccSStefan Behrens if (!sinfo) 34275af3e8ccSStefan Behrens continue; 34285af3e8ccSStefan Behrens 34295af3e8ccSStefan Behrens down_read(&sinfo->groups_sem); 34305af3e8ccSStefan Behrens for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 34315af3e8ccSStefan Behrens if (!list_empty(&sinfo->block_groups[c])) { 34325af3e8ccSStefan Behrens u64 flags; 34335af3e8ccSStefan Behrens 34345af3e8ccSStefan Behrens btrfs_get_block_group_info( 34355af3e8ccSStefan Behrens &sinfo->block_groups[c], &space); 34365af3e8ccSStefan Behrens if (space.total_bytes == 0 || 34375af3e8ccSStefan Behrens space.used_bytes == 0) 34385af3e8ccSStefan Behrens continue; 34395af3e8ccSStefan Behrens flags = space.flags; 34405af3e8ccSStefan Behrens /* 34415af3e8ccSStefan Behrens * return 34425af3e8ccSStefan Behrens * 0: if dup, single or RAID0 is configured for 34435af3e8ccSStefan Behrens * any of metadata, system or data, else 34445af3e8ccSStefan Behrens * 1: if RAID5 is configured, or if RAID1 or 34455af3e8ccSStefan Behrens * RAID10 is configured and only two mirrors 34465af3e8ccSStefan Behrens * are used, else 34475af3e8ccSStefan Behrens * 2: if RAID6 is configured, else 34485af3e8ccSStefan Behrens * num_mirrors - 1: if RAID1 or RAID10 is 34495af3e8ccSStefan Behrens * configured and more than 34505af3e8ccSStefan Behrens * 2 mirrors are used. 34515af3e8ccSStefan Behrens */ 34525af3e8ccSStefan Behrens if (num_tolerated_disk_barrier_failures > 0 && 34535af3e8ccSStefan Behrens ((flags & (BTRFS_BLOCK_GROUP_DUP | 34545af3e8ccSStefan Behrens BTRFS_BLOCK_GROUP_RAID0)) || 34555af3e8ccSStefan Behrens ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) 34565af3e8ccSStefan Behrens == 0))) 34575af3e8ccSStefan Behrens num_tolerated_disk_barrier_failures = 0; 345853b381b3SDavid Woodhouse else if (num_tolerated_disk_barrier_failures > 1) { 345953b381b3SDavid Woodhouse if (flags & (BTRFS_BLOCK_GROUP_RAID1 | 346053b381b3SDavid Woodhouse BTRFS_BLOCK_GROUP_RAID5 | 346153b381b3SDavid Woodhouse BTRFS_BLOCK_GROUP_RAID10)) { 34625af3e8ccSStefan Behrens num_tolerated_disk_barrier_failures = 1; 346353b381b3SDavid Woodhouse } else if (flags & 346415b0a89dSHenrik Nordvik BTRFS_BLOCK_GROUP_RAID6) { 346553b381b3SDavid Woodhouse num_tolerated_disk_barrier_failures = 2; 346653b381b3SDavid Woodhouse } 346753b381b3SDavid Woodhouse } 34685af3e8ccSStefan Behrens } 34695af3e8ccSStefan Behrens } 34705af3e8ccSStefan Behrens up_read(&sinfo->groups_sem); 34715af3e8ccSStefan Behrens } 34725af3e8ccSStefan Behrens 34735af3e8ccSStefan Behrens return num_tolerated_disk_barrier_failures; 34745af3e8ccSStefan Behrens } 34755af3e8ccSStefan Behrens 347648a3b636SEric Sandeen static int write_all_supers(struct btrfs_root *root, int max_mirrors) 3477f2984462SChris Mason { 3478e5e9a520SChris Mason struct list_head *head; 3479f2984462SChris Mason struct btrfs_device *dev; 3480a061fc8dSChris Mason struct btrfs_super_block *sb; 3481f2984462SChris Mason struct btrfs_dev_item *dev_item; 3482f2984462SChris Mason int ret; 3483f2984462SChris Mason int do_barriers; 3484a236aed1SChris Mason int max_errors; 3485a236aed1SChris Mason int total_errors = 0; 3486a061fc8dSChris Mason u64 flags; 3487f2984462SChris Mason 3488f2984462SChris Mason do_barriers = !btrfs_test_opt(root, NOBARRIER); 3489af31f5e5SChris Mason backup_super_roots(root->fs_info); 3490f2984462SChris Mason 34916c41761fSDavid Sterba sb = root->fs_info->super_for_commit; 3492a061fc8dSChris Mason dev_item = &sb->dev_item; 3493e5e9a520SChris Mason 3494174ba509SChris Mason mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 3495e5e9a520SChris Mason head = &root->fs_info->fs_devices->devices; 3496d7306801SFilipe David Borba Manana max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1; 3497387125fcSChris Mason 34985af3e8ccSStefan Behrens if (do_barriers) { 34995af3e8ccSStefan Behrens ret = barrier_all_devices(root->fs_info); 35005af3e8ccSStefan Behrens if (ret) { 35015af3e8ccSStefan Behrens mutex_unlock( 35025af3e8ccSStefan Behrens &root->fs_info->fs_devices->device_list_mutex); 35035af3e8ccSStefan Behrens btrfs_error(root->fs_info, ret, 35045af3e8ccSStefan Behrens "errors while submitting device barriers."); 35055af3e8ccSStefan Behrens return ret; 35065af3e8ccSStefan Behrens } 35075af3e8ccSStefan Behrens } 3508387125fcSChris Mason 35091f78160cSXiao Guangrong list_for_each_entry_rcu(dev, head, dev_list) { 3510dfe25020SChris Mason if (!dev->bdev) { 3511dfe25020SChris Mason total_errors++; 3512dfe25020SChris Mason continue; 3513dfe25020SChris Mason } 35142b82032cSYan Zheng if (!dev->in_fs_metadata || !dev->writeable) 3515dfe25020SChris Mason continue; 3516dfe25020SChris Mason 35172b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 3518a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 3519a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 35207df69d3eSMiao Xie btrfs_set_stack_device_total_bytes(dev_item, 3521935e5cc9SMiao Xie dev->commit_total_bytes); 3522ce7213c7SMiao Xie btrfs_set_stack_device_bytes_used(dev_item, 3523ce7213c7SMiao Xie dev->commit_bytes_used); 3524a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 3525a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 3526a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 3527a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 35282b82032cSYan Zheng memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE); 3529a512bbf8SYan Zheng 3530a061fc8dSChris Mason flags = btrfs_super_flags(sb); 3531a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 3532f2984462SChris Mason 3533a512bbf8SYan Zheng ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors); 3534a236aed1SChris Mason if (ret) 3535a236aed1SChris Mason total_errors++; 3536f2984462SChris Mason } 3537a236aed1SChris Mason if (total_errors > max_errors) { 3538efe120a0SFrank Holton btrfs_err(root->fs_info, "%d errors while writing supers", 3539d397712bSChris Mason total_errors); 3540a724b436SStefan Behrens mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 354179787eaaSJeff Mahoney 35429d565ba4SStefan Behrens /* FUA is masked off if unsupported and can't be the reason */ 35439d565ba4SStefan Behrens btrfs_error(root->fs_info, -EIO, 35449d565ba4SStefan Behrens "%d errors while writing supers", total_errors); 35459d565ba4SStefan Behrens return -EIO; 3546a236aed1SChris Mason } 3547f2984462SChris Mason 3548a512bbf8SYan Zheng total_errors = 0; 35491f78160cSXiao Guangrong list_for_each_entry_rcu(dev, head, dev_list) { 3550dfe25020SChris Mason if (!dev->bdev) 3551dfe25020SChris Mason continue; 35522b82032cSYan Zheng if (!dev->in_fs_metadata || !dev->writeable) 3553dfe25020SChris Mason continue; 3554dfe25020SChris Mason 3555a512bbf8SYan Zheng ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors); 3556a512bbf8SYan Zheng if (ret) 35571259ab75SChris Mason total_errors++; 3558f2984462SChris Mason } 3559174ba509SChris Mason mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 3560a236aed1SChris Mason if (total_errors > max_errors) { 356179787eaaSJeff Mahoney btrfs_error(root->fs_info, -EIO, 356279787eaaSJeff Mahoney "%d errors while writing supers", total_errors); 356379787eaaSJeff Mahoney return -EIO; 3564a236aed1SChris Mason } 3565f2984462SChris Mason return 0; 3566f2984462SChris Mason } 3567f2984462SChris Mason 3568a512bbf8SYan Zheng int write_ctree_super(struct btrfs_trans_handle *trans, 3569a512bbf8SYan Zheng struct btrfs_root *root, int max_mirrors) 3570cfaa7295SChris Mason { 3571f570e757SRashika return write_all_supers(root, max_mirrors); 3572cfaa7295SChris Mason } 3573cfaa7295SChris Mason 3574cb517eabSMiao Xie /* Drop a fs root from the radix tree and free it. */ 3575cb517eabSMiao Xie void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 3576cb517eabSMiao Xie struct btrfs_root *root) 35772619ba1fSChris Mason { 35784df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 35792619ba1fSChris Mason radix_tree_delete(&fs_info->fs_roots_radix, 35802619ba1fSChris Mason (unsigned long)root->root_key.objectid); 35814df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 358276dda93cSYan, Zheng 358376dda93cSYan, Zheng if (btrfs_root_refs(&root->root_item) == 0) 358476dda93cSYan, Zheng synchronize_srcu(&fs_info->subvol_srcu); 358576dda93cSYan, Zheng 35861a4319ccSLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) 35873321719eSLiu Bo btrfs_free_log(NULL, root); 35883321719eSLiu Bo 3589faa2dbf0SJosef Bacik if (root->free_ino_pinned) 3590581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_pinned); 3591faa2dbf0SJosef Bacik if (root->free_ino_ctl) 3592581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_ctl); 35934df27c4dSYan, Zheng free_fs_root(root); 35944df27c4dSYan, Zheng } 35954df27c4dSYan, Zheng 35964df27c4dSYan, Zheng static void free_fs_root(struct btrfs_root *root) 35974df27c4dSYan, Zheng { 359857cdc8dbSDavid Sterba iput(root->ino_cache_inode); 35994df27c4dSYan, Zheng WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 36001cb048f5SFilipe David Borba Manana btrfs_free_block_rsv(root, root->orphan_block_rsv); 36011cb048f5SFilipe David Borba Manana root->orphan_block_rsv = NULL; 36020ee5dc67SAl Viro if (root->anon_dev) 36030ee5dc67SAl Viro free_anon_bdev(root->anon_dev); 36048257b2dcSMiao Xie if (root->subv_writers) 36058257b2dcSMiao Xie btrfs_free_subvolume_writers(root->subv_writers); 36065f39d397SChris Mason free_extent_buffer(root->node); 36075f39d397SChris Mason free_extent_buffer(root->commit_root); 3608581bb050SLi Zefan kfree(root->free_ino_ctl); 3609581bb050SLi Zefan kfree(root->free_ino_pinned); 361058176a96SJosef Bacik kfree(root->name); 3611b0feb9d9SMiao Xie btrfs_put_fs_root(root); 36122619ba1fSChris Mason } 36132619ba1fSChris Mason 3614cb517eabSMiao Xie void btrfs_free_fs_root(struct btrfs_root *root) 3615cb517eabSMiao Xie { 3616cb517eabSMiao Xie free_fs_root(root); 3617eb60ceacSChris Mason } 3618eb60ceacSChris Mason 3619c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 3620c146afadSYan Zheng { 3621c146afadSYan Zheng u64 root_objectid = 0; 3622c146afadSYan Zheng struct btrfs_root *gang[8]; 362365d33fd7SQu Wenruo int i = 0; 362465d33fd7SQu Wenruo int err = 0; 362565d33fd7SQu Wenruo unsigned int ret = 0; 362665d33fd7SQu Wenruo int index; 3627c146afadSYan Zheng 3628c146afadSYan Zheng while (1) { 362965d33fd7SQu Wenruo index = srcu_read_lock(&fs_info->subvol_srcu); 3630c146afadSYan Zheng ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 3631c146afadSYan Zheng (void **)gang, root_objectid, 3632c146afadSYan Zheng ARRAY_SIZE(gang)); 363365d33fd7SQu Wenruo if (!ret) { 363465d33fd7SQu Wenruo srcu_read_unlock(&fs_info->subvol_srcu, index); 3635c146afadSYan Zheng break; 363665d33fd7SQu Wenruo } 36375d4f98a2SYan Zheng root_objectid = gang[ret - 1]->root_key.objectid + 1; 363866b4ffd1SJosef Bacik 363965d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 364065d33fd7SQu Wenruo /* Avoid to grab roots in dead_roots */ 364165d33fd7SQu Wenruo if (btrfs_root_refs(&gang[i]->root_item) == 0) { 364265d33fd7SQu Wenruo gang[i] = NULL; 364365d33fd7SQu Wenruo continue; 364465d33fd7SQu Wenruo } 364565d33fd7SQu Wenruo /* grab all the search result for later use */ 364665d33fd7SQu Wenruo gang[i] = btrfs_grab_fs_root(gang[i]); 364765d33fd7SQu Wenruo } 364865d33fd7SQu Wenruo srcu_read_unlock(&fs_info->subvol_srcu, index); 364965d33fd7SQu Wenruo 365065d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 365165d33fd7SQu Wenruo if (!gang[i]) 365265d33fd7SQu Wenruo continue; 3653c146afadSYan Zheng root_objectid = gang[i]->root_key.objectid; 365466b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(gang[i]); 365566b4ffd1SJosef Bacik if (err) 365665d33fd7SQu Wenruo break; 365765d33fd7SQu Wenruo btrfs_put_fs_root(gang[i]); 3658c146afadSYan Zheng } 3659c146afadSYan Zheng root_objectid++; 3660c146afadSYan Zheng } 366165d33fd7SQu Wenruo 366265d33fd7SQu Wenruo /* release the uncleaned roots due to error */ 366365d33fd7SQu Wenruo for (; i < ret; i++) { 366465d33fd7SQu Wenruo if (gang[i]) 366565d33fd7SQu Wenruo btrfs_put_fs_root(gang[i]); 366665d33fd7SQu Wenruo } 366765d33fd7SQu Wenruo return err; 3668c146afadSYan Zheng } 3669c146afadSYan Zheng 3670c146afadSYan Zheng int btrfs_commit_super(struct btrfs_root *root) 3671c146afadSYan Zheng { 3672c146afadSYan Zheng struct btrfs_trans_handle *trans; 3673c146afadSYan Zheng 3674c146afadSYan Zheng mutex_lock(&root->fs_info->cleaner_mutex); 367524bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 3676c146afadSYan Zheng mutex_unlock(&root->fs_info->cleaner_mutex); 36779d1a2a3aSDavid Sterba wake_up_process(root->fs_info->cleaner_kthread); 3678c71bf099SYan, Zheng 3679c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 3680c71bf099SYan, Zheng down_write(&root->fs_info->cleanup_work_sem); 3681c71bf099SYan, Zheng up_write(&root->fs_info->cleanup_work_sem); 3682c71bf099SYan, Zheng 36837a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 36843612b495STsutomu Itoh if (IS_ERR(trans)) 36853612b495STsutomu Itoh return PTR_ERR(trans); 3686d52c1bccSLiu Bo return btrfs_commit_transaction(trans, root); 3687c146afadSYan Zheng } 3688c146afadSYan Zheng 36893abdbd78SDavid Sterba void close_ctree(struct btrfs_root *root) 3690eb60ceacSChris Mason { 36910f7d52f4SChris Mason struct btrfs_fs_info *fs_info = root->fs_info; 3692c146afadSYan Zheng int ret; 3693e089f05cSChris Mason 3694facda1e7SChris Mason fs_info->closing = 1; 3695a2135011SChris Mason smp_mb(); 3696a2135011SChris Mason 3697803b2f54SStefan Behrens /* wait for the uuid_scan task to finish */ 3698803b2f54SStefan Behrens down(&fs_info->uuid_tree_rescan_sem); 3699803b2f54SStefan Behrens /* avoid complains from lockdep et al., set sem back to initial state */ 3700803b2f54SStefan Behrens up(&fs_info->uuid_tree_rescan_sem); 3701803b2f54SStefan Behrens 3702837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 3703aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 3704837d5b6eSIlya Dryomov 37058dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 37068dabb742SStefan Behrens 3707aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 37084cb5300bSChris Mason 37094cb5300bSChris Mason /* wait for any defraggers to finish */ 37104cb5300bSChris Mason wait_event(fs_info->transaction_wait, 37114cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 37124cb5300bSChris Mason 37134cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 371426176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 37154cb5300bSChris Mason 371621c7e756SMiao Xie cancel_work_sync(&fs_info->async_reclaim_work); 371721c7e756SMiao Xie 3718c146afadSYan Zheng if (!(fs_info->sb->s_flags & MS_RDONLY)) { 3719c146afadSYan Zheng ret = btrfs_commit_super(root); 3720d397712bSChris Mason if (ret) 372104892340SEric Sandeen btrfs_err(fs_info, "commit super ret %d", ret); 3722c146afadSYan Zheng } 3723ed2ff2cbSChris Mason 372487533c47SMiao Xie if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) 372568ce9682SStefan Behrens btrfs_error_commit_super(root); 3726acce952bSliubo 3727e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 3728e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 37298929ecfaSYan, Zheng 3730f25784b3SYan Zheng fs_info->closing = 2; 3731f25784b3SYan Zheng smp_mb(); 3732f25784b3SYan Zheng 373304892340SEric Sandeen btrfs_free_qgroup_config(fs_info); 3734bcef60f2SArne Jansen 3735963d678bSMiao Xie if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 373604892340SEric Sandeen btrfs_info(fs_info, "at unmount delalloc count %lld", 3737963d678bSMiao Xie percpu_counter_sum(&fs_info->delalloc_bytes)); 3738b0c68f8bSChris Mason } 373931153d81SYan Zheng 37405ac1d209SJeff Mahoney btrfs_sysfs_remove_one(fs_info); 37415ac1d209SJeff Mahoney 3742faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 3743d10c5f31SChris Mason 37441a4319ccSLiu Bo btrfs_put_block_group_cache(fs_info); 37451a4319ccSLiu Bo 37462b1360daSJosef Bacik btrfs_free_block_groups(fs_info); 37472b1360daSJosef Bacik 3748de348ee0SWang Shilong /* 3749de348ee0SWang Shilong * we must make sure there is not any read request to 3750de348ee0SWang Shilong * submit after we stopping all workers. 3751de348ee0SWang Shilong */ 3752de348ee0SWang Shilong invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 375396192499SJosef Bacik btrfs_stop_all_workers(fs_info); 375496192499SJosef Bacik 375547ab2a6cSJosef Bacik fs_info->open = 0; 375613e6c37bSJosef Bacik free_root_pointers(fs_info, 1); 37579ad6b7bcSChris Mason 375813e6c37bSJosef Bacik iput(fs_info->btree_inode); 3759d6bfde87SChris Mason 376021adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 376121adbd5cSStefan Behrens if (btrfs_test_opt(root, CHECK_INTEGRITY)) 376221adbd5cSStefan Behrens btrfsic_unmount(root, fs_info->fs_devices); 376321adbd5cSStefan Behrens #endif 376421adbd5cSStefan Behrens 3765dfe25020SChris Mason btrfs_close_devices(fs_info->fs_devices); 37660b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 3767b248a415SChris Mason 3768e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 3769963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 3770c404e0dcSMiao Xie percpu_counter_destroy(&fs_info->bio_counter); 377104160088SChris Mason bdi_destroy(&fs_info->bdi); 377276dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 37730b86a832SChris Mason 377453b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 377553b381b3SDavid Woodhouse 37761cb048f5SFilipe David Borba Manana btrfs_free_block_rsv(root, root->orphan_block_rsv); 37771cb048f5SFilipe David Borba Manana root->orphan_block_rsv = NULL; 377804216820SFilipe Manana 377904216820SFilipe Manana lock_chunks(root); 378004216820SFilipe Manana while (!list_empty(&fs_info->pinned_chunks)) { 378104216820SFilipe Manana struct extent_map *em; 378204216820SFilipe Manana 378304216820SFilipe Manana em = list_first_entry(&fs_info->pinned_chunks, 378404216820SFilipe Manana struct extent_map, list); 378504216820SFilipe Manana list_del_init(&em->list); 378604216820SFilipe Manana free_extent_map(em); 378704216820SFilipe Manana } 378804216820SFilipe Manana unlock_chunks(root); 3789eb60ceacSChris Mason } 3790eb60ceacSChris Mason 3791b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 3792b9fab919SChris Mason int atomic) 3793ccd467d6SChris Mason { 37941259ab75SChris Mason int ret; 3795727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 37961259ab75SChris Mason 37970b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 37981259ab75SChris Mason if (!ret) 37991259ab75SChris Mason return ret; 38001259ab75SChris Mason 38011259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 3802b9fab919SChris Mason parent_transid, atomic); 3803b9fab919SChris Mason if (ret == -EAGAIN) 3804b9fab919SChris Mason return ret; 38051259ab75SChris Mason return !ret; 38065f39d397SChris Mason } 38076702ed49SChris Mason 38085f39d397SChris Mason int btrfs_set_buffer_uptodate(struct extent_buffer *buf) 38095f39d397SChris Mason { 38100b32f4bbSJosef Bacik return set_extent_buffer_uptodate(buf); 38115f39d397SChris Mason } 38125f39d397SChris Mason 38135f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 38145f39d397SChris Mason { 381506ea65a3SJosef Bacik struct btrfs_root *root; 38165f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 3817b9473439SChris Mason int was_dirty; 3818b4ce94deSChris Mason 381906ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 382006ea65a3SJosef Bacik /* 382106ea65a3SJosef Bacik * This is a fast path so only do this check if we have sanity tests 382206ea65a3SJosef Bacik * enabled. Normal people shouldn't be marking dummy buffers as dirty 382306ea65a3SJosef Bacik * outside of the sanity tests. 382406ea65a3SJosef Bacik */ 382506ea65a3SJosef Bacik if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags))) 382606ea65a3SJosef Bacik return; 382706ea65a3SJosef Bacik #endif 382806ea65a3SJosef Bacik root = BTRFS_I(buf->pages[0]->mapping->host)->root; 3829b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 383031b1a2bdSJulia Lawall if (transid != root->fs_info->generation) 383131b1a2bdSJulia Lawall WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, " 3832d397712bSChris Mason "found %llu running %llu\n", 3833c1c9ff7cSGeert Uytterhoeven buf->start, transid, root->fs_info->generation); 38340b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 3835e2d84521SMiao Xie if (!was_dirty) 3836e2d84521SMiao Xie __percpu_counter_add(&root->fs_info->dirty_metadata_bytes, 3837e2d84521SMiao Xie buf->len, 3838e2d84521SMiao Xie root->fs_info->dirty_metadata_batch); 38391f21ef0aSFilipe Manana #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 38401f21ef0aSFilipe Manana if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) { 38411f21ef0aSFilipe Manana btrfs_print_leaf(root, buf); 38421f21ef0aSFilipe Manana ASSERT(0); 38431f21ef0aSFilipe Manana } 38441f21ef0aSFilipe Manana #endif 3845eb60ceacSChris Mason } 3846eb60ceacSChris Mason 3847b53d3f5dSLiu Bo static void __btrfs_btree_balance_dirty(struct btrfs_root *root, 3848b53d3f5dSLiu Bo int flush_delayed) 384935b7e476SChris Mason { 3850188de649SChris Mason /* 3851188de649SChris Mason * looks as though older kernels can get into trouble with 3852188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 3853188de649SChris Mason */ 3854e2d84521SMiao Xie int ret; 3855d6bfde87SChris Mason 38566933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 3857d6bfde87SChris Mason return; 3858d6bfde87SChris Mason 3859b53d3f5dSLiu Bo if (flush_delayed) 386016cdcec7SMiao Xie btrfs_balance_delayed_items(root); 386116cdcec7SMiao Xie 3862e2d84521SMiao Xie ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes, 3863e2d84521SMiao Xie BTRFS_DIRTY_METADATA_THRESH); 3864e2d84521SMiao Xie if (ret > 0) { 3865d0e1d66bSNamjae Jeon balance_dirty_pages_ratelimited( 3866d0e1d66bSNamjae Jeon root->fs_info->btree_inode->i_mapping); 386716cdcec7SMiao Xie } 386816cdcec7SMiao Xie return; 386916cdcec7SMiao Xie } 387016cdcec7SMiao Xie 3871b53d3f5dSLiu Bo void btrfs_btree_balance_dirty(struct btrfs_root *root) 387216cdcec7SMiao Xie { 3873b53d3f5dSLiu Bo __btrfs_btree_balance_dirty(root, 1); 387435b7e476SChris Mason } 3875b53d3f5dSLiu Bo 3876b53d3f5dSLiu Bo void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root) 3877b53d3f5dSLiu Bo { 3878b53d3f5dSLiu Bo __btrfs_btree_balance_dirty(root, 0); 3879d6bfde87SChris Mason } 38806b80053dSChris Mason 3881ca7a79adSChris Mason int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid) 38826b80053dSChris Mason { 3883727011e0SChris Mason struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root; 38840b32f4bbSJosef Bacik return btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 38856b80053dSChris Mason } 38860da5468fSChris Mason 3887fcd1f065SDavid Sterba static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info, 3888acce952bSliubo int read_only) 3889acce952bSliubo { 3890c926093eSDavid Sterba struct btrfs_super_block *sb = fs_info->super_copy; 3891c926093eSDavid Sterba int ret = 0; 3892c926093eSDavid Sterba 389321e7626bSDavid Sterba if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 389421e7626bSDavid Sterba printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n", 389521e7626bSDavid Sterba btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 3896c926093eSDavid Sterba ret = -EINVAL; 3897c926093eSDavid Sterba } 389821e7626bSDavid Sterba if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 389921e7626bSDavid Sterba printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n", 390021e7626bSDavid Sterba btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 3901c926093eSDavid Sterba ret = -EINVAL; 3902c926093eSDavid Sterba } 390321e7626bSDavid Sterba if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 390421e7626bSDavid Sterba printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n", 390521e7626bSDavid Sterba btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 3906c926093eSDavid Sterba ret = -EINVAL; 3907c926093eSDavid Sterba } 3908c926093eSDavid Sterba 39091104a885SDavid Sterba /* 3910c926093eSDavid Sterba * The common minimum, we don't know if we can trust the nodesize/sectorsize 3911c926093eSDavid Sterba * items yet, they'll be verified later. Issue just a warning. 39121104a885SDavid Sterba */ 391321e7626bSDavid Sterba if (!IS_ALIGNED(btrfs_super_root(sb), 4096)) 3914c926093eSDavid Sterba printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n", 3915cd743facSDavid Sterba btrfs_super_root(sb)); 391621e7626bSDavid Sterba if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096)) 3917cd743facSDavid Sterba printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n", 3918cd743facSDavid Sterba btrfs_super_chunk_root(sb)); 391921e7626bSDavid Sterba if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096)) 3920cd743facSDavid Sterba printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n", 392121e7626bSDavid Sterba btrfs_super_log_root(sb)); 3922c926093eSDavid Sterba 392375d6ad38SDavid Sterba /* 392475d6ad38SDavid Sterba * Check the lower bound, the alignment and other constraints are 392575d6ad38SDavid Sterba * checked later. 392675d6ad38SDavid Sterba */ 392775d6ad38SDavid Sterba if (btrfs_super_nodesize(sb) < 4096) { 392875d6ad38SDavid Sterba printk(KERN_ERR "BTRFS: nodesize too small: %u < 4096\n", 392975d6ad38SDavid Sterba btrfs_super_nodesize(sb)); 393075d6ad38SDavid Sterba ret = -EINVAL; 393175d6ad38SDavid Sterba } 393275d6ad38SDavid Sterba if (btrfs_super_sectorsize(sb) < 4096) { 393375d6ad38SDavid Sterba printk(KERN_ERR "BTRFS: sectorsize too small: %u < 4096\n", 393475d6ad38SDavid Sterba btrfs_super_sectorsize(sb)); 393575d6ad38SDavid Sterba ret = -EINVAL; 393675d6ad38SDavid Sterba } 393775d6ad38SDavid Sterba 3938c926093eSDavid Sterba if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) { 3939c926093eSDavid Sterba printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n", 3940c926093eSDavid Sterba fs_info->fsid, sb->dev_item.fsid); 3941c926093eSDavid Sterba ret = -EINVAL; 3942c926093eSDavid Sterba } 3943c926093eSDavid Sterba 3944c926093eSDavid Sterba /* 3945c926093eSDavid Sterba * Hint to catch really bogus numbers, bitflips or so, more exact checks are 3946c926093eSDavid Sterba * done later 3947c926093eSDavid Sterba */ 394821e7626bSDavid Sterba if (btrfs_super_num_devices(sb) > (1UL << 31)) 3949c926093eSDavid Sterba printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n", 395021e7626bSDavid Sterba btrfs_super_num_devices(sb)); 395175d6ad38SDavid Sterba if (btrfs_super_num_devices(sb) == 0) { 395275d6ad38SDavid Sterba printk(KERN_ERR "BTRFS: number of devices is 0\n"); 395375d6ad38SDavid Sterba ret = -EINVAL; 395475d6ad38SDavid Sterba } 3955c926093eSDavid Sterba 395621e7626bSDavid Sterba if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) { 3957c926093eSDavid Sterba printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n", 395821e7626bSDavid Sterba btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 3959c926093eSDavid Sterba ret = -EINVAL; 3960c926093eSDavid Sterba } 3961c926093eSDavid Sterba 3962c926093eSDavid Sterba /* 3963ce7fca5fSDavid Sterba * Obvious sys_chunk_array corruptions, it must hold at least one key 3964ce7fca5fSDavid Sterba * and one chunk 3965ce7fca5fSDavid Sterba */ 3966ce7fca5fSDavid Sterba if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 3967ce7fca5fSDavid Sterba printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n", 3968ce7fca5fSDavid Sterba btrfs_super_sys_array_size(sb), 3969ce7fca5fSDavid Sterba BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 3970ce7fca5fSDavid Sterba ret = -EINVAL; 3971ce7fca5fSDavid Sterba } 3972ce7fca5fSDavid Sterba if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 3973ce7fca5fSDavid Sterba + sizeof(struct btrfs_chunk)) { 397441d6b13eSFabian Frederick printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n", 3975ce7fca5fSDavid Sterba btrfs_super_sys_array_size(sb), 3976ce7fca5fSDavid Sterba sizeof(struct btrfs_disk_key) 3977ce7fca5fSDavid Sterba + sizeof(struct btrfs_chunk)); 3978ce7fca5fSDavid Sterba ret = -EINVAL; 3979ce7fca5fSDavid Sterba } 3980ce7fca5fSDavid Sterba 3981ce7fca5fSDavid Sterba /* 3982c926093eSDavid Sterba * The generation is a global counter, we'll trust it more than the others 3983c926093eSDavid Sterba * but it's still possible that it's the one that's wrong. 3984c926093eSDavid Sterba */ 398521e7626bSDavid Sterba if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 3986c926093eSDavid Sterba printk(KERN_WARNING 3987c926093eSDavid Sterba "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n", 398821e7626bSDavid Sterba btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb)); 398921e7626bSDavid Sterba if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 399021e7626bSDavid Sterba && btrfs_super_cache_generation(sb) != (u64)-1) 3991c926093eSDavid Sterba printk(KERN_WARNING 3992c926093eSDavid Sterba "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n", 399321e7626bSDavid Sterba btrfs_super_generation(sb), btrfs_super_cache_generation(sb)); 3994c926093eSDavid Sterba 3995c926093eSDavid Sterba return ret; 3996fcd1f065SDavid Sterba } 3997fcd1f065SDavid Sterba 399848a3b636SEric Sandeen static void btrfs_error_commit_super(struct btrfs_root *root) 3999acce952bSliubo { 4000acce952bSliubo mutex_lock(&root->fs_info->cleaner_mutex); 4001acce952bSliubo btrfs_run_delayed_iputs(root); 4002acce952bSliubo mutex_unlock(&root->fs_info->cleaner_mutex); 4003acce952bSliubo 4004acce952bSliubo down_write(&root->fs_info->cleanup_work_sem); 4005acce952bSliubo up_write(&root->fs_info->cleanup_work_sem); 4006acce952bSliubo 4007acce952bSliubo /* cleanup FS via transaction */ 4008acce952bSliubo btrfs_cleanup_transaction(root); 4009acce952bSliubo } 4010acce952bSliubo 4011143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4012acce952bSliubo { 4013acce952bSliubo struct btrfs_ordered_extent *ordered; 4014acce952bSliubo 4015199c2a9cSMiao Xie spin_lock(&root->ordered_extent_lock); 4016779880efSJosef Bacik /* 4017779880efSJosef Bacik * This will just short circuit the ordered completion stuff which will 4018779880efSJosef Bacik * make sure the ordered extent gets properly cleaned up. 4019779880efSJosef Bacik */ 4020199c2a9cSMiao Xie list_for_each_entry(ordered, &root->ordered_extents, 4021779880efSJosef Bacik root_extent_list) 4022779880efSJosef Bacik set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4023199c2a9cSMiao Xie spin_unlock(&root->ordered_extent_lock); 4024199c2a9cSMiao Xie } 4025199c2a9cSMiao Xie 4026199c2a9cSMiao Xie static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4027199c2a9cSMiao Xie { 4028199c2a9cSMiao Xie struct btrfs_root *root; 4029199c2a9cSMiao Xie struct list_head splice; 4030199c2a9cSMiao Xie 4031199c2a9cSMiao Xie INIT_LIST_HEAD(&splice); 4032199c2a9cSMiao Xie 4033199c2a9cSMiao Xie spin_lock(&fs_info->ordered_root_lock); 4034199c2a9cSMiao Xie list_splice_init(&fs_info->ordered_roots, &splice); 4035199c2a9cSMiao Xie while (!list_empty(&splice)) { 4036199c2a9cSMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4037199c2a9cSMiao Xie ordered_root); 40381de2cfdeSJosef Bacik list_move_tail(&root->ordered_root, 40391de2cfdeSJosef Bacik &fs_info->ordered_roots); 4040199c2a9cSMiao Xie 40412a85d9caSLiu Bo spin_unlock(&fs_info->ordered_root_lock); 4042199c2a9cSMiao Xie btrfs_destroy_ordered_extents(root); 4043199c2a9cSMiao Xie 40442a85d9caSLiu Bo cond_resched(); 40452a85d9caSLiu Bo spin_lock(&fs_info->ordered_root_lock); 4046199c2a9cSMiao Xie } 4047199c2a9cSMiao Xie spin_unlock(&fs_info->ordered_root_lock); 4048acce952bSliubo } 4049acce952bSliubo 405035a3621bSStefan Behrens static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 4051acce952bSliubo struct btrfs_root *root) 4052acce952bSliubo { 4053acce952bSliubo struct rb_node *node; 4054acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 4055acce952bSliubo struct btrfs_delayed_ref_node *ref; 4056acce952bSliubo int ret = 0; 4057acce952bSliubo 4058acce952bSliubo delayed_refs = &trans->delayed_refs; 4059acce952bSliubo 4060acce952bSliubo spin_lock(&delayed_refs->lock); 4061d7df2c79SJosef Bacik if (atomic_read(&delayed_refs->num_entries) == 0) { 4062cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 4063efe120a0SFrank Holton btrfs_info(root->fs_info, "delayed_refs has NO entry"); 4064acce952bSliubo return ret; 4065acce952bSliubo } 4066acce952bSliubo 4067d7df2c79SJosef Bacik while ((node = rb_first(&delayed_refs->href_root)) != NULL) { 4068d7df2c79SJosef Bacik struct btrfs_delayed_ref_head *head; 4069e78417d1SJosef Bacik bool pin_bytes = false; 4070acce952bSliubo 4071d7df2c79SJosef Bacik head = rb_entry(node, struct btrfs_delayed_ref_head, 4072d7df2c79SJosef Bacik href_node); 4073b939d1abSJosef Bacik if (!mutex_trylock(&head->mutex)) { 4074d7df2c79SJosef Bacik atomic_inc(&head->node.refs); 407579787eaaSJeff Mahoney spin_unlock(&delayed_refs->lock); 4076b939d1abSJosef Bacik 4077acce952bSliubo mutex_lock(&head->mutex); 4078b939d1abSJosef Bacik mutex_unlock(&head->mutex); 4079d7df2c79SJosef Bacik btrfs_put_delayed_ref(&head->node); 4080e18fca73SJosef Bacik spin_lock(&delayed_refs->lock); 4081b939d1abSJosef Bacik continue; 4082b939d1abSJosef Bacik } 4083d7df2c79SJosef Bacik spin_lock(&head->lock); 4084d7df2c79SJosef Bacik while ((node = rb_first(&head->ref_root)) != NULL) { 4085d7df2c79SJosef Bacik ref = rb_entry(node, struct btrfs_delayed_ref_node, 4086d7df2c79SJosef Bacik rb_node); 4087d7df2c79SJosef Bacik ref->in_tree = 0; 4088d7df2c79SJosef Bacik rb_erase(&ref->rb_node, &head->ref_root); 4089d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4090d7df2c79SJosef Bacik btrfs_put_delayed_ref(ref); 4091d7df2c79SJosef Bacik } 409254067ae9SJosef Bacik if (head->must_insert_reserved) 4093e78417d1SJosef Bacik pin_bytes = true; 409478a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 4095acce952bSliubo delayed_refs->num_heads--; 4096d7df2c79SJosef Bacik if (head->processing == 0) 4097acce952bSliubo delayed_refs->num_heads_ready--; 4098d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4099d7df2c79SJosef Bacik head->node.in_tree = 0; 4100c46effa6SLiu Bo rb_erase(&head->href_node, &delayed_refs->href_root); 4101d7df2c79SJosef Bacik spin_unlock(&head->lock); 4102acce952bSliubo spin_unlock(&delayed_refs->lock); 4103e78417d1SJosef Bacik mutex_unlock(&head->mutex); 4104acce952bSliubo 4105d7df2c79SJosef Bacik if (pin_bytes) 4106d7df2c79SJosef Bacik btrfs_pin_extent(root, head->node.bytenr, 4107d7df2c79SJosef Bacik head->node.num_bytes, 1); 4108d7df2c79SJosef Bacik btrfs_put_delayed_ref(&head->node); 4109acce952bSliubo cond_resched(); 4110acce952bSliubo spin_lock(&delayed_refs->lock); 4111acce952bSliubo } 4112acce952bSliubo 4113acce952bSliubo spin_unlock(&delayed_refs->lock); 4114acce952bSliubo 4115acce952bSliubo return ret; 4116acce952bSliubo } 4117acce952bSliubo 4118143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 4119acce952bSliubo { 4120acce952bSliubo struct btrfs_inode *btrfs_inode; 4121acce952bSliubo struct list_head splice; 4122acce952bSliubo 4123acce952bSliubo INIT_LIST_HEAD(&splice); 4124acce952bSliubo 4125eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4126eb73c1b7SMiao Xie list_splice_init(&root->delalloc_inodes, &splice); 4127acce952bSliubo 4128acce952bSliubo while (!list_empty(&splice)) { 4129eb73c1b7SMiao Xie btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 4130acce952bSliubo delalloc_inodes); 4131acce952bSliubo 4132acce952bSliubo list_del_init(&btrfs_inode->delalloc_inodes); 4133df0af1a5SMiao Xie clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, 4134df0af1a5SMiao Xie &btrfs_inode->runtime_flags); 4135eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4136acce952bSliubo 4137acce952bSliubo btrfs_invalidate_inodes(btrfs_inode->root); 4138b216cbfbSMiao Xie 4139eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4140acce952bSliubo } 4141acce952bSliubo 4142eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4143eb73c1b7SMiao Xie } 4144eb73c1b7SMiao Xie 4145eb73c1b7SMiao Xie static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 4146eb73c1b7SMiao Xie { 4147eb73c1b7SMiao Xie struct btrfs_root *root; 4148eb73c1b7SMiao Xie struct list_head splice; 4149eb73c1b7SMiao Xie 4150eb73c1b7SMiao Xie INIT_LIST_HEAD(&splice); 4151eb73c1b7SMiao Xie 4152eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4153eb73c1b7SMiao Xie list_splice_init(&fs_info->delalloc_roots, &splice); 4154eb73c1b7SMiao Xie while (!list_empty(&splice)) { 4155eb73c1b7SMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4156eb73c1b7SMiao Xie delalloc_root); 4157eb73c1b7SMiao Xie list_del_init(&root->delalloc_root); 4158eb73c1b7SMiao Xie root = btrfs_grab_fs_root(root); 4159eb73c1b7SMiao Xie BUG_ON(!root); 4160eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4161eb73c1b7SMiao Xie 4162eb73c1b7SMiao Xie btrfs_destroy_delalloc_inodes(root); 4163eb73c1b7SMiao Xie btrfs_put_fs_root(root); 4164eb73c1b7SMiao Xie 4165eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4166eb73c1b7SMiao Xie } 4167eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4168acce952bSliubo } 4169acce952bSliubo 4170acce952bSliubo static int btrfs_destroy_marked_extents(struct btrfs_root *root, 4171acce952bSliubo struct extent_io_tree *dirty_pages, 4172acce952bSliubo int mark) 4173acce952bSliubo { 4174acce952bSliubo int ret; 4175acce952bSliubo struct extent_buffer *eb; 4176acce952bSliubo u64 start = 0; 4177acce952bSliubo u64 end; 4178acce952bSliubo 4179acce952bSliubo while (1) { 4180acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 4181e6138876SJosef Bacik mark, NULL); 4182acce952bSliubo if (ret) 4183acce952bSliubo break; 4184acce952bSliubo 4185acce952bSliubo clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); 4186acce952bSliubo while (start <= end) { 418701d58472SDaniel Dressler eb = btrfs_find_tree_block(root->fs_info, start); 4188707e8a07SDavid Sterba start += root->nodesize; 4189fd8b2b61SJosef Bacik if (!eb) 4190acce952bSliubo continue; 4191fd8b2b61SJosef Bacik wait_on_extent_buffer_writeback(eb); 4192acce952bSliubo 4193fd8b2b61SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 4194fd8b2b61SJosef Bacik &eb->bflags)) 4195fd8b2b61SJosef Bacik clear_extent_buffer_dirty(eb); 4196fd8b2b61SJosef Bacik free_extent_buffer_stale(eb); 4197acce952bSliubo } 4198acce952bSliubo } 4199acce952bSliubo 4200acce952bSliubo return ret; 4201acce952bSliubo } 4202acce952bSliubo 4203acce952bSliubo static int btrfs_destroy_pinned_extent(struct btrfs_root *root, 4204acce952bSliubo struct extent_io_tree *pinned_extents) 4205acce952bSliubo { 4206acce952bSliubo struct extent_io_tree *unpin; 4207acce952bSliubo u64 start; 4208acce952bSliubo u64 end; 4209acce952bSliubo int ret; 4210ed0eaa14SLiu Bo bool loop = true; 4211acce952bSliubo 4212acce952bSliubo unpin = pinned_extents; 4213ed0eaa14SLiu Bo again: 4214acce952bSliubo while (1) { 4215acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 4216e6138876SJosef Bacik EXTENT_DIRTY, NULL); 4217acce952bSliubo if (ret) 4218acce952bSliubo break; 4219acce952bSliubo 4220acce952bSliubo clear_extent_dirty(unpin, start, end, GFP_NOFS); 4221acce952bSliubo btrfs_error_unpin_extent_range(root, start, end); 4222acce952bSliubo cond_resched(); 4223acce952bSliubo } 4224acce952bSliubo 4225ed0eaa14SLiu Bo if (loop) { 4226ed0eaa14SLiu Bo if (unpin == &root->fs_info->freed_extents[0]) 4227ed0eaa14SLiu Bo unpin = &root->fs_info->freed_extents[1]; 4228ed0eaa14SLiu Bo else 4229ed0eaa14SLiu Bo unpin = &root->fs_info->freed_extents[0]; 4230ed0eaa14SLiu Bo loop = false; 4231ed0eaa14SLiu Bo goto again; 4232ed0eaa14SLiu Bo } 4233ed0eaa14SLiu Bo 4234acce952bSliubo return 0; 4235acce952bSliubo } 4236acce952bSliubo 423750d9aa99SJosef Bacik static void btrfs_free_pending_ordered(struct btrfs_transaction *cur_trans, 423850d9aa99SJosef Bacik struct btrfs_fs_info *fs_info) 423950d9aa99SJosef Bacik { 424050d9aa99SJosef Bacik struct btrfs_ordered_extent *ordered; 424150d9aa99SJosef Bacik 424250d9aa99SJosef Bacik spin_lock(&fs_info->trans_lock); 424350d9aa99SJosef Bacik while (!list_empty(&cur_trans->pending_ordered)) { 424450d9aa99SJosef Bacik ordered = list_first_entry(&cur_trans->pending_ordered, 424550d9aa99SJosef Bacik struct btrfs_ordered_extent, 424650d9aa99SJosef Bacik trans_list); 424750d9aa99SJosef Bacik list_del_init(&ordered->trans_list); 424850d9aa99SJosef Bacik spin_unlock(&fs_info->trans_lock); 424950d9aa99SJosef Bacik 425050d9aa99SJosef Bacik btrfs_put_ordered_extent(ordered); 425150d9aa99SJosef Bacik spin_lock(&fs_info->trans_lock); 425250d9aa99SJosef Bacik } 425350d9aa99SJosef Bacik spin_unlock(&fs_info->trans_lock); 425450d9aa99SJosef Bacik } 425550d9aa99SJosef Bacik 425649b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 425749b25e05SJeff Mahoney struct btrfs_root *root) 425849b25e05SJeff Mahoney { 425949b25e05SJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, root); 426049b25e05SJeff Mahoney 42614a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 426249b25e05SJeff Mahoney wake_up(&root->fs_info->transaction_blocked_wait); 426349b25e05SJeff Mahoney 42644a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 426549b25e05SJeff Mahoney wake_up(&root->fs_info->transaction_wait); 426649b25e05SJeff Mahoney 426750d9aa99SJosef Bacik btrfs_free_pending_ordered(cur_trans, root->fs_info); 426867cde344SMiao Xie btrfs_destroy_delayed_inodes(root); 426967cde344SMiao Xie btrfs_assert_delayed_root_empty(root); 427067cde344SMiao Xie 427149b25e05SJeff Mahoney btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages, 427249b25e05SJeff Mahoney EXTENT_DIRTY); 42736e841e32SLiu Bo btrfs_destroy_pinned_extent(root, 42746e841e32SLiu Bo root->fs_info->pinned_extents); 427549b25e05SJeff Mahoney 42764a9d8bdeSMiao Xie cur_trans->state =TRANS_STATE_COMPLETED; 42774a9d8bdeSMiao Xie wake_up(&cur_trans->commit_wait); 42784a9d8bdeSMiao Xie 427949b25e05SJeff Mahoney /* 428049b25e05SJeff Mahoney memset(cur_trans, 0, sizeof(*cur_trans)); 428149b25e05SJeff Mahoney kmem_cache_free(btrfs_transaction_cachep, cur_trans); 428249b25e05SJeff Mahoney */ 428349b25e05SJeff Mahoney } 428449b25e05SJeff Mahoney 428548a3b636SEric Sandeen static int btrfs_cleanup_transaction(struct btrfs_root *root) 4286acce952bSliubo { 4287acce952bSliubo struct btrfs_transaction *t; 4288acce952bSliubo 4289acce952bSliubo mutex_lock(&root->fs_info->transaction_kthread_mutex); 4290acce952bSliubo 4291a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 4292724e2315SJosef Bacik while (!list_empty(&root->fs_info->trans_list)) { 4293724e2315SJosef Bacik t = list_first_entry(&root->fs_info->trans_list, 4294724e2315SJosef Bacik struct btrfs_transaction, list); 4295724e2315SJosef Bacik if (t->state >= TRANS_STATE_COMMIT_START) { 4296724e2315SJosef Bacik atomic_inc(&t->use_count); 4297724e2315SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 4298724e2315SJosef Bacik btrfs_wait_for_commit(root, t->transid); 4299724e2315SJosef Bacik btrfs_put_transaction(t); 4300724e2315SJosef Bacik spin_lock(&root->fs_info->trans_lock); 4301724e2315SJosef Bacik continue; 4302724e2315SJosef Bacik } 4303724e2315SJosef Bacik if (t == root->fs_info->running_transaction) { 4304724e2315SJosef Bacik t->state = TRANS_STATE_COMMIT_DOING; 4305724e2315SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 4306724e2315SJosef Bacik /* 4307724e2315SJosef Bacik * We wait for 0 num_writers since we don't hold a trans 4308724e2315SJosef Bacik * handle open currently for this transaction. 4309724e2315SJosef Bacik */ 4310724e2315SJosef Bacik wait_event(t->writer_wait, 4311724e2315SJosef Bacik atomic_read(&t->num_writers) == 0); 4312724e2315SJosef Bacik } else { 4313724e2315SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 4314724e2315SJosef Bacik } 4315724e2315SJosef Bacik btrfs_cleanup_one_transaction(t, root); 4316724e2315SJosef Bacik 4317724e2315SJosef Bacik spin_lock(&root->fs_info->trans_lock); 4318724e2315SJosef Bacik if (t == root->fs_info->running_transaction) 4319ac673879SMiao Xie root->fs_info->running_transaction = NULL; 4320724e2315SJosef Bacik list_del_init(&t->list); 4321a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 4322a4abeea4SJosef Bacik 4323724e2315SJosef Bacik btrfs_put_transaction(t); 4324724e2315SJosef Bacik trace_btrfs_transaction_commit(root); 4325724e2315SJosef Bacik spin_lock(&root->fs_info->trans_lock); 4326724e2315SJosef Bacik } 4327724e2315SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 4328199c2a9cSMiao Xie btrfs_destroy_all_ordered_extents(root->fs_info); 432967cde344SMiao Xie btrfs_destroy_delayed_inodes(root); 433067cde344SMiao Xie btrfs_assert_delayed_root_empty(root); 4331724e2315SJosef Bacik btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents); 4332eb73c1b7SMiao Xie btrfs_destroy_all_delalloc_inodes(root->fs_info); 4333acce952bSliubo mutex_unlock(&root->fs_info->transaction_kthread_mutex); 4334acce952bSliubo 4335acce952bSliubo return 0; 4336acce952bSliubo } 4337acce952bSliubo 4338e8c9f186SDavid Sterba static const struct extent_io_ops btree_extent_io_ops = { 4339ce9adaa5SChris Mason .readpage_end_io_hook = btree_readpage_end_io_hook, 43404bb31e92SArne Jansen .readpage_io_failed_hook = btree_io_failed_hook, 43410b86a832SChris Mason .submit_bio_hook = btree_submit_bio_hook, 4342239b14b3SChris Mason /* note we're sharing with inode.c for the merge bio hook */ 4343239b14b3SChris Mason .merge_bio_hook = btrfs_merge_bio_hook, 43440da5468fSChris Mason }; 4345