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)) { 30631e818feSDavid Sterba result = kzalloc(csum_size, 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; 4221104a885SDavid Sterba } 4231104a885SDavid Sterba 4241104a885SDavid Sterba if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) { 425efe120a0SFrank Holton printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n", 4261104a885SDavid Sterba csum_type); 4271104a885SDavid Sterba ret = 1; 4281104a885SDavid Sterba } 4291104a885SDavid Sterba 4301104a885SDavid Sterba return ret; 4311104a885SDavid Sterba } 4321104a885SDavid Sterba 4331104a885SDavid Sterba /* 434d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 435d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 436d352ac68SChris Mason */ 437f188591eSChris Mason static int btree_read_extent_buffer_pages(struct btrfs_root *root, 438f188591eSChris Mason struct extent_buffer *eb, 439ca7a79adSChris Mason u64 start, u64 parent_transid) 440f188591eSChris Mason { 441f188591eSChris Mason struct extent_io_tree *io_tree; 442ea466794SJosef Bacik int failed = 0; 443f188591eSChris Mason int ret; 444f188591eSChris Mason int num_copies = 0; 445f188591eSChris Mason int mirror_num = 0; 446ea466794SJosef Bacik int failed_mirror = 0; 447f188591eSChris Mason 448a826d6dcSJosef Bacik clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 449f188591eSChris Mason io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree; 450f188591eSChris Mason while (1) { 451bb82ab88SArne Jansen ret = read_extent_buffer_pages(io_tree, eb, start, 452bb82ab88SArne Jansen WAIT_COMPLETE, 453f188591eSChris Mason btree_get_extent, mirror_num); 454256dd1bbSStefan Behrens if (!ret) { 455256dd1bbSStefan Behrens if (!verify_parent_transid(io_tree, eb, 456b9fab919SChris Mason parent_transid, 0)) 457ea466794SJosef Bacik break; 458256dd1bbSStefan Behrens else 459256dd1bbSStefan Behrens ret = -EIO; 460256dd1bbSStefan Behrens } 461d397712bSChris Mason 462a826d6dcSJosef Bacik /* 463a826d6dcSJosef Bacik * This buffer's crc is fine, but its contents are corrupted, so 464a826d6dcSJosef Bacik * there is no reason to read the other copies, they won't be 465a826d6dcSJosef Bacik * any less wrong. 466a826d6dcSJosef Bacik */ 467a826d6dcSJosef Bacik if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags)) 468ea466794SJosef Bacik break; 469ea466794SJosef Bacik 4705d964051SStefan Behrens num_copies = btrfs_num_copies(root->fs_info, 471f188591eSChris Mason eb->start, eb->len); 4724235298eSChris Mason if (num_copies == 1) 473ea466794SJosef Bacik break; 4744235298eSChris Mason 4755cf1ab56SJosef Bacik if (!failed_mirror) { 4765cf1ab56SJosef Bacik failed = 1; 4775cf1ab56SJosef Bacik failed_mirror = eb->read_mirror; 4785cf1ab56SJosef Bacik } 4795cf1ab56SJosef Bacik 480f188591eSChris Mason mirror_num++; 481ea466794SJosef Bacik if (mirror_num == failed_mirror) 482ea466794SJosef Bacik mirror_num++; 483ea466794SJosef Bacik 4844235298eSChris Mason if (mirror_num > num_copies) 485ea466794SJosef Bacik break; 486f188591eSChris Mason } 487ea466794SJosef Bacik 488c0901581SStefan Behrens if (failed && !ret && failed_mirror) 489ea466794SJosef Bacik repair_eb_io_failure(root, eb, failed_mirror); 490ea466794SJosef Bacik 491ea466794SJosef Bacik return ret; 492f188591eSChris Mason } 49319c00ddcSChris Mason 494d352ac68SChris Mason /* 495d397712bSChris Mason * checksum a dirty tree block before IO. This has extra checks to make sure 496d397712bSChris Mason * we only fill in the checksum field in the first page of a multi-page block 497d352ac68SChris Mason */ 498d397712bSChris Mason 49901d58472SDaniel Dressler static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page) 50019c00ddcSChris Mason { 5014eee4fa4SMiao Xie u64 start = page_offset(page); 50219c00ddcSChris Mason u64 found_start; 50319c00ddcSChris Mason struct extent_buffer *eb; 504f188591eSChris Mason 5054f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 5064f2de97aSJosef Bacik if (page != eb->pages[0]) 5074f2de97aSJosef Bacik return 0; 50819c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 509fae7f21cSDulshani Gunawardhana if (WARN_ON(found_start != start || !PageUptodate(page))) 5104f2de97aSJosef Bacik return 0; 51101d58472SDaniel Dressler csum_tree_block(fs_info, eb, 0); 51219c00ddcSChris Mason return 0; 51319c00ddcSChris Mason } 51419c00ddcSChris Mason 51501d58472SDaniel Dressler static int check_tree_block_fsid(struct btrfs_fs_info *fs_info, 5162b82032cSYan Zheng struct extent_buffer *eb) 5172b82032cSYan Zheng { 51801d58472SDaniel Dressler struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 5192b82032cSYan Zheng u8 fsid[BTRFS_UUID_SIZE]; 5202b82032cSYan Zheng int ret = 1; 5212b82032cSYan Zheng 5220a4e5586SRoss Kirk read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE); 5232b82032cSYan Zheng while (fs_devices) { 5242b82032cSYan Zheng if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) { 5252b82032cSYan Zheng ret = 0; 5262b82032cSYan Zheng break; 5272b82032cSYan Zheng } 5282b82032cSYan Zheng fs_devices = fs_devices->seed; 5292b82032cSYan Zheng } 5302b82032cSYan Zheng return ret; 5312b82032cSYan Zheng } 5322b82032cSYan Zheng 533a826d6dcSJosef Bacik #define CORRUPT(reason, eb, root, slot) \ 534efe120a0SFrank Holton btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \ 535efe120a0SFrank Holton "root=%llu, slot=%d", reason, \ 536c1c9ff7cSGeert Uytterhoeven btrfs_header_bytenr(eb), root->objectid, slot) 537a826d6dcSJosef Bacik 538a826d6dcSJosef Bacik static noinline int check_leaf(struct btrfs_root *root, 539a826d6dcSJosef Bacik struct extent_buffer *leaf) 540a826d6dcSJosef Bacik { 541a826d6dcSJosef Bacik struct btrfs_key key; 542a826d6dcSJosef Bacik struct btrfs_key leaf_key; 543a826d6dcSJosef Bacik u32 nritems = btrfs_header_nritems(leaf); 544a826d6dcSJosef Bacik int slot; 545a826d6dcSJosef Bacik 546a826d6dcSJosef Bacik if (nritems == 0) 547a826d6dcSJosef Bacik return 0; 548a826d6dcSJosef Bacik 549a826d6dcSJosef Bacik /* Check the 0 item */ 550a826d6dcSJosef Bacik if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) != 551a826d6dcSJosef Bacik BTRFS_LEAF_DATA_SIZE(root)) { 552a826d6dcSJosef Bacik CORRUPT("invalid item offset size pair", leaf, root, 0); 553a826d6dcSJosef Bacik return -EIO; 554a826d6dcSJosef Bacik } 555a826d6dcSJosef Bacik 556a826d6dcSJosef Bacik /* 557a826d6dcSJosef Bacik * Check to make sure each items keys are in the correct order and their 558a826d6dcSJosef Bacik * offsets make sense. We only have to loop through nritems-1 because 559a826d6dcSJosef Bacik * we check the current slot against the next slot, which verifies the 560a826d6dcSJosef Bacik * next slot's offset+size makes sense and that the current's slot 561a826d6dcSJosef Bacik * offset is correct. 562a826d6dcSJosef Bacik */ 563a826d6dcSJosef Bacik for (slot = 0; slot < nritems - 1; slot++) { 564a826d6dcSJosef Bacik btrfs_item_key_to_cpu(leaf, &leaf_key, slot); 565a826d6dcSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, slot + 1); 566a826d6dcSJosef Bacik 567a826d6dcSJosef Bacik /* Make sure the keys are in the right order */ 568a826d6dcSJosef Bacik if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) { 569a826d6dcSJosef Bacik CORRUPT("bad key order", leaf, root, slot); 570a826d6dcSJosef Bacik return -EIO; 571a826d6dcSJosef Bacik } 572a826d6dcSJosef Bacik 573a826d6dcSJosef Bacik /* 574a826d6dcSJosef Bacik * Make sure the offset and ends are right, remember that the 575a826d6dcSJosef Bacik * item data starts at the end of the leaf and grows towards the 576a826d6dcSJosef Bacik * front. 577a826d6dcSJosef Bacik */ 578a826d6dcSJosef Bacik if (btrfs_item_offset_nr(leaf, slot) != 579a826d6dcSJosef Bacik btrfs_item_end_nr(leaf, slot + 1)) { 580a826d6dcSJosef Bacik CORRUPT("slot offset bad", leaf, root, slot); 581a826d6dcSJosef Bacik return -EIO; 582a826d6dcSJosef Bacik } 583a826d6dcSJosef Bacik 584a826d6dcSJosef Bacik /* 585a826d6dcSJosef Bacik * Check to make sure that we don't point outside of the leaf, 586a826d6dcSJosef Bacik * just incase all the items are consistent to eachother, but 587a826d6dcSJosef Bacik * all point outside of the leaf. 588a826d6dcSJosef Bacik */ 589a826d6dcSJosef Bacik if (btrfs_item_end_nr(leaf, slot) > 590a826d6dcSJosef Bacik BTRFS_LEAF_DATA_SIZE(root)) { 591a826d6dcSJosef Bacik CORRUPT("slot end outside of leaf", leaf, root, slot); 592a826d6dcSJosef Bacik return -EIO; 593a826d6dcSJosef Bacik } 594a826d6dcSJosef Bacik } 595a826d6dcSJosef Bacik 596a826d6dcSJosef Bacik return 0; 597a826d6dcSJosef Bacik } 598a826d6dcSJosef Bacik 599facc8a22SMiao Xie static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio, 600facc8a22SMiao Xie u64 phy_offset, struct page *page, 601facc8a22SMiao Xie u64 start, u64 end, int mirror) 602ce9adaa5SChris Mason { 603ce9adaa5SChris Mason u64 found_start; 604ce9adaa5SChris Mason int found_level; 605ce9adaa5SChris Mason struct extent_buffer *eb; 606ce9adaa5SChris Mason struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 607f188591eSChris Mason int ret = 0; 608727011e0SChris Mason int reads_done; 609ce9adaa5SChris Mason 610ce9adaa5SChris Mason if (!page->private) 611ce9adaa5SChris Mason goto out; 612d397712bSChris Mason 6134f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 614d397712bSChris Mason 6150b32f4bbSJosef Bacik /* the pending IO might have been the only thing that kept this buffer 6160b32f4bbSJosef Bacik * in memory. Make sure we have a ref for all this other checks 6170b32f4bbSJosef Bacik */ 6180b32f4bbSJosef Bacik extent_buffer_get(eb); 6190b32f4bbSJosef Bacik 6200b32f4bbSJosef Bacik reads_done = atomic_dec_and_test(&eb->io_pages); 621727011e0SChris Mason if (!reads_done) 622727011e0SChris Mason goto err; 623f188591eSChris Mason 6245cf1ab56SJosef Bacik eb->read_mirror = mirror; 625656f30dbSFilipe Manana if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 626ea466794SJosef Bacik ret = -EIO; 627ea466794SJosef Bacik goto err; 628ea466794SJosef Bacik } 629ea466794SJosef Bacik 630ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 631727011e0SChris Mason if (found_start != eb->start) { 63268b663d1SDavid Sterba printk_ratelimited(KERN_ERR "BTRFS (device %s): bad tree block start " 633193f284dSChris Mason "%llu %llu\n", 63429549aecSWang Shilong eb->fs_info->sb->s_id, found_start, eb->start); 635f188591eSChris Mason ret = -EIO; 636ce9adaa5SChris Mason goto err; 637ce9adaa5SChris Mason } 63801d58472SDaniel Dressler if (check_tree_block_fsid(root->fs_info, eb)) { 63968b663d1SDavid Sterba printk_ratelimited(KERN_ERR "BTRFS (device %s): bad fsid on block %llu\n", 64029549aecSWang Shilong eb->fs_info->sb->s_id, eb->start); 6411259ab75SChris Mason ret = -EIO; 6421259ab75SChris Mason goto err; 6431259ab75SChris Mason } 644ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 6451c24c3ceSJosef Bacik if (found_level >= BTRFS_MAX_LEVEL) { 64668b663d1SDavid Sterba btrfs_err(root->fs_info, "bad tree block level %d", 6471c24c3ceSJosef Bacik (int)btrfs_header_level(eb)); 6481c24c3ceSJosef Bacik ret = -EIO; 6491c24c3ceSJosef Bacik goto err; 6501c24c3ceSJosef Bacik } 651ce9adaa5SChris Mason 65285d4e461SChris Mason btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), 65385d4e461SChris Mason eb, found_level); 6544008c04aSChris Mason 65501d58472SDaniel Dressler ret = csum_tree_block(root->fs_info, eb, 1); 656a826d6dcSJosef Bacik if (ret) { 657f188591eSChris Mason ret = -EIO; 658a826d6dcSJosef Bacik goto err; 659a826d6dcSJosef Bacik } 660a826d6dcSJosef Bacik 661a826d6dcSJosef Bacik /* 662a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 663a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 664a826d6dcSJosef Bacik * return -EIO. 665a826d6dcSJosef Bacik */ 666a826d6dcSJosef Bacik if (found_level == 0 && check_leaf(root, eb)) { 667a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 668a826d6dcSJosef Bacik ret = -EIO; 669a826d6dcSJosef Bacik } 670ce9adaa5SChris Mason 6710b32f4bbSJosef Bacik if (!ret) 6720b32f4bbSJosef Bacik set_extent_buffer_uptodate(eb); 673ce9adaa5SChris Mason err: 67479fb65a1SJosef Bacik if (reads_done && 67579fb65a1SJosef Bacik test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 6764bb31e92SArne Jansen btree_readahead_hook(root, eb, eb->start, ret); 6774bb31e92SArne Jansen 67853b381b3SDavid Woodhouse if (ret) { 67953b381b3SDavid Woodhouse /* 68053b381b3SDavid Woodhouse * our io error hook is going to dec the io pages 68153b381b3SDavid Woodhouse * again, we have to make sure it has something 68253b381b3SDavid Woodhouse * to decrement 68353b381b3SDavid Woodhouse */ 68453b381b3SDavid Woodhouse atomic_inc(&eb->io_pages); 6850b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 68653b381b3SDavid Woodhouse } 6870b32f4bbSJosef Bacik free_extent_buffer(eb); 688ce9adaa5SChris Mason out: 689f188591eSChris Mason return ret; 690ce9adaa5SChris Mason } 691ce9adaa5SChris Mason 692ea466794SJosef Bacik static int btree_io_failed_hook(struct page *page, int failed_mirror) 6934bb31e92SArne Jansen { 6944bb31e92SArne Jansen struct extent_buffer *eb; 6954bb31e92SArne Jansen struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 6964bb31e92SArne Jansen 6974f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 698656f30dbSFilipe Manana set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); 6995cf1ab56SJosef Bacik eb->read_mirror = failed_mirror; 70053b381b3SDavid Woodhouse atomic_dec(&eb->io_pages); 701ea466794SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 7024bb31e92SArne Jansen btree_readahead_hook(root, eb, eb->start, -EIO); 7034bb31e92SArne Jansen return -EIO; /* we fixed nothing */ 7044bb31e92SArne Jansen } 7054bb31e92SArne Jansen 706ce9adaa5SChris Mason static void end_workqueue_bio(struct bio *bio, int err) 707ce9adaa5SChris Mason { 70897eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq = bio->bi_private; 709ce9adaa5SChris Mason struct btrfs_fs_info *fs_info; 7109e0af237SLiu Bo struct btrfs_workqueue *wq; 7119e0af237SLiu Bo btrfs_work_func_t func; 712ce9adaa5SChris Mason 713ce9adaa5SChris Mason fs_info = end_io_wq->info; 714ce9adaa5SChris Mason end_io_wq->error = err; 715d20f7043SChris Mason 7167b6d91daSChristoph Hellwig if (bio->bi_rw & REQ_WRITE) { 7179e0af237SLiu Bo if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) { 7189e0af237SLiu Bo wq = fs_info->endio_meta_write_workers; 7199e0af237SLiu Bo func = btrfs_endio_meta_write_helper; 7209e0af237SLiu Bo } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) { 7219e0af237SLiu Bo wq = fs_info->endio_freespace_worker; 7229e0af237SLiu Bo func = btrfs_freespace_write_helper; 7239e0af237SLiu Bo } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) { 7249e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 7259e0af237SLiu Bo func = btrfs_endio_raid56_helper; 726d20f7043SChris Mason } else { 7279e0af237SLiu Bo wq = fs_info->endio_write_workers; 7289e0af237SLiu Bo func = btrfs_endio_write_helper; 729d20f7043SChris Mason } 7309e0af237SLiu Bo } else { 7318b110e39SMiao Xie if (unlikely(end_io_wq->metadata == 7328b110e39SMiao Xie BTRFS_WQ_ENDIO_DIO_REPAIR)) { 7338b110e39SMiao Xie wq = fs_info->endio_repair_workers; 7348b110e39SMiao Xie func = btrfs_endio_repair_helper; 7358b110e39SMiao Xie } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) { 7369e0af237SLiu Bo wq = fs_info->endio_raid56_workers; 7379e0af237SLiu Bo func = btrfs_endio_raid56_helper; 7389e0af237SLiu Bo } else if (end_io_wq->metadata) { 7399e0af237SLiu Bo wq = fs_info->endio_meta_workers; 7409e0af237SLiu Bo func = btrfs_endio_meta_helper; 7419e0af237SLiu Bo } else { 7429e0af237SLiu Bo wq = fs_info->endio_workers; 7439e0af237SLiu Bo func = btrfs_endio_helper; 7449e0af237SLiu Bo } 7459e0af237SLiu Bo } 7469e0af237SLiu Bo 7479e0af237SLiu Bo btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL); 7489e0af237SLiu Bo btrfs_queue_work(wq, &end_io_wq->work); 749ce9adaa5SChris Mason } 750ce9adaa5SChris Mason 75122c59948SChris Mason int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 752bfebd8b5SDavid Sterba enum btrfs_wq_endio_type metadata) 7530b86a832SChris Mason { 75497eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 7558b110e39SMiao Xie 75697eb6b69SDavid Sterba end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS); 757ce9adaa5SChris Mason if (!end_io_wq) 758ce9adaa5SChris Mason return -ENOMEM; 759ce9adaa5SChris Mason 760ce9adaa5SChris Mason end_io_wq->private = bio->bi_private; 761ce9adaa5SChris Mason end_io_wq->end_io = bio->bi_end_io; 76222c59948SChris Mason end_io_wq->info = info; 763ce9adaa5SChris Mason end_io_wq->error = 0; 764ce9adaa5SChris Mason end_io_wq->bio = bio; 76522c59948SChris Mason end_io_wq->metadata = metadata; 766ce9adaa5SChris Mason 767ce9adaa5SChris Mason bio->bi_private = end_io_wq; 768ce9adaa5SChris Mason bio->bi_end_io = end_workqueue_bio; 76922c59948SChris Mason return 0; 77022c59948SChris Mason } 77122c59948SChris Mason 772b64a2851SChris Mason unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info) 7734854ddd0SChris Mason { 7744854ddd0SChris Mason unsigned long limit = min_t(unsigned long, 7755cdc7ad3SQu Wenruo info->thread_pool_size, 7764854ddd0SChris Mason info->fs_devices->open_devices); 7774854ddd0SChris Mason return 256 * limit; 7784854ddd0SChris Mason } 7794854ddd0SChris Mason 7804a69a410SChris Mason static void run_one_async_start(struct btrfs_work *work) 7814a69a410SChris Mason { 7824a69a410SChris Mason struct async_submit_bio *async; 78379787eaaSJeff Mahoney int ret; 7844a69a410SChris Mason 7854a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 78679787eaaSJeff Mahoney ret = async->submit_bio_start(async->inode, async->rw, async->bio, 787eaf25d93SChris Mason async->mirror_num, async->bio_flags, 788eaf25d93SChris Mason async->bio_offset); 78979787eaaSJeff Mahoney if (ret) 79079787eaaSJeff Mahoney async->error = ret; 7914a69a410SChris Mason } 7924a69a410SChris Mason 7934a69a410SChris Mason static void run_one_async_done(struct btrfs_work *work) 7948b712842SChris Mason { 7958b712842SChris Mason struct btrfs_fs_info *fs_info; 7968b712842SChris Mason struct async_submit_bio *async; 7974854ddd0SChris Mason int limit; 7988b712842SChris Mason 7998b712842SChris Mason async = container_of(work, struct async_submit_bio, work); 8008b712842SChris Mason fs_info = BTRFS_I(async->inode)->root->fs_info; 8014854ddd0SChris Mason 802b64a2851SChris Mason limit = btrfs_async_submit_limit(fs_info); 8034854ddd0SChris Mason limit = limit * 2 / 3; 8044854ddd0SChris Mason 80566657b31SJosef Bacik if (atomic_dec_return(&fs_info->nr_async_submits) < limit && 806b64a2851SChris Mason waitqueue_active(&fs_info->async_submit_wait)) 8074854ddd0SChris Mason wake_up(&fs_info->async_submit_wait); 8084854ddd0SChris Mason 80979787eaaSJeff Mahoney /* If an error occured we just want to clean up the bio and move on */ 81079787eaaSJeff Mahoney if (async->error) { 81179787eaaSJeff Mahoney bio_endio(async->bio, async->error); 81279787eaaSJeff Mahoney return; 81379787eaaSJeff Mahoney } 81479787eaaSJeff Mahoney 8154a69a410SChris Mason async->submit_bio_done(async->inode, async->rw, async->bio, 816eaf25d93SChris Mason async->mirror_num, async->bio_flags, 817eaf25d93SChris Mason async->bio_offset); 8184a69a410SChris Mason } 8194a69a410SChris Mason 8204a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 8214a69a410SChris Mason { 8224a69a410SChris Mason struct async_submit_bio *async; 8234a69a410SChris Mason 8244a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 8258b712842SChris Mason kfree(async); 8268b712842SChris Mason } 8278b712842SChris Mason 82844b8bd7eSChris Mason int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode, 82944b8bd7eSChris Mason int rw, struct bio *bio, int mirror_num, 830c8b97818SChris Mason unsigned long bio_flags, 831eaf25d93SChris Mason u64 bio_offset, 8324a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_start, 8334a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_done) 83444b8bd7eSChris Mason { 83544b8bd7eSChris Mason struct async_submit_bio *async; 83644b8bd7eSChris Mason 83744b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 83844b8bd7eSChris Mason if (!async) 83944b8bd7eSChris Mason return -ENOMEM; 84044b8bd7eSChris Mason 84144b8bd7eSChris Mason async->inode = inode; 84244b8bd7eSChris Mason async->rw = rw; 84344b8bd7eSChris Mason async->bio = bio; 84444b8bd7eSChris Mason async->mirror_num = mirror_num; 8454a69a410SChris Mason async->submit_bio_start = submit_bio_start; 8464a69a410SChris Mason async->submit_bio_done = submit_bio_done; 8474a69a410SChris Mason 8489e0af237SLiu Bo btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start, 8495cdc7ad3SQu Wenruo run_one_async_done, run_one_async_free); 8504a69a410SChris Mason 851c8b97818SChris Mason async->bio_flags = bio_flags; 852eaf25d93SChris Mason async->bio_offset = bio_offset; 8538c8bee1dSChris Mason 85479787eaaSJeff Mahoney async->error = 0; 85579787eaaSJeff Mahoney 856cb03c743SChris Mason atomic_inc(&fs_info->nr_async_submits); 857d313d7a3SChris Mason 8587b6d91daSChristoph Hellwig if (rw & REQ_SYNC) 8595cdc7ad3SQu Wenruo btrfs_set_work_high_priority(&async->work); 860d313d7a3SChris Mason 8615cdc7ad3SQu Wenruo btrfs_queue_work(fs_info->workers, &async->work); 8629473f16cSChris Mason 863771ed689SChris Mason while (atomic_read(&fs_info->async_submit_draining) && 864771ed689SChris Mason atomic_read(&fs_info->nr_async_submits)) { 865771ed689SChris Mason wait_event(fs_info->async_submit_wait, 866771ed689SChris Mason (atomic_read(&fs_info->nr_async_submits) == 0)); 867771ed689SChris Mason } 868771ed689SChris Mason 86944b8bd7eSChris Mason return 0; 87044b8bd7eSChris Mason } 87144b8bd7eSChris Mason 872ce3ed71aSChris Mason static int btree_csum_one_bio(struct bio *bio) 873ce3ed71aSChris Mason { 8742c30c71bSKent Overstreet struct bio_vec *bvec; 875ce3ed71aSChris Mason struct btrfs_root *root; 8762c30c71bSKent Overstreet int i, ret = 0; 877ce3ed71aSChris Mason 8782c30c71bSKent Overstreet bio_for_each_segment_all(bvec, bio, i) { 879ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 88001d58472SDaniel Dressler ret = csum_dirty_buffer(root->fs_info, bvec->bv_page); 88179787eaaSJeff Mahoney if (ret) 88279787eaaSJeff Mahoney break; 883ce3ed71aSChris Mason } 8842c30c71bSKent Overstreet 88579787eaaSJeff Mahoney return ret; 886ce3ed71aSChris Mason } 887ce3ed71aSChris Mason 8884a69a410SChris Mason static int __btree_submit_bio_start(struct inode *inode, int rw, 8894a69a410SChris Mason struct bio *bio, int mirror_num, 890eaf25d93SChris Mason unsigned long bio_flags, 891eaf25d93SChris Mason u64 bio_offset) 89222c59948SChris Mason { 8938b712842SChris Mason /* 8948b712842SChris Mason * when we're called for a write, we're already in the async 8955443be45SChris Mason * submission context. Just jump into btrfs_map_bio 8968b712842SChris Mason */ 89779787eaaSJeff Mahoney return btree_csum_one_bio(bio); 89822c59948SChris Mason } 89922c59948SChris Mason 9004a69a410SChris Mason static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio, 901eaf25d93SChris Mason int mirror_num, unsigned long bio_flags, 902eaf25d93SChris Mason u64 bio_offset) 9034a69a410SChris Mason { 90461891923SStefan Behrens int ret; 90561891923SStefan Behrens 9068b712842SChris Mason /* 9074a69a410SChris Mason * when we're called for a write, we're already in the async 9084a69a410SChris Mason * submission context. Just jump into btrfs_map_bio 9098b712842SChris Mason */ 91061891923SStefan Behrens ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1); 91161891923SStefan Behrens if (ret) 91261891923SStefan Behrens bio_endio(bio, ret); 91361891923SStefan Behrens return ret; 9140b86a832SChris Mason } 9150b86a832SChris Mason 916de0022b9SJosef Bacik static int check_async_write(struct inode *inode, unsigned long bio_flags) 917de0022b9SJosef Bacik { 918de0022b9SJosef Bacik if (bio_flags & EXTENT_BIO_TREE_LOG) 919de0022b9SJosef Bacik return 0; 920de0022b9SJosef Bacik #ifdef CONFIG_X86 921de0022b9SJosef Bacik if (cpu_has_xmm4_2) 922de0022b9SJosef Bacik return 0; 923de0022b9SJosef Bacik #endif 924de0022b9SJosef Bacik return 1; 925de0022b9SJosef Bacik } 926de0022b9SJosef Bacik 92744b8bd7eSChris Mason static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, 928eaf25d93SChris Mason int mirror_num, unsigned long bio_flags, 929eaf25d93SChris Mason u64 bio_offset) 93044b8bd7eSChris Mason { 931de0022b9SJosef Bacik int async = check_async_write(inode, bio_flags); 9324a69a410SChris Mason int ret; 933cad321adSChris Mason 9347b6d91daSChristoph Hellwig if (!(rw & REQ_WRITE)) { 935cad321adSChris Mason /* 936cad321adSChris Mason * called for a read, do the setup so that checksum validation 937cad321adSChris Mason * can happen in the async kernel threads 938cad321adSChris Mason */ 939f3f266abSChris Mason ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info, 940bfebd8b5SDavid Sterba bio, BTRFS_WQ_ENDIO_METADATA); 9411d4284bdSChris Mason if (ret) 94261891923SStefan Behrens goto out_w_error; 94361891923SStefan Behrens ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, 9446f3577bdSChris Mason mirror_num, 0); 945de0022b9SJosef Bacik } else if (!async) { 946de0022b9SJosef Bacik ret = btree_csum_one_bio(bio); 947de0022b9SJosef Bacik if (ret) 94861891923SStefan Behrens goto out_w_error; 94961891923SStefan Behrens ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, 950de0022b9SJosef Bacik mirror_num, 0); 95161891923SStefan Behrens } else { 952cad321adSChris Mason /* 95361891923SStefan Behrens * kthread helpers are used to submit writes so that 95461891923SStefan Behrens * checksumming can happen in parallel across all CPUs 955cad321adSChris Mason */ 95661891923SStefan Behrens ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, 957c8b97818SChris Mason inode, rw, bio, mirror_num, 0, 958eaf25d93SChris Mason bio_offset, 9594a69a410SChris Mason __btree_submit_bio_start, 9604a69a410SChris Mason __btree_submit_bio_done); 96144b8bd7eSChris Mason } 96244b8bd7eSChris Mason 96361891923SStefan Behrens if (ret) { 96461891923SStefan Behrens out_w_error: 96561891923SStefan Behrens bio_endio(bio, ret); 96661891923SStefan Behrens } 96761891923SStefan Behrens return ret; 96861891923SStefan Behrens } 96961891923SStefan Behrens 9703dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 971784b4e29SChris Mason static int btree_migratepage(struct address_space *mapping, 972a6bc32b8SMel Gorman struct page *newpage, struct page *page, 973a6bc32b8SMel Gorman enum migrate_mode mode) 974784b4e29SChris Mason { 975784b4e29SChris Mason /* 976784b4e29SChris Mason * we can't safely write a btree page from here, 977784b4e29SChris Mason * we haven't done the locking hook 978784b4e29SChris Mason */ 979784b4e29SChris Mason if (PageDirty(page)) 980784b4e29SChris Mason return -EAGAIN; 981784b4e29SChris Mason /* 982784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 983784b4e29SChris Mason * We must have no buffers or drop them. 984784b4e29SChris Mason */ 985784b4e29SChris Mason if (page_has_private(page) && 986784b4e29SChris Mason !try_to_release_page(page, GFP_KERNEL)) 987784b4e29SChris Mason return -EAGAIN; 988a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 989784b4e29SChris Mason } 9903dd1462eSJan Beulich #endif 991784b4e29SChris Mason 9920da5468fSChris Mason 9930da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 9940da5468fSChris Mason struct writeback_control *wbc) 9950da5468fSChris Mason { 996e2d84521SMiao Xie struct btrfs_fs_info *fs_info; 997e2d84521SMiao Xie int ret; 998e2d84521SMiao Xie 999d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 1000448d640bSChris Mason 1001448d640bSChris Mason if (wbc->for_kupdate) 1002448d640bSChris Mason return 0; 1003448d640bSChris Mason 1004e2d84521SMiao Xie fs_info = BTRFS_I(mapping->host)->root->fs_info; 1005b9473439SChris Mason /* this is a bit racy, but that's ok */ 1006e2d84521SMiao Xie ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes, 1007e2d84521SMiao Xie BTRFS_DIRTY_METADATA_THRESH); 1008e2d84521SMiao Xie if (ret < 0) 1009793955bcSChris Mason return 0; 1010793955bcSChris Mason } 10110b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 10120da5468fSChris Mason } 10130da5468fSChris Mason 1014b2950863SChristoph Hellwig static int btree_readpage(struct file *file, struct page *page) 10155f39d397SChris Mason { 1016d1310b2eSChris Mason struct extent_io_tree *tree; 1017d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 10188ddc7d9cSJan Schmidt return extent_read_full_page(tree, page, btree_get_extent, 0); 10195f39d397SChris Mason } 10205f39d397SChris Mason 102170dec807SChris Mason static int btree_releasepage(struct page *page, gfp_t gfp_flags) 10225f39d397SChris Mason { 102398509cfcSChris Mason if (PageWriteback(page) || PageDirty(page)) 102498509cfcSChris Mason return 0; 10250c4e538bSDavid Sterba 1026f7a52a40SDavid Sterba return try_release_extent_buffer(page); 1027d98237b3SChris Mason } 1028d98237b3SChris Mason 1029d47992f8SLukas Czerner static void btree_invalidatepage(struct page *page, unsigned int offset, 1030d47992f8SLukas Czerner unsigned int length) 1031d98237b3SChris Mason { 1032d1310b2eSChris Mason struct extent_io_tree *tree; 1033d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 10345f39d397SChris Mason extent_invalidatepage(tree, page, offset); 10355f39d397SChris Mason btree_releasepage(page, GFP_NOFS); 10369ad6b7bcSChris Mason if (PagePrivate(page)) { 1037efe120a0SFrank Holton btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info, 1038efe120a0SFrank Holton "page private not zero on page %llu", 1039efe120a0SFrank Holton (unsigned long long)page_offset(page)); 10409ad6b7bcSChris Mason ClearPagePrivate(page); 10419ad6b7bcSChris Mason set_page_private(page, 0); 10429ad6b7bcSChris Mason page_cache_release(page); 10439ad6b7bcSChris Mason } 1044d98237b3SChris Mason } 1045d98237b3SChris Mason 10460b32f4bbSJosef Bacik static int btree_set_page_dirty(struct page *page) 10470b32f4bbSJosef Bacik { 1048bb146eb2SJosef Bacik #ifdef DEBUG 10490b32f4bbSJosef Bacik struct extent_buffer *eb; 10500b32f4bbSJosef Bacik 10510b32f4bbSJosef Bacik BUG_ON(!PagePrivate(page)); 10520b32f4bbSJosef Bacik eb = (struct extent_buffer *)page->private; 10530b32f4bbSJosef Bacik BUG_ON(!eb); 10540b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 10550b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 10560b32f4bbSJosef Bacik btrfs_assert_tree_locked(eb); 1057bb146eb2SJosef Bacik #endif 10580b32f4bbSJosef Bacik return __set_page_dirty_nobuffers(page); 10590b32f4bbSJosef Bacik } 10600b32f4bbSJosef Bacik 10617f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 1062d98237b3SChris Mason .readpage = btree_readpage, 10630da5468fSChris Mason .writepages = btree_writepages, 10645f39d397SChris Mason .releasepage = btree_releasepage, 10655f39d397SChris Mason .invalidatepage = btree_invalidatepage, 10665a92bc88SChris Mason #ifdef CONFIG_MIGRATION 1067784b4e29SChris Mason .migratepage = btree_migratepage, 10685a92bc88SChris Mason #endif 10690b32f4bbSJosef Bacik .set_page_dirty = btree_set_page_dirty, 1070d98237b3SChris Mason }; 1071123abc88SChris Mason 1072d3e46feaSDavid Sterba void readahead_tree_block(struct btrfs_root *root, u64 bytenr) 1073090d1875SChris Mason { 10745f39d397SChris Mason struct extent_buffer *buf = NULL; 10755f39d397SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 1076090d1875SChris Mason 1077a83fffb7SDavid Sterba buf = btrfs_find_create_tree_block(root, bytenr); 10785f39d397SChris Mason if (!buf) 10796197d86eSDavid Sterba return; 1080d1310b2eSChris Mason read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, 1081bb82ab88SArne Jansen buf, 0, WAIT_NONE, btree_get_extent, 0); 10825f39d397SChris Mason free_extent_buffer(buf); 1083090d1875SChris Mason } 1084090d1875SChris Mason 1085c0dcaa4dSDavid Sterba int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, 1086ab0fff03SArne Jansen int mirror_num, struct extent_buffer **eb) 1087ab0fff03SArne Jansen { 1088ab0fff03SArne Jansen struct extent_buffer *buf = NULL; 1089ab0fff03SArne Jansen struct inode *btree_inode = root->fs_info->btree_inode; 1090ab0fff03SArne Jansen struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree; 1091ab0fff03SArne Jansen int ret; 1092ab0fff03SArne Jansen 1093a83fffb7SDavid Sterba buf = btrfs_find_create_tree_block(root, bytenr); 1094ab0fff03SArne Jansen if (!buf) 1095ab0fff03SArne Jansen return 0; 1096ab0fff03SArne Jansen 1097ab0fff03SArne Jansen set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags); 1098ab0fff03SArne Jansen 1099ab0fff03SArne Jansen ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK, 1100ab0fff03SArne Jansen btree_get_extent, mirror_num); 1101ab0fff03SArne Jansen if (ret) { 1102ab0fff03SArne Jansen free_extent_buffer(buf); 1103ab0fff03SArne Jansen return ret; 1104ab0fff03SArne Jansen } 1105ab0fff03SArne Jansen 1106ab0fff03SArne Jansen if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) { 1107ab0fff03SArne Jansen free_extent_buffer(buf); 1108ab0fff03SArne Jansen return -EIO; 11090b32f4bbSJosef Bacik } else if (extent_buffer_uptodate(buf)) { 1110ab0fff03SArne Jansen *eb = buf; 1111ab0fff03SArne Jansen } else { 1112ab0fff03SArne Jansen free_extent_buffer(buf); 1113ab0fff03SArne Jansen } 1114ab0fff03SArne Jansen return 0; 1115ab0fff03SArne Jansen } 1116ab0fff03SArne Jansen 111701d58472SDaniel Dressler struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info, 11180308af44SDavid Sterba u64 bytenr) 11190999df54SChris Mason { 112001d58472SDaniel Dressler return find_extent_buffer(fs_info, bytenr); 11210999df54SChris Mason } 11220999df54SChris Mason 11230999df54SChris Mason struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root, 1124a83fffb7SDavid Sterba u64 bytenr) 11250999df54SChris Mason { 1126fccb84c9SDavid Sterba if (btrfs_test_is_dummy_root(root)) 1127ce3e6984SDavid Sterba return alloc_test_extent_buffer(root->fs_info, bytenr); 1128ce3e6984SDavid Sterba return alloc_extent_buffer(root->fs_info, bytenr); 11290999df54SChris Mason } 11300999df54SChris Mason 11310999df54SChris Mason 1132e02119d5SChris Mason int btrfs_write_tree_block(struct extent_buffer *buf) 1133e02119d5SChris Mason { 1134727011e0SChris Mason return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start, 11358aa38c31SChristoph Hellwig buf->start + buf->len - 1); 1136e02119d5SChris Mason } 1137e02119d5SChris Mason 1138e02119d5SChris Mason int btrfs_wait_tree_block_writeback(struct extent_buffer *buf) 1139e02119d5SChris Mason { 1140727011e0SChris Mason return filemap_fdatawait_range(buf->pages[0]->mapping, 11418aa38c31SChristoph Hellwig buf->start, buf->start + buf->len - 1); 1142e02119d5SChris Mason } 1143e02119d5SChris Mason 1144db94535dSChris Mason struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr, 1145ce86cd59SDavid Sterba u64 parent_transid) 1146e20d96d6SChris Mason { 11475f39d397SChris Mason struct extent_buffer *buf = NULL; 114819c00ddcSChris Mason int ret; 114919c00ddcSChris Mason 1150a83fffb7SDavid Sterba buf = btrfs_find_create_tree_block(root, bytenr); 11515f39d397SChris Mason if (!buf) 115264c043deSLiu Bo return ERR_PTR(-ENOMEM); 1153e4204dedSChris Mason 1154ca7a79adSChris Mason ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 11550f0fe8f7SFilipe David Borba Manana if (ret) { 11560f0fe8f7SFilipe David Borba Manana free_extent_buffer(buf); 115764c043deSLiu Bo return ERR_PTR(ret); 11580f0fe8f7SFilipe David Borba Manana } 11595f39d397SChris Mason return buf; 1160ce9adaa5SChris Mason 1161eb60ceacSChris Mason } 1162eb60ceacSChris Mason 116301d58472SDaniel Dressler void clean_tree_block(struct btrfs_trans_handle *trans, 116401d58472SDaniel Dressler struct btrfs_fs_info *fs_info, 11655f39d397SChris Mason struct extent_buffer *buf) 1166ed2ff2cbSChris Mason { 116755c69072SChris Mason if (btrfs_header_generation(buf) == 1168e2d84521SMiao Xie fs_info->running_transaction->transid) { 1169b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1170b4ce94deSChris Mason 1171b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1172e2d84521SMiao Xie __percpu_counter_add(&fs_info->dirty_metadata_bytes, 1173e2d84521SMiao Xie -buf->len, 1174e2d84521SMiao Xie fs_info->dirty_metadata_batch); 1175b9473439SChris Mason /* ugh, clear_extent_buffer_dirty needs to lock the page */ 1176b9473439SChris Mason btrfs_set_lock_blocking(buf); 11770b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 1178925baeddSChris Mason } 11795f39d397SChris Mason } 1180ed7b63ebSJosef Bacik } 11815f39d397SChris Mason 11828257b2dcSMiao Xie static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void) 11838257b2dcSMiao Xie { 11848257b2dcSMiao Xie struct btrfs_subvolume_writers *writers; 11858257b2dcSMiao Xie int ret; 11868257b2dcSMiao Xie 11878257b2dcSMiao Xie writers = kmalloc(sizeof(*writers), GFP_NOFS); 11888257b2dcSMiao Xie if (!writers) 11898257b2dcSMiao Xie return ERR_PTR(-ENOMEM); 11908257b2dcSMiao Xie 1191908c7f19STejun Heo ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL); 11928257b2dcSMiao Xie if (ret < 0) { 11938257b2dcSMiao Xie kfree(writers); 11948257b2dcSMiao Xie return ERR_PTR(ret); 11958257b2dcSMiao Xie } 11968257b2dcSMiao Xie 11978257b2dcSMiao Xie init_waitqueue_head(&writers->wait); 11988257b2dcSMiao Xie return writers; 11998257b2dcSMiao Xie } 12008257b2dcSMiao Xie 12018257b2dcSMiao Xie static void 12028257b2dcSMiao Xie btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers) 12038257b2dcSMiao Xie { 12048257b2dcSMiao Xie percpu_counter_destroy(&writers->counter); 12058257b2dcSMiao Xie kfree(writers); 12068257b2dcSMiao Xie } 12078257b2dcSMiao Xie 1208707e8a07SDavid Sterba static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize, 1209707e8a07SDavid Sterba struct btrfs_root *root, struct btrfs_fs_info *fs_info, 1210e20d96d6SChris Mason u64 objectid) 1211d97e63b6SChris Mason { 1212cfaa7295SChris Mason root->node = NULL; 1213a28ec197SChris Mason root->commit_root = NULL; 1214db94535dSChris Mason root->sectorsize = sectorsize; 1215db94535dSChris Mason root->nodesize = nodesize; 121687ee04ebSChris Mason root->stripesize = stripesize; 121727cdeb70SMiao Xie root->state = 0; 1218d68fc57bSYan, Zheng root->orphan_cleanup_state = 0; 12190b86a832SChris Mason 12200f7d52f4SChris Mason root->objectid = objectid; 12210f7d52f4SChris Mason root->last_trans = 0; 122213a8a7c8SYan, Zheng root->highest_objectid = 0; 1223eb73c1b7SMiao Xie root->nr_delalloc_inodes = 0; 1224199c2a9cSMiao Xie root->nr_ordered_extents = 0; 122558176a96SJosef Bacik root->name = NULL; 12266bef4d31SEric Paris root->inode_tree = RB_ROOT; 122716cdcec7SMiao Xie INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1228f0486c68SYan, Zheng root->block_rsv = NULL; 1229d68fc57bSYan, Zheng root->orphan_block_rsv = NULL; 12300b86a832SChris Mason 12310b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 12325d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 1233eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_inodes); 1234eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_root); 1235199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_extents); 1236199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_root); 12372ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 12382ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 1239d68fc57bSYan, Zheng spin_lock_init(&root->orphan_lock); 12405d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1241eb73c1b7SMiao Xie spin_lock_init(&root->delalloc_lock); 1242199c2a9cSMiao Xie spin_lock_init(&root->ordered_extent_lock); 1243f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 12442ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 12452ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 1246a2135011SChris Mason mutex_init(&root->objectid_mutex); 1247e02119d5SChris Mason mutex_init(&root->log_mutex); 124831f3d255SMiao Xie mutex_init(&root->ordered_extent_mutex); 1249573bfb72SMiao Xie mutex_init(&root->delalloc_mutex); 12507237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 12517237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 12527237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 12538b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[0]); 12548b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[1]); 12557237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 12567237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 12577237f183SYan Zheng atomic_set(&root->log_writers, 0); 12582ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 12598a35d95fSJosef Bacik atomic_set(&root->orphan_inodes, 0); 1260b0feb9d9SMiao Xie atomic_set(&root->refs, 1); 12618257b2dcSMiao Xie atomic_set(&root->will_be_snapshoted, 0); 12627237f183SYan Zheng root->log_transid = 0; 1263d1433debSMiao Xie root->log_transid_committed = -1; 1264257c62e1SChris Mason root->last_log_commit = 0; 126506ea65a3SJosef Bacik if (fs_info) 1266d0c803c4SChris Mason extent_io_tree_init(&root->dirty_log_pages, 1267f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 1268017e5369SChris Mason 12693768f368SChris Mason memset(&root->root_key, 0, sizeof(root->root_key)); 12703768f368SChris Mason memset(&root->root_item, 0, sizeof(root->root_item)); 12716702ed49SChris Mason memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 127206ea65a3SJosef Bacik if (fs_info) 12733f157a2fSChris Mason root->defrag_trans_start = fs_info->generation; 127406ea65a3SJosef Bacik else 127506ea65a3SJosef Bacik root->defrag_trans_start = 0; 12764d775673SChris Mason root->root_key.objectid = objectid; 12770ee5dc67SAl Viro root->anon_dev = 0; 12788ea05e3aSAlexander Block 12795f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 12803768f368SChris Mason } 12813768f368SChris Mason 1282f84a8bd6SAl Viro static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info) 12836f07e42eSAl Viro { 12846f07e42eSAl Viro struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS); 12856f07e42eSAl Viro if (root) 12866f07e42eSAl Viro root->fs_info = fs_info; 12876f07e42eSAl Viro return root; 12886f07e42eSAl Viro } 12896f07e42eSAl Viro 129006ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 129106ea65a3SJosef Bacik /* Should only be used by the testing infrastructure */ 129206ea65a3SJosef Bacik struct btrfs_root *btrfs_alloc_dummy_root(void) 129306ea65a3SJosef Bacik { 129406ea65a3SJosef Bacik struct btrfs_root *root; 129506ea65a3SJosef Bacik 129606ea65a3SJosef Bacik root = btrfs_alloc_root(NULL); 129706ea65a3SJosef Bacik if (!root) 129806ea65a3SJosef Bacik return ERR_PTR(-ENOMEM); 1299707e8a07SDavid Sterba __setup_root(4096, 4096, 4096, root, NULL, 1); 130027cdeb70SMiao Xie set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state); 1301faa2dbf0SJosef Bacik root->alloc_bytenr = 0; 130206ea65a3SJosef Bacik 130306ea65a3SJosef Bacik return root; 130406ea65a3SJosef Bacik } 130506ea65a3SJosef Bacik #endif 130606ea65a3SJosef Bacik 130720897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 130820897f5cSArne Jansen struct btrfs_fs_info *fs_info, 130920897f5cSArne Jansen u64 objectid) 131020897f5cSArne Jansen { 131120897f5cSArne Jansen struct extent_buffer *leaf; 131220897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 131320897f5cSArne Jansen struct btrfs_root *root; 131420897f5cSArne Jansen struct btrfs_key key; 131520897f5cSArne Jansen int ret = 0; 13166463fe58SStefan Behrens uuid_le uuid; 131720897f5cSArne Jansen 131820897f5cSArne Jansen root = btrfs_alloc_root(fs_info); 131920897f5cSArne Jansen if (!root) 132020897f5cSArne Jansen return ERR_PTR(-ENOMEM); 132120897f5cSArne Jansen 1322707e8a07SDavid Sterba __setup_root(tree_root->nodesize, tree_root->sectorsize, 1323707e8a07SDavid Sterba tree_root->stripesize, root, fs_info, objectid); 132420897f5cSArne Jansen root->root_key.objectid = objectid; 132520897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 132620897f5cSArne Jansen root->root_key.offset = 0; 132720897f5cSArne Jansen 13284d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0); 132920897f5cSArne Jansen if (IS_ERR(leaf)) { 133020897f5cSArne Jansen ret = PTR_ERR(leaf); 13311dd05682STsutomu Itoh leaf = NULL; 133220897f5cSArne Jansen goto fail; 133320897f5cSArne Jansen } 133420897f5cSArne Jansen 133520897f5cSArne Jansen memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 133620897f5cSArne Jansen btrfs_set_header_bytenr(leaf, leaf->start); 133720897f5cSArne Jansen btrfs_set_header_generation(leaf, trans->transid); 133820897f5cSArne Jansen btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 133920897f5cSArne Jansen btrfs_set_header_owner(leaf, objectid); 134020897f5cSArne Jansen root->node = leaf; 134120897f5cSArne Jansen 13420a4e5586SRoss Kirk write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(), 134320897f5cSArne Jansen BTRFS_FSID_SIZE); 134420897f5cSArne Jansen write_extent_buffer(leaf, fs_info->chunk_tree_uuid, 1345b308bc2fSGeert Uytterhoeven btrfs_header_chunk_tree_uuid(leaf), 134620897f5cSArne Jansen BTRFS_UUID_SIZE); 134720897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 134820897f5cSArne Jansen 134920897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 135027cdeb70SMiao Xie set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 135120897f5cSArne Jansen 135220897f5cSArne Jansen root->root_item.flags = 0; 135320897f5cSArne Jansen root->root_item.byte_limit = 0; 135420897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 135520897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 135620897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 135720897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 135820897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 135920897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 136020897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 13616463fe58SStefan Behrens uuid_le_gen(&uuid); 13626463fe58SStefan Behrens memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE); 136320897f5cSArne Jansen root->root_item.drop_level = 0; 136420897f5cSArne Jansen 136520897f5cSArne Jansen key.objectid = objectid; 136620897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 136720897f5cSArne Jansen key.offset = 0; 136820897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 136920897f5cSArne Jansen if (ret) 137020897f5cSArne Jansen goto fail; 137120897f5cSArne Jansen 137220897f5cSArne Jansen btrfs_tree_unlock(leaf); 137320897f5cSArne Jansen 137420897f5cSArne Jansen return root; 13751dd05682STsutomu Itoh 13761dd05682STsutomu Itoh fail: 13771dd05682STsutomu Itoh if (leaf) { 13781dd05682STsutomu Itoh btrfs_tree_unlock(leaf); 137959885b39STsutomu Itoh free_extent_buffer(root->commit_root); 13801dd05682STsutomu Itoh free_extent_buffer(leaf); 13811dd05682STsutomu Itoh } 13821dd05682STsutomu Itoh kfree(root); 13831dd05682STsutomu Itoh 13841dd05682STsutomu Itoh return ERR_PTR(ret); 138520897f5cSArne Jansen } 138620897f5cSArne Jansen 13877237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1388e02119d5SChris Mason struct btrfs_fs_info *fs_info) 13890f7d52f4SChris Mason { 13900f7d52f4SChris Mason struct btrfs_root *root; 13910f7d52f4SChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 13927237f183SYan Zheng struct extent_buffer *leaf; 1393e02119d5SChris Mason 13946f07e42eSAl Viro root = btrfs_alloc_root(fs_info); 1395e02119d5SChris Mason if (!root) 13967237f183SYan Zheng return ERR_PTR(-ENOMEM); 1397e02119d5SChris Mason 1398707e8a07SDavid Sterba __setup_root(tree_root->nodesize, tree_root->sectorsize, 1399707e8a07SDavid Sterba tree_root->stripesize, root, fs_info, 1400707e8a07SDavid Sterba BTRFS_TREE_LOG_OBJECTID); 1401e02119d5SChris Mason 1402e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1403e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1404e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 140527cdeb70SMiao Xie 14067237f183SYan Zheng /* 140727cdeb70SMiao Xie * DON'T set REF_COWS for log trees 140827cdeb70SMiao Xie * 14097237f183SYan Zheng * log trees do not get reference counted because they go away 14107237f183SYan Zheng * before a real commit is actually done. They do store pointers 14117237f183SYan Zheng * to file data extents, and those reference counts still get 14127237f183SYan Zheng * updated (along with back refs to the log tree). 14137237f183SYan Zheng */ 1414e02119d5SChris Mason 14154d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 14164d75f8a9SDavid Sterba NULL, 0, 0, 0); 14177237f183SYan Zheng if (IS_ERR(leaf)) { 14187237f183SYan Zheng kfree(root); 14197237f183SYan Zheng return ERR_CAST(leaf); 14207237f183SYan Zheng } 1421e02119d5SChris Mason 14225d4f98a2SYan Zheng memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 14235d4f98a2SYan Zheng btrfs_set_header_bytenr(leaf, leaf->start); 14245d4f98a2SYan Zheng btrfs_set_header_generation(leaf, trans->transid); 14255d4f98a2SYan Zheng btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 14265d4f98a2SYan Zheng btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID); 14277237f183SYan Zheng root->node = leaf; 1428e02119d5SChris Mason 1429e02119d5SChris Mason write_extent_buffer(root->node, root->fs_info->fsid, 14300a4e5586SRoss Kirk btrfs_header_fsid(), BTRFS_FSID_SIZE); 1431e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1432e02119d5SChris Mason btrfs_tree_unlock(root->node); 14337237f183SYan Zheng return root; 14347237f183SYan Zheng } 14357237f183SYan Zheng 14367237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 14377237f183SYan Zheng struct btrfs_fs_info *fs_info) 14387237f183SYan Zheng { 14397237f183SYan Zheng struct btrfs_root *log_root; 14407237f183SYan Zheng 14417237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 14427237f183SYan Zheng if (IS_ERR(log_root)) 14437237f183SYan Zheng return PTR_ERR(log_root); 14447237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 14457237f183SYan Zheng fs_info->log_root_tree = log_root; 14467237f183SYan Zheng return 0; 14477237f183SYan Zheng } 14487237f183SYan Zheng 14497237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 14507237f183SYan Zheng struct btrfs_root *root) 14517237f183SYan Zheng { 14527237f183SYan Zheng struct btrfs_root *log_root; 14537237f183SYan Zheng struct btrfs_inode_item *inode_item; 14547237f183SYan Zheng 14557237f183SYan Zheng log_root = alloc_log_tree(trans, root->fs_info); 14567237f183SYan Zheng if (IS_ERR(log_root)) 14577237f183SYan Zheng return PTR_ERR(log_root); 14587237f183SYan Zheng 14597237f183SYan Zheng log_root->last_trans = trans->transid; 14607237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 14617237f183SYan Zheng 14627237f183SYan Zheng inode_item = &log_root->root_item.inode; 14633cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 14643cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 14653cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 1466707e8a07SDavid Sterba btrfs_set_stack_inode_nbytes(inode_item, root->nodesize); 14673cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 14687237f183SYan Zheng 14695d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 14707237f183SYan Zheng 14717237f183SYan Zheng WARN_ON(root->log_root); 14727237f183SYan Zheng root->log_root = log_root; 14737237f183SYan Zheng root->log_transid = 0; 1474d1433debSMiao Xie root->log_transid_committed = -1; 1475257c62e1SChris Mason root->last_log_commit = 0; 1476e02119d5SChris Mason return 0; 1477e02119d5SChris Mason } 1478e02119d5SChris Mason 147935a3621bSStefan Behrens static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 1480cb517eabSMiao Xie struct btrfs_key *key) 1481e02119d5SChris Mason { 1482e02119d5SChris Mason struct btrfs_root *root; 1483e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 14840f7d52f4SChris Mason struct btrfs_path *path; 148584234f3aSYan Zheng u64 generation; 1486cb517eabSMiao Xie int ret; 1487cb517eabSMiao Xie 1488cb517eabSMiao Xie path = btrfs_alloc_path(); 1489cb517eabSMiao Xie if (!path) 1490cb517eabSMiao Xie return ERR_PTR(-ENOMEM); 14910f7d52f4SChris Mason 14926f07e42eSAl Viro root = btrfs_alloc_root(fs_info); 1493cb517eabSMiao Xie if (!root) { 1494cb517eabSMiao Xie ret = -ENOMEM; 1495cb517eabSMiao Xie goto alloc_fail; 14960f7d52f4SChris Mason } 14970f7d52f4SChris Mason 1498707e8a07SDavid Sterba __setup_root(tree_root->nodesize, tree_root->sectorsize, 1499707e8a07SDavid Sterba tree_root->stripesize, root, fs_info, key->objectid); 15000f7d52f4SChris Mason 1501cb517eabSMiao Xie ret = btrfs_find_root(tree_root, key, path, 1502cb517eabSMiao Xie &root->root_item, &root->root_key); 15030f7d52f4SChris Mason if (ret) { 150413a8a7c8SYan, Zheng if (ret > 0) 150513a8a7c8SYan, Zheng ret = -ENOENT; 1506cb517eabSMiao Xie goto find_fail; 15070f7d52f4SChris Mason } 150813a8a7c8SYan, Zheng 150984234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1510db94535dSChris Mason root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item), 1511ce86cd59SDavid Sterba generation); 151264c043deSLiu Bo if (IS_ERR(root->node)) { 151364c043deSLiu Bo ret = PTR_ERR(root->node); 1514cb517eabSMiao Xie goto find_fail; 1515cb517eabSMiao Xie } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1516cb517eabSMiao Xie ret = -EIO; 151764c043deSLiu Bo free_extent_buffer(root->node); 151864c043deSLiu Bo goto find_fail; 1519416bc658SJosef Bacik } 15205d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 152113a8a7c8SYan, Zheng out: 1522cb517eabSMiao Xie btrfs_free_path(path); 1523cb517eabSMiao Xie return root; 1524cb517eabSMiao Xie 1525cb517eabSMiao Xie find_fail: 1526cb517eabSMiao Xie kfree(root); 1527cb517eabSMiao Xie alloc_fail: 1528cb517eabSMiao Xie root = ERR_PTR(ret); 1529cb517eabSMiao Xie goto out; 1530cb517eabSMiao Xie } 1531cb517eabSMiao Xie 1532cb517eabSMiao Xie struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root, 1533cb517eabSMiao Xie struct btrfs_key *location) 1534cb517eabSMiao Xie { 1535cb517eabSMiao Xie struct btrfs_root *root; 1536cb517eabSMiao Xie 1537cb517eabSMiao Xie root = btrfs_read_tree_root(tree_root, location); 1538cb517eabSMiao Xie if (IS_ERR(root)) 1539cb517eabSMiao Xie return root; 1540cb517eabSMiao Xie 1541cb517eabSMiao Xie if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { 154227cdeb70SMiao Xie set_bit(BTRFS_ROOT_REF_COWS, &root->state); 154308fe4db1SLi Zefan btrfs_check_and_init_root_item(&root->root_item); 154408fe4db1SLi Zefan } 154513a8a7c8SYan, Zheng 15465eda7b5eSChris Mason return root; 15475eda7b5eSChris Mason } 15485eda7b5eSChris Mason 1549cb517eabSMiao Xie int btrfs_init_fs_root(struct btrfs_root *root) 1550cb517eabSMiao Xie { 1551cb517eabSMiao Xie int ret; 15528257b2dcSMiao Xie struct btrfs_subvolume_writers *writers; 1553cb517eabSMiao Xie 1554cb517eabSMiao Xie root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); 1555cb517eabSMiao Xie root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), 1556cb517eabSMiao Xie GFP_NOFS); 1557cb517eabSMiao Xie if (!root->free_ino_pinned || !root->free_ino_ctl) { 1558cb517eabSMiao Xie ret = -ENOMEM; 1559cb517eabSMiao Xie goto fail; 1560cb517eabSMiao Xie } 1561cb517eabSMiao Xie 15628257b2dcSMiao Xie writers = btrfs_alloc_subvolume_writers(); 15638257b2dcSMiao Xie if (IS_ERR(writers)) { 15648257b2dcSMiao Xie ret = PTR_ERR(writers); 15658257b2dcSMiao Xie goto fail; 15668257b2dcSMiao Xie } 15678257b2dcSMiao Xie root->subv_writers = writers; 15688257b2dcSMiao Xie 1569cb517eabSMiao Xie btrfs_init_free_ino_ctl(root); 157057cdc8dbSDavid Sterba spin_lock_init(&root->ino_cache_lock); 157157cdc8dbSDavid Sterba init_waitqueue_head(&root->ino_cache_wait); 1572cb517eabSMiao Xie 1573cb517eabSMiao Xie ret = get_anon_bdev(&root->anon_dev); 1574cb517eabSMiao Xie if (ret) 15758257b2dcSMiao Xie goto free_writers; 1576cb517eabSMiao Xie return 0; 15778257b2dcSMiao Xie 15788257b2dcSMiao Xie free_writers: 15798257b2dcSMiao Xie btrfs_free_subvolume_writers(root->subv_writers); 1580cb517eabSMiao Xie fail: 1581cb517eabSMiao Xie kfree(root->free_ino_ctl); 1582cb517eabSMiao Xie kfree(root->free_ino_pinned); 1583cb517eabSMiao Xie return ret; 1584cb517eabSMiao Xie } 1585cb517eabSMiao Xie 1586171170c1SSergei Trofimovich static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1587cb517eabSMiao Xie u64 root_id) 1588cb517eabSMiao Xie { 1589cb517eabSMiao Xie struct btrfs_root *root; 1590cb517eabSMiao Xie 1591cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1592cb517eabSMiao Xie root = radix_tree_lookup(&fs_info->fs_roots_radix, 1593cb517eabSMiao Xie (unsigned long)root_id); 1594cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1595cb517eabSMiao Xie return root; 1596cb517eabSMiao Xie } 1597cb517eabSMiao Xie 1598cb517eabSMiao Xie int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1599cb517eabSMiao Xie struct btrfs_root *root) 1600cb517eabSMiao Xie { 1601cb517eabSMiao Xie int ret; 1602cb517eabSMiao Xie 1603cb517eabSMiao Xie ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); 1604cb517eabSMiao Xie if (ret) 1605cb517eabSMiao Xie return ret; 1606cb517eabSMiao Xie 1607cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1608cb517eabSMiao Xie ret = radix_tree_insert(&fs_info->fs_roots_radix, 1609cb517eabSMiao Xie (unsigned long)root->root_key.objectid, 1610cb517eabSMiao Xie root); 1611cb517eabSMiao Xie if (ret == 0) 161227cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1613cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1614cb517eabSMiao Xie radix_tree_preload_end(); 1615cb517eabSMiao Xie 1616cb517eabSMiao Xie return ret; 1617cb517eabSMiao Xie } 1618cb517eabSMiao Xie 1619c00869f1SMiao Xie struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 1620c00869f1SMiao Xie struct btrfs_key *location, 1621c00869f1SMiao Xie bool check_ref) 16225eda7b5eSChris Mason { 16235eda7b5eSChris Mason struct btrfs_root *root; 1624381cf658SDavid Sterba struct btrfs_path *path; 16251d4c08e0SDavid Sterba struct btrfs_key key; 16265eda7b5eSChris Mason int ret; 16275eda7b5eSChris Mason 1628edbd8d4eSChris Mason if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 1629edbd8d4eSChris Mason return fs_info->tree_root; 1630edbd8d4eSChris Mason if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID) 1631edbd8d4eSChris Mason return fs_info->extent_root; 16328f18cf13SChris Mason if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID) 16338f18cf13SChris Mason return fs_info->chunk_root; 16348f18cf13SChris Mason if (location->objectid == BTRFS_DEV_TREE_OBJECTID) 16358f18cf13SChris Mason return fs_info->dev_root; 16360403e47eSYan Zheng if (location->objectid == BTRFS_CSUM_TREE_OBJECTID) 16370403e47eSYan Zheng return fs_info->csum_root; 1638bcef60f2SArne Jansen if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID) 1639bcef60f2SArne Jansen return fs_info->quota_root ? fs_info->quota_root : 1640bcef60f2SArne Jansen ERR_PTR(-ENOENT); 1641f7a81ea4SStefan Behrens if (location->objectid == BTRFS_UUID_TREE_OBJECTID) 1642f7a81ea4SStefan Behrens return fs_info->uuid_root ? fs_info->uuid_root : 1643f7a81ea4SStefan Behrens ERR_PTR(-ENOENT); 16444df27c4dSYan, Zheng again: 1645cb517eabSMiao Xie root = btrfs_lookup_fs_root(fs_info, location->objectid); 164648475471SStefan Behrens if (root) { 1647c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) 164848475471SStefan Behrens return ERR_PTR(-ENOENT); 16495eda7b5eSChris Mason return root; 165048475471SStefan Behrens } 16515eda7b5eSChris Mason 1652cb517eabSMiao Xie root = btrfs_read_fs_root(fs_info->tree_root, location); 16535eda7b5eSChris Mason if (IS_ERR(root)) 16545eda7b5eSChris Mason return root; 16553394e160SChris Mason 1656c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1657d68fc57bSYan, Zheng ret = -ENOENT; 1658d68fc57bSYan, Zheng goto fail; 1659d68fc57bSYan, Zheng } 1660d68fc57bSYan, Zheng 1661cb517eabSMiao Xie ret = btrfs_init_fs_root(root); 1662cb517eabSMiao Xie if (ret) 1663cb517eabSMiao Xie goto fail; 1664cb517eabSMiao Xie 1665381cf658SDavid Sterba path = btrfs_alloc_path(); 1666381cf658SDavid Sterba if (!path) { 1667381cf658SDavid Sterba ret = -ENOMEM; 1668381cf658SDavid Sterba goto fail; 1669381cf658SDavid Sterba } 16701d4c08e0SDavid Sterba key.objectid = BTRFS_ORPHAN_OBJECTID; 16711d4c08e0SDavid Sterba key.type = BTRFS_ORPHAN_ITEM_KEY; 16721d4c08e0SDavid Sterba key.offset = location->objectid; 16731d4c08e0SDavid Sterba 16741d4c08e0SDavid Sterba ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1675381cf658SDavid Sterba btrfs_free_path(path); 1676d68fc57bSYan, Zheng if (ret < 0) 1677d68fc57bSYan, Zheng goto fail; 1678d68fc57bSYan, Zheng if (ret == 0) 167927cdeb70SMiao Xie set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1680d68fc57bSYan, Zheng 1681cb517eabSMiao Xie ret = btrfs_insert_fs_root(fs_info, root); 16820f7d52f4SChris Mason if (ret) { 16834df27c4dSYan, Zheng if (ret == -EEXIST) { 16844df27c4dSYan, Zheng free_fs_root(root); 16854df27c4dSYan, Zheng goto again; 16860f7d52f4SChris Mason } 16874df27c4dSYan, Zheng goto fail; 16884df27c4dSYan, Zheng } 1689edbd8d4eSChris Mason return root; 16904df27c4dSYan, Zheng fail: 16914df27c4dSYan, Zheng free_fs_root(root); 16924df27c4dSYan, Zheng return ERR_PTR(ret); 1693edbd8d4eSChris Mason } 1694edbd8d4eSChris Mason 169504160088SChris Mason static int btrfs_congested_fn(void *congested_data, int bdi_bits) 169604160088SChris Mason { 169704160088SChris Mason struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; 169804160088SChris Mason int ret = 0; 169904160088SChris Mason struct btrfs_device *device; 170004160088SChris Mason struct backing_dev_info *bdi; 1701b7967db7SChris Mason 17021f78160cSXiao Guangrong rcu_read_lock(); 17031f78160cSXiao Guangrong list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) { 1704dfe25020SChris Mason if (!device->bdev) 1705dfe25020SChris Mason continue; 170604160088SChris Mason bdi = blk_get_backing_dev_info(device->bdev); 1707ff9ea323STejun Heo if (bdi_congested(bdi, bdi_bits)) { 170804160088SChris Mason ret = 1; 170904160088SChris Mason break; 171004160088SChris Mason } 171104160088SChris Mason } 17121f78160cSXiao Guangrong rcu_read_unlock(); 171304160088SChris Mason return ret; 171404160088SChris Mason } 171504160088SChris Mason 171604160088SChris Mason static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi) 171704160088SChris Mason { 1718ad081f14SJens Axboe int err; 1719ad081f14SJens Axboe 1720b4caecd4SChristoph Hellwig err = bdi_setup_and_register(bdi, "btrfs"); 1721ad081f14SJens Axboe if (err) 1722ad081f14SJens Axboe return err; 1723ad081f14SJens Axboe 1724df0ce26cSChristoph Hellwig bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE; 172504160088SChris Mason bdi->congested_fn = btrfs_congested_fn; 172604160088SChris Mason bdi->congested_data = info; 1727da2f0f74SChris Mason bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK; 172804160088SChris Mason return 0; 172904160088SChris Mason } 173004160088SChris Mason 17318b712842SChris Mason /* 17328b712842SChris Mason * called by the kthread helper functions to finally call the bio end_io 17338b712842SChris Mason * functions. This is where read checksum verification actually happens 17348b712842SChris Mason */ 17358b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work) 1736ce9adaa5SChris Mason { 1737ce9adaa5SChris Mason struct bio *bio; 173897eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 1739ce9adaa5SChris Mason int error; 1740ce9adaa5SChris Mason 174197eb6b69SDavid Sterba end_io_wq = container_of(work, struct btrfs_end_io_wq, work); 1742ce9adaa5SChris Mason bio = end_io_wq->bio; 17438b712842SChris Mason 1744ce9adaa5SChris Mason error = end_io_wq->error; 1745ce9adaa5SChris Mason bio->bi_private = end_io_wq->private; 1746ce9adaa5SChris Mason bio->bi_end_io = end_io_wq->end_io; 174797eb6b69SDavid Sterba kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq); 1748326e1dbbSMike Snitzer bio_endio(bio, error); 1749ce9adaa5SChris Mason } 175044b8bd7eSChris Mason 1751a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1752a74a4b97SChris Mason { 1753a74a4b97SChris Mason struct btrfs_root *root = arg; 1754d0278245SMiao Xie int again; 1755da288d28SFilipe Manana struct btrfs_trans_handle *trans; 1756a74a4b97SChris Mason 1757a74a4b97SChris Mason do { 1758d0278245SMiao Xie again = 0; 17599d1a2a3aSDavid Sterba 1760d0278245SMiao Xie /* Make the cleaner go to sleep early. */ 1761babbf170SMiao Xie if (btrfs_need_cleaner_sleep(root)) 1762d0278245SMiao Xie goto sleep; 1763d0278245SMiao Xie 1764d0278245SMiao Xie if (!mutex_trylock(&root->fs_info->cleaner_mutex)) 1765d0278245SMiao Xie goto sleep; 1766d0278245SMiao Xie 1767dc7f370cSMiao Xie /* 1768dc7f370cSMiao Xie * Avoid the problem that we change the status of the fs 1769dc7f370cSMiao Xie * during the above check and trylock. 1770dc7f370cSMiao Xie */ 1771babbf170SMiao Xie if (btrfs_need_cleaner_sleep(root)) { 1772dc7f370cSMiao Xie mutex_unlock(&root->fs_info->cleaner_mutex); 1773dc7f370cSMiao Xie goto sleep; 1774dc7f370cSMiao Xie } 1775dc7f370cSMiao Xie 177624bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 17779d1a2a3aSDavid Sterba again = btrfs_clean_one_deleted_snapshot(root); 1778a74a4b97SChris Mason mutex_unlock(&root->fs_info->cleaner_mutex); 1779a74a4b97SChris Mason 1780d0278245SMiao Xie /* 178105323cd1SMiao Xie * The defragger has dealt with the R/O remount and umount, 178205323cd1SMiao Xie * needn't do anything special here. 1783d0278245SMiao Xie */ 1784d0278245SMiao Xie btrfs_run_defrag_inodes(root->fs_info); 178567c5e7d4SFilipe Manana 178667c5e7d4SFilipe Manana /* 178767c5e7d4SFilipe Manana * Acquires fs_info->delete_unused_bgs_mutex to avoid racing 178867c5e7d4SFilipe Manana * with relocation (btrfs_relocate_chunk) and relocation 178967c5e7d4SFilipe Manana * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 179067c5e7d4SFilipe Manana * after acquiring fs_info->delete_unused_bgs_mutex. So we 179167c5e7d4SFilipe Manana * can't hold, nor need to, fs_info->cleaner_mutex when deleting 179267c5e7d4SFilipe Manana * unused block groups. 179367c5e7d4SFilipe Manana */ 179467c5e7d4SFilipe Manana btrfs_delete_unused_bgs(root->fs_info); 1795d0278245SMiao Xie sleep: 17969d1a2a3aSDavid Sterba if (!try_to_freeze() && !again) { 1797a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 17988929ecfaSYan, Zheng if (!kthread_should_stop()) 1799a74a4b97SChris Mason schedule(); 1800a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1801a74a4b97SChris Mason } 1802a74a4b97SChris Mason } while (!kthread_should_stop()); 1803da288d28SFilipe Manana 1804da288d28SFilipe Manana /* 1805da288d28SFilipe Manana * Transaction kthread is stopped before us and wakes us up. 1806da288d28SFilipe Manana * However we might have started a new transaction and COWed some 1807da288d28SFilipe Manana * tree blocks when deleting unused block groups for example. So 1808da288d28SFilipe Manana * make sure we commit the transaction we started to have a clean 1809da288d28SFilipe Manana * shutdown when evicting the btree inode - if it has dirty pages 1810da288d28SFilipe Manana * when we do the final iput() on it, eviction will trigger a 1811da288d28SFilipe Manana * writeback for it which will fail with null pointer dereferences 1812da288d28SFilipe Manana * since work queues and other resources were already released and 1813da288d28SFilipe Manana * destroyed by the time the iput/eviction/writeback is made. 1814da288d28SFilipe Manana */ 1815da288d28SFilipe Manana trans = btrfs_attach_transaction(root); 1816da288d28SFilipe Manana if (IS_ERR(trans)) { 1817da288d28SFilipe Manana if (PTR_ERR(trans) != -ENOENT) 1818da288d28SFilipe Manana btrfs_err(root->fs_info, 1819da288d28SFilipe Manana "cleaner transaction attach returned %ld", 1820da288d28SFilipe Manana PTR_ERR(trans)); 1821da288d28SFilipe Manana } else { 1822da288d28SFilipe Manana int ret; 1823da288d28SFilipe Manana 1824da288d28SFilipe Manana ret = btrfs_commit_transaction(trans, root); 1825da288d28SFilipe Manana if (ret) 1826da288d28SFilipe Manana btrfs_err(root->fs_info, 1827da288d28SFilipe Manana "cleaner open transaction commit returned %d", 1828da288d28SFilipe Manana ret); 1829da288d28SFilipe Manana } 1830da288d28SFilipe Manana 1831a74a4b97SChris Mason return 0; 1832a74a4b97SChris Mason } 1833a74a4b97SChris Mason 1834a74a4b97SChris Mason static int transaction_kthread(void *arg) 1835a74a4b97SChris Mason { 1836a74a4b97SChris Mason struct btrfs_root *root = arg; 1837a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1838a74a4b97SChris Mason struct btrfs_transaction *cur; 18398929ecfaSYan, Zheng u64 transid; 1840a74a4b97SChris Mason unsigned long now; 1841a74a4b97SChris Mason unsigned long delay; 1842914b2007SJan Kara bool cannot_commit; 1843a74a4b97SChris Mason 1844a74a4b97SChris Mason do { 1845914b2007SJan Kara cannot_commit = false; 18468b87dc17SDavid Sterba delay = HZ * root->fs_info->commit_interval; 1847a74a4b97SChris Mason mutex_lock(&root->fs_info->transaction_kthread_mutex); 1848a74a4b97SChris Mason 1849a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 1850a74a4b97SChris Mason cur = root->fs_info->running_transaction; 1851a74a4b97SChris Mason if (!cur) { 1852a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1853a74a4b97SChris Mason goto sleep; 1854a74a4b97SChris Mason } 185531153d81SYan Zheng 1856a74a4b97SChris Mason now = get_seconds(); 18574a9d8bdeSMiao Xie if (cur->state < TRANS_STATE_BLOCKED && 18588b87dc17SDavid Sterba (now < cur->start_time || 18598b87dc17SDavid Sterba now - cur->start_time < root->fs_info->commit_interval)) { 1860a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1861a74a4b97SChris Mason delay = HZ * 5; 1862a74a4b97SChris Mason goto sleep; 1863a74a4b97SChris Mason } 18648929ecfaSYan, Zheng transid = cur->transid; 1865a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 186656bec294SChris Mason 186779787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 1868354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 1869914b2007SJan Kara if (IS_ERR(trans)) { 1870354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 1871914b2007SJan Kara cannot_commit = true; 187279787eaaSJeff Mahoney goto sleep; 1873914b2007SJan Kara } 18748929ecfaSYan, Zheng if (transid == trans->transid) { 187579787eaaSJeff Mahoney btrfs_commit_transaction(trans, root); 18768929ecfaSYan, Zheng } else { 18778929ecfaSYan, Zheng btrfs_end_transaction(trans, root); 18788929ecfaSYan, Zheng } 1879a74a4b97SChris Mason sleep: 1880a74a4b97SChris Mason wake_up_process(root->fs_info->cleaner_kthread); 1881a74a4b97SChris Mason mutex_unlock(&root->fs_info->transaction_kthread_mutex); 1882a74a4b97SChris Mason 18834e121c06SJosef Bacik if (unlikely(test_bit(BTRFS_FS_STATE_ERROR, 18844e121c06SJosef Bacik &root->fs_info->fs_state))) 18854e121c06SJosef Bacik btrfs_cleanup_transaction(root); 1886a0acae0eSTejun Heo if (!try_to_freeze()) { 1887a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 18888929ecfaSYan, Zheng if (!kthread_should_stop() && 1889914b2007SJan Kara (!btrfs_transaction_blocked(root->fs_info) || 1890914b2007SJan Kara cannot_commit)) 1891a74a4b97SChris Mason schedule_timeout(delay); 1892a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1893a74a4b97SChris Mason } 1894a74a4b97SChris Mason } while (!kthread_should_stop()); 1895a74a4b97SChris Mason return 0; 1896a74a4b97SChris Mason } 1897a74a4b97SChris Mason 1898af31f5e5SChris Mason /* 1899af31f5e5SChris Mason * this will find the highest generation in the array of 1900af31f5e5SChris Mason * root backups. The index of the highest array is returned, 1901af31f5e5SChris Mason * or -1 if we can't find anything. 1902af31f5e5SChris Mason * 1903af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1904af31f5e5SChris Mason * generation of the latest root in the array with the generation 1905af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1906af31f5e5SChris Mason */ 1907af31f5e5SChris Mason static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen) 1908af31f5e5SChris Mason { 1909af31f5e5SChris Mason u64 cur; 1910af31f5e5SChris Mason int newest_index = -1; 1911af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1912af31f5e5SChris Mason int i; 1913af31f5e5SChris Mason 1914af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1915af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1916af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1917af31f5e5SChris Mason if (cur == newest_gen) 1918af31f5e5SChris Mason newest_index = i; 1919af31f5e5SChris Mason } 1920af31f5e5SChris Mason 1921af31f5e5SChris Mason /* check to see if we actually wrapped around */ 1922af31f5e5SChris Mason if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) { 1923af31f5e5SChris Mason root_backup = info->super_copy->super_roots; 1924af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1925af31f5e5SChris Mason if (cur == newest_gen) 1926af31f5e5SChris Mason newest_index = 0; 1927af31f5e5SChris Mason } 1928af31f5e5SChris Mason return newest_index; 1929af31f5e5SChris Mason } 1930af31f5e5SChris Mason 1931af31f5e5SChris Mason 1932af31f5e5SChris Mason /* 1933af31f5e5SChris Mason * find the oldest backup so we know where to store new entries 1934af31f5e5SChris Mason * in the backup array. This will set the backup_root_index 1935af31f5e5SChris Mason * field in the fs_info struct 1936af31f5e5SChris Mason */ 1937af31f5e5SChris Mason static void find_oldest_super_backup(struct btrfs_fs_info *info, 1938af31f5e5SChris Mason u64 newest_gen) 1939af31f5e5SChris Mason { 1940af31f5e5SChris Mason int newest_index = -1; 1941af31f5e5SChris Mason 1942af31f5e5SChris Mason newest_index = find_newest_super_backup(info, newest_gen); 1943af31f5e5SChris Mason /* if there was garbage in there, just move along */ 1944af31f5e5SChris Mason if (newest_index == -1) { 1945af31f5e5SChris Mason info->backup_root_index = 0; 1946af31f5e5SChris Mason } else { 1947af31f5e5SChris Mason info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS; 1948af31f5e5SChris Mason } 1949af31f5e5SChris Mason } 1950af31f5e5SChris Mason 1951af31f5e5SChris Mason /* 1952af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1953af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1954af31f5e5SChris Mason * done 1955af31f5e5SChris Mason */ 1956af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1957af31f5e5SChris Mason { 1958af31f5e5SChris Mason int next_backup; 1959af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1960af31f5e5SChris Mason int last_backup; 1961af31f5e5SChris Mason 1962af31f5e5SChris Mason next_backup = info->backup_root_index; 1963af31f5e5SChris Mason last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) % 1964af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 1965af31f5e5SChris Mason 1966af31f5e5SChris Mason /* 1967af31f5e5SChris Mason * just overwrite the last backup if we're at the same generation 1968af31f5e5SChris Mason * this happens only at umount 1969af31f5e5SChris Mason */ 1970af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + last_backup; 1971af31f5e5SChris Mason if (btrfs_backup_tree_root_gen(root_backup) == 1972af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)) 1973af31f5e5SChris Mason next_backup = last_backup; 1974af31f5e5SChris Mason 1975af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1976af31f5e5SChris Mason 1977af31f5e5SChris Mason /* 1978af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1979af31f5e5SChris Mason * regardless of which ones we use today 1980af31f5e5SChris Mason */ 1981af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1982af31f5e5SChris Mason 1983af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1984af31f5e5SChris Mason 1985af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1986af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1987af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1988af31f5e5SChris Mason 1989af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1990af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1991af31f5e5SChris Mason 1992af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1993af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1994af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 1995af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 1996af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 1997af31f5e5SChris Mason 1998af31f5e5SChris Mason btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); 1999af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 2000af31f5e5SChris Mason btrfs_header_generation(info->extent_root->node)); 2001af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 2002af31f5e5SChris Mason btrfs_header_level(info->extent_root->node)); 2003af31f5e5SChris Mason 20047c7e82a7SChris Mason /* 20057c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 20067c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 20077c7e82a7SChris Mason */ 20087c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 20097c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 20107c7e82a7SChris Mason info->fs_root->node->start); 2011af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 2012af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 2013af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 2014af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 20157c7e82a7SChris Mason } 2016af31f5e5SChris Mason 2017af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 2018af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 2019af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 2020af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 2021af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 2022af31f5e5SChris Mason 2023af31f5e5SChris Mason btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); 2024af31f5e5SChris Mason btrfs_set_backup_csum_root_gen(root_backup, 2025af31f5e5SChris Mason btrfs_header_generation(info->csum_root->node)); 2026af31f5e5SChris Mason btrfs_set_backup_csum_root_level(root_backup, 2027af31f5e5SChris Mason btrfs_header_level(info->csum_root->node)); 2028af31f5e5SChris Mason 2029af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 2030af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 2031af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 2032af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 2033af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 2034af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 2035af31f5e5SChris Mason 2036af31f5e5SChris Mason /* 2037af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 2038af31f5e5SChris Mason * for the next commit 2039af31f5e5SChris Mason */ 2040af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 2041af31f5e5SChris Mason &info->super_for_commit->super_roots, 2042af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 2043af31f5e5SChris Mason } 2044af31f5e5SChris Mason 2045af31f5e5SChris Mason /* 2046af31f5e5SChris Mason * this copies info out of the root backup array and back into 2047af31f5e5SChris Mason * the in-memory super block. It is meant to help iterate through 2048af31f5e5SChris Mason * the array, so you send it the number of backups you've already 2049af31f5e5SChris Mason * tried and the last backup index you used. 2050af31f5e5SChris Mason * 2051af31f5e5SChris Mason * this returns -1 when it has tried all the backups 2052af31f5e5SChris Mason */ 2053af31f5e5SChris Mason static noinline int next_root_backup(struct btrfs_fs_info *info, 2054af31f5e5SChris Mason struct btrfs_super_block *super, 2055af31f5e5SChris Mason int *num_backups_tried, int *backup_index) 2056af31f5e5SChris Mason { 2057af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 2058af31f5e5SChris Mason int newest = *backup_index; 2059af31f5e5SChris Mason 2060af31f5e5SChris Mason if (*num_backups_tried == 0) { 2061af31f5e5SChris Mason u64 gen = btrfs_super_generation(super); 2062af31f5e5SChris Mason 2063af31f5e5SChris Mason newest = find_newest_super_backup(info, gen); 2064af31f5e5SChris Mason if (newest == -1) 2065af31f5e5SChris Mason return -1; 2066af31f5e5SChris Mason 2067af31f5e5SChris Mason *backup_index = newest; 2068af31f5e5SChris Mason *num_backups_tried = 1; 2069af31f5e5SChris Mason } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) { 2070af31f5e5SChris Mason /* we've tried all the backups, all done */ 2071af31f5e5SChris Mason return -1; 2072af31f5e5SChris Mason } else { 2073af31f5e5SChris Mason /* jump to the next oldest backup */ 2074af31f5e5SChris Mason newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) % 2075af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 2076af31f5e5SChris Mason *backup_index = newest; 2077af31f5e5SChris Mason *num_backups_tried += 1; 2078af31f5e5SChris Mason } 2079af31f5e5SChris Mason root_backup = super->super_roots + newest; 2080af31f5e5SChris Mason 2081af31f5e5SChris Mason btrfs_set_super_generation(super, 2082af31f5e5SChris Mason btrfs_backup_tree_root_gen(root_backup)); 2083af31f5e5SChris Mason btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 2084af31f5e5SChris Mason btrfs_set_super_root_level(super, 2085af31f5e5SChris Mason btrfs_backup_tree_root_level(root_backup)); 2086af31f5e5SChris Mason btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 2087af31f5e5SChris Mason 2088af31f5e5SChris Mason /* 2089af31f5e5SChris Mason * fixme: the total bytes and num_devices need to match or we should 2090af31f5e5SChris Mason * need a fsck 2091af31f5e5SChris Mason */ 2092af31f5e5SChris Mason btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 2093af31f5e5SChris Mason btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 2094af31f5e5SChris Mason return 0; 2095af31f5e5SChris Mason } 2096af31f5e5SChris Mason 20977abadb64SLiu Bo /* helper to cleanup workers */ 20987abadb64SLiu Bo static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 20997abadb64SLiu Bo { 2100dc6e3209SQu Wenruo btrfs_destroy_workqueue(fs_info->fixup_workers); 2101afe3d242SQu Wenruo btrfs_destroy_workqueue(fs_info->delalloc_workers); 21025cdc7ad3SQu Wenruo btrfs_destroy_workqueue(fs_info->workers); 2103fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_workers); 2104fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_meta_workers); 2105fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_raid56_workers); 21068b110e39SMiao Xie btrfs_destroy_workqueue(fs_info->endio_repair_workers); 2107d05a33acSQu Wenruo btrfs_destroy_workqueue(fs_info->rmw_workers); 2108fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_meta_write_workers); 2109fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_write_workers); 2110fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 2111a8c93d4eSQu Wenruo btrfs_destroy_workqueue(fs_info->submit_workers); 21125b3bc44eSQu Wenruo btrfs_destroy_workqueue(fs_info->delayed_workers); 2113e66f0bb1SQu Wenruo btrfs_destroy_workqueue(fs_info->caching_workers); 2114736cfa15SQu Wenruo btrfs_destroy_workqueue(fs_info->readahead_workers); 2115a44903abSQu Wenruo btrfs_destroy_workqueue(fs_info->flush_workers); 2116fc97fab0SQu Wenruo btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2117a79b7d4bSChris Mason btrfs_destroy_workqueue(fs_info->extent_workers); 21187abadb64SLiu Bo } 21197abadb64SLiu Bo 21202e9f5954SRashika static void free_root_extent_buffers(struct btrfs_root *root) 21212e9f5954SRashika { 21222e9f5954SRashika if (root) { 21232e9f5954SRashika free_extent_buffer(root->node); 21242e9f5954SRashika free_extent_buffer(root->commit_root); 21252e9f5954SRashika root->node = NULL; 21262e9f5954SRashika root->commit_root = NULL; 21272e9f5954SRashika } 21282e9f5954SRashika } 21292e9f5954SRashika 2130af31f5e5SChris Mason /* helper to cleanup tree roots */ 2131af31f5e5SChris Mason static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root) 2132af31f5e5SChris Mason { 21332e9f5954SRashika free_root_extent_buffers(info->tree_root); 2134655b09feSJosef Bacik 21352e9f5954SRashika free_root_extent_buffers(info->dev_root); 21362e9f5954SRashika free_root_extent_buffers(info->extent_root); 21372e9f5954SRashika free_root_extent_buffers(info->csum_root); 21382e9f5954SRashika free_root_extent_buffers(info->quota_root); 21392e9f5954SRashika free_root_extent_buffers(info->uuid_root); 21402e9f5954SRashika if (chunk_root) 21412e9f5954SRashika free_root_extent_buffers(info->chunk_root); 2142af31f5e5SChris Mason } 2143af31f5e5SChris Mason 2144faa2dbf0SJosef Bacik void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2145171f6537SJosef Bacik { 2146171f6537SJosef Bacik int ret; 2147171f6537SJosef Bacik struct btrfs_root *gang[8]; 2148171f6537SJosef Bacik int i; 2149171f6537SJosef Bacik 2150171f6537SJosef Bacik while (!list_empty(&fs_info->dead_roots)) { 2151171f6537SJosef Bacik gang[0] = list_entry(fs_info->dead_roots.next, 2152171f6537SJosef Bacik struct btrfs_root, root_list); 2153171f6537SJosef Bacik list_del(&gang[0]->root_list); 2154171f6537SJosef Bacik 215527cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) { 2156cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[0]); 2157171f6537SJosef Bacik } else { 2158171f6537SJosef Bacik free_extent_buffer(gang[0]->node); 2159171f6537SJosef Bacik free_extent_buffer(gang[0]->commit_root); 2160b0feb9d9SMiao Xie btrfs_put_fs_root(gang[0]); 2161171f6537SJosef Bacik } 2162171f6537SJosef Bacik } 2163171f6537SJosef Bacik 2164171f6537SJosef Bacik while (1) { 2165171f6537SJosef Bacik ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2166171f6537SJosef Bacik (void **)gang, 0, 2167171f6537SJosef Bacik ARRAY_SIZE(gang)); 2168171f6537SJosef Bacik if (!ret) 2169171f6537SJosef Bacik break; 2170171f6537SJosef Bacik for (i = 0; i < ret; i++) 2171cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2172171f6537SJosef Bacik } 21731a4319ccSLiu Bo 21741a4319ccSLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 21751a4319ccSLiu Bo btrfs_free_log_root_tree(NULL, fs_info); 21761a4319ccSLiu Bo btrfs_destroy_pinned_extent(fs_info->tree_root, 21771a4319ccSLiu Bo fs_info->pinned_extents); 21781a4319ccSLiu Bo } 2179171f6537SJosef Bacik } 2180af31f5e5SChris Mason 2181638aa7edSEric Sandeen static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2182638aa7edSEric Sandeen { 2183638aa7edSEric Sandeen mutex_init(&fs_info->scrub_lock); 2184638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_running, 0); 2185638aa7edSEric Sandeen atomic_set(&fs_info->scrub_pause_req, 0); 2186638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_paused, 0); 2187638aa7edSEric Sandeen atomic_set(&fs_info->scrub_cancel_req, 0); 2188638aa7edSEric Sandeen init_waitqueue_head(&fs_info->scrub_pause_wait); 2189638aa7edSEric Sandeen fs_info->scrub_workers_refcnt = 0; 2190638aa7edSEric Sandeen } 2191638aa7edSEric Sandeen 2192779a65a4SEric Sandeen static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2193779a65a4SEric Sandeen { 2194779a65a4SEric Sandeen spin_lock_init(&fs_info->balance_lock); 2195779a65a4SEric Sandeen mutex_init(&fs_info->balance_mutex); 2196779a65a4SEric Sandeen atomic_set(&fs_info->balance_running, 0); 2197779a65a4SEric Sandeen atomic_set(&fs_info->balance_pause_req, 0); 2198779a65a4SEric Sandeen atomic_set(&fs_info->balance_cancel_req, 0); 2199779a65a4SEric Sandeen fs_info->balance_ctl = NULL; 2200779a65a4SEric Sandeen init_waitqueue_head(&fs_info->balance_wait_q); 2201779a65a4SEric Sandeen } 2202779a65a4SEric Sandeen 2203f37938e0SEric Sandeen static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info, 2204f37938e0SEric Sandeen struct btrfs_root *tree_root) 2205f37938e0SEric Sandeen { 2206f37938e0SEric Sandeen fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 2207f37938e0SEric Sandeen set_nlink(fs_info->btree_inode, 1); 2208f37938e0SEric Sandeen /* 2209f37938e0SEric Sandeen * we set the i_size on the btree inode to the max possible int. 2210f37938e0SEric Sandeen * the real end of the address space is determined by all of 2211f37938e0SEric Sandeen * the devices in the system 2212f37938e0SEric Sandeen */ 2213f37938e0SEric Sandeen fs_info->btree_inode->i_size = OFFSET_MAX; 2214f37938e0SEric Sandeen fs_info->btree_inode->i_mapping->a_ops = &btree_aops; 2215f37938e0SEric Sandeen 2216f37938e0SEric Sandeen RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node); 2217f37938e0SEric Sandeen extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree, 2218f37938e0SEric Sandeen fs_info->btree_inode->i_mapping); 2219f37938e0SEric Sandeen BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0; 2220f37938e0SEric Sandeen extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree); 2221f37938e0SEric Sandeen 2222f37938e0SEric Sandeen BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops; 2223f37938e0SEric Sandeen 2224f37938e0SEric Sandeen BTRFS_I(fs_info->btree_inode)->root = tree_root; 2225f37938e0SEric Sandeen memset(&BTRFS_I(fs_info->btree_inode)->location, 0, 2226f37938e0SEric Sandeen sizeof(struct btrfs_key)); 2227f37938e0SEric Sandeen set_bit(BTRFS_INODE_DUMMY, 2228f37938e0SEric Sandeen &BTRFS_I(fs_info->btree_inode)->runtime_flags); 2229f37938e0SEric Sandeen btrfs_insert_inode_hash(fs_info->btree_inode); 2230f37938e0SEric Sandeen } 2231f37938e0SEric Sandeen 2232ad618368SEric Sandeen static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2233ad618368SEric Sandeen { 2234ad618368SEric Sandeen fs_info->dev_replace.lock_owner = 0; 2235ad618368SEric Sandeen atomic_set(&fs_info->dev_replace.nesting_level, 0); 2236ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2237ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_management_lock); 2238ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock); 2239ad618368SEric Sandeen init_waitqueue_head(&fs_info->replace_wait); 2240ad618368SEric Sandeen } 2241ad618368SEric Sandeen 2242f9e92e40SEric Sandeen static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2243f9e92e40SEric Sandeen { 2244f9e92e40SEric Sandeen spin_lock_init(&fs_info->qgroup_lock); 2245f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_ioctl_lock); 2246f9e92e40SEric Sandeen fs_info->qgroup_tree = RB_ROOT; 2247f9e92e40SEric Sandeen fs_info->qgroup_op_tree = RB_ROOT; 2248f9e92e40SEric Sandeen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2249f9e92e40SEric Sandeen fs_info->qgroup_seq = 1; 2250f9e92e40SEric Sandeen fs_info->quota_enabled = 0; 2251f9e92e40SEric Sandeen fs_info->pending_quota_state = 0; 2252f9e92e40SEric Sandeen fs_info->qgroup_ulist = NULL; 2253f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_rescan_lock); 2254f9e92e40SEric Sandeen } 2255f9e92e40SEric Sandeen 22562a458198SEric Sandeen static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info, 22572a458198SEric Sandeen struct btrfs_fs_devices *fs_devices) 22582a458198SEric Sandeen { 22592a458198SEric Sandeen int max_active = fs_info->thread_pool_size; 22606f011058SDavid Sterba unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 22612a458198SEric Sandeen 22622a458198SEric Sandeen fs_info->workers = 22632a458198SEric Sandeen btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI, 22642a458198SEric Sandeen max_active, 16); 22652a458198SEric Sandeen 22662a458198SEric Sandeen fs_info->delalloc_workers = 22672a458198SEric Sandeen btrfs_alloc_workqueue("delalloc", flags, max_active, 2); 22682a458198SEric Sandeen 22692a458198SEric Sandeen fs_info->flush_workers = 22702a458198SEric Sandeen btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0); 22712a458198SEric Sandeen 22722a458198SEric Sandeen fs_info->caching_workers = 22732a458198SEric Sandeen btrfs_alloc_workqueue("cache", flags, max_active, 0); 22742a458198SEric Sandeen 22752a458198SEric Sandeen /* 22762a458198SEric Sandeen * a higher idle thresh on the submit workers makes it much more 22772a458198SEric Sandeen * likely that bios will be send down in a sane order to the 22782a458198SEric Sandeen * devices 22792a458198SEric Sandeen */ 22802a458198SEric Sandeen fs_info->submit_workers = 22812a458198SEric Sandeen btrfs_alloc_workqueue("submit", flags, 22822a458198SEric Sandeen min_t(u64, fs_devices->num_devices, 22832a458198SEric Sandeen max_active), 64); 22842a458198SEric Sandeen 22852a458198SEric Sandeen fs_info->fixup_workers = 22862a458198SEric Sandeen btrfs_alloc_workqueue("fixup", flags, 1, 0); 22872a458198SEric Sandeen 22882a458198SEric Sandeen /* 22892a458198SEric Sandeen * endios are largely parallel and should have a very 22902a458198SEric Sandeen * low idle thresh 22912a458198SEric Sandeen */ 22922a458198SEric Sandeen fs_info->endio_workers = 22932a458198SEric Sandeen btrfs_alloc_workqueue("endio", flags, max_active, 4); 22942a458198SEric Sandeen fs_info->endio_meta_workers = 22952a458198SEric Sandeen btrfs_alloc_workqueue("endio-meta", flags, max_active, 4); 22962a458198SEric Sandeen fs_info->endio_meta_write_workers = 22972a458198SEric Sandeen btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2); 22982a458198SEric Sandeen fs_info->endio_raid56_workers = 22992a458198SEric Sandeen btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4); 23002a458198SEric Sandeen fs_info->endio_repair_workers = 23012a458198SEric Sandeen btrfs_alloc_workqueue("endio-repair", flags, 1, 0); 23022a458198SEric Sandeen fs_info->rmw_workers = 23032a458198SEric Sandeen btrfs_alloc_workqueue("rmw", flags, max_active, 2); 23042a458198SEric Sandeen fs_info->endio_write_workers = 23052a458198SEric Sandeen btrfs_alloc_workqueue("endio-write", flags, max_active, 2); 23062a458198SEric Sandeen fs_info->endio_freespace_worker = 23072a458198SEric Sandeen btrfs_alloc_workqueue("freespace-write", flags, max_active, 0); 23082a458198SEric Sandeen fs_info->delayed_workers = 23092a458198SEric Sandeen btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0); 23102a458198SEric Sandeen fs_info->readahead_workers = 23112a458198SEric Sandeen btrfs_alloc_workqueue("readahead", flags, max_active, 2); 23122a458198SEric Sandeen fs_info->qgroup_rescan_workers = 23132a458198SEric Sandeen btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0); 23142a458198SEric Sandeen fs_info->extent_workers = 23152a458198SEric Sandeen btrfs_alloc_workqueue("extent-refs", flags, 23162a458198SEric Sandeen min_t(u64, fs_devices->num_devices, 23172a458198SEric Sandeen max_active), 8); 23182a458198SEric Sandeen 23192a458198SEric Sandeen if (!(fs_info->workers && fs_info->delalloc_workers && 23202a458198SEric Sandeen fs_info->submit_workers && fs_info->flush_workers && 23212a458198SEric Sandeen fs_info->endio_workers && fs_info->endio_meta_workers && 23222a458198SEric Sandeen fs_info->endio_meta_write_workers && 23232a458198SEric Sandeen fs_info->endio_repair_workers && 23242a458198SEric Sandeen fs_info->endio_write_workers && fs_info->endio_raid56_workers && 23252a458198SEric Sandeen fs_info->endio_freespace_worker && fs_info->rmw_workers && 23262a458198SEric Sandeen fs_info->caching_workers && fs_info->readahead_workers && 23272a458198SEric Sandeen fs_info->fixup_workers && fs_info->delayed_workers && 23282a458198SEric Sandeen fs_info->extent_workers && 23292a458198SEric Sandeen fs_info->qgroup_rescan_workers)) { 23302a458198SEric Sandeen return -ENOMEM; 23312a458198SEric Sandeen } 23322a458198SEric Sandeen 23332a458198SEric Sandeen return 0; 23342a458198SEric Sandeen } 23352a458198SEric Sandeen 233663443bf5SEric Sandeen static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 233763443bf5SEric Sandeen struct btrfs_fs_devices *fs_devices) 233863443bf5SEric Sandeen { 233963443bf5SEric Sandeen int ret; 234063443bf5SEric Sandeen struct btrfs_root *tree_root = fs_info->tree_root; 234163443bf5SEric Sandeen struct btrfs_root *log_tree_root; 234263443bf5SEric Sandeen struct btrfs_super_block *disk_super = fs_info->super_copy; 234363443bf5SEric Sandeen u64 bytenr = btrfs_super_log_root(disk_super); 234463443bf5SEric Sandeen 234563443bf5SEric Sandeen if (fs_devices->rw_devices == 0) { 234663443bf5SEric Sandeen printk(KERN_WARNING "BTRFS: log replay required " 234763443bf5SEric Sandeen "on RO media\n"); 234863443bf5SEric Sandeen return -EIO; 234963443bf5SEric Sandeen } 235063443bf5SEric Sandeen 235163443bf5SEric Sandeen log_tree_root = btrfs_alloc_root(fs_info); 235263443bf5SEric Sandeen if (!log_tree_root) 235363443bf5SEric Sandeen return -ENOMEM; 235463443bf5SEric Sandeen 235563443bf5SEric Sandeen __setup_root(tree_root->nodesize, tree_root->sectorsize, 235663443bf5SEric Sandeen tree_root->stripesize, log_tree_root, fs_info, 235763443bf5SEric Sandeen BTRFS_TREE_LOG_OBJECTID); 235863443bf5SEric Sandeen 235963443bf5SEric Sandeen log_tree_root->node = read_tree_block(tree_root, bytenr, 236063443bf5SEric Sandeen fs_info->generation + 1); 236164c043deSLiu Bo if (IS_ERR(log_tree_root->node)) { 236264c043deSLiu Bo printk(KERN_ERR "BTRFS: failed to read log tree\n"); 23630eeff236SLiu Bo ret = PTR_ERR(log_tree_root->node); 236464c043deSLiu Bo kfree(log_tree_root); 23650eeff236SLiu Bo return ret; 236664c043deSLiu Bo } else if (!extent_buffer_uptodate(log_tree_root->node)) { 236763443bf5SEric Sandeen printk(KERN_ERR "BTRFS: failed to read log tree\n"); 236863443bf5SEric Sandeen free_extent_buffer(log_tree_root->node); 236963443bf5SEric Sandeen kfree(log_tree_root); 237063443bf5SEric Sandeen return -EIO; 237163443bf5SEric Sandeen } 237263443bf5SEric Sandeen /* returns with log_tree_root freed on success */ 237363443bf5SEric Sandeen ret = btrfs_recover_log_trees(log_tree_root); 237463443bf5SEric Sandeen if (ret) { 237563443bf5SEric Sandeen btrfs_error(tree_root->fs_info, ret, 237663443bf5SEric Sandeen "Failed to recover log tree"); 237763443bf5SEric Sandeen free_extent_buffer(log_tree_root->node); 237863443bf5SEric Sandeen kfree(log_tree_root); 237963443bf5SEric Sandeen return ret; 238063443bf5SEric Sandeen } 238163443bf5SEric Sandeen 238263443bf5SEric Sandeen if (fs_info->sb->s_flags & MS_RDONLY) { 238363443bf5SEric Sandeen ret = btrfs_commit_super(tree_root); 238463443bf5SEric Sandeen if (ret) 238563443bf5SEric Sandeen return ret; 238663443bf5SEric Sandeen } 238763443bf5SEric Sandeen 238863443bf5SEric Sandeen return 0; 238963443bf5SEric Sandeen } 239063443bf5SEric Sandeen 23914bbcaa64SEric Sandeen static int btrfs_read_roots(struct btrfs_fs_info *fs_info, 23924bbcaa64SEric Sandeen struct btrfs_root *tree_root) 23934bbcaa64SEric Sandeen { 2394a4f3d2c4SDavid Sterba struct btrfs_root *root; 23954bbcaa64SEric Sandeen struct btrfs_key location; 23964bbcaa64SEric Sandeen int ret; 23974bbcaa64SEric Sandeen 23984bbcaa64SEric Sandeen location.objectid = BTRFS_EXTENT_TREE_OBJECTID; 23994bbcaa64SEric Sandeen location.type = BTRFS_ROOT_ITEM_KEY; 24004bbcaa64SEric Sandeen location.offset = 0; 24014bbcaa64SEric Sandeen 2402a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2403a4f3d2c4SDavid Sterba if (IS_ERR(root)) 2404a4f3d2c4SDavid Sterba return PTR_ERR(root); 2405a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2406a4f3d2c4SDavid Sterba fs_info->extent_root = root; 24074bbcaa64SEric Sandeen 24084bbcaa64SEric Sandeen location.objectid = BTRFS_DEV_TREE_OBJECTID; 2409a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2410a4f3d2c4SDavid Sterba if (IS_ERR(root)) 2411a4f3d2c4SDavid Sterba return PTR_ERR(root); 2412a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2413a4f3d2c4SDavid Sterba fs_info->dev_root = root; 24144bbcaa64SEric Sandeen btrfs_init_devices_late(fs_info); 24154bbcaa64SEric Sandeen 24164bbcaa64SEric Sandeen location.objectid = BTRFS_CSUM_TREE_OBJECTID; 2417a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2418a4f3d2c4SDavid Sterba if (IS_ERR(root)) 2419a4f3d2c4SDavid Sterba return PTR_ERR(root); 2420a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2421a4f3d2c4SDavid Sterba fs_info->csum_root = root; 24224bbcaa64SEric Sandeen 24234bbcaa64SEric Sandeen location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2424a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2425a4f3d2c4SDavid Sterba if (!IS_ERR(root)) { 2426a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 24274bbcaa64SEric Sandeen fs_info->quota_enabled = 1; 24284bbcaa64SEric Sandeen fs_info->pending_quota_state = 1; 2429a4f3d2c4SDavid Sterba fs_info->quota_root = root; 24304bbcaa64SEric Sandeen } 24314bbcaa64SEric Sandeen 24324bbcaa64SEric Sandeen location.objectid = BTRFS_UUID_TREE_OBJECTID; 2433a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2434a4f3d2c4SDavid Sterba if (IS_ERR(root)) { 2435a4f3d2c4SDavid Sterba ret = PTR_ERR(root); 24364bbcaa64SEric Sandeen if (ret != -ENOENT) 24374bbcaa64SEric Sandeen return ret; 24384bbcaa64SEric Sandeen } else { 2439a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2440a4f3d2c4SDavid Sterba fs_info->uuid_root = root; 24414bbcaa64SEric Sandeen } 24424bbcaa64SEric Sandeen 24434bbcaa64SEric Sandeen return 0; 24444bbcaa64SEric Sandeen } 24454bbcaa64SEric Sandeen 2446ad2b2c80SAl Viro int open_ctree(struct super_block *sb, 2447dfe25020SChris Mason struct btrfs_fs_devices *fs_devices, 2448dfe25020SChris Mason char *options) 2449eb60ceacSChris Mason { 2450db94535dSChris Mason u32 sectorsize; 2451db94535dSChris Mason u32 nodesize; 245287ee04ebSChris Mason u32 stripesize; 245384234f3aSYan Zheng u64 generation; 2454f2b636e8SJosef Bacik u64 features; 24553de4586cSChris Mason struct btrfs_key location; 2456a061fc8dSChris Mason struct buffer_head *bh; 24574d34b278SIlya Dryomov struct btrfs_super_block *disk_super; 2458815745cfSAl Viro struct btrfs_fs_info *fs_info = btrfs_sb(sb); 2459f84a8bd6SAl Viro struct btrfs_root *tree_root; 24604d34b278SIlya Dryomov struct btrfs_root *chunk_root; 2461eb60ceacSChris Mason int ret; 2462e58ca020SYan int err = -EINVAL; 2463af31f5e5SChris Mason int num_backups_tried = 0; 2464af31f5e5SChris Mason int backup_index = 0; 24655cdc7ad3SQu Wenruo int max_active; 24664543df7eSChris Mason 2467f84a8bd6SAl Viro tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info); 24686f07e42eSAl Viro chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info); 2469cb517eabSMiao Xie if (!tree_root || !chunk_root) { 247039279cc3SChris Mason err = -ENOMEM; 247139279cc3SChris Mason goto fail; 247239279cc3SChris Mason } 247376dda93cSYan, Zheng 247476dda93cSYan, Zheng ret = init_srcu_struct(&fs_info->subvol_srcu); 247576dda93cSYan, Zheng if (ret) { 247676dda93cSYan, Zheng err = ret; 247776dda93cSYan, Zheng goto fail; 247876dda93cSYan, Zheng } 247976dda93cSYan, Zheng 248076dda93cSYan, Zheng ret = setup_bdi(fs_info, &fs_info->bdi); 248176dda93cSYan, Zheng if (ret) { 248276dda93cSYan, Zheng err = ret; 248376dda93cSYan, Zheng goto fail_srcu; 248476dda93cSYan, Zheng } 248576dda93cSYan, Zheng 2486908c7f19STejun Heo ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 2487e2d84521SMiao Xie if (ret) { 2488e2d84521SMiao Xie err = ret; 2489e2d84521SMiao Xie goto fail_bdi; 2490e2d84521SMiao Xie } 2491e2d84521SMiao Xie fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE * 2492e2d84521SMiao Xie (1 + ilog2(nr_cpu_ids)); 2493e2d84521SMiao Xie 2494908c7f19STejun Heo ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 2495963d678bSMiao Xie if (ret) { 2496963d678bSMiao Xie err = ret; 2497963d678bSMiao Xie goto fail_dirty_metadata_bytes; 2498963d678bSMiao Xie } 2499963d678bSMiao Xie 2500908c7f19STejun Heo ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL); 2501c404e0dcSMiao Xie if (ret) { 2502c404e0dcSMiao Xie err = ret; 2503c404e0dcSMiao Xie goto fail_delalloc_bytes; 2504c404e0dcSMiao Xie } 2505c404e0dcSMiao Xie 250676dda93cSYan, Zheng fs_info->btree_inode = new_inode(sb); 250776dda93cSYan, Zheng if (!fs_info->btree_inode) { 250876dda93cSYan, Zheng err = -ENOMEM; 2509c404e0dcSMiao Xie goto fail_bio_counter; 251076dda93cSYan, Zheng } 251176dda93cSYan, Zheng 2512a6591715SChris Mason mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 25131561dedaSMiao Xie 251476dda93cSYan, Zheng INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 2515f28491e0SJosef Bacik INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 25168fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 2517facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 251824bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 2519eb73c1b7SMiao Xie INIT_LIST_HEAD(&fs_info->delalloc_roots); 252011833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 2521eb73c1b7SMiao Xie spin_lock_init(&fs_info->delalloc_root_lock); 2522a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 252376dda93cSYan, Zheng spin_lock_init(&fs_info->fs_roots_radix_lock); 252424bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 25254cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 25262bf64758SJosef Bacik spin_lock_init(&fs_info->free_chunk_lock); 2527f29021b2SJan Schmidt spin_lock_init(&fs_info->tree_mod_seq_lock); 2528ceda0864SMiao Xie spin_lock_init(&fs_info->super_lock); 2529fcebe456SJosef Bacik spin_lock_init(&fs_info->qgroup_op_lock); 2530f28491e0SJosef Bacik spin_lock_init(&fs_info->buffer_lock); 253147ab2a6cSJosef Bacik spin_lock_init(&fs_info->unused_bgs_lock); 2532f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 2533d7c15171SZhao Lei mutex_init(&fs_info->unused_bg_unpin_mutex); 253467c5e7d4SFilipe Manana mutex_init(&fs_info->delete_unused_bgs_mutex); 25357585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 2536573bfb72SMiao Xie mutex_init(&fs_info->delalloc_root_mutex); 2537de98ced9SMiao Xie seqlock_init(&fs_info->profiles_lock); 2538d7c15171SZhao Lei init_rwsem(&fs_info->delayed_iput_sem); 253919c00ddcSChris Mason 25400b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 25416324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 2542f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 254347ab2a6cSJosef Bacik INIT_LIST_HEAD(&fs_info->unused_bgs); 25440b86a832SChris Mason btrfs_mapping_init(&fs_info->mapping_tree); 254566d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 254666d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 254766d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delalloc_block_rsv, 254866d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELALLOC); 254966d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 255066d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 255166d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 255266d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 255366d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 2554cb03c743SChris Mason atomic_set(&fs_info->nr_async_submits, 0); 2555771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 25568c8bee1dSChris Mason atomic_set(&fs_info->async_submit_draining, 0); 25570986fe9eSChris Mason atomic_set(&fs_info->nr_async_bios, 0); 25584cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 2559fcebe456SJosef Bacik atomic_set(&fs_info->qgroup_op_seq, 0); 2560fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 2561e20d96d6SChris Mason fs_info->sb = sb; 256295ac567aSFilipe David Borba Manana fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 25639ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 25644cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 25652bf64758SJosef Bacik fs_info->free_chunk_space = 0; 2566f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 25678b87dc17SDavid Sterba fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 2568f8c269d7SDavid Sterba fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 256990519d66SArne Jansen /* readahead state */ 257090519d66SArne Jansen INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT); 257190519d66SArne Jansen spin_lock_init(&fs_info->reada_lock); 2572c8b97818SChris Mason 2573b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 2574b34b086cSChris Mason num_online_cpus() + 2, 8); 25750afbaf8cSChris Mason 2576199c2a9cSMiao Xie INIT_LIST_HEAD(&fs_info->ordered_roots); 2577199c2a9cSMiao Xie spin_lock_init(&fs_info->ordered_root_lock); 257816cdcec7SMiao Xie fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 257916cdcec7SMiao Xie GFP_NOFS); 258016cdcec7SMiao Xie if (!fs_info->delayed_root) { 258116cdcec7SMiao Xie err = -ENOMEM; 258216cdcec7SMiao Xie goto fail_iput; 258316cdcec7SMiao Xie } 258416cdcec7SMiao Xie btrfs_init_delayed_root(fs_info->delayed_root); 25853eaa2885SChris Mason 2586638aa7edSEric Sandeen btrfs_init_scrub(fs_info); 258721adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 258821adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 258921adbd5cSStefan Behrens #endif 2590779a65a4SEric Sandeen btrfs_init_balance(fs_info); 259121c7e756SMiao Xie btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work); 2592a2de733cSArne Jansen 2593a061fc8dSChris Mason sb->s_blocksize = 4096; 2594a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(4096); 259532a88aa1SJens Axboe sb->s_bdi = &fs_info->bdi; 2596a061fc8dSChris Mason 2597f37938e0SEric Sandeen btrfs_init_btree_inode(fs_info, tree_root); 259839279cc3SChris Mason 2599e02119d5SChris Mason spin_lock_init(&fs_info->block_group_cache_lock); 26006bef4d31SEric Paris fs_info->block_group_cache_tree = RB_ROOT; 2601a1897fddSLiu Bo fs_info->first_logical_byte = (u64)-1; 2602925baeddSChris Mason 260311833d66SYan Zheng extent_io_tree_init(&fs_info->freed_extents[0], 2604f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 260511833d66SYan Zheng extent_io_tree_init(&fs_info->freed_extents[1], 2606f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 260711833d66SYan Zheng fs_info->pinned_extents = &fs_info->freed_extents[0]; 2608e6dcd2dcSChris Mason fs_info->do_barriers = 1; 2609f9295749SChris Mason 2610d98237b3SChris Mason 26115a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 261230ae8467SChris Mason mutex_init(&fs_info->tree_log_mutex); 2613925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 2614925baeddSChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 2615a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 2616a74a4b97SChris Mason mutex_init(&fs_info->volume_mutex); 26171bbc621eSChris Mason mutex_init(&fs_info->ro_block_group_mutex); 26189e351cc8SJosef Bacik init_rwsem(&fs_info->commit_root_sem); 2619c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 262076dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 2621803b2f54SStefan Behrens sema_init(&fs_info->uuid_tree_rescan_sem, 1); 2622fa9c0d79SChris Mason 2623ad618368SEric Sandeen btrfs_init_dev_replace_locks(fs_info); 2624f9e92e40SEric Sandeen btrfs_init_qgroup(fs_info); 2625416ac51dSArne Jansen 2626fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2627fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2628fa9c0d79SChris Mason 26297d9eb12cSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 26303768f368SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 2631bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 26324854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 26339a8dd150SChris Mason 263404216820SFilipe Manana INIT_LIST_HEAD(&fs_info->pinned_chunks); 263504216820SFilipe Manana 263653b381b3SDavid Woodhouse ret = btrfs_alloc_stripe_hash_table(fs_info); 263753b381b3SDavid Woodhouse if (ret) { 263883c8266aSDavid Sterba err = ret; 263953b381b3SDavid Woodhouse goto fail_alloc; 264053b381b3SDavid Woodhouse } 264153b381b3SDavid Woodhouse 2642707e8a07SDavid Sterba __setup_root(4096, 4096, 4096, tree_root, 26432c90e5d6SChris Mason fs_info, BTRFS_ROOT_TREE_OBJECTID); 26447eccb903SChris Mason 26453c4bb26bSChris Mason invalidate_bdev(fs_devices->latest_bdev); 26461104a885SDavid Sterba 26471104a885SDavid Sterba /* 26481104a885SDavid Sterba * Read super block and check the signature bytes only 26491104a885SDavid Sterba */ 2650a512bbf8SYan Zheng bh = btrfs_read_dev_super(fs_devices->latest_bdev); 265120b45077SDave Young if (!bh) { 265220b45077SDave Young err = -EINVAL; 265316cdcec7SMiao Xie goto fail_alloc; 265420b45077SDave Young } 265539279cc3SChris Mason 26561104a885SDavid Sterba /* 26571104a885SDavid Sterba * We want to check superblock checksum, the type is stored inside. 26581104a885SDavid Sterba * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 26591104a885SDavid Sterba */ 26601104a885SDavid Sterba if (btrfs_check_super_csum(bh->b_data)) { 2661efe120a0SFrank Holton printk(KERN_ERR "BTRFS: superblock checksum mismatch\n"); 26621104a885SDavid Sterba err = -EINVAL; 26631104a885SDavid Sterba goto fail_alloc; 26641104a885SDavid Sterba } 26651104a885SDavid Sterba 26661104a885SDavid Sterba /* 26671104a885SDavid Sterba * super_copy is zeroed at allocation time and we never touch the 26681104a885SDavid Sterba * following bytes up to INFO_SIZE, the checksum is calculated from 26691104a885SDavid Sterba * the whole block of INFO_SIZE 26701104a885SDavid Sterba */ 26716c41761fSDavid Sterba memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy)); 26726c41761fSDavid Sterba memcpy(fs_info->super_for_commit, fs_info->super_copy, 26736c41761fSDavid Sterba sizeof(*fs_info->super_for_commit)); 2674a061fc8dSChris Mason brelse(bh); 26755f39d397SChris Mason 26766c41761fSDavid Sterba memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE); 26770b86a832SChris Mason 26781104a885SDavid Sterba ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY); 26791104a885SDavid Sterba if (ret) { 2680efe120a0SFrank Holton printk(KERN_ERR "BTRFS: superblock contains fatal errors\n"); 26811104a885SDavid Sterba err = -EINVAL; 26821104a885SDavid Sterba goto fail_alloc; 26831104a885SDavid Sterba } 26841104a885SDavid Sterba 26856c41761fSDavid Sterba disk_super = fs_info->super_copy; 26860f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 268716cdcec7SMiao Xie goto fail_alloc; 26880f7d52f4SChris Mason 2689acce952bSliubo /* check FS state, whether FS is broken. */ 269087533c47SMiao Xie if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 269187533c47SMiao Xie set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 2692acce952bSliubo 269375e7cb7fSLiu Bo /* 2694af31f5e5SChris Mason * run through our array of backup supers and setup 2695af31f5e5SChris Mason * our ring pointer to the oldest one 2696af31f5e5SChris Mason */ 2697af31f5e5SChris Mason generation = btrfs_super_generation(disk_super); 2698af31f5e5SChris Mason find_oldest_super_backup(fs_info, generation); 2699af31f5e5SChris Mason 2700af31f5e5SChris Mason /* 270175e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 270275e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 270375e7cb7fSLiu Bo */ 270475e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 270575e7cb7fSLiu Bo 27062b82032cSYan Zheng ret = btrfs_parse_options(tree_root, options); 27072b82032cSYan Zheng if (ret) { 27082b82032cSYan Zheng err = ret; 270916cdcec7SMiao Xie goto fail_alloc; 27102b82032cSYan Zheng } 2711dfe25020SChris Mason 2712f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 2713f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 2714f2b636e8SJosef Bacik if (features) { 2715f2b636e8SJosef Bacik printk(KERN_ERR "BTRFS: couldn't mount because of " 2716f2b636e8SJosef Bacik "unsupported optional features (%Lx).\n", 2717c1c9ff7cSGeert Uytterhoeven features); 2718f2b636e8SJosef Bacik err = -EINVAL; 271916cdcec7SMiao Xie goto fail_alloc; 2720f2b636e8SJosef Bacik } 2721f2b636e8SJosef Bacik 2722707e8a07SDavid Sterba /* 2723707e8a07SDavid Sterba * Leafsize and nodesize were always equal, this is only a sanity check. 2724707e8a07SDavid Sterba */ 2725707e8a07SDavid Sterba if (le32_to_cpu(disk_super->__unused_leafsize) != 2726727011e0SChris Mason btrfs_super_nodesize(disk_super)) { 2727727011e0SChris Mason printk(KERN_ERR "BTRFS: couldn't mount because metadata " 2728727011e0SChris Mason "blocksizes don't match. node %d leaf %d\n", 2729727011e0SChris Mason btrfs_super_nodesize(disk_super), 2730707e8a07SDavid Sterba le32_to_cpu(disk_super->__unused_leafsize)); 2731727011e0SChris Mason err = -EINVAL; 2732727011e0SChris Mason goto fail_alloc; 2733727011e0SChris Mason } 2734707e8a07SDavid Sterba if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) { 2735727011e0SChris Mason printk(KERN_ERR "BTRFS: couldn't mount because metadata " 2736727011e0SChris Mason "blocksize (%d) was too large\n", 2737707e8a07SDavid Sterba btrfs_super_nodesize(disk_super)); 2738727011e0SChris Mason err = -EINVAL; 2739727011e0SChris Mason goto fail_alloc; 2740727011e0SChris Mason } 2741727011e0SChris Mason 27425d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 27435d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 274469e380d1SLi Zefan if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO) 2745a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 2746727011e0SChris Mason 27473173a18fSJosef Bacik if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) 27485efa0490SDavid Sterba printk(KERN_INFO "BTRFS: has skinny extents\n"); 27493173a18fSJosef Bacik 2750727011e0SChris Mason /* 2751727011e0SChris Mason * flag our filesystem as having big metadata blocks if 2752727011e0SChris Mason * they are bigger than the page size 2753727011e0SChris Mason */ 2754707e8a07SDavid Sterba if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) { 2755727011e0SChris Mason if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 2756efe120a0SFrank Holton printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n"); 2757727011e0SChris Mason features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 2758727011e0SChris Mason } 2759727011e0SChris Mason 2760bc3f116fSChris Mason nodesize = btrfs_super_nodesize(disk_super); 2761bc3f116fSChris Mason sectorsize = btrfs_super_sectorsize(disk_super); 2762bc3f116fSChris Mason stripesize = btrfs_super_stripesize(disk_super); 2763707e8a07SDavid Sterba fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 2764963d678bSMiao Xie fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 2765bc3f116fSChris Mason 2766bc3f116fSChris Mason /* 2767bc3f116fSChris Mason * mixed block groups end up with duplicate but slightly offset 2768bc3f116fSChris Mason * extent buffers for the same range. It leads to corruptions 2769bc3f116fSChris Mason */ 2770bc3f116fSChris Mason if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 2771707e8a07SDavid Sterba (sectorsize != nodesize)) { 2772aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: unequal leaf/node/sector sizes " 2773bc3f116fSChris Mason "are not allowed for mixed block groups on %s\n", 2774bc3f116fSChris Mason sb->s_id); 2775bc3f116fSChris Mason goto fail_alloc; 2776bc3f116fSChris Mason } 2777bc3f116fSChris Mason 2778ceda0864SMiao Xie /* 2779ceda0864SMiao Xie * Needn't use the lock because there is no other task which will 2780ceda0864SMiao Xie * update the flag. 2781ceda0864SMiao Xie */ 27825d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 27835d4f98a2SYan Zheng 2784f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 2785f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 2786f2b636e8SJosef Bacik if (!(sb->s_flags & MS_RDONLY) && features) { 2787f2b636e8SJosef Bacik printk(KERN_ERR "BTRFS: couldn't mount RDWR because of " 2788f2b636e8SJosef Bacik "unsupported option features (%Lx).\n", 2789c1c9ff7cSGeert Uytterhoeven features); 2790f2b636e8SJosef Bacik err = -EINVAL; 279116cdcec7SMiao Xie goto fail_alloc; 2792f2b636e8SJosef Bacik } 279361d92c32SChris Mason 27945cdc7ad3SQu Wenruo max_active = fs_info->thread_pool_size; 279561d92c32SChris Mason 27962a458198SEric Sandeen ret = btrfs_init_workqueues(fs_info, fs_devices); 27972a458198SEric Sandeen if (ret) { 27982a458198SEric Sandeen err = ret; 27990dc3b84aSJosef Bacik goto fail_sb_buffer; 28000dc3b84aSJosef Bacik } 28014543df7eSChris Mason 28024575c9ccSChris Mason fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super); 2803c8b97818SChris Mason fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages, 2804c8b97818SChris Mason 4 * 1024 * 1024 / PAGE_CACHE_SIZE); 28054575c9ccSChris Mason 2806db94535dSChris Mason tree_root->nodesize = nodesize; 2807db94535dSChris Mason tree_root->sectorsize = sectorsize; 280887ee04ebSChris Mason tree_root->stripesize = stripesize; 2809a061fc8dSChris Mason 2810a061fc8dSChris Mason sb->s_blocksize = sectorsize; 2811a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 2812db94535dSChris Mason 28133cae210fSQu Wenruo if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) { 2814aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: valid FS not found on %s\n", sb->s_id); 281539279cc3SChris Mason goto fail_sb_buffer; 281639279cc3SChris Mason } 281719c00ddcSChris Mason 28188d082fb7SLiu Bo if (sectorsize != PAGE_SIZE) { 2819aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: incompatible sector size (%lu) " 28208d082fb7SLiu Bo "found on %s\n", (unsigned long)sectorsize, sb->s_id); 2821941b2ddfSKeith Mannthey goto fail_sb_buffer; 2822941b2ddfSKeith Mannthey } 2823941b2ddfSKeith Mannthey 2824925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 2825e4404d6eSYan Zheng ret = btrfs_read_sys_array(tree_root); 2826925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 282784eed90fSChris Mason if (ret) { 2828aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: failed to read the system " 2829d397712bSChris Mason "array on %s\n", sb->s_id); 28305d4f98a2SYan Zheng goto fail_sb_buffer; 283184eed90fSChris Mason } 28320b86a832SChris Mason 283384234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 28340b86a832SChris Mason 2835707e8a07SDavid Sterba __setup_root(nodesize, sectorsize, stripesize, chunk_root, 2836707e8a07SDavid Sterba fs_info, BTRFS_CHUNK_TREE_OBJECTID); 28370b86a832SChris Mason 28380b86a832SChris Mason chunk_root->node = read_tree_block(chunk_root, 28390b86a832SChris Mason btrfs_super_chunk_root(disk_super), 2840ce86cd59SDavid Sterba generation); 284164c043deSLiu Bo if (IS_ERR(chunk_root->node) || 284264c043deSLiu Bo !extent_buffer_uptodate(chunk_root->node)) { 2843aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: failed to read chunk root on %s\n", 284483121942SDavid Woodhouse sb->s_id); 284595ab1f64SZhao Lei chunk_root->node = NULL; 2846af31f5e5SChris Mason goto fail_tree_roots; 284783121942SDavid Woodhouse } 28485d4f98a2SYan Zheng btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 28495d4f98a2SYan Zheng chunk_root->commit_root = btrfs_root_node(chunk_root); 28500b86a832SChris Mason 2851e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 2852b308bc2fSGeert Uytterhoeven btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE); 2853e17cade2SChris Mason 28540b86a832SChris Mason ret = btrfs_read_chunk_tree(chunk_root); 28552b82032cSYan Zheng if (ret) { 2856aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: failed to read chunk tree on %s\n", 2857d397712bSChris Mason sb->s_id); 2858af31f5e5SChris Mason goto fail_tree_roots; 28592b82032cSYan Zheng } 28600b86a832SChris Mason 28618dabb742SStefan Behrens /* 28628dabb742SStefan Behrens * keep the device that is marked to be the target device for the 28638dabb742SStefan Behrens * dev_replace procedure 28648dabb742SStefan Behrens */ 28659eaed21eSEric Sandeen btrfs_close_extra_devices(fs_devices, 0); 2866dfe25020SChris Mason 2867a6b0d5c8SChris Mason if (!fs_devices->latest_bdev) { 2868aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: failed to read devices on %s\n", 2869a6b0d5c8SChris Mason sb->s_id); 2870a6b0d5c8SChris Mason goto fail_tree_roots; 2871a6b0d5c8SChris Mason } 2872a6b0d5c8SChris Mason 2873af31f5e5SChris Mason retry_root_backup: 287484234f3aSYan Zheng generation = btrfs_super_generation(disk_super); 28750b86a832SChris Mason 2876e20d96d6SChris Mason tree_root->node = read_tree_block(tree_root, 2877db94535dSChris Mason btrfs_super_root(disk_super), 2878ce86cd59SDavid Sterba generation); 287964c043deSLiu Bo if (IS_ERR(tree_root->node) || 288064c043deSLiu Bo !extent_buffer_uptodate(tree_root->node)) { 2881efe120a0SFrank Holton printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n", 288283121942SDavid Woodhouse sb->s_id); 288395ab1f64SZhao Lei tree_root->node = NULL; 2884af31f5e5SChris Mason goto recovery_tree_root; 288583121942SDavid Woodhouse } 2886af31f5e5SChris Mason 28875d4f98a2SYan Zheng btrfs_set_root_node(&tree_root->root_item, tree_root->node); 28885d4f98a2SYan Zheng tree_root->commit_root = btrfs_root_node(tree_root); 288969e9c6c6SStefan Behrens btrfs_set_root_refs(&tree_root->root_item, 1); 2890db94535dSChris Mason 28914bbcaa64SEric Sandeen ret = btrfs_read_roots(fs_info, tree_root); 28924bbcaa64SEric Sandeen if (ret) 2893af31f5e5SChris Mason goto recovery_tree_root; 2894f7a81ea4SStefan Behrens 28958929ecfaSYan, Zheng fs_info->generation = generation; 28968929ecfaSYan, Zheng fs_info->last_trans_committed = generation; 28978929ecfaSYan, Zheng 289868310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 289968310a5eSIlya Dryomov if (ret) { 2900aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: failed to recover balance\n"); 290168310a5eSIlya Dryomov goto fail_block_groups; 290268310a5eSIlya Dryomov } 290368310a5eSIlya Dryomov 2904733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 2905733f4fbbSStefan Behrens if (ret) { 2906efe120a0SFrank Holton printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n", 2907733f4fbbSStefan Behrens ret); 2908733f4fbbSStefan Behrens goto fail_block_groups; 2909733f4fbbSStefan Behrens } 2910733f4fbbSStefan Behrens 29118dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 29128dabb742SStefan Behrens if (ret) { 2913efe120a0SFrank Holton pr_err("BTRFS: failed to init dev_replace: %d\n", ret); 29148dabb742SStefan Behrens goto fail_block_groups; 29158dabb742SStefan Behrens } 29168dabb742SStefan Behrens 29179eaed21eSEric Sandeen btrfs_close_extra_devices(fs_devices, 1); 29188dabb742SStefan Behrens 2919b7c35e81SAnand Jain ret = btrfs_sysfs_add_fsid(fs_devices, NULL); 2920b7c35e81SAnand Jain if (ret) { 2921b7c35e81SAnand Jain pr_err("BTRFS: failed to init sysfs fsid interface: %d\n", ret); 2922b7c35e81SAnand Jain goto fail_block_groups; 2923b7c35e81SAnand Jain } 2924b7c35e81SAnand Jain 2925b7c35e81SAnand Jain ret = btrfs_sysfs_add_device(fs_devices); 2926b7c35e81SAnand Jain if (ret) { 2927b7c35e81SAnand Jain pr_err("BTRFS: failed to init sysfs device interface: %d\n", ret); 2928b7c35e81SAnand Jain goto fail_fsdev_sysfs; 2929b7c35e81SAnand Jain } 2930b7c35e81SAnand Jain 29315ac1d209SJeff Mahoney ret = btrfs_sysfs_add_one(fs_info); 29325ac1d209SJeff Mahoney if (ret) { 2933efe120a0SFrank Holton pr_err("BTRFS: failed to init sysfs interface: %d\n", ret); 2934b7c35e81SAnand Jain goto fail_fsdev_sysfs; 29355ac1d209SJeff Mahoney } 29365ac1d209SJeff Mahoney 2937c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 2938c59021f8Sliubo if (ret) { 2939efe120a0SFrank Holton printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret); 29402365dd3cSAnand Jain goto fail_sysfs; 2941c59021f8Sliubo } 2942c59021f8Sliubo 29434bbcaa64SEric Sandeen ret = btrfs_read_block_groups(fs_info->extent_root); 29441b1d1f66SJosef Bacik if (ret) { 2945efe120a0SFrank Holton printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret); 29462365dd3cSAnand Jain goto fail_sysfs; 29471b1d1f66SJosef Bacik } 29485af3e8ccSStefan Behrens fs_info->num_tolerated_disk_barrier_failures = 29495af3e8ccSStefan Behrens btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); 2950292fd7fcSStefan Behrens if (fs_info->fs_devices->missing_devices > 2951292fd7fcSStefan Behrens fs_info->num_tolerated_disk_barrier_failures && 2952292fd7fcSStefan Behrens !(sb->s_flags & MS_RDONLY)) { 295378fa1770SZhao Lei pr_warn("BTRFS: missing devices(%llu) exceeds the limit(%d), writeable mount is not allowed\n", 295478fa1770SZhao Lei fs_info->fs_devices->missing_devices, 295578fa1770SZhao Lei fs_info->num_tolerated_disk_barrier_failures); 29562365dd3cSAnand Jain goto fail_sysfs; 2957292fd7fcSStefan Behrens } 29589078a3e1SChris Mason 2959a74a4b97SChris Mason fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 2960a74a4b97SChris Mason "btrfs-cleaner"); 296157506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 29622365dd3cSAnand Jain goto fail_sysfs; 2963a74a4b97SChris Mason 2964a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 2965a74a4b97SChris Mason tree_root, 2966a74a4b97SChris Mason "btrfs-transaction"); 296757506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 29683f157a2fSChris Mason goto fail_cleaner; 2969a74a4b97SChris Mason 2970c289811cSChris Mason if (!btrfs_test_opt(tree_root, SSD) && 2971c289811cSChris Mason !btrfs_test_opt(tree_root, NOSSD) && 2972c289811cSChris Mason !fs_info->fs_devices->rotating) { 2973efe120a0SFrank Holton printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD " 2974c289811cSChris Mason "mode\n"); 2975c289811cSChris Mason btrfs_set_opt(fs_info->mount_opt, SSD); 2976c289811cSChris Mason } 2977c289811cSChris Mason 2978572d9ab7SDavid Sterba /* 2979572d9ab7SDavid Sterba * Mount does not set all options immediatelly, we can do it now and do 2980572d9ab7SDavid Sterba * not have to wait for transaction commit 2981572d9ab7SDavid Sterba */ 2982572d9ab7SDavid Sterba btrfs_apply_pending_changes(fs_info); 29833818aea2SQu Wenruo 298421adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 298521adbd5cSStefan Behrens if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) { 298621adbd5cSStefan Behrens ret = btrfsic_mount(tree_root, fs_devices, 298721adbd5cSStefan Behrens btrfs_test_opt(tree_root, 298821adbd5cSStefan Behrens CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? 298921adbd5cSStefan Behrens 1 : 0, 299021adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 299121adbd5cSStefan Behrens if (ret) 2992efe120a0SFrank Holton printk(KERN_WARNING "BTRFS: failed to initialize" 299321adbd5cSStefan Behrens " integrity check module %s\n", sb->s_id); 299421adbd5cSStefan Behrens } 299521adbd5cSStefan Behrens #endif 2996bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 2997bcef60f2SArne Jansen if (ret) 2998bcef60f2SArne Jansen goto fail_trans_kthread; 299921adbd5cSStefan Behrens 3000acce952bSliubo /* do not make disk changes in broken FS */ 300168ce9682SStefan Behrens if (btrfs_super_log_root(disk_super) != 0) { 300263443bf5SEric Sandeen ret = btrfs_replay_log(fs_info, fs_devices); 300379787eaaSJeff Mahoney if (ret) { 300463443bf5SEric Sandeen err = ret; 300528c16cbbSWang Shilong goto fail_qgroup; 3006e556ce2cSYan Zheng } 3007e02119d5SChris Mason } 30081a40e23bSZheng Yan 300976dda93cSYan, Zheng ret = btrfs_find_orphan_roots(tree_root); 301079787eaaSJeff Mahoney if (ret) 301128c16cbbSWang Shilong goto fail_qgroup; 301276dda93cSYan, Zheng 30137c2ca468SChris Mason if (!(sb->s_flags & MS_RDONLY)) { 3014d68fc57bSYan, Zheng ret = btrfs_cleanup_fs_roots(fs_info); 301544c44af2SIlya Dryomov if (ret) 301628c16cbbSWang Shilong goto fail_qgroup; 3017d68fc57bSYan, Zheng 30185f316481SWang Shilong mutex_lock(&fs_info->cleaner_mutex); 30195d4f98a2SYan Zheng ret = btrfs_recover_relocation(tree_root); 30205f316481SWang Shilong mutex_unlock(&fs_info->cleaner_mutex); 3021d7ce5843SMiao Xie if (ret < 0) { 3022d7ce5843SMiao Xie printk(KERN_WARNING 3023efe120a0SFrank Holton "BTRFS: failed to recover relocation\n"); 3024d7ce5843SMiao Xie err = -EINVAL; 3025bcef60f2SArne Jansen goto fail_qgroup; 3026d7ce5843SMiao Xie } 30277c2ca468SChris Mason } 30281a40e23bSZheng Yan 30293de4586cSChris Mason location.objectid = BTRFS_FS_TREE_OBJECTID; 30303de4586cSChris Mason location.type = BTRFS_ROOT_ITEM_KEY; 3031cb517eabSMiao Xie location.offset = 0; 30323de4586cSChris Mason 30333de4586cSChris Mason fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location); 30343140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 30353140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 3036bcef60f2SArne Jansen goto fail_qgroup; 30373140c9a3SDan Carpenter } 3038c289811cSChris Mason 30392b6ba629SIlya Dryomov if (sb->s_flags & MS_RDONLY) 30402b6ba629SIlya Dryomov return 0; 30412b6ba629SIlya Dryomov 3042e3acc2a6SJosef Bacik down_read(&fs_info->cleanup_work_sem); 30432b6ba629SIlya Dryomov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 30442b6ba629SIlya Dryomov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 30452b6ba629SIlya Dryomov up_read(&fs_info->cleanup_work_sem); 30462b6ba629SIlya Dryomov close_ctree(tree_root); 30472b6ba629SIlya Dryomov return ret; 30482b6ba629SIlya Dryomov } 3049e3acc2a6SJosef Bacik up_read(&fs_info->cleanup_work_sem); 305059641015SIlya Dryomov 30512b6ba629SIlya Dryomov ret = btrfs_resume_balance_async(fs_info); 30522b6ba629SIlya Dryomov if (ret) { 3053efe120a0SFrank Holton printk(KERN_WARNING "BTRFS: failed to resume balance\n"); 305466b4ffd1SJosef Bacik close_ctree(tree_root); 30552b6ba629SIlya Dryomov return ret; 3056e3acc2a6SJosef Bacik } 3057e3acc2a6SJosef Bacik 30588dabb742SStefan Behrens ret = btrfs_resume_dev_replace_async(fs_info); 30598dabb742SStefan Behrens if (ret) { 3060efe120a0SFrank Holton pr_warn("BTRFS: failed to resume dev_replace\n"); 30618dabb742SStefan Behrens close_ctree(tree_root); 30628dabb742SStefan Behrens return ret; 30638dabb742SStefan Behrens } 30648dabb742SStefan Behrens 3065b382a324SJan Schmidt btrfs_qgroup_rescan_resume(fs_info); 3066b382a324SJan Schmidt 30674bbcaa64SEric Sandeen if (!fs_info->uuid_root) { 3068efe120a0SFrank Holton pr_info("BTRFS: creating UUID tree\n"); 3069f7a81ea4SStefan Behrens ret = btrfs_create_uuid_tree(fs_info); 3070f7a81ea4SStefan Behrens if (ret) { 3071efe120a0SFrank Holton pr_warn("BTRFS: failed to create the UUID tree %d\n", 3072f7a81ea4SStefan Behrens ret); 3073f7a81ea4SStefan Behrens close_ctree(tree_root); 3074f7a81ea4SStefan Behrens return ret; 3075f7a81ea4SStefan Behrens } 30764bbcaa64SEric Sandeen } else if (btrfs_test_opt(tree_root, RESCAN_UUID_TREE) || 30774bbcaa64SEric Sandeen fs_info->generation != 30784bbcaa64SEric Sandeen btrfs_super_uuid_tree_generation(disk_super)) { 3079efe120a0SFrank Holton pr_info("BTRFS: checking UUID tree\n"); 308070f80175SStefan Behrens ret = btrfs_check_uuid_tree(fs_info); 308170f80175SStefan Behrens if (ret) { 3082efe120a0SFrank Holton pr_warn("BTRFS: failed to check the UUID tree %d\n", 308370f80175SStefan Behrens ret); 308470f80175SStefan Behrens close_ctree(tree_root); 308570f80175SStefan Behrens return ret; 308670f80175SStefan Behrens } 308770f80175SStefan Behrens } else { 308870f80175SStefan Behrens fs_info->update_uuid_tree_gen = 1; 3089f7a81ea4SStefan Behrens } 3090f7a81ea4SStefan Behrens 309147ab2a6cSJosef Bacik fs_info->open = 1; 309247ab2a6cSJosef Bacik 3093ad2b2c80SAl Viro return 0; 309439279cc3SChris Mason 3095bcef60f2SArne Jansen fail_qgroup: 3096bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 30977c2ca468SChris Mason fail_trans_kthread: 30987c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 309954067ae9SJosef Bacik btrfs_cleanup_transaction(fs_info->tree_root); 3100faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 31013f157a2fSChris Mason fail_cleaner: 3102a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 31037c2ca468SChris Mason 31047c2ca468SChris Mason /* 31057c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 31067c2ca468SChris Mason * kthreads 31077c2ca468SChris Mason */ 31087c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 31097c2ca468SChris Mason 31102365dd3cSAnand Jain fail_sysfs: 31112365dd3cSAnand Jain btrfs_sysfs_remove_one(fs_info); 31122365dd3cSAnand Jain 3113b7c35e81SAnand Jain fail_fsdev_sysfs: 3114b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3115b7c35e81SAnand Jain 31161b1d1f66SJosef Bacik fail_block_groups: 311754067ae9SJosef Bacik btrfs_put_block_group_cache(fs_info); 31181b1d1f66SJosef Bacik btrfs_free_block_groups(fs_info); 3119af31f5e5SChris Mason 3120af31f5e5SChris Mason fail_tree_roots: 3121af31f5e5SChris Mason free_root_pointers(fs_info, 1); 31222b8195bbSMiao Xie invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3123af31f5e5SChris Mason 312439279cc3SChris Mason fail_sb_buffer: 31257abadb64SLiu Bo btrfs_stop_all_workers(fs_info); 312616cdcec7SMiao Xie fail_alloc: 31274543df7eSChris Mason fail_iput: 3128586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 3129586e46e2SIlya Dryomov 31304543df7eSChris Mason iput(fs_info->btree_inode); 3131c404e0dcSMiao Xie fail_bio_counter: 3132c404e0dcSMiao Xie percpu_counter_destroy(&fs_info->bio_counter); 3133963d678bSMiao Xie fail_delalloc_bytes: 3134963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 3135e2d84521SMiao Xie fail_dirty_metadata_bytes: 3136e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 3137ad081f14SJens Axboe fail_bdi: 31387e662854SQinghuang Feng bdi_destroy(&fs_info->bdi); 313976dda93cSYan, Zheng fail_srcu: 314076dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 31417e662854SQinghuang Feng fail: 314253b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 3143586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 3144ad2b2c80SAl Viro return err; 3145af31f5e5SChris Mason 3146af31f5e5SChris Mason recovery_tree_root: 3147af31f5e5SChris Mason if (!btrfs_test_opt(tree_root, RECOVERY)) 3148af31f5e5SChris Mason goto fail_tree_roots; 3149af31f5e5SChris Mason 3150af31f5e5SChris Mason free_root_pointers(fs_info, 0); 3151af31f5e5SChris Mason 3152af31f5e5SChris Mason /* don't use the log in recovery mode, it won't be valid */ 3153af31f5e5SChris Mason btrfs_set_super_log_root(disk_super, 0); 3154af31f5e5SChris Mason 3155af31f5e5SChris Mason /* we can't trust the free space cache either */ 3156af31f5e5SChris Mason btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 3157af31f5e5SChris Mason 3158af31f5e5SChris Mason ret = next_root_backup(fs_info, fs_info->super_copy, 3159af31f5e5SChris Mason &num_backups_tried, &backup_index); 3160af31f5e5SChris Mason if (ret == -1) 3161af31f5e5SChris Mason goto fail_block_groups; 3162af31f5e5SChris Mason goto retry_root_backup; 3163eb60ceacSChris Mason } 3164eb60ceacSChris Mason 3165f2984462SChris Mason static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate) 3166f2984462SChris Mason { 3167f2984462SChris Mason if (uptodate) { 3168f2984462SChris Mason set_buffer_uptodate(bh); 3169f2984462SChris Mason } else { 3170442a4f63SStefan Behrens struct btrfs_device *device = (struct btrfs_device *) 3171442a4f63SStefan Behrens bh->b_private; 3172442a4f63SStefan Behrens 3173efe120a0SFrank Holton printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to " 3174606686eeSJosef Bacik "I/O error on %s\n", 3175606686eeSJosef Bacik rcu_str_deref(device->name)); 31761259ab75SChris Mason /* note, we dont' set_buffer_write_io_error because we have 31771259ab75SChris Mason * our own ways of dealing with the IO errors 31781259ab75SChris Mason */ 3179f2984462SChris Mason clear_buffer_uptodate(bh); 3180442a4f63SStefan Behrens btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS); 3181f2984462SChris Mason } 3182f2984462SChris Mason unlock_buffer(bh); 3183f2984462SChris Mason put_bh(bh); 3184f2984462SChris Mason } 3185f2984462SChris Mason 3186a512bbf8SYan Zheng struct buffer_head *btrfs_read_dev_super(struct block_device *bdev) 3187a512bbf8SYan Zheng { 3188a512bbf8SYan Zheng struct buffer_head *bh; 3189a512bbf8SYan Zheng struct buffer_head *latest = NULL; 3190a512bbf8SYan Zheng struct btrfs_super_block *super; 3191a512bbf8SYan Zheng int i; 3192a512bbf8SYan Zheng u64 transid = 0; 3193a512bbf8SYan Zheng u64 bytenr; 3194a512bbf8SYan Zheng 3195a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 3196a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 3197a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 3198a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3199a512bbf8SYan Zheng */ 3200a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 3201a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 32028068a47eSAnand Jain if (bytenr + BTRFS_SUPER_INFO_SIZE >= 32038068a47eSAnand Jain i_size_read(bdev->bd_inode)) 3204a512bbf8SYan Zheng break; 32058068a47eSAnand Jain bh = __bread(bdev, bytenr / 4096, 32068068a47eSAnand Jain BTRFS_SUPER_INFO_SIZE); 3207a512bbf8SYan Zheng if (!bh) 3208a512bbf8SYan Zheng continue; 3209a512bbf8SYan Zheng 3210a512bbf8SYan Zheng super = (struct btrfs_super_block *)bh->b_data; 3211a512bbf8SYan Zheng if (btrfs_super_bytenr(super) != bytenr || 32123cae210fSQu Wenruo btrfs_super_magic(super) != BTRFS_MAGIC) { 3213a512bbf8SYan Zheng brelse(bh); 3214a512bbf8SYan Zheng continue; 3215a512bbf8SYan Zheng } 3216a512bbf8SYan Zheng 3217a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 3218a512bbf8SYan Zheng brelse(latest); 3219a512bbf8SYan Zheng latest = bh; 3220a512bbf8SYan Zheng transid = btrfs_super_generation(super); 3221a512bbf8SYan Zheng } else { 3222a512bbf8SYan Zheng brelse(bh); 3223a512bbf8SYan Zheng } 3224a512bbf8SYan Zheng } 3225a512bbf8SYan Zheng return latest; 3226a512bbf8SYan Zheng } 3227a512bbf8SYan Zheng 32284eedeb75SHisashi Hifumi /* 32294eedeb75SHisashi Hifumi * this should be called twice, once with wait == 0 and 32304eedeb75SHisashi Hifumi * once with wait == 1. When wait == 0 is done, all the buffer heads 32314eedeb75SHisashi Hifumi * we write are pinned. 32324eedeb75SHisashi Hifumi * 32334eedeb75SHisashi Hifumi * They are released when wait == 1 is done. 32344eedeb75SHisashi Hifumi * max_mirrors must be the same for both runs, and it indicates how 32354eedeb75SHisashi Hifumi * many supers on this one device should be written. 32364eedeb75SHisashi Hifumi * 32374eedeb75SHisashi Hifumi * max_mirrors == 0 means to write them all. 32384eedeb75SHisashi Hifumi */ 3239a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 3240a512bbf8SYan Zheng struct btrfs_super_block *sb, 3241a512bbf8SYan Zheng int do_barriers, int wait, int max_mirrors) 3242a512bbf8SYan Zheng { 3243a512bbf8SYan Zheng struct buffer_head *bh; 3244a512bbf8SYan Zheng int i; 3245a512bbf8SYan Zheng int ret; 3246a512bbf8SYan Zheng int errors = 0; 3247a512bbf8SYan Zheng u32 crc; 3248a512bbf8SYan Zheng u64 bytenr; 3249a512bbf8SYan Zheng 3250a512bbf8SYan Zheng if (max_mirrors == 0) 3251a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3252a512bbf8SYan Zheng 3253a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 3254a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 3255935e5cc9SMiao Xie if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3256935e5cc9SMiao Xie device->commit_total_bytes) 3257a512bbf8SYan Zheng break; 3258a512bbf8SYan Zheng 3259a512bbf8SYan Zheng if (wait) { 3260a512bbf8SYan Zheng bh = __find_get_block(device->bdev, bytenr / 4096, 3261a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 3262634554dcSJosef Bacik if (!bh) { 3263634554dcSJosef Bacik errors++; 3264634554dcSJosef Bacik continue; 3265634554dcSJosef Bacik } 3266a512bbf8SYan Zheng wait_on_buffer(bh); 32674eedeb75SHisashi Hifumi if (!buffer_uptodate(bh)) 32684eedeb75SHisashi Hifumi errors++; 32694eedeb75SHisashi Hifumi 32704eedeb75SHisashi Hifumi /* drop our reference */ 32714eedeb75SHisashi Hifumi brelse(bh); 32724eedeb75SHisashi Hifumi 32734eedeb75SHisashi Hifumi /* drop the reference from the wait == 0 run */ 3274a512bbf8SYan Zheng brelse(bh); 3275a512bbf8SYan Zheng continue; 3276a512bbf8SYan Zheng } else { 3277a512bbf8SYan Zheng btrfs_set_super_bytenr(sb, bytenr); 3278a512bbf8SYan Zheng 3279a512bbf8SYan Zheng crc = ~(u32)0; 3280b0496686SLiu Bo crc = btrfs_csum_data((char *)sb + 3281a512bbf8SYan Zheng BTRFS_CSUM_SIZE, crc, 3282a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE - 3283a512bbf8SYan Zheng BTRFS_CSUM_SIZE); 3284a512bbf8SYan Zheng btrfs_csum_final(crc, sb->csum); 3285a512bbf8SYan Zheng 32864eedeb75SHisashi Hifumi /* 32874eedeb75SHisashi Hifumi * one reference for us, and we leave it for the 32884eedeb75SHisashi Hifumi * caller 32894eedeb75SHisashi Hifumi */ 3290a512bbf8SYan Zheng bh = __getblk(device->bdev, bytenr / 4096, 3291a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 3292634554dcSJosef Bacik if (!bh) { 3293efe120a0SFrank Holton printk(KERN_ERR "BTRFS: couldn't get super " 3294634554dcSJosef Bacik "buffer head for bytenr %Lu\n", bytenr); 3295634554dcSJosef Bacik errors++; 3296634554dcSJosef Bacik continue; 3297634554dcSJosef Bacik } 3298634554dcSJosef Bacik 3299a512bbf8SYan Zheng memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE); 3300a512bbf8SYan Zheng 33014eedeb75SHisashi Hifumi /* one reference for submit_bh */ 3302a512bbf8SYan Zheng get_bh(bh); 33034eedeb75SHisashi Hifumi 33044eedeb75SHisashi Hifumi set_buffer_uptodate(bh); 3305a512bbf8SYan Zheng lock_buffer(bh); 3306a512bbf8SYan Zheng bh->b_end_io = btrfs_end_buffer_write_sync; 3307442a4f63SStefan Behrens bh->b_private = device; 3308a512bbf8SYan Zheng } 3309a512bbf8SYan Zheng 3310387125fcSChris Mason /* 3311387125fcSChris Mason * we fua the first super. The others we allow 3312387125fcSChris Mason * to go down lazy. 3313387125fcSChris Mason */ 3314e8117c26SWang Shilong if (i == 0) 331521adbd5cSStefan Behrens ret = btrfsic_submit_bh(WRITE_FUA, bh); 3316e8117c26SWang Shilong else 3317e8117c26SWang Shilong ret = btrfsic_submit_bh(WRITE_SYNC, bh); 33184eedeb75SHisashi Hifumi if (ret) 3319a512bbf8SYan Zheng errors++; 3320a512bbf8SYan Zheng } 3321a512bbf8SYan Zheng return errors < i ? 0 : -1; 3322a512bbf8SYan Zheng } 3323a512bbf8SYan Zheng 3324387125fcSChris Mason /* 3325387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 3326387125fcSChris Mason * for the barrier when it is done 3327387125fcSChris Mason */ 3328387125fcSChris Mason static void btrfs_end_empty_barrier(struct bio *bio, int err) 3329387125fcSChris Mason { 3330b25de9d6SChristoph Hellwig if (err) 3331387125fcSChris Mason clear_bit(BIO_UPTODATE, &bio->bi_flags); 3332387125fcSChris Mason if (bio->bi_private) 3333387125fcSChris Mason complete(bio->bi_private); 3334387125fcSChris Mason bio_put(bio); 3335387125fcSChris Mason } 3336387125fcSChris Mason 3337387125fcSChris Mason /* 3338387125fcSChris Mason * trigger flushes for one the devices. If you pass wait == 0, the flushes are 3339387125fcSChris Mason * sent down. With wait == 1, it waits for the previous flush. 3340387125fcSChris Mason * 3341387125fcSChris Mason * any device where the flush fails with eopnotsupp are flagged as not-barrier 3342387125fcSChris Mason * capable 3343387125fcSChris Mason */ 3344387125fcSChris Mason static int write_dev_flush(struct btrfs_device *device, int wait) 3345387125fcSChris Mason { 3346387125fcSChris Mason struct bio *bio; 3347387125fcSChris Mason int ret = 0; 3348387125fcSChris Mason 3349387125fcSChris Mason if (device->nobarriers) 3350387125fcSChris Mason return 0; 3351387125fcSChris Mason 3352387125fcSChris Mason if (wait) { 3353387125fcSChris Mason bio = device->flush_bio; 3354387125fcSChris Mason if (!bio) 3355387125fcSChris Mason return 0; 3356387125fcSChris Mason 3357387125fcSChris Mason wait_for_completion(&device->flush_wait); 3358387125fcSChris Mason 3359b25de9d6SChristoph Hellwig if (!bio_flagged(bio, BIO_UPTODATE)) { 3360387125fcSChris Mason ret = -EIO; 3361442a4f63SStefan Behrens btrfs_dev_stat_inc_and_print(device, 3362442a4f63SStefan Behrens BTRFS_DEV_STAT_FLUSH_ERRS); 3363387125fcSChris Mason } 3364387125fcSChris Mason 3365387125fcSChris Mason /* drop the reference from the wait == 0 run */ 3366387125fcSChris Mason bio_put(bio); 3367387125fcSChris Mason device->flush_bio = NULL; 3368387125fcSChris Mason 3369387125fcSChris Mason return ret; 3370387125fcSChris Mason } 3371387125fcSChris Mason 3372387125fcSChris Mason /* 3373387125fcSChris Mason * one reference for us, and we leave it for the 3374387125fcSChris Mason * caller 3375387125fcSChris Mason */ 33769c017abcSJesper Juhl device->flush_bio = NULL; 33779be3395bSChris Mason bio = btrfs_io_bio_alloc(GFP_NOFS, 0); 3378387125fcSChris Mason if (!bio) 3379387125fcSChris Mason return -ENOMEM; 3380387125fcSChris Mason 3381387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 3382387125fcSChris Mason bio->bi_bdev = device->bdev; 3383387125fcSChris Mason init_completion(&device->flush_wait); 3384387125fcSChris Mason bio->bi_private = &device->flush_wait; 3385387125fcSChris Mason device->flush_bio = bio; 3386387125fcSChris Mason 3387387125fcSChris Mason bio_get(bio); 338821adbd5cSStefan Behrens btrfsic_submit_bio(WRITE_FLUSH, bio); 3389387125fcSChris Mason 3390387125fcSChris Mason return 0; 3391387125fcSChris Mason } 3392387125fcSChris Mason 3393387125fcSChris Mason /* 3394387125fcSChris Mason * send an empty flush down to each device in parallel, 3395387125fcSChris Mason * then wait for them 3396387125fcSChris Mason */ 3397387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 3398387125fcSChris Mason { 3399387125fcSChris Mason struct list_head *head; 3400387125fcSChris Mason struct btrfs_device *dev; 34015af3e8ccSStefan Behrens int errors_send = 0; 34025af3e8ccSStefan Behrens int errors_wait = 0; 3403387125fcSChris Mason int ret; 3404387125fcSChris Mason 3405387125fcSChris Mason /* send down all the barriers */ 3406387125fcSChris Mason head = &info->fs_devices->devices; 3407387125fcSChris Mason list_for_each_entry_rcu(dev, head, dev_list) { 3408f88ba6a2SHidetoshi Seto if (dev->missing) 3409f88ba6a2SHidetoshi Seto continue; 3410387125fcSChris Mason if (!dev->bdev) { 34115af3e8ccSStefan Behrens errors_send++; 3412387125fcSChris Mason continue; 3413387125fcSChris Mason } 3414387125fcSChris Mason if (!dev->in_fs_metadata || !dev->writeable) 3415387125fcSChris Mason continue; 3416387125fcSChris Mason 3417387125fcSChris Mason ret = write_dev_flush(dev, 0); 3418387125fcSChris Mason if (ret) 34195af3e8ccSStefan Behrens errors_send++; 3420387125fcSChris Mason } 3421387125fcSChris Mason 3422387125fcSChris Mason /* wait for all the barriers */ 3423387125fcSChris Mason list_for_each_entry_rcu(dev, head, dev_list) { 3424f88ba6a2SHidetoshi Seto if (dev->missing) 3425f88ba6a2SHidetoshi Seto continue; 3426387125fcSChris Mason if (!dev->bdev) { 34275af3e8ccSStefan Behrens errors_wait++; 3428387125fcSChris Mason continue; 3429387125fcSChris Mason } 3430387125fcSChris Mason if (!dev->in_fs_metadata || !dev->writeable) 3431387125fcSChris Mason continue; 3432387125fcSChris Mason 3433387125fcSChris Mason ret = write_dev_flush(dev, 1); 3434387125fcSChris Mason if (ret) 34355af3e8ccSStefan Behrens errors_wait++; 3436387125fcSChris Mason } 34375af3e8ccSStefan Behrens if (errors_send > info->num_tolerated_disk_barrier_failures || 34385af3e8ccSStefan Behrens errors_wait > info->num_tolerated_disk_barrier_failures) 3439387125fcSChris Mason return -EIO; 3440387125fcSChris Mason return 0; 3441387125fcSChris Mason } 3442387125fcSChris Mason 3443943c6e99SZhao Lei int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 3444943c6e99SZhao Lei { 3445943c6e99SZhao Lei if ((flags & (BTRFS_BLOCK_GROUP_DUP | 3446943c6e99SZhao Lei BTRFS_BLOCK_GROUP_RAID0 | 3447943c6e99SZhao Lei BTRFS_AVAIL_ALLOC_BIT_SINGLE)) || 3448943c6e99SZhao Lei ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0)) 3449943c6e99SZhao Lei return 0; 3450943c6e99SZhao Lei 3451943c6e99SZhao Lei if (flags & (BTRFS_BLOCK_GROUP_RAID1 | 3452943c6e99SZhao Lei BTRFS_BLOCK_GROUP_RAID5 | 3453943c6e99SZhao Lei BTRFS_BLOCK_GROUP_RAID10)) 3454943c6e99SZhao Lei return 1; 3455943c6e99SZhao Lei 3456943c6e99SZhao Lei if (flags & BTRFS_BLOCK_GROUP_RAID6) 3457943c6e99SZhao Lei return 2; 3458943c6e99SZhao Lei 3459943c6e99SZhao Lei pr_warn("BTRFS: unknown raid type: %llu\n", flags); 3460943c6e99SZhao Lei return 0; 3461943c6e99SZhao Lei } 3462943c6e99SZhao Lei 34635af3e8ccSStefan Behrens int btrfs_calc_num_tolerated_disk_barrier_failures( 34645af3e8ccSStefan Behrens struct btrfs_fs_info *fs_info) 34655af3e8ccSStefan Behrens { 34665af3e8ccSStefan Behrens struct btrfs_ioctl_space_info space; 34675af3e8ccSStefan Behrens struct btrfs_space_info *sinfo; 34685af3e8ccSStefan Behrens u64 types[] = {BTRFS_BLOCK_GROUP_DATA, 34695af3e8ccSStefan Behrens BTRFS_BLOCK_GROUP_SYSTEM, 34705af3e8ccSStefan Behrens BTRFS_BLOCK_GROUP_METADATA, 34715af3e8ccSStefan Behrens BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA}; 34725af3e8ccSStefan Behrens int i; 34735af3e8ccSStefan Behrens int c; 34745af3e8ccSStefan Behrens int num_tolerated_disk_barrier_failures = 34755af3e8ccSStefan Behrens (int)fs_info->fs_devices->num_devices; 34765af3e8ccSStefan Behrens 34772c458045SZhao Lei for (i = 0; i < ARRAY_SIZE(types); i++) { 34785af3e8ccSStefan Behrens struct btrfs_space_info *tmp; 34795af3e8ccSStefan Behrens 34805af3e8ccSStefan Behrens sinfo = NULL; 34815af3e8ccSStefan Behrens rcu_read_lock(); 34825af3e8ccSStefan Behrens list_for_each_entry_rcu(tmp, &fs_info->space_info, list) { 34835af3e8ccSStefan Behrens if (tmp->flags == types[i]) { 34845af3e8ccSStefan Behrens sinfo = tmp; 34855af3e8ccSStefan Behrens break; 34865af3e8ccSStefan Behrens } 34875af3e8ccSStefan Behrens } 34885af3e8ccSStefan Behrens rcu_read_unlock(); 34895af3e8ccSStefan Behrens 34905af3e8ccSStefan Behrens if (!sinfo) 34915af3e8ccSStefan Behrens continue; 34925af3e8ccSStefan Behrens 34935af3e8ccSStefan Behrens down_read(&sinfo->groups_sem); 34945af3e8ccSStefan Behrens for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 34955af3e8ccSStefan Behrens u64 flags; 34965af3e8ccSStefan Behrens 34972c458045SZhao Lei if (list_empty(&sinfo->block_groups[c])) 34982c458045SZhao Lei continue; 34992c458045SZhao Lei 35002c458045SZhao Lei btrfs_get_block_group_info(&sinfo->block_groups[c], 35012c458045SZhao Lei &space); 35022c458045SZhao Lei if (space.total_bytes == 0 || space.used_bytes == 0) 35035af3e8ccSStefan Behrens continue; 35045af3e8ccSStefan Behrens flags = space.flags; 3505943c6e99SZhao Lei 3506943c6e99SZhao Lei num_tolerated_disk_barrier_failures = min( 3507943c6e99SZhao Lei num_tolerated_disk_barrier_failures, 3508943c6e99SZhao Lei btrfs_get_num_tolerated_disk_barrier_failures( 3509943c6e99SZhao Lei flags)); 351053b381b3SDavid Woodhouse } 35115af3e8ccSStefan Behrens up_read(&sinfo->groups_sem); 35125af3e8ccSStefan Behrens } 35135af3e8ccSStefan Behrens 35145af3e8ccSStefan Behrens return num_tolerated_disk_barrier_failures; 35155af3e8ccSStefan Behrens } 35165af3e8ccSStefan Behrens 351748a3b636SEric Sandeen static int write_all_supers(struct btrfs_root *root, int max_mirrors) 3518f2984462SChris Mason { 3519e5e9a520SChris Mason struct list_head *head; 3520f2984462SChris Mason struct btrfs_device *dev; 3521a061fc8dSChris Mason struct btrfs_super_block *sb; 3522f2984462SChris Mason struct btrfs_dev_item *dev_item; 3523f2984462SChris Mason int ret; 3524f2984462SChris Mason int do_barriers; 3525a236aed1SChris Mason int max_errors; 3526a236aed1SChris Mason int total_errors = 0; 3527a061fc8dSChris Mason u64 flags; 3528f2984462SChris Mason 3529f2984462SChris Mason do_barriers = !btrfs_test_opt(root, NOBARRIER); 3530af31f5e5SChris Mason backup_super_roots(root->fs_info); 3531f2984462SChris Mason 35326c41761fSDavid Sterba sb = root->fs_info->super_for_commit; 3533a061fc8dSChris Mason dev_item = &sb->dev_item; 3534e5e9a520SChris Mason 3535174ba509SChris Mason mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 3536e5e9a520SChris Mason head = &root->fs_info->fs_devices->devices; 3537d7306801SFilipe David Borba Manana max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1; 3538387125fcSChris Mason 35395af3e8ccSStefan Behrens if (do_barriers) { 35405af3e8ccSStefan Behrens ret = barrier_all_devices(root->fs_info); 35415af3e8ccSStefan Behrens if (ret) { 35425af3e8ccSStefan Behrens mutex_unlock( 35435af3e8ccSStefan Behrens &root->fs_info->fs_devices->device_list_mutex); 35445af3e8ccSStefan Behrens btrfs_error(root->fs_info, ret, 35455af3e8ccSStefan Behrens "errors while submitting device barriers."); 35465af3e8ccSStefan Behrens return ret; 35475af3e8ccSStefan Behrens } 35485af3e8ccSStefan Behrens } 3549387125fcSChris Mason 35501f78160cSXiao Guangrong list_for_each_entry_rcu(dev, head, dev_list) { 3551dfe25020SChris Mason if (!dev->bdev) { 3552dfe25020SChris Mason total_errors++; 3553dfe25020SChris Mason continue; 3554dfe25020SChris Mason } 35552b82032cSYan Zheng if (!dev->in_fs_metadata || !dev->writeable) 3556dfe25020SChris Mason continue; 3557dfe25020SChris Mason 35582b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 3559a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 3560a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 35617df69d3eSMiao Xie btrfs_set_stack_device_total_bytes(dev_item, 3562935e5cc9SMiao Xie dev->commit_total_bytes); 3563ce7213c7SMiao Xie btrfs_set_stack_device_bytes_used(dev_item, 3564ce7213c7SMiao Xie dev->commit_bytes_used); 3565a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 3566a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 3567a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 3568a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 35692b82032cSYan Zheng memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE); 3570a512bbf8SYan Zheng 3571a061fc8dSChris Mason flags = btrfs_super_flags(sb); 3572a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 3573f2984462SChris Mason 3574a512bbf8SYan Zheng ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors); 3575a236aed1SChris Mason if (ret) 3576a236aed1SChris Mason total_errors++; 3577f2984462SChris Mason } 3578a236aed1SChris Mason if (total_errors > max_errors) { 3579efe120a0SFrank Holton btrfs_err(root->fs_info, "%d errors while writing supers", 3580d397712bSChris Mason total_errors); 3581a724b436SStefan Behrens mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 358279787eaaSJeff Mahoney 35839d565ba4SStefan Behrens /* FUA is masked off if unsupported and can't be the reason */ 35849d565ba4SStefan Behrens btrfs_error(root->fs_info, -EIO, 35859d565ba4SStefan Behrens "%d errors while writing supers", total_errors); 35869d565ba4SStefan Behrens return -EIO; 3587a236aed1SChris Mason } 3588f2984462SChris Mason 3589a512bbf8SYan Zheng total_errors = 0; 35901f78160cSXiao Guangrong list_for_each_entry_rcu(dev, head, dev_list) { 3591dfe25020SChris Mason if (!dev->bdev) 3592dfe25020SChris Mason continue; 35932b82032cSYan Zheng if (!dev->in_fs_metadata || !dev->writeable) 3594dfe25020SChris Mason continue; 3595dfe25020SChris Mason 3596a512bbf8SYan Zheng ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors); 3597a512bbf8SYan Zheng if (ret) 35981259ab75SChris Mason total_errors++; 3599f2984462SChris Mason } 3600174ba509SChris Mason mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 3601a236aed1SChris Mason if (total_errors > max_errors) { 360279787eaaSJeff Mahoney btrfs_error(root->fs_info, -EIO, 360379787eaaSJeff Mahoney "%d errors while writing supers", total_errors); 360479787eaaSJeff Mahoney return -EIO; 3605a236aed1SChris Mason } 3606f2984462SChris Mason return 0; 3607f2984462SChris Mason } 3608f2984462SChris Mason 3609a512bbf8SYan Zheng int write_ctree_super(struct btrfs_trans_handle *trans, 3610a512bbf8SYan Zheng struct btrfs_root *root, int max_mirrors) 3611cfaa7295SChris Mason { 3612f570e757SRashika return write_all_supers(root, max_mirrors); 3613cfaa7295SChris Mason } 3614cfaa7295SChris Mason 3615cb517eabSMiao Xie /* Drop a fs root from the radix tree and free it. */ 3616cb517eabSMiao Xie void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 3617cb517eabSMiao Xie struct btrfs_root *root) 36182619ba1fSChris Mason { 36194df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 36202619ba1fSChris Mason radix_tree_delete(&fs_info->fs_roots_radix, 36212619ba1fSChris Mason (unsigned long)root->root_key.objectid); 36224df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 362376dda93cSYan, Zheng 362476dda93cSYan, Zheng if (btrfs_root_refs(&root->root_item) == 0) 362576dda93cSYan, Zheng synchronize_srcu(&fs_info->subvol_srcu); 362676dda93cSYan, Zheng 36271a4319ccSLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) 36283321719eSLiu Bo btrfs_free_log(NULL, root); 36293321719eSLiu Bo 3630faa2dbf0SJosef Bacik if (root->free_ino_pinned) 3631581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_pinned); 3632faa2dbf0SJosef Bacik if (root->free_ino_ctl) 3633581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_ctl); 36344df27c4dSYan, Zheng free_fs_root(root); 36354df27c4dSYan, Zheng } 36364df27c4dSYan, Zheng 36374df27c4dSYan, Zheng static void free_fs_root(struct btrfs_root *root) 36384df27c4dSYan, Zheng { 363957cdc8dbSDavid Sterba iput(root->ino_cache_inode); 36404df27c4dSYan, Zheng WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 36411cb048f5SFilipe David Borba Manana btrfs_free_block_rsv(root, root->orphan_block_rsv); 36421cb048f5SFilipe David Borba Manana root->orphan_block_rsv = NULL; 36430ee5dc67SAl Viro if (root->anon_dev) 36440ee5dc67SAl Viro free_anon_bdev(root->anon_dev); 36458257b2dcSMiao Xie if (root->subv_writers) 36468257b2dcSMiao Xie btrfs_free_subvolume_writers(root->subv_writers); 36475f39d397SChris Mason free_extent_buffer(root->node); 36485f39d397SChris Mason free_extent_buffer(root->commit_root); 3649581bb050SLi Zefan kfree(root->free_ino_ctl); 3650581bb050SLi Zefan kfree(root->free_ino_pinned); 365158176a96SJosef Bacik kfree(root->name); 3652b0feb9d9SMiao Xie btrfs_put_fs_root(root); 36532619ba1fSChris Mason } 36542619ba1fSChris Mason 3655cb517eabSMiao Xie void btrfs_free_fs_root(struct btrfs_root *root) 3656cb517eabSMiao Xie { 3657cb517eabSMiao Xie free_fs_root(root); 3658eb60ceacSChris Mason } 3659eb60ceacSChris Mason 3660c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 3661c146afadSYan Zheng { 3662c146afadSYan Zheng u64 root_objectid = 0; 3663c146afadSYan Zheng struct btrfs_root *gang[8]; 366465d33fd7SQu Wenruo int i = 0; 366565d33fd7SQu Wenruo int err = 0; 366665d33fd7SQu Wenruo unsigned int ret = 0; 366765d33fd7SQu Wenruo int index; 3668c146afadSYan Zheng 3669c146afadSYan Zheng while (1) { 367065d33fd7SQu Wenruo index = srcu_read_lock(&fs_info->subvol_srcu); 3671c146afadSYan Zheng ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 3672c146afadSYan Zheng (void **)gang, root_objectid, 3673c146afadSYan Zheng ARRAY_SIZE(gang)); 367465d33fd7SQu Wenruo if (!ret) { 367565d33fd7SQu Wenruo srcu_read_unlock(&fs_info->subvol_srcu, index); 3676c146afadSYan Zheng break; 367765d33fd7SQu Wenruo } 36785d4f98a2SYan Zheng root_objectid = gang[ret - 1]->root_key.objectid + 1; 367966b4ffd1SJosef Bacik 368065d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 368165d33fd7SQu Wenruo /* Avoid to grab roots in dead_roots */ 368265d33fd7SQu Wenruo if (btrfs_root_refs(&gang[i]->root_item) == 0) { 368365d33fd7SQu Wenruo gang[i] = NULL; 368465d33fd7SQu Wenruo continue; 368565d33fd7SQu Wenruo } 368665d33fd7SQu Wenruo /* grab all the search result for later use */ 368765d33fd7SQu Wenruo gang[i] = btrfs_grab_fs_root(gang[i]); 368865d33fd7SQu Wenruo } 368965d33fd7SQu Wenruo srcu_read_unlock(&fs_info->subvol_srcu, index); 369065d33fd7SQu Wenruo 369165d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 369265d33fd7SQu Wenruo if (!gang[i]) 369365d33fd7SQu Wenruo continue; 3694c146afadSYan Zheng root_objectid = gang[i]->root_key.objectid; 369566b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(gang[i]); 369666b4ffd1SJosef Bacik if (err) 369765d33fd7SQu Wenruo break; 369865d33fd7SQu Wenruo btrfs_put_fs_root(gang[i]); 3699c146afadSYan Zheng } 3700c146afadSYan Zheng root_objectid++; 3701c146afadSYan Zheng } 370265d33fd7SQu Wenruo 370365d33fd7SQu Wenruo /* release the uncleaned roots due to error */ 370465d33fd7SQu Wenruo for (; i < ret; i++) { 370565d33fd7SQu Wenruo if (gang[i]) 370665d33fd7SQu Wenruo btrfs_put_fs_root(gang[i]); 370765d33fd7SQu Wenruo } 370865d33fd7SQu Wenruo return err; 3709c146afadSYan Zheng } 3710c146afadSYan Zheng 3711c146afadSYan Zheng int btrfs_commit_super(struct btrfs_root *root) 3712c146afadSYan Zheng { 3713c146afadSYan Zheng struct btrfs_trans_handle *trans; 3714c146afadSYan Zheng 3715c146afadSYan Zheng mutex_lock(&root->fs_info->cleaner_mutex); 371624bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 3717c146afadSYan Zheng mutex_unlock(&root->fs_info->cleaner_mutex); 37189d1a2a3aSDavid Sterba wake_up_process(root->fs_info->cleaner_kthread); 3719c71bf099SYan, Zheng 3720c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 3721c71bf099SYan, Zheng down_write(&root->fs_info->cleanup_work_sem); 3722c71bf099SYan, Zheng up_write(&root->fs_info->cleanup_work_sem); 3723c71bf099SYan, Zheng 37247a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 37253612b495STsutomu Itoh if (IS_ERR(trans)) 37263612b495STsutomu Itoh return PTR_ERR(trans); 3727d52c1bccSLiu Bo return btrfs_commit_transaction(trans, root); 3728c146afadSYan Zheng } 3729c146afadSYan Zheng 37303abdbd78SDavid Sterba void close_ctree(struct btrfs_root *root) 3731eb60ceacSChris Mason { 37320f7d52f4SChris Mason struct btrfs_fs_info *fs_info = root->fs_info; 3733c146afadSYan Zheng int ret; 3734e089f05cSChris Mason 3735facda1e7SChris Mason fs_info->closing = 1; 3736a2135011SChris Mason smp_mb(); 3737a2135011SChris Mason 3738803b2f54SStefan Behrens /* wait for the uuid_scan task to finish */ 3739803b2f54SStefan Behrens down(&fs_info->uuid_tree_rescan_sem); 3740803b2f54SStefan Behrens /* avoid complains from lockdep et al., set sem back to initial state */ 3741803b2f54SStefan Behrens up(&fs_info->uuid_tree_rescan_sem); 3742803b2f54SStefan Behrens 3743837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 3744aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 3745837d5b6eSIlya Dryomov 37468dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 37478dabb742SStefan Behrens 3748aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 37494cb5300bSChris Mason 37504cb5300bSChris Mason /* wait for any defraggers to finish */ 37514cb5300bSChris Mason wait_event(fs_info->transaction_wait, 37524cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 37534cb5300bSChris Mason 37544cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 375526176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 37564cb5300bSChris Mason 375721c7e756SMiao Xie cancel_work_sync(&fs_info->async_reclaim_work); 375821c7e756SMiao Xie 3759c146afadSYan Zheng if (!(fs_info->sb->s_flags & MS_RDONLY)) { 3760e44163e1SJeff Mahoney /* 3761e44163e1SJeff Mahoney * If the cleaner thread is stopped and there are 3762e44163e1SJeff Mahoney * block groups queued for removal, the deletion will be 3763e44163e1SJeff Mahoney * skipped when we quit the cleaner thread. 3764e44163e1SJeff Mahoney */ 3765e44163e1SJeff Mahoney mutex_lock(&root->fs_info->cleaner_mutex); 3766e44163e1SJeff Mahoney btrfs_delete_unused_bgs(root->fs_info); 3767e44163e1SJeff Mahoney mutex_unlock(&root->fs_info->cleaner_mutex); 3768e44163e1SJeff Mahoney 3769c146afadSYan Zheng ret = btrfs_commit_super(root); 3770d397712bSChris Mason if (ret) 377104892340SEric Sandeen btrfs_err(fs_info, "commit super ret %d", ret); 3772c146afadSYan Zheng } 3773ed2ff2cbSChris Mason 377487533c47SMiao Xie if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) 377568ce9682SStefan Behrens btrfs_error_commit_super(root); 3776acce952bSliubo 3777e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 3778e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 37798929ecfaSYan, Zheng 3780f25784b3SYan Zheng fs_info->closing = 2; 3781f25784b3SYan Zheng smp_mb(); 3782f25784b3SYan Zheng 378304892340SEric Sandeen btrfs_free_qgroup_config(fs_info); 3784bcef60f2SArne Jansen 3785963d678bSMiao Xie if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 378604892340SEric Sandeen btrfs_info(fs_info, "at unmount delalloc count %lld", 3787963d678bSMiao Xie percpu_counter_sum(&fs_info->delalloc_bytes)); 3788b0c68f8bSChris Mason } 378931153d81SYan Zheng 37905ac1d209SJeff Mahoney btrfs_sysfs_remove_one(fs_info); 3791b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 37925ac1d209SJeff Mahoney 3793faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 3794d10c5f31SChris Mason 37951a4319ccSLiu Bo btrfs_put_block_group_cache(fs_info); 37961a4319ccSLiu Bo 37972b1360daSJosef Bacik btrfs_free_block_groups(fs_info); 37982b1360daSJosef Bacik 3799de348ee0SWang Shilong /* 3800de348ee0SWang Shilong * we must make sure there is not any read request to 3801de348ee0SWang Shilong * submit after we stopping all workers. 3802de348ee0SWang Shilong */ 3803de348ee0SWang Shilong invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 380496192499SJosef Bacik btrfs_stop_all_workers(fs_info); 380596192499SJosef Bacik 380647ab2a6cSJosef Bacik fs_info->open = 0; 380713e6c37bSJosef Bacik free_root_pointers(fs_info, 1); 38089ad6b7bcSChris Mason 380913e6c37bSJosef Bacik iput(fs_info->btree_inode); 3810d6bfde87SChris Mason 381121adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 381221adbd5cSStefan Behrens if (btrfs_test_opt(root, CHECK_INTEGRITY)) 381321adbd5cSStefan Behrens btrfsic_unmount(root, fs_info->fs_devices); 381421adbd5cSStefan Behrens #endif 381521adbd5cSStefan Behrens 3816dfe25020SChris Mason btrfs_close_devices(fs_info->fs_devices); 38170b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 3818b248a415SChris Mason 3819e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 3820963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 3821c404e0dcSMiao Xie percpu_counter_destroy(&fs_info->bio_counter); 382204160088SChris Mason bdi_destroy(&fs_info->bdi); 382376dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 38240b86a832SChris Mason 382553b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 382653b381b3SDavid Woodhouse 3827cdfb080eSChris Mason __btrfs_free_block_rsv(root->orphan_block_rsv); 38281cb048f5SFilipe David Borba Manana root->orphan_block_rsv = NULL; 382904216820SFilipe Manana 383004216820SFilipe Manana lock_chunks(root); 383104216820SFilipe Manana while (!list_empty(&fs_info->pinned_chunks)) { 383204216820SFilipe Manana struct extent_map *em; 383304216820SFilipe Manana 383404216820SFilipe Manana em = list_first_entry(&fs_info->pinned_chunks, 383504216820SFilipe Manana struct extent_map, list); 383604216820SFilipe Manana list_del_init(&em->list); 383704216820SFilipe Manana free_extent_map(em); 383804216820SFilipe Manana } 383904216820SFilipe Manana unlock_chunks(root); 3840eb60ceacSChris Mason } 3841eb60ceacSChris Mason 3842b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 3843b9fab919SChris Mason int atomic) 3844ccd467d6SChris Mason { 38451259ab75SChris Mason int ret; 3846727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 38471259ab75SChris Mason 38480b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 38491259ab75SChris Mason if (!ret) 38501259ab75SChris Mason return ret; 38511259ab75SChris Mason 38521259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 3853b9fab919SChris Mason parent_transid, atomic); 3854b9fab919SChris Mason if (ret == -EAGAIN) 3855b9fab919SChris Mason return ret; 38561259ab75SChris Mason return !ret; 38575f39d397SChris Mason } 38586702ed49SChris Mason 38595f39d397SChris Mason int btrfs_set_buffer_uptodate(struct extent_buffer *buf) 38605f39d397SChris Mason { 38610b32f4bbSJosef Bacik return set_extent_buffer_uptodate(buf); 38625f39d397SChris Mason } 38635f39d397SChris Mason 38645f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 38655f39d397SChris Mason { 386606ea65a3SJosef Bacik struct btrfs_root *root; 38675f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 3868b9473439SChris Mason int was_dirty; 3869b4ce94deSChris Mason 387006ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 387106ea65a3SJosef Bacik /* 387206ea65a3SJosef Bacik * This is a fast path so only do this check if we have sanity tests 387306ea65a3SJosef Bacik * enabled. Normal people shouldn't be marking dummy buffers as dirty 387406ea65a3SJosef Bacik * outside of the sanity tests. 387506ea65a3SJosef Bacik */ 387606ea65a3SJosef Bacik if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags))) 387706ea65a3SJosef Bacik return; 387806ea65a3SJosef Bacik #endif 387906ea65a3SJosef Bacik root = BTRFS_I(buf->pages[0]->mapping->host)->root; 3880b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 388131b1a2bdSJulia Lawall if (transid != root->fs_info->generation) 388231b1a2bdSJulia Lawall WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, " 3883d397712bSChris Mason "found %llu running %llu\n", 3884c1c9ff7cSGeert Uytterhoeven buf->start, transid, root->fs_info->generation); 38850b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 3886e2d84521SMiao Xie if (!was_dirty) 3887e2d84521SMiao Xie __percpu_counter_add(&root->fs_info->dirty_metadata_bytes, 3888e2d84521SMiao Xie buf->len, 3889e2d84521SMiao Xie root->fs_info->dirty_metadata_batch); 38901f21ef0aSFilipe Manana #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 38911f21ef0aSFilipe Manana if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) { 38921f21ef0aSFilipe Manana btrfs_print_leaf(root, buf); 38931f21ef0aSFilipe Manana ASSERT(0); 38941f21ef0aSFilipe Manana } 38951f21ef0aSFilipe Manana #endif 3896eb60ceacSChris Mason } 3897eb60ceacSChris Mason 3898b53d3f5dSLiu Bo static void __btrfs_btree_balance_dirty(struct btrfs_root *root, 3899b53d3f5dSLiu Bo int flush_delayed) 390035b7e476SChris Mason { 3901188de649SChris Mason /* 3902188de649SChris Mason * looks as though older kernels can get into trouble with 3903188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 3904188de649SChris Mason */ 3905e2d84521SMiao Xie int ret; 3906d6bfde87SChris Mason 39076933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 3908d6bfde87SChris Mason return; 3909d6bfde87SChris Mason 3910b53d3f5dSLiu Bo if (flush_delayed) 391116cdcec7SMiao Xie btrfs_balance_delayed_items(root); 391216cdcec7SMiao Xie 3913e2d84521SMiao Xie ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes, 3914e2d84521SMiao Xie BTRFS_DIRTY_METADATA_THRESH); 3915e2d84521SMiao Xie if (ret > 0) { 3916d0e1d66bSNamjae Jeon balance_dirty_pages_ratelimited( 3917d0e1d66bSNamjae Jeon root->fs_info->btree_inode->i_mapping); 391816cdcec7SMiao Xie } 391916cdcec7SMiao Xie return; 392016cdcec7SMiao Xie } 392116cdcec7SMiao Xie 3922b53d3f5dSLiu Bo void btrfs_btree_balance_dirty(struct btrfs_root *root) 392316cdcec7SMiao Xie { 3924b53d3f5dSLiu Bo __btrfs_btree_balance_dirty(root, 1); 392535b7e476SChris Mason } 3926b53d3f5dSLiu Bo 3927b53d3f5dSLiu Bo void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root) 3928b53d3f5dSLiu Bo { 3929b53d3f5dSLiu Bo __btrfs_btree_balance_dirty(root, 0); 3930d6bfde87SChris Mason } 39316b80053dSChris Mason 3932ca7a79adSChris Mason int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid) 39336b80053dSChris Mason { 3934727011e0SChris Mason struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root; 39350b32f4bbSJosef Bacik return btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 39366b80053dSChris Mason } 39370da5468fSChris Mason 3938fcd1f065SDavid Sterba static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info, 3939acce952bSliubo int read_only) 3940acce952bSliubo { 3941c926093eSDavid Sterba struct btrfs_super_block *sb = fs_info->super_copy; 3942c926093eSDavid Sterba int ret = 0; 3943c926093eSDavid Sterba 394421e7626bSDavid Sterba if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 394521e7626bSDavid Sterba printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n", 394621e7626bSDavid Sterba btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 3947c926093eSDavid Sterba ret = -EINVAL; 3948c926093eSDavid Sterba } 394921e7626bSDavid Sterba if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 395021e7626bSDavid Sterba printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n", 395121e7626bSDavid Sterba btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 3952c926093eSDavid Sterba ret = -EINVAL; 3953c926093eSDavid Sterba } 395421e7626bSDavid Sterba if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 395521e7626bSDavid Sterba printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n", 395621e7626bSDavid Sterba btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 3957c926093eSDavid Sterba ret = -EINVAL; 3958c926093eSDavid Sterba } 3959c926093eSDavid Sterba 39601104a885SDavid Sterba /* 3961c926093eSDavid Sterba * The common minimum, we don't know if we can trust the nodesize/sectorsize 3962c926093eSDavid Sterba * items yet, they'll be verified later. Issue just a warning. 39631104a885SDavid Sterba */ 396421e7626bSDavid Sterba if (!IS_ALIGNED(btrfs_super_root(sb), 4096)) 3965c926093eSDavid Sterba printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n", 3966cd743facSDavid Sterba btrfs_super_root(sb)); 396721e7626bSDavid Sterba if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096)) 3968cd743facSDavid Sterba printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n", 3969cd743facSDavid Sterba btrfs_super_chunk_root(sb)); 397021e7626bSDavid Sterba if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096)) 3971cd743facSDavid Sterba printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n", 397221e7626bSDavid Sterba btrfs_super_log_root(sb)); 3973c926093eSDavid Sterba 397475d6ad38SDavid Sterba /* 397575d6ad38SDavid Sterba * Check the lower bound, the alignment and other constraints are 397675d6ad38SDavid Sterba * checked later. 397775d6ad38SDavid Sterba */ 397875d6ad38SDavid Sterba if (btrfs_super_nodesize(sb) < 4096) { 397975d6ad38SDavid Sterba printk(KERN_ERR "BTRFS: nodesize too small: %u < 4096\n", 398075d6ad38SDavid Sterba btrfs_super_nodesize(sb)); 398175d6ad38SDavid Sterba ret = -EINVAL; 398275d6ad38SDavid Sterba } 398375d6ad38SDavid Sterba if (btrfs_super_sectorsize(sb) < 4096) { 398475d6ad38SDavid Sterba printk(KERN_ERR "BTRFS: sectorsize too small: %u < 4096\n", 398575d6ad38SDavid Sterba btrfs_super_sectorsize(sb)); 398675d6ad38SDavid Sterba ret = -EINVAL; 398775d6ad38SDavid Sterba } 398875d6ad38SDavid Sterba 3989c926093eSDavid Sterba if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) { 3990c926093eSDavid Sterba printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n", 3991c926093eSDavid Sterba fs_info->fsid, sb->dev_item.fsid); 3992c926093eSDavid Sterba ret = -EINVAL; 3993c926093eSDavid Sterba } 3994c926093eSDavid Sterba 3995c926093eSDavid Sterba /* 3996c926093eSDavid Sterba * Hint to catch really bogus numbers, bitflips or so, more exact checks are 3997c926093eSDavid Sterba * done later 3998c926093eSDavid Sterba */ 399921e7626bSDavid Sterba if (btrfs_super_num_devices(sb) > (1UL << 31)) 4000c926093eSDavid Sterba printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n", 400121e7626bSDavid Sterba btrfs_super_num_devices(sb)); 400275d6ad38SDavid Sterba if (btrfs_super_num_devices(sb) == 0) { 400375d6ad38SDavid Sterba printk(KERN_ERR "BTRFS: number of devices is 0\n"); 400475d6ad38SDavid Sterba ret = -EINVAL; 400575d6ad38SDavid Sterba } 4006c926093eSDavid Sterba 400721e7626bSDavid Sterba if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) { 4008c926093eSDavid Sterba printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n", 400921e7626bSDavid Sterba btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 4010c926093eSDavid Sterba ret = -EINVAL; 4011c926093eSDavid Sterba } 4012c926093eSDavid Sterba 4013c926093eSDavid Sterba /* 4014ce7fca5fSDavid Sterba * Obvious sys_chunk_array corruptions, it must hold at least one key 4015ce7fca5fSDavid Sterba * and one chunk 4016ce7fca5fSDavid Sterba */ 4017ce7fca5fSDavid Sterba if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 4018ce7fca5fSDavid Sterba printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n", 4019ce7fca5fSDavid Sterba btrfs_super_sys_array_size(sb), 4020ce7fca5fSDavid Sterba BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 4021ce7fca5fSDavid Sterba ret = -EINVAL; 4022ce7fca5fSDavid Sterba } 4023ce7fca5fSDavid Sterba if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 4024ce7fca5fSDavid Sterba + sizeof(struct btrfs_chunk)) { 4025d2207129SFabian Frederick printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n", 4026ce7fca5fSDavid Sterba btrfs_super_sys_array_size(sb), 4027ce7fca5fSDavid Sterba sizeof(struct btrfs_disk_key) 4028ce7fca5fSDavid Sterba + sizeof(struct btrfs_chunk)); 4029ce7fca5fSDavid Sterba ret = -EINVAL; 4030ce7fca5fSDavid Sterba } 4031ce7fca5fSDavid Sterba 4032ce7fca5fSDavid Sterba /* 4033c926093eSDavid Sterba * The generation is a global counter, we'll trust it more than the others 4034c926093eSDavid Sterba * but it's still possible that it's the one that's wrong. 4035c926093eSDavid Sterba */ 403621e7626bSDavid Sterba if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 4037c926093eSDavid Sterba printk(KERN_WARNING 4038c926093eSDavid Sterba "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n", 403921e7626bSDavid Sterba btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb)); 404021e7626bSDavid Sterba if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 404121e7626bSDavid Sterba && btrfs_super_cache_generation(sb) != (u64)-1) 4042c926093eSDavid Sterba printk(KERN_WARNING 4043c926093eSDavid Sterba "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n", 404421e7626bSDavid Sterba btrfs_super_generation(sb), btrfs_super_cache_generation(sb)); 4045c926093eSDavid Sterba 4046c926093eSDavid Sterba return ret; 4047fcd1f065SDavid Sterba } 4048fcd1f065SDavid Sterba 404948a3b636SEric Sandeen static void btrfs_error_commit_super(struct btrfs_root *root) 4050acce952bSliubo { 4051acce952bSliubo mutex_lock(&root->fs_info->cleaner_mutex); 4052acce952bSliubo btrfs_run_delayed_iputs(root); 4053acce952bSliubo mutex_unlock(&root->fs_info->cleaner_mutex); 4054acce952bSliubo 4055acce952bSliubo down_write(&root->fs_info->cleanup_work_sem); 4056acce952bSliubo up_write(&root->fs_info->cleanup_work_sem); 4057acce952bSliubo 4058acce952bSliubo /* cleanup FS via transaction */ 4059acce952bSliubo btrfs_cleanup_transaction(root); 4060acce952bSliubo } 4061acce952bSliubo 4062143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4063acce952bSliubo { 4064acce952bSliubo struct btrfs_ordered_extent *ordered; 4065acce952bSliubo 4066199c2a9cSMiao Xie spin_lock(&root->ordered_extent_lock); 4067779880efSJosef Bacik /* 4068779880efSJosef Bacik * This will just short circuit the ordered completion stuff which will 4069779880efSJosef Bacik * make sure the ordered extent gets properly cleaned up. 4070779880efSJosef Bacik */ 4071199c2a9cSMiao Xie list_for_each_entry(ordered, &root->ordered_extents, 4072779880efSJosef Bacik root_extent_list) 4073779880efSJosef Bacik set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4074199c2a9cSMiao Xie spin_unlock(&root->ordered_extent_lock); 4075199c2a9cSMiao Xie } 4076199c2a9cSMiao Xie 4077199c2a9cSMiao Xie static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4078199c2a9cSMiao Xie { 4079199c2a9cSMiao Xie struct btrfs_root *root; 4080199c2a9cSMiao Xie struct list_head splice; 4081199c2a9cSMiao Xie 4082199c2a9cSMiao Xie INIT_LIST_HEAD(&splice); 4083199c2a9cSMiao Xie 4084199c2a9cSMiao Xie spin_lock(&fs_info->ordered_root_lock); 4085199c2a9cSMiao Xie list_splice_init(&fs_info->ordered_roots, &splice); 4086199c2a9cSMiao Xie while (!list_empty(&splice)) { 4087199c2a9cSMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4088199c2a9cSMiao Xie ordered_root); 40891de2cfdeSJosef Bacik list_move_tail(&root->ordered_root, 40901de2cfdeSJosef Bacik &fs_info->ordered_roots); 4091199c2a9cSMiao Xie 40922a85d9caSLiu Bo spin_unlock(&fs_info->ordered_root_lock); 4093199c2a9cSMiao Xie btrfs_destroy_ordered_extents(root); 4094199c2a9cSMiao Xie 40952a85d9caSLiu Bo cond_resched(); 40962a85d9caSLiu Bo spin_lock(&fs_info->ordered_root_lock); 4097199c2a9cSMiao Xie } 4098199c2a9cSMiao Xie spin_unlock(&fs_info->ordered_root_lock); 4099acce952bSliubo } 4100acce952bSliubo 410135a3621bSStefan Behrens static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 4102acce952bSliubo struct btrfs_root *root) 4103acce952bSliubo { 4104acce952bSliubo struct rb_node *node; 4105acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 4106acce952bSliubo struct btrfs_delayed_ref_node *ref; 4107acce952bSliubo int ret = 0; 4108acce952bSliubo 4109acce952bSliubo delayed_refs = &trans->delayed_refs; 4110acce952bSliubo 4111acce952bSliubo spin_lock(&delayed_refs->lock); 4112d7df2c79SJosef Bacik if (atomic_read(&delayed_refs->num_entries) == 0) { 4113cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 4114efe120a0SFrank Holton btrfs_info(root->fs_info, "delayed_refs has NO entry"); 4115acce952bSliubo return ret; 4116acce952bSliubo } 4117acce952bSliubo 4118d7df2c79SJosef Bacik while ((node = rb_first(&delayed_refs->href_root)) != NULL) { 4119d7df2c79SJosef Bacik struct btrfs_delayed_ref_head *head; 4120c6fc2454SQu Wenruo struct btrfs_delayed_ref_node *tmp; 4121e78417d1SJosef Bacik bool pin_bytes = false; 4122acce952bSliubo 4123d7df2c79SJosef Bacik head = rb_entry(node, struct btrfs_delayed_ref_head, 4124d7df2c79SJosef Bacik href_node); 4125b939d1abSJosef Bacik if (!mutex_trylock(&head->mutex)) { 4126d7df2c79SJosef Bacik atomic_inc(&head->node.refs); 412779787eaaSJeff Mahoney spin_unlock(&delayed_refs->lock); 4128b939d1abSJosef Bacik 4129acce952bSliubo mutex_lock(&head->mutex); 4130b939d1abSJosef Bacik mutex_unlock(&head->mutex); 4131d7df2c79SJosef Bacik btrfs_put_delayed_ref(&head->node); 4132e18fca73SJosef Bacik spin_lock(&delayed_refs->lock); 4133b939d1abSJosef Bacik continue; 4134b939d1abSJosef Bacik } 4135d7df2c79SJosef Bacik spin_lock(&head->lock); 4136c6fc2454SQu Wenruo list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list, 4137c6fc2454SQu Wenruo list) { 4138d7df2c79SJosef Bacik ref->in_tree = 0; 4139c6fc2454SQu Wenruo list_del(&ref->list); 4140d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4141d7df2c79SJosef Bacik btrfs_put_delayed_ref(ref); 4142d7df2c79SJosef Bacik } 414354067ae9SJosef Bacik if (head->must_insert_reserved) 4144e78417d1SJosef Bacik pin_bytes = true; 414578a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 4146acce952bSliubo delayed_refs->num_heads--; 4147d7df2c79SJosef Bacik if (head->processing == 0) 4148acce952bSliubo delayed_refs->num_heads_ready--; 4149d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4150d7df2c79SJosef Bacik head->node.in_tree = 0; 4151c46effa6SLiu Bo rb_erase(&head->href_node, &delayed_refs->href_root); 4152d7df2c79SJosef Bacik spin_unlock(&head->lock); 4153acce952bSliubo spin_unlock(&delayed_refs->lock); 4154e78417d1SJosef Bacik mutex_unlock(&head->mutex); 4155acce952bSliubo 4156d7df2c79SJosef Bacik if (pin_bytes) 4157d7df2c79SJosef Bacik btrfs_pin_extent(root, head->node.bytenr, 4158d7df2c79SJosef Bacik head->node.num_bytes, 1); 4159d7df2c79SJosef Bacik btrfs_put_delayed_ref(&head->node); 4160acce952bSliubo cond_resched(); 4161acce952bSliubo spin_lock(&delayed_refs->lock); 4162acce952bSliubo } 4163acce952bSliubo 4164acce952bSliubo spin_unlock(&delayed_refs->lock); 4165acce952bSliubo 4166acce952bSliubo return ret; 4167acce952bSliubo } 4168acce952bSliubo 4169143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 4170acce952bSliubo { 4171acce952bSliubo struct btrfs_inode *btrfs_inode; 4172acce952bSliubo struct list_head splice; 4173acce952bSliubo 4174acce952bSliubo INIT_LIST_HEAD(&splice); 4175acce952bSliubo 4176eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4177eb73c1b7SMiao Xie list_splice_init(&root->delalloc_inodes, &splice); 4178acce952bSliubo 4179acce952bSliubo while (!list_empty(&splice)) { 4180eb73c1b7SMiao Xie btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 4181acce952bSliubo delalloc_inodes); 4182acce952bSliubo 4183acce952bSliubo list_del_init(&btrfs_inode->delalloc_inodes); 4184df0af1a5SMiao Xie clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, 4185df0af1a5SMiao Xie &btrfs_inode->runtime_flags); 4186eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4187acce952bSliubo 4188acce952bSliubo btrfs_invalidate_inodes(btrfs_inode->root); 4189b216cbfbSMiao Xie 4190eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4191acce952bSliubo } 4192acce952bSliubo 4193eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4194eb73c1b7SMiao Xie } 4195eb73c1b7SMiao Xie 4196eb73c1b7SMiao Xie static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 4197eb73c1b7SMiao Xie { 4198eb73c1b7SMiao Xie struct btrfs_root *root; 4199eb73c1b7SMiao Xie struct list_head splice; 4200eb73c1b7SMiao Xie 4201eb73c1b7SMiao Xie INIT_LIST_HEAD(&splice); 4202eb73c1b7SMiao Xie 4203eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4204eb73c1b7SMiao Xie list_splice_init(&fs_info->delalloc_roots, &splice); 4205eb73c1b7SMiao Xie while (!list_empty(&splice)) { 4206eb73c1b7SMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4207eb73c1b7SMiao Xie delalloc_root); 4208eb73c1b7SMiao Xie list_del_init(&root->delalloc_root); 4209eb73c1b7SMiao Xie root = btrfs_grab_fs_root(root); 4210eb73c1b7SMiao Xie BUG_ON(!root); 4211eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4212eb73c1b7SMiao Xie 4213eb73c1b7SMiao Xie btrfs_destroy_delalloc_inodes(root); 4214eb73c1b7SMiao Xie btrfs_put_fs_root(root); 4215eb73c1b7SMiao Xie 4216eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4217eb73c1b7SMiao Xie } 4218eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4219acce952bSliubo } 4220acce952bSliubo 4221acce952bSliubo static int btrfs_destroy_marked_extents(struct btrfs_root *root, 4222acce952bSliubo struct extent_io_tree *dirty_pages, 4223acce952bSliubo int mark) 4224acce952bSliubo { 4225acce952bSliubo int ret; 4226acce952bSliubo struct extent_buffer *eb; 4227acce952bSliubo u64 start = 0; 4228acce952bSliubo u64 end; 4229acce952bSliubo 4230acce952bSliubo while (1) { 4231acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 4232e6138876SJosef Bacik mark, NULL); 4233acce952bSliubo if (ret) 4234acce952bSliubo break; 4235acce952bSliubo 4236acce952bSliubo clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); 4237acce952bSliubo while (start <= end) { 423801d58472SDaniel Dressler eb = btrfs_find_tree_block(root->fs_info, start); 4239707e8a07SDavid Sterba start += root->nodesize; 4240fd8b2b61SJosef Bacik if (!eb) 4241acce952bSliubo continue; 4242fd8b2b61SJosef Bacik wait_on_extent_buffer_writeback(eb); 4243acce952bSliubo 4244fd8b2b61SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 4245fd8b2b61SJosef Bacik &eb->bflags)) 4246fd8b2b61SJosef Bacik clear_extent_buffer_dirty(eb); 4247fd8b2b61SJosef Bacik free_extent_buffer_stale(eb); 4248acce952bSliubo } 4249acce952bSliubo } 4250acce952bSliubo 4251acce952bSliubo return ret; 4252acce952bSliubo } 4253acce952bSliubo 4254acce952bSliubo static int btrfs_destroy_pinned_extent(struct btrfs_root *root, 4255acce952bSliubo struct extent_io_tree *pinned_extents) 4256acce952bSliubo { 4257acce952bSliubo struct extent_io_tree *unpin; 4258acce952bSliubo u64 start; 4259acce952bSliubo u64 end; 4260acce952bSliubo int ret; 4261ed0eaa14SLiu Bo bool loop = true; 4262acce952bSliubo 4263acce952bSliubo unpin = pinned_extents; 4264ed0eaa14SLiu Bo again: 4265acce952bSliubo while (1) { 4266acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 4267e6138876SJosef Bacik EXTENT_DIRTY, NULL); 4268acce952bSliubo if (ret) 4269acce952bSliubo break; 4270acce952bSliubo 4271acce952bSliubo clear_extent_dirty(unpin, start, end, GFP_NOFS); 4272acce952bSliubo btrfs_error_unpin_extent_range(root, start, end); 4273acce952bSliubo cond_resched(); 4274acce952bSliubo } 4275acce952bSliubo 4276ed0eaa14SLiu Bo if (loop) { 4277ed0eaa14SLiu Bo if (unpin == &root->fs_info->freed_extents[0]) 4278ed0eaa14SLiu Bo unpin = &root->fs_info->freed_extents[1]; 4279ed0eaa14SLiu Bo else 4280ed0eaa14SLiu Bo unpin = &root->fs_info->freed_extents[0]; 4281ed0eaa14SLiu Bo loop = false; 4282ed0eaa14SLiu Bo goto again; 4283ed0eaa14SLiu Bo } 4284ed0eaa14SLiu Bo 4285acce952bSliubo return 0; 4286acce952bSliubo } 4287acce952bSliubo 428850d9aa99SJosef Bacik static void btrfs_free_pending_ordered(struct btrfs_transaction *cur_trans, 428950d9aa99SJosef Bacik struct btrfs_fs_info *fs_info) 429050d9aa99SJosef Bacik { 429150d9aa99SJosef Bacik struct btrfs_ordered_extent *ordered; 429250d9aa99SJosef Bacik 429350d9aa99SJosef Bacik spin_lock(&fs_info->trans_lock); 429450d9aa99SJosef Bacik while (!list_empty(&cur_trans->pending_ordered)) { 429550d9aa99SJosef Bacik ordered = list_first_entry(&cur_trans->pending_ordered, 429650d9aa99SJosef Bacik struct btrfs_ordered_extent, 429750d9aa99SJosef Bacik trans_list); 429850d9aa99SJosef Bacik list_del_init(&ordered->trans_list); 429950d9aa99SJosef Bacik spin_unlock(&fs_info->trans_lock); 430050d9aa99SJosef Bacik 430150d9aa99SJosef Bacik btrfs_put_ordered_extent(ordered); 430250d9aa99SJosef Bacik spin_lock(&fs_info->trans_lock); 430350d9aa99SJosef Bacik } 430450d9aa99SJosef Bacik spin_unlock(&fs_info->trans_lock); 430550d9aa99SJosef Bacik } 430650d9aa99SJosef Bacik 430749b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 430849b25e05SJeff Mahoney struct btrfs_root *root) 430949b25e05SJeff Mahoney { 431049b25e05SJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, root); 431149b25e05SJeff Mahoney 43124a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 431349b25e05SJeff Mahoney wake_up(&root->fs_info->transaction_blocked_wait); 431449b25e05SJeff Mahoney 43154a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 431649b25e05SJeff Mahoney wake_up(&root->fs_info->transaction_wait); 431749b25e05SJeff Mahoney 431850d9aa99SJosef Bacik btrfs_free_pending_ordered(cur_trans, root->fs_info); 431967cde344SMiao Xie btrfs_destroy_delayed_inodes(root); 432067cde344SMiao Xie btrfs_assert_delayed_root_empty(root); 432167cde344SMiao Xie 432249b25e05SJeff Mahoney btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages, 432349b25e05SJeff Mahoney EXTENT_DIRTY); 43246e841e32SLiu Bo btrfs_destroy_pinned_extent(root, 43256e841e32SLiu Bo root->fs_info->pinned_extents); 432649b25e05SJeff Mahoney 43274a9d8bdeSMiao Xie cur_trans->state =TRANS_STATE_COMPLETED; 43284a9d8bdeSMiao Xie wake_up(&cur_trans->commit_wait); 43294a9d8bdeSMiao Xie 433049b25e05SJeff Mahoney /* 433149b25e05SJeff Mahoney memset(cur_trans, 0, sizeof(*cur_trans)); 433249b25e05SJeff Mahoney kmem_cache_free(btrfs_transaction_cachep, cur_trans); 433349b25e05SJeff Mahoney */ 433449b25e05SJeff Mahoney } 433549b25e05SJeff Mahoney 433648a3b636SEric Sandeen static int btrfs_cleanup_transaction(struct btrfs_root *root) 4337acce952bSliubo { 4338acce952bSliubo struct btrfs_transaction *t; 4339acce952bSliubo 4340acce952bSliubo mutex_lock(&root->fs_info->transaction_kthread_mutex); 4341acce952bSliubo 4342a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 4343724e2315SJosef Bacik while (!list_empty(&root->fs_info->trans_list)) { 4344724e2315SJosef Bacik t = list_first_entry(&root->fs_info->trans_list, 4345724e2315SJosef Bacik struct btrfs_transaction, list); 4346724e2315SJosef Bacik if (t->state >= TRANS_STATE_COMMIT_START) { 4347724e2315SJosef Bacik atomic_inc(&t->use_count); 4348724e2315SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 4349724e2315SJosef Bacik btrfs_wait_for_commit(root, t->transid); 4350724e2315SJosef Bacik btrfs_put_transaction(t); 4351724e2315SJosef Bacik spin_lock(&root->fs_info->trans_lock); 4352724e2315SJosef Bacik continue; 4353724e2315SJosef Bacik } 4354724e2315SJosef Bacik if (t == root->fs_info->running_transaction) { 4355724e2315SJosef Bacik t->state = TRANS_STATE_COMMIT_DOING; 4356724e2315SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 4357724e2315SJosef Bacik /* 4358724e2315SJosef Bacik * We wait for 0 num_writers since we don't hold a trans 4359724e2315SJosef Bacik * handle open currently for this transaction. 4360724e2315SJosef Bacik */ 4361724e2315SJosef Bacik wait_event(t->writer_wait, 4362724e2315SJosef Bacik atomic_read(&t->num_writers) == 0); 4363724e2315SJosef Bacik } else { 4364724e2315SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 4365724e2315SJosef Bacik } 4366724e2315SJosef Bacik btrfs_cleanup_one_transaction(t, root); 4367724e2315SJosef Bacik 4368724e2315SJosef Bacik spin_lock(&root->fs_info->trans_lock); 4369724e2315SJosef Bacik if (t == root->fs_info->running_transaction) 4370ac673879SMiao Xie root->fs_info->running_transaction = NULL; 4371724e2315SJosef Bacik list_del_init(&t->list); 4372a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 4373a4abeea4SJosef Bacik 4374724e2315SJosef Bacik btrfs_put_transaction(t); 4375724e2315SJosef Bacik trace_btrfs_transaction_commit(root); 4376724e2315SJosef Bacik spin_lock(&root->fs_info->trans_lock); 4377724e2315SJosef Bacik } 4378724e2315SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 4379199c2a9cSMiao Xie btrfs_destroy_all_ordered_extents(root->fs_info); 438067cde344SMiao Xie btrfs_destroy_delayed_inodes(root); 438167cde344SMiao Xie btrfs_assert_delayed_root_empty(root); 4382724e2315SJosef Bacik btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents); 4383eb73c1b7SMiao Xie btrfs_destroy_all_delalloc_inodes(root->fs_info); 4384acce952bSliubo mutex_unlock(&root->fs_info->transaction_kthread_mutex); 4385acce952bSliubo 4386acce952bSliubo return 0; 4387acce952bSliubo } 4388acce952bSliubo 4389e8c9f186SDavid Sterba static const struct extent_io_ops btree_extent_io_ops = { 4390ce9adaa5SChris Mason .readpage_end_io_hook = btree_readpage_end_io_hook, 43914bb31e92SArne Jansen .readpage_io_failed_hook = btree_io_failed_hook, 43920b86a832SChris Mason .submit_bio_hook = btree_submit_bio_hook, 4393239b14b3SChris Mason /* note we're sharing with inode.c for the merge bio hook */ 4394239b14b3SChris Mason .merge_bio_hook = btrfs_merge_bio_hook, 43950da5468fSChris Mason }; 4396