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 7064246a0b6SChristoph Hellwig static void end_workqueue_bio(struct bio *bio) 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; 7144246a0b6SChristoph Hellwig end_io_wq->error = bio->bi_error; 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) { 8114246a0b6SChristoph Hellwig async->bio->bi_error = async->error; 8124246a0b6SChristoph Hellwig bio_endio(async->bio); 81379787eaaSJeff Mahoney return; 81479787eaaSJeff Mahoney } 81579787eaaSJeff Mahoney 8164a69a410SChris Mason async->submit_bio_done(async->inode, async->rw, async->bio, 817eaf25d93SChris Mason async->mirror_num, async->bio_flags, 818eaf25d93SChris Mason async->bio_offset); 8194a69a410SChris Mason } 8204a69a410SChris Mason 8214a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 8224a69a410SChris Mason { 8234a69a410SChris Mason struct async_submit_bio *async; 8244a69a410SChris Mason 8254a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 8268b712842SChris Mason kfree(async); 8278b712842SChris Mason } 8288b712842SChris Mason 82944b8bd7eSChris Mason int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode, 83044b8bd7eSChris Mason int rw, struct bio *bio, int mirror_num, 831c8b97818SChris Mason unsigned long bio_flags, 832eaf25d93SChris Mason u64 bio_offset, 8334a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_start, 8344a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_done) 83544b8bd7eSChris Mason { 83644b8bd7eSChris Mason struct async_submit_bio *async; 83744b8bd7eSChris Mason 83844b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 83944b8bd7eSChris Mason if (!async) 84044b8bd7eSChris Mason return -ENOMEM; 84144b8bd7eSChris Mason 84244b8bd7eSChris Mason async->inode = inode; 84344b8bd7eSChris Mason async->rw = rw; 84444b8bd7eSChris Mason async->bio = bio; 84544b8bd7eSChris Mason async->mirror_num = mirror_num; 8464a69a410SChris Mason async->submit_bio_start = submit_bio_start; 8474a69a410SChris Mason async->submit_bio_done = submit_bio_done; 8484a69a410SChris Mason 8499e0af237SLiu Bo btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start, 8505cdc7ad3SQu Wenruo run_one_async_done, run_one_async_free); 8514a69a410SChris Mason 852c8b97818SChris Mason async->bio_flags = bio_flags; 853eaf25d93SChris Mason async->bio_offset = bio_offset; 8548c8bee1dSChris Mason 85579787eaaSJeff Mahoney async->error = 0; 85679787eaaSJeff Mahoney 857cb03c743SChris Mason atomic_inc(&fs_info->nr_async_submits); 858d313d7a3SChris Mason 8597b6d91daSChristoph Hellwig if (rw & REQ_SYNC) 8605cdc7ad3SQu Wenruo btrfs_set_work_high_priority(&async->work); 861d313d7a3SChris Mason 8625cdc7ad3SQu Wenruo btrfs_queue_work(fs_info->workers, &async->work); 8639473f16cSChris Mason 864771ed689SChris Mason while (atomic_read(&fs_info->async_submit_draining) && 865771ed689SChris Mason atomic_read(&fs_info->nr_async_submits)) { 866771ed689SChris Mason wait_event(fs_info->async_submit_wait, 867771ed689SChris Mason (atomic_read(&fs_info->nr_async_submits) == 0)); 868771ed689SChris Mason } 869771ed689SChris Mason 87044b8bd7eSChris Mason return 0; 87144b8bd7eSChris Mason } 87244b8bd7eSChris Mason 873ce3ed71aSChris Mason static int btree_csum_one_bio(struct bio *bio) 874ce3ed71aSChris Mason { 8752c30c71bSKent Overstreet struct bio_vec *bvec; 876ce3ed71aSChris Mason struct btrfs_root *root; 8772c30c71bSKent Overstreet int i, ret = 0; 878ce3ed71aSChris Mason 8792c30c71bSKent Overstreet bio_for_each_segment_all(bvec, bio, i) { 880ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 88101d58472SDaniel Dressler ret = csum_dirty_buffer(root->fs_info, bvec->bv_page); 88279787eaaSJeff Mahoney if (ret) 88379787eaaSJeff Mahoney break; 884ce3ed71aSChris Mason } 8852c30c71bSKent Overstreet 88679787eaaSJeff Mahoney return ret; 887ce3ed71aSChris Mason } 888ce3ed71aSChris Mason 8894a69a410SChris Mason static int __btree_submit_bio_start(struct inode *inode, int rw, 8904a69a410SChris Mason struct bio *bio, int mirror_num, 891eaf25d93SChris Mason unsigned long bio_flags, 892eaf25d93SChris Mason u64 bio_offset) 89322c59948SChris Mason { 8948b712842SChris Mason /* 8958b712842SChris Mason * when we're called for a write, we're already in the async 8965443be45SChris Mason * submission context. Just jump into btrfs_map_bio 8978b712842SChris Mason */ 89879787eaaSJeff Mahoney return btree_csum_one_bio(bio); 89922c59948SChris Mason } 90022c59948SChris Mason 9014a69a410SChris Mason static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio, 902eaf25d93SChris Mason int mirror_num, unsigned long bio_flags, 903eaf25d93SChris Mason u64 bio_offset) 9044a69a410SChris Mason { 90561891923SStefan Behrens int ret; 90661891923SStefan Behrens 9078b712842SChris Mason /* 9084a69a410SChris Mason * when we're called for a write, we're already in the async 9094a69a410SChris Mason * submission context. Just jump into btrfs_map_bio 9108b712842SChris Mason */ 91161891923SStefan Behrens ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1); 9124246a0b6SChristoph Hellwig if (ret) { 9134246a0b6SChristoph Hellwig bio->bi_error = ret; 9144246a0b6SChristoph Hellwig bio_endio(bio); 9154246a0b6SChristoph Hellwig } 91661891923SStefan Behrens return ret; 9170b86a832SChris Mason } 9180b86a832SChris Mason 919de0022b9SJosef Bacik static int check_async_write(struct inode *inode, unsigned long bio_flags) 920de0022b9SJosef Bacik { 921de0022b9SJosef Bacik if (bio_flags & EXTENT_BIO_TREE_LOG) 922de0022b9SJosef Bacik return 0; 923de0022b9SJosef Bacik #ifdef CONFIG_X86 924de0022b9SJosef Bacik if (cpu_has_xmm4_2) 925de0022b9SJosef Bacik return 0; 926de0022b9SJosef Bacik #endif 927de0022b9SJosef Bacik return 1; 928de0022b9SJosef Bacik } 929de0022b9SJosef Bacik 93044b8bd7eSChris Mason static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, 931eaf25d93SChris Mason int mirror_num, unsigned long bio_flags, 932eaf25d93SChris Mason u64 bio_offset) 93344b8bd7eSChris Mason { 934de0022b9SJosef Bacik int async = check_async_write(inode, bio_flags); 9354a69a410SChris Mason int ret; 936cad321adSChris Mason 9377b6d91daSChristoph Hellwig if (!(rw & REQ_WRITE)) { 938cad321adSChris Mason /* 939cad321adSChris Mason * called for a read, do the setup so that checksum validation 940cad321adSChris Mason * can happen in the async kernel threads 941cad321adSChris Mason */ 942f3f266abSChris Mason ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info, 943bfebd8b5SDavid Sterba bio, BTRFS_WQ_ENDIO_METADATA); 9441d4284bdSChris Mason if (ret) 94561891923SStefan Behrens goto out_w_error; 94661891923SStefan Behrens ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, 9476f3577bdSChris Mason mirror_num, 0); 948de0022b9SJosef Bacik } else if (!async) { 949de0022b9SJosef Bacik ret = btree_csum_one_bio(bio); 950de0022b9SJosef Bacik if (ret) 95161891923SStefan Behrens goto out_w_error; 95261891923SStefan Behrens ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, 953de0022b9SJosef Bacik mirror_num, 0); 95461891923SStefan Behrens } else { 955cad321adSChris Mason /* 95661891923SStefan Behrens * kthread helpers are used to submit writes so that 95761891923SStefan Behrens * checksumming can happen in parallel across all CPUs 958cad321adSChris Mason */ 95961891923SStefan Behrens ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, 960c8b97818SChris Mason inode, rw, bio, mirror_num, 0, 961eaf25d93SChris Mason bio_offset, 9624a69a410SChris Mason __btree_submit_bio_start, 9634a69a410SChris Mason __btree_submit_bio_done); 96444b8bd7eSChris Mason } 96544b8bd7eSChris Mason 9664246a0b6SChristoph Hellwig if (ret) 9674246a0b6SChristoph Hellwig goto out_w_error; 9684246a0b6SChristoph Hellwig return 0; 9694246a0b6SChristoph Hellwig 97061891923SStefan Behrens out_w_error: 9714246a0b6SChristoph Hellwig bio->bi_error = ret; 9724246a0b6SChristoph Hellwig bio_endio(bio); 97361891923SStefan Behrens return ret; 97461891923SStefan Behrens } 97561891923SStefan Behrens 9763dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 977784b4e29SChris Mason static int btree_migratepage(struct address_space *mapping, 978a6bc32b8SMel Gorman struct page *newpage, struct page *page, 979a6bc32b8SMel Gorman enum migrate_mode mode) 980784b4e29SChris Mason { 981784b4e29SChris Mason /* 982784b4e29SChris Mason * we can't safely write a btree page from here, 983784b4e29SChris Mason * we haven't done the locking hook 984784b4e29SChris Mason */ 985784b4e29SChris Mason if (PageDirty(page)) 986784b4e29SChris Mason return -EAGAIN; 987784b4e29SChris Mason /* 988784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 989784b4e29SChris Mason * We must have no buffers or drop them. 990784b4e29SChris Mason */ 991784b4e29SChris Mason if (page_has_private(page) && 992784b4e29SChris Mason !try_to_release_page(page, GFP_KERNEL)) 993784b4e29SChris Mason return -EAGAIN; 994a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 995784b4e29SChris Mason } 9963dd1462eSJan Beulich #endif 997784b4e29SChris Mason 9980da5468fSChris Mason 9990da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 10000da5468fSChris Mason struct writeback_control *wbc) 10010da5468fSChris Mason { 1002e2d84521SMiao Xie struct btrfs_fs_info *fs_info; 1003e2d84521SMiao Xie int ret; 1004e2d84521SMiao Xie 1005d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 1006448d640bSChris Mason 1007448d640bSChris Mason if (wbc->for_kupdate) 1008448d640bSChris Mason return 0; 1009448d640bSChris Mason 1010e2d84521SMiao Xie fs_info = BTRFS_I(mapping->host)->root->fs_info; 1011b9473439SChris Mason /* this is a bit racy, but that's ok */ 1012e2d84521SMiao Xie ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes, 1013e2d84521SMiao Xie BTRFS_DIRTY_METADATA_THRESH); 1014e2d84521SMiao Xie if (ret < 0) 1015793955bcSChris Mason return 0; 1016793955bcSChris Mason } 10170b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 10180da5468fSChris Mason } 10190da5468fSChris Mason 1020b2950863SChristoph Hellwig static int btree_readpage(struct file *file, struct page *page) 10215f39d397SChris Mason { 1022d1310b2eSChris Mason struct extent_io_tree *tree; 1023d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 10248ddc7d9cSJan Schmidt return extent_read_full_page(tree, page, btree_get_extent, 0); 10255f39d397SChris Mason } 10265f39d397SChris Mason 102770dec807SChris Mason static int btree_releasepage(struct page *page, gfp_t gfp_flags) 10285f39d397SChris Mason { 102998509cfcSChris Mason if (PageWriteback(page) || PageDirty(page)) 103098509cfcSChris Mason return 0; 10310c4e538bSDavid Sterba 1032f7a52a40SDavid Sterba return try_release_extent_buffer(page); 1033d98237b3SChris Mason } 1034d98237b3SChris Mason 1035d47992f8SLukas Czerner static void btree_invalidatepage(struct page *page, unsigned int offset, 1036d47992f8SLukas Czerner unsigned int length) 1037d98237b3SChris Mason { 1038d1310b2eSChris Mason struct extent_io_tree *tree; 1039d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 10405f39d397SChris Mason extent_invalidatepage(tree, page, offset); 10415f39d397SChris Mason btree_releasepage(page, GFP_NOFS); 10429ad6b7bcSChris Mason if (PagePrivate(page)) { 1043efe120a0SFrank Holton btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info, 1044efe120a0SFrank Holton "page private not zero on page %llu", 1045efe120a0SFrank Holton (unsigned long long)page_offset(page)); 10469ad6b7bcSChris Mason ClearPagePrivate(page); 10479ad6b7bcSChris Mason set_page_private(page, 0); 10489ad6b7bcSChris Mason page_cache_release(page); 10499ad6b7bcSChris Mason } 1050d98237b3SChris Mason } 1051d98237b3SChris Mason 10520b32f4bbSJosef Bacik static int btree_set_page_dirty(struct page *page) 10530b32f4bbSJosef Bacik { 1054bb146eb2SJosef Bacik #ifdef DEBUG 10550b32f4bbSJosef Bacik struct extent_buffer *eb; 10560b32f4bbSJosef Bacik 10570b32f4bbSJosef Bacik BUG_ON(!PagePrivate(page)); 10580b32f4bbSJosef Bacik eb = (struct extent_buffer *)page->private; 10590b32f4bbSJosef Bacik BUG_ON(!eb); 10600b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 10610b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 10620b32f4bbSJosef Bacik btrfs_assert_tree_locked(eb); 1063bb146eb2SJosef Bacik #endif 10640b32f4bbSJosef Bacik return __set_page_dirty_nobuffers(page); 10650b32f4bbSJosef Bacik } 10660b32f4bbSJosef Bacik 10677f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 1068d98237b3SChris Mason .readpage = btree_readpage, 10690da5468fSChris Mason .writepages = btree_writepages, 10705f39d397SChris Mason .releasepage = btree_releasepage, 10715f39d397SChris Mason .invalidatepage = btree_invalidatepage, 10725a92bc88SChris Mason #ifdef CONFIG_MIGRATION 1073784b4e29SChris Mason .migratepage = btree_migratepage, 10745a92bc88SChris Mason #endif 10750b32f4bbSJosef Bacik .set_page_dirty = btree_set_page_dirty, 1076d98237b3SChris Mason }; 1077123abc88SChris Mason 1078d3e46feaSDavid Sterba void readahead_tree_block(struct btrfs_root *root, u64 bytenr) 1079090d1875SChris Mason { 10805f39d397SChris Mason struct extent_buffer *buf = NULL; 10815f39d397SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 1082090d1875SChris Mason 1083a83fffb7SDavid Sterba buf = btrfs_find_create_tree_block(root, bytenr); 10845f39d397SChris Mason if (!buf) 10856197d86eSDavid Sterba return; 1086d1310b2eSChris Mason read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, 1087bb82ab88SArne Jansen buf, 0, WAIT_NONE, btree_get_extent, 0); 10885f39d397SChris Mason free_extent_buffer(buf); 1089090d1875SChris Mason } 1090090d1875SChris Mason 1091c0dcaa4dSDavid Sterba int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, 1092ab0fff03SArne Jansen int mirror_num, struct extent_buffer **eb) 1093ab0fff03SArne Jansen { 1094ab0fff03SArne Jansen struct extent_buffer *buf = NULL; 1095ab0fff03SArne Jansen struct inode *btree_inode = root->fs_info->btree_inode; 1096ab0fff03SArne Jansen struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree; 1097ab0fff03SArne Jansen int ret; 1098ab0fff03SArne Jansen 1099a83fffb7SDavid Sterba buf = btrfs_find_create_tree_block(root, bytenr); 1100ab0fff03SArne Jansen if (!buf) 1101ab0fff03SArne Jansen return 0; 1102ab0fff03SArne Jansen 1103ab0fff03SArne Jansen set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags); 1104ab0fff03SArne Jansen 1105ab0fff03SArne Jansen ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK, 1106ab0fff03SArne Jansen btree_get_extent, mirror_num); 1107ab0fff03SArne Jansen if (ret) { 1108ab0fff03SArne Jansen free_extent_buffer(buf); 1109ab0fff03SArne Jansen return ret; 1110ab0fff03SArne Jansen } 1111ab0fff03SArne Jansen 1112ab0fff03SArne Jansen if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) { 1113ab0fff03SArne Jansen free_extent_buffer(buf); 1114ab0fff03SArne Jansen return -EIO; 11150b32f4bbSJosef Bacik } else if (extent_buffer_uptodate(buf)) { 1116ab0fff03SArne Jansen *eb = buf; 1117ab0fff03SArne Jansen } else { 1118ab0fff03SArne Jansen free_extent_buffer(buf); 1119ab0fff03SArne Jansen } 1120ab0fff03SArne Jansen return 0; 1121ab0fff03SArne Jansen } 1122ab0fff03SArne Jansen 112301d58472SDaniel Dressler struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info, 11240308af44SDavid Sterba u64 bytenr) 11250999df54SChris Mason { 112601d58472SDaniel Dressler return find_extent_buffer(fs_info, bytenr); 11270999df54SChris Mason } 11280999df54SChris Mason 11290999df54SChris Mason struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root, 1130a83fffb7SDavid Sterba u64 bytenr) 11310999df54SChris Mason { 1132fccb84c9SDavid Sterba if (btrfs_test_is_dummy_root(root)) 1133ce3e6984SDavid Sterba return alloc_test_extent_buffer(root->fs_info, bytenr); 1134ce3e6984SDavid Sterba return alloc_extent_buffer(root->fs_info, bytenr); 11350999df54SChris Mason } 11360999df54SChris Mason 11370999df54SChris Mason 1138e02119d5SChris Mason int btrfs_write_tree_block(struct extent_buffer *buf) 1139e02119d5SChris Mason { 1140727011e0SChris Mason return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start, 11418aa38c31SChristoph Hellwig buf->start + buf->len - 1); 1142e02119d5SChris Mason } 1143e02119d5SChris Mason 1144e02119d5SChris Mason int btrfs_wait_tree_block_writeback(struct extent_buffer *buf) 1145e02119d5SChris Mason { 1146727011e0SChris Mason return filemap_fdatawait_range(buf->pages[0]->mapping, 11478aa38c31SChristoph Hellwig buf->start, buf->start + buf->len - 1); 1148e02119d5SChris Mason } 1149e02119d5SChris Mason 1150db94535dSChris Mason struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr, 1151ce86cd59SDavid Sterba u64 parent_transid) 1152e20d96d6SChris Mason { 11535f39d397SChris Mason struct extent_buffer *buf = NULL; 115419c00ddcSChris Mason int ret; 115519c00ddcSChris Mason 1156a83fffb7SDavid Sterba buf = btrfs_find_create_tree_block(root, bytenr); 11575f39d397SChris Mason if (!buf) 115864c043deSLiu Bo return ERR_PTR(-ENOMEM); 1159e4204dedSChris Mason 1160ca7a79adSChris Mason ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 11610f0fe8f7SFilipe David Borba Manana if (ret) { 11620f0fe8f7SFilipe David Borba Manana free_extent_buffer(buf); 116364c043deSLiu Bo return ERR_PTR(ret); 11640f0fe8f7SFilipe David Borba Manana } 11655f39d397SChris Mason return buf; 1166ce9adaa5SChris Mason 1167eb60ceacSChris Mason } 1168eb60ceacSChris Mason 116901d58472SDaniel Dressler void clean_tree_block(struct btrfs_trans_handle *trans, 117001d58472SDaniel Dressler struct btrfs_fs_info *fs_info, 11715f39d397SChris Mason struct extent_buffer *buf) 1172ed2ff2cbSChris Mason { 117355c69072SChris Mason if (btrfs_header_generation(buf) == 1174e2d84521SMiao Xie fs_info->running_transaction->transid) { 1175b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1176b4ce94deSChris Mason 1177b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1178e2d84521SMiao Xie __percpu_counter_add(&fs_info->dirty_metadata_bytes, 1179e2d84521SMiao Xie -buf->len, 1180e2d84521SMiao Xie fs_info->dirty_metadata_batch); 1181b9473439SChris Mason /* ugh, clear_extent_buffer_dirty needs to lock the page */ 1182b9473439SChris Mason btrfs_set_lock_blocking(buf); 11830b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 1184925baeddSChris Mason } 11855f39d397SChris Mason } 1186ed7b63ebSJosef Bacik } 11875f39d397SChris Mason 11888257b2dcSMiao Xie static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void) 11898257b2dcSMiao Xie { 11908257b2dcSMiao Xie struct btrfs_subvolume_writers *writers; 11918257b2dcSMiao Xie int ret; 11928257b2dcSMiao Xie 11938257b2dcSMiao Xie writers = kmalloc(sizeof(*writers), GFP_NOFS); 11948257b2dcSMiao Xie if (!writers) 11958257b2dcSMiao Xie return ERR_PTR(-ENOMEM); 11968257b2dcSMiao Xie 1197908c7f19STejun Heo ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL); 11988257b2dcSMiao Xie if (ret < 0) { 11998257b2dcSMiao Xie kfree(writers); 12008257b2dcSMiao Xie return ERR_PTR(ret); 12018257b2dcSMiao Xie } 12028257b2dcSMiao Xie 12038257b2dcSMiao Xie init_waitqueue_head(&writers->wait); 12048257b2dcSMiao Xie return writers; 12058257b2dcSMiao Xie } 12068257b2dcSMiao Xie 12078257b2dcSMiao Xie static void 12088257b2dcSMiao Xie btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers) 12098257b2dcSMiao Xie { 12108257b2dcSMiao Xie percpu_counter_destroy(&writers->counter); 12118257b2dcSMiao Xie kfree(writers); 12128257b2dcSMiao Xie } 12138257b2dcSMiao Xie 1214707e8a07SDavid Sterba static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize, 1215707e8a07SDavid Sterba struct btrfs_root *root, struct btrfs_fs_info *fs_info, 1216e20d96d6SChris Mason u64 objectid) 1217d97e63b6SChris Mason { 1218cfaa7295SChris Mason root->node = NULL; 1219a28ec197SChris Mason root->commit_root = NULL; 1220db94535dSChris Mason root->sectorsize = sectorsize; 1221db94535dSChris Mason root->nodesize = nodesize; 122287ee04ebSChris Mason root->stripesize = stripesize; 122327cdeb70SMiao Xie root->state = 0; 1224d68fc57bSYan, Zheng root->orphan_cleanup_state = 0; 12250b86a832SChris Mason 12260f7d52f4SChris Mason root->objectid = objectid; 12270f7d52f4SChris Mason root->last_trans = 0; 122813a8a7c8SYan, Zheng root->highest_objectid = 0; 1229eb73c1b7SMiao Xie root->nr_delalloc_inodes = 0; 1230199c2a9cSMiao Xie root->nr_ordered_extents = 0; 123158176a96SJosef Bacik root->name = NULL; 12326bef4d31SEric Paris root->inode_tree = RB_ROOT; 123316cdcec7SMiao Xie INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1234f0486c68SYan, Zheng root->block_rsv = NULL; 1235d68fc57bSYan, Zheng root->orphan_block_rsv = NULL; 12360b86a832SChris Mason 12370b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 12385d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 1239eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_inodes); 1240eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_root); 1241199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_extents); 1242199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_root); 12432ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 12442ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 1245d68fc57bSYan, Zheng spin_lock_init(&root->orphan_lock); 12465d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1247eb73c1b7SMiao Xie spin_lock_init(&root->delalloc_lock); 1248199c2a9cSMiao Xie spin_lock_init(&root->ordered_extent_lock); 1249f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 12502ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 12512ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 1252a2135011SChris Mason mutex_init(&root->objectid_mutex); 1253e02119d5SChris Mason mutex_init(&root->log_mutex); 125431f3d255SMiao Xie mutex_init(&root->ordered_extent_mutex); 1255573bfb72SMiao Xie mutex_init(&root->delalloc_mutex); 12567237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 12577237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 12587237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 12598b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[0]); 12608b050d35SMiao Xie INIT_LIST_HEAD(&root->log_ctxs[1]); 12617237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 12627237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 12637237f183SYan Zheng atomic_set(&root->log_writers, 0); 12642ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 12658a35d95fSJosef Bacik atomic_set(&root->orphan_inodes, 0); 1266b0feb9d9SMiao Xie atomic_set(&root->refs, 1); 12678257b2dcSMiao Xie atomic_set(&root->will_be_snapshoted, 0); 12687237f183SYan Zheng root->log_transid = 0; 1269d1433debSMiao Xie root->log_transid_committed = -1; 1270257c62e1SChris Mason root->last_log_commit = 0; 127106ea65a3SJosef Bacik if (fs_info) 1272d0c803c4SChris Mason extent_io_tree_init(&root->dirty_log_pages, 1273f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 1274017e5369SChris Mason 12753768f368SChris Mason memset(&root->root_key, 0, sizeof(root->root_key)); 12763768f368SChris Mason memset(&root->root_item, 0, sizeof(root->root_item)); 12776702ed49SChris Mason memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 127806ea65a3SJosef Bacik if (fs_info) 12793f157a2fSChris Mason root->defrag_trans_start = fs_info->generation; 128006ea65a3SJosef Bacik else 128106ea65a3SJosef Bacik root->defrag_trans_start = 0; 12824d775673SChris Mason root->root_key.objectid = objectid; 12830ee5dc67SAl Viro root->anon_dev = 0; 12848ea05e3aSAlexander Block 12855f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 12863768f368SChris Mason } 12873768f368SChris Mason 1288f84a8bd6SAl Viro static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info) 12896f07e42eSAl Viro { 12906f07e42eSAl Viro struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS); 12916f07e42eSAl Viro if (root) 12926f07e42eSAl Viro root->fs_info = fs_info; 12936f07e42eSAl Viro return root; 12946f07e42eSAl Viro } 12956f07e42eSAl Viro 129606ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 129706ea65a3SJosef Bacik /* Should only be used by the testing infrastructure */ 129806ea65a3SJosef Bacik struct btrfs_root *btrfs_alloc_dummy_root(void) 129906ea65a3SJosef Bacik { 130006ea65a3SJosef Bacik struct btrfs_root *root; 130106ea65a3SJosef Bacik 130206ea65a3SJosef Bacik root = btrfs_alloc_root(NULL); 130306ea65a3SJosef Bacik if (!root) 130406ea65a3SJosef Bacik return ERR_PTR(-ENOMEM); 1305707e8a07SDavid Sterba __setup_root(4096, 4096, 4096, root, NULL, 1); 130627cdeb70SMiao Xie set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state); 1307faa2dbf0SJosef Bacik root->alloc_bytenr = 0; 130806ea65a3SJosef Bacik 130906ea65a3SJosef Bacik return root; 131006ea65a3SJosef Bacik } 131106ea65a3SJosef Bacik #endif 131206ea65a3SJosef Bacik 131320897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 131420897f5cSArne Jansen struct btrfs_fs_info *fs_info, 131520897f5cSArne Jansen u64 objectid) 131620897f5cSArne Jansen { 131720897f5cSArne Jansen struct extent_buffer *leaf; 131820897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 131920897f5cSArne Jansen struct btrfs_root *root; 132020897f5cSArne Jansen struct btrfs_key key; 132120897f5cSArne Jansen int ret = 0; 13226463fe58SStefan Behrens uuid_le uuid; 132320897f5cSArne Jansen 132420897f5cSArne Jansen root = btrfs_alloc_root(fs_info); 132520897f5cSArne Jansen if (!root) 132620897f5cSArne Jansen return ERR_PTR(-ENOMEM); 132720897f5cSArne Jansen 1328707e8a07SDavid Sterba __setup_root(tree_root->nodesize, tree_root->sectorsize, 1329707e8a07SDavid Sterba tree_root->stripesize, root, fs_info, objectid); 133020897f5cSArne Jansen root->root_key.objectid = objectid; 133120897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 133220897f5cSArne Jansen root->root_key.offset = 0; 133320897f5cSArne Jansen 13344d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0); 133520897f5cSArne Jansen if (IS_ERR(leaf)) { 133620897f5cSArne Jansen ret = PTR_ERR(leaf); 13371dd05682STsutomu Itoh leaf = NULL; 133820897f5cSArne Jansen goto fail; 133920897f5cSArne Jansen } 134020897f5cSArne Jansen 134120897f5cSArne Jansen memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 134220897f5cSArne Jansen btrfs_set_header_bytenr(leaf, leaf->start); 134320897f5cSArne Jansen btrfs_set_header_generation(leaf, trans->transid); 134420897f5cSArne Jansen btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 134520897f5cSArne Jansen btrfs_set_header_owner(leaf, objectid); 134620897f5cSArne Jansen root->node = leaf; 134720897f5cSArne Jansen 13480a4e5586SRoss Kirk write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(), 134920897f5cSArne Jansen BTRFS_FSID_SIZE); 135020897f5cSArne Jansen write_extent_buffer(leaf, fs_info->chunk_tree_uuid, 1351b308bc2fSGeert Uytterhoeven btrfs_header_chunk_tree_uuid(leaf), 135220897f5cSArne Jansen BTRFS_UUID_SIZE); 135320897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 135420897f5cSArne Jansen 135520897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 135627cdeb70SMiao Xie set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 135720897f5cSArne Jansen 135820897f5cSArne Jansen root->root_item.flags = 0; 135920897f5cSArne Jansen root->root_item.byte_limit = 0; 136020897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 136120897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 136220897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 136320897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 136420897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 136520897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 136620897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 13676463fe58SStefan Behrens uuid_le_gen(&uuid); 13686463fe58SStefan Behrens memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE); 136920897f5cSArne Jansen root->root_item.drop_level = 0; 137020897f5cSArne Jansen 137120897f5cSArne Jansen key.objectid = objectid; 137220897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 137320897f5cSArne Jansen key.offset = 0; 137420897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 137520897f5cSArne Jansen if (ret) 137620897f5cSArne Jansen goto fail; 137720897f5cSArne Jansen 137820897f5cSArne Jansen btrfs_tree_unlock(leaf); 137920897f5cSArne Jansen 138020897f5cSArne Jansen return root; 13811dd05682STsutomu Itoh 13821dd05682STsutomu Itoh fail: 13831dd05682STsutomu Itoh if (leaf) { 13841dd05682STsutomu Itoh btrfs_tree_unlock(leaf); 138559885b39STsutomu Itoh free_extent_buffer(root->commit_root); 13861dd05682STsutomu Itoh free_extent_buffer(leaf); 13871dd05682STsutomu Itoh } 13881dd05682STsutomu Itoh kfree(root); 13891dd05682STsutomu Itoh 13901dd05682STsutomu Itoh return ERR_PTR(ret); 139120897f5cSArne Jansen } 139220897f5cSArne Jansen 13937237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1394e02119d5SChris Mason struct btrfs_fs_info *fs_info) 13950f7d52f4SChris Mason { 13960f7d52f4SChris Mason struct btrfs_root *root; 13970f7d52f4SChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 13987237f183SYan Zheng struct extent_buffer *leaf; 1399e02119d5SChris Mason 14006f07e42eSAl Viro root = btrfs_alloc_root(fs_info); 1401e02119d5SChris Mason if (!root) 14027237f183SYan Zheng return ERR_PTR(-ENOMEM); 1403e02119d5SChris Mason 1404707e8a07SDavid Sterba __setup_root(tree_root->nodesize, tree_root->sectorsize, 1405707e8a07SDavid Sterba tree_root->stripesize, root, fs_info, 1406707e8a07SDavid Sterba BTRFS_TREE_LOG_OBJECTID); 1407e02119d5SChris Mason 1408e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1409e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1410e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 141127cdeb70SMiao Xie 14127237f183SYan Zheng /* 141327cdeb70SMiao Xie * DON'T set REF_COWS for log trees 141427cdeb70SMiao Xie * 14157237f183SYan Zheng * log trees do not get reference counted because they go away 14167237f183SYan Zheng * before a real commit is actually done. They do store pointers 14177237f183SYan Zheng * to file data extents, and those reference counts still get 14187237f183SYan Zheng * updated (along with back refs to the log tree). 14197237f183SYan Zheng */ 1420e02119d5SChris Mason 14214d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 14224d75f8a9SDavid Sterba NULL, 0, 0, 0); 14237237f183SYan Zheng if (IS_ERR(leaf)) { 14247237f183SYan Zheng kfree(root); 14257237f183SYan Zheng return ERR_CAST(leaf); 14267237f183SYan Zheng } 1427e02119d5SChris Mason 14285d4f98a2SYan Zheng memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 14295d4f98a2SYan Zheng btrfs_set_header_bytenr(leaf, leaf->start); 14305d4f98a2SYan Zheng btrfs_set_header_generation(leaf, trans->transid); 14315d4f98a2SYan Zheng btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 14325d4f98a2SYan Zheng btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID); 14337237f183SYan Zheng root->node = leaf; 1434e02119d5SChris Mason 1435e02119d5SChris Mason write_extent_buffer(root->node, root->fs_info->fsid, 14360a4e5586SRoss Kirk btrfs_header_fsid(), BTRFS_FSID_SIZE); 1437e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1438e02119d5SChris Mason btrfs_tree_unlock(root->node); 14397237f183SYan Zheng return root; 14407237f183SYan Zheng } 14417237f183SYan Zheng 14427237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 14437237f183SYan Zheng struct btrfs_fs_info *fs_info) 14447237f183SYan Zheng { 14457237f183SYan Zheng struct btrfs_root *log_root; 14467237f183SYan Zheng 14477237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 14487237f183SYan Zheng if (IS_ERR(log_root)) 14497237f183SYan Zheng return PTR_ERR(log_root); 14507237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 14517237f183SYan Zheng fs_info->log_root_tree = log_root; 14527237f183SYan Zheng return 0; 14537237f183SYan Zheng } 14547237f183SYan Zheng 14557237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 14567237f183SYan Zheng struct btrfs_root *root) 14577237f183SYan Zheng { 14587237f183SYan Zheng struct btrfs_root *log_root; 14597237f183SYan Zheng struct btrfs_inode_item *inode_item; 14607237f183SYan Zheng 14617237f183SYan Zheng log_root = alloc_log_tree(trans, root->fs_info); 14627237f183SYan Zheng if (IS_ERR(log_root)) 14637237f183SYan Zheng return PTR_ERR(log_root); 14647237f183SYan Zheng 14657237f183SYan Zheng log_root->last_trans = trans->transid; 14667237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 14677237f183SYan Zheng 14687237f183SYan Zheng inode_item = &log_root->root_item.inode; 14693cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 14703cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 14713cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 1472707e8a07SDavid Sterba btrfs_set_stack_inode_nbytes(inode_item, root->nodesize); 14733cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 14747237f183SYan Zheng 14755d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 14767237f183SYan Zheng 14777237f183SYan Zheng WARN_ON(root->log_root); 14787237f183SYan Zheng root->log_root = log_root; 14797237f183SYan Zheng root->log_transid = 0; 1480d1433debSMiao Xie root->log_transid_committed = -1; 1481257c62e1SChris Mason root->last_log_commit = 0; 1482e02119d5SChris Mason return 0; 1483e02119d5SChris Mason } 1484e02119d5SChris Mason 148535a3621bSStefan Behrens static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 1486cb517eabSMiao Xie struct btrfs_key *key) 1487e02119d5SChris Mason { 1488e02119d5SChris Mason struct btrfs_root *root; 1489e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 14900f7d52f4SChris Mason struct btrfs_path *path; 149184234f3aSYan Zheng u64 generation; 1492cb517eabSMiao Xie int ret; 1493cb517eabSMiao Xie 1494cb517eabSMiao Xie path = btrfs_alloc_path(); 1495cb517eabSMiao Xie if (!path) 1496cb517eabSMiao Xie return ERR_PTR(-ENOMEM); 14970f7d52f4SChris Mason 14986f07e42eSAl Viro root = btrfs_alloc_root(fs_info); 1499cb517eabSMiao Xie if (!root) { 1500cb517eabSMiao Xie ret = -ENOMEM; 1501cb517eabSMiao Xie goto alloc_fail; 15020f7d52f4SChris Mason } 15030f7d52f4SChris Mason 1504707e8a07SDavid Sterba __setup_root(tree_root->nodesize, tree_root->sectorsize, 1505707e8a07SDavid Sterba tree_root->stripesize, root, fs_info, key->objectid); 15060f7d52f4SChris Mason 1507cb517eabSMiao Xie ret = btrfs_find_root(tree_root, key, path, 1508cb517eabSMiao Xie &root->root_item, &root->root_key); 15090f7d52f4SChris Mason if (ret) { 151013a8a7c8SYan, Zheng if (ret > 0) 151113a8a7c8SYan, Zheng ret = -ENOENT; 1512cb517eabSMiao Xie goto find_fail; 15130f7d52f4SChris Mason } 151413a8a7c8SYan, Zheng 151584234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1516db94535dSChris Mason root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item), 1517ce86cd59SDavid Sterba generation); 151864c043deSLiu Bo if (IS_ERR(root->node)) { 151964c043deSLiu Bo ret = PTR_ERR(root->node); 1520cb517eabSMiao Xie goto find_fail; 1521cb517eabSMiao Xie } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1522cb517eabSMiao Xie ret = -EIO; 152364c043deSLiu Bo free_extent_buffer(root->node); 152464c043deSLiu Bo goto find_fail; 1525416bc658SJosef Bacik } 15265d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 152713a8a7c8SYan, Zheng out: 1528cb517eabSMiao Xie btrfs_free_path(path); 1529cb517eabSMiao Xie return root; 1530cb517eabSMiao Xie 1531cb517eabSMiao Xie find_fail: 1532cb517eabSMiao Xie kfree(root); 1533cb517eabSMiao Xie alloc_fail: 1534cb517eabSMiao Xie root = ERR_PTR(ret); 1535cb517eabSMiao Xie goto out; 1536cb517eabSMiao Xie } 1537cb517eabSMiao Xie 1538cb517eabSMiao Xie struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root, 1539cb517eabSMiao Xie struct btrfs_key *location) 1540cb517eabSMiao Xie { 1541cb517eabSMiao Xie struct btrfs_root *root; 1542cb517eabSMiao Xie 1543cb517eabSMiao Xie root = btrfs_read_tree_root(tree_root, location); 1544cb517eabSMiao Xie if (IS_ERR(root)) 1545cb517eabSMiao Xie return root; 1546cb517eabSMiao Xie 1547cb517eabSMiao Xie if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { 154827cdeb70SMiao Xie set_bit(BTRFS_ROOT_REF_COWS, &root->state); 154908fe4db1SLi Zefan btrfs_check_and_init_root_item(&root->root_item); 155008fe4db1SLi Zefan } 155113a8a7c8SYan, Zheng 15525eda7b5eSChris Mason return root; 15535eda7b5eSChris Mason } 15545eda7b5eSChris Mason 1555cb517eabSMiao Xie int btrfs_init_fs_root(struct btrfs_root *root) 1556cb517eabSMiao Xie { 1557cb517eabSMiao Xie int ret; 15588257b2dcSMiao Xie struct btrfs_subvolume_writers *writers; 1559cb517eabSMiao Xie 1560cb517eabSMiao Xie root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); 1561cb517eabSMiao Xie root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), 1562cb517eabSMiao Xie GFP_NOFS); 1563cb517eabSMiao Xie if (!root->free_ino_pinned || !root->free_ino_ctl) { 1564cb517eabSMiao Xie ret = -ENOMEM; 1565cb517eabSMiao Xie goto fail; 1566cb517eabSMiao Xie } 1567cb517eabSMiao Xie 15688257b2dcSMiao Xie writers = btrfs_alloc_subvolume_writers(); 15698257b2dcSMiao Xie if (IS_ERR(writers)) { 15708257b2dcSMiao Xie ret = PTR_ERR(writers); 15718257b2dcSMiao Xie goto fail; 15728257b2dcSMiao Xie } 15738257b2dcSMiao Xie root->subv_writers = writers; 15748257b2dcSMiao Xie 1575cb517eabSMiao Xie btrfs_init_free_ino_ctl(root); 157657cdc8dbSDavid Sterba spin_lock_init(&root->ino_cache_lock); 157757cdc8dbSDavid Sterba init_waitqueue_head(&root->ino_cache_wait); 1578cb517eabSMiao Xie 1579cb517eabSMiao Xie ret = get_anon_bdev(&root->anon_dev); 1580cb517eabSMiao Xie if (ret) 15818257b2dcSMiao Xie goto free_writers; 1582cb517eabSMiao Xie return 0; 15838257b2dcSMiao Xie 15848257b2dcSMiao Xie free_writers: 15858257b2dcSMiao Xie btrfs_free_subvolume_writers(root->subv_writers); 1586cb517eabSMiao Xie fail: 1587cb517eabSMiao Xie kfree(root->free_ino_ctl); 1588cb517eabSMiao Xie kfree(root->free_ino_pinned); 1589cb517eabSMiao Xie return ret; 1590cb517eabSMiao Xie } 1591cb517eabSMiao Xie 1592171170c1SSergei Trofimovich static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1593cb517eabSMiao Xie u64 root_id) 1594cb517eabSMiao Xie { 1595cb517eabSMiao Xie struct btrfs_root *root; 1596cb517eabSMiao Xie 1597cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1598cb517eabSMiao Xie root = radix_tree_lookup(&fs_info->fs_roots_radix, 1599cb517eabSMiao Xie (unsigned long)root_id); 1600cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1601cb517eabSMiao Xie return root; 1602cb517eabSMiao Xie } 1603cb517eabSMiao Xie 1604cb517eabSMiao Xie int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1605cb517eabSMiao Xie struct btrfs_root *root) 1606cb517eabSMiao Xie { 1607cb517eabSMiao Xie int ret; 1608cb517eabSMiao Xie 1609cb517eabSMiao Xie ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); 1610cb517eabSMiao Xie if (ret) 1611cb517eabSMiao Xie return ret; 1612cb517eabSMiao Xie 1613cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1614cb517eabSMiao Xie ret = radix_tree_insert(&fs_info->fs_roots_radix, 1615cb517eabSMiao Xie (unsigned long)root->root_key.objectid, 1616cb517eabSMiao Xie root); 1617cb517eabSMiao Xie if (ret == 0) 161827cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1619cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1620cb517eabSMiao Xie radix_tree_preload_end(); 1621cb517eabSMiao Xie 1622cb517eabSMiao Xie return ret; 1623cb517eabSMiao Xie } 1624cb517eabSMiao Xie 1625c00869f1SMiao Xie struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 1626c00869f1SMiao Xie struct btrfs_key *location, 1627c00869f1SMiao Xie bool check_ref) 16285eda7b5eSChris Mason { 16295eda7b5eSChris Mason struct btrfs_root *root; 1630381cf658SDavid Sterba struct btrfs_path *path; 16311d4c08e0SDavid Sterba struct btrfs_key key; 16325eda7b5eSChris Mason int ret; 16335eda7b5eSChris Mason 1634edbd8d4eSChris Mason if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 1635edbd8d4eSChris Mason return fs_info->tree_root; 1636edbd8d4eSChris Mason if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID) 1637edbd8d4eSChris Mason return fs_info->extent_root; 16388f18cf13SChris Mason if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID) 16398f18cf13SChris Mason return fs_info->chunk_root; 16408f18cf13SChris Mason if (location->objectid == BTRFS_DEV_TREE_OBJECTID) 16418f18cf13SChris Mason return fs_info->dev_root; 16420403e47eSYan Zheng if (location->objectid == BTRFS_CSUM_TREE_OBJECTID) 16430403e47eSYan Zheng return fs_info->csum_root; 1644bcef60f2SArne Jansen if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID) 1645bcef60f2SArne Jansen return fs_info->quota_root ? fs_info->quota_root : 1646bcef60f2SArne Jansen ERR_PTR(-ENOENT); 1647f7a81ea4SStefan Behrens if (location->objectid == BTRFS_UUID_TREE_OBJECTID) 1648f7a81ea4SStefan Behrens return fs_info->uuid_root ? fs_info->uuid_root : 1649f7a81ea4SStefan Behrens ERR_PTR(-ENOENT); 16504df27c4dSYan, Zheng again: 1651cb517eabSMiao Xie root = btrfs_lookup_fs_root(fs_info, location->objectid); 165248475471SStefan Behrens if (root) { 1653c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) 165448475471SStefan Behrens return ERR_PTR(-ENOENT); 16555eda7b5eSChris Mason return root; 165648475471SStefan Behrens } 16575eda7b5eSChris Mason 1658cb517eabSMiao Xie root = btrfs_read_fs_root(fs_info->tree_root, location); 16595eda7b5eSChris Mason if (IS_ERR(root)) 16605eda7b5eSChris Mason return root; 16613394e160SChris Mason 1662c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1663d68fc57bSYan, Zheng ret = -ENOENT; 1664d68fc57bSYan, Zheng goto fail; 1665d68fc57bSYan, Zheng } 1666d68fc57bSYan, Zheng 1667cb517eabSMiao Xie ret = btrfs_init_fs_root(root); 1668cb517eabSMiao Xie if (ret) 1669cb517eabSMiao Xie goto fail; 1670cb517eabSMiao Xie 1671381cf658SDavid Sterba path = btrfs_alloc_path(); 1672381cf658SDavid Sterba if (!path) { 1673381cf658SDavid Sterba ret = -ENOMEM; 1674381cf658SDavid Sterba goto fail; 1675381cf658SDavid Sterba } 16761d4c08e0SDavid Sterba key.objectid = BTRFS_ORPHAN_OBJECTID; 16771d4c08e0SDavid Sterba key.type = BTRFS_ORPHAN_ITEM_KEY; 16781d4c08e0SDavid Sterba key.offset = location->objectid; 16791d4c08e0SDavid Sterba 16801d4c08e0SDavid Sterba ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1681381cf658SDavid Sterba btrfs_free_path(path); 1682d68fc57bSYan, Zheng if (ret < 0) 1683d68fc57bSYan, Zheng goto fail; 1684d68fc57bSYan, Zheng if (ret == 0) 168527cdeb70SMiao Xie set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1686d68fc57bSYan, Zheng 1687cb517eabSMiao Xie ret = btrfs_insert_fs_root(fs_info, root); 16880f7d52f4SChris Mason if (ret) { 16894df27c4dSYan, Zheng if (ret == -EEXIST) { 16904df27c4dSYan, Zheng free_fs_root(root); 16914df27c4dSYan, Zheng goto again; 16920f7d52f4SChris Mason } 16934df27c4dSYan, Zheng goto fail; 16944df27c4dSYan, Zheng } 1695edbd8d4eSChris Mason return root; 16964df27c4dSYan, Zheng fail: 16974df27c4dSYan, Zheng free_fs_root(root); 16984df27c4dSYan, Zheng return ERR_PTR(ret); 1699edbd8d4eSChris Mason } 1700edbd8d4eSChris Mason 170104160088SChris Mason static int btrfs_congested_fn(void *congested_data, int bdi_bits) 170204160088SChris Mason { 170304160088SChris Mason struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; 170404160088SChris Mason int ret = 0; 170504160088SChris Mason struct btrfs_device *device; 170604160088SChris Mason struct backing_dev_info *bdi; 1707b7967db7SChris Mason 17081f78160cSXiao Guangrong rcu_read_lock(); 17091f78160cSXiao Guangrong list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) { 1710dfe25020SChris Mason if (!device->bdev) 1711dfe25020SChris Mason continue; 171204160088SChris Mason bdi = blk_get_backing_dev_info(device->bdev); 1713ff9ea323STejun Heo if (bdi_congested(bdi, bdi_bits)) { 171404160088SChris Mason ret = 1; 171504160088SChris Mason break; 171604160088SChris Mason } 171704160088SChris Mason } 17181f78160cSXiao Guangrong rcu_read_unlock(); 171904160088SChris Mason return ret; 172004160088SChris Mason } 172104160088SChris Mason 172204160088SChris Mason static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi) 172304160088SChris Mason { 1724ad081f14SJens Axboe int err; 1725ad081f14SJens Axboe 1726b4caecd4SChristoph Hellwig err = bdi_setup_and_register(bdi, "btrfs"); 1727ad081f14SJens Axboe if (err) 1728ad081f14SJens Axboe return err; 1729ad081f14SJens Axboe 1730df0ce26cSChristoph Hellwig bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE; 173104160088SChris Mason bdi->congested_fn = btrfs_congested_fn; 173204160088SChris Mason bdi->congested_data = info; 1733da2f0f74SChris Mason bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK; 173404160088SChris Mason return 0; 173504160088SChris Mason } 173604160088SChris Mason 17378b712842SChris Mason /* 17388b712842SChris Mason * called by the kthread helper functions to finally call the bio end_io 17398b712842SChris Mason * functions. This is where read checksum verification actually happens 17408b712842SChris Mason */ 17418b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work) 1742ce9adaa5SChris Mason { 1743ce9adaa5SChris Mason struct bio *bio; 174497eb6b69SDavid Sterba struct btrfs_end_io_wq *end_io_wq; 1745ce9adaa5SChris Mason 174697eb6b69SDavid Sterba end_io_wq = container_of(work, struct btrfs_end_io_wq, work); 1747ce9adaa5SChris Mason bio = end_io_wq->bio; 17488b712842SChris Mason 17494246a0b6SChristoph Hellwig bio->bi_error = end_io_wq->error; 1750ce9adaa5SChris Mason bio->bi_private = end_io_wq->private; 1751ce9adaa5SChris Mason bio->bi_end_io = end_io_wq->end_io; 175297eb6b69SDavid Sterba kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq); 17534246a0b6SChristoph Hellwig bio_endio(bio); 1754ce9adaa5SChris Mason } 175544b8bd7eSChris Mason 1756a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1757a74a4b97SChris Mason { 1758a74a4b97SChris Mason struct btrfs_root *root = arg; 1759d0278245SMiao Xie int again; 1760da288d28SFilipe Manana struct btrfs_trans_handle *trans; 1761a74a4b97SChris Mason 1762a74a4b97SChris Mason do { 1763d0278245SMiao Xie again = 0; 17649d1a2a3aSDavid Sterba 1765d0278245SMiao Xie /* Make the cleaner go to sleep early. */ 1766babbf170SMiao Xie if (btrfs_need_cleaner_sleep(root)) 1767d0278245SMiao Xie goto sleep; 1768d0278245SMiao Xie 1769d0278245SMiao Xie if (!mutex_trylock(&root->fs_info->cleaner_mutex)) 1770d0278245SMiao Xie goto sleep; 1771d0278245SMiao Xie 1772dc7f370cSMiao Xie /* 1773dc7f370cSMiao Xie * Avoid the problem that we change the status of the fs 1774dc7f370cSMiao Xie * during the above check and trylock. 1775dc7f370cSMiao Xie */ 1776babbf170SMiao Xie if (btrfs_need_cleaner_sleep(root)) { 1777dc7f370cSMiao Xie mutex_unlock(&root->fs_info->cleaner_mutex); 1778dc7f370cSMiao Xie goto sleep; 1779dc7f370cSMiao Xie } 1780dc7f370cSMiao Xie 178124bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 17829d1a2a3aSDavid Sterba again = btrfs_clean_one_deleted_snapshot(root); 1783a74a4b97SChris Mason mutex_unlock(&root->fs_info->cleaner_mutex); 1784a74a4b97SChris Mason 1785d0278245SMiao Xie /* 178605323cd1SMiao Xie * The defragger has dealt with the R/O remount and umount, 178705323cd1SMiao Xie * needn't do anything special here. 1788d0278245SMiao Xie */ 1789d0278245SMiao Xie btrfs_run_defrag_inodes(root->fs_info); 179067c5e7d4SFilipe Manana 179167c5e7d4SFilipe Manana /* 179267c5e7d4SFilipe Manana * Acquires fs_info->delete_unused_bgs_mutex to avoid racing 179367c5e7d4SFilipe Manana * with relocation (btrfs_relocate_chunk) and relocation 179467c5e7d4SFilipe Manana * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 179567c5e7d4SFilipe Manana * after acquiring fs_info->delete_unused_bgs_mutex. So we 179667c5e7d4SFilipe Manana * can't hold, nor need to, fs_info->cleaner_mutex when deleting 179767c5e7d4SFilipe Manana * unused block groups. 179867c5e7d4SFilipe Manana */ 179967c5e7d4SFilipe Manana btrfs_delete_unused_bgs(root->fs_info); 1800d0278245SMiao Xie sleep: 18019d1a2a3aSDavid Sterba if (!try_to_freeze() && !again) { 1802a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 18038929ecfaSYan, Zheng if (!kthread_should_stop()) 1804a74a4b97SChris Mason schedule(); 1805a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1806a74a4b97SChris Mason } 1807a74a4b97SChris Mason } while (!kthread_should_stop()); 1808da288d28SFilipe Manana 1809da288d28SFilipe Manana /* 1810da288d28SFilipe Manana * Transaction kthread is stopped before us and wakes us up. 1811da288d28SFilipe Manana * However we might have started a new transaction and COWed some 1812da288d28SFilipe Manana * tree blocks when deleting unused block groups for example. So 1813da288d28SFilipe Manana * make sure we commit the transaction we started to have a clean 1814da288d28SFilipe Manana * shutdown when evicting the btree inode - if it has dirty pages 1815da288d28SFilipe Manana * when we do the final iput() on it, eviction will trigger a 1816da288d28SFilipe Manana * writeback for it which will fail with null pointer dereferences 1817da288d28SFilipe Manana * since work queues and other resources were already released and 1818da288d28SFilipe Manana * destroyed by the time the iput/eviction/writeback is made. 1819da288d28SFilipe Manana */ 1820da288d28SFilipe Manana trans = btrfs_attach_transaction(root); 1821da288d28SFilipe Manana if (IS_ERR(trans)) { 1822da288d28SFilipe Manana if (PTR_ERR(trans) != -ENOENT) 1823da288d28SFilipe Manana btrfs_err(root->fs_info, 1824da288d28SFilipe Manana "cleaner transaction attach returned %ld", 1825da288d28SFilipe Manana PTR_ERR(trans)); 1826da288d28SFilipe Manana } else { 1827da288d28SFilipe Manana int ret; 1828da288d28SFilipe Manana 1829da288d28SFilipe Manana ret = btrfs_commit_transaction(trans, root); 1830da288d28SFilipe Manana if (ret) 1831da288d28SFilipe Manana btrfs_err(root->fs_info, 1832da288d28SFilipe Manana "cleaner open transaction commit returned %d", 1833da288d28SFilipe Manana ret); 1834da288d28SFilipe Manana } 1835da288d28SFilipe Manana 1836a74a4b97SChris Mason return 0; 1837a74a4b97SChris Mason } 1838a74a4b97SChris Mason 1839a74a4b97SChris Mason static int transaction_kthread(void *arg) 1840a74a4b97SChris Mason { 1841a74a4b97SChris Mason struct btrfs_root *root = arg; 1842a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1843a74a4b97SChris Mason struct btrfs_transaction *cur; 18448929ecfaSYan, Zheng u64 transid; 1845a74a4b97SChris Mason unsigned long now; 1846a74a4b97SChris Mason unsigned long delay; 1847914b2007SJan Kara bool cannot_commit; 1848a74a4b97SChris Mason 1849a74a4b97SChris Mason do { 1850914b2007SJan Kara cannot_commit = false; 18518b87dc17SDavid Sterba delay = HZ * root->fs_info->commit_interval; 1852a74a4b97SChris Mason mutex_lock(&root->fs_info->transaction_kthread_mutex); 1853a74a4b97SChris Mason 1854a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 1855a74a4b97SChris Mason cur = root->fs_info->running_transaction; 1856a74a4b97SChris Mason if (!cur) { 1857a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1858a74a4b97SChris Mason goto sleep; 1859a74a4b97SChris Mason } 186031153d81SYan Zheng 1861a74a4b97SChris Mason now = get_seconds(); 18624a9d8bdeSMiao Xie if (cur->state < TRANS_STATE_BLOCKED && 18638b87dc17SDavid Sterba (now < cur->start_time || 18648b87dc17SDavid Sterba now - cur->start_time < root->fs_info->commit_interval)) { 1865a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1866a74a4b97SChris Mason delay = HZ * 5; 1867a74a4b97SChris Mason goto sleep; 1868a74a4b97SChris Mason } 18698929ecfaSYan, Zheng transid = cur->transid; 1870a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 187156bec294SChris Mason 187279787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 1873354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 1874914b2007SJan Kara if (IS_ERR(trans)) { 1875354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 1876914b2007SJan Kara cannot_commit = true; 187779787eaaSJeff Mahoney goto sleep; 1878914b2007SJan Kara } 18798929ecfaSYan, Zheng if (transid == trans->transid) { 188079787eaaSJeff Mahoney btrfs_commit_transaction(trans, root); 18818929ecfaSYan, Zheng } else { 18828929ecfaSYan, Zheng btrfs_end_transaction(trans, root); 18838929ecfaSYan, Zheng } 1884a74a4b97SChris Mason sleep: 1885a74a4b97SChris Mason wake_up_process(root->fs_info->cleaner_kthread); 1886a74a4b97SChris Mason mutex_unlock(&root->fs_info->transaction_kthread_mutex); 1887a74a4b97SChris Mason 18884e121c06SJosef Bacik if (unlikely(test_bit(BTRFS_FS_STATE_ERROR, 18894e121c06SJosef Bacik &root->fs_info->fs_state))) 18904e121c06SJosef Bacik btrfs_cleanup_transaction(root); 1891a0acae0eSTejun Heo if (!try_to_freeze()) { 1892a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 18938929ecfaSYan, Zheng if (!kthread_should_stop() && 1894914b2007SJan Kara (!btrfs_transaction_blocked(root->fs_info) || 1895914b2007SJan Kara cannot_commit)) 1896a74a4b97SChris Mason schedule_timeout(delay); 1897a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1898a74a4b97SChris Mason } 1899a74a4b97SChris Mason } while (!kthread_should_stop()); 1900a74a4b97SChris Mason return 0; 1901a74a4b97SChris Mason } 1902a74a4b97SChris Mason 1903af31f5e5SChris Mason /* 1904af31f5e5SChris Mason * this will find the highest generation in the array of 1905af31f5e5SChris Mason * root backups. The index of the highest array is returned, 1906af31f5e5SChris Mason * or -1 if we can't find anything. 1907af31f5e5SChris Mason * 1908af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1909af31f5e5SChris Mason * generation of the latest root in the array with the generation 1910af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1911af31f5e5SChris Mason */ 1912af31f5e5SChris Mason static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen) 1913af31f5e5SChris Mason { 1914af31f5e5SChris Mason u64 cur; 1915af31f5e5SChris Mason int newest_index = -1; 1916af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1917af31f5e5SChris Mason int i; 1918af31f5e5SChris Mason 1919af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1920af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1921af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1922af31f5e5SChris Mason if (cur == newest_gen) 1923af31f5e5SChris Mason newest_index = i; 1924af31f5e5SChris Mason } 1925af31f5e5SChris Mason 1926af31f5e5SChris Mason /* check to see if we actually wrapped around */ 1927af31f5e5SChris Mason if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) { 1928af31f5e5SChris Mason root_backup = info->super_copy->super_roots; 1929af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1930af31f5e5SChris Mason if (cur == newest_gen) 1931af31f5e5SChris Mason newest_index = 0; 1932af31f5e5SChris Mason } 1933af31f5e5SChris Mason return newest_index; 1934af31f5e5SChris Mason } 1935af31f5e5SChris Mason 1936af31f5e5SChris Mason 1937af31f5e5SChris Mason /* 1938af31f5e5SChris Mason * find the oldest backup so we know where to store new entries 1939af31f5e5SChris Mason * in the backup array. This will set the backup_root_index 1940af31f5e5SChris Mason * field in the fs_info struct 1941af31f5e5SChris Mason */ 1942af31f5e5SChris Mason static void find_oldest_super_backup(struct btrfs_fs_info *info, 1943af31f5e5SChris Mason u64 newest_gen) 1944af31f5e5SChris Mason { 1945af31f5e5SChris Mason int newest_index = -1; 1946af31f5e5SChris Mason 1947af31f5e5SChris Mason newest_index = find_newest_super_backup(info, newest_gen); 1948af31f5e5SChris Mason /* if there was garbage in there, just move along */ 1949af31f5e5SChris Mason if (newest_index == -1) { 1950af31f5e5SChris Mason info->backup_root_index = 0; 1951af31f5e5SChris Mason } else { 1952af31f5e5SChris Mason info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS; 1953af31f5e5SChris Mason } 1954af31f5e5SChris Mason } 1955af31f5e5SChris Mason 1956af31f5e5SChris Mason /* 1957af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1958af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1959af31f5e5SChris Mason * done 1960af31f5e5SChris Mason */ 1961af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1962af31f5e5SChris Mason { 1963af31f5e5SChris Mason int next_backup; 1964af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1965af31f5e5SChris Mason int last_backup; 1966af31f5e5SChris Mason 1967af31f5e5SChris Mason next_backup = info->backup_root_index; 1968af31f5e5SChris Mason last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) % 1969af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 1970af31f5e5SChris Mason 1971af31f5e5SChris Mason /* 1972af31f5e5SChris Mason * just overwrite the last backup if we're at the same generation 1973af31f5e5SChris Mason * this happens only at umount 1974af31f5e5SChris Mason */ 1975af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + last_backup; 1976af31f5e5SChris Mason if (btrfs_backup_tree_root_gen(root_backup) == 1977af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)) 1978af31f5e5SChris Mason next_backup = last_backup; 1979af31f5e5SChris Mason 1980af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1981af31f5e5SChris Mason 1982af31f5e5SChris Mason /* 1983af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1984af31f5e5SChris Mason * regardless of which ones we use today 1985af31f5e5SChris Mason */ 1986af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1987af31f5e5SChris Mason 1988af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1989af31f5e5SChris Mason 1990af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1991af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1992af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1993af31f5e5SChris Mason 1994af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1995af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1996af31f5e5SChris Mason 1997af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1998af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1999af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 2000af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 2001af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 2002af31f5e5SChris Mason 2003af31f5e5SChris Mason btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); 2004af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 2005af31f5e5SChris Mason btrfs_header_generation(info->extent_root->node)); 2006af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 2007af31f5e5SChris Mason btrfs_header_level(info->extent_root->node)); 2008af31f5e5SChris Mason 20097c7e82a7SChris Mason /* 20107c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 20117c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 20127c7e82a7SChris Mason */ 20137c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 20147c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 20157c7e82a7SChris Mason info->fs_root->node->start); 2016af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 2017af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 2018af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 2019af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 20207c7e82a7SChris Mason } 2021af31f5e5SChris Mason 2022af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 2023af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 2024af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 2025af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 2026af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 2027af31f5e5SChris Mason 2028af31f5e5SChris Mason btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); 2029af31f5e5SChris Mason btrfs_set_backup_csum_root_gen(root_backup, 2030af31f5e5SChris Mason btrfs_header_generation(info->csum_root->node)); 2031af31f5e5SChris Mason btrfs_set_backup_csum_root_level(root_backup, 2032af31f5e5SChris Mason btrfs_header_level(info->csum_root->node)); 2033af31f5e5SChris Mason 2034af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 2035af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 2036af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 2037af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 2038af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 2039af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 2040af31f5e5SChris Mason 2041af31f5e5SChris Mason /* 2042af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 2043af31f5e5SChris Mason * for the next commit 2044af31f5e5SChris Mason */ 2045af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 2046af31f5e5SChris Mason &info->super_for_commit->super_roots, 2047af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 2048af31f5e5SChris Mason } 2049af31f5e5SChris Mason 2050af31f5e5SChris Mason /* 2051af31f5e5SChris Mason * this copies info out of the root backup array and back into 2052af31f5e5SChris Mason * the in-memory super block. It is meant to help iterate through 2053af31f5e5SChris Mason * the array, so you send it the number of backups you've already 2054af31f5e5SChris Mason * tried and the last backup index you used. 2055af31f5e5SChris Mason * 2056af31f5e5SChris Mason * this returns -1 when it has tried all the backups 2057af31f5e5SChris Mason */ 2058af31f5e5SChris Mason static noinline int next_root_backup(struct btrfs_fs_info *info, 2059af31f5e5SChris Mason struct btrfs_super_block *super, 2060af31f5e5SChris Mason int *num_backups_tried, int *backup_index) 2061af31f5e5SChris Mason { 2062af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 2063af31f5e5SChris Mason int newest = *backup_index; 2064af31f5e5SChris Mason 2065af31f5e5SChris Mason if (*num_backups_tried == 0) { 2066af31f5e5SChris Mason u64 gen = btrfs_super_generation(super); 2067af31f5e5SChris Mason 2068af31f5e5SChris Mason newest = find_newest_super_backup(info, gen); 2069af31f5e5SChris Mason if (newest == -1) 2070af31f5e5SChris Mason return -1; 2071af31f5e5SChris Mason 2072af31f5e5SChris Mason *backup_index = newest; 2073af31f5e5SChris Mason *num_backups_tried = 1; 2074af31f5e5SChris Mason } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) { 2075af31f5e5SChris Mason /* we've tried all the backups, all done */ 2076af31f5e5SChris Mason return -1; 2077af31f5e5SChris Mason } else { 2078af31f5e5SChris Mason /* jump to the next oldest backup */ 2079af31f5e5SChris Mason newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) % 2080af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 2081af31f5e5SChris Mason *backup_index = newest; 2082af31f5e5SChris Mason *num_backups_tried += 1; 2083af31f5e5SChris Mason } 2084af31f5e5SChris Mason root_backup = super->super_roots + newest; 2085af31f5e5SChris Mason 2086af31f5e5SChris Mason btrfs_set_super_generation(super, 2087af31f5e5SChris Mason btrfs_backup_tree_root_gen(root_backup)); 2088af31f5e5SChris Mason btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 2089af31f5e5SChris Mason btrfs_set_super_root_level(super, 2090af31f5e5SChris Mason btrfs_backup_tree_root_level(root_backup)); 2091af31f5e5SChris Mason btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 2092af31f5e5SChris Mason 2093af31f5e5SChris Mason /* 2094af31f5e5SChris Mason * fixme: the total bytes and num_devices need to match or we should 2095af31f5e5SChris Mason * need a fsck 2096af31f5e5SChris Mason */ 2097af31f5e5SChris Mason btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 2098af31f5e5SChris Mason btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 2099af31f5e5SChris Mason return 0; 2100af31f5e5SChris Mason } 2101af31f5e5SChris Mason 21027abadb64SLiu Bo /* helper to cleanup workers */ 21037abadb64SLiu Bo static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 21047abadb64SLiu Bo { 2105dc6e3209SQu Wenruo btrfs_destroy_workqueue(fs_info->fixup_workers); 2106afe3d242SQu Wenruo btrfs_destroy_workqueue(fs_info->delalloc_workers); 21075cdc7ad3SQu Wenruo btrfs_destroy_workqueue(fs_info->workers); 2108fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_workers); 2109fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_meta_workers); 2110fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_raid56_workers); 21118b110e39SMiao Xie btrfs_destroy_workqueue(fs_info->endio_repair_workers); 2112d05a33acSQu Wenruo btrfs_destroy_workqueue(fs_info->rmw_workers); 2113fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_meta_write_workers); 2114fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_write_workers); 2115fccb5d86SQu Wenruo btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 2116a8c93d4eSQu Wenruo btrfs_destroy_workqueue(fs_info->submit_workers); 21175b3bc44eSQu Wenruo btrfs_destroy_workqueue(fs_info->delayed_workers); 2118e66f0bb1SQu Wenruo btrfs_destroy_workqueue(fs_info->caching_workers); 2119736cfa15SQu Wenruo btrfs_destroy_workqueue(fs_info->readahead_workers); 2120a44903abSQu Wenruo btrfs_destroy_workqueue(fs_info->flush_workers); 2121fc97fab0SQu Wenruo btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2122a79b7d4bSChris Mason btrfs_destroy_workqueue(fs_info->extent_workers); 21237abadb64SLiu Bo } 21247abadb64SLiu Bo 21252e9f5954SRashika static void free_root_extent_buffers(struct btrfs_root *root) 21262e9f5954SRashika { 21272e9f5954SRashika if (root) { 21282e9f5954SRashika free_extent_buffer(root->node); 21292e9f5954SRashika free_extent_buffer(root->commit_root); 21302e9f5954SRashika root->node = NULL; 21312e9f5954SRashika root->commit_root = NULL; 21322e9f5954SRashika } 21332e9f5954SRashika } 21342e9f5954SRashika 2135af31f5e5SChris Mason /* helper to cleanup tree roots */ 2136af31f5e5SChris Mason static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root) 2137af31f5e5SChris Mason { 21382e9f5954SRashika free_root_extent_buffers(info->tree_root); 2139655b09feSJosef Bacik 21402e9f5954SRashika free_root_extent_buffers(info->dev_root); 21412e9f5954SRashika free_root_extent_buffers(info->extent_root); 21422e9f5954SRashika free_root_extent_buffers(info->csum_root); 21432e9f5954SRashika free_root_extent_buffers(info->quota_root); 21442e9f5954SRashika free_root_extent_buffers(info->uuid_root); 21452e9f5954SRashika if (chunk_root) 21462e9f5954SRashika free_root_extent_buffers(info->chunk_root); 2147af31f5e5SChris Mason } 2148af31f5e5SChris Mason 2149faa2dbf0SJosef Bacik void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2150171f6537SJosef Bacik { 2151171f6537SJosef Bacik int ret; 2152171f6537SJosef Bacik struct btrfs_root *gang[8]; 2153171f6537SJosef Bacik int i; 2154171f6537SJosef Bacik 2155171f6537SJosef Bacik while (!list_empty(&fs_info->dead_roots)) { 2156171f6537SJosef Bacik gang[0] = list_entry(fs_info->dead_roots.next, 2157171f6537SJosef Bacik struct btrfs_root, root_list); 2158171f6537SJosef Bacik list_del(&gang[0]->root_list); 2159171f6537SJosef Bacik 216027cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) { 2161cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[0]); 2162171f6537SJosef Bacik } else { 2163171f6537SJosef Bacik free_extent_buffer(gang[0]->node); 2164171f6537SJosef Bacik free_extent_buffer(gang[0]->commit_root); 2165b0feb9d9SMiao Xie btrfs_put_fs_root(gang[0]); 2166171f6537SJosef Bacik } 2167171f6537SJosef Bacik } 2168171f6537SJosef Bacik 2169171f6537SJosef Bacik while (1) { 2170171f6537SJosef Bacik ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2171171f6537SJosef Bacik (void **)gang, 0, 2172171f6537SJosef Bacik ARRAY_SIZE(gang)); 2173171f6537SJosef Bacik if (!ret) 2174171f6537SJosef Bacik break; 2175171f6537SJosef Bacik for (i = 0; i < ret; i++) 2176cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2177171f6537SJosef Bacik } 21781a4319ccSLiu Bo 21791a4319ccSLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 21801a4319ccSLiu Bo btrfs_free_log_root_tree(NULL, fs_info); 21811a4319ccSLiu Bo btrfs_destroy_pinned_extent(fs_info->tree_root, 21821a4319ccSLiu Bo fs_info->pinned_extents); 21831a4319ccSLiu Bo } 2184171f6537SJosef Bacik } 2185af31f5e5SChris Mason 2186638aa7edSEric Sandeen static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2187638aa7edSEric Sandeen { 2188638aa7edSEric Sandeen mutex_init(&fs_info->scrub_lock); 2189638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_running, 0); 2190638aa7edSEric Sandeen atomic_set(&fs_info->scrub_pause_req, 0); 2191638aa7edSEric Sandeen atomic_set(&fs_info->scrubs_paused, 0); 2192638aa7edSEric Sandeen atomic_set(&fs_info->scrub_cancel_req, 0); 2193638aa7edSEric Sandeen init_waitqueue_head(&fs_info->scrub_pause_wait); 2194638aa7edSEric Sandeen fs_info->scrub_workers_refcnt = 0; 2195638aa7edSEric Sandeen } 2196638aa7edSEric Sandeen 2197779a65a4SEric Sandeen static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2198779a65a4SEric Sandeen { 2199779a65a4SEric Sandeen spin_lock_init(&fs_info->balance_lock); 2200779a65a4SEric Sandeen mutex_init(&fs_info->balance_mutex); 2201779a65a4SEric Sandeen atomic_set(&fs_info->balance_running, 0); 2202779a65a4SEric Sandeen atomic_set(&fs_info->balance_pause_req, 0); 2203779a65a4SEric Sandeen atomic_set(&fs_info->balance_cancel_req, 0); 2204779a65a4SEric Sandeen fs_info->balance_ctl = NULL; 2205779a65a4SEric Sandeen init_waitqueue_head(&fs_info->balance_wait_q); 2206779a65a4SEric Sandeen } 2207779a65a4SEric Sandeen 2208f37938e0SEric Sandeen static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info, 2209f37938e0SEric Sandeen struct btrfs_root *tree_root) 2210f37938e0SEric Sandeen { 2211f37938e0SEric Sandeen fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 2212f37938e0SEric Sandeen set_nlink(fs_info->btree_inode, 1); 2213f37938e0SEric Sandeen /* 2214f37938e0SEric Sandeen * we set the i_size on the btree inode to the max possible int. 2215f37938e0SEric Sandeen * the real end of the address space is determined by all of 2216f37938e0SEric Sandeen * the devices in the system 2217f37938e0SEric Sandeen */ 2218f37938e0SEric Sandeen fs_info->btree_inode->i_size = OFFSET_MAX; 2219f37938e0SEric Sandeen fs_info->btree_inode->i_mapping->a_ops = &btree_aops; 2220f37938e0SEric Sandeen 2221f37938e0SEric Sandeen RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node); 2222f37938e0SEric Sandeen extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree, 2223f37938e0SEric Sandeen fs_info->btree_inode->i_mapping); 2224f37938e0SEric Sandeen BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0; 2225f37938e0SEric Sandeen extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree); 2226f37938e0SEric Sandeen 2227f37938e0SEric Sandeen BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops; 2228f37938e0SEric Sandeen 2229f37938e0SEric Sandeen BTRFS_I(fs_info->btree_inode)->root = tree_root; 2230f37938e0SEric Sandeen memset(&BTRFS_I(fs_info->btree_inode)->location, 0, 2231f37938e0SEric Sandeen sizeof(struct btrfs_key)); 2232f37938e0SEric Sandeen set_bit(BTRFS_INODE_DUMMY, 2233f37938e0SEric Sandeen &BTRFS_I(fs_info->btree_inode)->runtime_flags); 2234f37938e0SEric Sandeen btrfs_insert_inode_hash(fs_info->btree_inode); 2235f37938e0SEric Sandeen } 2236f37938e0SEric Sandeen 2237ad618368SEric Sandeen static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2238ad618368SEric Sandeen { 2239ad618368SEric Sandeen fs_info->dev_replace.lock_owner = 0; 2240ad618368SEric Sandeen atomic_set(&fs_info->dev_replace.nesting_level, 0); 2241ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2242ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock_management_lock); 2243ad618368SEric Sandeen mutex_init(&fs_info->dev_replace.lock); 2244ad618368SEric Sandeen init_waitqueue_head(&fs_info->replace_wait); 2245ad618368SEric Sandeen } 2246ad618368SEric Sandeen 2247f9e92e40SEric Sandeen static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2248f9e92e40SEric Sandeen { 2249f9e92e40SEric Sandeen spin_lock_init(&fs_info->qgroup_lock); 2250f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_ioctl_lock); 2251f9e92e40SEric Sandeen fs_info->qgroup_tree = RB_ROOT; 2252f9e92e40SEric Sandeen fs_info->qgroup_op_tree = RB_ROOT; 2253f9e92e40SEric Sandeen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2254f9e92e40SEric Sandeen fs_info->qgroup_seq = 1; 2255f9e92e40SEric Sandeen fs_info->quota_enabled = 0; 2256f9e92e40SEric Sandeen fs_info->pending_quota_state = 0; 2257f9e92e40SEric Sandeen fs_info->qgroup_ulist = NULL; 2258f9e92e40SEric Sandeen mutex_init(&fs_info->qgroup_rescan_lock); 2259f9e92e40SEric Sandeen } 2260f9e92e40SEric Sandeen 22612a458198SEric Sandeen static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info, 22622a458198SEric Sandeen struct btrfs_fs_devices *fs_devices) 22632a458198SEric Sandeen { 22642a458198SEric Sandeen int max_active = fs_info->thread_pool_size; 22656f011058SDavid Sterba unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 22662a458198SEric Sandeen 22672a458198SEric Sandeen fs_info->workers = 22682a458198SEric Sandeen btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI, 22692a458198SEric Sandeen max_active, 16); 22702a458198SEric Sandeen 22712a458198SEric Sandeen fs_info->delalloc_workers = 22722a458198SEric Sandeen btrfs_alloc_workqueue("delalloc", flags, max_active, 2); 22732a458198SEric Sandeen 22742a458198SEric Sandeen fs_info->flush_workers = 22752a458198SEric Sandeen btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0); 22762a458198SEric Sandeen 22772a458198SEric Sandeen fs_info->caching_workers = 22782a458198SEric Sandeen btrfs_alloc_workqueue("cache", flags, max_active, 0); 22792a458198SEric Sandeen 22802a458198SEric Sandeen /* 22812a458198SEric Sandeen * a higher idle thresh on the submit workers makes it much more 22822a458198SEric Sandeen * likely that bios will be send down in a sane order to the 22832a458198SEric Sandeen * devices 22842a458198SEric Sandeen */ 22852a458198SEric Sandeen fs_info->submit_workers = 22862a458198SEric Sandeen btrfs_alloc_workqueue("submit", flags, 22872a458198SEric Sandeen min_t(u64, fs_devices->num_devices, 22882a458198SEric Sandeen max_active), 64); 22892a458198SEric Sandeen 22902a458198SEric Sandeen fs_info->fixup_workers = 22912a458198SEric Sandeen btrfs_alloc_workqueue("fixup", flags, 1, 0); 22922a458198SEric Sandeen 22932a458198SEric Sandeen /* 22942a458198SEric Sandeen * endios are largely parallel and should have a very 22952a458198SEric Sandeen * low idle thresh 22962a458198SEric Sandeen */ 22972a458198SEric Sandeen fs_info->endio_workers = 22982a458198SEric Sandeen btrfs_alloc_workqueue("endio", flags, max_active, 4); 22992a458198SEric Sandeen fs_info->endio_meta_workers = 23002a458198SEric Sandeen btrfs_alloc_workqueue("endio-meta", flags, max_active, 4); 23012a458198SEric Sandeen fs_info->endio_meta_write_workers = 23022a458198SEric Sandeen btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2); 23032a458198SEric Sandeen fs_info->endio_raid56_workers = 23042a458198SEric Sandeen btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4); 23052a458198SEric Sandeen fs_info->endio_repair_workers = 23062a458198SEric Sandeen btrfs_alloc_workqueue("endio-repair", flags, 1, 0); 23072a458198SEric Sandeen fs_info->rmw_workers = 23082a458198SEric Sandeen btrfs_alloc_workqueue("rmw", flags, max_active, 2); 23092a458198SEric Sandeen fs_info->endio_write_workers = 23102a458198SEric Sandeen btrfs_alloc_workqueue("endio-write", flags, max_active, 2); 23112a458198SEric Sandeen fs_info->endio_freespace_worker = 23122a458198SEric Sandeen btrfs_alloc_workqueue("freespace-write", flags, max_active, 0); 23132a458198SEric Sandeen fs_info->delayed_workers = 23142a458198SEric Sandeen btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0); 23152a458198SEric Sandeen fs_info->readahead_workers = 23162a458198SEric Sandeen btrfs_alloc_workqueue("readahead", flags, max_active, 2); 23172a458198SEric Sandeen fs_info->qgroup_rescan_workers = 23182a458198SEric Sandeen btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0); 23192a458198SEric Sandeen fs_info->extent_workers = 23202a458198SEric Sandeen btrfs_alloc_workqueue("extent-refs", flags, 23212a458198SEric Sandeen min_t(u64, fs_devices->num_devices, 23222a458198SEric Sandeen max_active), 8); 23232a458198SEric Sandeen 23242a458198SEric Sandeen if (!(fs_info->workers && fs_info->delalloc_workers && 23252a458198SEric Sandeen fs_info->submit_workers && fs_info->flush_workers && 23262a458198SEric Sandeen fs_info->endio_workers && fs_info->endio_meta_workers && 23272a458198SEric Sandeen fs_info->endio_meta_write_workers && 23282a458198SEric Sandeen fs_info->endio_repair_workers && 23292a458198SEric Sandeen fs_info->endio_write_workers && fs_info->endio_raid56_workers && 23302a458198SEric Sandeen fs_info->endio_freespace_worker && fs_info->rmw_workers && 23312a458198SEric Sandeen fs_info->caching_workers && fs_info->readahead_workers && 23322a458198SEric Sandeen fs_info->fixup_workers && fs_info->delayed_workers && 23332a458198SEric Sandeen fs_info->extent_workers && 23342a458198SEric Sandeen fs_info->qgroup_rescan_workers)) { 23352a458198SEric Sandeen return -ENOMEM; 23362a458198SEric Sandeen } 23372a458198SEric Sandeen 23382a458198SEric Sandeen return 0; 23392a458198SEric Sandeen } 23402a458198SEric Sandeen 234163443bf5SEric Sandeen static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 234263443bf5SEric Sandeen struct btrfs_fs_devices *fs_devices) 234363443bf5SEric Sandeen { 234463443bf5SEric Sandeen int ret; 234563443bf5SEric Sandeen struct btrfs_root *tree_root = fs_info->tree_root; 234663443bf5SEric Sandeen struct btrfs_root *log_tree_root; 234763443bf5SEric Sandeen struct btrfs_super_block *disk_super = fs_info->super_copy; 234863443bf5SEric Sandeen u64 bytenr = btrfs_super_log_root(disk_super); 234963443bf5SEric Sandeen 235063443bf5SEric Sandeen if (fs_devices->rw_devices == 0) { 235163443bf5SEric Sandeen printk(KERN_WARNING "BTRFS: log replay required " 235263443bf5SEric Sandeen "on RO media\n"); 235363443bf5SEric Sandeen return -EIO; 235463443bf5SEric Sandeen } 235563443bf5SEric Sandeen 235663443bf5SEric Sandeen log_tree_root = btrfs_alloc_root(fs_info); 235763443bf5SEric Sandeen if (!log_tree_root) 235863443bf5SEric Sandeen return -ENOMEM; 235963443bf5SEric Sandeen 236063443bf5SEric Sandeen __setup_root(tree_root->nodesize, tree_root->sectorsize, 236163443bf5SEric Sandeen tree_root->stripesize, log_tree_root, fs_info, 236263443bf5SEric Sandeen BTRFS_TREE_LOG_OBJECTID); 236363443bf5SEric Sandeen 236463443bf5SEric Sandeen log_tree_root->node = read_tree_block(tree_root, bytenr, 236563443bf5SEric Sandeen fs_info->generation + 1); 236664c043deSLiu Bo if (IS_ERR(log_tree_root->node)) { 236764c043deSLiu Bo printk(KERN_ERR "BTRFS: failed to read log tree\n"); 23680eeff236SLiu Bo ret = PTR_ERR(log_tree_root->node); 236964c043deSLiu Bo kfree(log_tree_root); 23700eeff236SLiu Bo return ret; 237164c043deSLiu Bo } else if (!extent_buffer_uptodate(log_tree_root->node)) { 237263443bf5SEric Sandeen printk(KERN_ERR "BTRFS: failed to read log tree\n"); 237363443bf5SEric Sandeen free_extent_buffer(log_tree_root->node); 237463443bf5SEric Sandeen kfree(log_tree_root); 237563443bf5SEric Sandeen return -EIO; 237663443bf5SEric Sandeen } 237763443bf5SEric Sandeen /* returns with log_tree_root freed on success */ 237863443bf5SEric Sandeen ret = btrfs_recover_log_trees(log_tree_root); 237963443bf5SEric Sandeen if (ret) { 238063443bf5SEric Sandeen btrfs_error(tree_root->fs_info, ret, 238163443bf5SEric Sandeen "Failed to recover log tree"); 238263443bf5SEric Sandeen free_extent_buffer(log_tree_root->node); 238363443bf5SEric Sandeen kfree(log_tree_root); 238463443bf5SEric Sandeen return ret; 238563443bf5SEric Sandeen } 238663443bf5SEric Sandeen 238763443bf5SEric Sandeen if (fs_info->sb->s_flags & MS_RDONLY) { 238863443bf5SEric Sandeen ret = btrfs_commit_super(tree_root); 238963443bf5SEric Sandeen if (ret) 239063443bf5SEric Sandeen return ret; 239163443bf5SEric Sandeen } 239263443bf5SEric Sandeen 239363443bf5SEric Sandeen return 0; 239463443bf5SEric Sandeen } 239563443bf5SEric Sandeen 23964bbcaa64SEric Sandeen static int btrfs_read_roots(struct btrfs_fs_info *fs_info, 23974bbcaa64SEric Sandeen struct btrfs_root *tree_root) 23984bbcaa64SEric Sandeen { 2399a4f3d2c4SDavid Sterba struct btrfs_root *root; 24004bbcaa64SEric Sandeen struct btrfs_key location; 24014bbcaa64SEric Sandeen int ret; 24024bbcaa64SEric Sandeen 24034bbcaa64SEric Sandeen location.objectid = BTRFS_EXTENT_TREE_OBJECTID; 24044bbcaa64SEric Sandeen location.type = BTRFS_ROOT_ITEM_KEY; 24054bbcaa64SEric Sandeen location.offset = 0; 24064bbcaa64SEric Sandeen 2407a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2408a4f3d2c4SDavid Sterba if (IS_ERR(root)) 2409a4f3d2c4SDavid Sterba return PTR_ERR(root); 2410a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2411a4f3d2c4SDavid Sterba fs_info->extent_root = root; 24124bbcaa64SEric Sandeen 24134bbcaa64SEric Sandeen location.objectid = BTRFS_DEV_TREE_OBJECTID; 2414a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2415a4f3d2c4SDavid Sterba if (IS_ERR(root)) 2416a4f3d2c4SDavid Sterba return PTR_ERR(root); 2417a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2418a4f3d2c4SDavid Sterba fs_info->dev_root = root; 24194bbcaa64SEric Sandeen btrfs_init_devices_late(fs_info); 24204bbcaa64SEric Sandeen 24214bbcaa64SEric Sandeen location.objectid = BTRFS_CSUM_TREE_OBJECTID; 2422a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2423a4f3d2c4SDavid Sterba if (IS_ERR(root)) 2424a4f3d2c4SDavid Sterba return PTR_ERR(root); 2425a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2426a4f3d2c4SDavid Sterba fs_info->csum_root = root; 24274bbcaa64SEric Sandeen 24284bbcaa64SEric Sandeen location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2429a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2430a4f3d2c4SDavid Sterba if (!IS_ERR(root)) { 2431a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 24324bbcaa64SEric Sandeen fs_info->quota_enabled = 1; 24334bbcaa64SEric Sandeen fs_info->pending_quota_state = 1; 2434a4f3d2c4SDavid Sterba fs_info->quota_root = root; 24354bbcaa64SEric Sandeen } 24364bbcaa64SEric Sandeen 24374bbcaa64SEric Sandeen location.objectid = BTRFS_UUID_TREE_OBJECTID; 2438a4f3d2c4SDavid Sterba root = btrfs_read_tree_root(tree_root, &location); 2439a4f3d2c4SDavid Sterba if (IS_ERR(root)) { 2440a4f3d2c4SDavid Sterba ret = PTR_ERR(root); 24414bbcaa64SEric Sandeen if (ret != -ENOENT) 24424bbcaa64SEric Sandeen return ret; 24434bbcaa64SEric Sandeen } else { 2444a4f3d2c4SDavid Sterba set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2445a4f3d2c4SDavid Sterba fs_info->uuid_root = root; 24464bbcaa64SEric Sandeen } 24474bbcaa64SEric Sandeen 24484bbcaa64SEric Sandeen return 0; 24494bbcaa64SEric Sandeen } 24504bbcaa64SEric Sandeen 2451ad2b2c80SAl Viro int open_ctree(struct super_block *sb, 2452dfe25020SChris Mason struct btrfs_fs_devices *fs_devices, 2453dfe25020SChris Mason char *options) 2454eb60ceacSChris Mason { 2455db94535dSChris Mason u32 sectorsize; 2456db94535dSChris Mason u32 nodesize; 245787ee04ebSChris Mason u32 stripesize; 245884234f3aSYan Zheng u64 generation; 2459f2b636e8SJosef Bacik u64 features; 24603de4586cSChris Mason struct btrfs_key location; 2461a061fc8dSChris Mason struct buffer_head *bh; 24624d34b278SIlya Dryomov struct btrfs_super_block *disk_super; 2463815745cfSAl Viro struct btrfs_fs_info *fs_info = btrfs_sb(sb); 2464f84a8bd6SAl Viro struct btrfs_root *tree_root; 24654d34b278SIlya Dryomov struct btrfs_root *chunk_root; 2466eb60ceacSChris Mason int ret; 2467e58ca020SYan int err = -EINVAL; 2468af31f5e5SChris Mason int num_backups_tried = 0; 2469af31f5e5SChris Mason int backup_index = 0; 24705cdc7ad3SQu Wenruo int max_active; 24714543df7eSChris Mason 2472f84a8bd6SAl Viro tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info); 24736f07e42eSAl Viro chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info); 2474cb517eabSMiao Xie if (!tree_root || !chunk_root) { 247539279cc3SChris Mason err = -ENOMEM; 247639279cc3SChris Mason goto fail; 247739279cc3SChris Mason } 247876dda93cSYan, Zheng 247976dda93cSYan, Zheng ret = init_srcu_struct(&fs_info->subvol_srcu); 248076dda93cSYan, Zheng if (ret) { 248176dda93cSYan, Zheng err = ret; 248276dda93cSYan, Zheng goto fail; 248376dda93cSYan, Zheng } 248476dda93cSYan, Zheng 248576dda93cSYan, Zheng ret = setup_bdi(fs_info, &fs_info->bdi); 248676dda93cSYan, Zheng if (ret) { 248776dda93cSYan, Zheng err = ret; 248876dda93cSYan, Zheng goto fail_srcu; 248976dda93cSYan, Zheng } 249076dda93cSYan, Zheng 2491908c7f19STejun Heo ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 2492e2d84521SMiao Xie if (ret) { 2493e2d84521SMiao Xie err = ret; 2494e2d84521SMiao Xie goto fail_bdi; 2495e2d84521SMiao Xie } 2496e2d84521SMiao Xie fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE * 2497e2d84521SMiao Xie (1 + ilog2(nr_cpu_ids)); 2498e2d84521SMiao Xie 2499908c7f19STejun Heo ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 2500963d678bSMiao Xie if (ret) { 2501963d678bSMiao Xie err = ret; 2502963d678bSMiao Xie goto fail_dirty_metadata_bytes; 2503963d678bSMiao Xie } 2504963d678bSMiao Xie 2505908c7f19STejun Heo ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL); 2506c404e0dcSMiao Xie if (ret) { 2507c404e0dcSMiao Xie err = ret; 2508c404e0dcSMiao Xie goto fail_delalloc_bytes; 2509c404e0dcSMiao Xie } 2510c404e0dcSMiao Xie 251176dda93cSYan, Zheng fs_info->btree_inode = new_inode(sb); 251276dda93cSYan, Zheng if (!fs_info->btree_inode) { 251376dda93cSYan, Zheng err = -ENOMEM; 2514c404e0dcSMiao Xie goto fail_bio_counter; 251576dda93cSYan, Zheng } 251676dda93cSYan, Zheng 2517a6591715SChris Mason mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 25181561dedaSMiao Xie 251976dda93cSYan, Zheng INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 2520f28491e0SJosef Bacik INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 25218fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 2522facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 252324bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 2524eb73c1b7SMiao Xie INIT_LIST_HEAD(&fs_info->delalloc_roots); 252511833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 2526eb73c1b7SMiao Xie spin_lock_init(&fs_info->delalloc_root_lock); 2527a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 252876dda93cSYan, Zheng spin_lock_init(&fs_info->fs_roots_radix_lock); 252924bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 25304cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 25312bf64758SJosef Bacik spin_lock_init(&fs_info->free_chunk_lock); 2532f29021b2SJan Schmidt spin_lock_init(&fs_info->tree_mod_seq_lock); 2533ceda0864SMiao Xie spin_lock_init(&fs_info->super_lock); 2534fcebe456SJosef Bacik spin_lock_init(&fs_info->qgroup_op_lock); 2535f28491e0SJosef Bacik spin_lock_init(&fs_info->buffer_lock); 253647ab2a6cSJosef Bacik spin_lock_init(&fs_info->unused_bgs_lock); 2537f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 2538d7c15171SZhao Lei mutex_init(&fs_info->unused_bg_unpin_mutex); 253967c5e7d4SFilipe Manana mutex_init(&fs_info->delete_unused_bgs_mutex); 25407585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 2541573bfb72SMiao Xie mutex_init(&fs_info->delalloc_root_mutex); 2542de98ced9SMiao Xie seqlock_init(&fs_info->profiles_lock); 2543d7c15171SZhao Lei init_rwsem(&fs_info->delayed_iput_sem); 254419c00ddcSChris Mason 25450b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 25466324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 2547f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 254847ab2a6cSJosef Bacik INIT_LIST_HEAD(&fs_info->unused_bgs); 25490b86a832SChris Mason btrfs_mapping_init(&fs_info->mapping_tree); 255066d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 255166d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 255266d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delalloc_block_rsv, 255366d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELALLOC); 255466d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 255566d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 255666d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 255766d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 255866d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 2559cb03c743SChris Mason atomic_set(&fs_info->nr_async_submits, 0); 2560771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 25618c8bee1dSChris Mason atomic_set(&fs_info->async_submit_draining, 0); 25620986fe9eSChris Mason atomic_set(&fs_info->nr_async_bios, 0); 25634cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 2564fcebe456SJosef Bacik atomic_set(&fs_info->qgroup_op_seq, 0); 2565fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 2566e20d96d6SChris Mason fs_info->sb = sb; 256795ac567aSFilipe David Borba Manana fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 25689ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 25694cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 25702bf64758SJosef Bacik fs_info->free_chunk_space = 0; 2571f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 25728b87dc17SDavid Sterba fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 2573f8c269d7SDavid Sterba fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 257490519d66SArne Jansen /* readahead state */ 257590519d66SArne Jansen INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT); 257690519d66SArne Jansen spin_lock_init(&fs_info->reada_lock); 2577c8b97818SChris Mason 2578b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 2579b34b086cSChris Mason num_online_cpus() + 2, 8); 25800afbaf8cSChris Mason 2581199c2a9cSMiao Xie INIT_LIST_HEAD(&fs_info->ordered_roots); 2582199c2a9cSMiao Xie spin_lock_init(&fs_info->ordered_root_lock); 258316cdcec7SMiao Xie fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 258416cdcec7SMiao Xie GFP_NOFS); 258516cdcec7SMiao Xie if (!fs_info->delayed_root) { 258616cdcec7SMiao Xie err = -ENOMEM; 258716cdcec7SMiao Xie goto fail_iput; 258816cdcec7SMiao Xie } 258916cdcec7SMiao Xie btrfs_init_delayed_root(fs_info->delayed_root); 25903eaa2885SChris Mason 2591638aa7edSEric Sandeen btrfs_init_scrub(fs_info); 259221adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 259321adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 259421adbd5cSStefan Behrens #endif 2595779a65a4SEric Sandeen btrfs_init_balance(fs_info); 259621c7e756SMiao Xie btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work); 2597a2de733cSArne Jansen 2598a061fc8dSChris Mason sb->s_blocksize = 4096; 2599a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(4096); 260032a88aa1SJens Axboe sb->s_bdi = &fs_info->bdi; 2601a061fc8dSChris Mason 2602f37938e0SEric Sandeen btrfs_init_btree_inode(fs_info, tree_root); 260339279cc3SChris Mason 2604e02119d5SChris Mason spin_lock_init(&fs_info->block_group_cache_lock); 26056bef4d31SEric Paris fs_info->block_group_cache_tree = RB_ROOT; 2606a1897fddSLiu Bo fs_info->first_logical_byte = (u64)-1; 2607925baeddSChris Mason 260811833d66SYan Zheng extent_io_tree_init(&fs_info->freed_extents[0], 2609f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 261011833d66SYan Zheng extent_io_tree_init(&fs_info->freed_extents[1], 2611f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 261211833d66SYan Zheng fs_info->pinned_extents = &fs_info->freed_extents[0]; 2613e6dcd2dcSChris Mason fs_info->do_barriers = 1; 2614f9295749SChris Mason 2615d98237b3SChris Mason 26165a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 261730ae8467SChris Mason mutex_init(&fs_info->tree_log_mutex); 2618925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 2619925baeddSChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 2620a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 2621a74a4b97SChris Mason mutex_init(&fs_info->volume_mutex); 26221bbc621eSChris Mason mutex_init(&fs_info->ro_block_group_mutex); 26239e351cc8SJosef Bacik init_rwsem(&fs_info->commit_root_sem); 2624c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 262576dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 2626803b2f54SStefan Behrens sema_init(&fs_info->uuid_tree_rescan_sem, 1); 2627fa9c0d79SChris Mason 2628ad618368SEric Sandeen btrfs_init_dev_replace_locks(fs_info); 2629f9e92e40SEric Sandeen btrfs_init_qgroup(fs_info); 2630416ac51dSArne Jansen 2631fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2632fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2633fa9c0d79SChris Mason 26347d9eb12cSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 26353768f368SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 2636bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 26374854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 26389a8dd150SChris Mason 263904216820SFilipe Manana INIT_LIST_HEAD(&fs_info->pinned_chunks); 264004216820SFilipe Manana 264153b381b3SDavid Woodhouse ret = btrfs_alloc_stripe_hash_table(fs_info); 264253b381b3SDavid Woodhouse if (ret) { 264383c8266aSDavid Sterba err = ret; 264453b381b3SDavid Woodhouse goto fail_alloc; 264553b381b3SDavid Woodhouse } 264653b381b3SDavid Woodhouse 2647707e8a07SDavid Sterba __setup_root(4096, 4096, 4096, tree_root, 26482c90e5d6SChris Mason fs_info, BTRFS_ROOT_TREE_OBJECTID); 26497eccb903SChris Mason 26503c4bb26bSChris Mason invalidate_bdev(fs_devices->latest_bdev); 26511104a885SDavid Sterba 26521104a885SDavid Sterba /* 26531104a885SDavid Sterba * Read super block and check the signature bytes only 26541104a885SDavid Sterba */ 2655a512bbf8SYan Zheng bh = btrfs_read_dev_super(fs_devices->latest_bdev); 265620b45077SDave Young if (!bh) { 265720b45077SDave Young err = -EINVAL; 265816cdcec7SMiao Xie goto fail_alloc; 265920b45077SDave Young } 266039279cc3SChris Mason 26611104a885SDavid Sterba /* 26621104a885SDavid Sterba * We want to check superblock checksum, the type is stored inside. 26631104a885SDavid Sterba * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 26641104a885SDavid Sterba */ 26651104a885SDavid Sterba if (btrfs_check_super_csum(bh->b_data)) { 2666efe120a0SFrank Holton printk(KERN_ERR "BTRFS: superblock checksum mismatch\n"); 26671104a885SDavid Sterba err = -EINVAL; 26681104a885SDavid Sterba goto fail_alloc; 26691104a885SDavid Sterba } 26701104a885SDavid Sterba 26711104a885SDavid Sterba /* 26721104a885SDavid Sterba * super_copy is zeroed at allocation time and we never touch the 26731104a885SDavid Sterba * following bytes up to INFO_SIZE, the checksum is calculated from 26741104a885SDavid Sterba * the whole block of INFO_SIZE 26751104a885SDavid Sterba */ 26766c41761fSDavid Sterba memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy)); 26776c41761fSDavid Sterba memcpy(fs_info->super_for_commit, fs_info->super_copy, 26786c41761fSDavid Sterba sizeof(*fs_info->super_for_commit)); 2679a061fc8dSChris Mason brelse(bh); 26805f39d397SChris Mason 26816c41761fSDavid Sterba memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE); 26820b86a832SChris Mason 26831104a885SDavid Sterba ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY); 26841104a885SDavid Sterba if (ret) { 2685efe120a0SFrank Holton printk(KERN_ERR "BTRFS: superblock contains fatal errors\n"); 26861104a885SDavid Sterba err = -EINVAL; 26871104a885SDavid Sterba goto fail_alloc; 26881104a885SDavid Sterba } 26891104a885SDavid Sterba 26906c41761fSDavid Sterba disk_super = fs_info->super_copy; 26910f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 269216cdcec7SMiao Xie goto fail_alloc; 26930f7d52f4SChris Mason 2694acce952bSliubo /* check FS state, whether FS is broken. */ 269587533c47SMiao Xie if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 269687533c47SMiao Xie set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 2697acce952bSliubo 269875e7cb7fSLiu Bo /* 2699af31f5e5SChris Mason * run through our array of backup supers and setup 2700af31f5e5SChris Mason * our ring pointer to the oldest one 2701af31f5e5SChris Mason */ 2702af31f5e5SChris Mason generation = btrfs_super_generation(disk_super); 2703af31f5e5SChris Mason find_oldest_super_backup(fs_info, generation); 2704af31f5e5SChris Mason 2705af31f5e5SChris Mason /* 270675e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 270775e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 270875e7cb7fSLiu Bo */ 270975e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 271075e7cb7fSLiu Bo 27112b82032cSYan Zheng ret = btrfs_parse_options(tree_root, options); 27122b82032cSYan Zheng if (ret) { 27132b82032cSYan Zheng err = ret; 271416cdcec7SMiao Xie goto fail_alloc; 27152b82032cSYan Zheng } 2716dfe25020SChris Mason 2717f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 2718f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 2719f2b636e8SJosef Bacik if (features) { 2720f2b636e8SJosef Bacik printk(KERN_ERR "BTRFS: couldn't mount because of " 2721f2b636e8SJosef Bacik "unsupported optional features (%Lx).\n", 2722c1c9ff7cSGeert Uytterhoeven features); 2723f2b636e8SJosef Bacik err = -EINVAL; 272416cdcec7SMiao Xie goto fail_alloc; 2725f2b636e8SJosef Bacik } 2726f2b636e8SJosef Bacik 2727707e8a07SDavid Sterba /* 2728707e8a07SDavid Sterba * Leafsize and nodesize were always equal, this is only a sanity check. 2729707e8a07SDavid Sterba */ 2730707e8a07SDavid Sterba if (le32_to_cpu(disk_super->__unused_leafsize) != 2731727011e0SChris Mason btrfs_super_nodesize(disk_super)) { 2732727011e0SChris Mason printk(KERN_ERR "BTRFS: couldn't mount because metadata " 2733727011e0SChris Mason "blocksizes don't match. node %d leaf %d\n", 2734727011e0SChris Mason btrfs_super_nodesize(disk_super), 2735707e8a07SDavid Sterba le32_to_cpu(disk_super->__unused_leafsize)); 2736727011e0SChris Mason err = -EINVAL; 2737727011e0SChris Mason goto fail_alloc; 2738727011e0SChris Mason } 2739707e8a07SDavid Sterba if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) { 2740727011e0SChris Mason printk(KERN_ERR "BTRFS: couldn't mount because metadata " 2741727011e0SChris Mason "blocksize (%d) was too large\n", 2742707e8a07SDavid Sterba btrfs_super_nodesize(disk_super)); 2743727011e0SChris Mason err = -EINVAL; 2744727011e0SChris Mason goto fail_alloc; 2745727011e0SChris Mason } 2746727011e0SChris Mason 27475d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 27485d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 274969e380d1SLi Zefan if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO) 2750a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 2751727011e0SChris Mason 27523173a18fSJosef Bacik if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) 27535efa0490SDavid Sterba printk(KERN_INFO "BTRFS: has skinny extents\n"); 27543173a18fSJosef Bacik 2755727011e0SChris Mason /* 2756727011e0SChris Mason * flag our filesystem as having big metadata blocks if 2757727011e0SChris Mason * they are bigger than the page size 2758727011e0SChris Mason */ 2759707e8a07SDavid Sterba if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) { 2760727011e0SChris Mason if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 2761efe120a0SFrank Holton printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n"); 2762727011e0SChris Mason features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 2763727011e0SChris Mason } 2764727011e0SChris Mason 2765bc3f116fSChris Mason nodesize = btrfs_super_nodesize(disk_super); 2766bc3f116fSChris Mason sectorsize = btrfs_super_sectorsize(disk_super); 2767bc3f116fSChris Mason stripesize = btrfs_super_stripesize(disk_super); 2768707e8a07SDavid Sterba fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 2769963d678bSMiao Xie fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 2770bc3f116fSChris Mason 2771bc3f116fSChris Mason /* 2772bc3f116fSChris Mason * mixed block groups end up with duplicate but slightly offset 2773bc3f116fSChris Mason * extent buffers for the same range. It leads to corruptions 2774bc3f116fSChris Mason */ 2775bc3f116fSChris Mason if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 2776707e8a07SDavid Sterba (sectorsize != nodesize)) { 2777aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: unequal leaf/node/sector sizes " 2778bc3f116fSChris Mason "are not allowed for mixed block groups on %s\n", 2779bc3f116fSChris Mason sb->s_id); 2780bc3f116fSChris Mason goto fail_alloc; 2781bc3f116fSChris Mason } 2782bc3f116fSChris Mason 2783ceda0864SMiao Xie /* 2784ceda0864SMiao Xie * Needn't use the lock because there is no other task which will 2785ceda0864SMiao Xie * update the flag. 2786ceda0864SMiao Xie */ 27875d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 27885d4f98a2SYan Zheng 2789f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 2790f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 2791f2b636e8SJosef Bacik if (!(sb->s_flags & MS_RDONLY) && features) { 2792f2b636e8SJosef Bacik printk(KERN_ERR "BTRFS: couldn't mount RDWR because of " 2793f2b636e8SJosef Bacik "unsupported option features (%Lx).\n", 2794c1c9ff7cSGeert Uytterhoeven features); 2795f2b636e8SJosef Bacik err = -EINVAL; 279616cdcec7SMiao Xie goto fail_alloc; 2797f2b636e8SJosef Bacik } 279861d92c32SChris Mason 27995cdc7ad3SQu Wenruo max_active = fs_info->thread_pool_size; 280061d92c32SChris Mason 28012a458198SEric Sandeen ret = btrfs_init_workqueues(fs_info, fs_devices); 28022a458198SEric Sandeen if (ret) { 28032a458198SEric Sandeen err = ret; 28040dc3b84aSJosef Bacik goto fail_sb_buffer; 28050dc3b84aSJosef Bacik } 28064543df7eSChris Mason 28074575c9ccSChris Mason fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super); 2808c8b97818SChris Mason fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages, 2809c8b97818SChris Mason 4 * 1024 * 1024 / PAGE_CACHE_SIZE); 28104575c9ccSChris Mason 2811db94535dSChris Mason tree_root->nodesize = nodesize; 2812db94535dSChris Mason tree_root->sectorsize = sectorsize; 281387ee04ebSChris Mason tree_root->stripesize = stripesize; 2814a061fc8dSChris Mason 2815a061fc8dSChris Mason sb->s_blocksize = sectorsize; 2816a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 2817db94535dSChris Mason 28183cae210fSQu Wenruo if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) { 2819aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: valid FS not found on %s\n", sb->s_id); 282039279cc3SChris Mason goto fail_sb_buffer; 282139279cc3SChris Mason } 282219c00ddcSChris Mason 28238d082fb7SLiu Bo if (sectorsize != PAGE_SIZE) { 2824aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: incompatible sector size (%lu) " 28258d082fb7SLiu Bo "found on %s\n", (unsigned long)sectorsize, sb->s_id); 2826941b2ddfSKeith Mannthey goto fail_sb_buffer; 2827941b2ddfSKeith Mannthey } 2828941b2ddfSKeith Mannthey 2829925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 2830e4404d6eSYan Zheng ret = btrfs_read_sys_array(tree_root); 2831925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 283284eed90fSChris Mason if (ret) { 2833aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: failed to read the system " 2834d397712bSChris Mason "array on %s\n", sb->s_id); 28355d4f98a2SYan Zheng goto fail_sb_buffer; 283684eed90fSChris Mason } 28370b86a832SChris Mason 283884234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 28390b86a832SChris Mason 2840707e8a07SDavid Sterba __setup_root(nodesize, sectorsize, stripesize, chunk_root, 2841707e8a07SDavid Sterba fs_info, BTRFS_CHUNK_TREE_OBJECTID); 28420b86a832SChris Mason 28430b86a832SChris Mason chunk_root->node = read_tree_block(chunk_root, 28440b86a832SChris Mason btrfs_super_chunk_root(disk_super), 2845ce86cd59SDavid Sterba generation); 284664c043deSLiu Bo if (IS_ERR(chunk_root->node) || 284764c043deSLiu Bo !extent_buffer_uptodate(chunk_root->node)) { 2848aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: failed to read chunk root on %s\n", 284983121942SDavid Woodhouse sb->s_id); 2850e5fffbacSchandan if (!IS_ERR(chunk_root->node)) 2851e5fffbacSchandan free_extent_buffer(chunk_root->node); 285295ab1f64SZhao Lei chunk_root->node = NULL; 2853af31f5e5SChris Mason goto fail_tree_roots; 285483121942SDavid Woodhouse } 28555d4f98a2SYan Zheng btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 28565d4f98a2SYan Zheng chunk_root->commit_root = btrfs_root_node(chunk_root); 28570b86a832SChris Mason 2858e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 2859b308bc2fSGeert Uytterhoeven btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE); 2860e17cade2SChris Mason 28610b86a832SChris Mason ret = btrfs_read_chunk_tree(chunk_root); 28622b82032cSYan Zheng if (ret) { 2863aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: failed to read chunk tree on %s\n", 2864d397712bSChris Mason sb->s_id); 2865af31f5e5SChris Mason goto fail_tree_roots; 28662b82032cSYan Zheng } 28670b86a832SChris Mason 28688dabb742SStefan Behrens /* 28698dabb742SStefan Behrens * keep the device that is marked to be the target device for the 28708dabb742SStefan Behrens * dev_replace procedure 28718dabb742SStefan Behrens */ 28729eaed21eSEric Sandeen btrfs_close_extra_devices(fs_devices, 0); 2873dfe25020SChris Mason 2874a6b0d5c8SChris Mason if (!fs_devices->latest_bdev) { 2875aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: failed to read devices on %s\n", 2876a6b0d5c8SChris Mason sb->s_id); 2877a6b0d5c8SChris Mason goto fail_tree_roots; 2878a6b0d5c8SChris Mason } 2879a6b0d5c8SChris Mason 2880af31f5e5SChris Mason retry_root_backup: 288184234f3aSYan Zheng generation = btrfs_super_generation(disk_super); 28820b86a832SChris Mason 2883e20d96d6SChris Mason tree_root->node = read_tree_block(tree_root, 2884db94535dSChris Mason btrfs_super_root(disk_super), 2885ce86cd59SDavid Sterba generation); 288664c043deSLiu Bo if (IS_ERR(tree_root->node) || 288764c043deSLiu Bo !extent_buffer_uptodate(tree_root->node)) { 2888efe120a0SFrank Holton printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n", 288983121942SDavid Woodhouse sb->s_id); 2890e5fffbacSchandan if (!IS_ERR(tree_root->node)) 2891e5fffbacSchandan free_extent_buffer(tree_root->node); 289295ab1f64SZhao Lei tree_root->node = NULL; 2893af31f5e5SChris Mason goto recovery_tree_root; 289483121942SDavid Woodhouse } 2895af31f5e5SChris Mason 28965d4f98a2SYan Zheng btrfs_set_root_node(&tree_root->root_item, tree_root->node); 28975d4f98a2SYan Zheng tree_root->commit_root = btrfs_root_node(tree_root); 289869e9c6c6SStefan Behrens btrfs_set_root_refs(&tree_root->root_item, 1); 2899db94535dSChris Mason 29004bbcaa64SEric Sandeen ret = btrfs_read_roots(fs_info, tree_root); 29014bbcaa64SEric Sandeen if (ret) 2902af31f5e5SChris Mason goto recovery_tree_root; 2903f7a81ea4SStefan Behrens 29048929ecfaSYan, Zheng fs_info->generation = generation; 29058929ecfaSYan, Zheng fs_info->last_trans_committed = generation; 29068929ecfaSYan, Zheng 290768310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 290868310a5eSIlya Dryomov if (ret) { 2909aa8ee312SDavid Sterba printk(KERN_ERR "BTRFS: failed to recover balance\n"); 291068310a5eSIlya Dryomov goto fail_block_groups; 291168310a5eSIlya Dryomov } 291268310a5eSIlya Dryomov 2913733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 2914733f4fbbSStefan Behrens if (ret) { 2915efe120a0SFrank Holton printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n", 2916733f4fbbSStefan Behrens ret); 2917733f4fbbSStefan Behrens goto fail_block_groups; 2918733f4fbbSStefan Behrens } 2919733f4fbbSStefan Behrens 29208dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 29218dabb742SStefan Behrens if (ret) { 2922efe120a0SFrank Holton pr_err("BTRFS: failed to init dev_replace: %d\n", ret); 29238dabb742SStefan Behrens goto fail_block_groups; 29248dabb742SStefan Behrens } 29258dabb742SStefan Behrens 29269eaed21eSEric Sandeen btrfs_close_extra_devices(fs_devices, 1); 29278dabb742SStefan Behrens 2928b7c35e81SAnand Jain ret = btrfs_sysfs_add_fsid(fs_devices, NULL); 2929b7c35e81SAnand Jain if (ret) { 2930b7c35e81SAnand Jain pr_err("BTRFS: failed to init sysfs fsid interface: %d\n", ret); 2931b7c35e81SAnand Jain goto fail_block_groups; 2932b7c35e81SAnand Jain } 2933b7c35e81SAnand Jain 2934b7c35e81SAnand Jain ret = btrfs_sysfs_add_device(fs_devices); 2935b7c35e81SAnand Jain if (ret) { 2936b7c35e81SAnand Jain pr_err("BTRFS: failed to init sysfs device interface: %d\n", ret); 2937b7c35e81SAnand Jain goto fail_fsdev_sysfs; 2938b7c35e81SAnand Jain } 2939b7c35e81SAnand Jain 29405ac1d209SJeff Mahoney ret = btrfs_sysfs_add_one(fs_info); 29415ac1d209SJeff Mahoney if (ret) { 2942efe120a0SFrank Holton pr_err("BTRFS: failed to init sysfs interface: %d\n", ret); 2943b7c35e81SAnand Jain goto fail_fsdev_sysfs; 29445ac1d209SJeff Mahoney } 29455ac1d209SJeff Mahoney 2946c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 2947c59021f8Sliubo if (ret) { 2948efe120a0SFrank Holton printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret); 29492365dd3cSAnand Jain goto fail_sysfs; 2950c59021f8Sliubo } 2951c59021f8Sliubo 29524bbcaa64SEric Sandeen ret = btrfs_read_block_groups(fs_info->extent_root); 29531b1d1f66SJosef Bacik if (ret) { 2954efe120a0SFrank Holton printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret); 29552365dd3cSAnand Jain goto fail_sysfs; 29561b1d1f66SJosef Bacik } 29575af3e8ccSStefan Behrens fs_info->num_tolerated_disk_barrier_failures = 29585af3e8ccSStefan Behrens btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); 2959292fd7fcSStefan Behrens if (fs_info->fs_devices->missing_devices > 2960292fd7fcSStefan Behrens fs_info->num_tolerated_disk_barrier_failures && 2961292fd7fcSStefan Behrens !(sb->s_flags & MS_RDONLY)) { 296278fa1770SZhao Lei pr_warn("BTRFS: missing devices(%llu) exceeds the limit(%d), writeable mount is not allowed\n", 296378fa1770SZhao Lei fs_info->fs_devices->missing_devices, 296478fa1770SZhao Lei fs_info->num_tolerated_disk_barrier_failures); 29652365dd3cSAnand Jain goto fail_sysfs; 2966292fd7fcSStefan Behrens } 29679078a3e1SChris Mason 2968a74a4b97SChris Mason fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 2969a74a4b97SChris Mason "btrfs-cleaner"); 297057506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 29712365dd3cSAnand Jain goto fail_sysfs; 2972a74a4b97SChris Mason 2973a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 2974a74a4b97SChris Mason tree_root, 2975a74a4b97SChris Mason "btrfs-transaction"); 297657506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 29773f157a2fSChris Mason goto fail_cleaner; 2978a74a4b97SChris Mason 2979c289811cSChris Mason if (!btrfs_test_opt(tree_root, SSD) && 2980c289811cSChris Mason !btrfs_test_opt(tree_root, NOSSD) && 2981c289811cSChris Mason !fs_info->fs_devices->rotating) { 2982efe120a0SFrank Holton printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD " 2983c289811cSChris Mason "mode\n"); 2984c289811cSChris Mason btrfs_set_opt(fs_info->mount_opt, SSD); 2985c289811cSChris Mason } 2986c289811cSChris Mason 2987572d9ab7SDavid Sterba /* 2988572d9ab7SDavid Sterba * Mount does not set all options immediatelly, we can do it now and do 2989572d9ab7SDavid Sterba * not have to wait for transaction commit 2990572d9ab7SDavid Sterba */ 2991572d9ab7SDavid Sterba btrfs_apply_pending_changes(fs_info); 29923818aea2SQu Wenruo 299321adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 299421adbd5cSStefan Behrens if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) { 299521adbd5cSStefan Behrens ret = btrfsic_mount(tree_root, fs_devices, 299621adbd5cSStefan Behrens btrfs_test_opt(tree_root, 299721adbd5cSStefan Behrens CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? 299821adbd5cSStefan Behrens 1 : 0, 299921adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 300021adbd5cSStefan Behrens if (ret) 3001efe120a0SFrank Holton printk(KERN_WARNING "BTRFS: failed to initialize" 300221adbd5cSStefan Behrens " integrity check module %s\n", sb->s_id); 300321adbd5cSStefan Behrens } 300421adbd5cSStefan Behrens #endif 3005bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 3006bcef60f2SArne Jansen if (ret) 3007bcef60f2SArne Jansen goto fail_trans_kthread; 300821adbd5cSStefan Behrens 3009acce952bSliubo /* do not make disk changes in broken FS */ 301068ce9682SStefan Behrens if (btrfs_super_log_root(disk_super) != 0) { 301163443bf5SEric Sandeen ret = btrfs_replay_log(fs_info, fs_devices); 301279787eaaSJeff Mahoney if (ret) { 301363443bf5SEric Sandeen err = ret; 301428c16cbbSWang Shilong goto fail_qgroup; 3015e556ce2cSYan Zheng } 3016e02119d5SChris Mason } 30171a40e23bSZheng Yan 301876dda93cSYan, Zheng ret = btrfs_find_orphan_roots(tree_root); 301979787eaaSJeff Mahoney if (ret) 302028c16cbbSWang Shilong goto fail_qgroup; 302176dda93cSYan, Zheng 30227c2ca468SChris Mason if (!(sb->s_flags & MS_RDONLY)) { 3023d68fc57bSYan, Zheng ret = btrfs_cleanup_fs_roots(fs_info); 302444c44af2SIlya Dryomov if (ret) 302528c16cbbSWang Shilong goto fail_qgroup; 3026d68fc57bSYan, Zheng 30275f316481SWang Shilong mutex_lock(&fs_info->cleaner_mutex); 30285d4f98a2SYan Zheng ret = btrfs_recover_relocation(tree_root); 30295f316481SWang Shilong mutex_unlock(&fs_info->cleaner_mutex); 3030d7ce5843SMiao Xie if (ret < 0) { 3031d7ce5843SMiao Xie printk(KERN_WARNING 3032efe120a0SFrank Holton "BTRFS: failed to recover relocation\n"); 3033d7ce5843SMiao Xie err = -EINVAL; 3034bcef60f2SArne Jansen goto fail_qgroup; 3035d7ce5843SMiao Xie } 30367c2ca468SChris Mason } 30371a40e23bSZheng Yan 30383de4586cSChris Mason location.objectid = BTRFS_FS_TREE_OBJECTID; 30393de4586cSChris Mason location.type = BTRFS_ROOT_ITEM_KEY; 3040cb517eabSMiao Xie location.offset = 0; 30413de4586cSChris Mason 30423de4586cSChris Mason fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location); 30433140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 30443140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 3045bcef60f2SArne Jansen goto fail_qgroup; 30463140c9a3SDan Carpenter } 3047c289811cSChris Mason 30482b6ba629SIlya Dryomov if (sb->s_flags & MS_RDONLY) 30492b6ba629SIlya Dryomov return 0; 30502b6ba629SIlya Dryomov 3051e3acc2a6SJosef Bacik down_read(&fs_info->cleanup_work_sem); 30522b6ba629SIlya Dryomov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 30532b6ba629SIlya Dryomov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 30542b6ba629SIlya Dryomov up_read(&fs_info->cleanup_work_sem); 30552b6ba629SIlya Dryomov close_ctree(tree_root); 30562b6ba629SIlya Dryomov return ret; 30572b6ba629SIlya Dryomov } 3058e3acc2a6SJosef Bacik up_read(&fs_info->cleanup_work_sem); 305959641015SIlya Dryomov 30602b6ba629SIlya Dryomov ret = btrfs_resume_balance_async(fs_info); 30612b6ba629SIlya Dryomov if (ret) { 3062efe120a0SFrank Holton printk(KERN_WARNING "BTRFS: failed to resume balance\n"); 306366b4ffd1SJosef Bacik close_ctree(tree_root); 30642b6ba629SIlya Dryomov return ret; 3065e3acc2a6SJosef Bacik } 3066e3acc2a6SJosef Bacik 30678dabb742SStefan Behrens ret = btrfs_resume_dev_replace_async(fs_info); 30688dabb742SStefan Behrens if (ret) { 3069efe120a0SFrank Holton pr_warn("BTRFS: failed to resume dev_replace\n"); 30708dabb742SStefan Behrens close_ctree(tree_root); 30718dabb742SStefan Behrens return ret; 30728dabb742SStefan Behrens } 30738dabb742SStefan Behrens 3074b382a324SJan Schmidt btrfs_qgroup_rescan_resume(fs_info); 3075b382a324SJan Schmidt 30764bbcaa64SEric Sandeen if (!fs_info->uuid_root) { 3077efe120a0SFrank Holton pr_info("BTRFS: creating UUID tree\n"); 3078f7a81ea4SStefan Behrens ret = btrfs_create_uuid_tree(fs_info); 3079f7a81ea4SStefan Behrens if (ret) { 3080efe120a0SFrank Holton pr_warn("BTRFS: failed to create the UUID tree %d\n", 3081f7a81ea4SStefan Behrens ret); 3082f7a81ea4SStefan Behrens close_ctree(tree_root); 3083f7a81ea4SStefan Behrens return ret; 3084f7a81ea4SStefan Behrens } 30854bbcaa64SEric Sandeen } else if (btrfs_test_opt(tree_root, RESCAN_UUID_TREE) || 30864bbcaa64SEric Sandeen fs_info->generation != 30874bbcaa64SEric Sandeen btrfs_super_uuid_tree_generation(disk_super)) { 3088efe120a0SFrank Holton pr_info("BTRFS: checking UUID tree\n"); 308970f80175SStefan Behrens ret = btrfs_check_uuid_tree(fs_info); 309070f80175SStefan Behrens if (ret) { 3091efe120a0SFrank Holton pr_warn("BTRFS: failed to check the UUID tree %d\n", 309270f80175SStefan Behrens ret); 309370f80175SStefan Behrens close_ctree(tree_root); 309470f80175SStefan Behrens return ret; 309570f80175SStefan Behrens } 309670f80175SStefan Behrens } else { 309770f80175SStefan Behrens fs_info->update_uuid_tree_gen = 1; 3098f7a81ea4SStefan Behrens } 3099f7a81ea4SStefan Behrens 310047ab2a6cSJosef Bacik fs_info->open = 1; 310147ab2a6cSJosef Bacik 3102ad2b2c80SAl Viro return 0; 310339279cc3SChris Mason 3104bcef60f2SArne Jansen fail_qgroup: 3105bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 31067c2ca468SChris Mason fail_trans_kthread: 31077c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 310854067ae9SJosef Bacik btrfs_cleanup_transaction(fs_info->tree_root); 3109faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 31103f157a2fSChris Mason fail_cleaner: 3111a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 31127c2ca468SChris Mason 31137c2ca468SChris Mason /* 31147c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 31157c2ca468SChris Mason * kthreads 31167c2ca468SChris Mason */ 31177c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 31187c2ca468SChris Mason 31192365dd3cSAnand Jain fail_sysfs: 31202365dd3cSAnand Jain btrfs_sysfs_remove_one(fs_info); 31212365dd3cSAnand Jain 3122b7c35e81SAnand Jain fail_fsdev_sysfs: 3123b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3124b7c35e81SAnand Jain 31251b1d1f66SJosef Bacik fail_block_groups: 312654067ae9SJosef Bacik btrfs_put_block_group_cache(fs_info); 31271b1d1f66SJosef Bacik btrfs_free_block_groups(fs_info); 3128af31f5e5SChris Mason 3129af31f5e5SChris Mason fail_tree_roots: 3130af31f5e5SChris Mason free_root_pointers(fs_info, 1); 31312b8195bbSMiao Xie invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3132af31f5e5SChris Mason 313339279cc3SChris Mason fail_sb_buffer: 31347abadb64SLiu Bo btrfs_stop_all_workers(fs_info); 313516cdcec7SMiao Xie fail_alloc: 31364543df7eSChris Mason fail_iput: 3137586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 3138586e46e2SIlya Dryomov 31394543df7eSChris Mason iput(fs_info->btree_inode); 3140c404e0dcSMiao Xie fail_bio_counter: 3141c404e0dcSMiao Xie percpu_counter_destroy(&fs_info->bio_counter); 3142963d678bSMiao Xie fail_delalloc_bytes: 3143963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 3144e2d84521SMiao Xie fail_dirty_metadata_bytes: 3145e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 3146ad081f14SJens Axboe fail_bdi: 31477e662854SQinghuang Feng bdi_destroy(&fs_info->bdi); 314876dda93cSYan, Zheng fail_srcu: 314976dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 31507e662854SQinghuang Feng fail: 315153b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 3152586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 3153ad2b2c80SAl Viro return err; 3154af31f5e5SChris Mason 3155af31f5e5SChris Mason recovery_tree_root: 3156af31f5e5SChris Mason if (!btrfs_test_opt(tree_root, RECOVERY)) 3157af31f5e5SChris Mason goto fail_tree_roots; 3158af31f5e5SChris Mason 3159af31f5e5SChris Mason free_root_pointers(fs_info, 0); 3160af31f5e5SChris Mason 3161af31f5e5SChris Mason /* don't use the log in recovery mode, it won't be valid */ 3162af31f5e5SChris Mason btrfs_set_super_log_root(disk_super, 0); 3163af31f5e5SChris Mason 3164af31f5e5SChris Mason /* we can't trust the free space cache either */ 3165af31f5e5SChris Mason btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 3166af31f5e5SChris Mason 3167af31f5e5SChris Mason ret = next_root_backup(fs_info, fs_info->super_copy, 3168af31f5e5SChris Mason &num_backups_tried, &backup_index); 3169af31f5e5SChris Mason if (ret == -1) 3170af31f5e5SChris Mason goto fail_block_groups; 3171af31f5e5SChris Mason goto retry_root_backup; 3172eb60ceacSChris Mason } 3173eb60ceacSChris Mason 3174f2984462SChris Mason static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate) 3175f2984462SChris Mason { 3176f2984462SChris Mason if (uptodate) { 3177f2984462SChris Mason set_buffer_uptodate(bh); 3178f2984462SChris Mason } else { 3179442a4f63SStefan Behrens struct btrfs_device *device = (struct btrfs_device *) 3180442a4f63SStefan Behrens bh->b_private; 3181442a4f63SStefan Behrens 3182efe120a0SFrank Holton printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to " 3183606686eeSJosef Bacik "I/O error on %s\n", 3184606686eeSJosef Bacik rcu_str_deref(device->name)); 31851259ab75SChris Mason /* note, we dont' set_buffer_write_io_error because we have 31861259ab75SChris Mason * our own ways of dealing with the IO errors 31871259ab75SChris Mason */ 3188f2984462SChris Mason clear_buffer_uptodate(bh); 3189442a4f63SStefan Behrens btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS); 3190f2984462SChris Mason } 3191f2984462SChris Mason unlock_buffer(bh); 3192f2984462SChris Mason put_bh(bh); 3193f2984462SChris Mason } 3194f2984462SChris Mason 3195a512bbf8SYan Zheng struct buffer_head *btrfs_read_dev_super(struct block_device *bdev) 3196a512bbf8SYan Zheng { 3197a512bbf8SYan Zheng struct buffer_head *bh; 3198a512bbf8SYan Zheng struct buffer_head *latest = NULL; 3199a512bbf8SYan Zheng struct btrfs_super_block *super; 3200a512bbf8SYan Zheng int i; 3201a512bbf8SYan Zheng u64 transid = 0; 3202a512bbf8SYan Zheng u64 bytenr; 3203a512bbf8SYan Zheng 3204a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 3205a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 3206a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 3207a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3208a512bbf8SYan Zheng */ 3209a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 3210a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 32118068a47eSAnand Jain if (bytenr + BTRFS_SUPER_INFO_SIZE >= 32128068a47eSAnand Jain i_size_read(bdev->bd_inode)) 3213a512bbf8SYan Zheng break; 32148068a47eSAnand Jain bh = __bread(bdev, bytenr / 4096, 32158068a47eSAnand Jain BTRFS_SUPER_INFO_SIZE); 3216a512bbf8SYan Zheng if (!bh) 3217a512bbf8SYan Zheng continue; 3218a512bbf8SYan Zheng 3219a512bbf8SYan Zheng super = (struct btrfs_super_block *)bh->b_data; 3220a512bbf8SYan Zheng if (btrfs_super_bytenr(super) != bytenr || 32213cae210fSQu Wenruo btrfs_super_magic(super) != BTRFS_MAGIC) { 3222a512bbf8SYan Zheng brelse(bh); 3223a512bbf8SYan Zheng continue; 3224a512bbf8SYan Zheng } 3225a512bbf8SYan Zheng 3226a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 3227a512bbf8SYan Zheng brelse(latest); 3228a512bbf8SYan Zheng latest = bh; 3229a512bbf8SYan Zheng transid = btrfs_super_generation(super); 3230a512bbf8SYan Zheng } else { 3231a512bbf8SYan Zheng brelse(bh); 3232a512bbf8SYan Zheng } 3233a512bbf8SYan Zheng } 3234a512bbf8SYan Zheng return latest; 3235a512bbf8SYan Zheng } 3236a512bbf8SYan Zheng 32374eedeb75SHisashi Hifumi /* 32384eedeb75SHisashi Hifumi * this should be called twice, once with wait == 0 and 32394eedeb75SHisashi Hifumi * once with wait == 1. When wait == 0 is done, all the buffer heads 32404eedeb75SHisashi Hifumi * we write are pinned. 32414eedeb75SHisashi Hifumi * 32424eedeb75SHisashi Hifumi * They are released when wait == 1 is done. 32434eedeb75SHisashi Hifumi * max_mirrors must be the same for both runs, and it indicates how 32444eedeb75SHisashi Hifumi * many supers on this one device should be written. 32454eedeb75SHisashi Hifumi * 32464eedeb75SHisashi Hifumi * max_mirrors == 0 means to write them all. 32474eedeb75SHisashi Hifumi */ 3248a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 3249a512bbf8SYan Zheng struct btrfs_super_block *sb, 3250a512bbf8SYan Zheng int do_barriers, int wait, int max_mirrors) 3251a512bbf8SYan Zheng { 3252a512bbf8SYan Zheng struct buffer_head *bh; 3253a512bbf8SYan Zheng int i; 3254a512bbf8SYan Zheng int ret; 3255a512bbf8SYan Zheng int errors = 0; 3256a512bbf8SYan Zheng u32 crc; 3257a512bbf8SYan Zheng u64 bytenr; 3258a512bbf8SYan Zheng 3259a512bbf8SYan Zheng if (max_mirrors == 0) 3260a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3261a512bbf8SYan Zheng 3262a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 3263a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 3264935e5cc9SMiao Xie if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3265935e5cc9SMiao Xie device->commit_total_bytes) 3266a512bbf8SYan Zheng break; 3267a512bbf8SYan Zheng 3268a512bbf8SYan Zheng if (wait) { 3269a512bbf8SYan Zheng bh = __find_get_block(device->bdev, bytenr / 4096, 3270a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 3271634554dcSJosef Bacik if (!bh) { 3272634554dcSJosef Bacik errors++; 3273634554dcSJosef Bacik continue; 3274634554dcSJosef Bacik } 3275a512bbf8SYan Zheng wait_on_buffer(bh); 32764eedeb75SHisashi Hifumi if (!buffer_uptodate(bh)) 32774eedeb75SHisashi Hifumi errors++; 32784eedeb75SHisashi Hifumi 32794eedeb75SHisashi Hifumi /* drop our reference */ 32804eedeb75SHisashi Hifumi brelse(bh); 32814eedeb75SHisashi Hifumi 32824eedeb75SHisashi Hifumi /* drop the reference from the wait == 0 run */ 3283a512bbf8SYan Zheng brelse(bh); 3284a512bbf8SYan Zheng continue; 3285a512bbf8SYan Zheng } else { 3286a512bbf8SYan Zheng btrfs_set_super_bytenr(sb, bytenr); 3287a512bbf8SYan Zheng 3288a512bbf8SYan Zheng crc = ~(u32)0; 3289b0496686SLiu Bo crc = btrfs_csum_data((char *)sb + 3290a512bbf8SYan Zheng BTRFS_CSUM_SIZE, crc, 3291a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE - 3292a512bbf8SYan Zheng BTRFS_CSUM_SIZE); 3293a512bbf8SYan Zheng btrfs_csum_final(crc, sb->csum); 3294a512bbf8SYan Zheng 32954eedeb75SHisashi Hifumi /* 32964eedeb75SHisashi Hifumi * one reference for us, and we leave it for the 32974eedeb75SHisashi Hifumi * caller 32984eedeb75SHisashi Hifumi */ 3299a512bbf8SYan Zheng bh = __getblk(device->bdev, bytenr / 4096, 3300a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 3301634554dcSJosef Bacik if (!bh) { 3302efe120a0SFrank Holton printk(KERN_ERR "BTRFS: couldn't get super " 3303634554dcSJosef Bacik "buffer head for bytenr %Lu\n", bytenr); 3304634554dcSJosef Bacik errors++; 3305634554dcSJosef Bacik continue; 3306634554dcSJosef Bacik } 3307634554dcSJosef Bacik 3308a512bbf8SYan Zheng memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE); 3309a512bbf8SYan Zheng 33104eedeb75SHisashi Hifumi /* one reference for submit_bh */ 3311a512bbf8SYan Zheng get_bh(bh); 33124eedeb75SHisashi Hifumi 33134eedeb75SHisashi Hifumi set_buffer_uptodate(bh); 3314a512bbf8SYan Zheng lock_buffer(bh); 3315a512bbf8SYan Zheng bh->b_end_io = btrfs_end_buffer_write_sync; 3316442a4f63SStefan Behrens bh->b_private = device; 3317a512bbf8SYan Zheng } 3318a512bbf8SYan Zheng 3319387125fcSChris Mason /* 3320387125fcSChris Mason * we fua the first super. The others we allow 3321387125fcSChris Mason * to go down lazy. 3322387125fcSChris Mason */ 3323e8117c26SWang Shilong if (i == 0) 332421adbd5cSStefan Behrens ret = btrfsic_submit_bh(WRITE_FUA, bh); 3325e8117c26SWang Shilong else 3326e8117c26SWang Shilong ret = btrfsic_submit_bh(WRITE_SYNC, bh); 33274eedeb75SHisashi Hifumi if (ret) 3328a512bbf8SYan Zheng errors++; 3329a512bbf8SYan Zheng } 3330a512bbf8SYan Zheng return errors < i ? 0 : -1; 3331a512bbf8SYan Zheng } 3332a512bbf8SYan Zheng 3333387125fcSChris Mason /* 3334387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 3335387125fcSChris Mason * for the barrier when it is done 3336387125fcSChris Mason */ 33374246a0b6SChristoph Hellwig static void btrfs_end_empty_barrier(struct bio *bio) 3338387125fcSChris Mason { 3339387125fcSChris Mason if (bio->bi_private) 3340387125fcSChris Mason complete(bio->bi_private); 3341387125fcSChris Mason bio_put(bio); 3342387125fcSChris Mason } 3343387125fcSChris Mason 3344387125fcSChris Mason /* 3345387125fcSChris Mason * trigger flushes for one the devices. If you pass wait == 0, the flushes are 3346387125fcSChris Mason * sent down. With wait == 1, it waits for the previous flush. 3347387125fcSChris Mason * 3348387125fcSChris Mason * any device where the flush fails with eopnotsupp are flagged as not-barrier 3349387125fcSChris Mason * capable 3350387125fcSChris Mason */ 3351387125fcSChris Mason static int write_dev_flush(struct btrfs_device *device, int wait) 3352387125fcSChris Mason { 3353387125fcSChris Mason struct bio *bio; 3354387125fcSChris Mason int ret = 0; 3355387125fcSChris Mason 3356387125fcSChris Mason if (device->nobarriers) 3357387125fcSChris Mason return 0; 3358387125fcSChris Mason 3359387125fcSChris Mason if (wait) { 3360387125fcSChris Mason bio = device->flush_bio; 3361387125fcSChris Mason if (!bio) 3362387125fcSChris Mason return 0; 3363387125fcSChris Mason 3364387125fcSChris Mason wait_for_completion(&device->flush_wait); 3365387125fcSChris Mason 33664246a0b6SChristoph Hellwig if (bio->bi_error) { 33674246a0b6SChristoph Hellwig ret = bio->bi_error; 3368442a4f63SStefan Behrens btrfs_dev_stat_inc_and_print(device, 3369442a4f63SStefan Behrens BTRFS_DEV_STAT_FLUSH_ERRS); 3370387125fcSChris Mason } 3371387125fcSChris Mason 3372387125fcSChris Mason /* drop the reference from the wait == 0 run */ 3373387125fcSChris Mason bio_put(bio); 3374387125fcSChris Mason device->flush_bio = NULL; 3375387125fcSChris Mason 3376387125fcSChris Mason return ret; 3377387125fcSChris Mason } 3378387125fcSChris Mason 3379387125fcSChris Mason /* 3380387125fcSChris Mason * one reference for us, and we leave it for the 3381387125fcSChris Mason * caller 3382387125fcSChris Mason */ 33839c017abcSJesper Juhl device->flush_bio = NULL; 33849be3395bSChris Mason bio = btrfs_io_bio_alloc(GFP_NOFS, 0); 3385387125fcSChris Mason if (!bio) 3386387125fcSChris Mason return -ENOMEM; 3387387125fcSChris Mason 3388387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 3389387125fcSChris Mason bio->bi_bdev = device->bdev; 3390387125fcSChris Mason init_completion(&device->flush_wait); 3391387125fcSChris Mason bio->bi_private = &device->flush_wait; 3392387125fcSChris Mason device->flush_bio = bio; 3393387125fcSChris Mason 3394387125fcSChris Mason bio_get(bio); 339521adbd5cSStefan Behrens btrfsic_submit_bio(WRITE_FLUSH, bio); 3396387125fcSChris Mason 3397387125fcSChris Mason return 0; 3398387125fcSChris Mason } 3399387125fcSChris Mason 3400387125fcSChris Mason /* 3401387125fcSChris Mason * send an empty flush down to each device in parallel, 3402387125fcSChris Mason * then wait for them 3403387125fcSChris Mason */ 3404387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 3405387125fcSChris Mason { 3406387125fcSChris Mason struct list_head *head; 3407387125fcSChris Mason struct btrfs_device *dev; 34085af3e8ccSStefan Behrens int errors_send = 0; 34095af3e8ccSStefan Behrens int errors_wait = 0; 3410387125fcSChris Mason int ret; 3411387125fcSChris Mason 3412387125fcSChris Mason /* send down all the barriers */ 3413387125fcSChris Mason head = &info->fs_devices->devices; 3414387125fcSChris Mason list_for_each_entry_rcu(dev, head, dev_list) { 3415f88ba6a2SHidetoshi Seto if (dev->missing) 3416f88ba6a2SHidetoshi Seto continue; 3417387125fcSChris Mason if (!dev->bdev) { 34185af3e8ccSStefan Behrens errors_send++; 3419387125fcSChris Mason continue; 3420387125fcSChris Mason } 3421387125fcSChris Mason if (!dev->in_fs_metadata || !dev->writeable) 3422387125fcSChris Mason continue; 3423387125fcSChris Mason 3424387125fcSChris Mason ret = write_dev_flush(dev, 0); 3425387125fcSChris Mason if (ret) 34265af3e8ccSStefan Behrens errors_send++; 3427387125fcSChris Mason } 3428387125fcSChris Mason 3429387125fcSChris Mason /* wait for all the barriers */ 3430387125fcSChris Mason list_for_each_entry_rcu(dev, head, dev_list) { 3431f88ba6a2SHidetoshi Seto if (dev->missing) 3432f88ba6a2SHidetoshi Seto continue; 3433387125fcSChris Mason if (!dev->bdev) { 34345af3e8ccSStefan Behrens errors_wait++; 3435387125fcSChris Mason continue; 3436387125fcSChris Mason } 3437387125fcSChris Mason if (!dev->in_fs_metadata || !dev->writeable) 3438387125fcSChris Mason continue; 3439387125fcSChris Mason 3440387125fcSChris Mason ret = write_dev_flush(dev, 1); 3441387125fcSChris Mason if (ret) 34425af3e8ccSStefan Behrens errors_wait++; 3443387125fcSChris Mason } 34445af3e8ccSStefan Behrens if (errors_send > info->num_tolerated_disk_barrier_failures || 34455af3e8ccSStefan Behrens errors_wait > info->num_tolerated_disk_barrier_failures) 3446387125fcSChris Mason return -EIO; 3447387125fcSChris Mason return 0; 3448387125fcSChris Mason } 3449387125fcSChris Mason 3450943c6e99SZhao Lei int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 3451943c6e99SZhao Lei { 34528789f4feSZhao Lei int raid_type; 34538789f4feSZhao Lei int min_tolerated = INT_MAX; 3454943c6e99SZhao Lei 34558789f4feSZhao Lei if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 34568789f4feSZhao Lei (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 34578789f4feSZhao Lei min_tolerated = min(min_tolerated, 34588789f4feSZhao Lei btrfs_raid_array[BTRFS_RAID_SINGLE]. 34598789f4feSZhao Lei tolerated_failures); 3460943c6e99SZhao Lei 34618789f4feSZhao Lei for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 34628789f4feSZhao Lei if (raid_type == BTRFS_RAID_SINGLE) 34638789f4feSZhao Lei continue; 34648789f4feSZhao Lei if (!(flags & btrfs_raid_group[raid_type])) 34658789f4feSZhao Lei continue; 34668789f4feSZhao Lei min_tolerated = min(min_tolerated, 34678789f4feSZhao Lei btrfs_raid_array[raid_type]. 34688789f4feSZhao Lei tolerated_failures); 34698789f4feSZhao Lei } 3470943c6e99SZhao Lei 34718789f4feSZhao Lei if (min_tolerated == INT_MAX) { 34728789f4feSZhao Lei pr_warn("BTRFS: unknown raid flag: %llu\n", flags); 34738789f4feSZhao Lei min_tolerated = 0; 34748789f4feSZhao Lei } 34758789f4feSZhao Lei 34768789f4feSZhao Lei return min_tolerated; 3477943c6e99SZhao Lei } 3478943c6e99SZhao Lei 34795af3e8ccSStefan Behrens int btrfs_calc_num_tolerated_disk_barrier_failures( 34805af3e8ccSStefan Behrens struct btrfs_fs_info *fs_info) 34815af3e8ccSStefan Behrens { 34825af3e8ccSStefan Behrens struct btrfs_ioctl_space_info space; 34835af3e8ccSStefan Behrens struct btrfs_space_info *sinfo; 34845af3e8ccSStefan Behrens u64 types[] = {BTRFS_BLOCK_GROUP_DATA, 34855af3e8ccSStefan Behrens BTRFS_BLOCK_GROUP_SYSTEM, 34865af3e8ccSStefan Behrens BTRFS_BLOCK_GROUP_METADATA, 34875af3e8ccSStefan Behrens BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA}; 34885af3e8ccSStefan Behrens int i; 34895af3e8ccSStefan Behrens int c; 34905af3e8ccSStefan Behrens int num_tolerated_disk_barrier_failures = 34915af3e8ccSStefan Behrens (int)fs_info->fs_devices->num_devices; 34925af3e8ccSStefan Behrens 34932c458045SZhao Lei for (i = 0; i < ARRAY_SIZE(types); i++) { 34945af3e8ccSStefan Behrens struct btrfs_space_info *tmp; 34955af3e8ccSStefan Behrens 34965af3e8ccSStefan Behrens sinfo = NULL; 34975af3e8ccSStefan Behrens rcu_read_lock(); 34985af3e8ccSStefan Behrens list_for_each_entry_rcu(tmp, &fs_info->space_info, list) { 34995af3e8ccSStefan Behrens if (tmp->flags == types[i]) { 35005af3e8ccSStefan Behrens sinfo = tmp; 35015af3e8ccSStefan Behrens break; 35025af3e8ccSStefan Behrens } 35035af3e8ccSStefan Behrens } 35045af3e8ccSStefan Behrens rcu_read_unlock(); 35055af3e8ccSStefan Behrens 35065af3e8ccSStefan Behrens if (!sinfo) 35075af3e8ccSStefan Behrens continue; 35085af3e8ccSStefan Behrens 35095af3e8ccSStefan Behrens down_read(&sinfo->groups_sem); 35105af3e8ccSStefan Behrens for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 35115af3e8ccSStefan Behrens u64 flags; 35125af3e8ccSStefan Behrens 35132c458045SZhao Lei if (list_empty(&sinfo->block_groups[c])) 35142c458045SZhao Lei continue; 35152c458045SZhao Lei 35162c458045SZhao Lei btrfs_get_block_group_info(&sinfo->block_groups[c], 35172c458045SZhao Lei &space); 35182c458045SZhao Lei if (space.total_bytes == 0 || space.used_bytes == 0) 35195af3e8ccSStefan Behrens continue; 35205af3e8ccSStefan Behrens flags = space.flags; 3521943c6e99SZhao Lei 3522943c6e99SZhao Lei num_tolerated_disk_barrier_failures = min( 3523943c6e99SZhao Lei num_tolerated_disk_barrier_failures, 3524943c6e99SZhao Lei btrfs_get_num_tolerated_disk_barrier_failures( 3525943c6e99SZhao Lei flags)); 35265af3e8ccSStefan Behrens } 35275af3e8ccSStefan Behrens up_read(&sinfo->groups_sem); 35285af3e8ccSStefan Behrens } 35295af3e8ccSStefan Behrens 35305af3e8ccSStefan Behrens return num_tolerated_disk_barrier_failures; 35315af3e8ccSStefan Behrens } 35325af3e8ccSStefan Behrens 353348a3b636SEric Sandeen static int write_all_supers(struct btrfs_root *root, int max_mirrors) 3534f2984462SChris Mason { 3535e5e9a520SChris Mason struct list_head *head; 3536f2984462SChris Mason struct btrfs_device *dev; 3537a061fc8dSChris Mason struct btrfs_super_block *sb; 3538f2984462SChris Mason struct btrfs_dev_item *dev_item; 3539f2984462SChris Mason int ret; 3540f2984462SChris Mason int do_barriers; 3541a236aed1SChris Mason int max_errors; 3542a236aed1SChris Mason int total_errors = 0; 3543a061fc8dSChris Mason u64 flags; 3544f2984462SChris Mason 3545f2984462SChris Mason do_barriers = !btrfs_test_opt(root, NOBARRIER); 3546af31f5e5SChris Mason backup_super_roots(root->fs_info); 3547f2984462SChris Mason 35486c41761fSDavid Sterba sb = root->fs_info->super_for_commit; 3549a061fc8dSChris Mason dev_item = &sb->dev_item; 3550e5e9a520SChris Mason 3551174ba509SChris Mason mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 3552e5e9a520SChris Mason head = &root->fs_info->fs_devices->devices; 3553d7306801SFilipe David Borba Manana max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1; 3554387125fcSChris Mason 35555af3e8ccSStefan Behrens if (do_barriers) { 35565af3e8ccSStefan Behrens ret = barrier_all_devices(root->fs_info); 35575af3e8ccSStefan Behrens if (ret) { 35585af3e8ccSStefan Behrens mutex_unlock( 35595af3e8ccSStefan Behrens &root->fs_info->fs_devices->device_list_mutex); 35605af3e8ccSStefan Behrens btrfs_error(root->fs_info, ret, 35615af3e8ccSStefan Behrens "errors while submitting device barriers."); 35625af3e8ccSStefan Behrens return ret; 35635af3e8ccSStefan Behrens } 35645af3e8ccSStefan Behrens } 3565387125fcSChris Mason 35661f78160cSXiao Guangrong list_for_each_entry_rcu(dev, head, dev_list) { 3567dfe25020SChris Mason if (!dev->bdev) { 3568dfe25020SChris Mason total_errors++; 3569dfe25020SChris Mason continue; 3570dfe25020SChris Mason } 35712b82032cSYan Zheng if (!dev->in_fs_metadata || !dev->writeable) 3572dfe25020SChris Mason continue; 3573dfe25020SChris Mason 35742b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 3575a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 3576a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 35777df69d3eSMiao Xie btrfs_set_stack_device_total_bytes(dev_item, 3578935e5cc9SMiao Xie dev->commit_total_bytes); 3579ce7213c7SMiao Xie btrfs_set_stack_device_bytes_used(dev_item, 3580ce7213c7SMiao Xie dev->commit_bytes_used); 3581a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 3582a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 3583a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 3584a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 35852b82032cSYan Zheng memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE); 3586a512bbf8SYan Zheng 3587a061fc8dSChris Mason flags = btrfs_super_flags(sb); 3588a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 3589f2984462SChris Mason 3590a512bbf8SYan Zheng ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors); 3591a236aed1SChris Mason if (ret) 3592a236aed1SChris Mason total_errors++; 3593f2984462SChris Mason } 3594a236aed1SChris Mason if (total_errors > max_errors) { 3595efe120a0SFrank Holton btrfs_err(root->fs_info, "%d errors while writing supers", 3596d397712bSChris Mason total_errors); 3597a724b436SStefan Behrens mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 359879787eaaSJeff Mahoney 35999d565ba4SStefan Behrens /* FUA is masked off if unsupported and can't be the reason */ 36009d565ba4SStefan Behrens btrfs_error(root->fs_info, -EIO, 36019d565ba4SStefan Behrens "%d errors while writing supers", total_errors); 36029d565ba4SStefan Behrens return -EIO; 3603a236aed1SChris Mason } 3604f2984462SChris Mason 3605a512bbf8SYan Zheng total_errors = 0; 36061f78160cSXiao Guangrong list_for_each_entry_rcu(dev, head, dev_list) { 3607dfe25020SChris Mason if (!dev->bdev) 3608dfe25020SChris Mason continue; 36092b82032cSYan Zheng if (!dev->in_fs_metadata || !dev->writeable) 3610dfe25020SChris Mason continue; 3611dfe25020SChris Mason 3612a512bbf8SYan Zheng ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors); 3613a512bbf8SYan Zheng if (ret) 36141259ab75SChris Mason total_errors++; 3615f2984462SChris Mason } 3616174ba509SChris Mason mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 3617a236aed1SChris Mason if (total_errors > max_errors) { 361879787eaaSJeff Mahoney btrfs_error(root->fs_info, -EIO, 361979787eaaSJeff Mahoney "%d errors while writing supers", total_errors); 362079787eaaSJeff Mahoney return -EIO; 3621a236aed1SChris Mason } 3622f2984462SChris Mason return 0; 3623f2984462SChris Mason } 3624f2984462SChris Mason 3625a512bbf8SYan Zheng int write_ctree_super(struct btrfs_trans_handle *trans, 3626a512bbf8SYan Zheng struct btrfs_root *root, int max_mirrors) 3627cfaa7295SChris Mason { 3628f570e757SRashika return write_all_supers(root, max_mirrors); 3629cfaa7295SChris Mason } 3630cfaa7295SChris Mason 3631cb517eabSMiao Xie /* Drop a fs root from the radix tree and free it. */ 3632cb517eabSMiao Xie void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 3633cb517eabSMiao Xie struct btrfs_root *root) 36342619ba1fSChris Mason { 36354df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 36362619ba1fSChris Mason radix_tree_delete(&fs_info->fs_roots_radix, 36372619ba1fSChris Mason (unsigned long)root->root_key.objectid); 36384df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 363976dda93cSYan, Zheng 364076dda93cSYan, Zheng if (btrfs_root_refs(&root->root_item) == 0) 364176dda93cSYan, Zheng synchronize_srcu(&fs_info->subvol_srcu); 364276dda93cSYan, Zheng 36431a4319ccSLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) 36443321719eSLiu Bo btrfs_free_log(NULL, root); 36453321719eSLiu Bo 3646faa2dbf0SJosef Bacik if (root->free_ino_pinned) 3647581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_pinned); 3648faa2dbf0SJosef Bacik if (root->free_ino_ctl) 3649581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_ctl); 36504df27c4dSYan, Zheng free_fs_root(root); 36514df27c4dSYan, Zheng } 36524df27c4dSYan, Zheng 36534df27c4dSYan, Zheng static void free_fs_root(struct btrfs_root *root) 36544df27c4dSYan, Zheng { 365557cdc8dbSDavid Sterba iput(root->ino_cache_inode); 36564df27c4dSYan, Zheng WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 36571cb048f5SFilipe David Borba Manana btrfs_free_block_rsv(root, root->orphan_block_rsv); 36581cb048f5SFilipe David Borba Manana root->orphan_block_rsv = NULL; 36590ee5dc67SAl Viro if (root->anon_dev) 36600ee5dc67SAl Viro free_anon_bdev(root->anon_dev); 36618257b2dcSMiao Xie if (root->subv_writers) 36628257b2dcSMiao Xie btrfs_free_subvolume_writers(root->subv_writers); 36635f39d397SChris Mason free_extent_buffer(root->node); 36645f39d397SChris Mason free_extent_buffer(root->commit_root); 3665581bb050SLi Zefan kfree(root->free_ino_ctl); 3666581bb050SLi Zefan kfree(root->free_ino_pinned); 366758176a96SJosef Bacik kfree(root->name); 3668b0feb9d9SMiao Xie btrfs_put_fs_root(root); 36692619ba1fSChris Mason } 36702619ba1fSChris Mason 3671cb517eabSMiao Xie void btrfs_free_fs_root(struct btrfs_root *root) 3672cb517eabSMiao Xie { 3673cb517eabSMiao Xie free_fs_root(root); 3674eb60ceacSChris Mason } 3675eb60ceacSChris Mason 3676c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 3677c146afadSYan Zheng { 3678c146afadSYan Zheng u64 root_objectid = 0; 3679c146afadSYan Zheng struct btrfs_root *gang[8]; 368065d33fd7SQu Wenruo int i = 0; 368165d33fd7SQu Wenruo int err = 0; 368265d33fd7SQu Wenruo unsigned int ret = 0; 368365d33fd7SQu Wenruo int index; 3684c146afadSYan Zheng 3685c146afadSYan Zheng while (1) { 368665d33fd7SQu Wenruo index = srcu_read_lock(&fs_info->subvol_srcu); 3687c146afadSYan Zheng ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 3688c146afadSYan Zheng (void **)gang, root_objectid, 3689c146afadSYan Zheng ARRAY_SIZE(gang)); 369065d33fd7SQu Wenruo if (!ret) { 369165d33fd7SQu Wenruo srcu_read_unlock(&fs_info->subvol_srcu, index); 3692c146afadSYan Zheng break; 369365d33fd7SQu Wenruo } 36945d4f98a2SYan Zheng root_objectid = gang[ret - 1]->root_key.objectid + 1; 369566b4ffd1SJosef Bacik 369665d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 369765d33fd7SQu Wenruo /* Avoid to grab roots in dead_roots */ 369865d33fd7SQu Wenruo if (btrfs_root_refs(&gang[i]->root_item) == 0) { 369965d33fd7SQu Wenruo gang[i] = NULL; 370065d33fd7SQu Wenruo continue; 370165d33fd7SQu Wenruo } 370265d33fd7SQu Wenruo /* grab all the search result for later use */ 370365d33fd7SQu Wenruo gang[i] = btrfs_grab_fs_root(gang[i]); 370465d33fd7SQu Wenruo } 370565d33fd7SQu Wenruo srcu_read_unlock(&fs_info->subvol_srcu, index); 370665d33fd7SQu Wenruo 370765d33fd7SQu Wenruo for (i = 0; i < ret; i++) { 370865d33fd7SQu Wenruo if (!gang[i]) 370965d33fd7SQu Wenruo continue; 3710c146afadSYan Zheng root_objectid = gang[i]->root_key.objectid; 371166b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(gang[i]); 371266b4ffd1SJosef Bacik if (err) 371365d33fd7SQu Wenruo break; 371465d33fd7SQu Wenruo btrfs_put_fs_root(gang[i]); 3715c146afadSYan Zheng } 3716c146afadSYan Zheng root_objectid++; 3717c146afadSYan Zheng } 371865d33fd7SQu Wenruo 371965d33fd7SQu Wenruo /* release the uncleaned roots due to error */ 372065d33fd7SQu Wenruo for (; i < ret; i++) { 372165d33fd7SQu Wenruo if (gang[i]) 372265d33fd7SQu Wenruo btrfs_put_fs_root(gang[i]); 372365d33fd7SQu Wenruo } 372465d33fd7SQu Wenruo return err; 3725c146afadSYan Zheng } 3726c146afadSYan Zheng 3727c146afadSYan Zheng int btrfs_commit_super(struct btrfs_root *root) 3728c146afadSYan Zheng { 3729c146afadSYan Zheng struct btrfs_trans_handle *trans; 3730c146afadSYan Zheng 3731c146afadSYan Zheng mutex_lock(&root->fs_info->cleaner_mutex); 373224bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 3733c146afadSYan Zheng mutex_unlock(&root->fs_info->cleaner_mutex); 37349d1a2a3aSDavid Sterba wake_up_process(root->fs_info->cleaner_kthread); 3735c71bf099SYan, Zheng 3736c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 3737c71bf099SYan, Zheng down_write(&root->fs_info->cleanup_work_sem); 3738c71bf099SYan, Zheng up_write(&root->fs_info->cleanup_work_sem); 3739c71bf099SYan, Zheng 37407a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 37413612b495STsutomu Itoh if (IS_ERR(trans)) 37423612b495STsutomu Itoh return PTR_ERR(trans); 3743d52c1bccSLiu Bo return btrfs_commit_transaction(trans, root); 3744c146afadSYan Zheng } 3745c146afadSYan Zheng 37463abdbd78SDavid Sterba void close_ctree(struct btrfs_root *root) 3747eb60ceacSChris Mason { 37480f7d52f4SChris Mason struct btrfs_fs_info *fs_info = root->fs_info; 3749c146afadSYan Zheng int ret; 3750e089f05cSChris Mason 3751facda1e7SChris Mason fs_info->closing = 1; 3752a2135011SChris Mason smp_mb(); 3753a2135011SChris Mason 3754803b2f54SStefan Behrens /* wait for the uuid_scan task to finish */ 3755803b2f54SStefan Behrens down(&fs_info->uuid_tree_rescan_sem); 3756803b2f54SStefan Behrens /* avoid complains from lockdep et al., set sem back to initial state */ 3757803b2f54SStefan Behrens up(&fs_info->uuid_tree_rescan_sem); 3758803b2f54SStefan Behrens 3759837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 3760aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 3761837d5b6eSIlya Dryomov 37628dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 37638dabb742SStefan Behrens 3764aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 37654cb5300bSChris Mason 37664cb5300bSChris Mason /* wait for any defraggers to finish */ 37674cb5300bSChris Mason wait_event(fs_info->transaction_wait, 37684cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 37694cb5300bSChris Mason 37704cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 377126176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 37724cb5300bSChris Mason 377321c7e756SMiao Xie cancel_work_sync(&fs_info->async_reclaim_work); 377421c7e756SMiao Xie 3775c146afadSYan Zheng if (!(fs_info->sb->s_flags & MS_RDONLY)) { 3776e44163e1SJeff Mahoney /* 3777e44163e1SJeff Mahoney * If the cleaner thread is stopped and there are 3778e44163e1SJeff Mahoney * block groups queued for removal, the deletion will be 3779e44163e1SJeff Mahoney * skipped when we quit the cleaner thread. 3780e44163e1SJeff Mahoney */ 3781e44163e1SJeff Mahoney btrfs_delete_unused_bgs(root->fs_info); 3782e44163e1SJeff Mahoney 3783c146afadSYan Zheng ret = btrfs_commit_super(root); 3784d397712bSChris Mason if (ret) 378504892340SEric Sandeen btrfs_err(fs_info, "commit super ret %d", ret); 3786c146afadSYan Zheng } 3787ed2ff2cbSChris Mason 378887533c47SMiao Xie if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) 378968ce9682SStefan Behrens btrfs_error_commit_super(root); 3790acce952bSliubo 3791e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 3792e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 37938929ecfaSYan, Zheng 3794f25784b3SYan Zheng fs_info->closing = 2; 3795f25784b3SYan Zheng smp_mb(); 3796f25784b3SYan Zheng 379704892340SEric Sandeen btrfs_free_qgroup_config(fs_info); 3798bcef60f2SArne Jansen 3799963d678bSMiao Xie if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 380004892340SEric Sandeen btrfs_info(fs_info, "at unmount delalloc count %lld", 3801963d678bSMiao Xie percpu_counter_sum(&fs_info->delalloc_bytes)); 3802b0c68f8bSChris Mason } 380331153d81SYan Zheng 38045ac1d209SJeff Mahoney btrfs_sysfs_remove_one(fs_info); 3805b7c35e81SAnand Jain btrfs_sysfs_remove_fsid(fs_info->fs_devices); 38065ac1d209SJeff Mahoney 3807faa2dbf0SJosef Bacik btrfs_free_fs_roots(fs_info); 3808d10c5f31SChris Mason 38091a4319ccSLiu Bo btrfs_put_block_group_cache(fs_info); 38101a4319ccSLiu Bo 38112b1360daSJosef Bacik btrfs_free_block_groups(fs_info); 38122b1360daSJosef Bacik 3813de348ee0SWang Shilong /* 3814de348ee0SWang Shilong * we must make sure there is not any read request to 3815de348ee0SWang Shilong * submit after we stopping all workers. 3816de348ee0SWang Shilong */ 3817de348ee0SWang Shilong invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 381896192499SJosef Bacik btrfs_stop_all_workers(fs_info); 381996192499SJosef Bacik 382047ab2a6cSJosef Bacik fs_info->open = 0; 382113e6c37bSJosef Bacik free_root_pointers(fs_info, 1); 38229ad6b7bcSChris Mason 382313e6c37bSJosef Bacik iput(fs_info->btree_inode); 3824d6bfde87SChris Mason 382521adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 382621adbd5cSStefan Behrens if (btrfs_test_opt(root, CHECK_INTEGRITY)) 382721adbd5cSStefan Behrens btrfsic_unmount(root, fs_info->fs_devices); 382821adbd5cSStefan Behrens #endif 382921adbd5cSStefan Behrens 3830dfe25020SChris Mason btrfs_close_devices(fs_info->fs_devices); 38310b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 3832b248a415SChris Mason 3833e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 3834963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 3835c404e0dcSMiao Xie percpu_counter_destroy(&fs_info->bio_counter); 383604160088SChris Mason bdi_destroy(&fs_info->bdi); 383776dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 38380b86a832SChris Mason 383953b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 384053b381b3SDavid Woodhouse 3841cdfb080eSChris Mason __btrfs_free_block_rsv(root->orphan_block_rsv); 38421cb048f5SFilipe David Borba Manana root->orphan_block_rsv = NULL; 384304216820SFilipe Manana 384404216820SFilipe Manana lock_chunks(root); 384504216820SFilipe Manana while (!list_empty(&fs_info->pinned_chunks)) { 384604216820SFilipe Manana struct extent_map *em; 384704216820SFilipe Manana 384804216820SFilipe Manana em = list_first_entry(&fs_info->pinned_chunks, 384904216820SFilipe Manana struct extent_map, list); 385004216820SFilipe Manana list_del_init(&em->list); 385104216820SFilipe Manana free_extent_map(em); 385204216820SFilipe Manana } 385304216820SFilipe Manana unlock_chunks(root); 3854eb60ceacSChris Mason } 3855eb60ceacSChris Mason 3856b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 3857b9fab919SChris Mason int atomic) 3858ccd467d6SChris Mason { 38591259ab75SChris Mason int ret; 3860727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 38611259ab75SChris Mason 38620b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 38631259ab75SChris Mason if (!ret) 38641259ab75SChris Mason return ret; 38651259ab75SChris Mason 38661259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 3867b9fab919SChris Mason parent_transid, atomic); 3868b9fab919SChris Mason if (ret == -EAGAIN) 3869b9fab919SChris Mason return ret; 38701259ab75SChris Mason return !ret; 38715f39d397SChris Mason } 38726702ed49SChris Mason 38735f39d397SChris Mason int btrfs_set_buffer_uptodate(struct extent_buffer *buf) 38745f39d397SChris Mason { 38750b32f4bbSJosef Bacik return set_extent_buffer_uptodate(buf); 38765f39d397SChris Mason } 38775f39d397SChris Mason 38785f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 38795f39d397SChris Mason { 388006ea65a3SJosef Bacik struct btrfs_root *root; 38815f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 3882b9473439SChris Mason int was_dirty; 3883b4ce94deSChris Mason 388406ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 388506ea65a3SJosef Bacik /* 388606ea65a3SJosef Bacik * This is a fast path so only do this check if we have sanity tests 388706ea65a3SJosef Bacik * enabled. Normal people shouldn't be marking dummy buffers as dirty 388806ea65a3SJosef Bacik * outside of the sanity tests. 388906ea65a3SJosef Bacik */ 389006ea65a3SJosef Bacik if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags))) 389106ea65a3SJosef Bacik return; 389206ea65a3SJosef Bacik #endif 389306ea65a3SJosef Bacik root = BTRFS_I(buf->pages[0]->mapping->host)->root; 3894b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 389531b1a2bdSJulia Lawall if (transid != root->fs_info->generation) 389631b1a2bdSJulia Lawall WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, " 3897d397712bSChris Mason "found %llu running %llu\n", 3898c1c9ff7cSGeert Uytterhoeven buf->start, transid, root->fs_info->generation); 38990b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 3900e2d84521SMiao Xie if (!was_dirty) 3901e2d84521SMiao Xie __percpu_counter_add(&root->fs_info->dirty_metadata_bytes, 3902e2d84521SMiao Xie buf->len, 3903e2d84521SMiao Xie root->fs_info->dirty_metadata_batch); 39041f21ef0aSFilipe Manana #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 39051f21ef0aSFilipe Manana if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) { 39061f21ef0aSFilipe Manana btrfs_print_leaf(root, buf); 39071f21ef0aSFilipe Manana ASSERT(0); 39081f21ef0aSFilipe Manana } 39091f21ef0aSFilipe Manana #endif 3910eb60ceacSChris Mason } 3911eb60ceacSChris Mason 3912b53d3f5dSLiu Bo static void __btrfs_btree_balance_dirty(struct btrfs_root *root, 3913b53d3f5dSLiu Bo int flush_delayed) 391435b7e476SChris Mason { 3915188de649SChris Mason /* 3916188de649SChris Mason * looks as though older kernels can get into trouble with 3917188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 3918188de649SChris Mason */ 3919e2d84521SMiao Xie int ret; 3920d6bfde87SChris Mason 39216933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 3922d6bfde87SChris Mason return; 3923d6bfde87SChris Mason 3924b53d3f5dSLiu Bo if (flush_delayed) 392516cdcec7SMiao Xie btrfs_balance_delayed_items(root); 392616cdcec7SMiao Xie 3927e2d84521SMiao Xie ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes, 3928e2d84521SMiao Xie BTRFS_DIRTY_METADATA_THRESH); 3929e2d84521SMiao Xie if (ret > 0) { 3930d0e1d66bSNamjae Jeon balance_dirty_pages_ratelimited( 3931d0e1d66bSNamjae Jeon root->fs_info->btree_inode->i_mapping); 393216cdcec7SMiao Xie } 393316cdcec7SMiao Xie return; 393416cdcec7SMiao Xie } 393516cdcec7SMiao Xie 3936b53d3f5dSLiu Bo void btrfs_btree_balance_dirty(struct btrfs_root *root) 393716cdcec7SMiao Xie { 3938b53d3f5dSLiu Bo __btrfs_btree_balance_dirty(root, 1); 393935b7e476SChris Mason } 3940b53d3f5dSLiu Bo 3941b53d3f5dSLiu Bo void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root) 3942b53d3f5dSLiu Bo { 3943b53d3f5dSLiu Bo __btrfs_btree_balance_dirty(root, 0); 3944d6bfde87SChris Mason } 39456b80053dSChris Mason 3946ca7a79adSChris Mason int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid) 39476b80053dSChris Mason { 3948727011e0SChris Mason struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root; 39490b32f4bbSJosef Bacik return btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 39506b80053dSChris Mason } 39510da5468fSChris Mason 3952fcd1f065SDavid Sterba static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info, 3953acce952bSliubo int read_only) 3954acce952bSliubo { 3955c926093eSDavid Sterba struct btrfs_super_block *sb = fs_info->super_copy; 3956c926093eSDavid Sterba int ret = 0; 3957c926093eSDavid Sterba 395821e7626bSDavid Sterba if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 395921e7626bSDavid Sterba printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n", 396021e7626bSDavid Sterba btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 3961c926093eSDavid Sterba ret = -EINVAL; 3962c926093eSDavid Sterba } 396321e7626bSDavid Sterba if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 396421e7626bSDavid Sterba printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n", 396521e7626bSDavid Sterba btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 3966c926093eSDavid Sterba ret = -EINVAL; 3967c926093eSDavid Sterba } 396821e7626bSDavid Sterba if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 396921e7626bSDavid Sterba printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n", 397021e7626bSDavid Sterba btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 3971c926093eSDavid Sterba ret = -EINVAL; 3972c926093eSDavid Sterba } 3973c926093eSDavid Sterba 39741104a885SDavid Sterba /* 3975c926093eSDavid Sterba * The common minimum, we don't know if we can trust the nodesize/sectorsize 3976c926093eSDavid Sterba * items yet, they'll be verified later. Issue just a warning. 39771104a885SDavid Sterba */ 397821e7626bSDavid Sterba if (!IS_ALIGNED(btrfs_super_root(sb), 4096)) 3979c926093eSDavid Sterba printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n", 3980cd743facSDavid Sterba btrfs_super_root(sb)); 398121e7626bSDavid Sterba if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096)) 3982cd743facSDavid Sterba printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n", 3983cd743facSDavid Sterba btrfs_super_chunk_root(sb)); 398421e7626bSDavid Sterba if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096)) 3985cd743facSDavid Sterba printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n", 398621e7626bSDavid Sterba btrfs_super_log_root(sb)); 3987c926093eSDavid Sterba 398875d6ad38SDavid Sterba /* 398975d6ad38SDavid Sterba * Check the lower bound, the alignment and other constraints are 399075d6ad38SDavid Sterba * checked later. 399175d6ad38SDavid Sterba */ 399275d6ad38SDavid Sterba if (btrfs_super_nodesize(sb) < 4096) { 399375d6ad38SDavid Sterba printk(KERN_ERR "BTRFS: nodesize too small: %u < 4096\n", 399475d6ad38SDavid Sterba btrfs_super_nodesize(sb)); 399575d6ad38SDavid Sterba ret = -EINVAL; 399675d6ad38SDavid Sterba } 399775d6ad38SDavid Sterba if (btrfs_super_sectorsize(sb) < 4096) { 399875d6ad38SDavid Sterba printk(KERN_ERR "BTRFS: sectorsize too small: %u < 4096\n", 399975d6ad38SDavid Sterba btrfs_super_sectorsize(sb)); 400075d6ad38SDavid Sterba ret = -EINVAL; 400175d6ad38SDavid Sterba } 400275d6ad38SDavid Sterba 4003c926093eSDavid Sterba if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) { 4004c926093eSDavid Sterba printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n", 4005c926093eSDavid Sterba fs_info->fsid, sb->dev_item.fsid); 4006c926093eSDavid Sterba ret = -EINVAL; 4007c926093eSDavid Sterba } 4008c926093eSDavid Sterba 4009c926093eSDavid Sterba /* 4010c926093eSDavid Sterba * Hint to catch really bogus numbers, bitflips or so, more exact checks are 4011c926093eSDavid Sterba * done later 4012c926093eSDavid Sterba */ 401321e7626bSDavid Sterba if (btrfs_super_num_devices(sb) > (1UL << 31)) 4014c926093eSDavid Sterba printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n", 401521e7626bSDavid Sterba btrfs_super_num_devices(sb)); 401675d6ad38SDavid Sterba if (btrfs_super_num_devices(sb) == 0) { 401775d6ad38SDavid Sterba printk(KERN_ERR "BTRFS: number of devices is 0\n"); 401875d6ad38SDavid Sterba ret = -EINVAL; 401975d6ad38SDavid Sterba } 4020c926093eSDavid Sterba 402121e7626bSDavid Sterba if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) { 4022c926093eSDavid Sterba printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n", 402321e7626bSDavid Sterba btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 4024c926093eSDavid Sterba ret = -EINVAL; 4025c926093eSDavid Sterba } 4026c926093eSDavid Sterba 4027c926093eSDavid Sterba /* 4028ce7fca5fSDavid Sterba * Obvious sys_chunk_array corruptions, it must hold at least one key 4029ce7fca5fSDavid Sterba * and one chunk 4030ce7fca5fSDavid Sterba */ 4031ce7fca5fSDavid Sterba if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 4032ce7fca5fSDavid Sterba printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n", 4033ce7fca5fSDavid Sterba btrfs_super_sys_array_size(sb), 4034ce7fca5fSDavid Sterba BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 4035ce7fca5fSDavid Sterba ret = -EINVAL; 4036ce7fca5fSDavid Sterba } 4037ce7fca5fSDavid Sterba if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 4038ce7fca5fSDavid Sterba + sizeof(struct btrfs_chunk)) { 4039d2207129SFabian Frederick printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n", 4040ce7fca5fSDavid Sterba btrfs_super_sys_array_size(sb), 4041ce7fca5fSDavid Sterba sizeof(struct btrfs_disk_key) 4042ce7fca5fSDavid Sterba + sizeof(struct btrfs_chunk)); 4043ce7fca5fSDavid Sterba ret = -EINVAL; 4044ce7fca5fSDavid Sterba } 4045ce7fca5fSDavid Sterba 4046ce7fca5fSDavid Sterba /* 4047c926093eSDavid Sterba * The generation is a global counter, we'll trust it more than the others 4048c926093eSDavid Sterba * but it's still possible that it's the one that's wrong. 4049c926093eSDavid Sterba */ 405021e7626bSDavid Sterba if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 4051c926093eSDavid Sterba printk(KERN_WARNING 4052c926093eSDavid Sterba "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n", 405321e7626bSDavid Sterba btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb)); 405421e7626bSDavid Sterba if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 405521e7626bSDavid Sterba && btrfs_super_cache_generation(sb) != (u64)-1) 4056c926093eSDavid Sterba printk(KERN_WARNING 4057c926093eSDavid Sterba "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n", 405821e7626bSDavid Sterba btrfs_super_generation(sb), btrfs_super_cache_generation(sb)); 4059c926093eSDavid Sterba 4060c926093eSDavid Sterba return ret; 4061fcd1f065SDavid Sterba } 4062fcd1f065SDavid Sterba 406348a3b636SEric Sandeen static void btrfs_error_commit_super(struct btrfs_root *root) 4064acce952bSliubo { 4065acce952bSliubo mutex_lock(&root->fs_info->cleaner_mutex); 4066acce952bSliubo btrfs_run_delayed_iputs(root); 4067acce952bSliubo mutex_unlock(&root->fs_info->cleaner_mutex); 4068acce952bSliubo 4069acce952bSliubo down_write(&root->fs_info->cleanup_work_sem); 4070acce952bSliubo up_write(&root->fs_info->cleanup_work_sem); 4071acce952bSliubo 4072acce952bSliubo /* cleanup FS via transaction */ 4073acce952bSliubo btrfs_cleanup_transaction(root); 4074acce952bSliubo } 4075acce952bSliubo 4076143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4077acce952bSliubo { 4078acce952bSliubo struct btrfs_ordered_extent *ordered; 4079acce952bSliubo 4080199c2a9cSMiao Xie spin_lock(&root->ordered_extent_lock); 4081779880efSJosef Bacik /* 4082779880efSJosef Bacik * This will just short circuit the ordered completion stuff which will 4083779880efSJosef Bacik * make sure the ordered extent gets properly cleaned up. 4084779880efSJosef Bacik */ 4085199c2a9cSMiao Xie list_for_each_entry(ordered, &root->ordered_extents, 4086779880efSJosef Bacik root_extent_list) 4087779880efSJosef Bacik set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4088199c2a9cSMiao Xie spin_unlock(&root->ordered_extent_lock); 4089199c2a9cSMiao Xie } 4090199c2a9cSMiao Xie 4091199c2a9cSMiao Xie static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4092199c2a9cSMiao Xie { 4093199c2a9cSMiao Xie struct btrfs_root *root; 4094199c2a9cSMiao Xie struct list_head splice; 4095199c2a9cSMiao Xie 4096199c2a9cSMiao Xie INIT_LIST_HEAD(&splice); 4097199c2a9cSMiao Xie 4098199c2a9cSMiao Xie spin_lock(&fs_info->ordered_root_lock); 4099199c2a9cSMiao Xie list_splice_init(&fs_info->ordered_roots, &splice); 4100199c2a9cSMiao Xie while (!list_empty(&splice)) { 4101199c2a9cSMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4102199c2a9cSMiao Xie ordered_root); 41031de2cfdeSJosef Bacik list_move_tail(&root->ordered_root, 41041de2cfdeSJosef Bacik &fs_info->ordered_roots); 4105199c2a9cSMiao Xie 41062a85d9caSLiu Bo spin_unlock(&fs_info->ordered_root_lock); 4107199c2a9cSMiao Xie btrfs_destroy_ordered_extents(root); 4108199c2a9cSMiao Xie 41092a85d9caSLiu Bo cond_resched(); 41102a85d9caSLiu Bo spin_lock(&fs_info->ordered_root_lock); 4111199c2a9cSMiao Xie } 4112199c2a9cSMiao Xie spin_unlock(&fs_info->ordered_root_lock); 4113acce952bSliubo } 4114acce952bSliubo 411535a3621bSStefan Behrens static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 4116acce952bSliubo struct btrfs_root *root) 4117acce952bSliubo { 4118acce952bSliubo struct rb_node *node; 4119acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 4120acce952bSliubo struct btrfs_delayed_ref_node *ref; 4121acce952bSliubo int ret = 0; 4122acce952bSliubo 4123acce952bSliubo delayed_refs = &trans->delayed_refs; 4124acce952bSliubo 4125acce952bSliubo spin_lock(&delayed_refs->lock); 4126d7df2c79SJosef Bacik if (atomic_read(&delayed_refs->num_entries) == 0) { 4127cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 4128efe120a0SFrank Holton btrfs_info(root->fs_info, "delayed_refs has NO entry"); 4129acce952bSliubo return ret; 4130acce952bSliubo } 4131acce952bSliubo 4132d7df2c79SJosef Bacik while ((node = rb_first(&delayed_refs->href_root)) != NULL) { 4133d7df2c79SJosef Bacik struct btrfs_delayed_ref_head *head; 4134c6fc2454SQu Wenruo struct btrfs_delayed_ref_node *tmp; 4135e78417d1SJosef Bacik bool pin_bytes = false; 4136acce952bSliubo 4137d7df2c79SJosef Bacik head = rb_entry(node, struct btrfs_delayed_ref_head, 4138d7df2c79SJosef Bacik href_node); 4139b939d1abSJosef Bacik if (!mutex_trylock(&head->mutex)) { 4140d7df2c79SJosef Bacik atomic_inc(&head->node.refs); 414179787eaaSJeff Mahoney spin_unlock(&delayed_refs->lock); 4142b939d1abSJosef Bacik 4143acce952bSliubo mutex_lock(&head->mutex); 4144b939d1abSJosef Bacik mutex_unlock(&head->mutex); 4145d7df2c79SJosef Bacik btrfs_put_delayed_ref(&head->node); 4146e18fca73SJosef Bacik spin_lock(&delayed_refs->lock); 4147b939d1abSJosef Bacik continue; 4148b939d1abSJosef Bacik } 4149d7df2c79SJosef Bacik spin_lock(&head->lock); 4150c6fc2454SQu Wenruo list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list, 4151c6fc2454SQu Wenruo list) { 4152d7df2c79SJosef Bacik ref->in_tree = 0; 4153c6fc2454SQu Wenruo list_del(&ref->list); 4154d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4155d7df2c79SJosef Bacik btrfs_put_delayed_ref(ref); 4156d7df2c79SJosef Bacik } 415754067ae9SJosef Bacik if (head->must_insert_reserved) 4158e78417d1SJosef Bacik pin_bytes = true; 415978a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 4160acce952bSliubo delayed_refs->num_heads--; 4161d7df2c79SJosef Bacik if (head->processing == 0) 4162acce952bSliubo delayed_refs->num_heads_ready--; 4163d7df2c79SJosef Bacik atomic_dec(&delayed_refs->num_entries); 4164d7df2c79SJosef Bacik head->node.in_tree = 0; 4165c46effa6SLiu Bo rb_erase(&head->href_node, &delayed_refs->href_root); 4166d7df2c79SJosef Bacik spin_unlock(&head->lock); 4167acce952bSliubo spin_unlock(&delayed_refs->lock); 4168e78417d1SJosef Bacik mutex_unlock(&head->mutex); 4169acce952bSliubo 4170d7df2c79SJosef Bacik if (pin_bytes) 4171d7df2c79SJosef Bacik btrfs_pin_extent(root, head->node.bytenr, 4172d7df2c79SJosef Bacik head->node.num_bytes, 1); 4173d7df2c79SJosef Bacik btrfs_put_delayed_ref(&head->node); 4174acce952bSliubo cond_resched(); 4175acce952bSliubo spin_lock(&delayed_refs->lock); 4176acce952bSliubo } 4177acce952bSliubo 4178acce952bSliubo spin_unlock(&delayed_refs->lock); 4179acce952bSliubo 4180acce952bSliubo return ret; 4181acce952bSliubo } 4182acce952bSliubo 4183143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 4184acce952bSliubo { 4185acce952bSliubo struct btrfs_inode *btrfs_inode; 4186acce952bSliubo struct list_head splice; 4187acce952bSliubo 4188acce952bSliubo INIT_LIST_HEAD(&splice); 4189acce952bSliubo 4190eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4191eb73c1b7SMiao Xie list_splice_init(&root->delalloc_inodes, &splice); 4192acce952bSliubo 4193acce952bSliubo while (!list_empty(&splice)) { 4194eb73c1b7SMiao Xie btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 4195acce952bSliubo delalloc_inodes); 4196acce952bSliubo 4197acce952bSliubo list_del_init(&btrfs_inode->delalloc_inodes); 4198df0af1a5SMiao Xie clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, 4199df0af1a5SMiao Xie &btrfs_inode->runtime_flags); 4200eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4201acce952bSliubo 4202acce952bSliubo btrfs_invalidate_inodes(btrfs_inode->root); 4203b216cbfbSMiao Xie 4204eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 4205acce952bSliubo } 4206acce952bSliubo 4207eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 4208eb73c1b7SMiao Xie } 4209eb73c1b7SMiao Xie 4210eb73c1b7SMiao Xie static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 4211eb73c1b7SMiao Xie { 4212eb73c1b7SMiao Xie struct btrfs_root *root; 4213eb73c1b7SMiao Xie struct list_head splice; 4214eb73c1b7SMiao Xie 4215eb73c1b7SMiao Xie INIT_LIST_HEAD(&splice); 4216eb73c1b7SMiao Xie 4217eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4218eb73c1b7SMiao Xie list_splice_init(&fs_info->delalloc_roots, &splice); 4219eb73c1b7SMiao Xie while (!list_empty(&splice)) { 4220eb73c1b7SMiao Xie root = list_first_entry(&splice, struct btrfs_root, 4221eb73c1b7SMiao Xie delalloc_root); 4222eb73c1b7SMiao Xie list_del_init(&root->delalloc_root); 4223eb73c1b7SMiao Xie root = btrfs_grab_fs_root(root); 4224eb73c1b7SMiao Xie BUG_ON(!root); 4225eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4226eb73c1b7SMiao Xie 4227eb73c1b7SMiao Xie btrfs_destroy_delalloc_inodes(root); 4228eb73c1b7SMiao Xie btrfs_put_fs_root(root); 4229eb73c1b7SMiao Xie 4230eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 4231eb73c1b7SMiao Xie } 4232eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 4233acce952bSliubo } 4234acce952bSliubo 4235acce952bSliubo static int btrfs_destroy_marked_extents(struct btrfs_root *root, 4236acce952bSliubo struct extent_io_tree *dirty_pages, 4237acce952bSliubo int mark) 4238acce952bSliubo { 4239acce952bSliubo int ret; 4240acce952bSliubo struct extent_buffer *eb; 4241acce952bSliubo u64 start = 0; 4242acce952bSliubo u64 end; 4243acce952bSliubo 4244acce952bSliubo while (1) { 4245acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 4246e6138876SJosef Bacik mark, NULL); 4247acce952bSliubo if (ret) 4248acce952bSliubo break; 4249acce952bSliubo 4250acce952bSliubo clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); 4251acce952bSliubo while (start <= end) { 425201d58472SDaniel Dressler eb = btrfs_find_tree_block(root->fs_info, start); 4253707e8a07SDavid Sterba start += root->nodesize; 4254fd8b2b61SJosef Bacik if (!eb) 4255acce952bSliubo continue; 4256fd8b2b61SJosef Bacik wait_on_extent_buffer_writeback(eb); 4257acce952bSliubo 4258fd8b2b61SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 4259fd8b2b61SJosef Bacik &eb->bflags)) 4260fd8b2b61SJosef Bacik clear_extent_buffer_dirty(eb); 4261fd8b2b61SJosef Bacik free_extent_buffer_stale(eb); 4262acce952bSliubo } 4263acce952bSliubo } 4264acce952bSliubo 4265acce952bSliubo return ret; 4266acce952bSliubo } 4267acce952bSliubo 4268acce952bSliubo static int btrfs_destroy_pinned_extent(struct btrfs_root *root, 4269acce952bSliubo struct extent_io_tree *pinned_extents) 4270acce952bSliubo { 4271acce952bSliubo struct extent_io_tree *unpin; 4272acce952bSliubo u64 start; 4273acce952bSliubo u64 end; 4274acce952bSliubo int ret; 4275ed0eaa14SLiu Bo bool loop = true; 4276acce952bSliubo 4277acce952bSliubo unpin = pinned_extents; 4278ed0eaa14SLiu Bo again: 4279acce952bSliubo while (1) { 4280acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 4281e6138876SJosef Bacik EXTENT_DIRTY, NULL); 4282acce952bSliubo if (ret) 4283acce952bSliubo break; 4284acce952bSliubo 4285acce952bSliubo clear_extent_dirty(unpin, start, end, GFP_NOFS); 4286acce952bSliubo btrfs_error_unpin_extent_range(root, start, end); 4287acce952bSliubo cond_resched(); 4288acce952bSliubo } 4289acce952bSliubo 4290ed0eaa14SLiu Bo if (loop) { 4291ed0eaa14SLiu Bo if (unpin == &root->fs_info->freed_extents[0]) 4292ed0eaa14SLiu Bo unpin = &root->fs_info->freed_extents[1]; 4293ed0eaa14SLiu Bo else 4294ed0eaa14SLiu Bo unpin = &root->fs_info->freed_extents[0]; 4295ed0eaa14SLiu Bo loop = false; 4296ed0eaa14SLiu Bo goto again; 4297ed0eaa14SLiu Bo } 4298ed0eaa14SLiu Bo 4299acce952bSliubo return 0; 4300acce952bSliubo } 4301acce952bSliubo 430250d9aa99SJosef Bacik static void btrfs_free_pending_ordered(struct btrfs_transaction *cur_trans, 430350d9aa99SJosef Bacik struct btrfs_fs_info *fs_info) 430450d9aa99SJosef Bacik { 430550d9aa99SJosef Bacik struct btrfs_ordered_extent *ordered; 430650d9aa99SJosef Bacik 430750d9aa99SJosef Bacik spin_lock(&fs_info->trans_lock); 430850d9aa99SJosef Bacik while (!list_empty(&cur_trans->pending_ordered)) { 430950d9aa99SJosef Bacik ordered = list_first_entry(&cur_trans->pending_ordered, 431050d9aa99SJosef Bacik struct btrfs_ordered_extent, 431150d9aa99SJosef Bacik trans_list); 431250d9aa99SJosef Bacik list_del_init(&ordered->trans_list); 431350d9aa99SJosef Bacik spin_unlock(&fs_info->trans_lock); 431450d9aa99SJosef Bacik 431550d9aa99SJosef Bacik btrfs_put_ordered_extent(ordered); 431650d9aa99SJosef Bacik spin_lock(&fs_info->trans_lock); 431750d9aa99SJosef Bacik } 431850d9aa99SJosef Bacik spin_unlock(&fs_info->trans_lock); 431950d9aa99SJosef Bacik } 432050d9aa99SJosef Bacik 432149b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 432249b25e05SJeff Mahoney struct btrfs_root *root) 432349b25e05SJeff Mahoney { 432449b25e05SJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, root); 432549b25e05SJeff Mahoney 43264a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 432749b25e05SJeff Mahoney wake_up(&root->fs_info->transaction_blocked_wait); 432849b25e05SJeff Mahoney 43294a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 433049b25e05SJeff Mahoney wake_up(&root->fs_info->transaction_wait); 433149b25e05SJeff Mahoney 433250d9aa99SJosef Bacik btrfs_free_pending_ordered(cur_trans, root->fs_info); 433367cde344SMiao Xie btrfs_destroy_delayed_inodes(root); 433467cde344SMiao Xie btrfs_assert_delayed_root_empty(root); 433567cde344SMiao Xie 433649b25e05SJeff Mahoney btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages, 433749b25e05SJeff Mahoney EXTENT_DIRTY); 43386e841e32SLiu Bo btrfs_destroy_pinned_extent(root, 43396e841e32SLiu Bo root->fs_info->pinned_extents); 434049b25e05SJeff Mahoney 43414a9d8bdeSMiao Xie cur_trans->state =TRANS_STATE_COMPLETED; 43424a9d8bdeSMiao Xie wake_up(&cur_trans->commit_wait); 43434a9d8bdeSMiao Xie 434449b25e05SJeff Mahoney /* 434549b25e05SJeff Mahoney memset(cur_trans, 0, sizeof(*cur_trans)); 434649b25e05SJeff Mahoney kmem_cache_free(btrfs_transaction_cachep, cur_trans); 434749b25e05SJeff Mahoney */ 434849b25e05SJeff Mahoney } 434949b25e05SJeff Mahoney 435048a3b636SEric Sandeen static int btrfs_cleanup_transaction(struct btrfs_root *root) 4351acce952bSliubo { 4352acce952bSliubo struct btrfs_transaction *t; 4353acce952bSliubo 4354acce952bSliubo mutex_lock(&root->fs_info->transaction_kthread_mutex); 4355acce952bSliubo 4356a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 4357724e2315SJosef Bacik while (!list_empty(&root->fs_info->trans_list)) { 4358724e2315SJosef Bacik t = list_first_entry(&root->fs_info->trans_list, 4359724e2315SJosef Bacik struct btrfs_transaction, list); 4360724e2315SJosef Bacik if (t->state >= TRANS_STATE_COMMIT_START) { 4361724e2315SJosef Bacik atomic_inc(&t->use_count); 4362724e2315SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 4363724e2315SJosef Bacik btrfs_wait_for_commit(root, t->transid); 4364724e2315SJosef Bacik btrfs_put_transaction(t); 4365724e2315SJosef Bacik spin_lock(&root->fs_info->trans_lock); 4366724e2315SJosef Bacik continue; 4367724e2315SJosef Bacik } 4368724e2315SJosef Bacik if (t == root->fs_info->running_transaction) { 4369724e2315SJosef Bacik t->state = TRANS_STATE_COMMIT_DOING; 4370724e2315SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 4371724e2315SJosef Bacik /* 4372724e2315SJosef Bacik * We wait for 0 num_writers since we don't hold a trans 4373724e2315SJosef Bacik * handle open currently for this transaction. 4374724e2315SJosef Bacik */ 4375724e2315SJosef Bacik wait_event(t->writer_wait, 4376724e2315SJosef Bacik atomic_read(&t->num_writers) == 0); 4377724e2315SJosef Bacik } else { 4378724e2315SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 4379724e2315SJosef Bacik } 4380724e2315SJosef Bacik btrfs_cleanup_one_transaction(t, root); 4381724e2315SJosef Bacik 4382724e2315SJosef Bacik spin_lock(&root->fs_info->trans_lock); 4383724e2315SJosef Bacik if (t == root->fs_info->running_transaction) 4384ac673879SMiao Xie root->fs_info->running_transaction = NULL; 4385724e2315SJosef Bacik list_del_init(&t->list); 4386a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 4387a4abeea4SJosef Bacik 4388724e2315SJosef Bacik btrfs_put_transaction(t); 4389724e2315SJosef Bacik trace_btrfs_transaction_commit(root); 4390724e2315SJosef Bacik spin_lock(&root->fs_info->trans_lock); 4391724e2315SJosef Bacik } 4392724e2315SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 4393199c2a9cSMiao Xie btrfs_destroy_all_ordered_extents(root->fs_info); 439467cde344SMiao Xie btrfs_destroy_delayed_inodes(root); 439567cde344SMiao Xie btrfs_assert_delayed_root_empty(root); 4396724e2315SJosef Bacik btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents); 4397eb73c1b7SMiao Xie btrfs_destroy_all_delalloc_inodes(root->fs_info); 4398acce952bSliubo mutex_unlock(&root->fs_info->transaction_kthread_mutex); 4399acce952bSliubo 4400acce952bSliubo return 0; 4401acce952bSliubo } 4402acce952bSliubo 4403e8c9f186SDavid Sterba static const struct extent_io_ops btree_extent_io_ops = { 4404ce9adaa5SChris Mason .readpage_end_io_hook = btree_readpage_end_io_hook, 44054bb31e92SArne Jansen .readpage_io_failed_hook = btree_io_failed_hook, 44060b86a832SChris Mason .submit_bio_hook = btree_submit_bio_hook, 4407239b14b3SChris Mason /* note we're sharing with inode.c for the merge bio hook */ 4408239b14b3SChris Mason .merge_bio_hook = btrfs_merge_bio_hook, 44090da5468fSChris Mason }; 4410