16cbd5570SChris Mason /* 26cbd5570SChris Mason * Copyright (C) 2007 Oracle. All rights reserved. 36cbd5570SChris Mason * 46cbd5570SChris Mason * This program is free software; you can redistribute it and/or 56cbd5570SChris Mason * modify it under the terms of the GNU General Public 66cbd5570SChris Mason * License v2 as published by the Free Software Foundation. 76cbd5570SChris Mason * 86cbd5570SChris Mason * This program is distributed in the hope that it will be useful, 96cbd5570SChris Mason * but WITHOUT ANY WARRANTY; without even the implied warranty of 106cbd5570SChris Mason * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 116cbd5570SChris Mason * General Public License for more details. 126cbd5570SChris Mason * 136cbd5570SChris Mason * You should have received a copy of the GNU General Public 146cbd5570SChris Mason * License along with this program; if not, write to the 156cbd5570SChris Mason * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 166cbd5570SChris Mason * Boston, MA 021110-1307, USA. 176cbd5570SChris Mason */ 186cbd5570SChris Mason 19e20d96d6SChris Mason #include <linux/fs.h> 20d98237b3SChris Mason #include <linux/blkdev.h> 2187cbda5cSChris Mason #include <linux/scatterlist.h> 2222b0ebdaSChris Mason #include <linux/swap.h> 230f7d52f4SChris Mason #include <linux/radix-tree.h> 2435b7e476SChris Mason #include <linux/writeback.h> 25d397712bSChris Mason #include <linux/buffer_head.h> 26ce9adaa5SChris Mason #include <linux/workqueue.h> 27a74a4b97SChris Mason #include <linux/kthread.h> 28a74a4b97SChris Mason #include <linux/freezer.h> 29163e783eSDavid Woodhouse #include <linux/crc32c.h> 305a0e3ad6STejun Heo #include <linux/slab.h> 31784b4e29SChris Mason #include <linux/migrate.h> 327a36ddecSDavid Sterba #include <linux/ratelimit.h> 337e75bf3fSDavid Sterba #include <asm/unaligned.h> 344b4e25f2SChris Mason #include "compat.h" 35eb60ceacSChris Mason #include "ctree.h" 36eb60ceacSChris Mason #include "disk-io.h" 37e089f05cSChris Mason #include "transaction.h" 380f7d52f4SChris Mason #include "btrfs_inode.h" 390b86a832SChris Mason #include "volumes.h" 40db94535dSChris Mason #include "print-tree.h" 418b712842SChris Mason #include "async-thread.h" 42925baeddSChris Mason #include "locking.h" 43e02119d5SChris Mason #include "tree-log.h" 44fa9c0d79SChris Mason #include "free-space-cache.h" 45581bb050SLi Zefan #include "inode-map.h" 46eb60ceacSChris Mason 47d1310b2eSChris Mason static struct extent_io_ops btree_extent_io_ops; 488b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work); 494df27c4dSYan, Zheng static void free_fs_root(struct btrfs_root *root); 50acce952bSliubo static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info, 51acce952bSliubo int read_only); 52acce952bSliubo static int btrfs_destroy_ordered_operations(struct btrfs_root *root); 53acce952bSliubo static int btrfs_destroy_ordered_extents(struct btrfs_root *root); 54acce952bSliubo static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 55acce952bSliubo struct btrfs_root *root); 56acce952bSliubo static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t); 57acce952bSliubo static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root); 58acce952bSliubo static int btrfs_destroy_marked_extents(struct btrfs_root *root, 59acce952bSliubo struct extent_io_tree *dirty_pages, 60acce952bSliubo int mark); 61acce952bSliubo static int btrfs_destroy_pinned_extent(struct btrfs_root *root, 62acce952bSliubo struct extent_io_tree *pinned_extents); 63acce952bSliubo static int btrfs_cleanup_transaction(struct btrfs_root *root); 64ce9adaa5SChris Mason 65d352ac68SChris Mason /* 66d352ac68SChris Mason * end_io_wq structs are used to do processing in task context when an IO is 67d352ac68SChris Mason * complete. This is used during reads to verify checksums, and it is used 68d352ac68SChris Mason * by writes to insert metadata for new file extents after IO is complete. 69d352ac68SChris Mason */ 70ce9adaa5SChris Mason struct end_io_wq { 71ce9adaa5SChris Mason struct bio *bio; 72ce9adaa5SChris Mason bio_end_io_t *end_io; 73ce9adaa5SChris Mason void *private; 74ce9adaa5SChris Mason struct btrfs_fs_info *info; 75ce9adaa5SChris Mason int error; 7622c59948SChris Mason int metadata; 77ce9adaa5SChris Mason struct list_head list; 788b712842SChris Mason struct btrfs_work work; 79ce9adaa5SChris Mason }; 800da5468fSChris Mason 81d352ac68SChris Mason /* 82d352ac68SChris Mason * async submit bios are used to offload expensive checksumming 83d352ac68SChris Mason * onto the worker threads. They checksum file and metadata bios 84d352ac68SChris Mason * just before they are sent down the IO stack. 85d352ac68SChris Mason */ 8644b8bd7eSChris Mason struct async_submit_bio { 8744b8bd7eSChris Mason struct inode *inode; 8844b8bd7eSChris Mason struct bio *bio; 8944b8bd7eSChris Mason struct list_head list; 904a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_start; 914a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_done; 9244b8bd7eSChris Mason int rw; 9344b8bd7eSChris Mason int mirror_num; 94c8b97818SChris Mason unsigned long bio_flags; 95eaf25d93SChris Mason /* 96eaf25d93SChris Mason * bio_offset is optional, can be used if the pages in the bio 97eaf25d93SChris Mason * can't tell us where in the file the bio should go 98eaf25d93SChris Mason */ 99eaf25d93SChris Mason u64 bio_offset; 1008b712842SChris Mason struct btrfs_work work; 10144b8bd7eSChris Mason }; 10244b8bd7eSChris Mason 1034008c04aSChris Mason /* These are used to set the lockdep class on the extent buffer locks. 1044008c04aSChris Mason * The class is set by the readpage_end_io_hook after the buffer has 1054008c04aSChris Mason * passed csum validation but before the pages are unlocked. 1064008c04aSChris Mason * 1074008c04aSChris Mason * The lockdep class is also set by btrfs_init_new_buffer on freshly 1084008c04aSChris Mason * allocated blocks. 1094008c04aSChris Mason * 1104008c04aSChris Mason * The class is based on the level in the tree block, which allows lockdep 1114008c04aSChris Mason * to know that lower nodes nest inside the locks of higher nodes. 1124008c04aSChris Mason * 1134008c04aSChris Mason * We also add a check to make sure the highest level of the tree is 1144008c04aSChris Mason * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this 1154008c04aSChris Mason * code needs update as well. 1164008c04aSChris Mason */ 1174008c04aSChris Mason #ifdef CONFIG_DEBUG_LOCK_ALLOC 1184008c04aSChris Mason # if BTRFS_MAX_LEVEL != 8 1194008c04aSChris Mason # error 1204008c04aSChris Mason # endif 1214008c04aSChris Mason static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1]; 1224008c04aSChris Mason static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = { 1234008c04aSChris Mason /* leaf */ 1244008c04aSChris Mason "btrfs-extent-00", 1254008c04aSChris Mason "btrfs-extent-01", 1264008c04aSChris Mason "btrfs-extent-02", 1274008c04aSChris Mason "btrfs-extent-03", 1284008c04aSChris Mason "btrfs-extent-04", 1294008c04aSChris Mason "btrfs-extent-05", 1304008c04aSChris Mason "btrfs-extent-06", 1314008c04aSChris Mason "btrfs-extent-07", 1324008c04aSChris Mason /* highest possible level */ 1334008c04aSChris Mason "btrfs-extent-08", 1344008c04aSChris Mason }; 1354008c04aSChris Mason #endif 1364008c04aSChris Mason 137d352ac68SChris Mason /* 138d352ac68SChris Mason * extents on the btree inode are pretty simple, there's one extent 139d352ac68SChris Mason * that covers the entire device 140d352ac68SChris Mason */ 141b2950863SChristoph Hellwig static struct extent_map *btree_get_extent(struct inode *inode, 142306e16ceSDavid Sterba struct page *page, size_t pg_offset, u64 start, u64 len, 1435f39d397SChris Mason int create) 1445f39d397SChris Mason { 1455f39d397SChris Mason struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 1465f39d397SChris Mason struct extent_map *em; 1475f39d397SChris Mason int ret; 1485f39d397SChris Mason 149890871beSChris Mason read_lock(&em_tree->lock); 150d1310b2eSChris Mason em = lookup_extent_mapping(em_tree, start, len); 151a061fc8dSChris Mason if (em) { 152a061fc8dSChris Mason em->bdev = 153a061fc8dSChris Mason BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 154890871beSChris Mason read_unlock(&em_tree->lock); 1555f39d397SChris Mason goto out; 156a061fc8dSChris Mason } 157890871beSChris Mason read_unlock(&em_tree->lock); 1587b13b7b1SChris Mason 159172ddd60SDavid Sterba em = alloc_extent_map(); 1605f39d397SChris Mason if (!em) { 1615f39d397SChris Mason em = ERR_PTR(-ENOMEM); 1625f39d397SChris Mason goto out; 1635f39d397SChris Mason } 1645f39d397SChris Mason em->start = 0; 1650afbaf8cSChris Mason em->len = (u64)-1; 166c8b97818SChris Mason em->block_len = (u64)-1; 1675f39d397SChris Mason em->block_start = 0; 168a061fc8dSChris Mason em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 169d1310b2eSChris Mason 170890871beSChris Mason write_lock(&em_tree->lock); 1715f39d397SChris Mason ret = add_extent_mapping(em_tree, em); 1725f39d397SChris Mason if (ret == -EEXIST) { 1730afbaf8cSChris Mason u64 failed_start = em->start; 1740afbaf8cSChris Mason u64 failed_len = em->len; 1750afbaf8cSChris Mason 1765f39d397SChris Mason free_extent_map(em); 1777b13b7b1SChris Mason em = lookup_extent_mapping(em_tree, start, len); 1780afbaf8cSChris Mason if (em) { 1797b13b7b1SChris Mason ret = 0; 1800afbaf8cSChris Mason } else { 1810afbaf8cSChris Mason em = lookup_extent_mapping(em_tree, failed_start, 1820afbaf8cSChris Mason failed_len); 1837b13b7b1SChris Mason ret = -EIO; 1840afbaf8cSChris Mason } 1855f39d397SChris Mason } else if (ret) { 1867b13b7b1SChris Mason free_extent_map(em); 1877b13b7b1SChris Mason em = NULL; 1885f39d397SChris Mason } 189890871beSChris Mason write_unlock(&em_tree->lock); 1907b13b7b1SChris Mason 1917b13b7b1SChris Mason if (ret) 1927b13b7b1SChris Mason em = ERR_PTR(ret); 1935f39d397SChris Mason out: 1945f39d397SChris Mason return em; 1955f39d397SChris Mason } 1965f39d397SChris Mason 19719c00ddcSChris Mason u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len) 19819c00ddcSChris Mason { 199163e783eSDavid Woodhouse return crc32c(seed, data, len); 20019c00ddcSChris Mason } 20119c00ddcSChris Mason 20219c00ddcSChris Mason void btrfs_csum_final(u32 crc, char *result) 20319c00ddcSChris Mason { 2047e75bf3fSDavid Sterba put_unaligned_le32(~crc, result); 20519c00ddcSChris Mason } 20619c00ddcSChris Mason 207d352ac68SChris Mason /* 208d352ac68SChris Mason * compute the csum for a btree block, and either verify it or write it 209d352ac68SChris Mason * into the csum field of the block. 210d352ac68SChris Mason */ 21119c00ddcSChris Mason static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf, 21219c00ddcSChris Mason int verify) 21319c00ddcSChris Mason { 214607d432dSJosef Bacik u16 csum_size = 215607d432dSJosef Bacik btrfs_super_csum_size(&root->fs_info->super_copy); 216607d432dSJosef Bacik char *result = NULL; 21719c00ddcSChris Mason unsigned long len; 21819c00ddcSChris Mason unsigned long cur_len; 21919c00ddcSChris Mason unsigned long offset = BTRFS_CSUM_SIZE; 22019c00ddcSChris Mason char *map_token = NULL; 22119c00ddcSChris Mason char *kaddr; 22219c00ddcSChris Mason unsigned long map_start; 22319c00ddcSChris Mason unsigned long map_len; 22419c00ddcSChris Mason int err; 22519c00ddcSChris Mason u32 crc = ~(u32)0; 226607d432dSJosef Bacik unsigned long inline_result; 22719c00ddcSChris Mason 22819c00ddcSChris Mason len = buf->len - offset; 22919c00ddcSChris Mason while (len > 0) { 23019c00ddcSChris Mason err = map_private_extent_buffer(buf, offset, 32, 23119c00ddcSChris Mason &map_token, &kaddr, 23219c00ddcSChris Mason &map_start, &map_len, KM_USER0); 233d397712bSChris Mason if (err) 23419c00ddcSChris Mason return 1; 23519c00ddcSChris Mason cur_len = min(len, map_len - (offset - map_start)); 23619c00ddcSChris Mason crc = btrfs_csum_data(root, kaddr + offset - map_start, 23719c00ddcSChris Mason crc, cur_len); 23819c00ddcSChris Mason len -= cur_len; 23919c00ddcSChris Mason offset += cur_len; 24019c00ddcSChris Mason unmap_extent_buffer(buf, map_token, KM_USER0); 24119c00ddcSChris Mason } 242607d432dSJosef Bacik if (csum_size > sizeof(inline_result)) { 243607d432dSJosef Bacik result = kzalloc(csum_size * sizeof(char), GFP_NOFS); 244607d432dSJosef Bacik if (!result) 245607d432dSJosef Bacik return 1; 246607d432dSJosef Bacik } else { 247607d432dSJosef Bacik result = (char *)&inline_result; 248607d432dSJosef Bacik } 249607d432dSJosef Bacik 25019c00ddcSChris Mason btrfs_csum_final(crc, result); 25119c00ddcSChris Mason 25219c00ddcSChris Mason if (verify) { 253607d432dSJosef Bacik if (memcmp_extent_buffer(buf, result, 0, csum_size)) { 254e4204dedSChris Mason u32 val; 255e4204dedSChris Mason u32 found = 0; 256607d432dSJosef Bacik memcpy(&found, result, csum_size); 257e4204dedSChris Mason 258607d432dSJosef Bacik read_extent_buffer(buf, &val, 0, csum_size); 2597a36ddecSDavid Sterba printk_ratelimited(KERN_INFO "btrfs: %s checksum verify " 260193f284dSChris Mason "failed on %llu wanted %X found %X " 261193f284dSChris Mason "level %d\n", 26219c00ddcSChris Mason root->fs_info->sb->s_id, 263193f284dSChris Mason (unsigned long long)buf->start, val, found, 264193f284dSChris Mason btrfs_header_level(buf)); 265607d432dSJosef Bacik if (result != (char *)&inline_result) 266607d432dSJosef Bacik kfree(result); 26719c00ddcSChris Mason return 1; 26819c00ddcSChris Mason } 26919c00ddcSChris Mason } else { 270607d432dSJosef Bacik write_extent_buffer(buf, result, 0, csum_size); 27119c00ddcSChris Mason } 272607d432dSJosef Bacik if (result != (char *)&inline_result) 273607d432dSJosef Bacik kfree(result); 27419c00ddcSChris Mason return 0; 27519c00ddcSChris Mason } 27619c00ddcSChris Mason 277d352ac68SChris Mason /* 278d352ac68SChris Mason * we can't consider a given block up to date unless the transid of the 279d352ac68SChris Mason * block matches the transid in the parent node's pointer. This is how we 280d352ac68SChris Mason * detect blocks that either didn't get written at all or got written 281d352ac68SChris Mason * in the wrong place. 282d352ac68SChris Mason */ 2831259ab75SChris Mason static int verify_parent_transid(struct extent_io_tree *io_tree, 2841259ab75SChris Mason struct extent_buffer *eb, u64 parent_transid) 2851259ab75SChris Mason { 2862ac55d41SJosef Bacik struct extent_state *cached_state = NULL; 2871259ab75SChris Mason int ret; 2881259ab75SChris Mason 2891259ab75SChris Mason if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 2901259ab75SChris Mason return 0; 2911259ab75SChris Mason 2922ac55d41SJosef Bacik lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 2932ac55d41SJosef Bacik 0, &cached_state, GFP_NOFS); 2942ac55d41SJosef Bacik if (extent_buffer_uptodate(io_tree, eb, cached_state) && 2951259ab75SChris Mason btrfs_header_generation(eb) == parent_transid) { 2961259ab75SChris Mason ret = 0; 2971259ab75SChris Mason goto out; 2981259ab75SChris Mason } 2997a36ddecSDavid Sterba printk_ratelimited("parent transid verify failed on %llu wanted %llu " 300193f284dSChris Mason "found %llu\n", 3011259ab75SChris Mason (unsigned long long)eb->start, 3021259ab75SChris Mason (unsigned long long)parent_transid, 3031259ab75SChris Mason (unsigned long long)btrfs_header_generation(eb)); 3041259ab75SChris Mason ret = 1; 3052ac55d41SJosef Bacik clear_extent_buffer_uptodate(io_tree, eb, &cached_state); 30633958dc6SChris Mason out: 3072ac55d41SJosef Bacik unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 3082ac55d41SJosef Bacik &cached_state, GFP_NOFS); 3091259ab75SChris Mason return ret; 3101259ab75SChris Mason } 3111259ab75SChris Mason 312d352ac68SChris Mason /* 313d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 314d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 315d352ac68SChris Mason */ 316f188591eSChris Mason static int btree_read_extent_buffer_pages(struct btrfs_root *root, 317f188591eSChris Mason struct extent_buffer *eb, 318ca7a79adSChris Mason u64 start, u64 parent_transid) 319f188591eSChris Mason { 320f188591eSChris Mason struct extent_io_tree *io_tree; 321f188591eSChris Mason int ret; 322f188591eSChris Mason int num_copies = 0; 323f188591eSChris Mason int mirror_num = 0; 324f188591eSChris Mason 325a826d6dcSJosef Bacik clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 326f188591eSChris Mason io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree; 327f188591eSChris Mason while (1) { 328f188591eSChris Mason ret = read_extent_buffer_pages(io_tree, eb, start, 1, 329f188591eSChris Mason btree_get_extent, mirror_num); 3301259ab75SChris Mason if (!ret && 3311259ab75SChris Mason !verify_parent_transid(io_tree, eb, parent_transid)) 332f188591eSChris Mason return ret; 333d397712bSChris Mason 334a826d6dcSJosef Bacik /* 335a826d6dcSJosef Bacik * This buffer's crc is fine, but its contents are corrupted, so 336a826d6dcSJosef Bacik * there is no reason to read the other copies, they won't be 337a826d6dcSJosef Bacik * any less wrong. 338a826d6dcSJosef Bacik */ 339a826d6dcSJosef Bacik if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags)) 340a826d6dcSJosef Bacik return ret; 341a826d6dcSJosef Bacik 342f188591eSChris Mason num_copies = btrfs_num_copies(&root->fs_info->mapping_tree, 343f188591eSChris Mason eb->start, eb->len); 3444235298eSChris Mason if (num_copies == 1) 345f188591eSChris Mason return ret; 3464235298eSChris Mason 347f188591eSChris Mason mirror_num++; 3484235298eSChris Mason if (mirror_num > num_copies) 349f188591eSChris Mason return ret; 350f188591eSChris Mason } 351f188591eSChris Mason return -EIO; 352f188591eSChris Mason } 35319c00ddcSChris Mason 354d352ac68SChris Mason /* 355d397712bSChris Mason * checksum a dirty tree block before IO. This has extra checks to make sure 356d397712bSChris Mason * we only fill in the checksum field in the first page of a multi-page block 357d352ac68SChris Mason */ 358d397712bSChris Mason 359b2950863SChristoph Hellwig static int csum_dirty_buffer(struct btrfs_root *root, struct page *page) 36019c00ddcSChris Mason { 361d1310b2eSChris Mason struct extent_io_tree *tree; 36235ebb934SChris Mason u64 start = (u64)page->index << PAGE_CACHE_SHIFT; 36319c00ddcSChris Mason u64 found_start; 36419c00ddcSChris Mason unsigned long len; 36519c00ddcSChris Mason struct extent_buffer *eb; 366f188591eSChris Mason int ret; 367f188591eSChris Mason 368d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 36919c00ddcSChris Mason 370eb14ab8eSChris Mason if (page->private == EXTENT_PAGE_PRIVATE) { 371eb14ab8eSChris Mason WARN_ON(1); 37219c00ddcSChris Mason goto out; 373eb14ab8eSChris Mason } 374eb14ab8eSChris Mason if (!page->private) { 375eb14ab8eSChris Mason WARN_ON(1); 37619c00ddcSChris Mason goto out; 377eb14ab8eSChris Mason } 37819c00ddcSChris Mason len = page->private >> 2; 379d397712bSChris Mason WARN_ON(len == 0); 380d397712bSChris Mason 381ba144192SDavid Sterba eb = alloc_extent_buffer(tree, start, len, page); 38291ca338dSTsutomu Itoh if (eb == NULL) { 38391ca338dSTsutomu Itoh WARN_ON(1); 38491ca338dSTsutomu Itoh goto out; 38591ca338dSTsutomu Itoh } 386ca7a79adSChris Mason ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE, 387ca7a79adSChris Mason btrfs_header_generation(eb)); 388f188591eSChris Mason BUG_ON(ret); 389784b4e29SChris Mason WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN)); 390784b4e29SChris Mason 39119c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 39219c00ddcSChris Mason if (found_start != start) { 39355c69072SChris Mason WARN_ON(1); 39455c69072SChris Mason goto err; 39555c69072SChris Mason } 39655c69072SChris Mason if (eb->first_page != page) { 39755c69072SChris Mason WARN_ON(1); 39855c69072SChris Mason goto err; 39955c69072SChris Mason } 40055c69072SChris Mason if (!PageUptodate(page)) { 40155c69072SChris Mason WARN_ON(1); 40255c69072SChris Mason goto err; 40319c00ddcSChris Mason } 40419c00ddcSChris Mason csum_tree_block(root, eb, 0); 40555c69072SChris Mason err: 40619c00ddcSChris Mason free_extent_buffer(eb); 40719c00ddcSChris Mason out: 40819c00ddcSChris Mason return 0; 40919c00ddcSChris Mason } 41019c00ddcSChris Mason 4112b82032cSYan Zheng static int check_tree_block_fsid(struct btrfs_root *root, 4122b82032cSYan Zheng struct extent_buffer *eb) 4132b82032cSYan Zheng { 4142b82032cSYan Zheng struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; 4152b82032cSYan Zheng u8 fsid[BTRFS_UUID_SIZE]; 4162b82032cSYan Zheng int ret = 1; 4172b82032cSYan Zheng 4182b82032cSYan Zheng read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb), 4192b82032cSYan Zheng BTRFS_FSID_SIZE); 4202b82032cSYan Zheng while (fs_devices) { 4212b82032cSYan Zheng if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) { 4222b82032cSYan Zheng ret = 0; 4232b82032cSYan Zheng break; 4242b82032cSYan Zheng } 4252b82032cSYan Zheng fs_devices = fs_devices->seed; 4262b82032cSYan Zheng } 4272b82032cSYan Zheng return ret; 4282b82032cSYan Zheng } 4292b82032cSYan Zheng 430a826d6dcSJosef Bacik #define CORRUPT(reason, eb, root, slot) \ 431a826d6dcSJosef Bacik printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \ 432a826d6dcSJosef Bacik "root=%llu, slot=%d\n", reason, \ 433a826d6dcSJosef Bacik (unsigned long long)btrfs_header_bytenr(eb), \ 434a826d6dcSJosef Bacik (unsigned long long)root->objectid, slot) 435a826d6dcSJosef Bacik 436a826d6dcSJosef Bacik static noinline int check_leaf(struct btrfs_root *root, 437a826d6dcSJosef Bacik struct extent_buffer *leaf) 438a826d6dcSJosef Bacik { 439a826d6dcSJosef Bacik struct btrfs_key key; 440a826d6dcSJosef Bacik struct btrfs_key leaf_key; 441a826d6dcSJosef Bacik u32 nritems = btrfs_header_nritems(leaf); 442a826d6dcSJosef Bacik int slot; 443a826d6dcSJosef Bacik 444a826d6dcSJosef Bacik if (nritems == 0) 445a826d6dcSJosef Bacik return 0; 446a826d6dcSJosef Bacik 447a826d6dcSJosef Bacik /* Check the 0 item */ 448a826d6dcSJosef Bacik if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) != 449a826d6dcSJosef Bacik BTRFS_LEAF_DATA_SIZE(root)) { 450a826d6dcSJosef Bacik CORRUPT("invalid item offset size pair", leaf, root, 0); 451a826d6dcSJosef Bacik return -EIO; 452a826d6dcSJosef Bacik } 453a826d6dcSJosef Bacik 454a826d6dcSJosef Bacik /* 455a826d6dcSJosef Bacik * Check to make sure each items keys are in the correct order and their 456a826d6dcSJosef Bacik * offsets make sense. We only have to loop through nritems-1 because 457a826d6dcSJosef Bacik * we check the current slot against the next slot, which verifies the 458a826d6dcSJosef Bacik * next slot's offset+size makes sense and that the current's slot 459a826d6dcSJosef Bacik * offset is correct. 460a826d6dcSJosef Bacik */ 461a826d6dcSJosef Bacik for (slot = 0; slot < nritems - 1; slot++) { 462a826d6dcSJosef Bacik btrfs_item_key_to_cpu(leaf, &leaf_key, slot); 463a826d6dcSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, slot + 1); 464a826d6dcSJosef Bacik 465a826d6dcSJosef Bacik /* Make sure the keys are in the right order */ 466a826d6dcSJosef Bacik if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) { 467a826d6dcSJosef Bacik CORRUPT("bad key order", leaf, root, slot); 468a826d6dcSJosef Bacik return -EIO; 469a826d6dcSJosef Bacik } 470a826d6dcSJosef Bacik 471a826d6dcSJosef Bacik /* 472a826d6dcSJosef Bacik * Make sure the offset and ends are right, remember that the 473a826d6dcSJosef Bacik * item data starts at the end of the leaf and grows towards the 474a826d6dcSJosef Bacik * front. 475a826d6dcSJosef Bacik */ 476a826d6dcSJosef Bacik if (btrfs_item_offset_nr(leaf, slot) != 477a826d6dcSJosef Bacik btrfs_item_end_nr(leaf, slot + 1)) { 478a826d6dcSJosef Bacik CORRUPT("slot offset bad", leaf, root, slot); 479a826d6dcSJosef Bacik return -EIO; 480a826d6dcSJosef Bacik } 481a826d6dcSJosef Bacik 482a826d6dcSJosef Bacik /* 483a826d6dcSJosef Bacik * Check to make sure that we don't point outside of the leaf, 484a826d6dcSJosef Bacik * just incase all the items are consistent to eachother, but 485a826d6dcSJosef Bacik * all point outside of the leaf. 486a826d6dcSJosef Bacik */ 487a826d6dcSJosef Bacik if (btrfs_item_end_nr(leaf, slot) > 488a826d6dcSJosef Bacik BTRFS_LEAF_DATA_SIZE(root)) { 489a826d6dcSJosef Bacik CORRUPT("slot end outside of leaf", leaf, root, slot); 490a826d6dcSJosef Bacik return -EIO; 491a826d6dcSJosef Bacik } 492a826d6dcSJosef Bacik } 493a826d6dcSJosef Bacik 494a826d6dcSJosef Bacik return 0; 495a826d6dcSJosef Bacik } 496a826d6dcSJosef Bacik 4974008c04aSChris Mason #ifdef CONFIG_DEBUG_LOCK_ALLOC 4984008c04aSChris Mason void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level) 4994008c04aSChris Mason { 5004008c04aSChris Mason lockdep_set_class_and_name(&eb->lock, 5014008c04aSChris Mason &btrfs_eb_class[level], 5024008c04aSChris Mason btrfs_eb_name[level]); 5034008c04aSChris Mason } 5044008c04aSChris Mason #endif 5054008c04aSChris Mason 506b2950863SChristoph Hellwig static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end, 507ce9adaa5SChris Mason struct extent_state *state) 508ce9adaa5SChris Mason { 509ce9adaa5SChris Mason struct extent_io_tree *tree; 510ce9adaa5SChris Mason u64 found_start; 511ce9adaa5SChris Mason int found_level; 512ce9adaa5SChris Mason unsigned long len; 513ce9adaa5SChris Mason struct extent_buffer *eb; 514ce9adaa5SChris Mason struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 515f188591eSChris Mason int ret = 0; 516ce9adaa5SChris Mason 517ce9adaa5SChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 518ce9adaa5SChris Mason if (page->private == EXTENT_PAGE_PRIVATE) 519ce9adaa5SChris Mason goto out; 520ce9adaa5SChris Mason if (!page->private) 521ce9adaa5SChris Mason goto out; 522d397712bSChris Mason 523ce9adaa5SChris Mason len = page->private >> 2; 524d397712bSChris Mason WARN_ON(len == 0); 525d397712bSChris Mason 526ba144192SDavid Sterba eb = alloc_extent_buffer(tree, start, len, page); 52791ca338dSTsutomu Itoh if (eb == NULL) { 52891ca338dSTsutomu Itoh ret = -EIO; 52991ca338dSTsutomu Itoh goto out; 53091ca338dSTsutomu Itoh } 531f188591eSChris Mason 532ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 53323a07867SChris Mason if (found_start != start) { 5347a36ddecSDavid Sterba printk_ratelimited(KERN_INFO "btrfs bad tree block start " 535193f284dSChris Mason "%llu %llu\n", 536a1b32a59SChris Mason (unsigned long long)found_start, 537a1b32a59SChris Mason (unsigned long long)eb->start); 538f188591eSChris Mason ret = -EIO; 539ce9adaa5SChris Mason goto err; 540ce9adaa5SChris Mason } 541ce9adaa5SChris Mason if (eb->first_page != page) { 542d397712bSChris Mason printk(KERN_INFO "btrfs bad first page %lu %lu\n", 543d397712bSChris Mason eb->first_page->index, page->index); 544ce9adaa5SChris Mason WARN_ON(1); 545f188591eSChris Mason ret = -EIO; 546ce9adaa5SChris Mason goto err; 547ce9adaa5SChris Mason } 5482b82032cSYan Zheng if (check_tree_block_fsid(root, eb)) { 5497a36ddecSDavid Sterba printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n", 550d397712bSChris Mason (unsigned long long)eb->start); 5511259ab75SChris Mason ret = -EIO; 5521259ab75SChris Mason goto err; 5531259ab75SChris Mason } 554ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 555ce9adaa5SChris Mason 5564008c04aSChris Mason btrfs_set_buffer_lockdep_class(eb, found_level); 5574008c04aSChris Mason 558ce9adaa5SChris Mason ret = csum_tree_block(root, eb, 1); 559a826d6dcSJosef Bacik if (ret) { 560f188591eSChris Mason ret = -EIO; 561a826d6dcSJosef Bacik goto err; 562a826d6dcSJosef Bacik } 563a826d6dcSJosef Bacik 564a826d6dcSJosef Bacik /* 565a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 566a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 567a826d6dcSJosef Bacik * return -EIO. 568a826d6dcSJosef Bacik */ 569a826d6dcSJosef Bacik if (found_level == 0 && check_leaf(root, eb)) { 570a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 571a826d6dcSJosef Bacik ret = -EIO; 572a826d6dcSJosef Bacik } 573ce9adaa5SChris Mason 574ce9adaa5SChris Mason end = min_t(u64, eb->len, PAGE_CACHE_SIZE); 575ce9adaa5SChris Mason end = eb->start + end - 1; 576ce9adaa5SChris Mason err: 577ce9adaa5SChris Mason free_extent_buffer(eb); 578ce9adaa5SChris Mason out: 579f188591eSChris Mason return ret; 580ce9adaa5SChris Mason } 581ce9adaa5SChris Mason 582ce9adaa5SChris Mason static void end_workqueue_bio(struct bio *bio, int err) 583ce9adaa5SChris Mason { 584ce9adaa5SChris Mason struct end_io_wq *end_io_wq = bio->bi_private; 585ce9adaa5SChris Mason struct btrfs_fs_info *fs_info; 586ce9adaa5SChris Mason 587ce9adaa5SChris Mason fs_info = end_io_wq->info; 588ce9adaa5SChris Mason end_io_wq->error = err; 5898b712842SChris Mason end_io_wq->work.func = end_workqueue_fn; 5908b712842SChris Mason end_io_wq->work.flags = 0; 591d20f7043SChris Mason 5927b6d91daSChristoph Hellwig if (bio->bi_rw & REQ_WRITE) { 5930cb59c99SJosef Bacik if (end_io_wq->metadata == 1) 594cad321adSChris Mason btrfs_queue_worker(&fs_info->endio_meta_write_workers, 595cad321adSChris Mason &end_io_wq->work); 5960cb59c99SJosef Bacik else if (end_io_wq->metadata == 2) 5970cb59c99SJosef Bacik btrfs_queue_worker(&fs_info->endio_freespace_worker, 5980cb59c99SJosef Bacik &end_io_wq->work); 599cad321adSChris Mason else 600e6dcd2dcSChris Mason btrfs_queue_worker(&fs_info->endio_write_workers, 601e6dcd2dcSChris Mason &end_io_wq->work); 602d20f7043SChris Mason } else { 603d20f7043SChris Mason if (end_io_wq->metadata) 604d20f7043SChris Mason btrfs_queue_worker(&fs_info->endio_meta_workers, 605d20f7043SChris Mason &end_io_wq->work); 606e6dcd2dcSChris Mason else 607d20f7043SChris Mason btrfs_queue_worker(&fs_info->endio_workers, 608d20f7043SChris Mason &end_io_wq->work); 609d20f7043SChris Mason } 610ce9adaa5SChris Mason } 611ce9adaa5SChris Mason 6120cb59c99SJosef Bacik /* 6130cb59c99SJosef Bacik * For the metadata arg you want 6140cb59c99SJosef Bacik * 6150cb59c99SJosef Bacik * 0 - if data 6160cb59c99SJosef Bacik * 1 - if normal metadta 6170cb59c99SJosef Bacik * 2 - if writing to the free space cache area 6180cb59c99SJosef Bacik */ 61922c59948SChris Mason int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 62022c59948SChris Mason int metadata) 6210b86a832SChris Mason { 622ce9adaa5SChris Mason struct end_io_wq *end_io_wq; 623ce9adaa5SChris Mason end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS); 624ce9adaa5SChris Mason if (!end_io_wq) 625ce9adaa5SChris Mason return -ENOMEM; 626ce9adaa5SChris Mason 627ce9adaa5SChris Mason end_io_wq->private = bio->bi_private; 628ce9adaa5SChris Mason end_io_wq->end_io = bio->bi_end_io; 62922c59948SChris Mason end_io_wq->info = info; 630ce9adaa5SChris Mason end_io_wq->error = 0; 631ce9adaa5SChris Mason end_io_wq->bio = bio; 63222c59948SChris Mason end_io_wq->metadata = metadata; 633ce9adaa5SChris Mason 634ce9adaa5SChris Mason bio->bi_private = end_io_wq; 635ce9adaa5SChris Mason bio->bi_end_io = end_workqueue_bio; 63622c59948SChris Mason return 0; 63722c59948SChris Mason } 63822c59948SChris Mason 639b64a2851SChris Mason unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info) 6404854ddd0SChris Mason { 6414854ddd0SChris Mason unsigned long limit = min_t(unsigned long, 6424854ddd0SChris Mason info->workers.max_workers, 6434854ddd0SChris Mason info->fs_devices->open_devices); 6444854ddd0SChris Mason return 256 * limit; 6454854ddd0SChris Mason } 6464854ddd0SChris Mason 6474a69a410SChris Mason static void run_one_async_start(struct btrfs_work *work) 6484a69a410SChris Mason { 6494a69a410SChris Mason struct async_submit_bio *async; 6504a69a410SChris Mason 6514a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 6524a69a410SChris Mason async->submit_bio_start(async->inode, async->rw, async->bio, 653eaf25d93SChris Mason async->mirror_num, async->bio_flags, 654eaf25d93SChris Mason async->bio_offset); 6554a69a410SChris Mason } 6564a69a410SChris Mason 6574a69a410SChris Mason static void run_one_async_done(struct btrfs_work *work) 6588b712842SChris Mason { 6598b712842SChris Mason struct btrfs_fs_info *fs_info; 6608b712842SChris Mason struct async_submit_bio *async; 6614854ddd0SChris Mason int limit; 6628b712842SChris Mason 6638b712842SChris Mason async = container_of(work, struct async_submit_bio, work); 6648b712842SChris Mason fs_info = BTRFS_I(async->inode)->root->fs_info; 6654854ddd0SChris Mason 666b64a2851SChris Mason limit = btrfs_async_submit_limit(fs_info); 6674854ddd0SChris Mason limit = limit * 2 / 3; 6684854ddd0SChris Mason 6698b712842SChris Mason atomic_dec(&fs_info->nr_async_submits); 6700986fe9eSChris Mason 671b64a2851SChris Mason if (atomic_read(&fs_info->nr_async_submits) < limit && 672b64a2851SChris Mason waitqueue_active(&fs_info->async_submit_wait)) 6734854ddd0SChris Mason wake_up(&fs_info->async_submit_wait); 6744854ddd0SChris Mason 6754a69a410SChris Mason async->submit_bio_done(async->inode, async->rw, async->bio, 676eaf25d93SChris Mason async->mirror_num, async->bio_flags, 677eaf25d93SChris Mason async->bio_offset); 6784a69a410SChris Mason } 6794a69a410SChris Mason 6804a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 6814a69a410SChris Mason { 6824a69a410SChris Mason struct async_submit_bio *async; 6834a69a410SChris Mason 6844a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 6858b712842SChris Mason kfree(async); 6868b712842SChris Mason } 6878b712842SChris Mason 68844b8bd7eSChris Mason int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode, 68944b8bd7eSChris Mason int rw, struct bio *bio, int mirror_num, 690c8b97818SChris Mason unsigned long bio_flags, 691eaf25d93SChris Mason u64 bio_offset, 6924a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_start, 6934a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_done) 69444b8bd7eSChris Mason { 69544b8bd7eSChris Mason struct async_submit_bio *async; 69644b8bd7eSChris Mason 69744b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 69844b8bd7eSChris Mason if (!async) 69944b8bd7eSChris Mason return -ENOMEM; 70044b8bd7eSChris Mason 70144b8bd7eSChris Mason async->inode = inode; 70244b8bd7eSChris Mason async->rw = rw; 70344b8bd7eSChris Mason async->bio = bio; 70444b8bd7eSChris Mason async->mirror_num = mirror_num; 7054a69a410SChris Mason async->submit_bio_start = submit_bio_start; 7064a69a410SChris Mason async->submit_bio_done = submit_bio_done; 7074a69a410SChris Mason 7084a69a410SChris Mason async->work.func = run_one_async_start; 7094a69a410SChris Mason async->work.ordered_func = run_one_async_done; 7104a69a410SChris Mason async->work.ordered_free = run_one_async_free; 7114a69a410SChris Mason 7128b712842SChris Mason async->work.flags = 0; 713c8b97818SChris Mason async->bio_flags = bio_flags; 714eaf25d93SChris Mason async->bio_offset = bio_offset; 7158c8bee1dSChris Mason 716cb03c743SChris Mason atomic_inc(&fs_info->nr_async_submits); 717d313d7a3SChris Mason 7187b6d91daSChristoph Hellwig if (rw & REQ_SYNC) 719d313d7a3SChris Mason btrfs_set_work_high_prio(&async->work); 720d313d7a3SChris Mason 7218b712842SChris Mason btrfs_queue_worker(&fs_info->workers, &async->work); 7229473f16cSChris Mason 723771ed689SChris Mason while (atomic_read(&fs_info->async_submit_draining) && 724771ed689SChris Mason atomic_read(&fs_info->nr_async_submits)) { 725771ed689SChris Mason wait_event(fs_info->async_submit_wait, 726771ed689SChris Mason (atomic_read(&fs_info->nr_async_submits) == 0)); 727771ed689SChris Mason } 728771ed689SChris Mason 72944b8bd7eSChris Mason return 0; 73044b8bd7eSChris Mason } 73144b8bd7eSChris Mason 732ce3ed71aSChris Mason static int btree_csum_one_bio(struct bio *bio) 733ce3ed71aSChris Mason { 734ce3ed71aSChris Mason struct bio_vec *bvec = bio->bi_io_vec; 735ce3ed71aSChris Mason int bio_index = 0; 736ce3ed71aSChris Mason struct btrfs_root *root; 737ce3ed71aSChris Mason 738ce3ed71aSChris Mason WARN_ON(bio->bi_vcnt <= 0); 739ce3ed71aSChris Mason while (bio_index < bio->bi_vcnt) { 740ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 741ce3ed71aSChris Mason csum_dirty_buffer(root, bvec->bv_page); 742ce3ed71aSChris Mason bio_index++; 743ce3ed71aSChris Mason bvec++; 744ce3ed71aSChris Mason } 745ce3ed71aSChris Mason return 0; 746ce3ed71aSChris Mason } 747ce3ed71aSChris Mason 7484a69a410SChris Mason static int __btree_submit_bio_start(struct inode *inode, int rw, 7494a69a410SChris Mason struct bio *bio, int mirror_num, 750eaf25d93SChris Mason unsigned long bio_flags, 751eaf25d93SChris Mason u64 bio_offset) 75222c59948SChris Mason { 7538b712842SChris Mason /* 7548b712842SChris Mason * when we're called for a write, we're already in the async 7555443be45SChris Mason * submission context. Just jump into btrfs_map_bio 7568b712842SChris Mason */ 757ce3ed71aSChris Mason btree_csum_one_bio(bio); 7584a69a410SChris Mason return 0; 75922c59948SChris Mason } 76022c59948SChris Mason 7614a69a410SChris Mason static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio, 762eaf25d93SChris Mason int mirror_num, unsigned long bio_flags, 763eaf25d93SChris Mason u64 bio_offset) 7644a69a410SChris Mason { 7658b712842SChris Mason /* 7664a69a410SChris Mason * when we're called for a write, we're already in the async 7674a69a410SChris Mason * submission context. Just jump into btrfs_map_bio 7688b712842SChris Mason */ 7698b712842SChris Mason return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1); 7700b86a832SChris Mason } 7710b86a832SChris Mason 77244b8bd7eSChris Mason static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, 773eaf25d93SChris Mason int mirror_num, unsigned long bio_flags, 774eaf25d93SChris Mason u64 bio_offset) 77544b8bd7eSChris Mason { 7764a69a410SChris Mason int ret; 777cad321adSChris Mason 7784a69a410SChris Mason ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info, 7794a69a410SChris Mason bio, 1); 7804a69a410SChris Mason BUG_ON(ret); 7814a69a410SChris Mason 7827b6d91daSChristoph Hellwig if (!(rw & REQ_WRITE)) { 783cad321adSChris Mason /* 784cad321adSChris Mason * called for a read, do the setup so that checksum validation 785cad321adSChris Mason * can happen in the async kernel threads 786cad321adSChris Mason */ 7874a69a410SChris Mason return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, 7886f3577bdSChris Mason mirror_num, 0); 78944b8bd7eSChris Mason } 790d313d7a3SChris Mason 791cad321adSChris Mason /* 792cad321adSChris Mason * kthread helpers are used to submit writes so that checksumming 793cad321adSChris Mason * can happen in parallel across all CPUs 794cad321adSChris Mason */ 79544b8bd7eSChris Mason return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, 796c8b97818SChris Mason inode, rw, bio, mirror_num, 0, 797eaf25d93SChris Mason bio_offset, 7984a69a410SChris Mason __btree_submit_bio_start, 7994a69a410SChris Mason __btree_submit_bio_done); 80044b8bd7eSChris Mason } 80144b8bd7eSChris Mason 8023dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 803784b4e29SChris Mason static int btree_migratepage(struct address_space *mapping, 804784b4e29SChris Mason struct page *newpage, struct page *page) 805784b4e29SChris Mason { 806784b4e29SChris Mason /* 807784b4e29SChris Mason * we can't safely write a btree page from here, 808784b4e29SChris Mason * we haven't done the locking hook 809784b4e29SChris Mason */ 810784b4e29SChris Mason if (PageDirty(page)) 811784b4e29SChris Mason return -EAGAIN; 812784b4e29SChris Mason /* 813784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 814784b4e29SChris Mason * We must have no buffers or drop them. 815784b4e29SChris Mason */ 816784b4e29SChris Mason if (page_has_private(page) && 817784b4e29SChris Mason !try_to_release_page(page, GFP_KERNEL)) 818784b4e29SChris Mason return -EAGAIN; 819784b4e29SChris Mason return migrate_page(mapping, newpage, page); 820784b4e29SChris Mason } 8213dd1462eSJan Beulich #endif 822784b4e29SChris Mason 8235f39d397SChris Mason static int btree_writepage(struct page *page, struct writeback_control *wbc) 8245f39d397SChris Mason { 825d1310b2eSChris Mason struct extent_io_tree *tree; 826b9473439SChris Mason struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 827b9473439SChris Mason struct extent_buffer *eb; 828b9473439SChris Mason int was_dirty; 8295443be45SChris Mason 830b9473439SChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 831b9473439SChris Mason if (!(current->flags & PF_MEMALLOC)) { 832b9473439SChris Mason return extent_write_full_page(tree, page, 833b9473439SChris Mason btree_get_extent, wbc); 834b9473439SChris Mason } 835b9473439SChris Mason 8365443be45SChris Mason redirty_page_for_writepage(wbc, page); 837784b4e29SChris Mason eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE); 838b9473439SChris Mason WARN_ON(!eb); 839b9473439SChris Mason 840b9473439SChris Mason was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); 841b9473439SChris Mason if (!was_dirty) { 842b9473439SChris Mason spin_lock(&root->fs_info->delalloc_lock); 843b9473439SChris Mason root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE; 844b9473439SChris Mason spin_unlock(&root->fs_info->delalloc_lock); 845b9473439SChris Mason } 846b9473439SChris Mason free_extent_buffer(eb); 847b9473439SChris Mason 8485443be45SChris Mason unlock_page(page); 8495443be45SChris Mason return 0; 8505443be45SChris Mason } 8510da5468fSChris Mason 8520da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 8530da5468fSChris Mason struct writeback_control *wbc) 8540da5468fSChris Mason { 855d1310b2eSChris Mason struct extent_io_tree *tree; 856d1310b2eSChris Mason tree = &BTRFS_I(mapping->host)->io_tree; 857d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 858b9473439SChris Mason struct btrfs_root *root = BTRFS_I(mapping->host)->root; 859793955bcSChris Mason u64 num_dirty; 86024ab9cd8SChris Mason unsigned long thresh = 32 * 1024 * 1024; 861448d640bSChris Mason 862448d640bSChris Mason if (wbc->for_kupdate) 863448d640bSChris Mason return 0; 864448d640bSChris Mason 865b9473439SChris Mason /* this is a bit racy, but that's ok */ 866b9473439SChris Mason num_dirty = root->fs_info->dirty_metadata_bytes; 867d397712bSChris Mason if (num_dirty < thresh) 868793955bcSChris Mason return 0; 869793955bcSChris Mason } 8700da5468fSChris Mason return extent_writepages(tree, mapping, btree_get_extent, wbc); 8710da5468fSChris Mason } 8720da5468fSChris Mason 873b2950863SChristoph Hellwig static int btree_readpage(struct file *file, struct page *page) 8745f39d397SChris Mason { 875d1310b2eSChris Mason struct extent_io_tree *tree; 876d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 8775f39d397SChris Mason return extent_read_full_page(tree, page, btree_get_extent); 8785f39d397SChris Mason } 8795f39d397SChris Mason 88070dec807SChris Mason static int btree_releasepage(struct page *page, gfp_t gfp_flags) 8815f39d397SChris Mason { 882d1310b2eSChris Mason struct extent_io_tree *tree; 883d1310b2eSChris Mason struct extent_map_tree *map; 8845f39d397SChris Mason int ret; 8855f39d397SChris Mason 88698509cfcSChris Mason if (PageWriteback(page) || PageDirty(page)) 88798509cfcSChris Mason return 0; 88898509cfcSChris Mason 889d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 890d1310b2eSChris Mason map = &BTRFS_I(page->mapping->host)->extent_tree; 8916af118ceSChris Mason 8927b13b7b1SChris Mason ret = try_release_extent_state(map, tree, page, gfp_flags); 893d397712bSChris Mason if (!ret) 8946af118ceSChris Mason return 0; 8956af118ceSChris Mason 8966af118ceSChris Mason ret = try_release_extent_buffer(tree, page); 8975f39d397SChris Mason if (ret == 1) { 8985f39d397SChris Mason ClearPagePrivate(page); 8995f39d397SChris Mason set_page_private(page, 0); 900d98237b3SChris Mason page_cache_release(page); 9015f39d397SChris Mason } 9026af118ceSChris Mason 903d98237b3SChris Mason return ret; 904d98237b3SChris Mason } 905d98237b3SChris Mason 9065f39d397SChris Mason static void btree_invalidatepage(struct page *page, unsigned long offset) 907d98237b3SChris Mason { 908d1310b2eSChris Mason struct extent_io_tree *tree; 909d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9105f39d397SChris Mason extent_invalidatepage(tree, page, offset); 9115f39d397SChris Mason btree_releasepage(page, GFP_NOFS); 9129ad6b7bcSChris Mason if (PagePrivate(page)) { 913d397712bSChris Mason printk(KERN_WARNING "btrfs warning page private not zero " 914d397712bSChris Mason "on page %llu\n", (unsigned long long)page_offset(page)); 9159ad6b7bcSChris Mason ClearPagePrivate(page); 9169ad6b7bcSChris Mason set_page_private(page, 0); 9179ad6b7bcSChris Mason page_cache_release(page); 9189ad6b7bcSChris Mason } 919d98237b3SChris Mason } 920d98237b3SChris Mason 9217f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 922d98237b3SChris Mason .readpage = btree_readpage, 923d98237b3SChris Mason .writepage = btree_writepage, 9240da5468fSChris Mason .writepages = btree_writepages, 9255f39d397SChris Mason .releasepage = btree_releasepage, 9265f39d397SChris Mason .invalidatepage = btree_invalidatepage, 9275a92bc88SChris Mason #ifdef CONFIG_MIGRATION 928784b4e29SChris Mason .migratepage = btree_migratepage, 9295a92bc88SChris Mason #endif 930d98237b3SChris Mason }; 931123abc88SChris Mason 932ca7a79adSChris Mason int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize, 933ca7a79adSChris Mason u64 parent_transid) 934090d1875SChris Mason { 9355f39d397SChris Mason struct extent_buffer *buf = NULL; 9365f39d397SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 937de428b63SChris Mason int ret = 0; 938090d1875SChris Mason 939db94535dSChris Mason buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 9405f39d397SChris Mason if (!buf) 941090d1875SChris Mason return 0; 942d1310b2eSChris Mason read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, 943f188591eSChris Mason buf, 0, 0, btree_get_extent, 0); 9445f39d397SChris Mason free_extent_buffer(buf); 945de428b63SChris Mason return ret; 946090d1875SChris Mason } 947090d1875SChris Mason 9480999df54SChris Mason struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root, 9490999df54SChris Mason u64 bytenr, u32 blocksize) 9500999df54SChris Mason { 9510999df54SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 9520999df54SChris Mason struct extent_buffer *eb; 9530999df54SChris Mason eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree, 954f09d1f60SDavid Sterba bytenr, blocksize); 9550999df54SChris Mason return eb; 9560999df54SChris Mason } 9570999df54SChris Mason 9580999df54SChris Mason struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root, 9590999df54SChris Mason u64 bytenr, u32 blocksize) 9600999df54SChris Mason { 9610999df54SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 9620999df54SChris Mason struct extent_buffer *eb; 9630999df54SChris Mason 9640999df54SChris Mason eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree, 965ba144192SDavid Sterba bytenr, blocksize, NULL); 9660999df54SChris Mason return eb; 9670999df54SChris Mason } 9680999df54SChris Mason 9690999df54SChris Mason 970e02119d5SChris Mason int btrfs_write_tree_block(struct extent_buffer *buf) 971e02119d5SChris Mason { 9728aa38c31SChristoph Hellwig return filemap_fdatawrite_range(buf->first_page->mapping, buf->start, 9738aa38c31SChristoph Hellwig buf->start + buf->len - 1); 974e02119d5SChris Mason } 975e02119d5SChris Mason 976e02119d5SChris Mason int btrfs_wait_tree_block_writeback(struct extent_buffer *buf) 977e02119d5SChris Mason { 9788aa38c31SChristoph Hellwig return filemap_fdatawait_range(buf->first_page->mapping, 9798aa38c31SChristoph Hellwig buf->start, buf->start + buf->len - 1); 980e02119d5SChris Mason } 981e02119d5SChris Mason 982db94535dSChris Mason struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr, 983ca7a79adSChris Mason u32 blocksize, u64 parent_transid) 984e20d96d6SChris Mason { 9855f39d397SChris Mason struct extent_buffer *buf = NULL; 98619c00ddcSChris Mason int ret; 98719c00ddcSChris Mason 988db94535dSChris Mason buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 9895f39d397SChris Mason if (!buf) 990d98237b3SChris Mason return NULL; 991e4204dedSChris Mason 992ca7a79adSChris Mason ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 993ce9adaa5SChris Mason 994d397712bSChris Mason if (ret == 0) 995b4ce94deSChris Mason set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags); 9965f39d397SChris Mason return buf; 997ce9adaa5SChris Mason 998eb60ceacSChris Mason } 999eb60ceacSChris Mason 1000e089f05cSChris Mason int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root, 10015f39d397SChris Mason struct extent_buffer *buf) 1002ed2ff2cbSChris Mason { 10035f39d397SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 100455c69072SChris Mason if (btrfs_header_generation(buf) == 1005925baeddSChris Mason root->fs_info->running_transaction->transid) { 1006b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1007b4ce94deSChris Mason 1008b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1009b9473439SChris Mason spin_lock(&root->fs_info->delalloc_lock); 1010b9473439SChris Mason if (root->fs_info->dirty_metadata_bytes >= buf->len) 1011b9473439SChris Mason root->fs_info->dirty_metadata_bytes -= buf->len; 1012b9473439SChris Mason else 1013b9473439SChris Mason WARN_ON(1); 1014b9473439SChris Mason spin_unlock(&root->fs_info->delalloc_lock); 1015b9473439SChris Mason } 1016b4ce94deSChris Mason 1017b9473439SChris Mason /* ugh, clear_extent_buffer_dirty needs to lock the page */ 1018b9473439SChris Mason btrfs_set_lock_blocking(buf); 1019d1310b2eSChris Mason clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, 102055c69072SChris Mason buf); 1021925baeddSChris Mason } 10225f39d397SChris Mason return 0; 10235f39d397SChris Mason } 10245f39d397SChris Mason 1025db94535dSChris Mason static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize, 102687ee04ebSChris Mason u32 stripesize, struct btrfs_root *root, 10279f5fae2fSChris Mason struct btrfs_fs_info *fs_info, 1028e20d96d6SChris Mason u64 objectid) 1029d97e63b6SChris Mason { 1030cfaa7295SChris Mason root->node = NULL; 1031a28ec197SChris Mason root->commit_root = NULL; 1032db94535dSChris Mason root->sectorsize = sectorsize; 1033db94535dSChris Mason root->nodesize = nodesize; 1034db94535dSChris Mason root->leafsize = leafsize; 103587ee04ebSChris Mason root->stripesize = stripesize; 1036123abc88SChris Mason root->ref_cows = 0; 10370b86a832SChris Mason root->track_dirty = 0; 1038c71bf099SYan, Zheng root->in_radix = 0; 1039d68fc57bSYan, Zheng root->orphan_item_inserted = 0; 1040d68fc57bSYan, Zheng root->orphan_cleanup_state = 0; 10410b86a832SChris Mason 10429f5fae2fSChris Mason root->fs_info = fs_info; 10430f7d52f4SChris Mason root->objectid = objectid; 10440f7d52f4SChris Mason root->last_trans = 0; 104513a8a7c8SYan, Zheng root->highest_objectid = 0; 104658176a96SJosef Bacik root->name = NULL; 10474313b399SChris Mason root->in_sysfs = 0; 10486bef4d31SEric Paris root->inode_tree = RB_ROOT; 104916cdcec7SMiao Xie INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1050f0486c68SYan, Zheng root->block_rsv = NULL; 1051d68fc57bSYan, Zheng root->orphan_block_rsv = NULL; 10520b86a832SChris Mason 10530b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 10547b128766SJosef Bacik INIT_LIST_HEAD(&root->orphan_list); 10555d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 1056d68fc57bSYan, Zheng spin_lock_init(&root->orphan_lock); 10575d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1058f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 1059a2135011SChris Mason mutex_init(&root->objectid_mutex); 1060e02119d5SChris Mason mutex_init(&root->log_mutex); 10617237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 10627237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 10637237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 10647237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 10657237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 10667237f183SYan Zheng atomic_set(&root->log_writers, 0); 10677237f183SYan Zheng root->log_batch = 0; 10687237f183SYan Zheng root->log_transid = 0; 1069257c62e1SChris Mason root->last_log_commit = 0; 1070d0c803c4SChris Mason extent_io_tree_init(&root->dirty_log_pages, 1071f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 1072017e5369SChris Mason 10733768f368SChris Mason memset(&root->root_key, 0, sizeof(root->root_key)); 10743768f368SChris Mason memset(&root->root_item, 0, sizeof(root->root_item)); 10756702ed49SChris Mason memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 107658176a96SJosef Bacik memset(&root->root_kobj, 0, sizeof(root->root_kobj)); 10773f157a2fSChris Mason root->defrag_trans_start = fs_info->generation; 107858176a96SJosef Bacik init_completion(&root->kobj_unregister); 10796702ed49SChris Mason root->defrag_running = 0; 10804d775673SChris Mason root->root_key.objectid = objectid; 10813394e160SChris Mason root->anon_super.s_root = NULL; 10823394e160SChris Mason root->anon_super.s_dev = 0; 10833394e160SChris Mason INIT_LIST_HEAD(&root->anon_super.s_list); 10843394e160SChris Mason INIT_LIST_HEAD(&root->anon_super.s_instances); 10853394e160SChris Mason init_rwsem(&root->anon_super.s_umount); 10863394e160SChris Mason 10873768f368SChris Mason return 0; 10883768f368SChris Mason } 10893768f368SChris Mason 1090db94535dSChris Mason static int find_and_setup_root(struct btrfs_root *tree_root, 10919f5fae2fSChris Mason struct btrfs_fs_info *fs_info, 10929f5fae2fSChris Mason u64 objectid, 1093e20d96d6SChris Mason struct btrfs_root *root) 10943768f368SChris Mason { 10953768f368SChris Mason int ret; 1096db94535dSChris Mason u32 blocksize; 109784234f3aSYan Zheng u64 generation; 10983768f368SChris Mason 1099db94535dSChris Mason __setup_root(tree_root->nodesize, tree_root->leafsize, 110087ee04ebSChris Mason tree_root->sectorsize, tree_root->stripesize, 110187ee04ebSChris Mason root, fs_info, objectid); 11023768f368SChris Mason ret = btrfs_find_last_root(tree_root, objectid, 11033768f368SChris Mason &root->root_item, &root->root_key); 11044df27c4dSYan, Zheng if (ret > 0) 11054df27c4dSYan, Zheng return -ENOENT; 11063768f368SChris Mason BUG_ON(ret); 11073768f368SChris Mason 110884234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1109db94535dSChris Mason blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item)); 1110db94535dSChris Mason root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item), 111184234f3aSYan Zheng blocksize, generation); 111268433b73SChris Mason if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) { 111368433b73SChris Mason free_extent_buffer(root->node); 111468433b73SChris Mason return -EIO; 111568433b73SChris Mason } 11164df27c4dSYan, Zheng root->commit_root = btrfs_root_node(root); 1117d97e63b6SChris Mason return 0; 1118d97e63b6SChris Mason } 1119d97e63b6SChris Mason 11207237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1121e02119d5SChris Mason struct btrfs_fs_info *fs_info) 11220f7d52f4SChris Mason { 11230f7d52f4SChris Mason struct btrfs_root *root; 11240f7d52f4SChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 11257237f183SYan Zheng struct extent_buffer *leaf; 1126e02119d5SChris Mason 1127e02119d5SChris Mason root = kzalloc(sizeof(*root), GFP_NOFS); 1128e02119d5SChris Mason if (!root) 11297237f183SYan Zheng return ERR_PTR(-ENOMEM); 1130e02119d5SChris Mason 1131e02119d5SChris Mason __setup_root(tree_root->nodesize, tree_root->leafsize, 1132e02119d5SChris Mason tree_root->sectorsize, tree_root->stripesize, 1133e02119d5SChris Mason root, fs_info, BTRFS_TREE_LOG_OBJECTID); 1134e02119d5SChris Mason 1135e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1136e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1137e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 11387237f183SYan Zheng /* 11397237f183SYan Zheng * log trees do not get reference counted because they go away 11407237f183SYan Zheng * before a real commit is actually done. They do store pointers 11417237f183SYan Zheng * to file data extents, and those reference counts still get 11427237f183SYan Zheng * updated (along with back refs to the log tree). 11437237f183SYan Zheng */ 1144e02119d5SChris Mason root->ref_cows = 0; 1145e02119d5SChris Mason 11465d4f98a2SYan Zheng leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0, 11475d4f98a2SYan Zheng BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0); 11487237f183SYan Zheng if (IS_ERR(leaf)) { 11497237f183SYan Zheng kfree(root); 11507237f183SYan Zheng return ERR_CAST(leaf); 11517237f183SYan Zheng } 1152e02119d5SChris Mason 11535d4f98a2SYan Zheng memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 11545d4f98a2SYan Zheng btrfs_set_header_bytenr(leaf, leaf->start); 11555d4f98a2SYan Zheng btrfs_set_header_generation(leaf, trans->transid); 11565d4f98a2SYan Zheng btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 11575d4f98a2SYan Zheng btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID); 11587237f183SYan Zheng root->node = leaf; 1159e02119d5SChris Mason 1160e02119d5SChris Mason write_extent_buffer(root->node, root->fs_info->fsid, 1161e02119d5SChris Mason (unsigned long)btrfs_header_fsid(root->node), 1162e02119d5SChris Mason BTRFS_FSID_SIZE); 1163e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1164e02119d5SChris Mason btrfs_tree_unlock(root->node); 11657237f183SYan Zheng return root; 11667237f183SYan Zheng } 11677237f183SYan Zheng 11687237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 11697237f183SYan Zheng struct btrfs_fs_info *fs_info) 11707237f183SYan Zheng { 11717237f183SYan Zheng struct btrfs_root *log_root; 11727237f183SYan Zheng 11737237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 11747237f183SYan Zheng if (IS_ERR(log_root)) 11757237f183SYan Zheng return PTR_ERR(log_root); 11767237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 11777237f183SYan Zheng fs_info->log_root_tree = log_root; 11787237f183SYan Zheng return 0; 11797237f183SYan Zheng } 11807237f183SYan Zheng 11817237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 11827237f183SYan Zheng struct btrfs_root *root) 11837237f183SYan Zheng { 11847237f183SYan Zheng struct btrfs_root *log_root; 11857237f183SYan Zheng struct btrfs_inode_item *inode_item; 11867237f183SYan Zheng 11877237f183SYan Zheng log_root = alloc_log_tree(trans, root->fs_info); 11887237f183SYan Zheng if (IS_ERR(log_root)) 11897237f183SYan Zheng return PTR_ERR(log_root); 11907237f183SYan Zheng 11917237f183SYan Zheng log_root->last_trans = trans->transid; 11927237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 11937237f183SYan Zheng 11947237f183SYan Zheng inode_item = &log_root->root_item.inode; 11957237f183SYan Zheng inode_item->generation = cpu_to_le64(1); 11967237f183SYan Zheng inode_item->size = cpu_to_le64(3); 11977237f183SYan Zheng inode_item->nlink = cpu_to_le32(1); 11987237f183SYan Zheng inode_item->nbytes = cpu_to_le64(root->leafsize); 11997237f183SYan Zheng inode_item->mode = cpu_to_le32(S_IFDIR | 0755); 12007237f183SYan Zheng 12015d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 12027237f183SYan Zheng 12037237f183SYan Zheng WARN_ON(root->log_root); 12047237f183SYan Zheng root->log_root = log_root; 12057237f183SYan Zheng root->log_transid = 0; 1206257c62e1SChris Mason root->last_log_commit = 0; 1207e02119d5SChris Mason return 0; 1208e02119d5SChris Mason } 1209e02119d5SChris Mason 1210e02119d5SChris Mason struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root, 1211e02119d5SChris Mason struct btrfs_key *location) 1212e02119d5SChris Mason { 1213e02119d5SChris Mason struct btrfs_root *root; 1214e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 12150f7d52f4SChris Mason struct btrfs_path *path; 12165f39d397SChris Mason struct extent_buffer *l; 121784234f3aSYan Zheng u64 generation; 1218db94535dSChris Mason u32 blocksize; 12190f7d52f4SChris Mason int ret = 0; 12200f7d52f4SChris Mason 12215eda7b5eSChris Mason root = kzalloc(sizeof(*root), GFP_NOFS); 12220cf6c620SChris Mason if (!root) 12230f7d52f4SChris Mason return ERR_PTR(-ENOMEM); 12240f7d52f4SChris Mason if (location->offset == (u64)-1) { 1225db94535dSChris Mason ret = find_and_setup_root(tree_root, fs_info, 12260f7d52f4SChris Mason location->objectid, root); 12270f7d52f4SChris Mason if (ret) { 12280f7d52f4SChris Mason kfree(root); 12290f7d52f4SChris Mason return ERR_PTR(ret); 12300f7d52f4SChris Mason } 123113a8a7c8SYan, Zheng goto out; 12320f7d52f4SChris Mason } 12330f7d52f4SChris Mason 1234db94535dSChris Mason __setup_root(tree_root->nodesize, tree_root->leafsize, 123587ee04ebSChris Mason tree_root->sectorsize, tree_root->stripesize, 123687ee04ebSChris Mason root, fs_info, location->objectid); 12370f7d52f4SChris Mason 12380f7d52f4SChris Mason path = btrfs_alloc_path(); 1239db5b493aSTsutomu Itoh if (!path) { 1240db5b493aSTsutomu Itoh kfree(root); 1241db5b493aSTsutomu Itoh return ERR_PTR(-ENOMEM); 1242db5b493aSTsutomu Itoh } 12430f7d52f4SChris Mason ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0); 124413a8a7c8SYan, Zheng if (ret == 0) { 12455f39d397SChris Mason l = path->nodes[0]; 12465f39d397SChris Mason read_extent_buffer(l, &root->root_item, 12475f39d397SChris Mason btrfs_item_ptr_offset(l, path->slots[0]), 12480f7d52f4SChris Mason sizeof(root->root_item)); 124944b36eb2SYan Zheng memcpy(&root->root_key, location, sizeof(*location)); 125013a8a7c8SYan, Zheng } 12510f7d52f4SChris Mason btrfs_free_path(path); 12520f7d52f4SChris Mason if (ret) { 12535e540f77STsutomu Itoh kfree(root); 125413a8a7c8SYan, Zheng if (ret > 0) 125513a8a7c8SYan, Zheng ret = -ENOENT; 12560f7d52f4SChris Mason return ERR_PTR(ret); 12570f7d52f4SChris Mason } 125813a8a7c8SYan, Zheng 125984234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1260db94535dSChris Mason blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item)); 1261db94535dSChris Mason root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item), 126284234f3aSYan Zheng blocksize, generation); 12635d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 12640f7d52f4SChris Mason BUG_ON(!root->node); 126513a8a7c8SYan, Zheng out: 126608fe4db1SLi Zefan if (location->objectid != BTRFS_TREE_LOG_OBJECTID) { 12670f7d52f4SChris Mason root->ref_cows = 1; 126808fe4db1SLi Zefan btrfs_check_and_init_root_item(&root->root_item); 126908fe4db1SLi Zefan } 127013a8a7c8SYan, Zheng 12715eda7b5eSChris Mason return root; 12725eda7b5eSChris Mason } 12735eda7b5eSChris Mason 1274edbd8d4eSChris Mason struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info, 1275edbd8d4eSChris Mason struct btrfs_key *location) 12765eda7b5eSChris Mason { 12775eda7b5eSChris Mason struct btrfs_root *root; 12785eda7b5eSChris Mason int ret; 12795eda7b5eSChris Mason 1280edbd8d4eSChris Mason if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 1281edbd8d4eSChris Mason return fs_info->tree_root; 1282edbd8d4eSChris Mason if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID) 1283edbd8d4eSChris Mason return fs_info->extent_root; 12848f18cf13SChris Mason if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID) 12858f18cf13SChris Mason return fs_info->chunk_root; 12868f18cf13SChris Mason if (location->objectid == BTRFS_DEV_TREE_OBJECTID) 12878f18cf13SChris Mason return fs_info->dev_root; 12880403e47eSYan Zheng if (location->objectid == BTRFS_CSUM_TREE_OBJECTID) 12890403e47eSYan Zheng return fs_info->csum_root; 12904df27c4dSYan, Zheng again: 12914df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 12925eda7b5eSChris Mason root = radix_tree_lookup(&fs_info->fs_roots_radix, 12935eda7b5eSChris Mason (unsigned long)location->objectid); 12944df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 12955eda7b5eSChris Mason if (root) 12965eda7b5eSChris Mason return root; 12975eda7b5eSChris Mason 1298e02119d5SChris Mason root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location); 12995eda7b5eSChris Mason if (IS_ERR(root)) 13005eda7b5eSChris Mason return root; 13013394e160SChris Mason 1302581bb050SLi Zefan root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); 1303581bb050SLi Zefan if (!root->free_ino_ctl) 1304581bb050SLi Zefan goto fail; 1305581bb050SLi Zefan root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), 1306581bb050SLi Zefan GFP_NOFS); 1307581bb050SLi Zefan if (!root->free_ino_pinned) 1308581bb050SLi Zefan goto fail; 1309581bb050SLi Zefan 1310581bb050SLi Zefan btrfs_init_free_ino_ctl(root); 1311581bb050SLi Zefan mutex_init(&root->fs_commit_mutex); 1312581bb050SLi Zefan spin_lock_init(&root->cache_lock); 1313581bb050SLi Zefan init_waitqueue_head(&root->cache_wait); 1314581bb050SLi Zefan 1315ac08aedfSChris Mason ret = set_anon_super(&root->anon_super, NULL); 1316ac08aedfSChris Mason if (ret) 1317ac08aedfSChris Mason goto fail; 13183394e160SChris Mason 1319d68fc57bSYan, Zheng if (btrfs_root_refs(&root->root_item) == 0) { 1320d68fc57bSYan, Zheng ret = -ENOENT; 1321d68fc57bSYan, Zheng goto fail; 1322d68fc57bSYan, Zheng } 1323d68fc57bSYan, Zheng 1324d68fc57bSYan, Zheng ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid); 1325d68fc57bSYan, Zheng if (ret < 0) 1326d68fc57bSYan, Zheng goto fail; 1327d68fc57bSYan, Zheng if (ret == 0) 1328d68fc57bSYan, Zheng root->orphan_item_inserted = 1; 1329d68fc57bSYan, Zheng 13304df27c4dSYan, Zheng ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); 13314df27c4dSYan, Zheng if (ret) 13324df27c4dSYan, Zheng goto fail; 13334df27c4dSYan, Zheng 13344df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 13352619ba1fSChris Mason ret = radix_tree_insert(&fs_info->fs_roots_radix, 13362619ba1fSChris Mason (unsigned long)root->root_key.objectid, 13370f7d52f4SChris Mason root); 1338d68fc57bSYan, Zheng if (ret == 0) 13394df27c4dSYan, Zheng root->in_radix = 1; 1340d68fc57bSYan, Zheng 13414df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 13424df27c4dSYan, Zheng radix_tree_preload_end(); 13430f7d52f4SChris Mason if (ret) { 13444df27c4dSYan, Zheng if (ret == -EEXIST) { 13454df27c4dSYan, Zheng free_fs_root(root); 13464df27c4dSYan, Zheng goto again; 13470f7d52f4SChris Mason } 13484df27c4dSYan, Zheng goto fail; 13494df27c4dSYan, Zheng } 13504df27c4dSYan, Zheng 1351edbd8d4eSChris Mason ret = btrfs_find_dead_roots(fs_info->tree_root, 13525d4f98a2SYan Zheng root->root_key.objectid); 13534df27c4dSYan, Zheng WARN_ON(ret); 1354edbd8d4eSChris Mason return root; 13554df27c4dSYan, Zheng fail: 13564df27c4dSYan, Zheng free_fs_root(root); 13574df27c4dSYan, Zheng return ERR_PTR(ret); 1358edbd8d4eSChris Mason } 1359edbd8d4eSChris Mason 136004160088SChris Mason static int btrfs_congested_fn(void *congested_data, int bdi_bits) 136104160088SChris Mason { 136204160088SChris Mason struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; 136304160088SChris Mason int ret = 0; 136404160088SChris Mason struct btrfs_device *device; 136504160088SChris Mason struct backing_dev_info *bdi; 1366b7967db7SChris Mason 13671f78160cSXiao Guangrong rcu_read_lock(); 13681f78160cSXiao Guangrong list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) { 1369dfe25020SChris Mason if (!device->bdev) 1370dfe25020SChris Mason continue; 137104160088SChris Mason bdi = blk_get_backing_dev_info(device->bdev); 137204160088SChris Mason if (bdi && bdi_congested(bdi, bdi_bits)) { 137304160088SChris Mason ret = 1; 137404160088SChris Mason break; 137504160088SChris Mason } 137604160088SChris Mason } 13771f78160cSXiao Guangrong rcu_read_unlock(); 137804160088SChris Mason return ret; 137904160088SChris Mason } 138004160088SChris Mason 138138b66988SChris Mason /* 1382ad081f14SJens Axboe * If this fails, caller must call bdi_destroy() to get rid of the 1383ad081f14SJens Axboe * bdi again. 1384ad081f14SJens Axboe */ 138504160088SChris Mason static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi) 138604160088SChris Mason { 1387ad081f14SJens Axboe int err; 1388ad081f14SJens Axboe 1389ad081f14SJens Axboe bdi->capabilities = BDI_CAP_MAP_COPY; 1390e6d086d8SJens Axboe err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY); 1391ad081f14SJens Axboe if (err) 1392ad081f14SJens Axboe return err; 1393ad081f14SJens Axboe 13944575c9ccSChris Mason bdi->ra_pages = default_backing_dev_info.ra_pages; 139504160088SChris Mason bdi->congested_fn = btrfs_congested_fn; 139604160088SChris Mason bdi->congested_data = info; 139704160088SChris Mason return 0; 139804160088SChris Mason } 139904160088SChris Mason 1400ce9adaa5SChris Mason static int bio_ready_for_csum(struct bio *bio) 1401ce9adaa5SChris Mason { 1402ce9adaa5SChris Mason u64 length = 0; 1403ce9adaa5SChris Mason u64 buf_len = 0; 1404ce9adaa5SChris Mason u64 start = 0; 1405ce9adaa5SChris Mason struct page *page; 1406ce9adaa5SChris Mason struct extent_io_tree *io_tree = NULL; 1407ce9adaa5SChris Mason struct bio_vec *bvec; 1408ce9adaa5SChris Mason int i; 1409ce9adaa5SChris Mason int ret; 1410ce9adaa5SChris Mason 1411ce9adaa5SChris Mason bio_for_each_segment(bvec, bio, i) { 1412ce9adaa5SChris Mason page = bvec->bv_page; 1413ce9adaa5SChris Mason if (page->private == EXTENT_PAGE_PRIVATE) { 1414ce9adaa5SChris Mason length += bvec->bv_len; 1415ce9adaa5SChris Mason continue; 1416ce9adaa5SChris Mason } 1417ce9adaa5SChris Mason if (!page->private) { 1418ce9adaa5SChris Mason length += bvec->bv_len; 1419ce9adaa5SChris Mason continue; 1420ce9adaa5SChris Mason } 1421ce9adaa5SChris Mason length = bvec->bv_len; 1422ce9adaa5SChris Mason buf_len = page->private >> 2; 1423ce9adaa5SChris Mason start = page_offset(page) + bvec->bv_offset; 1424ce9adaa5SChris Mason io_tree = &BTRFS_I(page->mapping->host)->io_tree; 1425ce9adaa5SChris Mason } 1426ce9adaa5SChris Mason /* are we fully contained in this bio? */ 1427ce9adaa5SChris Mason if (buf_len <= length) 1428ce9adaa5SChris Mason return 1; 1429ce9adaa5SChris Mason 1430ce9adaa5SChris Mason ret = extent_range_uptodate(io_tree, start + length, 1431ce9adaa5SChris Mason start + buf_len - 1); 1432ce9adaa5SChris Mason return ret; 1433ce9adaa5SChris Mason } 1434ce9adaa5SChris Mason 14358b712842SChris Mason /* 14368b712842SChris Mason * called by the kthread helper functions to finally call the bio end_io 14378b712842SChris Mason * functions. This is where read checksum verification actually happens 14388b712842SChris Mason */ 14398b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work) 1440ce9adaa5SChris Mason { 1441ce9adaa5SChris Mason struct bio *bio; 14428b712842SChris Mason struct end_io_wq *end_io_wq; 14438b712842SChris Mason struct btrfs_fs_info *fs_info; 1444ce9adaa5SChris Mason int error; 1445ce9adaa5SChris Mason 14468b712842SChris Mason end_io_wq = container_of(work, struct end_io_wq, work); 1447ce9adaa5SChris Mason bio = end_io_wq->bio; 14488b712842SChris Mason fs_info = end_io_wq->info; 14498b712842SChris Mason 1450cad321adSChris Mason /* metadata bio reads are special because the whole tree block must 14518b712842SChris Mason * be checksummed at once. This makes sure the entire block is in 14528b712842SChris Mason * ram and up to date before trying to verify things. For 14538b712842SChris Mason * blocksize <= pagesize, it is basically a noop 14548b712842SChris Mason */ 14557b6d91daSChristoph Hellwig if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata && 1456cad321adSChris Mason !bio_ready_for_csum(bio)) { 1457d20f7043SChris Mason btrfs_queue_worker(&fs_info->endio_meta_workers, 14588b712842SChris Mason &end_io_wq->work); 1459ce9adaa5SChris Mason return; 1460ce9adaa5SChris Mason } 1461ce9adaa5SChris Mason error = end_io_wq->error; 1462ce9adaa5SChris Mason bio->bi_private = end_io_wq->private; 1463ce9adaa5SChris Mason bio->bi_end_io = end_io_wq->end_io; 1464ce9adaa5SChris Mason kfree(end_io_wq); 1465ce9adaa5SChris Mason bio_endio(bio, error); 1466ce9adaa5SChris Mason } 146744b8bd7eSChris Mason 1468a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1469a74a4b97SChris Mason { 1470a74a4b97SChris Mason struct btrfs_root *root = arg; 1471a74a4b97SChris Mason 1472a74a4b97SChris Mason do { 1473a74a4b97SChris Mason vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE); 147476dda93cSYan, Zheng 147576dda93cSYan, Zheng if (!(root->fs_info->sb->s_flags & MS_RDONLY) && 147676dda93cSYan, Zheng mutex_trylock(&root->fs_info->cleaner_mutex)) { 147724bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 1478a74a4b97SChris Mason btrfs_clean_old_snapshots(root); 1479a74a4b97SChris Mason mutex_unlock(&root->fs_info->cleaner_mutex); 14804cb5300bSChris Mason btrfs_run_defrag_inodes(root->fs_info); 148176dda93cSYan, Zheng } 1482a74a4b97SChris Mason 1483a74a4b97SChris Mason if (freezing(current)) { 1484a74a4b97SChris Mason refrigerator(); 1485a74a4b97SChris Mason } else { 1486a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 14878929ecfaSYan, Zheng if (!kthread_should_stop()) 1488a74a4b97SChris Mason schedule(); 1489a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1490a74a4b97SChris Mason } 1491a74a4b97SChris Mason } while (!kthread_should_stop()); 1492a74a4b97SChris Mason return 0; 1493a74a4b97SChris Mason } 1494a74a4b97SChris Mason 1495a74a4b97SChris Mason static int transaction_kthread(void *arg) 1496a74a4b97SChris Mason { 1497a74a4b97SChris Mason struct btrfs_root *root = arg; 1498a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1499a74a4b97SChris Mason struct btrfs_transaction *cur; 15008929ecfaSYan, Zheng u64 transid; 1501a74a4b97SChris Mason unsigned long now; 1502a74a4b97SChris Mason unsigned long delay; 1503a74a4b97SChris Mason int ret; 1504a74a4b97SChris Mason 1505a74a4b97SChris Mason do { 1506a74a4b97SChris Mason delay = HZ * 30; 1507a74a4b97SChris Mason vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE); 1508a74a4b97SChris Mason mutex_lock(&root->fs_info->transaction_kthread_mutex); 1509a74a4b97SChris Mason 1510a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 1511a74a4b97SChris Mason cur = root->fs_info->running_transaction; 1512a74a4b97SChris Mason if (!cur) { 1513a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1514a74a4b97SChris Mason goto sleep; 1515a74a4b97SChris Mason } 151631153d81SYan Zheng 1517a74a4b97SChris Mason now = get_seconds(); 15188929ecfaSYan, Zheng if (!cur->blocked && 15198929ecfaSYan, Zheng (now < cur->start_time || now - cur->start_time < 30)) { 1520a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1521a74a4b97SChris Mason delay = HZ * 5; 1522a74a4b97SChris Mason goto sleep; 1523a74a4b97SChris Mason } 15248929ecfaSYan, Zheng transid = cur->transid; 1525a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 152656bec294SChris Mason 15277a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 15283612b495STsutomu Itoh BUG_ON(IS_ERR(trans)); 15298929ecfaSYan, Zheng if (transid == trans->transid) { 15308929ecfaSYan, Zheng ret = btrfs_commit_transaction(trans, root); 15318929ecfaSYan, Zheng BUG_ON(ret); 15328929ecfaSYan, Zheng } else { 15338929ecfaSYan, Zheng btrfs_end_transaction(trans, root); 15348929ecfaSYan, Zheng } 1535a74a4b97SChris Mason sleep: 1536a74a4b97SChris Mason wake_up_process(root->fs_info->cleaner_kthread); 1537a74a4b97SChris Mason mutex_unlock(&root->fs_info->transaction_kthread_mutex); 1538a74a4b97SChris Mason 1539a74a4b97SChris Mason if (freezing(current)) { 1540a74a4b97SChris Mason refrigerator(); 1541a74a4b97SChris Mason } else { 1542a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 15438929ecfaSYan, Zheng if (!kthread_should_stop() && 15448929ecfaSYan, Zheng !btrfs_transaction_blocked(root->fs_info)) 1545a74a4b97SChris Mason schedule_timeout(delay); 1546a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1547a74a4b97SChris Mason } 1548a74a4b97SChris Mason } while (!kthread_should_stop()); 1549a74a4b97SChris Mason return 0; 1550a74a4b97SChris Mason } 1551a74a4b97SChris Mason 15528a4b83ccSChris Mason struct btrfs_root *open_ctree(struct super_block *sb, 1553dfe25020SChris Mason struct btrfs_fs_devices *fs_devices, 1554dfe25020SChris Mason char *options) 1555eb60ceacSChris Mason { 1556db94535dSChris Mason u32 sectorsize; 1557db94535dSChris Mason u32 nodesize; 1558db94535dSChris Mason u32 leafsize; 1559db94535dSChris Mason u32 blocksize; 156087ee04ebSChris Mason u32 stripesize; 156184234f3aSYan Zheng u64 generation; 1562f2b636e8SJosef Bacik u64 features; 15633de4586cSChris Mason struct btrfs_key location; 1564a061fc8dSChris Mason struct buffer_head *bh; 1565e02119d5SChris Mason struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root), 1566e20d96d6SChris Mason GFP_NOFS); 1567d20f7043SChris Mason struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root), 1568d20f7043SChris Mason GFP_NOFS); 1569450ba0eaSJosef Bacik struct btrfs_root *tree_root = btrfs_sb(sb); 15704891aca2SDavid Sterba struct btrfs_fs_info *fs_info = NULL; 1571e02119d5SChris Mason struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root), 15720b86a832SChris Mason GFP_NOFS); 1573e02119d5SChris Mason struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root), 15740b86a832SChris Mason GFP_NOFS); 1575e02119d5SChris Mason struct btrfs_root *log_tree_root; 1576e02119d5SChris Mason 1577eb60ceacSChris Mason int ret; 1578e58ca020SYan int err = -EINVAL; 15794543df7eSChris Mason 15802c90e5d6SChris Mason struct btrfs_super_block *disk_super; 15818790d502SChris Mason 15824891aca2SDavid Sterba if (!extent_root || !tree_root || !tree_root->fs_info || 1583d20f7043SChris Mason !chunk_root || !dev_root || !csum_root) { 158439279cc3SChris Mason err = -ENOMEM; 158539279cc3SChris Mason goto fail; 158639279cc3SChris Mason } 15874891aca2SDavid Sterba fs_info = tree_root->fs_info; 158876dda93cSYan, Zheng 158976dda93cSYan, Zheng ret = init_srcu_struct(&fs_info->subvol_srcu); 159076dda93cSYan, Zheng if (ret) { 159176dda93cSYan, Zheng err = ret; 159276dda93cSYan, Zheng goto fail; 159376dda93cSYan, Zheng } 159476dda93cSYan, Zheng 159576dda93cSYan, Zheng ret = setup_bdi(fs_info, &fs_info->bdi); 159676dda93cSYan, Zheng if (ret) { 159776dda93cSYan, Zheng err = ret; 159876dda93cSYan, Zheng goto fail_srcu; 159976dda93cSYan, Zheng } 160076dda93cSYan, Zheng 160176dda93cSYan, Zheng fs_info->btree_inode = new_inode(sb); 160276dda93cSYan, Zheng if (!fs_info->btree_inode) { 160376dda93cSYan, Zheng err = -ENOMEM; 160476dda93cSYan, Zheng goto fail_bdi; 160576dda93cSYan, Zheng } 160676dda93cSYan, Zheng 16071561dedaSMiao Xie fs_info->btree_inode->i_mapping->flags &= ~__GFP_FS; 16081561dedaSMiao Xie 160976dda93cSYan, Zheng INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 16108fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 1611facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 161224bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 161319c00ddcSChris Mason INIT_LIST_HEAD(&fs_info->hashers); 1614ea8c2819SChris Mason INIT_LIST_HEAD(&fs_info->delalloc_inodes); 16155a3f23d5SChris Mason INIT_LIST_HEAD(&fs_info->ordered_operations); 161611833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 16171832a6d5SChris Mason spin_lock_init(&fs_info->delalloc_lock); 1618a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 161931153d81SYan Zheng spin_lock_init(&fs_info->ref_cache_lock); 162076dda93cSYan, Zheng spin_lock_init(&fs_info->fs_roots_radix_lock); 162124bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 16224cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 16237585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 162419c00ddcSChris Mason 162558176a96SJosef Bacik init_completion(&fs_info->kobj_unregister); 16269f5fae2fSChris Mason fs_info->tree_root = tree_root; 16279f5fae2fSChris Mason fs_info->extent_root = extent_root; 1628d20f7043SChris Mason fs_info->csum_root = csum_root; 16290b86a832SChris Mason fs_info->chunk_root = chunk_root; 16300b86a832SChris Mason fs_info->dev_root = dev_root; 16318a4b83ccSChris Mason fs_info->fs_devices = fs_devices; 16320b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 16336324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 16340b86a832SChris Mason btrfs_mapping_init(&fs_info->mapping_tree); 1635f0486c68SYan, Zheng btrfs_init_block_rsv(&fs_info->global_block_rsv); 1636f0486c68SYan, Zheng btrfs_init_block_rsv(&fs_info->delalloc_block_rsv); 1637f0486c68SYan, Zheng btrfs_init_block_rsv(&fs_info->trans_block_rsv); 1638f0486c68SYan, Zheng btrfs_init_block_rsv(&fs_info->chunk_block_rsv); 1639f0486c68SYan, Zheng btrfs_init_block_rsv(&fs_info->empty_block_rsv); 1640f0486c68SYan, Zheng INIT_LIST_HEAD(&fs_info->durable_block_rsv_list); 1641f0486c68SYan, Zheng mutex_init(&fs_info->durable_block_rsv_mutex); 1642cb03c743SChris Mason atomic_set(&fs_info->nr_async_submits, 0); 1643771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 16448c8bee1dSChris Mason atomic_set(&fs_info->async_submit_draining, 0); 16450986fe9eSChris Mason atomic_set(&fs_info->nr_async_bios, 0); 16464cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 1647e20d96d6SChris Mason fs_info->sb = sb; 16486f568d35SChris Mason fs_info->max_inline = 8192 * 1024; 16499ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 16504cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 1651a4abeea4SJosef Bacik fs_info->trans_no_join = 0; 1652c8b97818SChris Mason 1653b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 1654b34b086cSChris Mason num_online_cpus() + 2, 8); 16550afbaf8cSChris Mason 16563eaa2885SChris Mason INIT_LIST_HEAD(&fs_info->ordered_extents); 16573eaa2885SChris Mason spin_lock_init(&fs_info->ordered_extent_lock); 165816cdcec7SMiao Xie fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 165916cdcec7SMiao Xie GFP_NOFS); 166016cdcec7SMiao Xie if (!fs_info->delayed_root) { 166116cdcec7SMiao Xie err = -ENOMEM; 166216cdcec7SMiao Xie goto fail_iput; 166316cdcec7SMiao Xie } 166416cdcec7SMiao Xie btrfs_init_delayed_root(fs_info->delayed_root); 16653eaa2885SChris Mason 1666a2de733cSArne Jansen mutex_init(&fs_info->scrub_lock); 1667a2de733cSArne Jansen atomic_set(&fs_info->scrubs_running, 0); 1668a2de733cSArne Jansen atomic_set(&fs_info->scrub_pause_req, 0); 1669a2de733cSArne Jansen atomic_set(&fs_info->scrubs_paused, 0); 1670a2de733cSArne Jansen atomic_set(&fs_info->scrub_cancel_req, 0); 1671a2de733cSArne Jansen init_waitqueue_head(&fs_info->scrub_pause_wait); 1672a2de733cSArne Jansen init_rwsem(&fs_info->scrub_super_lock); 1673a2de733cSArne Jansen fs_info->scrub_workers_refcnt = 0; 1674a2de733cSArne Jansen 1675a061fc8dSChris Mason sb->s_blocksize = 4096; 1676a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(4096); 167732a88aa1SJens Axboe sb->s_bdi = &fs_info->bdi; 1678a061fc8dSChris Mason 167976dda93cSYan, Zheng fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 168076dda93cSYan, Zheng fs_info->btree_inode->i_nlink = 1; 16810afbaf8cSChris Mason /* 16820afbaf8cSChris Mason * we set the i_size on the btree inode to the max possible int. 16830afbaf8cSChris Mason * the real end of the address space is determined by all of 16840afbaf8cSChris Mason * the devices in the system 16850afbaf8cSChris Mason */ 16860afbaf8cSChris Mason fs_info->btree_inode->i_size = OFFSET_MAX; 1687d98237b3SChris Mason fs_info->btree_inode->i_mapping->a_ops = &btree_aops; 168804160088SChris Mason fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi; 168904160088SChris Mason 16905d4f98a2SYan Zheng RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node); 1691d1310b2eSChris Mason extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree, 1692f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 1693a8067e02SDavid Sterba extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree); 16940da5468fSChris Mason 1695d1310b2eSChris Mason BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops; 1696d1310b2eSChris Mason 16970f7d52f4SChris Mason BTRFS_I(fs_info->btree_inode)->root = tree_root; 16980f7d52f4SChris Mason memset(&BTRFS_I(fs_info->btree_inode)->location, 0, 16990f7d52f4SChris Mason sizeof(struct btrfs_key)); 170076dda93cSYan, Zheng BTRFS_I(fs_info->btree_inode)->dummy_inode = 1; 170122b0ebdaSChris Mason insert_inode_hash(fs_info->btree_inode); 170239279cc3SChris Mason 1703e02119d5SChris Mason spin_lock_init(&fs_info->block_group_cache_lock); 17046bef4d31SEric Paris fs_info->block_group_cache_tree = RB_ROOT; 1705925baeddSChris Mason 170611833d66SYan Zheng extent_io_tree_init(&fs_info->freed_extents[0], 1707f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 170811833d66SYan Zheng extent_io_tree_init(&fs_info->freed_extents[1], 1709f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 171011833d66SYan Zheng fs_info->pinned_extents = &fs_info->freed_extents[0]; 1711e6dcd2dcSChris Mason fs_info->do_barriers = 1; 1712f9295749SChris Mason 1713d98237b3SChris Mason 17145a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 171530ae8467SChris Mason mutex_init(&fs_info->tree_log_mutex); 1716925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 1717925baeddSChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 1718a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 1719a74a4b97SChris Mason mutex_init(&fs_info->volume_mutex); 1720276e680dSYan Zheng init_rwsem(&fs_info->extent_commit_sem); 1721c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 172276dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 1723fa9c0d79SChris Mason 1724fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 1725fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 1726fa9c0d79SChris Mason 17277d9eb12cSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 17283768f368SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 1729bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 17304854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 17319a8dd150SChris Mason 17320b86a832SChris Mason __setup_root(4096, 4096, 4096, 4096, tree_root, 17332c90e5d6SChris Mason fs_info, BTRFS_ROOT_TREE_OBJECTID); 17347eccb903SChris Mason 1735a512bbf8SYan Zheng bh = btrfs_read_dev_super(fs_devices->latest_bdev); 173620b45077SDave Young if (!bh) { 173720b45077SDave Young err = -EINVAL; 173816cdcec7SMiao Xie goto fail_alloc; 173920b45077SDave Young } 174039279cc3SChris Mason 1741a061fc8dSChris Mason memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy)); 17422d69a0f8SYan Zheng memcpy(&fs_info->super_for_commit, &fs_info->super_copy, 17432d69a0f8SYan Zheng sizeof(fs_info->super_for_commit)); 1744a061fc8dSChris Mason brelse(bh); 17455f39d397SChris Mason 1746a061fc8dSChris Mason memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE); 17470b86a832SChris Mason 17485f39d397SChris Mason disk_super = &fs_info->super_copy; 17490f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 175016cdcec7SMiao Xie goto fail_alloc; 17510f7d52f4SChris Mason 1752acce952bSliubo /* check FS state, whether FS is broken. */ 1753acce952bSliubo fs_info->fs_state |= btrfs_super_flags(disk_super); 1754acce952bSliubo 1755acce952bSliubo btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY); 1756acce952bSliubo 175775e7cb7fSLiu Bo /* 175875e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 175975e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 176075e7cb7fSLiu Bo */ 176175e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 176275e7cb7fSLiu Bo 17632b82032cSYan Zheng ret = btrfs_parse_options(tree_root, options); 17642b82032cSYan Zheng if (ret) { 17652b82032cSYan Zheng err = ret; 176616cdcec7SMiao Xie goto fail_alloc; 17672b82032cSYan Zheng } 1768dfe25020SChris Mason 1769f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 1770f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 1771f2b636e8SJosef Bacik if (features) { 1772f2b636e8SJosef Bacik printk(KERN_ERR "BTRFS: couldn't mount because of " 1773f2b636e8SJosef Bacik "unsupported optional features (%Lx).\n", 177421380931SJoel Becker (unsigned long long)features); 1775f2b636e8SJosef Bacik err = -EINVAL; 177616cdcec7SMiao Xie goto fail_alloc; 1777f2b636e8SJosef Bacik } 1778f2b636e8SJosef Bacik 17795d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 17805d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 1781a6fa6faeSLi Zefan if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO) 1782a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 17835d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 17845d4f98a2SYan Zheng 1785f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 1786f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 1787f2b636e8SJosef Bacik if (!(sb->s_flags & MS_RDONLY) && features) { 1788f2b636e8SJosef Bacik printk(KERN_ERR "BTRFS: couldn't mount RDWR because of " 1789f2b636e8SJosef Bacik "unsupported option features (%Lx).\n", 179021380931SJoel Becker (unsigned long long)features); 1791f2b636e8SJosef Bacik err = -EINVAL; 179216cdcec7SMiao Xie goto fail_alloc; 1793f2b636e8SJosef Bacik } 179461d92c32SChris Mason 179561d92c32SChris Mason btrfs_init_workers(&fs_info->generic_worker, 179661d92c32SChris Mason "genwork", 1, NULL); 179761d92c32SChris Mason 17985443be45SChris Mason btrfs_init_workers(&fs_info->workers, "worker", 179961d92c32SChris Mason fs_info->thread_pool_size, 180061d92c32SChris Mason &fs_info->generic_worker); 1801c8b97818SChris Mason 1802771ed689SChris Mason btrfs_init_workers(&fs_info->delalloc_workers, "delalloc", 180361d92c32SChris Mason fs_info->thread_pool_size, 180461d92c32SChris Mason &fs_info->generic_worker); 1805771ed689SChris Mason 18065443be45SChris Mason btrfs_init_workers(&fs_info->submit_workers, "submit", 1807b720d209SChris Mason min_t(u64, fs_devices->num_devices, 180861d92c32SChris Mason fs_info->thread_pool_size), 180961d92c32SChris Mason &fs_info->generic_worker); 181061b49440SChris Mason 181161b49440SChris Mason /* a higher idle thresh on the submit workers makes it much more 181261b49440SChris Mason * likely that bios will be send down in a sane order to the 181361b49440SChris Mason * devices 181461b49440SChris Mason */ 181561b49440SChris Mason fs_info->submit_workers.idle_thresh = 64; 181653863232SChris Mason 1817771ed689SChris Mason fs_info->workers.idle_thresh = 16; 18184a69a410SChris Mason fs_info->workers.ordered = 1; 181961b49440SChris Mason 1820771ed689SChris Mason fs_info->delalloc_workers.idle_thresh = 2; 1821771ed689SChris Mason fs_info->delalloc_workers.ordered = 1; 1822771ed689SChris Mason 182361d92c32SChris Mason btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1, 182461d92c32SChris Mason &fs_info->generic_worker); 18255443be45SChris Mason btrfs_init_workers(&fs_info->endio_workers, "endio", 182661d92c32SChris Mason fs_info->thread_pool_size, 182761d92c32SChris Mason &fs_info->generic_worker); 1828d20f7043SChris Mason btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta", 182961d92c32SChris Mason fs_info->thread_pool_size, 183061d92c32SChris Mason &fs_info->generic_worker); 1831cad321adSChris Mason btrfs_init_workers(&fs_info->endio_meta_write_workers, 183261d92c32SChris Mason "endio-meta-write", fs_info->thread_pool_size, 183361d92c32SChris Mason &fs_info->generic_worker); 18345443be45SChris Mason btrfs_init_workers(&fs_info->endio_write_workers, "endio-write", 183561d92c32SChris Mason fs_info->thread_pool_size, 183661d92c32SChris Mason &fs_info->generic_worker); 18370cb59c99SJosef Bacik btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write", 18380cb59c99SJosef Bacik 1, &fs_info->generic_worker); 183916cdcec7SMiao Xie btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta", 184016cdcec7SMiao Xie fs_info->thread_pool_size, 184116cdcec7SMiao Xie &fs_info->generic_worker); 184261b49440SChris Mason 184361b49440SChris Mason /* 184461b49440SChris Mason * endios are largely parallel and should have a very 184561b49440SChris Mason * low idle thresh 184661b49440SChris Mason */ 184761b49440SChris Mason fs_info->endio_workers.idle_thresh = 4; 1848b51912c9SChris Mason fs_info->endio_meta_workers.idle_thresh = 4; 1849b51912c9SChris Mason 18509042846bSChris Mason fs_info->endio_write_workers.idle_thresh = 2; 18519042846bSChris Mason fs_info->endio_meta_write_workers.idle_thresh = 2; 18529042846bSChris Mason 18534543df7eSChris Mason btrfs_start_workers(&fs_info->workers, 1); 185461d92c32SChris Mason btrfs_start_workers(&fs_info->generic_worker, 1); 18551cc127b5SChris Mason btrfs_start_workers(&fs_info->submit_workers, 1); 1856771ed689SChris Mason btrfs_start_workers(&fs_info->delalloc_workers, 1); 1857247e743cSChris Mason btrfs_start_workers(&fs_info->fixup_workers, 1); 18589042846bSChris Mason btrfs_start_workers(&fs_info->endio_workers, 1); 18599042846bSChris Mason btrfs_start_workers(&fs_info->endio_meta_workers, 1); 18609042846bSChris Mason btrfs_start_workers(&fs_info->endio_meta_write_workers, 1); 18619042846bSChris Mason btrfs_start_workers(&fs_info->endio_write_workers, 1); 18620cb59c99SJosef Bacik btrfs_start_workers(&fs_info->endio_freespace_worker, 1); 186316cdcec7SMiao Xie btrfs_start_workers(&fs_info->delayed_workers, 1); 18644543df7eSChris Mason 18654575c9ccSChris Mason fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super); 1866c8b97818SChris Mason fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages, 1867c8b97818SChris Mason 4 * 1024 * 1024 / PAGE_CACHE_SIZE); 18684575c9ccSChris Mason 1869db94535dSChris Mason nodesize = btrfs_super_nodesize(disk_super); 1870db94535dSChris Mason leafsize = btrfs_super_leafsize(disk_super); 1871db94535dSChris Mason sectorsize = btrfs_super_sectorsize(disk_super); 187287ee04ebSChris Mason stripesize = btrfs_super_stripesize(disk_super); 1873db94535dSChris Mason tree_root->nodesize = nodesize; 1874db94535dSChris Mason tree_root->leafsize = leafsize; 1875db94535dSChris Mason tree_root->sectorsize = sectorsize; 187687ee04ebSChris Mason tree_root->stripesize = stripesize; 1877a061fc8dSChris Mason 1878a061fc8dSChris Mason sb->s_blocksize = sectorsize; 1879a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 1880db94535dSChris Mason 188139279cc3SChris Mason if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC, 188239279cc3SChris Mason sizeof(disk_super->magic))) { 1883d397712bSChris Mason printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id); 188439279cc3SChris Mason goto fail_sb_buffer; 188539279cc3SChris Mason } 188619c00ddcSChris Mason 1887925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 1888e4404d6eSYan Zheng ret = btrfs_read_sys_array(tree_root); 1889925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 189084eed90fSChris Mason if (ret) { 1891d397712bSChris Mason printk(KERN_WARNING "btrfs: failed to read the system " 1892d397712bSChris Mason "array on %s\n", sb->s_id); 18935d4f98a2SYan Zheng goto fail_sb_buffer; 189484eed90fSChris Mason } 18950b86a832SChris Mason 18960b86a832SChris Mason blocksize = btrfs_level_size(tree_root, 18970b86a832SChris Mason btrfs_super_chunk_root_level(disk_super)); 189884234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 18990b86a832SChris Mason 19000b86a832SChris Mason __setup_root(nodesize, leafsize, sectorsize, stripesize, 19010b86a832SChris Mason chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID); 19020b86a832SChris Mason 19030b86a832SChris Mason chunk_root->node = read_tree_block(chunk_root, 19040b86a832SChris Mason btrfs_super_chunk_root(disk_super), 190584234f3aSYan Zheng blocksize, generation); 19060b86a832SChris Mason BUG_ON(!chunk_root->node); 190783121942SDavid Woodhouse if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) { 190883121942SDavid Woodhouse printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n", 190983121942SDavid Woodhouse sb->s_id); 191083121942SDavid Woodhouse goto fail_chunk_root; 191183121942SDavid Woodhouse } 19125d4f98a2SYan Zheng btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 19135d4f98a2SYan Zheng chunk_root->commit_root = btrfs_root_node(chunk_root); 19140b86a832SChris Mason 1915e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 1916e17cade2SChris Mason (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node), 1917e17cade2SChris Mason BTRFS_UUID_SIZE); 1918e17cade2SChris Mason 1919925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 19200b86a832SChris Mason ret = btrfs_read_chunk_tree(chunk_root); 1921925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 19222b82032cSYan Zheng if (ret) { 1923d397712bSChris Mason printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n", 1924d397712bSChris Mason sb->s_id); 19252b82032cSYan Zheng goto fail_chunk_root; 19262b82032cSYan Zheng } 19270b86a832SChris Mason 1928dfe25020SChris Mason btrfs_close_extra_devices(fs_devices); 1929dfe25020SChris Mason 1930db94535dSChris Mason blocksize = btrfs_level_size(tree_root, 1931db94535dSChris Mason btrfs_super_root_level(disk_super)); 193284234f3aSYan Zheng generation = btrfs_super_generation(disk_super); 19330b86a832SChris Mason 1934e20d96d6SChris Mason tree_root->node = read_tree_block(tree_root, 1935db94535dSChris Mason btrfs_super_root(disk_super), 193684234f3aSYan Zheng blocksize, generation); 193739279cc3SChris Mason if (!tree_root->node) 19382b82032cSYan Zheng goto fail_chunk_root; 193983121942SDavid Woodhouse if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) { 194083121942SDavid Woodhouse printk(KERN_WARNING "btrfs: failed to read tree root on %s\n", 194183121942SDavid Woodhouse sb->s_id); 194283121942SDavid Woodhouse goto fail_tree_root; 194383121942SDavid Woodhouse } 19445d4f98a2SYan Zheng btrfs_set_root_node(&tree_root->root_item, tree_root->node); 19455d4f98a2SYan Zheng tree_root->commit_root = btrfs_root_node(tree_root); 1946db94535dSChris Mason 1947db94535dSChris Mason ret = find_and_setup_root(tree_root, fs_info, 1948e20d96d6SChris Mason BTRFS_EXTENT_TREE_OBJECTID, extent_root); 19490b86a832SChris Mason if (ret) 195039279cc3SChris Mason goto fail_tree_root; 19510b86a832SChris Mason extent_root->track_dirty = 1; 19520b86a832SChris Mason 19530b86a832SChris Mason ret = find_and_setup_root(tree_root, fs_info, 19540b86a832SChris Mason BTRFS_DEV_TREE_OBJECTID, dev_root); 19550b86a832SChris Mason if (ret) 19560b86a832SChris Mason goto fail_extent_root; 19575d4f98a2SYan Zheng dev_root->track_dirty = 1; 19583768f368SChris Mason 1959d20f7043SChris Mason ret = find_and_setup_root(tree_root, fs_info, 1960d20f7043SChris Mason BTRFS_CSUM_TREE_OBJECTID, csum_root); 1961d20f7043SChris Mason if (ret) 19625d4f98a2SYan Zheng goto fail_dev_root; 1963d20f7043SChris Mason 1964d20f7043SChris Mason csum_root->track_dirty = 1; 1965d20f7043SChris Mason 19668929ecfaSYan, Zheng fs_info->generation = generation; 19678929ecfaSYan, Zheng fs_info->last_trans_committed = generation; 19688929ecfaSYan, Zheng fs_info->data_alloc_profile = (u64)-1; 19698929ecfaSYan, Zheng fs_info->metadata_alloc_profile = (u64)-1; 19708929ecfaSYan, Zheng fs_info->system_alloc_profile = fs_info->metadata_alloc_profile; 19718929ecfaSYan, Zheng 1972c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 1973c59021f8Sliubo if (ret) { 1974c59021f8Sliubo printk(KERN_ERR "Failed to initial space info: %d\n", ret); 1975c59021f8Sliubo goto fail_block_groups; 1976c59021f8Sliubo } 1977c59021f8Sliubo 19781b1d1f66SJosef Bacik ret = btrfs_read_block_groups(extent_root); 19791b1d1f66SJosef Bacik if (ret) { 19801b1d1f66SJosef Bacik printk(KERN_ERR "Failed to read block groups: %d\n", ret); 19811b1d1f66SJosef Bacik goto fail_block_groups; 19821b1d1f66SJosef Bacik } 19839078a3e1SChris Mason 1984a74a4b97SChris Mason fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 1985a74a4b97SChris Mason "btrfs-cleaner"); 198657506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 19871b1d1f66SJosef Bacik goto fail_block_groups; 1988a74a4b97SChris Mason 1989a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 1990a74a4b97SChris Mason tree_root, 1991a74a4b97SChris Mason "btrfs-transaction"); 199257506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 19933f157a2fSChris Mason goto fail_cleaner; 1994a74a4b97SChris Mason 1995c289811cSChris Mason if (!btrfs_test_opt(tree_root, SSD) && 1996c289811cSChris Mason !btrfs_test_opt(tree_root, NOSSD) && 1997c289811cSChris Mason !fs_info->fs_devices->rotating) { 1998c289811cSChris Mason printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD " 1999c289811cSChris Mason "mode\n"); 2000c289811cSChris Mason btrfs_set_opt(fs_info->mount_opt, SSD); 2001c289811cSChris Mason } 2002c289811cSChris Mason 2003acce952bSliubo /* do not make disk changes in broken FS */ 2004acce952bSliubo if (btrfs_super_log_root(disk_super) != 0 && 2005acce952bSliubo !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) { 2006e02119d5SChris Mason u64 bytenr = btrfs_super_log_root(disk_super); 2007d18a2c44SChris Mason 20087c2ca468SChris Mason if (fs_devices->rw_devices == 0) { 2009d397712bSChris Mason printk(KERN_WARNING "Btrfs log replay required " 2010d397712bSChris Mason "on RO media\n"); 20117c2ca468SChris Mason err = -EIO; 20127c2ca468SChris Mason goto fail_trans_kthread; 20137c2ca468SChris Mason } 2014e02119d5SChris Mason blocksize = 2015e02119d5SChris Mason btrfs_level_size(tree_root, 2016e02119d5SChris Mason btrfs_super_log_root_level(disk_super)); 2017e02119d5SChris Mason 2018676e4c86SDan Carpenter log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS); 2019676e4c86SDan Carpenter if (!log_tree_root) { 2020676e4c86SDan Carpenter err = -ENOMEM; 2021676e4c86SDan Carpenter goto fail_trans_kthread; 2022676e4c86SDan Carpenter } 2023e02119d5SChris Mason 2024e02119d5SChris Mason __setup_root(nodesize, leafsize, sectorsize, stripesize, 2025e02119d5SChris Mason log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID); 2026e02119d5SChris Mason 2027e02119d5SChris Mason log_tree_root->node = read_tree_block(tree_root, bytenr, 202884234f3aSYan Zheng blocksize, 202984234f3aSYan Zheng generation + 1); 2030e02119d5SChris Mason ret = btrfs_recover_log_trees(log_tree_root); 2031e02119d5SChris Mason BUG_ON(ret); 2032e556ce2cSYan Zheng 2033e556ce2cSYan Zheng if (sb->s_flags & MS_RDONLY) { 2034e556ce2cSYan Zheng ret = btrfs_commit_super(tree_root); 2035e556ce2cSYan Zheng BUG_ON(ret); 2036e556ce2cSYan Zheng } 2037e02119d5SChris Mason } 20381a40e23bSZheng Yan 203976dda93cSYan, Zheng ret = btrfs_find_orphan_roots(tree_root); 204076dda93cSYan, Zheng BUG_ON(ret); 204176dda93cSYan, Zheng 20427c2ca468SChris Mason if (!(sb->s_flags & MS_RDONLY)) { 2043d68fc57bSYan, Zheng ret = btrfs_cleanup_fs_roots(fs_info); 2044d68fc57bSYan, Zheng BUG_ON(ret); 2045d68fc57bSYan, Zheng 20465d4f98a2SYan Zheng ret = btrfs_recover_relocation(tree_root); 2047d7ce5843SMiao Xie if (ret < 0) { 2048d7ce5843SMiao Xie printk(KERN_WARNING 2049d7ce5843SMiao Xie "btrfs: failed to recover relocation\n"); 2050d7ce5843SMiao Xie err = -EINVAL; 2051d7ce5843SMiao Xie goto fail_trans_kthread; 2052d7ce5843SMiao Xie } 20537c2ca468SChris Mason } 20541a40e23bSZheng Yan 20553de4586cSChris Mason location.objectid = BTRFS_FS_TREE_OBJECTID; 20563de4586cSChris Mason location.type = BTRFS_ROOT_ITEM_KEY; 20573de4586cSChris Mason location.offset = (u64)-1; 20583de4586cSChris Mason 20593de4586cSChris Mason fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location); 20603de4586cSChris Mason if (!fs_info->fs_root) 20617c2ca468SChris Mason goto fail_trans_kthread; 20623140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 20633140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 20643140c9a3SDan Carpenter goto fail_trans_kthread; 20653140c9a3SDan Carpenter } 2066c289811cSChris Mason 2067e3acc2a6SJosef Bacik if (!(sb->s_flags & MS_RDONLY)) { 2068e3acc2a6SJosef Bacik down_read(&fs_info->cleanup_work_sem); 206966b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_info->fs_root); 207066b4ffd1SJosef Bacik if (!err) 207166b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_info->tree_root); 2072e3acc2a6SJosef Bacik up_read(&fs_info->cleanup_work_sem); 207366b4ffd1SJosef Bacik if (err) { 207466b4ffd1SJosef Bacik close_ctree(tree_root); 207566b4ffd1SJosef Bacik return ERR_PTR(err); 207666b4ffd1SJosef Bacik } 2077e3acc2a6SJosef Bacik } 2078e3acc2a6SJosef Bacik 20790f7d52f4SChris Mason return tree_root; 208039279cc3SChris Mason 20817c2ca468SChris Mason fail_trans_kthread: 20827c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 20833f157a2fSChris Mason fail_cleaner: 2084a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 20857c2ca468SChris Mason 20867c2ca468SChris Mason /* 20877c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 20887c2ca468SChris Mason * kthreads 20897c2ca468SChris Mason */ 20907c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 20917c2ca468SChris Mason invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 20927c2ca468SChris Mason 20931b1d1f66SJosef Bacik fail_block_groups: 20941b1d1f66SJosef Bacik btrfs_free_block_groups(fs_info); 2095d20f7043SChris Mason free_extent_buffer(csum_root->node); 20965d4f98a2SYan Zheng free_extent_buffer(csum_root->commit_root); 20975d4f98a2SYan Zheng fail_dev_root: 20985d4f98a2SYan Zheng free_extent_buffer(dev_root->node); 20995d4f98a2SYan Zheng free_extent_buffer(dev_root->commit_root); 21000b86a832SChris Mason fail_extent_root: 21010b86a832SChris Mason free_extent_buffer(extent_root->node); 21025d4f98a2SYan Zheng free_extent_buffer(extent_root->commit_root); 210339279cc3SChris Mason fail_tree_root: 21045f39d397SChris Mason free_extent_buffer(tree_root->node); 21055d4f98a2SYan Zheng free_extent_buffer(tree_root->commit_root); 21062b82032cSYan Zheng fail_chunk_root: 21072b82032cSYan Zheng free_extent_buffer(chunk_root->node); 21085d4f98a2SYan Zheng free_extent_buffer(chunk_root->commit_root); 210939279cc3SChris Mason fail_sb_buffer: 211061d92c32SChris Mason btrfs_stop_workers(&fs_info->generic_worker); 2111247e743cSChris Mason btrfs_stop_workers(&fs_info->fixup_workers); 2112771ed689SChris Mason btrfs_stop_workers(&fs_info->delalloc_workers); 21138b712842SChris Mason btrfs_stop_workers(&fs_info->workers); 21148b712842SChris Mason btrfs_stop_workers(&fs_info->endio_workers); 2115d20f7043SChris Mason btrfs_stop_workers(&fs_info->endio_meta_workers); 2116cad321adSChris Mason btrfs_stop_workers(&fs_info->endio_meta_write_workers); 2117e6dcd2dcSChris Mason btrfs_stop_workers(&fs_info->endio_write_workers); 21180cb59c99SJosef Bacik btrfs_stop_workers(&fs_info->endio_freespace_worker); 21191cc127b5SChris Mason btrfs_stop_workers(&fs_info->submit_workers); 212016cdcec7SMiao Xie btrfs_stop_workers(&fs_info->delayed_workers); 212116cdcec7SMiao Xie fail_alloc: 212216cdcec7SMiao Xie kfree(fs_info->delayed_root); 21234543df7eSChris Mason fail_iput: 21247c2ca468SChris Mason invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 21254543df7eSChris Mason iput(fs_info->btree_inode); 21267e662854SQinghuang Feng 2127dfe25020SChris Mason btrfs_close_devices(fs_info->fs_devices); 212884eed90fSChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 2129ad081f14SJens Axboe fail_bdi: 21307e662854SQinghuang Feng bdi_destroy(&fs_info->bdi); 213176dda93cSYan, Zheng fail_srcu: 213276dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 21337e662854SQinghuang Feng fail: 213439279cc3SChris Mason kfree(extent_root); 213539279cc3SChris Mason kfree(tree_root); 213639279cc3SChris Mason kfree(fs_info); 213783afeac4SJim Meyering kfree(chunk_root); 213883afeac4SJim Meyering kfree(dev_root); 2139d20f7043SChris Mason kfree(csum_root); 214039279cc3SChris Mason return ERR_PTR(err); 2141eb60ceacSChris Mason } 2142eb60ceacSChris Mason 2143f2984462SChris Mason static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate) 2144f2984462SChris Mason { 2145f2984462SChris Mason char b[BDEVNAME_SIZE]; 2146f2984462SChris Mason 2147f2984462SChris Mason if (uptodate) { 2148f2984462SChris Mason set_buffer_uptodate(bh); 2149f2984462SChris Mason } else { 21507a36ddecSDavid Sterba printk_ratelimited(KERN_WARNING "lost page write due to " 2151f2984462SChris Mason "I/O error on %s\n", 2152f2984462SChris Mason bdevname(bh->b_bdev, b)); 21531259ab75SChris Mason /* note, we dont' set_buffer_write_io_error because we have 21541259ab75SChris Mason * our own ways of dealing with the IO errors 21551259ab75SChris Mason */ 2156f2984462SChris Mason clear_buffer_uptodate(bh); 2157f2984462SChris Mason } 2158f2984462SChris Mason unlock_buffer(bh); 2159f2984462SChris Mason put_bh(bh); 2160f2984462SChris Mason } 2161f2984462SChris Mason 2162a512bbf8SYan Zheng struct buffer_head *btrfs_read_dev_super(struct block_device *bdev) 2163a512bbf8SYan Zheng { 2164a512bbf8SYan Zheng struct buffer_head *bh; 2165a512bbf8SYan Zheng struct buffer_head *latest = NULL; 2166a512bbf8SYan Zheng struct btrfs_super_block *super; 2167a512bbf8SYan Zheng int i; 2168a512bbf8SYan Zheng u64 transid = 0; 2169a512bbf8SYan Zheng u64 bytenr; 2170a512bbf8SYan Zheng 2171a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 2172a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 2173a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 2174a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 2175a512bbf8SYan Zheng */ 2176a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 2177a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 2178a512bbf8SYan Zheng if (bytenr + 4096 >= i_size_read(bdev->bd_inode)) 2179a512bbf8SYan Zheng break; 2180a512bbf8SYan Zheng bh = __bread(bdev, bytenr / 4096, 4096); 2181a512bbf8SYan Zheng if (!bh) 2182a512bbf8SYan Zheng continue; 2183a512bbf8SYan Zheng 2184a512bbf8SYan Zheng super = (struct btrfs_super_block *)bh->b_data; 2185a512bbf8SYan Zheng if (btrfs_super_bytenr(super) != bytenr || 2186a512bbf8SYan Zheng strncmp((char *)(&super->magic), BTRFS_MAGIC, 2187a512bbf8SYan Zheng sizeof(super->magic))) { 2188a512bbf8SYan Zheng brelse(bh); 2189a512bbf8SYan Zheng continue; 2190a512bbf8SYan Zheng } 2191a512bbf8SYan Zheng 2192a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 2193a512bbf8SYan Zheng brelse(latest); 2194a512bbf8SYan Zheng latest = bh; 2195a512bbf8SYan Zheng transid = btrfs_super_generation(super); 2196a512bbf8SYan Zheng } else { 2197a512bbf8SYan Zheng brelse(bh); 2198a512bbf8SYan Zheng } 2199a512bbf8SYan Zheng } 2200a512bbf8SYan Zheng return latest; 2201a512bbf8SYan Zheng } 2202a512bbf8SYan Zheng 22034eedeb75SHisashi Hifumi /* 22044eedeb75SHisashi Hifumi * this should be called twice, once with wait == 0 and 22054eedeb75SHisashi Hifumi * once with wait == 1. When wait == 0 is done, all the buffer heads 22064eedeb75SHisashi Hifumi * we write are pinned. 22074eedeb75SHisashi Hifumi * 22084eedeb75SHisashi Hifumi * They are released when wait == 1 is done. 22094eedeb75SHisashi Hifumi * max_mirrors must be the same for both runs, and it indicates how 22104eedeb75SHisashi Hifumi * many supers on this one device should be written. 22114eedeb75SHisashi Hifumi * 22124eedeb75SHisashi Hifumi * max_mirrors == 0 means to write them all. 22134eedeb75SHisashi Hifumi */ 2214a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 2215a512bbf8SYan Zheng struct btrfs_super_block *sb, 2216a512bbf8SYan Zheng int do_barriers, int wait, int max_mirrors) 2217a512bbf8SYan Zheng { 2218a512bbf8SYan Zheng struct buffer_head *bh; 2219a512bbf8SYan Zheng int i; 2220a512bbf8SYan Zheng int ret; 2221a512bbf8SYan Zheng int errors = 0; 2222a512bbf8SYan Zheng u32 crc; 2223a512bbf8SYan Zheng u64 bytenr; 2224a512bbf8SYan Zheng int last_barrier = 0; 2225a512bbf8SYan Zheng 2226a512bbf8SYan Zheng if (max_mirrors == 0) 2227a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 2228a512bbf8SYan Zheng 2229a512bbf8SYan Zheng /* make sure only the last submit_bh does a barrier */ 2230a512bbf8SYan Zheng if (do_barriers) { 2231a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 2232a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 2233a512bbf8SYan Zheng if (bytenr + BTRFS_SUPER_INFO_SIZE >= 2234a512bbf8SYan Zheng device->total_bytes) 2235a512bbf8SYan Zheng break; 2236a512bbf8SYan Zheng last_barrier = i; 2237a512bbf8SYan Zheng } 2238a512bbf8SYan Zheng } 2239a512bbf8SYan Zheng 2240a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 2241a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 2242a512bbf8SYan Zheng if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes) 2243a512bbf8SYan Zheng break; 2244a512bbf8SYan Zheng 2245a512bbf8SYan Zheng if (wait) { 2246a512bbf8SYan Zheng bh = __find_get_block(device->bdev, bytenr / 4096, 2247a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 2248a512bbf8SYan Zheng BUG_ON(!bh); 2249a512bbf8SYan Zheng wait_on_buffer(bh); 22504eedeb75SHisashi Hifumi if (!buffer_uptodate(bh)) 22514eedeb75SHisashi Hifumi errors++; 22524eedeb75SHisashi Hifumi 22534eedeb75SHisashi Hifumi /* drop our reference */ 22544eedeb75SHisashi Hifumi brelse(bh); 22554eedeb75SHisashi Hifumi 22564eedeb75SHisashi Hifumi /* drop the reference from the wait == 0 run */ 2257a512bbf8SYan Zheng brelse(bh); 2258a512bbf8SYan Zheng continue; 2259a512bbf8SYan Zheng } else { 2260a512bbf8SYan Zheng btrfs_set_super_bytenr(sb, bytenr); 2261a512bbf8SYan Zheng 2262a512bbf8SYan Zheng crc = ~(u32)0; 2263a512bbf8SYan Zheng crc = btrfs_csum_data(NULL, (char *)sb + 2264a512bbf8SYan Zheng BTRFS_CSUM_SIZE, crc, 2265a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE - 2266a512bbf8SYan Zheng BTRFS_CSUM_SIZE); 2267a512bbf8SYan Zheng btrfs_csum_final(crc, sb->csum); 2268a512bbf8SYan Zheng 22694eedeb75SHisashi Hifumi /* 22704eedeb75SHisashi Hifumi * one reference for us, and we leave it for the 22714eedeb75SHisashi Hifumi * caller 22724eedeb75SHisashi Hifumi */ 2273a512bbf8SYan Zheng bh = __getblk(device->bdev, bytenr / 4096, 2274a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 2275a512bbf8SYan Zheng memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE); 2276a512bbf8SYan Zheng 22774eedeb75SHisashi Hifumi /* one reference for submit_bh */ 2278a512bbf8SYan Zheng get_bh(bh); 22794eedeb75SHisashi Hifumi 22804eedeb75SHisashi Hifumi set_buffer_uptodate(bh); 2281a512bbf8SYan Zheng lock_buffer(bh); 2282a512bbf8SYan Zheng bh->b_end_io = btrfs_end_buffer_write_sync; 2283a512bbf8SYan Zheng } 2284a512bbf8SYan Zheng 2285c3b9a62cSChristoph Hellwig if (i == last_barrier && do_barriers) 2286c3b9a62cSChristoph Hellwig ret = submit_bh(WRITE_FLUSH_FUA, bh); 2287c3b9a62cSChristoph Hellwig else 2288ffbd517dSChris Mason ret = submit_bh(WRITE_SYNC, bh); 2289a512bbf8SYan Zheng 22904eedeb75SHisashi Hifumi if (ret) 2291a512bbf8SYan Zheng errors++; 2292a512bbf8SYan Zheng } 2293a512bbf8SYan Zheng return errors < i ? 0 : -1; 2294a512bbf8SYan Zheng } 2295a512bbf8SYan Zheng 2296a512bbf8SYan Zheng int write_all_supers(struct btrfs_root *root, int max_mirrors) 2297f2984462SChris Mason { 2298e5e9a520SChris Mason struct list_head *head; 2299f2984462SChris Mason struct btrfs_device *dev; 2300a061fc8dSChris Mason struct btrfs_super_block *sb; 2301f2984462SChris Mason struct btrfs_dev_item *dev_item; 2302f2984462SChris Mason int ret; 2303f2984462SChris Mason int do_barriers; 2304a236aed1SChris Mason int max_errors; 2305a236aed1SChris Mason int total_errors = 0; 2306a061fc8dSChris Mason u64 flags; 2307f2984462SChris Mason 2308a236aed1SChris Mason max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1; 2309f2984462SChris Mason do_barriers = !btrfs_test_opt(root, NOBARRIER); 2310f2984462SChris Mason 2311a061fc8dSChris Mason sb = &root->fs_info->super_for_commit; 2312a061fc8dSChris Mason dev_item = &sb->dev_item; 2313e5e9a520SChris Mason 2314174ba509SChris Mason mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 2315e5e9a520SChris Mason head = &root->fs_info->fs_devices->devices; 23161f78160cSXiao Guangrong list_for_each_entry_rcu(dev, head, dev_list) { 2317dfe25020SChris Mason if (!dev->bdev) { 2318dfe25020SChris Mason total_errors++; 2319dfe25020SChris Mason continue; 2320dfe25020SChris Mason } 23212b82032cSYan Zheng if (!dev->in_fs_metadata || !dev->writeable) 2322dfe25020SChris Mason continue; 2323dfe25020SChris Mason 23242b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 2325a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 2326a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 2327a061fc8dSChris Mason btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes); 2328a061fc8dSChris Mason btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used); 2329a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 2330a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 2331a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 2332a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 23332b82032cSYan Zheng memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE); 2334a512bbf8SYan Zheng 2335a061fc8dSChris Mason flags = btrfs_super_flags(sb); 2336a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 2337f2984462SChris Mason 2338a512bbf8SYan Zheng ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors); 2339a236aed1SChris Mason if (ret) 2340a236aed1SChris Mason total_errors++; 2341f2984462SChris Mason } 2342a236aed1SChris Mason if (total_errors > max_errors) { 2343d397712bSChris Mason printk(KERN_ERR "btrfs: %d errors while writing supers\n", 2344d397712bSChris Mason total_errors); 2345a236aed1SChris Mason BUG(); 2346a236aed1SChris Mason } 2347f2984462SChris Mason 2348a512bbf8SYan Zheng total_errors = 0; 23491f78160cSXiao Guangrong list_for_each_entry_rcu(dev, head, dev_list) { 2350dfe25020SChris Mason if (!dev->bdev) 2351dfe25020SChris Mason continue; 23522b82032cSYan Zheng if (!dev->in_fs_metadata || !dev->writeable) 2353dfe25020SChris Mason continue; 2354dfe25020SChris Mason 2355a512bbf8SYan Zheng ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors); 2356a512bbf8SYan Zheng if (ret) 23571259ab75SChris Mason total_errors++; 2358f2984462SChris Mason } 2359174ba509SChris Mason mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 2360a236aed1SChris Mason if (total_errors > max_errors) { 2361d397712bSChris Mason printk(KERN_ERR "btrfs: %d errors while writing supers\n", 2362d397712bSChris Mason total_errors); 2363a236aed1SChris Mason BUG(); 2364a236aed1SChris Mason } 2365f2984462SChris Mason return 0; 2366f2984462SChris Mason } 2367f2984462SChris Mason 2368a512bbf8SYan Zheng int write_ctree_super(struct btrfs_trans_handle *trans, 2369a512bbf8SYan Zheng struct btrfs_root *root, int max_mirrors) 2370cfaa7295SChris Mason { 2371e66f709bSChris Mason int ret; 23722c90e5d6SChris Mason 2373a512bbf8SYan Zheng ret = write_all_supers(root, max_mirrors); 23745f39d397SChris Mason return ret; 2375cfaa7295SChris Mason } 2376cfaa7295SChris Mason 23775eda7b5eSChris Mason int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root) 23782619ba1fSChris Mason { 23794df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 23802619ba1fSChris Mason radix_tree_delete(&fs_info->fs_roots_radix, 23812619ba1fSChris Mason (unsigned long)root->root_key.objectid); 23824df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 238376dda93cSYan, Zheng 238476dda93cSYan, Zheng if (btrfs_root_refs(&root->root_item) == 0) 238576dda93cSYan, Zheng synchronize_srcu(&fs_info->subvol_srcu); 238676dda93cSYan, Zheng 2387581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_pinned); 2388581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_ctl); 23894df27c4dSYan, Zheng free_fs_root(root); 23904df27c4dSYan, Zheng return 0; 23914df27c4dSYan, Zheng } 23924df27c4dSYan, Zheng 23934df27c4dSYan, Zheng static void free_fs_root(struct btrfs_root *root) 23944df27c4dSYan, Zheng { 239582d5902dSLi Zefan iput(root->cache_inode); 23964df27c4dSYan, Zheng WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 23973394e160SChris Mason if (root->anon_super.s_dev) { 23983394e160SChris Mason down_write(&root->anon_super.s_umount); 23993394e160SChris Mason kill_anon_super(&root->anon_super); 24003394e160SChris Mason } 24015f39d397SChris Mason free_extent_buffer(root->node); 24025f39d397SChris Mason free_extent_buffer(root->commit_root); 2403581bb050SLi Zefan kfree(root->free_ino_ctl); 2404581bb050SLi Zefan kfree(root->free_ino_pinned); 240558176a96SJosef Bacik kfree(root->name); 24062619ba1fSChris Mason kfree(root); 24072619ba1fSChris Mason } 24082619ba1fSChris Mason 240935b7e476SChris Mason static int del_fs_roots(struct btrfs_fs_info *fs_info) 24100f7d52f4SChris Mason { 24110f7d52f4SChris Mason int ret; 24120f7d52f4SChris Mason struct btrfs_root *gang[8]; 24130f7d52f4SChris Mason int i; 24140f7d52f4SChris Mason 241576dda93cSYan, Zheng while (!list_empty(&fs_info->dead_roots)) { 241676dda93cSYan, Zheng gang[0] = list_entry(fs_info->dead_roots.next, 241776dda93cSYan, Zheng struct btrfs_root, root_list); 241876dda93cSYan, Zheng list_del(&gang[0]->root_list); 241976dda93cSYan, Zheng 242076dda93cSYan, Zheng if (gang[0]->in_radix) { 242176dda93cSYan, Zheng btrfs_free_fs_root(fs_info, gang[0]); 242276dda93cSYan, Zheng } else { 242376dda93cSYan, Zheng free_extent_buffer(gang[0]->node); 242476dda93cSYan, Zheng free_extent_buffer(gang[0]->commit_root); 242576dda93cSYan, Zheng kfree(gang[0]); 242676dda93cSYan, Zheng } 242776dda93cSYan, Zheng } 242876dda93cSYan, Zheng 24290f7d52f4SChris Mason while (1) { 24300f7d52f4SChris Mason ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 24310f7d52f4SChris Mason (void **)gang, 0, 24320f7d52f4SChris Mason ARRAY_SIZE(gang)); 24330f7d52f4SChris Mason if (!ret) 24340f7d52f4SChris Mason break; 24352619ba1fSChris Mason for (i = 0; i < ret; i++) 24365eda7b5eSChris Mason btrfs_free_fs_root(fs_info, gang[i]); 24370f7d52f4SChris Mason } 24380f7d52f4SChris Mason return 0; 24390f7d52f4SChris Mason } 2440b4100d64SChris Mason 2441c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 2442c146afadSYan Zheng { 2443c146afadSYan Zheng u64 root_objectid = 0; 2444c146afadSYan Zheng struct btrfs_root *gang[8]; 2445c146afadSYan Zheng int i; 2446c146afadSYan Zheng int ret; 2447c146afadSYan Zheng 2448c146afadSYan Zheng while (1) { 2449c146afadSYan Zheng ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2450c146afadSYan Zheng (void **)gang, root_objectid, 2451c146afadSYan Zheng ARRAY_SIZE(gang)); 2452c146afadSYan Zheng if (!ret) 2453c146afadSYan Zheng break; 24545d4f98a2SYan Zheng 24555d4f98a2SYan Zheng root_objectid = gang[ret - 1]->root_key.objectid + 1; 2456c146afadSYan Zheng for (i = 0; i < ret; i++) { 245766b4ffd1SJosef Bacik int err; 245866b4ffd1SJosef Bacik 2459c146afadSYan Zheng root_objectid = gang[i]->root_key.objectid; 246066b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(gang[i]); 246166b4ffd1SJosef Bacik if (err) 246266b4ffd1SJosef Bacik return err; 2463c146afadSYan Zheng } 2464c146afadSYan Zheng root_objectid++; 2465c146afadSYan Zheng } 2466c146afadSYan Zheng return 0; 2467c146afadSYan Zheng } 2468c146afadSYan Zheng 2469c146afadSYan Zheng int btrfs_commit_super(struct btrfs_root *root) 2470c146afadSYan Zheng { 2471c146afadSYan Zheng struct btrfs_trans_handle *trans; 2472c146afadSYan Zheng int ret; 2473c146afadSYan Zheng 2474c146afadSYan Zheng mutex_lock(&root->fs_info->cleaner_mutex); 247524bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 2476c146afadSYan Zheng btrfs_clean_old_snapshots(root); 2477c146afadSYan Zheng mutex_unlock(&root->fs_info->cleaner_mutex); 2478c71bf099SYan, Zheng 2479c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 2480c71bf099SYan, Zheng down_write(&root->fs_info->cleanup_work_sem); 2481c71bf099SYan, Zheng up_write(&root->fs_info->cleanup_work_sem); 2482c71bf099SYan, Zheng 24837a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 24843612b495STsutomu Itoh if (IS_ERR(trans)) 24853612b495STsutomu Itoh return PTR_ERR(trans); 2486c146afadSYan Zheng ret = btrfs_commit_transaction(trans, root); 2487c146afadSYan Zheng BUG_ON(ret); 2488c146afadSYan Zheng /* run commit again to drop the original snapshot */ 24897a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 24903612b495STsutomu Itoh if (IS_ERR(trans)) 24913612b495STsutomu Itoh return PTR_ERR(trans); 2492c146afadSYan Zheng btrfs_commit_transaction(trans, root); 2493c146afadSYan Zheng ret = btrfs_write_and_wait_transaction(NULL, root); 2494c146afadSYan Zheng BUG_ON(ret); 2495c146afadSYan Zheng 2496a512bbf8SYan Zheng ret = write_ctree_super(NULL, root, 0); 2497c146afadSYan Zheng return ret; 2498c146afadSYan Zheng } 2499c146afadSYan Zheng 2500e20d96d6SChris Mason int close_ctree(struct btrfs_root *root) 2501eb60ceacSChris Mason { 25020f7d52f4SChris Mason struct btrfs_fs_info *fs_info = root->fs_info; 2503c146afadSYan Zheng int ret; 2504e089f05cSChris Mason 2505facda1e7SChris Mason fs_info->closing = 1; 2506a2135011SChris Mason smp_mb(); 2507a2135011SChris Mason 2508a2de733cSArne Jansen btrfs_scrub_cancel(root); 25094cb5300bSChris Mason 25104cb5300bSChris Mason /* wait for any defraggers to finish */ 25114cb5300bSChris Mason wait_event(fs_info->transaction_wait, 25124cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 25134cb5300bSChris Mason 25144cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 25154cb5300bSChris Mason btrfs_run_defrag_inodes(root->fs_info); 25164cb5300bSChris Mason 25170af3d00bSJosef Bacik btrfs_put_block_group_cache(fs_info); 2518acce952bSliubo 2519acce952bSliubo /* 2520acce952bSliubo * Here come 2 situations when btrfs is broken to flip readonly: 2521acce952bSliubo * 2522acce952bSliubo * 1. when btrfs flips readonly somewhere else before 2523acce952bSliubo * btrfs_commit_super, sb->s_flags has MS_RDONLY flag, 2524acce952bSliubo * and btrfs will skip to write sb directly to keep 2525acce952bSliubo * ERROR state on disk. 2526acce952bSliubo * 2527acce952bSliubo * 2. when btrfs flips readonly just in btrfs_commit_super, 2528ae0e47f0SJustin P. Mattock * and in such case, btrfs cannot write sb via btrfs_commit_super, 2529acce952bSliubo * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag, 2530acce952bSliubo * btrfs will cleanup all FS resources first and write sb then. 2531acce952bSliubo */ 2532c146afadSYan Zheng if (!(fs_info->sb->s_flags & MS_RDONLY)) { 2533c146afadSYan Zheng ret = btrfs_commit_super(root); 2534d397712bSChris Mason if (ret) 2535d397712bSChris Mason printk(KERN_ERR "btrfs: commit super ret %d\n", ret); 2536c146afadSYan Zheng } 2537ed2ff2cbSChris Mason 2538acce952bSliubo if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { 2539acce952bSliubo ret = btrfs_error_commit_super(root); 2540acce952bSliubo if (ret) 2541acce952bSliubo printk(KERN_ERR "btrfs: commit super ret %d\n", ret); 2542acce952bSliubo } 2543acce952bSliubo 25448929ecfaSYan, Zheng kthread_stop(root->fs_info->transaction_kthread); 25458929ecfaSYan, Zheng kthread_stop(root->fs_info->cleaner_kthread); 25468929ecfaSYan, Zheng 2547f25784b3SYan Zheng fs_info->closing = 2; 2548f25784b3SYan Zheng smp_mb(); 2549f25784b3SYan Zheng 2550b0c68f8bSChris Mason if (fs_info->delalloc_bytes) { 2551d397712bSChris Mason printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n", 255221380931SJoel Becker (unsigned long long)fs_info->delalloc_bytes); 2553b0c68f8bSChris Mason } 255431153d81SYan Zheng if (fs_info->total_ref_cache_size) { 2555d397712bSChris Mason printk(KERN_INFO "btrfs: at umount reference cache size %llu\n", 2556d397712bSChris Mason (unsigned long long)fs_info->total_ref_cache_size); 255731153d81SYan Zheng } 255831153d81SYan Zheng 25595f39d397SChris Mason free_extent_buffer(fs_info->extent_root->node); 25605d4f98a2SYan Zheng free_extent_buffer(fs_info->extent_root->commit_root); 25615f39d397SChris Mason free_extent_buffer(fs_info->tree_root->node); 25625d4f98a2SYan Zheng free_extent_buffer(fs_info->tree_root->commit_root); 25630b86a832SChris Mason free_extent_buffer(root->fs_info->chunk_root->node); 25645d4f98a2SYan Zheng free_extent_buffer(root->fs_info->chunk_root->commit_root); 25650b86a832SChris Mason free_extent_buffer(root->fs_info->dev_root->node); 25665d4f98a2SYan Zheng free_extent_buffer(root->fs_info->dev_root->commit_root); 2567d20f7043SChris Mason free_extent_buffer(root->fs_info->csum_root->node); 25685d4f98a2SYan Zheng free_extent_buffer(root->fs_info->csum_root->commit_root); 2569d20f7043SChris Mason 25709078a3e1SChris Mason btrfs_free_block_groups(root->fs_info); 2571c146afadSYan Zheng 25720f7d52f4SChris Mason del_fs_roots(fs_info); 2573d10c5f31SChris Mason 2574c146afadSYan Zheng iput(fs_info->btree_inode); 257516cdcec7SMiao Xie kfree(fs_info->delayed_root); 25769ad6b7bcSChris Mason 257761d92c32SChris Mason btrfs_stop_workers(&fs_info->generic_worker); 2578247e743cSChris Mason btrfs_stop_workers(&fs_info->fixup_workers); 2579771ed689SChris Mason btrfs_stop_workers(&fs_info->delalloc_workers); 25808b712842SChris Mason btrfs_stop_workers(&fs_info->workers); 25818b712842SChris Mason btrfs_stop_workers(&fs_info->endio_workers); 2582d20f7043SChris Mason btrfs_stop_workers(&fs_info->endio_meta_workers); 2583cad321adSChris Mason btrfs_stop_workers(&fs_info->endio_meta_write_workers); 2584e6dcd2dcSChris Mason btrfs_stop_workers(&fs_info->endio_write_workers); 25850cb59c99SJosef Bacik btrfs_stop_workers(&fs_info->endio_freespace_worker); 25861cc127b5SChris Mason btrfs_stop_workers(&fs_info->submit_workers); 258716cdcec7SMiao Xie btrfs_stop_workers(&fs_info->delayed_workers); 2588d6bfde87SChris Mason 2589dfe25020SChris Mason btrfs_close_devices(fs_info->fs_devices); 25900b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 2591b248a415SChris Mason 259204160088SChris Mason bdi_destroy(&fs_info->bdi); 259376dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 25940b86a832SChris Mason 25950f7d52f4SChris Mason kfree(fs_info->extent_root); 25960f7d52f4SChris Mason kfree(fs_info->tree_root); 25970b86a832SChris Mason kfree(fs_info->chunk_root); 25980b86a832SChris Mason kfree(fs_info->dev_root); 2599d20f7043SChris Mason kfree(fs_info->csum_root); 260083a4d548SLi Zefan kfree(fs_info); 260183a4d548SLi Zefan 2602eb60ceacSChris Mason return 0; 2603eb60ceacSChris Mason } 2604eb60ceacSChris Mason 26051259ab75SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid) 2606ccd467d6SChris Mason { 26071259ab75SChris Mason int ret; 2608810191ffSChris Mason struct inode *btree_inode = buf->first_page->mapping->host; 26091259ab75SChris Mason 26102ac55d41SJosef Bacik ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf, 26112ac55d41SJosef Bacik NULL); 26121259ab75SChris Mason if (!ret) 26131259ab75SChris Mason return ret; 26141259ab75SChris Mason 26151259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 26161259ab75SChris Mason parent_transid); 26171259ab75SChris Mason return !ret; 26185f39d397SChris Mason } 26196702ed49SChris Mason 26205f39d397SChris Mason int btrfs_set_buffer_uptodate(struct extent_buffer *buf) 26215f39d397SChris Mason { 2622810191ffSChris Mason struct inode *btree_inode = buf->first_page->mapping->host; 2623d1310b2eSChris Mason return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, 26245f39d397SChris Mason buf); 26255f39d397SChris Mason } 26265f39d397SChris Mason 26275f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 26285f39d397SChris Mason { 2629810191ffSChris Mason struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root; 26305f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 26315f39d397SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 2632b9473439SChris Mason int was_dirty; 2633b4ce94deSChris Mason 2634b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 2635ccd467d6SChris Mason if (transid != root->fs_info->generation) { 2636d397712bSChris Mason printk(KERN_CRIT "btrfs transid mismatch buffer %llu, " 2637d397712bSChris Mason "found %llu running %llu\n", 2638db94535dSChris Mason (unsigned long long)buf->start, 2639d397712bSChris Mason (unsigned long long)transid, 2640d397712bSChris Mason (unsigned long long)root->fs_info->generation); 2641ccd467d6SChris Mason WARN_ON(1); 2642ccd467d6SChris Mason } 2643b9473439SChris Mason was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, 2644b9473439SChris Mason buf); 2645b9473439SChris Mason if (!was_dirty) { 2646b9473439SChris Mason spin_lock(&root->fs_info->delalloc_lock); 2647b9473439SChris Mason root->fs_info->dirty_metadata_bytes += buf->len; 2648b9473439SChris Mason spin_unlock(&root->fs_info->delalloc_lock); 2649b9473439SChris Mason } 2650eb60ceacSChris Mason } 2651eb60ceacSChris Mason 2652d3c2fdcfSChris Mason void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr) 265335b7e476SChris Mason { 2654188de649SChris Mason /* 2655188de649SChris Mason * looks as though older kernels can get into trouble with 2656188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 2657188de649SChris Mason */ 2658d6bfde87SChris Mason u64 num_dirty; 2659771ed689SChris Mason unsigned long thresh = 32 * 1024 * 1024; 2660d6bfde87SChris Mason 26616933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 2662d6bfde87SChris Mason return; 2663d6bfde87SChris Mason 266416cdcec7SMiao Xie btrfs_balance_delayed_items(root); 266516cdcec7SMiao Xie 266616cdcec7SMiao Xie num_dirty = root->fs_info->dirty_metadata_bytes; 266716cdcec7SMiao Xie 266816cdcec7SMiao Xie if (num_dirty > thresh) { 266916cdcec7SMiao Xie balance_dirty_pages_ratelimited_nr( 267016cdcec7SMiao Xie root->fs_info->btree_inode->i_mapping, 1); 267116cdcec7SMiao Xie } 267216cdcec7SMiao Xie return; 267316cdcec7SMiao Xie } 267416cdcec7SMiao Xie 267516cdcec7SMiao Xie void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr) 267616cdcec7SMiao Xie { 267716cdcec7SMiao Xie /* 267816cdcec7SMiao Xie * looks as though older kernels can get into trouble with 267916cdcec7SMiao Xie * this code, they end up stuck in balance_dirty_pages forever 268016cdcec7SMiao Xie */ 268116cdcec7SMiao Xie u64 num_dirty; 268216cdcec7SMiao Xie unsigned long thresh = 32 * 1024 * 1024; 268316cdcec7SMiao Xie 268416cdcec7SMiao Xie if (current->flags & PF_MEMALLOC) 268516cdcec7SMiao Xie return; 268616cdcec7SMiao Xie 2687585ad2c3SChris Mason num_dirty = root->fs_info->dirty_metadata_bytes; 2688585ad2c3SChris Mason 2689d6bfde87SChris Mason if (num_dirty > thresh) { 2690d3c2fdcfSChris Mason balance_dirty_pages_ratelimited_nr( 2691304fced6SChris Mason root->fs_info->btree_inode->i_mapping, 1); 269235b7e476SChris Mason } 2693188de649SChris Mason return; 2694d6bfde87SChris Mason } 26956b80053dSChris Mason 2696ca7a79adSChris Mason int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid) 26976b80053dSChris Mason { 2698810191ffSChris Mason struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root; 2699ce9adaa5SChris Mason int ret; 2700ca7a79adSChris Mason ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 2701d397712bSChris Mason if (ret == 0) 2702b4ce94deSChris Mason set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags); 2703ce9adaa5SChris Mason return ret; 27046b80053dSChris Mason } 27050da5468fSChris Mason 27064bef0848SChris Mason int btree_lock_page_hook(struct page *page) 27074bef0848SChris Mason { 27084bef0848SChris Mason struct inode *inode = page->mapping->host; 2709b9473439SChris Mason struct btrfs_root *root = BTRFS_I(inode)->root; 27104bef0848SChris Mason struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 27114bef0848SChris Mason struct extent_buffer *eb; 27124bef0848SChris Mason unsigned long len; 27134bef0848SChris Mason u64 bytenr = page_offset(page); 27144bef0848SChris Mason 27154bef0848SChris Mason if (page->private == EXTENT_PAGE_PRIVATE) 27164bef0848SChris Mason goto out; 27174bef0848SChris Mason 27184bef0848SChris Mason len = page->private >> 2; 2719f09d1f60SDavid Sterba eb = find_extent_buffer(io_tree, bytenr, len); 27204bef0848SChris Mason if (!eb) 27214bef0848SChris Mason goto out; 27224bef0848SChris Mason 27234bef0848SChris Mason btrfs_tree_lock(eb); 27244bef0848SChris Mason btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); 2725b9473439SChris Mason 2726b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { 2727b9473439SChris Mason spin_lock(&root->fs_info->delalloc_lock); 2728b9473439SChris Mason if (root->fs_info->dirty_metadata_bytes >= eb->len) 2729b9473439SChris Mason root->fs_info->dirty_metadata_bytes -= eb->len; 2730b9473439SChris Mason else 2731b9473439SChris Mason WARN_ON(1); 2732b9473439SChris Mason spin_unlock(&root->fs_info->delalloc_lock); 2733b9473439SChris Mason } 2734b9473439SChris Mason 27354bef0848SChris Mason btrfs_tree_unlock(eb); 27364bef0848SChris Mason free_extent_buffer(eb); 27374bef0848SChris Mason out: 27384bef0848SChris Mason lock_page(page); 27394bef0848SChris Mason return 0; 27404bef0848SChris Mason } 27414bef0848SChris Mason 2742acce952bSliubo static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info, 2743acce952bSliubo int read_only) 2744acce952bSliubo { 2745acce952bSliubo if (read_only) 2746acce952bSliubo return; 2747acce952bSliubo 2748acce952bSliubo if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) 2749acce952bSliubo printk(KERN_WARNING "warning: mount fs with errors, " 2750acce952bSliubo "running btrfsck is recommended\n"); 2751acce952bSliubo } 2752acce952bSliubo 2753acce952bSliubo int btrfs_error_commit_super(struct btrfs_root *root) 2754acce952bSliubo { 2755acce952bSliubo int ret; 2756acce952bSliubo 2757acce952bSliubo mutex_lock(&root->fs_info->cleaner_mutex); 2758acce952bSliubo btrfs_run_delayed_iputs(root); 2759acce952bSliubo mutex_unlock(&root->fs_info->cleaner_mutex); 2760acce952bSliubo 2761acce952bSliubo down_write(&root->fs_info->cleanup_work_sem); 2762acce952bSliubo up_write(&root->fs_info->cleanup_work_sem); 2763acce952bSliubo 2764acce952bSliubo /* cleanup FS via transaction */ 2765acce952bSliubo btrfs_cleanup_transaction(root); 2766acce952bSliubo 2767acce952bSliubo ret = write_ctree_super(NULL, root, 0); 2768acce952bSliubo 2769acce952bSliubo return ret; 2770acce952bSliubo } 2771acce952bSliubo 2772acce952bSliubo static int btrfs_destroy_ordered_operations(struct btrfs_root *root) 2773acce952bSliubo { 2774acce952bSliubo struct btrfs_inode *btrfs_inode; 2775acce952bSliubo struct list_head splice; 2776acce952bSliubo 2777acce952bSliubo INIT_LIST_HEAD(&splice); 2778acce952bSliubo 2779acce952bSliubo mutex_lock(&root->fs_info->ordered_operations_mutex); 2780acce952bSliubo spin_lock(&root->fs_info->ordered_extent_lock); 2781acce952bSliubo 2782acce952bSliubo list_splice_init(&root->fs_info->ordered_operations, &splice); 2783acce952bSliubo while (!list_empty(&splice)) { 2784acce952bSliubo btrfs_inode = list_entry(splice.next, struct btrfs_inode, 2785acce952bSliubo ordered_operations); 2786acce952bSliubo 2787acce952bSliubo list_del_init(&btrfs_inode->ordered_operations); 2788acce952bSliubo 2789acce952bSliubo btrfs_invalidate_inodes(btrfs_inode->root); 2790acce952bSliubo } 2791acce952bSliubo 2792acce952bSliubo spin_unlock(&root->fs_info->ordered_extent_lock); 2793acce952bSliubo mutex_unlock(&root->fs_info->ordered_operations_mutex); 2794acce952bSliubo 2795acce952bSliubo return 0; 2796acce952bSliubo } 2797acce952bSliubo 2798acce952bSliubo static int btrfs_destroy_ordered_extents(struct btrfs_root *root) 2799acce952bSliubo { 2800acce952bSliubo struct list_head splice; 2801acce952bSliubo struct btrfs_ordered_extent *ordered; 2802acce952bSliubo struct inode *inode; 2803acce952bSliubo 2804acce952bSliubo INIT_LIST_HEAD(&splice); 2805acce952bSliubo 2806acce952bSliubo spin_lock(&root->fs_info->ordered_extent_lock); 2807acce952bSliubo 2808acce952bSliubo list_splice_init(&root->fs_info->ordered_extents, &splice); 2809acce952bSliubo while (!list_empty(&splice)) { 2810acce952bSliubo ordered = list_entry(splice.next, struct btrfs_ordered_extent, 2811acce952bSliubo root_extent_list); 2812acce952bSliubo 2813acce952bSliubo list_del_init(&ordered->root_extent_list); 2814acce952bSliubo atomic_inc(&ordered->refs); 2815acce952bSliubo 2816acce952bSliubo /* the inode may be getting freed (in sys_unlink path). */ 2817acce952bSliubo inode = igrab(ordered->inode); 2818acce952bSliubo 2819acce952bSliubo spin_unlock(&root->fs_info->ordered_extent_lock); 2820acce952bSliubo if (inode) 2821acce952bSliubo iput(inode); 2822acce952bSliubo 2823acce952bSliubo atomic_set(&ordered->refs, 1); 2824acce952bSliubo btrfs_put_ordered_extent(ordered); 2825acce952bSliubo 2826acce952bSliubo spin_lock(&root->fs_info->ordered_extent_lock); 2827acce952bSliubo } 2828acce952bSliubo 2829acce952bSliubo spin_unlock(&root->fs_info->ordered_extent_lock); 2830acce952bSliubo 2831acce952bSliubo return 0; 2832acce952bSliubo } 2833acce952bSliubo 2834acce952bSliubo static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 2835acce952bSliubo struct btrfs_root *root) 2836acce952bSliubo { 2837acce952bSliubo struct rb_node *node; 2838acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 2839acce952bSliubo struct btrfs_delayed_ref_node *ref; 2840acce952bSliubo int ret = 0; 2841acce952bSliubo 2842acce952bSliubo delayed_refs = &trans->delayed_refs; 2843acce952bSliubo 2844acce952bSliubo spin_lock(&delayed_refs->lock); 2845acce952bSliubo if (delayed_refs->num_entries == 0) { 2846cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 2847acce952bSliubo printk(KERN_INFO "delayed_refs has NO entry\n"); 2848acce952bSliubo return ret; 2849acce952bSliubo } 2850acce952bSliubo 2851acce952bSliubo node = rb_first(&delayed_refs->root); 2852acce952bSliubo while (node) { 2853acce952bSliubo ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); 2854acce952bSliubo node = rb_next(node); 2855acce952bSliubo 2856acce952bSliubo ref->in_tree = 0; 2857acce952bSliubo rb_erase(&ref->rb_node, &delayed_refs->root); 2858acce952bSliubo delayed_refs->num_entries--; 2859acce952bSliubo 2860acce952bSliubo atomic_set(&ref->refs, 1); 2861acce952bSliubo if (btrfs_delayed_ref_is_head(ref)) { 2862acce952bSliubo struct btrfs_delayed_ref_head *head; 2863acce952bSliubo 2864acce952bSliubo head = btrfs_delayed_node_to_head(ref); 2865acce952bSliubo mutex_lock(&head->mutex); 2866acce952bSliubo kfree(head->extent_op); 2867acce952bSliubo delayed_refs->num_heads--; 2868acce952bSliubo if (list_empty(&head->cluster)) 2869acce952bSliubo delayed_refs->num_heads_ready--; 2870acce952bSliubo list_del_init(&head->cluster); 2871acce952bSliubo mutex_unlock(&head->mutex); 2872acce952bSliubo } 2873acce952bSliubo 2874acce952bSliubo spin_unlock(&delayed_refs->lock); 2875acce952bSliubo btrfs_put_delayed_ref(ref); 2876acce952bSliubo 2877acce952bSliubo cond_resched(); 2878acce952bSliubo spin_lock(&delayed_refs->lock); 2879acce952bSliubo } 2880acce952bSliubo 2881acce952bSliubo spin_unlock(&delayed_refs->lock); 2882acce952bSliubo 2883acce952bSliubo return ret; 2884acce952bSliubo } 2885acce952bSliubo 2886acce952bSliubo static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t) 2887acce952bSliubo { 2888acce952bSliubo struct btrfs_pending_snapshot *snapshot; 2889acce952bSliubo struct list_head splice; 2890acce952bSliubo 2891acce952bSliubo INIT_LIST_HEAD(&splice); 2892acce952bSliubo 2893acce952bSliubo list_splice_init(&t->pending_snapshots, &splice); 2894acce952bSliubo 2895acce952bSliubo while (!list_empty(&splice)) { 2896acce952bSliubo snapshot = list_entry(splice.next, 2897acce952bSliubo struct btrfs_pending_snapshot, 2898acce952bSliubo list); 2899acce952bSliubo 2900acce952bSliubo list_del_init(&snapshot->list); 2901acce952bSliubo 2902acce952bSliubo kfree(snapshot); 2903acce952bSliubo } 2904acce952bSliubo 2905acce952bSliubo return 0; 2906acce952bSliubo } 2907acce952bSliubo 2908acce952bSliubo static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 2909acce952bSliubo { 2910acce952bSliubo struct btrfs_inode *btrfs_inode; 2911acce952bSliubo struct list_head splice; 2912acce952bSliubo 2913acce952bSliubo INIT_LIST_HEAD(&splice); 2914acce952bSliubo 2915acce952bSliubo spin_lock(&root->fs_info->delalloc_lock); 29165be76758SDavid Sterba list_splice_init(&root->fs_info->delalloc_inodes, &splice); 2917acce952bSliubo 2918acce952bSliubo while (!list_empty(&splice)) { 2919acce952bSliubo btrfs_inode = list_entry(splice.next, struct btrfs_inode, 2920acce952bSliubo delalloc_inodes); 2921acce952bSliubo 2922acce952bSliubo list_del_init(&btrfs_inode->delalloc_inodes); 2923acce952bSliubo 2924acce952bSliubo btrfs_invalidate_inodes(btrfs_inode->root); 2925acce952bSliubo } 2926acce952bSliubo 2927acce952bSliubo spin_unlock(&root->fs_info->delalloc_lock); 2928acce952bSliubo 2929acce952bSliubo return 0; 2930acce952bSliubo } 2931acce952bSliubo 2932acce952bSliubo static int btrfs_destroy_marked_extents(struct btrfs_root *root, 2933acce952bSliubo struct extent_io_tree *dirty_pages, 2934acce952bSliubo int mark) 2935acce952bSliubo { 2936acce952bSliubo int ret; 2937acce952bSliubo struct page *page; 2938acce952bSliubo struct inode *btree_inode = root->fs_info->btree_inode; 2939acce952bSliubo struct extent_buffer *eb; 2940acce952bSliubo u64 start = 0; 2941acce952bSliubo u64 end; 2942acce952bSliubo u64 offset; 2943acce952bSliubo unsigned long index; 2944acce952bSliubo 2945acce952bSliubo while (1) { 2946acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 2947acce952bSliubo mark); 2948acce952bSliubo if (ret) 2949acce952bSliubo break; 2950acce952bSliubo 2951acce952bSliubo clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); 2952acce952bSliubo while (start <= end) { 2953acce952bSliubo index = start >> PAGE_CACHE_SHIFT; 2954acce952bSliubo start = (u64)(index + 1) << PAGE_CACHE_SHIFT; 2955acce952bSliubo page = find_get_page(btree_inode->i_mapping, index); 2956acce952bSliubo if (!page) 2957acce952bSliubo continue; 2958acce952bSliubo offset = page_offset(page); 2959acce952bSliubo 2960acce952bSliubo spin_lock(&dirty_pages->buffer_lock); 2961acce952bSliubo eb = radix_tree_lookup( 2962acce952bSliubo &(&BTRFS_I(page->mapping->host)->io_tree)->buffer, 2963acce952bSliubo offset >> PAGE_CACHE_SHIFT); 2964acce952bSliubo spin_unlock(&dirty_pages->buffer_lock); 2965acce952bSliubo if (eb) { 2966acce952bSliubo ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY, 2967acce952bSliubo &eb->bflags); 2968acce952bSliubo atomic_set(&eb->refs, 1); 2969acce952bSliubo } 2970acce952bSliubo if (PageWriteback(page)) 2971acce952bSliubo end_page_writeback(page); 2972acce952bSliubo 2973acce952bSliubo lock_page(page); 2974acce952bSliubo if (PageDirty(page)) { 2975acce952bSliubo clear_page_dirty_for_io(page); 2976acce952bSliubo spin_lock_irq(&page->mapping->tree_lock); 2977acce952bSliubo radix_tree_tag_clear(&page->mapping->page_tree, 2978acce952bSliubo page_index(page), 2979acce952bSliubo PAGECACHE_TAG_DIRTY); 2980acce952bSliubo spin_unlock_irq(&page->mapping->tree_lock); 2981acce952bSliubo } 2982acce952bSliubo 2983acce952bSliubo page->mapping->a_ops->invalidatepage(page, 0); 2984acce952bSliubo unlock_page(page); 2985acce952bSliubo } 2986acce952bSliubo } 2987acce952bSliubo 2988acce952bSliubo return ret; 2989acce952bSliubo } 2990acce952bSliubo 2991acce952bSliubo static int btrfs_destroy_pinned_extent(struct btrfs_root *root, 2992acce952bSliubo struct extent_io_tree *pinned_extents) 2993acce952bSliubo { 2994acce952bSliubo struct extent_io_tree *unpin; 2995acce952bSliubo u64 start; 2996acce952bSliubo u64 end; 2997acce952bSliubo int ret; 2998acce952bSliubo 2999acce952bSliubo unpin = pinned_extents; 3000acce952bSliubo while (1) { 3001acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 3002acce952bSliubo EXTENT_DIRTY); 3003acce952bSliubo if (ret) 3004acce952bSliubo break; 3005acce952bSliubo 3006acce952bSliubo /* opt_discard */ 30075378e607SLi Dongyang if (btrfs_test_opt(root, DISCARD)) 30085378e607SLi Dongyang ret = btrfs_error_discard_extent(root, start, 30095378e607SLi Dongyang end + 1 - start, 30105378e607SLi Dongyang NULL); 3011acce952bSliubo 3012acce952bSliubo clear_extent_dirty(unpin, start, end, GFP_NOFS); 3013acce952bSliubo btrfs_error_unpin_extent_range(root, start, end); 3014acce952bSliubo cond_resched(); 3015acce952bSliubo } 3016acce952bSliubo 3017acce952bSliubo return 0; 3018acce952bSliubo } 3019acce952bSliubo 3020acce952bSliubo static int btrfs_cleanup_transaction(struct btrfs_root *root) 3021acce952bSliubo { 3022acce952bSliubo struct btrfs_transaction *t; 3023acce952bSliubo LIST_HEAD(list); 3024acce952bSliubo 3025acce952bSliubo WARN_ON(1); 3026acce952bSliubo 3027acce952bSliubo mutex_lock(&root->fs_info->transaction_kthread_mutex); 3028acce952bSliubo 3029a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 3030acce952bSliubo list_splice_init(&root->fs_info->trans_list, &list); 3031a4abeea4SJosef Bacik root->fs_info->trans_no_join = 1; 3032a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 3033a4abeea4SJosef Bacik 3034acce952bSliubo while (!list_empty(&list)) { 3035acce952bSliubo t = list_entry(list.next, struct btrfs_transaction, list); 3036acce952bSliubo if (!t) 3037acce952bSliubo break; 3038acce952bSliubo 3039acce952bSliubo btrfs_destroy_ordered_operations(root); 3040acce952bSliubo 3041acce952bSliubo btrfs_destroy_ordered_extents(root); 3042acce952bSliubo 3043acce952bSliubo btrfs_destroy_delayed_refs(t, root); 3044acce952bSliubo 3045acce952bSliubo btrfs_block_rsv_release(root, 3046acce952bSliubo &root->fs_info->trans_block_rsv, 3047acce952bSliubo t->dirty_pages.dirty_bytes); 3048acce952bSliubo 3049acce952bSliubo /* FIXME: cleanup wait for commit */ 3050acce952bSliubo t->in_commit = 1; 3051acce952bSliubo t->blocked = 1; 3052acce952bSliubo if (waitqueue_active(&root->fs_info->transaction_blocked_wait)) 3053acce952bSliubo wake_up(&root->fs_info->transaction_blocked_wait); 3054acce952bSliubo 3055acce952bSliubo t->blocked = 0; 3056acce952bSliubo if (waitqueue_active(&root->fs_info->transaction_wait)) 3057acce952bSliubo wake_up(&root->fs_info->transaction_wait); 3058acce952bSliubo 3059acce952bSliubo t->commit_done = 1; 3060acce952bSliubo if (waitqueue_active(&t->commit_wait)) 3061acce952bSliubo wake_up(&t->commit_wait); 3062acce952bSliubo 3063acce952bSliubo btrfs_destroy_pending_snapshots(t); 3064acce952bSliubo 3065acce952bSliubo btrfs_destroy_delalloc_inodes(root); 3066acce952bSliubo 3067a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 3068acce952bSliubo root->fs_info->running_transaction = NULL; 3069a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 3070acce952bSliubo 3071acce952bSliubo btrfs_destroy_marked_extents(root, &t->dirty_pages, 3072acce952bSliubo EXTENT_DIRTY); 3073acce952bSliubo 3074acce952bSliubo btrfs_destroy_pinned_extent(root, 3075acce952bSliubo root->fs_info->pinned_extents); 3076acce952bSliubo 307713c5a93eSJosef Bacik atomic_set(&t->use_count, 0); 3078acce952bSliubo list_del_init(&t->list); 3079acce952bSliubo memset(t, 0, sizeof(*t)); 3080acce952bSliubo kmem_cache_free(btrfs_transaction_cachep, t); 3081acce952bSliubo } 3082acce952bSliubo 3083a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 3084a4abeea4SJosef Bacik root->fs_info->trans_no_join = 0; 3085a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 3086acce952bSliubo mutex_unlock(&root->fs_info->transaction_kthread_mutex); 3087acce952bSliubo 3088acce952bSliubo return 0; 3089acce952bSliubo } 3090acce952bSliubo 3091d1310b2eSChris Mason static struct extent_io_ops btree_extent_io_ops = { 30924bef0848SChris Mason .write_cache_pages_lock_hook = btree_lock_page_hook, 3093ce9adaa5SChris Mason .readpage_end_io_hook = btree_readpage_end_io_hook, 30940b86a832SChris Mason .submit_bio_hook = btree_submit_bio_hook, 3095239b14b3SChris Mason /* note we're sharing with inode.c for the merge bio hook */ 3096239b14b3SChris Mason .merge_bio_hook = btrfs_merge_bio_hook, 30970da5468fSChris Mason }; 3098