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> 327e75bf3fSDavid Sterba #include <asm/unaligned.h> 334b4e25f2SChris Mason #include "compat.h" 34eb60ceacSChris Mason #include "ctree.h" 35eb60ceacSChris Mason #include "disk-io.h" 36e089f05cSChris Mason #include "transaction.h" 370f7d52f4SChris Mason #include "btrfs_inode.h" 380b86a832SChris Mason #include "volumes.h" 39db94535dSChris Mason #include "print-tree.h" 408b712842SChris Mason #include "async-thread.h" 41925baeddSChris Mason #include "locking.h" 42e02119d5SChris Mason #include "tree-log.h" 43fa9c0d79SChris Mason #include "free-space-cache.h" 44eb60ceacSChris Mason 45d1310b2eSChris Mason static struct extent_io_ops btree_extent_io_ops; 468b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work); 474df27c4dSYan, Zheng static void free_fs_root(struct btrfs_root *root); 48acce952bSliubo static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info, 49acce952bSliubo int read_only); 50acce952bSliubo static int btrfs_destroy_ordered_operations(struct btrfs_root *root); 51acce952bSliubo static int btrfs_destroy_ordered_extents(struct btrfs_root *root); 52acce952bSliubo static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 53acce952bSliubo struct btrfs_root *root); 54acce952bSliubo static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t); 55acce952bSliubo static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root); 56acce952bSliubo static int btrfs_destroy_marked_extents(struct btrfs_root *root, 57acce952bSliubo struct extent_io_tree *dirty_pages, 58acce952bSliubo int mark); 59acce952bSliubo static int btrfs_destroy_pinned_extent(struct btrfs_root *root, 60acce952bSliubo struct extent_io_tree *pinned_extents); 61acce952bSliubo static int btrfs_cleanup_transaction(struct btrfs_root *root); 62ce9adaa5SChris Mason 63d352ac68SChris Mason /* 64d352ac68SChris Mason * end_io_wq structs are used to do processing in task context when an IO is 65d352ac68SChris Mason * complete. This is used during reads to verify checksums, and it is used 66d352ac68SChris Mason * by writes to insert metadata for new file extents after IO is complete. 67d352ac68SChris Mason */ 68ce9adaa5SChris Mason struct end_io_wq { 69ce9adaa5SChris Mason struct bio *bio; 70ce9adaa5SChris Mason bio_end_io_t *end_io; 71ce9adaa5SChris Mason void *private; 72ce9adaa5SChris Mason struct btrfs_fs_info *info; 73ce9adaa5SChris Mason int error; 7422c59948SChris Mason int metadata; 75ce9adaa5SChris Mason struct list_head list; 768b712842SChris Mason struct btrfs_work work; 77ce9adaa5SChris Mason }; 780da5468fSChris Mason 79d352ac68SChris Mason /* 80d352ac68SChris Mason * async submit bios are used to offload expensive checksumming 81d352ac68SChris Mason * onto the worker threads. They checksum file and metadata bios 82d352ac68SChris Mason * just before they are sent down the IO stack. 83d352ac68SChris Mason */ 8444b8bd7eSChris Mason struct async_submit_bio { 8544b8bd7eSChris Mason struct inode *inode; 8644b8bd7eSChris Mason struct bio *bio; 8744b8bd7eSChris Mason struct list_head list; 884a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_start; 894a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_done; 9044b8bd7eSChris Mason int rw; 9144b8bd7eSChris Mason int mirror_num; 92c8b97818SChris Mason unsigned long bio_flags; 93eaf25d93SChris Mason /* 94eaf25d93SChris Mason * bio_offset is optional, can be used if the pages in the bio 95eaf25d93SChris Mason * can't tell us where in the file the bio should go 96eaf25d93SChris Mason */ 97eaf25d93SChris Mason u64 bio_offset; 988b712842SChris Mason struct btrfs_work work; 9944b8bd7eSChris Mason }; 10044b8bd7eSChris Mason 1014008c04aSChris Mason /* These are used to set the lockdep class on the extent buffer locks. 1024008c04aSChris Mason * The class is set by the readpage_end_io_hook after the buffer has 1034008c04aSChris Mason * passed csum validation but before the pages are unlocked. 1044008c04aSChris Mason * 1054008c04aSChris Mason * The lockdep class is also set by btrfs_init_new_buffer on freshly 1064008c04aSChris Mason * allocated blocks. 1074008c04aSChris Mason * 1084008c04aSChris Mason * The class is based on the level in the tree block, which allows lockdep 1094008c04aSChris Mason * to know that lower nodes nest inside the locks of higher nodes. 1104008c04aSChris Mason * 1114008c04aSChris Mason * We also add a check to make sure the highest level of the tree is 1124008c04aSChris Mason * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this 1134008c04aSChris Mason * code needs update as well. 1144008c04aSChris Mason */ 1154008c04aSChris Mason #ifdef CONFIG_DEBUG_LOCK_ALLOC 1164008c04aSChris Mason # if BTRFS_MAX_LEVEL != 8 1174008c04aSChris Mason # error 1184008c04aSChris Mason # endif 1194008c04aSChris Mason static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1]; 1204008c04aSChris Mason static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = { 1214008c04aSChris Mason /* leaf */ 1224008c04aSChris Mason "btrfs-extent-00", 1234008c04aSChris Mason "btrfs-extent-01", 1244008c04aSChris Mason "btrfs-extent-02", 1254008c04aSChris Mason "btrfs-extent-03", 1264008c04aSChris Mason "btrfs-extent-04", 1274008c04aSChris Mason "btrfs-extent-05", 1284008c04aSChris Mason "btrfs-extent-06", 1294008c04aSChris Mason "btrfs-extent-07", 1304008c04aSChris Mason /* highest possible level */ 1314008c04aSChris Mason "btrfs-extent-08", 1324008c04aSChris Mason }; 1334008c04aSChris Mason #endif 1344008c04aSChris Mason 135d352ac68SChris Mason /* 136d352ac68SChris Mason * extents on the btree inode are pretty simple, there's one extent 137d352ac68SChris Mason * that covers the entire device 138d352ac68SChris Mason */ 139b2950863SChristoph Hellwig static struct extent_map *btree_get_extent(struct inode *inode, 140b2950863SChristoph Hellwig struct page *page, size_t page_offset, u64 start, u64 len, 1415f39d397SChris Mason int create) 1425f39d397SChris Mason { 1435f39d397SChris Mason struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 1445f39d397SChris Mason struct extent_map *em; 1455f39d397SChris Mason int ret; 1465f39d397SChris Mason 147890871beSChris Mason read_lock(&em_tree->lock); 148d1310b2eSChris Mason em = lookup_extent_mapping(em_tree, start, len); 149a061fc8dSChris Mason if (em) { 150a061fc8dSChris Mason em->bdev = 151a061fc8dSChris Mason BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 152890871beSChris Mason read_unlock(&em_tree->lock); 1535f39d397SChris Mason goto out; 154a061fc8dSChris Mason } 155890871beSChris Mason read_unlock(&em_tree->lock); 1567b13b7b1SChris Mason 1575f39d397SChris Mason em = alloc_extent_map(GFP_NOFS); 1585f39d397SChris Mason if (!em) { 1595f39d397SChris Mason em = ERR_PTR(-ENOMEM); 1605f39d397SChris Mason goto out; 1615f39d397SChris Mason } 1625f39d397SChris Mason em->start = 0; 1630afbaf8cSChris Mason em->len = (u64)-1; 164c8b97818SChris Mason em->block_len = (u64)-1; 1655f39d397SChris Mason em->block_start = 0; 166a061fc8dSChris Mason em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 167d1310b2eSChris Mason 168890871beSChris Mason write_lock(&em_tree->lock); 1695f39d397SChris Mason ret = add_extent_mapping(em_tree, em); 1705f39d397SChris Mason if (ret == -EEXIST) { 1710afbaf8cSChris Mason u64 failed_start = em->start; 1720afbaf8cSChris Mason u64 failed_len = em->len; 1730afbaf8cSChris Mason 1745f39d397SChris Mason free_extent_map(em); 1757b13b7b1SChris Mason em = lookup_extent_mapping(em_tree, start, len); 1760afbaf8cSChris Mason if (em) { 1777b13b7b1SChris Mason ret = 0; 1780afbaf8cSChris Mason } else { 1790afbaf8cSChris Mason em = lookup_extent_mapping(em_tree, failed_start, 1800afbaf8cSChris Mason failed_len); 1817b13b7b1SChris Mason ret = -EIO; 1820afbaf8cSChris Mason } 1835f39d397SChris Mason } else if (ret) { 1847b13b7b1SChris Mason free_extent_map(em); 1857b13b7b1SChris Mason em = NULL; 1865f39d397SChris Mason } 187890871beSChris Mason write_unlock(&em_tree->lock); 1887b13b7b1SChris Mason 1897b13b7b1SChris Mason if (ret) 1907b13b7b1SChris Mason em = ERR_PTR(ret); 1915f39d397SChris Mason out: 1925f39d397SChris Mason return em; 1935f39d397SChris Mason } 1945f39d397SChris Mason 19519c00ddcSChris Mason u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len) 19619c00ddcSChris Mason { 197163e783eSDavid Woodhouse return crc32c(seed, data, len); 19819c00ddcSChris Mason } 19919c00ddcSChris Mason 20019c00ddcSChris Mason void btrfs_csum_final(u32 crc, char *result) 20119c00ddcSChris Mason { 2027e75bf3fSDavid Sterba put_unaligned_le32(~crc, result); 20319c00ddcSChris Mason } 20419c00ddcSChris Mason 205d352ac68SChris Mason /* 206d352ac68SChris Mason * compute the csum for a btree block, and either verify it or write it 207d352ac68SChris Mason * into the csum field of the block. 208d352ac68SChris Mason */ 20919c00ddcSChris Mason static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf, 21019c00ddcSChris Mason int verify) 21119c00ddcSChris Mason { 212607d432dSJosef Bacik u16 csum_size = 213607d432dSJosef Bacik btrfs_super_csum_size(&root->fs_info->super_copy); 214607d432dSJosef Bacik char *result = NULL; 21519c00ddcSChris Mason unsigned long len; 21619c00ddcSChris Mason unsigned long cur_len; 21719c00ddcSChris Mason unsigned long offset = BTRFS_CSUM_SIZE; 21819c00ddcSChris Mason char *map_token = NULL; 21919c00ddcSChris Mason char *kaddr; 22019c00ddcSChris Mason unsigned long map_start; 22119c00ddcSChris Mason unsigned long map_len; 22219c00ddcSChris Mason int err; 22319c00ddcSChris Mason u32 crc = ~(u32)0; 224607d432dSJosef Bacik unsigned long inline_result; 22519c00ddcSChris Mason 22619c00ddcSChris Mason len = buf->len - offset; 22719c00ddcSChris Mason while (len > 0) { 22819c00ddcSChris Mason err = map_private_extent_buffer(buf, offset, 32, 22919c00ddcSChris Mason &map_token, &kaddr, 23019c00ddcSChris Mason &map_start, &map_len, KM_USER0); 231d397712bSChris Mason if (err) 23219c00ddcSChris Mason return 1; 23319c00ddcSChris Mason cur_len = min(len, map_len - (offset - map_start)); 23419c00ddcSChris Mason crc = btrfs_csum_data(root, kaddr + offset - map_start, 23519c00ddcSChris Mason crc, cur_len); 23619c00ddcSChris Mason len -= cur_len; 23719c00ddcSChris Mason offset += cur_len; 23819c00ddcSChris Mason unmap_extent_buffer(buf, map_token, KM_USER0); 23919c00ddcSChris Mason } 240607d432dSJosef Bacik if (csum_size > sizeof(inline_result)) { 241607d432dSJosef Bacik result = kzalloc(csum_size * sizeof(char), GFP_NOFS); 242607d432dSJosef Bacik if (!result) 243607d432dSJosef Bacik return 1; 244607d432dSJosef Bacik } else { 245607d432dSJosef Bacik result = (char *)&inline_result; 246607d432dSJosef Bacik } 247607d432dSJosef Bacik 24819c00ddcSChris Mason btrfs_csum_final(crc, result); 24919c00ddcSChris Mason 25019c00ddcSChris Mason if (verify) { 251607d432dSJosef Bacik if (memcmp_extent_buffer(buf, result, 0, csum_size)) { 252e4204dedSChris Mason u32 val; 253e4204dedSChris Mason u32 found = 0; 254607d432dSJosef Bacik memcpy(&found, result, csum_size); 255e4204dedSChris Mason 256607d432dSJosef Bacik read_extent_buffer(buf, &val, 0, csum_size); 257193f284dSChris Mason if (printk_ratelimit()) { 258193f284dSChris Mason printk(KERN_INFO "btrfs: %s checksum verify " 259193f284dSChris Mason "failed on %llu wanted %X found %X " 260193f284dSChris Mason "level %d\n", 26119c00ddcSChris Mason root->fs_info->sb->s_id, 262193f284dSChris Mason (unsigned long long)buf->start, val, found, 263193f284dSChris Mason btrfs_header_level(buf)); 264193f284dSChris Mason } 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 } 299193f284dSChris Mason if (printk_ratelimit()) { 300193f284dSChris Mason printk("parent transid verify failed on %llu wanted %llu " 301193f284dSChris Mason "found %llu\n", 3021259ab75SChris Mason (unsigned long long)eb->start, 3031259ab75SChris Mason (unsigned long long)parent_transid, 3041259ab75SChris Mason (unsigned long long)btrfs_header_generation(eb)); 305193f284dSChris Mason } 3061259ab75SChris Mason ret = 1; 3072ac55d41SJosef Bacik clear_extent_buffer_uptodate(io_tree, eb, &cached_state); 30833958dc6SChris Mason out: 3092ac55d41SJosef Bacik unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 3102ac55d41SJosef Bacik &cached_state, GFP_NOFS); 3111259ab75SChris Mason return ret; 3121259ab75SChris Mason } 3131259ab75SChris Mason 314d352ac68SChris Mason /* 315d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 316d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 317d352ac68SChris Mason */ 318f188591eSChris Mason static int btree_read_extent_buffer_pages(struct btrfs_root *root, 319f188591eSChris Mason struct extent_buffer *eb, 320ca7a79adSChris Mason u64 start, u64 parent_transid) 321f188591eSChris Mason { 322f188591eSChris Mason struct extent_io_tree *io_tree; 323f188591eSChris Mason int ret; 324f188591eSChris Mason int num_copies = 0; 325f188591eSChris Mason int mirror_num = 0; 326f188591eSChris Mason 327a826d6dcSJosef Bacik clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 328f188591eSChris Mason io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree; 329f188591eSChris Mason while (1) { 330f188591eSChris Mason ret = read_extent_buffer_pages(io_tree, eb, start, 1, 331f188591eSChris Mason btree_get_extent, mirror_num); 3321259ab75SChris Mason if (!ret && 3331259ab75SChris Mason !verify_parent_transid(io_tree, eb, parent_transid)) 334f188591eSChris Mason return ret; 335d397712bSChris Mason 336a826d6dcSJosef Bacik /* 337a826d6dcSJosef Bacik * This buffer's crc is fine, but its contents are corrupted, so 338a826d6dcSJosef Bacik * there is no reason to read the other copies, they won't be 339a826d6dcSJosef Bacik * any less wrong. 340a826d6dcSJosef Bacik */ 341a826d6dcSJosef Bacik if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags)) 342a826d6dcSJosef Bacik return ret; 343a826d6dcSJosef Bacik 344f188591eSChris Mason num_copies = btrfs_num_copies(&root->fs_info->mapping_tree, 345f188591eSChris Mason eb->start, eb->len); 3464235298eSChris Mason if (num_copies == 1) 347f188591eSChris Mason return ret; 3484235298eSChris Mason 349f188591eSChris Mason mirror_num++; 3504235298eSChris Mason if (mirror_num > num_copies) 351f188591eSChris Mason return ret; 352f188591eSChris Mason } 353f188591eSChris Mason return -EIO; 354f188591eSChris Mason } 35519c00ddcSChris Mason 356d352ac68SChris Mason /* 357d397712bSChris Mason * checksum a dirty tree block before IO. This has extra checks to make sure 358d397712bSChris Mason * we only fill in the checksum field in the first page of a multi-page block 359d352ac68SChris Mason */ 360d397712bSChris Mason 361b2950863SChristoph Hellwig static int csum_dirty_buffer(struct btrfs_root *root, struct page *page) 36219c00ddcSChris Mason { 363d1310b2eSChris Mason struct extent_io_tree *tree; 36435ebb934SChris Mason u64 start = (u64)page->index << PAGE_CACHE_SHIFT; 36519c00ddcSChris Mason u64 found_start; 36619c00ddcSChris Mason unsigned long len; 36719c00ddcSChris Mason struct extent_buffer *eb; 368f188591eSChris Mason int ret; 369f188591eSChris Mason 370d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 37119c00ddcSChris Mason 372eb14ab8eSChris Mason if (page->private == EXTENT_PAGE_PRIVATE) { 373eb14ab8eSChris Mason WARN_ON(1); 37419c00ddcSChris Mason goto out; 375eb14ab8eSChris Mason } 376eb14ab8eSChris Mason if (!page->private) { 377eb14ab8eSChris Mason WARN_ON(1); 37819c00ddcSChris Mason goto out; 379eb14ab8eSChris Mason } 38019c00ddcSChris Mason len = page->private >> 2; 381d397712bSChris Mason WARN_ON(len == 0); 382d397712bSChris Mason 38319c00ddcSChris Mason eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS); 38491ca338dSTsutomu Itoh if (eb == NULL) { 38591ca338dSTsutomu Itoh WARN_ON(1); 38691ca338dSTsutomu Itoh goto out; 38791ca338dSTsutomu Itoh } 388ca7a79adSChris Mason ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE, 389ca7a79adSChris Mason btrfs_header_generation(eb)); 390f188591eSChris Mason BUG_ON(ret); 391784b4e29SChris Mason WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN)); 392784b4e29SChris Mason 39319c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 39419c00ddcSChris Mason if (found_start != start) { 39555c69072SChris Mason WARN_ON(1); 39655c69072SChris Mason goto err; 39755c69072SChris Mason } 39855c69072SChris Mason if (eb->first_page != page) { 39955c69072SChris Mason WARN_ON(1); 40055c69072SChris Mason goto err; 40155c69072SChris Mason } 40255c69072SChris Mason if (!PageUptodate(page)) { 40355c69072SChris Mason WARN_ON(1); 40455c69072SChris Mason goto err; 40519c00ddcSChris Mason } 40619c00ddcSChris Mason csum_tree_block(root, eb, 0); 40755c69072SChris Mason err: 40819c00ddcSChris Mason free_extent_buffer(eb); 40919c00ddcSChris Mason out: 41019c00ddcSChris Mason return 0; 41119c00ddcSChris Mason } 41219c00ddcSChris Mason 4132b82032cSYan Zheng static int check_tree_block_fsid(struct btrfs_root *root, 4142b82032cSYan Zheng struct extent_buffer *eb) 4152b82032cSYan Zheng { 4162b82032cSYan Zheng struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; 4172b82032cSYan Zheng u8 fsid[BTRFS_UUID_SIZE]; 4182b82032cSYan Zheng int ret = 1; 4192b82032cSYan Zheng 4202b82032cSYan Zheng read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb), 4212b82032cSYan Zheng BTRFS_FSID_SIZE); 4222b82032cSYan Zheng while (fs_devices) { 4232b82032cSYan Zheng if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) { 4242b82032cSYan Zheng ret = 0; 4252b82032cSYan Zheng break; 4262b82032cSYan Zheng } 4272b82032cSYan Zheng fs_devices = fs_devices->seed; 4282b82032cSYan Zheng } 4292b82032cSYan Zheng return ret; 4302b82032cSYan Zheng } 4312b82032cSYan Zheng 432a826d6dcSJosef Bacik #define CORRUPT(reason, eb, root, slot) \ 433a826d6dcSJosef Bacik printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \ 434a826d6dcSJosef Bacik "root=%llu, slot=%d\n", reason, \ 435a826d6dcSJosef Bacik (unsigned long long)btrfs_header_bytenr(eb), \ 436a826d6dcSJosef Bacik (unsigned long long)root->objectid, slot) 437a826d6dcSJosef Bacik 438a826d6dcSJosef Bacik static noinline int check_leaf(struct btrfs_root *root, 439a826d6dcSJosef Bacik struct extent_buffer *leaf) 440a826d6dcSJosef Bacik { 441a826d6dcSJosef Bacik struct btrfs_key key; 442a826d6dcSJosef Bacik struct btrfs_key leaf_key; 443a826d6dcSJosef Bacik u32 nritems = btrfs_header_nritems(leaf); 444a826d6dcSJosef Bacik int slot; 445a826d6dcSJosef Bacik 446a826d6dcSJosef Bacik if (nritems == 0) 447a826d6dcSJosef Bacik return 0; 448a826d6dcSJosef Bacik 449a826d6dcSJosef Bacik /* Check the 0 item */ 450a826d6dcSJosef Bacik if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) != 451a826d6dcSJosef Bacik BTRFS_LEAF_DATA_SIZE(root)) { 452a826d6dcSJosef Bacik CORRUPT("invalid item offset size pair", leaf, root, 0); 453a826d6dcSJosef Bacik return -EIO; 454a826d6dcSJosef Bacik } 455a826d6dcSJosef Bacik 456a826d6dcSJosef Bacik /* 457a826d6dcSJosef Bacik * Check to make sure each items keys are in the correct order and their 458a826d6dcSJosef Bacik * offsets make sense. We only have to loop through nritems-1 because 459a826d6dcSJosef Bacik * we check the current slot against the next slot, which verifies the 460a826d6dcSJosef Bacik * next slot's offset+size makes sense and that the current's slot 461a826d6dcSJosef Bacik * offset is correct. 462a826d6dcSJosef Bacik */ 463a826d6dcSJosef Bacik for (slot = 0; slot < nritems - 1; slot++) { 464a826d6dcSJosef Bacik btrfs_item_key_to_cpu(leaf, &leaf_key, slot); 465a826d6dcSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, slot + 1); 466a826d6dcSJosef Bacik 467a826d6dcSJosef Bacik /* Make sure the keys are in the right order */ 468a826d6dcSJosef Bacik if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) { 469a826d6dcSJosef Bacik CORRUPT("bad key order", leaf, root, slot); 470a826d6dcSJosef Bacik return -EIO; 471a826d6dcSJosef Bacik } 472a826d6dcSJosef Bacik 473a826d6dcSJosef Bacik /* 474a826d6dcSJosef Bacik * Make sure the offset and ends are right, remember that the 475a826d6dcSJosef Bacik * item data starts at the end of the leaf and grows towards the 476a826d6dcSJosef Bacik * front. 477a826d6dcSJosef Bacik */ 478a826d6dcSJosef Bacik if (btrfs_item_offset_nr(leaf, slot) != 479a826d6dcSJosef Bacik btrfs_item_end_nr(leaf, slot + 1)) { 480a826d6dcSJosef Bacik CORRUPT("slot offset bad", leaf, root, slot); 481a826d6dcSJosef Bacik return -EIO; 482a826d6dcSJosef Bacik } 483a826d6dcSJosef Bacik 484a826d6dcSJosef Bacik /* 485a826d6dcSJosef Bacik * Check to make sure that we don't point outside of the leaf, 486a826d6dcSJosef Bacik * just incase all the items are consistent to eachother, but 487a826d6dcSJosef Bacik * all point outside of the leaf. 488a826d6dcSJosef Bacik */ 489a826d6dcSJosef Bacik if (btrfs_item_end_nr(leaf, slot) > 490a826d6dcSJosef Bacik BTRFS_LEAF_DATA_SIZE(root)) { 491a826d6dcSJosef Bacik CORRUPT("slot end outside of leaf", leaf, root, slot); 492a826d6dcSJosef Bacik return -EIO; 493a826d6dcSJosef Bacik } 494a826d6dcSJosef Bacik } 495a826d6dcSJosef Bacik 496a826d6dcSJosef Bacik return 0; 497a826d6dcSJosef Bacik } 498a826d6dcSJosef Bacik 4994008c04aSChris Mason #ifdef CONFIG_DEBUG_LOCK_ALLOC 5004008c04aSChris Mason void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level) 5014008c04aSChris Mason { 5024008c04aSChris Mason lockdep_set_class_and_name(&eb->lock, 5034008c04aSChris Mason &btrfs_eb_class[level], 5044008c04aSChris Mason btrfs_eb_name[level]); 5054008c04aSChris Mason } 5064008c04aSChris Mason #endif 5074008c04aSChris Mason 508b2950863SChristoph Hellwig static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end, 509ce9adaa5SChris Mason struct extent_state *state) 510ce9adaa5SChris Mason { 511ce9adaa5SChris Mason struct extent_io_tree *tree; 512ce9adaa5SChris Mason u64 found_start; 513ce9adaa5SChris Mason int found_level; 514ce9adaa5SChris Mason unsigned long len; 515ce9adaa5SChris Mason struct extent_buffer *eb; 516ce9adaa5SChris Mason struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 517f188591eSChris Mason int ret = 0; 518ce9adaa5SChris Mason 519ce9adaa5SChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 520ce9adaa5SChris Mason if (page->private == EXTENT_PAGE_PRIVATE) 521ce9adaa5SChris Mason goto out; 522ce9adaa5SChris Mason if (!page->private) 523ce9adaa5SChris Mason goto out; 524d397712bSChris Mason 525ce9adaa5SChris Mason len = page->private >> 2; 526d397712bSChris Mason WARN_ON(len == 0); 527d397712bSChris Mason 528ce9adaa5SChris Mason eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS); 52991ca338dSTsutomu Itoh if (eb == NULL) { 53091ca338dSTsutomu Itoh ret = -EIO; 53191ca338dSTsutomu Itoh goto out; 53291ca338dSTsutomu Itoh } 533f188591eSChris Mason 534ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 53523a07867SChris Mason if (found_start != start) { 536193f284dSChris Mason if (printk_ratelimit()) { 537193f284dSChris Mason printk(KERN_INFO "btrfs bad tree block start " 538193f284dSChris Mason "%llu %llu\n", 539a1b32a59SChris Mason (unsigned long long)found_start, 540a1b32a59SChris Mason (unsigned long long)eb->start); 541193f284dSChris Mason } 542f188591eSChris Mason ret = -EIO; 543ce9adaa5SChris Mason goto err; 544ce9adaa5SChris Mason } 545ce9adaa5SChris Mason if (eb->first_page != page) { 546d397712bSChris Mason printk(KERN_INFO "btrfs bad first page %lu %lu\n", 547d397712bSChris Mason eb->first_page->index, page->index); 548ce9adaa5SChris Mason WARN_ON(1); 549f188591eSChris Mason ret = -EIO; 550ce9adaa5SChris Mason goto err; 551ce9adaa5SChris Mason } 5522b82032cSYan Zheng if (check_tree_block_fsid(root, eb)) { 553193f284dSChris Mason if (printk_ratelimit()) { 554d397712bSChris Mason printk(KERN_INFO "btrfs bad fsid on block %llu\n", 555d397712bSChris Mason (unsigned long long)eb->start); 556193f284dSChris Mason } 5571259ab75SChris Mason ret = -EIO; 5581259ab75SChris Mason goto err; 5591259ab75SChris Mason } 560ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 561ce9adaa5SChris Mason 5624008c04aSChris Mason btrfs_set_buffer_lockdep_class(eb, found_level); 5634008c04aSChris Mason 564ce9adaa5SChris Mason ret = csum_tree_block(root, eb, 1); 565a826d6dcSJosef Bacik if (ret) { 566f188591eSChris Mason ret = -EIO; 567a826d6dcSJosef Bacik goto err; 568a826d6dcSJosef Bacik } 569a826d6dcSJosef Bacik 570a826d6dcSJosef Bacik /* 571a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 572a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 573a826d6dcSJosef Bacik * return -EIO. 574a826d6dcSJosef Bacik */ 575a826d6dcSJosef Bacik if (found_level == 0 && check_leaf(root, eb)) { 576a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 577a826d6dcSJosef Bacik ret = -EIO; 578a826d6dcSJosef Bacik } 579ce9adaa5SChris Mason 580ce9adaa5SChris Mason end = min_t(u64, eb->len, PAGE_CACHE_SIZE); 581ce9adaa5SChris Mason end = eb->start + end - 1; 582ce9adaa5SChris Mason err: 583ce9adaa5SChris Mason free_extent_buffer(eb); 584ce9adaa5SChris Mason out: 585f188591eSChris Mason return ret; 586ce9adaa5SChris Mason } 587ce9adaa5SChris Mason 588ce9adaa5SChris Mason static void end_workqueue_bio(struct bio *bio, int err) 589ce9adaa5SChris Mason { 590ce9adaa5SChris Mason struct end_io_wq *end_io_wq = bio->bi_private; 591ce9adaa5SChris Mason struct btrfs_fs_info *fs_info; 592ce9adaa5SChris Mason 593ce9adaa5SChris Mason fs_info = end_io_wq->info; 594ce9adaa5SChris Mason end_io_wq->error = err; 5958b712842SChris Mason end_io_wq->work.func = end_workqueue_fn; 5968b712842SChris Mason end_io_wq->work.flags = 0; 597d20f7043SChris Mason 5987b6d91daSChristoph Hellwig if (bio->bi_rw & REQ_WRITE) { 5990cb59c99SJosef Bacik if (end_io_wq->metadata == 1) 600cad321adSChris Mason btrfs_queue_worker(&fs_info->endio_meta_write_workers, 601cad321adSChris Mason &end_io_wq->work); 6020cb59c99SJosef Bacik else if (end_io_wq->metadata == 2) 6030cb59c99SJosef Bacik btrfs_queue_worker(&fs_info->endio_freespace_worker, 6040cb59c99SJosef Bacik &end_io_wq->work); 605cad321adSChris Mason else 606e6dcd2dcSChris Mason btrfs_queue_worker(&fs_info->endio_write_workers, 607e6dcd2dcSChris Mason &end_io_wq->work); 608d20f7043SChris Mason } else { 609d20f7043SChris Mason if (end_io_wq->metadata) 610d20f7043SChris Mason btrfs_queue_worker(&fs_info->endio_meta_workers, 611d20f7043SChris Mason &end_io_wq->work); 612e6dcd2dcSChris Mason else 613d20f7043SChris Mason btrfs_queue_worker(&fs_info->endio_workers, 614d20f7043SChris Mason &end_io_wq->work); 615d20f7043SChris Mason } 616ce9adaa5SChris Mason } 617ce9adaa5SChris Mason 6180cb59c99SJosef Bacik /* 6190cb59c99SJosef Bacik * For the metadata arg you want 6200cb59c99SJosef Bacik * 6210cb59c99SJosef Bacik * 0 - if data 6220cb59c99SJosef Bacik * 1 - if normal metadta 6230cb59c99SJosef Bacik * 2 - if writing to the free space cache area 6240cb59c99SJosef Bacik */ 62522c59948SChris Mason int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 62622c59948SChris Mason int metadata) 6270b86a832SChris Mason { 628ce9adaa5SChris Mason struct end_io_wq *end_io_wq; 629ce9adaa5SChris Mason end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS); 630ce9adaa5SChris Mason if (!end_io_wq) 631ce9adaa5SChris Mason return -ENOMEM; 632ce9adaa5SChris Mason 633ce9adaa5SChris Mason end_io_wq->private = bio->bi_private; 634ce9adaa5SChris Mason end_io_wq->end_io = bio->bi_end_io; 63522c59948SChris Mason end_io_wq->info = info; 636ce9adaa5SChris Mason end_io_wq->error = 0; 637ce9adaa5SChris Mason end_io_wq->bio = bio; 63822c59948SChris Mason end_io_wq->metadata = metadata; 639ce9adaa5SChris Mason 640ce9adaa5SChris Mason bio->bi_private = end_io_wq; 641ce9adaa5SChris Mason bio->bi_end_io = end_workqueue_bio; 64222c59948SChris Mason return 0; 64322c59948SChris Mason } 64422c59948SChris Mason 645b64a2851SChris Mason unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info) 6464854ddd0SChris Mason { 6474854ddd0SChris Mason unsigned long limit = min_t(unsigned long, 6484854ddd0SChris Mason info->workers.max_workers, 6494854ddd0SChris Mason info->fs_devices->open_devices); 6504854ddd0SChris Mason return 256 * limit; 6514854ddd0SChris Mason } 6524854ddd0SChris Mason 653777e6bd7SChris Mason int btrfs_congested_async(struct btrfs_fs_info *info, int iodone) 6540986fe9eSChris Mason { 655b64a2851SChris Mason return atomic_read(&info->nr_async_bios) > 656b64a2851SChris Mason btrfs_async_submit_limit(info); 6570986fe9eSChris Mason } 6580986fe9eSChris Mason 6594a69a410SChris Mason static void run_one_async_start(struct btrfs_work *work) 6604a69a410SChris Mason { 6614a69a410SChris Mason struct async_submit_bio *async; 6624a69a410SChris Mason 6634a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 6644a69a410SChris Mason async->submit_bio_start(async->inode, async->rw, async->bio, 665eaf25d93SChris Mason async->mirror_num, async->bio_flags, 666eaf25d93SChris Mason async->bio_offset); 6674a69a410SChris Mason } 6684a69a410SChris Mason 6694a69a410SChris Mason static void run_one_async_done(struct btrfs_work *work) 6708b712842SChris Mason { 6718b712842SChris Mason struct btrfs_fs_info *fs_info; 6728b712842SChris Mason struct async_submit_bio *async; 6734854ddd0SChris Mason int limit; 6748b712842SChris Mason 6758b712842SChris Mason async = container_of(work, struct async_submit_bio, work); 6768b712842SChris Mason fs_info = BTRFS_I(async->inode)->root->fs_info; 6774854ddd0SChris Mason 678b64a2851SChris Mason limit = btrfs_async_submit_limit(fs_info); 6794854ddd0SChris Mason limit = limit * 2 / 3; 6804854ddd0SChris Mason 6818b712842SChris Mason atomic_dec(&fs_info->nr_async_submits); 6820986fe9eSChris Mason 683b64a2851SChris Mason if (atomic_read(&fs_info->nr_async_submits) < limit && 684b64a2851SChris Mason waitqueue_active(&fs_info->async_submit_wait)) 6854854ddd0SChris Mason wake_up(&fs_info->async_submit_wait); 6864854ddd0SChris Mason 6874a69a410SChris Mason async->submit_bio_done(async->inode, async->rw, async->bio, 688eaf25d93SChris Mason async->mirror_num, async->bio_flags, 689eaf25d93SChris Mason async->bio_offset); 6904a69a410SChris Mason } 6914a69a410SChris Mason 6924a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 6934a69a410SChris Mason { 6944a69a410SChris Mason struct async_submit_bio *async; 6954a69a410SChris Mason 6964a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 6978b712842SChris Mason kfree(async); 6988b712842SChris Mason } 6998b712842SChris Mason 70044b8bd7eSChris Mason int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode, 70144b8bd7eSChris Mason int rw, struct bio *bio, int mirror_num, 702c8b97818SChris Mason unsigned long bio_flags, 703eaf25d93SChris Mason u64 bio_offset, 7044a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_start, 7054a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_done) 70644b8bd7eSChris Mason { 70744b8bd7eSChris Mason struct async_submit_bio *async; 70844b8bd7eSChris Mason 70944b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 71044b8bd7eSChris Mason if (!async) 71144b8bd7eSChris Mason return -ENOMEM; 71244b8bd7eSChris Mason 71344b8bd7eSChris Mason async->inode = inode; 71444b8bd7eSChris Mason async->rw = rw; 71544b8bd7eSChris Mason async->bio = bio; 71644b8bd7eSChris Mason async->mirror_num = mirror_num; 7174a69a410SChris Mason async->submit_bio_start = submit_bio_start; 7184a69a410SChris Mason async->submit_bio_done = submit_bio_done; 7194a69a410SChris Mason 7204a69a410SChris Mason async->work.func = run_one_async_start; 7214a69a410SChris Mason async->work.ordered_func = run_one_async_done; 7224a69a410SChris Mason async->work.ordered_free = run_one_async_free; 7234a69a410SChris Mason 7248b712842SChris Mason async->work.flags = 0; 725c8b97818SChris Mason async->bio_flags = bio_flags; 726eaf25d93SChris Mason async->bio_offset = bio_offset; 7278c8bee1dSChris Mason 728cb03c743SChris Mason atomic_inc(&fs_info->nr_async_submits); 729d313d7a3SChris Mason 7307b6d91daSChristoph Hellwig if (rw & REQ_SYNC) 731d313d7a3SChris Mason btrfs_set_work_high_prio(&async->work); 732d313d7a3SChris Mason 7338b712842SChris Mason btrfs_queue_worker(&fs_info->workers, &async->work); 7349473f16cSChris Mason 735771ed689SChris Mason while (atomic_read(&fs_info->async_submit_draining) && 736771ed689SChris Mason atomic_read(&fs_info->nr_async_submits)) { 737771ed689SChris Mason wait_event(fs_info->async_submit_wait, 738771ed689SChris Mason (atomic_read(&fs_info->nr_async_submits) == 0)); 739771ed689SChris Mason } 740771ed689SChris Mason 74144b8bd7eSChris Mason return 0; 74244b8bd7eSChris Mason } 74344b8bd7eSChris Mason 744ce3ed71aSChris Mason static int btree_csum_one_bio(struct bio *bio) 745ce3ed71aSChris Mason { 746ce3ed71aSChris Mason struct bio_vec *bvec = bio->bi_io_vec; 747ce3ed71aSChris Mason int bio_index = 0; 748ce3ed71aSChris Mason struct btrfs_root *root; 749ce3ed71aSChris Mason 750ce3ed71aSChris Mason WARN_ON(bio->bi_vcnt <= 0); 751ce3ed71aSChris Mason while (bio_index < bio->bi_vcnt) { 752ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 753ce3ed71aSChris Mason csum_dirty_buffer(root, bvec->bv_page); 754ce3ed71aSChris Mason bio_index++; 755ce3ed71aSChris Mason bvec++; 756ce3ed71aSChris Mason } 757ce3ed71aSChris Mason return 0; 758ce3ed71aSChris Mason } 759ce3ed71aSChris Mason 7604a69a410SChris Mason static int __btree_submit_bio_start(struct inode *inode, int rw, 7614a69a410SChris Mason struct bio *bio, int mirror_num, 762eaf25d93SChris Mason unsigned long bio_flags, 763eaf25d93SChris Mason u64 bio_offset) 76422c59948SChris Mason { 7658b712842SChris Mason /* 7668b712842SChris Mason * when we're called for a write, we're already in the async 7675443be45SChris Mason * submission context. Just jump into btrfs_map_bio 7688b712842SChris Mason */ 769ce3ed71aSChris Mason btree_csum_one_bio(bio); 7704a69a410SChris Mason return 0; 77122c59948SChris Mason } 77222c59948SChris Mason 7734a69a410SChris Mason static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio, 774eaf25d93SChris Mason int mirror_num, unsigned long bio_flags, 775eaf25d93SChris Mason u64 bio_offset) 7764a69a410SChris Mason { 7778b712842SChris Mason /* 7784a69a410SChris Mason * when we're called for a write, we're already in the async 7794a69a410SChris Mason * submission context. Just jump into btrfs_map_bio 7808b712842SChris Mason */ 7818b712842SChris Mason return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1); 7820b86a832SChris Mason } 7830b86a832SChris Mason 78444b8bd7eSChris Mason static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, 785eaf25d93SChris Mason int mirror_num, unsigned long bio_flags, 786eaf25d93SChris Mason u64 bio_offset) 78744b8bd7eSChris Mason { 7884a69a410SChris Mason int ret; 789cad321adSChris Mason 7904a69a410SChris Mason ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info, 7914a69a410SChris Mason bio, 1); 7924a69a410SChris Mason BUG_ON(ret); 7934a69a410SChris Mason 7947b6d91daSChristoph Hellwig if (!(rw & REQ_WRITE)) { 795cad321adSChris Mason /* 796cad321adSChris Mason * called for a read, do the setup so that checksum validation 797cad321adSChris Mason * can happen in the async kernel threads 798cad321adSChris Mason */ 7994a69a410SChris Mason return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, 8006f3577bdSChris Mason mirror_num, 0); 80144b8bd7eSChris Mason } 802d313d7a3SChris Mason 803cad321adSChris Mason /* 804cad321adSChris Mason * kthread helpers are used to submit writes so that checksumming 805cad321adSChris Mason * can happen in parallel across all CPUs 806cad321adSChris Mason */ 80744b8bd7eSChris Mason return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, 808c8b97818SChris Mason inode, rw, bio, mirror_num, 0, 809eaf25d93SChris Mason bio_offset, 8104a69a410SChris Mason __btree_submit_bio_start, 8114a69a410SChris Mason __btree_submit_bio_done); 81244b8bd7eSChris Mason } 81344b8bd7eSChris Mason 8143dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 815784b4e29SChris Mason static int btree_migratepage(struct address_space *mapping, 816784b4e29SChris Mason struct page *newpage, struct page *page) 817784b4e29SChris Mason { 818784b4e29SChris Mason /* 819784b4e29SChris Mason * we can't safely write a btree page from here, 820784b4e29SChris Mason * we haven't done the locking hook 821784b4e29SChris Mason */ 822784b4e29SChris Mason if (PageDirty(page)) 823784b4e29SChris Mason return -EAGAIN; 824784b4e29SChris Mason /* 825784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 826784b4e29SChris Mason * We must have no buffers or drop them. 827784b4e29SChris Mason */ 828784b4e29SChris Mason if (page_has_private(page) && 829784b4e29SChris Mason !try_to_release_page(page, GFP_KERNEL)) 830784b4e29SChris Mason return -EAGAIN; 831784b4e29SChris Mason return migrate_page(mapping, newpage, page); 832784b4e29SChris Mason } 8333dd1462eSJan Beulich #endif 834784b4e29SChris Mason 8355f39d397SChris Mason static int btree_writepage(struct page *page, struct writeback_control *wbc) 8365f39d397SChris Mason { 837d1310b2eSChris Mason struct extent_io_tree *tree; 838b9473439SChris Mason struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 839b9473439SChris Mason struct extent_buffer *eb; 840b9473439SChris Mason int was_dirty; 8415443be45SChris Mason 842b9473439SChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 843b9473439SChris Mason if (!(current->flags & PF_MEMALLOC)) { 844b9473439SChris Mason return extent_write_full_page(tree, page, 845b9473439SChris Mason btree_get_extent, wbc); 846b9473439SChris Mason } 847b9473439SChris Mason 8485443be45SChris Mason redirty_page_for_writepage(wbc, page); 849784b4e29SChris Mason eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE); 850b9473439SChris Mason WARN_ON(!eb); 851b9473439SChris Mason 852b9473439SChris Mason was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); 853b9473439SChris Mason if (!was_dirty) { 854b9473439SChris Mason spin_lock(&root->fs_info->delalloc_lock); 855b9473439SChris Mason root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE; 856b9473439SChris Mason spin_unlock(&root->fs_info->delalloc_lock); 857b9473439SChris Mason } 858b9473439SChris Mason free_extent_buffer(eb); 859b9473439SChris Mason 8605443be45SChris Mason unlock_page(page); 8615443be45SChris Mason return 0; 8625443be45SChris Mason } 8630da5468fSChris Mason 8640da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 8650da5468fSChris Mason struct writeback_control *wbc) 8660da5468fSChris Mason { 867d1310b2eSChris Mason struct extent_io_tree *tree; 868d1310b2eSChris Mason tree = &BTRFS_I(mapping->host)->io_tree; 869d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 870b9473439SChris Mason struct btrfs_root *root = BTRFS_I(mapping->host)->root; 871793955bcSChris Mason u64 num_dirty; 87224ab9cd8SChris Mason unsigned long thresh = 32 * 1024 * 1024; 873448d640bSChris Mason 874448d640bSChris Mason if (wbc->for_kupdate) 875448d640bSChris Mason return 0; 876448d640bSChris Mason 877b9473439SChris Mason /* this is a bit racy, but that's ok */ 878b9473439SChris Mason num_dirty = root->fs_info->dirty_metadata_bytes; 879d397712bSChris Mason if (num_dirty < thresh) 880793955bcSChris Mason return 0; 881793955bcSChris Mason } 8820da5468fSChris Mason return extent_writepages(tree, mapping, btree_get_extent, wbc); 8830da5468fSChris Mason } 8840da5468fSChris Mason 885b2950863SChristoph Hellwig static int btree_readpage(struct file *file, struct page *page) 8865f39d397SChris Mason { 887d1310b2eSChris Mason struct extent_io_tree *tree; 888d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 8895f39d397SChris Mason return extent_read_full_page(tree, page, btree_get_extent); 8905f39d397SChris Mason } 8915f39d397SChris Mason 89270dec807SChris Mason static int btree_releasepage(struct page *page, gfp_t gfp_flags) 8935f39d397SChris Mason { 894d1310b2eSChris Mason struct extent_io_tree *tree; 895d1310b2eSChris Mason struct extent_map_tree *map; 8965f39d397SChris Mason int ret; 8975f39d397SChris Mason 89898509cfcSChris Mason if (PageWriteback(page) || PageDirty(page)) 89998509cfcSChris Mason return 0; 90098509cfcSChris Mason 901d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 902d1310b2eSChris Mason map = &BTRFS_I(page->mapping->host)->extent_tree; 9036af118ceSChris Mason 9047b13b7b1SChris Mason ret = try_release_extent_state(map, tree, page, gfp_flags); 905d397712bSChris Mason if (!ret) 9066af118ceSChris Mason return 0; 9076af118ceSChris Mason 9086af118ceSChris Mason ret = try_release_extent_buffer(tree, page); 9095f39d397SChris Mason if (ret == 1) { 9105f39d397SChris Mason ClearPagePrivate(page); 9115f39d397SChris Mason set_page_private(page, 0); 912d98237b3SChris Mason page_cache_release(page); 9135f39d397SChris Mason } 9146af118ceSChris Mason 915d98237b3SChris Mason return ret; 916d98237b3SChris Mason } 917d98237b3SChris Mason 9185f39d397SChris Mason static void btree_invalidatepage(struct page *page, unsigned long offset) 919d98237b3SChris Mason { 920d1310b2eSChris Mason struct extent_io_tree *tree; 921d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9225f39d397SChris Mason extent_invalidatepage(tree, page, offset); 9235f39d397SChris Mason btree_releasepage(page, GFP_NOFS); 9249ad6b7bcSChris Mason if (PagePrivate(page)) { 925d397712bSChris Mason printk(KERN_WARNING "btrfs warning page private not zero " 926d397712bSChris Mason "on page %llu\n", (unsigned long long)page_offset(page)); 9279ad6b7bcSChris Mason ClearPagePrivate(page); 9289ad6b7bcSChris Mason set_page_private(page, 0); 9299ad6b7bcSChris Mason page_cache_release(page); 9309ad6b7bcSChris Mason } 931d98237b3SChris Mason } 932d98237b3SChris Mason 9337f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 934d98237b3SChris Mason .readpage = btree_readpage, 935d98237b3SChris Mason .writepage = btree_writepage, 9360da5468fSChris Mason .writepages = btree_writepages, 9375f39d397SChris Mason .releasepage = btree_releasepage, 9385f39d397SChris Mason .invalidatepage = btree_invalidatepage, 939d98237b3SChris Mason .sync_page = block_sync_page, 9405a92bc88SChris Mason #ifdef CONFIG_MIGRATION 941784b4e29SChris Mason .migratepage = btree_migratepage, 9425a92bc88SChris Mason #endif 943d98237b3SChris Mason }; 944123abc88SChris Mason 945ca7a79adSChris Mason int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize, 946ca7a79adSChris Mason u64 parent_transid) 947090d1875SChris Mason { 9485f39d397SChris Mason struct extent_buffer *buf = NULL; 9495f39d397SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 950de428b63SChris Mason int ret = 0; 951090d1875SChris Mason 952db94535dSChris Mason buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 9535f39d397SChris Mason if (!buf) 954090d1875SChris Mason return 0; 955d1310b2eSChris Mason read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, 956f188591eSChris Mason buf, 0, 0, btree_get_extent, 0); 9575f39d397SChris Mason free_extent_buffer(buf); 958de428b63SChris Mason return ret; 959090d1875SChris Mason } 960090d1875SChris Mason 9610999df54SChris Mason struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root, 9620999df54SChris Mason u64 bytenr, u32 blocksize) 9630999df54SChris Mason { 9640999df54SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 9650999df54SChris Mason struct extent_buffer *eb; 9660999df54SChris Mason eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree, 9670999df54SChris Mason bytenr, blocksize, GFP_NOFS); 9680999df54SChris Mason return eb; 9690999df54SChris Mason } 9700999df54SChris Mason 9710999df54SChris Mason struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root, 9720999df54SChris Mason u64 bytenr, u32 blocksize) 9730999df54SChris Mason { 9740999df54SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 9750999df54SChris Mason struct extent_buffer *eb; 9760999df54SChris Mason 9770999df54SChris Mason eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree, 9780999df54SChris Mason bytenr, blocksize, NULL, GFP_NOFS); 9790999df54SChris Mason return eb; 9800999df54SChris Mason } 9810999df54SChris Mason 9820999df54SChris Mason 983e02119d5SChris Mason int btrfs_write_tree_block(struct extent_buffer *buf) 984e02119d5SChris Mason { 9858aa38c31SChristoph Hellwig return filemap_fdatawrite_range(buf->first_page->mapping, buf->start, 9868aa38c31SChristoph Hellwig buf->start + buf->len - 1); 987e02119d5SChris Mason } 988e02119d5SChris Mason 989e02119d5SChris Mason int btrfs_wait_tree_block_writeback(struct extent_buffer *buf) 990e02119d5SChris Mason { 9918aa38c31SChristoph Hellwig return filemap_fdatawait_range(buf->first_page->mapping, 9928aa38c31SChristoph Hellwig buf->start, buf->start + buf->len - 1); 993e02119d5SChris Mason } 994e02119d5SChris Mason 995db94535dSChris Mason struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr, 996ca7a79adSChris Mason u32 blocksize, u64 parent_transid) 997e20d96d6SChris Mason { 9985f39d397SChris Mason struct extent_buffer *buf = NULL; 99919c00ddcSChris Mason int ret; 100019c00ddcSChris Mason 1001db94535dSChris Mason buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 10025f39d397SChris Mason if (!buf) 1003d98237b3SChris Mason return NULL; 1004e4204dedSChris Mason 1005ca7a79adSChris Mason ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 1006ce9adaa5SChris Mason 1007d397712bSChris Mason if (ret == 0) 1008b4ce94deSChris Mason set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags); 10095f39d397SChris Mason return buf; 1010ce9adaa5SChris Mason 1011eb60ceacSChris Mason } 1012eb60ceacSChris Mason 1013e089f05cSChris Mason int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root, 10145f39d397SChris Mason struct extent_buffer *buf) 1015ed2ff2cbSChris Mason { 10165f39d397SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 101755c69072SChris Mason if (btrfs_header_generation(buf) == 1018925baeddSChris Mason root->fs_info->running_transaction->transid) { 1019b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1020b4ce94deSChris Mason 1021b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1022b9473439SChris Mason spin_lock(&root->fs_info->delalloc_lock); 1023b9473439SChris Mason if (root->fs_info->dirty_metadata_bytes >= buf->len) 1024b9473439SChris Mason root->fs_info->dirty_metadata_bytes -= buf->len; 1025b9473439SChris Mason else 1026b9473439SChris Mason WARN_ON(1); 1027b9473439SChris Mason spin_unlock(&root->fs_info->delalloc_lock); 1028b9473439SChris Mason } 1029b4ce94deSChris Mason 1030b9473439SChris Mason /* ugh, clear_extent_buffer_dirty needs to lock the page */ 1031b9473439SChris Mason btrfs_set_lock_blocking(buf); 1032d1310b2eSChris Mason clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, 103355c69072SChris Mason buf); 1034925baeddSChris Mason } 10355f39d397SChris Mason return 0; 10365f39d397SChris Mason } 10375f39d397SChris Mason 1038db94535dSChris Mason static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize, 103987ee04ebSChris Mason u32 stripesize, struct btrfs_root *root, 10409f5fae2fSChris Mason struct btrfs_fs_info *fs_info, 1041e20d96d6SChris Mason u64 objectid) 1042d97e63b6SChris Mason { 1043cfaa7295SChris Mason root->node = NULL; 1044a28ec197SChris Mason root->commit_root = NULL; 1045db94535dSChris Mason root->sectorsize = sectorsize; 1046db94535dSChris Mason root->nodesize = nodesize; 1047db94535dSChris Mason root->leafsize = leafsize; 104887ee04ebSChris Mason root->stripesize = stripesize; 1049123abc88SChris Mason root->ref_cows = 0; 10500b86a832SChris Mason root->track_dirty = 0; 1051c71bf099SYan, Zheng root->in_radix = 0; 1052d68fc57bSYan, Zheng root->orphan_item_inserted = 0; 1053d68fc57bSYan, Zheng root->orphan_cleanup_state = 0; 10540b86a832SChris Mason 10559f5fae2fSChris Mason root->fs_info = fs_info; 10560f7d52f4SChris Mason root->objectid = objectid; 10570f7d52f4SChris Mason root->last_trans = 0; 105813a8a7c8SYan, Zheng root->highest_objectid = 0; 105958176a96SJosef Bacik root->name = NULL; 10604313b399SChris Mason root->in_sysfs = 0; 10616bef4d31SEric Paris root->inode_tree = RB_ROOT; 1062f0486c68SYan, Zheng root->block_rsv = NULL; 1063d68fc57bSYan, Zheng root->orphan_block_rsv = NULL; 10640b86a832SChris Mason 10650b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 10667b128766SJosef Bacik INIT_LIST_HEAD(&root->orphan_list); 10675d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 1068925baeddSChris Mason spin_lock_init(&root->node_lock); 1069d68fc57bSYan, Zheng spin_lock_init(&root->orphan_lock); 10705d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1071f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 1072a2135011SChris Mason mutex_init(&root->objectid_mutex); 1073e02119d5SChris Mason mutex_init(&root->log_mutex); 10747237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 10757237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 10767237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 10777237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 10787237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 10797237f183SYan Zheng atomic_set(&root->log_writers, 0); 10807237f183SYan Zheng root->log_batch = 0; 10817237f183SYan Zheng root->log_transid = 0; 1082257c62e1SChris Mason root->last_log_commit = 0; 1083d0c803c4SChris Mason extent_io_tree_init(&root->dirty_log_pages, 1084d0c803c4SChris Mason fs_info->btree_inode->i_mapping, GFP_NOFS); 1085017e5369SChris Mason 10863768f368SChris Mason memset(&root->root_key, 0, sizeof(root->root_key)); 10873768f368SChris Mason memset(&root->root_item, 0, sizeof(root->root_item)); 10886702ed49SChris Mason memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 108958176a96SJosef Bacik memset(&root->root_kobj, 0, sizeof(root->root_kobj)); 10903f157a2fSChris Mason root->defrag_trans_start = fs_info->generation; 109158176a96SJosef Bacik init_completion(&root->kobj_unregister); 10926702ed49SChris Mason root->defrag_running = 0; 10934d775673SChris Mason root->root_key.objectid = objectid; 10943394e160SChris Mason root->anon_super.s_root = NULL; 10953394e160SChris Mason root->anon_super.s_dev = 0; 10963394e160SChris Mason INIT_LIST_HEAD(&root->anon_super.s_list); 10973394e160SChris Mason INIT_LIST_HEAD(&root->anon_super.s_instances); 10983394e160SChris Mason init_rwsem(&root->anon_super.s_umount); 10993394e160SChris Mason 11003768f368SChris Mason return 0; 11013768f368SChris Mason } 11023768f368SChris Mason 1103db94535dSChris Mason static int find_and_setup_root(struct btrfs_root *tree_root, 11049f5fae2fSChris Mason struct btrfs_fs_info *fs_info, 11059f5fae2fSChris Mason u64 objectid, 1106e20d96d6SChris Mason struct btrfs_root *root) 11073768f368SChris Mason { 11083768f368SChris Mason int ret; 1109db94535dSChris Mason u32 blocksize; 111084234f3aSYan Zheng u64 generation; 11113768f368SChris Mason 1112db94535dSChris Mason __setup_root(tree_root->nodesize, tree_root->leafsize, 111387ee04ebSChris Mason tree_root->sectorsize, tree_root->stripesize, 111487ee04ebSChris Mason root, fs_info, objectid); 11153768f368SChris Mason ret = btrfs_find_last_root(tree_root, objectid, 11163768f368SChris Mason &root->root_item, &root->root_key); 11174df27c4dSYan, Zheng if (ret > 0) 11184df27c4dSYan, Zheng return -ENOENT; 11193768f368SChris Mason BUG_ON(ret); 11203768f368SChris Mason 112184234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1122db94535dSChris Mason blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item)); 1123db94535dSChris Mason root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item), 112484234f3aSYan Zheng blocksize, generation); 112568433b73SChris Mason if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) { 112668433b73SChris Mason free_extent_buffer(root->node); 112768433b73SChris Mason return -EIO; 112868433b73SChris Mason } 11294df27c4dSYan, Zheng root->commit_root = btrfs_root_node(root); 1130d97e63b6SChris Mason return 0; 1131d97e63b6SChris Mason } 1132d97e63b6SChris Mason 11337237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1134e02119d5SChris Mason struct btrfs_fs_info *fs_info) 11350f7d52f4SChris Mason { 11360f7d52f4SChris Mason struct btrfs_root *root; 11370f7d52f4SChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 11387237f183SYan Zheng struct extent_buffer *leaf; 1139e02119d5SChris Mason 1140e02119d5SChris Mason root = kzalloc(sizeof(*root), GFP_NOFS); 1141e02119d5SChris Mason if (!root) 11427237f183SYan Zheng return ERR_PTR(-ENOMEM); 1143e02119d5SChris Mason 1144e02119d5SChris Mason __setup_root(tree_root->nodesize, tree_root->leafsize, 1145e02119d5SChris Mason tree_root->sectorsize, tree_root->stripesize, 1146e02119d5SChris Mason root, fs_info, BTRFS_TREE_LOG_OBJECTID); 1147e02119d5SChris Mason 1148e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1149e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1150e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 11517237f183SYan Zheng /* 11527237f183SYan Zheng * log trees do not get reference counted because they go away 11537237f183SYan Zheng * before a real commit is actually done. They do store pointers 11547237f183SYan Zheng * to file data extents, and those reference counts still get 11557237f183SYan Zheng * updated (along with back refs to the log tree). 11567237f183SYan Zheng */ 1157e02119d5SChris Mason root->ref_cows = 0; 1158e02119d5SChris Mason 11595d4f98a2SYan Zheng leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0, 11605d4f98a2SYan Zheng BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0); 11617237f183SYan Zheng if (IS_ERR(leaf)) { 11627237f183SYan Zheng kfree(root); 11637237f183SYan Zheng return ERR_CAST(leaf); 11647237f183SYan Zheng } 1165e02119d5SChris Mason 11665d4f98a2SYan Zheng memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 11675d4f98a2SYan Zheng btrfs_set_header_bytenr(leaf, leaf->start); 11685d4f98a2SYan Zheng btrfs_set_header_generation(leaf, trans->transid); 11695d4f98a2SYan Zheng btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 11705d4f98a2SYan Zheng btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID); 11717237f183SYan Zheng root->node = leaf; 1172e02119d5SChris Mason 1173e02119d5SChris Mason write_extent_buffer(root->node, root->fs_info->fsid, 1174e02119d5SChris Mason (unsigned long)btrfs_header_fsid(root->node), 1175e02119d5SChris Mason BTRFS_FSID_SIZE); 1176e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1177e02119d5SChris Mason btrfs_tree_unlock(root->node); 11787237f183SYan Zheng return root; 11797237f183SYan Zheng } 11807237f183SYan Zheng 11817237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 11827237f183SYan Zheng struct btrfs_fs_info *fs_info) 11837237f183SYan Zheng { 11847237f183SYan Zheng struct btrfs_root *log_root; 11857237f183SYan Zheng 11867237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 11877237f183SYan Zheng if (IS_ERR(log_root)) 11887237f183SYan Zheng return PTR_ERR(log_root); 11897237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 11907237f183SYan Zheng fs_info->log_root_tree = log_root; 11917237f183SYan Zheng return 0; 11927237f183SYan Zheng } 11937237f183SYan Zheng 11947237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 11957237f183SYan Zheng struct btrfs_root *root) 11967237f183SYan Zheng { 11977237f183SYan Zheng struct btrfs_root *log_root; 11987237f183SYan Zheng struct btrfs_inode_item *inode_item; 11997237f183SYan Zheng 12007237f183SYan Zheng log_root = alloc_log_tree(trans, root->fs_info); 12017237f183SYan Zheng if (IS_ERR(log_root)) 12027237f183SYan Zheng return PTR_ERR(log_root); 12037237f183SYan Zheng 12047237f183SYan Zheng log_root->last_trans = trans->transid; 12057237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 12067237f183SYan Zheng 12077237f183SYan Zheng inode_item = &log_root->root_item.inode; 12087237f183SYan Zheng inode_item->generation = cpu_to_le64(1); 12097237f183SYan Zheng inode_item->size = cpu_to_le64(3); 12107237f183SYan Zheng inode_item->nlink = cpu_to_le32(1); 12117237f183SYan Zheng inode_item->nbytes = cpu_to_le64(root->leafsize); 12127237f183SYan Zheng inode_item->mode = cpu_to_le32(S_IFDIR | 0755); 12137237f183SYan Zheng 12145d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 12157237f183SYan Zheng 12167237f183SYan Zheng WARN_ON(root->log_root); 12177237f183SYan Zheng root->log_root = log_root; 12187237f183SYan Zheng root->log_transid = 0; 1219257c62e1SChris Mason root->last_log_commit = 0; 1220e02119d5SChris Mason return 0; 1221e02119d5SChris Mason } 1222e02119d5SChris Mason 1223e02119d5SChris Mason struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root, 1224e02119d5SChris Mason struct btrfs_key *location) 1225e02119d5SChris Mason { 1226e02119d5SChris Mason struct btrfs_root *root; 1227e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 12280f7d52f4SChris Mason struct btrfs_path *path; 12295f39d397SChris Mason struct extent_buffer *l; 123084234f3aSYan Zheng u64 generation; 1231db94535dSChris Mason u32 blocksize; 12320f7d52f4SChris Mason int ret = 0; 12330f7d52f4SChris Mason 12345eda7b5eSChris Mason root = kzalloc(sizeof(*root), GFP_NOFS); 12350cf6c620SChris Mason if (!root) 12360f7d52f4SChris Mason return ERR_PTR(-ENOMEM); 12370f7d52f4SChris Mason if (location->offset == (u64)-1) { 1238db94535dSChris Mason ret = find_and_setup_root(tree_root, fs_info, 12390f7d52f4SChris Mason location->objectid, root); 12400f7d52f4SChris Mason if (ret) { 12410f7d52f4SChris Mason kfree(root); 12420f7d52f4SChris Mason return ERR_PTR(ret); 12430f7d52f4SChris Mason } 124413a8a7c8SYan, Zheng goto out; 12450f7d52f4SChris Mason } 12460f7d52f4SChris Mason 1247db94535dSChris Mason __setup_root(tree_root->nodesize, tree_root->leafsize, 124887ee04ebSChris Mason tree_root->sectorsize, tree_root->stripesize, 124987ee04ebSChris Mason root, fs_info, location->objectid); 12500f7d52f4SChris Mason 12510f7d52f4SChris Mason path = btrfs_alloc_path(); 1252db5b493aSTsutomu Itoh if (!path) { 1253db5b493aSTsutomu Itoh kfree(root); 1254db5b493aSTsutomu Itoh return ERR_PTR(-ENOMEM); 1255db5b493aSTsutomu Itoh } 12560f7d52f4SChris Mason ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0); 125713a8a7c8SYan, Zheng if (ret == 0) { 12585f39d397SChris Mason l = path->nodes[0]; 12595f39d397SChris Mason read_extent_buffer(l, &root->root_item, 12605f39d397SChris Mason btrfs_item_ptr_offset(l, path->slots[0]), 12610f7d52f4SChris Mason sizeof(root->root_item)); 126244b36eb2SYan Zheng memcpy(&root->root_key, location, sizeof(*location)); 126313a8a7c8SYan, Zheng } 12640f7d52f4SChris Mason btrfs_free_path(path); 12650f7d52f4SChris Mason if (ret) { 12665e540f77STsutomu Itoh kfree(root); 126713a8a7c8SYan, Zheng if (ret > 0) 126813a8a7c8SYan, Zheng ret = -ENOENT; 12690f7d52f4SChris Mason return ERR_PTR(ret); 12700f7d52f4SChris Mason } 127113a8a7c8SYan, Zheng 127284234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1273db94535dSChris Mason blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item)); 1274db94535dSChris Mason root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item), 127584234f3aSYan Zheng blocksize, generation); 12765d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 12770f7d52f4SChris Mason BUG_ON(!root->node); 127813a8a7c8SYan, Zheng out: 127913a8a7c8SYan, Zheng if (location->objectid != BTRFS_TREE_LOG_OBJECTID) 12800f7d52f4SChris Mason root->ref_cows = 1; 128113a8a7c8SYan, Zheng 12825eda7b5eSChris Mason return root; 12835eda7b5eSChris Mason } 12845eda7b5eSChris Mason 1285dc17ff8fSChris Mason struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1286dc17ff8fSChris Mason u64 root_objectid) 1287dc17ff8fSChris Mason { 1288dc17ff8fSChris Mason struct btrfs_root *root; 1289dc17ff8fSChris Mason 1290dc17ff8fSChris Mason if (root_objectid == BTRFS_ROOT_TREE_OBJECTID) 1291dc17ff8fSChris Mason return fs_info->tree_root; 1292dc17ff8fSChris Mason if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID) 1293dc17ff8fSChris Mason return fs_info->extent_root; 1294dc17ff8fSChris Mason 1295dc17ff8fSChris Mason root = radix_tree_lookup(&fs_info->fs_roots_radix, 1296dc17ff8fSChris Mason (unsigned long)root_objectid); 1297dc17ff8fSChris Mason return root; 1298dc17ff8fSChris Mason } 1299dc17ff8fSChris Mason 1300edbd8d4eSChris Mason struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info, 1301edbd8d4eSChris Mason struct btrfs_key *location) 13025eda7b5eSChris Mason { 13035eda7b5eSChris Mason struct btrfs_root *root; 13045eda7b5eSChris Mason int ret; 13055eda7b5eSChris Mason 1306edbd8d4eSChris Mason if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 1307edbd8d4eSChris Mason return fs_info->tree_root; 1308edbd8d4eSChris Mason if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID) 1309edbd8d4eSChris Mason return fs_info->extent_root; 13108f18cf13SChris Mason if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID) 13118f18cf13SChris Mason return fs_info->chunk_root; 13128f18cf13SChris Mason if (location->objectid == BTRFS_DEV_TREE_OBJECTID) 13138f18cf13SChris Mason return fs_info->dev_root; 13140403e47eSYan Zheng if (location->objectid == BTRFS_CSUM_TREE_OBJECTID) 13150403e47eSYan Zheng return fs_info->csum_root; 13164df27c4dSYan, Zheng again: 13174df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 13185eda7b5eSChris Mason root = radix_tree_lookup(&fs_info->fs_roots_radix, 13195eda7b5eSChris Mason (unsigned long)location->objectid); 13204df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 13215eda7b5eSChris Mason if (root) 13225eda7b5eSChris Mason return root; 13235eda7b5eSChris Mason 1324e02119d5SChris Mason root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location); 13255eda7b5eSChris Mason if (IS_ERR(root)) 13265eda7b5eSChris Mason return root; 13273394e160SChris Mason 13283394e160SChris Mason set_anon_super(&root->anon_super, NULL); 13293394e160SChris Mason 1330d68fc57bSYan, Zheng if (btrfs_root_refs(&root->root_item) == 0) { 1331d68fc57bSYan, Zheng ret = -ENOENT; 1332d68fc57bSYan, Zheng goto fail; 1333d68fc57bSYan, Zheng } 1334d68fc57bSYan, Zheng 1335d68fc57bSYan, Zheng ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid); 1336d68fc57bSYan, Zheng if (ret < 0) 1337d68fc57bSYan, Zheng goto fail; 1338d68fc57bSYan, Zheng if (ret == 0) 1339d68fc57bSYan, Zheng root->orphan_item_inserted = 1; 1340d68fc57bSYan, Zheng 13414df27c4dSYan, Zheng ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); 13424df27c4dSYan, Zheng if (ret) 13434df27c4dSYan, Zheng goto fail; 13444df27c4dSYan, Zheng 13454df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 13462619ba1fSChris Mason ret = radix_tree_insert(&fs_info->fs_roots_radix, 13472619ba1fSChris Mason (unsigned long)root->root_key.objectid, 13480f7d52f4SChris Mason root); 1349d68fc57bSYan, Zheng if (ret == 0) 13504df27c4dSYan, Zheng root->in_radix = 1; 1351d68fc57bSYan, Zheng 13524df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 13534df27c4dSYan, Zheng radix_tree_preload_end(); 13540f7d52f4SChris Mason if (ret) { 13554df27c4dSYan, Zheng if (ret == -EEXIST) { 13564df27c4dSYan, Zheng free_fs_root(root); 13574df27c4dSYan, Zheng goto again; 13580f7d52f4SChris Mason } 13594df27c4dSYan, Zheng goto fail; 13604df27c4dSYan, Zheng } 13614df27c4dSYan, Zheng 1362edbd8d4eSChris Mason ret = btrfs_find_dead_roots(fs_info->tree_root, 13635d4f98a2SYan Zheng root->root_key.objectid); 13644df27c4dSYan, Zheng WARN_ON(ret); 1365edbd8d4eSChris Mason return root; 13664df27c4dSYan, Zheng fail: 13674df27c4dSYan, Zheng free_fs_root(root); 13684df27c4dSYan, Zheng return ERR_PTR(ret); 1369edbd8d4eSChris Mason } 1370edbd8d4eSChris Mason 1371edbd8d4eSChris Mason struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info, 1372edbd8d4eSChris Mason struct btrfs_key *location, 1373edbd8d4eSChris Mason const char *name, int namelen) 1374edbd8d4eSChris Mason { 13754df27c4dSYan, Zheng return btrfs_read_fs_root_no_name(fs_info, location); 13764df27c4dSYan, Zheng #if 0 1377edbd8d4eSChris Mason struct btrfs_root *root; 1378edbd8d4eSChris Mason int ret; 1379edbd8d4eSChris Mason 1380edbd8d4eSChris Mason root = btrfs_read_fs_root_no_name(fs_info, location); 1381edbd8d4eSChris Mason if (!root) 1382edbd8d4eSChris Mason return NULL; 138358176a96SJosef Bacik 13844313b399SChris Mason if (root->in_sysfs) 13854313b399SChris Mason return root; 13864313b399SChris Mason 138758176a96SJosef Bacik ret = btrfs_set_root_name(root, name, namelen); 138858176a96SJosef Bacik if (ret) { 13895f39d397SChris Mason free_extent_buffer(root->node); 139058176a96SJosef Bacik kfree(root); 139158176a96SJosef Bacik return ERR_PTR(ret); 139258176a96SJosef Bacik } 13934df27c4dSYan, Zheng 139458176a96SJosef Bacik ret = btrfs_sysfs_add_root(root); 139558176a96SJosef Bacik if (ret) { 13965f39d397SChris Mason free_extent_buffer(root->node); 139758176a96SJosef Bacik kfree(root->name); 139858176a96SJosef Bacik kfree(root); 139958176a96SJosef Bacik return ERR_PTR(ret); 140058176a96SJosef Bacik } 14014313b399SChris Mason root->in_sysfs = 1; 14020f7d52f4SChris Mason return root; 14034df27c4dSYan, Zheng #endif 14040f7d52f4SChris Mason } 140504160088SChris Mason 140604160088SChris Mason static int btrfs_congested_fn(void *congested_data, int bdi_bits) 140704160088SChris Mason { 140804160088SChris Mason struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; 140904160088SChris Mason int ret = 0; 141004160088SChris Mason struct btrfs_device *device; 141104160088SChris Mason struct backing_dev_info *bdi; 1412b7967db7SChris Mason 1413c6e30871SQinghuang Feng list_for_each_entry(device, &info->fs_devices->devices, dev_list) { 1414dfe25020SChris Mason if (!device->bdev) 1415dfe25020SChris Mason continue; 141604160088SChris Mason bdi = blk_get_backing_dev_info(device->bdev); 141704160088SChris Mason if (bdi && bdi_congested(bdi, bdi_bits)) { 141804160088SChris Mason ret = 1; 141904160088SChris Mason break; 142004160088SChris Mason } 142104160088SChris Mason } 142204160088SChris Mason return ret; 142304160088SChris Mason } 142404160088SChris Mason 142538b66988SChris Mason /* 142638b66988SChris Mason * this unplugs every device on the box, and it is only used when page 142738b66988SChris Mason * is null 142838b66988SChris Mason */ 142938b66988SChris Mason static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page) 143038b66988SChris Mason { 143138b66988SChris Mason struct btrfs_device *device; 143238b66988SChris Mason struct btrfs_fs_info *info; 143338b66988SChris Mason 143438b66988SChris Mason info = (struct btrfs_fs_info *)bdi->unplug_io_data; 1435c6e30871SQinghuang Feng list_for_each_entry(device, &info->fs_devices->devices, dev_list) { 1436d20f7043SChris Mason if (!device->bdev) 1437d20f7043SChris Mason continue; 1438d20f7043SChris Mason 143938b66988SChris Mason bdi = blk_get_backing_dev_info(device->bdev); 1440d397712bSChris Mason if (bdi->unplug_io_fn) 144138b66988SChris Mason bdi->unplug_io_fn(bdi, page); 144238b66988SChris Mason } 144338b66988SChris Mason } 144438b66988SChris Mason 1445b2950863SChristoph Hellwig static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page) 144604160088SChris Mason { 144738b66988SChris Mason struct inode *inode; 1448f2d8d74dSChris Mason struct extent_map_tree *em_tree; 1449f2d8d74dSChris Mason struct extent_map *em; 1450bcbfce8aSChris Mason struct address_space *mapping; 145138b66988SChris Mason u64 offset; 145238b66988SChris Mason 1453bcbfce8aSChris Mason /* the generic O_DIRECT read code does this */ 14549f0ba5bdSChris Mason if (1 || !page) { 145538b66988SChris Mason __unplug_io_fn(bdi, page); 145638b66988SChris Mason return; 145738b66988SChris Mason } 145838b66988SChris Mason 1459bcbfce8aSChris Mason /* 1460bcbfce8aSChris Mason * page->mapping may change at any time. Get a consistent copy 1461bcbfce8aSChris Mason * and use that for everything below 1462bcbfce8aSChris Mason */ 1463bcbfce8aSChris Mason smp_mb(); 1464bcbfce8aSChris Mason mapping = page->mapping; 1465bcbfce8aSChris Mason if (!mapping) 1466bcbfce8aSChris Mason return; 1467bcbfce8aSChris Mason 1468bcbfce8aSChris Mason inode = mapping->host; 1469240d5d48SChris Mason 1470240d5d48SChris Mason /* 1471240d5d48SChris Mason * don't do the expensive searching for a small number of 1472240d5d48SChris Mason * devices 1473240d5d48SChris Mason */ 1474240d5d48SChris Mason if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) { 1475240d5d48SChris Mason __unplug_io_fn(bdi, page); 1476240d5d48SChris Mason return; 1477240d5d48SChris Mason } 1478240d5d48SChris Mason 147938b66988SChris Mason offset = page_offset(page); 148004160088SChris Mason 1481f2d8d74dSChris Mason em_tree = &BTRFS_I(inode)->extent_tree; 1482890871beSChris Mason read_lock(&em_tree->lock); 1483f2d8d74dSChris Mason em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE); 1484890871beSChris Mason read_unlock(&em_tree->lock); 148589642229SChris Mason if (!em) { 148689642229SChris Mason __unplug_io_fn(bdi, page); 1487f2d8d74dSChris Mason return; 148889642229SChris Mason } 1489f2d8d74dSChris Mason 149089642229SChris Mason if (em->block_start >= EXTENT_MAP_LAST_BYTE) { 149189642229SChris Mason free_extent_map(em); 149289642229SChris Mason __unplug_io_fn(bdi, page); 149389642229SChris Mason return; 149489642229SChris Mason } 1495f2d8d74dSChris Mason offset = offset - em->start; 1496f2d8d74dSChris Mason btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree, 1497f2d8d74dSChris Mason em->block_start + offset, page); 1498f2d8d74dSChris Mason free_extent_map(em); 149904160088SChris Mason } 150004160088SChris Mason 1501ad081f14SJens Axboe /* 1502ad081f14SJens Axboe * If this fails, caller must call bdi_destroy() to get rid of the 1503ad081f14SJens Axboe * bdi again. 1504ad081f14SJens Axboe */ 150504160088SChris Mason static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi) 150604160088SChris Mason { 1507ad081f14SJens Axboe int err; 1508ad081f14SJens Axboe 1509ad081f14SJens Axboe bdi->capabilities = BDI_CAP_MAP_COPY; 1510e6d086d8SJens Axboe err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY); 1511ad081f14SJens Axboe if (err) 1512ad081f14SJens Axboe return err; 1513ad081f14SJens Axboe 15144575c9ccSChris Mason bdi->ra_pages = default_backing_dev_info.ra_pages; 151504160088SChris Mason bdi->unplug_io_fn = btrfs_unplug_io_fn; 151604160088SChris Mason bdi->unplug_io_data = info; 151704160088SChris Mason bdi->congested_fn = btrfs_congested_fn; 151804160088SChris Mason bdi->congested_data = info; 151904160088SChris Mason return 0; 152004160088SChris Mason } 152104160088SChris Mason 1522ce9adaa5SChris Mason static int bio_ready_for_csum(struct bio *bio) 1523ce9adaa5SChris Mason { 1524ce9adaa5SChris Mason u64 length = 0; 1525ce9adaa5SChris Mason u64 buf_len = 0; 1526ce9adaa5SChris Mason u64 start = 0; 1527ce9adaa5SChris Mason struct page *page; 1528ce9adaa5SChris Mason struct extent_io_tree *io_tree = NULL; 1529ce9adaa5SChris Mason struct bio_vec *bvec; 1530ce9adaa5SChris Mason int i; 1531ce9adaa5SChris Mason int ret; 1532ce9adaa5SChris Mason 1533ce9adaa5SChris Mason bio_for_each_segment(bvec, bio, i) { 1534ce9adaa5SChris Mason page = bvec->bv_page; 1535ce9adaa5SChris Mason if (page->private == EXTENT_PAGE_PRIVATE) { 1536ce9adaa5SChris Mason length += bvec->bv_len; 1537ce9adaa5SChris Mason continue; 1538ce9adaa5SChris Mason } 1539ce9adaa5SChris Mason if (!page->private) { 1540ce9adaa5SChris Mason length += bvec->bv_len; 1541ce9adaa5SChris Mason continue; 1542ce9adaa5SChris Mason } 1543ce9adaa5SChris Mason length = bvec->bv_len; 1544ce9adaa5SChris Mason buf_len = page->private >> 2; 1545ce9adaa5SChris Mason start = page_offset(page) + bvec->bv_offset; 1546ce9adaa5SChris Mason io_tree = &BTRFS_I(page->mapping->host)->io_tree; 1547ce9adaa5SChris Mason } 1548ce9adaa5SChris Mason /* are we fully contained in this bio? */ 1549ce9adaa5SChris Mason if (buf_len <= length) 1550ce9adaa5SChris Mason return 1; 1551ce9adaa5SChris Mason 1552ce9adaa5SChris Mason ret = extent_range_uptodate(io_tree, start + length, 1553ce9adaa5SChris Mason start + buf_len - 1); 1554ce9adaa5SChris Mason return ret; 1555ce9adaa5SChris Mason } 1556ce9adaa5SChris Mason 15578b712842SChris Mason /* 15588b712842SChris Mason * called by the kthread helper functions to finally call the bio end_io 15598b712842SChris Mason * functions. This is where read checksum verification actually happens 15608b712842SChris Mason */ 15618b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work) 1562ce9adaa5SChris Mason { 1563ce9adaa5SChris Mason struct bio *bio; 15648b712842SChris Mason struct end_io_wq *end_io_wq; 15658b712842SChris Mason struct btrfs_fs_info *fs_info; 1566ce9adaa5SChris Mason int error; 1567ce9adaa5SChris Mason 15688b712842SChris Mason end_io_wq = container_of(work, struct end_io_wq, work); 1569ce9adaa5SChris Mason bio = end_io_wq->bio; 15708b712842SChris Mason fs_info = end_io_wq->info; 15718b712842SChris Mason 1572cad321adSChris Mason /* metadata bio reads are special because the whole tree block must 15738b712842SChris Mason * be checksummed at once. This makes sure the entire block is in 15748b712842SChris Mason * ram and up to date before trying to verify things. For 15758b712842SChris Mason * blocksize <= pagesize, it is basically a noop 15768b712842SChris Mason */ 15777b6d91daSChristoph Hellwig if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata && 1578cad321adSChris Mason !bio_ready_for_csum(bio)) { 1579d20f7043SChris Mason btrfs_queue_worker(&fs_info->endio_meta_workers, 15808b712842SChris Mason &end_io_wq->work); 1581ce9adaa5SChris Mason return; 1582ce9adaa5SChris Mason } 1583ce9adaa5SChris Mason error = end_io_wq->error; 1584ce9adaa5SChris Mason bio->bi_private = end_io_wq->private; 1585ce9adaa5SChris Mason bio->bi_end_io = end_io_wq->end_io; 1586ce9adaa5SChris Mason kfree(end_io_wq); 1587ce9adaa5SChris Mason bio_endio(bio, error); 1588ce9adaa5SChris Mason } 158944b8bd7eSChris Mason 1590a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1591a74a4b97SChris Mason { 1592a74a4b97SChris Mason struct btrfs_root *root = arg; 1593a74a4b97SChris Mason 1594a74a4b97SChris Mason do { 1595a74a4b97SChris Mason vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE); 159676dda93cSYan, Zheng 159776dda93cSYan, Zheng if (!(root->fs_info->sb->s_flags & MS_RDONLY) && 159876dda93cSYan, Zheng mutex_trylock(&root->fs_info->cleaner_mutex)) { 159924bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 1600a74a4b97SChris Mason btrfs_clean_old_snapshots(root); 1601a74a4b97SChris Mason mutex_unlock(&root->fs_info->cleaner_mutex); 160276dda93cSYan, Zheng } 1603a74a4b97SChris Mason 1604a74a4b97SChris Mason if (freezing(current)) { 1605a74a4b97SChris Mason refrigerator(); 1606a74a4b97SChris Mason } else { 1607a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 16088929ecfaSYan, Zheng if (!kthread_should_stop()) 1609a74a4b97SChris Mason schedule(); 1610a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1611a74a4b97SChris Mason } 1612a74a4b97SChris Mason } while (!kthread_should_stop()); 1613a74a4b97SChris Mason return 0; 1614a74a4b97SChris Mason } 1615a74a4b97SChris Mason 1616a74a4b97SChris Mason static int transaction_kthread(void *arg) 1617a74a4b97SChris Mason { 1618a74a4b97SChris Mason struct btrfs_root *root = arg; 1619a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1620a74a4b97SChris Mason struct btrfs_transaction *cur; 16218929ecfaSYan, Zheng u64 transid; 1622a74a4b97SChris Mason unsigned long now; 1623a74a4b97SChris Mason unsigned long delay; 1624a74a4b97SChris Mason int ret; 1625a74a4b97SChris Mason 1626a74a4b97SChris Mason do { 1627a74a4b97SChris Mason delay = HZ * 30; 1628a74a4b97SChris Mason vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE); 1629a74a4b97SChris Mason mutex_lock(&root->fs_info->transaction_kthread_mutex); 1630a74a4b97SChris Mason 16318929ecfaSYan, Zheng spin_lock(&root->fs_info->new_trans_lock); 1632a74a4b97SChris Mason cur = root->fs_info->running_transaction; 1633a74a4b97SChris Mason if (!cur) { 16348929ecfaSYan, Zheng spin_unlock(&root->fs_info->new_trans_lock); 1635a74a4b97SChris Mason goto sleep; 1636a74a4b97SChris Mason } 163731153d81SYan Zheng 1638a74a4b97SChris Mason now = get_seconds(); 16398929ecfaSYan, Zheng if (!cur->blocked && 16408929ecfaSYan, Zheng (now < cur->start_time || now - cur->start_time < 30)) { 16418929ecfaSYan, Zheng spin_unlock(&root->fs_info->new_trans_lock); 1642a74a4b97SChris Mason delay = HZ * 5; 1643a74a4b97SChris Mason goto sleep; 1644a74a4b97SChris Mason } 16458929ecfaSYan, Zheng transid = cur->transid; 16468929ecfaSYan, Zheng spin_unlock(&root->fs_info->new_trans_lock); 164756bec294SChris Mason 16488929ecfaSYan, Zheng trans = btrfs_join_transaction(root, 1); 16493612b495STsutomu Itoh BUG_ON(IS_ERR(trans)); 16508929ecfaSYan, Zheng if (transid == trans->transid) { 16518929ecfaSYan, Zheng ret = btrfs_commit_transaction(trans, root); 16528929ecfaSYan, Zheng BUG_ON(ret); 16538929ecfaSYan, Zheng } else { 16548929ecfaSYan, Zheng btrfs_end_transaction(trans, root); 16558929ecfaSYan, Zheng } 1656a74a4b97SChris Mason sleep: 1657a74a4b97SChris Mason wake_up_process(root->fs_info->cleaner_kthread); 1658a74a4b97SChris Mason mutex_unlock(&root->fs_info->transaction_kthread_mutex); 1659a74a4b97SChris Mason 1660a74a4b97SChris Mason if (freezing(current)) { 1661a74a4b97SChris Mason refrigerator(); 1662a74a4b97SChris Mason } else { 1663a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 16648929ecfaSYan, Zheng if (!kthread_should_stop() && 16658929ecfaSYan, Zheng !btrfs_transaction_blocked(root->fs_info)) 1666a74a4b97SChris Mason schedule_timeout(delay); 1667a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1668a74a4b97SChris Mason } 1669a74a4b97SChris Mason } while (!kthread_should_stop()); 1670a74a4b97SChris Mason return 0; 1671a74a4b97SChris Mason } 1672a74a4b97SChris Mason 16738a4b83ccSChris Mason struct btrfs_root *open_ctree(struct super_block *sb, 1674dfe25020SChris Mason struct btrfs_fs_devices *fs_devices, 1675dfe25020SChris Mason char *options) 1676eb60ceacSChris Mason { 1677db94535dSChris Mason u32 sectorsize; 1678db94535dSChris Mason u32 nodesize; 1679db94535dSChris Mason u32 leafsize; 1680db94535dSChris Mason u32 blocksize; 168187ee04ebSChris Mason u32 stripesize; 168284234f3aSYan Zheng u64 generation; 1683f2b636e8SJosef Bacik u64 features; 16843de4586cSChris Mason struct btrfs_key location; 1685a061fc8dSChris Mason struct buffer_head *bh; 1686e02119d5SChris Mason struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root), 1687e20d96d6SChris Mason GFP_NOFS); 1688d20f7043SChris Mason struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root), 1689d20f7043SChris Mason GFP_NOFS); 1690450ba0eaSJosef Bacik struct btrfs_root *tree_root = btrfs_sb(sb); 1691450ba0eaSJosef Bacik struct btrfs_fs_info *fs_info = tree_root->fs_info; 1692e02119d5SChris Mason struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root), 16930b86a832SChris Mason GFP_NOFS); 1694e02119d5SChris Mason struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root), 16950b86a832SChris Mason GFP_NOFS); 1696e02119d5SChris Mason struct btrfs_root *log_tree_root; 1697e02119d5SChris Mason 1698eb60ceacSChris Mason int ret; 1699e58ca020SYan int err = -EINVAL; 17004543df7eSChris Mason 17012c90e5d6SChris Mason struct btrfs_super_block *disk_super; 17028790d502SChris Mason 17030463bb4eSJim Meyering if (!extent_root || !tree_root || !fs_info || 1704d20f7043SChris Mason !chunk_root || !dev_root || !csum_root) { 170539279cc3SChris Mason err = -ENOMEM; 170639279cc3SChris Mason goto fail; 170739279cc3SChris Mason } 170876dda93cSYan, Zheng 170976dda93cSYan, Zheng ret = init_srcu_struct(&fs_info->subvol_srcu); 171076dda93cSYan, Zheng if (ret) { 171176dda93cSYan, Zheng err = ret; 171276dda93cSYan, Zheng goto fail; 171376dda93cSYan, Zheng } 171476dda93cSYan, Zheng 171576dda93cSYan, Zheng ret = setup_bdi(fs_info, &fs_info->bdi); 171676dda93cSYan, Zheng if (ret) { 171776dda93cSYan, Zheng err = ret; 171876dda93cSYan, Zheng goto fail_srcu; 171976dda93cSYan, Zheng } 172076dda93cSYan, Zheng 172176dda93cSYan, Zheng fs_info->btree_inode = new_inode(sb); 172276dda93cSYan, Zheng if (!fs_info->btree_inode) { 172376dda93cSYan, Zheng err = -ENOMEM; 172476dda93cSYan, Zheng goto fail_bdi; 172576dda93cSYan, Zheng } 172676dda93cSYan, Zheng 172776dda93cSYan, Zheng INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 17288fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 1729facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 173024bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 173119c00ddcSChris Mason INIT_LIST_HEAD(&fs_info->hashers); 1732ea8c2819SChris Mason INIT_LIST_HEAD(&fs_info->delalloc_inodes); 17335a3f23d5SChris Mason INIT_LIST_HEAD(&fs_info->ordered_operations); 173411833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 17351832a6d5SChris Mason spin_lock_init(&fs_info->delalloc_lock); 1736cee36a03SChris Mason spin_lock_init(&fs_info->new_trans_lock); 173731153d81SYan Zheng spin_lock_init(&fs_info->ref_cache_lock); 173876dda93cSYan, Zheng spin_lock_init(&fs_info->fs_roots_radix_lock); 173924bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 174019c00ddcSChris Mason 174158176a96SJosef Bacik init_completion(&fs_info->kobj_unregister); 17429f5fae2fSChris Mason fs_info->tree_root = tree_root; 17439f5fae2fSChris Mason fs_info->extent_root = extent_root; 1744d20f7043SChris Mason fs_info->csum_root = csum_root; 17450b86a832SChris Mason fs_info->chunk_root = chunk_root; 17460b86a832SChris Mason fs_info->dev_root = dev_root; 17478a4b83ccSChris Mason fs_info->fs_devices = fs_devices; 17480b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 17496324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 17500b86a832SChris Mason btrfs_mapping_init(&fs_info->mapping_tree); 1751f0486c68SYan, Zheng btrfs_init_block_rsv(&fs_info->global_block_rsv); 1752f0486c68SYan, Zheng btrfs_init_block_rsv(&fs_info->delalloc_block_rsv); 1753f0486c68SYan, Zheng btrfs_init_block_rsv(&fs_info->trans_block_rsv); 1754f0486c68SYan, Zheng btrfs_init_block_rsv(&fs_info->chunk_block_rsv); 1755f0486c68SYan, Zheng btrfs_init_block_rsv(&fs_info->empty_block_rsv); 1756f0486c68SYan, Zheng INIT_LIST_HEAD(&fs_info->durable_block_rsv_list); 1757f0486c68SYan, Zheng mutex_init(&fs_info->durable_block_rsv_mutex); 1758cb03c743SChris Mason atomic_set(&fs_info->nr_async_submits, 0); 1759771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 17608c8bee1dSChris Mason atomic_set(&fs_info->async_submit_draining, 0); 17610986fe9eSChris Mason atomic_set(&fs_info->nr_async_bios, 0); 1762e20d96d6SChris Mason fs_info->sb = sb; 17636f568d35SChris Mason fs_info->max_inline = 8192 * 1024; 17649ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 1765c8b97818SChris Mason 1766b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 1767b34b086cSChris Mason num_online_cpus() + 2, 8); 17680afbaf8cSChris Mason 17693eaa2885SChris Mason INIT_LIST_HEAD(&fs_info->ordered_extents); 17703eaa2885SChris Mason spin_lock_init(&fs_info->ordered_extent_lock); 17713eaa2885SChris Mason 1772a061fc8dSChris Mason sb->s_blocksize = 4096; 1773a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(4096); 177432a88aa1SJens Axboe sb->s_bdi = &fs_info->bdi; 1775a061fc8dSChris Mason 177676dda93cSYan, Zheng fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 177776dda93cSYan, Zheng fs_info->btree_inode->i_nlink = 1; 17780afbaf8cSChris Mason /* 17790afbaf8cSChris Mason * we set the i_size on the btree inode to the max possible int. 17800afbaf8cSChris Mason * the real end of the address space is determined by all of 17810afbaf8cSChris Mason * the devices in the system 17820afbaf8cSChris Mason */ 17830afbaf8cSChris Mason fs_info->btree_inode->i_size = OFFSET_MAX; 1784d98237b3SChris Mason fs_info->btree_inode->i_mapping->a_ops = &btree_aops; 178504160088SChris Mason fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi; 178604160088SChris Mason 17875d4f98a2SYan Zheng RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node); 1788d1310b2eSChris Mason extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree, 17895f39d397SChris Mason fs_info->btree_inode->i_mapping, 17905f39d397SChris Mason GFP_NOFS); 1791d1310b2eSChris Mason extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree, 1792d1310b2eSChris Mason GFP_NOFS); 17930da5468fSChris Mason 1794d1310b2eSChris Mason BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops; 1795d1310b2eSChris Mason 17960f7d52f4SChris Mason BTRFS_I(fs_info->btree_inode)->root = tree_root; 17970f7d52f4SChris Mason memset(&BTRFS_I(fs_info->btree_inode)->location, 0, 17980f7d52f4SChris Mason sizeof(struct btrfs_key)); 179976dda93cSYan, Zheng BTRFS_I(fs_info->btree_inode)->dummy_inode = 1; 180022b0ebdaSChris Mason insert_inode_hash(fs_info->btree_inode); 180139279cc3SChris Mason 1802e02119d5SChris Mason spin_lock_init(&fs_info->block_group_cache_lock); 18036bef4d31SEric Paris fs_info->block_group_cache_tree = RB_ROOT; 1804925baeddSChris Mason 180511833d66SYan Zheng extent_io_tree_init(&fs_info->freed_extents[0], 18067d9eb12cSChris Mason fs_info->btree_inode->i_mapping, GFP_NOFS); 180711833d66SYan Zheng extent_io_tree_init(&fs_info->freed_extents[1], 180811833d66SYan Zheng fs_info->btree_inode->i_mapping, GFP_NOFS); 180911833d66SYan Zheng fs_info->pinned_extents = &fs_info->freed_extents[0]; 1810e6dcd2dcSChris Mason fs_info->do_barriers = 1; 1811f9295749SChris Mason 1812d98237b3SChris Mason 1813509659cdSChris Mason mutex_init(&fs_info->trans_mutex); 18145a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 181530ae8467SChris Mason mutex_init(&fs_info->tree_log_mutex); 1816925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 1817925baeddSChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 1818a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 1819a74a4b97SChris Mason mutex_init(&fs_info->volume_mutex); 1820276e680dSYan Zheng init_rwsem(&fs_info->extent_commit_sem); 1821c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 182276dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 1823fa9c0d79SChris Mason 1824fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 1825fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 1826fa9c0d79SChris Mason 18273768f368SChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 182819c00ddcSChris Mason init_waitqueue_head(&fs_info->transaction_wait); 1829bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 18304854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 18319a8dd150SChris Mason 18320b86a832SChris Mason __setup_root(4096, 4096, 4096, 4096, tree_root, 18332c90e5d6SChris Mason fs_info, BTRFS_ROOT_TREE_OBJECTID); 18347eccb903SChris Mason 1835a512bbf8SYan Zheng bh = btrfs_read_dev_super(fs_devices->latest_bdev); 183620b45077SDave Young if (!bh) { 183720b45077SDave Young err = -EINVAL; 183839279cc3SChris Mason goto fail_iput; 183920b45077SDave Young } 184039279cc3SChris Mason 1841a061fc8dSChris Mason memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy)); 18422d69a0f8SYan Zheng memcpy(&fs_info->super_for_commit, &fs_info->super_copy, 18432d69a0f8SYan Zheng sizeof(fs_info->super_for_commit)); 1844a061fc8dSChris Mason brelse(bh); 18455f39d397SChris Mason 1846a061fc8dSChris Mason memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE); 18470b86a832SChris Mason 18485f39d397SChris Mason disk_super = &fs_info->super_copy; 18490f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 1850c6e2bac1SJosef Bacik goto fail_iput; 18510f7d52f4SChris Mason 1852acce952bSliubo /* check FS state, whether FS is broken. */ 1853acce952bSliubo fs_info->fs_state |= btrfs_super_flags(disk_super); 1854acce952bSliubo 1855acce952bSliubo btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY); 1856acce952bSliubo 1857*75e7cb7fSLiu Bo /* 1858*75e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 1859*75e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 1860*75e7cb7fSLiu Bo */ 1861*75e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 1862*75e7cb7fSLiu Bo 18632b82032cSYan Zheng ret = btrfs_parse_options(tree_root, options); 18642b82032cSYan Zheng if (ret) { 18652b82032cSYan Zheng err = ret; 1866c6e2bac1SJosef Bacik goto fail_iput; 18672b82032cSYan Zheng } 1868dfe25020SChris Mason 1869f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 1870f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 1871f2b636e8SJosef Bacik if (features) { 1872f2b636e8SJosef Bacik printk(KERN_ERR "BTRFS: couldn't mount because of " 1873f2b636e8SJosef Bacik "unsupported optional features (%Lx).\n", 187421380931SJoel Becker (unsigned long long)features); 1875f2b636e8SJosef Bacik err = -EINVAL; 1876c6e2bac1SJosef Bacik goto fail_iput; 1877f2b636e8SJosef Bacik } 1878f2b636e8SJosef Bacik 18795d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 18805d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 1881a6fa6faeSLi Zefan if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO) 1882a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 18835d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 18845d4f98a2SYan Zheng 1885f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 1886f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 1887f2b636e8SJosef Bacik if (!(sb->s_flags & MS_RDONLY) && features) { 1888f2b636e8SJosef Bacik printk(KERN_ERR "BTRFS: couldn't mount RDWR because of " 1889f2b636e8SJosef Bacik "unsupported option features (%Lx).\n", 189021380931SJoel Becker (unsigned long long)features); 1891f2b636e8SJosef Bacik err = -EINVAL; 1892c6e2bac1SJosef Bacik goto fail_iput; 1893f2b636e8SJosef Bacik } 189461d92c32SChris Mason 189561d92c32SChris Mason btrfs_init_workers(&fs_info->generic_worker, 189661d92c32SChris Mason "genwork", 1, NULL); 189761d92c32SChris Mason 18985443be45SChris Mason btrfs_init_workers(&fs_info->workers, "worker", 189961d92c32SChris Mason fs_info->thread_pool_size, 190061d92c32SChris Mason &fs_info->generic_worker); 1901c8b97818SChris Mason 1902771ed689SChris Mason btrfs_init_workers(&fs_info->delalloc_workers, "delalloc", 190361d92c32SChris Mason fs_info->thread_pool_size, 190461d92c32SChris Mason &fs_info->generic_worker); 1905771ed689SChris Mason 19065443be45SChris Mason btrfs_init_workers(&fs_info->submit_workers, "submit", 1907b720d209SChris Mason min_t(u64, fs_devices->num_devices, 190861d92c32SChris Mason fs_info->thread_pool_size), 190961d92c32SChris Mason &fs_info->generic_worker); 191061b49440SChris Mason 191161b49440SChris Mason /* a higher idle thresh on the submit workers makes it much more 191261b49440SChris Mason * likely that bios will be send down in a sane order to the 191361b49440SChris Mason * devices 191461b49440SChris Mason */ 191561b49440SChris Mason fs_info->submit_workers.idle_thresh = 64; 191653863232SChris Mason 1917771ed689SChris Mason fs_info->workers.idle_thresh = 16; 19184a69a410SChris Mason fs_info->workers.ordered = 1; 191961b49440SChris Mason 1920771ed689SChris Mason fs_info->delalloc_workers.idle_thresh = 2; 1921771ed689SChris Mason fs_info->delalloc_workers.ordered = 1; 1922771ed689SChris Mason 192361d92c32SChris Mason btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1, 192461d92c32SChris Mason &fs_info->generic_worker); 19255443be45SChris Mason btrfs_init_workers(&fs_info->endio_workers, "endio", 192661d92c32SChris Mason fs_info->thread_pool_size, 192761d92c32SChris Mason &fs_info->generic_worker); 1928d20f7043SChris Mason btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta", 192961d92c32SChris Mason fs_info->thread_pool_size, 193061d92c32SChris Mason &fs_info->generic_worker); 1931cad321adSChris Mason btrfs_init_workers(&fs_info->endio_meta_write_workers, 193261d92c32SChris Mason "endio-meta-write", fs_info->thread_pool_size, 193361d92c32SChris Mason &fs_info->generic_worker); 19345443be45SChris Mason btrfs_init_workers(&fs_info->endio_write_workers, "endio-write", 193561d92c32SChris Mason fs_info->thread_pool_size, 193661d92c32SChris Mason &fs_info->generic_worker); 19370cb59c99SJosef Bacik btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write", 19380cb59c99SJosef Bacik 1, &fs_info->generic_worker); 193961b49440SChris Mason 194061b49440SChris Mason /* 194161b49440SChris Mason * endios are largely parallel and should have a very 194261b49440SChris Mason * low idle thresh 194361b49440SChris Mason */ 194461b49440SChris Mason fs_info->endio_workers.idle_thresh = 4; 1945b51912c9SChris Mason fs_info->endio_meta_workers.idle_thresh = 4; 1946b51912c9SChris Mason 19479042846bSChris Mason fs_info->endio_write_workers.idle_thresh = 2; 19489042846bSChris Mason fs_info->endio_meta_write_workers.idle_thresh = 2; 19499042846bSChris Mason 19504543df7eSChris Mason btrfs_start_workers(&fs_info->workers, 1); 195161d92c32SChris Mason btrfs_start_workers(&fs_info->generic_worker, 1); 19521cc127b5SChris Mason btrfs_start_workers(&fs_info->submit_workers, 1); 1953771ed689SChris Mason btrfs_start_workers(&fs_info->delalloc_workers, 1); 1954247e743cSChris Mason btrfs_start_workers(&fs_info->fixup_workers, 1); 19559042846bSChris Mason btrfs_start_workers(&fs_info->endio_workers, 1); 19569042846bSChris Mason btrfs_start_workers(&fs_info->endio_meta_workers, 1); 19579042846bSChris Mason btrfs_start_workers(&fs_info->endio_meta_write_workers, 1); 19589042846bSChris Mason btrfs_start_workers(&fs_info->endio_write_workers, 1); 19590cb59c99SJosef Bacik btrfs_start_workers(&fs_info->endio_freespace_worker, 1); 19604543df7eSChris Mason 19614575c9ccSChris Mason fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super); 1962c8b97818SChris Mason fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages, 1963c8b97818SChris Mason 4 * 1024 * 1024 / PAGE_CACHE_SIZE); 19644575c9ccSChris Mason 1965db94535dSChris Mason nodesize = btrfs_super_nodesize(disk_super); 1966db94535dSChris Mason leafsize = btrfs_super_leafsize(disk_super); 1967db94535dSChris Mason sectorsize = btrfs_super_sectorsize(disk_super); 196887ee04ebSChris Mason stripesize = btrfs_super_stripesize(disk_super); 1969db94535dSChris Mason tree_root->nodesize = nodesize; 1970db94535dSChris Mason tree_root->leafsize = leafsize; 1971db94535dSChris Mason tree_root->sectorsize = sectorsize; 197287ee04ebSChris Mason tree_root->stripesize = stripesize; 1973a061fc8dSChris Mason 1974a061fc8dSChris Mason sb->s_blocksize = sectorsize; 1975a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 1976db94535dSChris Mason 197739279cc3SChris Mason if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC, 197839279cc3SChris Mason sizeof(disk_super->magic))) { 1979d397712bSChris Mason printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id); 198039279cc3SChris Mason goto fail_sb_buffer; 198139279cc3SChris Mason } 198219c00ddcSChris Mason 1983925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 1984e4404d6eSYan Zheng ret = btrfs_read_sys_array(tree_root); 1985925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 198684eed90fSChris Mason if (ret) { 1987d397712bSChris Mason printk(KERN_WARNING "btrfs: failed to read the system " 1988d397712bSChris Mason "array on %s\n", sb->s_id); 19895d4f98a2SYan Zheng goto fail_sb_buffer; 199084eed90fSChris Mason } 19910b86a832SChris Mason 19920b86a832SChris Mason blocksize = btrfs_level_size(tree_root, 19930b86a832SChris Mason btrfs_super_chunk_root_level(disk_super)); 199484234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 19950b86a832SChris Mason 19960b86a832SChris Mason __setup_root(nodesize, leafsize, sectorsize, stripesize, 19970b86a832SChris Mason chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID); 19980b86a832SChris Mason 19990b86a832SChris Mason chunk_root->node = read_tree_block(chunk_root, 20000b86a832SChris Mason btrfs_super_chunk_root(disk_super), 200184234f3aSYan Zheng blocksize, generation); 20020b86a832SChris Mason BUG_ON(!chunk_root->node); 200383121942SDavid Woodhouse if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) { 200483121942SDavid Woodhouse printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n", 200583121942SDavid Woodhouse sb->s_id); 200683121942SDavid Woodhouse goto fail_chunk_root; 200783121942SDavid Woodhouse } 20085d4f98a2SYan Zheng btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 20095d4f98a2SYan Zheng chunk_root->commit_root = btrfs_root_node(chunk_root); 20100b86a832SChris Mason 2011e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 2012e17cade2SChris Mason (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node), 2013e17cade2SChris Mason BTRFS_UUID_SIZE); 2014e17cade2SChris Mason 2015925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 20160b86a832SChris Mason ret = btrfs_read_chunk_tree(chunk_root); 2017925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 20182b82032cSYan Zheng if (ret) { 2019d397712bSChris Mason printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n", 2020d397712bSChris Mason sb->s_id); 20212b82032cSYan Zheng goto fail_chunk_root; 20222b82032cSYan Zheng } 20230b86a832SChris Mason 2024dfe25020SChris Mason btrfs_close_extra_devices(fs_devices); 2025dfe25020SChris Mason 2026db94535dSChris Mason blocksize = btrfs_level_size(tree_root, 2027db94535dSChris Mason btrfs_super_root_level(disk_super)); 202884234f3aSYan Zheng generation = btrfs_super_generation(disk_super); 20290b86a832SChris Mason 2030e20d96d6SChris Mason tree_root->node = read_tree_block(tree_root, 2031db94535dSChris Mason btrfs_super_root(disk_super), 203284234f3aSYan Zheng blocksize, generation); 203339279cc3SChris Mason if (!tree_root->node) 20342b82032cSYan Zheng goto fail_chunk_root; 203583121942SDavid Woodhouse if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) { 203683121942SDavid Woodhouse printk(KERN_WARNING "btrfs: failed to read tree root on %s\n", 203783121942SDavid Woodhouse sb->s_id); 203883121942SDavid Woodhouse goto fail_tree_root; 203983121942SDavid Woodhouse } 20405d4f98a2SYan Zheng btrfs_set_root_node(&tree_root->root_item, tree_root->node); 20415d4f98a2SYan Zheng tree_root->commit_root = btrfs_root_node(tree_root); 2042db94535dSChris Mason 2043db94535dSChris Mason ret = find_and_setup_root(tree_root, fs_info, 2044e20d96d6SChris Mason BTRFS_EXTENT_TREE_OBJECTID, extent_root); 20450b86a832SChris Mason if (ret) 204639279cc3SChris Mason goto fail_tree_root; 20470b86a832SChris Mason extent_root->track_dirty = 1; 20480b86a832SChris Mason 20490b86a832SChris Mason ret = find_and_setup_root(tree_root, fs_info, 20500b86a832SChris Mason BTRFS_DEV_TREE_OBJECTID, dev_root); 20510b86a832SChris Mason if (ret) 20520b86a832SChris Mason goto fail_extent_root; 20535d4f98a2SYan Zheng dev_root->track_dirty = 1; 20543768f368SChris Mason 2055d20f7043SChris Mason ret = find_and_setup_root(tree_root, fs_info, 2056d20f7043SChris Mason BTRFS_CSUM_TREE_OBJECTID, csum_root); 2057d20f7043SChris Mason if (ret) 20585d4f98a2SYan Zheng goto fail_dev_root; 2059d20f7043SChris Mason 2060d20f7043SChris Mason csum_root->track_dirty = 1; 2061d20f7043SChris Mason 20628929ecfaSYan, Zheng fs_info->generation = generation; 20638929ecfaSYan, Zheng fs_info->last_trans_committed = generation; 20648929ecfaSYan, Zheng fs_info->data_alloc_profile = (u64)-1; 20658929ecfaSYan, Zheng fs_info->metadata_alloc_profile = (u64)-1; 20668929ecfaSYan, Zheng fs_info->system_alloc_profile = fs_info->metadata_alloc_profile; 20678929ecfaSYan, Zheng 20681b1d1f66SJosef Bacik ret = btrfs_read_block_groups(extent_root); 20691b1d1f66SJosef Bacik if (ret) { 20701b1d1f66SJosef Bacik printk(KERN_ERR "Failed to read block groups: %d\n", ret); 20711b1d1f66SJosef Bacik goto fail_block_groups; 20721b1d1f66SJosef Bacik } 20739078a3e1SChris Mason 2074a74a4b97SChris Mason fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 2075a74a4b97SChris Mason "btrfs-cleaner"); 207657506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 20771b1d1f66SJosef Bacik goto fail_block_groups; 2078a74a4b97SChris Mason 2079a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 2080a74a4b97SChris Mason tree_root, 2081a74a4b97SChris Mason "btrfs-transaction"); 208257506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 20833f157a2fSChris Mason goto fail_cleaner; 2084a74a4b97SChris Mason 2085c289811cSChris Mason if (!btrfs_test_opt(tree_root, SSD) && 2086c289811cSChris Mason !btrfs_test_opt(tree_root, NOSSD) && 2087c289811cSChris Mason !fs_info->fs_devices->rotating) { 2088c289811cSChris Mason printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD " 2089c289811cSChris Mason "mode\n"); 2090c289811cSChris Mason btrfs_set_opt(fs_info->mount_opt, SSD); 2091c289811cSChris Mason } 2092c289811cSChris Mason 2093acce952bSliubo /* do not make disk changes in broken FS */ 2094acce952bSliubo if (btrfs_super_log_root(disk_super) != 0 && 2095acce952bSliubo !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) { 2096e02119d5SChris Mason u64 bytenr = btrfs_super_log_root(disk_super); 2097d18a2c44SChris Mason 20987c2ca468SChris Mason if (fs_devices->rw_devices == 0) { 2099d397712bSChris Mason printk(KERN_WARNING "Btrfs log replay required " 2100d397712bSChris Mason "on RO media\n"); 21017c2ca468SChris Mason err = -EIO; 21027c2ca468SChris Mason goto fail_trans_kthread; 21037c2ca468SChris Mason } 2104e02119d5SChris Mason blocksize = 2105e02119d5SChris Mason btrfs_level_size(tree_root, 2106e02119d5SChris Mason btrfs_super_log_root_level(disk_super)); 2107e02119d5SChris Mason 2108676e4c86SDan Carpenter log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS); 2109676e4c86SDan Carpenter if (!log_tree_root) { 2110676e4c86SDan Carpenter err = -ENOMEM; 2111676e4c86SDan Carpenter goto fail_trans_kthread; 2112676e4c86SDan Carpenter } 2113e02119d5SChris Mason 2114e02119d5SChris Mason __setup_root(nodesize, leafsize, sectorsize, stripesize, 2115e02119d5SChris Mason log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID); 2116e02119d5SChris Mason 2117e02119d5SChris Mason log_tree_root->node = read_tree_block(tree_root, bytenr, 211884234f3aSYan Zheng blocksize, 211984234f3aSYan Zheng generation + 1); 2120e02119d5SChris Mason ret = btrfs_recover_log_trees(log_tree_root); 2121e02119d5SChris Mason BUG_ON(ret); 2122e556ce2cSYan Zheng 2123e556ce2cSYan Zheng if (sb->s_flags & MS_RDONLY) { 2124e556ce2cSYan Zheng ret = btrfs_commit_super(tree_root); 2125e556ce2cSYan Zheng BUG_ON(ret); 2126e556ce2cSYan Zheng } 2127e02119d5SChris Mason } 21281a40e23bSZheng Yan 212976dda93cSYan, Zheng ret = btrfs_find_orphan_roots(tree_root); 213076dda93cSYan, Zheng BUG_ON(ret); 213176dda93cSYan, Zheng 21327c2ca468SChris Mason if (!(sb->s_flags & MS_RDONLY)) { 2133d68fc57bSYan, Zheng ret = btrfs_cleanup_fs_roots(fs_info); 2134d68fc57bSYan, Zheng BUG_ON(ret); 2135d68fc57bSYan, Zheng 21365d4f98a2SYan Zheng ret = btrfs_recover_relocation(tree_root); 2137d7ce5843SMiao Xie if (ret < 0) { 2138d7ce5843SMiao Xie printk(KERN_WARNING 2139d7ce5843SMiao Xie "btrfs: failed to recover relocation\n"); 2140d7ce5843SMiao Xie err = -EINVAL; 2141d7ce5843SMiao Xie goto fail_trans_kthread; 2142d7ce5843SMiao Xie } 21437c2ca468SChris Mason } 21441a40e23bSZheng Yan 21453de4586cSChris Mason location.objectid = BTRFS_FS_TREE_OBJECTID; 21463de4586cSChris Mason location.type = BTRFS_ROOT_ITEM_KEY; 21473de4586cSChris Mason location.offset = (u64)-1; 21483de4586cSChris Mason 21493de4586cSChris Mason fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location); 21503de4586cSChris Mason if (!fs_info->fs_root) 21517c2ca468SChris Mason goto fail_trans_kthread; 21523140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 21533140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 21543140c9a3SDan Carpenter goto fail_trans_kthread; 21553140c9a3SDan Carpenter } 2156c289811cSChris Mason 2157e3acc2a6SJosef Bacik if (!(sb->s_flags & MS_RDONLY)) { 2158e3acc2a6SJosef Bacik down_read(&fs_info->cleanup_work_sem); 215966b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_info->fs_root); 216066b4ffd1SJosef Bacik if (!err) 216166b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_info->tree_root); 2162e3acc2a6SJosef Bacik up_read(&fs_info->cleanup_work_sem); 216366b4ffd1SJosef Bacik if (err) { 216466b4ffd1SJosef Bacik close_ctree(tree_root); 216566b4ffd1SJosef Bacik return ERR_PTR(err); 216666b4ffd1SJosef Bacik } 2167e3acc2a6SJosef Bacik } 2168e3acc2a6SJosef Bacik 21690f7d52f4SChris Mason return tree_root; 217039279cc3SChris Mason 21717c2ca468SChris Mason fail_trans_kthread: 21727c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 21733f157a2fSChris Mason fail_cleaner: 2174a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 21757c2ca468SChris Mason 21767c2ca468SChris Mason /* 21777c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 21787c2ca468SChris Mason * kthreads 21797c2ca468SChris Mason */ 21807c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 21817c2ca468SChris Mason invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 21827c2ca468SChris Mason 21831b1d1f66SJosef Bacik fail_block_groups: 21841b1d1f66SJosef Bacik btrfs_free_block_groups(fs_info); 2185d20f7043SChris Mason free_extent_buffer(csum_root->node); 21865d4f98a2SYan Zheng free_extent_buffer(csum_root->commit_root); 21875d4f98a2SYan Zheng fail_dev_root: 21885d4f98a2SYan Zheng free_extent_buffer(dev_root->node); 21895d4f98a2SYan Zheng free_extent_buffer(dev_root->commit_root); 21900b86a832SChris Mason fail_extent_root: 21910b86a832SChris Mason free_extent_buffer(extent_root->node); 21925d4f98a2SYan Zheng free_extent_buffer(extent_root->commit_root); 219339279cc3SChris Mason fail_tree_root: 21945f39d397SChris Mason free_extent_buffer(tree_root->node); 21955d4f98a2SYan Zheng free_extent_buffer(tree_root->commit_root); 21962b82032cSYan Zheng fail_chunk_root: 21972b82032cSYan Zheng free_extent_buffer(chunk_root->node); 21985d4f98a2SYan Zheng free_extent_buffer(chunk_root->commit_root); 219939279cc3SChris Mason fail_sb_buffer: 220061d92c32SChris Mason btrfs_stop_workers(&fs_info->generic_worker); 2201247e743cSChris Mason btrfs_stop_workers(&fs_info->fixup_workers); 2202771ed689SChris Mason btrfs_stop_workers(&fs_info->delalloc_workers); 22038b712842SChris Mason btrfs_stop_workers(&fs_info->workers); 22048b712842SChris Mason btrfs_stop_workers(&fs_info->endio_workers); 2205d20f7043SChris Mason btrfs_stop_workers(&fs_info->endio_meta_workers); 2206cad321adSChris Mason btrfs_stop_workers(&fs_info->endio_meta_write_workers); 2207e6dcd2dcSChris Mason btrfs_stop_workers(&fs_info->endio_write_workers); 22080cb59c99SJosef Bacik btrfs_stop_workers(&fs_info->endio_freespace_worker); 22091cc127b5SChris Mason btrfs_stop_workers(&fs_info->submit_workers); 22104543df7eSChris Mason fail_iput: 22117c2ca468SChris Mason invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 22124543df7eSChris Mason iput(fs_info->btree_inode); 22137e662854SQinghuang Feng 2214dfe25020SChris Mason btrfs_close_devices(fs_info->fs_devices); 221584eed90fSChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 2216ad081f14SJens Axboe fail_bdi: 22177e662854SQinghuang Feng bdi_destroy(&fs_info->bdi); 221876dda93cSYan, Zheng fail_srcu: 221976dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 22207e662854SQinghuang Feng fail: 222139279cc3SChris Mason kfree(extent_root); 222239279cc3SChris Mason kfree(tree_root); 222339279cc3SChris Mason kfree(fs_info); 222483afeac4SJim Meyering kfree(chunk_root); 222583afeac4SJim Meyering kfree(dev_root); 2226d20f7043SChris Mason kfree(csum_root); 222739279cc3SChris Mason return ERR_PTR(err); 2228eb60ceacSChris Mason } 2229eb60ceacSChris Mason 2230f2984462SChris Mason static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate) 2231f2984462SChris Mason { 2232f2984462SChris Mason char b[BDEVNAME_SIZE]; 2233f2984462SChris Mason 2234f2984462SChris Mason if (uptodate) { 2235f2984462SChris Mason set_buffer_uptodate(bh); 2236f2984462SChris Mason } else { 2237c3b9a62cSChristoph Hellwig if (printk_ratelimit()) { 2238f2984462SChris Mason printk(KERN_WARNING "lost page write due to " 2239f2984462SChris Mason "I/O error on %s\n", 2240f2984462SChris Mason bdevname(bh->b_bdev, b)); 2241f2984462SChris Mason } 22421259ab75SChris Mason /* note, we dont' set_buffer_write_io_error because we have 22431259ab75SChris Mason * our own ways of dealing with the IO errors 22441259ab75SChris Mason */ 2245f2984462SChris Mason clear_buffer_uptodate(bh); 2246f2984462SChris Mason } 2247f2984462SChris Mason unlock_buffer(bh); 2248f2984462SChris Mason put_bh(bh); 2249f2984462SChris Mason } 2250f2984462SChris Mason 2251a512bbf8SYan Zheng struct buffer_head *btrfs_read_dev_super(struct block_device *bdev) 2252a512bbf8SYan Zheng { 2253a512bbf8SYan Zheng struct buffer_head *bh; 2254a512bbf8SYan Zheng struct buffer_head *latest = NULL; 2255a512bbf8SYan Zheng struct btrfs_super_block *super; 2256a512bbf8SYan Zheng int i; 2257a512bbf8SYan Zheng u64 transid = 0; 2258a512bbf8SYan Zheng u64 bytenr; 2259a512bbf8SYan Zheng 2260a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 2261a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 2262a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 2263a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 2264a512bbf8SYan Zheng */ 2265a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 2266a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 2267a512bbf8SYan Zheng if (bytenr + 4096 >= i_size_read(bdev->bd_inode)) 2268a512bbf8SYan Zheng break; 2269a512bbf8SYan Zheng bh = __bread(bdev, bytenr / 4096, 4096); 2270a512bbf8SYan Zheng if (!bh) 2271a512bbf8SYan Zheng continue; 2272a512bbf8SYan Zheng 2273a512bbf8SYan Zheng super = (struct btrfs_super_block *)bh->b_data; 2274a512bbf8SYan Zheng if (btrfs_super_bytenr(super) != bytenr || 2275a512bbf8SYan Zheng strncmp((char *)(&super->magic), BTRFS_MAGIC, 2276a512bbf8SYan Zheng sizeof(super->magic))) { 2277a512bbf8SYan Zheng brelse(bh); 2278a512bbf8SYan Zheng continue; 2279a512bbf8SYan Zheng } 2280a512bbf8SYan Zheng 2281a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 2282a512bbf8SYan Zheng brelse(latest); 2283a512bbf8SYan Zheng latest = bh; 2284a512bbf8SYan Zheng transid = btrfs_super_generation(super); 2285a512bbf8SYan Zheng } else { 2286a512bbf8SYan Zheng brelse(bh); 2287a512bbf8SYan Zheng } 2288a512bbf8SYan Zheng } 2289a512bbf8SYan Zheng return latest; 2290a512bbf8SYan Zheng } 2291a512bbf8SYan Zheng 22924eedeb75SHisashi Hifumi /* 22934eedeb75SHisashi Hifumi * this should be called twice, once with wait == 0 and 22944eedeb75SHisashi Hifumi * once with wait == 1. When wait == 0 is done, all the buffer heads 22954eedeb75SHisashi Hifumi * we write are pinned. 22964eedeb75SHisashi Hifumi * 22974eedeb75SHisashi Hifumi * They are released when wait == 1 is done. 22984eedeb75SHisashi Hifumi * max_mirrors must be the same for both runs, and it indicates how 22994eedeb75SHisashi Hifumi * many supers on this one device should be written. 23004eedeb75SHisashi Hifumi * 23014eedeb75SHisashi Hifumi * max_mirrors == 0 means to write them all. 23024eedeb75SHisashi Hifumi */ 2303a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 2304a512bbf8SYan Zheng struct btrfs_super_block *sb, 2305a512bbf8SYan Zheng int do_barriers, int wait, int max_mirrors) 2306a512bbf8SYan Zheng { 2307a512bbf8SYan Zheng struct buffer_head *bh; 2308a512bbf8SYan Zheng int i; 2309a512bbf8SYan Zheng int ret; 2310a512bbf8SYan Zheng int errors = 0; 2311a512bbf8SYan Zheng u32 crc; 2312a512bbf8SYan Zheng u64 bytenr; 2313a512bbf8SYan Zheng int last_barrier = 0; 2314a512bbf8SYan Zheng 2315a512bbf8SYan Zheng if (max_mirrors == 0) 2316a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 2317a512bbf8SYan Zheng 2318a512bbf8SYan Zheng /* make sure only the last submit_bh does a barrier */ 2319a512bbf8SYan Zheng if (do_barriers) { 2320a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 2321a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 2322a512bbf8SYan Zheng if (bytenr + BTRFS_SUPER_INFO_SIZE >= 2323a512bbf8SYan Zheng device->total_bytes) 2324a512bbf8SYan Zheng break; 2325a512bbf8SYan Zheng last_barrier = i; 2326a512bbf8SYan Zheng } 2327a512bbf8SYan Zheng } 2328a512bbf8SYan Zheng 2329a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 2330a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 2331a512bbf8SYan Zheng if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes) 2332a512bbf8SYan Zheng break; 2333a512bbf8SYan Zheng 2334a512bbf8SYan Zheng if (wait) { 2335a512bbf8SYan Zheng bh = __find_get_block(device->bdev, bytenr / 4096, 2336a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 2337a512bbf8SYan Zheng BUG_ON(!bh); 2338a512bbf8SYan Zheng wait_on_buffer(bh); 23394eedeb75SHisashi Hifumi if (!buffer_uptodate(bh)) 23404eedeb75SHisashi Hifumi errors++; 23414eedeb75SHisashi Hifumi 23424eedeb75SHisashi Hifumi /* drop our reference */ 23434eedeb75SHisashi Hifumi brelse(bh); 23444eedeb75SHisashi Hifumi 23454eedeb75SHisashi Hifumi /* drop the reference from the wait == 0 run */ 2346a512bbf8SYan Zheng brelse(bh); 2347a512bbf8SYan Zheng continue; 2348a512bbf8SYan Zheng } else { 2349a512bbf8SYan Zheng btrfs_set_super_bytenr(sb, bytenr); 2350a512bbf8SYan Zheng 2351a512bbf8SYan Zheng crc = ~(u32)0; 2352a512bbf8SYan Zheng crc = btrfs_csum_data(NULL, (char *)sb + 2353a512bbf8SYan Zheng BTRFS_CSUM_SIZE, crc, 2354a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE - 2355a512bbf8SYan Zheng BTRFS_CSUM_SIZE); 2356a512bbf8SYan Zheng btrfs_csum_final(crc, sb->csum); 2357a512bbf8SYan Zheng 23584eedeb75SHisashi Hifumi /* 23594eedeb75SHisashi Hifumi * one reference for us, and we leave it for the 23604eedeb75SHisashi Hifumi * caller 23614eedeb75SHisashi Hifumi */ 2362a512bbf8SYan Zheng bh = __getblk(device->bdev, bytenr / 4096, 2363a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 2364a512bbf8SYan Zheng memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE); 2365a512bbf8SYan Zheng 23664eedeb75SHisashi Hifumi /* one reference for submit_bh */ 2367a512bbf8SYan Zheng get_bh(bh); 23684eedeb75SHisashi Hifumi 23694eedeb75SHisashi Hifumi set_buffer_uptodate(bh); 2370a512bbf8SYan Zheng lock_buffer(bh); 2371a512bbf8SYan Zheng bh->b_end_io = btrfs_end_buffer_write_sync; 2372a512bbf8SYan Zheng } 2373a512bbf8SYan Zheng 2374c3b9a62cSChristoph Hellwig if (i == last_barrier && do_barriers) 2375c3b9a62cSChristoph Hellwig ret = submit_bh(WRITE_FLUSH_FUA, bh); 2376c3b9a62cSChristoph Hellwig else 2377ffbd517dSChris Mason ret = submit_bh(WRITE_SYNC, bh); 2378a512bbf8SYan Zheng 23794eedeb75SHisashi Hifumi if (ret) 2380a512bbf8SYan Zheng errors++; 2381a512bbf8SYan Zheng } 2382a512bbf8SYan Zheng return errors < i ? 0 : -1; 2383a512bbf8SYan Zheng } 2384a512bbf8SYan Zheng 2385a512bbf8SYan Zheng int write_all_supers(struct btrfs_root *root, int max_mirrors) 2386f2984462SChris Mason { 2387e5e9a520SChris Mason struct list_head *head; 2388f2984462SChris Mason struct btrfs_device *dev; 2389a061fc8dSChris Mason struct btrfs_super_block *sb; 2390f2984462SChris Mason struct btrfs_dev_item *dev_item; 2391f2984462SChris Mason int ret; 2392f2984462SChris Mason int do_barriers; 2393a236aed1SChris Mason int max_errors; 2394a236aed1SChris Mason int total_errors = 0; 2395a061fc8dSChris Mason u64 flags; 2396f2984462SChris Mason 2397a236aed1SChris Mason max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1; 2398f2984462SChris Mason do_barriers = !btrfs_test_opt(root, NOBARRIER); 2399f2984462SChris Mason 2400a061fc8dSChris Mason sb = &root->fs_info->super_for_commit; 2401a061fc8dSChris Mason dev_item = &sb->dev_item; 2402e5e9a520SChris Mason 2403e5e9a520SChris Mason mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 2404e5e9a520SChris Mason head = &root->fs_info->fs_devices->devices; 2405c6e30871SQinghuang Feng list_for_each_entry(dev, head, dev_list) { 2406dfe25020SChris Mason if (!dev->bdev) { 2407dfe25020SChris Mason total_errors++; 2408dfe25020SChris Mason continue; 2409dfe25020SChris Mason } 24102b82032cSYan Zheng if (!dev->in_fs_metadata || !dev->writeable) 2411dfe25020SChris Mason continue; 2412dfe25020SChris Mason 24132b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 2414a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 2415a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 2416a061fc8dSChris Mason btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes); 2417a061fc8dSChris Mason btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used); 2418a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 2419a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 2420a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 2421a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 24222b82032cSYan Zheng memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE); 2423a512bbf8SYan Zheng 2424a061fc8dSChris Mason flags = btrfs_super_flags(sb); 2425a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 2426f2984462SChris Mason 2427a512bbf8SYan Zheng ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors); 2428a236aed1SChris Mason if (ret) 2429a236aed1SChris Mason total_errors++; 2430f2984462SChris Mason } 2431a236aed1SChris Mason if (total_errors > max_errors) { 2432d397712bSChris Mason printk(KERN_ERR "btrfs: %d errors while writing supers\n", 2433d397712bSChris Mason total_errors); 2434a236aed1SChris Mason BUG(); 2435a236aed1SChris Mason } 2436f2984462SChris Mason 2437a512bbf8SYan Zheng total_errors = 0; 2438c6e30871SQinghuang Feng list_for_each_entry(dev, head, dev_list) { 2439dfe25020SChris Mason if (!dev->bdev) 2440dfe25020SChris Mason continue; 24412b82032cSYan Zheng if (!dev->in_fs_metadata || !dev->writeable) 2442dfe25020SChris Mason continue; 2443dfe25020SChris Mason 2444a512bbf8SYan Zheng ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors); 2445a512bbf8SYan Zheng if (ret) 24461259ab75SChris Mason total_errors++; 2447f2984462SChris Mason } 2448e5e9a520SChris Mason mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 2449a236aed1SChris Mason if (total_errors > max_errors) { 2450d397712bSChris Mason printk(KERN_ERR "btrfs: %d errors while writing supers\n", 2451d397712bSChris Mason total_errors); 2452a236aed1SChris Mason BUG(); 2453a236aed1SChris Mason } 2454f2984462SChris Mason return 0; 2455f2984462SChris Mason } 2456f2984462SChris Mason 2457a512bbf8SYan Zheng int write_ctree_super(struct btrfs_trans_handle *trans, 2458a512bbf8SYan Zheng struct btrfs_root *root, int max_mirrors) 2459cfaa7295SChris Mason { 2460e66f709bSChris Mason int ret; 24612c90e5d6SChris Mason 2462a512bbf8SYan Zheng ret = write_all_supers(root, max_mirrors); 24635f39d397SChris Mason return ret; 2464cfaa7295SChris Mason } 2465cfaa7295SChris Mason 24665eda7b5eSChris Mason int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root) 24672619ba1fSChris Mason { 24684df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 24692619ba1fSChris Mason radix_tree_delete(&fs_info->fs_roots_radix, 24702619ba1fSChris Mason (unsigned long)root->root_key.objectid); 24714df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 247276dda93cSYan, Zheng 247376dda93cSYan, Zheng if (btrfs_root_refs(&root->root_item) == 0) 247476dda93cSYan, Zheng synchronize_srcu(&fs_info->subvol_srcu); 247576dda93cSYan, Zheng 24764df27c4dSYan, Zheng free_fs_root(root); 24774df27c4dSYan, Zheng return 0; 24784df27c4dSYan, Zheng } 24794df27c4dSYan, Zheng 24804df27c4dSYan, Zheng static void free_fs_root(struct btrfs_root *root) 24814df27c4dSYan, Zheng { 24824df27c4dSYan, Zheng WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 24833394e160SChris Mason if (root->anon_super.s_dev) { 24843394e160SChris Mason down_write(&root->anon_super.s_umount); 24853394e160SChris Mason kill_anon_super(&root->anon_super); 24863394e160SChris Mason } 24875f39d397SChris Mason free_extent_buffer(root->node); 24885f39d397SChris Mason free_extent_buffer(root->commit_root); 248958176a96SJosef Bacik kfree(root->name); 24902619ba1fSChris Mason kfree(root); 24912619ba1fSChris Mason } 24922619ba1fSChris Mason 249335b7e476SChris Mason static int del_fs_roots(struct btrfs_fs_info *fs_info) 24940f7d52f4SChris Mason { 24950f7d52f4SChris Mason int ret; 24960f7d52f4SChris Mason struct btrfs_root *gang[8]; 24970f7d52f4SChris Mason int i; 24980f7d52f4SChris Mason 249976dda93cSYan, Zheng while (!list_empty(&fs_info->dead_roots)) { 250076dda93cSYan, Zheng gang[0] = list_entry(fs_info->dead_roots.next, 250176dda93cSYan, Zheng struct btrfs_root, root_list); 250276dda93cSYan, Zheng list_del(&gang[0]->root_list); 250376dda93cSYan, Zheng 250476dda93cSYan, Zheng if (gang[0]->in_radix) { 250576dda93cSYan, Zheng btrfs_free_fs_root(fs_info, gang[0]); 250676dda93cSYan, Zheng } else { 250776dda93cSYan, Zheng free_extent_buffer(gang[0]->node); 250876dda93cSYan, Zheng free_extent_buffer(gang[0]->commit_root); 250976dda93cSYan, Zheng kfree(gang[0]); 251076dda93cSYan, Zheng } 251176dda93cSYan, Zheng } 251276dda93cSYan, Zheng 25130f7d52f4SChris Mason while (1) { 25140f7d52f4SChris Mason ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 25150f7d52f4SChris Mason (void **)gang, 0, 25160f7d52f4SChris Mason ARRAY_SIZE(gang)); 25170f7d52f4SChris Mason if (!ret) 25180f7d52f4SChris Mason break; 25192619ba1fSChris Mason for (i = 0; i < ret; i++) 25205eda7b5eSChris Mason btrfs_free_fs_root(fs_info, gang[i]); 25210f7d52f4SChris Mason } 25220f7d52f4SChris Mason return 0; 25230f7d52f4SChris Mason } 2524b4100d64SChris Mason 2525c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 2526c146afadSYan Zheng { 2527c146afadSYan Zheng u64 root_objectid = 0; 2528c146afadSYan Zheng struct btrfs_root *gang[8]; 2529c146afadSYan Zheng int i; 2530c146afadSYan Zheng int ret; 2531c146afadSYan Zheng 2532c146afadSYan Zheng while (1) { 2533c146afadSYan Zheng ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2534c146afadSYan Zheng (void **)gang, root_objectid, 2535c146afadSYan Zheng ARRAY_SIZE(gang)); 2536c146afadSYan Zheng if (!ret) 2537c146afadSYan Zheng break; 25385d4f98a2SYan Zheng 25395d4f98a2SYan Zheng root_objectid = gang[ret - 1]->root_key.objectid + 1; 2540c146afadSYan Zheng for (i = 0; i < ret; i++) { 254166b4ffd1SJosef Bacik int err; 254266b4ffd1SJosef Bacik 2543c146afadSYan Zheng root_objectid = gang[i]->root_key.objectid; 254466b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(gang[i]); 254566b4ffd1SJosef Bacik if (err) 254666b4ffd1SJosef Bacik return err; 2547c146afadSYan Zheng } 2548c146afadSYan Zheng root_objectid++; 2549c146afadSYan Zheng } 2550c146afadSYan Zheng return 0; 2551c146afadSYan Zheng } 2552c146afadSYan Zheng 2553c146afadSYan Zheng int btrfs_commit_super(struct btrfs_root *root) 2554c146afadSYan Zheng { 2555c146afadSYan Zheng struct btrfs_trans_handle *trans; 2556c146afadSYan Zheng int ret; 2557c146afadSYan Zheng 2558c146afadSYan Zheng mutex_lock(&root->fs_info->cleaner_mutex); 255924bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 2560c146afadSYan Zheng btrfs_clean_old_snapshots(root); 2561c146afadSYan Zheng mutex_unlock(&root->fs_info->cleaner_mutex); 2562c71bf099SYan, Zheng 2563c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 2564c71bf099SYan, Zheng down_write(&root->fs_info->cleanup_work_sem); 2565c71bf099SYan, Zheng up_write(&root->fs_info->cleanup_work_sem); 2566c71bf099SYan, Zheng 2567a22285a6SYan, Zheng trans = btrfs_join_transaction(root, 1); 25683612b495STsutomu Itoh if (IS_ERR(trans)) 25693612b495STsutomu Itoh return PTR_ERR(trans); 2570c146afadSYan Zheng ret = btrfs_commit_transaction(trans, root); 2571c146afadSYan Zheng BUG_ON(ret); 2572c146afadSYan Zheng /* run commit again to drop the original snapshot */ 2573a22285a6SYan, Zheng trans = btrfs_join_transaction(root, 1); 25743612b495STsutomu Itoh if (IS_ERR(trans)) 25753612b495STsutomu Itoh return PTR_ERR(trans); 2576c146afadSYan Zheng btrfs_commit_transaction(trans, root); 2577c146afadSYan Zheng ret = btrfs_write_and_wait_transaction(NULL, root); 2578c146afadSYan Zheng BUG_ON(ret); 2579c146afadSYan Zheng 2580a512bbf8SYan Zheng ret = write_ctree_super(NULL, root, 0); 2581c146afadSYan Zheng return ret; 2582c146afadSYan Zheng } 2583c146afadSYan Zheng 2584e20d96d6SChris Mason int close_ctree(struct btrfs_root *root) 2585eb60ceacSChris Mason { 25860f7d52f4SChris Mason struct btrfs_fs_info *fs_info = root->fs_info; 2587c146afadSYan Zheng int ret; 2588e089f05cSChris Mason 2589facda1e7SChris Mason fs_info->closing = 1; 2590a2135011SChris Mason smp_mb(); 2591a2135011SChris Mason 25920af3d00bSJosef Bacik btrfs_put_block_group_cache(fs_info); 2593acce952bSliubo 2594acce952bSliubo /* 2595acce952bSliubo * Here come 2 situations when btrfs is broken to flip readonly: 2596acce952bSliubo * 2597acce952bSliubo * 1. when btrfs flips readonly somewhere else before 2598acce952bSliubo * btrfs_commit_super, sb->s_flags has MS_RDONLY flag, 2599acce952bSliubo * and btrfs will skip to write sb directly to keep 2600acce952bSliubo * ERROR state on disk. 2601acce952bSliubo * 2602acce952bSliubo * 2. when btrfs flips readonly just in btrfs_commit_super, 2603acce952bSliubo * and in such case, btrfs cannnot write sb via btrfs_commit_super, 2604acce952bSliubo * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag, 2605acce952bSliubo * btrfs will cleanup all FS resources first and write sb then. 2606acce952bSliubo */ 2607c146afadSYan Zheng if (!(fs_info->sb->s_flags & MS_RDONLY)) { 2608c146afadSYan Zheng ret = btrfs_commit_super(root); 2609d397712bSChris Mason if (ret) 2610d397712bSChris Mason printk(KERN_ERR "btrfs: commit super ret %d\n", ret); 2611c146afadSYan Zheng } 2612ed2ff2cbSChris Mason 2613acce952bSliubo if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { 2614acce952bSliubo ret = btrfs_error_commit_super(root); 2615acce952bSliubo if (ret) 2616acce952bSliubo printk(KERN_ERR "btrfs: commit super ret %d\n", ret); 2617acce952bSliubo } 2618acce952bSliubo 26198929ecfaSYan, Zheng kthread_stop(root->fs_info->transaction_kthread); 26208929ecfaSYan, Zheng kthread_stop(root->fs_info->cleaner_kthread); 26218929ecfaSYan, Zheng 2622f25784b3SYan Zheng fs_info->closing = 2; 2623f25784b3SYan Zheng smp_mb(); 2624f25784b3SYan Zheng 2625b0c68f8bSChris Mason if (fs_info->delalloc_bytes) { 2626d397712bSChris Mason printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n", 262721380931SJoel Becker (unsigned long long)fs_info->delalloc_bytes); 2628b0c68f8bSChris Mason } 262931153d81SYan Zheng if (fs_info->total_ref_cache_size) { 2630d397712bSChris Mason printk(KERN_INFO "btrfs: at umount reference cache size %llu\n", 2631d397712bSChris Mason (unsigned long long)fs_info->total_ref_cache_size); 263231153d81SYan Zheng } 263331153d81SYan Zheng 26345f39d397SChris Mason free_extent_buffer(fs_info->extent_root->node); 26355d4f98a2SYan Zheng free_extent_buffer(fs_info->extent_root->commit_root); 26365f39d397SChris Mason free_extent_buffer(fs_info->tree_root->node); 26375d4f98a2SYan Zheng free_extent_buffer(fs_info->tree_root->commit_root); 26380b86a832SChris Mason free_extent_buffer(root->fs_info->chunk_root->node); 26395d4f98a2SYan Zheng free_extent_buffer(root->fs_info->chunk_root->commit_root); 26400b86a832SChris Mason free_extent_buffer(root->fs_info->dev_root->node); 26415d4f98a2SYan Zheng free_extent_buffer(root->fs_info->dev_root->commit_root); 2642d20f7043SChris Mason free_extent_buffer(root->fs_info->csum_root->node); 26435d4f98a2SYan Zheng free_extent_buffer(root->fs_info->csum_root->commit_root); 2644d20f7043SChris Mason 26459078a3e1SChris Mason btrfs_free_block_groups(root->fs_info); 2646c146afadSYan Zheng 26470f7d52f4SChris Mason del_fs_roots(fs_info); 2648d10c5f31SChris Mason 2649c146afadSYan Zheng iput(fs_info->btree_inode); 26509ad6b7bcSChris Mason 265161d92c32SChris Mason btrfs_stop_workers(&fs_info->generic_worker); 2652247e743cSChris Mason btrfs_stop_workers(&fs_info->fixup_workers); 2653771ed689SChris Mason btrfs_stop_workers(&fs_info->delalloc_workers); 26548b712842SChris Mason btrfs_stop_workers(&fs_info->workers); 26558b712842SChris Mason btrfs_stop_workers(&fs_info->endio_workers); 2656d20f7043SChris Mason btrfs_stop_workers(&fs_info->endio_meta_workers); 2657cad321adSChris Mason btrfs_stop_workers(&fs_info->endio_meta_write_workers); 2658e6dcd2dcSChris Mason btrfs_stop_workers(&fs_info->endio_write_workers); 26590cb59c99SJosef Bacik btrfs_stop_workers(&fs_info->endio_freespace_worker); 26601cc127b5SChris Mason btrfs_stop_workers(&fs_info->submit_workers); 2661d6bfde87SChris Mason 2662dfe25020SChris Mason btrfs_close_devices(fs_info->fs_devices); 26630b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 2664b248a415SChris Mason 266504160088SChris Mason bdi_destroy(&fs_info->bdi); 266676dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 26670b86a832SChris Mason 26680f7d52f4SChris Mason kfree(fs_info->extent_root); 26690f7d52f4SChris Mason kfree(fs_info->tree_root); 26700b86a832SChris Mason kfree(fs_info->chunk_root); 26710b86a832SChris Mason kfree(fs_info->dev_root); 2672d20f7043SChris Mason kfree(fs_info->csum_root); 267383a4d548SLi Zefan kfree(fs_info); 267483a4d548SLi Zefan 2675eb60ceacSChris Mason return 0; 2676eb60ceacSChris Mason } 2677eb60ceacSChris Mason 26781259ab75SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid) 2679ccd467d6SChris Mason { 26801259ab75SChris Mason int ret; 2681810191ffSChris Mason struct inode *btree_inode = buf->first_page->mapping->host; 26821259ab75SChris Mason 26832ac55d41SJosef Bacik ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf, 26842ac55d41SJosef Bacik NULL); 26851259ab75SChris Mason if (!ret) 26861259ab75SChris Mason return ret; 26871259ab75SChris Mason 26881259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 26891259ab75SChris Mason parent_transid); 26901259ab75SChris Mason return !ret; 26915f39d397SChris Mason } 26926702ed49SChris Mason 26935f39d397SChris Mason int btrfs_set_buffer_uptodate(struct extent_buffer *buf) 26945f39d397SChris Mason { 2695810191ffSChris Mason struct inode *btree_inode = buf->first_page->mapping->host; 2696d1310b2eSChris Mason return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, 26975f39d397SChris Mason buf); 26985f39d397SChris Mason } 26995f39d397SChris Mason 27005f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 27015f39d397SChris Mason { 2702810191ffSChris Mason struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root; 27035f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 27045f39d397SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 2705b9473439SChris Mason int was_dirty; 2706b4ce94deSChris Mason 2707b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 2708ccd467d6SChris Mason if (transid != root->fs_info->generation) { 2709d397712bSChris Mason printk(KERN_CRIT "btrfs transid mismatch buffer %llu, " 2710d397712bSChris Mason "found %llu running %llu\n", 2711db94535dSChris Mason (unsigned long long)buf->start, 2712d397712bSChris Mason (unsigned long long)transid, 2713d397712bSChris Mason (unsigned long long)root->fs_info->generation); 2714ccd467d6SChris Mason WARN_ON(1); 2715ccd467d6SChris Mason } 2716b9473439SChris Mason was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, 2717b9473439SChris Mason buf); 2718b9473439SChris Mason if (!was_dirty) { 2719b9473439SChris Mason spin_lock(&root->fs_info->delalloc_lock); 2720b9473439SChris Mason root->fs_info->dirty_metadata_bytes += buf->len; 2721b9473439SChris Mason spin_unlock(&root->fs_info->delalloc_lock); 2722b9473439SChris Mason } 2723eb60ceacSChris Mason } 2724eb60ceacSChris Mason 2725d3c2fdcfSChris Mason void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr) 272635b7e476SChris Mason { 2727188de649SChris Mason /* 2728188de649SChris Mason * looks as though older kernels can get into trouble with 2729188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 2730188de649SChris Mason */ 2731d6bfde87SChris Mason u64 num_dirty; 2732771ed689SChris Mason unsigned long thresh = 32 * 1024 * 1024; 2733d6bfde87SChris Mason 27346933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 2735d6bfde87SChris Mason return; 2736d6bfde87SChris Mason 2737585ad2c3SChris Mason num_dirty = root->fs_info->dirty_metadata_bytes; 2738585ad2c3SChris Mason 2739d6bfde87SChris Mason if (num_dirty > thresh) { 2740d3c2fdcfSChris Mason balance_dirty_pages_ratelimited_nr( 2741304fced6SChris Mason root->fs_info->btree_inode->i_mapping, 1); 274235b7e476SChris Mason } 2743188de649SChris Mason return; 2744d6bfde87SChris Mason } 27456b80053dSChris Mason 2746ca7a79adSChris Mason int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid) 27476b80053dSChris Mason { 2748810191ffSChris Mason struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root; 2749ce9adaa5SChris Mason int ret; 2750ca7a79adSChris Mason ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 2751d397712bSChris Mason if (ret == 0) 2752b4ce94deSChris Mason set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags); 2753ce9adaa5SChris Mason return ret; 27546b80053dSChris Mason } 27550da5468fSChris Mason 27564bef0848SChris Mason int btree_lock_page_hook(struct page *page) 27574bef0848SChris Mason { 27584bef0848SChris Mason struct inode *inode = page->mapping->host; 2759b9473439SChris Mason struct btrfs_root *root = BTRFS_I(inode)->root; 27604bef0848SChris Mason struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 27614bef0848SChris Mason struct extent_buffer *eb; 27624bef0848SChris Mason unsigned long len; 27634bef0848SChris Mason u64 bytenr = page_offset(page); 27644bef0848SChris Mason 27654bef0848SChris Mason if (page->private == EXTENT_PAGE_PRIVATE) 27664bef0848SChris Mason goto out; 27674bef0848SChris Mason 27684bef0848SChris Mason len = page->private >> 2; 27694bef0848SChris Mason eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS); 27704bef0848SChris Mason if (!eb) 27714bef0848SChris Mason goto out; 27724bef0848SChris Mason 27734bef0848SChris Mason btrfs_tree_lock(eb); 27744bef0848SChris Mason btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); 2775b9473439SChris Mason 2776b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { 2777b9473439SChris Mason spin_lock(&root->fs_info->delalloc_lock); 2778b9473439SChris Mason if (root->fs_info->dirty_metadata_bytes >= eb->len) 2779b9473439SChris Mason root->fs_info->dirty_metadata_bytes -= eb->len; 2780b9473439SChris Mason else 2781b9473439SChris Mason WARN_ON(1); 2782b9473439SChris Mason spin_unlock(&root->fs_info->delalloc_lock); 2783b9473439SChris Mason } 2784b9473439SChris Mason 27854bef0848SChris Mason btrfs_tree_unlock(eb); 27864bef0848SChris Mason free_extent_buffer(eb); 27874bef0848SChris Mason out: 27884bef0848SChris Mason lock_page(page); 27894bef0848SChris Mason return 0; 27904bef0848SChris Mason } 27914bef0848SChris Mason 2792acce952bSliubo static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info, 2793acce952bSliubo int read_only) 2794acce952bSliubo { 2795acce952bSliubo if (read_only) 2796acce952bSliubo return; 2797acce952bSliubo 2798acce952bSliubo if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) 2799acce952bSliubo printk(KERN_WARNING "warning: mount fs with errors, " 2800acce952bSliubo "running btrfsck is recommended\n"); 2801acce952bSliubo } 2802acce952bSliubo 2803acce952bSliubo int btrfs_error_commit_super(struct btrfs_root *root) 2804acce952bSliubo { 2805acce952bSliubo int ret; 2806acce952bSliubo 2807acce952bSliubo mutex_lock(&root->fs_info->cleaner_mutex); 2808acce952bSliubo btrfs_run_delayed_iputs(root); 2809acce952bSliubo mutex_unlock(&root->fs_info->cleaner_mutex); 2810acce952bSliubo 2811acce952bSliubo down_write(&root->fs_info->cleanup_work_sem); 2812acce952bSliubo up_write(&root->fs_info->cleanup_work_sem); 2813acce952bSliubo 2814acce952bSliubo /* cleanup FS via transaction */ 2815acce952bSliubo btrfs_cleanup_transaction(root); 2816acce952bSliubo 2817acce952bSliubo ret = write_ctree_super(NULL, root, 0); 2818acce952bSliubo 2819acce952bSliubo return ret; 2820acce952bSliubo } 2821acce952bSliubo 2822acce952bSliubo static int btrfs_destroy_ordered_operations(struct btrfs_root *root) 2823acce952bSliubo { 2824acce952bSliubo struct btrfs_inode *btrfs_inode; 2825acce952bSliubo struct list_head splice; 2826acce952bSliubo 2827acce952bSliubo INIT_LIST_HEAD(&splice); 2828acce952bSliubo 2829acce952bSliubo mutex_lock(&root->fs_info->ordered_operations_mutex); 2830acce952bSliubo spin_lock(&root->fs_info->ordered_extent_lock); 2831acce952bSliubo 2832acce952bSliubo list_splice_init(&root->fs_info->ordered_operations, &splice); 2833acce952bSliubo while (!list_empty(&splice)) { 2834acce952bSliubo btrfs_inode = list_entry(splice.next, struct btrfs_inode, 2835acce952bSliubo ordered_operations); 2836acce952bSliubo 2837acce952bSliubo list_del_init(&btrfs_inode->ordered_operations); 2838acce952bSliubo 2839acce952bSliubo btrfs_invalidate_inodes(btrfs_inode->root); 2840acce952bSliubo } 2841acce952bSliubo 2842acce952bSliubo spin_unlock(&root->fs_info->ordered_extent_lock); 2843acce952bSliubo mutex_unlock(&root->fs_info->ordered_operations_mutex); 2844acce952bSliubo 2845acce952bSliubo return 0; 2846acce952bSliubo } 2847acce952bSliubo 2848acce952bSliubo static int btrfs_destroy_ordered_extents(struct btrfs_root *root) 2849acce952bSliubo { 2850acce952bSliubo struct list_head splice; 2851acce952bSliubo struct btrfs_ordered_extent *ordered; 2852acce952bSliubo struct inode *inode; 2853acce952bSliubo 2854acce952bSliubo INIT_LIST_HEAD(&splice); 2855acce952bSliubo 2856acce952bSliubo spin_lock(&root->fs_info->ordered_extent_lock); 2857acce952bSliubo 2858acce952bSliubo list_splice_init(&root->fs_info->ordered_extents, &splice); 2859acce952bSliubo while (!list_empty(&splice)) { 2860acce952bSliubo ordered = list_entry(splice.next, struct btrfs_ordered_extent, 2861acce952bSliubo root_extent_list); 2862acce952bSliubo 2863acce952bSliubo list_del_init(&ordered->root_extent_list); 2864acce952bSliubo atomic_inc(&ordered->refs); 2865acce952bSliubo 2866acce952bSliubo /* the inode may be getting freed (in sys_unlink path). */ 2867acce952bSliubo inode = igrab(ordered->inode); 2868acce952bSliubo 2869acce952bSliubo spin_unlock(&root->fs_info->ordered_extent_lock); 2870acce952bSliubo if (inode) 2871acce952bSliubo iput(inode); 2872acce952bSliubo 2873acce952bSliubo atomic_set(&ordered->refs, 1); 2874acce952bSliubo btrfs_put_ordered_extent(ordered); 2875acce952bSliubo 2876acce952bSliubo spin_lock(&root->fs_info->ordered_extent_lock); 2877acce952bSliubo } 2878acce952bSliubo 2879acce952bSliubo spin_unlock(&root->fs_info->ordered_extent_lock); 2880acce952bSliubo 2881acce952bSliubo return 0; 2882acce952bSliubo } 2883acce952bSliubo 2884acce952bSliubo static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 2885acce952bSliubo struct btrfs_root *root) 2886acce952bSliubo { 2887acce952bSliubo struct rb_node *node; 2888acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 2889acce952bSliubo struct btrfs_delayed_ref_node *ref; 2890acce952bSliubo int ret = 0; 2891acce952bSliubo 2892acce952bSliubo delayed_refs = &trans->delayed_refs; 2893acce952bSliubo 2894acce952bSliubo spin_lock(&delayed_refs->lock); 2895acce952bSliubo if (delayed_refs->num_entries == 0) { 2896acce952bSliubo printk(KERN_INFO "delayed_refs has NO entry\n"); 2897acce952bSliubo return ret; 2898acce952bSliubo } 2899acce952bSliubo 2900acce952bSliubo node = rb_first(&delayed_refs->root); 2901acce952bSliubo while (node) { 2902acce952bSliubo ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); 2903acce952bSliubo node = rb_next(node); 2904acce952bSliubo 2905acce952bSliubo ref->in_tree = 0; 2906acce952bSliubo rb_erase(&ref->rb_node, &delayed_refs->root); 2907acce952bSliubo delayed_refs->num_entries--; 2908acce952bSliubo 2909acce952bSliubo atomic_set(&ref->refs, 1); 2910acce952bSliubo if (btrfs_delayed_ref_is_head(ref)) { 2911acce952bSliubo struct btrfs_delayed_ref_head *head; 2912acce952bSliubo 2913acce952bSliubo head = btrfs_delayed_node_to_head(ref); 2914acce952bSliubo mutex_lock(&head->mutex); 2915acce952bSliubo kfree(head->extent_op); 2916acce952bSliubo delayed_refs->num_heads--; 2917acce952bSliubo if (list_empty(&head->cluster)) 2918acce952bSliubo delayed_refs->num_heads_ready--; 2919acce952bSliubo list_del_init(&head->cluster); 2920acce952bSliubo mutex_unlock(&head->mutex); 2921acce952bSliubo } 2922acce952bSliubo 2923acce952bSliubo spin_unlock(&delayed_refs->lock); 2924acce952bSliubo btrfs_put_delayed_ref(ref); 2925acce952bSliubo 2926acce952bSliubo cond_resched(); 2927acce952bSliubo spin_lock(&delayed_refs->lock); 2928acce952bSliubo } 2929acce952bSliubo 2930acce952bSliubo spin_unlock(&delayed_refs->lock); 2931acce952bSliubo 2932acce952bSliubo return ret; 2933acce952bSliubo } 2934acce952bSliubo 2935acce952bSliubo static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t) 2936acce952bSliubo { 2937acce952bSliubo struct btrfs_pending_snapshot *snapshot; 2938acce952bSliubo struct list_head splice; 2939acce952bSliubo 2940acce952bSliubo INIT_LIST_HEAD(&splice); 2941acce952bSliubo 2942acce952bSliubo list_splice_init(&t->pending_snapshots, &splice); 2943acce952bSliubo 2944acce952bSliubo while (!list_empty(&splice)) { 2945acce952bSliubo snapshot = list_entry(splice.next, 2946acce952bSliubo struct btrfs_pending_snapshot, 2947acce952bSliubo list); 2948acce952bSliubo 2949acce952bSliubo list_del_init(&snapshot->list); 2950acce952bSliubo 2951acce952bSliubo kfree(snapshot); 2952acce952bSliubo } 2953acce952bSliubo 2954acce952bSliubo return 0; 2955acce952bSliubo } 2956acce952bSliubo 2957acce952bSliubo static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 2958acce952bSliubo { 2959acce952bSliubo struct btrfs_inode *btrfs_inode; 2960acce952bSliubo struct list_head splice; 2961acce952bSliubo 2962acce952bSliubo INIT_LIST_HEAD(&splice); 2963acce952bSliubo 2964acce952bSliubo list_splice_init(&root->fs_info->delalloc_inodes, &splice); 2965acce952bSliubo 2966acce952bSliubo spin_lock(&root->fs_info->delalloc_lock); 2967acce952bSliubo 2968acce952bSliubo while (!list_empty(&splice)) { 2969acce952bSliubo btrfs_inode = list_entry(splice.next, struct btrfs_inode, 2970acce952bSliubo delalloc_inodes); 2971acce952bSliubo 2972acce952bSliubo list_del_init(&btrfs_inode->delalloc_inodes); 2973acce952bSliubo 2974acce952bSliubo btrfs_invalidate_inodes(btrfs_inode->root); 2975acce952bSliubo } 2976acce952bSliubo 2977acce952bSliubo spin_unlock(&root->fs_info->delalloc_lock); 2978acce952bSliubo 2979acce952bSliubo return 0; 2980acce952bSliubo } 2981acce952bSliubo 2982acce952bSliubo static int btrfs_destroy_marked_extents(struct btrfs_root *root, 2983acce952bSliubo struct extent_io_tree *dirty_pages, 2984acce952bSliubo int mark) 2985acce952bSliubo { 2986acce952bSliubo int ret; 2987acce952bSliubo struct page *page; 2988acce952bSliubo struct inode *btree_inode = root->fs_info->btree_inode; 2989acce952bSliubo struct extent_buffer *eb; 2990acce952bSliubo u64 start = 0; 2991acce952bSliubo u64 end; 2992acce952bSliubo u64 offset; 2993acce952bSliubo unsigned long index; 2994acce952bSliubo 2995acce952bSliubo while (1) { 2996acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 2997acce952bSliubo mark); 2998acce952bSliubo if (ret) 2999acce952bSliubo break; 3000acce952bSliubo 3001acce952bSliubo clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); 3002acce952bSliubo while (start <= end) { 3003acce952bSliubo index = start >> PAGE_CACHE_SHIFT; 3004acce952bSliubo start = (u64)(index + 1) << PAGE_CACHE_SHIFT; 3005acce952bSliubo page = find_get_page(btree_inode->i_mapping, index); 3006acce952bSliubo if (!page) 3007acce952bSliubo continue; 3008acce952bSliubo offset = page_offset(page); 3009acce952bSliubo 3010acce952bSliubo spin_lock(&dirty_pages->buffer_lock); 3011acce952bSliubo eb = radix_tree_lookup( 3012acce952bSliubo &(&BTRFS_I(page->mapping->host)->io_tree)->buffer, 3013acce952bSliubo offset >> PAGE_CACHE_SHIFT); 3014acce952bSliubo spin_unlock(&dirty_pages->buffer_lock); 3015acce952bSliubo if (eb) { 3016acce952bSliubo ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY, 3017acce952bSliubo &eb->bflags); 3018acce952bSliubo atomic_set(&eb->refs, 1); 3019acce952bSliubo } 3020acce952bSliubo if (PageWriteback(page)) 3021acce952bSliubo end_page_writeback(page); 3022acce952bSliubo 3023acce952bSliubo lock_page(page); 3024acce952bSliubo if (PageDirty(page)) { 3025acce952bSliubo clear_page_dirty_for_io(page); 3026acce952bSliubo spin_lock_irq(&page->mapping->tree_lock); 3027acce952bSliubo radix_tree_tag_clear(&page->mapping->page_tree, 3028acce952bSliubo page_index(page), 3029acce952bSliubo PAGECACHE_TAG_DIRTY); 3030acce952bSliubo spin_unlock_irq(&page->mapping->tree_lock); 3031acce952bSliubo } 3032acce952bSliubo 3033acce952bSliubo page->mapping->a_ops->invalidatepage(page, 0); 3034acce952bSliubo unlock_page(page); 3035acce952bSliubo } 3036acce952bSliubo } 3037acce952bSliubo 3038acce952bSliubo return ret; 3039acce952bSliubo } 3040acce952bSliubo 3041acce952bSliubo static int btrfs_destroy_pinned_extent(struct btrfs_root *root, 3042acce952bSliubo struct extent_io_tree *pinned_extents) 3043acce952bSliubo { 3044acce952bSliubo struct extent_io_tree *unpin; 3045acce952bSliubo u64 start; 3046acce952bSliubo u64 end; 3047acce952bSliubo int ret; 3048acce952bSliubo 3049acce952bSliubo unpin = pinned_extents; 3050acce952bSliubo while (1) { 3051acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 3052acce952bSliubo EXTENT_DIRTY); 3053acce952bSliubo if (ret) 3054acce952bSliubo break; 3055acce952bSliubo 3056acce952bSliubo /* opt_discard */ 3057acce952bSliubo ret = btrfs_error_discard_extent(root, start, end + 1 - start); 3058acce952bSliubo 3059acce952bSliubo clear_extent_dirty(unpin, start, end, GFP_NOFS); 3060acce952bSliubo btrfs_error_unpin_extent_range(root, start, end); 3061acce952bSliubo cond_resched(); 3062acce952bSliubo } 3063acce952bSliubo 3064acce952bSliubo return 0; 3065acce952bSliubo } 3066acce952bSliubo 3067acce952bSliubo static int btrfs_cleanup_transaction(struct btrfs_root *root) 3068acce952bSliubo { 3069acce952bSliubo struct btrfs_transaction *t; 3070acce952bSliubo LIST_HEAD(list); 3071acce952bSliubo 3072acce952bSliubo WARN_ON(1); 3073acce952bSliubo 3074acce952bSliubo mutex_lock(&root->fs_info->trans_mutex); 3075acce952bSliubo mutex_lock(&root->fs_info->transaction_kthread_mutex); 3076acce952bSliubo 3077acce952bSliubo list_splice_init(&root->fs_info->trans_list, &list); 3078acce952bSliubo while (!list_empty(&list)) { 3079acce952bSliubo t = list_entry(list.next, struct btrfs_transaction, list); 3080acce952bSliubo if (!t) 3081acce952bSliubo break; 3082acce952bSliubo 3083acce952bSliubo btrfs_destroy_ordered_operations(root); 3084acce952bSliubo 3085acce952bSliubo btrfs_destroy_ordered_extents(root); 3086acce952bSliubo 3087acce952bSliubo btrfs_destroy_delayed_refs(t, root); 3088acce952bSliubo 3089acce952bSliubo btrfs_block_rsv_release(root, 3090acce952bSliubo &root->fs_info->trans_block_rsv, 3091acce952bSliubo t->dirty_pages.dirty_bytes); 3092acce952bSliubo 3093acce952bSliubo /* FIXME: cleanup wait for commit */ 3094acce952bSliubo t->in_commit = 1; 3095acce952bSliubo t->blocked = 1; 3096acce952bSliubo if (waitqueue_active(&root->fs_info->transaction_blocked_wait)) 3097acce952bSliubo wake_up(&root->fs_info->transaction_blocked_wait); 3098acce952bSliubo 3099acce952bSliubo t->blocked = 0; 3100acce952bSliubo if (waitqueue_active(&root->fs_info->transaction_wait)) 3101acce952bSliubo wake_up(&root->fs_info->transaction_wait); 3102acce952bSliubo mutex_unlock(&root->fs_info->trans_mutex); 3103acce952bSliubo 3104acce952bSliubo mutex_lock(&root->fs_info->trans_mutex); 3105acce952bSliubo t->commit_done = 1; 3106acce952bSliubo if (waitqueue_active(&t->commit_wait)) 3107acce952bSliubo wake_up(&t->commit_wait); 3108acce952bSliubo mutex_unlock(&root->fs_info->trans_mutex); 3109acce952bSliubo 3110acce952bSliubo mutex_lock(&root->fs_info->trans_mutex); 3111acce952bSliubo 3112acce952bSliubo btrfs_destroy_pending_snapshots(t); 3113acce952bSliubo 3114acce952bSliubo btrfs_destroy_delalloc_inodes(root); 3115acce952bSliubo 3116acce952bSliubo spin_lock(&root->fs_info->new_trans_lock); 3117acce952bSliubo root->fs_info->running_transaction = NULL; 3118acce952bSliubo spin_unlock(&root->fs_info->new_trans_lock); 3119acce952bSliubo 3120acce952bSliubo btrfs_destroy_marked_extents(root, &t->dirty_pages, 3121acce952bSliubo EXTENT_DIRTY); 3122acce952bSliubo 3123acce952bSliubo btrfs_destroy_pinned_extent(root, 3124acce952bSliubo root->fs_info->pinned_extents); 3125acce952bSliubo 3126acce952bSliubo t->use_count = 0; 3127acce952bSliubo list_del_init(&t->list); 3128acce952bSliubo memset(t, 0, sizeof(*t)); 3129acce952bSliubo kmem_cache_free(btrfs_transaction_cachep, t); 3130acce952bSliubo } 3131acce952bSliubo 3132acce952bSliubo mutex_unlock(&root->fs_info->transaction_kthread_mutex); 3133acce952bSliubo mutex_unlock(&root->fs_info->trans_mutex); 3134acce952bSliubo 3135acce952bSliubo return 0; 3136acce952bSliubo } 3137acce952bSliubo 3138d1310b2eSChris Mason static struct extent_io_ops btree_extent_io_ops = { 31394bef0848SChris Mason .write_cache_pages_lock_hook = btree_lock_page_hook, 3140ce9adaa5SChris Mason .readpage_end_io_hook = btree_readpage_end_io_hook, 31410b86a832SChris Mason .submit_bio_hook = btree_submit_bio_hook, 3142239b14b3SChris Mason /* note we're sharing with inode.c for the merge bio hook */ 3143239b14b3SChris Mason .merge_bio_hook = btrfs_merge_bio_hook, 31440da5468fSChris Mason }; 3145